2 * This program is free software; you can redistribute it and/or modify
3 * it under the terms of the GNU General Public License, version 2, as
4 * published by the Free Software Foundation.
6 * This program is distributed in the hope that it will be useful,
7 * but WITHOUT ANY WARRANTY; without even the implied warranty of
8 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
9 * GNU General Public License for more details.
11 * You should have received a copy of the GNU General Public License
12 * along with this program; if not, write to the Free Software
13 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
15 * Copyright SUSE Linux Products GmbH 2009
20 #include <linux/types.h>
21 #include <linux/string.h>
22 #include <linux/kvm.h>
23 #include <linux/kvm_host.h>
24 #include <linux/highmem.h>
26 #include <asm/kvm_ppc.h>
27 #include <asm/kvm_book3s.h>
28 #include <asm/book3s/64/mmu-hash.h>
30 /* #define DEBUG_MMU */
33 #define dprintk(X...) printk(KERN_INFO X)
35 #define dprintk(X...) do { } while(0)
38 static void kvmppc_mmu_book3s_64_reset_msr(struct kvm_vcpu *vcpu)
40 unsigned long msr = vcpu->arch.intr_msr;
41 unsigned long cur_msr = kvmppc_get_msr(vcpu);
43 /* If transactional, change to suspend mode on IRQ delivery */
44 if (MSR_TM_TRANSACTIONAL(cur_msr))
47 msr |= cur_msr & MSR_TS_MASK;
49 kvmppc_set_msr(vcpu, msr);
52 static struct kvmppc_slb *kvmppc_mmu_book3s_64_find_slbe(
53 struct kvm_vcpu *vcpu,
57 u64 esid = GET_ESID(eaddr);
58 u64 esid_1t = GET_ESID_1T(eaddr);
60 for (i = 0; i < vcpu->arch.slb_nr; i++) {
63 if (!vcpu->arch.slb[i].valid)
66 if (vcpu->arch.slb[i].tb)
69 if (vcpu->arch.slb[i].esid == cmp_esid)
70 return &vcpu->arch.slb[i];
73 dprintk("KVM: No SLB entry found for 0x%lx [%llx | %llx]\n",
74 eaddr, esid, esid_1t);
75 for (i = 0; i < vcpu->arch.slb_nr; i++) {
76 if (vcpu->arch.slb[i].vsid)
77 dprintk(" %d: %c%c%c %llx %llx\n", i,
78 vcpu->arch.slb[i].valid ? 'v' : ' ',
79 vcpu->arch.slb[i].large ? 'l' : ' ',
80 vcpu->arch.slb[i].tb ? 't' : ' ',
81 vcpu->arch.slb[i].esid,
82 vcpu->arch.slb[i].vsid);
88 static int kvmppc_slb_sid_shift(struct kvmppc_slb *slbe)
90 return slbe->tb ? SID_SHIFT_1T : SID_SHIFT;
93 static u64 kvmppc_slb_offset_mask(struct kvmppc_slb *slbe)
95 return (1ul << kvmppc_slb_sid_shift(slbe)) - 1;
98 static u64 kvmppc_slb_calc_vpn(struct kvmppc_slb *slb, gva_t eaddr)
100 eaddr &= kvmppc_slb_offset_mask(slb);
102 return (eaddr >> VPN_SHIFT) |
103 ((slb->vsid) << (kvmppc_slb_sid_shift(slb) - VPN_SHIFT));
106 static u64 kvmppc_mmu_book3s_64_ea_to_vp(struct kvm_vcpu *vcpu, gva_t eaddr,
109 struct kvmppc_slb *slb;
111 slb = kvmppc_mmu_book3s_64_find_slbe(vcpu, eaddr);
115 return kvmppc_slb_calc_vpn(slb, eaddr);
118 static int mmu_pagesize(int mmu_pg)
129 static int kvmppc_mmu_book3s_64_get_pagesize(struct kvmppc_slb *slbe)
131 return mmu_pagesize(slbe->base_page_size);
134 static u32 kvmppc_mmu_book3s_64_get_page(struct kvmppc_slb *slbe, gva_t eaddr)
136 int p = kvmppc_mmu_book3s_64_get_pagesize(slbe);
138 return ((eaddr & kvmppc_slb_offset_mask(slbe)) >> p);
141 static hva_t kvmppc_mmu_book3s_64_get_pteg(struct kvm_vcpu *vcpu,
142 struct kvmppc_slb *slbe, gva_t eaddr,
145 struct kvmppc_vcpu_book3s *vcpu_book3s = to_book3s(vcpu);
146 u64 hash, pteg, htabsize;
151 htabsize = ((1 << ((vcpu_book3s->sdr1 & 0x1f) + 11)) - 1);
153 vpn = kvmppc_slb_calc_vpn(slbe, eaddr);
154 ssize = slbe->tb ? MMU_SEGSIZE_1T : MMU_SEGSIZE_256M;
155 hash = hpt_hash(vpn, kvmppc_mmu_book3s_64_get_pagesize(slbe), ssize);
158 hash &= ((1ULL << 39ULL) - 1ULL);
162 pteg = vcpu_book3s->sdr1 & 0xfffffffffffc0000ULL;
165 dprintk("MMU: page=0x%x sdr1=0x%llx pteg=0x%llx vsid=0x%llx\n",
166 page, vcpu_book3s->sdr1, pteg, slbe->vsid);
168 /* When running a PAPR guest, SDR1 contains a HVA address instead
170 if (vcpu->arch.papr_enabled)
173 r = gfn_to_hva(vcpu->kvm, pteg >> PAGE_SHIFT);
175 if (kvm_is_error_hva(r))
177 return r | (pteg & ~PAGE_MASK);
180 static u64 kvmppc_mmu_book3s_64_get_avpn(struct kvmppc_slb *slbe, gva_t eaddr)
182 int p = kvmppc_mmu_book3s_64_get_pagesize(slbe);
185 avpn = kvmppc_mmu_book3s_64_get_page(slbe, eaddr);
186 avpn |= slbe->vsid << (kvmppc_slb_sid_shift(slbe) - p);
189 avpn >>= ((80 - p) - 56) - 8; /* 16 - p */
197 * Return page size encoded in the second word of a HPTE, or
198 * -1 for an invalid encoding for the base page size indicated by
199 * the SLB entry. This doesn't handle mixed pagesize segments yet.
201 static int decode_pagesize(struct kvmppc_slb *slbe, u64 r)
203 switch (slbe->base_page_size) {
205 if ((r & 0xf000) == 0x1000)
209 if ((r & 0xff000) == 0)
216 static int kvmppc_mmu_book3s_64_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
217 struct kvmppc_pte *gpte, bool data,
220 struct kvmppc_slb *slbe;
232 ulong mp_ea = vcpu->arch.magic_page_ea;
234 /* Magic page override */
235 if (unlikely(mp_ea) &&
236 unlikely((eaddr & ~0xfffULL) == (mp_ea & ~0xfffULL)) &&
237 !(kvmppc_get_msr(vcpu) & MSR_PR)) {
239 gpte->vpage = kvmppc_mmu_book3s_64_ea_to_vp(vcpu, eaddr, data);
240 gpte->raddr = vcpu->arch.magic_page_pa | (gpte->raddr & 0xfff);
241 gpte->raddr &= KVM_PAM;
242 gpte->may_execute = true;
243 gpte->may_read = true;
244 gpte->may_write = true;
245 gpte->page_size = MMU_PAGE_4K;
246 gpte->wimg = HPTE_R_M;
251 slbe = kvmppc_mmu_book3s_64_find_slbe(vcpu, eaddr);
255 avpn = kvmppc_mmu_book3s_64_get_avpn(slbe, eaddr);
256 v_val = avpn & HPTE_V_AVPN;
259 v_val |= SLB_VSID_B_1T;
261 v_val |= HPTE_V_LARGE;
262 v_val |= HPTE_V_VALID;
264 v_mask = SLB_VSID_B | HPTE_V_AVPN | HPTE_V_LARGE | HPTE_V_VALID |
267 pgsize = slbe->large ? MMU_PAGE_16M : MMU_PAGE_4K;
269 mutex_lock(&vcpu->kvm->arch.hpt_mutex);
272 ptegp = kvmppc_mmu_book3s_64_get_pteg(vcpu, slbe, eaddr, second);
273 if (kvm_is_error_hva(ptegp))
276 if(copy_from_user(pteg, (void __user *)ptegp, sizeof(pteg))) {
277 printk_ratelimited(KERN_ERR
278 "KVM: Can't copy data from 0x%lx!\n", ptegp);
282 if ((kvmppc_get_msr(vcpu) & MSR_PR) && slbe->Kp)
284 else if (!(kvmppc_get_msr(vcpu) & MSR_PR) && slbe->Ks)
287 for (i=0; i<16; i+=2) {
288 u64 pte0 = be64_to_cpu(pteg[i]);
289 u64 pte1 = be64_to_cpu(pteg[i + 1]);
291 /* Check all relevant fields of 1st dword */
292 if ((pte0 & v_mask) == v_val) {
293 /* If large page bit is set, check pgsize encoding */
295 (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE)) {
296 pgsize = decode_pagesize(slbe, pte1);
308 v_val |= HPTE_V_SECONDARY;
313 v = be64_to_cpu(pteg[i]);
314 r = be64_to_cpu(pteg[i+1]);
315 pp = (r & HPTE_R_PP) | key;
320 gpte->vpage = kvmppc_mmu_book3s_64_ea_to_vp(vcpu, eaddr, data);
322 eaddr_mask = (1ull << mmu_pagesize(pgsize)) - 1;
323 gpte->raddr = (r & HPTE_R_RPN & ~eaddr_mask) | (eaddr & eaddr_mask);
324 gpte->page_size = pgsize;
325 gpte->may_execute = ((r & HPTE_R_N) ? false : true);
326 if (unlikely(vcpu->arch.disable_kernel_nx) &&
327 !(kvmppc_get_msr(vcpu) & MSR_PR))
328 gpte->may_execute = true;
329 gpte->may_read = false;
330 gpte->may_write = false;
331 gpte->wimg = r & HPTE_R_WIMG;
338 gpte->may_write = true;
344 gpte->may_read = true;
348 dprintk("KVM MMU: Translated 0x%lx [0x%llx] -> 0x%llx "
350 eaddr, avpn, gpte->vpage, gpte->raddr);
352 /* Update PTE R and C bits, so the guest's swapper knows we used the
354 if (gpte->may_read && !(r & HPTE_R_R)) {
356 * Set the accessed flag.
357 * We have to write this back with a single byte write
358 * because another vcpu may be accessing this on
359 * non-PAPR platforms such as mac99, and this is
360 * what real hardware does.
362 char __user *addr = (char __user *) (ptegp + (i + 1) * sizeof(u64));
364 put_user(r >> 8, addr + 6);
366 if (iswrite && gpte->may_write && !(r & HPTE_R_C)) {
367 /* Set the dirty flag */
368 /* Use a single byte write */
369 char __user *addr = (char __user *) (ptegp + (i + 1) * sizeof(u64));
371 put_user(r, addr + 7);
374 mutex_unlock(&vcpu->kvm->arch.hpt_mutex);
376 if (!gpte->may_read || (iswrite && !gpte->may_write))
381 mutex_unlock(&vcpu->kvm->arch.hpt_mutex);
385 dprintk("KVM MMU: Trigger segment fault\n");
389 static void kvmppc_mmu_book3s_64_slbmte(struct kvm_vcpu *vcpu, u64 rs, u64 rb)
393 struct kvmppc_slb *slbe;
395 dprintk("KVM MMU: slbmte(0x%llx, 0x%llx)\n", rs, rb);
398 esid_1t = GET_ESID_1T(rb);
401 if (slb_nr > vcpu->arch.slb_nr)
404 slbe = &vcpu->arch.slb[slb_nr];
406 slbe->large = (rs & SLB_VSID_L) ? 1 : 0;
407 slbe->tb = (rs & SLB_VSID_B_1T) ? 1 : 0;
408 slbe->esid = slbe->tb ? esid_1t : esid;
409 slbe->vsid = (rs & ~SLB_VSID_B) >> (kvmppc_slb_sid_shift(slbe) - 16);
410 slbe->valid = (rb & SLB_ESID_V) ? 1 : 0;
411 slbe->Ks = (rs & SLB_VSID_KS) ? 1 : 0;
412 slbe->Kp = (rs & SLB_VSID_KP) ? 1 : 0;
413 slbe->nx = (rs & SLB_VSID_N) ? 1 : 0;
414 slbe->class = (rs & SLB_VSID_C) ? 1 : 0;
416 slbe->base_page_size = MMU_PAGE_4K;
418 if (vcpu->arch.hflags & BOOK3S_HFLAG_MULTI_PGSIZE) {
419 switch (rs & SLB_VSID_LP) {
421 slbe->base_page_size = MMU_PAGE_16M;
424 slbe->base_page_size = MMU_PAGE_64K;
428 slbe->base_page_size = MMU_PAGE_16M;
431 slbe->orige = rb & (ESID_MASK | SLB_ESID_V);
434 /* Map the new segment */
435 kvmppc_mmu_map_segment(vcpu, esid << SID_SHIFT);
438 static u64 kvmppc_mmu_book3s_64_slbmfee(struct kvm_vcpu *vcpu, u64 slb_nr)
440 struct kvmppc_slb *slbe;
442 if (slb_nr > vcpu->arch.slb_nr)
445 slbe = &vcpu->arch.slb[slb_nr];
450 static u64 kvmppc_mmu_book3s_64_slbmfev(struct kvm_vcpu *vcpu, u64 slb_nr)
452 struct kvmppc_slb *slbe;
454 if (slb_nr > vcpu->arch.slb_nr)
457 slbe = &vcpu->arch.slb[slb_nr];
462 static void kvmppc_mmu_book3s_64_slbie(struct kvm_vcpu *vcpu, u64 ea)
464 struct kvmppc_slb *slbe;
467 dprintk("KVM MMU: slbie(0x%llx)\n", ea);
469 slbe = kvmppc_mmu_book3s_64_find_slbe(vcpu, ea);
474 dprintk("KVM MMU: slbie(0x%llx, 0x%llx)\n", ea, slbe->esid);
480 seg_size = 1ull << kvmppc_slb_sid_shift(slbe);
481 kvmppc_mmu_flush_segment(vcpu, ea & ~(seg_size - 1), seg_size);
484 static void kvmppc_mmu_book3s_64_slbia(struct kvm_vcpu *vcpu)
488 dprintk("KVM MMU: slbia()\n");
490 for (i = 1; i < vcpu->arch.slb_nr; i++) {
491 vcpu->arch.slb[i].valid = false;
492 vcpu->arch.slb[i].orige = 0;
493 vcpu->arch.slb[i].origv = 0;
496 if (kvmppc_get_msr(vcpu) & MSR_IR) {
497 kvmppc_mmu_flush_segments(vcpu);
498 kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
502 static void kvmppc_mmu_book3s_64_mtsrin(struct kvm_vcpu *vcpu, u32 srnum,
508 * According to Book3 2.01 mtsrin is implemented as:
510 * The SLB entry specified by (RB)32:35 is loaded from register
513 * SLBE Bit Source SLB Field
515 * 0:31 0x0000_0000 ESID-0:31
516 * 32:35 (RB)32:35 ESID-32:35
518 * 37:61 0x00_0000|| 0b0 VSID-0:24
519 * 62:88 (RS)37:63 VSID-25:51
520 * 89:91 (RS)33:35 Ks Kp N
521 * 92 (RS)36 L ((RS)36 must be 0b0)
525 dprintk("KVM MMU: mtsrin(0x%x, 0x%lx)\n", srnum, value);
528 rb |= (srnum & 0xf) << 28;
529 /* Set the valid bit */
535 rs |= (value & 0xfffffff) << 12;
537 rs |= ((value >> 28) & 0x7) << 9;
539 kvmppc_mmu_book3s_64_slbmte(vcpu, rs, rb);
542 static void kvmppc_mmu_book3s_64_tlbie(struct kvm_vcpu *vcpu, ulong va,
545 u64 mask = 0xFFFFFFFFFULL;
549 dprintk("KVM MMU: tlbie(0x%lx)\n", va);
552 * The tlbie instruction changed behaviour starting with
553 * POWER6. POWER6 and later don't have the large page flag
554 * in the instruction but in the RB value, along with bits
555 * indicating page and segment sizes.
557 if (vcpu->arch.hflags & BOOK3S_HFLAG_NEW_TLBIE) {
558 /* POWER6 or later */
559 if (va & 1) { /* L bit */
560 if ((va & 0xf000) == 0x1000)
561 mask = 0xFFFFFFFF0ULL; /* 64k page */
563 mask = 0xFFFFFF000ULL; /* 16M page */
566 /* older processors, e.g. PPC970 */
568 mask = 0xFFFFFF000ULL;
570 /* flush this VA on all vcpus */
571 kvm_for_each_vcpu(i, v, vcpu->kvm)
572 kvmppc_mmu_pte_vflush(v, va >> 12, mask);
575 #ifdef CONFIG_PPC_64K_PAGES
576 static int segment_contains_magic_page(struct kvm_vcpu *vcpu, ulong esid)
578 ulong mp_ea = vcpu->arch.magic_page_ea;
580 return mp_ea && !(kvmppc_get_msr(vcpu) & MSR_PR) &&
581 (mp_ea >> SID_SHIFT) == esid;
585 static int kvmppc_mmu_book3s_64_esid_to_vsid(struct kvm_vcpu *vcpu, ulong esid,
588 ulong ea = esid << SID_SHIFT;
589 struct kvmppc_slb *slb;
591 ulong mp_ea = vcpu->arch.magic_page_ea;
592 int pagesize = MMU_PAGE_64K;
593 u64 msr = kvmppc_get_msr(vcpu);
595 if (msr & (MSR_DR|MSR_IR)) {
596 slb = kvmppc_mmu_book3s_64_find_slbe(vcpu, ea);
599 pagesize = slb->base_page_size;
601 gvsid <<= SID_SHIFT_1T - SID_SHIFT;
602 gvsid |= esid & ((1ul << (SID_SHIFT_1T - SID_SHIFT)) - 1);
608 switch (msr & (MSR_DR|MSR_IR)) {
610 gvsid = VSID_REAL | esid;
613 gvsid |= VSID_REAL_IR;
616 gvsid |= VSID_REAL_DR;
628 #ifdef CONFIG_PPC_64K_PAGES
630 * Mark this as a 64k segment if the host is using
631 * 64k pages, the host MMU supports 64k pages and
632 * the guest segment page size is >= 64k,
633 * but not if this segment contains the magic page.
635 if (pagesize >= MMU_PAGE_64K &&
636 mmu_psize_defs[MMU_PAGE_64K].shift &&
637 !segment_contains_magic_page(vcpu, esid))
641 if (kvmppc_get_msr(vcpu) & MSR_PR)
648 /* Catch magic page case */
649 if (unlikely(mp_ea) &&
650 unlikely(esid == (mp_ea >> SID_SHIFT)) &&
651 !(kvmppc_get_msr(vcpu) & MSR_PR)) {
652 *vsid = VSID_REAL | esid;
659 static bool kvmppc_mmu_book3s_64_is_dcbz32(struct kvm_vcpu *vcpu)
661 return (to_book3s(vcpu)->hid[5] & 0x80);
664 void kvmppc_mmu_book3s_64_init(struct kvm_vcpu *vcpu)
666 struct kvmppc_mmu *mmu = &vcpu->arch.mmu;
669 mmu->mtsrin = kvmppc_mmu_book3s_64_mtsrin;
670 mmu->slbmte = kvmppc_mmu_book3s_64_slbmte;
671 mmu->slbmfee = kvmppc_mmu_book3s_64_slbmfee;
672 mmu->slbmfev = kvmppc_mmu_book3s_64_slbmfev;
673 mmu->slbie = kvmppc_mmu_book3s_64_slbie;
674 mmu->slbia = kvmppc_mmu_book3s_64_slbia;
675 mmu->xlate = kvmppc_mmu_book3s_64_xlate;
676 mmu->reset_msr = kvmppc_mmu_book3s_64_reset_msr;
677 mmu->tlbie = kvmppc_mmu_book3s_64_tlbie;
678 mmu->esid_to_vsid = kvmppc_mmu_book3s_64_esid_to_vsid;
679 mmu->ea_to_vp = kvmppc_mmu_book3s_64_ea_to_vp;
680 mmu->is_dcbz32 = kvmppc_mmu_book3s_64_is_dcbz32;
682 vcpu->arch.hflags |= BOOK3S_HFLAG_SLB;