]>
Commit | Line | Data |
---|---|---|
3b827c1b JF |
1 | /* |
2 | * Xen mmu operations | |
3 | * | |
4 | * This file contains the various mmu fetch and update operations. | |
5 | * The most important job they must perform is the mapping between the | |
6 | * domain's pfn and the overall machine mfns. | |
7 | * | |
8 | * Xen allows guests to directly update the pagetable, in a controlled | |
9 | * fashion. In other words, the guest modifies the same pagetable | |
10 | * that the CPU actually uses, which eliminates the overhead of having | |
11 | * a separate shadow pagetable. | |
12 | * | |
13 | * In order to allow this, it falls on the guest domain to map its | |
14 | * notion of a "physical" pfn - which is just a domain-local linear | |
15 | * address - into a real "machine address" which the CPU's MMU can | |
16 | * use. | |
17 | * | |
18 | * A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be | |
19 | * inserted directly into the pagetable. When creating a new | |
20 | * pte/pmd/pgd, it converts the passed pfn into an mfn. Conversely, | |
21 | * when reading the content back with __(pgd|pmd|pte)_val, it converts | |
22 | * the mfn back into a pfn. | |
23 | * | |
24 | * The other constraint is that all pages which make up a pagetable | |
25 | * must be mapped read-only in the guest. This prevents uncontrolled | |
26 | * guest updates to the pagetable. Xen strictly enforces this, and | |
27 | * will disallow any pagetable update which will end up mapping a | |
28 | * pagetable page RW, and will disallow using any writable page as a | |
29 | * pagetable. | |
30 | * | |
31 | * Naively, when loading %cr3 with the base of a new pagetable, Xen | |
32 | * would need to validate the whole pagetable before going on. | |
33 | * Naturally, this is quite slow. The solution is to "pin" a | |
34 | * pagetable, which enforces all the constraints on the pagetable even | |
35 | * when it is not actively in use. This menas that Xen can be assured | |
36 | * that it is still valid when you do load it into %cr3, and doesn't | |
37 | * need to revalidate it. | |
38 | * | |
39 | * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007 | |
40 | */ | |
f120f13e | 41 | #include <linux/sched.h> |
f4f97b3e | 42 | #include <linux/highmem.h> |
3b827c1b | 43 | #include <linux/bug.h> |
3b827c1b JF |
44 | |
45 | #include <asm/pgtable.h> | |
46 | #include <asm/tlbflush.h> | |
47 | #include <asm/mmu_context.h> | |
f4f97b3e | 48 | #include <asm/paravirt.h> |
3b827c1b JF |
49 | |
50 | #include <asm/xen/hypercall.h> | |
f4f97b3e | 51 | #include <asm/xen/hypervisor.h> |
3b827c1b JF |
52 | |
53 | #include <xen/page.h> | |
54 | #include <xen/interface/xen.h> | |
55 | ||
f4f97b3e | 56 | #include "multicalls.h" |
3b827c1b JF |
57 | #include "mmu.h" |
58 | ||
59 | xmaddr_t arbitrary_virt_to_machine(unsigned long address) | |
60 | { | |
da7bfc50 | 61 | unsigned int level; |
f0646e43 | 62 | pte_t *pte = lookup_address(address, &level); |
3b827c1b JF |
63 | unsigned offset = address & PAGE_MASK; |
64 | ||
65 | BUG_ON(pte == NULL); | |
66 | ||
67 | return XMADDR((pte_mfn(*pte) << PAGE_SHIFT) + offset); | |
68 | } | |
69 | ||
70 | void make_lowmem_page_readonly(void *vaddr) | |
71 | { | |
72 | pte_t *pte, ptev; | |
73 | unsigned long address = (unsigned long)vaddr; | |
da7bfc50 | 74 | unsigned int level; |
3b827c1b | 75 | |
f0646e43 | 76 | pte = lookup_address(address, &level); |
3b827c1b JF |
77 | BUG_ON(pte == NULL); |
78 | ||
79 | ptev = pte_wrprotect(*pte); | |
80 | ||
81 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
82 | BUG(); | |
83 | } | |
84 | ||
85 | void make_lowmem_page_readwrite(void *vaddr) | |
86 | { | |
87 | pte_t *pte, ptev; | |
88 | unsigned long address = (unsigned long)vaddr; | |
da7bfc50 | 89 | unsigned int level; |
3b827c1b | 90 | |
f0646e43 | 91 | pte = lookup_address(address, &level); |
3b827c1b JF |
92 | BUG_ON(pte == NULL); |
93 | ||
94 | ptev = pte_mkwrite(*pte); | |
95 | ||
96 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
97 | BUG(); | |
98 | } | |
99 | ||
100 | ||
3b827c1b JF |
101 | void xen_set_pmd(pmd_t *ptr, pmd_t val) |
102 | { | |
d66bf8fc JF |
103 | struct multicall_space mcs; |
104 | struct mmu_update *u; | |
3b827c1b | 105 | |
d66bf8fc JF |
106 | preempt_disable(); |
107 | ||
108 | mcs = xen_mc_entry(sizeof(*u)); | |
109 | u = mcs.args; | |
110 | u->ptr = virt_to_machine(ptr).maddr; | |
111 | u->val = pmd_val_ma(val); | |
112 | MULTI_mmu_update(mcs.mc, u, 1, NULL, DOMID_SELF); | |
113 | ||
114 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
115 | ||
116 | preempt_enable(); | |
3b827c1b JF |
117 | } |
118 | ||
3b827c1b JF |
119 | /* |
120 | * Associate a virtual page frame with a given physical page frame | |
121 | * and protection flags for that frame. | |
122 | */ | |
123 | void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags) | |
124 | { | |
125 | pgd_t *pgd; | |
126 | pud_t *pud; | |
127 | pmd_t *pmd; | |
128 | pte_t *pte; | |
129 | ||
130 | pgd = swapper_pg_dir + pgd_index(vaddr); | |
131 | if (pgd_none(*pgd)) { | |
132 | BUG(); | |
133 | return; | |
134 | } | |
135 | pud = pud_offset(pgd, vaddr); | |
136 | if (pud_none(*pud)) { | |
137 | BUG(); | |
138 | return; | |
139 | } | |
140 | pmd = pmd_offset(pud, vaddr); | |
141 | if (pmd_none(*pmd)) { | |
142 | BUG(); | |
143 | return; | |
144 | } | |
145 | pte = pte_offset_kernel(pmd, vaddr); | |
146 | /* <mfn,flags> stored as-is, to permit clearing entries */ | |
147 | xen_set_pte(pte, mfn_pte(mfn, flags)); | |
148 | ||
149 | /* | |
150 | * It's enough to flush this one mapping. | |
151 | * (PGE mappings get flushed as well) | |
152 | */ | |
153 | __flush_tlb_one(vaddr); | |
154 | } | |
155 | ||
156 | void xen_set_pte_at(struct mm_struct *mm, unsigned long addr, | |
157 | pte_t *ptep, pte_t pteval) | |
158 | { | |
d66bf8fc | 159 | if (mm == current->mm || mm == &init_mm) { |
8965c1c0 | 160 | if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) { |
d66bf8fc JF |
161 | struct multicall_space mcs; |
162 | mcs = xen_mc_entry(0); | |
163 | ||
164 | MULTI_update_va_mapping(mcs.mc, addr, pteval, 0); | |
165 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
166 | return; | |
167 | } else | |
168 | if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0) | |
169 | return; | |
170 | } | |
171 | xen_set_pte(ptep, pteval); | |
3b827c1b JF |
172 | } |
173 | ||
174 | #ifdef CONFIG_X86_PAE | |
f4f97b3e JF |
175 | void xen_set_pud(pud_t *ptr, pud_t val) |
176 | { | |
d66bf8fc JF |
177 | struct multicall_space mcs; |
178 | struct mmu_update *u; | |
f4f97b3e | 179 | |
d66bf8fc JF |
180 | preempt_disable(); |
181 | ||
182 | mcs = xen_mc_entry(sizeof(*u)); | |
183 | u = mcs.args; | |
184 | u->ptr = virt_to_machine(ptr).maddr; | |
185 | u->val = pud_val_ma(val); | |
186 | MULTI_mmu_update(mcs.mc, u, 1, NULL, DOMID_SELF); | |
187 | ||
188 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
189 | ||
190 | preempt_enable(); | |
f4f97b3e JF |
191 | } |
192 | ||
193 | void xen_set_pte(pte_t *ptep, pte_t pte) | |
194 | { | |
195 | ptep->pte_high = pte.pte_high; | |
196 | smp_wmb(); | |
197 | ptep->pte_low = pte.pte_low; | |
198 | } | |
199 | ||
3b827c1b JF |
200 | void xen_set_pte_atomic(pte_t *ptep, pte_t pte) |
201 | { | |
202 | set_64bit((u64 *)ptep, pte_val_ma(pte)); | |
203 | } | |
204 | ||
205 | void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep) | |
206 | { | |
207 | ptep->pte_low = 0; | |
208 | smp_wmb(); /* make sure low gets written first */ | |
209 | ptep->pte_high = 0; | |
210 | } | |
211 | ||
212 | void xen_pmd_clear(pmd_t *pmdp) | |
213 | { | |
214 | xen_set_pmd(pmdp, __pmd(0)); | |
215 | } | |
216 | ||
abf33038 | 217 | pteval_t xen_pte_val(pte_t pte) |
3b827c1b | 218 | { |
430442e3 | 219 | pteval_t ret = pte.pte; |
3b827c1b | 220 | |
430442e3 JF |
221 | if (ret & _PAGE_PRESENT) |
222 | ret = machine_to_phys(XMADDR(ret)).paddr | _PAGE_PRESENT; | |
3b827c1b JF |
223 | |
224 | return ret; | |
225 | } | |
226 | ||
abf33038 | 227 | pmdval_t xen_pmd_val(pmd_t pmd) |
3b827c1b | 228 | { |
abf33038 | 229 | pmdval_t ret = pmd.pmd; |
430442e3 JF |
230 | if (ret & _PAGE_PRESENT) |
231 | ret = machine_to_phys(XMADDR(ret)).paddr | _PAGE_PRESENT; | |
3b827c1b JF |
232 | return ret; |
233 | } | |
234 | ||
abf33038 | 235 | pgdval_t xen_pgd_val(pgd_t pgd) |
3b827c1b | 236 | { |
abf33038 | 237 | pgdval_t ret = pgd.pgd; |
430442e3 JF |
238 | if (ret & _PAGE_PRESENT) |
239 | ret = machine_to_phys(XMADDR(ret)).paddr | _PAGE_PRESENT; | |
3b827c1b JF |
240 | return ret; |
241 | } | |
242 | ||
abf33038 | 243 | pte_t xen_make_pte(pteval_t pte) |
3b827c1b | 244 | { |
a89780f3 | 245 | if (pte & _PAGE_PRESENT) { |
3b827c1b | 246 | pte = phys_to_machine(XPADDR(pte)).maddr; |
a89780f3 JF |
247 | pte &= ~(_PAGE_PCD | _PAGE_PWT); |
248 | } | |
3b827c1b | 249 | |
c8e5393a | 250 | return (pte_t){ .pte = pte }; |
3b827c1b JF |
251 | } |
252 | ||
abf33038 | 253 | pmd_t xen_make_pmd(pmdval_t pmd) |
3b827c1b | 254 | { |
430442e3 | 255 | if (pmd & _PAGE_PRESENT) |
3b827c1b JF |
256 | pmd = phys_to_machine(XPADDR(pmd)).maddr; |
257 | ||
258 | return (pmd_t){ pmd }; | |
259 | } | |
260 | ||
abf33038 | 261 | pgd_t xen_make_pgd(pgdval_t pgd) |
3b827c1b JF |
262 | { |
263 | if (pgd & _PAGE_PRESENT) | |
264 | pgd = phys_to_machine(XPADDR(pgd)).maddr; | |
265 | ||
266 | return (pgd_t){ pgd }; | |
267 | } | |
268 | #else /* !PAE */ | |
f4f97b3e JF |
269 | void xen_set_pte(pte_t *ptep, pte_t pte) |
270 | { | |
271 | *ptep = pte; | |
272 | } | |
273 | ||
abf33038 | 274 | pteval_t xen_pte_val(pte_t pte) |
3b827c1b | 275 | { |
430442e3 | 276 | pteval_t ret = pte.pte; |
3b827c1b JF |
277 | |
278 | if (ret & _PAGE_PRESENT) | |
279 | ret = machine_to_phys(XMADDR(ret)).paddr; | |
280 | ||
281 | return ret; | |
282 | } | |
283 | ||
abf33038 | 284 | pgdval_t xen_pgd_val(pgd_t pgd) |
3b827c1b | 285 | { |
abf33038 | 286 | pteval_t ret = pgd.pgd; |
430442e3 JF |
287 | if (ret & _PAGE_PRESENT) |
288 | ret = machine_to_phys(XMADDR(ret)).paddr | _PAGE_PRESENT; | |
3b827c1b JF |
289 | return ret; |
290 | } | |
291 | ||
abf33038 | 292 | pte_t xen_make_pte(pteval_t pte) |
3b827c1b | 293 | { |
a89780f3 | 294 | if (pte & _PAGE_PRESENT) { |
3b827c1b | 295 | pte = phys_to_machine(XPADDR(pte)).maddr; |
a89780f3 JF |
296 | pte &= ~(_PAGE_PCD | _PAGE_PWT); |
297 | } | |
2c80b01b | 298 | |
3b827c1b JF |
299 | return (pte_t){ pte }; |
300 | } | |
301 | ||
abf33038 | 302 | pgd_t xen_make_pgd(pgdval_t pgd) |
3b827c1b JF |
303 | { |
304 | if (pgd & _PAGE_PRESENT) | |
305 | pgd = phys_to_machine(XPADDR(pgd)).maddr; | |
306 | ||
307 | return (pgd_t){ pgd }; | |
308 | } | |
309 | #endif /* CONFIG_X86_PAE */ | |
310 | ||
f4f97b3e JF |
311 | /* |
312 | (Yet another) pagetable walker. This one is intended for pinning a | |
313 | pagetable. This means that it walks a pagetable and calls the | |
314 | callback function on each page it finds making up the page table, | |
315 | at every level. It walks the entire pagetable, but it only bothers | |
316 | pinning pte pages which are below pte_limit. In the normal case | |
317 | this will be TASK_SIZE, but at boot we need to pin up to | |
318 | FIXADDR_TOP. But the important bit is that we don't pin beyond | |
319 | there, because then we start getting into Xen's ptes. | |
320 | */ | |
74260714 | 321 | static int pgd_walk(pgd_t *pgd_base, int (*func)(struct page *, enum pt_level), |
f4f97b3e | 322 | unsigned long limit) |
3b827c1b JF |
323 | { |
324 | pgd_t *pgd = pgd_base; | |
f4f97b3e JF |
325 | int flush = 0; |
326 | unsigned long addr = 0; | |
327 | unsigned long pgd_next; | |
328 | ||
329 | BUG_ON(limit > FIXADDR_TOP); | |
3b827c1b JF |
330 | |
331 | if (xen_feature(XENFEAT_auto_translated_physmap)) | |
f4f97b3e JF |
332 | return 0; |
333 | ||
334 | for (; addr != FIXADDR_TOP; pgd++, addr = pgd_next) { | |
335 | pud_t *pud; | |
336 | unsigned long pud_limit, pud_next; | |
3b827c1b | 337 | |
f4f97b3e JF |
338 | pgd_next = pud_limit = pgd_addr_end(addr, FIXADDR_TOP); |
339 | ||
340 | if (!pgd_val(*pgd)) | |
3b827c1b | 341 | continue; |
f4f97b3e | 342 | |
3b827c1b JF |
343 | pud = pud_offset(pgd, 0); |
344 | ||
345 | if (PTRS_PER_PUD > 1) /* not folded */ | |
74260714 | 346 | flush |= (*func)(virt_to_page(pud), PT_PUD); |
f4f97b3e JF |
347 | |
348 | for (; addr != pud_limit; pud++, addr = pud_next) { | |
349 | pmd_t *pmd; | |
350 | unsigned long pmd_limit; | |
351 | ||
352 | pud_next = pud_addr_end(addr, pud_limit); | |
353 | ||
354 | if (pud_next < limit) | |
355 | pmd_limit = pud_next; | |
356 | else | |
357 | pmd_limit = limit; | |
3b827c1b | 358 | |
3b827c1b JF |
359 | if (pud_none(*pud)) |
360 | continue; | |
f4f97b3e | 361 | |
3b827c1b JF |
362 | pmd = pmd_offset(pud, 0); |
363 | ||
364 | if (PTRS_PER_PMD > 1) /* not folded */ | |
74260714 | 365 | flush |= (*func)(virt_to_page(pmd), PT_PMD); |
f4f97b3e JF |
366 | |
367 | for (; addr != pmd_limit; pmd++) { | |
368 | addr += (PAGE_SIZE * PTRS_PER_PTE); | |
369 | if ((pmd_limit-1) < (addr-1)) { | |
370 | addr = pmd_limit; | |
371 | break; | |
372 | } | |
3b827c1b | 373 | |
3b827c1b JF |
374 | if (pmd_none(*pmd)) |
375 | continue; | |
376 | ||
74260714 | 377 | flush |= (*func)(pmd_page(*pmd), PT_PTE); |
3b827c1b JF |
378 | } |
379 | } | |
380 | } | |
381 | ||
74260714 | 382 | flush |= (*func)(virt_to_page(pgd_base), PT_PGD); |
f4f97b3e JF |
383 | |
384 | return flush; | |
3b827c1b JF |
385 | } |
386 | ||
74260714 JF |
387 | static spinlock_t *lock_pte(struct page *page) |
388 | { | |
389 | spinlock_t *ptl = NULL; | |
390 | ||
391 | #if NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS | |
392 | ptl = __pte_lockptr(page); | |
393 | spin_lock(ptl); | |
394 | #endif | |
395 | ||
396 | return ptl; | |
397 | } | |
398 | ||
399 | static void do_unlock(void *v) | |
400 | { | |
401 | spinlock_t *ptl = v; | |
402 | spin_unlock(ptl); | |
403 | } | |
404 | ||
405 | static void xen_do_pin(unsigned level, unsigned long pfn) | |
406 | { | |
407 | struct mmuext_op *op; | |
408 | struct multicall_space mcs; | |
409 | ||
410 | mcs = __xen_mc_entry(sizeof(*op)); | |
411 | op = mcs.args; | |
412 | op->cmd = level; | |
413 | op->arg1.mfn = pfn_to_mfn(pfn); | |
414 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
415 | } | |
416 | ||
417 | static int pin_page(struct page *page, enum pt_level level) | |
f4f97b3e JF |
418 | { |
419 | unsigned pgfl = test_and_set_bit(PG_pinned, &page->flags); | |
420 | int flush; | |
421 | ||
422 | if (pgfl) | |
423 | flush = 0; /* already pinned */ | |
424 | else if (PageHighMem(page)) | |
425 | /* kmaps need flushing if we found an unpinned | |
426 | highpage */ | |
427 | flush = 1; | |
428 | else { | |
429 | void *pt = lowmem_page_address(page); | |
430 | unsigned long pfn = page_to_pfn(page); | |
431 | struct multicall_space mcs = __xen_mc_entry(0); | |
74260714 | 432 | spinlock_t *ptl; |
f4f97b3e JF |
433 | |
434 | flush = 0; | |
435 | ||
74260714 JF |
436 | ptl = NULL; |
437 | if (level == PT_PTE) | |
438 | ptl = lock_pte(page); | |
439 | ||
f4f97b3e JF |
440 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, |
441 | pfn_pte(pfn, PAGE_KERNEL_RO), | |
74260714 JF |
442 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
443 | ||
444 | if (level == PT_PTE) | |
445 | xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn); | |
446 | ||
447 | if (ptl) { | |
448 | /* Queue a deferred unlock for when this batch | |
449 | is completed. */ | |
450 | xen_mc_callback(do_unlock, ptl); | |
451 | } | |
f4f97b3e JF |
452 | } |
453 | ||
454 | return flush; | |
455 | } | |
3b827c1b | 456 | |
f4f97b3e JF |
457 | /* This is called just after a mm has been created, but it has not |
458 | been used yet. We need to make sure that its pagetable is all | |
459 | read-only, and can be pinned. */ | |
3b827c1b JF |
460 | void xen_pgd_pin(pgd_t *pgd) |
461 | { | |
74260714 | 462 | unsigned level; |
3b827c1b | 463 | |
f4f97b3e | 464 | xen_mc_batch(); |
3b827c1b | 465 | |
f87e4cac JF |
466 | if (pgd_walk(pgd, pin_page, TASK_SIZE)) { |
467 | /* re-enable interrupts for kmap_flush_unused */ | |
468 | xen_mc_issue(0); | |
f4f97b3e | 469 | kmap_flush_unused(); |
f87e4cac JF |
470 | xen_mc_batch(); |
471 | } | |
f4f97b3e | 472 | |
f4f97b3e | 473 | #ifdef CONFIG_X86_PAE |
74260714 | 474 | level = MMUEXT_PIN_L3_TABLE; |
3b827c1b | 475 | #else |
74260714 | 476 | level = MMUEXT_PIN_L2_TABLE; |
3b827c1b | 477 | #endif |
74260714 JF |
478 | |
479 | xen_do_pin(level, PFN_DOWN(__pa(pgd))); | |
f4f97b3e JF |
480 | |
481 | xen_mc_issue(0); | |
3b827c1b JF |
482 | } |
483 | ||
f4f97b3e JF |
484 | /* The init_mm pagetable is really pinned as soon as its created, but |
485 | that's before we have page structures to store the bits. So do all | |
486 | the book-keeping now. */ | |
74260714 | 487 | static __init int mark_pinned(struct page *page, enum pt_level level) |
3b827c1b | 488 | { |
f4f97b3e JF |
489 | SetPagePinned(page); |
490 | return 0; | |
491 | } | |
3b827c1b | 492 | |
f4f97b3e JF |
493 | void __init xen_mark_init_mm_pinned(void) |
494 | { | |
495 | pgd_walk(init_mm.pgd, mark_pinned, FIXADDR_TOP); | |
496 | } | |
3b827c1b | 497 | |
74260714 | 498 | static int unpin_page(struct page *page, enum pt_level level) |
f4f97b3e JF |
499 | { |
500 | unsigned pgfl = test_and_clear_bit(PG_pinned, &page->flags); | |
3b827c1b | 501 | |
f4f97b3e JF |
502 | if (pgfl && !PageHighMem(page)) { |
503 | void *pt = lowmem_page_address(page); | |
504 | unsigned long pfn = page_to_pfn(page); | |
74260714 JF |
505 | spinlock_t *ptl = NULL; |
506 | struct multicall_space mcs; | |
507 | ||
508 | if (level == PT_PTE) { | |
509 | ptl = lock_pte(page); | |
510 | ||
511 | xen_do_pin(MMUEXT_UNPIN_TABLE, pfn); | |
512 | } | |
513 | ||
514 | mcs = __xen_mc_entry(0); | |
f4f97b3e JF |
515 | |
516 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, | |
517 | pfn_pte(pfn, PAGE_KERNEL), | |
74260714 JF |
518 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
519 | ||
520 | if (ptl) { | |
521 | /* unlock when batch completed */ | |
522 | xen_mc_callback(do_unlock, ptl); | |
523 | } | |
f4f97b3e JF |
524 | } |
525 | ||
526 | return 0; /* never need to flush on unpin */ | |
3b827c1b JF |
527 | } |
528 | ||
f4f97b3e JF |
529 | /* Release a pagetables pages back as normal RW */ |
530 | static void xen_pgd_unpin(pgd_t *pgd) | |
531 | { | |
f4f97b3e JF |
532 | xen_mc_batch(); |
533 | ||
74260714 | 534 | xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); |
f4f97b3e JF |
535 | |
536 | pgd_walk(pgd, unpin_page, TASK_SIZE); | |
537 | ||
538 | xen_mc_issue(0); | |
539 | } | |
3b827c1b JF |
540 | |
541 | void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next) | |
542 | { | |
f4f97b3e | 543 | spin_lock(&next->page_table_lock); |
3b827c1b | 544 | xen_pgd_pin(next->pgd); |
f4f97b3e | 545 | spin_unlock(&next->page_table_lock); |
3b827c1b JF |
546 | } |
547 | ||
548 | void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm) | |
549 | { | |
f4f97b3e | 550 | spin_lock(&mm->page_table_lock); |
3b827c1b | 551 | xen_pgd_pin(mm->pgd); |
f4f97b3e | 552 | spin_unlock(&mm->page_table_lock); |
3b827c1b JF |
553 | } |
554 | ||
3b827c1b | 555 | |
f87e4cac JF |
556 | #ifdef CONFIG_SMP |
557 | /* Another cpu may still have their %cr3 pointing at the pagetable, so | |
558 | we need to repoint it somewhere else before we can unpin it. */ | |
559 | static void drop_other_mm_ref(void *info) | |
560 | { | |
561 | struct mm_struct *mm = info; | |
3b827c1b | 562 | |
f87e4cac JF |
563 | if (__get_cpu_var(cpu_tlbstate).active_mm == mm) |
564 | leave_mm(smp_processor_id()); | |
9f79991d JF |
565 | |
566 | /* If this cpu still has a stale cr3 reference, then make sure | |
567 | it has been flushed. */ | |
568 | if (x86_read_percpu(xen_current_cr3) == __pa(mm->pgd)) { | |
569 | load_cr3(swapper_pg_dir); | |
570 | arch_flush_lazy_cpu_mode(); | |
571 | } | |
f87e4cac | 572 | } |
3b827c1b | 573 | |
f87e4cac JF |
574 | static void drop_mm_ref(struct mm_struct *mm) |
575 | { | |
9f79991d JF |
576 | cpumask_t mask; |
577 | unsigned cpu; | |
578 | ||
f87e4cac JF |
579 | if (current->active_mm == mm) { |
580 | if (current->mm == mm) | |
581 | load_cr3(swapper_pg_dir); | |
582 | else | |
583 | leave_mm(smp_processor_id()); | |
9f79991d JF |
584 | arch_flush_lazy_cpu_mode(); |
585 | } | |
586 | ||
587 | /* Get the "official" set of cpus referring to our pagetable. */ | |
588 | mask = mm->cpu_vm_mask; | |
589 | ||
590 | /* It's possible that a vcpu may have a stale reference to our | |
591 | cr3, because its in lazy mode, and it hasn't yet flushed | |
592 | its set of pending hypercalls yet. In this case, we can | |
593 | look at its actual current cr3 value, and force it to flush | |
594 | if needed. */ | |
595 | for_each_online_cpu(cpu) { | |
596 | if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd)) | |
597 | cpu_set(cpu, mask); | |
3b827c1b JF |
598 | } |
599 | ||
9f79991d JF |
600 | if (!cpus_empty(mask)) |
601 | xen_smp_call_function_mask(mask, drop_other_mm_ref, mm, 1); | |
f87e4cac JF |
602 | } |
603 | #else | |
604 | static void drop_mm_ref(struct mm_struct *mm) | |
605 | { | |
606 | if (current->active_mm == mm) | |
607 | load_cr3(swapper_pg_dir); | |
608 | } | |
609 | #endif | |
610 | ||
611 | /* | |
612 | * While a process runs, Xen pins its pagetables, which means that the | |
613 | * hypervisor forces it to be read-only, and it controls all updates | |
614 | * to it. This means that all pagetable updates have to go via the | |
615 | * hypervisor, which is moderately expensive. | |
616 | * | |
617 | * Since we're pulling the pagetable down, we switch to use init_mm, | |
618 | * unpin old process pagetable and mark it all read-write, which | |
619 | * allows further operations on it to be simple memory accesses. | |
620 | * | |
621 | * The only subtle point is that another CPU may be still using the | |
622 | * pagetable because of lazy tlb flushing. This means we need need to | |
623 | * switch all CPUs off this pagetable before we can unpin it. | |
624 | */ | |
625 | void xen_exit_mmap(struct mm_struct *mm) | |
626 | { | |
627 | get_cpu(); /* make sure we don't move around */ | |
628 | drop_mm_ref(mm); | |
629 | put_cpu(); | |
3b827c1b | 630 | |
f120f13e | 631 | spin_lock(&mm->page_table_lock); |
df912ea4 JF |
632 | |
633 | /* pgd may not be pinned in the error exit path of execve */ | |
634 | if (PagePinned(virt_to_page(mm->pgd))) | |
635 | xen_pgd_unpin(mm->pgd); | |
74260714 | 636 | |
f120f13e | 637 | spin_unlock(&mm->page_table_lock); |
3b827c1b | 638 | } |