xref: /openbmc/linux/arch/s390/include/asm/pgtable.h (revision 6548d543)
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/sections.h>
21 #include <asm/bug.h>
22 #include <asm/page.h>
23 #include <asm/uv.h>
24 
25 extern pgd_t swapper_pg_dir[];
26 extern pgd_t invalid_pg_dir[];
27 extern void paging_init(void);
28 extern unsigned long s390_invalid_asce;
29 
30 enum {
31 	PG_DIRECT_MAP_4K = 0,
32 	PG_DIRECT_MAP_1M,
33 	PG_DIRECT_MAP_2G,
34 	PG_DIRECT_MAP_MAX
35 };
36 
37 extern atomic_long_t __bootdata_preserved(direct_pages_count[PG_DIRECT_MAP_MAX]);
38 
39 static inline void update_page_count(int level, long count)
40 {
41 	if (IS_ENABLED(CONFIG_PROC_FS))
42 		atomic_long_add(count, &direct_pages_count[level]);
43 }
44 
45 /*
46  * The S390 doesn't have any external MMU info: the kernel page
47  * tables contain all the necessary information.
48  */
49 #define update_mmu_cache(vma, address, ptep)     do { } while (0)
50 #define update_mmu_cache_range(vmf, vma, addr, ptep, nr) 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 pte_ERROR(e) \
69 	pr_err("%s:%d: bad pte %016lx.\n", __FILE__, __LINE__, pte_val(e))
70 #define pmd_ERROR(e) \
71 	pr_err("%s:%d: bad pmd %016lx.\n", __FILE__, __LINE__, pmd_val(e))
72 #define pud_ERROR(e) \
73 	pr_err("%s:%d: bad pud %016lx.\n", __FILE__, __LINE__, pud_val(e))
74 #define p4d_ERROR(e) \
75 	pr_err("%s:%d: bad p4d %016lx.\n", __FILE__, __LINE__, p4d_val(e))
76 #define pgd_ERROR(e) \
77 	pr_err("%s:%d: bad pgd %016lx.\n", __FILE__, __LINE__, pgd_val(e))
78 
79 /*
80  * The vmalloc and module area will always be on the topmost area of the
81  * kernel mapping. 512GB are reserved for vmalloc by default.
82  * At the top of the vmalloc area a 2GB area is reserved where modules
83  * will reside. That makes sure that inter module branches always
84  * happen without trampolines and in addition the placement within a
85  * 2GB frame is branch prediction unit friendly.
86  */
87 extern unsigned long __bootdata_preserved(VMALLOC_START);
88 extern unsigned long __bootdata_preserved(VMALLOC_END);
89 #define VMALLOC_DEFAULT_SIZE	((512UL << 30) - MODULES_LEN)
90 extern struct page *__bootdata_preserved(vmemmap);
91 extern unsigned long __bootdata_preserved(vmemmap_size);
92 
93 extern unsigned long __bootdata_preserved(MODULES_VADDR);
94 extern unsigned long __bootdata_preserved(MODULES_END);
95 #define MODULES_VADDR	MODULES_VADDR
96 #define MODULES_END	MODULES_END
97 #define MODULES_LEN	(1UL << 31)
98 
99 static inline int is_module_addr(void *addr)
100 {
101 	BUILD_BUG_ON(MODULES_LEN > (1UL << 31));
102 	if (addr < (void *)MODULES_VADDR)
103 		return 0;
104 	if (addr > (void *)MODULES_END)
105 		return 0;
106 	return 1;
107 }
108 
109 /*
110  * A 64 bit pagetable entry of S390 has following format:
111  * |			 PFRA			      |0IPC|  OS  |
112  * 0000000000111111111122222222223333333333444444444455555555556666
113  * 0123456789012345678901234567890123456789012345678901234567890123
114  *
115  * I Page-Invalid Bit:    Page is not available for address-translation
116  * P Page-Protection Bit: Store access not possible for page
117  * C Change-bit override: HW is not required to set change bit
118  *
119  * A 64 bit segmenttable entry of S390 has following format:
120  * |        P-table origin                              |      TT
121  * 0000000000111111111122222222223333333333444444444455555555556666
122  * 0123456789012345678901234567890123456789012345678901234567890123
123  *
124  * I Segment-Invalid Bit:    Segment is not available for address-translation
125  * C Common-Segment Bit:     Segment is not private (PoP 3-30)
126  * P Page-Protection Bit: Store access not possible for page
127  * TT Type 00
128  *
129  * A 64 bit region table entry of S390 has following format:
130  * |        S-table origin                             |   TF  TTTL
131  * 0000000000111111111122222222223333333333444444444455555555556666
132  * 0123456789012345678901234567890123456789012345678901234567890123
133  *
134  * I Segment-Invalid Bit:    Segment is not available for address-translation
135  * TT Type 01
136  * TF
137  * TL Table length
138  *
139  * The 64 bit regiontable origin of S390 has following format:
140  * |      region table origon                          |       DTTL
141  * 0000000000111111111122222222223333333333444444444455555555556666
142  * 0123456789012345678901234567890123456789012345678901234567890123
143  *
144  * X Space-Switch event:
145  * G Segment-Invalid Bit:
146  * P Private-Space Bit:
147  * S Storage-Alteration:
148  * R Real space
149  * TL Table-Length:
150  *
151  * A storage key has the following format:
152  * | ACC |F|R|C|0|
153  *  0   3 4 5 6 7
154  * ACC: access key
155  * F  : fetch protection bit
156  * R  : referenced bit
157  * C  : changed bit
158  */
159 
160 /* Hardware bits in the page table entry */
161 #define _PAGE_NOEXEC	0x100		/* HW no-execute bit  */
162 #define _PAGE_PROTECT	0x200		/* HW read-only bit  */
163 #define _PAGE_INVALID	0x400		/* HW invalid bit    */
164 #define _PAGE_LARGE	0x800		/* Bit to mark a large pte */
165 
166 /* Software bits in the page table entry */
167 #define _PAGE_PRESENT	0x001		/* SW pte present bit */
168 #define _PAGE_YOUNG	0x004		/* SW pte young bit */
169 #define _PAGE_DIRTY	0x008		/* SW pte dirty bit */
170 #define _PAGE_READ	0x010		/* SW pte read bit */
171 #define _PAGE_WRITE	0x020		/* SW pte write bit */
172 #define _PAGE_SPECIAL	0x040		/* SW associated with special page */
173 #define _PAGE_UNUSED	0x080		/* SW bit for pgste usage state */
174 
175 #ifdef CONFIG_MEM_SOFT_DIRTY
176 #define _PAGE_SOFT_DIRTY 0x002		/* SW pte soft dirty bit */
177 #else
178 #define _PAGE_SOFT_DIRTY 0x000
179 #endif
180 
181 #define _PAGE_SW_BITS	0xffUL		/* All SW bits */
182 
183 #define _PAGE_SWP_EXCLUSIVE _PAGE_LARGE	/* SW pte exclusive swap bit */
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  * Mask of bits that must not be changed with RDP. Allow only _PAGE_PROTECT
191  * HW bit and all SW bits.
192  */
193 #define _PAGE_RDP_MASK		~(_PAGE_PROTECT | _PAGE_SW_BITS)
194 
195 /*
196  * handle_pte_fault uses pte_present and pte_none to find out the pte type
197  * WITHOUT holding the page table lock. The _PAGE_PRESENT bit is used to
198  * distinguish present from not-present ptes. It is changed only with the page
199  * table lock held.
200  *
201  * The following table gives the different possible bit combinations for
202  * the pte hardware and software bits in the last 12 bits of a pte
203  * (. unassigned bit, x don't care, t swap type):
204  *
205  *				842100000000
206  *				000084210000
207  *				000000008421
208  *				.IR.uswrdy.p
209  * empty			.10.00000000
210  * swap				.11..ttttt.0
211  * prot-none, clean, old	.11.xx0000.1
212  * prot-none, clean, young	.11.xx0001.1
213  * prot-none, dirty, old	.11.xx0010.1
214  * prot-none, dirty, young	.11.xx0011.1
215  * read-only, clean, old	.11.xx0100.1
216  * read-only, clean, young	.01.xx0101.1
217  * read-only, dirty, old	.11.xx0110.1
218  * read-only, dirty, young	.01.xx0111.1
219  * read-write, clean, old	.11.xx1100.1
220  * read-write, clean, young	.01.xx1101.1
221  * read-write, dirty, old	.10.xx1110.1
222  * read-write, dirty, young	.00.xx1111.1
223  * HW-bits: R read-only, I invalid
224  * SW-bits: p present, y young, d dirty, r read, w write, s special,
225  *	    u unused, l large
226  *
227  * pte_none    is true for the bit pattern .10.00000000, pte == 0x400
228  * pte_swap    is true for the bit pattern .11..ooooo.0, (pte & 0x201) == 0x200
229  * pte_present is true for the bit pattern .xx.xxxxxx.1, (pte & 0x001) == 0x001
230  */
231 
232 /* Bits in the segment/region table address-space-control-element */
233 #define _ASCE_ORIGIN		~0xfffUL/* region/segment table origin	    */
234 #define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
235 #define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
236 #define _ASCE_SPACE_SWITCH	0x40	/* space switch event		    */
237 #define _ASCE_REAL_SPACE	0x20	/* real space control		    */
238 #define _ASCE_TYPE_MASK		0x0c	/* asce table type mask		    */
239 #define _ASCE_TYPE_REGION1	0x0c	/* region first table type	    */
240 #define _ASCE_TYPE_REGION2	0x08	/* region second table type	    */
241 #define _ASCE_TYPE_REGION3	0x04	/* region third table type	    */
242 #define _ASCE_TYPE_SEGMENT	0x00	/* segment table type		    */
243 #define _ASCE_TABLE_LENGTH	0x03	/* region table length		    */
244 
245 /* Bits in the region table entry */
246 #define _REGION_ENTRY_ORIGIN	~0xfffUL/* region/segment table origin	    */
247 #define _REGION_ENTRY_PROTECT	0x200	/* region protection bit	    */
248 #define _REGION_ENTRY_NOEXEC	0x100	/* region no-execute bit	    */
249 #define _REGION_ENTRY_OFFSET	0xc0	/* region table offset		    */
250 #define _REGION_ENTRY_INVALID	0x20	/* invalid region table entry	    */
251 #define _REGION_ENTRY_TYPE_MASK	0x0c	/* region table type mask	    */
252 #define _REGION_ENTRY_TYPE_R1	0x0c	/* region first table type	    */
253 #define _REGION_ENTRY_TYPE_R2	0x08	/* region second table type	    */
254 #define _REGION_ENTRY_TYPE_R3	0x04	/* region third table type	    */
255 #define _REGION_ENTRY_LENGTH	0x03	/* region third length		    */
256 
257 #define _REGION1_ENTRY		(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_LENGTH)
258 #define _REGION1_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INVALID)
259 #define _REGION2_ENTRY		(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_LENGTH)
260 #define _REGION2_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INVALID)
261 #define _REGION3_ENTRY		(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_LENGTH)
262 #define _REGION3_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INVALID)
263 
264 #define _REGION3_ENTRY_ORIGIN_LARGE ~0x7fffffffUL /* large page address	     */
265 #define _REGION3_ENTRY_DIRTY	0x2000	/* SW region dirty bit */
266 #define _REGION3_ENTRY_YOUNG	0x1000	/* SW region young bit */
267 #define _REGION3_ENTRY_LARGE	0x0400	/* RTTE-format control, large page  */
268 #define _REGION3_ENTRY_READ	0x0002	/* SW region read bit */
269 #define _REGION3_ENTRY_WRITE	0x0001	/* SW region write bit */
270 
271 #ifdef CONFIG_MEM_SOFT_DIRTY
272 #define _REGION3_ENTRY_SOFT_DIRTY 0x4000 /* SW region soft dirty bit */
273 #else
274 #define _REGION3_ENTRY_SOFT_DIRTY 0x0000 /* SW region soft dirty bit */
275 #endif
276 
277 #define _REGION_ENTRY_BITS	 0xfffffffffffff22fUL
278 
279 /* Bits in the segment table entry */
280 #define _SEGMENT_ENTRY_BITS			0xfffffffffffffe33UL
281 #define _SEGMENT_ENTRY_HARDWARE_BITS		0xfffffffffffffe30UL
282 #define _SEGMENT_ENTRY_HARDWARE_BITS_LARGE	0xfffffffffff00730UL
283 #define _SEGMENT_ENTRY_ORIGIN_LARGE ~0xfffffUL /* large page address	    */
284 #define _SEGMENT_ENTRY_ORIGIN	~0x7ffUL/* page table origin		    */
285 #define _SEGMENT_ENTRY_PROTECT	0x200	/* segment protection bit	    */
286 #define _SEGMENT_ENTRY_NOEXEC	0x100	/* segment no-execute bit	    */
287 #define _SEGMENT_ENTRY_INVALID	0x20	/* invalid segment table entry	    */
288 #define _SEGMENT_ENTRY_TYPE_MASK 0x0c	/* segment table type mask	    */
289 
290 #define _SEGMENT_ENTRY		(0)
291 #define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INVALID)
292 
293 #define _SEGMENT_ENTRY_DIRTY	0x2000	/* SW segment dirty bit */
294 #define _SEGMENT_ENTRY_YOUNG	0x1000	/* SW segment young bit */
295 #define _SEGMENT_ENTRY_LARGE	0x0400	/* STE-format control, large page */
296 #define _SEGMENT_ENTRY_WRITE	0x0002	/* SW segment write bit */
297 #define _SEGMENT_ENTRY_READ	0x0001	/* SW segment read bit */
298 
299 #ifdef CONFIG_MEM_SOFT_DIRTY
300 #define _SEGMENT_ENTRY_SOFT_DIRTY 0x4000 /* SW segment soft dirty bit */
301 #else
302 #define _SEGMENT_ENTRY_SOFT_DIRTY 0x0000 /* SW segment soft dirty bit */
303 #endif
304 
305 #define _CRST_ENTRIES	2048	/* number of region/segment table entries */
306 #define _PAGE_ENTRIES	256	/* number of page table entries	*/
307 
308 #define _CRST_TABLE_SIZE (_CRST_ENTRIES * 8)
309 #define _PAGE_TABLE_SIZE (_PAGE_ENTRIES * 8)
310 
311 #define _REGION1_SHIFT	53
312 #define _REGION2_SHIFT	42
313 #define _REGION3_SHIFT	31
314 #define _SEGMENT_SHIFT	20
315 
316 #define _REGION1_INDEX	(0x7ffUL << _REGION1_SHIFT)
317 #define _REGION2_INDEX	(0x7ffUL << _REGION2_SHIFT)
318 #define _REGION3_INDEX	(0x7ffUL << _REGION3_SHIFT)
319 #define _SEGMENT_INDEX	(0x7ffUL << _SEGMENT_SHIFT)
320 #define _PAGE_INDEX	(0xffUL  << _PAGE_SHIFT)
321 
322 #define _REGION1_SIZE	(1UL << _REGION1_SHIFT)
323 #define _REGION2_SIZE	(1UL << _REGION2_SHIFT)
324 #define _REGION3_SIZE	(1UL << _REGION3_SHIFT)
325 #define _SEGMENT_SIZE	(1UL << _SEGMENT_SHIFT)
326 
327 #define _REGION1_MASK	(~(_REGION1_SIZE - 1))
328 #define _REGION2_MASK	(~(_REGION2_SIZE - 1))
329 #define _REGION3_MASK	(~(_REGION3_SIZE - 1))
330 #define _SEGMENT_MASK	(~(_SEGMENT_SIZE - 1))
331 
332 #define PMD_SHIFT	_SEGMENT_SHIFT
333 #define PUD_SHIFT	_REGION3_SHIFT
334 #define P4D_SHIFT	_REGION2_SHIFT
335 #define PGDIR_SHIFT	_REGION1_SHIFT
336 
337 #define PMD_SIZE	_SEGMENT_SIZE
338 #define PUD_SIZE	_REGION3_SIZE
339 #define P4D_SIZE	_REGION2_SIZE
340 #define PGDIR_SIZE	_REGION1_SIZE
341 
342 #define PMD_MASK	_SEGMENT_MASK
343 #define PUD_MASK	_REGION3_MASK
344 #define P4D_MASK	_REGION2_MASK
345 #define PGDIR_MASK	_REGION1_MASK
346 
347 #define PTRS_PER_PTE	_PAGE_ENTRIES
348 #define PTRS_PER_PMD	_CRST_ENTRIES
349 #define PTRS_PER_PUD	_CRST_ENTRIES
350 #define PTRS_PER_P4D	_CRST_ENTRIES
351 #define PTRS_PER_PGD	_CRST_ENTRIES
352 
353 /*
354  * Segment table and region3 table entry encoding
355  * (R = read-only, I = invalid, y = young bit):
356  *				dy..R...I...wr
357  * prot-none, clean, old	00..1...1...00
358  * prot-none, clean, young	01..1...1...00
359  * prot-none, dirty, old	10..1...1...00
360  * prot-none, dirty, young	11..1...1...00
361  * read-only, clean, old	00..1...1...01
362  * read-only, clean, young	01..1...0...01
363  * read-only, dirty, old	10..1...1...01
364  * read-only, dirty, young	11..1...0...01
365  * read-write, clean, old	00..1...1...11
366  * read-write, clean, young	01..1...0...11
367  * read-write, dirty, old	10..0...1...11
368  * read-write, dirty, young	11..0...0...11
369  * The segment table origin is used to distinguish empty (origin==0) from
370  * read-write, old segment table entries (origin!=0)
371  * HW-bits: R read-only, I invalid
372  * SW-bits: y young, d dirty, r read, w write
373  */
374 
375 /* Page status table bits for virtualization */
376 #define PGSTE_ACC_BITS	0xf000000000000000UL
377 #define PGSTE_FP_BIT	0x0800000000000000UL
378 #define PGSTE_PCL_BIT	0x0080000000000000UL
379 #define PGSTE_HR_BIT	0x0040000000000000UL
380 #define PGSTE_HC_BIT	0x0020000000000000UL
381 #define PGSTE_GR_BIT	0x0004000000000000UL
382 #define PGSTE_GC_BIT	0x0002000000000000UL
383 #define PGSTE_UC_BIT	0x0000800000000000UL	/* user dirty (migration) */
384 #define PGSTE_IN_BIT	0x0000400000000000UL	/* IPTE notify bit */
385 #define PGSTE_VSIE_BIT	0x0000200000000000UL	/* ref'd in a shadow table */
386 
387 /* Guest Page State used for virtualization */
388 #define _PGSTE_GPS_ZERO			0x0000000080000000UL
389 #define _PGSTE_GPS_NODAT		0x0000000040000000UL
390 #define _PGSTE_GPS_USAGE_MASK		0x0000000003000000UL
391 #define _PGSTE_GPS_USAGE_STABLE		0x0000000000000000UL
392 #define _PGSTE_GPS_USAGE_UNUSED		0x0000000001000000UL
393 #define _PGSTE_GPS_USAGE_POT_VOLATILE	0x0000000002000000UL
394 #define _PGSTE_GPS_USAGE_VOLATILE	_PGSTE_GPS_USAGE_MASK
395 
396 /*
397  * A user page table pointer has the space-switch-event bit, the
398  * private-space-control bit and the storage-alteration-event-control
399  * bit set. A kernel page table pointer doesn't need them.
400  */
401 #define _ASCE_USER_BITS		(_ASCE_SPACE_SWITCH | _ASCE_PRIVATE_SPACE | \
402 				 _ASCE_ALT_EVENT)
403 
404 /*
405  * Page protection definitions.
406  */
407 #define PAGE_NONE	__pgprot(_PAGE_PRESENT | _PAGE_INVALID | _PAGE_PROTECT)
408 #define PAGE_RO		__pgprot(_PAGE_PRESENT | _PAGE_READ | \
409 				 _PAGE_NOEXEC  | _PAGE_INVALID | _PAGE_PROTECT)
410 #define PAGE_RX		__pgprot(_PAGE_PRESENT | _PAGE_READ | \
411 				 _PAGE_INVALID | _PAGE_PROTECT)
412 #define PAGE_RW		__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
413 				 _PAGE_NOEXEC  | _PAGE_INVALID | _PAGE_PROTECT)
414 #define PAGE_RWX	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
415 				 _PAGE_INVALID | _PAGE_PROTECT)
416 
417 #define PAGE_SHARED	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
418 				 _PAGE_YOUNG | _PAGE_DIRTY | _PAGE_NOEXEC)
419 #define PAGE_KERNEL	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
420 				 _PAGE_YOUNG | _PAGE_DIRTY | _PAGE_NOEXEC)
421 #define PAGE_KERNEL_RO	__pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_YOUNG | \
422 				 _PAGE_PROTECT | _PAGE_NOEXEC)
423 #define PAGE_KERNEL_EXEC __pgprot(_PAGE_PRESENT | _PAGE_READ | _PAGE_WRITE | \
424 				  _PAGE_YOUNG |	_PAGE_DIRTY)
425 
426 /*
427  * On s390 the page table entry has an invalid bit and a read-only bit.
428  * Read permission implies execute permission and write permission
429  * implies read permission.
430  */
431          /*xwr*/
432 
433 /*
434  * Segment entry (large page) protection definitions.
435  */
436 #define SEGMENT_NONE	__pgprot(_SEGMENT_ENTRY_INVALID | \
437 				 _SEGMENT_ENTRY_PROTECT)
438 #define SEGMENT_RO	__pgprot(_SEGMENT_ENTRY_PROTECT | \
439 				 _SEGMENT_ENTRY_READ | \
440 				 _SEGMENT_ENTRY_NOEXEC)
441 #define SEGMENT_RX	__pgprot(_SEGMENT_ENTRY_PROTECT | \
442 				 _SEGMENT_ENTRY_READ)
443 #define SEGMENT_RW	__pgprot(_SEGMENT_ENTRY_READ | \
444 				 _SEGMENT_ENTRY_WRITE | \
445 				 _SEGMENT_ENTRY_NOEXEC)
446 #define SEGMENT_RWX	__pgprot(_SEGMENT_ENTRY_READ | \
447 				 _SEGMENT_ENTRY_WRITE)
448 #define SEGMENT_KERNEL	__pgprot(_SEGMENT_ENTRY |	\
449 				 _SEGMENT_ENTRY_LARGE |	\
450 				 _SEGMENT_ENTRY_READ |	\
451 				 _SEGMENT_ENTRY_WRITE | \
452 				 _SEGMENT_ENTRY_YOUNG | \
453 				 _SEGMENT_ENTRY_DIRTY | \
454 				 _SEGMENT_ENTRY_NOEXEC)
455 #define SEGMENT_KERNEL_RO __pgprot(_SEGMENT_ENTRY |	\
456 				 _SEGMENT_ENTRY_LARGE |	\
457 				 _SEGMENT_ENTRY_READ |	\
458 				 _SEGMENT_ENTRY_YOUNG |	\
459 				 _SEGMENT_ENTRY_PROTECT | \
460 				 _SEGMENT_ENTRY_NOEXEC)
461 #define SEGMENT_KERNEL_EXEC __pgprot(_SEGMENT_ENTRY |	\
462 				 _SEGMENT_ENTRY_LARGE |	\
463 				 _SEGMENT_ENTRY_READ |	\
464 				 _SEGMENT_ENTRY_WRITE | \
465 				 _SEGMENT_ENTRY_YOUNG |	\
466 				 _SEGMENT_ENTRY_DIRTY)
467 
468 /*
469  * Region3 entry (large page) protection definitions.
470  */
471 
472 #define REGION3_KERNEL	__pgprot(_REGION_ENTRY_TYPE_R3 | \
473 				 _REGION3_ENTRY_LARGE |	 \
474 				 _REGION3_ENTRY_READ |	 \
475 				 _REGION3_ENTRY_WRITE |	 \
476 				 _REGION3_ENTRY_YOUNG |	 \
477 				 _REGION3_ENTRY_DIRTY | \
478 				 _REGION_ENTRY_NOEXEC)
479 #define REGION3_KERNEL_RO __pgprot(_REGION_ENTRY_TYPE_R3 | \
480 				   _REGION3_ENTRY_LARGE |  \
481 				   _REGION3_ENTRY_READ |   \
482 				   _REGION3_ENTRY_YOUNG |  \
483 				   _REGION_ENTRY_PROTECT | \
484 				   _REGION_ENTRY_NOEXEC)
485 #define REGION3_KERNEL_EXEC __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 
492 static inline bool mm_p4d_folded(struct mm_struct *mm)
493 {
494 	return mm->context.asce_limit <= _REGION1_SIZE;
495 }
496 #define mm_p4d_folded(mm) mm_p4d_folded(mm)
497 
498 static inline bool mm_pud_folded(struct mm_struct *mm)
499 {
500 	return mm->context.asce_limit <= _REGION2_SIZE;
501 }
502 #define mm_pud_folded(mm) mm_pud_folded(mm)
503 
504 static inline bool mm_pmd_folded(struct mm_struct *mm)
505 {
506 	return mm->context.asce_limit <= _REGION3_SIZE;
507 }
508 #define mm_pmd_folded(mm) mm_pmd_folded(mm)
509 
510 static inline int mm_has_pgste(struct mm_struct *mm)
511 {
512 #ifdef CONFIG_PGSTE
513 	if (unlikely(mm->context.has_pgste))
514 		return 1;
515 #endif
516 	return 0;
517 }
518 
519 static inline int mm_is_protected(struct mm_struct *mm)
520 {
521 #ifdef CONFIG_PGSTE
522 	if (unlikely(atomic_read(&mm->context.protected_count)))
523 		return 1;
524 #endif
525 	return 0;
526 }
527 
528 static inline int mm_alloc_pgste(struct mm_struct *mm)
529 {
530 #ifdef CONFIG_PGSTE
531 	if (unlikely(mm->context.alloc_pgste))
532 		return 1;
533 #endif
534 	return 0;
535 }
536 
537 static inline pte_t clear_pte_bit(pte_t pte, pgprot_t prot)
538 {
539 	return __pte(pte_val(pte) & ~pgprot_val(prot));
540 }
541 
542 static inline pte_t set_pte_bit(pte_t pte, pgprot_t prot)
543 {
544 	return __pte(pte_val(pte) | pgprot_val(prot));
545 }
546 
547 static inline pmd_t clear_pmd_bit(pmd_t pmd, pgprot_t prot)
548 {
549 	return __pmd(pmd_val(pmd) & ~pgprot_val(prot));
550 }
551 
552 static inline pmd_t set_pmd_bit(pmd_t pmd, pgprot_t prot)
553 {
554 	return __pmd(pmd_val(pmd) | pgprot_val(prot));
555 }
556 
557 static inline pud_t clear_pud_bit(pud_t pud, pgprot_t prot)
558 {
559 	return __pud(pud_val(pud) & ~pgprot_val(prot));
560 }
561 
562 static inline pud_t set_pud_bit(pud_t pud, pgprot_t prot)
563 {
564 	return __pud(pud_val(pud) | pgprot_val(prot));
565 }
566 
567 /*
568  * As soon as the guest uses storage keys or enables PV, we deduplicate all
569  * mapped shared zeropages and prevent new shared zeropages from getting
570  * mapped.
571  */
572 #define mm_forbids_zeropage mm_forbids_zeropage
573 static inline int mm_forbids_zeropage(struct mm_struct *mm)
574 {
575 #ifdef CONFIG_PGSTE
576 	if (!mm->context.allow_cow_sharing)
577 		return 1;
578 #endif
579 	return 0;
580 }
581 
582 static inline int mm_uses_skeys(struct mm_struct *mm)
583 {
584 #ifdef CONFIG_PGSTE
585 	if (mm->context.uses_skeys)
586 		return 1;
587 #endif
588 	return 0;
589 }
590 
591 static inline void csp(unsigned int *ptr, unsigned int old, unsigned int new)
592 {
593 	union register_pair r1 = { .even = old, .odd = new, };
594 	unsigned long address = (unsigned long)ptr | 1;
595 
596 	asm volatile(
597 		"	csp	%[r1],%[address]"
598 		: [r1] "+&d" (r1.pair), "+m" (*ptr)
599 		: [address] "d" (address)
600 		: "cc");
601 }
602 
603 static inline void cspg(unsigned long *ptr, unsigned long old, unsigned long new)
604 {
605 	union register_pair r1 = { .even = old, .odd = new, };
606 	unsigned long address = (unsigned long)ptr | 1;
607 
608 	asm volatile(
609 		"	cspg	%[r1],%[address]"
610 		: [r1] "+&d" (r1.pair), "+m" (*ptr)
611 		: [address] "d" (address)
612 		: "cc");
613 }
614 
615 #define CRDTE_DTT_PAGE		0x00UL
616 #define CRDTE_DTT_SEGMENT	0x10UL
617 #define CRDTE_DTT_REGION3	0x14UL
618 #define CRDTE_DTT_REGION2	0x18UL
619 #define CRDTE_DTT_REGION1	0x1cUL
620 
621 static inline void crdte(unsigned long old, unsigned long new,
622 			 unsigned long *table, unsigned long dtt,
623 			 unsigned long address, unsigned long asce)
624 {
625 	union register_pair r1 = { .even = old, .odd = new, };
626 	union register_pair r2 = { .even = __pa(table) | dtt, .odd = address, };
627 
628 	asm volatile(".insn rrf,0xb98f0000,%[r1],%[r2],%[asce],0"
629 		     : [r1] "+&d" (r1.pair)
630 		     : [r2] "d" (r2.pair), [asce] "a" (asce)
631 		     : "memory", "cc");
632 }
633 
634 /*
635  * pgd/p4d/pud/pmd/pte query functions
636  */
637 static inline int pgd_folded(pgd_t pgd)
638 {
639 	return (pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R1;
640 }
641 
642 static inline int pgd_present(pgd_t pgd)
643 {
644 	if (pgd_folded(pgd))
645 		return 1;
646 	return (pgd_val(pgd) & _REGION_ENTRY_ORIGIN) != 0UL;
647 }
648 
649 static inline int pgd_none(pgd_t pgd)
650 {
651 	if (pgd_folded(pgd))
652 		return 0;
653 	return (pgd_val(pgd) & _REGION_ENTRY_INVALID) != 0UL;
654 }
655 
656 static inline int pgd_bad(pgd_t pgd)
657 {
658 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R1)
659 		return 0;
660 	return (pgd_val(pgd) & ~_REGION_ENTRY_BITS) != 0;
661 }
662 
663 static inline unsigned long pgd_pfn(pgd_t pgd)
664 {
665 	unsigned long origin_mask;
666 
667 	origin_mask = _REGION_ENTRY_ORIGIN;
668 	return (pgd_val(pgd) & origin_mask) >> PAGE_SHIFT;
669 }
670 
671 static inline int p4d_folded(p4d_t p4d)
672 {
673 	return (p4d_val(p4d) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2;
674 }
675 
676 static inline int p4d_present(p4d_t p4d)
677 {
678 	if (p4d_folded(p4d))
679 		return 1;
680 	return (p4d_val(p4d) & _REGION_ENTRY_ORIGIN) != 0UL;
681 }
682 
683 static inline int p4d_none(p4d_t p4d)
684 {
685 	if (p4d_folded(p4d))
686 		return 0;
687 	return p4d_val(p4d) == _REGION2_ENTRY_EMPTY;
688 }
689 
690 static inline unsigned long p4d_pfn(p4d_t p4d)
691 {
692 	unsigned long origin_mask;
693 
694 	origin_mask = _REGION_ENTRY_ORIGIN;
695 	return (p4d_val(p4d) & origin_mask) >> PAGE_SHIFT;
696 }
697 
698 static inline int pud_folded(pud_t pud)
699 {
700 	return (pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3;
701 }
702 
703 static inline int pud_present(pud_t pud)
704 {
705 	if (pud_folded(pud))
706 		return 1;
707 	return (pud_val(pud) & _REGION_ENTRY_ORIGIN) != 0UL;
708 }
709 
710 static inline int pud_none(pud_t pud)
711 {
712 	if (pud_folded(pud))
713 		return 0;
714 	return pud_val(pud) == _REGION3_ENTRY_EMPTY;
715 }
716 
717 #define pud_leaf	pud_large
718 static inline int pud_large(pud_t pud)
719 {
720 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) != _REGION_ENTRY_TYPE_R3)
721 		return 0;
722 	return !!(pud_val(pud) & _REGION3_ENTRY_LARGE);
723 }
724 
725 #define pmd_leaf	pmd_large
726 static inline int pmd_large(pmd_t pmd)
727 {
728 	return (pmd_val(pmd) & _SEGMENT_ENTRY_LARGE) != 0;
729 }
730 
731 static inline int pmd_bad(pmd_t pmd)
732 {
733 	if ((pmd_val(pmd) & _SEGMENT_ENTRY_TYPE_MASK) > 0 || pmd_large(pmd))
734 		return 1;
735 	return (pmd_val(pmd) & ~_SEGMENT_ENTRY_BITS) != 0;
736 }
737 
738 static inline int pud_bad(pud_t pud)
739 {
740 	unsigned long type = pud_val(pud) & _REGION_ENTRY_TYPE_MASK;
741 
742 	if (type > _REGION_ENTRY_TYPE_R3 || pud_leaf(pud))
743 		return 1;
744 	if (type < _REGION_ENTRY_TYPE_R3)
745 		return 0;
746 	return (pud_val(pud) & ~_REGION_ENTRY_BITS) != 0;
747 }
748 
749 static inline int p4d_bad(p4d_t p4d)
750 {
751 	unsigned long type = p4d_val(p4d) & _REGION_ENTRY_TYPE_MASK;
752 
753 	if (type > _REGION_ENTRY_TYPE_R2)
754 		return 1;
755 	if (type < _REGION_ENTRY_TYPE_R2)
756 		return 0;
757 	return (p4d_val(p4d) & ~_REGION_ENTRY_BITS) != 0;
758 }
759 
760 static inline int pmd_present(pmd_t pmd)
761 {
762 	return pmd_val(pmd) != _SEGMENT_ENTRY_EMPTY;
763 }
764 
765 static inline int pmd_none(pmd_t pmd)
766 {
767 	return pmd_val(pmd) == _SEGMENT_ENTRY_EMPTY;
768 }
769 
770 #define pmd_write pmd_write
771 static inline int pmd_write(pmd_t pmd)
772 {
773 	return (pmd_val(pmd) & _SEGMENT_ENTRY_WRITE) != 0;
774 }
775 
776 #define pud_write pud_write
777 static inline int pud_write(pud_t pud)
778 {
779 	return (pud_val(pud) & _REGION3_ENTRY_WRITE) != 0;
780 }
781 
782 static inline int pmd_dirty(pmd_t pmd)
783 {
784 	return (pmd_val(pmd) & _SEGMENT_ENTRY_DIRTY) != 0;
785 }
786 
787 #define pmd_young pmd_young
788 static inline int pmd_young(pmd_t pmd)
789 {
790 	return (pmd_val(pmd) & _SEGMENT_ENTRY_YOUNG) != 0;
791 }
792 
793 static inline int pte_present(pte_t pte)
794 {
795 	/* Bit pattern: (pte & 0x001) == 0x001 */
796 	return (pte_val(pte) & _PAGE_PRESENT) != 0;
797 }
798 
799 static inline int pte_none(pte_t pte)
800 {
801 	/* Bit pattern: pte == 0x400 */
802 	return pte_val(pte) == _PAGE_INVALID;
803 }
804 
805 static inline int pte_swap(pte_t pte)
806 {
807 	/* Bit pattern: (pte & 0x201) == 0x200 */
808 	return (pte_val(pte) & (_PAGE_PROTECT | _PAGE_PRESENT))
809 		== _PAGE_PROTECT;
810 }
811 
812 static inline int pte_special(pte_t pte)
813 {
814 	return (pte_val(pte) & _PAGE_SPECIAL);
815 }
816 
817 #define __HAVE_ARCH_PTE_SAME
818 static inline int pte_same(pte_t a, pte_t b)
819 {
820 	return pte_val(a) == pte_val(b);
821 }
822 
823 #ifdef CONFIG_NUMA_BALANCING
824 static inline int pte_protnone(pte_t pte)
825 {
826 	return pte_present(pte) && !(pte_val(pte) & _PAGE_READ);
827 }
828 
829 static inline int pmd_protnone(pmd_t pmd)
830 {
831 	/* pmd_large(pmd) implies pmd_present(pmd) */
832 	return pmd_large(pmd) && !(pmd_val(pmd) & _SEGMENT_ENTRY_READ);
833 }
834 #endif
835 
836 static inline int pte_swp_exclusive(pte_t pte)
837 {
838 	return pte_val(pte) & _PAGE_SWP_EXCLUSIVE;
839 }
840 
841 static inline pte_t pte_swp_mkexclusive(pte_t pte)
842 {
843 	return set_pte_bit(pte, __pgprot(_PAGE_SWP_EXCLUSIVE));
844 }
845 
846 static inline pte_t pte_swp_clear_exclusive(pte_t pte)
847 {
848 	return clear_pte_bit(pte, __pgprot(_PAGE_SWP_EXCLUSIVE));
849 }
850 
851 static inline int pte_soft_dirty(pte_t pte)
852 {
853 	return pte_val(pte) & _PAGE_SOFT_DIRTY;
854 }
855 #define pte_swp_soft_dirty pte_soft_dirty
856 
857 static inline pte_t pte_mksoft_dirty(pte_t pte)
858 {
859 	return set_pte_bit(pte, __pgprot(_PAGE_SOFT_DIRTY));
860 }
861 #define pte_swp_mksoft_dirty pte_mksoft_dirty
862 
863 static inline pte_t pte_clear_soft_dirty(pte_t pte)
864 {
865 	return clear_pte_bit(pte, __pgprot(_PAGE_SOFT_DIRTY));
866 }
867 #define pte_swp_clear_soft_dirty pte_clear_soft_dirty
868 
869 static inline int pmd_soft_dirty(pmd_t pmd)
870 {
871 	return pmd_val(pmd) & _SEGMENT_ENTRY_SOFT_DIRTY;
872 }
873 
874 static inline pmd_t pmd_mksoft_dirty(pmd_t pmd)
875 {
876 	return set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_SOFT_DIRTY));
877 }
878 
879 static inline pmd_t pmd_clear_soft_dirty(pmd_t pmd)
880 {
881 	return clear_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_SOFT_DIRTY));
882 }
883 
884 /*
885  * query functions pte_write/pte_dirty/pte_young only work if
886  * pte_present() is true. Undefined behaviour if not..
887  */
888 static inline int pte_write(pte_t pte)
889 {
890 	return (pte_val(pte) & _PAGE_WRITE) != 0;
891 }
892 
893 static inline int pte_dirty(pte_t pte)
894 {
895 	return (pte_val(pte) & _PAGE_DIRTY) != 0;
896 }
897 
898 static inline int pte_young(pte_t pte)
899 {
900 	return (pte_val(pte) & _PAGE_YOUNG) != 0;
901 }
902 
903 #define __HAVE_ARCH_PTE_UNUSED
904 static inline int pte_unused(pte_t pte)
905 {
906 	return pte_val(pte) & _PAGE_UNUSED;
907 }
908 
909 /*
910  * Extract the pgprot value from the given pte while at the same time making it
911  * usable for kernel address space mappings where fault driven dirty and
912  * young/old accounting is not supported, i.e _PAGE_PROTECT and _PAGE_INVALID
913  * must not be set.
914  */
915 static inline pgprot_t pte_pgprot(pte_t pte)
916 {
917 	unsigned long pte_flags = pte_val(pte) & _PAGE_CHG_MASK;
918 
919 	if (pte_write(pte))
920 		pte_flags |= pgprot_val(PAGE_KERNEL);
921 	else
922 		pte_flags |= pgprot_val(PAGE_KERNEL_RO);
923 	pte_flags |= pte_val(pte) & mio_wb_bit_mask;
924 
925 	return __pgprot(pte_flags);
926 }
927 
928 /*
929  * pgd/pmd/pte modification functions
930  */
931 
932 static inline void set_pgd(pgd_t *pgdp, pgd_t pgd)
933 {
934 	WRITE_ONCE(*pgdp, pgd);
935 }
936 
937 static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
938 {
939 	WRITE_ONCE(*p4dp, p4d);
940 }
941 
942 static inline void set_pud(pud_t *pudp, pud_t pud)
943 {
944 	WRITE_ONCE(*pudp, pud);
945 }
946 
947 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
948 {
949 	WRITE_ONCE(*pmdp, pmd);
950 }
951 
952 static inline void set_pte(pte_t *ptep, pte_t pte)
953 {
954 	WRITE_ONCE(*ptep, pte);
955 }
956 
957 static inline void pgd_clear(pgd_t *pgd)
958 {
959 	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R1)
960 		set_pgd(pgd, __pgd(_REGION1_ENTRY_EMPTY));
961 }
962 
963 static inline void p4d_clear(p4d_t *p4d)
964 {
965 	if ((p4d_val(*p4d) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
966 		set_p4d(p4d, __p4d(_REGION2_ENTRY_EMPTY));
967 }
968 
969 static inline void pud_clear(pud_t *pud)
970 {
971 	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
972 		set_pud(pud, __pud(_REGION3_ENTRY_EMPTY));
973 }
974 
975 static inline void pmd_clear(pmd_t *pmdp)
976 {
977 	set_pmd(pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
978 }
979 
980 static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
981 {
982 	set_pte(ptep, __pte(_PAGE_INVALID));
983 }
984 
985 /*
986  * The following pte modification functions only work if
987  * pte_present() is true. Undefined behaviour if not..
988  */
989 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
990 {
991 	pte = clear_pte_bit(pte, __pgprot(~_PAGE_CHG_MASK));
992 	pte = set_pte_bit(pte, newprot);
993 	/*
994 	 * newprot for PAGE_NONE, PAGE_RO, PAGE_RX, PAGE_RW and PAGE_RWX
995 	 * has the invalid bit set, clear it again for readable, young pages
996 	 */
997 	if ((pte_val(pte) & _PAGE_YOUNG) && (pte_val(pte) & _PAGE_READ))
998 		pte = clear_pte_bit(pte, __pgprot(_PAGE_INVALID));
999 	/*
1000 	 * newprot for PAGE_RO, PAGE_RX, PAGE_RW and PAGE_RWX has the page
1001 	 * protection bit set, clear it again for writable, dirty pages
1002 	 */
1003 	if ((pte_val(pte) & _PAGE_DIRTY) && (pte_val(pte) & _PAGE_WRITE))
1004 		pte = clear_pte_bit(pte, __pgprot(_PAGE_PROTECT));
1005 	return pte;
1006 }
1007 
1008 static inline pte_t pte_wrprotect(pte_t pte)
1009 {
1010 	pte = clear_pte_bit(pte, __pgprot(_PAGE_WRITE));
1011 	return set_pte_bit(pte, __pgprot(_PAGE_PROTECT));
1012 }
1013 
1014 static inline pte_t pte_mkwrite_novma(pte_t pte)
1015 {
1016 	pte = set_pte_bit(pte, __pgprot(_PAGE_WRITE));
1017 	if (pte_val(pte) & _PAGE_DIRTY)
1018 		pte = clear_pte_bit(pte, __pgprot(_PAGE_PROTECT));
1019 	return pte;
1020 }
1021 
1022 static inline pte_t pte_mkclean(pte_t pte)
1023 {
1024 	pte = clear_pte_bit(pte, __pgprot(_PAGE_DIRTY));
1025 	return set_pte_bit(pte, __pgprot(_PAGE_PROTECT));
1026 }
1027 
1028 static inline pte_t pte_mkdirty(pte_t pte)
1029 {
1030 	pte = set_pte_bit(pte, __pgprot(_PAGE_DIRTY | _PAGE_SOFT_DIRTY));
1031 	if (pte_val(pte) & _PAGE_WRITE)
1032 		pte = clear_pte_bit(pte, __pgprot(_PAGE_PROTECT));
1033 	return pte;
1034 }
1035 
1036 static inline pte_t pte_mkold(pte_t pte)
1037 {
1038 	pte = clear_pte_bit(pte, __pgprot(_PAGE_YOUNG));
1039 	return set_pte_bit(pte, __pgprot(_PAGE_INVALID));
1040 }
1041 
1042 static inline pte_t pte_mkyoung(pte_t pte)
1043 {
1044 	pte = set_pte_bit(pte, __pgprot(_PAGE_YOUNG));
1045 	if (pte_val(pte) & _PAGE_READ)
1046 		pte = clear_pte_bit(pte, __pgprot(_PAGE_INVALID));
1047 	return pte;
1048 }
1049 
1050 static inline pte_t pte_mkspecial(pte_t pte)
1051 {
1052 	return set_pte_bit(pte, __pgprot(_PAGE_SPECIAL));
1053 }
1054 
1055 #ifdef CONFIG_HUGETLB_PAGE
1056 static inline pte_t pte_mkhuge(pte_t pte)
1057 {
1058 	return set_pte_bit(pte, __pgprot(_PAGE_LARGE));
1059 }
1060 #endif
1061 
1062 #define IPTE_GLOBAL	0
1063 #define	IPTE_LOCAL	1
1064 
1065 #define IPTE_NODAT	0x400
1066 #define IPTE_GUEST_ASCE	0x800
1067 
1068 static __always_inline void __ptep_rdp(unsigned long addr, pte_t *ptep,
1069 				       unsigned long opt, unsigned long asce,
1070 				       int local)
1071 {
1072 	unsigned long pto;
1073 
1074 	pto = __pa(ptep) & ~(PTRS_PER_PTE * sizeof(pte_t) - 1);
1075 	asm volatile(".insn rrf,0xb98b0000,%[r1],%[r2],%[asce],%[m4]"
1076 		     : "+m" (*ptep)
1077 		     : [r1] "a" (pto), [r2] "a" ((addr & PAGE_MASK) | opt),
1078 		       [asce] "a" (asce), [m4] "i" (local));
1079 }
1080 
1081 static __always_inline void __ptep_ipte(unsigned long address, pte_t *ptep,
1082 					unsigned long opt, unsigned long asce,
1083 					int local)
1084 {
1085 	unsigned long pto = __pa(ptep);
1086 
1087 	if (__builtin_constant_p(opt) && opt == 0) {
1088 		/* Invalidation + TLB flush for the pte */
1089 		asm volatile(
1090 			"	ipte	%[r1],%[r2],0,%[m4]"
1091 			: "+m" (*ptep) : [r1] "a" (pto), [r2] "a" (address),
1092 			  [m4] "i" (local));
1093 		return;
1094 	}
1095 
1096 	/* Invalidate ptes with options + TLB flush of the ptes */
1097 	opt = opt | (asce & _ASCE_ORIGIN);
1098 	asm volatile(
1099 		"	ipte	%[r1],%[r2],%[r3],%[m4]"
1100 		: [r2] "+a" (address), [r3] "+a" (opt)
1101 		: [r1] "a" (pto), [m4] "i" (local) : "memory");
1102 }
1103 
1104 static __always_inline void __ptep_ipte_range(unsigned long address, int nr,
1105 					      pte_t *ptep, int local)
1106 {
1107 	unsigned long pto = __pa(ptep);
1108 
1109 	/* Invalidate a range of ptes + TLB flush of the ptes */
1110 	do {
1111 		asm volatile(
1112 			"	ipte %[r1],%[r2],%[r3],%[m4]"
1113 			: [r2] "+a" (address), [r3] "+a" (nr)
1114 			: [r1] "a" (pto), [m4] "i" (local) : "memory");
1115 	} while (nr != 255);
1116 }
1117 
1118 /*
1119  * This is hard to understand. ptep_get_and_clear and ptep_clear_flush
1120  * both clear the TLB for the unmapped pte. The reason is that
1121  * ptep_get_and_clear is used in common code (e.g. change_pte_range)
1122  * to modify an active pte. The sequence is
1123  *   1) ptep_get_and_clear
1124  *   2) set_pte_at
1125  *   3) flush_tlb_range
1126  * On s390 the tlb needs to get flushed with the modification of the pte
1127  * if the pte is active. The only way how this can be implemented is to
1128  * have ptep_get_and_clear do the tlb flush. In exchange flush_tlb_range
1129  * is a nop.
1130  */
1131 pte_t ptep_xchg_direct(struct mm_struct *, unsigned long, pte_t *, pte_t);
1132 pte_t ptep_xchg_lazy(struct mm_struct *, unsigned long, pte_t *, pte_t);
1133 
1134 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1135 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
1136 					    unsigned long addr, pte_t *ptep)
1137 {
1138 	pte_t pte = *ptep;
1139 
1140 	pte = ptep_xchg_direct(vma->vm_mm, addr, ptep, pte_mkold(pte));
1141 	return pte_young(pte);
1142 }
1143 
1144 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1145 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
1146 					 unsigned long address, pte_t *ptep)
1147 {
1148 	return ptep_test_and_clear_young(vma, address, ptep);
1149 }
1150 
1151 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1152 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
1153 				       unsigned long addr, pte_t *ptep)
1154 {
1155 	pte_t res;
1156 
1157 	res = ptep_xchg_lazy(mm, addr, ptep, __pte(_PAGE_INVALID));
1158 	/* At this point the reference through the mapping is still present */
1159 	if (mm_is_protected(mm) && pte_present(res))
1160 		uv_convert_owned_from_secure(pte_val(res) & PAGE_MASK);
1161 	return res;
1162 }
1163 
1164 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1165 pte_t ptep_modify_prot_start(struct vm_area_struct *, unsigned long, pte_t *);
1166 void ptep_modify_prot_commit(struct vm_area_struct *, unsigned long,
1167 			     pte_t *, pte_t, pte_t);
1168 
1169 #define __HAVE_ARCH_PTEP_CLEAR_FLUSH
1170 static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
1171 				     unsigned long addr, pte_t *ptep)
1172 {
1173 	pte_t res;
1174 
1175 	res = ptep_xchg_direct(vma->vm_mm, addr, ptep, __pte(_PAGE_INVALID));
1176 	/* At this point the reference through the mapping is still present */
1177 	if (mm_is_protected(vma->vm_mm) && pte_present(res))
1178 		uv_convert_owned_from_secure(pte_val(res) & PAGE_MASK);
1179 	return res;
1180 }
1181 
1182 /*
1183  * The batched pte unmap code uses ptep_get_and_clear_full to clear the
1184  * ptes. Here an optimization is possible. tlb_gather_mmu flushes all
1185  * tlbs of an mm if it can guarantee that the ptes of the mm_struct
1186  * cannot be accessed while the batched unmap is running. In this case
1187  * full==1 and a simple pte_clear is enough. See tlb.h.
1188  */
1189 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
1190 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
1191 					    unsigned long addr,
1192 					    pte_t *ptep, int full)
1193 {
1194 	pte_t res;
1195 
1196 	if (full) {
1197 		res = *ptep;
1198 		set_pte(ptep, __pte(_PAGE_INVALID));
1199 	} else {
1200 		res = ptep_xchg_lazy(mm, addr, ptep, __pte(_PAGE_INVALID));
1201 	}
1202 	/* Nothing to do */
1203 	if (!mm_is_protected(mm) || !pte_present(res))
1204 		return res;
1205 	/*
1206 	 * At this point the reference through the mapping is still present.
1207 	 * The notifier should have destroyed all protected vCPUs at this
1208 	 * point, so the destroy should be successful.
1209 	 */
1210 	if (full && !uv_destroy_owned_page(pte_val(res) & PAGE_MASK))
1211 		return res;
1212 	/*
1213 	 * If something went wrong and the page could not be destroyed, or
1214 	 * if this is not a mm teardown, the slower export is used as
1215 	 * fallback instead.
1216 	 */
1217 	uv_convert_owned_from_secure(pte_val(res) & PAGE_MASK);
1218 	return res;
1219 }
1220 
1221 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1222 static inline void ptep_set_wrprotect(struct mm_struct *mm,
1223 				      unsigned long addr, pte_t *ptep)
1224 {
1225 	pte_t pte = *ptep;
1226 
1227 	if (pte_write(pte))
1228 		ptep_xchg_lazy(mm, addr, ptep, pte_wrprotect(pte));
1229 }
1230 
1231 /*
1232  * Check if PTEs only differ in _PAGE_PROTECT HW bit, but also allow SW PTE
1233  * bits in the comparison. Those might change e.g. because of dirty and young
1234  * tracking.
1235  */
1236 static inline int pte_allow_rdp(pte_t old, pte_t new)
1237 {
1238 	/*
1239 	 * Only allow changes from RO to RW
1240 	 */
1241 	if (!(pte_val(old) & _PAGE_PROTECT) || pte_val(new) & _PAGE_PROTECT)
1242 		return 0;
1243 
1244 	return (pte_val(old) & _PAGE_RDP_MASK) == (pte_val(new) & _PAGE_RDP_MASK);
1245 }
1246 
1247 static inline void flush_tlb_fix_spurious_fault(struct vm_area_struct *vma,
1248 						unsigned long address,
1249 						pte_t *ptep)
1250 {
1251 	/*
1252 	 * RDP might not have propagated the PTE protection reset to all CPUs,
1253 	 * so there could be spurious TLB protection faults.
1254 	 * NOTE: This will also be called when a racing pagetable update on
1255 	 * another thread already installed the correct PTE. Both cases cannot
1256 	 * really be distinguished.
1257 	 * Therefore, only do the local TLB flush when RDP can be used, and the
1258 	 * PTE does not have _PAGE_PROTECT set, to avoid unnecessary overhead.
1259 	 * A local RDP can be used to do the flush.
1260 	 */
1261 	if (MACHINE_HAS_RDP && !(pte_val(*ptep) & _PAGE_PROTECT))
1262 		__ptep_rdp(address, ptep, 0, 0, 1);
1263 }
1264 #define flush_tlb_fix_spurious_fault flush_tlb_fix_spurious_fault
1265 
1266 void ptep_reset_dat_prot(struct mm_struct *mm, unsigned long addr, pte_t *ptep,
1267 			 pte_t new);
1268 
1269 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1270 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
1271 					unsigned long addr, pte_t *ptep,
1272 					pte_t entry, int dirty)
1273 {
1274 	if (pte_same(*ptep, entry))
1275 		return 0;
1276 	if (MACHINE_HAS_RDP && !mm_has_pgste(vma->vm_mm) && pte_allow_rdp(*ptep, entry))
1277 		ptep_reset_dat_prot(vma->vm_mm, addr, ptep, entry);
1278 	else
1279 		ptep_xchg_direct(vma->vm_mm, addr, ptep, entry);
1280 	return 1;
1281 }
1282 
1283 /*
1284  * Additional functions to handle KVM guest page tables
1285  */
1286 void ptep_set_pte_at(struct mm_struct *mm, unsigned long addr,
1287 		     pte_t *ptep, pte_t entry);
1288 void ptep_set_notify(struct mm_struct *mm, unsigned long addr, pte_t *ptep);
1289 void ptep_notify(struct mm_struct *mm, unsigned long addr,
1290 		 pte_t *ptep, unsigned long bits);
1291 int ptep_force_prot(struct mm_struct *mm, unsigned long gaddr,
1292 		    pte_t *ptep, int prot, unsigned long bit);
1293 void ptep_zap_unused(struct mm_struct *mm, unsigned long addr,
1294 		     pte_t *ptep , int reset);
1295 void ptep_zap_key(struct mm_struct *mm, unsigned long addr, pte_t *ptep);
1296 int ptep_shadow_pte(struct mm_struct *mm, unsigned long saddr,
1297 		    pte_t *sptep, pte_t *tptep, pte_t pte);
1298 void ptep_unshadow_pte(struct mm_struct *mm, unsigned long saddr, pte_t *ptep);
1299 
1300 bool ptep_test_and_clear_uc(struct mm_struct *mm, unsigned long address,
1301 			    pte_t *ptep);
1302 int set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
1303 			  unsigned char key, bool nq);
1304 int cond_set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
1305 			       unsigned char key, unsigned char *oldkey,
1306 			       bool nq, bool mr, bool mc);
1307 int reset_guest_reference_bit(struct mm_struct *mm, unsigned long addr);
1308 int get_guest_storage_key(struct mm_struct *mm, unsigned long addr,
1309 			  unsigned char *key);
1310 
1311 int set_pgste_bits(struct mm_struct *mm, unsigned long addr,
1312 				unsigned long bits, unsigned long value);
1313 int get_pgste(struct mm_struct *mm, unsigned long hva, unsigned long *pgstep);
1314 int pgste_perform_essa(struct mm_struct *mm, unsigned long hva, int orc,
1315 			unsigned long *oldpte, unsigned long *oldpgste);
1316 void gmap_pmdp_csp(struct mm_struct *mm, unsigned long vmaddr);
1317 void gmap_pmdp_invalidate(struct mm_struct *mm, unsigned long vmaddr);
1318 void gmap_pmdp_idte_local(struct mm_struct *mm, unsigned long vmaddr);
1319 void gmap_pmdp_idte_global(struct mm_struct *mm, unsigned long vmaddr);
1320 
1321 #define pgprot_writecombine	pgprot_writecombine
1322 pgprot_t pgprot_writecombine(pgprot_t prot);
1323 
1324 #define pgprot_writethrough	pgprot_writethrough
1325 pgprot_t pgprot_writethrough(pgprot_t prot);
1326 
1327 /*
1328  * Set multiple PTEs to consecutive pages with a single call.  All PTEs
1329  * are within the same folio, PMD and VMA.
1330  */
1331 static inline void set_ptes(struct mm_struct *mm, unsigned long addr,
1332 			      pte_t *ptep, pte_t entry, unsigned int nr)
1333 {
1334 	if (pte_present(entry))
1335 		entry = clear_pte_bit(entry, __pgprot(_PAGE_UNUSED));
1336 	if (mm_has_pgste(mm)) {
1337 		for (;;) {
1338 			ptep_set_pte_at(mm, addr, ptep, entry);
1339 			if (--nr == 0)
1340 				break;
1341 			ptep++;
1342 			entry = __pte(pte_val(entry) + PAGE_SIZE);
1343 			addr += PAGE_SIZE;
1344 		}
1345 	} else {
1346 		for (;;) {
1347 			set_pte(ptep, entry);
1348 			if (--nr == 0)
1349 				break;
1350 			ptep++;
1351 			entry = __pte(pte_val(entry) + PAGE_SIZE);
1352 		}
1353 	}
1354 }
1355 #define set_ptes set_ptes
1356 
1357 /*
1358  * Conversion functions: convert a page and protection to a page entry,
1359  * and a page entry and page directory to the page they refer to.
1360  */
1361 static inline pte_t mk_pte_phys(unsigned long physpage, pgprot_t pgprot)
1362 {
1363 	pte_t __pte;
1364 
1365 	__pte = __pte(physpage | pgprot_val(pgprot));
1366 	if (!MACHINE_HAS_NX)
1367 		__pte = clear_pte_bit(__pte, __pgprot(_PAGE_NOEXEC));
1368 	return pte_mkyoung(__pte);
1369 }
1370 
1371 static inline pte_t mk_pte(struct page *page, pgprot_t pgprot)
1372 {
1373 	unsigned long physpage = page_to_phys(page);
1374 	pte_t __pte = mk_pte_phys(physpage, pgprot);
1375 
1376 	if (pte_write(__pte) && PageDirty(page))
1377 		__pte = pte_mkdirty(__pte);
1378 	return __pte;
1379 }
1380 
1381 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD-1))
1382 #define p4d_index(address) (((address) >> P4D_SHIFT) & (PTRS_PER_P4D-1))
1383 #define pud_index(address) (((address) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
1384 #define pmd_index(address) (((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1))
1385 
1386 #define p4d_deref(pud) ((unsigned long)__va(p4d_val(pud) & _REGION_ENTRY_ORIGIN))
1387 #define pgd_deref(pgd) ((unsigned long)__va(pgd_val(pgd) & _REGION_ENTRY_ORIGIN))
1388 
1389 static inline unsigned long pmd_deref(pmd_t pmd)
1390 {
1391 	unsigned long origin_mask;
1392 
1393 	origin_mask = _SEGMENT_ENTRY_ORIGIN;
1394 	if (pmd_large(pmd))
1395 		origin_mask = _SEGMENT_ENTRY_ORIGIN_LARGE;
1396 	return (unsigned long)__va(pmd_val(pmd) & origin_mask);
1397 }
1398 
1399 static inline unsigned long pmd_pfn(pmd_t pmd)
1400 {
1401 	return __pa(pmd_deref(pmd)) >> PAGE_SHIFT;
1402 }
1403 
1404 static inline unsigned long pud_deref(pud_t pud)
1405 {
1406 	unsigned long origin_mask;
1407 
1408 	origin_mask = _REGION_ENTRY_ORIGIN;
1409 	if (pud_leaf(pud))
1410 		origin_mask = _REGION3_ENTRY_ORIGIN_LARGE;
1411 	return (unsigned long)__va(pud_val(pud) & origin_mask);
1412 }
1413 
1414 static inline unsigned long pud_pfn(pud_t pud)
1415 {
1416 	return __pa(pud_deref(pud)) >> PAGE_SHIFT;
1417 }
1418 
1419 /*
1420  * The pgd_offset function *always* adds the index for the top-level
1421  * region/segment table. This is done to get a sequence like the
1422  * following to work:
1423  *	pgdp = pgd_offset(current->mm, addr);
1424  *	pgd = READ_ONCE(*pgdp);
1425  *	p4dp = p4d_offset(&pgd, addr);
1426  *	...
1427  * The subsequent p4d_offset, pud_offset and pmd_offset functions
1428  * only add an index if they dereferenced the pointer.
1429  */
1430 static inline pgd_t *pgd_offset_raw(pgd_t *pgd, unsigned long address)
1431 {
1432 	unsigned long rste;
1433 	unsigned int shift;
1434 
1435 	/* Get the first entry of the top level table */
1436 	rste = pgd_val(*pgd);
1437 	/* Pick up the shift from the table type of the first entry */
1438 	shift = ((rste & _REGION_ENTRY_TYPE_MASK) >> 2) * 11 + 20;
1439 	return pgd + ((address >> shift) & (PTRS_PER_PGD - 1));
1440 }
1441 
1442 #define pgd_offset(mm, address) pgd_offset_raw(READ_ONCE((mm)->pgd), address)
1443 
1444 static inline p4d_t *p4d_offset_lockless(pgd_t *pgdp, pgd_t pgd, unsigned long address)
1445 {
1446 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) >= _REGION_ENTRY_TYPE_R1)
1447 		return (p4d_t *) pgd_deref(pgd) + p4d_index(address);
1448 	return (p4d_t *) pgdp;
1449 }
1450 #define p4d_offset_lockless p4d_offset_lockless
1451 
1452 static inline p4d_t *p4d_offset(pgd_t *pgdp, unsigned long address)
1453 {
1454 	return p4d_offset_lockless(pgdp, *pgdp, address);
1455 }
1456 
1457 static inline pud_t *pud_offset_lockless(p4d_t *p4dp, p4d_t p4d, unsigned long address)
1458 {
1459 	if ((p4d_val(p4d) & _REGION_ENTRY_TYPE_MASK) >= _REGION_ENTRY_TYPE_R2)
1460 		return (pud_t *) p4d_deref(p4d) + pud_index(address);
1461 	return (pud_t *) p4dp;
1462 }
1463 #define pud_offset_lockless pud_offset_lockless
1464 
1465 static inline pud_t *pud_offset(p4d_t *p4dp, unsigned long address)
1466 {
1467 	return pud_offset_lockless(p4dp, *p4dp, address);
1468 }
1469 #define pud_offset pud_offset
1470 
1471 static inline pmd_t *pmd_offset_lockless(pud_t *pudp, pud_t pud, unsigned long address)
1472 {
1473 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) >= _REGION_ENTRY_TYPE_R3)
1474 		return (pmd_t *) pud_deref(pud) + pmd_index(address);
1475 	return (pmd_t *) pudp;
1476 }
1477 #define pmd_offset_lockless pmd_offset_lockless
1478 
1479 static inline pmd_t *pmd_offset(pud_t *pudp, unsigned long address)
1480 {
1481 	return pmd_offset_lockless(pudp, *pudp, address);
1482 }
1483 #define pmd_offset pmd_offset
1484 
1485 static inline unsigned long pmd_page_vaddr(pmd_t pmd)
1486 {
1487 	return (unsigned long) pmd_deref(pmd);
1488 }
1489 
1490 static inline bool gup_fast_permitted(unsigned long start, unsigned long end)
1491 {
1492 	return end <= current->mm->context.asce_limit;
1493 }
1494 #define gup_fast_permitted gup_fast_permitted
1495 
1496 #define pfn_pte(pfn, pgprot)	mk_pte_phys(((pfn) << PAGE_SHIFT), (pgprot))
1497 #define pte_pfn(x) (pte_val(x) >> PAGE_SHIFT)
1498 #define pte_page(x) pfn_to_page(pte_pfn(x))
1499 
1500 #define pmd_page(pmd) pfn_to_page(pmd_pfn(pmd))
1501 #define pud_page(pud) pfn_to_page(pud_pfn(pud))
1502 #define p4d_page(p4d) pfn_to_page(p4d_pfn(p4d))
1503 #define pgd_page(pgd) pfn_to_page(pgd_pfn(pgd))
1504 
1505 static inline pmd_t pmd_wrprotect(pmd_t pmd)
1506 {
1507 	pmd = clear_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_WRITE));
1508 	return set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_PROTECT));
1509 }
1510 
1511 static inline pmd_t pmd_mkwrite_novma(pmd_t pmd)
1512 {
1513 	pmd = set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_WRITE));
1514 	if (pmd_val(pmd) & _SEGMENT_ENTRY_DIRTY)
1515 		pmd = clear_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_PROTECT));
1516 	return pmd;
1517 }
1518 
1519 static inline pmd_t pmd_mkclean(pmd_t pmd)
1520 {
1521 	pmd = clear_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_DIRTY));
1522 	return set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_PROTECT));
1523 }
1524 
1525 static inline pmd_t pmd_mkdirty(pmd_t pmd)
1526 {
1527 	pmd = set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_DIRTY | _SEGMENT_ENTRY_SOFT_DIRTY));
1528 	if (pmd_val(pmd) & _SEGMENT_ENTRY_WRITE)
1529 		pmd = clear_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_PROTECT));
1530 	return pmd;
1531 }
1532 
1533 static inline pud_t pud_wrprotect(pud_t pud)
1534 {
1535 	pud = clear_pud_bit(pud, __pgprot(_REGION3_ENTRY_WRITE));
1536 	return set_pud_bit(pud, __pgprot(_REGION_ENTRY_PROTECT));
1537 }
1538 
1539 static inline pud_t pud_mkwrite(pud_t pud)
1540 {
1541 	pud = set_pud_bit(pud, __pgprot(_REGION3_ENTRY_WRITE));
1542 	if (pud_val(pud) & _REGION3_ENTRY_DIRTY)
1543 		pud = clear_pud_bit(pud, __pgprot(_REGION_ENTRY_PROTECT));
1544 	return pud;
1545 }
1546 
1547 static inline pud_t pud_mkclean(pud_t pud)
1548 {
1549 	pud = clear_pud_bit(pud, __pgprot(_REGION3_ENTRY_DIRTY));
1550 	return set_pud_bit(pud, __pgprot(_REGION_ENTRY_PROTECT));
1551 }
1552 
1553 static inline pud_t pud_mkdirty(pud_t pud)
1554 {
1555 	pud = set_pud_bit(pud, __pgprot(_REGION3_ENTRY_DIRTY | _REGION3_ENTRY_SOFT_DIRTY));
1556 	if (pud_val(pud) & _REGION3_ENTRY_WRITE)
1557 		pud = clear_pud_bit(pud, __pgprot(_REGION_ENTRY_PROTECT));
1558 	return pud;
1559 }
1560 
1561 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE)
1562 static inline unsigned long massage_pgprot_pmd(pgprot_t pgprot)
1563 {
1564 	/*
1565 	 * pgprot is PAGE_NONE, PAGE_RO, PAGE_RX, PAGE_RW or PAGE_RWX
1566 	 * (see __Pxxx / __Sxxx). Convert to segment table entry format.
1567 	 */
1568 	if (pgprot_val(pgprot) == pgprot_val(PAGE_NONE))
1569 		return pgprot_val(SEGMENT_NONE);
1570 	if (pgprot_val(pgprot) == pgprot_val(PAGE_RO))
1571 		return pgprot_val(SEGMENT_RO);
1572 	if (pgprot_val(pgprot) == pgprot_val(PAGE_RX))
1573 		return pgprot_val(SEGMENT_RX);
1574 	if (pgprot_val(pgprot) == pgprot_val(PAGE_RW))
1575 		return pgprot_val(SEGMENT_RW);
1576 	return pgprot_val(SEGMENT_RWX);
1577 }
1578 
1579 static inline pmd_t pmd_mkyoung(pmd_t pmd)
1580 {
1581 	pmd = set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_YOUNG));
1582 	if (pmd_val(pmd) & _SEGMENT_ENTRY_READ)
1583 		pmd = clear_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_INVALID));
1584 	return pmd;
1585 }
1586 
1587 static inline pmd_t pmd_mkold(pmd_t pmd)
1588 {
1589 	pmd = clear_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_YOUNG));
1590 	return set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_INVALID));
1591 }
1592 
1593 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
1594 {
1595 	unsigned long mask;
1596 
1597 	mask  = _SEGMENT_ENTRY_ORIGIN_LARGE;
1598 	mask |= _SEGMENT_ENTRY_DIRTY;
1599 	mask |= _SEGMENT_ENTRY_YOUNG;
1600 	mask |=	_SEGMENT_ENTRY_LARGE;
1601 	mask |= _SEGMENT_ENTRY_SOFT_DIRTY;
1602 	pmd = __pmd(pmd_val(pmd) & mask);
1603 	pmd = set_pmd_bit(pmd, __pgprot(massage_pgprot_pmd(newprot)));
1604 	if (!(pmd_val(pmd) & _SEGMENT_ENTRY_DIRTY))
1605 		pmd = set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_PROTECT));
1606 	if (!(pmd_val(pmd) & _SEGMENT_ENTRY_YOUNG))
1607 		pmd = set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_INVALID));
1608 	return pmd;
1609 }
1610 
1611 static inline pmd_t mk_pmd_phys(unsigned long physpage, pgprot_t pgprot)
1612 {
1613 	return __pmd(physpage + massage_pgprot_pmd(pgprot));
1614 }
1615 
1616 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLB_PAGE */
1617 
1618 static inline void __pmdp_csp(pmd_t *pmdp)
1619 {
1620 	csp((unsigned int *)pmdp + 1, pmd_val(*pmdp),
1621 	    pmd_val(*pmdp) | _SEGMENT_ENTRY_INVALID);
1622 }
1623 
1624 #define IDTE_GLOBAL	0
1625 #define IDTE_LOCAL	1
1626 
1627 #define IDTE_PTOA	0x0800
1628 #define IDTE_NODAT	0x1000
1629 #define IDTE_GUEST_ASCE	0x2000
1630 
1631 static __always_inline void __pmdp_idte(unsigned long addr, pmd_t *pmdp,
1632 					unsigned long opt, unsigned long asce,
1633 					int local)
1634 {
1635 	unsigned long sto;
1636 
1637 	sto = __pa(pmdp) - pmd_index(addr) * sizeof(pmd_t);
1638 	if (__builtin_constant_p(opt) && opt == 0) {
1639 		/* flush without guest asce */
1640 		asm volatile(
1641 			"	idte	%[r1],0,%[r2],%[m4]"
1642 			: "+m" (*pmdp)
1643 			: [r1] "a" (sto), [r2] "a" ((addr & HPAGE_MASK)),
1644 			  [m4] "i" (local)
1645 			: "cc" );
1646 	} else {
1647 		/* flush with guest asce */
1648 		asm volatile(
1649 			"	idte	%[r1],%[r3],%[r2],%[m4]"
1650 			: "+m" (*pmdp)
1651 			: [r1] "a" (sto), [r2] "a" ((addr & HPAGE_MASK) | opt),
1652 			  [r3] "a" (asce), [m4] "i" (local)
1653 			: "cc" );
1654 	}
1655 }
1656 
1657 static __always_inline void __pudp_idte(unsigned long addr, pud_t *pudp,
1658 					unsigned long opt, unsigned long asce,
1659 					int local)
1660 {
1661 	unsigned long r3o;
1662 
1663 	r3o = __pa(pudp) - pud_index(addr) * sizeof(pud_t);
1664 	r3o |= _ASCE_TYPE_REGION3;
1665 	if (__builtin_constant_p(opt) && opt == 0) {
1666 		/* flush without guest asce */
1667 		asm volatile(
1668 			"	idte	%[r1],0,%[r2],%[m4]"
1669 			: "+m" (*pudp)
1670 			: [r1] "a" (r3o), [r2] "a" ((addr & PUD_MASK)),
1671 			  [m4] "i" (local)
1672 			: "cc");
1673 	} else {
1674 		/* flush with guest asce */
1675 		asm volatile(
1676 			"	idte	%[r1],%[r3],%[r2],%[m4]"
1677 			: "+m" (*pudp)
1678 			: [r1] "a" (r3o), [r2] "a" ((addr & PUD_MASK) | opt),
1679 			  [r3] "a" (asce), [m4] "i" (local)
1680 			: "cc" );
1681 	}
1682 }
1683 
1684 pmd_t pmdp_xchg_direct(struct mm_struct *, unsigned long, pmd_t *, pmd_t);
1685 pmd_t pmdp_xchg_lazy(struct mm_struct *, unsigned long, pmd_t *, pmd_t);
1686 pud_t pudp_xchg_direct(struct mm_struct *, unsigned long, pud_t *, pud_t);
1687 
1688 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1689 
1690 #define __HAVE_ARCH_PGTABLE_DEPOSIT
1691 void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
1692 				pgtable_t pgtable);
1693 
1694 #define __HAVE_ARCH_PGTABLE_WITHDRAW
1695 pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp);
1696 
1697 #define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
1698 static inline int pmdp_set_access_flags(struct vm_area_struct *vma,
1699 					unsigned long addr, pmd_t *pmdp,
1700 					pmd_t entry, int dirty)
1701 {
1702 	VM_BUG_ON(addr & ~HPAGE_MASK);
1703 
1704 	entry = pmd_mkyoung(entry);
1705 	if (dirty)
1706 		entry = pmd_mkdirty(entry);
1707 	if (pmd_val(*pmdp) == pmd_val(entry))
1708 		return 0;
1709 	pmdp_xchg_direct(vma->vm_mm, addr, pmdp, entry);
1710 	return 1;
1711 }
1712 
1713 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1714 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1715 					    unsigned long addr, pmd_t *pmdp)
1716 {
1717 	pmd_t pmd = *pmdp;
1718 
1719 	pmd = pmdp_xchg_direct(vma->vm_mm, addr, pmdp, pmd_mkold(pmd));
1720 	return pmd_young(pmd);
1721 }
1722 
1723 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
1724 static inline int pmdp_clear_flush_young(struct vm_area_struct *vma,
1725 					 unsigned long addr, pmd_t *pmdp)
1726 {
1727 	VM_BUG_ON(addr & ~HPAGE_MASK);
1728 	return pmdp_test_and_clear_young(vma, addr, pmdp);
1729 }
1730 
1731 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1732 			      pmd_t *pmdp, pmd_t entry)
1733 {
1734 	if (!MACHINE_HAS_NX)
1735 		entry = clear_pmd_bit(entry, __pgprot(_SEGMENT_ENTRY_NOEXEC));
1736 	set_pmd(pmdp, entry);
1737 }
1738 
1739 static inline pmd_t pmd_mkhuge(pmd_t pmd)
1740 {
1741 	pmd = set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_LARGE));
1742 	pmd = set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_YOUNG));
1743 	return set_pmd_bit(pmd, __pgprot(_SEGMENT_ENTRY_PROTECT));
1744 }
1745 
1746 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR
1747 static inline pmd_t pmdp_huge_get_and_clear(struct mm_struct *mm,
1748 					    unsigned long addr, pmd_t *pmdp)
1749 {
1750 	return pmdp_xchg_direct(mm, addr, pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
1751 }
1752 
1753 #define __HAVE_ARCH_PMDP_HUGE_GET_AND_CLEAR_FULL
1754 static inline pmd_t pmdp_huge_get_and_clear_full(struct vm_area_struct *vma,
1755 						 unsigned long addr,
1756 						 pmd_t *pmdp, int full)
1757 {
1758 	if (full) {
1759 		pmd_t pmd = *pmdp;
1760 		set_pmd(pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
1761 		return pmd;
1762 	}
1763 	return pmdp_xchg_lazy(vma->vm_mm, addr, pmdp, __pmd(_SEGMENT_ENTRY_EMPTY));
1764 }
1765 
1766 #define __HAVE_ARCH_PMDP_HUGE_CLEAR_FLUSH
1767 static inline pmd_t pmdp_huge_clear_flush(struct vm_area_struct *vma,
1768 					  unsigned long addr, pmd_t *pmdp)
1769 {
1770 	return pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp);
1771 }
1772 
1773 #define __HAVE_ARCH_PMDP_INVALIDATE
1774 static inline pmd_t pmdp_invalidate(struct vm_area_struct *vma,
1775 				   unsigned long addr, pmd_t *pmdp)
1776 {
1777 	pmd_t pmd;
1778 
1779 	VM_WARN_ON_ONCE(!pmd_present(*pmdp));
1780 	pmd = __pmd(pmd_val(*pmdp) | _SEGMENT_ENTRY_INVALID);
1781 	return pmdp_xchg_direct(vma->vm_mm, addr, pmdp, pmd);
1782 }
1783 
1784 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1785 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1786 				      unsigned long addr, pmd_t *pmdp)
1787 {
1788 	pmd_t pmd = *pmdp;
1789 
1790 	if (pmd_write(pmd))
1791 		pmd = pmdp_xchg_lazy(mm, addr, pmdp, pmd_wrprotect(pmd));
1792 }
1793 
1794 static inline pmd_t pmdp_collapse_flush(struct vm_area_struct *vma,
1795 					unsigned long address,
1796 					pmd_t *pmdp)
1797 {
1798 	return pmdp_huge_get_and_clear(vma->vm_mm, address, pmdp);
1799 }
1800 #define pmdp_collapse_flush pmdp_collapse_flush
1801 
1802 #define pfn_pmd(pfn, pgprot)	mk_pmd_phys(((pfn) << PAGE_SHIFT), (pgprot))
1803 #define mk_pmd(page, pgprot)	pfn_pmd(page_to_pfn(page), (pgprot))
1804 
1805 static inline int pmd_trans_huge(pmd_t pmd)
1806 {
1807 	return pmd_val(pmd) & _SEGMENT_ENTRY_LARGE;
1808 }
1809 
1810 #define has_transparent_hugepage has_transparent_hugepage
1811 static inline int has_transparent_hugepage(void)
1812 {
1813 	return MACHINE_HAS_EDAT1 ? 1 : 0;
1814 }
1815 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1816 
1817 /*
1818  * 64 bit swap entry format:
1819  * A page-table entry has some bits we have to treat in a special way.
1820  * Bits 54 and 63 are used to indicate the page type. Bit 53 marks the pte
1821  * as invalid.
1822  * A swap pte is indicated by bit pattern (pte & 0x201) == 0x200
1823  * |			  offset			|E11XX|type |S0|
1824  * |0000000000111111111122222222223333333333444444444455|55555|55566|66|
1825  * |0123456789012345678901234567890123456789012345678901|23456|78901|23|
1826  *
1827  * Bits 0-51 store the offset.
1828  * Bit 52 (E) is used to remember PG_anon_exclusive.
1829  * Bits 57-61 store the type.
1830  * Bit 62 (S) is used for softdirty tracking.
1831  * Bits 55 and 56 (X) are unused.
1832  */
1833 
1834 #define __SWP_OFFSET_MASK	((1UL << 52) - 1)
1835 #define __SWP_OFFSET_SHIFT	12
1836 #define __SWP_TYPE_MASK		((1UL << 5) - 1)
1837 #define __SWP_TYPE_SHIFT	2
1838 
1839 static inline pte_t mk_swap_pte(unsigned long type, unsigned long offset)
1840 {
1841 	unsigned long pteval;
1842 
1843 	pteval = _PAGE_INVALID | _PAGE_PROTECT;
1844 	pteval |= (offset & __SWP_OFFSET_MASK) << __SWP_OFFSET_SHIFT;
1845 	pteval |= (type & __SWP_TYPE_MASK) << __SWP_TYPE_SHIFT;
1846 	return __pte(pteval);
1847 }
1848 
1849 static inline unsigned long __swp_type(swp_entry_t entry)
1850 {
1851 	return (entry.val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK;
1852 }
1853 
1854 static inline unsigned long __swp_offset(swp_entry_t entry)
1855 {
1856 	return (entry.val >> __SWP_OFFSET_SHIFT) & __SWP_OFFSET_MASK;
1857 }
1858 
1859 static inline swp_entry_t __swp_entry(unsigned long type, unsigned long offset)
1860 {
1861 	return (swp_entry_t) { pte_val(mk_swap_pte(type, offset)) };
1862 }
1863 
1864 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
1865 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
1866 
1867 extern int vmem_add_mapping(unsigned long start, unsigned long size);
1868 extern void vmem_remove_mapping(unsigned long start, unsigned long size);
1869 extern int __vmem_map_4k_page(unsigned long addr, unsigned long phys, pgprot_t prot, bool alloc);
1870 extern int vmem_map_4k_page(unsigned long addr, unsigned long phys, pgprot_t prot);
1871 extern void vmem_unmap_4k_page(unsigned long addr);
1872 extern pte_t *vmem_get_alloc_pte(unsigned long addr, bool alloc);
1873 extern int s390_enable_sie(void);
1874 extern int s390_enable_skey(void);
1875 extern void s390_reset_cmma(struct mm_struct *mm);
1876 
1877 /* s390 has a private copy of get unmapped area to deal with cache synonyms */
1878 #define HAVE_ARCH_UNMAPPED_AREA
1879 #define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1880 
1881 #define pmd_pgtable(pmd) \
1882 	((pgtable_t)__va(pmd_val(pmd) & -sizeof(pte_t)*PTRS_PER_PTE))
1883 
1884 #endif /* _S390_PAGE_H */
1885