aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorRusty Russell <rusty@rustcorp.com.au>2009-07-30 18:03:45 -0400
committerRusty Russell <rusty@rustcorp.com.au>2009-07-30 02:33:45 -0400
commit2e04ef76916d1e29a077ea9d0f2003c8fd86724d (patch)
tree2ff8d625d6e467be9f9f1b67a3674cb6e125e970
parente969fed542cae08cb11d666efac4f7c5d624d09f (diff)
lguest: fix comment style
I don't really notice it (except to begrudge the extra vertical space), but Ingo does. And he pointed out that one excuse of lguest is as a teaching tool, it should set a good example. Signed-off-by: Rusty Russell <rusty@rustcorp.com.au> Cc: Ingo Molnar <mingo@redhat.com>
-rw-r--r--Documentation/lguest/lguest.c540
-rw-r--r--arch/x86/include/asm/lguest.h3
-rw-r--r--arch/x86/include/asm/lguest_hcall.h10
-rw-r--r--arch/x86/lguest/boot.c428
-rw-r--r--arch/x86/lguest/i386_head.S110
-rw-r--r--drivers/lguest/core.c114
-rw-r--r--drivers/lguest/hypercalls.c141
-rw-r--r--drivers/lguest/interrupts_and_traps.c288
-rw-r--r--drivers/lguest/lg.h23
-rw-r--r--drivers/lguest/lguest_device.c150
-rw-r--r--drivers/lguest/lguest_user.c137
-rw-r--r--drivers/lguest/page_tables.c427
-rw-r--r--drivers/lguest/segments.c106
-rw-r--r--drivers/lguest/x86/core.c372
-rw-r--r--drivers/lguest/x86/switcher_32.S18
-rw-r--r--include/linux/lguest.h36
-rw-r--r--include/linux/lguest_launcher.h18
17 files changed, 1906 insertions, 1015 deletions
diff --git a/Documentation/lguest/lguest.c b/Documentation/lguest/lguest.c
index 45d7d6dcae7a..aa66a52b73e9 100644
--- a/Documentation/lguest/lguest.c
+++ b/Documentation/lguest/lguest.c
@@ -1,7 +1,9 @@
1/*P:100 This is the Launcher code, a simple program which lays out the 1/*P:100
2 * "physical" memory for the new Guest by mapping the kernel image and 2 * This is the Launcher code, a simple program which lays out the "physical"
3 * the virtual devices, then opens /dev/lguest to tell the kernel 3 * memory for the new Guest by mapping the kernel image and the virtual
4 * about the Guest and control it. :*/ 4 * devices, then opens /dev/lguest to tell the kernel about the Guest and
5 * control it.
6:*/
5#define _LARGEFILE64_SOURCE 7#define _LARGEFILE64_SOURCE
6#define _GNU_SOURCE 8#define _GNU_SOURCE
7#include <stdio.h> 9#include <stdio.h>
@@ -46,13 +48,15 @@
46#include "linux/virtio_rng.h" 48#include "linux/virtio_rng.h"
47#include "linux/virtio_ring.h" 49#include "linux/virtio_ring.h"
48#include "asm/bootparam.h" 50#include "asm/bootparam.h"
49/*L:110 We can ignore the 39 include files we need for this program, but I do 51/*L:110
50 * want to draw attention to the use of kernel-style types. 52 * We can ignore the 39 include files we need for this program, but I do want
53 * to draw attention to the use of kernel-style types.
51 * 54 *
52 * As Linus said, "C is a Spartan language, and so should your naming be." I 55 * As Linus said, "C is a Spartan language, and so should your naming be." I
53 * like these abbreviations, so we define them here. Note that u64 is always 56 * like these abbreviations, so we define them here. Note that u64 is always
54 * unsigned long long, which works on all Linux systems: this means that we can 57 * unsigned long long, which works on all Linux systems: this means that we can
55 * use %llu in printf for any u64. */ 58 * use %llu in printf for any u64.
59 */
56typedef unsigned long long u64; 60typedef unsigned long long u64;
57typedef uint32_t u32; 61typedef uint32_t u32;
58typedef uint16_t u16; 62typedef uint16_t u16;
@@ -69,8 +73,10 @@ typedef uint8_t u8;
69/* This will occupy 3 pages: it must be a power of 2. */ 73/* This will occupy 3 pages: it must be a power of 2. */
70#define VIRTQUEUE_NUM 256 74#define VIRTQUEUE_NUM 256
71 75
72/*L:120 verbose is both a global flag and a macro. The C preprocessor allows 76/*L:120
73 * this, and although I wouldn't recommend it, it works quite nicely here. */ 77 * verbose is both a global flag and a macro. The C preprocessor allows
78 * this, and although I wouldn't recommend it, it works quite nicely here.
79 */
74static bool verbose; 80static bool verbose;
75#define verbose(args...) \ 81#define verbose(args...) \
76 do { if (verbose) printf(args); } while(0) 82 do { if (verbose) printf(args); } while(0)
@@ -100,8 +106,7 @@ struct device_list
100 106
101 /* A single linked list of devices. */ 107 /* A single linked list of devices. */
102 struct device *dev; 108 struct device *dev;
103 /* And a pointer to the last device for easy append and also for 109 /* And a pointer to the last device for easy append. */
104 * configuration appending. */
105 struct device *lastdev; 110 struct device *lastdev;
106}; 111};
107 112
@@ -168,20 +173,24 @@ static char **main_args;
168/* The original tty settings to restore on exit. */ 173/* The original tty settings to restore on exit. */
169static struct termios orig_term; 174static struct termios orig_term;
170 175
171/* We have to be careful with barriers: our devices are all run in separate 176/*
177 * We have to be careful with barriers: our devices are all run in separate
172 * threads and so we need to make sure that changes visible to the Guest happen 178 * threads and so we need to make sure that changes visible to the Guest happen
173 * in precise order. */ 179 * in precise order.
180 */
174#define wmb() __asm__ __volatile__("" : : : "memory") 181#define wmb() __asm__ __volatile__("" : : : "memory")
175#define mb() __asm__ __volatile__("" : : : "memory") 182#define mb() __asm__ __volatile__("" : : : "memory")
176 183
177/* Convert an iovec element to the given type. 184/*
185 * Convert an iovec element to the given type.
178 * 186 *
179 * This is a fairly ugly trick: we need to know the size of the type and 187 * This is a fairly ugly trick: we need to know the size of the type and
180 * alignment requirement to check the pointer is kosher. It's also nice to 188 * alignment requirement to check the pointer is kosher. It's also nice to
181 * have the name of the type in case we report failure. 189 * have the name of the type in case we report failure.
182 * 190 *
183 * Typing those three things all the time is cumbersome and error prone, so we 191 * Typing those three things all the time is cumbersome and error prone, so we
184 * have a macro which sets them all up and passes to the real function. */ 192 * have a macro which sets them all up and passes to the real function.
193 */
185#define convert(iov, type) \ 194#define convert(iov, type) \
186 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type)) 195 ((type *)_convert((iov), sizeof(type), __alignof__(type), #type))
187 196
@@ -198,8 +207,10 @@ static void *_convert(struct iovec *iov, size_t size, size_t align,
198/* Wrapper for the last available index. Makes it easier to change. */ 207/* Wrapper for the last available index. Makes it easier to change. */
199#define lg_last_avail(vq) ((vq)->last_avail_idx) 208#define lg_last_avail(vq) ((vq)->last_avail_idx)
200 209
201/* The virtio configuration space is defined to be little-endian. x86 is 210/*
202 * little-endian too, but it's nice to be explicit so we have these helpers. */ 211 * The virtio configuration space is defined to be little-endian. x86 is
212 * little-endian too, but it's nice to be explicit so we have these helpers.
213 */
203#define cpu_to_le16(v16) (v16) 214#define cpu_to_le16(v16) (v16)
204#define cpu_to_le32(v32) (v32) 215#define cpu_to_le32(v32) (v32)
205#define cpu_to_le64(v64) (v64) 216#define cpu_to_le64(v64) (v64)
@@ -241,11 +252,12 @@ static u8 *get_feature_bits(struct device *dev)
241 + dev->num_vq * sizeof(struct lguest_vqconfig); 252 + dev->num_vq * sizeof(struct lguest_vqconfig);
242} 253}
243 254
244/*L:100 The Launcher code itself takes us out into userspace, that scary place 255/*L:100
245 * where pointers run wild and free! Unfortunately, like most userspace 256 * The Launcher code itself takes us out into userspace, that scary place where
246 * programs, it's quite boring (which is why everyone likes to hack on the 257 * pointers run wild and free! Unfortunately, like most userspace programs,
247 * kernel!). Perhaps if you make up an Lguest Drinking Game at this point, it 258 * it's quite boring (which is why everyone likes to hack on the kernel!).
248 * will get you through this section. Or, maybe not. 259 * Perhaps if you make up an Lguest Drinking Game at this point, it will get
260 * you through this section. Or, maybe not.
249 * 261 *
250 * The Launcher sets up a big chunk of memory to be the Guest's "physical" 262 * The Launcher sets up a big chunk of memory to be the Guest's "physical"
251 * memory and stores it in "guest_base". In other words, Guest physical == 263 * memory and stores it in "guest_base". In other words, Guest physical ==
@@ -253,7 +265,8 @@ static u8 *get_feature_bits(struct device *dev)
253 * 265 *
254 * This can be tough to get your head around, but usually it just means that we 266 * This can be tough to get your head around, but usually it just means that we
255 * use these trivial conversion functions when the Guest gives us it's 267 * use these trivial conversion functions when the Guest gives us it's
256 * "physical" addresses: */ 268 * "physical" addresses:
269 */
257static void *from_guest_phys(unsigned long addr) 270static void *from_guest_phys(unsigned long addr)
258{ 271{
259 return guest_base + addr; 272 return guest_base + addr;
@@ -268,7 +281,8 @@ static unsigned long to_guest_phys(const void *addr)
268 * Loading the Kernel. 281 * Loading the Kernel.
269 * 282 *
270 * We start with couple of simple helper routines. open_or_die() avoids 283 * We start with couple of simple helper routines. open_or_die() avoids
271 * error-checking code cluttering the callers: */ 284 * error-checking code cluttering the callers:
285 */
272static int open_or_die(const char *name, int flags) 286static int open_or_die(const char *name, int flags)
273{ 287{
274 int fd = open(name, flags); 288 int fd = open(name, flags);
@@ -283,8 +297,10 @@ static void *map_zeroed_pages(unsigned int num)
283 int fd = open_or_die("/dev/zero", O_RDONLY); 297 int fd = open_or_die("/dev/zero", O_RDONLY);
284 void *addr; 298 void *addr;
285 299
286 /* We use a private mapping (ie. if we write to the page, it will be 300 /*
287 * copied). */ 301 * We use a private mapping (ie. if we write to the page, it will be
302 * copied).
303 */
288 addr = mmap(NULL, getpagesize() * num, 304 addr = mmap(NULL, getpagesize() * num,
289 PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0); 305 PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, fd, 0);
290 if (addr == MAP_FAILED) 306 if (addr == MAP_FAILED)
@@ -305,20 +321,24 @@ static void *get_pages(unsigned int num)
305 return addr; 321 return addr;
306} 322}
307 323
308/* This routine is used to load the kernel or initrd. It tries mmap, but if 324/*
325 * This routine is used to load the kernel or initrd. It tries mmap, but if
309 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries), 326 * that fails (Plan 9's kernel file isn't nicely aligned on page boundaries),
310 * it falls back to reading the memory in. */ 327 * it falls back to reading the memory in.
328 */
311static void map_at(int fd, void *addr, unsigned long offset, unsigned long len) 329static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
312{ 330{
313 ssize_t r; 331 ssize_t r;
314 332
315 /* We map writable even though for some segments are marked read-only. 333 /*
334 * We map writable even though for some segments are marked read-only.
316 * The kernel really wants to be writable: it patches its own 335 * The kernel really wants to be writable: it patches its own
317 * instructions. 336 * instructions.
318 * 337 *
319 * MAP_PRIVATE means that the page won't be copied until a write is 338 * MAP_PRIVATE means that the page won't be copied until a write is
320 * done to it. This allows us to share untouched memory between 339 * done to it. This allows us to share untouched memory between
321 * Guests. */ 340 * Guests.
341 */
322 if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC, 342 if (mmap(addr, len, PROT_READ|PROT_WRITE|PROT_EXEC,
323 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED) 343 MAP_FIXED|MAP_PRIVATE, fd, offset) != MAP_FAILED)
324 return; 344 return;
@@ -329,7 +349,8 @@ static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
329 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r); 349 err(1, "Reading offset %lu len %lu gave %zi", offset, len, r);
330} 350}
331 351
332/* This routine takes an open vmlinux image, which is in ELF, and maps it into 352/*
353 * This routine takes an open vmlinux image, which is in ELF, and maps it into
333 * the Guest memory. ELF = Embedded Linking Format, which is the format used 354 * the Guest memory. ELF = Embedded Linking Format, which is the format used
334 * by all modern binaries on Linux including the kernel. 355 * by all modern binaries on Linux including the kernel.
335 * 356 *
@@ -337,23 +358,28 @@ static void map_at(int fd, void *addr, unsigned long offset, unsigned long len)
337 * address. We use the physical address; the Guest will map itself to the 358 * address. We use the physical address; the Guest will map itself to the
338 * virtual address. 359 * virtual address.
339 * 360 *
340 * We return the starting address. */ 361 * We return the starting address.
362 */
341static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr) 363static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
342{ 364{
343 Elf32_Phdr phdr[ehdr->e_phnum]; 365 Elf32_Phdr phdr[ehdr->e_phnum];
344 unsigned int i; 366 unsigned int i;
345 367
346 /* Sanity checks on the main ELF header: an x86 executable with a 368 /*
347 * reasonable number of correctly-sized program headers. */ 369 * Sanity checks on the main ELF header: an x86 executable with a
370 * reasonable number of correctly-sized program headers.
371 */
348 if (ehdr->e_type != ET_EXEC 372 if (ehdr->e_type != ET_EXEC
349 || ehdr->e_machine != EM_386 373 || ehdr->e_machine != EM_386
350 || ehdr->e_phentsize != sizeof(Elf32_Phdr) 374 || ehdr->e_phentsize != sizeof(Elf32_Phdr)
351 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr)) 375 || ehdr->e_phnum < 1 || ehdr->e_phnum > 65536U/sizeof(Elf32_Phdr))
352 errx(1, "Malformed elf header"); 376 errx(1, "Malformed elf header");
353 377
354 /* An ELF executable contains an ELF header and a number of "program" 378 /*
379 * An ELF executable contains an ELF header and a number of "program"
355 * headers which indicate which parts ("segments") of the program to 380 * headers which indicate which parts ("segments") of the program to
356 * load where. */ 381 * load where.
382 */
357 383
358 /* We read in all the program headers at once: */ 384 /* We read in all the program headers at once: */
359 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0) 385 if (lseek(elf_fd, ehdr->e_phoff, SEEK_SET) < 0)
@@ -361,8 +387,10 @@ static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
361 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr)) 387 if (read(elf_fd, phdr, sizeof(phdr)) != sizeof(phdr))
362 err(1, "Reading program headers"); 388 err(1, "Reading program headers");
363 389
364 /* Try all the headers: there are usually only three. A read-only one, 390 /*
365 * a read-write one, and a "note" section which we don't load. */ 391 * Try all the headers: there are usually only three. A read-only one,
392 * a read-write one, and a "note" section which we don't load.
393 */
366 for (i = 0; i < ehdr->e_phnum; i++) { 394 for (i = 0; i < ehdr->e_phnum; i++) {
367 /* If this isn't a loadable segment, we ignore it */ 395 /* If this isn't a loadable segment, we ignore it */
368 if (phdr[i].p_type != PT_LOAD) 396 if (phdr[i].p_type != PT_LOAD)
@@ -380,13 +408,15 @@ static unsigned long map_elf(int elf_fd, const Elf32_Ehdr *ehdr)
380 return ehdr->e_entry; 408 return ehdr->e_entry;
381} 409}
382 410
383/*L:150 A bzImage, unlike an ELF file, is not meant to be loaded. You're 411/*L:150
384 * supposed to jump into it and it will unpack itself. We used to have to 412 * A bzImage, unlike an ELF file, is not meant to be loaded. You're supposed
385 * perform some hairy magic because the unpacking code scared me. 413 * to jump into it and it will unpack itself. We used to have to perform some
414 * hairy magic because the unpacking code scared me.
386 * 415 *
387 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote 416 * Fortunately, Jeremy Fitzhardinge convinced me it wasn't that hard and wrote
388 * a small patch to jump over the tricky bits in the Guest, so now we just read 417 * a small patch to jump over the tricky bits in the Guest, so now we just read
389 * the funky header so we know where in the file to load, and away we go! */ 418 * the funky header so we know where in the file to load, and away we go!
419 */
390static unsigned long load_bzimage(int fd) 420static unsigned long load_bzimage(int fd)
391{ 421{
392 struct boot_params boot; 422 struct boot_params boot;
@@ -394,8 +424,10 @@ static unsigned long load_bzimage(int fd)
394 /* Modern bzImages get loaded at 1M. */ 424 /* Modern bzImages get loaded at 1M. */
395 void *p = from_guest_phys(0x100000); 425 void *p = from_guest_phys(0x100000);
396 426
397 /* Go back to the start of the file and read the header. It should be 427 /*
398 * a Linux boot header (see Documentation/x86/i386/boot.txt) */ 428 * Go back to the start of the file and read the header. It should be
429 * a Linux boot header (see Documentation/x86/i386/boot.txt)
430 */
399 lseek(fd, 0, SEEK_SET); 431 lseek(fd, 0, SEEK_SET);
400 read(fd, &boot, sizeof(boot)); 432 read(fd, &boot, sizeof(boot));
401 433
@@ -414,9 +446,11 @@ static unsigned long load_bzimage(int fd)
414 return boot.hdr.code32_start; 446 return boot.hdr.code32_start;
415} 447}
416 448
417/*L:140 Loading the kernel is easy when it's a "vmlinux", but most kernels 449/*L:140
450 * Loading the kernel is easy when it's a "vmlinux", but most kernels
418 * come wrapped up in the self-decompressing "bzImage" format. With a little 451 * come wrapped up in the self-decompressing "bzImage" format. With a little
419 * work, we can load those, too. */ 452 * work, we can load those, too.
453 */
420static unsigned long load_kernel(int fd) 454static unsigned long load_kernel(int fd)
421{ 455{
422 Elf32_Ehdr hdr; 456 Elf32_Ehdr hdr;
@@ -433,24 +467,28 @@ static unsigned long load_kernel(int fd)
433 return load_bzimage(fd); 467 return load_bzimage(fd);
434} 468}
435 469
436/* This is a trivial little helper to align pages. Andi Kleen hated it because 470/*
471 * This is a trivial little helper to align pages. Andi Kleen hated it because
437 * it calls getpagesize() twice: "it's dumb code." 472 * it calls getpagesize() twice: "it's dumb code."
438 * 473 *
439 * Kernel guys get really het up about optimization, even when it's not 474 * Kernel guys get really het up about optimization, even when it's not
440 * necessary. I leave this code as a reaction against that. */ 475 * necessary. I leave this code as a reaction against that.
476 */
441static inline unsigned long page_align(unsigned long addr) 477static inline unsigned long page_align(unsigned long addr)
442{ 478{
443 /* Add upwards and truncate downwards. */ 479 /* Add upwards and truncate downwards. */
444 return ((addr + getpagesize()-1) & ~(getpagesize()-1)); 480 return ((addr + getpagesize()-1) & ~(getpagesize()-1));
445} 481}
446 482
447/*L:180 An "initial ram disk" is a disk image loaded into memory along with 483/*L:180
448 * the kernel which the kernel can use to boot from without needing any 484 * An "initial ram disk" is a disk image loaded into memory along with the
449 * drivers. Most distributions now use this as standard: the initrd contains 485 * kernel which the kernel can use to boot from without needing any drivers.
450 * the code to load the appropriate driver modules for the current machine. 486 * Most distributions now use this as standard: the initrd contains the code to
487 * load the appropriate driver modules for the current machine.
451 * 488 *
452 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its 489 * Importantly, James Morris works for RedHat, and Fedora uses initrds for its
453 * kernels. He sent me this (and tells me when I break it). */ 490 * kernels. He sent me this (and tells me when I break it).
491 */
454static unsigned long load_initrd(const char *name, unsigned long mem) 492static unsigned long load_initrd(const char *name, unsigned long mem)
455{ 493{
456 int ifd; 494 int ifd;
@@ -462,12 +500,16 @@ static unsigned long load_initrd(const char *name, unsigned long mem)
462 if (fstat(ifd, &st) < 0) 500 if (fstat(ifd, &st) < 0)
463 err(1, "fstat() on initrd '%s'", name); 501 err(1, "fstat() on initrd '%s'", name);
464 502
465 /* We map the initrd at the top of memory, but mmap wants it to be 503 /*
466 * page-aligned, so we round the size up for that. */ 504 * We map the initrd at the top of memory, but mmap wants it to be
505 * page-aligned, so we round the size up for that.
506 */
467 len = page_align(st.st_size); 507 len = page_align(st.st_size);
468 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size); 508 map_at(ifd, from_guest_phys(mem - len), 0, st.st_size);
469 /* Once a file is mapped, you can close the file descriptor. It's a 509 /*
470 * little odd, but quite useful. */ 510 * Once a file is mapped, you can close the file descriptor. It's a
511 * little odd, but quite useful.
512 */
471 close(ifd); 513 close(ifd);
472 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len); 514 verbose("mapped initrd %s size=%lu @ %p\n", name, len, (void*)mem-len);
473 515
@@ -476,8 +518,10 @@ static unsigned long load_initrd(const char *name, unsigned long mem)
476} 518}
477/*:*/ 519/*:*/
478 520
479/* Simple routine to roll all the commandline arguments together with spaces 521/*
480 * between them. */ 522 * Simple routine to roll all the commandline arguments together with spaces
523 * between them.
524 */
481static void concat(char *dst, char *args[]) 525static void concat(char *dst, char *args[])
482{ 526{
483 unsigned int i, len = 0; 527 unsigned int i, len = 0;
@@ -494,10 +538,12 @@ static void concat(char *dst, char *args[])
494 dst[len] = '\0'; 538 dst[len] = '\0';
495} 539}
496 540
497/*L:185 This is where we actually tell the kernel to initialize the Guest. We 541/*L:185
542 * This is where we actually tell the kernel to initialize the Guest. We
498 * saw the arguments it expects when we looked at initialize() in lguest_user.c: 543 * saw the arguments it expects when we looked at initialize() in lguest_user.c:
499 * the base of Guest "physical" memory, the top physical page to allow and the 544 * the base of Guest "physical" memory, the top physical page to allow and the
500 * entry point for the Guest. */ 545 * entry point for the Guest.
546 */
501static void tell_kernel(unsigned long start) 547static void tell_kernel(unsigned long start)
502{ 548{
503 unsigned long args[] = { LHREQ_INITIALIZE, 549 unsigned long args[] = { LHREQ_INITIALIZE,
@@ -522,20 +568,26 @@ static void tell_kernel(unsigned long start)
522static void *_check_pointer(unsigned long addr, unsigned int size, 568static void *_check_pointer(unsigned long addr, unsigned int size,
523 unsigned int line) 569 unsigned int line)
524{ 570{
525 /* We have to separately check addr and addr+size, because size could 571 /*
526 * be huge and addr + size might wrap around. */ 572 * We have to separately check addr and addr+size, because size could
573 * be huge and addr + size might wrap around.
574 */
527 if (addr >= guest_limit || addr + size >= guest_limit) 575 if (addr >= guest_limit || addr + size >= guest_limit)
528 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr); 576 errx(1, "%s:%i: Invalid address %#lx", __FILE__, line, addr);
529 /* We return a pointer for the caller's convenience, now we know it's 577 /*
530 * safe to use. */ 578 * We return a pointer for the caller's convenience, now we know it's
579 * safe to use.
580 */
531 return from_guest_phys(addr); 581 return from_guest_phys(addr);
532} 582}
533/* A macro which transparently hands the line number to the real function. */ 583/* A macro which transparently hands the line number to the real function. */
534#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__) 584#define check_pointer(addr,size) _check_pointer(addr, size, __LINE__)
535 585
536/* Each buffer in the virtqueues is actually a chain of descriptors. This 586/*
587 * Each buffer in the virtqueues is actually a chain of descriptors. This
537 * function returns the next descriptor in the chain, or vq->vring.num if we're 588 * function returns the next descriptor in the chain, or vq->vring.num if we're
538 * at the end. */ 589 * at the end.
590 */
539static unsigned next_desc(struct vring_desc *desc, 591static unsigned next_desc(struct vring_desc *desc,
540 unsigned int i, unsigned int max) 592 unsigned int i, unsigned int max)
541{ 593{
@@ -576,12 +628,14 @@ static void trigger_irq(struct virtqueue *vq)
576 err(1, "Triggering irq %i", vq->config.irq); 628 err(1, "Triggering irq %i", vq->config.irq);
577} 629}
578 630
579/* This looks in the virtqueue and for the first available buffer, and converts 631/*
632 * This looks in the virtqueue and for the first available buffer, and converts
580 * it to an iovec for convenient access. Since descriptors consist of some 633 * it to an iovec for convenient access. Since descriptors consist of some
581 * number of output then some number of input descriptors, it's actually two 634 * number of output then some number of input descriptors, it's actually two
582 * iovecs, but we pack them into one and note how many of each there were. 635 * iovecs, but we pack them into one and note how many of each there were.
583 * 636 *
584 * This function returns the descriptor number found. */ 637 * This function returns the descriptor number found.
638 */
585static unsigned wait_for_vq_desc(struct virtqueue *vq, 639static unsigned wait_for_vq_desc(struct virtqueue *vq,
586 struct iovec iov[], 640 struct iovec iov[],
587 unsigned int *out_num, unsigned int *in_num) 641 unsigned int *out_num, unsigned int *in_num)
@@ -599,8 +653,10 @@ static unsigned wait_for_vq_desc(struct virtqueue *vq,
599 /* OK, now we need to know about added descriptors. */ 653 /* OK, now we need to know about added descriptors. */
600 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY; 654 vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
601 655
602 /* They could have slipped one in as we were doing that: make 656 /*
603 * sure it's written, then check again. */ 657 * They could have slipped one in as we were doing that: make
658 * sure it's written, then check again.
659 */
604 mb(); 660 mb();
605 if (last_avail != vq->vring.avail->idx) { 661 if (last_avail != vq->vring.avail->idx) {
606 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY; 662 vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
@@ -620,8 +676,10 @@ static unsigned wait_for_vq_desc(struct virtqueue *vq,
620 errx(1, "Guest moved used index from %u to %u", 676 errx(1, "Guest moved used index from %u to %u",
621 last_avail, vq->vring.avail->idx); 677 last_avail, vq->vring.avail->idx);
622 678
623 /* Grab the next descriptor number they're advertising, and increment 679 /*
624 * the index we've seen. */ 680 * Grab the next descriptor number they're advertising, and increment
681 * the index we've seen.
682 */
625 head = vq->vring.avail->ring[last_avail % vq->vring.num]; 683 head = vq->vring.avail->ring[last_avail % vq->vring.num];
626 lg_last_avail(vq)++; 684 lg_last_avail(vq)++;
627 685
@@ -636,8 +694,10 @@ static unsigned wait_for_vq_desc(struct virtqueue *vq,
636 desc = vq->vring.desc; 694 desc = vq->vring.desc;
637 i = head; 695 i = head;
638 696
639 /* If this is an indirect entry, then this buffer contains a descriptor 697 /*
640 * table which we handle as if it's any normal descriptor chain. */ 698 * If this is an indirect entry, then this buffer contains a descriptor
699 * table which we handle as if it's any normal descriptor chain.
700 */
641 if (desc[i].flags & VRING_DESC_F_INDIRECT) { 701 if (desc[i].flags & VRING_DESC_F_INDIRECT) {
642 if (desc[i].len % sizeof(struct vring_desc)) 702 if (desc[i].len % sizeof(struct vring_desc))
643 errx(1, "Invalid size for indirect buffer table"); 703 errx(1, "Invalid size for indirect buffer table");
@@ -656,8 +716,10 @@ static unsigned wait_for_vq_desc(struct virtqueue *vq,
656 if (desc[i].flags & VRING_DESC_F_WRITE) 716 if (desc[i].flags & VRING_DESC_F_WRITE)
657 (*in_num)++; 717 (*in_num)++;
658 else { 718 else {
659 /* If it's an output descriptor, they're all supposed 719 /*
660 * to come before any input descriptors. */ 720 * If it's an output descriptor, they're all supposed
721 * to come before any input descriptors.
722 */
661 if (*in_num) 723 if (*in_num)
662 errx(1, "Descriptor has out after in"); 724 errx(1, "Descriptor has out after in");
663 (*out_num)++; 725 (*out_num)++;
@@ -671,14 +733,18 @@ static unsigned wait_for_vq_desc(struct virtqueue *vq,
671 return head; 733 return head;
672} 734}
673 735
674/* After we've used one of their buffers, we tell them about it. We'll then 736/*
675 * want to send them an interrupt, using trigger_irq(). */ 737 * After we've used one of their buffers, we tell them about it. We'll then
738 * want to send them an interrupt, using trigger_irq().
739 */
676static void add_used(struct virtqueue *vq, unsigned int head, int len) 740static void add_used(struct virtqueue *vq, unsigned int head, int len)
677{ 741{
678 struct vring_used_elem *used; 742 struct vring_used_elem *used;
679 743
680 /* The virtqueue contains a ring of used buffers. Get a pointer to the 744 /*
681 * next entry in that used ring. */ 745 * The virtqueue contains a ring of used buffers. Get a pointer to the
746 * next entry in that used ring.
747 */
682 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num]; 748 used = &vq->vring.used->ring[vq->vring.used->idx % vq->vring.num];
683 used->id = head; 749 used->id = head;
684 used->len = len; 750 used->len = len;
@@ -698,7 +764,8 @@ static void add_used_and_trigger(struct virtqueue *vq, unsigned head, int len)
698/* 764/*
699 * The Console 765 * The Console
700 * 766 *
701 * We associate some data with the console for our exit hack. */ 767 * We associate some data with the console for our exit hack.
768 */
702struct console_abort 769struct console_abort
703{ 770{
704 /* How many times have they hit ^C? */ 771 /* How many times have they hit ^C? */
@@ -725,20 +792,24 @@ static void console_input(struct virtqueue *vq)
725 if (len <= 0) { 792 if (len <= 0) {
726 /* Ran out of input? */ 793 /* Ran out of input? */
727 warnx("Failed to get console input, ignoring console."); 794 warnx("Failed to get console input, ignoring console.");
728 /* For simplicity, dying threads kill the whole Launcher. So 795 /*
729 * just nap here. */ 796 * For simplicity, dying threads kill the whole Launcher. So
797 * just nap here.
798 */
730 for (;;) 799 for (;;)
731 pause(); 800 pause();
732 } 801 }
733 802
734 add_used_and_trigger(vq, head, len); 803 add_used_and_trigger(vq, head, len);
735 804
736 /* Three ^C within one second? Exit. 805 /*
806 * Three ^C within one second? Exit.
737 * 807 *
738 * This is such a hack, but works surprisingly well. Each ^C has to 808 * This is such a hack, but works surprisingly well. Each ^C has to
739 * be in a buffer by itself, so they can't be too fast. But we check 809 * be in a buffer by itself, so they can't be too fast. But we check
740 * that we get three within about a second, so they can't be too 810 * that we get three within about a second, so they can't be too
741 * slow. */ 811 * slow.
812 */
742 if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) { 813 if (len != 1 || ((char *)iov[0].iov_base)[0] != 3) {
743 abort->count = 0; 814 abort->count = 0;
744 return; 815 return;
@@ -809,8 +880,7 @@ static bool will_block(int fd)
809 return select(fd+1, &fdset, NULL, NULL, &zero) != 1; 880 return select(fd+1, &fdset, NULL, NULL, &zero) != 1;
810} 881}
811 882
812/* This is where we handle packets coming in from the tun device to our 883/* This handles packets coming in from the tun device to our Guest. */
813 * Guest. */
814static void net_input(struct virtqueue *vq) 884static void net_input(struct virtqueue *vq)
815{ 885{
816 int len; 886 int len;
@@ -842,8 +912,10 @@ static int do_thread(void *_vq)
842 return 0; 912 return 0;
843} 913}
844 914
845/* When a child dies, we kill our entire process group with SIGTERM. This 915/*
846 * also has the side effect that the shell restores the console for us! */ 916 * When a child dies, we kill our entire process group with SIGTERM. This
917 * also has the side effect that the shell restores the console for us!
918 */
847static void kill_launcher(int signal) 919static void kill_launcher(int signal)
848{ 920{
849 kill(0, SIGTERM); 921 kill(0, SIGTERM);
@@ -880,9 +952,10 @@ static void reset_device(struct device *dev)
880 952
881static void create_thread(struct virtqueue *vq) 953static void create_thread(struct virtqueue *vq)
882{ 954{
883 /* Create stack for thread and run it. Since stack grows 955 /*
884 * upwards, we point the stack pointer to the end of this 956 * Create stack for thread and run it. Since the stack grows upwards,
885 * region. */ 957 * we point the stack pointer to the end of this region.
958 */
886 char *stack = malloc(32768); 959 char *stack = malloc(32768);
887 unsigned long args[] = { LHREQ_EVENTFD, 960 unsigned long args[] = { LHREQ_EVENTFD,
888 vq->config.pfn*getpagesize(), 0 }; 961 vq->config.pfn*getpagesize(), 0 };
@@ -981,8 +1054,11 @@ static void handle_output(unsigned long addr)
981 } 1054 }
982 } 1055 }
983 1056
984 /* Early console write is done using notify on a nul-terminated string 1057 /*
985 * in Guest memory. */ 1058 * Early console write is done using notify on a nul-terminated string
1059 * in Guest memory. It's also great for hacking debugging messages
1060 * into a Guest.
1061 */
986 if (addr >= guest_limit) 1062 if (addr >= guest_limit)
987 errx(1, "Bad NOTIFY %#lx", addr); 1063 errx(1, "Bad NOTIFY %#lx", addr);
988 1064
@@ -998,10 +1074,12 @@ static void handle_output(unsigned long addr)
998 * routines to allocate and manage them. 1074 * routines to allocate and manage them.
999 */ 1075 */
1000 1076
1001/* The layout of the device page is a "struct lguest_device_desc" followed by a 1077/*
1078 * The layout of the device page is a "struct lguest_device_desc" followed by a
1002 * number of virtqueue descriptors, then two sets of feature bits, then an 1079 * number of virtqueue descriptors, then two sets of feature bits, then an
1003 * array of configuration bytes. This routine returns the configuration 1080 * array of configuration bytes. This routine returns the configuration
1004 * pointer. */ 1081 * pointer.
1082 */
1005static u8 *device_config(const struct device *dev) 1083static u8 *device_config(const struct device *dev)
1006{ 1084{
1007 return (void *)(dev->desc + 1) 1085 return (void *)(dev->desc + 1)
@@ -1009,9 +1087,11 @@ static u8 *device_config(const struct device *dev)
1009 + dev->feature_len * 2; 1087 + dev->feature_len * 2;
1010} 1088}
1011 1089
1012/* This routine allocates a new "struct lguest_device_desc" from descriptor 1090/*
1091 * This routine allocates a new "struct lguest_device_desc" from descriptor
1013 * table page just above the Guest's normal memory. It returns a pointer to 1092 * table page just above the Guest's normal memory. It returns a pointer to
1014 * that descriptor. */ 1093 * that descriptor.
1094 */
1015static struct lguest_device_desc *new_dev_desc(u16 type) 1095static struct lguest_device_desc *new_dev_desc(u16 type)
1016{ 1096{
1017 struct lguest_device_desc d = { .type = type }; 1097 struct lguest_device_desc d = { .type = type };
@@ -1032,8 +1112,10 @@ static struct lguest_device_desc *new_dev_desc(u16 type)
1032 return memcpy(p, &d, sizeof(d)); 1112 return memcpy(p, &d, sizeof(d));
1033} 1113}
1034 1114
1035/* Each device descriptor is followed by the description of its virtqueues. We 1115/*
1036 * specify how many descriptors the virtqueue is to have. */ 1116 * Each device descriptor is followed by the description of its virtqueues. We
1117 * specify how many descriptors the virtqueue is to have.
1118 */
1037static void add_virtqueue(struct device *dev, unsigned int num_descs, 1119static void add_virtqueue(struct device *dev, unsigned int num_descs,
1038 void (*service)(struct virtqueue *)) 1120 void (*service)(struct virtqueue *))
1039{ 1121{
@@ -1061,10 +1143,12 @@ static void add_virtqueue(struct device *dev, unsigned int num_descs,
1061 /* Initialize the vring. */ 1143 /* Initialize the vring. */
1062 vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN); 1144 vring_init(&vq->vring, num_descs, p, LGUEST_VRING_ALIGN);
1063 1145
1064 /* Append virtqueue to this device's descriptor. We use 1146 /*
1147 * Append virtqueue to this device's descriptor. We use
1065 * device_config() to get the end of the device's current virtqueues; 1148 * device_config() to get the end of the device's current virtqueues;
1066 * we check that we haven't added any config or feature information 1149 * we check that we haven't added any config or feature information
1067 * yet, otherwise we'd be overwriting them. */ 1150 * yet, otherwise we'd be overwriting them.
1151 */
1068 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0); 1152 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1069 memcpy(device_config(dev), &vq->config, sizeof(vq->config)); 1153 memcpy(device_config(dev), &vq->config, sizeof(vq->config));
1070 dev->num_vq++; 1154 dev->num_vq++;
@@ -1072,14 +1156,18 @@ static void add_virtqueue(struct device *dev, unsigned int num_descs,
1072 1156
1073 verbose("Virtqueue page %#lx\n", to_guest_phys(p)); 1157 verbose("Virtqueue page %#lx\n", to_guest_phys(p));
1074 1158
1075 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is 1159 /*
1076 * second. */ 1160 * Add to tail of list, so dev->vq is first vq, dev->vq->next is
1161 * second.
1162 */
1077 for (i = &dev->vq; *i; i = &(*i)->next); 1163 for (i = &dev->vq; *i; i = &(*i)->next);
1078 *i = vq; 1164 *i = vq;
1079} 1165}
1080 1166
1081/* The first half of the feature bitmask is for us to advertise features. The 1167/*
1082 * second half is for the Guest to accept features. */ 1168 * The first half of the feature bitmask is for us to advertise features. The
1169 * second half is for the Guest to accept features.
1170 */
1083static void add_feature(struct device *dev, unsigned bit) 1171static void add_feature(struct device *dev, unsigned bit)
1084{ 1172{
1085 u8 *features = get_feature_bits(dev); 1173 u8 *features = get_feature_bits(dev);
@@ -1093,9 +1181,11 @@ static void add_feature(struct device *dev, unsigned bit)
1093 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT)); 1181 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1094} 1182}
1095 1183
1096/* This routine sets the configuration fields for an existing device's 1184/*
1185 * This routine sets the configuration fields for an existing device's
1097 * descriptor. It only works for the last device, but that's OK because that's 1186 * descriptor. It only works for the last device, but that's OK because that's
1098 * how we use it. */ 1187 * how we use it.
1188 */
1099static void set_config(struct device *dev, unsigned len, const void *conf) 1189static void set_config(struct device *dev, unsigned len, const void *conf)
1100{ 1190{
1101 /* Check we haven't overflowed our single page. */ 1191 /* Check we haven't overflowed our single page. */
@@ -1110,10 +1200,12 @@ static void set_config(struct device *dev, unsigned len, const void *conf)
1110 assert(dev->desc->config_len == len); 1200 assert(dev->desc->config_len == len);
1111} 1201}
1112 1202
1113/* This routine does all the creation and setup of a new device, including 1203/*
1204 * This routine does all the creation and setup of a new device, including
1114 * calling new_dev_desc() to allocate the descriptor and device memory. 1205 * calling new_dev_desc() to allocate the descriptor and device memory.
1115 * 1206 *
1116 * See what I mean about userspace being boring? */ 1207 * See what I mean about userspace being boring?
1208 */
1117static struct device *new_device(const char *name, u16 type) 1209static struct device *new_device(const char *name, u16 type)
1118{ 1210{
1119 struct device *dev = malloc(sizeof(*dev)); 1211 struct device *dev = malloc(sizeof(*dev));
@@ -1126,10 +1218,12 @@ static struct device *new_device(const char *name, u16 type)
1126 dev->num_vq = 0; 1218 dev->num_vq = 0;
1127 dev->running = false; 1219 dev->running = false;
1128 1220
1129 /* Append to device list. Prepending to a single-linked list is 1221 /*
1222 * Append to device list. Prepending to a single-linked list is
1130 * easier, but the user expects the devices to be arranged on the bus 1223 * easier, but the user expects the devices to be arranged on the bus
1131 * in command-line order. The first network device on the command line 1224 * in command-line order. The first network device on the command line
1132 * is eth0, the first block device /dev/vda, etc. */ 1225 * is eth0, the first block device /dev/vda, etc.
1226 */
1133 if (devices.lastdev) 1227 if (devices.lastdev)
1134 devices.lastdev->next = dev; 1228 devices.lastdev->next = dev;
1135 else 1229 else
@@ -1139,8 +1233,10 @@ static struct device *new_device(const char *name, u16 type)
1139 return dev; 1233 return dev;
1140} 1234}
1141 1235
1142/* Our first setup routine is the console. It's a fairly simple device, but 1236/*
1143 * UNIX tty handling makes it uglier than it could be. */ 1237 * Our first setup routine is the console. It's a fairly simple device, but
1238 * UNIX tty handling makes it uglier than it could be.
1239 */
1144static void setup_console(void) 1240static void setup_console(void)
1145{ 1241{
1146 struct device *dev; 1242 struct device *dev;
@@ -1148,8 +1244,10 @@ static void setup_console(void)
1148 /* If we can save the initial standard input settings... */ 1244 /* If we can save the initial standard input settings... */
1149 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) { 1245 if (tcgetattr(STDIN_FILENO, &orig_term) == 0) {
1150 struct termios term = orig_term; 1246 struct termios term = orig_term;
1151 /* Then we turn off echo, line buffering and ^C etc. We want a 1247 /*
1152 * raw input stream to the Guest. */ 1248 * Then we turn off echo, line buffering and ^C etc: We want a
1249 * raw input stream to the Guest.
1250 */
1153 term.c_lflag &= ~(ISIG|ICANON|ECHO); 1251 term.c_lflag &= ~(ISIG|ICANON|ECHO);
1154 tcsetattr(STDIN_FILENO, TCSANOW, &term); 1252 tcsetattr(STDIN_FILENO, TCSANOW, &term);
1155 } 1253 }
@@ -1160,10 +1258,12 @@ static void setup_console(void)
1160 dev->priv = malloc(sizeof(struct console_abort)); 1258 dev->priv = malloc(sizeof(struct console_abort));
1161 ((struct console_abort *)dev->priv)->count = 0; 1259 ((struct console_abort *)dev->priv)->count = 0;
1162 1260
1163 /* The console needs two virtqueues: the input then the output. When 1261 /*
1262 * The console needs two virtqueues: the input then the output. When
1164 * they put something the input queue, we make sure we're listening to 1263 * they put something the input queue, we make sure we're listening to
1165 * stdin. When they put something in the output queue, we write it to 1264 * stdin. When they put something in the output queue, we write it to
1166 * stdout. */ 1265 * stdout.
1266 */
1167 add_virtqueue(dev, VIRTQUEUE_NUM, console_input); 1267 add_virtqueue(dev, VIRTQUEUE_NUM, console_input);
1168 add_virtqueue(dev, VIRTQUEUE_NUM, console_output); 1268 add_virtqueue(dev, VIRTQUEUE_NUM, console_output);
1169 1269
@@ -1171,7 +1271,8 @@ static void setup_console(void)
1171} 1271}
1172/*:*/ 1272/*:*/
1173 1273
1174/*M:010 Inter-guest networking is an interesting area. Simplest is to have a 1274/*M:010
1275 * Inter-guest networking is an interesting area. Simplest is to have a
1175 * --sharenet=<name> option which opens or creates a named pipe. This can be 1276 * --sharenet=<name> option which opens or creates a named pipe. This can be
1176 * used to send packets to another guest in a 1:1 manner. 1277 * used to send packets to another guest in a 1:1 manner.
1177 * 1278 *
@@ -1185,7 +1286,8 @@ static void setup_console(void)
1185 * multiple inter-guest channels behind one interface, although it would 1286 * multiple inter-guest channels behind one interface, although it would
1186 * require some manner of hotplugging new virtio channels. 1287 * require some manner of hotplugging new virtio channels.
1187 * 1288 *
1188 * Finally, we could implement a virtio network switch in the kernel. :*/ 1289 * Finally, we could implement a virtio network switch in the kernel.
1290:*/
1189 1291
1190static u32 str2ip(const char *ipaddr) 1292static u32 str2ip(const char *ipaddr)
1191{ 1293{
@@ -1210,11 +1312,13 @@ static void str2mac(const char *macaddr, unsigned char mac[6])
1210 mac[5] = m[5]; 1312 mac[5] = m[5];
1211} 1313}
1212 1314
1213/* This code is "adapted" from libbridge: it attaches the Host end of the 1315/*
1316 * This code is "adapted" from libbridge: it attaches the Host end of the
1214 * network device to the bridge device specified by the command line. 1317 * network device to the bridge device specified by the command line.
1215 * 1318 *
1216 * This is yet another James Morris contribution (I'm an IP-level guy, so I 1319 * This is yet another James Morris contribution (I'm an IP-level guy, so I
1217 * dislike bridging), and I just try not to break it. */ 1320 * dislike bridging), and I just try not to break it.
1321 */
1218static void add_to_bridge(int fd, const char *if_name, const char *br_name) 1322static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1219{ 1323{
1220 int ifidx; 1324 int ifidx;
@@ -1234,9 +1338,11 @@ static void add_to_bridge(int fd, const char *if_name, const char *br_name)
1234 err(1, "can't add %s to bridge %s", if_name, br_name); 1338 err(1, "can't add %s to bridge %s", if_name, br_name);
1235} 1339}
1236 1340
1237/* This sets up the Host end of the network device with an IP address, brings 1341/*
1342 * This sets up the Host end of the network device with an IP address, brings
1238 * it up so packets will flow, the copies the MAC address into the hwaddr 1343 * it up so packets will flow, the copies the MAC address into the hwaddr
1239 * pointer. */ 1344 * pointer.
1345 */
1240static void configure_device(int fd, const char *tapif, u32 ipaddr) 1346static void configure_device(int fd, const char *tapif, u32 ipaddr)
1241{ 1347{
1242 struct ifreq ifr; 1348 struct ifreq ifr;
@@ -1263,10 +1369,12 @@ static int get_tun_device(char tapif[IFNAMSIZ])
1263 /* Start with this zeroed. Messy but sure. */ 1369 /* Start with this zeroed. Messy but sure. */
1264 memset(&ifr, 0, sizeof(ifr)); 1370 memset(&ifr, 0, sizeof(ifr));
1265 1371
1266 /* We open the /dev/net/tun device and tell it we want a tap device. A 1372 /*
1373 * We open the /dev/net/tun device and tell it we want a tap device. A
1267 * tap device is like a tun device, only somehow different. To tell 1374 * tap device is like a tun device, only somehow different. To tell
1268 * the truth, I completely blundered my way through this code, but it 1375 * the truth, I completely blundered my way through this code, but it
1269 * works now! */ 1376 * works now!
1377 */
1270 netfd = open_or_die("/dev/net/tun", O_RDWR); 1378 netfd = open_or_die("/dev/net/tun", O_RDWR);
1271 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; 1379 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
1272 strcpy(ifr.ifr_name, "tap%d"); 1380 strcpy(ifr.ifr_name, "tap%d");
@@ -1277,18 +1385,22 @@ static int get_tun_device(char tapif[IFNAMSIZ])
1277 TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0) 1385 TUN_F_CSUM|TUN_F_TSO4|TUN_F_TSO6|TUN_F_TSO_ECN) != 0)
1278 err(1, "Could not set features for tun device"); 1386 err(1, "Could not set features for tun device");
1279 1387
1280 /* We don't need checksums calculated for packets coming in this 1388 /*
1281 * device: trust us! */ 1389 * We don't need checksums calculated for packets coming in this
1390 * device: trust us!
1391 */
1282 ioctl(netfd, TUNSETNOCSUM, 1); 1392 ioctl(netfd, TUNSETNOCSUM, 1);
1283 1393
1284 memcpy(tapif, ifr.ifr_name, IFNAMSIZ); 1394 memcpy(tapif, ifr.ifr_name, IFNAMSIZ);
1285 return netfd; 1395 return netfd;
1286} 1396}
1287 1397
1288/*L:195 Our network is a Host<->Guest network. This can either use bridging or 1398/*L:195
1399 * Our network is a Host<->Guest network. This can either use bridging or
1289 * routing, but the principle is the same: it uses the "tun" device to inject 1400 * routing, but the principle is the same: it uses the "tun" device to inject
1290 * packets into the Host as if they came in from a normal network card. We 1401 * packets into the Host as if they came in from a normal network card. We
1291 * just shunt packets between the Guest and the tun device. */ 1402 * just shunt packets between the Guest and the tun device.
1403 */
1292static void setup_tun_net(char *arg) 1404static void setup_tun_net(char *arg)
1293{ 1405{
1294 struct device *dev; 1406 struct device *dev;
@@ -1305,13 +1417,14 @@ static void setup_tun_net(char *arg)
1305 dev = new_device("net", VIRTIO_ID_NET); 1417 dev = new_device("net", VIRTIO_ID_NET);
1306 dev->priv = net_info; 1418 dev->priv = net_info;
1307 1419
1308 /* Network devices need a receive and a send queue, just like 1420 /* Network devices need a recv and a send queue, just like console. */
1309 * console. */
1310 add_virtqueue(dev, VIRTQUEUE_NUM, net_input); 1421 add_virtqueue(dev, VIRTQUEUE_NUM, net_input);
1311 add_virtqueue(dev, VIRTQUEUE_NUM, net_output); 1422 add_virtqueue(dev, VIRTQUEUE_NUM, net_output);
1312 1423
1313 /* We need a socket to perform the magic network ioctls to bring up the 1424 /*
1314 * tap interface, connect to the bridge etc. Any socket will do! */ 1425 * We need a socket to perform the magic network ioctls to bring up the
1426 * tap interface, connect to the bridge etc. Any socket will do!
1427 */
1315 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP); 1428 ipfd = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
1316 if (ipfd < 0) 1429 if (ipfd < 0)
1317 err(1, "opening IP socket"); 1430 err(1, "opening IP socket");
@@ -1366,7 +1479,8 @@ static void setup_tun_net(char *arg)
1366 devices.device_num, tapif, arg); 1479 devices.device_num, tapif, arg);
1367} 1480}
1368 1481
1369/* Our block (disk) device should be really simple: the Guest asks for a block 1482/*
1483 * Our block (disk) device should be really simple: the Guest asks for a block
1370 * number and we read or write that position in the file. Unfortunately, that 1484 * number and we read or write that position in the file. Unfortunately, that
1371 * was amazingly slow: the Guest waits until the read is finished before 1485 * was amazingly slow: the Guest waits until the read is finished before
1372 * running anything else, even if it could have been doing useful work. 1486 * running anything else, even if it could have been doing useful work.
@@ -1374,7 +1488,9 @@ static void setup_tun_net(char *arg)
1374 * We could use async I/O, except it's reputed to suck so hard that characters 1488 * We could use async I/O, except it's reputed to suck so hard that characters
1375 * actually go missing from your code when you try to use it. 1489 * actually go missing from your code when you try to use it.
1376 * 1490 *
1377 * So we farm the I/O out to thread, and communicate with it via a pipe. */ 1491 * So this was one reason why lguest now does all virtqueue servicing in
1492 * separate threads: it's more efficient and more like a real device.
1493 */
1378 1494
1379/* This hangs off device->priv. */ 1495/* This hangs off device->priv. */
1380struct vblk_info 1496struct vblk_info
@@ -1412,9 +1528,11 @@ static void blk_request(struct virtqueue *vq)
1412 /* Get the next request. */ 1528 /* Get the next request. */
1413 head = wait_for_vq_desc(vq, iov, &out_num, &in_num); 1529 head = wait_for_vq_desc(vq, iov, &out_num, &in_num);
1414 1530
1415 /* Every block request should contain at least one output buffer 1531 /*
1532 * Every block request should contain at least one output buffer
1416 * (detailing the location on disk and the type of request) and one 1533 * (detailing the location on disk and the type of request) and one
1417 * input buffer (to hold the result). */ 1534 * input buffer (to hold the result).
1535 */
1418 if (out_num == 0 || in_num == 0) 1536 if (out_num == 0 || in_num == 0)
1419 errx(1, "Bad virtblk cmd %u out=%u in=%u", 1537 errx(1, "Bad virtblk cmd %u out=%u in=%u",
1420 head, out_num, in_num); 1538 head, out_num, in_num);
@@ -1423,33 +1541,41 @@ static void blk_request(struct virtqueue *vq)
1423 in = convert(&iov[out_num+in_num-1], u8); 1541 in = convert(&iov[out_num+in_num-1], u8);
1424 off = out->sector * 512; 1542 off = out->sector * 512;
1425 1543
1426 /* The block device implements "barriers", where the Guest indicates 1544 /*
1545 * The block device implements "barriers", where the Guest indicates
1427 * that it wants all previous writes to occur before this write. We 1546 * that it wants all previous writes to occur before this write. We
1428 * don't have a way of asking our kernel to do a barrier, so we just 1547 * don't have a way of asking our kernel to do a barrier, so we just
1429 * synchronize all the data in the file. Pretty poor, no? */ 1548 * synchronize all the data in the file. Pretty poor, no?
1549 */
1430 if (out->type & VIRTIO_BLK_T_BARRIER) 1550 if (out->type & VIRTIO_BLK_T_BARRIER)
1431 fdatasync(vblk->fd); 1551 fdatasync(vblk->fd);
1432 1552
1433 /* In general the virtio block driver is allowed to try SCSI commands. 1553 /*
1434 * It'd be nice if we supported eject, for example, but we don't. */ 1554 * In general the virtio block driver is allowed to try SCSI commands.
1555 * It'd be nice if we supported eject, for example, but we don't.
1556 */
1435 if (out->type & VIRTIO_BLK_T_SCSI_CMD) { 1557 if (out->type & VIRTIO_BLK_T_SCSI_CMD) {
1436 fprintf(stderr, "Scsi commands unsupported\n"); 1558 fprintf(stderr, "Scsi commands unsupported\n");
1437 *in = VIRTIO_BLK_S_UNSUPP; 1559 *in = VIRTIO_BLK_S_UNSUPP;
1438 wlen = sizeof(*in); 1560 wlen = sizeof(*in);
1439 } else if (out->type & VIRTIO_BLK_T_OUT) { 1561 } else if (out->type & VIRTIO_BLK_T_OUT) {
1440 /* Write */ 1562 /*
1441 1563 * Write
1442 /* Move to the right location in the block file. This can fail 1564 *
1443 * if they try to write past end. */ 1565 * Move to the right location in the block file. This can fail
1566 * if they try to write past end.
1567 */
1444 if (lseek64(vblk->fd, off, SEEK_SET) != off) 1568 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1445 err(1, "Bad seek to sector %llu", out->sector); 1569 err(1, "Bad seek to sector %llu", out->sector);
1446 1570
1447 ret = writev(vblk->fd, iov+1, out_num-1); 1571 ret = writev(vblk->fd, iov+1, out_num-1);
1448 verbose("WRITE to sector %llu: %i\n", out->sector, ret); 1572 verbose("WRITE to sector %llu: %i\n", out->sector, ret);
1449 1573
1450 /* Grr... Now we know how long the descriptor they sent was, we 1574 /*
1575 * Grr... Now we know how long the descriptor they sent was, we
1451 * make sure they didn't try to write over the end of the block 1576 * make sure they didn't try to write over the end of the block
1452 * file (possibly extending it). */ 1577 * file (possibly extending it).
1578 */
1453 if (ret > 0 && off + ret > vblk->len) { 1579 if (ret > 0 && off + ret > vblk->len) {
1454 /* Trim it back to the correct length */ 1580 /* Trim it back to the correct length */
1455 ftruncate64(vblk->fd, vblk->len); 1581 ftruncate64(vblk->fd, vblk->len);
@@ -1459,10 +1585,12 @@ static void blk_request(struct virtqueue *vq)
1459 wlen = sizeof(*in); 1585 wlen = sizeof(*in);
1460 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR); 1586 *in = (ret >= 0 ? VIRTIO_BLK_S_OK : VIRTIO_BLK_S_IOERR);
1461 } else { 1587 } else {
1462 /* Read */ 1588 /*
1463 1589 * Read
1464 /* Move to the right location in the block file. This can fail 1590 *
1465 * if they try to read past end. */ 1591 * Move to the right location in the block file. This can fail
1592 * if they try to read past end.
1593 */
1466 if (lseek64(vblk->fd, off, SEEK_SET) != off) 1594 if (lseek64(vblk->fd, off, SEEK_SET) != off)
1467 err(1, "Bad seek to sector %llu", out->sector); 1595 err(1, "Bad seek to sector %llu", out->sector);
1468 1596
@@ -1477,10 +1605,12 @@ static void blk_request(struct virtqueue *vq)
1477 } 1605 }
1478 } 1606 }
1479 1607
1480 /* OK, so we noted that it was pretty poor to use an fdatasync as a 1608 /*
1609 * OK, so we noted that it was pretty poor to use an fdatasync as a
1481 * barrier. But Christoph Hellwig points out that we need a sync 1610 * barrier. But Christoph Hellwig points out that we need a sync
1482 * *afterwards* as well: "Barriers specify no reordering to the front 1611 * *afterwards* as well: "Barriers specify no reordering to the front
1483 * or the back." And Jens Axboe confirmed it, so here we are: */ 1612 * or the back." And Jens Axboe confirmed it, so here we are:
1613 */
1484 if (out->type & VIRTIO_BLK_T_BARRIER) 1614 if (out->type & VIRTIO_BLK_T_BARRIER)
1485 fdatasync(vblk->fd); 1615 fdatasync(vblk->fd);
1486 1616
@@ -1494,7 +1624,7 @@ static void setup_block_file(const char *filename)
1494 struct vblk_info *vblk; 1624 struct vblk_info *vblk;
1495 struct virtio_blk_config conf; 1625 struct virtio_blk_config conf;
1496 1626
1497 /* The device responds to return from I/O thread. */ 1627 /* Creat the device. */
1498 dev = new_device("block", VIRTIO_ID_BLOCK); 1628 dev = new_device("block", VIRTIO_ID_BLOCK);
1499 1629
1500 /* The device has one virtqueue, where the Guest places requests. */ 1630 /* The device has one virtqueue, where the Guest places requests. */
@@ -1513,8 +1643,10 @@ static void setup_block_file(const char *filename)
1513 /* Tell Guest how many sectors this device has. */ 1643 /* Tell Guest how many sectors this device has. */
1514 conf.capacity = cpu_to_le64(vblk->len / 512); 1644 conf.capacity = cpu_to_le64(vblk->len / 512);
1515 1645
1516 /* Tell Guest not to put in too many descriptors at once: two are used 1646 /*
1517 * for the in and out elements. */ 1647 * Tell Guest not to put in too many descriptors at once: two are used
1648 * for the in and out elements.
1649 */
1518 add_feature(dev, VIRTIO_BLK_F_SEG_MAX); 1650 add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1519 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2); 1651 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1520 1652
@@ -1525,16 +1657,18 @@ static void setup_block_file(const char *filename)
1525 ++devices.device_num, le64_to_cpu(conf.capacity)); 1657 ++devices.device_num, le64_to_cpu(conf.capacity));
1526} 1658}
1527 1659
1528struct rng_info { 1660/*L:211
1529 int rfd; 1661 * Our random number generator device reads from /dev/random into the Guest's
1530};
1531
1532/* Our random number generator device reads from /dev/random into the Guest's
1533 * input buffers. The usual case is that the Guest doesn't want random numbers 1662 * input buffers. The usual case is that the Guest doesn't want random numbers
1534 * and so has no buffers although /dev/random is still readable, whereas 1663 * and so has no buffers although /dev/random is still readable, whereas
1535 * console is the reverse. 1664 * console is the reverse.
1536 * 1665 *
1537 * The same logic applies, however. */ 1666 * The same logic applies, however.
1667 */
1668struct rng_info {
1669 int rfd;
1670};
1671
1538static void rng_input(struct virtqueue *vq) 1672static void rng_input(struct virtqueue *vq)
1539{ 1673{
1540 int len; 1674 int len;
@@ -1547,9 +1681,11 @@ static void rng_input(struct virtqueue *vq)
1547 if (out_num) 1681 if (out_num)
1548 errx(1, "Output buffers in rng?"); 1682 errx(1, "Output buffers in rng?");
1549 1683
1550 /* This is why we convert to iovecs: the readv() call uses them, and so 1684 /*
1685 * This is why we convert to iovecs: the readv() call uses them, and so
1551 * it reads straight into the Guest's buffer. We loop to make sure we 1686 * it reads straight into the Guest's buffer. We loop to make sure we
1552 * fill it. */ 1687 * fill it.
1688 */
1553 while (!iov_empty(iov, in_num)) { 1689 while (!iov_empty(iov, in_num)) {
1554 len = readv(rng_info->rfd, iov, in_num); 1690 len = readv(rng_info->rfd, iov, in_num);
1555 if (len <= 0) 1691 if (len <= 0)
@@ -1562,15 +1698,18 @@ static void rng_input(struct virtqueue *vq)
1562 add_used(vq, head, totlen); 1698 add_used(vq, head, totlen);
1563} 1699}
1564 1700
1565/* And this creates a "hardware" random number device for the Guest. */ 1701/*L:199
1702 * This creates a "hardware" random number device for the Guest.
1703 */
1566static void setup_rng(void) 1704static void setup_rng(void)
1567{ 1705{
1568 struct device *dev; 1706 struct device *dev;
1569 struct rng_info *rng_info = malloc(sizeof(*rng_info)); 1707 struct rng_info *rng_info = malloc(sizeof(*rng_info));
1570 1708
1709 /* Our device's privat info simply contains the /dev/random fd. */
1571 rng_info->rfd = open_or_die("/dev/random", O_RDONLY); 1710 rng_info->rfd = open_or_die("/dev/random", O_RDONLY);
1572 1711
1573 /* The device responds to return from I/O thread. */ 1712 /* Create the new device. */
1574 dev = new_device("rng", VIRTIO_ID_RNG); 1713 dev = new_device("rng", VIRTIO_ID_RNG);
1575 dev->priv = rng_info; 1714 dev->priv = rng_info;
1576 1715
@@ -1586,8 +1725,10 @@ static void __attribute__((noreturn)) restart_guest(void)
1586{ 1725{
1587 unsigned int i; 1726 unsigned int i;
1588 1727
1589 /* Since we don't track all open fds, we simply close everything beyond 1728 /*
1590 * stderr. */ 1729 * Since we don't track all open fds, we simply close everything beyond
1730 * stderr.
1731 */
1591 for (i = 3; i < FD_SETSIZE; i++) 1732 for (i = 3; i < FD_SETSIZE; i++)
1592 close(i); 1733 close(i);
1593 1734
@@ -1598,8 +1739,10 @@ static void __attribute__((noreturn)) restart_guest(void)
1598 err(1, "Could not exec %s", main_args[0]); 1739 err(1, "Could not exec %s", main_args[0]);
1599} 1740}
1600 1741
1601/*L:220 Finally we reach the core of the Launcher which runs the Guest, serves 1742/*L:220
1602 * its input and output, and finally, lays it to rest. */ 1743 * Finally we reach the core of the Launcher which runs the Guest, serves
1744 * its input and output, and finally, lays it to rest.
1745 */
1603static void __attribute__((noreturn)) run_guest(void) 1746static void __attribute__((noreturn)) run_guest(void)
1604{ 1747{
1605 for (;;) { 1748 for (;;) {
@@ -1634,7 +1777,7 @@ static void __attribute__((noreturn)) run_guest(void)
1634 * 1777 *
1635 * Are you ready? Take a deep breath and join me in the core of the Host, in 1778 * Are you ready? Take a deep breath and join me in the core of the Host, in
1636 * "make Host". 1779 * "make Host".
1637 :*/ 1780:*/
1638 1781
1639static struct option opts[] = { 1782static struct option opts[] = {
1640 { "verbose", 0, NULL, 'v' }, 1783 { "verbose", 0, NULL, 'v' },
@@ -1655,8 +1798,7 @@ static void usage(void)
1655/*L:105 The main routine is where the real work begins: */ 1798/*L:105 The main routine is where the real work begins: */
1656int main(int argc, char *argv[]) 1799int main(int argc, char *argv[])
1657{ 1800{
1658 /* Memory, top-level pagetable, code startpoint and size of the 1801 /* Memory, code startpoint and size of the (optional) initrd. */
1659 * (optional) initrd. */
1660 unsigned long mem = 0, start, initrd_size = 0; 1802 unsigned long mem = 0, start, initrd_size = 0;
1661 /* Two temporaries. */ 1803 /* Two temporaries. */
1662 int i, c; 1804 int i, c;
@@ -1668,24 +1810,30 @@ int main(int argc, char *argv[])
1668 /* Save the args: we "reboot" by execing ourselves again. */ 1810 /* Save the args: we "reboot" by execing ourselves again. */
1669 main_args = argv; 1811 main_args = argv;
1670 1812
1671 /* First we initialize the device list. We keep a pointer to the last 1813 /*
1814 * First we initialize the device list. We keep a pointer to the last
1672 * device, and the next interrupt number to use for devices (1: 1815 * device, and the next interrupt number to use for devices (1:
1673 * remember that 0 is used by the timer). */ 1816 * remember that 0 is used by the timer).
1817 */
1674 devices.lastdev = NULL; 1818 devices.lastdev = NULL;
1675 devices.next_irq = 1; 1819 devices.next_irq = 1;
1676 1820
1677 cpu_id = 0; 1821 cpu_id = 0;
1678 /* We need to know how much memory so we can set up the device 1822 /*
1823 * We need to know how much memory so we can set up the device
1679 * descriptor and memory pages for the devices as we parse the command 1824 * descriptor and memory pages for the devices as we parse the command
1680 * line. So we quickly look through the arguments to find the amount 1825 * line. So we quickly look through the arguments to find the amount
1681 * of memory now. */ 1826 * of memory now.
1827 */
1682 for (i = 1; i < argc; i++) { 1828 for (i = 1; i < argc; i++) {
1683 if (argv[i][0] != '-') { 1829 if (argv[i][0] != '-') {
1684 mem = atoi(argv[i]) * 1024 * 1024; 1830 mem = atoi(argv[i]) * 1024 * 1024;
1685 /* We start by mapping anonymous pages over all of 1831 /*
1832 * We start by mapping anonymous pages over all of
1686 * guest-physical memory range. This fills it with 0, 1833 * guest-physical memory range. This fills it with 0,
1687 * and ensures that the Guest won't be killed when it 1834 * and ensures that the Guest won't be killed when it
1688 * tries to access it. */ 1835 * tries to access it.
1836 */
1689 guest_base = map_zeroed_pages(mem / getpagesize() 1837 guest_base = map_zeroed_pages(mem / getpagesize()
1690 + DEVICE_PAGES); 1838 + DEVICE_PAGES);
1691 guest_limit = mem; 1839 guest_limit = mem;
@@ -1718,8 +1866,10 @@ int main(int argc, char *argv[])
1718 usage(); 1866 usage();
1719 } 1867 }
1720 } 1868 }
1721 /* After the other arguments we expect memory and kernel image name, 1869 /*
1722 * followed by command line arguments for the kernel. */ 1870 * After the other arguments we expect memory and kernel image name,
1871 * followed by command line arguments for the kernel.
1872 */
1723 if (optind + 2 > argc) 1873 if (optind + 2 > argc)
1724 usage(); 1874 usage();
1725 1875
@@ -1737,20 +1887,26 @@ int main(int argc, char *argv[])
1737 /* Map the initrd image if requested (at top of physical memory) */ 1887 /* Map the initrd image if requested (at top of physical memory) */
1738 if (initrd_name) { 1888 if (initrd_name) {
1739 initrd_size = load_initrd(initrd_name, mem); 1889 initrd_size = load_initrd(initrd_name, mem);
1740 /* These are the location in the Linux boot header where the 1890 /*
1741 * start and size of the initrd are expected to be found. */ 1891 * These are the location in the Linux boot header where the
1892 * start and size of the initrd are expected to be found.
1893 */
1742 boot->hdr.ramdisk_image = mem - initrd_size; 1894 boot->hdr.ramdisk_image = mem - initrd_size;
1743 boot->hdr.ramdisk_size = initrd_size; 1895 boot->hdr.ramdisk_size = initrd_size;
1744 /* The bootloader type 0xFF means "unknown"; that's OK. */ 1896 /* The bootloader type 0xFF means "unknown"; that's OK. */
1745 boot->hdr.type_of_loader = 0xFF; 1897 boot->hdr.type_of_loader = 0xFF;
1746 } 1898 }
1747 1899
1748 /* The Linux boot header contains an "E820" memory map: ours is a 1900 /*
1749 * simple, single region. */ 1901 * The Linux boot header contains an "E820" memory map: ours is a
1902 * simple, single region.
1903 */
1750 boot->e820_entries = 1; 1904 boot->e820_entries = 1;
1751 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM }); 1905 boot->e820_map[0] = ((struct e820entry) { 0, mem, E820_RAM });
1752 /* The boot header contains a command line pointer: we put the command 1906 /*
1753 * line after the boot header. */ 1907 * The boot header contains a command line pointer: we put the command
1908 * line after the boot header.
1909 */
1754 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1); 1910 boot->hdr.cmd_line_ptr = to_guest_phys(boot + 1);
1755 /* We use a simple helper to copy the arguments separated by spaces. */ 1911 /* We use a simple helper to copy the arguments separated by spaces. */
1756 concat((char *)(boot + 1), argv+optind+2); 1912 concat((char *)(boot + 1), argv+optind+2);
@@ -1764,8 +1920,10 @@ int main(int argc, char *argv[])
1764 /* Tell the entry path not to try to reload segment registers. */ 1920 /* Tell the entry path not to try to reload segment registers. */
1765 boot->hdr.loadflags |= KEEP_SEGMENTS; 1921 boot->hdr.loadflags |= KEEP_SEGMENTS;
1766 1922
1767 /* We tell the kernel to initialize the Guest: this returns the open 1923 /*
1768 * /dev/lguest file descriptor. */ 1924 * We tell the kernel to initialize the Guest: this returns the open
1925 * /dev/lguest file descriptor.
1926 */
1769 tell_kernel(start); 1927 tell_kernel(start);
1770 1928
1771 /* Ensure that we terminate if a child dies. */ 1929 /* Ensure that we terminate if a child dies. */
diff --git a/arch/x86/include/asm/lguest.h b/arch/x86/include/asm/lguest.h
index 313389cd50d2..5136dad57cbb 100644
--- a/arch/x86/include/asm/lguest.h
+++ b/arch/x86/include/asm/lguest.h
@@ -17,8 +17,7 @@
17/* Pages for switcher itself, then two pages per cpu */ 17/* Pages for switcher itself, then two pages per cpu */
18#define TOTAL_SWITCHER_PAGES (SHARED_SWITCHER_PAGES + 2 * nr_cpu_ids) 18#define TOTAL_SWITCHER_PAGES (SHARED_SWITCHER_PAGES + 2 * nr_cpu_ids)
19 19
20/* We map at -4M (-2M when PAE is activated) for ease of mapping 20/* We map at -4M (-2M for PAE) for ease of mapping (one PTE page). */
21 * into the guest (one PTE page). */
22#ifdef CONFIG_X86_PAE 21#ifdef CONFIG_X86_PAE
23#define SWITCHER_ADDR 0xFFE00000 22#define SWITCHER_ADDR 0xFFE00000
24#else 23#else
diff --git a/arch/x86/include/asm/lguest_hcall.h b/arch/x86/include/asm/lguest_hcall.h
index 33600a66755f..cceb73e12e50 100644
--- a/arch/x86/include/asm/lguest_hcall.h
+++ b/arch/x86/include/asm/lguest_hcall.h
@@ -30,7 +30,8 @@
30#include <asm/hw_irq.h> 30#include <asm/hw_irq.h>
31#include <asm/kvm_para.h> 31#include <asm/kvm_para.h>
32 32
33/*G:030 But first, how does our Guest contact the Host to ask for privileged 33/*G:030
34 * But first, how does our Guest contact the Host to ask for privileged
34 * operations? There are two ways: the direct way is to make a "hypercall", 35 * operations? There are two ways: the direct way is to make a "hypercall",
35 * to make requests of the Host Itself. 36 * to make requests of the Host Itself.
36 * 37 *
@@ -41,16 +42,15 @@
41 * 42 *
42 * Grossly invalid calls result in Sudden Death at the hands of the vengeful 43 * Grossly invalid calls result in Sudden Death at the hands of the vengeful
43 * Host, rather than returning failure. This reflects Winston Churchill's 44 * Host, rather than returning failure. This reflects Winston Churchill's
44 * definition of a gentleman: "someone who is only rude intentionally". */ 45 * definition of a gentleman: "someone who is only rude intentionally".
45/*:*/ 46:*/
46 47
47/* Can't use our min() macro here: needs to be a constant */ 48/* Can't use our min() macro here: needs to be a constant */
48#define LGUEST_IRQS (NR_IRQS < 32 ? NR_IRQS: 32) 49#define LGUEST_IRQS (NR_IRQS < 32 ? NR_IRQS: 32)
49 50
50#define LHCALL_RING_SIZE 64 51#define LHCALL_RING_SIZE 64
51struct hcall_args { 52struct hcall_args {
52 /* These map directly onto eax, ebx, ecx, edx and esi 53 /* These map directly onto eax/ebx/ecx/edx/esi in struct lguest_regs */
53 * in struct lguest_regs */
54 unsigned long arg0, arg1, arg2, arg3, arg4; 54 unsigned long arg0, arg1, arg2, arg3, arg4;
55}; 55};
56 56
diff --git a/arch/x86/lguest/boot.c b/arch/x86/lguest/boot.c
index f2bf1f73d468..025c04d18f2b 100644
--- a/arch/x86/lguest/boot.c
+++ b/arch/x86/lguest/boot.c
@@ -22,7 +22,8 @@
22 * 22 *
23 * So how does the kernel know it's a Guest? We'll see that later, but let's 23 * So how does the kernel know it's a Guest? We'll see that later, but let's
24 * just say that we end up here where we replace the native functions various 24 * just say that we end up here where we replace the native functions various
25 * "paravirt" structures with our Guest versions, then boot like normal. :*/ 25 * "paravirt" structures with our Guest versions, then boot like normal.
26:*/
26 27
27/* 28/*
28 * Copyright (C) 2006, Rusty Russell <rusty@rustcorp.com.au> IBM Corporation. 29 * Copyright (C) 2006, Rusty Russell <rusty@rustcorp.com.au> IBM Corporation.
@@ -74,7 +75,8 @@
74 * 75 *
75 * The Guest in our tale is a simple creature: identical to the Host but 76 * The Guest in our tale is a simple creature: identical to the Host but
76 * behaving in simplified but equivalent ways. In particular, the Guest is the 77 * behaving in simplified but equivalent ways. In particular, the Guest is the
77 * same kernel as the Host (or at least, built from the same source code). :*/ 78 * same kernel as the Host (or at least, built from the same source code).
79:*/
78 80
79struct lguest_data lguest_data = { 81struct lguest_data lguest_data = {
80 .hcall_status = { [0 ... LHCALL_RING_SIZE-1] = 0xFF }, 82 .hcall_status = { [0 ... LHCALL_RING_SIZE-1] = 0xFF },
@@ -85,7 +87,8 @@ struct lguest_data lguest_data = {
85 .syscall_vec = SYSCALL_VECTOR, 87 .syscall_vec = SYSCALL_VECTOR,
86}; 88};
87 89
88/*G:037 async_hcall() is pretty simple: I'm quite proud of it really. We have a 90/*G:037
91 * async_hcall() is pretty simple: I'm quite proud of it really. We have a
89 * ring buffer of stored hypercalls which the Host will run though next time we 92 * ring buffer of stored hypercalls which the Host will run though next time we
90 * do a normal hypercall. Each entry in the ring has 5 slots for the hypercall 93 * do a normal hypercall. Each entry in the ring has 5 slots for the hypercall
91 * arguments, and a "hcall_status" word which is 0 if the call is ready to go, 94 * arguments, and a "hcall_status" word which is 0 if the call is ready to go,
@@ -94,7 +97,8 @@ struct lguest_data lguest_data = {
94 * If we come around to a slot which hasn't been finished, then the table is 97 * If we come around to a slot which hasn't been finished, then the table is
95 * full and we just make the hypercall directly. This has the nice side 98 * full and we just make the hypercall directly. This has the nice side
96 * effect of causing the Host to run all the stored calls in the ring buffer 99 * effect of causing the Host to run all the stored calls in the ring buffer
97 * which empties it for next time! */ 100 * which empties it for next time!
101 */
98static void async_hcall(unsigned long call, unsigned long arg1, 102static void async_hcall(unsigned long call, unsigned long arg1,
99 unsigned long arg2, unsigned long arg3, 103 unsigned long arg2, unsigned long arg3,
100 unsigned long arg4) 104 unsigned long arg4)
@@ -103,9 +107,11 @@ static void async_hcall(unsigned long call, unsigned long arg1,
103 static unsigned int next_call; 107 static unsigned int next_call;
104 unsigned long flags; 108 unsigned long flags;
105 109
106 /* Disable interrupts if not already disabled: we don't want an 110 /*
111 * Disable interrupts if not already disabled: we don't want an
107 * interrupt handler making a hypercall while we're already doing 112 * interrupt handler making a hypercall while we're already doing
108 * one! */ 113 * one!
114 */
109 local_irq_save(flags); 115 local_irq_save(flags);
110 if (lguest_data.hcall_status[next_call] != 0xFF) { 116 if (lguest_data.hcall_status[next_call] != 0xFF) {
111 /* Table full, so do normal hcall which will flush table. */ 117 /* Table full, so do normal hcall which will flush table. */
@@ -125,8 +131,9 @@ static void async_hcall(unsigned long call, unsigned long arg1,
125 local_irq_restore(flags); 131 local_irq_restore(flags);
126} 132}
127 133
128/*G:035 Notice the lazy_hcall() above, rather than hcall(). This is our first 134/*G:035
129 * real optimization trick! 135 * Notice the lazy_hcall() above, rather than hcall(). This is our first real
136 * optimization trick!
130 * 137 *
131 * When lazy_mode is set, it means we're allowed to defer all hypercalls and do 138 * When lazy_mode is set, it means we're allowed to defer all hypercalls and do
132 * them as a batch when lazy_mode is eventually turned off. Because hypercalls 139 * them as a batch when lazy_mode is eventually turned off. Because hypercalls
@@ -136,7 +143,8 @@ static void async_hcall(unsigned long call, unsigned long arg1,
136 * lguest_leave_lazy_mode(). 143 * lguest_leave_lazy_mode().
137 * 144 *
138 * So, when we're in lazy mode, we call async_hcall() to store the call for 145 * So, when we're in lazy mode, we call async_hcall() to store the call for
139 * future processing: */ 146 * future processing:
147 */
140static void lazy_hcall1(unsigned long call, 148static void lazy_hcall1(unsigned long call,
141 unsigned long arg1) 149 unsigned long arg1)
142{ 150{
@@ -208,9 +216,11 @@ static void lguest_end_context_switch(struct task_struct *next)
208 * check there before it tries to deliver an interrupt. 216 * check there before it tries to deliver an interrupt.
209 */ 217 */
210 218
211/* save_flags() is expected to return the processor state (ie. "flags"). The 219/*
220 * save_flags() is expected to return the processor state (ie. "flags"). The
212 * flags word contains all kind of stuff, but in practice Linux only cares 221 * flags word contains all kind of stuff, but in practice Linux only cares
213 * about the interrupt flag. Our "save_flags()" just returns that. */ 222 * about the interrupt flag. Our "save_flags()" just returns that.
223 */
214static unsigned long save_fl(void) 224static unsigned long save_fl(void)
215{ 225{
216 return lguest_data.irq_enabled; 226 return lguest_data.irq_enabled;
@@ -222,13 +232,15 @@ static void irq_disable(void)
222 lguest_data.irq_enabled = 0; 232 lguest_data.irq_enabled = 0;
223} 233}
224 234
225/* Let's pause a moment. Remember how I said these are called so often? 235/*
236 * Let's pause a moment. Remember how I said these are called so often?
226 * Jeremy Fitzhardinge optimized them so hard early in 2009 that he had to 237 * Jeremy Fitzhardinge optimized them so hard early in 2009 that he had to
227 * break some rules. In particular, these functions are assumed to save their 238 * break some rules. In particular, these functions are assumed to save their
228 * own registers if they need to: normal C functions assume they can trash the 239 * own registers if they need to: normal C functions assume they can trash the
229 * eax register. To use normal C functions, we use 240 * eax register. To use normal C functions, we use
230 * PV_CALLEE_SAVE_REGS_THUNK(), which pushes %eax onto the stack, calls the 241 * PV_CALLEE_SAVE_REGS_THUNK(), which pushes %eax onto the stack, calls the
231 * C function, then restores it. */ 242 * C function, then restores it.
243 */
232PV_CALLEE_SAVE_REGS_THUNK(save_fl); 244PV_CALLEE_SAVE_REGS_THUNK(save_fl);
233PV_CALLEE_SAVE_REGS_THUNK(irq_disable); 245PV_CALLEE_SAVE_REGS_THUNK(irq_disable);
234/*:*/ 246/*:*/
@@ -237,18 +249,20 @@ PV_CALLEE_SAVE_REGS_THUNK(irq_disable);
237extern void lg_irq_enable(void); 249extern void lg_irq_enable(void);
238extern void lg_restore_fl(unsigned long flags); 250extern void lg_restore_fl(unsigned long flags);
239 251
240/*M:003 Note that we don't check for outstanding interrupts when we re-enable 252/*M:003
241 * them (or when we unmask an interrupt). This seems to work for the moment, 253 * Note that we don't check for outstanding interrupts when we re-enable them
242 * since interrupts are rare and we'll just get the interrupt on the next timer 254 * (or when we unmask an interrupt). This seems to work for the moment, since
243 * tick, but now we can run with CONFIG_NO_HZ, we should revisit this. One way 255 * interrupts are rare and we'll just get the interrupt on the next timer tick,
244 * would be to put the "irq_enabled" field in a page by itself, and have the 256 * but now we can run with CONFIG_NO_HZ, we should revisit this. One way would
245 * Host write-protect it when an interrupt comes in when irqs are disabled. 257 * be to put the "irq_enabled" field in a page by itself, and have the Host
246 * There will then be a page fault as soon as interrupts are re-enabled. 258 * write-protect it when an interrupt comes in when irqs are disabled. There
259 * will then be a page fault as soon as interrupts are re-enabled.
247 * 260 *
248 * A better method is to implement soft interrupt disable generally for x86: 261 * A better method is to implement soft interrupt disable generally for x86:
249 * instead of disabling interrupts, we set a flag. If an interrupt does come 262 * instead of disabling interrupts, we set a flag. If an interrupt does come
250 * in, we then disable them for real. This is uncommon, so we could simply use 263 * in, we then disable them for real. This is uncommon, so we could simply use
251 * a hypercall for interrupt control and not worry about efficiency. :*/ 264 * a hypercall for interrupt control and not worry about efficiency.
265:*/
252 266
253/*G:034 267/*G:034
254 * The Interrupt Descriptor Table (IDT). 268 * The Interrupt Descriptor Table (IDT).
@@ -261,10 +275,12 @@ extern void lg_restore_fl(unsigned long flags);
261static void lguest_write_idt_entry(gate_desc *dt, 275static void lguest_write_idt_entry(gate_desc *dt,
262 int entrynum, const gate_desc *g) 276 int entrynum, const gate_desc *g)
263{ 277{
264 /* The gate_desc structure is 8 bytes long: we hand it to the Host in 278 /*
279 * The gate_desc structure is 8 bytes long: we hand it to the Host in
265 * two 32-bit chunks. The whole 32-bit kernel used to hand descriptors 280 * two 32-bit chunks. The whole 32-bit kernel used to hand descriptors
266 * around like this; typesafety wasn't a big concern in Linux's early 281 * around like this; typesafety wasn't a big concern in Linux's early
267 * years. */ 282 * years.
283 */
268 u32 *desc = (u32 *)g; 284 u32 *desc = (u32 *)g;
269 /* Keep the local copy up to date. */ 285 /* Keep the local copy up to date. */
270 native_write_idt_entry(dt, entrynum, g); 286 native_write_idt_entry(dt, entrynum, g);
@@ -272,9 +288,11 @@ static void lguest_write_idt_entry(gate_desc *dt,
272 kvm_hypercall3(LHCALL_LOAD_IDT_ENTRY, entrynum, desc[0], desc[1]); 288 kvm_hypercall3(LHCALL_LOAD_IDT_ENTRY, entrynum, desc[0], desc[1]);
273} 289}
274 290
275/* Changing to a different IDT is very rare: we keep the IDT up-to-date every 291/*
292 * Changing to a different IDT is very rare: we keep the IDT up-to-date every
276 * time it is written, so we can simply loop through all entries and tell the 293 * time it is written, so we can simply loop through all entries and tell the
277 * Host about them. */ 294 * Host about them.
295 */
278static void lguest_load_idt(const struct desc_ptr *desc) 296static void lguest_load_idt(const struct desc_ptr *desc)
279{ 297{
280 unsigned int i; 298 unsigned int i;
@@ -305,9 +323,11 @@ static void lguest_load_gdt(const struct desc_ptr *desc)
305 kvm_hypercall3(LHCALL_LOAD_GDT_ENTRY, i, gdt[i].a, gdt[i].b); 323 kvm_hypercall3(LHCALL_LOAD_GDT_ENTRY, i, gdt[i].a, gdt[i].b);
306} 324}
307 325
308/* For a single GDT entry which changes, we do the lazy thing: alter our GDT, 326/*
327 * For a single GDT entry which changes, we do the lazy thing: alter our GDT,
309 * then tell the Host to reload the entire thing. This operation is so rare 328 * then tell the Host to reload the entire thing. This operation is so rare
310 * that this naive implementation is reasonable. */ 329 * that this naive implementation is reasonable.
330 */
311static void lguest_write_gdt_entry(struct desc_struct *dt, int entrynum, 331static void lguest_write_gdt_entry(struct desc_struct *dt, int entrynum,
312 const void *desc, int type) 332 const void *desc, int type)
313{ 333{
@@ -317,29 +337,36 @@ static void lguest_write_gdt_entry(struct desc_struct *dt, int entrynum,
317 dt[entrynum].a, dt[entrynum].b); 337 dt[entrynum].a, dt[entrynum].b);
318} 338}
319 339
320/* OK, I lied. There are three "thread local storage" GDT entries which change 340/*
341 * OK, I lied. There are three "thread local storage" GDT entries which change
321 * on every context switch (these three entries are how glibc implements 342 * on every context switch (these three entries are how glibc implements
322 * __thread variables). So we have a hypercall specifically for this case. */ 343 * __thread variables). So we have a hypercall specifically for this case.
344 */
323static void lguest_load_tls(struct thread_struct *t, unsigned int cpu) 345static void lguest_load_tls(struct thread_struct *t, unsigned int cpu)
324{ 346{
325 /* There's one problem which normal hardware doesn't have: the Host 347 /*
348 * There's one problem which normal hardware doesn't have: the Host
326 * can't handle us removing entries we're currently using. So we clear 349 * can't handle us removing entries we're currently using. So we clear
327 * the GS register here: if it's needed it'll be reloaded anyway. */ 350 * the GS register here: if it's needed it'll be reloaded anyway.
351 */
328 lazy_load_gs(0); 352 lazy_load_gs(0);
329 lazy_hcall2(LHCALL_LOAD_TLS, __pa(&t->tls_array), cpu); 353 lazy_hcall2(LHCALL_LOAD_TLS, __pa(&t->tls_array), cpu);
330} 354}
331 355
332/*G:038 That's enough excitement for now, back to ploughing through each of 356/*G:038
333 * the different pv_ops structures (we're about 1/3 of the way through). 357 * That's enough excitement for now, back to ploughing through each of the
358 * different pv_ops structures (we're about 1/3 of the way through).
334 * 359 *
335 * This is the Local Descriptor Table, another weird Intel thingy. Linux only 360 * This is the Local Descriptor Table, another weird Intel thingy. Linux only
336 * uses this for some strange applications like Wine. We don't do anything 361 * uses this for some strange applications like Wine. We don't do anything
337 * here, so they'll get an informative and friendly Segmentation Fault. */ 362 * here, so they'll get an informative and friendly Segmentation Fault.
363 */
338static void lguest_set_ldt(const void *addr, unsigned entries) 364static void lguest_set_ldt(const void *addr, unsigned entries)
339{ 365{
340} 366}
341 367
342/* This loads a GDT entry into the "Task Register": that entry points to a 368/*
369 * This loads a GDT entry into the "Task Register": that entry points to a
343 * structure called the Task State Segment. Some comments scattered though the 370 * structure called the Task State Segment. Some comments scattered though the
344 * kernel code indicate that this used for task switching in ages past, along 371 * kernel code indicate that this used for task switching in ages past, along
345 * with blood sacrifice and astrology. 372 * with blood sacrifice and astrology.
@@ -347,19 +374,21 @@ static void lguest_set_ldt(const void *addr, unsigned entries)
347 * Now there's nothing interesting in here that we don't get told elsewhere. 374 * Now there's nothing interesting in here that we don't get told elsewhere.
348 * But the native version uses the "ltr" instruction, which makes the Host 375 * But the native version uses the "ltr" instruction, which makes the Host
349 * complain to the Guest about a Segmentation Fault and it'll oops. So we 376 * complain to the Guest about a Segmentation Fault and it'll oops. So we
350 * override the native version with a do-nothing version. */ 377 * override the native version with a do-nothing version.
378 */
351static void lguest_load_tr_desc(void) 379static void lguest_load_tr_desc(void)
352{ 380{
353} 381}
354 382
355/* The "cpuid" instruction is a way of querying both the CPU identity 383/*
384 * The "cpuid" instruction is a way of querying both the CPU identity
356 * (manufacturer, model, etc) and its features. It was introduced before the 385 * (manufacturer, model, etc) and its features. It was introduced before the
357 * Pentium in 1993 and keeps getting extended by both Intel, AMD and others. 386 * Pentium in 1993 and keeps getting extended by both Intel, AMD and others.
358 * As you might imagine, after a decade and a half this treatment, it is now a 387 * As you might imagine, after a decade and a half this treatment, it is now a
359 * giant ball of hair. Its entry in the current Intel manual runs to 28 pages. 388 * giant ball of hair. Its entry in the current Intel manual runs to 28 pages.
360 * 389 *
361 * This instruction even it has its own Wikipedia entry. The Wikipedia entry 390 * This instruction even it has its own Wikipedia entry. The Wikipedia entry
362 * has been translated into 4 languages. I am not making this up! 391 * has been translated into 5 languages. I am not making this up!
363 * 392 *
364 * We could get funky here and identify ourselves as "GenuineLguest", but 393 * We could get funky here and identify ourselves as "GenuineLguest", but
365 * instead we just use the real "cpuid" instruction. Then I pretty much turned 394 * instead we just use the real "cpuid" instruction. Then I pretty much turned
@@ -371,7 +400,8 @@ static void lguest_load_tr_desc(void)
371 * Replacing the cpuid so we can turn features off is great for the kernel, but 400 * Replacing the cpuid so we can turn features off is great for the kernel, but
372 * anyone (including userspace) can just use the raw "cpuid" instruction and 401 * anyone (including userspace) can just use the raw "cpuid" instruction and
373 * the Host won't even notice since it isn't privileged. So we try not to get 402 * the Host won't even notice since it isn't privileged. So we try not to get
374 * too worked up about it. */ 403 * too worked up about it.
404 */
375static void lguest_cpuid(unsigned int *ax, unsigned int *bx, 405static void lguest_cpuid(unsigned int *ax, unsigned int *bx,
376 unsigned int *cx, unsigned int *dx) 406 unsigned int *cx, unsigned int *dx)
377{ 407{
@@ -379,43 +409,63 @@ static void lguest_cpuid(unsigned int *ax, unsigned int *bx,
379 409
380 native_cpuid(ax, bx, cx, dx); 410 native_cpuid(ax, bx, cx, dx);
381 switch (function) { 411 switch (function) {
382 case 0: /* ID and highest CPUID. Futureproof a little by sticking to 412 /*
383 * older ones. */ 413 * CPUID 0 gives the highest legal CPUID number (and the ID string).
414 * We futureproof our code a little by sticking to known CPUID values.
415 */
416 case 0:
384 if (*ax > 5) 417 if (*ax > 5)
385 *ax = 5; 418 *ax = 5;
386 break; 419 break;
387 case 1: /* Basic feature request. */ 420
388 /* We only allow kernel to see SSE3, CMPXCHG16B and SSSE3 */ 421 /*
422 * CPUID 1 is a basic feature request.
423 *
424 * CX: we only allow kernel to see SSE3, CMPXCHG16B and SSSE3
425 * DX: SSE, SSE2, FXSR, MMX, CMOV, CMPXCHG8B, TSC, FPU and PAE.
426 */
427 case 1:
389 *cx &= 0x00002201; 428 *cx &= 0x00002201;
390 /* SSE, SSE2, FXSR, MMX, CMOV, CMPXCHG8B, TSC, FPU, PAE. */
391 *dx &= 0x07808151; 429 *dx &= 0x07808151;
392 /* The Host can do a nice optimization if it knows that the 430 /*
431 * The Host can do a nice optimization if it knows that the
393 * kernel mappings (addresses above 0xC0000000 or whatever 432 * kernel mappings (addresses above 0xC0000000 or whatever
394 * PAGE_OFFSET is set to) haven't changed. But Linux calls 433 * PAGE_OFFSET is set to) haven't changed. But Linux calls
395 * flush_tlb_user() for both user and kernel mappings unless 434 * flush_tlb_user() for both user and kernel mappings unless
396 * the Page Global Enable (PGE) feature bit is set. */ 435 * the Page Global Enable (PGE) feature bit is set.
436 */
397 *dx |= 0x00002000; 437 *dx |= 0x00002000;
398 /* We also lie, and say we're family id 5. 6 or greater 438 /*
439 * We also lie, and say we're family id 5. 6 or greater
399 * leads to a rdmsr in early_init_intel which we can't handle. 440 * leads to a rdmsr in early_init_intel which we can't handle.
400 * Family ID is returned as bits 8-12 in ax. */ 441 * Family ID is returned as bits 8-12 in ax.
442 */
401 *ax &= 0xFFFFF0FF; 443 *ax &= 0xFFFFF0FF;
402 *ax |= 0x00000500; 444 *ax |= 0x00000500;
403 break; 445 break;
446 /*
447 * 0x80000000 returns the highest Extended Function, so we futureproof
448 * like we do above by limiting it to known fields.
449 */
404 case 0x80000000: 450 case 0x80000000:
405 /* Futureproof this a little: if they ask how much extended
406 * processor information there is, limit it to known fields. */
407 if (*ax > 0x80000008) 451 if (*ax > 0x80000008)
408 *ax = 0x80000008; 452 *ax = 0x80000008;
409 break; 453 break;
454
455 /*
456 * PAE systems can mark pages as non-executable. Linux calls this the
457 * NX bit. Intel calls it XD (eXecute Disable), AMD EVP (Enhanced
458 * Virus Protection). We just switch turn if off here, since we don't
459 * support it.
460 */
410 case 0x80000001: 461 case 0x80000001:
411 /* Here we should fix nx cap depending on host. */
412 /* For this version of PAE, we just clear NX bit. */
413 *dx &= ~(1 << 20); 462 *dx &= ~(1 << 20);
414 break; 463 break;
415 } 464 }
416} 465}
417 466
418/* Intel has four control registers, imaginatively named cr0, cr2, cr3 and cr4. 467/*
468 * Intel has four control registers, imaginatively named cr0, cr2, cr3 and cr4.
419 * I assume there's a cr1, but it hasn't bothered us yet, so we'll not bother 469 * I assume there's a cr1, but it hasn't bothered us yet, so we'll not bother
420 * it. The Host needs to know when the Guest wants to change them, so we have 470 * it. The Host needs to know when the Guest wants to change them, so we have
421 * a whole series of functions like read_cr0() and write_cr0(). 471 * a whole series of functions like read_cr0() and write_cr0().
@@ -430,7 +480,8 @@ static void lguest_cpuid(unsigned int *ax, unsigned int *bx,
430 * name like "FPUTRAP bit" be a little less cryptic? 480 * name like "FPUTRAP bit" be a little less cryptic?
431 * 481 *
432 * We store cr0 locally because the Host never changes it. The Guest sometimes 482 * We store cr0 locally because the Host never changes it. The Guest sometimes
433 * wants to read it and we'd prefer not to bother the Host unnecessarily. */ 483 * wants to read it and we'd prefer not to bother the Host unnecessarily.
484 */
434static unsigned long current_cr0; 485static unsigned long current_cr0;
435static void lguest_write_cr0(unsigned long val) 486static void lguest_write_cr0(unsigned long val)
436{ 487{
@@ -443,18 +494,22 @@ static unsigned long lguest_read_cr0(void)
443 return current_cr0; 494 return current_cr0;
444} 495}
445 496
446/* Intel provided a special instruction to clear the TS bit for people too cool 497/*
498 * Intel provided a special instruction to clear the TS bit for people too cool
447 * to use write_cr0() to do it. This "clts" instruction is faster, because all 499 * to use write_cr0() to do it. This "clts" instruction is faster, because all
448 * the vowels have been optimized out. */ 500 * the vowels have been optimized out.
501 */
449static void lguest_clts(void) 502static void lguest_clts(void)
450{ 503{
451 lazy_hcall1(LHCALL_TS, 0); 504 lazy_hcall1(LHCALL_TS, 0);
452 current_cr0 &= ~X86_CR0_TS; 505 current_cr0 &= ~X86_CR0_TS;
453} 506}
454 507
455/* cr2 is the virtual address of the last page fault, which the Guest only ever 508/*
509 * cr2 is the virtual address of the last page fault, which the Guest only ever
456 * reads. The Host kindly writes this into our "struct lguest_data", so we 510 * reads. The Host kindly writes this into our "struct lguest_data", so we
457 * just read it out of there. */ 511 * just read it out of there.
512 */
458static unsigned long lguest_read_cr2(void) 513static unsigned long lguest_read_cr2(void)
459{ 514{
460 return lguest_data.cr2; 515 return lguest_data.cr2;
@@ -463,10 +518,12 @@ static unsigned long lguest_read_cr2(void)
463/* See lguest_set_pte() below. */ 518/* See lguest_set_pte() below. */
464static bool cr3_changed = false; 519static bool cr3_changed = false;
465 520
466/* cr3 is the current toplevel pagetable page: the principle is the same as 521/*
522 * cr3 is the current toplevel pagetable page: the principle is the same as
467 * cr0. Keep a local copy, and tell the Host when it changes. The only 523 * cr0. Keep a local copy, and tell the Host when it changes. The only
468 * difference is that our local copy is in lguest_data because the Host needs 524 * difference is that our local copy is in lguest_data because the Host needs
469 * to set it upon our initial hypercall. */ 525 * to set it upon our initial hypercall.
526 */
470static void lguest_write_cr3(unsigned long cr3) 527static void lguest_write_cr3(unsigned long cr3)
471{ 528{
472 lguest_data.pgdir = cr3; 529 lguest_data.pgdir = cr3;
@@ -538,10 +595,12 @@ static void lguest_write_cr4(unsigned long val)
538 * the real page tables based on the Guests'. 595 * the real page tables based on the Guests'.
539 */ 596 */
540 597
541/* The Guest calls this to set a second-level entry (pte), ie. to map a page 598/*
599 * The Guest calls this to set a second-level entry (pte), ie. to map a page
542 * into a process' address space. We set the entry then tell the Host the 600 * into a process' address space. We set the entry then tell the Host the
543 * toplevel and address this corresponds to. The Guest uses one pagetable per 601 * toplevel and address this corresponds to. The Guest uses one pagetable per
544 * process, so we need to tell the Host which one we're changing (mm->pgd). */ 602 * process, so we need to tell the Host which one we're changing (mm->pgd).
603 */
545static void lguest_pte_update(struct mm_struct *mm, unsigned long addr, 604static void lguest_pte_update(struct mm_struct *mm, unsigned long addr,
546 pte_t *ptep) 605 pte_t *ptep)
547{ 606{
@@ -560,10 +619,13 @@ static void lguest_set_pte_at(struct mm_struct *mm, unsigned long addr,
560 lguest_pte_update(mm, addr, ptep); 619 lguest_pte_update(mm, addr, ptep);
561} 620}
562 621
563/* The Guest calls lguest_set_pud to set a top-level entry and lguest_set_pmd 622/*
623 * The Guest calls lguest_set_pud to set a top-level entry and lguest_set_pmd
564 * to set a middle-level entry when PAE is activated. 624 * to set a middle-level entry when PAE is activated.
625 *
565 * Again, we set the entry then tell the Host which page we changed, 626 * Again, we set the entry then tell the Host which page we changed,
566 * and the index of the entry we changed. */ 627 * and the index of the entry we changed.
628 */
567#ifdef CONFIG_X86_PAE 629#ifdef CONFIG_X86_PAE
568static void lguest_set_pud(pud_t *pudp, pud_t pudval) 630static void lguest_set_pud(pud_t *pudp, pud_t pudval)
569{ 631{
@@ -582,8 +644,7 @@ static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval)
582} 644}
583#else 645#else
584 646
585/* The Guest calls lguest_set_pmd to set a top-level entry when PAE is not 647/* The Guest calls lguest_set_pmd to set a top-level entry when !PAE. */
586 * activated. */
587static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval) 648static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval)
588{ 649{
589 native_set_pmd(pmdp, pmdval); 650 native_set_pmd(pmdp, pmdval);
@@ -592,7 +653,8 @@ static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval)
592} 653}
593#endif 654#endif
594 655
595/* There are a couple of legacy places where the kernel sets a PTE, but we 656/*
657 * There are a couple of legacy places where the kernel sets a PTE, but we
596 * don't know the top level any more. This is useless for us, since we don't 658 * don't know the top level any more. This is useless for us, since we don't
597 * know which pagetable is changing or what address, so we just tell the Host 659 * know which pagetable is changing or what address, so we just tell the Host
598 * to forget all of them. Fortunately, this is very rare. 660 * to forget all of them. Fortunately, this is very rare.
@@ -600,7 +662,8 @@ static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval)
600 * ... except in early boot when the kernel sets up the initial pagetables, 662 * ... except in early boot when the kernel sets up the initial pagetables,
601 * which makes booting astonishingly slow: 1.83 seconds! So we don't even tell 663 * which makes booting astonishingly slow: 1.83 seconds! So we don't even tell
602 * the Host anything changed until we've done the first page table switch, 664 * the Host anything changed until we've done the first page table switch,
603 * which brings boot back to 0.25 seconds. */ 665 * which brings boot back to 0.25 seconds.
666 */
604static void lguest_set_pte(pte_t *ptep, pte_t pteval) 667static void lguest_set_pte(pte_t *ptep, pte_t pteval)
605{ 668{
606 native_set_pte(ptep, pteval); 669 native_set_pte(ptep, pteval);
@@ -628,7 +691,8 @@ void lguest_pmd_clear(pmd_t *pmdp)
628} 691}
629#endif 692#endif
630 693
631/* Unfortunately for Lguest, the pv_mmu_ops for page tables were based on 694/*
695 * Unfortunately for Lguest, the pv_mmu_ops for page tables were based on
632 * native page table operations. On native hardware you can set a new page 696 * native page table operations. On native hardware you can set a new page
633 * table entry whenever you want, but if you want to remove one you have to do 697 * table entry whenever you want, but if you want to remove one you have to do
634 * a TLB flush (a TLB is a little cache of page table entries kept by the CPU). 698 * a TLB flush (a TLB is a little cache of page table entries kept by the CPU).
@@ -637,24 +701,29 @@ void lguest_pmd_clear(pmd_t *pmdp)
637 * called when a valid entry is written, not when it's removed (ie. marked not 701 * called when a valid entry is written, not when it's removed (ie. marked not
638 * present). Instead, this is where we come when the Guest wants to remove a 702 * present). Instead, this is where we come when the Guest wants to remove a
639 * page table entry: we tell the Host to set that entry to 0 (ie. the present 703 * page table entry: we tell the Host to set that entry to 0 (ie. the present
640 * bit is zero). */ 704 * bit is zero).
705 */
641static void lguest_flush_tlb_single(unsigned long addr) 706static void lguest_flush_tlb_single(unsigned long addr)
642{ 707{
643 /* Simply set it to zero: if it was not, it will fault back in. */ 708 /* Simply set it to zero: if it was not, it will fault back in. */
644 lazy_hcall3(LHCALL_SET_PTE, lguest_data.pgdir, addr, 0); 709 lazy_hcall3(LHCALL_SET_PTE, lguest_data.pgdir, addr, 0);
645} 710}
646 711
647/* This is what happens after the Guest has removed a large number of entries. 712/*
713 * This is what happens after the Guest has removed a large number of entries.
648 * This tells the Host that any of the page table entries for userspace might 714 * This tells the Host that any of the page table entries for userspace might
649 * have changed, ie. virtual addresses below PAGE_OFFSET. */ 715 * have changed, ie. virtual addresses below PAGE_OFFSET.
716 */
650static void lguest_flush_tlb_user(void) 717static void lguest_flush_tlb_user(void)
651{ 718{
652 lazy_hcall1(LHCALL_FLUSH_TLB, 0); 719 lazy_hcall1(LHCALL_FLUSH_TLB, 0);
653} 720}
654 721
655/* This is called when the kernel page tables have changed. That's not very 722/*
723 * This is called when the kernel page tables have changed. That's not very
656 * common (unless the Guest is using highmem, which makes the Guest extremely 724 * common (unless the Guest is using highmem, which makes the Guest extremely
657 * slow), so it's worth separating this from the user flushing above. */ 725 * slow), so it's worth separating this from the user flushing above.
726 */
658static void lguest_flush_tlb_kernel(void) 727static void lguest_flush_tlb_kernel(void)
659{ 728{
660 lazy_hcall1(LHCALL_FLUSH_TLB, 1); 729 lazy_hcall1(LHCALL_FLUSH_TLB, 1);
@@ -691,23 +760,27 @@ static struct irq_chip lguest_irq_controller = {
691 .unmask = enable_lguest_irq, 760 .unmask = enable_lguest_irq,
692}; 761};
693 762
694/* This sets up the Interrupt Descriptor Table (IDT) entry for each hardware 763/*
764 * This sets up the Interrupt Descriptor Table (IDT) entry for each hardware
695 * interrupt (except 128, which is used for system calls), and then tells the 765 * interrupt (except 128, which is used for system calls), and then tells the
696 * Linux infrastructure that each interrupt is controlled by our level-based 766 * Linux infrastructure that each interrupt is controlled by our level-based
697 * lguest interrupt controller. */ 767 * lguest interrupt controller.
768 */
698static void __init lguest_init_IRQ(void) 769static void __init lguest_init_IRQ(void)
699{ 770{
700 unsigned int i; 771 unsigned int i;
701 772
702 for (i = FIRST_EXTERNAL_VECTOR; i < NR_VECTORS; i++) { 773 for (i = FIRST_EXTERNAL_VECTOR; i < NR_VECTORS; i++) {
703 /* Some systems map "vectors" to interrupts weirdly. Lguest has 774 /* Some systems map "vectors" to interrupts weirdly. Not us! */
704 * a straightforward 1 to 1 mapping, so force that here. */
705 __get_cpu_var(vector_irq)[i] = i - FIRST_EXTERNAL_VECTOR; 775 __get_cpu_var(vector_irq)[i] = i - FIRST_EXTERNAL_VECTOR;
706 if (i != SYSCALL_VECTOR) 776 if (i != SYSCALL_VECTOR)
707 set_intr_gate(i, interrupt[i - FIRST_EXTERNAL_VECTOR]); 777 set_intr_gate(i, interrupt[i - FIRST_EXTERNAL_VECTOR]);
708 } 778 }
709 /* This call is required to set up for 4k stacks, where we have 779
710 * separate stacks for hard and soft interrupts. */ 780 /*
781 * This call is required to set up for 4k stacks, where we have
782 * separate stacks for hard and soft interrupts.
783 */
711 irq_ctx_init(smp_processor_id()); 784 irq_ctx_init(smp_processor_id());
712} 785}
713 786
@@ -729,31 +802,39 @@ static unsigned long lguest_get_wallclock(void)
729 return lguest_data.time.tv_sec; 802 return lguest_data.time.tv_sec;
730} 803}
731 804
732/* The TSC is an Intel thing called the Time Stamp Counter. The Host tells us 805/*
806 * The TSC is an Intel thing called the Time Stamp Counter. The Host tells us
733 * what speed it runs at, or 0 if it's unusable as a reliable clock source. 807 * what speed it runs at, or 0 if it's unusable as a reliable clock source.
734 * This matches what we want here: if we return 0 from this function, the x86 808 * This matches what we want here: if we return 0 from this function, the x86
735 * TSC clock will give up and not register itself. */ 809 * TSC clock will give up and not register itself.
810 */
736static unsigned long lguest_tsc_khz(void) 811static unsigned long lguest_tsc_khz(void)
737{ 812{
738 return lguest_data.tsc_khz; 813 return lguest_data.tsc_khz;
739} 814}
740 815
741/* If we can't use the TSC, the kernel falls back to our lower-priority 816/*
742 * "lguest_clock", where we read the time value given to us by the Host. */ 817 * If we can't use the TSC, the kernel falls back to our lower-priority
818 * "lguest_clock", where we read the time value given to us by the Host.
819 */
743static cycle_t lguest_clock_read(struct clocksource *cs) 820static cycle_t lguest_clock_read(struct clocksource *cs)
744{ 821{
745 unsigned long sec, nsec; 822 unsigned long sec, nsec;
746 823
747 /* Since the time is in two parts (seconds and nanoseconds), we risk 824 /*
825 * Since the time is in two parts (seconds and nanoseconds), we risk
748 * reading it just as it's changing from 99 & 0.999999999 to 100 and 0, 826 * reading it just as it's changing from 99 & 0.999999999 to 100 and 0,
749 * and getting 99 and 0. As Linux tends to come apart under the stress 827 * and getting 99 and 0. As Linux tends to come apart under the stress
750 * of time travel, we must be careful: */ 828 * of time travel, we must be careful:
829 */
751 do { 830 do {
752 /* First we read the seconds part. */ 831 /* First we read the seconds part. */
753 sec = lguest_data.time.tv_sec; 832 sec = lguest_data.time.tv_sec;
754 /* This read memory barrier tells the compiler and the CPU that 833 /*
834 * This read memory barrier tells the compiler and the CPU that
755 * this can't be reordered: we have to complete the above 835 * this can't be reordered: we have to complete the above
756 * before going on. */ 836 * before going on.
837 */
757 rmb(); 838 rmb();
758 /* Now we read the nanoseconds part. */ 839 /* Now we read the nanoseconds part. */
759 nsec = lguest_data.time.tv_nsec; 840 nsec = lguest_data.time.tv_nsec;
@@ -777,9 +858,11 @@ static struct clocksource lguest_clock = {
777 .flags = CLOCK_SOURCE_IS_CONTINUOUS, 858 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
778}; 859};
779 860
780/* We also need a "struct clock_event_device": Linux asks us to set it to go 861/*
862 * We also need a "struct clock_event_device": Linux asks us to set it to go
781 * off some time in the future. Actually, James Morris figured all this out, I 863 * off some time in the future. Actually, James Morris figured all this out, I
782 * just applied the patch. */ 864 * just applied the patch.
865 */
783static int lguest_clockevent_set_next_event(unsigned long delta, 866static int lguest_clockevent_set_next_event(unsigned long delta,
784 struct clock_event_device *evt) 867 struct clock_event_device *evt)
785{ 868{
@@ -829,8 +912,10 @@ static struct clock_event_device lguest_clockevent = {
829 .max_delta_ns = LG_CLOCK_MAX_DELTA, 912 .max_delta_ns = LG_CLOCK_MAX_DELTA,
830}; 913};
831 914
832/* This is the Guest timer interrupt handler (hardware interrupt 0). We just 915/*
833 * call the clockevent infrastructure and it does whatever needs doing. */ 916 * This is the Guest timer interrupt handler (hardware interrupt 0). We just
917 * call the clockevent infrastructure and it does whatever needs doing.
918 */
834static void lguest_time_irq(unsigned int irq, struct irq_desc *desc) 919static void lguest_time_irq(unsigned int irq, struct irq_desc *desc)
835{ 920{
836 unsigned long flags; 921 unsigned long flags;
@@ -841,10 +926,12 @@ static void lguest_time_irq(unsigned int irq, struct irq_desc *desc)
841 local_irq_restore(flags); 926 local_irq_restore(flags);
842} 927}
843 928
844/* At some point in the boot process, we get asked to set up our timing 929/*
930 * At some point in the boot process, we get asked to set up our timing
845 * infrastructure. The kernel doesn't expect timer interrupts before this, but 931 * infrastructure. The kernel doesn't expect timer interrupts before this, but
846 * we cleverly initialized the "blocked_interrupts" field of "struct 932 * we cleverly initialized the "blocked_interrupts" field of "struct
847 * lguest_data" so that timer interrupts were blocked until now. */ 933 * lguest_data" so that timer interrupts were blocked until now.
934 */
848static void lguest_time_init(void) 935static void lguest_time_init(void)
849{ 936{
850 /* Set up the timer interrupt (0) to go to our simple timer routine */ 937 /* Set up the timer interrupt (0) to go to our simple timer routine */
@@ -868,14 +955,16 @@ static void lguest_time_init(void)
868 * to work. They're pretty simple. 955 * to work. They're pretty simple.
869 */ 956 */
870 957
871/* The Guest needs to tell the Host what stack it expects traps to use. For 958/*
959 * The Guest needs to tell the Host what stack it expects traps to use. For
872 * native hardware, this is part of the Task State Segment mentioned above in 960 * native hardware, this is part of the Task State Segment mentioned above in
873 * lguest_load_tr_desc(), but to help hypervisors there's this special call. 961 * lguest_load_tr_desc(), but to help hypervisors there's this special call.
874 * 962 *
875 * We tell the Host the segment we want to use (__KERNEL_DS is the kernel data 963 * We tell the Host the segment we want to use (__KERNEL_DS is the kernel data
876 * segment), the privilege level (we're privilege level 1, the Host is 0 and 964 * segment), the privilege level (we're privilege level 1, the Host is 0 and
877 * will not tolerate us trying to use that), the stack pointer, and the number 965 * will not tolerate us trying to use that), the stack pointer, and the number
878 * of pages in the stack. */ 966 * of pages in the stack.
967 */
879static void lguest_load_sp0(struct tss_struct *tss, 968static void lguest_load_sp0(struct tss_struct *tss,
880 struct thread_struct *thread) 969 struct thread_struct *thread)
881{ 970{
@@ -889,7 +978,8 @@ static void lguest_set_debugreg(int regno, unsigned long value)
889 /* FIXME: Implement */ 978 /* FIXME: Implement */
890} 979}
891 980
892/* There are times when the kernel wants to make sure that no memory writes are 981/*
982 * There are times when the kernel wants to make sure that no memory writes are
893 * caught in the cache (that they've all reached real hardware devices). This 983 * caught in the cache (that they've all reached real hardware devices). This
894 * doesn't matter for the Guest which has virtual hardware. 984 * doesn't matter for the Guest which has virtual hardware.
895 * 985 *
@@ -903,11 +993,13 @@ static void lguest_wbinvd(void)
903{ 993{
904} 994}
905 995
906/* If the Guest expects to have an Advanced Programmable Interrupt Controller, 996/*
997 * If the Guest expects to have an Advanced Programmable Interrupt Controller,
907 * we play dumb by ignoring writes and returning 0 for reads. So it's no 998 * we play dumb by ignoring writes and returning 0 for reads. So it's no
908 * longer Programmable nor Controlling anything, and I don't think 8 lines of 999 * longer Programmable nor Controlling anything, and I don't think 8 lines of
909 * code qualifies for Advanced. It will also never interrupt anything. It 1000 * code qualifies for Advanced. It will also never interrupt anything. It
910 * does, however, allow us to get through the Linux boot code. */ 1001 * does, however, allow us to get through the Linux boot code.
1002 */
911#ifdef CONFIG_X86_LOCAL_APIC 1003#ifdef CONFIG_X86_LOCAL_APIC
912static void lguest_apic_write(u32 reg, u32 v) 1004static void lguest_apic_write(u32 reg, u32 v)
913{ 1005{
@@ -956,11 +1048,13 @@ static void lguest_safe_halt(void)
956 kvm_hypercall0(LHCALL_HALT); 1048 kvm_hypercall0(LHCALL_HALT);
957} 1049}
958 1050
959/* The SHUTDOWN hypercall takes a string to describe what's happening, and 1051/*
1052 * The SHUTDOWN hypercall takes a string to describe what's happening, and
960 * an argument which says whether this to restart (reboot) the Guest or not. 1053 * an argument which says whether this to restart (reboot) the Guest or not.
961 * 1054 *
962 * Note that the Host always prefers that the Guest speak in physical addresses 1055 * Note that the Host always prefers that the Guest speak in physical addresses
963 * rather than virtual addresses, so we use __pa() here. */ 1056 * rather than virtual addresses, so we use __pa() here.
1057 */
964static void lguest_power_off(void) 1058static void lguest_power_off(void)
965{ 1059{
966 kvm_hypercall2(LHCALL_SHUTDOWN, __pa("Power down"), 1060 kvm_hypercall2(LHCALL_SHUTDOWN, __pa("Power down"),
@@ -991,8 +1085,10 @@ static __init char *lguest_memory_setup(void)
991 * nice to move it back to lguest_init. Patch welcome... */ 1085 * nice to move it back to lguest_init. Patch welcome... */
992 atomic_notifier_chain_register(&panic_notifier_list, &paniced); 1086 atomic_notifier_chain_register(&panic_notifier_list, &paniced);
993 1087
994 /* The Linux bootloader header contains an "e820" memory map: the 1088 /*
995 * Launcher populated the first entry with our memory limit. */ 1089 *The Linux bootloader header contains an "e820" memory map: the
1090 * Launcher populated the first entry with our memory limit.
1091 */
996 e820_add_region(boot_params.e820_map[0].addr, 1092 e820_add_region(boot_params.e820_map[0].addr,
997 boot_params.e820_map[0].size, 1093 boot_params.e820_map[0].size,
998 boot_params.e820_map[0].type); 1094 boot_params.e820_map[0].type);
@@ -1001,16 +1097,17 @@ static __init char *lguest_memory_setup(void)
1001 return "LGUEST"; 1097 return "LGUEST";
1002} 1098}
1003 1099
1004/* We will eventually use the virtio console device to produce console output, 1100/*
1101 * We will eventually use the virtio console device to produce console output,
1005 * but before that is set up we use LHCALL_NOTIFY on normal memory to produce 1102 * but before that is set up we use LHCALL_NOTIFY on normal memory to produce
1006 * console output. */ 1103 * console output.
1104 */
1007static __init int early_put_chars(u32 vtermno, const char *buf, int count) 1105static __init int early_put_chars(u32 vtermno, const char *buf, int count)
1008{ 1106{
1009 char scratch[17]; 1107 char scratch[17];
1010 unsigned int len = count; 1108 unsigned int len = count;
1011 1109
1012 /* We use a nul-terminated string, so we have to make a copy. Icky, 1110 /* We use a nul-terminated string, so we make a copy. Icky, huh? */
1013 * huh? */
1014 if (len > sizeof(scratch) - 1) 1111 if (len > sizeof(scratch) - 1)
1015 len = sizeof(scratch) - 1; 1112 len = sizeof(scratch) - 1;
1016 scratch[len] = '\0'; 1113 scratch[len] = '\0';
@@ -1021,8 +1118,10 @@ static __init int early_put_chars(u32 vtermno, const char *buf, int count)
1021 return len; 1118 return len;
1022} 1119}
1023 1120
1024/* Rebooting also tells the Host we're finished, but the RESTART flag tells the 1121/*
1025 * Launcher to reboot us. */ 1122 * Rebooting also tells the Host we're finished, but the RESTART flag tells the
1123 * Launcher to reboot us.
1124 */
1026static void lguest_restart(char *reason) 1125static void lguest_restart(char *reason)
1027{ 1126{
1028 kvm_hypercall2(LHCALL_SHUTDOWN, __pa(reason), LGUEST_SHUTDOWN_RESTART); 1127 kvm_hypercall2(LHCALL_SHUTDOWN, __pa(reason), LGUEST_SHUTDOWN_RESTART);
@@ -1049,7 +1148,8 @@ static void lguest_restart(char *reason)
1049 * fit comfortably. 1148 * fit comfortably.
1050 * 1149 *
1051 * First we need assembly templates of each of the patchable Guest operations, 1150 * First we need assembly templates of each of the patchable Guest operations,
1052 * and these are in i386_head.S. */ 1151 * and these are in i386_head.S.
1152 */
1053 1153
1054/*G:060 We construct a table from the assembler templates: */ 1154/*G:060 We construct a table from the assembler templates: */
1055static const struct lguest_insns 1155static const struct lguest_insns
@@ -1060,9 +1160,11 @@ static const struct lguest_insns
1060 [PARAVIRT_PATCH(pv_irq_ops.save_fl)] = { lgstart_pushf, lgend_pushf }, 1160 [PARAVIRT_PATCH(pv_irq_ops.save_fl)] = { lgstart_pushf, lgend_pushf },
1061}; 1161};
1062 1162
1063/* Now our patch routine is fairly simple (based on the native one in 1163/*
1164 * Now our patch routine is fairly simple (based on the native one in
1064 * paravirt.c). If we have a replacement, we copy it in and return how much of 1165 * paravirt.c). If we have a replacement, we copy it in and return how much of
1065 * the available space we used. */ 1166 * the available space we used.
1167 */
1066static unsigned lguest_patch(u8 type, u16 clobber, void *ibuf, 1168static unsigned lguest_patch(u8 type, u16 clobber, void *ibuf,
1067 unsigned long addr, unsigned len) 1169 unsigned long addr, unsigned len)
1068{ 1170{
@@ -1074,8 +1176,7 @@ static unsigned lguest_patch(u8 type, u16 clobber, void *ibuf,
1074 1176
1075 insn_len = lguest_insns[type].end - lguest_insns[type].start; 1177 insn_len = lguest_insns[type].end - lguest_insns[type].start;
1076 1178
1077 /* Similarly if we can't fit replacement (shouldn't happen, but let's 1179 /* Similarly if it can't fit (doesn't happen, but let's be thorough). */
1078 * be thorough). */
1079 if (len < insn_len) 1180 if (len < insn_len)
1080 return paravirt_patch_default(type, clobber, ibuf, addr, len); 1181 return paravirt_patch_default(type, clobber, ibuf, addr, len);
1081 1182
@@ -1084,22 +1185,28 @@ static unsigned lguest_patch(u8 type, u16 clobber, void *ibuf,
1084 return insn_len; 1185 return insn_len;
1085} 1186}
1086 1187
1087/*G:029 Once we get to lguest_init(), we know we're a Guest. The various 1188/*G:029
1189 * Once we get to lguest_init(), we know we're a Guest. The various
1088 * pv_ops structures in the kernel provide points for (almost) every routine we 1190 * pv_ops structures in the kernel provide points for (almost) every routine we
1089 * have to override to avoid privileged instructions. */ 1191 * have to override to avoid privileged instructions.
1192 */
1090__init void lguest_init(void) 1193__init void lguest_init(void)
1091{ 1194{
1092 /* We're under lguest, paravirt is enabled, and we're running at 1195 /* We're under lguest. */
1093 * privilege level 1, not 0 as normal. */
1094 pv_info.name = "lguest"; 1196 pv_info.name = "lguest";
1197 /* Paravirt is enabled. */
1095 pv_info.paravirt_enabled = 1; 1198 pv_info.paravirt_enabled = 1;
1199 /* We're running at privilege level 1, not 0 as normal. */
1096 pv_info.kernel_rpl = 1; 1200 pv_info.kernel_rpl = 1;
1201 /* Everyone except Xen runs with this set. */
1097 pv_info.shared_kernel_pmd = 1; 1202 pv_info.shared_kernel_pmd = 1;
1098 1203
1099 /* We set up all the lguest overrides for sensitive operations. These 1204 /*
1100 * are detailed with the operations themselves. */ 1205 * We set up all the lguest overrides for sensitive operations. These
1206 * are detailed with the operations themselves.
1207 */
1101 1208
1102 /* interrupt-related operations */ 1209 /* Interrupt-related operations */
1103 pv_irq_ops.init_IRQ = lguest_init_IRQ; 1210 pv_irq_ops.init_IRQ = lguest_init_IRQ;
1104 pv_irq_ops.save_fl = PV_CALLEE_SAVE(save_fl); 1211 pv_irq_ops.save_fl = PV_CALLEE_SAVE(save_fl);
1105 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(lg_restore_fl); 1212 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(lg_restore_fl);
@@ -1107,11 +1214,11 @@ __init void lguest_init(void)
1107 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(lg_irq_enable); 1214 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(lg_irq_enable);
1108 pv_irq_ops.safe_halt = lguest_safe_halt; 1215 pv_irq_ops.safe_halt = lguest_safe_halt;
1109 1216
1110 /* init-time operations */ 1217 /* Setup operations */
1111 pv_init_ops.memory_setup = lguest_memory_setup; 1218 pv_init_ops.memory_setup = lguest_memory_setup;
1112 pv_init_ops.patch = lguest_patch; 1219 pv_init_ops.patch = lguest_patch;
1113 1220
1114 /* Intercepts of various cpu instructions */ 1221 /* Intercepts of various CPU instructions */
1115 pv_cpu_ops.load_gdt = lguest_load_gdt; 1222 pv_cpu_ops.load_gdt = lguest_load_gdt;
1116 pv_cpu_ops.cpuid = lguest_cpuid; 1223 pv_cpu_ops.cpuid = lguest_cpuid;
1117 pv_cpu_ops.load_idt = lguest_load_idt; 1224 pv_cpu_ops.load_idt = lguest_load_idt;
@@ -1132,7 +1239,7 @@ __init void lguest_init(void)
1132 pv_cpu_ops.start_context_switch = paravirt_start_context_switch; 1239 pv_cpu_ops.start_context_switch = paravirt_start_context_switch;
1133 pv_cpu_ops.end_context_switch = lguest_end_context_switch; 1240 pv_cpu_ops.end_context_switch = lguest_end_context_switch;
1134 1241
1135 /* pagetable management */ 1242 /* Pagetable management */
1136 pv_mmu_ops.write_cr3 = lguest_write_cr3; 1243 pv_mmu_ops.write_cr3 = lguest_write_cr3;
1137 pv_mmu_ops.flush_tlb_user = lguest_flush_tlb_user; 1244 pv_mmu_ops.flush_tlb_user = lguest_flush_tlb_user;
1138 pv_mmu_ops.flush_tlb_single = lguest_flush_tlb_single; 1245 pv_mmu_ops.flush_tlb_single = lguest_flush_tlb_single;
@@ -1154,54 +1261,71 @@ __init void lguest_init(void)
1154 pv_mmu_ops.pte_update_defer = lguest_pte_update; 1261 pv_mmu_ops.pte_update_defer = lguest_pte_update;
1155 1262
1156#ifdef CONFIG_X86_LOCAL_APIC 1263#ifdef CONFIG_X86_LOCAL_APIC
1157 /* apic read/write intercepts */ 1264 /* APIC read/write intercepts */
1158 set_lguest_basic_apic_ops(); 1265 set_lguest_basic_apic_ops();
1159#endif 1266#endif
1160 1267
1161 /* time operations */ 1268 /* Time operations */
1162 pv_time_ops.get_wallclock = lguest_get_wallclock; 1269 pv_time_ops.get_wallclock = lguest_get_wallclock;
1163 pv_time_ops.time_init = lguest_time_init; 1270 pv_time_ops.time_init = lguest_time_init;
1164 pv_time_ops.get_tsc_khz = lguest_tsc_khz; 1271 pv_time_ops.get_tsc_khz = lguest_tsc_khz;
1165 1272
1166 /* Now is a good time to look at the implementations of these functions 1273 /*
1167 * before returning to the rest of lguest_init(). */ 1274 * Now is a good time to look at the implementations of these functions
1275 * before returning to the rest of lguest_init().
1276 */
1168 1277
1169 /*G:070 Now we've seen all the paravirt_ops, we return to 1278 /*G:070
1279 * Now we've seen all the paravirt_ops, we return to
1170 * lguest_init() where the rest of the fairly chaotic boot setup 1280 * lguest_init() where the rest of the fairly chaotic boot setup
1171 * occurs. */ 1281 * occurs.
1282 */
1172 1283
1173 /* The stack protector is a weird thing where gcc places a canary 1284 /*
1285 * The stack protector is a weird thing where gcc places a canary
1174 * value on the stack and then checks it on return. This file is 1286 * value on the stack and then checks it on return. This file is
1175 * compiled with -fno-stack-protector it, so we got this far without 1287 * compiled with -fno-stack-protector it, so we got this far without
1176 * problems. The value of the canary is kept at offset 20 from the 1288 * problems. The value of the canary is kept at offset 20 from the
1177 * %gs register, so we need to set that up before calling C functions 1289 * %gs register, so we need to set that up before calling C functions
1178 * in other files. */ 1290 * in other files.
1291 */
1179 setup_stack_canary_segment(0); 1292 setup_stack_canary_segment(0);
1180 /* We could just call load_stack_canary_segment(), but we might as 1293
1181 * call switch_to_new_gdt() which loads the whole table and sets up 1294 /*
1182 * the per-cpu segment descriptor register %fs as well. */ 1295 * We could just call load_stack_canary_segment(), but we might as well
1296 * call switch_to_new_gdt() which loads the whole table and sets up the
1297 * per-cpu segment descriptor register %fs as well.
1298 */
1183 switch_to_new_gdt(0); 1299 switch_to_new_gdt(0);
1184 1300
1185 /* As described in head_32.S, we map the first 128M of memory. */ 1301 /* As described in head_32.S, we map the first 128M of memory. */
1186 max_pfn_mapped = (128*1024*1024) >> PAGE_SHIFT; 1302 max_pfn_mapped = (128*1024*1024) >> PAGE_SHIFT;
1187 1303
1188 /* The Host<->Guest Switcher lives at the top of our address space, and 1304 /*
1305 * The Host<->Guest Switcher lives at the top of our address space, and
1189 * the Host told us how big it is when we made LGUEST_INIT hypercall: 1306 * the Host told us how big it is when we made LGUEST_INIT hypercall:
1190 * it put the answer in lguest_data.reserve_mem */ 1307 * it put the answer in lguest_data.reserve_mem
1308 */
1191 reserve_top_address(lguest_data.reserve_mem); 1309 reserve_top_address(lguest_data.reserve_mem);
1192 1310
1193 /* If we don't initialize the lock dependency checker now, it crashes 1311 /*
1194 * paravirt_disable_iospace. */ 1312 * If we don't initialize the lock dependency checker now, it crashes
1313 * paravirt_disable_iospace.
1314 */
1195 lockdep_init(); 1315 lockdep_init();
1196 1316
1197 /* The IDE code spends about 3 seconds probing for disks: if we reserve 1317 /*
1318 * The IDE code spends about 3 seconds probing for disks: if we reserve
1198 * all the I/O ports up front it can't get them and so doesn't probe. 1319 * all the I/O ports up front it can't get them and so doesn't probe.
1199 * Other device drivers are similar (but less severe). This cuts the 1320 * Other device drivers are similar (but less severe). This cuts the
1200 * kernel boot time on my machine from 4.1 seconds to 0.45 seconds. */ 1321 * kernel boot time on my machine from 4.1 seconds to 0.45 seconds.
1322 */
1201 paravirt_disable_iospace(); 1323 paravirt_disable_iospace();
1202 1324
1203 /* This is messy CPU setup stuff which the native boot code does before 1325 /*
1204 * start_kernel, so we have to do, too: */ 1326 * This is messy CPU setup stuff which the native boot code does before
1327 * start_kernel, so we have to do, too:
1328 */
1205 cpu_detect(&new_cpu_data); 1329 cpu_detect(&new_cpu_data);
1206 /* head.S usually sets up the first capability word, so do it here. */ 1330 /* head.S usually sets up the first capability word, so do it here. */
1207 new_cpu_data.x86_capability[0] = cpuid_edx(1); 1331 new_cpu_data.x86_capability[0] = cpuid_edx(1);
@@ -1218,22 +1342,28 @@ __init void lguest_init(void)
1218 acpi_ht = 0; 1342 acpi_ht = 0;
1219#endif 1343#endif
1220 1344
1221 /* We set the preferred console to "hvc". This is the "hypervisor 1345 /*
1346 * We set the preferred console to "hvc". This is the "hypervisor
1222 * virtual console" driver written by the PowerPC people, which we also 1347 * virtual console" driver written by the PowerPC people, which we also
1223 * adapted for lguest's use. */ 1348 * adapted for lguest's use.
1349 */
1224 add_preferred_console("hvc", 0, NULL); 1350 add_preferred_console("hvc", 0, NULL);
1225 1351
1226 /* Register our very early console. */ 1352 /* Register our very early console. */
1227 virtio_cons_early_init(early_put_chars); 1353 virtio_cons_early_init(early_put_chars);
1228 1354
1229 /* Last of all, we set the power management poweroff hook to point to 1355 /*
1356 * Last of all, we set the power management poweroff hook to point to
1230 * the Guest routine to power off, and the reboot hook to our restart 1357 * the Guest routine to power off, and the reboot hook to our restart
1231 * routine. */ 1358 * routine.
1359 */
1232 pm_power_off = lguest_power_off; 1360 pm_power_off = lguest_power_off;
1233 machine_ops.restart = lguest_restart; 1361 machine_ops.restart = lguest_restart;
1234 1362
1235 /* Now we're set up, call i386_start_kernel() in head32.c and we proceed 1363 /*
1236 * to boot as normal. It never returns. */ 1364 * Now we're set up, call i386_start_kernel() in head32.c and we proceed
1365 * to boot as normal. It never returns.
1366 */
1237 i386_start_kernel(); 1367 i386_start_kernel();
1238} 1368}
1239/* 1369/*
diff --git a/arch/x86/lguest/i386_head.S b/arch/x86/lguest/i386_head.S
index a9c8cfe61cd4..db6aa95eb054 100644
--- a/arch/x86/lguest/i386_head.S
+++ b/arch/x86/lguest/i386_head.S
@@ -5,7 +5,8 @@
5#include <asm/thread_info.h> 5#include <asm/thread_info.h>
6#include <asm/processor-flags.h> 6#include <asm/processor-flags.h>
7 7
8/*G:020 Our story starts with the kernel booting into startup_32 in 8/*G:020
9 * Our story starts with the kernel booting into startup_32 in
9 * arch/x86/kernel/head_32.S. It expects a boot header, which is created by 10 * arch/x86/kernel/head_32.S. It expects a boot header, which is created by
10 * the bootloader (the Launcher in our case). 11 * the bootloader (the Launcher in our case).
11 * 12 *
@@ -21,11 +22,14 @@
21 * data without remembering to subtract __PAGE_OFFSET! 22 * data without remembering to subtract __PAGE_OFFSET!
22 * 23 *
23 * The .section line puts this code in .init.text so it will be discarded after 24 * The .section line puts this code in .init.text so it will be discarded after
24 * boot. */ 25 * boot.
26 */
25.section .init.text, "ax", @progbits 27.section .init.text, "ax", @progbits
26ENTRY(lguest_entry) 28ENTRY(lguest_entry)
27 /* We make the "initialization" hypercall now to tell the Host about 29 /*
28 * us, and also find out where it put our page tables. */ 30 * We make the "initialization" hypercall now to tell the Host about
31 * us, and also find out where it put our page tables.
32 */
29 movl $LHCALL_LGUEST_INIT, %eax 33 movl $LHCALL_LGUEST_INIT, %eax
30 movl $lguest_data - __PAGE_OFFSET, %ebx 34 movl $lguest_data - __PAGE_OFFSET, %ebx
31 .byte 0x0f,0x01,0xc1 /* KVM_HYPERCALL */ 35 .byte 0x0f,0x01,0xc1 /* KVM_HYPERCALL */
@@ -33,13 +37,14 @@ ENTRY(lguest_entry)
33 /* Set up the initial stack so we can run C code. */ 37 /* Set up the initial stack so we can run C code. */
34 movl $(init_thread_union+THREAD_SIZE),%esp 38 movl $(init_thread_union+THREAD_SIZE),%esp
35 39
36 /* Jumps are relative, and we're running __PAGE_OFFSET too low at the 40 /* Jumps are relative: we're running __PAGE_OFFSET too low. */
37 * moment. */
38 jmp lguest_init+__PAGE_OFFSET 41 jmp lguest_init+__PAGE_OFFSET
39 42
40/*G:055 We create a macro which puts the assembler code between lgstart_ and 43/*G:055
41 * lgend_ markers. These templates are put in the .text section: they can't be 44 * We create a macro which puts the assembler code between lgstart_ and lgend_
42 * discarded after boot as we may need to patch modules, too. */ 45 * markers. These templates are put in the .text section: they can't be
46 * discarded after boot as we may need to patch modules, too.
47 */
43.text 48.text
44#define LGUEST_PATCH(name, insns...) \ 49#define LGUEST_PATCH(name, insns...) \
45 lgstart_##name: insns; lgend_##name:; \ 50 lgstart_##name: insns; lgend_##name:; \
@@ -48,58 +53,74 @@ ENTRY(lguest_entry)
48LGUEST_PATCH(cli, movl $0, lguest_data+LGUEST_DATA_irq_enabled) 53LGUEST_PATCH(cli, movl $0, lguest_data+LGUEST_DATA_irq_enabled)
49LGUEST_PATCH(pushf, movl lguest_data+LGUEST_DATA_irq_enabled, %eax) 54LGUEST_PATCH(pushf, movl lguest_data+LGUEST_DATA_irq_enabled, %eax)
50 55
51/*G:033 But using those wrappers is inefficient (we'll see why that doesn't 56/*G:033
52 * matter for save_fl and irq_disable later). If we write our routines 57 * But using those wrappers is inefficient (we'll see why that doesn't matter
53 * carefully in assembler, we can avoid clobbering any registers and avoid 58 * for save_fl and irq_disable later). If we write our routines carefully in
54 * jumping through the wrapper functions. 59 * assembler, we can avoid clobbering any registers and avoid jumping through
60 * the wrapper functions.
55 * 61 *
56 * I skipped over our first piece of assembler, but this one is worth studying 62 * I skipped over our first piece of assembler, but this one is worth studying
57 * in a bit more detail so I'll describe in easy stages. First, the routine 63 * in a bit more detail so I'll describe in easy stages. First, the routine to
58 * to enable interrupts: */ 64 * enable interrupts:
65 */
59ENTRY(lg_irq_enable) 66ENTRY(lg_irq_enable)
60 /* The reverse of irq_disable, this sets lguest_data.irq_enabled to 67 /*
61 * X86_EFLAGS_IF (ie. "Interrupts enabled"). */ 68 * The reverse of irq_disable, this sets lguest_data.irq_enabled to
69 * X86_EFLAGS_IF (ie. "Interrupts enabled").
70 */
62 movl $X86_EFLAGS_IF, lguest_data+LGUEST_DATA_irq_enabled 71 movl $X86_EFLAGS_IF, lguest_data+LGUEST_DATA_irq_enabled
63 /* But now we need to check if the Host wants to know: there might have 72 /*
73 * But now we need to check if the Host wants to know: there might have
64 * been interrupts waiting to be delivered, in which case it will have 74 * been interrupts waiting to be delivered, in which case it will have
65 * set lguest_data.irq_pending to X86_EFLAGS_IF. If it's not zero, we 75 * set lguest_data.irq_pending to X86_EFLAGS_IF. If it's not zero, we
66 * jump to send_interrupts, otherwise we're done. */ 76 * jump to send_interrupts, otherwise we're done.
77 */
67 testl $0, lguest_data+LGUEST_DATA_irq_pending 78 testl $0, lguest_data+LGUEST_DATA_irq_pending
68 jnz send_interrupts 79 jnz send_interrupts
69 /* One cool thing about x86 is that you can do many things without using 80 /*
81 * One cool thing about x86 is that you can do many things without using
70 * a register. In this case, the normal path hasn't needed to save or 82 * a register. In this case, the normal path hasn't needed to save or
71 * restore any registers at all! */ 83 * restore any registers at all!
84 */
72 ret 85 ret
73send_interrupts: 86send_interrupts:
74 /* OK, now we need a register: eax is used for the hypercall number, 87 /*
88 * OK, now we need a register: eax is used for the hypercall number,
75 * which is LHCALL_SEND_INTERRUPTS. 89 * which is LHCALL_SEND_INTERRUPTS.
76 * 90 *
77 * We used not to bother with this pending detection at all, which was 91 * We used not to bother with this pending detection at all, which was
78 * much simpler. Sooner or later the Host would realize it had to 92 * much simpler. Sooner or later the Host would realize it had to
79 * send us an interrupt. But that turns out to make performance 7 93 * send us an interrupt. But that turns out to make performance 7
80 * times worse on a simple tcp benchmark. So now we do this the hard 94 * times worse on a simple tcp benchmark. So now we do this the hard
81 * way. */ 95 * way.
96 */
82 pushl %eax 97 pushl %eax
83 movl $LHCALL_SEND_INTERRUPTS, %eax 98 movl $LHCALL_SEND_INTERRUPTS, %eax
84 /* This is a vmcall instruction (same thing that KVM uses). Older 99 /*
100 * This is a vmcall instruction (same thing that KVM uses). Older
85 * assembler versions might not know the "vmcall" instruction, so we 101 * assembler versions might not know the "vmcall" instruction, so we
86 * create one manually here. */ 102 * create one manually here.
103 */
87 .byte 0x0f,0x01,0xc1 /* KVM_HYPERCALL */ 104 .byte 0x0f,0x01,0xc1 /* KVM_HYPERCALL */
88 popl %eax 105 popl %eax
89 ret 106 ret
90 107
91/* Finally, the "popf" or "restore flags" routine. The %eax register holds the 108/*
109 * Finally, the "popf" or "restore flags" routine. The %eax register holds the
92 * flags (in practice, either X86_EFLAGS_IF or 0): if it's X86_EFLAGS_IF we're 110 * flags (in practice, either X86_EFLAGS_IF or 0): if it's X86_EFLAGS_IF we're
93 * enabling interrupts again, if it's 0 we're leaving them off. */ 111 * enabling interrupts again, if it's 0 we're leaving them off.
112 */
94ENTRY(lg_restore_fl) 113ENTRY(lg_restore_fl)
95 /* This is just "lguest_data.irq_enabled = flags;" */ 114 /* This is just "lguest_data.irq_enabled = flags;" */
96 movl %eax, lguest_data+LGUEST_DATA_irq_enabled 115 movl %eax, lguest_data+LGUEST_DATA_irq_enabled
97 /* Now, if the %eax value has enabled interrupts and 116 /*
117 * Now, if the %eax value has enabled interrupts and
98 * lguest_data.irq_pending is set, we want to tell the Host so it can 118 * lguest_data.irq_pending is set, we want to tell the Host so it can
99 * deliver any outstanding interrupts. Fortunately, both values will 119 * deliver any outstanding interrupts. Fortunately, both values will
100 * be X86_EFLAGS_IF (ie. 512) in that case, and the "testl" 120 * be X86_EFLAGS_IF (ie. 512) in that case, and the "testl"
101 * instruction will AND them together for us. If both are set, we 121 * instruction will AND them together for us. If both are set, we
102 * jump to send_interrupts. */ 122 * jump to send_interrupts.
123 */
103 testl lguest_data+LGUEST_DATA_irq_pending, %eax 124 testl lguest_data+LGUEST_DATA_irq_pending, %eax
104 jnz send_interrupts 125 jnz send_interrupts
105 /* Again, the normal path has used no extra registers. Clever, huh? */ 126 /* Again, the normal path has used no extra registers. Clever, huh? */
@@ -109,22 +130,24 @@ ENTRY(lg_restore_fl)
109.global lguest_noirq_start 130.global lguest_noirq_start
110.global lguest_noirq_end 131.global lguest_noirq_end
111 132
112/*M:004 When the Host reflects a trap or injects an interrupt into the Guest, 133/*M:004
113 * it sets the eflags interrupt bit on the stack based on 134 * When the Host reflects a trap or injects an interrupt into the Guest, it
114 * lguest_data.irq_enabled, so the Guest iret logic does the right thing when 135 * sets the eflags interrupt bit on the stack based on lguest_data.irq_enabled,
115 * restoring it. However, when the Host sets the Guest up for direct traps, 136 * so the Guest iret logic does the right thing when restoring it. However,
116 * such as system calls, the processor is the one to push eflags onto the 137 * when the Host sets the Guest up for direct traps, such as system calls, the
117 * stack, and the interrupt bit will be 1 (in reality, interrupts are always 138 * processor is the one to push eflags onto the stack, and the interrupt bit
118 * enabled in the Guest). 139 * will be 1 (in reality, interrupts are always enabled in the Guest).
119 * 140 *
120 * This turns out to be harmless: the only trap which should happen under Linux 141 * This turns out to be harmless: the only trap which should happen under Linux
121 * with interrupts disabled is Page Fault (due to our lazy mapping of vmalloc 142 * with interrupts disabled is Page Fault (due to our lazy mapping of vmalloc
122 * regions), which has to be reflected through the Host anyway. If another 143 * regions), which has to be reflected through the Host anyway. If another
123 * trap *does* go off when interrupts are disabled, the Guest will panic, and 144 * trap *does* go off when interrupts are disabled, the Guest will panic, and
124 * we'll never get to this iret! :*/ 145 * we'll never get to this iret!
146:*/
125 147
126/*G:045 There is one final paravirt_op that the Guest implements, and glancing 148/*G:045
127 * at it you can see why I left it to last. It's *cool*! It's in *assembler*! 149 * There is one final paravirt_op that the Guest implements, and glancing at it
150 * you can see why I left it to last. It's *cool*! It's in *assembler*!
128 * 151 *
129 * The "iret" instruction is used to return from an interrupt or trap. The 152 * The "iret" instruction is used to return from an interrupt or trap. The
130 * stack looks like this: 153 * stack looks like this:
@@ -148,15 +171,18 @@ ENTRY(lg_restore_fl)
148 * return to userspace or wherever. Our solution to this is to surround the 171 * return to userspace or wherever. Our solution to this is to surround the
149 * code with lguest_noirq_start: and lguest_noirq_end: labels. We tell the 172 * code with lguest_noirq_start: and lguest_noirq_end: labels. We tell the
150 * Host that it is *never* to interrupt us there, even if interrupts seem to be 173 * Host that it is *never* to interrupt us there, even if interrupts seem to be
151 * enabled. */ 174 * enabled.
175 */
152ENTRY(lguest_iret) 176ENTRY(lguest_iret)
153 pushl %eax 177 pushl %eax
154 movl 12(%esp), %eax 178 movl 12(%esp), %eax
155lguest_noirq_start: 179lguest_noirq_start:
156 /* Note the %ss: segment prefix here. Normal data accesses use the 180 /*
181 * Note the %ss: segment prefix here. Normal data accesses use the
157 * "ds" segment, but that will have already been restored for whatever 182 * "ds" segment, but that will have already been restored for whatever
158 * we're returning to (such as userspace): we can't trust it. The %ss: 183 * we're returning to (such as userspace): we can't trust it. The %ss:
159 * prefix makes sure we use the stack segment, which is still valid. */ 184 * prefix makes sure we use the stack segment, which is still valid.
185 */
160 movl %eax,%ss:lguest_data+LGUEST_DATA_irq_enabled 186 movl %eax,%ss:lguest_data+LGUEST_DATA_irq_enabled
161 popl %eax 187 popl %eax
162 iret 188 iret
diff --git a/drivers/lguest/core.c b/drivers/lguest/core.c
index a6974e9b8ebf..cd058bc903ff 100644
--- a/drivers/lguest/core.c
+++ b/drivers/lguest/core.c
@@ -1,6 +1,8 @@
1/*P:400 This contains run_guest() which actually calls into the Host<->Guest 1/*P:400
2 * This contains run_guest() which actually calls into the Host<->Guest
2 * Switcher and analyzes the return, such as determining if the Guest wants the 3 * Switcher and analyzes the return, such as determining if the Guest wants the
3 * Host to do something. This file also contains useful helper routines. :*/ 4 * Host to do something. This file also contains useful helper routines.
5:*/
4#include <linux/module.h> 6#include <linux/module.h>
5#include <linux/stringify.h> 7#include <linux/stringify.h>
6#include <linux/stddef.h> 8#include <linux/stddef.h>
@@ -24,7 +26,8 @@ static struct page **switcher_page;
24/* This One Big lock protects all inter-guest data structures. */ 26/* This One Big lock protects all inter-guest data structures. */
25DEFINE_MUTEX(lguest_lock); 27DEFINE_MUTEX(lguest_lock);
26 28
27/*H:010 We need to set up the Switcher at a high virtual address. Remember the 29/*H:010
30 * We need to set up the Switcher at a high virtual address. Remember the
28 * Switcher is a few hundred bytes of assembler code which actually changes the 31 * Switcher is a few hundred bytes of assembler code which actually changes the
29 * CPU to run the Guest, and then changes back to the Host when a trap or 32 * CPU to run the Guest, and then changes back to the Host when a trap or
30 * interrupt happens. 33 * interrupt happens.
@@ -33,7 +36,8 @@ DEFINE_MUTEX(lguest_lock);
33 * Host since it will be running as the switchover occurs. 36 * Host since it will be running as the switchover occurs.
34 * 37 *
35 * Trying to map memory at a particular address is an unusual thing to do, so 38 * Trying to map memory at a particular address is an unusual thing to do, so
36 * it's not a simple one-liner. */ 39 * it's not a simple one-liner.
40 */
37static __init int map_switcher(void) 41static __init int map_switcher(void)
38{ 42{
39 int i, err; 43 int i, err;
@@ -47,8 +51,10 @@ static __init int map_switcher(void)
47 * easy. 51 * easy.
48 */ 52 */
49 53
50 /* We allocate an array of struct page pointers. map_vm_area() wants 54 /*
51 * this, rather than just an array of pages. */ 55 * We allocate an array of struct page pointers. map_vm_area() wants
56 * this, rather than just an array of pages.
57 */
52 switcher_page = kmalloc(sizeof(switcher_page[0])*TOTAL_SWITCHER_PAGES, 58 switcher_page = kmalloc(sizeof(switcher_page[0])*TOTAL_SWITCHER_PAGES,
53 GFP_KERNEL); 59 GFP_KERNEL);
54 if (!switcher_page) { 60 if (!switcher_page) {
@@ -56,8 +62,10 @@ static __init int map_switcher(void)
56 goto out; 62 goto out;
57 } 63 }
58 64
59 /* Now we actually allocate the pages. The Guest will see these pages, 65 /*
60 * so we make sure they're zeroed. */ 66 * Now we actually allocate the pages. The Guest will see these pages,
67 * so we make sure they're zeroed.
68 */
61 for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) { 69 for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) {
62 unsigned long addr = get_zeroed_page(GFP_KERNEL); 70 unsigned long addr = get_zeroed_page(GFP_KERNEL);
63 if (!addr) { 71 if (!addr) {
@@ -67,19 +75,23 @@ static __init int map_switcher(void)
67 switcher_page[i] = virt_to_page(addr); 75 switcher_page[i] = virt_to_page(addr);
68 } 76 }
69 77
70 /* First we check that the Switcher won't overlap the fixmap area at 78 /*
79 * First we check that the Switcher won't overlap the fixmap area at
71 * the top of memory. It's currently nowhere near, but it could have 80 * the top of memory. It's currently nowhere near, but it could have
72 * very strange effects if it ever happened. */ 81 * very strange effects if it ever happened.
82 */
73 if (SWITCHER_ADDR + (TOTAL_SWITCHER_PAGES+1)*PAGE_SIZE > FIXADDR_START){ 83 if (SWITCHER_ADDR + (TOTAL_SWITCHER_PAGES+1)*PAGE_SIZE > FIXADDR_START){
74 err = -ENOMEM; 84 err = -ENOMEM;
75 printk("lguest: mapping switcher would thwack fixmap\n"); 85 printk("lguest: mapping switcher would thwack fixmap\n");
76 goto free_pages; 86 goto free_pages;
77 } 87 }
78 88
79 /* Now we reserve the "virtual memory area" we want: 0xFFC00000 89 /*
90 * Now we reserve the "virtual memory area" we want: 0xFFC00000
80 * (SWITCHER_ADDR). We might not get it in theory, but in practice 91 * (SWITCHER_ADDR). We might not get it in theory, but in practice
81 * it's worked so far. The end address needs +1 because __get_vm_area 92 * it's worked so far. The end address needs +1 because __get_vm_area
82 * allocates an extra guard page, so we need space for that. */ 93 * allocates an extra guard page, so we need space for that.
94 */
83 switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE, 95 switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE,
84 VM_ALLOC, SWITCHER_ADDR, SWITCHER_ADDR 96 VM_ALLOC, SWITCHER_ADDR, SWITCHER_ADDR
85 + (TOTAL_SWITCHER_PAGES+1) * PAGE_SIZE); 97 + (TOTAL_SWITCHER_PAGES+1) * PAGE_SIZE);
@@ -89,11 +101,13 @@ static __init int map_switcher(void)
89 goto free_pages; 101 goto free_pages;
90 } 102 }
91 103
92 /* This code actually sets up the pages we've allocated to appear at 104 /*
105 * This code actually sets up the pages we've allocated to appear at
93 * SWITCHER_ADDR. map_vm_area() takes the vma we allocated above, the 106 * SWITCHER_ADDR. map_vm_area() takes the vma we allocated above, the
94 * kind of pages we're mapping (kernel pages), and a pointer to our 107 * kind of pages we're mapping (kernel pages), and a pointer to our
95 * array of struct pages. It increments that pointer, but we don't 108 * array of struct pages. It increments that pointer, but we don't
96 * care. */ 109 * care.
110 */
97 pagep = switcher_page; 111 pagep = switcher_page;
98 err = map_vm_area(switcher_vma, PAGE_KERNEL_EXEC, &pagep); 112 err = map_vm_area(switcher_vma, PAGE_KERNEL_EXEC, &pagep);
99 if (err) { 113 if (err) {
@@ -101,8 +115,10 @@ static __init int map_switcher(void)
101 goto free_vma; 115 goto free_vma;
102 } 116 }
103 117
104 /* Now the Switcher is mapped at the right address, we can't fail! 118 /*
105 * Copy in the compiled-in Switcher code (from <arch>_switcher.S). */ 119 * Now the Switcher is mapped at the right address, we can't fail!
120 * Copy in the compiled-in Switcher code (from <arch>_switcher.S).
121 */
106 memcpy(switcher_vma->addr, start_switcher_text, 122 memcpy(switcher_vma->addr, start_switcher_text,
107 end_switcher_text - start_switcher_text); 123 end_switcher_text - start_switcher_text);
108 124
@@ -124,8 +140,7 @@ out:
124} 140}
125/*:*/ 141/*:*/
126 142
127/* Cleaning up the mapping when the module is unloaded is almost... 143/* Cleaning up the mapping when the module is unloaded is almost... too easy. */
128 * too easy. */
129static void unmap_switcher(void) 144static void unmap_switcher(void)
130{ 145{
131 unsigned int i; 146 unsigned int i;
@@ -151,16 +166,19 @@ static void unmap_switcher(void)
151 * But we can't trust the Guest: it might be trying to access the Launcher 166 * But we can't trust the Guest: it might be trying to access the Launcher
152 * code. We have to check that the range is below the pfn_limit the Launcher 167 * code. We have to check that the range is below the pfn_limit the Launcher
153 * gave us. We have to make sure that addr + len doesn't give us a false 168 * gave us. We have to make sure that addr + len doesn't give us a false
154 * positive by overflowing, too. */ 169 * positive by overflowing, too.
170 */
155bool lguest_address_ok(const struct lguest *lg, 171bool lguest_address_ok(const struct lguest *lg,
156 unsigned long addr, unsigned long len) 172 unsigned long addr, unsigned long len)
157{ 173{
158 return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr); 174 return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr);
159} 175}
160 176
161/* This routine copies memory from the Guest. Here we can see how useful the 177/*
178 * This routine copies memory from the Guest. Here we can see how useful the
162 * kill_lguest() routine we met in the Launcher can be: we return a random 179 * kill_lguest() routine we met in the Launcher can be: we return a random
163 * value (all zeroes) instead of needing to return an error. */ 180 * value (all zeroes) instead of needing to return an error.
181 */
164void __lgread(struct lg_cpu *cpu, void *b, unsigned long addr, unsigned bytes) 182void __lgread(struct lg_cpu *cpu, void *b, unsigned long addr, unsigned bytes)
165{ 183{
166 if (!lguest_address_ok(cpu->lg, addr, bytes) 184 if (!lguest_address_ok(cpu->lg, addr, bytes)
@@ -181,9 +199,11 @@ void __lgwrite(struct lg_cpu *cpu, unsigned long addr, const void *b,
181} 199}
182/*:*/ 200/*:*/
183 201
184/*H:030 Let's jump straight to the the main loop which runs the Guest. 202/*H:030
203 * Let's jump straight to the the main loop which runs the Guest.
185 * Remember, this is called by the Launcher reading /dev/lguest, and we keep 204 * Remember, this is called by the Launcher reading /dev/lguest, and we keep
186 * going around and around until something interesting happens. */ 205 * going around and around until something interesting happens.
206 */
187int run_guest(struct lg_cpu *cpu, unsigned long __user *user) 207int run_guest(struct lg_cpu *cpu, unsigned long __user *user)
188{ 208{
189 /* We stop running once the Guest is dead. */ 209 /* We stop running once the Guest is dead. */
@@ -195,8 +215,10 @@ int run_guest(struct lg_cpu *cpu, unsigned long __user *user)
195 if (cpu->hcall) 215 if (cpu->hcall)
196 do_hypercalls(cpu); 216 do_hypercalls(cpu);
197 217
198 /* It's possible the Guest did a NOTIFY hypercall to the 218 /*
199 * Launcher, in which case we return from the read() now. */ 219 * It's possible the Guest did a NOTIFY hypercall to the
220 * Launcher, in which case we return from the read() now.
221 */
200 if (cpu->pending_notify) { 222 if (cpu->pending_notify) {
201 if (!send_notify_to_eventfd(cpu)) { 223 if (!send_notify_to_eventfd(cpu)) {
202 if (put_user(cpu->pending_notify, user)) 224 if (put_user(cpu->pending_notify, user))
@@ -209,29 +231,39 @@ int run_guest(struct lg_cpu *cpu, unsigned long __user *user)
209 if (signal_pending(current)) 231 if (signal_pending(current))
210 return -ERESTARTSYS; 232 return -ERESTARTSYS;
211 233
212 /* Check if there are any interrupts which can be delivered now: 234 /*
235 * Check if there are any interrupts which can be delivered now:
213 * if so, this sets up the hander to be executed when we next 236 * if so, this sets up the hander to be executed when we next
214 * run the Guest. */ 237 * run the Guest.
238 */
215 irq = interrupt_pending(cpu, &more); 239 irq = interrupt_pending(cpu, &more);
216 if (irq < LGUEST_IRQS) 240 if (irq < LGUEST_IRQS)
217 try_deliver_interrupt(cpu, irq, more); 241 try_deliver_interrupt(cpu, irq, more);
218 242
219 /* All long-lived kernel loops need to check with this horrible 243 /*
244 * All long-lived kernel loops need to check with this horrible
220 * thing called the freezer. If the Host is trying to suspend, 245 * thing called the freezer. If the Host is trying to suspend,
221 * it stops us. */ 246 * it stops us.
247 */
222 try_to_freeze(); 248 try_to_freeze();
223 249
224 /* Just make absolutely sure the Guest is still alive. One of 250 /*
225 * those hypercalls could have been fatal, for example. */ 251 * Just make absolutely sure the Guest is still alive. One of
252 * those hypercalls could have been fatal, for example.
253 */
226 if (cpu->lg->dead) 254 if (cpu->lg->dead)
227 break; 255 break;
228 256
229 /* If the Guest asked to be stopped, we sleep. The Guest's 257 /*
230 * clock timer will wake us. */ 258 * If the Guest asked to be stopped, we sleep. The Guest's
259 * clock timer will wake us.
260 */
231 if (cpu->halted) { 261 if (cpu->halted) {
232 set_current_state(TASK_INTERRUPTIBLE); 262 set_current_state(TASK_INTERRUPTIBLE);
233 /* Just before we sleep, make sure no interrupt snuck in 263 /*
234 * which we should be doing. */ 264 * Just before we sleep, make sure no interrupt snuck in
265 * which we should be doing.
266 */
235 if (interrupt_pending(cpu, &more) < LGUEST_IRQS) 267 if (interrupt_pending(cpu, &more) < LGUEST_IRQS)
236 set_current_state(TASK_RUNNING); 268 set_current_state(TASK_RUNNING);
237 else 269 else
@@ -239,8 +271,10 @@ int run_guest(struct lg_cpu *cpu, unsigned long __user *user)
239 continue; 271 continue;
240 } 272 }
241 273
242 /* OK, now we're ready to jump into the Guest. First we put up 274 /*
243 * the "Do Not Disturb" sign: */ 275 * OK, now we're ready to jump into the Guest. First we put up
276 * the "Do Not Disturb" sign:
277 */
244 local_irq_disable(); 278 local_irq_disable();
245 279
246 /* Actually run the Guest until something happens. */ 280 /* Actually run the Guest until something happens. */
@@ -327,8 +361,10 @@ static void __exit fini(void)
327} 361}
328/*:*/ 362/*:*/
329 363
330/* The Host side of lguest can be a module. This is a nice way for people to 364/*
331 * play with it. */ 365 * The Host side of lguest can be a module. This is a nice way for people to
366 * play with it.
367 */
332module_init(init); 368module_init(init);
333module_exit(fini); 369module_exit(fini);
334MODULE_LICENSE("GPL"); 370MODULE_LICENSE("GPL");
diff --git a/drivers/lguest/hypercalls.c b/drivers/lguest/hypercalls.c
index c29ffa19cb74..787ab4bc09f0 100644
--- a/drivers/lguest/hypercalls.c
+++ b/drivers/lguest/hypercalls.c
@@ -1,8 +1,10 @@
1/*P:500 Just as userspace programs request kernel operations through a system 1/*P:500
2 * Just as userspace programs request kernel operations through a system
2 * call, the Guest requests Host operations through a "hypercall". You might 3 * call, the Guest requests Host operations through a "hypercall". You might
3 * notice this nomenclature doesn't really follow any logic, but the name has 4 * notice this nomenclature doesn't really follow any logic, but the name has
4 * been around for long enough that we're stuck with it. As you'd expect, this 5 * been around for long enough that we're stuck with it. As you'd expect, this
5 * code is basically a one big switch statement. :*/ 6 * code is basically a one big switch statement.
7:*/
6 8
7/* Copyright (C) 2006 Rusty Russell IBM Corporation 9/* Copyright (C) 2006 Rusty Russell IBM Corporation
8 10
@@ -28,30 +30,41 @@
28#include <asm/pgtable.h> 30#include <asm/pgtable.h>
29#include "lg.h" 31#include "lg.h"
30 32
31/*H:120 This is the core hypercall routine: where the Guest gets what it wants. 33/*H:120
32 * Or gets killed. Or, in the case of LHCALL_SHUTDOWN, both. */ 34 * This is the core hypercall routine: where the Guest gets what it wants.
35 * Or gets killed. Or, in the case of LHCALL_SHUTDOWN, both.
36 */
33static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args) 37static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args)
34{ 38{
35 switch (args->arg0) { 39 switch (args->arg0) {
36 case LHCALL_FLUSH_ASYNC: 40 case LHCALL_FLUSH_ASYNC:
37 /* This call does nothing, except by breaking out of the Guest 41 /*
38 * it makes us process all the asynchronous hypercalls. */ 42 * This call does nothing, except by breaking out of the Guest
43 * it makes us process all the asynchronous hypercalls.
44 */
39 break; 45 break;
40 case LHCALL_SEND_INTERRUPTS: 46 case LHCALL_SEND_INTERRUPTS:
41 /* This call does nothing too, but by breaking out of the Guest 47 /*
42 * it makes us process any pending interrupts. */ 48 * This call does nothing too, but by breaking out of the Guest
49 * it makes us process any pending interrupts.
50 */
43 break; 51 break;
44 case LHCALL_LGUEST_INIT: 52 case LHCALL_LGUEST_INIT:
45 /* You can't get here unless you're already initialized. Don't 53 /*
46 * do that. */ 54 * You can't get here unless you're already initialized. Don't
55 * do that.
56 */
47 kill_guest(cpu, "already have lguest_data"); 57 kill_guest(cpu, "already have lguest_data");
48 break; 58 break;
49 case LHCALL_SHUTDOWN: { 59 case LHCALL_SHUTDOWN: {
50 /* Shutdown is such a trivial hypercall that we do it in four
51 * lines right here. */
52 char msg[128]; 60 char msg[128];
53 /* If the lgread fails, it will call kill_guest() itself; the 61 /*
54 * kill_guest() with the message will be ignored. */ 62 * Shutdown is such a trivial hypercall that we do it in four
63 * lines right here.
64 *
65 * If the lgread fails, it will call kill_guest() itself; the
66 * kill_guest() with the message will be ignored.
67 */
55 __lgread(cpu, msg, args->arg1, sizeof(msg)); 68 __lgread(cpu, msg, args->arg1, sizeof(msg));
56 msg[sizeof(msg)-1] = '\0'; 69 msg[sizeof(msg)-1] = '\0';
57 kill_guest(cpu, "CRASH: %s", msg); 70 kill_guest(cpu, "CRASH: %s", msg);
@@ -60,16 +73,17 @@ static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args)
60 break; 73 break;
61 } 74 }
62 case LHCALL_FLUSH_TLB: 75 case LHCALL_FLUSH_TLB:
63 /* FLUSH_TLB comes in two flavors, depending on the 76 /* FLUSH_TLB comes in two flavors, depending on the argument: */
64 * argument: */
65 if (args->arg1) 77 if (args->arg1)
66 guest_pagetable_clear_all(cpu); 78 guest_pagetable_clear_all(cpu);
67 else 79 else
68 guest_pagetable_flush_user(cpu); 80 guest_pagetable_flush_user(cpu);
69 break; 81 break;
70 82
71 /* All these calls simply pass the arguments through to the right 83 /*
72 * routines. */ 84 * All these calls simply pass the arguments through to the right
85 * routines.
86 */
73 case LHCALL_NEW_PGTABLE: 87 case LHCALL_NEW_PGTABLE:
74 guest_new_pagetable(cpu, args->arg1); 88 guest_new_pagetable(cpu, args->arg1);
75 break; 89 break;
@@ -112,15 +126,16 @@ static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args)
112 kill_guest(cpu, "Bad hypercall %li\n", args->arg0); 126 kill_guest(cpu, "Bad hypercall %li\n", args->arg0);
113 } 127 }
114} 128}
115/*:*/
116 129
117/*H:124 Asynchronous hypercalls are easy: we just look in the array in the 130/*H:124
131 * Asynchronous hypercalls are easy: we just look in the array in the
118 * Guest's "struct lguest_data" to see if any new ones are marked "ready". 132 * Guest's "struct lguest_data" to see if any new ones are marked "ready".
119 * 133 *
120 * We are careful to do these in order: obviously we respect the order the 134 * We are careful to do these in order: obviously we respect the order the
121 * Guest put them in the ring, but we also promise the Guest that they will 135 * Guest put them in the ring, but we also promise the Guest that they will
122 * happen before any normal hypercall (which is why we check this before 136 * happen before any normal hypercall (which is why we check this before
123 * checking for a normal hcall). */ 137 * checking for a normal hcall).
138 */
124static void do_async_hcalls(struct lg_cpu *cpu) 139static void do_async_hcalls(struct lg_cpu *cpu)
125{ 140{
126 unsigned int i; 141 unsigned int i;
@@ -133,22 +148,28 @@ static void do_async_hcalls(struct lg_cpu *cpu)
133 /* We process "struct lguest_data"s hcalls[] ring once. */ 148 /* We process "struct lguest_data"s hcalls[] ring once. */
134 for (i = 0; i < ARRAY_SIZE(st); i++) { 149 for (i = 0; i < ARRAY_SIZE(st); i++) {
135 struct hcall_args args; 150 struct hcall_args args;
136 /* We remember where we were up to from last time. This makes 151 /*
152 * We remember where we were up to from last time. This makes
137 * sure that the hypercalls are done in the order the Guest 153 * sure that the hypercalls are done in the order the Guest
138 * places them in the ring. */ 154 * places them in the ring.
155 */
139 unsigned int n = cpu->next_hcall; 156 unsigned int n = cpu->next_hcall;
140 157
141 /* 0xFF means there's no call here (yet). */ 158 /* 0xFF means there's no call here (yet). */
142 if (st[n] == 0xFF) 159 if (st[n] == 0xFF)
143 break; 160 break;
144 161
145 /* OK, we have hypercall. Increment the "next_hcall" cursor, 162 /*
146 * and wrap back to 0 if we reach the end. */ 163 * OK, we have hypercall. Increment the "next_hcall" cursor,
164 * and wrap back to 0 if we reach the end.
165 */
147 if (++cpu->next_hcall == LHCALL_RING_SIZE) 166 if (++cpu->next_hcall == LHCALL_RING_SIZE)
148 cpu->next_hcall = 0; 167 cpu->next_hcall = 0;
149 168
150 /* Copy the hypercall arguments into a local copy of 169 /*
151 * the hcall_args struct. */ 170 * Copy the hypercall arguments into a local copy of the
171 * hcall_args struct.
172 */
152 if (copy_from_user(&args, &cpu->lg->lguest_data->hcalls[n], 173 if (copy_from_user(&args, &cpu->lg->lguest_data->hcalls[n],
153 sizeof(struct hcall_args))) { 174 sizeof(struct hcall_args))) {
154 kill_guest(cpu, "Fetching async hypercalls"); 175 kill_guest(cpu, "Fetching async hypercalls");
@@ -164,19 +185,25 @@ static void do_async_hcalls(struct lg_cpu *cpu)
164 break; 185 break;
165 } 186 }
166 187
167 /* Stop doing hypercalls if they want to notify the Launcher: 188 /*
168 * it needs to service this first. */ 189 * Stop doing hypercalls if they want to notify the Launcher:
190 * it needs to service this first.
191 */
169 if (cpu->pending_notify) 192 if (cpu->pending_notify)
170 break; 193 break;
171 } 194 }
172} 195}
173 196
174/* Last of all, we look at what happens first of all. The very first time the 197/*
175 * Guest makes a hypercall, we end up here to set things up: */ 198 * Last of all, we look at what happens first of all. The very first time the
199 * Guest makes a hypercall, we end up here to set things up:
200 */
176static void initialize(struct lg_cpu *cpu) 201static void initialize(struct lg_cpu *cpu)
177{ 202{
178 /* You can't do anything until you're initialized. The Guest knows the 203 /*
179 * rules, so we're unforgiving here. */ 204 * You can't do anything until you're initialized. The Guest knows the
205 * rules, so we're unforgiving here.
206 */
180 if (cpu->hcall->arg0 != LHCALL_LGUEST_INIT) { 207 if (cpu->hcall->arg0 != LHCALL_LGUEST_INIT) {
181 kill_guest(cpu, "hypercall %li before INIT", cpu->hcall->arg0); 208 kill_guest(cpu, "hypercall %li before INIT", cpu->hcall->arg0);
182 return; 209 return;
@@ -185,32 +212,40 @@ static void initialize(struct lg_cpu *cpu)
185 if (lguest_arch_init_hypercalls(cpu)) 212 if (lguest_arch_init_hypercalls(cpu))
186 kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); 213 kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data);
187 214
188 /* The Guest tells us where we're not to deliver interrupts by putting 215 /*
189 * the range of addresses into "struct lguest_data". */ 216 * The Guest tells us where we're not to deliver interrupts by putting
217 * the range of addresses into "struct lguest_data".
218 */
190 if (get_user(cpu->lg->noirq_start, &cpu->lg->lguest_data->noirq_start) 219 if (get_user(cpu->lg->noirq_start, &cpu->lg->lguest_data->noirq_start)
191 || get_user(cpu->lg->noirq_end, &cpu->lg->lguest_data->noirq_end)) 220 || get_user(cpu->lg->noirq_end, &cpu->lg->lguest_data->noirq_end))
192 kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); 221 kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data);
193 222
194 /* We write the current time into the Guest's data page once so it can 223 /*
195 * set its clock. */ 224 * We write the current time into the Guest's data page once so it can
225 * set its clock.
226 */
196 write_timestamp(cpu); 227 write_timestamp(cpu);
197 228
198 /* page_tables.c will also do some setup. */ 229 /* page_tables.c will also do some setup. */
199 page_table_guest_data_init(cpu); 230 page_table_guest_data_init(cpu);
200 231
201 /* This is the one case where the above accesses might have been the 232 /*
233 * This is the one case where the above accesses might have been the
202 * first write to a Guest page. This may have caused a copy-on-write 234 * first write to a Guest page. This may have caused a copy-on-write
203 * fault, but the old page might be (read-only) in the Guest 235 * fault, but the old page might be (read-only) in the Guest
204 * pagetable. */ 236 * pagetable.
237 */
205 guest_pagetable_clear_all(cpu); 238 guest_pagetable_clear_all(cpu);
206} 239}
207/*:*/ 240/*:*/
208 241
209/*M:013 If a Guest reads from a page (so creates a mapping) that it has never 242/*M:013
243 * If a Guest reads from a page (so creates a mapping) that it has never
210 * written to, and then the Launcher writes to it (ie. the output of a virtual 244 * written to, and then the Launcher writes to it (ie. the output of a virtual
211 * device), the Guest will still see the old page. In practice, this never 245 * device), the Guest will still see the old page. In practice, this never
212 * happens: why would the Guest read a page which it has never written to? But 246 * happens: why would the Guest read a page which it has never written to? But
213 * a similar scenario might one day bite us, so it's worth mentioning. :*/ 247 * a similar scenario might one day bite us, so it's worth mentioning.
248:*/
214 249
215/*H:100 250/*H:100
216 * Hypercalls 251 * Hypercalls
@@ -229,17 +264,22 @@ void do_hypercalls(struct lg_cpu *cpu)
229 return; 264 return;
230 } 265 }
231 266
232 /* The Guest has initialized. 267 /*
268 * The Guest has initialized.
233 * 269 *
234 * Look in the hypercall ring for the async hypercalls: */ 270 * Look in the hypercall ring for the async hypercalls:
271 */
235 do_async_hcalls(cpu); 272 do_async_hcalls(cpu);
236 273
237 /* If we stopped reading the hypercall ring because the Guest did a 274 /*
275 * If we stopped reading the hypercall ring because the Guest did a
238 * NOTIFY to the Launcher, we want to return now. Otherwise we do 276 * NOTIFY to the Launcher, we want to return now. Otherwise we do
239 * the hypercall. */ 277 * the hypercall.
278 */
240 if (!cpu->pending_notify) { 279 if (!cpu->pending_notify) {
241 do_hcall(cpu, cpu->hcall); 280 do_hcall(cpu, cpu->hcall);
242 /* Tricky point: we reset the hcall pointer to mark the 281 /*
282 * Tricky point: we reset the hcall pointer to mark the
243 * hypercall as "done". We use the hcall pointer rather than 283 * hypercall as "done". We use the hcall pointer rather than
244 * the trap number to indicate a hypercall is pending. 284 * the trap number to indicate a hypercall is pending.
245 * Normally it doesn't matter: the Guest will run again and 285 * Normally it doesn't matter: the Guest will run again and
@@ -248,13 +288,16 @@ void do_hypercalls(struct lg_cpu *cpu)
248 * However, if we are signalled or the Guest sends I/O to the 288 * However, if we are signalled or the Guest sends I/O to the
249 * Launcher, the run_guest() loop will exit without running the 289 * Launcher, the run_guest() loop will exit without running the
250 * Guest. When it comes back it would try to re-run the 290 * Guest. When it comes back it would try to re-run the
251 * hypercall. Finding that bug sucked. */ 291 * hypercall. Finding that bug sucked.
292 */
252 cpu->hcall = NULL; 293 cpu->hcall = NULL;
253 } 294 }
254} 295}
255 296
256/* This routine supplies the Guest with time: it's used for wallclock time at 297/*
257 * initial boot and as a rough time source if the TSC isn't available. */ 298 * This routine supplies the Guest with time: it's used for wallclock time at
299 * initial boot and as a rough time source if the TSC isn't available.
300 */
258void write_timestamp(struct lg_cpu *cpu) 301void write_timestamp(struct lg_cpu *cpu)
259{ 302{
260 struct timespec now; 303 struct timespec now;
diff --git a/drivers/lguest/interrupts_and_traps.c b/drivers/lguest/interrupts_and_traps.c
index 0e9067b0d507..18648180db02 100644
--- a/drivers/lguest/interrupts_and_traps.c
+++ b/drivers/lguest/interrupts_and_traps.c
@@ -1,4 +1,5 @@
1/*P:800 Interrupts (traps) are complicated enough to earn their own file. 1/*P:800
2 * Interrupts (traps) are complicated enough to earn their own file.
2 * There are three classes of interrupts: 3 * There are three classes of interrupts:
3 * 4 *
4 * 1) Real hardware interrupts which occur while we're running the Guest, 5 * 1) Real hardware interrupts which occur while we're running the Guest,
@@ -10,7 +11,8 @@
10 * just like real hardware would deliver them. Traps from the Guest can be set 11 * just like real hardware would deliver them. Traps from the Guest can be set
11 * up to go directly back into the Guest, but sometimes the Host wants to see 12 * up to go directly back into the Guest, but sometimes the Host wants to see
12 * them first, so we also have a way of "reflecting" them into the Guest as if 13 * them first, so we also have a way of "reflecting" them into the Guest as if
13 * they had been delivered to it directly. :*/ 14 * they had been delivered to it directly.
15:*/
14#include <linux/uaccess.h> 16#include <linux/uaccess.h>
15#include <linux/interrupt.h> 17#include <linux/interrupt.h>
16#include <linux/module.h> 18#include <linux/module.h>
@@ -26,8 +28,10 @@ static unsigned long idt_address(u32 lo, u32 hi)
26 return (lo & 0x0000FFFF) | (hi & 0xFFFF0000); 28 return (lo & 0x0000FFFF) | (hi & 0xFFFF0000);
27} 29}
28 30
29/* The "type" of the interrupt handler is a 4 bit field: we only support a 31/*
30 * couple of types. */ 32 * The "type" of the interrupt handler is a 4 bit field: we only support a
33 * couple of types.
34 */
31static int idt_type(u32 lo, u32 hi) 35static int idt_type(u32 lo, u32 hi)
32{ 36{
33 return (hi >> 8) & 0xF; 37 return (hi >> 8) & 0xF;
@@ -39,8 +43,10 @@ static bool idt_present(u32 lo, u32 hi)
39 return (hi & 0x8000); 43 return (hi & 0x8000);
40} 44}
41 45
42/* We need a helper to "push" a value onto the Guest's stack, since that's a 46/*
43 * big part of what delivering an interrupt does. */ 47 * We need a helper to "push" a value onto the Guest's stack, since that's a
48 * big part of what delivering an interrupt does.
49 */
44static void push_guest_stack(struct lg_cpu *cpu, unsigned long *gstack, u32 val) 50static void push_guest_stack(struct lg_cpu *cpu, unsigned long *gstack, u32 val)
45{ 51{
46 /* Stack grows upwards: move stack then write value. */ 52 /* Stack grows upwards: move stack then write value. */
@@ -48,7 +54,8 @@ static void push_guest_stack(struct lg_cpu *cpu, unsigned long *gstack, u32 val)
48 lgwrite(cpu, *gstack, u32, val); 54 lgwrite(cpu, *gstack, u32, val);
49} 55}
50 56
51/*H:210 The set_guest_interrupt() routine actually delivers the interrupt or 57/*H:210
58 * The set_guest_interrupt() routine actually delivers the interrupt or
52 * trap. The mechanics of delivering traps and interrupts to the Guest are the 59 * trap. The mechanics of delivering traps and interrupts to the Guest are the
53 * same, except some traps have an "error code" which gets pushed onto the 60 * same, except some traps have an "error code" which gets pushed onto the
54 * stack as well: the caller tells us if this is one. 61 * stack as well: the caller tells us if this is one.
@@ -59,7 +66,8 @@ static void push_guest_stack(struct lg_cpu *cpu, unsigned long *gstack, u32 val)
59 * 66 *
60 * We set up the stack just like the CPU does for a real interrupt, so it's 67 * We set up the stack just like the CPU does for a real interrupt, so it's
61 * identical for the Guest (and the standard "iret" instruction will undo 68 * identical for the Guest (and the standard "iret" instruction will undo
62 * it). */ 69 * it).
70 */
63static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi, 71static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi,
64 bool has_err) 72 bool has_err)
65{ 73{
@@ -67,20 +75,26 @@ static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi,
67 u32 eflags, ss, irq_enable; 75 u32 eflags, ss, irq_enable;
68 unsigned long virtstack; 76 unsigned long virtstack;
69 77
70 /* There are two cases for interrupts: one where the Guest is already 78 /*
79 * There are two cases for interrupts: one where the Guest is already
71 * in the kernel, and a more complex one where the Guest is in 80 * in the kernel, and a more complex one where the Guest is in
72 * userspace. We check the privilege level to find out. */ 81 * userspace. We check the privilege level to find out.
82 */
73 if ((cpu->regs->ss&0x3) != GUEST_PL) { 83 if ((cpu->regs->ss&0x3) != GUEST_PL) {
74 /* The Guest told us their kernel stack with the SET_STACK 84 /*
75 * hypercall: both the virtual address and the segment */ 85 * The Guest told us their kernel stack with the SET_STACK
86 * hypercall: both the virtual address and the segment.
87 */
76 virtstack = cpu->esp1; 88 virtstack = cpu->esp1;
77 ss = cpu->ss1; 89 ss = cpu->ss1;
78 90
79 origstack = gstack = guest_pa(cpu, virtstack); 91 origstack = gstack = guest_pa(cpu, virtstack);
80 /* We push the old stack segment and pointer onto the new 92 /*
93 * We push the old stack segment and pointer onto the new
81 * stack: when the Guest does an "iret" back from the interrupt 94 * stack: when the Guest does an "iret" back from the interrupt
82 * handler the CPU will notice they're dropping privilege 95 * handler the CPU will notice they're dropping privilege
83 * levels and expect these here. */ 96 * levels and expect these here.
97 */
84 push_guest_stack(cpu, &gstack, cpu->regs->ss); 98 push_guest_stack(cpu, &gstack, cpu->regs->ss);
85 push_guest_stack(cpu, &gstack, cpu->regs->esp); 99 push_guest_stack(cpu, &gstack, cpu->regs->esp);
86 } else { 100 } else {
@@ -91,18 +105,22 @@ static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi,
91 origstack = gstack = guest_pa(cpu, virtstack); 105 origstack = gstack = guest_pa(cpu, virtstack);
92 } 106 }
93 107
94 /* Remember that we never let the Guest actually disable interrupts, so 108 /*
109 * Remember that we never let the Guest actually disable interrupts, so
95 * the "Interrupt Flag" bit is always set. We copy that bit from the 110 * the "Interrupt Flag" bit is always set. We copy that bit from the
96 * Guest's "irq_enabled" field into the eflags word: we saw the Guest 111 * Guest's "irq_enabled" field into the eflags word: we saw the Guest
97 * copy it back in "lguest_iret". */ 112 * copy it back in "lguest_iret".
113 */
98 eflags = cpu->regs->eflags; 114 eflags = cpu->regs->eflags;
99 if (get_user(irq_enable, &cpu->lg->lguest_data->irq_enabled) == 0 115 if (get_user(irq_enable, &cpu->lg->lguest_data->irq_enabled) == 0
100 && !(irq_enable & X86_EFLAGS_IF)) 116 && !(irq_enable & X86_EFLAGS_IF))
101 eflags &= ~X86_EFLAGS_IF; 117 eflags &= ~X86_EFLAGS_IF;
102 118
103 /* An interrupt is expected to push three things on the stack: the old 119 /*
120 * An interrupt is expected to push three things on the stack: the old
104 * "eflags" word, the old code segment, and the old instruction 121 * "eflags" word, the old code segment, and the old instruction
105 * pointer. */ 122 * pointer.
123 */
106 push_guest_stack(cpu, &gstack, eflags); 124 push_guest_stack(cpu, &gstack, eflags);
107 push_guest_stack(cpu, &gstack, cpu->regs->cs); 125 push_guest_stack(cpu, &gstack, cpu->regs->cs);
108 push_guest_stack(cpu, &gstack, cpu->regs->eip); 126 push_guest_stack(cpu, &gstack, cpu->regs->eip);
@@ -111,15 +129,19 @@ static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi,
111 if (has_err) 129 if (has_err)
112 push_guest_stack(cpu, &gstack, cpu->regs->errcode); 130 push_guest_stack(cpu, &gstack, cpu->regs->errcode);
113 131
114 /* Now we've pushed all the old state, we change the stack, the code 132 /*
115 * segment and the address to execute. */ 133 * Now we've pushed all the old state, we change the stack, the code
134 * segment and the address to execute.
135 */
116 cpu->regs->ss = ss; 136 cpu->regs->ss = ss;
117 cpu->regs->esp = virtstack + (gstack - origstack); 137 cpu->regs->esp = virtstack + (gstack - origstack);
118 cpu->regs->cs = (__KERNEL_CS|GUEST_PL); 138 cpu->regs->cs = (__KERNEL_CS|GUEST_PL);
119 cpu->regs->eip = idt_address(lo, hi); 139 cpu->regs->eip = idt_address(lo, hi);
120 140
121 /* There are two kinds of interrupt handlers: 0xE is an "interrupt 141 /*
122 * gate" which expects interrupts to be disabled on entry. */ 142 * There are two kinds of interrupt handlers: 0xE is an "interrupt
143 * gate" which expects interrupts to be disabled on entry.
144 */
123 if (idt_type(lo, hi) == 0xE) 145 if (idt_type(lo, hi) == 0xE)
124 if (put_user(0, &cpu->lg->lguest_data->irq_enabled)) 146 if (put_user(0, &cpu->lg->lguest_data->irq_enabled))
125 kill_guest(cpu, "Disabling interrupts"); 147 kill_guest(cpu, "Disabling interrupts");
@@ -130,7 +152,8 @@ static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi,
130 * 152 *
131 * interrupt_pending() returns the first pending interrupt which isn't blocked 153 * interrupt_pending() returns the first pending interrupt which isn't blocked
132 * by the Guest. It is called before every entry to the Guest, and just before 154 * by the Guest. It is called before every entry to the Guest, and just before
133 * we go to sleep when the Guest has halted itself. */ 155 * we go to sleep when the Guest has halted itself.
156 */
134unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more) 157unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more)
135{ 158{
136 unsigned int irq; 159 unsigned int irq;
@@ -140,8 +163,10 @@ unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more)
140 if (!cpu->lg->lguest_data) 163 if (!cpu->lg->lguest_data)
141 return LGUEST_IRQS; 164 return LGUEST_IRQS;
142 165
143 /* Take our "irqs_pending" array and remove any interrupts the Guest 166 /*
144 * wants blocked: the result ends up in "blk". */ 167 * Take our "irqs_pending" array and remove any interrupts the Guest
168 * wants blocked: the result ends up in "blk".
169 */
145 if (copy_from_user(&blk, cpu->lg->lguest_data->blocked_interrupts, 170 if (copy_from_user(&blk, cpu->lg->lguest_data->blocked_interrupts,
146 sizeof(blk))) 171 sizeof(blk)))
147 return LGUEST_IRQS; 172 return LGUEST_IRQS;
@@ -154,16 +179,20 @@ unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more)
154 return irq; 179 return irq;
155} 180}
156 181
157/* This actually diverts the Guest to running an interrupt handler, once an 182/*
158 * interrupt has been identified by interrupt_pending(). */ 183 * This actually diverts the Guest to running an interrupt handler, once an
184 * interrupt has been identified by interrupt_pending().
185 */
159void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more) 186void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more)
160{ 187{
161 struct desc_struct *idt; 188 struct desc_struct *idt;
162 189
163 BUG_ON(irq >= LGUEST_IRQS); 190 BUG_ON(irq >= LGUEST_IRQS);
164 191
165 /* They may be in the middle of an iret, where they asked us never to 192 /*
166 * deliver interrupts. */ 193 * They may be in the middle of an iret, where they asked us never to
194 * deliver interrupts.
195 */
167 if (cpu->regs->eip >= cpu->lg->noirq_start && 196 if (cpu->regs->eip >= cpu->lg->noirq_start &&
168 (cpu->regs->eip < cpu->lg->noirq_end)) 197 (cpu->regs->eip < cpu->lg->noirq_end))
169 return; 198 return;
@@ -187,29 +216,37 @@ void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more)
187 } 216 }
188 } 217 }
189 218
190 /* Look at the IDT entry the Guest gave us for this interrupt. The 219 /*
220 * Look at the IDT entry the Guest gave us for this interrupt. The
191 * first 32 (FIRST_EXTERNAL_VECTOR) entries are for traps, so we skip 221 * first 32 (FIRST_EXTERNAL_VECTOR) entries are for traps, so we skip
192 * over them. */ 222 * over them.
223 */
193 idt = &cpu->arch.idt[FIRST_EXTERNAL_VECTOR+irq]; 224 idt = &cpu->arch.idt[FIRST_EXTERNAL_VECTOR+irq];
194 /* If they don't have a handler (yet?), we just ignore it */ 225 /* If they don't have a handler (yet?), we just ignore it */
195 if (idt_present(idt->a, idt->b)) { 226 if (idt_present(idt->a, idt->b)) {
196 /* OK, mark it no longer pending and deliver it. */ 227 /* OK, mark it no longer pending and deliver it. */
197 clear_bit(irq, cpu->irqs_pending); 228 clear_bit(irq, cpu->irqs_pending);
198 /* set_guest_interrupt() takes the interrupt descriptor and a 229 /*
230 * set_guest_interrupt() takes the interrupt descriptor and a
199 * flag to say whether this interrupt pushes an error code onto 231 * flag to say whether this interrupt pushes an error code onto
200 * the stack as well: virtual interrupts never do. */ 232 * the stack as well: virtual interrupts never do.
233 */
201 set_guest_interrupt(cpu, idt->a, idt->b, false); 234 set_guest_interrupt(cpu, idt->a, idt->b, false);
202 } 235 }
203 236
204 /* Every time we deliver an interrupt, we update the timestamp in the 237 /*
238 * Every time we deliver an interrupt, we update the timestamp in the
205 * Guest's lguest_data struct. It would be better for the Guest if we 239 * Guest's lguest_data struct. It would be better for the Guest if we
206 * did this more often, but it can actually be quite slow: doing it 240 * did this more often, but it can actually be quite slow: doing it
207 * here is a compromise which means at least it gets updated every 241 * here is a compromise which means at least it gets updated every
208 * timer interrupt. */ 242 * timer interrupt.
243 */
209 write_timestamp(cpu); 244 write_timestamp(cpu);
210 245
211 /* If there are no other interrupts we want to deliver, clear 246 /*
212 * the pending flag. */ 247 * If there are no other interrupts we want to deliver, clear
248 * the pending flag.
249 */
213 if (!more) 250 if (!more)
214 put_user(0, &cpu->lg->lguest_data->irq_pending); 251 put_user(0, &cpu->lg->lguest_data->irq_pending);
215} 252}
@@ -217,24 +254,29 @@ void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more)
217/* And this is the routine when we want to set an interrupt for the Guest. */ 254/* And this is the routine when we want to set an interrupt for the Guest. */
218void set_interrupt(struct lg_cpu *cpu, unsigned int irq) 255void set_interrupt(struct lg_cpu *cpu, unsigned int irq)
219{ 256{
220 /* Next time the Guest runs, the core code will see if it can deliver 257 /*
221 * this interrupt. */ 258 * Next time the Guest runs, the core code will see if it can deliver
259 * this interrupt.
260 */
222 set_bit(irq, cpu->irqs_pending); 261 set_bit(irq, cpu->irqs_pending);
223 262
224 /* Make sure it sees it; it might be asleep (eg. halted), or 263 /*
225 * running the Guest right now, in which case kick_process() 264 * Make sure it sees it; it might be asleep (eg. halted), or running
226 * will knock it out. */ 265 * the Guest right now, in which case kick_process() will knock it out.
266 */
227 if (!wake_up_process(cpu->tsk)) 267 if (!wake_up_process(cpu->tsk))
228 kick_process(cpu->tsk); 268 kick_process(cpu->tsk);
229} 269}
230/*:*/ 270/*:*/
231 271
232/* Linux uses trap 128 for system calls. Plan9 uses 64, and Ron Minnich sent 272/*
273 * Linux uses trap 128 for system calls. Plan9 uses 64, and Ron Minnich sent
233 * me a patch, so we support that too. It'd be a big step for lguest if half 274 * me a patch, so we support that too. It'd be a big step for lguest if half
234 * the Plan 9 user base were to start using it. 275 * the Plan 9 user base were to start using it.
235 * 276 *
236 * Actually now I think of it, it's possible that Ron *is* half the Plan 9 277 * Actually now I think of it, it's possible that Ron *is* half the Plan 9
237 * userbase. Oh well. */ 278 * userbase. Oh well.
279 */
238static bool could_be_syscall(unsigned int num) 280static bool could_be_syscall(unsigned int num)
239{ 281{
240 /* Normal Linux SYSCALL_VECTOR or reserved vector? */ 282 /* Normal Linux SYSCALL_VECTOR or reserved vector? */
@@ -274,9 +316,11 @@ void free_interrupts(void)
274 clear_bit(syscall_vector, used_vectors); 316 clear_bit(syscall_vector, used_vectors);
275} 317}
276 318
277/*H:220 Now we've got the routines to deliver interrupts, delivering traps like 319/*H:220
320 * Now we've got the routines to deliver interrupts, delivering traps like
278 * page fault is easy. The only trick is that Intel decided that some traps 321 * page fault is easy. The only trick is that Intel decided that some traps
279 * should have error codes: */ 322 * should have error codes:
323 */
280static bool has_err(unsigned int trap) 324static bool has_err(unsigned int trap)
281{ 325{
282 return (trap == 8 || (trap >= 10 && trap <= 14) || trap == 17); 326 return (trap == 8 || (trap >= 10 && trap <= 14) || trap == 17);
@@ -285,13 +329,17 @@ static bool has_err(unsigned int trap)
285/* deliver_trap() returns true if it could deliver the trap. */ 329/* deliver_trap() returns true if it could deliver the trap. */
286bool deliver_trap(struct lg_cpu *cpu, unsigned int num) 330bool deliver_trap(struct lg_cpu *cpu, unsigned int num)
287{ 331{
288 /* Trap numbers are always 8 bit, but we set an impossible trap number 332 /*
289 * for traps inside the Switcher, so check that here. */ 333 * Trap numbers are always 8 bit, but we set an impossible trap number
334 * for traps inside the Switcher, so check that here.
335 */
290 if (num >= ARRAY_SIZE(cpu->arch.idt)) 336 if (num >= ARRAY_SIZE(cpu->arch.idt))
291 return false; 337 return false;
292 338
293 /* Early on the Guest hasn't set the IDT entries (or maybe it put a 339 /*
294 * bogus one in): if we fail here, the Guest will be killed. */ 340 * Early on the Guest hasn't set the IDT entries (or maybe it put a
341 * bogus one in): if we fail here, the Guest will be killed.
342 */
295 if (!idt_present(cpu->arch.idt[num].a, cpu->arch.idt[num].b)) 343 if (!idt_present(cpu->arch.idt[num].a, cpu->arch.idt[num].b))
296 return false; 344 return false;
297 set_guest_interrupt(cpu, cpu->arch.idt[num].a, 345 set_guest_interrupt(cpu, cpu->arch.idt[num].a,
@@ -299,7 +347,8 @@ bool deliver_trap(struct lg_cpu *cpu, unsigned int num)
299 return true; 347 return true;
300} 348}
301 349
302/*H:250 Here's the hard part: returning to the Host every time a trap happens 350/*H:250
351 * Here's the hard part: returning to the Host every time a trap happens
303 * and then calling deliver_trap() and re-entering the Guest is slow. 352 * and then calling deliver_trap() and re-entering the Guest is slow.
304 * Particularly because Guest userspace system calls are traps (usually trap 353 * Particularly because Guest userspace system calls are traps (usually trap
305 * 128). 354 * 128).
@@ -311,69 +360,87 @@ bool deliver_trap(struct lg_cpu *cpu, unsigned int num)
311 * the other hypervisors would beat it up at lunchtime. 360 * the other hypervisors would beat it up at lunchtime.
312 * 361 *
313 * This routine indicates if a particular trap number could be delivered 362 * This routine indicates if a particular trap number could be delivered
314 * directly. */ 363 * directly.
364 */
315static bool direct_trap(unsigned int num) 365static bool direct_trap(unsigned int num)
316{ 366{
317 /* Hardware interrupts don't go to the Guest at all (except system 367 /*
318 * call). */ 368 * Hardware interrupts don't go to the Guest at all (except system
369 * call).
370 */
319 if (num >= FIRST_EXTERNAL_VECTOR && !could_be_syscall(num)) 371 if (num >= FIRST_EXTERNAL_VECTOR && !could_be_syscall(num))
320 return false; 372 return false;
321 373
322 /* The Host needs to see page faults (for shadow paging and to save the 374 /*
375 * The Host needs to see page faults (for shadow paging and to save the
323 * fault address), general protection faults (in/out emulation) and 376 * fault address), general protection faults (in/out emulation) and
324 * device not available (TS handling), invalid opcode fault (kvm hcall), 377 * device not available (TS handling), invalid opcode fault (kvm hcall),
325 * and of course, the hypercall trap. */ 378 * and of course, the hypercall trap.
379 */
326 return num != 14 && num != 13 && num != 7 && 380 return num != 14 && num != 13 && num != 7 &&
327 num != 6 && num != LGUEST_TRAP_ENTRY; 381 num != 6 && num != LGUEST_TRAP_ENTRY;
328} 382}
329/*:*/ 383/*:*/
330 384
331/*M:005 The Guest has the ability to turn its interrupt gates into trap gates, 385/*M:005
386 * The Guest has the ability to turn its interrupt gates into trap gates,
332 * if it is careful. The Host will let trap gates can go directly to the 387 * if it is careful. The Host will let trap gates can go directly to the
333 * Guest, but the Guest needs the interrupts atomically disabled for an 388 * Guest, but the Guest needs the interrupts atomically disabled for an
334 * interrupt gate. It can do this by pointing the trap gate at instructions 389 * interrupt gate. It can do this by pointing the trap gate at instructions
335 * within noirq_start and noirq_end, where it can safely disable interrupts. */ 390 * within noirq_start and noirq_end, where it can safely disable interrupts.
391 */
336 392
337/*M:006 The Guests do not use the sysenter (fast system call) instruction, 393/*M:006
394 * The Guests do not use the sysenter (fast system call) instruction,
338 * because it's hardcoded to enter privilege level 0 and so can't go direct. 395 * because it's hardcoded to enter privilege level 0 and so can't go direct.
339 * It's about twice as fast as the older "int 0x80" system call, so it might 396 * It's about twice as fast as the older "int 0x80" system call, so it might
340 * still be worthwhile to handle it in the Switcher and lcall down to the 397 * still be worthwhile to handle it in the Switcher and lcall down to the
341 * Guest. The sysenter semantics are hairy tho: search for that keyword in 398 * Guest. The sysenter semantics are hairy tho: search for that keyword in
342 * entry.S :*/ 399 * entry.S
400:*/
343 401
344/*H:260 When we make traps go directly into the Guest, we need to make sure 402/*H:260
403 * When we make traps go directly into the Guest, we need to make sure
345 * the kernel stack is valid (ie. mapped in the page tables). Otherwise, the 404 * the kernel stack is valid (ie. mapped in the page tables). Otherwise, the
346 * CPU trying to deliver the trap will fault while trying to push the interrupt 405 * CPU trying to deliver the trap will fault while trying to push the interrupt
347 * words on the stack: this is called a double fault, and it forces us to kill 406 * words on the stack: this is called a double fault, and it forces us to kill
348 * the Guest. 407 * the Guest.
349 * 408 *
350 * Which is deeply unfair, because (literally!) it wasn't the Guests' fault. */ 409 * Which is deeply unfair, because (literally!) it wasn't the Guests' fault.
410 */
351void pin_stack_pages(struct lg_cpu *cpu) 411void pin_stack_pages(struct lg_cpu *cpu)
352{ 412{
353 unsigned int i; 413 unsigned int i;
354 414
355 /* Depending on the CONFIG_4KSTACKS option, the Guest can have one or 415 /*
356 * two pages of stack space. */ 416 * Depending on the CONFIG_4KSTACKS option, the Guest can have one or
417 * two pages of stack space.
418 */
357 for (i = 0; i < cpu->lg->stack_pages; i++) 419 for (i = 0; i < cpu->lg->stack_pages; i++)
358 /* The stack grows *upwards*, so the address we're given is the 420 /*
421 * The stack grows *upwards*, so the address we're given is the
359 * start of the page after the kernel stack. Subtract one to 422 * start of the page after the kernel stack. Subtract one to
360 * get back onto the first stack page, and keep subtracting to 423 * get back onto the first stack page, and keep subtracting to
361 * get to the rest of the stack pages. */ 424 * get to the rest of the stack pages.
425 */
362 pin_page(cpu, cpu->esp1 - 1 - i * PAGE_SIZE); 426 pin_page(cpu, cpu->esp1 - 1 - i * PAGE_SIZE);
363} 427}
364 428
365/* Direct traps also mean that we need to know whenever the Guest wants to use 429/*
430 * Direct traps also mean that we need to know whenever the Guest wants to use
366 * a different kernel stack, so we can change the IDT entries to use that 431 * a different kernel stack, so we can change the IDT entries to use that
367 * stack. The IDT entries expect a virtual address, so unlike most addresses 432 * stack. The IDT entries expect a virtual address, so unlike most addresses
368 * the Guest gives us, the "esp" (stack pointer) value here is virtual, not 433 * the Guest gives us, the "esp" (stack pointer) value here is virtual, not
369 * physical. 434 * physical.
370 * 435 *
371 * In Linux each process has its own kernel stack, so this happens a lot: we 436 * In Linux each process has its own kernel stack, so this happens a lot: we
372 * change stacks on each context switch. */ 437 * change stacks on each context switch.
438 */
373void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages) 439void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages)
374{ 440{
375 /* You are not allowed have a stack segment with privilege level 0: bad 441 /*
376 * Guest! */ 442 * You're not allowed a stack segment with privilege level 0: bad Guest!
443 */
377 if ((seg & 0x3) != GUEST_PL) 444 if ((seg & 0x3) != GUEST_PL)
378 kill_guest(cpu, "bad stack segment %i", seg); 445 kill_guest(cpu, "bad stack segment %i", seg);
379 /* We only expect one or two stack pages. */ 446 /* We only expect one or two stack pages. */
@@ -387,11 +454,15 @@ void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages)
387 pin_stack_pages(cpu); 454 pin_stack_pages(cpu);
388} 455}
389 456
390/* All this reference to mapping stacks leads us neatly into the other complex 457/*
391 * part of the Host: page table handling. */ 458 * All this reference to mapping stacks leads us neatly into the other complex
459 * part of the Host: page table handling.
460 */
392 461
393/*H:235 This is the routine which actually checks the Guest's IDT entry and 462/*H:235
394 * transfers it into the entry in "struct lguest": */ 463 * This is the routine which actually checks the Guest's IDT entry and
464 * transfers it into the entry in "struct lguest":
465 */
395static void set_trap(struct lg_cpu *cpu, struct desc_struct *trap, 466static void set_trap(struct lg_cpu *cpu, struct desc_struct *trap,
396 unsigned int num, u32 lo, u32 hi) 467 unsigned int num, u32 lo, u32 hi)
397{ 468{
@@ -407,30 +478,38 @@ static void set_trap(struct lg_cpu *cpu, struct desc_struct *trap,
407 if (type != 0xE && type != 0xF) 478 if (type != 0xE && type != 0xF)
408 kill_guest(cpu, "bad IDT type %i", type); 479 kill_guest(cpu, "bad IDT type %i", type);
409 480
410 /* We only copy the handler address, present bit, privilege level and 481 /*
482 * We only copy the handler address, present bit, privilege level and
411 * type. The privilege level controls where the trap can be triggered 483 * type. The privilege level controls where the trap can be triggered
412 * manually with an "int" instruction. This is usually GUEST_PL, 484 * manually with an "int" instruction. This is usually GUEST_PL,
413 * except for system calls which userspace can use. */ 485 * except for system calls which userspace can use.
486 */
414 trap->a = ((__KERNEL_CS|GUEST_PL)<<16) | (lo&0x0000FFFF); 487 trap->a = ((__KERNEL_CS|GUEST_PL)<<16) | (lo&0x0000FFFF);
415 trap->b = (hi&0xFFFFEF00); 488 trap->b = (hi&0xFFFFEF00);
416} 489}
417 490
418/*H:230 While we're here, dealing with delivering traps and interrupts to the 491/*H:230
492 * While we're here, dealing with delivering traps and interrupts to the
419 * Guest, we might as well complete the picture: how the Guest tells us where 493 * Guest, we might as well complete the picture: how the Guest tells us where
420 * it wants them to go. This would be simple, except making traps fast 494 * it wants them to go. This would be simple, except making traps fast
421 * requires some tricks. 495 * requires some tricks.
422 * 496 *
423 * We saw the Guest setting Interrupt Descriptor Table (IDT) entries with the 497 * We saw the Guest setting Interrupt Descriptor Table (IDT) entries with the
424 * LHCALL_LOAD_IDT_ENTRY hypercall before: that comes here. */ 498 * LHCALL_LOAD_IDT_ENTRY hypercall before: that comes here.
499 */
425void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int num, u32 lo, u32 hi) 500void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int num, u32 lo, u32 hi)
426{ 501{
427 /* Guest never handles: NMI, doublefault, spurious interrupt or 502 /*
428 * hypercall. We ignore when it tries to set them. */ 503 * Guest never handles: NMI, doublefault, spurious interrupt or
504 * hypercall. We ignore when it tries to set them.
505 */
429 if (num == 2 || num == 8 || num == 15 || num == LGUEST_TRAP_ENTRY) 506 if (num == 2 || num == 8 || num == 15 || num == LGUEST_TRAP_ENTRY)
430 return; 507 return;
431 508
432 /* Mark the IDT as changed: next time the Guest runs we'll know we have 509 /*
433 * to copy this again. */ 510 * Mark the IDT as changed: next time the Guest runs we'll know we have
511 * to copy this again.
512 */
434 cpu->changed |= CHANGED_IDT; 513 cpu->changed |= CHANGED_IDT;
435 514
436 /* Check that the Guest doesn't try to step outside the bounds. */ 515 /* Check that the Guest doesn't try to step outside the bounds. */
@@ -440,9 +519,11 @@ void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int num, u32 lo, u32 hi)
440 set_trap(cpu, &cpu->arch.idt[num], num, lo, hi); 519 set_trap(cpu, &cpu->arch.idt[num], num, lo, hi);
441} 520}
442 521
443/* The default entry for each interrupt points into the Switcher routines which 522/*
523 * The default entry for each interrupt points into the Switcher routines which
444 * simply return to the Host. The run_guest() loop will then call 524 * simply return to the Host. The run_guest() loop will then call
445 * deliver_trap() to bounce it back into the Guest. */ 525 * deliver_trap() to bounce it back into the Guest.
526 */
446static void default_idt_entry(struct desc_struct *idt, 527static void default_idt_entry(struct desc_struct *idt,
447 int trap, 528 int trap,
448 const unsigned long handler, 529 const unsigned long handler,
@@ -451,13 +532,17 @@ static void default_idt_entry(struct desc_struct *idt,
451 /* A present interrupt gate. */ 532 /* A present interrupt gate. */
452 u32 flags = 0x8e00; 533 u32 flags = 0x8e00;
453 534
454 /* Set the privilege level on the entry for the hypercall: this allows 535 /*
455 * the Guest to use the "int" instruction to trigger it. */ 536 * Set the privilege level on the entry for the hypercall: this allows
537 * the Guest to use the "int" instruction to trigger it.
538 */
456 if (trap == LGUEST_TRAP_ENTRY) 539 if (trap == LGUEST_TRAP_ENTRY)
457 flags |= (GUEST_PL << 13); 540 flags |= (GUEST_PL << 13);
458 else if (base) 541 else if (base)
459 /* Copy priv. level from what Guest asked for. This allows 542 /*
460 * debug (int 3) traps from Guest userspace, for example. */ 543 * Copy privilege level from what Guest asked for. This allows
544 * debug (int 3) traps from Guest userspace, for example.
545 */
461 flags |= (base->b & 0x6000); 546 flags |= (base->b & 0x6000);
462 547
463 /* Now pack it into the IDT entry in its weird format. */ 548 /* Now pack it into the IDT entry in its weird format. */
@@ -475,16 +560,20 @@ void setup_default_idt_entries(struct lguest_ro_state *state,
475 default_idt_entry(&state->guest_idt[i], i, def[i], NULL); 560 default_idt_entry(&state->guest_idt[i], i, def[i], NULL);
476} 561}
477 562
478/*H:240 We don't use the IDT entries in the "struct lguest" directly, instead 563/*H:240
564 * We don't use the IDT entries in the "struct lguest" directly, instead
479 * we copy them into the IDT which we've set up for Guests on this CPU, just 565 * we copy them into the IDT which we've set up for Guests on this CPU, just
480 * before we run the Guest. This routine does that copy. */ 566 * before we run the Guest. This routine does that copy.
567 */
481void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt, 568void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt,
482 const unsigned long *def) 569 const unsigned long *def)
483{ 570{
484 unsigned int i; 571 unsigned int i;
485 572
486 /* We can simply copy the direct traps, otherwise we use the default 573 /*
487 * ones in the Switcher: they will return to the Host. */ 574 * We can simply copy the direct traps, otherwise we use the default
575 * ones in the Switcher: they will return to the Host.
576 */
488 for (i = 0; i < ARRAY_SIZE(cpu->arch.idt); i++) { 577 for (i = 0; i < ARRAY_SIZE(cpu->arch.idt); i++) {
489 const struct desc_struct *gidt = &cpu->arch.idt[i]; 578 const struct desc_struct *gidt = &cpu->arch.idt[i];
490 579
@@ -492,14 +581,16 @@ void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt,
492 if (!direct_trap(i)) 581 if (!direct_trap(i))
493 continue; 582 continue;
494 583
495 /* Only trap gates (type 15) can go direct to the Guest. 584 /*
585 * Only trap gates (type 15) can go direct to the Guest.
496 * Interrupt gates (type 14) disable interrupts as they are 586 * Interrupt gates (type 14) disable interrupts as they are
497 * entered, which we never let the Guest do. Not present 587 * entered, which we never let the Guest do. Not present
498 * entries (type 0x0) also can't go direct, of course. 588 * entries (type 0x0) also can't go direct, of course.
499 * 589 *
500 * If it can't go direct, we still need to copy the priv. level: 590 * If it can't go direct, we still need to copy the priv. level:
501 * they might want to give userspace access to a software 591 * they might want to give userspace access to a software
502 * interrupt. */ 592 * interrupt.
593 */
503 if (idt_type(gidt->a, gidt->b) == 0xF) 594 if (idt_type(gidt->a, gidt->b) == 0xF)
504 idt[i] = *gidt; 595 idt[i] = *gidt;
505 else 596 else
@@ -518,7 +609,8 @@ void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt,
518 * the next timer interrupt (in nanoseconds). We use the high-resolution timer 609 * the next timer interrupt (in nanoseconds). We use the high-resolution timer
519 * infrastructure to set a callback at that time. 610 * infrastructure to set a callback at that time.
520 * 611 *
521 * 0 means "turn off the clock". */ 612 * 0 means "turn off the clock".
613 */
522void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta) 614void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta)
523{ 615{
524 ktime_t expires; 616 ktime_t expires;
@@ -529,9 +621,11 @@ void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta)
529 return; 621 return;
530 } 622 }
531 623
532 /* We use wallclock time here, so the Guest might not be running for 624 /*
625 * We use wallclock time here, so the Guest might not be running for
533 * all the time between now and the timer interrupt it asked for. This 626 * all the time between now and the timer interrupt it asked for. This
534 * is almost always the right thing to do. */ 627 * is almost always the right thing to do.
628 */
535 expires = ktime_add_ns(ktime_get_real(), delta); 629 expires = ktime_add_ns(ktime_get_real(), delta);
536 hrtimer_start(&cpu->hrt, expires, HRTIMER_MODE_ABS); 630 hrtimer_start(&cpu->hrt, expires, HRTIMER_MODE_ABS);
537} 631}
diff --git a/drivers/lguest/lg.h b/drivers/lguest/lg.h
index 01c591923793..74c0db691b53 100644
--- a/drivers/lguest/lg.h
+++ b/drivers/lguest/lg.h
@@ -54,13 +54,13 @@ struct lg_cpu {
54 54
55 unsigned long pending_notify; /* pfn from LHCALL_NOTIFY */ 55 unsigned long pending_notify; /* pfn from LHCALL_NOTIFY */
56 56
57 /* At end of a page shared mapped over lguest_pages in guest. */ 57 /* At end of a page shared mapped over lguest_pages in guest. */
58 unsigned long regs_page; 58 unsigned long regs_page;
59 struct lguest_regs *regs; 59 struct lguest_regs *regs;
60 60
61 struct lguest_pages *last_pages; 61 struct lguest_pages *last_pages;
62 62
63 int cpu_pgd; /* which pgd this cpu is currently using */ 63 int cpu_pgd; /* Which pgd this cpu is currently using */
64 64
65 /* If a hypercall was asked for, this points to the arguments. */ 65 /* If a hypercall was asked for, this points to the arguments. */
66 struct hcall_args *hcall; 66 struct hcall_args *hcall;
@@ -96,8 +96,11 @@ struct lguest
96 unsigned int nr_cpus; 96 unsigned int nr_cpus;
97 97
98 u32 pfn_limit; 98 u32 pfn_limit;
99 /* This provides the offset to the base of guest-physical 99
100 * memory in the Launcher. */ 100 /*
101 * This provides the offset to the base of guest-physical memory in the
102 * Launcher.
103 */
101 void __user *mem_base; 104 void __user *mem_base;
102 unsigned long kernel_address; 105 unsigned long kernel_address;
103 106
@@ -122,11 +125,13 @@ bool lguest_address_ok(const struct lguest *lg,
122void __lgread(struct lg_cpu *, void *, unsigned long, unsigned); 125void __lgread(struct lg_cpu *, void *, unsigned long, unsigned);
123void __lgwrite(struct lg_cpu *, unsigned long, const void *, unsigned); 126void __lgwrite(struct lg_cpu *, unsigned long, const void *, unsigned);
124 127
125/*H:035 Using memory-copy operations like that is usually inconvient, so we 128/*H:035
129 * Using memory-copy operations like that is usually inconvient, so we
126 * have the following helper macros which read and write a specific type (often 130 * have the following helper macros which read and write a specific type (often
127 * an unsigned long). 131 * an unsigned long).
128 * 132 *
129 * This reads into a variable of the given type then returns that. */ 133 * This reads into a variable of the given type then returns that.
134 */
130#define lgread(cpu, addr, type) \ 135#define lgread(cpu, addr, type) \
131 ({ type _v; __lgread((cpu), &_v, (addr), sizeof(_v)); _v; }) 136 ({ type _v; __lgread((cpu), &_v, (addr), sizeof(_v)); _v; })
132 137
@@ -140,9 +145,11 @@ void __lgwrite(struct lg_cpu *, unsigned long, const void *, unsigned);
140 145
141int run_guest(struct lg_cpu *cpu, unsigned long __user *user); 146int run_guest(struct lg_cpu *cpu, unsigned long __user *user);
142 147
143/* Helper macros to obtain the first 12 or the last 20 bits, this is only the 148/*
149 * Helper macros to obtain the first 12 or the last 20 bits, this is only the
144 * first step in the migration to the kernel types. pte_pfn is already defined 150 * first step in the migration to the kernel types. pte_pfn is already defined
145 * in the kernel. */ 151 * in the kernel.
152 */
146#define pgd_flags(x) (pgd_val(x) & ~PAGE_MASK) 153#define pgd_flags(x) (pgd_val(x) & ~PAGE_MASK)
147#define pgd_pfn(x) (pgd_val(x) >> PAGE_SHIFT) 154#define pgd_pfn(x) (pgd_val(x) >> PAGE_SHIFT)
148#define pmd_flags(x) (pmd_val(x) & ~PAGE_MASK) 155#define pmd_flags(x) (pmd_val(x) & ~PAGE_MASK)
diff --git a/drivers/lguest/lguest_device.c b/drivers/lguest/lguest_device.c
index e082cdac88b4..cc000e79c3d1 100644
--- a/drivers/lguest/lguest_device.c
+++ b/drivers/lguest/lguest_device.c
@@ -1,10 +1,12 @@
1/*P:050 Lguest guests use a very simple method to describe devices. It's a 1/*P:050
2 * Lguest guests use a very simple method to describe devices. It's a
2 * series of device descriptors contained just above the top of normal Guest 3 * series of device descriptors contained just above the top of normal Guest
3 * memory. 4 * memory.
4 * 5 *
5 * We use the standard "virtio" device infrastructure, which provides us with a 6 * We use the standard "virtio" device infrastructure, which provides us with a
6 * console, a network and a block driver. Each one expects some configuration 7 * console, a network and a block driver. Each one expects some configuration
7 * information and a "virtqueue" or two to send and receive data. :*/ 8 * information and a "virtqueue" or two to send and receive data.
9:*/
8#include <linux/init.h> 10#include <linux/init.h>
9#include <linux/bootmem.h> 11#include <linux/bootmem.h>
10#include <linux/lguest_launcher.h> 12#include <linux/lguest_launcher.h>
@@ -20,8 +22,10 @@
20/* The pointer to our (page) of device descriptions. */ 22/* The pointer to our (page) of device descriptions. */
21static void *lguest_devices; 23static void *lguest_devices;
22 24
23/* For Guests, device memory can be used as normal memory, so we cast away the 25/*
24 * __iomem to quieten sparse. */ 26 * For Guests, device memory can be used as normal memory, so we cast away the
27 * __iomem to quieten sparse.
28 */
25static inline void *lguest_map(unsigned long phys_addr, unsigned long pages) 29static inline void *lguest_map(unsigned long phys_addr, unsigned long pages)
26{ 30{
27 return (__force void *)ioremap_cache(phys_addr, PAGE_SIZE*pages); 31 return (__force void *)ioremap_cache(phys_addr, PAGE_SIZE*pages);
@@ -32,8 +36,10 @@ static inline void lguest_unmap(void *addr)
32 iounmap((__force void __iomem *)addr); 36 iounmap((__force void __iomem *)addr);
33} 37}
34 38
35/*D:100 Each lguest device is just a virtio device plus a pointer to its entry 39/*D:100
36 * in the lguest_devices page. */ 40 * Each lguest device is just a virtio device plus a pointer to its entry
41 * in the lguest_devices page.
42 */
37struct lguest_device { 43struct lguest_device {
38 struct virtio_device vdev; 44 struct virtio_device vdev;
39 45
@@ -41,9 +47,11 @@ struct lguest_device {
41 struct lguest_device_desc *desc; 47 struct lguest_device_desc *desc;
42}; 48};
43 49
44/* Since the virtio infrastructure hands us a pointer to the virtio_device all 50/*
51 * Since the virtio infrastructure hands us a pointer to the virtio_device all
45 * the time, it helps to have a curt macro to get a pointer to the struct 52 * the time, it helps to have a curt macro to get a pointer to the struct
46 * lguest_device it's enclosed in. */ 53 * lguest_device it's enclosed in.
54 */
47#define to_lgdev(vd) container_of(vd, struct lguest_device, vdev) 55#define to_lgdev(vd) container_of(vd, struct lguest_device, vdev)
48 56
49/*D:130 57/*D:130
@@ -55,7 +63,8 @@ struct lguest_device {
55 * the driver will look at them during setup. 63 * the driver will look at them during setup.
56 * 64 *
57 * A convenient routine to return the device's virtqueue config array: 65 * A convenient routine to return the device's virtqueue config array:
58 * immediately after the descriptor. */ 66 * immediately after the descriptor.
67 */
59static struct lguest_vqconfig *lg_vq(const struct lguest_device_desc *desc) 68static struct lguest_vqconfig *lg_vq(const struct lguest_device_desc *desc)
60{ 69{
61 return (void *)(desc + 1); 70 return (void *)(desc + 1);
@@ -98,10 +107,12 @@ static u32 lg_get_features(struct virtio_device *vdev)
98 return features; 107 return features;
99} 108}
100 109
101/* The virtio core takes the features the Host offers, and copies the 110/*
102 * ones supported by the driver into the vdev->features array. Once 111 * The virtio core takes the features the Host offers, and copies the ones
103 * that's all sorted out, this routine is called so we can tell the 112 * supported by the driver into the vdev->features array. Once that's all
104 * Host which features we understand and accept. */ 113 * sorted out, this routine is called so we can tell the Host which features we
114 * understand and accept.
115 */
105static void lg_finalize_features(struct virtio_device *vdev) 116static void lg_finalize_features(struct virtio_device *vdev)
106{ 117{
107 unsigned int i, bits; 118 unsigned int i, bits;
@@ -112,10 +123,11 @@ static void lg_finalize_features(struct virtio_device *vdev)
112 /* Give virtio_ring a chance to accept features. */ 123 /* Give virtio_ring a chance to accept features. */
113 vring_transport_features(vdev); 124 vring_transport_features(vdev);
114 125
115 /* The vdev->feature array is a Linux bitmask: this isn't the 126 /*
116 * same as a the simple array of bits used by lguest devices 127 * The vdev->feature array is a Linux bitmask: this isn't the same as a
117 * for features. So we do this slow, manual conversion which is 128 * the simple array of bits used by lguest devices for features. So we
118 * completely general. */ 129 * do this slow, manual conversion which is completely general.
130 */
119 memset(out_features, 0, desc->feature_len); 131 memset(out_features, 0, desc->feature_len);
120 bits = min_t(unsigned, desc->feature_len, sizeof(vdev->features)) * 8; 132 bits = min_t(unsigned, desc->feature_len, sizeof(vdev->features)) * 8;
121 for (i = 0; i < bits; i++) { 133 for (i = 0; i < bits; i++) {
@@ -146,15 +158,19 @@ static void lg_set(struct virtio_device *vdev, unsigned int offset,
146 memcpy(lg_config(desc) + offset, buf, len); 158 memcpy(lg_config(desc) + offset, buf, len);
147} 159}
148 160
149/* The operations to get and set the status word just access the status field 161/*
150 * of the device descriptor. */ 162 * The operations to get and set the status word just access the status field
163 * of the device descriptor.
164 */
151static u8 lg_get_status(struct virtio_device *vdev) 165static u8 lg_get_status(struct virtio_device *vdev)
152{ 166{
153 return to_lgdev(vdev)->desc->status; 167 return to_lgdev(vdev)->desc->status;
154} 168}
155 169
156/* To notify on status updates, we (ab)use the NOTIFY hypercall, with the 170/*
157 * descriptor address of the device. A zero status means "reset". */ 171 * To notify on status updates, we (ab)use the NOTIFY hypercall, with the
172 * descriptor address of the device. A zero status means "reset".
173 */
158static void set_status(struct virtio_device *vdev, u8 status) 174static void set_status(struct virtio_device *vdev, u8 status)
159{ 175{
160 unsigned long offset = (void *)to_lgdev(vdev)->desc - lguest_devices; 176 unsigned long offset = (void *)to_lgdev(vdev)->desc - lguest_devices;
@@ -200,13 +216,17 @@ struct lguest_vq_info
200 void *pages; 216 void *pages;
201}; 217};
202 218
203/* When the virtio_ring code wants to prod the Host, it calls us here and we 219/*
220 * When the virtio_ring code wants to prod the Host, it calls us here and we
204 * make a hypercall. We hand the physical address of the virtqueue so the Host 221 * make a hypercall. We hand the physical address of the virtqueue so the Host
205 * knows which virtqueue we're talking about. */ 222 * knows which virtqueue we're talking about.
223 */
206static void lg_notify(struct virtqueue *vq) 224static void lg_notify(struct virtqueue *vq)
207{ 225{
208 /* We store our virtqueue information in the "priv" pointer of the 226 /*
209 * virtqueue structure. */ 227 * We store our virtqueue information in the "priv" pointer of the
228 * virtqueue structure.
229 */
210 struct lguest_vq_info *lvq = vq->priv; 230 struct lguest_vq_info *lvq = vq->priv;
211 231
212 kvm_hypercall1(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT); 232 kvm_hypercall1(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT);
@@ -215,7 +235,8 @@ static void lg_notify(struct virtqueue *vq)
215/* An extern declaration inside a C file is bad form. Don't do it. */ 235/* An extern declaration inside a C file is bad form. Don't do it. */
216extern void lguest_setup_irq(unsigned int irq); 236extern void lguest_setup_irq(unsigned int irq);
217 237
218/* This routine finds the first virtqueue described in the configuration of 238/*
239 * This routine finds the first virtqueue described in the configuration of
219 * this device and sets it up. 240 * this device and sets it up.
220 * 241 *
221 * This is kind of an ugly duckling. It'd be nicer to have a standard 242 * This is kind of an ugly duckling. It'd be nicer to have a standard
@@ -225,7 +246,8 @@ extern void lguest_setup_irq(unsigned int irq);
225 * simpler for the Host to simply tell us where the pages are. 246 * simpler for the Host to simply tell us where the pages are.
226 * 247 *
227 * So we provide drivers with a "find the Nth virtqueue and set it up" 248 * So we provide drivers with a "find the Nth virtqueue and set it up"
228 * function. */ 249 * function.
250 */
229static struct virtqueue *lg_find_vq(struct virtio_device *vdev, 251static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
230 unsigned index, 252 unsigned index,
231 void (*callback)(struct virtqueue *vq), 253 void (*callback)(struct virtqueue *vq),
@@ -244,9 +266,11 @@ static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
244 if (!lvq) 266 if (!lvq)
245 return ERR_PTR(-ENOMEM); 267 return ERR_PTR(-ENOMEM);
246 268
247 /* Make a copy of the "struct lguest_vqconfig" entry, which sits after 269 /*
270 * Make a copy of the "struct lguest_vqconfig" entry, which sits after
248 * the descriptor. We need a copy because the config space might not 271 * the descriptor. We need a copy because the config space might not
249 * be aligned correctly. */ 272 * be aligned correctly.
273 */
250 memcpy(&lvq->config, lg_vq(ldev->desc)+index, sizeof(lvq->config)); 274 memcpy(&lvq->config, lg_vq(ldev->desc)+index, sizeof(lvq->config));
251 275
252 printk("Mapping virtqueue %i addr %lx\n", index, 276 printk("Mapping virtqueue %i addr %lx\n", index,
@@ -261,8 +285,10 @@ static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
261 goto free_lvq; 285 goto free_lvq;
262 } 286 }
263 287
264 /* OK, tell virtio_ring.c to set up a virtqueue now we know its size 288 /*
265 * and we've got a pointer to its pages. */ 289 * OK, tell virtio_ring.c to set up a virtqueue now we know its size
290 * and we've got a pointer to its pages.
291 */
266 vq = vring_new_virtqueue(lvq->config.num, LGUEST_VRING_ALIGN, 292 vq = vring_new_virtqueue(lvq->config.num, LGUEST_VRING_ALIGN,
267 vdev, lvq->pages, lg_notify, callback, name); 293 vdev, lvq->pages, lg_notify, callback, name);
268 if (!vq) { 294 if (!vq) {
@@ -273,18 +299,23 @@ static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
273 /* Make sure the interrupt is allocated. */ 299 /* Make sure the interrupt is allocated. */
274 lguest_setup_irq(lvq->config.irq); 300 lguest_setup_irq(lvq->config.irq);
275 301
276 /* Tell the interrupt for this virtqueue to go to the virtio_ring 302 /*
277 * interrupt handler. */ 303 * Tell the interrupt for this virtqueue to go to the virtio_ring
278 /* FIXME: We used to have a flag for the Host to tell us we could use 304 * interrupt handler.
305 *
306 * FIXME: We used to have a flag for the Host to tell us we could use
279 * the interrupt as a source of randomness: it'd be nice to have that 307 * the interrupt as a source of randomness: it'd be nice to have that
280 * back.. */ 308 * back.
309 */
281 err = request_irq(lvq->config.irq, vring_interrupt, IRQF_SHARED, 310 err = request_irq(lvq->config.irq, vring_interrupt, IRQF_SHARED,
282 dev_name(&vdev->dev), vq); 311 dev_name(&vdev->dev), vq);
283 if (err) 312 if (err)
284 goto destroy_vring; 313 goto destroy_vring;
285 314
286 /* Last of all we hook up our 'struct lguest_vq_info" to the 315 /*
287 * virtqueue's priv pointer. */ 316 * Last of all we hook up our 'struct lguest_vq_info" to the
317 * virtqueue's priv pointer.
318 */
288 vq->priv = lvq; 319 vq->priv = lvq;
289 return vq; 320 return vq;
290 321
@@ -358,11 +389,14 @@ static struct virtio_config_ops lguest_config_ops = {
358 .del_vqs = lg_del_vqs, 389 .del_vqs = lg_del_vqs,
359}; 390};
360 391
361/* The root device for the lguest virtio devices. This makes them appear as 392/*
362 * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2. */ 393 * The root device for the lguest virtio devices. This makes them appear as
394 * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2.
395 */
363static struct device *lguest_root; 396static struct device *lguest_root;
364 397
365/*D:120 This is the core of the lguest bus: actually adding a new device. 398/*D:120
399 * This is the core of the lguest bus: actually adding a new device.
366 * It's a separate function because it's neater that way, and because an 400 * It's a separate function because it's neater that way, and because an
367 * earlier version of the code supported hotplug and unplug. They were removed 401 * earlier version of the code supported hotplug and unplug. They were removed
368 * early on because they were never used. 402 * early on because they were never used.
@@ -371,14 +405,14 @@ static struct device *lguest_root;
371 * 405 *
372 * It's worth reading this carefully: we start with a pointer to the new device 406 * It's worth reading this carefully: we start with a pointer to the new device
373 * descriptor in the "lguest_devices" page, and the offset into the device 407 * descriptor in the "lguest_devices" page, and the offset into the device
374 * descriptor page so we can uniquely identify it if things go badly wrong. */ 408 * descriptor page so we can uniquely identify it if things go badly wrong.
409 */
375static void add_lguest_device(struct lguest_device_desc *d, 410static void add_lguest_device(struct lguest_device_desc *d,
376 unsigned int offset) 411 unsigned int offset)
377{ 412{
378 struct lguest_device *ldev; 413 struct lguest_device *ldev;
379 414
380 /* Start with zeroed memory; Linux's device layer seems to count on 415 /* Start with zeroed memory; Linux's device layer counts on it. */
381 * it. */
382 ldev = kzalloc(sizeof(*ldev), GFP_KERNEL); 416 ldev = kzalloc(sizeof(*ldev), GFP_KERNEL);
383 if (!ldev) { 417 if (!ldev) {
384 printk(KERN_EMERG "Cannot allocate lguest dev %u type %u\n", 418 printk(KERN_EMERG "Cannot allocate lguest dev %u type %u\n",
@@ -390,15 +424,19 @@ static void add_lguest_device(struct lguest_device_desc *d,
390 ldev->vdev.dev.parent = lguest_root; 424 ldev->vdev.dev.parent = lguest_root;
391 /* We have a unique device index thanks to the dev_index counter. */ 425 /* We have a unique device index thanks to the dev_index counter. */
392 ldev->vdev.id.device = d->type; 426 ldev->vdev.id.device = d->type;
393 /* We have a simple set of routines for querying the device's 427 /*
394 * configuration information and setting its status. */ 428 * We have a simple set of routines for querying the device's
429 * configuration information and setting its status.
430 */
395 ldev->vdev.config = &lguest_config_ops; 431 ldev->vdev.config = &lguest_config_ops;
396 /* And we remember the device's descriptor for lguest_config_ops. */ 432 /* And we remember the device's descriptor for lguest_config_ops. */
397 ldev->desc = d; 433 ldev->desc = d;
398 434
399 /* register_virtio_device() sets up the generic fields for the struct 435 /*
436 * register_virtio_device() sets up the generic fields for the struct
400 * virtio_device and calls device_register(). This makes the bus 437 * virtio_device and calls device_register(). This makes the bus
401 * infrastructure look for a matching driver. */ 438 * infrastructure look for a matching driver.
439 */
402 if (register_virtio_device(&ldev->vdev) != 0) { 440 if (register_virtio_device(&ldev->vdev) != 0) {
403 printk(KERN_ERR "Failed to register lguest dev %u type %u\n", 441 printk(KERN_ERR "Failed to register lguest dev %u type %u\n",
404 offset, d->type); 442 offset, d->type);
@@ -406,8 +444,10 @@ static void add_lguest_device(struct lguest_device_desc *d,
406 } 444 }
407} 445}
408 446
409/*D:110 scan_devices() simply iterates through the device page. The type 0 is 447/*D:110
410 * reserved to mean "end of devices". */ 448 * scan_devices() simply iterates through the device page. The type 0 is
449 * reserved to mean "end of devices".
450 */
411static void scan_devices(void) 451static void scan_devices(void)
412{ 452{
413 unsigned int i; 453 unsigned int i;
@@ -426,7 +466,8 @@ static void scan_devices(void)
426 } 466 }
427} 467}
428 468
429/*D:105 Fairly early in boot, lguest_devices_init() is called to set up the 469/*D:105
470 * Fairly early in boot, lguest_devices_init() is called to set up the
430 * lguest device infrastructure. We check that we are a Guest by checking 471 * lguest device infrastructure. We check that we are a Guest by checking
431 * pv_info.name: there are other ways of checking, but this seems most 472 * pv_info.name: there are other ways of checking, but this seems most
432 * obvious to me. 473 * obvious to me.
@@ -437,7 +478,8 @@ static void scan_devices(void)
437 * correct sysfs incantation). 478 * correct sysfs incantation).
438 * 479 *
439 * Finally we call scan_devices() which adds all the devices found in the 480 * Finally we call scan_devices() which adds all the devices found in the
440 * lguest_devices page. */ 481 * lguest_devices page.
482 */
441static int __init lguest_devices_init(void) 483static int __init lguest_devices_init(void)
442{ 484{
443 if (strcmp(pv_info.name, "lguest") != 0) 485 if (strcmp(pv_info.name, "lguest") != 0)
@@ -456,11 +498,13 @@ static int __init lguest_devices_init(void)
456/* We do this after core stuff, but before the drivers. */ 498/* We do this after core stuff, but before the drivers. */
457postcore_initcall(lguest_devices_init); 499postcore_initcall(lguest_devices_init);
458 500
459/*D:150 At this point in the journey we used to now wade through the lguest 501/*D:150
502 * At this point in the journey we used to now wade through the lguest
460 * devices themselves: net, block and console. Since they're all now virtio 503 * devices themselves: net, block and console. Since they're all now virtio
461 * devices rather than lguest-specific, I've decided to ignore them. Mostly, 504 * devices rather than lguest-specific, I've decided to ignore them. Mostly,
462 * they're kind of boring. But this does mean you'll never experience the 505 * they're kind of boring. But this does mean you'll never experience the
463 * thrill of reading the forbidden love scene buried deep in the block driver. 506 * thrill of reading the forbidden love scene buried deep in the block driver.
464 * 507 *
465 * "make Launcher" beckons, where we answer questions like "Where do Guests 508 * "make Launcher" beckons, where we answer questions like "Where do Guests
466 * come from?", and "What do you do when someone asks for optimization?". */ 509 * come from?", and "What do you do when someone asks for optimization?".
510 */
diff --git a/drivers/lguest/lguest_user.c b/drivers/lguest/lguest_user.c
index 407722a8e0c4..7e92017103dc 100644
--- a/drivers/lguest/lguest_user.c
+++ b/drivers/lguest/lguest_user.c
@@ -1,8 +1,10 @@
1/*P:200 This contains all the /dev/lguest code, whereby the userspace launcher 1/*P:200
2 * This contains all the /dev/lguest code, whereby the userspace launcher
2 * controls and communicates with the Guest. For example, the first write will 3 * controls and communicates with the Guest. For example, the first write will
3 * tell us the Guest's memory layout, pagetable, entry point and kernel address 4 * tell us the Guest's memory layout, pagetable, entry point and kernel address
4 * offset. A read will run the Guest until something happens, such as a signal 5 * offset. A read will run the Guest until something happens, such as a signal
5 * or the Guest doing a NOTIFY out to the Launcher. :*/ 6 * or the Guest doing a NOTIFY out to the Launcher.
7:*/
6#include <linux/uaccess.h> 8#include <linux/uaccess.h>
7#include <linux/miscdevice.h> 9#include <linux/miscdevice.h>
8#include <linux/fs.h> 10#include <linux/fs.h>
@@ -37,8 +39,10 @@ static int add_eventfd(struct lguest *lg, unsigned long addr, int fd)
37 if (!addr) 39 if (!addr)
38 return -EINVAL; 40 return -EINVAL;
39 41
40 /* Replace the old array with the new one, carefully: others can 42 /*
41 * be accessing it at the same time */ 43 * Replace the old array with the new one, carefully: others can
44 * be accessing it at the same time.
45 */
42 new = kmalloc(sizeof(*new) + sizeof(new->map[0]) * (old->num + 1), 46 new = kmalloc(sizeof(*new) + sizeof(new->map[0]) * (old->num + 1),
43 GFP_KERNEL); 47 GFP_KERNEL);
44 if (!new) 48 if (!new)
@@ -61,8 +65,10 @@ static int add_eventfd(struct lguest *lg, unsigned long addr, int fd)
61 /* Now put new one in place. */ 65 /* Now put new one in place. */
62 rcu_assign_pointer(lg->eventfds, new); 66 rcu_assign_pointer(lg->eventfds, new);
63 67
64 /* We're not in a big hurry. Wait until noone's looking at old 68 /*
65 * version, then delete it. */ 69 * We're not in a big hurry. Wait until noone's looking at old
70 * version, then delete it.
71 */
66 synchronize_rcu(); 72 synchronize_rcu();
67 kfree(old); 73 kfree(old);
68 74
@@ -87,8 +93,10 @@ static int attach_eventfd(struct lguest *lg, const unsigned long __user *input)
87 return err; 93 return err;
88} 94}
89 95
90/*L:050 Sending an interrupt is done by writing LHREQ_IRQ and an interrupt 96/*L:050
91 * number to /dev/lguest. */ 97 * Sending an interrupt is done by writing LHREQ_IRQ and an interrupt
98 * number to /dev/lguest.
99 */
92static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input) 100static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input)
93{ 101{
94 unsigned long irq; 102 unsigned long irq;
@@ -102,8 +110,10 @@ static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input)
102 return 0; 110 return 0;
103} 111}
104 112
105/*L:040 Once our Guest is initialized, the Launcher makes it run by reading 113/*L:040
106 * from /dev/lguest. */ 114 * Once our Guest is initialized, the Launcher makes it run by reading
115 * from /dev/lguest.
116 */
107static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o) 117static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
108{ 118{
109 struct lguest *lg = file->private_data; 119 struct lguest *lg = file->private_data;
@@ -139,8 +149,10 @@ static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
139 return len; 149 return len;
140 } 150 }
141 151
142 /* If we returned from read() last time because the Guest sent I/O, 152 /*
143 * clear the flag. */ 153 * If we returned from read() last time because the Guest sent I/O,
154 * clear the flag.
155 */
144 if (cpu->pending_notify) 156 if (cpu->pending_notify)
145 cpu->pending_notify = 0; 157 cpu->pending_notify = 0;
146 158
@@ -148,8 +160,10 @@ static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
148 return run_guest(cpu, (unsigned long __user *)user); 160 return run_guest(cpu, (unsigned long __user *)user);
149} 161}
150 162
151/*L:025 This actually initializes a CPU. For the moment, a Guest is only 163/*L:025
152 * uniprocessor, so "id" is always 0. */ 164 * This actually initializes a CPU. For the moment, a Guest is only
165 * uniprocessor, so "id" is always 0.
166 */
153static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip) 167static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip)
154{ 168{
155 /* We have a limited number the number of CPUs in the lguest struct. */ 169 /* We have a limited number the number of CPUs in the lguest struct. */
@@ -164,8 +178,10 @@ static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip)
164 /* Each CPU has a timer it can set. */ 178 /* Each CPU has a timer it can set. */
165 init_clockdev(cpu); 179 init_clockdev(cpu);
166 180
167 /* We need a complete page for the Guest registers: they are accessible 181 /*
168 * to the Guest and we can only grant it access to whole pages. */ 182 * We need a complete page for the Guest registers: they are accessible
183 * to the Guest and we can only grant it access to whole pages.
184 */
169 cpu->regs_page = get_zeroed_page(GFP_KERNEL); 185 cpu->regs_page = get_zeroed_page(GFP_KERNEL);
170 if (!cpu->regs_page) 186 if (!cpu->regs_page)
171 return -ENOMEM; 187 return -ENOMEM;
@@ -173,29 +189,38 @@ static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip)
173 /* We actually put the registers at the bottom of the page. */ 189 /* We actually put the registers at the bottom of the page. */
174 cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs); 190 cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs);
175 191
176 /* Now we initialize the Guest's registers, handing it the start 192 /*
177 * address. */ 193 * Now we initialize the Guest's registers, handing it the start
194 * address.
195 */
178 lguest_arch_setup_regs(cpu, start_ip); 196 lguest_arch_setup_regs(cpu, start_ip);
179 197
180 /* We keep a pointer to the Launcher task (ie. current task) for when 198 /*
181 * other Guests want to wake this one (eg. console input). */ 199 * We keep a pointer to the Launcher task (ie. current task) for when
200 * other Guests want to wake this one (eg. console input).
201 */
182 cpu->tsk = current; 202 cpu->tsk = current;
183 203
184 /* We need to keep a pointer to the Launcher's memory map, because if 204 /*
205 * We need to keep a pointer to the Launcher's memory map, because if
185 * the Launcher dies we need to clean it up. If we don't keep a 206 * the Launcher dies we need to clean it up. If we don't keep a
186 * reference, it is destroyed before close() is called. */ 207 * reference, it is destroyed before close() is called.
208 */
187 cpu->mm = get_task_mm(cpu->tsk); 209 cpu->mm = get_task_mm(cpu->tsk);
188 210
189 /* We remember which CPU's pages this Guest used last, for optimization 211 /*
190 * when the same Guest runs on the same CPU twice. */ 212 * We remember which CPU's pages this Guest used last, for optimization
213 * when the same Guest runs on the same CPU twice.
214 */
191 cpu->last_pages = NULL; 215 cpu->last_pages = NULL;
192 216
193 /* No error == success. */ 217 /* No error == success. */
194 return 0; 218 return 0;
195} 219}
196 220
197/*L:020 The initialization write supplies 3 pointer sized (32 or 64 bit) 221/*L:020
198 * values (in addition to the LHREQ_INITIALIZE value). These are: 222 * The initialization write supplies 3 pointer sized (32 or 64 bit) values (in
223 * addition to the LHREQ_INITIALIZE value). These are:
199 * 224 *
200 * base: The start of the Guest-physical memory inside the Launcher memory. 225 * base: The start of the Guest-physical memory inside the Launcher memory.
201 * 226 *
@@ -207,14 +232,15 @@ static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip)
207 */ 232 */
208static int initialize(struct file *file, const unsigned long __user *input) 233static int initialize(struct file *file, const unsigned long __user *input)
209{ 234{
210 /* "struct lguest" contains everything we (the Host) know about a 235 /* "struct lguest" contains all we (the Host) know about a Guest. */
211 * Guest. */
212 struct lguest *lg; 236 struct lguest *lg;
213 int err; 237 int err;
214 unsigned long args[3]; 238 unsigned long args[3];
215 239
216 /* We grab the Big Lguest lock, which protects against multiple 240 /*
217 * simultaneous initializations. */ 241 * We grab the Big Lguest lock, which protects against multiple
242 * simultaneous initializations.
243 */
218 mutex_lock(&lguest_lock); 244 mutex_lock(&lguest_lock);
219 /* You can't initialize twice! Close the device and start again... */ 245 /* You can't initialize twice! Close the device and start again... */
220 if (file->private_data) { 246 if (file->private_data) {
@@ -249,8 +275,10 @@ static int initialize(struct file *file, const unsigned long __user *input)
249 if (err) 275 if (err)
250 goto free_eventfds; 276 goto free_eventfds;
251 277
252 /* Initialize the Guest's shadow page tables, using the toplevel 278 /*
253 * address the Launcher gave us. This allocates memory, so can fail. */ 279 * Initialize the Guest's shadow page tables, using the toplevel
280 * address the Launcher gave us. This allocates memory, so can fail.
281 */
254 err = init_guest_pagetable(lg); 282 err = init_guest_pagetable(lg);
255 if (err) 283 if (err)
256 goto free_regs; 284 goto free_regs;
@@ -275,7 +303,8 @@ unlock:
275 return err; 303 return err;
276} 304}
277 305
278/*L:010 The first operation the Launcher does must be a write. All writes 306/*L:010
307 * The first operation the Launcher does must be a write. All writes
279 * start with an unsigned long number: for the first write this must be 308 * start with an unsigned long number: for the first write this must be
280 * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use 309 * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use
281 * writes of other values to send interrupts. 310 * writes of other values to send interrupts.
@@ -283,12 +312,15 @@ unlock:
283 * Note that we overload the "offset" in the /dev/lguest file to indicate what 312 * Note that we overload the "offset" in the /dev/lguest file to indicate what
284 * CPU number we're dealing with. Currently this is always 0, since we only 313 * CPU number we're dealing with. Currently this is always 0, since we only
285 * support uniprocessor Guests, but you can see the beginnings of SMP support 314 * support uniprocessor Guests, but you can see the beginnings of SMP support
286 * here. */ 315 * here.
316 */
287static ssize_t write(struct file *file, const char __user *in, 317static ssize_t write(struct file *file, const char __user *in,
288 size_t size, loff_t *off) 318 size_t size, loff_t *off)
289{ 319{
290 /* Once the Guest is initialized, we hold the "struct lguest" in the 320 /*
291 * file private data. */ 321 * Once the Guest is initialized, we hold the "struct lguest" in the
322 * file private data.
323 */
292 struct lguest *lg = file->private_data; 324 struct lguest *lg = file->private_data;
293 const unsigned long __user *input = (const unsigned long __user *)in; 325 const unsigned long __user *input = (const unsigned long __user *)in;
294 unsigned long req; 326 unsigned long req;
@@ -323,13 +355,15 @@ static ssize_t write(struct file *file, const char __user *in,
323 } 355 }
324} 356}
325 357
326/*L:060 The final piece of interface code is the close() routine. It reverses 358/*L:060
359 * The final piece of interface code is the close() routine. It reverses
327 * everything done in initialize(). This is usually called because the 360 * everything done in initialize(). This is usually called because the
328 * Launcher exited. 361 * Launcher exited.
329 * 362 *
330 * Note that the close routine returns 0 or a negative error number: it can't 363 * Note that the close routine returns 0 or a negative error number: it can't
331 * really fail, but it can whine. I blame Sun for this wart, and K&R C for 364 * really fail, but it can whine. I blame Sun for this wart, and K&R C for
332 * letting them do it. :*/ 365 * letting them do it.
366:*/
333static int close(struct inode *inode, struct file *file) 367static int close(struct inode *inode, struct file *file)
334{ 368{
335 struct lguest *lg = file->private_data; 369 struct lguest *lg = file->private_data;
@@ -339,8 +373,10 @@ static int close(struct inode *inode, struct file *file)
339 if (!lg) 373 if (!lg)
340 return 0; 374 return 0;
341 375
342 /* We need the big lock, to protect from inter-guest I/O and other 376 /*
343 * Launchers initializing guests. */ 377 * We need the big lock, to protect from inter-guest I/O and other
378 * Launchers initializing guests.
379 */
344 mutex_lock(&lguest_lock); 380 mutex_lock(&lguest_lock);
345 381
346 /* Free up the shadow page tables for the Guest. */ 382 /* Free up the shadow page tables for the Guest. */
@@ -351,8 +387,10 @@ static int close(struct inode *inode, struct file *file)
351 hrtimer_cancel(&lg->cpus[i].hrt); 387 hrtimer_cancel(&lg->cpus[i].hrt);
352 /* We can free up the register page we allocated. */ 388 /* We can free up the register page we allocated. */
353 free_page(lg->cpus[i].regs_page); 389 free_page(lg->cpus[i].regs_page);
354 /* Now all the memory cleanups are done, it's safe to release 390 /*
355 * the Launcher's memory management structure. */ 391 * Now all the memory cleanups are done, it's safe to release
392 * the Launcher's memory management structure.
393 */
356 mmput(lg->cpus[i].mm); 394 mmput(lg->cpus[i].mm);
357 } 395 }
358 396
@@ -361,8 +399,10 @@ static int close(struct inode *inode, struct file *file)
361 eventfd_ctx_put(lg->eventfds->map[i].event); 399 eventfd_ctx_put(lg->eventfds->map[i].event);
362 kfree(lg->eventfds); 400 kfree(lg->eventfds);
363 401
364 /* If lg->dead doesn't contain an error code it will be NULL or a 402 /*
365 * kmalloc()ed string, either of which is ok to hand to kfree(). */ 403 * If lg->dead doesn't contain an error code it will be NULL or a
404 * kmalloc()ed string, either of which is ok to hand to kfree().
405 */
366 if (!IS_ERR(lg->dead)) 406 if (!IS_ERR(lg->dead))
367 kfree(lg->dead); 407 kfree(lg->dead);
368 /* Free the memory allocated to the lguest_struct */ 408 /* Free the memory allocated to the lguest_struct */
@@ -386,7 +426,8 @@ static int close(struct inode *inode, struct file *file)
386 * 426 *
387 * We begin our understanding with the Host kernel interface which the Launcher 427 * We begin our understanding with the Host kernel interface which the Launcher
388 * uses: reading and writing a character device called /dev/lguest. All the 428 * uses: reading and writing a character device called /dev/lguest. All the
389 * work happens in the read(), write() and close() routines: */ 429 * work happens in the read(), write() and close() routines:
430 */
390static struct file_operations lguest_fops = { 431static struct file_operations lguest_fops = {
391 .owner = THIS_MODULE, 432 .owner = THIS_MODULE,
392 .release = close, 433 .release = close,
@@ -394,8 +435,10 @@ static struct file_operations lguest_fops = {
394 .read = read, 435 .read = read,
395}; 436};
396 437
397/* This is a textbook example of a "misc" character device. Populate a "struct 438/*
398 * miscdevice" and register it with misc_register(). */ 439 * This is a textbook example of a "misc" character device. Populate a "struct
440 * miscdevice" and register it with misc_register().
441 */
399static struct miscdevice lguest_dev = { 442static struct miscdevice lguest_dev = {
400 .minor = MISC_DYNAMIC_MINOR, 443 .minor = MISC_DYNAMIC_MINOR,
401 .name = "lguest", 444 .name = "lguest",
diff --git a/drivers/lguest/page_tables.c b/drivers/lguest/page_tables.c
index a6fe1abda240..3da902e4b4cb 100644
--- a/drivers/lguest/page_tables.c
+++ b/drivers/lguest/page_tables.c
@@ -1,9 +1,11 @@
1/*P:700 The pagetable code, on the other hand, still shows the scars of 1/*P:700
2 * The pagetable code, on the other hand, still shows the scars of
2 * previous encounters. It's functional, and as neat as it can be in the 3 * previous encounters. It's functional, and as neat as it can be in the
3 * circumstances, but be wary, for these things are subtle and break easily. 4 * circumstances, but be wary, for these things are subtle and break easily.
4 * The Guest provides a virtual to physical mapping, but we can neither trust 5 * The Guest provides a virtual to physical mapping, but we can neither trust
5 * it nor use it: we verify and convert it here then point the CPU to the 6 * it nor use it: we verify and convert it here then point the CPU to the
6 * converted Guest pages when running the Guest. :*/ 7 * converted Guest pages when running the Guest.
8:*/
7 9
8/* Copyright (C) Rusty Russell IBM Corporation 2006. 10/* Copyright (C) Rusty Russell IBM Corporation 2006.
9 * GPL v2 and any later version */ 11 * GPL v2 and any later version */
@@ -17,10 +19,12 @@
17#include <asm/bootparam.h> 19#include <asm/bootparam.h>
18#include "lg.h" 20#include "lg.h"
19 21
20/*M:008 We hold reference to pages, which prevents them from being swapped. 22/*M:008
23 * We hold reference to pages, which prevents them from being swapped.
21 * It'd be nice to have a callback in the "struct mm_struct" when Linux wants 24 * It'd be nice to have a callback in the "struct mm_struct" when Linux wants
22 * to swap out. If we had this, and a shrinker callback to trim PTE pages, we 25 * to swap out. If we had this, and a shrinker callback to trim PTE pages, we
23 * could probably consider launching Guests as non-root. :*/ 26 * could probably consider launching Guests as non-root.
27:*/
24 28
25/*H:300 29/*H:300
26 * The Page Table Code 30 * The Page Table Code
@@ -45,16 +49,19 @@
45 * (v) Flushing (throwing away) page tables, 49 * (v) Flushing (throwing away) page tables,
46 * (vi) Mapping the Switcher when the Guest is about to run, 50 * (vi) Mapping the Switcher when the Guest is about to run,
47 * (vii) Setting up the page tables initially. 51 * (vii) Setting up the page tables initially.
48 :*/ 52:*/
49 53
50 54/*
51/* 1024 entries in a page table page maps 1024 pages: 4MB. The Switcher is 55 * 1024 entries in a page table page maps 1024 pages: 4MB. The Switcher is
52 * conveniently placed at the top 4MB, so it uses a separate, complete PTE 56 * conveniently placed at the top 4MB, so it uses a separate, complete PTE
53 * page. */ 57 * page.
58 */
54#define SWITCHER_PGD_INDEX (PTRS_PER_PGD - 1) 59#define SWITCHER_PGD_INDEX (PTRS_PER_PGD - 1)
55 60
56/* For PAE we need the PMD index as well. We use the last 2MB, so we 61/*
57 * will need the last pmd entry of the last pmd page. */ 62 * For PAE we need the PMD index as well. We use the last 2MB, so we
63 * will need the last pmd entry of the last pmd page.
64 */
58#ifdef CONFIG_X86_PAE 65#ifdef CONFIG_X86_PAE
59#define SWITCHER_PMD_INDEX (PTRS_PER_PMD - 1) 66#define SWITCHER_PMD_INDEX (PTRS_PER_PMD - 1)
60#define RESERVE_MEM 2U 67#define RESERVE_MEM 2U
@@ -64,13 +71,16 @@
64#define CHECK_GPGD_MASK _PAGE_TABLE 71#define CHECK_GPGD_MASK _PAGE_TABLE
65#endif 72#endif
66 73
67/* We actually need a separate PTE page for each CPU. Remember that after the 74/*
75 * We actually need a separate PTE page for each CPU. Remember that after the
68 * Switcher code itself comes two pages for each CPU, and we don't want this 76 * Switcher code itself comes two pages for each CPU, and we don't want this
69 * CPU's guest to see the pages of any other CPU. */ 77 * CPU's guest to see the pages of any other CPU.
78 */
70static DEFINE_PER_CPU(pte_t *, switcher_pte_pages); 79static DEFINE_PER_CPU(pte_t *, switcher_pte_pages);
71#define switcher_pte_page(cpu) per_cpu(switcher_pte_pages, cpu) 80#define switcher_pte_page(cpu) per_cpu(switcher_pte_pages, cpu)
72 81
73/*H:320 The page table code is curly enough to need helper functions to keep it 82/*H:320
83 * The page table code is curly enough to need helper functions to keep it
74 * clear and clean. 84 * clear and clean.
75 * 85 *
76 * There are two functions which return pointers to the shadow (aka "real") 86 * There are two functions which return pointers to the shadow (aka "real")
@@ -79,7 +89,8 @@ static DEFINE_PER_CPU(pte_t *, switcher_pte_pages);
79 * spgd_addr() takes the virtual address and returns a pointer to the top-level 89 * spgd_addr() takes the virtual address and returns a pointer to the top-level
80 * page directory entry (PGD) for that address. Since we keep track of several 90 * page directory entry (PGD) for that address. Since we keep track of several
81 * page tables, the "i" argument tells us which one we're interested in (it's 91 * page tables, the "i" argument tells us which one we're interested in (it's
82 * usually the current one). */ 92 * usually the current one).
93 */
83static pgd_t *spgd_addr(struct lg_cpu *cpu, u32 i, unsigned long vaddr) 94static pgd_t *spgd_addr(struct lg_cpu *cpu, u32 i, unsigned long vaddr)
84{ 95{
85 unsigned int index = pgd_index(vaddr); 96 unsigned int index = pgd_index(vaddr);
@@ -96,9 +107,11 @@ static pgd_t *spgd_addr(struct lg_cpu *cpu, u32 i, unsigned long vaddr)
96} 107}
97 108
98#ifdef CONFIG_X86_PAE 109#ifdef CONFIG_X86_PAE
99/* This routine then takes the PGD entry given above, which contains the 110/*
111 * This routine then takes the PGD entry given above, which contains the
100 * address of the PMD page. It then returns a pointer to the PMD entry for the 112 * address of the PMD page. It then returns a pointer to the PMD entry for the
101 * given address. */ 113 * given address.
114 */
102static pmd_t *spmd_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr) 115static pmd_t *spmd_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr)
103{ 116{
104 unsigned int index = pmd_index(vaddr); 117 unsigned int index = pmd_index(vaddr);
@@ -119,9 +132,11 @@ static pmd_t *spmd_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr)
119} 132}
120#endif 133#endif
121 134
122/* This routine then takes the page directory entry returned above, which 135/*
136 * This routine then takes the page directory entry returned above, which
123 * contains the address of the page table entry (PTE) page. It then returns a 137 * contains the address of the page table entry (PTE) page. It then returns a
124 * pointer to the PTE entry for the given address. */ 138 * pointer to the PTE entry for the given address.
139 */
125static pte_t *spte_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr) 140static pte_t *spte_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr)
126{ 141{
127#ifdef CONFIG_X86_PAE 142#ifdef CONFIG_X86_PAE
@@ -139,8 +154,10 @@ static pte_t *spte_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr)
139 return &page[pte_index(vaddr)]; 154 return &page[pte_index(vaddr)];
140} 155}
141 156
142/* These two functions just like the above two, except they access the Guest 157/*
143 * page tables. Hence they return a Guest address. */ 158 * These two functions just like the above two, except they access the Guest
159 * page tables. Hence they return a Guest address.
160 */
144static unsigned long gpgd_addr(struct lg_cpu *cpu, unsigned long vaddr) 161static unsigned long gpgd_addr(struct lg_cpu *cpu, unsigned long vaddr)
145{ 162{
146 unsigned int index = vaddr >> (PGDIR_SHIFT); 163 unsigned int index = vaddr >> (PGDIR_SHIFT);
@@ -175,17 +192,21 @@ static unsigned long gpte_addr(struct lg_cpu *cpu,
175#endif 192#endif
176/*:*/ 193/*:*/
177 194
178/*M:014 get_pfn is slow: we could probably try to grab batches of pages here as 195/*M:014
179 * an optimization (ie. pre-faulting). :*/ 196 * get_pfn is slow: we could probably try to grab batches of pages here as
197 * an optimization (ie. pre-faulting).
198:*/
180 199
181/*H:350 This routine takes a page number given by the Guest and converts it to 200/*H:350
201 * This routine takes a page number given by the Guest and converts it to
182 * an actual, physical page number. It can fail for several reasons: the 202 * an actual, physical page number. It can fail for several reasons: the
183 * virtual address might not be mapped by the Launcher, the write flag is set 203 * virtual address might not be mapped by the Launcher, the write flag is set
184 * and the page is read-only, or the write flag was set and the page was 204 * and the page is read-only, or the write flag was set and the page was
185 * shared so had to be copied, but we ran out of memory. 205 * shared so had to be copied, but we ran out of memory.
186 * 206 *
187 * This holds a reference to the page, so release_pte() is careful to put that 207 * This holds a reference to the page, so release_pte() is careful to put that
188 * back. */ 208 * back.
209 */
189static unsigned long get_pfn(unsigned long virtpfn, int write) 210static unsigned long get_pfn(unsigned long virtpfn, int write)
190{ 211{
191 struct page *page; 212 struct page *page;
@@ -198,33 +219,41 @@ static unsigned long get_pfn(unsigned long virtpfn, int write)
198 return -1UL; 219 return -1UL;
199} 220}
200 221
201/*H:340 Converting a Guest page table entry to a shadow (ie. real) page table 222/*H:340
223 * Converting a Guest page table entry to a shadow (ie. real) page table
202 * entry can be a little tricky. The flags are (almost) the same, but the 224 * entry can be a little tricky. The flags are (almost) the same, but the
203 * Guest PTE contains a virtual page number: the CPU needs the real page 225 * Guest PTE contains a virtual page number: the CPU needs the real page
204 * number. */ 226 * number.
227 */
205static pte_t gpte_to_spte(struct lg_cpu *cpu, pte_t gpte, int write) 228static pte_t gpte_to_spte(struct lg_cpu *cpu, pte_t gpte, int write)
206{ 229{
207 unsigned long pfn, base, flags; 230 unsigned long pfn, base, flags;
208 231
209 /* The Guest sets the global flag, because it thinks that it is using 232 /*
233 * The Guest sets the global flag, because it thinks that it is using
210 * PGE. We only told it to use PGE so it would tell us whether it was 234 * PGE. We only told it to use PGE so it would tell us whether it was
211 * flushing a kernel mapping or a userspace mapping. We don't actually 235 * flushing a kernel mapping or a userspace mapping. We don't actually
212 * use the global bit, so throw it away. */ 236 * use the global bit, so throw it away.
237 */
213 flags = (pte_flags(gpte) & ~_PAGE_GLOBAL); 238 flags = (pte_flags(gpte) & ~_PAGE_GLOBAL);
214 239
215 /* The Guest's pages are offset inside the Launcher. */ 240 /* The Guest's pages are offset inside the Launcher. */
216 base = (unsigned long)cpu->lg->mem_base / PAGE_SIZE; 241 base = (unsigned long)cpu->lg->mem_base / PAGE_SIZE;
217 242
218 /* We need a temporary "unsigned long" variable to hold the answer from 243 /*
244 * We need a temporary "unsigned long" variable to hold the answer from
219 * get_pfn(), because it returns 0xFFFFFFFF on failure, which wouldn't 245 * get_pfn(), because it returns 0xFFFFFFFF on failure, which wouldn't
220 * fit in spte.pfn. get_pfn() finds the real physical number of the 246 * fit in spte.pfn. get_pfn() finds the real physical number of the
221 * page, given the virtual number. */ 247 * page, given the virtual number.
248 */
222 pfn = get_pfn(base + pte_pfn(gpte), write); 249 pfn = get_pfn(base + pte_pfn(gpte), write);
223 if (pfn == -1UL) { 250 if (pfn == -1UL) {
224 kill_guest(cpu, "failed to get page %lu", pte_pfn(gpte)); 251 kill_guest(cpu, "failed to get page %lu", pte_pfn(gpte));
225 /* When we destroy the Guest, we'll go through the shadow page 252 /*
253 * When we destroy the Guest, we'll go through the shadow page
226 * tables and release_pte() them. Make sure we don't think 254 * tables and release_pte() them. Make sure we don't think
227 * this one is valid! */ 255 * this one is valid!
256 */
228 flags = 0; 257 flags = 0;
229 } 258 }
230 /* Now we assemble our shadow PTE from the page number and flags. */ 259 /* Now we assemble our shadow PTE from the page number and flags. */
@@ -234,8 +263,10 @@ static pte_t gpte_to_spte(struct lg_cpu *cpu, pte_t gpte, int write)
234/*H:460 And to complete the chain, release_pte() looks like this: */ 263/*H:460 And to complete the chain, release_pte() looks like this: */
235static void release_pte(pte_t pte) 264static void release_pte(pte_t pte)
236{ 265{
237 /* Remember that get_user_pages_fast() took a reference to the page, in 266 /*
238 * get_pfn()? We have to put it back now. */ 267 * Remember that get_user_pages_fast() took a reference to the page, in
268 * get_pfn()? We have to put it back now.
269 */
239 if (pte_flags(pte) & _PAGE_PRESENT) 270 if (pte_flags(pte) & _PAGE_PRESENT)
240 put_page(pte_page(pte)); 271 put_page(pte_page(pte));
241} 272}
@@ -273,7 +304,8 @@ static void check_gpmd(struct lg_cpu *cpu, pmd_t gpmd)
273 * and return to the Guest without it knowing. 304 * and return to the Guest without it knowing.
274 * 305 *
275 * If we fixed up the fault (ie. we mapped the address), this routine returns 306 * If we fixed up the fault (ie. we mapped the address), this routine returns
276 * true. Otherwise, it was a real fault and we need to tell the Guest. */ 307 * true. Otherwise, it was a real fault and we need to tell the Guest.
308 */
277bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode) 309bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
278{ 310{
279 pgd_t gpgd; 311 pgd_t gpgd;
@@ -298,22 +330,26 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
298 if (!(pgd_flags(*spgd) & _PAGE_PRESENT)) { 330 if (!(pgd_flags(*spgd) & _PAGE_PRESENT)) {
299 /* No shadow entry: allocate a new shadow PTE page. */ 331 /* No shadow entry: allocate a new shadow PTE page. */
300 unsigned long ptepage = get_zeroed_page(GFP_KERNEL); 332 unsigned long ptepage = get_zeroed_page(GFP_KERNEL);
301 /* This is not really the Guest's fault, but killing it is 333 /*
302 * simple for this corner case. */ 334 * This is not really the Guest's fault, but killing it is
335 * simple for this corner case.
336 */
303 if (!ptepage) { 337 if (!ptepage) {
304 kill_guest(cpu, "out of memory allocating pte page"); 338 kill_guest(cpu, "out of memory allocating pte page");
305 return false; 339 return false;
306 } 340 }
307 /* We check that the Guest pgd is OK. */ 341 /* We check that the Guest pgd is OK. */
308 check_gpgd(cpu, gpgd); 342 check_gpgd(cpu, gpgd);
309 /* And we copy the flags to the shadow PGD entry. The page 343 /*
310 * number in the shadow PGD is the page we just allocated. */ 344 * And we copy the flags to the shadow PGD entry. The page
345 * number in the shadow PGD is the page we just allocated.
346 */
311 set_pgd(spgd, __pgd(__pa(ptepage) | pgd_flags(gpgd))); 347 set_pgd(spgd, __pgd(__pa(ptepage) | pgd_flags(gpgd)));
312 } 348 }
313 349
314#ifdef CONFIG_X86_PAE 350#ifdef CONFIG_X86_PAE
315 gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t); 351 gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t);
316 /* middle level not present? We can't map it in. */ 352 /* Middle level not present? We can't map it in. */
317 if (!(pmd_flags(gpmd) & _PAGE_PRESENT)) 353 if (!(pmd_flags(gpmd) & _PAGE_PRESENT))
318 return false; 354 return false;
319 355
@@ -324,8 +360,10 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
324 /* No shadow entry: allocate a new shadow PTE page. */ 360 /* No shadow entry: allocate a new shadow PTE page. */
325 unsigned long ptepage = get_zeroed_page(GFP_KERNEL); 361 unsigned long ptepage = get_zeroed_page(GFP_KERNEL);
326 362
327 /* This is not really the Guest's fault, but killing it is 363 /*
328 * simple for this corner case. */ 364 * This is not really the Guest's fault, but killing it is
365 * simple for this corner case.
366 */
329 if (!ptepage) { 367 if (!ptepage) {
330 kill_guest(cpu, "out of memory allocating pte page"); 368 kill_guest(cpu, "out of memory allocating pte page");
331 return false; 369 return false;
@@ -334,17 +372,23 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
334 /* We check that the Guest pmd is OK. */ 372 /* We check that the Guest pmd is OK. */
335 check_gpmd(cpu, gpmd); 373 check_gpmd(cpu, gpmd);
336 374
337 /* And we copy the flags to the shadow PMD entry. The page 375 /*
338 * number in the shadow PMD is the page we just allocated. */ 376 * And we copy the flags to the shadow PMD entry. The page
377 * number in the shadow PMD is the page we just allocated.
378 */
339 native_set_pmd(spmd, __pmd(__pa(ptepage) | pmd_flags(gpmd))); 379 native_set_pmd(spmd, __pmd(__pa(ptepage) | pmd_flags(gpmd)));
340 } 380 }
341 381
342 /* OK, now we look at the lower level in the Guest page table: keep its 382 /*
343 * address, because we might update it later. */ 383 * OK, now we look at the lower level in the Guest page table: keep its
384 * address, because we might update it later.
385 */
344 gpte_ptr = gpte_addr(cpu, gpmd, vaddr); 386 gpte_ptr = gpte_addr(cpu, gpmd, vaddr);
345#else 387#else
346 /* OK, now we look at the lower level in the Guest page table: keep its 388 /*
347 * address, because we might update it later. */ 389 * OK, now we look at the lower level in the Guest page table: keep its
390 * address, because we might update it later.
391 */
348 gpte_ptr = gpte_addr(cpu, gpgd, vaddr); 392 gpte_ptr = gpte_addr(cpu, gpgd, vaddr);
349#endif 393#endif
350 gpte = lgread(cpu, gpte_ptr, pte_t); 394 gpte = lgread(cpu, gpte_ptr, pte_t);
@@ -353,8 +397,10 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
353 if (!(pte_flags(gpte) & _PAGE_PRESENT)) 397 if (!(pte_flags(gpte) & _PAGE_PRESENT))
354 return false; 398 return false;
355 399
356 /* Check they're not trying to write to a page the Guest wants 400 /*
357 * read-only (bit 2 of errcode == write). */ 401 * Check they're not trying to write to a page the Guest wants
402 * read-only (bit 2 of errcode == write).
403 */
358 if ((errcode & 2) && !(pte_flags(gpte) & _PAGE_RW)) 404 if ((errcode & 2) && !(pte_flags(gpte) & _PAGE_RW))
359 return false; 405 return false;
360 406
@@ -362,8 +408,10 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
362 if ((errcode & 4) && !(pte_flags(gpte) & _PAGE_USER)) 408 if ((errcode & 4) && !(pte_flags(gpte) & _PAGE_USER))
363 return false; 409 return false;
364 410
365 /* Check that the Guest PTE flags are OK, and the page number is below 411 /*
366 * the pfn_limit (ie. not mapping the Launcher binary). */ 412 * Check that the Guest PTE flags are OK, and the page number is below
413 * the pfn_limit (ie. not mapping the Launcher binary).
414 */
367 check_gpte(cpu, gpte); 415 check_gpte(cpu, gpte);
368 416
369 /* Add the _PAGE_ACCESSED and (for a write) _PAGE_DIRTY flag */ 417 /* Add the _PAGE_ACCESSED and (for a write) _PAGE_DIRTY flag */
@@ -373,29 +421,40 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
373 421
374 /* Get the pointer to the shadow PTE entry we're going to set. */ 422 /* Get the pointer to the shadow PTE entry we're going to set. */
375 spte = spte_addr(cpu, *spgd, vaddr); 423 spte = spte_addr(cpu, *spgd, vaddr);
376 /* If there was a valid shadow PTE entry here before, we release it. 424
377 * This can happen with a write to a previously read-only entry. */ 425 /*
426 * If there was a valid shadow PTE entry here before, we release it.
427 * This can happen with a write to a previously read-only entry.
428 */
378 release_pte(*spte); 429 release_pte(*spte);
379 430
380 /* If this is a write, we insist that the Guest page is writable (the 431 /*
381 * final arg to gpte_to_spte()). */ 432 * If this is a write, we insist that the Guest page is writable (the
433 * final arg to gpte_to_spte()).
434 */
382 if (pte_dirty(gpte)) 435 if (pte_dirty(gpte))
383 *spte = gpte_to_spte(cpu, gpte, 1); 436 *spte = gpte_to_spte(cpu, gpte, 1);
384 else 437 else
385 /* If this is a read, don't set the "writable" bit in the page 438 /*
439 * If this is a read, don't set the "writable" bit in the page
386 * table entry, even if the Guest says it's writable. That way 440 * table entry, even if the Guest says it's writable. That way
387 * we will come back here when a write does actually occur, so 441 * we will come back here when a write does actually occur, so
388 * we can update the Guest's _PAGE_DIRTY flag. */ 442 * we can update the Guest's _PAGE_DIRTY flag.
443 */
389 native_set_pte(spte, gpte_to_spte(cpu, pte_wrprotect(gpte), 0)); 444 native_set_pte(spte, gpte_to_spte(cpu, pte_wrprotect(gpte), 0));
390 445
391 /* Finally, we write the Guest PTE entry back: we've set the 446 /*
392 * _PAGE_ACCESSED and maybe the _PAGE_DIRTY flags. */ 447 * Finally, we write the Guest PTE entry back: we've set the
448 * _PAGE_ACCESSED and maybe the _PAGE_DIRTY flags.
449 */
393 lgwrite(cpu, gpte_ptr, pte_t, gpte); 450 lgwrite(cpu, gpte_ptr, pte_t, gpte);
394 451
395 /* The fault is fixed, the page table is populated, the mapping 452 /*
453 * The fault is fixed, the page table is populated, the mapping
396 * manipulated, the result returned and the code complete. A small 454 * manipulated, the result returned and the code complete. A small
397 * delay and a trace of alliteration are the only indications the Guest 455 * delay and a trace of alliteration are the only indications the Guest
398 * has that a page fault occurred at all. */ 456 * has that a page fault occurred at all.
457 */
399 return true; 458 return true;
400} 459}
401 460
@@ -408,7 +467,8 @@ bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode)
408 * mapped, so it's overkill. 467 * mapped, so it's overkill.
409 * 468 *
410 * This is a quick version which answers the question: is this virtual address 469 * This is a quick version which answers the question: is this virtual address
411 * mapped by the shadow page tables, and is it writable? */ 470 * mapped by the shadow page tables, and is it writable?
471 */
412static bool page_writable(struct lg_cpu *cpu, unsigned long vaddr) 472static bool page_writable(struct lg_cpu *cpu, unsigned long vaddr)
413{ 473{
414 pgd_t *spgd; 474 pgd_t *spgd;
@@ -428,16 +488,20 @@ static bool page_writable(struct lg_cpu *cpu, unsigned long vaddr)
428 return false; 488 return false;
429#endif 489#endif
430 490
431 /* Check the flags on the pte entry itself: it must be present and 491 /*
432 * writable. */ 492 * Check the flags on the pte entry itself: it must be present and
493 * writable.
494 */
433 flags = pte_flags(*(spte_addr(cpu, *spgd, vaddr))); 495 flags = pte_flags(*(spte_addr(cpu, *spgd, vaddr)));
434 496
435 return (flags & (_PAGE_PRESENT|_PAGE_RW)) == (_PAGE_PRESENT|_PAGE_RW); 497 return (flags & (_PAGE_PRESENT|_PAGE_RW)) == (_PAGE_PRESENT|_PAGE_RW);
436} 498}
437 499
438/* So, when pin_stack_pages() asks us to pin a page, we check if it's already 500/*
501 * So, when pin_stack_pages() asks us to pin a page, we check if it's already
439 * in the page tables, and if not, we call demand_page() with error code 2 502 * in the page tables, and if not, we call demand_page() with error code 2
440 * (meaning "write"). */ 503 * (meaning "write").
504 */
441void pin_page(struct lg_cpu *cpu, unsigned long vaddr) 505void pin_page(struct lg_cpu *cpu, unsigned long vaddr)
442{ 506{
443 if (!page_writable(cpu, vaddr) && !demand_page(cpu, vaddr, 2)) 507 if (!page_writable(cpu, vaddr) && !demand_page(cpu, vaddr, 2))
@@ -485,9 +549,11 @@ static void release_pgd(pgd_t *spgd)
485 /* If the entry's not present, there's nothing to release. */ 549 /* If the entry's not present, there's nothing to release. */
486 if (pgd_flags(*spgd) & _PAGE_PRESENT) { 550 if (pgd_flags(*spgd) & _PAGE_PRESENT) {
487 unsigned int i; 551 unsigned int i;
488 /* Converting the pfn to find the actual PTE page is easy: turn 552 /*
553 * Converting the pfn to find the actual PTE page is easy: turn
489 * the page number into a physical address, then convert to a 554 * the page number into a physical address, then convert to a
490 * virtual address (easy for kernel pages like this one). */ 555 * virtual address (easy for kernel pages like this one).
556 */
491 pte_t *ptepage = __va(pgd_pfn(*spgd) << PAGE_SHIFT); 557 pte_t *ptepage = __va(pgd_pfn(*spgd) << PAGE_SHIFT);
492 /* For each entry in the page, we might need to release it. */ 558 /* For each entry in the page, we might need to release it. */
493 for (i = 0; i < PTRS_PER_PTE; i++) 559 for (i = 0; i < PTRS_PER_PTE; i++)
@@ -499,9 +565,12 @@ static void release_pgd(pgd_t *spgd)
499 } 565 }
500} 566}
501#endif 567#endif
502/*H:445 We saw flush_user_mappings() twice: once from the flush_user_mappings() 568
569/*H:445
570 * We saw flush_user_mappings() twice: once from the flush_user_mappings()
503 * hypercall and once in new_pgdir() when we re-used a top-level pgdir page. 571 * hypercall and once in new_pgdir() when we re-used a top-level pgdir page.
504 * It simply releases every PTE page from 0 up to the Guest's kernel address. */ 572 * It simply releases every PTE page from 0 up to the Guest's kernel address.
573 */
505static void flush_user_mappings(struct lguest *lg, int idx) 574static void flush_user_mappings(struct lguest *lg, int idx)
506{ 575{
507 unsigned int i; 576 unsigned int i;
@@ -510,10 +579,12 @@ static void flush_user_mappings(struct lguest *lg, int idx)
510 release_pgd(lg->pgdirs[idx].pgdir + i); 579 release_pgd(lg->pgdirs[idx].pgdir + i);
511} 580}
512 581
513/*H:440 (v) Flushing (throwing away) page tables, 582/*H:440
583 * (v) Flushing (throwing away) page tables,
514 * 584 *
515 * The Guest has a hypercall to throw away the page tables: it's used when a 585 * The Guest has a hypercall to throw away the page tables: it's used when a
516 * large number of mappings have been changed. */ 586 * large number of mappings have been changed.
587 */
517void guest_pagetable_flush_user(struct lg_cpu *cpu) 588void guest_pagetable_flush_user(struct lg_cpu *cpu)
518{ 589{
519 /* Drop the userspace part of the current page table. */ 590 /* Drop the userspace part of the current page table. */
@@ -551,9 +622,11 @@ unsigned long guest_pa(struct lg_cpu *cpu, unsigned long vaddr)
551 return pte_pfn(gpte) * PAGE_SIZE | (vaddr & ~PAGE_MASK); 622 return pte_pfn(gpte) * PAGE_SIZE | (vaddr & ~PAGE_MASK);
552} 623}
553 624
554/* We keep several page tables. This is a simple routine to find the page 625/*
626 * We keep several page tables. This is a simple routine to find the page
555 * table (if any) corresponding to this top-level address the Guest has given 627 * table (if any) corresponding to this top-level address the Guest has given
556 * us. */ 628 * us.
629 */
557static unsigned int find_pgdir(struct lguest *lg, unsigned long pgtable) 630static unsigned int find_pgdir(struct lguest *lg, unsigned long pgtable)
558{ 631{
559 unsigned int i; 632 unsigned int i;
@@ -563,9 +636,11 @@ static unsigned int find_pgdir(struct lguest *lg, unsigned long pgtable)
563 return i; 636 return i;
564} 637}
565 638
566/*H:435 And this is us, creating the new page directory. If we really do 639/*H:435
640 * And this is us, creating the new page directory. If we really do
567 * allocate a new one (and so the kernel parts are not there), we set 641 * allocate a new one (and so the kernel parts are not there), we set
568 * blank_pgdir. */ 642 * blank_pgdir.
643 */
569static unsigned int new_pgdir(struct lg_cpu *cpu, 644static unsigned int new_pgdir(struct lg_cpu *cpu,
570 unsigned long gpgdir, 645 unsigned long gpgdir,
571 int *blank_pgdir) 646 int *blank_pgdir)
@@ -575,8 +650,10 @@ static unsigned int new_pgdir(struct lg_cpu *cpu,
575 pmd_t *pmd_table; 650 pmd_t *pmd_table;
576#endif 651#endif
577 652
578 /* We pick one entry at random to throw out. Choosing the Least 653 /*
579 * Recently Used might be better, but this is easy. */ 654 * We pick one entry at random to throw out. Choosing the Least
655 * Recently Used might be better, but this is easy.
656 */
580 next = random32() % ARRAY_SIZE(cpu->lg->pgdirs); 657 next = random32() % ARRAY_SIZE(cpu->lg->pgdirs);
581 /* If it's never been allocated at all before, try now. */ 658 /* If it's never been allocated at all before, try now. */
582 if (!cpu->lg->pgdirs[next].pgdir) { 659 if (!cpu->lg->pgdirs[next].pgdir) {
@@ -587,8 +664,10 @@ static unsigned int new_pgdir(struct lg_cpu *cpu,
587 next = cpu->cpu_pgd; 664 next = cpu->cpu_pgd;
588 else { 665 else {
589#ifdef CONFIG_X86_PAE 666#ifdef CONFIG_X86_PAE
590 /* In PAE mode, allocate a pmd page and populate the 667 /*
591 * last pgd entry. */ 668 * In PAE mode, allocate a pmd page and populate the
669 * last pgd entry.
670 */
592 pmd_table = (pmd_t *)get_zeroed_page(GFP_KERNEL); 671 pmd_table = (pmd_t *)get_zeroed_page(GFP_KERNEL);
593 if (!pmd_table) { 672 if (!pmd_table) {
594 free_page((long)cpu->lg->pgdirs[next].pgdir); 673 free_page((long)cpu->lg->pgdirs[next].pgdir);
@@ -598,8 +677,10 @@ static unsigned int new_pgdir(struct lg_cpu *cpu,
598 set_pgd(cpu->lg->pgdirs[next].pgdir + 677 set_pgd(cpu->lg->pgdirs[next].pgdir +
599 SWITCHER_PGD_INDEX, 678 SWITCHER_PGD_INDEX,
600 __pgd(__pa(pmd_table) | _PAGE_PRESENT)); 679 __pgd(__pa(pmd_table) | _PAGE_PRESENT));
601 /* This is a blank page, so there are no kernel 680 /*
602 * mappings: caller must map the stack! */ 681 * This is a blank page, so there are no kernel
682 * mappings: caller must map the stack!
683 */
603 *blank_pgdir = 1; 684 *blank_pgdir = 1;
604 } 685 }
605#else 686#else
@@ -615,19 +696,23 @@ static unsigned int new_pgdir(struct lg_cpu *cpu,
615 return next; 696 return next;
616} 697}
617 698
618/*H:430 (iv) Switching page tables 699/*H:430
700 * (iv) Switching page tables
619 * 701 *
620 * Now we've seen all the page table setting and manipulation, let's see 702 * Now we've seen all the page table setting and manipulation, let's see
621 * what happens when the Guest changes page tables (ie. changes the top-level 703 * what happens when the Guest changes page tables (ie. changes the top-level
622 * pgdir). This occurs on almost every context switch. */ 704 * pgdir). This occurs on almost every context switch.
705 */
623void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable) 706void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable)
624{ 707{
625 int newpgdir, repin = 0; 708 int newpgdir, repin = 0;
626 709
627 /* Look to see if we have this one already. */ 710 /* Look to see if we have this one already. */
628 newpgdir = find_pgdir(cpu->lg, pgtable); 711 newpgdir = find_pgdir(cpu->lg, pgtable);
629 /* If not, we allocate or mug an existing one: if it's a fresh one, 712 /*
630 * repin gets set to 1. */ 713 * If not, we allocate or mug an existing one: if it's a fresh one,
714 * repin gets set to 1.
715 */
631 if (newpgdir == ARRAY_SIZE(cpu->lg->pgdirs)) 716 if (newpgdir == ARRAY_SIZE(cpu->lg->pgdirs))
632 newpgdir = new_pgdir(cpu, pgtable, &repin); 717 newpgdir = new_pgdir(cpu, pgtable, &repin);
633 /* Change the current pgd index to the new one. */ 718 /* Change the current pgd index to the new one. */
@@ -637,9 +722,11 @@ void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable)
637 pin_stack_pages(cpu); 722 pin_stack_pages(cpu);
638} 723}
639 724
640/*H:470 Finally, a routine which throws away everything: all PGD entries in all 725/*H:470
726 * Finally, a routine which throws away everything: all PGD entries in all
641 * the shadow page tables, including the Guest's kernel mappings. This is used 727 * the shadow page tables, including the Guest's kernel mappings. This is used
642 * when we destroy the Guest. */ 728 * when we destroy the Guest.
729 */
643static void release_all_pagetables(struct lguest *lg) 730static void release_all_pagetables(struct lguest *lg)
644{ 731{
645 unsigned int i, j; 732 unsigned int i, j;
@@ -656,8 +743,10 @@ static void release_all_pagetables(struct lguest *lg)
656 spgd = lg->pgdirs[i].pgdir + SWITCHER_PGD_INDEX; 743 spgd = lg->pgdirs[i].pgdir + SWITCHER_PGD_INDEX;
657 pmdpage = __va(pgd_pfn(*spgd) << PAGE_SHIFT); 744 pmdpage = __va(pgd_pfn(*spgd) << PAGE_SHIFT);
658 745
659 /* And release the pmd entries of that pmd page, 746 /*
660 * except for the switcher pmd. */ 747 * And release the pmd entries of that pmd page,
748 * except for the switcher pmd.
749 */
661 for (k = 0; k < SWITCHER_PMD_INDEX; k++) 750 for (k = 0; k < SWITCHER_PMD_INDEX; k++)
662 release_pmd(&pmdpage[k]); 751 release_pmd(&pmdpage[k]);
663#endif 752#endif
@@ -667,10 +756,12 @@ static void release_all_pagetables(struct lguest *lg)
667 } 756 }
668} 757}
669 758
670/* We also throw away everything when a Guest tells us it's changed a kernel 759/*
760 * We also throw away everything when a Guest tells us it's changed a kernel
671 * mapping. Since kernel mappings are in every page table, it's easiest to 761 * mapping. Since kernel mappings are in every page table, it's easiest to
672 * throw them all away. This traps the Guest in amber for a while as 762 * throw them all away. This traps the Guest in amber for a while as
673 * everything faults back in, but it's rare. */ 763 * everything faults back in, but it's rare.
764 */
674void guest_pagetable_clear_all(struct lg_cpu *cpu) 765void guest_pagetable_clear_all(struct lg_cpu *cpu)
675{ 766{
676 release_all_pagetables(cpu->lg); 767 release_all_pagetables(cpu->lg);
@@ -678,15 +769,19 @@ void guest_pagetable_clear_all(struct lg_cpu *cpu)
678 pin_stack_pages(cpu); 769 pin_stack_pages(cpu);
679} 770}
680/*:*/ 771/*:*/
681/*M:009 Since we throw away all mappings when a kernel mapping changes, our 772
773/*M:009
774 * Since we throw away all mappings when a kernel mapping changes, our
682 * performance sucks for guests using highmem. In fact, a guest with 775 * performance sucks for guests using highmem. In fact, a guest with
683 * PAGE_OFFSET 0xc0000000 (the default) and more than about 700MB of RAM is 776 * PAGE_OFFSET 0xc0000000 (the default) and more than about 700MB of RAM is
684 * usually slower than a Guest with less memory. 777 * usually slower than a Guest with less memory.
685 * 778 *
686 * This, of course, cannot be fixed. It would take some kind of... well, I 779 * This, of course, cannot be fixed. It would take some kind of... well, I
687 * don't know, but the term "puissant code-fu" comes to mind. :*/ 780 * don't know, but the term "puissant code-fu" comes to mind.
781:*/
688 782
689/*H:420 This is the routine which actually sets the page table entry for then 783/*H:420
784 * This is the routine which actually sets the page table entry for then
690 * "idx"'th shadow page table. 785 * "idx"'th shadow page table.
691 * 786 *
692 * Normally, we can just throw out the old entry and replace it with 0: if they 787 * Normally, we can just throw out the old entry and replace it with 0: if they
@@ -715,31 +810,36 @@ static void do_set_pte(struct lg_cpu *cpu, int idx,
715 spmd = spmd_addr(cpu, *spgd, vaddr); 810 spmd = spmd_addr(cpu, *spgd, vaddr);
716 if (pmd_flags(*spmd) & _PAGE_PRESENT) { 811 if (pmd_flags(*spmd) & _PAGE_PRESENT) {
717#endif 812#endif
718 /* Otherwise, we start by releasing 813 /* Otherwise, start by releasing the existing entry. */
719 * the existing entry. */
720 pte_t *spte = spte_addr(cpu, *spgd, vaddr); 814 pte_t *spte = spte_addr(cpu, *spgd, vaddr);
721 release_pte(*spte); 815 release_pte(*spte);
722 816
723 /* If they're setting this entry as dirty or accessed, 817 /*
724 * we might as well put that entry they've given us 818 * If they're setting this entry as dirty or accessed,
725 * in now. This shaves 10% off a 819 * we might as well put that entry they've given us in
726 * copy-on-write micro-benchmark. */ 820 * now. This shaves 10% off a copy-on-write
821 * micro-benchmark.
822 */
727 if (pte_flags(gpte) & (_PAGE_DIRTY | _PAGE_ACCESSED)) { 823 if (pte_flags(gpte) & (_PAGE_DIRTY | _PAGE_ACCESSED)) {
728 check_gpte(cpu, gpte); 824 check_gpte(cpu, gpte);
729 native_set_pte(spte, 825 native_set_pte(spte,
730 gpte_to_spte(cpu, gpte, 826 gpte_to_spte(cpu, gpte,
731 pte_flags(gpte) & _PAGE_DIRTY)); 827 pte_flags(gpte) & _PAGE_DIRTY));
732 } else 828 } else {
733 /* Otherwise kill it and we can demand_page() 829 /*
734 * it in later. */ 830 * Otherwise kill it and we can demand_page()
831 * it in later.
832 */
735 native_set_pte(spte, __pte(0)); 833 native_set_pte(spte, __pte(0));
834 }
736#ifdef CONFIG_X86_PAE 835#ifdef CONFIG_X86_PAE
737 } 836 }
738#endif 837#endif
739 } 838 }
740} 839}
741 840
742/*H:410 Updating a PTE entry is a little trickier. 841/*H:410
842 * Updating a PTE entry is a little trickier.
743 * 843 *
744 * We keep track of several different page tables (the Guest uses one for each 844 * We keep track of several different page tables (the Guest uses one for each
745 * process, so it makes sense to cache at least a few). Each of these have 845 * process, so it makes sense to cache at least a few). Each of these have
@@ -748,12 +848,15 @@ static void do_set_pte(struct lg_cpu *cpu, int idx,
748 * all the page tables, not just the current one. This is rare. 848 * all the page tables, not just the current one. This is rare.
749 * 849 *
750 * The benefit is that when we have to track a new page table, we can keep all 850 * The benefit is that when we have to track a new page table, we can keep all
751 * the kernel mappings. This speeds up context switch immensely. */ 851 * the kernel mappings. This speeds up context switch immensely.
852 */
752void guest_set_pte(struct lg_cpu *cpu, 853void guest_set_pte(struct lg_cpu *cpu,
753 unsigned long gpgdir, unsigned long vaddr, pte_t gpte) 854 unsigned long gpgdir, unsigned long vaddr, pte_t gpte)
754{ 855{
755 /* Kernel mappings must be changed on all top levels. Slow, but doesn't 856 /*
756 * happen often. */ 857 * Kernel mappings must be changed on all top levels. Slow, but doesn't
858 * happen often.
859 */
757 if (vaddr >= cpu->lg->kernel_address) { 860 if (vaddr >= cpu->lg->kernel_address) {
758 unsigned int i; 861 unsigned int i;
759 for (i = 0; i < ARRAY_SIZE(cpu->lg->pgdirs); i++) 862 for (i = 0; i < ARRAY_SIZE(cpu->lg->pgdirs); i++)
@@ -802,12 +905,14 @@ void guest_set_pmd(struct lguest *lg, unsigned long pmdp, u32 idx)
802} 905}
803#endif 906#endif
804 907
805/* Once we know how much memory we have we can construct simple identity 908/*
806 * (which set virtual == physical) and linear mappings 909 * Once we know how much memory we have we can construct simple identity (which
807 * which will get the Guest far enough into the boot to create its own. 910 * set virtual == physical) and linear mappings which will get the Guest far
911 * enough into the boot to create its own.
808 * 912 *
809 * We lay them out of the way, just below the initrd (which is why we need to 913 * We lay them out of the way, just below the initrd (which is why we need to
810 * know its size here). */ 914 * know its size here).
915 */
811static unsigned long setup_pagetables(struct lguest *lg, 916static unsigned long setup_pagetables(struct lguest *lg,
812 unsigned long mem, 917 unsigned long mem,
813 unsigned long initrd_size) 918 unsigned long initrd_size)
@@ -825,8 +930,10 @@ static unsigned long setup_pagetables(struct lguest *lg,
825 unsigned int phys_linear; 930 unsigned int phys_linear;
826#endif 931#endif
827 932
828 /* We have mapped_pages frames to map, so we need 933 /*
829 * linear_pages page tables to map them. */ 934 * We have mapped_pages frames to map, so we need linear_pages page
935 * tables to map them.
936 */
830 mapped_pages = mem / PAGE_SIZE; 937 mapped_pages = mem / PAGE_SIZE;
831 linear_pages = (mapped_pages + PTRS_PER_PTE - 1) / PTRS_PER_PTE; 938 linear_pages = (mapped_pages + PTRS_PER_PTE - 1) / PTRS_PER_PTE;
832 939
@@ -839,8 +946,10 @@ static unsigned long setup_pagetables(struct lguest *lg,
839#ifdef CONFIG_X86_PAE 946#ifdef CONFIG_X86_PAE
840 pmds = (void *)linear - PAGE_SIZE; 947 pmds = (void *)linear - PAGE_SIZE;
841#endif 948#endif
842 /* Linear mapping is easy: put every page's address into the 949 /*
843 * mapping in order. */ 950 * Linear mapping is easy: put every page's address into the
951 * mapping in order.
952 */
844 for (i = 0; i < mapped_pages; i++) { 953 for (i = 0; i < mapped_pages; i++) {
845 pte_t pte; 954 pte_t pte;
846 pte = pfn_pte(i, __pgprot(_PAGE_PRESENT|_PAGE_RW|_PAGE_USER)); 955 pte = pfn_pte(i, __pgprot(_PAGE_PRESENT|_PAGE_RW|_PAGE_USER));
@@ -848,8 +957,10 @@ static unsigned long setup_pagetables(struct lguest *lg,
848 return -EFAULT; 957 return -EFAULT;
849 } 958 }
850 959
851 /* The top level points to the linear page table pages above. 960 /*
852 * We setup the identity and linear mappings here. */ 961 * The top level points to the linear page table pages above.
962 * We setup the identity and linear mappings here.
963 */
853#ifdef CONFIG_X86_PAE 964#ifdef CONFIG_X86_PAE
854 for (i = j = 0; i < mapped_pages && j < PTRS_PER_PMD; 965 for (i = j = 0; i < mapped_pages && j < PTRS_PER_PMD;
855 i += PTRS_PER_PTE, j++) { 966 i += PTRS_PER_PTE, j++) {
@@ -880,15 +991,19 @@ static unsigned long setup_pagetables(struct lguest *lg,
880 } 991 }
881#endif 992#endif
882 993
883 /* We return the top level (guest-physical) address: remember where 994 /*
884 * this is. */ 995 * We return the top level (guest-physical) address: remember where
996 * this is.
997 */
885 return (unsigned long)pgdir - mem_base; 998 return (unsigned long)pgdir - mem_base;
886} 999}
887 1000
888/*H:500 (vii) Setting up the page tables initially. 1001/*H:500
1002 * (vii) Setting up the page tables initially.
889 * 1003 *
890 * When a Guest is first created, the Launcher tells us where the toplevel of 1004 * When a Guest is first created, the Launcher tells us where the toplevel of
891 * its first page table is. We set some things up here: */ 1005 * its first page table is. We set some things up here:
1006 */
892int init_guest_pagetable(struct lguest *lg) 1007int init_guest_pagetable(struct lguest *lg)
893{ 1008{
894 u64 mem; 1009 u64 mem;
@@ -898,14 +1013,18 @@ int init_guest_pagetable(struct lguest *lg)
898 pgd_t *pgd; 1013 pgd_t *pgd;
899 pmd_t *pmd_table; 1014 pmd_t *pmd_table;
900#endif 1015#endif
901 /* Get the Guest memory size and the ramdisk size from the boot header 1016 /*
902 * located at lg->mem_base (Guest address 0). */ 1017 * Get the Guest memory size and the ramdisk size from the boot header
1018 * located at lg->mem_base (Guest address 0).
1019 */
903 if (copy_from_user(&mem, &boot->e820_map[0].size, sizeof(mem)) 1020 if (copy_from_user(&mem, &boot->e820_map[0].size, sizeof(mem))
904 || get_user(initrd_size, &boot->hdr.ramdisk_size)) 1021 || get_user(initrd_size, &boot->hdr.ramdisk_size))
905 return -EFAULT; 1022 return -EFAULT;
906 1023
907 /* We start on the first shadow page table, and give it a blank PGD 1024 /*
908 * page. */ 1025 * We start on the first shadow page table, and give it a blank PGD
1026 * page.
1027 */
909 lg->pgdirs[0].gpgdir = setup_pagetables(lg, mem, initrd_size); 1028 lg->pgdirs[0].gpgdir = setup_pagetables(lg, mem, initrd_size);
910 if (IS_ERR_VALUE(lg->pgdirs[0].gpgdir)) 1029 if (IS_ERR_VALUE(lg->pgdirs[0].gpgdir))
911 return lg->pgdirs[0].gpgdir; 1030 return lg->pgdirs[0].gpgdir;
@@ -931,17 +1050,21 @@ void page_table_guest_data_init(struct lg_cpu *cpu)
931 /* We get the kernel address: above this is all kernel memory. */ 1050 /* We get the kernel address: above this is all kernel memory. */
932 if (get_user(cpu->lg->kernel_address, 1051 if (get_user(cpu->lg->kernel_address,
933 &cpu->lg->lguest_data->kernel_address) 1052 &cpu->lg->lguest_data->kernel_address)
934 /* We tell the Guest that it can't use the top 2 or 4 MB 1053 /*
935 * of virtual addresses used by the Switcher. */ 1054 * We tell the Guest that it can't use the top 2 or 4 MB
1055 * of virtual addresses used by the Switcher.
1056 */
936 || put_user(RESERVE_MEM * 1024 * 1024, 1057 || put_user(RESERVE_MEM * 1024 * 1024,
937 &cpu->lg->lguest_data->reserve_mem) 1058 &cpu->lg->lguest_data->reserve_mem)
938 || put_user(cpu->lg->pgdirs[0].gpgdir, 1059 || put_user(cpu->lg->pgdirs[0].gpgdir,
939 &cpu->lg->lguest_data->pgdir)) 1060 &cpu->lg->lguest_data->pgdir))
940 kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); 1061 kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data);
941 1062
942 /* In flush_user_mappings() we loop from 0 to 1063 /*
1064 * In flush_user_mappings() we loop from 0 to
943 * "pgd_index(lg->kernel_address)". This assumes it won't hit the 1065 * "pgd_index(lg->kernel_address)". This assumes it won't hit the
944 * Switcher mappings, so check that now. */ 1066 * Switcher mappings, so check that now.
1067 */
945#ifdef CONFIG_X86_PAE 1068#ifdef CONFIG_X86_PAE
946 if (pgd_index(cpu->lg->kernel_address) == SWITCHER_PGD_INDEX && 1069 if (pgd_index(cpu->lg->kernel_address) == SWITCHER_PGD_INDEX &&
947 pmd_index(cpu->lg->kernel_address) == SWITCHER_PMD_INDEX) 1070 pmd_index(cpu->lg->kernel_address) == SWITCHER_PMD_INDEX)
@@ -964,12 +1087,14 @@ void free_guest_pagetable(struct lguest *lg)
964 free_page((long)lg->pgdirs[i].pgdir); 1087 free_page((long)lg->pgdirs[i].pgdir);
965} 1088}
966 1089
967/*H:480 (vi) Mapping the Switcher when the Guest is about to run. 1090/*H:480
1091 * (vi) Mapping the Switcher when the Guest is about to run.
968 * 1092 *
969 * The Switcher and the two pages for this CPU need to be visible in the 1093 * The Switcher and the two pages for this CPU need to be visible in the
970 * Guest (and not the pages for other CPUs). We have the appropriate PTE pages 1094 * Guest (and not the pages for other CPUs). We have the appropriate PTE pages
971 * for each CPU already set up, we just need to hook them in now we know which 1095 * for each CPU already set up, we just need to hook them in now we know which
972 * Guest is about to run on this CPU. */ 1096 * Guest is about to run on this CPU.
1097 */
973void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages) 1098void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages)
974{ 1099{
975 pte_t *switcher_pte_page = __get_cpu_var(switcher_pte_pages); 1100 pte_t *switcher_pte_page = __get_cpu_var(switcher_pte_pages);
@@ -990,20 +1115,24 @@ void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages)
990#else 1115#else
991 pgd_t switcher_pgd; 1116 pgd_t switcher_pgd;
992 1117
993 /* Make the last PGD entry for this Guest point to the Switcher's PTE 1118 /*
994 * page for this CPU (with appropriate flags). */ 1119 * Make the last PGD entry for this Guest point to the Switcher's PTE
1120 * page for this CPU (with appropriate flags).
1121 */
995 switcher_pgd = __pgd(__pa(switcher_pte_page) | __PAGE_KERNEL_EXEC); 1122 switcher_pgd = __pgd(__pa(switcher_pte_page) | __PAGE_KERNEL_EXEC);
996 1123
997 cpu->lg->pgdirs[cpu->cpu_pgd].pgdir[SWITCHER_PGD_INDEX] = switcher_pgd; 1124 cpu->lg->pgdirs[cpu->cpu_pgd].pgdir[SWITCHER_PGD_INDEX] = switcher_pgd;
998 1125
999#endif 1126#endif
1000 /* We also change the Switcher PTE page. When we're running the Guest, 1127 /*
1128 * We also change the Switcher PTE page. When we're running the Guest,
1001 * we want the Guest's "regs" page to appear where the first Switcher 1129 * we want the Guest's "regs" page to appear where the first Switcher
1002 * page for this CPU is. This is an optimization: when the Switcher 1130 * page for this CPU is. This is an optimization: when the Switcher
1003 * saves the Guest registers, it saves them into the first page of this 1131 * saves the Guest registers, it saves them into the first page of this
1004 * CPU's "struct lguest_pages": if we make sure the Guest's register 1132 * CPU's "struct lguest_pages": if we make sure the Guest's register
1005 * page is already mapped there, we don't have to copy them out 1133 * page is already mapped there, we don't have to copy them out
1006 * again. */ 1134 * again.
1135 */
1007 pfn = __pa(cpu->regs_page) >> PAGE_SHIFT; 1136 pfn = __pa(cpu->regs_page) >> PAGE_SHIFT;
1008 native_set_pte(&regs_pte, pfn_pte(pfn, PAGE_KERNEL)); 1137 native_set_pte(&regs_pte, pfn_pte(pfn, PAGE_KERNEL));
1009 native_set_pte(&switcher_pte_page[pte_index((unsigned long)pages)], 1138 native_set_pte(&switcher_pte_page[pte_index((unsigned long)pages)],
@@ -1019,10 +1148,12 @@ static void free_switcher_pte_pages(void)
1019 free_page((long)switcher_pte_page(i)); 1148 free_page((long)switcher_pte_page(i));
1020} 1149}
1021 1150
1022/*H:520 Setting up the Switcher PTE page for given CPU is fairly easy, given 1151/*H:520
1152 * Setting up the Switcher PTE page for given CPU is fairly easy, given
1023 * the CPU number and the "struct page"s for the Switcher code itself. 1153 * the CPU number and the "struct page"s for the Switcher code itself.
1024 * 1154 *
1025 * Currently the Switcher is less than a page long, so "pages" is always 1. */ 1155 * Currently the Switcher is less than a page long, so "pages" is always 1.
1156 */
1026static __init void populate_switcher_pte_page(unsigned int cpu, 1157static __init void populate_switcher_pte_page(unsigned int cpu,
1027 struct page *switcher_page[], 1158 struct page *switcher_page[],
1028 unsigned int pages) 1159 unsigned int pages)
@@ -1043,13 +1174,16 @@ static __init void populate_switcher_pte_page(unsigned int cpu,
1043 native_set_pte(&pte[i], pfn_pte(page_to_pfn(switcher_page[i]), 1174 native_set_pte(&pte[i], pfn_pte(page_to_pfn(switcher_page[i]),
1044 __pgprot(_PAGE_PRESENT|_PAGE_ACCESSED|_PAGE_RW))); 1175 __pgprot(_PAGE_PRESENT|_PAGE_ACCESSED|_PAGE_RW)));
1045 1176
1046 /* The second page contains the "struct lguest_ro_state", and is 1177 /*
1047 * read-only. */ 1178 * The second page contains the "struct lguest_ro_state", and is
1179 * read-only.
1180 */
1048 native_set_pte(&pte[i+1], pfn_pte(page_to_pfn(switcher_page[i+1]), 1181 native_set_pte(&pte[i+1], pfn_pte(page_to_pfn(switcher_page[i+1]),
1049 __pgprot(_PAGE_PRESENT|_PAGE_ACCESSED))); 1182 __pgprot(_PAGE_PRESENT|_PAGE_ACCESSED)));
1050} 1183}
1051 1184
1052/* We've made it through the page table code. Perhaps our tired brains are 1185/*
1186 * We've made it through the page table code. Perhaps our tired brains are
1053 * still processing the details, or perhaps we're simply glad it's over. 1187 * still processing the details, or perhaps we're simply glad it's over.
1054 * 1188 *
1055 * If nothing else, note that all this complexity in juggling shadow page tables 1189 * If nothing else, note that all this complexity in juggling shadow page tables
@@ -1058,10 +1192,13 @@ static __init void populate_switcher_pte_page(unsigned int cpu,
1058 * uses exotic direct Guest pagetable manipulation, and why both Intel and AMD 1192 * uses exotic direct Guest pagetable manipulation, and why both Intel and AMD
1059 * have implemented shadow page table support directly into hardware. 1193 * have implemented shadow page table support directly into hardware.
1060 * 1194 *
1061 * There is just one file remaining in the Host. */ 1195 * There is just one file remaining in the Host.
1196 */
1062 1197
1063/*H:510 At boot or module load time, init_pagetables() allocates and populates 1198/*H:510
1064 * the Switcher PTE page for each CPU. */ 1199 * At boot or module load time, init_pagetables() allocates and populates
1200 * the Switcher PTE page for each CPU.
1201 */
1065__init int init_pagetables(struct page **switcher_page, unsigned int pages) 1202__init int init_pagetables(struct page **switcher_page, unsigned int pages)
1066{ 1203{
1067 unsigned int i; 1204 unsigned int i;
diff --git a/drivers/lguest/segments.c b/drivers/lguest/segments.c
index 482ed5a18750..951c57b0a7e0 100644
--- a/drivers/lguest/segments.c
+++ b/drivers/lguest/segments.c
@@ -1,4 +1,5 @@
1/*P:600 The x86 architecture has segments, which involve a table of descriptors 1/*P:600
2 * The x86 architecture has segments, which involve a table of descriptors
2 * which can be used to do funky things with virtual address interpretation. 3 * which can be used to do funky things with virtual address interpretation.
3 * We originally used to use segments so the Guest couldn't alter the 4 * We originally used to use segments so the Guest couldn't alter the
4 * Guest<->Host Switcher, and then we had to trim Guest segments, and restore 5 * Guest<->Host Switcher, and then we had to trim Guest segments, and restore
@@ -8,7 +9,8 @@
8 * 9 *
9 * In these modern times, the segment handling code consists of simple sanity 10 * In these modern times, the segment handling code consists of simple sanity
10 * checks, and the worst you'll experience reading this code is butterfly-rash 11 * checks, and the worst you'll experience reading this code is butterfly-rash
11 * from frolicking through its parklike serenity. :*/ 12 * from frolicking through its parklike serenity.
13:*/
12#include "lg.h" 14#include "lg.h"
13 15
14/*H:600 16/*H:600
@@ -41,10 +43,12 @@
41 * begin. 43 * begin.
42 */ 44 */
43 45
44/* There are several entries we don't let the Guest set. The TSS entry is the 46/*
47 * There are several entries we don't let the Guest set. The TSS entry is the
45 * "Task State Segment" which controls all kinds of delicate things. The 48 * "Task State Segment" which controls all kinds of delicate things. The
46 * LGUEST_CS and LGUEST_DS entries are reserved for the Switcher, and the 49 * LGUEST_CS and LGUEST_DS entries are reserved for the Switcher, and the
47 * the Guest can't be trusted to deal with double faults. */ 50 * the Guest can't be trusted to deal with double faults.
51 */
48static bool ignored_gdt(unsigned int num) 52static bool ignored_gdt(unsigned int num)
49{ 53{
50 return (num == GDT_ENTRY_TSS 54 return (num == GDT_ENTRY_TSS
@@ -53,42 +57,52 @@ static bool ignored_gdt(unsigned int num)
53 || num == GDT_ENTRY_DOUBLEFAULT_TSS); 57 || num == GDT_ENTRY_DOUBLEFAULT_TSS);
54} 58}
55 59
56/*H:630 Once the Guest gave us new GDT entries, we fix them up a little. We 60/*H:630
61 * Once the Guest gave us new GDT entries, we fix them up a little. We
57 * don't care if they're invalid: the worst that can happen is a General 62 * don't care if they're invalid: the worst that can happen is a General
58 * Protection Fault in the Switcher when it restores a Guest segment register 63 * Protection Fault in the Switcher when it restores a Guest segment register
59 * which tries to use that entry. Then we kill the Guest for causing such a 64 * which tries to use that entry. Then we kill the Guest for causing such a
60 * mess: the message will be "unhandled trap 256". */ 65 * mess: the message will be "unhandled trap 256".
66 */
61static void fixup_gdt_table(struct lg_cpu *cpu, unsigned start, unsigned end) 67static void fixup_gdt_table(struct lg_cpu *cpu, unsigned start, unsigned end)
62{ 68{
63 unsigned int i; 69 unsigned int i;
64 70
65 for (i = start; i < end; i++) { 71 for (i = start; i < end; i++) {
66 /* We never copy these ones to real GDT, so we don't care what 72 /*
67 * they say */ 73 * We never copy these ones to real GDT, so we don't care what
74 * they say
75 */
68 if (ignored_gdt(i)) 76 if (ignored_gdt(i))
69 continue; 77 continue;
70 78
71 /* Segment descriptors contain a privilege level: the Guest is 79 /*
80 * Segment descriptors contain a privilege level: the Guest is
72 * sometimes careless and leaves this as 0, even though it's 81 * sometimes careless and leaves this as 0, even though it's
73 * running at privilege level 1. If so, we fix it here. */ 82 * running at privilege level 1. If so, we fix it here.
83 */
74 if ((cpu->arch.gdt[i].b & 0x00006000) == 0) 84 if ((cpu->arch.gdt[i].b & 0x00006000) == 0)
75 cpu->arch.gdt[i].b |= (GUEST_PL << 13); 85 cpu->arch.gdt[i].b |= (GUEST_PL << 13);
76 86
77 /* Each descriptor has an "accessed" bit. If we don't set it 87 /*
88 * Each descriptor has an "accessed" bit. If we don't set it
78 * now, the CPU will try to set it when the Guest first loads 89 * now, the CPU will try to set it when the Guest first loads
79 * that entry into a segment register. But the GDT isn't 90 * that entry into a segment register. But the GDT isn't
80 * writable by the Guest, so bad things can happen. */ 91 * writable by the Guest, so bad things can happen.
92 */
81 cpu->arch.gdt[i].b |= 0x00000100; 93 cpu->arch.gdt[i].b |= 0x00000100;
82 } 94 }
83} 95}
84 96
85/*H:610 Like the IDT, we never simply use the GDT the Guest gives us. We keep 97/*H:610
98 * Like the IDT, we never simply use the GDT the Guest gives us. We keep
86 * a GDT for each CPU, and copy across the Guest's entries each time we want to 99 * a GDT for each CPU, and copy across the Guest's entries each time we want to
87 * run the Guest on that CPU. 100 * run the Guest on that CPU.
88 * 101 *
89 * This routine is called at boot or modprobe time for each CPU to set up the 102 * This routine is called at boot or modprobe time for each CPU to set up the
90 * constant GDT entries: the ones which are the same no matter what Guest we're 103 * constant GDT entries: the ones which are the same no matter what Guest we're
91 * running. */ 104 * running.
105 */
92void setup_default_gdt_entries(struct lguest_ro_state *state) 106void setup_default_gdt_entries(struct lguest_ro_state *state)
93{ 107{
94 struct desc_struct *gdt = state->guest_gdt; 108 struct desc_struct *gdt = state->guest_gdt;
@@ -98,30 +112,37 @@ void setup_default_gdt_entries(struct lguest_ro_state *state)
98 gdt[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT; 112 gdt[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
99 gdt[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT; 113 gdt[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
100 114
101 /* The TSS segment refers to the TSS entry for this particular CPU. 115 /*
116 * The TSS segment refers to the TSS entry for this particular CPU.
102 * Forgive the magic flags: the 0x8900 means the entry is Present, it's 117 * Forgive the magic flags: the 0x8900 means the entry is Present, it's
103 * privilege level 0 Available 386 TSS system segment, and the 0x67 118 * privilege level 0 Available 386 TSS system segment, and the 0x67
104 * means Saturn is eclipsed by Mercury in the twelfth house. */ 119 * means Saturn is eclipsed by Mercury in the twelfth house.
120 */
105 gdt[GDT_ENTRY_TSS].a = 0x00000067 | (tss << 16); 121 gdt[GDT_ENTRY_TSS].a = 0x00000067 | (tss << 16);
106 gdt[GDT_ENTRY_TSS].b = 0x00008900 | (tss & 0xFF000000) 122 gdt[GDT_ENTRY_TSS].b = 0x00008900 | (tss & 0xFF000000)
107 | ((tss >> 16) & 0x000000FF); 123 | ((tss >> 16) & 0x000000FF);
108} 124}
109 125
110/* This routine sets up the initial Guest GDT for booting. All entries start 126/*
111 * as 0 (unusable). */ 127 * This routine sets up the initial Guest GDT for booting. All entries start
128 * as 0 (unusable).
129 */
112void setup_guest_gdt(struct lg_cpu *cpu) 130void setup_guest_gdt(struct lg_cpu *cpu)
113{ 131{
114 /* Start with full 0-4G segments... */ 132 /*
133 * Start with full 0-4G segments...except the Guest is allowed to use
134 * them, so set the privilege level appropriately in the flags.
135 */
115 cpu->arch.gdt[GDT_ENTRY_KERNEL_CS] = FULL_EXEC_SEGMENT; 136 cpu->arch.gdt[GDT_ENTRY_KERNEL_CS] = FULL_EXEC_SEGMENT;
116 cpu->arch.gdt[GDT_ENTRY_KERNEL_DS] = FULL_SEGMENT; 137 cpu->arch.gdt[GDT_ENTRY_KERNEL_DS] = FULL_SEGMENT;
117 /* ...except the Guest is allowed to use them, so set the privilege
118 * level appropriately in the flags. */
119 cpu->arch.gdt[GDT_ENTRY_KERNEL_CS].b |= (GUEST_PL << 13); 138 cpu->arch.gdt[GDT_ENTRY_KERNEL_CS].b |= (GUEST_PL << 13);
120 cpu->arch.gdt[GDT_ENTRY_KERNEL_DS].b |= (GUEST_PL << 13); 139 cpu->arch.gdt[GDT_ENTRY_KERNEL_DS].b |= (GUEST_PL << 13);
121} 140}
122 141
123/*H:650 An optimization of copy_gdt(), for just the three "thead-local storage" 142/*H:650
124 * entries. */ 143 * An optimization of copy_gdt(), for just the three "thead-local storage"
144 * entries.
145 */
125void copy_gdt_tls(const struct lg_cpu *cpu, struct desc_struct *gdt) 146void copy_gdt_tls(const struct lg_cpu *cpu, struct desc_struct *gdt)
126{ 147{
127 unsigned int i; 148 unsigned int i;
@@ -130,26 +151,34 @@ void copy_gdt_tls(const struct lg_cpu *cpu, struct desc_struct *gdt)
130 gdt[i] = cpu->arch.gdt[i]; 151 gdt[i] = cpu->arch.gdt[i];
131} 152}
132 153
133/*H:640 When the Guest is run on a different CPU, or the GDT entries have 154/*H:640
134 * changed, copy_gdt() is called to copy the Guest's GDT entries across to this 155 * When the Guest is run on a different CPU, or the GDT entries have changed,
135 * CPU's GDT. */ 156 * copy_gdt() is called to copy the Guest's GDT entries across to this CPU's
157 * GDT.
158 */
136void copy_gdt(const struct lg_cpu *cpu, struct desc_struct *gdt) 159void copy_gdt(const struct lg_cpu *cpu, struct desc_struct *gdt)
137{ 160{
138 unsigned int i; 161 unsigned int i;
139 162
140 /* The default entries from setup_default_gdt_entries() are not 163 /*
141 * replaced. See ignored_gdt() above. */ 164 * The default entries from setup_default_gdt_entries() are not
165 * replaced. See ignored_gdt() above.
166 */
142 for (i = 0; i < GDT_ENTRIES; i++) 167 for (i = 0; i < GDT_ENTRIES; i++)
143 if (!ignored_gdt(i)) 168 if (!ignored_gdt(i))
144 gdt[i] = cpu->arch.gdt[i]; 169 gdt[i] = cpu->arch.gdt[i];
145} 170}
146 171
147/*H:620 This is where the Guest asks us to load a new GDT entry 172/*H:620
148 * (LHCALL_LOAD_GDT_ENTRY). We tweak the entry and copy it in. */ 173 * This is where the Guest asks us to load a new GDT entry
174 * (LHCALL_LOAD_GDT_ENTRY). We tweak the entry and copy it in.
175 */
149void load_guest_gdt_entry(struct lg_cpu *cpu, u32 num, u32 lo, u32 hi) 176void load_guest_gdt_entry(struct lg_cpu *cpu, u32 num, u32 lo, u32 hi)
150{ 177{
151 /* We assume the Guest has the same number of GDT entries as the 178 /*
152 * Host, otherwise we'd have to dynamically allocate the Guest GDT. */ 179 * We assume the Guest has the same number of GDT entries as the
180 * Host, otherwise we'd have to dynamically allocate the Guest GDT.
181 */
153 if (num >= ARRAY_SIZE(cpu->arch.gdt)) 182 if (num >= ARRAY_SIZE(cpu->arch.gdt))
154 kill_guest(cpu, "too many gdt entries %i", num); 183 kill_guest(cpu, "too many gdt entries %i", num);
155 184
@@ -157,15 +186,19 @@ void load_guest_gdt_entry(struct lg_cpu *cpu, u32 num, u32 lo, u32 hi)
157 cpu->arch.gdt[num].a = lo; 186 cpu->arch.gdt[num].a = lo;
158 cpu->arch.gdt[num].b = hi; 187 cpu->arch.gdt[num].b = hi;
159 fixup_gdt_table(cpu, num, num+1); 188 fixup_gdt_table(cpu, num, num+1);
160 /* Mark that the GDT changed so the core knows it has to copy it again, 189 /*
161 * even if the Guest is run on the same CPU. */ 190 * Mark that the GDT changed so the core knows it has to copy it again,
191 * even if the Guest is run on the same CPU.
192 */
162 cpu->changed |= CHANGED_GDT; 193 cpu->changed |= CHANGED_GDT;
163} 194}
164 195
165/* This is the fast-track version for just changing the three TLS entries. 196/*
197 * This is the fast-track version for just changing the three TLS entries.
166 * Remember that this happens on every context switch, so it's worth 198 * Remember that this happens on every context switch, so it's worth
167 * optimizing. But wouldn't it be neater to have a single hypercall to cover 199 * optimizing. But wouldn't it be neater to have a single hypercall to cover
168 * both cases? */ 200 * both cases?
201 */
169void guest_load_tls(struct lg_cpu *cpu, unsigned long gtls) 202void guest_load_tls(struct lg_cpu *cpu, unsigned long gtls)
170{ 203{
171 struct desc_struct *tls = &cpu->arch.gdt[GDT_ENTRY_TLS_MIN]; 204 struct desc_struct *tls = &cpu->arch.gdt[GDT_ENTRY_TLS_MIN];
@@ -175,7 +208,6 @@ void guest_load_tls(struct lg_cpu *cpu, unsigned long gtls)
175 /* Note that just the TLS entries have changed. */ 208 /* Note that just the TLS entries have changed. */
176 cpu->changed |= CHANGED_GDT_TLS; 209 cpu->changed |= CHANGED_GDT_TLS;
177} 210}
178/*:*/
179 211
180/*H:660 212/*H:660
181 * With this, we have finished the Host. 213 * With this, we have finished the Host.
diff --git a/drivers/lguest/x86/core.c b/drivers/lguest/x86/core.c
index eaf722fe309a..96f7d88ec7f8 100644
--- a/drivers/lguest/x86/core.c
+++ b/drivers/lguest/x86/core.c
@@ -17,13 +17,15 @@
17 * along with this program; if not, write to the Free Software 17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. 18 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
19 */ 19 */
20/*P:450 This file contains the x86-specific lguest code. It used to be all 20/*P:450
21 * This file contains the x86-specific lguest code. It used to be all
21 * mixed in with drivers/lguest/core.c but several foolhardy code slashers 22 * mixed in with drivers/lguest/core.c but several foolhardy code slashers
22 * wrestled most of the dependencies out to here in preparation for porting 23 * wrestled most of the dependencies out to here in preparation for porting
23 * lguest to other architectures (see what I mean by foolhardy?). 24 * lguest to other architectures (see what I mean by foolhardy?).
24 * 25 *
25 * This also contains a couple of non-obvious setup and teardown pieces which 26 * This also contains a couple of non-obvious setup and teardown pieces which
26 * were implemented after days of debugging pain. :*/ 27 * were implemented after days of debugging pain.
28:*/
27#include <linux/kernel.h> 29#include <linux/kernel.h>
28#include <linux/start_kernel.h> 30#include <linux/start_kernel.h>
29#include <linux/string.h> 31#include <linux/string.h>
@@ -82,25 +84,33 @@ static DEFINE_PER_CPU(struct lg_cpu *, last_cpu);
82 */ 84 */
83static void copy_in_guest_info(struct lg_cpu *cpu, struct lguest_pages *pages) 85static void copy_in_guest_info(struct lg_cpu *cpu, struct lguest_pages *pages)
84{ 86{
85 /* Copying all this data can be quite expensive. We usually run the 87 /*
88 * Copying all this data can be quite expensive. We usually run the
86 * same Guest we ran last time (and that Guest hasn't run anywhere else 89 * same Guest we ran last time (and that Guest hasn't run anywhere else
87 * meanwhile). If that's not the case, we pretend everything in the 90 * meanwhile). If that's not the case, we pretend everything in the
88 * Guest has changed. */ 91 * Guest has changed.
92 */
89 if (__get_cpu_var(last_cpu) != cpu || cpu->last_pages != pages) { 93 if (__get_cpu_var(last_cpu) != cpu || cpu->last_pages != pages) {
90 __get_cpu_var(last_cpu) = cpu; 94 __get_cpu_var(last_cpu) = cpu;
91 cpu->last_pages = pages; 95 cpu->last_pages = pages;
92 cpu->changed = CHANGED_ALL; 96 cpu->changed = CHANGED_ALL;
93 } 97 }
94 98
95 /* These copies are pretty cheap, so we do them unconditionally: */ 99 /*
96 /* Save the current Host top-level page directory. */ 100 * These copies are pretty cheap, so we do them unconditionally: */
101 /* Save the current Host top-level page directory.
102 */
97 pages->state.host_cr3 = __pa(current->mm->pgd); 103 pages->state.host_cr3 = __pa(current->mm->pgd);
98 /* Set up the Guest's page tables to see this CPU's pages (and no 104 /*
99 * other CPU's pages). */ 105 * Set up the Guest's page tables to see this CPU's pages (and no
106 * other CPU's pages).
107 */
100 map_switcher_in_guest(cpu, pages); 108 map_switcher_in_guest(cpu, pages);
101 /* Set up the two "TSS" members which tell the CPU what stack to use 109 /*
110 * Set up the two "TSS" members which tell the CPU what stack to use
102 * for traps which do directly into the Guest (ie. traps at privilege 111 * for traps which do directly into the Guest (ie. traps at privilege
103 * level 1). */ 112 * level 1).
113 */
104 pages->state.guest_tss.sp1 = cpu->esp1; 114 pages->state.guest_tss.sp1 = cpu->esp1;
105 pages->state.guest_tss.ss1 = cpu->ss1; 115 pages->state.guest_tss.ss1 = cpu->ss1;
106 116
@@ -125,40 +135,53 @@ static void run_guest_once(struct lg_cpu *cpu, struct lguest_pages *pages)
125 /* This is a dummy value we need for GCC's sake. */ 135 /* This is a dummy value we need for GCC's sake. */
126 unsigned int clobber; 136 unsigned int clobber;
127 137
128 /* Copy the guest-specific information into this CPU's "struct 138 /*
129 * lguest_pages". */ 139 * Copy the guest-specific information into this CPU's "struct
140 * lguest_pages".
141 */
130 copy_in_guest_info(cpu, pages); 142 copy_in_guest_info(cpu, pages);
131 143
132 /* Set the trap number to 256 (impossible value). If we fault while 144 /*
145 * Set the trap number to 256 (impossible value). If we fault while
133 * switching to the Guest (bad segment registers or bug), this will 146 * switching to the Guest (bad segment registers or bug), this will
134 * cause us to abort the Guest. */ 147 * cause us to abort the Guest.
148 */
135 cpu->regs->trapnum = 256; 149 cpu->regs->trapnum = 256;
136 150
137 /* Now: we push the "eflags" register on the stack, then do an "lcall". 151 /*
152 * Now: we push the "eflags" register on the stack, then do an "lcall".
138 * This is how we change from using the kernel code segment to using 153 * This is how we change from using the kernel code segment to using
139 * the dedicated lguest code segment, as well as jumping into the 154 * the dedicated lguest code segment, as well as jumping into the
140 * Switcher. 155 * Switcher.
141 * 156 *
142 * The lcall also pushes the old code segment (KERNEL_CS) onto the 157 * The lcall also pushes the old code segment (KERNEL_CS) onto the
143 * stack, then the address of this call. This stack layout happens to 158 * stack, then the address of this call. This stack layout happens to
144 * exactly match the stack layout created by an interrupt... */ 159 * exactly match the stack layout created by an interrupt...
160 */
145 asm volatile("pushf; lcall *lguest_entry" 161 asm volatile("pushf; lcall *lguest_entry"
146 /* This is how we tell GCC that %eax ("a") and %ebx ("b") 162 /*
147 * are changed by this routine. The "=" means output. */ 163 * This is how we tell GCC that %eax ("a") and %ebx ("b")
164 * are changed by this routine. The "=" means output.
165 */
148 : "=a"(clobber), "=b"(clobber) 166 : "=a"(clobber), "=b"(clobber)
149 /* %eax contains the pages pointer. ("0" refers to the 167 /*
168 * %eax contains the pages pointer. ("0" refers to the
150 * 0-th argument above, ie "a"). %ebx contains the 169 * 0-th argument above, ie "a"). %ebx contains the
151 * physical address of the Guest's top-level page 170 * physical address of the Guest's top-level page
152 * directory. */ 171 * directory.
172 */
153 : "0"(pages), "1"(__pa(cpu->lg->pgdirs[cpu->cpu_pgd].pgdir)) 173 : "0"(pages), "1"(__pa(cpu->lg->pgdirs[cpu->cpu_pgd].pgdir))
154 /* We tell gcc that all these registers could change, 174 /*
175 * We tell gcc that all these registers could change,
155 * which means we don't have to save and restore them in 176 * which means we don't have to save and restore them in
156 * the Switcher. */ 177 * the Switcher.
178 */
157 : "memory", "%edx", "%ecx", "%edi", "%esi"); 179 : "memory", "%edx", "%ecx", "%edi", "%esi");
158} 180}
159/*:*/ 181/*:*/
160 182
161/*M:002 There are hooks in the scheduler which we can register to tell when we 183/*M:002
184 * There are hooks in the scheduler which we can register to tell when we
162 * get kicked off the CPU (preempt_notifier_register()). This would allow us 185 * get kicked off the CPU (preempt_notifier_register()). This would allow us
163 * to lazily disable SYSENTER which would regain some performance, and should 186 * to lazily disable SYSENTER which would regain some performance, and should
164 * also simplify copy_in_guest_info(). Note that we'd still need to restore 187 * also simplify copy_in_guest_info(). Note that we'd still need to restore
@@ -166,56 +189,72 @@ static void run_guest_once(struct lg_cpu *cpu, struct lguest_pages *pages)
166 * 189 *
167 * We could also try using this hooks for PGE, but that might be too expensive. 190 * We could also try using this hooks for PGE, but that might be too expensive.
168 * 191 *
169 * The hooks were designed for KVM, but we can also put them to good use. :*/ 192 * The hooks were designed for KVM, but we can also put them to good use.
193:*/
170 194
171/*H:040 This is the i386-specific code to setup and run the Guest. Interrupts 195/*H:040
172 * are disabled: we own the CPU. */ 196 * This is the i386-specific code to setup and run the Guest. Interrupts
197 * are disabled: we own the CPU.
198 */
173void lguest_arch_run_guest(struct lg_cpu *cpu) 199void lguest_arch_run_guest(struct lg_cpu *cpu)
174{ 200{
175 /* Remember the awfully-named TS bit? If the Guest has asked to set it 201 /*
202 * Remember the awfully-named TS bit? If the Guest has asked to set it
176 * we set it now, so we can trap and pass that trap to the Guest if it 203 * we set it now, so we can trap and pass that trap to the Guest if it
177 * uses the FPU. */ 204 * uses the FPU.
205 */
178 if (cpu->ts) 206 if (cpu->ts)
179 unlazy_fpu(current); 207 unlazy_fpu(current);
180 208
181 /* SYSENTER is an optimized way of doing system calls. We can't allow 209 /*
210 * SYSENTER is an optimized way of doing system calls. We can't allow
182 * it because it always jumps to privilege level 0. A normal Guest 211 * it because it always jumps to privilege level 0. A normal Guest
183 * won't try it because we don't advertise it in CPUID, but a malicious 212 * won't try it because we don't advertise it in CPUID, but a malicious
184 * Guest (or malicious Guest userspace program) could, so we tell the 213 * Guest (or malicious Guest userspace program) could, so we tell the
185 * CPU to disable it before running the Guest. */ 214 * CPU to disable it before running the Guest.
215 */
186 if (boot_cpu_has(X86_FEATURE_SEP)) 216 if (boot_cpu_has(X86_FEATURE_SEP))
187 wrmsr(MSR_IA32_SYSENTER_CS, 0, 0); 217 wrmsr(MSR_IA32_SYSENTER_CS, 0, 0);
188 218
189 /* Now we actually run the Guest. It will return when something 219 /*
220 * Now we actually run the Guest. It will return when something
190 * interesting happens, and we can examine its registers to see what it 221 * interesting happens, and we can examine its registers to see what it
191 * was doing. */ 222 * was doing.
223 */
192 run_guest_once(cpu, lguest_pages(raw_smp_processor_id())); 224 run_guest_once(cpu, lguest_pages(raw_smp_processor_id()));
193 225
194 /* Note that the "regs" structure contains two extra entries which are 226 /*
227 * Note that the "regs" structure contains two extra entries which are
195 * not really registers: a trap number which says what interrupt or 228 * not really registers: a trap number which says what interrupt or
196 * trap made the switcher code come back, and an error code which some 229 * trap made the switcher code come back, and an error code which some
197 * traps set. */ 230 * traps set.
231 */
198 232
199 /* Restore SYSENTER if it's supposed to be on. */ 233 /* Restore SYSENTER if it's supposed to be on. */
200 if (boot_cpu_has(X86_FEATURE_SEP)) 234 if (boot_cpu_has(X86_FEATURE_SEP))
201 wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0); 235 wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
202 236
203 /* If the Guest page faulted, then the cr2 register will tell us the 237 /*
238 * If the Guest page faulted, then the cr2 register will tell us the
204 * bad virtual address. We have to grab this now, because once we 239 * bad virtual address. We have to grab this now, because once we
205 * re-enable interrupts an interrupt could fault and thus overwrite 240 * re-enable interrupts an interrupt could fault and thus overwrite
206 * cr2, or we could even move off to a different CPU. */ 241 * cr2, or we could even move off to a different CPU.
242 */
207 if (cpu->regs->trapnum == 14) 243 if (cpu->regs->trapnum == 14)
208 cpu->arch.last_pagefault = read_cr2(); 244 cpu->arch.last_pagefault = read_cr2();
209 /* Similarly, if we took a trap because the Guest used the FPU, 245 /*
246 * Similarly, if we took a trap because the Guest used the FPU,
210 * we have to restore the FPU it expects to see. 247 * we have to restore the FPU it expects to see.
211 * math_state_restore() may sleep and we may even move off to 248 * math_state_restore() may sleep and we may even move off to
212 * a different CPU. So all the critical stuff should be done 249 * a different CPU. So all the critical stuff should be done
213 * before this. */ 250 * before this.
251 */
214 else if (cpu->regs->trapnum == 7) 252 else if (cpu->regs->trapnum == 7)
215 math_state_restore(); 253 math_state_restore();
216} 254}
217 255
218/*H:130 Now we've examined the hypercall code; our Guest can make requests. 256/*H:130
257 * Now we've examined the hypercall code; our Guest can make requests.
219 * Our Guest is usually so well behaved; it never tries to do things it isn't 258 * Our Guest is usually so well behaved; it never tries to do things it isn't
220 * allowed to, and uses hypercalls instead. Unfortunately, Linux's paravirtual 259 * allowed to, and uses hypercalls instead. Unfortunately, Linux's paravirtual
221 * infrastructure isn't quite complete, because it doesn't contain replacements 260 * infrastructure isn't quite complete, because it doesn't contain replacements
@@ -225,26 +264,33 @@ void lguest_arch_run_guest(struct lg_cpu *cpu)
225 * 264 *
226 * When the Guest uses one of these instructions, we get a trap (General 265 * When the Guest uses one of these instructions, we get a trap (General
227 * Protection Fault) and come here. We see if it's one of those troublesome 266 * Protection Fault) and come here. We see if it's one of those troublesome
228 * instructions and skip over it. We return true if we did. */ 267 * instructions and skip over it. We return true if we did.
268 */
229static int emulate_insn(struct lg_cpu *cpu) 269static int emulate_insn(struct lg_cpu *cpu)
230{ 270{
231 u8 insn; 271 u8 insn;
232 unsigned int insnlen = 0, in = 0, shift = 0; 272 unsigned int insnlen = 0, in = 0, shift = 0;
233 /* The eip contains the *virtual* address of the Guest's instruction: 273 /*
234 * guest_pa just subtracts the Guest's page_offset. */ 274 * The eip contains the *virtual* address of the Guest's instruction:
275 * guest_pa just subtracts the Guest's page_offset.
276 */
235 unsigned long physaddr = guest_pa(cpu, cpu->regs->eip); 277 unsigned long physaddr = guest_pa(cpu, cpu->regs->eip);
236 278
237 /* This must be the Guest kernel trying to do something, not userspace! 279 /*
280 * This must be the Guest kernel trying to do something, not userspace!
238 * The bottom two bits of the CS segment register are the privilege 281 * The bottom two bits of the CS segment register are the privilege
239 * level. */ 282 * level.
283 */
240 if ((cpu->regs->cs & 3) != GUEST_PL) 284 if ((cpu->regs->cs & 3) != GUEST_PL)
241 return 0; 285 return 0;
242 286
243 /* Decoding x86 instructions is icky. */ 287 /* Decoding x86 instructions is icky. */
244 insn = lgread(cpu, physaddr, u8); 288 insn = lgread(cpu, physaddr, u8);
245 289
246 /* 0x66 is an "operand prefix". It means it's using the upper 16 bits 290 /*
247 of the eax register. */ 291 * 0x66 is an "operand prefix". It means it's using the upper 16 bits
292 * of the eax register.
293 */
248 if (insn == 0x66) { 294 if (insn == 0x66) {
249 shift = 16; 295 shift = 16;
250 /* The instruction is 1 byte so far, read the next byte. */ 296 /* The instruction is 1 byte so far, read the next byte. */
@@ -252,8 +298,10 @@ static int emulate_insn(struct lg_cpu *cpu)
252 insn = lgread(cpu, physaddr + insnlen, u8); 298 insn = lgread(cpu, physaddr + insnlen, u8);
253 } 299 }
254 300
255 /* We can ignore the lower bit for the moment and decode the 4 opcodes 301 /*
256 * we need to emulate. */ 302 * We can ignore the lower bit for the moment and decode the 4 opcodes
303 * we need to emulate.
304 */
257 switch (insn & 0xFE) { 305 switch (insn & 0xFE) {
258 case 0xE4: /* in <next byte>,%al */ 306 case 0xE4: /* in <next byte>,%al */
259 insnlen += 2; 307 insnlen += 2;
@@ -274,9 +322,11 @@ static int emulate_insn(struct lg_cpu *cpu)
274 return 0; 322 return 0;
275 } 323 }
276 324
277 /* If it was an "IN" instruction, they expect the result to be read 325 /*
326 * If it was an "IN" instruction, they expect the result to be read
278 * into %eax, so we change %eax. We always return all-ones, which 327 * into %eax, so we change %eax. We always return all-ones, which
279 * traditionally means "there's nothing there". */ 328 * traditionally means "there's nothing there".
329 */
280 if (in) { 330 if (in) {
281 /* Lower bit tells is whether it's a 16 or 32 bit access */ 331 /* Lower bit tells is whether it's a 16 or 32 bit access */
282 if (insn & 0x1) 332 if (insn & 0x1)
@@ -290,7 +340,8 @@ static int emulate_insn(struct lg_cpu *cpu)
290 return 1; 340 return 1;
291} 341}
292 342
293/* Our hypercalls mechanism used to be based on direct software interrupts. 343/*
344 * Our hypercalls mechanism used to be based on direct software interrupts.
294 * After Anthony's "Refactor hypercall infrastructure" kvm patch, we decided to 345 * After Anthony's "Refactor hypercall infrastructure" kvm patch, we decided to
295 * change over to using kvm hypercalls. 346 * change over to using kvm hypercalls.
296 * 347 *
@@ -318,16 +369,20 @@ static int emulate_insn(struct lg_cpu *cpu)
318 */ 369 */
319static void rewrite_hypercall(struct lg_cpu *cpu) 370static void rewrite_hypercall(struct lg_cpu *cpu)
320{ 371{
321 /* This are the opcodes we use to patch the Guest. The opcode for "int 372 /*
373 * This are the opcodes we use to patch the Guest. The opcode for "int
322 * $0x1f" is "0xcd 0x1f" but vmcall instruction is 3 bytes long, so we 374 * $0x1f" is "0xcd 0x1f" but vmcall instruction is 3 bytes long, so we
323 * complete the sequence with a NOP (0x90). */ 375 * complete the sequence with a NOP (0x90).
376 */
324 u8 insn[3] = {0xcd, 0x1f, 0x90}; 377 u8 insn[3] = {0xcd, 0x1f, 0x90};
325 378
326 __lgwrite(cpu, guest_pa(cpu, cpu->regs->eip), insn, sizeof(insn)); 379 __lgwrite(cpu, guest_pa(cpu, cpu->regs->eip), insn, sizeof(insn));
327 /* The above write might have caused a copy of that page to be made 380 /*
381 * The above write might have caused a copy of that page to be made
328 * (if it was read-only). We need to make sure the Guest has 382 * (if it was read-only). We need to make sure the Guest has
329 * up-to-date pagetables. As this doesn't happen often, we can just 383 * up-to-date pagetables. As this doesn't happen often, we can just
330 * drop them all. */ 384 * drop them all.
385 */
331 guest_pagetable_clear_all(cpu); 386 guest_pagetable_clear_all(cpu);
332} 387}
333 388
@@ -335,9 +390,11 @@ static bool is_hypercall(struct lg_cpu *cpu)
335{ 390{
336 u8 insn[3]; 391 u8 insn[3];
337 392
338 /* This must be the Guest kernel trying to do something. 393 /*
394 * This must be the Guest kernel trying to do something.
339 * The bottom two bits of the CS segment register are the privilege 395 * The bottom two bits of the CS segment register are the privilege
340 * level. */ 396 * level.
397 */
341 if ((cpu->regs->cs & 3) != GUEST_PL) 398 if ((cpu->regs->cs & 3) != GUEST_PL)
342 return false; 399 return false;
343 400
@@ -351,86 +408,105 @@ void lguest_arch_handle_trap(struct lg_cpu *cpu)
351{ 408{
352 switch (cpu->regs->trapnum) { 409 switch (cpu->regs->trapnum) {
353 case 13: /* We've intercepted a General Protection Fault. */ 410 case 13: /* We've intercepted a General Protection Fault. */
354 /* Check if this was one of those annoying IN or OUT 411 /*
412 * Check if this was one of those annoying IN or OUT
355 * instructions which we need to emulate. If so, we just go 413 * instructions which we need to emulate. If so, we just go
356 * back into the Guest after we've done it. */ 414 * back into the Guest after we've done it.
415 */
357 if (cpu->regs->errcode == 0) { 416 if (cpu->regs->errcode == 0) {
358 if (emulate_insn(cpu)) 417 if (emulate_insn(cpu))
359 return; 418 return;
360 } 419 }
361 /* If KVM is active, the vmcall instruction triggers a 420 /*
362 * General Protection Fault. Normally it triggers an 421 * If KVM is active, the vmcall instruction triggers a General
363 * invalid opcode fault (6): */ 422 * Protection Fault. Normally it triggers an invalid opcode
423 * fault (6):
424 */
364 case 6: 425 case 6:
365 /* We need to check if ring == GUEST_PL and 426 /*
366 * faulting instruction == vmcall. */ 427 * We need to check if ring == GUEST_PL and faulting
428 * instruction == vmcall.
429 */
367 if (is_hypercall(cpu)) { 430 if (is_hypercall(cpu)) {
368 rewrite_hypercall(cpu); 431 rewrite_hypercall(cpu);
369 return; 432 return;
370 } 433 }
371 break; 434 break;
372 case 14: /* We've intercepted a Page Fault. */ 435 case 14: /* We've intercepted a Page Fault. */
373 /* The Guest accessed a virtual address that wasn't mapped. 436 /*
437 * The Guest accessed a virtual address that wasn't mapped.
374 * This happens a lot: we don't actually set up most of the page 438 * This happens a lot: we don't actually set up most of the page
375 * tables for the Guest at all when we start: as it runs it asks 439 * tables for the Guest at all when we start: as it runs it asks
376 * for more and more, and we set them up as required. In this 440 * for more and more, and we set them up as required. In this
377 * case, we don't even tell the Guest that the fault happened. 441 * case, we don't even tell the Guest that the fault happened.
378 * 442 *
379 * The errcode tells whether this was a read or a write, and 443 * The errcode tells whether this was a read or a write, and
380 * whether kernel or userspace code. */ 444 * whether kernel or userspace code.
445 */
381 if (demand_page(cpu, cpu->arch.last_pagefault, 446 if (demand_page(cpu, cpu->arch.last_pagefault,
382 cpu->regs->errcode)) 447 cpu->regs->errcode))
383 return; 448 return;
384 449
385 /* OK, it's really not there (or not OK): the Guest needs to 450 /*
451 * OK, it's really not there (or not OK): the Guest needs to
386 * know. We write out the cr2 value so it knows where the 452 * know. We write out the cr2 value so it knows where the
387 * fault occurred. 453 * fault occurred.
388 * 454 *
389 * Note that if the Guest were really messed up, this could 455 * Note that if the Guest were really messed up, this could
390 * happen before it's done the LHCALL_LGUEST_INIT hypercall, so 456 * happen before it's done the LHCALL_LGUEST_INIT hypercall, so
391 * lg->lguest_data could be NULL */ 457 * lg->lguest_data could be NULL
458 */
392 if (cpu->lg->lguest_data && 459 if (cpu->lg->lguest_data &&
393 put_user(cpu->arch.last_pagefault, 460 put_user(cpu->arch.last_pagefault,
394 &cpu->lg->lguest_data->cr2)) 461 &cpu->lg->lguest_data->cr2))
395 kill_guest(cpu, "Writing cr2"); 462 kill_guest(cpu, "Writing cr2");
396 break; 463 break;
397 case 7: /* We've intercepted a Device Not Available fault. */ 464 case 7: /* We've intercepted a Device Not Available fault. */
398 /* If the Guest doesn't want to know, we already restored the 465 /*
399 * Floating Point Unit, so we just continue without telling 466 * If the Guest doesn't want to know, we already restored the
400 * it. */ 467 * Floating Point Unit, so we just continue without telling it.
468 */
401 if (!cpu->ts) 469 if (!cpu->ts)
402 return; 470 return;
403 break; 471 break;
404 case 32 ... 255: 472 case 32 ... 255:
405 /* These values mean a real interrupt occurred, in which case 473 /*
474 * These values mean a real interrupt occurred, in which case
406 * the Host handler has already been run. We just do a 475 * the Host handler has already been run. We just do a
407 * friendly check if another process should now be run, then 476 * friendly check if another process should now be run, then
408 * return to run the Guest again */ 477 * return to run the Guest again
478 */
409 cond_resched(); 479 cond_resched();
410 return; 480 return;
411 case LGUEST_TRAP_ENTRY: 481 case LGUEST_TRAP_ENTRY:
412 /* Our 'struct hcall_args' maps directly over our regs: we set 482 /*
413 * up the pointer now to indicate a hypercall is pending. */ 483 * Our 'struct hcall_args' maps directly over our regs: we set
484 * up the pointer now to indicate a hypercall is pending.
485 */
414 cpu->hcall = (struct hcall_args *)cpu->regs; 486 cpu->hcall = (struct hcall_args *)cpu->regs;
415 return; 487 return;
416 } 488 }
417 489
418 /* We didn't handle the trap, so it needs to go to the Guest. */ 490 /* We didn't handle the trap, so it needs to go to the Guest. */
419 if (!deliver_trap(cpu, cpu->regs->trapnum)) 491 if (!deliver_trap(cpu, cpu->regs->trapnum))
420 /* If the Guest doesn't have a handler (either it hasn't 492 /*
493 * If the Guest doesn't have a handler (either it hasn't
421 * registered any yet, or it's one of the faults we don't let 494 * registered any yet, or it's one of the faults we don't let
422 * it handle), it dies with this cryptic error message. */ 495 * it handle), it dies with this cryptic error message.
496 */
423 kill_guest(cpu, "unhandled trap %li at %#lx (%#lx)", 497 kill_guest(cpu, "unhandled trap %li at %#lx (%#lx)",
424 cpu->regs->trapnum, cpu->regs->eip, 498 cpu->regs->trapnum, cpu->regs->eip,
425 cpu->regs->trapnum == 14 ? cpu->arch.last_pagefault 499 cpu->regs->trapnum == 14 ? cpu->arch.last_pagefault
426 : cpu->regs->errcode); 500 : cpu->regs->errcode);
427} 501}
428 502
429/* Now we can look at each of the routines this calls, in increasing order of 503/*
504 * Now we can look at each of the routines this calls, in increasing order of
430 * complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(), 505 * complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(),
431 * deliver_trap() and demand_page(). After all those, we'll be ready to 506 * deliver_trap() and demand_page(). After all those, we'll be ready to
432 * examine the Switcher, and our philosophical understanding of the Host/Guest 507 * examine the Switcher, and our philosophical understanding of the Host/Guest
433 * duality will be complete. :*/ 508 * duality will be complete.
509:*/
434static void adjust_pge(void *on) 510static void adjust_pge(void *on)
435{ 511{
436 if (on) 512 if (on)
@@ -439,13 +515,16 @@ static void adjust_pge(void *on)
439 write_cr4(read_cr4() & ~X86_CR4_PGE); 515 write_cr4(read_cr4() & ~X86_CR4_PGE);
440} 516}
441 517
442/*H:020 Now the Switcher is mapped and every thing else is ready, we need to do 518/*H:020
443 * some more i386-specific initialization. */ 519 * Now the Switcher is mapped and every thing else is ready, we need to do
520 * some more i386-specific initialization.
521 */
444void __init lguest_arch_host_init(void) 522void __init lguest_arch_host_init(void)
445{ 523{
446 int i; 524 int i;
447 525
448 /* Most of the i386/switcher.S doesn't care that it's been moved; on 526 /*
527 * Most of the i386/switcher.S doesn't care that it's been moved; on
449 * Intel, jumps are relative, and it doesn't access any references to 528 * Intel, jumps are relative, and it doesn't access any references to
450 * external code or data. 529 * external code or data.
451 * 530 *
@@ -453,7 +532,8 @@ void __init lguest_arch_host_init(void)
453 * addresses are placed in a table (default_idt_entries), so we need to 532 * addresses are placed in a table (default_idt_entries), so we need to
454 * update the table with the new addresses. switcher_offset() is a 533 * update the table with the new addresses. switcher_offset() is a
455 * convenience function which returns the distance between the 534 * convenience function which returns the distance between the
456 * compiled-in switcher code and the high-mapped copy we just made. */ 535 * compiled-in switcher code and the high-mapped copy we just made.
536 */
457 for (i = 0; i < IDT_ENTRIES; i++) 537 for (i = 0; i < IDT_ENTRIES; i++)
458 default_idt_entries[i] += switcher_offset(); 538 default_idt_entries[i] += switcher_offset();
459 539
@@ -468,63 +548,81 @@ void __init lguest_arch_host_init(void)
468 for_each_possible_cpu(i) { 548 for_each_possible_cpu(i) {
469 /* lguest_pages() returns this CPU's two pages. */ 549 /* lguest_pages() returns this CPU's two pages. */
470 struct lguest_pages *pages = lguest_pages(i); 550 struct lguest_pages *pages = lguest_pages(i);
471 /* This is a convenience pointer to make the code fit one 551 /* This is a convenience pointer to make the code neater. */
472 * statement to a line. */
473 struct lguest_ro_state *state = &pages->state; 552 struct lguest_ro_state *state = &pages->state;
474 553
475 /* The Global Descriptor Table: the Host has a different one 554 /*
555 * The Global Descriptor Table: the Host has a different one
476 * for each CPU. We keep a descriptor for the GDT which says 556 * for each CPU. We keep a descriptor for the GDT which says
477 * where it is and how big it is (the size is actually the last 557 * where it is and how big it is (the size is actually the last
478 * byte, not the size, hence the "-1"). */ 558 * byte, not the size, hence the "-1").
559 */
479 state->host_gdt_desc.size = GDT_SIZE-1; 560 state->host_gdt_desc.size = GDT_SIZE-1;
480 state->host_gdt_desc.address = (long)get_cpu_gdt_table(i); 561 state->host_gdt_desc.address = (long)get_cpu_gdt_table(i);
481 562
482 /* All CPUs on the Host use the same Interrupt Descriptor 563 /*
564 * All CPUs on the Host use the same Interrupt Descriptor
483 * Table, so we just use store_idt(), which gets this CPU's IDT 565 * Table, so we just use store_idt(), which gets this CPU's IDT
484 * descriptor. */ 566 * descriptor.
567 */
485 store_idt(&state->host_idt_desc); 568 store_idt(&state->host_idt_desc);
486 569
487 /* The descriptors for the Guest's GDT and IDT can be filled 570 /*
571 * The descriptors for the Guest's GDT and IDT can be filled
488 * out now, too. We copy the GDT & IDT into ->guest_gdt and 572 * out now, too. We copy the GDT & IDT into ->guest_gdt and
489 * ->guest_idt before actually running the Guest. */ 573 * ->guest_idt before actually running the Guest.
574 */
490 state->guest_idt_desc.size = sizeof(state->guest_idt)-1; 575 state->guest_idt_desc.size = sizeof(state->guest_idt)-1;
491 state->guest_idt_desc.address = (long)&state->guest_idt; 576 state->guest_idt_desc.address = (long)&state->guest_idt;
492 state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1; 577 state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1;
493 state->guest_gdt_desc.address = (long)&state->guest_gdt; 578 state->guest_gdt_desc.address = (long)&state->guest_gdt;
494 579
495 /* We know where we want the stack to be when the Guest enters 580 /*
581 * We know where we want the stack to be when the Guest enters
496 * the Switcher: in pages->regs. The stack grows upwards, so 582 * the Switcher: in pages->regs. The stack grows upwards, so
497 * we start it at the end of that structure. */ 583 * we start it at the end of that structure.
584 */
498 state->guest_tss.sp0 = (long)(&pages->regs + 1); 585 state->guest_tss.sp0 = (long)(&pages->regs + 1);
499 /* And this is the GDT entry to use for the stack: we keep a 586 /*
500 * couple of special LGUEST entries. */ 587 * And this is the GDT entry to use for the stack: we keep a
588 * couple of special LGUEST entries.
589 */
501 state->guest_tss.ss0 = LGUEST_DS; 590 state->guest_tss.ss0 = LGUEST_DS;
502 591
503 /* x86 can have a finegrained bitmap which indicates what I/O 592 /*
593 * x86 can have a finegrained bitmap which indicates what I/O
504 * ports the process can use. We set it to the end of our 594 * ports the process can use. We set it to the end of our
505 * structure, meaning "none". */ 595 * structure, meaning "none".
596 */
506 state->guest_tss.io_bitmap_base = sizeof(state->guest_tss); 597 state->guest_tss.io_bitmap_base = sizeof(state->guest_tss);
507 598
508 /* Some GDT entries are the same across all Guests, so we can 599 /*
509 * set them up now. */ 600 * Some GDT entries are the same across all Guests, so we can
601 * set them up now.
602 */
510 setup_default_gdt_entries(state); 603 setup_default_gdt_entries(state);
511 /* Most IDT entries are the same for all Guests, too.*/ 604 /* Most IDT entries are the same for all Guests, too.*/
512 setup_default_idt_entries(state, default_idt_entries); 605 setup_default_idt_entries(state, default_idt_entries);
513 606
514 /* The Host needs to be able to use the LGUEST segments on this 607 /*
515 * CPU, too, so put them in the Host GDT. */ 608 * The Host needs to be able to use the LGUEST segments on this
609 * CPU, too, so put them in the Host GDT.
610 */
516 get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT; 611 get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
517 get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT; 612 get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
518 } 613 }
519 614
520 /* In the Switcher, we want the %cs segment register to use the 615 /*
616 * In the Switcher, we want the %cs segment register to use the
521 * LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so 617 * LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so
522 * it will be undisturbed when we switch. To change %cs and jump we 618 * it will be undisturbed when we switch. To change %cs and jump we
523 * need this structure to feed to Intel's "lcall" instruction. */ 619 * need this structure to feed to Intel's "lcall" instruction.
620 */
524 lguest_entry.offset = (long)switch_to_guest + switcher_offset(); 621 lguest_entry.offset = (long)switch_to_guest + switcher_offset();
525 lguest_entry.segment = LGUEST_CS; 622 lguest_entry.segment = LGUEST_CS;
526 623
527 /* Finally, we need to turn off "Page Global Enable". PGE is an 624 /*
625 * Finally, we need to turn off "Page Global Enable". PGE is an
528 * optimization where page table entries are specially marked to show 626 * optimization where page table entries are specially marked to show
529 * they never change. The Host kernel marks all the kernel pages this 627 * they never change. The Host kernel marks all the kernel pages this
530 * way because it's always present, even when userspace is running. 628 * way because it's always present, even when userspace is running.
@@ -534,16 +632,21 @@ void __init lguest_arch_host_init(void)
534 * you'll get really weird bugs that you'll chase for two days. 632 * you'll get really weird bugs that you'll chase for two days.
535 * 633 *
536 * I used to turn PGE off every time we switched to the Guest and back 634 * I used to turn PGE off every time we switched to the Guest and back
537 * on when we return, but that slowed the Switcher down noticibly. */ 635 * on when we return, but that slowed the Switcher down noticibly.
636 */
538 637
539 /* We don't need the complexity of CPUs coming and going while we're 638 /*
540 * doing this. */ 639 * We don't need the complexity of CPUs coming and going while we're
640 * doing this.
641 */
541 get_online_cpus(); 642 get_online_cpus();
542 if (cpu_has_pge) { /* We have a broader idea of "global". */ 643 if (cpu_has_pge) { /* We have a broader idea of "global". */
543 /* Remember that this was originally set (for cleanup). */ 644 /* Remember that this was originally set (for cleanup). */
544 cpu_had_pge = 1; 645 cpu_had_pge = 1;
545 /* adjust_pge is a helper function which sets or unsets the PGE 646 /*
546 * bit on its CPU, depending on the argument (0 == unset). */ 647 * adjust_pge is a helper function which sets or unsets the PGE
648 * bit on its CPU, depending on the argument (0 == unset).
649 */
547 on_each_cpu(adjust_pge, (void *)0, 1); 650 on_each_cpu(adjust_pge, (void *)0, 1);
548 /* Turn off the feature in the global feature set. */ 651 /* Turn off the feature in the global feature set. */
549 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE); 652 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE);
@@ -590,26 +693,32 @@ int lguest_arch_init_hypercalls(struct lg_cpu *cpu)
590{ 693{
591 u32 tsc_speed; 694 u32 tsc_speed;
592 695
593 /* The pointer to the Guest's "struct lguest_data" is the only argument. 696 /*
594 * We check that address now. */ 697 * The pointer to the Guest's "struct lguest_data" is the only argument.
698 * We check that address now.
699 */
595 if (!lguest_address_ok(cpu->lg, cpu->hcall->arg1, 700 if (!lguest_address_ok(cpu->lg, cpu->hcall->arg1,
596 sizeof(*cpu->lg->lguest_data))) 701 sizeof(*cpu->lg->lguest_data)))
597 return -EFAULT; 702 return -EFAULT;
598 703
599 /* Having checked it, we simply set lg->lguest_data to point straight 704 /*
705 * Having checked it, we simply set lg->lguest_data to point straight
600 * into the Launcher's memory at the right place and then use 706 * into the Launcher's memory at the right place and then use
601 * copy_to_user/from_user from now on, instead of lgread/write. I put 707 * copy_to_user/from_user from now on, instead of lgread/write. I put
602 * this in to show that I'm not immune to writing stupid 708 * this in to show that I'm not immune to writing stupid
603 * optimizations. */ 709 * optimizations.
710 */
604 cpu->lg->lguest_data = cpu->lg->mem_base + cpu->hcall->arg1; 711 cpu->lg->lguest_data = cpu->lg->mem_base + cpu->hcall->arg1;
605 712
606 /* We insist that the Time Stamp Counter exist and doesn't change with 713 /*
714 * We insist that the Time Stamp Counter exist and doesn't change with
607 * cpu frequency. Some devious chip manufacturers decided that TSC 715 * cpu frequency. Some devious chip manufacturers decided that TSC
608 * changes could be handled in software. I decided that time going 716 * changes could be handled in software. I decided that time going
609 * backwards might be good for benchmarks, but it's bad for users. 717 * backwards might be good for benchmarks, but it's bad for users.
610 * 718 *
611 * We also insist that the TSC be stable: the kernel detects unreliable 719 * We also insist that the TSC be stable: the kernel detects unreliable
612 * TSCs for its own purposes, and we use that here. */ 720 * TSCs for its own purposes, and we use that here.
721 */
613 if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) && !check_tsc_unstable()) 722 if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) && !check_tsc_unstable())
614 tsc_speed = tsc_khz; 723 tsc_speed = tsc_khz;
615 else 724 else
@@ -625,38 +734,47 @@ int lguest_arch_init_hypercalls(struct lg_cpu *cpu)
625} 734}
626/*:*/ 735/*:*/
627 736
628/*L:030 lguest_arch_setup_regs() 737/*L:030
738 * lguest_arch_setup_regs()
629 * 739 *
630 * Most of the Guest's registers are left alone: we used get_zeroed_page() to 740 * Most of the Guest's registers are left alone: we used get_zeroed_page() to
631 * allocate the structure, so they will be 0. */ 741 * allocate the structure, so they will be 0.
742 */
632void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start) 743void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start)
633{ 744{
634 struct lguest_regs *regs = cpu->regs; 745 struct lguest_regs *regs = cpu->regs;
635 746
636 /* There are four "segment" registers which the Guest needs to boot: 747 /*
748 * There are four "segment" registers which the Guest needs to boot:
637 * The "code segment" register (cs) refers to the kernel code segment 749 * The "code segment" register (cs) refers to the kernel code segment
638 * __KERNEL_CS, and the "data", "extra" and "stack" segment registers 750 * __KERNEL_CS, and the "data", "extra" and "stack" segment registers
639 * refer to the kernel data segment __KERNEL_DS. 751 * refer to the kernel data segment __KERNEL_DS.
640 * 752 *
641 * The privilege level is packed into the lower bits. The Guest runs 753 * The privilege level is packed into the lower bits. The Guest runs
642 * at privilege level 1 (GUEST_PL).*/ 754 * at privilege level 1 (GUEST_PL).
755 */
643 regs->ds = regs->es = regs->ss = __KERNEL_DS|GUEST_PL; 756 regs->ds = regs->es = regs->ss = __KERNEL_DS|GUEST_PL;
644 regs->cs = __KERNEL_CS|GUEST_PL; 757 regs->cs = __KERNEL_CS|GUEST_PL;
645 758
646 /* The "eflags" register contains miscellaneous flags. Bit 1 (0x002) 759 /*
760 * The "eflags" register contains miscellaneous flags. Bit 1 (0x002)
647 * is supposed to always be "1". Bit 9 (0x200) controls whether 761 * is supposed to always be "1". Bit 9 (0x200) controls whether
648 * interrupts are enabled. We always leave interrupts enabled while 762 * interrupts are enabled. We always leave interrupts enabled while
649 * running the Guest. */ 763 * running the Guest.
764 */
650 regs->eflags = X86_EFLAGS_IF | 0x2; 765 regs->eflags = X86_EFLAGS_IF | 0x2;
651 766
652 /* The "Extended Instruction Pointer" register says where the Guest is 767 /*
653 * running. */ 768 * The "Extended Instruction Pointer" register says where the Guest is
769 * running.
770 */
654 regs->eip = start; 771 regs->eip = start;
655 772
656 /* %esi points to our boot information, at physical address 0, so don't 773 /*
657 * touch it. */ 774 * %esi points to our boot information, at physical address 0, so don't
775 * touch it.
776 */
658 777
659 /* There are a couple of GDT entries the Guest expects when first 778 /* There are a couple of GDT entries the Guest expects at boot. */
660 * booting. */
661 setup_guest_gdt(cpu); 779 setup_guest_gdt(cpu);
662} 780}
diff --git a/drivers/lguest/x86/switcher_32.S b/drivers/lguest/x86/switcher_32.S
index 3fc15318a80f..6dec09793836 100644
--- a/drivers/lguest/x86/switcher_32.S
+++ b/drivers/lguest/x86/switcher_32.S
@@ -1,12 +1,15 @@
1/*P:900 This is the Switcher: code which sits at 0xFFC00000 astride both the 1/*P:900
2 * This is the Switcher: code which sits at 0xFFC00000 astride both the
2 * Host and Guest to do the low-level Guest<->Host switch. It is as simple as 3 * Host and Guest to do the low-level Guest<->Host switch. It is as simple as
3 * it can be made, but it's naturally very specific to x86. 4 * it can be made, but it's naturally very specific to x86.
4 * 5 *
5 * You have now completed Preparation. If this has whet your appetite; if you 6 * You have now completed Preparation. If this has whet your appetite; if you
6 * are feeling invigorated and refreshed then the next, more challenging stage 7 * are feeling invigorated and refreshed then the next, more challenging stage
7 * can be found in "make Guest". :*/ 8 * can be found in "make Guest".
9 :*/
8 10
9/*M:012 Lguest is meant to be simple: my rule of thumb is that 1% more LOC must 11/*M:012
12 * Lguest is meant to be simple: my rule of thumb is that 1% more LOC must
10 * gain at least 1% more performance. Since neither LOC nor performance can be 13 * gain at least 1% more performance. Since neither LOC nor performance can be
11 * measured beforehand, it generally means implementing a feature then deciding 14 * measured beforehand, it generally means implementing a feature then deciding
12 * if it's worth it. And once it's implemented, who can say no? 15 * if it's worth it. And once it's implemented, who can say no?
@@ -31,11 +34,14 @@
31 * Host (which is actually really easy). 34 * Host (which is actually really easy).
32 * 35 *
33 * Two questions remain. Would the performance gain outweigh the complexity? 36 * Two questions remain. Would the performance gain outweigh the complexity?
34 * And who would write the verse documenting it? :*/ 37 * And who would write the verse documenting it?
38:*/
35 39
36/*M:011 Lguest64 handles NMI. This gave me NMI envy (until I looked at their 40/*M:011
41 * Lguest64 handles NMI. This gave me NMI envy (until I looked at their
37 * code). It's worth doing though, since it would let us use oprofile in the 42 * code). It's worth doing though, since it would let us use oprofile in the
38 * Host when a Guest is running. :*/ 43 * Host when a Guest is running.
44:*/
39 45
40/*S:100 46/*S:100
41 * Welcome to the Switcher itself! 47 * Welcome to the Switcher itself!
diff --git a/include/linux/lguest.h b/include/linux/lguest.h
index dbf2479e808e..0a3a11afd64b 100644
--- a/include/linux/lguest.h
+++ b/include/linux/lguest.h
@@ -1,5 +1,7 @@
1/* Things the lguest guest needs to know. Note: like all lguest interfaces, 1/*
2 * this is subject to wild and random change between versions. */ 2 * Things the lguest guest needs to know. Note: like all lguest interfaces,
3 * this is subject to wild and random change between versions.
4 */
3#ifndef _LINUX_LGUEST_H 5#ifndef _LINUX_LGUEST_H
4#define _LINUX_LGUEST_H 6#define _LINUX_LGUEST_H
5 7
@@ -11,32 +13,42 @@
11#define LG_CLOCK_MIN_DELTA 100UL 13#define LG_CLOCK_MIN_DELTA 100UL
12#define LG_CLOCK_MAX_DELTA ULONG_MAX 14#define LG_CLOCK_MAX_DELTA ULONG_MAX
13 15
14/*G:031 The second method of communicating with the Host is to via "struct 16/*G:031
17 * The second method of communicating with the Host is to via "struct
15 * lguest_data". Once the Guest's initialization hypercall tells the Host where 18 * lguest_data". Once the Guest's initialization hypercall tells the Host where
16 * this is, the Guest and Host both publish information in it. :*/ 19 * this is, the Guest and Host both publish information in it.
20:*/
17struct lguest_data 21struct lguest_data
18{ 22{
19 /* 512 == enabled (same as eflags in normal hardware). The Guest 23 /*
20 * changes interrupts so often that a hypercall is too slow. */ 24 * 512 == enabled (same as eflags in normal hardware). The Guest
25 * changes interrupts so often that a hypercall is too slow.
26 */
21 unsigned int irq_enabled; 27 unsigned int irq_enabled;
22 /* Fine-grained interrupt disabling by the Guest */ 28 /* Fine-grained interrupt disabling by the Guest */
23 DECLARE_BITMAP(blocked_interrupts, LGUEST_IRQS); 29 DECLARE_BITMAP(blocked_interrupts, LGUEST_IRQS);
24 30
25 /* The Host writes the virtual address of the last page fault here, 31 /*
32 * The Host writes the virtual address of the last page fault here,
26 * which saves the Guest a hypercall. CR2 is the native register where 33 * which saves the Guest a hypercall. CR2 is the native register where
27 * this address would normally be found. */ 34 * this address would normally be found.
35 */
28 unsigned long cr2; 36 unsigned long cr2;
29 37
30 /* Wallclock time set by the Host. */ 38 /* Wallclock time set by the Host. */
31 struct timespec time; 39 struct timespec time;
32 40
33 /* Interrupt pending set by the Host. The Guest should do a hypercall 41 /*
34 * if it re-enables interrupts and sees this set (to X86_EFLAGS_IF). */ 42 * Interrupt pending set by the Host. The Guest should do a hypercall
43 * if it re-enables interrupts and sees this set (to X86_EFLAGS_IF).
44 */
35 int irq_pending; 45 int irq_pending;
36 46
37 /* Async hypercall ring. Instead of directly making hypercalls, we can 47 /*
48 * Async hypercall ring. Instead of directly making hypercalls, we can
38 * place them in here for processing the next time the Host wants. 49 * place them in here for processing the next time the Host wants.
39 * This batching can be quite efficient. */ 50 * This batching can be quite efficient.
51 */
40 52
41 /* 0xFF == done (set by Host), 0 == pending (set by Guest). */ 53 /* 0xFF == done (set by Host), 0 == pending (set by Guest). */
42 u8 hcall_status[LHCALL_RING_SIZE]; 54 u8 hcall_status[LHCALL_RING_SIZE];
diff --git a/include/linux/lguest_launcher.h b/include/linux/lguest_launcher.h
index bfefbdf7498a..495203ff221c 100644
--- a/include/linux/lguest_launcher.h
+++ b/include/linux/lguest_launcher.h
@@ -29,8 +29,10 @@ struct lguest_device_desc {
29 __u8 type; 29 __u8 type;
30 /* The number of virtqueues (first in config array) */ 30 /* The number of virtqueues (first in config array) */
31 __u8 num_vq; 31 __u8 num_vq;
32 /* The number of bytes of feature bits. Multiply by 2: one for host 32 /*
33 * features and one for Guest acknowledgements. */ 33 * The number of bytes of feature bits. Multiply by 2: one for host
34 * features and one for Guest acknowledgements.
35 */
34 __u8 feature_len; 36 __u8 feature_len;
35 /* The number of bytes of the config array after virtqueues. */ 37 /* The number of bytes of the config array after virtqueues. */
36 __u8 config_len; 38 __u8 config_len;
@@ -39,8 +41,10 @@ struct lguest_device_desc {
39 __u8 config[0]; 41 __u8 config[0];
40}; 42};
41 43
42/*D:135 This is how we expect the device configuration field for a virtqueue 44/*D:135
43 * to be laid out in config space. */ 45 * This is how we expect the device configuration field for a virtqueue
46 * to be laid out in config space.
47 */
44struct lguest_vqconfig { 48struct lguest_vqconfig {
45 /* The number of entries in the virtio_ring */ 49 /* The number of entries in the virtio_ring */
46 __u16 num; 50 __u16 num;
@@ -61,7 +65,9 @@ enum lguest_req
61 LHREQ_EVENTFD, /* + address, fd. */ 65 LHREQ_EVENTFD, /* + address, fd. */
62}; 66};
63 67
64/* The alignment to use between consumer and producer parts of vring. 68/*
65 * x86 pagesize for historical reasons. */ 69 * The alignment to use between consumer and producer parts of vring.
70 * x86 pagesize for historical reasons.
71 */
66#define LGUEST_VRING_ALIGN 4096 72#define LGUEST_VRING_ALIGN 4096
67#endif /* _LINUX_LGUEST_LAUNCHER */ 73#endif /* _LINUX_LGUEST_LAUNCHER */