xref: /openbmc/linux/arch/s390/include/asm/pgtable.h (revision d0bd7f2a)
1 /*
2  *  S390 version
3  *    Copyright IBM Corp. 1999, 2000
4  *    Author(s): Hartmut Penner (hp@de.ibm.com)
5  *               Ulrich Weigand (weigand@de.ibm.com)
6  *               Martin Schwidefsky (schwidefsky@de.ibm.com)
7  *
8  *  Derived from "include/asm-i386/pgtable.h"
9  */
10 
11 #ifndef _ASM_S390_PGTABLE_H
12 #define _ASM_S390_PGTABLE_H
13 
14 /*
15  * The Linux memory management assumes a three-level page table setup.
16  * For s390 64 bit we use up to four of the five levels the hardware
17  * provides (region first tables are not used).
18  *
19  * The "pgd_xxx()" functions are trivial for a folded two-level
20  * setup: the pgd is never bad, and a pmd always exists (as it's folded
21  * into the pgd entry)
22  *
23  * This file contains the functions and defines necessary to modify and use
24  * the S390 page table tree.
25  */
26 #ifndef __ASSEMBLY__
27 #include <linux/sched.h>
28 #include <linux/mm_types.h>
29 #include <linux/page-flags.h>
30 #include <linux/radix-tree.h>
31 #include <asm/bug.h>
32 #include <asm/page.h>
33 
34 extern pgd_t swapper_pg_dir[] __attribute__ ((aligned (4096)));
35 extern void paging_init(void);
36 extern void vmem_map_init(void);
37 
38 /*
39  * The S390 doesn't have any external MMU info: the kernel page
40  * tables contain all the necessary information.
41  */
42 #define update_mmu_cache(vma, address, ptep)     do { } while (0)
43 #define update_mmu_cache_pmd(vma, address, ptep) do { } while (0)
44 
45 /*
46  * ZERO_PAGE is a global shared page that is always zero; used
47  * for zero-mapped memory areas etc..
48  */
49 
50 extern unsigned long empty_zero_page;
51 extern unsigned long zero_page_mask;
52 
53 #define ZERO_PAGE(vaddr) \
54 	(virt_to_page((void *)(empty_zero_page + \
55 	 (((unsigned long)(vaddr)) &zero_page_mask))))
56 #define __HAVE_COLOR_ZERO_PAGE
57 
58 /* TODO: s390 cannot support io_remap_pfn_range... */
59 #endif /* !__ASSEMBLY__ */
60 
61 /*
62  * PMD_SHIFT determines the size of the area a second-level page
63  * table can map
64  * PGDIR_SHIFT determines what a third-level page table entry can map
65  */
66 #define PMD_SHIFT	20
67 #define PUD_SHIFT	31
68 #define PGDIR_SHIFT	42
69 
70 #define PMD_SIZE        (1UL << PMD_SHIFT)
71 #define PMD_MASK        (~(PMD_SIZE-1))
72 #define PUD_SIZE	(1UL << PUD_SHIFT)
73 #define PUD_MASK	(~(PUD_SIZE-1))
74 #define PGDIR_SIZE	(1UL << PGDIR_SHIFT)
75 #define PGDIR_MASK	(~(PGDIR_SIZE-1))
76 
77 /*
78  * entries per page directory level: the S390 is two-level, so
79  * we don't really have any PMD directory physically.
80  * for S390 segment-table entries are combined to one PGD
81  * that leads to 1024 pte per pgd
82  */
83 #define PTRS_PER_PTE	256
84 #define PTRS_PER_PMD	2048
85 #define PTRS_PER_PUD	2048
86 #define PTRS_PER_PGD	2048
87 
88 #define FIRST_USER_ADDRESS  0UL
89 
90 #define pte_ERROR(e) \
91 	printk("%s:%d: bad pte %p.\n", __FILE__, __LINE__, (void *) pte_val(e))
92 #define pmd_ERROR(e) \
93 	printk("%s:%d: bad pmd %p.\n", __FILE__, __LINE__, (void *) pmd_val(e))
94 #define pud_ERROR(e) \
95 	printk("%s:%d: bad pud %p.\n", __FILE__, __LINE__, (void *) pud_val(e))
96 #define pgd_ERROR(e) \
97 	printk("%s:%d: bad pgd %p.\n", __FILE__, __LINE__, (void *) pgd_val(e))
98 
99 #ifndef __ASSEMBLY__
100 /*
101  * The vmalloc and module area will always be on the topmost area of the
102  * kernel mapping. We reserve 128GB (64bit) for vmalloc and modules.
103  * On 64 bit kernels we have a 2GB area at the top of the vmalloc area where
104  * modules will reside. That makes sure that inter module branches always
105  * happen without trampolines and in addition the placement within a 2GB frame
106  * is branch prediction unit friendly.
107  */
108 extern unsigned long VMALLOC_START;
109 extern unsigned long VMALLOC_END;
110 extern struct page *vmemmap;
111 
112 #define VMEM_MAX_PHYS ((unsigned long) vmemmap)
113 
114 extern unsigned long MODULES_VADDR;
115 extern unsigned long MODULES_END;
116 #define MODULES_VADDR	MODULES_VADDR
117 #define MODULES_END	MODULES_END
118 #define MODULES_LEN	(1UL << 31)
119 
120 static inline int is_module_addr(void *addr)
121 {
122 	BUILD_BUG_ON(MODULES_LEN > (1UL << 31));
123 	if (addr < (void *)MODULES_VADDR)
124 		return 0;
125 	if (addr > (void *)MODULES_END)
126 		return 0;
127 	return 1;
128 }
129 
130 /*
131  * A 64 bit pagetable entry of S390 has following format:
132  * |			 PFRA			      |0IPC|  OS  |
133  * 0000000000111111111122222222223333333333444444444455555555556666
134  * 0123456789012345678901234567890123456789012345678901234567890123
135  *
136  * I Page-Invalid Bit:    Page is not available for address-translation
137  * P Page-Protection Bit: Store access not possible for page
138  * C Change-bit override: HW is not required to set change bit
139  *
140  * A 64 bit segmenttable entry of S390 has following format:
141  * |        P-table origin                              |      TT
142  * 0000000000111111111122222222223333333333444444444455555555556666
143  * 0123456789012345678901234567890123456789012345678901234567890123
144  *
145  * I Segment-Invalid Bit:    Segment is not available for address-translation
146  * C Common-Segment Bit:     Segment is not private (PoP 3-30)
147  * P Page-Protection Bit: Store access not possible for page
148  * TT Type 00
149  *
150  * A 64 bit region table entry of S390 has following format:
151  * |        S-table origin                             |   TF  TTTL
152  * 0000000000111111111122222222223333333333444444444455555555556666
153  * 0123456789012345678901234567890123456789012345678901234567890123
154  *
155  * I Segment-Invalid Bit:    Segment is not available for address-translation
156  * TT Type 01
157  * TF
158  * TL Table length
159  *
160  * The 64 bit regiontable origin of S390 has following format:
161  * |      region table origon                          |       DTTL
162  * 0000000000111111111122222222223333333333444444444455555555556666
163  * 0123456789012345678901234567890123456789012345678901234567890123
164  *
165  * X Space-Switch event:
166  * G Segment-Invalid Bit:
167  * P Private-Space Bit:
168  * S Storage-Alteration:
169  * R Real space
170  * TL Table-Length:
171  *
172  * A storage key has the following format:
173  * | ACC |F|R|C|0|
174  *  0   3 4 5 6 7
175  * ACC: access key
176  * F  : fetch protection bit
177  * R  : referenced bit
178  * C  : changed bit
179  */
180 
181 /* Hardware bits in the page table entry */
182 #define _PAGE_PROTECT	0x200		/* HW read-only bit  */
183 #define _PAGE_INVALID	0x400		/* HW invalid bit    */
184 #define _PAGE_LARGE	0x800		/* Bit to mark a large pte */
185 
186 /* Software bits in the page table entry */
187 #define _PAGE_PRESENT	0x001		/* SW pte present bit */
188 #define _PAGE_YOUNG	0x004		/* SW pte young bit */
189 #define _PAGE_DIRTY	0x008		/* SW pte dirty bit */
190 #define _PAGE_READ	0x010		/* SW pte read bit */
191 #define _PAGE_WRITE	0x020		/* SW pte write bit */
192 #define _PAGE_SPECIAL	0x040		/* SW associated with special page */
193 #define _PAGE_UNUSED	0x080		/* SW bit for pgste usage state */
194 #define __HAVE_ARCH_PTE_SPECIAL
195 
196 /* Set of bits not changed in pte_modify */
197 #define _PAGE_CHG_MASK		(PAGE_MASK | _PAGE_SPECIAL | _PAGE_DIRTY | \
198 				 _PAGE_YOUNG)
199 
200 /*
201  * handle_pte_fault uses pte_present and pte_none to find out the pte type
202  * WITHOUT holding the page table lock. The _PAGE_PRESENT bit is used to
203  * distinguish present from not-present ptes. It is changed only with the page
204  * table lock held.
205  *
206  * The following table gives the different possible bit combinations for
207  * the pte hardware and software bits in the last 12 bits of a pte
208  * (. unassigned bit, x don't care, t swap type):
209  *
210  *				842100000000
211  *				000084210000
212  *				000000008421
213  *				.IR.uswrdy.p
214  * empty			.10.00000000
215  * swap				.11..ttttt.0
216  * prot-none, clean, old	.11.xx0000.1
217  * prot-none, clean, young	.11.xx0001.1
218  * prot-none, dirty, old	.10.xx0010.1
219  * prot-none, dirty, young	.10.xx0011.1
220  * read-only, clean, old	.11.xx0100.1
221  * read-only, clean, young	.01.xx0101.1
222  * read-only, dirty, old	.11.xx0110.1
223  * read-only, dirty, young	.01.xx0111.1
224  * read-write, clean, old	.11.xx1100.1
225  * read-write, clean, young	.01.xx1101.1
226  * read-write, dirty, old	.10.xx1110.1
227  * read-write, dirty, young	.00.xx1111.1
228  * HW-bits: R read-only, I invalid
229  * SW-bits: p present, y young, d dirty, r read, w write, s special,
230  *	    u unused, l large
231  *
232  * pte_none    is true for the bit pattern .10.00000000, pte == 0x400
233  * pte_swap    is true for the bit pattern .11..ooooo.0, (pte & 0x201) == 0x200
234  * pte_present is true for the bit pattern .xx.xxxxxx.1, (pte & 0x001) == 0x001
235  */
236 
237 /* Bits in the segment/region table address-space-control-element */
238 #define _ASCE_ORIGIN		~0xfffUL/* segment table origin		    */
239 #define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
240 #define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
241 #define _ASCE_SPACE_SWITCH	0x40	/* space switch event		    */
242 #define _ASCE_REAL_SPACE	0x20	/* real space control		    */
243 #define _ASCE_TYPE_MASK		0x0c	/* asce table type mask		    */
244 #define _ASCE_TYPE_REGION1	0x0c	/* region first table type	    */
245 #define _ASCE_TYPE_REGION2	0x08	/* region second table type	    */
246 #define _ASCE_TYPE_REGION3	0x04	/* region third table type	    */
247 #define _ASCE_TYPE_SEGMENT	0x00	/* segment table type		    */
248 #define _ASCE_TABLE_LENGTH	0x03	/* region table length		    */
249 
250 /* Bits in the region table entry */
251 #define _REGION_ENTRY_ORIGIN	~0xfffUL/* region/segment table origin	    */
252 #define _REGION_ENTRY_PROTECT	0x200	/* region protection bit	    */
253 #define _REGION_ENTRY_INVALID	0x20	/* invalid region table entry	    */
254 #define _REGION_ENTRY_TYPE_MASK	0x0c	/* region/segment table type mask   */
255 #define _REGION_ENTRY_TYPE_R1	0x0c	/* region first table type	    */
256 #define _REGION_ENTRY_TYPE_R2	0x08	/* region second table type	    */
257 #define _REGION_ENTRY_TYPE_R3	0x04	/* region third table type	    */
258 #define _REGION_ENTRY_LENGTH	0x03	/* region third length		    */
259 
260 #define _REGION1_ENTRY		(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_LENGTH)
261 #define _REGION1_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID)
262 #define _REGION2_ENTRY		(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_LENGTH)
263 #define _REGION2_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID)
264 #define _REGION3_ENTRY		(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_LENGTH)
265 #define _REGION3_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID)
266 
267 #define _REGION3_ENTRY_LARGE	0x400	/* RTTE-format control, large page  */
268 #define _REGION3_ENTRY_RO	0x200	/* page protection bit		    */
269 
270 /* Bits in the segment table entry */
271 #define _SEGMENT_ENTRY_BITS	0xfffffffffffffe33UL
272 #define _SEGMENT_ENTRY_BITS_LARGE 0xfffffffffff0ff33UL
273 #define _SEGMENT_ENTRY_ORIGIN_LARGE ~0xfffffUL /* large page address	    */
274 #define _SEGMENT_ENTRY_ORIGIN	~0x7ffUL/* segment table origin		    */
275 #define _SEGMENT_ENTRY_PROTECT	0x200	/* page protection bit		    */
276 #define _SEGMENT_ENTRY_INVALID	0x20	/* invalid segment table entry	    */
277 
278 #define _SEGMENT_ENTRY		(0)
279 #define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INVALID)
280 
281 #define _SEGMENT_ENTRY_DIRTY	0x2000	/* SW segment dirty bit */
282 #define _SEGMENT_ENTRY_YOUNG	0x1000	/* SW segment young bit */
283 #define _SEGMENT_ENTRY_SPLIT	0x0800	/* THP splitting bit */
284 #define _SEGMENT_ENTRY_LARGE	0x0400	/* STE-format control, large page */
285 #define _SEGMENT_ENTRY_READ	0x0002	/* SW segment read bit */
286 #define _SEGMENT_ENTRY_WRITE	0x0001	/* SW segment write bit */
287 
288 /*
289  * Segment table entry encoding (R = read-only, I = invalid, y = young bit):
290  *				dy..R...I...wr
291  * prot-none, clean, old	00..1...1...00
292  * prot-none, clean, young	01..1...1...00
293  * prot-none, dirty, old	10..1...1...00
294  * prot-none, dirty, young	11..1...1...00
295  * read-only, clean, old	00..1...1...01
296  * read-only, clean, young	01..1...0...01
297  * read-only, dirty, old	10..1...1...01
298  * read-only, dirty, young	11..1...0...01
299  * read-write, clean, old	00..1...1...11
300  * read-write, clean, young	01..1...0...11
301  * read-write, dirty, old	10..0...1...11
302  * read-write, dirty, young	11..0...0...11
303  * The segment table origin is used to distinguish empty (origin==0) from
304  * read-write, old segment table entries (origin!=0)
305  * HW-bits: R read-only, I invalid
306  * SW-bits: y young, d dirty, r read, w write
307  */
308 
309 #define _SEGMENT_ENTRY_SPLIT_BIT 11	/* THP splitting bit number */
310 
311 /* Page status table bits for virtualization */
312 #define PGSTE_ACC_BITS	0xf000000000000000UL
313 #define PGSTE_FP_BIT	0x0800000000000000UL
314 #define PGSTE_PCL_BIT	0x0080000000000000UL
315 #define PGSTE_HR_BIT	0x0040000000000000UL
316 #define PGSTE_HC_BIT	0x0020000000000000UL
317 #define PGSTE_GR_BIT	0x0004000000000000UL
318 #define PGSTE_GC_BIT	0x0002000000000000UL
319 #define PGSTE_UC_BIT	0x0000800000000000UL	/* user dirty (migration) */
320 #define PGSTE_IN_BIT	0x0000400000000000UL	/* IPTE notify bit */
321 
322 /* Guest Page State used for virtualization */
323 #define _PGSTE_GPS_ZERO		0x0000000080000000UL
324 #define _PGSTE_GPS_USAGE_MASK	0x0000000003000000UL
325 #define _PGSTE_GPS_USAGE_STABLE 0x0000000000000000UL
326 #define _PGSTE_GPS_USAGE_UNUSED 0x0000000001000000UL
327 
328 /*
329  * A user page table pointer has the space-switch-event bit, the
330  * private-space-control bit and the storage-alteration-event-control
331  * bit set. A kernel page table pointer doesn't need them.
332  */
333 #define _ASCE_USER_BITS		(_ASCE_SPACE_SWITCH | _ASCE_PRIVATE_SPACE | \
334 				 _ASCE_ALT_EVENT)
335 
336 /*
337  * Page protection definitions.
338  */
339 #define PAGE_NONE	__pgprot(_PAGE_PRESENT | _PAGE_INVALID)
340 #define PAGE_READ	__pgprot(_PAGE_PRESENT | _PAGE_READ | \
341 				 _PAGE_INVALID | _PAGE_PROTECT)
342 #define PAGE_WRITE	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
343 				 _PAGE_INVALID | _PAGE_PROTECT)
344 
345 #define PAGE_SHARED	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
346 				 _PAGE_YOUNG | _PAGE_DIRTY)
347 #define PAGE_KERNEL	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
348 				 _PAGE_YOUNG | _PAGE_DIRTY)
349 #define PAGE_KERNEL_RO	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_YOUNG | \
350 				 _PAGE_PROTECT)
351 
352 /*
353  * On s390 the page table entry has an invalid bit and a read-only bit.
354  * Read permission implies execute permission and write permission
355  * implies read permission.
356  */
357          /*xwr*/
358 #define __P000	PAGE_NONE
359 #define __P001	PAGE_READ
360 #define __P010	PAGE_READ
361 #define __P011	PAGE_READ
362 #define __P100	PAGE_READ
363 #define __P101	PAGE_READ
364 #define __P110	PAGE_READ
365 #define __P111	PAGE_READ
366 
367 #define __S000	PAGE_NONE
368 #define __S001	PAGE_READ
369 #define __S010	PAGE_WRITE
370 #define __S011	PAGE_WRITE
371 #define __S100	PAGE_READ
372 #define __S101	PAGE_READ
373 #define __S110	PAGE_WRITE
374 #define __S111	PAGE_WRITE
375 
376 /*
377  * Segment entry (large page) protection definitions.
378  */
379 #define SEGMENT_NONE	__pgprot(_SEGMENT_ENTRY_INVALID | \
380 				 _SEGMENT_ENTRY_PROTECT)
381 #define SEGMENT_READ	__pgprot(_SEGMENT_ENTRY_PROTECT | \
382 				 _SEGMENT_ENTRY_READ)
383 #define SEGMENT_WRITE	__pgprot(_SEGMENT_ENTRY_READ | \
384 				 _SEGMENT_ENTRY_WRITE)
385 
386 static inline int mm_has_pgste(struct mm_struct *mm)
387 {
388 #ifdef CONFIG_PGSTE
389 	if (unlikely(mm->context.has_pgste))
390 		return 1;
391 #endif
392 	return 0;
393 }
394 
395 static inline int mm_alloc_pgste(struct mm_struct *mm)
396 {
397 #ifdef CONFIG_PGSTE
398 	if (unlikely(mm->context.alloc_pgste))
399 		return 1;
400 #endif
401 	return 0;
402 }
403 
404 /*
405  * In the case that a guest uses storage keys
406  * faults should no longer be backed by zero pages
407  */
408 #define mm_forbids_zeropage mm_use_skey
409 static inline int mm_use_skey(struct mm_struct *mm)
410 {
411 #ifdef CONFIG_PGSTE
412 	if (mm->context.use_skey)
413 		return 1;
414 #endif
415 	return 0;
416 }
417 
418 /*
419  * pgd/pmd/pte query functions
420  */
421 static inline int pgd_present(pgd_t pgd)
422 {
423 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
424 		return 1;
425 	return (pgd_val(pgd) & _REGION_ENTRY_ORIGIN) != 0UL;
426 }
427 
428 static inline int pgd_none(pgd_t pgd)
429 {
430 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
431 		return 0;
432 	return (pgd_val(pgd) & _REGION_ENTRY_INVALID) != 0UL;
433 }
434 
435 static inline int pgd_bad(pgd_t pgd)
436 {
437 	/*
438 	 * With dynamic page table levels the pgd can be a region table
439 	 * entry or a segment table entry. Check for the bit that are
440 	 * invalid for either table entry.
441 	 */
442 	unsigned long mask =
443 		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INVALID &
444 		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
445 	return (pgd_val(pgd) & mask) != 0;
446 }
447 
448 static inline int pud_present(pud_t pud)
449 {
450 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
451 		return 1;
452 	return (pud_val(pud) & _REGION_ENTRY_ORIGIN) != 0UL;
453 }
454 
455 static inline int pud_none(pud_t pud)
456 {
457 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
458 		return 0;
459 	return (pud_val(pud) & _REGION_ENTRY_INVALID) != 0UL;
460 }
461 
462 static inline int pud_large(pud_t pud)
463 {
464 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) != _REGION_ENTRY_TYPE_R3)
465 		return 0;
466 	return !!(pud_val(pud) & _REGION3_ENTRY_LARGE);
467 }
468 
469 static inline int pud_bad(pud_t pud)
470 {
471 	/*
472 	 * With dynamic page table levels the pud can be a region table
473 	 * entry or a segment table entry. Check for the bit that are
474 	 * invalid for either table entry.
475 	 */
476 	unsigned long mask =
477 		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INVALID &
478 		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
479 	return (pud_val(pud) & mask) != 0;
480 }
481 
482 static inline int pmd_present(pmd_t pmd)
483 {
484 	return pmd_val(pmd) != _SEGMENT_ENTRY_INVALID;
485 }
486 
487 static inline int pmd_none(pmd_t pmd)
488 {
489 	return pmd_val(pmd) == _SEGMENT_ENTRY_INVALID;
490 }
491 
492 static inline int pmd_large(pmd_t pmd)
493 {
494 	return (pmd_val(pmd) & _SEGMENT_ENTRY_LARGE) != 0;
495 }
496 
497 static inline unsigned long pmd_pfn(pmd_t pmd)
498 {
499 	unsigned long origin_mask;
500 
501 	origin_mask = _SEGMENT_ENTRY_ORIGIN;
502 	if (pmd_large(pmd))
503 		origin_mask = _SEGMENT_ENTRY_ORIGIN_LARGE;
504 	return (pmd_val(pmd) & origin_mask) >> PAGE_SHIFT;
505 }
506 
507 static inline int pmd_bad(pmd_t pmd)
508 {
509 	if (pmd_large(pmd))
510 		return (pmd_val(pmd) & ~_SEGMENT_ENTRY_BITS_LARGE) != 0;
511 	return (pmd_val(pmd) & ~_SEGMENT_ENTRY_BITS) != 0;
512 }
513 
514 #define __HAVE_ARCH_PMDP_SPLITTING_FLUSH
515 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
516 				 unsigned long addr, pmd_t *pmdp);
517 
518 #define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
519 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
520 				 unsigned long address, pmd_t *pmdp,
521 				 pmd_t entry, int dirty);
522 
523 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
524 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
525 				  unsigned long address, pmd_t *pmdp);
526 
527 #define __HAVE_ARCH_PMD_WRITE
528 static inline int pmd_write(pmd_t pmd)
529 {
530 	return (pmd_val(pmd) & _SEGMENT_ENTRY_WRITE) != 0;
531 }
532 
533 static inline int pmd_dirty(pmd_t pmd)
534 {
535 	int dirty = 1;
536 	if (pmd_large(pmd))
537 		dirty = (pmd_val(pmd) & _SEGMENT_ENTRY_DIRTY) != 0;
538 	return dirty;
539 }
540 
541 static inline int pmd_young(pmd_t pmd)
542 {
543 	int young = 1;
544 	if (pmd_large(pmd))
545 		young = (pmd_val(pmd) & _SEGMENT_ENTRY_YOUNG) != 0;
546 	return young;
547 }
548 
549 static inline int pte_present(pte_t pte)
550 {
551 	/* Bit pattern: (pte & 0x001) == 0x001 */
552 	return (pte_val(pte) & _PAGE_PRESENT) != 0;
553 }
554 
555 static inline int pte_none(pte_t pte)
556 {
557 	/* Bit pattern: pte == 0x400 */
558 	return pte_val(pte) == _PAGE_INVALID;
559 }
560 
561 static inline int pte_swap(pte_t pte)
562 {
563 	/* Bit pattern: (pte & 0x201) == 0x200 */
564 	return (pte_val(pte) & (_PAGE_PROTECT | _PAGE_PRESENT))
565 		== _PAGE_PROTECT;
566 }
567 
568 static inline int pte_special(pte_t pte)
569 {
570 	return (pte_val(pte) & _PAGE_SPECIAL);
571 }
572 
573 #define __HAVE_ARCH_PTE_SAME
574 static inline int pte_same(pte_t a, pte_t b)
575 {
576 	return pte_val(a) == pte_val(b);
577 }
578 
579 static inline pgste_t pgste_get_lock(pte_t *ptep)
580 {
581 	unsigned long new = 0;
582 #ifdef CONFIG_PGSTE
583 	unsigned long old;
584 
585 	preempt_disable();
586 	asm(
587 		"	lg	%0,%2\n"
588 		"0:	lgr	%1,%0\n"
589 		"	nihh	%0,0xff7f\n"	/* clear PCL bit in old */
590 		"	oihh	%1,0x0080\n"	/* set PCL bit in new */
591 		"	csg	%0,%1,%2\n"
592 		"	jl	0b\n"
593 		: "=&d" (old), "=&d" (new), "=Q" (ptep[PTRS_PER_PTE])
594 		: "Q" (ptep[PTRS_PER_PTE]) : "cc", "memory");
595 #endif
596 	return __pgste(new);
597 }
598 
599 static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste)
600 {
601 #ifdef CONFIG_PGSTE
602 	asm(
603 		"	nihh	%1,0xff7f\n"	/* clear PCL bit */
604 		"	stg	%1,%0\n"
605 		: "=Q" (ptep[PTRS_PER_PTE])
606 		: "d" (pgste_val(pgste)), "Q" (ptep[PTRS_PER_PTE])
607 		: "cc", "memory");
608 	preempt_enable();
609 #endif
610 }
611 
612 static inline pgste_t pgste_get(pte_t *ptep)
613 {
614 	unsigned long pgste = 0;
615 #ifdef CONFIG_PGSTE
616 	pgste = *(unsigned long *)(ptep + PTRS_PER_PTE);
617 #endif
618 	return __pgste(pgste);
619 }
620 
621 static inline void pgste_set(pte_t *ptep, pgste_t pgste)
622 {
623 #ifdef CONFIG_PGSTE
624 	*(pgste_t *)(ptep + PTRS_PER_PTE) = pgste;
625 #endif
626 }
627 
628 static inline pgste_t pgste_update_all(pte_t *ptep, pgste_t pgste,
629 				       struct mm_struct *mm)
630 {
631 #ifdef CONFIG_PGSTE
632 	unsigned long address, bits, skey;
633 
634 	if (!mm_use_skey(mm) || pte_val(*ptep) & _PAGE_INVALID)
635 		return pgste;
636 	address = pte_val(*ptep) & PAGE_MASK;
637 	skey = (unsigned long) page_get_storage_key(address);
638 	bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
639 	/* Transfer page changed & referenced bit to guest bits in pgste */
640 	pgste_val(pgste) |= bits << 48;		/* GR bit & GC bit */
641 	/* Copy page access key and fetch protection bit to pgste */
642 	pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
643 	pgste_val(pgste) |= (skey & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
644 #endif
645 	return pgste;
646 
647 }
648 
649 static inline void pgste_set_key(pte_t *ptep, pgste_t pgste, pte_t entry,
650 				 struct mm_struct *mm)
651 {
652 #ifdef CONFIG_PGSTE
653 	unsigned long address;
654 	unsigned long nkey;
655 
656 	if (!mm_use_skey(mm) || pte_val(entry) & _PAGE_INVALID)
657 		return;
658 	VM_BUG_ON(!(pte_val(*ptep) & _PAGE_INVALID));
659 	address = pte_val(entry) & PAGE_MASK;
660 	/*
661 	 * Set page access key and fetch protection bit from pgste.
662 	 * The guest C/R information is still in the PGSTE, set real
663 	 * key C/R to 0.
664 	 */
665 	nkey = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
666 	nkey |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
667 	page_set_storage_key(address, nkey, 0);
668 #endif
669 }
670 
671 static inline pgste_t pgste_set_pte(pte_t *ptep, pgste_t pgste, pte_t entry)
672 {
673 	if ((pte_val(entry) & _PAGE_PRESENT) &&
674 	    (pte_val(entry) & _PAGE_WRITE) &&
675 	    !(pte_val(entry) & _PAGE_INVALID)) {
676 		if (!MACHINE_HAS_ESOP) {
677 			/*
678 			 * Without enhanced suppression-on-protection force
679 			 * the dirty bit on for all writable ptes.
680 			 */
681 			pte_val(entry) |= _PAGE_DIRTY;
682 			pte_val(entry) &= ~_PAGE_PROTECT;
683 		}
684 		if (!(pte_val(entry) & _PAGE_PROTECT))
685 			/* This pte allows write access, set user-dirty */
686 			pgste_val(pgste) |= PGSTE_UC_BIT;
687 	}
688 	*ptep = entry;
689 	return pgste;
690 }
691 
692 /**
693  * struct gmap_struct - guest address space
694  * @crst_list: list of all crst tables used in the guest address space
695  * @mm: pointer to the parent mm_struct
696  * @guest_to_host: radix tree with guest to host address translation
697  * @host_to_guest: radix tree with pointer to segment table entries
698  * @guest_table_lock: spinlock to protect all entries in the guest page table
699  * @table: pointer to the page directory
700  * @asce: address space control element for gmap page table
701  * @pfault_enabled: defines if pfaults are applicable for the guest
702  */
703 struct gmap {
704 	struct list_head list;
705 	struct list_head crst_list;
706 	struct mm_struct *mm;
707 	struct radix_tree_root guest_to_host;
708 	struct radix_tree_root host_to_guest;
709 	spinlock_t guest_table_lock;
710 	unsigned long *table;
711 	unsigned long asce;
712 	unsigned long asce_end;
713 	void *private;
714 	bool pfault_enabled;
715 };
716 
717 /**
718  * struct gmap_notifier - notify function block for page invalidation
719  * @notifier_call: address of callback function
720  */
721 struct gmap_notifier {
722 	struct list_head list;
723 	void (*notifier_call)(struct gmap *gmap, unsigned long gaddr);
724 };
725 
726 struct gmap *gmap_alloc(struct mm_struct *mm, unsigned long limit);
727 void gmap_free(struct gmap *gmap);
728 void gmap_enable(struct gmap *gmap);
729 void gmap_disable(struct gmap *gmap);
730 int gmap_map_segment(struct gmap *gmap, unsigned long from,
731 		     unsigned long to, unsigned long len);
732 int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len);
733 unsigned long __gmap_translate(struct gmap *, unsigned long gaddr);
734 unsigned long gmap_translate(struct gmap *, unsigned long gaddr);
735 int __gmap_link(struct gmap *gmap, unsigned long gaddr, unsigned long vmaddr);
736 int gmap_fault(struct gmap *, unsigned long gaddr, unsigned int fault_flags);
737 void gmap_discard(struct gmap *, unsigned long from, unsigned long to);
738 void __gmap_zap(struct gmap *, unsigned long gaddr);
739 bool gmap_test_and_clear_dirty(unsigned long address, struct gmap *);
740 
741 
742 void gmap_register_ipte_notifier(struct gmap_notifier *);
743 void gmap_unregister_ipte_notifier(struct gmap_notifier *);
744 int gmap_ipte_notify(struct gmap *, unsigned long start, unsigned long len);
745 void gmap_do_ipte_notify(struct mm_struct *, unsigned long addr, pte_t *);
746 
747 static inline pgste_t pgste_ipte_notify(struct mm_struct *mm,
748 					unsigned long addr,
749 					pte_t *ptep, pgste_t pgste)
750 {
751 #ifdef CONFIG_PGSTE
752 	if (pgste_val(pgste) & PGSTE_IN_BIT) {
753 		pgste_val(pgste) &= ~PGSTE_IN_BIT;
754 		gmap_do_ipte_notify(mm, addr, ptep);
755 	}
756 #endif
757 	return pgste;
758 }
759 
760 /*
761  * Certain architectures need to do special things when PTEs
762  * within a page table are directly modified.  Thus, the following
763  * hook is made available.
764  */
765 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
766 			      pte_t *ptep, pte_t entry)
767 {
768 	pgste_t pgste;
769 
770 	if (mm_has_pgste(mm)) {
771 		pgste = pgste_get_lock(ptep);
772 		pgste_val(pgste) &= ~_PGSTE_GPS_ZERO;
773 		pgste_set_key(ptep, pgste, entry, mm);
774 		pgste = pgste_set_pte(ptep, pgste, entry);
775 		pgste_set_unlock(ptep, pgste);
776 	} else {
777 		*ptep = entry;
778 	}
779 }
780 
781 /*
782  * query functions pte_write/pte_dirty/pte_young only work if
783  * pte_present() is true. Undefined behaviour if not..
784  */
785 static inline int pte_write(pte_t pte)
786 {
787 	return (pte_val(pte) & _PAGE_WRITE) != 0;
788 }
789 
790 static inline int pte_dirty(pte_t pte)
791 {
792 	return (pte_val(pte) & _PAGE_DIRTY) != 0;
793 }
794 
795 static inline int pte_young(pte_t pte)
796 {
797 	return (pte_val(pte) & _PAGE_YOUNG) != 0;
798 }
799 
800 #define __HAVE_ARCH_PTE_UNUSED
801 static inline int pte_unused(pte_t pte)
802 {
803 	return pte_val(pte) & _PAGE_UNUSED;
804 }
805 
806 /*
807  * pgd/pmd/pte modification functions
808  */
809 
810 static inline void pgd_clear(pgd_t *pgd)
811 {
812 	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
813 		pgd_val(*pgd) = _REGION2_ENTRY_EMPTY;
814 }
815 
816 static inline void pud_clear(pud_t *pud)
817 {
818 	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
819 		pud_val(*pud) = _REGION3_ENTRY_EMPTY;
820 }
821 
822 static inline void pmd_clear(pmd_t *pmdp)
823 {
824 	pmd_val(*pmdp) = _SEGMENT_ENTRY_INVALID;
825 }
826 
827 static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
828 {
829 	pte_val(*ptep) = _PAGE_INVALID;
830 }
831 
832 /*
833  * The following pte modification functions only work if
834  * pte_present() is true. Undefined behaviour if not..
835  */
836 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
837 {
838 	pte_val(pte) &= _PAGE_CHG_MASK;
839 	pte_val(pte) |= pgprot_val(newprot);
840 	/*
841 	 * newprot for PAGE_NONE, PAGE_READ and PAGE_WRITE has the
842 	 * invalid bit set, clear it again for readable, young pages
843 	 */
844 	if ((pte_val(pte) & _PAGE_YOUNG) && (pte_val(pte) & _PAGE_READ))
845 		pte_val(pte) &= ~_PAGE_INVALID;
846 	/*
847 	 * newprot for PAGE_READ and PAGE_WRITE has the page protection
848 	 * bit set, clear it again for writable, dirty pages
849 	 */
850 	if ((pte_val(pte) & _PAGE_DIRTY) && (pte_val(pte) & _PAGE_WRITE))
851 		pte_val(pte) &= ~_PAGE_PROTECT;
852 	return pte;
853 }
854 
855 static inline pte_t pte_wrprotect(pte_t pte)
856 {
857 	pte_val(pte) &= ~_PAGE_WRITE;
858 	pte_val(pte) |= _PAGE_PROTECT;
859 	return pte;
860 }
861 
862 static inline pte_t pte_mkwrite(pte_t pte)
863 {
864 	pte_val(pte) |= _PAGE_WRITE;
865 	if (pte_val(pte) & _PAGE_DIRTY)
866 		pte_val(pte) &= ~_PAGE_PROTECT;
867 	return pte;
868 }
869 
870 static inline pte_t pte_mkclean(pte_t pte)
871 {
872 	pte_val(pte) &= ~_PAGE_DIRTY;
873 	pte_val(pte) |= _PAGE_PROTECT;
874 	return pte;
875 }
876 
877 static inline pte_t pte_mkdirty(pte_t pte)
878 {
879 	pte_val(pte) |= _PAGE_DIRTY;
880 	if (pte_val(pte) & _PAGE_WRITE)
881 		pte_val(pte) &= ~_PAGE_PROTECT;
882 	return pte;
883 }
884 
885 static inline pte_t pte_mkold(pte_t pte)
886 {
887 	pte_val(pte) &= ~_PAGE_YOUNG;
888 	pte_val(pte) |= _PAGE_INVALID;
889 	return pte;
890 }
891 
892 static inline pte_t pte_mkyoung(pte_t pte)
893 {
894 	pte_val(pte) |= _PAGE_YOUNG;
895 	if (pte_val(pte) & _PAGE_READ)
896 		pte_val(pte) &= ~_PAGE_INVALID;
897 	return pte;
898 }
899 
900 static inline pte_t pte_mkspecial(pte_t pte)
901 {
902 	pte_val(pte) |= _PAGE_SPECIAL;
903 	return pte;
904 }
905 
906 #ifdef CONFIG_HUGETLB_PAGE
907 static inline pte_t pte_mkhuge(pte_t pte)
908 {
909 	pte_val(pte) |= _PAGE_LARGE;
910 	return pte;
911 }
912 #endif
913 
914 static inline void __ptep_ipte(unsigned long address, pte_t *ptep)
915 {
916 	unsigned long pto = (unsigned long) ptep;
917 
918 	/* Invalidation + global TLB flush for the pte */
919 	asm volatile(
920 		"	ipte	%2,%3"
921 		: "=m" (*ptep) : "m" (*ptep), "a" (pto), "a" (address));
922 }
923 
924 static inline void __ptep_ipte_local(unsigned long address, pte_t *ptep)
925 {
926 	unsigned long pto = (unsigned long) ptep;
927 
928 	/* Invalidation + local TLB flush for the pte */
929 	asm volatile(
930 		"	.insn rrf,0xb2210000,%2,%3,0,1"
931 		: "=m" (*ptep) : "m" (*ptep), "a" (pto), "a" (address));
932 }
933 
934 static inline void __ptep_ipte_range(unsigned long address, int nr, pte_t *ptep)
935 {
936 	unsigned long pto = (unsigned long) ptep;
937 
938 	/* Invalidate a range of ptes + global TLB flush of the ptes */
939 	do {
940 		asm volatile(
941 			"	.insn rrf,0xb2210000,%2,%0,%1,0"
942 			: "+a" (address), "+a" (nr) : "a" (pto) : "memory");
943 	} while (nr != 255);
944 }
945 
946 static inline void ptep_flush_direct(struct mm_struct *mm,
947 				     unsigned long address, pte_t *ptep)
948 {
949 	int active, count;
950 
951 	if (pte_val(*ptep) & _PAGE_INVALID)
952 		return;
953 	active = (mm == current->active_mm) ? 1 : 0;
954 	count = atomic_add_return(0x10000, &mm->context.attach_count);
955 	if (MACHINE_HAS_TLB_LC && (count & 0xffff) <= active &&
956 	    cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
957 		__ptep_ipte_local(address, ptep);
958 	else
959 		__ptep_ipte(address, ptep);
960 	atomic_sub(0x10000, &mm->context.attach_count);
961 }
962 
963 static inline void ptep_flush_lazy(struct mm_struct *mm,
964 				   unsigned long address, pte_t *ptep)
965 {
966 	int active, count;
967 
968 	if (pte_val(*ptep) & _PAGE_INVALID)
969 		return;
970 	active = (mm == current->active_mm) ? 1 : 0;
971 	count = atomic_add_return(0x10000, &mm->context.attach_count);
972 	if ((count & 0xffff) <= active) {
973 		pte_val(*ptep) |= _PAGE_INVALID;
974 		mm->context.flush_mm = 1;
975 	} else
976 		__ptep_ipte(address, ptep);
977 	atomic_sub(0x10000, &mm->context.attach_count);
978 }
979 
980 /*
981  * Get (and clear) the user dirty bit for a pte.
982  */
983 static inline int ptep_test_and_clear_user_dirty(struct mm_struct *mm,
984 						 unsigned long addr,
985 						 pte_t *ptep)
986 {
987 	pgste_t pgste;
988 	pte_t pte;
989 	int dirty;
990 
991 	if (!mm_has_pgste(mm))
992 		return 0;
993 	pgste = pgste_get_lock(ptep);
994 	dirty = !!(pgste_val(pgste) & PGSTE_UC_BIT);
995 	pgste_val(pgste) &= ~PGSTE_UC_BIT;
996 	pte = *ptep;
997 	if (dirty && (pte_val(pte) & _PAGE_PRESENT)) {
998 		pgste = pgste_ipte_notify(mm, addr, ptep, pgste);
999 		__ptep_ipte(addr, ptep);
1000 		if (MACHINE_HAS_ESOP || !(pte_val(pte) & _PAGE_WRITE))
1001 			pte_val(pte) |= _PAGE_PROTECT;
1002 		else
1003 			pte_val(pte) |= _PAGE_INVALID;
1004 		*ptep = pte;
1005 	}
1006 	pgste_set_unlock(ptep, pgste);
1007 	return dirty;
1008 }
1009 
1010 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1011 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
1012 					    unsigned long addr, pte_t *ptep)
1013 {
1014 	pgste_t pgste;
1015 	pte_t pte, oldpte;
1016 	int young;
1017 
1018 	if (mm_has_pgste(vma->vm_mm)) {
1019 		pgste = pgste_get_lock(ptep);
1020 		pgste = pgste_ipte_notify(vma->vm_mm, addr, ptep, pgste);
1021 	}
1022 
1023 	oldpte = pte = *ptep;
1024 	ptep_flush_direct(vma->vm_mm, addr, ptep);
1025 	young = pte_young(pte);
1026 	pte = pte_mkold(pte);
1027 
1028 	if (mm_has_pgste(vma->vm_mm)) {
1029 		pgste = pgste_update_all(&oldpte, pgste, vma->vm_mm);
1030 		pgste = pgste_set_pte(ptep, pgste, pte);
1031 		pgste_set_unlock(ptep, pgste);
1032 	} else
1033 		*ptep = pte;
1034 
1035 	return young;
1036 }
1037 
1038 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1039 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
1040 					 unsigned long address, pte_t *ptep)
1041 {
1042 	return ptep_test_and_clear_young(vma, address, ptep);
1043 }
1044 
1045 /*
1046  * This is hard to understand. ptep_get_and_clear and ptep_clear_flush
1047  * both clear the TLB for the unmapped pte. The reason is that
1048  * ptep_get_and_clear is used in common code (e.g. change_pte_range)
1049  * to modify an active pte. The sequence is
1050  *   1) ptep_get_and_clear
1051  *   2) set_pte_at
1052  *   3) flush_tlb_range
1053  * On s390 the tlb needs to get flushed with the modification of the pte
1054  * if the pte is active. The only way how this can be implemented is to
1055  * have ptep_get_and_clear do the tlb flush. In exchange flush_tlb_range
1056  * is a nop.
1057  */
1058 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1059 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
1060 				       unsigned long address, pte_t *ptep)
1061 {
1062 	pgste_t pgste;
1063 	pte_t pte;
1064 
1065 	if (mm_has_pgste(mm)) {
1066 		pgste = pgste_get_lock(ptep);
1067 		pgste = pgste_ipte_notify(mm, address, ptep, pgste);
1068 	}
1069 
1070 	pte = *ptep;
1071 	ptep_flush_lazy(mm, address, ptep);
1072 	pte_val(*ptep) = _PAGE_INVALID;
1073 
1074 	if (mm_has_pgste(mm)) {
1075 		pgste = pgste_update_all(&pte, pgste, mm);
1076 		pgste_set_unlock(ptep, pgste);
1077 	}
1078 	return pte;
1079 }
1080 
1081 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1082 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
1083 					   unsigned long address,
1084 					   pte_t *ptep)
1085 {
1086 	pgste_t pgste;
1087 	pte_t pte;
1088 
1089 	if (mm_has_pgste(mm)) {
1090 		pgste = pgste_get_lock(ptep);
1091 		pgste_ipte_notify(mm, address, ptep, pgste);
1092 	}
1093 
1094 	pte = *ptep;
1095 	ptep_flush_lazy(mm, address, ptep);
1096 
1097 	if (mm_has_pgste(mm)) {
1098 		pgste = pgste_update_all(&pte, pgste, mm);
1099 		pgste_set(ptep, pgste);
1100 	}
1101 	return pte;
1102 }
1103 
1104 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
1105 					   unsigned long address,
1106 					   pte_t *ptep, pte_t pte)
1107 {
1108 	pgste_t pgste;
1109 
1110 	if (mm_has_pgste(mm)) {
1111 		pgste = pgste_get(ptep);
1112 		pgste_set_key(ptep, pgste, pte, mm);
1113 		pgste = pgste_set_pte(ptep, pgste, pte);
1114 		pgste_set_unlock(ptep, pgste);
1115 	} else
1116 		*ptep = pte;
1117 }
1118 
1119 #define __HAVE_ARCH_PTEP_CLEAR_FLUSH
1120 static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
1121 				     unsigned long address, pte_t *ptep)
1122 {
1123 	pgste_t pgste;
1124 	pte_t pte;
1125 
1126 	if (mm_has_pgste(vma->vm_mm)) {
1127 		pgste = pgste_get_lock(ptep);
1128 		pgste = pgste_ipte_notify(vma->vm_mm, address, ptep, pgste);
1129 	}
1130 
1131 	pte = *ptep;
1132 	ptep_flush_direct(vma->vm_mm, address, ptep);
1133 	pte_val(*ptep) = _PAGE_INVALID;
1134 
1135 	if (mm_has_pgste(vma->vm_mm)) {
1136 		if ((pgste_val(pgste) & _PGSTE_GPS_USAGE_MASK) ==
1137 		    _PGSTE_GPS_USAGE_UNUSED)
1138 			pte_val(pte) |= _PAGE_UNUSED;
1139 		pgste = pgste_update_all(&pte, pgste, vma->vm_mm);
1140 		pgste_set_unlock(ptep, pgste);
1141 	}
1142 	return pte;
1143 }
1144 
1145 /*
1146  * The batched pte unmap code uses ptep_get_and_clear_full to clear the
1147  * ptes. Here an optimization is possible. tlb_gather_mmu flushes all
1148  * tlbs of an mm if it can guarantee that the ptes of the mm_struct
1149  * cannot be accessed while the batched unmap is running. In this case
1150  * full==1 and a simple pte_clear is enough. See tlb.h.
1151  */
1152 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
1153 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
1154 					    unsigned long address,
1155 					    pte_t *ptep, int full)
1156 {
1157 	pgste_t pgste;
1158 	pte_t pte;
1159 
1160 	if (!full && mm_has_pgste(mm)) {
1161 		pgste = pgste_get_lock(ptep);
1162 		pgste = pgste_ipte_notify(mm, address, ptep, pgste);
1163 	}
1164 
1165 	pte = *ptep;
1166 	if (!full)
1167 		ptep_flush_lazy(mm, address, ptep);
1168 	pte_val(*ptep) = _PAGE_INVALID;
1169 
1170 	if (!full && mm_has_pgste(mm)) {
1171 		pgste = pgste_update_all(&pte, pgste, mm);
1172 		pgste_set_unlock(ptep, pgste);
1173 	}
1174 	return pte;
1175 }
1176 
1177 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1178 static inline pte_t ptep_set_wrprotect(struct mm_struct *mm,
1179 				       unsigned long address, pte_t *ptep)
1180 {
1181 	pgste_t pgste;
1182 	pte_t pte = *ptep;
1183 
1184 	if (pte_write(pte)) {
1185 		if (mm_has_pgste(mm)) {
1186 			pgste = pgste_get_lock(ptep);
1187 			pgste = pgste_ipte_notify(mm, address, ptep, pgste);
1188 		}
1189 
1190 		ptep_flush_lazy(mm, address, ptep);
1191 		pte = pte_wrprotect(pte);
1192 
1193 		if (mm_has_pgste(mm)) {
1194 			pgste = pgste_set_pte(ptep, pgste, pte);
1195 			pgste_set_unlock(ptep, pgste);
1196 		} else
1197 			*ptep = pte;
1198 	}
1199 	return pte;
1200 }
1201 
1202 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1203 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
1204 					unsigned long address, pte_t *ptep,
1205 					pte_t entry, int dirty)
1206 {
1207 	pgste_t pgste;
1208 
1209 	if (pte_same(*ptep, entry))
1210 		return 0;
1211 	if (mm_has_pgste(vma->vm_mm)) {
1212 		pgste = pgste_get_lock(ptep);
1213 		pgste = pgste_ipte_notify(vma->vm_mm, address, ptep, pgste);
1214 	}
1215 
1216 	ptep_flush_direct(vma->vm_mm, address, ptep);
1217 
1218 	if (mm_has_pgste(vma->vm_mm)) {
1219 		pgste_set_key(ptep, pgste, entry, vma->vm_mm);
1220 		pgste = pgste_set_pte(ptep, pgste, entry);
1221 		pgste_set_unlock(ptep, pgste);
1222 	} else
1223 		*ptep = entry;
1224 	return 1;
1225 }
1226 
1227 /*
1228  * Conversion functions: convert a page and protection to a page entry,
1229  * and a page entry and page directory to the page they refer to.
1230  */
1231 static inline pte_t mk_pte_phys(unsigned long physpage, pgprot_t pgprot)
1232 {
1233 	pte_t __pte;
1234 	pte_val(__pte) = physpage + pgprot_val(pgprot);
1235 	return pte_mkyoung(__pte);
1236 }
1237 
1238 static inline pte_t mk_pte(struct page *page, pgprot_t pgprot)
1239 {
1240 	unsigned long physpage = page_to_phys(page);
1241 	pte_t __pte = mk_pte_phys(physpage, pgprot);
1242 
1243 	if (pte_write(__pte) && PageDirty(page))
1244 		__pte = pte_mkdirty(__pte);
1245 	return __pte;
1246 }
1247 
1248 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD-1))
1249 #define pud_index(address) (((address) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
1250 #define pmd_index(address) (((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1))
1251 #define pte_index(address) (((address) >> PAGE_SHIFT) & (PTRS_PER_PTE-1))
1252 
1253 #define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address))
1254 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
1255 
1256 #define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
1257 #define pud_deref(pud) (pud_val(pud) & _REGION_ENTRY_ORIGIN)
1258 #define pgd_deref(pgd) (pgd_val(pgd) & _REGION_ENTRY_ORIGIN)
1259 
1260 static inline pud_t *pud_offset(pgd_t *pgd, unsigned long address)
1261 {
1262 	pud_t *pud = (pud_t *) pgd;
1263 	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
1264 		pud = (pud_t *) pgd_deref(*pgd);
1265 	return pud  + pud_index(address);
1266 }
1267 
1268 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
1269 {
1270 	pmd_t *pmd = (pmd_t *) pud;
1271 	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
1272 		pmd = (pmd_t *) pud_deref(*pud);
1273 	return pmd + pmd_index(address);
1274 }
1275 
1276 #define pfn_pte(pfn,pgprot) mk_pte_phys(__pa((pfn) << PAGE_SHIFT),(pgprot))
1277 #define pte_pfn(x) (pte_val(x) >> PAGE_SHIFT)
1278 #define pte_page(x) pfn_to_page(pte_pfn(x))
1279 
1280 #define pmd_page(pmd) pfn_to_page(pmd_pfn(pmd))
1281 
1282 /* Find an entry in the lowest level page table.. */
1283 #define pte_offset(pmd, addr) ((pte_t *) pmd_deref(*(pmd)) + pte_index(addr))
1284 #define pte_offset_kernel(pmd, address) pte_offset(pmd,address)
1285 #define pte_offset_map(pmd, address) pte_offset_kernel(pmd, address)
1286 #define pte_unmap(pte) do { } while (0)
1287 
1288 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE)
1289 static inline unsigned long massage_pgprot_pmd(pgprot_t pgprot)
1290 {
1291 	/*
1292 	 * pgprot is PAGE_NONE, PAGE_READ, or PAGE_WRITE (see __Pxxx / __Sxxx)
1293 	 * Convert to segment table entry format.
1294 	 */
1295 	if (pgprot_val(pgprot) == pgprot_val(PAGE_NONE))
1296 		return pgprot_val(SEGMENT_NONE);
1297 	if (pgprot_val(pgprot) == pgprot_val(PAGE_READ))
1298 		return pgprot_val(SEGMENT_READ);
1299 	return pgprot_val(SEGMENT_WRITE);
1300 }
1301 
1302 static inline pmd_t pmd_wrprotect(pmd_t pmd)
1303 {
1304 	pmd_val(pmd) &= ~_SEGMENT_ENTRY_WRITE;
1305 	pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
1306 	return pmd;
1307 }
1308 
1309 static inline pmd_t pmd_mkwrite(pmd_t pmd)
1310 {
1311 	pmd_val(pmd) |= _SEGMENT_ENTRY_WRITE;
1312 	if (pmd_large(pmd) && !(pmd_val(pmd) & _SEGMENT_ENTRY_DIRTY))
1313 		return pmd;
1314 	pmd_val(pmd) &= ~_SEGMENT_ENTRY_PROTECT;
1315 	return pmd;
1316 }
1317 
1318 static inline pmd_t pmd_mkclean(pmd_t pmd)
1319 {
1320 	if (pmd_large(pmd)) {
1321 		pmd_val(pmd) &= ~_SEGMENT_ENTRY_DIRTY;
1322 		pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
1323 	}
1324 	return pmd;
1325 }
1326 
1327 static inline pmd_t pmd_mkdirty(pmd_t pmd)
1328 {
1329 	if (pmd_large(pmd)) {
1330 		pmd_val(pmd) |= _SEGMENT_ENTRY_DIRTY;
1331 		if (pmd_val(pmd) & _SEGMENT_ENTRY_WRITE)
1332 			pmd_val(pmd) &= ~_SEGMENT_ENTRY_PROTECT;
1333 	}
1334 	return pmd;
1335 }
1336 
1337 static inline pmd_t pmd_mkyoung(pmd_t pmd)
1338 {
1339 	if (pmd_large(pmd)) {
1340 		pmd_val(pmd) |= _SEGMENT_ENTRY_YOUNG;
1341 		if (pmd_val(pmd) & _SEGMENT_ENTRY_READ)
1342 			pmd_val(pmd) &= ~_SEGMENT_ENTRY_INVALID;
1343 	}
1344 	return pmd;
1345 }
1346 
1347 static inline pmd_t pmd_mkold(pmd_t pmd)
1348 {
1349 	if (pmd_large(pmd)) {
1350 		pmd_val(pmd) &= ~_SEGMENT_ENTRY_YOUNG;
1351 		pmd_val(pmd) |= _SEGMENT_ENTRY_INVALID;
1352 	}
1353 	return pmd;
1354 }
1355 
1356 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
1357 {
1358 	if (pmd_large(pmd)) {
1359 		pmd_val(pmd) &= _SEGMENT_ENTRY_ORIGIN_LARGE |
1360 			_SEGMENT_ENTRY_DIRTY | _SEGMENT_ENTRY_YOUNG |
1361 			_SEGMENT_ENTRY_LARGE | _SEGMENT_ENTRY_SPLIT;
1362 		pmd_val(pmd) |= massage_pgprot_pmd(newprot);
1363 		if (!(pmd_val(pmd) & _SEGMENT_ENTRY_DIRTY))
1364 			pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
1365 		if (!(pmd_val(pmd) & _SEGMENT_ENTRY_YOUNG))
1366 			pmd_val(pmd) |= _SEGMENT_ENTRY_INVALID;
1367 		return pmd;
1368 	}
1369 	pmd_val(pmd) &= _SEGMENT_ENTRY_ORIGIN;
1370 	pmd_val(pmd) |= massage_pgprot_pmd(newprot);
1371 	return pmd;
1372 }
1373 
1374 static inline pmd_t mk_pmd_phys(unsigned long physpage, pgprot_t pgprot)
1375 {
1376 	pmd_t __pmd;
1377 	pmd_val(__pmd) = physpage + massage_pgprot_pmd(pgprot);
1378 	return __pmd;
1379 }
1380 
1381 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLB_PAGE */
1382 
1383 static inline void __pmdp_csp(pmd_t *pmdp)
1384 {
1385 	register unsigned long reg2 asm("2") = pmd_val(*pmdp);
1386 	register unsigned long reg3 asm("3") = pmd_val(*pmdp) |
1387 					       _SEGMENT_ENTRY_INVALID;
1388 	register unsigned long reg4 asm("4") = ((unsigned long) pmdp) + 5;
1389 
1390 	asm volatile(
1391 		"	csp %1,%3"
1392 		: "=m" (*pmdp)
1393 		: "d" (reg2), "d" (reg3), "d" (reg4), "m" (*pmdp) : "cc");
1394 }
1395 
1396 static inline void __pmdp_idte(unsigned long address, pmd_t *pmdp)
1397 {
1398 	unsigned long sto;
1399 
1400 	sto = (unsigned long) pmdp - pmd_index(address) * sizeof(pmd_t);
1401 	asm volatile(
1402 		"	.insn	rrf,0xb98e0000,%2,%3,0,0"
1403 		: "=m" (*pmdp)
1404 		: "m" (*pmdp), "a" (sto), "a" ((address & HPAGE_MASK))
1405 		: "cc" );
1406 }
1407 
1408 static inline void __pmdp_idte_local(unsigned long address, pmd_t *pmdp)
1409 {
1410 	unsigned long sto;
1411 
1412 	sto = (unsigned long) pmdp - pmd_index(address) * sizeof(pmd_t);
1413 	asm volatile(
1414 		"	.insn	rrf,0xb98e0000,%2,%3,0,1"
1415 		: "=m" (*pmdp)
1416 		: "m" (*pmdp), "a" (sto), "a" ((address & HPAGE_MASK))
1417 		: "cc" );
1418 }
1419 
1420 static inline void pmdp_flush_direct(struct mm_struct *mm,
1421 				     unsigned long address, pmd_t *pmdp)
1422 {
1423 	int active, count;
1424 
1425 	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
1426 		return;
1427 	if (!MACHINE_HAS_IDTE) {
1428 		__pmdp_csp(pmdp);
1429 		return;
1430 	}
1431 	active = (mm == current->active_mm) ? 1 : 0;
1432 	count = atomic_add_return(0x10000, &mm->context.attach_count);
1433 	if (MACHINE_HAS_TLB_LC && (count & 0xffff) <= active &&
1434 	    cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
1435 		__pmdp_idte_local(address, pmdp);
1436 	else
1437 		__pmdp_idte(address, pmdp);
1438 	atomic_sub(0x10000, &mm->context.attach_count);
1439 }
1440 
1441 static inline void pmdp_flush_lazy(struct mm_struct *mm,
1442 				   unsigned long address, pmd_t *pmdp)
1443 {
1444 	int active, count;
1445 
1446 	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
1447 		return;
1448 	active = (mm == current->active_mm) ? 1 : 0;
1449 	count = atomic_add_return(0x10000, &mm->context.attach_count);
1450 	if ((count & 0xffff) <= active) {
1451 		pmd_val(*pmdp) |= _SEGMENT_ENTRY_INVALID;
1452 		mm->context.flush_mm = 1;
1453 	} else if (MACHINE_HAS_IDTE)
1454 		__pmdp_idte(address, pmdp);
1455 	else
1456 		__pmdp_csp(pmdp);
1457 	atomic_sub(0x10000, &mm->context.attach_count);
1458 }
1459 
1460 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1461 
1462 #define __HAVE_ARCH_PGTABLE_DEPOSIT
1463 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
1464 				       pgtable_t pgtable);
1465 
1466 #define __HAVE_ARCH_PGTABLE_WITHDRAW
1467 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
1468 
1469 static inline int pmd_trans_splitting(pmd_t pmd)
1470 {
1471 	return (pmd_val(pmd) & _SEGMENT_ENTRY_LARGE) &&
1472 		(pmd_val(pmd) & _SEGMENT_ENTRY_SPLIT);
1473 }
1474 
1475 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1476 			      pmd_t *pmdp, pmd_t entry)
1477 {
1478 	*pmdp = entry;
1479 }
1480 
1481 static inline pmd_t pmd_mkhuge(pmd_t pmd)
1482 {
1483 	pmd_val(pmd) |= _SEGMENT_ENTRY_LARGE;
1484 	pmd_val(pmd) |= _SEGMENT_ENTRY_YOUNG;
1485 	pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
1486 	return pmd;
1487 }
1488 
1489 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1490 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1491 					    unsigned long address, pmd_t *pmdp)
1492 {
1493 	pmd_t pmd;
1494 
1495 	pmd = *pmdp;
1496 	pmdp_flush_direct(vma->vm_mm, address, pmdp);
1497 	*pmdp = pmd_mkold(pmd);
1498 	return pmd_young(pmd);
1499 }
1500 
1501 #define __HAVE_ARCH_PMDP_GET_AND_CLEAR
1502 static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
1503 				       unsigned long address, pmd_t *pmdp)
1504 {
1505 	pmd_t pmd = *pmdp;
1506 
1507 	pmdp_flush_direct(mm, address, pmdp);
1508 	pmd_clear(pmdp);
1509 	return pmd;
1510 }
1511 
1512 #define __HAVE_ARCH_PMDP_GET_AND_CLEAR_FULL
1513 static inline pmd_t pmdp_get_and_clear_full(struct mm_struct *mm,
1514 					    unsigned long address,
1515 					    pmd_t *pmdp, int full)
1516 {
1517 	pmd_t pmd = *pmdp;
1518 
1519 	if (!full)
1520 		pmdp_flush_lazy(mm, address, pmdp);
1521 	pmd_clear(pmdp);
1522 	return pmd;
1523 }
1524 
1525 #define __HAVE_ARCH_PMDP_CLEAR_FLUSH
1526 static inline pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
1527 				     unsigned long address, pmd_t *pmdp)
1528 {
1529 	return pmdp_get_and_clear(vma->vm_mm, address, pmdp);
1530 }
1531 
1532 #define __HAVE_ARCH_PMDP_INVALIDATE
1533 static inline void pmdp_invalidate(struct vm_area_struct *vma,
1534 				   unsigned long address, pmd_t *pmdp)
1535 {
1536 	pmdp_flush_direct(vma->vm_mm, address, pmdp);
1537 }
1538 
1539 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1540 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1541 				      unsigned long address, pmd_t *pmdp)
1542 {
1543 	pmd_t pmd = *pmdp;
1544 
1545 	if (pmd_write(pmd)) {
1546 		pmdp_flush_direct(mm, address, pmdp);
1547 		set_pmd_at(mm, address, pmdp, pmd_wrprotect(pmd));
1548 	}
1549 }
1550 
1551 #define pfn_pmd(pfn, pgprot)	mk_pmd_phys(__pa((pfn) << PAGE_SHIFT), (pgprot))
1552 #define mk_pmd(page, pgprot)	pfn_pmd(page_to_pfn(page), (pgprot))
1553 
1554 static inline int pmd_trans_huge(pmd_t pmd)
1555 {
1556 	return pmd_val(pmd) & _SEGMENT_ENTRY_LARGE;
1557 }
1558 
1559 static inline int has_transparent_hugepage(void)
1560 {
1561 	return MACHINE_HAS_HPAGE ? 1 : 0;
1562 }
1563 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1564 
1565 /*
1566  * 64 bit swap entry format:
1567  * A page-table entry has some bits we have to treat in a special way.
1568  * Bits 52 and bit 55 have to be zero, otherwise an specification
1569  * exception will occur instead of a page translation exception. The
1570  * specifiation exception has the bad habit not to store necessary
1571  * information in the lowcore.
1572  * Bits 54 and 63 are used to indicate the page type.
1573  * A swap pte is indicated by bit pattern (pte & 0x201) == 0x200
1574  * This leaves the bits 0-51 and bits 56-62 to store type and offset.
1575  * We use the 5 bits from 57-61 for the type and the 52 bits from 0-51
1576  * for the offset.
1577  * |			  offset			|01100|type |00|
1578  * |0000000000111111111122222222223333333333444444444455|55555|55566|66|
1579  * |0123456789012345678901234567890123456789012345678901|23456|78901|23|
1580  */
1581 
1582 #define __SWP_OFFSET_MASK	((1UL << 52) - 1)
1583 #define __SWP_OFFSET_SHIFT	12
1584 #define __SWP_TYPE_MASK		((1UL << 5) - 1)
1585 #define __SWP_TYPE_SHIFT	2
1586 
1587 static inline pte_t mk_swap_pte(unsigned long type, unsigned long offset)
1588 {
1589 	pte_t pte;
1590 
1591 	pte_val(pte) = _PAGE_INVALID | _PAGE_PROTECT;
1592 	pte_val(pte) |= (offset & __SWP_OFFSET_MASK) << __SWP_OFFSET_SHIFT;
1593 	pte_val(pte) |= (type & __SWP_TYPE_MASK) << __SWP_TYPE_SHIFT;
1594 	return pte;
1595 }
1596 
1597 static inline unsigned long __swp_type(swp_entry_t entry)
1598 {
1599 	return (entry.val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK;
1600 }
1601 
1602 static inline unsigned long __swp_offset(swp_entry_t entry)
1603 {
1604 	return (entry.val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK;
1605 }
1606 
1607 static inline swp_entry_t __swp_entry(unsigned long type, unsigned long offset)
1608 {
1609 	return (swp_entry_t) { pte_val(mk_swap_pte(type, offset)) };
1610 }
1611 
1612 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
1613 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
1614 
1615 #endif /* !__ASSEMBLY__ */
1616 
1617 #define kern_addr_valid(addr)   (1)
1618 
1619 extern int vmem_add_mapping(unsigned long start, unsigned long size);
1620 extern int vmem_remove_mapping(unsigned long start, unsigned long size);
1621 extern int s390_enable_sie(void);
1622 extern int s390_enable_skey(void);
1623 extern void s390_reset_cmma(struct mm_struct *mm);
1624 
1625 /* s390 has a private copy of get unmapped area to deal with cache synonyms */
1626 #define HAVE_ARCH_UNMAPPED_AREA
1627 #define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1628 
1629 /*
1630  * No page table caches to initialise
1631  */
1632 static inline void pgtable_cache_init(void) { }
1633 static inline void check_pgt_cache(void) { }
1634 
1635 #include <asm-generic/pgtable.h>
1636 
1637 #endif /* _S390_PAGE_H */
1638