Commit | Line | Data |
---|---|---|
3610cce8 | 1 | /* |
a53c8fab | 2 | * Copyright IBM Corp. 2007, 2011 |
3610cce8 MS |
3 | * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com> |
4 | */ | |
5 | ||
6 | #include <linux/sched.h> | |
7 | #include <linux/kernel.h> | |
8 | #include <linux/errno.h> | |
5a0e3ad6 | 9 | #include <linux/gfp.h> |
3610cce8 MS |
10 | #include <linux/mm.h> |
11 | #include <linux/swap.h> | |
12 | #include <linux/smp.h> | |
13 | #include <linux/highmem.h> | |
3610cce8 MS |
14 | #include <linux/pagemap.h> |
15 | #include <linux/spinlock.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/quicklist.h> | |
80217147 | 18 | #include <linux/rcupdate.h> |
e5992f2e | 19 | #include <linux/slab.h> |
b31288fa | 20 | #include <linux/swapops.h> |
3610cce8 | 21 | |
3610cce8 MS |
22 | #include <asm/pgtable.h> |
23 | #include <asm/pgalloc.h> | |
24 | #include <asm/tlb.h> | |
25 | #include <asm/tlbflush.h> | |
6252d702 | 26 | #include <asm/mmu_context.h> |
3610cce8 MS |
27 | |
28 | #ifndef CONFIG_64BIT | |
29 | #define ALLOC_ORDER 1 | |
36409f63 | 30 | #define FRAG_MASK 0x0f |
3610cce8 MS |
31 | #else |
32 | #define ALLOC_ORDER 2 | |
36409f63 | 33 | #define FRAG_MASK 0x03 |
3610cce8 MS |
34 | #endif |
35 | ||
239a6425 | 36 | |
043d0708 | 37 | unsigned long *crst_table_alloc(struct mm_struct *mm) |
3610cce8 MS |
38 | { |
39 | struct page *page = alloc_pages(GFP_KERNEL, ALLOC_ORDER); | |
40 | ||
41 | if (!page) | |
42 | return NULL; | |
3610cce8 MS |
43 | return (unsigned long *) page_to_phys(page); |
44 | } | |
45 | ||
80217147 MS |
46 | void crst_table_free(struct mm_struct *mm, unsigned long *table) |
47 | { | |
043d0708 | 48 | free_pages((unsigned long) table, ALLOC_ORDER); |
80217147 MS |
49 | } |
50 | ||
6252d702 | 51 | #ifdef CONFIG_64BIT |
10607864 MS |
52 | static void __crst_table_upgrade(void *arg) |
53 | { | |
54 | struct mm_struct *mm = arg; | |
55 | ||
56 | if (current->active_mm == mm) | |
57 | update_mm(mm, current); | |
58 | __tlb_flush_local(); | |
59 | } | |
60 | ||
6252d702 MS |
61 | int crst_table_upgrade(struct mm_struct *mm, unsigned long limit) |
62 | { | |
63 | unsigned long *table, *pgd; | |
64 | unsigned long entry; | |
10607864 | 65 | int flush; |
6252d702 MS |
66 | |
67 | BUG_ON(limit > (1UL << 53)); | |
10607864 | 68 | flush = 0; |
6252d702 | 69 | repeat: |
043d0708 | 70 | table = crst_table_alloc(mm); |
6252d702 MS |
71 | if (!table) |
72 | return -ENOMEM; | |
80217147 | 73 | spin_lock_bh(&mm->page_table_lock); |
6252d702 MS |
74 | if (mm->context.asce_limit < limit) { |
75 | pgd = (unsigned long *) mm->pgd; | |
76 | if (mm->context.asce_limit <= (1UL << 31)) { | |
77 | entry = _REGION3_ENTRY_EMPTY; | |
78 | mm->context.asce_limit = 1UL << 42; | |
79 | mm->context.asce_bits = _ASCE_TABLE_LENGTH | | |
80 | _ASCE_USER_BITS | | |
81 | _ASCE_TYPE_REGION3; | |
82 | } else { | |
83 | entry = _REGION2_ENTRY_EMPTY; | |
84 | mm->context.asce_limit = 1UL << 53; | |
85 | mm->context.asce_bits = _ASCE_TABLE_LENGTH | | |
86 | _ASCE_USER_BITS | | |
87 | _ASCE_TYPE_REGION2; | |
88 | } | |
89 | crst_table_init(table, entry); | |
90 | pgd_populate(mm, (pgd_t *) table, (pud_t *) pgd); | |
91 | mm->pgd = (pgd_t *) table; | |
f481bfaf | 92 | mm->task_size = mm->context.asce_limit; |
6252d702 | 93 | table = NULL; |
10607864 | 94 | flush = 1; |
6252d702 | 95 | } |
80217147 | 96 | spin_unlock_bh(&mm->page_table_lock); |
6252d702 MS |
97 | if (table) |
98 | crst_table_free(mm, table); | |
99 | if (mm->context.asce_limit < limit) | |
100 | goto repeat; | |
10607864 MS |
101 | if (flush) |
102 | on_each_cpu(__crst_table_upgrade, mm, 0); | |
6252d702 MS |
103 | return 0; |
104 | } | |
105 | ||
106 | void crst_table_downgrade(struct mm_struct *mm, unsigned long limit) | |
107 | { | |
108 | pgd_t *pgd; | |
109 | ||
10607864 MS |
110 | if (current->active_mm == mm) |
111 | __tlb_flush_mm(mm); | |
6252d702 MS |
112 | while (mm->context.asce_limit > limit) { |
113 | pgd = mm->pgd; | |
114 | switch (pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) { | |
115 | case _REGION_ENTRY_TYPE_R2: | |
116 | mm->context.asce_limit = 1UL << 42; | |
117 | mm->context.asce_bits = _ASCE_TABLE_LENGTH | | |
118 | _ASCE_USER_BITS | | |
119 | _ASCE_TYPE_REGION3; | |
120 | break; | |
121 | case _REGION_ENTRY_TYPE_R3: | |
122 | mm->context.asce_limit = 1UL << 31; | |
123 | mm->context.asce_bits = _ASCE_TABLE_LENGTH | | |
124 | _ASCE_USER_BITS | | |
125 | _ASCE_TYPE_SEGMENT; | |
126 | break; | |
127 | default: | |
128 | BUG(); | |
129 | } | |
130 | mm->pgd = (pgd_t *) (pgd_val(*pgd) & _REGION_ENTRY_ORIGIN); | |
f481bfaf | 131 | mm->task_size = mm->context.asce_limit; |
6252d702 MS |
132 | crst_table_free(mm, (unsigned long *) pgd); |
133 | } | |
10607864 MS |
134 | if (current->active_mm == mm) |
135 | update_mm(mm, current); | |
6252d702 MS |
136 | } |
137 | #endif | |
138 | ||
e5992f2e MS |
139 | #ifdef CONFIG_PGSTE |
140 | ||
141 | /** | |
142 | * gmap_alloc - allocate a guest address space | |
143 | * @mm: pointer to the parent mm_struct | |
144 | * | |
145 | * Returns a guest address space structure. | |
146 | */ | |
147 | struct gmap *gmap_alloc(struct mm_struct *mm) | |
36409f63 | 148 | { |
e5992f2e MS |
149 | struct gmap *gmap; |
150 | struct page *page; | |
151 | unsigned long *table; | |
36409f63 | 152 | |
e5992f2e MS |
153 | gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL); |
154 | if (!gmap) | |
155 | goto out; | |
156 | INIT_LIST_HEAD(&gmap->crst_list); | |
157 | gmap->mm = mm; | |
158 | page = alloc_pages(GFP_KERNEL, ALLOC_ORDER); | |
159 | if (!page) | |
160 | goto out_free; | |
161 | list_add(&page->lru, &gmap->crst_list); | |
162 | table = (unsigned long *) page_to_phys(page); | |
163 | crst_table_init(table, _REGION1_ENTRY_EMPTY); | |
164 | gmap->table = table; | |
480e5926 CB |
165 | gmap->asce = _ASCE_TYPE_REGION1 | _ASCE_TABLE_LENGTH | |
166 | _ASCE_USER_BITS | __pa(table); | |
e5992f2e MS |
167 | list_add(&gmap->list, &mm->context.gmap_list); |
168 | return gmap; | |
169 | ||
170 | out_free: | |
171 | kfree(gmap); | |
172 | out: | |
173 | return NULL; | |
36409f63 | 174 | } |
e5992f2e | 175 | EXPORT_SYMBOL_GPL(gmap_alloc); |
36409f63 | 176 | |
e5992f2e MS |
177 | static int gmap_unlink_segment(struct gmap *gmap, unsigned long *table) |
178 | { | |
179 | struct gmap_pgtable *mp; | |
180 | struct gmap_rmap *rmap; | |
181 | struct page *page; | |
182 | ||
e5098611 | 183 | if (*table & _SEGMENT_ENTRY_INVALID) |
e5992f2e MS |
184 | return 0; |
185 | page = pfn_to_page(*table >> PAGE_SHIFT); | |
186 | mp = (struct gmap_pgtable *) page->index; | |
187 | list_for_each_entry(rmap, &mp->mapper, list) { | |
188 | if (rmap->entry != table) | |
189 | continue; | |
190 | list_del(&rmap->list); | |
191 | kfree(rmap); | |
192 | break; | |
193 | } | |
e5098611 | 194 | *table = mp->vmaddr | _SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_PROTECT; |
e5992f2e MS |
195 | return 1; |
196 | } | |
197 | ||
198 | static void gmap_flush_tlb(struct gmap *gmap) | |
199 | { | |
200 | if (MACHINE_HAS_IDTE) | |
201 | __tlb_flush_idte((unsigned long) gmap->table | | |
202 | _ASCE_TYPE_REGION1); | |
203 | else | |
204 | __tlb_flush_global(); | |
205 | } | |
206 | ||
207 | /** | |
208 | * gmap_free - free a guest address space | |
209 | * @gmap: pointer to the guest address space structure | |
3610cce8 | 210 | */ |
e5992f2e MS |
211 | void gmap_free(struct gmap *gmap) |
212 | { | |
213 | struct page *page, *next; | |
214 | unsigned long *table; | |
215 | int i; | |
216 | ||
217 | ||
218 | /* Flush tlb. */ | |
219 | if (MACHINE_HAS_IDTE) | |
220 | __tlb_flush_idte((unsigned long) gmap->table | | |
221 | _ASCE_TYPE_REGION1); | |
222 | else | |
223 | __tlb_flush_global(); | |
224 | ||
225 | /* Free all segment & region tables. */ | |
226 | down_read(&gmap->mm->mmap_sem); | |
cc772456 | 227 | spin_lock(&gmap->mm->page_table_lock); |
e5992f2e MS |
228 | list_for_each_entry_safe(page, next, &gmap->crst_list, lru) { |
229 | table = (unsigned long *) page_to_phys(page); | |
230 | if ((*table & _REGION_ENTRY_TYPE_MASK) == 0) | |
231 | /* Remove gmap rmap structures for segment table. */ | |
232 | for (i = 0; i < PTRS_PER_PMD; i++, table++) | |
233 | gmap_unlink_segment(gmap, table); | |
234 | __free_pages(page, ALLOC_ORDER); | |
235 | } | |
cc772456 | 236 | spin_unlock(&gmap->mm->page_table_lock); |
e5992f2e MS |
237 | up_read(&gmap->mm->mmap_sem); |
238 | list_del(&gmap->list); | |
239 | kfree(gmap); | |
240 | } | |
241 | EXPORT_SYMBOL_GPL(gmap_free); | |
242 | ||
243 | /** | |
244 | * gmap_enable - switch primary space to the guest address space | |
245 | * @gmap: pointer to the guest address space structure | |
246 | */ | |
247 | void gmap_enable(struct gmap *gmap) | |
248 | { | |
e5992f2e MS |
249 | S390_lowcore.gmap = (unsigned long) gmap; |
250 | } | |
251 | EXPORT_SYMBOL_GPL(gmap_enable); | |
252 | ||
253 | /** | |
254 | * gmap_disable - switch back to the standard primary address space | |
255 | * @gmap: pointer to the guest address space structure | |
256 | */ | |
257 | void gmap_disable(struct gmap *gmap) | |
258 | { | |
e5992f2e MS |
259 | S390_lowcore.gmap = 0UL; |
260 | } | |
261 | EXPORT_SYMBOL_GPL(gmap_disable); | |
262 | ||
a9162f23 CO |
263 | /* |
264 | * gmap_alloc_table is assumed to be called with mmap_sem held | |
265 | */ | |
e5992f2e | 266 | static int gmap_alloc_table(struct gmap *gmap, |
984e2a59 HC |
267 | unsigned long *table, unsigned long init) |
268 | __releases(&gmap->mm->page_table_lock) | |
269 | __acquires(&gmap->mm->page_table_lock) | |
e5992f2e MS |
270 | { |
271 | struct page *page; | |
272 | unsigned long *new; | |
273 | ||
c86cce2a CB |
274 | /* since we dont free the gmap table until gmap_free we can unlock */ |
275 | spin_unlock(&gmap->mm->page_table_lock); | |
e5992f2e | 276 | page = alloc_pages(GFP_KERNEL, ALLOC_ORDER); |
c86cce2a | 277 | spin_lock(&gmap->mm->page_table_lock); |
e5992f2e MS |
278 | if (!page) |
279 | return -ENOMEM; | |
280 | new = (unsigned long *) page_to_phys(page); | |
281 | crst_table_init(new, init); | |
e5098611 | 282 | if (*table & _REGION_ENTRY_INVALID) { |
e5992f2e MS |
283 | list_add(&page->lru, &gmap->crst_list); |
284 | *table = (unsigned long) new | _REGION_ENTRY_LENGTH | | |
285 | (*table & _REGION_ENTRY_TYPE_MASK); | |
286 | } else | |
287 | __free_pages(page, ALLOC_ORDER); | |
e5992f2e MS |
288 | return 0; |
289 | } | |
290 | ||
291 | /** | |
292 | * gmap_unmap_segment - unmap segment from the guest address space | |
293 | * @gmap: pointer to the guest address space structure | |
294 | * @addr: address in the guest address space | |
295 | * @len: length of the memory area to unmap | |
296 | * | |
b4a96015 | 297 | * Returns 0 if the unmap succeeded, -EINVAL if not. |
e5992f2e MS |
298 | */ |
299 | int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len) | |
300 | { | |
301 | unsigned long *table; | |
302 | unsigned long off; | |
303 | int flush; | |
304 | ||
305 | if ((to | len) & (PMD_SIZE - 1)) | |
306 | return -EINVAL; | |
307 | if (len == 0 || to + len < to) | |
308 | return -EINVAL; | |
309 | ||
310 | flush = 0; | |
311 | down_read(&gmap->mm->mmap_sem); | |
cc772456 | 312 | spin_lock(&gmap->mm->page_table_lock); |
e5992f2e MS |
313 | for (off = 0; off < len; off += PMD_SIZE) { |
314 | /* Walk the guest addr space page table */ | |
315 | table = gmap->table + (((to + off) >> 53) & 0x7ff); | |
e5098611 | 316 | if (*table & _REGION_ENTRY_INVALID) |
05873df9 | 317 | goto out; |
e5992f2e MS |
318 | table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); |
319 | table = table + (((to + off) >> 42) & 0x7ff); | |
e5098611 | 320 | if (*table & _REGION_ENTRY_INVALID) |
05873df9 | 321 | goto out; |
e5992f2e MS |
322 | table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); |
323 | table = table + (((to + off) >> 31) & 0x7ff); | |
e5098611 | 324 | if (*table & _REGION_ENTRY_INVALID) |
05873df9 | 325 | goto out; |
e5992f2e MS |
326 | table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); |
327 | table = table + (((to + off) >> 20) & 0x7ff); | |
328 | ||
329 | /* Clear segment table entry in guest address space. */ | |
330 | flush |= gmap_unlink_segment(gmap, table); | |
e5098611 | 331 | *table = _SEGMENT_ENTRY_INVALID; |
e5992f2e | 332 | } |
05873df9 | 333 | out: |
cc772456 | 334 | spin_unlock(&gmap->mm->page_table_lock); |
e5992f2e MS |
335 | up_read(&gmap->mm->mmap_sem); |
336 | if (flush) | |
337 | gmap_flush_tlb(gmap); | |
338 | return 0; | |
339 | } | |
340 | EXPORT_SYMBOL_GPL(gmap_unmap_segment); | |
341 | ||
342 | /** | |
343 | * gmap_mmap_segment - map a segment to the guest address space | |
344 | * @gmap: pointer to the guest address space structure | |
345 | * @from: source address in the parent address space | |
346 | * @to: target address in the guest address space | |
347 | * | |
b4a96015 | 348 | * Returns 0 if the mmap succeeded, -EINVAL or -ENOMEM if not. |
e5992f2e MS |
349 | */ |
350 | int gmap_map_segment(struct gmap *gmap, unsigned long from, | |
351 | unsigned long to, unsigned long len) | |
352 | { | |
353 | unsigned long *table; | |
354 | unsigned long off; | |
355 | int flush; | |
356 | ||
357 | if ((from | to | len) & (PMD_SIZE - 1)) | |
358 | return -EINVAL; | |
ee6ee55b | 359 | if (len == 0 || from + len > TASK_MAX_SIZE || |
e5992f2e MS |
360 | from + len < from || to + len < to) |
361 | return -EINVAL; | |
362 | ||
363 | flush = 0; | |
364 | down_read(&gmap->mm->mmap_sem); | |
cc772456 | 365 | spin_lock(&gmap->mm->page_table_lock); |
e5992f2e MS |
366 | for (off = 0; off < len; off += PMD_SIZE) { |
367 | /* Walk the gmap address space page table */ | |
368 | table = gmap->table + (((to + off) >> 53) & 0x7ff); | |
e5098611 | 369 | if ((*table & _REGION_ENTRY_INVALID) && |
e5992f2e MS |
370 | gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY)) |
371 | goto out_unmap; | |
372 | table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); | |
373 | table = table + (((to + off) >> 42) & 0x7ff); | |
e5098611 | 374 | if ((*table & _REGION_ENTRY_INVALID) && |
e5992f2e MS |
375 | gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY)) |
376 | goto out_unmap; | |
377 | table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); | |
378 | table = table + (((to + off) >> 31) & 0x7ff); | |
e5098611 | 379 | if ((*table & _REGION_ENTRY_INVALID) && |
e5992f2e MS |
380 | gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY)) |
381 | goto out_unmap; | |
382 | table = (unsigned long *) (*table & _REGION_ENTRY_ORIGIN); | |
383 | table = table + (((to + off) >> 20) & 0x7ff); | |
384 | ||
385 | /* Store 'from' address in an invalid segment table entry. */ | |
386 | flush |= gmap_unlink_segment(gmap, table); | |
e5098611 MS |
387 | *table = (from + off) | (_SEGMENT_ENTRY_INVALID | |
388 | _SEGMENT_ENTRY_PROTECT); | |
e5992f2e | 389 | } |
cc772456 | 390 | spin_unlock(&gmap->mm->page_table_lock); |
e5992f2e MS |
391 | up_read(&gmap->mm->mmap_sem); |
392 | if (flush) | |
393 | gmap_flush_tlb(gmap); | |
394 | return 0; | |
395 | ||
396 | out_unmap: | |
cc772456 | 397 | spin_unlock(&gmap->mm->page_table_lock); |
e5992f2e MS |
398 | up_read(&gmap->mm->mmap_sem); |
399 | gmap_unmap_segment(gmap, to, len); | |
400 | return -ENOMEM; | |
401 | } | |
402 | EXPORT_SYMBOL_GPL(gmap_map_segment); | |
403 | ||
c5034945 HC |
404 | static unsigned long *gmap_table_walk(unsigned long address, struct gmap *gmap) |
405 | { | |
406 | unsigned long *table; | |
407 | ||
408 | table = gmap->table + ((address >> 53) & 0x7ff); | |
e5098611 | 409 | if (unlikely(*table & _REGION_ENTRY_INVALID)) |
c5034945 HC |
410 | return ERR_PTR(-EFAULT); |
411 | table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); | |
412 | table = table + ((address >> 42) & 0x7ff); | |
e5098611 | 413 | if (unlikely(*table & _REGION_ENTRY_INVALID)) |
c5034945 HC |
414 | return ERR_PTR(-EFAULT); |
415 | table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); | |
416 | table = table + ((address >> 31) & 0x7ff); | |
e5098611 | 417 | if (unlikely(*table & _REGION_ENTRY_INVALID)) |
c5034945 HC |
418 | return ERR_PTR(-EFAULT); |
419 | table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); | |
420 | table = table + ((address >> 20) & 0x7ff); | |
421 | return table; | |
422 | } | |
423 | ||
424 | /** | |
425 | * __gmap_translate - translate a guest address to a user space address | |
426 | * @address: guest address | |
427 | * @gmap: pointer to guest mapping meta data structure | |
428 | * | |
429 | * Returns user space address which corresponds to the guest address or | |
430 | * -EFAULT if no such mapping exists. | |
431 | * This function does not establish potentially missing page table entries. | |
432 | * The mmap_sem of the mm that belongs to the address space must be held | |
433 | * when this function gets called. | |
434 | */ | |
435 | unsigned long __gmap_translate(unsigned long address, struct gmap *gmap) | |
436 | { | |
437 | unsigned long *segment_ptr, vmaddr, segment; | |
438 | struct gmap_pgtable *mp; | |
439 | struct page *page; | |
440 | ||
441 | current->thread.gmap_addr = address; | |
442 | segment_ptr = gmap_table_walk(address, gmap); | |
443 | if (IS_ERR(segment_ptr)) | |
444 | return PTR_ERR(segment_ptr); | |
445 | /* Convert the gmap address to an mm address. */ | |
446 | segment = *segment_ptr; | |
e5098611 | 447 | if (!(segment & _SEGMENT_ENTRY_INVALID)) { |
c5034945 HC |
448 | page = pfn_to_page(segment >> PAGE_SHIFT); |
449 | mp = (struct gmap_pgtable *) page->index; | |
450 | return mp->vmaddr | (address & ~PMD_MASK); | |
e5098611 | 451 | } else if (segment & _SEGMENT_ENTRY_PROTECT) { |
c5034945 HC |
452 | vmaddr = segment & _SEGMENT_ENTRY_ORIGIN; |
453 | return vmaddr | (address & ~PMD_MASK); | |
454 | } | |
455 | return -EFAULT; | |
456 | } | |
457 | EXPORT_SYMBOL_GPL(__gmap_translate); | |
458 | ||
459 | /** | |
460 | * gmap_translate - translate a guest address to a user space address | |
461 | * @address: guest address | |
462 | * @gmap: pointer to guest mapping meta data structure | |
463 | * | |
464 | * Returns user space address which corresponds to the guest address or | |
465 | * -EFAULT if no such mapping exists. | |
466 | * This function does not establish potentially missing page table entries. | |
467 | */ | |
468 | unsigned long gmap_translate(unsigned long address, struct gmap *gmap) | |
469 | { | |
470 | unsigned long rc; | |
471 | ||
472 | down_read(&gmap->mm->mmap_sem); | |
473 | rc = __gmap_translate(address, gmap); | |
474 | up_read(&gmap->mm->mmap_sem); | |
475 | return rc; | |
476 | } | |
477 | EXPORT_SYMBOL_GPL(gmap_translate); | |
478 | ||
d3383632 MS |
479 | static int gmap_connect_pgtable(unsigned long address, unsigned long segment, |
480 | unsigned long *segment_ptr, struct gmap *gmap) | |
e5992f2e | 481 | { |
ab8e5235 | 482 | unsigned long vmaddr; |
c5034945 | 483 | struct vm_area_struct *vma; |
e5992f2e MS |
484 | struct gmap_pgtable *mp; |
485 | struct gmap_rmap *rmap; | |
c5034945 | 486 | struct mm_struct *mm; |
e5992f2e MS |
487 | struct page *page; |
488 | pgd_t *pgd; | |
489 | pud_t *pud; | |
490 | pmd_t *pmd; | |
491 | ||
ab8e5235 MS |
492 | mm = gmap->mm; |
493 | vmaddr = segment & _SEGMENT_ENTRY_ORIGIN; | |
494 | vma = find_vma(mm, vmaddr); | |
495 | if (!vma || vma->vm_start > vmaddr) | |
496 | return -EFAULT; | |
497 | /* Walk the parent mm page table */ | |
498 | pgd = pgd_offset(mm, vmaddr); | |
499 | pud = pud_alloc(mm, pgd, vmaddr); | |
500 | if (!pud) | |
501 | return -ENOMEM; | |
502 | pmd = pmd_alloc(mm, pud, vmaddr); | |
503 | if (!pmd) | |
504 | return -ENOMEM; | |
505 | if (!pmd_present(*pmd) && | |
506 | __pte_alloc(mm, vma, pmd, vmaddr)) | |
507 | return -ENOMEM; | |
508 | /* pmd now points to a valid segment table entry. */ | |
509 | rmap = kmalloc(sizeof(*rmap), GFP_KERNEL|__GFP_REPEAT); | |
510 | if (!rmap) | |
511 | return -ENOMEM; | |
512 | /* Link gmap segment table entry location to page table. */ | |
513 | page = pmd_page(*pmd); | |
514 | mp = (struct gmap_pgtable *) page->index; | |
d3383632 | 515 | rmap->gmap = gmap; |
ab8e5235 | 516 | rmap->entry = segment_ptr; |
e86cbd87 | 517 | rmap->vmaddr = address & PMD_MASK; |
ab8e5235 MS |
518 | spin_lock(&mm->page_table_lock); |
519 | if (*segment_ptr == segment) { | |
520 | list_add(&rmap->list, &mp->mapper); | |
521 | /* Set gmap segment table entry to page table. */ | |
522 | *segment_ptr = pmd_val(*pmd) & PAGE_MASK; | |
523 | rmap = NULL; | |
524 | } | |
525 | spin_unlock(&mm->page_table_lock); | |
526 | kfree(rmap); | |
527 | return 0; | |
528 | } | |
529 | ||
530 | static void gmap_disconnect_pgtable(struct mm_struct *mm, unsigned long *table) | |
531 | { | |
532 | struct gmap_rmap *rmap, *next; | |
533 | struct gmap_pgtable *mp; | |
534 | struct page *page; | |
535 | int flush; | |
536 | ||
537 | flush = 0; | |
538 | spin_lock(&mm->page_table_lock); | |
539 | page = pfn_to_page(__pa(table) >> PAGE_SHIFT); | |
540 | mp = (struct gmap_pgtable *) page->index; | |
541 | list_for_each_entry_safe(rmap, next, &mp->mapper, list) { | |
e5098611 MS |
542 | *rmap->entry = mp->vmaddr | (_SEGMENT_ENTRY_INVALID | |
543 | _SEGMENT_ENTRY_PROTECT); | |
ab8e5235 MS |
544 | list_del(&rmap->list); |
545 | kfree(rmap); | |
546 | flush = 1; | |
547 | } | |
548 | spin_unlock(&mm->page_table_lock); | |
549 | if (flush) | |
550 | __tlb_flush_global(); | |
551 | } | |
552 | ||
553 | /* | |
554 | * this function is assumed to be called with mmap_sem held | |
555 | */ | |
556 | unsigned long __gmap_fault(unsigned long address, struct gmap *gmap) | |
557 | { | |
558 | unsigned long *segment_ptr, segment; | |
559 | struct gmap_pgtable *mp; | |
560 | struct page *page; | |
561 | int rc; | |
562 | ||
e5992f2e | 563 | current->thread.gmap_addr = address; |
c5034945 HC |
564 | segment_ptr = gmap_table_walk(address, gmap); |
565 | if (IS_ERR(segment_ptr)) | |
e5992f2e | 566 | return -EFAULT; |
e5992f2e | 567 | /* Convert the gmap address to an mm address. */ |
ab8e5235 MS |
568 | while (1) { |
569 | segment = *segment_ptr; | |
e5098611 | 570 | if (!(segment & _SEGMENT_ENTRY_INVALID)) { |
ab8e5235 MS |
571 | /* Page table is present */ |
572 | page = pfn_to_page(segment >> PAGE_SHIFT); | |
573 | mp = (struct gmap_pgtable *) page->index; | |
574 | return mp->vmaddr | (address & ~PMD_MASK); | |
575 | } | |
e5098611 | 576 | if (!(segment & _SEGMENT_ENTRY_PROTECT)) |
ab8e5235 MS |
577 | /* Nothing mapped in the gmap address space. */ |
578 | break; | |
d3383632 | 579 | rc = gmap_connect_pgtable(address, segment, segment_ptr, gmap); |
ab8e5235 MS |
580 | if (rc) |
581 | return rc; | |
e5992f2e MS |
582 | } |
583 | return -EFAULT; | |
499069e1 CO |
584 | } |
585 | ||
586 | unsigned long gmap_fault(unsigned long address, struct gmap *gmap) | |
587 | { | |
588 | unsigned long rc; | |
589 | ||
590 | down_read(&gmap->mm->mmap_sem); | |
591 | rc = __gmap_fault(address, gmap); | |
592 | up_read(&gmap->mm->mmap_sem); | |
e5992f2e | 593 | |
499069e1 | 594 | return rc; |
e5992f2e MS |
595 | } |
596 | EXPORT_SYMBOL_GPL(gmap_fault); | |
597 | ||
b31288fa KW |
598 | static void gmap_zap_swap_entry(swp_entry_t entry, struct mm_struct *mm) |
599 | { | |
600 | if (!non_swap_entry(entry)) | |
601 | dec_mm_counter(mm, MM_SWAPENTS); | |
602 | else if (is_migration_entry(entry)) { | |
603 | struct page *page = migration_entry_to_page(entry); | |
604 | ||
605 | if (PageAnon(page)) | |
606 | dec_mm_counter(mm, MM_ANONPAGES); | |
607 | else | |
608 | dec_mm_counter(mm, MM_FILEPAGES); | |
609 | } | |
610 | free_swap_and_cache(entry); | |
611 | } | |
612 | ||
613 | /** | |
614 | * The mm->mmap_sem lock must be held | |
615 | */ | |
616 | static void gmap_zap_unused(struct mm_struct *mm, unsigned long address) | |
617 | { | |
618 | unsigned long ptev, pgstev; | |
619 | spinlock_t *ptl; | |
620 | pgste_t pgste; | |
621 | pte_t *ptep, pte; | |
622 | ||
623 | ptep = get_locked_pte(mm, address, &ptl); | |
624 | if (unlikely(!ptep)) | |
625 | return; | |
626 | pte = *ptep; | |
627 | if (!pte_swap(pte)) | |
628 | goto out_pte; | |
629 | /* Zap unused and logically-zero pages */ | |
630 | pgste = pgste_get_lock(ptep); | |
631 | pgstev = pgste_val(pgste); | |
632 | ptev = pte_val(pte); | |
633 | if (((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED) || | |
634 | ((pgstev & _PGSTE_GPS_ZERO) && (ptev & _PAGE_INVALID))) { | |
635 | gmap_zap_swap_entry(pte_to_swp_entry(pte), mm); | |
636 | pte_clear(mm, address, ptep); | |
637 | } | |
638 | pgste_set_unlock(ptep, pgste); | |
639 | out_pte: | |
640 | pte_unmap_unlock(*ptep, ptl); | |
641 | } | |
642 | ||
643 | /* | |
644 | * this function is assumed to be called with mmap_sem held | |
645 | */ | |
646 | void __gmap_zap(unsigned long address, struct gmap *gmap) | |
647 | { | |
648 | unsigned long *table, *segment_ptr; | |
649 | unsigned long segment, pgstev, ptev; | |
650 | struct gmap_pgtable *mp; | |
651 | struct page *page; | |
652 | ||
653 | segment_ptr = gmap_table_walk(address, gmap); | |
654 | if (IS_ERR(segment_ptr)) | |
655 | return; | |
656 | segment = *segment_ptr; | |
657 | if (segment & _SEGMENT_ENTRY_INVALID) | |
658 | return; | |
659 | page = pfn_to_page(segment >> PAGE_SHIFT); | |
660 | mp = (struct gmap_pgtable *) page->index; | |
661 | address = mp->vmaddr | (address & ~PMD_MASK); | |
662 | /* Page table is present */ | |
663 | table = (unsigned long *)(segment & _SEGMENT_ENTRY_ORIGIN); | |
664 | table = table + ((address >> 12) & 0xff); | |
665 | pgstev = table[PTRS_PER_PTE]; | |
666 | ptev = table[0]; | |
667 | /* quick check, checked again with locks held */ | |
668 | if (((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED) || | |
669 | ((pgstev & _PGSTE_GPS_ZERO) && (ptev & _PAGE_INVALID))) | |
670 | gmap_zap_unused(gmap->mm, address); | |
671 | } | |
672 | EXPORT_SYMBOL_GPL(__gmap_zap); | |
673 | ||
388186bc CB |
674 | void gmap_discard(unsigned long from, unsigned long to, struct gmap *gmap) |
675 | { | |
676 | ||
677 | unsigned long *table, address, size; | |
678 | struct vm_area_struct *vma; | |
679 | struct gmap_pgtable *mp; | |
680 | struct page *page; | |
681 | ||
682 | down_read(&gmap->mm->mmap_sem); | |
683 | address = from; | |
684 | while (address < to) { | |
685 | /* Walk the gmap address space page table */ | |
686 | table = gmap->table + ((address >> 53) & 0x7ff); | |
e5098611 | 687 | if (unlikely(*table & _REGION_ENTRY_INVALID)) { |
388186bc CB |
688 | address = (address + PMD_SIZE) & PMD_MASK; |
689 | continue; | |
690 | } | |
691 | table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); | |
692 | table = table + ((address >> 42) & 0x7ff); | |
e5098611 | 693 | if (unlikely(*table & _REGION_ENTRY_INVALID)) { |
388186bc CB |
694 | address = (address + PMD_SIZE) & PMD_MASK; |
695 | continue; | |
696 | } | |
697 | table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); | |
698 | table = table + ((address >> 31) & 0x7ff); | |
e5098611 | 699 | if (unlikely(*table & _REGION_ENTRY_INVALID)) { |
388186bc CB |
700 | address = (address + PMD_SIZE) & PMD_MASK; |
701 | continue; | |
702 | } | |
703 | table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); | |
704 | table = table + ((address >> 20) & 0x7ff); | |
e5098611 | 705 | if (unlikely(*table & _SEGMENT_ENTRY_INVALID)) { |
388186bc CB |
706 | address = (address + PMD_SIZE) & PMD_MASK; |
707 | continue; | |
708 | } | |
709 | page = pfn_to_page(*table >> PAGE_SHIFT); | |
710 | mp = (struct gmap_pgtable *) page->index; | |
711 | vma = find_vma(gmap->mm, mp->vmaddr); | |
712 | size = min(to - address, PMD_SIZE - (address & ~PMD_MASK)); | |
713 | zap_page_range(vma, mp->vmaddr | (address & ~PMD_MASK), | |
714 | size, NULL); | |
715 | address = (address + PMD_SIZE) & PMD_MASK; | |
716 | } | |
717 | up_read(&gmap->mm->mmap_sem); | |
718 | } | |
719 | EXPORT_SYMBOL_GPL(gmap_discard); | |
720 | ||
d3383632 MS |
721 | static LIST_HEAD(gmap_notifier_list); |
722 | static DEFINE_SPINLOCK(gmap_notifier_lock); | |
723 | ||
724 | /** | |
725 | * gmap_register_ipte_notifier - register a pte invalidation callback | |
726 | * @nb: pointer to the gmap notifier block | |
727 | */ | |
728 | void gmap_register_ipte_notifier(struct gmap_notifier *nb) | |
729 | { | |
730 | spin_lock(&gmap_notifier_lock); | |
731 | list_add(&nb->list, &gmap_notifier_list); | |
732 | spin_unlock(&gmap_notifier_lock); | |
733 | } | |
734 | EXPORT_SYMBOL_GPL(gmap_register_ipte_notifier); | |
735 | ||
736 | /** | |
737 | * gmap_unregister_ipte_notifier - remove a pte invalidation callback | |
738 | * @nb: pointer to the gmap notifier block | |
739 | */ | |
740 | void gmap_unregister_ipte_notifier(struct gmap_notifier *nb) | |
741 | { | |
742 | spin_lock(&gmap_notifier_lock); | |
743 | list_del_init(&nb->list); | |
744 | spin_unlock(&gmap_notifier_lock); | |
745 | } | |
746 | EXPORT_SYMBOL_GPL(gmap_unregister_ipte_notifier); | |
747 | ||
748 | /** | |
749 | * gmap_ipte_notify - mark a range of ptes for invalidation notification | |
750 | * @gmap: pointer to guest mapping meta data structure | |
c7c5be73 | 751 | * @start: virtual address in the guest address space |
d3383632 MS |
752 | * @len: size of area |
753 | * | |
754 | * Returns 0 if for each page in the given range a gmap mapping exists and | |
755 | * the invalidation notification could be set. If the gmap mapping is missing | |
756 | * for one or more pages -EFAULT is returned. If no memory could be allocated | |
757 | * -ENOMEM is returned. This function establishes missing page table entries. | |
758 | */ | |
759 | int gmap_ipte_notify(struct gmap *gmap, unsigned long start, unsigned long len) | |
760 | { | |
761 | unsigned long addr; | |
762 | spinlock_t *ptl; | |
763 | pte_t *ptep, entry; | |
764 | pgste_t pgste; | |
765 | int rc = 0; | |
766 | ||
767 | if ((start & ~PAGE_MASK) || (len & ~PAGE_MASK)) | |
768 | return -EINVAL; | |
769 | down_read(&gmap->mm->mmap_sem); | |
770 | while (len) { | |
771 | /* Convert gmap address and connect the page tables */ | |
772 | addr = __gmap_fault(start, gmap); | |
773 | if (IS_ERR_VALUE(addr)) { | |
774 | rc = addr; | |
775 | break; | |
776 | } | |
777 | /* Get the page mapped */ | |
bb4b42ce | 778 | if (fixup_user_fault(current, gmap->mm, addr, FAULT_FLAG_WRITE)) { |
d3383632 MS |
779 | rc = -EFAULT; |
780 | break; | |
781 | } | |
782 | /* Walk the process page table, lock and get pte pointer */ | |
783 | ptep = get_locked_pte(gmap->mm, addr, &ptl); | |
784 | if (unlikely(!ptep)) | |
785 | continue; | |
786 | /* Set notification bit in the pgste of the pte */ | |
787 | entry = *ptep; | |
e5098611 | 788 | if ((pte_val(entry) & (_PAGE_INVALID | _PAGE_PROTECT)) == 0) { |
d3383632 | 789 | pgste = pgste_get_lock(ptep); |
0d0dafc1 | 790 | pgste_val(pgste) |= PGSTE_IN_BIT; |
d3383632 MS |
791 | pgste_set_unlock(ptep, pgste); |
792 | start += PAGE_SIZE; | |
793 | len -= PAGE_SIZE; | |
794 | } | |
795 | spin_unlock(ptl); | |
796 | } | |
797 | up_read(&gmap->mm->mmap_sem); | |
798 | return rc; | |
799 | } | |
800 | EXPORT_SYMBOL_GPL(gmap_ipte_notify); | |
801 | ||
802 | /** | |
803 | * gmap_do_ipte_notify - call all invalidation callbacks for a specific pte. | |
804 | * @mm: pointer to the process mm_struct | |
d3383632 MS |
805 | * @pte: pointer to the page table entry |
806 | * | |
807 | * This function is assumed to be called with the page table lock held | |
808 | * for the pte to notify. | |
809 | */ | |
aaeff84a | 810 | void gmap_do_ipte_notify(struct mm_struct *mm, pte_t *pte) |
d3383632 MS |
811 | { |
812 | unsigned long segment_offset; | |
813 | struct gmap_notifier *nb; | |
814 | struct gmap_pgtable *mp; | |
815 | struct gmap_rmap *rmap; | |
816 | struct page *page; | |
817 | ||
818 | segment_offset = ((unsigned long) pte) & (255 * sizeof(pte_t)); | |
819 | segment_offset = segment_offset * (4096 / sizeof(pte_t)); | |
820 | page = pfn_to_page(__pa(pte) >> PAGE_SHIFT); | |
821 | mp = (struct gmap_pgtable *) page->index; | |
822 | spin_lock(&gmap_notifier_lock); | |
823 | list_for_each_entry(rmap, &mp->mapper, list) { | |
824 | list_for_each_entry(nb, &gmap_notifier_list, list) | |
825 | nb->notifier_call(rmap->gmap, | |
826 | rmap->vmaddr + segment_offset); | |
827 | } | |
828 | spin_unlock(&gmap_notifier_lock); | |
829 | } | |
830 | ||
3eabaee9 MS |
831 | static inline int page_table_with_pgste(struct page *page) |
832 | { | |
833 | return atomic_read(&page->_mapcount) == 0; | |
834 | } | |
835 | ||
e5992f2e MS |
836 | static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm, |
837 | unsigned long vmaddr) | |
36409f63 MS |
838 | { |
839 | struct page *page; | |
840 | unsigned long *table; | |
e5992f2e | 841 | struct gmap_pgtable *mp; |
36409f63 MS |
842 | |
843 | page = alloc_page(GFP_KERNEL|__GFP_REPEAT); | |
844 | if (!page) | |
845 | return NULL; | |
e5992f2e MS |
846 | mp = kmalloc(sizeof(*mp), GFP_KERNEL|__GFP_REPEAT); |
847 | if (!mp) { | |
848 | __free_page(page); | |
849 | return NULL; | |
850 | } | |
e89cfa58 KS |
851 | if (!pgtable_page_ctor(page)) { |
852 | kfree(mp); | |
853 | __free_page(page); | |
854 | return NULL; | |
855 | } | |
e5992f2e MS |
856 | mp->vmaddr = vmaddr & PMD_MASK; |
857 | INIT_LIST_HEAD(&mp->mapper); | |
858 | page->index = (unsigned long) mp; | |
3eabaee9 | 859 | atomic_set(&page->_mapcount, 0); |
36409f63 | 860 | table = (unsigned long *) page_to_phys(page); |
e5098611 | 861 | clear_table(table, _PAGE_INVALID, PAGE_SIZE/2); |
0944fe3f MS |
862 | clear_table(table + PTRS_PER_PTE, PGSTE_HR_BIT | PGSTE_HC_BIT, |
863 | PAGE_SIZE/2); | |
36409f63 MS |
864 | return table; |
865 | } | |
866 | ||
867 | static inline void page_table_free_pgste(unsigned long *table) | |
868 | { | |
869 | struct page *page; | |
e5992f2e | 870 | struct gmap_pgtable *mp; |
36409f63 MS |
871 | |
872 | page = pfn_to_page(__pa(table) >> PAGE_SHIFT); | |
e5992f2e MS |
873 | mp = (struct gmap_pgtable *) page->index; |
874 | BUG_ON(!list_empty(&mp->mapper)); | |
2320c579 | 875 | pgtable_page_dtor(page); |
36409f63 | 876 | atomic_set(&page->_mapcount, -1); |
e5992f2e | 877 | kfree(mp); |
36409f63 MS |
878 | __free_page(page); |
879 | } | |
36409f63 | 880 | |
deedabb2 MS |
881 | static inline unsigned long page_table_reset_pte(struct mm_struct *mm, |
882 | pmd_t *pmd, unsigned long addr, unsigned long end) | |
883 | { | |
884 | pte_t *start_pte, *pte; | |
885 | spinlock_t *ptl; | |
886 | pgste_t pgste; | |
887 | ||
888 | start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); | |
889 | pte = start_pte; | |
890 | do { | |
891 | pgste = pgste_get_lock(pte); | |
892 | pgste_val(pgste) &= ~_PGSTE_GPS_USAGE_MASK; | |
893 | pgste_set_unlock(pte, pgste); | |
894 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
895 | pte_unmap_unlock(start_pte, ptl); | |
896 | ||
897 | return addr; | |
898 | } | |
899 | ||
900 | static inline unsigned long page_table_reset_pmd(struct mm_struct *mm, | |
901 | pud_t *pud, unsigned long addr, unsigned long end) | |
902 | { | |
903 | unsigned long next; | |
904 | pmd_t *pmd; | |
905 | ||
906 | pmd = pmd_offset(pud, addr); | |
907 | do { | |
908 | next = pmd_addr_end(addr, end); | |
909 | if (pmd_none_or_clear_bad(pmd)) | |
910 | continue; | |
911 | next = page_table_reset_pte(mm, pmd, addr, next); | |
912 | } while (pmd++, addr = next, addr != end); | |
913 | ||
914 | return addr; | |
915 | } | |
916 | ||
917 | static inline unsigned long page_table_reset_pud(struct mm_struct *mm, | |
918 | pgd_t *pgd, unsigned long addr, unsigned long end) | |
919 | { | |
920 | unsigned long next; | |
921 | pud_t *pud; | |
922 | ||
923 | pud = pud_offset(pgd, addr); | |
924 | do { | |
925 | next = pud_addr_end(addr, end); | |
926 | if (pud_none_or_clear_bad(pud)) | |
927 | continue; | |
928 | next = page_table_reset_pmd(mm, pud, addr, next); | |
929 | } while (pud++, addr = next, addr != end); | |
930 | ||
931 | return addr; | |
932 | } | |
933 | ||
934 | void page_table_reset_pgste(struct mm_struct *mm, | |
935 | unsigned long start, unsigned long end) | |
936 | { | |
937 | unsigned long addr, next; | |
938 | pgd_t *pgd; | |
939 | ||
940 | addr = start; | |
941 | down_read(&mm->mmap_sem); | |
942 | pgd = pgd_offset(mm, addr); | |
943 | do { | |
944 | next = pgd_addr_end(addr, end); | |
945 | if (pgd_none_or_clear_bad(pgd)) | |
946 | continue; | |
947 | next = page_table_reset_pud(mm, pgd, addr, next); | |
948 | } while (pgd++, addr = next, addr != end); | |
949 | up_read(&mm->mmap_sem); | |
950 | } | |
951 | EXPORT_SYMBOL(page_table_reset_pgste); | |
952 | ||
24d5dd02 CB |
953 | int set_guest_storage_key(struct mm_struct *mm, unsigned long addr, |
954 | unsigned long key, bool nq) | |
955 | { | |
956 | spinlock_t *ptl; | |
957 | pgste_t old, new; | |
958 | pte_t *ptep; | |
959 | ||
960 | down_read(&mm->mmap_sem); | |
961 | ptep = get_locked_pte(current->mm, addr, &ptl); | |
962 | if (unlikely(!ptep)) { | |
963 | up_read(&mm->mmap_sem); | |
964 | return -EFAULT; | |
965 | } | |
966 | ||
967 | new = old = pgste_get_lock(ptep); | |
968 | pgste_val(new) &= ~(PGSTE_GR_BIT | PGSTE_GC_BIT | | |
969 | PGSTE_ACC_BITS | PGSTE_FP_BIT); | |
970 | pgste_val(new) |= (key & (_PAGE_CHANGED | _PAGE_REFERENCED)) << 48; | |
971 | pgste_val(new) |= (key & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56; | |
972 | if (!(pte_val(*ptep) & _PAGE_INVALID)) { | |
0944fe3f | 973 | unsigned long address, bits, skey; |
24d5dd02 CB |
974 | |
975 | address = pte_val(*ptep) & PAGE_MASK; | |
0944fe3f | 976 | skey = (unsigned long) page_get_storage_key(address); |
24d5dd02 | 977 | bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED); |
0944fe3f | 978 | skey = key & (_PAGE_ACC_BITS | _PAGE_FP_BIT); |
24d5dd02 | 979 | /* Set storage key ACC and FP */ |
0944fe3f | 980 | page_set_storage_key(address, skey, !nq); |
24d5dd02 CB |
981 | /* Merge host changed & referenced into pgste */ |
982 | pgste_val(new) |= bits << 52; | |
24d5dd02 CB |
983 | } |
984 | /* changing the guest storage key is considered a change of the page */ | |
985 | if ((pgste_val(new) ^ pgste_val(old)) & | |
986 | (PGSTE_ACC_BITS | PGSTE_FP_BIT | PGSTE_GR_BIT | PGSTE_GC_BIT)) | |
0944fe3f | 987 | pgste_val(new) |= PGSTE_HC_BIT; |
24d5dd02 CB |
988 | |
989 | pgste_set_unlock(ptep, new); | |
990 | pte_unmap_unlock(*ptep, ptl); | |
991 | up_read(&mm->mmap_sem); | |
992 | return 0; | |
993 | } | |
994 | EXPORT_SYMBOL(set_guest_storage_key); | |
995 | ||
e5992f2e MS |
996 | #else /* CONFIG_PGSTE */ |
997 | ||
3eabaee9 MS |
998 | static inline int page_table_with_pgste(struct page *page) |
999 | { | |
1000 | return 0; | |
1001 | } | |
1002 | ||
e5992f2e MS |
1003 | static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm, |
1004 | unsigned long vmaddr) | |
1005 | { | |
944291de | 1006 | return NULL; |
e5992f2e MS |
1007 | } |
1008 | ||
1009 | static inline void page_table_free_pgste(unsigned long *table) | |
1010 | { | |
1011 | } | |
1012 | ||
ab8e5235 MS |
1013 | static inline void gmap_disconnect_pgtable(struct mm_struct *mm, |
1014 | unsigned long *table) | |
e5992f2e MS |
1015 | { |
1016 | } | |
1017 | ||
1018 | #endif /* CONFIG_PGSTE */ | |
1019 | ||
1020 | static inline unsigned int atomic_xor_bits(atomic_t *v, unsigned int bits) | |
1021 | { | |
1022 | unsigned int old, new; | |
1023 | ||
1024 | do { | |
1025 | old = atomic_read(v); | |
1026 | new = old ^ bits; | |
1027 | } while (atomic_cmpxchg(v, old, new) != old); | |
1028 | return new; | |
1029 | } | |
1030 | ||
1031 | /* | |
1032 | * page table entry allocation/free routines. | |
1033 | */ | |
1034 | unsigned long *page_table_alloc(struct mm_struct *mm, unsigned long vmaddr) | |
3610cce8 | 1035 | { |
41459d36 HC |
1036 | unsigned long *uninitialized_var(table); |
1037 | struct page *uninitialized_var(page); | |
36409f63 | 1038 | unsigned int mask, bit; |
3610cce8 | 1039 | |
36409f63 | 1040 | if (mm_has_pgste(mm)) |
e5992f2e | 1041 | return page_table_alloc_pgste(mm, vmaddr); |
36409f63 | 1042 | /* Allocate fragments of a 4K page as 1K/2K page table */ |
80217147 | 1043 | spin_lock_bh(&mm->context.list_lock); |
36409f63 | 1044 | mask = FRAG_MASK; |
146e4b3c MS |
1045 | if (!list_empty(&mm->context.pgtable_list)) { |
1046 | page = list_first_entry(&mm->context.pgtable_list, | |
1047 | struct page, lru); | |
36409f63 MS |
1048 | table = (unsigned long *) page_to_phys(page); |
1049 | mask = atomic_read(&page->_mapcount); | |
1050 | mask = mask | (mask >> 4); | |
146e4b3c | 1051 | } |
36409f63 | 1052 | if ((mask & FRAG_MASK) == FRAG_MASK) { |
80217147 | 1053 | spin_unlock_bh(&mm->context.list_lock); |
146e4b3c MS |
1054 | page = alloc_page(GFP_KERNEL|__GFP_REPEAT); |
1055 | if (!page) | |
3610cce8 | 1056 | return NULL; |
e89cfa58 KS |
1057 | if (!pgtable_page_ctor(page)) { |
1058 | __free_page(page); | |
1059 | return NULL; | |
1060 | } | |
36409f63 | 1061 | atomic_set(&page->_mapcount, 1); |
146e4b3c | 1062 | table = (unsigned long *) page_to_phys(page); |
e5098611 | 1063 | clear_table(table, _PAGE_INVALID, PAGE_SIZE); |
80217147 | 1064 | spin_lock_bh(&mm->context.list_lock); |
146e4b3c | 1065 | list_add(&page->lru, &mm->context.pgtable_list); |
36409f63 MS |
1066 | } else { |
1067 | for (bit = 1; mask & bit; bit <<= 1) | |
1068 | table += PTRS_PER_PTE; | |
1069 | mask = atomic_xor_bits(&page->_mapcount, bit); | |
1070 | if ((mask & FRAG_MASK) == FRAG_MASK) | |
1071 | list_del(&page->lru); | |
3610cce8 | 1072 | } |
80217147 | 1073 | spin_unlock_bh(&mm->context.list_lock); |
3610cce8 MS |
1074 | return table; |
1075 | } | |
1076 | ||
36409f63 | 1077 | void page_table_free(struct mm_struct *mm, unsigned long *table) |
80217147 MS |
1078 | { |
1079 | struct page *page; | |
36409f63 | 1080 | unsigned int bit, mask; |
80217147 | 1081 | |
3eabaee9 MS |
1082 | page = pfn_to_page(__pa(table) >> PAGE_SHIFT); |
1083 | if (page_table_with_pgste(page)) { | |
ab8e5235 | 1084 | gmap_disconnect_pgtable(mm, table); |
36409f63 | 1085 | return page_table_free_pgste(table); |
e5992f2e | 1086 | } |
36409f63 | 1087 | /* Free 1K/2K page table fragment of a 4K page */ |
36409f63 MS |
1088 | bit = 1 << ((__pa(table) & ~PAGE_MASK)/(PTRS_PER_PTE*sizeof(pte_t))); |
1089 | spin_lock_bh(&mm->context.list_lock); | |
1090 | if ((atomic_read(&page->_mapcount) & FRAG_MASK) != FRAG_MASK) | |
1091 | list_del(&page->lru); | |
1092 | mask = atomic_xor_bits(&page->_mapcount, bit); | |
1093 | if (mask & FRAG_MASK) | |
1094 | list_add(&page->lru, &mm->context.pgtable_list); | |
1095 | spin_unlock_bh(&mm->context.list_lock); | |
1096 | if (mask == 0) { | |
80217147 | 1097 | pgtable_page_dtor(page); |
36409f63 | 1098 | atomic_set(&page->_mapcount, -1); |
80217147 MS |
1099 | __free_page(page); |
1100 | } | |
1101 | } | |
1102 | ||
36409f63 | 1103 | static void __page_table_free_rcu(void *table, unsigned bit) |
3610cce8 | 1104 | { |
146e4b3c | 1105 | struct page *page; |
3610cce8 | 1106 | |
36409f63 MS |
1107 | if (bit == FRAG_MASK) |
1108 | return page_table_free_pgste(table); | |
36409f63 | 1109 | /* Free 1K/2K page table fragment of a 4K page */ |
146e4b3c | 1110 | page = pfn_to_page(__pa(table) >> PAGE_SHIFT); |
36409f63 | 1111 | if (atomic_xor_bits(&page->_mapcount, bit) == 0) { |
146e4b3c | 1112 | pgtable_page_dtor(page); |
36409f63 | 1113 | atomic_set(&page->_mapcount, -1); |
146e4b3c MS |
1114 | __free_page(page); |
1115 | } | |
1116 | } | |
3610cce8 | 1117 | |
36409f63 | 1118 | void page_table_free_rcu(struct mmu_gather *tlb, unsigned long *table) |
80217147 | 1119 | { |
36409f63 | 1120 | struct mm_struct *mm; |
80217147 | 1121 | struct page *page; |
36409f63 | 1122 | unsigned int bit, mask; |
80217147 | 1123 | |
36409f63 | 1124 | mm = tlb->mm; |
3eabaee9 MS |
1125 | page = pfn_to_page(__pa(table) >> PAGE_SHIFT); |
1126 | if (page_table_with_pgste(page)) { | |
ab8e5235 | 1127 | gmap_disconnect_pgtable(mm, table); |
36409f63 MS |
1128 | table = (unsigned long *) (__pa(table) | FRAG_MASK); |
1129 | tlb_remove_table(tlb, table); | |
1130 | return; | |
80217147 | 1131 | } |
36409f63 | 1132 | bit = 1 << ((__pa(table) & ~PAGE_MASK) / (PTRS_PER_PTE*sizeof(pte_t))); |
80217147 | 1133 | spin_lock_bh(&mm->context.list_lock); |
36409f63 MS |
1134 | if ((atomic_read(&page->_mapcount) & FRAG_MASK) != FRAG_MASK) |
1135 | list_del(&page->lru); | |
1136 | mask = atomic_xor_bits(&page->_mapcount, bit | (bit << 4)); | |
1137 | if (mask & FRAG_MASK) | |
1138 | list_add_tail(&page->lru, &mm->context.pgtable_list); | |
80217147 | 1139 | spin_unlock_bh(&mm->context.list_lock); |
36409f63 MS |
1140 | table = (unsigned long *) (__pa(table) | (bit << 4)); |
1141 | tlb_remove_table(tlb, table); | |
1142 | } | |
1143 | ||
63df41d6 | 1144 | static void __tlb_remove_table(void *_table) |
36409f63 | 1145 | { |
e73b7fff MS |
1146 | const unsigned long mask = (FRAG_MASK << 4) | FRAG_MASK; |
1147 | void *table = (void *)((unsigned long) _table & ~mask); | |
1148 | unsigned type = (unsigned long) _table & mask; | |
36409f63 MS |
1149 | |
1150 | if (type) | |
1151 | __page_table_free_rcu(table, type); | |
1152 | else | |
1153 | free_pages((unsigned long) table, ALLOC_ORDER); | |
80217147 MS |
1154 | } |
1155 | ||
cd94154c MS |
1156 | static void tlb_remove_table_smp_sync(void *arg) |
1157 | { | |
1158 | /* Simply deliver the interrupt */ | |
1159 | } | |
1160 | ||
1161 | static void tlb_remove_table_one(void *table) | |
1162 | { | |
1163 | /* | |
1164 | * This isn't an RCU grace period and hence the page-tables cannot be | |
1165 | * assumed to be actually RCU-freed. | |
1166 | * | |
1167 | * It is however sufficient for software page-table walkers that rely | |
1168 | * on IRQ disabling. See the comment near struct mmu_table_batch. | |
1169 | */ | |
1170 | smp_call_function(tlb_remove_table_smp_sync, NULL, 1); | |
1171 | __tlb_remove_table(table); | |
1172 | } | |
1173 | ||
1174 | static void tlb_remove_table_rcu(struct rcu_head *head) | |
1175 | { | |
1176 | struct mmu_table_batch *batch; | |
1177 | int i; | |
1178 | ||
1179 | batch = container_of(head, struct mmu_table_batch, rcu); | |
1180 | ||
1181 | for (i = 0; i < batch->nr; i++) | |
1182 | __tlb_remove_table(batch->tables[i]); | |
1183 | ||
1184 | free_page((unsigned long)batch); | |
1185 | } | |
1186 | ||
1187 | void tlb_table_flush(struct mmu_gather *tlb) | |
1188 | { | |
1189 | struct mmu_table_batch **batch = &tlb->batch; | |
1190 | ||
1191 | if (*batch) { | |
cd94154c MS |
1192 | call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); |
1193 | *batch = NULL; | |
1194 | } | |
1195 | } | |
1196 | ||
1197 | void tlb_remove_table(struct mmu_gather *tlb, void *table) | |
1198 | { | |
1199 | struct mmu_table_batch **batch = &tlb->batch; | |
1200 | ||
5c474a1e | 1201 | tlb->mm->context.flush_mm = 1; |
cd94154c MS |
1202 | if (*batch == NULL) { |
1203 | *batch = (struct mmu_table_batch *) | |
1204 | __get_free_page(GFP_NOWAIT | __GFP_NOWARN); | |
1205 | if (*batch == NULL) { | |
5c474a1e | 1206 | __tlb_flush_mm_lazy(tlb->mm); |
cd94154c MS |
1207 | tlb_remove_table_one(table); |
1208 | return; | |
1209 | } | |
1210 | (*batch)->nr = 0; | |
1211 | } | |
1212 | (*batch)->tables[(*batch)->nr++] = table; | |
1213 | if ((*batch)->nr == MAX_TABLE_BATCH) | |
5c474a1e | 1214 | tlb_flush_mmu(tlb); |
cd94154c | 1215 | } |
36409f63 | 1216 | |
274023da | 1217 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3eabaee9 | 1218 | static inline void thp_split_vma(struct vm_area_struct *vma) |
274023da GS |
1219 | { |
1220 | unsigned long addr; | |
274023da | 1221 | |
3eabaee9 MS |
1222 | for (addr = vma->vm_start; addr < vma->vm_end; addr += PAGE_SIZE) |
1223 | follow_page(vma, addr, FOLL_SPLIT); | |
274023da GS |
1224 | } |
1225 | ||
3eabaee9 | 1226 | static inline void thp_split_mm(struct mm_struct *mm) |
274023da | 1227 | { |
3eabaee9 | 1228 | struct vm_area_struct *vma; |
274023da | 1229 | |
3eabaee9 | 1230 | for (vma = mm->mmap; vma != NULL; vma = vma->vm_next) { |
274023da GS |
1231 | thp_split_vma(vma); |
1232 | vma->vm_flags &= ~VM_HUGEPAGE; | |
1233 | vma->vm_flags |= VM_NOHUGEPAGE; | |
274023da | 1234 | } |
3eabaee9 MS |
1235 | mm->def_flags |= VM_NOHUGEPAGE; |
1236 | } | |
1237 | #else | |
1238 | static inline void thp_split_mm(struct mm_struct *mm) | |
1239 | { | |
274023da GS |
1240 | } |
1241 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1242 | ||
3eabaee9 MS |
1243 | static unsigned long page_table_realloc_pmd(struct mmu_gather *tlb, |
1244 | struct mm_struct *mm, pud_t *pud, | |
1245 | unsigned long addr, unsigned long end) | |
1246 | { | |
1247 | unsigned long next, *table, *new; | |
1248 | struct page *page; | |
1249 | pmd_t *pmd; | |
1250 | ||
1251 | pmd = pmd_offset(pud, addr); | |
1252 | do { | |
1253 | next = pmd_addr_end(addr, end); | |
1254 | again: | |
1255 | if (pmd_none_or_clear_bad(pmd)) | |
1256 | continue; | |
1257 | table = (unsigned long *) pmd_deref(*pmd); | |
1258 | page = pfn_to_page(__pa(table) >> PAGE_SHIFT); | |
1259 | if (page_table_with_pgste(page)) | |
1260 | continue; | |
1261 | /* Allocate new page table with pgstes */ | |
1262 | new = page_table_alloc_pgste(mm, addr); | |
be39f196 DD |
1263 | if (!new) |
1264 | return -ENOMEM; | |
1265 | ||
3eabaee9 MS |
1266 | spin_lock(&mm->page_table_lock); |
1267 | if (likely((unsigned long *) pmd_deref(*pmd) == table)) { | |
1268 | /* Nuke pmd entry pointing to the "short" page table */ | |
1269 | pmdp_flush_lazy(mm, addr, pmd); | |
1270 | pmd_clear(pmd); | |
1271 | /* Copy ptes from old table to new table */ | |
1272 | memcpy(new, table, PAGE_SIZE/2); | |
1273 | clear_table(table, _PAGE_INVALID, PAGE_SIZE/2); | |
1274 | /* Establish new table */ | |
1275 | pmd_populate(mm, pmd, (pte_t *) new); | |
1276 | /* Free old table with rcu, there might be a walker! */ | |
1277 | page_table_free_rcu(tlb, table); | |
1278 | new = NULL; | |
1279 | } | |
1280 | spin_unlock(&mm->page_table_lock); | |
1281 | if (new) { | |
1282 | page_table_free_pgste(new); | |
1283 | goto again; | |
1284 | } | |
1285 | } while (pmd++, addr = next, addr != end); | |
1286 | ||
1287 | return addr; | |
1288 | } | |
1289 | ||
1290 | static unsigned long page_table_realloc_pud(struct mmu_gather *tlb, | |
1291 | struct mm_struct *mm, pgd_t *pgd, | |
1292 | unsigned long addr, unsigned long end) | |
1293 | { | |
1294 | unsigned long next; | |
1295 | pud_t *pud; | |
1296 | ||
1297 | pud = pud_offset(pgd, addr); | |
1298 | do { | |
1299 | next = pud_addr_end(addr, end); | |
1300 | if (pud_none_or_clear_bad(pud)) | |
1301 | continue; | |
1302 | next = page_table_realloc_pmd(tlb, mm, pud, addr, next); | |
be39f196 DD |
1303 | if (unlikely(IS_ERR_VALUE(next))) |
1304 | return next; | |
3eabaee9 MS |
1305 | } while (pud++, addr = next, addr != end); |
1306 | ||
1307 | return addr; | |
1308 | } | |
1309 | ||
be39f196 DD |
1310 | static unsigned long page_table_realloc(struct mmu_gather *tlb, struct mm_struct *mm, |
1311 | unsigned long addr, unsigned long end) | |
3eabaee9 MS |
1312 | { |
1313 | unsigned long next; | |
1314 | pgd_t *pgd; | |
1315 | ||
1316 | pgd = pgd_offset(mm, addr); | |
1317 | do { | |
1318 | next = pgd_addr_end(addr, end); | |
1319 | if (pgd_none_or_clear_bad(pgd)) | |
1320 | continue; | |
1321 | next = page_table_realloc_pud(tlb, mm, pgd, addr, next); | |
be39f196 DD |
1322 | if (unlikely(IS_ERR_VALUE(next))) |
1323 | return next; | |
3eabaee9 | 1324 | } while (pgd++, addr = next, addr != end); |
be39f196 DD |
1325 | |
1326 | return 0; | |
3eabaee9 MS |
1327 | } |
1328 | ||
402b0862 CO |
1329 | /* |
1330 | * switch on pgstes for its userspace process (for kvm) | |
1331 | */ | |
1332 | int s390_enable_sie(void) | |
1333 | { | |
1334 | struct task_struct *tsk = current; | |
3eabaee9 MS |
1335 | struct mm_struct *mm = tsk->mm; |
1336 | struct mmu_gather tlb; | |
402b0862 | 1337 | |
74b6b522 | 1338 | /* Do we have pgstes? if yes, we are done */ |
36409f63 | 1339 | if (mm_has_pgste(tsk->mm)) |
74b6b522 | 1340 | return 0; |
402b0862 | 1341 | |
3eabaee9 | 1342 | down_write(&mm->mmap_sem); |
274023da GS |
1343 | /* split thp mappings and disable thp for future mappings */ |
1344 | thp_split_mm(mm); | |
3eabaee9 | 1345 | /* Reallocate the page tables with pgstes */ |
ae7a835c | 1346 | tlb_gather_mmu(&tlb, mm, 0, TASK_SIZE); |
be39f196 DD |
1347 | if (!page_table_realloc(&tlb, mm, 0, TASK_SIZE)) |
1348 | mm->context.has_pgste = 1; | |
ae7a835c | 1349 | tlb_finish_mmu(&tlb, 0, TASK_SIZE); |
3eabaee9 MS |
1350 | up_write(&mm->mmap_sem); |
1351 | return mm->context.has_pgste ? 0 : -ENOMEM; | |
402b0862 CO |
1352 | } |
1353 | EXPORT_SYMBOL_GPL(s390_enable_sie); | |
7db11a36 | 1354 | |
75077afb | 1355 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
1ae1c1d0 GS |
1356 | int pmdp_clear_flush_young(struct vm_area_struct *vma, unsigned long address, |
1357 | pmd_t *pmdp) | |
1358 | { | |
1359 | VM_BUG_ON(address & ~HPAGE_PMD_MASK); | |
1360 | /* No need to flush TLB | |
1361 | * On s390 reference bits are in storage key and never in TLB */ | |
1362 | return pmdp_test_and_clear_young(vma, address, pmdp); | |
1363 | } | |
1364 | ||
1365 | int pmdp_set_access_flags(struct vm_area_struct *vma, | |
1366 | unsigned long address, pmd_t *pmdp, | |
1367 | pmd_t entry, int dirty) | |
1368 | { | |
1369 | VM_BUG_ON(address & ~HPAGE_PMD_MASK); | |
1370 | ||
1371 | if (pmd_same(*pmdp, entry)) | |
1372 | return 0; | |
1373 | pmdp_invalidate(vma, address, pmdp); | |
1374 | set_pmd_at(vma->vm_mm, address, pmdp, entry); | |
1375 | return 1; | |
1376 | } | |
1377 | ||
75077afb GS |
1378 | static void pmdp_splitting_flush_sync(void *arg) |
1379 | { | |
1380 | /* Simply deliver the interrupt */ | |
1381 | } | |
1382 | ||
1383 | void pmdp_splitting_flush(struct vm_area_struct *vma, unsigned long address, | |
1384 | pmd_t *pmdp) | |
1385 | { | |
1386 | VM_BUG_ON(address & ~HPAGE_PMD_MASK); | |
1387 | if (!test_and_set_bit(_SEGMENT_ENTRY_SPLIT_BIT, | |
1388 | (unsigned long *) pmdp)) { | |
1389 | /* need to serialize against gup-fast (IRQ disabled) */ | |
1390 | smp_call_function(pmdp_splitting_flush_sync, NULL, 1); | |
1391 | } | |
1392 | } | |
9501d09f | 1393 | |
6b0b50b0 AK |
1394 | void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp, |
1395 | pgtable_t pgtable) | |
9501d09f GS |
1396 | { |
1397 | struct list_head *lh = (struct list_head *) pgtable; | |
1398 | ||
ec66ad66 | 1399 | assert_spin_locked(pmd_lockptr(mm, pmdp)); |
9501d09f GS |
1400 | |
1401 | /* FIFO */ | |
c389a250 | 1402 | if (!pmd_huge_pte(mm, pmdp)) |
9501d09f GS |
1403 | INIT_LIST_HEAD(lh); |
1404 | else | |
c389a250 KS |
1405 | list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp)); |
1406 | pmd_huge_pte(mm, pmdp) = pgtable; | |
9501d09f GS |
1407 | } |
1408 | ||
6b0b50b0 | 1409 | pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp) |
9501d09f GS |
1410 | { |
1411 | struct list_head *lh; | |
1412 | pgtable_t pgtable; | |
1413 | pte_t *ptep; | |
1414 | ||
ec66ad66 | 1415 | assert_spin_locked(pmd_lockptr(mm, pmdp)); |
9501d09f GS |
1416 | |
1417 | /* FIFO */ | |
c389a250 | 1418 | pgtable = pmd_huge_pte(mm, pmdp); |
9501d09f GS |
1419 | lh = (struct list_head *) pgtable; |
1420 | if (list_empty(lh)) | |
c389a250 | 1421 | pmd_huge_pte(mm, pmdp) = NULL; |
9501d09f | 1422 | else { |
c389a250 | 1423 | pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next; |
9501d09f GS |
1424 | list_del(lh); |
1425 | } | |
1426 | ptep = (pte_t *) pgtable; | |
e5098611 | 1427 | pte_val(*ptep) = _PAGE_INVALID; |
9501d09f | 1428 | ptep++; |
e5098611 | 1429 | pte_val(*ptep) = _PAGE_INVALID; |
9501d09f GS |
1430 | return pgtable; |
1431 | } | |
75077afb | 1432 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |