2 * Copyright (C) 2008-2011 Freescale Semiconductor, Inc. All rights reserved.
4 * Author: Yu Liu, yu.liu@freescale.com
7 * This file is based on arch/powerpc/kvm/44x_tlb.c,
8 * by Hollis Blanchard <hollisb@us.ibm.com>.
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License, version 2, as
12 * published by the Free Software Foundation.
15 #include <linux/kernel.h>
16 #include <linux/types.h>
17 #include <linux/slab.h>
18 #include <linux/string.h>
19 #include <linux/kvm.h>
20 #include <linux/kvm_host.h>
21 #include <linux/highmem.h>
22 #include <linux/log2.h>
23 #include <linux/uaccess.h>
24 #include <linux/sched.h>
25 #include <linux/rwsem.h>
26 #include <linux/vmalloc.h>
27 #include <linux/hugetlb.h>
28 #include <asm/kvm_ppc.h>
29 #include <asm/kvm_e500.h>
31 #include "../mm/mmu_decl.h"
36 #define to_htlb1_esel(esel) (host_tlb_params[1].entries - (esel) - 1)
46 * This table provide mappings from:
47 * (guestAS,guestTID,guestPR) --> ID of physical cpu
52 * Each vcpu keeps one vcpu_id_table.
54 struct vcpu_id_table {
55 struct id id[2][NUM_TIDS][2];
59 * This table provide reversed mappings of vcpu_id_table:
60 * ID --> address of vcpu_id_table item.
61 * Each physical core has one pcpu_id_table.
63 struct pcpu_id_table {
64 struct id *entry[NUM_TIDS];
67 static DEFINE_PER_CPU(struct pcpu_id_table, pcpu_sids);
69 /* This variable keeps last used shadow ID on local core.
70 * The valid range of shadow ID is [1..255] */
71 static DEFINE_PER_CPU(unsigned long, pcpu_last_used_sid);
73 static struct kvmppc_e500_tlb_params host_tlb_params[E500_TLB_NUM];
75 static struct kvm_book3e_206_tlb_entry *get_entry(
76 struct kvmppc_vcpu_e500 *vcpu_e500, int tlbsel, int entry)
78 int offset = vcpu_e500->gtlb_offset[tlbsel];
79 return &vcpu_e500->gtlb_arch[offset + entry];
83 * Allocate a free shadow id and setup a valid sid mapping in given entry.
84 * A mapping is only valid when vcpu_id_table and pcpu_id_table are match.
86 * The caller must have preemption disabled, and keep it that way until
87 * it has finished with the returned shadow id (either written into the
88 * TLB or arch.shadow_pid, or discarded).
90 static inline int local_sid_setup_one(struct id *entry)
95 sid = ++(__get_cpu_var(pcpu_last_used_sid));
97 __get_cpu_var(pcpu_sids).entry[sid] = entry;
99 entry->pentry = &__get_cpu_var(pcpu_sids).entry[sid];
104 * If sid == NUM_TIDS, we've run out of sids. We return -1, and
105 * the caller will invalidate everything and start over.
107 * sid > NUM_TIDS indicates a race, which we disable preemption to
110 WARN_ON(sid > NUM_TIDS);
116 * Check if given entry contain a valid shadow id mapping.
117 * An ID mapping is considered valid only if
118 * both vcpu and pcpu know this mapping.
120 * The caller must have preemption disabled, and keep it that way until
121 * it has finished with the returned shadow id (either written into the
122 * TLB or arch.shadow_pid, or discarded).
124 static inline int local_sid_lookup(struct id *entry)
126 if (entry && entry->val != 0 &&
127 __get_cpu_var(pcpu_sids).entry[entry->val] == entry &&
128 entry->pentry == &__get_cpu_var(pcpu_sids).entry[entry->val])
133 /* Invalidate all id mappings on local core -- call with preempt disabled */
134 static inline void local_sid_destroy_all(void)
136 __get_cpu_var(pcpu_last_used_sid) = 0;
137 memset(&__get_cpu_var(pcpu_sids), 0, sizeof(__get_cpu_var(pcpu_sids)));
140 static void *kvmppc_e500_id_table_alloc(struct kvmppc_vcpu_e500 *vcpu_e500)
142 vcpu_e500->idt = kzalloc(sizeof(struct vcpu_id_table), GFP_KERNEL);
143 return vcpu_e500->idt;
146 static void kvmppc_e500_id_table_free(struct kvmppc_vcpu_e500 *vcpu_e500)
148 kfree(vcpu_e500->idt);
151 /* Invalidate all mappings on vcpu */
152 static void kvmppc_e500_id_table_reset_all(struct kvmppc_vcpu_e500 *vcpu_e500)
154 memset(vcpu_e500->idt, 0, sizeof(struct vcpu_id_table));
156 /* Update shadow pid when mappings are changed */
157 kvmppc_e500_recalc_shadow_pid(vcpu_e500);
160 /* Invalidate one ID mapping on vcpu */
161 static inline void kvmppc_e500_id_table_reset_one(
162 struct kvmppc_vcpu_e500 *vcpu_e500,
163 int as, int pid, int pr)
165 struct vcpu_id_table *idt = vcpu_e500->idt;
168 BUG_ON(pid >= NUM_TIDS);
171 idt->id[as][pid][pr].val = 0;
172 idt->id[as][pid][pr].pentry = NULL;
174 /* Update shadow pid when mappings are changed */
175 kvmppc_e500_recalc_shadow_pid(vcpu_e500);
179 * Map guest (vcpu,AS,ID,PR) to physical core shadow id.
180 * This function first lookup if a valid mapping exists,
181 * if not, then creates a new one.
183 * The caller must have preemption disabled, and keep it that way until
184 * it has finished with the returned shadow id (either written into the
185 * TLB or arch.shadow_pid, or discarded).
187 static unsigned int kvmppc_e500_get_sid(struct kvmppc_vcpu_e500 *vcpu_e500,
188 unsigned int as, unsigned int gid,
189 unsigned int pr, int avoid_recursion)
191 struct vcpu_id_table *idt = vcpu_e500->idt;
195 BUG_ON(gid >= NUM_TIDS);
198 sid = local_sid_lookup(&idt->id[as][gid][pr]);
202 sid = local_sid_setup_one(&idt->id[as][gid][pr]);
205 local_sid_destroy_all();
208 /* Update shadow pid when mappings are changed */
209 if (!avoid_recursion)
210 kvmppc_e500_recalc_shadow_pid(vcpu_e500);
216 /* Map guest pid to shadow.
217 * We use PID to keep shadow of current guest non-zero PID,
218 * and use PID1 to keep shadow of guest zero PID.
219 * So that guest tlbe with TID=0 can be accessed at any time */
220 void kvmppc_e500_recalc_shadow_pid(struct kvmppc_vcpu_e500 *vcpu_e500)
223 vcpu_e500->vcpu.arch.shadow_pid = kvmppc_e500_get_sid(vcpu_e500,
224 get_cur_as(&vcpu_e500->vcpu),
225 get_cur_pid(&vcpu_e500->vcpu),
226 get_cur_pr(&vcpu_e500->vcpu), 1);
227 vcpu_e500->vcpu.arch.shadow_pid1 = kvmppc_e500_get_sid(vcpu_e500,
228 get_cur_as(&vcpu_e500->vcpu), 0,
229 get_cur_pr(&vcpu_e500->vcpu), 1);
233 static inline unsigned int gtlb0_get_next_victim(
234 struct kvmppc_vcpu_e500 *vcpu_e500)
238 victim = vcpu_e500->gtlb_nv[0]++;
239 if (unlikely(vcpu_e500->gtlb_nv[0] >= vcpu_e500->gtlb_params[0].ways))
240 vcpu_e500->gtlb_nv[0] = 0;
245 static inline unsigned int tlb1_max_shadow_size(void)
247 /* reserve one entry for magic page */
248 return host_tlb_params[1].entries - tlbcam_index - 1;
251 static inline int tlbe_is_writable(struct kvm_book3e_206_tlb_entry *tlbe)
253 return tlbe->mas7_3 & (MAS3_SW|MAS3_UW);
256 static inline u32 e500_shadow_mas3_attrib(u32 mas3, int usermode)
258 /* Mask off reserved bits. */
259 mas3 &= MAS3_ATTRIB_MASK;
262 /* Guest is in supervisor mode,
263 * so we need to translate guest
264 * supervisor permissions into user permissions. */
265 mas3 &= ~E500_TLB_USER_PERM_MASK;
266 mas3 |= (mas3 & E500_TLB_SUPER_PERM_MASK) << 1;
269 return mas3 | E500_TLB_SUPER_PERM_MASK;
272 static inline u32 e500_shadow_mas2_attrib(u32 mas2, int usermode)
275 return (mas2 & MAS2_ATTRIB_MASK) | MAS2_M;
277 return mas2 & MAS2_ATTRIB_MASK;
282 * writing shadow tlb entry to host TLB
284 static inline void __write_host_tlbe(struct kvm_book3e_206_tlb_entry *stlbe,
289 local_irq_save(flags);
290 mtspr(SPRN_MAS0, mas0);
291 mtspr(SPRN_MAS1, stlbe->mas1);
292 mtspr(SPRN_MAS2, (unsigned long)stlbe->mas2);
293 mtspr(SPRN_MAS3, (u32)stlbe->mas7_3);
294 mtspr(SPRN_MAS7, (u32)(stlbe->mas7_3 >> 32));
295 asm volatile("isync; tlbwe" : : : "memory");
296 local_irq_restore(flags);
299 /* esel is index into set, not whole array */
300 static inline void write_host_tlbe(struct kvmppc_vcpu_e500 *vcpu_e500,
301 int tlbsel, int esel, struct kvm_book3e_206_tlb_entry *stlbe)
304 int way = esel & (vcpu_e500->gtlb_params[0].ways - 1);
305 __write_host_tlbe(stlbe, MAS0_TLBSEL(0) | MAS0_ESEL(way));
307 __write_host_tlbe(stlbe,
309 MAS0_ESEL(to_htlb1_esel(esel)));
311 trace_kvm_stlb_write(index_of(tlbsel, esel), stlbe->mas1, stlbe->mas2,
312 (u32)stlbe->mas7_3, (u32)(stlbe->mas7_3 >> 32));
315 void kvmppc_map_magic(struct kvm_vcpu *vcpu)
317 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
318 struct kvm_book3e_206_tlb_entry magic;
319 ulong shared_page = ((ulong)vcpu->arch.shared) & PAGE_MASK;
323 pfn = (pfn_t)virt_to_phys((void *)shared_page) >> PAGE_SHIFT;
324 get_page(pfn_to_page(pfn));
327 stid = kvmppc_e500_get_sid(vcpu_e500, 0, 0, 0, 0);
329 magic.mas1 = MAS1_VALID | MAS1_TS | MAS1_TID(stid) |
330 MAS1_TSIZE(BOOK3E_PAGESZ_4K);
331 magic.mas2 = vcpu->arch.magic_page_ea | MAS2_M;
332 magic.mas7_3 = ((u64)pfn << PAGE_SHIFT) |
333 MAS3_SW | MAS3_SR | MAS3_UW | MAS3_UR;
335 __write_host_tlbe(&magic, MAS0_TLBSEL(1) | MAS0_ESEL(tlbcam_index));
339 void kvmppc_e500_tlb_load(struct kvm_vcpu *vcpu, int cpu)
341 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
343 /* Shadow PID may be expired on local core */
344 kvmppc_e500_recalc_shadow_pid(vcpu_e500);
347 void kvmppc_e500_tlb_put(struct kvm_vcpu *vcpu)
351 static void inval_gtlbe_on_host(struct kvmppc_vcpu_e500 *vcpu_e500,
352 int tlbsel, int esel)
354 struct kvm_book3e_206_tlb_entry *gtlbe =
355 get_entry(vcpu_e500, tlbsel, esel);
356 struct vcpu_id_table *idt = vcpu_e500->idt;
357 unsigned int pr, tid, ts, pid;
361 ts = get_tlb_ts(gtlbe);
362 tid = get_tlb_tid(gtlbe);
366 /* One guest ID may be mapped to two shadow IDs */
367 for (pr = 0; pr < 2; pr++) {
369 * The shadow PID can have a valid mapping on at most one
370 * host CPU. In the common case, it will be valid on this
371 * CPU, in which case (for TLB0) we do a local invalidation
372 * of the specific address.
374 * If the shadow PID is not valid on the current host CPU, or
375 * if we're invalidating a TLB1 entry, we invalidate the
379 (pid = local_sid_lookup(&idt->id[ts][tid][pr])) <= 0) {
380 kvmppc_e500_id_table_reset_one(vcpu_e500, ts, tid, pr);
385 * The guest is invalidating a TLB0 entry which is in a PID
386 * that has a valid shadow mapping on this host CPU. We
387 * search host TLB0 to invalidate it's shadow TLB entry,
388 * similar to __tlbil_va except that we need to look in AS1.
390 val = (pid << MAS6_SPID_SHIFT) | MAS6_SAS;
391 eaddr = get_tlb_eaddr(gtlbe);
393 local_irq_save(flags);
395 mtspr(SPRN_MAS6, val);
396 asm volatile("tlbsx 0, %[eaddr]" : : [eaddr] "r" (eaddr));
397 val = mfspr(SPRN_MAS1);
398 if (val & MAS1_VALID) {
399 mtspr(SPRN_MAS1, val & ~MAS1_VALID);
400 asm volatile("tlbwe");
403 local_irq_restore(flags);
409 static int tlb0_set_base(gva_t addr, int sets, int ways)
413 set_base = (addr >> PAGE_SHIFT) & (sets - 1);
419 static int gtlb0_set_base(struct kvmppc_vcpu_e500 *vcpu_e500, gva_t addr)
421 return tlb0_set_base(addr, vcpu_e500->gtlb_params[0].sets,
422 vcpu_e500->gtlb_params[0].ways);
425 static int htlb0_set_base(gva_t addr)
427 return tlb0_set_base(addr, host_tlb_params[0].sets,
428 host_tlb_params[0].ways);
431 static unsigned int get_tlb_esel(struct kvmppc_vcpu_e500 *vcpu_e500, int tlbsel)
433 unsigned int esel = get_tlb_esel_bit(vcpu_e500);
436 esel &= vcpu_e500->gtlb_params[0].ways - 1;
437 esel += gtlb0_set_base(vcpu_e500, vcpu_e500->mas2);
439 esel &= vcpu_e500->gtlb_params[tlbsel].entries - 1;
445 /* Search the guest TLB for a matching entry. */
446 static int kvmppc_e500_tlb_index(struct kvmppc_vcpu_e500 *vcpu_e500,
447 gva_t eaddr, int tlbsel, unsigned int pid, int as)
449 int size = vcpu_e500->gtlb_params[tlbsel].entries;
450 unsigned int set_base, offset;
454 set_base = gtlb0_set_base(vcpu_e500, eaddr);
455 size = vcpu_e500->gtlb_params[0].ways;
460 offset = vcpu_e500->gtlb_offset[tlbsel];
462 for (i = 0; i < size; i++) {
463 struct kvm_book3e_206_tlb_entry *tlbe =
464 &vcpu_e500->gtlb_arch[offset + set_base + i];
467 if (eaddr < get_tlb_eaddr(tlbe))
470 if (eaddr > get_tlb_end(tlbe))
473 tid = get_tlb_tid(tlbe);
474 if (tid && (tid != pid))
477 if (!get_tlb_v(tlbe))
480 if (get_tlb_ts(tlbe) != as && as != -1)
489 static inline void kvmppc_e500_ref_setup(struct tlbe_ref *ref,
490 struct kvm_book3e_206_tlb_entry *gtlbe,
494 ref->flags = E500_TLB_VALID;
496 if (tlbe_is_writable(gtlbe))
497 ref->flags |= E500_TLB_DIRTY;
500 static inline void kvmppc_e500_ref_release(struct tlbe_ref *ref)
502 if (ref->flags & E500_TLB_VALID) {
503 if (ref->flags & E500_TLB_DIRTY)
504 kvm_release_pfn_dirty(ref->pfn);
506 kvm_release_pfn_clean(ref->pfn);
512 static void clear_tlb_privs(struct kvmppc_vcpu_e500 *vcpu_e500)
517 for (i = 0; i < vcpu_e500->gtlb_params[tlbsel].entries; i++) {
518 struct tlbe_ref *ref =
519 &vcpu_e500->gtlb_priv[tlbsel][i].ref;
520 kvmppc_e500_ref_release(ref);
524 static void clear_tlb_refs(struct kvmppc_vcpu_e500 *vcpu_e500)
529 kvmppc_e500_id_table_reset_all(vcpu_e500);
531 for (i = 0; i < host_tlb_params[stlbsel].entries; i++) {
532 struct tlbe_ref *ref =
533 &vcpu_e500->tlb_refs[stlbsel][i];
534 kvmppc_e500_ref_release(ref);
537 clear_tlb_privs(vcpu_e500);
540 static inline void kvmppc_e500_deliver_tlb_miss(struct kvm_vcpu *vcpu,
541 unsigned int eaddr, int as)
543 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
544 unsigned int victim, pidsel, tsized;
547 /* since we only have two TLBs, only lower bit is used. */
548 tlbsel = (vcpu_e500->mas4 >> 28) & 0x1;
549 victim = (tlbsel == 0) ? gtlb0_get_next_victim(vcpu_e500) : 0;
550 pidsel = (vcpu_e500->mas4 >> 16) & 0xf;
551 tsized = (vcpu_e500->mas4 >> 7) & 0x1f;
553 vcpu_e500->mas0 = MAS0_TLBSEL(tlbsel) | MAS0_ESEL(victim)
554 | MAS0_NV(vcpu_e500->gtlb_nv[tlbsel]);
555 vcpu_e500->mas1 = MAS1_VALID | (as ? MAS1_TS : 0)
556 | MAS1_TID(vcpu_e500->pid[pidsel])
557 | MAS1_TSIZE(tsized);
558 vcpu_e500->mas2 = (eaddr & MAS2_EPN)
559 | (vcpu_e500->mas4 & MAS2_ATTRIB_MASK);
560 vcpu_e500->mas7_3 &= MAS3_U0 | MAS3_U1 | MAS3_U2 | MAS3_U3;
561 vcpu_e500->mas6 = (vcpu_e500->mas6 & MAS6_SPID1)
562 | (get_cur_pid(vcpu) << 16)
563 | (as ? MAS6_SAS : 0);
566 /* TID must be supplied by the caller */
567 static inline void kvmppc_e500_setup_stlbe(
568 struct kvmppc_vcpu_e500 *vcpu_e500,
569 struct kvm_book3e_206_tlb_entry *gtlbe,
570 int tsize, struct tlbe_ref *ref, u64 gvaddr,
571 struct kvm_book3e_206_tlb_entry *stlbe)
573 pfn_t pfn = ref->pfn;
575 BUG_ON(!(ref->flags & E500_TLB_VALID));
577 /* Force TS=1 IPROT=0 for all guest mappings. */
578 stlbe->mas1 = MAS1_TSIZE(tsize) | MAS1_TS | MAS1_VALID;
579 stlbe->mas2 = (gvaddr & MAS2_EPN)
580 | e500_shadow_mas2_attrib(gtlbe->mas2,
581 vcpu_e500->vcpu.arch.shared->msr & MSR_PR);
582 stlbe->mas7_3 = ((u64)pfn << PAGE_SHIFT)
583 | e500_shadow_mas3_attrib(gtlbe->mas7_3,
584 vcpu_e500->vcpu.arch.shared->msr & MSR_PR);
587 /* sesel is an index into the entire array, not just the set */
588 static inline void kvmppc_e500_shadow_map(struct kvmppc_vcpu_e500 *vcpu_e500,
589 u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe,
590 int tlbsel, int sesel, struct kvm_book3e_206_tlb_entry *stlbe,
591 struct tlbe_ref *ref)
593 struct kvm_memory_slot *slot;
594 unsigned long pfn, hva;
596 int tsize = BOOK3E_PAGESZ_4K;
599 * Translate guest physical to true physical, acquiring
600 * a page reference if it is normal, non-reserved memory.
602 * gfn_to_memslot() must succeed because otherwise we wouldn't
603 * have gotten this far. Eventually we should just pass the slot
604 * pointer through from the first lookup.
606 slot = gfn_to_memslot(vcpu_e500->vcpu.kvm, gfn);
607 hva = gfn_to_hva_memslot(slot, gfn);
610 struct vm_area_struct *vma;
611 down_read(¤t->mm->mmap_sem);
613 vma = find_vma(current->mm, hva);
614 if (vma && hva >= vma->vm_start &&
615 (vma->vm_flags & VM_PFNMAP)) {
617 * This VMA is a physically contiguous region (e.g.
618 * /dev/mem) that bypasses normal Linux page
619 * management. Find the overlap between the
620 * vma and the memslot.
623 unsigned long start, end;
624 unsigned long slot_start, slot_end;
628 start = vma->vm_pgoff;
630 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT);
632 pfn = start + ((hva - vma->vm_start) >> PAGE_SHIFT);
634 slot_start = pfn - (gfn - slot->base_gfn);
635 slot_end = slot_start + slot->npages;
637 if (start < slot_start)
642 tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >>
646 * e500 doesn't implement the lowest tsize bit,
649 tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1);
652 * Now find the largest tsize (up to what the guest
653 * requested) that will cover gfn, stay within the
654 * range, and for which gfn and pfn are mutually
658 for (; tsize > BOOK3E_PAGESZ_4K; tsize -= 2) {
659 unsigned long gfn_start, gfn_end, tsize_pages;
660 tsize_pages = 1 << (tsize - 2);
662 gfn_start = gfn & ~(tsize_pages - 1);
663 gfn_end = gfn_start + tsize_pages;
665 if (gfn_start + pfn - gfn < start)
667 if (gfn_end + pfn - gfn > end)
669 if ((gfn & (tsize_pages - 1)) !=
670 (pfn & (tsize_pages - 1)))
673 gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1);
674 pfn &= ~(tsize_pages - 1);
677 } else if (vma && hva >= vma->vm_start &&
678 (vma->vm_flags & VM_HUGETLB)) {
679 unsigned long psize = vma_kernel_pagesize(vma);
681 tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >>
685 * Take the largest page size that satisfies both host
688 tsize = min(__ilog2(psize) - 10, tsize);
691 * e500 doesn't implement the lowest tsize bit,
694 tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1);
697 up_read(¤t->mm->mmap_sem);
700 if (likely(!pfnmap)) {
701 unsigned long tsize_pages = 1 << (tsize + 10 - PAGE_SHIFT);
702 pfn = gfn_to_pfn_memslot(vcpu_e500->vcpu.kvm, slot, gfn);
703 if (is_error_pfn(pfn)) {
704 printk(KERN_ERR "Couldn't get real page for gfn %lx!\n",
706 kvm_release_pfn_clean(pfn);
710 /* Align guest and physical address to page map boundaries */
711 pfn &= ~(tsize_pages - 1);
712 gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1);
715 /* Drop old ref and setup new one. */
716 kvmppc_e500_ref_release(ref);
717 kvmppc_e500_ref_setup(ref, gtlbe, pfn);
719 kvmppc_e500_setup_stlbe(vcpu_e500, gtlbe, tsize, ref, gvaddr, stlbe);
722 /* XXX only map the one-one case, for now use TLB0 */
723 static int kvmppc_e500_tlb0_map(struct kvmppc_vcpu_e500 *vcpu_e500,
725 struct kvm_book3e_206_tlb_entry *stlbe)
727 struct kvm_book3e_206_tlb_entry *gtlbe;
728 struct tlbe_ref *ref;
729 int sesel = esel & (host_tlb_params[0].ways - 1);
733 gtlbe = get_entry(vcpu_e500, 0, esel);
734 ref = &vcpu_e500->gtlb_priv[0][esel].ref;
736 ea = get_tlb_eaddr(gtlbe);
737 sesel_base = htlb0_set_base(ea);
739 kvmppc_e500_shadow_map(vcpu_e500, get_tlb_eaddr(gtlbe),
740 get_tlb_raddr(gtlbe) >> PAGE_SHIFT,
741 gtlbe, 0, sesel_base + sesel, stlbe, ref);
746 /* Caller must ensure that the specified guest TLB entry is safe to insert into
748 /* XXX for both one-one and one-to-many , for now use TLB1 */
749 static int kvmppc_e500_tlb1_map(struct kvmppc_vcpu_e500 *vcpu_e500,
750 u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe,
751 struct kvm_book3e_206_tlb_entry *stlbe)
753 struct tlbe_ref *ref;
756 victim = vcpu_e500->host_tlb1_nv++;
758 if (unlikely(vcpu_e500->host_tlb1_nv >= tlb1_max_shadow_size()))
759 vcpu_e500->host_tlb1_nv = 0;
761 ref = &vcpu_e500->tlb_refs[1][victim];
762 kvmppc_e500_shadow_map(vcpu_e500, gvaddr, gfn, gtlbe, 1,
768 void kvmppc_mmu_msr_notify(struct kvm_vcpu *vcpu, u32 old_msr)
770 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
772 /* Recalc shadow pid since MSR changes */
773 kvmppc_e500_recalc_shadow_pid(vcpu_e500);
776 static inline int kvmppc_e500_gtlbe_invalidate(
777 struct kvmppc_vcpu_e500 *vcpu_e500,
778 int tlbsel, int esel)
780 struct kvm_book3e_206_tlb_entry *gtlbe =
781 get_entry(vcpu_e500, tlbsel, esel);
783 if (unlikely(get_tlb_iprot(gtlbe)))
791 int kvmppc_e500_emul_mt_mmucsr0(struct kvmppc_vcpu_e500 *vcpu_e500, ulong value)
795 if (value & MMUCSR0_TLB0FI)
796 for (esel = 0; esel < vcpu_e500->gtlb_params[0].entries; esel++)
797 kvmppc_e500_gtlbe_invalidate(vcpu_e500, 0, esel);
798 if (value & MMUCSR0_TLB1FI)
799 for (esel = 0; esel < vcpu_e500->gtlb_params[1].entries; esel++)
800 kvmppc_e500_gtlbe_invalidate(vcpu_e500, 1, esel);
802 /* Invalidate all vcpu id mappings */
803 kvmppc_e500_id_table_reset_all(vcpu_e500);
808 int kvmppc_e500_emul_tlbivax(struct kvm_vcpu *vcpu, int ra, int rb)
810 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
815 ea = ((ra) ? kvmppc_get_gpr(vcpu, ra) : 0) + kvmppc_get_gpr(vcpu, rb);
817 ia = (ea >> 2) & 0x1;
819 /* since we only have two TLBs, only lower bit is used. */
820 tlbsel = (ea >> 3) & 0x1;
823 /* invalidate all entries */
824 for (esel = 0; esel < vcpu_e500->gtlb_params[tlbsel].entries;
826 kvmppc_e500_gtlbe_invalidate(vcpu_e500, tlbsel, esel);
829 esel = kvmppc_e500_tlb_index(vcpu_e500, ea, tlbsel,
830 get_cur_pid(vcpu), -1);
832 kvmppc_e500_gtlbe_invalidate(vcpu_e500, tlbsel, esel);
835 /* Invalidate all vcpu id mappings */
836 kvmppc_e500_id_table_reset_all(vcpu_e500);
841 int kvmppc_e500_emul_tlbre(struct kvm_vcpu *vcpu)
843 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
845 struct kvm_book3e_206_tlb_entry *gtlbe;
847 tlbsel = get_tlb_tlbsel(vcpu_e500);
848 esel = get_tlb_esel(vcpu_e500, tlbsel);
850 gtlbe = get_entry(vcpu_e500, tlbsel, esel);
851 vcpu_e500->mas0 &= ~MAS0_NV(~0);
852 vcpu_e500->mas0 |= MAS0_NV(vcpu_e500->gtlb_nv[tlbsel]);
853 vcpu_e500->mas1 = gtlbe->mas1;
854 vcpu_e500->mas2 = gtlbe->mas2;
855 vcpu_e500->mas7_3 = gtlbe->mas7_3;
860 int kvmppc_e500_emul_tlbsx(struct kvm_vcpu *vcpu, int rb)
862 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
863 int as = !!get_cur_sas(vcpu_e500);
864 unsigned int pid = get_cur_spid(vcpu_e500);
866 struct kvm_book3e_206_tlb_entry *gtlbe = NULL;
869 ea = kvmppc_get_gpr(vcpu, rb);
871 for (tlbsel = 0; tlbsel < 2; tlbsel++) {
872 esel = kvmppc_e500_tlb_index(vcpu_e500, ea, tlbsel, pid, as);
874 gtlbe = get_entry(vcpu_e500, tlbsel, esel);
880 esel &= vcpu_e500->gtlb_params[tlbsel].ways - 1;
882 vcpu_e500->mas0 = MAS0_TLBSEL(tlbsel) | MAS0_ESEL(esel)
883 | MAS0_NV(vcpu_e500->gtlb_nv[tlbsel]);
884 vcpu_e500->mas1 = gtlbe->mas1;
885 vcpu_e500->mas2 = gtlbe->mas2;
886 vcpu_e500->mas7_3 = gtlbe->mas7_3;
890 /* since we only have two TLBs, only lower bit is used. */
891 tlbsel = vcpu_e500->mas4 >> 28 & 0x1;
892 victim = (tlbsel == 0) ? gtlb0_get_next_victim(vcpu_e500) : 0;
894 vcpu_e500->mas0 = MAS0_TLBSEL(tlbsel) | MAS0_ESEL(victim)
895 | MAS0_NV(vcpu_e500->gtlb_nv[tlbsel]);
896 vcpu_e500->mas1 = (vcpu_e500->mas6 & MAS6_SPID0)
897 | (vcpu_e500->mas6 & (MAS6_SAS ? MAS1_TS : 0))
898 | (vcpu_e500->mas4 & MAS4_TSIZED(~0));
899 vcpu_e500->mas2 &= MAS2_EPN;
900 vcpu_e500->mas2 |= vcpu_e500->mas4 & MAS2_ATTRIB_MASK;
901 vcpu_e500->mas7_3 &= MAS3_U0 | MAS3_U1 | MAS3_U2 | MAS3_U3;
904 kvmppc_set_exit_type(vcpu, EMULATED_TLBSX_EXITS);
908 /* sesel is index into the set, not the whole array */
909 static void write_stlbe(struct kvmppc_vcpu_e500 *vcpu_e500,
910 struct kvm_book3e_206_tlb_entry *gtlbe,
911 struct kvm_book3e_206_tlb_entry *stlbe,
912 int stlbsel, int sesel)
917 stid = kvmppc_e500_get_sid(vcpu_e500, get_tlb_ts(gtlbe),
919 get_cur_pr(&vcpu_e500->vcpu), 0);
921 stlbe->mas1 |= MAS1_TID(stid);
922 write_host_tlbe(vcpu_e500, stlbsel, sesel, stlbe);
926 int kvmppc_e500_emul_tlbwe(struct kvm_vcpu *vcpu)
928 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
929 struct kvm_book3e_206_tlb_entry *gtlbe;
932 tlbsel = get_tlb_tlbsel(vcpu_e500);
933 esel = get_tlb_esel(vcpu_e500, tlbsel);
935 gtlbe = get_entry(vcpu_e500, tlbsel, esel);
937 if (get_tlb_v(gtlbe))
938 inval_gtlbe_on_host(vcpu_e500, tlbsel, esel);
940 gtlbe->mas1 = vcpu_e500->mas1;
941 gtlbe->mas2 = vcpu_e500->mas2;
942 gtlbe->mas7_3 = vcpu_e500->mas7_3;
944 trace_kvm_gtlb_write(vcpu_e500->mas0, gtlbe->mas1, gtlbe->mas2,
945 (u32)gtlbe->mas7_3, (u32)(gtlbe->mas7_3 >> 32));
947 /* Invalidate shadow mappings for the about-to-be-clobbered TLBE. */
948 if (tlbe_is_host_safe(vcpu, gtlbe)) {
949 struct kvm_book3e_206_tlb_entry stlbe;
957 gtlbe->mas1 &= ~MAS1_TSIZE(~0);
958 gtlbe->mas1 |= MAS1_TSIZE(BOOK3E_PAGESZ_4K);
961 sesel = kvmppc_e500_tlb0_map(vcpu_e500, esel, &stlbe);
967 eaddr = get_tlb_eaddr(gtlbe);
968 raddr = get_tlb_raddr(gtlbe);
970 /* Create a 4KB mapping on the host.
971 * If the guest wanted a large page,
972 * only the first 4KB is mapped here and the rest
973 * are mapped on the fly. */
975 sesel = kvmppc_e500_tlb1_map(vcpu_e500, eaddr,
976 raddr >> PAGE_SHIFT, gtlbe, &stlbe);
983 write_stlbe(vcpu_e500, gtlbe, &stlbe, stlbsel, sesel);
986 kvmppc_set_exit_type(vcpu, EMULATED_TLBWE_EXITS);
990 int kvmppc_mmu_itlb_index(struct kvm_vcpu *vcpu, gva_t eaddr)
992 unsigned int as = !!(vcpu->arch.shared->msr & MSR_IS);
994 return kvmppc_e500_tlb_search(vcpu, eaddr, get_cur_pid(vcpu), as);
997 int kvmppc_mmu_dtlb_index(struct kvm_vcpu *vcpu, gva_t eaddr)
999 unsigned int as = !!(vcpu->arch.shared->msr & MSR_DS);
1001 return kvmppc_e500_tlb_search(vcpu, eaddr, get_cur_pid(vcpu), as);
1004 void kvmppc_mmu_itlb_miss(struct kvm_vcpu *vcpu)
1006 unsigned int as = !!(vcpu->arch.shared->msr & MSR_IS);
1008 kvmppc_e500_deliver_tlb_miss(vcpu, vcpu->arch.pc, as);
1011 void kvmppc_mmu_dtlb_miss(struct kvm_vcpu *vcpu)
1013 unsigned int as = !!(vcpu->arch.shared->msr & MSR_DS);
1015 kvmppc_e500_deliver_tlb_miss(vcpu, vcpu->arch.fault_dear, as);
1018 gpa_t kvmppc_mmu_xlate(struct kvm_vcpu *vcpu, unsigned int index,
1021 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
1022 struct kvm_book3e_206_tlb_entry *gtlbe;
1025 gtlbe = get_entry(vcpu_e500, tlbsel_of(index), esel_of(index));
1026 pgmask = get_tlb_bytes(gtlbe) - 1;
1028 return get_tlb_raddr(gtlbe) | (eaddr & pgmask);
1031 void kvmppc_mmu_destroy(struct kvm_vcpu *vcpu)
1035 void kvmppc_mmu_map(struct kvm_vcpu *vcpu, u64 eaddr, gpa_t gpaddr,
1038 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
1039 struct tlbe_priv *priv;
1040 struct kvm_book3e_206_tlb_entry *gtlbe, stlbe;
1041 int tlbsel = tlbsel_of(index);
1042 int esel = esel_of(index);
1045 gtlbe = get_entry(vcpu_e500, tlbsel, esel);
1050 sesel = esel & (host_tlb_params[0].ways - 1);
1051 priv = &vcpu_e500->gtlb_priv[tlbsel][esel];
1053 kvmppc_e500_setup_stlbe(vcpu_e500, gtlbe, BOOK3E_PAGESZ_4K,
1054 &priv->ref, eaddr, &stlbe);
1058 gfn_t gfn = gpaddr >> PAGE_SHIFT;
1061 sesel = kvmppc_e500_tlb1_map(vcpu_e500, eaddr, gfn,
1071 write_stlbe(vcpu_e500, gtlbe, &stlbe, stlbsel, sesel);
1074 int kvmppc_e500_tlb_search(struct kvm_vcpu *vcpu,
1075 gva_t eaddr, unsigned int pid, int as)
1077 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
1080 for (tlbsel = 0; tlbsel < 2; tlbsel++) {
1081 esel = kvmppc_e500_tlb_index(vcpu_e500, eaddr, tlbsel, pid, as);
1083 return index_of(tlbsel, esel);
1089 void kvmppc_set_pid(struct kvm_vcpu *vcpu, u32 pid)
1091 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
1093 if (vcpu->arch.pid != pid) {
1094 vcpu_e500->pid[0] = vcpu->arch.pid = pid;
1095 kvmppc_e500_recalc_shadow_pid(vcpu_e500);
1099 void kvmppc_e500_tlb_setup(struct kvmppc_vcpu_e500 *vcpu_e500)
1101 struct kvm_book3e_206_tlb_entry *tlbe;
1103 /* Insert large initial mapping for guest. */
1104 tlbe = get_entry(vcpu_e500, 1, 0);
1105 tlbe->mas1 = MAS1_VALID | MAS1_TSIZE(BOOK3E_PAGESZ_256M);
1107 tlbe->mas7_3 = E500_TLB_SUPER_PERM_MASK;
1109 /* 4K map for serial output. Used by kernel wrapper. */
1110 tlbe = get_entry(vcpu_e500, 1, 1);
1111 tlbe->mas1 = MAS1_VALID | MAS1_TSIZE(BOOK3E_PAGESZ_4K);
1112 tlbe->mas2 = (0xe0004500 & 0xFFFFF000) | MAS2_I | MAS2_G;
1113 tlbe->mas7_3 = (0xe0004500 & 0xFFFFF000) | E500_TLB_SUPER_PERM_MASK;
1116 static void free_gtlb(struct kvmppc_vcpu_e500 *vcpu_e500)
1120 clear_tlb_refs(vcpu_e500);
1121 kfree(vcpu_e500->gtlb_priv[0]);
1122 kfree(vcpu_e500->gtlb_priv[1]);
1124 if (vcpu_e500->shared_tlb_pages) {
1125 vfree((void *)(round_down((uintptr_t)vcpu_e500->gtlb_arch,
1128 for (i = 0; i < vcpu_e500->num_shared_tlb_pages; i++) {
1129 set_page_dirty_lock(vcpu_e500->shared_tlb_pages[i]);
1130 put_page(vcpu_e500->shared_tlb_pages[i]);
1133 vcpu_e500->num_shared_tlb_pages = 0;
1134 vcpu_e500->shared_tlb_pages = NULL;
1136 kfree(vcpu_e500->gtlb_arch);
1139 vcpu_e500->gtlb_arch = NULL;
1142 int kvm_vcpu_ioctl_config_tlb(struct kvm_vcpu *vcpu,
1143 struct kvm_config_tlb *cfg)
1145 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
1146 struct kvm_book3e_206_tlb_params params;
1148 struct page **pages;
1149 struct tlbe_priv *privs[2] = {};
1152 int num_pages, ret, i;
1154 if (cfg->mmu_type != KVM_MMU_FSL_BOOKE_NOHV)
1157 if (copy_from_user(¶ms, (void __user *)(uintptr_t)cfg->params,
1161 if (params.tlb_sizes[1] > 64)
1163 if (params.tlb_ways[1] != params.tlb_sizes[1])
1165 if (params.tlb_sizes[2] != 0 || params.tlb_sizes[3] != 0)
1167 if (params.tlb_ways[2] != 0 || params.tlb_ways[3] != 0)
1170 if (!is_power_of_2(params.tlb_ways[0]))
1173 sets = params.tlb_sizes[0] >> ilog2(params.tlb_ways[0]);
1174 if (!is_power_of_2(sets))
1177 array_len = params.tlb_sizes[0] + params.tlb_sizes[1];
1178 array_len *= sizeof(struct kvm_book3e_206_tlb_entry);
1180 if (cfg->array_len < array_len)
1183 num_pages = DIV_ROUND_UP(cfg->array + array_len - 1, PAGE_SIZE) -
1184 cfg->array / PAGE_SIZE;
1185 pages = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL);
1189 ret = get_user_pages_fast(cfg->array, num_pages, 1, pages);
1193 if (ret != num_pages) {
1199 virt = vmap(pages, num_pages, VM_MAP, PAGE_KERNEL);
1203 privs[0] = kzalloc(sizeof(struct tlbe_priv) * params.tlb_sizes[0],
1205 privs[1] = kzalloc(sizeof(struct tlbe_priv) * params.tlb_sizes[1],
1208 if (!privs[0] || !privs[1])
1211 free_gtlb(vcpu_e500);
1213 vcpu_e500->gtlb_priv[0] = privs[0];
1214 vcpu_e500->gtlb_priv[1] = privs[1];
1216 vcpu_e500->gtlb_arch = (struct kvm_book3e_206_tlb_entry *)
1217 (virt + (cfg->array & (PAGE_SIZE - 1)));
1219 vcpu_e500->gtlb_params[0].entries = params.tlb_sizes[0];
1220 vcpu_e500->gtlb_params[1].entries = params.tlb_sizes[1];
1222 vcpu_e500->gtlb_offset[0] = 0;
1223 vcpu_e500->gtlb_offset[1] = params.tlb_sizes[0];
1225 vcpu_e500->tlb0cfg = mfspr(SPRN_TLB0CFG) & ~0xfffUL;
1226 if (params.tlb_sizes[0] <= 2048)
1227 vcpu_e500->tlb0cfg |= params.tlb_sizes[0];
1229 vcpu_e500->tlb1cfg = mfspr(SPRN_TLB1CFG) & ~0xfffUL;
1230 vcpu_e500->tlb1cfg |= params.tlb_sizes[1];
1232 vcpu_e500->shared_tlb_pages = pages;
1233 vcpu_e500->num_shared_tlb_pages = num_pages;
1235 vcpu_e500->gtlb_params[0].ways = params.tlb_ways[0];
1236 vcpu_e500->gtlb_params[0].sets = sets;
1238 vcpu_e500->gtlb_params[1].ways = params.tlb_sizes[1];
1239 vcpu_e500->gtlb_params[1].sets = 1;
1247 for (i = 0; i < num_pages; i++)
1255 int kvm_vcpu_ioctl_dirty_tlb(struct kvm_vcpu *vcpu,
1256 struct kvm_dirty_tlb *dirty)
1258 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
1260 clear_tlb_refs(vcpu_e500);
1264 int kvmppc_e500_tlb_init(struct kvmppc_vcpu_e500 *vcpu_e500)
1266 int entry_size = sizeof(struct kvm_book3e_206_tlb_entry);
1267 int entries = KVM_E500_TLB0_SIZE + KVM_E500_TLB1_SIZE;
1269 host_tlb_params[0].entries = mfspr(SPRN_TLB0CFG) & TLBnCFG_N_ENTRY;
1270 host_tlb_params[1].entries = mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY;
1273 * This should never happen on real e500 hardware, but is
1274 * architecturally possible -- e.g. in some weird nested
1275 * virtualization case.
1277 if (host_tlb_params[0].entries == 0 ||
1278 host_tlb_params[1].entries == 0) {
1279 pr_err("%s: need to know host tlb size\n", __func__);
1283 host_tlb_params[0].ways = (mfspr(SPRN_TLB0CFG) & TLBnCFG_ASSOC) >>
1284 TLBnCFG_ASSOC_SHIFT;
1285 host_tlb_params[1].ways = host_tlb_params[1].entries;
1287 if (!is_power_of_2(host_tlb_params[0].entries) ||
1288 !is_power_of_2(host_tlb_params[0].ways) ||
1289 host_tlb_params[0].entries < host_tlb_params[0].ways ||
1290 host_tlb_params[0].ways == 0) {
1291 pr_err("%s: bad tlb0 host config: %u entries %u ways\n",
1292 __func__, host_tlb_params[0].entries,
1293 host_tlb_params[0].ways);
1297 host_tlb_params[0].sets =
1298 host_tlb_params[0].entries / host_tlb_params[0].ways;
1299 host_tlb_params[1].sets = 1;
1301 vcpu_e500->gtlb_params[0].entries = KVM_E500_TLB0_SIZE;
1302 vcpu_e500->gtlb_params[1].entries = KVM_E500_TLB1_SIZE;
1304 vcpu_e500->gtlb_params[0].ways = KVM_E500_TLB0_WAY_NUM;
1305 vcpu_e500->gtlb_params[0].sets =
1306 KVM_E500_TLB0_SIZE / KVM_E500_TLB0_WAY_NUM;
1308 vcpu_e500->gtlb_params[1].ways = KVM_E500_TLB1_SIZE;
1309 vcpu_e500->gtlb_params[1].sets = 1;
1311 vcpu_e500->gtlb_arch = kmalloc(entries * entry_size, GFP_KERNEL);
1312 if (!vcpu_e500->gtlb_arch)
1315 vcpu_e500->gtlb_offset[0] = 0;
1316 vcpu_e500->gtlb_offset[1] = KVM_E500_TLB0_SIZE;
1318 vcpu_e500->tlb_refs[0] =
1319 kzalloc(sizeof(struct tlbe_ref) * host_tlb_params[0].entries,
1321 if (!vcpu_e500->tlb_refs[0])
1324 vcpu_e500->tlb_refs[1] =
1325 kzalloc(sizeof(struct tlbe_ref) * host_tlb_params[1].entries,
1327 if (!vcpu_e500->tlb_refs[1])
1330 vcpu_e500->gtlb_priv[0] = kzalloc(sizeof(struct tlbe_ref) *
1331 vcpu_e500->gtlb_params[0].entries,
1333 if (!vcpu_e500->gtlb_priv[0])
1336 vcpu_e500->gtlb_priv[1] = kzalloc(sizeof(struct tlbe_ref) *
1337 vcpu_e500->gtlb_params[1].entries,
1339 if (!vcpu_e500->gtlb_priv[1])
1342 if (kvmppc_e500_id_table_alloc(vcpu_e500) == NULL)
1345 /* Init TLB configuration register */
1346 vcpu_e500->tlb0cfg = mfspr(SPRN_TLB0CFG) & ~0xfffUL;
1347 vcpu_e500->tlb0cfg |= vcpu_e500->gtlb_params[0].entries;
1348 vcpu_e500->tlb1cfg = mfspr(SPRN_TLB1CFG) & ~0xfffUL;
1349 vcpu_e500->tlb1cfg |= vcpu_e500->gtlb_params[1].entries;
1354 free_gtlb(vcpu_e500);
1355 kfree(vcpu_e500->tlb_refs[0]);
1356 kfree(vcpu_e500->tlb_refs[1]);
1360 void kvmppc_e500_tlb_uninit(struct kvmppc_vcpu_e500 *vcpu_e500)
1362 free_gtlb(vcpu_e500);
1363 kvmppc_e500_id_table_free(vcpu_e500);
1365 kfree(vcpu_e500->tlb_refs[0]);
1366 kfree(vcpu_e500->tlb_refs[1]);