xref: /linux/arch/powerpc/include/asm/book3s/64/radix.h (revision 32786fdc9506aeba98278c1844d4bfb766863832)
1 #ifndef _ASM_POWERPC_PGTABLE_RADIX_H
2 #define _ASM_POWERPC_PGTABLE_RADIX_H
3 
4 #ifndef __ASSEMBLY__
5 #include <asm/cmpxchg.h>
6 #endif
7 
8 #ifdef CONFIG_PPC_64K_PAGES
9 #include <asm/book3s/64/radix-64k.h>
10 #else
11 #include <asm/book3s/64/radix-4k.h>
12 #endif
13 
14 #ifndef __ASSEMBLY__
15 #include <asm/book3s/64/tlbflush-radix.h>
16 #include <asm/cpu_has_feature.h>
17 #endif
18 
19 /* An empty PTE can still have a R or C writeback */
20 #define RADIX_PTE_NONE_MASK		(_PAGE_DIRTY | _PAGE_ACCESSED)
21 
22 /* Bits to set in a RPMD/RPUD/RPGD */
23 #define RADIX_PMD_VAL_BITS		(0x8000000000000000UL | RADIX_PTE_INDEX_SIZE)
24 #define RADIX_PUD_VAL_BITS		(0x8000000000000000UL | RADIX_PMD_INDEX_SIZE)
25 #define RADIX_PGD_VAL_BITS		(0x8000000000000000UL | RADIX_PUD_INDEX_SIZE)
26 
27 /* Don't have anything in the reserved bits and leaf bits */
28 #define RADIX_PMD_BAD_BITS		0x60000000000000e0UL
29 #define RADIX_PUD_BAD_BITS		0x60000000000000e0UL
30 #define RADIX_PGD_BAD_BITS		0x60000000000000e0UL
31 
32 /*
33  * Size of EA range mapped by our pagetables.
34  */
35 #define RADIX_PGTABLE_EADDR_SIZE (RADIX_PTE_INDEX_SIZE + RADIX_PMD_INDEX_SIZE +	\
36 			      RADIX_PUD_INDEX_SIZE + RADIX_PGD_INDEX_SIZE + PAGE_SHIFT)
37 #define RADIX_PGTABLE_RANGE (ASM_CONST(1) << RADIX_PGTABLE_EADDR_SIZE)
38 
39 /*
40  * We support 52 bit address space, Use top bit for kernel
41  * virtual mapping. Also make sure kernel fit in the top
42  * quadrant.
43  *
44  *           +------------------+
45  *           +------------------+  Kernel virtual map (0xc008000000000000)
46  *           |                  |
47  *           |                  |
48  *           |                  |
49  * 0b11......+------------------+  Kernel linear map (0xc....)
50  *           |                  |
51  *           |     2 quadrant   |
52  *           |                  |
53  * 0b10......+------------------+
54  *           |                  |
55  *           |    1 quadrant    |
56  *           |                  |
57  * 0b01......+------------------+
58  *           |                  |
59  *           |    0 quadrant    |
60  *           |                  |
61  * 0b00......+------------------+
62  *
63  *
64  * 3rd quadrant expanded:
65  * +------------------------------+
66  * |                              |
67  * |                              |
68  * |                              |
69  * +------------------------------+  Kernel IO map end (0xc010000000000000)
70  * |                              |
71  * |                              |
72  * |      1/2 of virtual map      |
73  * |                              |
74  * |                              |
75  * +------------------------------+  Kernel IO map start
76  * |                              |
77  * |      1/4 of virtual map      |
78  * |                              |
79  * +------------------------------+  Kernel vmemap start
80  * |                              |
81  * |     1/4 of virtual map       |
82  * |                              |
83  * +------------------------------+  Kernel virt start (0xc008000000000000)
84  * |                              |
85  * |                              |
86  * |                              |
87  * +------------------------------+  Kernel linear (0xc.....)
88  */
89 
90 #define RADIX_KERN_VIRT_START ASM_CONST(0xc008000000000000)
91 #define RADIX_KERN_VIRT_SIZE  ASM_CONST(0x0008000000000000)
92 
93 /*
94  * The vmalloc space starts at the beginning of that region, and
95  * occupies a quarter of it on radix config.
96  * (we keep a quarter for the virtual memmap)
97  */
98 #define RADIX_VMALLOC_START	RADIX_KERN_VIRT_START
99 #define RADIX_VMALLOC_SIZE	(RADIX_KERN_VIRT_SIZE >> 2)
100 #define RADIX_VMALLOC_END	(RADIX_VMALLOC_START + RADIX_VMALLOC_SIZE)
101 /*
102  * Defines the address of the vmemap area, in its own region on
103  * hash table CPUs.
104  */
105 #define RADIX_VMEMMAP_BASE		(RADIX_VMALLOC_END)
106 
107 #ifndef __ASSEMBLY__
108 #define RADIX_PTE_TABLE_SIZE	(sizeof(pte_t) << RADIX_PTE_INDEX_SIZE)
109 #define RADIX_PMD_TABLE_SIZE	(sizeof(pmd_t) << RADIX_PMD_INDEX_SIZE)
110 #define RADIX_PUD_TABLE_SIZE	(sizeof(pud_t) << RADIX_PUD_INDEX_SIZE)
111 #define RADIX_PGD_TABLE_SIZE	(sizeof(pgd_t) << RADIX_PGD_INDEX_SIZE)
112 
113 static inline unsigned long __radix_pte_update(pte_t *ptep, unsigned long clr,
114 					       unsigned long set)
115 {
116 	pte_t pte;
117 	unsigned long old_pte, new_pte;
118 
119 	do {
120 		pte = READ_ONCE(*ptep);
121 		old_pte = pte_val(pte);
122 		new_pte = (old_pte | set) & ~clr;
123 
124 	} while (!pte_xchg(ptep, __pte(old_pte), __pte(new_pte)));
125 
126 	return old_pte;
127 }
128 
129 
130 static inline unsigned long radix__pte_update(struct mm_struct *mm,
131 					unsigned long addr,
132 					pte_t *ptep, unsigned long clr,
133 					unsigned long set,
134 					int huge)
135 {
136 	unsigned long old_pte;
137 
138 	if (cpu_has_feature(CPU_FTR_POWER9_DD1)) {
139 
140 		unsigned long new_pte;
141 
142 		old_pte = __radix_pte_update(ptep, ~0, 0);
143 		/*
144 		 * new value of pte
145 		 */
146 		new_pte = (old_pte | set) & ~clr;
147 		/*
148 		 * If we are trying to clear the pte, we can skip
149 		 * the below sequence and batch the tlb flush. The
150 		 * tlb flush batching is done by mmu gather code
151 		 */
152 		if (new_pte) {
153 			asm volatile("ptesync" : : : "memory");
154 			radix__flush_tlb_pte_p9_dd1(old_pte, mm, addr);
155 			__radix_pte_update(ptep, 0, new_pte);
156 		}
157 	} else
158 		old_pte = __radix_pte_update(ptep, clr, set);
159 	asm volatile("ptesync" : : : "memory");
160 	if (!huge)
161 		assert_pte_locked(mm, addr);
162 
163 	return old_pte;
164 }
165 
166 /*
167  * Set the dirty and/or accessed bits atomically in a linux PTE, this
168  * function doesn't need to invalidate tlb.
169  */
170 static inline void radix__ptep_set_access_flags(struct mm_struct *mm,
171 						pte_t *ptep, pte_t entry,
172 						unsigned long address)
173 {
174 
175 	unsigned long set = pte_val(entry) & (_PAGE_DIRTY | _PAGE_ACCESSED |
176 					      _PAGE_RW | _PAGE_EXEC);
177 
178 	if (cpu_has_feature(CPU_FTR_POWER9_DD1)) {
179 
180 		unsigned long old_pte, new_pte;
181 
182 		old_pte = __radix_pte_update(ptep, ~0, 0);
183 		asm volatile("ptesync" : : : "memory");
184 		/*
185 		 * new value of pte
186 		 */
187 		new_pte = old_pte | set;
188 		radix__flush_tlb_pte_p9_dd1(old_pte, mm, address);
189 		__radix_pte_update(ptep, 0, new_pte);
190 	} else
191 		__radix_pte_update(ptep, 0, set);
192 	asm volatile("ptesync" : : : "memory");
193 }
194 
195 static inline int radix__pte_same(pte_t pte_a, pte_t pte_b)
196 {
197 	return ((pte_raw(pte_a) ^ pte_raw(pte_b)) == 0);
198 }
199 
200 static inline int radix__pte_none(pte_t pte)
201 {
202 	return (pte_val(pte) & ~RADIX_PTE_NONE_MASK) == 0;
203 }
204 
205 static inline void radix__set_pte_at(struct mm_struct *mm, unsigned long addr,
206 				 pte_t *ptep, pte_t pte, int percpu)
207 {
208 	*ptep = pte;
209 	asm volatile("ptesync" : : : "memory");
210 }
211 
212 static inline int radix__pmd_bad(pmd_t pmd)
213 {
214 	return !!(pmd_val(pmd) & RADIX_PMD_BAD_BITS);
215 }
216 
217 static inline int radix__pmd_same(pmd_t pmd_a, pmd_t pmd_b)
218 {
219 	return ((pmd_raw(pmd_a) ^ pmd_raw(pmd_b)) == 0);
220 }
221 
222 static inline int radix__pud_bad(pud_t pud)
223 {
224 	return !!(pud_val(pud) & RADIX_PUD_BAD_BITS);
225 }
226 
227 
228 static inline int radix__pgd_bad(pgd_t pgd)
229 {
230 	return !!(pgd_val(pgd) & RADIX_PGD_BAD_BITS);
231 }
232 
233 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
234 
235 static inline int radix__pmd_trans_huge(pmd_t pmd)
236 {
237 	return !!(pmd_val(pmd) & _PAGE_PTE);
238 }
239 
240 static inline pmd_t radix__pmd_mkhuge(pmd_t pmd)
241 {
242 	if (cpu_has_feature(CPU_FTR_POWER9_DD1))
243 		return __pmd(pmd_val(pmd) | _PAGE_PTE | _PAGE_LARGE);
244 	return __pmd(pmd_val(pmd) | _PAGE_PTE);
245 }
246 static inline void radix__pmdp_huge_split_prepare(struct vm_area_struct *vma,
247 					    unsigned long address, pmd_t *pmdp)
248 {
249 	/* Nothing to do for radix. */
250 	return;
251 }
252 
253 extern unsigned long radix__pmd_hugepage_update(struct mm_struct *mm, unsigned long addr,
254 					  pmd_t *pmdp, unsigned long clr,
255 					  unsigned long set);
256 extern pmd_t radix__pmdp_collapse_flush(struct vm_area_struct *vma,
257 				  unsigned long address, pmd_t *pmdp);
258 extern void radix__pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
259 					pgtable_t pgtable);
260 extern pgtable_t radix__pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
261 extern pmd_t radix__pmdp_huge_get_and_clear(struct mm_struct *mm,
262 				      unsigned long addr, pmd_t *pmdp);
263 extern int radix__has_transparent_hugepage(void);
264 #endif
265 
266 extern int __meminit radix__vmemmap_create_mapping(unsigned long start,
267 					     unsigned long page_size,
268 					     unsigned long phys);
269 extern void radix__vmemmap_remove_mapping(unsigned long start,
270 				    unsigned long page_size);
271 
272 extern int radix__map_kernel_page(unsigned long ea, unsigned long pa,
273 				 pgprot_t flags, unsigned int psz);
274 
275 static inline unsigned long radix__get_tree_size(void)
276 {
277 	unsigned long rts_field;
278 	/*
279 	 * We support 52 bits, hence:
280 	 *  DD1    52-28 = 24, 0b11000
281 	 *  Others 52-31 = 21, 0b10101
282 	 * RTS encoding details
283 	 * bits 0 - 3 of rts -> bits 6 - 8 unsigned long
284 	 * bits 4 - 5 of rts -> bits 62 - 63 of unsigned long
285 	 */
286 	if (cpu_has_feature(CPU_FTR_POWER9_DD1))
287 		rts_field = (0x3UL << 61);
288 	else {
289 		rts_field = (0x5UL << 5); /* 6 - 8 bits */
290 		rts_field |= (0x2UL << 61);
291 	}
292 	return rts_field;
293 }
294 #endif /* __ASSEMBLY__ */
295 #endif
296