xref: /linux/arch/s390/include/asm/pgtable.h (revision 9307c29524502c21f0e8a6d96d850b2f5bc0bd9a)
1 /*
2  *  S390 version
3  *    Copyright IBM Corp. 1999, 2000
4  *    Author(s): Hartmut Penner (hp@de.ibm.com)
5  *               Ulrich Weigand (weigand@de.ibm.com)
6  *               Martin Schwidefsky (schwidefsky@de.ibm.com)
7  *
8  *  Derived from "include/asm-i386/pgtable.h"
9  */
10 
11 #ifndef _ASM_S390_PGTABLE_H
12 #define _ASM_S390_PGTABLE_H
13 
14 /*
15  * The Linux memory management assumes a three-level page table setup. For
16  * s390 31 bit we "fold" the mid level into the top-level page table, so
17  * that we physically have the same two-level page table as the s390 mmu
18  * expects in 31 bit mode. For s390 64 bit we use three of the five levels
19  * the hardware provides (region first and region second tables are not
20  * used).
21  *
22  * The "pgd_xxx()" functions are trivial for a folded two-level
23  * setup: the pgd is never bad, and a pmd always exists (as it's folded
24  * into the pgd entry)
25  *
26  * This file contains the functions and defines necessary to modify and use
27  * the S390 page table tree.
28  */
29 #ifndef __ASSEMBLY__
30 #include <linux/sched.h>
31 #include <linux/mm_types.h>
32 #include <linux/page-flags.h>
33 #include <asm/bug.h>
34 #include <asm/page.h>
35 
36 extern pgd_t swapper_pg_dir[] __attribute__ ((aligned (4096)));
37 extern void paging_init(void);
38 extern void vmem_map_init(void);
39 
40 /*
41  * The S390 doesn't have any external MMU info: the kernel page
42  * tables contain all the necessary information.
43  */
44 #define update_mmu_cache(vma, address, ptep)     do { } while (0)
45 #define update_mmu_cache_pmd(vma, address, ptep) do { } while (0)
46 
47 /*
48  * ZERO_PAGE is a global shared page that is always zero; used
49  * for zero-mapped memory areas etc..
50  */
51 
52 extern unsigned long empty_zero_page;
53 extern unsigned long zero_page_mask;
54 
55 #define ZERO_PAGE(vaddr) \
56 	(virt_to_page((void *)(empty_zero_page + \
57 	 (((unsigned long)(vaddr)) &zero_page_mask))))
58 #define __HAVE_COLOR_ZERO_PAGE
59 
60 /* TODO: s390 cannot support io_remap_pfn_range... */
61 #endif /* !__ASSEMBLY__ */
62 
63 /*
64  * PMD_SHIFT determines the size of the area a second-level page
65  * table can map
66  * PGDIR_SHIFT determines what a third-level page table entry can map
67  */
68 #ifndef CONFIG_64BIT
69 # define PMD_SHIFT	20
70 # define PUD_SHIFT	20
71 # define PGDIR_SHIFT	20
72 #else /* CONFIG_64BIT */
73 # define PMD_SHIFT	20
74 # define PUD_SHIFT	31
75 # define PGDIR_SHIFT	42
76 #endif /* CONFIG_64BIT */
77 
78 #define PMD_SIZE        (1UL << PMD_SHIFT)
79 #define PMD_MASK        (~(PMD_SIZE-1))
80 #define PUD_SIZE	(1UL << PUD_SHIFT)
81 #define PUD_MASK	(~(PUD_SIZE-1))
82 #define PGDIR_SIZE	(1UL << PGDIR_SHIFT)
83 #define PGDIR_MASK	(~(PGDIR_SIZE-1))
84 
85 /*
86  * entries per page directory level: the S390 is two-level, so
87  * we don't really have any PMD directory physically.
88  * for S390 segment-table entries are combined to one PGD
89  * that leads to 1024 pte per pgd
90  */
91 #define PTRS_PER_PTE	256
92 #ifndef CONFIG_64BIT
93 #define PTRS_PER_PMD	1
94 #define PTRS_PER_PUD	1
95 #else /* CONFIG_64BIT */
96 #define PTRS_PER_PMD	2048
97 #define PTRS_PER_PUD	2048
98 #endif /* CONFIG_64BIT */
99 #define PTRS_PER_PGD	2048
100 
101 #define FIRST_USER_ADDRESS  0
102 
103 #define pte_ERROR(e) \
104 	printk("%s:%d: bad pte %p.\n", __FILE__, __LINE__, (void *) pte_val(e))
105 #define pmd_ERROR(e) \
106 	printk("%s:%d: bad pmd %p.\n", __FILE__, __LINE__, (void *) pmd_val(e))
107 #define pud_ERROR(e) \
108 	printk("%s:%d: bad pud %p.\n", __FILE__, __LINE__, (void *) pud_val(e))
109 #define pgd_ERROR(e) \
110 	printk("%s:%d: bad pgd %p.\n", __FILE__, __LINE__, (void *) pgd_val(e))
111 
112 #ifndef __ASSEMBLY__
113 /*
114  * The vmalloc and module area will always be on the topmost area of the kernel
115  * mapping. We reserve 96MB (31bit) / 128GB (64bit) for vmalloc and modules.
116  * On 64 bit kernels we have a 2GB area at the top of the vmalloc area where
117  * modules will reside. That makes sure that inter module branches always
118  * happen without trampolines and in addition the placement within a 2GB frame
119  * is branch prediction unit friendly.
120  */
121 extern unsigned long VMALLOC_START;
122 extern unsigned long VMALLOC_END;
123 extern struct page *vmemmap;
124 
125 #define VMEM_MAX_PHYS ((unsigned long) vmemmap)
126 
127 #ifdef CONFIG_64BIT
128 extern unsigned long MODULES_VADDR;
129 extern unsigned long MODULES_END;
130 #define MODULES_VADDR	MODULES_VADDR
131 #define MODULES_END	MODULES_END
132 #define MODULES_LEN	(1UL << 31)
133 #endif
134 
135 /*
136  * A 31 bit pagetable entry of S390 has following format:
137  *  |   PFRA          |    |  OS  |
138  * 0                   0IP0
139  * 00000000001111111111222222222233
140  * 01234567890123456789012345678901
141  *
142  * I Page-Invalid Bit:    Page is not available for address-translation
143  * P Page-Protection Bit: Store access not possible for page
144  *
145  * A 31 bit segmenttable entry of S390 has following format:
146  *  |   P-table origin      |  |PTL
147  * 0                         IC
148  * 00000000001111111111222222222233
149  * 01234567890123456789012345678901
150  *
151  * I Segment-Invalid Bit:    Segment is not available for address-translation
152  * C Common-Segment Bit:     Segment is not private (PoP 3-30)
153  * PTL Page-Table-Length:    Page-table length (PTL+1*16 entries -> up to 256)
154  *
155  * The 31 bit segmenttable origin of S390 has following format:
156  *
157  *  |S-table origin   |     | STL |
158  * X                   **GPS
159  * 00000000001111111111222222222233
160  * 01234567890123456789012345678901
161  *
162  * X Space-Switch event:
163  * G Segment-Invalid Bit:     *
164  * P Private-Space Bit:       Segment is not private (PoP 3-30)
165  * S Storage-Alteration:
166  * STL Segment-Table-Length:  Segment-table length (STL+1*16 entries -> up to 2048)
167  *
168  * A 64 bit pagetable entry of S390 has following format:
169  * |			 PFRA			      |0IPC|  OS  |
170  * 0000000000111111111122222222223333333333444444444455555555556666
171  * 0123456789012345678901234567890123456789012345678901234567890123
172  *
173  * I Page-Invalid Bit:    Page is not available for address-translation
174  * P Page-Protection Bit: Store access not possible for page
175  * C Change-bit override: HW is not required to set change bit
176  *
177  * A 64 bit segmenttable entry of S390 has following format:
178  * |        P-table origin                              |      TT
179  * 0000000000111111111122222222223333333333444444444455555555556666
180  * 0123456789012345678901234567890123456789012345678901234567890123
181  *
182  * I Segment-Invalid Bit:    Segment is not available for address-translation
183  * C Common-Segment Bit:     Segment is not private (PoP 3-30)
184  * P Page-Protection Bit: Store access not possible for page
185  * TT Type 00
186  *
187  * A 64 bit region table entry of S390 has following format:
188  * |        S-table origin                             |   TF  TTTL
189  * 0000000000111111111122222222223333333333444444444455555555556666
190  * 0123456789012345678901234567890123456789012345678901234567890123
191  *
192  * I Segment-Invalid Bit:    Segment is not available for address-translation
193  * TT Type 01
194  * TF
195  * TL Table length
196  *
197  * The 64 bit regiontable origin of S390 has following format:
198  * |      region table origon                          |       DTTL
199  * 0000000000111111111122222222223333333333444444444455555555556666
200  * 0123456789012345678901234567890123456789012345678901234567890123
201  *
202  * X Space-Switch event:
203  * G Segment-Invalid Bit:
204  * P Private-Space Bit:
205  * S Storage-Alteration:
206  * R Real space
207  * TL Table-Length:
208  *
209  * A storage key has the following format:
210  * | ACC |F|R|C|0|
211  *  0   3 4 5 6 7
212  * ACC: access key
213  * F  : fetch protection bit
214  * R  : referenced bit
215  * C  : changed bit
216  */
217 
218 /* Hardware bits in the page table entry */
219 #define _PAGE_CO	0x100		/* HW Change-bit override */
220 #define _PAGE_RO	0x200		/* HW read-only bit  */
221 #define _PAGE_INVALID	0x400		/* HW invalid bit    */
222 
223 /* Software bits in the page table entry */
224 #define _PAGE_SWT	0x001		/* SW pte type bit t */
225 #define _PAGE_SWX	0x002		/* SW pte type bit x */
226 #define _PAGE_SWC	0x004		/* SW pte changed bit */
227 #define _PAGE_SWR	0x008		/* SW pte referenced bit */
228 #define _PAGE_SWW	0x010		/* SW pte write bit */
229 #define _PAGE_SPECIAL	0x020		/* SW associated with special page */
230 #define __HAVE_ARCH_PTE_SPECIAL
231 
232 /* Set of bits not changed in pte_modify */
233 #define _PAGE_CHG_MASK		(PAGE_MASK | _PAGE_SPECIAL | _PAGE_CO | \
234 				 _PAGE_SWC | _PAGE_SWR)
235 
236 /* Six different types of pages. */
237 #define _PAGE_TYPE_EMPTY	0x400
238 #define _PAGE_TYPE_NONE		0x401
239 #define _PAGE_TYPE_SWAP		0x403
240 #define _PAGE_TYPE_FILE		0x601	/* bit 0x002 is used for offset !! */
241 #define _PAGE_TYPE_RO		0x200
242 #define _PAGE_TYPE_RW		0x000
243 
244 /*
245  * Only four types for huge pages, using the invalid bit and protection bit
246  * of a segment table entry.
247  */
248 #define _HPAGE_TYPE_EMPTY	0x020	/* _SEGMENT_ENTRY_INV */
249 #define _HPAGE_TYPE_NONE	0x220
250 #define _HPAGE_TYPE_RO		0x200	/* _SEGMENT_ENTRY_RO  */
251 #define _HPAGE_TYPE_RW		0x000
252 
253 /*
254  * PTE type bits are rather complicated. handle_pte_fault uses pte_present,
255  * pte_none and pte_file to find out the pte type WITHOUT holding the page
256  * table lock. ptep_clear_flush on the other hand uses ptep_clear_flush to
257  * invalidate a given pte. ipte sets the hw invalid bit and clears all tlbs
258  * for the page. The page table entry is set to _PAGE_TYPE_EMPTY afterwards.
259  * This change is done while holding the lock, but the intermediate step
260  * of a previously valid pte with the hw invalid bit set can be observed by
261  * handle_pte_fault. That makes it necessary that all valid pte types with
262  * the hw invalid bit set must be distinguishable from the four pte types
263  * empty, none, swap and file.
264  *
265  *			irxt  ipte  irxt
266  * _PAGE_TYPE_EMPTY	1000   ->   1000
267  * _PAGE_TYPE_NONE	1001   ->   1001
268  * _PAGE_TYPE_SWAP	1011   ->   1011
269  * _PAGE_TYPE_FILE	11?1   ->   11?1
270  * _PAGE_TYPE_RO	0100   ->   1100
271  * _PAGE_TYPE_RW	0000   ->   1000
272  *
273  * pte_none is true for bits combinations 1000, 1010, 1100, 1110
274  * pte_present is true for bits combinations 0000, 0010, 0100, 0110, 1001
275  * pte_file is true for bits combinations 1101, 1111
276  * swap pte is 1011 and 0001, 0011, 0101, 0111 are invalid.
277  */
278 
279 #ifndef CONFIG_64BIT
280 
281 /* Bits in the segment table address-space-control-element */
282 #define _ASCE_SPACE_SWITCH	0x80000000UL	/* space switch event	    */
283 #define _ASCE_ORIGIN_MASK	0x7ffff000UL	/* segment table origin	    */
284 #define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
285 #define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
286 #define _ASCE_TABLE_LENGTH	0x7f	/* 128 x 64 entries = 8k	    */
287 
288 /* Bits in the segment table entry */
289 #define _SEGMENT_ENTRY_ORIGIN	0x7fffffc0UL	/* page table origin	    */
290 #define _SEGMENT_ENTRY_RO	0x200	/* page protection bit		    */
291 #define _SEGMENT_ENTRY_INV	0x20	/* invalid segment table entry	    */
292 #define _SEGMENT_ENTRY_COMMON	0x10	/* common segment bit		    */
293 #define _SEGMENT_ENTRY_PTL	0x0f	/* page table length		    */
294 
295 #define _SEGMENT_ENTRY		(_SEGMENT_ENTRY_PTL)
296 #define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INV)
297 
298 /* Page status table bits for virtualization */
299 #define RCP_ACC_BITS	0xf0000000UL
300 #define RCP_FP_BIT	0x08000000UL
301 #define RCP_PCL_BIT	0x00800000UL
302 #define RCP_HR_BIT	0x00400000UL
303 #define RCP_HC_BIT	0x00200000UL
304 #define RCP_GR_BIT	0x00040000UL
305 #define RCP_GC_BIT	0x00020000UL
306 #define RCP_IN_BIT	0x00002000UL	/* IPTE notify bit */
307 
308 /* User dirty / referenced bit for KVM's migration feature */
309 #define KVM_UR_BIT	0x00008000UL
310 #define KVM_UC_BIT	0x00004000UL
311 
312 #else /* CONFIG_64BIT */
313 
314 /* Bits in the segment/region table address-space-control-element */
315 #define _ASCE_ORIGIN		~0xfffUL/* segment table origin		    */
316 #define _ASCE_PRIVATE_SPACE	0x100	/* private space control	    */
317 #define _ASCE_ALT_EVENT		0x80	/* storage alteration event control */
318 #define _ASCE_SPACE_SWITCH	0x40	/* space switch event		    */
319 #define _ASCE_REAL_SPACE	0x20	/* real space control		    */
320 #define _ASCE_TYPE_MASK		0x0c	/* asce table type mask		    */
321 #define _ASCE_TYPE_REGION1	0x0c	/* region first table type	    */
322 #define _ASCE_TYPE_REGION2	0x08	/* region second table type	    */
323 #define _ASCE_TYPE_REGION3	0x04	/* region third table type	    */
324 #define _ASCE_TYPE_SEGMENT	0x00	/* segment table type		    */
325 #define _ASCE_TABLE_LENGTH	0x03	/* region table length		    */
326 
327 /* Bits in the region table entry */
328 #define _REGION_ENTRY_ORIGIN	~0xfffUL/* region/segment table origin	    */
329 #define _REGION_ENTRY_RO	0x200	/* region protection bit	    */
330 #define _REGION_ENTRY_INV	0x20	/* invalid region table entry	    */
331 #define _REGION_ENTRY_TYPE_MASK	0x0c	/* region/segment table type mask   */
332 #define _REGION_ENTRY_TYPE_R1	0x0c	/* region first table type	    */
333 #define _REGION_ENTRY_TYPE_R2	0x08	/* region second table type	    */
334 #define _REGION_ENTRY_TYPE_R3	0x04	/* region third table type	    */
335 #define _REGION_ENTRY_LENGTH	0x03	/* region third length		    */
336 
337 #define _REGION1_ENTRY		(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_LENGTH)
338 #define _REGION1_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R1 | _REGION_ENTRY_INV)
339 #define _REGION2_ENTRY		(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_LENGTH)
340 #define _REGION2_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R2 | _REGION_ENTRY_INV)
341 #define _REGION3_ENTRY		(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_LENGTH)
342 #define _REGION3_ENTRY_EMPTY	(_REGION_ENTRY_TYPE_R3 | _REGION_ENTRY_INV)
343 
344 #define _REGION3_ENTRY_LARGE	0x400	/* RTTE-format control, large page  */
345 #define _REGION3_ENTRY_RO	0x200	/* page protection bit		    */
346 #define _REGION3_ENTRY_CO	0x100	/* change-recording override	    */
347 
348 /* Bits in the segment table entry */
349 #define _SEGMENT_ENTRY_ORIGIN_LARGE ~0xfffffUL /* large page address	    */
350 #define _SEGMENT_ENTRY_ORIGIN	~0x7ffUL/* segment table origin		    */
351 #define _SEGMENT_ENTRY_RO	0x200	/* page protection bit		    */
352 #define _SEGMENT_ENTRY_INV	0x20	/* invalid segment table entry	    */
353 
354 #define _SEGMENT_ENTRY		(0)
355 #define _SEGMENT_ENTRY_EMPTY	(_SEGMENT_ENTRY_INV)
356 
357 #define _SEGMENT_ENTRY_LARGE	0x400	/* STE-format control, large page   */
358 #define _SEGMENT_ENTRY_CO	0x100	/* change-recording override   */
359 #define _SEGMENT_ENTRY_SPLIT_BIT 0	/* THP splitting bit number */
360 #define _SEGMENT_ENTRY_SPLIT	(1UL << _SEGMENT_ENTRY_SPLIT_BIT)
361 
362 /* Set of bits not changed in pmd_modify */
363 #define _SEGMENT_CHG_MASK	(_SEGMENT_ENTRY_ORIGIN | _SEGMENT_ENTRY_LARGE \
364 				 | _SEGMENT_ENTRY_SPLIT | _SEGMENT_ENTRY_CO)
365 
366 /* Page status table bits for virtualization */
367 #define RCP_ACC_BITS	0xf000000000000000UL
368 #define RCP_FP_BIT	0x0800000000000000UL
369 #define RCP_PCL_BIT	0x0080000000000000UL
370 #define RCP_HR_BIT	0x0040000000000000UL
371 #define RCP_HC_BIT	0x0020000000000000UL
372 #define RCP_GR_BIT	0x0004000000000000UL
373 #define RCP_GC_BIT	0x0002000000000000UL
374 #define RCP_IN_BIT	0x0000200000000000UL	/* IPTE notify bit */
375 
376 /* User dirty / referenced bit for KVM's migration feature */
377 #define KVM_UR_BIT	0x0000800000000000UL
378 #define KVM_UC_BIT	0x0000400000000000UL
379 
380 #endif /* CONFIG_64BIT */
381 
382 /*
383  * A user page table pointer has the space-switch-event bit, the
384  * private-space-control bit and the storage-alteration-event-control
385  * bit set. A kernel page table pointer doesn't need them.
386  */
387 #define _ASCE_USER_BITS		(_ASCE_SPACE_SWITCH | _ASCE_PRIVATE_SPACE | \
388 				 _ASCE_ALT_EVENT)
389 
390 /*
391  * Page protection definitions.
392  */
393 #define PAGE_NONE	__pgprot(_PAGE_TYPE_NONE)
394 #define PAGE_RO		__pgprot(_PAGE_TYPE_RO)
395 #define PAGE_RW		__pgprot(_PAGE_TYPE_RO | _PAGE_SWW)
396 #define PAGE_RWC	__pgprot(_PAGE_TYPE_RW | _PAGE_SWW | _PAGE_SWC)
397 
398 #define PAGE_KERNEL	PAGE_RWC
399 #define PAGE_SHARED	PAGE_KERNEL
400 #define PAGE_COPY	PAGE_RO
401 
402 /*
403  * On s390 the page table entry has an invalid bit and a read-only bit.
404  * Read permission implies execute permission and write permission
405  * implies read permission.
406  */
407          /*xwr*/
408 #define __P000	PAGE_NONE
409 #define __P001	PAGE_RO
410 #define __P010	PAGE_RO
411 #define __P011	PAGE_RO
412 #define __P100	PAGE_RO
413 #define __P101	PAGE_RO
414 #define __P110	PAGE_RO
415 #define __P111	PAGE_RO
416 
417 #define __S000	PAGE_NONE
418 #define __S001	PAGE_RO
419 #define __S010	PAGE_RW
420 #define __S011	PAGE_RW
421 #define __S100	PAGE_RO
422 #define __S101	PAGE_RO
423 #define __S110	PAGE_RW
424 #define __S111	PAGE_RW
425 
426 /*
427  * Segment entry (large page) protection definitions.
428  */
429 #define SEGMENT_NONE	__pgprot(_HPAGE_TYPE_NONE)
430 #define SEGMENT_RO	__pgprot(_HPAGE_TYPE_RO)
431 #define SEGMENT_RW	__pgprot(_HPAGE_TYPE_RW)
432 
433 static inline int mm_exclusive(struct mm_struct *mm)
434 {
435 	return likely(mm == current->active_mm &&
436 		      atomic_read(&mm->context.attach_count) <= 1);
437 }
438 
439 static inline int mm_has_pgste(struct mm_struct *mm)
440 {
441 #ifdef CONFIG_PGSTE
442 	if (unlikely(mm->context.has_pgste))
443 		return 1;
444 #endif
445 	return 0;
446 }
447 /*
448  * pgd/pmd/pte query functions
449  */
450 #ifndef CONFIG_64BIT
451 
452 static inline int pgd_present(pgd_t pgd) { return 1; }
453 static inline int pgd_none(pgd_t pgd)    { return 0; }
454 static inline int pgd_bad(pgd_t pgd)     { return 0; }
455 
456 static inline int pud_present(pud_t pud) { return 1; }
457 static inline int pud_none(pud_t pud)	 { return 0; }
458 static inline int pud_large(pud_t pud)	 { return 0; }
459 static inline int pud_bad(pud_t pud)	 { return 0; }
460 
461 #else /* CONFIG_64BIT */
462 
463 static inline int pgd_present(pgd_t pgd)
464 {
465 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
466 		return 1;
467 	return (pgd_val(pgd) & _REGION_ENTRY_ORIGIN) != 0UL;
468 }
469 
470 static inline int pgd_none(pgd_t pgd)
471 {
472 	if ((pgd_val(pgd) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R2)
473 		return 0;
474 	return (pgd_val(pgd) & _REGION_ENTRY_INV) != 0UL;
475 }
476 
477 static inline int pgd_bad(pgd_t pgd)
478 {
479 	/*
480 	 * With dynamic page table levels the pgd can be a region table
481 	 * entry or a segment table entry. Check for the bit that are
482 	 * invalid for either table entry.
483 	 */
484 	unsigned long mask =
485 		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INV &
486 		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
487 	return (pgd_val(pgd) & mask) != 0;
488 }
489 
490 static inline int pud_present(pud_t pud)
491 {
492 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
493 		return 1;
494 	return (pud_val(pud) & _REGION_ENTRY_ORIGIN) != 0UL;
495 }
496 
497 static inline int pud_none(pud_t pud)
498 {
499 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) < _REGION_ENTRY_TYPE_R3)
500 		return 0;
501 	return (pud_val(pud) & _REGION_ENTRY_INV) != 0UL;
502 }
503 
504 static inline int pud_large(pud_t pud)
505 {
506 	if ((pud_val(pud) & _REGION_ENTRY_TYPE_MASK) != _REGION_ENTRY_TYPE_R3)
507 		return 0;
508 	return !!(pud_val(pud) & _REGION3_ENTRY_LARGE);
509 }
510 
511 static inline int pud_bad(pud_t pud)
512 {
513 	/*
514 	 * With dynamic page table levels the pud can be a region table
515 	 * entry or a segment table entry. Check for the bit that are
516 	 * invalid for either table entry.
517 	 */
518 	unsigned long mask =
519 		~_SEGMENT_ENTRY_ORIGIN & ~_REGION_ENTRY_INV &
520 		~_REGION_ENTRY_TYPE_MASK & ~_REGION_ENTRY_LENGTH;
521 	return (pud_val(pud) & mask) != 0;
522 }
523 
524 #endif /* CONFIG_64BIT */
525 
526 static inline int pmd_present(pmd_t pmd)
527 {
528 	unsigned long mask = _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO;
529 	return (pmd_val(pmd) & mask) == _HPAGE_TYPE_NONE ||
530 	       !(pmd_val(pmd) & _SEGMENT_ENTRY_INV);
531 }
532 
533 static inline int pmd_none(pmd_t pmd)
534 {
535 	return (pmd_val(pmd) & _SEGMENT_ENTRY_INV) &&
536 	       !(pmd_val(pmd) & _SEGMENT_ENTRY_RO);
537 }
538 
539 static inline int pmd_large(pmd_t pmd)
540 {
541 #ifdef CONFIG_64BIT
542 	return !!(pmd_val(pmd) & _SEGMENT_ENTRY_LARGE);
543 #else
544 	return 0;
545 #endif
546 }
547 
548 static inline int pmd_bad(pmd_t pmd)
549 {
550 	unsigned long mask = ~_SEGMENT_ENTRY_ORIGIN & ~_SEGMENT_ENTRY_INV;
551 	return (pmd_val(pmd) & mask) != _SEGMENT_ENTRY;
552 }
553 
554 #define __HAVE_ARCH_PMDP_SPLITTING_FLUSH
555 extern void pmdp_splitting_flush(struct vm_area_struct *vma,
556 				 unsigned long addr, pmd_t *pmdp);
557 
558 #define  __HAVE_ARCH_PMDP_SET_ACCESS_FLAGS
559 extern int pmdp_set_access_flags(struct vm_area_struct *vma,
560 				 unsigned long address, pmd_t *pmdp,
561 				 pmd_t entry, int dirty);
562 
563 #define __HAVE_ARCH_PMDP_CLEAR_YOUNG_FLUSH
564 extern int pmdp_clear_flush_young(struct vm_area_struct *vma,
565 				  unsigned long address, pmd_t *pmdp);
566 
567 #define __HAVE_ARCH_PMD_WRITE
568 static inline int pmd_write(pmd_t pmd)
569 {
570 	return (pmd_val(pmd) & _SEGMENT_ENTRY_RO) == 0;
571 }
572 
573 static inline int pmd_young(pmd_t pmd)
574 {
575 	return 0;
576 }
577 
578 static inline int pte_none(pte_t pte)
579 {
580 	return (pte_val(pte) & _PAGE_INVALID) && !(pte_val(pte) & _PAGE_SWT);
581 }
582 
583 static inline int pte_present(pte_t pte)
584 {
585 	unsigned long mask = _PAGE_RO | _PAGE_INVALID | _PAGE_SWT | _PAGE_SWX;
586 	return (pte_val(pte) & mask) == _PAGE_TYPE_NONE ||
587 		(!(pte_val(pte) & _PAGE_INVALID) &&
588 		 !(pte_val(pte) & _PAGE_SWT));
589 }
590 
591 static inline int pte_file(pte_t pte)
592 {
593 	unsigned long mask = _PAGE_RO | _PAGE_INVALID | _PAGE_SWT;
594 	return (pte_val(pte) & mask) == _PAGE_TYPE_FILE;
595 }
596 
597 static inline int pte_special(pte_t pte)
598 {
599 	return (pte_val(pte) & _PAGE_SPECIAL);
600 }
601 
602 #define __HAVE_ARCH_PTE_SAME
603 static inline int pte_same(pte_t a, pte_t b)
604 {
605 	return pte_val(a) == pte_val(b);
606 }
607 
608 static inline pgste_t pgste_get_lock(pte_t *ptep)
609 {
610 	unsigned long new = 0;
611 #ifdef CONFIG_PGSTE
612 	unsigned long old;
613 
614 	preempt_disable();
615 	asm(
616 		"	lg	%0,%2\n"
617 		"0:	lgr	%1,%0\n"
618 		"	nihh	%0,0xff7f\n"	/* clear RCP_PCL_BIT in old */
619 		"	oihh	%1,0x0080\n"	/* set RCP_PCL_BIT in new */
620 		"	csg	%0,%1,%2\n"
621 		"	jl	0b\n"
622 		: "=&d" (old), "=&d" (new), "=Q" (ptep[PTRS_PER_PTE])
623 		: "Q" (ptep[PTRS_PER_PTE]) : "cc", "memory");
624 #endif
625 	return __pgste(new);
626 }
627 
628 static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste)
629 {
630 #ifdef CONFIG_PGSTE
631 	asm(
632 		"	nihh	%1,0xff7f\n"	/* clear RCP_PCL_BIT */
633 		"	stg	%1,%0\n"
634 		: "=Q" (ptep[PTRS_PER_PTE])
635 		: "d" (pgste_val(pgste)), "Q" (ptep[PTRS_PER_PTE])
636 		: "cc", "memory");
637 	preempt_enable();
638 #endif
639 }
640 
641 static inline void pgste_set(pte_t *ptep, pgste_t pgste)
642 {
643 #ifdef CONFIG_PGSTE
644 	*(pgste_t *)(ptep + PTRS_PER_PTE) = pgste;
645 #endif
646 }
647 
648 static inline pgste_t pgste_update_all(pte_t *ptep, pgste_t pgste)
649 {
650 #ifdef CONFIG_PGSTE
651 	unsigned long address, bits;
652 	unsigned char skey;
653 
654 	if (pte_val(*ptep) & _PAGE_INVALID)
655 		return pgste;
656 	address = pte_val(*ptep) & PAGE_MASK;
657 	skey = page_get_storage_key(address);
658 	bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
659 	/* Clear page changed & referenced bit in the storage key */
660 	if (bits & _PAGE_CHANGED)
661 		page_set_storage_key(address, skey ^ bits, 0);
662 	else if (bits)
663 		page_reset_referenced(address);
664 	/* Transfer page changed & referenced bit to guest bits in pgste */
665 	pgste_val(pgste) |= bits << 48;		/* RCP_GR_BIT & RCP_GC_BIT */
666 	/* Get host changed & referenced bits from pgste */
667 	bits |= (pgste_val(pgste) & (RCP_HR_BIT | RCP_HC_BIT)) >> 52;
668 	/* Transfer page changed & referenced bit to kvm user bits */
669 	pgste_val(pgste) |= bits << 45;		/* KVM_UR_BIT & KVM_UC_BIT */
670 	/* Clear relevant host bits in pgste. */
671 	pgste_val(pgste) &= ~(RCP_HR_BIT | RCP_HC_BIT);
672 	pgste_val(pgste) &= ~(RCP_ACC_BITS | RCP_FP_BIT);
673 	/* Copy page access key and fetch protection bit to pgste */
674 	pgste_val(pgste) |=
675 		(unsigned long) (skey & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
676 	/* Transfer referenced bit to pte */
677 	pte_val(*ptep) |= (bits & _PAGE_REFERENCED) << 1;
678 #endif
679 	return pgste;
680 
681 }
682 
683 static inline pgste_t pgste_update_young(pte_t *ptep, pgste_t pgste)
684 {
685 #ifdef CONFIG_PGSTE
686 	int young;
687 
688 	if (pte_val(*ptep) & _PAGE_INVALID)
689 		return pgste;
690 	/* Get referenced bit from storage key */
691 	young = page_reset_referenced(pte_val(*ptep) & PAGE_MASK);
692 	if (young)
693 		pgste_val(pgste) |= RCP_GR_BIT;
694 	/* Get host referenced bit from pgste */
695 	if (pgste_val(pgste) & RCP_HR_BIT) {
696 		pgste_val(pgste) &= ~RCP_HR_BIT;
697 		young = 1;
698 	}
699 	/* Transfer referenced bit to kvm user bits and pte */
700 	if (young) {
701 		pgste_val(pgste) |= KVM_UR_BIT;
702 		pte_val(*ptep) |= _PAGE_SWR;
703 	}
704 #endif
705 	return pgste;
706 }
707 
708 static inline void pgste_set_key(pte_t *ptep, pgste_t pgste, pte_t entry)
709 {
710 #ifdef CONFIG_PGSTE
711 	unsigned long address;
712 	unsigned long nkey;
713 
714 	if (pte_val(entry) & _PAGE_INVALID)
715 		return;
716 	VM_BUG_ON(!(pte_val(*ptep) & _PAGE_INVALID));
717 	address = pte_val(entry) & PAGE_MASK;
718 	/*
719 	 * Set page access key and fetch protection bit from pgste.
720 	 * The guest C/R information is still in the PGSTE, set real
721 	 * key C/R to 0.
722 	 */
723 	nkey = (pgste_val(pgste) & (RCP_ACC_BITS | RCP_FP_BIT)) >> 56;
724 	page_set_storage_key(address, nkey, 0);
725 #endif
726 }
727 
728 static inline void pgste_set_pte(pte_t *ptep, pte_t entry)
729 {
730 	if (!MACHINE_HAS_ESOP && (pte_val(entry) & _PAGE_SWW)) {
731 		/*
732 		 * Without enhanced suppression-on-protection force
733 		 * the dirty bit on for all writable ptes.
734 		 */
735 		pte_val(entry) |= _PAGE_SWC;
736 		pte_val(entry) &= ~_PAGE_RO;
737 	}
738 	*ptep = entry;
739 }
740 
741 /**
742  * struct gmap_struct - guest address space
743  * @mm: pointer to the parent mm_struct
744  * @table: pointer to the page directory
745  * @asce: address space control element for gmap page table
746  * @crst_list: list of all crst tables used in the guest address space
747  */
748 struct gmap {
749 	struct list_head list;
750 	struct mm_struct *mm;
751 	unsigned long *table;
752 	unsigned long asce;
753 	struct list_head crst_list;
754 };
755 
756 /**
757  * struct gmap_rmap - reverse mapping for segment table entries
758  * @gmap: pointer to the gmap_struct
759  * @entry: pointer to a segment table entry
760  * @vmaddr: virtual address in the guest address space
761  */
762 struct gmap_rmap {
763 	struct list_head list;
764 	struct gmap *gmap;
765 	unsigned long *entry;
766 	unsigned long vmaddr;
767 };
768 
769 /**
770  * struct gmap_pgtable - gmap information attached to a page table
771  * @vmaddr: address of the 1MB segment in the process virtual memory
772  * @mapper: list of segment table entries mapping a page table
773  */
774 struct gmap_pgtable {
775 	unsigned long vmaddr;
776 	struct list_head mapper;
777 };
778 
779 /**
780  * struct gmap_notifier - notify function block for page invalidation
781  * @notifier_call: address of callback function
782  */
783 struct gmap_notifier {
784 	struct list_head list;
785 	void (*notifier_call)(struct gmap *gmap, unsigned long address);
786 };
787 
788 struct gmap *gmap_alloc(struct mm_struct *mm);
789 void gmap_free(struct gmap *gmap);
790 void gmap_enable(struct gmap *gmap);
791 void gmap_disable(struct gmap *gmap);
792 int gmap_map_segment(struct gmap *gmap, unsigned long from,
793 		     unsigned long to, unsigned long len);
794 int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len);
795 unsigned long __gmap_translate(unsigned long address, struct gmap *);
796 unsigned long gmap_translate(unsigned long address, struct gmap *);
797 unsigned long __gmap_fault(unsigned long address, struct gmap *);
798 unsigned long gmap_fault(unsigned long address, struct gmap *);
799 void gmap_discard(unsigned long from, unsigned long to, struct gmap *);
800 
801 void gmap_register_ipte_notifier(struct gmap_notifier *);
802 void gmap_unregister_ipte_notifier(struct gmap_notifier *);
803 int gmap_ipte_notify(struct gmap *, unsigned long start, unsigned long len);
804 void gmap_do_ipte_notify(struct mm_struct *, unsigned long addr, pte_t *);
805 
806 static inline pgste_t pgste_ipte_notify(struct mm_struct *mm,
807 					unsigned long addr,
808 					pte_t *ptep, pgste_t pgste)
809 {
810 #ifdef CONFIG_PGSTE
811 	if (pgste_val(pgste) & RCP_IN_BIT) {
812 		pgste_val(pgste) &= ~RCP_IN_BIT;
813 		gmap_do_ipte_notify(mm, addr, ptep);
814 	}
815 #endif
816 	return pgste;
817 }
818 
819 /*
820  * Certain architectures need to do special things when PTEs
821  * within a page table are directly modified.  Thus, the following
822  * hook is made available.
823  */
824 static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
825 			      pte_t *ptep, pte_t entry)
826 {
827 	pgste_t pgste;
828 
829 	if (mm_has_pgste(mm)) {
830 		pgste = pgste_get_lock(ptep);
831 		pgste_set_key(ptep, pgste, entry);
832 		pgste_set_pte(ptep, entry);
833 		pgste_set_unlock(ptep, pgste);
834 	} else {
835 		if (!(pte_val(entry) & _PAGE_INVALID) && MACHINE_HAS_EDAT1)
836 			pte_val(entry) |= _PAGE_CO;
837 		*ptep = entry;
838 	}
839 }
840 
841 /*
842  * query functions pte_write/pte_dirty/pte_young only work if
843  * pte_present() is true. Undefined behaviour if not..
844  */
845 static inline int pte_write(pte_t pte)
846 {
847 	return (pte_val(pte) & _PAGE_SWW) != 0;
848 }
849 
850 static inline int pte_dirty(pte_t pte)
851 {
852 	return (pte_val(pte) & _PAGE_SWC) != 0;
853 }
854 
855 static inline int pte_young(pte_t pte)
856 {
857 #ifdef CONFIG_PGSTE
858 	if (pte_val(pte) & _PAGE_SWR)
859 		return 1;
860 #endif
861 	return 0;
862 }
863 
864 /*
865  * pgd/pmd/pte modification functions
866  */
867 
868 static inline void pgd_clear(pgd_t *pgd)
869 {
870 #ifdef CONFIG_64BIT
871 	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
872 		pgd_val(*pgd) = _REGION2_ENTRY_EMPTY;
873 #endif
874 }
875 
876 static inline void pud_clear(pud_t *pud)
877 {
878 #ifdef CONFIG_64BIT
879 	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
880 		pud_val(*pud) = _REGION3_ENTRY_EMPTY;
881 #endif
882 }
883 
884 static inline void pmd_clear(pmd_t *pmdp)
885 {
886 	pmd_val(*pmdp) = _SEGMENT_ENTRY_EMPTY;
887 }
888 
889 static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
890 {
891 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
892 }
893 
894 /*
895  * The following pte modification functions only work if
896  * pte_present() is true. Undefined behaviour if not..
897  */
898 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
899 {
900 	pte_val(pte) &= _PAGE_CHG_MASK;
901 	pte_val(pte) |= pgprot_val(newprot);
902 	if ((pte_val(pte) & _PAGE_SWC) && (pte_val(pte) & _PAGE_SWW))
903 		pte_val(pte) &= ~_PAGE_RO;
904 	return pte;
905 }
906 
907 static inline pte_t pte_wrprotect(pte_t pte)
908 {
909 	pte_val(pte) &= ~_PAGE_SWW;
910 	/* Do not clobber _PAGE_TYPE_NONE pages!  */
911 	if (!(pte_val(pte) & _PAGE_INVALID))
912 		pte_val(pte) |= _PAGE_RO;
913 	return pte;
914 }
915 
916 static inline pte_t pte_mkwrite(pte_t pte)
917 {
918 	pte_val(pte) |= _PAGE_SWW;
919 	if (pte_val(pte) & _PAGE_SWC)
920 		pte_val(pte) &= ~_PAGE_RO;
921 	return pte;
922 }
923 
924 static inline pte_t pte_mkclean(pte_t pte)
925 {
926 	pte_val(pte) &= ~_PAGE_SWC;
927 	/* Do not clobber _PAGE_TYPE_NONE pages!  */
928 	if (!(pte_val(pte) & _PAGE_INVALID))
929 		pte_val(pte) |= _PAGE_RO;
930 	return pte;
931 }
932 
933 static inline pte_t pte_mkdirty(pte_t pte)
934 {
935 	pte_val(pte) |= _PAGE_SWC;
936 	if (pte_val(pte) & _PAGE_SWW)
937 		pte_val(pte) &= ~_PAGE_RO;
938 	return pte;
939 }
940 
941 static inline pte_t pte_mkold(pte_t pte)
942 {
943 #ifdef CONFIG_PGSTE
944 	pte_val(pte) &= ~_PAGE_SWR;
945 #endif
946 	return pte;
947 }
948 
949 static inline pte_t pte_mkyoung(pte_t pte)
950 {
951 	return pte;
952 }
953 
954 static inline pte_t pte_mkspecial(pte_t pte)
955 {
956 	pte_val(pte) |= _PAGE_SPECIAL;
957 	return pte;
958 }
959 
960 #ifdef CONFIG_HUGETLB_PAGE
961 static inline pte_t pte_mkhuge(pte_t pte)
962 {
963 	pte_val(pte) |= (_SEGMENT_ENTRY_LARGE | _SEGMENT_ENTRY_CO);
964 	return pte;
965 }
966 #endif
967 
968 /*
969  * Get (and clear) the user dirty bit for a pte.
970  */
971 static inline int ptep_test_and_clear_user_dirty(struct mm_struct *mm,
972 						 pte_t *ptep)
973 {
974 	pgste_t pgste;
975 	int dirty = 0;
976 
977 	if (mm_has_pgste(mm)) {
978 		pgste = pgste_get_lock(ptep);
979 		pgste = pgste_update_all(ptep, pgste);
980 		dirty = !!(pgste_val(pgste) & KVM_UC_BIT);
981 		pgste_val(pgste) &= ~KVM_UC_BIT;
982 		pgste_set_unlock(ptep, pgste);
983 		return dirty;
984 	}
985 	return dirty;
986 }
987 
988 /*
989  * Get (and clear) the user referenced bit for a pte.
990  */
991 static inline int ptep_test_and_clear_user_young(struct mm_struct *mm,
992 						 pte_t *ptep)
993 {
994 	pgste_t pgste;
995 	int young = 0;
996 
997 	if (mm_has_pgste(mm)) {
998 		pgste = pgste_get_lock(ptep);
999 		pgste = pgste_update_young(ptep, pgste);
1000 		young = !!(pgste_val(pgste) & KVM_UR_BIT);
1001 		pgste_val(pgste) &= ~KVM_UR_BIT;
1002 		pgste_set_unlock(ptep, pgste);
1003 	}
1004 	return young;
1005 }
1006 
1007 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
1008 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma,
1009 					    unsigned long addr, pte_t *ptep)
1010 {
1011 	pgste_t pgste;
1012 	pte_t pte;
1013 
1014 	if (mm_has_pgste(vma->vm_mm)) {
1015 		pgste = pgste_get_lock(ptep);
1016 		pgste = pgste_update_young(ptep, pgste);
1017 		pte = *ptep;
1018 		*ptep = pte_mkold(pte);
1019 		pgste_set_unlock(ptep, pgste);
1020 		return pte_young(pte);
1021 	}
1022 	return 0;
1023 }
1024 
1025 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
1026 static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
1027 					 unsigned long address, pte_t *ptep)
1028 {
1029 	/* No need to flush TLB
1030 	 * On s390 reference bits are in storage key and never in TLB
1031 	 * With virtualization we handle the reference bit, without we
1032 	 * we can simply return */
1033 	return ptep_test_and_clear_young(vma, address, ptep);
1034 }
1035 
1036 static inline void __ptep_ipte(unsigned long address, pte_t *ptep)
1037 {
1038 	if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1039 #ifndef CONFIG_64BIT
1040 		/* pto must point to the start of the segment table */
1041 		pte_t *pto = (pte_t *) (((unsigned long) ptep) & 0x7ffffc00);
1042 #else
1043 		/* ipte in zarch mode can do the math */
1044 		pte_t *pto = ptep;
1045 #endif
1046 		asm volatile(
1047 			"	ipte	%2,%3"
1048 			: "=m" (*ptep) : "m" (*ptep),
1049 			  "a" (pto), "a" (address));
1050 	}
1051 }
1052 
1053 /*
1054  * This is hard to understand. ptep_get_and_clear and ptep_clear_flush
1055  * both clear the TLB for the unmapped pte. The reason is that
1056  * ptep_get_and_clear is used in common code (e.g. change_pte_range)
1057  * to modify an active pte. The sequence is
1058  *   1) ptep_get_and_clear
1059  *   2) set_pte_at
1060  *   3) flush_tlb_range
1061  * On s390 the tlb needs to get flushed with the modification of the pte
1062  * if the pte is active. The only way how this can be implemented is to
1063  * have ptep_get_and_clear do the tlb flush. In exchange flush_tlb_range
1064  * is a nop.
1065  */
1066 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
1067 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
1068 				       unsigned long address, pte_t *ptep)
1069 {
1070 	pgste_t pgste;
1071 	pte_t pte;
1072 
1073 	mm->context.flush_mm = 1;
1074 	if (mm_has_pgste(mm)) {
1075 		pgste = pgste_get_lock(ptep);
1076 		pgste = pgste_ipte_notify(mm, address, ptep, pgste);
1077 	}
1078 
1079 	pte = *ptep;
1080 	if (!mm_exclusive(mm))
1081 		__ptep_ipte(address, ptep);
1082 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
1083 
1084 	if (mm_has_pgste(mm)) {
1085 		pgste = pgste_update_all(&pte, pgste);
1086 		pgste_set_unlock(ptep, pgste);
1087 	}
1088 	return pte;
1089 }
1090 
1091 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
1092 static inline pte_t ptep_modify_prot_start(struct mm_struct *mm,
1093 					   unsigned long address,
1094 					   pte_t *ptep)
1095 {
1096 	pgste_t pgste;
1097 	pte_t pte;
1098 
1099 	mm->context.flush_mm = 1;
1100 	if (mm_has_pgste(mm)) {
1101 		pgste = pgste_get_lock(ptep);
1102 		pgste_ipte_notify(mm, address, ptep, pgste);
1103 	}
1104 
1105 	pte = *ptep;
1106 	if (!mm_exclusive(mm))
1107 		__ptep_ipte(address, ptep);
1108 
1109 	if (mm_has_pgste(mm)) {
1110 		pgste = pgste_update_all(&pte, pgste);
1111 		pgste_set(ptep, pgste);
1112 	}
1113 	return pte;
1114 }
1115 
1116 static inline void ptep_modify_prot_commit(struct mm_struct *mm,
1117 					   unsigned long address,
1118 					   pte_t *ptep, pte_t pte)
1119 {
1120 	pgste_t pgste;
1121 
1122 	if (mm_has_pgste(mm)) {
1123 		pgste = *(pgste_t *)(ptep + PTRS_PER_PTE);
1124 		pgste_set_key(ptep, pgste, pte);
1125 		pgste_set_pte(ptep, pte);
1126 		pgste_set_unlock(ptep, pgste);
1127 	} else
1128 		*ptep = pte;
1129 }
1130 
1131 #define __HAVE_ARCH_PTEP_CLEAR_FLUSH
1132 static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
1133 				     unsigned long address, pte_t *ptep)
1134 {
1135 	pgste_t pgste;
1136 	pte_t pte;
1137 
1138 	if (mm_has_pgste(vma->vm_mm)) {
1139 		pgste = pgste_get_lock(ptep);
1140 		pgste = pgste_ipte_notify(vma->vm_mm, address, ptep, pgste);
1141 	}
1142 
1143 	pte = *ptep;
1144 	__ptep_ipte(address, ptep);
1145 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
1146 
1147 	if (mm_has_pgste(vma->vm_mm)) {
1148 		pgste = pgste_update_all(&pte, pgste);
1149 		pgste_set_unlock(ptep, pgste);
1150 	}
1151 	return pte;
1152 }
1153 
1154 /*
1155  * The batched pte unmap code uses ptep_get_and_clear_full to clear the
1156  * ptes. Here an optimization is possible. tlb_gather_mmu flushes all
1157  * tlbs of an mm if it can guarantee that the ptes of the mm_struct
1158  * cannot be accessed while the batched unmap is running. In this case
1159  * full==1 and a simple pte_clear is enough. See tlb.h.
1160  */
1161 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
1162 static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
1163 					    unsigned long address,
1164 					    pte_t *ptep, int full)
1165 {
1166 	pgste_t pgste;
1167 	pte_t pte;
1168 
1169 	if (mm_has_pgste(mm)) {
1170 		pgste = pgste_get_lock(ptep);
1171 		if (!full)
1172 			pgste = pgste_ipte_notify(mm, address, ptep, pgste);
1173 	}
1174 
1175 	pte = *ptep;
1176 	if (!full)
1177 		__ptep_ipte(address, ptep);
1178 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
1179 
1180 	if (mm_has_pgste(mm)) {
1181 		pgste = pgste_update_all(&pte, pgste);
1182 		pgste_set_unlock(ptep, pgste);
1183 	}
1184 	return pte;
1185 }
1186 
1187 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
1188 static inline pte_t ptep_set_wrprotect(struct mm_struct *mm,
1189 				       unsigned long address, pte_t *ptep)
1190 {
1191 	pgste_t pgste;
1192 	pte_t pte = *ptep;
1193 
1194 	if (pte_write(pte)) {
1195 		mm->context.flush_mm = 1;
1196 		if (mm_has_pgste(mm)) {
1197 			pgste = pgste_get_lock(ptep);
1198 			pgste = pgste_ipte_notify(mm, address, ptep, pgste);
1199 		}
1200 
1201 		if (!mm_exclusive(mm))
1202 			__ptep_ipte(address, ptep);
1203 		pte = pte_wrprotect(pte);
1204 
1205 		if (mm_has_pgste(mm)) {
1206 			pgste_set_pte(ptep, pte);
1207 			pgste_set_unlock(ptep, pgste);
1208 		} else
1209 			*ptep = pte;
1210 	}
1211 	return pte;
1212 }
1213 
1214 #define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
1215 static inline int ptep_set_access_flags(struct vm_area_struct *vma,
1216 					unsigned long address, pte_t *ptep,
1217 					pte_t entry, int dirty)
1218 {
1219 	pgste_t pgste;
1220 
1221 	if (pte_same(*ptep, entry))
1222 		return 0;
1223 	if (mm_has_pgste(vma->vm_mm)) {
1224 		pgste = pgste_get_lock(ptep);
1225 		pgste = pgste_ipte_notify(vma->vm_mm, address, ptep, pgste);
1226 	}
1227 
1228 	__ptep_ipte(address, ptep);
1229 
1230 	if (mm_has_pgste(vma->vm_mm)) {
1231 		pgste_set_pte(ptep, entry);
1232 		pgste_set_unlock(ptep, pgste);
1233 	} else
1234 		*ptep = entry;
1235 	return 1;
1236 }
1237 
1238 /*
1239  * Conversion functions: convert a page and protection to a page entry,
1240  * and a page entry and page directory to the page they refer to.
1241  */
1242 static inline pte_t mk_pte_phys(unsigned long physpage, pgprot_t pgprot)
1243 {
1244 	pte_t __pte;
1245 	pte_val(__pte) = physpage + pgprot_val(pgprot);
1246 	return __pte;
1247 }
1248 
1249 static inline pte_t mk_pte(struct page *page, pgprot_t pgprot)
1250 {
1251 	unsigned long physpage = page_to_phys(page);
1252 	pte_t __pte = mk_pte_phys(physpage, pgprot);
1253 
1254 	if ((pte_val(__pte) & _PAGE_SWW) && PageDirty(page)) {
1255 		pte_val(__pte) |= _PAGE_SWC;
1256 		pte_val(__pte) &= ~_PAGE_RO;
1257 	}
1258 	return __pte;
1259 }
1260 
1261 #define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD-1))
1262 #define pud_index(address) (((address) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
1263 #define pmd_index(address) (((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1))
1264 #define pte_index(address) (((address) >> PAGE_SHIFT) & (PTRS_PER_PTE-1))
1265 
1266 #define pgd_offset(mm, address) ((mm)->pgd + pgd_index(address))
1267 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
1268 
1269 #ifndef CONFIG_64BIT
1270 
1271 #define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
1272 #define pud_deref(pmd) ({ BUG(); 0UL; })
1273 #define pgd_deref(pmd) ({ BUG(); 0UL; })
1274 
1275 #define pud_offset(pgd, address) ((pud_t *) pgd)
1276 #define pmd_offset(pud, address) ((pmd_t *) pud + pmd_index(address))
1277 
1278 #else /* CONFIG_64BIT */
1279 
1280 #define pmd_deref(pmd) (pmd_val(pmd) & _SEGMENT_ENTRY_ORIGIN)
1281 #define pud_deref(pud) (pud_val(pud) & _REGION_ENTRY_ORIGIN)
1282 #define pgd_deref(pgd) (pgd_val(pgd) & _REGION_ENTRY_ORIGIN)
1283 
1284 static inline pud_t *pud_offset(pgd_t *pgd, unsigned long address)
1285 {
1286 	pud_t *pud = (pud_t *) pgd;
1287 	if ((pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R2)
1288 		pud = (pud_t *) pgd_deref(*pgd);
1289 	return pud  + pud_index(address);
1290 }
1291 
1292 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long address)
1293 {
1294 	pmd_t *pmd = (pmd_t *) pud;
1295 	if ((pud_val(*pud) & _REGION_ENTRY_TYPE_MASK) == _REGION_ENTRY_TYPE_R3)
1296 		pmd = (pmd_t *) pud_deref(*pud);
1297 	return pmd + pmd_index(address);
1298 }
1299 
1300 #endif /* CONFIG_64BIT */
1301 
1302 #define pfn_pte(pfn,pgprot) mk_pte_phys(__pa((pfn) << PAGE_SHIFT),(pgprot))
1303 #define pte_pfn(x) (pte_val(x) >> PAGE_SHIFT)
1304 #define pte_page(x) pfn_to_page(pte_pfn(x))
1305 
1306 #define pmd_page(pmd) pfn_to_page(pmd_val(pmd) >> PAGE_SHIFT)
1307 
1308 /* Find an entry in the lowest level page table.. */
1309 #define pte_offset(pmd, addr) ((pte_t *) pmd_deref(*(pmd)) + pte_index(addr))
1310 #define pte_offset_kernel(pmd, address) pte_offset(pmd,address)
1311 #define pte_offset_map(pmd, address) pte_offset_kernel(pmd, address)
1312 #define pte_unmap(pte) do { } while (0)
1313 
1314 static inline void __pmd_idte(unsigned long address, pmd_t *pmdp)
1315 {
1316 	unsigned long sto = (unsigned long) pmdp -
1317 			    pmd_index(address) * sizeof(pmd_t);
1318 
1319 	if (!(pmd_val(*pmdp) & _SEGMENT_ENTRY_INV)) {
1320 		asm volatile(
1321 			"	.insn	rrf,0xb98e0000,%2,%3,0,0"
1322 			: "=m" (*pmdp)
1323 			: "m" (*pmdp), "a" (sto),
1324 			  "a" ((address & HPAGE_MASK))
1325 			: "cc"
1326 		);
1327 	}
1328 }
1329 
1330 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLB_PAGE)
1331 static inline unsigned long massage_pgprot_pmd(pgprot_t pgprot)
1332 {
1333 	/*
1334 	 * pgprot is PAGE_NONE, PAGE_RO, or PAGE_RW (see __Pxxx / __Sxxx)
1335 	 * Convert to segment table entry format.
1336 	 */
1337 	if (pgprot_val(pgprot) == pgprot_val(PAGE_NONE))
1338 		return pgprot_val(SEGMENT_NONE);
1339 	if (pgprot_val(pgprot) == pgprot_val(PAGE_RO))
1340 		return pgprot_val(SEGMENT_RO);
1341 	return pgprot_val(SEGMENT_RW);
1342 }
1343 
1344 static inline pmd_t pmd_modify(pmd_t pmd, pgprot_t newprot)
1345 {
1346 	pmd_val(pmd) &= _SEGMENT_CHG_MASK;
1347 	pmd_val(pmd) |= massage_pgprot_pmd(newprot);
1348 	return pmd;
1349 }
1350 
1351 static inline pmd_t mk_pmd_phys(unsigned long physpage, pgprot_t pgprot)
1352 {
1353 	pmd_t __pmd;
1354 	pmd_val(__pmd) = physpage + massage_pgprot_pmd(pgprot);
1355 	return __pmd;
1356 }
1357 
1358 static inline pmd_t pmd_mkwrite(pmd_t pmd)
1359 {
1360 	/* Do not clobber _HPAGE_TYPE_NONE pages! */
1361 	if (!(pmd_val(pmd) & _SEGMENT_ENTRY_INV))
1362 		pmd_val(pmd) &= ~_SEGMENT_ENTRY_RO;
1363 	return pmd;
1364 }
1365 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLB_PAGE */
1366 
1367 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1368 
1369 #define __HAVE_ARCH_PGTABLE_DEPOSIT
1370 extern void pgtable_trans_huge_deposit(struct mm_struct *mm, pgtable_t pgtable);
1371 
1372 #define __HAVE_ARCH_PGTABLE_WITHDRAW
1373 extern pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm);
1374 
1375 static inline int pmd_trans_splitting(pmd_t pmd)
1376 {
1377 	return pmd_val(pmd) & _SEGMENT_ENTRY_SPLIT;
1378 }
1379 
1380 static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
1381 			      pmd_t *pmdp, pmd_t entry)
1382 {
1383 	if (!(pmd_val(entry) & _SEGMENT_ENTRY_INV) && MACHINE_HAS_EDAT1)
1384 		pmd_val(entry) |= _SEGMENT_ENTRY_CO;
1385 	*pmdp = entry;
1386 }
1387 
1388 static inline pmd_t pmd_mkhuge(pmd_t pmd)
1389 {
1390 	pmd_val(pmd) |= _SEGMENT_ENTRY_LARGE;
1391 	return pmd;
1392 }
1393 
1394 static inline pmd_t pmd_wrprotect(pmd_t pmd)
1395 {
1396 	pmd_val(pmd) |= _SEGMENT_ENTRY_RO;
1397 	return pmd;
1398 }
1399 
1400 static inline pmd_t pmd_mkdirty(pmd_t pmd)
1401 {
1402 	/* No dirty bit in the segment table entry. */
1403 	return pmd;
1404 }
1405 
1406 static inline pmd_t pmd_mkold(pmd_t pmd)
1407 {
1408 	/* No referenced bit in the segment table entry. */
1409 	return pmd;
1410 }
1411 
1412 static inline pmd_t pmd_mkyoung(pmd_t pmd)
1413 {
1414 	/* No referenced bit in the segment table entry. */
1415 	return pmd;
1416 }
1417 
1418 #define __HAVE_ARCH_PMDP_TEST_AND_CLEAR_YOUNG
1419 static inline int pmdp_test_and_clear_young(struct vm_area_struct *vma,
1420 					    unsigned long address, pmd_t *pmdp)
1421 {
1422 	unsigned long pmd_addr = pmd_val(*pmdp) & HPAGE_MASK;
1423 	long tmp, rc;
1424 	int counter;
1425 
1426 	rc = 0;
1427 	if (MACHINE_HAS_RRBM) {
1428 		counter = PTRS_PER_PTE >> 6;
1429 		asm volatile(
1430 			"0:	.insn	rre,0xb9ae0000,%0,%3\n"	/* rrbm */
1431 			"	ogr	%1,%0\n"
1432 			"	la	%3,0(%4,%3)\n"
1433 			"	brct	%2,0b\n"
1434 			: "=&d" (tmp), "+&d" (rc), "+d" (counter),
1435 			  "+a" (pmd_addr)
1436 			: "a" (64 * 4096UL) : "cc");
1437 		rc = !!rc;
1438 	} else {
1439 		counter = PTRS_PER_PTE;
1440 		asm volatile(
1441 			"0:	rrbe	0,%2\n"
1442 			"	la	%2,0(%3,%2)\n"
1443 			"	brc	12,1f\n"
1444 			"	lhi	%0,1\n"
1445 			"1:	brct	%1,0b\n"
1446 			: "+d" (rc), "+d" (counter), "+a" (pmd_addr)
1447 			: "a" (4096UL) : "cc");
1448 	}
1449 	return rc;
1450 }
1451 
1452 #define __HAVE_ARCH_PMDP_GET_AND_CLEAR
1453 static inline pmd_t pmdp_get_and_clear(struct mm_struct *mm,
1454 				       unsigned long address, pmd_t *pmdp)
1455 {
1456 	pmd_t pmd = *pmdp;
1457 
1458 	__pmd_idte(address, pmdp);
1459 	pmd_clear(pmdp);
1460 	return pmd;
1461 }
1462 
1463 #define __HAVE_ARCH_PMDP_CLEAR_FLUSH
1464 static inline pmd_t pmdp_clear_flush(struct vm_area_struct *vma,
1465 				     unsigned long address, pmd_t *pmdp)
1466 {
1467 	return pmdp_get_and_clear(vma->vm_mm, address, pmdp);
1468 }
1469 
1470 #define __HAVE_ARCH_PMDP_INVALIDATE
1471 static inline void pmdp_invalidate(struct vm_area_struct *vma,
1472 				   unsigned long address, pmd_t *pmdp)
1473 {
1474 	__pmd_idte(address, pmdp);
1475 }
1476 
1477 #define __HAVE_ARCH_PMDP_SET_WRPROTECT
1478 static inline void pmdp_set_wrprotect(struct mm_struct *mm,
1479 				      unsigned long address, pmd_t *pmdp)
1480 {
1481 	pmd_t pmd = *pmdp;
1482 
1483 	if (pmd_write(pmd)) {
1484 		__pmd_idte(address, pmdp);
1485 		set_pmd_at(mm, address, pmdp, pmd_wrprotect(pmd));
1486 	}
1487 }
1488 
1489 #define pfn_pmd(pfn, pgprot)	mk_pmd_phys(__pa((pfn) << PAGE_SHIFT), (pgprot))
1490 #define mk_pmd(page, pgprot)	pfn_pmd(page_to_pfn(page), (pgprot))
1491 
1492 static inline int pmd_trans_huge(pmd_t pmd)
1493 {
1494 	return pmd_val(pmd) & _SEGMENT_ENTRY_LARGE;
1495 }
1496 
1497 static inline int has_transparent_hugepage(void)
1498 {
1499 	return MACHINE_HAS_HPAGE ? 1 : 0;
1500 }
1501 
1502 static inline unsigned long pmd_pfn(pmd_t pmd)
1503 {
1504 	return pmd_val(pmd) >> PAGE_SHIFT;
1505 }
1506 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1507 
1508 /*
1509  * 31 bit swap entry format:
1510  * A page-table entry has some bits we have to treat in a special way.
1511  * Bits 0, 20 and bit 23 have to be zero, otherwise an specification
1512  * exception will occur instead of a page translation exception. The
1513  * specifiation exception has the bad habit not to store necessary
1514  * information in the lowcore.
1515  * Bit 21 and bit 22 are the page invalid bit and the page protection
1516  * bit. We set both to indicate a swapped page.
1517  * Bit 30 and 31 are used to distinguish the different page types. For
1518  * a swapped page these bits need to be zero.
1519  * This leaves the bits 1-19 and bits 24-29 to store type and offset.
1520  * We use the 5 bits from 25-29 for the type and the 20 bits from 1-19
1521  * plus 24 for the offset.
1522  * 0|     offset        |0110|o|type |00|
1523  * 0 0000000001111111111 2222 2 22222 33
1524  * 0 1234567890123456789 0123 4 56789 01
1525  *
1526  * 64 bit swap entry format:
1527  * A page-table entry has some bits we have to treat in a special way.
1528  * Bits 52 and bit 55 have to be zero, otherwise an specification
1529  * exception will occur instead of a page translation exception. The
1530  * specifiation exception has the bad habit not to store necessary
1531  * information in the lowcore.
1532  * Bit 53 and bit 54 are the page invalid bit and the page protection
1533  * bit. We set both to indicate a swapped page.
1534  * Bit 62 and 63 are used to distinguish the different page types. For
1535  * a swapped page these bits need to be zero.
1536  * This leaves the bits 0-51 and bits 56-61 to store type and offset.
1537  * We use the 5 bits from 57-61 for the type and the 53 bits from 0-51
1538  * plus 56 for the offset.
1539  * |                      offset                        |0110|o|type |00|
1540  *  0000000000111111111122222222223333333333444444444455 5555 5 55566 66
1541  *  0123456789012345678901234567890123456789012345678901 2345 6 78901 23
1542  */
1543 #ifndef CONFIG_64BIT
1544 #define __SWP_OFFSET_MASK (~0UL >> 12)
1545 #else
1546 #define __SWP_OFFSET_MASK (~0UL >> 11)
1547 #endif
1548 static inline pte_t mk_swap_pte(unsigned long type, unsigned long offset)
1549 {
1550 	pte_t pte;
1551 	offset &= __SWP_OFFSET_MASK;
1552 	pte_val(pte) = _PAGE_TYPE_SWAP | ((type & 0x1f) << 2) |
1553 		((offset & 1UL) << 7) | ((offset & ~1UL) << 11);
1554 	return pte;
1555 }
1556 
1557 #define __swp_type(entry)	(((entry).val >> 2) & 0x1f)
1558 #define __swp_offset(entry)	(((entry).val >> 11) | (((entry).val >> 7) & 1))
1559 #define __swp_entry(type,offset) ((swp_entry_t) { pte_val(mk_swap_pte((type),(offset))) })
1560 
1561 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
1562 #define __swp_entry_to_pte(x)	((pte_t) { (x).val })
1563 
1564 #ifndef CONFIG_64BIT
1565 # define PTE_FILE_MAX_BITS	26
1566 #else /* CONFIG_64BIT */
1567 # define PTE_FILE_MAX_BITS	59
1568 #endif /* CONFIG_64BIT */
1569 
1570 #define pte_to_pgoff(__pte) \
1571 	((((__pte).pte >> 12) << 7) + (((__pte).pte >> 1) & 0x7f))
1572 
1573 #define pgoff_to_pte(__off) \
1574 	((pte_t) { ((((__off) & 0x7f) << 1) + (((__off) >> 7) << 12)) \
1575 		   | _PAGE_TYPE_FILE })
1576 
1577 #endif /* !__ASSEMBLY__ */
1578 
1579 #define kern_addr_valid(addr)   (1)
1580 
1581 extern int vmem_add_mapping(unsigned long start, unsigned long size);
1582 extern int vmem_remove_mapping(unsigned long start, unsigned long size);
1583 extern int s390_enable_sie(void);
1584 
1585 /*
1586  * No page table caches to initialise
1587  */
1588 static inline void pgtable_cache_init(void) { }
1589 static inline void check_pgt_cache(void) { }
1590 
1591 #include <asm-generic/pgtable.h>
1592 
1593 #endif /* _S390_PAGE_H */
1594