1/*
  2 * Copyright 2015-2016, Aneesh Kumar K.V, IBM Corporation.
  3 *
  4 * This program is free software; you can redistribute it and/or
  5 * modify it under the terms of the GNU General Public License
  6 * as published by the Free Software Foundation; either version
  7 * 2 of the License, or (at your option) any later version.
  8 */
  9
 10#include <linux/sched.h>
 11#include <linux/mm_types.h>
 12#include <misc/cxl-base.h>
 13
 14#include <asm/pgalloc.h>
 15#include <asm/tlb.h>
 16
 17#include "mmu_decl.h"
 18#include <trace/events/thp.h>
 19
 20int (*register_process_table)(unsigned long base, unsigned long page_size,
 21			      unsigned long tbl_size);
 22
 23#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 24/*
 25 * This is called when relaxing access to a hugepage. It's also called in the page
 26 * fault path when we don't hit any of the major fault cases, ie, a minor
 27 * update of _PAGE_ACCESSED, _PAGE_DIRTY, etc... The generic code will have
 28 * handled those two for us, we additionally deal with missing execute
 29 * permission here on some processors
 30 */
 31int pmdp_set_access_flags(struct vm_area_struct *vma, unsigned long address,
 32			  pmd_t *pmdp, pmd_t entry, int dirty)
 33{
 34	int changed;
 35#ifdef CONFIG_DEBUG_VM
 36	WARN_ON(!pmd_trans_huge(*pmdp) && !pmd_devmap(*pmdp));
 37	assert_spin_locked(&vma->vm_mm->page_table_lock);
 38#endif
 39	changed = !pmd_same(*(pmdp), entry);
 40	if (changed) {
 41		__ptep_set_access_flags(vma->vm_mm, pmdp_ptep(pmdp),
 42					pmd_pte(entry), address);
 43		flush_pmd_tlb_range(vma, address, address + HPAGE_PMD_SIZE);
 44	}
 45	return changed;
 46}
 47
 48int pmdp_test_and_clear_young(struct vm_area_struct *vma,
 49			      unsigned long address, pmd_t *pmdp)
 50{
 51	return __pmdp_test_and_clear_young(vma->vm_mm, address, pmdp);
 52}
 53/*
 54 * set a new huge pmd. We should not be called for updating
 55 * an existing pmd entry. That should go via pmd_hugepage_update.
 56 */
 57void set_pmd_at(struct mm_struct *mm, unsigned long addr,
 58		pmd_t *pmdp, pmd_t pmd)
 59{
 60#ifdef CONFIG_DEBUG_VM
 61	WARN_ON(pte_present(pmd_pte(*pmdp)) && !pte_protnone(pmd_pte(*pmdp)));
 62	assert_spin_locked(&mm->page_table_lock);
 63	WARN_ON(!(pmd_trans_huge(pmd) || pmd_devmap(pmd)));
 64#endif
 65	trace_hugepage_set_pmd(addr, pmd_val(pmd));
 66	return set_pte_at(mm, addr, pmdp_ptep(pmdp), pmd_pte(pmd));
 67}
 68
 69static void do_nothing(void *unused)
 70{
 71
 72}
 73/*
 74 * Serialize against find_current_mm_pte which does lock-less
 75 * lookup in page tables with local interrupts disabled. For huge pages
 76 * it casts pmd_t to pte_t. Since format of pte_t is different from
 77 * pmd_t we want to prevent transit from pmd pointing to page table
 78 * to pmd pointing to huge page (and back) while interrupts are disabled.
 79 * We clear pmd to possibly replace it with page table pointer in
 80 * different code paths. So make sure we wait for the parallel
 81 * find_current_mm_pte to finish.
 82 */
 83void serialize_against_pte_lookup(struct mm_struct *mm)
 84{
 85	smp_mb();
 86	smp_call_function_many(mm_cpumask(mm), do_nothing, NULL, 1);
 87}
 88
 89/*
 90 * We use this to invalidate a pmdp entry before switching from a
 91 * hugepte to regular pmd entry.
 92 */
 93pmd_t pmdp_invalidate(struct vm_area_struct *vma, unsigned long address,
 94		     pmd_t *pmdp)
 95{
 96	unsigned long old_pmd;
 97
 98	old_pmd = pmd_hugepage_update(vma->vm_mm, address, pmdp, _PAGE_PRESENT, 0);
 99	flush_pmd_tlb_range(vma, address, address + HPAGE_PMD_SIZE);
100	/*
101	 * This ensures that generic code that rely on IRQ disabling
102	 * to prevent a parallel THP split work as expected.
103	 */
104	serialize_against_pte_lookup(vma->vm_mm);
105	return __pmd(old_pmd);
106}
107
108static pmd_t pmd_set_protbits(pmd_t pmd, pgprot_t pgprot)
109{
110	return __pmd(pmd_val(pmd) | pgprot_val(pgprot));
111}
112
113pmd_t pfn_pmd(unsigned long pfn, pgprot_t pgprot)
114{
115	unsigned long pmdv;
116
117	pmdv = (pfn << PAGE_SHIFT) & PTE_RPN_MASK;
118	return pmd_set_protbits(__pmd(pmdv), pgprot);
119}
120
121pmd_t mk_pmd(struct page *page, pgprot_t pgprot)
122{
123	return pfn_pmd(page_to_pfn(page), pgprot);
124}
125
126pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
127{
128	unsigned long pmdv;
129
130	pmdv = pmd_val(pmd);
131	pmdv &= _HPAGE_CHG_MASK;
132	return pmd_set_protbits(__pmd(pmdv), newprot);
133}
134
135/*
136 * This is called at the end of handling a user page fault, when the
137 * fault has been handled by updating a HUGE PMD entry in the linux page tables.
138 * We use it to preload an HPTE into the hash table corresponding to
139 * the updated linux HUGE PMD entry.
140 */
141void update_mmu_cache_pmd(struct vm_area_struct *vma, unsigned long addr,
142			  pmd_t *pmd)
143{
144	return;
145}
146#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
147
148/* For use by kexec */
149void mmu_cleanup_all(void)
150{
151	if (radix_enabled())
152		radix__mmu_cleanup_all();
153	else if (mmu_hash_ops.hpte_clear_all)
154		mmu_hash_ops.hpte_clear_all();
155}
156
157#ifdef CONFIG_MEMORY_HOTPLUG
158int __meminit create_section_mapping(unsigned long start, unsigned long end, int nid)
159{
160	if (radix_enabled())
161		return radix__create_section_mapping(start, end, nid);
162
163	return hash__create_section_mapping(start, end, nid);
164}
165
166int __meminit remove_section_mapping(unsigned long start, unsigned long end)
167{
168	if (radix_enabled())
169		return radix__remove_section_mapping(start, end);
170
171	return hash__remove_section_mapping(start, end);
172}
173#endif /* CONFIG_MEMORY_HOTPLUG */