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> | |
45e96ea6 | 22 | #include <trace/events/kvm.h> |
342cd0ab | 23 | #include <asm/pgalloc.h> |
94f8e641 | 24 | #include <asm/cacheflush.h> |
342cd0ab CD |
25 | #include <asm/kvm_arm.h> |
26 | #include <asm/kvm_mmu.h> | |
45e96ea6 | 27 | #include <asm/kvm_mmio.h> |
d5d8184d | 28 | #include <asm/kvm_asm.h> |
94f8e641 | 29 | #include <asm/kvm_emulate.h> |
d5d8184d CD |
30 | |
31 | #include "trace.h" | |
342cd0ab CD |
32 | |
33 | extern char __hyp_idmap_text_start[], __hyp_idmap_text_end[]; | |
34 | ||
2fb41059 | 35 | static pgd_t *hyp_pgd; |
342cd0ab CD |
36 | static DEFINE_MUTEX(kvm_hyp_pgd_mutex); |
37 | ||
48762767 | 38 | static void kvm_tlb_flush_vmid_ipa(struct kvm *kvm, phys_addr_t ipa) |
d5d8184d | 39 | { |
48762767 | 40 | kvm_call_hyp(__kvm_tlb_flush_vmid_ipa, kvm, ipa); |
d5d8184d CD |
41 | } |
42 | ||
d5d8184d CD |
43 | static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache, |
44 | int min, int max) | |
45 | { | |
46 | void *page; | |
47 | ||
48 | BUG_ON(max > KVM_NR_MEM_OBJS); | |
49 | if (cache->nobjs >= min) | |
50 | return 0; | |
51 | while (cache->nobjs < max) { | |
52 | page = (void *)__get_free_page(PGALLOC_GFP); | |
53 | if (!page) | |
54 | return -ENOMEM; | |
55 | cache->objects[cache->nobjs++] = page; | |
56 | } | |
57 | return 0; | |
58 | } | |
59 | ||
60 | static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
61 | { | |
62 | while (mc->nobjs) | |
63 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
64 | } | |
65 | ||
66 | static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
67 | { | |
68 | void *p; | |
69 | ||
70 | BUG_ON(!mc || !mc->nobjs); | |
71 | p = mc->objects[--mc->nobjs]; | |
72 | return p; | |
73 | } | |
74 | ||
4f728276 | 75 | static void clear_pud_entry(pud_t *pud) |
342cd0ab | 76 | { |
4f728276 MZ |
77 | pmd_t *pmd_table = pmd_offset(pud, 0); |
78 | pud_clear(pud); | |
79 | pmd_free(NULL, pmd_table); | |
80 | put_page(virt_to_page(pud)); | |
81 | } | |
342cd0ab | 82 | |
4f728276 MZ |
83 | static void clear_pmd_entry(pmd_t *pmd) |
84 | { | |
85 | pte_t *pte_table = pte_offset_kernel(pmd, 0); | |
86 | pmd_clear(pmd); | |
87 | pte_free_kernel(NULL, pte_table); | |
88 | put_page(virt_to_page(pmd)); | |
89 | } | |
90 | ||
91 | static bool pmd_empty(pmd_t *pmd) | |
92 | { | |
93 | struct page *pmd_page = virt_to_page(pmd); | |
94 | return page_count(pmd_page) == 1; | |
95 | } | |
96 | ||
97 | static void clear_pte_entry(pte_t *pte) | |
98 | { | |
99 | if (pte_present(*pte)) { | |
100 | kvm_set_pte(pte, __pte(0)); | |
101 | put_page(virt_to_page(pte)); | |
342cd0ab CD |
102 | } |
103 | } | |
104 | ||
4f728276 MZ |
105 | static bool pte_empty(pte_t *pte) |
106 | { | |
107 | struct page *pte_page = virt_to_page(pte); | |
108 | return page_count(pte_page) == 1; | |
109 | } | |
110 | ||
111 | static void unmap_range(pgd_t *pgdp, unsigned long long start, u64 size) | |
000d3996 MZ |
112 | { |
113 | pgd_t *pgd; | |
114 | pud_t *pud; | |
115 | pmd_t *pmd; | |
4f728276 MZ |
116 | pte_t *pte; |
117 | unsigned long long addr = start, end = start + size; | |
118 | u64 range; | |
000d3996 | 119 | |
4f728276 MZ |
120 | while (addr < end) { |
121 | pgd = pgdp + pgd_index(addr); | |
122 | pud = pud_offset(pgd, addr); | |
123 | if (pud_none(*pud)) { | |
124 | addr += PUD_SIZE; | |
125 | continue; | |
126 | } | |
000d3996 | 127 | |
4f728276 MZ |
128 | pmd = pmd_offset(pud, addr); |
129 | if (pmd_none(*pmd)) { | |
130 | addr += PMD_SIZE; | |
131 | continue; | |
132 | } | |
000d3996 | 133 | |
4f728276 MZ |
134 | pte = pte_offset_kernel(pmd, addr); |
135 | clear_pte_entry(pte); | |
136 | range = PAGE_SIZE; | |
137 | ||
138 | /* If we emptied the pte, walk back up the ladder */ | |
139 | if (pte_empty(pte)) { | |
140 | clear_pmd_entry(pmd); | |
141 | range = PMD_SIZE; | |
142 | if (pmd_empty(pmd)) { | |
143 | clear_pud_entry(pud); | |
144 | range = PUD_SIZE; | |
145 | } | |
146 | } | |
147 | ||
148 | addr += range; | |
149 | } | |
000d3996 MZ |
150 | } |
151 | ||
342cd0ab | 152 | /** |
4f728276 | 153 | * free_hyp_pgds - free Hyp-mode page tables |
342cd0ab | 154 | * |
4f728276 | 155 | * Assumes hyp_pgd is a page table used strictly in Hyp-mode and therefore contains |
000d3996 MZ |
156 | * either mappings in the kernel memory area (above PAGE_OFFSET), or |
157 | * device mappings in the vmalloc range (from VMALLOC_START to VMALLOC_END). | |
342cd0ab | 158 | */ |
4f728276 | 159 | void free_hyp_pgds(void) |
342cd0ab | 160 | { |
342cd0ab CD |
161 | unsigned long addr; |
162 | ||
163 | mutex_lock(&kvm_hyp_pgd_mutex); | |
4f728276 MZ |
164 | |
165 | if (hyp_pgd) { | |
166 | for (addr = PAGE_OFFSET; virt_addr_valid(addr); addr += PGDIR_SIZE) | |
167 | unmap_range(hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE); | |
168 | for (addr = VMALLOC_START; is_vmalloc_addr((void*)addr); addr += PGDIR_SIZE) | |
169 | unmap_range(hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE); | |
170 | kfree(hyp_pgd); | |
171 | } | |
172 | ||
342cd0ab CD |
173 | mutex_unlock(&kvm_hyp_pgd_mutex); |
174 | } | |
175 | ||
176 | static void create_hyp_pte_mappings(pmd_t *pmd, unsigned long start, | |
6060df84 MZ |
177 | unsigned long end, unsigned long pfn, |
178 | pgprot_t prot) | |
342cd0ab CD |
179 | { |
180 | pte_t *pte; | |
181 | unsigned long addr; | |
342cd0ab | 182 | |
3562c76d MZ |
183 | addr = start; |
184 | do { | |
6060df84 MZ |
185 | pte = pte_offset_kernel(pmd, addr); |
186 | kvm_set_pte(pte, pfn_pte(pfn, prot)); | |
4f728276 | 187 | get_page(virt_to_page(pte)); |
6060df84 | 188 | pfn++; |
3562c76d | 189 | } while (addr += PAGE_SIZE, addr != end); |
342cd0ab CD |
190 | } |
191 | ||
192 | static int create_hyp_pmd_mappings(pud_t *pud, unsigned long start, | |
6060df84 MZ |
193 | unsigned long end, unsigned long pfn, |
194 | pgprot_t prot) | |
342cd0ab CD |
195 | { |
196 | pmd_t *pmd; | |
197 | pte_t *pte; | |
198 | unsigned long addr, next; | |
199 | ||
3562c76d MZ |
200 | addr = start; |
201 | do { | |
6060df84 | 202 | pmd = pmd_offset(pud, addr); |
342cd0ab CD |
203 | |
204 | BUG_ON(pmd_sect(*pmd)); | |
205 | ||
206 | if (pmd_none(*pmd)) { | |
6060df84 | 207 | pte = pte_alloc_one_kernel(NULL, addr); |
342cd0ab CD |
208 | if (!pte) { |
209 | kvm_err("Cannot allocate Hyp pte\n"); | |
210 | return -ENOMEM; | |
211 | } | |
212 | pmd_populate_kernel(NULL, pmd, pte); | |
4f728276 | 213 | get_page(virt_to_page(pmd)); |
342cd0ab CD |
214 | } |
215 | ||
216 | next = pmd_addr_end(addr, end); | |
217 | ||
6060df84 MZ |
218 | create_hyp_pte_mappings(pmd, addr, next, pfn, prot); |
219 | pfn += (next - addr) >> PAGE_SHIFT; | |
3562c76d | 220 | } while (addr = next, addr != end); |
342cd0ab CD |
221 | |
222 | return 0; | |
223 | } | |
224 | ||
6060df84 MZ |
225 | static int __create_hyp_mappings(pgd_t *pgdp, |
226 | unsigned long start, unsigned long end, | |
227 | unsigned long pfn, pgprot_t prot) | |
342cd0ab | 228 | { |
342cd0ab CD |
229 | pgd_t *pgd; |
230 | pud_t *pud; | |
231 | pmd_t *pmd; | |
232 | unsigned long addr, next; | |
233 | int err = 0; | |
234 | ||
342cd0ab | 235 | mutex_lock(&kvm_hyp_pgd_mutex); |
3562c76d MZ |
236 | addr = start & PAGE_MASK; |
237 | end = PAGE_ALIGN(end); | |
238 | do { | |
6060df84 MZ |
239 | pgd = pgdp + pgd_index(addr); |
240 | pud = pud_offset(pgd, addr); | |
342cd0ab CD |
241 | |
242 | if (pud_none_or_clear_bad(pud)) { | |
6060df84 | 243 | pmd = pmd_alloc_one(NULL, addr); |
342cd0ab CD |
244 | if (!pmd) { |
245 | kvm_err("Cannot allocate Hyp pmd\n"); | |
246 | err = -ENOMEM; | |
247 | goto out; | |
248 | } | |
249 | pud_populate(NULL, pud, pmd); | |
4f728276 | 250 | get_page(virt_to_page(pud)); |
342cd0ab CD |
251 | } |
252 | ||
253 | next = pgd_addr_end(addr, end); | |
6060df84 | 254 | err = create_hyp_pmd_mappings(pud, addr, next, pfn, prot); |
342cd0ab CD |
255 | if (err) |
256 | goto out; | |
6060df84 | 257 | pfn += (next - addr) >> PAGE_SHIFT; |
3562c76d | 258 | } while (addr = next, addr != end); |
342cd0ab CD |
259 | out: |
260 | mutex_unlock(&kvm_hyp_pgd_mutex); | |
261 | return err; | |
262 | } | |
263 | ||
264 | /** | |
06e8c3b0 | 265 | * create_hyp_mappings - duplicate a kernel virtual address range in Hyp mode |
342cd0ab CD |
266 | * @from: The virtual kernel start address of the range |
267 | * @to: The virtual kernel end address of the range (exclusive) | |
268 | * | |
06e8c3b0 MZ |
269 | * The same virtual address as the kernel virtual address is also used |
270 | * in Hyp-mode mapping (modulo HYP_PAGE_OFFSET) to the same underlying | |
271 | * physical pages. | |
342cd0ab CD |
272 | */ |
273 | int create_hyp_mappings(void *from, void *to) | |
274 | { | |
6060df84 MZ |
275 | unsigned long phys_addr = virt_to_phys(from); |
276 | unsigned long start = KERN_TO_HYP((unsigned long)from); | |
277 | unsigned long end = KERN_TO_HYP((unsigned long)to); | |
278 | ||
279 | /* Check for a valid kernel memory mapping */ | |
280 | if (!virt_addr_valid(from) || !virt_addr_valid(to - 1)) | |
281 | return -EINVAL; | |
282 | ||
283 | return __create_hyp_mappings(hyp_pgd, start, end, | |
284 | __phys_to_pfn(phys_addr), PAGE_HYP); | |
342cd0ab CD |
285 | } |
286 | ||
287 | /** | |
06e8c3b0 MZ |
288 | * create_hyp_io_mappings - duplicate a kernel IO mapping into Hyp mode |
289 | * @from: The kernel start VA of the range | |
290 | * @to: The kernel end VA of the range (exclusive) | |
6060df84 | 291 | * @phys_addr: The physical start address which gets mapped |
06e8c3b0 MZ |
292 | * |
293 | * The resulting HYP VA is the same as the kernel VA, modulo | |
294 | * HYP_PAGE_OFFSET. | |
342cd0ab | 295 | */ |
6060df84 | 296 | int create_hyp_io_mappings(void *from, void *to, phys_addr_t phys_addr) |
342cd0ab | 297 | { |
6060df84 MZ |
298 | unsigned long start = KERN_TO_HYP((unsigned long)from); |
299 | unsigned long end = KERN_TO_HYP((unsigned long)to); | |
300 | ||
301 | /* Check for a valid kernel IO mapping */ | |
302 | if (!is_vmalloc_addr(from) || !is_vmalloc_addr(to - 1)) | |
303 | return -EINVAL; | |
304 | ||
305 | return __create_hyp_mappings(hyp_pgd, start, end, | |
306 | __phys_to_pfn(phys_addr), PAGE_HYP_DEVICE); | |
342cd0ab CD |
307 | } |
308 | ||
d5d8184d CD |
309 | /** |
310 | * kvm_alloc_stage2_pgd - allocate level-1 table for stage-2 translation. | |
311 | * @kvm: The KVM struct pointer for the VM. | |
312 | * | |
313 | * Allocates the 1st level table only of size defined by S2_PGD_ORDER (can | |
314 | * support either full 40-bit input addresses or limited to 32-bit input | |
315 | * addresses). Clears the allocated pages. | |
316 | * | |
317 | * Note we don't need locking here as this is only called when the VM is | |
318 | * created, which can only be done once. | |
319 | */ | |
320 | int kvm_alloc_stage2_pgd(struct kvm *kvm) | |
321 | { | |
322 | pgd_t *pgd; | |
323 | ||
324 | if (kvm->arch.pgd != NULL) { | |
325 | kvm_err("kvm_arch already initialized?\n"); | |
326 | return -EINVAL; | |
327 | } | |
328 | ||
329 | pgd = (pgd_t *)__get_free_pages(GFP_KERNEL, S2_PGD_ORDER); | |
330 | if (!pgd) | |
331 | return -ENOMEM; | |
332 | ||
333 | /* stage-2 pgd must be aligned to its size */ | |
334 | VM_BUG_ON((unsigned long)pgd & (S2_PGD_SIZE - 1)); | |
335 | ||
336 | memset(pgd, 0, PTRS_PER_S2_PGD * sizeof(pgd_t)); | |
c62ee2b2 | 337 | kvm_clean_pgd(pgd); |
d5d8184d CD |
338 | kvm->arch.pgd = pgd; |
339 | ||
340 | return 0; | |
341 | } | |
342 | ||
d5d8184d CD |
343 | /** |
344 | * unmap_stage2_range -- Clear stage2 page table entries to unmap a range | |
345 | * @kvm: The VM pointer | |
346 | * @start: The intermediate physical base address of the range to unmap | |
347 | * @size: The size of the area to unmap | |
348 | * | |
349 | * Clear a range of stage-2 mappings, lowering the various ref-counts. Must | |
350 | * be called while holding mmu_lock (unless for freeing the stage2 pgd before | |
351 | * destroying the VM), otherwise another faulting VCPU may come in and mess | |
352 | * with things behind our backs. | |
353 | */ | |
354 | static void unmap_stage2_range(struct kvm *kvm, phys_addr_t start, u64 size) | |
355 | { | |
4f728276 | 356 | unmap_range(kvm->arch.pgd, start, size); |
d5d8184d CD |
357 | } |
358 | ||
359 | /** | |
360 | * kvm_free_stage2_pgd - free all stage-2 tables | |
361 | * @kvm: The KVM struct pointer for the VM. | |
362 | * | |
363 | * Walks the level-1 page table pointed to by kvm->arch.pgd and frees all | |
364 | * underlying level-2 and level-3 tables before freeing the actual level-1 table | |
365 | * and setting the struct pointer to NULL. | |
366 | * | |
367 | * Note we don't need locking here as this is only called when the VM is | |
368 | * destroyed, which can only be done once. | |
369 | */ | |
370 | void kvm_free_stage2_pgd(struct kvm *kvm) | |
371 | { | |
372 | if (kvm->arch.pgd == NULL) | |
373 | return; | |
374 | ||
375 | unmap_stage2_range(kvm, 0, KVM_PHYS_SIZE); | |
376 | free_pages((unsigned long)kvm->arch.pgd, S2_PGD_ORDER); | |
377 | kvm->arch.pgd = NULL; | |
378 | } | |
379 | ||
380 | ||
381 | static int stage2_set_pte(struct kvm *kvm, struct kvm_mmu_memory_cache *cache, | |
382 | phys_addr_t addr, const pte_t *new_pte, bool iomap) | |
383 | { | |
384 | pgd_t *pgd; | |
385 | pud_t *pud; | |
386 | pmd_t *pmd; | |
387 | pte_t *pte, old_pte; | |
388 | ||
389 | /* Create 2nd stage page table mapping - Level 1 */ | |
390 | pgd = kvm->arch.pgd + pgd_index(addr); | |
391 | pud = pud_offset(pgd, addr); | |
392 | if (pud_none(*pud)) { | |
393 | if (!cache) | |
394 | return 0; /* ignore calls from kvm_set_spte_hva */ | |
395 | pmd = mmu_memory_cache_alloc(cache); | |
396 | pud_populate(NULL, pud, pmd); | |
d5d8184d | 397 | get_page(virt_to_page(pud)); |
c62ee2b2 MZ |
398 | } |
399 | ||
400 | pmd = pmd_offset(pud, addr); | |
d5d8184d CD |
401 | |
402 | /* Create 2nd stage page table mapping - Level 2 */ | |
403 | if (pmd_none(*pmd)) { | |
404 | if (!cache) | |
405 | return 0; /* ignore calls from kvm_set_spte_hva */ | |
406 | pte = mmu_memory_cache_alloc(cache); | |
c62ee2b2 | 407 | kvm_clean_pte(pte); |
d5d8184d | 408 | pmd_populate_kernel(NULL, pmd, pte); |
d5d8184d | 409 | get_page(virt_to_page(pmd)); |
c62ee2b2 MZ |
410 | } |
411 | ||
412 | pte = pte_offset_kernel(pmd, addr); | |
d5d8184d CD |
413 | |
414 | if (iomap && pte_present(*pte)) | |
415 | return -EFAULT; | |
416 | ||
417 | /* Create 2nd stage page table mapping - Level 3 */ | |
418 | old_pte = *pte; | |
419 | kvm_set_pte(pte, *new_pte); | |
420 | if (pte_present(old_pte)) | |
48762767 | 421 | kvm_tlb_flush_vmid_ipa(kvm, addr); |
d5d8184d CD |
422 | else |
423 | get_page(virt_to_page(pte)); | |
424 | ||
425 | return 0; | |
426 | } | |
427 | ||
428 | /** | |
429 | * kvm_phys_addr_ioremap - map a device range to guest IPA | |
430 | * | |
431 | * @kvm: The KVM pointer | |
432 | * @guest_ipa: The IPA at which to insert the mapping | |
433 | * @pa: The physical address of the device | |
434 | * @size: The size of the mapping | |
435 | */ | |
436 | int kvm_phys_addr_ioremap(struct kvm *kvm, phys_addr_t guest_ipa, | |
437 | phys_addr_t pa, unsigned long size) | |
438 | { | |
439 | phys_addr_t addr, end; | |
440 | int ret = 0; | |
441 | unsigned long pfn; | |
442 | struct kvm_mmu_memory_cache cache = { 0, }; | |
443 | ||
444 | end = (guest_ipa + size + PAGE_SIZE - 1) & PAGE_MASK; | |
445 | pfn = __phys_to_pfn(pa); | |
446 | ||
447 | for (addr = guest_ipa; addr < end; addr += PAGE_SIZE) { | |
c62ee2b2 MZ |
448 | pte_t pte = pfn_pte(pfn, PAGE_S2_DEVICE); |
449 | kvm_set_s2pte_writable(&pte); | |
d5d8184d CD |
450 | |
451 | ret = mmu_topup_memory_cache(&cache, 2, 2); | |
452 | if (ret) | |
453 | goto out; | |
454 | spin_lock(&kvm->mmu_lock); | |
455 | ret = stage2_set_pte(kvm, &cache, addr, &pte, true); | |
456 | spin_unlock(&kvm->mmu_lock); | |
457 | if (ret) | |
458 | goto out; | |
459 | ||
460 | pfn++; | |
461 | } | |
462 | ||
463 | out: | |
464 | mmu_free_memory_cache(&cache); | |
465 | return ret; | |
466 | } | |
467 | ||
94f8e641 CD |
468 | static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa, |
469 | gfn_t gfn, struct kvm_memory_slot *memslot, | |
470 | unsigned long fault_status) | |
471 | { | |
472 | pte_t new_pte; | |
473 | pfn_t pfn; | |
474 | int ret; | |
475 | bool write_fault, writable; | |
476 | unsigned long mmu_seq; | |
477 | struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache; | |
478 | ||
7393b599 | 479 | write_fault = kvm_is_write_fault(kvm_vcpu_get_hsr(vcpu)); |
94f8e641 CD |
480 | if (fault_status == FSC_PERM && !write_fault) { |
481 | kvm_err("Unexpected L2 read permission error\n"); | |
482 | return -EFAULT; | |
483 | } | |
484 | ||
485 | /* We need minimum second+third level pages */ | |
486 | ret = mmu_topup_memory_cache(memcache, 2, KVM_NR_MEM_OBJS); | |
487 | if (ret) | |
488 | return ret; | |
489 | ||
490 | mmu_seq = vcpu->kvm->mmu_notifier_seq; | |
491 | /* | |
492 | * Ensure the read of mmu_notifier_seq happens before we call | |
493 | * gfn_to_pfn_prot (which calls get_user_pages), so that we don't risk | |
494 | * the page we just got a reference to gets unmapped before we have a | |
495 | * chance to grab the mmu_lock, which ensure that if the page gets | |
496 | * unmapped afterwards, the call to kvm_unmap_hva will take it away | |
497 | * from us again properly. This smp_rmb() interacts with the smp_wmb() | |
498 | * in kvm_mmu_notifier_invalidate_<page|range_end>. | |
499 | */ | |
500 | smp_rmb(); | |
501 | ||
502 | pfn = gfn_to_pfn_prot(vcpu->kvm, gfn, write_fault, &writable); | |
503 | if (is_error_pfn(pfn)) | |
504 | return -EFAULT; | |
505 | ||
506 | new_pte = pfn_pte(pfn, PAGE_S2); | |
507 | coherent_icache_guest_page(vcpu->kvm, gfn); | |
508 | ||
509 | spin_lock(&vcpu->kvm->mmu_lock); | |
510 | if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) | |
511 | goto out_unlock; | |
512 | if (writable) { | |
c62ee2b2 | 513 | kvm_set_s2pte_writable(&new_pte); |
94f8e641 CD |
514 | kvm_set_pfn_dirty(pfn); |
515 | } | |
516 | stage2_set_pte(vcpu->kvm, memcache, fault_ipa, &new_pte, false); | |
517 | ||
518 | out_unlock: | |
519 | spin_unlock(&vcpu->kvm->mmu_lock); | |
520 | kvm_release_pfn_clean(pfn); | |
521 | return 0; | |
522 | } | |
523 | ||
524 | /** | |
525 | * kvm_handle_guest_abort - handles all 2nd stage aborts | |
526 | * @vcpu: the VCPU pointer | |
527 | * @run: the kvm_run structure | |
528 | * | |
529 | * Any abort that gets to the host is almost guaranteed to be caused by a | |
530 | * missing second stage translation table entry, which can mean that either the | |
531 | * guest simply needs more memory and we must allocate an appropriate page or it | |
532 | * can mean that the guest tried to access I/O memory, which is emulated by user | |
533 | * space. The distinction is based on the IPA causing the fault and whether this | |
534 | * memory region has been registered as standard RAM by user space. | |
535 | */ | |
342cd0ab CD |
536 | int kvm_handle_guest_abort(struct kvm_vcpu *vcpu, struct kvm_run *run) |
537 | { | |
94f8e641 CD |
538 | unsigned long fault_status; |
539 | phys_addr_t fault_ipa; | |
540 | struct kvm_memory_slot *memslot; | |
541 | bool is_iabt; | |
542 | gfn_t gfn; | |
543 | int ret, idx; | |
544 | ||
52d1dba9 | 545 | is_iabt = kvm_vcpu_trap_is_iabt(vcpu); |
7393b599 | 546 | fault_ipa = kvm_vcpu_get_fault_ipa(vcpu); |
94f8e641 | 547 | |
7393b599 MZ |
548 | trace_kvm_guest_fault(*vcpu_pc(vcpu), kvm_vcpu_get_hsr(vcpu), |
549 | kvm_vcpu_get_hfar(vcpu), fault_ipa); | |
94f8e641 CD |
550 | |
551 | /* Check the stage-2 fault is trans. fault or write fault */ | |
1cc287dd | 552 | fault_status = kvm_vcpu_trap_get_fault(vcpu); |
94f8e641 | 553 | if (fault_status != FSC_FAULT && fault_status != FSC_PERM) { |
52d1dba9 MZ |
554 | kvm_err("Unsupported fault status: EC=%#x DFCS=%#lx\n", |
555 | kvm_vcpu_trap_get_class(vcpu), fault_status); | |
94f8e641 CD |
556 | return -EFAULT; |
557 | } | |
558 | ||
559 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
560 | ||
561 | gfn = fault_ipa >> PAGE_SHIFT; | |
562 | if (!kvm_is_visible_gfn(vcpu->kvm, gfn)) { | |
563 | if (is_iabt) { | |
564 | /* Prefetch Abort on I/O address */ | |
7393b599 | 565 | kvm_inject_pabt(vcpu, kvm_vcpu_get_hfar(vcpu)); |
94f8e641 CD |
566 | ret = 1; |
567 | goto out_unlock; | |
568 | } | |
569 | ||
570 | if (fault_status != FSC_FAULT) { | |
571 | kvm_err("Unsupported fault status on io memory: %#lx\n", | |
572 | fault_status); | |
573 | ret = -EFAULT; | |
574 | goto out_unlock; | |
575 | } | |
576 | ||
cfe3950c MZ |
577 | /* |
578 | * The IPA is reported as [MAX:12], so we need to | |
579 | * complement it with the bottom 12 bits from the | |
580 | * faulting VA. This is always 12 bits, irrespective | |
581 | * of the page size. | |
582 | */ | |
583 | fault_ipa |= kvm_vcpu_get_hfar(vcpu) & ((1 << 12) - 1); | |
45e96ea6 | 584 | ret = io_mem_abort(vcpu, run, fault_ipa); |
94f8e641 CD |
585 | goto out_unlock; |
586 | } | |
587 | ||
588 | memslot = gfn_to_memslot(vcpu->kvm, gfn); | |
94f8e641 CD |
589 | |
590 | ret = user_mem_abort(vcpu, fault_ipa, gfn, memslot, fault_status); | |
591 | if (ret == 0) | |
592 | ret = 1; | |
593 | out_unlock: | |
594 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
595 | return ret; | |
342cd0ab CD |
596 | } |
597 | ||
d5d8184d CD |
598 | static void handle_hva_to_gpa(struct kvm *kvm, |
599 | unsigned long start, | |
600 | unsigned long end, | |
601 | void (*handler)(struct kvm *kvm, | |
602 | gpa_t gpa, void *data), | |
603 | void *data) | |
604 | { | |
605 | struct kvm_memslots *slots; | |
606 | struct kvm_memory_slot *memslot; | |
607 | ||
608 | slots = kvm_memslots(kvm); | |
609 | ||
610 | /* we only care about the pages that the guest sees */ | |
611 | kvm_for_each_memslot(memslot, slots) { | |
612 | unsigned long hva_start, hva_end; | |
613 | gfn_t gfn, gfn_end; | |
614 | ||
615 | hva_start = max(start, memslot->userspace_addr); | |
616 | hva_end = min(end, memslot->userspace_addr + | |
617 | (memslot->npages << PAGE_SHIFT)); | |
618 | if (hva_start >= hva_end) | |
619 | continue; | |
620 | ||
621 | /* | |
622 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
623 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
624 | */ | |
625 | gfn = hva_to_gfn_memslot(hva_start, memslot); | |
626 | gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot); | |
627 | ||
628 | for (; gfn < gfn_end; ++gfn) { | |
629 | gpa_t gpa = gfn << PAGE_SHIFT; | |
630 | handler(kvm, gpa, data); | |
631 | } | |
632 | } | |
633 | } | |
634 | ||
635 | static void kvm_unmap_hva_handler(struct kvm *kvm, gpa_t gpa, void *data) | |
636 | { | |
637 | unmap_stage2_range(kvm, gpa, PAGE_SIZE); | |
48762767 | 638 | kvm_tlb_flush_vmid_ipa(kvm, gpa); |
d5d8184d CD |
639 | } |
640 | ||
641 | int kvm_unmap_hva(struct kvm *kvm, unsigned long hva) | |
642 | { | |
643 | unsigned long end = hva + PAGE_SIZE; | |
644 | ||
645 | if (!kvm->arch.pgd) | |
646 | return 0; | |
647 | ||
648 | trace_kvm_unmap_hva(hva); | |
649 | handle_hva_to_gpa(kvm, hva, end, &kvm_unmap_hva_handler, NULL); | |
650 | return 0; | |
651 | } | |
652 | ||
653 | int kvm_unmap_hva_range(struct kvm *kvm, | |
654 | unsigned long start, unsigned long end) | |
655 | { | |
656 | if (!kvm->arch.pgd) | |
657 | return 0; | |
658 | ||
659 | trace_kvm_unmap_hva_range(start, end); | |
660 | handle_hva_to_gpa(kvm, start, end, &kvm_unmap_hva_handler, NULL); | |
661 | return 0; | |
662 | } | |
663 | ||
664 | static void kvm_set_spte_handler(struct kvm *kvm, gpa_t gpa, void *data) | |
665 | { | |
666 | pte_t *pte = (pte_t *)data; | |
667 | ||
668 | stage2_set_pte(kvm, NULL, gpa, pte, false); | |
669 | } | |
670 | ||
671 | ||
672 | void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte) | |
673 | { | |
674 | unsigned long end = hva + PAGE_SIZE; | |
675 | pte_t stage2_pte; | |
676 | ||
677 | if (!kvm->arch.pgd) | |
678 | return; | |
679 | ||
680 | trace_kvm_set_spte_hva(hva); | |
681 | stage2_pte = pfn_pte(pte_pfn(pte), PAGE_S2); | |
682 | handle_hva_to_gpa(kvm, hva, end, &kvm_set_spte_handler, &stage2_pte); | |
683 | } | |
684 | ||
685 | void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu) | |
686 | { | |
687 | mmu_free_memory_cache(&vcpu->arch.mmu_page_cache); | |
688 | } | |
689 | ||
342cd0ab CD |
690 | phys_addr_t kvm_mmu_get_httbr(void) |
691 | { | |
692 | VM_BUG_ON(!virt_addr_valid(hyp_pgd)); | |
693 | return virt_to_phys(hyp_pgd); | |
694 | } | |
695 | ||
696 | int kvm_mmu_init(void) | |
697 | { | |
2fb41059 MZ |
698 | unsigned long hyp_idmap_start = virt_to_phys(__hyp_idmap_text_start); |
699 | unsigned long hyp_idmap_end = virt_to_phys(__hyp_idmap_text_end); | |
700 | int err; | |
701 | ||
702 | hyp_pgd = kzalloc(PTRS_PER_PGD * sizeof(pgd_t), GFP_KERNEL); | |
d5d8184d CD |
703 | if (!hyp_pgd) { |
704 | kvm_err("Hyp mode PGD not allocated\n"); | |
2fb41059 MZ |
705 | err = -ENOMEM; |
706 | goto out; | |
707 | } | |
708 | ||
709 | /* Create the idmap in the boot page tables */ | |
710 | err = __create_hyp_mappings(boot_hyp_pgd, | |
711 | hyp_idmap_start, hyp_idmap_end, | |
712 | __phys_to_pfn(hyp_idmap_start), | |
713 | PAGE_HYP); | |
714 | ||
715 | if (err) { | |
716 | kvm_err("Failed to idmap %lx-%lx\n", | |
717 | hyp_idmap_start, hyp_idmap_end); | |
718 | goto out; | |
d5d8184d CD |
719 | } |
720 | ||
721 | return 0; | |
2fb41059 | 722 | out: |
4f728276 | 723 | free_hyp_pgds(); |
2fb41059 | 724 | return err; |
342cd0ab CD |
725 | } |
726 | ||
727 | /** | |
728 | * kvm_clear_idmap - remove all idmaps from the hyp pgd | |
729 | * | |
730 | * Free the underlying pmds for all pgds in range and clear the pgds (but | |
731 | * don't free them) afterwards. | |
732 | */ | |
733 | void kvm_clear_hyp_idmap(void) | |
734 | { | |
735 | unsigned long addr, end; | |
736 | unsigned long next; | |
737 | pgd_t *pgd = hyp_pgd; | |
738 | pud_t *pud; | |
739 | pmd_t *pmd; | |
740 | ||
741 | addr = virt_to_phys(__hyp_idmap_text_start); | |
742 | end = virt_to_phys(__hyp_idmap_text_end); | |
743 | ||
744 | pgd += pgd_index(addr); | |
745 | do { | |
746 | next = pgd_addr_end(addr, end); | |
747 | if (pgd_none_or_clear_bad(pgd)) | |
748 | continue; | |
749 | pud = pud_offset(pgd, addr); | |
750 | pmd = pmd_offset(pud, addr); | |
751 | ||
752 | pud_clear(pud); | |
c62ee2b2 | 753 | kvm_clean_pmd_entry(pmd); |
342cd0ab CD |
754 | pmd_free(NULL, (pmd_t *)((unsigned long)pmd & PAGE_MASK)); |
755 | } while (pgd++, addr = next, addr < end); | |
756 | } |