]>
Commit | Line | Data |
---|---|---|
749cf76c CD |
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 | */ | |
342cd0ab CD |
18 | |
19 | #include <linux/mman.h> | |
20 | #include <linux/kvm_host.h> | |
21 | #include <linux/io.h> | |
ad361f09 | 22 | #include <linux/hugetlb.h> |
45e96ea6 | 23 | #include <trace/events/kvm.h> |
342cd0ab | 24 | #include <asm/pgalloc.h> |
94f8e641 | 25 | #include <asm/cacheflush.h> |
342cd0ab CD |
26 | #include <asm/kvm_arm.h> |
27 | #include <asm/kvm_mmu.h> | |
45e96ea6 | 28 | #include <asm/kvm_mmio.h> |
d5d8184d | 29 | #include <asm/kvm_asm.h> |
94f8e641 | 30 | #include <asm/kvm_emulate.h> |
d5d8184d CD |
31 | |
32 | #include "trace.h" | |
342cd0ab CD |
33 | |
34 | extern char __hyp_idmap_text_start[], __hyp_idmap_text_end[]; | |
35 | ||
5a677ce0 | 36 | static pgd_t *boot_hyp_pgd; |
2fb41059 | 37 | static pgd_t *hyp_pgd; |
e4c5a685 | 38 | static pgd_t *merged_hyp_pgd; |
342cd0ab CD |
39 | static DEFINE_MUTEX(kvm_hyp_pgd_mutex); |
40 | ||
5a677ce0 MZ |
41 | static unsigned long hyp_idmap_start; |
42 | static unsigned long hyp_idmap_end; | |
43 | static phys_addr_t hyp_idmap_vector; | |
44 | ||
38f791a4 | 45 | #define hyp_pgd_order get_order(PTRS_PER_PGD * sizeof(pgd_t)) |
5d4e08c4 | 46 | |
9b5fdb97 | 47 | #define kvm_pmd_huge(_x) (pmd_huge(_x) || pmd_trans_huge(_x)) |
c6473555 | 48 | #define kvm_pud_huge(_x) pud_huge(_x) |
ad361f09 | 49 | |
15a49a44 MS |
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 | { | |
15a49a44 | 55 | return memslot->dirty_bitmap && !(memslot->flags & KVM_MEM_READONLY); |
7276030a MS |
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); | |
15a49a44 | 67 | } |
ad361f09 | 68 | |
48762767 | 69 | static void kvm_tlb_flush_vmid_ipa(struct kvm *kvm, phys_addr_t ipa) |
d5d8184d | 70 | { |
d4cb9df5 MZ |
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); | |
d5d8184d CD |
79 | } |
80 | ||
363ef89f MZ |
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 | ||
15a49a44 MS |
101 | /** |
102 | * stage2_dissolve_pmd() - clear and flush huge PMD entry | |
103 | * @kvm: pointer to kvm structure. | |
104 | * @addr: IPA | |
105 | * @pmd: pmd pointer for IPA | |
106 | * | |
107 | * Function clears a PMD entry, flushes addr 1st and 2nd stage TLBs. Marks all | |
108 | * pages in the range dirty. | |
109 | */ | |
110 | static void stage2_dissolve_pmd(struct kvm *kvm, phys_addr_t addr, pmd_t *pmd) | |
111 | { | |
112 | if (!kvm_pmd_huge(*pmd)) | |
113 | return; | |
114 | ||
115 | pmd_clear(pmd); | |
116 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
117 | put_page(virt_to_page(pmd)); | |
118 | } | |
119 | ||
d5d8184d CD |
120 | static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache, |
121 | int min, int max) | |
122 | { | |
123 | void *page; | |
124 | ||
125 | BUG_ON(max > KVM_NR_MEM_OBJS); | |
126 | if (cache->nobjs >= min) | |
127 | return 0; | |
128 | while (cache->nobjs < max) { | |
129 | page = (void *)__get_free_page(PGALLOC_GFP); | |
130 | if (!page) | |
131 | return -ENOMEM; | |
132 | cache->objects[cache->nobjs++] = page; | |
133 | } | |
134 | return 0; | |
135 | } | |
136 | ||
137 | static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
138 | { | |
139 | while (mc->nobjs) | |
140 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
141 | } | |
142 | ||
143 | static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
144 | { | |
145 | void *p; | |
146 | ||
147 | BUG_ON(!mc || !mc->nobjs); | |
148 | p = mc->objects[--mc->nobjs]; | |
149 | return p; | |
150 | } | |
151 | ||
4f853a71 | 152 | static void clear_pgd_entry(struct kvm *kvm, pgd_t *pgd, phys_addr_t addr) |
979acd5e | 153 | { |
4f853a71 CD |
154 | pud_t *pud_table __maybe_unused = pud_offset(pgd, 0); |
155 | pgd_clear(pgd); | |
156 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
157 | pud_free(NULL, pud_table); | |
158 | put_page(virt_to_page(pgd)); | |
979acd5e MZ |
159 | } |
160 | ||
d4cb9df5 | 161 | static void clear_pud_entry(struct kvm *kvm, pud_t *pud, phys_addr_t addr) |
342cd0ab | 162 | { |
4f853a71 CD |
163 | pmd_t *pmd_table = pmd_offset(pud, 0); |
164 | VM_BUG_ON(pud_huge(*pud)); | |
165 | pud_clear(pud); | |
166 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
167 | pmd_free(NULL, pmd_table); | |
4f728276 MZ |
168 | put_page(virt_to_page(pud)); |
169 | } | |
342cd0ab | 170 | |
d4cb9df5 | 171 | static void clear_pmd_entry(struct kvm *kvm, pmd_t *pmd, phys_addr_t addr) |
4f728276 | 172 | { |
4f853a71 CD |
173 | pte_t *pte_table = pte_offset_kernel(pmd, 0); |
174 | VM_BUG_ON(kvm_pmd_huge(*pmd)); | |
175 | pmd_clear(pmd); | |
176 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
177 | pte_free_kernel(NULL, pte_table); | |
4f728276 MZ |
178 | put_page(virt_to_page(pmd)); |
179 | } | |
180 | ||
363ef89f MZ |
181 | /* |
182 | * Unmapping vs dcache management: | |
183 | * | |
184 | * If a guest maps certain memory pages as uncached, all writes will | |
185 | * bypass the data cache and go directly to RAM. However, the CPUs | |
186 | * can still speculate reads (not writes) and fill cache lines with | |
187 | * data. | |
188 | * | |
189 | * Those cache lines will be *clean* cache lines though, so a | |
190 | * clean+invalidate operation is equivalent to an invalidate | |
191 | * operation, because no cache lines are marked dirty. | |
192 | * | |
193 | * Those clean cache lines could be filled prior to an uncached write | |
194 | * by the guest, and the cache coherent IO subsystem would therefore | |
195 | * end up writing old data to disk. | |
196 | * | |
197 | * This is why right after unmapping a page/section and invalidating | |
198 | * the corresponding TLBs, we call kvm_flush_dcache_p*() to make sure | |
199 | * the IO subsystem will never hit in the cache. | |
200 | */ | |
4f853a71 CD |
201 | static void unmap_ptes(struct kvm *kvm, pmd_t *pmd, |
202 | phys_addr_t addr, phys_addr_t end) | |
4f728276 | 203 | { |
4f853a71 CD |
204 | phys_addr_t start_addr = addr; |
205 | pte_t *pte, *start_pte; | |
206 | ||
207 | start_pte = pte = pte_offset_kernel(pmd, addr); | |
208 | do { | |
209 | if (!pte_none(*pte)) { | |
363ef89f MZ |
210 | pte_t old_pte = *pte; |
211 | ||
4f853a71 | 212 | kvm_set_pte(pte, __pte(0)); |
4f853a71 | 213 | kvm_tlb_flush_vmid_ipa(kvm, addr); |
363ef89f MZ |
214 | |
215 | /* No need to invalidate the cache for device mappings */ | |
216 | if ((pte_val(old_pte) & PAGE_S2_DEVICE) != PAGE_S2_DEVICE) | |
217 | kvm_flush_dcache_pte(old_pte); | |
218 | ||
219 | put_page(virt_to_page(pte)); | |
4f853a71 CD |
220 | } |
221 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
222 | ||
38f791a4 | 223 | if (kvm_pte_table_empty(kvm, start_pte)) |
4f853a71 | 224 | clear_pmd_entry(kvm, pmd, start_addr); |
342cd0ab CD |
225 | } |
226 | ||
4f853a71 CD |
227 | static void unmap_pmds(struct kvm *kvm, pud_t *pud, |
228 | phys_addr_t addr, phys_addr_t end) | |
000d3996 | 229 | { |
4f853a71 CD |
230 | phys_addr_t next, start_addr = addr; |
231 | pmd_t *pmd, *start_pmd; | |
000d3996 | 232 | |
4f853a71 CD |
233 | start_pmd = pmd = pmd_offset(pud, addr); |
234 | do { | |
235 | next = kvm_pmd_addr_end(addr, end); | |
236 | if (!pmd_none(*pmd)) { | |
237 | if (kvm_pmd_huge(*pmd)) { | |
363ef89f MZ |
238 | pmd_t old_pmd = *pmd; |
239 | ||
4f853a71 CD |
240 | pmd_clear(pmd); |
241 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
363ef89f MZ |
242 | |
243 | kvm_flush_dcache_pmd(old_pmd); | |
244 | ||
4f853a71 CD |
245 | put_page(virt_to_page(pmd)); |
246 | } else { | |
247 | unmap_ptes(kvm, pmd, addr, next); | |
248 | } | |
ad361f09 | 249 | } |
4f853a71 | 250 | } while (pmd++, addr = next, addr != end); |
ad361f09 | 251 | |
38f791a4 | 252 | if (kvm_pmd_table_empty(kvm, start_pmd)) |
4f853a71 CD |
253 | clear_pud_entry(kvm, pud, start_addr); |
254 | } | |
000d3996 | 255 | |
4f853a71 CD |
256 | static void unmap_puds(struct kvm *kvm, pgd_t *pgd, |
257 | phys_addr_t addr, phys_addr_t end) | |
258 | { | |
259 | phys_addr_t next, start_addr = addr; | |
260 | pud_t *pud, *start_pud; | |
4f728276 | 261 | |
4f853a71 CD |
262 | start_pud = pud = pud_offset(pgd, addr); |
263 | do { | |
264 | next = kvm_pud_addr_end(addr, end); | |
265 | if (!pud_none(*pud)) { | |
266 | if (pud_huge(*pud)) { | |
363ef89f MZ |
267 | pud_t old_pud = *pud; |
268 | ||
4f853a71 CD |
269 | pud_clear(pud); |
270 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
363ef89f MZ |
271 | |
272 | kvm_flush_dcache_pud(old_pud); | |
273 | ||
4f853a71 CD |
274 | put_page(virt_to_page(pud)); |
275 | } else { | |
276 | unmap_pmds(kvm, pud, addr, next); | |
4f728276 MZ |
277 | } |
278 | } | |
4f853a71 | 279 | } while (pud++, addr = next, addr != end); |
4f728276 | 280 | |
38f791a4 | 281 | if (kvm_pud_table_empty(kvm, start_pud)) |
4f853a71 CD |
282 | clear_pgd_entry(kvm, pgd, start_addr); |
283 | } | |
284 | ||
285 | ||
286 | static void unmap_range(struct kvm *kvm, pgd_t *pgdp, | |
287 | phys_addr_t start, u64 size) | |
288 | { | |
289 | pgd_t *pgd; | |
290 | phys_addr_t addr = start, end = start + size; | |
291 | phys_addr_t next; | |
292 | ||
04b8dc85 | 293 | pgd = pgdp + kvm_pgd_index(addr); |
4f853a71 CD |
294 | do { |
295 | next = kvm_pgd_addr_end(addr, end); | |
7cbb87d6 MR |
296 | if (!pgd_none(*pgd)) |
297 | unmap_puds(kvm, pgd, addr, next); | |
4f853a71 | 298 | } while (pgd++, addr = next, addr != end); |
000d3996 MZ |
299 | } |
300 | ||
9d218a1f MZ |
301 | static void stage2_flush_ptes(struct kvm *kvm, pmd_t *pmd, |
302 | phys_addr_t addr, phys_addr_t end) | |
303 | { | |
304 | pte_t *pte; | |
305 | ||
306 | pte = pte_offset_kernel(pmd, addr); | |
307 | do { | |
363ef89f MZ |
308 | if (!pte_none(*pte) && |
309 | (pte_val(*pte) & PAGE_S2_DEVICE) != PAGE_S2_DEVICE) | |
310 | kvm_flush_dcache_pte(*pte); | |
9d218a1f MZ |
311 | } while (pte++, addr += PAGE_SIZE, addr != end); |
312 | } | |
313 | ||
314 | static void stage2_flush_pmds(struct kvm *kvm, pud_t *pud, | |
315 | phys_addr_t addr, phys_addr_t end) | |
316 | { | |
317 | pmd_t *pmd; | |
318 | phys_addr_t next; | |
319 | ||
320 | pmd = pmd_offset(pud, addr); | |
321 | do { | |
322 | next = kvm_pmd_addr_end(addr, end); | |
323 | if (!pmd_none(*pmd)) { | |
363ef89f MZ |
324 | if (kvm_pmd_huge(*pmd)) |
325 | kvm_flush_dcache_pmd(*pmd); | |
326 | else | |
9d218a1f | 327 | stage2_flush_ptes(kvm, pmd, addr, next); |
9d218a1f MZ |
328 | } |
329 | } while (pmd++, addr = next, addr != end); | |
330 | } | |
331 | ||
332 | static void stage2_flush_puds(struct kvm *kvm, pgd_t *pgd, | |
333 | phys_addr_t addr, phys_addr_t end) | |
334 | { | |
335 | pud_t *pud; | |
336 | phys_addr_t next; | |
337 | ||
338 | pud = pud_offset(pgd, addr); | |
339 | do { | |
340 | next = kvm_pud_addr_end(addr, end); | |
341 | if (!pud_none(*pud)) { | |
363ef89f MZ |
342 | if (pud_huge(*pud)) |
343 | kvm_flush_dcache_pud(*pud); | |
344 | else | |
9d218a1f | 345 | stage2_flush_pmds(kvm, pud, addr, next); |
9d218a1f MZ |
346 | } |
347 | } while (pud++, addr = next, addr != end); | |
348 | } | |
349 | ||
350 | static void stage2_flush_memslot(struct kvm *kvm, | |
351 | struct kvm_memory_slot *memslot) | |
352 | { | |
353 | phys_addr_t addr = memslot->base_gfn << PAGE_SHIFT; | |
354 | phys_addr_t end = addr + PAGE_SIZE * memslot->npages; | |
355 | phys_addr_t next; | |
356 | pgd_t *pgd; | |
357 | ||
04b8dc85 | 358 | pgd = kvm->arch.pgd + kvm_pgd_index(addr); |
9d218a1f MZ |
359 | do { |
360 | next = kvm_pgd_addr_end(addr, end); | |
361 | stage2_flush_puds(kvm, pgd, addr, next); | |
362 | } while (pgd++, addr = next, addr != end); | |
363 | } | |
364 | ||
365 | /** | |
366 | * stage2_flush_vm - Invalidate cache for pages mapped in stage 2 | |
367 | * @kvm: The struct kvm pointer | |
368 | * | |
369 | * Go through the stage 2 page tables and invalidate any cache lines | |
370 | * backing memory already mapped to the VM. | |
371 | */ | |
3c1e7165 | 372 | static void stage2_flush_vm(struct kvm *kvm) |
9d218a1f MZ |
373 | { |
374 | struct kvm_memslots *slots; | |
375 | struct kvm_memory_slot *memslot; | |
376 | int idx; | |
377 | ||
378 | idx = srcu_read_lock(&kvm->srcu); | |
379 | spin_lock(&kvm->mmu_lock); | |
380 | ||
381 | slots = kvm_memslots(kvm); | |
382 | kvm_for_each_memslot(memslot, slots) | |
383 | stage2_flush_memslot(kvm, memslot); | |
384 | ||
385 | spin_unlock(&kvm->mmu_lock); | |
386 | srcu_read_unlock(&kvm->srcu, idx); | |
387 | } | |
388 | ||
d157f4a5 MZ |
389 | /** |
390 | * free_boot_hyp_pgd - free HYP boot page tables | |
391 | * | |
392 | * Free the HYP boot page tables. The bounce page is also freed. | |
393 | */ | |
394 | void free_boot_hyp_pgd(void) | |
395 | { | |
396 | mutex_lock(&kvm_hyp_pgd_mutex); | |
397 | ||
398 | if (boot_hyp_pgd) { | |
d4cb9df5 MZ |
399 | unmap_range(NULL, boot_hyp_pgd, hyp_idmap_start, PAGE_SIZE); |
400 | unmap_range(NULL, boot_hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE); | |
38f791a4 | 401 | free_pages((unsigned long)boot_hyp_pgd, hyp_pgd_order); |
d157f4a5 MZ |
402 | boot_hyp_pgd = NULL; |
403 | } | |
404 | ||
405 | if (hyp_pgd) | |
d4cb9df5 | 406 | unmap_range(NULL, hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE); |
d157f4a5 | 407 | |
d157f4a5 MZ |
408 | mutex_unlock(&kvm_hyp_pgd_mutex); |
409 | } | |
410 | ||
342cd0ab | 411 | /** |
4f728276 | 412 | * free_hyp_pgds - free Hyp-mode page tables |
342cd0ab | 413 | * |
5a677ce0 MZ |
414 | * Assumes hyp_pgd is a page table used strictly in Hyp-mode and |
415 | * therefore contains either mappings in the kernel memory area (above | |
416 | * PAGE_OFFSET), or device mappings in the vmalloc range (from | |
417 | * VMALLOC_START to VMALLOC_END). | |
418 | * | |
419 | * boot_hyp_pgd should only map two pages for the init code. | |
342cd0ab | 420 | */ |
4f728276 | 421 | void free_hyp_pgds(void) |
342cd0ab | 422 | { |
342cd0ab CD |
423 | unsigned long addr; |
424 | ||
d157f4a5 | 425 | free_boot_hyp_pgd(); |
4f728276 | 426 | |
d157f4a5 | 427 | mutex_lock(&kvm_hyp_pgd_mutex); |
5a677ce0 | 428 | |
4f728276 MZ |
429 | if (hyp_pgd) { |
430 | for (addr = PAGE_OFFSET; virt_addr_valid(addr); addr += PGDIR_SIZE) | |
d4cb9df5 | 431 | unmap_range(NULL, hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE); |
4f728276 | 432 | for (addr = VMALLOC_START; is_vmalloc_addr((void*)addr); addr += PGDIR_SIZE) |
d4cb9df5 MZ |
433 | unmap_range(NULL, hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE); |
434 | ||
38f791a4 | 435 | free_pages((unsigned long)hyp_pgd, hyp_pgd_order); |
d157f4a5 | 436 | hyp_pgd = NULL; |
4f728276 | 437 | } |
e4c5a685 AB |
438 | if (merged_hyp_pgd) { |
439 | clear_page(merged_hyp_pgd); | |
440 | free_page((unsigned long)merged_hyp_pgd); | |
441 | merged_hyp_pgd = NULL; | |
442 | } | |
4f728276 | 443 | |
342cd0ab CD |
444 | mutex_unlock(&kvm_hyp_pgd_mutex); |
445 | } | |
446 | ||
447 | static void create_hyp_pte_mappings(pmd_t *pmd, unsigned long start, | |
6060df84 MZ |
448 | unsigned long end, unsigned long pfn, |
449 | pgprot_t prot) | |
342cd0ab CD |
450 | { |
451 | pte_t *pte; | |
452 | unsigned long addr; | |
342cd0ab | 453 | |
3562c76d MZ |
454 | addr = start; |
455 | do { | |
6060df84 MZ |
456 | pte = pte_offset_kernel(pmd, addr); |
457 | kvm_set_pte(pte, pfn_pte(pfn, prot)); | |
4f728276 | 458 | get_page(virt_to_page(pte)); |
5a677ce0 | 459 | kvm_flush_dcache_to_poc(pte, sizeof(*pte)); |
6060df84 | 460 | pfn++; |
3562c76d | 461 | } while (addr += PAGE_SIZE, addr != end); |
342cd0ab CD |
462 | } |
463 | ||
464 | static int create_hyp_pmd_mappings(pud_t *pud, unsigned long start, | |
6060df84 MZ |
465 | unsigned long end, unsigned long pfn, |
466 | pgprot_t prot) | |
342cd0ab CD |
467 | { |
468 | pmd_t *pmd; | |
469 | pte_t *pte; | |
470 | unsigned long addr, next; | |
471 | ||
3562c76d MZ |
472 | addr = start; |
473 | do { | |
6060df84 | 474 | pmd = pmd_offset(pud, addr); |
342cd0ab CD |
475 | |
476 | BUG_ON(pmd_sect(*pmd)); | |
477 | ||
478 | if (pmd_none(*pmd)) { | |
6060df84 | 479 | pte = pte_alloc_one_kernel(NULL, addr); |
342cd0ab CD |
480 | if (!pte) { |
481 | kvm_err("Cannot allocate Hyp pte\n"); | |
482 | return -ENOMEM; | |
483 | } | |
484 | pmd_populate_kernel(NULL, pmd, pte); | |
4f728276 | 485 | get_page(virt_to_page(pmd)); |
5a677ce0 | 486 | kvm_flush_dcache_to_poc(pmd, sizeof(*pmd)); |
342cd0ab CD |
487 | } |
488 | ||
489 | next = pmd_addr_end(addr, end); | |
490 | ||
6060df84 MZ |
491 | create_hyp_pte_mappings(pmd, addr, next, pfn, prot); |
492 | pfn += (next - addr) >> PAGE_SHIFT; | |
3562c76d | 493 | } while (addr = next, addr != end); |
342cd0ab CD |
494 | |
495 | return 0; | |
496 | } | |
497 | ||
38f791a4 CD |
498 | static int create_hyp_pud_mappings(pgd_t *pgd, unsigned long start, |
499 | unsigned long end, unsigned long pfn, | |
500 | pgprot_t prot) | |
501 | { | |
502 | pud_t *pud; | |
503 | pmd_t *pmd; | |
504 | unsigned long addr, next; | |
505 | int ret; | |
506 | ||
507 | addr = start; | |
508 | do { | |
509 | pud = pud_offset(pgd, addr); | |
510 | ||
511 | if (pud_none_or_clear_bad(pud)) { | |
512 | pmd = pmd_alloc_one(NULL, addr); | |
513 | if (!pmd) { | |
514 | kvm_err("Cannot allocate Hyp pmd\n"); | |
515 | return -ENOMEM; | |
516 | } | |
517 | pud_populate(NULL, pud, pmd); | |
518 | get_page(virt_to_page(pud)); | |
519 | kvm_flush_dcache_to_poc(pud, sizeof(*pud)); | |
520 | } | |
521 | ||
522 | next = pud_addr_end(addr, end); | |
523 | ret = create_hyp_pmd_mappings(pud, addr, next, pfn, prot); | |
524 | if (ret) | |
525 | return ret; | |
526 | pfn += (next - addr) >> PAGE_SHIFT; | |
527 | } while (addr = next, addr != end); | |
528 | ||
529 | return 0; | |
530 | } | |
531 | ||
6060df84 MZ |
532 | static int __create_hyp_mappings(pgd_t *pgdp, |
533 | unsigned long start, unsigned long end, | |
534 | unsigned long pfn, pgprot_t prot) | |
342cd0ab | 535 | { |
342cd0ab CD |
536 | pgd_t *pgd; |
537 | pud_t *pud; | |
342cd0ab CD |
538 | unsigned long addr, next; |
539 | int err = 0; | |
540 | ||
342cd0ab | 541 | mutex_lock(&kvm_hyp_pgd_mutex); |
3562c76d MZ |
542 | addr = start & PAGE_MASK; |
543 | end = PAGE_ALIGN(end); | |
544 | do { | |
6060df84 | 545 | pgd = pgdp + pgd_index(addr); |
342cd0ab | 546 | |
38f791a4 CD |
547 | if (pgd_none(*pgd)) { |
548 | pud = pud_alloc_one(NULL, addr); | |
549 | if (!pud) { | |
550 | kvm_err("Cannot allocate Hyp pud\n"); | |
342cd0ab CD |
551 | err = -ENOMEM; |
552 | goto out; | |
553 | } | |
38f791a4 CD |
554 | pgd_populate(NULL, pgd, pud); |
555 | get_page(virt_to_page(pgd)); | |
556 | kvm_flush_dcache_to_poc(pgd, sizeof(*pgd)); | |
342cd0ab CD |
557 | } |
558 | ||
559 | next = pgd_addr_end(addr, end); | |
38f791a4 | 560 | err = create_hyp_pud_mappings(pgd, addr, next, pfn, prot); |
342cd0ab CD |
561 | if (err) |
562 | goto out; | |
6060df84 | 563 | pfn += (next - addr) >> PAGE_SHIFT; |
3562c76d | 564 | } while (addr = next, addr != end); |
342cd0ab CD |
565 | out: |
566 | mutex_unlock(&kvm_hyp_pgd_mutex); | |
567 | return err; | |
568 | } | |
569 | ||
40c2729b CD |
570 | static phys_addr_t kvm_kaddr_to_phys(void *kaddr) |
571 | { | |
572 | if (!is_vmalloc_addr(kaddr)) { | |
573 | BUG_ON(!virt_addr_valid(kaddr)); | |
574 | return __pa(kaddr); | |
575 | } else { | |
576 | return page_to_phys(vmalloc_to_page(kaddr)) + | |
577 | offset_in_page(kaddr); | |
578 | } | |
579 | } | |
580 | ||
342cd0ab | 581 | /** |
06e8c3b0 | 582 | * create_hyp_mappings - duplicate a kernel virtual address range in Hyp mode |
342cd0ab CD |
583 | * @from: The virtual kernel start address of the range |
584 | * @to: The virtual kernel end address of the range (exclusive) | |
585 | * | |
06e8c3b0 MZ |
586 | * The same virtual address as the kernel virtual address is also used |
587 | * in Hyp-mode mapping (modulo HYP_PAGE_OFFSET) to the same underlying | |
588 | * physical pages. | |
342cd0ab CD |
589 | */ |
590 | int create_hyp_mappings(void *from, void *to) | |
591 | { | |
40c2729b CD |
592 | phys_addr_t phys_addr; |
593 | unsigned long virt_addr; | |
6060df84 MZ |
594 | unsigned long start = KERN_TO_HYP((unsigned long)from); |
595 | unsigned long end = KERN_TO_HYP((unsigned long)to); | |
596 | ||
40c2729b CD |
597 | start = start & PAGE_MASK; |
598 | end = PAGE_ALIGN(end); | |
6060df84 | 599 | |
40c2729b CD |
600 | for (virt_addr = start; virt_addr < end; virt_addr += PAGE_SIZE) { |
601 | int err; | |
6060df84 | 602 | |
40c2729b CD |
603 | phys_addr = kvm_kaddr_to_phys(from + virt_addr - start); |
604 | err = __create_hyp_mappings(hyp_pgd, virt_addr, | |
605 | virt_addr + PAGE_SIZE, | |
606 | __phys_to_pfn(phys_addr), | |
607 | PAGE_HYP); | |
608 | if (err) | |
609 | return err; | |
610 | } | |
611 | ||
612 | return 0; | |
342cd0ab CD |
613 | } |
614 | ||
615 | /** | |
06e8c3b0 MZ |
616 | * create_hyp_io_mappings - duplicate a kernel IO mapping into Hyp mode |
617 | * @from: The kernel start VA of the range | |
618 | * @to: The kernel end VA of the range (exclusive) | |
6060df84 | 619 | * @phys_addr: The physical start address which gets mapped |
06e8c3b0 MZ |
620 | * |
621 | * The resulting HYP VA is the same as the kernel VA, modulo | |
622 | * HYP_PAGE_OFFSET. | |
342cd0ab | 623 | */ |
6060df84 | 624 | int create_hyp_io_mappings(void *from, void *to, phys_addr_t phys_addr) |
342cd0ab | 625 | { |
6060df84 MZ |
626 | unsigned long start = KERN_TO_HYP((unsigned long)from); |
627 | unsigned long end = KERN_TO_HYP((unsigned long)to); | |
628 | ||
629 | /* Check for a valid kernel IO mapping */ | |
630 | if (!is_vmalloc_addr(from) || !is_vmalloc_addr(to - 1)) | |
631 | return -EINVAL; | |
632 | ||
633 | return __create_hyp_mappings(hyp_pgd, start, end, | |
634 | __phys_to_pfn(phys_addr), PAGE_HYP_DEVICE); | |
342cd0ab CD |
635 | } |
636 | ||
a987370f MZ |
637 | /* Free the HW pgd, one page at a time */ |
638 | static void kvm_free_hwpgd(void *hwpgd) | |
639 | { | |
640 | free_pages_exact(hwpgd, kvm_get_hwpgd_size()); | |
641 | } | |
642 | ||
643 | /* Allocate the HW PGD, making sure that each page gets its own refcount */ | |
644 | static void *kvm_alloc_hwpgd(void) | |
645 | { | |
646 | unsigned int size = kvm_get_hwpgd_size(); | |
647 | ||
648 | return alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO); | |
649 | } | |
650 | ||
d5d8184d CD |
651 | /** |
652 | * kvm_alloc_stage2_pgd - allocate level-1 table for stage-2 translation. | |
653 | * @kvm: The KVM struct pointer for the VM. | |
654 | * | |
655 | * Allocates the 1st level table only of size defined by S2_PGD_ORDER (can | |
656 | * support either full 40-bit input addresses or limited to 32-bit input | |
657 | * addresses). Clears the allocated pages. | |
658 | * | |
659 | * Note we don't need locking here as this is only called when the VM is | |
660 | * created, which can only be done once. | |
661 | */ | |
662 | int kvm_alloc_stage2_pgd(struct kvm *kvm) | |
663 | { | |
664 | pgd_t *pgd; | |
a987370f | 665 | void *hwpgd; |
d5d8184d CD |
666 | |
667 | if (kvm->arch.pgd != NULL) { | |
668 | kvm_err("kvm_arch already initialized?\n"); | |
669 | return -EINVAL; | |
670 | } | |
671 | ||
a987370f MZ |
672 | hwpgd = kvm_alloc_hwpgd(); |
673 | if (!hwpgd) | |
674 | return -ENOMEM; | |
675 | ||
676 | /* When the kernel uses more levels of page tables than the | |
677 | * guest, we allocate a fake PGD and pre-populate it to point | |
678 | * to the next-level page table, which will be the real | |
679 | * initial page table pointed to by the VTTBR. | |
680 | * | |
681 | * When KVM_PREALLOC_LEVEL==2, we allocate a single page for | |
682 | * the PMD and the kernel will use folded pud. | |
683 | * When KVM_PREALLOC_LEVEL==1, we allocate 2 consecutive PUD | |
684 | * pages. | |
685 | */ | |
38f791a4 | 686 | if (KVM_PREALLOC_LEVEL > 0) { |
a987370f MZ |
687 | int i; |
688 | ||
38f791a4 CD |
689 | /* |
690 | * Allocate fake pgd for the page table manipulation macros to | |
691 | * work. This is not used by the hardware and we have no | |
692 | * alignment requirement for this allocation. | |
693 | */ | |
694 | pgd = (pgd_t *)kmalloc(PTRS_PER_S2_PGD * sizeof(pgd_t), | |
695 | GFP_KERNEL | __GFP_ZERO); | |
a987370f MZ |
696 | |
697 | if (!pgd) { | |
698 | kvm_free_hwpgd(hwpgd); | |
699 | return -ENOMEM; | |
700 | } | |
701 | ||
702 | /* Plug the HW PGD into the fake one. */ | |
703 | for (i = 0; i < PTRS_PER_S2_PGD; i++) { | |
704 | if (KVM_PREALLOC_LEVEL == 1) | |
705 | pgd_populate(NULL, pgd + i, | |
706 | (pud_t *)hwpgd + i * PTRS_PER_PUD); | |
707 | else if (KVM_PREALLOC_LEVEL == 2) | |
708 | pud_populate(NULL, pud_offset(pgd, 0) + i, | |
709 | (pmd_t *)hwpgd + i * PTRS_PER_PMD); | |
710 | } | |
38f791a4 CD |
711 | } else { |
712 | /* | |
713 | * Allocate actual first-level Stage-2 page table used by the | |
714 | * hardware for Stage-2 page table walks. | |
715 | */ | |
a987370f | 716 | pgd = (pgd_t *)hwpgd; |
38f791a4 CD |
717 | } |
718 | ||
c62ee2b2 | 719 | kvm_clean_pgd(pgd); |
d5d8184d | 720 | kvm->arch.pgd = pgd; |
d5d8184d CD |
721 | return 0; |
722 | } | |
723 | ||
d5d8184d CD |
724 | /** |
725 | * unmap_stage2_range -- Clear stage2 page table entries to unmap a range | |
726 | * @kvm: The VM pointer | |
727 | * @start: The intermediate physical base address of the range to unmap | |
728 | * @size: The size of the area to unmap | |
729 | * | |
730 | * Clear a range of stage-2 mappings, lowering the various ref-counts. Must | |
731 | * be called while holding mmu_lock (unless for freeing the stage2 pgd before | |
732 | * destroying the VM), otherwise another faulting VCPU may come in and mess | |
733 | * with things behind our backs. | |
734 | */ | |
735 | static void unmap_stage2_range(struct kvm *kvm, phys_addr_t start, u64 size) | |
736 | { | |
d4cb9df5 | 737 | unmap_range(kvm, kvm->arch.pgd, start, size); |
d5d8184d CD |
738 | } |
739 | ||
957db105 CD |
740 | static void stage2_unmap_memslot(struct kvm *kvm, |
741 | struct kvm_memory_slot *memslot) | |
742 | { | |
743 | hva_t hva = memslot->userspace_addr; | |
744 | phys_addr_t addr = memslot->base_gfn << PAGE_SHIFT; | |
745 | phys_addr_t size = PAGE_SIZE * memslot->npages; | |
746 | hva_t reg_end = hva + size; | |
747 | ||
748 | /* | |
749 | * A memory region could potentially cover multiple VMAs, and any holes | |
750 | * between them, so iterate over all of them to find out if we should | |
751 | * unmap any of them. | |
752 | * | |
753 | * +--------------------------------------------+ | |
754 | * +---------------+----------------+ +----------------+ | |
755 | * | : VMA 1 | VMA 2 | | VMA 3 : | | |
756 | * +---------------+----------------+ +----------------+ | |
757 | * | memory region | | |
758 | * +--------------------------------------------+ | |
759 | */ | |
760 | do { | |
761 | struct vm_area_struct *vma = find_vma(current->mm, hva); | |
762 | hva_t vm_start, vm_end; | |
763 | ||
764 | if (!vma || vma->vm_start >= reg_end) | |
765 | break; | |
766 | ||
767 | /* | |
768 | * Take the intersection of this VMA with the memory region | |
769 | */ | |
770 | vm_start = max(hva, vma->vm_start); | |
771 | vm_end = min(reg_end, vma->vm_end); | |
772 | ||
773 | if (!(vma->vm_flags & VM_PFNMAP)) { | |
774 | gpa_t gpa = addr + (vm_start - memslot->userspace_addr); | |
775 | unmap_stage2_range(kvm, gpa, vm_end - vm_start); | |
776 | } | |
777 | hva = vm_end; | |
778 | } while (hva < reg_end); | |
779 | } | |
780 | ||
781 | /** | |
782 | * stage2_unmap_vm - Unmap Stage-2 RAM mappings | |
783 | * @kvm: The struct kvm pointer | |
784 | * | |
785 | * Go through the memregions and unmap any reguler RAM | |
786 | * backing memory already mapped to the VM. | |
787 | */ | |
788 | void stage2_unmap_vm(struct kvm *kvm) | |
789 | { | |
790 | struct kvm_memslots *slots; | |
791 | struct kvm_memory_slot *memslot; | |
792 | int idx; | |
793 | ||
794 | idx = srcu_read_lock(&kvm->srcu); | |
795 | spin_lock(&kvm->mmu_lock); | |
796 | ||
797 | slots = kvm_memslots(kvm); | |
798 | kvm_for_each_memslot(memslot, slots) | |
799 | stage2_unmap_memslot(kvm, memslot); | |
800 | ||
801 | spin_unlock(&kvm->mmu_lock); | |
802 | srcu_read_unlock(&kvm->srcu, idx); | |
803 | } | |
804 | ||
d5d8184d CD |
805 | /** |
806 | * kvm_free_stage2_pgd - free all stage-2 tables | |
807 | * @kvm: The KVM struct pointer for the VM. | |
808 | * | |
809 | * Walks the level-1 page table pointed to by kvm->arch.pgd and frees all | |
810 | * underlying level-2 and level-3 tables before freeing the actual level-1 table | |
811 | * and setting the struct pointer to NULL. | |
812 | * | |
813 | * Note we don't need locking here as this is only called when the VM is | |
814 | * destroyed, which can only be done once. | |
815 | */ | |
816 | void kvm_free_stage2_pgd(struct kvm *kvm) | |
817 | { | |
818 | if (kvm->arch.pgd == NULL) | |
819 | return; | |
820 | ||
821 | unmap_stage2_range(kvm, 0, KVM_PHYS_SIZE); | |
a987370f | 822 | kvm_free_hwpgd(kvm_get_hwpgd(kvm)); |
38f791a4 CD |
823 | if (KVM_PREALLOC_LEVEL > 0) |
824 | kfree(kvm->arch.pgd); | |
a987370f | 825 | |
d5d8184d CD |
826 | kvm->arch.pgd = NULL; |
827 | } | |
828 | ||
38f791a4 | 829 | static pud_t *stage2_get_pud(struct kvm *kvm, struct kvm_mmu_memory_cache *cache, |
ad361f09 | 830 | phys_addr_t addr) |
d5d8184d CD |
831 | { |
832 | pgd_t *pgd; | |
833 | pud_t *pud; | |
d5d8184d | 834 | |
04b8dc85 | 835 | pgd = kvm->arch.pgd + kvm_pgd_index(addr); |
38f791a4 CD |
836 | if (WARN_ON(pgd_none(*pgd))) { |
837 | if (!cache) | |
838 | return NULL; | |
839 | pud = mmu_memory_cache_alloc(cache); | |
840 | pgd_populate(NULL, pgd, pud); | |
841 | get_page(virt_to_page(pgd)); | |
842 | } | |
843 | ||
844 | return pud_offset(pgd, addr); | |
845 | } | |
846 | ||
847 | static pmd_t *stage2_get_pmd(struct kvm *kvm, struct kvm_mmu_memory_cache *cache, | |
848 | phys_addr_t addr) | |
849 | { | |
850 | pud_t *pud; | |
851 | pmd_t *pmd; | |
852 | ||
853 | pud = stage2_get_pud(kvm, cache, addr); | |
d5d8184d CD |
854 | if (pud_none(*pud)) { |
855 | if (!cache) | |
ad361f09 | 856 | return NULL; |
d5d8184d CD |
857 | pmd = mmu_memory_cache_alloc(cache); |
858 | pud_populate(NULL, pud, pmd); | |
d5d8184d | 859 | get_page(virt_to_page(pud)); |
c62ee2b2 MZ |
860 | } |
861 | ||
ad361f09 CD |
862 | return pmd_offset(pud, addr); |
863 | } | |
864 | ||
865 | static int stage2_set_pmd_huge(struct kvm *kvm, struct kvm_mmu_memory_cache | |
866 | *cache, phys_addr_t addr, const pmd_t *new_pmd) | |
867 | { | |
868 | pmd_t *pmd, old_pmd; | |
869 | ||
870 | pmd = stage2_get_pmd(kvm, cache, addr); | |
871 | VM_BUG_ON(!pmd); | |
d5d8184d | 872 | |
ad361f09 CD |
873 | /* |
874 | * Mapping in huge pages should only happen through a fault. If a | |
875 | * page is merged into a transparent huge page, the individual | |
876 | * subpages of that huge page should be unmapped through MMU | |
877 | * notifiers before we get here. | |
878 | * | |
879 | * Merging of CompoundPages is not supported; they should become | |
880 | * splitting first, unmapped, merged, and mapped back in on-demand. | |
881 | */ | |
882 | VM_BUG_ON(pmd_present(*pmd) && pmd_pfn(*pmd) != pmd_pfn(*new_pmd)); | |
883 | ||
884 | old_pmd = *pmd; | |
885 | kvm_set_pmd(pmd, *new_pmd); | |
886 | if (pmd_present(old_pmd)) | |
887 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
888 | else | |
889 | get_page(virt_to_page(pmd)); | |
890 | return 0; | |
891 | } | |
892 | ||
893 | static int stage2_set_pte(struct kvm *kvm, struct kvm_mmu_memory_cache *cache, | |
15a49a44 MS |
894 | phys_addr_t addr, const pte_t *new_pte, |
895 | unsigned long flags) | |
ad361f09 CD |
896 | { |
897 | pmd_t *pmd; | |
898 | pte_t *pte, old_pte; | |
15a49a44 MS |
899 | bool iomap = flags & KVM_S2PTE_FLAG_IS_IOMAP; |
900 | bool logging_active = flags & KVM_S2_FLAG_LOGGING_ACTIVE; | |
901 | ||
902 | VM_BUG_ON(logging_active && !cache); | |
ad361f09 | 903 | |
38f791a4 | 904 | /* Create stage-2 page table mapping - Levels 0 and 1 */ |
ad361f09 CD |
905 | pmd = stage2_get_pmd(kvm, cache, addr); |
906 | if (!pmd) { | |
907 | /* | |
908 | * Ignore calls from kvm_set_spte_hva for unallocated | |
909 | * address ranges. | |
910 | */ | |
911 | return 0; | |
912 | } | |
913 | ||
15a49a44 MS |
914 | /* |
915 | * While dirty page logging - dissolve huge PMD, then continue on to | |
916 | * allocate page. | |
917 | */ | |
918 | if (logging_active) | |
919 | stage2_dissolve_pmd(kvm, addr, pmd); | |
920 | ||
ad361f09 | 921 | /* Create stage-2 page mappings - Level 2 */ |
d5d8184d CD |
922 | if (pmd_none(*pmd)) { |
923 | if (!cache) | |
924 | return 0; /* ignore calls from kvm_set_spte_hva */ | |
925 | pte = mmu_memory_cache_alloc(cache); | |
c62ee2b2 | 926 | kvm_clean_pte(pte); |
d5d8184d | 927 | pmd_populate_kernel(NULL, pmd, pte); |
d5d8184d | 928 | get_page(virt_to_page(pmd)); |
c62ee2b2 MZ |
929 | } |
930 | ||
931 | pte = pte_offset_kernel(pmd, addr); | |
d5d8184d CD |
932 | |
933 | if (iomap && pte_present(*pte)) | |
934 | return -EFAULT; | |
935 | ||
936 | /* Create 2nd stage page table mapping - Level 3 */ | |
937 | old_pte = *pte; | |
938 | kvm_set_pte(pte, *new_pte); | |
939 | if (pte_present(old_pte)) | |
48762767 | 940 | kvm_tlb_flush_vmid_ipa(kvm, addr); |
d5d8184d CD |
941 | else |
942 | get_page(virt_to_page(pte)); | |
943 | ||
944 | return 0; | |
945 | } | |
946 | ||
947 | /** | |
948 | * kvm_phys_addr_ioremap - map a device range to guest IPA | |
949 | * | |
950 | * @kvm: The KVM pointer | |
951 | * @guest_ipa: The IPA at which to insert the mapping | |
952 | * @pa: The physical address of the device | |
953 | * @size: The size of the mapping | |
954 | */ | |
955 | int kvm_phys_addr_ioremap(struct kvm *kvm, phys_addr_t guest_ipa, | |
c40f2f8f | 956 | phys_addr_t pa, unsigned long size, bool writable) |
d5d8184d CD |
957 | { |
958 | phys_addr_t addr, end; | |
959 | int ret = 0; | |
960 | unsigned long pfn; | |
961 | struct kvm_mmu_memory_cache cache = { 0, }; | |
962 | ||
963 | end = (guest_ipa + size + PAGE_SIZE - 1) & PAGE_MASK; | |
964 | pfn = __phys_to_pfn(pa); | |
965 | ||
966 | for (addr = guest_ipa; addr < end; addr += PAGE_SIZE) { | |
c62ee2b2 | 967 | pte_t pte = pfn_pte(pfn, PAGE_S2_DEVICE); |
d5d8184d | 968 | |
c40f2f8f AB |
969 | if (writable) |
970 | kvm_set_s2pte_writable(&pte); | |
971 | ||
38f791a4 CD |
972 | ret = mmu_topup_memory_cache(&cache, KVM_MMU_CACHE_MIN_PAGES, |
973 | KVM_NR_MEM_OBJS); | |
d5d8184d CD |
974 | if (ret) |
975 | goto out; | |
976 | spin_lock(&kvm->mmu_lock); | |
15a49a44 MS |
977 | ret = stage2_set_pte(kvm, &cache, addr, &pte, |
978 | KVM_S2PTE_FLAG_IS_IOMAP); | |
d5d8184d CD |
979 | spin_unlock(&kvm->mmu_lock); |
980 | if (ret) | |
981 | goto out; | |
982 | ||
983 | pfn++; | |
984 | } | |
985 | ||
986 | out: | |
987 | mmu_free_memory_cache(&cache); | |
988 | return ret; | |
989 | } | |
990 | ||
9b5fdb97 CD |
991 | static bool transparent_hugepage_adjust(pfn_t *pfnp, phys_addr_t *ipap) |
992 | { | |
993 | pfn_t pfn = *pfnp; | |
994 | gfn_t gfn = *ipap >> PAGE_SHIFT; | |
995 | ||
996 | if (PageTransCompound(pfn_to_page(pfn))) { | |
997 | unsigned long mask; | |
998 | /* | |
999 | * The address we faulted on is backed by a transparent huge | |
1000 | * page. However, because we map the compound huge page and | |
1001 | * not the individual tail page, we need to transfer the | |
1002 | * refcount to the head page. We have to be careful that the | |
1003 | * THP doesn't start to split while we are adjusting the | |
1004 | * refcounts. | |
1005 | * | |
1006 | * We are sure this doesn't happen, because mmu_notifier_retry | |
1007 | * was successful and we are holding the mmu_lock, so if this | |
1008 | * THP is trying to split, it will be blocked in the mmu | |
1009 | * notifier before touching any of the pages, specifically | |
1010 | * before being able to call __split_huge_page_refcount(). | |
1011 | * | |
1012 | * We can therefore safely transfer the refcount from PG_tail | |
1013 | * to PG_head and switch the pfn from a tail page to the head | |
1014 | * page accordingly. | |
1015 | */ | |
1016 | mask = PTRS_PER_PMD - 1; | |
1017 | VM_BUG_ON((gfn & mask) != (pfn & mask)); | |
1018 | if (pfn & mask) { | |
1019 | *ipap &= PMD_MASK; | |
1020 | kvm_release_pfn_clean(pfn); | |
1021 | pfn &= ~mask; | |
1022 | kvm_get_pfn(pfn); | |
1023 | *pfnp = pfn; | |
1024 | } | |
1025 | ||
1026 | return true; | |
1027 | } | |
1028 | ||
1029 | return false; | |
1030 | } | |
1031 | ||
a7d079ce AB |
1032 | static bool kvm_is_write_fault(struct kvm_vcpu *vcpu) |
1033 | { | |
1034 | if (kvm_vcpu_trap_is_iabt(vcpu)) | |
1035 | return false; | |
1036 | ||
1037 | return kvm_vcpu_dabt_iswrite(vcpu); | |
1038 | } | |
1039 | ||
bb55e9b1 AB |
1040 | static bool kvm_is_device_pfn(unsigned long pfn) |
1041 | { | |
1042 | return !pfn_valid(pfn); | |
1043 | } | |
1044 | ||
c6473555 MS |
1045 | /** |
1046 | * stage2_wp_ptes - write protect PMD range | |
1047 | * @pmd: pointer to pmd entry | |
1048 | * @addr: range start address | |
1049 | * @end: range end address | |
1050 | */ | |
1051 | static void stage2_wp_ptes(pmd_t *pmd, phys_addr_t addr, phys_addr_t end) | |
1052 | { | |
1053 | pte_t *pte; | |
1054 | ||
1055 | pte = pte_offset_kernel(pmd, addr); | |
1056 | do { | |
1057 | if (!pte_none(*pte)) { | |
1058 | if (!kvm_s2pte_readonly(pte)) | |
1059 | kvm_set_s2pte_readonly(pte); | |
1060 | } | |
1061 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
1062 | } | |
1063 | ||
1064 | /** | |
1065 | * stage2_wp_pmds - write protect PUD range | |
1066 | * @pud: pointer to pud entry | |
1067 | * @addr: range start address | |
1068 | * @end: range end address | |
1069 | */ | |
1070 | static void stage2_wp_pmds(pud_t *pud, phys_addr_t addr, phys_addr_t end) | |
1071 | { | |
1072 | pmd_t *pmd; | |
1073 | phys_addr_t next; | |
1074 | ||
1075 | pmd = pmd_offset(pud, addr); | |
1076 | ||
1077 | do { | |
1078 | next = kvm_pmd_addr_end(addr, end); | |
1079 | if (!pmd_none(*pmd)) { | |
1080 | if (kvm_pmd_huge(*pmd)) { | |
1081 | if (!kvm_s2pmd_readonly(pmd)) | |
1082 | kvm_set_s2pmd_readonly(pmd); | |
1083 | } else { | |
1084 | stage2_wp_ptes(pmd, addr, next); | |
1085 | } | |
1086 | } | |
1087 | } while (pmd++, addr = next, addr != end); | |
1088 | } | |
1089 | ||
1090 | /** | |
1091 | * stage2_wp_puds - write protect PGD range | |
1092 | * @pgd: pointer to pgd entry | |
1093 | * @addr: range start address | |
1094 | * @end: range end address | |
1095 | * | |
1096 | * Process PUD entries, for a huge PUD we cause a panic. | |
1097 | */ | |
1098 | static void stage2_wp_puds(pgd_t *pgd, phys_addr_t addr, phys_addr_t end) | |
1099 | { | |
1100 | pud_t *pud; | |
1101 | phys_addr_t next; | |
1102 | ||
1103 | pud = pud_offset(pgd, addr); | |
1104 | do { | |
1105 | next = kvm_pud_addr_end(addr, end); | |
1106 | if (!pud_none(*pud)) { | |
1107 | /* TODO:PUD not supported, revisit later if supported */ | |
1108 | BUG_ON(kvm_pud_huge(*pud)); | |
1109 | stage2_wp_pmds(pud, addr, next); | |
1110 | } | |
1111 | } while (pud++, addr = next, addr != end); | |
1112 | } | |
1113 | ||
1114 | /** | |
1115 | * stage2_wp_range() - write protect stage2 memory region range | |
1116 | * @kvm: The KVM pointer | |
1117 | * @addr: Start address of range | |
1118 | * @end: End address of range | |
1119 | */ | |
1120 | static void stage2_wp_range(struct kvm *kvm, phys_addr_t addr, phys_addr_t end) | |
1121 | { | |
1122 | pgd_t *pgd; | |
1123 | phys_addr_t next; | |
1124 | ||
04b8dc85 | 1125 | pgd = kvm->arch.pgd + kvm_pgd_index(addr); |
c6473555 MS |
1126 | do { |
1127 | /* | |
1128 | * Release kvm_mmu_lock periodically if the memory region is | |
1129 | * large. Otherwise, we may see kernel panics with | |
227ea818 CD |
1130 | * CONFIG_DETECT_HUNG_TASK, CONFIG_LOCKUP_DETECTOR, |
1131 | * CONFIG_LOCKDEP. Additionally, holding the lock too long | |
c6473555 MS |
1132 | * will also starve other vCPUs. |
1133 | */ | |
1134 | if (need_resched() || spin_needbreak(&kvm->mmu_lock)) | |
1135 | cond_resched_lock(&kvm->mmu_lock); | |
1136 | ||
1137 | next = kvm_pgd_addr_end(addr, end); | |
1138 | if (pgd_present(*pgd)) | |
1139 | stage2_wp_puds(pgd, addr, next); | |
1140 | } while (pgd++, addr = next, addr != end); | |
1141 | } | |
1142 | ||
1143 | /** | |
1144 | * kvm_mmu_wp_memory_region() - write protect stage 2 entries for memory slot | |
1145 | * @kvm: The KVM pointer | |
1146 | * @slot: The memory slot to write protect | |
1147 | * | |
1148 | * Called to start logging dirty pages after memory region | |
1149 | * KVM_MEM_LOG_DIRTY_PAGES operation is called. After this function returns | |
1150 | * all present PMD and PTEs are write protected in the memory region. | |
1151 | * Afterwards read of dirty page log can be called. | |
1152 | * | |
1153 | * Acquires kvm_mmu_lock. Called with kvm->slots_lock mutex acquired, | |
1154 | * serializing operations for VM memory regions. | |
1155 | */ | |
1156 | void kvm_mmu_wp_memory_region(struct kvm *kvm, int slot) | |
1157 | { | |
9f6b8029 PB |
1158 | struct kvm_memslots *slots = kvm_memslots(kvm); |
1159 | struct kvm_memory_slot *memslot = id_to_memslot(slots, slot); | |
c6473555 MS |
1160 | phys_addr_t start = memslot->base_gfn << PAGE_SHIFT; |
1161 | phys_addr_t end = (memslot->base_gfn + memslot->npages) << PAGE_SHIFT; | |
1162 | ||
1163 | spin_lock(&kvm->mmu_lock); | |
1164 | stage2_wp_range(kvm, start, end); | |
1165 | spin_unlock(&kvm->mmu_lock); | |
1166 | kvm_flush_remote_tlbs(kvm); | |
1167 | } | |
53c810c3 MS |
1168 | |
1169 | /** | |
3b0f1d01 | 1170 | * kvm_mmu_write_protect_pt_masked() - write protect dirty pages |
53c810c3 MS |
1171 | * @kvm: The KVM pointer |
1172 | * @slot: The memory slot associated with mask | |
1173 | * @gfn_offset: The gfn offset in memory slot | |
1174 | * @mask: The mask of dirty pages at offset 'gfn_offset' in this memory | |
1175 | * slot to be write protected | |
1176 | * | |
1177 | * Walks bits set in mask write protects the associated pte's. Caller must | |
1178 | * acquire kvm_mmu_lock. | |
1179 | */ | |
3b0f1d01 | 1180 | static void kvm_mmu_write_protect_pt_masked(struct kvm *kvm, |
53c810c3 MS |
1181 | struct kvm_memory_slot *slot, |
1182 | gfn_t gfn_offset, unsigned long mask) | |
1183 | { | |
1184 | phys_addr_t base_gfn = slot->base_gfn + gfn_offset; | |
1185 | phys_addr_t start = (base_gfn + __ffs(mask)) << PAGE_SHIFT; | |
1186 | phys_addr_t end = (base_gfn + __fls(mask) + 1) << PAGE_SHIFT; | |
1187 | ||
1188 | stage2_wp_range(kvm, start, end); | |
1189 | } | |
c6473555 | 1190 | |
3b0f1d01 KH |
1191 | /* |
1192 | * kvm_arch_mmu_enable_log_dirty_pt_masked - enable dirty logging for selected | |
1193 | * dirty pages. | |
1194 | * | |
1195 | * It calls kvm_mmu_write_protect_pt_masked to write protect selected pages to | |
1196 | * enable dirty logging for them. | |
1197 | */ | |
1198 | void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm, | |
1199 | struct kvm_memory_slot *slot, | |
1200 | gfn_t gfn_offset, unsigned long mask) | |
1201 | { | |
1202 | kvm_mmu_write_protect_pt_masked(kvm, slot, gfn_offset, mask); | |
1203 | } | |
1204 | ||
0d3e4d4f MZ |
1205 | static void coherent_cache_guest_page(struct kvm_vcpu *vcpu, pfn_t pfn, |
1206 | unsigned long size, bool uncached) | |
1207 | { | |
1208 | __coherent_cache_guest_page(vcpu, pfn, size, uncached); | |
1209 | } | |
1210 | ||
94f8e641 | 1211 | static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa, |
98047888 | 1212 | struct kvm_memory_slot *memslot, unsigned long hva, |
94f8e641 CD |
1213 | unsigned long fault_status) |
1214 | { | |
94f8e641 | 1215 | int ret; |
9b5fdb97 | 1216 | bool write_fault, writable, hugetlb = false, force_pte = false; |
94f8e641 | 1217 | unsigned long mmu_seq; |
ad361f09 | 1218 | gfn_t gfn = fault_ipa >> PAGE_SHIFT; |
ad361f09 | 1219 | struct kvm *kvm = vcpu->kvm; |
94f8e641 | 1220 | struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache; |
ad361f09 CD |
1221 | struct vm_area_struct *vma; |
1222 | pfn_t pfn; | |
b8865767 | 1223 | pgprot_t mem_type = PAGE_S2; |
840f4bfb | 1224 | bool fault_ipa_uncached; |
15a49a44 MS |
1225 | bool logging_active = memslot_is_logging(memslot); |
1226 | unsigned long flags = 0; | |
94f8e641 | 1227 | |
a7d079ce | 1228 | write_fault = kvm_is_write_fault(vcpu); |
94f8e641 CD |
1229 | if (fault_status == FSC_PERM && !write_fault) { |
1230 | kvm_err("Unexpected L2 read permission error\n"); | |
1231 | return -EFAULT; | |
1232 | } | |
1233 | ||
ad361f09 CD |
1234 | /* Let's check if we will get back a huge page backed by hugetlbfs */ |
1235 | down_read(¤t->mm->mmap_sem); | |
1236 | vma = find_vma_intersection(current->mm, hva, hva + 1); | |
37b54408 AB |
1237 | if (unlikely(!vma)) { |
1238 | kvm_err("Failed to find VMA for hva 0x%lx\n", hva); | |
1239 | up_read(¤t->mm->mmap_sem); | |
1240 | return -EFAULT; | |
1241 | } | |
1242 | ||
15a49a44 | 1243 | if (is_vm_hugetlb_page(vma) && !logging_active) { |
ad361f09 CD |
1244 | hugetlb = true; |
1245 | gfn = (fault_ipa & PMD_MASK) >> PAGE_SHIFT; | |
9b5fdb97 CD |
1246 | } else { |
1247 | /* | |
136d737f MZ |
1248 | * Pages belonging to memslots that don't have the same |
1249 | * alignment for userspace and IPA cannot be mapped using | |
1250 | * block descriptors even if the pages belong to a THP for | |
1251 | * the process, because the stage-2 block descriptor will | |
1252 | * cover more than a single THP and we loose atomicity for | |
1253 | * unmapping, updates, and splits of the THP or other pages | |
1254 | * in the stage-2 block range. | |
9b5fdb97 | 1255 | */ |
136d737f MZ |
1256 | if ((memslot->userspace_addr & ~PMD_MASK) != |
1257 | ((memslot->base_gfn << PAGE_SHIFT) & ~PMD_MASK)) | |
9b5fdb97 | 1258 | force_pte = true; |
ad361f09 CD |
1259 | } |
1260 | up_read(¤t->mm->mmap_sem); | |
1261 | ||
94f8e641 | 1262 | /* We need minimum second+third level pages */ |
38f791a4 CD |
1263 | ret = mmu_topup_memory_cache(memcache, KVM_MMU_CACHE_MIN_PAGES, |
1264 | KVM_NR_MEM_OBJS); | |
94f8e641 CD |
1265 | if (ret) |
1266 | return ret; | |
1267 | ||
1268 | mmu_seq = vcpu->kvm->mmu_notifier_seq; | |
1269 | /* | |
1270 | * Ensure the read of mmu_notifier_seq happens before we call | |
1271 | * gfn_to_pfn_prot (which calls get_user_pages), so that we don't risk | |
1272 | * the page we just got a reference to gets unmapped before we have a | |
1273 | * chance to grab the mmu_lock, which ensure that if the page gets | |
1274 | * unmapped afterwards, the call to kvm_unmap_hva will take it away | |
1275 | * from us again properly. This smp_rmb() interacts with the smp_wmb() | |
1276 | * in kvm_mmu_notifier_invalidate_<page|range_end>. | |
1277 | */ | |
1278 | smp_rmb(); | |
1279 | ||
ad361f09 | 1280 | pfn = gfn_to_pfn_prot(kvm, gfn, write_fault, &writable); |
94f8e641 CD |
1281 | if (is_error_pfn(pfn)) |
1282 | return -EFAULT; | |
1283 | ||
15a49a44 | 1284 | if (kvm_is_device_pfn(pfn)) { |
b8865767 | 1285 | mem_type = PAGE_S2_DEVICE; |
15a49a44 MS |
1286 | flags |= KVM_S2PTE_FLAG_IS_IOMAP; |
1287 | } else if (logging_active) { | |
1288 | /* | |
1289 | * Faults on pages in a memslot with logging enabled | |
1290 | * should not be mapped with huge pages (it introduces churn | |
1291 | * and performance degradation), so force a pte mapping. | |
1292 | */ | |
1293 | force_pte = true; | |
1294 | flags |= KVM_S2_FLAG_LOGGING_ACTIVE; | |
1295 | ||
1296 | /* | |
1297 | * Only actually map the page as writable if this was a write | |
1298 | * fault. | |
1299 | */ | |
1300 | if (!write_fault) | |
1301 | writable = false; | |
1302 | } | |
b8865767 | 1303 | |
ad361f09 CD |
1304 | spin_lock(&kvm->mmu_lock); |
1305 | if (mmu_notifier_retry(kvm, mmu_seq)) | |
94f8e641 | 1306 | goto out_unlock; |
15a49a44 | 1307 | |
9b5fdb97 CD |
1308 | if (!hugetlb && !force_pte) |
1309 | hugetlb = transparent_hugepage_adjust(&pfn, &fault_ipa); | |
ad361f09 | 1310 | |
849260c7 | 1311 | fault_ipa_uncached = memslot->flags & KVM_MEMSLOT_INCOHERENT; |
840f4bfb | 1312 | |
ad361f09 | 1313 | if (hugetlb) { |
b8865767 | 1314 | pmd_t new_pmd = pfn_pmd(pfn, mem_type); |
ad361f09 CD |
1315 | new_pmd = pmd_mkhuge(new_pmd); |
1316 | if (writable) { | |
1317 | kvm_set_s2pmd_writable(&new_pmd); | |
1318 | kvm_set_pfn_dirty(pfn); | |
1319 | } | |
0d3e4d4f | 1320 | coherent_cache_guest_page(vcpu, pfn, PMD_SIZE, fault_ipa_uncached); |
ad361f09 CD |
1321 | ret = stage2_set_pmd_huge(kvm, memcache, fault_ipa, &new_pmd); |
1322 | } else { | |
b8865767 | 1323 | pte_t new_pte = pfn_pte(pfn, mem_type); |
15a49a44 | 1324 | |
ad361f09 CD |
1325 | if (writable) { |
1326 | kvm_set_s2pte_writable(&new_pte); | |
1327 | kvm_set_pfn_dirty(pfn); | |
15a49a44 | 1328 | mark_page_dirty(kvm, gfn); |
ad361f09 | 1329 | } |
0d3e4d4f | 1330 | coherent_cache_guest_page(vcpu, pfn, PAGE_SIZE, fault_ipa_uncached); |
15a49a44 | 1331 | ret = stage2_set_pte(kvm, memcache, fault_ipa, &new_pte, flags); |
94f8e641 | 1332 | } |
ad361f09 | 1333 | |
94f8e641 | 1334 | out_unlock: |
ad361f09 | 1335 | spin_unlock(&kvm->mmu_lock); |
35307b9a | 1336 | kvm_set_pfn_accessed(pfn); |
94f8e641 | 1337 | kvm_release_pfn_clean(pfn); |
ad361f09 | 1338 | return ret; |
94f8e641 CD |
1339 | } |
1340 | ||
aeda9130 MZ |
1341 | /* |
1342 | * Resolve the access fault by making the page young again. | |
1343 | * Note that because the faulting entry is guaranteed not to be | |
1344 | * cached in the TLB, we don't need to invalidate anything. | |
1345 | */ | |
1346 | static void handle_access_fault(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa) | |
1347 | { | |
1348 | pmd_t *pmd; | |
1349 | pte_t *pte; | |
1350 | pfn_t pfn; | |
1351 | bool pfn_valid = false; | |
1352 | ||
1353 | trace_kvm_access_fault(fault_ipa); | |
1354 | ||
1355 | spin_lock(&vcpu->kvm->mmu_lock); | |
1356 | ||
1357 | pmd = stage2_get_pmd(vcpu->kvm, NULL, fault_ipa); | |
1358 | if (!pmd || pmd_none(*pmd)) /* Nothing there */ | |
1359 | goto out; | |
1360 | ||
1361 | if (kvm_pmd_huge(*pmd)) { /* THP, HugeTLB */ | |
1362 | *pmd = pmd_mkyoung(*pmd); | |
1363 | pfn = pmd_pfn(*pmd); | |
1364 | pfn_valid = true; | |
1365 | goto out; | |
1366 | } | |
1367 | ||
1368 | pte = pte_offset_kernel(pmd, fault_ipa); | |
1369 | if (pte_none(*pte)) /* Nothing there either */ | |
1370 | goto out; | |
1371 | ||
1372 | *pte = pte_mkyoung(*pte); /* Just a page... */ | |
1373 | pfn = pte_pfn(*pte); | |
1374 | pfn_valid = true; | |
1375 | out: | |
1376 | spin_unlock(&vcpu->kvm->mmu_lock); | |
1377 | if (pfn_valid) | |
1378 | kvm_set_pfn_accessed(pfn); | |
1379 | } | |
1380 | ||
94f8e641 CD |
1381 | /** |
1382 | * kvm_handle_guest_abort - handles all 2nd stage aborts | |
1383 | * @vcpu: the VCPU pointer | |
1384 | * @run: the kvm_run structure | |
1385 | * | |
1386 | * Any abort that gets to the host is almost guaranteed to be caused by a | |
1387 | * missing second stage translation table entry, which can mean that either the | |
1388 | * guest simply needs more memory and we must allocate an appropriate page or it | |
1389 | * can mean that the guest tried to access I/O memory, which is emulated by user | |
1390 | * space. The distinction is based on the IPA causing the fault and whether this | |
1391 | * memory region has been registered as standard RAM by user space. | |
1392 | */ | |
342cd0ab CD |
1393 | int kvm_handle_guest_abort(struct kvm_vcpu *vcpu, struct kvm_run *run) |
1394 | { | |
94f8e641 CD |
1395 | unsigned long fault_status; |
1396 | phys_addr_t fault_ipa; | |
1397 | struct kvm_memory_slot *memslot; | |
98047888 CD |
1398 | unsigned long hva; |
1399 | bool is_iabt, write_fault, writable; | |
94f8e641 CD |
1400 | gfn_t gfn; |
1401 | int ret, idx; | |
1402 | ||
52d1dba9 | 1403 | is_iabt = kvm_vcpu_trap_is_iabt(vcpu); |
7393b599 | 1404 | fault_ipa = kvm_vcpu_get_fault_ipa(vcpu); |
94f8e641 | 1405 | |
7393b599 MZ |
1406 | trace_kvm_guest_fault(*vcpu_pc(vcpu), kvm_vcpu_get_hsr(vcpu), |
1407 | kvm_vcpu_get_hfar(vcpu), fault_ipa); | |
94f8e641 CD |
1408 | |
1409 | /* Check the stage-2 fault is trans. fault or write fault */ | |
0496daa5 | 1410 | fault_status = kvm_vcpu_trap_get_fault_type(vcpu); |
35307b9a MZ |
1411 | if (fault_status != FSC_FAULT && fault_status != FSC_PERM && |
1412 | fault_status != FSC_ACCESS) { | |
0496daa5 CD |
1413 | kvm_err("Unsupported FSC: EC=%#x xFSC=%#lx ESR_EL2=%#lx\n", |
1414 | kvm_vcpu_trap_get_class(vcpu), | |
1415 | (unsigned long)kvm_vcpu_trap_get_fault(vcpu), | |
1416 | (unsigned long)kvm_vcpu_get_hsr(vcpu)); | |
94f8e641 CD |
1417 | return -EFAULT; |
1418 | } | |
1419 | ||
1420 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
1421 | ||
1422 | gfn = fault_ipa >> PAGE_SHIFT; | |
98047888 CD |
1423 | memslot = gfn_to_memslot(vcpu->kvm, gfn); |
1424 | hva = gfn_to_hva_memslot_prot(memslot, gfn, &writable); | |
a7d079ce | 1425 | write_fault = kvm_is_write_fault(vcpu); |
98047888 | 1426 | if (kvm_is_error_hva(hva) || (write_fault && !writable)) { |
94f8e641 CD |
1427 | if (is_iabt) { |
1428 | /* Prefetch Abort on I/O address */ | |
7393b599 | 1429 | kvm_inject_pabt(vcpu, kvm_vcpu_get_hfar(vcpu)); |
94f8e641 CD |
1430 | ret = 1; |
1431 | goto out_unlock; | |
1432 | } | |
1433 | ||
cfe3950c MZ |
1434 | /* |
1435 | * The IPA is reported as [MAX:12], so we need to | |
1436 | * complement it with the bottom 12 bits from the | |
1437 | * faulting VA. This is always 12 bits, irrespective | |
1438 | * of the page size. | |
1439 | */ | |
1440 | fault_ipa |= kvm_vcpu_get_hfar(vcpu) & ((1 << 12) - 1); | |
45e96ea6 | 1441 | ret = io_mem_abort(vcpu, run, fault_ipa); |
94f8e641 CD |
1442 | goto out_unlock; |
1443 | } | |
1444 | ||
c3058d5d CD |
1445 | /* Userspace should not be able to register out-of-bounds IPAs */ |
1446 | VM_BUG_ON(fault_ipa >= KVM_PHYS_SIZE); | |
1447 | ||
aeda9130 MZ |
1448 | if (fault_status == FSC_ACCESS) { |
1449 | handle_access_fault(vcpu, fault_ipa); | |
1450 | ret = 1; | |
1451 | goto out_unlock; | |
1452 | } | |
1453 | ||
98047888 | 1454 | ret = user_mem_abort(vcpu, fault_ipa, memslot, hva, fault_status); |
94f8e641 CD |
1455 | if (ret == 0) |
1456 | ret = 1; | |
1457 | out_unlock: | |
1458 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
1459 | return ret; | |
342cd0ab CD |
1460 | } |
1461 | ||
1d2ebacc MZ |
1462 | static int handle_hva_to_gpa(struct kvm *kvm, |
1463 | unsigned long start, | |
1464 | unsigned long end, | |
1465 | int (*handler)(struct kvm *kvm, | |
1466 | gpa_t gpa, void *data), | |
1467 | void *data) | |
d5d8184d CD |
1468 | { |
1469 | struct kvm_memslots *slots; | |
1470 | struct kvm_memory_slot *memslot; | |
1d2ebacc | 1471 | int ret = 0; |
d5d8184d CD |
1472 | |
1473 | slots = kvm_memslots(kvm); | |
1474 | ||
1475 | /* we only care about the pages that the guest sees */ | |
1476 | kvm_for_each_memslot(memslot, slots) { | |
1477 | unsigned long hva_start, hva_end; | |
1478 | gfn_t gfn, gfn_end; | |
1479 | ||
1480 | hva_start = max(start, memslot->userspace_addr); | |
1481 | hva_end = min(end, memslot->userspace_addr + | |
1482 | (memslot->npages << PAGE_SHIFT)); | |
1483 | if (hva_start >= hva_end) | |
1484 | continue; | |
1485 | ||
1486 | /* | |
1487 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
1488 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
1489 | */ | |
1490 | gfn = hva_to_gfn_memslot(hva_start, memslot); | |
1491 | gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot); | |
1492 | ||
1493 | for (; gfn < gfn_end; ++gfn) { | |
1494 | gpa_t gpa = gfn << PAGE_SHIFT; | |
1d2ebacc | 1495 | ret |= handler(kvm, gpa, data); |
d5d8184d CD |
1496 | } |
1497 | } | |
1d2ebacc MZ |
1498 | |
1499 | return ret; | |
d5d8184d CD |
1500 | } |
1501 | ||
1d2ebacc | 1502 | static int kvm_unmap_hva_handler(struct kvm *kvm, gpa_t gpa, void *data) |
d5d8184d CD |
1503 | { |
1504 | unmap_stage2_range(kvm, gpa, PAGE_SIZE); | |
1d2ebacc | 1505 | return 0; |
d5d8184d CD |
1506 | } |
1507 | ||
1508 | int kvm_unmap_hva(struct kvm *kvm, unsigned long hva) | |
1509 | { | |
1510 | unsigned long end = hva + PAGE_SIZE; | |
1511 | ||
1512 | if (!kvm->arch.pgd) | |
1513 | return 0; | |
1514 | ||
1515 | trace_kvm_unmap_hva(hva); | |
1516 | handle_hva_to_gpa(kvm, hva, end, &kvm_unmap_hva_handler, NULL); | |
1517 | return 0; | |
1518 | } | |
1519 | ||
1520 | int kvm_unmap_hva_range(struct kvm *kvm, | |
1521 | unsigned long start, unsigned long end) | |
1522 | { | |
1523 | if (!kvm->arch.pgd) | |
1524 | return 0; | |
1525 | ||
1526 | trace_kvm_unmap_hva_range(start, end); | |
1527 | handle_hva_to_gpa(kvm, start, end, &kvm_unmap_hva_handler, NULL); | |
1528 | return 0; | |
1529 | } | |
1530 | ||
1d2ebacc | 1531 | static int kvm_set_spte_handler(struct kvm *kvm, gpa_t gpa, void *data) |
d5d8184d CD |
1532 | { |
1533 | pte_t *pte = (pte_t *)data; | |
1534 | ||
15a49a44 MS |
1535 | /* |
1536 | * We can always call stage2_set_pte with KVM_S2PTE_FLAG_LOGGING_ACTIVE | |
1537 | * flag clear because MMU notifiers will have unmapped a huge PMD before | |
1538 | * calling ->change_pte() (which in turn calls kvm_set_spte_hva()) and | |
1539 | * therefore stage2_set_pte() never needs to clear out a huge PMD | |
1540 | * through this calling path. | |
1541 | */ | |
1542 | stage2_set_pte(kvm, NULL, gpa, pte, 0); | |
1d2ebacc | 1543 | return 0; |
d5d8184d CD |
1544 | } |
1545 | ||
1546 | ||
1547 | void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte) | |
1548 | { | |
1549 | unsigned long end = hva + PAGE_SIZE; | |
1550 | pte_t stage2_pte; | |
1551 | ||
1552 | if (!kvm->arch.pgd) | |
1553 | return; | |
1554 | ||
1555 | trace_kvm_set_spte_hva(hva); | |
1556 | stage2_pte = pfn_pte(pte_pfn(pte), PAGE_S2); | |
1557 | handle_hva_to_gpa(kvm, hva, end, &kvm_set_spte_handler, &stage2_pte); | |
1558 | } | |
1559 | ||
35307b9a MZ |
1560 | static int kvm_age_hva_handler(struct kvm *kvm, gpa_t gpa, void *data) |
1561 | { | |
1562 | pmd_t *pmd; | |
1563 | pte_t *pte; | |
1564 | ||
1565 | pmd = stage2_get_pmd(kvm, NULL, gpa); | |
1566 | if (!pmd || pmd_none(*pmd)) /* Nothing there */ | |
1567 | return 0; | |
1568 | ||
1569 | if (kvm_pmd_huge(*pmd)) { /* THP, HugeTLB */ | |
1570 | if (pmd_young(*pmd)) { | |
1571 | *pmd = pmd_mkold(*pmd); | |
1572 | return 1; | |
1573 | } | |
1574 | ||
1575 | return 0; | |
1576 | } | |
1577 | ||
1578 | pte = pte_offset_kernel(pmd, gpa); | |
1579 | if (pte_none(*pte)) | |
1580 | return 0; | |
1581 | ||
1582 | if (pte_young(*pte)) { | |
1583 | *pte = pte_mkold(*pte); /* Just a page... */ | |
1584 | return 1; | |
1585 | } | |
1586 | ||
1587 | return 0; | |
1588 | } | |
1589 | ||
1590 | static int kvm_test_age_hva_handler(struct kvm *kvm, gpa_t gpa, void *data) | |
1591 | { | |
1592 | pmd_t *pmd; | |
1593 | pte_t *pte; | |
1594 | ||
1595 | pmd = stage2_get_pmd(kvm, NULL, gpa); | |
1596 | if (!pmd || pmd_none(*pmd)) /* Nothing there */ | |
1597 | return 0; | |
1598 | ||
1599 | if (kvm_pmd_huge(*pmd)) /* THP, HugeTLB */ | |
1600 | return pmd_young(*pmd); | |
1601 | ||
1602 | pte = pte_offset_kernel(pmd, gpa); | |
1603 | if (!pte_none(*pte)) /* Just a page... */ | |
1604 | return pte_young(*pte); | |
1605 | ||
1606 | return 0; | |
1607 | } | |
1608 | ||
1609 | int kvm_age_hva(struct kvm *kvm, unsigned long start, unsigned long end) | |
1610 | { | |
1611 | trace_kvm_age_hva(start, end); | |
1612 | return handle_hva_to_gpa(kvm, start, end, kvm_age_hva_handler, NULL); | |
1613 | } | |
1614 | ||
1615 | int kvm_test_age_hva(struct kvm *kvm, unsigned long hva) | |
1616 | { | |
1617 | trace_kvm_test_age_hva(hva); | |
1618 | return handle_hva_to_gpa(kvm, hva, hva, kvm_test_age_hva_handler, NULL); | |
1619 | } | |
1620 | ||
d5d8184d CD |
1621 | void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu) |
1622 | { | |
1623 | mmu_free_memory_cache(&vcpu->arch.mmu_page_cache); | |
1624 | } | |
1625 | ||
342cd0ab CD |
1626 | phys_addr_t kvm_mmu_get_httbr(void) |
1627 | { | |
e4c5a685 AB |
1628 | if (__kvm_cpu_uses_extended_idmap()) |
1629 | return virt_to_phys(merged_hyp_pgd); | |
1630 | else | |
1631 | return virt_to_phys(hyp_pgd); | |
342cd0ab CD |
1632 | } |
1633 | ||
5a677ce0 MZ |
1634 | phys_addr_t kvm_mmu_get_boot_httbr(void) |
1635 | { | |
e4c5a685 AB |
1636 | if (__kvm_cpu_uses_extended_idmap()) |
1637 | return virt_to_phys(merged_hyp_pgd); | |
1638 | else | |
1639 | return virt_to_phys(boot_hyp_pgd); | |
5a677ce0 MZ |
1640 | } |
1641 | ||
1642 | phys_addr_t kvm_get_idmap_vector(void) | |
1643 | { | |
1644 | return hyp_idmap_vector; | |
1645 | } | |
1646 | ||
342cd0ab CD |
1647 | int kvm_mmu_init(void) |
1648 | { | |
2fb41059 MZ |
1649 | int err; |
1650 | ||
4fda342c SS |
1651 | hyp_idmap_start = kvm_virt_to_phys(__hyp_idmap_text_start); |
1652 | hyp_idmap_end = kvm_virt_to_phys(__hyp_idmap_text_end); | |
1653 | hyp_idmap_vector = kvm_virt_to_phys(__kvm_hyp_init); | |
5a677ce0 | 1654 | |
06f75a1f AB |
1655 | /* |
1656 | * We rely on the linker script to ensure at build time that the HYP | |
1657 | * init code does not cross a page boundary. | |
1658 | */ | |
1659 | BUG_ON((hyp_idmap_start ^ (hyp_idmap_end - 1)) & PAGE_MASK); | |
5a677ce0 | 1660 | |
38f791a4 CD |
1661 | hyp_pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, hyp_pgd_order); |
1662 | boot_hyp_pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, hyp_pgd_order); | |
5d4e08c4 | 1663 | |
5a677ce0 | 1664 | if (!hyp_pgd || !boot_hyp_pgd) { |
d5d8184d | 1665 | kvm_err("Hyp mode PGD not allocated\n"); |
2fb41059 MZ |
1666 | err = -ENOMEM; |
1667 | goto out; | |
1668 | } | |
1669 | ||
1670 | /* Create the idmap in the boot page tables */ | |
1671 | err = __create_hyp_mappings(boot_hyp_pgd, | |
1672 | hyp_idmap_start, hyp_idmap_end, | |
1673 | __phys_to_pfn(hyp_idmap_start), | |
1674 | PAGE_HYP); | |
1675 | ||
1676 | if (err) { | |
1677 | kvm_err("Failed to idmap %lx-%lx\n", | |
1678 | hyp_idmap_start, hyp_idmap_end); | |
1679 | goto out; | |
d5d8184d CD |
1680 | } |
1681 | ||
e4c5a685 AB |
1682 | if (__kvm_cpu_uses_extended_idmap()) { |
1683 | merged_hyp_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO); | |
1684 | if (!merged_hyp_pgd) { | |
1685 | kvm_err("Failed to allocate extra HYP pgd\n"); | |
1686 | goto out; | |
1687 | } | |
1688 | __kvm_extend_hypmap(boot_hyp_pgd, hyp_pgd, merged_hyp_pgd, | |
1689 | hyp_idmap_start); | |
1690 | return 0; | |
1691 | } | |
1692 | ||
5a677ce0 MZ |
1693 | /* Map the very same page at the trampoline VA */ |
1694 | err = __create_hyp_mappings(boot_hyp_pgd, | |
1695 | TRAMPOLINE_VA, TRAMPOLINE_VA + PAGE_SIZE, | |
1696 | __phys_to_pfn(hyp_idmap_start), | |
1697 | PAGE_HYP); | |
1698 | if (err) { | |
1699 | kvm_err("Failed to map trampoline @%lx into boot HYP pgd\n", | |
1700 | TRAMPOLINE_VA); | |
1701 | goto out; | |
1702 | } | |
1703 | ||
1704 | /* Map the same page again into the runtime page tables */ | |
1705 | err = __create_hyp_mappings(hyp_pgd, | |
1706 | TRAMPOLINE_VA, TRAMPOLINE_VA + PAGE_SIZE, | |
1707 | __phys_to_pfn(hyp_idmap_start), | |
1708 | PAGE_HYP); | |
1709 | if (err) { | |
1710 | kvm_err("Failed to map trampoline @%lx into runtime HYP pgd\n", | |
1711 | TRAMPOLINE_VA); | |
1712 | goto out; | |
1713 | } | |
1714 | ||
d5d8184d | 1715 | return 0; |
2fb41059 | 1716 | out: |
4f728276 | 1717 | free_hyp_pgds(); |
2fb41059 | 1718 | return err; |
342cd0ab | 1719 | } |
df6ce24f EA |
1720 | |
1721 | void kvm_arch_commit_memory_region(struct kvm *kvm, | |
1722 | struct kvm_userspace_memory_region *mem, | |
1723 | const struct kvm_memory_slot *old, | |
1724 | enum kvm_mr_change change) | |
1725 | { | |
c6473555 MS |
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); | |
df6ce24f EA |
1733 | } |
1734 | ||
1735 | int kvm_arch_prepare_memory_region(struct kvm *kvm, | |
1736 | struct kvm_memory_slot *memslot, | |
1737 | struct kvm_userspace_memory_region *mem, | |
1738 | enum kvm_mr_change change) | |
1739 | { | |
8eef9123 AB |
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 | ||
15a49a44 MS |
1745 | if (change != KVM_MR_CREATE && change != KVM_MR_MOVE && |
1746 | change != KVM_MR_FLAGS_ONLY) | |
8eef9123 AB |
1747 | return 0; |
1748 | ||
c3058d5d CD |
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 | ||
8eef9123 AB |
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 = (vma->vm_pgoff << PAGE_SHIFT) + | |
1795 | vm_start - vma->vm_start; | |
1796 | ||
15a49a44 MS |
1797 | /* IO region dirty page logging not allowed */ |
1798 | if (memslot->flags & KVM_MEM_LOG_DIRTY_PAGES) | |
1799 | return -EINVAL; | |
1800 | ||
8eef9123 AB |
1801 | ret = kvm_phys_addr_ioremap(kvm, gpa, pa, |
1802 | vm_end - vm_start, | |
1803 | writable); | |
1804 | if (ret) | |
1805 | break; | |
1806 | } | |
1807 | hva = vm_end; | |
1808 | } while (hva < reg_end); | |
1809 | ||
15a49a44 MS |
1810 | if (change == KVM_MR_FLAGS_ONLY) |
1811 | return ret; | |
1812 | ||
849260c7 AB |
1813 | spin_lock(&kvm->mmu_lock); |
1814 | if (ret) | |
8eef9123 | 1815 | unmap_stage2_range(kvm, mem->guest_phys_addr, mem->memory_size); |
849260c7 AB |
1816 | else |
1817 | stage2_flush_memslot(kvm, memslot); | |
1818 | spin_unlock(&kvm->mmu_lock); | |
8eef9123 | 1819 | return ret; |
df6ce24f EA |
1820 | } |
1821 | ||
1822 | void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, | |
1823 | struct kvm_memory_slot *dont) | |
1824 | { | |
1825 | } | |
1826 | ||
1827 | int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot, | |
1828 | unsigned long npages) | |
1829 | { | |
849260c7 AB |
1830 | /* |
1831 | * Readonly memslots are not incoherent with the caches by definition, | |
1832 | * but in practice, they are used mostly to emulate ROMs or NOR flashes | |
1833 | * that the guest may consider devices and hence map as uncached. | |
1834 | * To prevent incoherency issues in these cases, tag all readonly | |
1835 | * regions as incoherent. | |
1836 | */ | |
1837 | if (slot->flags & KVM_MEM_READONLY) | |
1838 | slot->flags |= KVM_MEMSLOT_INCOHERENT; | |
df6ce24f EA |
1839 | return 0; |
1840 | } | |
1841 | ||
1842 | void kvm_arch_memslots_updated(struct kvm *kvm) | |
1843 | { | |
1844 | } | |
1845 | ||
1846 | void kvm_arch_flush_shadow_all(struct kvm *kvm) | |
1847 | { | |
1848 | } | |
1849 | ||
1850 | void kvm_arch_flush_shadow_memslot(struct kvm *kvm, | |
1851 | struct kvm_memory_slot *slot) | |
1852 | { | |
8eef9123 AB |
1853 | gpa_t gpa = slot->base_gfn << PAGE_SHIFT; |
1854 | phys_addr_t size = slot->npages << PAGE_SHIFT; | |
1855 | ||
1856 | spin_lock(&kvm->mmu_lock); | |
1857 | unmap_stage2_range(kvm, gpa, size); | |
1858 | spin_unlock(&kvm->mmu_lock); | |
df6ce24f | 1859 | } |
3c1e7165 MZ |
1860 | |
1861 | /* | |
1862 | * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized). | |
1863 | * | |
1864 | * Main problems: | |
1865 | * - S/W ops are local to a CPU (not broadcast) | |
1866 | * - We have line migration behind our back (speculation) | |
1867 | * - System caches don't support S/W at all (damn!) | |
1868 | * | |
1869 | * In the face of the above, the best we can do is to try and convert | |
1870 | * S/W ops to VA ops. Because the guest is not allowed to infer the | |
1871 | * S/W to PA mapping, it can only use S/W to nuke the whole cache, | |
1872 | * which is a rather good thing for us. | |
1873 | * | |
1874 | * Also, it is only used when turning caches on/off ("The expected | |
1875 | * usage of the cache maintenance instructions that operate by set/way | |
1876 | * is associated with the cache maintenance instructions associated | |
1877 | * with the powerdown and powerup of caches, if this is required by | |
1878 | * the implementation."). | |
1879 | * | |
1880 | * We use the following policy: | |
1881 | * | |
1882 | * - If we trap a S/W operation, we enable VM trapping to detect | |
1883 | * caches being turned on/off, and do a full clean. | |
1884 | * | |
1885 | * - We flush the caches on both caches being turned on and off. | |
1886 | * | |
1887 | * - Once the caches are enabled, we stop trapping VM ops. | |
1888 | */ | |
1889 | void kvm_set_way_flush(struct kvm_vcpu *vcpu) | |
1890 | { | |
1891 | unsigned long hcr = vcpu_get_hcr(vcpu); | |
1892 | ||
1893 | /* | |
1894 | * If this is the first time we do a S/W operation | |
1895 | * (i.e. HCR_TVM not set) flush the whole memory, and set the | |
1896 | * VM trapping. | |
1897 | * | |
1898 | * Otherwise, rely on the VM trapping to wait for the MMU + | |
1899 | * Caches to be turned off. At that point, we'll be able to | |
1900 | * clean the caches again. | |
1901 | */ | |
1902 | if (!(hcr & HCR_TVM)) { | |
1903 | trace_kvm_set_way_flush(*vcpu_pc(vcpu), | |
1904 | vcpu_has_cache_enabled(vcpu)); | |
1905 | stage2_flush_vm(vcpu->kvm); | |
1906 | vcpu_set_hcr(vcpu, hcr | HCR_TVM); | |
1907 | } | |
1908 | } | |
1909 | ||
1910 | void kvm_toggle_cache(struct kvm_vcpu *vcpu, bool was_enabled) | |
1911 | { | |
1912 | bool now_enabled = vcpu_has_cache_enabled(vcpu); | |
1913 | ||
1914 | /* | |
1915 | * If switching the MMU+caches on, need to invalidate the caches. | |
1916 | * If switching it off, need to clean the caches. | |
1917 | * Clean + invalidate does the trick always. | |
1918 | */ | |
1919 | if (now_enabled != was_enabled) | |
1920 | stage2_flush_vm(vcpu->kvm); | |
1921 | ||
1922 | /* Caches are now on, stop trapping VM ops (until a S/W op) */ | |
1923 | if (now_enabled) | |
1924 | vcpu_set_hcr(vcpu, vcpu_get_hcr(vcpu) & ~HCR_TVM); | |
1925 | ||
1926 | trace_kvm_toggle_cache(*vcpu_pc(vcpu), was_enabled, now_enabled); | |
1927 | } |