projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: PPC: booke: Paravirtualize wrtee
[linux-2.6-block.git] / arch / powerpc / kvm / e500_tlb.c
CommitLineData
bc8080cb 1/*
49ea0695 2 * Copyright (C) 2008-2011 Freescale Semiconductor, Inc. All rights reserved.
bc8080cb
HB		 3 *
4 * Author: Yu Liu, yu.liu@freescale.com
5 *
6 * Description:
7 * This file is based on arch/powerpc/kvm/44x_tlb.c,
8 * by Hollis Blanchard <hollisb@us.ibm.com>.
9 *
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.
13 */
14
0164c0f0 15#include <linux/kernel.h>
bc8080cb 16#include <linux/types.h>
5a0e3ad6 17#include <linux/slab.h>
bc8080cb
HB		 18#include <linux/string.h>
19#include <linux/kvm.h>
20#include <linux/kvm_host.h>
21#include <linux/highmem.h>
dc83b8bc
SW		 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>
95325e6b 27#include <linux/hugetlb.h>
bc8080cb
HB		 28#include <asm/kvm_ppc.h>
29#include <asm/kvm_e500.h>
30
9aa4dd5e 31#include "../mm/mmu_decl.h"
bc8080cb 32#include "e500_tlb.h"
46f43c6e 33#include "trace.h"
49ea0695 34#include "timing.h"
bc8080cb 35
0164c0f0 36#define to_htlb1_esel(esel) (host_tlb_params[1].entries - (esel) - 1)
bc8080cb 37
dd9ebf1f
LY		 38struct id {
39 unsigned long val;
40 struct id **pentry;
41};
42
43#define NUM_TIDS 256
44
45/*
46 * This table provide mappings from:
47 * (guestAS,guestTID,guestPR) --> ID of physical cpu
48 * guestAS [0..1]
49 * guestTID [0..255]
50 * guestPR [0..1]
51 * ID [1..255]
52 * Each vcpu keeps one vcpu_id_table.
53 */
54struct vcpu_id_table {
55 struct id id[2][NUM_TIDS][2];
56};
57
58/*
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.
62 */
63struct pcpu_id_table {
64 struct id *entry[NUM_TIDS];
65};
66
67static DEFINE_PER_CPU(struct pcpu_id_table, pcpu_sids);
68
69/* This variable keeps last used shadow ID on local core.
70 * The valid range of shadow ID is [1..255] */
71static DEFINE_PER_CPU(unsigned long, pcpu_last_used_sid);
72
0164c0f0 73static struct kvmppc_e500_tlb_params host_tlb_params[E500_TLB_NUM];
bc8080cb 74
dc83b8bc
SW		 75static struct kvm_book3e_206_tlb_entry *get_entry(
76 struct kvmppc_vcpu_e500 *vcpu_e500, int tlbsel, int entry)
77{
78 int offset = vcpu_e500->gtlb_offset[tlbsel];
79 return &vcpu_e500->gtlb_arch[offset + entry];
80}
81
dd9ebf1f
LY		 82/*
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.
85 *
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).
89 */
90static inline int local_sid_setup_one(struct id *entry)
91{
92 unsigned long sid;
93 int ret = -1;
94
95 sid = ++(__get_cpu_var(pcpu_last_used_sid));
96 if (sid < NUM_TIDS) {
97 __get_cpu_var(pcpu_sids).entry[sid] = entry;
98 entry->val = sid;
99 entry->pentry = &__get_cpu_var(pcpu_sids).entry[sid];
100 ret = sid;
101 }
102
103 /*
104 * If sid == NUM_TIDS, we've run out of sids. We return -1, and
105 * the caller will invalidate everything and start over.
106 *
107 * sid > NUM_TIDS indicates a race, which we disable preemption to
108 * avoid.
109 */
110 WARN_ON(sid > NUM_TIDS);
111
112 return ret;
113}
114
115/*
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.
119 *
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).
123 */
124static inline int local_sid_lookup(struct id *entry)
125{
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])
129 return entry->val;
130 return -1;
131}
132
90b92a6f 133/* Invalidate all id mappings on local core -- call with preempt disabled */
dd9ebf1f
LY		 134static inline void local_sid_destroy_all(void)
135{
dd9ebf1f
LY		 136 __get_cpu_var(pcpu_last_used_sid) = 0;
137 memset(&__get_cpu_var(pcpu_sids), 0, sizeof(__get_cpu_var(pcpu_sids)));
dd9ebf1f
LY		 138}
139
140static void *kvmppc_e500_id_table_alloc(struct kvmppc_vcpu_e500 *vcpu_e500)
141{
142 vcpu_e500->idt = kzalloc(sizeof(struct vcpu_id_table), GFP_KERNEL);
143 return vcpu_e500->idt;
144}
145
146static void kvmppc_e500_id_table_free(struct kvmppc_vcpu_e500 *vcpu_e500)
147{
148 kfree(vcpu_e500->idt);
149}
150
151/* Invalidate all mappings on vcpu */
152static void kvmppc_e500_id_table_reset_all(struct kvmppc_vcpu_e500 *vcpu_e500)
153{
154 memset(vcpu_e500->idt, 0, sizeof(struct vcpu_id_table));
155
156 /* Update shadow pid when mappings are changed */
157 kvmppc_e500_recalc_shadow_pid(vcpu_e500);
158}
159
160/* Invalidate one ID mapping on vcpu */
161static inline void kvmppc_e500_id_table_reset_one(
162 struct kvmppc_vcpu_e500 *vcpu_e500,
163 int as, int pid, int pr)
164{
165 struct vcpu_id_table *idt = vcpu_e500->idt;
166
167 BUG_ON(as >= 2);
168 BUG_ON(pid >= NUM_TIDS);
169 BUG_ON(pr >= 2);
170
171 idt->id[as][pid][pr].val = 0;
172 idt->id[as][pid][pr].pentry = NULL;
173
174 /* Update shadow pid when mappings are changed */
175 kvmppc_e500_recalc_shadow_pid(vcpu_e500);
176}
177
178/*
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.
182 *
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).
186 */
187static 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)
190{
191 struct vcpu_id_table *idt = vcpu_e500->idt;
192 int sid;
193
194 BUG_ON(as >= 2);
195 BUG_ON(gid >= NUM_TIDS);
196 BUG_ON(pr >= 2);
197
198 sid = local_sid_lookup(&idt->id[as][gid][pr]);
199
200 while (sid <= 0) {
201 /* No mapping yet */
202 sid = local_sid_setup_one(&idt->id[as][gid][pr]);
203 if (sid <= 0) {
204 _tlbil_all();
205 local_sid_destroy_all();
206 }
207
208 /* Update shadow pid when mappings are changed */
209 if (!avoid_recursion)
210 kvmppc_e500_recalc_shadow_pid(vcpu_e500);
211 }
212
213 return sid;
214}
215
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 */
220void kvmppc_e500_recalc_shadow_pid(struct kvmppc_vcpu_e500 *vcpu_e500)
221{
222 preempt_disable();
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);
230 preempt_enable();
231}
232
0164c0f0 233static inline unsigned int gtlb0_get_next_victim(
bc8080cb
HB		 234 struct kvmppc_vcpu_e500 *vcpu_e500)
235{
236 unsigned int victim;
237
08b7fa92 238 victim = vcpu_e500->gtlb_nv[0]++;
dc83b8bc 239 if (unlikely(vcpu_e500->gtlb_nv[0] >= vcpu_e500->gtlb_params[0].ways))
08b7fa92 240 vcpu_e500->gtlb_nv[0] = 0;
bc8080cb
HB		 241
242 return victim;
243}
244
245static inline unsigned int tlb1_max_shadow_size(void)
246{
a4cd8b23 247 /* reserve one entry for magic page */
0164c0f0 248 return host_tlb_params[1].entries - tlbcam_index - 1;
bc8080cb
HB		 249}
250
dc83b8bc 251static inline int tlbe_is_writable(struct kvm_book3e_206_tlb_entry *tlbe)
bc8080cb 252{
dc83b8bc 253 return tlbe->mas7_3 & (MAS3_SW|MAS3_UW);
bc8080cb
HB		 254}
255
256static inline u32 e500_shadow_mas3_attrib(u32 mas3, int usermode)
257{
258 /* Mask off reserved bits. */
259 mas3 &= MAS3_ATTRIB_MASK;
260
261 if (!usermode) {
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;
267 }
268
269 return mas3 | E500_TLB_SUPER_PERM_MASK;
270}
271
272static inline u32 e500_shadow_mas2_attrib(u32 mas2, int usermode)
273{
046a48b3
LY		 274#ifdef CONFIG_SMP
275 return (mas2 & MAS2_ATTRIB_MASK) | MAS2_M;
276#else
bc8080cb 277 return mas2 & MAS2_ATTRIB_MASK;
046a48b3 278#endif
bc8080cb
HB		 279}
280
281/*
282 * writing shadow tlb entry to host TLB
283 */
dc83b8bc
SW		 284static inline void __write_host_tlbe(struct kvm_book3e_206_tlb_entry *stlbe,
285 uint32_t mas0)
bc8080cb 286{
0ef30995
SW		 287 unsigned long flags;
288
289 local_irq_save(flags);
290 mtspr(SPRN_MAS0, mas0);
bc8080cb 291 mtspr(SPRN_MAS1, stlbe->mas1);
dc83b8bc
SW		 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));
0ef30995
SW		 295 asm volatile("isync; tlbwe" : : : "memory");
296 local_irq_restore(flags);
bc8080cb
HB		 297}
298
0164c0f0 299/* esel is index into set, not whole array */
bc8080cb 300static inline void write_host_tlbe(struct kvmppc_vcpu_e500 *vcpu_e500,
dc83b8bc 301 int tlbsel, int esel, struct kvm_book3e_206_tlb_entry *stlbe)
bc8080cb 302{
bc8080cb 303 if (tlbsel == 0) {
dc83b8bc
SW		 304 int way = esel & (vcpu_e500->gtlb_params[0].ways - 1);
305 __write_host_tlbe(stlbe, MAS0_TLBSEL(0) | MAS0_ESEL(way));
bc8080cb 306 } else {
0ef30995
SW		 307 __write_host_tlbe(stlbe,
308 MAS0_TLBSEL(1) |
309 MAS0_ESEL(to_htlb1_esel(esel)));
bc8080cb 310 }
08b7fa92 311 trace_kvm_stlb_write(index_of(tlbsel, esel), stlbe->mas1, stlbe->mas2,
dc83b8bc 312 (u32)stlbe->mas7_3, (u32)(stlbe->mas7_3 >> 32));
bc8080cb
HB		 313}
314
a4cd8b23
SW		 315void kvmppc_map_magic(struct kvm_vcpu *vcpu)
316{
dd9ebf1f 317 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
dc83b8bc 318 struct kvm_book3e_206_tlb_entry magic;
a4cd8b23 319 ulong shared_page = ((ulong)vcpu->arch.shared) & PAGE_MASK;
dd9ebf1f 320 unsigned int stid;
a4cd8b23
SW		 321 pfn_t pfn;
322
323 pfn = (pfn_t)virt_to_phys((void *)shared_page) >> PAGE_SHIFT;
324 get_page(pfn_to_page(pfn));
325
dd9ebf1f
LY		 326 preempt_disable();
327 stid = kvmppc_e500_get_sid(vcpu_e500, 0, 0, 0, 0);
328
329 magic.mas1 = MAS1_VALID | MAS1_TS | MAS1_TID(stid) |
a4cd8b23
SW		 330 MAS1_TSIZE(BOOK3E_PAGESZ_4K);
331 magic.mas2 = vcpu->arch.magic_page_ea | MAS2_M;
dc83b8bc
SW		 332 magic.mas7_3 = ((u64)pfn << PAGE_SHIFT) |
333 MAS3_SW | MAS3_SR | MAS3_UW | MAS3_UR;
a4cd8b23
SW		 334
335 __write_host_tlbe(&magic, MAS0_TLBSEL(1) | MAS0_ESEL(tlbcam_index));
dd9ebf1f 336 preempt_enable();
a4cd8b23
SW		 337}
338
bc8080cb
HB		 339void kvmppc_e500_tlb_load(struct kvm_vcpu *vcpu, int cpu)
340{
dd9ebf1f
LY		 341 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
342
343 /* Shadow PID may be expired on local core */
344 kvmppc_e500_recalc_shadow_pid(vcpu_e500);
bc8080cb
HB		 345}
346
347void kvmppc_e500_tlb_put(struct kvm_vcpu *vcpu)
348{
dd9ebf1f
LY		 349}
350
0164c0f0
SW		 351static void inval_gtlbe_on_host(struct kvmppc_vcpu_e500 *vcpu_e500,
352 int tlbsel, int esel)
dd9ebf1f 353{
dc83b8bc
SW		 354 struct kvm_book3e_206_tlb_entry *gtlbe =
355 get_entry(vcpu_e500, tlbsel, esel);
dd9ebf1f
LY		 356 struct vcpu_id_table *idt = vcpu_e500->idt;
357 unsigned int pr, tid, ts, pid;
358 u32 val, eaddr;
359 unsigned long flags;
360
361 ts = get_tlb_ts(gtlbe);
362 tid = get_tlb_tid(gtlbe);
363
364 preempt_disable();
365
366 /* One guest ID may be mapped to two shadow IDs */
367 for (pr = 0; pr < 2; pr++) {
368 /*
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.
373 *
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
376 * entire shadow PID.
377 */
378 if (tlbsel == 1 ||
379 (pid = local_sid_lookup(&idt->id[ts][tid][pr])) <= 0) {
380 kvmppc_e500_id_table_reset_one(vcpu_e500, ts, tid, pr);
381 continue;
382 }
383
384 /*
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.
389 */
390 val = (pid << MAS6_SPID_SHIFT) | MAS6_SAS;
391 eaddr = get_tlb_eaddr(gtlbe);
392
393 local_irq_save(flags);
394
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");
401 }
402
403 local_irq_restore(flags);
404 }
405
406 preempt_enable();
bc8080cb
HB		 407}
408
0164c0f0
SW		 409static int tlb0_set_base(gva_t addr, int sets, int ways)
410{
411 int set_base;
412
413 set_base = (addr >> PAGE_SHIFT) & (sets - 1);
414 set_base *= ways;
415
416 return set_base;
417}
418
419static int gtlb0_set_base(struct kvmppc_vcpu_e500 *vcpu_e500, gva_t addr)
420{
dc83b8bc
SW		 421 return tlb0_set_base(addr, vcpu_e500->gtlb_params[0].sets,
422 vcpu_e500->gtlb_params[0].ways);
0164c0f0
SW		 423}
424
425static int htlb0_set_base(gva_t addr)
426{
427 return tlb0_set_base(addr, host_tlb_params[0].sets,
428 host_tlb_params[0].ways);
429}
430
431static unsigned int get_tlb_esel(struct kvmppc_vcpu_e500 *vcpu_e500, int tlbsel)
432{
433 unsigned int esel = get_tlb_esel_bit(vcpu_e500);
434
435 if (tlbsel == 0) {
dc83b8bc 436 esel &= vcpu_e500->gtlb_params[0].ways - 1;
0164c0f0
SW		 437 esel += gtlb0_set_base(vcpu_e500, vcpu_e500->mas2);
438 } else {
dc83b8bc 439 esel &= vcpu_e500->gtlb_params[tlbsel].entries - 1;
0164c0f0
SW		 440 }
441
442 return esel;
443}
444
bc8080cb
HB		 445/* Search the guest TLB for a matching entry. */
446static int kvmppc_e500_tlb_index(struct kvmppc_vcpu_e500 *vcpu_e500,
447 gva_t eaddr, int tlbsel, unsigned int pid, int as)
448{
dc83b8bc
SW		 449 int size = vcpu_e500->gtlb_params[tlbsel].entries;
450 unsigned int set_base, offset;
bc8080cb
HB		 451 int i;
452
1aee47a0 453 if (tlbsel == 0) {
0164c0f0 454 set_base = gtlb0_set_base(vcpu_e500, eaddr);
dc83b8bc 455 size = vcpu_e500->gtlb_params[0].ways;
1aee47a0
SW		 456 } else {
457 set_base = 0;
458 }
459
dc83b8bc
SW		 460 offset = vcpu_e500->gtlb_offset[tlbsel];
461
1aee47a0 462 for (i = 0; i < size; i++) {
dc83b8bc
SW		 463 struct kvm_book3e_206_tlb_entry *tlbe =
464 &vcpu_e500->gtlb_arch[offset + set_base + i];
bc8080cb
HB		 465 unsigned int tid;
466
467 if (eaddr < get_tlb_eaddr(tlbe))
468 continue;
469
470 if (eaddr > get_tlb_end(tlbe))
471 continue;
472
473 tid = get_tlb_tid(tlbe);
474 if (tid && (tid != pid))
475 continue;
476
477 if (!get_tlb_v(tlbe))
478 continue;
479
480 if (get_tlb_ts(tlbe) != as && as != -1)
481 continue;
482
1aee47a0 483 return set_base + i;
bc8080cb
HB		 484 }
485
486 return -1;
487}
488
0164c0f0 489static inline void kvmppc_e500_ref_setup(struct tlbe_ref *ref,
dc83b8bc 490 struct kvm_book3e_206_tlb_entry *gtlbe,
0164c0f0 491 pfn_t pfn)
bc8080cb 492{
0164c0f0
SW		 493 ref->pfn = pfn;
494 ref->flags = E500_TLB_VALID;
bc8080cb 495
08b7fa92 496 if (tlbe_is_writable(gtlbe))
0164c0f0 497 ref->flags |= E500_TLB_DIRTY;
bc8080cb
HB		 498}
499
0164c0f0 500static inline void kvmppc_e500_ref_release(struct tlbe_ref *ref)
bc8080cb 501{
0164c0f0
SW		 502 if (ref->flags & E500_TLB_VALID) {
503 if (ref->flags & E500_TLB_DIRTY)
504 kvm_release_pfn_dirty(ref->pfn);
08b7fa92 505 else
0164c0f0
SW		 506 kvm_release_pfn_clean(ref->pfn);
507
508 ref->flags = 0;
509 }
510}
511
512static void clear_tlb_privs(struct kvmppc_vcpu_e500 *vcpu_e500)
513{
514 int tlbsel = 0;
515 int i;
bc8080cb 516
dc83b8bc 517 for (i = 0; i < vcpu_e500->gtlb_params[tlbsel].entries; i++) {
0164c0f0
SW		 518 struct tlbe_ref *ref =
519 &vcpu_e500->gtlb_priv[tlbsel][i].ref;
520 kvmppc_e500_ref_release(ref);
08b7fa92 521 }
bc8080cb
HB		 522}
523
0164c0f0
SW		 524static void clear_tlb_refs(struct kvmppc_vcpu_e500 *vcpu_e500)
525{
526 int stlbsel = 1;
527 int i;
528
dc83b8bc
SW		 529 kvmppc_e500_id_table_reset_all(vcpu_e500);
530
0164c0f0
SW		 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);
535 }
536
537 clear_tlb_privs(vcpu_e500);
538}
539
bc8080cb
HB		 540static inline void kvmppc_e500_deliver_tlb_miss(struct kvm_vcpu *vcpu,
541 unsigned int eaddr, int as)
542{
543 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
544 unsigned int victim, pidsel, tsized;
545 int tlbsel;
546
fb2838d4 547 /* since we only have two TLBs, only lower bit is used. */
bc8080cb 548 tlbsel = (vcpu_e500->mas4 >> 28) & 0x1;
0164c0f0 549 victim = (tlbsel == 0) ? gtlb0_get_next_victim(vcpu_e500) : 0;
bc8080cb 550 pidsel = (vcpu_e500->mas4 >> 16) & 0xf;
0cfb50e5 551 tsized = (vcpu_e500->mas4 >> 7) & 0x1f;
bc8080cb
HB		 552
553 vcpu_e500->mas0 = MAS0_TLBSEL(tlbsel) | MAS0_ESEL(victim)
08b7fa92 554 | MAS0_NV(vcpu_e500->gtlb_nv[tlbsel]);
bc8080cb
HB		 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);
dc83b8bc 560 vcpu_e500->mas7_3 &= MAS3_U0 | MAS3_U1 | MAS3_U2 | MAS3_U3;
bc8080cb
HB		 561 vcpu_e500->mas6 = (vcpu_e500->mas6 & MAS6_SPID1)
562 | (get_cur_pid(vcpu) << 16)
563 | (as ? MAS6_SAS : 0);
bc8080cb
HB		 564}
565
3bf3cdcc 566/* TID must be supplied by the caller */
dc83b8bc
SW		 567static 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)
08b7fa92 572{
0164c0f0
SW		 573 pfn_t pfn = ref->pfn;
574
575 BUG_ON(!(ref->flags & E500_TLB_VALID));
08b7fa92
LY		 576
577 /* Force TS=1 IPROT=0 for all guest mappings. */
3bf3cdcc 578 stlbe->mas1 = MAS1_TSIZE(tsize) | MAS1_TS | MAS1_VALID;
08b7fa92
LY		 579 stlbe->mas2 = (gvaddr & MAS2_EPN)
580 | e500_shadow_mas2_attrib(gtlbe->mas2,
581 vcpu_e500->vcpu.arch.shared->msr & MSR_PR);
dc83b8bc
SW		 582 stlbe->mas7_3 = ((u64)pfn << PAGE_SHIFT)
583 | e500_shadow_mas3_attrib(gtlbe->mas7_3,
08b7fa92 584 vcpu_e500->vcpu.arch.shared->msr & MSR_PR);
08b7fa92
LY		 585}
586
0164c0f0 587/* sesel is an index into the entire array, not just the set */
bc8080cb 588static inline void kvmppc_e500_shadow_map(struct kvmppc_vcpu_e500 *vcpu_e500,
dc83b8bc
SW		 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)
bc8080cb 592{
9973d54e 593 struct kvm_memory_slot *slot;
9973d54e
SW		 594 unsigned long pfn, hva;
595 int pfnmap = 0;
596 int tsize = BOOK3E_PAGESZ_4K;
bc8080cb 597
59c1f4e3
SW		 598 /*
599 * Translate guest physical to true physical, acquiring
600 * a page reference if it is normal, non-reserved memory.
9973d54e
SW		 601 *
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.
59c1f4e3 605 */
9973d54e
SW		 606 slot = gfn_to_memslot(vcpu_e500->vcpu.kvm, gfn);
607 hva = gfn_to_hva_memslot(slot, gfn);
608
609 if (tlbsel == 1) {
610 struct vm_area_struct *vma;
611 down_read(&current->mm->mmap_sem);
612
613 vma = find_vma(current->mm, hva);
614 if (vma && hva >= vma->vm_start &&
615 (vma->vm_flags & VM_PFNMAP)) {
616 /*
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.
621 */
622
623 unsigned long start, end;
624 unsigned long slot_start, slot_end;
625
626 pfnmap = 1;
627
628 start = vma->vm_pgoff;
629 end = start +
630 ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT);
631
632 pfn = start + ((hva - vma->vm_start) >> PAGE_SHIFT);
633
634 slot_start = pfn - (gfn - slot->base_gfn);
635 slot_end = slot_start + slot->npages;
636
637 if (start < slot_start)
638 start = slot_start;
639 if (end > slot_end)
640 end = slot_end;
641
642 tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >>
643 MAS1_TSIZE_SHIFT;
644
645 /*
646 * e500 doesn't implement the lowest tsize bit,
647 * or 1K pages.
648 */
649 tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1);
650
651 /*
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
655 * aligned.
656 */
657
658 for (; tsize > BOOK3E_PAGESZ_4K; tsize -= 2) {
659 unsigned long gfn_start, gfn_end, tsize_pages;
660 tsize_pages = 1 << (tsize - 2);
661
662 gfn_start = gfn & ~(tsize_pages - 1);
663 gfn_end = gfn_start + tsize_pages;
664
665 if (gfn_start + pfn - gfn < start)
666 continue;
667 if (gfn_end + pfn - gfn > end)
668 continue;
669 if ((gfn & (tsize_pages - 1)) !=
670 (pfn & (tsize_pages - 1)))
671 continue;
672
673 gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1);
674 pfn &= ~(tsize_pages - 1);
675 break;
676 }
95325e6b
AG		 677 } else if (vma && hva >= vma->vm_start &&
678 (vma->vm_flags & VM_HUGETLB)) {
679 unsigned long psize = vma_kernel_pagesize(vma);
680
681 tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >>
682 MAS1_TSIZE_SHIFT;
683
684 /*
685 * Take the largest page size that satisfies both host
686 * and guest mapping
687 */
688 tsize = min(__ilog2(psize) - 10, tsize);
689
690 /*
691 * e500 doesn't implement the lowest tsize bit,
692 * or 1K pages.
693 */
694 tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1);
9973d54e
SW		 695 }
696
697 up_read(&current->mm->mmap_sem);
698 }
699
700 if (likely(!pfnmap)) {
95325e6b 701 unsigned long tsize_pages = 1 << (tsize + 10 - PAGE_SHIFT);
9973d54e
SW		 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",
705 (long)gfn);
706 kvm_release_pfn_clean(pfn);
707 return;
708 }
95325e6b
AG		 709
710 /* Align guest and physical address to page map boundaries */
711 pfn &= ~(tsize_pages - 1);
712 gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1);
bc8080cb 713 }
bc8080cb 714
0164c0f0
SW		 715 /* Drop old ref and setup new one. */
716 kvmppc_e500_ref_release(ref);
717 kvmppc_e500_ref_setup(ref, gtlbe, pfn);
bc8080cb 718
0164c0f0 719 kvmppc_e500_setup_stlbe(vcpu_e500, gtlbe, tsize, ref, gvaddr, stlbe);
bc8080cb
HB		 720}
721
722/* XXX only map the one-one case, for now use TLB0 */
08b7fa92 723static int kvmppc_e500_tlb0_map(struct kvmppc_vcpu_e500 *vcpu_e500,
dc83b8bc
SW		 724 int esel,
725 struct kvm_book3e_206_tlb_entry *stlbe)
bc8080cb 726{
dc83b8bc 727 struct kvm_book3e_206_tlb_entry *gtlbe;
0164c0f0
SW		 728 struct tlbe_ref *ref;
729 int sesel = esel & (host_tlb_params[0].ways - 1);
730 int sesel_base;
731 gva_t ea;
bc8080cb 732
dc83b8bc 733 gtlbe = get_entry(vcpu_e500, 0, esel);
0164c0f0
SW		 734 ref = &vcpu_e500->gtlb_priv[0][esel].ref;
735
736 ea = get_tlb_eaddr(gtlbe);
737 sesel_base = htlb0_set_base(ea);
bc8080cb
HB		 738
739 kvmppc_e500_shadow_map(vcpu_e500, get_tlb_eaddr(gtlbe),
740 get_tlb_raddr(gtlbe) >> PAGE_SHIFT,
0164c0f0 741 gtlbe, 0, sesel_base + sesel, stlbe, ref);
bc8080cb 742
0164c0f0 743 return sesel;
bc8080cb
HB		 744}
745
746/* Caller must ensure that the specified guest TLB entry is safe to insert into
747 * the shadow TLB. */
748/* XXX for both one-one and one-to-many , for now use TLB1 */
749static int kvmppc_e500_tlb1_map(struct kvmppc_vcpu_e500 *vcpu_e500,
dc83b8bc
SW		 750 u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe,
751 struct kvm_book3e_206_tlb_entry *stlbe)
bc8080cb 752{
0164c0f0 753 struct tlbe_ref *ref;
bc8080cb
HB		 754 unsigned int victim;
755
0164c0f0 756 victim = vcpu_e500->host_tlb1_nv++;
bc8080cb 757
0164c0f0
SW		 758 if (unlikely(vcpu_e500->host_tlb1_nv >= tlb1_max_shadow_size()))
759 vcpu_e500->host_tlb1_nv = 0;
bc8080cb 760
0164c0f0
SW		 761 ref = &vcpu_e500->tlb_refs[1][victim];
762 kvmppc_e500_shadow_map(vcpu_e500, gvaddr, gfn, gtlbe, 1,
763 victim, stlbe, ref);
bc8080cb
HB		 764
765 return victim;
766}
767
dd9ebf1f 768void kvmppc_mmu_msr_notify(struct kvm_vcpu *vcpu, u32 old_msr)
bc8080cb 769{
dd9ebf1f
LY		 770 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
771
772 /* Recalc shadow pid since MSR changes */
773 kvmppc_e500_recalc_shadow_pid(vcpu_e500);
bc8080cb
HB		 774}
775
08b7fa92
LY		 776static inline int kvmppc_e500_gtlbe_invalidate(
777 struct kvmppc_vcpu_e500 *vcpu_e500,
778 int tlbsel, int esel)
bc8080cb 779{
dc83b8bc
SW		 780 struct kvm_book3e_206_tlb_entry *gtlbe =
781 get_entry(vcpu_e500, tlbsel, esel);
bc8080cb
HB		 782
783 if (unlikely(get_tlb_iprot(gtlbe)))
784 return -1;
785
bc8080cb
HB		 786 gtlbe->mas1 = 0;
787
788 return 0;
789}
790
b0a1835d
LY		 791int kvmppc_e500_emul_mt_mmucsr0(struct kvmppc_vcpu_e500 *vcpu_e500, ulong value)
792{
793 int esel;
794
795 if (value & MMUCSR0_TLB0FI)
dc83b8bc 796 for (esel = 0; esel < vcpu_e500->gtlb_params[0].entries; esel++)
b0a1835d
LY		 797 kvmppc_e500_gtlbe_invalidate(vcpu_e500, 0, esel);
798 if (value & MMUCSR0_TLB1FI)
dc83b8bc 799 for (esel = 0; esel < vcpu_e500->gtlb_params[1].entries; esel++)
b0a1835d
LY		 800 kvmppc_e500_gtlbe_invalidate(vcpu_e500, 1, esel);
801
dd9ebf1f
LY		 802 /* Invalidate all vcpu id mappings */
803 kvmppc_e500_id_table_reset_all(vcpu_e500);
b0a1835d
LY		 804
805 return EMULATE_DONE;
806}
807
bc8080cb
HB		 808int kvmppc_e500_emul_tlbivax(struct kvm_vcpu *vcpu, int ra, int rb)
809{
810 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
811 unsigned int ia;
812 int esel, tlbsel;
813 gva_t ea;
814
8e5b26b5 815 ea = ((ra) ? kvmppc_get_gpr(vcpu, ra) : 0) + kvmppc_get_gpr(vcpu, rb);
bc8080cb
HB		 816
817 ia = (ea >> 2) & 0x1;
818
fb2838d4 819 /* since we only have two TLBs, only lower bit is used. */
bc8080cb
HB		 820 tlbsel = (ea >> 3) & 0x1;
821
822 if (ia) {
823 /* invalidate all entries */
dc83b8bc
SW		 824 for (esel = 0; esel < vcpu_e500->gtlb_params[tlbsel].entries;
825 esel++)
bc8080cb
HB		 826 kvmppc_e500_gtlbe_invalidate(vcpu_e500, tlbsel, esel);
827 } else {
828 ea &= 0xfffff000;
829 esel = kvmppc_e500_tlb_index(vcpu_e500, ea, tlbsel,
830 get_cur_pid(vcpu), -1);
831 if (esel >= 0)
832 kvmppc_e500_gtlbe_invalidate(vcpu_e500, tlbsel, esel);
833 }
834
dd9ebf1f
LY		 835 /* Invalidate all vcpu id mappings */
836 kvmppc_e500_id_table_reset_all(vcpu_e500);
bc8080cb
HB		 837
838 return EMULATE_DONE;
839}
840
841int kvmppc_e500_emul_tlbre(struct kvm_vcpu *vcpu)
842{
843 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
844 int tlbsel, esel;
dc83b8bc 845 struct kvm_book3e_206_tlb_entry *gtlbe;
bc8080cb
HB		 846
847 tlbsel = get_tlb_tlbsel(vcpu_e500);
848 esel = get_tlb_esel(vcpu_e500, tlbsel);
849
dc83b8bc 850 gtlbe = get_entry(vcpu_e500, tlbsel, esel);
bc35cbc8 851 vcpu_e500->mas0 &= ~MAS0_NV(~0);
08b7fa92 852 vcpu_e500->mas0 |= MAS0_NV(vcpu_e500->gtlb_nv[tlbsel]);
bc8080cb
HB		 853 vcpu_e500->mas1 = gtlbe->mas1;
854 vcpu_e500->mas2 = gtlbe->mas2;
dc83b8bc 855 vcpu_e500->mas7_3 = gtlbe->mas7_3;
bc8080cb
HB		 856
857 return EMULATE_DONE;
858}
859
860int kvmppc_e500_emul_tlbsx(struct kvm_vcpu *vcpu, int rb)
861{
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);
865 int esel, tlbsel;
dc83b8bc 866 struct kvm_book3e_206_tlb_entry *gtlbe = NULL;
bc8080cb
HB		 867 gva_t ea;
868
8e5b26b5 869 ea = kvmppc_get_gpr(vcpu, rb);
bc8080cb
HB		 870
871 for (tlbsel = 0; tlbsel < 2; tlbsel++) {
872 esel = kvmppc_e500_tlb_index(vcpu_e500, ea, tlbsel, pid, as);
873 if (esel >= 0) {
dc83b8bc 874 gtlbe = get_entry(vcpu_e500, tlbsel, esel);
bc8080cb
HB		 875 break;
876 }
877 }
878
879 if (gtlbe) {
303b7c97
SW		 880 esel &= vcpu_e500->gtlb_params[tlbsel].ways - 1;
881
bc8080cb 882 vcpu_e500->mas0 = MAS0_TLBSEL(tlbsel) | MAS0_ESEL(esel)
08b7fa92 883 | MAS0_NV(vcpu_e500->gtlb_nv[tlbsel]);
bc8080cb
HB		 884 vcpu_e500->mas1 = gtlbe->mas1;
885 vcpu_e500->mas2 = gtlbe->mas2;
dc83b8bc 886 vcpu_e500->mas7_3 = gtlbe->mas7_3;
bc8080cb
HB		 887 } else {
888 int victim;
889
fb2838d4 890 /* since we only have two TLBs, only lower bit is used. */
bc8080cb 891 tlbsel = vcpu_e500->mas4 >> 28 & 0x1;
0164c0f0 892 victim = (tlbsel == 0) ? gtlb0_get_next_victim(vcpu_e500) : 0;
bc8080cb
HB		 893
894 vcpu_e500->mas0 = MAS0_TLBSEL(tlbsel) | MAS0_ESEL(victim)
08b7fa92 895 | MAS0_NV(vcpu_e500->gtlb_nv[tlbsel]);
bc8080cb
HB		 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;
dc83b8bc 901 vcpu_e500->mas7_3 &= MAS3_U0 | MAS3_U1 | MAS3_U2 | MAS3_U3;
bc8080cb
HB		 902 }
903
49ea0695 904 kvmppc_set_exit_type(vcpu, EMULATED_TLBSX_EXITS);
bc8080cb
HB		 905 return EMULATE_DONE;
906}
907
3bf3cdcc
SW		 908/* sesel is index into the set, not the whole array */
909static void write_stlbe(struct kvmppc_vcpu_e500 *vcpu_e500,
dc83b8bc
SW		 910 struct kvm_book3e_206_tlb_entry *gtlbe,
911 struct kvm_book3e_206_tlb_entry *stlbe,
3bf3cdcc
SW		 912 int stlbsel, int sesel)
913{
914 int stid;
915
916 preempt_disable();
917 stid = kvmppc_e500_get_sid(vcpu_e500, get_tlb_ts(gtlbe),
918 get_tlb_tid(gtlbe),
919 get_cur_pr(&vcpu_e500->vcpu), 0);
920
921 stlbe->mas1 |= MAS1_TID(stid);
922 write_host_tlbe(vcpu_e500, stlbsel, sesel, stlbe);
923 preempt_enable();
924}
925
bc8080cb
HB		 926int kvmppc_e500_emul_tlbwe(struct kvm_vcpu *vcpu)
927{
928 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
dc83b8bc 929 struct kvm_book3e_206_tlb_entry *gtlbe;
08b7fa92 930 int tlbsel, esel;
bc8080cb
HB		 931
932 tlbsel = get_tlb_tlbsel(vcpu_e500);
933 esel = get_tlb_esel(vcpu_e500, tlbsel);
934
dc83b8bc 935 gtlbe = get_entry(vcpu_e500, tlbsel, esel);
bc8080cb 936
dd9ebf1f 937 if (get_tlb_v(gtlbe))
0164c0f0 938 inval_gtlbe_on_host(vcpu_e500, tlbsel, esel);
bc8080cb
HB		 939
940 gtlbe->mas1 = vcpu_e500->mas1;
941 gtlbe->mas2 = vcpu_e500->mas2;
dc83b8bc 942 gtlbe->mas7_3 = vcpu_e500->mas7_3;
bc8080cb 943
46f43c6e 944 trace_kvm_gtlb_write(vcpu_e500->mas0, gtlbe->mas1, gtlbe->mas2,
dc83b8bc 945 (u32)gtlbe->mas7_3, (u32)(gtlbe->mas7_3 >> 32));
bc8080cb
HB		 946
947 /* Invalidate shadow mappings for the about-to-be-clobbered TLBE. */
948 if (tlbe_is_host_safe(vcpu, gtlbe)) {
dc83b8bc 949 struct kvm_book3e_206_tlb_entry stlbe;
08b7fa92
LY		 950 int stlbsel, sesel;
951 u64 eaddr;
952 u64 raddr;
953
bc8080cb
HB		 954 switch (tlbsel) {
955 case 0:
956 /* TLB0 */
957 gtlbe->mas1 &= ~MAS1_TSIZE(~0);
0cfb50e5 958 gtlbe->mas1 |= MAS1_TSIZE(BOOK3E_PAGESZ_4K);
bc8080cb
HB		 959
960 stlbsel = 0;
08b7fa92 961 sesel = kvmppc_e500_tlb0_map(vcpu_e500, esel, &stlbe);
bc8080cb
HB		 962
963 break;
964
965 case 1:
966 /* TLB1 */
967 eaddr = get_tlb_eaddr(gtlbe);
968 raddr = get_tlb_raddr(gtlbe);
969
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. */
974 stlbsel = 1;
975 sesel = kvmppc_e500_tlb1_map(vcpu_e500, eaddr,
08b7fa92 976 raddr >> PAGE_SHIFT, gtlbe, &stlbe);
bc8080cb
HB		 977 break;
978
979 default:
980 BUG();
981 }
3bf3cdcc
SW		 982
983 write_stlbe(vcpu_e500, gtlbe, &stlbe, stlbsel, sesel);
bc8080cb
HB		 984 }
985
49ea0695 986 kvmppc_set_exit_type(vcpu, EMULATED_TLBWE_EXITS);
bc8080cb
HB		 987 return EMULATE_DONE;
988}
989
990int kvmppc_mmu_itlb_index(struct kvm_vcpu *vcpu, gva_t eaddr)
991{
666e7252 992 unsigned int as = !!(vcpu->arch.shared->msr & MSR_IS);
bc8080cb
HB		 993
994 return kvmppc_e500_tlb_search(vcpu, eaddr, get_cur_pid(vcpu), as);
995}
996
997int kvmppc_mmu_dtlb_index(struct kvm_vcpu *vcpu, gva_t eaddr)
998{
666e7252 999 unsigned int as = !!(vcpu->arch.shared->msr & MSR_DS);
bc8080cb
HB		 1000
1001 return kvmppc_e500_tlb_search(vcpu, eaddr, get_cur_pid(vcpu), as);
1002}
1003
1004void kvmppc_mmu_itlb_miss(struct kvm_vcpu *vcpu)
1005{
666e7252 1006 unsigned int as = !!(vcpu->arch.shared->msr & MSR_IS);
bc8080cb
HB		 1007
1008 kvmppc_e500_deliver_tlb_miss(vcpu, vcpu->arch.pc, as);
1009}
1010
1011void kvmppc_mmu_dtlb_miss(struct kvm_vcpu *vcpu)
1012{
666e7252 1013 unsigned int as = !!(vcpu->arch.shared->msr & MSR_DS);
bc8080cb
HB		 1014
1015 kvmppc_e500_deliver_tlb_miss(vcpu, vcpu->arch.fault_dear, as);
1016}
1017
1018gpa_t kvmppc_mmu_xlate(struct kvm_vcpu *vcpu, unsigned int index,
1019 gva_t eaddr)
1020{
1021 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
dc83b8bc
SW		 1022 struct kvm_book3e_206_tlb_entry *gtlbe;
1023 u64 pgmask;
1024
1025 gtlbe = get_entry(vcpu_e500, tlbsel_of(index), esel_of(index));
1026 pgmask = get_tlb_bytes(gtlbe) - 1;
bc8080cb
HB		 1027
1028 return get_tlb_raddr(gtlbe) | (eaddr & pgmask);
1029}
1030
1031void kvmppc_mmu_destroy(struct kvm_vcpu *vcpu)
1032{
bc8080cb
HB		 1033}
1034
1035void kvmppc_mmu_map(struct kvm_vcpu *vcpu, u64 eaddr, gpa_t gpaddr,
1036 unsigned int index)
1037{
1038 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
08b7fa92 1039 struct tlbe_priv *priv;
dc83b8bc 1040 struct kvm_book3e_206_tlb_entry *gtlbe, stlbe;
bc8080cb
HB		 1041 int tlbsel = tlbsel_of(index);
1042 int esel = esel_of(index);
1043 int stlbsel, sesel;
1044
dc83b8bc 1045 gtlbe = get_entry(vcpu_e500, tlbsel, esel);
08b7fa92 1046
bc8080cb
HB		 1047 switch (tlbsel) {
1048 case 0:
1049 stlbsel = 0;
0164c0f0
SW		 1050 sesel = esel & (host_tlb_params[0].ways - 1);
1051 priv = &vcpu_e500->gtlb_priv[tlbsel][esel];
08b7fa92
LY		 1052
1053 kvmppc_e500_setup_stlbe(vcpu_e500, gtlbe, BOOK3E_PAGESZ_4K,
0164c0f0 1054 &priv->ref, eaddr, &stlbe);
bc8080cb
HB		 1055 break;
1056
1057 case 1: {
1058 gfn_t gfn = gpaddr >> PAGE_SHIFT;
bc8080cb
HB		 1059
1060 stlbsel = 1;
08b7fa92
LY		 1061 sesel = kvmppc_e500_tlb1_map(vcpu_e500, eaddr, gfn,
1062 gtlbe, &stlbe);
bc8080cb
HB		 1063 break;
1064 }
1065
1066 default:
1067 BUG();
1068 break;
1069 }
08b7fa92 1070
3bf3cdcc 1071 write_stlbe(vcpu_e500, gtlbe, &stlbe, stlbsel, sesel);
bc8080cb
HB		 1072}
1073
1074int kvmppc_e500_tlb_search(struct kvm_vcpu *vcpu,
1075 gva_t eaddr, unsigned int pid, int as)
1076{
1077 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
1078 int esel, tlbsel;
1079
1080 for (tlbsel = 0; tlbsel < 2; tlbsel++) {
1081 esel = kvmppc_e500_tlb_index(vcpu_e500, eaddr, tlbsel, pid, as);
1082 if (esel >= 0)
1083 return index_of(tlbsel, esel);
1084 }
1085
1086 return -1;
1087}
1088
5ce941ee
SW		 1089void kvmppc_set_pid(struct kvm_vcpu *vcpu, u32 pid)
1090{
1091 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
1092
dd9ebf1f
LY		 1093 if (vcpu->arch.pid != pid) {
1094 vcpu_e500->pid[0] = vcpu->arch.pid = pid;
1095 kvmppc_e500_recalc_shadow_pid(vcpu_e500);
1096 }
5ce941ee
SW		 1097}
1098
bc8080cb
HB		 1099void kvmppc_e500_tlb_setup(struct kvmppc_vcpu_e500 *vcpu_e500)
1100{
dc83b8bc 1101 struct kvm_book3e_206_tlb_entry *tlbe;
bc8080cb
HB		 1102
1103 /* Insert large initial mapping for guest. */
dc83b8bc 1104 tlbe = get_entry(vcpu_e500, 1, 0);
0cfb50e5 1105 tlbe->mas1 = MAS1_VALID | MAS1_TSIZE(BOOK3E_PAGESZ_256M);
bc8080cb 1106 tlbe->mas2 = 0;
dc83b8bc 1107 tlbe->mas7_3 = E500_TLB_SUPER_PERM_MASK;
bc8080cb
HB		 1108
1109 /* 4K map for serial output. Used by kernel wrapper. */
dc83b8bc 1110 tlbe = get_entry(vcpu_e500, 1, 1);
0cfb50e5 1111 tlbe->mas1 = MAS1_VALID | MAS1_TSIZE(BOOK3E_PAGESZ_4K);
bc8080cb 1112 tlbe->mas2 = (0xe0004500 & 0xFFFFF000) | MAS2_I | MAS2_G;
dc83b8bc
SW		 1113 tlbe->mas7_3 = (0xe0004500 & 0xFFFFF000) | E500_TLB_SUPER_PERM_MASK;
1114}
1115
1116static void free_gtlb(struct kvmppc_vcpu_e500 *vcpu_e500)
1117{
1118 int i;
1119
1120 clear_tlb_refs(vcpu_e500);
1121 kfree(vcpu_e500->gtlb_priv[0]);
1122 kfree(vcpu_e500->gtlb_priv[1]);
1123
1124 if (vcpu_e500->shared_tlb_pages) {
1125 vfree((void *)(round_down((uintptr_t)vcpu_e500->gtlb_arch,
1126 PAGE_SIZE)));
1127
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]);
1131 }
1132
1133 vcpu_e500->num_shared_tlb_pages = 0;
1134 vcpu_e500->shared_tlb_pages = NULL;
1135 } else {
1136 kfree(vcpu_e500->gtlb_arch);
1137 }
1138
1139 vcpu_e500->gtlb_arch = NULL;
1140}
1141
1142int kvm_vcpu_ioctl_config_tlb(struct kvm_vcpu *vcpu,
1143 struct kvm_config_tlb *cfg)
1144{
1145 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
1146 struct kvm_book3e_206_tlb_params params;
1147 char *virt;
1148 struct page **pages;
1149 struct tlbe_priv *privs[2] = {};
1150 size_t array_len;
1151 u32 sets;
1152 int num_pages, ret, i;
1153
1154 if (cfg->mmu_type != KVM_MMU_FSL_BOOKE_NOHV)
1155 return -EINVAL;
1156
1157 if (copy_from_user(&params, (void __user *)(uintptr_t)cfg->params,
1158 sizeof(params)))
1159 return -EFAULT;
1160
1161 if (params.tlb_sizes[1] > 64)
1162 return -EINVAL;
1163 if (params.tlb_ways[1] != params.tlb_sizes[1])
1164 return -EINVAL;
1165 if (params.tlb_sizes[2] != 0 || params.tlb_sizes[3] != 0)
1166 return -EINVAL;
1167 if (params.tlb_ways[2] != 0 || params.tlb_ways[3] != 0)
1168 return -EINVAL;
1169
1170 if (!is_power_of_2(params.tlb_ways[0]))
1171 return -EINVAL;
1172
1173 sets = params.tlb_sizes[0] >> ilog2(params.tlb_ways[0]);
1174 if (!is_power_of_2(sets))
1175 return -EINVAL;
1176
1177 array_len = params.tlb_sizes[0] + params.tlb_sizes[1];
1178 array_len *= sizeof(struct kvm_book3e_206_tlb_entry);
1179
1180 if (cfg->array_len < array_len)
1181 return -EINVAL;
1182
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);
1186 if (!pages)
1187 return -ENOMEM;
1188
1189 ret = get_user_pages_fast(cfg->array, num_pages, 1, pages);
1190 if (ret < 0)
1191 goto err_pages;
1192
1193 if (ret != num_pages) {
1194 num_pages = ret;
1195 ret = -EFAULT;
1196 goto err_put_page;
1197 }
1198
1199 virt = vmap(pages, num_pages, VM_MAP, PAGE_KERNEL);
1200 if (!virt)
1201 goto err_put_page;
1202
1203 privs[0] = kzalloc(sizeof(struct tlbe_priv) * params.tlb_sizes[0],
1204 GFP_KERNEL);
1205 privs[1] = kzalloc(sizeof(struct tlbe_priv) * params.tlb_sizes[1],
1206 GFP_KERNEL);
1207
1208 if (!privs[0] || !privs[1])
1209 goto err_put_page;
1210
1211 free_gtlb(vcpu_e500);
1212
1213 vcpu_e500->gtlb_priv[0] = privs[0];
1214 vcpu_e500->gtlb_priv[1] = privs[1];
1215
1216 vcpu_e500->gtlb_arch = (struct kvm_book3e_206_tlb_entry *)
1217 (virt + (cfg->array & (PAGE_SIZE - 1)));
1218
1219 vcpu_e500->gtlb_params[0].entries = params.tlb_sizes[0];
1220 vcpu_e500->gtlb_params[1].entries = params.tlb_sizes[1];
1221
1222 vcpu_e500->gtlb_offset[0] = 0;
1223 vcpu_e500->gtlb_offset[1] = params.tlb_sizes[0];
1224
1225 vcpu_e500->tlb0cfg = mfspr(SPRN_TLB0CFG) & ~0xfffUL;
1226 if (params.tlb_sizes[0] <= 2048)
1227 vcpu_e500->tlb0cfg |= params.tlb_sizes[0];
1228
1229 vcpu_e500->tlb1cfg = mfspr(SPRN_TLB1CFG) & ~0xfffUL;
1230 vcpu_e500->tlb1cfg |= params.tlb_sizes[1];
1231
1232 vcpu_e500->shared_tlb_pages = pages;
1233 vcpu_e500->num_shared_tlb_pages = num_pages;
1234
1235 vcpu_e500->gtlb_params[0].ways = params.tlb_ways[0];
1236 vcpu_e500->gtlb_params[0].sets = sets;
1237
1238 vcpu_e500->gtlb_params[1].ways = params.tlb_sizes[1];
1239 vcpu_e500->gtlb_params[1].sets = 1;
1240
1241 return 0;
1242
1243err_put_page:
1244 kfree(privs[0]);
1245 kfree(privs[1]);
1246
1247 for (i = 0; i < num_pages; i++)
1248 put_page(pages[i]);
1249
1250err_pages:
1251 kfree(pages);
1252 return ret;
1253}
1254
1255int kvm_vcpu_ioctl_dirty_tlb(struct kvm_vcpu *vcpu,
1256 struct kvm_dirty_tlb *dirty)
1257{
1258 struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
1259
1260 clear_tlb_refs(vcpu_e500);
1261 return 0;
bc8080cb
HB		 1262}
1263
1264int kvmppc_e500_tlb_init(struct kvmppc_vcpu_e500 *vcpu_e500)
1265{
dc83b8bc
SW		 1266 int entry_size = sizeof(struct kvm_book3e_206_tlb_entry);
1267 int entries = KVM_E500_TLB0_SIZE + KVM_E500_TLB1_SIZE;
1268
0164c0f0
SW		 1269 host_tlb_params[0].entries = mfspr(SPRN_TLB0CFG) & TLBnCFG_N_ENTRY;
1270 host_tlb_params[1].entries = mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY;
1271
1272 /*
1273 * This should never happen on real e500 hardware, but is
1274 * architecturally possible -- e.g. in some weird nested
1275 * virtualization case.
1276 */
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__);
1280 return -ENODEV;
1281 }
1282
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;
1286
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);
1294 return -ENODEV;
1295 }
1296
1297 host_tlb_params[0].sets =
1298 host_tlb_params[0].entries / host_tlb_params[0].ways;
1299 host_tlb_params[1].sets = 1;
bc8080cb 1300
dc83b8bc
SW		 1301 vcpu_e500->gtlb_params[0].entries = KVM_E500_TLB0_SIZE;
1302 vcpu_e500->gtlb_params[1].entries = KVM_E500_TLB1_SIZE;
bc8080cb 1303
dc83b8bc
SW		 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;
1307
1308 vcpu_e500->gtlb_params[1].ways = KVM_E500_TLB1_SIZE;
1309 vcpu_e500->gtlb_params[1].sets = 1;
1310
1311 vcpu_e500->gtlb_arch = kmalloc(entries * entry_size, GFP_KERNEL);
1312 if (!vcpu_e500->gtlb_arch)
1313 return -ENOMEM;
1314
1315 vcpu_e500->gtlb_offset[0] = 0;
1316 vcpu_e500->gtlb_offset[1] = KVM_E500_TLB0_SIZE;
0164c0f0
SW		 1317
1318 vcpu_e500->tlb_refs[0] =
1319 kzalloc(sizeof(struct tlbe_ref) * host_tlb_params[0].entries,
1320 GFP_KERNEL);
1321 if (!vcpu_e500->tlb_refs[0])
1322 goto err;
1323
1324 vcpu_e500->tlb_refs[1] =
1325 kzalloc(sizeof(struct tlbe_ref) * host_tlb_params[1].entries,
1326 GFP_KERNEL);
1327 if (!vcpu_e500->tlb_refs[1])
1328 goto err;
1329
dc83b8bc
SW		 1330 vcpu_e500->gtlb_priv[0] = kzalloc(sizeof(struct tlbe_ref) *
1331 vcpu_e500->gtlb_params[0].entries,
1332 GFP_KERNEL);
0164c0f0
SW		 1333 if (!vcpu_e500->gtlb_priv[0])
1334 goto err;
1335
dc83b8bc
SW		 1336 vcpu_e500->gtlb_priv[1] = kzalloc(sizeof(struct tlbe_ref) *
1337 vcpu_e500->gtlb_params[1].entries,
1338 GFP_KERNEL);
0164c0f0
SW		 1339 if (!vcpu_e500->gtlb_priv[1])
1340 goto err;
bc8080cb 1341
dd9ebf1f 1342 if (kvmppc_e500_id_table_alloc(vcpu_e500) == NULL)
0164c0f0 1343 goto err;
dd9ebf1f 1344
da15bf43
LY		 1345 /* Init TLB configuration register */
1346 vcpu_e500->tlb0cfg = mfspr(SPRN_TLB0CFG) & ~0xfffUL;
dc83b8bc 1347 vcpu_e500->tlb0cfg |= vcpu_e500->gtlb_params[0].entries;
da15bf43 1348 vcpu_e500->tlb1cfg = mfspr(SPRN_TLB1CFG) & ~0xfffUL;
dc83b8bc 1349 vcpu_e500->tlb1cfg |= vcpu_e500->gtlb_params[1].entries;
da15bf43 1350
bc8080cb
HB		 1351 return 0;
1352
0164c0f0 1353err:
dc83b8bc 1354 free_gtlb(vcpu_e500);
0164c0f0
SW		 1355 kfree(vcpu_e500->tlb_refs[0]);
1356 kfree(vcpu_e500->tlb_refs[1]);
bc8080cb
HB		 1357 return -1;
1358}
1359
1360void kvmppc_e500_tlb_uninit(struct kvmppc_vcpu_e500 *vcpu_e500)
1361{
dc83b8bc 1362 free_gtlb(vcpu_e500);
dd9ebf1f 1363 kvmppc_e500_id_table_free(vcpu_e500);
0164c0f0
SW		 1364
1365 kfree(vcpu_e500->tlb_refs[0]);
1366 kfree(vcpu_e500->tlb_refs[1]);
bc8080cb 1367}
Linux 6.x block layer and io_uring tree(s)