]>
Commit | Line | Data |
---|---|---|
d94d71cb | 1 | // SPDX-License-Identifier: GPL-2.0-only |
de56a948 | 2 | /* |
de56a948 PM |
3 | * |
4 | * Copyright 2010 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com> | |
5 | */ | |
6 | ||
7 | #include <linux/types.h> | |
8 | #include <linux/string.h> | |
9 | #include <linux/kvm.h> | |
10 | #include <linux/kvm_host.h> | |
11 | #include <linux/highmem.h> | |
12 | #include <linux/gfp.h> | |
13 | #include <linux/slab.h> | |
14 | #include <linux/hugetlb.h> | |
8936dda4 | 15 | #include <linux/vmalloc.h> |
2c9097e4 | 16 | #include <linux/srcu.h> |
a2932923 PM |
17 | #include <linux/anon_inodes.h> |
18 | #include <linux/file.h> | |
e23a808b | 19 | #include <linux/debugfs.h> |
de56a948 | 20 | |
de56a948 PM |
21 | #include <asm/kvm_ppc.h> |
22 | #include <asm/kvm_book3s.h> | |
f64e8084 | 23 | #include <asm/book3s/64/mmu-hash.h> |
de56a948 PM |
24 | #include <asm/hvcall.h> |
25 | #include <asm/synch.h> | |
26 | #include <asm/ppc-opcode.h> | |
27 | #include <asm/cputable.h> | |
94171b19 | 28 | #include <asm/pte-walk.h> |
de56a948 | 29 | |
3c78f78a SW |
30 | #include "trace_hv.h" |
31 | ||
5e985969 DG |
32 | //#define DEBUG_RESIZE_HPT 1 |
33 | ||
34 | #ifdef DEBUG_RESIZE_HPT | |
35 | #define resize_hpt_debug(resize, ...) \ | |
36 | do { \ | |
37 | printk(KERN_DEBUG "RESIZE HPT %p: ", resize); \ | |
38 | printk(__VA_ARGS__); \ | |
39 | } while (0) | |
40 | #else | |
41 | #define resize_hpt_debug(resize, ...) \ | |
42 | do { } while (0) | |
43 | #endif | |
44 | ||
7ed661bf PM |
45 | static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags, |
46 | long pte_index, unsigned long pteh, | |
47 | unsigned long ptel, unsigned long *pte_idx_ret); | |
5e985969 DG |
48 | |
49 | struct kvm_resize_hpt { | |
50 | /* These fields read-only after init */ | |
51 | struct kvm *kvm; | |
52 | struct work_struct work; | |
53 | u32 order; | |
54 | ||
0d4ee88d | 55 | /* These fields protected by kvm->arch.mmu_setup_lock */ |
3073774e SP |
56 | |
57 | /* Possible values and their usage: | |
58 | * <0 an error occurred during allocation, | |
59 | * -EBUSY allocation is in the progress, | |
60 | * 0 allocation made successfuly. | |
61 | */ | |
5e985969 | 62 | int error; |
b5baa687 | 63 | |
3073774e | 64 | /* Private to the work thread, until error != -EBUSY, |
0d4ee88d | 65 | * then protected by kvm->arch.mmu_setup_lock. |
3073774e | 66 | */ |
b5baa687 | 67 | struct kvm_hpt_info hpt; |
5e985969 DG |
68 | }; |
69 | ||
aae0777f | 70 | int kvmppc_allocate_hpt(struct kvm_hpt_info *info, u32 order) |
de56a948 | 71 | { |
792fc497 | 72 | unsigned long hpt = 0; |
aae0777f | 73 | int cma = 0; |
fa61a4e3 | 74 | struct page *page = NULL; |
aae0777f DG |
75 | struct revmap_entry *rev; |
76 | unsigned long npte; | |
de56a948 | 77 | |
aae0777f DG |
78 | if ((order < PPC_MIN_HPT_ORDER) || (order > PPC_MAX_HPT_ORDER)) |
79 | return -EINVAL; | |
32fad281 | 80 | |
db9a290d | 81 | page = kvm_alloc_hpt_cma(1ul << (order - PAGE_SHIFT)); |
792fc497 AK |
82 | if (page) { |
83 | hpt = (unsigned long)pfn_to_kaddr(page_to_pfn(page)); | |
02a68d05 | 84 | memset((void *)hpt, 0, (1ul << order)); |
aae0777f | 85 | cma = 1; |
de56a948 | 86 | } |
32fad281 | 87 | |
aae0777f | 88 | if (!hpt) |
dcda9b04 | 89 | hpt = __get_free_pages(GFP_KERNEL|__GFP_ZERO|__GFP_RETRY_MAYFAIL |
aae0777f | 90 | |__GFP_NOWARN, order - PAGE_SHIFT); |
32fad281 PM |
91 | |
92 | if (!hpt) | |
93 | return -ENOMEM; | |
94 | ||
aae0777f DG |
95 | /* HPTEs are 2**4 bytes long */ |
96 | npte = 1ul << (order - 4); | |
a56ee9f8 | 97 | |
8936dda4 | 98 | /* Allocate reverse map array */ |
42bc47b3 | 99 | rev = vmalloc(array_size(npte, sizeof(struct revmap_entry))); |
8936dda4 | 100 | if (!rev) { |
aae0777f DG |
101 | if (cma) |
102 | kvm_free_hpt_cma(page, 1 << (order - PAGE_SHIFT)); | |
103 | else | |
104 | free_pages(hpt, order - PAGE_SHIFT); | |
105 | return -ENOMEM; | |
8936dda4 | 106 | } |
8936dda4 | 107 | |
aae0777f DG |
108 | info->order = order; |
109 | info->virt = hpt; | |
110 | info->cma = cma; | |
111 | info->rev = rev; | |
de56a948 | 112 | |
de56a948 | 113 | return 0; |
aae0777f | 114 | } |
8936dda4 | 115 | |
aae0777f DG |
116 | void kvmppc_set_hpt(struct kvm *kvm, struct kvm_hpt_info *info) |
117 | { | |
118 | atomic64_set(&kvm->arch.mmio_update, 0); | |
119 | kvm->arch.hpt = *info; | |
120 | kvm->arch.sdr1 = __pa(info->virt) | (info->order - 18); | |
121 | ||
3a4f1760 TH |
122 | pr_debug("KVM guest htab at %lx (order %ld), LPID %x\n", |
123 | info->virt, (long)info->order, kvm->arch.lpid); | |
de56a948 PM |
124 | } |
125 | ||
f98a8bf9 | 126 | long kvmppc_alloc_reset_hpt(struct kvm *kvm, int order) |
32fad281 PM |
127 | { |
128 | long err = -EBUSY; | |
f98a8bf9 | 129 | struct kvm_hpt_info info; |
32fad281 | 130 | |
0d4ee88d | 131 | mutex_lock(&kvm->arch.mmu_setup_lock); |
1b151ce4 PM |
132 | if (kvm->arch.mmu_ready) { |
133 | kvm->arch.mmu_ready = 0; | |
134 | /* order mmu_ready vs. vcpus_running */ | |
32fad281 PM |
135 | smp_mb(); |
136 | if (atomic_read(&kvm->arch.vcpus_running)) { | |
1b151ce4 | 137 | kvm->arch.mmu_ready = 1; |
32fad281 PM |
138 | goto out; |
139 | } | |
140 | } | |
18c3640c PM |
141 | if (kvm_is_radix(kvm)) { |
142 | err = kvmppc_switch_mmu_to_hpt(kvm); | |
143 | if (err) | |
144 | goto out; | |
145 | } | |
146 | ||
f98a8bf9 DG |
147 | if (kvm->arch.hpt.order == order) { |
148 | /* We already have a suitable HPT */ | |
149 | ||
32fad281 | 150 | /* Set the entire HPT to 0, i.e. invalid HPTEs */ |
3f9d4f5a | 151 | memset((void *)kvm->arch.hpt.virt, 0, 1ul << order); |
a64fd707 PM |
152 | /* |
153 | * Reset all the reverse-mapping chains for all memslots | |
154 | */ | |
155 | kvmppc_rmap_reset(kvm); | |
32fad281 | 156 | err = 0; |
f98a8bf9 | 157 | goto out; |
32fad281 | 158 | } |
f98a8bf9 | 159 | |
ef427198 | 160 | if (kvm->arch.hpt.virt) { |
f98a8bf9 | 161 | kvmppc_free_hpt(&kvm->arch.hpt); |
ef427198 PM |
162 | kvmppc_rmap_reset(kvm); |
163 | } | |
f98a8bf9 DG |
164 | |
165 | err = kvmppc_allocate_hpt(&info, order); | |
166 | if (err < 0) | |
167 | goto out; | |
168 | kvmppc_set_hpt(kvm, &info); | |
169 | ||
170 | out: | |
ecba8297 DG |
171 | if (err == 0) |
172 | /* Ensure that each vcpu will flush its TLB on next entry. */ | |
173 | cpumask_setall(&kvm->arch.need_tlb_flush); | |
174 | ||
0d4ee88d | 175 | mutex_unlock(&kvm->arch.mmu_setup_lock); |
32fad281 PM |
176 | return err; |
177 | } | |
178 | ||
aae0777f | 179 | void kvmppc_free_hpt(struct kvm_hpt_info *info) |
de56a948 | 180 | { |
aae0777f | 181 | vfree(info->rev); |
18c3640c | 182 | info->rev = NULL; |
aae0777f DG |
183 | if (info->cma) |
184 | kvm_free_hpt_cma(virt_to_page(info->virt), | |
185 | 1 << (info->order - PAGE_SHIFT)); | |
186 | else if (info->virt) | |
187 | free_pages(info->virt, info->order - PAGE_SHIFT); | |
188 | info->virt = 0; | |
189 | info->order = 0; | |
de56a948 PM |
190 | } |
191 | ||
da9d1d7f PM |
192 | /* Bits in first HPTE dword for pagesize 4k, 64k or 16M */ |
193 | static inline unsigned long hpte0_pgsize_encoding(unsigned long pgsize) | |
194 | { | |
195 | return (pgsize > 0x1000) ? HPTE_V_LARGE : 0; | |
196 | } | |
197 | ||
198 | /* Bits in second HPTE dword for pagesize 4k, 64k or 16M */ | |
199 | static inline unsigned long hpte1_pgsize_encoding(unsigned long pgsize) | |
200 | { | |
201 | return (pgsize == 0x10000) ? 0x1000 : 0; | |
202 | } | |
203 | ||
204 | void kvmppc_map_vrma(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot, | |
205 | unsigned long porder) | |
de56a948 PM |
206 | { |
207 | unsigned long i; | |
b2b2f165 | 208 | unsigned long npages; |
c77162de PM |
209 | unsigned long hp_v, hp_r; |
210 | unsigned long addr, hash; | |
da9d1d7f PM |
211 | unsigned long psize; |
212 | unsigned long hp0, hp1; | |
7ed661bf | 213 | unsigned long idx_ret; |
c77162de | 214 | long ret; |
32fad281 | 215 | struct kvm *kvm = vcpu->kvm; |
de56a948 | 216 | |
da9d1d7f PM |
217 | psize = 1ul << porder; |
218 | npages = memslot->npages >> (porder - PAGE_SHIFT); | |
de56a948 PM |
219 | |
220 | /* VRMA can't be > 1TB */ | |
8936dda4 PM |
221 | if (npages > 1ul << (40 - porder)) |
222 | npages = 1ul << (40 - porder); | |
de56a948 | 223 | /* Can't use more than 1 HPTE per HPTEG */ |
3d089f84 DG |
224 | if (npages > kvmppc_hpt_mask(&kvm->arch.hpt) + 1) |
225 | npages = kvmppc_hpt_mask(&kvm->arch.hpt) + 1; | |
de56a948 | 226 | |
da9d1d7f PM |
227 | hp0 = HPTE_V_1TB_SEG | (VRMA_VSID << (40 - 16)) | |
228 | HPTE_V_BOLTED | hpte0_pgsize_encoding(psize); | |
229 | hp1 = hpte1_pgsize_encoding(psize) | | |
230 | HPTE_R_R | HPTE_R_C | HPTE_R_M | PP_RWXX; | |
231 | ||
de56a948 | 232 | for (i = 0; i < npages; ++i) { |
c77162de | 233 | addr = i << porder; |
de56a948 | 234 | /* can't use hpt_hash since va > 64 bits */ |
3d089f84 DG |
235 | hash = (i ^ (VRMA_VSID ^ (VRMA_VSID << 25))) |
236 | & kvmppc_hpt_mask(&kvm->arch.hpt); | |
de56a948 PM |
237 | /* |
238 | * We assume that the hash table is empty and no | |
239 | * vcpus are using it at this stage. Since we create | |
240 | * at most one HPTE per HPTEG, we just assume entry 7 | |
241 | * is available and use it. | |
242 | */ | |
8936dda4 | 243 | hash = (hash << 3) + 7; |
da9d1d7f PM |
244 | hp_v = hp0 | ((addr >> 16) & ~0x7fUL); |
245 | hp_r = hp1 | addr; | |
7ed661bf PM |
246 | ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, hash, hp_v, hp_r, |
247 | &idx_ret); | |
c77162de PM |
248 | if (ret != H_SUCCESS) { |
249 | pr_err("KVM: map_vrma at %lx failed, ret=%ld\n", | |
250 | addr, ret); | |
251 | break; | |
252 | } | |
de56a948 PM |
253 | } |
254 | } | |
255 | ||
256 | int kvmppc_mmu_hv_init(void) | |
257 | { | |
9e368f29 PM |
258 | unsigned long host_lpid, rsvd_lpid; |
259 | ||
b7557451 NP |
260 | if (!mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE)) |
261 | return -EINVAL; | |
262 | ||
c17b98cf | 263 | /* POWER7 has 10-bit LPIDs (12-bit in POWER8) */ |
f3c99f97 PM |
264 | host_lpid = 0; |
265 | if (cpu_has_feature(CPU_FTR_HVMODE)) | |
266 | host_lpid = mfspr(SPRN_LPID); | |
c17b98cf | 267 | rsvd_lpid = LPID_RSVD; |
9e368f29 | 268 | |
043cc4d7 SW |
269 | kvmppc_init_lpid(rsvd_lpid + 1); |
270 | ||
271 | kvmppc_claim_lpid(host_lpid); | |
9e368f29 | 272 | /* rsvd_lpid is reserved for use in partition switching */ |
043cc4d7 | 273 | kvmppc_claim_lpid(rsvd_lpid); |
de56a948 PM |
274 | |
275 | return 0; | |
276 | } | |
277 | ||
025c9511 | 278 | static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags, |
7ed661bf PM |
279 | long pte_index, unsigned long pteh, |
280 | unsigned long ptel, unsigned long *pte_idx_ret) | |
c77162de | 281 | { |
c77162de PM |
282 | long ret; |
283 | ||
342d3db7 PM |
284 | /* Protect linux PTE lookup from page table destruction */ |
285 | rcu_read_lock_sched(); /* this disables preemption too */ | |
7ed661bf | 286 | ret = kvmppc_do_h_enter(kvm, flags, pte_index, pteh, ptel, |
8a9c8925 | 287 | kvm->mm->pgd, false, pte_idx_ret); |
342d3db7 | 288 | rcu_read_unlock_sched(); |
c77162de PM |
289 | if (ret == H_TOO_HARD) { |
290 | /* this can't happen */ | |
291 | pr_err("KVM: Oops, kvmppc_h_enter returned too hard!\n"); | |
292 | ret = H_RESOURCE; /* or something */ | |
293 | } | |
294 | return ret; | |
295 | ||
296 | } | |
297 | ||
697d3899 PM |
298 | static struct kvmppc_slb *kvmppc_mmu_book3s_hv_find_slbe(struct kvm_vcpu *vcpu, |
299 | gva_t eaddr) | |
300 | { | |
301 | u64 mask; | |
302 | int i; | |
303 | ||
304 | for (i = 0; i < vcpu->arch.slb_nr; i++) { | |
305 | if (!(vcpu->arch.slb[i].orige & SLB_ESID_V)) | |
306 | continue; | |
307 | ||
308 | if (vcpu->arch.slb[i].origv & SLB_VSID_B_1T) | |
309 | mask = ESID_MASK_1T; | |
310 | else | |
311 | mask = ESID_MASK; | |
312 | ||
313 | if (((vcpu->arch.slb[i].orige ^ eaddr) & mask) == 0) | |
314 | return &vcpu->arch.slb[i]; | |
315 | } | |
316 | return NULL; | |
317 | } | |
318 | ||
319 | static unsigned long kvmppc_mmu_get_real_addr(unsigned long v, unsigned long r, | |
320 | unsigned long ea) | |
321 | { | |
322 | unsigned long ra_mask; | |
323 | ||
8dc6cca5 | 324 | ra_mask = kvmppc_actual_pgsz(v, r) - 1; |
697d3899 PM |
325 | return (r & HPTE_R_RPN & ~ra_mask) | (ea & ra_mask); |
326 | } | |
327 | ||
de56a948 | 328 | static int kvmppc_mmu_book3s_64_hv_xlate(struct kvm_vcpu *vcpu, gva_t eaddr, |
93b159b4 | 329 | struct kvmppc_pte *gpte, bool data, bool iswrite) |
de56a948 | 330 | { |
697d3899 PM |
331 | struct kvm *kvm = vcpu->kvm; |
332 | struct kvmppc_slb *slbe; | |
333 | unsigned long slb_v; | |
334 | unsigned long pp, key; | |
abb7c7dd | 335 | unsigned long v, orig_v, gr; |
6f22bd32 | 336 | __be64 *hptep; |
46dec40f | 337 | long int index; |
697d3899 PM |
338 | int virtmode = vcpu->arch.shregs.msr & (data ? MSR_DR : MSR_IR); |
339 | ||
18c3640c PM |
340 | if (kvm_is_radix(vcpu->kvm)) |
341 | return kvmppc_mmu_radix_xlate(vcpu, eaddr, gpte, data, iswrite); | |
342 | ||
697d3899 PM |
343 | /* Get SLB entry */ |
344 | if (virtmode) { | |
345 | slbe = kvmppc_mmu_book3s_hv_find_slbe(vcpu, eaddr); | |
346 | if (!slbe) | |
347 | return -EINVAL; | |
348 | slb_v = slbe->origv; | |
349 | } else { | |
350 | /* real mode access */ | |
351 | slb_v = vcpu->kvm->arch.vrma_slb_v; | |
352 | } | |
353 | ||
91648ec0 | 354 | preempt_disable(); |
697d3899 PM |
355 | /* Find the HPTE in the hash table */ |
356 | index = kvmppc_hv_find_lock_hpte(kvm, eaddr, slb_v, | |
357 | HPTE_V_VALID | HPTE_V_ABSENT); | |
91648ec0 | 358 | if (index < 0) { |
359 | preempt_enable(); | |
697d3899 | 360 | return -ENOENT; |
91648ec0 | 361 | } |
3f9d4f5a | 362 | hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4)); |
abb7c7dd PM |
363 | v = orig_v = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK; |
364 | if (cpu_has_feature(CPU_FTR_ARCH_300)) | |
365 | v = hpte_new_to_old_v(v, be64_to_cpu(hptep[1])); | |
3f9d4f5a | 366 | gr = kvm->arch.hpt.rev[index].guest_rpte; |
697d3899 | 367 | |
abb7c7dd | 368 | unlock_hpte(hptep, orig_v); |
91648ec0 | 369 | preempt_enable(); |
697d3899 PM |
370 | |
371 | gpte->eaddr = eaddr; | |
372 | gpte->vpage = ((v & HPTE_V_AVPN) << 4) | ((eaddr >> 12) & 0xfff); | |
373 | ||
374 | /* Get PP bits and key for permission check */ | |
375 | pp = gr & (HPTE_R_PP0 | HPTE_R_PP); | |
376 | key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS; | |
377 | key &= slb_v; | |
378 | ||
379 | /* Calculate permissions */ | |
380 | gpte->may_read = hpte_read_permission(pp, key); | |
381 | gpte->may_write = hpte_write_permission(pp, key); | |
382 | gpte->may_execute = gpte->may_read && !(gr & (HPTE_R_N | HPTE_R_G)); | |
383 | ||
384 | /* Storage key permission check for POWER7 */ | |
c17b98cf | 385 | if (data && virtmode) { |
697d3899 PM |
386 | int amrfield = hpte_get_skey_perm(gr, vcpu->arch.amr); |
387 | if (amrfield & 1) | |
388 | gpte->may_read = 0; | |
389 | if (amrfield & 2) | |
390 | gpte->may_write = 0; | |
391 | } | |
392 | ||
393 | /* Get the guest physical address */ | |
394 | gpte->raddr = kvmppc_mmu_get_real_addr(v, gr, eaddr); | |
395 | return 0; | |
396 | } | |
397 | ||
398 | /* | |
399 | * Quick test for whether an instruction is a load or a store. | |
400 | * If the instruction is a load or a store, then this will indicate | |
401 | * which it is, at least on server processors. (Embedded processors | |
402 | * have some external PID instructions that don't follow the rule | |
403 | * embodied here.) If the instruction isn't a load or store, then | |
404 | * this doesn't return anything useful. | |
405 | */ | |
406 | static int instruction_is_store(unsigned int instr) | |
407 | { | |
408 | unsigned int mask; | |
409 | ||
410 | mask = 0x10000000; | |
411 | if ((instr & 0xfc000000) == 0x7c000000) | |
412 | mask = 0x100; /* major opcode 31 */ | |
413 | return (instr & mask) != 0; | |
414 | } | |
415 | ||
5a319350 PM |
416 | int kvmppc_hv_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu, |
417 | unsigned long gpa, gva_t ea, int is_store) | |
697d3899 | 418 | { |
697d3899 | 419 | u32 last_inst; |
697d3899 | 420 | |
1b642257 SJS |
421 | /* |
422 | * Fast path - check if the guest physical address corresponds to a | |
423 | * device on the FAST_MMIO_BUS, if so we can avoid loading the | |
424 | * instruction all together, then we can just handle it and return. | |
425 | */ | |
426 | if (is_store) { | |
427 | int idx, ret; | |
428 | ||
429 | idx = srcu_read_lock(&vcpu->kvm->srcu); | |
430 | ret = kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, (gpa_t) gpa, 0, | |
431 | NULL); | |
432 | srcu_read_unlock(&vcpu->kvm->srcu, idx); | |
433 | if (!ret) { | |
434 | kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4); | |
435 | return RESUME_GUEST; | |
436 | } | |
437 | } | |
438 | ||
51f04726 | 439 | /* |
697d3899 PM |
440 | * If we fail, we just return to the guest and try executing it again. |
441 | */ | |
51f04726 MC |
442 | if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) != |
443 | EMULATE_DONE) | |
444 | return RESUME_GUEST; | |
697d3899 PM |
445 | |
446 | /* | |
447 | * WARNING: We do not know for sure whether the instruction we just | |
448 | * read from memory is the same that caused the fault in the first | |
449 | * place. If the instruction we read is neither an load or a store, | |
450 | * then it can't access memory, so we don't need to worry about | |
451 | * enforcing access permissions. So, assuming it is a load or | |
452 | * store, we just check that its direction (load or store) is | |
453 | * consistent with the original fault, since that's what we | |
454 | * checked the access permissions against. If there is a mismatch | |
455 | * we just return and retry the instruction. | |
456 | */ | |
457 | ||
51f04726 | 458 | if (instruction_is_store(last_inst) != !!is_store) |
697d3899 PM |
459 | return RESUME_GUEST; |
460 | ||
461 | /* | |
462 | * Emulated accesses are emulated by looking at the hash for | |
463 | * translation once, then performing the access later. The | |
464 | * translation could be invalidated in the meantime in which | |
465 | * point performing the subsequent memory access on the old | |
466 | * physical address could possibly be a security hole for the | |
467 | * guest (but not the host). | |
468 | * | |
469 | * This is less of an issue for MMIO stores since they aren't | |
470 | * globally visible. It could be an issue for MMIO loads to | |
471 | * a certain extent but we'll ignore it for now. | |
472 | */ | |
473 | ||
474 | vcpu->arch.paddr_accessed = gpa; | |
6020c0f6 | 475 | vcpu->arch.vaddr_accessed = ea; |
697d3899 PM |
476 | return kvmppc_emulate_mmio(run, vcpu); |
477 | } | |
478 | ||
479 | int kvmppc_book3s_hv_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu, | |
480 | unsigned long ea, unsigned long dsisr) | |
481 | { | |
482 | struct kvm *kvm = vcpu->kvm; | |
6f22bd32 | 483 | unsigned long hpte[3], r; |
abb7c7dd | 484 | unsigned long hnow_v, hnow_r; |
6f22bd32 | 485 | __be64 *hptep; |
342d3db7 | 486 | unsigned long mmu_seq, psize, pte_size; |
1066f772 | 487 | unsigned long gpa_base, gfn_base; |
cd758a9b | 488 | unsigned long gpa, gfn, hva, pfn, hpa; |
697d3899 | 489 | struct kvm_memory_slot *memslot; |
342d3db7 | 490 | unsigned long *rmap; |
697d3899 | 491 | struct revmap_entry *rev; |
cd758a9b PM |
492 | struct page *page; |
493 | long index, ret; | |
30bda41a | 494 | bool is_ci; |
cd758a9b PM |
495 | bool writing, write_ok; |
496 | unsigned int shift; | |
bad3b507 | 497 | unsigned long rcbits; |
a56ee9f8 | 498 | long mmio_update; |
cd758a9b | 499 | pte_t pte, *ptep; |
697d3899 | 500 | |
5a319350 PM |
501 | if (kvm_is_radix(kvm)) |
502 | return kvmppc_book3s_radix_page_fault(run, vcpu, ea, dsisr); | |
503 | ||
697d3899 PM |
504 | /* |
505 | * Real-mode code has already searched the HPT and found the | |
506 | * entry we're interested in. Lock the entry and check that | |
507 | * it hasn't changed. If it has, just return and re-execute the | |
508 | * instruction. | |
509 | */ | |
510 | if (ea != vcpu->arch.pgfault_addr) | |
511 | return RESUME_GUEST; | |
a56ee9f8 YX |
512 | |
513 | if (vcpu->arch.pgfault_cache) { | |
514 | mmio_update = atomic64_read(&kvm->arch.mmio_update); | |
515 | if (mmio_update == vcpu->arch.pgfault_cache->mmio_update) { | |
516 | r = vcpu->arch.pgfault_cache->rpte; | |
8dc6cca5 PM |
517 | psize = kvmppc_actual_pgsz(vcpu->arch.pgfault_hpte[0], |
518 | r); | |
a56ee9f8 YX |
519 | gpa_base = r & HPTE_R_RPN & ~(psize - 1); |
520 | gfn_base = gpa_base >> PAGE_SHIFT; | |
521 | gpa = gpa_base | (ea & (psize - 1)); | |
522 | return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea, | |
523 | dsisr & DSISR_ISSTORE); | |
524 | } | |
525 | } | |
697d3899 | 526 | index = vcpu->arch.pgfault_index; |
3f9d4f5a DG |
527 | hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4)); |
528 | rev = &kvm->arch.hpt.rev[index]; | |
697d3899 PM |
529 | preempt_disable(); |
530 | while (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) | |
531 | cpu_relax(); | |
6f22bd32 AG |
532 | hpte[0] = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK; |
533 | hpte[1] = be64_to_cpu(hptep[1]); | |
342d3db7 | 534 | hpte[2] = r = rev->guest_rpte; |
a4bd6eb0 | 535 | unlock_hpte(hptep, hpte[0]); |
697d3899 PM |
536 | preempt_enable(); |
537 | ||
abb7c7dd PM |
538 | if (cpu_has_feature(CPU_FTR_ARCH_300)) { |
539 | hpte[0] = hpte_new_to_old_v(hpte[0], hpte[1]); | |
540 | hpte[1] = hpte_new_to_old_r(hpte[1]); | |
541 | } | |
697d3899 PM |
542 | if (hpte[0] != vcpu->arch.pgfault_hpte[0] || |
543 | hpte[1] != vcpu->arch.pgfault_hpte[1]) | |
544 | return RESUME_GUEST; | |
545 | ||
546 | /* Translate the logical address and get the page */ | |
8dc6cca5 | 547 | psize = kvmppc_actual_pgsz(hpte[0], r); |
1066f772 PM |
548 | gpa_base = r & HPTE_R_RPN & ~(psize - 1); |
549 | gfn_base = gpa_base >> PAGE_SHIFT; | |
550 | gpa = gpa_base | (ea & (psize - 1)); | |
70bddfef | 551 | gfn = gpa >> PAGE_SHIFT; |
697d3899 PM |
552 | memslot = gfn_to_memslot(kvm, gfn); |
553 | ||
3c78f78a SW |
554 | trace_kvm_page_fault_enter(vcpu, hpte, memslot, ea, dsisr); |
555 | ||
697d3899 | 556 | /* No memslot means it's an emulated MMIO region */ |
70bddfef | 557 | if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) |
6020c0f6 | 558 | return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea, |
697d3899 | 559 | dsisr & DSISR_ISSTORE); |
697d3899 | 560 | |
1066f772 PM |
561 | /* |
562 | * This should never happen, because of the slot_is_aligned() | |
563 | * check in kvmppc_do_h_enter(). | |
564 | */ | |
565 | if (gfn_base < memslot->base_gfn) | |
566 | return -EFAULT; | |
567 | ||
342d3db7 PM |
568 | /* used to check for invalidations in progress */ |
569 | mmu_seq = kvm->mmu_notifier_seq; | |
570 | smp_rmb(); | |
571 | ||
3c78f78a | 572 | ret = -EFAULT; |
342d3db7 | 573 | page = NULL; |
4cf302bc PM |
574 | writing = (dsisr & DSISR_ISSTORE) != 0; |
575 | /* If writing != 0, then the HPTE must allow writing, if we get here */ | |
576 | write_ok = writing; | |
342d3db7 | 577 | hva = gfn_to_hva_memslot(memslot, gfn); |
cd758a9b PM |
578 | |
579 | /* | |
580 | * Do a fast check first, since __gfn_to_pfn_memslot doesn't | |
581 | * do it with !atomic && !async, which is how we call it. | |
582 | * We always ask for write permission since the common case | |
583 | * is that the page is writable. | |
584 | */ | |
585 | if (__get_user_pages_fast(hva, 1, 1, &page) == 1) { | |
586 | write_ok = true; | |
342d3db7 | 587 | } else { |
cd758a9b PM |
588 | /* Call KVM generic code to do the slow-path check */ |
589 | pfn = __gfn_to_pfn_memslot(memslot, gfn, false, NULL, | |
590 | writing, &write_ok); | |
591 | if (is_error_noslot_pfn(pfn)) | |
592 | return -EFAULT; | |
593 | page = NULL; | |
594 | if (pfn_valid(pfn)) { | |
595 | page = pfn_to_page(pfn); | |
596 | if (PageReserved(page)) | |
597 | page = NULL; | |
4cf302bc | 598 | } |
342d3db7 PM |
599 | } |
600 | ||
cd758a9b PM |
601 | /* |
602 | * Read the PTE from the process' radix tree and use that | |
603 | * so we get the shift and attribute bits. | |
604 | */ | |
605 | local_irq_disable(); | |
606 | ptep = __find_linux_pte(vcpu->arch.pgdir, hva, NULL, &shift); | |
ae49deda PM |
607 | pte = __pte(0); |
608 | if (ptep) | |
609 | pte = *ptep; | |
610 | local_irq_enable(); | |
cd758a9b PM |
611 | /* |
612 | * If the PTE disappeared temporarily due to a THP | |
613 | * collapse, just return and let the guest try again. | |
614 | */ | |
ae49deda | 615 | if (!pte_present(pte)) { |
cd758a9b PM |
616 | if (page) |
617 | put_page(page); | |
618 | return RESUME_GUEST; | |
619 | } | |
cd758a9b PM |
620 | hpa = pte_pfn(pte) << PAGE_SHIFT; |
621 | pte_size = PAGE_SIZE; | |
622 | if (shift) | |
623 | pte_size = 1ul << shift; | |
624 | is_ci = pte_ci(pte); | |
625 | ||
342d3db7 PM |
626 | if (psize > pte_size) |
627 | goto out_put; | |
cd758a9b PM |
628 | if (pte_size > psize) |
629 | hpa |= hva & (pte_size - psize); | |
342d3db7 PM |
630 | |
631 | /* Check WIMG vs. the actual page we're accessing */ | |
30bda41a AK |
632 | if (!hpte_cache_flags_ok(r, is_ci)) { |
633 | if (is_ci) | |
3c78f78a | 634 | goto out_put; |
342d3db7 PM |
635 | /* |
636 | * Allow guest to map emulated device memory as | |
637 | * uncacheable, but actually make it cacheable. | |
638 | */ | |
639 | r = (r & ~(HPTE_R_W|HPTE_R_I|HPTE_R_G)) | HPTE_R_M; | |
640 | } | |
641 | ||
caaa4c80 | 642 | /* |
cd758a9b PM |
643 | * Set the HPTE to point to hpa. |
644 | * Since the hpa is at PAGE_SIZE granularity, make sure we | |
caaa4c80 PM |
645 | * don't mask out lower-order bits if psize < PAGE_SIZE. |
646 | */ | |
647 | if (psize < PAGE_SIZE) | |
648 | psize = PAGE_SIZE; | |
cd758a9b | 649 | r = (r & HPTE_R_KEY_HI) | (r & ~(HPTE_R_PP0 - psize)) | hpa; |
4cf302bc PM |
650 | if (hpte_is_writable(r) && !write_ok) |
651 | r = hpte_make_readonly(r); | |
342d3db7 PM |
652 | ret = RESUME_GUEST; |
653 | preempt_disable(); | |
654 | while (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) | |
655 | cpu_relax(); | |
abb7c7dd PM |
656 | hnow_v = be64_to_cpu(hptep[0]); |
657 | hnow_r = be64_to_cpu(hptep[1]); | |
658 | if (cpu_has_feature(CPU_FTR_ARCH_300)) { | |
659 | hnow_v = hpte_new_to_old_v(hnow_v, hnow_r); | |
660 | hnow_r = hpte_new_to_old_r(hnow_r); | |
661 | } | |
38c53af8 PM |
662 | |
663 | /* | |
664 | * If the HPT is being resized, don't update the HPTE, | |
665 | * instead let the guest retry after the resize operation is complete. | |
072df813 | 666 | * The synchronization for mmu_ready test vs. set is provided |
38c53af8 PM |
667 | * by the HPTE lock. |
668 | */ | |
072df813 | 669 | if (!kvm->arch.mmu_ready) |
38c53af8 PM |
670 | goto out_unlock; |
671 | ||
abb7c7dd PM |
672 | if ((hnow_v & ~HPTE_V_HVLOCK) != hpte[0] || hnow_r != hpte[1] || |
673 | rev->guest_rpte != hpte[2]) | |
342d3db7 PM |
674 | /* HPTE has been changed under us; let the guest retry */ |
675 | goto out_unlock; | |
676 | hpte[0] = (hpte[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID; | |
677 | ||
1066f772 PM |
678 | /* Always put the HPTE in the rmap chain for the page base address */ |
679 | rmap = &memslot->arch.rmap[gfn_base - memslot->base_gfn]; | |
342d3db7 PM |
680 | lock_rmap(rmap); |
681 | ||
682 | /* Check if we might have been invalidated; let the guest retry if so */ | |
683 | ret = RESUME_GUEST; | |
8ca40a70 | 684 | if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) { |
342d3db7 PM |
685 | unlock_rmap(rmap); |
686 | goto out_unlock; | |
687 | } | |
4cf302bc | 688 | |
bad3b507 PM |
689 | /* Only set R/C in real HPTE if set in both *rmap and guest_rpte */ |
690 | rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT; | |
691 | r &= rcbits | ~(HPTE_R_R | HPTE_R_C); | |
692 | ||
6f22bd32 | 693 | if (be64_to_cpu(hptep[0]) & HPTE_V_VALID) { |
4cf302bc PM |
694 | /* HPTE was previously valid, so we need to invalidate it */ |
695 | unlock_rmap(rmap); | |
6f22bd32 | 696 | hptep[0] |= cpu_to_be64(HPTE_V_ABSENT); |
4cf302bc | 697 | kvmppc_invalidate_hpte(kvm, hptep, index); |
bad3b507 | 698 | /* don't lose previous R and C bits */ |
6f22bd32 | 699 | r |= be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C); |
4cf302bc PM |
700 | } else { |
701 | kvmppc_add_revmap_chain(kvm, rev, rmap, index, 0); | |
702 | } | |
342d3db7 | 703 | |
abb7c7dd PM |
704 | if (cpu_has_feature(CPU_FTR_ARCH_300)) { |
705 | r = hpte_old_to_new_r(hpte[0], r); | |
706 | hpte[0] = hpte_old_to_new_v(hpte[0]); | |
707 | } | |
6f22bd32 | 708 | hptep[1] = cpu_to_be64(r); |
342d3db7 | 709 | eieio(); |
a4bd6eb0 | 710 | __unlock_hpte(hptep, hpte[0]); |
342d3db7 PM |
711 | asm volatile("ptesync" : : : "memory"); |
712 | preempt_enable(); | |
4cf302bc | 713 | if (page && hpte_is_writable(r)) |
cd758a9b | 714 | set_page_dirty_lock(page); |
342d3db7 PM |
715 | |
716 | out_put: | |
3c78f78a SW |
717 | trace_kvm_page_fault_exit(vcpu, hpte, ret); |
718 | ||
cd758a9b PM |
719 | if (page) |
720 | put_page(page); | |
342d3db7 PM |
721 | return ret; |
722 | ||
723 | out_unlock: | |
a4bd6eb0 | 724 | __unlock_hpte(hptep, be64_to_cpu(hptep[0])); |
342d3db7 PM |
725 | preempt_enable(); |
726 | goto out_put; | |
727 | } | |
728 | ||
18c3640c | 729 | void kvmppc_rmap_reset(struct kvm *kvm) |
a64fd707 PM |
730 | { |
731 | struct kvm_memslots *slots; | |
732 | struct kvm_memory_slot *memslot; | |
733 | int srcu_idx; | |
734 | ||
735 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
9f6b8029 | 736 | slots = kvm_memslots(kvm); |
a64fd707 | 737 | kvm_for_each_memslot(memslot, slots) { |
234ff0b7 PM |
738 | /* Mutual exclusion with kvm_unmap_hva_range etc. */ |
739 | spin_lock(&kvm->mmu_lock); | |
a64fd707 PM |
740 | /* |
741 | * This assumes it is acceptable to lose reference and | |
742 | * change bits across a reset. | |
743 | */ | |
744 | memset(memslot->arch.rmap, 0, | |
745 | memslot->npages * sizeof(*memslot->arch.rmap)); | |
234ff0b7 | 746 | spin_unlock(&kvm->mmu_lock); |
a64fd707 PM |
747 | } |
748 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
749 | } | |
750 | ||
01756099 PM |
751 | typedef int (*hva_handler_fn)(struct kvm *kvm, struct kvm_memory_slot *memslot, |
752 | unsigned long gfn); | |
753 | ||
84504ef3 TY |
754 | static int kvm_handle_hva_range(struct kvm *kvm, |
755 | unsigned long start, | |
756 | unsigned long end, | |
01756099 | 757 | hva_handler_fn handler) |
342d3db7 PM |
758 | { |
759 | int ret; | |
760 | int retval = 0; | |
761 | struct kvm_memslots *slots; | |
762 | struct kvm_memory_slot *memslot; | |
763 | ||
764 | slots = kvm_memslots(kvm); | |
765 | kvm_for_each_memslot(memslot, slots) { | |
84504ef3 TY |
766 | unsigned long hva_start, hva_end; |
767 | gfn_t gfn, gfn_end; | |
768 | ||
769 | hva_start = max(start, memslot->userspace_addr); | |
770 | hva_end = min(end, memslot->userspace_addr + | |
771 | (memslot->npages << PAGE_SHIFT)); | |
772 | if (hva_start >= hva_end) | |
773 | continue; | |
774 | /* | |
775 | * {gfn(page) | page intersects with [hva_start, hva_end)} = | |
776 | * {gfn, gfn+1, ..., gfn_end-1}. | |
777 | */ | |
778 | gfn = hva_to_gfn_memslot(hva_start, memslot); | |
779 | gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot); | |
342d3db7 | 780 | |
84504ef3 | 781 | for (; gfn < gfn_end; ++gfn) { |
01756099 | 782 | ret = handler(kvm, memslot, gfn); |
342d3db7 PM |
783 | retval |= ret; |
784 | } | |
785 | } | |
786 | ||
787 | return retval; | |
788 | } | |
789 | ||
84504ef3 | 790 | static int kvm_handle_hva(struct kvm *kvm, unsigned long hva, |
01756099 | 791 | hva_handler_fn handler) |
84504ef3 TY |
792 | { |
793 | return kvm_handle_hva_range(kvm, hva, hva + 1, handler); | |
794 | } | |
795 | ||
639e4597 DG |
796 | /* Must be called with both HPTE and rmap locked */ |
797 | static void kvmppc_unmap_hpte(struct kvm *kvm, unsigned long i, | |
e641a317 | 798 | struct kvm_memory_slot *memslot, |
639e4597 DG |
799 | unsigned long *rmapp, unsigned long gfn) |
800 | { | |
801 | __be64 *hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4)); | |
802 | struct revmap_entry *rev = kvm->arch.hpt.rev; | |
803 | unsigned long j, h; | |
804 | unsigned long ptel, psize, rcbits; | |
805 | ||
806 | j = rev[i].forw; | |
807 | if (j == i) { | |
808 | /* chain is now empty */ | |
809 | *rmapp &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX); | |
810 | } else { | |
811 | /* remove i from chain */ | |
812 | h = rev[i].back; | |
813 | rev[h].forw = j; | |
814 | rev[j].back = h; | |
815 | rev[i].forw = rev[i].back = i; | |
816 | *rmapp = (*rmapp & ~KVMPPC_RMAP_INDEX) | j; | |
817 | } | |
818 | ||
819 | /* Now check and modify the HPTE */ | |
820 | ptel = rev[i].guest_rpte; | |
8dc6cca5 | 821 | psize = kvmppc_actual_pgsz(be64_to_cpu(hptep[0]), ptel); |
639e4597 DG |
822 | if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) && |
823 | hpte_rpn(ptel, psize) == gfn) { | |
824 | hptep[0] |= cpu_to_be64(HPTE_V_ABSENT); | |
825 | kvmppc_invalidate_hpte(kvm, hptep, i); | |
826 | hptep[1] &= ~cpu_to_be64(HPTE_R_KEY_HI | HPTE_R_KEY_LO); | |
827 | /* Harvest R and C */ | |
828 | rcbits = be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C); | |
829 | *rmapp |= rcbits << KVMPPC_RMAP_RC_SHIFT; | |
e641a317 PM |
830 | if ((rcbits & HPTE_R_C) && memslot->dirty_bitmap) |
831 | kvmppc_update_dirty_map(memslot, gfn, psize); | |
639e4597 DG |
832 | if (rcbits & ~rev[i].guest_rpte) { |
833 | rev[i].guest_rpte = ptel | rcbits; | |
834 | note_hpte_modification(kvm, &rev[i]); | |
835 | } | |
836 | } | |
837 | } | |
838 | ||
01756099 | 839 | static int kvm_unmap_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot, |
342d3db7 PM |
840 | unsigned long gfn) |
841 | { | |
639e4597 | 842 | unsigned long i; |
6f22bd32 | 843 | __be64 *hptep; |
01756099 | 844 | unsigned long *rmapp; |
342d3db7 | 845 | |
01756099 | 846 | rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn]; |
342d3db7 | 847 | for (;;) { |
bad3b507 | 848 | lock_rmap(rmapp); |
342d3db7 | 849 | if (!(*rmapp & KVMPPC_RMAP_PRESENT)) { |
bad3b507 | 850 | unlock_rmap(rmapp); |
342d3db7 PM |
851 | break; |
852 | } | |
853 | ||
854 | /* | |
855 | * To avoid an ABBA deadlock with the HPTE lock bit, | |
bad3b507 PM |
856 | * we can't spin on the HPTE lock while holding the |
857 | * rmap chain lock. | |
342d3db7 PM |
858 | */ |
859 | i = *rmapp & KVMPPC_RMAP_INDEX; | |
3f9d4f5a | 860 | hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4)); |
bad3b507 PM |
861 | if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) { |
862 | /* unlock rmap before spinning on the HPTE lock */ | |
863 | unlock_rmap(rmapp); | |
6f22bd32 | 864 | while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK) |
bad3b507 PM |
865 | cpu_relax(); |
866 | continue; | |
867 | } | |
342d3db7 | 868 | |
e641a317 | 869 | kvmppc_unmap_hpte(kvm, i, memslot, rmapp, gfn); |
bad3b507 | 870 | unlock_rmap(rmapp); |
a4bd6eb0 | 871 | __unlock_hpte(hptep, be64_to_cpu(hptep[0])); |
342d3db7 PM |
872 | } |
873 | return 0; | |
874 | } | |
875 | ||
3a167bea | 876 | int kvm_unmap_hva_range_hv(struct kvm *kvm, unsigned long start, unsigned long end) |
b3ae2096 | 877 | { |
01756099 PM |
878 | hva_handler_fn handler; |
879 | ||
880 | handler = kvm_is_radix(kvm) ? kvm_unmap_radix : kvm_unmap_rmapp; | |
881 | kvm_handle_hva_range(kvm, start, end, handler); | |
b3ae2096 TY |
882 | return 0; |
883 | } | |
884 | ||
3a167bea AK |
885 | void kvmppc_core_flush_memslot_hv(struct kvm *kvm, |
886 | struct kvm_memory_slot *memslot) | |
dfe49dbd | 887 | { |
dfe49dbd PM |
888 | unsigned long gfn; |
889 | unsigned long n; | |
01756099 | 890 | unsigned long *rmapp; |
dfe49dbd | 891 | |
dfe49dbd | 892 | gfn = memslot->base_gfn; |
01756099 | 893 | rmapp = memslot->arch.rmap; |
5af3e9d0 PM |
894 | if (kvm_is_radix(kvm)) { |
895 | kvmppc_radix_flush_memslot(kvm, memslot); | |
896 | return; | |
897 | } | |
898 | ||
01756099 | 899 | for (n = memslot->npages; n; --n, ++gfn) { |
dfe49dbd PM |
900 | /* |
901 | * Testing the present bit without locking is OK because | |
902 | * the memslot has been marked invalid already, and hence | |
903 | * no new HPTEs referencing this page can be created, | |
904 | * thus the present bit can't go from 0 to 1. | |
905 | */ | |
906 | if (*rmapp & KVMPPC_RMAP_PRESENT) | |
01756099 | 907 | kvm_unmap_rmapp(kvm, memslot, gfn); |
dfe49dbd | 908 | ++rmapp; |
dfe49dbd PM |
909 | } |
910 | } | |
911 | ||
01756099 | 912 | static int kvm_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot, |
342d3db7 PM |
913 | unsigned long gfn) |
914 | { | |
3f9d4f5a | 915 | struct revmap_entry *rev = kvm->arch.hpt.rev; |
55514893 | 916 | unsigned long head, i, j; |
6f22bd32 | 917 | __be64 *hptep; |
55514893 | 918 | int ret = 0; |
01756099 | 919 | unsigned long *rmapp; |
55514893 | 920 | |
01756099 | 921 | rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn]; |
55514893 PM |
922 | retry: |
923 | lock_rmap(rmapp); | |
924 | if (*rmapp & KVMPPC_RMAP_REFERENCED) { | |
925 | *rmapp &= ~KVMPPC_RMAP_REFERENCED; | |
926 | ret = 1; | |
927 | } | |
928 | if (!(*rmapp & KVMPPC_RMAP_PRESENT)) { | |
929 | unlock_rmap(rmapp); | |
930 | return ret; | |
931 | } | |
932 | ||
933 | i = head = *rmapp & KVMPPC_RMAP_INDEX; | |
934 | do { | |
3f9d4f5a | 935 | hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4)); |
55514893 PM |
936 | j = rev[i].forw; |
937 | ||
938 | /* If this HPTE isn't referenced, ignore it */ | |
6f22bd32 | 939 | if (!(be64_to_cpu(hptep[1]) & HPTE_R_R)) |
55514893 PM |
940 | continue; |
941 | ||
942 | if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) { | |
943 | /* unlock rmap before spinning on the HPTE lock */ | |
944 | unlock_rmap(rmapp); | |
6f22bd32 | 945 | while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK) |
55514893 PM |
946 | cpu_relax(); |
947 | goto retry; | |
948 | } | |
949 | ||
950 | /* Now check and modify the HPTE */ | |
6f22bd32 AG |
951 | if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) && |
952 | (be64_to_cpu(hptep[1]) & HPTE_R_R)) { | |
55514893 | 953 | kvmppc_clear_ref_hpte(kvm, hptep, i); |
a1b4a0f6 PM |
954 | if (!(rev[i].guest_rpte & HPTE_R_R)) { |
955 | rev[i].guest_rpte |= HPTE_R_R; | |
956 | note_hpte_modification(kvm, &rev[i]); | |
957 | } | |
55514893 PM |
958 | ret = 1; |
959 | } | |
a4bd6eb0 | 960 | __unlock_hpte(hptep, be64_to_cpu(hptep[0])); |
55514893 PM |
961 | } while ((i = j) != head); |
962 | ||
963 | unlock_rmap(rmapp); | |
964 | return ret; | |
342d3db7 PM |
965 | } |
966 | ||
57128468 | 967 | int kvm_age_hva_hv(struct kvm *kvm, unsigned long start, unsigned long end) |
342d3db7 | 968 | { |
01756099 PM |
969 | hva_handler_fn handler; |
970 | ||
971 | handler = kvm_is_radix(kvm) ? kvm_age_radix : kvm_age_rmapp; | |
972 | return kvm_handle_hva_range(kvm, start, end, handler); | |
342d3db7 PM |
973 | } |
974 | ||
01756099 | 975 | static int kvm_test_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot, |
342d3db7 PM |
976 | unsigned long gfn) |
977 | { | |
3f9d4f5a | 978 | struct revmap_entry *rev = kvm->arch.hpt.rev; |
55514893 PM |
979 | unsigned long head, i, j; |
980 | unsigned long *hp; | |
981 | int ret = 1; | |
01756099 | 982 | unsigned long *rmapp; |
55514893 | 983 | |
01756099 | 984 | rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn]; |
55514893 PM |
985 | if (*rmapp & KVMPPC_RMAP_REFERENCED) |
986 | return 1; | |
987 | ||
988 | lock_rmap(rmapp); | |
989 | if (*rmapp & KVMPPC_RMAP_REFERENCED) | |
990 | goto out; | |
991 | ||
992 | if (*rmapp & KVMPPC_RMAP_PRESENT) { | |
993 | i = head = *rmapp & KVMPPC_RMAP_INDEX; | |
994 | do { | |
3f9d4f5a | 995 | hp = (unsigned long *)(kvm->arch.hpt.virt + (i << 4)); |
55514893 | 996 | j = rev[i].forw; |
6f22bd32 | 997 | if (be64_to_cpu(hp[1]) & HPTE_R_R) |
55514893 PM |
998 | goto out; |
999 | } while ((i = j) != head); | |
1000 | } | |
1001 | ret = 0; | |
1002 | ||
1003 | out: | |
1004 | unlock_rmap(rmapp); | |
1005 | return ret; | |
342d3db7 PM |
1006 | } |
1007 | ||
3a167bea | 1008 | int kvm_test_age_hva_hv(struct kvm *kvm, unsigned long hva) |
342d3db7 | 1009 | { |
01756099 PM |
1010 | hva_handler_fn handler; |
1011 | ||
1012 | handler = kvm_is_radix(kvm) ? kvm_test_age_radix : kvm_test_age_rmapp; | |
1013 | return kvm_handle_hva(kvm, hva, handler); | |
342d3db7 PM |
1014 | } |
1015 | ||
3a167bea | 1016 | void kvm_set_spte_hva_hv(struct kvm *kvm, unsigned long hva, pte_t pte) |
342d3db7 | 1017 | { |
01756099 PM |
1018 | hva_handler_fn handler; |
1019 | ||
1020 | handler = kvm_is_radix(kvm) ? kvm_unmap_radix : kvm_unmap_rmapp; | |
1021 | kvm_handle_hva(kvm, hva, handler); | |
de56a948 PM |
1022 | } |
1023 | ||
6c576e74 PM |
1024 | static int vcpus_running(struct kvm *kvm) |
1025 | { | |
1026 | return atomic_read(&kvm->arch.vcpus_running) != 0; | |
1027 | } | |
1028 | ||
687414be AK |
1029 | /* |
1030 | * Returns the number of system pages that are dirty. | |
1031 | * This can be more than 1 if we find a huge-page HPTE. | |
1032 | */ | |
1033 | static int kvm_test_clear_dirty_npages(struct kvm *kvm, unsigned long *rmapp) | |
82ed3616 | 1034 | { |
3f9d4f5a | 1035 | struct revmap_entry *rev = kvm->arch.hpt.rev; |
82ed3616 | 1036 | unsigned long head, i, j; |
687414be | 1037 | unsigned long n; |
6c576e74 | 1038 | unsigned long v, r; |
6f22bd32 | 1039 | __be64 *hptep; |
687414be | 1040 | int npages_dirty = 0; |
82ed3616 PM |
1041 | |
1042 | retry: | |
1043 | lock_rmap(rmapp); | |
82ed3616 PM |
1044 | if (!(*rmapp & KVMPPC_RMAP_PRESENT)) { |
1045 | unlock_rmap(rmapp); | |
687414be | 1046 | return npages_dirty; |
82ed3616 PM |
1047 | } |
1048 | ||
1049 | i = head = *rmapp & KVMPPC_RMAP_INDEX; | |
1050 | do { | |
6f22bd32 | 1051 | unsigned long hptep1; |
3f9d4f5a | 1052 | hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4)); |
82ed3616 PM |
1053 | j = rev[i].forw; |
1054 | ||
6c576e74 PM |
1055 | /* |
1056 | * Checking the C (changed) bit here is racy since there | |
1057 | * is no guarantee about when the hardware writes it back. | |
1058 | * If the HPTE is not writable then it is stable since the | |
1059 | * page can't be written to, and we would have done a tlbie | |
1060 | * (which forces the hardware to complete any writeback) | |
1061 | * when making the HPTE read-only. | |
1062 | * If vcpus are running then this call is racy anyway | |
1063 | * since the page could get dirtied subsequently, so we | |
1064 | * expect there to be a further call which would pick up | |
1065 | * any delayed C bit writeback. | |
1066 | * Otherwise we need to do the tlbie even if C==0 in | |
1067 | * order to pick up any delayed writeback of C. | |
1068 | */ | |
6f22bd32 AG |
1069 | hptep1 = be64_to_cpu(hptep[1]); |
1070 | if (!(hptep1 & HPTE_R_C) && | |
1071 | (!hpte_is_writable(hptep1) || vcpus_running(kvm))) | |
82ed3616 PM |
1072 | continue; |
1073 | ||
1074 | if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) { | |
1075 | /* unlock rmap before spinning on the HPTE lock */ | |
1076 | unlock_rmap(rmapp); | |
6f22bd32 | 1077 | while (hptep[0] & cpu_to_be64(HPTE_V_HVLOCK)) |
82ed3616 PM |
1078 | cpu_relax(); |
1079 | goto retry; | |
1080 | } | |
1081 | ||
1082 | /* Now check and modify the HPTE */ | |
f6fb9e84 | 1083 | if (!(hptep[0] & cpu_to_be64(HPTE_V_VALID))) { |
a4bd6eb0 | 1084 | __unlock_hpte(hptep, be64_to_cpu(hptep[0])); |
6c576e74 | 1085 | continue; |
f6fb9e84 | 1086 | } |
6c576e74 PM |
1087 | |
1088 | /* need to make it temporarily absent so C is stable */ | |
6f22bd32 | 1089 | hptep[0] |= cpu_to_be64(HPTE_V_ABSENT); |
6c576e74 | 1090 | kvmppc_invalidate_hpte(kvm, hptep, i); |
6f22bd32 AG |
1091 | v = be64_to_cpu(hptep[0]); |
1092 | r = be64_to_cpu(hptep[1]); | |
6c576e74 | 1093 | if (r & HPTE_R_C) { |
6f22bd32 | 1094 | hptep[1] = cpu_to_be64(r & ~HPTE_R_C); |
a1b4a0f6 PM |
1095 | if (!(rev[i].guest_rpte & HPTE_R_C)) { |
1096 | rev[i].guest_rpte |= HPTE_R_C; | |
1097 | note_hpte_modification(kvm, &rev[i]); | |
1098 | } | |
8dc6cca5 | 1099 | n = kvmppc_actual_pgsz(v, r); |
687414be AK |
1100 | n = (n + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1101 | if (n > npages_dirty) | |
1102 | npages_dirty = n; | |
6c576e74 | 1103 | eieio(); |
82ed3616 | 1104 | } |
a4bd6eb0 | 1105 | v &= ~HPTE_V_ABSENT; |
6c576e74 | 1106 | v |= HPTE_V_VALID; |
a4bd6eb0 | 1107 | __unlock_hpte(hptep, v); |
82ed3616 PM |
1108 | } while ((i = j) != head); |
1109 | ||
1110 | unlock_rmap(rmapp); | |
687414be | 1111 | return npages_dirty; |
82ed3616 PM |
1112 | } |
1113 | ||
8f7b79b8 | 1114 | void kvmppc_harvest_vpa_dirty(struct kvmppc_vpa *vpa, |
c35635ef PM |
1115 | struct kvm_memory_slot *memslot, |
1116 | unsigned long *map) | |
1117 | { | |
1118 | unsigned long gfn; | |
1119 | ||
1120 | if (!vpa->dirty || !vpa->pinned_addr) | |
1121 | return; | |
1122 | gfn = vpa->gpa >> PAGE_SHIFT; | |
1123 | if (gfn < memslot->base_gfn || | |
1124 | gfn >= memslot->base_gfn + memslot->npages) | |
1125 | return; | |
1126 | ||
1127 | vpa->dirty = false; | |
1128 | if (map) | |
1129 | __set_bit_le(gfn - memslot->base_gfn, map); | |
1130 | } | |
1131 | ||
8f7b79b8 PM |
1132 | long kvmppc_hv_get_dirty_log_hpt(struct kvm *kvm, |
1133 | struct kvm_memory_slot *memslot, unsigned long *map) | |
82ed3616 | 1134 | { |
e641a317 | 1135 | unsigned long i; |
dfe49dbd | 1136 | unsigned long *rmapp; |
82ed3616 PM |
1137 | |
1138 | preempt_disable(); | |
d89cc617 | 1139 | rmapp = memslot->arch.rmap; |
82ed3616 | 1140 | for (i = 0; i < memslot->npages; ++i) { |
687414be AK |
1141 | int npages = kvm_test_clear_dirty_npages(kvm, rmapp); |
1142 | /* | |
1143 | * Note that if npages > 0 then i must be a multiple of npages, | |
1144 | * since we always put huge-page HPTEs in the rmap chain | |
1145 | * corresponding to their page base address. | |
1146 | */ | |
e641a317 PM |
1147 | if (npages) |
1148 | set_dirty_bits(map, i, npages); | |
82ed3616 PM |
1149 | ++rmapp; |
1150 | } | |
1151 | preempt_enable(); | |
1152 | return 0; | |
1153 | } | |
1154 | ||
93e60249 PM |
1155 | void *kvmppc_pin_guest_page(struct kvm *kvm, unsigned long gpa, |
1156 | unsigned long *nb_ret) | |
1157 | { | |
1158 | struct kvm_memory_slot *memslot; | |
1159 | unsigned long gfn = gpa >> PAGE_SHIFT; | |
342d3db7 PM |
1160 | struct page *page, *pages[1]; |
1161 | int npages; | |
c35635ef | 1162 | unsigned long hva, offset; |
2c9097e4 | 1163 | int srcu_idx; |
93e60249 | 1164 | |
2c9097e4 | 1165 | srcu_idx = srcu_read_lock(&kvm->srcu); |
93e60249 PM |
1166 | memslot = gfn_to_memslot(kvm, gfn); |
1167 | if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) | |
2c9097e4 | 1168 | goto err; |
c17b98cf | 1169 | hva = gfn_to_hva_memslot(memslot, gfn); |
73b0140b | 1170 | npages = get_user_pages_fast(hva, 1, FOLL_WRITE, pages); |
c17b98cf PM |
1171 | if (npages < 1) |
1172 | goto err; | |
1173 | page = pages[0]; | |
2c9097e4 PM |
1174 | srcu_read_unlock(&kvm->srcu, srcu_idx); |
1175 | ||
c35635ef | 1176 | offset = gpa & (PAGE_SIZE - 1); |
93e60249 | 1177 | if (nb_ret) |
c35635ef | 1178 | *nb_ret = PAGE_SIZE - offset; |
93e60249 | 1179 | return page_address(page) + offset; |
2c9097e4 PM |
1180 | |
1181 | err: | |
1182 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
1183 | return NULL; | |
93e60249 PM |
1184 | } |
1185 | ||
c35635ef PM |
1186 | void kvmppc_unpin_guest_page(struct kvm *kvm, void *va, unsigned long gpa, |
1187 | bool dirty) | |
93e60249 PM |
1188 | { |
1189 | struct page *page = virt_to_page(va); | |
c35635ef PM |
1190 | struct kvm_memory_slot *memslot; |
1191 | unsigned long gfn; | |
c35635ef | 1192 | int srcu_idx; |
93e60249 | 1193 | |
93e60249 | 1194 | put_page(page); |
c35635ef | 1195 | |
c17b98cf | 1196 | if (!dirty) |
c35635ef PM |
1197 | return; |
1198 | ||
e641a317 | 1199 | /* We need to mark this page dirty in the memslot dirty_bitmap, if any */ |
c35635ef PM |
1200 | gfn = gpa >> PAGE_SHIFT; |
1201 | srcu_idx = srcu_read_lock(&kvm->srcu); | |
1202 | memslot = gfn_to_memslot(kvm, gfn); | |
e641a317 PM |
1203 | if (memslot && memslot->dirty_bitmap) |
1204 | set_bit_le(gfn - memslot->base_gfn, memslot->dirty_bitmap); | |
c35635ef | 1205 | srcu_read_unlock(&kvm->srcu, srcu_idx); |
93e60249 PM |
1206 | } |
1207 | ||
5e985969 DG |
1208 | /* |
1209 | * HPT resizing | |
1210 | */ | |
1211 | static int resize_hpt_allocate(struct kvm_resize_hpt *resize) | |
1212 | { | |
b5baa687 DG |
1213 | int rc; |
1214 | ||
1215 | rc = kvmppc_allocate_hpt(&resize->hpt, resize->order); | |
1216 | if (rc < 0) | |
1217 | return rc; | |
1218 | ||
1219 | resize_hpt_debug(resize, "resize_hpt_allocate(): HPT @ 0x%lx\n", | |
1220 | resize->hpt.virt); | |
1221 | ||
5e985969 DG |
1222 | return 0; |
1223 | } | |
1224 | ||
b5baa687 DG |
1225 | static unsigned long resize_hpt_rehash_hpte(struct kvm_resize_hpt *resize, |
1226 | unsigned long idx) | |
1227 | { | |
1228 | struct kvm *kvm = resize->kvm; | |
1229 | struct kvm_hpt_info *old = &kvm->arch.hpt; | |
1230 | struct kvm_hpt_info *new = &resize->hpt; | |
1231 | unsigned long old_hash_mask = (1ULL << (old->order - 7)) - 1; | |
1232 | unsigned long new_hash_mask = (1ULL << (new->order - 7)) - 1; | |
1233 | __be64 *hptep, *new_hptep; | |
1234 | unsigned long vpte, rpte, guest_rpte; | |
1235 | int ret; | |
1236 | struct revmap_entry *rev; | |
ded13fc1 | 1237 | unsigned long apsize, avpn, pteg, hash; |
b5baa687 | 1238 | unsigned long new_idx, new_pteg, replace_vpte; |
ded13fc1 | 1239 | int pshift; |
b5baa687 DG |
1240 | |
1241 | hptep = (__be64 *)(old->virt + (idx << 4)); | |
1242 | ||
1243 | /* Guest is stopped, so new HPTEs can't be added or faulted | |
1244 | * in, only unmapped or altered by host actions. So, it's | |
1245 | * safe to check this before we take the HPTE lock */ | |
1246 | vpte = be64_to_cpu(hptep[0]); | |
1247 | if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT)) | |
1248 | return 0; /* nothing to do */ | |
1249 | ||
1250 | while (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) | |
1251 | cpu_relax(); | |
1252 | ||
1253 | vpte = be64_to_cpu(hptep[0]); | |
1254 | ||
1255 | ret = 0; | |
1256 | if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT)) | |
1257 | /* Nothing to do */ | |
1258 | goto out; | |
1259 | ||
790a9df5 DG |
1260 | if (cpu_has_feature(CPU_FTR_ARCH_300)) { |
1261 | rpte = be64_to_cpu(hptep[1]); | |
1262 | vpte = hpte_new_to_old_v(vpte, rpte); | |
1263 | } | |
1264 | ||
b5baa687 DG |
1265 | /* Unmap */ |
1266 | rev = &old->rev[idx]; | |
1267 | guest_rpte = rev->guest_rpte; | |
1268 | ||
1269 | ret = -EIO; | |
8dc6cca5 | 1270 | apsize = kvmppc_actual_pgsz(vpte, guest_rpte); |
b5baa687 DG |
1271 | if (!apsize) |
1272 | goto out; | |
1273 | ||
1274 | if (vpte & HPTE_V_VALID) { | |
1275 | unsigned long gfn = hpte_rpn(guest_rpte, apsize); | |
1276 | int srcu_idx = srcu_read_lock(&kvm->srcu); | |
1277 | struct kvm_memory_slot *memslot = | |
1278 | __gfn_to_memslot(kvm_memslots(kvm), gfn); | |
1279 | ||
1280 | if (memslot) { | |
1281 | unsigned long *rmapp; | |
1282 | rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn]; | |
1283 | ||
1284 | lock_rmap(rmapp); | |
e641a317 | 1285 | kvmppc_unmap_hpte(kvm, idx, memslot, rmapp, gfn); |
b5baa687 DG |
1286 | unlock_rmap(rmapp); |
1287 | } | |
1288 | ||
1289 | srcu_read_unlock(&kvm->srcu, srcu_idx); | |
1290 | } | |
1291 | ||
1292 | /* Reload PTE after unmap */ | |
1293 | vpte = be64_to_cpu(hptep[0]); | |
b5baa687 DG |
1294 | BUG_ON(vpte & HPTE_V_VALID); |
1295 | BUG_ON(!(vpte & HPTE_V_ABSENT)); | |
1296 | ||
1297 | ret = 0; | |
1298 | if (!(vpte & HPTE_V_BOLTED)) | |
1299 | goto out; | |
1300 | ||
1301 | rpte = be64_to_cpu(hptep[1]); | |
790a9df5 DG |
1302 | |
1303 | if (cpu_has_feature(CPU_FTR_ARCH_300)) { | |
1304 | vpte = hpte_new_to_old_v(vpte, rpte); | |
1305 | rpte = hpte_new_to_old_r(rpte); | |
1306 | } | |
1307 | ||
ded13fc1 PM |
1308 | pshift = kvmppc_hpte_base_page_shift(vpte, rpte); |
1309 | avpn = HPTE_V_AVPN_VAL(vpte) & ~(((1ul << pshift) - 1) >> 23); | |
b5baa687 DG |
1310 | pteg = idx / HPTES_PER_GROUP; |
1311 | if (vpte & HPTE_V_SECONDARY) | |
1312 | pteg = ~pteg; | |
1313 | ||
1314 | if (!(vpte & HPTE_V_1TB_SEG)) { | |
1315 | unsigned long offset, vsid; | |
1316 | ||
1317 | /* We only have 28 - 23 bits of offset in avpn */ | |
1318 | offset = (avpn & 0x1f) << 23; | |
1319 | vsid = avpn >> 5; | |
1320 | /* We can find more bits from the pteg value */ | |
ded13fc1 PM |
1321 | if (pshift < 23) |
1322 | offset |= ((vsid ^ pteg) & old_hash_mask) << pshift; | |
b5baa687 | 1323 | |
ded13fc1 | 1324 | hash = vsid ^ (offset >> pshift); |
b5baa687 DG |
1325 | } else { |
1326 | unsigned long offset, vsid; | |
1327 | ||
1328 | /* We only have 40 - 23 bits of seg_off in avpn */ | |
1329 | offset = (avpn & 0x1ffff) << 23; | |
1330 | vsid = avpn >> 17; | |
ded13fc1 PM |
1331 | if (pshift < 23) |
1332 | offset |= ((vsid ^ (vsid << 25) ^ pteg) & old_hash_mask) << pshift; | |
b5baa687 | 1333 | |
ded13fc1 | 1334 | hash = vsid ^ (vsid << 25) ^ (offset >> pshift); |
b5baa687 DG |
1335 | } |
1336 | ||
1337 | new_pteg = hash & new_hash_mask; | |
05f2bb03 PM |
1338 | if (vpte & HPTE_V_SECONDARY) |
1339 | new_pteg = ~hash & new_hash_mask; | |
b5baa687 DG |
1340 | |
1341 | new_idx = new_pteg * HPTES_PER_GROUP + (idx % HPTES_PER_GROUP); | |
1342 | new_hptep = (__be64 *)(new->virt + (new_idx << 4)); | |
1343 | ||
1344 | replace_vpte = be64_to_cpu(new_hptep[0]); | |
790a9df5 DG |
1345 | if (cpu_has_feature(CPU_FTR_ARCH_300)) { |
1346 | unsigned long replace_rpte = be64_to_cpu(new_hptep[1]); | |
1347 | replace_vpte = hpte_new_to_old_v(replace_vpte, replace_rpte); | |
1348 | } | |
b5baa687 DG |
1349 | |
1350 | if (replace_vpte & (HPTE_V_VALID | HPTE_V_ABSENT)) { | |
1351 | BUG_ON(new->order >= old->order); | |
1352 | ||
1353 | if (replace_vpte & HPTE_V_BOLTED) { | |
1354 | if (vpte & HPTE_V_BOLTED) | |
1355 | /* Bolted collision, nothing we can do */ | |
1356 | ret = -ENOSPC; | |
1357 | /* Discard the new HPTE */ | |
1358 | goto out; | |
1359 | } | |
1360 | ||
1361 | /* Discard the previous HPTE */ | |
1362 | } | |
1363 | ||
790a9df5 DG |
1364 | if (cpu_has_feature(CPU_FTR_ARCH_300)) { |
1365 | rpte = hpte_old_to_new_r(vpte, rpte); | |
1366 | vpte = hpte_old_to_new_v(vpte); | |
1367 | } | |
1368 | ||
b5baa687 DG |
1369 | new_hptep[1] = cpu_to_be64(rpte); |
1370 | new->rev[new_idx].guest_rpte = guest_rpte; | |
1371 | /* No need for a barrier, since new HPT isn't active */ | |
1372 | new_hptep[0] = cpu_to_be64(vpte); | |
1373 | unlock_hpte(new_hptep, vpte); | |
1374 | ||
1375 | out: | |
1376 | unlock_hpte(hptep, vpte); | |
1377 | return ret; | |
1378 | } | |
1379 | ||
5e985969 DG |
1380 | static int resize_hpt_rehash(struct kvm_resize_hpt *resize) |
1381 | { | |
b5baa687 DG |
1382 | struct kvm *kvm = resize->kvm; |
1383 | unsigned long i; | |
1384 | int rc; | |
1385 | ||
1386 | for (i = 0; i < kvmppc_hpt_npte(&kvm->arch.hpt); i++) { | |
1387 | rc = resize_hpt_rehash_hpte(resize, i); | |
1388 | if (rc != 0) | |
1389 | return rc; | |
1390 | } | |
1391 | ||
1392 | return 0; | |
5e985969 DG |
1393 | } |
1394 | ||
1395 | static void resize_hpt_pivot(struct kvm_resize_hpt *resize) | |
1396 | { | |
b5baa687 DG |
1397 | struct kvm *kvm = resize->kvm; |
1398 | struct kvm_hpt_info hpt_tmp; | |
1399 | ||
1400 | /* Exchange the pending tables in the resize structure with | |
1401 | * the active tables */ | |
1402 | ||
1403 | resize_hpt_debug(resize, "resize_hpt_pivot()\n"); | |
1404 | ||
1405 | spin_lock(&kvm->mmu_lock); | |
1406 | asm volatile("ptesync" : : : "memory"); | |
1407 | ||
1408 | hpt_tmp = kvm->arch.hpt; | |
1409 | kvmppc_set_hpt(kvm, &resize->hpt); | |
1410 | resize->hpt = hpt_tmp; | |
1411 | ||
1412 | spin_unlock(&kvm->mmu_lock); | |
1413 | ||
1414 | synchronize_srcu_expedited(&kvm->srcu); | |
1415 | ||
790a9df5 DG |
1416 | if (cpu_has_feature(CPU_FTR_ARCH_300)) |
1417 | kvmppc_setup_partition_table(kvm); | |
1418 | ||
b5baa687 | 1419 | resize_hpt_debug(resize, "resize_hpt_pivot() done\n"); |
5e985969 DG |
1420 | } |
1421 | ||
1422 | static void resize_hpt_release(struct kvm *kvm, struct kvm_resize_hpt *resize) | |
1423 | { | |
0d4ee88d | 1424 | if (WARN_ON(!mutex_is_locked(&kvm->arch.mmu_setup_lock))) |
4ed11aee | 1425 | return; |
b5baa687 | 1426 | |
5b73d634 DG |
1427 | if (!resize) |
1428 | return; | |
1429 | ||
4ed11aee SP |
1430 | if (resize->error != -EBUSY) { |
1431 | if (resize->hpt.virt) | |
1432 | kvmppc_free_hpt(&resize->hpt); | |
1433 | kfree(resize); | |
1434 | } | |
b5baa687 | 1435 | |
4ed11aee SP |
1436 | if (kvm->arch.resize_hpt == resize) |
1437 | kvm->arch.resize_hpt = NULL; | |
5e985969 DG |
1438 | } |
1439 | ||
1440 | static void resize_hpt_prepare_work(struct work_struct *work) | |
1441 | { | |
1442 | struct kvm_resize_hpt *resize = container_of(work, | |
1443 | struct kvm_resize_hpt, | |
1444 | work); | |
1445 | struct kvm *kvm = resize->kvm; | |
4ed11aee | 1446 | int err = 0; |
5e985969 | 1447 | |
3073774e SP |
1448 | if (WARN_ON(resize->error != -EBUSY)) |
1449 | return; | |
1450 | ||
0d4ee88d | 1451 | mutex_lock(&kvm->arch.mmu_setup_lock); |
5e985969 | 1452 | |
4ed11aee SP |
1453 | /* Request is still current? */ |
1454 | if (kvm->arch.resize_hpt == resize) { | |
1455 | /* We may request large allocations here: | |
0d4ee88d | 1456 | * do not sleep with kvm->arch.mmu_setup_lock held for a while. |
4ed11aee | 1457 | */ |
0d4ee88d | 1458 | mutex_unlock(&kvm->arch.mmu_setup_lock); |
5e985969 | 1459 | |
4ed11aee SP |
1460 | resize_hpt_debug(resize, "resize_hpt_prepare_work(): order = %d\n", |
1461 | resize->order); | |
3073774e | 1462 | |
4ed11aee SP |
1463 | err = resize_hpt_allocate(resize); |
1464 | ||
1465 | /* We have strict assumption about -EBUSY | |
1466 | * when preparing for HPT resize. | |
1467 | */ | |
1468 | if (WARN_ON(err == -EBUSY)) | |
1469 | err = -EINPROGRESS; | |
1470 | ||
0d4ee88d | 1471 | mutex_lock(&kvm->arch.mmu_setup_lock); |
4ed11aee | 1472 | /* It is possible that kvm->arch.resize_hpt != resize |
0d4ee88d | 1473 | * after we grab kvm->arch.mmu_setup_lock again. |
4ed11aee SP |
1474 | */ |
1475 | } | |
5e985969 DG |
1476 | |
1477 | resize->error = err; | |
5e985969 | 1478 | |
4ed11aee SP |
1479 | if (kvm->arch.resize_hpt != resize) |
1480 | resize_hpt_release(kvm, resize); | |
1481 | ||
0d4ee88d | 1482 | mutex_unlock(&kvm->arch.mmu_setup_lock); |
5e985969 DG |
1483 | } |
1484 | ||
1485 | long kvm_vm_ioctl_resize_hpt_prepare(struct kvm *kvm, | |
1486 | struct kvm_ppc_resize_hpt *rhpt) | |
1487 | { | |
1488 | unsigned long flags = rhpt->flags; | |
1489 | unsigned long shift = rhpt->shift; | |
1490 | struct kvm_resize_hpt *resize; | |
1491 | int ret; | |
1492 | ||
891f1ebf | 1493 | if (flags != 0 || kvm_is_radix(kvm)) |
5e985969 DG |
1494 | return -EINVAL; |
1495 | ||
1496 | if (shift && ((shift < 18) || (shift > 46))) | |
1497 | return -EINVAL; | |
1498 | ||
0d4ee88d | 1499 | mutex_lock(&kvm->arch.mmu_setup_lock); |
5e985969 DG |
1500 | |
1501 | resize = kvm->arch.resize_hpt; | |
1502 | ||
1503 | if (resize) { | |
1504 | if (resize->order == shift) { | |
3073774e SP |
1505 | /* Suitable resize in progress? */ |
1506 | ret = resize->error; | |
1507 | if (ret == -EBUSY) | |
5e985969 | 1508 | ret = 100; /* estimated time in ms */ |
3073774e SP |
1509 | else if (ret) |
1510 | resize_hpt_release(kvm, resize); | |
5e985969 DG |
1511 | |
1512 | goto out; | |
1513 | } | |
1514 | ||
1515 | /* not suitable, cancel it */ | |
1516 | resize_hpt_release(kvm, resize); | |
1517 | } | |
1518 | ||
1519 | ret = 0; | |
1520 | if (!shift) | |
1521 | goto out; /* nothing to do */ | |
1522 | ||
1523 | /* start new resize */ | |
1524 | ||
1525 | resize = kzalloc(sizeof(*resize), GFP_KERNEL); | |
abd80dcb DC |
1526 | if (!resize) { |
1527 | ret = -ENOMEM; | |
1528 | goto out; | |
1529 | } | |
3073774e SP |
1530 | |
1531 | resize->error = -EBUSY; | |
5e985969 DG |
1532 | resize->order = shift; |
1533 | resize->kvm = kvm; | |
1534 | INIT_WORK(&resize->work, resize_hpt_prepare_work); | |
1535 | kvm->arch.resize_hpt = resize; | |
1536 | ||
1537 | schedule_work(&resize->work); | |
1538 | ||
1539 | ret = 100; /* estimated time in ms */ | |
1540 | ||
1541 | out: | |
0d4ee88d | 1542 | mutex_unlock(&kvm->arch.mmu_setup_lock); |
5e985969 DG |
1543 | return ret; |
1544 | } | |
1545 | ||
1546 | static void resize_hpt_boot_vcpu(void *opaque) | |
1547 | { | |
1548 | /* Nothing to do, just force a KVM exit */ | |
1549 | } | |
1550 | ||
1551 | long kvm_vm_ioctl_resize_hpt_commit(struct kvm *kvm, | |
1552 | struct kvm_ppc_resize_hpt *rhpt) | |
1553 | { | |
1554 | unsigned long flags = rhpt->flags; | |
1555 | unsigned long shift = rhpt->shift; | |
1556 | struct kvm_resize_hpt *resize; | |
1557 | long ret; | |
1558 | ||
891f1ebf | 1559 | if (flags != 0 || kvm_is_radix(kvm)) |
5e985969 DG |
1560 | return -EINVAL; |
1561 | ||
1562 | if (shift && ((shift < 18) || (shift > 46))) | |
1563 | return -EINVAL; | |
1564 | ||
0d4ee88d | 1565 | mutex_lock(&kvm->arch.mmu_setup_lock); |
5e985969 DG |
1566 | |
1567 | resize = kvm->arch.resize_hpt; | |
1568 | ||
1569 | /* This shouldn't be possible */ | |
1570 | ret = -EIO; | |
1b151ce4 | 1571 | if (WARN_ON(!kvm->arch.mmu_ready)) |
5e985969 DG |
1572 | goto out_no_hpt; |
1573 | ||
1574 | /* Stop VCPUs from running while we mess with the HPT */ | |
1b151ce4 | 1575 | kvm->arch.mmu_ready = 0; |
5e985969 DG |
1576 | smp_mb(); |
1577 | ||
1578 | /* Boot all CPUs out of the guest so they re-read | |
1b151ce4 | 1579 | * mmu_ready */ |
5e985969 DG |
1580 | on_each_cpu(resize_hpt_boot_vcpu, NULL, 1); |
1581 | ||
1582 | ret = -ENXIO; | |
1583 | if (!resize || (resize->order != shift)) | |
1584 | goto out; | |
1585 | ||
5e985969 | 1586 | ret = resize->error; |
3073774e | 1587 | if (ret) |
5e985969 DG |
1588 | goto out; |
1589 | ||
1590 | ret = resize_hpt_rehash(resize); | |
3073774e | 1591 | if (ret) |
5e985969 DG |
1592 | goto out; |
1593 | ||
1594 | resize_hpt_pivot(resize); | |
1595 | ||
1596 | out: | |
1597 | /* Let VCPUs run again */ | |
1b151ce4 | 1598 | kvm->arch.mmu_ready = 1; |
5e985969 DG |
1599 | smp_mb(); |
1600 | out_no_hpt: | |
1601 | resize_hpt_release(kvm, resize); | |
0d4ee88d | 1602 | mutex_unlock(&kvm->arch.mmu_setup_lock); |
5e985969 DG |
1603 | return ret; |
1604 | } | |
1605 | ||
a2932923 PM |
1606 | /* |
1607 | * Functions for reading and writing the hash table via reads and | |
1608 | * writes on a file descriptor. | |
1609 | * | |
1610 | * Reads return the guest view of the hash table, which has to be | |
1611 | * pieced together from the real hash table and the guest_rpte | |
1612 | * values in the revmap array. | |
1613 | * | |
1614 | * On writes, each HPTE written is considered in turn, and if it | |
1615 | * is valid, it is written to the HPT as if an H_ENTER with the | |
1616 | * exact flag set was done. When the invalid count is non-zero | |
1617 | * in the header written to the stream, the kernel will make | |
1618 | * sure that that many HPTEs are invalid, and invalidate them | |
1619 | * if not. | |
1620 | */ | |
1621 | ||
1622 | struct kvm_htab_ctx { | |
1623 | unsigned long index; | |
1624 | unsigned long flags; | |
1625 | struct kvm *kvm; | |
1626 | int first_pass; | |
1627 | }; | |
1628 | ||
1629 | #define HPTE_SIZE (2 * sizeof(unsigned long)) | |
1630 | ||
a1b4a0f6 PM |
1631 | /* |
1632 | * Returns 1 if this HPT entry has been modified or has pending | |
1633 | * R/C bit changes. | |
1634 | */ | |
6f22bd32 | 1635 | static int hpte_dirty(struct revmap_entry *revp, __be64 *hptp) |
a1b4a0f6 PM |
1636 | { |
1637 | unsigned long rcbits_unset; | |
1638 | ||
1639 | if (revp->guest_rpte & HPTE_GR_MODIFIED) | |
1640 | return 1; | |
1641 | ||
1642 | /* Also need to consider changes in reference and changed bits */ | |
1643 | rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C); | |
6f22bd32 AG |
1644 | if ((be64_to_cpu(hptp[0]) & HPTE_V_VALID) && |
1645 | (be64_to_cpu(hptp[1]) & rcbits_unset)) | |
a1b4a0f6 PM |
1646 | return 1; |
1647 | ||
1648 | return 0; | |
1649 | } | |
1650 | ||
6f22bd32 | 1651 | static long record_hpte(unsigned long flags, __be64 *hptp, |
a2932923 PM |
1652 | unsigned long *hpte, struct revmap_entry *revp, |
1653 | int want_valid, int first_pass) | |
1654 | { | |
abb7c7dd | 1655 | unsigned long v, r, hr; |
a1b4a0f6 | 1656 | unsigned long rcbits_unset; |
a2932923 PM |
1657 | int ok = 1; |
1658 | int valid, dirty; | |
1659 | ||
1660 | /* Unmodified entries are uninteresting except on the first pass */ | |
a1b4a0f6 | 1661 | dirty = hpte_dirty(revp, hptp); |
a2932923 PM |
1662 | if (!first_pass && !dirty) |
1663 | return 0; | |
1664 | ||
1665 | valid = 0; | |
6f22bd32 | 1666 | if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)) { |
a2932923 PM |
1667 | valid = 1; |
1668 | if ((flags & KVM_GET_HTAB_BOLTED_ONLY) && | |
6f22bd32 | 1669 | !(be64_to_cpu(hptp[0]) & HPTE_V_BOLTED)) |
a2932923 PM |
1670 | valid = 0; |
1671 | } | |
1672 | if (valid != want_valid) | |
1673 | return 0; | |
1674 | ||
1675 | v = r = 0; | |
1676 | if (valid || dirty) { | |
1677 | /* lock the HPTE so it's stable and read it */ | |
1678 | preempt_disable(); | |
1679 | while (!try_lock_hpte(hptp, HPTE_V_HVLOCK)) | |
1680 | cpu_relax(); | |
6f22bd32 | 1681 | v = be64_to_cpu(hptp[0]); |
abb7c7dd PM |
1682 | hr = be64_to_cpu(hptp[1]); |
1683 | if (cpu_has_feature(CPU_FTR_ARCH_300)) { | |
1684 | v = hpte_new_to_old_v(v, hr); | |
1685 | hr = hpte_new_to_old_r(hr); | |
1686 | } | |
a1b4a0f6 PM |
1687 | |
1688 | /* re-evaluate valid and dirty from synchronized HPTE value */ | |
1689 | valid = !!(v & HPTE_V_VALID); | |
1690 | dirty = !!(revp->guest_rpte & HPTE_GR_MODIFIED); | |
1691 | ||
1692 | /* Harvest R and C into guest view if necessary */ | |
1693 | rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C); | |
abb7c7dd PM |
1694 | if (valid && (rcbits_unset & hr)) { |
1695 | revp->guest_rpte |= (hr & | |
6f22bd32 | 1696 | (HPTE_R_R | HPTE_R_C)) | HPTE_GR_MODIFIED; |
a1b4a0f6 PM |
1697 | dirty = 1; |
1698 | } | |
1699 | ||
a2932923 PM |
1700 | if (v & HPTE_V_ABSENT) { |
1701 | v &= ~HPTE_V_ABSENT; | |
1702 | v |= HPTE_V_VALID; | |
a1b4a0f6 | 1703 | valid = 1; |
a2932923 | 1704 | } |
a2932923 PM |
1705 | if ((flags & KVM_GET_HTAB_BOLTED_ONLY) && !(v & HPTE_V_BOLTED)) |
1706 | valid = 0; | |
a1b4a0f6 PM |
1707 | |
1708 | r = revp->guest_rpte; | |
a2932923 PM |
1709 | /* only clear modified if this is the right sort of entry */ |
1710 | if (valid == want_valid && dirty) { | |
1711 | r &= ~HPTE_GR_MODIFIED; | |
1712 | revp->guest_rpte = r; | |
1713 | } | |
a4bd6eb0 | 1714 | unlock_hpte(hptp, be64_to_cpu(hptp[0])); |
a2932923 PM |
1715 | preempt_enable(); |
1716 | if (!(valid == want_valid && (first_pass || dirty))) | |
1717 | ok = 0; | |
1718 | } | |
6f22bd32 AG |
1719 | hpte[0] = cpu_to_be64(v); |
1720 | hpte[1] = cpu_to_be64(r); | |
a2932923 PM |
1721 | return ok; |
1722 | } | |
1723 | ||
1724 | static ssize_t kvm_htab_read(struct file *file, char __user *buf, | |
1725 | size_t count, loff_t *ppos) | |
1726 | { | |
1727 | struct kvm_htab_ctx *ctx = file->private_data; | |
1728 | struct kvm *kvm = ctx->kvm; | |
1729 | struct kvm_get_htab_header hdr; | |
6f22bd32 | 1730 | __be64 *hptp; |
a2932923 PM |
1731 | struct revmap_entry *revp; |
1732 | unsigned long i, nb, nw; | |
1733 | unsigned long __user *lbuf; | |
1734 | struct kvm_get_htab_header __user *hptr; | |
1735 | unsigned long flags; | |
1736 | int first_pass; | |
1737 | unsigned long hpte[2]; | |
1738 | ||
96d4f267 | 1739 | if (!access_ok(buf, count)) |
a2932923 | 1740 | return -EFAULT; |
891f1ebf PM |
1741 | if (kvm_is_radix(kvm)) |
1742 | return 0; | |
a2932923 PM |
1743 | |
1744 | first_pass = ctx->first_pass; | |
1745 | flags = ctx->flags; | |
1746 | ||
1747 | i = ctx->index; | |
3f9d4f5a DG |
1748 | hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE)); |
1749 | revp = kvm->arch.hpt.rev + i; | |
a2932923 PM |
1750 | lbuf = (unsigned long __user *)buf; |
1751 | ||
1752 | nb = 0; | |
1753 | while (nb + sizeof(hdr) + HPTE_SIZE < count) { | |
1754 | /* Initialize header */ | |
1755 | hptr = (struct kvm_get_htab_header __user *)buf; | |
a2932923 PM |
1756 | hdr.n_valid = 0; |
1757 | hdr.n_invalid = 0; | |
1758 | nw = nb; | |
1759 | nb += sizeof(hdr); | |
1760 | lbuf = (unsigned long __user *)(buf + sizeof(hdr)); | |
1761 | ||
1762 | /* Skip uninteresting entries, i.e. clean on not-first pass */ | |
1763 | if (!first_pass) { | |
3d089f84 | 1764 | while (i < kvmppc_hpt_npte(&kvm->arch.hpt) && |
a1b4a0f6 | 1765 | !hpte_dirty(revp, hptp)) { |
a2932923 PM |
1766 | ++i; |
1767 | hptp += 2; | |
1768 | ++revp; | |
1769 | } | |
1770 | } | |
05dd85f7 | 1771 | hdr.index = i; |
a2932923 PM |
1772 | |
1773 | /* Grab a series of valid entries */ | |
3d089f84 | 1774 | while (i < kvmppc_hpt_npte(&kvm->arch.hpt) && |
a2932923 PM |
1775 | hdr.n_valid < 0xffff && |
1776 | nb + HPTE_SIZE < count && | |
1777 | record_hpte(flags, hptp, hpte, revp, 1, first_pass)) { | |
1778 | /* valid entry, write it out */ | |
1779 | ++hdr.n_valid; | |
1780 | if (__put_user(hpte[0], lbuf) || | |
1781 | __put_user(hpte[1], lbuf + 1)) | |
1782 | return -EFAULT; | |
1783 | nb += HPTE_SIZE; | |
1784 | lbuf += 2; | |
1785 | ++i; | |
1786 | hptp += 2; | |
1787 | ++revp; | |
1788 | } | |
1789 | /* Now skip invalid entries while we can */ | |
3d089f84 | 1790 | while (i < kvmppc_hpt_npte(&kvm->arch.hpt) && |
a2932923 PM |
1791 | hdr.n_invalid < 0xffff && |
1792 | record_hpte(flags, hptp, hpte, revp, 0, first_pass)) { | |
1793 | /* found an invalid entry */ | |
1794 | ++hdr.n_invalid; | |
1795 | ++i; | |
1796 | hptp += 2; | |
1797 | ++revp; | |
1798 | } | |
1799 | ||
1800 | if (hdr.n_valid || hdr.n_invalid) { | |
1801 | /* write back the header */ | |
1802 | if (__copy_to_user(hptr, &hdr, sizeof(hdr))) | |
1803 | return -EFAULT; | |
1804 | nw = nb; | |
1805 | buf = (char __user *)lbuf; | |
1806 | } else { | |
1807 | nb = nw; | |
1808 | } | |
1809 | ||
1810 | /* Check if we've wrapped around the hash table */ | |
3d089f84 | 1811 | if (i >= kvmppc_hpt_npte(&kvm->arch.hpt)) { |
a2932923 PM |
1812 | i = 0; |
1813 | ctx->first_pass = 0; | |
1814 | break; | |
1815 | } | |
1816 | } | |
1817 | ||
1818 | ctx->index = i; | |
1819 | ||
1820 | return nb; | |
1821 | } | |
1822 | ||
1823 | static ssize_t kvm_htab_write(struct file *file, const char __user *buf, | |
1824 | size_t count, loff_t *ppos) | |
1825 | { | |
1826 | struct kvm_htab_ctx *ctx = file->private_data; | |
1827 | struct kvm *kvm = ctx->kvm; | |
1828 | struct kvm_get_htab_header hdr; | |
1829 | unsigned long i, j; | |
1830 | unsigned long v, r; | |
1831 | unsigned long __user *lbuf; | |
6f22bd32 | 1832 | __be64 *hptp; |
a2932923 PM |
1833 | unsigned long tmp[2]; |
1834 | ssize_t nb; | |
1835 | long int err, ret; | |
1b151ce4 | 1836 | int mmu_ready; |
ded13fc1 | 1837 | int pshift; |
a2932923 | 1838 | |
96d4f267 | 1839 | if (!access_ok(buf, count)) |
a2932923 | 1840 | return -EFAULT; |
891f1ebf PM |
1841 | if (kvm_is_radix(kvm)) |
1842 | return -EINVAL; | |
a2932923 PM |
1843 | |
1844 | /* lock out vcpus from running while we're doing this */ | |
0d4ee88d | 1845 | mutex_lock(&kvm->arch.mmu_setup_lock); |
1b151ce4 PM |
1846 | mmu_ready = kvm->arch.mmu_ready; |
1847 | if (mmu_ready) { | |
1848 | kvm->arch.mmu_ready = 0; /* temporarily */ | |
1849 | /* order mmu_ready vs. vcpus_running */ | |
a2932923 PM |
1850 | smp_mb(); |
1851 | if (atomic_read(&kvm->arch.vcpus_running)) { | |
1b151ce4 | 1852 | kvm->arch.mmu_ready = 1; |
0d4ee88d | 1853 | mutex_unlock(&kvm->arch.mmu_setup_lock); |
a2932923 PM |
1854 | return -EBUSY; |
1855 | } | |
1856 | } | |
1857 | ||
1858 | err = 0; | |
1859 | for (nb = 0; nb + sizeof(hdr) <= count; ) { | |
1860 | err = -EFAULT; | |
1861 | if (__copy_from_user(&hdr, buf, sizeof(hdr))) | |
1862 | break; | |
1863 | ||
1864 | err = 0; | |
1865 | if (nb + hdr.n_valid * HPTE_SIZE > count) | |
1866 | break; | |
1867 | ||
1868 | nb += sizeof(hdr); | |
1869 | buf += sizeof(hdr); | |
1870 | ||
1871 | err = -EINVAL; | |
1872 | i = hdr.index; | |
3d089f84 DG |
1873 | if (i >= kvmppc_hpt_npte(&kvm->arch.hpt) || |
1874 | i + hdr.n_valid + hdr.n_invalid > kvmppc_hpt_npte(&kvm->arch.hpt)) | |
a2932923 PM |
1875 | break; |
1876 | ||
3f9d4f5a | 1877 | hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE)); |
a2932923 PM |
1878 | lbuf = (unsigned long __user *)buf; |
1879 | for (j = 0; j < hdr.n_valid; ++j) { | |
ffada016 CLG |
1880 | __be64 hpte_v; |
1881 | __be64 hpte_r; | |
1882 | ||
a2932923 | 1883 | err = -EFAULT; |
ffada016 CLG |
1884 | if (__get_user(hpte_v, lbuf) || |
1885 | __get_user(hpte_r, lbuf + 1)) | |
a2932923 | 1886 | goto out; |
ffada016 CLG |
1887 | v = be64_to_cpu(hpte_v); |
1888 | r = be64_to_cpu(hpte_r); | |
a2932923 PM |
1889 | err = -EINVAL; |
1890 | if (!(v & HPTE_V_VALID)) | |
1891 | goto out; | |
ded13fc1 PM |
1892 | pshift = kvmppc_hpte_base_page_shift(v, r); |
1893 | if (pshift <= 0) | |
1894 | goto out; | |
a2932923 PM |
1895 | lbuf += 2; |
1896 | nb += HPTE_SIZE; | |
1897 | ||
6f22bd32 | 1898 | if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)) |
a2932923 PM |
1899 | kvmppc_do_h_remove(kvm, 0, i, 0, tmp); |
1900 | err = -EIO; | |
1901 | ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, i, v, r, | |
1902 | tmp); | |
1903 | if (ret != H_SUCCESS) { | |
1904 | pr_err("kvm_htab_write ret %ld i=%ld v=%lx " | |
1905 | "r=%lx\n", ret, i, v, r); | |
1906 | goto out; | |
1907 | } | |
1b151ce4 | 1908 | if (!mmu_ready && is_vrma_hpte(v)) { |
ded13fc1 | 1909 | unsigned long senc, lpcr; |
a2932923 | 1910 | |
ded13fc1 | 1911 | senc = slb_pgsize_encoding(1ul << pshift); |
a2932923 PM |
1912 | kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T | |
1913 | (VRMA_VSID << SLB_VSID_SHIFT_1T); | |
ded13fc1 PM |
1914 | if (!cpu_has_feature(CPU_FTR_ARCH_300)) { |
1915 | lpcr = senc << (LPCR_VRMASD_SH - 4); | |
1916 | kvmppc_update_lpcr(kvm, lpcr, | |
1917 | LPCR_VRMASD); | |
1918 | } else { | |
1919 | kvmppc_setup_partition_table(kvm); | |
1920 | } | |
1b151ce4 | 1921 | mmu_ready = 1; |
a2932923 PM |
1922 | } |
1923 | ++i; | |
1924 | hptp += 2; | |
1925 | } | |
1926 | ||
1927 | for (j = 0; j < hdr.n_invalid; ++j) { | |
6f22bd32 | 1928 | if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)) |
a2932923 PM |
1929 | kvmppc_do_h_remove(kvm, 0, i, 0, tmp); |
1930 | ++i; | |
1931 | hptp += 2; | |
1932 | } | |
1933 | err = 0; | |
1934 | } | |
1935 | ||
1936 | out: | |
1b151ce4 | 1937 | /* Order HPTE updates vs. mmu_ready */ |
a2932923 | 1938 | smp_wmb(); |
1b151ce4 | 1939 | kvm->arch.mmu_ready = mmu_ready; |
0d4ee88d | 1940 | mutex_unlock(&kvm->arch.mmu_setup_lock); |
a2932923 PM |
1941 | |
1942 | if (err) | |
1943 | return err; | |
1944 | return nb; | |
1945 | } | |
1946 | ||
1947 | static int kvm_htab_release(struct inode *inode, struct file *filp) | |
1948 | { | |
1949 | struct kvm_htab_ctx *ctx = filp->private_data; | |
1950 | ||
1951 | filp->private_data = NULL; | |
1952 | if (!(ctx->flags & KVM_GET_HTAB_WRITE)) | |
1953 | atomic_dec(&ctx->kvm->arch.hpte_mod_interest); | |
1954 | kvm_put_kvm(ctx->kvm); | |
1955 | kfree(ctx); | |
1956 | return 0; | |
1957 | } | |
1958 | ||
75ef9de1 | 1959 | static const struct file_operations kvm_htab_fops = { |
a2932923 PM |
1960 | .read = kvm_htab_read, |
1961 | .write = kvm_htab_write, | |
1962 | .llseek = default_llseek, | |
1963 | .release = kvm_htab_release, | |
1964 | }; | |
1965 | ||
1966 | int kvm_vm_ioctl_get_htab_fd(struct kvm *kvm, struct kvm_get_htab_fd *ghf) | |
1967 | { | |
1968 | int ret; | |
1969 | struct kvm_htab_ctx *ctx; | |
1970 | int rwflag; | |
1971 | ||
1972 | /* reject flags we don't recognize */ | |
1973 | if (ghf->flags & ~(KVM_GET_HTAB_BOLTED_ONLY | KVM_GET_HTAB_WRITE)) | |
1974 | return -EINVAL; | |
1975 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); | |
1976 | if (!ctx) | |
1977 | return -ENOMEM; | |
1978 | kvm_get_kvm(kvm); | |
1979 | ctx->kvm = kvm; | |
1980 | ctx->index = ghf->start_index; | |
1981 | ctx->flags = ghf->flags; | |
1982 | ctx->first_pass = 1; | |
1983 | ||
1984 | rwflag = (ghf->flags & KVM_GET_HTAB_WRITE) ? O_WRONLY : O_RDONLY; | |
2f84d5ea | 1985 | ret = anon_inode_getfd("kvm-htab", &kvm_htab_fops, ctx, rwflag | O_CLOEXEC); |
a2932923 | 1986 | if (ret < 0) { |
43f6b0cf | 1987 | kfree(ctx); |
149487bd | 1988 | kvm_put_kvm_no_destroy(kvm); |
a2932923 PM |
1989 | return ret; |
1990 | } | |
1991 | ||
1992 | if (rwflag == O_RDONLY) { | |
1993 | mutex_lock(&kvm->slots_lock); | |
1994 | atomic_inc(&kvm->arch.hpte_mod_interest); | |
1995 | /* make sure kvmppc_do_h_enter etc. see the increment */ | |
1996 | synchronize_srcu_expedited(&kvm->srcu); | |
1997 | mutex_unlock(&kvm->slots_lock); | |
1998 | } | |
1999 | ||
2000 | return ret; | |
2001 | } | |
2002 | ||
e23a808b PM |
2003 | struct debugfs_htab_state { |
2004 | struct kvm *kvm; | |
2005 | struct mutex mutex; | |
2006 | unsigned long hpt_index; | |
2007 | int chars_left; | |
2008 | int buf_index; | |
2009 | char buf[64]; | |
2010 | }; | |
2011 | ||
2012 | static int debugfs_htab_open(struct inode *inode, struct file *file) | |
2013 | { | |
2014 | struct kvm *kvm = inode->i_private; | |
2015 | struct debugfs_htab_state *p; | |
2016 | ||
2017 | p = kzalloc(sizeof(*p), GFP_KERNEL); | |
2018 | if (!p) | |
2019 | return -ENOMEM; | |
2020 | ||
2021 | kvm_get_kvm(kvm); | |
2022 | p->kvm = kvm; | |
2023 | mutex_init(&p->mutex); | |
2024 | file->private_data = p; | |
2025 | ||
2026 | return nonseekable_open(inode, file); | |
2027 | } | |
2028 | ||
2029 | static int debugfs_htab_release(struct inode *inode, struct file *file) | |
2030 | { | |
2031 | struct debugfs_htab_state *p = file->private_data; | |
2032 | ||
2033 | kvm_put_kvm(p->kvm); | |
2034 | kfree(p); | |
2035 | return 0; | |
2036 | } | |
2037 | ||
2038 | static ssize_t debugfs_htab_read(struct file *file, char __user *buf, | |
2039 | size_t len, loff_t *ppos) | |
2040 | { | |
2041 | struct debugfs_htab_state *p = file->private_data; | |
2042 | ssize_t ret, r; | |
2043 | unsigned long i, n; | |
2044 | unsigned long v, hr, gr; | |
2045 | struct kvm *kvm; | |
2046 | __be64 *hptp; | |
2047 | ||
891f1ebf PM |
2048 | kvm = p->kvm; |
2049 | if (kvm_is_radix(kvm)) | |
2050 | return 0; | |
2051 | ||
e23a808b PM |
2052 | ret = mutex_lock_interruptible(&p->mutex); |
2053 | if (ret) | |
2054 | return ret; | |
2055 | ||
2056 | if (p->chars_left) { | |
2057 | n = p->chars_left; | |
2058 | if (n > len) | |
2059 | n = len; | |
2060 | r = copy_to_user(buf, p->buf + p->buf_index, n); | |
2061 | n -= r; | |
2062 | p->chars_left -= n; | |
2063 | p->buf_index += n; | |
2064 | buf += n; | |
2065 | len -= n; | |
2066 | ret = n; | |
2067 | if (r) { | |
2068 | if (!n) | |
2069 | ret = -EFAULT; | |
2070 | goto out; | |
2071 | } | |
2072 | } | |
2073 | ||
e23a808b | 2074 | i = p->hpt_index; |
3f9d4f5a | 2075 | hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE)); |
3d089f84 DG |
2076 | for (; len != 0 && i < kvmppc_hpt_npte(&kvm->arch.hpt); |
2077 | ++i, hptp += 2) { | |
e23a808b PM |
2078 | if (!(be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))) |
2079 | continue; | |
2080 | ||
2081 | /* lock the HPTE so it's stable and read it */ | |
2082 | preempt_disable(); | |
2083 | while (!try_lock_hpte(hptp, HPTE_V_HVLOCK)) | |
2084 | cpu_relax(); | |
2085 | v = be64_to_cpu(hptp[0]) & ~HPTE_V_HVLOCK; | |
2086 | hr = be64_to_cpu(hptp[1]); | |
3f9d4f5a | 2087 | gr = kvm->arch.hpt.rev[i].guest_rpte; |
e23a808b PM |
2088 | unlock_hpte(hptp, v); |
2089 | preempt_enable(); | |
2090 | ||
2091 | if (!(v & (HPTE_V_VALID | HPTE_V_ABSENT))) | |
2092 | continue; | |
2093 | ||
2094 | n = scnprintf(p->buf, sizeof(p->buf), | |
2095 | "%6lx %.16lx %.16lx %.16lx\n", | |
2096 | i, v, hr, gr); | |
2097 | p->chars_left = n; | |
2098 | if (n > len) | |
2099 | n = len; | |
2100 | r = copy_to_user(buf, p->buf, n); | |
2101 | n -= r; | |
2102 | p->chars_left -= n; | |
2103 | p->buf_index = n; | |
2104 | buf += n; | |
2105 | len -= n; | |
2106 | ret += n; | |
2107 | if (r) { | |
2108 | if (!ret) | |
2109 | ret = -EFAULT; | |
2110 | goto out; | |
2111 | } | |
2112 | } | |
2113 | p->hpt_index = i; | |
2114 | ||
2115 | out: | |
2116 | mutex_unlock(&p->mutex); | |
2117 | return ret; | |
2118 | } | |
2119 | ||
025c9511 | 2120 | static ssize_t debugfs_htab_write(struct file *file, const char __user *buf, |
e23a808b PM |
2121 | size_t len, loff_t *ppos) |
2122 | { | |
2123 | return -EACCES; | |
2124 | } | |
2125 | ||
2126 | static const struct file_operations debugfs_htab_fops = { | |
2127 | .owner = THIS_MODULE, | |
2128 | .open = debugfs_htab_open, | |
2129 | .release = debugfs_htab_release, | |
2130 | .read = debugfs_htab_read, | |
2131 | .write = debugfs_htab_write, | |
2132 | .llseek = generic_file_llseek, | |
2133 | }; | |
2134 | ||
2135 | void kvmppc_mmu_debugfs_init(struct kvm *kvm) | |
2136 | { | |
c4fd527f GKH |
2137 | debugfs_create_file("htab", 0400, kvm->arch.debugfs_dir, kvm, |
2138 | &debugfs_htab_fops); | |
e23a808b PM |
2139 | } |
2140 | ||
de56a948 PM |
2141 | void kvmppc_mmu_book3s_hv_init(struct kvm_vcpu *vcpu) |
2142 | { | |
2143 | struct kvmppc_mmu *mmu = &vcpu->arch.mmu; | |
2144 | ||
c17b98cf | 2145 | vcpu->arch.slb_nr = 32; /* POWER7/POWER8 */ |
de56a948 | 2146 | |
18c3640c | 2147 | mmu->xlate = kvmppc_mmu_book3s_64_hv_xlate; |
de56a948 PM |
2148 | |
2149 | vcpu->arch.hflags |= BOOK3S_HFLAG_SLB; | |
2150 | } |