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 /* Some sanity checking */
58 #if TASK_SIZE_USER64 > PGTABLE_RANGE
59 #error TASK_SIZE_USER64 exceeds pagetable range
62 #ifdef CONFIG_PPC_STD_MMU_64
63 #if TASK_SIZE_USER64 > (1UL << (ESID_BITS + SID_SHIFT))
64 #error TASK_SIZE_USER64 exceeds user VSID range
68 unsigned long ioremap_bot = IOREMAP_BASE;
70 #ifdef CONFIG_PPC_MMU_NOHASH
71 static void *early_alloc_pgtable(unsigned long size)
75 if (init_bootmem_done)
76 pt = __alloc_bootmem(size, size, __pa(MAX_DMA_ADDRESS));
78 pt = __va(memblock_alloc_base(size, size,
79 __pa(MAX_DMA_ADDRESS)));
84 #endif /* CONFIG_PPC_MMU_NOHASH */
87 * map_kernel_page currently only called by __ioremap
88 * map_kernel_page adds an entry to the ioremap page table
89 * and adds an entry to the HPT, possibly bolting it
91 int map_kernel_page(unsigned long ea, unsigned long pa, int flags)
98 if (slab_is_available()) {
99 pgdp = pgd_offset_k(ea);
100 pudp = pud_alloc(&init_mm, pgdp, ea);
103 pmdp = pmd_alloc(&init_mm, pudp, ea);
106 ptep = pte_alloc_kernel(pmdp, ea);
109 set_pte_at(&init_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT,
112 #ifdef CONFIG_PPC_MMU_NOHASH
113 /* Warning ! This will blow up if bootmem is not initialized
114 * which our ppc64 code is keen to do that, we'll need to
115 * fix it and/or be more careful
117 pgdp = pgd_offset_k(ea);
118 #ifdef PUD_TABLE_SIZE
119 if (pgd_none(*pgdp)) {
120 pudp = early_alloc_pgtable(PUD_TABLE_SIZE);
121 BUG_ON(pudp == NULL);
122 pgd_populate(&init_mm, pgdp, pudp);
124 #endif /* PUD_TABLE_SIZE */
125 pudp = pud_offset(pgdp, ea);
126 if (pud_none(*pudp)) {
127 pmdp = early_alloc_pgtable(PMD_TABLE_SIZE);
128 BUG_ON(pmdp == NULL);
129 pud_populate(&init_mm, pudp, pmdp);
131 pmdp = pmd_offset(pudp, ea);
132 if (!pmd_present(*pmdp)) {
133 ptep = early_alloc_pgtable(PAGE_SIZE);
134 BUG_ON(ptep == NULL);
135 pmd_populate_kernel(&init_mm, pmdp, ptep);
137 ptep = pte_offset_kernel(pmdp, ea);
138 set_pte_at(&init_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT,
140 #else /* CONFIG_PPC_MMU_NOHASH */
142 * If the mm subsystem is not fully up, we cannot create a
143 * linux page table entry for this mapping. Simply bolt an
144 * entry in the hardware page table.
147 if (htab_bolt_mapping(ea, ea + PAGE_SIZE, pa, flags,
148 mmu_io_psize, mmu_kernel_ssize)) {
149 printk(KERN_ERR "Failed to do bolted mapping IO "
150 "memory at %016lx !\n", pa);
153 #endif /* !CONFIG_PPC_MMU_NOHASH */
156 #ifdef CONFIG_PPC_BOOK3E_64
158 * With hardware tablewalk, a sync is needed to ensure that
159 * subsequent accesses see the PTE we just wrote. Unlike userspace
160 * mappings, we can't tolerate spurious faults, so make sure
161 * the new PTE will be seen the first time.
172 * __ioremap_at - Low level function to establish the page tables
175 void __iomem * __ioremap_at(phys_addr_t pa, void *ea, unsigned long size,
180 /* Make sure we have the base flags */
181 if ((flags & _PAGE_PRESENT) == 0)
182 flags |= pgprot_val(PAGE_KERNEL);
184 /* Non-cacheable page cannot be coherent */
185 if (flags & _PAGE_NO_CACHE)
186 flags &= ~_PAGE_COHERENT;
188 /* We don't support the 4K PFN hack with ioremap */
189 if (flags & _PAGE_4K_PFN)
192 WARN_ON(pa & ~PAGE_MASK);
193 WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
194 WARN_ON(size & ~PAGE_MASK);
196 for (i = 0; i < size; i += PAGE_SIZE)
197 if (map_kernel_page((unsigned long)ea+i, pa+i, flags))
200 return (void __iomem *)ea;
204 * __iounmap_from - Low level function to tear down the page tables
205 * for an IO mapping. This is used for mappings that
206 * are manipulated manually, like partial unmapping of
207 * PCI IOs or ISA space.
209 void __iounmap_at(void *ea, unsigned long size)
211 WARN_ON(((unsigned long)ea) & ~PAGE_MASK);
212 WARN_ON(size & ~PAGE_MASK);
214 unmap_kernel_range((unsigned long)ea, size);
217 void __iomem * __ioremap_caller(phys_addr_t addr, unsigned long size,
218 unsigned long flags, void *caller)
220 phys_addr_t paligned;
224 * Choose an address to map it to.
225 * Once the imalloc system is running, we use it.
226 * Before that, we map using addresses going
227 * up from ioremap_bot. imalloc will use
228 * the addresses from ioremap_bot through
232 paligned = addr & PAGE_MASK;
233 size = PAGE_ALIGN(addr + size) - paligned;
235 if ((size == 0) || (paligned == 0))
239 struct vm_struct *area;
241 area = __get_vm_area_caller(size, VM_IOREMAP,
242 ioremap_bot, IOREMAP_END,
247 area->phys_addr = paligned;
248 ret = __ioremap_at(paligned, area->addr, size, flags);
252 ret = __ioremap_at(paligned, (void *)ioremap_bot, size, flags);
258 ret += addr & ~PAGE_MASK;
262 void __iomem * __ioremap(phys_addr_t addr, unsigned long size,
265 return __ioremap_caller(addr, size, flags, __builtin_return_address(0));
268 void __iomem * ioremap(phys_addr_t addr, unsigned long size)
270 unsigned long flags = _PAGE_NO_CACHE | _PAGE_GUARDED;
271 void *caller = __builtin_return_address(0);
274 return ppc_md.ioremap(addr, size, flags, caller);
275 return __ioremap_caller(addr, size, flags, caller);
278 void __iomem * ioremap_wc(phys_addr_t addr, unsigned long size)
280 unsigned long flags = _PAGE_NO_CACHE;
281 void *caller = __builtin_return_address(0);
284 return ppc_md.ioremap(addr, size, flags, caller);
285 return __ioremap_caller(addr, size, flags, caller);
288 void __iomem * ioremap_prot(phys_addr_t addr, unsigned long size,
291 void *caller = __builtin_return_address(0);
293 /* writeable implies dirty for kernel addresses */
294 if (flags & _PAGE_RW)
295 flags |= _PAGE_DIRTY;
297 /* we don't want to let _PAGE_USER and _PAGE_EXEC leak out */
298 flags &= ~(_PAGE_USER | _PAGE_EXEC);
301 /* _PAGE_USER contains _PAGE_BAP_SR on BookE using the new PTE format
302 * which means that we just cleared supervisor access... oops ;-) This
305 flags |= _PAGE_BAP_SR;
309 return ppc_md.ioremap(addr, size, flags, caller);
310 return __ioremap_caller(addr, size, flags, caller);
315 * Unmap an IO region and remove it from imalloc'd list.
316 * Access to IO memory should be serialized by driver.
318 void __iounmap(volatile void __iomem *token)
325 addr = (void *) ((unsigned long __force)
326 PCI_FIX_ADDR(token) & PAGE_MASK);
327 if ((unsigned long)addr < ioremap_bot) {
328 printk(KERN_WARNING "Attempt to iounmap early bolted mapping"
335 void iounmap(volatile void __iomem *token)
338 ppc_md.iounmap(token);
343 EXPORT_SYMBOL(ioremap);
344 EXPORT_SYMBOL(ioremap_wc);
345 EXPORT_SYMBOL(ioremap_prot);
346 EXPORT_SYMBOL(__ioremap);
347 EXPORT_SYMBOL(__ioremap_at);
348 EXPORT_SYMBOL(iounmap);
349 EXPORT_SYMBOL(__iounmap);
350 EXPORT_SYMBOL(__iounmap_at);
353 * For hugepage we have pfn in the pmd, we use PTE_RPN_SHIFT bits for flags
354 * For PTE page, we have a PTE_FRAG_SIZE (4K) aligned virtual address.
356 struct page *pmd_page(pmd_t pmd)
358 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
359 if (pmd_trans_huge(pmd))
360 return pfn_to_page(pmd_pfn(pmd));
362 return virt_to_page(pmd_page_vaddr(pmd));
365 #ifdef CONFIG_PPC_64K_PAGES
366 static pte_t *get_from_cache(struct mm_struct *mm)
368 void *pte_frag, *ret;
370 spin_lock(&mm->page_table_lock);
371 ret = mm->context.pte_frag;
373 pte_frag = ret + PTE_FRAG_SIZE;
375 * If we have taken up all the fragments mark PTE page NULL
377 if (((unsigned long)pte_frag & ~PAGE_MASK) == 0)
379 mm->context.pte_frag = pte_frag;
381 spin_unlock(&mm->page_table_lock);
385 static pte_t *__alloc_for_cache(struct mm_struct *mm, int kernel)
388 struct page *page = alloc_page(GFP_KERNEL | __GFP_NOTRACK |
389 __GFP_REPEAT | __GFP_ZERO);
392 if (!kernel && !pgtable_page_ctor(page)) {
397 ret = page_address(page);
398 spin_lock(&mm->page_table_lock);
400 * If we find pgtable_page set, we return
401 * the allocated page with single fragement
404 if (likely(!mm->context.pte_frag)) {
405 atomic_set(&page->_count, PTE_FRAG_NR);
406 mm->context.pte_frag = ret + PTE_FRAG_SIZE;
408 spin_unlock(&mm->page_table_lock);
413 pte_t *page_table_alloc(struct mm_struct *mm, unsigned long vmaddr, int kernel)
417 pte = get_from_cache(mm);
421 return __alloc_for_cache(mm, kernel);
424 void page_table_free(struct mm_struct *mm, unsigned long *table, int kernel)
426 struct page *page = virt_to_page(table);
427 if (put_page_testzero(page)) {
429 pgtable_page_dtor(page);
430 free_hot_cold_page(page, 0);
435 static void page_table_free_rcu(void *table)
437 struct page *page = virt_to_page(table);
438 if (put_page_testzero(page)) {
439 pgtable_page_dtor(page);
440 free_hot_cold_page(page, 0);
444 void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift)
446 unsigned long pgf = (unsigned long)table;
448 BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
450 tlb_remove_table(tlb, (void *)pgf);
453 void __tlb_remove_table(void *_table)
455 void *table = (void *)((unsigned long)_table & ~MAX_PGTABLE_INDEX_SIZE);
456 unsigned shift = (unsigned long)_table & MAX_PGTABLE_INDEX_SIZE;
459 /* PTE page needs special handling */
460 page_table_free_rcu(table);
462 BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
463 kmem_cache_free(PGT_CACHE(shift), table);
467 void pgtable_free_tlb(struct mmu_gather *tlb, void *table, int shift)
470 /* PTE page needs special handling */
471 struct page *page = virt_to_page(table);
472 if (put_page_testzero(page)) {
473 pgtable_page_dtor(page);
474 free_hot_cold_page(page, 0);
477 BUG_ON(shift > MAX_PGTABLE_INDEX_SIZE);
478 kmem_cache_free(PGT_CACHE(shift), table);
482 #endif /* CONFIG_PPC_64K_PAGES */
484 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
487 * This is called when relaxing access to a hugepage. It's also called in the page
488 * fault path when we don't hit any of the major fault cases, ie, a minor
489 * update of _PAGE_ACCESSED, _PAGE_DIRTY, etc... The generic code will have
490 * handled those two for us, we additionally deal with missing execute
491 * permission here on some processors
493 int pmdp_set_access_flags(struct vm_area_struct *vma, unsigned long address,
494 pmd_t *pmdp, pmd_t entry, int dirty)
497 #ifdef CONFIG_DEBUG_VM
498 WARN_ON(!pmd_trans_huge(*pmdp));
499 assert_spin_locked(&vma->vm_mm->page_table_lock);
501 changed = !pmd_same(*(pmdp), entry);
503 __ptep_set_access_flags(pmdp_ptep(pmdp), pmd_pte(entry));
505 * Since we are not supporting SW TLB systems, we don't
506 * have any thing similar to flush_tlb_page_nohash()
512 unsigned long pmd_hugepage_update(struct mm_struct *mm, unsigned long addr,
513 pmd_t *pmdp, unsigned long clr,
517 unsigned long old, tmp;
519 #ifdef CONFIG_DEBUG_VM
520 WARN_ON(!pmd_trans_huge(*pmdp));
521 assert_spin_locked(&mm->page_table_lock);
524 #ifdef PTE_ATOMIC_UPDATES
525 __asm__ __volatile__(
533 : "=&r" (old), "=&r" (tmp), "=m" (*pmdp)
534 : "r" (pmdp), "r" (clr), "m" (*pmdp), "i" (_PAGE_BUSY), "r" (set)
537 old = pmd_val(*pmdp);
538 *pmdp = __pmd((old & ~clr) | set);
540 if (old & _PAGE_HASHPTE)
541 hpte_do_hugepage_flush(mm, addr, pmdp);
545 pmd_t pmdp_clear_flush(struct vm_area_struct *vma, unsigned long address,
550 VM_BUG_ON(address & ~HPAGE_PMD_MASK);
551 if (pmd_trans_huge(*pmdp)) {
552 pmd = pmdp_get_and_clear(vma->vm_mm, address, pmdp);
555 * khugepaged calls this for normal pmd
560 * Wait for all pending hash_page to finish. This is needed
561 * in case of subpage collapse. When we collapse normal pages
562 * to hugepage, we first clear the pmd, then invalidate all
563 * the PTE entries. The assumption here is that any low level
564 * page fault will see a none pmd and take the slow path that
565 * will wait on mmap_sem. But we could very well be in a
566 * hash_page with local ptep pointer value. Such a hash page
567 * can result in adding new HPTE entries for normal subpages.
568 * That means we could be modifying the page content as we
569 * copy them to a huge page. So wait for parallel hash_page
570 * to finish before invalidating HPTE entries. We can do this
571 * by sending an IPI to all the cpus and executing a dummy
574 kick_all_cpus_sync();
576 * Now invalidate the hpte entries in the range
577 * covered by pmd. This make sure we take a
578 * fault and will find the pmd as none, which will
579 * result in a major fault which takes mmap_sem and
580 * hence wait for collapse to complete. Without this
581 * the __collapse_huge_page_copy can result in copying
584 flush_tlb_pmd_range(vma->vm_mm, &pmd, address);
589 int pmdp_test_and_clear_young(struct vm_area_struct *vma,
590 unsigned long address, pmd_t *pmdp)
592 return __pmdp_test_and_clear_young(vma->vm_mm, address, pmdp);
596 * We currently remove entries from the hashtable regardless of whether
597 * the entry was young or dirty. The generic routines only flush if the
598 * entry was young or dirty which is not good enough.
600 * We should be more intelligent about this but for the moment we override
601 * these functions and force a tlb flush unconditionally
603 int pmdp_clear_flush_young(struct vm_area_struct *vma,
604 unsigned long address, pmd_t *pmdp)
606 return __pmdp_test_and_clear_young(vma->vm_mm, address, pmdp);
610 * We mark the pmd splitting and invalidate all the hpte
611 * entries for this hugepage.
613 void pmdp_splitting_flush(struct vm_area_struct *vma,
614 unsigned long address, pmd_t *pmdp)
616 unsigned long old, tmp;
618 VM_BUG_ON(address & ~HPAGE_PMD_MASK);
620 #ifdef CONFIG_DEBUG_VM
621 WARN_ON(!pmd_trans_huge(*pmdp));
622 assert_spin_locked(&vma->vm_mm->page_table_lock);
625 #ifdef PTE_ATOMIC_UPDATES
627 __asm__ __volatile__(
634 : "=&r" (old), "=&r" (tmp), "=m" (*pmdp)
635 : "r" (pmdp), "i" (_PAGE_SPLITTING), "m" (*pmdp), "i" (_PAGE_BUSY)
638 old = pmd_val(*pmdp);
639 *pmdp = __pmd(old | _PAGE_SPLITTING);
642 * If we didn't had the splitting flag set, go and flush the
645 if (!(old & _PAGE_SPLITTING)) {
646 /* We need to flush the hpte */
647 if (old & _PAGE_HASHPTE)
648 hpte_do_hugepage_flush(vma->vm_mm, address, pmdp);
653 * We want to put the pgtable in pmd and use pgtable for tracking
654 * the base page size hptes
656 void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
659 pgtable_t *pgtable_slot;
660 assert_spin_locked(&mm->page_table_lock);
662 * we store the pgtable in the second half of PMD
664 pgtable_slot = (pgtable_t *)pmdp + PTRS_PER_PMD;
665 *pgtable_slot = pgtable;
667 * expose the deposited pgtable to other cpus.
668 * before we set the hugepage PTE at pmd level
669 * hash fault code looks at the deposted pgtable
670 * to store hash index values.
675 pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
678 pgtable_t *pgtable_slot;
680 assert_spin_locked(&mm->page_table_lock);
681 pgtable_slot = (pgtable_t *)pmdp + PTRS_PER_PMD;
682 pgtable = *pgtable_slot;
684 * Once we withdraw, mark the entry NULL.
686 *pgtable_slot = NULL;
688 * We store HPTE information in the deposited PTE fragment.
689 * zero out the content on withdraw.
691 memset(pgtable, 0, PTE_FRAG_SIZE);
696 * set a new huge pmd. We should not be called for updating
697 * an existing pmd entry. That should go via pmd_hugepage_update.
699 void set_pmd_at(struct mm_struct *mm, unsigned long addr,
700 pmd_t *pmdp, pmd_t pmd)
702 #ifdef CONFIG_DEBUG_VM
703 WARN_ON(pmd_val(*pmdp) & _PAGE_PRESENT);
704 assert_spin_locked(&mm->page_table_lock);
705 WARN_ON(!pmd_trans_huge(pmd));
707 return set_pte_at(mm, addr, pmdp_ptep(pmdp), pmd_pte(pmd));
710 void pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
713 pmd_hugepage_update(vma->vm_mm, address, pmdp, _PAGE_PRESENT, 0);
717 * A linux hugepage PMD was changed and the corresponding hash table entries
718 * neesd to be flushed.
720 void hpte_do_hugepage_flush(struct mm_struct *mm, unsigned long addr,
724 unsigned long s_addr;
726 unsigned int psize, valid;
727 unsigned char *hpte_slot_array;
728 unsigned long hidx, vpn, vsid, hash, shift, slot;
731 * Flush all the hptes mapping this hugepage
733 s_addr = addr & HPAGE_PMD_MASK;
734 hpte_slot_array = get_hpte_slot_array(pmdp);
736 * IF we try to do a HUGE PTE update after a withdraw is done.
737 * we will find the below NULL. This happens when we do
738 * split_huge_page_pmd
740 if (!hpte_slot_array)
743 /* get the base page size */
744 psize = get_slice_psize(mm, s_addr);
746 if (ppc_md.hugepage_invalidate)
747 return ppc_md.hugepage_invalidate(mm, hpte_slot_array,
750 * No bluk hpte removal support, invalidate each entry
752 shift = mmu_psize_defs[psize].shift;
753 max_hpte_count = HPAGE_PMD_SIZE >> shift;
754 for (i = 0; i < max_hpte_count; i++) {
756 * 8 bits per each hpte entries
757 * 000| [ secondary group (one bit) | hidx (3 bits) | valid bit]
759 valid = hpte_valid(hpte_slot_array, i);
762 hidx = hpte_hash_index(hpte_slot_array, i);
765 addr = s_addr + (i * (1ul << shift));
766 if (!is_kernel_addr(addr)) {
767 ssize = user_segment_size(addr);
768 vsid = get_vsid(mm->context.id, addr, ssize);
771 vsid = get_kernel_vsid(addr, mmu_kernel_ssize);
772 ssize = mmu_kernel_ssize;
775 vpn = hpt_vpn(addr, vsid, ssize);
776 hash = hpt_hash(vpn, shift, ssize);
777 if (hidx & _PTEIDX_SECONDARY)
780 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
781 slot += hidx & _PTEIDX_GROUP_IX;
782 ppc_md.hpte_invalidate(slot, vpn, psize,
783 MMU_PAGE_16M, ssize, 0);
787 static pmd_t pmd_set_protbits(pmd_t pmd, pgprot_t pgprot)
789 pmd_val(pmd) |= pgprot_val(pgprot);
793 pmd_t pfn_pmd(unsigned long pfn, pgprot_t pgprot)
797 * For a valid pte, we would have _PAGE_PRESENT or _PAGE_FILE always
798 * set. We use this to check THP page at pmd level.
799 * leaf pte for huge page, bottom two bits != 00
801 pmd_val(pmd) = pfn << PTE_RPN_SHIFT;
802 pmd_val(pmd) |= _PAGE_THP_HUGE;
803 pmd = pmd_set_protbits(pmd, pgprot);
807 pmd_t mk_pmd(struct page *page, pgprot_t pgprot)
809 return pfn_pmd(page_to_pfn(page), pgprot);
812 pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
815 pmd_val(pmd) &= _HPAGE_CHG_MASK;
816 pmd = pmd_set_protbits(pmd, newprot);
821 * This is called at the end of handling a user page fault, when the
822 * fault has been handled by updating a HUGE PMD entry in the linux page tables.
823 * We use it to preload an HPTE into the hash table corresponding to
824 * the updated linux HUGE PMD entry.
826 void update_mmu_cache_pmd(struct vm_area_struct *vma, unsigned long addr,
832 pmd_t pmdp_get_and_clear(struct mm_struct *mm,
833 unsigned long addr, pmd_t *pmdp)
838 pgtable_t *pgtable_slot;
840 old = pmd_hugepage_update(mm, addr, pmdp, ~0UL, 0);
841 old_pmd = __pmd(old);
843 * We have pmd == none and we are holding page_table_lock.
844 * So we can safely go and clear the pgtable hash
847 pgtable_slot = (pgtable_t *)pmdp + PTRS_PER_PMD;
848 pgtable = *pgtable_slot;
850 * Let's zero out old valid and hash index details
851 * hash fault look at them.
853 memset(pgtable, 0, PTE_FRAG_SIZE);
857 int has_transparent_hugepage(void)
859 if (!mmu_has_feature(MMU_FTR_16M_PAGE))
862 * We support THP only if PMD_SIZE is 16MB.
864 if (mmu_psize_defs[MMU_PAGE_16M].shift != PMD_SHIFT)
867 * We need to make sure that we support 16MB hugepage in a segement
868 * with base page size 64K or 4K. We only enable THP with a PAGE_SIZE
872 * If we have 64K HPTE, we will be using that by default
874 if (mmu_psize_defs[MMU_PAGE_64K].shift &&
875 (mmu_psize_defs[MMU_PAGE_64K].penc[MMU_PAGE_16M] == -1))
878 * Ok we only have 4K HPTE
880 if (mmu_psize_defs[MMU_PAGE_4K].penc[MMU_PAGE_16M] == -1)
885 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */