xref: /openbmc/linux/arch/s390/include/asm/pgtable.h (revision ee89bd6b)
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 /* TODO: s390 cannot support io_remap_pfn_range... */
61 #define io_remap_pfn_range(vma, vaddr, pfn, size, prot) 	       \
62 	remap_pfn_range(vma, vaddr, pfn, size, prot)
63 
64 #endif /* !__ASSEMBLY__ */
65 
66 /*
67  * PMD_SHIFT determines the size of the area a second-level page
68  * table can map
69  * PGDIR_SHIFT determines what a third-level page table entry can map
70  */
71 #ifndef CONFIG_64BIT
72 # define PMD_SHIFT	20
73 # define PUD_SHIFT	20
74 # define PGDIR_SHIFT	20
75 #else /* CONFIG_64BIT */
76 # define PMD_SHIFT	20
77 # define PUD_SHIFT	31
78 # define PGDIR_SHIFT	42
79 #endif /* CONFIG_64BIT */
80 
81 #define PMD_SIZE        (1UL << PMD_SHIFT)
82 #define PMD_MASK        (~(PMD_SIZE-1))
83 #define PUD_SIZE	(1UL << PUD_SHIFT)
84 #define PUD_MASK	(~(PUD_SIZE-1))
85 #define PGDIR_SIZE	(1UL << PGDIR_SHIFT)
86 #define PGDIR_MASK	(~(PGDIR_SIZE-1))
87 
88 /*
89  * entries per page directory level: the S390 is two-level, so
90  * we don't really have any PMD directory physically.
91  * for S390 segment-table entries are combined to one PGD
92  * that leads to 1024 pte per pgd
93  */
94 #define PTRS_PER_PTE	256
95 #ifndef CONFIG_64BIT
96 #define PTRS_PER_PMD	1
97 #define PTRS_PER_PUD	1
98 #else /* CONFIG_64BIT */
99 #define PTRS_PER_PMD	2048
100 #define PTRS_PER_PUD	2048
101 #endif /* CONFIG_64BIT */
102 #define PTRS_PER_PGD	2048
103 
104 #define FIRST_USER_ADDRESS  0
105 
106 #define pte_ERROR(e) \
107 	printk("%s:%d: bad pte %p.\n", __FILE__, __LINE__, (void *) pte_val(e))
108 #define pmd_ERROR(e) \
109 	printk("%s:%d: bad pmd %p.\n", __FILE__, __LINE__, (void *) pmd_val(e))
110 #define pud_ERROR(e) \
111 	printk("%s:%d: bad pud %p.\n", __FILE__, __LINE__, (void *) pud_val(e))
112 #define pgd_ERROR(e) \
113 	printk("%s:%d: bad pgd %p.\n", __FILE__, __LINE__, (void *) pgd_val(e))
114 
115 #ifndef __ASSEMBLY__
116 /*
117  * The vmalloc and module area will always be on the topmost area of the kernel
118  * mapping. We reserve 96MB (31bit) / 128GB (64bit) for vmalloc and modules.
119  * On 64 bit kernels we have a 2GB area at the top of the vmalloc area where
120  * modules will reside. That makes sure that inter module branches always
121  * happen without trampolines and in addition the placement within a 2GB frame
122  * is branch prediction unit friendly.
123  */
124 extern unsigned long VMALLOC_START;
125 extern unsigned long VMALLOC_END;
126 extern struct page *vmemmap;
127 
128 #define VMEM_MAX_PHYS ((unsigned long) vmemmap)
129 
130 #ifdef CONFIG_64BIT
131 extern unsigned long MODULES_VADDR;
132 extern unsigned long MODULES_END;
133 #define MODULES_VADDR	MODULES_VADDR
134 #define MODULES_END	MODULES_END
135 #define MODULES_LEN	(1UL << 31)
136 #endif
137 
138 /*
139  * A 31 bit pagetable entry of S390 has following format:
140  *  |   PFRA          |    |  OS  |
141  * 0                   0IP0
142  * 00000000001111111111222222222233
143  * 01234567890123456789012345678901
144  *
145  * I Page-Invalid Bit:    Page is not available for address-translation
146  * P Page-Protection Bit: Store access not possible for page
147  *
148  * A 31 bit segmenttable entry of S390 has following format:
149  *  |   P-table origin      |  |PTL
150  * 0                         IC
151  * 00000000001111111111222222222233
152  * 01234567890123456789012345678901
153  *
154  * I Segment-Invalid Bit:    Segment is not available for address-translation
155  * C Common-Segment Bit:     Segment is not private (PoP 3-30)
156  * PTL Page-Table-Length:    Page-table length (PTL+1*16 entries -> up to 256)
157  *
158  * The 31 bit segmenttable origin of S390 has following format:
159  *
160  *  |S-table origin   |     | STL |
161  * X                   **GPS
162  * 00000000001111111111222222222233
163  * 01234567890123456789012345678901
164  *
165  * X Space-Switch event:
166  * G Segment-Invalid Bit:     *
167  * P Private-Space Bit:       Segment is not private (PoP 3-30)
168  * S Storage-Alteration:
169  * STL Segment-Table-Length:  Segment-table length (STL+1*16 entries -> up to 2048)
170  *
171  * A 64 bit pagetable entry of S390 has following format:
172  * |			 PFRA			      |0IPC|  OS  |
173  * 0000000000111111111122222222223333333333444444444455555555556666
174  * 0123456789012345678901234567890123456789012345678901234567890123
175  *
176  * I Page-Invalid Bit:    Page is not available for address-translation
177  * P Page-Protection Bit: Store access not possible for page
178  * C Change-bit override: HW is not required to set change bit
179  *
180  * A 64 bit segmenttable entry of S390 has following format:
181  * |        P-table origin                              |      TT
182  * 0000000000111111111122222222223333333333444444444455555555556666
183  * 0123456789012345678901234567890123456789012345678901234567890123
184  *
185  * I Segment-Invalid Bit:    Segment is not available for address-translation
186  * C Common-Segment Bit:     Segment is not private (PoP 3-30)
187  * P Page-Protection Bit: Store access not possible for page
188  * TT Type 00
189  *
190  * A 64 bit region table entry of S390 has following format:
191  * |        S-table origin                             |   TF  TTTL
192  * 0000000000111111111122222222223333333333444444444455555555556666
193  * 0123456789012345678901234567890123456789012345678901234567890123
194  *
195  * I Segment-Invalid Bit:    Segment is not available for address-translation
196  * TT Type 01
197  * TF
198  * TL Table length
199  *
200  * The 64 bit regiontable origin of S390 has following format:
201  * |      region table origon                          |       DTTL
202  * 0000000000111111111122222222223333333333444444444455555555556666
203  * 0123456789012345678901234567890123456789012345678901234567890123
204  *
205  * X Space-Switch event:
206  * G Segment-Invalid Bit:
207  * P Private-Space Bit:
208  * S Storage-Alteration:
209  * R Real space
210  * TL Table-Length:
211  *
212  * A storage key has the following format:
213  * | ACC |F|R|C|0|
214  *  0   3 4 5 6 7
215  * ACC: access key
216  * F  : fetch protection bit
217  * R  : referenced bit
218  * C  : changed bit
219  */
220 
221 /* Hardware bits in the page table entry */
222 #define _PAGE_CO	0x100		/* HW Change-bit override */
223 #define _PAGE_RO	0x200		/* HW read-only bit  */
224 #define _PAGE_INVALID	0x400		/* HW invalid bit    */
225 
226 /* Software bits in the page table entry */
227 #define _PAGE_SWT	0x001		/* SW pte type bit t */
228 #define _PAGE_SWX	0x002		/* SW pte type bit x */
229 #define _PAGE_SWC	0x004		/* SW pte changed bit */
230 #define _PAGE_SWR	0x008		/* SW pte referenced bit */
231 #define _PAGE_SWW	0x010		/* SW pte write bit */
232 #define _PAGE_SPECIAL	0x020		/* SW associated with special page */
233 #define __HAVE_ARCH_PTE_SPECIAL
234 
235 /* Set of bits not changed in pte_modify */
236 #define _PAGE_CHG_MASK		(PAGE_MASK | _PAGE_SPECIAL | _PAGE_CO | \
237 				 _PAGE_SWC | _PAGE_SWR)
238 
239 /* Six different types of pages. */
240 #define _PAGE_TYPE_EMPTY	0x400
241 #define _PAGE_TYPE_NONE		0x401
242 #define _PAGE_TYPE_SWAP		0x403
243 #define _PAGE_TYPE_FILE		0x601	/* bit 0x002 is used for offset !! */
244 #define _PAGE_TYPE_RO		0x200
245 #define _PAGE_TYPE_RW		0x000
246 
247 /*
248  * Only four types for huge pages, using the invalid bit and protection bit
249  * of a segment table entry.
250  */
251 #define _HPAGE_TYPE_EMPTY	0x020	/* _SEGMENT_ENTRY_INV */
252 #define _HPAGE_TYPE_NONE	0x220
253 #define _HPAGE_TYPE_RO		0x200	/* _SEGMENT_ENTRY_RO  */
254 #define _HPAGE_TYPE_RW		0x000
255 
256 /*
257  * PTE type bits are rather complicated. handle_pte_fault uses pte_present,
258  * pte_none and pte_file to find out the pte type WITHOUT holding the page
259  * table lock. ptep_clear_flush on the other hand uses ptep_clear_flush to
260  * invalidate a given pte. ipte sets the hw invalid bit and clears all tlbs
261  * for the page. The page table entry is set to _PAGE_TYPE_EMPTY afterwards.
262  * This change is done while holding the lock, but the intermediate step
263  * of a previously valid pte with the hw invalid bit set can be observed by
264  * handle_pte_fault. That makes it necessary that all valid pte types with
265  * the hw invalid bit set must be distinguishable from the four pte types
266  * empty, none, swap and file.
267  *
268  *			irxt  ipte  irxt
269  * _PAGE_TYPE_EMPTY	1000   ->   1000
270  * _PAGE_TYPE_NONE	1001   ->   1001
271  * _PAGE_TYPE_SWAP	1011   ->   1011
272  * _PAGE_TYPE_FILE	11?1   ->   11?1
273  * _PAGE_TYPE_RO	0100   ->   1100
274  * _PAGE_TYPE_RW	0000   ->   1000
275  *
276  * pte_none is true for bits combinations 1000, 1010, 1100, 1110
277  * pte_present is true for bits combinations 0000, 0010, 0100, 0110, 1001
278  * pte_file is true for bits combinations 1101, 1111
279  * swap pte is 1011 and 0001, 0011, 0101, 0111 are invalid.
280  */
281 
282 #ifndef CONFIG_64BIT
283 
284 /* Bits in the segment table address-space-control-element */
285 #define _ASCE_SPACE_SWITCH	0x80000000UL	/* space switch event	    */
286 #define _ASCE_ORIGIN_MASK	0x7ffff000UL	/* segment table origin	    */
287 #define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
288 #define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
289 #define _ASCE_TABLE_LENGTH	0x7f	/* 128 x 64 entries = 8k	    */
290 
291 /* Bits in the segment table entry */
292 #define _SEGMENT_ENTRY_ORIGIN	0x7fffffc0UL	/* page table origin	    */
293 #define _SEGMENT_ENTRY_RO	0x200	/* page protection bit		    */
294 #define _SEGMENT_ENTRY_INV	0x20	/* invalid segment table entry	    */
295 #define _SEGMENT_ENTRY_COMMON	0x10	/* common segment bit		    */
296 #define _SEGMENT_ENTRY_PTL	0x0f	/* page table length		    */
297 
298 #define _SEGMENT_ENTRY		(_SEGMENT_ENTRY_PTL)
299 #define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INV)
300 
301 /* Page status table bits for virtualization */
302 #define RCP_ACC_BITS	0xf0000000UL
303 #define RCP_FP_BIT	0x08000000UL
304 #define RCP_PCL_BIT	0x00800000UL
305 #define RCP_HR_BIT	0x00400000UL
306 #define RCP_HC_BIT	0x00200000UL
307 #define RCP_GR_BIT	0x00040000UL
308 #define RCP_GC_BIT	0x00020000UL
309 #define RCP_IN_BIT	0x00002000UL	/* IPTE notify bit */
310 
311 /* User dirty / referenced bit for KVM's migration feature */
312 #define KVM_UR_BIT	0x00008000UL
313 #define KVM_UC_BIT	0x00004000UL
314 
315 #else /* CONFIG_64BIT */
316 
317 /* Bits in the segment/region table address-space-control-element */
318 #define _ASCE_ORIGIN		~0xfffUL/* segment table origin		    */
319 #define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
320 #define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
321 #define _ASCE_SPACE_SWITCH	0x40	/* space switch event		    */
322 #define _ASCE_REAL_SPACE	0x20	/* real space control		    */
323 #define _ASCE_TYPE_MASK		0x0c	/* asce table type mask		    */
324 #define _ASCE_TYPE_REGION1	0x0c	/* region first table type	    */
325 #define _ASCE_TYPE_REGION2	0x08	/* region second table type	    */
326 #define _ASCE_TYPE_REGION3	0x04	/* region third table type	    */
327 #define _ASCE_TYPE_SEGMENT	0x00	/* segment table type		    */
328 #define _ASCE_TABLE_LENGTH	0x03	/* region table length		    */
329 
330 /* Bits in the region table entry */
331 #define _REGION_ENTRY_ORIGIN	~0xfffUL/* region/segment table origin	    */
332 #define _REGION_ENTRY_RO	0x200	/* region protection bit	    */
333 #define _REGION_ENTRY_INV	0x20	/* invalid region table entry	    */
334 #define _REGION_ENTRY_TYPE_MASK	0x0c	/* region/segment table type mask   */
335 #define _REGION_ENTRY_TYPE_R1	0x0c	/* region first table type	    */
336 #define _REGION_ENTRY_TYPE_R2	0x08	/* region second table type	    */
337 #define _REGION_ENTRY_TYPE_R3	0x04	/* region third table type	    */
338 #define _REGION_ENTRY_LENGTH	0x03	/* region third length		    */
339 
340 #define _REGION1_ENTRY		(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_LENGTH)
341 #define _REGION1_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INV)
342 #define _REGION2_ENTRY		(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_LENGTH)
343 #define _REGION2_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INV)
344 #define _REGION3_ENTRY		(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_LENGTH)
345 #define _REGION3_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INV)
346 
347 #define _REGION3_ENTRY_LARGE	0x400	/* RTTE-format control, large page  */
348 #define _REGION3_ENTRY_RO	0x200	/* page protection bit		    */
349 #define _REGION3_ENTRY_CO	0x100	/* change-recording override	    */
350 
351 /* Bits in the segment table entry */
352 #define _SEGMENT_ENTRY_ORIGIN_LARGE ~0xfffffUL /* large page address	    */
353 #define _SEGMENT_ENTRY_ORIGIN	~0x7ffUL/* segment table origin		    */
354 #define _SEGMENT_ENTRY_RO	0x200	/* page protection bit		    */
355 #define _SEGMENT_ENTRY_INV	0x20	/* invalid segment table entry	    */
356 
357 #define _SEGMENT_ENTRY		(0)
358 #define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INV)
359 
360 #define _SEGMENT_ENTRY_LARGE	0x400	/* STE-format control, large page   */
361 #define _SEGMENT_ENTRY_CO	0x100	/* change-recording override   */
362 #define _SEGMENT_ENTRY_SPLIT_BIT 0	/* THP splitting bit number */
363 #define _SEGMENT_ENTRY_SPLIT	(1UL << _SEGMENT_ENTRY_SPLIT_BIT)
364 
365 /* Set of bits not changed in pmd_modify */
366 #define _SEGMENT_CHG_MASK	(_SEGMENT_ENTRY_ORIGIN | _SEGMENT_ENTRY_LARGE \
367 				 | _SEGMENT_ENTRY_SPLIT | _SEGMENT_ENTRY_CO)
368 
369 /* Page status table bits for virtualization */
370 #define RCP_ACC_BITS	0xf000000000000000UL
371 #define RCP_FP_BIT	0x0800000000000000UL
372 #define RCP_PCL_BIT	0x0080000000000000UL
373 #define RCP_HR_BIT	0x0040000000000000UL
374 #define RCP_HC_BIT	0x0020000000000000UL
375 #define RCP_GR_BIT	0x0004000000000000UL
376 #define RCP_GC_BIT	0x0002000000000000UL
377 #define RCP_IN_BIT	0x0000200000000000UL	/* IPTE notify bit */
378 
379 /* User dirty / referenced bit for KVM's migration feature */
380 #define KVM_UR_BIT	0x0000800000000000UL
381 #define KVM_UC_BIT	0x0000400000000000UL
382 
383 #endif /* CONFIG_64BIT */
384 
385 /*
386  * A user page table pointer has the space-switch-event bit, the
387  * private-space-control bit and the storage-alteration-event-control
388  * bit set. A kernel page table pointer doesn't need them.
389  */
390 #define _ASCE_USER_BITS		(_ASCE_SPACE_SWITCH | _ASCE_PRIVATE_SPACE | \
391 				 _ASCE_ALT_EVENT)
392 
393 /*
394  * Page protection definitions.
395  */
396 #define PAGE_NONE	__pgprot(_PAGE_TYPE_NONE)
397 #define PAGE_RO		__pgprot(_PAGE_TYPE_RO)
398 #define PAGE_RW		__pgprot(_PAGE_TYPE_RO | _PAGE_SWW)
399 #define PAGE_RWC	__pgprot(_PAGE_TYPE_RW | _PAGE_SWW | _PAGE_SWC)
400 
401 #define PAGE_KERNEL	PAGE_RWC
402 #define PAGE_SHARED	PAGE_KERNEL
403 #define PAGE_COPY	PAGE_RO
404 
405 /*
406  * On s390 the page table entry has an invalid bit and a read-only bit.
407  * Read permission implies execute permission and write permission
408  * implies read permission.
409  */
410          /*xwr*/
411 #define __P000	PAGE_NONE
412 #define __P001	PAGE_RO
413 #define __P010	PAGE_RO
414 #define __P011	PAGE_RO
415 #define __P100	PAGE_RO
416 #define __P101	PAGE_RO
417 #define __P110	PAGE_RO
418 #define __P111	PAGE_RO
419 
420 #define __S000	PAGE_NONE
421 #define __S001	PAGE_RO
422 #define __S010	PAGE_RW
423 #define __S011	PAGE_RW
424 #define __S100	PAGE_RO
425 #define __S101	PAGE_RO
426 #define __S110	PAGE_RW
427 #define __S111	PAGE_RW
428 
429 /*
430  * Segment entry (large page) protection definitions.
431  */
432 #define SEGMENT_NONE	__pgprot(_HPAGE_TYPE_NONE)
433 #define SEGMENT_RO	__pgprot(_HPAGE_TYPE_RO)
434 #define SEGMENT_RW	__pgprot(_HPAGE_TYPE_RW)
435 
436 static inline int mm_exclusive(struct mm_struct *mm)
437 {
438 	return likely(mm == current->active_mm &&
439 		      atomic_read(&mm->context.attach_count) <= 1);
440 }
441 
442 static inline int mm_has_pgste(struct mm_struct *mm)
443 {
444 #ifdef CONFIG_PGSTE
445 	if (unlikely(mm->context.has_pgste))
446 		return 1;
447 #endif
448 	return 0;
449 }
450 /*
451  * pgd/pmd/pte query functions
452  */
453 #ifndef CONFIG_64BIT
454 
455 static inline int pgd_present(pgd_t pgd) { return 1; }
456 static inline int pgd_none(pgd_t pgd)    { return 0; }
457 static inline int pgd_bad(pgd_t pgd)     { return 0; }
458 
459 static inline int pud_present(pud_t pud) { return 1; }
460 static inline int pud_none(pud_t pud)	 { return 0; }
461 static inline int pud_large(pud_t pud)	 { return 0; }
462 static inline int pud_bad(pud_t pud)	 { return 0; }
463 
464 #else /* CONFIG_64BIT */
465 
466 static inline int pgd_present(pgd_t pgd)
467 {
468 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
469 		return 1;
470 	return (pgd_val(pgd) & _REGION_ENTRY_ORIGIN) != 0UL;
471 }
472 
473 static inline int pgd_none(pgd_t pgd)
474 {
475 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
476 		return 0;
477 	return (pgd_val(pgd) & _REGION_ENTRY_INV) != 0UL;
478 }
479 
480 static inline int pgd_bad(pgd_t pgd)
481 {
482 	/*
483 	 * With dynamic page table levels the pgd can be a region table
484 	 * entry or a segment table entry. Check for the bit that are
485 	 * invalid for either table entry.
486 	 */
487 	unsigned long mask =
488 		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INV &
489 		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
490 	return (pgd_val(pgd) & mask) != 0;
491 }
492 
493 static inline int pud_present(pud_t pud)
494 {
495 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
496 		return 1;
497 	return (pud_val(pud) & _REGION_ENTRY_ORIGIN) != 0UL;
498 }
499 
500 static inline int pud_none(pud_t pud)
501 {
502 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
503 		return 0;
504 	return (pud_val(pud) & _REGION_ENTRY_INV) != 0UL;
505 }
506 
507 static inline int pud_large(pud_t pud)
508 {
509 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) != _REGION_ENTRY_TYPE_R3)
510 		return 0;
511 	return !!(pud_val(pud) & _REGION3_ENTRY_LARGE);
512 }
513 
514 static inline int pud_bad(pud_t pud)
515 {
516 	/*
517 	 * With dynamic page table levels the pud can be a region table
518 	 * entry or a segment table entry. Check for the bit that are
519 	 * invalid for either table entry.
520 	 */
521 	unsigned long mask =
522 		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INV &
523 		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
524 	return (pud_val(pud) & mask) != 0;
525 }
526 
527 #endif /* CONFIG_64BIT */
528 
529 static inline int pmd_present(pmd_t pmd)
530 {
531 	unsigned long mask = _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO;
532 	return (pmd_val(pmd) & mask) == _HPAGE_TYPE_NONE ||
533 	       !(pmd_val(pmd) & _SEGMENT_ENTRY_INV);
534 }
535 
536 static inline int pmd_none(pmd_t pmd)
537 {
538 	return (pmd_val(pmd) & _SEGMENT_ENTRY_INV) &&
539 	       !(pmd_val(pmd) & _SEGMENT_ENTRY_RO);
540 }
541 
542 static inline int pmd_large(pmd_t pmd)
543 {
544 #ifdef CONFIG_64BIT
545 	return !!(pmd_val(pmd) & _SEGMENT_ENTRY_LARGE);
546 #else
547 	return 0;
548 #endif
549 }
550 
551 static inline int pmd_bad(pmd_t pmd)
552 {
553 	unsigned long mask = ~_SEGMENT_ENTRY_ORIGIN & ~_SEGMENT_ENTRY_INV;
554 	return (pmd_val(pmd) & mask) != _SEGMENT_ENTRY;
555 }
556 
557 #define __HAVE_ARCH_PMDP_SPLITTING_FLUSH
558 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
559 				 unsigned long addr, pmd_t *pmdp);
560 
561 #define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
562 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
563 				 unsigned long address, pmd_t *pmdp,
564 				 pmd_t entry, int dirty);
565 
566 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
567 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
568 				  unsigned long address, pmd_t *pmdp);
569 
570 #define __HAVE_ARCH_PMD_WRITE
571 static inline int pmd_write(pmd_t pmd)
572 {
573 	return (pmd_val(pmd) & _SEGMENT_ENTRY_RO) == 0;
574 }
575 
576 static inline int pmd_young(pmd_t pmd)
577 {
578 	return 0;
579 }
580 
581 static inline int pte_none(pte_t pte)
582 {
583 	return (pte_val(pte) & _PAGE_INVALID) && !(pte_val(pte) & _PAGE_SWT);
584 }
585 
586 static inline int pte_present(pte_t pte)
587 {
588 	unsigned long mask = _PAGE_RO | _PAGE_INVALID | _PAGE_SWT | _PAGE_SWX;
589 	return (pte_val(pte) & mask) == _PAGE_TYPE_NONE ||
590 		(!(pte_val(pte) & _PAGE_INVALID) &&
591 		 !(pte_val(pte) & _PAGE_SWT));
592 }
593 
594 static inline int pte_file(pte_t pte)
595 {
596 	unsigned long mask = _PAGE_RO | _PAGE_INVALID | _PAGE_SWT;
597 	return (pte_val(pte) & mask) == _PAGE_TYPE_FILE;
598 }
599 
600 static inline int pte_special(pte_t pte)
601 {
602 	return (pte_val(pte) & _PAGE_SPECIAL);
603 }
604 
605 #define __HAVE_ARCH_PTE_SAME
606 static inline int pte_same(pte_t a, pte_t b)
607 {
608 	return pte_val(a) == pte_val(b);
609 }
610 
611 static inline pgste_t pgste_get_lock(pte_t *ptep)
612 {
613 	unsigned long new = 0;
614 #ifdef CONFIG_PGSTE
615 	unsigned long old;
616 
617 	preempt_disable();
618 	asm(
619 		"	lg	%0,%2\n"
620 		"0:	lgr	%1,%0\n"
621 		"	nihh	%0,0xff7f\n"	/* clear RCP_PCL_BIT in old */
622 		"	oihh	%1,0x0080\n"	/* set RCP_PCL_BIT in new */
623 		"	csg	%0,%1,%2\n"
624 		"	jl	0b\n"
625 		: "=&d" (old), "=&d" (new), "=Q" (ptep[PTRS_PER_PTE])
626 		: "Q" (ptep[PTRS_PER_PTE]) : "cc");
627 #endif
628 	return __pgste(new);
629 }
630 
631 static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste)
632 {
633 #ifdef CONFIG_PGSTE
634 	asm(
635 		"	nihh	%1,0xff7f\n"	/* clear RCP_PCL_BIT */
636 		"	stg	%1,%0\n"
637 		: "=Q" (ptep[PTRS_PER_PTE])
638 		: "d" (pgste_val(pgste)), "Q" (ptep[PTRS_PER_PTE]) : "cc");
639 	preempt_enable();
640 #endif
641 }
642 
643 static inline pgste_t pgste_update_all(pte_t *ptep, pgste_t pgste)
644 {
645 #ifdef CONFIG_PGSTE
646 	unsigned long address, bits;
647 	unsigned char skey;
648 
649 	if (pte_val(*ptep) & _PAGE_INVALID)
650 		return pgste;
651 	address = pte_val(*ptep) & PAGE_MASK;
652 	skey = page_get_storage_key(address);
653 	bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
654 	/* Clear page changed & referenced bit in the storage key */
655 	if (bits & _PAGE_CHANGED)
656 		page_set_storage_key(address, skey ^ bits, 0);
657 	else if (bits)
658 		page_reset_referenced(address);
659 	/* Transfer page changed & referenced bit to guest bits in pgste */
660 	pgste_val(pgste) |= bits << 48;		/* RCP_GR_BIT & RCP_GC_BIT */
661 	/* Get host changed & referenced bits from pgste */
662 	bits |= (pgste_val(pgste) & (RCP_HR_BIT | RCP_HC_BIT)) >> 52;
663 	/* Transfer page changed & referenced bit to kvm user bits */
664 	pgste_val(pgste) |= bits << 45;		/* KVM_UR_BIT & KVM_UC_BIT */
665 	/* Clear relevant host bits in pgste. */
666 	pgste_val(pgste) &= ~(RCP_HR_BIT | RCP_HC_BIT);
667 	pgste_val(pgste) &= ~(RCP_ACC_BITS | RCP_FP_BIT);
668 	/* Copy page access key and fetch protection bit to pgste */
669 	pgste_val(pgste) |=
670 		(unsigned long) (skey & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
671 	/* Transfer referenced bit to pte */
672 	pte_val(*ptep) |= (bits & _PAGE_REFERENCED) << 1;
673 #endif
674 	return pgste;
675 
676 }
677 
678 static inline pgste_t pgste_update_young(pte_t *ptep, pgste_t pgste)
679 {
680 #ifdef CONFIG_PGSTE
681 	int young;
682 
683 	if (pte_val(*ptep) & _PAGE_INVALID)
684 		return pgste;
685 	/* Get referenced bit from storage key */
686 	young = page_reset_referenced(pte_val(*ptep) & PAGE_MASK);
687 	if (young)
688 		pgste_val(pgste) |= RCP_GR_BIT;
689 	/* Get host referenced bit from pgste */
690 	if (pgste_val(pgste) & RCP_HR_BIT) {
691 		pgste_val(pgste) &= ~RCP_HR_BIT;
692 		young = 1;
693 	}
694 	/* Transfer referenced bit to kvm user bits and pte */
695 	if (young) {
696 		pgste_val(pgste) |= KVM_UR_BIT;
697 		pte_val(*ptep) |= _PAGE_SWR;
698 	}
699 #endif
700 	return pgste;
701 }
702 
703 static inline void pgste_set_key(pte_t *ptep, pgste_t pgste, pte_t entry)
704 {
705 #ifdef CONFIG_PGSTE
706 	unsigned long address;
707 	unsigned long okey, nkey;
708 
709 	if (pte_val(entry) & _PAGE_INVALID)
710 		return;
711 	address = pte_val(entry) & PAGE_MASK;
712 	okey = nkey = page_get_storage_key(address);
713 	nkey &= ~(_PAGE_ACC_BITS | _PAGE_FP_BIT);
714 	/* Set page access key and fetch protection bit from pgste */
715 	nkey |= (pgste_val(pgste) & (RCP_ACC_BITS | RCP_FP_BIT)) >> 56;
716 	if (okey != nkey)
717 		page_set_storage_key(address, nkey, 0);
718 #endif
719 }
720 
721 static inline void pgste_set_pte(pte_t *ptep, pte_t entry)
722 {
723 	if (!MACHINE_HAS_ESOP && (pte_val(entry) & _PAGE_SWW)) {
724 		/*
725 		 * Without enhanced suppression-on-protection force
726 		 * the dirty bit on for all writable ptes.
727 		 */
728 		pte_val(entry) |= _PAGE_SWC;
729 		pte_val(entry) &= ~_PAGE_RO;
730 	}
731 	*ptep = entry;
732 }
733 
734 /**
735  * struct gmap_struct - guest address space
736  * @mm: pointer to the parent mm_struct
737  * @table: pointer to the page directory
738  * @asce: address space control element for gmap page table
739  * @crst_list: list of all crst tables used in the guest address space
740  */
741 struct gmap {
742 	struct list_head list;
743 	struct mm_struct *mm;
744 	unsigned long *table;
745 	unsigned long asce;
746 	struct list_head crst_list;
747 };
748 
749 /**
750  * struct gmap_rmap - reverse mapping for segment table entries
751  * @gmap: pointer to the gmap_struct
752  * @entry: pointer to a segment table entry
753  * @vmaddr: virtual address in the guest address space
754  */
755 struct gmap_rmap {
756 	struct list_head list;
757 	struct gmap *gmap;
758 	unsigned long *entry;
759 	unsigned long vmaddr;
760 };
761 
762 /**
763  * struct gmap_pgtable - gmap information attached to a page table
764  * @vmaddr: address of the 1MB segment in the process virtual memory
765  * @mapper: list of segment table entries mapping a page table
766  */
767 struct gmap_pgtable {
768 	unsigned long vmaddr;
769 	struct list_head mapper;
770 };
771 
772 /**
773  * struct gmap_notifier - notify function block for page invalidation
774  * @notifier_call: address of callback function
775  */
776 struct gmap_notifier {
777 	struct list_head list;
778 	void (*notifier_call)(struct gmap *gmap, unsigned long address);
779 };
780 
781 struct gmap *gmap_alloc(struct mm_struct *mm);
782 void gmap_free(struct gmap *gmap);
783 void gmap_enable(struct gmap *gmap);
784 void gmap_disable(struct gmap *gmap);
785 int gmap_map_segment(struct gmap *gmap, unsigned long from,
786 		     unsigned long to, unsigned long len);
787 int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len);
788 unsigned long __gmap_translate(unsigned long address, struct gmap *);
789 unsigned long gmap_translate(unsigned long address, struct gmap *);
790 unsigned long __gmap_fault(unsigned long address, struct gmap *);
791 unsigned long gmap_fault(unsigned long address, struct gmap *);
792 void gmap_discard(unsigned long from, unsigned long to, struct gmap *);
793 
794 void gmap_register_ipte_notifier(struct gmap_notifier *);
795 void gmap_unregister_ipte_notifier(struct gmap_notifier *);
796 int gmap_ipte_notify(struct gmap *, unsigned long start, unsigned long len);
797 void gmap_do_ipte_notify(struct mm_struct *, unsigned long addr, pte_t *);
798 
799 static inline pgste_t pgste_ipte_notify(struct mm_struct *mm,
800 					unsigned long addr,
801 					pte_t *ptep, pgste_t pgste)
802 {
803 #ifdef CONFIG_PGSTE
804 	if (pgste_val(pgste) & RCP_IN_BIT) {
805 		pgste_val(pgste) &= ~RCP_IN_BIT;
806 		gmap_do_ipte_notify(mm, addr, ptep);
807 	}
808 #endif
809 	return pgste;
810 }
811 
812 /*
813  * Certain architectures need to do special things when PTEs
814  * within a page table are directly modified.  Thus, the following
815  * hook is made available.
816  */
817 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
818 			      pte_t *ptep, pte_t entry)
819 {
820 	pgste_t pgste;
821 
822 	if (mm_has_pgste(mm)) {
823 		pgste = pgste_get_lock(ptep);
824 		pgste_set_key(ptep, pgste, entry);
825 		pgste_set_pte(ptep, entry);
826 		pgste_set_unlock(ptep, pgste);
827 	} else {
828 		if (!(pte_val(entry) & _PAGE_INVALID) && MACHINE_HAS_EDAT1)
829 			pte_val(entry) |= _PAGE_CO;
830 		*ptep = entry;
831 	}
832 }
833 
834 /*
835  * query functions pte_write/pte_dirty/pte_young only work if
836  * pte_present() is true. Undefined behaviour if not..
837  */
838 static inline int pte_write(pte_t pte)
839 {
840 	return (pte_val(pte) & _PAGE_SWW) != 0;
841 }
842 
843 static inline int pte_dirty(pte_t pte)
844 {
845 	return (pte_val(pte) & _PAGE_SWC) != 0;
846 }
847 
848 static inline int pte_young(pte_t pte)
849 {
850 #ifdef CONFIG_PGSTE
851 	if (pte_val(pte) & _PAGE_SWR)
852 		return 1;
853 #endif
854 	return 0;
855 }
856 
857 /*
858  * pgd/pmd/pte modification functions
859  */
860 
861 static inline void pgd_clear(pgd_t *pgd)
862 {
863 #ifdef CONFIG_64BIT
864 	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
865 		pgd_val(*pgd) = _REGION2_ENTRY_EMPTY;
866 #endif
867 }
868 
869 static inline void pud_clear(pud_t *pud)
870 {
871 #ifdef CONFIG_64BIT
872 	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
873 		pud_val(*pud) = _REGION3_ENTRY_EMPTY;
874 #endif
875 }
876 
877 static inline void pmd_clear(pmd_t *pmdp)
878 {
879 	pmd_val(*pmdp) = _SEGMENT_ENTRY_EMPTY;
880 }
881 
882 static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
883 {
884 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
885 }
886 
887 /*
888  * The following pte modification functions only work if
889  * pte_present() is true. Undefined behaviour if not..
890  */
891 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
892 {
893 	pte_val(pte) &= _PAGE_CHG_MASK;
894 	pte_val(pte) |= pgprot_val(newprot);
895 	if ((pte_val(pte) & _PAGE_SWC) && (pte_val(pte) & _PAGE_SWW))
896 		pte_val(pte) &= ~_PAGE_RO;
897 	return pte;
898 }
899 
900 static inline pte_t pte_wrprotect(pte_t pte)
901 {
902 	pte_val(pte) &= ~_PAGE_SWW;
903 	/* Do not clobber _PAGE_TYPE_NONE pages!  */
904 	if (!(pte_val(pte) & _PAGE_INVALID))
905 		pte_val(pte) |= _PAGE_RO;
906 	return pte;
907 }
908 
909 static inline pte_t pte_mkwrite(pte_t pte)
910 {
911 	pte_val(pte) |= _PAGE_SWW;
912 	if (pte_val(pte) & _PAGE_SWC)
913 		pte_val(pte) &= ~_PAGE_RO;
914 	return pte;
915 }
916 
917 static inline pte_t pte_mkclean(pte_t pte)
918 {
919 	pte_val(pte) &= ~_PAGE_SWC;
920 	/* Do not clobber _PAGE_TYPE_NONE pages!  */
921 	if (!(pte_val(pte) & _PAGE_INVALID))
922 		pte_val(pte) |= _PAGE_RO;
923 	return pte;
924 }
925 
926 static inline pte_t pte_mkdirty(pte_t pte)
927 {
928 	pte_val(pte) |= _PAGE_SWC;
929 	if (pte_val(pte) & _PAGE_SWW)
930 		pte_val(pte) &= ~_PAGE_RO;
931 	return pte;
932 }
933 
934 static inline pte_t pte_mkold(pte_t pte)
935 {
936 #ifdef CONFIG_PGSTE
937 	pte_val(pte) &= ~_PAGE_SWR;
938 #endif
939 	return pte;
940 }
941 
942 static inline pte_t pte_mkyoung(pte_t pte)
943 {
944 	return pte;
945 }
946 
947 static inline pte_t pte_mkspecial(pte_t pte)
948 {
949 	pte_val(pte) |= _PAGE_SPECIAL;
950 	return pte;
951 }
952 
953 #ifdef CONFIG_HUGETLB_PAGE
954 static inline pte_t pte_mkhuge(pte_t pte)
955 {
956 	pte_val(pte) |= (_SEGMENT_ENTRY_LARGE | _SEGMENT_ENTRY_CO);
957 	return pte;
958 }
959 #endif
960 
961 /*
962  * Get (and clear) the user dirty bit for a pte.
963  */
964 static inline int ptep_test_and_clear_user_dirty(struct mm_struct *mm,
965 						 pte_t *ptep)
966 {
967 	pgste_t pgste;
968 	int dirty = 0;
969 
970 	if (mm_has_pgste(mm)) {
971 		pgste = pgste_get_lock(ptep);
972 		pgste = pgste_update_all(ptep, pgste);
973 		dirty = !!(pgste_val(pgste) & KVM_UC_BIT);
974 		pgste_val(pgste) &= ~KVM_UC_BIT;
975 		pgste_set_unlock(ptep, pgste);
976 		return dirty;
977 	}
978 	return dirty;
979 }
980 
981 /*
982  * Get (and clear) the user referenced bit for a pte.
983  */
984 static inline int ptep_test_and_clear_user_young(struct mm_struct *mm,
985 						 pte_t *ptep)
986 {
987 	pgste_t pgste;
988 	int young = 0;
989 
990 	if (mm_has_pgste(mm)) {
991 		pgste = pgste_get_lock(ptep);
992 		pgste = pgste_update_young(ptep, pgste);
993 		young = !!(pgste_val(pgste) & KVM_UR_BIT);
994 		pgste_val(pgste) &= ~KVM_UR_BIT;
995 		pgste_set_unlock(ptep, pgste);
996 	}
997 	return young;
998 }
999 
1000 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1001 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
1002 					    unsigned long addr, pte_t *ptep)
1003 {
1004 	pgste_t pgste;
1005 	pte_t pte;
1006 
1007 	if (mm_has_pgste(vma->vm_mm)) {
1008 		pgste = pgste_get_lock(ptep);
1009 		pgste = pgste_update_young(ptep, pgste);
1010 		pte = *ptep;
1011 		*ptep = pte_mkold(pte);
1012 		pgste_set_unlock(ptep, pgste);
1013 		return pte_young(pte);
1014 	}
1015 	return 0;
1016 }
1017 
1018 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1019 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
1020 					 unsigned long address, pte_t *ptep)
1021 {
1022 	/* No need to flush TLB
1023 	 * On s390 reference bits are in storage key and never in TLB
1024 	 * With virtualization we handle the reference bit, without we
1025 	 * we can simply return */
1026 	return ptep_test_and_clear_young(vma, address, ptep);
1027 }
1028 
1029 static inline void __ptep_ipte(unsigned long address, pte_t *ptep)
1030 {
1031 	if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1032 #ifndef CONFIG_64BIT
1033 		/* pto must point to the start of the segment table */
1034 		pte_t *pto = (pte_t *) (((unsigned long) ptep) & 0x7ffffc00);
1035 #else
1036 		/* ipte in zarch mode can do the math */
1037 		pte_t *pto = ptep;
1038 #endif
1039 		asm volatile(
1040 			"	ipte	%2,%3"
1041 			: "=m" (*ptep) : "m" (*ptep),
1042 			  "a" (pto), "a" (address));
1043 	}
1044 }
1045 
1046 /*
1047  * This is hard to understand. ptep_get_and_clear and ptep_clear_flush
1048  * both clear the TLB for the unmapped pte. The reason is that
1049  * ptep_get_and_clear is used in common code (e.g. change_pte_range)
1050  * to modify an active pte. The sequence is
1051  *   1) ptep_get_and_clear
1052  *   2) set_pte_at
1053  *   3) flush_tlb_range
1054  * On s390 the tlb needs to get flushed with the modification of the pte
1055  * if the pte is active. The only way how this can be implemented is to
1056  * have ptep_get_and_clear do the tlb flush. In exchange flush_tlb_range
1057  * is a nop.
1058  */
1059 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1060 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
1061 				       unsigned long address, pte_t *ptep)
1062 {
1063 	pgste_t pgste;
1064 	pte_t pte;
1065 
1066 	mm->context.flush_mm = 1;
1067 	if (mm_has_pgste(mm)) {
1068 		pgste = pgste_get_lock(ptep);
1069 		pgste = pgste_ipte_notify(mm, address, ptep, pgste);
1070 	}
1071 
1072 	pte = *ptep;
1073 	if (!mm_exclusive(mm))
1074 		__ptep_ipte(address, ptep);
1075 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
1076 
1077 	if (mm_has_pgste(mm)) {
1078 		pgste = pgste_update_all(&pte, pgste);
1079 		pgste_set_unlock(ptep, pgste);
1080 	}
1081 	return pte;
1082 }
1083 
1084 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1085 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
1086 					   unsigned long address,
1087 					   pte_t *ptep)
1088 {
1089 	pgste_t pgste;
1090 	pte_t pte;
1091 
1092 	mm->context.flush_mm = 1;
1093 	if (mm_has_pgste(mm)) {
1094 		pgste = pgste_get_lock(ptep);
1095 		pgste_ipte_notify(mm, address, ptep, pgste);
1096 	}
1097 
1098 	pte = *ptep;
1099 	if (!mm_exclusive(mm))
1100 		__ptep_ipte(address, ptep);
1101 
1102 	if (mm_has_pgste(mm))
1103 		pgste = pgste_update_all(&pte, pgste);
1104 	return pte;
1105 }
1106 
1107 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
1108 					   unsigned long address,
1109 					   pte_t *ptep, pte_t pte)
1110 {
1111 	pgste_t pgste;
1112 
1113 	if (mm_has_pgste(mm)) {
1114 		pgste = *(pgste_t *)(ptep + PTRS_PER_PTE);
1115 		pgste_set_key(ptep, pgste, pte);
1116 		pgste_set_pte(ptep, pte);
1117 		pgste_set_unlock(ptep, pgste);
1118 	} else
1119 		*ptep = pte;
1120 }
1121 
1122 #define __HAVE_ARCH_PTEP_CLEAR_FLUSH
1123 static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
1124 				     unsigned long address, pte_t *ptep)
1125 {
1126 	pgste_t pgste;
1127 	pte_t pte;
1128 
1129 	if (mm_has_pgste(vma->vm_mm)) {
1130 		pgste = pgste_get_lock(ptep);
1131 		pgste = pgste_ipte_notify(vma->vm_mm, address, ptep, pgste);
1132 	}
1133 
1134 	pte = *ptep;
1135 	__ptep_ipte(address, ptep);
1136 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
1137 
1138 	if (mm_has_pgste(vma->vm_mm)) {
1139 		pgste = pgste_update_all(&pte, pgste);
1140 		pgste_set_unlock(ptep, pgste);
1141 	}
1142 	return pte;
1143 }
1144 
1145 /*
1146  * The batched pte unmap code uses ptep_get_and_clear_full to clear the
1147  * ptes. Here an optimization is possible. tlb_gather_mmu flushes all
1148  * tlbs of an mm if it can guarantee that the ptes of the mm_struct
1149  * cannot be accessed while the batched unmap is running. In this case
1150  * full==1 and a simple pte_clear is enough. See tlb.h.
1151  */
1152 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
1153 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
1154 					    unsigned long address,
1155 					    pte_t *ptep, int full)
1156 {
1157 	pgste_t pgste;
1158 	pte_t pte;
1159 
1160 	if (mm_has_pgste(mm)) {
1161 		pgste = pgste_get_lock(ptep);
1162 		if (!full)
1163 			pgste = pgste_ipte_notify(mm, address, ptep, pgste);
1164 	}
1165 
1166 	pte = *ptep;
1167 	if (!full)
1168 		__ptep_ipte(address, ptep);
1169 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
1170 
1171 	if (mm_has_pgste(mm)) {
1172 		pgste = pgste_update_all(&pte, pgste);
1173 		pgste_set_unlock(ptep, pgste);
1174 	}
1175 	return pte;
1176 }
1177 
1178 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1179 static inline pte_t ptep_set_wrprotect(struct mm_struct *mm,
1180 				       unsigned long address, pte_t *ptep)
1181 {
1182 	pgste_t pgste;
1183 	pte_t pte = *ptep;
1184 
1185 	if (pte_write(pte)) {
1186 		mm->context.flush_mm = 1;
1187 		if (mm_has_pgste(mm)) {
1188 			pgste = pgste_get_lock(ptep);
1189 			pgste = pgste_ipte_notify(mm, address, ptep, pgste);
1190 		}
1191 
1192 		if (!mm_exclusive(mm))
1193 			__ptep_ipte(address, ptep);
1194 		pte = pte_wrprotect(pte);
1195 
1196 		if (mm_has_pgste(mm)) {
1197 			pgste_set_pte(ptep, pte);
1198 			pgste_set_unlock(ptep, pgste);
1199 		} else
1200 			*ptep = pte;
1201 	}
1202 	return pte;
1203 }
1204 
1205 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1206 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
1207 					unsigned long address, pte_t *ptep,
1208 					pte_t entry, int dirty)
1209 {
1210 	pgste_t pgste;
1211 
1212 	if (pte_same(*ptep, entry))
1213 		return 0;
1214 	if (mm_has_pgste(vma->vm_mm)) {
1215 		pgste = pgste_get_lock(ptep);
1216 		pgste = pgste_ipte_notify(vma->vm_mm, address, ptep, pgste);
1217 	}
1218 
1219 	__ptep_ipte(address, ptep);
1220 
1221 	if (mm_has_pgste(vma->vm_mm)) {
1222 		pgste_set_pte(ptep, entry);
1223 		pgste_set_unlock(ptep, pgste);
1224 	} else
1225 		*ptep = entry;
1226 	return 1;
1227 }
1228 
1229 /*
1230  * Conversion functions: convert a page and protection to a page entry,
1231  * and a page entry and page directory to the page they refer to.
1232  */
1233 static inline pte_t mk_pte_phys(unsigned long physpage, pgprot_t pgprot)
1234 {
1235 	pte_t __pte;
1236 	pte_val(__pte) = physpage + pgprot_val(pgprot);
1237 	return __pte;
1238 }
1239 
1240 static inline pte_t mk_pte(struct page *page, pgprot_t pgprot)
1241 {
1242 	unsigned long physpage = page_to_phys(page);
1243 	pte_t __pte = mk_pte_phys(physpage, pgprot);
1244 
1245 	if ((pte_val(__pte) & _PAGE_SWW) && PageDirty(page)) {
1246 		pte_val(__pte) |= _PAGE_SWC;
1247 		pte_val(__pte) &= ~_PAGE_RO;
1248 	}
1249 	return __pte;
1250 }
1251 
1252 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD-1))
1253 #define pud_index(address) (((address) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
1254 #define pmd_index(address) (((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1))
1255 #define pte_index(address) (((address) >> PAGE_SHIFT) & (PTRS_PER_PTE-1))
1256 
1257 #define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address))
1258 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
1259 
1260 #ifndef CONFIG_64BIT
1261 
1262 #define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
1263 #define pud_deref(pmd) ({ BUG(); 0UL; })
1264 #define pgd_deref(pmd) ({ BUG(); 0UL; })
1265 
1266 #define pud_offset(pgd, address) ((pud_t *) pgd)
1267 #define pmd_offset(pud, address) ((pmd_t *) pud + pmd_index(address))
1268 
1269 #else /* CONFIG_64BIT */
1270 
1271 #define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
1272 #define pud_deref(pud) (pud_val(pud) & _REGION_ENTRY_ORIGIN)
1273 #define pgd_deref(pgd) (pgd_val(pgd) & _REGION_ENTRY_ORIGIN)
1274 
1275 static inline pud_t *pud_offset(pgd_t *pgd, unsigned long address)
1276 {
1277 	pud_t *pud = (pud_t *) pgd;
1278 	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
1279 		pud = (pud_t *) pgd_deref(*pgd);
1280 	return pud  + pud_index(address);
1281 }
1282 
1283 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
1284 {
1285 	pmd_t *pmd = (pmd_t *) pud;
1286 	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
1287 		pmd = (pmd_t *) pud_deref(*pud);
1288 	return pmd + pmd_index(address);
1289 }
1290 
1291 #endif /* CONFIG_64BIT */
1292 
1293 #define pfn_pte(pfn,pgprot) mk_pte_phys(__pa((pfn) << PAGE_SHIFT),(pgprot))
1294 #define pte_pfn(x) (pte_val(x) >> PAGE_SHIFT)
1295 #define pte_page(x) pfn_to_page(pte_pfn(x))
1296 
1297 #define pmd_page(pmd) pfn_to_page(pmd_val(pmd) >> PAGE_SHIFT)
1298 
1299 /* Find an entry in the lowest level page table.. */
1300 #define pte_offset(pmd, addr) ((pte_t *) pmd_deref(*(pmd)) + pte_index(addr))
1301 #define pte_offset_kernel(pmd, address) pte_offset(pmd,address)
1302 #define pte_offset_map(pmd, address) pte_offset_kernel(pmd, address)
1303 #define pte_unmap(pte) do { } while (0)
1304 
1305 static inline void __pmd_idte(unsigned long address, pmd_t *pmdp)
1306 {
1307 	unsigned long sto = (unsigned long) pmdp -
1308 			    pmd_index(address) * sizeof(pmd_t);
1309 
1310 	if (!(pmd_val(*pmdp) & _SEGMENT_ENTRY_INV)) {
1311 		asm volatile(
1312 			"	.insn	rrf,0xb98e0000,%2,%3,0,0"
1313 			: "=m" (*pmdp)
1314 			: "m" (*pmdp), "a" (sto),
1315 			  "a" ((address & HPAGE_MASK))
1316 			: "cc"
1317 		);
1318 	}
1319 }
1320 
1321 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE)
1322 static inline unsigned long massage_pgprot_pmd(pgprot_t pgprot)
1323 {
1324 	/*
1325 	 * pgprot is PAGE_NONE, PAGE_RO, or PAGE_RW (see __Pxxx / __Sxxx)
1326 	 * Convert to segment table entry format.
1327 	 */
1328 	if (pgprot_val(pgprot) == pgprot_val(PAGE_NONE))
1329 		return pgprot_val(SEGMENT_NONE);
1330 	if (pgprot_val(pgprot) == pgprot_val(PAGE_RO))
1331 		return pgprot_val(SEGMENT_RO);
1332 	return pgprot_val(SEGMENT_RW);
1333 }
1334 
1335 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
1336 {
1337 	pmd_val(pmd) &= _SEGMENT_CHG_MASK;
1338 	pmd_val(pmd) |= massage_pgprot_pmd(newprot);
1339 	return pmd;
1340 }
1341 
1342 static inline pmd_t mk_pmd_phys(unsigned long physpage, pgprot_t pgprot)
1343 {
1344 	pmd_t __pmd;
1345 	pmd_val(__pmd) = physpage + massage_pgprot_pmd(pgprot);
1346 	return __pmd;
1347 }
1348 
1349 static inline pmd_t pmd_mkwrite(pmd_t pmd)
1350 {
1351 	/* Do not clobber _HPAGE_TYPE_NONE pages! */
1352 	if (!(pmd_val(pmd) & _SEGMENT_ENTRY_INV))
1353 		pmd_val(pmd) &= ~_SEGMENT_ENTRY_RO;
1354 	return pmd;
1355 }
1356 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLB_PAGE */
1357 
1358 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1359 
1360 #define __HAVE_ARCH_PGTABLE_DEPOSIT
1361 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pgtable_t pgtable);
1362 
1363 #define __HAVE_ARCH_PGTABLE_WITHDRAW
1364 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm);
1365 
1366 static inline int pmd_trans_splitting(pmd_t pmd)
1367 {
1368 	return pmd_val(pmd) & _SEGMENT_ENTRY_SPLIT;
1369 }
1370 
1371 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1372 			      pmd_t *pmdp, pmd_t entry)
1373 {
1374 	if (!(pmd_val(entry) & _SEGMENT_ENTRY_INV) && MACHINE_HAS_EDAT1)
1375 		pmd_val(entry) |= _SEGMENT_ENTRY_CO;
1376 	*pmdp = entry;
1377 }
1378 
1379 static inline pmd_t pmd_mkhuge(pmd_t pmd)
1380 {
1381 	pmd_val(pmd) |= _SEGMENT_ENTRY_LARGE;
1382 	return pmd;
1383 }
1384 
1385 static inline pmd_t pmd_wrprotect(pmd_t pmd)
1386 {
1387 	pmd_val(pmd) |= _SEGMENT_ENTRY_RO;
1388 	return pmd;
1389 }
1390 
1391 static inline pmd_t pmd_mkdirty(pmd_t pmd)
1392 {
1393 	/* No dirty bit in the segment table entry. */
1394 	return pmd;
1395 }
1396 
1397 static inline pmd_t pmd_mkold(pmd_t pmd)
1398 {
1399 	/* No referenced bit in the segment table entry. */
1400 	return pmd;
1401 }
1402 
1403 static inline pmd_t pmd_mkyoung(pmd_t pmd)
1404 {
1405 	/* No referenced bit in the segment table entry. */
1406 	return pmd;
1407 }
1408 
1409 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1410 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1411 					    unsigned long address, pmd_t *pmdp)
1412 {
1413 	unsigned long pmd_addr = pmd_val(*pmdp) & HPAGE_MASK;
1414 	long tmp, rc;
1415 	int counter;
1416 
1417 	rc = 0;
1418 	if (MACHINE_HAS_RRBM) {
1419 		counter = PTRS_PER_PTE >> 6;
1420 		asm volatile(
1421 			"0:	.insn	rre,0xb9ae0000,%0,%3\n"	/* rrbm */
1422 			"	ogr	%1,%0\n"
1423 			"	la	%3,0(%4,%3)\n"
1424 			"	brct	%2,0b\n"
1425 			: "=&d" (tmp), "+&d" (rc), "+d" (counter),
1426 			  "+a" (pmd_addr)
1427 			: "a" (64 * 4096UL) : "cc");
1428 		rc = !!rc;
1429 	} else {
1430 		counter = PTRS_PER_PTE;
1431 		asm volatile(
1432 			"0:	rrbe	0,%2\n"
1433 			"	la	%2,0(%3,%2)\n"
1434 			"	brc	12,1f\n"
1435 			"	lhi	%0,1\n"
1436 			"1:	brct	%1,0b\n"
1437 			: "+d" (rc), "+d" (counter), "+a" (pmd_addr)
1438 			: "a" (4096UL) : "cc");
1439 	}
1440 	return rc;
1441 }
1442 
1443 #define __HAVE_ARCH_PMDP_GET_AND_CLEAR
1444 static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
1445 				       unsigned long address, pmd_t *pmdp)
1446 {
1447 	pmd_t pmd = *pmdp;
1448 
1449 	__pmd_idte(address, pmdp);
1450 	pmd_clear(pmdp);
1451 	return pmd;
1452 }
1453 
1454 #define __HAVE_ARCH_PMDP_CLEAR_FLUSH
1455 static inline pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
1456 				     unsigned long address, pmd_t *pmdp)
1457 {
1458 	return pmdp_get_and_clear(vma->vm_mm, address, pmdp);
1459 }
1460 
1461 #define __HAVE_ARCH_PMDP_INVALIDATE
1462 static inline void pmdp_invalidate(struct vm_area_struct *vma,
1463 				   unsigned long address, pmd_t *pmdp)
1464 {
1465 	__pmd_idte(address, pmdp);
1466 }
1467 
1468 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1469 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1470 				      unsigned long address, pmd_t *pmdp)
1471 {
1472 	pmd_t pmd = *pmdp;
1473 
1474 	if (pmd_write(pmd)) {
1475 		__pmd_idte(address, pmdp);
1476 		set_pmd_at(mm, address, pmdp, pmd_wrprotect(pmd));
1477 	}
1478 }
1479 
1480 #define pfn_pmd(pfn, pgprot)	mk_pmd_phys(__pa((pfn) << PAGE_SHIFT), (pgprot))
1481 #define mk_pmd(page, pgprot)	pfn_pmd(page_to_pfn(page), (pgprot))
1482 
1483 static inline int pmd_trans_huge(pmd_t pmd)
1484 {
1485 	return pmd_val(pmd) & _SEGMENT_ENTRY_LARGE;
1486 }
1487 
1488 static inline int has_transparent_hugepage(void)
1489 {
1490 	return MACHINE_HAS_HPAGE ? 1 : 0;
1491 }
1492 
1493 static inline unsigned long pmd_pfn(pmd_t pmd)
1494 {
1495 	return pmd_val(pmd) >> PAGE_SHIFT;
1496 }
1497 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1498 
1499 /*
1500  * 31 bit swap entry format:
1501  * A page-table entry has some bits we have to treat in a special way.
1502  * Bits 0, 20 and bit 23 have to be zero, otherwise an specification
1503  * exception will occur instead of a page translation exception. The
1504  * specifiation exception has the bad habit not to store necessary
1505  * information in the lowcore.
1506  * Bit 21 and bit 22 are the page invalid bit and the page protection
1507  * bit. We set both to indicate a swapped page.
1508  * Bit 30 and 31 are used to distinguish the different page types. For
1509  * a swapped page these bits need to be zero.
1510  * This leaves the bits 1-19 and bits 24-29 to store type and offset.
1511  * We use the 5 bits from 25-29 for the type and the 20 bits from 1-19
1512  * plus 24 for the offset.
1513  * 0|     offset        |0110|o|type |00|
1514  * 0 0000000001111111111 2222 2 22222 33
1515  * 0 1234567890123456789 0123 4 56789 01
1516  *
1517  * 64 bit swap entry format:
1518  * A page-table entry has some bits we have to treat in a special way.
1519  * Bits 52 and bit 55 have to be zero, otherwise an specification
1520  * exception will occur instead of a page translation exception. The
1521  * specifiation exception has the bad habit not to store necessary
1522  * information in the lowcore.
1523  * Bit 53 and bit 54 are the page invalid bit and the page protection
1524  * bit. We set both to indicate a swapped page.
1525  * Bit 62 and 63 are used to distinguish the different page types. For
1526  * a swapped page these bits need to be zero.
1527  * This leaves the bits 0-51 and bits 56-61 to store type and offset.
1528  * We use the 5 bits from 57-61 for the type and the 53 bits from 0-51
1529  * plus 56 for the offset.
1530  * |                      offset                        |0110|o|type |00|
1531  *  0000000000111111111122222222223333333333444444444455 5555 5 55566 66
1532  *  0123456789012345678901234567890123456789012345678901 2345 6 78901 23
1533  */
1534 #ifndef CONFIG_64BIT
1535 #define __SWP_OFFSET_MASK (~0UL >> 12)
1536 #else
1537 #define __SWP_OFFSET_MASK (~0UL >> 11)
1538 #endif
1539 static inline pte_t mk_swap_pte(unsigned long type, unsigned long offset)
1540 {
1541 	pte_t pte;
1542 	offset &= __SWP_OFFSET_MASK;
1543 	pte_val(pte) = _PAGE_TYPE_SWAP | ((type & 0x1f) << 2) |
1544 		((offset & 1UL) << 7) | ((offset & ~1UL) << 11);
1545 	return pte;
1546 }
1547 
1548 #define __swp_type(entry)	(((entry).val >> 2) & 0x1f)
1549 #define __swp_offset(entry)	(((entry).val >> 11) | (((entry).val >> 7) & 1))
1550 #define __swp_entry(type,offset) ((swp_entry_t) { pte_val(mk_swap_pte((type),(offset))) })
1551 
1552 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
1553 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
1554 
1555 #ifndef CONFIG_64BIT
1556 # define PTE_FILE_MAX_BITS	26
1557 #else /* CONFIG_64BIT */
1558 # define PTE_FILE_MAX_BITS	59
1559 #endif /* CONFIG_64BIT */
1560 
1561 #define pte_to_pgoff(__pte) \
1562 	((((__pte).pte >> 12) << 7) + (((__pte).pte >> 1) & 0x7f))
1563 
1564 #define pgoff_to_pte(__off) \
1565 	((pte_t) { ((((__off) & 0x7f) << 1) + (((__off) >> 7) << 12)) \
1566 		   | _PAGE_TYPE_FILE })
1567 
1568 #endif /* !__ASSEMBLY__ */
1569 
1570 #define kern_addr_valid(addr)   (1)
1571 
1572 extern int vmem_add_mapping(unsigned long start, unsigned long size);
1573 extern int vmem_remove_mapping(unsigned long start, unsigned long size);
1574 extern int s390_enable_sie(void);
1575 
1576 /*
1577  * No page table caches to initialise
1578  */
1579 static inline void pgtable_cache_init(void) { }
1580 static inline void check_pgt_cache(void) { }
1581 
1582 #include <asm-generic/pgtable.h>
1583 
1584 #endif /* _S390_PAGE_H */
1585