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; |
342cd0ab CD |
38 | static DEFINE_MUTEX(kvm_hyp_pgd_mutex); |
39 | ||
5a677ce0 MZ |
40 | static void *init_bounce_page; |
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)) |
ad361f09 | 48 | |
48762767 | 49 | static void kvm_tlb_flush_vmid_ipa(struct kvm *kvm, phys_addr_t ipa) |
d5d8184d | 50 | { |
d4cb9df5 MZ |
51 | /* |
52 | * This function also gets called when dealing with HYP page | |
53 | * tables. As HYP doesn't have an associated struct kvm (and | |
54 | * the HYP page tables are fairly static), we don't do | |
55 | * anything there. | |
56 | */ | |
57 | if (kvm) | |
58 | kvm_call_hyp(__kvm_tlb_flush_vmid_ipa, kvm, ipa); | |
d5d8184d CD |
59 | } |
60 | ||
d5d8184d CD |
61 | static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache, |
62 | int min, int max) | |
63 | { | |
64 | void *page; | |
65 | ||
66 | BUG_ON(max > KVM_NR_MEM_OBJS); | |
67 | if (cache->nobjs >= min) | |
68 | return 0; | |
69 | while (cache->nobjs < max) { | |
70 | page = (void *)__get_free_page(PGALLOC_GFP); | |
71 | if (!page) | |
72 | return -ENOMEM; | |
73 | cache->objects[cache->nobjs++] = page; | |
74 | } | |
75 | return 0; | |
76 | } | |
77 | ||
78 | static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
79 | { | |
80 | while (mc->nobjs) | |
81 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
82 | } | |
83 | ||
84 | static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
85 | { | |
86 | void *p; | |
87 | ||
88 | BUG_ON(!mc || !mc->nobjs); | |
89 | p = mc->objects[--mc->nobjs]; | |
90 | return p; | |
91 | } | |
92 | ||
4f853a71 | 93 | static void clear_pgd_entry(struct kvm *kvm, pgd_t *pgd, phys_addr_t addr) |
979acd5e | 94 | { |
4f853a71 CD |
95 | pud_t *pud_table __maybe_unused = pud_offset(pgd, 0); |
96 | pgd_clear(pgd); | |
97 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
98 | pud_free(NULL, pud_table); | |
99 | put_page(virt_to_page(pgd)); | |
979acd5e MZ |
100 | } |
101 | ||
d4cb9df5 | 102 | static void clear_pud_entry(struct kvm *kvm, pud_t *pud, phys_addr_t addr) |
342cd0ab | 103 | { |
4f853a71 CD |
104 | pmd_t *pmd_table = pmd_offset(pud, 0); |
105 | VM_BUG_ON(pud_huge(*pud)); | |
106 | pud_clear(pud); | |
107 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
108 | pmd_free(NULL, pmd_table); | |
4f728276 MZ |
109 | put_page(virt_to_page(pud)); |
110 | } | |
342cd0ab | 111 | |
d4cb9df5 | 112 | static void clear_pmd_entry(struct kvm *kvm, pmd_t *pmd, phys_addr_t addr) |
4f728276 | 113 | { |
4f853a71 CD |
114 | pte_t *pte_table = pte_offset_kernel(pmd, 0); |
115 | VM_BUG_ON(kvm_pmd_huge(*pmd)); | |
116 | pmd_clear(pmd); | |
117 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
118 | pte_free_kernel(NULL, pte_table); | |
4f728276 MZ |
119 | put_page(virt_to_page(pmd)); |
120 | } | |
121 | ||
4f853a71 CD |
122 | static void unmap_ptes(struct kvm *kvm, pmd_t *pmd, |
123 | phys_addr_t addr, phys_addr_t end) | |
4f728276 | 124 | { |
4f853a71 CD |
125 | phys_addr_t start_addr = addr; |
126 | pte_t *pte, *start_pte; | |
127 | ||
128 | start_pte = pte = pte_offset_kernel(pmd, addr); | |
129 | do { | |
130 | if (!pte_none(*pte)) { | |
131 | kvm_set_pte(pte, __pte(0)); | |
132 | put_page(virt_to_page(pte)); | |
133 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
134 | } | |
135 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
136 | ||
38f791a4 | 137 | if (kvm_pte_table_empty(kvm, start_pte)) |
4f853a71 | 138 | clear_pmd_entry(kvm, pmd, start_addr); |
342cd0ab CD |
139 | } |
140 | ||
4f853a71 CD |
141 | static void unmap_pmds(struct kvm *kvm, pud_t *pud, |
142 | phys_addr_t addr, phys_addr_t end) | |
000d3996 | 143 | { |
4f853a71 CD |
144 | phys_addr_t next, start_addr = addr; |
145 | pmd_t *pmd, *start_pmd; | |
000d3996 | 146 | |
4f853a71 CD |
147 | start_pmd = pmd = pmd_offset(pud, addr); |
148 | do { | |
149 | next = kvm_pmd_addr_end(addr, end); | |
150 | if (!pmd_none(*pmd)) { | |
151 | if (kvm_pmd_huge(*pmd)) { | |
152 | pmd_clear(pmd); | |
153 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
154 | put_page(virt_to_page(pmd)); | |
155 | } else { | |
156 | unmap_ptes(kvm, pmd, addr, next); | |
157 | } | |
ad361f09 | 158 | } |
4f853a71 | 159 | } while (pmd++, addr = next, addr != end); |
ad361f09 | 160 | |
38f791a4 | 161 | if (kvm_pmd_table_empty(kvm, start_pmd)) |
4f853a71 CD |
162 | clear_pud_entry(kvm, pud, start_addr); |
163 | } | |
000d3996 | 164 | |
4f853a71 CD |
165 | static void unmap_puds(struct kvm *kvm, pgd_t *pgd, |
166 | phys_addr_t addr, phys_addr_t end) | |
167 | { | |
168 | phys_addr_t next, start_addr = addr; | |
169 | pud_t *pud, *start_pud; | |
4f728276 | 170 | |
4f853a71 CD |
171 | start_pud = pud = pud_offset(pgd, addr); |
172 | do { | |
173 | next = kvm_pud_addr_end(addr, end); | |
174 | if (!pud_none(*pud)) { | |
175 | if (pud_huge(*pud)) { | |
176 | pud_clear(pud); | |
177 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
178 | put_page(virt_to_page(pud)); | |
179 | } else { | |
180 | unmap_pmds(kvm, pud, addr, next); | |
4f728276 MZ |
181 | } |
182 | } | |
4f853a71 | 183 | } while (pud++, addr = next, addr != end); |
4f728276 | 184 | |
38f791a4 | 185 | if (kvm_pud_table_empty(kvm, start_pud)) |
4f853a71 CD |
186 | clear_pgd_entry(kvm, pgd, start_addr); |
187 | } | |
188 | ||
189 | ||
190 | static void unmap_range(struct kvm *kvm, pgd_t *pgdp, | |
191 | phys_addr_t start, u64 size) | |
192 | { | |
193 | pgd_t *pgd; | |
194 | phys_addr_t addr = start, end = start + size; | |
195 | phys_addr_t next; | |
196 | ||
197 | pgd = pgdp + pgd_index(addr); | |
198 | do { | |
199 | next = kvm_pgd_addr_end(addr, end); | |
7cbb87d6 MR |
200 | if (!pgd_none(*pgd)) |
201 | unmap_puds(kvm, pgd, addr, next); | |
4f853a71 | 202 | } while (pgd++, addr = next, addr != end); |
000d3996 MZ |
203 | } |
204 | ||
9d218a1f MZ |
205 | static void stage2_flush_ptes(struct kvm *kvm, pmd_t *pmd, |
206 | phys_addr_t addr, phys_addr_t end) | |
207 | { | |
208 | pte_t *pte; | |
209 | ||
210 | pte = pte_offset_kernel(pmd, addr); | |
211 | do { | |
212 | if (!pte_none(*pte)) { | |
213 | hva_t hva = gfn_to_hva(kvm, addr >> PAGE_SHIFT); | |
214 | kvm_flush_dcache_to_poc((void*)hva, PAGE_SIZE); | |
215 | } | |
216 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
217 | } | |
218 | ||
219 | static void stage2_flush_pmds(struct kvm *kvm, pud_t *pud, | |
220 | phys_addr_t addr, phys_addr_t end) | |
221 | { | |
222 | pmd_t *pmd; | |
223 | phys_addr_t next; | |
224 | ||
225 | pmd = pmd_offset(pud, addr); | |
226 | do { | |
227 | next = kvm_pmd_addr_end(addr, end); | |
228 | if (!pmd_none(*pmd)) { | |
229 | if (kvm_pmd_huge(*pmd)) { | |
230 | hva_t hva = gfn_to_hva(kvm, addr >> PAGE_SHIFT); | |
231 | kvm_flush_dcache_to_poc((void*)hva, PMD_SIZE); | |
232 | } else { | |
233 | stage2_flush_ptes(kvm, pmd, addr, next); | |
234 | } | |
235 | } | |
236 | } while (pmd++, addr = next, addr != end); | |
237 | } | |
238 | ||
239 | static void stage2_flush_puds(struct kvm *kvm, pgd_t *pgd, | |
240 | phys_addr_t addr, phys_addr_t end) | |
241 | { | |
242 | pud_t *pud; | |
243 | phys_addr_t next; | |
244 | ||
245 | pud = pud_offset(pgd, addr); | |
246 | do { | |
247 | next = kvm_pud_addr_end(addr, end); | |
248 | if (!pud_none(*pud)) { | |
249 | if (pud_huge(*pud)) { | |
250 | hva_t hva = gfn_to_hva(kvm, addr >> PAGE_SHIFT); | |
251 | kvm_flush_dcache_to_poc((void*)hva, PUD_SIZE); | |
252 | } else { | |
253 | stage2_flush_pmds(kvm, pud, addr, next); | |
254 | } | |
255 | } | |
256 | } while (pud++, addr = next, addr != end); | |
257 | } | |
258 | ||
259 | static void stage2_flush_memslot(struct kvm *kvm, | |
260 | struct kvm_memory_slot *memslot) | |
261 | { | |
262 | phys_addr_t addr = memslot->base_gfn << PAGE_SHIFT; | |
263 | phys_addr_t end = addr + PAGE_SIZE * memslot->npages; | |
264 | phys_addr_t next; | |
265 | pgd_t *pgd; | |
266 | ||
267 | pgd = kvm->arch.pgd + pgd_index(addr); | |
268 | do { | |
269 | next = kvm_pgd_addr_end(addr, end); | |
270 | stage2_flush_puds(kvm, pgd, addr, next); | |
271 | } while (pgd++, addr = next, addr != end); | |
272 | } | |
273 | ||
274 | /** | |
275 | * stage2_flush_vm - Invalidate cache for pages mapped in stage 2 | |
276 | * @kvm: The struct kvm pointer | |
277 | * | |
278 | * Go through the stage 2 page tables and invalidate any cache lines | |
279 | * backing memory already mapped to the VM. | |
280 | */ | |
281 | void stage2_flush_vm(struct kvm *kvm) | |
282 | { | |
283 | struct kvm_memslots *slots; | |
284 | struct kvm_memory_slot *memslot; | |
285 | int idx; | |
286 | ||
287 | idx = srcu_read_lock(&kvm->srcu); | |
288 | spin_lock(&kvm->mmu_lock); | |
289 | ||
290 | slots = kvm_memslots(kvm); | |
291 | kvm_for_each_memslot(memslot, slots) | |
292 | stage2_flush_memslot(kvm, memslot); | |
293 | ||
294 | spin_unlock(&kvm->mmu_lock); | |
295 | srcu_read_unlock(&kvm->srcu, idx); | |
296 | } | |
297 | ||
d157f4a5 MZ |
298 | /** |
299 | * free_boot_hyp_pgd - free HYP boot page tables | |
300 | * | |
301 | * Free the HYP boot page tables. The bounce page is also freed. | |
302 | */ | |
303 | void free_boot_hyp_pgd(void) | |
304 | { | |
305 | mutex_lock(&kvm_hyp_pgd_mutex); | |
306 | ||
307 | if (boot_hyp_pgd) { | |
d4cb9df5 MZ |
308 | unmap_range(NULL, boot_hyp_pgd, hyp_idmap_start, PAGE_SIZE); |
309 | unmap_range(NULL, boot_hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE); | |
38f791a4 | 310 | free_pages((unsigned long)boot_hyp_pgd, hyp_pgd_order); |
d157f4a5 MZ |
311 | boot_hyp_pgd = NULL; |
312 | } | |
313 | ||
314 | if (hyp_pgd) | |
d4cb9df5 | 315 | unmap_range(NULL, hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE); |
d157f4a5 | 316 | |
5d4e08c4 | 317 | free_page((unsigned long)init_bounce_page); |
d157f4a5 MZ |
318 | init_bounce_page = NULL; |
319 | ||
320 | mutex_unlock(&kvm_hyp_pgd_mutex); | |
321 | } | |
322 | ||
342cd0ab | 323 | /** |
4f728276 | 324 | * free_hyp_pgds - free Hyp-mode page tables |
342cd0ab | 325 | * |
5a677ce0 MZ |
326 | * Assumes hyp_pgd is a page table used strictly in Hyp-mode and |
327 | * therefore contains either mappings in the kernel memory area (above | |
328 | * PAGE_OFFSET), or device mappings in the vmalloc range (from | |
329 | * VMALLOC_START to VMALLOC_END). | |
330 | * | |
331 | * boot_hyp_pgd should only map two pages for the init code. | |
342cd0ab | 332 | */ |
4f728276 | 333 | void free_hyp_pgds(void) |
342cd0ab | 334 | { |
342cd0ab CD |
335 | unsigned long addr; |
336 | ||
d157f4a5 | 337 | free_boot_hyp_pgd(); |
4f728276 | 338 | |
d157f4a5 | 339 | mutex_lock(&kvm_hyp_pgd_mutex); |
5a677ce0 | 340 | |
4f728276 MZ |
341 | if (hyp_pgd) { |
342 | for (addr = PAGE_OFFSET; virt_addr_valid(addr); addr += PGDIR_SIZE) | |
d4cb9df5 | 343 | unmap_range(NULL, hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE); |
4f728276 | 344 | for (addr = VMALLOC_START; is_vmalloc_addr((void*)addr); addr += PGDIR_SIZE) |
d4cb9df5 MZ |
345 | unmap_range(NULL, hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE); |
346 | ||
38f791a4 | 347 | free_pages((unsigned long)hyp_pgd, hyp_pgd_order); |
d157f4a5 | 348 | hyp_pgd = NULL; |
4f728276 MZ |
349 | } |
350 | ||
342cd0ab CD |
351 | mutex_unlock(&kvm_hyp_pgd_mutex); |
352 | } | |
353 | ||
354 | static void create_hyp_pte_mappings(pmd_t *pmd, unsigned long start, | |
6060df84 MZ |
355 | unsigned long end, unsigned long pfn, |
356 | pgprot_t prot) | |
342cd0ab CD |
357 | { |
358 | pte_t *pte; | |
359 | unsigned long addr; | |
342cd0ab | 360 | |
3562c76d MZ |
361 | addr = start; |
362 | do { | |
6060df84 MZ |
363 | pte = pte_offset_kernel(pmd, addr); |
364 | kvm_set_pte(pte, pfn_pte(pfn, prot)); | |
4f728276 | 365 | get_page(virt_to_page(pte)); |
5a677ce0 | 366 | kvm_flush_dcache_to_poc(pte, sizeof(*pte)); |
6060df84 | 367 | pfn++; |
3562c76d | 368 | } while (addr += PAGE_SIZE, addr != end); |
342cd0ab CD |
369 | } |
370 | ||
371 | static int create_hyp_pmd_mappings(pud_t *pud, unsigned long start, | |
6060df84 MZ |
372 | unsigned long end, unsigned long pfn, |
373 | pgprot_t prot) | |
342cd0ab CD |
374 | { |
375 | pmd_t *pmd; | |
376 | pte_t *pte; | |
377 | unsigned long addr, next; | |
378 | ||
3562c76d MZ |
379 | addr = start; |
380 | do { | |
6060df84 | 381 | pmd = pmd_offset(pud, addr); |
342cd0ab CD |
382 | |
383 | BUG_ON(pmd_sect(*pmd)); | |
384 | ||
385 | if (pmd_none(*pmd)) { | |
6060df84 | 386 | pte = pte_alloc_one_kernel(NULL, addr); |
342cd0ab CD |
387 | if (!pte) { |
388 | kvm_err("Cannot allocate Hyp pte\n"); | |
389 | return -ENOMEM; | |
390 | } | |
391 | pmd_populate_kernel(NULL, pmd, pte); | |
4f728276 | 392 | get_page(virt_to_page(pmd)); |
5a677ce0 | 393 | kvm_flush_dcache_to_poc(pmd, sizeof(*pmd)); |
342cd0ab CD |
394 | } |
395 | ||
396 | next = pmd_addr_end(addr, end); | |
397 | ||
6060df84 MZ |
398 | create_hyp_pte_mappings(pmd, addr, next, pfn, prot); |
399 | pfn += (next - addr) >> PAGE_SHIFT; | |
3562c76d | 400 | } while (addr = next, addr != end); |
342cd0ab CD |
401 | |
402 | return 0; | |
403 | } | |
404 | ||
38f791a4 CD |
405 | static int create_hyp_pud_mappings(pgd_t *pgd, unsigned long start, |
406 | unsigned long end, unsigned long pfn, | |
407 | pgprot_t prot) | |
408 | { | |
409 | pud_t *pud; | |
410 | pmd_t *pmd; | |
411 | unsigned long addr, next; | |
412 | int ret; | |
413 | ||
414 | addr = start; | |
415 | do { | |
416 | pud = pud_offset(pgd, addr); | |
417 | ||
418 | if (pud_none_or_clear_bad(pud)) { | |
419 | pmd = pmd_alloc_one(NULL, addr); | |
420 | if (!pmd) { | |
421 | kvm_err("Cannot allocate Hyp pmd\n"); | |
422 | return -ENOMEM; | |
423 | } | |
424 | pud_populate(NULL, pud, pmd); | |
425 | get_page(virt_to_page(pud)); | |
426 | kvm_flush_dcache_to_poc(pud, sizeof(*pud)); | |
427 | } | |
428 | ||
429 | next = pud_addr_end(addr, end); | |
430 | ret = create_hyp_pmd_mappings(pud, addr, next, pfn, prot); | |
431 | if (ret) | |
432 | return ret; | |
433 | pfn += (next - addr) >> PAGE_SHIFT; | |
434 | } while (addr = next, addr != end); | |
435 | ||
436 | return 0; | |
437 | } | |
438 | ||
6060df84 MZ |
439 | static int __create_hyp_mappings(pgd_t *pgdp, |
440 | unsigned long start, unsigned long end, | |
441 | unsigned long pfn, pgprot_t prot) | |
342cd0ab | 442 | { |
342cd0ab CD |
443 | pgd_t *pgd; |
444 | pud_t *pud; | |
342cd0ab CD |
445 | unsigned long addr, next; |
446 | int err = 0; | |
447 | ||
342cd0ab | 448 | mutex_lock(&kvm_hyp_pgd_mutex); |
3562c76d MZ |
449 | addr = start & PAGE_MASK; |
450 | end = PAGE_ALIGN(end); | |
451 | do { | |
6060df84 | 452 | pgd = pgdp + pgd_index(addr); |
342cd0ab | 453 | |
38f791a4 CD |
454 | if (pgd_none(*pgd)) { |
455 | pud = pud_alloc_one(NULL, addr); | |
456 | if (!pud) { | |
457 | kvm_err("Cannot allocate Hyp pud\n"); | |
342cd0ab CD |
458 | err = -ENOMEM; |
459 | goto out; | |
460 | } | |
38f791a4 CD |
461 | pgd_populate(NULL, pgd, pud); |
462 | get_page(virt_to_page(pgd)); | |
463 | kvm_flush_dcache_to_poc(pgd, sizeof(*pgd)); | |
342cd0ab CD |
464 | } |
465 | ||
466 | next = pgd_addr_end(addr, end); | |
38f791a4 | 467 | err = create_hyp_pud_mappings(pgd, addr, next, pfn, prot); |
342cd0ab CD |
468 | if (err) |
469 | goto out; | |
6060df84 | 470 | pfn += (next - addr) >> PAGE_SHIFT; |
3562c76d | 471 | } while (addr = next, addr != end); |
342cd0ab CD |
472 | out: |
473 | mutex_unlock(&kvm_hyp_pgd_mutex); | |
474 | return err; | |
475 | } | |
476 | ||
40c2729b CD |
477 | static phys_addr_t kvm_kaddr_to_phys(void *kaddr) |
478 | { | |
479 | if (!is_vmalloc_addr(kaddr)) { | |
480 | BUG_ON(!virt_addr_valid(kaddr)); | |
481 | return __pa(kaddr); | |
482 | } else { | |
483 | return page_to_phys(vmalloc_to_page(kaddr)) + | |
484 | offset_in_page(kaddr); | |
485 | } | |
486 | } | |
487 | ||
342cd0ab | 488 | /** |
06e8c3b0 | 489 | * create_hyp_mappings - duplicate a kernel virtual address range in Hyp mode |
342cd0ab CD |
490 | * @from: The virtual kernel start address of the range |
491 | * @to: The virtual kernel end address of the range (exclusive) | |
492 | * | |
06e8c3b0 MZ |
493 | * The same virtual address as the kernel virtual address is also used |
494 | * in Hyp-mode mapping (modulo HYP_PAGE_OFFSET) to the same underlying | |
495 | * physical pages. | |
342cd0ab CD |
496 | */ |
497 | int create_hyp_mappings(void *from, void *to) | |
498 | { | |
40c2729b CD |
499 | phys_addr_t phys_addr; |
500 | unsigned long virt_addr; | |
6060df84 MZ |
501 | unsigned long start = KERN_TO_HYP((unsigned long)from); |
502 | unsigned long end = KERN_TO_HYP((unsigned long)to); | |
503 | ||
40c2729b CD |
504 | start = start & PAGE_MASK; |
505 | end = PAGE_ALIGN(end); | |
6060df84 | 506 | |
40c2729b CD |
507 | for (virt_addr = start; virt_addr < end; virt_addr += PAGE_SIZE) { |
508 | int err; | |
6060df84 | 509 | |
40c2729b CD |
510 | phys_addr = kvm_kaddr_to_phys(from + virt_addr - start); |
511 | err = __create_hyp_mappings(hyp_pgd, virt_addr, | |
512 | virt_addr + PAGE_SIZE, | |
513 | __phys_to_pfn(phys_addr), | |
514 | PAGE_HYP); | |
515 | if (err) | |
516 | return err; | |
517 | } | |
518 | ||
519 | return 0; | |
342cd0ab CD |
520 | } |
521 | ||
522 | /** | |
06e8c3b0 MZ |
523 | * create_hyp_io_mappings - duplicate a kernel IO mapping into Hyp mode |
524 | * @from: The kernel start VA of the range | |
525 | * @to: The kernel end VA of the range (exclusive) | |
6060df84 | 526 | * @phys_addr: The physical start address which gets mapped |
06e8c3b0 MZ |
527 | * |
528 | * The resulting HYP VA is the same as the kernel VA, modulo | |
529 | * HYP_PAGE_OFFSET. | |
342cd0ab | 530 | */ |
6060df84 | 531 | int create_hyp_io_mappings(void *from, void *to, phys_addr_t phys_addr) |
342cd0ab | 532 | { |
6060df84 MZ |
533 | unsigned long start = KERN_TO_HYP((unsigned long)from); |
534 | unsigned long end = KERN_TO_HYP((unsigned long)to); | |
535 | ||
536 | /* Check for a valid kernel IO mapping */ | |
537 | if (!is_vmalloc_addr(from) || !is_vmalloc_addr(to - 1)) | |
538 | return -EINVAL; | |
539 | ||
540 | return __create_hyp_mappings(hyp_pgd, start, end, | |
541 | __phys_to_pfn(phys_addr), PAGE_HYP_DEVICE); | |
342cd0ab CD |
542 | } |
543 | ||
d5d8184d CD |
544 | /** |
545 | * kvm_alloc_stage2_pgd - allocate level-1 table for stage-2 translation. | |
546 | * @kvm: The KVM struct pointer for the VM. | |
547 | * | |
548 | * Allocates the 1st level table only of size defined by S2_PGD_ORDER (can | |
549 | * support either full 40-bit input addresses or limited to 32-bit input | |
550 | * addresses). Clears the allocated pages. | |
551 | * | |
552 | * Note we don't need locking here as this is only called when the VM is | |
553 | * created, which can only be done once. | |
554 | */ | |
555 | int kvm_alloc_stage2_pgd(struct kvm *kvm) | |
556 | { | |
38f791a4 | 557 | int ret; |
d5d8184d CD |
558 | pgd_t *pgd; |
559 | ||
560 | if (kvm->arch.pgd != NULL) { | |
561 | kvm_err("kvm_arch already initialized?\n"); | |
562 | return -EINVAL; | |
563 | } | |
564 | ||
38f791a4 CD |
565 | if (KVM_PREALLOC_LEVEL > 0) { |
566 | /* | |
567 | * Allocate fake pgd for the page table manipulation macros to | |
568 | * work. This is not used by the hardware and we have no | |
569 | * alignment requirement for this allocation. | |
570 | */ | |
571 | pgd = (pgd_t *)kmalloc(PTRS_PER_S2_PGD * sizeof(pgd_t), | |
572 | GFP_KERNEL | __GFP_ZERO); | |
573 | } else { | |
574 | /* | |
575 | * Allocate actual first-level Stage-2 page table used by the | |
576 | * hardware for Stage-2 page table walks. | |
577 | */ | |
578 | pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, S2_PGD_ORDER); | |
579 | } | |
580 | ||
d5d8184d CD |
581 | if (!pgd) |
582 | return -ENOMEM; | |
583 | ||
38f791a4 CD |
584 | ret = kvm_prealloc_hwpgd(kvm, pgd); |
585 | if (ret) | |
586 | goto out_err; | |
587 | ||
c62ee2b2 | 588 | kvm_clean_pgd(pgd); |
d5d8184d | 589 | kvm->arch.pgd = pgd; |
d5d8184d | 590 | return 0; |
38f791a4 CD |
591 | out_err: |
592 | if (KVM_PREALLOC_LEVEL > 0) | |
593 | kfree(pgd); | |
594 | else | |
595 | free_pages((unsigned long)pgd, S2_PGD_ORDER); | |
596 | return ret; | |
d5d8184d CD |
597 | } |
598 | ||
d5d8184d CD |
599 | /** |
600 | * unmap_stage2_range -- Clear stage2 page table entries to unmap a range | |
601 | * @kvm: The VM pointer | |
602 | * @start: The intermediate physical base address of the range to unmap | |
603 | * @size: The size of the area to unmap | |
604 | * | |
605 | * Clear a range of stage-2 mappings, lowering the various ref-counts. Must | |
606 | * be called while holding mmu_lock (unless for freeing the stage2 pgd before | |
607 | * destroying the VM), otherwise another faulting VCPU may come in and mess | |
608 | * with things behind our backs. | |
609 | */ | |
610 | static void unmap_stage2_range(struct kvm *kvm, phys_addr_t start, u64 size) | |
611 | { | |
d4cb9df5 | 612 | unmap_range(kvm, kvm->arch.pgd, start, size); |
d5d8184d CD |
613 | } |
614 | ||
615 | /** | |
616 | * kvm_free_stage2_pgd - free all stage-2 tables | |
617 | * @kvm: The KVM struct pointer for the VM. | |
618 | * | |
619 | * Walks the level-1 page table pointed to by kvm->arch.pgd and frees all | |
620 | * underlying level-2 and level-3 tables before freeing the actual level-1 table | |
621 | * and setting the struct pointer to NULL. | |
622 | * | |
623 | * Note we don't need locking here as this is only called when the VM is | |
624 | * destroyed, which can only be done once. | |
625 | */ | |
626 | void kvm_free_stage2_pgd(struct kvm *kvm) | |
627 | { | |
628 | if (kvm->arch.pgd == NULL) | |
629 | return; | |
630 | ||
631 | unmap_stage2_range(kvm, 0, KVM_PHYS_SIZE); | |
38f791a4 CD |
632 | kvm_free_hwpgd(kvm); |
633 | if (KVM_PREALLOC_LEVEL > 0) | |
634 | kfree(kvm->arch.pgd); | |
635 | else | |
636 | free_pages((unsigned long)kvm->arch.pgd, S2_PGD_ORDER); | |
d5d8184d CD |
637 | kvm->arch.pgd = NULL; |
638 | } | |
639 | ||
38f791a4 | 640 | static pud_t *stage2_get_pud(struct kvm *kvm, struct kvm_mmu_memory_cache *cache, |
ad361f09 | 641 | phys_addr_t addr) |
d5d8184d CD |
642 | { |
643 | pgd_t *pgd; | |
644 | pud_t *pud; | |
d5d8184d | 645 | |
d5d8184d | 646 | pgd = kvm->arch.pgd + pgd_index(addr); |
38f791a4 CD |
647 | if (WARN_ON(pgd_none(*pgd))) { |
648 | if (!cache) | |
649 | return NULL; | |
650 | pud = mmu_memory_cache_alloc(cache); | |
651 | pgd_populate(NULL, pgd, pud); | |
652 | get_page(virt_to_page(pgd)); | |
653 | } | |
654 | ||
655 | return pud_offset(pgd, addr); | |
656 | } | |
657 | ||
658 | static pmd_t *stage2_get_pmd(struct kvm *kvm, struct kvm_mmu_memory_cache *cache, | |
659 | phys_addr_t addr) | |
660 | { | |
661 | pud_t *pud; | |
662 | pmd_t *pmd; | |
663 | ||
664 | pud = stage2_get_pud(kvm, cache, addr); | |
d5d8184d CD |
665 | if (pud_none(*pud)) { |
666 | if (!cache) | |
ad361f09 | 667 | return NULL; |
d5d8184d CD |
668 | pmd = mmu_memory_cache_alloc(cache); |
669 | pud_populate(NULL, pud, pmd); | |
d5d8184d | 670 | get_page(virt_to_page(pud)); |
c62ee2b2 MZ |
671 | } |
672 | ||
ad361f09 CD |
673 | return pmd_offset(pud, addr); |
674 | } | |
675 | ||
676 | static int stage2_set_pmd_huge(struct kvm *kvm, struct kvm_mmu_memory_cache | |
677 | *cache, phys_addr_t addr, const pmd_t *new_pmd) | |
678 | { | |
679 | pmd_t *pmd, old_pmd; | |
680 | ||
681 | pmd = stage2_get_pmd(kvm, cache, addr); | |
682 | VM_BUG_ON(!pmd); | |
d5d8184d | 683 | |
ad361f09 CD |
684 | /* |
685 | * Mapping in huge pages should only happen through a fault. If a | |
686 | * page is merged into a transparent huge page, the individual | |
687 | * subpages of that huge page should be unmapped through MMU | |
688 | * notifiers before we get here. | |
689 | * | |
690 | * Merging of CompoundPages is not supported; they should become | |
691 | * splitting first, unmapped, merged, and mapped back in on-demand. | |
692 | */ | |
693 | VM_BUG_ON(pmd_present(*pmd) && pmd_pfn(*pmd) != pmd_pfn(*new_pmd)); | |
694 | ||
695 | old_pmd = *pmd; | |
696 | kvm_set_pmd(pmd, *new_pmd); | |
697 | if (pmd_present(old_pmd)) | |
698 | kvm_tlb_flush_vmid_ipa(kvm, addr); | |
699 | else | |
700 | get_page(virt_to_page(pmd)); | |
701 | return 0; | |
702 | } | |
703 | ||
704 | static int stage2_set_pte(struct kvm *kvm, struct kvm_mmu_memory_cache *cache, | |
705 | phys_addr_t addr, const pte_t *new_pte, bool iomap) | |
706 | { | |
707 | pmd_t *pmd; | |
708 | pte_t *pte, old_pte; | |
709 | ||
38f791a4 | 710 | /* Create stage-2 page table mapping - Levels 0 and 1 */ |
ad361f09 CD |
711 | pmd = stage2_get_pmd(kvm, cache, addr); |
712 | if (!pmd) { | |
713 | /* | |
714 | * Ignore calls from kvm_set_spte_hva for unallocated | |
715 | * address ranges. | |
716 | */ | |
717 | return 0; | |
718 | } | |
719 | ||
720 | /* Create stage-2 page mappings - Level 2 */ | |
d5d8184d CD |
721 | if (pmd_none(*pmd)) { |
722 | if (!cache) | |
723 | return 0; /* ignore calls from kvm_set_spte_hva */ | |
724 | pte = mmu_memory_cache_alloc(cache); | |
c62ee2b2 | 725 | kvm_clean_pte(pte); |
d5d8184d | 726 | pmd_populate_kernel(NULL, pmd, pte); |
d5d8184d | 727 | get_page(virt_to_page(pmd)); |
c62ee2b2 MZ |
728 | } |
729 | ||
730 | pte = pte_offset_kernel(pmd, addr); | |
d5d8184d CD |
731 | |
732 | if (iomap && pte_present(*pte)) | |
733 | return -EFAULT; | |
734 | ||
735 | /* Create 2nd stage page table mapping - Level 3 */ | |
736 | old_pte = *pte; | |
737 | kvm_set_pte(pte, *new_pte); | |
738 | if (pte_present(old_pte)) | |
48762767 | 739 | kvm_tlb_flush_vmid_ipa(kvm, addr); |
d5d8184d CD |
740 | else |
741 | get_page(virt_to_page(pte)); | |
742 | ||
743 | return 0; | |
744 | } | |
745 | ||
746 | /** | |
747 | * kvm_phys_addr_ioremap - map a device range to guest IPA | |
748 | * | |
749 | * @kvm: The KVM pointer | |
750 | * @guest_ipa: The IPA at which to insert the mapping | |
751 | * @pa: The physical address of the device | |
752 | * @size: The size of the mapping | |
753 | */ | |
754 | int kvm_phys_addr_ioremap(struct kvm *kvm, phys_addr_t guest_ipa, | |
c40f2f8f | 755 | phys_addr_t pa, unsigned long size, bool writable) |
d5d8184d CD |
756 | { |
757 | phys_addr_t addr, end; | |
758 | int ret = 0; | |
759 | unsigned long pfn; | |
760 | struct kvm_mmu_memory_cache cache = { 0, }; | |
761 | ||
762 | end = (guest_ipa + size + PAGE_SIZE - 1) & PAGE_MASK; | |
763 | pfn = __phys_to_pfn(pa); | |
764 | ||
765 | for (addr = guest_ipa; addr < end; addr += PAGE_SIZE) { | |
c62ee2b2 | 766 | pte_t pte = pfn_pte(pfn, PAGE_S2_DEVICE); |
d5d8184d | 767 | |
c40f2f8f AB |
768 | if (writable) |
769 | kvm_set_s2pte_writable(&pte); | |
770 | ||
38f791a4 CD |
771 | ret = mmu_topup_memory_cache(&cache, KVM_MMU_CACHE_MIN_PAGES, |
772 | KVM_NR_MEM_OBJS); | |
d5d8184d CD |
773 | if (ret) |
774 | goto out; | |
775 | spin_lock(&kvm->mmu_lock); | |
776 | ret = stage2_set_pte(kvm, &cache, addr, &pte, true); | |
777 | spin_unlock(&kvm->mmu_lock); | |
778 | if (ret) | |
779 | goto out; | |
780 | ||
781 | pfn++; | |
782 | } | |
783 | ||
784 | out: | |
785 | mmu_free_memory_cache(&cache); | |
786 | return ret; | |
787 | } | |
788 | ||
9b5fdb97 CD |
789 | static bool transparent_hugepage_adjust(pfn_t *pfnp, phys_addr_t *ipap) |
790 | { | |
791 | pfn_t pfn = *pfnp; | |
792 | gfn_t gfn = *ipap >> PAGE_SHIFT; | |
793 | ||
794 | if (PageTransCompound(pfn_to_page(pfn))) { | |
795 | unsigned long mask; | |
796 | /* | |
797 | * The address we faulted on is backed by a transparent huge | |
798 | * page. However, because we map the compound huge page and | |
799 | * not the individual tail page, we need to transfer the | |
800 | * refcount to the head page. We have to be careful that the | |
801 | * THP doesn't start to split while we are adjusting the | |
802 | * refcounts. | |
803 | * | |
804 | * We are sure this doesn't happen, because mmu_notifier_retry | |
805 | * was successful and we are holding the mmu_lock, so if this | |
806 | * THP is trying to split, it will be blocked in the mmu | |
807 | * notifier before touching any of the pages, specifically | |
808 | * before being able to call __split_huge_page_refcount(). | |
809 | * | |
810 | * We can therefore safely transfer the refcount from PG_tail | |
811 | * to PG_head and switch the pfn from a tail page to the head | |
812 | * page accordingly. | |
813 | */ | |
814 | mask = PTRS_PER_PMD - 1; | |
815 | VM_BUG_ON((gfn & mask) != (pfn & mask)); | |
816 | if (pfn & mask) { | |
817 | *ipap &= PMD_MASK; | |
818 | kvm_release_pfn_clean(pfn); | |
819 | pfn &= ~mask; | |
820 | kvm_get_pfn(pfn); | |
821 | *pfnp = pfn; | |
822 | } | |
823 | ||
824 | return true; | |
825 | } | |
826 | ||
827 | return false; | |
828 | } | |
829 | ||
a7d079ce AB |
830 | static bool kvm_is_write_fault(struct kvm_vcpu *vcpu) |
831 | { | |
832 | if (kvm_vcpu_trap_is_iabt(vcpu)) | |
833 | return false; | |
834 | ||
835 | return kvm_vcpu_dabt_iswrite(vcpu); | |
836 | } | |
837 | ||
bb55e9b1 AB |
838 | static bool kvm_is_device_pfn(unsigned long pfn) |
839 | { | |
840 | return !pfn_valid(pfn); | |
841 | } | |
842 | ||
94f8e641 | 843 | static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa, |
98047888 | 844 | struct kvm_memory_slot *memslot, unsigned long hva, |
94f8e641 CD |
845 | unsigned long fault_status) |
846 | { | |
94f8e641 | 847 | int ret; |
9b5fdb97 | 848 | bool write_fault, writable, hugetlb = false, force_pte = false; |
94f8e641 | 849 | unsigned long mmu_seq; |
ad361f09 | 850 | gfn_t gfn = fault_ipa >> PAGE_SHIFT; |
ad361f09 | 851 | struct kvm *kvm = vcpu->kvm; |
94f8e641 | 852 | struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache; |
ad361f09 CD |
853 | struct vm_area_struct *vma; |
854 | pfn_t pfn; | |
b8865767 | 855 | pgprot_t mem_type = PAGE_S2; |
94f8e641 | 856 | |
a7d079ce | 857 | write_fault = kvm_is_write_fault(vcpu); |
94f8e641 CD |
858 | if (fault_status == FSC_PERM && !write_fault) { |
859 | kvm_err("Unexpected L2 read permission error\n"); | |
860 | return -EFAULT; | |
861 | } | |
862 | ||
ad361f09 CD |
863 | /* Let's check if we will get back a huge page backed by hugetlbfs */ |
864 | down_read(¤t->mm->mmap_sem); | |
865 | vma = find_vma_intersection(current->mm, hva, hva + 1); | |
37b54408 AB |
866 | if (unlikely(!vma)) { |
867 | kvm_err("Failed to find VMA for hva 0x%lx\n", hva); | |
868 | up_read(¤t->mm->mmap_sem); | |
869 | return -EFAULT; | |
870 | } | |
871 | ||
ad361f09 CD |
872 | if (is_vm_hugetlb_page(vma)) { |
873 | hugetlb = true; | |
874 | gfn = (fault_ipa & PMD_MASK) >> PAGE_SHIFT; | |
9b5fdb97 CD |
875 | } else { |
876 | /* | |
136d737f MZ |
877 | * Pages belonging to memslots that don't have the same |
878 | * alignment for userspace and IPA cannot be mapped using | |
879 | * block descriptors even if the pages belong to a THP for | |
880 | * the process, because the stage-2 block descriptor will | |
881 | * cover more than a single THP and we loose atomicity for | |
882 | * unmapping, updates, and splits of the THP or other pages | |
883 | * in the stage-2 block range. | |
9b5fdb97 | 884 | */ |
136d737f MZ |
885 | if ((memslot->userspace_addr & ~PMD_MASK) != |
886 | ((memslot->base_gfn << PAGE_SHIFT) & ~PMD_MASK)) | |
9b5fdb97 | 887 | force_pte = true; |
ad361f09 CD |
888 | } |
889 | up_read(¤t->mm->mmap_sem); | |
890 | ||
94f8e641 | 891 | /* We need minimum second+third level pages */ |
38f791a4 CD |
892 | ret = mmu_topup_memory_cache(memcache, KVM_MMU_CACHE_MIN_PAGES, |
893 | KVM_NR_MEM_OBJS); | |
94f8e641 CD |
894 | if (ret) |
895 | return ret; | |
896 | ||
897 | mmu_seq = vcpu->kvm->mmu_notifier_seq; | |
898 | /* | |
899 | * Ensure the read of mmu_notifier_seq happens before we call | |
900 | * gfn_to_pfn_prot (which calls get_user_pages), so that we don't risk | |
901 | * the page we just got a reference to gets unmapped before we have a | |
902 | * chance to grab the mmu_lock, which ensure that if the page gets | |
903 | * unmapped afterwards, the call to kvm_unmap_hva will take it away | |
904 | * from us again properly. This smp_rmb() interacts with the smp_wmb() | |
905 | * in kvm_mmu_notifier_invalidate_<page|range_end>. | |
906 | */ | |
907 | smp_rmb(); | |
908 | ||
ad361f09 | 909 | pfn = gfn_to_pfn_prot(kvm, gfn, write_fault, &writable); |
94f8e641 CD |
910 | if (is_error_pfn(pfn)) |
911 | return -EFAULT; | |
912 | ||
bb55e9b1 | 913 | if (kvm_is_device_pfn(pfn)) |
b8865767 KP |
914 | mem_type = PAGE_S2_DEVICE; |
915 | ||
ad361f09 CD |
916 | spin_lock(&kvm->mmu_lock); |
917 | if (mmu_notifier_retry(kvm, mmu_seq)) | |
94f8e641 | 918 | goto out_unlock; |
9b5fdb97 CD |
919 | if (!hugetlb && !force_pte) |
920 | hugetlb = transparent_hugepage_adjust(&pfn, &fault_ipa); | |
ad361f09 CD |
921 | |
922 | if (hugetlb) { | |
b8865767 | 923 | pmd_t new_pmd = pfn_pmd(pfn, mem_type); |
ad361f09 CD |
924 | new_pmd = pmd_mkhuge(new_pmd); |
925 | if (writable) { | |
926 | kvm_set_s2pmd_writable(&new_pmd); | |
927 | kvm_set_pfn_dirty(pfn); | |
928 | } | |
2d58b733 | 929 | coherent_cache_guest_page(vcpu, hva & PMD_MASK, PMD_SIZE); |
ad361f09 CD |
930 | ret = stage2_set_pmd_huge(kvm, memcache, fault_ipa, &new_pmd); |
931 | } else { | |
b8865767 | 932 | pte_t new_pte = pfn_pte(pfn, mem_type); |
ad361f09 CD |
933 | if (writable) { |
934 | kvm_set_s2pte_writable(&new_pte); | |
935 | kvm_set_pfn_dirty(pfn); | |
936 | } | |
2d58b733 | 937 | coherent_cache_guest_page(vcpu, hva, PAGE_SIZE); |
b8865767 | 938 | ret = stage2_set_pte(kvm, memcache, fault_ipa, &new_pte, |
3d08c629 | 939 | pgprot_val(mem_type) == pgprot_val(PAGE_S2_DEVICE)); |
94f8e641 | 940 | } |
ad361f09 | 941 | |
94f8e641 CD |
942 | |
943 | out_unlock: | |
ad361f09 | 944 | spin_unlock(&kvm->mmu_lock); |
94f8e641 | 945 | kvm_release_pfn_clean(pfn); |
ad361f09 | 946 | return ret; |
94f8e641 CD |
947 | } |
948 | ||
949 | /** | |
950 | * kvm_handle_guest_abort - handles all 2nd stage aborts | |
951 | * @vcpu: the VCPU pointer | |
952 | * @run: the kvm_run structure | |
953 | * | |
954 | * Any abort that gets to the host is almost guaranteed to be caused by a | |
955 | * missing second stage translation table entry, which can mean that either the | |
956 | * guest simply needs more memory and we must allocate an appropriate page or it | |
957 | * can mean that the guest tried to access I/O memory, which is emulated by user | |
958 | * space. The distinction is based on the IPA causing the fault and whether this | |
959 | * memory region has been registered as standard RAM by user space. | |
960 | */ | |
342cd0ab CD |
961 | int kvm_handle_guest_abort(struct kvm_vcpu *vcpu, struct kvm_run *run) |
962 | { | |
94f8e641 CD |
963 | unsigned long fault_status; |
964 | phys_addr_t fault_ipa; | |
965 | struct kvm_memory_slot *memslot; | |
98047888 CD |
966 | unsigned long hva; |
967 | bool is_iabt, write_fault, writable; | |
94f8e641 CD |
968 | gfn_t gfn; |
969 | int ret, idx; | |
970 | ||
52d1dba9 | 971 | is_iabt = kvm_vcpu_trap_is_iabt(vcpu); |
7393b599 | 972 | fault_ipa = kvm_vcpu_get_fault_ipa(vcpu); |
94f8e641 | 973 | |
7393b599 MZ |
974 | trace_kvm_guest_fault(*vcpu_pc(vcpu), kvm_vcpu_get_hsr(vcpu), |
975 | kvm_vcpu_get_hfar(vcpu), fault_ipa); | |
94f8e641 CD |
976 | |
977 | /* Check the stage-2 fault is trans. fault or write fault */ | |
0496daa5 | 978 | fault_status = kvm_vcpu_trap_get_fault_type(vcpu); |
94f8e641 | 979 | if (fault_status != FSC_FAULT && fault_status != FSC_PERM) { |
0496daa5 CD |
980 | kvm_err("Unsupported FSC: EC=%#x xFSC=%#lx ESR_EL2=%#lx\n", |
981 | kvm_vcpu_trap_get_class(vcpu), | |
982 | (unsigned long)kvm_vcpu_trap_get_fault(vcpu), | |
983 | (unsigned long)kvm_vcpu_get_hsr(vcpu)); | |
94f8e641 CD |
984 | return -EFAULT; |
985 | } | |
986 | ||
987 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
988 | ||
989 | gfn = fault_ipa >> PAGE_SHIFT; | |
98047888 CD |
990 | memslot = gfn_to_memslot(vcpu->kvm, gfn); |
991 | hva = gfn_to_hva_memslot_prot(memslot, gfn, &writable); | |
a7d079ce | 992 | write_fault = kvm_is_write_fault(vcpu); |
98047888 | 993 | if (kvm_is_error_hva(hva) || (write_fault && !writable)) { |
94f8e641 CD |
994 | if (is_iabt) { |
995 | /* Prefetch Abort on I/O address */ | |
7393b599 | 996 | kvm_inject_pabt(vcpu, kvm_vcpu_get_hfar(vcpu)); |
94f8e641 CD |
997 | ret = 1; |
998 | goto out_unlock; | |
999 | } | |
1000 | ||
cfe3950c MZ |
1001 | /* |
1002 | * The IPA is reported as [MAX:12], so we need to | |
1003 | * complement it with the bottom 12 bits from the | |
1004 | * faulting VA. This is always 12 bits, irrespective | |
1005 | * of the page size. | |
1006 | */ | |
1007 | fault_ipa |= kvm_vcpu_get_hfar(vcpu) & ((1 << 12) - 1); | |
45e96ea6 | 1008 | ret = io_mem_abort(vcpu, run, fault_ipa); |
94f8e641 CD |
1009 | goto out_unlock; |
1010 | } | |
1011 | ||
c3058d5d CD |
1012 | /* Userspace should not be able to register out-of-bounds IPAs */ |
1013 | VM_BUG_ON(fault_ipa >= KVM_PHYS_SIZE); | |
1014 | ||
98047888 | 1015 | ret = user_mem_abort(vcpu, fault_ipa, memslot, hva, fault_status); |
94f8e641 CD |
1016 | if (ret == 0) |
1017 | ret = 1; | |
1018 | out_unlock: | |
1019 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
1020 | return ret; | |
342cd0ab CD |
1021 | } |
1022 | ||
d5d8184d CD |
1023 | static void handle_hva_to_gpa(struct kvm *kvm, |
1024 | unsigned long start, | |
1025 | unsigned long end, | |
1026 | void (*handler)(struct kvm *kvm, | |
1027 | gpa_t gpa, void *data), | |
1028 | void *data) | |
1029 | { | |
1030 | struct kvm_memslots *slots; | |
1031 | struct kvm_memory_slot *memslot; | |
1032 | ||
1033 | slots = kvm_memslots(kvm); | |
1034 | ||
1035 | /* we only care about the pages that the guest sees */ | |
1036 | kvm_for_each_memslot(memslot, slots) { | |
1037 | unsigned long hva_start, hva_end; | |
1038 | gfn_t gfn, gfn_end; | |
1039 | ||
1040 | hva_start = max(start, memslot->userspace_addr); | |
1041 | hva_end = min(end, memslot->userspace_addr + | |
1042 | (memslot->npages << PAGE_SHIFT)); | |
1043 | if (hva_start >= hva_end) | |
1044 | continue; | |
1045 | ||
1046 | /* | |
1047 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
1048 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
1049 | */ | |
1050 | gfn = hva_to_gfn_memslot(hva_start, memslot); | |
1051 | gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot); | |
1052 | ||
1053 | for (; gfn < gfn_end; ++gfn) { | |
1054 | gpa_t gpa = gfn << PAGE_SHIFT; | |
1055 | handler(kvm, gpa, data); | |
1056 | } | |
1057 | } | |
1058 | } | |
1059 | ||
1060 | static void kvm_unmap_hva_handler(struct kvm *kvm, gpa_t gpa, void *data) | |
1061 | { | |
1062 | unmap_stage2_range(kvm, gpa, PAGE_SIZE); | |
d5d8184d CD |
1063 | } |
1064 | ||
1065 | int kvm_unmap_hva(struct kvm *kvm, unsigned long hva) | |
1066 | { | |
1067 | unsigned long end = hva + PAGE_SIZE; | |
1068 | ||
1069 | if (!kvm->arch.pgd) | |
1070 | return 0; | |
1071 | ||
1072 | trace_kvm_unmap_hva(hva); | |
1073 | handle_hva_to_gpa(kvm, hva, end, &kvm_unmap_hva_handler, NULL); | |
1074 | return 0; | |
1075 | } | |
1076 | ||
1077 | int kvm_unmap_hva_range(struct kvm *kvm, | |
1078 | unsigned long start, unsigned long end) | |
1079 | { | |
1080 | if (!kvm->arch.pgd) | |
1081 | return 0; | |
1082 | ||
1083 | trace_kvm_unmap_hva_range(start, end); | |
1084 | handle_hva_to_gpa(kvm, start, end, &kvm_unmap_hva_handler, NULL); | |
1085 | return 0; | |
1086 | } | |
1087 | ||
1088 | static void kvm_set_spte_handler(struct kvm *kvm, gpa_t gpa, void *data) | |
1089 | { | |
1090 | pte_t *pte = (pte_t *)data; | |
1091 | ||
1092 | stage2_set_pte(kvm, NULL, gpa, pte, false); | |
1093 | } | |
1094 | ||
1095 | ||
1096 | void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte) | |
1097 | { | |
1098 | unsigned long end = hva + PAGE_SIZE; | |
1099 | pte_t stage2_pte; | |
1100 | ||
1101 | if (!kvm->arch.pgd) | |
1102 | return; | |
1103 | ||
1104 | trace_kvm_set_spte_hva(hva); | |
1105 | stage2_pte = pfn_pte(pte_pfn(pte), PAGE_S2); | |
1106 | handle_hva_to_gpa(kvm, hva, end, &kvm_set_spte_handler, &stage2_pte); | |
1107 | } | |
1108 | ||
1109 | void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu) | |
1110 | { | |
1111 | mmu_free_memory_cache(&vcpu->arch.mmu_page_cache); | |
1112 | } | |
1113 | ||
342cd0ab CD |
1114 | phys_addr_t kvm_mmu_get_httbr(void) |
1115 | { | |
342cd0ab CD |
1116 | return virt_to_phys(hyp_pgd); |
1117 | } | |
1118 | ||
5a677ce0 MZ |
1119 | phys_addr_t kvm_mmu_get_boot_httbr(void) |
1120 | { | |
1121 | return virt_to_phys(boot_hyp_pgd); | |
1122 | } | |
1123 | ||
1124 | phys_addr_t kvm_get_idmap_vector(void) | |
1125 | { | |
1126 | return hyp_idmap_vector; | |
1127 | } | |
1128 | ||
342cd0ab CD |
1129 | int kvm_mmu_init(void) |
1130 | { | |
2fb41059 MZ |
1131 | int err; |
1132 | ||
4fda342c SS |
1133 | hyp_idmap_start = kvm_virt_to_phys(__hyp_idmap_text_start); |
1134 | hyp_idmap_end = kvm_virt_to_phys(__hyp_idmap_text_end); | |
1135 | hyp_idmap_vector = kvm_virt_to_phys(__kvm_hyp_init); | |
5a677ce0 MZ |
1136 | |
1137 | if ((hyp_idmap_start ^ hyp_idmap_end) & PAGE_MASK) { | |
1138 | /* | |
1139 | * Our init code is crossing a page boundary. Allocate | |
1140 | * a bounce page, copy the code over and use that. | |
1141 | */ | |
1142 | size_t len = __hyp_idmap_text_end - __hyp_idmap_text_start; | |
1143 | phys_addr_t phys_base; | |
1144 | ||
5d4e08c4 | 1145 | init_bounce_page = (void *)__get_free_page(GFP_KERNEL); |
5a677ce0 MZ |
1146 | if (!init_bounce_page) { |
1147 | kvm_err("Couldn't allocate HYP init bounce page\n"); | |
1148 | err = -ENOMEM; | |
1149 | goto out; | |
1150 | } | |
1151 | ||
1152 | memcpy(init_bounce_page, __hyp_idmap_text_start, len); | |
1153 | /* | |
1154 | * Warning: the code we just copied to the bounce page | |
1155 | * must be flushed to the point of coherency. | |
1156 | * Otherwise, the data may be sitting in L2, and HYP | |
1157 | * mode won't be able to observe it as it runs with | |
1158 | * caches off at that point. | |
1159 | */ | |
1160 | kvm_flush_dcache_to_poc(init_bounce_page, len); | |
1161 | ||
4fda342c | 1162 | phys_base = kvm_virt_to_phys(init_bounce_page); |
5a677ce0 MZ |
1163 | hyp_idmap_vector += phys_base - hyp_idmap_start; |
1164 | hyp_idmap_start = phys_base; | |
1165 | hyp_idmap_end = phys_base + len; | |
1166 | ||
1167 | kvm_info("Using HYP init bounce page @%lx\n", | |
1168 | (unsigned long)phys_base); | |
1169 | } | |
1170 | ||
38f791a4 CD |
1171 | hyp_pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, hyp_pgd_order); |
1172 | boot_hyp_pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, hyp_pgd_order); | |
5d4e08c4 | 1173 | |
5a677ce0 | 1174 | if (!hyp_pgd || !boot_hyp_pgd) { |
d5d8184d | 1175 | kvm_err("Hyp mode PGD not allocated\n"); |
2fb41059 MZ |
1176 | err = -ENOMEM; |
1177 | goto out; | |
1178 | } | |
1179 | ||
1180 | /* Create the idmap in the boot page tables */ | |
1181 | err = __create_hyp_mappings(boot_hyp_pgd, | |
1182 | hyp_idmap_start, hyp_idmap_end, | |
1183 | __phys_to_pfn(hyp_idmap_start), | |
1184 | PAGE_HYP); | |
1185 | ||
1186 | if (err) { | |
1187 | kvm_err("Failed to idmap %lx-%lx\n", | |
1188 | hyp_idmap_start, hyp_idmap_end); | |
1189 | goto out; | |
d5d8184d CD |
1190 | } |
1191 | ||
5a677ce0 MZ |
1192 | /* Map the very same page at the trampoline VA */ |
1193 | err = __create_hyp_mappings(boot_hyp_pgd, | |
1194 | TRAMPOLINE_VA, TRAMPOLINE_VA + PAGE_SIZE, | |
1195 | __phys_to_pfn(hyp_idmap_start), | |
1196 | PAGE_HYP); | |
1197 | if (err) { | |
1198 | kvm_err("Failed to map trampoline @%lx into boot HYP pgd\n", | |
1199 | TRAMPOLINE_VA); | |
1200 | goto out; | |
1201 | } | |
1202 | ||
1203 | /* Map the same page again into the runtime page tables */ | |
1204 | err = __create_hyp_mappings(hyp_pgd, | |
1205 | TRAMPOLINE_VA, TRAMPOLINE_VA + PAGE_SIZE, | |
1206 | __phys_to_pfn(hyp_idmap_start), | |
1207 | PAGE_HYP); | |
1208 | if (err) { | |
1209 | kvm_err("Failed to map trampoline @%lx into runtime HYP pgd\n", | |
1210 | TRAMPOLINE_VA); | |
1211 | goto out; | |
1212 | } | |
1213 | ||
d5d8184d | 1214 | return 0; |
2fb41059 | 1215 | out: |
4f728276 | 1216 | free_hyp_pgds(); |
2fb41059 | 1217 | return err; |
342cd0ab | 1218 | } |
df6ce24f EA |
1219 | |
1220 | void kvm_arch_commit_memory_region(struct kvm *kvm, | |
1221 | struct kvm_userspace_memory_region *mem, | |
1222 | const struct kvm_memory_slot *old, | |
1223 | enum kvm_mr_change change) | |
1224 | { | |
df6ce24f EA |
1225 | } |
1226 | ||
1227 | int kvm_arch_prepare_memory_region(struct kvm *kvm, | |
1228 | struct kvm_memory_slot *memslot, | |
1229 | struct kvm_userspace_memory_region *mem, | |
1230 | enum kvm_mr_change change) | |
1231 | { | |
8eef9123 AB |
1232 | hva_t hva = mem->userspace_addr; |
1233 | hva_t reg_end = hva + mem->memory_size; | |
1234 | bool writable = !(mem->flags & KVM_MEM_READONLY); | |
1235 | int ret = 0; | |
1236 | ||
1237 | if (change != KVM_MR_CREATE && change != KVM_MR_MOVE) | |
1238 | return 0; | |
1239 | ||
c3058d5d CD |
1240 | /* |
1241 | * Prevent userspace from creating a memory region outside of the IPA | |
1242 | * space addressable by the KVM guest IPA space. | |
1243 | */ | |
1244 | if (memslot->base_gfn + memslot->npages >= | |
1245 | (KVM_PHYS_SIZE >> PAGE_SHIFT)) | |
1246 | return -EFAULT; | |
1247 | ||
8eef9123 AB |
1248 | /* |
1249 | * A memory region could potentially cover multiple VMAs, and any holes | |
1250 | * between them, so iterate over all of them to find out if we can map | |
1251 | * any of them right now. | |
1252 | * | |
1253 | * +--------------------------------------------+ | |
1254 | * +---------------+----------------+ +----------------+ | |
1255 | * | : VMA 1 | VMA 2 | | VMA 3 : | | |
1256 | * +---------------+----------------+ +----------------+ | |
1257 | * | memory region | | |
1258 | * +--------------------------------------------+ | |
1259 | */ | |
1260 | do { | |
1261 | struct vm_area_struct *vma = find_vma(current->mm, hva); | |
1262 | hva_t vm_start, vm_end; | |
1263 | ||
1264 | if (!vma || vma->vm_start >= reg_end) | |
1265 | break; | |
1266 | ||
1267 | /* | |
1268 | * Mapping a read-only VMA is only allowed if the | |
1269 | * memory region is configured as read-only. | |
1270 | */ | |
1271 | if (writable && !(vma->vm_flags & VM_WRITE)) { | |
1272 | ret = -EPERM; | |
1273 | break; | |
1274 | } | |
1275 | ||
1276 | /* | |
1277 | * Take the intersection of this VMA with the memory region | |
1278 | */ | |
1279 | vm_start = max(hva, vma->vm_start); | |
1280 | vm_end = min(reg_end, vma->vm_end); | |
1281 | ||
1282 | if (vma->vm_flags & VM_PFNMAP) { | |
1283 | gpa_t gpa = mem->guest_phys_addr + | |
1284 | (vm_start - mem->userspace_addr); | |
1285 | phys_addr_t pa = (vma->vm_pgoff << PAGE_SHIFT) + | |
1286 | vm_start - vma->vm_start; | |
1287 | ||
1288 | ret = kvm_phys_addr_ioremap(kvm, gpa, pa, | |
1289 | vm_end - vm_start, | |
1290 | writable); | |
1291 | if (ret) | |
1292 | break; | |
1293 | } | |
1294 | hva = vm_end; | |
1295 | } while (hva < reg_end); | |
1296 | ||
1297 | if (ret) { | |
1298 | spin_lock(&kvm->mmu_lock); | |
1299 | unmap_stage2_range(kvm, mem->guest_phys_addr, mem->memory_size); | |
1300 | spin_unlock(&kvm->mmu_lock); | |
1301 | } | |
1302 | return ret; | |
df6ce24f EA |
1303 | } |
1304 | ||
1305 | void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, | |
1306 | struct kvm_memory_slot *dont) | |
1307 | { | |
1308 | } | |
1309 | ||
1310 | int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot, | |
1311 | unsigned long npages) | |
1312 | { | |
1313 | return 0; | |
1314 | } | |
1315 | ||
1316 | void kvm_arch_memslots_updated(struct kvm *kvm) | |
1317 | { | |
1318 | } | |
1319 | ||
1320 | void kvm_arch_flush_shadow_all(struct kvm *kvm) | |
1321 | { | |
1322 | } | |
1323 | ||
1324 | void kvm_arch_flush_shadow_memslot(struct kvm *kvm, | |
1325 | struct kvm_memory_slot *slot) | |
1326 | { | |
8eef9123 AB |
1327 | gpa_t gpa = slot->base_gfn << PAGE_SHIFT; |
1328 | phys_addr_t size = slot->npages << PAGE_SHIFT; | |
1329 | ||
1330 | spin_lock(&kvm->mmu_lock); | |
1331 | unmap_stage2_range(kvm, gpa, size); | |
1332 | spin_unlock(&kvm->mmu_lock); | |
df6ce24f | 1333 | } |