1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
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>
15 #include <linux/vmalloc.h>
16 #include <linux/srcu.h>
17 #include <linux/anon_inodes.h>
18 #include <linux/file.h>
19 #include <linux/debugfs.h>
20
21 #include <asm/kvm_ppc.h>
22 #include <asm/kvm_book3s.h>
23 #include <asm/book3s/64/mmu-hash.h>
24 #include <asm/hvcall.h>
25 #include <asm/synch.h>
26 #include <asm/ppc-opcode.h>
27 #include <asm/cputable.h>
28 #include <asm/pte-walk.h>
29
30 #include "book3s.h"
31 #include "book3s_hv.h"
32 #include "trace_hv.h"
33
34 //#define DEBUG_RESIZE_HPT 1
35
36 #ifdef DEBUG_RESIZE_HPT
37 #define resize_hpt_debug(resize, ...) \
38 do { \
39 printk(KERN_DEBUG "RESIZE HPT %p: ", resize); \
40 printk(__VA_ARGS__); \
41 } while (0)
42 #else
43 #define resize_hpt_debug(resize, ...) \
44 do { } while (0)
45 #endif
46
47 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
48 long pte_index, unsigned long pteh,
49 unsigned long ptel, unsigned long *pte_idx_ret);
50
51 struct kvm_resize_hpt {
52 /* These fields read-only after init */
53 struct kvm *kvm;
54 struct work_struct work;
55 u32 order;
56
57 /* These fields protected by kvm->arch.mmu_setup_lock */
58
59 /* Possible values and their usage:
60 * <0 an error occurred during allocation,
61 * -EBUSY allocation is in the progress,
62 * 0 allocation made successfully.
63 */
64 int error;
65
66 /* Private to the work thread, until error != -EBUSY,
67 * then protected by kvm->arch.mmu_setup_lock.
68 */
69 struct kvm_hpt_info hpt;
70 };
71
72 int kvmppc_allocate_hpt(struct kvm_hpt_info *info, u32 order)
73 {
74 unsigned long hpt = 0;
75 int cma = 0;
76 struct page *page = NULL;
77 struct revmap_entry *rev;
78 unsigned long npte;
79
80 if ((order < PPC_MIN_HPT_ORDER) || (order > PPC_MAX_HPT_ORDER))
81 return -EINVAL;
82
83 page = kvm_alloc_hpt_cma(1ul << (order - PAGE_SHIFT));
84 if (page) {
85 hpt = (unsigned long)pfn_to_kaddr(page_to_pfn(page));
86 memset((void *)hpt, 0, (1ul << order));
87 cma = 1;
88 }
89
90 if (!hpt)
91 hpt = __get_free_pages(GFP_KERNEL|__GFP_ZERO|__GFP_RETRY_MAYFAIL
92 |__GFP_NOWARN, order - PAGE_SHIFT);
93
94 if (!hpt)
95 return -ENOMEM;
96
97 /* HPTEs are 2**4 bytes long */
98 npte = 1ul << (order - 4);
99
100 /* Allocate reverse map array */
101 rev = vmalloc(array_size(npte, sizeof(struct revmap_entry)));
102 if (!rev) {
103 if (cma)
104 kvm_free_hpt_cma(page, 1 << (order - PAGE_SHIFT));
105 else
106 free_pages(hpt, order - PAGE_SHIFT);
107 return -ENOMEM;
108 }
109
110 info->order = order;
111 info->virt = hpt;
112 info->cma = cma;
113 info->rev = rev;
114
115 return 0;
116 }
117
118 void kvmppc_set_hpt(struct kvm *kvm, struct kvm_hpt_info *info)
119 {
120 atomic64_set(&kvm->arch.mmio_update, 0);
121 kvm->arch.hpt = *info;
122 kvm->arch.sdr1 = __pa(info->virt) | (info->order - 18);
123
124 pr_debug("KVM guest htab at %lx (order %ld), LPID %llx\n",
125 info->virt, (long)info->order, kvm->arch.lpid);
126 }
127
128 int kvmppc_alloc_reset_hpt(struct kvm *kvm, int order)
129 {
130 int err = -EBUSY;
131 struct kvm_hpt_info info;
132
133 mutex_lock(&kvm->arch.mmu_setup_lock);
134 if (kvm->arch.mmu_ready) {
135 kvm->arch.mmu_ready = 0;
136 /* order mmu_ready vs. vcpus_running */
137 smp_mb();
138 if (atomic_read(&kvm->arch.vcpus_running)) {
139 kvm->arch.mmu_ready = 1;
140 goto out;
141 }
142 }
143 if (kvm_is_radix(kvm)) {
144 err = kvmppc_switch_mmu_to_hpt(kvm);
145 if (err)
146 goto out;
147 }
148
149 if (kvm->arch.hpt.order == order) {
150 /* We already have a suitable HPT */
151
152 /* Set the entire HPT to 0, i.e. invalid HPTEs */
153 memset((void *)kvm->arch.hpt.virt, 0, 1ul << order);
154 /*
155 * Reset all the reverse-mapping chains for all memslots
156 */
157 kvmppc_rmap_reset(kvm);
158 err = 0;
159 goto out;
160 }
161
162 if (kvm->arch.hpt.virt) {
163 kvmppc_free_hpt(&kvm->arch.hpt);
164 kvmppc_rmap_reset(kvm);
165 }
166
167 err = kvmppc_allocate_hpt(&info, order);
168 if (err < 0)
169 goto out;
170 kvmppc_set_hpt(kvm, &info);
171
172 out:
173 if (err == 0)
174 /* Ensure that each vcpu will flush its TLB on next entry. */
175 cpumask_setall(&kvm->arch.need_tlb_flush);
176
177 mutex_unlock(&kvm->arch.mmu_setup_lock);
178 return err;
179 }
180
181 void kvmppc_free_hpt(struct kvm_hpt_info *info)
182 {
183 vfree(info->rev);
184 info->rev = NULL;
185 if (info->cma)
186 kvm_free_hpt_cma(virt_to_page((void *)info->virt),
187 1 << (info->order - PAGE_SHIFT));
188 else if (info->virt)
189 free_pages(info->virt, info->order - PAGE_SHIFT);
190 info->virt = 0;
191 info->order = 0;
192 }
193
194 /* Bits in first HPTE dword for pagesize 4k, 64k or 16M */
195 static inline unsigned long hpte0_pgsize_encoding(unsigned long pgsize)
196 {
197 return (pgsize > 0x1000) ? HPTE_V_LARGE : 0;
198 }
199
200 /* Bits in second HPTE dword for pagesize 4k, 64k or 16M */
201 static inline unsigned long hpte1_pgsize_encoding(unsigned long pgsize)
202 {
203 return (pgsize == 0x10000) ? 0x1000 : 0;
204 }
205
206 void kvmppc_map_vrma(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
207 unsigned long porder)
208 {
209 unsigned long i;
210 unsigned long npages;
211 unsigned long hp_v, hp_r;
212 unsigned long addr, hash;
213 unsigned long psize;
214 unsigned long hp0, hp1;
215 unsigned long idx_ret;
216 long ret;
217 struct kvm *kvm = vcpu->kvm;
218
219 psize = 1ul << porder;
220 npages = memslot->npages >> (porder - PAGE_SHIFT);
221
222 /* VRMA can't be > 1TB */
223 if (npages > 1ul << (40 - porder))
224 npages = 1ul << (40 - porder);
225 /* Can't use more than 1 HPTE per HPTEG */
226 if (npages > kvmppc_hpt_mask(&kvm->arch.hpt) + 1)
227 npages = kvmppc_hpt_mask(&kvm->arch.hpt) + 1;
228
229 hp0 = HPTE_V_1TB_SEG | (VRMA_VSID << (40 - 16)) |
230 HPTE_V_BOLTED | hpte0_pgsize_encoding(psize);
231 hp1 = hpte1_pgsize_encoding(psize) |
232 HPTE_R_R | HPTE_R_C | HPTE_R_M | PP_RWXX;
233
234 for (i = 0; i < npages; ++i) {
235 addr = i << porder;
236 /* can't use hpt_hash since va > 64 bits */
237 hash = (i ^ (VRMA_VSID ^ (VRMA_VSID << 25)))
238 & kvmppc_hpt_mask(&kvm->arch.hpt);
239 /*
240 * We assume that the hash table is empty and no
241 * vcpus are using it at this stage. Since we create
242 * at most one HPTE per HPTEG, we just assume entry 7
243 * is available and use it.
244 */
245 hash = (hash << 3) + 7;
246 hp_v = hp0 | ((addr >> 16) & ~0x7fUL);
247 hp_r = hp1 | addr;
248 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, hash, hp_v, hp_r,
249 &idx_ret);
250 if (ret != H_SUCCESS) {
251 pr_err("KVM: map_vrma at %lx failed, ret=%ld\n",
252 addr, ret);
253 break;
254 }
255 }
256 }
257
258 int kvmppc_mmu_hv_init(void)
259 {
260 unsigned long nr_lpids;
261
262 if (!mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE))
263 return -EINVAL;
264
265 if (cpu_has_feature(CPU_FTR_HVMODE)) {
266 if (WARN_ON(mfspr(SPRN_LPID) != 0))
267 return -EINVAL;
268 nr_lpids = 1UL << mmu_lpid_bits;
269 } else {
270 nr_lpids = 1UL << KVM_MAX_NESTED_GUESTS_SHIFT;
271 }
272
273 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
274 /* POWER7 has 10-bit LPIDs, POWER8 has 12-bit LPIDs */
275 if (cpu_has_feature(CPU_FTR_ARCH_207S))
276 WARN_ON(nr_lpids != 1UL << 12);
277 else
278 WARN_ON(nr_lpids != 1UL << 10);
279
280 /*
281 * Reserve the last implemented LPID use in partition
282 * switching for POWER7 and POWER8.
283 */
284 nr_lpids -= 1;
285 }
286
287 kvmppc_init_lpid(nr_lpids);
288
289 return 0;
290 }
291
292 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
293 long pte_index, unsigned long pteh,
294 unsigned long ptel, unsigned long *pte_idx_ret)
295 {
296 long ret;
297
298 preempt_disable();
299 ret = kvmppc_do_h_enter(kvm, flags, pte_index, pteh, ptel,
300 kvm->mm->pgd, false, pte_idx_ret);
301 preempt_enable();
302 if (ret == H_TOO_HARD) {
303 /* this can't happen */
304 pr_err("KVM: Oops, kvmppc_h_enter returned too hard!\n");
305 ret = H_RESOURCE; /* or something */
306 }
307 return ret;
308
309 }
310
311 static struct kvmppc_slb *kvmppc_mmu_book3s_hv_find_slbe(struct kvm_vcpu *vcpu,
312 gva_t eaddr)
313 {
314 u64 mask;
315 int i;
316
317 for (i = 0; i < vcpu->arch.slb_nr; i++) {
318 if (!(vcpu->arch.slb[i].orige & SLB_ESID_V))
319 continue;
320
321 if (vcpu->arch.slb[i].origv & SLB_VSID_B_1T)
322 mask = ESID_MASK_1T;
323 else
324 mask = ESID_MASK;
325
326 if (((vcpu->arch.slb[i].orige ^ eaddr) & mask) == 0)
327 return &vcpu->arch.slb[i];
328 }
329 return NULL;
330 }
331
332 static unsigned long kvmppc_mmu_get_real_addr(unsigned long v, unsigned long r,
333 unsigned long ea)
334 {
335 unsigned long ra_mask;
336
337 ra_mask = kvmppc_actual_pgsz(v, r) - 1;
338 return (r & HPTE_R_RPN & ~ra_mask) | (ea & ra_mask);
339 }
340
341 static int kvmppc_mmu_book3s_64_hv_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
342 struct kvmppc_pte *gpte, bool data, bool iswrite)
343 {
344 struct kvm *kvm = vcpu->kvm;
345 struct kvmppc_slb *slbe;
346 unsigned long slb_v;
347 unsigned long pp, key;
348 unsigned long v, orig_v, gr;
349 __be64 *hptep;
350 long int index;
351 int virtmode = __kvmppc_get_msr_hv(vcpu) & (data ? MSR_DR : MSR_IR);
352
353 if (kvm_is_radix(vcpu->kvm))
354 return kvmppc_mmu_radix_xlate(vcpu, eaddr, gpte, data, iswrite);
355
356 /* Get SLB entry */
357 if (virtmode) {
358 slbe = kvmppc_mmu_book3s_hv_find_slbe(vcpu, eaddr);
359 if (!slbe)
360 return -EINVAL;
361 slb_v = slbe->origv;
362 } else {
363 /* real mode access */
364 slb_v = vcpu->kvm->arch.vrma_slb_v;
365 }
366
367 preempt_disable();
368 /* Find the HPTE in the hash table */
369 index = kvmppc_hv_find_lock_hpte(kvm, eaddr, slb_v,
370 HPTE_V_VALID | HPTE_V_ABSENT);
371 if (index < 0) {
372 preempt_enable();
373 return -ENOENT;
374 }
375 hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
376 v = orig_v = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
377 if (cpu_has_feature(CPU_FTR_ARCH_300))
378 v = hpte_new_to_old_v(v, be64_to_cpu(hptep[1]));
379 gr = kvm->arch.hpt.rev[index].guest_rpte;
380
381 unlock_hpte(hptep, orig_v);
382 preempt_enable();
383
384 gpte->eaddr = eaddr;
385 gpte->vpage = ((v & HPTE_V_AVPN) << 4) | ((eaddr >> 12) & 0xfff);
386
387 /* Get PP bits and key for permission check */
388 pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
389 key = (__kvmppc_get_msr_hv(vcpu) & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
390 key &= slb_v;
391
392 /* Calculate permissions */
393 gpte->may_read = hpte_read_permission(pp, key);
394 gpte->may_write = hpte_write_permission(pp, key);
395 gpte->may_execute = gpte->may_read && !(gr & (HPTE_R_N | HPTE_R_G));
396
397 /* Storage key permission check for POWER7 */
398 if (data && virtmode) {
399 int amrfield = hpte_get_skey_perm(gr, vcpu->arch.amr);
400 if (amrfield & 1)
401 gpte->may_read = 0;
402 if (amrfield & 2)
403 gpte->may_write = 0;
404 }
405
406 /* Get the guest physical address */
407 gpte->raddr = kvmppc_mmu_get_real_addr(v, gr, eaddr);
408 return 0;
409 }
410
411 /*
412 * Quick test for whether an instruction is a load or a store.
413 * If the instruction is a load or a store, then this will indicate
414 * which it is, at least on server processors. (Embedded processors
415 * have some external PID instructions that don't follow the rule
416 * embodied here.) If the instruction isn't a load or store, then
417 * this doesn't return anything useful.
418 */
419 static int instruction_is_store(ppc_inst_t instr)
420 {
421 unsigned int mask;
422 unsigned int suffix;
423
424 mask = 0x10000000;
425 suffix = ppc_inst_val(instr);
426 if (ppc_inst_prefixed(instr))
427 suffix = ppc_inst_suffix(instr);
428 else if ((suffix & 0xfc000000) == 0x7c000000)
429 mask = 0x100; /* major opcode 31 */
430 return (suffix & mask) != 0;
431 }
432
433 int kvmppc_hv_emulate_mmio(struct kvm_vcpu *vcpu,
434 unsigned long gpa, gva_t ea, int is_store)
435 {
436 ppc_inst_t last_inst;
437 bool is_prefixed = !!(kvmppc_get_msr(vcpu) & SRR1_PREFIXED);
438
439 /*
440 * Fast path - check if the guest physical address corresponds to a
441 * device on the FAST_MMIO_BUS, if so we can avoid loading the
442 * instruction all together, then we can just handle it and return.
443 */
444 if (is_store) {
445 int idx, ret;
446
447 idx = srcu_read_lock(&vcpu->kvm->srcu);
448 ret = kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, (gpa_t) gpa, 0,
449 NULL);
450 srcu_read_unlock(&vcpu->kvm->srcu, idx);
451 if (!ret) {
452 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + (is_prefixed ? 8 : 4));
453 return RESUME_GUEST;
454 }
455 }
456
457 /*
458 * If we fail, we just return to the guest and try executing it again.
459 */
460 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
461 EMULATE_DONE)
462 return RESUME_GUEST;
463
464 /*
465 * WARNING: We do not know for sure whether the instruction we just
466 * read from memory is the same that caused the fault in the first
467 * place.
468 *
469 * If the fault is prefixed but the instruction is not or vice
470 * versa, try again so that we don't advance pc the wrong amount.
471 */
472 if (ppc_inst_prefixed(last_inst) != is_prefixed)
473 return RESUME_GUEST;
474
475 /*
476 * If the instruction we read is neither an load or a store,
477 * then it can't access memory, so we don't need to worry about
478 * enforcing access permissions. So, assuming it is a load or
479 * store, we just check that its direction (load or store) is
480 * consistent with the original fault, since that's what we
481 * checked the access permissions against. If there is a mismatch
482 * we just return and retry the instruction.
483 */
484
485 if (instruction_is_store(last_inst) != !!is_store)
486 return RESUME_GUEST;
487
488 /*
489 * Emulated accesses are emulated by looking at the hash for
490 * translation once, then performing the access later. The
491 * translation could be invalidated in the meantime in which
492 * point performing the subsequent memory access on the old
493 * physical address could possibly be a security hole for the
494 * guest (but not the host).
495 *
496 * This is less of an issue for MMIO stores since they aren't
497 * globally visible. It could be an issue for MMIO loads to
498 * a certain extent but we'll ignore it for now.
499 */
500
501 vcpu->arch.paddr_accessed = gpa;
502 vcpu->arch.vaddr_accessed = ea;
503 return kvmppc_emulate_mmio(vcpu);
504 }
505
506 int kvmppc_book3s_hv_page_fault(struct kvm_vcpu *vcpu,
507 unsigned long ea, unsigned long dsisr)
508 {
509 struct kvm *kvm = vcpu->kvm;
510 unsigned long hpte[3], r;
511 unsigned long hnow_v, hnow_r;
512 __be64 *hptep;
513 unsigned long mmu_seq, psize, pte_size;
514 unsigned long gpa_base, gfn_base;
515 unsigned long gpa, gfn, hva, pfn, hpa;
516 struct kvm_memory_slot *memslot;
517 unsigned long *rmap;
518 struct revmap_entry *rev;
519 struct page *page;
520 long index, ret;
521 bool is_ci;
522 bool writing, write_ok;
523 unsigned int shift;
524 unsigned long rcbits;
525 long mmio_update;
526 pte_t pte, *ptep;
527
528 if (kvm_is_radix(kvm))
529 return kvmppc_book3s_radix_page_fault(vcpu, ea, dsisr);
530
531 /*
532 * Real-mode code has already searched the HPT and found the
533 * entry we're interested in. Lock the entry and check that
534 * it hasn't changed. If it has, just return and re-execute the
535 * instruction.
536 */
537 if (ea != vcpu->arch.pgfault_addr)
538 return RESUME_GUEST;
539
540 if (vcpu->arch.pgfault_cache) {
541 mmio_update = atomic64_read(&kvm->arch.mmio_update);
542 if (mmio_update == vcpu->arch.pgfault_cache->mmio_update) {
543 r = vcpu->arch.pgfault_cache->rpte;
544 psize = kvmppc_actual_pgsz(vcpu->arch.pgfault_hpte[0],
545 r);
546 gpa_base = r & HPTE_R_RPN & ~(psize - 1);
547 gfn_base = gpa_base >> PAGE_SHIFT;
548 gpa = gpa_base | (ea & (psize - 1));
549 return kvmppc_hv_emulate_mmio(vcpu, gpa, ea,
550 dsisr & DSISR_ISSTORE);
551 }
552 }
553 index = vcpu->arch.pgfault_index;
554 hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
555 rev = &kvm->arch.hpt.rev[index];
556 preempt_disable();
557 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
558 cpu_relax();
559 hpte[0] = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
560 hpte[1] = be64_to_cpu(hptep[1]);
561 hpte[2] = r = rev->guest_rpte;
562 unlock_hpte(hptep, hpte[0]);
563 preempt_enable();
564
565 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
566 hpte[0] = hpte_new_to_old_v(hpte[0], hpte[1]);
567 hpte[1] = hpte_new_to_old_r(hpte[1]);
568 }
569 if (hpte[0] != vcpu->arch.pgfault_hpte[0] ||
570 hpte[1] != vcpu->arch.pgfault_hpte[1])
571 return RESUME_GUEST;
572
573 /* Translate the logical address and get the page */
574 psize = kvmppc_actual_pgsz(hpte[0], r);
575 gpa_base = r & HPTE_R_RPN & ~(psize - 1);
576 gfn_base = gpa_base >> PAGE_SHIFT;
577 gpa = gpa_base | (ea & (psize - 1));
578 gfn = gpa >> PAGE_SHIFT;
579 memslot = gfn_to_memslot(kvm, gfn);
580
581 trace_kvm_page_fault_enter(vcpu, hpte, memslot, ea, dsisr);
582
583 /* No memslot means it's an emulated MMIO region */
584 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
585 return kvmppc_hv_emulate_mmio(vcpu, gpa, ea,
586 dsisr & DSISR_ISSTORE);
587
588 /*
589 * This should never happen, because of the slot_is_aligned()
590 * check in kvmppc_do_h_enter().
591 */
592 if (gfn_base < memslot->base_gfn)
593 return -EFAULT;
594
595 /* used to check for invalidations in progress */
596 mmu_seq = kvm->mmu_invalidate_seq;
597 smp_rmb();
598
599 ret = -EFAULT;
600 page = NULL;
601 writing = (dsisr & DSISR_ISSTORE) != 0;
602 /* If writing != 0, then the HPTE must allow writing, if we get here */
603 write_ok = writing;
604 hva = gfn_to_hva_memslot(memslot, gfn);
605
606 /*
607 * Do a fast check first, since __gfn_to_pfn_memslot doesn't
608 * do it with !atomic && !async, which is how we call it.
609 * We always ask for write permission since the common case
610 * is that the page is writable.
611 */
612 if (get_user_page_fast_only(hva, FOLL_WRITE, &page)) {
613 write_ok = true;
614 } else {
615 /* Call KVM generic code to do the slow-path check */
616 pfn = __gfn_to_pfn_memslot(memslot, gfn, false, false, NULL,
617 writing, &write_ok, NULL);
618 if (is_error_noslot_pfn(pfn))
619 return -EFAULT;
620 page = NULL;
621 if (pfn_valid(pfn)) {
622 page = pfn_to_page(pfn);
623 if (PageReserved(page))
624 page = NULL;
625 }
626 }
627
628 /*
629 * Read the PTE from the process' radix tree and use that
630 * so we get the shift and attribute bits.
631 */
632 spin_lock(&kvm->mmu_lock);
633 ptep = find_kvm_host_pte(kvm, mmu_seq, hva, &shift);
634 pte = __pte(0);
635 if (ptep)
636 pte = READ_ONCE(*ptep);
637 spin_unlock(&kvm->mmu_lock);
638 /*
639 * If the PTE disappeared temporarily due to a THP
640 * collapse, just return and let the guest try again.
641 */
642 if (!pte_present(pte)) {
643 if (page)
644 put_page(page);
645 return RESUME_GUEST;
646 }
647 hpa = pte_pfn(pte) << PAGE_SHIFT;
648 pte_size = PAGE_SIZE;
649 if (shift)
650 pte_size = 1ul << shift;
651 is_ci = pte_ci(pte);
652
653 if (psize > pte_size)
654 goto out_put;
655 if (pte_size > psize)
656 hpa |= hva & (pte_size - psize);
657
658 /* Check WIMG vs. the actual page we're accessing */
659 if (!hpte_cache_flags_ok(r, is_ci)) {
660 if (is_ci)
661 goto out_put;
662 /*
663 * Allow guest to map emulated device memory as
664 * uncacheable, but actually make it cacheable.
665 */
666 r = (r & ~(HPTE_R_W|HPTE_R_I|HPTE_R_G)) | HPTE_R_M;
667 }
668
669 /*
670 * Set the HPTE to point to hpa.
671 * Since the hpa is at PAGE_SIZE granularity, make sure we
672 * don't mask out lower-order bits if psize < PAGE_SIZE.
673 */
674 if (psize < PAGE_SIZE)
675 psize = PAGE_SIZE;
676 r = (r & HPTE_R_KEY_HI) | (r & ~(HPTE_R_PP0 - psize)) | hpa;
677 if (hpte_is_writable(r) && !write_ok)
678 r = hpte_make_readonly(r);
679 ret = RESUME_GUEST;
680 preempt_disable();
681 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
682 cpu_relax();
683 hnow_v = be64_to_cpu(hptep[0]);
684 hnow_r = be64_to_cpu(hptep[1]);
685 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
686 hnow_v = hpte_new_to_old_v(hnow_v, hnow_r);
687 hnow_r = hpte_new_to_old_r(hnow_r);
688 }
689
690 /*
691 * If the HPT is being resized, don't update the HPTE,
692 * instead let the guest retry after the resize operation is complete.
693 * The synchronization for mmu_ready test vs. set is provided
694 * by the HPTE lock.
695 */
696 if (!kvm->arch.mmu_ready)
697 goto out_unlock;
698
699 if ((hnow_v & ~HPTE_V_HVLOCK) != hpte[0] || hnow_r != hpte[1] ||
700 rev->guest_rpte != hpte[2])
701 /* HPTE has been changed under us; let the guest retry */
702 goto out_unlock;
703 hpte[0] = (hpte[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID;
704
705 /* Always put the HPTE in the rmap chain for the page base address */
706 rmap = &memslot->arch.rmap[gfn_base - memslot->base_gfn];
707 lock_rmap(rmap);
708
709 /* Check if we might have been invalidated; let the guest retry if so */
710 ret = RESUME_GUEST;
711 if (mmu_invalidate_retry(vcpu->kvm, mmu_seq)) {
712 unlock_rmap(rmap);
713 goto out_unlock;
714 }
715
716 /* Only set R/C in real HPTE if set in both *rmap and guest_rpte */
717 rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
718 r &= rcbits | ~(HPTE_R_R | HPTE_R_C);
719
720 if (be64_to_cpu(hptep[0]) & HPTE_V_VALID) {
721 /* HPTE was previously valid, so we need to invalidate it */
722 unlock_rmap(rmap);
723 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
724 kvmppc_invalidate_hpte(kvm, hptep, index);
725 /* don't lose previous R and C bits */
726 r |= be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
727 } else {
728 kvmppc_add_revmap_chain(kvm, rev, rmap, index, 0);
729 }
730
731 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
732 r = hpte_old_to_new_r(hpte[0], r);
733 hpte[0] = hpte_old_to_new_v(hpte[0]);
734 }
735 hptep[1] = cpu_to_be64(r);
736 eieio();
737 __unlock_hpte(hptep, hpte[0]);
738 asm volatile("ptesync" : : : "memory");
739 preempt_enable();
740 if (page && hpte_is_writable(r))
741 set_page_dirty_lock(page);
742
743 out_put:
744 trace_kvm_page_fault_exit(vcpu, hpte, ret);
745
746 if (page)
747 put_page(page);
748 return ret;
749
750 out_unlock:
751 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
752 preempt_enable();
753 goto out_put;
754 }
755
756 void kvmppc_rmap_reset(struct kvm *kvm)
757 {
758 struct kvm_memslots *slots;
759 struct kvm_memory_slot *memslot;
760 int srcu_idx, bkt;
761
762 srcu_idx = srcu_read_lock(&kvm->srcu);
763 slots = kvm_memslots(kvm);
764 kvm_for_each_memslot(memslot, bkt, slots) {
765 /* Mutual exclusion with kvm_unmap_hva_range etc. */
766 spin_lock(&kvm->mmu_lock);
767 /*
768 * This assumes it is acceptable to lose reference and
769 * change bits across a reset.
770 */
771 memset(memslot->arch.rmap, 0,
772 memslot->npages * sizeof(*memslot->arch.rmap));
773 spin_unlock(&kvm->mmu_lock);
774 }
775 srcu_read_unlock(&kvm->srcu, srcu_idx);
776 }
777
778 /* Must be called with both HPTE and rmap locked */
779 static void kvmppc_unmap_hpte(struct kvm *kvm, unsigned long i,
780 struct kvm_memory_slot *memslot,
781 unsigned long *rmapp, unsigned long gfn)
782 {
783 __be64 *hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
784 struct revmap_entry *rev = kvm->arch.hpt.rev;
785 unsigned long j, h;
786 unsigned long ptel, psize, rcbits;
787
788 j = rev[i].forw;
789 if (j == i) {
790 /* chain is now empty */
791 *rmapp &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
792 } else {
793 /* remove i from chain */
794 h = rev[i].back;
795 rev[h].forw = j;
796 rev[j].back = h;
797 rev[i].forw = rev[i].back = i;
798 *rmapp = (*rmapp & ~KVMPPC_RMAP_INDEX) | j;
799 }
800
801 /* Now check and modify the HPTE */
802 ptel = rev[i].guest_rpte;
803 psize = kvmppc_actual_pgsz(be64_to_cpu(hptep[0]), ptel);
804 if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
805 hpte_rpn(ptel, psize) == gfn) {
806 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
807 kvmppc_invalidate_hpte(kvm, hptep, i);
808 hptep[1] &= ~cpu_to_be64(HPTE_R_KEY_HI | HPTE_R_KEY_LO);
809 /* Harvest R and C */
810 rcbits = be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
811 *rmapp |= rcbits << KVMPPC_RMAP_RC_SHIFT;
812 if ((rcbits & HPTE_R_C) && memslot->dirty_bitmap)
813 kvmppc_update_dirty_map(memslot, gfn, psize);
814 if (rcbits & ~rev[i].guest_rpte) {
815 rev[i].guest_rpte = ptel | rcbits;
816 note_hpte_modification(kvm, &rev[i]);
817 }
818 }
819 }
820
821 static void kvm_unmap_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
822 unsigned long gfn)
823 {
824 unsigned long i;
825 __be64 *hptep;
826 unsigned long *rmapp;
827
828 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
829 for (;;) {
830 lock_rmap(rmapp);
831 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
832 unlock_rmap(rmapp);
833 break;
834 }
835
836 /*
837 * To avoid an ABBA deadlock with the HPTE lock bit,
838 * we can't spin on the HPTE lock while holding the
839 * rmap chain lock.
840 */
841 i = *rmapp & KVMPPC_RMAP_INDEX;
842 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
843 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
844 /* unlock rmap before spinning on the HPTE lock */
845 unlock_rmap(rmapp);
846 while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
847 cpu_relax();
848 continue;
849 }
850
851 kvmppc_unmap_hpte(kvm, i, memslot, rmapp, gfn);
852 unlock_rmap(rmapp);
853 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
854 }
855 }
856
857 bool kvm_unmap_gfn_range_hv(struct kvm *kvm, struct kvm_gfn_range *range)
858 {
859 gfn_t gfn;
860
861 if (kvm_is_radix(kvm)) {
862 for (gfn = range->start; gfn < range->end; gfn++)
863 kvm_unmap_radix(kvm, range->slot, gfn);
864 } else {
865 for (gfn = range->start; gfn < range->end; gfn++)
866 kvm_unmap_rmapp(kvm, range->slot, gfn);
867 }
868
869 return false;
870 }
871
872 void kvmppc_core_flush_memslot_hv(struct kvm *kvm,
873 struct kvm_memory_slot *memslot)
874 {
875 unsigned long gfn;
876 unsigned long n;
877 unsigned long *rmapp;
878
879 gfn = memslot->base_gfn;
880 rmapp = memslot->arch.rmap;
881 if (kvm_is_radix(kvm)) {
882 kvmppc_radix_flush_memslot(kvm, memslot);
883 return;
884 }
885
886 for (n = memslot->npages; n; --n, ++gfn) {
887 /*
888 * Testing the present bit without locking is OK because
889 * the memslot has been marked invalid already, and hence
890 * no new HPTEs referencing this page can be created,
891 * thus the present bit can't go from 0 to 1.
892 */
893 if (*rmapp & KVMPPC_RMAP_PRESENT)
894 kvm_unmap_rmapp(kvm, memslot, gfn);
895 ++rmapp;
896 }
897 }
898
899 static bool kvm_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
900 unsigned long gfn)
901 {
902 struct revmap_entry *rev = kvm->arch.hpt.rev;
903 unsigned long head, i, j;
904 __be64 *hptep;
905 bool ret = false;
906 unsigned long *rmapp;
907
908 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
909 retry:
910 lock_rmap(rmapp);
911 if (*rmapp & KVMPPC_RMAP_REFERENCED) {
912 *rmapp &= ~KVMPPC_RMAP_REFERENCED;
913 ret = true;
914 }
915 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
916 unlock_rmap(rmapp);
917 return ret;
918 }
919
920 i = head = *rmapp & KVMPPC_RMAP_INDEX;
921 do {
922 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
923 j = rev[i].forw;
924
925 /* If this HPTE isn't referenced, ignore it */
926 if (!(be64_to_cpu(hptep[1]) & HPTE_R_R))
927 continue;
928
929 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
930 /* unlock rmap before spinning on the HPTE lock */
931 unlock_rmap(rmapp);
932 while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
933 cpu_relax();
934 goto retry;
935 }
936
937 /* Now check and modify the HPTE */
938 if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
939 (be64_to_cpu(hptep[1]) & HPTE_R_R)) {
940 kvmppc_clear_ref_hpte(kvm, hptep, i);
941 if (!(rev[i].guest_rpte & HPTE_R_R)) {
942 rev[i].guest_rpte |= HPTE_R_R;
943 note_hpte_modification(kvm, &rev[i]);
944 }
945 ret = true;
946 }
947 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
948 } while ((i = j) != head);
949
950 unlock_rmap(rmapp);
951 return ret;
952 }
953
954 bool kvm_age_gfn_hv(struct kvm *kvm, struct kvm_gfn_range *range)
955 {
956 gfn_t gfn;
957 bool ret = false;
958
959 if (kvm_is_radix(kvm)) {
960 for (gfn = range->start; gfn < range->end; gfn++)
961 ret |= kvm_age_radix(kvm, range->slot, gfn);
962 } else {
963 for (gfn = range->start; gfn < range->end; gfn++)
964 ret |= kvm_age_rmapp(kvm, range->slot, gfn);
965 }
966
967 return ret;
968 }
969
970 static bool kvm_test_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
971 unsigned long gfn)
972 {
973 struct revmap_entry *rev = kvm->arch.hpt.rev;
974 unsigned long head, i, j;
975 unsigned long *hp;
976 bool ret = true;
977 unsigned long *rmapp;
978
979 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
980 if (*rmapp & KVMPPC_RMAP_REFERENCED)
981 return true;
982
983 lock_rmap(rmapp);
984 if (*rmapp & KVMPPC_RMAP_REFERENCED)
985 goto out;
986
987 if (*rmapp & KVMPPC_RMAP_PRESENT) {
988 i = head = *rmapp & KVMPPC_RMAP_INDEX;
989 do {
990 hp = (unsigned long *)(kvm->arch.hpt.virt + (i << 4));
991 j = rev[i].forw;
992 if (be64_to_cpu(hp[1]) & HPTE_R_R)
993 goto out;
994 } while ((i = j) != head);
995 }
996 ret = false;
997
998 out:
999 unlock_rmap(rmapp);
1000 return ret;
1001 }
1002
1003 bool kvm_test_age_gfn_hv(struct kvm *kvm, struct kvm_gfn_range *range)
1004 {
1005 WARN_ON(range->start + 1 != range->end);
1006
1007 if (kvm_is_radix(kvm))
1008 return kvm_test_age_radix(kvm, range->slot, range->start);
1009 else
1010 return kvm_test_age_rmapp(kvm, range->slot, range->start);
1011 }
1012
1013 static int vcpus_running(struct kvm *kvm)
1014 {
1015 return atomic_read(&kvm->arch.vcpus_running) != 0;
1016 }
1017
1018 /*
1019 * Returns the number of system pages that are dirty.
1020 * This can be more than 1 if we find a huge-page HPTE.
1021 */
1022 static int kvm_test_clear_dirty_npages(struct kvm *kvm, unsigned long *rmapp)
1023 {
1024 struct revmap_entry *rev = kvm->arch.hpt.rev;
1025 unsigned long head, i, j;
1026 unsigned long n;
1027 unsigned long v, r;
1028 __be64 *hptep;
1029 int npages_dirty = 0;
1030
1031 retry:
1032 lock_rmap(rmapp);
1033 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
1034 unlock_rmap(rmapp);
1035 return npages_dirty;
1036 }
1037
1038 i = head = *rmapp & KVMPPC_RMAP_INDEX;
1039 do {
1040 unsigned long hptep1;
1041 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
1042 j = rev[i].forw;
1043
1044 /*
1045 * Checking the C (changed) bit here is racy since there
1046 * is no guarantee about when the hardware writes it back.
1047 * If the HPTE is not writable then it is stable since the
1048 * page can't be written to, and we would have done a tlbie
1049 * (which forces the hardware to complete any writeback)
1050 * when making the HPTE read-only.
1051 * If vcpus are running then this call is racy anyway
1052 * since the page could get dirtied subsequently, so we
1053 * expect there to be a further call which would pick up
1054 * any delayed C bit writeback.
1055 * Otherwise we need to do the tlbie even if C==0 in
1056 * order to pick up any delayed writeback of C.
1057 */
1058 hptep1 = be64_to_cpu(hptep[1]);
1059 if (!(hptep1 & HPTE_R_C) &&
1060 (!hpte_is_writable(hptep1) || vcpus_running(kvm)))
1061 continue;
1062
1063 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
1064 /* unlock rmap before spinning on the HPTE lock */
1065 unlock_rmap(rmapp);
1066 while (hptep[0] & cpu_to_be64(HPTE_V_HVLOCK))
1067 cpu_relax();
1068 goto retry;
1069 }
1070
1071 /* Now check and modify the HPTE */
1072 if (!(hptep[0] & cpu_to_be64(HPTE_V_VALID))) {
1073 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
1074 continue;
1075 }
1076
1077 /* need to make it temporarily absent so C is stable */
1078 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
1079 kvmppc_invalidate_hpte(kvm, hptep, i);
1080 v = be64_to_cpu(hptep[0]);
1081 r = be64_to_cpu(hptep[1]);
1082 if (r & HPTE_R_C) {
1083 hptep[1] = cpu_to_be64(r & ~HPTE_R_C);
1084 if (!(rev[i].guest_rpte & HPTE_R_C)) {
1085 rev[i].guest_rpte |= HPTE_R_C;
1086 note_hpte_modification(kvm, &rev[i]);
1087 }
1088 n = kvmppc_actual_pgsz(v, r);
1089 n = (n + PAGE_SIZE - 1) >> PAGE_SHIFT;
1090 if (n > npages_dirty)
1091 npages_dirty = n;
1092 eieio();
1093 }
1094 v &= ~HPTE_V_ABSENT;
1095 v |= HPTE_V_VALID;
1096 __unlock_hpte(hptep, v);
1097 } while ((i = j) != head);
1098
1099 unlock_rmap(rmapp);
1100 return npages_dirty;
1101 }
1102
1103 void kvmppc_harvest_vpa_dirty(struct kvmppc_vpa *vpa,
1104 struct kvm_memory_slot *memslot,
1105 unsigned long *map)
1106 {
1107 unsigned long gfn;
1108
1109 if (!vpa->dirty || !vpa->pinned_addr)
1110 return;
1111 gfn = vpa->gpa >> PAGE_SHIFT;
1112 if (gfn < memslot->base_gfn ||
1113 gfn >= memslot->base_gfn + memslot->npages)
1114 return;
1115
1116 vpa->dirty = false;
1117 if (map)
1118 __set_bit_le(gfn - memslot->base_gfn, map);
1119 }
1120
1121 long kvmppc_hv_get_dirty_log_hpt(struct kvm *kvm,
1122 struct kvm_memory_slot *memslot, unsigned long *map)
1123 {
1124 unsigned long i;
1125 unsigned long *rmapp;
1126
1127 preempt_disable();
1128 rmapp = memslot->arch.rmap;
1129 for (i = 0; i < memslot->npages; ++i) {
1130 int npages = kvm_test_clear_dirty_npages(kvm, rmapp);
1131 /*
1132 * Note that if npages > 0 then i must be a multiple of npages,
1133 * since we always put huge-page HPTEs in the rmap chain
1134 * corresponding to their page base address.
1135 */
1136 if (npages)
1137 set_dirty_bits(map, i, npages);
1138 ++rmapp;
1139 }
1140 preempt_enable();
1141 return 0;
1142 }
1143
1144 void *kvmppc_pin_guest_page(struct kvm *kvm, unsigned long gpa,
1145 unsigned long *nb_ret)
1146 {
1147 struct kvm_memory_slot *memslot;
1148 unsigned long gfn = gpa >> PAGE_SHIFT;
1149 struct page *page, *pages[1];
1150 int npages;
1151 unsigned long hva, offset;
1152 int srcu_idx;
1153
1154 srcu_idx = srcu_read_lock(&kvm->srcu);
1155 memslot = gfn_to_memslot(kvm, gfn);
1156 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1157 goto err;
1158 hva = gfn_to_hva_memslot(memslot, gfn);
1159 npages = get_user_pages_fast(hva, 1, FOLL_WRITE, pages);
1160 if (npages < 1)
1161 goto err;
1162 page = pages[0];
1163 srcu_read_unlock(&kvm->srcu, srcu_idx);
1164
1165 offset = gpa & (PAGE_SIZE - 1);
1166 if (nb_ret)
1167 *nb_ret = PAGE_SIZE - offset;
1168 return page_address(page) + offset;
1169
1170 err:
1171 srcu_read_unlock(&kvm->srcu, srcu_idx);
1172 return NULL;
1173 }
1174
1175 void kvmppc_unpin_guest_page(struct kvm *kvm, void *va, unsigned long gpa,
1176 bool dirty)
1177 {
1178 struct page *page = virt_to_page(va);
1179 struct kvm_memory_slot *memslot;
1180 unsigned long gfn;
1181 int srcu_idx;
1182
1183 put_page(page);
1184
1185 if (!dirty)
1186 return;
1187
1188 /* We need to mark this page dirty in the memslot dirty_bitmap, if any */
1189 gfn = gpa >> PAGE_SHIFT;
1190 srcu_idx = srcu_read_lock(&kvm->srcu);
1191 memslot = gfn_to_memslot(kvm, gfn);
1192 if (memslot && memslot->dirty_bitmap)
1193 set_bit_le(gfn - memslot->base_gfn, memslot->dirty_bitmap);
1194 srcu_read_unlock(&kvm->srcu, srcu_idx);
1195 }
1196
1197 /*
1198 * HPT resizing
1199 */
1200 static int resize_hpt_allocate(struct kvm_resize_hpt *resize)
1201 {
1202 int rc;
1203
1204 rc = kvmppc_allocate_hpt(&resize->hpt, resize->order);
1205 if (rc < 0)
1206 return rc;
1207
1208 resize_hpt_debug(resize, "%s(): HPT @ 0x%lx\n", __func__,
1209 resize->hpt.virt);
1210
1211 return 0;
1212 }
1213
1214 static unsigned long resize_hpt_rehash_hpte(struct kvm_resize_hpt *resize,
1215 unsigned long idx)
1216 {
1217 struct kvm *kvm = resize->kvm;
1218 struct kvm_hpt_info *old = &kvm->arch.hpt;
1219 struct kvm_hpt_info *new = &resize->hpt;
1220 unsigned long old_hash_mask = (1ULL << (old->order - 7)) - 1;
1221 unsigned long new_hash_mask = (1ULL << (new->order - 7)) - 1;
1222 __be64 *hptep, *new_hptep;
1223 unsigned long vpte, rpte, guest_rpte;
1224 int ret;
1225 struct revmap_entry *rev;
1226 unsigned long apsize, avpn, pteg, hash;
1227 unsigned long new_idx, new_pteg, replace_vpte;
1228 int pshift;
1229
1230 hptep = (__be64 *)(old->virt + (idx << 4));
1231
1232 /* Guest is stopped, so new HPTEs can't be added or faulted
1233 * in, only unmapped or altered by host actions. So, it's
1234 * safe to check this before we take the HPTE lock */
1235 vpte = be64_to_cpu(hptep[0]);
1236 if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1237 return 0; /* nothing to do */
1238
1239 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
1240 cpu_relax();
1241
1242 vpte = be64_to_cpu(hptep[0]);
1243
1244 ret = 0;
1245 if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1246 /* Nothing to do */
1247 goto out;
1248
1249 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1250 rpte = be64_to_cpu(hptep[1]);
1251 vpte = hpte_new_to_old_v(vpte, rpte);
1252 }
1253
1254 /* Unmap */
1255 rev = &old->rev[idx];
1256 guest_rpte = rev->guest_rpte;
1257
1258 ret = -EIO;
1259 apsize = kvmppc_actual_pgsz(vpte, guest_rpte);
1260 if (!apsize)
1261 goto out;
1262
1263 if (vpte & HPTE_V_VALID) {
1264 unsigned long gfn = hpte_rpn(guest_rpte, apsize);
1265 int srcu_idx = srcu_read_lock(&kvm->srcu);
1266 struct kvm_memory_slot *memslot =
1267 __gfn_to_memslot(kvm_memslots(kvm), gfn);
1268
1269 if (memslot) {
1270 unsigned long *rmapp;
1271 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1272
1273 lock_rmap(rmapp);
1274 kvmppc_unmap_hpte(kvm, idx, memslot, rmapp, gfn);
1275 unlock_rmap(rmapp);
1276 }
1277
1278 srcu_read_unlock(&kvm->srcu, srcu_idx);
1279 }
1280
1281 /* Reload PTE after unmap */
1282 vpte = be64_to_cpu(hptep[0]);
1283 BUG_ON(vpte & HPTE_V_VALID);
1284 BUG_ON(!(vpte & HPTE_V_ABSENT));
1285
1286 ret = 0;
1287 if (!(vpte & HPTE_V_BOLTED))
1288 goto out;
1289
1290 rpte = be64_to_cpu(hptep[1]);
1291
1292 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1293 vpte = hpte_new_to_old_v(vpte, rpte);
1294 rpte = hpte_new_to_old_r(rpte);
1295 }
1296
1297 pshift = kvmppc_hpte_base_page_shift(vpte, rpte);
1298 avpn = HPTE_V_AVPN_VAL(vpte) & ~(((1ul << pshift) - 1) >> 23);
1299 pteg = idx / HPTES_PER_GROUP;
1300 if (vpte & HPTE_V_SECONDARY)
1301 pteg = ~pteg;
1302
1303 if (!(vpte & HPTE_V_1TB_SEG)) {
1304 unsigned long offset, vsid;
1305
1306 /* We only have 28 - 23 bits of offset in avpn */
1307 offset = (avpn & 0x1f) << 23;
1308 vsid = avpn >> 5;
1309 /* We can find more bits from the pteg value */
1310 if (pshift < 23)
1311 offset |= ((vsid ^ pteg) & old_hash_mask) << pshift;
1312
1313 hash = vsid ^ (offset >> pshift);
1314 } else {
1315 unsigned long offset, vsid;
1316
1317 /* We only have 40 - 23 bits of seg_off in avpn */
1318 offset = (avpn & 0x1ffff) << 23;
1319 vsid = avpn >> 17;
1320 if (pshift < 23)
1321 offset |= ((vsid ^ (vsid << 25) ^ pteg) & old_hash_mask) << pshift;
1322
1323 hash = vsid ^ (vsid << 25) ^ (offset >> pshift);
1324 }
1325
1326 new_pteg = hash & new_hash_mask;
1327 if (vpte & HPTE_V_SECONDARY)
1328 new_pteg = ~hash & new_hash_mask;
1329
1330 new_idx = new_pteg * HPTES_PER_GROUP + (idx % HPTES_PER_GROUP);
1331 new_hptep = (__be64 *)(new->virt + (new_idx << 4));
1332
1333 replace_vpte = be64_to_cpu(new_hptep[0]);
1334 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1335 unsigned long replace_rpte = be64_to_cpu(new_hptep[1]);
1336 replace_vpte = hpte_new_to_old_v(replace_vpte, replace_rpte);
1337 }
1338
1339 if (replace_vpte & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1340 BUG_ON(new->order >= old->order);
1341
1342 if (replace_vpte & HPTE_V_BOLTED) {
1343 if (vpte & HPTE_V_BOLTED)
1344 /* Bolted collision, nothing we can do */
1345 ret = -ENOSPC;
1346 /* Discard the new HPTE */
1347 goto out;
1348 }
1349
1350 /* Discard the previous HPTE */
1351 }
1352
1353 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1354 rpte = hpte_old_to_new_r(vpte, rpte);
1355 vpte = hpte_old_to_new_v(vpte);
1356 }
1357
1358 new_hptep[1] = cpu_to_be64(rpte);
1359 new->rev[new_idx].guest_rpte = guest_rpte;
1360 /* No need for a barrier, since new HPT isn't active */
1361 new_hptep[0] = cpu_to_be64(vpte);
1362 unlock_hpte(new_hptep, vpte);
1363
1364 out:
1365 unlock_hpte(hptep, vpte);
1366 return ret;
1367 }
1368
1369 static int resize_hpt_rehash(struct kvm_resize_hpt *resize)
1370 {
1371 struct kvm *kvm = resize->kvm;
1372 unsigned long i;
1373 int rc;
1374
1375 for (i = 0; i < kvmppc_hpt_npte(&kvm->arch.hpt); i++) {
1376 rc = resize_hpt_rehash_hpte(resize, i);
1377 if (rc != 0)
1378 return rc;
1379 }
1380
1381 return 0;
1382 }
1383
1384 static void resize_hpt_pivot(struct kvm_resize_hpt *resize)
1385 {
1386 struct kvm *kvm = resize->kvm;
1387 struct kvm_hpt_info hpt_tmp;
1388
1389 /* Exchange the pending tables in the resize structure with
1390 * the active tables */
1391
1392 resize_hpt_debug(resize, "resize_hpt_pivot()\n");
1393
1394 spin_lock(&kvm->mmu_lock);
1395 asm volatile("ptesync" : : : "memory");
1396
1397 hpt_tmp = kvm->arch.hpt;
1398 kvmppc_set_hpt(kvm, &resize->hpt);
1399 resize->hpt = hpt_tmp;
1400
1401 spin_unlock(&kvm->mmu_lock);
1402
1403 synchronize_srcu_expedited(&kvm->srcu);
1404
1405 if (cpu_has_feature(CPU_FTR_ARCH_300))
1406 kvmppc_setup_partition_table(kvm);
1407
1408 resize_hpt_debug(resize, "resize_hpt_pivot() done\n");
1409 }
1410
1411 static void resize_hpt_release(struct kvm *kvm, struct kvm_resize_hpt *resize)
1412 {
1413 if (WARN_ON(!mutex_is_locked(&kvm->arch.mmu_setup_lock)))
1414 return;
1415
1416 if (!resize)
1417 return;
1418
1419 if (resize->error != -EBUSY) {
1420 if (resize->hpt.virt)
1421 kvmppc_free_hpt(&resize->hpt);
1422 kfree(resize);
1423 }
1424
1425 if (kvm->arch.resize_hpt == resize)
1426 kvm->arch.resize_hpt = NULL;
1427 }
1428
1429 static void resize_hpt_prepare_work(struct work_struct *work)
1430 {
1431 struct kvm_resize_hpt *resize = container_of(work,
1432 struct kvm_resize_hpt,
1433 work);
1434 struct kvm *kvm = resize->kvm;
1435 int err = 0;
1436
1437 if (WARN_ON(resize->error != -EBUSY))
1438 return;
1439
1440 mutex_lock(&kvm->arch.mmu_setup_lock);
1441
1442 /* Request is still current? */
1443 if (kvm->arch.resize_hpt == resize) {
1444 /* We may request large allocations here:
1445 * do not sleep with kvm->arch.mmu_setup_lock held for a while.
1446 */
1447 mutex_unlock(&kvm->arch.mmu_setup_lock);
1448
1449 resize_hpt_debug(resize, "%s(): order = %d\n", __func__,
1450 resize->order);
1451
1452 err = resize_hpt_allocate(resize);
1453
1454 /* We have strict assumption about -EBUSY
1455 * when preparing for HPT resize.
1456 */
1457 if (WARN_ON(err == -EBUSY))
1458 err = -EINPROGRESS;
1459
1460 mutex_lock(&kvm->arch.mmu_setup_lock);
1461 /* It is possible that kvm->arch.resize_hpt != resize
1462 * after we grab kvm->arch.mmu_setup_lock again.
1463 */
1464 }
1465
1466 resize->error = err;
1467
1468 if (kvm->arch.resize_hpt != resize)
1469 resize_hpt_release(kvm, resize);
1470
1471 mutex_unlock(&kvm->arch.mmu_setup_lock);
1472 }
1473
1474 int kvm_vm_ioctl_resize_hpt_prepare(struct kvm *kvm,
1475 struct kvm_ppc_resize_hpt *rhpt)
1476 {
1477 unsigned long flags = rhpt->flags;
1478 unsigned long shift = rhpt->shift;
1479 struct kvm_resize_hpt *resize;
1480 int ret;
1481
1482 if (flags != 0 || kvm_is_radix(kvm))
1483 return -EINVAL;
1484
1485 if (shift && ((shift < 18) || (shift > 46)))
1486 return -EINVAL;
1487
1488 mutex_lock(&kvm->arch.mmu_setup_lock);
1489
1490 resize = kvm->arch.resize_hpt;
1491
1492 if (resize) {
1493 if (resize->order == shift) {
1494 /* Suitable resize in progress? */
1495 ret = resize->error;
1496 if (ret == -EBUSY)
1497 ret = 100; /* estimated time in ms */
1498 else if (ret)
1499 resize_hpt_release(kvm, resize);
1500
1501 goto out;
1502 }
1503
1504 /* not suitable, cancel it */
1505 resize_hpt_release(kvm, resize);
1506 }
1507
1508 ret = 0;
1509 if (!shift)
1510 goto out; /* nothing to do */
1511
1512 /* start new resize */
1513
1514 resize = kzalloc(sizeof(*resize), GFP_KERNEL);
1515 if (!resize) {
1516 ret = -ENOMEM;
1517 goto out;
1518 }
1519
1520 resize->error = -EBUSY;
1521 resize->order = shift;
1522 resize->kvm = kvm;
1523 INIT_WORK(&resize->work, resize_hpt_prepare_work);
1524 kvm->arch.resize_hpt = resize;
1525
1526 schedule_work(&resize->work);
1527
1528 ret = 100; /* estimated time in ms */
1529
1530 out:
1531 mutex_unlock(&kvm->arch.mmu_setup_lock);
1532 return ret;
1533 }
1534
1535 static void resize_hpt_boot_vcpu(void *opaque)
1536 {
1537 /* Nothing to do, just force a KVM exit */
1538 }
1539
1540 int kvm_vm_ioctl_resize_hpt_commit(struct kvm *kvm,
1541 struct kvm_ppc_resize_hpt *rhpt)
1542 {
1543 unsigned long flags = rhpt->flags;
1544 unsigned long shift = rhpt->shift;
1545 struct kvm_resize_hpt *resize;
1546 int ret;
1547
1548 if (flags != 0 || kvm_is_radix(kvm))
1549 return -EINVAL;
1550
1551 if (shift && ((shift < 18) || (shift > 46)))
1552 return -EINVAL;
1553
1554 mutex_lock(&kvm->arch.mmu_setup_lock);
1555
1556 resize = kvm->arch.resize_hpt;
1557
1558 /* This shouldn't be possible */
1559 ret = -EIO;
1560 if (WARN_ON(!kvm->arch.mmu_ready))
1561 goto out_no_hpt;
1562
1563 /* Stop VCPUs from running while we mess with the HPT */
1564 kvm->arch.mmu_ready = 0;
1565 smp_mb();
1566
1567 /* Boot all CPUs out of the guest so they re-read
1568 * mmu_ready */
1569 on_each_cpu(resize_hpt_boot_vcpu, NULL, 1);
1570
1571 ret = -ENXIO;
1572 if (!resize || (resize->order != shift))
1573 goto out;
1574
1575 ret = resize->error;
1576 if (ret)
1577 goto out;
1578
1579 ret = resize_hpt_rehash(resize);
1580 if (ret)
1581 goto out;
1582
1583 resize_hpt_pivot(resize);
1584
1585 out:
1586 /* Let VCPUs run again */
1587 kvm->arch.mmu_ready = 1;
1588 smp_mb();
1589 out_no_hpt:
1590 resize_hpt_release(kvm, resize);
1591 mutex_unlock(&kvm->arch.mmu_setup_lock);
1592 return ret;
1593 }
1594
1595 /*
1596 * Functions for reading and writing the hash table via reads and
1597 * writes on a file descriptor.
1598 *
1599 * Reads return the guest view of the hash table, which has to be
1600 * pieced together from the real hash table and the guest_rpte
1601 * values in the revmap array.
1602 *
1603 * On writes, each HPTE written is considered in turn, and if it
1604 * is valid, it is written to the HPT as if an H_ENTER with the
1605 * exact flag set was done. When the invalid count is non-zero
1606 * in the header written to the stream, the kernel will make
1607 * sure that that many HPTEs are invalid, and invalidate them
1608 * if not.
1609 */
1610
1611 struct kvm_htab_ctx {
1612 unsigned long index;
1613 unsigned long flags;
1614 struct kvm *kvm;
1615 int first_pass;
1616 };
1617
1618 #define HPTE_SIZE (2 * sizeof(unsigned long))
1619
1620 /*
1621 * Returns 1 if this HPT entry has been modified or has pending
1622 * R/C bit changes.
1623 */
1624 static int hpte_dirty(struct revmap_entry *revp, __be64 *hptp)
1625 {
1626 unsigned long rcbits_unset;
1627
1628 if (revp->guest_rpte & HPTE_GR_MODIFIED)
1629 return 1;
1630
1631 /* Also need to consider changes in reference and changed bits */
1632 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1633 if ((be64_to_cpu(hptp[0]) & HPTE_V_VALID) &&
1634 (be64_to_cpu(hptp[1]) & rcbits_unset))
1635 return 1;
1636
1637 return 0;
1638 }
1639
1640 static long record_hpte(unsigned long flags, __be64 *hptp,
1641 unsigned long *hpte, struct revmap_entry *revp,
1642 int want_valid, int first_pass)
1643 {
1644 unsigned long v, r, hr;
1645 unsigned long rcbits_unset;
1646 int ok = 1;
1647 int valid, dirty;
1648
1649 /* Unmodified entries are uninteresting except on the first pass */
1650 dirty = hpte_dirty(revp, hptp);
1651 if (!first_pass && !dirty)
1652 return 0;
1653
1654 valid = 0;
1655 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1656 valid = 1;
1657 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) &&
1658 !(be64_to_cpu(hptp[0]) & HPTE_V_BOLTED))
1659 valid = 0;
1660 }
1661 if (valid != want_valid)
1662 return 0;
1663
1664 v = r = 0;
1665 if (valid || dirty) {
1666 /* lock the HPTE so it's stable and read it */
1667 preempt_disable();
1668 while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
1669 cpu_relax();
1670 v = be64_to_cpu(hptp[0]);
1671 hr = be64_to_cpu(hptp[1]);
1672 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1673 v = hpte_new_to_old_v(v, hr);
1674 hr = hpte_new_to_old_r(hr);
1675 }
1676
1677 /* re-evaluate valid and dirty from synchronized HPTE value */
1678 valid = !!(v & HPTE_V_VALID);
1679 dirty = !!(revp->guest_rpte & HPTE_GR_MODIFIED);
1680
1681 /* Harvest R and C into guest view if necessary */
1682 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1683 if (valid && (rcbits_unset & hr)) {
1684 revp->guest_rpte |= (hr &
1685 (HPTE_R_R | HPTE_R_C)) | HPTE_GR_MODIFIED;
1686 dirty = 1;
1687 }
1688
1689 if (v & HPTE_V_ABSENT) {
1690 v &= ~HPTE_V_ABSENT;
1691 v |= HPTE_V_VALID;
1692 valid = 1;
1693 }
1694 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) && !(v & HPTE_V_BOLTED))
1695 valid = 0;
1696
1697 r = revp->guest_rpte;
1698 /* only clear modified if this is the right sort of entry */
1699 if (valid == want_valid && dirty) {
1700 r &= ~HPTE_GR_MODIFIED;
1701 revp->guest_rpte = r;
1702 }
1703 unlock_hpte(hptp, be64_to_cpu(hptp[0]));
1704 preempt_enable();
1705 if (!(valid == want_valid && (first_pass || dirty)))
1706 ok = 0;
1707 }
1708 hpte[0] = cpu_to_be64(v);
1709 hpte[1] = cpu_to_be64(r);
1710 return ok;
1711 }
1712
1713 static ssize_t kvm_htab_read(struct file *file, char __user *buf,
1714 size_t count, loff_t *ppos)
1715 {
1716 struct kvm_htab_ctx *ctx = file->private_data;
1717 struct kvm *kvm = ctx->kvm;
1718 struct kvm_get_htab_header hdr;
1719 __be64 *hptp;
1720 struct revmap_entry *revp;
1721 unsigned long i, nb, nw;
1722 unsigned long __user *lbuf;
1723 struct kvm_get_htab_header __user *hptr;
1724 unsigned long flags;
1725 int first_pass;
1726 unsigned long hpte[2];
1727
1728 if (!access_ok(buf, count))
1729 return -EFAULT;
1730 if (kvm_is_radix(kvm))
1731 return 0;
1732
1733 first_pass = ctx->first_pass;
1734 flags = ctx->flags;
1735
1736 i = ctx->index;
1737 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1738 revp = kvm->arch.hpt.rev + i;
1739 lbuf = (unsigned long __user *)buf;
1740
1741 nb = 0;
1742 while (nb + sizeof(hdr) + HPTE_SIZE < count) {
1743 /* Initialize header */
1744 hptr = (struct kvm_get_htab_header __user *)buf;
1745 hdr.n_valid = 0;
1746 hdr.n_invalid = 0;
1747 nw = nb;
1748 nb += sizeof(hdr);
1749 lbuf = (unsigned long __user *)(buf + sizeof(hdr));
1750
1751 /* Skip uninteresting entries, i.e. clean on not-first pass */
1752 if (!first_pass) {
1753 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1754 !hpte_dirty(revp, hptp)) {
1755 ++i;
1756 hptp += 2;
1757 ++revp;
1758 }
1759 }
1760 hdr.index = i;
1761
1762 /* Grab a series of valid entries */
1763 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1764 hdr.n_valid < 0xffff &&
1765 nb + HPTE_SIZE < count &&
1766 record_hpte(flags, hptp, hpte, revp, 1, first_pass)) {
1767 /* valid entry, write it out */
1768 ++hdr.n_valid;
1769 if (__put_user(hpte[0], lbuf) ||
1770 __put_user(hpte[1], lbuf + 1))
1771 return -EFAULT;
1772 nb += HPTE_SIZE;
1773 lbuf += 2;
1774 ++i;
1775 hptp += 2;
1776 ++revp;
1777 }
1778 /* Now skip invalid entries while we can */
1779 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1780 hdr.n_invalid < 0xffff &&
1781 record_hpte(flags, hptp, hpte, revp, 0, first_pass)) {
1782 /* found an invalid entry */
1783 ++hdr.n_invalid;
1784 ++i;
1785 hptp += 2;
1786 ++revp;
1787 }
1788
1789 if (hdr.n_valid || hdr.n_invalid) {
1790 /* write back the header */
1791 if (__copy_to_user(hptr, &hdr, sizeof(hdr)))
1792 return -EFAULT;
1793 nw = nb;
1794 buf = (char __user *)lbuf;
1795 } else {
1796 nb = nw;
1797 }
1798
1799 /* Check if we've wrapped around the hash table */
1800 if (i >= kvmppc_hpt_npte(&kvm->arch.hpt)) {
1801 i = 0;
1802 ctx->first_pass = 0;
1803 break;
1804 }
1805 }
1806
1807 ctx->index = i;
1808
1809 return nb;
1810 }
1811
1812 static ssize_t kvm_htab_write(struct file *file, const char __user *buf,
1813 size_t count, loff_t *ppos)
1814 {
1815 struct kvm_htab_ctx *ctx = file->private_data;
1816 struct kvm *kvm = ctx->kvm;
1817 struct kvm_get_htab_header hdr;
1818 unsigned long i, j;
1819 unsigned long v, r;
1820 unsigned long __user *lbuf;
1821 __be64 *hptp;
1822 unsigned long tmp[2];
1823 ssize_t nb;
1824 long int err, ret;
1825 int mmu_ready;
1826 int pshift;
1827
1828 if (!access_ok(buf, count))
1829 return -EFAULT;
1830 if (kvm_is_radix(kvm))
1831 return -EINVAL;
1832
1833 /* lock out vcpus from running while we're doing this */
1834 mutex_lock(&kvm->arch.mmu_setup_lock);
1835 mmu_ready = kvm->arch.mmu_ready;
1836 if (mmu_ready) {
1837 kvm->arch.mmu_ready = 0; /* temporarily */
1838 /* order mmu_ready vs. vcpus_running */
1839 smp_mb();
1840 if (atomic_read(&kvm->arch.vcpus_running)) {
1841 kvm->arch.mmu_ready = 1;
1842 mutex_unlock(&kvm->arch.mmu_setup_lock);
1843 return -EBUSY;
1844 }
1845 }
1846
1847 err = 0;
1848 for (nb = 0; nb + sizeof(hdr) <= count; ) {
1849 err = -EFAULT;
1850 if (__copy_from_user(&hdr, buf, sizeof(hdr)))
1851 break;
1852
1853 err = 0;
1854 if (nb + hdr.n_valid * HPTE_SIZE > count)
1855 break;
1856
1857 nb += sizeof(hdr);
1858 buf += sizeof(hdr);
1859
1860 err = -EINVAL;
1861 i = hdr.index;
1862 if (i >= kvmppc_hpt_npte(&kvm->arch.hpt) ||
1863 i + hdr.n_valid + hdr.n_invalid > kvmppc_hpt_npte(&kvm->arch.hpt))
1864 break;
1865
1866 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1867 lbuf = (unsigned long __user *)buf;
1868 for (j = 0; j < hdr.n_valid; ++j) {
1869 __be64 hpte_v;
1870 __be64 hpte_r;
1871
1872 err = -EFAULT;
1873 if (__get_user(hpte_v, lbuf) ||
1874 __get_user(hpte_r, lbuf + 1))
1875 goto out;
1876 v = be64_to_cpu(hpte_v);
1877 r = be64_to_cpu(hpte_r);
1878 err = -EINVAL;
1879 if (!(v & HPTE_V_VALID))
1880 goto out;
1881 pshift = kvmppc_hpte_base_page_shift(v, r);
1882 if (pshift <= 0)
1883 goto out;
1884 lbuf += 2;
1885 nb += HPTE_SIZE;
1886
1887 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1888 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1889 err = -EIO;
1890 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, i, v, r,
1891 tmp);
1892 if (ret != H_SUCCESS) {
1893 pr_err("%s ret %ld i=%ld v=%lx r=%lx\n", __func__, ret, i, v, r);
1894 goto out;
1895 }
1896 if (!mmu_ready && is_vrma_hpte(v)) {
1897 unsigned long senc, lpcr;
1898
1899 senc = slb_pgsize_encoding(1ul << pshift);
1900 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1901 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1902 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1903 lpcr = senc << (LPCR_VRMASD_SH - 4);
1904 kvmppc_update_lpcr(kvm, lpcr,
1905 LPCR_VRMASD);
1906 } else {
1907 kvmppc_setup_partition_table(kvm);
1908 }
1909 mmu_ready = 1;
1910 }
1911 ++i;
1912 hptp += 2;
1913 }
1914
1915 for (j = 0; j < hdr.n_invalid; ++j) {
1916 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1917 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1918 ++i;
1919 hptp += 2;
1920 }
1921 err = 0;
1922 }
1923
1924 out:
1925 /* Order HPTE updates vs. mmu_ready */
1926 smp_wmb();
1927 kvm->arch.mmu_ready = mmu_ready;
1928 mutex_unlock(&kvm->arch.mmu_setup_lock);
1929
1930 if (err)
1931 return err;
1932 return nb;
1933 }
1934
1935 static int kvm_htab_release(struct inode *inode, struct file *filp)
1936 {
1937 struct kvm_htab_ctx *ctx = filp->private_data;
1938
1939 filp->private_data = NULL;
1940 if (!(ctx->flags & KVM_GET_HTAB_WRITE))
1941 atomic_dec(&ctx->kvm->arch.hpte_mod_interest);
1942 kvm_put_kvm(ctx->kvm);
1943 kfree(ctx);
1944 return 0;
1945 }
1946
1947 static const struct file_operations kvm_htab_fops = {
1948 .read = kvm_htab_read,
1949 .write = kvm_htab_write,
1950 .llseek = default_llseek,
1951 .release = kvm_htab_release,
1952 };
1953
1954 int kvm_vm_ioctl_get_htab_fd(struct kvm *kvm, struct kvm_get_htab_fd *ghf)
1955 {
1956 int ret;
1957 struct kvm_htab_ctx *ctx;
1958 int rwflag;
1959
1960 /* reject flags we don't recognize */
1961 if (ghf->flags & ~(KVM_GET_HTAB_BOLTED_ONLY | KVM_GET_HTAB_WRITE))
1962 return -EINVAL;
1963 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1964 if (!ctx)
1965 return -ENOMEM;
1966 kvm_get_kvm(kvm);
1967 ctx->kvm = kvm;
1968 ctx->index = ghf->start_index;
1969 ctx->flags = ghf->flags;
1970 ctx->first_pass = 1;
1971
1972 rwflag = (ghf->flags & KVM_GET_HTAB_WRITE) ? O_WRONLY : O_RDONLY;
1973 ret = anon_inode_getfd("kvm-htab", &kvm_htab_fops, ctx, rwflag | O_CLOEXEC);
1974 if (ret < 0) {
1975 kfree(ctx);
1976 kvm_put_kvm_no_destroy(kvm);
1977 return ret;
1978 }
1979
1980 if (rwflag == O_RDONLY) {
1981 mutex_lock(&kvm->slots_lock);
1982 atomic_inc(&kvm->arch.hpte_mod_interest);
1983 /* make sure kvmppc_do_h_enter etc. see the increment */
1984 synchronize_srcu_expedited(&kvm->srcu);
1985 mutex_unlock(&kvm->slots_lock);
1986 }
1987
1988 return ret;
1989 }
1990
1991 struct debugfs_htab_state {
1992 struct kvm *kvm;
1993 struct mutex mutex;
1994 unsigned long hpt_index;
1995 int chars_left;
1996 int buf_index;
1997 char buf[64];
1998 };
1999
2000 static int debugfs_htab_open(struct inode *inode, struct file *file)
2001 {
2002 struct kvm *kvm = inode->i_private;
2003 struct debugfs_htab_state *p;
2004
2005 p = kzalloc(sizeof(*p), GFP_KERNEL);
2006 if (!p)
2007 return -ENOMEM;
2008
2009 kvm_get_kvm(kvm);
2010 p->kvm = kvm;
2011 mutex_init(&p->mutex);
2012 file->private_data = p;
2013
2014 return nonseekable_open(inode, file);
2015 }
2016
2017 static int debugfs_htab_release(struct inode *inode, struct file *file)
2018 {
2019 struct debugfs_htab_state *p = file->private_data;
2020
2021 kvm_put_kvm(p->kvm);
2022 kfree(p);
2023 return 0;
2024 }
2025
2026 static ssize_t debugfs_htab_read(struct file *file, char __user *buf,
2027 size_t len, loff_t *ppos)
2028 {
2029 struct debugfs_htab_state *p = file->private_data;
2030 ssize_t ret, r;
2031 unsigned long i, n;
2032 unsigned long v, hr, gr;
2033 struct kvm *kvm;
2034 __be64 *hptp;
2035
2036 kvm = p->kvm;
2037 if (kvm_is_radix(kvm))
2038 return 0;
2039
2040 ret = mutex_lock_interruptible(&p->mutex);
2041 if (ret)
2042 return ret;
2043
2044 if (p->chars_left) {
2045 n = p->chars_left;
2046 if (n > len)
2047 n = len;
2048 r = copy_to_user(buf, p->buf + p->buf_index, n);
2049 n -= r;
2050 p->chars_left -= n;
2051 p->buf_index += n;
2052 buf += n;
2053 len -= n;
2054 ret = n;
2055 if (r) {
2056 if (!n)
2057 ret = -EFAULT;
2058 goto out;
2059 }
2060 }
2061
2062 i = p->hpt_index;
2063 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
2064 for (; len != 0 && i < kvmppc_hpt_npte(&kvm->arch.hpt);
2065 ++i, hptp += 2) {
2066 if (!(be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)))
2067 continue;
2068
2069 /* lock the HPTE so it's stable and read it */
2070 preempt_disable();
2071 while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
2072 cpu_relax();
2073 v = be64_to_cpu(hptp[0]) & ~HPTE_V_HVLOCK;
2074 hr = be64_to_cpu(hptp[1]);
2075 gr = kvm->arch.hpt.rev[i].guest_rpte;
2076 unlock_hpte(hptp, v);
2077 preempt_enable();
2078
2079 if (!(v & (HPTE_V_VALID | HPTE_V_ABSENT)))
2080 continue;
2081
2082 n = scnprintf(p->buf, sizeof(p->buf),
2083 "%6lx %.16lx %.16lx %.16lx\n",
2084 i, v, hr, gr);
2085 p->chars_left = n;
2086 if (n > len)
2087 n = len;
2088 r = copy_to_user(buf, p->buf, n);
2089 n -= r;
2090 p->chars_left -= n;
2091 p->buf_index = n;
2092 buf += n;
2093 len -= n;
2094 ret += n;
2095 if (r) {
2096 if (!ret)
2097 ret = -EFAULT;
2098 goto out;
2099 }
2100 }
2101 p->hpt_index = i;
2102
2103 out:
2104 mutex_unlock(&p->mutex);
2105 return ret;
2106 }
2107
2108 static ssize_t debugfs_htab_write(struct file *file, const char __user *buf,
2109 size_t len, loff_t *ppos)
2110 {
2111 return -EACCES;
2112 }
2113
2114 static const struct file_operations debugfs_htab_fops = {
2115 .owner = THIS_MODULE,
2116 .open = debugfs_htab_open,
2117 .release = debugfs_htab_release,
2118 .read = debugfs_htab_read,
2119 .write = debugfs_htab_write,
2120 .llseek = generic_file_llseek,
2121 };
2122
2123 void kvmppc_mmu_debugfs_init(struct kvm *kvm)
2124 {
2125 debugfs_create_file("htab", 0400, kvm->debugfs_dentry, kvm,
2126 &debugfs_htab_fops);
2127 }
2128
2129 void kvmppc_mmu_book3s_hv_init(struct kvm_vcpu *vcpu)
2130 {
2131 struct kvmppc_mmu *mmu = &vcpu->arch.mmu;
2132
2133 vcpu->arch.slb_nr = 32; /* POWER7/POWER8 */
2134
2135 mmu->xlate = kvmppc_mmu_book3s_64_hv_xlate;
2136
2137 vcpu->arch.hflags |= BOOK3S_HFLAG_SLB;
2138 }
2139
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.