1 /*
2 * Copyright IBM Corporation, 2013
3 * Author Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of version 2.1 of the GNU Lesser General Public License
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful, but
10 * WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 *
13 */
14
15 /*
16 * PPC64 THP Support for hash based MMUs
17 */
18 #include <linux/mm.h>
19 #include <asm/machdep.h>
20
21 int __hash_page_thp(unsigned long ea, unsigned long access, unsigned long vsid,
22 pmd_t *pmdp, unsigned long trap, unsigned long flags,
23 int ssize, unsigned int psize)
24 {
25 unsigned int index, valid;
26 unsigned char *hpte_slot_array;
27 unsigned long rflags, pa, hidx;
28 unsigned long old_pmd, new_pmd;
29 int ret, lpsize = MMU_PAGE_16M;
30 unsigned long vpn, hash, shift, slot;
31
32 /*
33 * atomically mark the linux large page PMD busy and dirty
34 */
35 do {
36 pmd_t pmd = READ_ONCE(*pmdp);
37
38 old_pmd = pmd_val(pmd);
39 /* If PMD busy, retry the access */
40 if (unlikely(old_pmd & H_PAGE_BUSY))
41 return 0;
42 /* If PMD permissions don't match, take page fault */
43 if (unlikely(!check_pte_access(access, old_pmd)))
44 return 1;
45 /*
46 * Try to lock the PTE, add ACCESSED and DIRTY if it was
47 * a write access
48 */
49 new_pmd = old_pmd | H_PAGE_BUSY | _PAGE_ACCESSED;
50 if (access & _PAGE_WRITE)
51 new_pmd |= _PAGE_DIRTY;
52 } while (!pmd_xchg(pmdp, __pmd(old_pmd), __pmd(new_pmd)));
53
54 /*
55 * Make sure this is thp or devmap entry
56 */
57 if (!(old_pmd & (H_PAGE_THP_HUGE | _PAGE_DEVMAP)))
58 return 0;
59
60 rflags = htab_convert_pte_flags(new_pmd);
61
62 #if 0
63 if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE)) {
64
65 /*
66 * No CPU has hugepages but lacks no execute, so we
67 * don't need to worry about that case
68 */
69 rflags = hash_page_do_lazy_icache(rflags, __pte(old_pte), trap);
70 }
71 #endif
72 /*
73 * Find the slot index details for this ea, using base page size.
74 */
75 shift = mmu_psize_defs[psize].shift;
76 index = (ea & ~HPAGE_PMD_MASK) >> shift;
77 BUG_ON(index >= PTE_FRAG_SIZE);
78
79 vpn = hpt_vpn(ea, vsid, ssize);
80 hpte_slot_array = get_hpte_slot_array(pmdp);
81 if (psize == MMU_PAGE_4K) {
82 /*
83 * invalidate the old hpte entry if we have that mapped via 64K
84 * base page size. This is because demote_segment won't flush
85 * hash page table entries.
86 */
87 if ((old_pmd & H_PAGE_HASHPTE) && !(old_pmd & H_PAGE_COMBO)) {
88 flush_hash_hugepage(vsid, ea, pmdp, MMU_PAGE_64K,
89 ssize, flags);
90 /*
91 * With THP, we also clear the slot information with
92 * respect to all the 64K hash pte mapping the 16MB
93 * page. They are all invalid now. This make sure we
94 * don't find the slot valid when we fault with 4k
95 * base page size.
96 *
97 */
98 memset(hpte_slot_array, 0, PTE_FRAG_SIZE);
99 }
100 }
101
102 valid = hpte_valid(hpte_slot_array, index);
103 if (valid) {
104 /* update the hpte bits */
105 hash = hpt_hash(vpn, shift, ssize);
106 hidx = hpte_hash_index(hpte_slot_array, index);
107 if (hidx & _PTEIDX_SECONDARY)
108 hash = ~hash;
109 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
110 slot += hidx & _PTEIDX_GROUP_IX;
111
112 ret = mmu_hash_ops.hpte_updatepp(slot, rflags, vpn,
113 psize, lpsize, ssize, flags);
114 /*
115 * We failed to update, try to insert a new entry.
116 */
117 if (ret == -1) {
118 /*
119 * large pte is marked busy, so we can be sure
120 * nobody is looking at hpte_slot_array. hence we can
121 * safely update this here.
122 */
123 valid = 0;
124 hpte_slot_array[index] = 0;
125 }
126 }
127
128 if (!valid) {
129 unsigned long hpte_group;
130
131 hash = hpt_hash(vpn, shift, ssize);
132 /* insert new entry */
133 pa = pmd_pfn(__pmd(old_pmd)) << PAGE_SHIFT;
134 new_pmd |= H_PAGE_HASHPTE;
135
136 repeat:
137 hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
138
139 /* Insert into the hash table, primary slot */
140 slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags, 0,
141 psize, lpsize, ssize);
142 /*
143 * Primary is full, try the secondary
144 */
145 if (unlikely(slot == -1)) {
146 hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
147 slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa,
148 rflags,
149 HPTE_V_SECONDARY,
150 psize, lpsize, ssize);
151 if (slot == -1) {
152 if (mftb() & 0x1)
153 hpte_group = (hash & htab_hash_mask) *
154 HPTES_PER_GROUP;
155
156 mmu_hash_ops.hpte_remove(hpte_group);
157 goto repeat;
158 }
159 }
160 /*
161 * Hypervisor failure. Restore old pmd and return -1
162 * similar to __hash_page_*
163 */
164 if (unlikely(slot == -2)) {
165 *pmdp = __pmd(old_pmd);
166 hash_failure_debug(ea, access, vsid, trap, ssize,
167 psize, lpsize, old_pmd);
168 return -1;
169 }
170 /*
171 * large pte is marked busy, so we can be sure
172 * nobody is looking at hpte_slot_array. hence we can
173 * safely update this here.
174 */
175 mark_hpte_slot_valid(hpte_slot_array, index, slot);
176 }
177 /*
178 * Mark the pte with H_PAGE_COMBO, if we are trying to hash it with
179 * base page size 4k.
180 */
181 if (psize == MMU_PAGE_4K)
182 new_pmd |= H_PAGE_COMBO;
183 /*
184 * The hpte valid is stored in the pgtable whose address is in the
185 * second half of the PMD. Order this against clearing of the busy bit in
186 * huge pmd.
187 */
188 smp_wmb();
189 *pmdp = __pmd(new_pmd & ~H_PAGE_BUSY);
190 return 0;
191 }