aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/lguest
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/lguest')
-rw-r--r--drivers/lguest/Makefile2
-rw-r--r--drivers/lguest/core.c459
-rw-r--r--drivers/lguest/interrupts_and_traps.c18
-rw-r--r--drivers/lguest/lg.h63
-rw-r--r--drivers/lguest/segments.c26
-rw-r--r--drivers/lguest/x86/core.c476
-rw-r--r--drivers/lguest/x86/switcher_32.S3
7 files changed, 525 insertions, 522 deletions
diff --git a/drivers/lguest/Makefile b/drivers/lguest/Makefile
index a4567c99991b..d330f5b8c456 100644
--- a/drivers/lguest/Makefile
+++ b/drivers/lguest/Makefile
@@ -6,7 +6,7 @@ obj-$(CONFIG_LGUEST) += lg.o
6lg-y = core.o hypercalls.o page_tables.o interrupts_and_traps.o \ 6lg-y = core.o hypercalls.o page_tables.o interrupts_and_traps.o \
7 segments.o io.o lguest_user.o 7 segments.o io.o lguest_user.o
8 8
9lg-$(CONFIG_X86_32) += x86/switcher_32.o 9lg-$(CONFIG_X86_32) += x86/switcher_32.o x86/core.o
10 10
11Preparation Preparation!: PREFIX=P 11Preparation Preparation!: PREFIX=P
12Guest: PREFIX=G 12Guest: PREFIX=G
diff --git a/drivers/lguest/core.c b/drivers/lguest/core.c
index ca581ef591e8..06869a2d3b40 100644
--- a/drivers/lguest/core.c
+++ b/drivers/lguest/core.c
@@ -11,54 +11,20 @@
11#include <linux/vmalloc.h> 11#include <linux/vmalloc.h>
12#include <linux/cpu.h> 12#include <linux/cpu.h>
13#include <linux/freezer.h> 13#include <linux/freezer.h>
14#include <linux/highmem.h>
14#include <asm/paravirt.h> 15#include <asm/paravirt.h>
15#include <asm/desc.h>
16#include <asm/pgtable.h> 16#include <asm/pgtable.h>
17#include <asm/uaccess.h> 17#include <asm/uaccess.h>
18#include <asm/poll.h> 18#include <asm/poll.h>
19#include <asm/highmem.h>
20#include <asm/asm-offsets.h> 19#include <asm/asm-offsets.h>
21#include <asm/i387.h>
22#include "lg.h" 20#include "lg.h"
23 21
24/* Found in switcher.S */
25extern char start_switcher_text[], end_switcher_text[], switch_to_guest[];
26extern unsigned long default_idt_entries[];
27
28/* Every guest maps the core switcher code. */
29#define SHARED_SWITCHER_PAGES \
30 DIV_ROUND_UP(end_switcher_text - start_switcher_text, PAGE_SIZE)
31/* Pages for switcher itself, then two pages per cpu */
32#define TOTAL_SWITCHER_PAGES (SHARED_SWITCHER_PAGES + 2 * NR_CPUS)
33
34/* We map at -4M for ease of mapping into the guest (one PTE page). */
35#define SWITCHER_ADDR 0xFFC00000
36 22
37static struct vm_struct *switcher_vma; 23static struct vm_struct *switcher_vma;
38static struct page **switcher_page; 24static struct page **switcher_page;
39 25
40static int cpu_had_pge;
41static struct {
42 unsigned long offset;
43 unsigned short segment;
44} lguest_entry;
45
46/* This One Big lock protects all inter-guest data structures. */ 26/* This One Big lock protects all inter-guest data structures. */
47DEFINE_MUTEX(lguest_lock); 27DEFINE_MUTEX(lguest_lock);
48static DEFINE_PER_CPU(struct lguest *, last_guest);
49
50/* Offset from where switcher.S was compiled to where we've copied it */
51static unsigned long switcher_offset(void)
52{
53 return SWITCHER_ADDR - (unsigned long)start_switcher_text;
54}
55
56/* This cpu's struct lguest_pages. */
57static struct lguest_pages *lguest_pages(unsigned int cpu)
58{
59 return &(((struct lguest_pages *)
60 (SWITCHER_ADDR + SHARED_SWITCHER_PAGES*PAGE_SIZE))[cpu]);
61}
62 28
63/*H:010 We need to set up the Switcher at a high virtual address. Remember the 29/*H:010 We need to set up the Switcher at a high virtual address. Remember the
64 * Switcher is a few hundred bytes of assembler code which actually changes the 30 * Switcher is a few hundred bytes of assembler code which actually changes the
@@ -69,9 +35,7 @@ static struct lguest_pages *lguest_pages(unsigned int cpu)
69 * Host since it will be running as the switchover occurs. 35 * Host since it will be running as the switchover occurs.
70 * 36 *
71 * Trying to map memory at a particular address is an unusual thing to do, so 37 * Trying to map memory at a particular address is an unusual thing to do, so
72 * it's not a simple one-liner. We also set up the per-cpu parts of the 38 * it's not a simple one-liner. */
73 * Switcher here.
74 */
75static __init int map_switcher(void) 39static __init int map_switcher(void)
76{ 40{
77 int i, err; 41 int i, err;
@@ -128,90 +92,11 @@ static __init int map_switcher(void)
128 goto free_vma; 92 goto free_vma;
129 } 93 }
130 94
131 /* Now the switcher is mapped at the right address, we can't fail! 95 /* Now the Switcher is mapped at the right address, we can't fail!
132 * Copy in the compiled-in Switcher code (from switcher.S). */ 96 * Copy in the compiled-in Switcher code (from <arch>_switcher.S). */
133 memcpy(switcher_vma->addr, start_switcher_text, 97 memcpy(switcher_vma->addr, start_switcher_text,
134 end_switcher_text - start_switcher_text); 98 end_switcher_text - start_switcher_text);
135 99
136 /* Most of the switcher.S doesn't care that it's been moved; on Intel,
137 * jumps are relative, and it doesn't access any references to external
138 * code or data.
139 *
140 * The only exception is the interrupt handlers in switcher.S: their
141 * addresses are placed in a table (default_idt_entries), so we need to
142 * update the table with the new addresses. switcher_offset() is a
143 * convenience function which returns the distance between the builtin
144 * switcher code and the high-mapped copy we just made. */
145 for (i = 0; i < IDT_ENTRIES; i++)
146 default_idt_entries[i] += switcher_offset();
147
148 /*
149 * Set up the Switcher's per-cpu areas.
150 *
151 * Each CPU gets two pages of its own within the high-mapped region
152 * (aka. "struct lguest_pages"). Much of this can be initialized now,
153 * but some depends on what Guest we are running (which is set up in
154 * copy_in_guest_info()).
155 */
156 for_each_possible_cpu(i) {
157 /* lguest_pages() returns this CPU's two pages. */
158 struct lguest_pages *pages = lguest_pages(i);
159 /* This is a convenience pointer to make the code fit one
160 * statement to a line. */
161 struct lguest_ro_state *state = &pages->state;
162
163 /* The Global Descriptor Table: the Host has a different one
164 * for each CPU. We keep a descriptor for the GDT which says
165 * where it is and how big it is (the size is actually the last
166 * byte, not the size, hence the "-1"). */
167 state->host_gdt_desc.size = GDT_SIZE-1;
168 state->host_gdt_desc.address = (long)get_cpu_gdt_table(i);
169
170 /* All CPUs on the Host use the same Interrupt Descriptor
171 * Table, so we just use store_idt(), which gets this CPU's IDT
172 * descriptor. */
173 store_idt(&state->host_idt_desc);
174
175 /* The descriptors for the Guest's GDT and IDT can be filled
176 * out now, too. We copy the GDT & IDT into ->guest_gdt and
177 * ->guest_idt before actually running the Guest. */
178 state->guest_idt_desc.size = sizeof(state->guest_idt)-1;
179 state->guest_idt_desc.address = (long)&state->guest_idt;
180 state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1;
181 state->guest_gdt_desc.address = (long)&state->guest_gdt;
182
183 /* We know where we want the stack to be when the Guest enters
184 * the switcher: in pages->regs. The stack grows upwards, so
185 * we start it at the end of that structure. */
186 state->guest_tss.esp0 = (long)(&pages->regs + 1);
187 /* And this is the GDT entry to use for the stack: we keep a
188 * couple of special LGUEST entries. */
189 state->guest_tss.ss0 = LGUEST_DS;
190
191 /* x86 can have a finegrained bitmap which indicates what I/O
192 * ports the process can use. We set it to the end of our
193 * structure, meaning "none". */
194 state->guest_tss.io_bitmap_base = sizeof(state->guest_tss);
195
196 /* Some GDT entries are the same across all Guests, so we can
197 * set them up now. */
198 setup_default_gdt_entries(state);
199 /* Most IDT entries are the same for all Guests, too.*/
200 setup_default_idt_entries(state, default_idt_entries);
201
202 /* The Host needs to be able to use the LGUEST segments on this
203 * CPU, too, so put them in the Host GDT. */
204 get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
205 get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
206 }
207
208 /* In the Switcher, we want the %cs segment register to use the
209 * LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so
210 * it will be undisturbed when we switch. To change %cs and jump we
211 * need this structure to feed to Intel's "lcall" instruction. */
212 lguest_entry.offset = (long)switch_to_guest + switcher_offset();
213 lguest_entry.segment = LGUEST_CS;
214
215 printk(KERN_INFO "lguest: mapped switcher at %p\n", 100 printk(KERN_INFO "lguest: mapped switcher at %p\n",
216 switcher_vma->addr); 101 switcher_vma->addr);
217 /* And we succeeded... */ 102 /* And we succeeded... */
@@ -243,80 +128,6 @@ static void unmap_switcher(void)
243 __free_pages(switcher_page[i], 0); 128 __free_pages(switcher_page[i], 0);
244} 129}
245 130
246/*H:130 Our Guest is usually so well behaved; it never tries to do things it
247 * isn't allowed to. Unfortunately, Linux's paravirtual infrastructure isn't
248 * quite complete, because it doesn't contain replacements for the Intel I/O
249 * instructions. As a result, the Guest sometimes fumbles across one during
250 * the boot process as it probes for various things which are usually attached
251 * to a PC.
252 *
253 * When the Guest uses one of these instructions, we get trap #13 (General
254 * Protection Fault) and come here. We see if it's one of those troublesome
255 * instructions and skip over it. We return true if we did. */
256static int emulate_insn(struct lguest *lg)
257{
258 u8 insn;
259 unsigned int insnlen = 0, in = 0, shift = 0;
260 /* The eip contains the *virtual* address of the Guest's instruction:
261 * guest_pa just subtracts the Guest's page_offset. */
262 unsigned long physaddr = guest_pa(lg, lg->regs->eip);
263
264 /* The guest_pa() function only works for Guest kernel addresses, but
265 * that's all we're trying to do anyway. */
266 if (lg->regs->eip < lg->page_offset)
267 return 0;
268
269 /* Decoding x86 instructions is icky. */
270 lgread(lg, &insn, physaddr, 1);
271
272 /* 0x66 is an "operand prefix". It means it's using the upper 16 bits
273 of the eax register. */
274 if (insn == 0x66) {
275 shift = 16;
276 /* The instruction is 1 byte so far, read the next byte. */
277 insnlen = 1;
278 lgread(lg, &insn, physaddr + insnlen, 1);
279 }
280
281 /* We can ignore the lower bit for the moment and decode the 4 opcodes
282 * we need to emulate. */
283 switch (insn & 0xFE) {
284 case 0xE4: /* in <next byte>,%al */
285 insnlen += 2;
286 in = 1;
287 break;
288 case 0xEC: /* in (%dx),%al */
289 insnlen += 1;
290 in = 1;
291 break;
292 case 0xE6: /* out %al,<next byte> */
293 insnlen += 2;
294 break;
295 case 0xEE: /* out %al,(%dx) */
296 insnlen += 1;
297 break;
298 default:
299 /* OK, we don't know what this is, can't emulate. */
300 return 0;
301 }
302
303 /* If it was an "IN" instruction, they expect the result to be read
304 * into %eax, so we change %eax. We always return all-ones, which
305 * traditionally means "there's nothing there". */
306 if (in) {
307 /* Lower bit tells is whether it's a 16 or 32 bit access */
308 if (insn & 0x1)
309 lg->regs->eax = 0xFFFFFFFF;
310 else
311 lg->regs->eax |= (0xFFFF << shift);
312 }
313 /* Finally, we've "done" the instruction, so move past it. */
314 lg->regs->eip += insnlen;
315 /* Success! */
316 return 1;
317}
318/*:*/
319
320/*L:305 131/*L:305
321 * Dealing With Guest Memory. 132 * Dealing With Guest Memory.
322 * 133 *
@@ -380,104 +191,6 @@ void lgwrite(struct lguest *lg, unsigned long addr, const void *b,
380} 191}
381/* (end of memory access helper routines) :*/ 192/* (end of memory access helper routines) :*/
382 193
383static void set_ts(void)
384{
385 u32 cr0;
386
387 cr0 = read_cr0();
388 if (!(cr0 & 8))
389 write_cr0(cr0|8);
390}
391
392/*S:010
393 * We are getting close to the Switcher.
394 *
395 * Remember that each CPU has two pages which are visible to the Guest when it
396 * runs on that CPU. This has to contain the state for that Guest: we copy the
397 * state in just before we run the Guest.
398 *
399 * Each Guest has "changed" flags which indicate what has changed in the Guest
400 * since it last ran. We saw this set in interrupts_and_traps.c and
401 * segments.c.
402 */
403static void copy_in_guest_info(struct lguest *lg, struct lguest_pages *pages)
404{
405 /* Copying all this data can be quite expensive. We usually run the
406 * same Guest we ran last time (and that Guest hasn't run anywhere else
407 * meanwhile). If that's not the case, we pretend everything in the
408 * Guest has changed. */
409 if (__get_cpu_var(last_guest) != lg || lg->last_pages != pages) {
410 __get_cpu_var(last_guest) = lg;
411 lg->last_pages = pages;
412 lg->changed = CHANGED_ALL;
413 }
414
415 /* These copies are pretty cheap, so we do them unconditionally: */
416 /* Save the current Host top-level page directory. */
417 pages->state.host_cr3 = __pa(current->mm->pgd);
418 /* Set up the Guest's page tables to see this CPU's pages (and no
419 * other CPU's pages). */
420 map_switcher_in_guest(lg, pages);
421 /* Set up the two "TSS" members which tell the CPU what stack to use
422 * for traps which do directly into the Guest (ie. traps at privilege
423 * level 1). */
424 pages->state.guest_tss.esp1 = lg->esp1;
425 pages->state.guest_tss.ss1 = lg->ss1;
426
427 /* Copy direct-to-Guest trap entries. */
428 if (lg->changed & CHANGED_IDT)
429 copy_traps(lg, pages->state.guest_idt, default_idt_entries);
430
431 /* Copy all GDT entries which the Guest can change. */
432 if (lg->changed & CHANGED_GDT)
433 copy_gdt(lg, pages->state.guest_gdt);
434 /* If only the TLS entries have changed, copy them. */
435 else if (lg->changed & CHANGED_GDT_TLS)
436 copy_gdt_tls(lg, pages->state.guest_gdt);
437
438 /* Mark the Guest as unchanged for next time. */
439 lg->changed = 0;
440}
441
442/* Finally: the code to actually call into the Switcher to run the Guest. */
443static void run_guest_once(struct lguest *lg, struct lguest_pages *pages)
444{
445 /* This is a dummy value we need for GCC's sake. */
446 unsigned int clobber;
447
448 /* Copy the guest-specific information into this CPU's "struct
449 * lguest_pages". */
450 copy_in_guest_info(lg, pages);
451
452 /* Set the trap number to 256 (impossible value). If we fault while
453 * switching to the Guest (bad segment registers or bug), this will
454 * cause us to abort the Guest. */
455 lg->regs->trapnum = 256;
456
457 /* Now: we push the "eflags" register on the stack, then do an "lcall".
458 * This is how we change from using the kernel code segment to using
459 * the dedicated lguest code segment, as well as jumping into the
460 * Switcher.
461 *
462 * The lcall also pushes the old code segment (KERNEL_CS) onto the
463 * stack, then the address of this call. This stack layout happens to
464 * exactly match the stack of an interrupt... */
465 asm volatile("pushf; lcall *lguest_entry"
466 /* This is how we tell GCC that %eax ("a") and %ebx ("b")
467 * are changed by this routine. The "=" means output. */
468 : "=a"(clobber), "=b"(clobber)
469 /* %eax contains the pages pointer. ("0" refers to the
470 * 0-th argument above, ie "a"). %ebx contains the
471 * physical address of the Guest's top-level page
472 * directory. */
473 : "0"(pages), "1"(__pa(lg->pgdirs[lg->pgdidx].pgdir))
474 /* We tell gcc that all these registers could change,
475 * which means we don't have to save and restore them in
476 * the Switcher. */
477 : "memory", "%edx", "%ecx", "%edi", "%esi");
478}
479/*:*/
480
481/*H:030 Let's jump straight to the the main loop which runs the Guest. 194/*H:030 Let's jump straight to the the main loop which runs the Guest.
482 * Remember, this is called by the Launcher reading /dev/lguest, and we keep 195 * Remember, this is called by the Launcher reading /dev/lguest, and we keep
483 * going around and around until something interesting happens. */ 196 * going around and around until something interesting happens. */
@@ -485,11 +198,6 @@ int run_guest(struct lguest *lg, unsigned long __user *user)
485{ 198{
486 /* We stop running once the Guest is dead. */ 199 /* We stop running once the Guest is dead. */
487 while (!lg->dead) { 200 while (!lg->dead) {
488 /* We need to initialize this, otherwise gcc complains. It's
489 * not (yet) clever enough to see that it's initialized when we
490 * need it. */
491 unsigned int cr2 = 0; /* Damn gcc */
492
493 /* First we run any hypercalls the Guest wants done: either in 201 /* First we run any hypercalls the Guest wants done: either in
494 * the hypercall ring in "struct lguest_data", or directly by 202 * the hypercall ring in "struct lguest_data", or directly by
495 * using int 31 (LGUEST_TRAP_ENTRY). */ 203 * using int 31 (LGUEST_TRAP_ENTRY). */
@@ -538,132 +246,20 @@ int run_guest(struct lguest *lg, unsigned long __user *user)
538 * the "Do Not Disturb" sign: */ 246 * the "Do Not Disturb" sign: */
539 local_irq_disable(); 247 local_irq_disable();
540 248
541 /* Remember the awfully-named TS bit? If the Guest has asked 249 /* Actually run the Guest until something happens. */
542 * to set it we set it now, so we can trap and pass that trap 250 lguest_arch_run_guest(lg);
543 * to the Guest if it uses the FPU. */
544 if (lg->ts)
545 set_ts();
546
547 /* SYSENTER is an optimized way of doing system calls. We
548 * can't allow it because it always jumps to privilege level 0.
549 * A normal Guest won't try it because we don't advertise it in
550 * CPUID, but a malicious Guest (or malicious Guest userspace
551 * program) could, so we tell the CPU to disable it before
552 * running the Guest. */
553 if (boot_cpu_has(X86_FEATURE_SEP))
554 wrmsr(MSR_IA32_SYSENTER_CS, 0, 0);
555
556 /* Now we actually run the Guest. It will pop back out when
557 * something interesting happens, and we can examine its
558 * registers to see what it was doing. */
559 run_guest_once(lg, lguest_pages(raw_smp_processor_id()));
560
561 /* The "regs" pointer contains two extra entries which are not
562 * really registers: a trap number which says what interrupt or
563 * trap made the switcher code come back, and an error code
564 * which some traps set. */
565
566 /* If the Guest page faulted, then the cr2 register will tell
567 * us the bad virtual address. We have to grab this now,
568 * because once we re-enable interrupts an interrupt could
569 * fault and thus overwrite cr2, or we could even move off to a
570 * different CPU. */
571 if (lg->regs->trapnum == 14)
572 cr2 = read_cr2();
573 /* Similarly, if we took a trap because the Guest used the FPU,
574 * we have to restore the FPU it expects to see. */
575 else if (lg->regs->trapnum == 7)
576 math_state_restore();
577
578 /* Restore SYSENTER if it's supposed to be on. */
579 if (boot_cpu_has(X86_FEATURE_SEP))
580 wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
581 251
582 /* Now we're ready to be interrupted or moved to other CPUs */ 252 /* Now we're ready to be interrupted or moved to other CPUs */
583 local_irq_enable(); 253 local_irq_enable();
584 254
585 /* OK, so what happened? */ 255 /* Now we deal with whatever happened to the Guest. */
586 switch (lg->regs->trapnum) { 256 lguest_arch_handle_trap(lg);
587 case 13: /* We've intercepted a GPF. */
588 /* Check if this was one of those annoying IN or OUT
589 * instructions which we need to emulate. If so, we
590 * just go back into the Guest after we've done it. */
591 if (lg->regs->errcode == 0) {
592 if (emulate_insn(lg))
593 continue;
594 }
595 break;
596 case 14: /* We've intercepted a page fault. */
597 /* The Guest accessed a virtual address that wasn't
598 * mapped. This happens a lot: we don't actually set
599 * up most of the page tables for the Guest at all when
600 * we start: as it runs it asks for more and more, and
601 * we set them up as required. In this case, we don't
602 * even tell the Guest that the fault happened.
603 *
604 * The errcode tells whether this was a read or a
605 * write, and whether kernel or userspace code. */
606 if (demand_page(lg, cr2, lg->regs->errcode))
607 continue;
608
609 /* OK, it's really not there (or not OK): the Guest
610 * needs to know. We write out the cr2 value so it
611 * knows where the fault occurred.
612 *
613 * Note that if the Guest were really messed up, this
614 * could happen before it's done the INITIALIZE
615 * hypercall, so lg->lguest_data will be NULL */
616 if (lg->lguest_data
617 && put_user(cr2, &lg->lguest_data->cr2))
618 kill_guest(lg, "Writing cr2");
619 break;
620 case 7: /* We've intercepted a Device Not Available fault. */
621 /* If the Guest doesn't want to know, we already
622 * restored the Floating Point Unit, so we just
623 * continue without telling it. */
624 if (!lg->ts)
625 continue;
626 break;
627 case 32 ... 255:
628 /* These values mean a real interrupt occurred, in
629 * which case the Host handler has already been run.
630 * We just do a friendly check if another process
631 * should now be run, then fall through to loop
632 * around: */
633 cond_resched();
634 case LGUEST_TRAP_ENTRY: /* Handled at top of loop */
635 continue;
636 }
637
638 /* If we get here, it's a trap the Guest wants to know
639 * about. */
640 if (deliver_trap(lg, lg->regs->trapnum))
641 continue;
642
643 /* If the Guest doesn't have a handler (either it hasn't
644 * registered any yet, or it's one of the faults we don't let
645 * it handle), it dies with a cryptic error message. */
646 kill_guest(lg, "unhandled trap %li at %#lx (%#lx)",
647 lg->regs->trapnum, lg->regs->eip,
648 lg->regs->trapnum == 14 ? cr2 : lg->regs->errcode);
649 } 257 }
258
650 /* The Guest is dead => "No such file or directory" */ 259 /* The Guest is dead => "No such file or directory" */
651 return -ENOENT; 260 return -ENOENT;
652} 261}
653 262
654/* Now we can look at each of the routines this calls, in increasing order of
655 * complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(),
656 * deliver_trap() and demand_page(). After all those, we'll be ready to
657 * examine the Switcher, and our philosophical understanding of the Host/Guest
658 * duality will be complete. :*/
659static void adjust_pge(void *on)
660{
661 if (on)
662 write_cr4(read_cr4() | X86_CR4_PGE);
663 else
664 write_cr4(read_cr4() & ~X86_CR4_PGE);
665}
666
667/*H:000 263/*H:000
668 * Welcome to the Host! 264 * Welcome to the Host!
669 * 265 *
@@ -705,31 +301,8 @@ static int __init init(void)
705 return err; 301 return err;
706 } 302 }
707 303
708 /* Finally, we need to turn off "Page Global Enable". PGE is an 304 /* Finally we do some architecture-specific setup. */
709 * optimization where page table entries are specially marked to show 305 lguest_arch_host_init();
710 * they never change. The Host kernel marks all the kernel pages this
711 * way because it's always present, even when userspace is running.
712 *
713 * Lguest breaks this: unbeknownst to the rest of the Host kernel, we
714 * switch to the Guest kernel. If you don't disable this on all CPUs,
715 * you'll get really weird bugs that you'll chase for two days.
716 *
717 * I used to turn PGE off every time we switched to the Guest and back
718 * on when we return, but that slowed the Switcher down noticibly. */
719
720 /* We don't need the complexity of CPUs coming and going while we're
721 * doing this. */
722 lock_cpu_hotplug();
723 if (cpu_has_pge) { /* We have a broader idea of "global". */
724 /* Remember that this was originally set (for cleanup). */
725 cpu_had_pge = 1;
726 /* adjust_pge is a helper function which sets or unsets the PGE
727 * bit on its CPU, depending on the argument (0 == unset). */
728 on_each_cpu(adjust_pge, (void *)0, 0, 1);
729 /* Turn off the feature in the global feature set. */
730 clear_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
731 }
732 unlock_cpu_hotplug();
733 306
734 /* All good! */ 307 /* All good! */
735 return 0; 308 return 0;
@@ -742,15 +315,9 @@ static void __exit fini(void)
742 free_pagetables(); 315 free_pagetables();
743 unmap_switcher(); 316 unmap_switcher();
744 317
745 /* If we had PGE before we started, turn it back on now. */ 318 lguest_arch_host_fini();
746 lock_cpu_hotplug();
747 if (cpu_had_pge) {
748 set_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
749 /* adjust_pge's argument "1" means set PGE. */
750 on_each_cpu(adjust_pge, (void *)1, 0, 1);
751 }
752 unlock_cpu_hotplug();
753} 319}
320/*:*/
754 321
755/* The Host side of lguest can be a module. This is a nice way for people to 322/* The Host side of lguest can be a module. This is a nice way for people to
756 * play with it. */ 323 * play with it. */
diff --git a/drivers/lguest/interrupts_and_traps.c b/drivers/lguest/interrupts_and_traps.c
index 0dfb0903aa69..fdefc0afc38c 100644
--- a/drivers/lguest/interrupts_and_traps.c
+++ b/drivers/lguest/interrupts_and_traps.c
@@ -165,7 +165,7 @@ void maybe_do_interrupt(struct lguest *lg)
165 /* Look at the IDT entry the Guest gave us for this interrupt. The 165 /* Look at the IDT entry the Guest gave us for this interrupt. The
166 * first 32 (FIRST_EXTERNAL_VECTOR) entries are for traps, so we skip 166 * first 32 (FIRST_EXTERNAL_VECTOR) entries are for traps, so we skip
167 * over them. */ 167 * over them. */
168 idt = &lg->idt[FIRST_EXTERNAL_VECTOR+irq]; 168 idt = &lg->arch.idt[FIRST_EXTERNAL_VECTOR+irq];
169 /* If they don't have a handler (yet?), we just ignore it */ 169 /* If they don't have a handler (yet?), we just ignore it */
170 if (idt_present(idt->a, idt->b)) { 170 if (idt_present(idt->a, idt->b)) {
171 /* OK, mark it no longer pending and deliver it. */ 171 /* OK, mark it no longer pending and deliver it. */
@@ -197,14 +197,14 @@ int deliver_trap(struct lguest *lg, unsigned int num)
197{ 197{
198 /* Trap numbers are always 8 bit, but we set an impossible trap number 198 /* Trap numbers are always 8 bit, but we set an impossible trap number
199 * for traps inside the Switcher, so check that here. */ 199 * for traps inside the Switcher, so check that here. */
200 if (num >= ARRAY_SIZE(lg->idt)) 200 if (num >= ARRAY_SIZE(lg->arch.idt))
201 return 0; 201 return 0;
202 202
203 /* Early on the Guest hasn't set the IDT entries (or maybe it put a 203 /* Early on the Guest hasn't set the IDT entries (or maybe it put a
204 * bogus one in): if we fail here, the Guest will be killed. */ 204 * bogus one in): if we fail here, the Guest will be killed. */
205 if (!idt_present(lg->idt[num].a, lg->idt[num].b)) 205 if (!idt_present(lg->arch.idt[num].a, lg->arch.idt[num].b))
206 return 0; 206 return 0;
207 set_guest_interrupt(lg, lg->idt[num].a, lg->idt[num].b, has_err(num)); 207 set_guest_interrupt(lg, lg->arch.idt[num].a, lg->arch.idt[num].b, has_err(num));
208 return 1; 208 return 1;
209} 209}
210 210
@@ -341,10 +341,10 @@ void load_guest_idt_entry(struct lguest *lg, unsigned int num, u32 lo, u32 hi)
341 lg->changed |= CHANGED_IDT; 341 lg->changed |= CHANGED_IDT;
342 342
343 /* Check that the Guest doesn't try to step outside the bounds. */ 343 /* Check that the Guest doesn't try to step outside the bounds. */
344 if (num >= ARRAY_SIZE(lg->idt)) 344 if (num >= ARRAY_SIZE(lg->arch.idt))
345 kill_guest(lg, "Setting idt entry %u", num); 345 kill_guest(lg, "Setting idt entry %u", num);
346 else 346 else
347 set_trap(lg, &lg->idt[num], num, lo, hi); 347 set_trap(lg, &lg->arch.idt[num], num, lo, hi);
348} 348}
349 349
350/* The default entry for each interrupt points into the Switcher routines which 350/* The default entry for each interrupt points into the Switcher routines which
@@ -387,7 +387,7 @@ void copy_traps(const struct lguest *lg, struct desc_struct *idt,
387 387
388 /* We can simply copy the direct traps, otherwise we use the default 388 /* We can simply copy the direct traps, otherwise we use the default
389 * ones in the Switcher: they will return to the Host. */ 389 * ones in the Switcher: they will return to the Host. */
390 for (i = 0; i < ARRAY_SIZE(lg->idt); i++) { 390 for (i = 0; i < ARRAY_SIZE(lg->arch.idt); i++) {
391 /* If no Guest can ever override this trap, leave it alone. */ 391 /* If no Guest can ever override this trap, leave it alone. */
392 if (!direct_trap(i)) 392 if (!direct_trap(i))
393 continue; 393 continue;
@@ -396,8 +396,8 @@ void copy_traps(const struct lguest *lg, struct desc_struct *idt,
396 * Interrupt gates (type 14) disable interrupts as they are 396 * Interrupt gates (type 14) disable interrupts as they are
397 * entered, which we never let the Guest do. Not present 397 * entered, which we never let the Guest do. Not present
398 * entries (type 0x0) also can't go direct, of course. */ 398 * entries (type 0x0) also can't go direct, of course. */
399 if (idt_type(lg->idt[i].a, lg->idt[i].b) == 0xF) 399 if (idt_type(lg->arch.idt[i].a, lg->arch.idt[i].b) == 0xF)
400 idt[i] = lg->idt[i]; 400 idt[i] = lg->arch.idt[i];
401 else 401 else
402 /* Reset it to the default. */ 402 /* Reset it to the default. */
403 default_idt_entry(&idt[i], i, def[i]); 403 default_idt_entry(&idt[i], i, def[i]);
diff --git a/drivers/lguest/lg.h b/drivers/lguest/lg.h
index c1ca127ddece..203d3100c3b4 100644
--- a/drivers/lguest/lg.h
+++ b/drivers/lguest/lg.h
@@ -1,13 +1,6 @@
1#ifndef _LGUEST_H 1#ifndef _LGUEST_H
2#define _LGUEST_H 2#define _LGUEST_H
3 3
4#include <asm/desc.h>
5
6#define GDT_ENTRY_LGUEST_CS 10
7#define GDT_ENTRY_LGUEST_DS 11
8#define LGUEST_CS (GDT_ENTRY_LGUEST_CS * 8)
9#define LGUEST_DS (GDT_ENTRY_LGUEST_DS * 8)
10
11#ifndef __ASSEMBLY__ 4#ifndef __ASSEMBLY__
12#include <linux/types.h> 5#include <linux/types.h>
13#include <linux/init.h> 6#include <linux/init.h>
@@ -18,34 +11,12 @@
18#include <linux/wait.h> 11#include <linux/wait.h>
19#include <linux/err.h> 12#include <linux/err.h>
20#include <asm/semaphore.h> 13#include <asm/semaphore.h>
21#include "irq_vectors.h"
22
23#define GUEST_PL 1
24 14
25struct lguest_regs 15#include <asm/lguest.h>
26{
27 /* Manually saved part. */
28 unsigned long ebx, ecx, edx;
29 unsigned long esi, edi, ebp;
30 unsigned long gs;
31 unsigned long eax;
32 unsigned long fs, ds, es;
33 unsigned long trapnum, errcode;
34 /* Trap pushed part */
35 unsigned long eip;
36 unsigned long cs;
37 unsigned long eflags;
38 unsigned long esp;
39 unsigned long ss;
40};
41 16
42void free_pagetables(void); 17void free_pagetables(void);
43int init_pagetables(struct page **switcher_page, unsigned int pages); 18int init_pagetables(struct page **switcher_page, unsigned int pages);
44 19
45/* Full 4G segment descriptors, suitable for CS and DS. */
46#define FULL_EXEC_SEGMENT ((struct desc_struct){0x0000ffff, 0x00cf9b00})
47#define FULL_SEGMENT ((struct desc_struct){0x0000ffff, 0x00cf9300})
48
49struct lguest_dma_info 20struct lguest_dma_info
50{ 21{
51 struct list_head list; 22 struct list_head list;
@@ -98,23 +69,6 @@ struct pgdir
98 spgd_t *pgdir; 69 spgd_t *pgdir;
99}; 70};
100 71
101/* This is a guest-specific page (mapped ro) into the guest. */
102struct lguest_ro_state
103{
104 /* Host information we need to restore when we switch back. */
105 u32 host_cr3;
106 struct Xgt_desc_struct host_idt_desc;
107 struct Xgt_desc_struct host_gdt_desc;
108 u32 host_sp;
109
110 /* Fields which are used when guest is running. */
111 struct Xgt_desc_struct guest_idt_desc;
112 struct Xgt_desc_struct guest_gdt_desc;
113 struct i386_hw_tss guest_tss;
114 struct desc_struct guest_idt[IDT_ENTRIES];
115 struct desc_struct guest_gdt[GDT_ENTRIES];
116};
117
118/* We have two pages shared with guests, per cpu. */ 72/* We have two pages shared with guests, per cpu. */
119struct lguest_pages 73struct lguest_pages
120{ 74{
@@ -180,11 +134,7 @@ struct lguest
180 /* Dead? */ 134 /* Dead? */
181 const char *dead; 135 const char *dead;
182 136
183 /* The GDT entries copied into lguest_ro_state when running. */ 137 struct lguest_arch arch;
184 struct desc_struct gdt[GDT_ENTRIES];
185
186 /* The IDT entries: some copied into lguest_ro_state when running. */
187 struct desc_struct idt[IDT_ENTRIES];
188 138
189 /* Virtual clock device */ 139 /* Virtual clock device */
190 struct hrtimer hrt; 140 struct hrtimer hrt;
@@ -239,6 +189,15 @@ void map_switcher_in_guest(struct lguest *lg, struct lguest_pages *pages);
239int demand_page(struct lguest *info, unsigned long cr2, int errcode); 189int demand_page(struct lguest *info, unsigned long cr2, int errcode);
240void pin_page(struct lguest *lg, unsigned long vaddr); 190void pin_page(struct lguest *lg, unsigned long vaddr);
241 191
192/* <arch>/core.c: */
193void lguest_arch_host_init(void);
194void lguest_arch_host_fini(void);
195void lguest_arch_run_guest(struct lguest *lg);
196void lguest_arch_handle_trap(struct lguest *lg);
197
198/* <arch>/switcher.S: */
199extern char start_switcher_text[], end_switcher_text[], switch_to_guest[];
200
242/* lguest_user.c: */ 201/* lguest_user.c: */
243int lguest_device_init(void); 202int lguest_device_init(void);
244void lguest_device_remove(void); 203void lguest_device_remove(void);
diff --git a/drivers/lguest/segments.c b/drivers/lguest/segments.c
index 9b81119f46e9..95eb9cf297bf 100644
--- a/drivers/lguest/segments.c
+++ b/drivers/lguest/segments.c
@@ -73,14 +73,14 @@ static void fixup_gdt_table(struct lguest *lg, unsigned start, unsigned end)
73 /* Segment descriptors contain a privilege level: the Guest is 73 /* Segment descriptors contain a privilege level: the Guest is
74 * sometimes careless and leaves this as 0, even though it's 74 * sometimes careless and leaves this as 0, even though it's
75 * running at privilege level 1. If so, we fix it here. */ 75 * running at privilege level 1. If so, we fix it here. */
76 if ((lg->gdt[i].b & 0x00006000) == 0) 76 if ((lg->arch.gdt[i].b & 0x00006000) == 0)
77 lg->gdt[i].b |= (GUEST_PL << 13); 77 lg->arch.gdt[i].b |= (GUEST_PL << 13);
78 78
79 /* Each descriptor has an "accessed" bit. If we don't set it 79 /* Each descriptor has an "accessed" bit. If we don't set it
80 * now, the CPU will try to set it when the Guest first loads 80 * now, the CPU will try to set it when the Guest first loads
81 * that entry into a segment register. But the GDT isn't 81 * that entry into a segment register. But the GDT isn't
82 * writable by the Guest, so bad things can happen. */ 82 * writable by the Guest, so bad things can happen. */
83 lg->gdt[i].b |= 0x00000100; 83 lg->arch.gdt[i].b |= 0x00000100;
84 } 84 }
85} 85}
86 86
@@ -106,12 +106,12 @@ void setup_default_gdt_entries(struct lguest_ro_state *state)
106void setup_guest_gdt(struct lguest *lg) 106void setup_guest_gdt(struct lguest *lg)
107{ 107{
108 /* Start with full 0-4G segments... */ 108 /* Start with full 0-4G segments... */
109 lg->gdt[GDT_ENTRY_KERNEL_CS] = FULL_EXEC_SEGMENT; 109 lg->arch.gdt[GDT_ENTRY_KERNEL_CS] = FULL_EXEC_SEGMENT;
110 lg->gdt[GDT_ENTRY_KERNEL_DS] = FULL_SEGMENT; 110 lg->arch.gdt[GDT_ENTRY_KERNEL_DS] = FULL_SEGMENT;
111 /* ...except the Guest is allowed to use them, so set the privilege 111 /* ...except the Guest is allowed to use them, so set the privilege
112 * level appropriately in the flags. */ 112 * level appropriately in the flags. */
113 lg->gdt[GDT_ENTRY_KERNEL_CS].b |= (GUEST_PL << 13); 113 lg->arch.gdt[GDT_ENTRY_KERNEL_CS].b |= (GUEST_PL << 13);
114 lg->gdt[GDT_ENTRY_KERNEL_DS].b |= (GUEST_PL << 13); 114 lg->arch.gdt[GDT_ENTRY_KERNEL_DS].b |= (GUEST_PL << 13);
115} 115}
116 116
117/* Like the IDT, we never simply use the GDT the Guest gives us. We set up the 117/* Like the IDT, we never simply use the GDT the Guest gives us. We set up the
@@ -126,7 +126,7 @@ void copy_gdt_tls(const struct lguest *lg, struct desc_struct *gdt)
126 unsigned int i; 126 unsigned int i;
127 127
128 for (i = GDT_ENTRY_TLS_MIN; i <= GDT_ENTRY_TLS_MAX; i++) 128 for (i = GDT_ENTRY_TLS_MIN; i <= GDT_ENTRY_TLS_MAX; i++)
129 gdt[i] = lg->gdt[i]; 129 gdt[i] = lg->arch.gdt[i];
130} 130}
131 131
132/* This is the full version */ 132/* This is the full version */
@@ -138,7 +138,7 @@ void copy_gdt(const struct lguest *lg, struct desc_struct *gdt)
138 * replaced. See ignored_gdt() above. */ 138 * replaced. See ignored_gdt() above. */
139 for (i = 0; i < GDT_ENTRIES; i++) 139 for (i = 0; i < GDT_ENTRIES; i++)
140 if (!ignored_gdt(i)) 140 if (!ignored_gdt(i))
141 gdt[i] = lg->gdt[i]; 141 gdt[i] = lg->arch.gdt[i];
142} 142}
143 143
144/* This is where the Guest asks us to load a new GDT (LHCALL_LOAD_GDT). */ 144/* This is where the Guest asks us to load a new GDT (LHCALL_LOAD_GDT). */
@@ -146,12 +146,12 @@ void load_guest_gdt(struct lguest *lg, unsigned long table, u32 num)
146{ 146{
147 /* We assume the Guest has the same number of GDT entries as the 147 /* We assume the Guest has the same number of GDT entries as the
148 * Host, otherwise we'd have to dynamically allocate the Guest GDT. */ 148 * Host, otherwise we'd have to dynamically allocate the Guest GDT. */
149 if (num > ARRAY_SIZE(lg->gdt)) 149 if (num > ARRAY_SIZE(lg->arch.gdt))
150 kill_guest(lg, "too many gdt entries %i", num); 150 kill_guest(lg, "too many gdt entries %i", num);
151 151
152 /* We read the whole thing in, then fix it up. */ 152 /* We read the whole thing in, then fix it up. */
153 lgread(lg, lg->gdt, table, num * sizeof(lg->gdt[0])); 153 lgread(lg, lg->arch.gdt, table, num * sizeof(lg->arch.gdt[0]));
154 fixup_gdt_table(lg, 0, ARRAY_SIZE(lg->gdt)); 154 fixup_gdt_table(lg, 0, ARRAY_SIZE(lg->arch.gdt));
155 /* Mark that the GDT changed so the core knows it has to copy it again, 155 /* Mark that the GDT changed so the core knows it has to copy it again,
156 * even if the Guest is run on the same CPU. */ 156 * even if the Guest is run on the same CPU. */
157 lg->changed |= CHANGED_GDT; 157 lg->changed |= CHANGED_GDT;
@@ -159,7 +159,7 @@ void load_guest_gdt(struct lguest *lg, unsigned long table, u32 num)
159 159
160void guest_load_tls(struct lguest *lg, unsigned long gtls) 160void guest_load_tls(struct lguest *lg, unsigned long gtls)
161{ 161{
162 struct desc_struct *tls = &lg->gdt[GDT_ENTRY_TLS_MIN]; 162 struct desc_struct *tls = &lg->arch.gdt[GDT_ENTRY_TLS_MIN];
163 163
164 lgread(lg, tls, gtls, sizeof(*tls)*GDT_ENTRY_TLS_ENTRIES); 164 lgread(lg, tls, gtls, sizeof(*tls)*GDT_ENTRY_TLS_ENTRIES);
165 fixup_gdt_table(lg, GDT_ENTRY_TLS_MIN, GDT_ENTRY_TLS_MAX+1); 165 fixup_gdt_table(lg, GDT_ENTRY_TLS_MIN, GDT_ENTRY_TLS_MAX+1);
diff --git a/drivers/lguest/x86/core.c b/drivers/lguest/x86/core.c
new file mode 100644
index 000000000000..e2f46b16ce31
--- /dev/null
+++ b/drivers/lguest/x86/core.c
@@ -0,0 +1,476 @@
1/*
2 * Copyright (C) 2006, Rusty Russell <rusty@rustcorp.com.au> IBM Corporation.
3 * Copyright (C) 2007, Jes Sorensen <jes@sgi.com> SGI.
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13 * NON INFRINGEMENT. See the GNU General Public License for more
14 * details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
19 */
20#include <linux/kernel.h>
21#include <linux/start_kernel.h>
22#include <linux/string.h>
23#include <linux/console.h>
24#include <linux/screen_info.h>
25#include <linux/irq.h>
26#include <linux/interrupt.h>
27#include <linux/clocksource.h>
28#include <linux/clockchips.h>
29#include <linux/cpu.h>
30#include <linux/lguest.h>
31#include <linux/lguest_launcher.h>
32#include <linux/lguest_bus.h>
33#include <asm/paravirt.h>
34#include <asm/param.h>
35#include <asm/page.h>
36#include <asm/pgtable.h>
37#include <asm/desc.h>
38#include <asm/setup.h>
39#include <asm/lguest.h>
40#include <asm/uaccess.h>
41#include <asm/i387.h>
42#include "../lg.h"
43
44static int cpu_had_pge;
45
46static struct {
47 unsigned long offset;
48 unsigned short segment;
49} lguest_entry;
50
51/* Offset from where switcher.S was compiled to where we've copied it */
52static unsigned long switcher_offset(void)
53{
54 return SWITCHER_ADDR - (unsigned long)start_switcher_text;
55}
56
57/* This cpu's struct lguest_pages. */
58static struct lguest_pages *lguest_pages(unsigned int cpu)
59{
60 return &(((struct lguest_pages *)
61 (SWITCHER_ADDR + SHARED_SWITCHER_PAGES*PAGE_SIZE))[cpu]);
62}
63
64static DEFINE_PER_CPU(struct lguest *, last_guest);
65
66/*S:010
67 * We are getting close to the Switcher.
68 *
69 * Remember that each CPU has two pages which are visible to the Guest when it
70 * runs on that CPU. This has to contain the state for that Guest: we copy the
71 * state in just before we run the Guest.
72 *
73 * Each Guest has "changed" flags which indicate what has changed in the Guest
74 * since it last ran. We saw this set in interrupts_and_traps.c and
75 * segments.c.
76 */
77static void copy_in_guest_info(struct lguest *lg, struct lguest_pages *pages)
78{
79 /* Copying all this data can be quite expensive. We usually run the
80 * same Guest we ran last time (and that Guest hasn't run anywhere else
81 * meanwhile). If that's not the case, we pretend everything in the
82 * Guest has changed. */
83 if (__get_cpu_var(last_guest) != lg || lg->last_pages != pages) {
84 __get_cpu_var(last_guest) = lg;
85 lg->last_pages = pages;
86 lg->changed = CHANGED_ALL;
87 }
88
89 /* These copies are pretty cheap, so we do them unconditionally: */
90 /* Save the current Host top-level page directory. */
91 pages->state.host_cr3 = __pa(current->mm->pgd);
92 /* Set up the Guest's page tables to see this CPU's pages (and no
93 * other CPU's pages). */
94 map_switcher_in_guest(lg, pages);
95 /* Set up the two "TSS" members which tell the CPU what stack to use
96 * for traps which do directly into the Guest (ie. traps at privilege
97 * level 1). */
98 pages->state.guest_tss.esp1 = lg->esp1;
99 pages->state.guest_tss.ss1 = lg->ss1;
100
101 /* Copy direct-to-Guest trap entries. */
102 if (lg->changed & CHANGED_IDT)
103 copy_traps(lg, pages->state.guest_idt, default_idt_entries);
104
105 /* Copy all GDT entries which the Guest can change. */
106 if (lg->changed & CHANGED_GDT)
107 copy_gdt(lg, pages->state.guest_gdt);
108 /* If only the TLS entries have changed, copy them. */
109 else if (lg->changed & CHANGED_GDT_TLS)
110 copy_gdt_tls(lg, pages->state.guest_gdt);
111
112 /* Mark the Guest as unchanged for next time. */
113 lg->changed = 0;
114}
115
116/* Finally: the code to actually call into the Switcher to run the Guest. */
117static void run_guest_once(struct lguest *lg, struct lguest_pages *pages)
118{
119 /* This is a dummy value we need for GCC's sake. */
120 unsigned int clobber;
121
122 /* Copy the guest-specific information into this CPU's "struct
123 * lguest_pages". */
124 copy_in_guest_info(lg, pages);
125
126 /* Set the trap number to 256 (impossible value). If we fault while
127 * switching to the Guest (bad segment registers or bug), this will
128 * cause us to abort the Guest. */
129 lg->regs->trapnum = 256;
130
131 /* Now: we push the "eflags" register on the stack, then do an "lcall".
132 * This is how we change from using the kernel code segment to using
133 * the dedicated lguest code segment, as well as jumping into the
134 * Switcher.
135 *
136 * The lcall also pushes the old code segment (KERNEL_CS) onto the
137 * stack, then the address of this call. This stack layout happens to
138 * exactly match the stack of an interrupt... */
139 asm volatile("pushf; lcall *lguest_entry"
140 /* This is how we tell GCC that %eax ("a") and %ebx ("b")
141 * are changed by this routine. The "=" means output. */
142 : "=a"(clobber), "=b"(clobber)
143 /* %eax contains the pages pointer. ("0" refers to the
144 * 0-th argument above, ie "a"). %ebx contains the
145 * physical address of the Guest's top-level page
146 * directory. */
147 : "0"(pages), "1"(__pa(lg->pgdirs[lg->pgdidx].pgdir))
148 /* We tell gcc that all these registers could change,
149 * which means we don't have to save and restore them in
150 * the Switcher. */
151 : "memory", "%edx", "%ecx", "%edi", "%esi");
152}
153/*:*/
154
155/*H:040 This is the i386-specific code to setup and run the Guest. Interrupts
156 * are disabled: we own the CPU. */
157void lguest_arch_run_guest(struct lguest *lg)
158{
159 /* Remember the awfully-named TS bit? If the Guest has asked
160 * to set it we set it now, so we can trap and pass that trap
161 * to the Guest if it uses the FPU. */
162 if (lg->ts)
163 lguest_set_ts();
164
165 /* SYSENTER is an optimized way of doing system calls. We
166 * can't allow it because it always jumps to privilege level 0.
167 * A normal Guest won't try it because we don't advertise it in
168 * CPUID, but a malicious Guest (or malicious Guest userspace
169 * program) could, so we tell the CPU to disable it before
170 * running the Guest. */
171 if (boot_cpu_has(X86_FEATURE_SEP))
172 wrmsr(MSR_IA32_SYSENTER_CS, 0, 0);
173
174 /* Now we actually run the Guest. It will pop back out when
175 * something interesting happens, and we can examine its
176 * registers to see what it was doing. */
177 run_guest_once(lg, lguest_pages(raw_smp_processor_id()));
178
179 /* The "regs" pointer contains two extra entries which are not
180 * really registers: a trap number which says what interrupt or
181 * trap made the switcher code come back, and an error code
182 * which some traps set. */
183
184 /* If the Guest page faulted, then the cr2 register will tell
185 * us the bad virtual address. We have to grab this now,
186 * because once we re-enable interrupts an interrupt could
187 * fault and thus overwrite cr2, or we could even move off to a
188 * different CPU. */
189 if (lg->regs->trapnum == 14)
190 lg->arch.last_pagefault = read_cr2();
191 /* Similarly, if we took a trap because the Guest used the FPU,
192 * we have to restore the FPU it expects to see. */
193 else if (lg->regs->trapnum == 7)
194 math_state_restore();
195
196 /* Restore SYSENTER if it's supposed to be on. */
197 if (boot_cpu_has(X86_FEATURE_SEP))
198 wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
199}
200
201/*H:130 Our Guest is usually so well behaved; it never tries to do things it
202 * isn't allowed to. Unfortunately, Linux's paravirtual infrastructure isn't
203 * quite complete, because it doesn't contain replacements for the Intel I/O
204 * instructions. As a result, the Guest sometimes fumbles across one during
205 * the boot process as it probes for various things which are usually attached
206 * to a PC.
207 *
208 * When the Guest uses one of these instructions, we get trap #13 (General
209 * Protection Fault) and come here. We see if it's one of those troublesome
210 * instructions and skip over it. We return true if we did. */
211static int emulate_insn(struct lguest *lg)
212{
213 u8 insn;
214 unsigned int insnlen = 0, in = 0, shift = 0;
215 /* The eip contains the *virtual* address of the Guest's instruction:
216 * guest_pa just subtracts the Guest's page_offset. */
217 unsigned long physaddr = guest_pa(lg, lg->regs->eip);
218
219 /* The guest_pa() function only works for Guest kernel addresses, but
220 * that's all we're trying to do anyway. */
221 if (lg->regs->eip < lg->page_offset)
222 return 0;
223
224 /* Decoding x86 instructions is icky. */
225 lgread(lg, &insn, physaddr, 1);
226
227 /* 0x66 is an "operand prefix". It means it's using the upper 16 bits
228 of the eax register. */
229 if (insn == 0x66) {
230 shift = 16;
231 /* The instruction is 1 byte so far, read the next byte. */
232 insnlen = 1;
233 lgread(lg, &insn, physaddr + insnlen, 1);
234 }
235
236 /* We can ignore the lower bit for the moment and decode the 4 opcodes
237 * we need to emulate. */
238 switch (insn & 0xFE) {
239 case 0xE4: /* in <next byte>,%al */
240 insnlen += 2;
241 in = 1;
242 break;
243 case 0xEC: /* in (%dx),%al */
244 insnlen += 1;
245 in = 1;
246 break;
247 case 0xE6: /* out %al,<next byte> */
248 insnlen += 2;
249 break;
250 case 0xEE: /* out %al,(%dx) */
251 insnlen += 1;
252 break;
253 default:
254 /* OK, we don't know what this is, can't emulate. */
255 return 0;
256 }
257
258 /* If it was an "IN" instruction, they expect the result to be read
259 * into %eax, so we change %eax. We always return all-ones, which
260 * traditionally means "there's nothing there". */
261 if (in) {
262 /* Lower bit tells is whether it's a 16 or 32 bit access */
263 if (insn & 0x1)
264 lg->regs->eax = 0xFFFFFFFF;
265 else
266 lg->regs->eax |= (0xFFFF << shift);
267 }
268 /* Finally, we've "done" the instruction, so move past it. */
269 lg->regs->eip += insnlen;
270 /* Success! */
271 return 1;
272}
273
274/*H:050 Once we've re-enabled interrupts, we look at why the Guest exited. */
275void lguest_arch_handle_trap(struct lguest *lg)
276{
277 switch (lg->regs->trapnum) {
278 case 13: /* We've intercepted a GPF. */
279 /* Check if this was one of those annoying IN or OUT
280 * instructions which we need to emulate. If so, we
281 * just go back into the Guest after we've done it. */
282 if (lg->regs->errcode == 0) {
283 if (emulate_insn(lg))
284 return;
285 }
286 break;
287 case 14: /* We've intercepted a page fault. */
288 /* The Guest accessed a virtual address that wasn't
289 * mapped. This happens a lot: we don't actually set
290 * up most of the page tables for the Guest at all when
291 * we start: as it runs it asks for more and more, and
292 * we set them up as required. In this case, we don't
293 * even tell the Guest that the fault happened.
294 *
295 * The errcode tells whether this was a read or a
296 * write, and whether kernel or userspace code. */
297 if (demand_page(lg, lg->arch.last_pagefault, lg->regs->errcode))
298 return;
299
300 /* OK, it's really not there (or not OK): the Guest
301 * needs to know. We write out the cr2 value so it
302 * knows where the fault occurred.
303 *
304 * Note that if the Guest were really messed up, this
305 * could happen before it's done the INITIALIZE
306 * hypercall, so lg->lguest_data will be NULL */
307 if (lg->lguest_data &&
308 put_user(lg->arch.last_pagefault, &lg->lguest_data->cr2))
309 kill_guest(lg, "Writing cr2");
310 break;
311 case 7: /* We've intercepted a Device Not Available fault. */
312 /* If the Guest doesn't want to know, we already
313 * restored the Floating Point Unit, so we just
314 * continue without telling it. */
315 if (!lg->ts)
316 return;
317 break;
318 case 32 ... 255:
319 /* These values mean a real interrupt occurred, in
320 * which case the Host handler has already been run.
321 * We just do a friendly check if another process
322 * should now be run, then fall through to loop
323 * around: */
324 cond_resched();
325 case LGUEST_TRAP_ENTRY: /* Handled before re-entering Guest */
326 return;
327 }
328
329 /* We didn't handle the trap, so it needs to go to the Guest. */
330 if (!deliver_trap(lg, lg->regs->trapnum))
331 /* If the Guest doesn't have a handler (either it hasn't
332 * registered any yet, or it's one of the faults we don't let
333 * it handle), it dies with a cryptic error message. */
334 kill_guest(lg, "unhandled trap %li at %#lx (%#lx)",
335 lg->regs->trapnum, lg->regs->eip,
336 lg->regs->trapnum == 14 ? lg->arch.last_pagefault
337 : lg->regs->errcode);
338}
339
340/* Now we can look at each of the routines this calls, in increasing order of
341 * complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(),
342 * deliver_trap() and demand_page(). After all those, we'll be ready to
343 * examine the Switcher, and our philosophical understanding of the Host/Guest
344 * duality will be complete. :*/
345static void adjust_pge(void *on)
346{
347 if (on)
348 write_cr4(read_cr4() | X86_CR4_PGE);
349 else
350 write_cr4(read_cr4() & ~X86_CR4_PGE);
351}
352
353/*H:020 Now the Switcher is mapped and every thing else is ready, we need to do
354 * some more i386-specific initialization. */
355void __init lguest_arch_host_init(void)
356{
357 int i;
358
359 /* Most of the i386/switcher.S doesn't care that it's been moved; on
360 * Intel, jumps are relative, and it doesn't access any references to
361 * external code or data.
362 *
363 * The only exception is the interrupt handlers in switcher.S: their
364 * addresses are placed in a table (default_idt_entries), so we need to
365 * update the table with the new addresses. switcher_offset() is a
366 * convenience function which returns the distance between the builtin
367 * switcher code and the high-mapped copy we just made. */
368 for (i = 0; i < IDT_ENTRIES; i++)
369 default_idt_entries[i] += switcher_offset();
370
371 /*
372 * Set up the Switcher's per-cpu areas.
373 *
374 * Each CPU gets two pages of its own within the high-mapped region
375 * (aka. "struct lguest_pages"). Much of this can be initialized now,
376 * but some depends on what Guest we are running (which is set up in
377 * copy_in_guest_info()).
378 */
379 for_each_possible_cpu(i) {
380 /* lguest_pages() returns this CPU's two pages. */
381 struct lguest_pages *pages = lguest_pages(i);
382 /* This is a convenience pointer to make the code fit one
383 * statement to a line. */
384 struct lguest_ro_state *state = &pages->state;
385
386 /* The Global Descriptor Table: the Host has a different one
387 * for each CPU. We keep a descriptor for the GDT which says
388 * where it is and how big it is (the size is actually the last
389 * byte, not the size, hence the "-1"). */
390 state->host_gdt_desc.size = GDT_SIZE-1;
391 state->host_gdt_desc.address = (long)get_cpu_gdt_table(i);
392
393 /* All CPUs on the Host use the same Interrupt Descriptor
394 * Table, so we just use store_idt(), which gets this CPU's IDT
395 * descriptor. */
396 store_idt(&state->host_idt_desc);
397
398 /* The descriptors for the Guest's GDT and IDT can be filled
399 * out now, too. We copy the GDT & IDT into ->guest_gdt and
400 * ->guest_idt before actually running the Guest. */
401 state->guest_idt_desc.size = sizeof(state->guest_idt)-1;
402 state->guest_idt_desc.address = (long)&state->guest_idt;
403 state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1;
404 state->guest_gdt_desc.address = (long)&state->guest_gdt;
405
406 /* We know where we want the stack to be when the Guest enters
407 * the switcher: in pages->regs. The stack grows upwards, so
408 * we start it at the end of that structure. */
409 state->guest_tss.esp0 = (long)(&pages->regs + 1);
410 /* And this is the GDT entry to use for the stack: we keep a
411 * couple of special LGUEST entries. */
412 state->guest_tss.ss0 = LGUEST_DS;
413
414 /* x86 can have a finegrained bitmap which indicates what I/O
415 * ports the process can use. We set it to the end of our
416 * structure, meaning "none". */
417 state->guest_tss.io_bitmap_base = sizeof(state->guest_tss);
418
419 /* Some GDT entries are the same across all Guests, so we can
420 * set them up now. */
421 setup_default_gdt_entries(state);
422 /* Most IDT entries are the same for all Guests, too.*/
423 setup_default_idt_entries(state, default_idt_entries);
424
425 /* The Host needs to be able to use the LGUEST segments on this
426 * CPU, too, so put them in the Host GDT. */
427 get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
428 get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
429 }
430
431 /* In the Switcher, we want the %cs segment register to use the
432 * LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so
433 * it will be undisturbed when we switch. To change %cs and jump we
434 * need this structure to feed to Intel's "lcall" instruction. */
435 lguest_entry.offset = (long)switch_to_guest + switcher_offset();
436 lguest_entry.segment = LGUEST_CS;
437
438 /* Finally, we need to turn off "Page Global Enable". PGE is an
439 * optimization where page table entries are specially marked to show
440 * they never change. The Host kernel marks all the kernel pages this
441 * way because it's always present, even when userspace is running.
442 *
443 * Lguest breaks this: unbeknownst to the rest of the Host kernel, we
444 * switch to the Guest kernel. If you don't disable this on all CPUs,
445 * you'll get really weird bugs that you'll chase for two days.
446 *
447 * I used to turn PGE off every time we switched to the Guest and back
448 * on when we return, but that slowed the Switcher down noticibly. */
449
450 /* We don't need the complexity of CPUs coming and going while we're
451 * doing this. */
452 lock_cpu_hotplug();
453 if (cpu_has_pge) { /* We have a broader idea of "global". */
454 /* Remember that this was originally set (for cleanup). */
455 cpu_had_pge = 1;
456 /* adjust_pge is a helper function which sets or unsets the PGE
457 * bit on its CPU, depending on the argument (0 == unset). */
458 on_each_cpu(adjust_pge, (void *)0, 0, 1);
459 /* Turn off the feature in the global feature set. */
460 clear_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
461 }
462 unlock_cpu_hotplug();
463};
464/*:*/
465
466void __exit lguest_arch_host_fini(void)
467{
468 /* If we had PGE before we started, turn it back on now. */
469 lock_cpu_hotplug();
470 if (cpu_had_pge) {
471 set_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
472 /* adjust_pge's argument "1" means set PGE. */
473 on_each_cpu(adjust_pge, (void *)1, 0, 1);
474 }
475 unlock_cpu_hotplug();
476}
diff --git a/drivers/lguest/x86/switcher_32.S b/drivers/lguest/x86/switcher_32.S
index a3d23f79cba4..e66cec5ac240 100644
--- a/drivers/lguest/x86/switcher_32.S
+++ b/drivers/lguest/x86/switcher_32.S
@@ -48,7 +48,8 @@
48#include <linux/linkage.h> 48#include <linux/linkage.h>
49#include <asm/asm-offsets.h> 49#include <asm/asm-offsets.h>
50#include <asm/page.h> 50#include <asm/page.h>
51#include "../lg.h" 51#include <asm/segment.h>
52#include <asm/lguest.h>
52 53
53// We mark the start of the code to copy 54// We mark the start of the code to copy
54// It's placed in .text tho it's never run here 55// It's placed in .text tho it's never run here