diff options
Diffstat (limited to 'drivers/lguest/page_tables.c')
-rw-r--r-- | drivers/lguest/page_tables.c | 427 |
1 files changed, 282 insertions, 145 deletions
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 | */ | ||
70 | static DEFINE_PER_CPU(pte_t *, switcher_pte_pages); | 79 | static 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 | */ | ||
83 | static pgd_t *spgd_addr(struct lg_cpu *cpu, u32 i, unsigned long vaddr) | 94 | static 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 | */ | ||
102 | static pmd_t *spmd_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr) | 115 | static 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 | */ | ||
125 | static pte_t *spte_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr) | 140 | static 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 | */ | ||
144 | static unsigned long gpgd_addr(struct lg_cpu *cpu, unsigned long vaddr) | 161 | static 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 | */ | ||
189 | static unsigned long get_pfn(unsigned long virtpfn, int write) | 210 | static 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 | */ | ||
205 | static pte_t gpte_to_spte(struct lg_cpu *cpu, pte_t gpte, int write) | 228 | static 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: */ |
235 | static void release_pte(pte_t pte) | 264 | static 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 | */ | ||
277 | bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode) | 309 | bool 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 | */ | ||
412 | static bool page_writable(struct lg_cpu *cpu, unsigned long vaddr) | 472 | static 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 | */ | ||
441 | void pin_page(struct lg_cpu *cpu, unsigned long vaddr) | 505 | void 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 | */ | ||
505 | static void flush_user_mappings(struct lguest *lg, int idx) | 574 | static 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 | */ | ||
517 | void guest_pagetable_flush_user(struct lg_cpu *cpu) | 588 | void 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 | */ | ||
557 | static unsigned int find_pgdir(struct lguest *lg, unsigned long pgtable) | 630 | static 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 | */ | ||
569 | static unsigned int new_pgdir(struct lg_cpu *cpu, | 644 | static 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 | */ | ||
623 | void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable) | 706 | void 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 | */ | ||
643 | static void release_all_pagetables(struct lguest *lg) | 730 | static 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 | */ | ||
674 | void guest_pagetable_clear_all(struct lg_cpu *cpu) | 765 | void 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 | */ | ||
752 | void guest_set_pte(struct lg_cpu *cpu, | 853 | void 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 | */ | ||
811 | static unsigned long setup_pagetables(struct lguest *lg, | 916 | static 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 | */ | ||
892 | int init_guest_pagetable(struct lguest *lg) | 1007 | int 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 | */ | ||
973 | void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages) | 1098 | void 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(®s_pte, pfn_pte(pfn, PAGE_KERNEL)); | 1137 | native_set_pte(®s_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 | */ | ||
1026 | static __init void populate_switcher_pte_page(unsigned int cpu, | 1157 | static __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; |