1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* |
3 | * Copyright (C) 2009 SUSE Linux Products GmbH. All rights reserved. |
4 | * |
5 | * Authors: |
6 | * Alexander Graf <agraf@suse.de> |
7 | * Kevin Wolf <mail@kevin-wolf.de> |
8 | */ |
9 | |
10 | #include <linux/kvm_host.h> |
11 | #include <linux/pkeys.h> |
12 | |
13 | #include <asm/kvm_ppc.h> |
14 | #include <asm/kvm_book3s.h> |
15 | #include <asm/book3s/64/mmu-hash.h> |
16 | #include <asm/machdep.h> |
17 | #include <asm/mmu_context.h> |
18 | #include <asm/hw_irq.h> |
19 | #include "trace_pr.h" |
20 | #include "book3s.h" |
21 | |
22 | #define PTE_SIZE 12 |
23 | |
24 | void kvmppc_mmu_invalidate_pte(struct kvm_vcpu *vcpu, struct hpte_cache *pte) |
25 | { |
26 | mmu_hash_ops.hpte_invalidate(pte->slot, pte->host_vpn, |
27 | pte->pagesize, pte->pagesize, |
28 | MMU_SEGSIZE_256M, false); |
29 | } |
30 | |
31 | /* We keep 512 gvsid->hvsid entries, mapping the guest ones to the array using |
32 | * a hash, so we don't waste cycles on looping */ |
33 | static u16 kvmppc_sid_hash(struct kvm_vcpu *vcpu, u64 gvsid) |
34 | { |
35 | return (u16)(((gvsid >> (SID_MAP_BITS * 7)) & SID_MAP_MASK) ^ |
36 | ((gvsid >> (SID_MAP_BITS * 6)) & SID_MAP_MASK) ^ |
37 | ((gvsid >> (SID_MAP_BITS * 5)) & SID_MAP_MASK) ^ |
38 | ((gvsid >> (SID_MAP_BITS * 4)) & SID_MAP_MASK) ^ |
39 | ((gvsid >> (SID_MAP_BITS * 3)) & SID_MAP_MASK) ^ |
40 | ((gvsid >> (SID_MAP_BITS * 2)) & SID_MAP_MASK) ^ |
41 | ((gvsid >> (SID_MAP_BITS * 1)) & SID_MAP_MASK) ^ |
42 | ((gvsid >> (SID_MAP_BITS * 0)) & SID_MAP_MASK)); |
43 | } |
44 | |
45 | |
46 | static struct kvmppc_sid_map *find_sid_vsid(struct kvm_vcpu *vcpu, u64 gvsid) |
47 | { |
48 | struct kvmppc_sid_map *map; |
49 | u16 sid_map_mask; |
50 | |
51 | if (kvmppc_get_msr(vcpu) & MSR_PR) |
52 | gvsid |= VSID_PR; |
53 | |
54 | sid_map_mask = kvmppc_sid_hash(vcpu, gvsid); |
55 | map = &to_book3s(vcpu)->sid_map[sid_map_mask]; |
56 | if (map->valid && (map->guest_vsid == gvsid)) { |
57 | trace_kvm_book3s_slb_found(gvsid, hvsid: map->host_vsid); |
58 | return map; |
59 | } |
60 | |
61 | map = &to_book3s(vcpu)->sid_map[SID_MAP_MASK - sid_map_mask]; |
62 | if (map->valid && (map->guest_vsid == gvsid)) { |
63 | trace_kvm_book3s_slb_found(gvsid, hvsid: map->host_vsid); |
64 | return map; |
65 | } |
66 | |
67 | trace_kvm_book3s_slb_fail(sid_map_mask, gvsid); |
68 | return NULL; |
69 | } |
70 | |
71 | int kvmppc_mmu_map_page(struct kvm_vcpu *vcpu, struct kvmppc_pte *orig_pte, |
72 | bool iswrite) |
73 | { |
74 | unsigned long vpn; |
75 | kvm_pfn_t hpaddr; |
76 | ulong hash, hpteg; |
77 | u64 vsid; |
78 | int ret; |
79 | int rflags = 0x192; |
80 | int vflags = 0; |
81 | int attempt = 0; |
82 | struct kvmppc_sid_map *map; |
83 | int r = 0; |
84 | int hpsize = MMU_PAGE_4K; |
85 | bool writable; |
86 | unsigned long mmu_seq; |
87 | struct kvm *kvm = vcpu->kvm; |
88 | struct hpte_cache *cpte; |
89 | unsigned long gfn = orig_pte->raddr >> PAGE_SHIFT; |
90 | unsigned long pfn; |
91 | |
92 | /* used to check for invalidations in progress */ |
93 | mmu_seq = kvm->mmu_invalidate_seq; |
94 | smp_rmb(); |
95 | |
96 | /* Get host physical address for gpa */ |
97 | pfn = kvmppc_gpa_to_pfn(vcpu, orig_pte->raddr, iswrite, &writable); |
98 | if (is_error_noslot_pfn(pfn)) { |
99 | printk(KERN_INFO "Couldn't get guest page for gpa %lx!\n" , |
100 | orig_pte->raddr); |
101 | r = -EINVAL; |
102 | goto out; |
103 | } |
104 | hpaddr = pfn << PAGE_SHIFT; |
105 | |
106 | /* and write the mapping ea -> hpa into the pt */ |
107 | vcpu->arch.mmu.esid_to_vsid(vcpu, orig_pte->eaddr >> SID_SHIFT, &vsid); |
108 | map = find_sid_vsid(vcpu, gvsid: vsid); |
109 | if (!map) { |
110 | ret = kvmppc_mmu_map_segment(vcpu, orig_pte->eaddr); |
111 | WARN_ON(ret < 0); |
112 | map = find_sid_vsid(vcpu, gvsid: vsid); |
113 | } |
114 | if (!map) { |
115 | printk(KERN_ERR "KVM: Segment map for 0x%llx (0x%lx) failed\n" , |
116 | vsid, orig_pte->eaddr); |
117 | WARN_ON(true); |
118 | r = -EINVAL; |
119 | goto out; |
120 | } |
121 | |
122 | vpn = hpt_vpn(orig_pte->eaddr, map->host_vsid, MMU_SEGSIZE_256M); |
123 | |
124 | kvm_set_pfn_accessed(pfn); |
125 | if (!orig_pte->may_write || !writable) |
126 | rflags |= PP_RXRX; |
127 | else { |
128 | mark_page_dirty(kvm: vcpu->kvm, gfn); |
129 | kvm_set_pfn_dirty(pfn); |
130 | } |
131 | |
132 | if (!orig_pte->may_execute) |
133 | rflags |= HPTE_R_N; |
134 | else |
135 | kvmppc_mmu_flush_icache(pfn); |
136 | |
137 | rflags |= pte_to_hpte_pkey_bits(0, HPTE_USE_KERNEL_KEY); |
138 | rflags = (rflags & ~HPTE_R_WIMG) | orig_pte->wimg; |
139 | |
140 | /* |
141 | * Use 64K pages if possible; otherwise, on 64K page kernels, |
142 | * we need to transfer 4 more bits from guest real to host real addr. |
143 | */ |
144 | if (vsid & VSID_64K) |
145 | hpsize = MMU_PAGE_64K; |
146 | else |
147 | hpaddr |= orig_pte->raddr & (~0xfffULL & ~PAGE_MASK); |
148 | |
149 | hash = hpt_hash(vpn, mmu_psize_defs[hpsize].shift, MMU_SEGSIZE_256M); |
150 | |
151 | cpte = kvmppc_mmu_hpte_cache_next(vcpu); |
152 | |
153 | spin_lock(lock: &kvm->mmu_lock); |
154 | if (!cpte || mmu_invalidate_retry(kvm, mmu_seq)) { |
155 | r = -EAGAIN; |
156 | goto out_unlock; |
157 | } |
158 | |
159 | map_again: |
160 | hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP); |
161 | |
162 | /* In case we tried normal mapping already, let's nuke old entries */ |
163 | if (attempt > 1) |
164 | if (mmu_hash_ops.hpte_remove(hpteg) < 0) { |
165 | r = -1; |
166 | goto out_unlock; |
167 | } |
168 | |
169 | ret = mmu_hash_ops.hpte_insert(hpteg, vpn, hpaddr, rflags, vflags, |
170 | hpsize, hpsize, MMU_SEGSIZE_256M); |
171 | |
172 | if (ret == -1) { |
173 | /* If we couldn't map a primary PTE, try a secondary */ |
174 | hash = ~hash; |
175 | vflags ^= HPTE_V_SECONDARY; |
176 | attempt++; |
177 | goto map_again; |
178 | } else if (ret < 0) { |
179 | r = -EIO; |
180 | goto out_unlock; |
181 | } else { |
182 | trace_kvm_book3s_64_mmu_map(rflags, hpteg, |
183 | vpn, hpaddr, orig_pte); |
184 | |
185 | /* |
186 | * The mmu_hash_ops code may give us a secondary entry even |
187 | * though we asked for a primary. Fix up. |
188 | */ |
189 | if ((ret & _PTEIDX_SECONDARY) && !(vflags & HPTE_V_SECONDARY)) { |
190 | hash = ~hash; |
191 | hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP); |
192 | } |
193 | |
194 | cpte->slot = hpteg + (ret & 7); |
195 | cpte->host_vpn = vpn; |
196 | cpte->pte = *orig_pte; |
197 | cpte->pfn = pfn; |
198 | cpte->pagesize = hpsize; |
199 | |
200 | kvmppc_mmu_hpte_cache_map(vcpu, cpte); |
201 | cpte = NULL; |
202 | } |
203 | |
204 | out_unlock: |
205 | spin_unlock(lock: &kvm->mmu_lock); |
206 | kvm_release_pfn_clean(pfn); |
207 | if (cpte) |
208 | kvmppc_mmu_hpte_cache_free(cpte); |
209 | |
210 | out: |
211 | return r; |
212 | } |
213 | |
214 | void kvmppc_mmu_unmap_page(struct kvm_vcpu *vcpu, struct kvmppc_pte *pte) |
215 | { |
216 | u64 mask = 0xfffffffffULL; |
217 | u64 vsid; |
218 | |
219 | vcpu->arch.mmu.esid_to_vsid(vcpu, pte->eaddr >> SID_SHIFT, &vsid); |
220 | if (vsid & VSID_64K) |
221 | mask = 0xffffffff0ULL; |
222 | kvmppc_mmu_pte_vflush(vcpu, pte->vpage, mask); |
223 | } |
224 | |
225 | static struct kvmppc_sid_map *create_sid_map(struct kvm_vcpu *vcpu, u64 gvsid) |
226 | { |
227 | unsigned long vsid_bits = VSID_BITS_65_256M; |
228 | struct kvmppc_sid_map *map; |
229 | struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu); |
230 | u16 sid_map_mask; |
231 | static int backwards_map; |
232 | |
233 | if (kvmppc_get_msr(vcpu) & MSR_PR) |
234 | gvsid |= VSID_PR; |
235 | |
236 | /* We might get collisions that trap in preceding order, so let's |
237 | map them differently */ |
238 | |
239 | sid_map_mask = kvmppc_sid_hash(vcpu, gvsid); |
240 | if (backwards_map) |
241 | sid_map_mask = SID_MAP_MASK - sid_map_mask; |
242 | |
243 | map = &to_book3s(vcpu)->sid_map[sid_map_mask]; |
244 | |
245 | /* Make sure we're taking the other map next time */ |
246 | backwards_map = !backwards_map; |
247 | |
248 | /* Uh-oh ... out of mappings. Let's flush! */ |
249 | if (vcpu_book3s->proto_vsid_next == vcpu_book3s->proto_vsid_max) { |
250 | vcpu_book3s->proto_vsid_next = vcpu_book3s->proto_vsid_first; |
251 | memset(vcpu_book3s->sid_map, 0, |
252 | sizeof(struct kvmppc_sid_map) * SID_MAP_NUM); |
253 | kvmppc_mmu_pte_flush(vcpu, 0, 0); |
254 | kvmppc_mmu_flush_segments(vcpu); |
255 | } |
256 | |
257 | if (mmu_has_feature(MMU_FTR_68_BIT_VA)) |
258 | vsid_bits = VSID_BITS_256M; |
259 | |
260 | map->host_vsid = vsid_scramble(vcpu_book3s->proto_vsid_next++, |
261 | VSID_MULTIPLIER_256M, vsid_bits); |
262 | |
263 | map->guest_vsid = gvsid; |
264 | map->valid = true; |
265 | |
266 | trace_kvm_book3s_slb_map(sid_map_mask, gvsid, hvsid: map->host_vsid); |
267 | |
268 | return map; |
269 | } |
270 | |
271 | static int kvmppc_mmu_next_segment(struct kvm_vcpu *vcpu, ulong esid) |
272 | { |
273 | struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu); |
274 | int i; |
275 | int max_slb_size = 64; |
276 | int found_inval = -1; |
277 | int r; |
278 | |
279 | /* Are we overwriting? */ |
280 | for (i = 0; i < svcpu->slb_max; i++) { |
281 | if (!(svcpu->slb[i].esid & SLB_ESID_V)) |
282 | found_inval = i; |
283 | else if ((svcpu->slb[i].esid & ESID_MASK) == esid) { |
284 | r = i; |
285 | goto out; |
286 | } |
287 | } |
288 | |
289 | /* Found a spare entry that was invalidated before */ |
290 | if (found_inval >= 0) { |
291 | r = found_inval; |
292 | goto out; |
293 | } |
294 | |
295 | /* No spare invalid entry, so create one */ |
296 | |
297 | if (mmu_slb_size < 64) |
298 | max_slb_size = mmu_slb_size; |
299 | |
300 | /* Overflowing -> purge */ |
301 | if ((svcpu->slb_max) == max_slb_size) |
302 | kvmppc_mmu_flush_segments(vcpu); |
303 | |
304 | r = svcpu->slb_max; |
305 | svcpu->slb_max++; |
306 | |
307 | out: |
308 | svcpu_put(svcpu); |
309 | return r; |
310 | } |
311 | |
312 | int kvmppc_mmu_map_segment(struct kvm_vcpu *vcpu, ulong eaddr) |
313 | { |
314 | struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu); |
315 | u64 esid = eaddr >> SID_SHIFT; |
316 | u64 slb_esid = (eaddr & ESID_MASK) | SLB_ESID_V; |
317 | u64 slb_vsid = SLB_VSID_USER; |
318 | u64 gvsid; |
319 | int slb_index; |
320 | struct kvmppc_sid_map *map; |
321 | int r = 0; |
322 | |
323 | slb_index = kvmppc_mmu_next_segment(vcpu, eaddr & ESID_MASK); |
324 | |
325 | if (vcpu->arch.mmu.esid_to_vsid(vcpu, esid, &gvsid)) { |
326 | /* Invalidate an entry */ |
327 | svcpu->slb[slb_index].esid = 0; |
328 | r = -ENOENT; |
329 | goto out; |
330 | } |
331 | |
332 | map = find_sid_vsid(vcpu, gvsid); |
333 | if (!map) |
334 | map = create_sid_map(vcpu, gvsid); |
335 | |
336 | map->guest_esid = esid; |
337 | |
338 | slb_vsid |= (map->host_vsid << 12); |
339 | slb_vsid &= ~SLB_VSID_KP; |
340 | slb_esid |= slb_index; |
341 | |
342 | #ifdef CONFIG_PPC_64K_PAGES |
343 | /* Set host segment base page size to 64K if possible */ |
344 | if (gvsid & VSID_64K) |
345 | slb_vsid |= mmu_psize_defs[MMU_PAGE_64K].sllp; |
346 | #endif |
347 | |
348 | svcpu->slb[slb_index].esid = slb_esid; |
349 | svcpu->slb[slb_index].vsid = slb_vsid; |
350 | |
351 | trace_kvm_book3s_slbmte(slb_vsid, slb_esid); |
352 | |
353 | out: |
354 | svcpu_put(svcpu); |
355 | return r; |
356 | } |
357 | |
358 | void kvmppc_mmu_flush_segment(struct kvm_vcpu *vcpu, ulong ea, ulong seg_size) |
359 | { |
360 | struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu); |
361 | ulong seg_mask = -seg_size; |
362 | int i; |
363 | |
364 | for (i = 0; i < svcpu->slb_max; i++) { |
365 | if ((svcpu->slb[i].esid & SLB_ESID_V) && |
366 | (svcpu->slb[i].esid & seg_mask) == ea) { |
367 | /* Invalidate this entry */ |
368 | svcpu->slb[i].esid = 0; |
369 | } |
370 | } |
371 | |
372 | svcpu_put(svcpu); |
373 | } |
374 | |
375 | void kvmppc_mmu_flush_segments(struct kvm_vcpu *vcpu) |
376 | { |
377 | struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu); |
378 | svcpu->slb_max = 0; |
379 | svcpu->slb[0].esid = 0; |
380 | svcpu_put(svcpu); |
381 | } |
382 | |
383 | void kvmppc_mmu_destroy_pr(struct kvm_vcpu *vcpu) |
384 | { |
385 | kvmppc_mmu_hpte_destroy(vcpu); |
386 | __destroy_context(to_book3s(vcpu)->context_id[0]); |
387 | } |
388 | |
389 | int kvmppc_mmu_init_pr(struct kvm_vcpu *vcpu) |
390 | { |
391 | struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu); |
392 | int err; |
393 | |
394 | err = hash__alloc_context_id(); |
395 | if (err < 0) |
396 | return -1; |
397 | vcpu3s->context_id[0] = err; |
398 | |
399 | vcpu3s->proto_vsid_max = ((u64)(vcpu3s->context_id[0] + 1) |
400 | << ESID_BITS) - 1; |
401 | vcpu3s->proto_vsid_first = (u64)vcpu3s->context_id[0] << ESID_BITS; |
402 | vcpu3s->proto_vsid_next = vcpu3s->proto_vsid_first; |
403 | |
404 | kvmppc_mmu_hpte_init(vcpu); |
405 | |
406 | return 0; |
407 | } |
408 | |