1 /*
2 * Copyright (C) 2008-2013 Freescale Semiconductor, Inc. All rights reserved.
3 *
4 * Author: Yu Liu, yu.liu@freescale.com
5 * Scott Wood, scottwood@freescale.com
6 * Ashish Kalra, ashish.kalra@freescale.com
7 * Varun Sethi, varun.sethi@freescale.com
8 * Alexander Graf, agraf@suse.de
9 *
10 * Description:
11 * This file is based on arch/powerpc/kvm/44x_tlb.c,
12 * by Hollis Blanchard <hollisb@us.ibm.com>.
13 *
14 * This program is free software; you can redistribute it and/or modify
15 * it under the terms of the GNU General Public License, version 2, as
16 * published by the Free Software Foundation.
17 */
18
19 #include <linux/kernel.h>
20 #include <linux/types.h>
21 #include <linux/slab.h>
22 #include <linux/string.h>
23 #include <linux/kvm.h>
24 #include <linux/kvm_host.h>
25 #include <linux/highmem.h>
26 #include <linux/log2.h>
27 #include <linux/uaccess.h>
28 #include <linux/sched.h>
29 #include <linux/rwsem.h>
30 #include <linux/vmalloc.h>
31 #include <linux/hugetlb.h>
32 #include <asm/kvm_ppc.h>
33
34 #include "e500.h"
35 #include "timing.h"
36 #include "e500_mmu_host.h"
37
38 #include "trace_booke.h"
39
40 #define to_htlb1_esel(esel) (host_tlb_params[1].entries - (esel) - 1)
41
42 static struct kvmppc_e500_tlb_params host_tlb_params[E500_TLB_NUM];
43
tlb1_max_shadow_size(void)44 static inline unsigned int tlb1_max_shadow_size(void)
45 {
46 /* reserve one entry for magic page */
47 return host_tlb_params[1].entries - tlbcam_index - 1;
48 }
49
e500_shadow_mas3_attrib(u32 mas3,int usermode)50 static inline u32 e500_shadow_mas3_attrib(u32 mas3, int usermode)
51 {
52 /* Mask off reserved bits. */
53 mas3 &= MAS3_ATTRIB_MASK;
54
55 #ifndef CONFIG_KVM_BOOKE_HV
56 if (!usermode) {
57 /* Guest is in supervisor mode,
58 * so we need to translate guest
59 * supervisor permissions into user permissions. */
60 mas3 &= ~E500_TLB_USER_PERM_MASK;
61 mas3 |= (mas3 & E500_TLB_SUPER_PERM_MASK) << 1;
62 }
63 mas3 |= E500_TLB_SUPER_PERM_MASK;
64 #endif
65 return mas3;
66 }
67
68 /*
69 * writing shadow tlb entry to host TLB
70 */
__write_host_tlbe(struct kvm_book3e_206_tlb_entry * stlbe,uint32_t mas0,uint32_t lpid)71 static inline void __write_host_tlbe(struct kvm_book3e_206_tlb_entry *stlbe,
72 uint32_t mas0,
73 uint32_t lpid)
74 {
75 unsigned long flags;
76
77 local_irq_save(flags);
78 mtspr(SPRN_MAS0, mas0);
79 mtspr(SPRN_MAS1, stlbe->mas1);
80 mtspr(SPRN_MAS2, (unsigned long)stlbe->mas2);
81 mtspr(SPRN_MAS3, (u32)stlbe->mas7_3);
82 mtspr(SPRN_MAS7, (u32)(stlbe->mas7_3 >> 32));
83 #ifdef CONFIG_KVM_BOOKE_HV
84 mtspr(SPRN_MAS8, MAS8_TGS | get_thread_specific_lpid(lpid));
85 #endif
86 asm volatile("isync; tlbwe" : : : "memory");
87
88 #ifdef CONFIG_KVM_BOOKE_HV
89 /* Must clear mas8 for other host tlbwe's */
90 mtspr(SPRN_MAS8, 0);
91 isync();
92 #endif
93 local_irq_restore(flags);
94
95 trace_kvm_booke206_stlb_write(mas0, stlbe->mas8, stlbe->mas1,
96 stlbe->mas2, stlbe->mas7_3);
97 }
98
99 /*
100 * Acquire a mas0 with victim hint, as if we just took a TLB miss.
101 *
102 * We don't care about the address we're searching for, other than that it's
103 * in the right set and is not present in the TLB. Using a zero PID and a
104 * userspace address means we don't have to set and then restore MAS5, or
105 * calculate a proper MAS6 value.
106 */
get_host_mas0(unsigned long eaddr)107 static u32 get_host_mas0(unsigned long eaddr)
108 {
109 unsigned long flags;
110 u32 mas0;
111 u32 mas4;
112
113 local_irq_save(flags);
114 mtspr(SPRN_MAS6, 0);
115 mas4 = mfspr(SPRN_MAS4);
116 mtspr(SPRN_MAS4, mas4 & ~MAS4_TLBSEL_MASK);
117 asm volatile("tlbsx 0, %0" : : "b" (eaddr & ~CONFIG_PAGE_OFFSET));
118 mas0 = mfspr(SPRN_MAS0);
119 mtspr(SPRN_MAS4, mas4);
120 local_irq_restore(flags);
121
122 return mas0;
123 }
124
125 /* sesel is for tlb1 only */
write_host_tlbe(struct kvmppc_vcpu_e500 * vcpu_e500,int tlbsel,int sesel,struct kvm_book3e_206_tlb_entry * stlbe)126 static inline void write_host_tlbe(struct kvmppc_vcpu_e500 *vcpu_e500,
127 int tlbsel, int sesel, struct kvm_book3e_206_tlb_entry *stlbe)
128 {
129 u32 mas0;
130
131 if (tlbsel == 0) {
132 mas0 = get_host_mas0(stlbe->mas2);
133 __write_host_tlbe(stlbe, mas0, vcpu_e500->vcpu.kvm->arch.lpid);
134 } else {
135 __write_host_tlbe(stlbe,
136 MAS0_TLBSEL(1) |
137 MAS0_ESEL(to_htlb1_esel(sesel)),
138 vcpu_e500->vcpu.kvm->arch.lpid);
139 }
140 }
141
142 /* sesel is for tlb1 only */
write_stlbe(struct kvmppc_vcpu_e500 * vcpu_e500,struct kvm_book3e_206_tlb_entry * gtlbe,struct kvm_book3e_206_tlb_entry * stlbe,int stlbsel,int sesel)143 static void write_stlbe(struct kvmppc_vcpu_e500 *vcpu_e500,
144 struct kvm_book3e_206_tlb_entry *gtlbe,
145 struct kvm_book3e_206_tlb_entry *stlbe,
146 int stlbsel, int sesel)
147 {
148 int stid;
149
150 preempt_disable();
151 stid = kvmppc_e500_get_tlb_stid(&vcpu_e500->vcpu, gtlbe);
152
153 stlbe->mas1 |= MAS1_TID(stid);
154 write_host_tlbe(vcpu_e500, stlbsel, sesel, stlbe);
155 preempt_enable();
156 }
157
158 #ifdef CONFIG_KVM_E500V2
159 /* XXX should be a hook in the gva2hpa translation */
kvmppc_map_magic(struct kvm_vcpu * vcpu)160 void kvmppc_map_magic(struct kvm_vcpu *vcpu)
161 {
162 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
163 struct kvm_book3e_206_tlb_entry magic;
164 ulong shared_page = ((ulong)vcpu->arch.shared) & PAGE_MASK;
165 unsigned int stid;
166 pfn_t pfn;
167
168 pfn = (pfn_t)virt_to_phys((void *)shared_page) >> PAGE_SHIFT;
169 get_page(pfn_to_page(pfn));
170
171 preempt_disable();
172 stid = kvmppc_e500_get_sid(vcpu_e500, 0, 0, 0, 0);
173
174 magic.mas1 = MAS1_VALID | MAS1_TS | MAS1_TID(stid) |
175 MAS1_TSIZE(BOOK3E_PAGESZ_4K);
176 magic.mas2 = vcpu->arch.magic_page_ea | MAS2_M;
177 magic.mas7_3 = ((u64)pfn << PAGE_SHIFT) |
178 MAS3_SW | MAS3_SR | MAS3_UW | MAS3_UR;
179 magic.mas8 = 0;
180
181 __write_host_tlbe(&magic, MAS0_TLBSEL(1) | MAS0_ESEL(tlbcam_index), 0);
182 preempt_enable();
183 }
184 #endif
185
inval_gtlbe_on_host(struct kvmppc_vcpu_e500 * vcpu_e500,int tlbsel,int esel)186 void inval_gtlbe_on_host(struct kvmppc_vcpu_e500 *vcpu_e500, int tlbsel,
187 int esel)
188 {
189 struct kvm_book3e_206_tlb_entry *gtlbe =
190 get_entry(vcpu_e500, tlbsel, esel);
191 struct tlbe_ref *ref = &vcpu_e500->gtlb_priv[tlbsel][esel].ref;
192
193 /* Don't bother with unmapped entries */
194 if (!(ref->flags & E500_TLB_VALID)) {
195 WARN(ref->flags & (E500_TLB_BITMAP | E500_TLB_TLB0),
196 "%s: flags %x\n", __func__, ref->flags);
197 WARN_ON(tlbsel == 1 && vcpu_e500->g2h_tlb1_map[esel]);
198 }
199
200 if (tlbsel == 1 && ref->flags & E500_TLB_BITMAP) {
201 u64 tmp = vcpu_e500->g2h_tlb1_map[esel];
202 int hw_tlb_indx;
203 unsigned long flags;
204
205 local_irq_save(flags);
206 while (tmp) {
207 hw_tlb_indx = __ilog2_u64(tmp & -tmp);
208 mtspr(SPRN_MAS0,
209 MAS0_TLBSEL(1) |
210 MAS0_ESEL(to_htlb1_esel(hw_tlb_indx)));
211 mtspr(SPRN_MAS1, 0);
212 asm volatile("tlbwe");
213 vcpu_e500->h2g_tlb1_rmap[hw_tlb_indx] = 0;
214 tmp &= tmp - 1;
215 }
216 mb();
217 vcpu_e500->g2h_tlb1_map[esel] = 0;
218 ref->flags &= ~(E500_TLB_BITMAP | E500_TLB_VALID);
219 local_irq_restore(flags);
220 }
221
222 if (tlbsel == 1 && ref->flags & E500_TLB_TLB0) {
223 /*
224 * TLB1 entry is backed by 4k pages. This should happen
225 * rarely and is not worth optimizing. Invalidate everything.
226 */
227 kvmppc_e500_tlbil_all(vcpu_e500);
228 ref->flags &= ~(E500_TLB_TLB0 | E500_TLB_VALID);
229 }
230
231 /*
232 * If TLB entry is still valid then it's a TLB0 entry, and thus
233 * backed by at most one host tlbe per shadow pid
234 */
235 if (ref->flags & E500_TLB_VALID)
236 kvmppc_e500_tlbil_one(vcpu_e500, gtlbe);
237
238 /* Mark the TLB as not backed by the host anymore */
239 ref->flags = 0;
240 }
241
tlbe_is_writable(struct kvm_book3e_206_tlb_entry * tlbe)242 static inline int tlbe_is_writable(struct kvm_book3e_206_tlb_entry *tlbe)
243 {
244 return tlbe->mas7_3 & (MAS3_SW|MAS3_UW);
245 }
246
kvmppc_e500_ref_setup(struct tlbe_ref * ref,struct kvm_book3e_206_tlb_entry * gtlbe,pfn_t pfn,unsigned int wimg)247 static inline void kvmppc_e500_ref_setup(struct tlbe_ref *ref,
248 struct kvm_book3e_206_tlb_entry *gtlbe,
249 pfn_t pfn, unsigned int wimg)
250 {
251 ref->pfn = pfn;
252 ref->flags = E500_TLB_VALID;
253
254 /* Use guest supplied MAS2_G and MAS2_E */
255 ref->flags |= (gtlbe->mas2 & MAS2_ATTRIB_MASK) | wimg;
256
257 /* Mark the page accessed */
258 kvm_set_pfn_accessed(pfn);
259
260 if (tlbe_is_writable(gtlbe))
261 kvm_set_pfn_dirty(pfn);
262 }
263
kvmppc_e500_ref_release(struct tlbe_ref * ref)264 static inline void kvmppc_e500_ref_release(struct tlbe_ref *ref)
265 {
266 if (ref->flags & E500_TLB_VALID) {
267 /* FIXME: don't log bogus pfn for TLB1 */
268 trace_kvm_booke206_ref_release(ref->pfn, ref->flags);
269 ref->flags = 0;
270 }
271 }
272
clear_tlb1_bitmap(struct kvmppc_vcpu_e500 * vcpu_e500)273 static void clear_tlb1_bitmap(struct kvmppc_vcpu_e500 *vcpu_e500)
274 {
275 if (vcpu_e500->g2h_tlb1_map)
276 memset(vcpu_e500->g2h_tlb1_map, 0,
277 sizeof(u64) * vcpu_e500->gtlb_params[1].entries);
278 if (vcpu_e500->h2g_tlb1_rmap)
279 memset(vcpu_e500->h2g_tlb1_rmap, 0,
280 sizeof(unsigned int) * host_tlb_params[1].entries);
281 }
282
clear_tlb_privs(struct kvmppc_vcpu_e500 * vcpu_e500)283 static void clear_tlb_privs(struct kvmppc_vcpu_e500 *vcpu_e500)
284 {
285 int tlbsel;
286 int i;
287
288 for (tlbsel = 0; tlbsel <= 1; tlbsel++) {
289 for (i = 0; i < vcpu_e500->gtlb_params[tlbsel].entries; i++) {
290 struct tlbe_ref *ref =
291 &vcpu_e500->gtlb_priv[tlbsel][i].ref;
292 kvmppc_e500_ref_release(ref);
293 }
294 }
295 }
296
kvmppc_core_flush_tlb(struct kvm_vcpu * vcpu)297 void kvmppc_core_flush_tlb(struct kvm_vcpu *vcpu)
298 {
299 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
300 kvmppc_e500_tlbil_all(vcpu_e500);
301 clear_tlb_privs(vcpu_e500);
302 clear_tlb1_bitmap(vcpu_e500);
303 }
304
305 /* TID must be supplied by the caller */
kvmppc_e500_setup_stlbe(struct kvm_vcpu * vcpu,struct kvm_book3e_206_tlb_entry * gtlbe,int tsize,struct tlbe_ref * ref,u64 gvaddr,struct kvm_book3e_206_tlb_entry * stlbe)306 static void kvmppc_e500_setup_stlbe(
307 struct kvm_vcpu *vcpu,
308 struct kvm_book3e_206_tlb_entry *gtlbe,
309 int tsize, struct tlbe_ref *ref, u64 gvaddr,
310 struct kvm_book3e_206_tlb_entry *stlbe)
311 {
312 pfn_t pfn = ref->pfn;
313 u32 pr = vcpu->arch.shared->msr & MSR_PR;
314
315 BUG_ON(!(ref->flags & E500_TLB_VALID));
316
317 /* Force IPROT=0 for all guest mappings. */
318 stlbe->mas1 = MAS1_TSIZE(tsize) | get_tlb_sts(gtlbe) | MAS1_VALID;
319 stlbe->mas2 = (gvaddr & MAS2_EPN) | (ref->flags & E500_TLB_MAS2_ATTR);
320 stlbe->mas7_3 = ((u64)pfn << PAGE_SHIFT) |
321 e500_shadow_mas3_attrib(gtlbe->mas7_3, pr);
322 }
323
kvmppc_e500_shadow_map(struct kvmppc_vcpu_e500 * vcpu_e500,u64 gvaddr,gfn_t gfn,struct kvm_book3e_206_tlb_entry * gtlbe,int tlbsel,struct kvm_book3e_206_tlb_entry * stlbe,struct tlbe_ref * ref)324 static inline int kvmppc_e500_shadow_map(struct kvmppc_vcpu_e500 *vcpu_e500,
325 u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe,
326 int tlbsel, struct kvm_book3e_206_tlb_entry *stlbe,
327 struct tlbe_ref *ref)
328 {
329 struct kvm_memory_slot *slot;
330 unsigned long pfn = 0; /* silence GCC warning */
331 unsigned long hva;
332 int pfnmap = 0;
333 int tsize = BOOK3E_PAGESZ_4K;
334 int ret = 0;
335 unsigned long mmu_seq;
336 struct kvm *kvm = vcpu_e500->vcpu.kvm;
337 unsigned long tsize_pages = 0;
338 pte_t *ptep;
339 unsigned int wimg = 0;
340 pgd_t *pgdir;
341 unsigned long flags;
342
343 /* used to check for invalidations in progress */
344 mmu_seq = kvm->mmu_notifier_seq;
345 smp_rmb();
346
347 /*
348 * Translate guest physical to true physical, acquiring
349 * a page reference if it is normal, non-reserved memory.
350 *
351 * gfn_to_memslot() must succeed because otherwise we wouldn't
352 * have gotten this far. Eventually we should just pass the slot
353 * pointer through from the first lookup.
354 */
355 slot = gfn_to_memslot(vcpu_e500->vcpu.kvm, gfn);
356 hva = gfn_to_hva_memslot(slot, gfn);
357
358 if (tlbsel == 1) {
359 struct vm_area_struct *vma;
360 down_read(¤t->mm->mmap_sem);
361
362 vma = find_vma(current->mm, hva);
363 if (vma && hva >= vma->vm_start &&
364 (vma->vm_flags & VM_PFNMAP)) {
365 /*
366 * This VMA is a physically contiguous region (e.g.
367 * /dev/mem) that bypasses normal Linux page
368 * management. Find the overlap between the
369 * vma and the memslot.
370 */
371
372 unsigned long start, end;
373 unsigned long slot_start, slot_end;
374
375 pfnmap = 1;
376
377 start = vma->vm_pgoff;
378 end = start +
379 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT);
380
381 pfn = start + ((hva - vma->vm_start) >> PAGE_SHIFT);
382
383 slot_start = pfn - (gfn - slot->base_gfn);
384 slot_end = slot_start + slot->npages;
385
386 if (start < slot_start)
387 start = slot_start;
388 if (end > slot_end)
389 end = slot_end;
390
391 tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >>
392 MAS1_TSIZE_SHIFT;
393
394 /*
395 * e500 doesn't implement the lowest tsize bit,
396 * or 1K pages.
397 */
398 tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1);
399
400 /*
401 * Now find the largest tsize (up to what the guest
402 * requested) that will cover gfn, stay within the
403 * range, and for which gfn and pfn are mutually
404 * aligned.
405 */
406
407 for (; tsize > BOOK3E_PAGESZ_4K; tsize -= 2) {
408 unsigned long gfn_start, gfn_end;
409 tsize_pages = 1UL << (tsize - 2);
410
411 gfn_start = gfn & ~(tsize_pages - 1);
412 gfn_end = gfn_start + tsize_pages;
413
414 if (gfn_start + pfn - gfn < start)
415 continue;
416 if (gfn_end + pfn - gfn > end)
417 continue;
418 if ((gfn & (tsize_pages - 1)) !=
419 (pfn & (tsize_pages - 1)))
420 continue;
421
422 gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1);
423 pfn &= ~(tsize_pages - 1);
424 break;
425 }
426 } else if (vma && hva >= vma->vm_start &&
427 (vma->vm_flags & VM_HUGETLB)) {
428 unsigned long psize = vma_kernel_pagesize(vma);
429
430 tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >>
431 MAS1_TSIZE_SHIFT;
432
433 /*
434 * Take the largest page size that satisfies both host
435 * and guest mapping
436 */
437 tsize = min(__ilog2(psize) - 10, tsize);
438
439 /*
440 * e500 doesn't implement the lowest tsize bit,
441 * or 1K pages.
442 */
443 tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1);
444 }
445
446 up_read(¤t->mm->mmap_sem);
447 }
448
449 if (likely(!pfnmap)) {
450 tsize_pages = 1UL << (tsize + 10 - PAGE_SHIFT);
451 pfn = gfn_to_pfn_memslot(slot, gfn);
452 if (is_error_noslot_pfn(pfn)) {
453 if (printk_ratelimit())
454 pr_err("%s: real page not found for gfn %lx\n",
455 __func__, (long)gfn);
456 return -EINVAL;
457 }
458
459 /* Align guest and physical address to page map boundaries */
460 pfn &= ~(tsize_pages - 1);
461 gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1);
462 }
463
464 spin_lock(&kvm->mmu_lock);
465 if (mmu_notifier_retry(kvm, mmu_seq)) {
466 ret = -EAGAIN;
467 goto out;
468 }
469
470
471 pgdir = vcpu_e500->vcpu.arch.pgdir;
472 /*
473 * We are just looking at the wimg bits, so we don't
474 * care much about the trans splitting bit.
475 * We are holding kvm->mmu_lock so a notifier invalidate
476 * can't run hence pfn won't change.
477 */
478 local_irq_save(flags);
479 ptep = find_linux_pte_or_hugepte(pgdir, hva, NULL, NULL);
480 if (ptep) {
481 pte_t pte = READ_ONCE(*ptep);
482
483 if (pte_present(pte)) {
484 wimg = (pte_val(pte) >> PTE_WIMGE_SHIFT) &
485 MAS2_WIMGE_MASK;
486 local_irq_restore(flags);
487 } else {
488 local_irq_restore(flags);
489 pr_err_ratelimited("%s: pte not present: gfn %lx,pfn %lx\n",
490 __func__, (long)gfn, pfn);
491 ret = -EINVAL;
492 goto out;
493 }
494 }
495 kvmppc_e500_ref_setup(ref, gtlbe, pfn, wimg);
496
497 kvmppc_e500_setup_stlbe(&vcpu_e500->vcpu, gtlbe, tsize,
498 ref, gvaddr, stlbe);
499
500 /* Clear i-cache for new pages */
501 kvmppc_mmu_flush_icache(pfn);
502
503 out:
504 spin_unlock(&kvm->mmu_lock);
505
506 /* Drop refcount on page, so that mmu notifiers can clear it */
507 kvm_release_pfn_clean(pfn);
508
509 return ret;
510 }
511
512 /* XXX only map the one-one case, for now use TLB0 */
kvmppc_e500_tlb0_map(struct kvmppc_vcpu_e500 * vcpu_e500,int esel,struct kvm_book3e_206_tlb_entry * stlbe)513 static int kvmppc_e500_tlb0_map(struct kvmppc_vcpu_e500 *vcpu_e500, int esel,
514 struct kvm_book3e_206_tlb_entry *stlbe)
515 {
516 struct kvm_book3e_206_tlb_entry *gtlbe;
517 struct tlbe_ref *ref;
518 int stlbsel = 0;
519 int sesel = 0;
520 int r;
521
522 gtlbe = get_entry(vcpu_e500, 0, esel);
523 ref = &vcpu_e500->gtlb_priv[0][esel].ref;
524
525 r = kvmppc_e500_shadow_map(vcpu_e500, get_tlb_eaddr(gtlbe),
526 get_tlb_raddr(gtlbe) >> PAGE_SHIFT,
527 gtlbe, 0, stlbe, ref);
528 if (r)
529 return r;
530
531 write_stlbe(vcpu_e500, gtlbe, stlbe, stlbsel, sesel);
532
533 return 0;
534 }
535
kvmppc_e500_tlb1_map_tlb1(struct kvmppc_vcpu_e500 * vcpu_e500,struct tlbe_ref * ref,int esel)536 static int kvmppc_e500_tlb1_map_tlb1(struct kvmppc_vcpu_e500 *vcpu_e500,
537 struct tlbe_ref *ref,
538 int esel)
539 {
540 unsigned int sesel = vcpu_e500->host_tlb1_nv++;
541
542 if (unlikely(vcpu_e500->host_tlb1_nv >= tlb1_max_shadow_size()))
543 vcpu_e500->host_tlb1_nv = 0;
544
545 if (vcpu_e500->h2g_tlb1_rmap[sesel]) {
546 unsigned int idx = vcpu_e500->h2g_tlb1_rmap[sesel] - 1;
547 vcpu_e500->g2h_tlb1_map[idx] &= ~(1ULL << sesel);
548 }
549
550 vcpu_e500->gtlb_priv[1][esel].ref.flags |= E500_TLB_BITMAP;
551 vcpu_e500->g2h_tlb1_map[esel] |= (u64)1 << sesel;
552 vcpu_e500->h2g_tlb1_rmap[sesel] = esel + 1;
553 WARN_ON(!(ref->flags & E500_TLB_VALID));
554
555 return sesel;
556 }
557
558 /* Caller must ensure that the specified guest TLB entry is safe to insert into
559 * the shadow TLB. */
560 /* For both one-one and one-to-many */
kvmppc_e500_tlb1_map(struct kvmppc_vcpu_e500 * vcpu_e500,u64 gvaddr,gfn_t gfn,struct kvm_book3e_206_tlb_entry * gtlbe,struct kvm_book3e_206_tlb_entry * stlbe,int esel)561 static int kvmppc_e500_tlb1_map(struct kvmppc_vcpu_e500 *vcpu_e500,
562 u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe,
563 struct kvm_book3e_206_tlb_entry *stlbe, int esel)
564 {
565 struct tlbe_ref *ref = &vcpu_e500->gtlb_priv[1][esel].ref;
566 int sesel;
567 int r;
568
569 r = kvmppc_e500_shadow_map(vcpu_e500, gvaddr, gfn, gtlbe, 1, stlbe,
570 ref);
571 if (r)
572 return r;
573
574 /* Use TLB0 when we can only map a page with 4k */
575 if (get_tlb_tsize(stlbe) == BOOK3E_PAGESZ_4K) {
576 vcpu_e500->gtlb_priv[1][esel].ref.flags |= E500_TLB_TLB0;
577 write_stlbe(vcpu_e500, gtlbe, stlbe, 0, 0);
578 return 0;
579 }
580
581 /* Otherwise map into TLB1 */
582 sesel = kvmppc_e500_tlb1_map_tlb1(vcpu_e500, ref, esel);
583 write_stlbe(vcpu_e500, gtlbe, stlbe, 1, sesel);
584
585 return 0;
586 }
587
kvmppc_mmu_map(struct kvm_vcpu * vcpu,u64 eaddr,gpa_t gpaddr,unsigned int index)588 void kvmppc_mmu_map(struct kvm_vcpu *vcpu, u64 eaddr, gpa_t gpaddr,
589 unsigned int index)
590 {
591 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
592 struct tlbe_priv *priv;
593 struct kvm_book3e_206_tlb_entry *gtlbe, stlbe;
594 int tlbsel = tlbsel_of(index);
595 int esel = esel_of(index);
596
597 gtlbe = get_entry(vcpu_e500, tlbsel, esel);
598
599 switch (tlbsel) {
600 case 0:
601 priv = &vcpu_e500->gtlb_priv[tlbsel][esel];
602
603 /* Triggers after clear_tlb_privs or on initial mapping */
604 if (!(priv->ref.flags & E500_TLB_VALID)) {
605 kvmppc_e500_tlb0_map(vcpu_e500, esel, &stlbe);
606 } else {
607 kvmppc_e500_setup_stlbe(vcpu, gtlbe, BOOK3E_PAGESZ_4K,
608 &priv->ref, eaddr, &stlbe);
609 write_stlbe(vcpu_e500, gtlbe, &stlbe, 0, 0);
610 }
611 break;
612
613 case 1: {
614 gfn_t gfn = gpaddr >> PAGE_SHIFT;
615 kvmppc_e500_tlb1_map(vcpu_e500, eaddr, gfn, gtlbe, &stlbe,
616 esel);
617 break;
618 }
619
620 default:
621 BUG();
622 break;
623 }
624 }
625
626 #ifdef CONFIG_KVM_BOOKE_HV
kvmppc_load_last_inst(struct kvm_vcpu * vcpu,enum instruction_type type,u32 * instr)627 int kvmppc_load_last_inst(struct kvm_vcpu *vcpu, enum instruction_type type,
628 u32 *instr)
629 {
630 gva_t geaddr;
631 hpa_t addr;
632 hfn_t pfn;
633 hva_t eaddr;
634 u32 mas1, mas2, mas3;
635 u64 mas7_mas3;
636 struct page *page;
637 unsigned int addr_space, psize_shift;
638 bool pr;
639 unsigned long flags;
640
641 /* Search TLB for guest pc to get the real address */
642 geaddr = kvmppc_get_pc(vcpu);
643
644 addr_space = (vcpu->arch.shared->msr & MSR_IS) >> MSR_IR_LG;
645
646 local_irq_save(flags);
647 mtspr(SPRN_MAS6, (vcpu->arch.pid << MAS6_SPID_SHIFT) | addr_space);
648 mtspr(SPRN_MAS5, MAS5_SGS | get_lpid(vcpu));
649 asm volatile("tlbsx 0, %[geaddr]\n" : :
650 [geaddr] "r" (geaddr));
651 mtspr(SPRN_MAS5, 0);
652 mtspr(SPRN_MAS8, 0);
653 mas1 = mfspr(SPRN_MAS1);
654 mas2 = mfspr(SPRN_MAS2);
655 mas3 = mfspr(SPRN_MAS3);
656 #ifdef CONFIG_64BIT
657 mas7_mas3 = mfspr(SPRN_MAS7_MAS3);
658 #else
659 mas7_mas3 = ((u64)mfspr(SPRN_MAS7) << 32) | mas3;
660 #endif
661 local_irq_restore(flags);
662
663 /*
664 * If the TLB entry for guest pc was evicted, return to the guest.
665 * There are high chances to find a valid TLB entry next time.
666 */
667 if (!(mas1 & MAS1_VALID))
668 return EMULATE_AGAIN;
669
670 /*
671 * Another thread may rewrite the TLB entry in parallel, don't
672 * execute from the address if the execute permission is not set
673 */
674 pr = vcpu->arch.shared->msr & MSR_PR;
675 if (unlikely((pr && !(mas3 & MAS3_UX)) ||
676 (!pr && !(mas3 & MAS3_SX)))) {
677 pr_err_ratelimited(
678 "%s: Instruction emulation from guest address %08lx without execute permission\n",
679 __func__, geaddr);
680 return EMULATE_AGAIN;
681 }
682
683 /*
684 * The real address will be mapped by a cacheable, memory coherent,
685 * write-back page. Check for mismatches when LRAT is used.
686 */
687 if (has_feature(vcpu, VCPU_FTR_MMU_V2) &&
688 unlikely((mas2 & MAS2_I) || (mas2 & MAS2_W) || !(mas2 & MAS2_M))) {
689 pr_err_ratelimited(
690 "%s: Instruction emulation from guest address %08lx mismatches storage attributes\n",
691 __func__, geaddr);
692 return EMULATE_AGAIN;
693 }
694
695 /* Get pfn */
696 psize_shift = MAS1_GET_TSIZE(mas1) + 10;
697 addr = (mas7_mas3 & (~0ULL << psize_shift)) |
698 (geaddr & ((1ULL << psize_shift) - 1ULL));
699 pfn = addr >> PAGE_SHIFT;
700
701 /* Guard against emulation from devices area */
702 if (unlikely(!page_is_ram(pfn))) {
703 pr_err_ratelimited("%s: Instruction emulation from non-RAM host address %08llx is not supported\n",
704 __func__, addr);
705 return EMULATE_AGAIN;
706 }
707
708 /* Map a page and get guest's instruction */
709 page = pfn_to_page(pfn);
710 eaddr = (unsigned long)kmap_atomic(page);
711 *instr = *(u32 *)(eaddr | (unsigned long)(addr & ~PAGE_MASK));
712 kunmap_atomic((u32 *)eaddr);
713
714 return EMULATE_DONE;
715 }
716 #else
kvmppc_load_last_inst(struct kvm_vcpu * vcpu,enum instruction_type type,u32 * instr)717 int kvmppc_load_last_inst(struct kvm_vcpu *vcpu, enum instruction_type type,
718 u32 *instr)
719 {
720 return EMULATE_AGAIN;
721 }
722 #endif
723
724 /************* MMU Notifiers *************/
725
kvm_unmap_hva(struct kvm * kvm,unsigned long hva)726 int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
727 {
728 trace_kvm_unmap_hva(hva);
729
730 /*
731 * Flush all shadow tlb entries everywhere. This is slow, but
732 * we are 100% sure that we catch the to be unmapped page
733 */
734 kvm_flush_remote_tlbs(kvm);
735
736 return 0;
737 }
738
kvm_unmap_hva_range(struct kvm * kvm,unsigned long start,unsigned long end)739 int kvm_unmap_hva_range(struct kvm *kvm, unsigned long start, unsigned long end)
740 {
741 /* kvm_unmap_hva flushes everything anyways */
742 kvm_unmap_hva(kvm, start);
743
744 return 0;
745 }
746
kvm_age_hva(struct kvm * kvm,unsigned long start,unsigned long end)747 int kvm_age_hva(struct kvm *kvm, unsigned long start, unsigned long end)
748 {
749 /* XXX could be more clever ;) */
750 return 0;
751 }
752
kvm_test_age_hva(struct kvm * kvm,unsigned long hva)753 int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
754 {
755 /* XXX could be more clever ;) */
756 return 0;
757 }
758
kvm_set_spte_hva(struct kvm * kvm,unsigned long hva,pte_t pte)759 void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
760 {
761 /* The page will get remapped properly on its next fault */
762 kvm_unmap_hva(kvm, hva);
763 }
764
765 /*****************************************/
766
e500_mmu_host_init(struct kvmppc_vcpu_e500 * vcpu_e500)767 int e500_mmu_host_init(struct kvmppc_vcpu_e500 *vcpu_e500)
768 {
769 host_tlb_params[0].entries = mfspr(SPRN_TLB0CFG) & TLBnCFG_N_ENTRY;
770 host_tlb_params[1].entries = mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY;
771
772 /*
773 * This should never happen on real e500 hardware, but is
774 * architecturally possible -- e.g. in some weird nested
775 * virtualization case.
776 */
777 if (host_tlb_params[0].entries == 0 ||
778 host_tlb_params[1].entries == 0) {
779 pr_err("%s: need to know host tlb size\n", __func__);
780 return -ENODEV;
781 }
782
783 host_tlb_params[0].ways = (mfspr(SPRN_TLB0CFG) & TLBnCFG_ASSOC) >>
784 TLBnCFG_ASSOC_SHIFT;
785 host_tlb_params[1].ways = host_tlb_params[1].entries;
786
787 if (!is_power_of_2(host_tlb_params[0].entries) ||
788 !is_power_of_2(host_tlb_params[0].ways) ||
789 host_tlb_params[0].entries < host_tlb_params[0].ways ||
790 host_tlb_params[0].ways == 0) {
791 pr_err("%s: bad tlb0 host config: %u entries %u ways\n",
792 __func__, host_tlb_params[0].entries,
793 host_tlb_params[0].ways);
794 return -ENODEV;
795 }
796
797 host_tlb_params[0].sets =
798 host_tlb_params[0].entries / host_tlb_params[0].ways;
799 host_tlb_params[1].sets = 1;
800
801 vcpu_e500->h2g_tlb1_rmap = kzalloc(sizeof(unsigned int) *
802 host_tlb_params[1].entries,
803 GFP_KERNEL);
804 if (!vcpu_e500->h2g_tlb1_rmap)
805 return -EINVAL;
806
807 return 0;
808 }
809
e500_mmu_host_uninit(struct kvmppc_vcpu_e500 * vcpu_e500)810 void e500_mmu_host_uninit(struct kvmppc_vcpu_e500 *vcpu_e500)
811 {
812 kfree(vcpu_e500->h2g_tlb1_rmap);
813 }
814