]>
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> |
cbcd79c2 | 49 | #include <asm/linkage.h> |
3b827c1b JF |
50 | |
51 | #include <asm/xen/hypercall.h> | |
f4f97b3e | 52 | #include <asm/xen/hypervisor.h> |
3b827c1b JF |
53 | |
54 | #include <xen/page.h> | |
55 | #include <xen/interface/xen.h> | |
56 | ||
f4f97b3e | 57 | #include "multicalls.h" |
3b827c1b JF |
58 | #include "mmu.h" |
59 | ||
d451bb7a | 60 | #define P2M_ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(unsigned long)) |
cf0923ea | 61 | #define TOP_ENTRIES (MAX_DOMAIN_PAGES / P2M_ENTRIES_PER_PAGE) |
d451bb7a | 62 | |
cf0923ea | 63 | /* Placeholder for holes in the address space */ |
cbcd79c2 | 64 | static unsigned long p2m_missing[P2M_ENTRIES_PER_PAGE] __page_aligned_data = |
cf0923ea JF |
65 | { [ 0 ... P2M_ENTRIES_PER_PAGE-1 ] = ~0UL }; |
66 | ||
67 | /* Array of pointers to pages containing p2m entries */ | |
cbcd79c2 | 68 | static unsigned long *p2m_top[TOP_ENTRIES] __page_aligned_data = |
cf0923ea | 69 | { [ 0 ... TOP_ENTRIES - 1] = &p2m_missing[0] }; |
d451bb7a | 70 | |
d5edbc1f | 71 | /* Arrays of p2m arrays expressed in mfns used for save/restore */ |
cbcd79c2 | 72 | static unsigned long p2m_top_mfn[TOP_ENTRIES] __page_aligned_bss; |
d5edbc1f | 73 | |
cbcd79c2 JF |
74 | static unsigned long p2m_top_mfn_list[TOP_ENTRIES / P2M_ENTRIES_PER_PAGE] |
75 | __page_aligned_bss; | |
d5edbc1f | 76 | |
d451bb7a JF |
77 | static inline unsigned p2m_top_index(unsigned long pfn) |
78 | { | |
8006ec3e | 79 | BUG_ON(pfn >= MAX_DOMAIN_PAGES); |
d451bb7a JF |
80 | return pfn / P2M_ENTRIES_PER_PAGE; |
81 | } | |
82 | ||
83 | static inline unsigned p2m_index(unsigned long pfn) | |
84 | { | |
85 | return pfn % P2M_ENTRIES_PER_PAGE; | |
86 | } | |
87 | ||
d5edbc1f JF |
88 | /* Build the parallel p2m_top_mfn structures */ |
89 | void xen_setup_mfn_list_list(void) | |
90 | { | |
91 | unsigned pfn, idx; | |
92 | ||
93 | for(pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn += P2M_ENTRIES_PER_PAGE) { | |
94 | unsigned topidx = p2m_top_index(pfn); | |
95 | ||
96 | p2m_top_mfn[topidx] = virt_to_mfn(p2m_top[topidx]); | |
97 | } | |
98 | ||
99 | for(idx = 0; idx < ARRAY_SIZE(p2m_top_mfn_list); idx++) { | |
100 | unsigned topidx = idx * P2M_ENTRIES_PER_PAGE; | |
101 | p2m_top_mfn_list[idx] = virt_to_mfn(&p2m_top_mfn[topidx]); | |
102 | } | |
103 | ||
104 | BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); | |
105 | ||
106 | HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list = | |
107 | virt_to_mfn(p2m_top_mfn_list); | |
108 | HYPERVISOR_shared_info->arch.max_pfn = xen_start_info->nr_pages; | |
109 | } | |
110 | ||
111 | /* Set up p2m_top to point to the domain-builder provided p2m pages */ | |
d451bb7a JF |
112 | void __init xen_build_dynamic_phys_to_machine(void) |
113 | { | |
d451bb7a | 114 | unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list; |
8006ec3e | 115 | unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages); |
d5edbc1f | 116 | unsigned pfn; |
d451bb7a | 117 | |
8006ec3e | 118 | for(pfn = 0; pfn < max_pfn; pfn += P2M_ENTRIES_PER_PAGE) { |
d451bb7a JF |
119 | unsigned topidx = p2m_top_index(pfn); |
120 | ||
121 | p2m_top[topidx] = &mfn_list[pfn]; | |
122 | } | |
123 | } | |
124 | ||
125 | unsigned long get_phys_to_machine(unsigned long pfn) | |
126 | { | |
127 | unsigned topidx, idx; | |
128 | ||
8006ec3e JF |
129 | if (unlikely(pfn >= MAX_DOMAIN_PAGES)) |
130 | return INVALID_P2M_ENTRY; | |
131 | ||
d451bb7a | 132 | topidx = p2m_top_index(pfn); |
d451bb7a JF |
133 | idx = p2m_index(pfn); |
134 | return p2m_top[topidx][idx]; | |
135 | } | |
15ce6005 | 136 | EXPORT_SYMBOL_GPL(get_phys_to_machine); |
d451bb7a | 137 | |
d5edbc1f | 138 | static void alloc_p2m(unsigned long **pp, unsigned long *mfnp) |
d451bb7a JF |
139 | { |
140 | unsigned long *p; | |
141 | unsigned i; | |
142 | ||
143 | p = (void *)__get_free_page(GFP_KERNEL | __GFP_NOFAIL); | |
144 | BUG_ON(p == NULL); | |
145 | ||
146 | for(i = 0; i < P2M_ENTRIES_PER_PAGE; i++) | |
147 | p[i] = INVALID_P2M_ENTRY; | |
148 | ||
cf0923ea | 149 | if (cmpxchg(pp, p2m_missing, p) != p2m_missing) |
d451bb7a | 150 | free_page((unsigned long)p); |
d5edbc1f JF |
151 | else |
152 | *mfnp = virt_to_mfn(p); | |
d451bb7a JF |
153 | } |
154 | ||
155 | void set_phys_to_machine(unsigned long pfn, unsigned long mfn) | |
156 | { | |
157 | unsigned topidx, idx; | |
158 | ||
159 | if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) { | |
160 | BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY); | |
8006ec3e JF |
161 | return; |
162 | } | |
163 | ||
164 | if (unlikely(pfn >= MAX_DOMAIN_PAGES)) { | |
165 | BUG_ON(mfn != INVALID_P2M_ENTRY); | |
d451bb7a JF |
166 | return; |
167 | } | |
168 | ||
169 | topidx = p2m_top_index(pfn); | |
cf0923ea | 170 | if (p2m_top[topidx] == p2m_missing) { |
d451bb7a JF |
171 | /* no need to allocate a page to store an invalid entry */ |
172 | if (mfn == INVALID_P2M_ENTRY) | |
173 | return; | |
d5edbc1f | 174 | alloc_p2m(&p2m_top[topidx], &p2m_top_mfn[topidx]); |
d451bb7a JF |
175 | } |
176 | ||
177 | idx = p2m_index(pfn); | |
178 | p2m_top[topidx][idx] = mfn; | |
179 | } | |
180 | ||
ce803e70 | 181 | xmaddr_t arbitrary_virt_to_machine(void *vaddr) |
3b827c1b | 182 | { |
ce803e70 | 183 | unsigned long address = (unsigned long)vaddr; |
da7bfc50 | 184 | unsigned int level; |
f0646e43 | 185 | pte_t *pte = lookup_address(address, &level); |
de067814 | 186 | unsigned offset = address & ~PAGE_MASK; |
3b827c1b JF |
187 | |
188 | BUG_ON(pte == NULL); | |
189 | ||
ebd879e3 | 190 | return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset); |
3b827c1b JF |
191 | } |
192 | ||
193 | void make_lowmem_page_readonly(void *vaddr) | |
194 | { | |
195 | pte_t *pte, ptev; | |
196 | unsigned long address = (unsigned long)vaddr; | |
da7bfc50 | 197 | unsigned int level; |
3b827c1b | 198 | |
f0646e43 | 199 | pte = lookup_address(address, &level); |
3b827c1b JF |
200 | BUG_ON(pte == NULL); |
201 | ||
202 | ptev = pte_wrprotect(*pte); | |
203 | ||
204 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
205 | BUG(); | |
206 | } | |
207 | ||
208 | void make_lowmem_page_readwrite(void *vaddr) | |
209 | { | |
210 | pte_t *pte, ptev; | |
211 | unsigned long address = (unsigned long)vaddr; | |
da7bfc50 | 212 | unsigned int level; |
3b827c1b | 213 | |
f0646e43 | 214 | pte = lookup_address(address, &level); |
3b827c1b JF |
215 | BUG_ON(pte == NULL); |
216 | ||
217 | ptev = pte_mkwrite(*pte); | |
218 | ||
219 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
220 | BUG(); | |
221 | } | |
222 | ||
223 | ||
e2426cf8 JF |
224 | static bool page_pinned(void *ptr) |
225 | { | |
226 | struct page *page = virt_to_page(ptr); | |
227 | ||
228 | return PagePinned(page); | |
229 | } | |
230 | ||
400d3494 | 231 | static void extend_mmu_update(const struct mmu_update *update) |
3b827c1b | 232 | { |
d66bf8fc JF |
233 | struct multicall_space mcs; |
234 | struct mmu_update *u; | |
3b827c1b | 235 | |
400d3494 JF |
236 | mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u)); |
237 | ||
238 | if (mcs.mc != NULL) | |
239 | mcs.mc->args[1]++; | |
240 | else { | |
241 | mcs = __xen_mc_entry(sizeof(*u)); | |
242 | MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); | |
243 | } | |
d66bf8fc | 244 | |
d66bf8fc | 245 | u = mcs.args; |
400d3494 JF |
246 | *u = *update; |
247 | } | |
248 | ||
249 | void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val) | |
250 | { | |
251 | struct mmu_update u; | |
252 | ||
253 | preempt_disable(); | |
254 | ||
255 | xen_mc_batch(); | |
256 | ||
ce803e70 JF |
257 | /* ptr may be ioremapped for 64-bit pagetable setup */ |
258 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
400d3494 JF |
259 | u.val = pmd_val_ma(val); |
260 | extend_mmu_update(&u); | |
d66bf8fc JF |
261 | |
262 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
263 | ||
264 | preempt_enable(); | |
3b827c1b JF |
265 | } |
266 | ||
e2426cf8 JF |
267 | void xen_set_pmd(pmd_t *ptr, pmd_t val) |
268 | { | |
269 | /* If page is not pinned, we can just update the entry | |
270 | directly */ | |
271 | if (!page_pinned(ptr)) { | |
272 | *ptr = val; | |
273 | return; | |
274 | } | |
275 | ||
276 | xen_set_pmd_hyper(ptr, val); | |
277 | } | |
278 | ||
3b827c1b JF |
279 | /* |
280 | * Associate a virtual page frame with a given physical page frame | |
281 | * and protection flags for that frame. | |
282 | */ | |
283 | void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags) | |
284 | { | |
836fe2f2 | 285 | set_pte_vaddr(vaddr, mfn_pte(mfn, flags)); |
3b827c1b JF |
286 | } |
287 | ||
288 | void xen_set_pte_at(struct mm_struct *mm, unsigned long addr, | |
289 | pte_t *ptep, pte_t pteval) | |
290 | { | |
2bd50036 JF |
291 | /* updates to init_mm may be done without lock */ |
292 | if (mm == &init_mm) | |
293 | preempt_disable(); | |
294 | ||
d66bf8fc | 295 | if (mm == current->mm || mm == &init_mm) { |
8965c1c0 | 296 | if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) { |
d66bf8fc JF |
297 | struct multicall_space mcs; |
298 | mcs = xen_mc_entry(0); | |
299 | ||
300 | MULTI_update_va_mapping(mcs.mc, addr, pteval, 0); | |
301 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
2bd50036 | 302 | goto out; |
d66bf8fc JF |
303 | } else |
304 | if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0) | |
2bd50036 | 305 | goto out; |
d66bf8fc JF |
306 | } |
307 | xen_set_pte(ptep, pteval); | |
2bd50036 JF |
308 | |
309 | out: | |
310 | if (mm == &init_mm) | |
311 | preempt_enable(); | |
3b827c1b JF |
312 | } |
313 | ||
e57778a1 | 314 | pte_t xen_ptep_modify_prot_start(struct mm_struct *mm, unsigned long addr, pte_t *ptep) |
947a69c9 | 315 | { |
e57778a1 JF |
316 | /* Just return the pte as-is. We preserve the bits on commit */ |
317 | return *ptep; | |
318 | } | |
319 | ||
320 | void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr, | |
321 | pte_t *ptep, pte_t pte) | |
322 | { | |
400d3494 | 323 | struct mmu_update u; |
e57778a1 | 324 | |
400d3494 | 325 | xen_mc_batch(); |
947a69c9 | 326 | |
400d3494 JF |
327 | u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD; |
328 | u.val = pte_val_ma(pte); | |
329 | extend_mmu_update(&u); | |
947a69c9 | 330 | |
e57778a1 | 331 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
947a69c9 JF |
332 | } |
333 | ||
ebb9cfe2 JF |
334 | /* Assume pteval_t is equivalent to all the other *val_t types. */ |
335 | static pteval_t pte_mfn_to_pfn(pteval_t val) | |
947a69c9 | 336 | { |
ebb9cfe2 JF |
337 | if (val & _PAGE_PRESENT) { |
338 | unsigned long mfn = (val & PTE_MASK) >> PAGE_SHIFT; | |
339 | pteval_t flags = val & ~PTE_MASK; | |
d8355aca | 340 | val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags; |
ebb9cfe2 | 341 | } |
947a69c9 | 342 | |
ebb9cfe2 | 343 | return val; |
947a69c9 JF |
344 | } |
345 | ||
ebb9cfe2 | 346 | static pteval_t pte_pfn_to_mfn(pteval_t val) |
947a69c9 | 347 | { |
ebb9cfe2 JF |
348 | if (val & _PAGE_PRESENT) { |
349 | unsigned long pfn = (val & PTE_MASK) >> PAGE_SHIFT; | |
350 | pteval_t flags = val & ~PTE_MASK; | |
d8355aca | 351 | val = ((pteval_t)pfn_to_mfn(pfn) << PAGE_SHIFT) | flags; |
947a69c9 JF |
352 | } |
353 | ||
ebb9cfe2 | 354 | return val; |
947a69c9 JF |
355 | } |
356 | ||
ebb9cfe2 | 357 | pteval_t xen_pte_val(pte_t pte) |
947a69c9 | 358 | { |
ebb9cfe2 | 359 | return pte_mfn_to_pfn(pte.pte); |
947a69c9 | 360 | } |
947a69c9 | 361 | |
947a69c9 JF |
362 | pgdval_t xen_pgd_val(pgd_t pgd) |
363 | { | |
ebb9cfe2 | 364 | return pte_mfn_to_pfn(pgd.pgd); |
947a69c9 JF |
365 | } |
366 | ||
367 | pte_t xen_make_pte(pteval_t pte) | |
368 | { | |
ebb9cfe2 JF |
369 | pte = pte_pfn_to_mfn(pte); |
370 | return native_make_pte(pte); | |
947a69c9 JF |
371 | } |
372 | ||
373 | pgd_t xen_make_pgd(pgdval_t pgd) | |
374 | { | |
ebb9cfe2 JF |
375 | pgd = pte_pfn_to_mfn(pgd); |
376 | return native_make_pgd(pgd); | |
947a69c9 JF |
377 | } |
378 | ||
379 | pmdval_t xen_pmd_val(pmd_t pmd) | |
380 | { | |
ebb9cfe2 | 381 | return pte_mfn_to_pfn(pmd.pmd); |
947a69c9 | 382 | } |
28499143 | 383 | |
e2426cf8 | 384 | void xen_set_pud_hyper(pud_t *ptr, pud_t val) |
f4f97b3e | 385 | { |
400d3494 | 386 | struct mmu_update u; |
f4f97b3e | 387 | |
d66bf8fc JF |
388 | preempt_disable(); |
389 | ||
400d3494 JF |
390 | xen_mc_batch(); |
391 | ||
ce803e70 JF |
392 | /* ptr may be ioremapped for 64-bit pagetable setup */ |
393 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
400d3494 JF |
394 | u.val = pud_val_ma(val); |
395 | extend_mmu_update(&u); | |
d66bf8fc JF |
396 | |
397 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
398 | ||
399 | preempt_enable(); | |
f4f97b3e JF |
400 | } |
401 | ||
e2426cf8 JF |
402 | void xen_set_pud(pud_t *ptr, pud_t val) |
403 | { | |
404 | /* If page is not pinned, we can just update the entry | |
405 | directly */ | |
406 | if (!page_pinned(ptr)) { | |
407 | *ptr = val; | |
408 | return; | |
409 | } | |
410 | ||
411 | xen_set_pud_hyper(ptr, val); | |
412 | } | |
413 | ||
f4f97b3e JF |
414 | void xen_set_pte(pte_t *ptep, pte_t pte) |
415 | { | |
f6e58732 | 416 | #ifdef CONFIG_X86_PAE |
f4f97b3e JF |
417 | ptep->pte_high = pte.pte_high; |
418 | smp_wmb(); | |
419 | ptep->pte_low = pte.pte_low; | |
f6e58732 JF |
420 | #else |
421 | *ptep = pte; | |
422 | #endif | |
f4f97b3e JF |
423 | } |
424 | ||
f6e58732 | 425 | #ifdef CONFIG_X86_PAE |
3b827c1b JF |
426 | void xen_set_pte_atomic(pte_t *ptep, pte_t pte) |
427 | { | |
f6e58732 | 428 | set_64bit((u64 *)ptep, native_pte_val(pte)); |
3b827c1b JF |
429 | } |
430 | ||
431 | void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep) | |
432 | { | |
433 | ptep->pte_low = 0; | |
434 | smp_wmb(); /* make sure low gets written first */ | |
435 | ptep->pte_high = 0; | |
436 | } | |
437 | ||
438 | void xen_pmd_clear(pmd_t *pmdp) | |
439 | { | |
e2426cf8 | 440 | set_pmd(pmdp, __pmd(0)); |
3b827c1b | 441 | } |
f6e58732 | 442 | #endif /* CONFIG_X86_PAE */ |
3b827c1b | 443 | |
abf33038 | 444 | pmd_t xen_make_pmd(pmdval_t pmd) |
3b827c1b | 445 | { |
ebb9cfe2 | 446 | pmd = pte_pfn_to_mfn(pmd); |
947a69c9 | 447 | return native_make_pmd(pmd); |
3b827c1b | 448 | } |
3b827c1b | 449 | |
f6e58732 JF |
450 | #if PAGETABLE_LEVELS == 4 |
451 | pudval_t xen_pud_val(pud_t pud) | |
452 | { | |
453 | return pte_mfn_to_pfn(pud.pud); | |
454 | } | |
455 | ||
456 | pud_t xen_make_pud(pudval_t pud) | |
457 | { | |
458 | pud = pte_pfn_to_mfn(pud); | |
459 | ||
460 | return native_make_pud(pud); | |
461 | } | |
462 | ||
463 | void xen_set_pgd_hyper(pgd_t *ptr, pgd_t val) | |
464 | { | |
465 | struct mmu_update u; | |
466 | ||
467 | preempt_disable(); | |
468 | ||
469 | xen_mc_batch(); | |
470 | ||
471 | u.ptr = virt_to_machine(ptr).maddr; | |
472 | u.val = pgd_val_ma(val); | |
473 | extend_mmu_update(&u); | |
474 | ||
475 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
476 | ||
477 | preempt_enable(); | |
478 | } | |
479 | ||
480 | void xen_set_pgd(pgd_t *ptr, pgd_t val) | |
481 | { | |
482 | /* If page is not pinned, we can just update the entry | |
483 | directly */ | |
484 | if (!page_pinned(ptr)) { | |
485 | *ptr = val; | |
486 | return; | |
487 | } | |
488 | ||
489 | xen_set_pgd_hyper(ptr, val); | |
490 | } | |
491 | #endif /* PAGETABLE_LEVELS == 4 */ | |
492 | ||
f4f97b3e JF |
493 | /* |
494 | (Yet another) pagetable walker. This one is intended for pinning a | |
495 | pagetable. This means that it walks a pagetable and calls the | |
496 | callback function on each page it finds making up the page table, | |
497 | at every level. It walks the entire pagetable, but it only bothers | |
498 | pinning pte pages which are below pte_limit. In the normal case | |
499 | this will be TASK_SIZE, but at boot we need to pin up to | |
500 | FIXADDR_TOP. But the important bit is that we don't pin beyond | |
501 | there, because then we start getting into Xen's ptes. | |
502 | */ | |
74260714 | 503 | static int pgd_walk(pgd_t *pgd_base, int (*func)(struct page *, enum pt_level), |
f4f97b3e | 504 | unsigned long limit) |
3b827c1b JF |
505 | { |
506 | pgd_t *pgd = pgd_base; | |
f4f97b3e JF |
507 | int flush = 0; |
508 | unsigned long addr = 0; | |
509 | unsigned long pgd_next; | |
510 | ||
511 | BUG_ON(limit > FIXADDR_TOP); | |
3b827c1b JF |
512 | |
513 | if (xen_feature(XENFEAT_auto_translated_physmap)) | |
f4f97b3e JF |
514 | return 0; |
515 | ||
516 | for (; addr != FIXADDR_TOP; pgd++, addr = pgd_next) { | |
517 | pud_t *pud; | |
518 | unsigned long pud_limit, pud_next; | |
3b827c1b | 519 | |
f4f97b3e JF |
520 | pgd_next = pud_limit = pgd_addr_end(addr, FIXADDR_TOP); |
521 | ||
522 | if (!pgd_val(*pgd)) | |
3b827c1b | 523 | continue; |
f4f97b3e | 524 | |
3b827c1b JF |
525 | pud = pud_offset(pgd, 0); |
526 | ||
527 | if (PTRS_PER_PUD > 1) /* not folded */ | |
74260714 | 528 | flush |= (*func)(virt_to_page(pud), PT_PUD); |
f4f97b3e JF |
529 | |
530 | for (; addr != pud_limit; pud++, addr = pud_next) { | |
531 | pmd_t *pmd; | |
532 | unsigned long pmd_limit; | |
533 | ||
534 | pud_next = pud_addr_end(addr, pud_limit); | |
535 | ||
536 | if (pud_next < limit) | |
537 | pmd_limit = pud_next; | |
538 | else | |
539 | pmd_limit = limit; | |
3b827c1b | 540 | |
3b827c1b JF |
541 | if (pud_none(*pud)) |
542 | continue; | |
f4f97b3e | 543 | |
3b827c1b JF |
544 | pmd = pmd_offset(pud, 0); |
545 | ||
546 | if (PTRS_PER_PMD > 1) /* not folded */ | |
74260714 | 547 | flush |= (*func)(virt_to_page(pmd), PT_PMD); |
f4f97b3e JF |
548 | |
549 | for (; addr != pmd_limit; pmd++) { | |
550 | addr += (PAGE_SIZE * PTRS_PER_PTE); | |
551 | if ((pmd_limit-1) < (addr-1)) { | |
552 | addr = pmd_limit; | |
553 | break; | |
554 | } | |
3b827c1b | 555 | |
3b827c1b JF |
556 | if (pmd_none(*pmd)) |
557 | continue; | |
558 | ||
74260714 | 559 | flush |= (*func)(pmd_page(*pmd), PT_PTE); |
3b827c1b JF |
560 | } |
561 | } | |
562 | } | |
563 | ||
74260714 | 564 | flush |= (*func)(virt_to_page(pgd_base), PT_PGD); |
f4f97b3e JF |
565 | |
566 | return flush; | |
3b827c1b JF |
567 | } |
568 | ||
74260714 JF |
569 | static spinlock_t *lock_pte(struct page *page) |
570 | { | |
571 | spinlock_t *ptl = NULL; | |
572 | ||
573 | #if NR_CPUS >= CONFIG_SPLIT_PTLOCK_CPUS | |
574 | ptl = __pte_lockptr(page); | |
575 | spin_lock(ptl); | |
576 | #endif | |
577 | ||
578 | return ptl; | |
579 | } | |
580 | ||
581 | static void do_unlock(void *v) | |
582 | { | |
583 | spinlock_t *ptl = v; | |
584 | spin_unlock(ptl); | |
585 | } | |
586 | ||
587 | static void xen_do_pin(unsigned level, unsigned long pfn) | |
588 | { | |
589 | struct mmuext_op *op; | |
590 | struct multicall_space mcs; | |
591 | ||
592 | mcs = __xen_mc_entry(sizeof(*op)); | |
593 | op = mcs.args; | |
594 | op->cmd = level; | |
595 | op->arg1.mfn = pfn_to_mfn(pfn); | |
596 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
597 | } | |
598 | ||
599 | static int pin_page(struct page *page, enum pt_level level) | |
f4f97b3e | 600 | { |
d60cd46b | 601 | unsigned pgfl = TestSetPagePinned(page); |
f4f97b3e JF |
602 | int flush; |
603 | ||
604 | if (pgfl) | |
605 | flush = 0; /* already pinned */ | |
606 | else if (PageHighMem(page)) | |
607 | /* kmaps need flushing if we found an unpinned | |
608 | highpage */ | |
609 | flush = 1; | |
610 | else { | |
611 | void *pt = lowmem_page_address(page); | |
612 | unsigned long pfn = page_to_pfn(page); | |
613 | struct multicall_space mcs = __xen_mc_entry(0); | |
74260714 | 614 | spinlock_t *ptl; |
f4f97b3e JF |
615 | |
616 | flush = 0; | |
617 | ||
74260714 JF |
618 | ptl = NULL; |
619 | if (level == PT_PTE) | |
620 | ptl = lock_pte(page); | |
621 | ||
f4f97b3e JF |
622 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, |
623 | pfn_pte(pfn, PAGE_KERNEL_RO), | |
74260714 JF |
624 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
625 | ||
626 | if (level == PT_PTE) | |
627 | xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn); | |
628 | ||
629 | if (ptl) { | |
630 | /* Queue a deferred unlock for when this batch | |
631 | is completed. */ | |
632 | xen_mc_callback(do_unlock, ptl); | |
633 | } | |
f4f97b3e JF |
634 | } |
635 | ||
636 | return flush; | |
637 | } | |
3b827c1b | 638 | |
f4f97b3e JF |
639 | /* This is called just after a mm has been created, but it has not |
640 | been used yet. We need to make sure that its pagetable is all | |
641 | read-only, and can be pinned. */ | |
3b827c1b JF |
642 | void xen_pgd_pin(pgd_t *pgd) |
643 | { | |
f4f97b3e | 644 | xen_mc_batch(); |
3b827c1b | 645 | |
f87e4cac JF |
646 | if (pgd_walk(pgd, pin_page, TASK_SIZE)) { |
647 | /* re-enable interrupts for kmap_flush_unused */ | |
648 | xen_mc_issue(0); | |
f4f97b3e | 649 | kmap_flush_unused(); |
f87e4cac JF |
650 | xen_mc_batch(); |
651 | } | |
f4f97b3e | 652 | |
28499143 | 653 | xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd))); |
f4f97b3e | 654 | xen_mc_issue(0); |
3b827c1b JF |
655 | } |
656 | ||
0e91398f JF |
657 | /* |
658 | * On save, we need to pin all pagetables to make sure they get their | |
659 | * mfns turned into pfns. Search the list for any unpinned pgds and pin | |
660 | * them (unpinned pgds are not currently in use, probably because the | |
661 | * process is under construction or destruction). | |
662 | */ | |
663 | void xen_mm_pin_all(void) | |
664 | { | |
665 | unsigned long flags; | |
666 | struct page *page; | |
74260714 | 667 | |
0e91398f | 668 | spin_lock_irqsave(&pgd_lock, flags); |
f4f97b3e | 669 | |
0e91398f JF |
670 | list_for_each_entry(page, &pgd_list, lru) { |
671 | if (!PagePinned(page)) { | |
672 | xen_pgd_pin((pgd_t *)page_address(page)); | |
673 | SetPageSavePinned(page); | |
674 | } | |
675 | } | |
676 | ||
677 | spin_unlock_irqrestore(&pgd_lock, flags); | |
3b827c1b JF |
678 | } |
679 | ||
c1f2f09e EH |
680 | /* |
681 | * The init_mm pagetable is really pinned as soon as its created, but | |
682 | * that's before we have page structures to store the bits. So do all | |
683 | * the book-keeping now. | |
684 | */ | |
74260714 | 685 | static __init int mark_pinned(struct page *page, enum pt_level level) |
3b827c1b | 686 | { |
f4f97b3e JF |
687 | SetPagePinned(page); |
688 | return 0; | |
689 | } | |
3b827c1b | 690 | |
f4f97b3e JF |
691 | void __init xen_mark_init_mm_pinned(void) |
692 | { | |
693 | pgd_walk(init_mm.pgd, mark_pinned, FIXADDR_TOP); | |
694 | } | |
3b827c1b | 695 | |
74260714 | 696 | static int unpin_page(struct page *page, enum pt_level level) |
f4f97b3e | 697 | { |
d60cd46b | 698 | unsigned pgfl = TestClearPagePinned(page); |
3b827c1b | 699 | |
f4f97b3e JF |
700 | if (pgfl && !PageHighMem(page)) { |
701 | void *pt = lowmem_page_address(page); | |
702 | unsigned long pfn = page_to_pfn(page); | |
74260714 JF |
703 | spinlock_t *ptl = NULL; |
704 | struct multicall_space mcs; | |
705 | ||
706 | if (level == PT_PTE) { | |
707 | ptl = lock_pte(page); | |
708 | ||
709 | xen_do_pin(MMUEXT_UNPIN_TABLE, pfn); | |
710 | } | |
711 | ||
712 | mcs = __xen_mc_entry(0); | |
f4f97b3e JF |
713 | |
714 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, | |
715 | pfn_pte(pfn, PAGE_KERNEL), | |
74260714 JF |
716 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
717 | ||
718 | if (ptl) { | |
719 | /* unlock when batch completed */ | |
720 | xen_mc_callback(do_unlock, ptl); | |
721 | } | |
f4f97b3e JF |
722 | } |
723 | ||
724 | return 0; /* never need to flush on unpin */ | |
3b827c1b JF |
725 | } |
726 | ||
f4f97b3e JF |
727 | /* Release a pagetables pages back as normal RW */ |
728 | static void xen_pgd_unpin(pgd_t *pgd) | |
729 | { | |
f4f97b3e JF |
730 | xen_mc_batch(); |
731 | ||
74260714 | 732 | xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); |
f4f97b3e JF |
733 | |
734 | pgd_walk(pgd, unpin_page, TASK_SIZE); | |
735 | ||
736 | xen_mc_issue(0); | |
737 | } | |
3b827c1b | 738 | |
0e91398f JF |
739 | /* |
740 | * On resume, undo any pinning done at save, so that the rest of the | |
741 | * kernel doesn't see any unexpected pinned pagetables. | |
742 | */ | |
743 | void xen_mm_unpin_all(void) | |
744 | { | |
745 | unsigned long flags; | |
746 | struct page *page; | |
747 | ||
748 | spin_lock_irqsave(&pgd_lock, flags); | |
749 | ||
750 | list_for_each_entry(page, &pgd_list, lru) { | |
751 | if (PageSavePinned(page)) { | |
752 | BUG_ON(!PagePinned(page)); | |
753 | printk("unpinning pinned %p\n", page_address(page)); | |
754 | xen_pgd_unpin((pgd_t *)page_address(page)); | |
755 | ClearPageSavePinned(page); | |
756 | } | |
757 | } | |
758 | ||
759 | spin_unlock_irqrestore(&pgd_lock, flags); | |
760 | } | |
761 | ||
3b827c1b JF |
762 | void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next) |
763 | { | |
f4f97b3e | 764 | spin_lock(&next->page_table_lock); |
3b827c1b | 765 | xen_pgd_pin(next->pgd); |
f4f97b3e | 766 | spin_unlock(&next->page_table_lock); |
3b827c1b JF |
767 | } |
768 | ||
769 | void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm) | |
770 | { | |
f4f97b3e | 771 | spin_lock(&mm->page_table_lock); |
3b827c1b | 772 | xen_pgd_pin(mm->pgd); |
f4f97b3e | 773 | spin_unlock(&mm->page_table_lock); |
3b827c1b JF |
774 | } |
775 | ||
3b827c1b | 776 | |
f87e4cac JF |
777 | #ifdef CONFIG_SMP |
778 | /* Another cpu may still have their %cr3 pointing at the pagetable, so | |
779 | we need to repoint it somewhere else before we can unpin it. */ | |
780 | static void drop_other_mm_ref(void *info) | |
781 | { | |
782 | struct mm_struct *mm = info; | |
ce87b3d3 | 783 | struct mm_struct *active_mm; |
3b827c1b | 784 | |
ce87b3d3 JF |
785 | #ifdef CONFIG_X86_64 |
786 | active_mm = read_pda(active_mm); | |
787 | #else | |
788 | active_mm = __get_cpu_var(cpu_tlbstate).active_mm; | |
789 | #endif | |
790 | ||
791 | if (active_mm == mm) | |
f87e4cac | 792 | leave_mm(smp_processor_id()); |
9f79991d JF |
793 | |
794 | /* If this cpu still has a stale cr3 reference, then make sure | |
795 | it has been flushed. */ | |
796 | if (x86_read_percpu(xen_current_cr3) == __pa(mm->pgd)) { | |
797 | load_cr3(swapper_pg_dir); | |
798 | arch_flush_lazy_cpu_mode(); | |
799 | } | |
f87e4cac | 800 | } |
3b827c1b | 801 | |
f87e4cac JF |
802 | static void drop_mm_ref(struct mm_struct *mm) |
803 | { | |
9f79991d JF |
804 | cpumask_t mask; |
805 | unsigned cpu; | |
806 | ||
f87e4cac JF |
807 | if (current->active_mm == mm) { |
808 | if (current->mm == mm) | |
809 | load_cr3(swapper_pg_dir); | |
810 | else | |
811 | leave_mm(smp_processor_id()); | |
9f79991d JF |
812 | arch_flush_lazy_cpu_mode(); |
813 | } | |
814 | ||
815 | /* Get the "official" set of cpus referring to our pagetable. */ | |
816 | mask = mm->cpu_vm_mask; | |
817 | ||
818 | /* It's possible that a vcpu may have a stale reference to our | |
819 | cr3, because its in lazy mode, and it hasn't yet flushed | |
820 | its set of pending hypercalls yet. In this case, we can | |
821 | look at its actual current cr3 value, and force it to flush | |
822 | if needed. */ | |
823 | for_each_online_cpu(cpu) { | |
824 | if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd)) | |
825 | cpu_set(cpu, mask); | |
3b827c1b JF |
826 | } |
827 | ||
9f79991d | 828 | if (!cpus_empty(mask)) |
3b16cf87 | 829 | smp_call_function_mask(mask, drop_other_mm_ref, mm, 1); |
f87e4cac JF |
830 | } |
831 | #else | |
832 | static void drop_mm_ref(struct mm_struct *mm) | |
833 | { | |
834 | if (current->active_mm == mm) | |
835 | load_cr3(swapper_pg_dir); | |
836 | } | |
837 | #endif | |
838 | ||
839 | /* | |
840 | * While a process runs, Xen pins its pagetables, which means that the | |
841 | * hypervisor forces it to be read-only, and it controls all updates | |
842 | * to it. This means that all pagetable updates have to go via the | |
843 | * hypervisor, which is moderately expensive. | |
844 | * | |
845 | * Since we're pulling the pagetable down, we switch to use init_mm, | |
846 | * unpin old process pagetable and mark it all read-write, which | |
847 | * allows further operations on it to be simple memory accesses. | |
848 | * | |
849 | * The only subtle point is that another CPU may be still using the | |
850 | * pagetable because of lazy tlb flushing. This means we need need to | |
851 | * switch all CPUs off this pagetable before we can unpin it. | |
852 | */ | |
853 | void xen_exit_mmap(struct mm_struct *mm) | |
854 | { | |
855 | get_cpu(); /* make sure we don't move around */ | |
856 | drop_mm_ref(mm); | |
857 | put_cpu(); | |
3b827c1b | 858 | |
f120f13e | 859 | spin_lock(&mm->page_table_lock); |
df912ea4 JF |
860 | |
861 | /* pgd may not be pinned in the error exit path of execve */ | |
e2426cf8 | 862 | if (page_pinned(mm->pgd)) |
df912ea4 | 863 | xen_pgd_unpin(mm->pgd); |
74260714 | 864 | |
f120f13e | 865 | spin_unlock(&mm->page_table_lock); |
3b827c1b | 866 | } |