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> |
994025ca | 43 | #include <linux/debugfs.h> |
3b827c1b | 44 | #include <linux/bug.h> |
3b827c1b JF |
45 | |
46 | #include <asm/pgtable.h> | |
47 | #include <asm/tlbflush.h> | |
5deb30d1 | 48 | #include <asm/fixmap.h> |
3b827c1b | 49 | #include <asm/mmu_context.h> |
319f3ba5 | 50 | #include <asm/setup.h> |
f4f97b3e | 51 | #include <asm/paravirt.h> |
cbcd79c2 | 52 | #include <asm/linkage.h> |
3b827c1b JF |
53 | |
54 | #include <asm/xen/hypercall.h> | |
f4f97b3e | 55 | #include <asm/xen/hypervisor.h> |
3b827c1b JF |
56 | |
57 | #include <xen/page.h> | |
58 | #include <xen/interface/xen.h> | |
319f3ba5 JF |
59 | #include <xen/interface/version.h> |
60 | #include <xen/hvc-console.h> | |
3b827c1b | 61 | |
f4f97b3e | 62 | #include "multicalls.h" |
3b827c1b | 63 | #include "mmu.h" |
994025ca JF |
64 | #include "debugfs.h" |
65 | ||
66 | #define MMU_UPDATE_HISTO 30 | |
67 | ||
68 | #ifdef CONFIG_XEN_DEBUG_FS | |
69 | ||
70 | static struct { | |
71 | u32 pgd_update; | |
72 | u32 pgd_update_pinned; | |
73 | u32 pgd_update_batched; | |
74 | ||
75 | u32 pud_update; | |
76 | u32 pud_update_pinned; | |
77 | u32 pud_update_batched; | |
78 | ||
79 | u32 pmd_update; | |
80 | u32 pmd_update_pinned; | |
81 | u32 pmd_update_batched; | |
82 | ||
83 | u32 pte_update; | |
84 | u32 pte_update_pinned; | |
85 | u32 pte_update_batched; | |
86 | ||
87 | u32 mmu_update; | |
88 | u32 mmu_update_extended; | |
89 | u32 mmu_update_histo[MMU_UPDATE_HISTO]; | |
90 | ||
91 | u32 prot_commit; | |
92 | u32 prot_commit_batched; | |
93 | ||
94 | u32 set_pte_at; | |
95 | u32 set_pte_at_batched; | |
96 | u32 set_pte_at_pinned; | |
97 | u32 set_pte_at_current; | |
98 | u32 set_pte_at_kernel; | |
99 | } mmu_stats; | |
100 | ||
101 | static u8 zero_stats; | |
102 | ||
103 | static inline void check_zero(void) | |
104 | { | |
105 | if (unlikely(zero_stats)) { | |
106 | memset(&mmu_stats, 0, sizeof(mmu_stats)); | |
107 | zero_stats = 0; | |
108 | } | |
109 | } | |
110 | ||
111 | #define ADD_STATS(elem, val) \ | |
112 | do { check_zero(); mmu_stats.elem += (val); } while(0) | |
113 | ||
114 | #else /* !CONFIG_XEN_DEBUG_FS */ | |
115 | ||
116 | #define ADD_STATS(elem, val) do { (void)(val); } while(0) | |
117 | ||
118 | #endif /* CONFIG_XEN_DEBUG_FS */ | |
3b827c1b | 119 | |
319f3ba5 JF |
120 | |
121 | /* | |
122 | * Identity map, in addition to plain kernel map. This needs to be | |
123 | * large enough to allocate page table pages to allocate the rest. | |
124 | * Each page can map 2MB. | |
125 | */ | |
126 | static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss; | |
127 | ||
128 | #ifdef CONFIG_X86_64 | |
129 | /* l3 pud for userspace vsyscall mapping */ | |
130 | static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss; | |
131 | #endif /* CONFIG_X86_64 */ | |
132 | ||
133 | /* | |
134 | * Note about cr3 (pagetable base) values: | |
135 | * | |
136 | * xen_cr3 contains the current logical cr3 value; it contains the | |
137 | * last set cr3. This may not be the current effective cr3, because | |
138 | * its update may be being lazily deferred. However, a vcpu looking | |
139 | * at its own cr3 can use this value knowing that it everything will | |
140 | * be self-consistent. | |
141 | * | |
142 | * xen_current_cr3 contains the actual vcpu cr3; it is set once the | |
143 | * hypercall to set the vcpu cr3 is complete (so it may be a little | |
144 | * out of date, but it will never be set early). If one vcpu is | |
145 | * looking at another vcpu's cr3 value, it should use this variable. | |
146 | */ | |
147 | DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */ | |
148 | DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */ | |
149 | ||
150 | ||
d6182fbf JF |
151 | /* |
152 | * Just beyond the highest usermode address. STACK_TOP_MAX has a | |
153 | * redzone above it, so round it up to a PGD boundary. | |
154 | */ | |
155 | #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK) | |
156 | ||
157 | ||
d451bb7a | 158 | #define P2M_ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(unsigned long)) |
cf0923ea | 159 | #define TOP_ENTRIES (MAX_DOMAIN_PAGES / P2M_ENTRIES_PER_PAGE) |
d451bb7a | 160 | |
cf0923ea | 161 | /* Placeholder for holes in the address space */ |
cbcd79c2 | 162 | static unsigned long p2m_missing[P2M_ENTRIES_PER_PAGE] __page_aligned_data = |
cf0923ea JF |
163 | { [ 0 ... P2M_ENTRIES_PER_PAGE-1 ] = ~0UL }; |
164 | ||
165 | /* Array of pointers to pages containing p2m entries */ | |
cbcd79c2 | 166 | static unsigned long *p2m_top[TOP_ENTRIES] __page_aligned_data = |
cf0923ea | 167 | { [ 0 ... TOP_ENTRIES - 1] = &p2m_missing[0] }; |
d451bb7a | 168 | |
d5edbc1f | 169 | /* Arrays of p2m arrays expressed in mfns used for save/restore */ |
cbcd79c2 | 170 | static unsigned long p2m_top_mfn[TOP_ENTRIES] __page_aligned_bss; |
d5edbc1f | 171 | |
cbcd79c2 JF |
172 | static unsigned long p2m_top_mfn_list[TOP_ENTRIES / P2M_ENTRIES_PER_PAGE] |
173 | __page_aligned_bss; | |
d5edbc1f | 174 | |
d451bb7a JF |
175 | static inline unsigned p2m_top_index(unsigned long pfn) |
176 | { | |
8006ec3e | 177 | BUG_ON(pfn >= MAX_DOMAIN_PAGES); |
d451bb7a JF |
178 | return pfn / P2M_ENTRIES_PER_PAGE; |
179 | } | |
180 | ||
181 | static inline unsigned p2m_index(unsigned long pfn) | |
182 | { | |
183 | return pfn % P2M_ENTRIES_PER_PAGE; | |
184 | } | |
185 | ||
d5edbc1f JF |
186 | /* Build the parallel p2m_top_mfn structures */ |
187 | void xen_setup_mfn_list_list(void) | |
188 | { | |
189 | unsigned pfn, idx; | |
190 | ||
f63c2f24 | 191 | for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn += P2M_ENTRIES_PER_PAGE) { |
d5edbc1f JF |
192 | unsigned topidx = p2m_top_index(pfn); |
193 | ||
194 | p2m_top_mfn[topidx] = virt_to_mfn(p2m_top[topidx]); | |
195 | } | |
196 | ||
f63c2f24 | 197 | for (idx = 0; idx < ARRAY_SIZE(p2m_top_mfn_list); idx++) { |
d5edbc1f JF |
198 | unsigned topidx = idx * P2M_ENTRIES_PER_PAGE; |
199 | p2m_top_mfn_list[idx] = virt_to_mfn(&p2m_top_mfn[topidx]); | |
200 | } | |
201 | ||
202 | BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); | |
203 | ||
204 | HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list = | |
205 | virt_to_mfn(p2m_top_mfn_list); | |
206 | HYPERVISOR_shared_info->arch.max_pfn = xen_start_info->nr_pages; | |
207 | } | |
208 | ||
209 | /* Set up p2m_top to point to the domain-builder provided p2m pages */ | |
d451bb7a JF |
210 | void __init xen_build_dynamic_phys_to_machine(void) |
211 | { | |
d451bb7a | 212 | unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list; |
8006ec3e | 213 | unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages); |
d5edbc1f | 214 | unsigned pfn; |
d451bb7a | 215 | |
f63c2f24 | 216 | for (pfn = 0; pfn < max_pfn; pfn += P2M_ENTRIES_PER_PAGE) { |
d451bb7a JF |
217 | unsigned topidx = p2m_top_index(pfn); |
218 | ||
219 | p2m_top[topidx] = &mfn_list[pfn]; | |
220 | } | |
221 | } | |
222 | ||
223 | unsigned long get_phys_to_machine(unsigned long pfn) | |
224 | { | |
225 | unsigned topidx, idx; | |
226 | ||
8006ec3e JF |
227 | if (unlikely(pfn >= MAX_DOMAIN_PAGES)) |
228 | return INVALID_P2M_ENTRY; | |
229 | ||
d451bb7a | 230 | topidx = p2m_top_index(pfn); |
d451bb7a JF |
231 | idx = p2m_index(pfn); |
232 | return p2m_top[topidx][idx]; | |
233 | } | |
15ce6005 | 234 | EXPORT_SYMBOL_GPL(get_phys_to_machine); |
d451bb7a | 235 | |
d5edbc1f | 236 | static void alloc_p2m(unsigned long **pp, unsigned long *mfnp) |
d451bb7a JF |
237 | { |
238 | unsigned long *p; | |
239 | unsigned i; | |
240 | ||
241 | p = (void *)__get_free_page(GFP_KERNEL | __GFP_NOFAIL); | |
242 | BUG_ON(p == NULL); | |
243 | ||
f63c2f24 | 244 | for (i = 0; i < P2M_ENTRIES_PER_PAGE; i++) |
d451bb7a JF |
245 | p[i] = INVALID_P2M_ENTRY; |
246 | ||
cf0923ea | 247 | if (cmpxchg(pp, p2m_missing, p) != p2m_missing) |
d451bb7a | 248 | free_page((unsigned long)p); |
d5edbc1f JF |
249 | else |
250 | *mfnp = virt_to_mfn(p); | |
d451bb7a JF |
251 | } |
252 | ||
253 | void set_phys_to_machine(unsigned long pfn, unsigned long mfn) | |
254 | { | |
255 | unsigned topidx, idx; | |
256 | ||
257 | if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) { | |
258 | BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY); | |
8006ec3e JF |
259 | return; |
260 | } | |
261 | ||
262 | if (unlikely(pfn >= MAX_DOMAIN_PAGES)) { | |
263 | BUG_ON(mfn != INVALID_P2M_ENTRY); | |
d451bb7a JF |
264 | return; |
265 | } | |
266 | ||
267 | topidx = p2m_top_index(pfn); | |
cf0923ea | 268 | if (p2m_top[topidx] == p2m_missing) { |
d451bb7a JF |
269 | /* no need to allocate a page to store an invalid entry */ |
270 | if (mfn == INVALID_P2M_ENTRY) | |
271 | return; | |
d5edbc1f | 272 | alloc_p2m(&p2m_top[topidx], &p2m_top_mfn[topidx]); |
d451bb7a JF |
273 | } |
274 | ||
275 | idx = p2m_index(pfn); | |
276 | p2m_top[topidx][idx] = mfn; | |
277 | } | |
278 | ||
9976b39b JF |
279 | unsigned long arbitrary_virt_to_mfn(void *vaddr) |
280 | { | |
281 | xmaddr_t maddr = arbitrary_virt_to_machine(vaddr); | |
282 | ||
283 | return PFN_DOWN(maddr.maddr); | |
284 | } | |
285 | ||
ce803e70 | 286 | xmaddr_t arbitrary_virt_to_machine(void *vaddr) |
3b827c1b | 287 | { |
ce803e70 | 288 | unsigned long address = (unsigned long)vaddr; |
da7bfc50 | 289 | unsigned int level; |
9f32d21c CL |
290 | pte_t *pte; |
291 | unsigned offset; | |
3b827c1b | 292 | |
9f32d21c CL |
293 | /* |
294 | * if the PFN is in the linear mapped vaddr range, we can just use | |
295 | * the (quick) virt_to_machine() p2m lookup | |
296 | */ | |
297 | if (virt_addr_valid(vaddr)) | |
298 | return virt_to_machine(vaddr); | |
299 | ||
300 | /* otherwise we have to do a (slower) full page-table walk */ | |
3b827c1b | 301 | |
9f32d21c CL |
302 | pte = lookup_address(address, &level); |
303 | BUG_ON(pte == NULL); | |
304 | offset = address & ~PAGE_MASK; | |
ebd879e3 | 305 | return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset); |
3b827c1b JF |
306 | } |
307 | ||
308 | void make_lowmem_page_readonly(void *vaddr) | |
309 | { | |
310 | pte_t *pte, ptev; | |
311 | unsigned long address = (unsigned long)vaddr; | |
da7bfc50 | 312 | unsigned int level; |
3b827c1b | 313 | |
f0646e43 | 314 | pte = lookup_address(address, &level); |
3b827c1b JF |
315 | BUG_ON(pte == NULL); |
316 | ||
317 | ptev = pte_wrprotect(*pte); | |
318 | ||
319 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
320 | BUG(); | |
321 | } | |
322 | ||
323 | void make_lowmem_page_readwrite(void *vaddr) | |
324 | { | |
325 | pte_t *pte, ptev; | |
326 | unsigned long address = (unsigned long)vaddr; | |
da7bfc50 | 327 | unsigned int level; |
3b827c1b | 328 | |
f0646e43 | 329 | pte = lookup_address(address, &level); |
3b827c1b JF |
330 | BUG_ON(pte == NULL); |
331 | ||
332 | ptev = pte_mkwrite(*pte); | |
333 | ||
334 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
335 | BUG(); | |
336 | } | |
337 | ||
338 | ||
7708ad64 | 339 | static bool xen_page_pinned(void *ptr) |
e2426cf8 JF |
340 | { |
341 | struct page *page = virt_to_page(ptr); | |
342 | ||
343 | return PagePinned(page); | |
344 | } | |
345 | ||
7708ad64 | 346 | static void xen_extend_mmu_update(const struct mmu_update *update) |
3b827c1b | 347 | { |
d66bf8fc JF |
348 | struct multicall_space mcs; |
349 | struct mmu_update *u; | |
3b827c1b | 350 | |
400d3494 JF |
351 | mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u)); |
352 | ||
994025ca JF |
353 | if (mcs.mc != NULL) { |
354 | ADD_STATS(mmu_update_extended, 1); | |
355 | ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1); | |
356 | ||
400d3494 | 357 | mcs.mc->args[1]++; |
994025ca JF |
358 | |
359 | if (mcs.mc->args[1] < MMU_UPDATE_HISTO) | |
360 | ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1); | |
361 | else | |
362 | ADD_STATS(mmu_update_histo[0], 1); | |
363 | } else { | |
364 | ADD_STATS(mmu_update, 1); | |
400d3494 JF |
365 | mcs = __xen_mc_entry(sizeof(*u)); |
366 | MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); | |
994025ca | 367 | ADD_STATS(mmu_update_histo[1], 1); |
400d3494 | 368 | } |
d66bf8fc | 369 | |
d66bf8fc | 370 | u = mcs.args; |
400d3494 JF |
371 | *u = *update; |
372 | } | |
373 | ||
374 | void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val) | |
375 | { | |
376 | struct mmu_update u; | |
377 | ||
378 | preempt_disable(); | |
379 | ||
380 | xen_mc_batch(); | |
381 | ||
ce803e70 JF |
382 | /* ptr may be ioremapped for 64-bit pagetable setup */ |
383 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
400d3494 | 384 | u.val = pmd_val_ma(val); |
7708ad64 | 385 | xen_extend_mmu_update(&u); |
d66bf8fc | 386 | |
994025ca JF |
387 | ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); |
388 | ||
d66bf8fc JF |
389 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
390 | ||
391 | preempt_enable(); | |
3b827c1b JF |
392 | } |
393 | ||
e2426cf8 JF |
394 | void xen_set_pmd(pmd_t *ptr, pmd_t val) |
395 | { | |
994025ca JF |
396 | ADD_STATS(pmd_update, 1); |
397 | ||
e2426cf8 JF |
398 | /* If page is not pinned, we can just update the entry |
399 | directly */ | |
7708ad64 | 400 | if (!xen_page_pinned(ptr)) { |
e2426cf8 JF |
401 | *ptr = val; |
402 | return; | |
403 | } | |
404 | ||
994025ca JF |
405 | ADD_STATS(pmd_update_pinned, 1); |
406 | ||
e2426cf8 JF |
407 | xen_set_pmd_hyper(ptr, val); |
408 | } | |
409 | ||
3b827c1b JF |
410 | /* |
411 | * Associate a virtual page frame with a given physical page frame | |
412 | * and protection flags for that frame. | |
413 | */ | |
414 | void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags) | |
415 | { | |
836fe2f2 | 416 | set_pte_vaddr(vaddr, mfn_pte(mfn, flags)); |
3b827c1b JF |
417 | } |
418 | ||
419 | void xen_set_pte_at(struct mm_struct *mm, unsigned long addr, | |
420 | pte_t *ptep, pte_t pteval) | |
421 | { | |
994025ca JF |
422 | ADD_STATS(set_pte_at, 1); |
423 | // ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep)); | |
424 | ADD_STATS(set_pte_at_current, mm == current->mm); | |
425 | ADD_STATS(set_pte_at_kernel, mm == &init_mm); | |
426 | ||
d66bf8fc | 427 | if (mm == current->mm || mm == &init_mm) { |
8965c1c0 | 428 | if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) { |
d66bf8fc JF |
429 | struct multicall_space mcs; |
430 | mcs = xen_mc_entry(0); | |
431 | ||
432 | MULTI_update_va_mapping(mcs.mc, addr, pteval, 0); | |
994025ca | 433 | ADD_STATS(set_pte_at_batched, 1); |
d66bf8fc | 434 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
2bd50036 | 435 | goto out; |
d66bf8fc JF |
436 | } else |
437 | if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0) | |
2bd50036 | 438 | goto out; |
d66bf8fc JF |
439 | } |
440 | xen_set_pte(ptep, pteval); | |
2bd50036 | 441 | |
2829b449 | 442 | out: return; |
3b827c1b JF |
443 | } |
444 | ||
f63c2f24 T |
445 | pte_t xen_ptep_modify_prot_start(struct mm_struct *mm, |
446 | unsigned long addr, pte_t *ptep) | |
947a69c9 | 447 | { |
e57778a1 JF |
448 | /* Just return the pte as-is. We preserve the bits on commit */ |
449 | return *ptep; | |
450 | } | |
451 | ||
452 | void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr, | |
453 | pte_t *ptep, pte_t pte) | |
454 | { | |
400d3494 | 455 | struct mmu_update u; |
e57778a1 | 456 | |
400d3494 | 457 | xen_mc_batch(); |
947a69c9 | 458 | |
9f32d21c | 459 | u.ptr = arbitrary_virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD; |
400d3494 | 460 | u.val = pte_val_ma(pte); |
7708ad64 | 461 | xen_extend_mmu_update(&u); |
947a69c9 | 462 | |
994025ca JF |
463 | ADD_STATS(prot_commit, 1); |
464 | ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); | |
465 | ||
e57778a1 | 466 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
947a69c9 JF |
467 | } |
468 | ||
ebb9cfe2 JF |
469 | /* Assume pteval_t is equivalent to all the other *val_t types. */ |
470 | static pteval_t pte_mfn_to_pfn(pteval_t val) | |
947a69c9 | 471 | { |
ebb9cfe2 | 472 | if (val & _PAGE_PRESENT) { |
59438c9f | 473 | unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; |
77be1fab | 474 | pteval_t flags = val & PTE_FLAGS_MASK; |
d8355aca | 475 | val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags; |
ebb9cfe2 | 476 | } |
947a69c9 | 477 | |
ebb9cfe2 | 478 | return val; |
947a69c9 JF |
479 | } |
480 | ||
ebb9cfe2 | 481 | static pteval_t pte_pfn_to_mfn(pteval_t val) |
947a69c9 | 482 | { |
ebb9cfe2 | 483 | if (val & _PAGE_PRESENT) { |
59438c9f | 484 | unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; |
77be1fab | 485 | pteval_t flags = val & PTE_FLAGS_MASK; |
d8355aca | 486 | val = ((pteval_t)pfn_to_mfn(pfn) << PAGE_SHIFT) | flags; |
947a69c9 JF |
487 | } |
488 | ||
ebb9cfe2 | 489 | return val; |
947a69c9 JF |
490 | } |
491 | ||
ebb9cfe2 | 492 | pteval_t xen_pte_val(pte_t pte) |
947a69c9 | 493 | { |
ebb9cfe2 | 494 | return pte_mfn_to_pfn(pte.pte); |
947a69c9 | 495 | } |
da5de7c2 | 496 | PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val); |
947a69c9 | 497 | |
947a69c9 JF |
498 | pgdval_t xen_pgd_val(pgd_t pgd) |
499 | { | |
ebb9cfe2 | 500 | return pte_mfn_to_pfn(pgd.pgd); |
947a69c9 | 501 | } |
da5de7c2 | 502 | PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val); |
947a69c9 JF |
503 | |
504 | pte_t xen_make_pte(pteval_t pte) | |
505 | { | |
ebb9cfe2 JF |
506 | pte = pte_pfn_to_mfn(pte); |
507 | return native_make_pte(pte); | |
947a69c9 | 508 | } |
da5de7c2 | 509 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte); |
947a69c9 JF |
510 | |
511 | pgd_t xen_make_pgd(pgdval_t pgd) | |
512 | { | |
ebb9cfe2 JF |
513 | pgd = pte_pfn_to_mfn(pgd); |
514 | return native_make_pgd(pgd); | |
947a69c9 | 515 | } |
da5de7c2 | 516 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd); |
947a69c9 JF |
517 | |
518 | pmdval_t xen_pmd_val(pmd_t pmd) | |
519 | { | |
ebb9cfe2 | 520 | return pte_mfn_to_pfn(pmd.pmd); |
947a69c9 | 521 | } |
da5de7c2 | 522 | PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val); |
28499143 | 523 | |
e2426cf8 | 524 | void xen_set_pud_hyper(pud_t *ptr, pud_t val) |
f4f97b3e | 525 | { |
400d3494 | 526 | struct mmu_update u; |
f4f97b3e | 527 | |
d66bf8fc JF |
528 | preempt_disable(); |
529 | ||
400d3494 JF |
530 | xen_mc_batch(); |
531 | ||
ce803e70 JF |
532 | /* ptr may be ioremapped for 64-bit pagetable setup */ |
533 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
400d3494 | 534 | u.val = pud_val_ma(val); |
7708ad64 | 535 | xen_extend_mmu_update(&u); |
d66bf8fc | 536 | |
994025ca JF |
537 | ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); |
538 | ||
d66bf8fc JF |
539 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
540 | ||
541 | preempt_enable(); | |
f4f97b3e JF |
542 | } |
543 | ||
e2426cf8 JF |
544 | void xen_set_pud(pud_t *ptr, pud_t val) |
545 | { | |
994025ca JF |
546 | ADD_STATS(pud_update, 1); |
547 | ||
e2426cf8 JF |
548 | /* If page is not pinned, we can just update the entry |
549 | directly */ | |
7708ad64 | 550 | if (!xen_page_pinned(ptr)) { |
e2426cf8 JF |
551 | *ptr = val; |
552 | return; | |
553 | } | |
554 | ||
994025ca JF |
555 | ADD_STATS(pud_update_pinned, 1); |
556 | ||
e2426cf8 JF |
557 | xen_set_pud_hyper(ptr, val); |
558 | } | |
559 | ||
f4f97b3e JF |
560 | void xen_set_pte(pte_t *ptep, pte_t pte) |
561 | { | |
994025ca JF |
562 | ADD_STATS(pte_update, 1); |
563 | // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep)); | |
564 | ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); | |
565 | ||
f6e58732 | 566 | #ifdef CONFIG_X86_PAE |
f4f97b3e JF |
567 | ptep->pte_high = pte.pte_high; |
568 | smp_wmb(); | |
569 | ptep->pte_low = pte.pte_low; | |
f6e58732 JF |
570 | #else |
571 | *ptep = pte; | |
572 | #endif | |
f4f97b3e JF |
573 | } |
574 | ||
f6e58732 | 575 | #ifdef CONFIG_X86_PAE |
3b827c1b JF |
576 | void xen_set_pte_atomic(pte_t *ptep, pte_t pte) |
577 | { | |
f6e58732 | 578 | set_64bit((u64 *)ptep, native_pte_val(pte)); |
3b827c1b JF |
579 | } |
580 | ||
581 | void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep) | |
582 | { | |
583 | ptep->pte_low = 0; | |
584 | smp_wmb(); /* make sure low gets written first */ | |
585 | ptep->pte_high = 0; | |
586 | } | |
587 | ||
588 | void xen_pmd_clear(pmd_t *pmdp) | |
589 | { | |
e2426cf8 | 590 | set_pmd(pmdp, __pmd(0)); |
3b827c1b | 591 | } |
f6e58732 | 592 | #endif /* CONFIG_X86_PAE */ |
3b827c1b | 593 | |
abf33038 | 594 | pmd_t xen_make_pmd(pmdval_t pmd) |
3b827c1b | 595 | { |
ebb9cfe2 | 596 | pmd = pte_pfn_to_mfn(pmd); |
947a69c9 | 597 | return native_make_pmd(pmd); |
3b827c1b | 598 | } |
da5de7c2 | 599 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd); |
3b827c1b | 600 | |
f6e58732 JF |
601 | #if PAGETABLE_LEVELS == 4 |
602 | pudval_t xen_pud_val(pud_t pud) | |
603 | { | |
604 | return pte_mfn_to_pfn(pud.pud); | |
605 | } | |
da5de7c2 | 606 | PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val); |
f6e58732 JF |
607 | |
608 | pud_t xen_make_pud(pudval_t pud) | |
609 | { | |
610 | pud = pte_pfn_to_mfn(pud); | |
611 | ||
612 | return native_make_pud(pud); | |
613 | } | |
da5de7c2 | 614 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud); |
f6e58732 | 615 | |
d6182fbf | 616 | pgd_t *xen_get_user_pgd(pgd_t *pgd) |
f6e58732 | 617 | { |
d6182fbf JF |
618 | pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK); |
619 | unsigned offset = pgd - pgd_page; | |
620 | pgd_t *user_ptr = NULL; | |
f6e58732 | 621 | |
d6182fbf JF |
622 | if (offset < pgd_index(USER_LIMIT)) { |
623 | struct page *page = virt_to_page(pgd_page); | |
624 | user_ptr = (pgd_t *)page->private; | |
625 | if (user_ptr) | |
626 | user_ptr += offset; | |
627 | } | |
f6e58732 | 628 | |
d6182fbf JF |
629 | return user_ptr; |
630 | } | |
631 | ||
632 | static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val) | |
633 | { | |
634 | struct mmu_update u; | |
f6e58732 JF |
635 | |
636 | u.ptr = virt_to_machine(ptr).maddr; | |
637 | u.val = pgd_val_ma(val); | |
7708ad64 | 638 | xen_extend_mmu_update(&u); |
d6182fbf JF |
639 | } |
640 | ||
641 | /* | |
642 | * Raw hypercall-based set_pgd, intended for in early boot before | |
643 | * there's a page structure. This implies: | |
644 | * 1. The only existing pagetable is the kernel's | |
645 | * 2. It is always pinned | |
646 | * 3. It has no user pagetable attached to it | |
647 | */ | |
648 | void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val) | |
649 | { | |
650 | preempt_disable(); | |
651 | ||
652 | xen_mc_batch(); | |
653 | ||
654 | __xen_set_pgd_hyper(ptr, val); | |
f6e58732 JF |
655 | |
656 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
657 | ||
658 | preempt_enable(); | |
659 | } | |
660 | ||
661 | void xen_set_pgd(pgd_t *ptr, pgd_t val) | |
662 | { | |
d6182fbf JF |
663 | pgd_t *user_ptr = xen_get_user_pgd(ptr); |
664 | ||
994025ca JF |
665 | ADD_STATS(pgd_update, 1); |
666 | ||
f6e58732 JF |
667 | /* If page is not pinned, we can just update the entry |
668 | directly */ | |
7708ad64 | 669 | if (!xen_page_pinned(ptr)) { |
f6e58732 | 670 | *ptr = val; |
d6182fbf | 671 | if (user_ptr) { |
7708ad64 | 672 | WARN_ON(xen_page_pinned(user_ptr)); |
d6182fbf JF |
673 | *user_ptr = val; |
674 | } | |
f6e58732 JF |
675 | return; |
676 | } | |
677 | ||
994025ca JF |
678 | ADD_STATS(pgd_update_pinned, 1); |
679 | ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); | |
680 | ||
d6182fbf JF |
681 | /* If it's pinned, then we can at least batch the kernel and |
682 | user updates together. */ | |
683 | xen_mc_batch(); | |
684 | ||
685 | __xen_set_pgd_hyper(ptr, val); | |
686 | if (user_ptr) | |
687 | __xen_set_pgd_hyper(user_ptr, val); | |
688 | ||
689 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
f6e58732 JF |
690 | } |
691 | #endif /* PAGETABLE_LEVELS == 4 */ | |
692 | ||
f4f97b3e | 693 | /* |
5deb30d1 JF |
694 | * (Yet another) pagetable walker. This one is intended for pinning a |
695 | * pagetable. This means that it walks a pagetable and calls the | |
696 | * callback function on each page it finds making up the page table, | |
697 | * at every level. It walks the entire pagetable, but it only bothers | |
698 | * pinning pte pages which are below limit. In the normal case this | |
699 | * will be STACK_TOP_MAX, but at boot we need to pin up to | |
700 | * FIXADDR_TOP. | |
701 | * | |
702 | * For 32-bit the important bit is that we don't pin beyond there, | |
703 | * because then we start getting into Xen's ptes. | |
704 | * | |
705 | * For 64-bit, we must skip the Xen hole in the middle of the address | |
706 | * space, just after the big x86-64 virtual hole. | |
707 | */ | |
86bbc2c2 IC |
708 | static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd, |
709 | int (*func)(struct mm_struct *mm, struct page *, | |
710 | enum pt_level), | |
711 | unsigned long limit) | |
3b827c1b | 712 | { |
f4f97b3e | 713 | int flush = 0; |
5deb30d1 JF |
714 | unsigned hole_low, hole_high; |
715 | unsigned pgdidx_limit, pudidx_limit, pmdidx_limit; | |
716 | unsigned pgdidx, pudidx, pmdidx; | |
f4f97b3e | 717 | |
5deb30d1 JF |
718 | /* The limit is the last byte to be touched */ |
719 | limit--; | |
720 | BUG_ON(limit >= FIXADDR_TOP); | |
3b827c1b JF |
721 | |
722 | if (xen_feature(XENFEAT_auto_translated_physmap)) | |
f4f97b3e JF |
723 | return 0; |
724 | ||
5deb30d1 JF |
725 | /* |
726 | * 64-bit has a great big hole in the middle of the address | |
727 | * space, which contains the Xen mappings. On 32-bit these | |
728 | * will end up making a zero-sized hole and so is a no-op. | |
729 | */ | |
d6182fbf | 730 | hole_low = pgd_index(USER_LIMIT); |
5deb30d1 JF |
731 | hole_high = pgd_index(PAGE_OFFSET); |
732 | ||
733 | pgdidx_limit = pgd_index(limit); | |
734 | #if PTRS_PER_PUD > 1 | |
735 | pudidx_limit = pud_index(limit); | |
736 | #else | |
737 | pudidx_limit = 0; | |
738 | #endif | |
739 | #if PTRS_PER_PMD > 1 | |
740 | pmdidx_limit = pmd_index(limit); | |
741 | #else | |
742 | pmdidx_limit = 0; | |
743 | #endif | |
744 | ||
5deb30d1 | 745 | for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) { |
f4f97b3e | 746 | pud_t *pud; |
3b827c1b | 747 | |
5deb30d1 JF |
748 | if (pgdidx >= hole_low && pgdidx < hole_high) |
749 | continue; | |
f4f97b3e | 750 | |
5deb30d1 | 751 | if (!pgd_val(pgd[pgdidx])) |
3b827c1b | 752 | continue; |
f4f97b3e | 753 | |
5deb30d1 | 754 | pud = pud_offset(&pgd[pgdidx], 0); |
3b827c1b JF |
755 | |
756 | if (PTRS_PER_PUD > 1) /* not folded */ | |
eefb47f6 | 757 | flush |= (*func)(mm, virt_to_page(pud), PT_PUD); |
f4f97b3e | 758 | |
5deb30d1 | 759 | for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) { |
f4f97b3e | 760 | pmd_t *pmd; |
f4f97b3e | 761 | |
5deb30d1 JF |
762 | if (pgdidx == pgdidx_limit && |
763 | pudidx > pudidx_limit) | |
764 | goto out; | |
3b827c1b | 765 | |
5deb30d1 | 766 | if (pud_none(pud[pudidx])) |
3b827c1b | 767 | continue; |
f4f97b3e | 768 | |
5deb30d1 | 769 | pmd = pmd_offset(&pud[pudidx], 0); |
3b827c1b JF |
770 | |
771 | if (PTRS_PER_PMD > 1) /* not folded */ | |
eefb47f6 | 772 | flush |= (*func)(mm, virt_to_page(pmd), PT_PMD); |
f4f97b3e | 773 | |
5deb30d1 JF |
774 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) { |
775 | struct page *pte; | |
776 | ||
777 | if (pgdidx == pgdidx_limit && | |
778 | pudidx == pudidx_limit && | |
779 | pmdidx > pmdidx_limit) | |
780 | goto out; | |
3b827c1b | 781 | |
5deb30d1 | 782 | if (pmd_none(pmd[pmdidx])) |
3b827c1b JF |
783 | continue; |
784 | ||
5deb30d1 | 785 | pte = pmd_page(pmd[pmdidx]); |
eefb47f6 | 786 | flush |= (*func)(mm, pte, PT_PTE); |
3b827c1b JF |
787 | } |
788 | } | |
789 | } | |
11ad93e5 | 790 | |
5deb30d1 | 791 | out: |
11ad93e5 JF |
792 | /* Do the top level last, so that the callbacks can use it as |
793 | a cue to do final things like tlb flushes. */ | |
eefb47f6 | 794 | flush |= (*func)(mm, virt_to_page(pgd), PT_PGD); |
f4f97b3e JF |
795 | |
796 | return flush; | |
3b827c1b JF |
797 | } |
798 | ||
86bbc2c2 IC |
799 | static int xen_pgd_walk(struct mm_struct *mm, |
800 | int (*func)(struct mm_struct *mm, struct page *, | |
801 | enum pt_level), | |
802 | unsigned long limit) | |
803 | { | |
804 | return __xen_pgd_walk(mm, mm->pgd, func, limit); | |
805 | } | |
806 | ||
7708ad64 JF |
807 | /* If we're using split pte locks, then take the page's lock and |
808 | return a pointer to it. Otherwise return NULL. */ | |
eefb47f6 | 809 | static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm) |
74260714 JF |
810 | { |
811 | spinlock_t *ptl = NULL; | |
812 | ||
f7d0b926 | 813 | #if USE_SPLIT_PTLOCKS |
74260714 | 814 | ptl = __pte_lockptr(page); |
eefb47f6 | 815 | spin_lock_nest_lock(ptl, &mm->page_table_lock); |
74260714 JF |
816 | #endif |
817 | ||
818 | return ptl; | |
819 | } | |
820 | ||
7708ad64 | 821 | static void xen_pte_unlock(void *v) |
74260714 JF |
822 | { |
823 | spinlock_t *ptl = v; | |
824 | spin_unlock(ptl); | |
825 | } | |
826 | ||
827 | static void xen_do_pin(unsigned level, unsigned long pfn) | |
828 | { | |
829 | struct mmuext_op *op; | |
830 | struct multicall_space mcs; | |
831 | ||
832 | mcs = __xen_mc_entry(sizeof(*op)); | |
833 | op = mcs.args; | |
834 | op->cmd = level; | |
835 | op->arg1.mfn = pfn_to_mfn(pfn); | |
836 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
837 | } | |
838 | ||
eefb47f6 JF |
839 | static int xen_pin_page(struct mm_struct *mm, struct page *page, |
840 | enum pt_level level) | |
f4f97b3e | 841 | { |
d60cd46b | 842 | unsigned pgfl = TestSetPagePinned(page); |
f4f97b3e JF |
843 | int flush; |
844 | ||
845 | if (pgfl) | |
846 | flush = 0; /* already pinned */ | |
847 | else if (PageHighMem(page)) | |
848 | /* kmaps need flushing if we found an unpinned | |
849 | highpage */ | |
850 | flush = 1; | |
851 | else { | |
852 | void *pt = lowmem_page_address(page); | |
853 | unsigned long pfn = page_to_pfn(page); | |
854 | struct multicall_space mcs = __xen_mc_entry(0); | |
74260714 | 855 | spinlock_t *ptl; |
f4f97b3e JF |
856 | |
857 | flush = 0; | |
858 | ||
11ad93e5 JF |
859 | /* |
860 | * We need to hold the pagetable lock between the time | |
861 | * we make the pagetable RO and when we actually pin | |
862 | * it. If we don't, then other users may come in and | |
863 | * attempt to update the pagetable by writing it, | |
864 | * which will fail because the memory is RO but not | |
865 | * pinned, so Xen won't do the trap'n'emulate. | |
866 | * | |
867 | * If we're using split pte locks, we can't hold the | |
868 | * entire pagetable's worth of locks during the | |
869 | * traverse, because we may wrap the preempt count (8 | |
870 | * bits). The solution is to mark RO and pin each PTE | |
871 | * page while holding the lock. This means the number | |
872 | * of locks we end up holding is never more than a | |
873 | * batch size (~32 entries, at present). | |
874 | * | |
875 | * If we're not using split pte locks, we needn't pin | |
876 | * the PTE pages independently, because we're | |
877 | * protected by the overall pagetable lock. | |
878 | */ | |
74260714 JF |
879 | ptl = NULL; |
880 | if (level == PT_PTE) | |
eefb47f6 | 881 | ptl = xen_pte_lock(page, mm); |
74260714 | 882 | |
f4f97b3e JF |
883 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, |
884 | pfn_pte(pfn, PAGE_KERNEL_RO), | |
74260714 JF |
885 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
886 | ||
11ad93e5 | 887 | if (ptl) { |
74260714 JF |
888 | xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn); |
889 | ||
74260714 JF |
890 | /* Queue a deferred unlock for when this batch |
891 | is completed. */ | |
7708ad64 | 892 | xen_mc_callback(xen_pte_unlock, ptl); |
74260714 | 893 | } |
f4f97b3e JF |
894 | } |
895 | ||
896 | return flush; | |
897 | } | |
3b827c1b | 898 | |
f4f97b3e JF |
899 | /* This is called just after a mm has been created, but it has not |
900 | been used yet. We need to make sure that its pagetable is all | |
901 | read-only, and can be pinned. */ | |
eefb47f6 | 902 | static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd) |
3b827c1b | 903 | { |
d05fdf31 JF |
904 | vm_unmap_aliases(); |
905 | ||
f4f97b3e | 906 | xen_mc_batch(); |
3b827c1b | 907 | |
86bbc2c2 | 908 | if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) { |
d05fdf31 | 909 | /* re-enable interrupts for flushing */ |
f87e4cac | 910 | xen_mc_issue(0); |
d05fdf31 | 911 | |
f4f97b3e | 912 | kmap_flush_unused(); |
d05fdf31 | 913 | |
f87e4cac JF |
914 | xen_mc_batch(); |
915 | } | |
f4f97b3e | 916 | |
d6182fbf JF |
917 | #ifdef CONFIG_X86_64 |
918 | { | |
919 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
920 | ||
921 | xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd))); | |
922 | ||
923 | if (user_pgd) { | |
eefb47f6 | 924 | xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD); |
f63c2f24 T |
925 | xen_do_pin(MMUEXT_PIN_L4_TABLE, |
926 | PFN_DOWN(__pa(user_pgd))); | |
d6182fbf JF |
927 | } |
928 | } | |
929 | #else /* CONFIG_X86_32 */ | |
5deb30d1 JF |
930 | #ifdef CONFIG_X86_PAE |
931 | /* Need to make sure unshared kernel PMD is pinnable */ | |
47cb2ed9 | 932 | xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), |
eefb47f6 | 933 | PT_PMD); |
5deb30d1 | 934 | #endif |
28499143 | 935 | xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd))); |
d6182fbf | 936 | #endif /* CONFIG_X86_64 */ |
f4f97b3e | 937 | xen_mc_issue(0); |
3b827c1b JF |
938 | } |
939 | ||
eefb47f6 JF |
940 | static void xen_pgd_pin(struct mm_struct *mm) |
941 | { | |
942 | __xen_pgd_pin(mm, mm->pgd); | |
943 | } | |
944 | ||
0e91398f JF |
945 | /* |
946 | * On save, we need to pin all pagetables to make sure they get their | |
947 | * mfns turned into pfns. Search the list for any unpinned pgds and pin | |
948 | * them (unpinned pgds are not currently in use, probably because the | |
949 | * process is under construction or destruction). | |
eefb47f6 JF |
950 | * |
951 | * Expected to be called in stop_machine() ("equivalent to taking | |
952 | * every spinlock in the system"), so the locking doesn't really | |
953 | * matter all that much. | |
0e91398f JF |
954 | */ |
955 | void xen_mm_pin_all(void) | |
956 | { | |
957 | unsigned long flags; | |
958 | struct page *page; | |
74260714 | 959 | |
0e91398f | 960 | spin_lock_irqsave(&pgd_lock, flags); |
f4f97b3e | 961 | |
0e91398f JF |
962 | list_for_each_entry(page, &pgd_list, lru) { |
963 | if (!PagePinned(page)) { | |
eefb47f6 | 964 | __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page)); |
0e91398f JF |
965 | SetPageSavePinned(page); |
966 | } | |
967 | } | |
968 | ||
969 | spin_unlock_irqrestore(&pgd_lock, flags); | |
3b827c1b JF |
970 | } |
971 | ||
c1f2f09e EH |
972 | /* |
973 | * The init_mm pagetable is really pinned as soon as its created, but | |
974 | * that's before we have page structures to store the bits. So do all | |
975 | * the book-keeping now. | |
976 | */ | |
eefb47f6 JF |
977 | static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page, |
978 | enum pt_level level) | |
3b827c1b | 979 | { |
f4f97b3e JF |
980 | SetPagePinned(page); |
981 | return 0; | |
982 | } | |
3b827c1b | 983 | |
f4f97b3e JF |
984 | void __init xen_mark_init_mm_pinned(void) |
985 | { | |
eefb47f6 | 986 | xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP); |
f4f97b3e | 987 | } |
3b827c1b | 988 | |
eefb47f6 JF |
989 | static int xen_unpin_page(struct mm_struct *mm, struct page *page, |
990 | enum pt_level level) | |
f4f97b3e | 991 | { |
d60cd46b | 992 | unsigned pgfl = TestClearPagePinned(page); |
3b827c1b | 993 | |
f4f97b3e JF |
994 | if (pgfl && !PageHighMem(page)) { |
995 | void *pt = lowmem_page_address(page); | |
996 | unsigned long pfn = page_to_pfn(page); | |
74260714 JF |
997 | spinlock_t *ptl = NULL; |
998 | struct multicall_space mcs; | |
999 | ||
11ad93e5 JF |
1000 | /* |
1001 | * Do the converse to pin_page. If we're using split | |
1002 | * pte locks, we must be holding the lock for while | |
1003 | * the pte page is unpinned but still RO to prevent | |
1004 | * concurrent updates from seeing it in this | |
1005 | * partially-pinned state. | |
1006 | */ | |
74260714 | 1007 | if (level == PT_PTE) { |
eefb47f6 | 1008 | ptl = xen_pte_lock(page, mm); |
74260714 | 1009 | |
11ad93e5 JF |
1010 | if (ptl) |
1011 | xen_do_pin(MMUEXT_UNPIN_TABLE, pfn); | |
74260714 JF |
1012 | } |
1013 | ||
1014 | mcs = __xen_mc_entry(0); | |
f4f97b3e JF |
1015 | |
1016 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, | |
1017 | pfn_pte(pfn, PAGE_KERNEL), | |
74260714 JF |
1018 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
1019 | ||
1020 | if (ptl) { | |
1021 | /* unlock when batch completed */ | |
7708ad64 | 1022 | xen_mc_callback(xen_pte_unlock, ptl); |
74260714 | 1023 | } |
f4f97b3e JF |
1024 | } |
1025 | ||
1026 | return 0; /* never need to flush on unpin */ | |
3b827c1b JF |
1027 | } |
1028 | ||
f4f97b3e | 1029 | /* Release a pagetables pages back as normal RW */ |
eefb47f6 | 1030 | static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd) |
f4f97b3e | 1031 | { |
f4f97b3e JF |
1032 | xen_mc_batch(); |
1033 | ||
74260714 | 1034 | xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); |
f4f97b3e | 1035 | |
d6182fbf JF |
1036 | #ifdef CONFIG_X86_64 |
1037 | { | |
1038 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
1039 | ||
1040 | if (user_pgd) { | |
f63c2f24 T |
1041 | xen_do_pin(MMUEXT_UNPIN_TABLE, |
1042 | PFN_DOWN(__pa(user_pgd))); | |
eefb47f6 | 1043 | xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD); |
d6182fbf JF |
1044 | } |
1045 | } | |
1046 | #endif | |
1047 | ||
5deb30d1 JF |
1048 | #ifdef CONFIG_X86_PAE |
1049 | /* Need to make sure unshared kernel PMD is unpinned */ | |
47cb2ed9 | 1050 | xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), |
eefb47f6 | 1051 | PT_PMD); |
5deb30d1 | 1052 | #endif |
d6182fbf | 1053 | |
86bbc2c2 | 1054 | __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT); |
f4f97b3e JF |
1055 | |
1056 | xen_mc_issue(0); | |
1057 | } | |
3b827c1b | 1058 | |
eefb47f6 JF |
1059 | static void xen_pgd_unpin(struct mm_struct *mm) |
1060 | { | |
1061 | __xen_pgd_unpin(mm, mm->pgd); | |
1062 | } | |
1063 | ||
0e91398f JF |
1064 | /* |
1065 | * On resume, undo any pinning done at save, so that the rest of the | |
1066 | * kernel doesn't see any unexpected pinned pagetables. | |
1067 | */ | |
1068 | void xen_mm_unpin_all(void) | |
1069 | { | |
1070 | unsigned long flags; | |
1071 | struct page *page; | |
1072 | ||
1073 | spin_lock_irqsave(&pgd_lock, flags); | |
1074 | ||
1075 | list_for_each_entry(page, &pgd_list, lru) { | |
1076 | if (PageSavePinned(page)) { | |
1077 | BUG_ON(!PagePinned(page)); | |
eefb47f6 | 1078 | __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page)); |
0e91398f JF |
1079 | ClearPageSavePinned(page); |
1080 | } | |
1081 | } | |
1082 | ||
1083 | spin_unlock_irqrestore(&pgd_lock, flags); | |
1084 | } | |
1085 | ||
3b827c1b JF |
1086 | void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next) |
1087 | { | |
f4f97b3e | 1088 | spin_lock(&next->page_table_lock); |
eefb47f6 | 1089 | xen_pgd_pin(next); |
f4f97b3e | 1090 | spin_unlock(&next->page_table_lock); |
3b827c1b JF |
1091 | } |
1092 | ||
1093 | void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm) | |
1094 | { | |
f4f97b3e | 1095 | spin_lock(&mm->page_table_lock); |
eefb47f6 | 1096 | xen_pgd_pin(mm); |
f4f97b3e | 1097 | spin_unlock(&mm->page_table_lock); |
3b827c1b JF |
1098 | } |
1099 | ||
3b827c1b | 1100 | |
f87e4cac JF |
1101 | #ifdef CONFIG_SMP |
1102 | /* Another cpu may still have their %cr3 pointing at the pagetable, so | |
1103 | we need to repoint it somewhere else before we can unpin it. */ | |
1104 | static void drop_other_mm_ref(void *info) | |
1105 | { | |
1106 | struct mm_struct *mm = info; | |
ce87b3d3 | 1107 | struct mm_struct *active_mm; |
3b827c1b | 1108 | |
9eb912d1 | 1109 | active_mm = percpu_read(cpu_tlbstate.active_mm); |
ce87b3d3 JF |
1110 | |
1111 | if (active_mm == mm) | |
f87e4cac | 1112 | leave_mm(smp_processor_id()); |
9f79991d JF |
1113 | |
1114 | /* If this cpu still has a stale cr3 reference, then make sure | |
1115 | it has been flushed. */ | |
7fd7d83d | 1116 | if (percpu_read(xen_current_cr3) == __pa(mm->pgd)) |
9f79991d | 1117 | load_cr3(swapper_pg_dir); |
f87e4cac | 1118 | } |
3b827c1b | 1119 | |
7708ad64 | 1120 | static void xen_drop_mm_ref(struct mm_struct *mm) |
f87e4cac | 1121 | { |
e4d98207 | 1122 | cpumask_var_t mask; |
9f79991d JF |
1123 | unsigned cpu; |
1124 | ||
f87e4cac JF |
1125 | if (current->active_mm == mm) { |
1126 | if (current->mm == mm) | |
1127 | load_cr3(swapper_pg_dir); | |
1128 | else | |
1129 | leave_mm(smp_processor_id()); | |
9f79991d JF |
1130 | } |
1131 | ||
1132 | /* Get the "official" set of cpus referring to our pagetable. */ | |
e4d98207 MT |
1133 | if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) { |
1134 | for_each_online_cpu(cpu) { | |
1135 | if (!cpumask_test_cpu(cpu, &mm->cpu_vm_mask) | |
1136 | && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd)) | |
1137 | continue; | |
1138 | smp_call_function_single(cpu, drop_other_mm_ref, mm, 1); | |
1139 | } | |
1140 | return; | |
1141 | } | |
1142 | cpumask_copy(mask, &mm->cpu_vm_mask); | |
9f79991d JF |
1143 | |
1144 | /* It's possible that a vcpu may have a stale reference to our | |
1145 | cr3, because its in lazy mode, and it hasn't yet flushed | |
1146 | its set of pending hypercalls yet. In this case, we can | |
1147 | look at its actual current cr3 value, and force it to flush | |
1148 | if needed. */ | |
1149 | for_each_online_cpu(cpu) { | |
1150 | if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd)) | |
e4d98207 | 1151 | cpumask_set_cpu(cpu, mask); |
3b827c1b JF |
1152 | } |
1153 | ||
e4d98207 MT |
1154 | if (!cpumask_empty(mask)) |
1155 | smp_call_function_many(mask, drop_other_mm_ref, mm, 1); | |
1156 | free_cpumask_var(mask); | |
f87e4cac JF |
1157 | } |
1158 | #else | |
7708ad64 | 1159 | static void xen_drop_mm_ref(struct mm_struct *mm) |
f87e4cac JF |
1160 | { |
1161 | if (current->active_mm == mm) | |
1162 | load_cr3(swapper_pg_dir); | |
1163 | } | |
1164 | #endif | |
1165 | ||
1166 | /* | |
1167 | * While a process runs, Xen pins its pagetables, which means that the | |
1168 | * hypervisor forces it to be read-only, and it controls all updates | |
1169 | * to it. This means that all pagetable updates have to go via the | |
1170 | * hypervisor, which is moderately expensive. | |
1171 | * | |
1172 | * Since we're pulling the pagetable down, we switch to use init_mm, | |
1173 | * unpin old process pagetable and mark it all read-write, which | |
1174 | * allows further operations on it to be simple memory accesses. | |
1175 | * | |
1176 | * The only subtle point is that another CPU may be still using the | |
1177 | * pagetable because of lazy tlb flushing. This means we need need to | |
1178 | * switch all CPUs off this pagetable before we can unpin it. | |
1179 | */ | |
1180 | void xen_exit_mmap(struct mm_struct *mm) | |
1181 | { | |
1182 | get_cpu(); /* make sure we don't move around */ | |
7708ad64 | 1183 | xen_drop_mm_ref(mm); |
f87e4cac | 1184 | put_cpu(); |
3b827c1b | 1185 | |
f120f13e | 1186 | spin_lock(&mm->page_table_lock); |
df912ea4 JF |
1187 | |
1188 | /* pgd may not be pinned in the error exit path of execve */ | |
7708ad64 | 1189 | if (xen_page_pinned(mm->pgd)) |
eefb47f6 | 1190 | xen_pgd_unpin(mm); |
74260714 | 1191 | |
f120f13e | 1192 | spin_unlock(&mm->page_table_lock); |
3b827c1b | 1193 | } |
994025ca | 1194 | |
319f3ba5 JF |
1195 | static __init void xen_pagetable_setup_start(pgd_t *base) |
1196 | { | |
1197 | } | |
1198 | ||
1199 | static __init void xen_pagetable_setup_done(pgd_t *base) | |
1200 | { | |
1201 | xen_setup_shared_info(); | |
1202 | } | |
1203 | ||
1204 | static void xen_write_cr2(unsigned long cr2) | |
1205 | { | |
1206 | percpu_read(xen_vcpu)->arch.cr2 = cr2; | |
1207 | } | |
1208 | ||
1209 | static unsigned long xen_read_cr2(void) | |
1210 | { | |
1211 | return percpu_read(xen_vcpu)->arch.cr2; | |
1212 | } | |
1213 | ||
1214 | unsigned long xen_read_cr2_direct(void) | |
1215 | { | |
1216 | return percpu_read(xen_vcpu_info.arch.cr2); | |
1217 | } | |
1218 | ||
1219 | static void xen_flush_tlb(void) | |
1220 | { | |
1221 | struct mmuext_op *op; | |
1222 | struct multicall_space mcs; | |
1223 | ||
1224 | preempt_disable(); | |
1225 | ||
1226 | mcs = xen_mc_entry(sizeof(*op)); | |
1227 | ||
1228 | op = mcs.args; | |
1229 | op->cmd = MMUEXT_TLB_FLUSH_LOCAL; | |
1230 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1231 | ||
1232 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1233 | ||
1234 | preempt_enable(); | |
1235 | } | |
1236 | ||
1237 | static void xen_flush_tlb_single(unsigned long addr) | |
1238 | { | |
1239 | struct mmuext_op *op; | |
1240 | struct multicall_space mcs; | |
1241 | ||
1242 | preempt_disable(); | |
1243 | ||
1244 | mcs = xen_mc_entry(sizeof(*op)); | |
1245 | op = mcs.args; | |
1246 | op->cmd = MMUEXT_INVLPG_LOCAL; | |
1247 | op->arg1.linear_addr = addr & PAGE_MASK; | |
1248 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1249 | ||
1250 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1251 | ||
1252 | preempt_enable(); | |
1253 | } | |
1254 | ||
1255 | static void xen_flush_tlb_others(const struct cpumask *cpus, | |
1256 | struct mm_struct *mm, unsigned long va) | |
1257 | { | |
1258 | struct { | |
1259 | struct mmuext_op op; | |
1260 | DECLARE_BITMAP(mask, NR_CPUS); | |
1261 | } *args; | |
1262 | struct multicall_space mcs; | |
1263 | ||
1264 | BUG_ON(cpumask_empty(cpus)); | |
1265 | BUG_ON(!mm); | |
1266 | ||
1267 | mcs = xen_mc_entry(sizeof(*args)); | |
1268 | args = mcs.args; | |
1269 | args->op.arg2.vcpumask = to_cpumask(args->mask); | |
1270 | ||
1271 | /* Remove us, and any offline CPUS. */ | |
1272 | cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask); | |
1273 | cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask)); | |
319f3ba5 JF |
1274 | |
1275 | if (va == TLB_FLUSH_ALL) { | |
1276 | args->op.cmd = MMUEXT_TLB_FLUSH_MULTI; | |
1277 | } else { | |
1278 | args->op.cmd = MMUEXT_INVLPG_MULTI; | |
1279 | args->op.arg1.linear_addr = va; | |
1280 | } | |
1281 | ||
1282 | MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF); | |
1283 | ||
319f3ba5 JF |
1284 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
1285 | } | |
1286 | ||
1287 | static unsigned long xen_read_cr3(void) | |
1288 | { | |
1289 | return percpu_read(xen_cr3); | |
1290 | } | |
1291 | ||
1292 | static void set_current_cr3(void *v) | |
1293 | { | |
1294 | percpu_write(xen_current_cr3, (unsigned long)v); | |
1295 | } | |
1296 | ||
1297 | static void __xen_write_cr3(bool kernel, unsigned long cr3) | |
1298 | { | |
1299 | struct mmuext_op *op; | |
1300 | struct multicall_space mcs; | |
1301 | unsigned long mfn; | |
1302 | ||
1303 | if (cr3) | |
1304 | mfn = pfn_to_mfn(PFN_DOWN(cr3)); | |
1305 | else | |
1306 | mfn = 0; | |
1307 | ||
1308 | WARN_ON(mfn == 0 && kernel); | |
1309 | ||
1310 | mcs = __xen_mc_entry(sizeof(*op)); | |
1311 | ||
1312 | op = mcs.args; | |
1313 | op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR; | |
1314 | op->arg1.mfn = mfn; | |
1315 | ||
1316 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1317 | ||
1318 | if (kernel) { | |
1319 | percpu_write(xen_cr3, cr3); | |
1320 | ||
1321 | /* Update xen_current_cr3 once the batch has actually | |
1322 | been submitted. */ | |
1323 | xen_mc_callback(set_current_cr3, (void *)cr3); | |
1324 | } | |
1325 | } | |
1326 | ||
1327 | static void xen_write_cr3(unsigned long cr3) | |
1328 | { | |
1329 | BUG_ON(preemptible()); | |
1330 | ||
1331 | xen_mc_batch(); /* disables interrupts */ | |
1332 | ||
1333 | /* Update while interrupts are disabled, so its atomic with | |
1334 | respect to ipis */ | |
1335 | percpu_write(xen_cr3, cr3); | |
1336 | ||
1337 | __xen_write_cr3(true, cr3); | |
1338 | ||
1339 | #ifdef CONFIG_X86_64 | |
1340 | { | |
1341 | pgd_t *user_pgd = xen_get_user_pgd(__va(cr3)); | |
1342 | if (user_pgd) | |
1343 | __xen_write_cr3(false, __pa(user_pgd)); | |
1344 | else | |
1345 | __xen_write_cr3(false, 0); | |
1346 | } | |
1347 | #endif | |
1348 | ||
1349 | xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ | |
1350 | } | |
1351 | ||
1352 | static int xen_pgd_alloc(struct mm_struct *mm) | |
1353 | { | |
1354 | pgd_t *pgd = mm->pgd; | |
1355 | int ret = 0; | |
1356 | ||
1357 | BUG_ON(PagePinned(virt_to_page(pgd))); | |
1358 | ||
1359 | #ifdef CONFIG_X86_64 | |
1360 | { | |
1361 | struct page *page = virt_to_page(pgd); | |
1362 | pgd_t *user_pgd; | |
1363 | ||
1364 | BUG_ON(page->private != 0); | |
1365 | ||
1366 | ret = -ENOMEM; | |
1367 | ||
1368 | user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO); | |
1369 | page->private = (unsigned long)user_pgd; | |
1370 | ||
1371 | if (user_pgd != NULL) { | |
1372 | user_pgd[pgd_index(VSYSCALL_START)] = | |
1373 | __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE); | |
1374 | ret = 0; | |
1375 | } | |
1376 | ||
1377 | BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd)))); | |
1378 | } | |
1379 | #endif | |
1380 | ||
1381 | return ret; | |
1382 | } | |
1383 | ||
1384 | static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd) | |
1385 | { | |
1386 | #ifdef CONFIG_X86_64 | |
1387 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
1388 | ||
1389 | if (user_pgd) | |
1390 | free_page((unsigned long)user_pgd); | |
1391 | #endif | |
1392 | } | |
1393 | ||
1f4f9315 JF |
1394 | #ifdef CONFIG_HIGHPTE |
1395 | static void *xen_kmap_atomic_pte(struct page *page, enum km_type type) | |
1396 | { | |
1397 | pgprot_t prot = PAGE_KERNEL; | |
1398 | ||
1399 | if (PagePinned(page)) | |
1400 | prot = PAGE_KERNEL_RO; | |
1401 | ||
1402 | if (0 && PageHighMem(page)) | |
1403 | printk("mapping highpte %lx type %d prot %s\n", | |
1404 | page_to_pfn(page), type, | |
1405 | (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ"); | |
1406 | ||
1407 | return kmap_atomic_prot(page, type, prot); | |
1408 | } | |
1409 | #endif | |
1410 | ||
1411 | #ifdef CONFIG_X86_32 | |
1412 | static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte) | |
1413 | { | |
1414 | /* If there's an existing pte, then don't allow _PAGE_RW to be set */ | |
1415 | if (pte_val_ma(*ptep) & _PAGE_PRESENT) | |
1416 | pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) & | |
1417 | pte_val_ma(pte)); | |
1418 | ||
1419 | return pte; | |
1420 | } | |
1421 | ||
1422 | /* Init-time set_pte while constructing initial pagetables, which | |
1423 | doesn't allow RO pagetable pages to be remapped RW */ | |
1424 | static __init void xen_set_pte_init(pte_t *ptep, pte_t pte) | |
1425 | { | |
1426 | pte = mask_rw_pte(ptep, pte); | |
1427 | ||
1428 | xen_set_pte(ptep, pte); | |
1429 | } | |
1430 | #endif | |
319f3ba5 JF |
1431 | |
1432 | /* Early in boot, while setting up the initial pagetable, assume | |
1433 | everything is pinned. */ | |
1434 | static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn) | |
1435 | { | |
1436 | #ifdef CONFIG_FLATMEM | |
1437 | BUG_ON(mem_map); /* should only be used early */ | |
1438 | #endif | |
1439 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1440 | } | |
1441 | ||
1442 | /* Early release_pte assumes that all pts are pinned, since there's | |
1443 | only init_mm and anything attached to that is pinned. */ | |
1444 | static void xen_release_pte_init(unsigned long pfn) | |
1445 | { | |
1446 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
1447 | } | |
1448 | ||
1449 | static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn) | |
1450 | { | |
1451 | struct mmuext_op op; | |
1452 | op.cmd = cmd; | |
1453 | op.arg1.mfn = pfn_to_mfn(pfn); | |
1454 | if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) | |
1455 | BUG(); | |
1456 | } | |
1457 | ||
1458 | /* This needs to make sure the new pte page is pinned iff its being | |
1459 | attached to a pinned pagetable. */ | |
1460 | static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level) | |
1461 | { | |
1462 | struct page *page = pfn_to_page(pfn); | |
1463 | ||
1464 | if (PagePinned(virt_to_page(mm->pgd))) { | |
1465 | SetPagePinned(page); | |
1466 | ||
1467 | vm_unmap_aliases(); | |
1468 | if (!PageHighMem(page)) { | |
1469 | make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn))); | |
1470 | if (level == PT_PTE && USE_SPLIT_PTLOCKS) | |
1471 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
1472 | } else { | |
1473 | /* make sure there are no stray mappings of | |
1474 | this page */ | |
1475 | kmap_flush_unused(); | |
1476 | } | |
1477 | } | |
1478 | } | |
1479 | ||
1480 | static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn) | |
1481 | { | |
1482 | xen_alloc_ptpage(mm, pfn, PT_PTE); | |
1483 | } | |
1484 | ||
1485 | static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn) | |
1486 | { | |
1487 | xen_alloc_ptpage(mm, pfn, PT_PMD); | |
1488 | } | |
1489 | ||
1490 | /* This should never happen until we're OK to use struct page */ | |
1491 | static void xen_release_ptpage(unsigned long pfn, unsigned level) | |
1492 | { | |
1493 | struct page *page = pfn_to_page(pfn); | |
1494 | ||
1495 | if (PagePinned(page)) { | |
1496 | if (!PageHighMem(page)) { | |
1497 | if (level == PT_PTE && USE_SPLIT_PTLOCKS) | |
1498 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); | |
1499 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
1500 | } | |
1501 | ClearPagePinned(page); | |
1502 | } | |
1503 | } | |
1504 | ||
1505 | static void xen_release_pte(unsigned long pfn) | |
1506 | { | |
1507 | xen_release_ptpage(pfn, PT_PTE); | |
1508 | } | |
1509 | ||
1510 | static void xen_release_pmd(unsigned long pfn) | |
1511 | { | |
1512 | xen_release_ptpage(pfn, PT_PMD); | |
1513 | } | |
1514 | ||
1515 | #if PAGETABLE_LEVELS == 4 | |
1516 | static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn) | |
1517 | { | |
1518 | xen_alloc_ptpage(mm, pfn, PT_PUD); | |
1519 | } | |
1520 | ||
1521 | static void xen_release_pud(unsigned long pfn) | |
1522 | { | |
1523 | xen_release_ptpage(pfn, PT_PUD); | |
1524 | } | |
1525 | #endif | |
1526 | ||
1527 | void __init xen_reserve_top(void) | |
1528 | { | |
1529 | #ifdef CONFIG_X86_32 | |
1530 | unsigned long top = HYPERVISOR_VIRT_START; | |
1531 | struct xen_platform_parameters pp; | |
1532 | ||
1533 | if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0) | |
1534 | top = pp.virt_start; | |
1535 | ||
1536 | reserve_top_address(-top); | |
1537 | #endif /* CONFIG_X86_32 */ | |
1538 | } | |
1539 | ||
1540 | /* | |
1541 | * Like __va(), but returns address in the kernel mapping (which is | |
1542 | * all we have until the physical memory mapping has been set up. | |
1543 | */ | |
1544 | static void *__ka(phys_addr_t paddr) | |
1545 | { | |
1546 | #ifdef CONFIG_X86_64 | |
1547 | return (void *)(paddr + __START_KERNEL_map); | |
1548 | #else | |
1549 | return __va(paddr); | |
1550 | #endif | |
1551 | } | |
1552 | ||
1553 | /* Convert a machine address to physical address */ | |
1554 | static unsigned long m2p(phys_addr_t maddr) | |
1555 | { | |
1556 | phys_addr_t paddr; | |
1557 | ||
1558 | maddr &= PTE_PFN_MASK; | |
1559 | paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT; | |
1560 | ||
1561 | return paddr; | |
1562 | } | |
1563 | ||
1564 | /* Convert a machine address to kernel virtual */ | |
1565 | static void *m2v(phys_addr_t maddr) | |
1566 | { | |
1567 | return __ka(m2p(maddr)); | |
1568 | } | |
1569 | ||
1570 | static void set_page_prot(void *addr, pgprot_t prot) | |
1571 | { | |
1572 | unsigned long pfn = __pa(addr) >> PAGE_SHIFT; | |
1573 | pte_t pte = pfn_pte(pfn, prot); | |
1574 | ||
1575 | if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0)) | |
1576 | BUG(); | |
1577 | } | |
1578 | ||
1579 | static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn) | |
1580 | { | |
1581 | unsigned pmdidx, pteidx; | |
1582 | unsigned ident_pte; | |
1583 | unsigned long pfn; | |
1584 | ||
1585 | ident_pte = 0; | |
1586 | pfn = 0; | |
1587 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) { | |
1588 | pte_t *pte_page; | |
1589 | ||
1590 | /* Reuse or allocate a page of ptes */ | |
1591 | if (pmd_present(pmd[pmdidx])) | |
1592 | pte_page = m2v(pmd[pmdidx].pmd); | |
1593 | else { | |
1594 | /* Check for free pte pages */ | |
1595 | if (ident_pte == ARRAY_SIZE(level1_ident_pgt)) | |
1596 | break; | |
1597 | ||
1598 | pte_page = &level1_ident_pgt[ident_pte]; | |
1599 | ident_pte += PTRS_PER_PTE; | |
1600 | ||
1601 | pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE); | |
1602 | } | |
1603 | ||
1604 | /* Install mappings */ | |
1605 | for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) { | |
1606 | pte_t pte; | |
1607 | ||
1608 | if (pfn > max_pfn_mapped) | |
1609 | max_pfn_mapped = pfn; | |
1610 | ||
1611 | if (!pte_none(pte_page[pteidx])) | |
1612 | continue; | |
1613 | ||
1614 | pte = pfn_pte(pfn, PAGE_KERNEL_EXEC); | |
1615 | pte_page[pteidx] = pte; | |
1616 | } | |
1617 | } | |
1618 | ||
1619 | for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE) | |
1620 | set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO); | |
1621 | ||
1622 | set_page_prot(pmd, PAGE_KERNEL_RO); | |
1623 | } | |
1624 | ||
1625 | #ifdef CONFIG_X86_64 | |
1626 | static void convert_pfn_mfn(void *v) | |
1627 | { | |
1628 | pte_t *pte = v; | |
1629 | int i; | |
1630 | ||
1631 | /* All levels are converted the same way, so just treat them | |
1632 | as ptes. */ | |
1633 | for (i = 0; i < PTRS_PER_PTE; i++) | |
1634 | pte[i] = xen_make_pte(pte[i].pte); | |
1635 | } | |
1636 | ||
1637 | /* | |
1638 | * Set up the inital kernel pagetable. | |
1639 | * | |
1640 | * We can construct this by grafting the Xen provided pagetable into | |
1641 | * head_64.S's preconstructed pagetables. We copy the Xen L2's into | |
1642 | * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This | |
1643 | * means that only the kernel has a physical mapping to start with - | |
1644 | * but that's enough to get __va working. We need to fill in the rest | |
1645 | * of the physical mapping once some sort of allocator has been set | |
1646 | * up. | |
1647 | */ | |
1648 | __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, | |
1649 | unsigned long max_pfn) | |
1650 | { | |
1651 | pud_t *l3; | |
1652 | pmd_t *l2; | |
1653 | ||
1654 | /* Zap identity mapping */ | |
1655 | init_level4_pgt[0] = __pgd(0); | |
1656 | ||
1657 | /* Pre-constructed entries are in pfn, so convert to mfn */ | |
1658 | convert_pfn_mfn(init_level4_pgt); | |
1659 | convert_pfn_mfn(level3_ident_pgt); | |
1660 | convert_pfn_mfn(level3_kernel_pgt); | |
1661 | ||
1662 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd); | |
1663 | l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud); | |
1664 | ||
1665 | memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1666 | memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1667 | ||
1668 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd); | |
1669 | l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud); | |
1670 | memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1671 | ||
1672 | /* Set up identity map */ | |
1673 | xen_map_identity_early(level2_ident_pgt, max_pfn); | |
1674 | ||
1675 | /* Make pagetable pieces RO */ | |
1676 | set_page_prot(init_level4_pgt, PAGE_KERNEL_RO); | |
1677 | set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO); | |
1678 | set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO); | |
1679 | set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO); | |
1680 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); | |
1681 | set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO); | |
1682 | ||
1683 | /* Pin down new L4 */ | |
1684 | pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE, | |
1685 | PFN_DOWN(__pa_symbol(init_level4_pgt))); | |
1686 | ||
1687 | /* Unpin Xen-provided one */ | |
1688 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
1689 | ||
1690 | /* Switch over */ | |
1691 | pgd = init_level4_pgt; | |
1692 | ||
1693 | /* | |
1694 | * At this stage there can be no user pgd, and no page | |
1695 | * structure to attach it to, so make sure we just set kernel | |
1696 | * pgd. | |
1697 | */ | |
1698 | xen_mc_batch(); | |
1699 | __xen_write_cr3(true, __pa(pgd)); | |
1700 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
1701 | ||
1702 | reserve_early(__pa(xen_start_info->pt_base), | |
1703 | __pa(xen_start_info->pt_base + | |
1704 | xen_start_info->nr_pt_frames * PAGE_SIZE), | |
1705 | "XEN PAGETABLES"); | |
1706 | ||
1707 | return pgd; | |
1708 | } | |
1709 | #else /* !CONFIG_X86_64 */ | |
1710 | static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss; | |
1711 | ||
1712 | __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, | |
1713 | unsigned long max_pfn) | |
1714 | { | |
1715 | pmd_t *kernel_pmd; | |
1716 | ||
1717 | init_pg_tables_start = __pa(pgd); | |
1718 | init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE; | |
1719 | max_pfn_mapped = PFN_DOWN(init_pg_tables_end + 512*1024); | |
1720 | ||
1721 | kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd); | |
1722 | memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD); | |
1723 | ||
1724 | xen_map_identity_early(level2_kernel_pgt, max_pfn); | |
1725 | ||
1726 | memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD); | |
1727 | set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY], | |
1728 | __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT)); | |
1729 | ||
1730 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); | |
1731 | set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO); | |
1732 | set_page_prot(empty_zero_page, PAGE_KERNEL_RO); | |
1733 | ||
1734 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
1735 | ||
1736 | xen_write_cr3(__pa(swapper_pg_dir)); | |
1737 | ||
1738 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir))); | |
1739 | ||
1740 | return swapper_pg_dir; | |
1741 | } | |
1742 | #endif /* CONFIG_X86_64 */ | |
1743 | ||
1744 | static void xen_set_fixmap(unsigned idx, unsigned long phys, pgprot_t prot) | |
1745 | { | |
1746 | pte_t pte; | |
1747 | ||
1748 | phys >>= PAGE_SHIFT; | |
1749 | ||
1750 | switch (idx) { | |
1751 | case FIX_BTMAP_END ... FIX_BTMAP_BEGIN: | |
1752 | #ifdef CONFIG_X86_F00F_BUG | |
1753 | case FIX_F00F_IDT: | |
1754 | #endif | |
1755 | #ifdef CONFIG_X86_32 | |
1756 | case FIX_WP_TEST: | |
1757 | case FIX_VDSO: | |
1758 | # ifdef CONFIG_HIGHMEM | |
1759 | case FIX_KMAP_BEGIN ... FIX_KMAP_END: | |
1760 | # endif | |
1761 | #else | |
1762 | case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE: | |
1763 | #endif | |
1764 | #ifdef CONFIG_X86_LOCAL_APIC | |
1765 | case FIX_APIC_BASE: /* maps dummy local APIC */ | |
1766 | #endif | |
1767 | pte = pfn_pte(phys, prot); | |
1768 | break; | |
1769 | ||
1770 | default: | |
1771 | pte = mfn_pte(phys, prot); | |
1772 | break; | |
1773 | } | |
1774 | ||
1775 | __native_set_fixmap(idx, pte); | |
1776 | ||
1777 | #ifdef CONFIG_X86_64 | |
1778 | /* Replicate changes to map the vsyscall page into the user | |
1779 | pagetable vsyscall mapping. */ | |
1780 | if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) { | |
1781 | unsigned long vaddr = __fix_to_virt(idx); | |
1782 | set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte); | |
1783 | } | |
1784 | #endif | |
1785 | } | |
1786 | ||
1787 | __init void xen_post_allocator_init(void) | |
1788 | { | |
1789 | pv_mmu_ops.set_pte = xen_set_pte; | |
1790 | pv_mmu_ops.set_pmd = xen_set_pmd; | |
1791 | pv_mmu_ops.set_pud = xen_set_pud; | |
1792 | #if PAGETABLE_LEVELS == 4 | |
1793 | pv_mmu_ops.set_pgd = xen_set_pgd; | |
1794 | #endif | |
1795 | ||
1796 | /* This will work as long as patching hasn't happened yet | |
1797 | (which it hasn't) */ | |
1798 | pv_mmu_ops.alloc_pte = xen_alloc_pte; | |
1799 | pv_mmu_ops.alloc_pmd = xen_alloc_pmd; | |
1800 | pv_mmu_ops.release_pte = xen_release_pte; | |
1801 | pv_mmu_ops.release_pmd = xen_release_pmd; | |
1802 | #if PAGETABLE_LEVELS == 4 | |
1803 | pv_mmu_ops.alloc_pud = xen_alloc_pud; | |
1804 | pv_mmu_ops.release_pud = xen_release_pud; | |
1805 | #endif | |
1806 | ||
1807 | #ifdef CONFIG_X86_64 | |
1808 | SetPagePinned(virt_to_page(level3_user_vsyscall)); | |
1809 | #endif | |
1810 | xen_mark_init_mm_pinned(); | |
1811 | } | |
1812 | ||
b407fc57 JF |
1813 | static void xen_leave_lazy_mmu(void) |
1814 | { | |
1815 | xen_mc_flush(); | |
1816 | paravirt_leave_lazy_mmu(); | |
1817 | } | |
319f3ba5 JF |
1818 | |
1819 | const struct pv_mmu_ops xen_mmu_ops __initdata = { | |
1820 | .pagetable_setup_start = xen_pagetable_setup_start, | |
1821 | .pagetable_setup_done = xen_pagetable_setup_done, | |
1822 | ||
1823 | .read_cr2 = xen_read_cr2, | |
1824 | .write_cr2 = xen_write_cr2, | |
1825 | ||
1826 | .read_cr3 = xen_read_cr3, | |
1827 | .write_cr3 = xen_write_cr3, | |
1828 | ||
1829 | .flush_tlb_user = xen_flush_tlb, | |
1830 | .flush_tlb_kernel = xen_flush_tlb, | |
1831 | .flush_tlb_single = xen_flush_tlb_single, | |
1832 | .flush_tlb_others = xen_flush_tlb_others, | |
1833 | ||
1834 | .pte_update = paravirt_nop, | |
1835 | .pte_update_defer = paravirt_nop, | |
1836 | ||
1837 | .pgd_alloc = xen_pgd_alloc, | |
1838 | .pgd_free = xen_pgd_free, | |
1839 | ||
1840 | .alloc_pte = xen_alloc_pte_init, | |
1841 | .release_pte = xen_release_pte_init, | |
1842 | .alloc_pmd = xen_alloc_pte_init, | |
1843 | .alloc_pmd_clone = paravirt_nop, | |
1844 | .release_pmd = xen_release_pte_init, | |
1845 | ||
1846 | #ifdef CONFIG_HIGHPTE | |
1847 | .kmap_atomic_pte = xen_kmap_atomic_pte, | |
1848 | #endif | |
1849 | ||
1850 | #ifdef CONFIG_X86_64 | |
1851 | .set_pte = xen_set_pte, | |
1852 | #else | |
1853 | .set_pte = xen_set_pte_init, | |
1854 | #endif | |
1855 | .set_pte_at = xen_set_pte_at, | |
1856 | .set_pmd = xen_set_pmd_hyper, | |
1857 | ||
1858 | .ptep_modify_prot_start = __ptep_modify_prot_start, | |
1859 | .ptep_modify_prot_commit = __ptep_modify_prot_commit, | |
1860 | ||
da5de7c2 JF |
1861 | .pte_val = PV_CALLEE_SAVE(xen_pte_val), |
1862 | .pgd_val = PV_CALLEE_SAVE(xen_pgd_val), | |
319f3ba5 | 1863 | |
da5de7c2 JF |
1864 | .make_pte = PV_CALLEE_SAVE(xen_make_pte), |
1865 | .make_pgd = PV_CALLEE_SAVE(xen_make_pgd), | |
319f3ba5 JF |
1866 | |
1867 | #ifdef CONFIG_X86_PAE | |
1868 | .set_pte_atomic = xen_set_pte_atomic, | |
1869 | .set_pte_present = xen_set_pte_at, | |
1870 | .pte_clear = xen_pte_clear, | |
1871 | .pmd_clear = xen_pmd_clear, | |
1872 | #endif /* CONFIG_X86_PAE */ | |
1873 | .set_pud = xen_set_pud_hyper, | |
1874 | ||
da5de7c2 JF |
1875 | .make_pmd = PV_CALLEE_SAVE(xen_make_pmd), |
1876 | .pmd_val = PV_CALLEE_SAVE(xen_pmd_val), | |
319f3ba5 JF |
1877 | |
1878 | #if PAGETABLE_LEVELS == 4 | |
da5de7c2 JF |
1879 | .pud_val = PV_CALLEE_SAVE(xen_pud_val), |
1880 | .make_pud = PV_CALLEE_SAVE(xen_make_pud), | |
319f3ba5 JF |
1881 | .set_pgd = xen_set_pgd_hyper, |
1882 | ||
1883 | .alloc_pud = xen_alloc_pte_init, | |
1884 | .release_pud = xen_release_pte_init, | |
1885 | #endif /* PAGETABLE_LEVELS == 4 */ | |
1886 | ||
1887 | .activate_mm = xen_activate_mm, | |
1888 | .dup_mmap = xen_dup_mmap, | |
1889 | .exit_mmap = xen_exit_mmap, | |
1890 | ||
1891 | .lazy_mode = { | |
1892 | .enter = paravirt_enter_lazy_mmu, | |
b407fc57 | 1893 | .leave = xen_leave_lazy_mmu, |
319f3ba5 JF |
1894 | }, |
1895 | ||
1896 | .set_fixmap = xen_set_fixmap, | |
1897 | }; | |
1898 | ||
1899 | ||
994025ca JF |
1900 | #ifdef CONFIG_XEN_DEBUG_FS |
1901 | ||
1902 | static struct dentry *d_mmu_debug; | |
1903 | ||
1904 | static int __init xen_mmu_debugfs(void) | |
1905 | { | |
1906 | struct dentry *d_xen = xen_init_debugfs(); | |
1907 | ||
1908 | if (d_xen == NULL) | |
1909 | return -ENOMEM; | |
1910 | ||
1911 | d_mmu_debug = debugfs_create_dir("mmu", d_xen); | |
1912 | ||
1913 | debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats); | |
1914 | ||
1915 | debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update); | |
1916 | debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug, | |
1917 | &mmu_stats.pgd_update_pinned); | |
1918 | debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug, | |
1919 | &mmu_stats.pgd_update_pinned); | |
1920 | ||
1921 | debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update); | |
1922 | debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug, | |
1923 | &mmu_stats.pud_update_pinned); | |
1924 | debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug, | |
1925 | &mmu_stats.pud_update_pinned); | |
1926 | ||
1927 | debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update); | |
1928 | debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug, | |
1929 | &mmu_stats.pmd_update_pinned); | |
1930 | debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug, | |
1931 | &mmu_stats.pmd_update_pinned); | |
1932 | ||
1933 | debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update); | |
1934 | // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug, | |
1935 | // &mmu_stats.pte_update_pinned); | |
1936 | debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug, | |
1937 | &mmu_stats.pte_update_pinned); | |
1938 | ||
1939 | debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update); | |
1940 | debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug, | |
1941 | &mmu_stats.mmu_update_extended); | |
1942 | xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug, | |
1943 | mmu_stats.mmu_update_histo, 20); | |
1944 | ||
1945 | debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at); | |
1946 | debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug, | |
1947 | &mmu_stats.set_pte_at_batched); | |
1948 | debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug, | |
1949 | &mmu_stats.set_pte_at_current); | |
1950 | debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug, | |
1951 | &mmu_stats.set_pte_at_kernel); | |
1952 | ||
1953 | debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit); | |
1954 | debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug, | |
1955 | &mmu_stats.prot_commit_batched); | |
1956 | ||
1957 | return 0; | |
1958 | } | |
1959 | fs_initcall(xen_mmu_debugfs); | |
1960 | ||
1961 | #endif /* CONFIG_XEN_DEBUG_FS */ |