xref: /linux/arch/s390/include/asm/pgtable.h (revision f9c41a62bba3f3f7ef3541b2a025e3371bcbba97)
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. For
16  * s390 31 bit we "fold" the mid level into the top-level page table, so
17  * that we physically have the same two-level page table as the s390 mmu
18  * expects in 31 bit mode. For s390 64 bit we use three of the five levels
19  * the hardware provides (region first and region second tables are not
20  * used).
21  *
22  * The "pgd_xxx()" functions are trivial for a folded two-level
23  * setup: the pgd is never bad, and a pmd always exists (as it's folded
24  * into the pgd entry)
25  *
26  * This file contains the functions and defines necessary to modify and use
27  * the S390 page table tree.
28  */
29 #ifndef __ASSEMBLY__
30 #include <linux/sched.h>
31 #include <linux/mm_types.h>
32 #include <linux/page-flags.h>
33 #include <asm/bug.h>
34 #include <asm/page.h>
35 
36 extern pgd_t swapper_pg_dir[] __attribute__ ((aligned (4096)));
37 extern void paging_init(void);
38 extern void vmem_map_init(void);
39 
40 /*
41  * The S390 doesn't have any external MMU info: the kernel page
42  * tables contain all the necessary information.
43  */
44 #define update_mmu_cache(vma, address, ptep)     do { } while (0)
45 #define update_mmu_cache_pmd(vma, address, ptep) do { } while (0)
46 
47 /*
48  * ZERO_PAGE is a global shared page that is always zero; used
49  * for zero-mapped memory areas etc..
50  */
51 
52 extern unsigned long empty_zero_page;
53 extern unsigned long zero_page_mask;
54 
55 #define ZERO_PAGE(vaddr) \
56 	(virt_to_page((void *)(empty_zero_page + \
57 	 (((unsigned long)(vaddr)) &zero_page_mask))))
58 #define __HAVE_COLOR_ZERO_PAGE
59 
60 #endif /* !__ASSEMBLY__ */
61 
62 /*
63  * PMD_SHIFT determines the size of the area a second-level page
64  * table can map
65  * PGDIR_SHIFT determines what a third-level page table entry can map
66  */
67 #ifndef CONFIG_64BIT
68 # define PMD_SHIFT	20
69 # define PUD_SHIFT	20
70 # define PGDIR_SHIFT	20
71 #else /* CONFIG_64BIT */
72 # define PMD_SHIFT	20
73 # define PUD_SHIFT	31
74 # define PGDIR_SHIFT	42
75 #endif /* CONFIG_64BIT */
76 
77 #define PMD_SIZE        (1UL << PMD_SHIFT)
78 #define PMD_MASK        (~(PMD_SIZE-1))
79 #define PUD_SIZE	(1UL << PUD_SHIFT)
80 #define PUD_MASK	(~(PUD_SIZE-1))
81 #define PGDIR_SIZE	(1UL << PGDIR_SHIFT)
82 #define PGDIR_MASK	(~(PGDIR_SIZE-1))
83 
84 /*
85  * entries per page directory level: the S390 is two-level, so
86  * we don't really have any PMD directory physically.
87  * for S390 segment-table entries are combined to one PGD
88  * that leads to 1024 pte per pgd
89  */
90 #define PTRS_PER_PTE	256
91 #ifndef CONFIG_64BIT
92 #define PTRS_PER_PMD	1
93 #define PTRS_PER_PUD	1
94 #else /* CONFIG_64BIT */
95 #define PTRS_PER_PMD	2048
96 #define PTRS_PER_PUD	2048
97 #endif /* CONFIG_64BIT */
98 #define PTRS_PER_PGD	2048
99 
100 #define FIRST_USER_ADDRESS  0
101 
102 #define pte_ERROR(e) \
103 	printk("%s:%d: bad pte %p.\n", __FILE__, __LINE__, (void *) pte_val(e))
104 #define pmd_ERROR(e) \
105 	printk("%s:%d: bad pmd %p.\n", __FILE__, __LINE__, (void *) pmd_val(e))
106 #define pud_ERROR(e) \
107 	printk("%s:%d: bad pud %p.\n", __FILE__, __LINE__, (void *) pud_val(e))
108 #define pgd_ERROR(e) \
109 	printk("%s:%d: bad pgd %p.\n", __FILE__, __LINE__, (void *) pgd_val(e))
110 
111 #ifndef __ASSEMBLY__
112 /*
113  * The vmalloc and module area will always be on the topmost area of the kernel
114  * mapping. We reserve 96MB (31bit) / 128GB (64bit) for vmalloc and modules.
115  * On 64 bit kernels we have a 2GB area at the top of the vmalloc area where
116  * modules will reside. That makes sure that inter module branches always
117  * happen without trampolines and in addition the placement within a 2GB frame
118  * is branch prediction unit friendly.
119  */
120 extern unsigned long VMALLOC_START;
121 extern unsigned long VMALLOC_END;
122 extern struct page *vmemmap;
123 
124 #define VMEM_MAX_PHYS ((unsigned long) vmemmap)
125 
126 #ifdef CONFIG_64BIT
127 extern unsigned long MODULES_VADDR;
128 extern unsigned long MODULES_END;
129 #define MODULES_VADDR	MODULES_VADDR
130 #define MODULES_END	MODULES_END
131 #define MODULES_LEN	(1UL << 31)
132 #endif
133 
134 /*
135  * A 31 bit pagetable entry of S390 has following format:
136  *  |   PFRA          |    |  OS  |
137  * 0                   0IP0
138  * 00000000001111111111222222222233
139  * 01234567890123456789012345678901
140  *
141  * I Page-Invalid Bit:    Page is not available for address-translation
142  * P Page-Protection Bit: Store access not possible for page
143  *
144  * A 31 bit segmenttable entry of S390 has following format:
145  *  |   P-table origin      |  |PTL
146  * 0                         IC
147  * 00000000001111111111222222222233
148  * 01234567890123456789012345678901
149  *
150  * I Segment-Invalid Bit:    Segment is not available for address-translation
151  * C Common-Segment Bit:     Segment is not private (PoP 3-30)
152  * PTL Page-Table-Length:    Page-table length (PTL+1*16 entries -> up to 256)
153  *
154  * The 31 bit segmenttable origin of S390 has following format:
155  *
156  *  |S-table origin   |     | STL |
157  * X                   **GPS
158  * 00000000001111111111222222222233
159  * 01234567890123456789012345678901
160  *
161  * X Space-Switch event:
162  * G Segment-Invalid Bit:     *
163  * P Private-Space Bit:       Segment is not private (PoP 3-30)
164  * S Storage-Alteration:
165  * STL Segment-Table-Length:  Segment-table length (STL+1*16 entries -> up to 2048)
166  *
167  * A 64 bit pagetable entry of S390 has following format:
168  * |			 PFRA			      |0IPC|  OS  |
169  * 0000000000111111111122222222223333333333444444444455555555556666
170  * 0123456789012345678901234567890123456789012345678901234567890123
171  *
172  * I Page-Invalid Bit:    Page is not available for address-translation
173  * P Page-Protection Bit: Store access not possible for page
174  * C Change-bit override: HW is not required to set change bit
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_CO	0x100		/* HW Change-bit override */
219 #define _PAGE_RO	0x200		/* HW read-only bit  */
220 #define _PAGE_INVALID	0x400		/* HW invalid bit    */
221 
222 /* Software bits in the page table entry */
223 #define _PAGE_SWT	0x001		/* SW pte type bit t */
224 #define _PAGE_SWX	0x002		/* SW pte type bit x */
225 #define _PAGE_SWC	0x004		/* SW pte changed bit */
226 #define _PAGE_SWR	0x008		/* SW pte referenced bit */
227 #define _PAGE_SWW	0x010		/* SW pte write bit */
228 #define _PAGE_SPECIAL	0x020		/* SW associated with special page */
229 #define __HAVE_ARCH_PTE_SPECIAL
230 
231 /* Set of bits not changed in pte_modify */
232 #define _PAGE_CHG_MASK		(PAGE_MASK | _PAGE_SPECIAL | _PAGE_CO | \
233 				 _PAGE_SWC | _PAGE_SWR)
234 
235 /* Six different types of pages. */
236 #define _PAGE_TYPE_EMPTY	0x400
237 #define _PAGE_TYPE_NONE		0x401
238 #define _PAGE_TYPE_SWAP		0x403
239 #define _PAGE_TYPE_FILE		0x601	/* bit 0x002 is used for offset !! */
240 #define _PAGE_TYPE_RO		0x200
241 #define _PAGE_TYPE_RW		0x000
242 
243 /*
244  * Only four types for huge pages, using the invalid bit and protection bit
245  * of a segment table entry.
246  */
247 #define _HPAGE_TYPE_EMPTY	0x020	/* _SEGMENT_ENTRY_INV */
248 #define _HPAGE_TYPE_NONE	0x220
249 #define _HPAGE_TYPE_RO		0x200	/* _SEGMENT_ENTRY_RO  */
250 #define _HPAGE_TYPE_RW		0x000
251 
252 /*
253  * PTE type bits are rather complicated. handle_pte_fault uses pte_present,
254  * pte_none and pte_file to find out the pte type WITHOUT holding the page
255  * table lock. ptep_clear_flush on the other hand uses ptep_clear_flush to
256  * invalidate a given pte. ipte sets the hw invalid bit and clears all tlbs
257  * for the page. The page table entry is set to _PAGE_TYPE_EMPTY afterwards.
258  * This change is done while holding the lock, but the intermediate step
259  * of a previously valid pte with the hw invalid bit set can be observed by
260  * handle_pte_fault. That makes it necessary that all valid pte types with
261  * the hw invalid bit set must be distinguishable from the four pte types
262  * empty, none, swap and file.
263  *
264  *			irxt  ipte  irxt
265  * _PAGE_TYPE_EMPTY	1000   ->   1000
266  * _PAGE_TYPE_NONE	1001   ->   1001
267  * _PAGE_TYPE_SWAP	1011   ->   1011
268  * _PAGE_TYPE_FILE	11?1   ->   11?1
269  * _PAGE_TYPE_RO	0100   ->   1100
270  * _PAGE_TYPE_RW	0000   ->   1000
271  *
272  * pte_none is true for bits combinations 1000, 1010, 1100, 1110
273  * pte_present is true for bits combinations 0000, 0010, 0100, 0110, 1001
274  * pte_file is true for bits combinations 1101, 1111
275  * swap pte is 1011 and 0001, 0011, 0101, 0111 are invalid.
276  */
277 
278 #ifndef CONFIG_64BIT
279 
280 /* Bits in the segment table address-space-control-element */
281 #define _ASCE_SPACE_SWITCH	0x80000000UL	/* space switch event	    */
282 #define _ASCE_ORIGIN_MASK	0x7ffff000UL	/* segment table origin	    */
283 #define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
284 #define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
285 #define _ASCE_TABLE_LENGTH	0x7f	/* 128 x 64 entries = 8k	    */
286 
287 /* Bits in the segment table entry */
288 #define _SEGMENT_ENTRY_ORIGIN	0x7fffffc0UL	/* page table origin	    */
289 #define _SEGMENT_ENTRY_RO	0x200	/* page protection bit		    */
290 #define _SEGMENT_ENTRY_INV	0x20	/* invalid segment table entry	    */
291 #define _SEGMENT_ENTRY_COMMON	0x10	/* common segment bit		    */
292 #define _SEGMENT_ENTRY_PTL	0x0f	/* page table length		    */
293 
294 #define _SEGMENT_ENTRY		(_SEGMENT_ENTRY_PTL)
295 #define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INV)
296 
297 /* Page status table bits for virtualization */
298 #define RCP_ACC_BITS	0xf0000000UL
299 #define RCP_FP_BIT	0x08000000UL
300 #define RCP_PCL_BIT	0x00800000UL
301 #define RCP_HR_BIT	0x00400000UL
302 #define RCP_HC_BIT	0x00200000UL
303 #define RCP_GR_BIT	0x00040000UL
304 #define RCP_GC_BIT	0x00020000UL
305 
306 /* User dirty / referenced bit for KVM's migration feature */
307 #define KVM_UR_BIT	0x00008000UL
308 #define KVM_UC_BIT	0x00004000UL
309 
310 #else /* CONFIG_64BIT */
311 
312 /* Bits in the segment/region table address-space-control-element */
313 #define _ASCE_ORIGIN		~0xfffUL/* segment table origin		    */
314 #define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
315 #define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
316 #define _ASCE_SPACE_SWITCH	0x40	/* space switch event		    */
317 #define _ASCE_REAL_SPACE	0x20	/* real space control		    */
318 #define _ASCE_TYPE_MASK		0x0c	/* asce table type mask		    */
319 #define _ASCE_TYPE_REGION1	0x0c	/* region first table type	    */
320 #define _ASCE_TYPE_REGION2	0x08	/* region second table type	    */
321 #define _ASCE_TYPE_REGION3	0x04	/* region third table type	    */
322 #define _ASCE_TYPE_SEGMENT	0x00	/* segment table type		    */
323 #define _ASCE_TABLE_LENGTH	0x03	/* region table length		    */
324 
325 /* Bits in the region table entry */
326 #define _REGION_ENTRY_ORIGIN	~0xfffUL/* region/segment table origin	    */
327 #define _REGION_ENTRY_RO	0x200	/* region protection bit	    */
328 #define _REGION_ENTRY_INV	0x20	/* invalid region table entry	    */
329 #define _REGION_ENTRY_TYPE_MASK	0x0c	/* region/segment table type mask   */
330 #define _REGION_ENTRY_TYPE_R1	0x0c	/* region first table type	    */
331 #define _REGION_ENTRY_TYPE_R2	0x08	/* region second table type	    */
332 #define _REGION_ENTRY_TYPE_R3	0x04	/* region third table type	    */
333 #define _REGION_ENTRY_LENGTH	0x03	/* region third length		    */
334 
335 #define _REGION1_ENTRY		(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_LENGTH)
336 #define _REGION1_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INV)
337 #define _REGION2_ENTRY		(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_LENGTH)
338 #define _REGION2_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INV)
339 #define _REGION3_ENTRY		(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_LENGTH)
340 #define _REGION3_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INV)
341 
342 #define _REGION3_ENTRY_LARGE	0x400	/* RTTE-format control, large page  */
343 #define _REGION3_ENTRY_RO	0x200	/* page protection bit		    */
344 #define _REGION3_ENTRY_CO	0x100	/* change-recording override	    */
345 
346 /* Bits in the segment table entry */
347 #define _SEGMENT_ENTRY_ORIGIN_LARGE ~0xfffffUL /* large page address	    */
348 #define _SEGMENT_ENTRY_ORIGIN	~0x7ffUL/* segment table origin		    */
349 #define _SEGMENT_ENTRY_RO	0x200	/* page protection bit		    */
350 #define _SEGMENT_ENTRY_INV	0x20	/* invalid segment table entry	    */
351 
352 #define _SEGMENT_ENTRY		(0)
353 #define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INV)
354 
355 #define _SEGMENT_ENTRY_LARGE	0x400	/* STE-format control, large page   */
356 #define _SEGMENT_ENTRY_CO	0x100	/* change-recording override   */
357 #define _SEGMENT_ENTRY_SPLIT_BIT 0	/* THP splitting bit number */
358 #define _SEGMENT_ENTRY_SPLIT	(1UL << _SEGMENT_ENTRY_SPLIT_BIT)
359 
360 /* Set of bits not changed in pmd_modify */
361 #define _SEGMENT_CHG_MASK	(_SEGMENT_ENTRY_ORIGIN | _SEGMENT_ENTRY_LARGE \
362 				 | _SEGMENT_ENTRY_SPLIT | _SEGMENT_ENTRY_CO)
363 
364 /* Page status table bits for virtualization */
365 #define RCP_ACC_BITS	0xf000000000000000UL
366 #define RCP_FP_BIT	0x0800000000000000UL
367 #define RCP_PCL_BIT	0x0080000000000000UL
368 #define RCP_HR_BIT	0x0040000000000000UL
369 #define RCP_HC_BIT	0x0020000000000000UL
370 #define RCP_GR_BIT	0x0004000000000000UL
371 #define RCP_GC_BIT	0x0002000000000000UL
372 
373 /* User dirty / referenced bit for KVM's migration feature */
374 #define KVM_UR_BIT	0x0000800000000000UL
375 #define KVM_UC_BIT	0x0000400000000000UL
376 
377 #endif /* CONFIG_64BIT */
378 
379 /*
380  * A user page table pointer has the space-switch-event bit, the
381  * private-space-control bit and the storage-alteration-event-control
382  * bit set. A kernel page table pointer doesn't need them.
383  */
384 #define _ASCE_USER_BITS		(_ASCE_SPACE_SWITCH | _ASCE_PRIVATE_SPACE | \
385 				 _ASCE_ALT_EVENT)
386 
387 /*
388  * Page protection definitions.
389  */
390 #define PAGE_NONE	__pgprot(_PAGE_TYPE_NONE)
391 #define PAGE_RO		__pgprot(_PAGE_TYPE_RO)
392 #define PAGE_RW		__pgprot(_PAGE_TYPE_RO | _PAGE_SWW)
393 #define PAGE_RWC	__pgprot(_PAGE_TYPE_RW | _PAGE_SWW | _PAGE_SWC)
394 
395 #define PAGE_KERNEL	PAGE_RWC
396 #define PAGE_SHARED	PAGE_KERNEL
397 #define PAGE_COPY	PAGE_RO
398 
399 /*
400  * On s390 the page table entry has an invalid bit and a read-only bit.
401  * Read permission implies execute permission and write permission
402  * implies read permission.
403  */
404          /*xwr*/
405 #define __P000	PAGE_NONE
406 #define __P001	PAGE_RO
407 #define __P010	PAGE_RO
408 #define __P011	PAGE_RO
409 #define __P100	PAGE_RO
410 #define __P101	PAGE_RO
411 #define __P110	PAGE_RO
412 #define __P111	PAGE_RO
413 
414 #define __S000	PAGE_NONE
415 #define __S001	PAGE_RO
416 #define __S010	PAGE_RW
417 #define __S011	PAGE_RW
418 #define __S100	PAGE_RO
419 #define __S101	PAGE_RO
420 #define __S110	PAGE_RW
421 #define __S111	PAGE_RW
422 
423 static inline int mm_exclusive(struct mm_struct *mm)
424 {
425 	return likely(mm == current->active_mm &&
426 		      atomic_read(&mm->context.attach_count) <= 1);
427 }
428 
429 static inline int mm_has_pgste(struct mm_struct *mm)
430 {
431 #ifdef CONFIG_PGSTE
432 	if (unlikely(mm->context.has_pgste))
433 		return 1;
434 #endif
435 	return 0;
436 }
437 /*
438  * pgd/pmd/pte query functions
439  */
440 #ifndef CONFIG_64BIT
441 
442 static inline int pgd_present(pgd_t pgd) { return 1; }
443 static inline int pgd_none(pgd_t pgd)    { return 0; }
444 static inline int pgd_bad(pgd_t pgd)     { return 0; }
445 
446 static inline int pud_present(pud_t pud) { return 1; }
447 static inline int pud_none(pud_t pud)	 { return 0; }
448 static inline int pud_large(pud_t pud)	 { return 0; }
449 static inline int pud_bad(pud_t pud)	 { return 0; }
450 
451 #else /* CONFIG_64BIT */
452 
453 static inline int pgd_present(pgd_t pgd)
454 {
455 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
456 		return 1;
457 	return (pgd_val(pgd) & _REGION_ENTRY_ORIGIN) != 0UL;
458 }
459 
460 static inline int pgd_none(pgd_t pgd)
461 {
462 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
463 		return 0;
464 	return (pgd_val(pgd) & _REGION_ENTRY_INV) != 0UL;
465 }
466 
467 static inline int pgd_bad(pgd_t pgd)
468 {
469 	/*
470 	 * With dynamic page table levels the pgd can be a region table
471 	 * entry or a segment table entry. Check for the bit that are
472 	 * invalid for either table entry.
473 	 */
474 	unsigned long mask =
475 		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INV &
476 		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
477 	return (pgd_val(pgd) & mask) != 0;
478 }
479 
480 static inline int pud_present(pud_t pud)
481 {
482 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
483 		return 1;
484 	return (pud_val(pud) & _REGION_ENTRY_ORIGIN) != 0UL;
485 }
486 
487 static inline int pud_none(pud_t pud)
488 {
489 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
490 		return 0;
491 	return (pud_val(pud) & _REGION_ENTRY_INV) != 0UL;
492 }
493 
494 static inline int pud_large(pud_t pud)
495 {
496 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) != _REGION_ENTRY_TYPE_R3)
497 		return 0;
498 	return !!(pud_val(pud) & _REGION3_ENTRY_LARGE);
499 }
500 
501 static inline int pud_bad(pud_t pud)
502 {
503 	/*
504 	 * With dynamic page table levels the pud can be a region table
505 	 * entry or a segment table entry. Check for the bit that are
506 	 * invalid for either table entry.
507 	 */
508 	unsigned long mask =
509 		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INV &
510 		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
511 	return (pud_val(pud) & mask) != 0;
512 }
513 
514 #endif /* CONFIG_64BIT */
515 
516 static inline int pmd_present(pmd_t pmd)
517 {
518 	unsigned long mask = _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO;
519 	return (pmd_val(pmd) & mask) == _HPAGE_TYPE_NONE ||
520 	       !(pmd_val(pmd) & _SEGMENT_ENTRY_INV);
521 }
522 
523 static inline int pmd_none(pmd_t pmd)
524 {
525 	return (pmd_val(pmd) & _SEGMENT_ENTRY_INV) &&
526 	       !(pmd_val(pmd) & _SEGMENT_ENTRY_RO);
527 }
528 
529 static inline int pmd_large(pmd_t pmd)
530 {
531 #ifdef CONFIG_64BIT
532 	return !!(pmd_val(pmd) & _SEGMENT_ENTRY_LARGE);
533 #else
534 	return 0;
535 #endif
536 }
537 
538 static inline int pmd_bad(pmd_t pmd)
539 {
540 	unsigned long mask = ~_SEGMENT_ENTRY_ORIGIN & ~_SEGMENT_ENTRY_INV;
541 	return (pmd_val(pmd) & mask) != _SEGMENT_ENTRY;
542 }
543 
544 #define __HAVE_ARCH_PMDP_SPLITTING_FLUSH
545 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
546 				 unsigned long addr, pmd_t *pmdp);
547 
548 #define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
549 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
550 				 unsigned long address, pmd_t *pmdp,
551 				 pmd_t entry, int dirty);
552 
553 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
554 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
555 				  unsigned long address, pmd_t *pmdp);
556 
557 #define __HAVE_ARCH_PMD_WRITE
558 static inline int pmd_write(pmd_t pmd)
559 {
560 	return (pmd_val(pmd) & _SEGMENT_ENTRY_RO) == 0;
561 }
562 
563 static inline int pmd_young(pmd_t pmd)
564 {
565 	return 0;
566 }
567 
568 static inline int pte_none(pte_t pte)
569 {
570 	return (pte_val(pte) & _PAGE_INVALID) && !(pte_val(pte) & _PAGE_SWT);
571 }
572 
573 static inline int pte_present(pte_t pte)
574 {
575 	unsigned long mask = _PAGE_RO | _PAGE_INVALID | _PAGE_SWT | _PAGE_SWX;
576 	return (pte_val(pte) & mask) == _PAGE_TYPE_NONE ||
577 		(!(pte_val(pte) & _PAGE_INVALID) &&
578 		 !(pte_val(pte) & _PAGE_SWT));
579 }
580 
581 static inline int pte_file(pte_t pte)
582 {
583 	unsigned long mask = _PAGE_RO | _PAGE_INVALID | _PAGE_SWT;
584 	return (pte_val(pte) & mask) == _PAGE_TYPE_FILE;
585 }
586 
587 static inline int pte_special(pte_t pte)
588 {
589 	return (pte_val(pte) & _PAGE_SPECIAL);
590 }
591 
592 #define __HAVE_ARCH_PTE_SAME
593 static inline int pte_same(pte_t a, pte_t b)
594 {
595 	return pte_val(a) == pte_val(b);
596 }
597 
598 static inline pgste_t pgste_get_lock(pte_t *ptep)
599 {
600 	unsigned long new = 0;
601 #ifdef CONFIG_PGSTE
602 	unsigned long old;
603 
604 	preempt_disable();
605 	asm(
606 		"	lg	%0,%2\n"
607 		"0:	lgr	%1,%0\n"
608 		"	nihh	%0,0xff7f\n"	/* clear RCP_PCL_BIT in old */
609 		"	oihh	%1,0x0080\n"	/* set RCP_PCL_BIT in new */
610 		"	csg	%0,%1,%2\n"
611 		"	jl	0b\n"
612 		: "=&d" (old), "=&d" (new), "=Q" (ptep[PTRS_PER_PTE])
613 		: "Q" (ptep[PTRS_PER_PTE]) : "cc");
614 #endif
615 	return __pgste(new);
616 }
617 
618 static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste)
619 {
620 #ifdef CONFIG_PGSTE
621 	asm(
622 		"	nihh	%1,0xff7f\n"	/* clear RCP_PCL_BIT */
623 		"	stg	%1,%0\n"
624 		: "=Q" (ptep[PTRS_PER_PTE])
625 		: "d" (pgste_val(pgste)), "Q" (ptep[PTRS_PER_PTE]) : "cc");
626 	preempt_enable();
627 #endif
628 }
629 
630 static inline pgste_t pgste_update_all(pte_t *ptep, pgste_t pgste)
631 {
632 #ifdef CONFIG_PGSTE
633 	unsigned long address, bits;
634 	unsigned char skey;
635 
636 	if (!pte_present(*ptep))
637 		return pgste;
638 	address = pte_val(*ptep) & PAGE_MASK;
639 	skey = page_get_storage_key(address);
640 	bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
641 	/* Clear page changed & referenced bit in the storage key */
642 	if (bits & _PAGE_CHANGED)
643 		page_set_storage_key(address, skey ^ bits, 0);
644 	else if (bits)
645 		page_reset_referenced(address);
646 	/* Transfer page changed & referenced bit to guest bits in pgste */
647 	pgste_val(pgste) |= bits << 48;		/* RCP_GR_BIT & RCP_GC_BIT */
648 	/* Get host changed & referenced bits from pgste */
649 	bits |= (pgste_val(pgste) & (RCP_HR_BIT | RCP_HC_BIT)) >> 52;
650 	/* Transfer page changed & referenced bit to kvm user bits */
651 	pgste_val(pgste) |= bits << 45;		/* KVM_UR_BIT & KVM_UC_BIT */
652 	/* Clear relevant host bits in pgste. */
653 	pgste_val(pgste) &= ~(RCP_HR_BIT | RCP_HC_BIT);
654 	pgste_val(pgste) &= ~(RCP_ACC_BITS | RCP_FP_BIT);
655 	/* Copy page access key and fetch protection bit to pgste */
656 	pgste_val(pgste) |=
657 		(unsigned long) (skey & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
658 	/* Transfer referenced bit to pte */
659 	pte_val(*ptep) |= (bits & _PAGE_REFERENCED) << 1;
660 #endif
661 	return pgste;
662 
663 }
664 
665 static inline pgste_t pgste_update_young(pte_t *ptep, pgste_t pgste)
666 {
667 #ifdef CONFIG_PGSTE
668 	int young;
669 
670 	if (!pte_present(*ptep))
671 		return pgste;
672 	/* Get referenced bit from storage key */
673 	young = page_reset_referenced(pte_val(*ptep) & PAGE_MASK);
674 	if (young)
675 		pgste_val(pgste) |= RCP_GR_BIT;
676 	/* Get host referenced bit from pgste */
677 	if (pgste_val(pgste) & RCP_HR_BIT) {
678 		pgste_val(pgste) &= ~RCP_HR_BIT;
679 		young = 1;
680 	}
681 	/* Transfer referenced bit to kvm user bits and pte */
682 	if (young) {
683 		pgste_val(pgste) |= KVM_UR_BIT;
684 		pte_val(*ptep) |= _PAGE_SWR;
685 	}
686 #endif
687 	return pgste;
688 }
689 
690 static inline void pgste_set_key(pte_t *ptep, pgste_t pgste, pte_t entry)
691 {
692 #ifdef CONFIG_PGSTE
693 	unsigned long address;
694 	unsigned long okey, nkey;
695 
696 	if (!pte_present(entry))
697 		return;
698 	address = pte_val(entry) & PAGE_MASK;
699 	okey = nkey = page_get_storage_key(address);
700 	nkey &= ~(_PAGE_ACC_BITS | _PAGE_FP_BIT);
701 	/* Set page access key and fetch protection bit from pgste */
702 	nkey |= (pgste_val(pgste) & (RCP_ACC_BITS | RCP_FP_BIT)) >> 56;
703 	if (okey != nkey)
704 		page_set_storage_key(address, nkey, 0);
705 #endif
706 }
707 
708 static inline void pgste_set_pte(pte_t *ptep, pte_t entry)
709 {
710 	if (!MACHINE_HAS_ESOP && (pte_val(entry) & _PAGE_SWW)) {
711 		/*
712 		 * Without enhanced suppression-on-protection force
713 		 * the dirty bit on for all writable ptes.
714 		 */
715 		pte_val(entry) |= _PAGE_SWC;
716 		pte_val(entry) &= ~_PAGE_RO;
717 	}
718 	*ptep = entry;
719 }
720 
721 /**
722  * struct gmap_struct - guest address space
723  * @mm: pointer to the parent mm_struct
724  * @table: pointer to the page directory
725  * @asce: address space control element for gmap page table
726  * @crst_list: list of all crst tables used in the guest address space
727  */
728 struct gmap {
729 	struct list_head list;
730 	struct mm_struct *mm;
731 	unsigned long *table;
732 	unsigned long asce;
733 	struct list_head crst_list;
734 };
735 
736 /**
737  * struct gmap_rmap - reverse mapping for segment table entries
738  * @next: pointer to the next gmap_rmap structure in the list
739  * @entry: pointer to a segment table entry
740  */
741 struct gmap_rmap {
742 	struct list_head list;
743 	unsigned long *entry;
744 };
745 
746 /**
747  * struct gmap_pgtable - gmap information attached to a page table
748  * @vmaddr: address of the 1MB segment in the process virtual memory
749  * @mapper: list of segment table entries maping a page table
750  */
751 struct gmap_pgtable {
752 	unsigned long vmaddr;
753 	struct list_head mapper;
754 };
755 
756 struct gmap *gmap_alloc(struct mm_struct *mm);
757 void gmap_free(struct gmap *gmap);
758 void gmap_enable(struct gmap *gmap);
759 void gmap_disable(struct gmap *gmap);
760 int gmap_map_segment(struct gmap *gmap, unsigned long from,
761 		     unsigned long to, unsigned long length);
762 int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len);
763 unsigned long __gmap_fault(unsigned long address, struct gmap *);
764 unsigned long gmap_fault(unsigned long address, struct gmap *);
765 void gmap_discard(unsigned long from, unsigned long to, struct gmap *);
766 
767 /*
768  * Certain architectures need to do special things when PTEs
769  * within a page table are directly modified.  Thus, the following
770  * hook is made available.
771  */
772 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
773 			      pte_t *ptep, pte_t entry)
774 {
775 	pgste_t pgste;
776 
777 	if (mm_has_pgste(mm)) {
778 		pgste = pgste_get_lock(ptep);
779 		pgste_set_key(ptep, pgste, entry);
780 		pgste_set_pte(ptep, entry);
781 		pgste_set_unlock(ptep, pgste);
782 	} else {
783 		if (!(pte_val(entry) & _PAGE_INVALID) && MACHINE_HAS_EDAT1)
784 			pte_val(entry) |= _PAGE_CO;
785 		*ptep = entry;
786 	}
787 }
788 
789 /*
790  * query functions pte_write/pte_dirty/pte_young only work if
791  * pte_present() is true. Undefined behaviour if not..
792  */
793 static inline int pte_write(pte_t pte)
794 {
795 	return (pte_val(pte) & _PAGE_SWW) != 0;
796 }
797 
798 static inline int pte_dirty(pte_t pte)
799 {
800 	return (pte_val(pte) & _PAGE_SWC) != 0;
801 }
802 
803 static inline int pte_young(pte_t pte)
804 {
805 #ifdef CONFIG_PGSTE
806 	if (pte_val(pte) & _PAGE_SWR)
807 		return 1;
808 #endif
809 	return 0;
810 }
811 
812 /*
813  * pgd/pmd/pte modification functions
814  */
815 
816 static inline void pgd_clear(pgd_t *pgd)
817 {
818 #ifdef CONFIG_64BIT
819 	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
820 		pgd_val(*pgd) = _REGION2_ENTRY_EMPTY;
821 #endif
822 }
823 
824 static inline void pud_clear(pud_t *pud)
825 {
826 #ifdef CONFIG_64BIT
827 	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
828 		pud_val(*pud) = _REGION3_ENTRY_EMPTY;
829 #endif
830 }
831 
832 static inline void pmd_clear(pmd_t *pmdp)
833 {
834 	pmd_val(*pmdp) = _SEGMENT_ENTRY_EMPTY;
835 }
836 
837 static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
838 {
839 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
840 }
841 
842 /*
843  * The following pte modification functions only work if
844  * pte_present() is true. Undefined behaviour if not..
845  */
846 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
847 {
848 	pte_val(pte) &= _PAGE_CHG_MASK;
849 	pte_val(pte) |= pgprot_val(newprot);
850 	if ((pte_val(pte) & _PAGE_SWC) && (pte_val(pte) & _PAGE_SWW))
851 		pte_val(pte) &= ~_PAGE_RO;
852 	return pte;
853 }
854 
855 static inline pte_t pte_wrprotect(pte_t pte)
856 {
857 	pte_val(pte) &= ~_PAGE_SWW;
858 	/* Do not clobber _PAGE_TYPE_NONE pages!  */
859 	if (!(pte_val(pte) & _PAGE_INVALID))
860 		pte_val(pte) |= _PAGE_RO;
861 	return pte;
862 }
863 
864 static inline pte_t pte_mkwrite(pte_t pte)
865 {
866 	pte_val(pte) |= _PAGE_SWW;
867 	if (pte_val(pte) & _PAGE_SWC)
868 		pte_val(pte) &= ~_PAGE_RO;
869 	return pte;
870 }
871 
872 static inline pte_t pte_mkclean(pte_t pte)
873 {
874 	pte_val(pte) &= ~_PAGE_SWC;
875 	/* Do not clobber _PAGE_TYPE_NONE pages!  */
876 	if (!(pte_val(pte) & _PAGE_INVALID))
877 		pte_val(pte) |= _PAGE_RO;
878 	return pte;
879 }
880 
881 static inline pte_t pte_mkdirty(pte_t pte)
882 {
883 	pte_val(pte) |= _PAGE_SWC;
884 	if (pte_val(pte) & _PAGE_SWW)
885 		pte_val(pte) &= ~_PAGE_RO;
886 	return pte;
887 }
888 
889 static inline pte_t pte_mkold(pte_t pte)
890 {
891 #ifdef CONFIG_PGSTE
892 	pte_val(pte) &= ~_PAGE_SWR;
893 #endif
894 	return pte;
895 }
896 
897 static inline pte_t pte_mkyoung(pte_t pte)
898 {
899 	return pte;
900 }
901 
902 static inline pte_t pte_mkspecial(pte_t pte)
903 {
904 	pte_val(pte) |= _PAGE_SPECIAL;
905 	return pte;
906 }
907 
908 #ifdef CONFIG_HUGETLB_PAGE
909 static inline pte_t pte_mkhuge(pte_t pte)
910 {
911 	/*
912 	 * PROT_NONE needs to be remapped from the pte type to the ste type.
913 	 * The HW invalid bit is also different for pte and ste. The pte
914 	 * invalid bit happens to be the same as the ste _SEGMENT_ENTRY_LARGE
915 	 * bit, so we don't have to clear it.
916 	 */
917 	if (pte_val(pte) & _PAGE_INVALID) {
918 		if (pte_val(pte) & _PAGE_SWT)
919 			pte_val(pte) |= _HPAGE_TYPE_NONE;
920 		pte_val(pte) |= _SEGMENT_ENTRY_INV;
921 	}
922 	/*
923 	 * Clear SW pte bits, there are no SW bits in a segment table entry.
924 	 */
925 	pte_val(pte) &= ~(_PAGE_SWT | _PAGE_SWX | _PAGE_SWC |
926 			  _PAGE_SWR | _PAGE_SWW);
927 	/*
928 	 * Also set the change-override bit because we don't need dirty bit
929 	 * tracking for hugetlbfs pages.
930 	 */
931 	pte_val(pte) |= (_SEGMENT_ENTRY_LARGE | _SEGMENT_ENTRY_CO);
932 	return pte;
933 }
934 #endif
935 
936 /*
937  * Get (and clear) the user dirty bit for a pte.
938  */
939 static inline int ptep_test_and_clear_user_dirty(struct mm_struct *mm,
940 						 pte_t *ptep)
941 {
942 	pgste_t pgste;
943 	int dirty = 0;
944 
945 	if (mm_has_pgste(mm)) {
946 		pgste = pgste_get_lock(ptep);
947 		pgste = pgste_update_all(ptep, pgste);
948 		dirty = !!(pgste_val(pgste) & KVM_UC_BIT);
949 		pgste_val(pgste) &= ~KVM_UC_BIT;
950 		pgste_set_unlock(ptep, pgste);
951 		return dirty;
952 	}
953 	return dirty;
954 }
955 
956 /*
957  * Get (and clear) the user referenced bit for a pte.
958  */
959 static inline int ptep_test_and_clear_user_young(struct mm_struct *mm,
960 						 pte_t *ptep)
961 {
962 	pgste_t pgste;
963 	int young = 0;
964 
965 	if (mm_has_pgste(mm)) {
966 		pgste = pgste_get_lock(ptep);
967 		pgste = pgste_update_young(ptep, pgste);
968 		young = !!(pgste_val(pgste) & KVM_UR_BIT);
969 		pgste_val(pgste) &= ~KVM_UR_BIT;
970 		pgste_set_unlock(ptep, pgste);
971 	}
972 	return young;
973 }
974 
975 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
976 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
977 					    unsigned long addr, pte_t *ptep)
978 {
979 	pgste_t pgste;
980 	pte_t pte;
981 
982 	if (mm_has_pgste(vma->vm_mm)) {
983 		pgste = pgste_get_lock(ptep);
984 		pgste = pgste_update_young(ptep, pgste);
985 		pte = *ptep;
986 		*ptep = pte_mkold(pte);
987 		pgste_set_unlock(ptep, pgste);
988 		return pte_young(pte);
989 	}
990 	return 0;
991 }
992 
993 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
994 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
995 					 unsigned long address, pte_t *ptep)
996 {
997 	/* No need to flush TLB
998 	 * On s390 reference bits are in storage key and never in TLB
999 	 * With virtualization we handle the reference bit, without we
1000 	 * we can simply return */
1001 	return ptep_test_and_clear_young(vma, address, ptep);
1002 }
1003 
1004 static inline void __ptep_ipte(unsigned long address, pte_t *ptep)
1005 {
1006 	if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1007 #ifndef CONFIG_64BIT
1008 		/* pto must point to the start of the segment table */
1009 		pte_t *pto = (pte_t *) (((unsigned long) ptep) & 0x7ffffc00);
1010 #else
1011 		/* ipte in zarch mode can do the math */
1012 		pte_t *pto = ptep;
1013 #endif
1014 		asm volatile(
1015 			"	ipte	%2,%3"
1016 			: "=m" (*ptep) : "m" (*ptep),
1017 			  "a" (pto), "a" (address));
1018 	}
1019 }
1020 
1021 /*
1022  * This is hard to understand. ptep_get_and_clear and ptep_clear_flush
1023  * both clear the TLB for the unmapped pte. The reason is that
1024  * ptep_get_and_clear is used in common code (e.g. change_pte_range)
1025  * to modify an active pte. The sequence is
1026  *   1) ptep_get_and_clear
1027  *   2) set_pte_at
1028  *   3) flush_tlb_range
1029  * On s390 the tlb needs to get flushed with the modification of the pte
1030  * if the pte is active. The only way how this can be implemented is to
1031  * have ptep_get_and_clear do the tlb flush. In exchange flush_tlb_range
1032  * is a nop.
1033  */
1034 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1035 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
1036 				       unsigned long address, pte_t *ptep)
1037 {
1038 	pgste_t pgste;
1039 	pte_t pte;
1040 
1041 	mm->context.flush_mm = 1;
1042 	if (mm_has_pgste(mm))
1043 		pgste = pgste_get_lock(ptep);
1044 
1045 	pte = *ptep;
1046 	if (!mm_exclusive(mm))
1047 		__ptep_ipte(address, ptep);
1048 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
1049 
1050 	if (mm_has_pgste(mm)) {
1051 		pgste = pgste_update_all(&pte, pgste);
1052 		pgste_set_unlock(ptep, pgste);
1053 	}
1054 	return pte;
1055 }
1056 
1057 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1058 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
1059 					   unsigned long address,
1060 					   pte_t *ptep)
1061 {
1062 	pte_t pte;
1063 
1064 	mm->context.flush_mm = 1;
1065 	if (mm_has_pgste(mm))
1066 		pgste_get_lock(ptep);
1067 
1068 	pte = *ptep;
1069 	if (!mm_exclusive(mm))
1070 		__ptep_ipte(address, ptep);
1071 	return pte;
1072 }
1073 
1074 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
1075 					   unsigned long address,
1076 					   pte_t *ptep, pte_t pte)
1077 {
1078 	if (mm_has_pgste(mm)) {
1079 		pgste_set_pte(ptep, pte);
1080 		pgste_set_unlock(ptep, *(pgste_t *)(ptep + PTRS_PER_PTE));
1081 	} else
1082 		*ptep = pte;
1083 }
1084 
1085 #define __HAVE_ARCH_PTEP_CLEAR_FLUSH
1086 static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
1087 				     unsigned long address, pte_t *ptep)
1088 {
1089 	pgste_t pgste;
1090 	pte_t pte;
1091 
1092 	if (mm_has_pgste(vma->vm_mm))
1093 		pgste = pgste_get_lock(ptep);
1094 
1095 	pte = *ptep;
1096 	__ptep_ipte(address, ptep);
1097 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
1098 
1099 	if (mm_has_pgste(vma->vm_mm)) {
1100 		pgste = pgste_update_all(&pte, pgste);
1101 		pgste_set_unlock(ptep, pgste);
1102 	}
1103 	return pte;
1104 }
1105 
1106 /*
1107  * The batched pte unmap code uses ptep_get_and_clear_full to clear the
1108  * ptes. Here an optimization is possible. tlb_gather_mmu flushes all
1109  * tlbs of an mm if it can guarantee that the ptes of the mm_struct
1110  * cannot be accessed while the batched unmap is running. In this case
1111  * full==1 and a simple pte_clear is enough. See tlb.h.
1112  */
1113 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
1114 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
1115 					    unsigned long address,
1116 					    pte_t *ptep, int full)
1117 {
1118 	pgste_t pgste;
1119 	pte_t pte;
1120 
1121 	if (mm_has_pgste(mm))
1122 		pgste = pgste_get_lock(ptep);
1123 
1124 	pte = *ptep;
1125 	if (!full)
1126 		__ptep_ipte(address, ptep);
1127 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
1128 
1129 	if (mm_has_pgste(mm)) {
1130 		pgste = pgste_update_all(&pte, pgste);
1131 		pgste_set_unlock(ptep, pgste);
1132 	}
1133 	return pte;
1134 }
1135 
1136 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1137 static inline pte_t ptep_set_wrprotect(struct mm_struct *mm,
1138 				       unsigned long address, pte_t *ptep)
1139 {
1140 	pgste_t pgste;
1141 	pte_t pte = *ptep;
1142 
1143 	if (pte_write(pte)) {
1144 		mm->context.flush_mm = 1;
1145 		if (mm_has_pgste(mm))
1146 			pgste = pgste_get_lock(ptep);
1147 
1148 		if (!mm_exclusive(mm))
1149 			__ptep_ipte(address, ptep);
1150 		pte = pte_wrprotect(pte);
1151 
1152 		if (mm_has_pgste(mm)) {
1153 			pgste_set_pte(ptep, pte);
1154 			pgste_set_unlock(ptep, pgste);
1155 		} else
1156 			*ptep = pte;
1157 	}
1158 	return pte;
1159 }
1160 
1161 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1162 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
1163 					unsigned long address, pte_t *ptep,
1164 					pte_t entry, int dirty)
1165 {
1166 	pgste_t pgste;
1167 
1168 	if (pte_same(*ptep, entry))
1169 		return 0;
1170 	if (mm_has_pgste(vma->vm_mm))
1171 		pgste = pgste_get_lock(ptep);
1172 
1173 	__ptep_ipte(address, ptep);
1174 
1175 	if (mm_has_pgste(vma->vm_mm)) {
1176 		pgste_set_pte(ptep, entry);
1177 		pgste_set_unlock(ptep, pgste);
1178 	} else
1179 		*ptep = entry;
1180 	return 1;
1181 }
1182 
1183 /*
1184  * Conversion functions: convert a page and protection to a page entry,
1185  * and a page entry and page directory to the page they refer to.
1186  */
1187 static inline pte_t mk_pte_phys(unsigned long physpage, pgprot_t pgprot)
1188 {
1189 	pte_t __pte;
1190 	pte_val(__pte) = physpage + pgprot_val(pgprot);
1191 	return __pte;
1192 }
1193 
1194 static inline pte_t mk_pte(struct page *page, pgprot_t pgprot)
1195 {
1196 	unsigned long physpage = page_to_phys(page);
1197 	pte_t __pte = mk_pte_phys(physpage, pgprot);
1198 
1199 	if ((pte_val(__pte) & _PAGE_SWW) && PageDirty(page)) {
1200 		pte_val(__pte) |= _PAGE_SWC;
1201 		pte_val(__pte) &= ~_PAGE_RO;
1202 	}
1203 	return __pte;
1204 }
1205 
1206 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD-1))
1207 #define pud_index(address) (((address) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
1208 #define pmd_index(address) (((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1))
1209 #define pte_index(address) (((address) >> PAGE_SHIFT) & (PTRS_PER_PTE-1))
1210 
1211 #define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address))
1212 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
1213 
1214 #ifndef CONFIG_64BIT
1215 
1216 #define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
1217 #define pud_deref(pmd) ({ BUG(); 0UL; })
1218 #define pgd_deref(pmd) ({ BUG(); 0UL; })
1219 
1220 #define pud_offset(pgd, address) ((pud_t *) pgd)
1221 #define pmd_offset(pud, address) ((pmd_t *) pud + pmd_index(address))
1222 
1223 #else /* CONFIG_64BIT */
1224 
1225 #define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
1226 #define pud_deref(pud) (pud_val(pud) & _REGION_ENTRY_ORIGIN)
1227 #define pgd_deref(pgd) (pgd_val(pgd) & _REGION_ENTRY_ORIGIN)
1228 
1229 static inline pud_t *pud_offset(pgd_t *pgd, unsigned long address)
1230 {
1231 	pud_t *pud = (pud_t *) pgd;
1232 	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
1233 		pud = (pud_t *) pgd_deref(*pgd);
1234 	return pud  + pud_index(address);
1235 }
1236 
1237 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
1238 {
1239 	pmd_t *pmd = (pmd_t *) pud;
1240 	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
1241 		pmd = (pmd_t *) pud_deref(*pud);
1242 	return pmd + pmd_index(address);
1243 }
1244 
1245 #endif /* CONFIG_64BIT */
1246 
1247 #define pfn_pte(pfn,pgprot) mk_pte_phys(__pa((pfn) << PAGE_SHIFT),(pgprot))
1248 #define pte_pfn(x) (pte_val(x) >> PAGE_SHIFT)
1249 #define pte_page(x) pfn_to_page(pte_pfn(x))
1250 
1251 #define pmd_page(pmd) pfn_to_page(pmd_val(pmd) >> PAGE_SHIFT)
1252 
1253 /* Find an entry in the lowest level page table.. */
1254 #define pte_offset(pmd, addr) ((pte_t *) pmd_deref(*(pmd)) + pte_index(addr))
1255 #define pte_offset_kernel(pmd, address) pte_offset(pmd,address)
1256 #define pte_offset_map(pmd, address) pte_offset_kernel(pmd, address)
1257 #define pte_unmap(pte) do { } while (0)
1258 
1259 static inline void __pmd_idte(unsigned long address, pmd_t *pmdp)
1260 {
1261 	unsigned long sto = (unsigned long) pmdp -
1262 			    pmd_index(address) * sizeof(pmd_t);
1263 
1264 	if (!(pmd_val(*pmdp) & _SEGMENT_ENTRY_INV)) {
1265 		asm volatile(
1266 			"	.insn	rrf,0xb98e0000,%2,%3,0,0"
1267 			: "=m" (*pmdp)
1268 			: "m" (*pmdp), "a" (sto),
1269 			  "a" ((address & HPAGE_MASK))
1270 			: "cc"
1271 		);
1272 	}
1273 }
1274 
1275 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1276 
1277 #define SEGMENT_NONE	__pgprot(_HPAGE_TYPE_NONE)
1278 #define SEGMENT_RO	__pgprot(_HPAGE_TYPE_RO)
1279 #define SEGMENT_RW	__pgprot(_HPAGE_TYPE_RW)
1280 
1281 #define __HAVE_ARCH_PGTABLE_DEPOSIT
1282 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pgtable_t pgtable);
1283 
1284 #define __HAVE_ARCH_PGTABLE_WITHDRAW
1285 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm);
1286 
1287 static inline int pmd_trans_splitting(pmd_t pmd)
1288 {
1289 	return pmd_val(pmd) & _SEGMENT_ENTRY_SPLIT;
1290 }
1291 
1292 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1293 			      pmd_t *pmdp, pmd_t entry)
1294 {
1295 	if (!(pmd_val(entry) & _SEGMENT_ENTRY_INV) && MACHINE_HAS_EDAT1)
1296 		pmd_val(entry) |= _SEGMENT_ENTRY_CO;
1297 	*pmdp = entry;
1298 }
1299 
1300 static inline unsigned long massage_pgprot_pmd(pgprot_t pgprot)
1301 {
1302 	/*
1303 	 * pgprot is PAGE_NONE, PAGE_RO, or PAGE_RW (see __Pxxx / __Sxxx)
1304 	 * Convert to segment table entry format.
1305 	 */
1306 	if (pgprot_val(pgprot) == pgprot_val(PAGE_NONE))
1307 		return pgprot_val(SEGMENT_NONE);
1308 	if (pgprot_val(pgprot) == pgprot_val(PAGE_RO))
1309 		return pgprot_val(SEGMENT_RO);
1310 	return pgprot_val(SEGMENT_RW);
1311 }
1312 
1313 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
1314 {
1315 	pmd_val(pmd) &= _SEGMENT_CHG_MASK;
1316 	pmd_val(pmd) |= massage_pgprot_pmd(newprot);
1317 	return pmd;
1318 }
1319 
1320 static inline pmd_t pmd_mkhuge(pmd_t pmd)
1321 {
1322 	pmd_val(pmd) |= _SEGMENT_ENTRY_LARGE;
1323 	return pmd;
1324 }
1325 
1326 static inline pmd_t pmd_mkwrite(pmd_t pmd)
1327 {
1328 	/* Do not clobber _HPAGE_TYPE_NONE pages! */
1329 	if (!(pmd_val(pmd) & _SEGMENT_ENTRY_INV))
1330 		pmd_val(pmd) &= ~_SEGMENT_ENTRY_RO;
1331 	return pmd;
1332 }
1333 
1334 static inline pmd_t pmd_wrprotect(pmd_t pmd)
1335 {
1336 	pmd_val(pmd) |= _SEGMENT_ENTRY_RO;
1337 	return pmd;
1338 }
1339 
1340 static inline pmd_t pmd_mkdirty(pmd_t pmd)
1341 {
1342 	/* No dirty bit in the segment table entry. */
1343 	return pmd;
1344 }
1345 
1346 static inline pmd_t pmd_mkold(pmd_t pmd)
1347 {
1348 	/* No referenced bit in the segment table entry. */
1349 	return pmd;
1350 }
1351 
1352 static inline pmd_t pmd_mkyoung(pmd_t pmd)
1353 {
1354 	/* No referenced bit in the segment table entry. */
1355 	return pmd;
1356 }
1357 
1358 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1359 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1360 					    unsigned long address, pmd_t *pmdp)
1361 {
1362 	unsigned long pmd_addr = pmd_val(*pmdp) & HPAGE_MASK;
1363 	long tmp, rc;
1364 	int counter;
1365 
1366 	rc = 0;
1367 	if (MACHINE_HAS_RRBM) {
1368 		counter = PTRS_PER_PTE >> 6;
1369 		asm volatile(
1370 			"0:	.insn	rre,0xb9ae0000,%0,%3\n"	/* rrbm */
1371 			"	ogr	%1,%0\n"
1372 			"	la	%3,0(%4,%3)\n"
1373 			"	brct	%2,0b\n"
1374 			: "=&d" (tmp), "+&d" (rc), "+d" (counter),
1375 			  "+a" (pmd_addr)
1376 			: "a" (64 * 4096UL) : "cc");
1377 		rc = !!rc;
1378 	} else {
1379 		counter = PTRS_PER_PTE;
1380 		asm volatile(
1381 			"0:	rrbe	0,%2\n"
1382 			"	la	%2,0(%3,%2)\n"
1383 			"	brc	12,1f\n"
1384 			"	lhi	%0,1\n"
1385 			"1:	brct	%1,0b\n"
1386 			: "+d" (rc), "+d" (counter), "+a" (pmd_addr)
1387 			: "a" (4096UL) : "cc");
1388 	}
1389 	return rc;
1390 }
1391 
1392 #define __HAVE_ARCH_PMDP_GET_AND_CLEAR
1393 static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
1394 				       unsigned long address, pmd_t *pmdp)
1395 {
1396 	pmd_t pmd = *pmdp;
1397 
1398 	__pmd_idte(address, pmdp);
1399 	pmd_clear(pmdp);
1400 	return pmd;
1401 }
1402 
1403 #define __HAVE_ARCH_PMDP_CLEAR_FLUSH
1404 static inline pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
1405 				     unsigned long address, pmd_t *pmdp)
1406 {
1407 	return pmdp_get_and_clear(vma->vm_mm, address, pmdp);
1408 }
1409 
1410 #define __HAVE_ARCH_PMDP_INVALIDATE
1411 static inline void pmdp_invalidate(struct vm_area_struct *vma,
1412 				   unsigned long address, pmd_t *pmdp)
1413 {
1414 	__pmd_idte(address, pmdp);
1415 }
1416 
1417 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1418 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1419 				      unsigned long address, pmd_t *pmdp)
1420 {
1421 	pmd_t pmd = *pmdp;
1422 
1423 	if (pmd_write(pmd)) {
1424 		__pmd_idte(address, pmdp);
1425 		set_pmd_at(mm, address, pmdp, pmd_wrprotect(pmd));
1426 	}
1427 }
1428 
1429 static inline pmd_t mk_pmd_phys(unsigned long physpage, pgprot_t pgprot)
1430 {
1431 	pmd_t __pmd;
1432 	pmd_val(__pmd) = physpage + massage_pgprot_pmd(pgprot);
1433 	return __pmd;
1434 }
1435 
1436 #define pfn_pmd(pfn, pgprot)	mk_pmd_phys(__pa((pfn) << PAGE_SHIFT), (pgprot))
1437 #define mk_pmd(page, pgprot)	pfn_pmd(page_to_pfn(page), (pgprot))
1438 
1439 static inline int pmd_trans_huge(pmd_t pmd)
1440 {
1441 	return pmd_val(pmd) & _SEGMENT_ENTRY_LARGE;
1442 }
1443 
1444 static inline int has_transparent_hugepage(void)
1445 {
1446 	return MACHINE_HAS_HPAGE ? 1 : 0;
1447 }
1448 
1449 static inline unsigned long pmd_pfn(pmd_t pmd)
1450 {
1451 	return pmd_val(pmd) >> PAGE_SHIFT;
1452 }
1453 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1454 
1455 /*
1456  * 31 bit swap entry format:
1457  * A page-table entry has some bits we have to treat in a special way.
1458  * Bits 0, 20 and bit 23 have to be zero, otherwise an specification
1459  * exception will occur instead of a page translation exception. The
1460  * specifiation exception has the bad habit not to store necessary
1461  * information in the lowcore.
1462  * Bit 21 and bit 22 are the page invalid bit and the page protection
1463  * bit. We set both to indicate a swapped page.
1464  * Bit 30 and 31 are used to distinguish the different page types. For
1465  * a swapped page these bits need to be zero.
1466  * This leaves the bits 1-19 and bits 24-29 to store type and offset.
1467  * We use the 5 bits from 25-29 for the type and the 20 bits from 1-19
1468  * plus 24 for the offset.
1469  * 0|     offset        |0110|o|type |00|
1470  * 0 0000000001111111111 2222 2 22222 33
1471  * 0 1234567890123456789 0123 4 56789 01
1472  *
1473  * 64 bit swap entry format:
1474  * A page-table entry has some bits we have to treat in a special way.
1475  * Bits 52 and bit 55 have to be zero, otherwise an specification
1476  * exception will occur instead of a page translation exception. The
1477  * specifiation exception has the bad habit not to store necessary
1478  * information in the lowcore.
1479  * Bit 53 and bit 54 are the page invalid bit and the page protection
1480  * bit. We set both to indicate a swapped page.
1481  * Bit 62 and 63 are used to distinguish the different page types. For
1482  * a swapped page these bits need to be zero.
1483  * This leaves the bits 0-51 and bits 56-61 to store type and offset.
1484  * We use the 5 bits from 57-61 for the type and the 53 bits from 0-51
1485  * plus 56 for the offset.
1486  * |                      offset                        |0110|o|type |00|
1487  *  0000000000111111111122222222223333333333444444444455 5555 5 55566 66
1488  *  0123456789012345678901234567890123456789012345678901 2345 6 78901 23
1489  */
1490 #ifndef CONFIG_64BIT
1491 #define __SWP_OFFSET_MASK (~0UL >> 12)
1492 #else
1493 #define __SWP_OFFSET_MASK (~0UL >> 11)
1494 #endif
1495 static inline pte_t mk_swap_pte(unsigned long type, unsigned long offset)
1496 {
1497 	pte_t pte;
1498 	offset &= __SWP_OFFSET_MASK;
1499 	pte_val(pte) = _PAGE_TYPE_SWAP | ((type & 0x1f) << 2) |
1500 		((offset & 1UL) << 7) | ((offset & ~1UL) << 11);
1501 	return pte;
1502 }
1503 
1504 #define __swp_type(entry)	(((entry).val >> 2) & 0x1f)
1505 #define __swp_offset(entry)	(((entry).val >> 11) | (((entry).val >> 7) & 1))
1506 #define __swp_entry(type,offset) ((swp_entry_t) { pte_val(mk_swap_pte((type),(offset))) })
1507 
1508 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
1509 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
1510 
1511 #ifndef CONFIG_64BIT
1512 # define PTE_FILE_MAX_BITS	26
1513 #else /* CONFIG_64BIT */
1514 # define PTE_FILE_MAX_BITS	59
1515 #endif /* CONFIG_64BIT */
1516 
1517 #define pte_to_pgoff(__pte) \
1518 	((((__pte).pte >> 12) << 7) + (((__pte).pte >> 1) & 0x7f))
1519 
1520 #define pgoff_to_pte(__off) \
1521 	((pte_t) { ((((__off) & 0x7f) << 1) + (((__off) >> 7) << 12)) \
1522 		   | _PAGE_TYPE_FILE })
1523 
1524 #endif /* !__ASSEMBLY__ */
1525 
1526 #define kern_addr_valid(addr)   (1)
1527 
1528 extern int vmem_add_mapping(unsigned long start, unsigned long size);
1529 extern int vmem_remove_mapping(unsigned long start, unsigned long size);
1530 extern int s390_enable_sie(void);
1531 
1532 /*
1533  * No page table caches to initialise
1534  */
1535 static inline void pgtable_cache_init(void) { }
1536 static inline void check_pgt_cache(void) { }
1537 
1538 #include <asm-generic/pgtable.h>
1539 
1540 #endif /* _S390_PAGE_H */
1541