xref: /openbmc/linux/arch/s390/include/asm/pgtable.h (revision 97da55fc)
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	~0x7ffUL/* segment table origin		    */
348 #define _SEGMENT_ENTRY_RO	0x200	/* page protection bit		    */
349 #define _SEGMENT_ENTRY_INV	0x20	/* invalid segment table entry	    */
350 
351 #define _SEGMENT_ENTRY		(0)
352 #define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INV)
353 
354 #define _SEGMENT_ENTRY_LARGE	0x400	/* STE-format control, large page   */
355 #define _SEGMENT_ENTRY_CO	0x100	/* change-recording override   */
356 #define _SEGMENT_ENTRY_SPLIT_BIT 0	/* THP splitting bit number */
357 #define _SEGMENT_ENTRY_SPLIT	(1UL << _SEGMENT_ENTRY_SPLIT_BIT)
358 
359 /* Set of bits not changed in pmd_modify */
360 #define _SEGMENT_CHG_MASK	(_SEGMENT_ENTRY_ORIGIN | _SEGMENT_ENTRY_LARGE \
361 				 | _SEGMENT_ENTRY_SPLIT | _SEGMENT_ENTRY_CO)
362 
363 /* Page status table bits for virtualization */
364 #define RCP_ACC_BITS	0xf000000000000000UL
365 #define RCP_FP_BIT	0x0800000000000000UL
366 #define RCP_PCL_BIT	0x0080000000000000UL
367 #define RCP_HR_BIT	0x0040000000000000UL
368 #define RCP_HC_BIT	0x0020000000000000UL
369 #define RCP_GR_BIT	0x0004000000000000UL
370 #define RCP_GC_BIT	0x0002000000000000UL
371 
372 /* User dirty / referenced bit for KVM's migration feature */
373 #define KVM_UR_BIT	0x0000800000000000UL
374 #define KVM_UC_BIT	0x0000400000000000UL
375 
376 #endif /* CONFIG_64BIT */
377 
378 /*
379  * A user page table pointer has the space-switch-event bit, the
380  * private-space-control bit and the storage-alteration-event-control
381  * bit set. A kernel page table pointer doesn't need them.
382  */
383 #define _ASCE_USER_BITS		(_ASCE_SPACE_SWITCH | _ASCE_PRIVATE_SPACE | \
384 				 _ASCE_ALT_EVENT)
385 
386 /*
387  * Page protection definitions.
388  */
389 #define PAGE_NONE	__pgprot(_PAGE_TYPE_NONE)
390 #define PAGE_RO		__pgprot(_PAGE_TYPE_RO)
391 #define PAGE_RW		__pgprot(_PAGE_TYPE_RO | _PAGE_SWW)
392 #define PAGE_RWC	__pgprot(_PAGE_TYPE_RW | _PAGE_SWW | _PAGE_SWC)
393 
394 #define PAGE_KERNEL	PAGE_RWC
395 #define PAGE_SHARED	PAGE_KERNEL
396 #define PAGE_COPY	PAGE_RO
397 
398 /*
399  * On s390 the page table entry has an invalid bit and a read-only bit.
400  * Read permission implies execute permission and write permission
401  * implies read permission.
402  */
403          /*xwr*/
404 #define __P000	PAGE_NONE
405 #define __P001	PAGE_RO
406 #define __P010	PAGE_RO
407 #define __P011	PAGE_RO
408 #define __P100	PAGE_RO
409 #define __P101	PAGE_RO
410 #define __P110	PAGE_RO
411 #define __P111	PAGE_RO
412 
413 #define __S000	PAGE_NONE
414 #define __S001	PAGE_RO
415 #define __S010	PAGE_RW
416 #define __S011	PAGE_RW
417 #define __S100	PAGE_RO
418 #define __S101	PAGE_RO
419 #define __S110	PAGE_RW
420 #define __S111	PAGE_RW
421 
422 static inline int mm_exclusive(struct mm_struct *mm)
423 {
424 	return likely(mm == current->active_mm &&
425 		      atomic_read(&mm->context.attach_count) <= 1);
426 }
427 
428 static inline int mm_has_pgste(struct mm_struct *mm)
429 {
430 #ifdef CONFIG_PGSTE
431 	if (unlikely(mm->context.has_pgste))
432 		return 1;
433 #endif
434 	return 0;
435 }
436 /*
437  * pgd/pmd/pte query functions
438  */
439 #ifndef CONFIG_64BIT
440 
441 static inline int pgd_present(pgd_t pgd) { return 1; }
442 static inline int pgd_none(pgd_t pgd)    { return 0; }
443 static inline int pgd_bad(pgd_t pgd)     { return 0; }
444 
445 static inline int pud_present(pud_t pud) { return 1; }
446 static inline int pud_none(pud_t pud)	 { return 0; }
447 static inline int pud_large(pud_t pud)	 { return 0; }
448 static inline int pud_bad(pud_t pud)	 { return 0; }
449 
450 #else /* CONFIG_64BIT */
451 
452 static inline int pgd_present(pgd_t pgd)
453 {
454 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
455 		return 1;
456 	return (pgd_val(pgd) & _REGION_ENTRY_ORIGIN) != 0UL;
457 }
458 
459 static inline int pgd_none(pgd_t pgd)
460 {
461 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
462 		return 0;
463 	return (pgd_val(pgd) & _REGION_ENTRY_INV) != 0UL;
464 }
465 
466 static inline int pgd_bad(pgd_t pgd)
467 {
468 	/*
469 	 * With dynamic page table levels the pgd can be a region table
470 	 * entry or a segment table entry. Check for the bit that are
471 	 * invalid for either table entry.
472 	 */
473 	unsigned long mask =
474 		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INV &
475 		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
476 	return (pgd_val(pgd) & mask) != 0;
477 }
478 
479 static inline int pud_present(pud_t pud)
480 {
481 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
482 		return 1;
483 	return (pud_val(pud) & _REGION_ENTRY_ORIGIN) != 0UL;
484 }
485 
486 static inline int pud_none(pud_t pud)
487 {
488 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
489 		return 0;
490 	return (pud_val(pud) & _REGION_ENTRY_INV) != 0UL;
491 }
492 
493 static inline int pud_large(pud_t pud)
494 {
495 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) != _REGION_ENTRY_TYPE_R3)
496 		return 0;
497 	return !!(pud_val(pud) & _REGION3_ENTRY_LARGE);
498 }
499 
500 static inline int pud_bad(pud_t pud)
501 {
502 	/*
503 	 * With dynamic page table levels the pud can be a region table
504 	 * entry or a segment table entry. Check for the bit that are
505 	 * invalid for either table entry.
506 	 */
507 	unsigned long mask =
508 		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INV &
509 		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
510 	return (pud_val(pud) & mask) != 0;
511 }
512 
513 #endif /* CONFIG_64BIT */
514 
515 static inline int pmd_present(pmd_t pmd)
516 {
517 	unsigned long mask = _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO;
518 	return (pmd_val(pmd) & mask) == _HPAGE_TYPE_NONE ||
519 	       !(pmd_val(pmd) & _SEGMENT_ENTRY_INV);
520 }
521 
522 static inline int pmd_none(pmd_t pmd)
523 {
524 	return (pmd_val(pmd) & _SEGMENT_ENTRY_INV) &&
525 	       !(pmd_val(pmd) & _SEGMENT_ENTRY_RO);
526 }
527 
528 static inline int pmd_large(pmd_t pmd)
529 {
530 #ifdef CONFIG_64BIT
531 	return !!(pmd_val(pmd) & _SEGMENT_ENTRY_LARGE);
532 #else
533 	return 0;
534 #endif
535 }
536 
537 static inline int pmd_bad(pmd_t pmd)
538 {
539 	unsigned long mask = ~_SEGMENT_ENTRY_ORIGIN & ~_SEGMENT_ENTRY_INV;
540 	return (pmd_val(pmd) & mask) != _SEGMENT_ENTRY;
541 }
542 
543 #define __HAVE_ARCH_PMDP_SPLITTING_FLUSH
544 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
545 				 unsigned long addr, pmd_t *pmdp);
546 
547 #define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
548 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
549 				 unsigned long address, pmd_t *pmdp,
550 				 pmd_t entry, int dirty);
551 
552 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
553 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
554 				  unsigned long address, pmd_t *pmdp);
555 
556 #define __HAVE_ARCH_PMD_WRITE
557 static inline int pmd_write(pmd_t pmd)
558 {
559 	return (pmd_val(pmd) & _SEGMENT_ENTRY_RO) == 0;
560 }
561 
562 static inline int pmd_young(pmd_t pmd)
563 {
564 	return 0;
565 }
566 
567 static inline int pte_none(pte_t pte)
568 {
569 	return (pte_val(pte) & _PAGE_INVALID) && !(pte_val(pte) & _PAGE_SWT);
570 }
571 
572 static inline int pte_present(pte_t pte)
573 {
574 	unsigned long mask = _PAGE_RO | _PAGE_INVALID | _PAGE_SWT | _PAGE_SWX;
575 	return (pte_val(pte) & mask) == _PAGE_TYPE_NONE ||
576 		(!(pte_val(pte) & _PAGE_INVALID) &&
577 		 !(pte_val(pte) & _PAGE_SWT));
578 }
579 
580 static inline int pte_file(pte_t pte)
581 {
582 	unsigned long mask = _PAGE_RO | _PAGE_INVALID | _PAGE_SWT;
583 	return (pte_val(pte) & mask) == _PAGE_TYPE_FILE;
584 }
585 
586 static inline int pte_special(pte_t pte)
587 {
588 	return (pte_val(pte) & _PAGE_SPECIAL);
589 }
590 
591 #define __HAVE_ARCH_PTE_SAME
592 static inline int pte_same(pte_t a, pte_t b)
593 {
594 	return pte_val(a) == pte_val(b);
595 }
596 
597 static inline pgste_t pgste_get_lock(pte_t *ptep)
598 {
599 	unsigned long new = 0;
600 #ifdef CONFIG_PGSTE
601 	unsigned long old;
602 
603 	preempt_disable();
604 	asm(
605 		"	lg	%0,%2\n"
606 		"0:	lgr	%1,%0\n"
607 		"	nihh	%0,0xff7f\n"	/* clear RCP_PCL_BIT in old */
608 		"	oihh	%1,0x0080\n"	/* set RCP_PCL_BIT in new */
609 		"	csg	%0,%1,%2\n"
610 		"	jl	0b\n"
611 		: "=&d" (old), "=&d" (new), "=Q" (ptep[PTRS_PER_PTE])
612 		: "Q" (ptep[PTRS_PER_PTE]) : "cc");
613 #endif
614 	return __pgste(new);
615 }
616 
617 static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste)
618 {
619 #ifdef CONFIG_PGSTE
620 	asm(
621 		"	nihh	%1,0xff7f\n"	/* clear RCP_PCL_BIT */
622 		"	stg	%1,%0\n"
623 		: "=Q" (ptep[PTRS_PER_PTE])
624 		: "d" (pgste_val(pgste)), "Q" (ptep[PTRS_PER_PTE]) : "cc");
625 	preempt_enable();
626 #endif
627 }
628 
629 static inline pgste_t pgste_update_all(pte_t *ptep, pgste_t pgste)
630 {
631 #ifdef CONFIG_PGSTE
632 	unsigned long address, bits;
633 	unsigned char skey;
634 
635 	if (!pte_present(*ptep))
636 		return pgste;
637 	address = pte_val(*ptep) & PAGE_MASK;
638 	skey = page_get_storage_key(address);
639 	bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
640 	/* Clear page changed & referenced bit in the storage key */
641 	if (bits & _PAGE_CHANGED)
642 		page_set_storage_key(address, skey ^ bits, 0);
643 	else if (bits)
644 		page_reset_referenced(address);
645 	/* Transfer page changed & referenced bit to guest bits in pgste */
646 	pgste_val(pgste) |= bits << 48;		/* RCP_GR_BIT & RCP_GC_BIT */
647 	/* Get host changed & referenced bits from pgste */
648 	bits |= (pgste_val(pgste) & (RCP_HR_BIT | RCP_HC_BIT)) >> 52;
649 	/* Transfer page changed & referenced bit to kvm user bits */
650 	pgste_val(pgste) |= bits << 45;		/* KVM_UR_BIT & KVM_UC_BIT */
651 	/* Clear relevant host bits in pgste. */
652 	pgste_val(pgste) &= ~(RCP_HR_BIT | RCP_HC_BIT);
653 	pgste_val(pgste) &= ~(RCP_ACC_BITS | RCP_FP_BIT);
654 	/* Copy page access key and fetch protection bit to pgste */
655 	pgste_val(pgste) |=
656 		(unsigned long) (skey & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
657 	/* Transfer referenced bit to pte */
658 	pte_val(*ptep) |= (bits & _PAGE_REFERENCED) << 1;
659 #endif
660 	return pgste;
661 
662 }
663 
664 static inline pgste_t pgste_update_young(pte_t *ptep, pgste_t pgste)
665 {
666 #ifdef CONFIG_PGSTE
667 	int young;
668 
669 	if (!pte_present(*ptep))
670 		return pgste;
671 	/* Get referenced bit from storage key */
672 	young = page_reset_referenced(pte_val(*ptep) & PAGE_MASK);
673 	if (young)
674 		pgste_val(pgste) |= RCP_GR_BIT;
675 	/* Get host referenced bit from pgste */
676 	if (pgste_val(pgste) & RCP_HR_BIT) {
677 		pgste_val(pgste) &= ~RCP_HR_BIT;
678 		young = 1;
679 	}
680 	/* Transfer referenced bit to kvm user bits and pte */
681 	if (young) {
682 		pgste_val(pgste) |= KVM_UR_BIT;
683 		pte_val(*ptep) |= _PAGE_SWR;
684 	}
685 #endif
686 	return pgste;
687 }
688 
689 static inline void pgste_set_key(pte_t *ptep, pgste_t pgste, pte_t entry)
690 {
691 #ifdef CONFIG_PGSTE
692 	unsigned long address;
693 	unsigned long okey, nkey;
694 
695 	if (!pte_present(entry))
696 		return;
697 	address = pte_val(entry) & PAGE_MASK;
698 	okey = nkey = page_get_storage_key(address);
699 	nkey &= ~(_PAGE_ACC_BITS | _PAGE_FP_BIT);
700 	/* Set page access key and fetch protection bit from pgste */
701 	nkey |= (pgste_val(pgste) & (RCP_ACC_BITS | RCP_FP_BIT)) >> 56;
702 	if (okey != nkey)
703 		page_set_storage_key(address, nkey, 0);
704 #endif
705 }
706 
707 static inline void pgste_set_pte(pte_t *ptep, pte_t entry)
708 {
709 	if (!MACHINE_HAS_ESOP && (pte_val(entry) & _PAGE_SWW)) {
710 		/*
711 		 * Without enhanced suppression-on-protection force
712 		 * the dirty bit on for all writable ptes.
713 		 */
714 		pte_val(entry) |= _PAGE_SWC;
715 		pte_val(entry) &= ~_PAGE_RO;
716 	}
717 	*ptep = entry;
718 }
719 
720 /**
721  * struct gmap_struct - guest address space
722  * @mm: pointer to the parent mm_struct
723  * @table: pointer to the page directory
724  * @asce: address space control element for gmap page table
725  * @crst_list: list of all crst tables used in the guest address space
726  */
727 struct gmap {
728 	struct list_head list;
729 	struct mm_struct *mm;
730 	unsigned long *table;
731 	unsigned long asce;
732 	struct list_head crst_list;
733 };
734 
735 /**
736  * struct gmap_rmap - reverse mapping for segment table entries
737  * @next: pointer to the next gmap_rmap structure in the list
738  * @entry: pointer to a segment table entry
739  */
740 struct gmap_rmap {
741 	struct list_head list;
742 	unsigned long *entry;
743 };
744 
745 /**
746  * struct gmap_pgtable - gmap information attached to a page table
747  * @vmaddr: address of the 1MB segment in the process virtual memory
748  * @mapper: list of segment table entries maping a page table
749  */
750 struct gmap_pgtable {
751 	unsigned long vmaddr;
752 	struct list_head mapper;
753 };
754 
755 struct gmap *gmap_alloc(struct mm_struct *mm);
756 void gmap_free(struct gmap *gmap);
757 void gmap_enable(struct gmap *gmap);
758 void gmap_disable(struct gmap *gmap);
759 int gmap_map_segment(struct gmap *gmap, unsigned long from,
760 		     unsigned long to, unsigned long length);
761 int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len);
762 unsigned long __gmap_fault(unsigned long address, struct gmap *);
763 unsigned long gmap_fault(unsigned long address, struct gmap *);
764 void gmap_discard(unsigned long from, unsigned long to, struct gmap *);
765 
766 /*
767  * Certain architectures need to do special things when PTEs
768  * within a page table are directly modified.  Thus, the following
769  * hook is made available.
770  */
771 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
772 			      pte_t *ptep, pte_t entry)
773 {
774 	pgste_t pgste;
775 
776 	if (mm_has_pgste(mm)) {
777 		pgste = pgste_get_lock(ptep);
778 		pgste_set_key(ptep, pgste, entry);
779 		pgste_set_pte(ptep, entry);
780 		pgste_set_unlock(ptep, pgste);
781 	} else {
782 		if (!(pte_val(entry) & _PAGE_INVALID) && MACHINE_HAS_EDAT1)
783 			pte_val(entry) |= _PAGE_CO;
784 		*ptep = entry;
785 	}
786 }
787 
788 /*
789  * query functions pte_write/pte_dirty/pte_young only work if
790  * pte_present() is true. Undefined behaviour if not..
791  */
792 static inline int pte_write(pte_t pte)
793 {
794 	return (pte_val(pte) & _PAGE_SWW) != 0;
795 }
796 
797 static inline int pte_dirty(pte_t pte)
798 {
799 	return (pte_val(pte) & _PAGE_SWC) != 0;
800 }
801 
802 static inline int pte_young(pte_t pte)
803 {
804 #ifdef CONFIG_PGSTE
805 	if (pte_val(pte) & _PAGE_SWR)
806 		return 1;
807 #endif
808 	return 0;
809 }
810 
811 /*
812  * pgd/pmd/pte modification functions
813  */
814 
815 static inline void pgd_clear(pgd_t *pgd)
816 {
817 #ifdef CONFIG_64BIT
818 	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
819 		pgd_val(*pgd) = _REGION2_ENTRY_EMPTY;
820 #endif
821 }
822 
823 static inline void pud_clear(pud_t *pud)
824 {
825 #ifdef CONFIG_64BIT
826 	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
827 		pud_val(*pud) = _REGION3_ENTRY_EMPTY;
828 #endif
829 }
830 
831 static inline void pmd_clear(pmd_t *pmdp)
832 {
833 	pmd_val(*pmdp) = _SEGMENT_ENTRY_EMPTY;
834 }
835 
836 static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
837 {
838 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
839 }
840 
841 /*
842  * The following pte modification functions only work if
843  * pte_present() is true. Undefined behaviour if not..
844  */
845 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
846 {
847 	pte_val(pte) &= _PAGE_CHG_MASK;
848 	pte_val(pte) |= pgprot_val(newprot);
849 	if ((pte_val(pte) & _PAGE_SWC) && (pte_val(pte) & _PAGE_SWW))
850 		pte_val(pte) &= ~_PAGE_RO;
851 	return pte;
852 }
853 
854 static inline pte_t pte_wrprotect(pte_t pte)
855 {
856 	pte_val(pte) &= ~_PAGE_SWW;
857 	/* Do not clobber _PAGE_TYPE_NONE pages!  */
858 	if (!(pte_val(pte) & _PAGE_INVALID))
859 		pte_val(pte) |= _PAGE_RO;
860 	return pte;
861 }
862 
863 static inline pte_t pte_mkwrite(pte_t pte)
864 {
865 	pte_val(pte) |= _PAGE_SWW;
866 	if (pte_val(pte) & _PAGE_SWC)
867 		pte_val(pte) &= ~_PAGE_RO;
868 	return pte;
869 }
870 
871 static inline pte_t pte_mkclean(pte_t pte)
872 {
873 	pte_val(pte) &= ~_PAGE_SWC;
874 	/* Do not clobber _PAGE_TYPE_NONE pages!  */
875 	if (!(pte_val(pte) & _PAGE_INVALID))
876 		pte_val(pte) |= _PAGE_RO;
877 	return pte;
878 }
879 
880 static inline pte_t pte_mkdirty(pte_t pte)
881 {
882 	pte_val(pte) |= _PAGE_SWC;
883 	if (pte_val(pte) & _PAGE_SWW)
884 		pte_val(pte) &= ~_PAGE_RO;
885 	return pte;
886 }
887 
888 static inline pte_t pte_mkold(pte_t pte)
889 {
890 #ifdef CONFIG_PGSTE
891 	pte_val(pte) &= ~_PAGE_SWR;
892 #endif
893 	return pte;
894 }
895 
896 static inline pte_t pte_mkyoung(pte_t pte)
897 {
898 	return pte;
899 }
900 
901 static inline pte_t pte_mkspecial(pte_t pte)
902 {
903 	pte_val(pte) |= _PAGE_SPECIAL;
904 	return pte;
905 }
906 
907 #ifdef CONFIG_HUGETLB_PAGE
908 static inline pte_t pte_mkhuge(pte_t pte)
909 {
910 	/*
911 	 * PROT_NONE needs to be remapped from the pte type to the ste type.
912 	 * The HW invalid bit is also different for pte and ste. The pte
913 	 * invalid bit happens to be the same as the ste _SEGMENT_ENTRY_LARGE
914 	 * bit, so we don't have to clear it.
915 	 */
916 	if (pte_val(pte) & _PAGE_INVALID) {
917 		if (pte_val(pte) & _PAGE_SWT)
918 			pte_val(pte) |= _HPAGE_TYPE_NONE;
919 		pte_val(pte) |= _SEGMENT_ENTRY_INV;
920 	}
921 	/*
922 	 * Clear SW pte bits, there are no SW bits in a segment table entry.
923 	 */
924 	pte_val(pte) &= ~(_PAGE_SWT | _PAGE_SWX | _PAGE_SWC |
925 			  _PAGE_SWR | _PAGE_SWW);
926 	/*
927 	 * Also set the change-override bit because we don't need dirty bit
928 	 * tracking for hugetlbfs pages.
929 	 */
930 	pte_val(pte) |= (_SEGMENT_ENTRY_LARGE | _SEGMENT_ENTRY_CO);
931 	return pte;
932 }
933 #endif
934 
935 /*
936  * Get (and clear) the user dirty bit for a pte.
937  */
938 static inline int ptep_test_and_clear_user_dirty(struct mm_struct *mm,
939 						 pte_t *ptep)
940 {
941 	pgste_t pgste;
942 	int dirty = 0;
943 
944 	if (mm_has_pgste(mm)) {
945 		pgste = pgste_get_lock(ptep);
946 		pgste = pgste_update_all(ptep, pgste);
947 		dirty = !!(pgste_val(pgste) & KVM_UC_BIT);
948 		pgste_val(pgste) &= ~KVM_UC_BIT;
949 		pgste_set_unlock(ptep, pgste);
950 		return dirty;
951 	}
952 	return dirty;
953 }
954 
955 /*
956  * Get (and clear) the user referenced bit for a pte.
957  */
958 static inline int ptep_test_and_clear_user_young(struct mm_struct *mm,
959 						 pte_t *ptep)
960 {
961 	pgste_t pgste;
962 	int young = 0;
963 
964 	if (mm_has_pgste(mm)) {
965 		pgste = pgste_get_lock(ptep);
966 		pgste = pgste_update_young(ptep, pgste);
967 		young = !!(pgste_val(pgste) & KVM_UR_BIT);
968 		pgste_val(pgste) &= ~KVM_UR_BIT;
969 		pgste_set_unlock(ptep, pgste);
970 	}
971 	return young;
972 }
973 
974 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
975 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
976 					    unsigned long addr, pte_t *ptep)
977 {
978 	pgste_t pgste;
979 	pte_t pte;
980 
981 	if (mm_has_pgste(vma->vm_mm)) {
982 		pgste = pgste_get_lock(ptep);
983 		pgste = pgste_update_young(ptep, pgste);
984 		pte = *ptep;
985 		*ptep = pte_mkold(pte);
986 		pgste_set_unlock(ptep, pgste);
987 		return pte_young(pte);
988 	}
989 	return 0;
990 }
991 
992 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
993 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
994 					 unsigned long address, pte_t *ptep)
995 {
996 	/* No need to flush TLB
997 	 * On s390 reference bits are in storage key and never in TLB
998 	 * With virtualization we handle the reference bit, without we
999 	 * we can simply return */
1000 	return ptep_test_and_clear_young(vma, address, ptep);
1001 }
1002 
1003 static inline void __ptep_ipte(unsigned long address, pte_t *ptep)
1004 {
1005 	if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1006 #ifndef CONFIG_64BIT
1007 		/* pto must point to the start of the segment table */
1008 		pte_t *pto = (pte_t *) (((unsigned long) ptep) & 0x7ffffc00);
1009 #else
1010 		/* ipte in zarch mode can do the math */
1011 		pte_t *pto = ptep;
1012 #endif
1013 		asm volatile(
1014 			"	ipte	%2,%3"
1015 			: "=m" (*ptep) : "m" (*ptep),
1016 			  "a" (pto), "a" (address));
1017 	}
1018 }
1019 
1020 /*
1021  * This is hard to understand. ptep_get_and_clear and ptep_clear_flush
1022  * both clear the TLB for the unmapped pte. The reason is that
1023  * ptep_get_and_clear is used in common code (e.g. change_pte_range)
1024  * to modify an active pte. The sequence is
1025  *   1) ptep_get_and_clear
1026  *   2) set_pte_at
1027  *   3) flush_tlb_range
1028  * On s390 the tlb needs to get flushed with the modification of the pte
1029  * if the pte is active. The only way how this can be implemented is to
1030  * have ptep_get_and_clear do the tlb flush. In exchange flush_tlb_range
1031  * is a nop.
1032  */
1033 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1034 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
1035 				       unsigned long address, pte_t *ptep)
1036 {
1037 	pgste_t pgste;
1038 	pte_t pte;
1039 
1040 	mm->context.flush_mm = 1;
1041 	if (mm_has_pgste(mm))
1042 		pgste = pgste_get_lock(ptep);
1043 
1044 	pte = *ptep;
1045 	if (!mm_exclusive(mm))
1046 		__ptep_ipte(address, ptep);
1047 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
1048 
1049 	if (mm_has_pgste(mm)) {
1050 		pgste = pgste_update_all(&pte, pgste);
1051 		pgste_set_unlock(ptep, pgste);
1052 	}
1053 	return pte;
1054 }
1055 
1056 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1057 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
1058 					   unsigned long address,
1059 					   pte_t *ptep)
1060 {
1061 	pte_t pte;
1062 
1063 	mm->context.flush_mm = 1;
1064 	if (mm_has_pgste(mm))
1065 		pgste_get_lock(ptep);
1066 
1067 	pte = *ptep;
1068 	if (!mm_exclusive(mm))
1069 		__ptep_ipte(address, ptep);
1070 	return pte;
1071 }
1072 
1073 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
1074 					   unsigned long address,
1075 					   pte_t *ptep, pte_t pte)
1076 {
1077 	if (mm_has_pgste(mm)) {
1078 		pgste_set_pte(ptep, pte);
1079 		pgste_set_unlock(ptep, *(pgste_t *)(ptep + PTRS_PER_PTE));
1080 	} else
1081 		*ptep = pte;
1082 }
1083 
1084 #define __HAVE_ARCH_PTEP_CLEAR_FLUSH
1085 static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
1086 				     unsigned long address, pte_t *ptep)
1087 {
1088 	pgste_t pgste;
1089 	pte_t pte;
1090 
1091 	if (mm_has_pgste(vma->vm_mm))
1092 		pgste = pgste_get_lock(ptep);
1093 
1094 	pte = *ptep;
1095 	__ptep_ipte(address, ptep);
1096 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
1097 
1098 	if (mm_has_pgste(vma->vm_mm)) {
1099 		pgste = pgste_update_all(&pte, pgste);
1100 		pgste_set_unlock(ptep, pgste);
1101 	}
1102 	return pte;
1103 }
1104 
1105 /*
1106  * The batched pte unmap code uses ptep_get_and_clear_full to clear the
1107  * ptes. Here an optimization is possible. tlb_gather_mmu flushes all
1108  * tlbs of an mm if it can guarantee that the ptes of the mm_struct
1109  * cannot be accessed while the batched unmap is running. In this case
1110  * full==1 and a simple pte_clear is enough. See tlb.h.
1111  */
1112 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
1113 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
1114 					    unsigned long address,
1115 					    pte_t *ptep, int full)
1116 {
1117 	pgste_t pgste;
1118 	pte_t pte;
1119 
1120 	if (mm_has_pgste(mm))
1121 		pgste = pgste_get_lock(ptep);
1122 
1123 	pte = *ptep;
1124 	if (!full)
1125 		__ptep_ipte(address, ptep);
1126 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
1127 
1128 	if (mm_has_pgste(mm)) {
1129 		pgste = pgste_update_all(&pte, pgste);
1130 		pgste_set_unlock(ptep, pgste);
1131 	}
1132 	return pte;
1133 }
1134 
1135 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1136 static inline pte_t ptep_set_wrprotect(struct mm_struct *mm,
1137 				       unsigned long address, pte_t *ptep)
1138 {
1139 	pgste_t pgste;
1140 	pte_t pte = *ptep;
1141 
1142 	if (pte_write(pte)) {
1143 		mm->context.flush_mm = 1;
1144 		if (mm_has_pgste(mm))
1145 			pgste = pgste_get_lock(ptep);
1146 
1147 		if (!mm_exclusive(mm))
1148 			__ptep_ipte(address, ptep);
1149 		pte = pte_wrprotect(pte);
1150 
1151 		if (mm_has_pgste(mm)) {
1152 			pgste_set_pte(ptep, pte);
1153 			pgste_set_unlock(ptep, pgste);
1154 		} else
1155 			*ptep = pte;
1156 	}
1157 	return pte;
1158 }
1159 
1160 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1161 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
1162 					unsigned long address, pte_t *ptep,
1163 					pte_t entry, int dirty)
1164 {
1165 	pgste_t pgste;
1166 
1167 	if (pte_same(*ptep, entry))
1168 		return 0;
1169 	if (mm_has_pgste(vma->vm_mm))
1170 		pgste = pgste_get_lock(ptep);
1171 
1172 	__ptep_ipte(address, ptep);
1173 
1174 	if (mm_has_pgste(vma->vm_mm)) {
1175 		pgste_set_pte(ptep, entry);
1176 		pgste_set_unlock(ptep, pgste);
1177 	} else
1178 		*ptep = entry;
1179 	return 1;
1180 }
1181 
1182 /*
1183  * Conversion functions: convert a page and protection to a page entry,
1184  * and a page entry and page directory to the page they refer to.
1185  */
1186 static inline pte_t mk_pte_phys(unsigned long physpage, pgprot_t pgprot)
1187 {
1188 	pte_t __pte;
1189 	pte_val(__pte) = physpage + pgprot_val(pgprot);
1190 	return __pte;
1191 }
1192 
1193 static inline pte_t mk_pte(struct page *page, pgprot_t pgprot)
1194 {
1195 	unsigned long physpage = page_to_phys(page);
1196 	pte_t __pte = mk_pte_phys(physpage, pgprot);
1197 
1198 	if ((pte_val(__pte) & _PAGE_SWW) && PageDirty(page)) {
1199 		pte_val(__pte) |= _PAGE_SWC;
1200 		pte_val(__pte) &= ~_PAGE_RO;
1201 	}
1202 	return __pte;
1203 }
1204 
1205 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD-1))
1206 #define pud_index(address) (((address) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
1207 #define pmd_index(address) (((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1))
1208 #define pte_index(address) (((address) >> PAGE_SHIFT) & (PTRS_PER_PTE-1))
1209 
1210 #define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address))
1211 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
1212 
1213 #ifndef CONFIG_64BIT
1214 
1215 #define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
1216 #define pud_deref(pmd) ({ BUG(); 0UL; })
1217 #define pgd_deref(pmd) ({ BUG(); 0UL; })
1218 
1219 #define pud_offset(pgd, address) ((pud_t *) pgd)
1220 #define pmd_offset(pud, address) ((pmd_t *) pud + pmd_index(address))
1221 
1222 #else /* CONFIG_64BIT */
1223 
1224 #define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
1225 #define pud_deref(pud) (pud_val(pud) & _REGION_ENTRY_ORIGIN)
1226 #define pgd_deref(pgd) (pgd_val(pgd) & _REGION_ENTRY_ORIGIN)
1227 
1228 static inline pud_t *pud_offset(pgd_t *pgd, unsigned long address)
1229 {
1230 	pud_t *pud = (pud_t *) pgd;
1231 	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
1232 		pud = (pud_t *) pgd_deref(*pgd);
1233 	return pud  + pud_index(address);
1234 }
1235 
1236 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
1237 {
1238 	pmd_t *pmd = (pmd_t *) pud;
1239 	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
1240 		pmd = (pmd_t *) pud_deref(*pud);
1241 	return pmd + pmd_index(address);
1242 }
1243 
1244 #endif /* CONFIG_64BIT */
1245 
1246 #define pfn_pte(pfn,pgprot) mk_pte_phys(__pa((pfn) << PAGE_SHIFT),(pgprot))
1247 #define pte_pfn(x) (pte_val(x) >> PAGE_SHIFT)
1248 #define pte_page(x) pfn_to_page(pte_pfn(x))
1249 
1250 #define pmd_page(pmd) pfn_to_page(pmd_val(pmd) >> PAGE_SHIFT)
1251 
1252 /* Find an entry in the lowest level page table.. */
1253 #define pte_offset(pmd, addr) ((pte_t *) pmd_deref(*(pmd)) + pte_index(addr))
1254 #define pte_offset_kernel(pmd, address) pte_offset(pmd,address)
1255 #define pte_offset_map(pmd, address) pte_offset_kernel(pmd, address)
1256 #define pte_unmap(pte) do { } while (0)
1257 
1258 static inline void __pmd_idte(unsigned long address, pmd_t *pmdp)
1259 {
1260 	unsigned long sto = (unsigned long) pmdp -
1261 			    pmd_index(address) * sizeof(pmd_t);
1262 
1263 	if (!(pmd_val(*pmdp) & _SEGMENT_ENTRY_INV)) {
1264 		asm volatile(
1265 			"	.insn	rrf,0xb98e0000,%2,%3,0,0"
1266 			: "=m" (*pmdp)
1267 			: "m" (*pmdp), "a" (sto),
1268 			  "a" ((address & HPAGE_MASK))
1269 			: "cc"
1270 		);
1271 	}
1272 }
1273 
1274 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1275 
1276 #define SEGMENT_NONE	__pgprot(_HPAGE_TYPE_NONE)
1277 #define SEGMENT_RO	__pgprot(_HPAGE_TYPE_RO)
1278 #define SEGMENT_RW	__pgprot(_HPAGE_TYPE_RW)
1279 
1280 #define __HAVE_ARCH_PGTABLE_DEPOSIT
1281 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pgtable_t pgtable);
1282 
1283 #define __HAVE_ARCH_PGTABLE_WITHDRAW
1284 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm);
1285 
1286 static inline int pmd_trans_splitting(pmd_t pmd)
1287 {
1288 	return pmd_val(pmd) & _SEGMENT_ENTRY_SPLIT;
1289 }
1290 
1291 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1292 			      pmd_t *pmdp, pmd_t entry)
1293 {
1294 	if (!(pmd_val(entry) & _SEGMENT_ENTRY_INV) && MACHINE_HAS_EDAT1)
1295 		pmd_val(entry) |= _SEGMENT_ENTRY_CO;
1296 	*pmdp = entry;
1297 }
1298 
1299 static inline unsigned long massage_pgprot_pmd(pgprot_t pgprot)
1300 {
1301 	/*
1302 	 * pgprot is PAGE_NONE, PAGE_RO, or PAGE_RW (see __Pxxx / __Sxxx)
1303 	 * Convert to segment table entry format.
1304 	 */
1305 	if (pgprot_val(pgprot) == pgprot_val(PAGE_NONE))
1306 		return pgprot_val(SEGMENT_NONE);
1307 	if (pgprot_val(pgprot) == pgprot_val(PAGE_RO))
1308 		return pgprot_val(SEGMENT_RO);
1309 	return pgprot_val(SEGMENT_RW);
1310 }
1311 
1312 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
1313 {
1314 	pmd_val(pmd) &= _SEGMENT_CHG_MASK;
1315 	pmd_val(pmd) |= massage_pgprot_pmd(newprot);
1316 	return pmd;
1317 }
1318 
1319 static inline pmd_t pmd_mkhuge(pmd_t pmd)
1320 {
1321 	pmd_val(pmd) |= _SEGMENT_ENTRY_LARGE;
1322 	return pmd;
1323 }
1324 
1325 static inline pmd_t pmd_mkwrite(pmd_t pmd)
1326 {
1327 	/* Do not clobber _HPAGE_TYPE_NONE pages! */
1328 	if (!(pmd_val(pmd) & _SEGMENT_ENTRY_INV))
1329 		pmd_val(pmd) &= ~_SEGMENT_ENTRY_RO;
1330 	return pmd;
1331 }
1332 
1333 static inline pmd_t pmd_wrprotect(pmd_t pmd)
1334 {
1335 	pmd_val(pmd) |= _SEGMENT_ENTRY_RO;
1336 	return pmd;
1337 }
1338 
1339 static inline pmd_t pmd_mkdirty(pmd_t pmd)
1340 {
1341 	/* No dirty bit in the segment table entry. */
1342 	return pmd;
1343 }
1344 
1345 static inline pmd_t pmd_mkold(pmd_t pmd)
1346 {
1347 	/* No referenced bit in the segment table entry. */
1348 	return pmd;
1349 }
1350 
1351 static inline pmd_t pmd_mkyoung(pmd_t pmd)
1352 {
1353 	/* No referenced bit in the segment table entry. */
1354 	return pmd;
1355 }
1356 
1357 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1358 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1359 					    unsigned long address, pmd_t *pmdp)
1360 {
1361 	unsigned long pmd_addr = pmd_val(*pmdp) & HPAGE_MASK;
1362 	long tmp, rc;
1363 	int counter;
1364 
1365 	rc = 0;
1366 	if (MACHINE_HAS_RRBM) {
1367 		counter = PTRS_PER_PTE >> 6;
1368 		asm volatile(
1369 			"0:	.insn	rre,0xb9ae0000,%0,%3\n"	/* rrbm */
1370 			"	ogr	%1,%0\n"
1371 			"	la	%3,0(%4,%3)\n"
1372 			"	brct	%2,0b\n"
1373 			: "=&d" (tmp), "+&d" (rc), "+d" (counter),
1374 			  "+a" (pmd_addr)
1375 			: "a" (64 * 4096UL) : "cc");
1376 		rc = !!rc;
1377 	} else {
1378 		counter = PTRS_PER_PTE;
1379 		asm volatile(
1380 			"0:	rrbe	0,%2\n"
1381 			"	la	%2,0(%3,%2)\n"
1382 			"	brc	12,1f\n"
1383 			"	lhi	%0,1\n"
1384 			"1:	brct	%1,0b\n"
1385 			: "+d" (rc), "+d" (counter), "+a" (pmd_addr)
1386 			: "a" (4096UL) : "cc");
1387 	}
1388 	return rc;
1389 }
1390 
1391 #define __HAVE_ARCH_PMDP_GET_AND_CLEAR
1392 static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
1393 				       unsigned long address, pmd_t *pmdp)
1394 {
1395 	pmd_t pmd = *pmdp;
1396 
1397 	__pmd_idte(address, pmdp);
1398 	pmd_clear(pmdp);
1399 	return pmd;
1400 }
1401 
1402 #define __HAVE_ARCH_PMDP_CLEAR_FLUSH
1403 static inline pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
1404 				     unsigned long address, pmd_t *pmdp)
1405 {
1406 	return pmdp_get_and_clear(vma->vm_mm, address, pmdp);
1407 }
1408 
1409 #define __HAVE_ARCH_PMDP_INVALIDATE
1410 static inline void pmdp_invalidate(struct vm_area_struct *vma,
1411 				   unsigned long address, pmd_t *pmdp)
1412 {
1413 	__pmd_idte(address, pmdp);
1414 }
1415 
1416 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1417 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1418 				      unsigned long address, pmd_t *pmdp)
1419 {
1420 	pmd_t pmd = *pmdp;
1421 
1422 	if (pmd_write(pmd)) {
1423 		__pmd_idte(address, pmdp);
1424 		set_pmd_at(mm, address, pmdp, pmd_wrprotect(pmd));
1425 	}
1426 }
1427 
1428 static inline pmd_t mk_pmd_phys(unsigned long physpage, pgprot_t pgprot)
1429 {
1430 	pmd_t __pmd;
1431 	pmd_val(__pmd) = physpage + massage_pgprot_pmd(pgprot);
1432 	return __pmd;
1433 }
1434 
1435 #define pfn_pmd(pfn, pgprot)	mk_pmd_phys(__pa((pfn) << PAGE_SHIFT), (pgprot))
1436 #define mk_pmd(page, pgprot)	pfn_pmd(page_to_pfn(page), (pgprot))
1437 
1438 static inline int pmd_trans_huge(pmd_t pmd)
1439 {
1440 	return pmd_val(pmd) & _SEGMENT_ENTRY_LARGE;
1441 }
1442 
1443 static inline int has_transparent_hugepage(void)
1444 {
1445 	return MACHINE_HAS_HPAGE ? 1 : 0;
1446 }
1447 
1448 static inline unsigned long pmd_pfn(pmd_t pmd)
1449 {
1450 	return pmd_val(pmd) >> PAGE_SHIFT;
1451 }
1452 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1453 
1454 /*
1455  * 31 bit swap entry format:
1456  * A page-table entry has some bits we have to treat in a special way.
1457  * Bits 0, 20 and bit 23 have to be zero, otherwise an specification
1458  * exception will occur instead of a page translation exception. The
1459  * specifiation exception has the bad habit not to store necessary
1460  * information in the lowcore.
1461  * Bit 21 and bit 22 are the page invalid bit and the page protection
1462  * bit. We set both to indicate a swapped page.
1463  * Bit 30 and 31 are used to distinguish the different page types. For
1464  * a swapped page these bits need to be zero.
1465  * This leaves the bits 1-19 and bits 24-29 to store type and offset.
1466  * We use the 5 bits from 25-29 for the type and the 20 bits from 1-19
1467  * plus 24 for the offset.
1468  * 0|     offset        |0110|o|type |00|
1469  * 0 0000000001111111111 2222 2 22222 33
1470  * 0 1234567890123456789 0123 4 56789 01
1471  *
1472  * 64 bit swap entry format:
1473  * A page-table entry has some bits we have to treat in a special way.
1474  * Bits 52 and bit 55 have to be zero, otherwise an specification
1475  * exception will occur instead of a page translation exception. The
1476  * specifiation exception has the bad habit not to store necessary
1477  * information in the lowcore.
1478  * Bit 53 and bit 54 are the page invalid bit and the page protection
1479  * bit. We set both to indicate a swapped page.
1480  * Bit 62 and 63 are used to distinguish the different page types. For
1481  * a swapped page these bits need to be zero.
1482  * This leaves the bits 0-51 and bits 56-61 to store type and offset.
1483  * We use the 5 bits from 57-61 for the type and the 53 bits from 0-51
1484  * plus 56 for the offset.
1485  * |                      offset                        |0110|o|type |00|
1486  *  0000000000111111111122222222223333333333444444444455 5555 5 55566 66
1487  *  0123456789012345678901234567890123456789012345678901 2345 6 78901 23
1488  */
1489 #ifndef CONFIG_64BIT
1490 #define __SWP_OFFSET_MASK (~0UL >> 12)
1491 #else
1492 #define __SWP_OFFSET_MASK (~0UL >> 11)
1493 #endif
1494 static inline pte_t mk_swap_pte(unsigned long type, unsigned long offset)
1495 {
1496 	pte_t pte;
1497 	offset &= __SWP_OFFSET_MASK;
1498 	pte_val(pte) = _PAGE_TYPE_SWAP | ((type & 0x1f) << 2) |
1499 		((offset & 1UL) << 7) | ((offset & ~1UL) << 11);
1500 	return pte;
1501 }
1502 
1503 #define __swp_type(entry)	(((entry).val >> 2) & 0x1f)
1504 #define __swp_offset(entry)	(((entry).val >> 11) | (((entry).val >> 7) & 1))
1505 #define __swp_entry(type,offset) ((swp_entry_t) { pte_val(mk_swap_pte((type),(offset))) })
1506 
1507 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
1508 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
1509 
1510 #ifndef CONFIG_64BIT
1511 # define PTE_FILE_MAX_BITS	26
1512 #else /* CONFIG_64BIT */
1513 # define PTE_FILE_MAX_BITS	59
1514 #endif /* CONFIG_64BIT */
1515 
1516 #define pte_to_pgoff(__pte) \
1517 	((((__pte).pte >> 12) << 7) + (((__pte).pte >> 1) & 0x7f))
1518 
1519 #define pgoff_to_pte(__off) \
1520 	((pte_t) { ((((__off) & 0x7f) << 1) + (((__off) >> 7) << 12)) \
1521 		   | _PAGE_TYPE_FILE })
1522 
1523 #endif /* !__ASSEMBLY__ */
1524 
1525 #define kern_addr_valid(addr)   (1)
1526 
1527 extern int vmem_add_mapping(unsigned long start, unsigned long size);
1528 extern int vmem_remove_mapping(unsigned long start, unsigned long size);
1529 extern int s390_enable_sie(void);
1530 
1531 /*
1532  * No page table caches to initialise
1533  */
1534 #define pgtable_cache_init()	do { } while (0)
1535 
1536 #include <asm-generic/pgtable.h>
1537 
1538 #endif /* _S390_PAGE_H */
1539