1 #ifndef _ASM_POWERPC_BOOK3S_64_HASH_64K_H
2 #define _ASM_POWERPC_BOOK3S_64_HASH_64K_H
3 
4 #include <asm-generic/pgtable-nopud.h>
5 
6 #define PTE_INDEX_SIZE  8
7 #define PMD_INDEX_SIZE  10
8 #define PUD_INDEX_SIZE	0
9 #define PGD_INDEX_SIZE  12
10 
11 #define PTRS_PER_PTE	(1 << PTE_INDEX_SIZE)
12 #define PTRS_PER_PMD	(1 << PMD_INDEX_SIZE)
13 #define PTRS_PER_PGD	(1 << PGD_INDEX_SIZE)
14 
15 /* With 4k base page size, hugepage PTEs go at the PMD level */
16 #define MIN_HUGEPTE_SHIFT	PAGE_SHIFT
17 
18 /* PMD_SHIFT determines what a second-level page table entry can map */
19 #define PMD_SHIFT	(PAGE_SHIFT + PTE_INDEX_SIZE)
20 #define PMD_SIZE	(1UL << PMD_SHIFT)
21 #define PMD_MASK	(~(PMD_SIZE-1))
22 
23 /* PGDIR_SHIFT determines what a third-level page table entry can map */
24 #define PGDIR_SHIFT	(PMD_SHIFT + PMD_INDEX_SIZE)
25 #define PGDIR_SIZE	(1UL << PGDIR_SHIFT)
26 #define PGDIR_MASK	(~(PGDIR_SIZE-1))
27 
28 #define _PAGE_COMBO	0x00040000 /* this is a combo 4k page */
29 #define _PAGE_4K_PFN	0x00080000 /* PFN is for a single 4k page */
30 /*
31  * Used to track subpage group valid if _PAGE_COMBO is set
32  * This overloads _PAGE_F_GIX and _PAGE_F_SECOND
33  */
34 #define _PAGE_COMBO_VALID	(_PAGE_F_GIX | _PAGE_F_SECOND)
35 
36 /* PTE flags to conserve for HPTE identification */
37 #define _PAGE_HPTEFLAGS (_PAGE_BUSY | _PAGE_F_SECOND | \
38 			 _PAGE_F_GIX | _PAGE_HASHPTE | _PAGE_COMBO)
39 
40 /* Shift to put page number into pte.
41  *
42  * That gives us a max RPN of 34 bits, which means a max of 50 bits
43  * of addressable physical space, or 46 bits for the special 4k PFNs.
44  */
45 #define PTE_RPN_SHIFT	(30)
46 /*
47  * we support 16 fragments per PTE page of 64K size.
48  */
49 #define PTE_FRAG_NR	16
50 /*
51  * We use a 2K PTE page fragment and another 2K for storing
52  * real_pte_t hash index
53  */
54 #define PTE_FRAG_SIZE_SHIFT  12
55 #define PTE_FRAG_SIZE (1UL << PTE_FRAG_SIZE_SHIFT)
56 
57 /*
58  * Bits to mask out from a PMD to get to the PTE page
59  * PMDs point to PTE table fragments which are PTE_FRAG_SIZE aligned.
60  */
61 #define PMD_MASKED_BITS		(PTE_FRAG_SIZE - 1)
62 /* Bits to mask out from a PGD/PUD to get to the PMD page */
63 #define PUD_MASKED_BITS		0x1ff
64 
65 #ifndef __ASSEMBLY__
66 
67 /*
68  * With 64K pages on hash table, we have a special PTE format that
69  * uses a second "half" of the page table to encode sub-page information
70  * in order to deal with 64K made of 4K HW pages. Thus we override the
71  * generic accessors and iterators here
72  */
73 #define __real_pte __real_pte
74 static inline real_pte_t __real_pte(pte_t pte, pte_t *ptep)
75 {
76 	real_pte_t rpte;
77 	unsigned long *hidxp;
78 
79 	rpte.pte = pte;
80 	rpte.hidx = 0;
81 	if (pte_val(pte) & _PAGE_COMBO) {
82 		/*
83 		 * Make sure we order the hidx load against the _PAGE_COMBO
84 		 * check. The store side ordering is done in __hash_page_4K
85 		 */
86 		smp_rmb();
87 		hidxp = (unsigned long *)(ptep + PTRS_PER_PTE);
88 		rpte.hidx = *hidxp;
89 	}
90 	return rpte;
91 }
92 
93 static inline unsigned long __rpte_to_hidx(real_pte_t rpte, unsigned long index)
94 {
95 	if ((pte_val(rpte.pte) & _PAGE_COMBO))
96 		return (rpte.hidx >> (index<<2)) & 0xf;
97 	return (pte_val(rpte.pte) >> _PAGE_F_GIX_SHIFT) & 0xf;
98 }
99 
100 #define __rpte_to_pte(r)	((r).pte)
101 extern bool __rpte_sub_valid(real_pte_t rpte, unsigned long index);
102 /*
103  * Trick: we set __end to va + 64k, which happens works for
104  * a 16M page as well as we want only one iteration
105  */
106 #define pte_iterate_hashed_subpages(rpte, psize, vpn, index, shift)	\
107 	do {								\
108 		unsigned long __end = vpn + (1UL << (PAGE_SHIFT - VPN_SHIFT));	\
109 		unsigned __split = (psize == MMU_PAGE_4K ||		\
110 				    psize == MMU_PAGE_64K_AP);		\
111 		shift = mmu_psize_defs[psize].shift;			\
112 		for (index = 0; vpn < __end; index++,			\
113 			     vpn += (1L << (shift - VPN_SHIFT))) {	\
114 			if (!__split || __rpte_sub_valid(rpte, index))	\
115 				do {
116 
117 #define pte_iterate_hashed_end() } while(0); } } while(0)
118 
119 #define pte_pagesize_index(mm, addr, pte)	\
120 	(((pte) & _PAGE_COMBO)? MMU_PAGE_4K: MMU_PAGE_64K)
121 
122 #define remap_4k_pfn(vma, addr, pfn, prot)				\
123 	(WARN_ON(((pfn) >= (1UL << (64 - PTE_RPN_SHIFT)))) ? -EINVAL :	\
124 		remap_pfn_range((vma), (addr), (pfn), PAGE_SIZE,	\
125 			__pgprot(pgprot_val((prot)) | _PAGE_4K_PFN)))
126 
127 #define PTE_TABLE_SIZE	PTE_FRAG_SIZE
128 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
129 #define PMD_TABLE_SIZE	((sizeof(pmd_t) << PMD_INDEX_SIZE) + (sizeof(unsigned long) << PMD_INDEX_SIZE))
130 #else
131 #define PMD_TABLE_SIZE	(sizeof(pmd_t) << PMD_INDEX_SIZE)
132 #endif
133 #define PGD_TABLE_SIZE	(sizeof(pgd_t) << PGD_INDEX_SIZE)
134 
135 #define pgd_pte(pgd)	(pud_pte(((pud_t){ pgd })))
136 #define pte_pgd(pte)	((pgd_t)pte_pud(pte))
137 
138 #ifdef CONFIG_HUGETLB_PAGE
139 /*
140  * We have PGD_INDEX_SIZ = 12 and PTE_INDEX_SIZE = 8, so that we can have
141  * 16GB hugepage pte in PGD and 16MB hugepage pte at PMD;
142  *
143  * Defined in such a way that we can optimize away code block at build time
144  * if CONFIG_HUGETLB_PAGE=n.
145  */
146 static inline int pmd_huge(pmd_t pmd)
147 {
148 	/*
149 	 * leaf pte for huge page
150 	 */
151 	return !!(pmd_val(pmd) & _PAGE_PTE);
152 }
153 
154 static inline int pud_huge(pud_t pud)
155 {
156 	/*
157 	 * leaf pte for huge page
158 	 */
159 	return !!(pud_val(pud) & _PAGE_PTE);
160 }
161 
162 static inline int pgd_huge(pgd_t pgd)
163 {
164 	/*
165 	 * leaf pte for huge page
166 	 */
167 	return !!(pgd_val(pgd) & _PAGE_PTE);
168 }
169 #define pgd_huge pgd_huge
170 
171 #ifdef CONFIG_DEBUG_VM
172 extern int hugepd_ok(hugepd_t hpd);
173 #define is_hugepd(hpd)               (hugepd_ok(hpd))
174 #else
175 /*
176  * With 64k page size, we have hugepage ptes in the pgd and pmd entries. We don't
177  * need to setup hugepage directory for them. Our pte and page directory format
178  * enable us to have this enabled.
179  */
180 static inline int hugepd_ok(hugepd_t hpd)
181 {
182 	return 0;
183 }
184 #define is_hugepd(pdep)			0
185 #endif /* CONFIG_DEBUG_VM */
186 
187 #endif /* CONFIG_HUGETLB_PAGE */
188 
189 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
190 extern unsigned long pmd_hugepage_update(struct mm_struct *mm,
191 					 unsigned long addr,
192 					 pmd_t *pmdp,
193 					 unsigned long clr,
194 					 unsigned long set);
195 static inline char *get_hpte_slot_array(pmd_t *pmdp)
196 {
197 	/*
198 	 * The hpte hindex is stored in the pgtable whose address is in the
199 	 * second half of the PMD
200 	 *
201 	 * Order this load with the test for pmd_trans_huge in the caller
202 	 */
203 	smp_rmb();
204 	return *(char **)(pmdp + PTRS_PER_PMD);
205 
206 
207 }
208 /*
209  * The linux hugepage PMD now include the pmd entries followed by the address
210  * to the stashed pgtable_t. The stashed pgtable_t contains the hpte bits.
211  * [ 1 bit secondary | 3 bit hidx | 1 bit valid | 000]. We use one byte per
212  * each HPTE entry. With 16MB hugepage and 64K HPTE we need 256 entries and
213  * with 4K HPTE we need 4096 entries. Both will fit in a 4K pgtable_t.
214  *
215  * The last three bits are intentionally left to zero. This memory location
216  * are also used as normal page PTE pointers. So if we have any pointers
217  * left around while we collapse a hugepage, we need to make sure
218  * _PAGE_PRESENT bit of that is zero when we look at them
219  */
220 static inline unsigned int hpte_valid(unsigned char *hpte_slot_array, int index)
221 {
222 	return (hpte_slot_array[index] >> 3) & 0x1;
223 }
224 
225 static inline unsigned int hpte_hash_index(unsigned char *hpte_slot_array,
226 					   int index)
227 {
228 	return hpte_slot_array[index] >> 4;
229 }
230 
231 static inline void mark_hpte_slot_valid(unsigned char *hpte_slot_array,
232 					unsigned int index, unsigned int hidx)
233 {
234 	hpte_slot_array[index] = hidx << 4 | 0x1 << 3;
235 }
236 
237 /*
238  *
239  * For core kernel code by design pmd_trans_huge is never run on any hugetlbfs
240  * page. The hugetlbfs page table walking and mangling paths are totally
241  * separated form the core VM paths and they're differentiated by
242  *  VM_HUGETLB being set on vm_flags well before any pmd_trans_huge could run.
243  *
244  * pmd_trans_huge() is defined as false at build time if
245  * CONFIG_TRANSPARENT_HUGEPAGE=n to optimize away code blocks at build
246  * time in such case.
247  *
248  * For ppc64 we need to differntiate from explicit hugepages from THP, because
249  * for THP we also track the subpage details at the pmd level. We don't do
250  * that for explicit huge pages.
251  *
252  */
253 static inline int pmd_trans_huge(pmd_t pmd)
254 {
255 	return !!((pmd_val(pmd) & (_PAGE_PTE | _PAGE_THP_HUGE)) ==
256 		  (_PAGE_PTE | _PAGE_THP_HUGE));
257 }
258 
259 static inline int pmd_large(pmd_t pmd)
260 {
261 	return !!(pmd_val(pmd) & _PAGE_PTE);
262 }
263 
264 static inline pmd_t pmd_mknotpresent(pmd_t pmd)
265 {
266 	return __pmd(pmd_val(pmd) & ~_PAGE_PRESENT);
267 }
268 
269 #define __HAVE_ARCH_PMD_SAME
270 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
271 {
272 	return (((pmd_val(pmd_a) ^ pmd_val(pmd_b)) & ~_PAGE_HPTEFLAGS) == 0);
273 }
274 
275 static inline int __pmdp_test_and_clear_young(struct mm_struct *mm,
276 					      unsigned long addr, pmd_t *pmdp)
277 {
278 	unsigned long old;
279 
280 	if ((pmd_val(*pmdp) & (_PAGE_ACCESSED | _PAGE_HASHPTE)) == 0)
281 		return 0;
282 	old = pmd_hugepage_update(mm, addr, pmdp, _PAGE_ACCESSED, 0);
283 	return ((old & _PAGE_ACCESSED) != 0);
284 }
285 
286 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
287 static inline void pmdp_set_wrprotect(struct mm_struct *mm, unsigned long addr,
288 				      pmd_t *pmdp)
289 {
290 
291 	if ((pmd_val(*pmdp) & _PAGE_RW) == 0)
292 		return;
293 
294 	pmd_hugepage_update(mm, addr, pmdp, _PAGE_RW, 0);
295 }
296 
297 #endif /*  CONFIG_TRANSPARENT_HUGEPAGE */
298 #endif	/* __ASSEMBLY__ */
299 
300 #endif /* _ASM_POWERPC_BOOK3S_64_HASH_64K_H */
301