xref: /linux/arch/s390/include/asm/pgtable.h (revision bbcd53c960713507ae764bf81970651b5577b95a)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  *  S390 version
4  *    Copyright IBM Corp. 1999, 2000
5  *    Author(s): Hartmut Penner (hp@de.ibm.com)
6  *               Ulrich Weigand (weigand@de.ibm.com)
7  *               Martin Schwidefsky (schwidefsky@de.ibm.com)
8  *
9  *  Derived from "include/asm-i386/pgtable.h"
10  */
11 
12 #ifndef _ASM_S390_PGTABLE_H
13 #define _ASM_S390_PGTABLE_H
14 
15 #include <linux/sched.h>
16 #include <linux/mm_types.h>
17 #include <linux/page-flags.h>
18 #include <linux/radix-tree.h>
19 #include <linux/atomic.h>
20 #include <asm/bug.h>
21 #include <asm/page.h>
22 #include <asm/uv.h>
23 
24 extern pgd_t swapper_pg_dir[];
25 extern void paging_init(void);
26 extern unsigned long s390_invalid_asce;
27 
28 enum {
29 	PG_DIRECT_MAP_4K = 0,
30 	PG_DIRECT_MAP_1M,
31 	PG_DIRECT_MAP_2G,
32 	PG_DIRECT_MAP_MAX
33 };
34 
35 extern atomic_long_t direct_pages_count[PG_DIRECT_MAP_MAX];
36 
37 static inline void update_page_count(int level, long count)
38 {
39 	if (IS_ENABLED(CONFIG_PROC_FS))
40 		atomic_long_add(count, &direct_pages_count[level]);
41 }
42 
43 struct seq_file;
44 void arch_report_meminfo(struct seq_file *m);
45 
46 /*
47  * The S390 doesn't have any external MMU info: the kernel page
48  * tables contain all the necessary information.
49  */
50 #define update_mmu_cache(vma, address, ptep)     do { } while (0)
51 #define update_mmu_cache_pmd(vma, address, ptep) do { } while (0)
52 
53 /*
54  * ZERO_PAGE is a global shared page that is always zero; used
55  * for zero-mapped memory areas etc..
56  */
57 
58 extern unsigned long empty_zero_page;
59 extern unsigned long zero_page_mask;
60 
61 #define ZERO_PAGE(vaddr) \
62 	(virt_to_page((void *)(empty_zero_page + \
63 	 (((unsigned long)(vaddr)) &zero_page_mask))))
64 #define __HAVE_COLOR_ZERO_PAGE
65 
66 /* TODO: s390 cannot support io_remap_pfn_range... */
67 
68 #define FIRST_USER_ADDRESS  0UL
69 
70 #define pte_ERROR(e) \
71 	printk("%s:%d: bad pte %p.\n", __FILE__, __LINE__, (void *) pte_val(e))
72 #define pmd_ERROR(e) \
73 	printk("%s:%d: bad pmd %p.\n", __FILE__, __LINE__, (void *) pmd_val(e))
74 #define pud_ERROR(e) \
75 	printk("%s:%d: bad pud %p.\n", __FILE__, __LINE__, (void *) pud_val(e))
76 #define p4d_ERROR(e) \
77 	printk("%s:%d: bad p4d %p.\n", __FILE__, __LINE__, (void *) p4d_val(e))
78 #define pgd_ERROR(e) \
79 	printk("%s:%d: bad pgd %p.\n", __FILE__, __LINE__, (void *) pgd_val(e))
80 
81 /*
82  * The vmalloc and module area will always be on the topmost area of the
83  * kernel mapping. 512GB are reserved for vmalloc by default.
84  * At the top of the vmalloc area a 2GB area is reserved where modules
85  * will reside. That makes sure that inter module branches always
86  * happen without trampolines and in addition the placement within a
87  * 2GB frame is branch prediction unit friendly.
88  */
89 extern unsigned long VMALLOC_START;
90 extern unsigned long VMALLOC_END;
91 #define VMALLOC_DEFAULT_SIZE	((512UL << 30) - MODULES_LEN)
92 extern struct page *vmemmap;
93 extern unsigned long vmemmap_size;
94 
95 #define VMEM_MAX_PHYS ((unsigned long) vmemmap)
96 
97 extern unsigned long MODULES_VADDR;
98 extern unsigned long MODULES_END;
99 #define MODULES_VADDR	MODULES_VADDR
100 #define MODULES_END	MODULES_END
101 #define MODULES_LEN	(1UL << 31)
102 
103 static inline int is_module_addr(void *addr)
104 {
105 	BUILD_BUG_ON(MODULES_LEN > (1UL << 31));
106 	if (addr < (void *)MODULES_VADDR)
107 		return 0;
108 	if (addr > (void *)MODULES_END)
109 		return 0;
110 	return 1;
111 }
112 
113 /*
114  * A 64 bit pagetable entry of S390 has following format:
115  * |			 PFRA			      |0IPC|  OS  |
116  * 0000000000111111111122222222223333333333444444444455555555556666
117  * 0123456789012345678901234567890123456789012345678901234567890123
118  *
119  * I Page-Invalid Bit:    Page is not available for address-translation
120  * P Page-Protection Bit: Store access not possible for page
121  * C Change-bit override: HW is not required to set change bit
122  *
123  * A 64 bit segmenttable entry of S390 has following format:
124  * |        P-table origin                              |      TT
125  * 0000000000111111111122222222223333333333444444444455555555556666
126  * 0123456789012345678901234567890123456789012345678901234567890123
127  *
128  * I Segment-Invalid Bit:    Segment is not available for address-translation
129  * C Common-Segment Bit:     Segment is not private (PoP 3-30)
130  * P Page-Protection Bit: Store access not possible for page
131  * TT Type 00
132  *
133  * A 64 bit region table entry of S390 has following format:
134  * |        S-table origin                             |   TF  TTTL
135  * 0000000000111111111122222222223333333333444444444455555555556666
136  * 0123456789012345678901234567890123456789012345678901234567890123
137  *
138  * I Segment-Invalid Bit:    Segment is not available for address-translation
139  * TT Type 01
140  * TF
141  * TL Table length
142  *
143  * The 64 bit regiontable origin of S390 has following format:
144  * |      region table origon                          |       DTTL
145  * 0000000000111111111122222222223333333333444444444455555555556666
146  * 0123456789012345678901234567890123456789012345678901234567890123
147  *
148  * X Space-Switch event:
149  * G Segment-Invalid Bit:
150  * P Private-Space Bit:
151  * S Storage-Alteration:
152  * R Real space
153  * TL Table-Length:
154  *
155  * A storage key has the following format:
156  * | ACC |F|R|C|0|
157  *  0   3 4 5 6 7
158  * ACC: access key
159  * F  : fetch protection bit
160  * R  : referenced bit
161  * C  : changed bit
162  */
163 
164 /* Hardware bits in the page table entry */
165 #define _PAGE_NOEXEC	0x100		/* HW no-execute bit  */
166 #define _PAGE_PROTECT	0x200		/* HW read-only bit  */
167 #define _PAGE_INVALID	0x400		/* HW invalid bit    */
168 #define _PAGE_LARGE	0x800		/* Bit to mark a large pte */
169 
170 /* Software bits in the page table entry */
171 #define _PAGE_PRESENT	0x001		/* SW pte present bit */
172 #define _PAGE_YOUNG	0x004		/* SW pte young bit */
173 #define _PAGE_DIRTY	0x008		/* SW pte dirty bit */
174 #define _PAGE_READ	0x010		/* SW pte read bit */
175 #define _PAGE_WRITE	0x020		/* SW pte write bit */
176 #define _PAGE_SPECIAL	0x040		/* SW associated with special page */
177 #define _PAGE_UNUSED	0x080		/* SW bit for pgste usage state */
178 
179 #ifdef CONFIG_MEM_SOFT_DIRTY
180 #define _PAGE_SOFT_DIRTY 0x002		/* SW pte soft dirty bit */
181 #else
182 #define _PAGE_SOFT_DIRTY 0x000
183 #endif
184 
185 /* Set of bits not changed in pte_modify */
186 #define _PAGE_CHG_MASK		(PAGE_MASK | _PAGE_SPECIAL | _PAGE_DIRTY | \
187 				 _PAGE_YOUNG | _PAGE_SOFT_DIRTY)
188 
189 /*
190  * handle_pte_fault uses pte_present and pte_none to find out the pte type
191  * WITHOUT holding the page table lock. The _PAGE_PRESENT bit is used to
192  * distinguish present from not-present ptes. It is changed only with the page
193  * table lock held.
194  *
195  * The following table gives the different possible bit combinations for
196  * the pte hardware and software bits in the last 12 bits of a pte
197  * (. unassigned bit, x don't care, t swap type):
198  *
199  *				842100000000
200  *				000084210000
201  *				000000008421
202  *				.IR.uswrdy.p
203  * empty			.10.00000000
204  * swap				.11..ttttt.0
205  * prot-none, clean, old	.11.xx0000.1
206  * prot-none, clean, young	.11.xx0001.1
207  * prot-none, dirty, old	.11.xx0010.1
208  * prot-none, dirty, young	.11.xx0011.1
209  * read-only, clean, old	.11.xx0100.1
210  * read-only, clean, young	.01.xx0101.1
211  * read-only, dirty, old	.11.xx0110.1
212  * read-only, dirty, young	.01.xx0111.1
213  * read-write, clean, old	.11.xx1100.1
214  * read-write, clean, young	.01.xx1101.1
215  * read-write, dirty, old	.10.xx1110.1
216  * read-write, dirty, young	.00.xx1111.1
217  * HW-bits: R read-only, I invalid
218  * SW-bits: p present, y young, d dirty, r read, w write, s special,
219  *	    u unused, l large
220  *
221  * pte_none    is true for the bit pattern .10.00000000, pte == 0x400
222  * pte_swap    is true for the bit pattern .11..ooooo.0, (pte & 0x201) == 0x200
223  * pte_present is true for the bit pattern .xx.xxxxxx.1, (pte & 0x001) == 0x001
224  */
225 
226 /* Bits in the segment/region table address-space-control-element */
227 #define _ASCE_ORIGIN		~0xfffUL/* region/segment table origin	    */
228 #define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
229 #define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
230 #define _ASCE_SPACE_SWITCH	0x40	/* space switch event		    */
231 #define _ASCE_REAL_SPACE	0x20	/* real space control		    */
232 #define _ASCE_TYPE_MASK		0x0c	/* asce table type mask		    */
233 #define _ASCE_TYPE_REGION1	0x0c	/* region first table type	    */
234 #define _ASCE_TYPE_REGION2	0x08	/* region second table type	    */
235 #define _ASCE_TYPE_REGION3	0x04	/* region third table type	    */
236 #define _ASCE_TYPE_SEGMENT	0x00	/* segment table type		    */
237 #define _ASCE_TABLE_LENGTH	0x03	/* region table length		    */
238 
239 /* Bits in the region table entry */
240 #define _REGION_ENTRY_ORIGIN	~0xfffUL/* region/segment table origin	    */
241 #define _REGION_ENTRY_PROTECT	0x200	/* region protection bit	    */
242 #define _REGION_ENTRY_NOEXEC	0x100	/* region no-execute bit	    */
243 #define _REGION_ENTRY_OFFSET	0xc0	/* region table offset		    */
244 #define _REGION_ENTRY_INVALID	0x20	/* invalid region table entry	    */
245 #define _REGION_ENTRY_TYPE_MASK	0x0c	/* region table type mask	    */
246 #define _REGION_ENTRY_TYPE_R1	0x0c	/* region first table type	    */
247 #define _REGION_ENTRY_TYPE_R2	0x08	/* region second table type	    */
248 #define _REGION_ENTRY_TYPE_R3	0x04	/* region third table type	    */
249 #define _REGION_ENTRY_LENGTH	0x03	/* region third length		    */
250 
251 #define _REGION1_ENTRY		(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_LENGTH)
252 #define _REGION1_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID)
253 #define _REGION2_ENTRY		(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_LENGTH)
254 #define _REGION2_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID)
255 #define _REGION3_ENTRY		(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_LENGTH)
256 #define _REGION3_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID)
257 
258 #define _REGION3_ENTRY_ORIGIN_LARGE ~0x7fffffffUL /* large page address	     */
259 #define _REGION3_ENTRY_DIRTY	0x2000	/* SW region dirty bit */
260 #define _REGION3_ENTRY_YOUNG	0x1000	/* SW region young bit */
261 #define _REGION3_ENTRY_LARGE	0x0400	/* RTTE-format control, large page  */
262 #define _REGION3_ENTRY_READ	0x0002	/* SW region read bit */
263 #define _REGION3_ENTRY_WRITE	0x0001	/* SW region write bit */
264 
265 #ifdef CONFIG_MEM_SOFT_DIRTY
266 #define _REGION3_ENTRY_SOFT_DIRTY 0x4000 /* SW region soft dirty bit */
267 #else
268 #define _REGION3_ENTRY_SOFT_DIRTY 0x0000 /* SW region soft dirty bit */
269 #endif
270 
271 #define _REGION_ENTRY_BITS	 0xfffffffffffff22fUL
272 
273 /* Bits in the segment table entry */
274 #define _SEGMENT_ENTRY_BITS			0xfffffffffffffe33UL
275 #define _SEGMENT_ENTRY_HARDWARE_BITS		0xfffffffffffffe30UL
276 #define _SEGMENT_ENTRY_HARDWARE_BITS_LARGE	0xfffffffffff00730UL
277 #define _SEGMENT_ENTRY_ORIGIN_LARGE ~0xfffffUL /* large page address	    */
278 #define _SEGMENT_ENTRY_ORIGIN	~0x7ffUL/* page table origin		    */
279 #define _SEGMENT_ENTRY_PROTECT	0x200	/* segment protection bit	    */
280 #define _SEGMENT_ENTRY_NOEXEC	0x100	/* segment no-execute bit	    */
281 #define _SEGMENT_ENTRY_INVALID	0x20	/* invalid segment table entry	    */
282 #define _SEGMENT_ENTRY_TYPE_MASK 0x0c	/* segment table type mask	    */
283 
284 #define _SEGMENT_ENTRY		(0)
285 #define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INVALID)
286 
287 #define _SEGMENT_ENTRY_DIRTY	0x2000	/* SW segment dirty bit */
288 #define _SEGMENT_ENTRY_YOUNG	0x1000	/* SW segment young bit */
289 #define _SEGMENT_ENTRY_LARGE	0x0400	/* STE-format control, large page */
290 #define _SEGMENT_ENTRY_WRITE	0x0002	/* SW segment write bit */
291 #define _SEGMENT_ENTRY_READ	0x0001	/* SW segment read bit */
292 
293 #ifdef CONFIG_MEM_SOFT_DIRTY
294 #define _SEGMENT_ENTRY_SOFT_DIRTY 0x4000 /* SW segment soft dirty bit */
295 #else
296 #define _SEGMENT_ENTRY_SOFT_DIRTY 0x0000 /* SW segment soft dirty bit */
297 #endif
298 
299 #define _CRST_ENTRIES	2048	/* number of region/segment table entries */
300 #define _PAGE_ENTRIES	256	/* number of page table entries	*/
301 
302 #define _CRST_TABLE_SIZE (_CRST_ENTRIES * 8)
303 #define _PAGE_TABLE_SIZE (_PAGE_ENTRIES * 8)
304 
305 #define _REGION1_SHIFT	53
306 #define _REGION2_SHIFT	42
307 #define _REGION3_SHIFT	31
308 #define _SEGMENT_SHIFT	20
309 
310 #define _REGION1_INDEX	(0x7ffUL << _REGION1_SHIFT)
311 #define _REGION2_INDEX	(0x7ffUL << _REGION2_SHIFT)
312 #define _REGION3_INDEX	(0x7ffUL << _REGION3_SHIFT)
313 #define _SEGMENT_INDEX	(0x7ffUL << _SEGMENT_SHIFT)
314 #define _PAGE_INDEX	(0xffUL  << _PAGE_SHIFT)
315 
316 #define _REGION1_SIZE	(1UL << _REGION1_SHIFT)
317 #define _REGION2_SIZE	(1UL << _REGION2_SHIFT)
318 #define _REGION3_SIZE	(1UL << _REGION3_SHIFT)
319 #define _SEGMENT_SIZE	(1UL << _SEGMENT_SHIFT)
320 
321 #define _REGION1_MASK	(~(_REGION1_SIZE - 1))
322 #define _REGION2_MASK	(~(_REGION2_SIZE - 1))
323 #define _REGION3_MASK	(~(_REGION3_SIZE - 1))
324 #define _SEGMENT_MASK	(~(_SEGMENT_SIZE - 1))
325 
326 #define PMD_SHIFT	_SEGMENT_SHIFT
327 #define PUD_SHIFT	_REGION3_SHIFT
328 #define P4D_SHIFT	_REGION2_SHIFT
329 #define PGDIR_SHIFT	_REGION1_SHIFT
330 
331 #define PMD_SIZE	_SEGMENT_SIZE
332 #define PUD_SIZE	_REGION3_SIZE
333 #define P4D_SIZE	_REGION2_SIZE
334 #define PGDIR_SIZE	_REGION1_SIZE
335 
336 #define PMD_MASK	_SEGMENT_MASK
337 #define PUD_MASK	_REGION3_MASK
338 #define P4D_MASK	_REGION2_MASK
339 #define PGDIR_MASK	_REGION1_MASK
340 
341 #define PTRS_PER_PTE	_PAGE_ENTRIES
342 #define PTRS_PER_PMD	_CRST_ENTRIES
343 #define PTRS_PER_PUD	_CRST_ENTRIES
344 #define PTRS_PER_P4D	_CRST_ENTRIES
345 #define PTRS_PER_PGD	_CRST_ENTRIES
346 
347 #define MAX_PTRS_PER_P4D	PTRS_PER_P4D
348 
349 /*
350  * Segment table and region3 table entry encoding
351  * (R = read-only, I = invalid, y = young bit):
352  *				dy..R...I...wr
353  * prot-none, clean, old	00..1...1...00
354  * prot-none, clean, young	01..1...1...00
355  * prot-none, dirty, old	10..1...1...00
356  * prot-none, dirty, young	11..1...1...00
357  * read-only, clean, old	00..1...1...01
358  * read-only, clean, young	01..1...0...01
359  * read-only, dirty, old	10..1...1...01
360  * read-only, dirty, young	11..1...0...01
361  * read-write, clean, old	00..1...1...11
362  * read-write, clean, young	01..1...0...11
363  * read-write, dirty, old	10..0...1...11
364  * read-write, dirty, young	11..0...0...11
365  * The segment table origin is used to distinguish empty (origin==0) from
366  * read-write, old segment table entries (origin!=0)
367  * HW-bits: R read-only, I invalid
368  * SW-bits: y young, d dirty, r read, w write
369  */
370 
371 /* Page status table bits for virtualization */
372 #define PGSTE_ACC_BITS	0xf000000000000000UL
373 #define PGSTE_FP_BIT	0x0800000000000000UL
374 #define PGSTE_PCL_BIT	0x0080000000000000UL
375 #define PGSTE_HR_BIT	0x0040000000000000UL
376 #define PGSTE_HC_BIT	0x0020000000000000UL
377 #define PGSTE_GR_BIT	0x0004000000000000UL
378 #define PGSTE_GC_BIT	0x0002000000000000UL
379 #define PGSTE_UC_BIT	0x0000800000000000UL	/* user dirty (migration) */
380 #define PGSTE_IN_BIT	0x0000400000000000UL	/* IPTE notify bit */
381 #define PGSTE_VSIE_BIT	0x0000200000000000UL	/* ref'd in a shadow table */
382 
383 /* Guest Page State used for virtualization */
384 #define _PGSTE_GPS_ZERO			0x0000000080000000UL
385 #define _PGSTE_GPS_NODAT		0x0000000040000000UL
386 #define _PGSTE_GPS_USAGE_MASK		0x0000000003000000UL
387 #define _PGSTE_GPS_USAGE_STABLE		0x0000000000000000UL
388 #define _PGSTE_GPS_USAGE_UNUSED		0x0000000001000000UL
389 #define _PGSTE_GPS_USAGE_POT_VOLATILE	0x0000000002000000UL
390 #define _PGSTE_GPS_USAGE_VOLATILE	_PGSTE_GPS_USAGE_MASK
391 
392 /*
393  * A user page table pointer has the space-switch-event bit, the
394  * private-space-control bit and the storage-alteration-event-control
395  * bit set. A kernel page table pointer doesn't need them.
396  */
397 #define _ASCE_USER_BITS		(_ASCE_SPACE_SWITCH | _ASCE_PRIVATE_SPACE | \
398 				 _ASCE_ALT_EVENT)
399 
400 /*
401  * Page protection definitions.
402  */
403 #define PAGE_NONE	__pgprot(_PAGE_PRESENT | _PAGE_INVALID | _PAGE_PROTECT)
404 #define PAGE_RO		__pgprot(_PAGE_PRESENT | _PAGE_READ | \
405 				 _PAGE_NOEXEC  | _PAGE_INVALID | _PAGE_PROTECT)
406 #define PAGE_RX		__pgprot(_PAGE_PRESENT | _PAGE_READ | \
407 				 _PAGE_INVALID | _PAGE_PROTECT)
408 #define PAGE_RW		__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
409 				 _PAGE_NOEXEC  | _PAGE_INVALID | _PAGE_PROTECT)
410 #define PAGE_RWX	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
411 				 _PAGE_INVALID | _PAGE_PROTECT)
412 
413 #define PAGE_SHARED	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
414 				 _PAGE_YOUNG | _PAGE_DIRTY | _PAGE_NOEXEC)
415 #define PAGE_KERNEL	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
416 				 _PAGE_YOUNG | _PAGE_DIRTY | _PAGE_NOEXEC)
417 #define PAGE_KERNEL_RO	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_YOUNG | \
418 				 _PAGE_PROTECT | _PAGE_NOEXEC)
419 #define PAGE_KERNEL_EXEC __pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
420 				  _PAGE_YOUNG |	_PAGE_DIRTY)
421 
422 /*
423  * On s390 the page table entry has an invalid bit and a read-only bit.
424  * Read permission implies execute permission and write permission
425  * implies read permission.
426  */
427          /*xwr*/
428 #define __P000	PAGE_NONE
429 #define __P001	PAGE_RO
430 #define __P010	PAGE_RO
431 #define __P011	PAGE_RO
432 #define __P100	PAGE_RX
433 #define __P101	PAGE_RX
434 #define __P110	PAGE_RX
435 #define __P111	PAGE_RX
436 
437 #define __S000	PAGE_NONE
438 #define __S001	PAGE_RO
439 #define __S010	PAGE_RW
440 #define __S011	PAGE_RW
441 #define __S100	PAGE_RX
442 #define __S101	PAGE_RX
443 #define __S110	PAGE_RWX
444 #define __S111	PAGE_RWX
445 
446 /*
447  * Segment entry (large page) protection definitions.
448  */
449 #define SEGMENT_NONE	__pgprot(_SEGMENT_ENTRY_INVALID | \
450 				 _SEGMENT_ENTRY_PROTECT)
451 #define SEGMENT_RO	__pgprot(_SEGMENT_ENTRY_PROTECT | \
452 				 _SEGMENT_ENTRY_READ | \
453 				 _SEGMENT_ENTRY_NOEXEC)
454 #define SEGMENT_RX	__pgprot(_SEGMENT_ENTRY_PROTECT | \
455 				 _SEGMENT_ENTRY_READ)
456 #define SEGMENT_RW	__pgprot(_SEGMENT_ENTRY_READ | \
457 				 _SEGMENT_ENTRY_WRITE | \
458 				 _SEGMENT_ENTRY_NOEXEC)
459 #define SEGMENT_RWX	__pgprot(_SEGMENT_ENTRY_READ | \
460 				 _SEGMENT_ENTRY_WRITE)
461 #define SEGMENT_KERNEL	__pgprot(_SEGMENT_ENTRY |	\
462 				 _SEGMENT_ENTRY_LARGE |	\
463 				 _SEGMENT_ENTRY_READ |	\
464 				 _SEGMENT_ENTRY_WRITE | \
465 				 _SEGMENT_ENTRY_YOUNG | \
466 				 _SEGMENT_ENTRY_DIRTY | \
467 				 _SEGMENT_ENTRY_NOEXEC)
468 #define SEGMENT_KERNEL_RO __pgprot(_SEGMENT_ENTRY |	\
469 				 _SEGMENT_ENTRY_LARGE |	\
470 				 _SEGMENT_ENTRY_READ |	\
471 				 _SEGMENT_ENTRY_YOUNG |	\
472 				 _SEGMENT_ENTRY_PROTECT | \
473 				 _SEGMENT_ENTRY_NOEXEC)
474 #define SEGMENT_KERNEL_EXEC __pgprot(_SEGMENT_ENTRY |	\
475 				 _SEGMENT_ENTRY_LARGE |	\
476 				 _SEGMENT_ENTRY_READ |	\
477 				 _SEGMENT_ENTRY_WRITE | \
478 				 _SEGMENT_ENTRY_YOUNG |	\
479 				 _SEGMENT_ENTRY_DIRTY)
480 
481 /*
482  * Region3 entry (large page) protection definitions.
483  */
484 
485 #define REGION3_KERNEL	__pgprot(_REGION_ENTRY_TYPE_R3 | \
486 				 _REGION3_ENTRY_LARGE |	 \
487 				 _REGION3_ENTRY_READ |	 \
488 				 _REGION3_ENTRY_WRITE |	 \
489 				 _REGION3_ENTRY_YOUNG |	 \
490 				 _REGION3_ENTRY_DIRTY | \
491 				 _REGION_ENTRY_NOEXEC)
492 #define REGION3_KERNEL_RO __pgprot(_REGION_ENTRY_TYPE_R3 | \
493 				   _REGION3_ENTRY_LARGE |  \
494 				   _REGION3_ENTRY_READ |   \
495 				   _REGION3_ENTRY_YOUNG |  \
496 				   _REGION_ENTRY_PROTECT | \
497 				   _REGION_ENTRY_NOEXEC)
498 
499 static inline bool mm_p4d_folded(struct mm_struct *mm)
500 {
501 	return mm->context.asce_limit <= _REGION1_SIZE;
502 }
503 #define mm_p4d_folded(mm) mm_p4d_folded(mm)
504 
505 static inline bool mm_pud_folded(struct mm_struct *mm)
506 {
507 	return mm->context.asce_limit <= _REGION2_SIZE;
508 }
509 #define mm_pud_folded(mm) mm_pud_folded(mm)
510 
511 static inline bool mm_pmd_folded(struct mm_struct *mm)
512 {
513 	return mm->context.asce_limit <= _REGION3_SIZE;
514 }
515 #define mm_pmd_folded(mm) mm_pmd_folded(mm)
516 
517 static inline int mm_has_pgste(struct mm_struct *mm)
518 {
519 #ifdef CONFIG_PGSTE
520 	if (unlikely(mm->context.has_pgste))
521 		return 1;
522 #endif
523 	return 0;
524 }
525 
526 static inline int mm_is_protected(struct mm_struct *mm)
527 {
528 #ifdef CONFIG_PGSTE
529 	if (unlikely(atomic_read(&mm->context.is_protected)))
530 		return 1;
531 #endif
532 	return 0;
533 }
534 
535 static inline int mm_alloc_pgste(struct mm_struct *mm)
536 {
537 #ifdef CONFIG_PGSTE
538 	if (unlikely(mm->context.alloc_pgste))
539 		return 1;
540 #endif
541 	return 0;
542 }
543 
544 /*
545  * In the case that a guest uses storage keys
546  * faults should no longer be backed by zero pages
547  */
548 #define mm_forbids_zeropage mm_has_pgste
549 static inline int mm_uses_skeys(struct mm_struct *mm)
550 {
551 #ifdef CONFIG_PGSTE
552 	if (mm->context.uses_skeys)
553 		return 1;
554 #endif
555 	return 0;
556 }
557 
558 static inline void csp(unsigned int *ptr, unsigned int old, unsigned int new)
559 {
560 	register unsigned long reg2 asm("2") = old;
561 	register unsigned long reg3 asm("3") = new;
562 	unsigned long address = (unsigned long)ptr | 1;
563 
564 	asm volatile(
565 		"	csp	%0,%3"
566 		: "+d" (reg2), "+m" (*ptr)
567 		: "d" (reg3), "d" (address)
568 		: "cc");
569 }
570 
571 static inline void cspg(unsigned long *ptr, unsigned long old, unsigned long new)
572 {
573 	register unsigned long reg2 asm("2") = old;
574 	register unsigned long reg3 asm("3") = new;
575 	unsigned long address = (unsigned long)ptr | 1;
576 
577 	asm volatile(
578 		"	.insn	rre,0xb98a0000,%0,%3"
579 		: "+d" (reg2), "+m" (*ptr)
580 		: "d" (reg3), "d" (address)
581 		: "cc");
582 }
583 
584 #define CRDTE_DTT_PAGE		0x00UL
585 #define CRDTE_DTT_SEGMENT	0x10UL
586 #define CRDTE_DTT_REGION3	0x14UL
587 #define CRDTE_DTT_REGION2	0x18UL
588 #define CRDTE_DTT_REGION1	0x1cUL
589 
590 static inline void crdte(unsigned long old, unsigned long new,
591 			 unsigned long table, unsigned long dtt,
592 			 unsigned long address, unsigned long asce)
593 {
594 	register unsigned long reg2 asm("2") = old;
595 	register unsigned long reg3 asm("3") = new;
596 	register unsigned long reg4 asm("4") = table | dtt;
597 	register unsigned long reg5 asm("5") = address;
598 
599 	asm volatile(".insn rrf,0xb98f0000,%0,%2,%4,0"
600 		     : "+d" (reg2)
601 		     : "d" (reg3), "d" (reg4), "d" (reg5), "a" (asce)
602 		     : "memory", "cc");
603 }
604 
605 /*
606  * pgd/p4d/pud/pmd/pte query functions
607  */
608 static inline int pgd_folded(pgd_t pgd)
609 {
610 	return (pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R1;
611 }
612 
613 static inline int pgd_present(pgd_t pgd)
614 {
615 	if (pgd_folded(pgd))
616 		return 1;
617 	return (pgd_val(pgd) & _REGION_ENTRY_ORIGIN) != 0UL;
618 }
619 
620 static inline int pgd_none(pgd_t pgd)
621 {
622 	if (pgd_folded(pgd))
623 		return 0;
624 	return (pgd_val(pgd) & _REGION_ENTRY_INVALID) != 0UL;
625 }
626 
627 static inline int pgd_bad(pgd_t pgd)
628 {
629 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R1)
630 		return 0;
631 	return (pgd_val(pgd) & ~_REGION_ENTRY_BITS) != 0;
632 }
633 
634 static inline unsigned long pgd_pfn(pgd_t pgd)
635 {
636 	unsigned long origin_mask;
637 
638 	origin_mask = _REGION_ENTRY_ORIGIN;
639 	return (pgd_val(pgd) & origin_mask) >> PAGE_SHIFT;
640 }
641 
642 static inline int p4d_folded(p4d_t p4d)
643 {
644 	return (p4d_val(p4d) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2;
645 }
646 
647 static inline int p4d_present(p4d_t p4d)
648 {
649 	if (p4d_folded(p4d))
650 		return 1;
651 	return (p4d_val(p4d) & _REGION_ENTRY_ORIGIN) != 0UL;
652 }
653 
654 static inline int p4d_none(p4d_t p4d)
655 {
656 	if (p4d_folded(p4d))
657 		return 0;
658 	return p4d_val(p4d) == _REGION2_ENTRY_EMPTY;
659 }
660 
661 static inline unsigned long p4d_pfn(p4d_t p4d)
662 {
663 	unsigned long origin_mask;
664 
665 	origin_mask = _REGION_ENTRY_ORIGIN;
666 	return (p4d_val(p4d) & origin_mask) >> PAGE_SHIFT;
667 }
668 
669 static inline int pud_folded(pud_t pud)
670 {
671 	return (pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3;
672 }
673 
674 static inline int pud_present(pud_t pud)
675 {
676 	if (pud_folded(pud))
677 		return 1;
678 	return (pud_val(pud) & _REGION_ENTRY_ORIGIN) != 0UL;
679 }
680 
681 static inline int pud_none(pud_t pud)
682 {
683 	if (pud_folded(pud))
684 		return 0;
685 	return pud_val(pud) == _REGION3_ENTRY_EMPTY;
686 }
687 
688 #define pud_leaf	pud_large
689 static inline int pud_large(pud_t pud)
690 {
691 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) != _REGION_ENTRY_TYPE_R3)
692 		return 0;
693 	return !!(pud_val(pud) & _REGION3_ENTRY_LARGE);
694 }
695 
696 #define pmd_leaf	pmd_large
697 static inline int pmd_large(pmd_t pmd)
698 {
699 	return (pmd_val(pmd) & _SEGMENT_ENTRY_LARGE) != 0;
700 }
701 
702 static inline int pmd_bad(pmd_t pmd)
703 {
704 	if ((pmd_val(pmd) & _SEGMENT_ENTRY_TYPE_MASK) > 0 || pmd_large(pmd))
705 		return 1;
706 	return (pmd_val(pmd) & ~_SEGMENT_ENTRY_BITS) != 0;
707 }
708 
709 static inline int pud_bad(pud_t pud)
710 {
711 	unsigned long type = pud_val(pud) & _REGION_ENTRY_TYPE_MASK;
712 
713 	if (type > _REGION_ENTRY_TYPE_R3 || pud_large(pud))
714 		return 1;
715 	if (type < _REGION_ENTRY_TYPE_R3)
716 		return 0;
717 	return (pud_val(pud) & ~_REGION_ENTRY_BITS) != 0;
718 }
719 
720 static inline int p4d_bad(p4d_t p4d)
721 {
722 	unsigned long type = p4d_val(p4d) & _REGION_ENTRY_TYPE_MASK;
723 
724 	if (type > _REGION_ENTRY_TYPE_R2)
725 		return 1;
726 	if (type < _REGION_ENTRY_TYPE_R2)
727 		return 0;
728 	return (p4d_val(p4d) & ~_REGION_ENTRY_BITS) != 0;
729 }
730 
731 static inline int pmd_present(pmd_t pmd)
732 {
733 	return pmd_val(pmd) != _SEGMENT_ENTRY_EMPTY;
734 }
735 
736 static inline int pmd_none(pmd_t pmd)
737 {
738 	return pmd_val(pmd) == _SEGMENT_ENTRY_EMPTY;
739 }
740 
741 #define pmd_write pmd_write
742 static inline int pmd_write(pmd_t pmd)
743 {
744 	return (pmd_val(pmd) & _SEGMENT_ENTRY_WRITE) != 0;
745 }
746 
747 #define pud_write pud_write
748 static inline int pud_write(pud_t pud)
749 {
750 	return (pud_val(pud) & _REGION3_ENTRY_WRITE) != 0;
751 }
752 
753 static inline int pmd_dirty(pmd_t pmd)
754 {
755 	return (pmd_val(pmd) & _SEGMENT_ENTRY_DIRTY) != 0;
756 }
757 
758 static inline int pmd_young(pmd_t pmd)
759 {
760 	return (pmd_val(pmd) & _SEGMENT_ENTRY_YOUNG) != 0;
761 }
762 
763 static inline int pte_present(pte_t pte)
764 {
765 	/* Bit pattern: (pte & 0x001) == 0x001 */
766 	return (pte_val(pte) & _PAGE_PRESENT) != 0;
767 }
768 
769 static inline int pte_none(pte_t pte)
770 {
771 	/* Bit pattern: pte == 0x400 */
772 	return pte_val(pte) == _PAGE_INVALID;
773 }
774 
775 static inline int pte_swap(pte_t pte)
776 {
777 	/* Bit pattern: (pte & 0x201) == 0x200 */
778 	return (pte_val(pte) & (_PAGE_PROTECT | _PAGE_PRESENT))
779 		== _PAGE_PROTECT;
780 }
781 
782 static inline int pte_special(pte_t pte)
783 {
784 	return (pte_val(pte) & _PAGE_SPECIAL);
785 }
786 
787 #define __HAVE_ARCH_PTE_SAME
788 static inline int pte_same(pte_t a, pte_t b)
789 {
790 	return pte_val(a) == pte_val(b);
791 }
792 
793 #ifdef CONFIG_NUMA_BALANCING
794 static inline int pte_protnone(pte_t pte)
795 {
796 	return pte_present(pte) && !(pte_val(pte) & _PAGE_READ);
797 }
798 
799 static inline int pmd_protnone(pmd_t pmd)
800 {
801 	/* pmd_large(pmd) implies pmd_present(pmd) */
802 	return pmd_large(pmd) && !(pmd_val(pmd) & _SEGMENT_ENTRY_READ);
803 }
804 #endif
805 
806 static inline int pte_soft_dirty(pte_t pte)
807 {
808 	return pte_val(pte) & _PAGE_SOFT_DIRTY;
809 }
810 #define pte_swp_soft_dirty pte_soft_dirty
811 
812 static inline pte_t pte_mksoft_dirty(pte_t pte)
813 {
814 	pte_val(pte) |= _PAGE_SOFT_DIRTY;
815 	return pte;
816 }
817 #define pte_swp_mksoft_dirty pte_mksoft_dirty
818 
819 static inline pte_t pte_clear_soft_dirty(pte_t pte)
820 {
821 	pte_val(pte) &= ~_PAGE_SOFT_DIRTY;
822 	return pte;
823 }
824 #define pte_swp_clear_soft_dirty pte_clear_soft_dirty
825 
826 static inline int pmd_soft_dirty(pmd_t pmd)
827 {
828 	return pmd_val(pmd) & _SEGMENT_ENTRY_SOFT_DIRTY;
829 }
830 
831 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
832 {
833 	pmd_val(pmd) |= _SEGMENT_ENTRY_SOFT_DIRTY;
834 	return pmd;
835 }
836 
837 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
838 {
839 	pmd_val(pmd) &= ~_SEGMENT_ENTRY_SOFT_DIRTY;
840 	return pmd;
841 }
842 
843 /*
844  * query functions pte_write/pte_dirty/pte_young only work if
845  * pte_present() is true. Undefined behaviour if not..
846  */
847 static inline int pte_write(pte_t pte)
848 {
849 	return (pte_val(pte) & _PAGE_WRITE) != 0;
850 }
851 
852 static inline int pte_dirty(pte_t pte)
853 {
854 	return (pte_val(pte) & _PAGE_DIRTY) != 0;
855 }
856 
857 static inline int pte_young(pte_t pte)
858 {
859 	return (pte_val(pte) & _PAGE_YOUNG) != 0;
860 }
861 
862 #define __HAVE_ARCH_PTE_UNUSED
863 static inline int pte_unused(pte_t pte)
864 {
865 	return pte_val(pte) & _PAGE_UNUSED;
866 }
867 
868 /*
869  * pgd/pmd/pte modification functions
870  */
871 
872 static inline void pgd_clear(pgd_t *pgd)
873 {
874 	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R1)
875 		pgd_val(*pgd) = _REGION1_ENTRY_EMPTY;
876 }
877 
878 static inline void p4d_clear(p4d_t *p4d)
879 {
880 	if ((p4d_val(*p4d) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
881 		p4d_val(*p4d) = _REGION2_ENTRY_EMPTY;
882 }
883 
884 static inline void pud_clear(pud_t *pud)
885 {
886 	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
887 		pud_val(*pud) = _REGION3_ENTRY_EMPTY;
888 }
889 
890 static inline void pmd_clear(pmd_t *pmdp)
891 {
892 	pmd_val(*pmdp) = _SEGMENT_ENTRY_EMPTY;
893 }
894 
895 static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
896 {
897 	pte_val(*ptep) = _PAGE_INVALID;
898 }
899 
900 /*
901  * The following pte modification functions only work if
902  * pte_present() is true. Undefined behaviour if not..
903  */
904 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
905 {
906 	pte_val(pte) &= _PAGE_CHG_MASK;
907 	pte_val(pte) |= pgprot_val(newprot);
908 	/*
909 	 * newprot for PAGE_NONE, PAGE_RO, PAGE_RX, PAGE_RW and PAGE_RWX
910 	 * has the invalid bit set, clear it again for readable, young pages
911 	 */
912 	if ((pte_val(pte) & _PAGE_YOUNG) && (pte_val(pte) & _PAGE_READ))
913 		pte_val(pte) &= ~_PAGE_INVALID;
914 	/*
915 	 * newprot for PAGE_RO, PAGE_RX, PAGE_RW and PAGE_RWX has the page
916 	 * protection bit set, clear it again for writable, dirty pages
917 	 */
918 	if ((pte_val(pte) & _PAGE_DIRTY) && (pte_val(pte) & _PAGE_WRITE))
919 		pte_val(pte) &= ~_PAGE_PROTECT;
920 	return pte;
921 }
922 
923 static inline pte_t pte_wrprotect(pte_t pte)
924 {
925 	pte_val(pte) &= ~_PAGE_WRITE;
926 	pte_val(pte) |= _PAGE_PROTECT;
927 	return pte;
928 }
929 
930 static inline pte_t pte_mkwrite(pte_t pte)
931 {
932 	pte_val(pte) |= _PAGE_WRITE;
933 	if (pte_val(pte) & _PAGE_DIRTY)
934 		pte_val(pte) &= ~_PAGE_PROTECT;
935 	return pte;
936 }
937 
938 static inline pte_t pte_mkclean(pte_t pte)
939 {
940 	pte_val(pte) &= ~_PAGE_DIRTY;
941 	pte_val(pte) |= _PAGE_PROTECT;
942 	return pte;
943 }
944 
945 static inline pte_t pte_mkdirty(pte_t pte)
946 {
947 	pte_val(pte) |= _PAGE_DIRTY | _PAGE_SOFT_DIRTY;
948 	if (pte_val(pte) & _PAGE_WRITE)
949 		pte_val(pte) &= ~_PAGE_PROTECT;
950 	return pte;
951 }
952 
953 static inline pte_t pte_mkold(pte_t pte)
954 {
955 	pte_val(pte) &= ~_PAGE_YOUNG;
956 	pte_val(pte) |= _PAGE_INVALID;
957 	return pte;
958 }
959 
960 static inline pte_t pte_mkyoung(pte_t pte)
961 {
962 	pte_val(pte) |= _PAGE_YOUNG;
963 	if (pte_val(pte) & _PAGE_READ)
964 		pte_val(pte) &= ~_PAGE_INVALID;
965 	return pte;
966 }
967 
968 static inline pte_t pte_mkspecial(pte_t pte)
969 {
970 	pte_val(pte) |= _PAGE_SPECIAL;
971 	return pte;
972 }
973 
974 #ifdef CONFIG_HUGETLB_PAGE
975 static inline pte_t pte_mkhuge(pte_t pte)
976 {
977 	pte_val(pte) |= _PAGE_LARGE;
978 	return pte;
979 }
980 #endif
981 
982 #define IPTE_GLOBAL	0
983 #define	IPTE_LOCAL	1
984 
985 #define IPTE_NODAT	0x400
986 #define IPTE_GUEST_ASCE	0x800
987 
988 static __always_inline void __ptep_ipte(unsigned long address, pte_t *ptep,
989 					unsigned long opt, unsigned long asce,
990 					int local)
991 {
992 	unsigned long pto = (unsigned long) ptep;
993 
994 	if (__builtin_constant_p(opt) && opt == 0) {
995 		/* Invalidation + TLB flush for the pte */
996 		asm volatile(
997 			"	.insn	rrf,0xb2210000,%[r1],%[r2],0,%[m4]"
998 			: "+m" (*ptep) : [r1] "a" (pto), [r2] "a" (address),
999 			  [m4] "i" (local));
1000 		return;
1001 	}
1002 
1003 	/* Invalidate ptes with options + TLB flush of the ptes */
1004 	opt = opt | (asce & _ASCE_ORIGIN);
1005 	asm volatile(
1006 		"	.insn	rrf,0xb2210000,%[r1],%[r2],%[r3],%[m4]"
1007 		: [r2] "+a" (address), [r3] "+a" (opt)
1008 		: [r1] "a" (pto), [m4] "i" (local) : "memory");
1009 }
1010 
1011 static __always_inline void __ptep_ipte_range(unsigned long address, int nr,
1012 					      pte_t *ptep, int local)
1013 {
1014 	unsigned long pto = (unsigned long) ptep;
1015 
1016 	/* Invalidate a range of ptes + TLB flush of the ptes */
1017 	do {
1018 		asm volatile(
1019 			"       .insn rrf,0xb2210000,%[r1],%[r2],%[r3],%[m4]"
1020 			: [r2] "+a" (address), [r3] "+a" (nr)
1021 			: [r1] "a" (pto), [m4] "i" (local) : "memory");
1022 	} while (nr != 255);
1023 }
1024 
1025 /*
1026  * This is hard to understand. ptep_get_and_clear and ptep_clear_flush
1027  * both clear the TLB for the unmapped pte. The reason is that
1028  * ptep_get_and_clear is used in common code (e.g. change_pte_range)
1029  * to modify an active pte. The sequence is
1030  *   1) ptep_get_and_clear
1031  *   2) set_pte_at
1032  *   3) flush_tlb_range
1033  * On s390 the tlb needs to get flushed with the modification of the pte
1034  * if the pte is active. The only way how this can be implemented is to
1035  * have ptep_get_and_clear do the tlb flush. In exchange flush_tlb_range
1036  * is a nop.
1037  */
1038 pte_t ptep_xchg_direct(struct mm_struct *, unsigned long, pte_t *, pte_t);
1039 pte_t ptep_xchg_lazy(struct mm_struct *, unsigned long, pte_t *, pte_t);
1040 
1041 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1042 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
1043 					    unsigned long addr, pte_t *ptep)
1044 {
1045 	pte_t pte = *ptep;
1046 
1047 	pte = ptep_xchg_direct(vma->vm_mm, addr, ptep, pte_mkold(pte));
1048 	return pte_young(pte);
1049 }
1050 
1051 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1052 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
1053 					 unsigned long address, pte_t *ptep)
1054 {
1055 	return ptep_test_and_clear_young(vma, address, ptep);
1056 }
1057 
1058 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1059 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
1060 				       unsigned long addr, pte_t *ptep)
1061 {
1062 	pte_t res;
1063 
1064 	res = ptep_xchg_lazy(mm, addr, ptep, __pte(_PAGE_INVALID));
1065 	if (mm_is_protected(mm) && pte_present(res))
1066 		uv_convert_from_secure(pte_val(res) & PAGE_MASK);
1067 	return res;
1068 }
1069 
1070 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1071 pte_t ptep_modify_prot_start(struct vm_area_struct *, unsigned long, pte_t *);
1072 void ptep_modify_prot_commit(struct vm_area_struct *, unsigned long,
1073 			     pte_t *, pte_t, pte_t);
1074 
1075 #define __HAVE_ARCH_PTEP_CLEAR_FLUSH
1076 static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
1077 				     unsigned long addr, pte_t *ptep)
1078 {
1079 	pte_t res;
1080 
1081 	res = ptep_xchg_direct(vma->vm_mm, addr, ptep, __pte(_PAGE_INVALID));
1082 	if (mm_is_protected(vma->vm_mm) && pte_present(res))
1083 		uv_convert_from_secure(pte_val(res) & PAGE_MASK);
1084 	return res;
1085 }
1086 
1087 /*
1088  * The batched pte unmap code uses ptep_get_and_clear_full to clear the
1089  * ptes. Here an optimization is possible. tlb_gather_mmu flushes all
1090  * tlbs of an mm if it can guarantee that the ptes of the mm_struct
1091  * cannot be accessed while the batched unmap is running. In this case
1092  * full==1 and a simple pte_clear is enough. See tlb.h.
1093  */
1094 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
1095 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
1096 					    unsigned long addr,
1097 					    pte_t *ptep, int full)
1098 {
1099 	pte_t res;
1100 
1101 	if (full) {
1102 		res = *ptep;
1103 		*ptep = __pte(_PAGE_INVALID);
1104 	} else {
1105 		res = ptep_xchg_lazy(mm, addr, ptep, __pte(_PAGE_INVALID));
1106 	}
1107 	if (mm_is_protected(mm) && pte_present(res))
1108 		uv_convert_from_secure(pte_val(res) & PAGE_MASK);
1109 	return res;
1110 }
1111 
1112 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1113 static inline void ptep_set_wrprotect(struct mm_struct *mm,
1114 				      unsigned long addr, pte_t *ptep)
1115 {
1116 	pte_t pte = *ptep;
1117 
1118 	if (pte_write(pte))
1119 		ptep_xchg_lazy(mm, addr, ptep, pte_wrprotect(pte));
1120 }
1121 
1122 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1123 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
1124 					unsigned long addr, pte_t *ptep,
1125 					pte_t entry, int dirty)
1126 {
1127 	if (pte_same(*ptep, entry))
1128 		return 0;
1129 	ptep_xchg_direct(vma->vm_mm, addr, ptep, entry);
1130 	return 1;
1131 }
1132 
1133 /*
1134  * Additional functions to handle KVM guest page tables
1135  */
1136 void ptep_set_pte_at(struct mm_struct *mm, unsigned long addr,
1137 		     pte_t *ptep, pte_t entry);
1138 void ptep_set_notify(struct mm_struct *mm, unsigned long addr, pte_t *ptep);
1139 void ptep_notify(struct mm_struct *mm, unsigned long addr,
1140 		 pte_t *ptep, unsigned long bits);
1141 int ptep_force_prot(struct mm_struct *mm, unsigned long gaddr,
1142 		    pte_t *ptep, int prot, unsigned long bit);
1143 void ptep_zap_unused(struct mm_struct *mm, unsigned long addr,
1144 		     pte_t *ptep , int reset);
1145 void ptep_zap_key(struct mm_struct *mm, unsigned long addr, pte_t *ptep);
1146 int ptep_shadow_pte(struct mm_struct *mm, unsigned long saddr,
1147 		    pte_t *sptep, pte_t *tptep, pte_t pte);
1148 void ptep_unshadow_pte(struct mm_struct *mm, unsigned long saddr, pte_t *ptep);
1149 
1150 bool ptep_test_and_clear_uc(struct mm_struct *mm, unsigned long address,
1151 			    pte_t *ptep);
1152 int set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
1153 			  unsigned char key, bool nq);
1154 int cond_set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
1155 			       unsigned char key, unsigned char *oldkey,
1156 			       bool nq, bool mr, bool mc);
1157 int reset_guest_reference_bit(struct mm_struct *mm, unsigned long addr);
1158 int get_guest_storage_key(struct mm_struct *mm, unsigned long addr,
1159 			  unsigned char *key);
1160 
1161 int set_pgste_bits(struct mm_struct *mm, unsigned long addr,
1162 				unsigned long bits, unsigned long value);
1163 int get_pgste(struct mm_struct *mm, unsigned long hva, unsigned long *pgstep);
1164 int pgste_perform_essa(struct mm_struct *mm, unsigned long hva, int orc,
1165 			unsigned long *oldpte, unsigned long *oldpgste);
1166 void gmap_pmdp_csp(struct mm_struct *mm, unsigned long vmaddr);
1167 void gmap_pmdp_invalidate(struct mm_struct *mm, unsigned long vmaddr);
1168 void gmap_pmdp_idte_local(struct mm_struct *mm, unsigned long vmaddr);
1169 void gmap_pmdp_idte_global(struct mm_struct *mm, unsigned long vmaddr);
1170 
1171 #define pgprot_writecombine	pgprot_writecombine
1172 pgprot_t pgprot_writecombine(pgprot_t prot);
1173 
1174 #define pgprot_writethrough	pgprot_writethrough
1175 pgprot_t pgprot_writethrough(pgprot_t prot);
1176 
1177 /*
1178  * Certain architectures need to do special things when PTEs
1179  * within a page table are directly modified.  Thus, the following
1180  * hook is made available.
1181  */
1182 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
1183 			      pte_t *ptep, pte_t entry)
1184 {
1185 	if (pte_present(entry))
1186 		pte_val(entry) &= ~_PAGE_UNUSED;
1187 	if (mm_has_pgste(mm))
1188 		ptep_set_pte_at(mm, addr, ptep, entry);
1189 	else
1190 		*ptep = entry;
1191 }
1192 
1193 /*
1194  * Conversion functions: convert a page and protection to a page entry,
1195  * and a page entry and page directory to the page they refer to.
1196  */
1197 static inline pte_t mk_pte_phys(unsigned long physpage, pgprot_t pgprot)
1198 {
1199 	pte_t __pte;
1200 
1201 	pte_val(__pte) = physpage | pgprot_val(pgprot);
1202 	if (!MACHINE_HAS_NX)
1203 		pte_val(__pte) &= ~_PAGE_NOEXEC;
1204 	return pte_mkyoung(__pte);
1205 }
1206 
1207 static inline pte_t mk_pte(struct page *page, pgprot_t pgprot)
1208 {
1209 	unsigned long physpage = page_to_phys(page);
1210 	pte_t __pte = mk_pte_phys(physpage, pgprot);
1211 
1212 	if (pte_write(__pte) && PageDirty(page))
1213 		__pte = pte_mkdirty(__pte);
1214 	return __pte;
1215 }
1216 
1217 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD-1))
1218 #define p4d_index(address) (((address) >> P4D_SHIFT) & (PTRS_PER_P4D-1))
1219 #define pud_index(address) (((address) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
1220 #define pmd_index(address) (((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1))
1221 
1222 #define p4d_deref(pud) ((unsigned long)__va(p4d_val(pud) & _REGION_ENTRY_ORIGIN))
1223 #define pgd_deref(pgd) ((unsigned long)__va(pgd_val(pgd) & _REGION_ENTRY_ORIGIN))
1224 
1225 static inline unsigned long pmd_deref(pmd_t pmd)
1226 {
1227 	unsigned long origin_mask;
1228 
1229 	origin_mask = _SEGMENT_ENTRY_ORIGIN;
1230 	if (pmd_large(pmd))
1231 		origin_mask = _SEGMENT_ENTRY_ORIGIN_LARGE;
1232 	return (unsigned long)__va(pmd_val(pmd) & origin_mask);
1233 }
1234 
1235 static inline unsigned long pmd_pfn(pmd_t pmd)
1236 {
1237 	return __pa(pmd_deref(pmd)) >> PAGE_SHIFT;
1238 }
1239 
1240 static inline unsigned long pud_deref(pud_t pud)
1241 {
1242 	unsigned long origin_mask;
1243 
1244 	origin_mask = _REGION_ENTRY_ORIGIN;
1245 	if (pud_large(pud))
1246 		origin_mask = _REGION3_ENTRY_ORIGIN_LARGE;
1247 	return (unsigned long)__va(pud_val(pud) & origin_mask);
1248 }
1249 
1250 static inline unsigned long pud_pfn(pud_t pud)
1251 {
1252 	return __pa(pud_deref(pud)) >> PAGE_SHIFT;
1253 }
1254 
1255 /*
1256  * The pgd_offset function *always* adds the index for the top-level
1257  * region/segment table. This is done to get a sequence like the
1258  * following to work:
1259  *	pgdp = pgd_offset(current->mm, addr);
1260  *	pgd = READ_ONCE(*pgdp);
1261  *	p4dp = p4d_offset(&pgd, addr);
1262  *	...
1263  * The subsequent p4d_offset, pud_offset and pmd_offset functions
1264  * only add an index if they dereferenced the pointer.
1265  */
1266 static inline pgd_t *pgd_offset_raw(pgd_t *pgd, unsigned long address)
1267 {
1268 	unsigned long rste;
1269 	unsigned int shift;
1270 
1271 	/* Get the first entry of the top level table */
1272 	rste = pgd_val(*pgd);
1273 	/* Pick up the shift from the table type of the first entry */
1274 	shift = ((rste & _REGION_ENTRY_TYPE_MASK) >> 2) * 11 + 20;
1275 	return pgd + ((address >> shift) & (PTRS_PER_PGD - 1));
1276 }
1277 
1278 #define pgd_offset(mm, address) pgd_offset_raw(READ_ONCE((mm)->pgd), address)
1279 
1280 static inline p4d_t *p4d_offset_lockless(pgd_t *pgdp, pgd_t pgd, unsigned long address)
1281 {
1282 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) >= _REGION_ENTRY_TYPE_R1)
1283 		return (p4d_t *) pgd_deref(pgd) + p4d_index(address);
1284 	return (p4d_t *) pgdp;
1285 }
1286 #define p4d_offset_lockless p4d_offset_lockless
1287 
1288 static inline p4d_t *p4d_offset(pgd_t *pgdp, unsigned long address)
1289 {
1290 	return p4d_offset_lockless(pgdp, *pgdp, address);
1291 }
1292 
1293 static inline pud_t *pud_offset_lockless(p4d_t *p4dp, p4d_t p4d, unsigned long address)
1294 {
1295 	if ((p4d_val(p4d) & _REGION_ENTRY_TYPE_MASK) >= _REGION_ENTRY_TYPE_R2)
1296 		return (pud_t *) p4d_deref(p4d) + pud_index(address);
1297 	return (pud_t *) p4dp;
1298 }
1299 #define pud_offset_lockless pud_offset_lockless
1300 
1301 static inline pud_t *pud_offset(p4d_t *p4dp, unsigned long address)
1302 {
1303 	return pud_offset_lockless(p4dp, *p4dp, address);
1304 }
1305 #define pud_offset pud_offset
1306 
1307 static inline pmd_t *pmd_offset_lockless(pud_t *pudp, pud_t pud, unsigned long address)
1308 {
1309 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) >= _REGION_ENTRY_TYPE_R3)
1310 		return (pmd_t *) pud_deref(pud) + pmd_index(address);
1311 	return (pmd_t *) pudp;
1312 }
1313 #define pmd_offset_lockless pmd_offset_lockless
1314 
1315 static inline pmd_t *pmd_offset(pud_t *pudp, unsigned long address)
1316 {
1317 	return pmd_offset_lockless(pudp, *pudp, address);
1318 }
1319 #define pmd_offset pmd_offset
1320 
1321 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
1322 {
1323 	return (unsigned long) pmd_deref(pmd);
1324 }
1325 
1326 static inline bool gup_fast_permitted(unsigned long start, unsigned long end)
1327 {
1328 	return end <= current->mm->context.asce_limit;
1329 }
1330 #define gup_fast_permitted gup_fast_permitted
1331 
1332 #define pfn_pte(pfn, pgprot)	mk_pte_phys(((pfn) << PAGE_SHIFT), (pgprot))
1333 #define pte_pfn(x) (pte_val(x) >> PAGE_SHIFT)
1334 #define pte_page(x) pfn_to_page(pte_pfn(x))
1335 
1336 #define pmd_page(pmd) pfn_to_page(pmd_pfn(pmd))
1337 #define pud_page(pud) pfn_to_page(pud_pfn(pud))
1338 #define p4d_page(p4d) pfn_to_page(p4d_pfn(p4d))
1339 #define pgd_page(pgd) pfn_to_page(pgd_pfn(pgd))
1340 
1341 static inline pmd_t pmd_wrprotect(pmd_t pmd)
1342 {
1343 	pmd_val(pmd) &= ~_SEGMENT_ENTRY_WRITE;
1344 	pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
1345 	return pmd;
1346 }
1347 
1348 static inline pmd_t pmd_mkwrite(pmd_t pmd)
1349 {
1350 	pmd_val(pmd) |= _SEGMENT_ENTRY_WRITE;
1351 	if (pmd_val(pmd) & _SEGMENT_ENTRY_DIRTY)
1352 		pmd_val(pmd) &= ~_SEGMENT_ENTRY_PROTECT;
1353 	return pmd;
1354 }
1355 
1356 static inline pmd_t pmd_mkclean(pmd_t pmd)
1357 {
1358 	pmd_val(pmd) &= ~_SEGMENT_ENTRY_DIRTY;
1359 	pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
1360 	return pmd;
1361 }
1362 
1363 static inline pmd_t pmd_mkdirty(pmd_t pmd)
1364 {
1365 	pmd_val(pmd) |= _SEGMENT_ENTRY_DIRTY | _SEGMENT_ENTRY_SOFT_DIRTY;
1366 	if (pmd_val(pmd) & _SEGMENT_ENTRY_WRITE)
1367 		pmd_val(pmd) &= ~_SEGMENT_ENTRY_PROTECT;
1368 	return pmd;
1369 }
1370 
1371 static inline pud_t pud_wrprotect(pud_t pud)
1372 {
1373 	pud_val(pud) &= ~_REGION3_ENTRY_WRITE;
1374 	pud_val(pud) |= _REGION_ENTRY_PROTECT;
1375 	return pud;
1376 }
1377 
1378 static inline pud_t pud_mkwrite(pud_t pud)
1379 {
1380 	pud_val(pud) |= _REGION3_ENTRY_WRITE;
1381 	if (pud_val(pud) & _REGION3_ENTRY_DIRTY)
1382 		pud_val(pud) &= ~_REGION_ENTRY_PROTECT;
1383 	return pud;
1384 }
1385 
1386 static inline pud_t pud_mkclean(pud_t pud)
1387 {
1388 	pud_val(pud) &= ~_REGION3_ENTRY_DIRTY;
1389 	pud_val(pud) |= _REGION_ENTRY_PROTECT;
1390 	return pud;
1391 }
1392 
1393 static inline pud_t pud_mkdirty(pud_t pud)
1394 {
1395 	pud_val(pud) |= _REGION3_ENTRY_DIRTY | _REGION3_ENTRY_SOFT_DIRTY;
1396 	if (pud_val(pud) & _REGION3_ENTRY_WRITE)
1397 		pud_val(pud) &= ~_REGION_ENTRY_PROTECT;
1398 	return pud;
1399 }
1400 
1401 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE)
1402 static inline unsigned long massage_pgprot_pmd(pgprot_t pgprot)
1403 {
1404 	/*
1405 	 * pgprot is PAGE_NONE, PAGE_RO, PAGE_RX, PAGE_RW or PAGE_RWX
1406 	 * (see __Pxxx / __Sxxx). Convert to segment table entry format.
1407 	 */
1408 	if (pgprot_val(pgprot) == pgprot_val(PAGE_NONE))
1409 		return pgprot_val(SEGMENT_NONE);
1410 	if (pgprot_val(pgprot) == pgprot_val(PAGE_RO))
1411 		return pgprot_val(SEGMENT_RO);
1412 	if (pgprot_val(pgprot) == pgprot_val(PAGE_RX))
1413 		return pgprot_val(SEGMENT_RX);
1414 	if (pgprot_val(pgprot) == pgprot_val(PAGE_RW))
1415 		return pgprot_val(SEGMENT_RW);
1416 	return pgprot_val(SEGMENT_RWX);
1417 }
1418 
1419 static inline pmd_t pmd_mkyoung(pmd_t pmd)
1420 {
1421 	pmd_val(pmd) |= _SEGMENT_ENTRY_YOUNG;
1422 	if (pmd_val(pmd) & _SEGMENT_ENTRY_READ)
1423 		pmd_val(pmd) &= ~_SEGMENT_ENTRY_INVALID;
1424 	return pmd;
1425 }
1426 
1427 static inline pmd_t pmd_mkold(pmd_t pmd)
1428 {
1429 	pmd_val(pmd) &= ~_SEGMENT_ENTRY_YOUNG;
1430 	pmd_val(pmd) |= _SEGMENT_ENTRY_INVALID;
1431 	return pmd;
1432 }
1433 
1434 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
1435 {
1436 	pmd_val(pmd) &= _SEGMENT_ENTRY_ORIGIN_LARGE |
1437 		_SEGMENT_ENTRY_DIRTY | _SEGMENT_ENTRY_YOUNG |
1438 		_SEGMENT_ENTRY_LARGE | _SEGMENT_ENTRY_SOFT_DIRTY;
1439 	pmd_val(pmd) |= massage_pgprot_pmd(newprot);
1440 	if (!(pmd_val(pmd) & _SEGMENT_ENTRY_DIRTY))
1441 		pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
1442 	if (!(pmd_val(pmd) & _SEGMENT_ENTRY_YOUNG))
1443 		pmd_val(pmd) |= _SEGMENT_ENTRY_INVALID;
1444 	return pmd;
1445 }
1446 
1447 static inline pmd_t mk_pmd_phys(unsigned long physpage, pgprot_t pgprot)
1448 {
1449 	pmd_t __pmd;
1450 	pmd_val(__pmd) = physpage + massage_pgprot_pmd(pgprot);
1451 	return __pmd;
1452 }
1453 
1454 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLB_PAGE */
1455 
1456 static inline void __pmdp_csp(pmd_t *pmdp)
1457 {
1458 	csp((unsigned int *)pmdp + 1, pmd_val(*pmdp),
1459 	    pmd_val(*pmdp) | _SEGMENT_ENTRY_INVALID);
1460 }
1461 
1462 #define IDTE_GLOBAL	0
1463 #define IDTE_LOCAL	1
1464 
1465 #define IDTE_PTOA	0x0800
1466 #define IDTE_NODAT	0x1000
1467 #define IDTE_GUEST_ASCE	0x2000
1468 
1469 static __always_inline void __pmdp_idte(unsigned long addr, pmd_t *pmdp,
1470 					unsigned long opt, unsigned long asce,
1471 					int local)
1472 {
1473 	unsigned long sto;
1474 
1475 	sto = (unsigned long) pmdp - pmd_index(addr) * sizeof(pmd_t);
1476 	if (__builtin_constant_p(opt) && opt == 0) {
1477 		/* flush without guest asce */
1478 		asm volatile(
1479 			"	.insn	rrf,0xb98e0000,%[r1],%[r2],0,%[m4]"
1480 			: "+m" (*pmdp)
1481 			: [r1] "a" (sto), [r2] "a" ((addr & HPAGE_MASK)),
1482 			  [m4] "i" (local)
1483 			: "cc" );
1484 	} else {
1485 		/* flush with guest asce */
1486 		asm volatile(
1487 			"	.insn	rrf,0xb98e0000,%[r1],%[r2],%[r3],%[m4]"
1488 			: "+m" (*pmdp)
1489 			: [r1] "a" (sto), [r2] "a" ((addr & HPAGE_MASK) | opt),
1490 			  [r3] "a" (asce), [m4] "i" (local)
1491 			: "cc" );
1492 	}
1493 }
1494 
1495 static __always_inline void __pudp_idte(unsigned long addr, pud_t *pudp,
1496 					unsigned long opt, unsigned long asce,
1497 					int local)
1498 {
1499 	unsigned long r3o;
1500 
1501 	r3o = (unsigned long) pudp - pud_index(addr) * sizeof(pud_t);
1502 	r3o |= _ASCE_TYPE_REGION3;
1503 	if (__builtin_constant_p(opt) && opt == 0) {
1504 		/* flush without guest asce */
1505 		asm volatile(
1506 			"	.insn	rrf,0xb98e0000,%[r1],%[r2],0,%[m4]"
1507 			: "+m" (*pudp)
1508 			: [r1] "a" (r3o), [r2] "a" ((addr & PUD_MASK)),
1509 			  [m4] "i" (local)
1510 			: "cc");
1511 	} else {
1512 		/* flush with guest asce */
1513 		asm volatile(
1514 			"	.insn	rrf,0xb98e0000,%[r1],%[r2],%[r3],%[m4]"
1515 			: "+m" (*pudp)
1516 			: [r1] "a" (r3o), [r2] "a" ((addr & PUD_MASK) | opt),
1517 			  [r3] "a" (asce), [m4] "i" (local)
1518 			: "cc" );
1519 	}
1520 }
1521 
1522 pmd_t pmdp_xchg_direct(struct mm_struct *, unsigned long, pmd_t *, pmd_t);
1523 pmd_t pmdp_xchg_lazy(struct mm_struct *, unsigned long, pmd_t *, pmd_t);
1524 pud_t pudp_xchg_direct(struct mm_struct *, unsigned long, pud_t *, pud_t);
1525 
1526 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1527 
1528 #define __HAVE_ARCH_PGTABLE_DEPOSIT
1529 void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
1530 				pgtable_t pgtable);
1531 
1532 #define __HAVE_ARCH_PGTABLE_WITHDRAW
1533 pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
1534 
1535 #define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
1536 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
1537 					unsigned long addr, pmd_t *pmdp,
1538 					pmd_t entry, int dirty)
1539 {
1540 	VM_BUG_ON(addr & ~HPAGE_MASK);
1541 
1542 	entry = pmd_mkyoung(entry);
1543 	if (dirty)
1544 		entry = pmd_mkdirty(entry);
1545 	if (pmd_val(*pmdp) == pmd_val(entry))
1546 		return 0;
1547 	pmdp_xchg_direct(vma->vm_mm, addr, pmdp, entry);
1548 	return 1;
1549 }
1550 
1551 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1552 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1553 					    unsigned long addr, pmd_t *pmdp)
1554 {
1555 	pmd_t pmd = *pmdp;
1556 
1557 	pmd = pmdp_xchg_direct(vma->vm_mm, addr, pmdp, pmd_mkold(pmd));
1558 	return pmd_young(pmd);
1559 }
1560 
1561 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
1562 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
1563 					 unsigned long addr, pmd_t *pmdp)
1564 {
1565 	VM_BUG_ON(addr & ~HPAGE_MASK);
1566 	return pmdp_test_and_clear_young(vma, addr, pmdp);
1567 }
1568 
1569 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1570 			      pmd_t *pmdp, pmd_t entry)
1571 {
1572 	if (!MACHINE_HAS_NX)
1573 		pmd_val(entry) &= ~_SEGMENT_ENTRY_NOEXEC;
1574 	*pmdp = entry;
1575 }
1576 
1577 static inline pmd_t pmd_mkhuge(pmd_t pmd)
1578 {
1579 	pmd_val(pmd) |= _SEGMENT_ENTRY_LARGE;
1580 	pmd_val(pmd) |= _SEGMENT_ENTRY_YOUNG;
1581 	pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
1582 	return pmd;
1583 }
1584 
1585 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1586 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
1587 					    unsigned long addr, pmd_t *pmdp)
1588 {
1589 	return pmdp_xchg_direct(mm, addr, pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
1590 }
1591 
1592 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
1593 static inline pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
1594 						 unsigned long addr,
1595 						 pmd_t *pmdp, int full)
1596 {
1597 	if (full) {
1598 		pmd_t pmd = *pmdp;
1599 		*pmdp = __pmd(_SEGMENT_ENTRY_EMPTY);
1600 		return pmd;
1601 	}
1602 	return pmdp_xchg_lazy(vma->vm_mm, addr, pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
1603 }
1604 
1605 #define __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
1606 static inline pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
1607 					  unsigned long addr, pmd_t *pmdp)
1608 {
1609 	return pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
1610 }
1611 
1612 #define __HAVE_ARCH_PMDP_INVALIDATE
1613 static inline pmd_t pmdp_invalidate(struct vm_area_struct *vma,
1614 				   unsigned long addr, pmd_t *pmdp)
1615 {
1616 	pmd_t pmd = __pmd(pmd_val(*pmdp) | _SEGMENT_ENTRY_INVALID);
1617 
1618 	return pmdp_xchg_direct(vma->vm_mm, addr, pmdp, pmd);
1619 }
1620 
1621 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1622 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1623 				      unsigned long addr, pmd_t *pmdp)
1624 {
1625 	pmd_t pmd = *pmdp;
1626 
1627 	if (pmd_write(pmd))
1628 		pmd = pmdp_xchg_lazy(mm, addr, pmdp, pmd_wrprotect(pmd));
1629 }
1630 
1631 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
1632 					unsigned long address,
1633 					pmd_t *pmdp)
1634 {
1635 	return pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
1636 }
1637 #define pmdp_collapse_flush pmdp_collapse_flush
1638 
1639 #define pfn_pmd(pfn, pgprot)	mk_pmd_phys(((pfn) << PAGE_SHIFT), (pgprot))
1640 #define mk_pmd(page, pgprot)	pfn_pmd(page_to_pfn(page), (pgprot))
1641 
1642 static inline int pmd_trans_huge(pmd_t pmd)
1643 {
1644 	return pmd_val(pmd) & _SEGMENT_ENTRY_LARGE;
1645 }
1646 
1647 #define has_transparent_hugepage has_transparent_hugepage
1648 static inline int has_transparent_hugepage(void)
1649 {
1650 	return MACHINE_HAS_EDAT1 ? 1 : 0;
1651 }
1652 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1653 
1654 /*
1655  * 64 bit swap entry format:
1656  * A page-table entry has some bits we have to treat in a special way.
1657  * Bits 52 and bit 55 have to be zero, otherwise a specification
1658  * exception will occur instead of a page translation exception. The
1659  * specification exception has the bad habit not to store necessary
1660  * information in the lowcore.
1661  * Bits 54 and 63 are used to indicate the page type.
1662  * A swap pte is indicated by bit pattern (pte & 0x201) == 0x200
1663  * This leaves the bits 0-51 and bits 56-62 to store type and offset.
1664  * We use the 5 bits from 57-61 for the type and the 52 bits from 0-51
1665  * for the offset.
1666  * |			  offset			|01100|type |00|
1667  * |0000000000111111111122222222223333333333444444444455|55555|55566|66|
1668  * |0123456789012345678901234567890123456789012345678901|23456|78901|23|
1669  */
1670 
1671 #define __SWP_OFFSET_MASK	((1UL << 52) - 1)
1672 #define __SWP_OFFSET_SHIFT	12
1673 #define __SWP_TYPE_MASK		((1UL << 5) - 1)
1674 #define __SWP_TYPE_SHIFT	2
1675 
1676 static inline pte_t mk_swap_pte(unsigned long type, unsigned long offset)
1677 {
1678 	pte_t pte;
1679 
1680 	pte_val(pte) = _PAGE_INVALID | _PAGE_PROTECT;
1681 	pte_val(pte) |= (offset & __SWP_OFFSET_MASK) << __SWP_OFFSET_SHIFT;
1682 	pte_val(pte) |= (type & __SWP_TYPE_MASK) << __SWP_TYPE_SHIFT;
1683 	return pte;
1684 }
1685 
1686 static inline unsigned long __swp_type(swp_entry_t entry)
1687 {
1688 	return (entry.val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK;
1689 }
1690 
1691 static inline unsigned long __swp_offset(swp_entry_t entry)
1692 {
1693 	return (entry.val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK;
1694 }
1695 
1696 static inline swp_entry_t __swp_entry(unsigned long type, unsigned long offset)
1697 {
1698 	return (swp_entry_t) { pte_val(mk_swap_pte(type, offset)) };
1699 }
1700 
1701 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
1702 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
1703 
1704 #define kern_addr_valid(addr)   (1)
1705 
1706 extern int vmem_add_mapping(unsigned long start, unsigned long size);
1707 extern void vmem_remove_mapping(unsigned long start, unsigned long size);
1708 extern int s390_enable_sie(void);
1709 extern int s390_enable_skey(void);
1710 extern void s390_reset_cmma(struct mm_struct *mm);
1711 
1712 /* s390 has a private copy of get unmapped area to deal with cache synonyms */
1713 #define HAVE_ARCH_UNMAPPED_AREA
1714 #define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1715 
1716 #endif /* _S390_PAGE_H */
1717