1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_POWERPC_PGTABLE_RADIX_H
3 #define _ASM_POWERPC_PGTABLE_RADIX_H
4 
5 #include <asm/asm-const.h>
6 
7 #ifndef __ASSEMBLY__
8 #include <asm/cmpxchg.h>
9 #endif
10 
11 #ifdef CONFIG_PPC_64K_PAGES
12 #include <asm/book3s/64/radix-64k.h>
13 #else
14 #include <asm/book3s/64/radix-4k.h>
15 #endif
16 
17 #ifndef __ASSEMBLY__
18 #include <asm/book3s/64/tlbflush-radix.h>
19 #include <asm/cpu_has_feature.h>
20 #endif
21 
22 /* An empty PTE can still have a R or C writeback */
23 #define RADIX_PTE_NONE_MASK		(_PAGE_DIRTY | _PAGE_ACCESSED)
24 
25 /* Bits to set in a RPMD/RPUD/RPGD */
26 #define RADIX_PMD_VAL_BITS		(0x8000000000000000UL | RADIX_PTE_INDEX_SIZE)
27 #define RADIX_PUD_VAL_BITS		(0x8000000000000000UL | RADIX_PMD_INDEX_SIZE)
28 #define RADIX_PGD_VAL_BITS		(0x8000000000000000UL | RADIX_PUD_INDEX_SIZE)
29 
30 /* Don't have anything in the reserved bits and leaf bits */
31 #define RADIX_PMD_BAD_BITS		0x60000000000000e0UL
32 #define RADIX_PUD_BAD_BITS		0x60000000000000e0UL
33 #define RADIX_P4D_BAD_BITS		0x60000000000000e0UL
34 
35 #define RADIX_PMD_SHIFT		(PAGE_SHIFT + RADIX_PTE_INDEX_SIZE)
36 #define RADIX_PUD_SHIFT		(RADIX_PMD_SHIFT + RADIX_PMD_INDEX_SIZE)
37 #define RADIX_PGD_SHIFT		(RADIX_PUD_SHIFT + RADIX_PUD_INDEX_SIZE)
38 
39 #define R_PTRS_PER_PTE		(1 << RADIX_PTE_INDEX_SIZE)
40 #define R_PTRS_PER_PMD		(1 << RADIX_PMD_INDEX_SIZE)
41 #define R_PTRS_PER_PUD		(1 << RADIX_PUD_INDEX_SIZE)
42 
43 /*
44  * Size of EA range mapped by our pagetables.
45  */
46 #define RADIX_PGTABLE_EADDR_SIZE (RADIX_PTE_INDEX_SIZE + RADIX_PMD_INDEX_SIZE +	\
47 			      RADIX_PUD_INDEX_SIZE + RADIX_PGD_INDEX_SIZE + PAGE_SHIFT)
48 #define RADIX_PGTABLE_RANGE (ASM_CONST(1) << RADIX_PGTABLE_EADDR_SIZE)
49 
50 /*
51  * We support 52 bit address space, Use top bit for kernel
52  * virtual mapping. Also make sure kernel fit in the top
53  * quadrant.
54  *
55  *           +------------------+
56  *           +------------------+  Kernel virtual map (0xc008000000000000)
57  *           |                  |
58  *           |                  |
59  *           |                  |
60  * 0b11......+------------------+  Kernel linear map (0xc....)
61  *           |                  |
62  *           |     2 quadrant   |
63  *           |                  |
64  * 0b10......+------------------+
65  *           |                  |
66  *           |    1 quadrant    |
67  *           |                  |
68  * 0b01......+------------------+
69  *           |                  |
70  *           |    0 quadrant    |
71  *           |                  |
72  * 0b00......+------------------+
73  *
74  *
75  * 3rd quadrant expanded:
76  * +------------------------------+  Highest address (0xc010000000000000)
77  * +------------------------------+  KASAN shadow end (0xc00fc00000000000)
78  * |                              |
79  * |                              |
80  * +------------------------------+  Kernel vmemmap end/shadow start (0xc00e000000000000)
81  * |                              |
82  * |           512TB		  |
83  * |                              |
84  * +------------------------------+  Kernel IO map end/vmemap start
85  * |                              |
86  * |           512TB		  |
87  * |                              |
88  * +------------------------------+  Kernel vmap end/ IO map start
89  * |                              |
90  * |           512TB		  |
91  * |                              |
92  * +------------------------------+  Kernel virt start (0xc008000000000000)
93  * |                              |
94  * |                              |
95  * |                              |
96  * +------------------------------+  Kernel linear (0xc.....)
97  */
98 
99 /* For the sizes of the shadow area, see kasan.h */
100 
101 /*
102  * If we store section details in page->flags we can't increase the MAX_PHYSMEM_BITS
103  * if we increase SECTIONS_WIDTH we will not store node details in page->flags and
104  * page_to_nid does a page->section->node lookup
105  * Hence only increase for VMEMMAP. Further depending on SPARSEMEM_EXTREME reduce
106  * memory requirements with large number of sections.
107  * 51 bits is the max physical real address on POWER9
108  */
109 
110 #if defined(CONFIG_SPARSEMEM_VMEMMAP) && defined(CONFIG_SPARSEMEM_EXTREME)
111 #define R_MAX_PHYSMEM_BITS	51
112 #else
113 #define R_MAX_PHYSMEM_BITS	46
114 #endif
115 
116 #define RADIX_KERN_VIRT_START	ASM_CONST(0xc008000000000000)
117 /*
118  * 49 =  MAX_EA_BITS_PER_CONTEXT (hash specific). To make sure we pick
119  * the same value as hash.
120  */
121 #define RADIX_KERN_MAP_SIZE	(1UL << 49)
122 
123 #define RADIX_VMALLOC_START	RADIX_KERN_VIRT_START
124 #define RADIX_VMALLOC_SIZE	RADIX_KERN_MAP_SIZE
125 #define RADIX_VMALLOC_END	(RADIX_VMALLOC_START + RADIX_VMALLOC_SIZE)
126 
127 #define RADIX_KERN_IO_START	RADIX_VMALLOC_END
128 #define RADIX_KERN_IO_SIZE	RADIX_KERN_MAP_SIZE
129 #define RADIX_KERN_IO_END	(RADIX_KERN_IO_START + RADIX_KERN_IO_SIZE)
130 
131 #define RADIX_VMEMMAP_START	RADIX_KERN_IO_END
132 #define RADIX_VMEMMAP_SIZE	RADIX_KERN_MAP_SIZE
133 #define RADIX_VMEMMAP_END	(RADIX_VMEMMAP_START + RADIX_VMEMMAP_SIZE)
134 
135 #ifndef __ASSEMBLY__
136 #define RADIX_PTE_TABLE_SIZE	(sizeof(pte_t) << RADIX_PTE_INDEX_SIZE)
137 #define RADIX_PMD_TABLE_SIZE	(sizeof(pmd_t) << RADIX_PMD_INDEX_SIZE)
138 #define RADIX_PUD_TABLE_SIZE	(sizeof(pud_t) << RADIX_PUD_INDEX_SIZE)
139 #define RADIX_PGD_TABLE_SIZE	(sizeof(pgd_t) << RADIX_PGD_INDEX_SIZE)
140 
141 #ifdef CONFIG_STRICT_KERNEL_RWX
142 extern void radix__mark_rodata_ro(void);
143 extern void radix__mark_initmem_nx(void);
144 #endif
145 
146 extern void radix__ptep_set_access_flags(struct vm_area_struct *vma, pte_t *ptep,
147 					 pte_t entry, unsigned long address,
148 					 int psize);
149 
150 extern void radix__ptep_modify_prot_commit(struct vm_area_struct *vma,
151 					   unsigned long addr, pte_t *ptep,
152 					   pte_t old_pte, pte_t pte);
153 
154 static inline unsigned long __radix_pte_update(pte_t *ptep, unsigned long clr,
155 					       unsigned long set)
156 {
157 	__be64 old_be, tmp_be;
158 
159 	__asm__ __volatile__(
160 	"1:	ldarx	%0,0,%3		# pte_update\n"
161 	"	andc	%1,%0,%5	\n"
162 	"	or	%1,%1,%4	\n"
163 	"	stdcx.	%1,0,%3		\n"
164 	"	bne-	1b"
165 	: "=&r" (old_be), "=&r" (tmp_be), "=m" (*ptep)
166 	: "r" (ptep), "r" (cpu_to_be64(set)), "r" (cpu_to_be64(clr))
167 	: "cc" );
168 
169 	return be64_to_cpu(old_be);
170 }
171 
172 static inline unsigned long radix__pte_update(struct mm_struct *mm,
173 					unsigned long addr,
174 					pte_t *ptep, unsigned long clr,
175 					unsigned long set,
176 					int huge)
177 {
178 	unsigned long old_pte;
179 
180 	old_pte = __radix_pte_update(ptep, clr, set);
181 	if (!huge)
182 		assert_pte_locked(mm, addr);
183 
184 	return old_pte;
185 }
186 
187 static inline pte_t radix__ptep_get_and_clear_full(struct mm_struct *mm,
188 						   unsigned long addr,
189 						   pte_t *ptep, int full)
190 {
191 	unsigned long old_pte;
192 
193 	if (full) {
194 		old_pte = pte_val(*ptep);
195 		*ptep = __pte(0);
196 	} else
197 		old_pte = radix__pte_update(mm, addr, ptep, ~0ul, 0, 0);
198 
199 	return __pte(old_pte);
200 }
201 
202 static inline int radix__pte_same(pte_t pte_a, pte_t pte_b)
203 {
204 	return ((pte_raw(pte_a) ^ pte_raw(pte_b)) == 0);
205 }
206 
207 static inline int radix__pte_none(pte_t pte)
208 {
209 	return (pte_val(pte) & ~RADIX_PTE_NONE_MASK) == 0;
210 }
211 
212 static inline void radix__set_pte_at(struct mm_struct *mm, unsigned long addr,
213 				 pte_t *ptep, pte_t pte, int percpu)
214 {
215 	*ptep = pte;
216 
217 	/*
218 	 * The architecture suggests a ptesync after setting the pte, which
219 	 * orders the store that updates the pte with subsequent page table
220 	 * walk accesses which may load the pte. Without this it may be
221 	 * possible for a subsequent access to result in spurious fault.
222 	 *
223 	 * This is not necessary for correctness, because a spurious fault
224 	 * is tolerated by the page fault handler, and this store will
225 	 * eventually be seen. In testing, there was no noticable increase
226 	 * in user faults on POWER9. Avoiding ptesync here is a significant
227 	 * win for things like fork. If a future microarchitecture benefits
228 	 * from ptesync, it should probably go into update_mmu_cache, rather
229 	 * than set_pte_at (which is used to set ptes unrelated to faults).
230 	 *
231 	 * Spurious faults from the kernel memory are not tolerated, so there
232 	 * is a ptesync in flush_cache_vmap, and __map_kernel_page() follows
233 	 * the pte update sequence from ISA Book III 6.10 Translation Table
234 	 * Update Synchronization Requirements.
235 	 */
236 }
237 
238 static inline int radix__pmd_bad(pmd_t pmd)
239 {
240 	return !!(pmd_val(pmd) & RADIX_PMD_BAD_BITS);
241 }
242 
243 static inline int radix__pmd_same(pmd_t pmd_a, pmd_t pmd_b)
244 {
245 	return ((pmd_raw(pmd_a) ^ pmd_raw(pmd_b)) == 0);
246 }
247 
248 static inline int radix__pud_bad(pud_t pud)
249 {
250 	return !!(pud_val(pud) & RADIX_PUD_BAD_BITS);
251 }
252 
253 static inline int radix__pud_same(pud_t pud_a, pud_t pud_b)
254 {
255 	return ((pud_raw(pud_a) ^ pud_raw(pud_b)) == 0);
256 }
257 
258 static inline int radix__p4d_bad(p4d_t p4d)
259 {
260 	return !!(p4d_val(p4d) & RADIX_P4D_BAD_BITS);
261 }
262 
263 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
264 
265 static inline int radix__pmd_trans_huge(pmd_t pmd)
266 {
267 	return (pmd_val(pmd) & (_PAGE_PTE | _PAGE_DEVMAP)) == _PAGE_PTE;
268 }
269 
270 static inline pmd_t radix__pmd_mkhuge(pmd_t pmd)
271 {
272 	return __pmd(pmd_val(pmd) | _PAGE_PTE);
273 }
274 
275 static inline int radix__pud_trans_huge(pud_t pud)
276 {
277 	return (pud_val(pud) & (_PAGE_PTE | _PAGE_DEVMAP)) == _PAGE_PTE;
278 }
279 
280 static inline pud_t radix__pud_mkhuge(pud_t pud)
281 {
282 	return __pud(pud_val(pud) | _PAGE_PTE);
283 }
284 
285 extern unsigned long radix__pmd_hugepage_update(struct mm_struct *mm, unsigned long addr,
286 					  pmd_t *pmdp, unsigned long clr,
287 					  unsigned long set);
288 extern unsigned long radix__pud_hugepage_update(struct mm_struct *mm, unsigned long addr,
289 						pud_t *pudp, unsigned long clr,
290 						unsigned long set);
291 extern pmd_t radix__pmdp_collapse_flush(struct vm_area_struct *vma,
292 				  unsigned long address, pmd_t *pmdp);
293 extern void radix__pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
294 					pgtable_t pgtable);
295 extern pgtable_t radix__pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
296 extern pmd_t radix__pmdp_huge_get_and_clear(struct mm_struct *mm,
297 				      unsigned long addr, pmd_t *pmdp);
298 pud_t radix__pudp_huge_get_and_clear(struct mm_struct *mm,
299 				     unsigned long addr, pud_t *pudp);
300 
301 static inline int radix__has_transparent_hugepage(void)
302 {
303 	/* For radix 2M at PMD level means thp */
304 	if (mmu_psize_defs[MMU_PAGE_2M].shift == PMD_SHIFT)
305 		return 1;
306 	return 0;
307 }
308 
309 static inline int radix__has_transparent_pud_hugepage(void)
310 {
311 	/* For radix 1G at PUD level means pud hugepage support */
312 	if (mmu_psize_defs[MMU_PAGE_1G].shift == PUD_SHIFT)
313 		return 1;
314 	return 0;
315 }
316 #endif
317 
318 static inline pmd_t radix__pmd_mkdevmap(pmd_t pmd)
319 {
320 	return __pmd(pmd_val(pmd) | (_PAGE_PTE | _PAGE_DEVMAP));
321 }
322 
323 static inline pud_t radix__pud_mkdevmap(pud_t pud)
324 {
325 	return __pud(pud_val(pud) | (_PAGE_PTE | _PAGE_DEVMAP));
326 }
327 
328 struct vmem_altmap;
329 struct dev_pagemap;
330 extern int __meminit radix__vmemmap_create_mapping(unsigned long start,
331 					     unsigned long page_size,
332 					     unsigned long phys);
333 int __meminit radix__vmemmap_populate(unsigned long start, unsigned long end,
334 				      int node, struct vmem_altmap *altmap);
335 void __ref radix__vmemmap_free(unsigned long start, unsigned long end,
336 			       struct vmem_altmap *altmap);
337 extern void radix__vmemmap_remove_mapping(unsigned long start,
338 				    unsigned long page_size);
339 
340 extern int radix__map_kernel_page(unsigned long ea, unsigned long pa,
341 				 pgprot_t flags, unsigned int psz);
342 
343 static inline unsigned long radix__get_tree_size(void)
344 {
345 	unsigned long rts_field;
346 	/*
347 	 * We support 52 bits, hence:
348 	 * bits 52 - 31 = 21, 0b10101
349 	 * RTS encoding details
350 	 * bits 0 - 3 of rts -> bits 6 - 8 unsigned long
351 	 * bits 4 - 5 of rts -> bits 62 - 63 of unsigned long
352 	 */
353 	rts_field = (0x5UL << 5); /* 6 - 8 bits */
354 	rts_field |= (0x2UL << 61);
355 
356 	return rts_field;
357 }
358 
359 #ifdef CONFIG_MEMORY_HOTPLUG
360 int radix__create_section_mapping(unsigned long start, unsigned long end,
361 				  int nid, pgprot_t prot);
362 int radix__remove_section_mapping(unsigned long start, unsigned long end);
363 #endif /* CONFIG_MEMORY_HOTPLUG */
364 
365 void radix__kernel_map_pages(struct page *page, int numpages, int enable);
366 
367 #ifdef CONFIG_ARCH_WANT_OPTIMIZE_DAX_VMEMMAP
368 #define vmemmap_can_optimize vmemmap_can_optimize
369 bool vmemmap_can_optimize(struct vmem_altmap *altmap, struct dev_pagemap *pgmap);
370 #endif
371 
372 #define vmemmap_populate_compound_pages vmemmap_populate_compound_pages
373 int __meminit vmemmap_populate_compound_pages(unsigned long start_pfn,
374 					      unsigned long start,
375 					      unsigned long end, int node,
376 					      struct dev_pagemap *pgmap);
377 #endif /* __ASSEMBLY__ */
378 #endif
379