xref: /openbmc/linux/arch/s390/include/asm/pgtable.h (revision 384740dc)
1 /*
2  *  include/asm-s390/pgtable.h
3  *
4  *  S390 version
5  *    Copyright (C) 1999,2000 IBM Deutschland Entwicklung GmbH, IBM Corporation
6  *    Author(s): Hartmut Penner (hp@de.ibm.com)
7  *               Ulrich Weigand (weigand@de.ibm.com)
8  *               Martin Schwidefsky (schwidefsky@de.ibm.com)
9  *
10  *  Derived from "include/asm-i386/pgtable.h"
11  */
12 
13 #ifndef _ASM_S390_PGTABLE_H
14 #define _ASM_S390_PGTABLE_H
15 
16 /*
17  * The Linux memory management assumes a three-level page table setup. For
18  * s390 31 bit we "fold" the mid level into the top-level page table, so
19  * that we physically have the same two-level page table as the s390 mmu
20  * expects in 31 bit mode. For s390 64 bit we use three of the five levels
21  * the hardware provides (region first and region second tables are not
22  * used).
23  *
24  * The "pgd_xxx()" functions are trivial for a folded two-level
25  * setup: the pgd is never bad, and a pmd always exists (as it's folded
26  * into the pgd entry)
27  *
28  * This file contains the functions and defines necessary to modify and use
29  * the S390 page table tree.
30  */
31 #ifndef __ASSEMBLY__
32 #include <linux/sched.h>
33 #include <linux/mm_types.h>
34 #include <asm/bitops.h>
35 #include <asm/bug.h>
36 #include <asm/processor.h>
37 
38 extern pgd_t swapper_pg_dir[] __attribute__ ((aligned (4096)));
39 extern void paging_init(void);
40 extern void vmem_map_init(void);
41 
42 /*
43  * The S390 doesn't have any external MMU info: the kernel page
44  * tables contain all the necessary information.
45  */
46 #define update_mmu_cache(vma, address, pte)     do { } while (0)
47 
48 /*
49  * ZERO_PAGE is a global shared page that is always zero: used
50  * for zero-mapped memory areas etc..
51  */
52 extern char empty_zero_page[PAGE_SIZE];
53 #define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
54 #endif /* !__ASSEMBLY__ */
55 
56 /*
57  * PMD_SHIFT determines the size of the area a second-level page
58  * table can map
59  * PGDIR_SHIFT determines what a third-level page table entry can map
60  */
61 #ifndef __s390x__
62 # define PMD_SHIFT	20
63 # define PUD_SHIFT	20
64 # define PGDIR_SHIFT	20
65 #else /* __s390x__ */
66 # define PMD_SHIFT	20
67 # define PUD_SHIFT	31
68 # define PGDIR_SHIFT	42
69 #endif /* __s390x__ */
70 
71 #define PMD_SIZE        (1UL << PMD_SHIFT)
72 #define PMD_MASK        (~(PMD_SIZE-1))
73 #define PUD_SIZE	(1UL << PUD_SHIFT)
74 #define PUD_MASK	(~(PUD_SIZE-1))
75 #define PGDIR_SIZE	(1UL << PGDIR_SHIFT)
76 #define PGDIR_MASK	(~(PGDIR_SIZE-1))
77 
78 /*
79  * entries per page directory level: the S390 is two-level, so
80  * we don't really have any PMD directory physically.
81  * for S390 segment-table entries are combined to one PGD
82  * that leads to 1024 pte per pgd
83  */
84 #define PTRS_PER_PTE	256
85 #ifndef __s390x__
86 #define PTRS_PER_PMD	1
87 #define PTRS_PER_PUD	1
88 #else /* __s390x__ */
89 #define PTRS_PER_PMD	2048
90 #define PTRS_PER_PUD	2048
91 #endif /* __s390x__ */
92 #define PTRS_PER_PGD	2048
93 
94 #define FIRST_USER_ADDRESS  0
95 
96 #define pte_ERROR(e) \
97 	printk("%s:%d: bad pte %p.\n", __FILE__, __LINE__, (void *) pte_val(e))
98 #define pmd_ERROR(e) \
99 	printk("%s:%d: bad pmd %p.\n", __FILE__, __LINE__, (void *) pmd_val(e))
100 #define pud_ERROR(e) \
101 	printk("%s:%d: bad pud %p.\n", __FILE__, __LINE__, (void *) pud_val(e))
102 #define pgd_ERROR(e) \
103 	printk("%s:%d: bad pgd %p.\n", __FILE__, __LINE__, (void *) pgd_val(e))
104 
105 #ifndef __ASSEMBLY__
106 /*
107  * The vmalloc area will always be on the topmost area of the kernel
108  * mapping. We reserve 96MB (31bit) / 1GB (64bit) for vmalloc,
109  * which should be enough for any sane case.
110  * By putting vmalloc at the top, we maximise the gap between physical
111  * memory and vmalloc to catch misplaced memory accesses. As a side
112  * effect, this also makes sure that 64 bit module code cannot be used
113  * as system call address.
114  */
115 #ifndef __s390x__
116 #define VMALLOC_START	0x78000000UL
117 #define VMALLOC_END	0x7e000000UL
118 #define VMEM_MAP_END	0x80000000UL
119 #else /* __s390x__ */
120 #define VMALLOC_START	0x3e000000000UL
121 #define VMALLOC_END	0x3e040000000UL
122 #define VMEM_MAP_END	0x40000000000UL
123 #endif /* __s390x__ */
124 
125 /*
126  * VMEM_MAX_PHYS is the highest physical address that can be added to the 1:1
127  * mapping. This needs to be calculated at compile time since the size of the
128  * VMEM_MAP is static but the size of struct page can change.
129  */
130 #define VMEM_MAX_PAGES	((VMEM_MAP_END - VMALLOC_END) / sizeof(struct page))
131 #define VMEM_MAX_PFN	min(VMALLOC_START >> PAGE_SHIFT, VMEM_MAX_PAGES)
132 #define VMEM_MAX_PHYS	((VMEM_MAX_PFN << PAGE_SHIFT) & ~((16 << 20) - 1))
133 #define vmemmap		((struct page *) VMALLOC_END)
134 
135 /*
136  * A 31 bit pagetable entry of S390 has following format:
137  *  |   PFRA          |    |  OS  |
138  * 0                   0IP0
139  * 00000000001111111111222222222233
140  * 01234567890123456789012345678901
141  *
142  * I Page-Invalid Bit:    Page is not available for address-translation
143  * P Page-Protection Bit: Store access not possible for page
144  *
145  * A 31 bit segmenttable entry of S390 has following format:
146  *  |   P-table origin      |  |PTL
147  * 0                         IC
148  * 00000000001111111111222222222233
149  * 01234567890123456789012345678901
150  *
151  * I Segment-Invalid Bit:    Segment is not available for address-translation
152  * C Common-Segment Bit:     Segment is not private (PoP 3-30)
153  * PTL Page-Table-Length:    Page-table length (PTL+1*16 entries -> up to 256)
154  *
155  * The 31 bit segmenttable origin of S390 has following format:
156  *
157  *  |S-table origin   |     | STL |
158  * X                   **GPS
159  * 00000000001111111111222222222233
160  * 01234567890123456789012345678901
161  *
162  * X Space-Switch event:
163  * G Segment-Invalid Bit:     *
164  * P Private-Space Bit:       Segment is not private (PoP 3-30)
165  * S Storage-Alteration:
166  * STL Segment-Table-Length:  Segment-table length (STL+1*16 entries -> up to 2048)
167  *
168  * A 64 bit pagetable entry of S390 has following format:
169  * |                     PFRA                         |0IP0|  OS  |
170  * 0000000000111111111122222222223333333333444444444455555555556666
171  * 0123456789012345678901234567890123456789012345678901234567890123
172  *
173  * I Page-Invalid Bit:    Page is not available for address-translation
174  * P Page-Protection Bit: Store access not possible for page
175  *
176  * A 64 bit segmenttable entry of S390 has following format:
177  * |        P-table origin                              |      TT
178  * 0000000000111111111122222222223333333333444444444455555555556666
179  * 0123456789012345678901234567890123456789012345678901234567890123
180  *
181  * I Segment-Invalid Bit:    Segment is not available for address-translation
182  * C Common-Segment Bit:     Segment is not private (PoP 3-30)
183  * P Page-Protection Bit: Store access not possible for page
184  * TT Type 00
185  *
186  * A 64 bit region table entry of S390 has following format:
187  * |        S-table origin                             |   TF  TTTL
188  * 0000000000111111111122222222223333333333444444444455555555556666
189  * 0123456789012345678901234567890123456789012345678901234567890123
190  *
191  * I Segment-Invalid Bit:    Segment is not available for address-translation
192  * TT Type 01
193  * TF
194  * TL Table length
195  *
196  * The 64 bit regiontable origin of S390 has following format:
197  * |      region table origon                          |       DTTL
198  * 0000000000111111111122222222223333333333444444444455555555556666
199  * 0123456789012345678901234567890123456789012345678901234567890123
200  *
201  * X Space-Switch event:
202  * G Segment-Invalid Bit:
203  * P Private-Space Bit:
204  * S Storage-Alteration:
205  * R Real space
206  * TL Table-Length:
207  *
208  * A storage key has the following format:
209  * | ACC |F|R|C|0|
210  *  0   3 4 5 6 7
211  * ACC: access key
212  * F  : fetch protection bit
213  * R  : referenced bit
214  * C  : changed bit
215  */
216 
217 /* Hardware bits in the page table entry */
218 #define _PAGE_RO	0x200		/* HW read-only bit  */
219 #define _PAGE_INVALID	0x400		/* HW invalid bit    */
220 
221 /* Software bits in the page table entry */
222 #define _PAGE_SWT	0x001		/* SW pte type bit t */
223 #define _PAGE_SWX	0x002		/* SW pte type bit x */
224 #define _PAGE_SPECIAL	0x004		/* SW associated with special page */
225 #define __HAVE_ARCH_PTE_SPECIAL
226 
227 /* Set of bits not changed in pte_modify */
228 #define _PAGE_CHG_MASK	(PAGE_MASK | _PAGE_SPECIAL)
229 
230 /* Six different types of pages. */
231 #define _PAGE_TYPE_EMPTY	0x400
232 #define _PAGE_TYPE_NONE		0x401
233 #define _PAGE_TYPE_SWAP		0x403
234 #define _PAGE_TYPE_FILE		0x601	/* bit 0x002 is used for offset !! */
235 #define _PAGE_TYPE_RO		0x200
236 #define _PAGE_TYPE_RW		0x000
237 #define _PAGE_TYPE_EX_RO	0x202
238 #define _PAGE_TYPE_EX_RW	0x002
239 
240 /*
241  * Only four types for huge pages, using the invalid bit and protection bit
242  * of a segment table entry.
243  */
244 #define _HPAGE_TYPE_EMPTY	0x020	/* _SEGMENT_ENTRY_INV */
245 #define _HPAGE_TYPE_NONE	0x220
246 #define _HPAGE_TYPE_RO		0x200	/* _SEGMENT_ENTRY_RO  */
247 #define _HPAGE_TYPE_RW		0x000
248 
249 /*
250  * PTE type bits are rather complicated. handle_pte_fault uses pte_present,
251  * pte_none and pte_file to find out the pte type WITHOUT holding the page
252  * table lock. ptep_clear_flush on the other hand uses ptep_clear_flush to
253  * invalidate a given pte. ipte sets the hw invalid bit and clears all tlbs
254  * for the page. The page table entry is set to _PAGE_TYPE_EMPTY afterwards.
255  * This change is done while holding the lock, but the intermediate step
256  * of a previously valid pte with the hw invalid bit set can be observed by
257  * handle_pte_fault. That makes it necessary that all valid pte types with
258  * the hw invalid bit set must be distinguishable from the four pte types
259  * empty, none, swap and file.
260  *
261  *			irxt  ipte  irxt
262  * _PAGE_TYPE_EMPTY	1000   ->   1000
263  * _PAGE_TYPE_NONE	1001   ->   1001
264  * _PAGE_TYPE_SWAP	1011   ->   1011
265  * _PAGE_TYPE_FILE	11?1   ->   11?1
266  * _PAGE_TYPE_RO	0100   ->   1100
267  * _PAGE_TYPE_RW	0000   ->   1000
268  * _PAGE_TYPE_EX_RO	0110   ->   1110
269  * _PAGE_TYPE_EX_RW	0010   ->   1010
270  *
271  * pte_none is true for bits combinations 1000, 1010, 1100, 1110
272  * pte_present is true for bits combinations 0000, 0010, 0100, 0110, 1001
273  * pte_file is true for bits combinations 1101, 1111
274  * swap pte is 1011 and 0001, 0011, 0101, 0111 are invalid.
275  */
276 
277 /* Page status table bits for virtualization */
278 #define RCP_PCL_BIT	55
279 #define RCP_HR_BIT	54
280 #define RCP_HC_BIT	53
281 #define RCP_GR_BIT	50
282 #define RCP_GC_BIT	49
283 
284 #ifndef __s390x__
285 
286 /* Bits in the segment table address-space-control-element */
287 #define _ASCE_SPACE_SWITCH	0x80000000UL	/* space switch event	    */
288 #define _ASCE_ORIGIN_MASK	0x7ffff000UL	/* segment table origin	    */
289 #define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
290 #define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
291 #define _ASCE_TABLE_LENGTH	0x7f	/* 128 x 64 entries = 8k	    */
292 
293 /* Bits in the segment table entry */
294 #define _SEGMENT_ENTRY_ORIGIN	0x7fffffc0UL	/* page table origin	    */
295 #define _SEGMENT_ENTRY_INV	0x20	/* invalid segment table entry	    */
296 #define _SEGMENT_ENTRY_COMMON	0x10	/* common segment bit		    */
297 #define _SEGMENT_ENTRY_PTL	0x0f	/* page table length		    */
298 
299 #define _SEGMENT_ENTRY		(_SEGMENT_ENTRY_PTL)
300 #define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INV)
301 
302 #else /* __s390x__ */
303 
304 /* Bits in the segment/region table address-space-control-element */
305 #define _ASCE_ORIGIN		~0xfffUL/* segment table origin		    */
306 #define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
307 #define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
308 #define _ASCE_SPACE_SWITCH	0x40	/* space switch event		    */
309 #define _ASCE_REAL_SPACE	0x20	/* real space control		    */
310 #define _ASCE_TYPE_MASK		0x0c	/* asce table type mask		    */
311 #define _ASCE_TYPE_REGION1	0x0c	/* region first table type	    */
312 #define _ASCE_TYPE_REGION2	0x08	/* region second table type	    */
313 #define _ASCE_TYPE_REGION3	0x04	/* region third table type	    */
314 #define _ASCE_TYPE_SEGMENT	0x00	/* segment table type		    */
315 #define _ASCE_TABLE_LENGTH	0x03	/* region table length		    */
316 
317 /* Bits in the region table entry */
318 #define _REGION_ENTRY_ORIGIN	~0xfffUL/* region/segment table origin	    */
319 #define _REGION_ENTRY_INV	0x20	/* invalid region table entry	    */
320 #define _REGION_ENTRY_TYPE_MASK	0x0c	/* region/segment table type mask   */
321 #define _REGION_ENTRY_TYPE_R1	0x0c	/* region first table type	    */
322 #define _REGION_ENTRY_TYPE_R2	0x08	/* region second table type	    */
323 #define _REGION_ENTRY_TYPE_R3	0x04	/* region third table type	    */
324 #define _REGION_ENTRY_LENGTH	0x03	/* region third length		    */
325 
326 #define _REGION1_ENTRY		(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_LENGTH)
327 #define _REGION1_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INV)
328 #define _REGION2_ENTRY		(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_LENGTH)
329 #define _REGION2_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INV)
330 #define _REGION3_ENTRY		(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_LENGTH)
331 #define _REGION3_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INV)
332 
333 /* Bits in the segment table entry */
334 #define _SEGMENT_ENTRY_ORIGIN	~0x7ffUL/* segment table origin		    */
335 #define _SEGMENT_ENTRY_RO	0x200	/* page protection bit		    */
336 #define _SEGMENT_ENTRY_INV	0x20	/* invalid segment table entry	    */
337 
338 #define _SEGMENT_ENTRY		(0)
339 #define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INV)
340 
341 #define _SEGMENT_ENTRY_LARGE	0x400	/* STE-format control, large page   */
342 #define _SEGMENT_ENTRY_CO	0x100	/* change-recording override   */
343 
344 #endif /* __s390x__ */
345 
346 /*
347  * A user page table pointer has the space-switch-event bit, the
348  * private-space-control bit and the storage-alteration-event-control
349  * bit set. A kernel page table pointer doesn't need them.
350  */
351 #define _ASCE_USER_BITS		(_ASCE_SPACE_SWITCH | _ASCE_PRIVATE_SPACE | \
352 				 _ASCE_ALT_EVENT)
353 
354 /* Bits int the storage key */
355 #define _PAGE_CHANGED    0x02          /* HW changed bit                   */
356 #define _PAGE_REFERENCED 0x04          /* HW referenced bit                */
357 
358 /*
359  * Page protection definitions.
360  */
361 #define PAGE_NONE	__pgprot(_PAGE_TYPE_NONE)
362 #define PAGE_RO		__pgprot(_PAGE_TYPE_RO)
363 #define PAGE_RW		__pgprot(_PAGE_TYPE_RW)
364 #define PAGE_EX_RO	__pgprot(_PAGE_TYPE_EX_RO)
365 #define PAGE_EX_RW	__pgprot(_PAGE_TYPE_EX_RW)
366 
367 #define PAGE_KERNEL	PAGE_RW
368 #define PAGE_COPY	PAGE_RO
369 
370 /*
371  * Dependent on the EXEC_PROTECT option s390 can do execute protection.
372  * Write permission always implies read permission. In theory with a
373  * primary/secondary page table execute only can be implemented but
374  * it would cost an additional bit in the pte to distinguish all the
375  * different pte types. To avoid that execute permission currently
376  * implies read permission as well.
377  */
378          /*xwr*/
379 #define __P000	PAGE_NONE
380 #define __P001	PAGE_RO
381 #define __P010	PAGE_RO
382 #define __P011	PAGE_RO
383 #define __P100	PAGE_EX_RO
384 #define __P101	PAGE_EX_RO
385 #define __P110	PAGE_EX_RO
386 #define __P111	PAGE_EX_RO
387 
388 #define __S000	PAGE_NONE
389 #define __S001	PAGE_RO
390 #define __S010	PAGE_RW
391 #define __S011	PAGE_RW
392 #define __S100	PAGE_EX_RO
393 #define __S101	PAGE_EX_RO
394 #define __S110	PAGE_EX_RW
395 #define __S111	PAGE_EX_RW
396 
397 #ifndef __s390x__
398 # define PxD_SHADOW_SHIFT	1
399 #else /* __s390x__ */
400 # define PxD_SHADOW_SHIFT	2
401 #endif /* __s390x__ */
402 
403 static inline void *get_shadow_table(void *table)
404 {
405 	unsigned long addr, offset;
406 	struct page *page;
407 
408 	addr = (unsigned long) table;
409 	offset = addr & ((PAGE_SIZE << PxD_SHADOW_SHIFT) - 1);
410 	page = virt_to_page((void *)(addr ^ offset));
411 	return (void *)(addr_t)(page->index ? (page->index | offset) : 0UL);
412 }
413 
414 /*
415  * Certain architectures need to do special things when PTEs
416  * within a page table are directly modified.  Thus, the following
417  * hook is made available.
418  */
419 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
420 			      pte_t *ptep, pte_t entry)
421 {
422 	*ptep = entry;
423 	if (mm->context.noexec) {
424 		if (!(pte_val(entry) & _PAGE_INVALID) &&
425 		    (pte_val(entry) & _PAGE_SWX))
426 			pte_val(entry) |= _PAGE_RO;
427 		else
428 			pte_val(entry) = _PAGE_TYPE_EMPTY;
429 		ptep[PTRS_PER_PTE] = entry;
430 	}
431 }
432 
433 /*
434  * pgd/pmd/pte query functions
435  */
436 #ifndef __s390x__
437 
438 static inline int pgd_present(pgd_t pgd) { return 1; }
439 static inline int pgd_none(pgd_t pgd)    { return 0; }
440 static inline int pgd_bad(pgd_t pgd)     { return 0; }
441 
442 static inline int pud_present(pud_t pud) { return 1; }
443 static inline int pud_none(pud_t pud)	 { return 0; }
444 static inline int pud_bad(pud_t pud)	 { return 0; }
445 
446 #else /* __s390x__ */
447 
448 static inline int pgd_present(pgd_t pgd)
449 {
450 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
451 		return 1;
452 	return (pgd_val(pgd) & _REGION_ENTRY_ORIGIN) != 0UL;
453 }
454 
455 static inline int pgd_none(pgd_t pgd)
456 {
457 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
458 		return 0;
459 	return (pgd_val(pgd) & _REGION_ENTRY_INV) != 0UL;
460 }
461 
462 static inline int pgd_bad(pgd_t pgd)
463 {
464 	/*
465 	 * With dynamic page table levels the pgd can be a region table
466 	 * entry or a segment table entry. Check for the bit that are
467 	 * invalid for either table entry.
468 	 */
469 	unsigned long mask =
470 		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INV &
471 		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
472 	return (pgd_val(pgd) & mask) != 0;
473 }
474 
475 static inline int pud_present(pud_t pud)
476 {
477 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
478 		return 1;
479 	return (pud_val(pud) & _REGION_ENTRY_ORIGIN) != 0UL;
480 }
481 
482 static inline int pud_none(pud_t pud)
483 {
484 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
485 		return 0;
486 	return (pud_val(pud) & _REGION_ENTRY_INV) != 0UL;
487 }
488 
489 static inline int pud_bad(pud_t pud)
490 {
491 	/*
492 	 * With dynamic page table levels the pud can be a region table
493 	 * entry or a segment table entry. Check for the bit that are
494 	 * invalid for either table entry.
495 	 */
496 	unsigned long mask =
497 		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INV &
498 		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
499 	return (pud_val(pud) & mask) != 0;
500 }
501 
502 #endif /* __s390x__ */
503 
504 static inline int pmd_present(pmd_t pmd)
505 {
506 	return (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN) != 0UL;
507 }
508 
509 static inline int pmd_none(pmd_t pmd)
510 {
511 	return (pmd_val(pmd) & _SEGMENT_ENTRY_INV) != 0UL;
512 }
513 
514 static inline int pmd_bad(pmd_t pmd)
515 {
516 	unsigned long mask = ~_SEGMENT_ENTRY_ORIGIN & ~_SEGMENT_ENTRY_INV;
517 	return (pmd_val(pmd) & mask) != _SEGMENT_ENTRY;
518 }
519 
520 static inline int pte_none(pte_t pte)
521 {
522 	return (pte_val(pte) & _PAGE_INVALID) && !(pte_val(pte) & _PAGE_SWT);
523 }
524 
525 static inline int pte_present(pte_t pte)
526 {
527 	unsigned long mask = _PAGE_RO | _PAGE_INVALID | _PAGE_SWT | _PAGE_SWX;
528 	return (pte_val(pte) & mask) == _PAGE_TYPE_NONE ||
529 		(!(pte_val(pte) & _PAGE_INVALID) &&
530 		 !(pte_val(pte) & _PAGE_SWT));
531 }
532 
533 static inline int pte_file(pte_t pte)
534 {
535 	unsigned long mask = _PAGE_RO | _PAGE_INVALID | _PAGE_SWT;
536 	return (pte_val(pte) & mask) == _PAGE_TYPE_FILE;
537 }
538 
539 static inline int pte_special(pte_t pte)
540 {
541 	return (pte_val(pte) & _PAGE_SPECIAL);
542 }
543 
544 #define __HAVE_ARCH_PTE_SAME
545 #define pte_same(a,b)  (pte_val(a) == pte_val(b))
546 
547 static inline void rcp_lock(pte_t *ptep)
548 {
549 #ifdef CONFIG_PGSTE
550 	unsigned long *pgste = (unsigned long *) (ptep + PTRS_PER_PTE);
551 	preempt_disable();
552 	while (test_and_set_bit(RCP_PCL_BIT, pgste))
553 		;
554 #endif
555 }
556 
557 static inline void rcp_unlock(pte_t *ptep)
558 {
559 #ifdef CONFIG_PGSTE
560 	unsigned long *pgste = (unsigned long *) (ptep + PTRS_PER_PTE);
561 	clear_bit(RCP_PCL_BIT, pgste);
562 	preempt_enable();
563 #endif
564 }
565 
566 /* forward declaration for SetPageUptodate in page-flags.h*/
567 static inline void page_clear_dirty(struct page *page);
568 #include <linux/page-flags.h>
569 
570 static inline void ptep_rcp_copy(pte_t *ptep)
571 {
572 #ifdef CONFIG_PGSTE
573 	struct page *page = virt_to_page(pte_val(*ptep));
574 	unsigned int skey;
575 	unsigned long *pgste = (unsigned long *) (ptep + PTRS_PER_PTE);
576 
577 	skey = page_get_storage_key(page_to_phys(page));
578 	if (skey & _PAGE_CHANGED)
579 		set_bit_simple(RCP_GC_BIT, pgste);
580 	if (skey & _PAGE_REFERENCED)
581 		set_bit_simple(RCP_GR_BIT, pgste);
582 	if (test_and_clear_bit_simple(RCP_HC_BIT, pgste))
583 		SetPageDirty(page);
584 	if (test_and_clear_bit_simple(RCP_HR_BIT, pgste))
585 		SetPageReferenced(page);
586 #endif
587 }
588 
589 /*
590  * query functions pte_write/pte_dirty/pte_young only work if
591  * pte_present() is true. Undefined behaviour if not..
592  */
593 static inline int pte_write(pte_t pte)
594 {
595 	return (pte_val(pte) & _PAGE_RO) == 0;
596 }
597 
598 static inline int pte_dirty(pte_t pte)
599 {
600 	/* A pte is neither clean nor dirty on s/390. The dirty bit
601 	 * is in the storage key. See page_test_and_clear_dirty for
602 	 * details.
603 	 */
604 	return 0;
605 }
606 
607 static inline int pte_young(pte_t pte)
608 {
609 	/* A pte is neither young nor old on s/390. The young bit
610 	 * is in the storage key. See page_test_and_clear_young for
611 	 * details.
612 	 */
613 	return 0;
614 }
615 
616 /*
617  * pgd/pmd/pte modification functions
618  */
619 
620 #ifndef __s390x__
621 
622 #define pgd_clear(pgd)		do { } while (0)
623 #define pud_clear(pud)		do { } while (0)
624 
625 #else /* __s390x__ */
626 
627 static inline void pgd_clear_kernel(pgd_t * pgd)
628 {
629 	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
630 		pgd_val(*pgd) = _REGION2_ENTRY_EMPTY;
631 }
632 
633 static inline void pgd_clear(pgd_t * pgd)
634 {
635 	pgd_t *shadow = get_shadow_table(pgd);
636 
637 	pgd_clear_kernel(pgd);
638 	if (shadow)
639 		pgd_clear_kernel(shadow);
640 }
641 
642 static inline void pud_clear_kernel(pud_t *pud)
643 {
644 	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
645 		pud_val(*pud) = _REGION3_ENTRY_EMPTY;
646 }
647 
648 static inline void pud_clear(pud_t *pud)
649 {
650 	pud_t *shadow = get_shadow_table(pud);
651 
652 	pud_clear_kernel(pud);
653 	if (shadow)
654 		pud_clear_kernel(shadow);
655 }
656 
657 #endif /* __s390x__ */
658 
659 static inline void pmd_clear_kernel(pmd_t * pmdp)
660 {
661 	pmd_val(*pmdp) = _SEGMENT_ENTRY_EMPTY;
662 }
663 
664 static inline void pmd_clear(pmd_t *pmd)
665 {
666 	pmd_t *shadow = get_shadow_table(pmd);
667 
668 	pmd_clear_kernel(pmd);
669 	if (shadow)
670 		pmd_clear_kernel(shadow);
671 }
672 
673 static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
674 {
675 	if (mm->context.pgstes)
676 		ptep_rcp_copy(ptep);
677 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
678 	if (mm->context.noexec)
679 		pte_val(ptep[PTRS_PER_PTE]) = _PAGE_TYPE_EMPTY;
680 }
681 
682 /*
683  * The following pte modification functions only work if
684  * pte_present() is true. Undefined behaviour if not..
685  */
686 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
687 {
688 	pte_val(pte) &= _PAGE_CHG_MASK;
689 	pte_val(pte) |= pgprot_val(newprot);
690 	return pte;
691 }
692 
693 static inline pte_t pte_wrprotect(pte_t pte)
694 {
695 	/* Do not clobber _PAGE_TYPE_NONE pages!  */
696 	if (!(pte_val(pte) & _PAGE_INVALID))
697 		pte_val(pte) |= _PAGE_RO;
698 	return pte;
699 }
700 
701 static inline pte_t pte_mkwrite(pte_t pte)
702 {
703 	pte_val(pte) &= ~_PAGE_RO;
704 	return pte;
705 }
706 
707 static inline pte_t pte_mkclean(pte_t pte)
708 {
709 	/* The only user of pte_mkclean is the fork() code.
710 	   We must *not* clear the *physical* page dirty bit
711 	   just because fork() wants to clear the dirty bit in
712 	   *one* of the page's mappings.  So we just do nothing. */
713 	return pte;
714 }
715 
716 static inline pte_t pte_mkdirty(pte_t pte)
717 {
718 	/* We do not explicitly set the dirty bit because the
719 	 * sske instruction is slow. It is faster to let the
720 	 * next instruction set the dirty bit.
721 	 */
722 	return pte;
723 }
724 
725 static inline pte_t pte_mkold(pte_t pte)
726 {
727 	/* S/390 doesn't keep its dirty/referenced bit in the pte.
728 	 * There is no point in clearing the real referenced bit.
729 	 */
730 	return pte;
731 }
732 
733 static inline pte_t pte_mkyoung(pte_t pte)
734 {
735 	/* S/390 doesn't keep its dirty/referenced bit in the pte.
736 	 * There is no point in setting the real referenced bit.
737 	 */
738 	return pte;
739 }
740 
741 static inline pte_t pte_mkspecial(pte_t pte)
742 {
743 	pte_val(pte) |= _PAGE_SPECIAL;
744 	return pte;
745 }
746 
747 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
748 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
749 					    unsigned long addr, pte_t *ptep)
750 {
751 #ifdef CONFIG_PGSTE
752 	unsigned long physpage;
753 	int young;
754 	unsigned long *pgste;
755 
756 	if (!vma->vm_mm->context.pgstes)
757 		return 0;
758 	physpage = pte_val(*ptep) & PAGE_MASK;
759 	pgste = (unsigned long *) (ptep + PTRS_PER_PTE);
760 
761 	young = ((page_get_storage_key(physpage) & _PAGE_REFERENCED) != 0);
762 	rcp_lock(ptep);
763 	if (young)
764 		set_bit_simple(RCP_GR_BIT, pgste);
765 	young |= test_and_clear_bit_simple(RCP_HR_BIT, pgste);
766 	rcp_unlock(ptep);
767 	return young;
768 #endif
769 	return 0;
770 }
771 
772 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
773 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
774 					 unsigned long address, pte_t *ptep)
775 {
776 	/* No need to flush TLB
777 	 * On s390 reference bits are in storage key and never in TLB
778 	 * With virtualization we handle the reference bit, without we
779 	 * we can simply return */
780 #ifdef CONFIG_PGSTE
781 	return ptep_test_and_clear_young(vma, address, ptep);
782 #endif
783 	return 0;
784 }
785 
786 static inline void __ptep_ipte(unsigned long address, pte_t *ptep)
787 {
788 	if (!(pte_val(*ptep) & _PAGE_INVALID)) {
789 #ifndef __s390x__
790 		/* pto must point to the start of the segment table */
791 		pte_t *pto = (pte_t *) (((unsigned long) ptep) & 0x7ffffc00);
792 #else
793 		/* ipte in zarch mode can do the math */
794 		pte_t *pto = ptep;
795 #endif
796 		asm volatile(
797 			"	ipte	%2,%3"
798 			: "=m" (*ptep) : "m" (*ptep),
799 			  "a" (pto), "a" (address));
800 	}
801 }
802 
803 static inline void ptep_invalidate(struct mm_struct *mm,
804 				   unsigned long address, pte_t *ptep)
805 {
806 	if (mm->context.pgstes) {
807 		rcp_lock(ptep);
808 		__ptep_ipte(address, ptep);
809 		ptep_rcp_copy(ptep);
810 		pte_val(*ptep) = _PAGE_TYPE_EMPTY;
811 		rcp_unlock(ptep);
812 		return;
813 	}
814 	__ptep_ipte(address, ptep);
815 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
816 	if (mm->context.noexec) {
817 		__ptep_ipte(address, ptep + PTRS_PER_PTE);
818 		pte_val(*(ptep + PTRS_PER_PTE)) = _PAGE_TYPE_EMPTY;
819 	}
820 }
821 
822 /*
823  * This is hard to understand. ptep_get_and_clear and ptep_clear_flush
824  * both clear the TLB for the unmapped pte. The reason is that
825  * ptep_get_and_clear is used in common code (e.g. change_pte_range)
826  * to modify an active pte. The sequence is
827  *   1) ptep_get_and_clear
828  *   2) set_pte_at
829  *   3) flush_tlb_range
830  * On s390 the tlb needs to get flushed with the modification of the pte
831  * if the pte is active. The only way how this can be implemented is to
832  * have ptep_get_and_clear do the tlb flush. In exchange flush_tlb_range
833  * is a nop.
834  */
835 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
836 #define ptep_get_and_clear(__mm, __address, __ptep)			\
837 ({									\
838 	pte_t __pte = *(__ptep);					\
839 	if (atomic_read(&(__mm)->mm_users) > 1 ||			\
840 	    (__mm) != current->active_mm)				\
841 		ptep_invalidate(__mm, __address, __ptep);		\
842 	else								\
843 		pte_clear((__mm), (__address), (__ptep));		\
844 	__pte;								\
845 })
846 
847 #define __HAVE_ARCH_PTEP_CLEAR_FLUSH
848 static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
849 				     unsigned long address, pte_t *ptep)
850 {
851 	pte_t pte = *ptep;
852 	ptep_invalidate(vma->vm_mm, address, ptep);
853 	return pte;
854 }
855 
856 /*
857  * The batched pte unmap code uses ptep_get_and_clear_full to clear the
858  * ptes. Here an optimization is possible. tlb_gather_mmu flushes all
859  * tlbs of an mm if it can guarantee that the ptes of the mm_struct
860  * cannot be accessed while the batched unmap is running. In this case
861  * full==1 and a simple pte_clear is enough. See tlb.h.
862  */
863 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
864 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
865 					    unsigned long addr,
866 					    pte_t *ptep, int full)
867 {
868 	pte_t pte = *ptep;
869 
870 	if (full)
871 		pte_clear(mm, addr, ptep);
872 	else
873 		ptep_invalidate(mm, addr, ptep);
874 	return pte;
875 }
876 
877 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
878 #define ptep_set_wrprotect(__mm, __addr, __ptep)			\
879 ({									\
880 	pte_t __pte = *(__ptep);					\
881 	if (pte_write(__pte)) {						\
882 		if (atomic_read(&(__mm)->mm_users) > 1 ||		\
883 		    (__mm) != current->active_mm)			\
884 			ptep_invalidate(__mm, __addr, __ptep);		\
885 		set_pte_at(__mm, __addr, __ptep, pte_wrprotect(__pte));	\
886 	}								\
887 })
888 
889 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
890 #define ptep_set_access_flags(__vma, __addr, __ptep, __entry, __dirty)	\
891 ({									\
892 	int __changed = !pte_same(*(__ptep), __entry);			\
893 	if (__changed) {						\
894 		ptep_invalidate((__vma)->vm_mm, __addr, __ptep);	\
895 		set_pte_at((__vma)->vm_mm, __addr, __ptep, __entry);	\
896 	}								\
897 	__changed;							\
898 })
899 
900 /*
901  * Test and clear dirty bit in storage key.
902  * We can't clear the changed bit atomically. This is a potential
903  * race against modification of the referenced bit. This function
904  * should therefore only be called if it is not mapped in any
905  * address space.
906  */
907 #define __HAVE_ARCH_PAGE_TEST_DIRTY
908 static inline int page_test_dirty(struct page *page)
909 {
910 	return (page_get_storage_key(page_to_phys(page)) & _PAGE_CHANGED) != 0;
911 }
912 
913 #define __HAVE_ARCH_PAGE_CLEAR_DIRTY
914 static inline void page_clear_dirty(struct page *page)
915 {
916 	page_set_storage_key(page_to_phys(page), PAGE_DEFAULT_KEY);
917 }
918 
919 /*
920  * Test and clear referenced bit in storage key.
921  */
922 #define __HAVE_ARCH_PAGE_TEST_AND_CLEAR_YOUNG
923 static inline int page_test_and_clear_young(struct page *page)
924 {
925 	unsigned long physpage = page_to_phys(page);
926 	int ccode;
927 
928 	asm volatile(
929 		"	rrbe	0,%1\n"
930 		"	ipm	%0\n"
931 		"	srl	%0,28\n"
932 		: "=d" (ccode) : "a" (physpage) : "cc" );
933 	return ccode & 2;
934 }
935 
936 /*
937  * Conversion functions: convert a page and protection to a page entry,
938  * and a page entry and page directory to the page they refer to.
939  */
940 static inline pte_t mk_pte_phys(unsigned long physpage, pgprot_t pgprot)
941 {
942 	pte_t __pte;
943 	pte_val(__pte) = physpage + pgprot_val(pgprot);
944 	return __pte;
945 }
946 
947 static inline pte_t mk_pte(struct page *page, pgprot_t pgprot)
948 {
949 	unsigned long physpage = page_to_phys(page);
950 
951 	return mk_pte_phys(physpage, pgprot);
952 }
953 
954 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD-1))
955 #define pud_index(address) (((address) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
956 #define pmd_index(address) (((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1))
957 #define pte_index(address) (((address) >> PAGE_SHIFT) & (PTRS_PER_PTE-1))
958 
959 #define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address))
960 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
961 
962 #ifndef __s390x__
963 
964 #define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
965 #define pud_deref(pmd) ({ BUG(); 0UL; })
966 #define pgd_deref(pmd) ({ BUG(); 0UL; })
967 
968 #define pud_offset(pgd, address) ((pud_t *) pgd)
969 #define pmd_offset(pud, address) ((pmd_t *) pud + pmd_index(address))
970 
971 #else /* __s390x__ */
972 
973 #define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
974 #define pud_deref(pud) (pud_val(pud) & _REGION_ENTRY_ORIGIN)
975 #define pgd_deref(pgd) (pgd_val(pgd) & _REGION_ENTRY_ORIGIN)
976 
977 static inline pud_t *pud_offset(pgd_t *pgd, unsigned long address)
978 {
979 	pud_t *pud = (pud_t *) pgd;
980 	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
981 		pud = (pud_t *) pgd_deref(*pgd);
982 	return pud  + pud_index(address);
983 }
984 
985 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
986 {
987 	pmd_t *pmd = (pmd_t *) pud;
988 	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
989 		pmd = (pmd_t *) pud_deref(*pud);
990 	return pmd + pmd_index(address);
991 }
992 
993 #endif /* __s390x__ */
994 
995 #define pfn_pte(pfn,pgprot) mk_pte_phys(__pa((pfn) << PAGE_SHIFT),(pgprot))
996 #define pte_pfn(x) (pte_val(x) >> PAGE_SHIFT)
997 #define pte_page(x) pfn_to_page(pte_pfn(x))
998 
999 #define pmd_page(pmd) pfn_to_page(pmd_val(pmd) >> PAGE_SHIFT)
1000 
1001 /* Find an entry in the lowest level page table.. */
1002 #define pte_offset(pmd, addr) ((pte_t *) pmd_deref(*(pmd)) + pte_index(addr))
1003 #define pte_offset_kernel(pmd, address) pte_offset(pmd,address)
1004 #define pte_offset_map(pmd, address) pte_offset_kernel(pmd, address)
1005 #define pte_offset_map_nested(pmd, address) pte_offset_kernel(pmd, address)
1006 #define pte_unmap(pte) do { } while (0)
1007 #define pte_unmap_nested(pte) do { } while (0)
1008 
1009 /*
1010  * 31 bit swap entry format:
1011  * A page-table entry has some bits we have to treat in a special way.
1012  * Bits 0, 20 and bit 23 have to be zero, otherwise an specification
1013  * exception will occur instead of a page translation exception. The
1014  * specifiation exception has the bad habit not to store necessary
1015  * information in the lowcore.
1016  * Bit 21 and bit 22 are the page invalid bit and the page protection
1017  * bit. We set both to indicate a swapped page.
1018  * Bit 30 and 31 are used to distinguish the different page types. For
1019  * a swapped page these bits need to be zero.
1020  * This leaves the bits 1-19 and bits 24-29 to store type and offset.
1021  * We use the 5 bits from 25-29 for the type and the 20 bits from 1-19
1022  * plus 24 for the offset.
1023  * 0|     offset        |0110|o|type |00|
1024  * 0 0000000001111111111 2222 2 22222 33
1025  * 0 1234567890123456789 0123 4 56789 01
1026  *
1027  * 64 bit swap entry format:
1028  * A page-table entry has some bits we have to treat in a special way.
1029  * Bits 52 and bit 55 have to be zero, otherwise an specification
1030  * exception will occur instead of a page translation exception. The
1031  * specifiation exception has the bad habit not to store necessary
1032  * information in the lowcore.
1033  * Bit 53 and bit 54 are the page invalid bit and the page protection
1034  * bit. We set both to indicate a swapped page.
1035  * Bit 62 and 63 are used to distinguish the different page types. For
1036  * a swapped page these bits need to be zero.
1037  * This leaves the bits 0-51 and bits 56-61 to store type and offset.
1038  * We use the 5 bits from 57-61 for the type and the 53 bits from 0-51
1039  * plus 56 for the offset.
1040  * |                      offset                        |0110|o|type |00|
1041  *  0000000000111111111122222222223333333333444444444455 5555 5 55566 66
1042  *  0123456789012345678901234567890123456789012345678901 2345 6 78901 23
1043  */
1044 #ifndef __s390x__
1045 #define __SWP_OFFSET_MASK (~0UL >> 12)
1046 #else
1047 #define __SWP_OFFSET_MASK (~0UL >> 11)
1048 #endif
1049 static inline pte_t mk_swap_pte(unsigned long type, unsigned long offset)
1050 {
1051 	pte_t pte;
1052 	offset &= __SWP_OFFSET_MASK;
1053 	pte_val(pte) = _PAGE_TYPE_SWAP | ((type & 0x1f) << 2) |
1054 		((offset & 1UL) << 7) | ((offset & ~1UL) << 11);
1055 	return pte;
1056 }
1057 
1058 #define __swp_type(entry)	(((entry).val >> 2) & 0x1f)
1059 #define __swp_offset(entry)	(((entry).val >> 11) | (((entry).val >> 7) & 1))
1060 #define __swp_entry(type,offset) ((swp_entry_t) { pte_val(mk_swap_pte((type),(offset))) })
1061 
1062 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
1063 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
1064 
1065 #ifndef __s390x__
1066 # define PTE_FILE_MAX_BITS	26
1067 #else /* __s390x__ */
1068 # define PTE_FILE_MAX_BITS	59
1069 #endif /* __s390x__ */
1070 
1071 #define pte_to_pgoff(__pte) \
1072 	((((__pte).pte >> 12) << 7) + (((__pte).pte >> 1) & 0x7f))
1073 
1074 #define pgoff_to_pte(__off) \
1075 	((pte_t) { ((((__off) & 0x7f) << 1) + (((__off) >> 7) << 12)) \
1076 		   | _PAGE_TYPE_FILE })
1077 
1078 #endif /* !__ASSEMBLY__ */
1079 
1080 #define kern_addr_valid(addr)   (1)
1081 
1082 extern int vmem_add_mapping(unsigned long start, unsigned long size);
1083 extern int vmem_remove_mapping(unsigned long start, unsigned long size);
1084 extern int s390_enable_sie(void);
1085 
1086 /*
1087  * No page table caches to initialise
1088  */
1089 #define pgtable_cache_init()	do { } while (0)
1090 
1091 #include <asm-generic/pgtable.h>
1092 
1093 #endif /* _S390_PAGE_H */
1094