2 * PowerPC64 port by Mike Corrigan and Dave Engebretsen
3 * {mikejc|engebret}@us.ibm.com
5 * Copyright (c) 2000 Mike Corrigan <mikejc@us.ibm.com>
7 * SMP scalability work:
8 * Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
13 * PowerPC Hashed Page Table functions
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.
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>
35 #include <asm/processor.h>
36 #include <asm/pgtable.h>
38 #include <asm/mmu_context.h>
40 #include <asm/types.h>
41 #include <asm/system.h>
42 #include <asm/uaccess.h>
43 #include <asm/machdep.h>
45 #include <asm/abs_addr.h>
46 #include <asm/tlbflush.h>
50 #include <asm/cacheflush.h>
51 #include <asm/cputable.h>
52 #include <asm/sections.h>
56 #define DBG(fmt...) udbg_printf(fmt)
62 #define DBG_LOW(fmt...) udbg_printf(fmt)
64 #define DBG_LOW(fmt...)
71 * Note: pte --> Linux PTE
72 * HPTE --> PowerPC Hashed Page Table Entry
75 * htab_initialize is called with the MMU off (of course), but
76 * the kernel has been copied down to zero so it can directly
77 * reference global data. At this point it is very difficult
78 * to print debug info.
83 extern unsigned long dart_tablebase
;
84 #endif /* CONFIG_U3_DART */
86 static unsigned long _SDR1
;
87 struct mmu_psize_def mmu_psize_defs
[MMU_PAGE_COUNT
];
89 struct hash_pte
*htab_address
;
90 unsigned long htab_size_bytes
;
91 unsigned long htab_hash_mask
;
92 int mmu_linear_psize
= MMU_PAGE_4K
;
93 int mmu_virtual_psize
= MMU_PAGE_4K
;
94 int mmu_vmalloc_psize
= MMU_PAGE_4K
;
95 int mmu_io_psize
= MMU_PAGE_4K
;
96 #ifdef CONFIG_HUGETLB_PAGE
97 int mmu_huge_psize
= MMU_PAGE_16M
;
98 unsigned int HPAGE_SHIFT
;
100 #ifdef CONFIG_PPC_64K_PAGES
101 int mmu_ci_restrictions
;
103 #ifdef CONFIG_DEBUG_PAGEALLOC
104 static u8
*linear_map_hash_slots
;
105 static unsigned long linear_map_hash_count
;
106 static DEFINE_SPINLOCK(linear_map_hash_lock
);
107 #endif /* CONFIG_DEBUG_PAGEALLOC */
109 /* There are definitions of page sizes arrays to be used when none
110 * is provided by the firmware.
113 /* Pre-POWER4 CPUs (4k pages only)
115 struct mmu_psize_def mmu_psize_defaults_old
[] = {
125 /* POWER4, GPUL, POWER5
127 * Support for 16Mb large pages
129 struct mmu_psize_def mmu_psize_defaults_gp
[] = {
147 int htab_bolt_mapping(unsigned long vstart
, unsigned long vend
,
148 unsigned long pstart
, unsigned long mode
, int psize
)
150 unsigned long vaddr
, paddr
;
151 unsigned int step
, shift
;
152 unsigned long tmp_mode
;
155 shift
= mmu_psize_defs
[psize
].shift
;
158 for (vaddr
= vstart
, paddr
= pstart
; vaddr
< vend
;
159 vaddr
+= step
, paddr
+= step
) {
160 unsigned long hash
, hpteg
;
161 unsigned long vsid
= get_kernel_vsid(vaddr
);
162 unsigned long va
= (vsid
<< 28) | (vaddr
& 0x0fffffff);
166 /* Make non-kernel text non-executable */
167 if (!in_kernel_text(vaddr
))
168 tmp_mode
= mode
| HPTE_R_N
;
170 hash
= hpt_hash(va
, shift
);
171 hpteg
= ((hash
& htab_hash_mask
) * HPTES_PER_GROUP
);
173 DBG("htab_bolt_mapping: calling %p\n", ppc_md
.hpte_insert
);
175 BUG_ON(!ppc_md
.hpte_insert
);
176 ret
= ppc_md
.hpte_insert(hpteg
, va
, paddr
,
177 tmp_mode
, HPTE_V_BOLTED
, psize
);
181 #ifdef CONFIG_DEBUG_PAGEALLOC
182 if ((paddr
>> PAGE_SHIFT
) < linear_map_hash_count
)
183 linear_map_hash_slots
[paddr
>> PAGE_SHIFT
] = ret
| 0x80;
184 #endif /* CONFIG_DEBUG_PAGEALLOC */
186 return ret
< 0 ? ret
: 0;
189 static int __init
htab_dt_scan_page_sizes(unsigned long node
,
190 const char *uname
, int depth
,
193 char *type
= of_get_flat_dt_prop(node
, "device_type", NULL
);
195 unsigned long size
= 0;
197 /* We are scanning "cpu" nodes only */
198 if (type
== NULL
|| strcmp(type
, "cpu") != 0)
201 prop
= (u32
*)of_get_flat_dt_prop(node
,
202 "ibm,segment-page-sizes", &size
);
204 DBG("Page sizes from device-tree:\n");
206 cur_cpu_spec
->cpu_features
&= ~(CPU_FTR_16M_PAGE
);
208 unsigned int shift
= prop
[0];
209 unsigned int slbenc
= prop
[1];
210 unsigned int lpnum
= prop
[2];
211 unsigned int lpenc
= 0;
212 struct mmu_psize_def
*def
;
215 size
-= 3; prop
+= 3;
216 while(size
> 0 && lpnum
) {
217 if (prop
[0] == shift
)
219 prop
+= 2; size
-= 2;
234 cur_cpu_spec
->cpu_features
|= CPU_FTR_16M_PAGE
;
242 def
= &mmu_psize_defs
[idx
];
247 def
->avpnm
= (1 << (shift
- 23)) - 1;
250 /* We don't know for sure what's up with tlbiel, so
251 * for now we only set it for 4K and 64K pages
253 if (idx
== MMU_PAGE_4K
|| idx
== MMU_PAGE_64K
)
258 DBG(" %d: shift=%02x, sllp=%04x, avpnm=%08x, "
259 "tlbiel=%d, penc=%d\n",
260 idx
, shift
, def
->sllp
, def
->avpnm
, def
->tlbiel
,
269 static void __init
htab_init_page_sizes(void)
273 /* Default to 4K pages only */
274 memcpy(mmu_psize_defs
, mmu_psize_defaults_old
,
275 sizeof(mmu_psize_defaults_old
));
278 * Try to find the available page sizes in the device-tree
280 rc
= of_scan_flat_dt(htab_dt_scan_page_sizes
, NULL
);
281 if (rc
!= 0) /* Found */
285 * Not in the device-tree, let's fallback on known size
286 * list for 16M capable GP & GR
288 if (cpu_has_feature(CPU_FTR_16M_PAGE
))
289 memcpy(mmu_psize_defs
, mmu_psize_defaults_gp
,
290 sizeof(mmu_psize_defaults_gp
));
292 #ifndef CONFIG_DEBUG_PAGEALLOC
294 * Pick a size for the linear mapping. Currently, we only support
295 * 16M, 1M and 4K which is the default
297 if (mmu_psize_defs
[MMU_PAGE_16M
].shift
)
298 mmu_linear_psize
= MMU_PAGE_16M
;
299 else if (mmu_psize_defs
[MMU_PAGE_1M
].shift
)
300 mmu_linear_psize
= MMU_PAGE_1M
;
301 #endif /* CONFIG_DEBUG_PAGEALLOC */
303 #ifdef CONFIG_PPC_64K_PAGES
305 * Pick a size for the ordinary pages. Default is 4K, we support
306 * 64K for user mappings and vmalloc if supported by the processor.
307 * We only use 64k for ioremap if the processor
308 * (and firmware) support cache-inhibited large pages.
309 * If not, we use 4k and set mmu_ci_restrictions so that
310 * hash_page knows to switch processes that use cache-inhibited
311 * mappings to 4k pages.
313 if (mmu_psize_defs
[MMU_PAGE_64K
].shift
) {
314 mmu_virtual_psize
= MMU_PAGE_64K
;
315 mmu_vmalloc_psize
= MMU_PAGE_64K
;
316 if (mmu_linear_psize
== MMU_PAGE_4K
)
317 mmu_linear_psize
= MMU_PAGE_64K
;
318 if (cpu_has_feature(CPU_FTR_CI_LARGE_PAGE
))
319 mmu_io_psize
= MMU_PAGE_64K
;
321 mmu_ci_restrictions
= 1;
323 #endif /* CONFIG_PPC_64K_PAGES */
325 printk(KERN_DEBUG
"Page orders: linear mapping = %d, "
326 "virtual = %d, io = %d\n",
327 mmu_psize_defs
[mmu_linear_psize
].shift
,
328 mmu_psize_defs
[mmu_virtual_psize
].shift
,
329 mmu_psize_defs
[mmu_io_psize
].shift
);
331 #ifdef CONFIG_HUGETLB_PAGE
332 /* Init large page size. Currently, we pick 16M or 1M depending
333 * on what is available
335 if (mmu_psize_defs
[MMU_PAGE_16M
].shift
)
336 mmu_huge_psize
= MMU_PAGE_16M
;
337 /* With 4k/4level pagetables, we can't (for now) cope with a
338 * huge page size < PMD_SIZE */
339 else if (mmu_psize_defs
[MMU_PAGE_1M
].shift
)
340 mmu_huge_psize
= MMU_PAGE_1M
;
342 /* Calculate HPAGE_SHIFT and sanity check it */
343 if (mmu_psize_defs
[mmu_huge_psize
].shift
> MIN_HUGEPTE_SHIFT
&&
344 mmu_psize_defs
[mmu_huge_psize
].shift
< SID_SHIFT
)
345 HPAGE_SHIFT
= mmu_psize_defs
[mmu_huge_psize
].shift
;
347 HPAGE_SHIFT
= 0; /* No huge pages dude ! */
348 #endif /* CONFIG_HUGETLB_PAGE */
351 static int __init
htab_dt_scan_pftsize(unsigned long node
,
352 const char *uname
, int depth
,
355 char *type
= of_get_flat_dt_prop(node
, "device_type", NULL
);
358 /* We are scanning "cpu" nodes only */
359 if (type
== NULL
|| strcmp(type
, "cpu") != 0)
362 prop
= (u32
*)of_get_flat_dt_prop(node
, "ibm,pft-size", NULL
);
364 /* pft_size[0] is the NUMA CEC cookie */
365 ppc64_pft_size
= prop
[1];
371 static unsigned long __init
htab_get_table_size(void)
373 unsigned long mem_size
, rnd_mem_size
, pteg_count
;
375 /* If hash size isn't already provided by the platform, we try to
376 * retrieve it from the device-tree. If it's not there neither, we
377 * calculate it now based on the total RAM size
379 if (ppc64_pft_size
== 0)
380 of_scan_flat_dt(htab_dt_scan_pftsize
, NULL
);
382 return 1UL << ppc64_pft_size
;
384 /* round mem_size up to next power of 2 */
385 mem_size
= lmb_phys_mem_size();
386 rnd_mem_size
= 1UL << __ilog2(mem_size
);
387 if (rnd_mem_size
< mem_size
)
391 pteg_count
= max(rnd_mem_size
>> (12 + 1), 1UL << 11);
393 return pteg_count
<< 7;
396 #ifdef CONFIG_MEMORY_HOTPLUG
397 void create_section_mapping(unsigned long start
, unsigned long end
)
399 BUG_ON(htab_bolt_mapping(start
, end
, __pa(start
),
400 _PAGE_ACCESSED
| _PAGE_DIRTY
| _PAGE_COHERENT
| PP_RWXX
,
403 #endif /* CONFIG_MEMORY_HOTPLUG */
405 static inline void make_bl(unsigned int *insn_addr
, void *func
)
407 unsigned long funcp
= *((unsigned long *)func
);
408 int offset
= funcp
- (unsigned long)insn_addr
;
410 *insn_addr
= (unsigned int)(0x48000001 | (offset
& 0x03fffffc));
411 flush_icache_range((unsigned long)insn_addr
, 4+
412 (unsigned long)insn_addr
);
415 static void __init
htab_finish_init(void)
417 extern unsigned int *htab_call_hpte_insert1
;
418 extern unsigned int *htab_call_hpte_insert2
;
419 extern unsigned int *htab_call_hpte_remove
;
420 extern unsigned int *htab_call_hpte_updatepp
;
422 #ifdef CONFIG_PPC_HAS_HASH_64K
423 extern unsigned int *ht64_call_hpte_insert1
;
424 extern unsigned int *ht64_call_hpte_insert2
;
425 extern unsigned int *ht64_call_hpte_remove
;
426 extern unsigned int *ht64_call_hpte_updatepp
;
428 make_bl(ht64_call_hpte_insert1
, ppc_md
.hpte_insert
);
429 make_bl(ht64_call_hpte_insert2
, ppc_md
.hpte_insert
);
430 make_bl(ht64_call_hpte_remove
, ppc_md
.hpte_remove
);
431 make_bl(ht64_call_hpte_updatepp
, ppc_md
.hpte_updatepp
);
432 #endif /* CONFIG_PPC_HAS_HASH_64K */
434 make_bl(htab_call_hpte_insert1
, ppc_md
.hpte_insert
);
435 make_bl(htab_call_hpte_insert2
, ppc_md
.hpte_insert
);
436 make_bl(htab_call_hpte_remove
, ppc_md
.hpte_remove
);
437 make_bl(htab_call_hpte_updatepp
, ppc_md
.hpte_updatepp
);
440 void __init
htab_initialize(void)
443 unsigned long pteg_count
;
444 unsigned long mode_rw
;
445 unsigned long base
= 0, size
= 0;
448 extern unsigned long tce_alloc_start
, tce_alloc_end
;
450 DBG(" -> htab_initialize()\n");
452 /* Initialize page sizes */
453 htab_init_page_sizes();
456 * Calculate the required size of the htab. We want the number of
457 * PTEGs to equal one half the number of real pages.
459 htab_size_bytes
= htab_get_table_size();
460 pteg_count
= htab_size_bytes
>> 7;
462 htab_hash_mask
= pteg_count
- 1;
464 if (firmware_has_feature(FW_FEATURE_LPAR
)) {
465 /* Using a hypervisor which owns the htab */
469 /* Find storage for the HPT. Must be contiguous in
470 * the absolute address space.
472 table
= lmb_alloc(htab_size_bytes
, htab_size_bytes
);
474 DBG("Hash table allocated at %lx, size: %lx\n", table
,
477 htab_address
= abs_to_virt(table
);
479 /* htab absolute addr + encoded htabsize */
480 _SDR1
= table
+ __ilog2(pteg_count
) - 11;
482 /* Initialize the HPT with no entries */
483 memset((void *)table
, 0, htab_size_bytes
);
486 mtspr(SPRN_SDR1
, _SDR1
);
489 mode_rw
= _PAGE_ACCESSED
| _PAGE_DIRTY
| _PAGE_COHERENT
| PP_RWXX
;
491 #ifdef CONFIG_DEBUG_PAGEALLOC
492 linear_map_hash_count
= lmb_end_of_DRAM() >> PAGE_SHIFT
;
493 linear_map_hash_slots
= __va(lmb_alloc_base(linear_map_hash_count
,
495 memset(linear_map_hash_slots
, 0, linear_map_hash_count
);
496 #endif /* CONFIG_DEBUG_PAGEALLOC */
498 /* On U3 based machines, we need to reserve the DART area and
499 * _NOT_ map it to avoid cache paradoxes as it's remapped non
503 /* create bolted the linear mapping in the hash table */
504 for (i
=0; i
< lmb
.memory
.cnt
; i
++) {
505 base
= (unsigned long)__va(lmb
.memory
.region
[i
].base
);
506 size
= lmb
.memory
.region
[i
].size
;
508 DBG("creating mapping for region: %lx : %lx\n", base
, size
);
510 #ifdef CONFIG_U3_DART
511 /* Do not map the DART space. Fortunately, it will be aligned
512 * in such a way that it will not cross two lmb regions and
513 * will fit within a single 16Mb page.
514 * The DART space is assumed to be a full 16Mb region even if
515 * we only use 2Mb of that space. We will use more of it later
516 * for AGP GART. We have to use a full 16Mb large page.
518 DBG("DART base: %lx\n", dart_tablebase
);
520 if (dart_tablebase
!= 0 && dart_tablebase
>= base
521 && dart_tablebase
< (base
+ size
)) {
522 unsigned long dart_table_end
= dart_tablebase
+ 16 * MB
;
523 if (base
!= dart_tablebase
)
524 BUG_ON(htab_bolt_mapping(base
, dart_tablebase
,
527 if ((base
+ size
) > dart_table_end
)
528 BUG_ON(htab_bolt_mapping(dart_tablebase
+16*MB
,
530 __pa(dart_table_end
),
535 #endif /* CONFIG_U3_DART */
536 BUG_ON(htab_bolt_mapping(base
, base
+ size
, __pa(base
),
537 mode_rw
, mmu_linear_psize
));
541 * If we have a memory_limit and we've allocated TCEs then we need to
542 * explicitly map the TCE area at the top of RAM. We also cope with the
543 * case that the TCEs start below memory_limit.
544 * tce_alloc_start/end are 16MB aligned so the mapping should work
545 * for either 4K or 16MB pages.
547 if (tce_alloc_start
) {
548 tce_alloc_start
= (unsigned long)__va(tce_alloc_start
);
549 tce_alloc_end
= (unsigned long)__va(tce_alloc_end
);
551 if (base
+ size
>= tce_alloc_start
)
552 tce_alloc_start
= base
+ size
+ 1;
554 BUG_ON(htab_bolt_mapping(tce_alloc_start
, tce_alloc_end
,
555 __pa(tce_alloc_start
), mode_rw
,
561 DBG(" <- htab_initialize()\n");
566 void htab_initialize_secondary(void)
568 if (!firmware_has_feature(FW_FEATURE_LPAR
))
569 mtspr(SPRN_SDR1
, _SDR1
);
573 * Called by asm hashtable.S for doing lazy icache flush
575 unsigned int hash_page_do_lazy_icache(unsigned int pp
, pte_t pte
, int trap
)
579 if (!pfn_valid(pte_pfn(pte
)))
582 page
= pte_page(pte
);
585 if (!test_bit(PG_arch_1
, &page
->flags
) && !PageReserved(page
)) {
587 __flush_dcache_icache(page_address(page
));
588 set_bit(PG_arch_1
, &page
->flags
);
596 * Demote a segment to using 4k pages.
597 * For now this makes the whole process use 4k pages.
599 #ifdef CONFIG_PPC_64K_PAGES
600 static void demote_segment_4k(struct mm_struct
*mm
, unsigned long addr
)
602 if (mm
->context
.user_psize
== MMU_PAGE_4K
)
604 slice_set_user_psize(mm
, MMU_PAGE_4K
);
605 #ifdef CONFIG_SPU_BASE
606 spu_flush_all_slbs(mm
);
609 #endif /* CONFIG_PPC_64K_PAGES */
613 * 1 - normal page fault
614 * -1 - critical hash insertion error
616 int hash_page(unsigned long ea
, unsigned long access
, unsigned long trap
)
620 struct mm_struct
*mm
;
623 int rc
, user_region
= 0, local
= 0;
626 DBG_LOW("hash_page(ea=%016lx, access=%lx, trap=%lx\n",
629 if ((ea
& ~REGION_MASK
) >= PGTABLE_RANGE
) {
630 DBG_LOW(" out of pgtable range !\n");
634 /* Get region & vsid */
635 switch (REGION_ID(ea
)) {
640 DBG_LOW(" user region with no mm !\n");
643 vsid
= get_vsid(mm
->context
.id
, ea
);
644 #ifdef CONFIG_PPC_MM_SLICES
645 psize
= get_slice_psize(mm
, ea
);
647 psize
= mm
->context
.user_psize
;
650 case VMALLOC_REGION_ID
:
652 vsid
= get_kernel_vsid(ea
);
653 if (ea
< VMALLOC_END
)
654 psize
= mmu_vmalloc_psize
;
656 psize
= mmu_io_psize
;
660 * Send the problem up to do_page_fault
664 DBG_LOW(" mm=%p, mm->pgdir=%p, vsid=%016lx\n", mm
, mm
->pgd
, vsid
);
671 /* Check CPU locality */
672 tmp
= cpumask_of_cpu(smp_processor_id());
673 if (user_region
&& cpus_equal(mm
->cpu_vm_mask
, tmp
))
676 #ifdef CONFIG_HUGETLB_PAGE
677 /* Handle hugepage regions */
678 if (HPAGE_SHIFT
&& psize
== mmu_huge_psize
) {
679 DBG_LOW(" -> huge page !\n");
680 return hash_huge_page(mm
, access
, ea
, vsid
, local
, trap
);
682 #endif /* CONFIG_HUGETLB_PAGE */
684 #ifndef CONFIG_PPC_64K_PAGES
685 /* If we use 4K pages and our psize is not 4K, then we are hitting
686 * a special driver mapping, we need to align the address before
689 if (psize
!= MMU_PAGE_4K
)
690 ea
&= ~((1ul << mmu_psize_defs
[psize
].shift
) - 1);
691 #endif /* CONFIG_PPC_64K_PAGES */
693 /* Get PTE and page size from page tables */
694 ptep
= find_linux_pte(pgdir
, ea
);
695 if (ptep
== NULL
|| !pte_present(*ptep
)) {
696 DBG_LOW(" no PTE !\n");
700 #ifndef CONFIG_PPC_64K_PAGES
701 DBG_LOW(" i-pte: %016lx\n", pte_val(*ptep
));
703 DBG_LOW(" i-pte: %016lx %016lx\n", pte_val(*ptep
),
704 pte_val(*(ptep
+ PTRS_PER_PTE
)));
706 /* Pre-check access permissions (will be re-checked atomically
707 * in __hash_page_XX but this pre-check is a fast path
709 if (access
& ~pte_val(*ptep
)) {
710 DBG_LOW(" no access !\n");
714 /* Do actual hashing */
715 #ifdef CONFIG_PPC_64K_PAGES
716 /* If _PAGE_4K_PFN is set, make sure this is a 4k segment */
717 if (pte_val(*ptep
) & _PAGE_4K_PFN
) {
718 demote_segment_4k(mm
, ea
);
722 /* If this PTE is non-cacheable and we have restrictions on
723 * using non cacheable large pages, then we switch to 4k
725 if (mmu_ci_restrictions
&& psize
== MMU_PAGE_64K
&&
726 (pte_val(*ptep
) & _PAGE_NO_CACHE
)) {
728 demote_segment_4k(mm
, ea
);
730 } else if (ea
< VMALLOC_END
) {
732 * some driver did a non-cacheable mapping
733 * in vmalloc space, so switch vmalloc
736 printk(KERN_ALERT
"Reducing vmalloc segment "
737 "to 4kB pages because of "
738 "non-cacheable mapping\n");
739 psize
= mmu_vmalloc_psize
= MMU_PAGE_4K
;
740 #ifdef CONFIG_SPU_BASE
741 spu_flush_all_slbs(mm
);
746 if (psize
!= get_paca()->context
.user_psize
) {
747 get_paca()->context
.user_psize
=
748 mm
->context
.user_psize
;
749 slb_flush_and_rebolt();
751 } else if (get_paca()->vmalloc_sllp
!=
752 mmu_psize_defs
[mmu_vmalloc_psize
].sllp
) {
753 get_paca()->vmalloc_sllp
=
754 mmu_psize_defs
[mmu_vmalloc_psize
].sllp
;
755 slb_vmalloc_update();
757 #endif /* CONFIG_PPC_64K_PAGES */
759 #ifdef CONFIG_PPC_HAS_HASH_64K
760 if (psize
== MMU_PAGE_64K
)
761 rc
= __hash_page_64K(ea
, access
, vsid
, ptep
, trap
, local
);
763 #endif /* CONFIG_PPC_HAS_HASH_64K */
764 rc
= __hash_page_4K(ea
, access
, vsid
, ptep
, trap
, local
);
766 #ifndef CONFIG_PPC_64K_PAGES
767 DBG_LOW(" o-pte: %016lx\n", pte_val(*ptep
));
769 DBG_LOW(" o-pte: %016lx %016lx\n", pte_val(*ptep
),
770 pte_val(*(ptep
+ PTRS_PER_PTE
)));
772 DBG_LOW(" -> rc=%d\n", rc
);
775 EXPORT_SYMBOL_GPL(hash_page
);
777 void hash_preload(struct mm_struct
*mm
, unsigned long ea
,
778 unsigned long access
, unsigned long trap
)
787 BUG_ON(REGION_ID(ea
) != USER_REGION_ID
);
789 #ifdef CONFIG_PPC_MM_SLICES
790 /* We only prefault standard pages for now */
791 if (unlikely(get_slice_psize(mm
, ea
) != mm
->context
.user_psize
))
795 DBG_LOW("hash_preload(mm=%p, mm->pgdir=%p, ea=%016lx, access=%lx,"
796 " trap=%lx\n", mm
, mm
->pgd
, ea
, access
, trap
);
798 /* Get Linux PTE if available */
802 ptep
= find_linux_pte(pgdir
, ea
);
806 #ifdef CONFIG_PPC_64K_PAGES
807 /* If either _PAGE_4K_PFN or _PAGE_NO_CACHE is set (and we are on
808 * a 64K kernel), then we don't preload, hash_page() will take
809 * care of it once we actually try to access the page.
810 * That way we don't have to duplicate all of the logic for segment
811 * page size demotion here
813 if (pte_val(*ptep
) & (_PAGE_4K_PFN
| _PAGE_NO_CACHE
))
815 #endif /* CONFIG_PPC_64K_PAGES */
818 vsid
= get_vsid(mm
->context
.id
, ea
);
820 /* Hash doesn't like irqs */
821 local_irq_save(flags
);
823 /* Is that local to this CPU ? */
824 mask
= cpumask_of_cpu(smp_processor_id());
825 if (cpus_equal(mm
->cpu_vm_mask
, mask
))
829 #ifdef CONFIG_PPC_HAS_HASH_64K
830 if (mm
->context
.user_psize
== MMU_PAGE_64K
)
831 __hash_page_64K(ea
, access
, vsid
, ptep
, trap
, local
);
833 #endif /* CONFIG_PPC_HAS_HASH_64K */
834 __hash_page_4K(ea
, access
, vsid
, ptep
, trap
, local
);
836 local_irq_restore(flags
);
839 void flush_hash_page(unsigned long va
, real_pte_t pte
, int psize
, int local
)
841 unsigned long hash
, index
, shift
, hidx
, slot
;
843 DBG_LOW("flush_hash_page(va=%016x)\n", va
);
844 pte_iterate_hashed_subpages(pte
, psize
, va
, index
, shift
) {
845 hash
= hpt_hash(va
, shift
);
846 hidx
= __rpte_to_hidx(pte
, index
);
847 if (hidx
& _PTEIDX_SECONDARY
)
849 slot
= (hash
& htab_hash_mask
) * HPTES_PER_GROUP
;
850 slot
+= hidx
& _PTEIDX_GROUP_IX
;
851 DBG_LOW(" sub %d: hash=%x, hidx=%x\n", index
, slot
, hidx
);
852 ppc_md
.hpte_invalidate(slot
, va
, psize
, local
);
853 } pte_iterate_hashed_end();
856 void flush_hash_range(unsigned long number
, int local
)
858 if (ppc_md
.flush_hash_range
)
859 ppc_md
.flush_hash_range(number
, local
);
862 struct ppc64_tlb_batch
*batch
=
863 &__get_cpu_var(ppc64_tlb_batch
);
865 for (i
= 0; i
< number
; i
++)
866 flush_hash_page(batch
->vaddr
[i
], batch
->pte
[i
],
867 batch
->psize
, local
);
872 * low_hash_fault is called when we the low level hash code failed
873 * to instert a PTE due to an hypervisor error
875 void low_hash_fault(struct pt_regs
*regs
, unsigned long address
)
877 if (user_mode(regs
)) {
880 info
.si_signo
= SIGBUS
;
882 info
.si_code
= BUS_ADRERR
;
883 info
.si_addr
= (void __user
*)address
;
884 force_sig_info(SIGBUS
, &info
, current
);
887 bad_page_fault(regs
, address
, SIGBUS
);
890 #ifdef CONFIG_DEBUG_PAGEALLOC
891 static void kernel_map_linear_page(unsigned long vaddr
, unsigned long lmi
)
893 unsigned long hash
, hpteg
, vsid
= get_kernel_vsid(vaddr
);
894 unsigned long va
= (vsid
<< 28) | (vaddr
& 0x0fffffff);
895 unsigned long mode
= _PAGE_ACCESSED
| _PAGE_DIRTY
|
896 _PAGE_COHERENT
| PP_RWXX
| HPTE_R_N
;
899 hash
= hpt_hash(va
, PAGE_SHIFT
);
900 hpteg
= ((hash
& htab_hash_mask
) * HPTES_PER_GROUP
);
902 ret
= ppc_md
.hpte_insert(hpteg
, va
, __pa(vaddr
),
903 mode
, HPTE_V_BOLTED
, mmu_linear_psize
);
905 spin_lock(&linear_map_hash_lock
);
906 BUG_ON(linear_map_hash_slots
[lmi
] & 0x80);
907 linear_map_hash_slots
[lmi
] = ret
| 0x80;
908 spin_unlock(&linear_map_hash_lock
);
911 static void kernel_unmap_linear_page(unsigned long vaddr
, unsigned long lmi
)
913 unsigned long hash
, hidx
, slot
, vsid
= get_kernel_vsid(vaddr
);
914 unsigned long va
= (vsid
<< 28) | (vaddr
& 0x0fffffff);
916 hash
= hpt_hash(va
, PAGE_SHIFT
);
917 spin_lock(&linear_map_hash_lock
);
918 BUG_ON(!(linear_map_hash_slots
[lmi
] & 0x80));
919 hidx
= linear_map_hash_slots
[lmi
] & 0x7f;
920 linear_map_hash_slots
[lmi
] = 0;
921 spin_unlock(&linear_map_hash_lock
);
922 if (hidx
& _PTEIDX_SECONDARY
)
924 slot
= (hash
& htab_hash_mask
) * HPTES_PER_GROUP
;
925 slot
+= hidx
& _PTEIDX_GROUP_IX
;
926 ppc_md
.hpte_invalidate(slot
, va
, mmu_linear_psize
, 0);
929 void kernel_map_pages(struct page
*page
, int numpages
, int enable
)
931 unsigned long flags
, vaddr
, lmi
;
934 local_irq_save(flags
);
935 for (i
= 0; i
< numpages
; i
++, page
++) {
936 vaddr
= (unsigned long)page_address(page
);
937 lmi
= __pa(vaddr
) >> PAGE_SHIFT
;
938 if (lmi
>= linear_map_hash_count
)
941 kernel_map_linear_page(vaddr
, lmi
);
943 kernel_unmap_linear_page(vaddr
, lmi
);
945 local_irq_restore(flags
);
947 #endif /* CONFIG_DEBUG_PAGEALLOC */