projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-rc-fixes-2.6
[linux-2.6-block.git] / arch / powerpc / mm / hash_utils_64.c
CommitLineData
1da177e4
LT		 1/*
2 * PowerPC64 port by Mike Corrigan and Dave Engebretsen
3 * {mikejc|engebret}@us.ibm.com
4 *
5 * Copyright (c) 2000 Mike Corrigan <mikejc@us.ibm.com>
6 *
7 * SMP scalability work:
8 * Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
9 *
10 * Module name: htab.c
11 *
12 * Description:
13 * PowerPC Hashed Page Table functions
14 *
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version
18 * 2 of the License, or (at your option) any later version.
19 */
20
21#undef DEBUG
3c726f8d 22#undef DEBUG_LOW
1da177e4 23
1da177e4
LT		 24#include <linux/spinlock.h>
25#include <linux/errno.h>
26#include <linux/sched.h>
27#include <linux/proc_fs.h>
28#include <linux/stat.h>
29#include <linux/sysctl.h>
30#include <linux/ctype.h>
31#include <linux/cache.h>
32#include <linux/init.h>
33#include <linux/signal.h>
34
1da177e4
LT		 35#include <asm/processor.h>
36#include <asm/pgtable.h>
37#include <asm/mmu.h>
38#include <asm/mmu_context.h>
39#include <asm/page.h>
40#include <asm/types.h>
41#include <asm/system.h>
42#include <asm/uaccess.h>
43#include <asm/machdep.h>
44#include <asm/lmb.h>
45#include <asm/abs_addr.h>
46#include <asm/tlbflush.h>
47#include <asm/io.h>
48#include <asm/eeh.h>
49#include <asm/tlb.h>
50#include <asm/cacheflush.h>
51#include <asm/cputable.h>
1da177e4 52#include <asm/sections.h>
d0f13e3c 53#include <asm/spu.h>
aa39be09 54#include <asm/udbg.h>
1da177e4
LT		 55
56#ifdef DEBUG
57#define DBG(fmt...) udbg_printf(fmt)
58#else
59#define DBG(fmt...)
60#endif
61
3c726f8d
BH		 62#ifdef DEBUG_LOW
63#define DBG_LOW(fmt...) udbg_printf(fmt)
64#else
65#define DBG_LOW(fmt...)
66#endif
67
68#define KB (1024)
69#define MB (1024*KB)
70
1da177e4
LT		 71/*
72 * Note: pte --> Linux PTE
73 * HPTE --> PowerPC Hashed Page Table Entry
74 *
75 * Execution context:
76 * htab_initialize is called with the MMU off (of course), but
77 * the kernel has been copied down to zero so it can directly
78 * reference global data. At this point it is very difficult
79 * to print debug info.
80 *
81 */
82
83#ifdef CONFIG_U3_DART
84extern unsigned long dart_tablebase;
85#endif /* CONFIG_U3_DART */
86
799d6046
PM		 87static unsigned long _SDR1;
88struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT];
89
8e561e7e 90struct hash_pte *htab_address;
337a7128 91unsigned long htab_size_bytes;
96e28449 92unsigned long htab_hash_mask;
3c726f8d
BH		 93int mmu_linear_psize = MMU_PAGE_4K;
94int mmu_virtual_psize = MMU_PAGE_4K;
bf72aeba
PM		 95int mmu_vmalloc_psize = MMU_PAGE_4K;
96int mmu_io_psize = MMU_PAGE_4K;
1189be65
PM		 97int mmu_kernel_ssize = MMU_SEGSIZE_256M;
98int mmu_highuser_ssize = MMU_SEGSIZE_256M;
584f8b71 99u16 mmu_slb_size = 64;
3c726f8d
BH		 100#ifdef CONFIG_HUGETLB_PAGE
101int mmu_huge_psize = MMU_PAGE_16M;
102unsigned int HPAGE_SHIFT;
103#endif
bf72aeba
PM		 104#ifdef CONFIG_PPC_64K_PAGES
105int mmu_ci_restrictions;
106#endif
370a908d
BH		 107#ifdef CONFIG_DEBUG_PAGEALLOC
108static u8 *linear_map_hash_slots;
109static unsigned long linear_map_hash_count;
ed166692 110static DEFINE_SPINLOCK(linear_map_hash_lock);
370a908d 111#endif /* CONFIG_DEBUG_PAGEALLOC */
1da177e4 112
3c726f8d
BH		 113/* There are definitions of page sizes arrays to be used when none
114 * is provided by the firmware.
115 */
1da177e4 116
3c726f8d
BH		 117/* Pre-POWER4 CPUs (4k pages only)
118 */
119struct mmu_psize_def mmu_psize_defaults_old[] = {
120 [MMU_PAGE_4K] = {
121 .shift = 12,
122 .sllp = 0,
123 .penc = 0,
124 .avpnm = 0,
125 .tlbiel = 0,
126 },
127};
128
129/* POWER4, GPUL, POWER5
130 *
131 * Support for 16Mb large pages
132 */
133struct mmu_psize_def mmu_psize_defaults_gp[] = {
134 [MMU_PAGE_4K] = {
135 .shift = 12,
136 .sllp = 0,
137 .penc = 0,
138 .avpnm = 0,
139 .tlbiel = 1,
140 },
141 [MMU_PAGE_16M] = {
142 .shift = 24,
143 .sllp = SLB_VSID_L,
144 .penc = 0,
145 .avpnm = 0x1UL,
146 .tlbiel = 0,
147 },
148};
149
150
151int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
1189be65
PM		 152 unsigned long pstart, unsigned long mode,
153 int psize, int ssize)
1da177e4 154{
3c726f8d
BH		 155 unsigned long vaddr, paddr;
156 unsigned int step, shift;
1da177e4 157 unsigned long tmp_mode;
3c726f8d 158 int ret = 0;
1da177e4 159
3c726f8d
BH		 160 shift = mmu_psize_defs[psize].shift;
161 step = 1 << shift;
1da177e4 162
3c726f8d
BH		 163 for (vaddr = vstart, paddr = pstart; vaddr < vend;
164 vaddr += step, paddr += step) {
370a908d 165 unsigned long hash, hpteg;
1189be65
PM		 166 unsigned long vsid = get_kernel_vsid(vaddr, ssize);
167 unsigned long va = hpt_va(vaddr, vsid, ssize);
1da177e4
LT		 168
169 tmp_mode = mode;
170
171 /* Make non-kernel text non-executable */
3c726f8d
BH		 172 if (!in_kernel_text(vaddr))
173 tmp_mode = mode | HPTE_R_N;
1da177e4 174
1189be65 175 hash = hpt_hash(va, shift, ssize);
1da177e4
LT		 176 hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
177
c30a4df3
ME		 178 DBG("htab_bolt_mapping: calling %p\n", ppc_md.hpte_insert);
179
180 BUG_ON(!ppc_md.hpte_insert);
181 ret = ppc_md.hpte_insert(hpteg, va, paddr,
1189be65 182 tmp_mode, HPTE_V_BOLTED, psize, ssize);
c30a4df3 183
3c726f8d
BH		 184 if (ret < 0)
185 break;
370a908d
BH		 186#ifdef CONFIG_DEBUG_PAGEALLOC
187 if ((paddr >> PAGE_SHIFT) < linear_map_hash_count)
188 linear_map_hash_slots[paddr >> PAGE_SHIFT] = ret | 0x80;
189#endif /* CONFIG_DEBUG_PAGEALLOC */
3c726f8d
BH		 190 }
191 return ret < 0 ? ret : 0;
192}
1da177e4 193
1189be65
PM		 194static int __init htab_dt_scan_seg_sizes(unsigned long node,
195 const char *uname, int depth,
196 void *data)
197{
198 char *type = of_get_flat_dt_prop(node, "device_type", NULL);
199 u32 *prop;
200 unsigned long size = 0;
201
202 /* We are scanning "cpu" nodes only */
203 if (type == NULL || strcmp(type, "cpu") != 0)
204 return 0;
205
206 prop = (u32 *)of_get_flat_dt_prop(node, "ibm,processor-segment-sizes",
207 &size);
208 if (prop == NULL)
209 return 0;
210 for (; size >= 4; size -= 4, ++prop) {
211 if (prop[0] == 40) {
212 DBG("1T segment support detected\n");
213 cur_cpu_spec->cpu_features |= CPU_FTR_1T_SEGMENT;
f5534004 214 return 1;
1189be65 215 }
1189be65 216 }
f66bce5e 217 cur_cpu_spec->cpu_features &= ~CPU_FTR_NO_SLBIE_B;
1189be65
PM		 218 return 0;
219}
220
221static void __init htab_init_seg_sizes(void)
222{
223 of_scan_flat_dt(htab_dt_scan_seg_sizes, NULL);
224}
225
3c726f8d
BH		 226static int __init htab_dt_scan_page_sizes(unsigned long node,
227 const char *uname, int depth,
228 void *data)
229{
230 char *type = of_get_flat_dt_prop(node, "device_type", NULL);
231 u32 *prop;
232 unsigned long size = 0;
233
234 /* We are scanning "cpu" nodes only */
235 if (type == NULL || strcmp(type, "cpu") != 0)
236 return 0;
237
238 prop = (u32 *)of_get_flat_dt_prop(node,
239 "ibm,segment-page-sizes", &size);
240 if (prop != NULL) {
241 DBG("Page sizes from device-tree:\n");
242 size /= 4;
243 cur_cpu_spec->cpu_features &= ~(CPU_FTR_16M_PAGE);
244 while(size > 0) {
245 unsigned int shift = prop[0];
246 unsigned int slbenc = prop[1];
247 unsigned int lpnum = prop[2];
248 unsigned int lpenc = 0;
249 struct mmu_psize_def *def;
250 int idx = -1;
251
252 size -= 3; prop += 3;
253 while(size > 0 && lpnum) {
254 if (prop[0] == shift)
255 lpenc = prop[1];
256 prop += 2; size -= 2;
257 lpnum--;
258 }
259 switch(shift) {
260 case 0xc:
261 idx = MMU_PAGE_4K;
262 break;
263 case 0x10:
264 idx = MMU_PAGE_64K;
265 break;
266 case 0x14:
267 idx = MMU_PAGE_1M;
268 break;
269 case 0x18:
270 idx = MMU_PAGE_16M;
271 cur_cpu_spec->cpu_features |= CPU_FTR_16M_PAGE;
272 break;
273 case 0x22:
274 idx = MMU_PAGE_16G;
275 break;
276 }
277 if (idx < 0)
278 continue;
279 def = &mmu_psize_defs[idx];
280 def->shift = shift;
281 if (shift <= 23)
282 def->avpnm = 0;
283 else
284 def->avpnm = (1 << (shift - 23)) - 1;
285 def->sllp = slbenc;
286 def->penc = lpenc;
287 /* We don't know for sure what's up with tlbiel, so
288 * for now we only set it for 4K and 64K pages
289 */
290 if (idx == MMU_PAGE_4K || idx == MMU_PAGE_64K)
291 def->tlbiel = 1;
292 else
293 def->tlbiel = 0;
294
295 DBG(" %d: shift=%02x, sllp=%04x, avpnm=%08x, "
296 "tlbiel=%d, penc=%d\n",
297 idx, shift, def->sllp, def->avpnm, def->tlbiel,
298 def->penc);
1da177e4 299 }
3c726f8d
BH		 300 return 1;
301 }
302 return 0;
303}
304
3c726f8d
BH		 305static void __init htab_init_page_sizes(void)
306{
307 int rc;
308
309 /* Default to 4K pages only */
310 memcpy(mmu_psize_defs, mmu_psize_defaults_old,
311 sizeof(mmu_psize_defaults_old));
312
313 /*
314 * Try to find the available page sizes in the device-tree
315 */
316 rc = of_scan_flat_dt(htab_dt_scan_page_sizes, NULL);
317 if (rc != 0) /* Found */
318 goto found;
319
320 /*
321 * Not in the device-tree, let's fallback on known size
322 * list for 16M capable GP & GR
323 */
0470466d 324 if (cpu_has_feature(CPU_FTR_16M_PAGE))
3c726f8d
BH		 325 memcpy(mmu_psize_defs, mmu_psize_defaults_gp,
326 sizeof(mmu_psize_defaults_gp));
327 found:
370a908d 328#ifndef CONFIG_DEBUG_PAGEALLOC
3c726f8d
BH		 329 /*
330 * Pick a size for the linear mapping. Currently, we only support
331 * 16M, 1M and 4K which is the default
332 */
333 if (mmu_psize_defs[MMU_PAGE_16M].shift)
334 mmu_linear_psize = MMU_PAGE_16M;
335 else if (mmu_psize_defs[MMU_PAGE_1M].shift)
336 mmu_linear_psize = MMU_PAGE_1M;
370a908d 337#endif /* CONFIG_DEBUG_PAGEALLOC */
3c726f8d 338
bf72aeba 339#ifdef CONFIG_PPC_64K_PAGES
3c726f8d
BH		 340 /*
341 * Pick a size for the ordinary pages. Default is 4K, we support
bf72aeba
PM		 342 * 64K for user mappings and vmalloc if supported by the processor.
343 * We only use 64k for ioremap if the processor
344 * (and firmware) support cache-inhibited large pages.
345 * If not, we use 4k and set mmu_ci_restrictions so that
346 * hash_page knows to switch processes that use cache-inhibited
347 * mappings to 4k pages.
3c726f8d 348 */
bf72aeba 349 if (mmu_psize_defs[MMU_PAGE_64K].shift) {
3c726f8d 350 mmu_virtual_psize = MMU_PAGE_64K;
bf72aeba 351 mmu_vmalloc_psize = MMU_PAGE_64K;
370a908d
BH		 352 if (mmu_linear_psize == MMU_PAGE_4K)
353 mmu_linear_psize = MMU_PAGE_64K;
bf72aeba
PM		 354 if (cpu_has_feature(CPU_FTR_CI_LARGE_PAGE))
355 mmu_io_psize = MMU_PAGE_64K;
356 else
357 mmu_ci_restrictions = 1;
358 }
370a908d 359#endif /* CONFIG_PPC_64K_PAGES */
3c726f8d 360
bf72aeba
PM		 361 printk(KERN_DEBUG "Page orders: linear mapping = %d, "
362 "virtual = %d, io = %d\n",
3c726f8d 363 mmu_psize_defs[mmu_linear_psize].shift,
bf72aeba
PM		 364 mmu_psize_defs[mmu_virtual_psize].shift,
365 mmu_psize_defs[mmu_io_psize].shift);
3c726f8d
BH		 366
367#ifdef CONFIG_HUGETLB_PAGE
368 /* Init large page size. Currently, we pick 16M or 1M depending
369 * on what is available
370 */
371 if (mmu_psize_defs[MMU_PAGE_16M].shift)
4ec161cf 372 set_huge_psize(MMU_PAGE_16M);
7d24f0b8
DG		 373 /* With 4k/4level pagetables, we can't (for now) cope with a
374 * huge page size < PMD_SIZE */
3c726f8d 375 else if (mmu_psize_defs[MMU_PAGE_1M].shift)
4ec161cf 376 set_huge_psize(MMU_PAGE_1M);
3c726f8d
BH		 377#endif /* CONFIG_HUGETLB_PAGE */
378}
379
380static int __init htab_dt_scan_pftsize(unsigned long node,
381 const char *uname, int depth,
382 void *data)
383{
384 char *type = of_get_flat_dt_prop(node, "device_type", NULL);
385 u32 *prop;
386
387 /* We are scanning "cpu" nodes only */
388 if (type == NULL || strcmp(type, "cpu") != 0)
389 return 0;
390
391 prop = (u32 *)of_get_flat_dt_prop(node, "ibm,pft-size", NULL);
392 if (prop != NULL) {
393 /* pft_size[0] is the NUMA CEC cookie */
394 ppc64_pft_size = prop[1];
395 return 1;
1da177e4 396 }
3c726f8d 397 return 0;
1da177e4
LT		 398}
399
3c726f8d 400static unsigned long __init htab_get_table_size(void)
3eac8c69 401{
799d6046 402 unsigned long mem_size, rnd_mem_size, pteg_count;
3eac8c69 403
3c726f8d 404 /* If hash size isn't already provided by the platform, we try to
943ffb58 405 * retrieve it from the device-tree. If it's not there neither, we
3c726f8d 406 * calculate it now based on the total RAM size
3eac8c69 407 */
3c726f8d
BH		 408 if (ppc64_pft_size == 0)
409 of_scan_flat_dt(htab_dt_scan_pftsize, NULL);
3eac8c69
PM		 410 if (ppc64_pft_size)
411 return 1UL << ppc64_pft_size;
412
413 /* round mem_size up to next power of 2 */
799d6046
PM		 414 mem_size = lmb_phys_mem_size();
415 rnd_mem_size = 1UL << __ilog2(mem_size);
416 if (rnd_mem_size < mem_size)
3eac8c69
PM		 417 rnd_mem_size <<= 1;
418
419 /* # pages / 2 */
420 pteg_count = max(rnd_mem_size >> (12 + 1), 1UL << 11);
421
422 return pteg_count << 7;
423}
424
54b79248
MK		 425#ifdef CONFIG_MEMORY_HOTPLUG
426void create_section_mapping(unsigned long start, unsigned long end)
427{
caf80e57 428 BUG_ON(htab_bolt_mapping(start, end, __pa(start),
54b79248 429 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_COHERENT | PP_RWXX,
1189be65 430 mmu_linear_psize, mmu_kernel_ssize));
54b79248
MK		 431}
432#endif /* CONFIG_MEMORY_HOTPLUG */
433
7d0daae4
ME		 434static inline void make_bl(unsigned int *insn_addr, void *func)
435{
436 unsigned long funcp = *((unsigned long *)func);
437 int offset = funcp - (unsigned long)insn_addr;
438
439 *insn_addr = (unsigned int)(0x48000001 | (offset & 0x03fffffc));
440 flush_icache_range((unsigned long)insn_addr, 4+
441 (unsigned long)insn_addr);
442}
443
444static void __init htab_finish_init(void)
445{
446 extern unsigned int *htab_call_hpte_insert1;
447 extern unsigned int *htab_call_hpte_insert2;
448 extern unsigned int *htab_call_hpte_remove;
449 extern unsigned int *htab_call_hpte_updatepp;
450
16c2d476 451#ifdef CONFIG_PPC_HAS_HASH_64K
7d0daae4
ME		 452 extern unsigned int *ht64_call_hpte_insert1;
453 extern unsigned int *ht64_call_hpte_insert2;
454 extern unsigned int *ht64_call_hpte_remove;
455 extern unsigned int *ht64_call_hpte_updatepp;
456
457 make_bl(ht64_call_hpte_insert1, ppc_md.hpte_insert);
458 make_bl(ht64_call_hpte_insert2, ppc_md.hpte_insert);
459 make_bl(ht64_call_hpte_remove, ppc_md.hpte_remove);
460 make_bl(ht64_call_hpte_updatepp, ppc_md.hpte_updatepp);
5b825831 461#endif /* CONFIG_PPC_HAS_HASH_64K */
7d0daae4
ME		 462
463 make_bl(htab_call_hpte_insert1, ppc_md.hpte_insert);
464 make_bl(htab_call_hpte_insert2, ppc_md.hpte_insert);
465 make_bl(htab_call_hpte_remove, ppc_md.hpte_remove);
466 make_bl(htab_call_hpte_updatepp, ppc_md.hpte_updatepp);
467}
468
1da177e4
LT		 469void __init htab_initialize(void)
470{
337a7128 471 unsigned long table;
1da177e4
LT		 472 unsigned long pteg_count;
473 unsigned long mode_rw;
41d824bf 474 unsigned long base = 0, size = 0, limit;
3c726f8d
BH		 475 int i;
476
1da177e4
LT		 477 extern unsigned long tce_alloc_start, tce_alloc_end;
478
479 DBG(" -> htab_initialize()\n");
480
1189be65
PM		 481 /* Initialize segment sizes */
482 htab_init_seg_sizes();
483
3c726f8d
BH		 484 /* Initialize page sizes */
485 htab_init_page_sizes();
486
1189be65
PM		 487 if (cpu_has_feature(CPU_FTR_1T_SEGMENT)) {
488 mmu_kernel_ssize = MMU_SEGSIZE_1T;
489 mmu_highuser_ssize = MMU_SEGSIZE_1T;
490 printk(KERN_INFO "Using 1TB segments\n");
491 }
492
1da177e4
LT		 493 /*
494 * Calculate the required size of the htab. We want the number of
495 * PTEGs to equal one half the number of real pages.
496 */
3c726f8d 497 htab_size_bytes = htab_get_table_size();
1da177e4
LT		 498 pteg_count = htab_size_bytes >> 7;
499
1da177e4
LT		 500 htab_hash_mask = pteg_count - 1;
501
57cfb814 502 if (firmware_has_feature(FW_FEATURE_LPAR)) {
1da177e4
LT		 503 /* Using a hypervisor which owns the htab */
504 htab_address = NULL;
505 _SDR1 = 0;
506 } else {
507 /* Find storage for the HPT. Must be contiguous in
41d824bf
ME		 508 * the absolute address space. On cell we want it to be
509 * in the first 1 Gig.
1da177e4 510 */
41d824bf
ME		 511 if (machine_is(cell))
512 limit = 0x40000000;
513 else
514 limit = 0;
515
516 table = lmb_alloc_base(htab_size_bytes, htab_size_bytes, limit);
1da177e4
LT		 517
518 DBG("Hash table allocated at %lx, size: %lx\n", table,
519 htab_size_bytes);
520
1da177e4
LT		 521 htab_address = abs_to_virt(table);
522
523 /* htab absolute addr + encoded htabsize */
524 _SDR1 = table + __ilog2(pteg_count) - 11;
525
526 /* Initialize the HPT with no entries */
527 memset((void *)table, 0, htab_size_bytes);
799d6046
PM		 528
529 /* Set SDR1 */
530 mtspr(SPRN_SDR1, _SDR1);
1da177e4
LT		 531 }
532
515bae9c 533 mode_rw = _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_COHERENT | PP_RWXX;
1da177e4 534
370a908d
BH		 535#ifdef CONFIG_DEBUG_PAGEALLOC
536 linear_map_hash_count = lmb_end_of_DRAM() >> PAGE_SHIFT;
537 linear_map_hash_slots = __va(lmb_alloc_base(linear_map_hash_count,
538 1, lmb.rmo_size));
539 memset(linear_map_hash_slots, 0, linear_map_hash_count);
540#endif /* CONFIG_DEBUG_PAGEALLOC */
541
1da177e4
LT		 542 /* On U3 based machines, we need to reserve the DART area and
543 * _NOT_ map it to avoid cache paradoxes as it's remapped non
544 * cacheable later on
545 */
1da177e4
LT		 546
547 /* create bolted the linear mapping in the hash table */
548 for (i=0; i < lmb.memory.cnt; i++) {
b5666f70 549 base = (unsigned long)__va(lmb.memory.region[i].base);
1da177e4
LT		 550 size = lmb.memory.region[i].size;
551
552 DBG("creating mapping for region: %lx : %lx\n", base, size);
553
554#ifdef CONFIG_U3_DART
555 /* Do not map the DART space. Fortunately, it will be aligned
3c726f8d
BH		 556 * in such a way that it will not cross two lmb regions and
557 * will fit within a single 16Mb page.
558 * The DART space is assumed to be a full 16Mb region even if
559 * we only use 2Mb of that space. We will use more of it later
560 * for AGP GART. We have to use a full 16Mb large page.
1da177e4
LT		 561 */
562 DBG("DART base: %lx\n", dart_tablebase);
563
564 if (dart_tablebase != 0 && dart_tablebase >= base
565 && dart_tablebase < (base + size)) {
caf80e57 566 unsigned long dart_table_end = dart_tablebase + 16 * MB;
1da177e4 567 if (base != dart_tablebase)
3c726f8d 568 BUG_ON(htab_bolt_mapping(base, dart_tablebase,
caf80e57 569 __pa(base), mode_rw,
1189be65
PM		 570 mmu_linear_psize,
571 mmu_kernel_ssize));
caf80e57 572 if ((base + size) > dart_table_end)
3c726f8d 573 BUG_ON(htab_bolt_mapping(dart_tablebase+16*MB,
caf80e57
ME		 574 base + size,
575 __pa(dart_table_end),
3c726f8d 576 mode_rw,
1189be65
PM		 577 mmu_linear_psize,
578 mmu_kernel_ssize));
1da177e4
LT		 579 continue;
580 }
581#endif /* CONFIG_U3_DART */
caf80e57 582 BUG_ON(htab_bolt_mapping(base, base + size, __pa(base),
1189be65 583 mode_rw, mmu_linear_psize, mmu_kernel_ssize));
3c726f8d 584 }
1da177e4
LT		 585
586 /*
587 * If we have a memory_limit and we've allocated TCEs then we need to
588 * explicitly map the TCE area at the top of RAM. We also cope with the
589 * case that the TCEs start below memory_limit.
590 * tce_alloc_start/end are 16MB aligned so the mapping should work
591 * for either 4K or 16MB pages.
592 */
593 if (tce_alloc_start) {
b5666f70
ME		 594 tce_alloc_start = (unsigned long)__va(tce_alloc_start);
595 tce_alloc_end = (unsigned long)__va(tce_alloc_end);
1da177e4
LT		 596
597 if (base + size >= tce_alloc_start)
598 tce_alloc_start = base + size + 1;
599
caf80e57
ME		 600 BUG_ON(htab_bolt_mapping(tce_alloc_start, tce_alloc_end,
601 __pa(tce_alloc_start), mode_rw,
1189be65 602 mmu_linear_psize, mmu_kernel_ssize));
1da177e4
LT		 603 }
604
7d0daae4
ME		 605 htab_finish_init();
606
1da177e4
LT		 607 DBG(" <- htab_initialize()\n");
608}
609#undef KB
610#undef MB
1da177e4 611
e597cb32 612void htab_initialize_secondary(void)
799d6046 613{
57cfb814 614 if (!firmware_has_feature(FW_FEATURE_LPAR))
799d6046
PM		 615 mtspr(SPRN_SDR1, _SDR1);
616}
617
1da177e4
LT		 618/*
619 * Called by asm hashtable.S for doing lazy icache flush
620 */
621unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap)
622{
623 struct page *page;
624
76c8e25b
BH		 625 if (!pfn_valid(pte_pfn(pte)))
626 return pp;
627
1da177e4
LT		 628 page = pte_page(pte);
629
630 /* page is dirty */
631 if (!test_bit(PG_arch_1, &page->flags) && !PageReserved(page)) {
632 if (trap == 0x400) {
633 __flush_dcache_icache(page_address(page));
634 set_bit(PG_arch_1, &page->flags);
635 } else
3c726f8d 636 pp |= HPTE_R_N;
1da177e4
LT		 637 }
638 return pp;
639}
640
721151d0
PM		 641/*
642 * Demote a segment to using 4k pages.
643 * For now this makes the whole process use 4k pages.
644 */
721151d0 645#ifdef CONFIG_PPC_64K_PAGES
fa28237c 646void demote_segment_4k(struct mm_struct *mm, unsigned long addr)
16f1c746 647{
721151d0
PM		 648 if (mm->context.user_psize == MMU_PAGE_4K)
649 return;
d0f13e3c 650 slice_set_user_psize(mm, MMU_PAGE_4K);
1e57ba8d 651#ifdef CONFIG_SPU_BASE
721151d0
PM		 652 spu_flush_all_slbs(mm);
653#endif
fa28237c
PM		 654 if (get_paca()->context.user_psize != MMU_PAGE_4K) {
655 get_paca()->context = mm->context;
656 slb_flush_and_rebolt();
657 }
721151d0 658}
16f1c746 659#endif /* CONFIG_PPC_64K_PAGES */
721151d0 660
fa28237c
PM		 661#ifdef CONFIG_PPC_SUBPAGE_PROT
662/*
663 * This looks up a 2-bit protection code for a 4k subpage of a 64k page.
664 * Userspace sets the subpage permissions using the subpage_prot system call.
665 *
666 * Result is 0: full permissions, _PAGE_RW: read-only,
667 * _PAGE_USER or _PAGE_USER|_PAGE_RW: no access.
668 */
669static int subpage_protection(pgd_t *pgdir, unsigned long ea)
670{
671 struct subpage_prot_table *spt = pgd_subpage_prot(pgdir);
672 u32 spp = 0;
673 u32 **sbpm, *sbpp;
674
675 if (ea >= spt->maxaddr)
676 return 0;
677 if (ea < 0x100000000) {
678 /* addresses below 4GB use spt->low_prot */
679 sbpm = spt->low_prot;
680 } else {
681 sbpm = spt->protptrs[ea >> SBP_L3_SHIFT];
682 if (!sbpm)
683 return 0;
684 }
685 sbpp = sbpm[(ea >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1)];
686 if (!sbpp)
687 return 0;
688 spp = sbpp[(ea >> PAGE_SHIFT) & (SBP_L1_COUNT - 1)];
689
690 /* extract 2-bit bitfield for this 4k subpage */
691 spp >>= 30 - 2 * ((ea >> 12) & 0xf);
692
693 /* turn 0,1,2,3 into combination of _PAGE_USER and _PAGE_RW */
694 spp = ((spp & 2) ? _PAGE_USER : 0) | ((spp & 1) ? _PAGE_RW : 0);
695 return spp;
696}
697
698#else /* CONFIG_PPC_SUBPAGE_PROT */
699static inline int subpage_protection(pgd_t *pgdir, unsigned long ea)
700{
701 return 0;
702}
703#endif
704
1da177e4
LT		 705/* Result code is:
706 * 0 - handled
707 * 1 - normal page fault
708 * -1 - critical hash insertion error
fa28237c 709 * -2 - access not permitted by subpage protection mechanism
1da177e4
LT		 710 */
711int hash_page(unsigned long ea, unsigned long access, unsigned long trap)
712{
713 void *pgdir;
714 unsigned long vsid;
715 struct mm_struct *mm;
716 pte_t *ptep;
1da177e4 717 cpumask_t tmp;
3c726f8d 718 int rc, user_region = 0, local = 0;
1189be65 719 int psize, ssize;
1da177e4 720
3c726f8d
BH		 721 DBG_LOW("hash_page(ea=%016lx, access=%lx, trap=%lx\n",
722 ea, access, trap);
1f8d419e 723
3c726f8d
BH		 724 if ((ea & ~REGION_MASK) >= PGTABLE_RANGE) {
725 DBG_LOW(" out of pgtable range !\n");
726 return 1;
727 }
728
729 /* Get region & vsid */
1da177e4
LT		 730 switch (REGION_ID(ea)) {
731 case USER_REGION_ID:
732 user_region = 1;
733 mm = current->mm;
3c726f8d
BH		 734 if (! mm) {
735 DBG_LOW(" user region with no mm !\n");
1da177e4 736 return 1;
3c726f8d 737 }
16c2d476
BH		 738#ifdef CONFIG_PPC_MM_SLICES
739 psize = get_slice_psize(mm, ea);
740#else
bf72aeba 741 psize = mm->context.user_psize;
16c2d476 742#endif
1189be65
PM		 743 ssize = user_segment_size(ea);
744 vsid = get_vsid(mm->context.id, ea, ssize);
1da177e4 745 break;
1da177e4 746 case VMALLOC_REGION_ID:
1da177e4 747 mm = &init_mm;
1189be65 748 vsid = get_kernel_vsid(ea, mmu_kernel_ssize);
bf72aeba
PM		 749 if (ea < VMALLOC_END)
750 psize = mmu_vmalloc_psize;
751 else
752 psize = mmu_io_psize;
1189be65 753 ssize = mmu_kernel_ssize;
1da177e4 754 break;
1da177e4
LT		 755 default:
756 /* Not a valid range
757 * Send the problem up to do_page_fault
758 */
759 return 1;
1da177e4 760 }
3c726f8d 761 DBG_LOW(" mm=%p, mm->pgdir=%p, vsid=%016lx\n", mm, mm->pgd, vsid);
1da177e4 762
3c726f8d 763 /* Get pgdir */
1da177e4 764 pgdir = mm->pgd;
1da177e4
LT		 765 if (pgdir == NULL)
766 return 1;
767
3c726f8d 768 /* Check CPU locality */
1da177e4
LT		 769 tmp = cpumask_of_cpu(smp_processor_id());
770 if (user_region && cpus_equal(mm->cpu_vm_mask, tmp))
771 local = 1;
772
d0f13e3c 773#ifdef CONFIG_HUGETLB_PAGE
3c726f8d 774 /* Handle hugepage regions */
16c2d476 775 if (HPAGE_SHIFT && psize == mmu_huge_psize) {
3c726f8d 776 DBG_LOW(" -> huge page !\n");
cbf52afd 777 return hash_huge_page(mm, access, ea, vsid, local, trap);
3c726f8d 778 }
d0f13e3c 779#endif /* CONFIG_HUGETLB_PAGE */
3c726f8d 780
16c2d476
BH		 781#ifndef CONFIG_PPC_64K_PAGES
782 /* If we use 4K pages and our psize is not 4K, then we are hitting
783 * a special driver mapping, we need to align the address before
784 * we fetch the PTE
785 */
786 if (psize != MMU_PAGE_4K)
787 ea &= ~((1ul << mmu_psize_defs[psize].shift) - 1);
788#endif /* CONFIG_PPC_64K_PAGES */
789
3c726f8d
BH		 790 /* Get PTE and page size from page tables */
791 ptep = find_linux_pte(pgdir, ea);
792 if (ptep == NULL || !pte_present(*ptep)) {
793 DBG_LOW(" no PTE !\n");
794 return 1;
795 }
796
797#ifndef CONFIG_PPC_64K_PAGES
798 DBG_LOW(" i-pte: %016lx\n", pte_val(*ptep));
799#else
800 DBG_LOW(" i-pte: %016lx %016lx\n", pte_val(*ptep),
801 pte_val(*(ptep + PTRS_PER_PTE)));
802#endif
803 /* Pre-check access permissions (will be re-checked atomically
804 * in __hash_page_XX but this pre-check is a fast path
805 */
806 if (access & ~pte_val(*ptep)) {
807 DBG_LOW(" no access !\n");
808 return 1;
1da177e4
LT		 809 }
810
3c726f8d 811 /* Do actual hashing */
16c2d476 812#ifdef CONFIG_PPC_64K_PAGES
721151d0
PM		 813 /* If _PAGE_4K_PFN is set, make sure this is a 4k segment */
814 if (pte_val(*ptep) & _PAGE_4K_PFN) {
815 demote_segment_4k(mm, ea);
816 psize = MMU_PAGE_4K;
817 }
818
16f1c746
BH		 819 /* If this PTE is non-cacheable and we have restrictions on
820 * using non cacheable large pages, then we switch to 4k
821 */
822 if (mmu_ci_restrictions && psize == MMU_PAGE_64K &&
823 (pte_val(*ptep) & _PAGE_NO_CACHE)) {
824 if (user_region) {
825 demote_segment_4k(mm, ea);
826 psize = MMU_PAGE_4K;
827 } else if (ea < VMALLOC_END) {
828 /*
829 * some driver did a non-cacheable mapping
830 * in vmalloc space, so switch vmalloc
831 * to 4k pages
832 */
833 printk(KERN_ALERT "Reducing vmalloc segment "
834 "to 4kB pages because of "
835 "non-cacheable mapping\n");
836 psize = mmu_vmalloc_psize = MMU_PAGE_4K;
1e57ba8d 837#ifdef CONFIG_SPU_BASE
94b2a439
BH		 838 spu_flush_all_slbs(mm);
839#endif
bf72aeba 840 }
16f1c746
BH		 841 }
842 if (user_region) {
843 if (psize != get_paca()->context.user_psize) {
f6ab0b92 844 get_paca()->context = mm->context;
bf72aeba
PM		 845 slb_flush_and_rebolt();
846 }
16f1c746
BH		 847 } else if (get_paca()->vmalloc_sllp !=
848 mmu_psize_defs[mmu_vmalloc_psize].sllp) {
849 get_paca()->vmalloc_sllp =
850 mmu_psize_defs[mmu_vmalloc_psize].sllp;
67439b76 851 slb_vmalloc_update();
bf72aeba 852 }
16c2d476 853#endif /* CONFIG_PPC_64K_PAGES */
16f1c746 854
16c2d476 855#ifdef CONFIG_PPC_HAS_HASH_64K
bf72aeba 856 if (psize == MMU_PAGE_64K)
1189be65 857 rc = __hash_page_64K(ea, access, vsid, ptep, trap, local, ssize);
3c726f8d 858 else
16c2d476 859#endif /* CONFIG_PPC_HAS_HASH_64K */
fa28237c
PM		 860 {
861 int spp = subpage_protection(pgdir, ea);
862 if (access & spp)
863 rc = -2;
864 else
865 rc = __hash_page_4K(ea, access, vsid, ptep, trap,
866 local, ssize, spp);
867 }
3c726f8d
BH		 868
869#ifndef CONFIG_PPC_64K_PAGES
870 DBG_LOW(" o-pte: %016lx\n", pte_val(*ptep));
871#else
872 DBG_LOW(" o-pte: %016lx %016lx\n", pte_val(*ptep),
873 pte_val(*(ptep + PTRS_PER_PTE)));
874#endif
875 DBG_LOW(" -> rc=%d\n", rc);
876 return rc;
1da177e4 877}
67207b96 878EXPORT_SYMBOL_GPL(hash_page);
1da177e4 879
3c726f8d
BH		 880void hash_preload(struct mm_struct *mm, unsigned long ea,
881 unsigned long access, unsigned long trap)
1da177e4 882{
3c726f8d
BH		 883 unsigned long vsid;
884 void *pgdir;
885 pte_t *ptep;
886 cpumask_t mask;
887 unsigned long flags;
888 int local = 0;
1189be65 889 int ssize;
3c726f8d 890
d0f13e3c
BH		 891 BUG_ON(REGION_ID(ea) != USER_REGION_ID);
892
893#ifdef CONFIG_PPC_MM_SLICES
894 /* We only prefault standard pages for now */
2b02d139 895 if (unlikely(get_slice_psize(mm, ea) != mm->context.user_psize))
3c726f8d 896 return;
d0f13e3c 897#endif
3c726f8d
BH		 898
899 DBG_LOW("hash_preload(mm=%p, mm->pgdir=%p, ea=%016lx, access=%lx,"
900 " trap=%lx\n", mm, mm->pgd, ea, access, trap);
1da177e4 901
16f1c746 902 /* Get Linux PTE if available */
3c726f8d
BH		 903 pgdir = mm->pgd;
904 if (pgdir == NULL)
905 return;
906 ptep = find_linux_pte(pgdir, ea);
907 if (!ptep)
908 return;
16f1c746
BH		 909
910#ifdef CONFIG_PPC_64K_PAGES
911 /* If either _PAGE_4K_PFN or _PAGE_NO_CACHE is set (and we are on
912 * a 64K kernel), then we don't preload, hash_page() will take
913 * care of it once we actually try to access the page.
914 * That way we don't have to duplicate all of the logic for segment
915 * page size demotion here
916 */
917 if (pte_val(*ptep) & (_PAGE_4K_PFN | _PAGE_NO_CACHE))
918 return;
919#endif /* CONFIG_PPC_64K_PAGES */
920
921 /* Get VSID */
1189be65
PM		 922 ssize = user_segment_size(ea);
923 vsid = get_vsid(mm->context.id, ea, ssize);
3c726f8d 924
16c2d476 925 /* Hash doesn't like irqs */
3c726f8d 926 local_irq_save(flags);
16c2d476
BH		 927
928 /* Is that local to this CPU ? */
3c726f8d
BH		 929 mask = cpumask_of_cpu(smp_processor_id());
930 if (cpus_equal(mm->cpu_vm_mask, mask))
931 local = 1;
16c2d476
BH		 932
933 /* Hash it in */
934#ifdef CONFIG_PPC_HAS_HASH_64K
bf72aeba 935 if (mm->context.user_psize == MMU_PAGE_64K)
1189be65 936 __hash_page_64K(ea, access, vsid, ptep, trap, local, ssize);
1da177e4 937 else
5b825831 938#endif /* CONFIG_PPC_HAS_HASH_64K */
fa28237c
PM		 939 __hash_page_4K(ea, access, vsid, ptep, trap, local, ssize,
940 subpage_protection(pgdir, ea));
16c2d476 941
3c726f8d
BH		 942 local_irq_restore(flags);
943}
944
f6ab0b92
BH		 945/* WARNING: This is called from hash_low_64.S, if you change this prototype,
946 * do not forget to update the assembly call site !
947 */
1189be65
PM		 948void flush_hash_page(unsigned long va, real_pte_t pte, int psize, int ssize,
949 int local)
3c726f8d
BH		 950{
951 unsigned long hash, index, shift, hidx, slot;
952
953 DBG_LOW("flush_hash_page(va=%016x)\n", va);
954 pte_iterate_hashed_subpages(pte, psize, va, index, shift) {
1189be65 955 hash = hpt_hash(va, shift, ssize);
3c726f8d
BH		 956 hidx = __rpte_to_hidx(pte, index);
957 if (hidx & _PTEIDX_SECONDARY)
958 hash = ~hash;
959 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
960 slot += hidx & _PTEIDX_GROUP_IX;
961 DBG_LOW(" sub %d: hash=%x, hidx=%x\n", index, slot, hidx);
1189be65 962 ppc_md.hpte_invalidate(slot, va, psize, ssize, local);
3c726f8d 963 } pte_iterate_hashed_end();
1da177e4
LT		 964}
965
61b1a942 966void flush_hash_range(unsigned long number, int local)
1da177e4 967{
3c726f8d 968 if (ppc_md.flush_hash_range)
61b1a942 969 ppc_md.flush_hash_range(number, local);
3c726f8d 970 else {
1da177e4 971 int i;
61b1a942
BH		 972 struct ppc64_tlb_batch *batch =
973 &__get_cpu_var(ppc64_tlb_batch);
1da177e4
LT		 974
975 for (i = 0; i < number; i++)
3c726f8d 976 flush_hash_page(batch->vaddr[i], batch->pte[i],
1189be65 977 batch->psize, batch->ssize, local);
1da177e4
LT		 978 }
979}
980
1da177e4
LT		 981/*
982 * low_hash_fault is called when we the low level hash code failed
983 * to instert a PTE due to an hypervisor error
984 */
fa28237c 985void low_hash_fault(struct pt_regs *regs, unsigned long address, int rc)
1da177e4
LT		 986{
987 if (user_mode(regs)) {
fa28237c
PM		 988#ifdef CONFIG_PPC_SUBPAGE_PROT
989 if (rc == -2)
990 _exception(SIGSEGV, regs, SEGV_ACCERR, address);
991 else
992#endif
993 _exception(SIGBUS, regs, BUS_ADRERR, address);
994 } else
995 bad_page_fault(regs, address, SIGBUS);
1da177e4 996}
370a908d
BH		 997
998#ifdef CONFIG_DEBUG_PAGEALLOC
999static void kernel_map_linear_page(unsigned long vaddr, unsigned long lmi)
1000{
1189be65
PM		 1001 unsigned long hash, hpteg;
1002 unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize);
1003 unsigned long va = hpt_va(vaddr, vsid, mmu_kernel_ssize);
370a908d
BH		 1004 unsigned long mode = _PAGE_ACCESSED | _PAGE_DIRTY |
1005 _PAGE_COHERENT | PP_RWXX | HPTE_R_N;
1006 int ret;
1007
1189be65 1008 hash = hpt_hash(va, PAGE_SHIFT, mmu_kernel_ssize);
370a908d
BH		 1009 hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
1010
1011 ret = ppc_md.hpte_insert(hpteg, va, __pa(vaddr),
1189be65
PM		 1012 mode, HPTE_V_BOLTED,
1013 mmu_linear_psize, mmu_kernel_ssize);
370a908d
BH		 1014 BUG_ON (ret < 0);
1015 spin_lock(&linear_map_hash_lock);
1016 BUG_ON(linear_map_hash_slots[lmi] & 0x80);
1017 linear_map_hash_slots[lmi] = ret | 0x80;
1018 spin_unlock(&linear_map_hash_lock);
1019}
1020
1021static void kernel_unmap_linear_page(unsigned long vaddr, unsigned long lmi)
1022{
1189be65
PM		 1023 unsigned long hash, hidx, slot;
1024 unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize);
1025 unsigned long va = hpt_va(vaddr, vsid, mmu_kernel_ssize);
370a908d 1026
1189be65 1027 hash = hpt_hash(va, PAGE_SHIFT, mmu_kernel_ssize);
370a908d
BH		 1028 spin_lock(&linear_map_hash_lock);
1029 BUG_ON(!(linear_map_hash_slots[lmi] & 0x80));
1030 hidx = linear_map_hash_slots[lmi] & 0x7f;
1031 linear_map_hash_slots[lmi] = 0;
1032 spin_unlock(&linear_map_hash_lock);
1033 if (hidx & _PTEIDX_SECONDARY)
1034 hash = ~hash;
1035 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1036 slot += hidx & _PTEIDX_GROUP_IX;
1189be65 1037 ppc_md.hpte_invalidate(slot, va, mmu_linear_psize, mmu_kernel_ssize, 0);
370a908d
BH		 1038}
1039
1040void kernel_map_pages(struct page *page, int numpages, int enable)
1041{
1042 unsigned long flags, vaddr, lmi;
1043 int i;
1044
1045 local_irq_save(flags);
1046 for (i = 0; i < numpages; i++, page++) {
1047 vaddr = (unsigned long)page_address(page);
1048 lmi = __pa(vaddr) >> PAGE_SHIFT;
1049 if (lmi >= linear_map_hash_count)
1050 continue;
1051 if (enable)
1052 kernel_map_linear_page(vaddr, lmi);
1053 else
1054 kernel_unmap_linear_page(vaddr, lmi);
1055 }
1056 local_irq_restore(flags);
1057}
1058#endif /* CONFIG_DEBUG_PAGEALLOC */
Linux 6.x block layer and io_uring tree(s)