?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge remote-tracking branch 'upstream/master' into kvm-devel
[linux-2.6/kvm.git] / arch / s390 / mm / pgtable.c
blob5d56c2b95b14a1298c4b3dffa0759e69588fa7c4
1 /*
2 * Copyright IBM Corp. 2007,2009
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>
9 #include <linux/gfp.h>
10 #include <linux/mm.h>
11 #include <linux/swap.h>
12 #include <linux/smp.h>
13 #include <linux/highmem.h>
14 #include <linux/pagemap.h>
15 #include <linux/spinlock.h>
16 #include <linux/module.h>
17 #include <linux/quicklist.h>
18 #include <linux/rcupdate.h>
19 #include <linux/slab.h>
20
21 #include <asm/system.h>
22 #include <asm/pgtable.h>
23 #include <asm/pgalloc.h>
24 #include <asm/tlb.h>
25 #include <asm/tlbflush.h>
26 #include <asm/mmu_context.h>
27
28 #ifndef CONFIG_64BIT
29 #define ALLOC_ORDER 1
30 #define FRAG_MASK 0x0f
31 #else
32 #define ALLOC_ORDER 2
33 #define FRAG_MASK 0x03
34 #endif
35
36 unsigned long VMALLOC_START = VMALLOC_END - VMALLOC_SIZE;
37 EXPORT_SYMBOL(VMALLOC_START);
38
39 static int __init parse_vmalloc(char *arg)
40 {
41 if (!arg)
42 return -EINVAL;
43 VMALLOC_START = (VMALLOC_END - memparse(arg, &arg)) & PAGE_MASK;
44 return 0;
45 }
46 early_param("vmalloc", parse_vmalloc);
47
48 unsigned long *crst_table_alloc(struct mm_struct *mm)
49 {
50 struct page *page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
51
52 if (!page)
53 return NULL;
54 return (unsigned long *) page_to_phys(page);
55 }
56
57 void crst_table_free(struct mm_struct *mm, unsigned long *table)
58 {
59 free_pages((unsigned long) table, ALLOC_ORDER);
60 }
61
62 #ifdef CONFIG_64BIT
63 int crst_table_upgrade(struct mm_struct *mm, unsigned long limit)
64 {
65 unsigned long *table, *pgd;
66 unsigned long entry;
67
68 BUG_ON(limit > (1UL << 53));
69 repeat:
70 table = crst_table_alloc(mm);
71 if (!table)
72 return -ENOMEM;
73 spin_lock_bh(&mm->page_table_lock);
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;
92 mm->task_size = mm->context.asce_limit;
93 table = NULL;
94 }
95 spin_unlock_bh(&mm->page_table_lock);
96 if (table)
97 crst_table_free(mm, table);
98 if (mm->context.asce_limit < limit)
99 goto repeat;
100 update_mm(mm, current);
101 return 0;
102 }
103
104 void crst_table_downgrade(struct mm_struct *mm, unsigned long limit)
105 {
106 pgd_t *pgd;
107
108 if (mm->context.asce_limit <= limit)
109 return;
110 __tlb_flush_mm(mm);
111 while (mm->context.asce_limit > limit) {
112 pgd = mm->pgd;
113 switch (pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) {
114 case _REGION_ENTRY_TYPE_R2:
115 mm->context.asce_limit = 1UL << 42;
116 mm->context.asce_bits = _ASCE_TABLE_LENGTH |
117 _ASCE_USER_BITS |
118 _ASCE_TYPE_REGION3;
119 break;
120 case _REGION_ENTRY_TYPE_R3:
121 mm->context.asce_limit = 1UL << 31;
122 mm->context.asce_bits = _ASCE_TABLE_LENGTH |
123 _ASCE_USER_BITS |
124 _ASCE_TYPE_SEGMENT;
125 break;
126 default:
127 BUG();
128 }
129 mm->pgd = (pgd_t *) (pgd_val(*pgd) & _REGION_ENTRY_ORIGIN);
130 mm->task_size = mm->context.asce_limit;
131 crst_table_free(mm, (unsigned long *) pgd);
132 }
133 update_mm(mm, current);
134 }
135 #endif
136
137 #ifdef CONFIG_PGSTE
138
139 /**
140 * gmap_alloc - allocate a guest address space
141 * @mm: pointer to the parent mm_struct
142 *
143 * Returns a guest address space structure.
144 */
145 struct gmap *gmap_alloc(struct mm_struct *mm)
146 {
147 struct gmap *gmap;
148 struct page *page;
149 unsigned long *table;
150
151 gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL);
152 if (!gmap)
153 goto out;
154 INIT_LIST_HEAD(&gmap->crst_list);
155 gmap->mm = mm;
156 page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
157 if (!page)
158 goto out_free;
159 list_add(&page->lru, &gmap->crst_list);
160 table = (unsigned long *) page_to_phys(page);
161 crst_table_init(table, _REGION1_ENTRY_EMPTY);
162 gmap->table = table;
163 gmap->asce = _ASCE_TYPE_REGION1 | _ASCE_TABLE_LENGTH |
164 _ASCE_USER_BITS | __pa(table);
165 list_add(&gmap->list, &mm->context.gmap_list);
166 return gmap;
167
168 out_free:
169 kfree(gmap);
170 out:
171 return NULL;
172 }
173 EXPORT_SYMBOL_GPL(gmap_alloc);
174
175 static int gmap_unlink_segment(struct gmap *gmap, unsigned long *table)
176 {
177 struct gmap_pgtable *mp;
178 struct gmap_rmap *rmap;
179 struct page *page;
180
181 if (*table & _SEGMENT_ENTRY_INV)
182 return 0;
183 page = pfn_to_page(*table >> PAGE_SHIFT);
184 mp = (struct gmap_pgtable *) page->index;
185 list_for_each_entry(rmap, &mp->mapper, list) {
186 if (rmap->entry != table)
187 continue;
188 list_del(&rmap->list);
189 kfree(rmap);
190 break;
191 }
192 *table = _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO | mp->vmaddr;
193 return 1;
194 }
195
196 static void gmap_flush_tlb(struct gmap *gmap)
197 {
198 if (MACHINE_HAS_IDTE)
199 __tlb_flush_idte((unsigned long) gmap->table |
200 _ASCE_TYPE_REGION1);
201 else
202 __tlb_flush_global();
203 }
204
205 /**
206 * gmap_free - free a guest address space
207 * @gmap: pointer to the guest address space structure
208 */
209 void gmap_free(struct gmap *gmap)
210 {
211 struct page *page, *next;
212 unsigned long *table;
213 int i;
214
215
216 /* Flush tlb. */
217 if (MACHINE_HAS_IDTE)
218 __tlb_flush_idte((unsigned long) gmap->table |
219 _ASCE_TYPE_REGION1);
220 else
221 __tlb_flush_global();
222
223 /* Free all segment & region tables. */
224 down_read(&gmap->mm->mmap_sem);
225 list_for_each_entry_safe(page, next, &gmap->crst_list, lru) {
226 table = (unsigned long *) page_to_phys(page);
227 if ((*table & _REGION_ENTRY_TYPE_MASK) == 0)
228 /* Remove gmap rmap structures for segment table. */
229 for (i = 0; i < PTRS_PER_PMD; i++, table++)
230 gmap_unlink_segment(gmap, table);
231 __free_pages(page, ALLOC_ORDER);
232 }
233 up_read(&gmap->mm->mmap_sem);
234 list_del(&gmap->list);
235 kfree(gmap);
236 }
237 EXPORT_SYMBOL_GPL(gmap_free);
238
239 /**
240 * gmap_enable - switch primary space to the guest address space
241 * @gmap: pointer to the guest address space structure
242 */
243 void gmap_enable(struct gmap *gmap)
244 {
245 S390_lowcore.gmap = (unsigned long) gmap;
246 }
247 EXPORT_SYMBOL_GPL(gmap_enable);
248
249 /**
250 * gmap_disable - switch back to the standard primary address space
251 * @gmap: pointer to the guest address space structure
252 */
253 void gmap_disable(struct gmap *gmap)
254 {
255 S390_lowcore.gmap = 0UL;
256 }
257 EXPORT_SYMBOL_GPL(gmap_disable);
258
259 static int gmap_alloc_table(struct gmap *gmap,
260 unsigned long *table, unsigned long init)
261 {
262 struct page *page;
263 unsigned long *new;
264
265 page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
266 if (!page)
267 return -ENOMEM;
268 new = (unsigned long *) page_to_phys(page);
269 crst_table_init(new, init);
270 down_read(&gmap->mm->mmap_sem);
271 if (*table & _REGION_ENTRY_INV) {
272 list_add(&page->lru, &gmap->crst_list);
273 *table = (unsigned long) new | _REGION_ENTRY_LENGTH |
274 (*table & _REGION_ENTRY_TYPE_MASK);
275 } else
276 __free_pages(page, ALLOC_ORDER);
277 up_read(&gmap->mm->mmap_sem);
278 return 0;
279 }
280
281 /**
282 * gmap_unmap_segment - unmap segment from the guest address space
283 * @gmap: pointer to the guest address space structure
284 * @addr: address in the guest address space
285 * @len: length of the memory area to unmap
286 *
287 * Returns 0 if the unmap succeded, -EINVAL if not.
288 */
289 int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
290 {
291 unsigned long *table;
292 unsigned long off;
293 int flush;
294
295 if ((to | len) & (PMD_SIZE - 1))
296 return -EINVAL;
297 if (len == 0 || to + len < to)
298 return -EINVAL;
299
300 flush = 0;
301 down_read(&gmap->mm->mmap_sem);
302 for (off = 0; off < len; off += PMD_SIZE) {
303 /* Walk the guest addr space page table */
304 table = gmap->table + (((to + off) >> 53) & 0x7ff);
305 if (*table & _REGION_ENTRY_INV)
306 goto out;
307 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
308 table = table + (((to + off) >> 42) & 0x7ff);
309 if (*table & _REGION_ENTRY_INV)
310 goto out;
311 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
312 table = table + (((to + off) >> 31) & 0x7ff);
313 if (*table & _REGION_ENTRY_INV)
314 goto out;
315 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
316 table = table + (((to + off) >> 20) & 0x7ff);
317
318 /* Clear segment table entry in guest address space. */
319 flush |= gmap_unlink_segment(gmap, table);
320 *table = _SEGMENT_ENTRY_INV;
321 }
322 out:
323 up_read(&gmap->mm->mmap_sem);
324 if (flush)
325 gmap_flush_tlb(gmap);
326 return 0;
327 }
328 EXPORT_SYMBOL_GPL(gmap_unmap_segment);
329
330 /**
331 * gmap_mmap_segment - map a segment to the guest address space
332 * @gmap: pointer to the guest address space structure
333 * @from: source address in the parent address space
334 * @to: target address in the guest address space
335 *
336 * Returns 0 if the mmap succeded, -EINVAL or -ENOMEM if not.
337 */
338 int gmap_map_segment(struct gmap *gmap, unsigned long from,
339 unsigned long to, unsigned long len)
340 {
341 unsigned long *table;
342 unsigned long off;
343 int flush;
344
345 if ((from | to | len) & (PMD_SIZE - 1))
346 return -EINVAL;
347 if (len == 0 || from + len > PGDIR_SIZE ||
348 from + len < from || to + len < to)
349 return -EINVAL;
350
351 flush = 0;
352 down_read(&gmap->mm->mmap_sem);
353 for (off = 0; off < len; off += PMD_SIZE) {
354 /* Walk the gmap address space page table */
355 table = gmap->table + (((to + off) >> 53) & 0x7ff);
356 if ((*table & _REGION_ENTRY_INV) &&
357 gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY))
358 goto out_unmap;
359 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
360 table = table + (((to + off) >> 42) & 0x7ff);
361 if ((*table & _REGION_ENTRY_INV) &&
362 gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY))
363 goto out_unmap;
364 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
365 table = table + (((to + off) >> 31) & 0x7ff);
366 if ((*table & _REGION_ENTRY_INV) &&
367 gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY))
368 goto out_unmap;
369 table = (unsigned long *) (*table & _REGION_ENTRY_ORIGIN);
370 table = table + (((to + off) >> 20) & 0x7ff);
371
372 /* Store 'from' address in an invalid segment table entry. */
373 flush |= gmap_unlink_segment(gmap, table);
374 *table = _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO | (from + off);
375 }
376 up_read(&gmap->mm->mmap_sem);
377 if (flush)
378 gmap_flush_tlb(gmap);
379 return 0;
380
381 out_unmap:
382 up_read(&gmap->mm->mmap_sem);
383 gmap_unmap_segment(gmap, to, len);
384 return -ENOMEM;
385 }
386 EXPORT_SYMBOL_GPL(gmap_map_segment);
387
388 unsigned long gmap_fault(unsigned long address, struct gmap *gmap)
389 {
390 unsigned long *table, vmaddr, segment;
391 struct mm_struct *mm;
392 struct gmap_pgtable *mp;
393 struct gmap_rmap *rmap;
394 struct vm_area_struct *vma;
395 struct page *page;
396 pgd_t *pgd;
397 pud_t *pud;
398 pmd_t *pmd;
399
400 current->thread.gmap_addr = address;
401 mm = gmap->mm;
402 /* Walk the gmap address space page table */
403 table = gmap->table + ((address >> 53) & 0x7ff);
404 if (unlikely(*table & _REGION_ENTRY_INV))
405 return -EFAULT;
406 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
407 table = table + ((address >> 42) & 0x7ff);
408 if (unlikely(*table & _REGION_ENTRY_INV))
409 return -EFAULT;
410 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
411 table = table + ((address >> 31) & 0x7ff);
412 if (unlikely(*table & _REGION_ENTRY_INV))
413 return -EFAULT;
414 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
415 table = table + ((address >> 20) & 0x7ff);
416
417 /* Convert the gmap address to an mm address. */
418 segment = *table;
419 if (likely(!(segment & _SEGMENT_ENTRY_INV))) {
420 page = pfn_to_page(segment >> PAGE_SHIFT);
421 mp = (struct gmap_pgtable *) page->index;
422 return mp->vmaddr | (address & ~PMD_MASK);
423 } else if (segment & _SEGMENT_ENTRY_RO) {
424 vmaddr = segment & _SEGMENT_ENTRY_ORIGIN;
425 vma = find_vma(mm, vmaddr);
426 if (!vma || vma->vm_start > vmaddr)
427 return -EFAULT;
428
429 /* Walk the parent mm page table */
430 pgd = pgd_offset(mm, vmaddr);
431 pud = pud_alloc(mm, pgd, vmaddr);
432 if (!pud)
433 return -ENOMEM;
434 pmd = pmd_alloc(mm, pud, vmaddr);
435 if (!pmd)
436 return -ENOMEM;
437 if (!pmd_present(*pmd) &&
438 __pte_alloc(mm, vma, pmd, vmaddr))
439 return -ENOMEM;
440 /* pmd now points to a valid segment table entry. */
441 rmap = kmalloc(sizeof(*rmap), GFP_KERNEL|__GFP_REPEAT);
442 if (!rmap)
443 return -ENOMEM;
444 /* Link gmap segment table entry location to page table. */
445 page = pmd_page(*pmd);
446 mp = (struct gmap_pgtable *) page->index;
447 rmap->entry = table;
448 list_add(&rmap->list, &mp->mapper);
449 /* Set gmap segment table entry to page table. */
450 *table = pmd_val(*pmd) & PAGE_MASK;
451 return vmaddr | (address & ~PMD_MASK);
452 }
453 return -EFAULT;
454
455 }
456 EXPORT_SYMBOL_GPL(gmap_fault);
457
458 void gmap_unmap_notifier(struct mm_struct *mm, unsigned long *table)
459 {
460 struct gmap_rmap *rmap, *next;
461 struct gmap_pgtable *mp;
462 struct page *page;
463 int flush;
464
465 flush = 0;
466 spin_lock(&mm->page_table_lock);
467 page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
468 mp = (struct gmap_pgtable *) page->index;
469 list_for_each_entry_safe(rmap, next, &mp->mapper, list) {
470 *rmap->entry =
471 _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO | mp->vmaddr;
472 list_del(&rmap->list);
473 kfree(rmap);
474 flush = 1;
475 }
476 spin_unlock(&mm->page_table_lock);
477 if (flush)
478 __tlb_flush_global();
479 }
480
481 static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm,
482 unsigned long vmaddr)
483 {
484 struct page *page;
485 unsigned long *table;
486 struct gmap_pgtable *mp;
487
488 page = alloc_page(GFP_KERNEL|__GFP_REPEAT);
489 if (!page)
490 return NULL;
491 mp = kmalloc(sizeof(*mp), GFP_KERNEL|__GFP_REPEAT);
492 if (!mp) {
493 __free_page(page);
494 return NULL;
495 }
496 pgtable_page_ctor(page);
497 mp->vmaddr = vmaddr & PMD_MASK;
498 INIT_LIST_HEAD(&mp->mapper);
499 page->index = (unsigned long) mp;
500 atomic_set(&page->_mapcount, 3);
501 table = (unsigned long *) page_to_phys(page);
502 clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE/2);
503 clear_table(table + PTRS_PER_PTE, 0, PAGE_SIZE/2);
504 return table;
505 }
506
507 static inline void page_table_free_pgste(unsigned long *table)
508 {
509 struct page *page;
510 struct gmap_pgtable *mp;
511
512 page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
513 mp = (struct gmap_pgtable *) page->index;
514 BUG_ON(!list_empty(&mp->mapper));
515 pgtable_page_ctor(page);
516 atomic_set(&page->_mapcount, -1);
517 kfree(mp);
518 __free_page(page);
519 }
520
521 #else /* CONFIG_PGSTE */
522
523 static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm,
524 unsigned long vmaddr)
525 {
526 return NULL;
527 }
528
529 static inline void page_table_free_pgste(unsigned long *table)
530 {
531 }
532
533 static inline void gmap_unmap_notifier(struct mm_struct *mm,
534 unsigned long *table)
535 {
536 }
537
538 #endif /* CONFIG_PGSTE */
539
540 static inline unsigned int atomic_xor_bits(atomic_t *v, unsigned int bits)
541 {
542 unsigned int old, new;
543
544 do {
545 old = atomic_read(v);
546 new = old ^ bits;
547 } while (atomic_cmpxchg(v, old, new) != old);
548 return new;
549 }
550
551 /*
552 * page table entry allocation/free routines.
553 */
554 unsigned long *page_table_alloc(struct mm_struct *mm, unsigned long vmaddr)
555 {
556 struct page *page;
557 unsigned long *table;
558 unsigned int mask, bit;
559
560 if (mm_has_pgste(mm))
561 return page_table_alloc_pgste(mm, vmaddr);
562 /* Allocate fragments of a 4K page as 1K/2K page table */
563 spin_lock_bh(&mm->context.list_lock);
564 mask = FRAG_MASK;
565 if (!list_empty(&mm->context.pgtable_list)) {
566 page = list_first_entry(&mm->context.pgtable_list,
567 struct page, lru);
568 table = (unsigned long *) page_to_phys(page);
569 mask = atomic_read(&page->_mapcount);
570 mask = mask | (mask >> 4);
571 }
572 if ((mask & FRAG_MASK) == FRAG_MASK) {
573 spin_unlock_bh(&mm->context.list_lock);
574 page = alloc_page(GFP_KERNEL|__GFP_REPEAT);
575 if (!page)
576 return NULL;
577 pgtable_page_ctor(page);
578 atomic_set(&page->_mapcount, 1);
579 table = (unsigned long *) page_to_phys(page);
580 clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE);
581 spin_lock_bh(&mm->context.list_lock);
582 list_add(&page->lru, &mm->context.pgtable_list);
583 } else {
584 for (bit = 1; mask & bit; bit <<= 1)
585 table += PTRS_PER_PTE;
586 mask = atomic_xor_bits(&page->_mapcount, bit);
587 if ((mask & FRAG_MASK) == FRAG_MASK)
588 list_del(&page->lru);
589 }
590 spin_unlock_bh(&mm->context.list_lock);
591 return table;
592 }
593
594 void page_table_free(struct mm_struct *mm, unsigned long *table)
595 {
596 struct page *page;
597 unsigned int bit, mask;
598
599 if (mm_has_pgste(mm)) {
600 gmap_unmap_notifier(mm, table);
601 return page_table_free_pgste(table);
602 }
603 /* Free 1K/2K page table fragment of a 4K page */
604 page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
605 bit = 1 << ((__pa(table) & ~PAGE_MASK)/(PTRS_PER_PTE*sizeof(pte_t)));
606 spin_lock_bh(&mm->context.list_lock);
607 if ((atomic_read(&page->_mapcount) & FRAG_MASK) != FRAG_MASK)
608 list_del(&page->lru);
609 mask = atomic_xor_bits(&page->_mapcount, bit);
610 if (mask & FRAG_MASK)
611 list_add(&page->lru, &mm->context.pgtable_list);
612 spin_unlock_bh(&mm->context.list_lock);
613 if (mask == 0) {
614 pgtable_page_dtor(page);
615 atomic_set(&page->_mapcount, -1);
616 __free_page(page);
617 }
618 }
619
620 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
621
622 static void __page_table_free_rcu(void *table, unsigned bit)
623 {
624 struct page *page;
625
626 if (bit == FRAG_MASK)
627 return page_table_free_pgste(table);
628 /* Free 1K/2K page table fragment of a 4K page */
629 page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
630 if (atomic_xor_bits(&page->_mapcount, bit) == 0) {
631 pgtable_page_dtor(page);
632 atomic_set(&page->_mapcount, -1);
633 __free_page(page);
634 }
635 }
636
637 void page_table_free_rcu(struct mmu_gather *tlb, unsigned long *table)
638 {
639 struct mm_struct *mm;
640 struct page *page;
641 unsigned int bit, mask;
642
643 mm = tlb->mm;
644 if (mm_has_pgste(mm)) {
645 gmap_unmap_notifier(mm, table);
646 table = (unsigned long *) (__pa(table) | FRAG_MASK);
647 tlb_remove_table(tlb, table);
648 return;
649 }
650 bit = 1 << ((__pa(table) & ~PAGE_MASK) / (PTRS_PER_PTE*sizeof(pte_t)));
651 page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
652 spin_lock_bh(&mm->context.list_lock);
653 if ((atomic_read(&page->_mapcount) & FRAG_MASK) != FRAG_MASK)
654 list_del(&page->lru);
655 mask = atomic_xor_bits(&page->_mapcount, bit | (bit << 4));
656 if (mask & FRAG_MASK)
657 list_add_tail(&page->lru, &mm->context.pgtable_list);
658 spin_unlock_bh(&mm->context.list_lock);
659 table = (unsigned long *) (__pa(table) | (bit << 4));
660 tlb_remove_table(tlb, table);
661 }
662
663 void __tlb_remove_table(void *_table)
664 {
665 void *table = (void *)((unsigned long) _table & PAGE_MASK);
666 unsigned type = (unsigned long) _table & ~PAGE_MASK;
667
668 if (type)
669 __page_table_free_rcu(table, type);
670 else
671 free_pages((unsigned long) table, ALLOC_ORDER);
672 }
673
674 #endif
675
676 /*
677 * switch on pgstes for its userspace process (for kvm)
678 */
679 int s390_enable_sie(void)
680 {
681 struct task_struct *tsk = current;
682 struct mm_struct *mm, *old_mm;
683
684 /* Do we have switched amode? If no, we cannot do sie */
685 if (user_mode == HOME_SPACE_MODE)
686 return -EINVAL;
687
688 /* Do we have pgstes? if yes, we are done */
689 if (mm_has_pgste(tsk->mm))
690 return 0;
691
692 /* lets check if we are allowed to replace the mm */
693 task_lock(tsk);
694 if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 ||
695 #ifdef CONFIG_AIO
696 !hlist_empty(&tsk->mm->ioctx_list) ||
697 #endif
698 tsk->mm != tsk->active_mm) {
699 task_unlock(tsk);
700 return -EINVAL;
701 }
702 task_unlock(tsk);
703
704 /* we copy the mm and let dup_mm create the page tables with_pgstes */
705 tsk->mm->context.alloc_pgste = 1;
706 mm = dup_mm(tsk);
707 tsk->mm->context.alloc_pgste = 0;
708 if (!mm)
709 return -ENOMEM;
710
711 /* Now lets check again if something happened */
712 task_lock(tsk);
713 if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 ||
714 #ifdef CONFIG_AIO
715 !hlist_empty(&tsk->mm->ioctx_list) ||
716 #endif
717 tsk->mm != tsk->active_mm) {
718 mmput(mm);
719 task_unlock(tsk);
720 return -EINVAL;
721 }
722
723 /* ok, we are alone. No ptrace, no threads, etc. */
724 old_mm = tsk->mm;
725 tsk->mm = tsk->active_mm = mm;
726 preempt_disable();
727 update_mm(mm, tsk);
728 atomic_inc(&mm->context.attach_count);
729 atomic_dec(&old_mm->context.attach_count);
730 cpumask_set_cpu(smp_processor_id(), mm_cpumask(mm));
731 preempt_enable();
732 task_unlock(tsk);
733 mmput(old_mm);
734 return 0;
735 }
736 EXPORT_SYMBOL_GPL(s390_enable_sie);
737
738 #if defined(CONFIG_DEBUG_PAGEALLOC) && defined(CONFIG_HIBERNATION)
739 bool kernel_page_present(struct page *page)
740 {
741 unsigned long addr;
742 int cc;
743
744 addr = page_to_phys(page);
745 asm volatile(
746 " lra %1,0(%1)\n"
747 " ipm %0\n"
748 " srl %0,28"
749 : "=d" (cc), "+a" (addr) : : "cc");
750 return cc == 0;
751 }
752 #endif /* CONFIG_HIBERNATION && CONFIG_DEBUG_PAGEALLOC */
kernel-based virtual machine