Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright IBM Corp. 2007, 2011
4 * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
5 */
6
7#include <linux/sched.h>
8#include <linux/kernel.h>
9#include <linux/errno.h>
10#include <linux/gfp.h>
11#include <linux/mm.h>
12#include <linux/swap.h>
13#include <linux/smp.h>
14#include <linux/spinlock.h>
15#include <linux/rcupdate.h>
16#include <linux/slab.h>
17#include <linux/swapops.h>
18#include <linux/sysctl.h>
19#include <linux/ksm.h>
20#include <linux/mman.h>
21
22#include <asm/tlb.h>
23#include <asm/tlbflush.h>
24#include <asm/mmu_context.h>
25#include <asm/page-states.h>
26
27pgprot_t pgprot_writecombine(pgprot_t prot)
28{
29 /*
30 * mio_wb_bit_mask may be set on a different CPU, but it is only set
31 * once at init and only read afterwards.
32 */
33 return __pgprot(pgprot_val(prot) | mio_wb_bit_mask);
34}
35EXPORT_SYMBOL_GPL(pgprot_writecombine);
36
37pgprot_t pgprot_writethrough(pgprot_t prot)
38{
39 /*
40 * mio_wb_bit_mask may be set on a different CPU, but it is only set
41 * once at init and only read afterwards.
42 */
43 return __pgprot(pgprot_val(prot) & ~mio_wb_bit_mask);
44}
45EXPORT_SYMBOL_GPL(pgprot_writethrough);
46
47static inline void ptep_ipte_local(struct mm_struct *mm, unsigned long addr,
48 pte_t *ptep, int nodat)
49{
50 unsigned long opt, asce;
51
52 if (MACHINE_HAS_TLB_GUEST) {
53 opt = 0;
54 asce = READ_ONCE(mm->context.gmap_asce);
55 if (asce == 0UL || nodat)
56 opt |= IPTE_NODAT;
57 if (asce != -1UL) {
58 asce = asce ? : mm->context.asce;
59 opt |= IPTE_GUEST_ASCE;
60 }
61 __ptep_ipte(addr, ptep, opt, asce, IPTE_LOCAL);
62 } else {
63 __ptep_ipte(addr, ptep, 0, 0, IPTE_LOCAL);
64 }
65}
66
67static inline void ptep_ipte_global(struct mm_struct *mm, unsigned long addr,
68 pte_t *ptep, int nodat)
69{
70 unsigned long opt, asce;
71
72 if (MACHINE_HAS_TLB_GUEST) {
73 opt = 0;
74 asce = READ_ONCE(mm->context.gmap_asce);
75 if (asce == 0UL || nodat)
76 opt |= IPTE_NODAT;
77 if (asce != -1UL) {
78 asce = asce ? : mm->context.asce;
79 opt |= IPTE_GUEST_ASCE;
80 }
81 __ptep_ipte(addr, ptep, opt, asce, IPTE_GLOBAL);
82 } else {
83 __ptep_ipte(addr, ptep, 0, 0, IPTE_GLOBAL);
84 }
85}
86
87static inline pte_t ptep_flush_direct(struct mm_struct *mm,
88 unsigned long addr, pte_t *ptep,
89 int nodat)
90{
91 pte_t old;
92
93 old = *ptep;
94 if (unlikely(pte_val(old) & _PAGE_INVALID))
95 return old;
96 atomic_inc(&mm->context.flush_count);
97 if (MACHINE_HAS_TLB_LC &&
98 cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
99 ptep_ipte_local(mm, addr, ptep, nodat);
100 else
101 ptep_ipte_global(mm, addr, ptep, nodat);
102 atomic_dec(&mm->context.flush_count);
103 return old;
104}
105
106static inline pte_t ptep_flush_lazy(struct mm_struct *mm,
107 unsigned long addr, pte_t *ptep,
108 int nodat)
109{
110 pte_t old;
111
112 old = *ptep;
113 if (unlikely(pte_val(old) & _PAGE_INVALID))
114 return old;
115 atomic_inc(&mm->context.flush_count);
116 if (cpumask_equal(&mm->context.cpu_attach_mask,
117 cpumask_of(smp_processor_id()))) {
118 set_pte(ptep, set_pte_bit(*ptep, __pgprot(_PAGE_INVALID)));
119 mm->context.flush_mm = 1;
120 } else
121 ptep_ipte_global(mm, addr, ptep, nodat);
122 atomic_dec(&mm->context.flush_count);
123 return old;
124}
125
126static inline pgste_t pgste_get_lock(pte_t *ptep)
127{
128 unsigned long new = 0;
129#ifdef CONFIG_PGSTE
130 unsigned long old;
131
132 asm(
133 " lg %0,%2\n"
134 "0: lgr %1,%0\n"
135 " nihh %0,0xff7f\n" /* clear PCL bit in old */
136 " oihh %1,0x0080\n" /* set PCL bit in new */
137 " csg %0,%1,%2\n"
138 " jl 0b\n"
139 : "=&d" (old), "=&d" (new), "=Q" (ptep[PTRS_PER_PTE])
140 : "Q" (ptep[PTRS_PER_PTE]) : "cc", "memory");
141#endif
142 return __pgste(new);
143}
144
145static inline void pgste_set_unlock(pte_t *ptep, pgste_t pgste)
146{
147#ifdef CONFIG_PGSTE
148 asm(
149 " nihh %1,0xff7f\n" /* clear PCL bit */
150 " stg %1,%0\n"
151 : "=Q" (ptep[PTRS_PER_PTE])
152 : "d" (pgste_val(pgste)), "Q" (ptep[PTRS_PER_PTE])
153 : "cc", "memory");
154#endif
155}
156
157static inline pgste_t pgste_get(pte_t *ptep)
158{
159 unsigned long pgste = 0;
160#ifdef CONFIG_PGSTE
161 pgste = *(unsigned long *)(ptep + PTRS_PER_PTE);
162#endif
163 return __pgste(pgste);
164}
165
166static inline void pgste_set(pte_t *ptep, pgste_t pgste)
167{
168#ifdef CONFIG_PGSTE
169 *(pgste_t *)(ptep + PTRS_PER_PTE) = pgste;
170#endif
171}
172
173static inline pgste_t pgste_update_all(pte_t pte, pgste_t pgste,
174 struct mm_struct *mm)
175{
176#ifdef CONFIG_PGSTE
177 unsigned long address, bits, skey;
178
179 if (!mm_uses_skeys(mm) || pte_val(pte) & _PAGE_INVALID)
180 return pgste;
181 address = pte_val(pte) & PAGE_MASK;
182 skey = (unsigned long) page_get_storage_key(address);
183 bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
184 /* Transfer page changed & referenced bit to guest bits in pgste */
185 pgste_val(pgste) |= bits << 48; /* GR bit & GC bit */
186 /* Copy page access key and fetch protection bit to pgste */
187 pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
188 pgste_val(pgste) |= (skey & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
189#endif
190 return pgste;
191
192}
193
194static inline void pgste_set_key(pte_t *ptep, pgste_t pgste, pte_t entry,
195 struct mm_struct *mm)
196{
197#ifdef CONFIG_PGSTE
198 unsigned long address;
199 unsigned long nkey;
200
201 if (!mm_uses_skeys(mm) || pte_val(entry) & _PAGE_INVALID)
202 return;
203 VM_BUG_ON(!(pte_val(*ptep) & _PAGE_INVALID));
204 address = pte_val(entry) & PAGE_MASK;
205 /*
206 * Set page access key and fetch protection bit from pgste.
207 * The guest C/R information is still in the PGSTE, set real
208 * key C/R to 0.
209 */
210 nkey = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
211 nkey |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
212 page_set_storage_key(address, nkey, 0);
213#endif
214}
215
216static inline pgste_t pgste_set_pte(pte_t *ptep, pgste_t pgste, pte_t entry)
217{
218#ifdef CONFIG_PGSTE
219 if ((pte_val(entry) & _PAGE_PRESENT) &&
220 (pte_val(entry) & _PAGE_WRITE) &&
221 !(pte_val(entry) & _PAGE_INVALID)) {
222 if (!MACHINE_HAS_ESOP) {
223 /*
224 * Without enhanced suppression-on-protection force
225 * the dirty bit on for all writable ptes.
226 */
227 entry = set_pte_bit(entry, __pgprot(_PAGE_DIRTY));
228 entry = clear_pte_bit(entry, __pgprot(_PAGE_PROTECT));
229 }
230 if (!(pte_val(entry) & _PAGE_PROTECT))
231 /* This pte allows write access, set user-dirty */
232 pgste_val(pgste) |= PGSTE_UC_BIT;
233 }
234#endif
235 set_pte(ptep, entry);
236 return pgste;
237}
238
239static inline pgste_t pgste_pte_notify(struct mm_struct *mm,
240 unsigned long addr,
241 pte_t *ptep, pgste_t pgste)
242{
243#ifdef CONFIG_PGSTE
244 unsigned long bits;
245
246 bits = pgste_val(pgste) & (PGSTE_IN_BIT | PGSTE_VSIE_BIT);
247 if (bits) {
248 pgste_val(pgste) ^= bits;
249 ptep_notify(mm, addr, ptep, bits);
250 }
251#endif
252 return pgste;
253}
254
255static inline pgste_t ptep_xchg_start(struct mm_struct *mm,
256 unsigned long addr, pte_t *ptep)
257{
258 pgste_t pgste = __pgste(0);
259
260 if (mm_has_pgste(mm)) {
261 pgste = pgste_get_lock(ptep);
262 pgste = pgste_pte_notify(mm, addr, ptep, pgste);
263 }
264 return pgste;
265}
266
267static inline pte_t ptep_xchg_commit(struct mm_struct *mm,
268 unsigned long addr, pte_t *ptep,
269 pgste_t pgste, pte_t old, pte_t new)
270{
271 if (mm_has_pgste(mm)) {
272 if (pte_val(old) & _PAGE_INVALID)
273 pgste_set_key(ptep, pgste, new, mm);
274 if (pte_val(new) & _PAGE_INVALID) {
275 pgste = pgste_update_all(old, pgste, mm);
276 if ((pgste_val(pgste) & _PGSTE_GPS_USAGE_MASK) ==
277 _PGSTE_GPS_USAGE_UNUSED)
278 old = set_pte_bit(old, __pgprot(_PAGE_UNUSED));
279 }
280 pgste = pgste_set_pte(ptep, pgste, new);
281 pgste_set_unlock(ptep, pgste);
282 } else {
283 set_pte(ptep, new);
284 }
285 return old;
286}
287
288pte_t ptep_xchg_direct(struct mm_struct *mm, unsigned long addr,
289 pte_t *ptep, pte_t new)
290{
291 pgste_t pgste;
292 pte_t old;
293 int nodat;
294
295 preempt_disable();
296 pgste = ptep_xchg_start(mm, addr, ptep);
297 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
298 old = ptep_flush_direct(mm, addr, ptep, nodat);
299 old = ptep_xchg_commit(mm, addr, ptep, pgste, old, new);
300 preempt_enable();
301 return old;
302}
303EXPORT_SYMBOL(ptep_xchg_direct);
304
305pte_t ptep_xchg_lazy(struct mm_struct *mm, unsigned long addr,
306 pte_t *ptep, pte_t new)
307{
308 pgste_t pgste;
309 pte_t old;
310 int nodat;
311
312 preempt_disable();
313 pgste = ptep_xchg_start(mm, addr, ptep);
314 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
315 old = ptep_flush_lazy(mm, addr, ptep, nodat);
316 old = ptep_xchg_commit(mm, addr, ptep, pgste, old, new);
317 preempt_enable();
318 return old;
319}
320EXPORT_SYMBOL(ptep_xchg_lazy);
321
322pte_t ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr,
323 pte_t *ptep)
324{
325 pgste_t pgste;
326 pte_t old;
327 int nodat;
328 struct mm_struct *mm = vma->vm_mm;
329
330 preempt_disable();
331 pgste = ptep_xchg_start(mm, addr, ptep);
332 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
333 old = ptep_flush_lazy(mm, addr, ptep, nodat);
334 if (mm_has_pgste(mm)) {
335 pgste = pgste_update_all(old, pgste, mm);
336 pgste_set(ptep, pgste);
337 }
338 return old;
339}
340
341void ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr,
342 pte_t *ptep, pte_t old_pte, pte_t pte)
343{
344 pgste_t pgste;
345 struct mm_struct *mm = vma->vm_mm;
346
347 if (!MACHINE_HAS_NX)
348 pte = clear_pte_bit(pte, __pgprot(_PAGE_NOEXEC));
349 if (mm_has_pgste(mm)) {
350 pgste = pgste_get(ptep);
351 pgste_set_key(ptep, pgste, pte, mm);
352 pgste = pgste_set_pte(ptep, pgste, pte);
353 pgste_set_unlock(ptep, pgste);
354 } else {
355 set_pte(ptep, pte);
356 }
357 preempt_enable();
358}
359
360static inline void pmdp_idte_local(struct mm_struct *mm,
361 unsigned long addr, pmd_t *pmdp)
362{
363 if (MACHINE_HAS_TLB_GUEST)
364 __pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
365 mm->context.asce, IDTE_LOCAL);
366 else
367 __pmdp_idte(addr, pmdp, 0, 0, IDTE_LOCAL);
368 if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
369 gmap_pmdp_idte_local(mm, addr);
370}
371
372static inline void pmdp_idte_global(struct mm_struct *mm,
373 unsigned long addr, pmd_t *pmdp)
374{
375 if (MACHINE_HAS_TLB_GUEST) {
376 __pmdp_idte(addr, pmdp, IDTE_NODAT | IDTE_GUEST_ASCE,
377 mm->context.asce, IDTE_GLOBAL);
378 if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
379 gmap_pmdp_idte_global(mm, addr);
380 } else if (MACHINE_HAS_IDTE) {
381 __pmdp_idte(addr, pmdp, 0, 0, IDTE_GLOBAL);
382 if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
383 gmap_pmdp_idte_global(mm, addr);
384 } else {
385 __pmdp_csp(pmdp);
386 if (mm_has_pgste(mm) && mm->context.allow_gmap_hpage_1m)
387 gmap_pmdp_csp(mm, addr);
388 }
389}
390
391static inline pmd_t pmdp_flush_direct(struct mm_struct *mm,
392 unsigned long addr, pmd_t *pmdp)
393{
394 pmd_t old;
395
396 old = *pmdp;
397 if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
398 return old;
399 atomic_inc(&mm->context.flush_count);
400 if (MACHINE_HAS_TLB_LC &&
401 cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
402 pmdp_idte_local(mm, addr, pmdp);
403 else
404 pmdp_idte_global(mm, addr, pmdp);
405 atomic_dec(&mm->context.flush_count);
406 return old;
407}
408
409static inline pmd_t pmdp_flush_lazy(struct mm_struct *mm,
410 unsigned long addr, pmd_t *pmdp)
411{
412 pmd_t old;
413
414 old = *pmdp;
415 if (pmd_val(old) & _SEGMENT_ENTRY_INVALID)
416 return old;
417 atomic_inc(&mm->context.flush_count);
418 if (cpumask_equal(&mm->context.cpu_attach_mask,
419 cpumask_of(smp_processor_id()))) {
420 set_pmd(pmdp, set_pmd_bit(*pmdp, __pgprot(_SEGMENT_ENTRY_INVALID)));
421 mm->context.flush_mm = 1;
422 if (mm_has_pgste(mm))
423 gmap_pmdp_invalidate(mm, addr);
424 } else {
425 pmdp_idte_global(mm, addr, pmdp);
426 }
427 atomic_dec(&mm->context.flush_count);
428 return old;
429}
430
431#ifdef CONFIG_PGSTE
432static int pmd_lookup(struct mm_struct *mm, unsigned long addr, pmd_t **pmdp)
433{
434 struct vm_area_struct *vma;
435 pgd_t *pgd;
436 p4d_t *p4d;
437 pud_t *pud;
438
439 /* We need a valid VMA, otherwise this is clearly a fault. */
440 vma = vma_lookup(mm, addr);
441 if (!vma)
442 return -EFAULT;
443
444 pgd = pgd_offset(mm, addr);
445 if (!pgd_present(*pgd))
446 return -ENOENT;
447
448 p4d = p4d_offset(pgd, addr);
449 if (!p4d_present(*p4d))
450 return -ENOENT;
451
452 pud = pud_offset(p4d, addr);
453 if (!pud_present(*pud))
454 return -ENOENT;
455
456 /* Large PUDs are not supported yet. */
457 if (pud_large(*pud))
458 return -EFAULT;
459
460 *pmdp = pmd_offset(pud, addr);
461 return 0;
462}
463#endif
464
465pmd_t pmdp_xchg_direct(struct mm_struct *mm, unsigned long addr,
466 pmd_t *pmdp, pmd_t new)
467{
468 pmd_t old;
469
470 preempt_disable();
471 old = pmdp_flush_direct(mm, addr, pmdp);
472 set_pmd(pmdp, new);
473 preempt_enable();
474 return old;
475}
476EXPORT_SYMBOL(pmdp_xchg_direct);
477
478pmd_t pmdp_xchg_lazy(struct mm_struct *mm, unsigned long addr,
479 pmd_t *pmdp, pmd_t new)
480{
481 pmd_t old;
482
483 preempt_disable();
484 old = pmdp_flush_lazy(mm, addr, pmdp);
485 set_pmd(pmdp, new);
486 preempt_enable();
487 return old;
488}
489EXPORT_SYMBOL(pmdp_xchg_lazy);
490
491static inline void pudp_idte_local(struct mm_struct *mm,
492 unsigned long addr, pud_t *pudp)
493{
494 if (MACHINE_HAS_TLB_GUEST)
495 __pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
496 mm->context.asce, IDTE_LOCAL);
497 else
498 __pudp_idte(addr, pudp, 0, 0, IDTE_LOCAL);
499}
500
501static inline void pudp_idte_global(struct mm_struct *mm,
502 unsigned long addr, pud_t *pudp)
503{
504 if (MACHINE_HAS_TLB_GUEST)
505 __pudp_idte(addr, pudp, IDTE_NODAT | IDTE_GUEST_ASCE,
506 mm->context.asce, IDTE_GLOBAL);
507 else if (MACHINE_HAS_IDTE)
508 __pudp_idte(addr, pudp, 0, 0, IDTE_GLOBAL);
509 else
510 /*
511 * Invalid bit position is the same for pmd and pud, so we can
512 * re-use _pmd_csp() here
513 */
514 __pmdp_csp((pmd_t *) pudp);
515}
516
517static inline pud_t pudp_flush_direct(struct mm_struct *mm,
518 unsigned long addr, pud_t *pudp)
519{
520 pud_t old;
521
522 old = *pudp;
523 if (pud_val(old) & _REGION_ENTRY_INVALID)
524 return old;
525 atomic_inc(&mm->context.flush_count);
526 if (MACHINE_HAS_TLB_LC &&
527 cpumask_equal(mm_cpumask(mm), cpumask_of(smp_processor_id())))
528 pudp_idte_local(mm, addr, pudp);
529 else
530 pudp_idte_global(mm, addr, pudp);
531 atomic_dec(&mm->context.flush_count);
532 return old;
533}
534
535pud_t pudp_xchg_direct(struct mm_struct *mm, unsigned long addr,
536 pud_t *pudp, pud_t new)
537{
538 pud_t old;
539
540 preempt_disable();
541 old = pudp_flush_direct(mm, addr, pudp);
542 set_pud(pudp, new);
543 preempt_enable();
544 return old;
545}
546EXPORT_SYMBOL(pudp_xchg_direct);
547
548#ifdef CONFIG_TRANSPARENT_HUGEPAGE
549void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
550 pgtable_t pgtable)
551{
552 struct list_head *lh = (struct list_head *) pgtable;
553
554 assert_spin_locked(pmd_lockptr(mm, pmdp));
555
556 /* FIFO */
557 if (!pmd_huge_pte(mm, pmdp))
558 INIT_LIST_HEAD(lh);
559 else
560 list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp));
561 pmd_huge_pte(mm, pmdp) = pgtable;
562}
563
564pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
565{
566 struct list_head *lh;
567 pgtable_t pgtable;
568 pte_t *ptep;
569
570 assert_spin_locked(pmd_lockptr(mm, pmdp));
571
572 /* FIFO */
573 pgtable = pmd_huge_pte(mm, pmdp);
574 lh = (struct list_head *) pgtable;
575 if (list_empty(lh))
576 pmd_huge_pte(mm, pmdp) = NULL;
577 else {
578 pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next;
579 list_del(lh);
580 }
581 ptep = (pte_t *) pgtable;
582 set_pte(ptep, __pte(_PAGE_INVALID));
583 ptep++;
584 set_pte(ptep, __pte(_PAGE_INVALID));
585 return pgtable;
586}
587#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
588
589#ifdef CONFIG_PGSTE
590void ptep_set_pte_at(struct mm_struct *mm, unsigned long addr,
591 pte_t *ptep, pte_t entry)
592{
593 pgste_t pgste;
594
595 /* the mm_has_pgste() check is done in set_pte_at() */
596 preempt_disable();
597 pgste = pgste_get_lock(ptep);
598 pgste_val(pgste) &= ~_PGSTE_GPS_ZERO;
599 pgste_set_key(ptep, pgste, entry, mm);
600 pgste = pgste_set_pte(ptep, pgste, entry);
601 pgste_set_unlock(ptep, pgste);
602 preempt_enable();
603}
604
605void ptep_set_notify(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
606{
607 pgste_t pgste;
608
609 preempt_disable();
610 pgste = pgste_get_lock(ptep);
611 pgste_val(pgste) |= PGSTE_IN_BIT;
612 pgste_set_unlock(ptep, pgste);
613 preempt_enable();
614}
615
616/**
617 * ptep_force_prot - change access rights of a locked pte
618 * @mm: pointer to the process mm_struct
619 * @addr: virtual address in the guest address space
620 * @ptep: pointer to the page table entry
621 * @prot: indicates guest access rights: PROT_NONE, PROT_READ or PROT_WRITE
622 * @bit: pgste bit to set (e.g. for notification)
623 *
624 * Returns 0 if the access rights were changed and -EAGAIN if the current
625 * and requested access rights are incompatible.
626 */
627int ptep_force_prot(struct mm_struct *mm, unsigned long addr,
628 pte_t *ptep, int prot, unsigned long bit)
629{
630 pte_t entry;
631 pgste_t pgste;
632 int pte_i, pte_p, nodat;
633
634 pgste = pgste_get_lock(ptep);
635 entry = *ptep;
636 /* Check pte entry after all locks have been acquired */
637 pte_i = pte_val(entry) & _PAGE_INVALID;
638 pte_p = pte_val(entry) & _PAGE_PROTECT;
639 if ((pte_i && (prot != PROT_NONE)) ||
640 (pte_p && (prot & PROT_WRITE))) {
641 pgste_set_unlock(ptep, pgste);
642 return -EAGAIN;
643 }
644 /* Change access rights and set pgste bit */
645 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
646 if (prot == PROT_NONE && !pte_i) {
647 ptep_flush_direct(mm, addr, ptep, nodat);
648 pgste = pgste_update_all(entry, pgste, mm);
649 entry = set_pte_bit(entry, __pgprot(_PAGE_INVALID));
650 }
651 if (prot == PROT_READ && !pte_p) {
652 ptep_flush_direct(mm, addr, ptep, nodat);
653 entry = clear_pte_bit(entry, __pgprot(_PAGE_INVALID));
654 entry = set_pte_bit(entry, __pgprot(_PAGE_PROTECT));
655 }
656 pgste_val(pgste) |= bit;
657 pgste = pgste_set_pte(ptep, pgste, entry);
658 pgste_set_unlock(ptep, pgste);
659 return 0;
660}
661
662int ptep_shadow_pte(struct mm_struct *mm, unsigned long saddr,
663 pte_t *sptep, pte_t *tptep, pte_t pte)
664{
665 pgste_t spgste, tpgste;
666 pte_t spte, tpte;
667 int rc = -EAGAIN;
668
669 if (!(pte_val(*tptep) & _PAGE_INVALID))
670 return 0; /* already shadowed */
671 spgste = pgste_get_lock(sptep);
672 spte = *sptep;
673 if (!(pte_val(spte) & _PAGE_INVALID) &&
674 !((pte_val(spte) & _PAGE_PROTECT) &&
675 !(pte_val(pte) & _PAGE_PROTECT))) {
676 pgste_val(spgste) |= PGSTE_VSIE_BIT;
677 tpgste = pgste_get_lock(tptep);
678 tpte = __pte((pte_val(spte) & PAGE_MASK) |
679 (pte_val(pte) & _PAGE_PROTECT));
680 /* don't touch the storage key - it belongs to parent pgste */
681 tpgste = pgste_set_pte(tptep, tpgste, tpte);
682 pgste_set_unlock(tptep, tpgste);
683 rc = 1;
684 }
685 pgste_set_unlock(sptep, spgste);
686 return rc;
687}
688
689void ptep_unshadow_pte(struct mm_struct *mm, unsigned long saddr, pte_t *ptep)
690{
691 pgste_t pgste;
692 int nodat;
693
694 pgste = pgste_get_lock(ptep);
695 /* notifier is called by the caller */
696 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
697 ptep_flush_direct(mm, saddr, ptep, nodat);
698 /* don't touch the storage key - it belongs to parent pgste */
699 pgste = pgste_set_pte(ptep, pgste, __pte(_PAGE_INVALID));
700 pgste_set_unlock(ptep, pgste);
701}
702
703static void ptep_zap_swap_entry(struct mm_struct *mm, swp_entry_t entry)
704{
705 if (!non_swap_entry(entry))
706 dec_mm_counter(mm, MM_SWAPENTS);
707 else if (is_migration_entry(entry)) {
708 struct page *page = pfn_swap_entry_to_page(entry);
709
710 dec_mm_counter(mm, mm_counter(page));
711 }
712 free_swap_and_cache(entry);
713}
714
715void ptep_zap_unused(struct mm_struct *mm, unsigned long addr,
716 pte_t *ptep, int reset)
717{
718 unsigned long pgstev;
719 pgste_t pgste;
720 pte_t pte;
721
722 /* Zap unused and logically-zero pages */
723 preempt_disable();
724 pgste = pgste_get_lock(ptep);
725 pgstev = pgste_val(pgste);
726 pte = *ptep;
727 if (!reset && pte_swap(pte) &&
728 ((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED ||
729 (pgstev & _PGSTE_GPS_ZERO))) {
730 ptep_zap_swap_entry(mm, pte_to_swp_entry(pte));
731 pte_clear(mm, addr, ptep);
732 }
733 if (reset)
734 pgste_val(pgste) &= ~_PGSTE_GPS_USAGE_MASK;
735 pgste_set_unlock(ptep, pgste);
736 preempt_enable();
737}
738
739void ptep_zap_key(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
740{
741 unsigned long ptev;
742 pgste_t pgste;
743
744 /* Clear storage key ACC and F, but set R/C */
745 preempt_disable();
746 pgste = pgste_get_lock(ptep);
747 pgste_val(pgste) &= ~(PGSTE_ACC_BITS | PGSTE_FP_BIT);
748 pgste_val(pgste) |= PGSTE_GR_BIT | PGSTE_GC_BIT;
749 ptev = pte_val(*ptep);
750 if (!(ptev & _PAGE_INVALID) && (ptev & _PAGE_WRITE))
751 page_set_storage_key(ptev & PAGE_MASK, PAGE_DEFAULT_KEY, 0);
752 pgste_set_unlock(ptep, pgste);
753 preempt_enable();
754}
755
756/*
757 * Test and reset if a guest page is dirty
758 */
759bool ptep_test_and_clear_uc(struct mm_struct *mm, unsigned long addr,
760 pte_t *ptep)
761{
762 pgste_t pgste;
763 pte_t pte;
764 bool dirty;
765 int nodat;
766
767 pgste = pgste_get_lock(ptep);
768 dirty = !!(pgste_val(pgste) & PGSTE_UC_BIT);
769 pgste_val(pgste) &= ~PGSTE_UC_BIT;
770 pte = *ptep;
771 if (dirty && (pte_val(pte) & _PAGE_PRESENT)) {
772 pgste = pgste_pte_notify(mm, addr, ptep, pgste);
773 nodat = !!(pgste_val(pgste) & _PGSTE_GPS_NODAT);
774 ptep_ipte_global(mm, addr, ptep, nodat);
775 if (MACHINE_HAS_ESOP || !(pte_val(pte) & _PAGE_WRITE))
776 pte = set_pte_bit(pte, __pgprot(_PAGE_PROTECT));
777 else
778 pte = set_pte_bit(pte, __pgprot(_PAGE_INVALID));
779 set_pte(ptep, pte);
780 }
781 pgste_set_unlock(ptep, pgste);
782 return dirty;
783}
784EXPORT_SYMBOL_GPL(ptep_test_and_clear_uc);
785
786int set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
787 unsigned char key, bool nq)
788{
789 unsigned long keyul, paddr;
790 spinlock_t *ptl;
791 pgste_t old, new;
792 pmd_t *pmdp;
793 pte_t *ptep;
794
795 /*
796 * If we don't have a PTE table and if there is no huge page mapped,
797 * we can ignore attempts to set the key to 0, because it already is 0.
798 */
799 switch (pmd_lookup(mm, addr, &pmdp)) {
800 case -ENOENT:
801 return key ? -EFAULT : 0;
802 case 0:
803 break;
804 default:
805 return -EFAULT;
806 }
807
808 ptl = pmd_lock(mm, pmdp);
809 if (!pmd_present(*pmdp)) {
810 spin_unlock(ptl);
811 return key ? -EFAULT : 0;
812 }
813
814 if (pmd_large(*pmdp)) {
815 paddr = pmd_val(*pmdp) & HPAGE_MASK;
816 paddr |= addr & ~HPAGE_MASK;
817 /*
818 * Huge pmds need quiescing operations, they are
819 * always mapped.
820 */
821 page_set_storage_key(paddr, key, 1);
822 spin_unlock(ptl);
823 return 0;
824 }
825 spin_unlock(ptl);
826
827 ptep = pte_offset_map_lock(mm, pmdp, addr, &ptl);
828 new = old = pgste_get_lock(ptep);
829 pgste_val(new) &= ~(PGSTE_GR_BIT | PGSTE_GC_BIT |
830 PGSTE_ACC_BITS | PGSTE_FP_BIT);
831 keyul = (unsigned long) key;
832 pgste_val(new) |= (keyul & (_PAGE_CHANGED | _PAGE_REFERENCED)) << 48;
833 pgste_val(new) |= (keyul & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
834 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
835 unsigned long bits, skey;
836
837 paddr = pte_val(*ptep) & PAGE_MASK;
838 skey = (unsigned long) page_get_storage_key(paddr);
839 bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
840 skey = key & (_PAGE_ACC_BITS | _PAGE_FP_BIT);
841 /* Set storage key ACC and FP */
842 page_set_storage_key(paddr, skey, !nq);
843 /* Merge host changed & referenced into pgste */
844 pgste_val(new) |= bits << 52;
845 }
846 /* changing the guest storage key is considered a change of the page */
847 if ((pgste_val(new) ^ pgste_val(old)) &
848 (PGSTE_ACC_BITS | PGSTE_FP_BIT | PGSTE_GR_BIT | PGSTE_GC_BIT))
849 pgste_val(new) |= PGSTE_UC_BIT;
850
851 pgste_set_unlock(ptep, new);
852 pte_unmap_unlock(ptep, ptl);
853 return 0;
854}
855EXPORT_SYMBOL(set_guest_storage_key);
856
857/*
858 * Conditionally set a guest storage key (handling csske).
859 * oldkey will be updated when either mr or mc is set and a pointer is given.
860 *
861 * Returns 0 if a guests storage key update wasn't necessary, 1 if the guest
862 * storage key was updated and -EFAULT on access errors.
863 */
864int cond_set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
865 unsigned char key, unsigned char *oldkey,
866 bool nq, bool mr, bool mc)
867{
868 unsigned char tmp, mask = _PAGE_ACC_BITS | _PAGE_FP_BIT;
869 int rc;
870
871 /* we can drop the pgste lock between getting and setting the key */
872 if (mr | mc) {
873 rc = get_guest_storage_key(current->mm, addr, &tmp);
874 if (rc)
875 return rc;
876 if (oldkey)
877 *oldkey = tmp;
878 if (!mr)
879 mask |= _PAGE_REFERENCED;
880 if (!mc)
881 mask |= _PAGE_CHANGED;
882 if (!((tmp ^ key) & mask))
883 return 0;
884 }
885 rc = set_guest_storage_key(current->mm, addr, key, nq);
886 return rc < 0 ? rc : 1;
887}
888EXPORT_SYMBOL(cond_set_guest_storage_key);
889
890/*
891 * Reset a guest reference bit (rrbe), returning the reference and changed bit.
892 *
893 * Returns < 0 in case of error, otherwise the cc to be reported to the guest.
894 */
895int reset_guest_reference_bit(struct mm_struct *mm, unsigned long addr)
896{
897 spinlock_t *ptl;
898 unsigned long paddr;
899 pgste_t old, new;
900 pmd_t *pmdp;
901 pte_t *ptep;
902 int cc = 0;
903
904 /*
905 * If we don't have a PTE table and if there is no huge page mapped,
906 * the storage key is 0 and there is nothing for us to do.
907 */
908 switch (pmd_lookup(mm, addr, &pmdp)) {
909 case -ENOENT:
910 return 0;
911 case 0:
912 break;
913 default:
914 return -EFAULT;
915 }
916
917 ptl = pmd_lock(mm, pmdp);
918 if (!pmd_present(*pmdp)) {
919 spin_unlock(ptl);
920 return 0;
921 }
922
923 if (pmd_large(*pmdp)) {
924 paddr = pmd_val(*pmdp) & HPAGE_MASK;
925 paddr |= addr & ~HPAGE_MASK;
926 cc = page_reset_referenced(paddr);
927 spin_unlock(ptl);
928 return cc;
929 }
930 spin_unlock(ptl);
931
932 ptep = pte_offset_map_lock(mm, pmdp, addr, &ptl);
933 new = old = pgste_get_lock(ptep);
934 /* Reset guest reference bit only */
935 pgste_val(new) &= ~PGSTE_GR_BIT;
936
937 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
938 paddr = pte_val(*ptep) & PAGE_MASK;
939 cc = page_reset_referenced(paddr);
940 /* Merge real referenced bit into host-set */
941 pgste_val(new) |= ((unsigned long) cc << 53) & PGSTE_HR_BIT;
942 }
943 /* Reflect guest's logical view, not physical */
944 cc |= (pgste_val(old) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 49;
945 /* Changing the guest storage key is considered a change of the page */
946 if ((pgste_val(new) ^ pgste_val(old)) & PGSTE_GR_BIT)
947 pgste_val(new) |= PGSTE_UC_BIT;
948
949 pgste_set_unlock(ptep, new);
950 pte_unmap_unlock(ptep, ptl);
951 return cc;
952}
953EXPORT_SYMBOL(reset_guest_reference_bit);
954
955int get_guest_storage_key(struct mm_struct *mm, unsigned long addr,
956 unsigned char *key)
957{
958 unsigned long paddr;
959 spinlock_t *ptl;
960 pgste_t pgste;
961 pmd_t *pmdp;
962 pte_t *ptep;
963
964 /*
965 * If we don't have a PTE table and if there is no huge page mapped,
966 * the storage key is 0.
967 */
968 *key = 0;
969
970 switch (pmd_lookup(mm, addr, &pmdp)) {
971 case -ENOENT:
972 return 0;
973 case 0:
974 break;
975 default:
976 return -EFAULT;
977 }
978
979 ptl = pmd_lock(mm, pmdp);
980 if (!pmd_present(*pmdp)) {
981 spin_unlock(ptl);
982 return 0;
983 }
984
985 if (pmd_large(*pmdp)) {
986 paddr = pmd_val(*pmdp) & HPAGE_MASK;
987 paddr |= addr & ~HPAGE_MASK;
988 *key = page_get_storage_key(paddr);
989 spin_unlock(ptl);
990 return 0;
991 }
992 spin_unlock(ptl);
993
994 ptep = pte_offset_map_lock(mm, pmdp, addr, &ptl);
995 pgste = pgste_get_lock(ptep);
996 *key = (pgste_val(pgste) & (PGSTE_ACC_BITS | PGSTE_FP_BIT)) >> 56;
997 paddr = pte_val(*ptep) & PAGE_MASK;
998 if (!(pte_val(*ptep) & _PAGE_INVALID))
999 *key = page_get_storage_key(paddr);
1000 /* Reflect guest's logical view, not physical */
1001 *key |= (pgste_val(pgste) & (PGSTE_GR_BIT | PGSTE_GC_BIT)) >> 48;
1002 pgste_set_unlock(ptep, pgste);
1003 pte_unmap_unlock(ptep, ptl);
1004 return 0;
1005}
1006EXPORT_SYMBOL(get_guest_storage_key);
1007
1008/**
1009 * pgste_perform_essa - perform ESSA actions on the PGSTE.
1010 * @mm: the memory context. It must have PGSTEs, no check is performed here!
1011 * @hva: the host virtual address of the page whose PGSTE is to be processed
1012 * @orc: the specific action to perform, see the ESSA_SET_* macros.
1013 * @oldpte: the PTE will be saved there if the pointer is not NULL.
1014 * @oldpgste: the old PGSTE will be saved there if the pointer is not NULL.
1015 *
1016 * Return: 1 if the page is to be added to the CBRL, otherwise 0,
1017 * or < 0 in case of error. -EINVAL is returned for invalid values
1018 * of orc, -EFAULT for invalid addresses.
1019 */
1020int pgste_perform_essa(struct mm_struct *mm, unsigned long hva, int orc,
1021 unsigned long *oldpte, unsigned long *oldpgste)
1022{
1023 struct vm_area_struct *vma;
1024 unsigned long pgstev;
1025 spinlock_t *ptl;
1026 pgste_t pgste;
1027 pte_t *ptep;
1028 int res = 0;
1029
1030 WARN_ON_ONCE(orc > ESSA_MAX);
1031 if (unlikely(orc > ESSA_MAX))
1032 return -EINVAL;
1033
1034 vma = vma_lookup(mm, hva);
1035 if (!vma || is_vm_hugetlb_page(vma))
1036 return -EFAULT;
1037 ptep = get_locked_pte(mm, hva, &ptl);
1038 if (unlikely(!ptep))
1039 return -EFAULT;
1040 pgste = pgste_get_lock(ptep);
1041 pgstev = pgste_val(pgste);
1042 if (oldpte)
1043 *oldpte = pte_val(*ptep);
1044 if (oldpgste)
1045 *oldpgste = pgstev;
1046
1047 switch (orc) {
1048 case ESSA_GET_STATE:
1049 break;
1050 case ESSA_SET_STABLE:
1051 pgstev &= ~(_PGSTE_GPS_USAGE_MASK | _PGSTE_GPS_NODAT);
1052 pgstev |= _PGSTE_GPS_USAGE_STABLE;
1053 break;
1054 case ESSA_SET_UNUSED:
1055 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1056 pgstev |= _PGSTE_GPS_USAGE_UNUSED;
1057 if (pte_val(*ptep) & _PAGE_INVALID)
1058 res = 1;
1059 break;
1060 case ESSA_SET_VOLATILE:
1061 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1062 pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1063 if (pte_val(*ptep) & _PAGE_INVALID)
1064 res = 1;
1065 break;
1066 case ESSA_SET_POT_VOLATILE:
1067 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1068 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1069 pgstev |= _PGSTE_GPS_USAGE_POT_VOLATILE;
1070 break;
1071 }
1072 if (pgstev & _PGSTE_GPS_ZERO) {
1073 pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1074 break;
1075 }
1076 if (!(pgstev & PGSTE_GC_BIT)) {
1077 pgstev |= _PGSTE_GPS_USAGE_VOLATILE;
1078 res = 1;
1079 break;
1080 }
1081 break;
1082 case ESSA_SET_STABLE_RESIDENT:
1083 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1084 pgstev |= _PGSTE_GPS_USAGE_STABLE;
1085 /*
1086 * Since the resident state can go away any time after this
1087 * call, we will not make this page resident. We can revisit
1088 * this decision if a guest will ever start using this.
1089 */
1090 break;
1091 case ESSA_SET_STABLE_IF_RESIDENT:
1092 if (!(pte_val(*ptep) & _PAGE_INVALID)) {
1093 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1094 pgstev |= _PGSTE_GPS_USAGE_STABLE;
1095 }
1096 break;
1097 case ESSA_SET_STABLE_NODAT:
1098 pgstev &= ~_PGSTE_GPS_USAGE_MASK;
1099 pgstev |= _PGSTE_GPS_USAGE_STABLE | _PGSTE_GPS_NODAT;
1100 break;
1101 default:
1102 /* we should never get here! */
1103 break;
1104 }
1105 /* If we are discarding a page, set it to logical zero */
1106 if (res)
1107 pgstev |= _PGSTE_GPS_ZERO;
1108
1109 pgste_val(pgste) = pgstev;
1110 pgste_set_unlock(ptep, pgste);
1111 pte_unmap_unlock(ptep, ptl);
1112 return res;
1113}
1114EXPORT_SYMBOL(pgste_perform_essa);
1115
1116/**
1117 * set_pgste_bits - set specific PGSTE bits.
1118 * @mm: the memory context. It must have PGSTEs, no check is performed here!
1119 * @hva: the host virtual address of the page whose PGSTE is to be processed
1120 * @bits: a bitmask representing the bits that will be touched
1121 * @value: the values of the bits to be written. Only the bits in the mask
1122 * will be written.
1123 *
1124 * Return: 0 on success, < 0 in case of error.
1125 */
1126int set_pgste_bits(struct mm_struct *mm, unsigned long hva,
1127 unsigned long bits, unsigned long value)
1128{
1129 struct vm_area_struct *vma;
1130 spinlock_t *ptl;
1131 pgste_t new;
1132 pte_t *ptep;
1133
1134 vma = vma_lookup(mm, hva);
1135 if (!vma || is_vm_hugetlb_page(vma))
1136 return -EFAULT;
1137 ptep = get_locked_pte(mm, hva, &ptl);
1138 if (unlikely(!ptep))
1139 return -EFAULT;
1140 new = pgste_get_lock(ptep);
1141
1142 pgste_val(new) &= ~bits;
1143 pgste_val(new) |= value & bits;
1144
1145 pgste_set_unlock(ptep, new);
1146 pte_unmap_unlock(ptep, ptl);
1147 return 0;
1148}
1149EXPORT_SYMBOL(set_pgste_bits);
1150
1151/**
1152 * get_pgste - get the current PGSTE for the given address.
1153 * @mm: the memory context. It must have PGSTEs, no check is performed here!
1154 * @hva: the host virtual address of the page whose PGSTE is to be processed
1155 * @pgstep: will be written with the current PGSTE for the given address.
1156 *
1157 * Return: 0 on success, < 0 in case of error.
1158 */
1159int get_pgste(struct mm_struct *mm, unsigned long hva, unsigned long *pgstep)
1160{
1161 struct vm_area_struct *vma;
1162 spinlock_t *ptl;
1163 pte_t *ptep;
1164
1165 vma = vma_lookup(mm, hva);
1166 if (!vma || is_vm_hugetlb_page(vma))
1167 return -EFAULT;
1168 ptep = get_locked_pte(mm, hva, &ptl);
1169 if (unlikely(!ptep))
1170 return -EFAULT;
1171 *pgstep = pgste_val(pgste_get(ptep));
1172 pte_unmap_unlock(ptep, ptl);
1173 return 0;
1174}
1175EXPORT_SYMBOL(get_pgste);
1176#endif
1/*
2 * Copyright IBM Corp. 2007,2011
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/pgtable.h>
22#include <asm/pgalloc.h>
23#include <asm/tlb.h>
24#include <asm/tlbflush.h>
25#include <asm/mmu_context.h>
26
27#ifndef CONFIG_64BIT
28#define ALLOC_ORDER 1
29#define FRAG_MASK 0x0f
30#else
31#define ALLOC_ORDER 2
32#define FRAG_MASK 0x03
33#endif
34
35
36unsigned long *crst_table_alloc(struct mm_struct *mm)
37{
38 struct page *page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
39
40 if (!page)
41 return NULL;
42 return (unsigned long *) page_to_phys(page);
43}
44
45void crst_table_free(struct mm_struct *mm, unsigned long *table)
46{
47 free_pages((unsigned long) table, ALLOC_ORDER);
48}
49
50#ifdef CONFIG_64BIT
51int crst_table_upgrade(struct mm_struct *mm, unsigned long limit)
52{
53 unsigned long *table, *pgd;
54 unsigned long entry;
55
56 BUG_ON(limit > (1UL << 53));
57repeat:
58 table = crst_table_alloc(mm);
59 if (!table)
60 return -ENOMEM;
61 spin_lock_bh(&mm->page_table_lock);
62 if (mm->context.asce_limit < limit) {
63 pgd = (unsigned long *) mm->pgd;
64 if (mm->context.asce_limit <= (1UL << 31)) {
65 entry = _REGION3_ENTRY_EMPTY;
66 mm->context.asce_limit = 1UL << 42;
67 mm->context.asce_bits = _ASCE_TABLE_LENGTH |
68 _ASCE_USER_BITS |
69 _ASCE_TYPE_REGION3;
70 } else {
71 entry = _REGION2_ENTRY_EMPTY;
72 mm->context.asce_limit = 1UL << 53;
73 mm->context.asce_bits = _ASCE_TABLE_LENGTH |
74 _ASCE_USER_BITS |
75 _ASCE_TYPE_REGION2;
76 }
77 crst_table_init(table, entry);
78 pgd_populate(mm, (pgd_t *) table, (pud_t *) pgd);
79 mm->pgd = (pgd_t *) table;
80 mm->task_size = mm->context.asce_limit;
81 table = NULL;
82 }
83 spin_unlock_bh(&mm->page_table_lock);
84 if (table)
85 crst_table_free(mm, table);
86 if (mm->context.asce_limit < limit)
87 goto repeat;
88 return 0;
89}
90
91void crst_table_downgrade(struct mm_struct *mm, unsigned long limit)
92{
93 pgd_t *pgd;
94
95 while (mm->context.asce_limit > limit) {
96 pgd = mm->pgd;
97 switch (pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) {
98 case _REGION_ENTRY_TYPE_R2:
99 mm->context.asce_limit = 1UL << 42;
100 mm->context.asce_bits = _ASCE_TABLE_LENGTH |
101 _ASCE_USER_BITS |
102 _ASCE_TYPE_REGION3;
103 break;
104 case _REGION_ENTRY_TYPE_R3:
105 mm->context.asce_limit = 1UL << 31;
106 mm->context.asce_bits = _ASCE_TABLE_LENGTH |
107 _ASCE_USER_BITS |
108 _ASCE_TYPE_SEGMENT;
109 break;
110 default:
111 BUG();
112 }
113 mm->pgd = (pgd_t *) (pgd_val(*pgd) & _REGION_ENTRY_ORIGIN);
114 mm->task_size = mm->context.asce_limit;
115 crst_table_free(mm, (unsigned long *) pgd);
116 }
117}
118#endif
119
120#ifdef CONFIG_PGSTE
121
122/**
123 * gmap_alloc - allocate a guest address space
124 * @mm: pointer to the parent mm_struct
125 *
126 * Returns a guest address space structure.
127 */
128struct gmap *gmap_alloc(struct mm_struct *mm)
129{
130 struct gmap *gmap;
131 struct page *page;
132 unsigned long *table;
133
134 gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL);
135 if (!gmap)
136 goto out;
137 INIT_LIST_HEAD(&gmap->crst_list);
138 gmap->mm = mm;
139 page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
140 if (!page)
141 goto out_free;
142 list_add(&page->lru, &gmap->crst_list);
143 table = (unsigned long *) page_to_phys(page);
144 crst_table_init(table, _REGION1_ENTRY_EMPTY);
145 gmap->table = table;
146 gmap->asce = _ASCE_TYPE_REGION1 | _ASCE_TABLE_LENGTH |
147 _ASCE_USER_BITS | __pa(table);
148 list_add(&gmap->list, &mm->context.gmap_list);
149 return gmap;
150
151out_free:
152 kfree(gmap);
153out:
154 return NULL;
155}
156EXPORT_SYMBOL_GPL(gmap_alloc);
157
158static int gmap_unlink_segment(struct gmap *gmap, unsigned long *table)
159{
160 struct gmap_pgtable *mp;
161 struct gmap_rmap *rmap;
162 struct page *page;
163
164 if (*table & _SEGMENT_ENTRY_INV)
165 return 0;
166 page = pfn_to_page(*table >> PAGE_SHIFT);
167 mp = (struct gmap_pgtable *) page->index;
168 list_for_each_entry(rmap, &mp->mapper, list) {
169 if (rmap->entry != table)
170 continue;
171 list_del(&rmap->list);
172 kfree(rmap);
173 break;
174 }
175 *table = _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO | mp->vmaddr;
176 return 1;
177}
178
179static void gmap_flush_tlb(struct gmap *gmap)
180{
181 if (MACHINE_HAS_IDTE)
182 __tlb_flush_idte((unsigned long) gmap->table |
183 _ASCE_TYPE_REGION1);
184 else
185 __tlb_flush_global();
186}
187
188/**
189 * gmap_free - free a guest address space
190 * @gmap: pointer to the guest address space structure
191 */
192void gmap_free(struct gmap *gmap)
193{
194 struct page *page, *next;
195 unsigned long *table;
196 int i;
197
198
199 /* Flush tlb. */
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 /* Free all segment & region tables. */
207 down_read(&gmap->mm->mmap_sem);
208 spin_lock(&gmap->mm->page_table_lock);
209 list_for_each_entry_safe(page, next, &gmap->crst_list, lru) {
210 table = (unsigned long *) page_to_phys(page);
211 if ((*table & _REGION_ENTRY_TYPE_MASK) == 0)
212 /* Remove gmap rmap structures for segment table. */
213 for (i = 0; i < PTRS_PER_PMD; i++, table++)
214 gmap_unlink_segment(gmap, table);
215 __free_pages(page, ALLOC_ORDER);
216 }
217 spin_unlock(&gmap->mm->page_table_lock);
218 up_read(&gmap->mm->mmap_sem);
219 list_del(&gmap->list);
220 kfree(gmap);
221}
222EXPORT_SYMBOL_GPL(gmap_free);
223
224/**
225 * gmap_enable - switch primary space to the guest address space
226 * @gmap: pointer to the guest address space structure
227 */
228void gmap_enable(struct gmap *gmap)
229{
230 S390_lowcore.gmap = (unsigned long) gmap;
231}
232EXPORT_SYMBOL_GPL(gmap_enable);
233
234/**
235 * gmap_disable - switch back to the standard primary address space
236 * @gmap: pointer to the guest address space structure
237 */
238void gmap_disable(struct gmap *gmap)
239{
240 S390_lowcore.gmap = 0UL;
241}
242EXPORT_SYMBOL_GPL(gmap_disable);
243
244/*
245 * gmap_alloc_table is assumed to be called with mmap_sem held
246 */
247static int gmap_alloc_table(struct gmap *gmap,
248 unsigned long *table, unsigned long init)
249{
250 struct page *page;
251 unsigned long *new;
252
253 /* since we dont free the gmap table until gmap_free we can unlock */
254 spin_unlock(&gmap->mm->page_table_lock);
255 page = alloc_pages(GFP_KERNEL, ALLOC_ORDER);
256 spin_lock(&gmap->mm->page_table_lock);
257 if (!page)
258 return -ENOMEM;
259 new = (unsigned long *) page_to_phys(page);
260 crst_table_init(new, init);
261 if (*table & _REGION_ENTRY_INV) {
262 list_add(&page->lru, &gmap->crst_list);
263 *table = (unsigned long) new | _REGION_ENTRY_LENGTH |
264 (*table & _REGION_ENTRY_TYPE_MASK);
265 } else
266 __free_pages(page, ALLOC_ORDER);
267 return 0;
268}
269
270/**
271 * gmap_unmap_segment - unmap segment from the guest address space
272 * @gmap: pointer to the guest address space structure
273 * @addr: address in the guest address space
274 * @len: length of the memory area to unmap
275 *
276 * Returns 0 if the unmap succeded, -EINVAL if not.
277 */
278int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len)
279{
280 unsigned long *table;
281 unsigned long off;
282 int flush;
283
284 if ((to | len) & (PMD_SIZE - 1))
285 return -EINVAL;
286 if (len == 0 || to + len < to)
287 return -EINVAL;
288
289 flush = 0;
290 down_read(&gmap->mm->mmap_sem);
291 spin_lock(&gmap->mm->page_table_lock);
292 for (off = 0; off < len; off += PMD_SIZE) {
293 /* Walk the guest addr space page table */
294 table = gmap->table + (((to + off) >> 53) & 0x7ff);
295 if (*table & _REGION_ENTRY_INV)
296 goto out;
297 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
298 table = table + (((to + off) >> 42) & 0x7ff);
299 if (*table & _REGION_ENTRY_INV)
300 goto out;
301 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
302 table = table + (((to + off) >> 31) & 0x7ff);
303 if (*table & _REGION_ENTRY_INV)
304 goto out;
305 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
306 table = table + (((to + off) >> 20) & 0x7ff);
307
308 /* Clear segment table entry in guest address space. */
309 flush |= gmap_unlink_segment(gmap, table);
310 *table = _SEGMENT_ENTRY_INV;
311 }
312out:
313 spin_unlock(&gmap->mm->page_table_lock);
314 up_read(&gmap->mm->mmap_sem);
315 if (flush)
316 gmap_flush_tlb(gmap);
317 return 0;
318}
319EXPORT_SYMBOL_GPL(gmap_unmap_segment);
320
321/**
322 * gmap_mmap_segment - map a segment to the guest address space
323 * @gmap: pointer to the guest address space structure
324 * @from: source address in the parent address space
325 * @to: target address in the guest address space
326 *
327 * Returns 0 if the mmap succeded, -EINVAL or -ENOMEM if not.
328 */
329int gmap_map_segment(struct gmap *gmap, unsigned long from,
330 unsigned long to, unsigned long len)
331{
332 unsigned long *table;
333 unsigned long off;
334 int flush;
335
336 if ((from | to | len) & (PMD_SIZE - 1))
337 return -EINVAL;
338 if (len == 0 || from + len > PGDIR_SIZE ||
339 from + len < from || to + len < to)
340 return -EINVAL;
341
342 flush = 0;
343 down_read(&gmap->mm->mmap_sem);
344 spin_lock(&gmap->mm->page_table_lock);
345 for (off = 0; off < len; off += PMD_SIZE) {
346 /* Walk the gmap address space page table */
347 table = gmap->table + (((to + off) >> 53) & 0x7ff);
348 if ((*table & _REGION_ENTRY_INV) &&
349 gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY))
350 goto out_unmap;
351 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
352 table = table + (((to + off) >> 42) & 0x7ff);
353 if ((*table & _REGION_ENTRY_INV) &&
354 gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY))
355 goto out_unmap;
356 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
357 table = table + (((to + off) >> 31) & 0x7ff);
358 if ((*table & _REGION_ENTRY_INV) &&
359 gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY))
360 goto out_unmap;
361 table = (unsigned long *) (*table & _REGION_ENTRY_ORIGIN);
362 table = table + (((to + off) >> 20) & 0x7ff);
363
364 /* Store 'from' address in an invalid segment table entry. */
365 flush |= gmap_unlink_segment(gmap, table);
366 *table = _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO | (from + off);
367 }
368 spin_unlock(&gmap->mm->page_table_lock);
369 up_read(&gmap->mm->mmap_sem);
370 if (flush)
371 gmap_flush_tlb(gmap);
372 return 0;
373
374out_unmap:
375 spin_unlock(&gmap->mm->page_table_lock);
376 up_read(&gmap->mm->mmap_sem);
377 gmap_unmap_segment(gmap, to, len);
378 return -ENOMEM;
379}
380EXPORT_SYMBOL_GPL(gmap_map_segment);
381
382/*
383 * this function is assumed to be called with mmap_sem held
384 */
385unsigned long __gmap_fault(unsigned long address, struct gmap *gmap)
386{
387 unsigned long *table, vmaddr, segment;
388 struct mm_struct *mm;
389 struct gmap_pgtable *mp;
390 struct gmap_rmap *rmap;
391 struct vm_area_struct *vma;
392 struct page *page;
393 pgd_t *pgd;
394 pud_t *pud;
395 pmd_t *pmd;
396
397 current->thread.gmap_addr = address;
398 mm = gmap->mm;
399 /* Walk the gmap address space page table */
400 table = gmap->table + ((address >> 53) & 0x7ff);
401 if (unlikely(*table & _REGION_ENTRY_INV))
402 return -EFAULT;
403 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
404 table = table + ((address >> 42) & 0x7ff);
405 if (unlikely(*table & _REGION_ENTRY_INV))
406 return -EFAULT;
407 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
408 table = table + ((address >> 31) & 0x7ff);
409 if (unlikely(*table & _REGION_ENTRY_INV))
410 return -EFAULT;
411 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
412 table = table + ((address >> 20) & 0x7ff);
413
414 /* Convert the gmap address to an mm address. */
415 segment = *table;
416 if (likely(!(segment & _SEGMENT_ENTRY_INV))) {
417 page = pfn_to_page(segment >> PAGE_SHIFT);
418 mp = (struct gmap_pgtable *) page->index;
419 return mp->vmaddr | (address & ~PMD_MASK);
420 } else if (segment & _SEGMENT_ENTRY_RO) {
421 vmaddr = segment & _SEGMENT_ENTRY_ORIGIN;
422 vma = find_vma(mm, vmaddr);
423 if (!vma || vma->vm_start > vmaddr)
424 return -EFAULT;
425
426 /* Walk the parent mm page table */
427 pgd = pgd_offset(mm, vmaddr);
428 pud = pud_alloc(mm, pgd, vmaddr);
429 if (!pud)
430 return -ENOMEM;
431 pmd = pmd_alloc(mm, pud, vmaddr);
432 if (!pmd)
433 return -ENOMEM;
434 if (!pmd_present(*pmd) &&
435 __pte_alloc(mm, vma, pmd, vmaddr))
436 return -ENOMEM;
437 /* pmd now points to a valid segment table entry. */
438 rmap = kmalloc(sizeof(*rmap), GFP_KERNEL|__GFP_REPEAT);
439 if (!rmap)
440 return -ENOMEM;
441 /* Link gmap segment table entry location to page table. */
442 page = pmd_page(*pmd);
443 mp = (struct gmap_pgtable *) page->index;
444 rmap->entry = table;
445 spin_lock(&mm->page_table_lock);
446 list_add(&rmap->list, &mp->mapper);
447 spin_unlock(&mm->page_table_lock);
448 /* Set gmap segment table entry to page table. */
449 *table = pmd_val(*pmd) & PAGE_MASK;
450 return vmaddr | (address & ~PMD_MASK);
451 }
452 return -EFAULT;
453}
454
455unsigned long gmap_fault(unsigned long address, struct gmap *gmap)
456{
457 unsigned long rc;
458
459 down_read(&gmap->mm->mmap_sem);
460 rc = __gmap_fault(address, gmap);
461 up_read(&gmap->mm->mmap_sem);
462
463 return rc;
464}
465EXPORT_SYMBOL_GPL(gmap_fault);
466
467void gmap_discard(unsigned long from, unsigned long to, struct gmap *gmap)
468{
469
470 unsigned long *table, address, size;
471 struct vm_area_struct *vma;
472 struct gmap_pgtable *mp;
473 struct page *page;
474
475 down_read(&gmap->mm->mmap_sem);
476 address = from;
477 while (address < to) {
478 /* Walk the gmap address space page table */
479 table = gmap->table + ((address >> 53) & 0x7ff);
480 if (unlikely(*table & _REGION_ENTRY_INV)) {
481 address = (address + PMD_SIZE) & PMD_MASK;
482 continue;
483 }
484 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
485 table = table + ((address >> 42) & 0x7ff);
486 if (unlikely(*table & _REGION_ENTRY_INV)) {
487 address = (address + PMD_SIZE) & PMD_MASK;
488 continue;
489 }
490 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
491 table = table + ((address >> 31) & 0x7ff);
492 if (unlikely(*table & _REGION_ENTRY_INV)) {
493 address = (address + PMD_SIZE) & PMD_MASK;
494 continue;
495 }
496 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN);
497 table = table + ((address >> 20) & 0x7ff);
498 if (unlikely(*table & _SEGMENT_ENTRY_INV)) {
499 address = (address + PMD_SIZE) & PMD_MASK;
500 continue;
501 }
502 page = pfn_to_page(*table >> PAGE_SHIFT);
503 mp = (struct gmap_pgtable *) page->index;
504 vma = find_vma(gmap->mm, mp->vmaddr);
505 size = min(to - address, PMD_SIZE - (address & ~PMD_MASK));
506 zap_page_range(vma, mp->vmaddr | (address & ~PMD_MASK),
507 size, NULL);
508 address = (address + PMD_SIZE) & PMD_MASK;
509 }
510 up_read(&gmap->mm->mmap_sem);
511}
512EXPORT_SYMBOL_GPL(gmap_discard);
513
514void gmap_unmap_notifier(struct mm_struct *mm, unsigned long *table)
515{
516 struct gmap_rmap *rmap, *next;
517 struct gmap_pgtable *mp;
518 struct page *page;
519 int flush;
520
521 flush = 0;
522 spin_lock(&mm->page_table_lock);
523 page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
524 mp = (struct gmap_pgtable *) page->index;
525 list_for_each_entry_safe(rmap, next, &mp->mapper, list) {
526 *rmap->entry =
527 _SEGMENT_ENTRY_INV | _SEGMENT_ENTRY_RO | mp->vmaddr;
528 list_del(&rmap->list);
529 kfree(rmap);
530 flush = 1;
531 }
532 spin_unlock(&mm->page_table_lock);
533 if (flush)
534 __tlb_flush_global();
535}
536
537static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm,
538 unsigned long vmaddr)
539{
540 struct page *page;
541 unsigned long *table;
542 struct gmap_pgtable *mp;
543
544 page = alloc_page(GFP_KERNEL|__GFP_REPEAT);
545 if (!page)
546 return NULL;
547 mp = kmalloc(sizeof(*mp), GFP_KERNEL|__GFP_REPEAT);
548 if (!mp) {
549 __free_page(page);
550 return NULL;
551 }
552 pgtable_page_ctor(page);
553 mp->vmaddr = vmaddr & PMD_MASK;
554 INIT_LIST_HEAD(&mp->mapper);
555 page->index = (unsigned long) mp;
556 atomic_set(&page->_mapcount, 3);
557 table = (unsigned long *) page_to_phys(page);
558 clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE/2);
559 clear_table(table + PTRS_PER_PTE, 0, PAGE_SIZE/2);
560 return table;
561}
562
563static inline void page_table_free_pgste(unsigned long *table)
564{
565 struct page *page;
566 struct gmap_pgtable *mp;
567
568 page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
569 mp = (struct gmap_pgtable *) page->index;
570 BUG_ON(!list_empty(&mp->mapper));
571 pgtable_page_dtor(page);
572 atomic_set(&page->_mapcount, -1);
573 kfree(mp);
574 __free_page(page);
575}
576
577#else /* CONFIG_PGSTE */
578
579static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm,
580 unsigned long vmaddr)
581{
582 return NULL;
583}
584
585static inline void page_table_free_pgste(unsigned long *table)
586{
587}
588
589static inline void gmap_unmap_notifier(struct mm_struct *mm,
590 unsigned long *table)
591{
592}
593
594#endif /* CONFIG_PGSTE */
595
596static inline unsigned int atomic_xor_bits(atomic_t *v, unsigned int bits)
597{
598 unsigned int old, new;
599
600 do {
601 old = atomic_read(v);
602 new = old ^ bits;
603 } while (atomic_cmpxchg(v, old, new) != old);
604 return new;
605}
606
607/*
608 * page table entry allocation/free routines.
609 */
610unsigned long *page_table_alloc(struct mm_struct *mm, unsigned long vmaddr)
611{
612 struct page *page;
613 unsigned long *table;
614 unsigned int mask, bit;
615
616 if (mm_has_pgste(mm))
617 return page_table_alloc_pgste(mm, vmaddr);
618 /* Allocate fragments of a 4K page as 1K/2K page table */
619 spin_lock_bh(&mm->context.list_lock);
620 mask = FRAG_MASK;
621 if (!list_empty(&mm->context.pgtable_list)) {
622 page = list_first_entry(&mm->context.pgtable_list,
623 struct page, lru);
624 table = (unsigned long *) page_to_phys(page);
625 mask = atomic_read(&page->_mapcount);
626 mask = mask | (mask >> 4);
627 }
628 if ((mask & FRAG_MASK) == FRAG_MASK) {
629 spin_unlock_bh(&mm->context.list_lock);
630 page = alloc_page(GFP_KERNEL|__GFP_REPEAT);
631 if (!page)
632 return NULL;
633 pgtable_page_ctor(page);
634 atomic_set(&page->_mapcount, 1);
635 table = (unsigned long *) page_to_phys(page);
636 clear_table(table, _PAGE_TYPE_EMPTY, PAGE_SIZE);
637 spin_lock_bh(&mm->context.list_lock);
638 list_add(&page->lru, &mm->context.pgtable_list);
639 } else {
640 for (bit = 1; mask & bit; bit <<= 1)
641 table += PTRS_PER_PTE;
642 mask = atomic_xor_bits(&page->_mapcount, bit);
643 if ((mask & FRAG_MASK) == FRAG_MASK)
644 list_del(&page->lru);
645 }
646 spin_unlock_bh(&mm->context.list_lock);
647 return table;
648}
649
650void page_table_free(struct mm_struct *mm, unsigned long *table)
651{
652 struct page *page;
653 unsigned int bit, mask;
654
655 if (mm_has_pgste(mm)) {
656 gmap_unmap_notifier(mm, table);
657 return page_table_free_pgste(table);
658 }
659 /* Free 1K/2K page table fragment of a 4K page */
660 page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
661 bit = 1 << ((__pa(table) & ~PAGE_MASK)/(PTRS_PER_PTE*sizeof(pte_t)));
662 spin_lock_bh(&mm->context.list_lock);
663 if ((atomic_read(&page->_mapcount) & FRAG_MASK) != FRAG_MASK)
664 list_del(&page->lru);
665 mask = atomic_xor_bits(&page->_mapcount, bit);
666 if (mask & FRAG_MASK)
667 list_add(&page->lru, &mm->context.pgtable_list);
668 spin_unlock_bh(&mm->context.list_lock);
669 if (mask == 0) {
670 pgtable_page_dtor(page);
671 atomic_set(&page->_mapcount, -1);
672 __free_page(page);
673 }
674}
675
676static void __page_table_free_rcu(void *table, unsigned bit)
677{
678 struct page *page;
679
680 if (bit == FRAG_MASK)
681 return page_table_free_pgste(table);
682 /* Free 1K/2K page table fragment of a 4K page */
683 page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
684 if (atomic_xor_bits(&page->_mapcount, bit) == 0) {
685 pgtable_page_dtor(page);
686 atomic_set(&page->_mapcount, -1);
687 __free_page(page);
688 }
689}
690
691void page_table_free_rcu(struct mmu_gather *tlb, unsigned long *table)
692{
693 struct mm_struct *mm;
694 struct page *page;
695 unsigned int bit, mask;
696
697 mm = tlb->mm;
698 if (mm_has_pgste(mm)) {
699 gmap_unmap_notifier(mm, table);
700 table = (unsigned long *) (__pa(table) | FRAG_MASK);
701 tlb_remove_table(tlb, table);
702 return;
703 }
704 bit = 1 << ((__pa(table) & ~PAGE_MASK) / (PTRS_PER_PTE*sizeof(pte_t)));
705 page = pfn_to_page(__pa(table) >> PAGE_SHIFT);
706 spin_lock_bh(&mm->context.list_lock);
707 if ((atomic_read(&page->_mapcount) & FRAG_MASK) != FRAG_MASK)
708 list_del(&page->lru);
709 mask = atomic_xor_bits(&page->_mapcount, bit | (bit << 4));
710 if (mask & FRAG_MASK)
711 list_add_tail(&page->lru, &mm->context.pgtable_list);
712 spin_unlock_bh(&mm->context.list_lock);
713 table = (unsigned long *) (__pa(table) | (bit << 4));
714 tlb_remove_table(tlb, table);
715}
716
717void __tlb_remove_table(void *_table)
718{
719 const unsigned long mask = (FRAG_MASK << 4) | FRAG_MASK;
720 void *table = (void *)((unsigned long) _table & ~mask);
721 unsigned type = (unsigned long) _table & mask;
722
723 if (type)
724 __page_table_free_rcu(table, type);
725 else
726 free_pages((unsigned long) table, ALLOC_ORDER);
727}
728
729static void tlb_remove_table_smp_sync(void *arg)
730{
731 /* Simply deliver the interrupt */
732}
733
734static void tlb_remove_table_one(void *table)
735{
736 /*
737 * This isn't an RCU grace period and hence the page-tables cannot be
738 * assumed to be actually RCU-freed.
739 *
740 * It is however sufficient for software page-table walkers that rely
741 * on IRQ disabling. See the comment near struct mmu_table_batch.
742 */
743 smp_call_function(tlb_remove_table_smp_sync, NULL, 1);
744 __tlb_remove_table(table);
745}
746
747static void tlb_remove_table_rcu(struct rcu_head *head)
748{
749 struct mmu_table_batch *batch;
750 int i;
751
752 batch = container_of(head, struct mmu_table_batch, rcu);
753
754 for (i = 0; i < batch->nr; i++)
755 __tlb_remove_table(batch->tables[i]);
756
757 free_page((unsigned long)batch);
758}
759
760void tlb_table_flush(struct mmu_gather *tlb)
761{
762 struct mmu_table_batch **batch = &tlb->batch;
763
764 if (*batch) {
765 __tlb_flush_mm(tlb->mm);
766 call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu);
767 *batch = NULL;
768 }
769}
770
771void tlb_remove_table(struct mmu_gather *tlb, void *table)
772{
773 struct mmu_table_batch **batch = &tlb->batch;
774
775 if (*batch == NULL) {
776 *batch = (struct mmu_table_batch *)
777 __get_free_page(GFP_NOWAIT | __GFP_NOWARN);
778 if (*batch == NULL) {
779 __tlb_flush_mm(tlb->mm);
780 tlb_remove_table_one(table);
781 return;
782 }
783 (*batch)->nr = 0;
784 }
785 (*batch)->tables[(*batch)->nr++] = table;
786 if ((*batch)->nr == MAX_TABLE_BATCH)
787 tlb_table_flush(tlb);
788}
789
790/*
791 * switch on pgstes for its userspace process (for kvm)
792 */
793int s390_enable_sie(void)
794{
795 struct task_struct *tsk = current;
796 struct mm_struct *mm, *old_mm;
797
798 /* Do we have switched amode? If no, we cannot do sie */
799 if (user_mode == HOME_SPACE_MODE)
800 return -EINVAL;
801
802 /* Do we have pgstes? if yes, we are done */
803 if (mm_has_pgste(tsk->mm))
804 return 0;
805
806 /* lets check if we are allowed to replace the mm */
807 task_lock(tsk);
808 if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 ||
809#ifdef CONFIG_AIO
810 !hlist_empty(&tsk->mm->ioctx_list) ||
811#endif
812 tsk->mm != tsk->active_mm) {
813 task_unlock(tsk);
814 return -EINVAL;
815 }
816 task_unlock(tsk);
817
818 /* we copy the mm and let dup_mm create the page tables with_pgstes */
819 tsk->mm->context.alloc_pgste = 1;
820 /* make sure that both mms have a correct rss state */
821 sync_mm_rss(tsk->mm);
822 mm = dup_mm(tsk);
823 tsk->mm->context.alloc_pgste = 0;
824 if (!mm)
825 return -ENOMEM;
826
827 /* Now lets check again if something happened */
828 task_lock(tsk);
829 if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 ||
830#ifdef CONFIG_AIO
831 !hlist_empty(&tsk->mm->ioctx_list) ||
832#endif
833 tsk->mm != tsk->active_mm) {
834 mmput(mm);
835 task_unlock(tsk);
836 return -EINVAL;
837 }
838
839 /* ok, we are alone. No ptrace, no threads, etc. */
840 old_mm = tsk->mm;
841 tsk->mm = tsk->active_mm = mm;
842 preempt_disable();
843 update_mm(mm, tsk);
844 atomic_inc(&mm->context.attach_count);
845 atomic_dec(&old_mm->context.attach_count);
846 cpumask_set_cpu(smp_processor_id(), mm_cpumask(mm));
847 preempt_enable();
848 task_unlock(tsk);
849 mmput(old_mm);
850 return 0;
851}
852EXPORT_SYMBOL_GPL(s390_enable_sie);
853
854#if defined(CONFIG_DEBUG_PAGEALLOC) && defined(CONFIG_HIBERNATION)
855bool kernel_page_present(struct page *page)
856{
857 unsigned long addr;
858 int cc;
859
860 addr = page_to_phys(page);
861 asm volatile(
862 " lra %1,0(%1)\n"
863 " ipm %0\n"
864 " srl %0,28"
865 : "=d" (cc), "+a" (addr) : : "cc");
866 return cc == 0;
867}
868#endif /* CONFIG_HIBERNATION && CONFIG_DEBUG_PAGEALLOC */