blob: 89b35c1813141c41172295325d8684c4d88c568b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +110022#undef DEBUG_LOW
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
24#include <linux/config.h>
25#include <linux/spinlock.h>
26#include <linux/errno.h>
27#include <linux/sched.h>
28#include <linux/proc_fs.h>
29#include <linux/stat.h>
30#include <linux/sysctl.h>
31#include <linux/ctype.h>
32#include <linux/cache.h>
33#include <linux/init.h>
34#include <linux/signal.h>
35
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <asm/processor.h>
37#include <asm/pgtable.h>
38#include <asm/mmu.h>
39#include <asm/mmu_context.h>
40#include <asm/page.h>
41#include <asm/types.h>
42#include <asm/system.h>
43#include <asm/uaccess.h>
44#include <asm/machdep.h>
45#include <asm/lmb.h>
46#include <asm/abs_addr.h>
47#include <asm/tlbflush.h>
48#include <asm/io.h>
49#include <asm/eeh.h>
50#include <asm/tlb.h>
51#include <asm/cacheflush.h>
52#include <asm/cputable.h>
53#include <asm/abs_addr.h>
54#include <asm/sections.h>
55
56#ifdef DEBUG
57#define DBG(fmt...) udbg_printf(fmt)
58#else
59#define DBG(fmt...)
60#endif
61
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +110062#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
Linus Torvalds1da177e2005-04-16 15:20:36 -070071/*
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
Paul Mackerras799d6042005-11-10 13:37:51 +110087static unsigned long _SDR1;
88struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT];
89
David Gibson96e28442005-07-13 01:11:42 -070090hpte_t *htab_address;
Michael Ellerman337a7122006-02-21 17:22:55 +110091unsigned long htab_size_bytes;
David Gibson96e28442005-07-13 01:11:42 -070092unsigned long htab_hash_mask;
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +110093int mmu_linear_psize = MMU_PAGE_4K;
94int mmu_virtual_psize = MMU_PAGE_4K;
95#ifdef CONFIG_HUGETLB_PAGE
96int mmu_huge_psize = MMU_PAGE_16M;
97unsigned int HPAGE_SHIFT;
98#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100100/* There are definitions of page sizes arrays to be used when none
101 * is provided by the firmware.
102 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100104/* Pre-POWER4 CPUs (4k pages only)
105 */
106struct mmu_psize_def mmu_psize_defaults_old[] = {
107 [MMU_PAGE_4K] = {
108 .shift = 12,
109 .sllp = 0,
110 .penc = 0,
111 .avpnm = 0,
112 .tlbiel = 0,
113 },
114};
115
116/* POWER4, GPUL, POWER5
117 *
118 * Support for 16Mb large pages
119 */
120struct mmu_psize_def mmu_psize_defaults_gp[] = {
121 [MMU_PAGE_4K] = {
122 .shift = 12,
123 .sllp = 0,
124 .penc = 0,
125 .avpnm = 0,
126 .tlbiel = 1,
127 },
128 [MMU_PAGE_16M] = {
129 .shift = 24,
130 .sllp = SLB_VSID_L,
131 .penc = 0,
132 .avpnm = 0x1UL,
133 .tlbiel = 0,
134 },
135};
136
137
138int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
139 unsigned long pstart, unsigned long mode, int psize)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140{
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100141 unsigned long vaddr, paddr;
142 unsigned int step, shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 unsigned long tmp_mode;
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100144 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100146 shift = mmu_psize_defs[psize].shift;
147 step = 1 << shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100149 for (vaddr = vstart, paddr = pstart; vaddr < vend;
150 vaddr += step, paddr += step) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 unsigned long vpn, hash, hpteg;
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100152 unsigned long vsid = get_kernel_vsid(vaddr);
153 unsigned long va = (vsid << 28) | (vaddr & 0x0fffffff);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100155 vpn = va >> shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 tmp_mode = mode;
157
158 /* Make non-kernel text non-executable */
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100159 if (!in_kernel_text(vaddr))
160 tmp_mode = mode | HPTE_R_N;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100162 hash = hpt_hash(va, shift);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
164
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100165 /* The crap below can be cleaned once ppd_md.probe() can
166 * set up the hash callbacks, thus we can just used the
167 * normal insert callback here.
168 */
Michael Ellerman4c551302005-09-23 14:47:58 +1000169#ifdef CONFIG_PPC_ISERIES
Paul Mackerras799d6042005-11-10 13:37:51 +1100170 if (_machine == PLATFORM_ISERIES_LPAR)
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100171 ret = iSeries_hpte_insert(hpteg, va,
Michael Ellermancaf80e52006-03-21 20:45:51 +1100172 paddr,
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100173 tmp_mode,
174 HPTE_V_BOLTED,
175 psize);
Michael Ellerman4c551302005-09-23 14:47:58 +1000176 else
177#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178#ifdef CONFIG_PPC_PSERIES
Paul Mackerras799d6042005-11-10 13:37:51 +1100179 if (_machine & PLATFORM_LPAR)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 ret = pSeries_lpar_hpte_insert(hpteg, va,
Michael Ellermancaf80e52006-03-21 20:45:51 +1100181 paddr,
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100182 tmp_mode,
183 HPTE_V_BOLTED,
184 psize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 else
Michael Ellerman4c551302005-09-23 14:47:58 +1000186#endif
187#ifdef CONFIG_PPC_MULTIPLATFORM
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 ret = native_hpte_insert(hpteg, va,
Michael Ellermancaf80e52006-03-21 20:45:51 +1100189 paddr,
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100190 tmp_mode, HPTE_V_BOLTED,
191 psize);
Michael Ellerman4c551302005-09-23 14:47:58 +1000192#endif
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100193 if (ret < 0)
194 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 }
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100196 return ret < 0 ? ret : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197}
198
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100199static int __init htab_dt_scan_page_sizes(unsigned long node,
200 const char *uname, int depth,
201 void *data)
202{
203 char *type = of_get_flat_dt_prop(node, "device_type", NULL);
204 u32 *prop;
205 unsigned long size = 0;
206
207 /* We are scanning "cpu" nodes only */
208 if (type == NULL || strcmp(type, "cpu") != 0)
209 return 0;
210
211 prop = (u32 *)of_get_flat_dt_prop(node,
212 "ibm,segment-page-sizes", &size);
213 if (prop != NULL) {
214 DBG("Page sizes from device-tree:\n");
215 size /= 4;
216 cur_cpu_spec->cpu_features &= ~(CPU_FTR_16M_PAGE);
217 while(size > 0) {
218 unsigned int shift = prop[0];
219 unsigned int slbenc = prop[1];
220 unsigned int lpnum = prop[2];
221 unsigned int lpenc = 0;
222 struct mmu_psize_def *def;
223 int idx = -1;
224
225 size -= 3; prop += 3;
226 while(size > 0 && lpnum) {
227 if (prop[0] == shift)
228 lpenc = prop[1];
229 prop += 2; size -= 2;
230 lpnum--;
231 }
232 switch(shift) {
233 case 0xc:
234 idx = MMU_PAGE_4K;
235 break;
236 case 0x10:
237 idx = MMU_PAGE_64K;
238 break;
239 case 0x14:
240 idx = MMU_PAGE_1M;
241 break;
242 case 0x18:
243 idx = MMU_PAGE_16M;
244 cur_cpu_spec->cpu_features |= CPU_FTR_16M_PAGE;
245 break;
246 case 0x22:
247 idx = MMU_PAGE_16G;
248 break;
249 }
250 if (idx < 0)
251 continue;
252 def = &mmu_psize_defs[idx];
253 def->shift = shift;
254 if (shift <= 23)
255 def->avpnm = 0;
256 else
257 def->avpnm = (1 << (shift - 23)) - 1;
258 def->sllp = slbenc;
259 def->penc = lpenc;
260 /* We don't know for sure what's up with tlbiel, so
261 * for now we only set it for 4K and 64K pages
262 */
263 if (idx == MMU_PAGE_4K || idx == MMU_PAGE_64K)
264 def->tlbiel = 1;
265 else
266 def->tlbiel = 0;
267
268 DBG(" %d: shift=%02x, sllp=%04x, avpnm=%08x, "
269 "tlbiel=%d, penc=%d\n",
270 idx, shift, def->sllp, def->avpnm, def->tlbiel,
271 def->penc);
272 }
273 return 1;
274 }
275 return 0;
276}
277
278
279static void __init htab_init_page_sizes(void)
280{
281 int rc;
282
283 /* Default to 4K pages only */
284 memcpy(mmu_psize_defs, mmu_psize_defaults_old,
285 sizeof(mmu_psize_defaults_old));
286
287 /*
288 * Try to find the available page sizes in the device-tree
289 */
290 rc = of_scan_flat_dt(htab_dt_scan_page_sizes, NULL);
291 if (rc != 0) /* Found */
292 goto found;
293
294 /*
295 * Not in the device-tree, let's fallback on known size
296 * list for 16M capable GP & GR
297 */
Paul Mackerras799d6042005-11-10 13:37:51 +1100298 if ((_machine != PLATFORM_ISERIES_LPAR) &&
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100299 cpu_has_feature(CPU_FTR_16M_PAGE))
300 memcpy(mmu_psize_defs, mmu_psize_defaults_gp,
301 sizeof(mmu_psize_defaults_gp));
302 found:
303 /*
304 * Pick a size for the linear mapping. Currently, we only support
305 * 16M, 1M and 4K which is the default
306 */
307 if (mmu_psize_defs[MMU_PAGE_16M].shift)
308 mmu_linear_psize = MMU_PAGE_16M;
309 else if (mmu_psize_defs[MMU_PAGE_1M].shift)
310 mmu_linear_psize = MMU_PAGE_1M;
311
312 /*
313 * Pick a size for the ordinary pages. Default is 4K, we support
314 * 64K if cache inhibited large pages are supported by the
315 * processor
316 */
317#ifdef CONFIG_PPC_64K_PAGES
318 if (mmu_psize_defs[MMU_PAGE_64K].shift &&
319 cpu_has_feature(CPU_FTR_CI_LARGE_PAGE))
320 mmu_virtual_psize = MMU_PAGE_64K;
321#endif
322
323 printk(KERN_INFO "Page orders: linear mapping = %d, others = %d\n",
324 mmu_psize_defs[mmu_linear_psize].shift,
325 mmu_psize_defs[mmu_virtual_psize].shift);
326
327#ifdef CONFIG_HUGETLB_PAGE
328 /* Init large page size. Currently, we pick 16M or 1M depending
329 * on what is available
330 */
331 if (mmu_psize_defs[MMU_PAGE_16M].shift)
332 mmu_huge_psize = MMU_PAGE_16M;
David Gibson7d24f0b2005-11-07 00:57:52 -0800333 /* With 4k/4level pagetables, we can't (for now) cope with a
334 * huge page size < PMD_SIZE */
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100335 else if (mmu_psize_defs[MMU_PAGE_1M].shift)
336 mmu_huge_psize = MMU_PAGE_1M;
337
338 /* Calculate HPAGE_SHIFT and sanity check it */
David Gibson7d24f0b2005-11-07 00:57:52 -0800339 if (mmu_psize_defs[mmu_huge_psize].shift > MIN_HUGEPTE_SHIFT &&
340 mmu_psize_defs[mmu_huge_psize].shift < SID_SHIFT)
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100341 HPAGE_SHIFT = mmu_psize_defs[mmu_huge_psize].shift;
342 else
343 HPAGE_SHIFT = 0; /* No huge pages dude ! */
344#endif /* CONFIG_HUGETLB_PAGE */
345}
346
347static int __init htab_dt_scan_pftsize(unsigned long node,
348 const char *uname, int depth,
349 void *data)
350{
351 char *type = of_get_flat_dt_prop(node, "device_type", NULL);
352 u32 *prop;
353
354 /* We are scanning "cpu" nodes only */
355 if (type == NULL || strcmp(type, "cpu") != 0)
356 return 0;
357
358 prop = (u32 *)of_get_flat_dt_prop(node, "ibm,pft-size", NULL);
359 if (prop != NULL) {
360 /* pft_size[0] is the NUMA CEC cookie */
361 ppc64_pft_size = prop[1];
362 return 1;
363 }
364 return 0;
365}
366
367static unsigned long __init htab_get_table_size(void)
Paul Mackerras3eac8c62005-10-12 16:58:53 +1000368{
Paul Mackerras799d6042005-11-10 13:37:51 +1100369 unsigned long mem_size, rnd_mem_size, pteg_count;
Paul Mackerras3eac8c62005-10-12 16:58:53 +1000370
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100371 /* If hash size isn't already provided by the platform, we try to
Adrian Bunk943ffb52006-01-10 00:10:13 +0100372 * retrieve it from the device-tree. If it's not there neither, we
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100373 * calculate it now based on the total RAM size
Paul Mackerras3eac8c62005-10-12 16:58:53 +1000374 */
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100375 if (ppc64_pft_size == 0)
376 of_scan_flat_dt(htab_dt_scan_pftsize, NULL);
Paul Mackerras3eac8c62005-10-12 16:58:53 +1000377 if (ppc64_pft_size)
378 return 1UL << ppc64_pft_size;
379
380 /* round mem_size up to next power of 2 */
Paul Mackerras799d6042005-11-10 13:37:51 +1100381 mem_size = lmb_phys_mem_size();
382 rnd_mem_size = 1UL << __ilog2(mem_size);
383 if (rnd_mem_size < mem_size)
Paul Mackerras3eac8c62005-10-12 16:58:53 +1000384 rnd_mem_size <<= 1;
385
386 /* # pages / 2 */
387 pteg_count = max(rnd_mem_size >> (12 + 1), 1UL << 11);
388
389 return pteg_count << 7;
390}
391
Mike Kravetz54b79242005-11-07 16:25:48 -0800392#ifdef CONFIG_MEMORY_HOTPLUG
393void create_section_mapping(unsigned long start, unsigned long end)
394{
Michael Ellermancaf80e52006-03-21 20:45:51 +1100395 BUG_ON(htab_bolt_mapping(start, end, __pa(start),
Mike Kravetz54b79242005-11-07 16:25:48 -0800396 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_COHERENT | PP_RWXX,
397 mmu_linear_psize));
398}
399#endif /* CONFIG_MEMORY_HOTPLUG */
400
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401void __init htab_initialize(void)
402{
Michael Ellerman337a7122006-02-21 17:22:55 +1100403 unsigned long table;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 unsigned long pteg_count;
405 unsigned long mode_rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 unsigned long base = 0, size = 0;
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100407 int i;
408
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 extern unsigned long tce_alloc_start, tce_alloc_end;
410
411 DBG(" -> htab_initialize()\n");
412
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100413 /* Initialize page sizes */
414 htab_init_page_sizes();
415
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 /*
417 * Calculate the required size of the htab. We want the number of
418 * PTEGs to equal one half the number of real pages.
419 */
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100420 htab_size_bytes = htab_get_table_size();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 pteg_count = htab_size_bytes >> 7;
422
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 htab_hash_mask = pteg_count - 1;
424
Michael Ellerman57cfb812006-03-21 20:45:59 +1100425 if (firmware_has_feature(FW_FEATURE_LPAR)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 /* Using a hypervisor which owns the htab */
427 htab_address = NULL;
428 _SDR1 = 0;
429 } else {
430 /* Find storage for the HPT. Must be contiguous in
431 * the absolute address space.
432 */
433 table = lmb_alloc(htab_size_bytes, htab_size_bytes);
434
435 DBG("Hash table allocated at %lx, size: %lx\n", table,
436 htab_size_bytes);
437
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 htab_address = abs_to_virt(table);
439
440 /* htab absolute addr + encoded htabsize */
441 _SDR1 = table + __ilog2(pteg_count) - 11;
442
443 /* Initialize the HPT with no entries */
444 memset((void *)table, 0, htab_size_bytes);
Paul Mackerras799d6042005-11-10 13:37:51 +1100445
446 /* Set SDR1 */
447 mtspr(SPRN_SDR1, _SDR1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 }
449
Anton Blanchard515bae92005-06-21 17:15:55 -0700450 mode_rw = _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_COHERENT | PP_RWXX;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
452 /* On U3 based machines, we need to reserve the DART area and
453 * _NOT_ map it to avoid cache paradoxes as it's remapped non
454 * cacheable later on
455 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
457 /* create bolted the linear mapping in the hash table */
458 for (i=0; i < lmb.memory.cnt; i++) {
Michael Ellermanb5666f72005-12-05 10:24:33 -0600459 base = (unsigned long)__va(lmb.memory.region[i].base);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 size = lmb.memory.region[i].size;
461
462 DBG("creating mapping for region: %lx : %lx\n", base, size);
463
464#ifdef CONFIG_U3_DART
465 /* Do not map the DART space. Fortunately, it will be aligned
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100466 * in such a way that it will not cross two lmb regions and
467 * will fit within a single 16Mb page.
468 * The DART space is assumed to be a full 16Mb region even if
469 * we only use 2Mb of that space. We will use more of it later
470 * for AGP GART. We have to use a full 16Mb large page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 */
472 DBG("DART base: %lx\n", dart_tablebase);
473
474 if (dart_tablebase != 0 && dart_tablebase >= base
475 && dart_tablebase < (base + size)) {
Michael Ellermancaf80e52006-03-21 20:45:51 +1100476 unsigned long dart_table_end = dart_tablebase + 16 * MB;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 if (base != dart_tablebase)
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100478 BUG_ON(htab_bolt_mapping(base, dart_tablebase,
Michael Ellermancaf80e52006-03-21 20:45:51 +1100479 __pa(base), mode_rw,
480 mmu_linear_psize));
481 if ((base + size) > dart_table_end)
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100482 BUG_ON(htab_bolt_mapping(dart_tablebase+16*MB,
Michael Ellermancaf80e52006-03-21 20:45:51 +1100483 base + size,
484 __pa(dart_table_end),
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100485 mode_rw,
486 mmu_linear_psize));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 continue;
488 }
489#endif /* CONFIG_U3_DART */
Michael Ellermancaf80e52006-03-21 20:45:51 +1100490 BUG_ON(htab_bolt_mapping(base, base + size, __pa(base),
491 mode_rw, mmu_linear_psize));
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100492 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
494 /*
495 * If we have a memory_limit and we've allocated TCEs then we need to
496 * explicitly map the TCE area at the top of RAM. We also cope with the
497 * case that the TCEs start below memory_limit.
498 * tce_alloc_start/end are 16MB aligned so the mapping should work
499 * for either 4K or 16MB pages.
500 */
501 if (tce_alloc_start) {
Michael Ellermanb5666f72005-12-05 10:24:33 -0600502 tce_alloc_start = (unsigned long)__va(tce_alloc_start);
503 tce_alloc_end = (unsigned long)__va(tce_alloc_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
505 if (base + size >= tce_alloc_start)
506 tce_alloc_start = base + size + 1;
507
Michael Ellermancaf80e52006-03-21 20:45:51 +1100508 BUG_ON(htab_bolt_mapping(tce_alloc_start, tce_alloc_end,
509 __pa(tce_alloc_start), mode_rw,
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100510 mmu_linear_psize));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 }
512
513 DBG(" <- htab_initialize()\n");
514}
515#undef KB
516#undef MB
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517
Anton Blancharde597cb322005-12-29 10:46:29 +1100518void htab_initialize_secondary(void)
Paul Mackerras799d6042005-11-10 13:37:51 +1100519{
Michael Ellerman57cfb812006-03-21 20:45:59 +1100520 if (!firmware_has_feature(FW_FEATURE_LPAR))
Paul Mackerras799d6042005-11-10 13:37:51 +1100521 mtspr(SPRN_SDR1, _SDR1);
522}
523
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524/*
525 * Called by asm hashtable.S for doing lazy icache flush
526 */
527unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap)
528{
529 struct page *page;
530
Benjamin Herrenschmidt76c8e252005-11-08 11:21:05 +1100531 if (!pfn_valid(pte_pfn(pte)))
532 return pp;
533
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 page = pte_page(pte);
535
536 /* page is dirty */
537 if (!test_bit(PG_arch_1, &page->flags) && !PageReserved(page)) {
538 if (trap == 0x400) {
539 __flush_dcache_icache(page_address(page));
540 set_bit(PG_arch_1, &page->flags);
541 } else
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100542 pp |= HPTE_R_N;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 }
544 return pp;
545}
546
547/* Result code is:
548 * 0 - handled
549 * 1 - normal page fault
550 * -1 - critical hash insertion error
551 */
552int hash_page(unsigned long ea, unsigned long access, unsigned long trap)
553{
554 void *pgdir;
555 unsigned long vsid;
556 struct mm_struct *mm;
557 pte_t *ptep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 cpumask_t tmp;
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100559 int rc, user_region = 0, local = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100561 DBG_LOW("hash_page(ea=%016lx, access=%lx, trap=%lx\n",
562 ea, access, trap);
David Gibson1f8d4192005-05-05 16:15:13 -0700563
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100564 if ((ea & ~REGION_MASK) >= PGTABLE_RANGE) {
565 DBG_LOW(" out of pgtable range !\n");
566 return 1;
567 }
568
569 /* Get region & vsid */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 switch (REGION_ID(ea)) {
571 case USER_REGION_ID:
572 user_region = 1;
573 mm = current->mm;
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100574 if (! mm) {
575 DBG_LOW(" user region with no mm !\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 return 1;
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100577 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 vsid = get_vsid(mm->context.id, ea);
579 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 case VMALLOC_REGION_ID:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 mm = &init_mm;
582 vsid = get_kernel_vsid(ea);
583 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 default:
585 /* Not a valid range
586 * Send the problem up to do_page_fault
587 */
588 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 }
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100590 DBG_LOW(" mm=%p, mm->pgdir=%p, vsid=%016lx\n", mm, mm->pgd, vsid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100592 /* Get pgdir */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 pgdir = mm->pgd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 if (pgdir == NULL)
595 return 1;
596
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100597 /* Check CPU locality */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 tmp = cpumask_of_cpu(smp_processor_id());
599 if (user_region && cpus_equal(mm->cpu_vm_mask, tmp))
600 local = 1;
601
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100602 /* Handle hugepage regions */
603 if (unlikely(in_hugepage_area(mm->context, ea))) {
604 DBG_LOW(" -> huge page !\n");
David Gibsoncbf52af2005-12-09 14:20:52 +1100605 return hash_huge_page(mm, access, ea, vsid, local, trap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 }
607
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100608 /* Get PTE and page size from page tables */
609 ptep = find_linux_pte(pgdir, ea);
610 if (ptep == NULL || !pte_present(*ptep)) {
611 DBG_LOW(" no PTE !\n");
612 return 1;
613 }
614
615#ifndef CONFIG_PPC_64K_PAGES
616 DBG_LOW(" i-pte: %016lx\n", pte_val(*ptep));
617#else
618 DBG_LOW(" i-pte: %016lx %016lx\n", pte_val(*ptep),
619 pte_val(*(ptep + PTRS_PER_PTE)));
620#endif
621 /* Pre-check access permissions (will be re-checked atomically
622 * in __hash_page_XX but this pre-check is a fast path
623 */
624 if (access & ~pte_val(*ptep)) {
625 DBG_LOW(" no access !\n");
626 return 1;
627 }
628
629 /* Do actual hashing */
630#ifndef CONFIG_PPC_64K_PAGES
631 rc = __hash_page_4K(ea, access, vsid, ptep, trap, local);
632#else
633 if (mmu_virtual_psize == MMU_PAGE_64K)
634 rc = __hash_page_64K(ea, access, vsid, ptep, trap, local);
635 else
636 rc = __hash_page_4K(ea, access, vsid, ptep, trap, local);
637#endif /* CONFIG_PPC_64K_PAGES */
638
639#ifndef CONFIG_PPC_64K_PAGES
640 DBG_LOW(" o-pte: %016lx\n", pte_val(*ptep));
641#else
642 DBG_LOW(" o-pte: %016lx %016lx\n", pte_val(*ptep),
643 pte_val(*(ptep + PTRS_PER_PTE)));
644#endif
645 DBG_LOW(" -> rc=%d\n", rc);
646 return rc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647}
Arnd Bergmann67207b92005-11-15 15:53:48 -0500648EXPORT_SYMBOL_GPL(hash_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100650void hash_preload(struct mm_struct *mm, unsigned long ea,
651 unsigned long access, unsigned long trap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652{
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100653 unsigned long vsid;
654 void *pgdir;
655 pte_t *ptep;
656 cpumask_t mask;
657 unsigned long flags;
658 int local = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100660 /* We don't want huge pages prefaulted for now
661 */
662 if (unlikely(in_hugepage_area(mm->context, ea)))
663 return;
664
665 DBG_LOW("hash_preload(mm=%p, mm->pgdir=%p, ea=%016lx, access=%lx,"
666 " trap=%lx\n", mm, mm->pgd, ea, access, trap);
667
668 /* Get PTE, VSID, access mask */
669 pgdir = mm->pgd;
670 if (pgdir == NULL)
671 return;
672 ptep = find_linux_pte(pgdir, ea);
673 if (!ptep)
674 return;
675 vsid = get_vsid(mm->context.id, ea);
676
677 /* Hash it in */
678 local_irq_save(flags);
679 mask = cpumask_of_cpu(smp_processor_id());
680 if (cpus_equal(mm->cpu_vm_mask, mask))
681 local = 1;
682#ifndef CONFIG_PPC_64K_PAGES
683 __hash_page_4K(ea, access, vsid, ptep, trap, local);
684#else
685 if (mmu_virtual_psize == MMU_PAGE_64K)
686 __hash_page_64K(ea, access, vsid, ptep, trap, local);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 else
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100688 __hash_page_4K(ea, access, vsid, ptep, trap, local);
689#endif /* CONFIG_PPC_64K_PAGES */
690 local_irq_restore(flags);
691}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100693void flush_hash_page(unsigned long va, real_pte_t pte, int psize, int local)
694{
695 unsigned long hash, index, shift, hidx, slot;
696
697 DBG_LOW("flush_hash_page(va=%016x)\n", va);
698 pte_iterate_hashed_subpages(pte, psize, va, index, shift) {
699 hash = hpt_hash(va, shift);
700 hidx = __rpte_to_hidx(pte, index);
701 if (hidx & _PTEIDX_SECONDARY)
702 hash = ~hash;
703 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
704 slot += hidx & _PTEIDX_GROUP_IX;
705 DBG_LOW(" sub %d: hash=%x, hidx=%x\n", index, slot, hidx);
706 ppc_md.hpte_invalidate(slot, va, psize, local);
707 } pte_iterate_hashed_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708}
709
Benjamin Herrenschmidt61b1a942005-09-20 13:52:50 +1000710void flush_hash_range(unsigned long number, int local)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711{
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100712 if (ppc_md.flush_hash_range)
Benjamin Herrenschmidt61b1a942005-09-20 13:52:50 +1000713 ppc_md.flush_hash_range(number, local);
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100714 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 int i;
Benjamin Herrenschmidt61b1a942005-09-20 13:52:50 +1000716 struct ppc64_tlb_batch *batch =
717 &__get_cpu_var(ppc64_tlb_batch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
719 for (i = 0; i < number; i++)
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100720 flush_hash_page(batch->vaddr[i], batch->pte[i],
721 batch->psize, local);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 }
723}
724
725static inline void make_bl(unsigned int *insn_addr, void *func)
726{
727 unsigned long funcp = *((unsigned long *)func);
728 int offset = funcp - (unsigned long)insn_addr;
729
730 *insn_addr = (unsigned int)(0x48000001 | (offset & 0x03fffffc));
731 flush_icache_range((unsigned long)insn_addr, 4+
732 (unsigned long)insn_addr);
733}
734
735/*
736 * low_hash_fault is called when we the low level hash code failed
737 * to instert a PTE due to an hypervisor error
738 */
739void low_hash_fault(struct pt_regs *regs, unsigned long address)
740{
741 if (user_mode(regs)) {
742 siginfo_t info;
743
744 info.si_signo = SIGBUS;
745 info.si_errno = 0;
746 info.si_code = BUS_ADRERR;
747 info.si_addr = (void __user *)address;
748 force_sig_info(SIGBUS, &info, current);
749 return;
750 }
751 bad_page_fault(regs, address, SIGBUS);
752}
753
754void __init htab_finish_init(void)
755{
756 extern unsigned int *htab_call_hpte_insert1;
757 extern unsigned int *htab_call_hpte_insert2;
758 extern unsigned int *htab_call_hpte_remove;
759 extern unsigned int *htab_call_hpte_updatepp;
760
Benjamin Herrenschmidt3c726f82005-11-07 11:06:55 +1100761#ifdef CONFIG_PPC_64K_PAGES
762 extern unsigned int *ht64_call_hpte_insert1;
763 extern unsigned int *ht64_call_hpte_insert2;
764 extern unsigned int *ht64_call_hpte_remove;
765 extern unsigned int *ht64_call_hpte_updatepp;
766
767 make_bl(ht64_call_hpte_insert1, ppc_md.hpte_insert);
768 make_bl(ht64_call_hpte_insert2, ppc_md.hpte_insert);
769 make_bl(ht64_call_hpte_remove, ppc_md.hpte_remove);
770 make_bl(ht64_call_hpte_updatepp, ppc_md.hpte_updatepp);
771#endif /* CONFIG_PPC_64K_PAGES */
772
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 make_bl(htab_call_hpte_insert1, ppc_md.hpte_insert);
774 make_bl(htab_call_hpte_insert2, ppc_md.hpte_insert);
775 make_bl(htab_call_hpte_remove, ppc_md.hpte_remove);
776 make_bl(htab_call_hpte_updatepp, ppc_md.hpte_updatepp);
777}