kvm tools, setup: Create private directory
[linux-2.6/next.git] / arch / powerpc / mm / hash_native_64.c
blobdfd764896db07bd26255648ec60a465006061ad2
1 /*
2 * native hashtable management.
4 * SMP scalability work:
5 * Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
13 #undef DEBUG_LOW
15 #include <linux/spinlock.h>
16 #include <linux/bitops.h>
17 #include <linux/threads.h>
18 #include <linux/smp.h>
20 #include <asm/abs_addr.h>
21 #include <asm/machdep.h>
22 #include <asm/mmu.h>
23 #include <asm/mmu_context.h>
24 #include <asm/pgtable.h>
25 #include <asm/tlbflush.h>
26 #include <asm/tlb.h>
27 #include <asm/cputable.h>
28 #include <asm/udbg.h>
29 #include <asm/kexec.h>
30 #include <asm/ppc-opcode.h>
32 #ifdef DEBUG_LOW
33 #define DBG_LOW(fmt...) udbg_printf(fmt)
34 #else
35 #define DBG_LOW(fmt...)
36 #endif
38 #define HPTE_LOCK_BIT 3
40 static DEFINE_RAW_SPINLOCK(native_tlbie_lock);
42 static inline void __tlbie(unsigned long va, int psize, int ssize)
44 unsigned int penc;
46 /* clear top 16 bits, non SLS segment */
47 va &= ~(0xffffULL << 48);
49 switch (psize) {
50 case MMU_PAGE_4K:
51 va &= ~0xffful;
52 va |= ssize << 8;
53 asm volatile(ASM_FTR_IFCLR("tlbie %0,0", PPC_TLBIE(%1,%0), %2)
54 : : "r" (va), "r"(0), "i" (CPU_FTR_HVMODE_206)
55 : "memory");
56 break;
57 default:
58 penc = mmu_psize_defs[psize].penc;
59 va &= ~((1ul << mmu_psize_defs[psize].shift) - 1);
60 va |= penc << 12;
61 va |= ssize << 8;
62 va |= 1; /* L */
63 asm volatile(ASM_FTR_IFCLR("tlbie %0,1", PPC_TLBIE(%1,%0), %2)
64 : : "r" (va), "r"(0), "i" (CPU_FTR_HVMODE_206)
65 : "memory");
66 break;
70 static inline void __tlbiel(unsigned long va, int psize, int ssize)
72 unsigned int penc;
74 /* clear top 16 bits, non SLS segment */
75 va &= ~(0xffffULL << 48);
77 switch (psize) {
78 case MMU_PAGE_4K:
79 va &= ~0xffful;
80 va |= ssize << 8;
81 asm volatile(".long 0x7c000224 | (%0 << 11) | (0 << 21)"
82 : : "r"(va) : "memory");
83 break;
84 default:
85 penc = mmu_psize_defs[psize].penc;
86 va &= ~((1ul << mmu_psize_defs[psize].shift) - 1);
87 va |= penc << 12;
88 va |= ssize << 8;
89 va |= 1; /* L */
90 asm volatile(".long 0x7c000224 | (%0 << 11) | (1 << 21)"
91 : : "r"(va) : "memory");
92 break;
97 static inline void tlbie(unsigned long va, int psize, int ssize, int local)
99 unsigned int use_local = local && mmu_has_feature(MMU_FTR_TLBIEL);
100 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
102 if (use_local)
103 use_local = mmu_psize_defs[psize].tlbiel;
104 if (lock_tlbie && !use_local)
105 raw_spin_lock(&native_tlbie_lock);
106 asm volatile("ptesync": : :"memory");
107 if (use_local) {
108 __tlbiel(va, psize, ssize);
109 asm volatile("ptesync": : :"memory");
110 } else {
111 __tlbie(va, psize, ssize);
112 asm volatile("eieio; tlbsync; ptesync": : :"memory");
114 if (lock_tlbie && !use_local)
115 raw_spin_unlock(&native_tlbie_lock);
118 static inline void native_lock_hpte(struct hash_pte *hptep)
120 unsigned long *word = &hptep->v;
122 while (1) {
123 if (!test_and_set_bit_lock(HPTE_LOCK_BIT, word))
124 break;
125 while(test_bit(HPTE_LOCK_BIT, word))
126 cpu_relax();
130 static inline void native_unlock_hpte(struct hash_pte *hptep)
132 unsigned long *word = &hptep->v;
134 clear_bit_unlock(HPTE_LOCK_BIT, word);
137 static long native_hpte_insert(unsigned long hpte_group, unsigned long va,
138 unsigned long pa, unsigned long rflags,
139 unsigned long vflags, int psize, int ssize)
141 struct hash_pte *hptep = htab_address + hpte_group;
142 unsigned long hpte_v, hpte_r;
143 int i;
145 if (!(vflags & HPTE_V_BOLTED)) {
146 DBG_LOW(" insert(group=%lx, va=%016lx, pa=%016lx,"
147 " rflags=%lx, vflags=%lx, psize=%d)\n",
148 hpte_group, va, pa, rflags, vflags, psize);
151 for (i = 0; i < HPTES_PER_GROUP; i++) {
152 if (! (hptep->v & HPTE_V_VALID)) {
153 /* retry with lock held */
154 native_lock_hpte(hptep);
155 if (! (hptep->v & HPTE_V_VALID))
156 break;
157 native_unlock_hpte(hptep);
160 hptep++;
163 if (i == HPTES_PER_GROUP)
164 return -1;
166 hpte_v = hpte_encode_v(va, psize, ssize) | vflags | HPTE_V_VALID;
167 hpte_r = hpte_encode_r(pa, psize) | rflags;
169 if (!(vflags & HPTE_V_BOLTED)) {
170 DBG_LOW(" i=%x hpte_v=%016lx, hpte_r=%016lx\n",
171 i, hpte_v, hpte_r);
174 hptep->r = hpte_r;
175 /* Guarantee the second dword is visible before the valid bit */
176 eieio();
178 * Now set the first dword including the valid bit
179 * NOTE: this also unlocks the hpte
181 hptep->v = hpte_v;
183 __asm__ __volatile__ ("ptesync" : : : "memory");
185 return i | (!!(vflags & HPTE_V_SECONDARY) << 3);
188 static long native_hpte_remove(unsigned long hpte_group)
190 struct hash_pte *hptep;
191 int i;
192 int slot_offset;
193 unsigned long hpte_v;
195 DBG_LOW(" remove(group=%lx)\n", hpte_group);
197 /* pick a random entry to start at */
198 slot_offset = mftb() & 0x7;
200 for (i = 0; i < HPTES_PER_GROUP; i++) {
201 hptep = htab_address + hpte_group + slot_offset;
202 hpte_v = hptep->v;
204 if ((hpte_v & HPTE_V_VALID) && !(hpte_v & HPTE_V_BOLTED)) {
205 /* retry with lock held */
206 native_lock_hpte(hptep);
207 hpte_v = hptep->v;
208 if ((hpte_v & HPTE_V_VALID)
209 && !(hpte_v & HPTE_V_BOLTED))
210 break;
211 native_unlock_hpte(hptep);
214 slot_offset++;
215 slot_offset &= 0x7;
218 if (i == HPTES_PER_GROUP)
219 return -1;
221 /* Invalidate the hpte. NOTE: this also unlocks it */
222 hptep->v = 0;
224 return i;
227 static long native_hpte_updatepp(unsigned long slot, unsigned long newpp,
228 unsigned long va, int psize, int ssize,
229 int local)
231 struct hash_pte *hptep = htab_address + slot;
232 unsigned long hpte_v, want_v;
233 int ret = 0;
235 want_v = hpte_encode_v(va, psize, ssize);
237 DBG_LOW(" update(va=%016lx, avpnv=%016lx, hash=%016lx, newpp=%x)",
238 va, want_v & HPTE_V_AVPN, slot, newpp);
240 native_lock_hpte(hptep);
242 hpte_v = hptep->v;
244 /* Even if we miss, we need to invalidate the TLB */
245 if (!HPTE_V_COMPARE(hpte_v, want_v) || !(hpte_v & HPTE_V_VALID)) {
246 DBG_LOW(" -> miss\n");
247 ret = -1;
248 } else {
249 DBG_LOW(" -> hit\n");
250 /* Update the HPTE */
251 hptep->r = (hptep->r & ~(HPTE_R_PP | HPTE_R_N)) |
252 (newpp & (HPTE_R_PP | HPTE_R_N | HPTE_R_C));
254 native_unlock_hpte(hptep);
256 /* Ensure it is out of the tlb too. */
257 tlbie(va, psize, ssize, local);
259 return ret;
262 static long native_hpte_find(unsigned long va, int psize, int ssize)
264 struct hash_pte *hptep;
265 unsigned long hash;
266 unsigned long i;
267 long slot;
268 unsigned long want_v, hpte_v;
270 hash = hpt_hash(va, mmu_psize_defs[psize].shift, ssize);
271 want_v = hpte_encode_v(va, psize, ssize);
273 /* Bolted mappings are only ever in the primary group */
274 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
275 for (i = 0; i < HPTES_PER_GROUP; i++) {
276 hptep = htab_address + slot;
277 hpte_v = hptep->v;
279 if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
280 /* HPTE matches */
281 return slot;
282 ++slot;
285 return -1;
289 * Update the page protection bits. Intended to be used to create
290 * guard pages for kernel data structures on pages which are bolted
291 * in the HPT. Assumes pages being operated on will not be stolen.
293 * No need to lock here because we should be the only user.
295 static void native_hpte_updateboltedpp(unsigned long newpp, unsigned long ea,
296 int psize, int ssize)
298 unsigned long vsid, va;
299 long slot;
300 struct hash_pte *hptep;
302 vsid = get_kernel_vsid(ea, ssize);
303 va = hpt_va(ea, vsid, ssize);
305 slot = native_hpte_find(va, psize, ssize);
306 if (slot == -1)
307 panic("could not find page to bolt\n");
308 hptep = htab_address + slot;
310 /* Update the HPTE */
311 hptep->r = (hptep->r & ~(HPTE_R_PP | HPTE_R_N)) |
312 (newpp & (HPTE_R_PP | HPTE_R_N));
314 /* Ensure it is out of the tlb too. */
315 tlbie(va, psize, ssize, 0);
318 static void native_hpte_invalidate(unsigned long slot, unsigned long va,
319 int psize, int ssize, int local)
321 struct hash_pte *hptep = htab_address + slot;
322 unsigned long hpte_v;
323 unsigned long want_v;
324 unsigned long flags;
326 local_irq_save(flags);
328 DBG_LOW(" invalidate(va=%016lx, hash: %x)\n", va, slot);
330 want_v = hpte_encode_v(va, psize, ssize);
331 native_lock_hpte(hptep);
332 hpte_v = hptep->v;
334 /* Even if we miss, we need to invalidate the TLB */
335 if (!HPTE_V_COMPARE(hpte_v, want_v) || !(hpte_v & HPTE_V_VALID))
336 native_unlock_hpte(hptep);
337 else
338 /* Invalidate the hpte. NOTE: this also unlocks it */
339 hptep->v = 0;
341 /* Invalidate the TLB */
342 tlbie(va, psize, ssize, local);
344 local_irq_restore(flags);
347 #define LP_SHIFT 12
348 #define LP_BITS 8
349 #define LP_MASK(i) ((0xFF >> (i)) << LP_SHIFT)
351 static void hpte_decode(struct hash_pte *hpte, unsigned long slot,
352 int *psize, int *ssize, unsigned long *va)
354 unsigned long hpte_r = hpte->r;
355 unsigned long hpte_v = hpte->v;
356 unsigned long avpn;
357 int i, size, shift, penc;
359 if (!(hpte_v & HPTE_V_LARGE))
360 size = MMU_PAGE_4K;
361 else {
362 for (i = 0; i < LP_BITS; i++) {
363 if ((hpte_r & LP_MASK(i+1)) == LP_MASK(i+1))
364 break;
366 penc = LP_MASK(i+1) >> LP_SHIFT;
367 for (size = 0; size < MMU_PAGE_COUNT; size++) {
369 /* 4K pages are not represented by LP */
370 if (size == MMU_PAGE_4K)
371 continue;
373 /* valid entries have a shift value */
374 if (!mmu_psize_defs[size].shift)
375 continue;
377 if (penc == mmu_psize_defs[size].penc)
378 break;
382 /* This works for all page sizes, and for 256M and 1T segments */
383 shift = mmu_psize_defs[size].shift;
384 avpn = (HPTE_V_AVPN_VAL(hpte_v) & ~mmu_psize_defs[size].avpnm) << 23;
386 if (shift < 23) {
387 unsigned long vpi, vsid, pteg;
389 pteg = slot / HPTES_PER_GROUP;
390 if (hpte_v & HPTE_V_SECONDARY)
391 pteg = ~pteg;
392 switch (hpte_v >> HPTE_V_SSIZE_SHIFT) {
393 case MMU_SEGSIZE_256M:
394 vpi = ((avpn >> 28) ^ pteg) & htab_hash_mask;
395 break;
396 case MMU_SEGSIZE_1T:
397 vsid = avpn >> 40;
398 vpi = (vsid ^ (vsid << 25) ^ pteg) & htab_hash_mask;
399 break;
400 default:
401 avpn = vpi = size = 0;
403 avpn |= (vpi << mmu_psize_defs[size].shift);
406 *va = avpn;
407 *psize = size;
408 *ssize = hpte_v >> HPTE_V_SSIZE_SHIFT;
412 * clear all mappings on kexec. All cpus are in real mode (or they will
413 * be when they isi), and we are the only one left. We rely on our kernel
414 * mapping being 0xC0's and the hardware ignoring those two real bits.
416 * TODO: add batching support when enabled. remember, no dynamic memory here,
417 * athough there is the control page available...
419 static void native_hpte_clear(void)
421 unsigned long slot, slots, flags;
422 struct hash_pte *hptep = htab_address;
423 unsigned long hpte_v, va;
424 unsigned long pteg_count;
425 int psize, ssize;
427 pteg_count = htab_hash_mask + 1;
429 local_irq_save(flags);
431 /* we take the tlbie lock and hold it. Some hardware will
432 * deadlock if we try to tlbie from two processors at once.
434 raw_spin_lock(&native_tlbie_lock);
436 slots = pteg_count * HPTES_PER_GROUP;
438 for (slot = 0; slot < slots; slot++, hptep++) {
440 * we could lock the pte here, but we are the only cpu
441 * running, right? and for crash dump, we probably
442 * don't want to wait for a maybe bad cpu.
444 hpte_v = hptep->v;
447 * Call __tlbie() here rather than tlbie() since we
448 * already hold the native_tlbie_lock.
450 if (hpte_v & HPTE_V_VALID) {
451 hpte_decode(hptep, slot, &psize, &ssize, &va);
452 hptep->v = 0;
453 __tlbie(va, psize, ssize);
457 asm volatile("eieio; tlbsync; ptesync":::"memory");
458 raw_spin_unlock(&native_tlbie_lock);
459 local_irq_restore(flags);
463 * Batched hash table flush, we batch the tlbie's to avoid taking/releasing
464 * the lock all the time
466 static void native_flush_hash_range(unsigned long number, int local)
468 unsigned long va, hash, index, hidx, shift, slot;
469 struct hash_pte *hptep;
470 unsigned long hpte_v;
471 unsigned long want_v;
472 unsigned long flags;
473 real_pte_t pte;
474 struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
475 unsigned long psize = batch->psize;
476 int ssize = batch->ssize;
477 int i;
479 local_irq_save(flags);
481 for (i = 0; i < number; i++) {
482 va = batch->vaddr[i];
483 pte = batch->pte[i];
485 pte_iterate_hashed_subpages(pte, psize, va, index, shift) {
486 hash = hpt_hash(va, shift, ssize);
487 hidx = __rpte_to_hidx(pte, index);
488 if (hidx & _PTEIDX_SECONDARY)
489 hash = ~hash;
490 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
491 slot += hidx & _PTEIDX_GROUP_IX;
492 hptep = htab_address + slot;
493 want_v = hpte_encode_v(va, psize, ssize);
494 native_lock_hpte(hptep);
495 hpte_v = hptep->v;
496 if (!HPTE_V_COMPARE(hpte_v, want_v) ||
497 !(hpte_v & HPTE_V_VALID))
498 native_unlock_hpte(hptep);
499 else
500 hptep->v = 0;
501 } pte_iterate_hashed_end();
504 if (mmu_has_feature(MMU_FTR_TLBIEL) &&
505 mmu_psize_defs[psize].tlbiel && local) {
506 asm volatile("ptesync":::"memory");
507 for (i = 0; i < number; i++) {
508 va = batch->vaddr[i];
509 pte = batch->pte[i];
511 pte_iterate_hashed_subpages(pte, psize, va, index,
512 shift) {
513 __tlbiel(va, psize, ssize);
514 } pte_iterate_hashed_end();
516 asm volatile("ptesync":::"memory");
517 } else {
518 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
520 if (lock_tlbie)
521 raw_spin_lock(&native_tlbie_lock);
523 asm volatile("ptesync":::"memory");
524 for (i = 0; i < number; i++) {
525 va = batch->vaddr[i];
526 pte = batch->pte[i];
528 pte_iterate_hashed_subpages(pte, psize, va, index,
529 shift) {
530 __tlbie(va, psize, ssize);
531 } pte_iterate_hashed_end();
533 asm volatile("eieio; tlbsync; ptesync":::"memory");
535 if (lock_tlbie)
536 raw_spin_unlock(&native_tlbie_lock);
539 local_irq_restore(flags);
542 #ifdef CONFIG_PPC_PSERIES
543 /* Disable TLB batching on nighthawk */
544 static inline int tlb_batching_enabled(void)
546 struct device_node *root = of_find_node_by_path("/");
547 int enabled = 1;
549 if (root) {
550 const char *model = of_get_property(root, "model", NULL);
551 if (model && !strcmp(model, "IBM,9076-N81"))
552 enabled = 0;
553 of_node_put(root);
556 return enabled;
558 #else
559 static inline int tlb_batching_enabled(void)
561 return 1;
563 #endif
565 void __init hpte_init_native(void)
567 ppc_md.hpte_invalidate = native_hpte_invalidate;
568 ppc_md.hpte_updatepp = native_hpte_updatepp;
569 ppc_md.hpte_updateboltedpp = native_hpte_updateboltedpp;
570 ppc_md.hpte_insert = native_hpte_insert;
571 ppc_md.hpte_remove = native_hpte_remove;
572 ppc_md.hpte_clear_all = native_hpte_clear;
573 if (tlb_batching_enabled())
574 ppc_md.flush_hash_range = native_flush_hash_range;