1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) 2020 Google LLC
4 * Author: Quentin Perret <qperret@google.com>
5 */
6
7 #include <linux/kvm_host.h>
8 #include <asm/kvm_emulate.h>
9 #include <asm/kvm_hyp.h>
10 #include <asm/kvm_mmu.h>
11 #include <asm/kvm_pgtable.h>
12 #include <asm/kvm_pkvm.h>
13 #include <asm/stage2_pgtable.h>
14
15 #include <hyp/fault.h>
16
17 #include <nvhe/gfp.h>
18 #include <nvhe/memory.h>
19 #include <nvhe/mem_protect.h>
20 #include <nvhe/mm.h>
21
22 #define KVM_HOST_S2_FLAGS (KVM_PGTABLE_S2_AS_S1 | KVM_PGTABLE_S2_IDMAP)
23
24 struct host_mmu host_mmu;
25
26 static struct hyp_pool host_s2_pool;
27
28 static DEFINE_PER_CPU(struct pkvm_hyp_vm *, __current_vm);
29 #define current_vm (*this_cpu_ptr(&__current_vm))
30
guest_lock_component(struct pkvm_hyp_vm * vm)31 static void guest_lock_component(struct pkvm_hyp_vm *vm)
32 {
33 hyp_spin_lock(&vm->lock);
34 current_vm = vm;
35 }
36
guest_unlock_component(struct pkvm_hyp_vm * vm)37 static void guest_unlock_component(struct pkvm_hyp_vm *vm)
38 {
39 current_vm = NULL;
40 hyp_spin_unlock(&vm->lock);
41 }
42
host_lock_component(void)43 static void host_lock_component(void)
44 {
45 hyp_spin_lock(&host_mmu.lock);
46 }
47
host_unlock_component(void)48 static void host_unlock_component(void)
49 {
50 hyp_spin_unlock(&host_mmu.lock);
51 }
52
hyp_lock_component(void)53 static void hyp_lock_component(void)
54 {
55 hyp_spin_lock(&pkvm_pgd_lock);
56 }
57
hyp_unlock_component(void)58 static void hyp_unlock_component(void)
59 {
60 hyp_spin_unlock(&pkvm_pgd_lock);
61 }
62
63 #define for_each_hyp_page(__p, __st, __sz) \
64 for (struct hyp_page *__p = hyp_phys_to_page(__st), \
65 *__e = __p + ((__sz) >> PAGE_SHIFT); \
66 __p < __e; __p++)
67
host_s2_zalloc_pages_exact(size_t size)68 static void *host_s2_zalloc_pages_exact(size_t size)
69 {
70 void *addr = hyp_alloc_pages(&host_s2_pool, get_order(size));
71
72 hyp_split_page(hyp_virt_to_page(addr));
73
74 /*
75 * The size of concatenated PGDs is always a power of two of PAGE_SIZE,
76 * so there should be no need to free any of the tail pages to make the
77 * allocation exact.
78 */
79 WARN_ON(size != (PAGE_SIZE << get_order(size)));
80
81 return addr;
82 }
83
host_s2_zalloc_page(void * pool)84 static void *host_s2_zalloc_page(void *pool)
85 {
86 return hyp_alloc_pages(pool, 0);
87 }
88
host_s2_get_page(void * addr)89 static void host_s2_get_page(void *addr)
90 {
91 hyp_get_page(&host_s2_pool, addr);
92 }
93
host_s2_put_page(void * addr)94 static void host_s2_put_page(void *addr)
95 {
96 hyp_put_page(&host_s2_pool, addr);
97 }
98
host_s2_free_unlinked_table(void * addr,s8 level)99 static void host_s2_free_unlinked_table(void *addr, s8 level)
100 {
101 kvm_pgtable_stage2_free_unlinked(&host_mmu.mm_ops, addr, level);
102 }
103
prepare_s2_pool(void * pgt_pool_base)104 static int prepare_s2_pool(void *pgt_pool_base)
105 {
106 unsigned long nr_pages, pfn;
107 int ret;
108
109 pfn = hyp_virt_to_pfn(pgt_pool_base);
110 nr_pages = host_s2_pgtable_pages();
111 ret = hyp_pool_init(&host_s2_pool, pfn, nr_pages, 0);
112 if (ret)
113 return ret;
114
115 host_mmu.mm_ops = (struct kvm_pgtable_mm_ops) {
116 .zalloc_pages_exact = host_s2_zalloc_pages_exact,
117 .zalloc_page = host_s2_zalloc_page,
118 .free_unlinked_table = host_s2_free_unlinked_table,
119 .phys_to_virt = hyp_phys_to_virt,
120 .virt_to_phys = hyp_virt_to_phys,
121 .page_count = hyp_page_count,
122 .get_page = host_s2_get_page,
123 .put_page = host_s2_put_page,
124 };
125
126 return 0;
127 }
128
prepare_host_vtcr(void)129 static void prepare_host_vtcr(void)
130 {
131 u32 parange, phys_shift;
132
133 /* The host stage 2 is id-mapped, so use parange for T0SZ */
134 parange = kvm_get_parange(id_aa64mmfr0_el1_sys_val);
135 phys_shift = id_aa64mmfr0_parange_to_phys_shift(parange);
136
137 host_mmu.arch.mmu.vtcr = kvm_get_vtcr(id_aa64mmfr0_el1_sys_val,
138 id_aa64mmfr1_el1_sys_val, phys_shift);
139 }
140
141 static bool host_stage2_force_pte_cb(u64 addr, u64 end, enum kvm_pgtable_prot prot);
142
kvm_host_prepare_stage2(void * pgt_pool_base)143 int kvm_host_prepare_stage2(void *pgt_pool_base)
144 {
145 struct kvm_s2_mmu *mmu = &host_mmu.arch.mmu;
146 int ret;
147
148 prepare_host_vtcr();
149 hyp_spin_lock_init(&host_mmu.lock);
150 mmu->arch = &host_mmu.arch;
151
152 ret = prepare_s2_pool(pgt_pool_base);
153 if (ret)
154 return ret;
155
156 ret = __kvm_pgtable_stage2_init(&host_mmu.pgt, mmu,
157 &host_mmu.mm_ops, KVM_HOST_S2_FLAGS,
158 host_stage2_force_pte_cb);
159 if (ret)
160 return ret;
161
162 mmu->pgd_phys = __hyp_pa(host_mmu.pgt.pgd);
163 mmu->pgt = &host_mmu.pgt;
164 atomic64_set(&mmu->vmid.id, 0);
165
166 return 0;
167 }
168
guest_s2_zalloc_pages_exact(size_t size)169 static void *guest_s2_zalloc_pages_exact(size_t size)
170 {
171 void *addr = hyp_alloc_pages(¤t_vm->pool, get_order(size));
172
173 WARN_ON(size != (PAGE_SIZE << get_order(size)));
174 hyp_split_page(hyp_virt_to_page(addr));
175
176 return addr;
177 }
178
guest_s2_free_pages_exact(void * addr,unsigned long size)179 static void guest_s2_free_pages_exact(void *addr, unsigned long size)
180 {
181 u8 order = get_order(size);
182 unsigned int i;
183
184 for (i = 0; i < (1 << order); i++)
185 hyp_put_page(¤t_vm->pool, addr + (i * PAGE_SIZE));
186 }
187
guest_s2_zalloc_page(void * mc)188 static void *guest_s2_zalloc_page(void *mc)
189 {
190 struct hyp_page *p;
191 void *addr;
192
193 addr = hyp_alloc_pages(¤t_vm->pool, 0);
194 if (addr)
195 return addr;
196
197 addr = pop_hyp_memcache(mc, hyp_phys_to_virt);
198 if (!addr)
199 return addr;
200
201 memset(addr, 0, PAGE_SIZE);
202 p = hyp_virt_to_page(addr);
203 p->refcount = 1;
204 p->order = 0;
205
206 return addr;
207 }
208
guest_s2_get_page(void * addr)209 static void guest_s2_get_page(void *addr)
210 {
211 hyp_get_page(¤t_vm->pool, addr);
212 }
213
guest_s2_put_page(void * addr)214 static void guest_s2_put_page(void *addr)
215 {
216 hyp_put_page(¤t_vm->pool, addr);
217 }
218
__apply_guest_page(void * va,size_t size,void (* func)(void * addr,size_t size))219 static void __apply_guest_page(void *va, size_t size,
220 void (*func)(void *addr, size_t size))
221 {
222 size += va - PTR_ALIGN_DOWN(va, PAGE_SIZE);
223 va = PTR_ALIGN_DOWN(va, PAGE_SIZE);
224 size = PAGE_ALIGN(size);
225
226 while (size) {
227 size_t map_size = PAGE_SIZE;
228 void *map;
229
230 if (IS_ALIGNED((unsigned long)va, PMD_SIZE) && size >= PMD_SIZE)
231 map = hyp_fixblock_map(__hyp_pa(va), &map_size);
232 else
233 map = hyp_fixmap_map(__hyp_pa(va));
234
235 func(map, map_size);
236
237 if (map_size == PMD_SIZE)
238 hyp_fixblock_unmap();
239 else
240 hyp_fixmap_unmap();
241
242 size -= map_size;
243 va += map_size;
244 }
245 }
246
clean_dcache_guest_page(void * va,size_t size)247 static void clean_dcache_guest_page(void *va, size_t size)
248 {
249 __apply_guest_page(va, size, __clean_dcache_guest_page);
250 }
251
invalidate_icache_guest_page(void * va,size_t size)252 static void invalidate_icache_guest_page(void *va, size_t size)
253 {
254 __apply_guest_page(va, size, __invalidate_icache_guest_page);
255 }
256
kvm_guest_prepare_stage2(struct pkvm_hyp_vm * vm,void * pgd)257 int kvm_guest_prepare_stage2(struct pkvm_hyp_vm *vm, void *pgd)
258 {
259 struct kvm_s2_mmu *mmu = &vm->kvm.arch.mmu;
260 unsigned long nr_pages;
261 int ret;
262
263 nr_pages = kvm_pgtable_stage2_pgd_size(mmu->vtcr) >> PAGE_SHIFT;
264 ret = hyp_pool_init(&vm->pool, hyp_virt_to_pfn(pgd), nr_pages, 0);
265 if (ret)
266 return ret;
267
268 hyp_spin_lock_init(&vm->lock);
269 vm->mm_ops = (struct kvm_pgtable_mm_ops) {
270 .zalloc_pages_exact = guest_s2_zalloc_pages_exact,
271 .free_pages_exact = guest_s2_free_pages_exact,
272 .zalloc_page = guest_s2_zalloc_page,
273 .phys_to_virt = hyp_phys_to_virt,
274 .virt_to_phys = hyp_virt_to_phys,
275 .page_count = hyp_page_count,
276 .get_page = guest_s2_get_page,
277 .put_page = guest_s2_put_page,
278 .dcache_clean_inval_poc = clean_dcache_guest_page,
279 .icache_inval_pou = invalidate_icache_guest_page,
280 };
281
282 guest_lock_component(vm);
283 ret = __kvm_pgtable_stage2_init(mmu->pgt, mmu, &vm->mm_ops, 0, NULL);
284 guest_unlock_component(vm);
285 if (ret)
286 return ret;
287
288 vm->kvm.arch.mmu.pgd_phys = __hyp_pa(vm->pgt.pgd);
289
290 return 0;
291 }
292
reclaim_pgtable_pages(struct pkvm_hyp_vm * vm,struct kvm_hyp_memcache * mc)293 void reclaim_pgtable_pages(struct pkvm_hyp_vm *vm, struct kvm_hyp_memcache *mc)
294 {
295 struct hyp_page *page;
296 void *addr;
297
298 /* Dump all pgtable pages in the hyp_pool */
299 guest_lock_component(vm);
300 kvm_pgtable_stage2_destroy(&vm->pgt);
301 vm->kvm.arch.mmu.pgd_phys = 0ULL;
302 guest_unlock_component(vm);
303
304 /* Drain the hyp_pool into the memcache */
305 addr = hyp_alloc_pages(&vm->pool, 0);
306 while (addr) {
307 page = hyp_virt_to_page(addr);
308 page->refcount = 0;
309 page->order = 0;
310 push_hyp_memcache(mc, addr, hyp_virt_to_phys);
311 WARN_ON(__pkvm_hyp_donate_host(hyp_virt_to_pfn(addr), 1));
312 addr = hyp_alloc_pages(&vm->pool, 0);
313 }
314 }
315
__pkvm_prot_finalize(void)316 int __pkvm_prot_finalize(void)
317 {
318 struct kvm_s2_mmu *mmu = &host_mmu.arch.mmu;
319 struct kvm_nvhe_init_params *params = this_cpu_ptr(&kvm_init_params);
320
321 if (params->hcr_el2 & HCR_VM)
322 return -EPERM;
323
324 params->vttbr = kvm_get_vttbr(mmu);
325 params->vtcr = mmu->vtcr;
326 params->hcr_el2 |= HCR_VM;
327 if (cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
328 params->hcr_el2 |= HCR_FWB;
329
330 /*
331 * The CMO below not only cleans the updated params to the
332 * PoC, but also provides the DSB that ensures ongoing
333 * page-table walks that have started before we trapped to EL2
334 * have completed.
335 */
336 kvm_flush_dcache_to_poc(params, sizeof(*params));
337
338 write_sysreg_hcr(params->hcr_el2);
339 __load_stage2(&host_mmu.arch.mmu, &host_mmu.arch);
340
341 /*
342 * Make sure to have an ISB before the TLB maintenance below but only
343 * when __load_stage2() doesn't include one already.
344 */
345 asm(ALTERNATIVE("isb", "nop", ARM64_WORKAROUND_SPECULATIVE_AT));
346
347 /* Invalidate stale HCR bits that may be cached in TLBs */
348 __tlbi(vmalls12e1);
349 dsb(nsh);
350 isb();
351
352 return 0;
353 }
354
host_stage2_unmap_dev_all(void)355 static int host_stage2_unmap_dev_all(void)
356 {
357 struct kvm_pgtable *pgt = &host_mmu.pgt;
358 struct memblock_region *reg;
359 u64 addr = 0;
360 int i, ret;
361
362 /* Unmap all non-memory regions to recycle the pages */
363 for (i = 0; i < hyp_memblock_nr; i++, addr = reg->base + reg->size) {
364 reg = &hyp_memory[i];
365 ret = kvm_pgtable_stage2_unmap(pgt, addr, reg->base - addr);
366 if (ret)
367 return ret;
368 }
369 return kvm_pgtable_stage2_unmap(pgt, addr, BIT(pgt->ia_bits) - addr);
370 }
371
372 /*
373 * Ensure the PFN range is contained within PA-range.
374 *
375 * This check is also robust to overflows and is therefore a requirement before
376 * using a pfn/nr_pages pair from an untrusted source.
377 */
pfn_range_is_valid(u64 pfn,u64 nr_pages)378 static bool pfn_range_is_valid(u64 pfn, u64 nr_pages)
379 {
380 u64 limit = BIT(kvm_phys_shift(&host_mmu.arch.mmu) - PAGE_SHIFT);
381
382 return pfn < limit && ((limit - pfn) >= nr_pages);
383 }
384
385 struct kvm_mem_range {
386 u64 start;
387 u64 end;
388 };
389
find_mem_range(phys_addr_t addr,struct kvm_mem_range * range)390 static struct memblock_region *find_mem_range(phys_addr_t addr, struct kvm_mem_range *range)
391 {
392 int cur, left = 0, right = hyp_memblock_nr;
393 struct memblock_region *reg;
394 phys_addr_t end;
395
396 range->start = 0;
397 range->end = ULONG_MAX;
398
399 /* The list of memblock regions is sorted, binary search it */
400 while (left < right) {
401 cur = (left + right) >> 1;
402 reg = &hyp_memory[cur];
403 end = reg->base + reg->size;
404 if (addr < reg->base) {
405 right = cur;
406 range->end = reg->base;
407 } else if (addr >= end) {
408 left = cur + 1;
409 range->start = end;
410 } else {
411 range->start = reg->base;
412 range->end = end;
413 return reg;
414 }
415 }
416
417 return NULL;
418 }
419
addr_is_memory(phys_addr_t phys)420 bool addr_is_memory(phys_addr_t phys)
421 {
422 struct kvm_mem_range range;
423
424 return !!find_mem_range(phys, &range);
425 }
426
is_in_mem_range(u64 addr,struct kvm_mem_range * range)427 static bool is_in_mem_range(u64 addr, struct kvm_mem_range *range)
428 {
429 return range->start <= addr && addr < range->end;
430 }
431
check_range_allowed_memory(u64 start,u64 end)432 static int check_range_allowed_memory(u64 start, u64 end)
433 {
434 struct memblock_region *reg;
435 struct kvm_mem_range range;
436
437 /*
438 * Callers can't check the state of a range that overlaps memory and
439 * MMIO regions, so ensure [start, end[ is in the same kvm_mem_range.
440 */
441 reg = find_mem_range(start, &range);
442 if (!is_in_mem_range(end - 1, &range))
443 return -EINVAL;
444
445 if (!reg || reg->flags & MEMBLOCK_NOMAP)
446 return -EPERM;
447
448 return 0;
449 }
450
range_is_memory(u64 start,u64 end)451 static bool range_is_memory(u64 start, u64 end)
452 {
453 struct kvm_mem_range r;
454
455 if (!find_mem_range(start, &r))
456 return false;
457
458 return is_in_mem_range(end - 1, &r);
459 }
460
__host_stage2_idmap(u64 start,u64 end,enum kvm_pgtable_prot prot)461 static inline int __host_stage2_idmap(u64 start, u64 end,
462 enum kvm_pgtable_prot prot)
463 {
464 return kvm_pgtable_stage2_map(&host_mmu.pgt, start, end - start, start,
465 prot, &host_s2_pool, 0);
466 }
467
468 /*
469 * The pool has been provided with enough pages to cover all of memory with
470 * page granularity, but it is difficult to know how much of the MMIO range
471 * we will need to cover upfront, so we may need to 'recycle' the pages if we
472 * run out.
473 */
474 #define host_stage2_try(fn, ...) \
475 ({ \
476 int __ret; \
477 hyp_assert_lock_held(&host_mmu.lock); \
478 __ret = fn(__VA_ARGS__); \
479 if (__ret == -ENOMEM) { \
480 __ret = host_stage2_unmap_dev_all(); \
481 if (!__ret) \
482 __ret = fn(__VA_ARGS__); \
483 } \
484 __ret; \
485 })
486
range_included(struct kvm_mem_range * child,struct kvm_mem_range * parent)487 static inline bool range_included(struct kvm_mem_range *child,
488 struct kvm_mem_range *parent)
489 {
490 return parent->start <= child->start && child->end <= parent->end;
491 }
492
host_stage2_adjust_range(u64 addr,struct kvm_mem_range * range)493 static int host_stage2_adjust_range(u64 addr, struct kvm_mem_range *range)
494 {
495 struct kvm_mem_range cur;
496 kvm_pte_t pte;
497 u64 granule;
498 s8 level;
499 int ret;
500
501 hyp_assert_lock_held(&host_mmu.lock);
502 ret = kvm_pgtable_get_leaf(&host_mmu.pgt, addr, &pte, &level);
503 if (ret)
504 return ret;
505
506 if (kvm_pte_valid(pte))
507 return -EAGAIN;
508
509 if (pte) {
510 WARN_ON(addr_is_memory(addr) &&
511 get_host_state(hyp_phys_to_page(addr)) != PKVM_NOPAGE);
512 return -EPERM;
513 }
514
515 for (; level <= KVM_PGTABLE_LAST_LEVEL; level++) {
516 if (!kvm_level_supports_block_mapping(level))
517 continue;
518 granule = kvm_granule_size(level);
519 cur.start = ALIGN_DOWN(addr, granule);
520 cur.end = cur.start + granule;
521 if (!range_included(&cur, range) && level < KVM_PGTABLE_LAST_LEVEL)
522 continue;
523 *range = cur;
524 return 0;
525 }
526
527 WARN_ON(1);
528
529 return -EINVAL;
530 }
531
host_stage2_idmap_locked(phys_addr_t addr,u64 size,enum kvm_pgtable_prot prot)532 int host_stage2_idmap_locked(phys_addr_t addr, u64 size,
533 enum kvm_pgtable_prot prot)
534 {
535 return host_stage2_try(__host_stage2_idmap, addr, addr + size, prot);
536 }
537
__host_update_page_state(phys_addr_t addr,u64 size,enum pkvm_page_state state)538 static void __host_update_page_state(phys_addr_t addr, u64 size, enum pkvm_page_state state)
539 {
540 for_each_hyp_page(page, addr, size)
541 set_host_state(page, state);
542 }
543
host_stage2_set_owner_locked(phys_addr_t addr,u64 size,u8 owner_id)544 int host_stage2_set_owner_locked(phys_addr_t addr, u64 size, u8 owner_id)
545 {
546 int ret;
547
548 if (!range_is_memory(addr, addr + size))
549 return -EPERM;
550
551 ret = host_stage2_try(kvm_pgtable_stage2_set_owner, &host_mmu.pgt,
552 addr, size, &host_s2_pool, owner_id);
553 if (ret)
554 return ret;
555
556 /* Don't forget to update the vmemmap tracking for the host */
557 if (owner_id == PKVM_ID_HOST)
558 __host_update_page_state(addr, size, PKVM_PAGE_OWNED);
559 else
560 __host_update_page_state(addr, size, PKVM_NOPAGE);
561
562 return 0;
563 }
564
host_stage2_force_pte_cb(u64 addr,u64 end,enum kvm_pgtable_prot prot)565 static bool host_stage2_force_pte_cb(u64 addr, u64 end, enum kvm_pgtable_prot prot)
566 {
567 /*
568 * Block mappings must be used with care in the host stage-2 as a
569 * kvm_pgtable_stage2_map() operation targeting a page in the range of
570 * an existing block will delete the block under the assumption that
571 * mappings in the rest of the block range can always be rebuilt lazily.
572 * That assumption is correct for the host stage-2 with RWX mappings
573 * targeting memory or RW mappings targeting MMIO ranges (see
574 * host_stage2_idmap() below which implements some of the host memory
575 * abort logic). However, this is not safe for any other mappings where
576 * the host stage-2 page-table is in fact the only place where this
577 * state is stored. In all those cases, it is safer to use page-level
578 * mappings, hence avoiding to lose the state because of side-effects in
579 * kvm_pgtable_stage2_map().
580 */
581 if (range_is_memory(addr, end))
582 return prot != PKVM_HOST_MEM_PROT;
583 else
584 return prot != PKVM_HOST_MMIO_PROT;
585 }
586
host_stage2_idmap(u64 addr)587 static int host_stage2_idmap(u64 addr)
588 {
589 struct kvm_mem_range range;
590 bool is_memory = !!find_mem_range(addr, &range);
591 enum kvm_pgtable_prot prot;
592 int ret;
593
594 prot = is_memory ? PKVM_HOST_MEM_PROT : PKVM_HOST_MMIO_PROT;
595
596 host_lock_component();
597 ret = host_stage2_adjust_range(addr, &range);
598 if (ret)
599 goto unlock;
600
601 ret = host_stage2_idmap_locked(range.start, range.end - range.start, prot);
602 unlock:
603 host_unlock_component();
604
605 return ret;
606 }
607
handle_host_mem_abort(struct kvm_cpu_context * host_ctxt)608 void handle_host_mem_abort(struct kvm_cpu_context *host_ctxt)
609 {
610 struct kvm_vcpu_fault_info fault;
611 u64 esr, addr;
612 int ret = 0;
613
614 esr = read_sysreg_el2(SYS_ESR);
615 if (!__get_fault_info(esr, &fault)) {
616 /*
617 * We've presumably raced with a page-table change which caused
618 * AT to fail, try again.
619 */
620 return;
621 }
622
623
624 /*
625 * Yikes, we couldn't resolve the fault IPA. This should reinject an
626 * abort into the host when we figure out how to do that.
627 */
628 BUG_ON(!(fault.hpfar_el2 & HPFAR_EL2_NS));
629 addr = FIELD_GET(HPFAR_EL2_FIPA, fault.hpfar_el2) << 12;
630
631 ret = host_stage2_idmap(addr);
632 BUG_ON(ret && ret != -EAGAIN);
633 }
634
635 struct check_walk_data {
636 enum pkvm_page_state desired;
637 enum pkvm_page_state (*get_page_state)(kvm_pte_t pte, u64 addr);
638 };
639
__check_page_state_visitor(const struct kvm_pgtable_visit_ctx * ctx,enum kvm_pgtable_walk_flags visit)640 static int __check_page_state_visitor(const struct kvm_pgtable_visit_ctx *ctx,
641 enum kvm_pgtable_walk_flags visit)
642 {
643 struct check_walk_data *d = ctx->arg;
644
645 return d->get_page_state(ctx->old, ctx->addr) == d->desired ? 0 : -EPERM;
646 }
647
check_page_state_range(struct kvm_pgtable * pgt,u64 addr,u64 size,struct check_walk_data * data)648 static int check_page_state_range(struct kvm_pgtable *pgt, u64 addr, u64 size,
649 struct check_walk_data *data)
650 {
651 struct kvm_pgtable_walker walker = {
652 .cb = __check_page_state_visitor,
653 .arg = data,
654 .flags = KVM_PGTABLE_WALK_LEAF,
655 };
656
657 return kvm_pgtable_walk(pgt, addr, size, &walker);
658 }
659
__host_check_page_state_range(u64 addr,u64 size,enum pkvm_page_state state)660 static int __host_check_page_state_range(u64 addr, u64 size,
661 enum pkvm_page_state state)
662 {
663 int ret;
664
665 ret = check_range_allowed_memory(addr, addr + size);
666 if (ret)
667 return ret;
668
669 hyp_assert_lock_held(&host_mmu.lock);
670
671 for_each_hyp_page(page, addr, size) {
672 if (get_host_state(page) != state)
673 return -EPERM;
674 }
675
676 return 0;
677 }
678
__host_set_page_state_range(u64 addr,u64 size,enum pkvm_page_state state)679 static int __host_set_page_state_range(u64 addr, u64 size,
680 enum pkvm_page_state state)
681 {
682 if (get_host_state(hyp_phys_to_page(addr)) == PKVM_NOPAGE) {
683 int ret = host_stage2_idmap_locked(addr, size, PKVM_HOST_MEM_PROT);
684
685 if (ret)
686 return ret;
687 }
688
689 __host_update_page_state(addr, size, state);
690
691 return 0;
692 }
693
__hyp_set_page_state_range(phys_addr_t phys,u64 size,enum pkvm_page_state state)694 static void __hyp_set_page_state_range(phys_addr_t phys, u64 size, enum pkvm_page_state state)
695 {
696 for_each_hyp_page(page, phys, size)
697 set_hyp_state(page, state);
698 }
699
__hyp_check_page_state_range(phys_addr_t phys,u64 size,enum pkvm_page_state state)700 static int __hyp_check_page_state_range(phys_addr_t phys, u64 size, enum pkvm_page_state state)
701 {
702 for_each_hyp_page(page, phys, size) {
703 if (get_hyp_state(page) != state)
704 return -EPERM;
705 }
706
707 return 0;
708 }
709
guest_get_page_state(kvm_pte_t pte,u64 addr)710 static enum pkvm_page_state guest_get_page_state(kvm_pte_t pte, u64 addr)
711 {
712 if (!kvm_pte_valid(pte))
713 return PKVM_NOPAGE;
714
715 return pkvm_getstate(kvm_pgtable_stage2_pte_prot(pte));
716 }
717
__guest_check_page_state_range(struct pkvm_hyp_vm * vm,u64 addr,u64 size,enum pkvm_page_state state)718 static int __guest_check_page_state_range(struct pkvm_hyp_vm *vm, u64 addr,
719 u64 size, enum pkvm_page_state state)
720 {
721 struct check_walk_data d = {
722 .desired = state,
723 .get_page_state = guest_get_page_state,
724 };
725
726 hyp_assert_lock_held(&vm->lock);
727 return check_page_state_range(&vm->pgt, addr, size, &d);
728 }
729
__pkvm_host_share_hyp(u64 pfn)730 int __pkvm_host_share_hyp(u64 pfn)
731 {
732 u64 phys = hyp_pfn_to_phys(pfn);
733 u64 size = PAGE_SIZE;
734 int ret;
735
736 host_lock_component();
737 hyp_lock_component();
738
739 ret = __host_check_page_state_range(phys, size, PKVM_PAGE_OWNED);
740 if (ret)
741 goto unlock;
742 ret = __hyp_check_page_state_range(phys, size, PKVM_NOPAGE);
743 if (ret)
744 goto unlock;
745
746 __hyp_set_page_state_range(phys, size, PKVM_PAGE_SHARED_BORROWED);
747 WARN_ON(__host_set_page_state_range(phys, size, PKVM_PAGE_SHARED_OWNED));
748
749 unlock:
750 hyp_unlock_component();
751 host_unlock_component();
752
753 return ret;
754 }
755
__pkvm_host_unshare_hyp(u64 pfn)756 int __pkvm_host_unshare_hyp(u64 pfn)
757 {
758 u64 phys = hyp_pfn_to_phys(pfn);
759 u64 virt = (u64)__hyp_va(phys);
760 u64 size = PAGE_SIZE;
761 int ret;
762
763 host_lock_component();
764 hyp_lock_component();
765
766 ret = __host_check_page_state_range(phys, size, PKVM_PAGE_SHARED_OWNED);
767 if (ret)
768 goto unlock;
769 ret = __hyp_check_page_state_range(phys, size, PKVM_PAGE_SHARED_BORROWED);
770 if (ret)
771 goto unlock;
772 if (hyp_page_count((void *)virt)) {
773 ret = -EBUSY;
774 goto unlock;
775 }
776
777 __hyp_set_page_state_range(phys, size, PKVM_NOPAGE);
778 WARN_ON(__host_set_page_state_range(phys, size, PKVM_PAGE_OWNED));
779
780 unlock:
781 hyp_unlock_component();
782 host_unlock_component();
783
784 return ret;
785 }
786
__pkvm_host_donate_hyp(u64 pfn,u64 nr_pages)787 int __pkvm_host_donate_hyp(u64 pfn, u64 nr_pages)
788 {
789 u64 phys = hyp_pfn_to_phys(pfn);
790 u64 size = PAGE_SIZE * nr_pages;
791 void *virt = __hyp_va(phys);
792 int ret;
793
794 if (!pfn_range_is_valid(pfn, nr_pages))
795 return -EINVAL;
796
797 host_lock_component();
798 hyp_lock_component();
799
800 ret = __host_check_page_state_range(phys, size, PKVM_PAGE_OWNED);
801 if (ret)
802 goto unlock;
803 ret = __hyp_check_page_state_range(phys, size, PKVM_NOPAGE);
804 if (ret)
805 goto unlock;
806
807 __hyp_set_page_state_range(phys, size, PKVM_PAGE_OWNED);
808 WARN_ON(pkvm_create_mappings_locked(virt, virt + size, PAGE_HYP));
809 WARN_ON(host_stage2_set_owner_locked(phys, size, PKVM_ID_HYP));
810
811 unlock:
812 hyp_unlock_component();
813 host_unlock_component();
814
815 return ret;
816 }
817
__pkvm_hyp_donate_host(u64 pfn,u64 nr_pages)818 int __pkvm_hyp_donate_host(u64 pfn, u64 nr_pages)
819 {
820 u64 phys = hyp_pfn_to_phys(pfn);
821 u64 size = PAGE_SIZE * nr_pages;
822 u64 virt = (u64)__hyp_va(phys);
823 int ret;
824
825 if (!pfn_range_is_valid(pfn, nr_pages))
826 return -EINVAL;
827
828 host_lock_component();
829 hyp_lock_component();
830
831 ret = __hyp_check_page_state_range(phys, size, PKVM_PAGE_OWNED);
832 if (ret)
833 goto unlock;
834 ret = __host_check_page_state_range(phys, size, PKVM_NOPAGE);
835 if (ret)
836 goto unlock;
837
838 __hyp_set_page_state_range(phys, size, PKVM_NOPAGE);
839 WARN_ON(kvm_pgtable_hyp_unmap(&pkvm_pgtable, virt, size) != size);
840 WARN_ON(host_stage2_set_owner_locked(phys, size, PKVM_ID_HOST));
841
842 unlock:
843 hyp_unlock_component();
844 host_unlock_component();
845
846 return ret;
847 }
848
hyp_pin_shared_mem(void * from,void * to)849 int hyp_pin_shared_mem(void *from, void *to)
850 {
851 u64 cur, start = ALIGN_DOWN((u64)from, PAGE_SIZE);
852 u64 end = PAGE_ALIGN((u64)to);
853 u64 phys = __hyp_pa(start);
854 u64 size = end - start;
855 struct hyp_page *p;
856 int ret;
857
858 host_lock_component();
859 hyp_lock_component();
860
861 ret = __host_check_page_state_range(phys, size, PKVM_PAGE_SHARED_OWNED);
862 if (ret)
863 goto unlock;
864
865 ret = __hyp_check_page_state_range(phys, size, PKVM_PAGE_SHARED_BORROWED);
866 if (ret)
867 goto unlock;
868
869 for (cur = start; cur < end; cur += PAGE_SIZE) {
870 p = hyp_virt_to_page(cur);
871 hyp_page_ref_inc(p);
872 if (p->refcount == 1)
873 WARN_ON(pkvm_create_mappings_locked((void *)cur,
874 (void *)cur + PAGE_SIZE,
875 PAGE_HYP));
876 }
877
878 unlock:
879 hyp_unlock_component();
880 host_unlock_component();
881
882 return ret;
883 }
884
hyp_unpin_shared_mem(void * from,void * to)885 void hyp_unpin_shared_mem(void *from, void *to)
886 {
887 u64 cur, start = ALIGN_DOWN((u64)from, PAGE_SIZE);
888 u64 end = PAGE_ALIGN((u64)to);
889 struct hyp_page *p;
890
891 host_lock_component();
892 hyp_lock_component();
893
894 for (cur = start; cur < end; cur += PAGE_SIZE) {
895 p = hyp_virt_to_page(cur);
896 if (p->refcount == 1)
897 WARN_ON(kvm_pgtable_hyp_unmap(&pkvm_pgtable, cur, PAGE_SIZE) != PAGE_SIZE);
898 hyp_page_ref_dec(p);
899 }
900
901 hyp_unlock_component();
902 host_unlock_component();
903 }
904
__pkvm_host_share_ffa(u64 pfn,u64 nr_pages)905 int __pkvm_host_share_ffa(u64 pfn, u64 nr_pages)
906 {
907 u64 phys = hyp_pfn_to_phys(pfn);
908 u64 size = PAGE_SIZE * nr_pages;
909 int ret;
910
911 if (!pfn_range_is_valid(pfn, nr_pages))
912 return -EINVAL;
913
914 host_lock_component();
915 ret = __host_check_page_state_range(phys, size, PKVM_PAGE_OWNED);
916 if (!ret)
917 ret = __host_set_page_state_range(phys, size, PKVM_PAGE_SHARED_OWNED);
918 host_unlock_component();
919
920 return ret;
921 }
922
__pkvm_host_unshare_ffa(u64 pfn,u64 nr_pages)923 int __pkvm_host_unshare_ffa(u64 pfn, u64 nr_pages)
924 {
925 u64 phys = hyp_pfn_to_phys(pfn);
926 u64 size = PAGE_SIZE * nr_pages;
927 int ret;
928
929 if (!pfn_range_is_valid(pfn, nr_pages))
930 return -EINVAL;
931
932 host_lock_component();
933 ret = __host_check_page_state_range(phys, size, PKVM_PAGE_SHARED_OWNED);
934 if (!ret)
935 ret = __host_set_page_state_range(phys, size, PKVM_PAGE_OWNED);
936 host_unlock_component();
937
938 return ret;
939 }
940
__guest_check_transition_size(u64 phys,u64 ipa,u64 nr_pages,u64 * size)941 static int __guest_check_transition_size(u64 phys, u64 ipa, u64 nr_pages, u64 *size)
942 {
943 size_t block_size;
944
945 if (nr_pages == 1) {
946 *size = PAGE_SIZE;
947 return 0;
948 }
949
950 /* We solely support second to last level huge mapping */
951 block_size = kvm_granule_size(KVM_PGTABLE_LAST_LEVEL - 1);
952
953 if (nr_pages != block_size >> PAGE_SHIFT)
954 return -EINVAL;
955
956 if (!IS_ALIGNED(phys | ipa, block_size))
957 return -EINVAL;
958
959 *size = block_size;
960 return 0;
961 }
962
__pkvm_host_share_guest(u64 pfn,u64 gfn,u64 nr_pages,struct pkvm_hyp_vcpu * vcpu,enum kvm_pgtable_prot prot)963 int __pkvm_host_share_guest(u64 pfn, u64 gfn, u64 nr_pages, struct pkvm_hyp_vcpu *vcpu,
964 enum kvm_pgtable_prot prot)
965 {
966 struct pkvm_hyp_vm *vm = pkvm_hyp_vcpu_to_hyp_vm(vcpu);
967 u64 phys = hyp_pfn_to_phys(pfn);
968 u64 ipa = hyp_pfn_to_phys(gfn);
969 u64 size;
970 int ret;
971
972 if (prot & ~KVM_PGTABLE_PROT_RWX)
973 return -EINVAL;
974
975 if (!pfn_range_is_valid(pfn, nr_pages))
976 return -EINVAL;
977
978 ret = __guest_check_transition_size(phys, ipa, nr_pages, &size);
979 if (ret)
980 return ret;
981
982 ret = check_range_allowed_memory(phys, phys + size);
983 if (ret)
984 return ret;
985
986 host_lock_component();
987 guest_lock_component(vm);
988
989 ret = __guest_check_page_state_range(vm, ipa, size, PKVM_NOPAGE);
990 if (ret)
991 goto unlock;
992
993 for_each_hyp_page(page, phys, size) {
994 switch (get_host_state(page)) {
995 case PKVM_PAGE_OWNED:
996 continue;
997 case PKVM_PAGE_SHARED_OWNED:
998 if (page->host_share_guest_count == U32_MAX) {
999 ret = -EBUSY;
1000 goto unlock;
1001 }
1002
1003 /* Only host to np-guest multi-sharing is tolerated */
1004 if (page->host_share_guest_count)
1005 continue;
1006
1007 fallthrough;
1008 default:
1009 ret = -EPERM;
1010 goto unlock;
1011 }
1012 }
1013
1014 for_each_hyp_page(page, phys, size) {
1015 set_host_state(page, PKVM_PAGE_SHARED_OWNED);
1016 page->host_share_guest_count++;
1017 }
1018
1019 WARN_ON(kvm_pgtable_stage2_map(&vm->pgt, ipa, size, phys,
1020 pkvm_mkstate(prot, PKVM_PAGE_SHARED_BORROWED),
1021 &vcpu->vcpu.arch.pkvm_memcache, 0));
1022
1023 unlock:
1024 guest_unlock_component(vm);
1025 host_unlock_component();
1026
1027 return ret;
1028 }
1029
__check_host_shared_guest(struct pkvm_hyp_vm * vm,u64 * __phys,u64 ipa,u64 size)1030 static int __check_host_shared_guest(struct pkvm_hyp_vm *vm, u64 *__phys, u64 ipa, u64 size)
1031 {
1032 enum pkvm_page_state state;
1033 kvm_pte_t pte;
1034 u64 phys;
1035 s8 level;
1036 int ret;
1037
1038 ret = kvm_pgtable_get_leaf(&vm->pgt, ipa, &pte, &level);
1039 if (ret)
1040 return ret;
1041 if (!kvm_pte_valid(pte))
1042 return -ENOENT;
1043 if (size && kvm_granule_size(level) != size)
1044 return -E2BIG;
1045
1046 if (!size)
1047 size = kvm_granule_size(level);
1048
1049 state = guest_get_page_state(pte, ipa);
1050 if (state != PKVM_PAGE_SHARED_BORROWED)
1051 return -EPERM;
1052
1053 phys = kvm_pte_to_phys(pte);
1054 ret = check_range_allowed_memory(phys, phys + size);
1055 if (WARN_ON(ret))
1056 return ret;
1057
1058 for_each_hyp_page(page, phys, size) {
1059 if (get_host_state(page) != PKVM_PAGE_SHARED_OWNED)
1060 return -EPERM;
1061 if (WARN_ON(!page->host_share_guest_count))
1062 return -EINVAL;
1063 }
1064
1065 *__phys = phys;
1066
1067 return 0;
1068 }
1069
__pkvm_host_unshare_guest(u64 gfn,u64 nr_pages,struct pkvm_hyp_vm * vm)1070 int __pkvm_host_unshare_guest(u64 gfn, u64 nr_pages, struct pkvm_hyp_vm *vm)
1071 {
1072 u64 ipa = hyp_pfn_to_phys(gfn);
1073 u64 size, phys;
1074 int ret;
1075
1076 ret = __guest_check_transition_size(0, ipa, nr_pages, &size);
1077 if (ret)
1078 return ret;
1079
1080 host_lock_component();
1081 guest_lock_component(vm);
1082
1083 ret = __check_host_shared_guest(vm, &phys, ipa, size);
1084 if (ret)
1085 goto unlock;
1086
1087 ret = kvm_pgtable_stage2_unmap(&vm->pgt, ipa, size);
1088 if (ret)
1089 goto unlock;
1090
1091 for_each_hyp_page(page, phys, size) {
1092 /* __check_host_shared_guest() protects against underflow */
1093 page->host_share_guest_count--;
1094 if (!page->host_share_guest_count)
1095 set_host_state(page, PKVM_PAGE_OWNED);
1096 }
1097
1098 unlock:
1099 guest_unlock_component(vm);
1100 host_unlock_component();
1101
1102 return ret;
1103 }
1104
assert_host_shared_guest(struct pkvm_hyp_vm * vm,u64 ipa,u64 size)1105 static void assert_host_shared_guest(struct pkvm_hyp_vm *vm, u64 ipa, u64 size)
1106 {
1107 u64 phys;
1108 int ret;
1109
1110 if (!IS_ENABLED(CONFIG_NVHE_EL2_DEBUG))
1111 return;
1112
1113 host_lock_component();
1114 guest_lock_component(vm);
1115
1116 ret = __check_host_shared_guest(vm, &phys, ipa, size);
1117
1118 guest_unlock_component(vm);
1119 host_unlock_component();
1120
1121 WARN_ON(ret && ret != -ENOENT);
1122 }
1123
__pkvm_host_relax_perms_guest(u64 gfn,struct pkvm_hyp_vcpu * vcpu,enum kvm_pgtable_prot prot)1124 int __pkvm_host_relax_perms_guest(u64 gfn, struct pkvm_hyp_vcpu *vcpu, enum kvm_pgtable_prot prot)
1125 {
1126 struct pkvm_hyp_vm *vm = pkvm_hyp_vcpu_to_hyp_vm(vcpu);
1127 u64 ipa = hyp_pfn_to_phys(gfn);
1128 int ret;
1129
1130 if (pkvm_hyp_vm_is_protected(vm))
1131 return -EPERM;
1132
1133 if (prot & ~KVM_PGTABLE_PROT_RWX)
1134 return -EINVAL;
1135
1136 assert_host_shared_guest(vm, ipa, 0);
1137 guest_lock_component(vm);
1138 ret = kvm_pgtable_stage2_relax_perms(&vm->pgt, ipa, prot, 0);
1139 guest_unlock_component(vm);
1140
1141 return ret;
1142 }
1143
__pkvm_host_wrprotect_guest(u64 gfn,u64 nr_pages,struct pkvm_hyp_vm * vm)1144 int __pkvm_host_wrprotect_guest(u64 gfn, u64 nr_pages, struct pkvm_hyp_vm *vm)
1145 {
1146 u64 size, ipa = hyp_pfn_to_phys(gfn);
1147 int ret;
1148
1149 if (pkvm_hyp_vm_is_protected(vm))
1150 return -EPERM;
1151
1152 ret = __guest_check_transition_size(0, ipa, nr_pages, &size);
1153 if (ret)
1154 return ret;
1155
1156 assert_host_shared_guest(vm, ipa, size);
1157 guest_lock_component(vm);
1158 ret = kvm_pgtable_stage2_wrprotect(&vm->pgt, ipa, size);
1159 guest_unlock_component(vm);
1160
1161 return ret;
1162 }
1163
__pkvm_host_test_clear_young_guest(u64 gfn,u64 nr_pages,bool mkold,struct pkvm_hyp_vm * vm)1164 int __pkvm_host_test_clear_young_guest(u64 gfn, u64 nr_pages, bool mkold, struct pkvm_hyp_vm *vm)
1165 {
1166 u64 size, ipa = hyp_pfn_to_phys(gfn);
1167 int ret;
1168
1169 if (pkvm_hyp_vm_is_protected(vm))
1170 return -EPERM;
1171
1172 ret = __guest_check_transition_size(0, ipa, nr_pages, &size);
1173 if (ret)
1174 return ret;
1175
1176 assert_host_shared_guest(vm, ipa, size);
1177 guest_lock_component(vm);
1178 ret = kvm_pgtable_stage2_test_clear_young(&vm->pgt, ipa, size, mkold);
1179 guest_unlock_component(vm);
1180
1181 return ret;
1182 }
1183
__pkvm_host_mkyoung_guest(u64 gfn,struct pkvm_hyp_vcpu * vcpu)1184 int __pkvm_host_mkyoung_guest(u64 gfn, struct pkvm_hyp_vcpu *vcpu)
1185 {
1186 struct pkvm_hyp_vm *vm = pkvm_hyp_vcpu_to_hyp_vm(vcpu);
1187 u64 ipa = hyp_pfn_to_phys(gfn);
1188
1189 if (pkvm_hyp_vm_is_protected(vm))
1190 return -EPERM;
1191
1192 assert_host_shared_guest(vm, ipa, 0);
1193 guest_lock_component(vm);
1194 kvm_pgtable_stage2_mkyoung(&vm->pgt, ipa, 0);
1195 guest_unlock_component(vm);
1196
1197 return 0;
1198 }
1199
1200 #ifdef CONFIG_NVHE_EL2_DEBUG
1201 struct pkvm_expected_state {
1202 enum pkvm_page_state host;
1203 enum pkvm_page_state hyp;
1204 enum pkvm_page_state guest[2]; /* [ gfn, gfn + 1 ] */
1205 };
1206
1207 static struct pkvm_expected_state selftest_state;
1208 static struct hyp_page *selftest_page;
1209
1210 static struct pkvm_hyp_vm selftest_vm = {
1211 .kvm = {
1212 .arch = {
1213 .mmu = {
1214 .arch = &selftest_vm.kvm.arch,
1215 .pgt = &selftest_vm.pgt,
1216 },
1217 },
1218 },
1219 };
1220
1221 static struct pkvm_hyp_vcpu selftest_vcpu = {
1222 .vcpu = {
1223 .arch = {
1224 .hw_mmu = &selftest_vm.kvm.arch.mmu,
1225 },
1226 .kvm = &selftest_vm.kvm,
1227 },
1228 };
1229
init_selftest_vm(void * virt)1230 static void init_selftest_vm(void *virt)
1231 {
1232 struct hyp_page *p = hyp_virt_to_page(virt);
1233 int i;
1234
1235 selftest_vm.kvm.arch.mmu.vtcr = host_mmu.arch.mmu.vtcr;
1236 WARN_ON(kvm_guest_prepare_stage2(&selftest_vm, virt));
1237
1238 for (i = 0; i < pkvm_selftest_pages(); i++) {
1239 if (p[i].refcount)
1240 continue;
1241 p[i].refcount = 1;
1242 hyp_put_page(&selftest_vm.pool, hyp_page_to_virt(&p[i]));
1243 }
1244 }
1245
selftest_ipa(void)1246 static u64 selftest_ipa(void)
1247 {
1248 return BIT(selftest_vm.pgt.ia_bits - 1);
1249 }
1250
assert_page_state(void)1251 static void assert_page_state(void)
1252 {
1253 void *virt = hyp_page_to_virt(selftest_page);
1254 u64 size = PAGE_SIZE << selftest_page->order;
1255 struct pkvm_hyp_vcpu *vcpu = &selftest_vcpu;
1256 u64 phys = hyp_virt_to_phys(virt);
1257 u64 ipa[2] = { selftest_ipa(), selftest_ipa() + PAGE_SIZE };
1258 struct pkvm_hyp_vm *vm;
1259
1260 vm = pkvm_hyp_vcpu_to_hyp_vm(vcpu);
1261
1262 host_lock_component();
1263 WARN_ON(__host_check_page_state_range(phys, size, selftest_state.host));
1264 host_unlock_component();
1265
1266 hyp_lock_component();
1267 WARN_ON(__hyp_check_page_state_range(phys, size, selftest_state.hyp));
1268 hyp_unlock_component();
1269
1270 guest_lock_component(&selftest_vm);
1271 WARN_ON(__guest_check_page_state_range(vm, ipa[0], size, selftest_state.guest[0]));
1272 WARN_ON(__guest_check_page_state_range(vm, ipa[1], size, selftest_state.guest[1]));
1273 guest_unlock_component(&selftest_vm);
1274 }
1275
1276 #define assert_transition_res(res, fn, ...) \
1277 do { \
1278 WARN_ON(fn(__VA_ARGS__) != res); \
1279 assert_page_state(); \
1280 } while (0)
1281
pkvm_ownership_selftest(void * base)1282 void pkvm_ownership_selftest(void *base)
1283 {
1284 enum kvm_pgtable_prot prot = KVM_PGTABLE_PROT_RWX;
1285 void *virt = hyp_alloc_pages(&host_s2_pool, 0);
1286 struct pkvm_hyp_vcpu *vcpu = &selftest_vcpu;
1287 struct pkvm_hyp_vm *vm = &selftest_vm;
1288 u64 phys, size, pfn, gfn;
1289
1290 WARN_ON(!virt);
1291 selftest_page = hyp_virt_to_page(virt);
1292 selftest_page->refcount = 0;
1293 init_selftest_vm(base);
1294
1295 size = PAGE_SIZE << selftest_page->order;
1296 phys = hyp_virt_to_phys(virt);
1297 pfn = hyp_phys_to_pfn(phys);
1298 gfn = hyp_phys_to_pfn(selftest_ipa());
1299
1300 selftest_state.host = PKVM_NOPAGE;
1301 selftest_state.hyp = PKVM_PAGE_OWNED;
1302 selftest_state.guest[0] = selftest_state.guest[1] = PKVM_NOPAGE;
1303 assert_page_state();
1304 assert_transition_res(-EPERM, __pkvm_host_donate_hyp, pfn, 1);
1305 assert_transition_res(-EPERM, __pkvm_host_share_hyp, pfn);
1306 assert_transition_res(-EPERM, __pkvm_host_unshare_hyp, pfn);
1307 assert_transition_res(-EPERM, __pkvm_host_share_ffa, pfn, 1);
1308 assert_transition_res(-EPERM, __pkvm_host_unshare_ffa, pfn, 1);
1309 assert_transition_res(-EPERM, hyp_pin_shared_mem, virt, virt + size);
1310 assert_transition_res(-EPERM, __pkvm_host_share_guest, pfn, gfn, 1, vcpu, prot);
1311 assert_transition_res(-ENOENT, __pkvm_host_unshare_guest, gfn, 1, vm);
1312
1313 selftest_state.host = PKVM_PAGE_OWNED;
1314 selftest_state.hyp = PKVM_NOPAGE;
1315 assert_transition_res(0, __pkvm_hyp_donate_host, pfn, 1);
1316 assert_transition_res(-EPERM, __pkvm_hyp_donate_host, pfn, 1);
1317 assert_transition_res(-EPERM, __pkvm_host_unshare_hyp, pfn);
1318 assert_transition_res(-EPERM, __pkvm_host_unshare_ffa, pfn, 1);
1319 assert_transition_res(-ENOENT, __pkvm_host_unshare_guest, gfn, 1, vm);
1320 assert_transition_res(-EPERM, hyp_pin_shared_mem, virt, virt + size);
1321
1322 selftest_state.host = PKVM_PAGE_SHARED_OWNED;
1323 selftest_state.hyp = PKVM_PAGE_SHARED_BORROWED;
1324 assert_transition_res(0, __pkvm_host_share_hyp, pfn);
1325 assert_transition_res(-EPERM, __pkvm_host_share_hyp, pfn);
1326 assert_transition_res(-EPERM, __pkvm_host_donate_hyp, pfn, 1);
1327 assert_transition_res(-EPERM, __pkvm_host_share_ffa, pfn, 1);
1328 assert_transition_res(-EPERM, __pkvm_hyp_donate_host, pfn, 1);
1329 assert_transition_res(-EPERM, __pkvm_host_share_guest, pfn, gfn, 1, vcpu, prot);
1330 assert_transition_res(-ENOENT, __pkvm_host_unshare_guest, gfn, 1, vm);
1331
1332 assert_transition_res(0, hyp_pin_shared_mem, virt, virt + size);
1333 assert_transition_res(0, hyp_pin_shared_mem, virt, virt + size);
1334 hyp_unpin_shared_mem(virt, virt + size);
1335 WARN_ON(hyp_page_count(virt) != 1);
1336 assert_transition_res(-EBUSY, __pkvm_host_unshare_hyp, pfn);
1337 assert_transition_res(-EPERM, __pkvm_host_share_hyp, pfn);
1338 assert_transition_res(-EPERM, __pkvm_host_donate_hyp, pfn, 1);
1339 assert_transition_res(-EPERM, __pkvm_host_share_ffa, pfn, 1);
1340 assert_transition_res(-EPERM, __pkvm_hyp_donate_host, pfn, 1);
1341 assert_transition_res(-EPERM, __pkvm_host_share_guest, pfn, gfn, 1, vcpu, prot);
1342 assert_transition_res(-ENOENT, __pkvm_host_unshare_guest, gfn, 1, vm);
1343
1344 hyp_unpin_shared_mem(virt, virt + size);
1345 assert_page_state();
1346 WARN_ON(hyp_page_count(virt));
1347
1348 selftest_state.host = PKVM_PAGE_OWNED;
1349 selftest_state.hyp = PKVM_NOPAGE;
1350 assert_transition_res(0, __pkvm_host_unshare_hyp, pfn);
1351
1352 selftest_state.host = PKVM_PAGE_SHARED_OWNED;
1353 selftest_state.hyp = PKVM_NOPAGE;
1354 assert_transition_res(0, __pkvm_host_share_ffa, pfn, 1);
1355 assert_transition_res(-EPERM, __pkvm_host_share_ffa, pfn, 1);
1356 assert_transition_res(-EPERM, __pkvm_host_donate_hyp, pfn, 1);
1357 assert_transition_res(-EPERM, __pkvm_host_share_hyp, pfn);
1358 assert_transition_res(-EPERM, __pkvm_host_unshare_hyp, pfn);
1359 assert_transition_res(-EPERM, __pkvm_hyp_donate_host, pfn, 1);
1360 assert_transition_res(-EPERM, __pkvm_host_share_guest, pfn, gfn, 1, vcpu, prot);
1361 assert_transition_res(-ENOENT, __pkvm_host_unshare_guest, gfn, 1, vm);
1362 assert_transition_res(-EPERM, hyp_pin_shared_mem, virt, virt + size);
1363
1364 selftest_state.host = PKVM_PAGE_OWNED;
1365 selftest_state.hyp = PKVM_NOPAGE;
1366 assert_transition_res(0, __pkvm_host_unshare_ffa, pfn, 1);
1367 assert_transition_res(-EPERM, __pkvm_host_unshare_ffa, pfn, 1);
1368
1369 selftest_state.host = PKVM_PAGE_SHARED_OWNED;
1370 selftest_state.guest[0] = PKVM_PAGE_SHARED_BORROWED;
1371 assert_transition_res(0, __pkvm_host_share_guest, pfn, gfn, 1, vcpu, prot);
1372 assert_transition_res(-EPERM, __pkvm_host_share_guest, pfn, gfn, 1, vcpu, prot);
1373 assert_transition_res(-EPERM, __pkvm_host_share_ffa, pfn, 1);
1374 assert_transition_res(-EPERM, __pkvm_host_donate_hyp, pfn, 1);
1375 assert_transition_res(-EPERM, __pkvm_host_share_hyp, pfn);
1376 assert_transition_res(-EPERM, __pkvm_host_unshare_hyp, pfn);
1377 assert_transition_res(-EPERM, __pkvm_hyp_donate_host, pfn, 1);
1378 assert_transition_res(-EPERM, hyp_pin_shared_mem, virt, virt + size);
1379
1380 selftest_state.guest[1] = PKVM_PAGE_SHARED_BORROWED;
1381 assert_transition_res(0, __pkvm_host_share_guest, pfn, gfn + 1, 1, vcpu, prot);
1382 WARN_ON(hyp_virt_to_page(virt)->host_share_guest_count != 2);
1383
1384 selftest_state.guest[0] = PKVM_NOPAGE;
1385 assert_transition_res(0, __pkvm_host_unshare_guest, gfn, 1, vm);
1386
1387 selftest_state.guest[1] = PKVM_NOPAGE;
1388 selftest_state.host = PKVM_PAGE_OWNED;
1389 assert_transition_res(0, __pkvm_host_unshare_guest, gfn + 1, 1, vm);
1390
1391 selftest_state.host = PKVM_NOPAGE;
1392 selftest_state.hyp = PKVM_PAGE_OWNED;
1393 assert_transition_res(0, __pkvm_host_donate_hyp, pfn, 1);
1394
1395 selftest_page->refcount = 1;
1396 hyp_put_page(&host_s2_pool, virt);
1397 }
1398 #endif
1399