xref: /openbmc/linux/arch/xtensa/include/asm/pgtable.h (revision fb799447)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * include/asm-xtensa/pgtable.h
4  *
5  * Copyright (C) 2001 - 2013 Tensilica Inc.
6  */
7 
8 #ifndef _XTENSA_PGTABLE_H
9 #define _XTENSA_PGTABLE_H
10 
11 #include <asm/page.h>
12 #include <asm/kmem_layout.h>
13 #include <asm-generic/pgtable-nopmd.h>
14 
15 /*
16  * We only use two ring levels, user and kernel space.
17  */
18 
19 #ifdef CONFIG_MMU
20 #define USER_RING		1	/* user ring level */
21 #else
22 #define USER_RING		0
23 #endif
24 #define KERNEL_RING		0	/* kernel ring level */
25 
26 /*
27  * The Xtensa architecture port of Linux has a two-level page table system,
28  * i.e. the logical three-level Linux page table layout is folded.
29  * Each task has the following memory page tables:
30  *
31  *   PGD table (page directory), ie. 3rd-level page table:
32  *	One page (4 kB) of 1024 (PTRS_PER_PGD) pointers to PTE tables
33  *	(Architectures that don't have the PMD folded point to the PMD tables)
34  *
35  *	The pointer to the PGD table for a given task can be retrieved from
36  *	the task structure (struct task_struct*) t, e.g. current():
37  *	  (t->mm ? t->mm : t->active_mm)->pgd
38  *
39  *   PMD tables (page middle-directory), ie. 2nd-level page tables:
40  *	Absent for the Xtensa architecture (folded, PTRS_PER_PMD == 1).
41  *
42  *   PTE tables (page table entry), ie. 1st-level page tables:
43  *	One page (4 kB) of 1024 (PTRS_PER_PTE) PTEs with a special PTE
44  *	invalid_pte_table for absent mappings.
45  *
46  * The individual pages are 4 kB big with special pages for the empty_zero_page.
47  */
48 
49 #define PGDIR_SHIFT	22
50 #define PGDIR_SIZE	(1UL << PGDIR_SHIFT)
51 #define PGDIR_MASK	(~(PGDIR_SIZE-1))
52 
53 /*
54  * Entries per page directory level: we use two-level, so
55  * we don't really have any PMD directory physically.
56  */
57 #define PTRS_PER_PTE		1024
58 #define PTRS_PER_PTE_SHIFT	10
59 #define PTRS_PER_PGD		1024
60 #define USER_PTRS_PER_PGD	(TASK_SIZE/PGDIR_SIZE)
61 #define FIRST_USER_PGD_NR	(FIRST_USER_ADDRESS >> PGDIR_SHIFT)
62 
63 #ifdef CONFIG_MMU
64 /*
65  * Virtual memory area. We keep a distance to other memory regions to be
66  * on the safe side. We also use this area for cache aliasing.
67  */
68 #define VMALLOC_START		(XCHAL_KSEG_CACHED_VADDR - 0x10000000)
69 #define VMALLOC_END		(VMALLOC_START + 0x07FEFFFF)
70 #define TLBTEMP_BASE_1		(VMALLOC_START + 0x08000000)
71 #define TLBTEMP_BASE_2		(TLBTEMP_BASE_1 + DCACHE_WAY_SIZE)
72 #if 2 * DCACHE_WAY_SIZE > ICACHE_WAY_SIZE
73 #define TLBTEMP_SIZE		(2 * DCACHE_WAY_SIZE)
74 #else
75 #define TLBTEMP_SIZE		ICACHE_WAY_SIZE
76 #endif
77 
78 #else
79 
80 #define VMALLOC_START		__XTENSA_UL_CONST(0)
81 #define VMALLOC_END		__XTENSA_UL_CONST(0xffffffff)
82 
83 #endif
84 
85 /*
86  * For the Xtensa architecture, the PTE layout is as follows:
87  *
88  *		31------12  11  10-9   8-6  5-4  3-2  1-0
89  *		+-----------------------------------------+
90  *		|           |   Software   |   HARDWARE   |
91  *		|    PPN    |          ADW | RI |Attribute|
92  *		+-----------------------------------------+
93  *   pte_none	|             MBZ          | 01 | 11 | 00 |
94  *		+-----------------------------------------+
95  *   present	|    PPN    | 0 | 00 | ADW | RI | CA | wx |
96  *		+- - - - - - - - - - - - - - - - - - - - -+
97  *   (PAGE_NONE)|    PPN    | 0 | 00 | ADW | 01 | 11 | 11 |
98  *		+-----------------------------------------+
99  *   swap	|     index     |   type   | 01 | 11 | e0 |
100  *		+-----------------------------------------+
101  *
102  * For T1050 hardware and earlier the layout differs for present and (PAGE_NONE)
103  *		+-----------------------------------------+
104  *   present	|    PPN    | 0 | 00 | ADW | RI | CA | w1 |
105  *		+-----------------------------------------+
106  *   (PAGE_NONE)|    PPN    | 0 | 00 | ADW | 01 | 01 | 00 |
107  *		+-----------------------------------------+
108  *
109  *  Legend:
110  *   PPN        Physical Page Number
111  *   ADW	software: accessed (young) / dirty / writable
112  *   RI         ring (0=privileged, 1=user, 2 and 3 are unused)
113  *   CA		cache attribute: 00 bypass, 01 writeback, 10 writethrough
114  *		(11 is invalid and used to mark pages that are not present)
115  *   e		exclusive marker in swap PTEs
116  *   w		page is writable (hw)
117  *   x		page is executable (hw)
118  *   index      swap offset / PAGE_SIZE (bit 11-31: 21 bits -> 8 GB)
119  *		(note that the index is always non-zero)
120  *   type       swap type (5 bits -> 32 types)
121  *
122  *  Notes:
123  *   - (PROT_NONE) is a special case of 'present' but causes an exception for
124  *     any access (read, write, and execute).
125  *   - 'multihit-exception' has the highest priority of all MMU exceptions,
126  *     so the ring must be set to 'RING_USER' even for 'non-present' pages.
127  *   - on older hardware, the exectuable flag was not supported and
128  *     used as a 'valid' flag, so it needs to be always set.
129  *   - we need to keep track of certain flags in software (dirty and young)
130  *     to do this, we use write exceptions and have a separate software w-flag.
131  *   - attribute value 1101 (and 1111 on T1050 and earlier) is reserved
132  */
133 
134 #define _PAGE_ATTRIB_MASK	0xf
135 
136 #define _PAGE_HW_EXEC		(1<<0)	/* hardware: page is executable */
137 #define _PAGE_HW_WRITE		(1<<1)	/* hardware: page is writable */
138 
139 #define _PAGE_CA_BYPASS		(0<<2)	/* bypass, non-speculative */
140 #define _PAGE_CA_WB		(1<<2)	/* write-back */
141 #define _PAGE_CA_WT		(2<<2)	/* write-through */
142 #define _PAGE_CA_MASK		(3<<2)
143 #define _PAGE_CA_INVALID	(3<<2)
144 
145 /* We use invalid attribute values to distinguish special pte entries */
146 #if XCHAL_HW_VERSION_MAJOR < 2000
147 #define _PAGE_HW_VALID		0x01	/* older HW needed this bit set */
148 #define _PAGE_NONE		0x04
149 #else
150 #define _PAGE_HW_VALID		0x00
151 #define _PAGE_NONE		0x0f
152 #endif
153 
154 #define _PAGE_USER		(1<<4)	/* user access (ring=1) */
155 
156 /* Software */
157 #define _PAGE_WRITABLE_BIT	6
158 #define _PAGE_WRITABLE		(1<<6)	/* software: page writable */
159 #define _PAGE_DIRTY		(1<<7)	/* software: page dirty */
160 #define _PAGE_ACCESSED		(1<<8)	/* software: page accessed (read) */
161 
162 /* We borrow bit 1 to store the exclusive marker in swap PTEs. */
163 #define _PAGE_SWP_EXCLUSIVE	(1<<1)
164 
165 #ifdef CONFIG_MMU
166 
167 #define _PAGE_CHG_MASK	   (PAGE_MASK | _PAGE_ACCESSED | _PAGE_DIRTY)
168 #define _PAGE_PRESENT	   (_PAGE_HW_VALID | _PAGE_CA_WB | _PAGE_ACCESSED)
169 
170 #define PAGE_NONE	   __pgprot(_PAGE_NONE | _PAGE_USER)
171 #define PAGE_COPY	   __pgprot(_PAGE_PRESENT | _PAGE_USER)
172 #define PAGE_COPY_EXEC	   __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_HW_EXEC)
173 #define PAGE_READONLY	   __pgprot(_PAGE_PRESENT | _PAGE_USER)
174 #define PAGE_READONLY_EXEC __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_HW_EXEC)
175 #define PAGE_SHARED	   __pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_WRITABLE)
176 #define PAGE_SHARED_EXEC \
177 	__pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_WRITABLE | _PAGE_HW_EXEC)
178 #define PAGE_KERNEL	   __pgprot(_PAGE_PRESENT | _PAGE_HW_WRITE)
179 #define PAGE_KERNEL_RO	   __pgprot(_PAGE_PRESENT)
180 #define PAGE_KERNEL_EXEC   __pgprot(_PAGE_PRESENT|_PAGE_HW_WRITE|_PAGE_HW_EXEC)
181 
182 #if (DCACHE_WAY_SIZE > PAGE_SIZE)
183 # define _PAGE_DIRECTORY   (_PAGE_HW_VALID | _PAGE_ACCESSED | _PAGE_CA_BYPASS)
184 #else
185 # define _PAGE_DIRECTORY   (_PAGE_HW_VALID | _PAGE_ACCESSED | _PAGE_CA_WB)
186 #endif
187 
188 #else /* no mmu */
189 
190 # define _PAGE_CHG_MASK  (PAGE_MASK | _PAGE_ACCESSED | _PAGE_DIRTY)
191 # define PAGE_NONE       __pgprot(0)
192 # define PAGE_SHARED     __pgprot(0)
193 # define PAGE_COPY       __pgprot(0)
194 # define PAGE_READONLY   __pgprot(0)
195 # define PAGE_KERNEL     __pgprot(0)
196 
197 #endif
198 
199 /*
200  * On certain configurations of Xtensa MMUs (eg. the initial Linux config),
201  * the MMU can't do page protection for execute, and considers that the same as
202  * read.  Also, write permissions may imply read permissions.
203  * What follows is the closest we can get by reasonable means..
204  * See linux/mm/mmap.c for protection_map[] array that uses these definitions.
205  */
206 #ifndef __ASSEMBLY__
207 
208 #define pte_ERROR(e) \
209 	printk("%s:%d: bad pte %08lx.\n", __FILE__, __LINE__, pte_val(e))
210 #define pgd_ERROR(e) \
211 	printk("%s:%d: bad pgd entry %08lx.\n", __FILE__, __LINE__, pgd_val(e))
212 
213 extern unsigned long empty_zero_page[1024];
214 
215 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
216 
217 #ifdef CONFIG_MMU
218 extern pgd_t swapper_pg_dir[PAGE_SIZE/sizeof(pgd_t)];
219 extern void paging_init(void);
220 #else
221 # define swapper_pg_dir NULL
222 static inline void paging_init(void) { }
223 #endif
224 
225 /*
226  * The pmd contains the kernel virtual address of the pte page.
227  */
228 #define pmd_page_vaddr(pmd) ((unsigned long)(pmd_val(pmd) & PAGE_MASK))
229 #define pmd_pfn(pmd) (__pa(pmd_val(pmd)) >> PAGE_SHIFT)
230 #define pmd_page(pmd) virt_to_page(pmd_val(pmd))
231 
232 /*
233  * pte status.
234  */
235 # define pte_none(pte)	 (pte_val(pte) == (_PAGE_CA_INVALID | _PAGE_USER))
236 #if XCHAL_HW_VERSION_MAJOR < 2000
237 # define pte_present(pte) ((pte_val(pte) & _PAGE_CA_MASK) != _PAGE_CA_INVALID)
238 #else
239 # define pte_present(pte)						\
240 	(((pte_val(pte) & _PAGE_CA_MASK) != _PAGE_CA_INVALID)		\
241 	 || ((pte_val(pte) & _PAGE_ATTRIB_MASK) == _PAGE_NONE))
242 #endif
243 #define pte_clear(mm,addr,ptep)						\
244 	do { update_pte(ptep, __pte(_PAGE_CA_INVALID | _PAGE_USER)); } while (0)
245 
246 #define pmd_none(pmd)	 (!pmd_val(pmd))
247 #define pmd_present(pmd) (pmd_val(pmd) & PAGE_MASK)
248 #define pmd_bad(pmd)	 (pmd_val(pmd) & ~PAGE_MASK)
249 #define pmd_clear(pmdp)	 do { set_pmd(pmdp, __pmd(0)); } while (0)
250 
251 static inline int pte_write(pte_t pte) { return pte_val(pte) & _PAGE_WRITABLE; }
252 static inline int pte_dirty(pte_t pte) { return pte_val(pte) & _PAGE_DIRTY; }
253 static inline int pte_young(pte_t pte) { return pte_val(pte) & _PAGE_ACCESSED; }
254 
255 static inline pte_t pte_wrprotect(pte_t pte)
256 	{ pte_val(pte) &= ~(_PAGE_WRITABLE | _PAGE_HW_WRITE); return pte; }
257 static inline pte_t pte_mkclean(pte_t pte)
258 	{ pte_val(pte) &= ~(_PAGE_DIRTY | _PAGE_HW_WRITE); return pte; }
259 static inline pte_t pte_mkold(pte_t pte)
260 	{ pte_val(pte) &= ~_PAGE_ACCESSED; return pte; }
261 static inline pte_t pte_mkdirty(pte_t pte)
262 	{ pte_val(pte) |= _PAGE_DIRTY; return pte; }
263 static inline pte_t pte_mkyoung(pte_t pte)
264 	{ pte_val(pte) |= _PAGE_ACCESSED; return pte; }
265 static inline pte_t pte_mkwrite(pte_t pte)
266 	{ pte_val(pte) |= _PAGE_WRITABLE; return pte; }
267 
268 #define pgprot_noncached(prot) \
269 		((__pgprot((pgprot_val(prot) & ~_PAGE_CA_MASK) | \
270 			   _PAGE_CA_BYPASS)))
271 
272 /*
273  * Conversion functions: convert a page and protection to a page entry,
274  * and a page entry and page directory to the page they refer to.
275  */
276 
277 #define pte_pfn(pte)		(pte_val(pte) >> PAGE_SHIFT)
278 #define pte_same(a,b)		(pte_val(a) == pte_val(b))
279 #define pte_page(x)		pfn_to_page(pte_pfn(x))
280 #define pfn_pte(pfn, prot)	__pte(((pfn) << PAGE_SHIFT) | pgprot_val(prot))
281 #define mk_pte(page, prot)	pfn_pte(page_to_pfn(page), prot)
282 
283 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
284 {
285 	return __pte((pte_val(pte) & _PAGE_CHG_MASK) | pgprot_val(newprot));
286 }
287 
288 /*
289  * Certain architectures need to do special things when pte's
290  * within a page table are directly modified.  Thus, the following
291  * hook is made available.
292  */
293 static inline void update_pte(pte_t *ptep, pte_t pteval)
294 {
295 	*ptep = pteval;
296 #if (DCACHE_WAY_SIZE > PAGE_SIZE) && XCHAL_DCACHE_IS_WRITEBACK
297 	__asm__ __volatile__ ("dhwb %0, 0" :: "a" (ptep));
298 #endif
299 
300 }
301 
302 struct mm_struct;
303 
304 static inline void
305 set_pte_at(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pteval)
306 {
307 	update_pte(ptep, pteval);
308 }
309 
310 static inline void set_pte(pte_t *ptep, pte_t pteval)
311 {
312 	update_pte(ptep, pteval);
313 }
314 
315 static inline void
316 set_pmd(pmd_t *pmdp, pmd_t pmdval)
317 {
318 	*pmdp = pmdval;
319 }
320 
321 struct vm_area_struct;
322 
323 static inline int
324 ptep_test_and_clear_young(struct vm_area_struct *vma, unsigned long addr,
325 			  pte_t *ptep)
326 {
327 	pte_t pte = *ptep;
328 	if (!pte_young(pte))
329 		return 0;
330 	update_pte(ptep, pte_mkold(pte));
331 	return 1;
332 }
333 
334 static inline pte_t
335 ptep_get_and_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
336 {
337 	pte_t pte = *ptep;
338 	pte_clear(mm, addr, ptep);
339 	return pte;
340 }
341 
342 static inline void
343 ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
344 {
345 	pte_t pte = *ptep;
346 	update_pte(ptep, pte_wrprotect(pte));
347 }
348 
349 /*
350  * Encode/decode swap entries and swap PTEs. Swap PTEs are all PTEs that
351  * are !pte_none() && !pte_present().
352  */
353 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > 5)
354 
355 #define __swp_type(entry)	(((entry).val >> 6) & 0x1f)
356 #define __swp_offset(entry)	((entry).val >> 11)
357 #define __swp_entry(type,offs)	\
358 	((swp_entry_t){(((type) & 0x1f) << 6) | ((offs) << 11) | \
359 	 _PAGE_CA_INVALID | _PAGE_USER})
360 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
361 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
362 
363 static inline int pte_swp_exclusive(pte_t pte)
364 {
365 	return pte_val(pte) & _PAGE_SWP_EXCLUSIVE;
366 }
367 
368 static inline pte_t pte_swp_mkexclusive(pte_t pte)
369 {
370 	pte_val(pte) |= _PAGE_SWP_EXCLUSIVE;
371 	return pte;
372 }
373 
374 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
375 {
376 	pte_val(pte) &= ~_PAGE_SWP_EXCLUSIVE;
377 	return pte;
378 }
379 
380 #endif /*  !defined (__ASSEMBLY__) */
381 
382 
383 #ifdef __ASSEMBLY__
384 
385 /* Assembly macro _PGD_INDEX is the same as C pgd_index(unsigned long),
386  *                _PGD_OFFSET as C pgd_offset(struct mm_struct*, unsigned long),
387  *                _PMD_OFFSET as C pmd_offset(pgd_t*, unsigned long)
388  *                _PTE_OFFSET as C pte_offset(pmd_t*, unsigned long)
389  *
390  * Note: We require an additional temporary register which can be the same as
391  *       the register that holds the address.
392  *
393  * ((pte_t*) ((unsigned long)(pmd_val(*pmd) & PAGE_MASK)) + pte_index(addr))
394  *
395  */
396 #define _PGD_INDEX(rt,rs)	extui	rt, rs, PGDIR_SHIFT, 32-PGDIR_SHIFT
397 #define _PTE_INDEX(rt,rs)	extui	rt, rs, PAGE_SHIFT, PTRS_PER_PTE_SHIFT
398 
399 #define _PGD_OFFSET(mm,adr,tmp)		l32i	mm, mm, MM_PGD;		\
400 					_PGD_INDEX(tmp, adr);		\
401 					addx4	mm, tmp, mm
402 
403 #define _PTE_OFFSET(pmd,adr,tmp)	_PTE_INDEX(tmp, adr);		\
404 					srli	pmd, pmd, PAGE_SHIFT;	\
405 					slli	pmd, pmd, PAGE_SHIFT;	\
406 					addx4	pmd, tmp, pmd
407 
408 #else
409 
410 extern  void update_mmu_cache(struct vm_area_struct * vma,
411 			      unsigned long address, pte_t *ptep);
412 
413 typedef pte_t *pte_addr_t;
414 
415 void update_mmu_tlb(struct vm_area_struct *vma,
416 		    unsigned long address, pte_t *ptep);
417 #define __HAVE_ARCH_UPDATE_MMU_TLB
418 
419 #endif /* !defined (__ASSEMBLY__) */
420 
421 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
422 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
423 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
424 #define __HAVE_ARCH_PTEP_MKDIRTY
425 #define __HAVE_ARCH_PTE_SAME
426 /* We provide our own get_unmapped_area to cope with
427  * SHM area cache aliasing for userland.
428  */
429 #define HAVE_ARCH_UNMAPPED_AREA
430 
431 #endif /* _XTENSA_PGTABLE_H */
432