]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (C) 2012 - Virtual Open Systems and Columbia University | |
3 | * Author: Christoffer Dall <c.dall@virtualopensystems.com> | |
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, version 2, as | |
7 | * published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. | |
17 | */ | |
18 | ||
19 | #include <linux/mman.h> | |
20 | #include <linux/kvm_host.h> | |
21 | #include <linux/io.h> | |
22 | #include <linux/hugetlb.h> | |
23 | #include <trace/events/kvm.h> | |
24 | #include <asm/pgalloc.h> | |
25 | #include <asm/cacheflush.h> | |
26 | #include <asm/kvm_arm.h> | |
27 | #include <asm/kvm_mmu.h> | |
28 | #include <asm/kvm_mmio.h> | |
29 | #include <asm/kvm_asm.h> | |
30 | #include <asm/kvm_emulate.h> | |
31 | ||
32 | #include "trace.h" | |
33 | ||
34 | extern char __hyp_idmap_text_start[], __hyp_idmap_text_end[]; | |
35 | ||
36 | static pgd_t *boot_hyp_pgd; | |
37 | static pgd_t *hyp_pgd; | |
38 | static pgd_t *merged_hyp_pgd; | |
39 | static DEFINE_MUTEX(kvm_hyp_pgd_mutex); | |
40 | ||
41 | static unsigned long hyp_idmap_start; | |
42 | static unsigned long hyp_idmap_end; | |
43 | static phys_addr_t hyp_idmap_vector; | |
44 | ||
45 | #define hyp_pgd_order get_order(PTRS_PER_PGD * sizeof(pgd_t)) | |
46 | ||
47 | #define kvm_pmd_huge(_x) (pmd_huge(_x) || pmd_trans_huge(_x)) | |
48 | #define kvm_pud_huge(_x) pud_huge(_x) | |
49 | ||
50 | #define KVM_S2PTE_FLAG_IS_IOMAP (1UL << 0) | |
51 | #define KVM_S2_FLAG_LOGGING_ACTIVE (1UL << 1) | |
52 | ||
53 | static bool memslot_is_logging(struct kvm_memory_slot *memslot) | |
54 | { | |
55 | return memslot->dirty_bitmap && !(memslot->flags & KVM_MEM_READONLY); | |
56 | } | |
57 | ||
58 | /** | |
59 | * kvm_flush_remote_tlbs() - flush all VM TLB entries for v7/8 | |
60 | * @kvm: pointer to kvm structure. | |
61 | * | |
62 | * Interface to HYP function to flush all VM TLB entries | |
63 | */ | |
64 | void kvm_flush_remote_tlbs(struct kvm *kvm) | |
65 | { | |
66 | kvm_call_hyp(__kvm_tlb_flush_vmid, kvm); | |
67 | } | |
68 | ||
69 | static void kvm_tlb_flush_vmid_ipa(struct kvm *kvm, phys_addr_t ipa) | |
70 | { | |
71 | /* | |
72 | * This function also gets called when dealing with HYP page | |
73 | * tables. As HYP doesn't have an associated struct kvm (and | |
74 | * the HYP page tables are fairly static), we don't do | |
75 | * anything there. | |
76 | */ | |
77 | if (kvm) | |
78 | kvm_call_hyp(__kvm_tlb_flush_vmid_ipa, kvm, ipa); | |
79 | } | |
80 | ||
81 | /* | |
82 | * D-Cache management functions. They take the page table entries by | |
83 | * value, as they are flushing the cache using the kernel mapping (or | |
84 | * kmap on 32bit). | |
85 | */ | |
86 | static void kvm_flush_dcache_pte(pte_t pte) | |
87 | { | |
88 | __kvm_flush_dcache_pte(pte); | |
89 | } | |
90 | ||
91 | static void kvm_flush_dcache_pmd(pmd_t pmd) | |
92 | { | |
93 | __kvm_flush_dcache_pmd(pmd); | |
94 | } | |
95 | ||
96 | static void kvm_flush_dcache_pud(pud_t pud) | |
97 | { | |
98 | __kvm_flush_dcache_pud(pud); | |
99 | } | |
100 | ||
101 | static bool kvm_is_device_pfn(unsigned long pfn) | |
102 | { | |
103 | return !pfn_valid(pfn); | |
104 | } | |
105 | ||
106 | /** | |
107 | * stage2_dissolve_pmd() - clear and flush huge PMD entry | |
108 | * @kvm: pointer to kvm structure. | |
109 | * @addr: IPA | |
110 | * @pmd: pmd pointer for IPA | |
111 | * | |
112 | * Function clears a PMD entry, flushes addr 1st and 2nd stage TLBs. Marks all | |
113 | * pages in the range dirty. | |
114 | */ | |
115 | static void stage2_dissolve_pmd(struct kvm *kvm, phys_addr_t addr, pmd_t *pmd) | |
116 | { | |
117 | if (!kvm_pmd_huge(*pmd)) | |
118 | return; | |
119 | ||
120 | pmd_clear(pmd); | |
121 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
122 | put_page(virt_to_page(pmd)); | |
123 | } | |
124 | ||
125 | static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache, | |
126 | int min, int max) | |
127 | { | |
128 | void *page; | |
129 | ||
130 | BUG_ON(max > KVM_NR_MEM_OBJS); | |
131 | if (cache->nobjs >= min) | |
132 | return 0; | |
133 | while (cache->nobjs < max) { | |
134 | page = (void *)__get_free_page(PGALLOC_GFP); | |
135 | if (!page) | |
136 | return -ENOMEM; | |
137 | cache->objects[cache->nobjs++] = page; | |
138 | } | |
139 | return 0; | |
140 | } | |
141 | ||
142 | static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
143 | { | |
144 | while (mc->nobjs) | |
145 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
146 | } | |
147 | ||
148 | static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
149 | { | |
150 | void *p; | |
151 | ||
152 | BUG_ON(!mc || !mc->nobjs); | |
153 | p = mc->objects[--mc->nobjs]; | |
154 | return p; | |
155 | } | |
156 | ||
157 | static void clear_pgd_entry(struct kvm *kvm, pgd_t *pgd, phys_addr_t addr) | |
158 | { | |
159 | pud_t *pud_table __maybe_unused = pud_offset(pgd, 0); | |
160 | pgd_clear(pgd); | |
161 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
162 | pud_free(NULL, pud_table); | |
163 | put_page(virt_to_page(pgd)); | |
164 | } | |
165 | ||
166 | static void clear_pud_entry(struct kvm *kvm, pud_t *pud, phys_addr_t addr) | |
167 | { | |
168 | pmd_t *pmd_table = pmd_offset(pud, 0); | |
169 | VM_BUG_ON(pud_huge(*pud)); | |
170 | pud_clear(pud); | |
171 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
172 | pmd_free(NULL, pmd_table); | |
173 | put_page(virt_to_page(pud)); | |
174 | } | |
175 | ||
176 | static void clear_pmd_entry(struct kvm *kvm, pmd_t *pmd, phys_addr_t addr) | |
177 | { | |
178 | pte_t *pte_table = pte_offset_kernel(pmd, 0); | |
179 | VM_BUG_ON(kvm_pmd_huge(*pmd)); | |
180 | pmd_clear(pmd); | |
181 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
182 | pte_free_kernel(NULL, pte_table); | |
183 | put_page(virt_to_page(pmd)); | |
184 | } | |
185 | ||
186 | /* | |
187 | * Unmapping vs dcache management: | |
188 | * | |
189 | * If a guest maps certain memory pages as uncached, all writes will | |
190 | * bypass the data cache and go directly to RAM. However, the CPUs | |
191 | * can still speculate reads (not writes) and fill cache lines with | |
192 | * data. | |
193 | * | |
194 | * Those cache lines will be *clean* cache lines though, so a | |
195 | * clean+invalidate operation is equivalent to an invalidate | |
196 | * operation, because no cache lines are marked dirty. | |
197 | * | |
198 | * Those clean cache lines could be filled prior to an uncached write | |
199 | * by the guest, and the cache coherent IO subsystem would therefore | |
200 | * end up writing old data to disk. | |
201 | * | |
202 | * This is why right after unmapping a page/section and invalidating | |
203 | * the corresponding TLBs, we call kvm_flush_dcache_p*() to make sure | |
204 | * the IO subsystem will never hit in the cache. | |
205 | */ | |
206 | static void unmap_ptes(struct kvm *kvm, pmd_t *pmd, | |
207 | phys_addr_t addr, phys_addr_t end) | |
208 | { | |
209 | phys_addr_t start_addr = addr; | |
210 | pte_t *pte, *start_pte; | |
211 | ||
212 | start_pte = pte = pte_offset_kernel(pmd, addr); | |
213 | do { | |
214 | if (!pte_none(*pte)) { | |
215 | pte_t old_pte = *pte; | |
216 | ||
217 | kvm_set_pte(pte, __pte(0)); | |
218 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
219 | ||
220 | /* No need to invalidate the cache for device mappings */ | |
221 | if (!kvm_is_device_pfn(pte_pfn(old_pte))) | |
222 | kvm_flush_dcache_pte(old_pte); | |
223 | ||
224 | put_page(virt_to_page(pte)); | |
225 | } | |
226 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
227 | ||
228 | if (kvm_pte_table_empty(kvm, start_pte)) | |
229 | clear_pmd_entry(kvm, pmd, start_addr); | |
230 | } | |
231 | ||
232 | static void unmap_pmds(struct kvm *kvm, pud_t *pud, | |
233 | phys_addr_t addr, phys_addr_t end) | |
234 | { | |
235 | phys_addr_t next, start_addr = addr; | |
236 | pmd_t *pmd, *start_pmd; | |
237 | ||
238 | start_pmd = pmd = pmd_offset(pud, addr); | |
239 | do { | |
240 | next = kvm_pmd_addr_end(addr, end); | |
241 | if (!pmd_none(*pmd)) { | |
242 | if (kvm_pmd_huge(*pmd)) { | |
243 | pmd_t old_pmd = *pmd; | |
244 | ||
245 | pmd_clear(pmd); | |
246 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
247 | ||
248 | kvm_flush_dcache_pmd(old_pmd); | |
249 | ||
250 | put_page(virt_to_page(pmd)); | |
251 | } else { | |
252 | unmap_ptes(kvm, pmd, addr, next); | |
253 | } | |
254 | } | |
255 | } while (pmd++, addr = next, addr != end); | |
256 | ||
257 | if (kvm_pmd_table_empty(kvm, start_pmd)) | |
258 | clear_pud_entry(kvm, pud, start_addr); | |
259 | } | |
260 | ||
261 | static void unmap_puds(struct kvm *kvm, pgd_t *pgd, | |
262 | phys_addr_t addr, phys_addr_t end) | |
263 | { | |
264 | phys_addr_t next, start_addr = addr; | |
265 | pud_t *pud, *start_pud; | |
266 | ||
267 | start_pud = pud = pud_offset(pgd, addr); | |
268 | do { | |
269 | next = kvm_pud_addr_end(addr, end); | |
270 | if (!pud_none(*pud)) { | |
271 | if (pud_huge(*pud)) { | |
272 | pud_t old_pud = *pud; | |
273 | ||
274 | pud_clear(pud); | |
275 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
276 | ||
277 | kvm_flush_dcache_pud(old_pud); | |
278 | ||
279 | put_page(virt_to_page(pud)); | |
280 | } else { | |
281 | unmap_pmds(kvm, pud, addr, next); | |
282 | } | |
283 | } | |
284 | } while (pud++, addr = next, addr != end); | |
285 | ||
286 | if (kvm_pud_table_empty(kvm, start_pud)) | |
287 | clear_pgd_entry(kvm, pgd, start_addr); | |
288 | } | |
289 | ||
290 | ||
291 | static void unmap_range(struct kvm *kvm, pgd_t *pgdp, | |
292 | phys_addr_t start, u64 size) | |
293 | { | |
294 | pgd_t *pgd; | |
295 | phys_addr_t addr = start, end = start + size; | |
296 | phys_addr_t next; | |
297 | ||
298 | pgd = pgdp + kvm_pgd_index(addr); | |
299 | do { | |
300 | next = kvm_pgd_addr_end(addr, end); | |
301 | if (!pgd_none(*pgd)) | |
302 | unmap_puds(kvm, pgd, addr, next); | |
303 | } while (pgd++, addr = next, addr != end); | |
304 | } | |
305 | ||
306 | static void stage2_flush_ptes(struct kvm *kvm, pmd_t *pmd, | |
307 | phys_addr_t addr, phys_addr_t end) | |
308 | { | |
309 | pte_t *pte; | |
310 | ||
311 | pte = pte_offset_kernel(pmd, addr); | |
312 | do { | |
313 | if (!pte_none(*pte) && !kvm_is_device_pfn(pte_pfn(*pte))) | |
314 | kvm_flush_dcache_pte(*pte); | |
315 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
316 | } | |
317 | ||
318 | static void stage2_flush_pmds(struct kvm *kvm, pud_t *pud, | |
319 | phys_addr_t addr, phys_addr_t end) | |
320 | { | |
321 | pmd_t *pmd; | |
322 | phys_addr_t next; | |
323 | ||
324 | pmd = pmd_offset(pud, addr); | |
325 | do { | |
326 | next = kvm_pmd_addr_end(addr, end); | |
327 | if (!pmd_none(*pmd)) { | |
328 | if (kvm_pmd_huge(*pmd)) | |
329 | kvm_flush_dcache_pmd(*pmd); | |
330 | else | |
331 | stage2_flush_ptes(kvm, pmd, addr, next); | |
332 | } | |
333 | } while (pmd++, addr = next, addr != end); | |
334 | } | |
335 | ||
336 | static void stage2_flush_puds(struct kvm *kvm, pgd_t *pgd, | |
337 | phys_addr_t addr, phys_addr_t end) | |
338 | { | |
339 | pud_t *pud; | |
340 | phys_addr_t next; | |
341 | ||
342 | pud = pud_offset(pgd, addr); | |
343 | do { | |
344 | next = kvm_pud_addr_end(addr, end); | |
345 | if (!pud_none(*pud)) { | |
346 | if (pud_huge(*pud)) | |
347 | kvm_flush_dcache_pud(*pud); | |
348 | else | |
349 | stage2_flush_pmds(kvm, pud, addr, next); | |
350 | } | |
351 | } while (pud++, addr = next, addr != end); | |
352 | } | |
353 | ||
354 | static void stage2_flush_memslot(struct kvm *kvm, | |
355 | struct kvm_memory_slot *memslot) | |
356 | { | |
357 | phys_addr_t addr = memslot->base_gfn << PAGE_SHIFT; | |
358 | phys_addr_t end = addr + PAGE_SIZE * memslot->npages; | |
359 | phys_addr_t next; | |
360 | pgd_t *pgd; | |
361 | ||
362 | pgd = kvm->arch.pgd + kvm_pgd_index(addr); | |
363 | do { | |
364 | next = kvm_pgd_addr_end(addr, end); | |
365 | stage2_flush_puds(kvm, pgd, addr, next); | |
366 | } while (pgd++, addr = next, addr != end); | |
367 | } | |
368 | ||
369 | /** | |
370 | * stage2_flush_vm - Invalidate cache for pages mapped in stage 2 | |
371 | * @kvm: The struct kvm pointer | |
372 | * | |
373 | * Go through the stage 2 page tables and invalidate any cache lines | |
374 | * backing memory already mapped to the VM. | |
375 | */ | |
376 | static void stage2_flush_vm(struct kvm *kvm) | |
377 | { | |
378 | struct kvm_memslots *slots; | |
379 | struct kvm_memory_slot *memslot; | |
380 | int idx; | |
381 | ||
382 | idx = srcu_read_lock(&kvm->srcu); | |
383 | spin_lock(&kvm->mmu_lock); | |
384 | ||
385 | slots = kvm_memslots(kvm); | |
386 | kvm_for_each_memslot(memslot, slots) | |
387 | stage2_flush_memslot(kvm, memslot); | |
388 | ||
389 | spin_unlock(&kvm->mmu_lock); | |
390 | srcu_read_unlock(&kvm->srcu, idx); | |
391 | } | |
392 | ||
393 | /** | |
394 | * free_boot_hyp_pgd - free HYP boot page tables | |
395 | * | |
396 | * Free the HYP boot page tables. The bounce page is also freed. | |
397 | */ | |
398 | void free_boot_hyp_pgd(void) | |
399 | { | |
400 | mutex_lock(&kvm_hyp_pgd_mutex); | |
401 | ||
402 | if (boot_hyp_pgd) { | |
403 | unmap_range(NULL, boot_hyp_pgd, hyp_idmap_start, PAGE_SIZE); | |
404 | unmap_range(NULL, boot_hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE); | |
405 | free_pages((unsigned long)boot_hyp_pgd, hyp_pgd_order); | |
406 | boot_hyp_pgd = NULL; | |
407 | } | |
408 | ||
409 | if (hyp_pgd) | |
410 | unmap_range(NULL, hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE); | |
411 | ||
412 | mutex_unlock(&kvm_hyp_pgd_mutex); | |
413 | } | |
414 | ||
415 | /** | |
416 | * free_hyp_pgds - free Hyp-mode page tables | |
417 | * | |
418 | * Assumes hyp_pgd is a page table used strictly in Hyp-mode and | |
419 | * therefore contains either mappings in the kernel memory area (above | |
420 | * PAGE_OFFSET), or device mappings in the vmalloc range (from | |
421 | * VMALLOC_START to VMALLOC_END). | |
422 | * | |
423 | * boot_hyp_pgd should only map two pages for the init code. | |
424 | */ | |
425 | void free_hyp_pgds(void) | |
426 | { | |
427 | unsigned long addr; | |
428 | ||
429 | free_boot_hyp_pgd(); | |
430 | ||
431 | mutex_lock(&kvm_hyp_pgd_mutex); | |
432 | ||
433 | if (hyp_pgd) { | |
434 | for (addr = PAGE_OFFSET; virt_addr_valid(addr); addr += PGDIR_SIZE) | |
435 | unmap_range(NULL, hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE); | |
436 | for (addr = VMALLOC_START; is_vmalloc_addr((void*)addr); addr += PGDIR_SIZE) | |
437 | unmap_range(NULL, hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE); | |
438 | ||
439 | free_pages((unsigned long)hyp_pgd, hyp_pgd_order); | |
440 | hyp_pgd = NULL; | |
441 | } | |
442 | if (merged_hyp_pgd) { | |
443 | clear_page(merged_hyp_pgd); | |
444 | free_page((unsigned long)merged_hyp_pgd); | |
445 | merged_hyp_pgd = NULL; | |
446 | } | |
447 | ||
448 | mutex_unlock(&kvm_hyp_pgd_mutex); | |
449 | } | |
450 | ||
451 | static void create_hyp_pte_mappings(pmd_t *pmd, unsigned long start, | |
452 | unsigned long end, unsigned long pfn, | |
453 | pgprot_t prot) | |
454 | { | |
455 | pte_t *pte; | |
456 | unsigned long addr; | |
457 | ||
458 | addr = start; | |
459 | do { | |
460 | pte = pte_offset_kernel(pmd, addr); | |
461 | kvm_set_pte(pte, pfn_pte(pfn, prot)); | |
462 | get_page(virt_to_page(pte)); | |
463 | kvm_flush_dcache_to_poc(pte, sizeof(*pte)); | |
464 | pfn++; | |
465 | } while (addr += PAGE_SIZE, addr != end); | |
466 | } | |
467 | ||
468 | static int create_hyp_pmd_mappings(pud_t *pud, unsigned long start, | |
469 | unsigned long end, unsigned long pfn, | |
470 | pgprot_t prot) | |
471 | { | |
472 | pmd_t *pmd; | |
473 | pte_t *pte; | |
474 | unsigned long addr, next; | |
475 | ||
476 | addr = start; | |
477 | do { | |
478 | pmd = pmd_offset(pud, addr); | |
479 | ||
480 | BUG_ON(pmd_sect(*pmd)); | |
481 | ||
482 | if (pmd_none(*pmd)) { | |
483 | pte = pte_alloc_one_kernel(NULL, addr); | |
484 | if (!pte) { | |
485 | kvm_err("Cannot allocate Hyp pte\n"); | |
486 | return -ENOMEM; | |
487 | } | |
488 | pmd_populate_kernel(NULL, pmd, pte); | |
489 | get_page(virt_to_page(pmd)); | |
490 | kvm_flush_dcache_to_poc(pmd, sizeof(*pmd)); | |
491 | } | |
492 | ||
493 | next = pmd_addr_end(addr, end); | |
494 | ||
495 | create_hyp_pte_mappings(pmd, addr, next, pfn, prot); | |
496 | pfn += (next - addr) >> PAGE_SHIFT; | |
497 | } while (addr = next, addr != end); | |
498 | ||
499 | return 0; | |
500 | } | |
501 | ||
502 | static int create_hyp_pud_mappings(pgd_t *pgd, unsigned long start, | |
503 | unsigned long end, unsigned long pfn, | |
504 | pgprot_t prot) | |
505 | { | |
506 | pud_t *pud; | |
507 | pmd_t *pmd; | |
508 | unsigned long addr, next; | |
509 | int ret; | |
510 | ||
511 | addr = start; | |
512 | do { | |
513 | pud = pud_offset(pgd, addr); | |
514 | ||
515 | if (pud_none_or_clear_bad(pud)) { | |
516 | pmd = pmd_alloc_one(NULL, addr); | |
517 | if (!pmd) { | |
518 | kvm_err("Cannot allocate Hyp pmd\n"); | |
519 | return -ENOMEM; | |
520 | } | |
521 | pud_populate(NULL, pud, pmd); | |
522 | get_page(virt_to_page(pud)); | |
523 | kvm_flush_dcache_to_poc(pud, sizeof(*pud)); | |
524 | } | |
525 | ||
526 | next = pud_addr_end(addr, end); | |
527 | ret = create_hyp_pmd_mappings(pud, addr, next, pfn, prot); | |
528 | if (ret) | |
529 | return ret; | |
530 | pfn += (next - addr) >> PAGE_SHIFT; | |
531 | } while (addr = next, addr != end); | |
532 | ||
533 | return 0; | |
534 | } | |
535 | ||
536 | static int __create_hyp_mappings(pgd_t *pgdp, | |
537 | unsigned long start, unsigned long end, | |
538 | unsigned long pfn, pgprot_t prot) | |
539 | { | |
540 | pgd_t *pgd; | |
541 | pud_t *pud; | |
542 | unsigned long addr, next; | |
543 | int err = 0; | |
544 | ||
545 | mutex_lock(&kvm_hyp_pgd_mutex); | |
546 | addr = start & PAGE_MASK; | |
547 | end = PAGE_ALIGN(end); | |
548 | do { | |
549 | pgd = pgdp + pgd_index(addr); | |
550 | ||
551 | if (pgd_none(*pgd)) { | |
552 | pud = pud_alloc_one(NULL, addr); | |
553 | if (!pud) { | |
554 | kvm_err("Cannot allocate Hyp pud\n"); | |
555 | err = -ENOMEM; | |
556 | goto out; | |
557 | } | |
558 | pgd_populate(NULL, pgd, pud); | |
559 | get_page(virt_to_page(pgd)); | |
560 | kvm_flush_dcache_to_poc(pgd, sizeof(*pgd)); | |
561 | } | |
562 | ||
563 | next = pgd_addr_end(addr, end); | |
564 | err = create_hyp_pud_mappings(pgd, addr, next, pfn, prot); | |
565 | if (err) | |
566 | goto out; | |
567 | pfn += (next - addr) >> PAGE_SHIFT; | |
568 | } while (addr = next, addr != end); | |
569 | out: | |
570 | mutex_unlock(&kvm_hyp_pgd_mutex); | |
571 | return err; | |
572 | } | |
573 | ||
574 | static phys_addr_t kvm_kaddr_to_phys(void *kaddr) | |
575 | { | |
576 | if (!is_vmalloc_addr(kaddr)) { | |
577 | BUG_ON(!virt_addr_valid(kaddr)); | |
578 | return __pa(kaddr); | |
579 | } else { | |
580 | return page_to_phys(vmalloc_to_page(kaddr)) + | |
581 | offset_in_page(kaddr); | |
582 | } | |
583 | } | |
584 | ||
585 | /** | |
586 | * create_hyp_mappings - duplicate a kernel virtual address range in Hyp mode | |
587 | * @from: The virtual kernel start address of the range | |
588 | * @to: The virtual kernel end address of the range (exclusive) | |
589 | * | |
590 | * The same virtual address as the kernel virtual address is also used | |
591 | * in Hyp-mode mapping (modulo HYP_PAGE_OFFSET) to the same underlying | |
592 | * physical pages. | |
593 | */ | |
594 | int create_hyp_mappings(void *from, void *to) | |
595 | { | |
596 | phys_addr_t phys_addr; | |
597 | unsigned long virt_addr; | |
598 | unsigned long start = KERN_TO_HYP((unsigned long)from); | |
599 | unsigned long end = KERN_TO_HYP((unsigned long)to); | |
600 | ||
601 | start = start & PAGE_MASK; | |
602 | end = PAGE_ALIGN(end); | |
603 | ||
604 | for (virt_addr = start; virt_addr < end; virt_addr += PAGE_SIZE) { | |
605 | int err; | |
606 | ||
607 | phys_addr = kvm_kaddr_to_phys(from + virt_addr - start); | |
608 | err = __create_hyp_mappings(hyp_pgd, virt_addr, | |
609 | virt_addr + PAGE_SIZE, | |
610 | __phys_to_pfn(phys_addr), | |
611 | PAGE_HYP); | |
612 | if (err) | |
613 | return err; | |
614 | } | |
615 | ||
616 | return 0; | |
617 | } | |
618 | ||
619 | /** | |
620 | * create_hyp_io_mappings - duplicate a kernel IO mapping into Hyp mode | |
621 | * @from: The kernel start VA of the range | |
622 | * @to: The kernel end VA of the range (exclusive) | |
623 | * @phys_addr: The physical start address which gets mapped | |
624 | * | |
625 | * The resulting HYP VA is the same as the kernel VA, modulo | |
626 | * HYP_PAGE_OFFSET. | |
627 | */ | |
628 | int create_hyp_io_mappings(void *from, void *to, phys_addr_t phys_addr) | |
629 | { | |
630 | unsigned long start = KERN_TO_HYP((unsigned long)from); | |
631 | unsigned long end = KERN_TO_HYP((unsigned long)to); | |
632 | ||
633 | /* Check for a valid kernel IO mapping */ | |
634 | if (!is_vmalloc_addr(from) || !is_vmalloc_addr(to - 1)) | |
635 | return -EINVAL; | |
636 | ||
637 | return __create_hyp_mappings(hyp_pgd, start, end, | |
638 | __phys_to_pfn(phys_addr), PAGE_HYP_DEVICE); | |
639 | } | |
640 | ||
641 | /* Free the HW pgd, one page at a time */ | |
642 | static void kvm_free_hwpgd(void *hwpgd) | |
643 | { | |
644 | free_pages_exact(hwpgd, kvm_get_hwpgd_size()); | |
645 | } | |
646 | ||
647 | /* Allocate the HW PGD, making sure that each page gets its own refcount */ | |
648 | static void *kvm_alloc_hwpgd(void) | |
649 | { | |
650 | unsigned int size = kvm_get_hwpgd_size(); | |
651 | ||
652 | return alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO); | |
653 | } | |
654 | ||
655 | /** | |
656 | * kvm_alloc_stage2_pgd - allocate level-1 table for stage-2 translation. | |
657 | * @kvm: The KVM struct pointer for the VM. | |
658 | * | |
659 | * Allocates the 1st level table only of size defined by S2_PGD_ORDER (can | |
660 | * support either full 40-bit input addresses or limited to 32-bit input | |
661 | * addresses). Clears the allocated pages. | |
662 | * | |
663 | * Note we don't need locking here as this is only called when the VM is | |
664 | * created, which can only be done once. | |
665 | */ | |
666 | int kvm_alloc_stage2_pgd(struct kvm *kvm) | |
667 | { | |
668 | pgd_t *pgd; | |
669 | void *hwpgd; | |
670 | ||
671 | if (kvm->arch.pgd != NULL) { | |
672 | kvm_err("kvm_arch already initialized?\n"); | |
673 | return -EINVAL; | |
674 | } | |
675 | ||
676 | hwpgd = kvm_alloc_hwpgd(); | |
677 | if (!hwpgd) | |
678 | return -ENOMEM; | |
679 | ||
680 | /* When the kernel uses more levels of page tables than the | |
681 | * guest, we allocate a fake PGD and pre-populate it to point | |
682 | * to the next-level page table, which will be the real | |
683 | * initial page table pointed to by the VTTBR. | |
684 | * | |
685 | * When KVM_PREALLOC_LEVEL==2, we allocate a single page for | |
686 | * the PMD and the kernel will use folded pud. | |
687 | * When KVM_PREALLOC_LEVEL==1, we allocate 2 consecutive PUD | |
688 | * pages. | |
689 | */ | |
690 | if (KVM_PREALLOC_LEVEL > 0) { | |
691 | int i; | |
692 | ||
693 | /* | |
694 | * Allocate fake pgd for the page table manipulation macros to | |
695 | * work. This is not used by the hardware and we have no | |
696 | * alignment requirement for this allocation. | |
697 | */ | |
698 | pgd = kmalloc(PTRS_PER_S2_PGD * sizeof(pgd_t), | |
699 | GFP_KERNEL | __GFP_ZERO); | |
700 | ||
701 | if (!pgd) { | |
702 | kvm_free_hwpgd(hwpgd); | |
703 | return -ENOMEM; | |
704 | } | |
705 | ||
706 | /* Plug the HW PGD into the fake one. */ | |
707 | for (i = 0; i < PTRS_PER_S2_PGD; i++) { | |
708 | if (KVM_PREALLOC_LEVEL == 1) | |
709 | pgd_populate(NULL, pgd + i, | |
710 | (pud_t *)hwpgd + i * PTRS_PER_PUD); | |
711 | else if (KVM_PREALLOC_LEVEL == 2) | |
712 | pud_populate(NULL, pud_offset(pgd, 0) + i, | |
713 | (pmd_t *)hwpgd + i * PTRS_PER_PMD); | |
714 | } | |
715 | } else { | |
716 | /* | |
717 | * Allocate actual first-level Stage-2 page table used by the | |
718 | * hardware for Stage-2 page table walks. | |
719 | */ | |
720 | pgd = (pgd_t *)hwpgd; | |
721 | } | |
722 | ||
723 | kvm_clean_pgd(pgd); | |
724 | kvm->arch.pgd = pgd; | |
725 | return 0; | |
726 | } | |
727 | ||
728 | /** | |
729 | * unmap_stage2_range -- Clear stage2 page table entries to unmap a range | |
730 | * @kvm: The VM pointer | |
731 | * @start: The intermediate physical base address of the range to unmap | |
732 | * @size: The size of the area to unmap | |
733 | * | |
734 | * Clear a range of stage-2 mappings, lowering the various ref-counts. Must | |
735 | * be called while holding mmu_lock (unless for freeing the stage2 pgd before | |
736 | * destroying the VM), otherwise another faulting VCPU may come in and mess | |
737 | * with things behind our backs. | |
738 | */ | |
739 | static void unmap_stage2_range(struct kvm *kvm, phys_addr_t start, u64 size) | |
740 | { | |
741 | unmap_range(kvm, kvm->arch.pgd, start, size); | |
742 | } | |
743 | ||
744 | static void stage2_unmap_memslot(struct kvm *kvm, | |
745 | struct kvm_memory_slot *memslot) | |
746 | { | |
747 | hva_t hva = memslot->userspace_addr; | |
748 | phys_addr_t addr = memslot->base_gfn << PAGE_SHIFT; | |
749 | phys_addr_t size = PAGE_SIZE * memslot->npages; | |
750 | hva_t reg_end = hva + size; | |
751 | ||
752 | /* | |
753 | * A memory region could potentially cover multiple VMAs, and any holes | |
754 | * between them, so iterate over all of them to find out if we should | |
755 | * unmap any of them. | |
756 | * | |
757 | * +--------------------------------------------+ | |
758 | * +---------------+----------------+ +----------------+ | |
759 | * | : VMA 1 | VMA 2 | | VMA 3 : | | |
760 | * +---------------+----------------+ +----------------+ | |
761 | * | memory region | | |
762 | * +--------------------------------------------+ | |
763 | */ | |
764 | do { | |
765 | struct vm_area_struct *vma = find_vma(current->mm, hva); | |
766 | hva_t vm_start, vm_end; | |
767 | ||
768 | if (!vma || vma->vm_start >= reg_end) | |
769 | break; | |
770 | ||
771 | /* | |
772 | * Take the intersection of this VMA with the memory region | |
773 | */ | |
774 | vm_start = max(hva, vma->vm_start); | |
775 | vm_end = min(reg_end, vma->vm_end); | |
776 | ||
777 | if (!(vma->vm_flags & VM_PFNMAP)) { | |
778 | gpa_t gpa = addr + (vm_start - memslot->userspace_addr); | |
779 | unmap_stage2_range(kvm, gpa, vm_end - vm_start); | |
780 | } | |
781 | hva = vm_end; | |
782 | } while (hva < reg_end); | |
783 | } | |
784 | ||
785 | /** | |
786 | * stage2_unmap_vm - Unmap Stage-2 RAM mappings | |
787 | * @kvm: The struct kvm pointer | |
788 | * | |
789 | * Go through the memregions and unmap any reguler RAM | |
790 | * backing memory already mapped to the VM. | |
791 | */ | |
792 | void stage2_unmap_vm(struct kvm *kvm) | |
793 | { | |
794 | struct kvm_memslots *slots; | |
795 | struct kvm_memory_slot *memslot; | |
796 | int idx; | |
797 | ||
798 | idx = srcu_read_lock(&kvm->srcu); | |
799 | spin_lock(&kvm->mmu_lock); | |
800 | ||
801 | slots = kvm_memslots(kvm); | |
802 | kvm_for_each_memslot(memslot, slots) | |
803 | stage2_unmap_memslot(kvm, memslot); | |
804 | ||
805 | spin_unlock(&kvm->mmu_lock); | |
806 | srcu_read_unlock(&kvm->srcu, idx); | |
807 | } | |
808 | ||
809 | /** | |
810 | * kvm_free_stage2_pgd - free all stage-2 tables | |
811 | * @kvm: The KVM struct pointer for the VM. | |
812 | * | |
813 | * Walks the level-1 page table pointed to by kvm->arch.pgd and frees all | |
814 | * underlying level-2 and level-3 tables before freeing the actual level-1 table | |
815 | * and setting the struct pointer to NULL. | |
816 | * | |
817 | * Note we don't need locking here as this is only called when the VM is | |
818 | * destroyed, which can only be done once. | |
819 | */ | |
820 | void kvm_free_stage2_pgd(struct kvm *kvm) | |
821 | { | |
822 | if (kvm->arch.pgd == NULL) | |
823 | return; | |
824 | ||
825 | unmap_stage2_range(kvm, 0, KVM_PHYS_SIZE); | |
826 | kvm_free_hwpgd(kvm_get_hwpgd(kvm)); | |
827 | if (KVM_PREALLOC_LEVEL > 0) | |
828 | kfree(kvm->arch.pgd); | |
829 | ||
830 | kvm->arch.pgd = NULL; | |
831 | } | |
832 | ||
833 | static pud_t *stage2_get_pud(struct kvm *kvm, struct kvm_mmu_memory_cache *cache, | |
834 | phys_addr_t addr) | |
835 | { | |
836 | pgd_t *pgd; | |
837 | pud_t *pud; | |
838 | ||
839 | pgd = kvm->arch.pgd + kvm_pgd_index(addr); | |
840 | if (WARN_ON(pgd_none(*pgd))) { | |
841 | if (!cache) | |
842 | return NULL; | |
843 | pud = mmu_memory_cache_alloc(cache); | |
844 | pgd_populate(NULL, pgd, pud); | |
845 | get_page(virt_to_page(pgd)); | |
846 | } | |
847 | ||
848 | return pud_offset(pgd, addr); | |
849 | } | |
850 | ||
851 | static pmd_t *stage2_get_pmd(struct kvm *kvm, struct kvm_mmu_memory_cache *cache, | |
852 | phys_addr_t addr) | |
853 | { | |
854 | pud_t *pud; | |
855 | pmd_t *pmd; | |
856 | ||
857 | pud = stage2_get_pud(kvm, cache, addr); | |
858 | if (pud_none(*pud)) { | |
859 | if (!cache) | |
860 | return NULL; | |
861 | pmd = mmu_memory_cache_alloc(cache); | |
862 | pud_populate(NULL, pud, pmd); | |
863 | get_page(virt_to_page(pud)); | |
864 | } | |
865 | ||
866 | return pmd_offset(pud, addr); | |
867 | } | |
868 | ||
869 | static int stage2_set_pmd_huge(struct kvm *kvm, struct kvm_mmu_memory_cache | |
870 | *cache, phys_addr_t addr, const pmd_t *new_pmd) | |
871 | { | |
872 | pmd_t *pmd, old_pmd; | |
873 | ||
874 | pmd = stage2_get_pmd(kvm, cache, addr); | |
875 | VM_BUG_ON(!pmd); | |
876 | ||
877 | /* | |
878 | * Mapping in huge pages should only happen through a fault. If a | |
879 | * page is merged into a transparent huge page, the individual | |
880 | * subpages of that huge page should be unmapped through MMU | |
881 | * notifiers before we get here. | |
882 | * | |
883 | * Merging of CompoundPages is not supported; they should become | |
884 | * splitting first, unmapped, merged, and mapped back in on-demand. | |
885 | */ | |
886 | VM_BUG_ON(pmd_present(*pmd) && pmd_pfn(*pmd) != pmd_pfn(*new_pmd)); | |
887 | ||
888 | old_pmd = *pmd; | |
889 | kvm_set_pmd(pmd, *new_pmd); | |
890 | if (pmd_present(old_pmd)) | |
891 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
892 | else | |
893 | get_page(virt_to_page(pmd)); | |
894 | return 0; | |
895 | } | |
896 | ||
897 | static int stage2_set_pte(struct kvm *kvm, struct kvm_mmu_memory_cache *cache, | |
898 | phys_addr_t addr, const pte_t *new_pte, | |
899 | unsigned long flags) | |
900 | { | |
901 | pmd_t *pmd; | |
902 | pte_t *pte, old_pte; | |
903 | bool iomap = flags & KVM_S2PTE_FLAG_IS_IOMAP; | |
904 | bool logging_active = flags & KVM_S2_FLAG_LOGGING_ACTIVE; | |
905 | ||
906 | VM_BUG_ON(logging_active && !cache); | |
907 | ||
908 | /* Create stage-2 page table mapping - Levels 0 and 1 */ | |
909 | pmd = stage2_get_pmd(kvm, cache, addr); | |
910 | if (!pmd) { | |
911 | /* | |
912 | * Ignore calls from kvm_set_spte_hva for unallocated | |
913 | * address ranges. | |
914 | */ | |
915 | return 0; | |
916 | } | |
917 | ||
918 | /* | |
919 | * While dirty page logging - dissolve huge PMD, then continue on to | |
920 | * allocate page. | |
921 | */ | |
922 | if (logging_active) | |
923 | stage2_dissolve_pmd(kvm, addr, pmd); | |
924 | ||
925 | /* Create stage-2 page mappings - Level 2 */ | |
926 | if (pmd_none(*pmd)) { | |
927 | if (!cache) | |
928 | return 0; /* ignore calls from kvm_set_spte_hva */ | |
929 | pte = mmu_memory_cache_alloc(cache); | |
930 | kvm_clean_pte(pte); | |
931 | pmd_populate_kernel(NULL, pmd, pte); | |
932 | get_page(virt_to_page(pmd)); | |
933 | } | |
934 | ||
935 | pte = pte_offset_kernel(pmd, addr); | |
936 | ||
937 | if (iomap && pte_present(*pte)) | |
938 | return -EFAULT; | |
939 | ||
940 | /* Create 2nd stage page table mapping - Level 3 */ | |
941 | old_pte = *pte; | |
942 | kvm_set_pte(pte, *new_pte); | |
943 | if (pte_present(old_pte)) | |
944 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
945 | else | |
946 | get_page(virt_to_page(pte)); | |
947 | ||
948 | return 0; | |
949 | } | |
950 | ||
951 | /** | |
952 | * kvm_phys_addr_ioremap - map a device range to guest IPA | |
953 | * | |
954 | * @kvm: The KVM pointer | |
955 | * @guest_ipa: The IPA at which to insert the mapping | |
956 | * @pa: The physical address of the device | |
957 | * @size: The size of the mapping | |
958 | */ | |
959 | int kvm_phys_addr_ioremap(struct kvm *kvm, phys_addr_t guest_ipa, | |
960 | phys_addr_t pa, unsigned long size, bool writable) | |
961 | { | |
962 | phys_addr_t addr, end; | |
963 | int ret = 0; | |
964 | unsigned long pfn; | |
965 | struct kvm_mmu_memory_cache cache = { 0, }; | |
966 | ||
967 | end = (guest_ipa + size + PAGE_SIZE - 1) & PAGE_MASK; | |
968 | pfn = __phys_to_pfn(pa); | |
969 | ||
970 | for (addr = guest_ipa; addr < end; addr += PAGE_SIZE) { | |
971 | pte_t pte = pfn_pte(pfn, PAGE_S2_DEVICE); | |
972 | ||
973 | if (writable) | |
974 | kvm_set_s2pte_writable(&pte); | |
975 | ||
976 | ret = mmu_topup_memory_cache(&cache, KVM_MMU_CACHE_MIN_PAGES, | |
977 | KVM_NR_MEM_OBJS); | |
978 | if (ret) | |
979 | goto out; | |
980 | spin_lock(&kvm->mmu_lock); | |
981 | ret = stage2_set_pte(kvm, &cache, addr, &pte, | |
982 | KVM_S2PTE_FLAG_IS_IOMAP); | |
983 | spin_unlock(&kvm->mmu_lock); | |
984 | if (ret) | |
985 | goto out; | |
986 | ||
987 | pfn++; | |
988 | } | |
989 | ||
990 | out: | |
991 | mmu_free_memory_cache(&cache); | |
992 | return ret; | |
993 | } | |
994 | ||
995 | static bool transparent_hugepage_adjust(pfn_t *pfnp, phys_addr_t *ipap) | |
996 | { | |
997 | pfn_t pfn = *pfnp; | |
998 | gfn_t gfn = *ipap >> PAGE_SHIFT; | |
999 | ||
1000 | if (PageTransCompound(pfn_to_page(pfn))) { | |
1001 | unsigned long mask; | |
1002 | /* | |
1003 | * The address we faulted on is backed by a transparent huge | |
1004 | * page. However, because we map the compound huge page and | |
1005 | * not the individual tail page, we need to transfer the | |
1006 | * refcount to the head page. We have to be careful that the | |
1007 | * THP doesn't start to split while we are adjusting the | |
1008 | * refcounts. | |
1009 | * | |
1010 | * We are sure this doesn't happen, because mmu_notifier_retry | |
1011 | * was successful and we are holding the mmu_lock, so if this | |
1012 | * THP is trying to split, it will be blocked in the mmu | |
1013 | * notifier before touching any of the pages, specifically | |
1014 | * before being able to call __split_huge_page_refcount(). | |
1015 | * | |
1016 | * We can therefore safely transfer the refcount from PG_tail | |
1017 | * to PG_head and switch the pfn from a tail page to the head | |
1018 | * page accordingly. | |
1019 | */ | |
1020 | mask = PTRS_PER_PMD - 1; | |
1021 | VM_BUG_ON((gfn & mask) != (pfn & mask)); | |
1022 | if (pfn & mask) { | |
1023 | *ipap &= PMD_MASK; | |
1024 | kvm_release_pfn_clean(pfn); | |
1025 | pfn &= ~mask; | |
1026 | kvm_get_pfn(pfn); | |
1027 | *pfnp = pfn; | |
1028 | } | |
1029 | ||
1030 | return true; | |
1031 | } | |
1032 | ||
1033 | return false; | |
1034 | } | |
1035 | ||
1036 | static bool kvm_is_write_fault(struct kvm_vcpu *vcpu) | |
1037 | { | |
1038 | if (kvm_vcpu_trap_is_iabt(vcpu)) | |
1039 | return false; | |
1040 | ||
1041 | return kvm_vcpu_dabt_iswrite(vcpu); | |
1042 | } | |
1043 | ||
1044 | /** | |
1045 | * stage2_wp_ptes - write protect PMD range | |
1046 | * @pmd: pointer to pmd entry | |
1047 | * @addr: range start address | |
1048 | * @end: range end address | |
1049 | */ | |
1050 | static void stage2_wp_ptes(pmd_t *pmd, phys_addr_t addr, phys_addr_t end) | |
1051 | { | |
1052 | pte_t *pte; | |
1053 | ||
1054 | pte = pte_offset_kernel(pmd, addr); | |
1055 | do { | |
1056 | if (!pte_none(*pte)) { | |
1057 | if (!kvm_s2pte_readonly(pte)) | |
1058 | kvm_set_s2pte_readonly(pte); | |
1059 | } | |
1060 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
1061 | } | |
1062 | ||
1063 | /** | |
1064 | * stage2_wp_pmds - write protect PUD range | |
1065 | * @pud: pointer to pud entry | |
1066 | * @addr: range start address | |
1067 | * @end: range end address | |
1068 | */ | |
1069 | static void stage2_wp_pmds(pud_t *pud, phys_addr_t addr, phys_addr_t end) | |
1070 | { | |
1071 | pmd_t *pmd; | |
1072 | phys_addr_t next; | |
1073 | ||
1074 | pmd = pmd_offset(pud, addr); | |
1075 | ||
1076 | do { | |
1077 | next = kvm_pmd_addr_end(addr, end); | |
1078 | if (!pmd_none(*pmd)) { | |
1079 | if (kvm_pmd_huge(*pmd)) { | |
1080 | if (!kvm_s2pmd_readonly(pmd)) | |
1081 | kvm_set_s2pmd_readonly(pmd); | |
1082 | } else { | |
1083 | stage2_wp_ptes(pmd, addr, next); | |
1084 | } | |
1085 | } | |
1086 | } while (pmd++, addr = next, addr != end); | |
1087 | } | |
1088 | ||
1089 | /** | |
1090 | * stage2_wp_puds - write protect PGD range | |
1091 | * @pgd: pointer to pgd entry | |
1092 | * @addr: range start address | |
1093 | * @end: range end address | |
1094 | * | |
1095 | * Process PUD entries, for a huge PUD we cause a panic. | |
1096 | */ | |
1097 | static void stage2_wp_puds(pgd_t *pgd, phys_addr_t addr, phys_addr_t end) | |
1098 | { | |
1099 | pud_t *pud; | |
1100 | phys_addr_t next; | |
1101 | ||
1102 | pud = pud_offset(pgd, addr); | |
1103 | do { | |
1104 | next = kvm_pud_addr_end(addr, end); | |
1105 | if (!pud_none(*pud)) { | |
1106 | /* TODO:PUD not supported, revisit later if supported */ | |
1107 | BUG_ON(kvm_pud_huge(*pud)); | |
1108 | stage2_wp_pmds(pud, addr, next); | |
1109 | } | |
1110 | } while (pud++, addr = next, addr != end); | |
1111 | } | |
1112 | ||
1113 | /** | |
1114 | * stage2_wp_range() - write protect stage2 memory region range | |
1115 | * @kvm: The KVM pointer | |
1116 | * @addr: Start address of range | |
1117 | * @end: End address of range | |
1118 | */ | |
1119 | static void stage2_wp_range(struct kvm *kvm, phys_addr_t addr, phys_addr_t end) | |
1120 | { | |
1121 | pgd_t *pgd; | |
1122 | phys_addr_t next; | |
1123 | ||
1124 | pgd = kvm->arch.pgd + kvm_pgd_index(addr); | |
1125 | do { | |
1126 | /* | |
1127 | * Release kvm_mmu_lock periodically if the memory region is | |
1128 | * large. Otherwise, we may see kernel panics with | |
1129 | * CONFIG_DETECT_HUNG_TASK, CONFIG_LOCKUP_DETECTOR, | |
1130 | * CONFIG_LOCKDEP. Additionally, holding the lock too long | |
1131 | * will also starve other vCPUs. | |
1132 | */ | |
1133 | if (need_resched() || spin_needbreak(&kvm->mmu_lock)) | |
1134 | cond_resched_lock(&kvm->mmu_lock); | |
1135 | ||
1136 | next = kvm_pgd_addr_end(addr, end); | |
1137 | if (pgd_present(*pgd)) | |
1138 | stage2_wp_puds(pgd, addr, next); | |
1139 | } while (pgd++, addr = next, addr != end); | |
1140 | } | |
1141 | ||
1142 | /** | |
1143 | * kvm_mmu_wp_memory_region() - write protect stage 2 entries for memory slot | |
1144 | * @kvm: The KVM pointer | |
1145 | * @slot: The memory slot to write protect | |
1146 | * | |
1147 | * Called to start logging dirty pages after memory region | |
1148 | * KVM_MEM_LOG_DIRTY_PAGES operation is called. After this function returns | |
1149 | * all present PMD and PTEs are write protected in the memory region. | |
1150 | * Afterwards read of dirty page log can be called. | |
1151 | * | |
1152 | * Acquires kvm_mmu_lock. Called with kvm->slots_lock mutex acquired, | |
1153 | * serializing operations for VM memory regions. | |
1154 | */ | |
1155 | void kvm_mmu_wp_memory_region(struct kvm *kvm, int slot) | |
1156 | { | |
1157 | struct kvm_memslots *slots = kvm_memslots(kvm); | |
1158 | struct kvm_memory_slot *memslot = id_to_memslot(slots, slot); | |
1159 | phys_addr_t start = memslot->base_gfn << PAGE_SHIFT; | |
1160 | phys_addr_t end = (memslot->base_gfn + memslot->npages) << PAGE_SHIFT; | |
1161 | ||
1162 | spin_lock(&kvm->mmu_lock); | |
1163 | stage2_wp_range(kvm, start, end); | |
1164 | spin_unlock(&kvm->mmu_lock); | |
1165 | kvm_flush_remote_tlbs(kvm); | |
1166 | } | |
1167 | ||
1168 | /** | |
1169 | * kvm_mmu_write_protect_pt_masked() - write protect dirty pages | |
1170 | * @kvm: The KVM pointer | |
1171 | * @slot: The memory slot associated with mask | |
1172 | * @gfn_offset: The gfn offset in memory slot | |
1173 | * @mask: The mask of dirty pages at offset 'gfn_offset' in this memory | |
1174 | * slot to be write protected | |
1175 | * | |
1176 | * Walks bits set in mask write protects the associated pte's. Caller must | |
1177 | * acquire kvm_mmu_lock. | |
1178 | */ | |
1179 | static void kvm_mmu_write_protect_pt_masked(struct kvm *kvm, | |
1180 | struct kvm_memory_slot *slot, | |
1181 | gfn_t gfn_offset, unsigned long mask) | |
1182 | { | |
1183 | phys_addr_t base_gfn = slot->base_gfn + gfn_offset; | |
1184 | phys_addr_t start = (base_gfn + __ffs(mask)) << PAGE_SHIFT; | |
1185 | phys_addr_t end = (base_gfn + __fls(mask) + 1) << PAGE_SHIFT; | |
1186 | ||
1187 | stage2_wp_range(kvm, start, end); | |
1188 | } | |
1189 | ||
1190 | /* | |
1191 | * kvm_arch_mmu_enable_log_dirty_pt_masked - enable dirty logging for selected | |
1192 | * dirty pages. | |
1193 | * | |
1194 | * It calls kvm_mmu_write_protect_pt_masked to write protect selected pages to | |
1195 | * enable dirty logging for them. | |
1196 | */ | |
1197 | void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm, | |
1198 | struct kvm_memory_slot *slot, | |
1199 | gfn_t gfn_offset, unsigned long mask) | |
1200 | { | |
1201 | kvm_mmu_write_protect_pt_masked(kvm, slot, gfn_offset, mask); | |
1202 | } | |
1203 | ||
1204 | static void coherent_cache_guest_page(struct kvm_vcpu *vcpu, pfn_t pfn, | |
1205 | unsigned long size, bool uncached) | |
1206 | { | |
1207 | __coherent_cache_guest_page(vcpu, pfn, size, uncached); | |
1208 | } | |
1209 | ||
1210 | static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa, | |
1211 | struct kvm_memory_slot *memslot, unsigned long hva, | |
1212 | unsigned long fault_status) | |
1213 | { | |
1214 | int ret; | |
1215 | bool write_fault, writable, hugetlb = false, force_pte = false; | |
1216 | unsigned long mmu_seq; | |
1217 | gfn_t gfn = fault_ipa >> PAGE_SHIFT; | |
1218 | struct kvm *kvm = vcpu->kvm; | |
1219 | struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache; | |
1220 | struct vm_area_struct *vma; | |
1221 | pfn_t pfn; | |
1222 | pgprot_t mem_type = PAGE_S2; | |
1223 | bool fault_ipa_uncached; | |
1224 | bool logging_active = memslot_is_logging(memslot); | |
1225 | unsigned long flags = 0; | |
1226 | ||
1227 | write_fault = kvm_is_write_fault(vcpu); | |
1228 | if (fault_status == FSC_PERM && !write_fault) { | |
1229 | kvm_err("Unexpected L2 read permission error\n"); | |
1230 | return -EFAULT; | |
1231 | } | |
1232 | ||
1233 | /* Let's check if we will get back a huge page backed by hugetlbfs */ | |
1234 | down_read(¤t->mm->mmap_sem); | |
1235 | vma = find_vma_intersection(current->mm, hva, hva + 1); | |
1236 | if (unlikely(!vma)) { | |
1237 | kvm_err("Failed to find VMA for hva 0x%lx\n", hva); | |
1238 | up_read(¤t->mm->mmap_sem); | |
1239 | return -EFAULT; | |
1240 | } | |
1241 | ||
1242 | if (is_vm_hugetlb_page(vma) && !logging_active) { | |
1243 | hugetlb = true; | |
1244 | gfn = (fault_ipa & PMD_MASK) >> PAGE_SHIFT; | |
1245 | } else { | |
1246 | /* | |
1247 | * Pages belonging to memslots that don't have the same | |
1248 | * alignment for userspace and IPA cannot be mapped using | |
1249 | * block descriptors even if the pages belong to a THP for | |
1250 | * the process, because the stage-2 block descriptor will | |
1251 | * cover more than a single THP and we loose atomicity for | |
1252 | * unmapping, updates, and splits of the THP or other pages | |
1253 | * in the stage-2 block range. | |
1254 | */ | |
1255 | if ((memslot->userspace_addr & ~PMD_MASK) != | |
1256 | ((memslot->base_gfn << PAGE_SHIFT) & ~PMD_MASK)) | |
1257 | force_pte = true; | |
1258 | } | |
1259 | up_read(¤t->mm->mmap_sem); | |
1260 | ||
1261 | /* We need minimum second+third level pages */ | |
1262 | ret = mmu_topup_memory_cache(memcache, KVM_MMU_CACHE_MIN_PAGES, | |
1263 | KVM_NR_MEM_OBJS); | |
1264 | if (ret) | |
1265 | return ret; | |
1266 | ||
1267 | mmu_seq = vcpu->kvm->mmu_notifier_seq; | |
1268 | /* | |
1269 | * Ensure the read of mmu_notifier_seq happens before we call | |
1270 | * gfn_to_pfn_prot (which calls get_user_pages), so that we don't risk | |
1271 | * the page we just got a reference to gets unmapped before we have a | |
1272 | * chance to grab the mmu_lock, which ensure that if the page gets | |
1273 | * unmapped afterwards, the call to kvm_unmap_hva will take it away | |
1274 | * from us again properly. This smp_rmb() interacts with the smp_wmb() | |
1275 | * in kvm_mmu_notifier_invalidate_<page|range_end>. | |
1276 | */ | |
1277 | smp_rmb(); | |
1278 | ||
1279 | pfn = gfn_to_pfn_prot(kvm, gfn, write_fault, &writable); | |
1280 | if (is_error_pfn(pfn)) | |
1281 | return -EFAULT; | |
1282 | ||
1283 | if (kvm_is_device_pfn(pfn)) { | |
1284 | mem_type = PAGE_S2_DEVICE; | |
1285 | flags |= KVM_S2PTE_FLAG_IS_IOMAP; | |
1286 | } else if (logging_active) { | |
1287 | /* | |
1288 | * Faults on pages in a memslot with logging enabled | |
1289 | * should not be mapped with huge pages (it introduces churn | |
1290 | * and performance degradation), so force a pte mapping. | |
1291 | */ | |
1292 | force_pte = true; | |
1293 | flags |= KVM_S2_FLAG_LOGGING_ACTIVE; | |
1294 | ||
1295 | /* | |
1296 | * Only actually map the page as writable if this was a write | |
1297 | * fault. | |
1298 | */ | |
1299 | if (!write_fault) | |
1300 | writable = false; | |
1301 | } | |
1302 | ||
1303 | spin_lock(&kvm->mmu_lock); | |
1304 | if (mmu_notifier_retry(kvm, mmu_seq)) | |
1305 | goto out_unlock; | |
1306 | ||
1307 | if (!hugetlb && !force_pte) | |
1308 | hugetlb = transparent_hugepage_adjust(&pfn, &fault_ipa); | |
1309 | ||
1310 | fault_ipa_uncached = memslot->flags & KVM_MEMSLOT_INCOHERENT; | |
1311 | ||
1312 | if (hugetlb) { | |
1313 | pmd_t new_pmd = pfn_pmd(pfn, mem_type); | |
1314 | new_pmd = pmd_mkhuge(new_pmd); | |
1315 | if (writable) { | |
1316 | kvm_set_s2pmd_writable(&new_pmd); | |
1317 | kvm_set_pfn_dirty(pfn); | |
1318 | } | |
1319 | coherent_cache_guest_page(vcpu, pfn, PMD_SIZE, fault_ipa_uncached); | |
1320 | ret = stage2_set_pmd_huge(kvm, memcache, fault_ipa, &new_pmd); | |
1321 | } else { | |
1322 | pte_t new_pte = pfn_pte(pfn, mem_type); | |
1323 | ||
1324 | if (writable) { | |
1325 | kvm_set_s2pte_writable(&new_pte); | |
1326 | kvm_set_pfn_dirty(pfn); | |
1327 | mark_page_dirty(kvm, gfn); | |
1328 | } | |
1329 | coherent_cache_guest_page(vcpu, pfn, PAGE_SIZE, fault_ipa_uncached); | |
1330 | ret = stage2_set_pte(kvm, memcache, fault_ipa, &new_pte, flags); | |
1331 | } | |
1332 | ||
1333 | out_unlock: | |
1334 | spin_unlock(&kvm->mmu_lock); | |
1335 | kvm_set_pfn_accessed(pfn); | |
1336 | kvm_release_pfn_clean(pfn); | |
1337 | return ret; | |
1338 | } | |
1339 | ||
1340 | /* | |
1341 | * Resolve the access fault by making the page young again. | |
1342 | * Note that because the faulting entry is guaranteed not to be | |
1343 | * cached in the TLB, we don't need to invalidate anything. | |
1344 | */ | |
1345 | static void handle_access_fault(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa) | |
1346 | { | |
1347 | pmd_t *pmd; | |
1348 | pte_t *pte; | |
1349 | pfn_t pfn; | |
1350 | bool pfn_valid = false; | |
1351 | ||
1352 | trace_kvm_access_fault(fault_ipa); | |
1353 | ||
1354 | spin_lock(&vcpu->kvm->mmu_lock); | |
1355 | ||
1356 | pmd = stage2_get_pmd(vcpu->kvm, NULL, fault_ipa); | |
1357 | if (!pmd || pmd_none(*pmd)) /* Nothing there */ | |
1358 | goto out; | |
1359 | ||
1360 | if (kvm_pmd_huge(*pmd)) { /* THP, HugeTLB */ | |
1361 | *pmd = pmd_mkyoung(*pmd); | |
1362 | pfn = pmd_pfn(*pmd); | |
1363 | pfn_valid = true; | |
1364 | goto out; | |
1365 | } | |
1366 | ||
1367 | pte = pte_offset_kernel(pmd, fault_ipa); | |
1368 | if (pte_none(*pte)) /* Nothing there either */ | |
1369 | goto out; | |
1370 | ||
1371 | *pte = pte_mkyoung(*pte); /* Just a page... */ | |
1372 | pfn = pte_pfn(*pte); | |
1373 | pfn_valid = true; | |
1374 | out: | |
1375 | spin_unlock(&vcpu->kvm->mmu_lock); | |
1376 | if (pfn_valid) | |
1377 | kvm_set_pfn_accessed(pfn); | |
1378 | } | |
1379 | ||
1380 | /** | |
1381 | * kvm_handle_guest_abort - handles all 2nd stage aborts | |
1382 | * @vcpu: the VCPU pointer | |
1383 | * @run: the kvm_run structure | |
1384 | * | |
1385 | * Any abort that gets to the host is almost guaranteed to be caused by a | |
1386 | * missing second stage translation table entry, which can mean that either the | |
1387 | * guest simply needs more memory and we must allocate an appropriate page or it | |
1388 | * can mean that the guest tried to access I/O memory, which is emulated by user | |
1389 | * space. The distinction is based on the IPA causing the fault and whether this | |
1390 | * memory region has been registered as standard RAM by user space. | |
1391 | */ | |
1392 | int kvm_handle_guest_abort(struct kvm_vcpu *vcpu, struct kvm_run *run) | |
1393 | { | |
1394 | unsigned long fault_status; | |
1395 | phys_addr_t fault_ipa; | |
1396 | struct kvm_memory_slot *memslot; | |
1397 | unsigned long hva; | |
1398 | bool is_iabt, write_fault, writable; | |
1399 | gfn_t gfn; | |
1400 | int ret, idx; | |
1401 | ||
1402 | is_iabt = kvm_vcpu_trap_is_iabt(vcpu); | |
1403 | fault_ipa = kvm_vcpu_get_fault_ipa(vcpu); | |
1404 | ||
1405 | trace_kvm_guest_fault(*vcpu_pc(vcpu), kvm_vcpu_get_hsr(vcpu), | |
1406 | kvm_vcpu_get_hfar(vcpu), fault_ipa); | |
1407 | ||
1408 | /* Check the stage-2 fault is trans. fault or write fault */ | |
1409 | fault_status = kvm_vcpu_trap_get_fault_type(vcpu); | |
1410 | if (fault_status != FSC_FAULT && fault_status != FSC_PERM && | |
1411 | fault_status != FSC_ACCESS) { | |
1412 | kvm_err("Unsupported FSC: EC=%#x xFSC=%#lx ESR_EL2=%#lx\n", | |
1413 | kvm_vcpu_trap_get_class(vcpu), | |
1414 | (unsigned long)kvm_vcpu_trap_get_fault(vcpu), | |
1415 | (unsigned long)kvm_vcpu_get_hsr(vcpu)); | |
1416 | return -EFAULT; | |
1417 | } | |
1418 | ||
1419 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
1420 | ||
1421 | gfn = fault_ipa >> PAGE_SHIFT; | |
1422 | memslot = gfn_to_memslot(vcpu->kvm, gfn); | |
1423 | hva = gfn_to_hva_memslot_prot(memslot, gfn, &writable); | |
1424 | write_fault = kvm_is_write_fault(vcpu); | |
1425 | if (kvm_is_error_hva(hva) || (write_fault && !writable)) { | |
1426 | if (is_iabt) { | |
1427 | /* Prefetch Abort on I/O address */ | |
1428 | kvm_inject_pabt(vcpu, kvm_vcpu_get_hfar(vcpu)); | |
1429 | ret = 1; | |
1430 | goto out_unlock; | |
1431 | } | |
1432 | ||
1433 | /* | |
1434 | * The IPA is reported as [MAX:12], so we need to | |
1435 | * complement it with the bottom 12 bits from the | |
1436 | * faulting VA. This is always 12 bits, irrespective | |
1437 | * of the page size. | |
1438 | */ | |
1439 | fault_ipa |= kvm_vcpu_get_hfar(vcpu) & ((1 << 12) - 1); | |
1440 | ret = io_mem_abort(vcpu, run, fault_ipa); | |
1441 | goto out_unlock; | |
1442 | } | |
1443 | ||
1444 | /* Userspace should not be able to register out-of-bounds IPAs */ | |
1445 | VM_BUG_ON(fault_ipa >= KVM_PHYS_SIZE); | |
1446 | ||
1447 | if (fault_status == FSC_ACCESS) { | |
1448 | handle_access_fault(vcpu, fault_ipa); | |
1449 | ret = 1; | |
1450 | goto out_unlock; | |
1451 | } | |
1452 | ||
1453 | ret = user_mem_abort(vcpu, fault_ipa, memslot, hva, fault_status); | |
1454 | if (ret == 0) | |
1455 | ret = 1; | |
1456 | out_unlock: | |
1457 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
1458 | return ret; | |
1459 | } | |
1460 | ||
1461 | static int handle_hva_to_gpa(struct kvm *kvm, | |
1462 | unsigned long start, | |
1463 | unsigned long end, | |
1464 | int (*handler)(struct kvm *kvm, | |
1465 | gpa_t gpa, void *data), | |
1466 | void *data) | |
1467 | { | |
1468 | struct kvm_memslots *slots; | |
1469 | struct kvm_memory_slot *memslot; | |
1470 | int ret = 0; | |
1471 | ||
1472 | slots = kvm_memslots(kvm); | |
1473 | ||
1474 | /* we only care about the pages that the guest sees */ | |
1475 | kvm_for_each_memslot(memslot, slots) { | |
1476 | unsigned long hva_start, hva_end; | |
1477 | gfn_t gfn, gfn_end; | |
1478 | ||
1479 | hva_start = max(start, memslot->userspace_addr); | |
1480 | hva_end = min(end, memslot->userspace_addr + | |
1481 | (memslot->npages << PAGE_SHIFT)); | |
1482 | if (hva_start >= hva_end) | |
1483 | continue; | |
1484 | ||
1485 | /* | |
1486 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
1487 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
1488 | */ | |
1489 | gfn = hva_to_gfn_memslot(hva_start, memslot); | |
1490 | gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot); | |
1491 | ||
1492 | for (; gfn < gfn_end; ++gfn) { | |
1493 | gpa_t gpa = gfn << PAGE_SHIFT; | |
1494 | ret |= handler(kvm, gpa, data); | |
1495 | } | |
1496 | } | |
1497 | ||
1498 | return ret; | |
1499 | } | |
1500 | ||
1501 | static int kvm_unmap_hva_handler(struct kvm *kvm, gpa_t gpa, void *data) | |
1502 | { | |
1503 | unmap_stage2_range(kvm, gpa, PAGE_SIZE); | |
1504 | return 0; | |
1505 | } | |
1506 | ||
1507 | int kvm_unmap_hva(struct kvm *kvm, unsigned long hva) | |
1508 | { | |
1509 | unsigned long end = hva + PAGE_SIZE; | |
1510 | ||
1511 | if (!kvm->arch.pgd) | |
1512 | return 0; | |
1513 | ||
1514 | trace_kvm_unmap_hva(hva); | |
1515 | handle_hva_to_gpa(kvm, hva, end, &kvm_unmap_hva_handler, NULL); | |
1516 | return 0; | |
1517 | } | |
1518 | ||
1519 | int kvm_unmap_hva_range(struct kvm *kvm, | |
1520 | unsigned long start, unsigned long end) | |
1521 | { | |
1522 | if (!kvm->arch.pgd) | |
1523 | return 0; | |
1524 | ||
1525 | trace_kvm_unmap_hva_range(start, end); | |
1526 | handle_hva_to_gpa(kvm, start, end, &kvm_unmap_hva_handler, NULL); | |
1527 | return 0; | |
1528 | } | |
1529 | ||
1530 | static int kvm_set_spte_handler(struct kvm *kvm, gpa_t gpa, void *data) | |
1531 | { | |
1532 | pte_t *pte = (pte_t *)data; | |
1533 | ||
1534 | /* | |
1535 | * We can always call stage2_set_pte with KVM_S2PTE_FLAG_LOGGING_ACTIVE | |
1536 | * flag clear because MMU notifiers will have unmapped a huge PMD before | |
1537 | * calling ->change_pte() (which in turn calls kvm_set_spte_hva()) and | |
1538 | * therefore stage2_set_pte() never needs to clear out a huge PMD | |
1539 | * through this calling path. | |
1540 | */ | |
1541 | stage2_set_pte(kvm, NULL, gpa, pte, 0); | |
1542 | return 0; | |
1543 | } | |
1544 | ||
1545 | ||
1546 | void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte) | |
1547 | { | |
1548 | unsigned long end = hva + PAGE_SIZE; | |
1549 | pte_t stage2_pte; | |
1550 | ||
1551 | if (!kvm->arch.pgd) | |
1552 | return; | |
1553 | ||
1554 | trace_kvm_set_spte_hva(hva); | |
1555 | stage2_pte = pfn_pte(pte_pfn(pte), PAGE_S2); | |
1556 | handle_hva_to_gpa(kvm, hva, end, &kvm_set_spte_handler, &stage2_pte); | |
1557 | } | |
1558 | ||
1559 | static int kvm_age_hva_handler(struct kvm *kvm, gpa_t gpa, void *data) | |
1560 | { | |
1561 | pmd_t *pmd; | |
1562 | pte_t *pte; | |
1563 | ||
1564 | pmd = stage2_get_pmd(kvm, NULL, gpa); | |
1565 | if (!pmd || pmd_none(*pmd)) /* Nothing there */ | |
1566 | return 0; | |
1567 | ||
1568 | if (kvm_pmd_huge(*pmd)) { /* THP, HugeTLB */ | |
1569 | if (pmd_young(*pmd)) { | |
1570 | *pmd = pmd_mkold(*pmd); | |
1571 | return 1; | |
1572 | } | |
1573 | ||
1574 | return 0; | |
1575 | } | |
1576 | ||
1577 | pte = pte_offset_kernel(pmd, gpa); | |
1578 | if (pte_none(*pte)) | |
1579 | return 0; | |
1580 | ||
1581 | if (pte_young(*pte)) { | |
1582 | *pte = pte_mkold(*pte); /* Just a page... */ | |
1583 | return 1; | |
1584 | } | |
1585 | ||
1586 | return 0; | |
1587 | } | |
1588 | ||
1589 | static int kvm_test_age_hva_handler(struct kvm *kvm, gpa_t gpa, void *data) | |
1590 | { | |
1591 | pmd_t *pmd; | |
1592 | pte_t *pte; | |
1593 | ||
1594 | pmd = stage2_get_pmd(kvm, NULL, gpa); | |
1595 | if (!pmd || pmd_none(*pmd)) /* Nothing there */ | |
1596 | return 0; | |
1597 | ||
1598 | if (kvm_pmd_huge(*pmd)) /* THP, HugeTLB */ | |
1599 | return pmd_young(*pmd); | |
1600 | ||
1601 | pte = pte_offset_kernel(pmd, gpa); | |
1602 | if (!pte_none(*pte)) /* Just a page... */ | |
1603 | return pte_young(*pte); | |
1604 | ||
1605 | return 0; | |
1606 | } | |
1607 | ||
1608 | int kvm_age_hva(struct kvm *kvm, unsigned long start, unsigned long end) | |
1609 | { | |
1610 | trace_kvm_age_hva(start, end); | |
1611 | return handle_hva_to_gpa(kvm, start, end, kvm_age_hva_handler, NULL); | |
1612 | } | |
1613 | ||
1614 | int kvm_test_age_hva(struct kvm *kvm, unsigned long hva) | |
1615 | { | |
1616 | trace_kvm_test_age_hva(hva); | |
1617 | return handle_hva_to_gpa(kvm, hva, hva, kvm_test_age_hva_handler, NULL); | |
1618 | } | |
1619 | ||
1620 | void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu) | |
1621 | { | |
1622 | mmu_free_memory_cache(&vcpu->arch.mmu_page_cache); | |
1623 | } | |
1624 | ||
1625 | phys_addr_t kvm_mmu_get_httbr(void) | |
1626 | { | |
1627 | if (__kvm_cpu_uses_extended_idmap()) | |
1628 | return virt_to_phys(merged_hyp_pgd); | |
1629 | else | |
1630 | return virt_to_phys(hyp_pgd); | |
1631 | } | |
1632 | ||
1633 | phys_addr_t kvm_mmu_get_boot_httbr(void) | |
1634 | { | |
1635 | if (__kvm_cpu_uses_extended_idmap()) | |
1636 | return virt_to_phys(merged_hyp_pgd); | |
1637 | else | |
1638 | return virt_to_phys(boot_hyp_pgd); | |
1639 | } | |
1640 | ||
1641 | phys_addr_t kvm_get_idmap_vector(void) | |
1642 | { | |
1643 | return hyp_idmap_vector; | |
1644 | } | |
1645 | ||
1646 | int kvm_mmu_init(void) | |
1647 | { | |
1648 | int err; | |
1649 | ||
1650 | hyp_idmap_start = kvm_virt_to_phys(__hyp_idmap_text_start); | |
1651 | hyp_idmap_end = kvm_virt_to_phys(__hyp_idmap_text_end); | |
1652 | hyp_idmap_vector = kvm_virt_to_phys(__kvm_hyp_init); | |
1653 | ||
1654 | /* | |
1655 | * We rely on the linker script to ensure at build time that the HYP | |
1656 | * init code does not cross a page boundary. | |
1657 | */ | |
1658 | BUG_ON((hyp_idmap_start ^ (hyp_idmap_end - 1)) & PAGE_MASK); | |
1659 | ||
1660 | hyp_pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, hyp_pgd_order); | |
1661 | boot_hyp_pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, hyp_pgd_order); | |
1662 | ||
1663 | if (!hyp_pgd || !boot_hyp_pgd) { | |
1664 | kvm_err("Hyp mode PGD not allocated\n"); | |
1665 | err = -ENOMEM; | |
1666 | goto out; | |
1667 | } | |
1668 | ||
1669 | /* Create the idmap in the boot page tables */ | |
1670 | err = __create_hyp_mappings(boot_hyp_pgd, | |
1671 | hyp_idmap_start, hyp_idmap_end, | |
1672 | __phys_to_pfn(hyp_idmap_start), | |
1673 | PAGE_HYP); | |
1674 | ||
1675 | if (err) { | |
1676 | kvm_err("Failed to idmap %lx-%lx\n", | |
1677 | hyp_idmap_start, hyp_idmap_end); | |
1678 | goto out; | |
1679 | } | |
1680 | ||
1681 | if (__kvm_cpu_uses_extended_idmap()) { | |
1682 | merged_hyp_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO); | |
1683 | if (!merged_hyp_pgd) { | |
1684 | kvm_err("Failed to allocate extra HYP pgd\n"); | |
1685 | goto out; | |
1686 | } | |
1687 | __kvm_extend_hypmap(boot_hyp_pgd, hyp_pgd, merged_hyp_pgd, | |
1688 | hyp_idmap_start); | |
1689 | return 0; | |
1690 | } | |
1691 | ||
1692 | /* Map the very same page at the trampoline VA */ | |
1693 | err = __create_hyp_mappings(boot_hyp_pgd, | |
1694 | TRAMPOLINE_VA, TRAMPOLINE_VA + PAGE_SIZE, | |
1695 | __phys_to_pfn(hyp_idmap_start), | |
1696 | PAGE_HYP); | |
1697 | if (err) { | |
1698 | kvm_err("Failed to map trampoline @%lx into boot HYP pgd\n", | |
1699 | TRAMPOLINE_VA); | |
1700 | goto out; | |
1701 | } | |
1702 | ||
1703 | /* Map the same page again into the runtime page tables */ | |
1704 | err = __create_hyp_mappings(hyp_pgd, | |
1705 | TRAMPOLINE_VA, TRAMPOLINE_VA + PAGE_SIZE, | |
1706 | __phys_to_pfn(hyp_idmap_start), | |
1707 | PAGE_HYP); | |
1708 | if (err) { | |
1709 | kvm_err("Failed to map trampoline @%lx into runtime HYP pgd\n", | |
1710 | TRAMPOLINE_VA); | |
1711 | goto out; | |
1712 | } | |
1713 | ||
1714 | return 0; | |
1715 | out: | |
1716 | free_hyp_pgds(); | |
1717 | return err; | |
1718 | } | |
1719 | ||
1720 | void kvm_arch_commit_memory_region(struct kvm *kvm, | |
1721 | const struct kvm_userspace_memory_region *mem, | |
1722 | const struct kvm_memory_slot *old, | |
1723 | const struct kvm_memory_slot *new, | |
1724 | enum kvm_mr_change change) | |
1725 | { | |
1726 | /* | |
1727 | * At this point memslot has been committed and there is an | |
1728 | * allocated dirty_bitmap[], dirty pages will be be tracked while the | |
1729 | * memory slot is write protected. | |
1730 | */ | |
1731 | if (change != KVM_MR_DELETE && mem->flags & KVM_MEM_LOG_DIRTY_PAGES) | |
1732 | kvm_mmu_wp_memory_region(kvm, mem->slot); | |
1733 | } | |
1734 | ||
1735 | int kvm_arch_prepare_memory_region(struct kvm *kvm, | |
1736 | struct kvm_memory_slot *memslot, | |
1737 | const struct kvm_userspace_memory_region *mem, | |
1738 | enum kvm_mr_change change) | |
1739 | { | |
1740 | hva_t hva = mem->userspace_addr; | |
1741 | hva_t reg_end = hva + mem->memory_size; | |
1742 | bool writable = !(mem->flags & KVM_MEM_READONLY); | |
1743 | int ret = 0; | |
1744 | ||
1745 | if (change != KVM_MR_CREATE && change != KVM_MR_MOVE && | |
1746 | change != KVM_MR_FLAGS_ONLY) | |
1747 | return 0; | |
1748 | ||
1749 | /* | |
1750 | * Prevent userspace from creating a memory region outside of the IPA | |
1751 | * space addressable by the KVM guest IPA space. | |
1752 | */ | |
1753 | if (memslot->base_gfn + memslot->npages >= | |
1754 | (KVM_PHYS_SIZE >> PAGE_SHIFT)) | |
1755 | return -EFAULT; | |
1756 | ||
1757 | /* | |
1758 | * A memory region could potentially cover multiple VMAs, and any holes | |
1759 | * between them, so iterate over all of them to find out if we can map | |
1760 | * any of them right now. | |
1761 | * | |
1762 | * +--------------------------------------------+ | |
1763 | * +---------------+----------------+ +----------------+ | |
1764 | * | : VMA 1 | VMA 2 | | VMA 3 : | | |
1765 | * +---------------+----------------+ +----------------+ | |
1766 | * | memory region | | |
1767 | * +--------------------------------------------+ | |
1768 | */ | |
1769 | do { | |
1770 | struct vm_area_struct *vma = find_vma(current->mm, hva); | |
1771 | hva_t vm_start, vm_end; | |
1772 | ||
1773 | if (!vma || vma->vm_start >= reg_end) | |
1774 | break; | |
1775 | ||
1776 | /* | |
1777 | * Mapping a read-only VMA is only allowed if the | |
1778 | * memory region is configured as read-only. | |
1779 | */ | |
1780 | if (writable && !(vma->vm_flags & VM_WRITE)) { | |
1781 | ret = -EPERM; | |
1782 | break; | |
1783 | } | |
1784 | ||
1785 | /* | |
1786 | * Take the intersection of this VMA with the memory region | |
1787 | */ | |
1788 | vm_start = max(hva, vma->vm_start); | |
1789 | vm_end = min(reg_end, vma->vm_end); | |
1790 | ||
1791 | if (vma->vm_flags & VM_PFNMAP) { | |
1792 | gpa_t gpa = mem->guest_phys_addr + | |
1793 | (vm_start - mem->userspace_addr); | |
1794 | phys_addr_t pa; | |
1795 | ||
1796 | pa = (phys_addr_t)vma->vm_pgoff << PAGE_SHIFT; | |
1797 | pa += vm_start - vma->vm_start; | |
1798 | ||
1799 | /* IO region dirty page logging not allowed */ | |
1800 | if (memslot->flags & KVM_MEM_LOG_DIRTY_PAGES) | |
1801 | return -EINVAL; | |
1802 | ||
1803 | ret = kvm_phys_addr_ioremap(kvm, gpa, pa, | |
1804 | vm_end - vm_start, | |
1805 | writable); | |
1806 | if (ret) | |
1807 | break; | |
1808 | } | |
1809 | hva = vm_end; | |
1810 | } while (hva < reg_end); | |
1811 | ||
1812 | if (change == KVM_MR_FLAGS_ONLY) | |
1813 | return ret; | |
1814 | ||
1815 | spin_lock(&kvm->mmu_lock); | |
1816 | if (ret) | |
1817 | unmap_stage2_range(kvm, mem->guest_phys_addr, mem->memory_size); | |
1818 | else | |
1819 | stage2_flush_memslot(kvm, memslot); | |
1820 | spin_unlock(&kvm->mmu_lock); | |
1821 | return ret; | |
1822 | } | |
1823 | ||
1824 | void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, | |
1825 | struct kvm_memory_slot *dont) | |
1826 | { | |
1827 | } | |
1828 | ||
1829 | int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot, | |
1830 | unsigned long npages) | |
1831 | { | |
1832 | /* | |
1833 | * Readonly memslots are not incoherent with the caches by definition, | |
1834 | * but in practice, they are used mostly to emulate ROMs or NOR flashes | |
1835 | * that the guest may consider devices and hence map as uncached. | |
1836 | * To prevent incoherency issues in these cases, tag all readonly | |
1837 | * regions as incoherent. | |
1838 | */ | |
1839 | if (slot->flags & KVM_MEM_READONLY) | |
1840 | slot->flags |= KVM_MEMSLOT_INCOHERENT; | |
1841 | return 0; | |
1842 | } | |
1843 | ||
1844 | void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots) | |
1845 | { | |
1846 | } | |
1847 | ||
1848 | void kvm_arch_flush_shadow_all(struct kvm *kvm) | |
1849 | { | |
1850 | } | |
1851 | ||
1852 | void kvm_arch_flush_shadow_memslot(struct kvm *kvm, | |
1853 | struct kvm_memory_slot *slot) | |
1854 | { | |
1855 | gpa_t gpa = slot->base_gfn << PAGE_SHIFT; | |
1856 | phys_addr_t size = slot->npages << PAGE_SHIFT; | |
1857 | ||
1858 | spin_lock(&kvm->mmu_lock); | |
1859 | unmap_stage2_range(kvm, gpa, size); | |
1860 | spin_unlock(&kvm->mmu_lock); | |
1861 | } | |
1862 | ||
1863 | /* | |
1864 | * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized). | |
1865 | * | |
1866 | * Main problems: | |
1867 | * - S/W ops are local to a CPU (not broadcast) | |
1868 | * - We have line migration behind our back (speculation) | |
1869 | * - System caches don't support S/W at all (damn!) | |
1870 | * | |
1871 | * In the face of the above, the best we can do is to try and convert | |
1872 | * S/W ops to VA ops. Because the guest is not allowed to infer the | |
1873 | * S/W to PA mapping, it can only use S/W to nuke the whole cache, | |
1874 | * which is a rather good thing for us. | |
1875 | * | |
1876 | * Also, it is only used when turning caches on/off ("The expected | |
1877 | * usage of the cache maintenance instructions that operate by set/way | |
1878 | * is associated with the cache maintenance instructions associated | |
1879 | * with the powerdown and powerup of caches, if this is required by | |
1880 | * the implementation."). | |
1881 | * | |
1882 | * We use the following policy: | |
1883 | * | |
1884 | * - If we trap a S/W operation, we enable VM trapping to detect | |
1885 | * caches being turned on/off, and do a full clean. | |
1886 | * | |
1887 | * - We flush the caches on both caches being turned on and off. | |
1888 | * | |
1889 | * - Once the caches are enabled, we stop trapping VM ops. | |
1890 | */ | |
1891 | void kvm_set_way_flush(struct kvm_vcpu *vcpu) | |
1892 | { | |
1893 | unsigned long hcr = vcpu_get_hcr(vcpu); | |
1894 | ||
1895 | /* | |
1896 | * If this is the first time we do a S/W operation | |
1897 | * (i.e. HCR_TVM not set) flush the whole memory, and set the | |
1898 | * VM trapping. | |
1899 | * | |
1900 | * Otherwise, rely on the VM trapping to wait for the MMU + | |
1901 | * Caches to be turned off. At that point, we'll be able to | |
1902 | * clean the caches again. | |
1903 | */ | |
1904 | if (!(hcr & HCR_TVM)) { | |
1905 | trace_kvm_set_way_flush(*vcpu_pc(vcpu), | |
1906 | vcpu_has_cache_enabled(vcpu)); | |
1907 | stage2_flush_vm(vcpu->kvm); | |
1908 | vcpu_set_hcr(vcpu, hcr | HCR_TVM); | |
1909 | } | |
1910 | } | |
1911 | ||
1912 | void kvm_toggle_cache(struct kvm_vcpu *vcpu, bool was_enabled) | |
1913 | { | |
1914 | bool now_enabled = vcpu_has_cache_enabled(vcpu); | |
1915 | ||
1916 | /* | |
1917 | * If switching the MMU+caches on, need to invalidate the caches. | |
1918 | * If switching it off, need to clean the caches. | |
1919 | * Clean + invalidate does the trick always. | |
1920 | */ | |
1921 | if (now_enabled != was_enabled) | |
1922 | stage2_flush_vm(vcpu->kvm); | |
1923 | ||
1924 | /* Caches are now on, stop trapping VM ops (until a S/W op) */ | |
1925 | if (now_enabled) | |
1926 | vcpu_set_hcr(vcpu, vcpu_get_hcr(vcpu) & ~HCR_TVM); | |
1927 | ||
1928 | trace_kvm_toggle_cache(*vcpu_pc(vcpu), was_enabled, now_enabled); | |
1929 | } |