xref: /openbmc/linux/arch/s390/include/asm/pgtable.h (revision df3305156f989339529b3d6744b898d498fb1f7b)
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 <linux/radix-tree.h>
34 #include <asm/bug.h>
35 #include <asm/page.h>
36 
37 extern pgd_t swapper_pg_dir[] __attribute__ ((aligned (4096)));
38 extern void paging_init(void);
39 extern void vmem_map_init(void);
40 
41 /*
42  * The S390 doesn't have any external MMU info: the kernel page
43  * tables contain all the necessary information.
44  */
45 #define update_mmu_cache(vma, address, ptep)     do { } while (0)
46 #define update_mmu_cache_pmd(vma, address, ptep) do { } while (0)
47 
48 /*
49  * ZERO_PAGE is a global shared page that is always zero; used
50  * for zero-mapped memory areas etc..
51  */
52 
53 extern unsigned long empty_zero_page;
54 extern unsigned long zero_page_mask;
55 
56 #define ZERO_PAGE(vaddr) \
57 	(virt_to_page((void *)(empty_zero_page + \
58 	 (((unsigned long)(vaddr)) &zero_page_mask))))
59 #define __HAVE_COLOR_ZERO_PAGE
60 
61 /* TODO: s390 cannot support io_remap_pfn_range... */
62 #endif /* !__ASSEMBLY__ */
63 
64 /*
65  * PMD_SHIFT determines the size of the area a second-level page
66  * table can map
67  * PGDIR_SHIFT determines what a third-level page table entry can map
68  */
69 #ifndef CONFIG_64BIT
70 # define PMD_SHIFT	20
71 # define PUD_SHIFT	20
72 # define PGDIR_SHIFT	20
73 #else /* CONFIG_64BIT */
74 # define PMD_SHIFT	20
75 # define PUD_SHIFT	31
76 # define PGDIR_SHIFT	42
77 #endif /* CONFIG_64BIT */
78 
79 #define PMD_SIZE        (1UL << PMD_SHIFT)
80 #define PMD_MASK        (~(PMD_SIZE-1))
81 #define PUD_SIZE	(1UL << PUD_SHIFT)
82 #define PUD_MASK	(~(PUD_SIZE-1))
83 #define PGDIR_SIZE	(1UL << PGDIR_SHIFT)
84 #define PGDIR_MASK	(~(PGDIR_SIZE-1))
85 
86 /*
87  * entries per page directory level: the S390 is two-level, so
88  * we don't really have any PMD directory physically.
89  * for S390 segment-table entries are combined to one PGD
90  * that leads to 1024 pte per pgd
91  */
92 #define PTRS_PER_PTE	256
93 #ifndef CONFIG_64BIT
94 #define PTRS_PER_PMD	1
95 #define PTRS_PER_PUD	1
96 #else /* CONFIG_64BIT */
97 #define PTRS_PER_PMD	2048
98 #define PTRS_PER_PUD	2048
99 #endif /* CONFIG_64BIT */
100 #define PTRS_PER_PGD	2048
101 
102 #define FIRST_USER_ADDRESS  0UL
103 
104 #define pte_ERROR(e) \
105 	printk("%s:%d: bad pte %p.\n", __FILE__, __LINE__, (void *) pte_val(e))
106 #define pmd_ERROR(e) \
107 	printk("%s:%d: bad pmd %p.\n", __FILE__, __LINE__, (void *) pmd_val(e))
108 #define pud_ERROR(e) \
109 	printk("%s:%d: bad pud %p.\n", __FILE__, __LINE__, (void *) pud_val(e))
110 #define pgd_ERROR(e) \
111 	printk("%s:%d: bad pgd %p.\n", __FILE__, __LINE__, (void *) pgd_val(e))
112 
113 #ifndef __ASSEMBLY__
114 /*
115  * The vmalloc and module area will always be on the topmost area of the kernel
116  * mapping. We reserve 96MB (31bit) / 128GB (64bit) for vmalloc and modules.
117  * On 64 bit kernels we have a 2GB area at the top of the vmalloc area where
118  * modules will reside. That makes sure that inter module branches always
119  * happen without trampolines and in addition the placement within a 2GB frame
120  * is branch prediction unit friendly.
121  */
122 extern unsigned long VMALLOC_START;
123 extern unsigned long VMALLOC_END;
124 extern struct page *vmemmap;
125 
126 #define VMEM_MAX_PHYS ((unsigned long) vmemmap)
127 
128 #ifdef CONFIG_64BIT
129 extern unsigned long MODULES_VADDR;
130 extern unsigned long MODULES_END;
131 #define MODULES_VADDR	MODULES_VADDR
132 #define MODULES_END	MODULES_END
133 #define MODULES_LEN	(1UL << 31)
134 #endif
135 
136 static inline int is_module_addr(void *addr)
137 {
138 #ifdef CONFIG_64BIT
139 	BUILD_BUG_ON(MODULES_LEN > (1UL << 31));
140 	if (addr < (void *)MODULES_VADDR)
141 		return 0;
142 	if (addr > (void *)MODULES_END)
143 		return 0;
144 #endif
145 	return 1;
146 }
147 
148 /*
149  * A 31 bit pagetable entry of S390 has following format:
150  *  |   PFRA          |    |  OS  |
151  * 0                   0IP0
152  * 00000000001111111111222222222233
153  * 01234567890123456789012345678901
154  *
155  * I Page-Invalid Bit:    Page is not available for address-translation
156  * P Page-Protection Bit: Store access not possible for page
157  *
158  * A 31 bit segmenttable entry of S390 has following format:
159  *  |   P-table origin      |  |PTL
160  * 0                         IC
161  * 00000000001111111111222222222233
162  * 01234567890123456789012345678901
163  *
164  * I Segment-Invalid Bit:    Segment is not available for address-translation
165  * C Common-Segment Bit:     Segment is not private (PoP 3-30)
166  * PTL Page-Table-Length:    Page-table length (PTL+1*16 entries -> up to 256)
167  *
168  * The 31 bit segmenttable origin of S390 has following format:
169  *
170  *  |S-table origin   |     | STL |
171  * X                   **GPS
172  * 00000000001111111111222222222233
173  * 01234567890123456789012345678901
174  *
175  * X Space-Switch event:
176  * G Segment-Invalid Bit:     *
177  * P Private-Space Bit:       Segment is not private (PoP 3-30)
178  * S Storage-Alteration:
179  * STL Segment-Table-Length:  Segment-table length (STL+1*16 entries -> up to 2048)
180  *
181  * A 64 bit pagetable entry of S390 has following format:
182  * |			 PFRA			      |0IPC|  OS  |
183  * 0000000000111111111122222222223333333333444444444455555555556666
184  * 0123456789012345678901234567890123456789012345678901234567890123
185  *
186  * I Page-Invalid Bit:    Page is not available for address-translation
187  * P Page-Protection Bit: Store access not possible for page
188  * C Change-bit override: HW is not required to set change bit
189  *
190  * A 64 bit segmenttable entry of S390 has following format:
191  * |        P-table origin                              |      TT
192  * 0000000000111111111122222222223333333333444444444455555555556666
193  * 0123456789012345678901234567890123456789012345678901234567890123
194  *
195  * I Segment-Invalid Bit:    Segment is not available for address-translation
196  * C Common-Segment Bit:     Segment is not private (PoP 3-30)
197  * P Page-Protection Bit: Store access not possible for page
198  * TT Type 00
199  *
200  * A 64 bit region table entry of S390 has following format:
201  * |        S-table origin                             |   TF  TTTL
202  * 0000000000111111111122222222223333333333444444444455555555556666
203  * 0123456789012345678901234567890123456789012345678901234567890123
204  *
205  * I Segment-Invalid Bit:    Segment is not available for address-translation
206  * TT Type 01
207  * TF
208  * TL Table length
209  *
210  * The 64 bit regiontable origin of S390 has following format:
211  * |      region table origon                          |       DTTL
212  * 0000000000111111111122222222223333333333444444444455555555556666
213  * 0123456789012345678901234567890123456789012345678901234567890123
214  *
215  * X Space-Switch event:
216  * G Segment-Invalid Bit:
217  * P Private-Space Bit:
218  * S Storage-Alteration:
219  * R Real space
220  * TL Table-Length:
221  *
222  * A storage key has the following format:
223  * | ACC |F|R|C|0|
224  *  0   3 4 5 6 7
225  * ACC: access key
226  * F  : fetch protection bit
227  * R  : referenced bit
228  * C  : changed bit
229  */
230 
231 /* Hardware bits in the page table entry */
232 #define _PAGE_PROTECT	0x200		/* HW read-only bit  */
233 #define _PAGE_INVALID	0x400		/* HW invalid bit    */
234 #define _PAGE_LARGE	0x800		/* Bit to mark a large pte */
235 
236 /* Software bits in the page table entry */
237 #define _PAGE_PRESENT	0x001		/* SW pte present bit */
238 #define _PAGE_TYPE	0x002		/* SW pte type bit */
239 #define _PAGE_YOUNG	0x004		/* SW pte young bit */
240 #define _PAGE_DIRTY	0x008		/* SW pte dirty bit */
241 #define _PAGE_READ	0x010		/* SW pte read bit */
242 #define _PAGE_WRITE	0x020		/* SW pte write bit */
243 #define _PAGE_SPECIAL	0x040		/* SW associated with special page */
244 #define _PAGE_UNUSED	0x080		/* SW bit for pgste usage state */
245 #define __HAVE_ARCH_PTE_SPECIAL
246 
247 /* Set of bits not changed in pte_modify */
248 #define _PAGE_CHG_MASK		(PAGE_MASK | _PAGE_SPECIAL | _PAGE_DIRTY | \
249 				 _PAGE_YOUNG)
250 
251 /*
252  * handle_pte_fault uses pte_present and pte_none to find out the pte type
253  * WITHOUT holding the page table lock. The _PAGE_PRESENT bit is used to
254  * distinguish present from not-present ptes. It is changed only with the page
255  * table lock held.
256  *
257  * The following table gives the different possible bit combinations for
258  * the pte hardware and software bits in the last 12 bits of a pte:
259  *
260  *				842100000000
261  *				000084210000
262  *				000000008421
263  *				.IR...wrdytp
264  * empty			.10...000000
265  * swap				.10...xxxx10
266  * file				.11...xxxxx0
267  * prot-none, clean, old	.11...000001
268  * prot-none, clean, young	.11...000101
269  * prot-none, dirty, old	.10...001001
270  * prot-none, dirty, young	.10...001101
271  * read-only, clean, old	.11...010001
272  * read-only, clean, young	.01...010101
273  * read-only, dirty, old	.11...011001
274  * read-only, dirty, young	.01...011101
275  * read-write, clean, old	.11...110001
276  * read-write, clean, young	.01...110101
277  * read-write, dirty, old	.10...111001
278  * read-write, dirty, young	.00...111101
279  *
280  * pte_present is true for the bit pattern .xx...xxxxx1, (pte & 0x001) == 0x001
281  * pte_none    is true for the bit pattern .10...xxxx00, (pte & 0x603) == 0x400
282  * pte_swap    is true for the bit pattern .10...xxxx10, (pte & 0x603) == 0x402
283  */
284 
285 #ifndef CONFIG_64BIT
286 
287 /* Bits in the segment table address-space-control-element */
288 #define _ASCE_SPACE_SWITCH	0x80000000UL	/* space switch event	    */
289 #define _ASCE_ORIGIN_MASK	0x7ffff000UL	/* segment table origin	    */
290 #define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
291 #define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
292 #define _ASCE_TABLE_LENGTH	0x7f	/* 128 x 64 entries = 8k	    */
293 
294 /* Bits in the segment table entry */
295 #define _SEGMENT_ENTRY_BITS	0x7fffffffUL	/* Valid segment table bits */
296 #define _SEGMENT_ENTRY_ORIGIN	0x7fffffc0UL	/* page table origin	    */
297 #define _SEGMENT_ENTRY_PROTECT	0x200	/* page protection bit		    */
298 #define _SEGMENT_ENTRY_INVALID	0x20	/* invalid segment table entry	    */
299 #define _SEGMENT_ENTRY_COMMON	0x10	/* common segment bit		    */
300 #define _SEGMENT_ENTRY_PTL	0x0f	/* page table length		    */
301 
302 #define _SEGMENT_ENTRY_DIRTY	0	/* No sw dirty bit for 31-bit */
303 #define _SEGMENT_ENTRY_YOUNG	0	/* No sw young bit for 31-bit */
304 #define _SEGMENT_ENTRY_READ	0	/* No sw read bit for 31-bit */
305 #define _SEGMENT_ENTRY_WRITE	0	/* No sw write bit for 31-bit */
306 #define _SEGMENT_ENTRY_LARGE	0	/* No large pages for 31-bit */
307 #define _SEGMENT_ENTRY_BITS_LARGE 0
308 #define _SEGMENT_ENTRY_ORIGIN_LARGE 0
309 
310 #define _SEGMENT_ENTRY		(_SEGMENT_ENTRY_PTL)
311 #define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INVALID)
312 
313 /*
314  * Segment table entry encoding (I = invalid, R = read-only bit):
315  *		..R...I.....
316  * prot-none	..1...1.....
317  * read-only	..1...0.....
318  * read-write	..0...0.....
319  * empty	..0...1.....
320  */
321 
322 /* Page status table bits for virtualization */
323 #define PGSTE_ACC_BITS	0xf0000000UL
324 #define PGSTE_FP_BIT	0x08000000UL
325 #define PGSTE_PCL_BIT	0x00800000UL
326 #define PGSTE_HR_BIT	0x00400000UL
327 #define PGSTE_HC_BIT	0x00200000UL
328 #define PGSTE_GR_BIT	0x00040000UL
329 #define PGSTE_GC_BIT	0x00020000UL
330 #define PGSTE_UC_BIT	0x00008000UL	/* user dirty (migration) */
331 #define PGSTE_IN_BIT	0x00004000UL	/* IPTE notify bit */
332 
333 #else /* CONFIG_64BIT */
334 
335 /* Bits in the segment/region table address-space-control-element */
336 #define _ASCE_ORIGIN		~0xfffUL/* segment table origin		    */
337 #define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
338 #define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
339 #define _ASCE_SPACE_SWITCH	0x40	/* space switch event		    */
340 #define _ASCE_REAL_SPACE	0x20	/* real space control		    */
341 #define _ASCE_TYPE_MASK		0x0c	/* asce table type mask		    */
342 #define _ASCE_TYPE_REGION1	0x0c	/* region first table type	    */
343 #define _ASCE_TYPE_REGION2	0x08	/* region second table type	    */
344 #define _ASCE_TYPE_REGION3	0x04	/* region third table type	    */
345 #define _ASCE_TYPE_SEGMENT	0x00	/* segment table type		    */
346 #define _ASCE_TABLE_LENGTH	0x03	/* region table length		    */
347 
348 /* Bits in the region table entry */
349 #define _REGION_ENTRY_ORIGIN	~0xfffUL/* region/segment table origin	    */
350 #define _REGION_ENTRY_PROTECT	0x200	/* region protection bit	    */
351 #define _REGION_ENTRY_INVALID	0x20	/* invalid region table entry	    */
352 #define _REGION_ENTRY_TYPE_MASK	0x0c	/* region/segment table type mask   */
353 #define _REGION_ENTRY_TYPE_R1	0x0c	/* region first table type	    */
354 #define _REGION_ENTRY_TYPE_R2	0x08	/* region second table type	    */
355 #define _REGION_ENTRY_TYPE_R3	0x04	/* region third table type	    */
356 #define _REGION_ENTRY_LENGTH	0x03	/* region third length		    */
357 
358 #define _REGION1_ENTRY		(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_LENGTH)
359 #define _REGION1_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID)
360 #define _REGION2_ENTRY		(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_LENGTH)
361 #define _REGION2_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID)
362 #define _REGION3_ENTRY		(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_LENGTH)
363 #define _REGION3_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID)
364 
365 #define _REGION3_ENTRY_LARGE	0x400	/* RTTE-format control, large page  */
366 #define _REGION3_ENTRY_RO	0x200	/* page protection bit		    */
367 
368 /* Bits in the segment table entry */
369 #define _SEGMENT_ENTRY_BITS	0xfffffffffffffe33UL
370 #define _SEGMENT_ENTRY_BITS_LARGE 0xfffffffffff0ff33UL
371 #define _SEGMENT_ENTRY_ORIGIN_LARGE ~0xfffffUL /* large page address	    */
372 #define _SEGMENT_ENTRY_ORIGIN	~0x7ffUL/* segment table origin		    */
373 #define _SEGMENT_ENTRY_PROTECT	0x200	/* page protection bit		    */
374 #define _SEGMENT_ENTRY_INVALID	0x20	/* invalid segment table entry	    */
375 
376 #define _SEGMENT_ENTRY		(0)
377 #define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INVALID)
378 
379 #define _SEGMENT_ENTRY_DIRTY	0x2000	/* SW segment dirty bit */
380 #define _SEGMENT_ENTRY_YOUNG	0x1000	/* SW segment young bit */
381 #define _SEGMENT_ENTRY_SPLIT	0x0800	/* THP splitting bit */
382 #define _SEGMENT_ENTRY_LARGE	0x0400	/* STE-format control, large page */
383 #define _SEGMENT_ENTRY_READ	0x0002	/* SW segment read bit */
384 #define _SEGMENT_ENTRY_WRITE	0x0001	/* SW segment write bit */
385 
386 /*
387  * Segment table entry encoding (R = read-only, I = invalid, y = young bit):
388  *				dy..R...I...wr
389  * prot-none, clean, old	00..1...1...00
390  * prot-none, clean, young	01..1...1...00
391  * prot-none, dirty, old	10..1...1...00
392  * prot-none, dirty, young	11..1...1...00
393  * read-only, clean, old	00..1...1...01
394  * read-only, clean, young	01..1...0...01
395  * read-only, dirty, old	10..1...1...01
396  * read-only, dirty, young	11..1...0...01
397  * read-write, clean, old	00..1...1...11
398  * read-write, clean, young	01..1...0...11
399  * read-write, dirty, old	10..0...1...11
400  * read-write, dirty, young	11..0...0...11
401  * The segment table origin is used to distinguish empty (origin==0) from
402  * read-write, old segment table entries (origin!=0)
403  */
404 
405 #define _SEGMENT_ENTRY_SPLIT_BIT 11	/* THP splitting bit number */
406 
407 /* Page status table bits for virtualization */
408 #define PGSTE_ACC_BITS	0xf000000000000000UL
409 #define PGSTE_FP_BIT	0x0800000000000000UL
410 #define PGSTE_PCL_BIT	0x0080000000000000UL
411 #define PGSTE_HR_BIT	0x0040000000000000UL
412 #define PGSTE_HC_BIT	0x0020000000000000UL
413 #define PGSTE_GR_BIT	0x0004000000000000UL
414 #define PGSTE_GC_BIT	0x0002000000000000UL
415 #define PGSTE_UC_BIT	0x0000800000000000UL	/* user dirty (migration) */
416 #define PGSTE_IN_BIT	0x0000400000000000UL	/* IPTE notify bit */
417 
418 #endif /* CONFIG_64BIT */
419 
420 /* Guest Page State used for virtualization */
421 #define _PGSTE_GPS_ZERO		0x0000000080000000UL
422 #define _PGSTE_GPS_USAGE_MASK	0x0000000003000000UL
423 #define _PGSTE_GPS_USAGE_STABLE 0x0000000000000000UL
424 #define _PGSTE_GPS_USAGE_UNUSED 0x0000000001000000UL
425 
426 /*
427  * A user page table pointer has the space-switch-event bit, the
428  * private-space-control bit and the storage-alteration-event-control
429  * bit set. A kernel page table pointer doesn't need them.
430  */
431 #define _ASCE_USER_BITS		(_ASCE_SPACE_SWITCH | _ASCE_PRIVATE_SPACE | \
432 				 _ASCE_ALT_EVENT)
433 
434 /*
435  * Page protection definitions.
436  */
437 #define PAGE_NONE	__pgprot(_PAGE_PRESENT | _PAGE_INVALID)
438 #define PAGE_READ	__pgprot(_PAGE_PRESENT | _PAGE_READ | \
439 				 _PAGE_INVALID | _PAGE_PROTECT)
440 #define PAGE_WRITE	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
441 				 _PAGE_INVALID | _PAGE_PROTECT)
442 
443 #define PAGE_SHARED	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
444 				 _PAGE_YOUNG | _PAGE_DIRTY)
445 #define PAGE_KERNEL	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
446 				 _PAGE_YOUNG | _PAGE_DIRTY)
447 #define PAGE_KERNEL_RO	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_YOUNG | \
448 				 _PAGE_PROTECT)
449 
450 /*
451  * On s390 the page table entry has an invalid bit and a read-only bit.
452  * Read permission implies execute permission and write permission
453  * implies read permission.
454  */
455          /*xwr*/
456 #define __P000	PAGE_NONE
457 #define __P001	PAGE_READ
458 #define __P010	PAGE_READ
459 #define __P011	PAGE_READ
460 #define __P100	PAGE_READ
461 #define __P101	PAGE_READ
462 #define __P110	PAGE_READ
463 #define __P111	PAGE_READ
464 
465 #define __S000	PAGE_NONE
466 #define __S001	PAGE_READ
467 #define __S010	PAGE_WRITE
468 #define __S011	PAGE_WRITE
469 #define __S100	PAGE_READ
470 #define __S101	PAGE_READ
471 #define __S110	PAGE_WRITE
472 #define __S111	PAGE_WRITE
473 
474 /*
475  * Segment entry (large page) protection definitions.
476  */
477 #define SEGMENT_NONE	__pgprot(_SEGMENT_ENTRY_INVALID | \
478 				 _SEGMENT_ENTRY_PROTECT)
479 #define SEGMENT_READ	__pgprot(_SEGMENT_ENTRY_PROTECT | \
480 				 _SEGMENT_ENTRY_READ)
481 #define SEGMENT_WRITE	__pgprot(_SEGMENT_ENTRY_READ | \
482 				 _SEGMENT_ENTRY_WRITE)
483 
484 static inline int mm_has_pgste(struct mm_struct *mm)
485 {
486 #ifdef CONFIG_PGSTE
487 	if (unlikely(mm->context.has_pgste))
488 		return 1;
489 #endif
490 	return 0;
491 }
492 
493 /*
494  * In the case that a guest uses storage keys
495  * faults should no longer be backed by zero pages
496  */
497 #define mm_forbids_zeropage mm_use_skey
498 static inline int mm_use_skey(struct mm_struct *mm)
499 {
500 #ifdef CONFIG_PGSTE
501 	if (mm->context.use_skey)
502 		return 1;
503 #endif
504 	return 0;
505 }
506 
507 /*
508  * pgd/pmd/pte query functions
509  */
510 #ifndef CONFIG_64BIT
511 
512 static inline int pgd_present(pgd_t pgd) { return 1; }
513 static inline int pgd_none(pgd_t pgd)    { return 0; }
514 static inline int pgd_bad(pgd_t pgd)     { return 0; }
515 
516 static inline int pud_present(pud_t pud) { return 1; }
517 static inline int pud_none(pud_t pud)	 { return 0; }
518 static inline int pud_large(pud_t pud)	 { return 0; }
519 static inline int pud_bad(pud_t pud)	 { return 0; }
520 
521 #else /* CONFIG_64BIT */
522 
523 static inline int pgd_present(pgd_t pgd)
524 {
525 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
526 		return 1;
527 	return (pgd_val(pgd) & _REGION_ENTRY_ORIGIN) != 0UL;
528 }
529 
530 static inline int pgd_none(pgd_t pgd)
531 {
532 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
533 		return 0;
534 	return (pgd_val(pgd) & _REGION_ENTRY_INVALID) != 0UL;
535 }
536 
537 static inline int pgd_bad(pgd_t pgd)
538 {
539 	/*
540 	 * With dynamic page table levels the pgd can be a region table
541 	 * entry or a segment table entry. Check for the bit that are
542 	 * invalid for either table entry.
543 	 */
544 	unsigned long mask =
545 		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INVALID &
546 		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
547 	return (pgd_val(pgd) & mask) != 0;
548 }
549 
550 static inline int pud_present(pud_t pud)
551 {
552 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
553 		return 1;
554 	return (pud_val(pud) & _REGION_ENTRY_ORIGIN) != 0UL;
555 }
556 
557 static inline int pud_none(pud_t pud)
558 {
559 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
560 		return 0;
561 	return (pud_val(pud) & _REGION_ENTRY_INVALID) != 0UL;
562 }
563 
564 static inline int pud_large(pud_t pud)
565 {
566 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) != _REGION_ENTRY_TYPE_R3)
567 		return 0;
568 	return !!(pud_val(pud) & _REGION3_ENTRY_LARGE);
569 }
570 
571 static inline int pud_bad(pud_t pud)
572 {
573 	/*
574 	 * With dynamic page table levels the pud can be a region table
575 	 * entry or a segment table entry. Check for the bit that are
576 	 * invalid for either table entry.
577 	 */
578 	unsigned long mask =
579 		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INVALID &
580 		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
581 	return (pud_val(pud) & mask) != 0;
582 }
583 
584 #endif /* CONFIG_64BIT */
585 
586 static inline int pmd_present(pmd_t pmd)
587 {
588 	return pmd_val(pmd) != _SEGMENT_ENTRY_INVALID;
589 }
590 
591 static inline int pmd_none(pmd_t pmd)
592 {
593 	return pmd_val(pmd) == _SEGMENT_ENTRY_INVALID;
594 }
595 
596 static inline int pmd_large(pmd_t pmd)
597 {
598 	return (pmd_val(pmd) & _SEGMENT_ENTRY_LARGE) != 0;
599 }
600 
601 static inline int pmd_pfn(pmd_t pmd)
602 {
603 	unsigned long origin_mask;
604 
605 	origin_mask = _SEGMENT_ENTRY_ORIGIN;
606 	if (pmd_large(pmd))
607 		origin_mask = _SEGMENT_ENTRY_ORIGIN_LARGE;
608 	return (pmd_val(pmd) & origin_mask) >> PAGE_SHIFT;
609 }
610 
611 static inline int pmd_bad(pmd_t pmd)
612 {
613 	if (pmd_large(pmd))
614 		return (pmd_val(pmd) & ~_SEGMENT_ENTRY_BITS_LARGE) != 0;
615 	return (pmd_val(pmd) & ~_SEGMENT_ENTRY_BITS) != 0;
616 }
617 
618 #define __HAVE_ARCH_PMDP_SPLITTING_FLUSH
619 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
620 				 unsigned long addr, pmd_t *pmdp);
621 
622 #define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
623 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
624 				 unsigned long address, pmd_t *pmdp,
625 				 pmd_t entry, int dirty);
626 
627 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
628 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
629 				  unsigned long address, pmd_t *pmdp);
630 
631 #define __HAVE_ARCH_PMD_WRITE
632 static inline int pmd_write(pmd_t pmd)
633 {
634 	return (pmd_val(pmd) & _SEGMENT_ENTRY_WRITE) != 0;
635 }
636 
637 static inline int pmd_dirty(pmd_t pmd)
638 {
639 	int dirty = 1;
640 	if (pmd_large(pmd))
641 		dirty = (pmd_val(pmd) & _SEGMENT_ENTRY_DIRTY) != 0;
642 	return dirty;
643 }
644 
645 static inline int pmd_young(pmd_t pmd)
646 {
647 	int young = 1;
648 	if (pmd_large(pmd))
649 		young = (pmd_val(pmd) & _SEGMENT_ENTRY_YOUNG) != 0;
650 	return young;
651 }
652 
653 static inline int pte_present(pte_t pte)
654 {
655 	/* Bit pattern: (pte & 0x001) == 0x001 */
656 	return (pte_val(pte) & _PAGE_PRESENT) != 0;
657 }
658 
659 static inline int pte_none(pte_t pte)
660 {
661 	/* Bit pattern: pte == 0x400 */
662 	return pte_val(pte) == _PAGE_INVALID;
663 }
664 
665 static inline int pte_swap(pte_t pte)
666 {
667 	/* Bit pattern: (pte & 0x603) == 0x402 */
668 	return (pte_val(pte) & (_PAGE_INVALID | _PAGE_PROTECT |
669 				_PAGE_TYPE | _PAGE_PRESENT))
670 		== (_PAGE_INVALID | _PAGE_TYPE);
671 }
672 
673 static inline int pte_special(pte_t pte)
674 {
675 	return (pte_val(pte) & _PAGE_SPECIAL);
676 }
677 
678 #define __HAVE_ARCH_PTE_SAME
679 static inline int pte_same(pte_t a, pte_t b)
680 {
681 	return pte_val(a) == pte_val(b);
682 }
683 
684 static inline pgste_t pgste_get_lock(pte_t *ptep)
685 {
686 	unsigned long new = 0;
687 #ifdef CONFIG_PGSTE
688 	unsigned long old;
689 
690 	preempt_disable();
691 	asm(
692 		"	lg	%0,%2\n"
693 		"0:	lgr	%1,%0\n"
694 		"	nihh	%0,0xff7f\n"	/* clear PCL bit in old */
695 		"	oihh	%1,0x0080\n"	/* set PCL bit in new */
696 		"	csg	%0,%1,%2\n"
697 		"	jl	0b\n"
698 		: "=&d" (old), "=&d" (new), "=Q" (ptep[PTRS_PER_PTE])
699 		: "Q" (ptep[PTRS_PER_PTE]) : "cc", "memory");
700 #endif
701 	return __pgste(new);
702 }
703 
704 static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste)
705 {
706 #ifdef CONFIG_PGSTE
707 	asm(
708 		"	nihh	%1,0xff7f\n"	/* clear PCL bit */
709 		"	stg	%1,%0\n"
710 		: "=Q" (ptep[PTRS_PER_PTE])
711 		: "d" (pgste_val(pgste)), "Q" (ptep[PTRS_PER_PTE])
712 		: "cc", "memory");
713 	preempt_enable();
714 #endif
715 }
716 
717 static inline pgste_t pgste_get(pte_t *ptep)
718 {
719 	unsigned long pgste = 0;
720 #ifdef CONFIG_PGSTE
721 	pgste = *(unsigned long *)(ptep + PTRS_PER_PTE);
722 #endif
723 	return __pgste(pgste);
724 }
725 
726 static inline void pgste_set(pte_t *ptep, pgste_t pgste)
727 {
728 #ifdef CONFIG_PGSTE
729 	*(pgste_t *)(ptep + PTRS_PER_PTE) = pgste;
730 #endif
731 }
732 
733 static inline pgste_t pgste_update_all(pte_t *ptep, pgste_t pgste,
734 				       struct mm_struct *mm)
735 {
736 #ifdef CONFIG_PGSTE
737 	unsigned long address, bits, skey;
738 
739 	if (!mm_use_skey(mm) || pte_val(*ptep) & _PAGE_INVALID)
740 		return pgste;
741 	address = pte_val(*ptep) & PAGE_MASK;
742 	skey = (unsigned long) page_get_storage_key(address);
743 	bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
744 	/* Transfer page changed & referenced bit to guest bits in pgste */
745 	pgste_val(pgste) |= bits << 48;		/* GR bit & GC bit */
746 	/* Copy page access key and fetch protection bit to pgste */
747 	pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
748 	pgste_val(pgste) |= (skey & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
749 #endif
750 	return pgste;
751 
752 }
753 
754 static inline void pgste_set_key(pte_t *ptep, pgste_t pgste, pte_t entry,
755 				 struct mm_struct *mm)
756 {
757 #ifdef CONFIG_PGSTE
758 	unsigned long address;
759 	unsigned long nkey;
760 
761 	if (!mm_use_skey(mm) || pte_val(entry) & _PAGE_INVALID)
762 		return;
763 	VM_BUG_ON(!(pte_val(*ptep) & _PAGE_INVALID));
764 	address = pte_val(entry) & PAGE_MASK;
765 	/*
766 	 * Set page access key and fetch protection bit from pgste.
767 	 * The guest C/R information is still in the PGSTE, set real
768 	 * key C/R to 0.
769 	 */
770 	nkey = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
771 	nkey |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
772 	page_set_storage_key(address, nkey, 0);
773 #endif
774 }
775 
776 static inline pgste_t pgste_set_pte(pte_t *ptep, pgste_t pgste, pte_t entry)
777 {
778 	if ((pte_val(entry) & _PAGE_PRESENT) &&
779 	    (pte_val(entry) & _PAGE_WRITE) &&
780 	    !(pte_val(entry) & _PAGE_INVALID)) {
781 		if (!MACHINE_HAS_ESOP) {
782 			/*
783 			 * Without enhanced suppression-on-protection force
784 			 * the dirty bit on for all writable ptes.
785 			 */
786 			pte_val(entry) |= _PAGE_DIRTY;
787 			pte_val(entry) &= ~_PAGE_PROTECT;
788 		}
789 		if (!(pte_val(entry) & _PAGE_PROTECT))
790 			/* This pte allows write access, set user-dirty */
791 			pgste_val(pgste) |= PGSTE_UC_BIT;
792 	}
793 	*ptep = entry;
794 	return pgste;
795 }
796 
797 /**
798  * struct gmap_struct - guest address space
799  * @crst_list: list of all crst tables used in the guest address space
800  * @mm: pointer to the parent mm_struct
801  * @guest_to_host: radix tree with guest to host address translation
802  * @host_to_guest: radix tree with pointer to segment table entries
803  * @guest_table_lock: spinlock to protect all entries in the guest page table
804  * @table: pointer to the page directory
805  * @asce: address space control element for gmap page table
806  * @pfault_enabled: defines if pfaults are applicable for the guest
807  */
808 struct gmap {
809 	struct list_head list;
810 	struct list_head crst_list;
811 	struct mm_struct *mm;
812 	struct radix_tree_root guest_to_host;
813 	struct radix_tree_root host_to_guest;
814 	spinlock_t guest_table_lock;
815 	unsigned long *table;
816 	unsigned long asce;
817 	unsigned long asce_end;
818 	void *private;
819 	bool pfault_enabled;
820 };
821 
822 /**
823  * struct gmap_notifier - notify function block for page invalidation
824  * @notifier_call: address of callback function
825  */
826 struct gmap_notifier {
827 	struct list_head list;
828 	void (*notifier_call)(struct gmap *gmap, unsigned long gaddr);
829 };
830 
831 struct gmap *gmap_alloc(struct mm_struct *mm, unsigned long limit);
832 void gmap_free(struct gmap *gmap);
833 void gmap_enable(struct gmap *gmap);
834 void gmap_disable(struct gmap *gmap);
835 int gmap_map_segment(struct gmap *gmap, unsigned long from,
836 		     unsigned long to, unsigned long len);
837 int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len);
838 unsigned long __gmap_translate(struct gmap *, unsigned long gaddr);
839 unsigned long gmap_translate(struct gmap *, unsigned long gaddr);
840 int __gmap_link(struct gmap *gmap, unsigned long gaddr, unsigned long vmaddr);
841 int gmap_fault(struct gmap *, unsigned long gaddr, unsigned int fault_flags);
842 void gmap_discard(struct gmap *, unsigned long from, unsigned long to);
843 void __gmap_zap(struct gmap *, unsigned long gaddr);
844 bool gmap_test_and_clear_dirty(unsigned long address, struct gmap *);
845 
846 
847 void gmap_register_ipte_notifier(struct gmap_notifier *);
848 void gmap_unregister_ipte_notifier(struct gmap_notifier *);
849 int gmap_ipte_notify(struct gmap *, unsigned long start, unsigned long len);
850 void gmap_do_ipte_notify(struct mm_struct *, unsigned long addr, pte_t *);
851 
852 static inline pgste_t pgste_ipte_notify(struct mm_struct *mm,
853 					unsigned long addr,
854 					pte_t *ptep, pgste_t pgste)
855 {
856 #ifdef CONFIG_PGSTE
857 	if (pgste_val(pgste) & PGSTE_IN_BIT) {
858 		pgste_val(pgste) &= ~PGSTE_IN_BIT;
859 		gmap_do_ipte_notify(mm, addr, ptep);
860 	}
861 #endif
862 	return pgste;
863 }
864 
865 /*
866  * Certain architectures need to do special things when PTEs
867  * within a page table are directly modified.  Thus, the following
868  * hook is made available.
869  */
870 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
871 			      pte_t *ptep, pte_t entry)
872 {
873 	pgste_t pgste;
874 
875 	if (mm_has_pgste(mm)) {
876 		pgste = pgste_get_lock(ptep);
877 		pgste_val(pgste) &= ~_PGSTE_GPS_ZERO;
878 		pgste_set_key(ptep, pgste, entry, mm);
879 		pgste = pgste_set_pte(ptep, pgste, entry);
880 		pgste_set_unlock(ptep, pgste);
881 	} else {
882 		*ptep = entry;
883 	}
884 }
885 
886 /*
887  * query functions pte_write/pte_dirty/pte_young only work if
888  * pte_present() is true. Undefined behaviour if not..
889  */
890 static inline int pte_write(pte_t pte)
891 {
892 	return (pte_val(pte) & _PAGE_WRITE) != 0;
893 }
894 
895 static inline int pte_dirty(pte_t pte)
896 {
897 	return (pte_val(pte) & _PAGE_DIRTY) != 0;
898 }
899 
900 static inline int pte_young(pte_t pte)
901 {
902 	return (pte_val(pte) & _PAGE_YOUNG) != 0;
903 }
904 
905 #define __HAVE_ARCH_PTE_UNUSED
906 static inline int pte_unused(pte_t pte)
907 {
908 	return pte_val(pte) & _PAGE_UNUSED;
909 }
910 
911 /*
912  * pgd/pmd/pte modification functions
913  */
914 
915 static inline void pgd_clear(pgd_t *pgd)
916 {
917 #ifdef CONFIG_64BIT
918 	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
919 		pgd_val(*pgd) = _REGION2_ENTRY_EMPTY;
920 #endif
921 }
922 
923 static inline void pud_clear(pud_t *pud)
924 {
925 #ifdef CONFIG_64BIT
926 	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
927 		pud_val(*pud) = _REGION3_ENTRY_EMPTY;
928 #endif
929 }
930 
931 static inline void pmd_clear(pmd_t *pmdp)
932 {
933 	pmd_val(*pmdp) = _SEGMENT_ENTRY_INVALID;
934 }
935 
936 static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
937 {
938 	pte_val(*ptep) = _PAGE_INVALID;
939 }
940 
941 /*
942  * The following pte modification functions only work if
943  * pte_present() is true. Undefined behaviour if not..
944  */
945 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
946 {
947 	pte_val(pte) &= _PAGE_CHG_MASK;
948 	pte_val(pte) |= pgprot_val(newprot);
949 	/*
950 	 * newprot for PAGE_NONE, PAGE_READ and PAGE_WRITE has the
951 	 * invalid bit set, clear it again for readable, young pages
952 	 */
953 	if ((pte_val(pte) & _PAGE_YOUNG) && (pte_val(pte) & _PAGE_READ))
954 		pte_val(pte) &= ~_PAGE_INVALID;
955 	/*
956 	 * newprot for PAGE_READ and PAGE_WRITE has the page protection
957 	 * bit set, clear it again for writable, dirty pages
958 	 */
959 	if ((pte_val(pte) & _PAGE_DIRTY) && (pte_val(pte) & _PAGE_WRITE))
960 		pte_val(pte) &= ~_PAGE_PROTECT;
961 	return pte;
962 }
963 
964 static inline pte_t pte_wrprotect(pte_t pte)
965 {
966 	pte_val(pte) &= ~_PAGE_WRITE;
967 	pte_val(pte) |= _PAGE_PROTECT;
968 	return pte;
969 }
970 
971 static inline pte_t pte_mkwrite(pte_t pte)
972 {
973 	pte_val(pte) |= _PAGE_WRITE;
974 	if (pte_val(pte) & _PAGE_DIRTY)
975 		pte_val(pte) &= ~_PAGE_PROTECT;
976 	return pte;
977 }
978 
979 static inline pte_t pte_mkclean(pte_t pte)
980 {
981 	pte_val(pte) &= ~_PAGE_DIRTY;
982 	pte_val(pte) |= _PAGE_PROTECT;
983 	return pte;
984 }
985 
986 static inline pte_t pte_mkdirty(pte_t pte)
987 {
988 	pte_val(pte) |= _PAGE_DIRTY;
989 	if (pte_val(pte) & _PAGE_WRITE)
990 		pte_val(pte) &= ~_PAGE_PROTECT;
991 	return pte;
992 }
993 
994 static inline pte_t pte_mkold(pte_t pte)
995 {
996 	pte_val(pte) &= ~_PAGE_YOUNG;
997 	pte_val(pte) |= _PAGE_INVALID;
998 	return pte;
999 }
1000 
1001 static inline pte_t pte_mkyoung(pte_t pte)
1002 {
1003 	pte_val(pte) |= _PAGE_YOUNG;
1004 	if (pte_val(pte) & _PAGE_READ)
1005 		pte_val(pte) &= ~_PAGE_INVALID;
1006 	return pte;
1007 }
1008 
1009 static inline pte_t pte_mkspecial(pte_t pte)
1010 {
1011 	pte_val(pte) |= _PAGE_SPECIAL;
1012 	return pte;
1013 }
1014 
1015 #ifdef CONFIG_HUGETLB_PAGE
1016 static inline pte_t pte_mkhuge(pte_t pte)
1017 {
1018 	pte_val(pte) |= _PAGE_LARGE;
1019 	return pte;
1020 }
1021 #endif
1022 
1023 static inline void __ptep_ipte(unsigned long address, pte_t *ptep)
1024 {
1025 	unsigned long pto = (unsigned long) ptep;
1026 
1027 #ifndef CONFIG_64BIT
1028 	/* pto in ESA mode must point to the start of the segment table */
1029 	pto &= 0x7ffffc00;
1030 #endif
1031 	/* Invalidation + global TLB flush for the pte */
1032 	asm volatile(
1033 		"	ipte	%2,%3"
1034 		: "=m" (*ptep) : "m" (*ptep), "a" (pto), "a" (address));
1035 }
1036 
1037 static inline void __ptep_ipte_local(unsigned long address, pte_t *ptep)
1038 {
1039 	unsigned long pto = (unsigned long) ptep;
1040 
1041 #ifndef CONFIG_64BIT
1042 	/* pto in ESA mode must point to the start of the segment table */
1043 	pto &= 0x7ffffc00;
1044 #endif
1045 	/* Invalidation + local TLB flush for the pte */
1046 	asm volatile(
1047 		"	.insn rrf,0xb2210000,%2,%3,0,1"
1048 		: "=m" (*ptep) : "m" (*ptep), "a" (pto), "a" (address));
1049 }
1050 
1051 static inline void __ptep_ipte_range(unsigned long address, int nr, pte_t *ptep)
1052 {
1053 	unsigned long pto = (unsigned long) ptep;
1054 
1055 #ifndef CONFIG_64BIT
1056 	/* pto in ESA mode must point to the start of the segment table */
1057 	pto &= 0x7ffffc00;
1058 #endif
1059 	/* Invalidate a range of ptes + global TLB flush of the ptes */
1060 	do {
1061 		asm volatile(
1062 			"	.insn rrf,0xb2210000,%2,%0,%1,0"
1063 			: "+a" (address), "+a" (nr) : "a" (pto) : "memory");
1064 	} while (nr != 255);
1065 }
1066 
1067 static inline void ptep_flush_direct(struct mm_struct *mm,
1068 				     unsigned long address, pte_t *ptep)
1069 {
1070 	int active, count;
1071 
1072 	if (pte_val(*ptep) & _PAGE_INVALID)
1073 		return;
1074 	active = (mm == current->active_mm) ? 1 : 0;
1075 	count = atomic_add_return(0x10000, &mm->context.attach_count);
1076 	if (MACHINE_HAS_TLB_LC && (count & 0xffff) <= active &&
1077 	    cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
1078 		__ptep_ipte_local(address, ptep);
1079 	else
1080 		__ptep_ipte(address, ptep);
1081 	atomic_sub(0x10000, &mm->context.attach_count);
1082 }
1083 
1084 static inline void ptep_flush_lazy(struct mm_struct *mm,
1085 				   unsigned long address, pte_t *ptep)
1086 {
1087 	int active, count;
1088 
1089 	if (pte_val(*ptep) & _PAGE_INVALID)
1090 		return;
1091 	active = (mm == current->active_mm) ? 1 : 0;
1092 	count = atomic_add_return(0x10000, &mm->context.attach_count);
1093 	if ((count & 0xffff) <= active) {
1094 		pte_val(*ptep) |= _PAGE_INVALID;
1095 		mm->context.flush_mm = 1;
1096 	} else
1097 		__ptep_ipte(address, ptep);
1098 	atomic_sub(0x10000, &mm->context.attach_count);
1099 }
1100 
1101 /*
1102  * Get (and clear) the user dirty bit for a pte.
1103  */
1104 static inline int ptep_test_and_clear_user_dirty(struct mm_struct *mm,
1105 						 unsigned long addr,
1106 						 pte_t *ptep)
1107 {
1108 	pgste_t pgste;
1109 	pte_t pte;
1110 	int dirty;
1111 
1112 	if (!mm_has_pgste(mm))
1113 		return 0;
1114 	pgste = pgste_get_lock(ptep);
1115 	dirty = !!(pgste_val(pgste) & PGSTE_UC_BIT);
1116 	pgste_val(pgste) &= ~PGSTE_UC_BIT;
1117 	pte = *ptep;
1118 	if (dirty && (pte_val(pte) & _PAGE_PRESENT)) {
1119 		pgste = pgste_ipte_notify(mm, addr, ptep, pgste);
1120 		__ptep_ipte(addr, ptep);
1121 		if (MACHINE_HAS_ESOP || !(pte_val(pte) & _PAGE_WRITE))
1122 			pte_val(pte) |= _PAGE_PROTECT;
1123 		else
1124 			pte_val(pte) |= _PAGE_INVALID;
1125 		*ptep = pte;
1126 	}
1127 	pgste_set_unlock(ptep, pgste);
1128 	return dirty;
1129 }
1130 
1131 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1132 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
1133 					    unsigned long addr, pte_t *ptep)
1134 {
1135 	pgste_t pgste;
1136 	pte_t pte, oldpte;
1137 	int young;
1138 
1139 	if (mm_has_pgste(vma->vm_mm)) {
1140 		pgste = pgste_get_lock(ptep);
1141 		pgste = pgste_ipte_notify(vma->vm_mm, addr, ptep, pgste);
1142 	}
1143 
1144 	oldpte = pte = *ptep;
1145 	ptep_flush_direct(vma->vm_mm, addr, ptep);
1146 	young = pte_young(pte);
1147 	pte = pte_mkold(pte);
1148 
1149 	if (mm_has_pgste(vma->vm_mm)) {
1150 		pgste = pgste_update_all(&oldpte, pgste, vma->vm_mm);
1151 		pgste = pgste_set_pte(ptep, pgste, pte);
1152 		pgste_set_unlock(ptep, pgste);
1153 	} else
1154 		*ptep = pte;
1155 
1156 	return young;
1157 }
1158 
1159 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1160 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
1161 					 unsigned long address, pte_t *ptep)
1162 {
1163 	return ptep_test_and_clear_young(vma, address, ptep);
1164 }
1165 
1166 /*
1167  * This is hard to understand. ptep_get_and_clear and ptep_clear_flush
1168  * both clear the TLB for the unmapped pte. The reason is that
1169  * ptep_get_and_clear is used in common code (e.g. change_pte_range)
1170  * to modify an active pte. The sequence is
1171  *   1) ptep_get_and_clear
1172  *   2) set_pte_at
1173  *   3) flush_tlb_range
1174  * On s390 the tlb needs to get flushed with the modification of the pte
1175  * if the pte is active. The only way how this can be implemented is to
1176  * have ptep_get_and_clear do the tlb flush. In exchange flush_tlb_range
1177  * is a nop.
1178  */
1179 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1180 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
1181 				       unsigned long address, pte_t *ptep)
1182 {
1183 	pgste_t pgste;
1184 	pte_t pte;
1185 
1186 	if (mm_has_pgste(mm)) {
1187 		pgste = pgste_get_lock(ptep);
1188 		pgste = pgste_ipte_notify(mm, address, ptep, pgste);
1189 	}
1190 
1191 	pte = *ptep;
1192 	ptep_flush_lazy(mm, address, ptep);
1193 	pte_val(*ptep) = _PAGE_INVALID;
1194 
1195 	if (mm_has_pgste(mm)) {
1196 		pgste = pgste_update_all(&pte, pgste, mm);
1197 		pgste_set_unlock(ptep, pgste);
1198 	}
1199 	return pte;
1200 }
1201 
1202 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1203 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
1204 					   unsigned long address,
1205 					   pte_t *ptep)
1206 {
1207 	pgste_t pgste;
1208 	pte_t pte;
1209 
1210 	if (mm_has_pgste(mm)) {
1211 		pgste = pgste_get_lock(ptep);
1212 		pgste_ipte_notify(mm, address, ptep, pgste);
1213 	}
1214 
1215 	pte = *ptep;
1216 	ptep_flush_lazy(mm, address, ptep);
1217 
1218 	if (mm_has_pgste(mm)) {
1219 		pgste = pgste_update_all(&pte, pgste, mm);
1220 		pgste_set(ptep, pgste);
1221 	}
1222 	return pte;
1223 }
1224 
1225 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
1226 					   unsigned long address,
1227 					   pte_t *ptep, pte_t pte)
1228 {
1229 	pgste_t pgste;
1230 
1231 	if (mm_has_pgste(mm)) {
1232 		pgste = pgste_get(ptep);
1233 		pgste_set_key(ptep, pgste, pte, mm);
1234 		pgste = pgste_set_pte(ptep, pgste, pte);
1235 		pgste_set_unlock(ptep, pgste);
1236 	} else
1237 		*ptep = pte;
1238 }
1239 
1240 #define __HAVE_ARCH_PTEP_CLEAR_FLUSH
1241 static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
1242 				     unsigned long address, pte_t *ptep)
1243 {
1244 	pgste_t pgste;
1245 	pte_t pte;
1246 
1247 	if (mm_has_pgste(vma->vm_mm)) {
1248 		pgste = pgste_get_lock(ptep);
1249 		pgste = pgste_ipte_notify(vma->vm_mm, address, ptep, pgste);
1250 	}
1251 
1252 	pte = *ptep;
1253 	ptep_flush_direct(vma->vm_mm, address, ptep);
1254 	pte_val(*ptep) = _PAGE_INVALID;
1255 
1256 	if (mm_has_pgste(vma->vm_mm)) {
1257 		if ((pgste_val(pgste) & _PGSTE_GPS_USAGE_MASK) ==
1258 		    _PGSTE_GPS_USAGE_UNUSED)
1259 			pte_val(pte) |= _PAGE_UNUSED;
1260 		pgste = pgste_update_all(&pte, pgste, vma->vm_mm);
1261 		pgste_set_unlock(ptep, pgste);
1262 	}
1263 	return pte;
1264 }
1265 
1266 /*
1267  * The batched pte unmap code uses ptep_get_and_clear_full to clear the
1268  * ptes. Here an optimization is possible. tlb_gather_mmu flushes all
1269  * tlbs of an mm if it can guarantee that the ptes of the mm_struct
1270  * cannot be accessed while the batched unmap is running. In this case
1271  * full==1 and a simple pte_clear is enough. See tlb.h.
1272  */
1273 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
1274 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
1275 					    unsigned long address,
1276 					    pte_t *ptep, int full)
1277 {
1278 	pgste_t pgste;
1279 	pte_t pte;
1280 
1281 	if (!full && mm_has_pgste(mm)) {
1282 		pgste = pgste_get_lock(ptep);
1283 		pgste = pgste_ipte_notify(mm, address, ptep, pgste);
1284 	}
1285 
1286 	pte = *ptep;
1287 	if (!full)
1288 		ptep_flush_lazy(mm, address, ptep);
1289 	pte_val(*ptep) = _PAGE_INVALID;
1290 
1291 	if (!full && mm_has_pgste(mm)) {
1292 		pgste = pgste_update_all(&pte, pgste, mm);
1293 		pgste_set_unlock(ptep, pgste);
1294 	}
1295 	return pte;
1296 }
1297 
1298 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1299 static inline pte_t ptep_set_wrprotect(struct mm_struct *mm,
1300 				       unsigned long address, pte_t *ptep)
1301 {
1302 	pgste_t pgste;
1303 	pte_t pte = *ptep;
1304 
1305 	if (pte_write(pte)) {
1306 		if (mm_has_pgste(mm)) {
1307 			pgste = pgste_get_lock(ptep);
1308 			pgste = pgste_ipte_notify(mm, address, ptep, pgste);
1309 		}
1310 
1311 		ptep_flush_lazy(mm, address, ptep);
1312 		pte = pte_wrprotect(pte);
1313 
1314 		if (mm_has_pgste(mm)) {
1315 			pgste = pgste_set_pte(ptep, pgste, pte);
1316 			pgste_set_unlock(ptep, pgste);
1317 		} else
1318 			*ptep = pte;
1319 	}
1320 	return pte;
1321 }
1322 
1323 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1324 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
1325 					unsigned long address, pte_t *ptep,
1326 					pte_t entry, int dirty)
1327 {
1328 	pgste_t pgste;
1329 
1330 	if (pte_same(*ptep, entry))
1331 		return 0;
1332 	if (mm_has_pgste(vma->vm_mm)) {
1333 		pgste = pgste_get_lock(ptep);
1334 		pgste = pgste_ipte_notify(vma->vm_mm, address, ptep, pgste);
1335 	}
1336 
1337 	ptep_flush_direct(vma->vm_mm, address, ptep);
1338 
1339 	if (mm_has_pgste(vma->vm_mm)) {
1340 		pgste_set_key(ptep, pgste, entry, vma->vm_mm);
1341 		pgste = pgste_set_pte(ptep, pgste, entry);
1342 		pgste_set_unlock(ptep, pgste);
1343 	} else
1344 		*ptep = entry;
1345 	return 1;
1346 }
1347 
1348 /*
1349  * Conversion functions: convert a page and protection to a page entry,
1350  * and a page entry and page directory to the page they refer to.
1351  */
1352 static inline pte_t mk_pte_phys(unsigned long physpage, pgprot_t pgprot)
1353 {
1354 	pte_t __pte;
1355 	pte_val(__pte) = physpage + pgprot_val(pgprot);
1356 	return pte_mkyoung(__pte);
1357 }
1358 
1359 static inline pte_t mk_pte(struct page *page, pgprot_t pgprot)
1360 {
1361 	unsigned long physpage = page_to_phys(page);
1362 	pte_t __pte = mk_pte_phys(physpage, pgprot);
1363 
1364 	if (pte_write(__pte) && PageDirty(page))
1365 		__pte = pte_mkdirty(__pte);
1366 	return __pte;
1367 }
1368 
1369 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD-1))
1370 #define pud_index(address) (((address) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
1371 #define pmd_index(address) (((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1))
1372 #define pte_index(address) (((address) >> PAGE_SHIFT) & (PTRS_PER_PTE-1))
1373 
1374 #define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address))
1375 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
1376 
1377 #ifndef CONFIG_64BIT
1378 
1379 #define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
1380 #define pud_deref(pmd) ({ BUG(); 0UL; })
1381 #define pgd_deref(pmd) ({ BUG(); 0UL; })
1382 
1383 #define pud_offset(pgd, address) ((pud_t *) pgd)
1384 #define pmd_offset(pud, address) ((pmd_t *) pud + pmd_index(address))
1385 
1386 #else /* CONFIG_64BIT */
1387 
1388 #define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
1389 #define pud_deref(pud) (pud_val(pud) & _REGION_ENTRY_ORIGIN)
1390 #define pgd_deref(pgd) (pgd_val(pgd) & _REGION_ENTRY_ORIGIN)
1391 
1392 static inline pud_t *pud_offset(pgd_t *pgd, unsigned long address)
1393 {
1394 	pud_t *pud = (pud_t *) pgd;
1395 	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
1396 		pud = (pud_t *) pgd_deref(*pgd);
1397 	return pud  + pud_index(address);
1398 }
1399 
1400 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
1401 {
1402 	pmd_t *pmd = (pmd_t *) pud;
1403 	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
1404 		pmd = (pmd_t *) pud_deref(*pud);
1405 	return pmd + pmd_index(address);
1406 }
1407 
1408 #endif /* CONFIG_64BIT */
1409 
1410 #define pfn_pte(pfn,pgprot) mk_pte_phys(__pa((pfn) << PAGE_SHIFT),(pgprot))
1411 #define pte_pfn(x) (pte_val(x) >> PAGE_SHIFT)
1412 #define pte_page(x) pfn_to_page(pte_pfn(x))
1413 
1414 #define pmd_page(pmd) pfn_to_page(pmd_pfn(pmd))
1415 
1416 /* Find an entry in the lowest level page table.. */
1417 #define pte_offset(pmd, addr) ((pte_t *) pmd_deref(*(pmd)) + pte_index(addr))
1418 #define pte_offset_kernel(pmd, address) pte_offset(pmd,address)
1419 #define pte_offset_map(pmd, address) pte_offset_kernel(pmd, address)
1420 #define pte_unmap(pte) do { } while (0)
1421 
1422 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE)
1423 static inline unsigned long massage_pgprot_pmd(pgprot_t pgprot)
1424 {
1425 	/*
1426 	 * pgprot is PAGE_NONE, PAGE_READ, or PAGE_WRITE (see __Pxxx / __Sxxx)
1427 	 * Convert to segment table entry format.
1428 	 */
1429 	if (pgprot_val(pgprot) == pgprot_val(PAGE_NONE))
1430 		return pgprot_val(SEGMENT_NONE);
1431 	if (pgprot_val(pgprot) == pgprot_val(PAGE_READ))
1432 		return pgprot_val(SEGMENT_READ);
1433 	return pgprot_val(SEGMENT_WRITE);
1434 }
1435 
1436 static inline pmd_t pmd_wrprotect(pmd_t pmd)
1437 {
1438 	pmd_val(pmd) &= ~_SEGMENT_ENTRY_WRITE;
1439 	pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
1440 	return pmd;
1441 }
1442 
1443 static inline pmd_t pmd_mkwrite(pmd_t pmd)
1444 {
1445 	pmd_val(pmd) |= _SEGMENT_ENTRY_WRITE;
1446 	if (pmd_large(pmd) && !(pmd_val(pmd) & _SEGMENT_ENTRY_DIRTY))
1447 		return pmd;
1448 	pmd_val(pmd) &= ~_SEGMENT_ENTRY_PROTECT;
1449 	return pmd;
1450 }
1451 
1452 static inline pmd_t pmd_mkclean(pmd_t pmd)
1453 {
1454 	if (pmd_large(pmd)) {
1455 		pmd_val(pmd) &= ~_SEGMENT_ENTRY_DIRTY;
1456 		pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
1457 	}
1458 	return pmd;
1459 }
1460 
1461 static inline pmd_t pmd_mkdirty(pmd_t pmd)
1462 {
1463 	if (pmd_large(pmd)) {
1464 		pmd_val(pmd) |= _SEGMENT_ENTRY_DIRTY;
1465 		if (pmd_val(pmd) & _SEGMENT_ENTRY_WRITE)
1466 			pmd_val(pmd) &= ~_SEGMENT_ENTRY_PROTECT;
1467 	}
1468 	return pmd;
1469 }
1470 
1471 static inline pmd_t pmd_mkyoung(pmd_t pmd)
1472 {
1473 	if (pmd_large(pmd)) {
1474 		pmd_val(pmd) |= _SEGMENT_ENTRY_YOUNG;
1475 		if (pmd_val(pmd) & _SEGMENT_ENTRY_READ)
1476 			pmd_val(pmd) &= ~_SEGMENT_ENTRY_INVALID;
1477 	}
1478 	return pmd;
1479 }
1480 
1481 static inline pmd_t pmd_mkold(pmd_t pmd)
1482 {
1483 	if (pmd_large(pmd)) {
1484 		pmd_val(pmd) &= ~_SEGMENT_ENTRY_YOUNG;
1485 		pmd_val(pmd) |= _SEGMENT_ENTRY_INVALID;
1486 	}
1487 	return pmd;
1488 }
1489 
1490 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
1491 {
1492 	if (pmd_large(pmd)) {
1493 		pmd_val(pmd) &= _SEGMENT_ENTRY_ORIGIN_LARGE |
1494 			_SEGMENT_ENTRY_DIRTY | _SEGMENT_ENTRY_YOUNG |
1495 			_SEGMENT_ENTRY_LARGE | _SEGMENT_ENTRY_SPLIT;
1496 		pmd_val(pmd) |= massage_pgprot_pmd(newprot);
1497 		if (!(pmd_val(pmd) & _SEGMENT_ENTRY_DIRTY))
1498 			pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
1499 		if (!(pmd_val(pmd) & _SEGMENT_ENTRY_YOUNG))
1500 			pmd_val(pmd) |= _SEGMENT_ENTRY_INVALID;
1501 		return pmd;
1502 	}
1503 	pmd_val(pmd) &= _SEGMENT_ENTRY_ORIGIN;
1504 	pmd_val(pmd) |= massage_pgprot_pmd(newprot);
1505 	return pmd;
1506 }
1507 
1508 static inline pmd_t mk_pmd_phys(unsigned long physpage, pgprot_t pgprot)
1509 {
1510 	pmd_t __pmd;
1511 	pmd_val(__pmd) = physpage + massage_pgprot_pmd(pgprot);
1512 	return __pmd;
1513 }
1514 
1515 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLB_PAGE */
1516 
1517 static inline void __pmdp_csp(pmd_t *pmdp)
1518 {
1519 	register unsigned long reg2 asm("2") = pmd_val(*pmdp);
1520 	register unsigned long reg3 asm("3") = pmd_val(*pmdp) |
1521 					       _SEGMENT_ENTRY_INVALID;
1522 	register unsigned long reg4 asm("4") = ((unsigned long) pmdp) + 5;
1523 
1524 	asm volatile(
1525 		"	csp %1,%3"
1526 		: "=m" (*pmdp)
1527 		: "d" (reg2), "d" (reg3), "d" (reg4), "m" (*pmdp) : "cc");
1528 }
1529 
1530 static inline void __pmdp_idte(unsigned long address, pmd_t *pmdp)
1531 {
1532 	unsigned long sto;
1533 
1534 	sto = (unsigned long) pmdp - pmd_index(address) * sizeof(pmd_t);
1535 	asm volatile(
1536 		"	.insn	rrf,0xb98e0000,%2,%3,0,0"
1537 		: "=m" (*pmdp)
1538 		: "m" (*pmdp), "a" (sto), "a" ((address & HPAGE_MASK))
1539 		: "cc" );
1540 }
1541 
1542 static inline void __pmdp_idte_local(unsigned long address, pmd_t *pmdp)
1543 {
1544 	unsigned long sto;
1545 
1546 	sto = (unsigned long) pmdp - pmd_index(address) * sizeof(pmd_t);
1547 	asm volatile(
1548 		"	.insn	rrf,0xb98e0000,%2,%3,0,1"
1549 		: "=m" (*pmdp)
1550 		: "m" (*pmdp), "a" (sto), "a" ((address & HPAGE_MASK))
1551 		: "cc" );
1552 }
1553 
1554 static inline void pmdp_flush_direct(struct mm_struct *mm,
1555 				     unsigned long address, pmd_t *pmdp)
1556 {
1557 	int active, count;
1558 
1559 	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
1560 		return;
1561 	if (!MACHINE_HAS_IDTE) {
1562 		__pmdp_csp(pmdp);
1563 		return;
1564 	}
1565 	active = (mm == current->active_mm) ? 1 : 0;
1566 	count = atomic_add_return(0x10000, &mm->context.attach_count);
1567 	if (MACHINE_HAS_TLB_LC && (count & 0xffff) <= active &&
1568 	    cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
1569 		__pmdp_idte_local(address, pmdp);
1570 	else
1571 		__pmdp_idte(address, pmdp);
1572 	atomic_sub(0x10000, &mm->context.attach_count);
1573 }
1574 
1575 static inline void pmdp_flush_lazy(struct mm_struct *mm,
1576 				   unsigned long address, pmd_t *pmdp)
1577 {
1578 	int active, count;
1579 
1580 	if (pmd_val(*pmdp) & _SEGMENT_ENTRY_INVALID)
1581 		return;
1582 	active = (mm == current->active_mm) ? 1 : 0;
1583 	count = atomic_add_return(0x10000, &mm->context.attach_count);
1584 	if ((count & 0xffff) <= active) {
1585 		pmd_val(*pmdp) |= _SEGMENT_ENTRY_INVALID;
1586 		mm->context.flush_mm = 1;
1587 	} else if (MACHINE_HAS_IDTE)
1588 		__pmdp_idte(address, pmdp);
1589 	else
1590 		__pmdp_csp(pmdp);
1591 	atomic_sub(0x10000, &mm->context.attach_count);
1592 }
1593 
1594 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1595 
1596 #define __HAVE_ARCH_PGTABLE_DEPOSIT
1597 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
1598 				       pgtable_t pgtable);
1599 
1600 #define __HAVE_ARCH_PGTABLE_WITHDRAW
1601 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
1602 
1603 static inline int pmd_trans_splitting(pmd_t pmd)
1604 {
1605 	return (pmd_val(pmd) & _SEGMENT_ENTRY_LARGE) &&
1606 		(pmd_val(pmd) & _SEGMENT_ENTRY_SPLIT);
1607 }
1608 
1609 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1610 			      pmd_t *pmdp, pmd_t entry)
1611 {
1612 	*pmdp = entry;
1613 }
1614 
1615 static inline pmd_t pmd_mkhuge(pmd_t pmd)
1616 {
1617 	pmd_val(pmd) |= _SEGMENT_ENTRY_LARGE;
1618 	pmd_val(pmd) |= _SEGMENT_ENTRY_YOUNG;
1619 	pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
1620 	return pmd;
1621 }
1622 
1623 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1624 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1625 					    unsigned long address, pmd_t *pmdp)
1626 {
1627 	pmd_t pmd;
1628 
1629 	pmd = *pmdp;
1630 	pmdp_flush_direct(vma->vm_mm, address, pmdp);
1631 	*pmdp = pmd_mkold(pmd);
1632 	return pmd_young(pmd);
1633 }
1634 
1635 #define __HAVE_ARCH_PMDP_GET_AND_CLEAR
1636 static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
1637 				       unsigned long address, pmd_t *pmdp)
1638 {
1639 	pmd_t pmd = *pmdp;
1640 
1641 	pmdp_flush_direct(mm, address, pmdp);
1642 	pmd_clear(pmdp);
1643 	return pmd;
1644 }
1645 
1646 #define __HAVE_ARCH_PMDP_GET_AND_CLEAR_FULL
1647 static inline pmd_t pmdp_get_and_clear_full(struct mm_struct *mm,
1648 					    unsigned long address,
1649 					    pmd_t *pmdp, int full)
1650 {
1651 	pmd_t pmd = *pmdp;
1652 
1653 	if (!full)
1654 		pmdp_flush_lazy(mm, address, pmdp);
1655 	pmd_clear(pmdp);
1656 	return pmd;
1657 }
1658 
1659 #define __HAVE_ARCH_PMDP_CLEAR_FLUSH
1660 static inline pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
1661 				     unsigned long address, pmd_t *pmdp)
1662 {
1663 	return pmdp_get_and_clear(vma->vm_mm, address, pmdp);
1664 }
1665 
1666 #define __HAVE_ARCH_PMDP_INVALIDATE
1667 static inline void pmdp_invalidate(struct vm_area_struct *vma,
1668 				   unsigned long address, pmd_t *pmdp)
1669 {
1670 	pmdp_flush_direct(vma->vm_mm, address, pmdp);
1671 }
1672 
1673 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1674 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1675 				      unsigned long address, pmd_t *pmdp)
1676 {
1677 	pmd_t pmd = *pmdp;
1678 
1679 	if (pmd_write(pmd)) {
1680 		pmdp_flush_direct(mm, address, pmdp);
1681 		set_pmd_at(mm, address, pmdp, pmd_wrprotect(pmd));
1682 	}
1683 }
1684 
1685 #define pfn_pmd(pfn, pgprot)	mk_pmd_phys(__pa((pfn) << PAGE_SHIFT), (pgprot))
1686 #define mk_pmd(page, pgprot)	pfn_pmd(page_to_pfn(page), (pgprot))
1687 
1688 static inline int pmd_trans_huge(pmd_t pmd)
1689 {
1690 	return pmd_val(pmd) & _SEGMENT_ENTRY_LARGE;
1691 }
1692 
1693 static inline int has_transparent_hugepage(void)
1694 {
1695 	return MACHINE_HAS_HPAGE ? 1 : 0;
1696 }
1697 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1698 
1699 /*
1700  * 31 bit swap entry format:
1701  * A page-table entry has some bits we have to treat in a special way.
1702  * Bits 0, 20 and bit 23 have to be zero, otherwise an specification
1703  * exception will occur instead of a page translation exception. The
1704  * specifiation exception has the bad habit not to store necessary
1705  * information in the lowcore.
1706  * Bits 21, 22, 30 and 31 are used to indicate the page type.
1707  * A swap pte is indicated by bit pattern (pte & 0x603) == 0x402
1708  * This leaves the bits 1-19 and bits 24-29 to store type and offset.
1709  * We use the 5 bits from 25-29 for the type and the 20 bits from 1-19
1710  * plus 24 for the offset.
1711  * 0|     offset        |0110|o|type |00|
1712  * 0 0000000001111111111 2222 2 22222 33
1713  * 0 1234567890123456789 0123 4 56789 01
1714  *
1715  * 64 bit swap entry format:
1716  * A page-table entry has some bits we have to treat in a special way.
1717  * Bits 52 and bit 55 have to be zero, otherwise an specification
1718  * exception will occur instead of a page translation exception. The
1719  * specifiation exception has the bad habit not to store necessary
1720  * information in the lowcore.
1721  * Bits 53, 54, 62 and 63 are used to indicate the page type.
1722  * A swap pte is indicated by bit pattern (pte & 0x603) == 0x402
1723  * This leaves the bits 0-51 and bits 56-61 to store type and offset.
1724  * We use the 5 bits from 57-61 for the type and the 53 bits from 0-51
1725  * plus 56 for the offset.
1726  * |                      offset                        |0110|o|type |00|
1727  *  0000000000111111111122222222223333333333444444444455 5555 5 55566 66
1728  *  0123456789012345678901234567890123456789012345678901 2345 6 78901 23
1729  */
1730 #ifndef CONFIG_64BIT
1731 #define __SWP_OFFSET_MASK (~0UL >> 12)
1732 #else
1733 #define __SWP_OFFSET_MASK (~0UL >> 11)
1734 #endif
1735 static inline pte_t mk_swap_pte(unsigned long type, unsigned long offset)
1736 {
1737 	pte_t pte;
1738 	offset &= __SWP_OFFSET_MASK;
1739 	pte_val(pte) = _PAGE_INVALID | _PAGE_TYPE | ((type & 0x1f) << 2) |
1740 		((offset & 1UL) << 7) | ((offset & ~1UL) << 11);
1741 	return pte;
1742 }
1743 
1744 #define __swp_type(entry)	(((entry).val >> 2) & 0x1f)
1745 #define __swp_offset(entry)	(((entry).val >> 11) | (((entry).val >> 7) & 1))
1746 #define __swp_entry(type,offset) ((swp_entry_t) { pte_val(mk_swap_pte((type),(offset))) })
1747 
1748 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
1749 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
1750 
1751 #endif /* !__ASSEMBLY__ */
1752 
1753 #define kern_addr_valid(addr)   (1)
1754 
1755 extern int vmem_add_mapping(unsigned long start, unsigned long size);
1756 extern int vmem_remove_mapping(unsigned long start, unsigned long size);
1757 extern int s390_enable_sie(void);
1758 extern int s390_enable_skey(void);
1759 extern void s390_reset_cmma(struct mm_struct *mm);
1760 
1761 /* s390 has a private copy of get unmapped area to deal with cache synonyms */
1762 #define HAVE_ARCH_UNMAPPED_AREA
1763 #define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1764 
1765 /*
1766  * No page table caches to initialise
1767  */
1768 static inline void pgtable_cache_init(void) { }
1769 static inline void check_pgt_cache(void) { }
1770 
1771 #include <asm-generic/pgtable.h>
1772 
1773 #endif /* _S390_PAGE_H */
1774