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 | ||
5a677ce0 | 35 | static pgd_t *boot_hyp_pgd; |
2fb41059 | 36 | static pgd_t *hyp_pgd; |
342cd0ab CD |
37 | static DEFINE_MUTEX(kvm_hyp_pgd_mutex); |
38 | ||
5a677ce0 MZ |
39 | static void *init_bounce_page; |
40 | static unsigned long hyp_idmap_start; | |
41 | static unsigned long hyp_idmap_end; | |
42 | static phys_addr_t hyp_idmap_vector; | |
43 | ||
48762767 | 44 | static void kvm_tlb_flush_vmid_ipa(struct kvm *kvm, phys_addr_t ipa) |
d5d8184d | 45 | { |
d4cb9df5 MZ |
46 | /* |
47 | * This function also gets called when dealing with HYP page | |
48 | * tables. As HYP doesn't have an associated struct kvm (and | |
49 | * the HYP page tables are fairly static), we don't do | |
50 | * anything there. | |
51 | */ | |
52 | if (kvm) | |
53 | kvm_call_hyp(__kvm_tlb_flush_vmid_ipa, kvm, ipa); | |
d5d8184d CD |
54 | } |
55 | ||
d5d8184d CD |
56 | static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache, |
57 | int min, int max) | |
58 | { | |
59 | void *page; | |
60 | ||
61 | BUG_ON(max > KVM_NR_MEM_OBJS); | |
62 | if (cache->nobjs >= min) | |
63 | return 0; | |
64 | while (cache->nobjs < max) { | |
65 | page = (void *)__get_free_page(PGALLOC_GFP); | |
66 | if (!page) | |
67 | return -ENOMEM; | |
68 | cache->objects[cache->nobjs++] = page; | |
69 | } | |
70 | return 0; | |
71 | } | |
72 | ||
73 | static void mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc) | |
74 | { | |
75 | while (mc->nobjs) | |
76 | free_page((unsigned long)mc->objects[--mc->nobjs]); | |
77 | } | |
78 | ||
79 | static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc) | |
80 | { | |
81 | void *p; | |
82 | ||
83 | BUG_ON(!mc || !mc->nobjs); | |
84 | p = mc->objects[--mc->nobjs]; | |
85 | return p; | |
86 | } | |
87 | ||
979acd5e MZ |
88 | static bool page_empty(void *ptr) |
89 | { | |
90 | struct page *ptr_page = virt_to_page(ptr); | |
91 | return page_count(ptr_page) == 1; | |
92 | } | |
93 | ||
d4cb9df5 | 94 | static void clear_pud_entry(struct kvm *kvm, pud_t *pud, phys_addr_t addr) |
342cd0ab | 95 | { |
4f728276 MZ |
96 | pmd_t *pmd_table = pmd_offset(pud, 0); |
97 | pud_clear(pud); | |
d4cb9df5 | 98 | kvm_tlb_flush_vmid_ipa(kvm, addr); |
4f728276 MZ |
99 | pmd_free(NULL, pmd_table); |
100 | put_page(virt_to_page(pud)); | |
101 | } | |
342cd0ab | 102 | |
d4cb9df5 | 103 | static void clear_pmd_entry(struct kvm *kvm, pmd_t *pmd, phys_addr_t addr) |
4f728276 MZ |
104 | { |
105 | pte_t *pte_table = pte_offset_kernel(pmd, 0); | |
106 | pmd_clear(pmd); | |
d4cb9df5 | 107 | kvm_tlb_flush_vmid_ipa(kvm, addr); |
4f728276 MZ |
108 | pte_free_kernel(NULL, pte_table); |
109 | put_page(virt_to_page(pmd)); | |
110 | } | |
111 | ||
d4cb9df5 | 112 | static void clear_pte_entry(struct kvm *kvm, pte_t *pte, phys_addr_t addr) |
4f728276 MZ |
113 | { |
114 | if (pte_present(*pte)) { | |
115 | kvm_set_pte(pte, __pte(0)); | |
116 | put_page(virt_to_page(pte)); | |
d4cb9df5 | 117 | kvm_tlb_flush_vmid_ipa(kvm, addr); |
342cd0ab CD |
118 | } |
119 | } | |
120 | ||
d4cb9df5 MZ |
121 | static void unmap_range(struct kvm *kvm, pgd_t *pgdp, |
122 | unsigned long long start, u64 size) | |
000d3996 MZ |
123 | { |
124 | pgd_t *pgd; | |
125 | pud_t *pud; | |
126 | pmd_t *pmd; | |
4f728276 MZ |
127 | pte_t *pte; |
128 | unsigned long long addr = start, end = start + size; | |
d3840b26 | 129 | u64 next; |
000d3996 | 130 | |
4f728276 MZ |
131 | while (addr < end) { |
132 | pgd = pgdp + pgd_index(addr); | |
133 | pud = pud_offset(pgd, addr); | |
134 | if (pud_none(*pud)) { | |
d3840b26 | 135 | addr = pud_addr_end(addr, end); |
4f728276 MZ |
136 | continue; |
137 | } | |
000d3996 | 138 | |
4f728276 MZ |
139 | pmd = pmd_offset(pud, addr); |
140 | if (pmd_none(*pmd)) { | |
d3840b26 | 141 | addr = pmd_addr_end(addr, end); |
4f728276 MZ |
142 | continue; |
143 | } | |
000d3996 | 144 | |
4f728276 | 145 | pte = pte_offset_kernel(pmd, addr); |
d4cb9df5 | 146 | clear_pte_entry(kvm, pte, addr); |
d3840b26 | 147 | next = addr + PAGE_SIZE; |
4f728276 MZ |
148 | |
149 | /* If we emptied the pte, walk back up the ladder */ | |
979acd5e | 150 | if (page_empty(pte)) { |
d4cb9df5 | 151 | clear_pmd_entry(kvm, pmd, addr); |
d3840b26 | 152 | next = pmd_addr_end(addr, end); |
979acd5e | 153 | if (page_empty(pmd) && !page_empty(pud)) { |
d4cb9df5 | 154 | clear_pud_entry(kvm, pud, addr); |
d3840b26 | 155 | next = pud_addr_end(addr, end); |
4f728276 MZ |
156 | } |
157 | } | |
158 | ||
d3840b26 | 159 | addr = next; |
4f728276 | 160 | } |
000d3996 MZ |
161 | } |
162 | ||
d157f4a5 MZ |
163 | /** |
164 | * free_boot_hyp_pgd - free HYP boot page tables | |
165 | * | |
166 | * Free the HYP boot page tables. The bounce page is also freed. | |
167 | */ | |
168 | void free_boot_hyp_pgd(void) | |
169 | { | |
170 | mutex_lock(&kvm_hyp_pgd_mutex); | |
171 | ||
172 | if (boot_hyp_pgd) { | |
d4cb9df5 MZ |
173 | unmap_range(NULL, boot_hyp_pgd, hyp_idmap_start, PAGE_SIZE); |
174 | unmap_range(NULL, boot_hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE); | |
d157f4a5 MZ |
175 | kfree(boot_hyp_pgd); |
176 | boot_hyp_pgd = NULL; | |
177 | } | |
178 | ||
179 | if (hyp_pgd) | |
d4cb9df5 | 180 | unmap_range(NULL, hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE); |
d157f4a5 MZ |
181 | |
182 | kfree(init_bounce_page); | |
183 | init_bounce_page = NULL; | |
184 | ||
185 | mutex_unlock(&kvm_hyp_pgd_mutex); | |
186 | } | |
187 | ||
342cd0ab | 188 | /** |
4f728276 | 189 | * free_hyp_pgds - free Hyp-mode page tables |
342cd0ab | 190 | * |
5a677ce0 MZ |
191 | * Assumes hyp_pgd is a page table used strictly in Hyp-mode and |
192 | * therefore contains either mappings in the kernel memory area (above | |
193 | * PAGE_OFFSET), or device mappings in the vmalloc range (from | |
194 | * VMALLOC_START to VMALLOC_END). | |
195 | * | |
196 | * boot_hyp_pgd should only map two pages for the init code. | |
342cd0ab | 197 | */ |
4f728276 | 198 | void free_hyp_pgds(void) |
342cd0ab | 199 | { |
342cd0ab CD |
200 | unsigned long addr; |
201 | ||
d157f4a5 | 202 | free_boot_hyp_pgd(); |
4f728276 | 203 | |
d157f4a5 | 204 | mutex_lock(&kvm_hyp_pgd_mutex); |
5a677ce0 | 205 | |
4f728276 MZ |
206 | if (hyp_pgd) { |
207 | for (addr = PAGE_OFFSET; virt_addr_valid(addr); addr += PGDIR_SIZE) | |
d4cb9df5 | 208 | unmap_range(NULL, hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE); |
4f728276 | 209 | for (addr = VMALLOC_START; is_vmalloc_addr((void*)addr); addr += PGDIR_SIZE) |
d4cb9df5 MZ |
210 | unmap_range(NULL, hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE); |
211 | ||
4f728276 | 212 | kfree(hyp_pgd); |
d157f4a5 | 213 | hyp_pgd = NULL; |
4f728276 MZ |
214 | } |
215 | ||
342cd0ab CD |
216 | mutex_unlock(&kvm_hyp_pgd_mutex); |
217 | } | |
218 | ||
219 | static void create_hyp_pte_mappings(pmd_t *pmd, unsigned long start, | |
6060df84 MZ |
220 | unsigned long end, unsigned long pfn, |
221 | pgprot_t prot) | |
342cd0ab CD |
222 | { |
223 | pte_t *pte; | |
224 | unsigned long addr; | |
342cd0ab | 225 | |
3562c76d MZ |
226 | addr = start; |
227 | do { | |
6060df84 MZ |
228 | pte = pte_offset_kernel(pmd, addr); |
229 | kvm_set_pte(pte, pfn_pte(pfn, prot)); | |
4f728276 | 230 | get_page(virt_to_page(pte)); |
5a677ce0 | 231 | kvm_flush_dcache_to_poc(pte, sizeof(*pte)); |
6060df84 | 232 | pfn++; |
3562c76d | 233 | } while (addr += PAGE_SIZE, addr != end); |
342cd0ab CD |
234 | } |
235 | ||
236 | static int create_hyp_pmd_mappings(pud_t *pud, unsigned long start, | |
6060df84 MZ |
237 | unsigned long end, unsigned long pfn, |
238 | pgprot_t prot) | |
342cd0ab CD |
239 | { |
240 | pmd_t *pmd; | |
241 | pte_t *pte; | |
242 | unsigned long addr, next; | |
243 | ||
3562c76d MZ |
244 | addr = start; |
245 | do { | |
6060df84 | 246 | pmd = pmd_offset(pud, addr); |
342cd0ab CD |
247 | |
248 | BUG_ON(pmd_sect(*pmd)); | |
249 | ||
250 | if (pmd_none(*pmd)) { | |
6060df84 | 251 | pte = pte_alloc_one_kernel(NULL, addr); |
342cd0ab CD |
252 | if (!pte) { |
253 | kvm_err("Cannot allocate Hyp pte\n"); | |
254 | return -ENOMEM; | |
255 | } | |
256 | pmd_populate_kernel(NULL, pmd, pte); | |
4f728276 | 257 | get_page(virt_to_page(pmd)); |
5a677ce0 | 258 | kvm_flush_dcache_to_poc(pmd, sizeof(*pmd)); |
342cd0ab CD |
259 | } |
260 | ||
261 | next = pmd_addr_end(addr, end); | |
262 | ||
6060df84 MZ |
263 | create_hyp_pte_mappings(pmd, addr, next, pfn, prot); |
264 | pfn += (next - addr) >> PAGE_SHIFT; | |
3562c76d | 265 | } while (addr = next, addr != end); |
342cd0ab CD |
266 | |
267 | return 0; | |
268 | } | |
269 | ||
6060df84 MZ |
270 | static int __create_hyp_mappings(pgd_t *pgdp, |
271 | unsigned long start, unsigned long end, | |
272 | unsigned long pfn, pgprot_t prot) | |
342cd0ab | 273 | { |
342cd0ab CD |
274 | pgd_t *pgd; |
275 | pud_t *pud; | |
276 | pmd_t *pmd; | |
277 | unsigned long addr, next; | |
278 | int err = 0; | |
279 | ||
342cd0ab | 280 | mutex_lock(&kvm_hyp_pgd_mutex); |
3562c76d MZ |
281 | addr = start & PAGE_MASK; |
282 | end = PAGE_ALIGN(end); | |
283 | do { | |
6060df84 MZ |
284 | pgd = pgdp + pgd_index(addr); |
285 | pud = pud_offset(pgd, addr); | |
342cd0ab CD |
286 | |
287 | if (pud_none_or_clear_bad(pud)) { | |
6060df84 | 288 | pmd = pmd_alloc_one(NULL, addr); |
342cd0ab CD |
289 | if (!pmd) { |
290 | kvm_err("Cannot allocate Hyp pmd\n"); | |
291 | err = -ENOMEM; | |
292 | goto out; | |
293 | } | |
294 | pud_populate(NULL, pud, pmd); | |
4f728276 | 295 | get_page(virt_to_page(pud)); |
5a677ce0 | 296 | kvm_flush_dcache_to_poc(pud, sizeof(*pud)); |
342cd0ab CD |
297 | } |
298 | ||
299 | next = pgd_addr_end(addr, end); | |
6060df84 | 300 | err = create_hyp_pmd_mappings(pud, addr, next, pfn, prot); |
342cd0ab CD |
301 | if (err) |
302 | goto out; | |
6060df84 | 303 | pfn += (next - addr) >> PAGE_SHIFT; |
3562c76d | 304 | } while (addr = next, addr != end); |
342cd0ab CD |
305 | out: |
306 | mutex_unlock(&kvm_hyp_pgd_mutex); | |
307 | return err; | |
308 | } | |
309 | ||
310 | /** | |
06e8c3b0 | 311 | * create_hyp_mappings - duplicate a kernel virtual address range in Hyp mode |
342cd0ab CD |
312 | * @from: The virtual kernel start address of the range |
313 | * @to: The virtual kernel end address of the range (exclusive) | |
314 | * | |
06e8c3b0 MZ |
315 | * The same virtual address as the kernel virtual address is also used |
316 | * in Hyp-mode mapping (modulo HYP_PAGE_OFFSET) to the same underlying | |
317 | * physical pages. | |
342cd0ab CD |
318 | */ |
319 | int create_hyp_mappings(void *from, void *to) | |
320 | { | |
6060df84 MZ |
321 | unsigned long phys_addr = virt_to_phys(from); |
322 | unsigned long start = KERN_TO_HYP((unsigned long)from); | |
323 | unsigned long end = KERN_TO_HYP((unsigned long)to); | |
324 | ||
325 | /* Check for a valid kernel memory mapping */ | |
326 | if (!virt_addr_valid(from) || !virt_addr_valid(to - 1)) | |
327 | return -EINVAL; | |
328 | ||
329 | return __create_hyp_mappings(hyp_pgd, start, end, | |
330 | __phys_to_pfn(phys_addr), PAGE_HYP); | |
342cd0ab CD |
331 | } |
332 | ||
333 | /** | |
06e8c3b0 MZ |
334 | * create_hyp_io_mappings - duplicate a kernel IO mapping into Hyp mode |
335 | * @from: The kernel start VA of the range | |
336 | * @to: The kernel end VA of the range (exclusive) | |
6060df84 | 337 | * @phys_addr: The physical start address which gets mapped |
06e8c3b0 MZ |
338 | * |
339 | * The resulting HYP VA is the same as the kernel VA, modulo | |
340 | * HYP_PAGE_OFFSET. | |
342cd0ab | 341 | */ |
6060df84 | 342 | int create_hyp_io_mappings(void *from, void *to, phys_addr_t phys_addr) |
342cd0ab | 343 | { |
6060df84 MZ |
344 | unsigned long start = KERN_TO_HYP((unsigned long)from); |
345 | unsigned long end = KERN_TO_HYP((unsigned long)to); | |
346 | ||
347 | /* Check for a valid kernel IO mapping */ | |
348 | if (!is_vmalloc_addr(from) || !is_vmalloc_addr(to - 1)) | |
349 | return -EINVAL; | |
350 | ||
351 | return __create_hyp_mappings(hyp_pgd, start, end, | |
352 | __phys_to_pfn(phys_addr), PAGE_HYP_DEVICE); | |
342cd0ab CD |
353 | } |
354 | ||
d5d8184d CD |
355 | /** |
356 | * kvm_alloc_stage2_pgd - allocate level-1 table for stage-2 translation. | |
357 | * @kvm: The KVM struct pointer for the VM. | |
358 | * | |
359 | * Allocates the 1st level table only of size defined by S2_PGD_ORDER (can | |
360 | * support either full 40-bit input addresses or limited to 32-bit input | |
361 | * addresses). Clears the allocated pages. | |
362 | * | |
363 | * Note we don't need locking here as this is only called when the VM is | |
364 | * created, which can only be done once. | |
365 | */ | |
366 | int kvm_alloc_stage2_pgd(struct kvm *kvm) | |
367 | { | |
368 | pgd_t *pgd; | |
369 | ||
370 | if (kvm->arch.pgd != NULL) { | |
371 | kvm_err("kvm_arch already initialized?\n"); | |
372 | return -EINVAL; | |
373 | } | |
374 | ||
375 | pgd = (pgd_t *)__get_free_pages(GFP_KERNEL, S2_PGD_ORDER); | |
376 | if (!pgd) | |
377 | return -ENOMEM; | |
378 | ||
d5d8184d | 379 | memset(pgd, 0, PTRS_PER_S2_PGD * sizeof(pgd_t)); |
c62ee2b2 | 380 | kvm_clean_pgd(pgd); |
d5d8184d CD |
381 | kvm->arch.pgd = pgd; |
382 | ||
383 | return 0; | |
384 | } | |
385 | ||
d5d8184d CD |
386 | /** |
387 | * unmap_stage2_range -- Clear stage2 page table entries to unmap a range | |
388 | * @kvm: The VM pointer | |
389 | * @start: The intermediate physical base address of the range to unmap | |
390 | * @size: The size of the area to unmap | |
391 | * | |
392 | * Clear a range of stage-2 mappings, lowering the various ref-counts. Must | |
393 | * be called while holding mmu_lock (unless for freeing the stage2 pgd before | |
394 | * destroying the VM), otherwise another faulting VCPU may come in and mess | |
395 | * with things behind our backs. | |
396 | */ | |
397 | static void unmap_stage2_range(struct kvm *kvm, phys_addr_t start, u64 size) | |
398 | { | |
d4cb9df5 | 399 | unmap_range(kvm, kvm->arch.pgd, start, size); |
d5d8184d CD |
400 | } |
401 | ||
402 | /** | |
403 | * kvm_free_stage2_pgd - free all stage-2 tables | |
404 | * @kvm: The KVM struct pointer for the VM. | |
405 | * | |
406 | * Walks the level-1 page table pointed to by kvm->arch.pgd and frees all | |
407 | * underlying level-2 and level-3 tables before freeing the actual level-1 table | |
408 | * and setting the struct pointer to NULL. | |
409 | * | |
410 | * Note we don't need locking here as this is only called when the VM is | |
411 | * destroyed, which can only be done once. | |
412 | */ | |
413 | void kvm_free_stage2_pgd(struct kvm *kvm) | |
414 | { | |
415 | if (kvm->arch.pgd == NULL) | |
416 | return; | |
417 | ||
418 | unmap_stage2_range(kvm, 0, KVM_PHYS_SIZE); | |
419 | free_pages((unsigned long)kvm->arch.pgd, S2_PGD_ORDER); | |
420 | kvm->arch.pgd = NULL; | |
421 | } | |
422 | ||
423 | ||
424 | static int stage2_set_pte(struct kvm *kvm, struct kvm_mmu_memory_cache *cache, | |
425 | phys_addr_t addr, const pte_t *new_pte, bool iomap) | |
426 | { | |
427 | pgd_t *pgd; | |
428 | pud_t *pud; | |
429 | pmd_t *pmd; | |
430 | pte_t *pte, old_pte; | |
431 | ||
432 | /* Create 2nd stage page table mapping - Level 1 */ | |
433 | pgd = kvm->arch.pgd + pgd_index(addr); | |
434 | pud = pud_offset(pgd, addr); | |
435 | if (pud_none(*pud)) { | |
436 | if (!cache) | |
437 | return 0; /* ignore calls from kvm_set_spte_hva */ | |
438 | pmd = mmu_memory_cache_alloc(cache); | |
439 | pud_populate(NULL, pud, pmd); | |
d5d8184d | 440 | get_page(virt_to_page(pud)); |
c62ee2b2 MZ |
441 | } |
442 | ||
443 | pmd = pmd_offset(pud, addr); | |
d5d8184d CD |
444 | |
445 | /* Create 2nd stage page table mapping - Level 2 */ | |
446 | if (pmd_none(*pmd)) { | |
447 | if (!cache) | |
448 | return 0; /* ignore calls from kvm_set_spte_hva */ | |
449 | pte = mmu_memory_cache_alloc(cache); | |
c62ee2b2 | 450 | kvm_clean_pte(pte); |
d5d8184d | 451 | pmd_populate_kernel(NULL, pmd, pte); |
d5d8184d | 452 | get_page(virt_to_page(pmd)); |
c62ee2b2 MZ |
453 | } |
454 | ||
455 | pte = pte_offset_kernel(pmd, addr); | |
d5d8184d CD |
456 | |
457 | if (iomap && pte_present(*pte)) | |
458 | return -EFAULT; | |
459 | ||
460 | /* Create 2nd stage page table mapping - Level 3 */ | |
461 | old_pte = *pte; | |
462 | kvm_set_pte(pte, *new_pte); | |
463 | if (pte_present(old_pte)) | |
48762767 | 464 | kvm_tlb_flush_vmid_ipa(kvm, addr); |
d5d8184d CD |
465 | else |
466 | get_page(virt_to_page(pte)); | |
467 | ||
468 | return 0; | |
469 | } | |
470 | ||
471 | /** | |
472 | * kvm_phys_addr_ioremap - map a device range to guest IPA | |
473 | * | |
474 | * @kvm: The KVM pointer | |
475 | * @guest_ipa: The IPA at which to insert the mapping | |
476 | * @pa: The physical address of the device | |
477 | * @size: The size of the mapping | |
478 | */ | |
479 | int kvm_phys_addr_ioremap(struct kvm *kvm, phys_addr_t guest_ipa, | |
480 | phys_addr_t pa, unsigned long size) | |
481 | { | |
482 | phys_addr_t addr, end; | |
483 | int ret = 0; | |
484 | unsigned long pfn; | |
485 | struct kvm_mmu_memory_cache cache = { 0, }; | |
486 | ||
487 | end = (guest_ipa + size + PAGE_SIZE - 1) & PAGE_MASK; | |
488 | pfn = __phys_to_pfn(pa); | |
489 | ||
490 | for (addr = guest_ipa; addr < end; addr += PAGE_SIZE) { | |
c62ee2b2 | 491 | pte_t pte = pfn_pte(pfn, PAGE_S2_DEVICE); |
d5d8184d CD |
492 | |
493 | ret = mmu_topup_memory_cache(&cache, 2, 2); | |
494 | if (ret) | |
495 | goto out; | |
496 | spin_lock(&kvm->mmu_lock); | |
497 | ret = stage2_set_pte(kvm, &cache, addr, &pte, true); | |
498 | spin_unlock(&kvm->mmu_lock); | |
499 | if (ret) | |
500 | goto out; | |
501 | ||
502 | pfn++; | |
503 | } | |
504 | ||
505 | out: | |
506 | mmu_free_memory_cache(&cache); | |
507 | return ret; | |
508 | } | |
509 | ||
94f8e641 CD |
510 | static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa, |
511 | gfn_t gfn, struct kvm_memory_slot *memslot, | |
512 | unsigned long fault_status) | |
513 | { | |
514 | pte_t new_pte; | |
515 | pfn_t pfn; | |
516 | int ret; | |
517 | bool write_fault, writable; | |
518 | unsigned long mmu_seq; | |
519 | struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache; | |
520 | ||
7393b599 | 521 | write_fault = kvm_is_write_fault(kvm_vcpu_get_hsr(vcpu)); |
94f8e641 CD |
522 | if (fault_status == FSC_PERM && !write_fault) { |
523 | kvm_err("Unexpected L2 read permission error\n"); | |
524 | return -EFAULT; | |
525 | } | |
526 | ||
527 | /* We need minimum second+third level pages */ | |
528 | ret = mmu_topup_memory_cache(memcache, 2, KVM_NR_MEM_OBJS); | |
529 | if (ret) | |
530 | return ret; | |
531 | ||
532 | mmu_seq = vcpu->kvm->mmu_notifier_seq; | |
533 | /* | |
534 | * Ensure the read of mmu_notifier_seq happens before we call | |
535 | * gfn_to_pfn_prot (which calls get_user_pages), so that we don't risk | |
536 | * the page we just got a reference to gets unmapped before we have a | |
537 | * chance to grab the mmu_lock, which ensure that if the page gets | |
538 | * unmapped afterwards, the call to kvm_unmap_hva will take it away | |
539 | * from us again properly. This smp_rmb() interacts with the smp_wmb() | |
540 | * in kvm_mmu_notifier_invalidate_<page|range_end>. | |
541 | */ | |
542 | smp_rmb(); | |
543 | ||
544 | pfn = gfn_to_pfn_prot(vcpu->kvm, gfn, write_fault, &writable); | |
545 | if (is_error_pfn(pfn)) | |
546 | return -EFAULT; | |
547 | ||
548 | new_pte = pfn_pte(pfn, PAGE_S2); | |
549 | coherent_icache_guest_page(vcpu->kvm, gfn); | |
550 | ||
551 | spin_lock(&vcpu->kvm->mmu_lock); | |
552 | if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) | |
553 | goto out_unlock; | |
554 | if (writable) { | |
c62ee2b2 | 555 | kvm_set_s2pte_writable(&new_pte); |
94f8e641 CD |
556 | kvm_set_pfn_dirty(pfn); |
557 | } | |
558 | stage2_set_pte(vcpu->kvm, memcache, fault_ipa, &new_pte, false); | |
559 | ||
560 | out_unlock: | |
561 | spin_unlock(&vcpu->kvm->mmu_lock); | |
562 | kvm_release_pfn_clean(pfn); | |
563 | return 0; | |
564 | } | |
565 | ||
566 | /** | |
567 | * kvm_handle_guest_abort - handles all 2nd stage aborts | |
568 | * @vcpu: the VCPU pointer | |
569 | * @run: the kvm_run structure | |
570 | * | |
571 | * Any abort that gets to the host is almost guaranteed to be caused by a | |
572 | * missing second stage translation table entry, which can mean that either the | |
573 | * guest simply needs more memory and we must allocate an appropriate page or it | |
574 | * can mean that the guest tried to access I/O memory, which is emulated by user | |
575 | * space. The distinction is based on the IPA causing the fault and whether this | |
576 | * memory region has been registered as standard RAM by user space. | |
577 | */ | |
342cd0ab CD |
578 | int kvm_handle_guest_abort(struct kvm_vcpu *vcpu, struct kvm_run *run) |
579 | { | |
94f8e641 CD |
580 | unsigned long fault_status; |
581 | phys_addr_t fault_ipa; | |
582 | struct kvm_memory_slot *memslot; | |
583 | bool is_iabt; | |
584 | gfn_t gfn; | |
585 | int ret, idx; | |
586 | ||
52d1dba9 | 587 | is_iabt = kvm_vcpu_trap_is_iabt(vcpu); |
7393b599 | 588 | fault_ipa = kvm_vcpu_get_fault_ipa(vcpu); |
94f8e641 | 589 | |
7393b599 MZ |
590 | trace_kvm_guest_fault(*vcpu_pc(vcpu), kvm_vcpu_get_hsr(vcpu), |
591 | kvm_vcpu_get_hfar(vcpu), fault_ipa); | |
94f8e641 CD |
592 | |
593 | /* Check the stage-2 fault is trans. fault or write fault */ | |
1cc287dd | 594 | fault_status = kvm_vcpu_trap_get_fault(vcpu); |
94f8e641 | 595 | if (fault_status != FSC_FAULT && fault_status != FSC_PERM) { |
52d1dba9 MZ |
596 | kvm_err("Unsupported fault status: EC=%#x DFCS=%#lx\n", |
597 | kvm_vcpu_trap_get_class(vcpu), fault_status); | |
94f8e641 CD |
598 | return -EFAULT; |
599 | } | |
600 | ||
601 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
602 | ||
603 | gfn = fault_ipa >> PAGE_SHIFT; | |
604 | if (!kvm_is_visible_gfn(vcpu->kvm, gfn)) { | |
605 | if (is_iabt) { | |
606 | /* Prefetch Abort on I/O address */ | |
7393b599 | 607 | kvm_inject_pabt(vcpu, kvm_vcpu_get_hfar(vcpu)); |
94f8e641 CD |
608 | ret = 1; |
609 | goto out_unlock; | |
610 | } | |
611 | ||
612 | if (fault_status != FSC_FAULT) { | |
613 | kvm_err("Unsupported fault status on io memory: %#lx\n", | |
614 | fault_status); | |
615 | ret = -EFAULT; | |
616 | goto out_unlock; | |
617 | } | |
618 | ||
cfe3950c MZ |
619 | /* |
620 | * The IPA is reported as [MAX:12], so we need to | |
621 | * complement it with the bottom 12 bits from the | |
622 | * faulting VA. This is always 12 bits, irrespective | |
623 | * of the page size. | |
624 | */ | |
625 | fault_ipa |= kvm_vcpu_get_hfar(vcpu) & ((1 << 12) - 1); | |
45e96ea6 | 626 | ret = io_mem_abort(vcpu, run, fault_ipa); |
94f8e641 CD |
627 | goto out_unlock; |
628 | } | |
629 | ||
630 | memslot = gfn_to_memslot(vcpu->kvm, gfn); | |
94f8e641 CD |
631 | |
632 | ret = user_mem_abort(vcpu, fault_ipa, gfn, memslot, fault_status); | |
633 | if (ret == 0) | |
634 | ret = 1; | |
635 | out_unlock: | |
636 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
637 | return ret; | |
342cd0ab CD |
638 | } |
639 | ||
d5d8184d CD |
640 | static void handle_hva_to_gpa(struct kvm *kvm, |
641 | unsigned long start, | |
642 | unsigned long end, | |
643 | void (*handler)(struct kvm *kvm, | |
644 | gpa_t gpa, void *data), | |
645 | void *data) | |
646 | { | |
647 | struct kvm_memslots *slots; | |
648 | struct kvm_memory_slot *memslot; | |
649 | ||
650 | slots = kvm_memslots(kvm); | |
651 | ||
652 | /* we only care about the pages that the guest sees */ | |
653 | kvm_for_each_memslot(memslot, slots) { | |
654 | unsigned long hva_start, hva_end; | |
655 | gfn_t gfn, gfn_end; | |
656 | ||
657 | hva_start = max(start, memslot->userspace_addr); | |
658 | hva_end = min(end, memslot->userspace_addr + | |
659 | (memslot->npages << PAGE_SHIFT)); | |
660 | if (hva_start >= hva_end) | |
661 | continue; | |
662 | ||
663 | /* | |
664 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
665 | * {gfn_start, gfn_start+1, ..., gfn_end-1}. | |
666 | */ | |
667 | gfn = hva_to_gfn_memslot(hva_start, memslot); | |
668 | gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot); | |
669 | ||
670 | for (; gfn < gfn_end; ++gfn) { | |
671 | gpa_t gpa = gfn << PAGE_SHIFT; | |
672 | handler(kvm, gpa, data); | |
673 | } | |
674 | } | |
675 | } | |
676 | ||
677 | static void kvm_unmap_hva_handler(struct kvm *kvm, gpa_t gpa, void *data) | |
678 | { | |
679 | unmap_stage2_range(kvm, gpa, PAGE_SIZE); | |
d5d8184d CD |
680 | } |
681 | ||
682 | int kvm_unmap_hva(struct kvm *kvm, unsigned long hva) | |
683 | { | |
684 | unsigned long end = hva + PAGE_SIZE; | |
685 | ||
686 | if (!kvm->arch.pgd) | |
687 | return 0; | |
688 | ||
689 | trace_kvm_unmap_hva(hva); | |
690 | handle_hva_to_gpa(kvm, hva, end, &kvm_unmap_hva_handler, NULL); | |
691 | return 0; | |
692 | } | |
693 | ||
694 | int kvm_unmap_hva_range(struct kvm *kvm, | |
695 | unsigned long start, unsigned long end) | |
696 | { | |
697 | if (!kvm->arch.pgd) | |
698 | return 0; | |
699 | ||
700 | trace_kvm_unmap_hva_range(start, end); | |
701 | handle_hva_to_gpa(kvm, start, end, &kvm_unmap_hva_handler, NULL); | |
702 | return 0; | |
703 | } | |
704 | ||
705 | static void kvm_set_spte_handler(struct kvm *kvm, gpa_t gpa, void *data) | |
706 | { | |
707 | pte_t *pte = (pte_t *)data; | |
708 | ||
709 | stage2_set_pte(kvm, NULL, gpa, pte, false); | |
710 | } | |
711 | ||
712 | ||
713 | void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte) | |
714 | { | |
715 | unsigned long end = hva + PAGE_SIZE; | |
716 | pte_t stage2_pte; | |
717 | ||
718 | if (!kvm->arch.pgd) | |
719 | return; | |
720 | ||
721 | trace_kvm_set_spte_hva(hva); | |
722 | stage2_pte = pfn_pte(pte_pfn(pte), PAGE_S2); | |
723 | handle_hva_to_gpa(kvm, hva, end, &kvm_set_spte_handler, &stage2_pte); | |
724 | } | |
725 | ||
726 | void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu) | |
727 | { | |
728 | mmu_free_memory_cache(&vcpu->arch.mmu_page_cache); | |
729 | } | |
730 | ||
342cd0ab CD |
731 | phys_addr_t kvm_mmu_get_httbr(void) |
732 | { | |
342cd0ab CD |
733 | return virt_to_phys(hyp_pgd); |
734 | } | |
735 | ||
5a677ce0 MZ |
736 | phys_addr_t kvm_mmu_get_boot_httbr(void) |
737 | { | |
738 | return virt_to_phys(boot_hyp_pgd); | |
739 | } | |
740 | ||
741 | phys_addr_t kvm_get_idmap_vector(void) | |
742 | { | |
743 | return hyp_idmap_vector; | |
744 | } | |
745 | ||
342cd0ab CD |
746 | int kvm_mmu_init(void) |
747 | { | |
2fb41059 MZ |
748 | int err; |
749 | ||
5a677ce0 MZ |
750 | hyp_idmap_start = virt_to_phys(__hyp_idmap_text_start); |
751 | hyp_idmap_end = virt_to_phys(__hyp_idmap_text_end); | |
752 | hyp_idmap_vector = virt_to_phys(__kvm_hyp_init); | |
753 | ||
754 | if ((hyp_idmap_start ^ hyp_idmap_end) & PAGE_MASK) { | |
755 | /* | |
756 | * Our init code is crossing a page boundary. Allocate | |
757 | * a bounce page, copy the code over and use that. | |
758 | */ | |
759 | size_t len = __hyp_idmap_text_end - __hyp_idmap_text_start; | |
760 | phys_addr_t phys_base; | |
761 | ||
762 | init_bounce_page = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
763 | if (!init_bounce_page) { | |
764 | kvm_err("Couldn't allocate HYP init bounce page\n"); | |
765 | err = -ENOMEM; | |
766 | goto out; | |
767 | } | |
768 | ||
769 | memcpy(init_bounce_page, __hyp_idmap_text_start, len); | |
770 | /* | |
771 | * Warning: the code we just copied to the bounce page | |
772 | * must be flushed to the point of coherency. | |
773 | * Otherwise, the data may be sitting in L2, and HYP | |
774 | * mode won't be able to observe it as it runs with | |
775 | * caches off at that point. | |
776 | */ | |
777 | kvm_flush_dcache_to_poc(init_bounce_page, len); | |
778 | ||
779 | phys_base = virt_to_phys(init_bounce_page); | |
780 | hyp_idmap_vector += phys_base - hyp_idmap_start; | |
781 | hyp_idmap_start = phys_base; | |
782 | hyp_idmap_end = phys_base + len; | |
783 | ||
784 | kvm_info("Using HYP init bounce page @%lx\n", | |
785 | (unsigned long)phys_base); | |
786 | } | |
787 | ||
2fb41059 | 788 | hyp_pgd = kzalloc(PTRS_PER_PGD * sizeof(pgd_t), GFP_KERNEL); |
5a677ce0 MZ |
789 | boot_hyp_pgd = kzalloc(PTRS_PER_PGD * sizeof(pgd_t), GFP_KERNEL); |
790 | if (!hyp_pgd || !boot_hyp_pgd) { | |
d5d8184d | 791 | kvm_err("Hyp mode PGD not allocated\n"); |
2fb41059 MZ |
792 | err = -ENOMEM; |
793 | goto out; | |
794 | } | |
795 | ||
796 | /* Create the idmap in the boot page tables */ | |
797 | err = __create_hyp_mappings(boot_hyp_pgd, | |
798 | hyp_idmap_start, hyp_idmap_end, | |
799 | __phys_to_pfn(hyp_idmap_start), | |
800 | PAGE_HYP); | |
801 | ||
802 | if (err) { | |
803 | kvm_err("Failed to idmap %lx-%lx\n", | |
804 | hyp_idmap_start, hyp_idmap_end); | |
805 | goto out; | |
d5d8184d CD |
806 | } |
807 | ||
5a677ce0 MZ |
808 | /* Map the very same page at the trampoline VA */ |
809 | err = __create_hyp_mappings(boot_hyp_pgd, | |
810 | TRAMPOLINE_VA, TRAMPOLINE_VA + PAGE_SIZE, | |
811 | __phys_to_pfn(hyp_idmap_start), | |
812 | PAGE_HYP); | |
813 | if (err) { | |
814 | kvm_err("Failed to map trampoline @%lx into boot HYP pgd\n", | |
815 | TRAMPOLINE_VA); | |
816 | goto out; | |
817 | } | |
818 | ||
819 | /* Map the same page again into the runtime page tables */ | |
820 | err = __create_hyp_mappings(hyp_pgd, | |
821 | TRAMPOLINE_VA, TRAMPOLINE_VA + PAGE_SIZE, | |
822 | __phys_to_pfn(hyp_idmap_start), | |
823 | PAGE_HYP); | |
824 | if (err) { | |
825 | kvm_err("Failed to map trampoline @%lx into runtime HYP pgd\n", | |
826 | TRAMPOLINE_VA); | |
827 | goto out; | |
828 | } | |
829 | ||
d5d8184d | 830 | return 0; |
2fb41059 | 831 | out: |
4f728276 | 832 | free_hyp_pgds(); |
2fb41059 | 833 | return err; |
342cd0ab | 834 | } |