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>
34 #include <linux/lmb.h>
36 #include <asm/processor.h>
37 #include <asm/pgtable.h>
39 #include <asm/mmu_context.h>
41 #include <asm/types.h>
42 #include <asm/system.h>
43 #include <asm/uaccess.h>
44 #include <asm/machdep.h>
46 #include <asm/abs_addr.h>
47 #include <asm/tlbflush.h>
51 #include <asm/cacheflush.h>
52 #include <asm/cputable.h>
53 #include <asm/sections.h>
58 #define DBG(fmt...) udbg_printf(fmt)
64 #define DBG_LOW(fmt...) udbg_printf(fmt)
66 #define DBG_LOW(fmt...)
73 * Note: pte --> Linux PTE
74 * HPTE --> PowerPC Hashed Page Table Entry
77 * htab_initialize is called with the MMU off (of course), but
78 * the kernel has been copied down to zero so it can directly
79 * reference global data. At this point it is very difficult
80 * to print debug info.
85 extern unsigned long dart_tablebase;
86 #endif /* CONFIG_U3_DART */
88 static unsigned long _SDR1;
89 struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT];
91 struct hash_pte *htab_address;
92 unsigned long htab_size_bytes;
93 unsigned long htab_hash_mask;
94 int mmu_linear_psize = MMU_PAGE_4K;
95 int mmu_virtual_psize = MMU_PAGE_4K;
96 int mmu_vmalloc_psize = MMU_PAGE_4K;
97 int mmu_io_psize = MMU_PAGE_4K;
98 int mmu_kernel_ssize = MMU_SEGSIZE_256M;
99 int mmu_highuser_ssize = MMU_SEGSIZE_256M;
100 u16 mmu_slb_size = 64;
101 #ifdef CONFIG_HUGETLB_PAGE
102 int mmu_huge_psize = MMU_PAGE_16M;
103 unsigned int HPAGE_SHIFT;
105 #ifdef CONFIG_PPC_64K_PAGES
106 int mmu_ci_restrictions;
108 #ifdef CONFIG_DEBUG_PAGEALLOC
109 static u8 *linear_map_hash_slots;
110 static unsigned long linear_map_hash_count;
111 static DEFINE_SPINLOCK(linear_map_hash_lock);
112 #endif /* CONFIG_DEBUG_PAGEALLOC */
114 /* There are definitions of page sizes arrays to be used when none
115 * is provided by the firmware.
118 /* Pre-POWER4 CPUs (4k pages only)
120 struct mmu_psize_def mmu_psize_defaults_old[] = {
130 /* POWER4, GPUL, POWER5
132 * Support for 16Mb large pages
134 struct mmu_psize_def mmu_psize_defaults_gp[] = {
152 int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
153 unsigned long pstart, unsigned long mode,
154 int psize, int ssize)
156 unsigned long vaddr, paddr;
157 unsigned int step, shift;
158 unsigned long tmp_mode;
161 shift = mmu_psize_defs[psize].shift;
164 for (vaddr = vstart, paddr = pstart; vaddr < vend;
165 vaddr += step, paddr += step) {
166 unsigned long hash, hpteg;
167 unsigned long vsid = get_kernel_vsid(vaddr, ssize);
168 unsigned long va = hpt_va(vaddr, vsid, ssize);
172 /* Make non-kernel text non-executable */
173 if (!in_kernel_text(vaddr))
174 tmp_mode = mode | HPTE_R_N;
176 hash = hpt_hash(va, shift, ssize);
177 hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
179 DBG("htab_bolt_mapping: calling %p\n", ppc_md.hpte_insert);
181 BUG_ON(!ppc_md.hpte_insert);
182 ret = ppc_md.hpte_insert(hpteg, va, paddr,
183 tmp_mode, HPTE_V_BOLTED, psize, ssize);
187 #ifdef CONFIG_DEBUG_PAGEALLOC
188 if ((paddr >> PAGE_SHIFT) < linear_map_hash_count)
189 linear_map_hash_slots[paddr >> PAGE_SHIFT] = ret | 0x80;
190 #endif /* CONFIG_DEBUG_PAGEALLOC */
192 return ret < 0 ? ret : 0;
195 static int __init htab_dt_scan_seg_sizes(unsigned long node,
196 const char *uname, int depth,
199 char *type = of_get_flat_dt_prop(node, "device_type", NULL);
201 unsigned long size = 0;
203 /* We are scanning "cpu" nodes only */
204 if (type == NULL || strcmp(type, "cpu") != 0)
207 prop = (u32 *)of_get_flat_dt_prop(node, "ibm,processor-segment-sizes",
211 for (; size >= 4; size -= 4, ++prop) {
213 DBG("1T segment support detected\n");
214 cur_cpu_spec->cpu_features |= CPU_FTR_1T_SEGMENT;
218 cur_cpu_spec->cpu_features &= ~CPU_FTR_NO_SLBIE_B;
222 static void __init htab_init_seg_sizes(void)
224 of_scan_flat_dt(htab_dt_scan_seg_sizes, NULL);
227 static int __init htab_dt_scan_page_sizes(unsigned long node,
228 const char *uname, int depth,
231 char *type = of_get_flat_dt_prop(node, "device_type", NULL);
233 unsigned long size = 0;
235 /* We are scanning "cpu" nodes only */
236 if (type == NULL || strcmp(type, "cpu") != 0)
239 prop = (u32 *)of_get_flat_dt_prop(node,
240 "ibm,segment-page-sizes", &size);
242 DBG("Page sizes from device-tree:\n");
244 cur_cpu_spec->cpu_features &= ~(CPU_FTR_16M_PAGE);
246 unsigned int shift = prop[0];
247 unsigned int slbenc = prop[1];
248 unsigned int lpnum = prop[2];
249 unsigned int lpenc = 0;
250 struct mmu_psize_def *def;
253 size -= 3; prop += 3;
254 while(size > 0 && lpnum) {
255 if (prop[0] == shift)
257 prop += 2; size -= 2;
272 cur_cpu_spec->cpu_features |= CPU_FTR_16M_PAGE;
280 def = &mmu_psize_defs[idx];
285 def->avpnm = (1 << (shift - 23)) - 1;
288 /* We don't know for sure what's up with tlbiel, so
289 * for now we only set it for 4K and 64K pages
291 if (idx == MMU_PAGE_4K || idx == MMU_PAGE_64K)
296 DBG(" %d: shift=%02x, sllp=%04x, avpnm=%08x, "
297 "tlbiel=%d, penc=%d\n",
298 idx, shift, def->sllp, def->avpnm, def->tlbiel,
306 static void __init htab_init_page_sizes(void)
310 /* Default to 4K pages only */
311 memcpy(mmu_psize_defs, mmu_psize_defaults_old,
312 sizeof(mmu_psize_defaults_old));
315 * Try to find the available page sizes in the device-tree
317 rc = of_scan_flat_dt(htab_dt_scan_page_sizes, NULL);
318 if (rc != 0) /* Found */
322 * Not in the device-tree, let's fallback on known size
323 * list for 16M capable GP & GR
325 if (cpu_has_feature(CPU_FTR_16M_PAGE))
326 memcpy(mmu_psize_defs, mmu_psize_defaults_gp,
327 sizeof(mmu_psize_defaults_gp));
329 #ifndef CONFIG_DEBUG_PAGEALLOC
331 * Pick a size for the linear mapping. Currently, we only support
332 * 16M, 1M and 4K which is the default
334 if (mmu_psize_defs[MMU_PAGE_16M].shift)
335 mmu_linear_psize = MMU_PAGE_16M;
336 else if (mmu_psize_defs[MMU_PAGE_1M].shift)
337 mmu_linear_psize = MMU_PAGE_1M;
338 #endif /* CONFIG_DEBUG_PAGEALLOC */
340 #ifdef CONFIG_PPC_64K_PAGES
342 * Pick a size for the ordinary pages. Default is 4K, we support
343 * 64K for user mappings and vmalloc if supported by the processor.
344 * We only use 64k for ioremap if the processor
345 * (and firmware) support cache-inhibited large pages.
346 * If not, we use 4k and set mmu_ci_restrictions so that
347 * hash_page knows to switch processes that use cache-inhibited
348 * mappings to 4k pages.
350 if (mmu_psize_defs[MMU_PAGE_64K].shift) {
351 mmu_virtual_psize = MMU_PAGE_64K;
352 mmu_vmalloc_psize = MMU_PAGE_64K;
353 if (mmu_linear_psize == MMU_PAGE_4K)
354 mmu_linear_psize = MMU_PAGE_64K;
355 if (cpu_has_feature(CPU_FTR_CI_LARGE_PAGE))
356 mmu_io_psize = MMU_PAGE_64K;
358 mmu_ci_restrictions = 1;
360 #endif /* CONFIG_PPC_64K_PAGES */
362 printk(KERN_DEBUG "Page orders: linear mapping = %d, "
363 "virtual = %d, io = %d\n",
364 mmu_psize_defs[mmu_linear_psize].shift,
365 mmu_psize_defs[mmu_virtual_psize].shift,
366 mmu_psize_defs[mmu_io_psize].shift);
368 #ifdef CONFIG_HUGETLB_PAGE
369 /* Init large page size. Currently, we pick 16M or 1M depending
370 * on what is available
372 if (mmu_psize_defs[MMU_PAGE_16M].shift)
373 set_huge_psize(MMU_PAGE_16M);
374 /* With 4k/4level pagetables, we can't (for now) cope with a
375 * huge page size < PMD_SIZE */
376 else if (mmu_psize_defs[MMU_PAGE_1M].shift)
377 set_huge_psize(MMU_PAGE_1M);
378 #endif /* CONFIG_HUGETLB_PAGE */
381 static int __init htab_dt_scan_pftsize(unsigned long node,
382 const char *uname, int depth,
385 char *type = of_get_flat_dt_prop(node, "device_type", NULL);
388 /* We are scanning "cpu" nodes only */
389 if (type == NULL || strcmp(type, "cpu") != 0)
392 prop = (u32 *)of_get_flat_dt_prop(node, "ibm,pft-size", NULL);
394 /* pft_size[0] is the NUMA CEC cookie */
395 ppc64_pft_size = prop[1];
401 static unsigned long __init htab_get_table_size(void)
403 unsigned long mem_size, rnd_mem_size, pteg_count;
405 /* If hash size isn't already provided by the platform, we try to
406 * retrieve it from the device-tree. If it's not there neither, we
407 * calculate it now based on the total RAM size
409 if (ppc64_pft_size == 0)
410 of_scan_flat_dt(htab_dt_scan_pftsize, NULL);
412 return 1UL << ppc64_pft_size;
414 /* round mem_size up to next power of 2 */
415 mem_size = lmb_phys_mem_size();
416 rnd_mem_size = 1UL << __ilog2(mem_size);
417 if (rnd_mem_size < mem_size)
421 pteg_count = max(rnd_mem_size >> (12 + 1), 1UL << 11);
423 return pteg_count << 7;
426 #ifdef CONFIG_MEMORY_HOTPLUG
427 void create_section_mapping(unsigned long start, unsigned long end)
429 BUG_ON(htab_bolt_mapping(start, end, __pa(start),
430 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_COHERENT | PP_RWXX,
431 mmu_linear_psize, mmu_kernel_ssize));
433 #endif /* CONFIG_MEMORY_HOTPLUG */
435 static inline void make_bl(unsigned int *insn_addr, void *func)
437 unsigned long funcp = *((unsigned long *)func);
438 int offset = funcp - (unsigned long)insn_addr;
440 *insn_addr = (unsigned int)(0x48000001 | (offset & 0x03fffffc));
441 flush_icache_range((unsigned long)insn_addr, 4+
442 (unsigned long)insn_addr);
445 static void __init htab_finish_init(void)
447 extern unsigned int *htab_call_hpte_insert1;
448 extern unsigned int *htab_call_hpte_insert2;
449 extern unsigned int *htab_call_hpte_remove;
450 extern unsigned int *htab_call_hpte_updatepp;
452 #ifdef CONFIG_PPC_HAS_HASH_64K
453 extern unsigned int *ht64_call_hpte_insert1;
454 extern unsigned int *ht64_call_hpte_insert2;
455 extern unsigned int *ht64_call_hpte_remove;
456 extern unsigned int *ht64_call_hpte_updatepp;
458 make_bl(ht64_call_hpte_insert1, ppc_md.hpte_insert);
459 make_bl(ht64_call_hpte_insert2, ppc_md.hpte_insert);
460 make_bl(ht64_call_hpte_remove, ppc_md.hpte_remove);
461 make_bl(ht64_call_hpte_updatepp, ppc_md.hpte_updatepp);
462 #endif /* CONFIG_PPC_HAS_HASH_64K */
464 make_bl(htab_call_hpte_insert1, ppc_md.hpte_insert);
465 make_bl(htab_call_hpte_insert2, ppc_md.hpte_insert);
466 make_bl(htab_call_hpte_remove, ppc_md.hpte_remove);
467 make_bl(htab_call_hpte_updatepp, ppc_md.hpte_updatepp);
470 void __init htab_initialize(void)
473 unsigned long pteg_count;
474 unsigned long mode_rw;
475 unsigned long base = 0, size = 0, limit;
478 extern unsigned long tce_alloc_start, tce_alloc_end;
480 DBG(" -> htab_initialize()\n");
482 /* Initialize segment sizes */
483 htab_init_seg_sizes();
485 /* Initialize page sizes */
486 htab_init_page_sizes();
488 if (cpu_has_feature(CPU_FTR_1T_SEGMENT)) {
489 mmu_kernel_ssize = MMU_SEGSIZE_1T;
490 mmu_highuser_ssize = MMU_SEGSIZE_1T;
491 printk(KERN_INFO "Using 1TB segments\n");
495 * Calculate the required size of the htab. We want the number of
496 * PTEGs to equal one half the number of real pages.
498 htab_size_bytes = htab_get_table_size();
499 pteg_count = htab_size_bytes >> 7;
501 htab_hash_mask = pteg_count - 1;
503 if (firmware_has_feature(FW_FEATURE_LPAR)) {
504 /* Using a hypervisor which owns the htab */
508 /* Find storage for the HPT. Must be contiguous in
509 * the absolute address space. On cell we want it to be
510 * in the first 1 Gig.
512 if (machine_is(cell))
517 table = lmb_alloc_base(htab_size_bytes, htab_size_bytes, limit);
519 DBG("Hash table allocated at %lx, size: %lx\n", table,
522 htab_address = abs_to_virt(table);
524 /* htab absolute addr + encoded htabsize */
525 _SDR1 = table + __ilog2(pteg_count) - 11;
527 /* Initialize the HPT with no entries */
528 memset((void *)table, 0, htab_size_bytes);
531 mtspr(SPRN_SDR1, _SDR1);
534 mode_rw = _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_COHERENT | PP_RWXX;
536 #ifdef CONFIG_DEBUG_PAGEALLOC
537 linear_map_hash_count = lmb_end_of_DRAM() >> PAGE_SHIFT;
538 linear_map_hash_slots = __va(lmb_alloc_base(linear_map_hash_count,
540 memset(linear_map_hash_slots, 0, linear_map_hash_count);
541 #endif /* CONFIG_DEBUG_PAGEALLOC */
543 /* On U3 based machines, we need to reserve the DART area and
544 * _NOT_ map it to avoid cache paradoxes as it's remapped non
548 /* create bolted the linear mapping in the hash table */
549 for (i=0; i < lmb.memory.cnt; i++) {
550 base = (unsigned long)__va(lmb.memory.region[i].base);
551 size = lmb.memory.region[i].size;
553 DBG("creating mapping for region: %lx : %lx\n", base, size);
555 #ifdef CONFIG_U3_DART
556 /* Do not map the DART space. Fortunately, it will be aligned
557 * in such a way that it will not cross two lmb regions and
558 * will fit within a single 16Mb page.
559 * The DART space is assumed to be a full 16Mb region even if
560 * we only use 2Mb of that space. We will use more of it later
561 * for AGP GART. We have to use a full 16Mb large page.
563 DBG("DART base: %lx\n", dart_tablebase);
565 if (dart_tablebase != 0 && dart_tablebase >= base
566 && dart_tablebase < (base + size)) {
567 unsigned long dart_table_end = dart_tablebase + 16 * MB;
568 if (base != dart_tablebase)
569 BUG_ON(htab_bolt_mapping(base, dart_tablebase,
573 if ((base + size) > dart_table_end)
574 BUG_ON(htab_bolt_mapping(dart_tablebase+16*MB,
576 __pa(dart_table_end),
582 #endif /* CONFIG_U3_DART */
583 BUG_ON(htab_bolt_mapping(base, base + size, __pa(base),
584 mode_rw, mmu_linear_psize, mmu_kernel_ssize));
588 * If we have a memory_limit and we've allocated TCEs then we need to
589 * explicitly map the TCE area at the top of RAM. We also cope with the
590 * case that the TCEs start below memory_limit.
591 * tce_alloc_start/end are 16MB aligned so the mapping should work
592 * for either 4K or 16MB pages.
594 if (tce_alloc_start) {
595 tce_alloc_start = (unsigned long)__va(tce_alloc_start);
596 tce_alloc_end = (unsigned long)__va(tce_alloc_end);
598 if (base + size >= tce_alloc_start)
599 tce_alloc_start = base + size + 1;
601 BUG_ON(htab_bolt_mapping(tce_alloc_start, tce_alloc_end,
602 __pa(tce_alloc_start), mode_rw,
603 mmu_linear_psize, mmu_kernel_ssize));
608 DBG(" <- htab_initialize()\n");
613 void htab_initialize_secondary(void)
615 if (!firmware_has_feature(FW_FEATURE_LPAR))
616 mtspr(SPRN_SDR1, _SDR1);
620 * Called by asm hashtable.S for doing lazy icache flush
622 unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap)
626 if (!pfn_valid(pte_pfn(pte)))
629 page = pte_page(pte);
632 if (!test_bit(PG_arch_1, &page->flags) && !PageReserved(page)) {
634 __flush_dcache_icache(page_address(page));
635 set_bit(PG_arch_1, &page->flags);
643 * Demote a segment to using 4k pages.
644 * For now this makes the whole process use 4k pages.
646 #ifdef CONFIG_PPC_64K_PAGES
647 void demote_segment_4k(struct mm_struct *mm, unsigned long addr)
649 if (mm->context.user_psize == MMU_PAGE_4K)
651 slice_set_user_psize(mm, MMU_PAGE_4K);
652 #ifdef CONFIG_SPU_BASE
653 spu_flush_all_slbs(mm);
655 if (get_paca()->context.user_psize != MMU_PAGE_4K) {
656 get_paca()->context = mm->context;
657 slb_flush_and_rebolt();
660 #endif /* CONFIG_PPC_64K_PAGES */
662 #ifdef CONFIG_PPC_SUBPAGE_PROT
664 * This looks up a 2-bit protection code for a 4k subpage of a 64k page.
665 * Userspace sets the subpage permissions using the subpage_prot system call.
667 * Result is 0: full permissions, _PAGE_RW: read-only,
668 * _PAGE_USER or _PAGE_USER|_PAGE_RW: no access.
670 static int subpage_protection(pgd_t *pgdir, unsigned long ea)
672 struct subpage_prot_table *spt = pgd_subpage_prot(pgdir);
676 if (ea >= spt->maxaddr)
678 if (ea < 0x100000000) {
679 /* addresses below 4GB use spt->low_prot */
680 sbpm = spt->low_prot;
682 sbpm = spt->protptrs[ea >> SBP_L3_SHIFT];
686 sbpp = sbpm[(ea >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1)];
689 spp = sbpp[(ea >> PAGE_SHIFT) & (SBP_L1_COUNT - 1)];
691 /* extract 2-bit bitfield for this 4k subpage */
692 spp >>= 30 - 2 * ((ea >> 12) & 0xf);
694 /* turn 0,1,2,3 into combination of _PAGE_USER and _PAGE_RW */
695 spp = ((spp & 2) ? _PAGE_USER : 0) | ((spp & 1) ? _PAGE_RW : 0);
699 #else /* CONFIG_PPC_SUBPAGE_PROT */
700 static inline int subpage_protection(pgd_t *pgdir, unsigned long ea)
708 * 1 - normal page fault
709 * -1 - critical hash insertion error
710 * -2 - access not permitted by subpage protection mechanism
712 int hash_page(unsigned long ea, unsigned long access, unsigned long trap)
716 struct mm_struct *mm;
719 int rc, user_region = 0, local = 0;
722 DBG_LOW("hash_page(ea=%016lx, access=%lx, trap=%lx\n",
725 if ((ea & ~REGION_MASK) >= PGTABLE_RANGE) {
726 DBG_LOW(" out of pgtable range !\n");
730 /* Get region & vsid */
731 switch (REGION_ID(ea)) {
736 DBG_LOW(" user region with no mm !\n");
739 #ifdef CONFIG_PPC_MM_SLICES
740 psize = get_slice_psize(mm, ea);
742 psize = mm->context.user_psize;
744 ssize = user_segment_size(ea);
745 vsid = get_vsid(mm->context.id, ea, ssize);
747 case VMALLOC_REGION_ID:
749 vsid = get_kernel_vsid(ea, mmu_kernel_ssize);
750 if (ea < VMALLOC_END)
751 psize = mmu_vmalloc_psize;
753 psize = mmu_io_psize;
754 ssize = mmu_kernel_ssize;
758 * Send the problem up to do_page_fault
762 DBG_LOW(" mm=%p, mm->pgdir=%p, vsid=%016lx\n", mm, mm->pgd, vsid);
769 /* Check CPU locality */
770 tmp = cpumask_of_cpu(smp_processor_id());
771 if (user_region && cpus_equal(mm->cpu_vm_mask, tmp))
774 #ifdef CONFIG_HUGETLB_PAGE
775 /* Handle hugepage regions */
776 if (HPAGE_SHIFT && psize == mmu_huge_psize) {
777 DBG_LOW(" -> huge page !\n");
778 return hash_huge_page(mm, access, ea, vsid, local, trap);
780 #endif /* CONFIG_HUGETLB_PAGE */
782 #ifndef CONFIG_PPC_64K_PAGES
783 /* If we use 4K pages and our psize is not 4K, then we are hitting
784 * a special driver mapping, we need to align the address before
787 if (psize != MMU_PAGE_4K)
788 ea &= ~((1ul << mmu_psize_defs[psize].shift) - 1);
789 #endif /* CONFIG_PPC_64K_PAGES */
791 /* Get PTE and page size from page tables */
792 ptep = find_linux_pte(pgdir, ea);
793 if (ptep == NULL || !pte_present(*ptep)) {
794 DBG_LOW(" no PTE !\n");
798 #ifndef CONFIG_PPC_64K_PAGES
799 DBG_LOW(" i-pte: %016lx\n", pte_val(*ptep));
801 DBG_LOW(" i-pte: %016lx %016lx\n", pte_val(*ptep),
802 pte_val(*(ptep + PTRS_PER_PTE)));
804 /* Pre-check access permissions (will be re-checked atomically
805 * in __hash_page_XX but this pre-check is a fast path
807 if (access & ~pte_val(*ptep)) {
808 DBG_LOW(" no access !\n");
812 /* Do actual hashing */
813 #ifdef CONFIG_PPC_64K_PAGES
814 /* If _PAGE_4K_PFN is set, make sure this is a 4k segment */
815 if (pte_val(*ptep) & _PAGE_4K_PFN) {
816 demote_segment_4k(mm, ea);
820 /* If this PTE is non-cacheable and we have restrictions on
821 * using non cacheable large pages, then we switch to 4k
823 if (mmu_ci_restrictions && psize == MMU_PAGE_64K &&
824 (pte_val(*ptep) & _PAGE_NO_CACHE)) {
826 demote_segment_4k(mm, ea);
828 } else if (ea < VMALLOC_END) {
830 * some driver did a non-cacheable mapping
831 * in vmalloc space, so switch vmalloc
834 printk(KERN_ALERT "Reducing vmalloc segment "
835 "to 4kB pages because of "
836 "non-cacheable mapping\n");
837 psize = mmu_vmalloc_psize = MMU_PAGE_4K;
838 #ifdef CONFIG_SPU_BASE
839 spu_flush_all_slbs(mm);
844 if (psize != get_paca()->context.user_psize) {
845 get_paca()->context = mm->context;
846 slb_flush_and_rebolt();
848 } else if (get_paca()->vmalloc_sllp !=
849 mmu_psize_defs[mmu_vmalloc_psize].sllp) {
850 get_paca()->vmalloc_sllp =
851 mmu_psize_defs[mmu_vmalloc_psize].sllp;
852 slb_vmalloc_update();
854 #endif /* CONFIG_PPC_64K_PAGES */
856 #ifdef CONFIG_PPC_HAS_HASH_64K
857 if (psize == MMU_PAGE_64K)
858 rc = __hash_page_64K(ea, access, vsid, ptep, trap, local, ssize);
860 #endif /* CONFIG_PPC_HAS_HASH_64K */
862 int spp = subpage_protection(pgdir, ea);
866 rc = __hash_page_4K(ea, access, vsid, ptep, trap,
870 #ifndef CONFIG_PPC_64K_PAGES
871 DBG_LOW(" o-pte: %016lx\n", pte_val(*ptep));
873 DBG_LOW(" o-pte: %016lx %016lx\n", pte_val(*ptep),
874 pte_val(*(ptep + PTRS_PER_PTE)));
876 DBG_LOW(" -> rc=%d\n", rc);
879 EXPORT_SYMBOL_GPL(hash_page);
881 void hash_preload(struct mm_struct *mm, unsigned long ea,
882 unsigned long access, unsigned long trap)
892 BUG_ON(REGION_ID(ea) != USER_REGION_ID);
894 #ifdef CONFIG_PPC_MM_SLICES
895 /* We only prefault standard pages for now */
896 if (unlikely(get_slice_psize(mm, ea) != mm->context.user_psize))
900 DBG_LOW("hash_preload(mm=%p, mm->pgdir=%p, ea=%016lx, access=%lx,"
901 " trap=%lx\n", mm, mm->pgd, ea, access, trap);
903 /* Get Linux PTE if available */
907 ptep = find_linux_pte(pgdir, ea);
911 #ifdef CONFIG_PPC_64K_PAGES
912 /* If either _PAGE_4K_PFN or _PAGE_NO_CACHE is set (and we are on
913 * a 64K kernel), then we don't preload, hash_page() will take
914 * care of it once we actually try to access the page.
915 * That way we don't have to duplicate all of the logic for segment
916 * page size demotion here
918 if (pte_val(*ptep) & (_PAGE_4K_PFN | _PAGE_NO_CACHE))
920 #endif /* CONFIG_PPC_64K_PAGES */
923 ssize = user_segment_size(ea);
924 vsid = get_vsid(mm->context.id, ea, ssize);
926 /* Hash doesn't like irqs */
927 local_irq_save(flags);
929 /* Is that local to this CPU ? */
930 mask = cpumask_of_cpu(smp_processor_id());
931 if (cpus_equal(mm->cpu_vm_mask, mask))
935 #ifdef CONFIG_PPC_HAS_HASH_64K
936 if (mm->context.user_psize == MMU_PAGE_64K)
937 __hash_page_64K(ea, access, vsid, ptep, trap, local, ssize);
939 #endif /* CONFIG_PPC_HAS_HASH_64K */
940 __hash_page_4K(ea, access, vsid, ptep, trap, local, ssize,
941 subpage_protection(pgdir, ea));
943 local_irq_restore(flags);
946 /* WARNING: This is called from hash_low_64.S, if you change this prototype,
947 * do not forget to update the assembly call site !
949 void flush_hash_page(unsigned long va, real_pte_t pte, int psize, int ssize,
952 unsigned long hash, index, shift, hidx, slot;
954 DBG_LOW("flush_hash_page(va=%016x)\n", va);
955 pte_iterate_hashed_subpages(pte, psize, va, index, shift) {
956 hash = hpt_hash(va, shift, ssize);
957 hidx = __rpte_to_hidx(pte, index);
958 if (hidx & _PTEIDX_SECONDARY)
960 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
961 slot += hidx & _PTEIDX_GROUP_IX;
962 DBG_LOW(" sub %d: hash=%x, hidx=%x\n", index, slot, hidx);
963 ppc_md.hpte_invalidate(slot, va, psize, ssize, local);
964 } pte_iterate_hashed_end();
967 void flush_hash_range(unsigned long number, int local)
969 if (ppc_md.flush_hash_range)
970 ppc_md.flush_hash_range(number, local);
973 struct ppc64_tlb_batch *batch =
974 &__get_cpu_var(ppc64_tlb_batch);
976 for (i = 0; i < number; i++)
977 flush_hash_page(batch->vaddr[i], batch->pte[i],
978 batch->psize, batch->ssize, local);
983 * low_hash_fault is called when we the low level hash code failed
984 * to instert a PTE due to an hypervisor error
986 void low_hash_fault(struct pt_regs *regs, unsigned long address, int rc)
988 if (user_mode(regs)) {
989 #ifdef CONFIG_PPC_SUBPAGE_PROT
991 _exception(SIGSEGV, regs, SEGV_ACCERR, address);
994 _exception(SIGBUS, regs, BUS_ADRERR, address);
996 bad_page_fault(regs, address, SIGBUS);
999 #ifdef CONFIG_DEBUG_PAGEALLOC
1000 static void kernel_map_linear_page(unsigned long vaddr, unsigned long lmi)
1002 unsigned long hash, hpteg;
1003 unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize);
1004 unsigned long va = hpt_va(vaddr, vsid, mmu_kernel_ssize);
1005 unsigned long mode = _PAGE_ACCESSED | _PAGE_DIRTY |
1006 _PAGE_COHERENT | PP_RWXX | HPTE_R_N;
1009 hash = hpt_hash(va, PAGE_SHIFT, mmu_kernel_ssize);
1010 hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
1012 ret = ppc_md.hpte_insert(hpteg, va, __pa(vaddr),
1013 mode, HPTE_V_BOLTED,
1014 mmu_linear_psize, mmu_kernel_ssize);
1016 spin_lock(&linear_map_hash_lock);
1017 BUG_ON(linear_map_hash_slots[lmi] & 0x80);
1018 linear_map_hash_slots[lmi] = ret | 0x80;
1019 spin_unlock(&linear_map_hash_lock);
1022 static void kernel_unmap_linear_page(unsigned long vaddr, unsigned long lmi)
1024 unsigned long hash, hidx, slot;
1025 unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize);
1026 unsigned long va = hpt_va(vaddr, vsid, mmu_kernel_ssize);
1028 hash = hpt_hash(va, PAGE_SHIFT, mmu_kernel_ssize);
1029 spin_lock(&linear_map_hash_lock);
1030 BUG_ON(!(linear_map_hash_slots[lmi] & 0x80));
1031 hidx = linear_map_hash_slots[lmi] & 0x7f;
1032 linear_map_hash_slots[lmi] = 0;
1033 spin_unlock(&linear_map_hash_lock);
1034 if (hidx & _PTEIDX_SECONDARY)
1036 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1037 slot += hidx & _PTEIDX_GROUP_IX;
1038 ppc_md.hpte_invalidate(slot, va, mmu_linear_psize, mmu_kernel_ssize, 0);
1041 void kernel_map_pages(struct page *page, int numpages, int enable)
1043 unsigned long flags, vaddr, lmi;
1046 local_irq_save(flags);
1047 for (i = 0; i < numpages; i++, page++) {
1048 vaddr = (unsigned long)page_address(page);
1049 lmi = __pa(vaddr) >> PAGE_SHIFT;
1050 if (lmi >= linear_map_hash_count)
1053 kernel_map_linear_page(vaddr, lmi);
1055 kernel_unmap_linear_page(vaddr, lmi);
1057 local_irq_restore(flags);
1059 #endif /* CONFIG_DEBUG_PAGEALLOC */