Fix CPU spinlock lockups on secondary CPU bringup
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / arch / powerpc / kvm / book3s_32_mmu_host.c
blob9fecbfbce773205650a9d6d9cd77704a61757a38
1 /*
2 * Copyright (C) 2010 SUSE Linux Products GmbH. All rights reserved.
4 * Authors:
5 * Alexander Graf <agraf@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License, version 2, as
9 * published by the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
21 #include <linux/kvm_host.h>
23 #include <asm/kvm_ppc.h>
24 #include <asm/kvm_book3s.h>
25 #include <asm/mmu-hash32.h>
26 #include <asm/machdep.h>
27 #include <asm/mmu_context.h>
28 #include <asm/hw_irq.h>
30 /* #define DEBUG_MMU */
31 /* #define DEBUG_SR */
33 #ifdef DEBUG_MMU
34 #define dprintk_mmu(a, ...) printk(KERN_INFO a, __VA_ARGS__)
35 #else
36 #define dprintk_mmu(a, ...) do { } while(0)
37 #endif
39 #ifdef DEBUG_SR
40 #define dprintk_sr(a, ...) printk(KERN_INFO a, __VA_ARGS__)
41 #else
42 #define dprintk_sr(a, ...) do { } while(0)
43 #endif
45 #if PAGE_SHIFT != 12
46 #error Unknown page size
47 #endif
49 #ifdef CONFIG_SMP
50 #error XXX need to grab mmu_hash_lock
51 #endif
53 #ifdef CONFIG_PTE_64BIT
54 #error Only 32 bit pages are supported for now
55 #endif
57 static ulong htab;
58 static u32 htabmask;
60 void kvmppc_mmu_invalidate_pte(struct kvm_vcpu *vcpu, struct hpte_cache *pte)
62 volatile u32 *pteg;
64 /* Remove from host HTAB */
65 pteg = (u32*)pte->slot;
66 pteg[0] = 0;
68 /* And make sure it's gone from the TLB too */
69 asm volatile ("sync");
70 asm volatile ("tlbie %0" : : "r" (pte->pte.eaddr) : "memory");
71 asm volatile ("sync");
72 asm volatile ("tlbsync");
75 /* We keep 512 gvsid->hvsid entries, mapping the guest ones to the array using
76 * a hash, so we don't waste cycles on looping */
77 static u16 kvmppc_sid_hash(struct kvm_vcpu *vcpu, u64 gvsid)
79 return (u16)(((gvsid >> (SID_MAP_BITS * 7)) & SID_MAP_MASK) ^
80 ((gvsid >> (SID_MAP_BITS * 6)) & SID_MAP_MASK) ^
81 ((gvsid >> (SID_MAP_BITS * 5)) & SID_MAP_MASK) ^
82 ((gvsid >> (SID_MAP_BITS * 4)) & SID_MAP_MASK) ^
83 ((gvsid >> (SID_MAP_BITS * 3)) & SID_MAP_MASK) ^
84 ((gvsid >> (SID_MAP_BITS * 2)) & SID_MAP_MASK) ^
85 ((gvsid >> (SID_MAP_BITS * 1)) & SID_MAP_MASK) ^
86 ((gvsid >> (SID_MAP_BITS * 0)) & SID_MAP_MASK));
90 static struct kvmppc_sid_map *find_sid_vsid(struct kvm_vcpu *vcpu, u64 gvsid)
92 struct kvmppc_sid_map *map;
93 u16 sid_map_mask;
95 if (vcpu->arch.shared->msr & MSR_PR)
96 gvsid |= VSID_PR;
98 sid_map_mask = kvmppc_sid_hash(vcpu, gvsid);
99 map = &to_book3s(vcpu)->sid_map[sid_map_mask];
100 if (map->guest_vsid == gvsid) {
101 dprintk_sr("SR: Searching 0x%llx -> 0x%llx\n",
102 gvsid, map->host_vsid);
103 return map;
106 map = &to_book3s(vcpu)->sid_map[SID_MAP_MASK - sid_map_mask];
107 if (map->guest_vsid == gvsid) {
108 dprintk_sr("SR: Searching 0x%llx -> 0x%llx\n",
109 gvsid, map->host_vsid);
110 return map;
113 dprintk_sr("SR: Searching 0x%llx -> not found\n", gvsid);
114 return NULL;
117 static u32 *kvmppc_mmu_get_pteg(struct kvm_vcpu *vcpu, u32 vsid, u32 eaddr,
118 bool primary)
120 u32 page, hash;
121 ulong pteg = htab;
123 page = (eaddr & ~ESID_MASK) >> 12;
125 hash = ((vsid ^ page) << 6);
126 if (!primary)
127 hash = ~hash;
129 hash &= htabmask;
131 pteg |= hash;
133 dprintk_mmu("htab: %lx | hash: %x | htabmask: %x | pteg: %lx\n",
134 htab, hash, htabmask, pteg);
136 return (u32*)pteg;
139 extern char etext[];
141 int kvmppc_mmu_map_page(struct kvm_vcpu *vcpu, struct kvmppc_pte *orig_pte)
143 pfn_t hpaddr;
144 u64 va;
145 u64 vsid;
146 struct kvmppc_sid_map *map;
147 volatile u32 *pteg;
148 u32 eaddr = orig_pte->eaddr;
149 u32 pteg0, pteg1;
150 register int rr = 0;
151 bool primary = false;
152 bool evict = false;
153 struct hpte_cache *pte;
155 /* Get host physical address for gpa */
156 hpaddr = kvmppc_gfn_to_pfn(vcpu, orig_pte->raddr >> PAGE_SHIFT);
157 if (is_error_pfn(hpaddr)) {
158 printk(KERN_INFO "Couldn't get guest page for gfn %lx!\n",
159 orig_pte->eaddr);
160 return -EINVAL;
162 hpaddr <<= PAGE_SHIFT;
164 /* and write the mapping ea -> hpa into the pt */
165 vcpu->arch.mmu.esid_to_vsid(vcpu, orig_pte->eaddr >> SID_SHIFT, &vsid);
166 map = find_sid_vsid(vcpu, vsid);
167 if (!map) {
168 kvmppc_mmu_map_segment(vcpu, eaddr);
169 map = find_sid_vsid(vcpu, vsid);
171 BUG_ON(!map);
173 vsid = map->host_vsid;
174 va = (vsid << SID_SHIFT) | (eaddr & ~ESID_MASK);
176 next_pteg:
177 if (rr == 16) {
178 primary = !primary;
179 evict = true;
180 rr = 0;
183 pteg = kvmppc_mmu_get_pteg(vcpu, vsid, eaddr, primary);
185 /* not evicting yet */
186 if (!evict && (pteg[rr] & PTE_V)) {
187 rr += 2;
188 goto next_pteg;
191 dprintk_mmu("KVM: old PTEG: %p (%d)\n", pteg, rr);
192 dprintk_mmu("KVM: %08x - %08x\n", pteg[0], pteg[1]);
193 dprintk_mmu("KVM: %08x - %08x\n", pteg[2], pteg[3]);
194 dprintk_mmu("KVM: %08x - %08x\n", pteg[4], pteg[5]);
195 dprintk_mmu("KVM: %08x - %08x\n", pteg[6], pteg[7]);
196 dprintk_mmu("KVM: %08x - %08x\n", pteg[8], pteg[9]);
197 dprintk_mmu("KVM: %08x - %08x\n", pteg[10], pteg[11]);
198 dprintk_mmu("KVM: %08x - %08x\n", pteg[12], pteg[13]);
199 dprintk_mmu("KVM: %08x - %08x\n", pteg[14], pteg[15]);
201 pteg0 = ((eaddr & 0x0fffffff) >> 22) | (vsid << 7) | PTE_V |
202 (primary ? 0 : PTE_SEC);
203 pteg1 = hpaddr | PTE_M | PTE_R | PTE_C;
205 if (orig_pte->may_write) {
206 pteg1 |= PP_RWRW;
207 mark_page_dirty(vcpu->kvm, orig_pte->raddr >> PAGE_SHIFT);
208 } else {
209 pteg1 |= PP_RWRX;
212 local_irq_disable();
214 if (pteg[rr]) {
215 pteg[rr] = 0;
216 asm volatile ("sync");
218 pteg[rr + 1] = pteg1;
219 pteg[rr] = pteg0;
220 asm volatile ("sync");
222 local_irq_enable();
224 dprintk_mmu("KVM: new PTEG: %p\n", pteg);
225 dprintk_mmu("KVM: %08x - %08x\n", pteg[0], pteg[1]);
226 dprintk_mmu("KVM: %08x - %08x\n", pteg[2], pteg[3]);
227 dprintk_mmu("KVM: %08x - %08x\n", pteg[4], pteg[5]);
228 dprintk_mmu("KVM: %08x - %08x\n", pteg[6], pteg[7]);
229 dprintk_mmu("KVM: %08x - %08x\n", pteg[8], pteg[9]);
230 dprintk_mmu("KVM: %08x - %08x\n", pteg[10], pteg[11]);
231 dprintk_mmu("KVM: %08x - %08x\n", pteg[12], pteg[13]);
232 dprintk_mmu("KVM: %08x - %08x\n", pteg[14], pteg[15]);
235 /* Now tell our Shadow PTE code about the new page */
237 pte = kvmppc_mmu_hpte_cache_next(vcpu);
239 dprintk_mmu("KVM: %c%c Map 0x%llx: [%lx] 0x%llx (0x%llx) -> %lx\n",
240 orig_pte->may_write ? 'w' : '-',
241 orig_pte->may_execute ? 'x' : '-',
242 orig_pte->eaddr, (ulong)pteg, va,
243 orig_pte->vpage, hpaddr);
245 pte->slot = (ulong)&pteg[rr];
246 pte->host_va = va;
247 pte->pte = *orig_pte;
248 pte->pfn = hpaddr >> PAGE_SHIFT;
250 kvmppc_mmu_hpte_cache_map(vcpu, pte);
252 return 0;
255 static struct kvmppc_sid_map *create_sid_map(struct kvm_vcpu *vcpu, u64 gvsid)
257 struct kvmppc_sid_map *map;
258 struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
259 u16 sid_map_mask;
260 static int backwards_map = 0;
262 if (vcpu->arch.shared->msr & MSR_PR)
263 gvsid |= VSID_PR;
265 /* We might get collisions that trap in preceding order, so let's
266 map them differently */
268 sid_map_mask = kvmppc_sid_hash(vcpu, gvsid);
269 if (backwards_map)
270 sid_map_mask = SID_MAP_MASK - sid_map_mask;
272 map = &to_book3s(vcpu)->sid_map[sid_map_mask];
274 /* Make sure we're taking the other map next time */
275 backwards_map = !backwards_map;
277 /* Uh-oh ... out of mappings. Let's flush! */
278 if (vcpu_book3s->vsid_next >= VSID_POOL_SIZE) {
279 vcpu_book3s->vsid_next = 0;
280 memset(vcpu_book3s->sid_map, 0,
281 sizeof(struct kvmppc_sid_map) * SID_MAP_NUM);
282 kvmppc_mmu_pte_flush(vcpu, 0, 0);
283 kvmppc_mmu_flush_segments(vcpu);
285 map->host_vsid = vcpu_book3s->vsid_pool[vcpu_book3s->vsid_next];
286 vcpu_book3s->vsid_next++;
288 map->guest_vsid = gvsid;
289 map->valid = true;
291 return map;
294 int kvmppc_mmu_map_segment(struct kvm_vcpu *vcpu, ulong eaddr)
296 u32 esid = eaddr >> SID_SHIFT;
297 u64 gvsid;
298 u32 sr;
299 struct kvmppc_sid_map *map;
300 struct kvmppc_book3s_shadow_vcpu *svcpu = to_svcpu(vcpu);
302 if (vcpu->arch.mmu.esid_to_vsid(vcpu, esid, &gvsid)) {
303 /* Invalidate an entry */
304 svcpu->sr[esid] = SR_INVALID;
305 return -ENOENT;
308 map = find_sid_vsid(vcpu, gvsid);
309 if (!map)
310 map = create_sid_map(vcpu, gvsid);
312 map->guest_esid = esid;
313 sr = map->host_vsid | SR_KP;
314 svcpu->sr[esid] = sr;
316 dprintk_sr("MMU: mtsr %d, 0x%x\n", esid, sr);
318 return 0;
321 void kvmppc_mmu_flush_segments(struct kvm_vcpu *vcpu)
323 int i;
324 struct kvmppc_book3s_shadow_vcpu *svcpu = to_svcpu(vcpu);
326 dprintk_sr("MMU: flushing all segments (%d)\n", ARRAY_SIZE(svcpu->sr));
327 for (i = 0; i < ARRAY_SIZE(svcpu->sr); i++)
328 svcpu->sr[i] = SR_INVALID;
331 void kvmppc_mmu_destroy(struct kvm_vcpu *vcpu)
333 int i;
335 kvmppc_mmu_hpte_destroy(vcpu);
336 preempt_disable();
337 for (i = 0; i < SID_CONTEXTS; i++)
338 __destroy_context(to_book3s(vcpu)->context_id[i]);
339 preempt_enable();
342 /* From mm/mmu_context_hash32.c */
343 #define CTX_TO_VSID(c, id) ((((c) * (897 * 16)) + (id * 0x111)) & 0xffffff)
345 int kvmppc_mmu_init(struct kvm_vcpu *vcpu)
347 struct kvmppc_vcpu_book3s *vcpu3s = to_book3s(vcpu);
348 int err;
349 ulong sdr1;
350 int i;
351 int j;
353 for (i = 0; i < SID_CONTEXTS; i++) {
354 err = __init_new_context();
355 if (err < 0)
356 goto init_fail;
357 vcpu3s->context_id[i] = err;
359 /* Remember context id for this combination */
360 for (j = 0; j < 16; j++)
361 vcpu3s->vsid_pool[(i * 16) + j] = CTX_TO_VSID(err, j);
364 vcpu3s->vsid_next = 0;
366 /* Remember where the HTAB is */
367 asm ( "mfsdr1 %0" : "=r"(sdr1) );
368 htabmask = ((sdr1 & 0x1FF) << 16) | 0xFFC0;
369 htab = (ulong)__va(sdr1 & 0xffff0000);
371 kvmppc_mmu_hpte_init(vcpu);
373 return 0;
375 init_fail:
376 for (j = 0; j < i; j++) {
377 if (!vcpu3s->context_id[j])
378 continue;
380 __destroy_context(to_book3s(vcpu)->context_id[j]);
383 return -1;