2 * This file contains ioremap and related functions for 64-bit machines.
4 * Derived from arch/ppc64/mm/init.c
5 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
7 * Modifications by Paul Mackerras (PowerMac) (paulus@samba.org)
8 * and Cort Dougan (PReP) (cort@cs.nmt.edu)
9 * Copyright (C) 1996 Paul Mackerras
11 * Derived from "arch/i386/mm/init.c"
12 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
14 * Dave Engebretsen <engebret@us.ibm.com>
15 * Rework for PPC64 port.
17 * This program is free software; you can redistribute it and/or
18 * modify it under the terms of the GNU General Public License
19 * as published by the Free Software Foundation; either version
20 * 2 of the License, or (at your option) any later version.
24 #include <linux/signal.h>
25 #include <linux/sched.h>
26 #include <linux/kernel.h>
27 #include <linux/errno.h>
28 #include <linux/string.h>
29 #include <linux/export.h>
30 #include <linux/types.h>
31 #include <linux/mman.h>
33 #include <linux/swap.h>
34 #include <linux/stddef.h>
35 #include <linux/vmalloc.h>
36 #include <linux/bootmem.h>
37 #include <linux/memblock.h>
38 #include <linux/slab.h>
40 #include <asm/pgalloc.h>
44 #include <asm/mmu_context.h>
45 #include <asm/pgtable.h>
48 #include <asm/machdep.h>
50 #include <asm/processor.h>
51 #include <asm/cputable.h>
52 #include <asm/sections.h>
53 #include <asm/firmware.h>
57 #define CREATE_TRACE_POINTS
58 #include <trace/events/thp.h>
60 /* Some sanity checking */
61 #if TASK_SIZE_USER64 > PGTABLE_RANGE
62 #error TASK_SIZE_USER64 exceeds pagetable range
65 #ifdef CONFIG_PPC_STD_MMU_64
66 #if TASK_SIZE_USER64 > (1UL << (ESID_BITS + SID_SHIFT))
67 #error TASK_SIZE_USER64 exceeds user VSID range
71 unsigned long ioremap_bot = IOREMAP_BASE;
73 #ifdef CONFIG_PPC_MMU_NOHASH
74 static __ref void *early_alloc_pgtable(unsigned long size)
78 if (init_bootmem_done)
79 pt = __alloc_bootmem(size, size, __pa(MAX_DMA_ADDRESS));
81 pt = __va(memblock_alloc_base(size, size,
82 __pa(MAX_DMA_ADDRESS)));
87 #endif /* CONFIG_PPC_MMU_NOHASH */
90 * map_kernel_page currently only called by __ioremap
91 * map_kernel_page adds an entry to the ioremap page table
92 * and adds an entry to the HPT, possibly bolting it
94 int map_kernel_page(unsigned long ea, unsigned long pa, int flags)
101 if (slab_is_available()) {
102 pgdp = pgd_offset_k(ea);
103 pudp = pud_alloc(&init_mm, pgdp, ea);
106 pmdp = pmd_alloc(&init_mm, pudp, ea);
109 ptep = pte_alloc_kernel(pmdp, ea);
112 set_pte_at(&init_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT,
115 #ifdef CONFIG_PPC_MMU_NOHASH
116 /* Warning ! This will blow up if bootmem is not initialized
117 * which our ppc64 code is keen to do that, we'll need to
118 * fix it and/or be more careful
120 pgdp = pgd_offset_k(ea);
121 #ifdef PUD_TABLE_SIZE
122 if (pgd_none(*pgdp)) {
123 pudp = early_alloc_pgtable(PUD_TABLE_SIZE);
124 BUG_ON(pudp == NULL);
125 pgd_populate(&init_mm, pgdp, pudp);
127 #endif /* PUD_TABLE_SIZE */
128 pudp = pud_offset(pgdp, ea);
129 if (pud_none(*pudp)) {
130 pmdp = early_alloc_pgtable(PMD_TABLE_SIZE);
131 BUG_ON(pmdp == NULL);
132 pud_populate(&init_mm, pudp, pmdp);
134 pmdp = pmd_offset(pudp, ea);
135 if (!pmd_present(*pmdp)) {
136 ptep = early_alloc_pgtable(PAGE_SIZE);
137 BUG_ON(ptep == NULL);
138 pmd_populate_kernel(&init_mm, pmdp, ptep);
140 ptep = pte_offset_kernel(pmdp, ea);
141 set_pte_at(&init_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT,
143 #else /* CONFIG_PPC_MMU_NOHASH */
145 * If the mm subsystem is not fully up, we cannot create a
146 * linux page table entry for this mapping. Simply bolt an
147 * entry in the hardware page table.
150 if (htab_bolt_mapping(ea, ea + PAGE_SIZE, pa, flags,
151 mmu_io_psize, mmu_kernel_ssize)) {
152 printk(KERN_ERR "Failed to do bolted mapping IO "
153 "memory at %016lx !\n", pa);
156 #endif /* !CONFIG_PPC_MMU_NOHASH */
159 #ifdef CONFIG_PPC_BOOK3E_64
161 * With hardware tablewalk, a sync is needed to ensure that
162 * subsequent accesses see the PTE we just wrote. Unlike userspace
163 * mappings, we can't tolerate spurious faults, so make sure
164 * the new PTE will be seen the first time.
175 * __ioremap_at - Low level function to establish the page tables
178 void __iomem * __ioremap_at(phys_addr_t pa, void *ea, unsigned long size,
183 /* Make sure we have the base flags */
184 if ((flags & _PAGE_PRESENT) == 0)
185 flags |= pgprot_val(PAGE_KERNEL);
187 /* Non-cacheable page cannot be coherent */
188 if (flags & _PAGE_NO_CACHE)
189 flags &= ~_PAGE_COHERENT;
191 /* We don't support the 4K PFN hack with ioremap */
192 if (flags & _PAGE_4K_PFN)
195 WARN_ON(pa & ~PAGE_MASK);
196 WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
197 WARN_ON(size & ~PAGE_MASK);
199 for (i = 0; i < size; i += PAGE_SIZE)
200 if (map_kernel_page((unsigned long)ea+i, pa+i, flags))
203 return (void __iomem *)ea;
207 * __iounmap_from - Low level function to tear down the page tables
208 * for an IO mapping. This is used for mappings that
209 * are manipulated manually, like partial unmapping of
210 * PCI IOs or ISA space.
212 void __iounmap_at(void *ea, unsigned long size)
214 WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
215 WARN_ON(size & ~PAGE_MASK);
217 unmap_kernel_range((unsigned long)ea, size);
220 void __iomem * __ioremap_caller(phys_addr_t addr, unsigned long size,
221 unsigned long flags, void *caller)
223 phys_addr_t paligned;
227 * Choose an address to map it to.
228 * Once the imalloc system is running, we use it.
229 * Before that, we map using addresses going
230 * up from ioremap_bot. imalloc will use
231 * the addresses from ioremap_bot through
235 paligned = addr & PAGE_MASK;
236 size = PAGE_ALIGN(addr + size) - paligned;
238 if ((size == 0) || (paligned == 0))
242 struct vm_struct *area;
244 area = __get_vm_area_caller(size, VM_IOREMAP,
245 ioremap_bot, IOREMAP_END,
250 area->phys_addr = paligned;
251 ret = __ioremap_at(paligned, area->addr, size, flags);
255 ret = __ioremap_at(paligned, (void *)ioremap_bot, size, flags);
261 ret += addr & ~PAGE_MASK;
265 void __iomem * __ioremap(phys_addr_t addr, unsigned long size,
268 return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
271 void __iomem * ioremap(phys_addr_t addr, unsigned long size)
273 unsigned long flags = _PAGE_NO_CACHE | _PAGE_GUARDED;
274 void *caller = __builtin_return_address(0);
277 return ppc_md.ioremap(addr, size, flags, caller);
278 return __ioremap_caller(addr, size, flags, caller);
281 void __iomem * ioremap_wc(phys_addr_t addr, unsigned long size)
283 unsigned long flags = _PAGE_NO_CACHE;
284 void *caller = __builtin_return_address(0);
287 return ppc_md.ioremap(addr, size, flags, caller);
288 return __ioremap_caller(addr, size, flags, caller);
291 void __iomem * ioremap_prot(phys_addr_t addr, unsigned long size,
294 void *caller = __builtin_return_address(0);
296 /* writeable implies dirty for kernel addresses */
297 if (flags & _PAGE_RW)
298 flags |= _PAGE_DIRTY;
300 /* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */
301 flags &= ~(_PAGE_USER | _PAGE_EXEC);
304 /* _PAGE_USER contains _PAGE_BAP_SR on BookE using the new PTE format
305 * which means that we just cleared supervisor access... oops ;-) This
308 flags |= _PAGE_BAP_SR;
312 return ppc_md.ioremap(addr, size, flags, caller);
313 return __ioremap_caller(addr, size, flags, caller);
318 * Unmap an IO region and remove it from imalloc'd list.
319 * Access to IO memory should be serialized by driver.
321 void __iounmap(volatile void __iomem *token)
328 addr = (void *) ((unsigned long __force)
329 PCI_FIX_ADDR(token) & PAGE_MASK);
330 if ((unsigned long)addr < ioremap_bot) {
331 printk(KERN_WARNING "Attempt to iounmap early bolted mapping"
338 void iounmap(volatile void __iomem *token)
341 ppc_md.iounmap(token);
346 EXPORT_SYMBOL(ioremap);
347 EXPORT_SYMBOL(ioremap_wc);
348 EXPORT_SYMBOL(ioremap_prot);
349 EXPORT_SYMBOL(__ioremap);
350 EXPORT_SYMBOL(__ioremap_at);
351 EXPORT_SYMBOL(iounmap);
352 EXPORT_SYMBOL(__iounmap);
353 EXPORT_SYMBOL(__iounmap_at);
356 * For hugepage we have pfn in the pmd, we use PTE_RPN_SHIFT bits for flags
357 * For PTE page, we have a PTE_FRAG_SIZE (4K) aligned virtual address.
359 struct page *pmd_page(pmd_t pmd)
361 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
362 if (pmd_trans_huge(pmd))
363 return pfn_to_page(pmd_pfn(pmd));
365 return virt_to_page(pmd_page_vaddr(pmd));
368 #ifdef CONFIG_PPC_64K_PAGES
369 static pte_t *get_from_cache(struct mm_struct *mm)
371 void *pte_frag, *ret;
373 spin_lock(&mm->page_table_lock);
374 ret = mm->context.pte_frag;
376 pte_frag = ret + PTE_FRAG_SIZE;
378 * If we have taken up all the fragments mark PTE page NULL
380 if (((unsigned long)pte_frag & ~PAGE_MASK) == 0)
382 mm->context.pte_frag = pte_frag;
384 spin_unlock(&mm->page_table_lock);
388 static pte_t *__alloc_for_cache(struct mm_struct *mm, int kernel)
391 struct page *page = alloc_page(GFP_KERNEL | __GFP_NOTRACK |
392 __GFP_REPEAT | __GFP_ZERO);
395 if (!kernel && !pgtable_page_ctor(page)) {
400 ret = page_address(page);
401 spin_lock(&mm->page_table_lock);
403 * If we find pgtable_page set, we return
404 * the allocated page with single fragement
407 if (likely(!mm->context.pte_frag)) {
408 atomic_set(&page->_count, PTE_FRAG_NR);
409 mm->context.pte_frag = ret + PTE_FRAG_SIZE;
411 spin_unlock(&mm->page_table_lock);
416 pte_t *page_table_alloc(struct mm_struct *mm, unsigned long vmaddr, int kernel)
420 pte = get_from_cache(mm);
424 return __alloc_for_cache(mm, kernel);
427 void page_table_free(struct mm_struct *mm, unsigned long *table, int kernel)
429 struct page *page = virt_to_page(table);
430 if (put_page_testzero(page)) {
432 pgtable_page_dtor(page);
433 free_hot_cold_page(page, 0);
438 static void page_table_free_rcu(void *table)
440 struct page *page = virt_to_page(table);
441 if (put_page_testzero(page)) {
442 pgtable_page_dtor(page);
443 free_hot_cold_page(page, 0);
447 void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift)
449 unsigned long pgf = (unsigned long)table;
451 BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
453 tlb_remove_table(tlb, (void *)pgf);
456 void __tlb_remove_table(void *_table)
458 void *table = (void *)((unsigned long)_table & ~MAX_PGTABLE_INDEX_SIZE);
459 unsigned shift = (unsigned long)_table & MAX_PGTABLE_INDEX_SIZE;
462 /* PTE page needs special handling */
463 page_table_free_rcu(table);
465 BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
466 kmem_cache_free(PGT_CACHE(shift), table);
470 void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift)
473 /* PTE page needs special handling */
474 struct page *page = virt_to_page(table);
475 if (put_page_testzero(page)) {
476 pgtable_page_dtor(page);
477 free_hot_cold_page(page, 0);
480 BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
481 kmem_cache_free(PGT_CACHE(shift), table);
485 #endif /* CONFIG_PPC_64K_PAGES */
487 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
490 * This is called when relaxing access to a hugepage. It's also called in the page
491 * fault path when we don't hit any of the major fault cases, ie, a minor
492 * update of _PAGE_ACCESSED, _PAGE_DIRTY, etc... The generic code will have
493 * handled those two for us, we additionally deal with missing execute
494 * permission here on some processors
496 int pmdp_set_access_flags(struct vm_area_struct *vma, unsigned long address,
497 pmd_t *pmdp, pmd_t entry, int dirty)
500 #ifdef CONFIG_DEBUG_VM
501 WARN_ON(!pmd_trans_huge(*pmdp));
502 assert_spin_locked(&vma->vm_mm->page_table_lock);
504 changed = !pmd_same(*(pmdp), entry);
506 __ptep_set_access_flags(pmdp_ptep(pmdp), pmd_pte(entry));
508 * Since we are not supporting SW TLB systems, we don't
509 * have any thing similar to flush_tlb_page_nohash()
515 unsigned long pmd_hugepage_update(struct mm_struct *mm, unsigned long addr,
516 pmd_t *pmdp, unsigned long clr,
520 unsigned long old, tmp;
522 #ifdef CONFIG_DEBUG_VM
523 WARN_ON(!pmd_trans_huge(*pmdp));
524 assert_spin_locked(&mm->page_table_lock);
527 #ifdef PTE_ATOMIC_UPDATES
528 __asm__ __volatile__(
536 : "=&r" (old), "=&r" (tmp), "=m" (*pmdp)
537 : "r" (pmdp), "r" (clr), "m" (*pmdp), "i" (_PAGE_BUSY), "r" (set)
540 old = pmd_val(*pmdp);
541 *pmdp = __pmd((old & ~clr) | set);
543 trace_hugepage_update(addr, old, clr, set);
544 if (old & _PAGE_HASHPTE)
545 hpte_do_hugepage_flush(mm, addr, pmdp, old);
549 pmd_t pmdp_clear_flush(struct vm_area_struct *vma, unsigned long address,
554 VM_BUG_ON(address & ~HPAGE_PMD_MASK);
555 if (pmd_trans_huge(*pmdp)) {
556 pmd = pmdp_get_and_clear(vma->vm_mm, address, pmdp);
559 * khugepaged calls this for normal pmd
564 * Wait for all pending hash_page to finish. This is needed
565 * in case of subpage collapse. When we collapse normal pages
566 * to hugepage, we first clear the pmd, then invalidate all
567 * the PTE entries. The assumption here is that any low level
568 * page fault will see a none pmd and take the slow path that
569 * will wait on mmap_sem. But we could very well be in a
570 * hash_page with local ptep pointer value. Such a hash page
571 * can result in adding new HPTE entries for normal subpages.
572 * That means we could be modifying the page content as we
573 * copy them to a huge page. So wait for parallel hash_page
574 * to finish before invalidating HPTE entries. We can do this
575 * by sending an IPI to all the cpus and executing a dummy
578 kick_all_cpus_sync();
580 * Now invalidate the hpte entries in the range
581 * covered by pmd. This make sure we take a
582 * fault and will find the pmd as none, which will
583 * result in a major fault which takes mmap_sem and
584 * hence wait for collapse to complete. Without this
585 * the __collapse_huge_page_copy can result in copying
588 flush_tlb_pmd_range(vma->vm_mm, &pmd, address);
593 int pmdp_test_and_clear_young(struct vm_area_struct *vma,
594 unsigned long address, pmd_t *pmdp)
596 return __pmdp_test_and_clear_young(vma->vm_mm, address, pmdp);
600 * We currently remove entries from the hashtable regardless of whether
601 * the entry was young or dirty. The generic routines only flush if the
602 * entry was young or dirty which is not good enough.
604 * We should be more intelligent about this but for the moment we override
605 * these functions and force a tlb flush unconditionally
607 int pmdp_clear_flush_young(struct vm_area_struct *vma,
608 unsigned long address, pmd_t *pmdp)
610 return __pmdp_test_and_clear_young(vma->vm_mm, address, pmdp);
614 * We mark the pmd splitting and invalidate all the hpte
615 * entries for this hugepage.
617 void pmdp_splitting_flush(struct vm_area_struct *vma,
618 unsigned long address, pmd_t *pmdp)
620 unsigned long old, tmp;
622 VM_BUG_ON(address & ~HPAGE_PMD_MASK);
624 #ifdef CONFIG_DEBUG_VM
625 WARN_ON(!pmd_trans_huge(*pmdp));
626 assert_spin_locked(&vma->vm_mm->page_table_lock);
629 #ifdef PTE_ATOMIC_UPDATES
631 __asm__ __volatile__(
638 : "=&r" (old), "=&r" (tmp), "=m" (*pmdp)
639 : "r" (pmdp), "i" (_PAGE_SPLITTING), "m" (*pmdp), "i" (_PAGE_BUSY)
642 old = pmd_val(*pmdp);
643 *pmdp = __pmd(old | _PAGE_SPLITTING);
646 * If we didn't had the splitting flag set, go and flush the
649 trace_hugepage_splitting(address, old);
650 if (!(old & _PAGE_SPLITTING)) {
651 /* We need to flush the hpte */
652 if (old & _PAGE_HASHPTE)
653 hpte_do_hugepage_flush(vma->vm_mm, address, pmdp, old);
656 * This ensures that generic code that rely on IRQ disabling
657 * to prevent a parallel THP split work as expected.
659 kick_all_cpus_sync();
663 * We want to put the pgtable in pmd and use pgtable for tracking
664 * the base page size hptes
666 void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
669 pgtable_t *pgtable_slot;
670 assert_spin_locked(&mm->page_table_lock);
672 * we store the pgtable in the second half of PMD
674 pgtable_slot = (pgtable_t *)pmdp + PTRS_PER_PMD;
675 *pgtable_slot = pgtable;
677 * expose the deposited pgtable to other cpus.
678 * before we set the hugepage PTE at pmd level
679 * hash fault code looks at the deposted pgtable
680 * to store hash index values.
685 pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
688 pgtable_t *pgtable_slot;
690 assert_spin_locked(&mm->page_table_lock);
691 pgtable_slot = (pgtable_t *)pmdp + PTRS_PER_PMD;
692 pgtable = *pgtable_slot;
694 * Once we withdraw, mark the entry NULL.
696 *pgtable_slot = NULL;
698 * We store HPTE information in the deposited PTE fragment.
699 * zero out the content on withdraw.
701 memset(pgtable, 0, PTE_FRAG_SIZE);
706 * set a new huge pmd. We should not be called for updating
707 * an existing pmd entry. That should go via pmd_hugepage_update.
709 void set_pmd_at(struct mm_struct *mm, unsigned long addr,
710 pmd_t *pmdp, pmd_t pmd)
712 #ifdef CONFIG_DEBUG_VM
713 WARN_ON(pmd_val(*pmdp) & _PAGE_PRESENT);
714 assert_spin_locked(&mm->page_table_lock);
715 WARN_ON(!pmd_trans_huge(pmd));
717 trace_hugepage_set_pmd(addr, pmd);
718 return set_pte_at(mm, addr, pmdp_ptep(pmdp), pmd_pte(pmd));
721 void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
724 pmd_hugepage_update(vma->vm_mm, address, pmdp, _PAGE_PRESENT, 0);
728 * A linux hugepage PMD was changed and the corresponding hash table entries
729 * neesd to be flushed.
731 void hpte_do_hugepage_flush(struct mm_struct *mm, unsigned long addr,
732 pmd_t *pmdp, unsigned long old_pmd)
735 unsigned long s_addr;
737 unsigned int psize, valid;
738 unsigned char *hpte_slot_array;
739 unsigned long hidx, vpn, vsid, hash, shift, slot;
742 * Flush all the hptes mapping this hugepage
744 s_addr = addr & HPAGE_PMD_MASK;
745 hpte_slot_array = get_hpte_slot_array(pmdp);
747 * IF we try to do a HUGE PTE update after a withdraw is done.
748 * we will find the below NULL. This happens when we do
749 * split_huge_page_pmd
751 if (!hpte_slot_array)
754 /* get the base page size,vsid and segment size */
755 #ifdef CONFIG_DEBUG_VM
756 psize = get_slice_psize(mm, s_addr);
757 BUG_ON(psize == MMU_PAGE_16M);
759 if (old_pmd & _PAGE_COMBO)
762 psize = MMU_PAGE_64K;
764 if (!is_kernel_addr(s_addr)) {
765 ssize = user_segment_size(s_addr);
766 vsid = get_vsid(mm->context.id, s_addr, ssize);
769 vsid = get_kernel_vsid(s_addr, mmu_kernel_ssize);
770 ssize = mmu_kernel_ssize;
773 if (ppc_md.hugepage_invalidate)
774 return ppc_md.hugepage_invalidate(vsid, s_addr,
778 * No bluk hpte removal support, invalidate each entry
780 shift = mmu_psize_defs[psize].shift;
781 max_hpte_count = HPAGE_PMD_SIZE >> shift;
782 for (i = 0; i < max_hpte_count; i++) {
784 * 8 bits per each hpte entries
785 * 000| [ secondary group (one bit) | hidx (3 bits) | valid bit]
787 valid = hpte_valid(hpte_slot_array, i);
790 hidx = hpte_hash_index(hpte_slot_array, i);
793 addr = s_addr + (i * (1ul << shift));
794 vpn = hpt_vpn(addr, vsid, ssize);
795 hash = hpt_hash(vpn, shift, ssize);
796 if (hidx & _PTEIDX_SECONDARY)
799 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
800 slot += hidx & _PTEIDX_GROUP_IX;
801 ppc_md.hpte_invalidate(slot, vpn, psize,
802 MMU_PAGE_16M, ssize, 0);
806 static pmd_t pmd_set_protbits(pmd_t pmd, pgprot_t pgprot)
808 pmd_val(pmd) |= pgprot_val(pgprot);
812 pmd_t pfn_pmd(unsigned long pfn, pgprot_t pgprot)
816 * For a valid pte, we would have _PAGE_PRESENT or _PAGE_FILE always
817 * set. We use this to check THP page at pmd level.
818 * leaf pte for huge page, bottom two bits != 00
820 pmd_val(pmd) = pfn << PTE_RPN_SHIFT;
821 pmd_val(pmd) |= _PAGE_THP_HUGE;
822 pmd = pmd_set_protbits(pmd, pgprot);
826 pmd_t mk_pmd(struct page *page, pgprot_t pgprot)
828 return pfn_pmd(page_to_pfn(page), pgprot);
831 pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
834 pmd_val(pmd) &= _HPAGE_CHG_MASK;
835 pmd = pmd_set_protbits(pmd, newprot);
840 * This is called at the end of handling a user page fault, when the
841 * fault has been handled by updating a HUGE PMD entry in the linux page tables.
842 * We use it to preload an HPTE into the hash table corresponding to
843 * the updated linux HUGE PMD entry.
845 void update_mmu_cache_pmd(struct vm_area_struct *vma, unsigned long addr,
851 pmd_t pmdp_get_and_clear(struct mm_struct *mm,
852 unsigned long addr, pmd_t *pmdp)
857 pgtable_t *pgtable_slot;
859 old = pmd_hugepage_update(mm, addr, pmdp, ~0UL, 0);
860 old_pmd = __pmd(old);
862 * We have pmd == none and we are holding page_table_lock.
863 * So we can safely go and clear the pgtable hash
866 pgtable_slot = (pgtable_t *)pmdp + PTRS_PER_PMD;
867 pgtable = *pgtable_slot;
869 * Let's zero out old valid and hash index details
870 * hash fault look at them.
872 memset(pgtable, 0, PTE_FRAG_SIZE);
876 int has_transparent_hugepage(void)
878 if (!mmu_has_feature(MMU_FTR_16M_PAGE))
881 * We support THP only if PMD_SIZE is 16MB.
883 if (mmu_psize_defs[MMU_PAGE_16M].shift != PMD_SHIFT)
886 * We need to make sure that we support 16MB hugepage in a segement
887 * with base page size 64K or 4K. We only enable THP with a PAGE_SIZE
891 * If we have 64K HPTE, we will be using that by default
893 if (mmu_psize_defs[MMU_PAGE_64K].shift &&
894 (mmu_psize_defs[MMU_PAGE_64K].penc[MMU_PAGE_16M] == -1))
897 * Ok we only have 4K HPTE
899 if (mmu_psize_defs[MMU_PAGE_4K].penc[MMU_PAGE_16M] == -1)
904 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */