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> |
44408ad7 | 45 | #include <linux/module.h> |
3b827c1b JF |
46 | |
47 | #include <asm/pgtable.h> | |
48 | #include <asm/tlbflush.h> | |
5deb30d1 | 49 | #include <asm/fixmap.h> |
3b827c1b | 50 | #include <asm/mmu_context.h> |
319f3ba5 | 51 | #include <asm/setup.h> |
f4f97b3e | 52 | #include <asm/paravirt.h> |
cbcd79c2 | 53 | #include <asm/linkage.h> |
3b827c1b JF |
54 | |
55 | #include <asm/xen/hypercall.h> | |
f4f97b3e | 56 | #include <asm/xen/hypervisor.h> |
3b827c1b JF |
57 | |
58 | #include <xen/page.h> | |
59 | #include <xen/interface/xen.h> | |
319f3ba5 JF |
60 | #include <xen/interface/version.h> |
61 | #include <xen/hvc-console.h> | |
3b827c1b | 62 | |
f4f97b3e | 63 | #include "multicalls.h" |
3b827c1b | 64 | #include "mmu.h" |
994025ca JF |
65 | #include "debugfs.h" |
66 | ||
67 | #define MMU_UPDATE_HISTO 30 | |
68 | ||
69 | #ifdef CONFIG_XEN_DEBUG_FS | |
70 | ||
71 | static struct { | |
72 | u32 pgd_update; | |
73 | u32 pgd_update_pinned; | |
74 | u32 pgd_update_batched; | |
75 | ||
76 | u32 pud_update; | |
77 | u32 pud_update_pinned; | |
78 | u32 pud_update_batched; | |
79 | ||
80 | u32 pmd_update; | |
81 | u32 pmd_update_pinned; | |
82 | u32 pmd_update_batched; | |
83 | ||
84 | u32 pte_update; | |
85 | u32 pte_update_pinned; | |
86 | u32 pte_update_batched; | |
87 | ||
88 | u32 mmu_update; | |
89 | u32 mmu_update_extended; | |
90 | u32 mmu_update_histo[MMU_UPDATE_HISTO]; | |
91 | ||
92 | u32 prot_commit; | |
93 | u32 prot_commit_batched; | |
94 | ||
95 | u32 set_pte_at; | |
96 | u32 set_pte_at_batched; | |
97 | u32 set_pte_at_pinned; | |
98 | u32 set_pte_at_current; | |
99 | u32 set_pte_at_kernel; | |
100 | } mmu_stats; | |
101 | ||
102 | static u8 zero_stats; | |
103 | ||
104 | static inline void check_zero(void) | |
105 | { | |
106 | if (unlikely(zero_stats)) { | |
107 | memset(&mmu_stats, 0, sizeof(mmu_stats)); | |
108 | zero_stats = 0; | |
109 | } | |
110 | } | |
111 | ||
112 | #define ADD_STATS(elem, val) \ | |
113 | do { check_zero(); mmu_stats.elem += (val); } while(0) | |
114 | ||
115 | #else /* !CONFIG_XEN_DEBUG_FS */ | |
116 | ||
117 | #define ADD_STATS(elem, val) do { (void)(val); } while(0) | |
118 | ||
119 | #endif /* CONFIG_XEN_DEBUG_FS */ | |
3b827c1b | 120 | |
319f3ba5 JF |
121 | |
122 | /* | |
123 | * Identity map, in addition to plain kernel map. This needs to be | |
124 | * large enough to allocate page table pages to allocate the rest. | |
125 | * Each page can map 2MB. | |
126 | */ | |
127 | static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss; | |
128 | ||
129 | #ifdef CONFIG_X86_64 | |
130 | /* l3 pud for userspace vsyscall mapping */ | |
131 | static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss; | |
132 | #endif /* CONFIG_X86_64 */ | |
133 | ||
134 | /* | |
135 | * Note about cr3 (pagetable base) values: | |
136 | * | |
137 | * xen_cr3 contains the current logical cr3 value; it contains the | |
138 | * last set cr3. This may not be the current effective cr3, because | |
139 | * its update may be being lazily deferred. However, a vcpu looking | |
140 | * at its own cr3 can use this value knowing that it everything will | |
141 | * be self-consistent. | |
142 | * | |
143 | * xen_current_cr3 contains the actual vcpu cr3; it is set once the | |
144 | * hypercall to set the vcpu cr3 is complete (so it may be a little | |
145 | * out of date, but it will never be set early). If one vcpu is | |
146 | * looking at another vcpu's cr3 value, it should use this variable. | |
147 | */ | |
148 | DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */ | |
149 | DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */ | |
150 | ||
151 | ||
d6182fbf JF |
152 | /* |
153 | * Just beyond the highest usermode address. STACK_TOP_MAX has a | |
154 | * redzone above it, so round it up to a PGD boundary. | |
155 | */ | |
156 | #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK) | |
157 | ||
158 | ||
d451bb7a | 159 | #define P2M_ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(unsigned long)) |
cf0923ea | 160 | #define TOP_ENTRIES (MAX_DOMAIN_PAGES / P2M_ENTRIES_PER_PAGE) |
d451bb7a | 161 | |
cf0923ea | 162 | /* Placeholder for holes in the address space */ |
cbcd79c2 | 163 | static unsigned long p2m_missing[P2M_ENTRIES_PER_PAGE] __page_aligned_data = |
cf0923ea JF |
164 | { [ 0 ... P2M_ENTRIES_PER_PAGE-1 ] = ~0UL }; |
165 | ||
166 | /* Array of pointers to pages containing p2m entries */ | |
cbcd79c2 | 167 | static unsigned long *p2m_top[TOP_ENTRIES] __page_aligned_data = |
cf0923ea | 168 | { [ 0 ... TOP_ENTRIES - 1] = &p2m_missing[0] }; |
d451bb7a | 169 | |
d5edbc1f | 170 | /* Arrays of p2m arrays expressed in mfns used for save/restore */ |
cbcd79c2 | 171 | static unsigned long p2m_top_mfn[TOP_ENTRIES] __page_aligned_bss; |
d5edbc1f | 172 | |
cbcd79c2 JF |
173 | static unsigned long p2m_top_mfn_list[TOP_ENTRIES / P2M_ENTRIES_PER_PAGE] |
174 | __page_aligned_bss; | |
d5edbc1f | 175 | |
d451bb7a JF |
176 | static inline unsigned p2m_top_index(unsigned long pfn) |
177 | { | |
8006ec3e | 178 | BUG_ON(pfn >= MAX_DOMAIN_PAGES); |
d451bb7a JF |
179 | return pfn / P2M_ENTRIES_PER_PAGE; |
180 | } | |
181 | ||
182 | static inline unsigned p2m_index(unsigned long pfn) | |
183 | { | |
184 | return pfn % P2M_ENTRIES_PER_PAGE; | |
185 | } | |
186 | ||
d5edbc1f | 187 | /* Build the parallel p2m_top_mfn structures */ |
cdaead6b | 188 | static void __init xen_build_mfn_list_list(void) |
d5edbc1f JF |
189 | { |
190 | unsigned pfn, idx; | |
191 | ||
f63c2f24 | 192 | for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn += P2M_ENTRIES_PER_PAGE) { |
d5edbc1f JF |
193 | unsigned topidx = p2m_top_index(pfn); |
194 | ||
195 | p2m_top_mfn[topidx] = virt_to_mfn(p2m_top[topidx]); | |
196 | } | |
197 | ||
f63c2f24 | 198 | for (idx = 0; idx < ARRAY_SIZE(p2m_top_mfn_list); idx++) { |
d5edbc1f JF |
199 | unsigned topidx = idx * P2M_ENTRIES_PER_PAGE; |
200 | p2m_top_mfn_list[idx] = virt_to_mfn(&p2m_top_mfn[topidx]); | |
201 | } | |
cdaead6b | 202 | } |
d5edbc1f | 203 | |
cdaead6b JF |
204 | void xen_setup_mfn_list_list(void) |
205 | { | |
d5edbc1f JF |
206 | BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); |
207 | ||
208 | HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list = | |
209 | virt_to_mfn(p2m_top_mfn_list); | |
210 | HYPERVISOR_shared_info->arch.max_pfn = xen_start_info->nr_pages; | |
211 | } | |
212 | ||
213 | /* Set up p2m_top to point to the domain-builder provided p2m pages */ | |
d451bb7a JF |
214 | void __init xen_build_dynamic_phys_to_machine(void) |
215 | { | |
d451bb7a | 216 | unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list; |
8006ec3e | 217 | unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages); |
d5edbc1f | 218 | unsigned pfn; |
d451bb7a | 219 | |
f63c2f24 | 220 | for (pfn = 0; pfn < max_pfn; pfn += P2M_ENTRIES_PER_PAGE) { |
d451bb7a JF |
221 | unsigned topidx = p2m_top_index(pfn); |
222 | ||
223 | p2m_top[topidx] = &mfn_list[pfn]; | |
224 | } | |
cdaead6b JF |
225 | |
226 | xen_build_mfn_list_list(); | |
d451bb7a JF |
227 | } |
228 | ||
229 | unsigned long get_phys_to_machine(unsigned long pfn) | |
230 | { | |
231 | unsigned topidx, idx; | |
232 | ||
8006ec3e JF |
233 | if (unlikely(pfn >= MAX_DOMAIN_PAGES)) |
234 | return INVALID_P2M_ENTRY; | |
235 | ||
d451bb7a | 236 | topidx = p2m_top_index(pfn); |
d451bb7a JF |
237 | idx = p2m_index(pfn); |
238 | return p2m_top[topidx][idx]; | |
239 | } | |
15ce6005 | 240 | EXPORT_SYMBOL_GPL(get_phys_to_machine); |
d451bb7a | 241 | |
e791ca0f JF |
242 | /* install a new p2m_top page */ |
243 | bool install_p2mtop_page(unsigned long pfn, unsigned long *p) | |
d451bb7a | 244 | { |
e791ca0f JF |
245 | unsigned topidx = p2m_top_index(pfn); |
246 | unsigned long **pfnp, *mfnp; | |
d451bb7a JF |
247 | unsigned i; |
248 | ||
e791ca0f JF |
249 | pfnp = &p2m_top[topidx]; |
250 | mfnp = &p2m_top_mfn[topidx]; | |
d451bb7a | 251 | |
f63c2f24 | 252 | for (i = 0; i < P2M_ENTRIES_PER_PAGE; i++) |
d451bb7a JF |
253 | p[i] = INVALID_P2M_ENTRY; |
254 | ||
e791ca0f | 255 | if (cmpxchg(pfnp, p2m_missing, p) == p2m_missing) { |
d5edbc1f | 256 | *mfnp = virt_to_mfn(p); |
e791ca0f JF |
257 | return true; |
258 | } | |
259 | ||
260 | return false; | |
d451bb7a JF |
261 | } |
262 | ||
e791ca0f | 263 | static void alloc_p2m(unsigned long pfn) |
d451bb7a | 264 | { |
e791ca0f | 265 | unsigned long *p; |
d451bb7a | 266 | |
e791ca0f JF |
267 | p = (void *)__get_free_page(GFP_KERNEL | __GFP_NOFAIL); |
268 | BUG_ON(p == NULL); | |
269 | ||
270 | if (!install_p2mtop_page(pfn, p)) | |
271 | free_page((unsigned long)p); | |
272 | } | |
273 | ||
274 | /* Try to install p2m mapping; fail if intermediate bits missing */ | |
275 | bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn) | |
276 | { | |
277 | unsigned topidx, idx; | |
8006ec3e JF |
278 | |
279 | if (unlikely(pfn >= MAX_DOMAIN_PAGES)) { | |
280 | BUG_ON(mfn != INVALID_P2M_ENTRY); | |
e791ca0f | 281 | return true; |
d451bb7a JF |
282 | } |
283 | ||
284 | topidx = p2m_top_index(pfn); | |
cf0923ea | 285 | if (p2m_top[topidx] == p2m_missing) { |
d451bb7a | 286 | if (mfn == INVALID_P2M_ENTRY) |
e791ca0f JF |
287 | return true; |
288 | return false; | |
d451bb7a JF |
289 | } |
290 | ||
291 | idx = p2m_index(pfn); | |
292 | p2m_top[topidx][idx] = mfn; | |
e791ca0f JF |
293 | |
294 | return true; | |
295 | } | |
296 | ||
297 | void set_phys_to_machine(unsigned long pfn, unsigned long mfn) | |
298 | { | |
299 | if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) { | |
300 | BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY); | |
301 | return; | |
302 | } | |
303 | ||
304 | if (unlikely(!__set_phys_to_machine(pfn, mfn))) { | |
305 | alloc_p2m(pfn); | |
306 | ||
307 | if (!__set_phys_to_machine(pfn, mfn)) | |
308 | BUG(); | |
309 | } | |
d451bb7a JF |
310 | } |
311 | ||
9976b39b JF |
312 | unsigned long arbitrary_virt_to_mfn(void *vaddr) |
313 | { | |
314 | xmaddr_t maddr = arbitrary_virt_to_machine(vaddr); | |
315 | ||
316 | return PFN_DOWN(maddr.maddr); | |
317 | } | |
318 | ||
ce803e70 | 319 | xmaddr_t arbitrary_virt_to_machine(void *vaddr) |
3b827c1b | 320 | { |
ce803e70 | 321 | unsigned long address = (unsigned long)vaddr; |
da7bfc50 | 322 | unsigned int level; |
9f32d21c CL |
323 | pte_t *pte; |
324 | unsigned offset; | |
3b827c1b | 325 | |
9f32d21c CL |
326 | /* |
327 | * if the PFN is in the linear mapped vaddr range, we can just use | |
328 | * the (quick) virt_to_machine() p2m lookup | |
329 | */ | |
330 | if (virt_addr_valid(vaddr)) | |
331 | return virt_to_machine(vaddr); | |
332 | ||
333 | /* otherwise we have to do a (slower) full page-table walk */ | |
3b827c1b | 334 | |
9f32d21c CL |
335 | pte = lookup_address(address, &level); |
336 | BUG_ON(pte == NULL); | |
337 | offset = address & ~PAGE_MASK; | |
ebd879e3 | 338 | return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset); |
3b827c1b JF |
339 | } |
340 | ||
341 | void make_lowmem_page_readonly(void *vaddr) | |
342 | { | |
343 | pte_t *pte, ptev; | |
344 | unsigned long address = (unsigned long)vaddr; | |
da7bfc50 | 345 | unsigned int level; |
3b827c1b | 346 | |
f0646e43 | 347 | pte = lookup_address(address, &level); |
3b827c1b JF |
348 | BUG_ON(pte == NULL); |
349 | ||
350 | ptev = pte_wrprotect(*pte); | |
351 | ||
352 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
353 | BUG(); | |
354 | } | |
355 | ||
356 | void make_lowmem_page_readwrite(void *vaddr) | |
357 | { | |
358 | pte_t *pte, ptev; | |
359 | unsigned long address = (unsigned long)vaddr; | |
da7bfc50 | 360 | unsigned int level; |
3b827c1b | 361 | |
f0646e43 | 362 | pte = lookup_address(address, &level); |
3b827c1b JF |
363 | BUG_ON(pte == NULL); |
364 | ||
365 | ptev = pte_mkwrite(*pte); | |
366 | ||
367 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
368 | BUG(); | |
369 | } | |
370 | ||
371 | ||
7708ad64 | 372 | static bool xen_page_pinned(void *ptr) |
e2426cf8 JF |
373 | { |
374 | struct page *page = virt_to_page(ptr); | |
375 | ||
376 | return PagePinned(page); | |
377 | } | |
378 | ||
7708ad64 | 379 | static void xen_extend_mmu_update(const struct mmu_update *update) |
3b827c1b | 380 | { |
d66bf8fc JF |
381 | struct multicall_space mcs; |
382 | struct mmu_update *u; | |
3b827c1b | 383 | |
400d3494 JF |
384 | mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u)); |
385 | ||
994025ca JF |
386 | if (mcs.mc != NULL) { |
387 | ADD_STATS(mmu_update_extended, 1); | |
388 | ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1); | |
389 | ||
400d3494 | 390 | mcs.mc->args[1]++; |
994025ca JF |
391 | |
392 | if (mcs.mc->args[1] < MMU_UPDATE_HISTO) | |
393 | ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1); | |
394 | else | |
395 | ADD_STATS(mmu_update_histo[0], 1); | |
396 | } else { | |
397 | ADD_STATS(mmu_update, 1); | |
400d3494 JF |
398 | mcs = __xen_mc_entry(sizeof(*u)); |
399 | MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); | |
994025ca | 400 | ADD_STATS(mmu_update_histo[1], 1); |
400d3494 | 401 | } |
d66bf8fc | 402 | |
d66bf8fc | 403 | u = mcs.args; |
400d3494 JF |
404 | *u = *update; |
405 | } | |
406 | ||
407 | void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val) | |
408 | { | |
409 | struct mmu_update u; | |
410 | ||
411 | preempt_disable(); | |
412 | ||
413 | xen_mc_batch(); | |
414 | ||
ce803e70 JF |
415 | /* ptr may be ioremapped for 64-bit pagetable setup */ |
416 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
400d3494 | 417 | u.val = pmd_val_ma(val); |
7708ad64 | 418 | xen_extend_mmu_update(&u); |
d66bf8fc | 419 | |
994025ca JF |
420 | ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); |
421 | ||
d66bf8fc JF |
422 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
423 | ||
424 | preempt_enable(); | |
3b827c1b JF |
425 | } |
426 | ||
e2426cf8 JF |
427 | void xen_set_pmd(pmd_t *ptr, pmd_t val) |
428 | { | |
994025ca JF |
429 | ADD_STATS(pmd_update, 1); |
430 | ||
e2426cf8 JF |
431 | /* If page is not pinned, we can just update the entry |
432 | directly */ | |
7708ad64 | 433 | if (!xen_page_pinned(ptr)) { |
e2426cf8 JF |
434 | *ptr = val; |
435 | return; | |
436 | } | |
437 | ||
994025ca JF |
438 | ADD_STATS(pmd_update_pinned, 1); |
439 | ||
e2426cf8 JF |
440 | xen_set_pmd_hyper(ptr, val); |
441 | } | |
442 | ||
3b827c1b JF |
443 | /* |
444 | * Associate a virtual page frame with a given physical page frame | |
445 | * and protection flags for that frame. | |
446 | */ | |
447 | void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags) | |
448 | { | |
836fe2f2 | 449 | set_pte_vaddr(vaddr, mfn_pte(mfn, flags)); |
3b827c1b JF |
450 | } |
451 | ||
452 | void xen_set_pte_at(struct mm_struct *mm, unsigned long addr, | |
453 | pte_t *ptep, pte_t pteval) | |
454 | { | |
994025ca JF |
455 | ADD_STATS(set_pte_at, 1); |
456 | // ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep)); | |
457 | ADD_STATS(set_pte_at_current, mm == current->mm); | |
458 | ADD_STATS(set_pte_at_kernel, mm == &init_mm); | |
459 | ||
d66bf8fc | 460 | if (mm == current->mm || mm == &init_mm) { |
8965c1c0 | 461 | if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) { |
d66bf8fc JF |
462 | struct multicall_space mcs; |
463 | mcs = xen_mc_entry(0); | |
464 | ||
465 | MULTI_update_va_mapping(mcs.mc, addr, pteval, 0); | |
994025ca | 466 | ADD_STATS(set_pte_at_batched, 1); |
d66bf8fc | 467 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
2bd50036 | 468 | goto out; |
d66bf8fc JF |
469 | } else |
470 | if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0) | |
2bd50036 | 471 | goto out; |
d66bf8fc JF |
472 | } |
473 | xen_set_pte(ptep, pteval); | |
2bd50036 | 474 | |
2829b449 | 475 | out: return; |
3b827c1b JF |
476 | } |
477 | ||
f63c2f24 T |
478 | pte_t xen_ptep_modify_prot_start(struct mm_struct *mm, |
479 | unsigned long addr, pte_t *ptep) | |
947a69c9 | 480 | { |
e57778a1 JF |
481 | /* Just return the pte as-is. We preserve the bits on commit */ |
482 | return *ptep; | |
483 | } | |
484 | ||
485 | void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr, | |
486 | pte_t *ptep, pte_t pte) | |
487 | { | |
400d3494 | 488 | struct mmu_update u; |
e57778a1 | 489 | |
400d3494 | 490 | xen_mc_batch(); |
947a69c9 | 491 | |
9f32d21c | 492 | u.ptr = arbitrary_virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD; |
400d3494 | 493 | u.val = pte_val_ma(pte); |
7708ad64 | 494 | xen_extend_mmu_update(&u); |
947a69c9 | 495 | |
994025ca JF |
496 | ADD_STATS(prot_commit, 1); |
497 | ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); | |
498 | ||
e57778a1 | 499 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
947a69c9 JF |
500 | } |
501 | ||
ebb9cfe2 JF |
502 | /* Assume pteval_t is equivalent to all the other *val_t types. */ |
503 | static pteval_t pte_mfn_to_pfn(pteval_t val) | |
947a69c9 | 504 | { |
ebb9cfe2 | 505 | if (val & _PAGE_PRESENT) { |
59438c9f | 506 | unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; |
77be1fab | 507 | pteval_t flags = val & PTE_FLAGS_MASK; |
d8355aca | 508 | val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags; |
ebb9cfe2 | 509 | } |
947a69c9 | 510 | |
ebb9cfe2 | 511 | return val; |
947a69c9 JF |
512 | } |
513 | ||
ebb9cfe2 | 514 | static pteval_t pte_pfn_to_mfn(pteval_t val) |
947a69c9 | 515 | { |
ebb9cfe2 | 516 | if (val & _PAGE_PRESENT) { |
59438c9f | 517 | unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; |
77be1fab | 518 | pteval_t flags = val & PTE_FLAGS_MASK; |
d8355aca | 519 | val = ((pteval_t)pfn_to_mfn(pfn) << PAGE_SHIFT) | flags; |
947a69c9 JF |
520 | } |
521 | ||
ebb9cfe2 | 522 | return val; |
947a69c9 JF |
523 | } |
524 | ||
ebb9cfe2 | 525 | pteval_t xen_pte_val(pte_t pte) |
947a69c9 | 526 | { |
ebb9cfe2 | 527 | return pte_mfn_to_pfn(pte.pte); |
947a69c9 | 528 | } |
da5de7c2 | 529 | PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val); |
947a69c9 | 530 | |
947a69c9 JF |
531 | pgdval_t xen_pgd_val(pgd_t pgd) |
532 | { | |
ebb9cfe2 | 533 | return pte_mfn_to_pfn(pgd.pgd); |
947a69c9 | 534 | } |
da5de7c2 | 535 | PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val); |
947a69c9 JF |
536 | |
537 | pte_t xen_make_pte(pteval_t pte) | |
538 | { | |
ebb9cfe2 JF |
539 | pte = pte_pfn_to_mfn(pte); |
540 | return native_make_pte(pte); | |
947a69c9 | 541 | } |
da5de7c2 | 542 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte); |
947a69c9 JF |
543 | |
544 | pgd_t xen_make_pgd(pgdval_t pgd) | |
545 | { | |
ebb9cfe2 JF |
546 | pgd = pte_pfn_to_mfn(pgd); |
547 | return native_make_pgd(pgd); | |
947a69c9 | 548 | } |
da5de7c2 | 549 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd); |
947a69c9 JF |
550 | |
551 | pmdval_t xen_pmd_val(pmd_t pmd) | |
552 | { | |
ebb9cfe2 | 553 | return pte_mfn_to_pfn(pmd.pmd); |
947a69c9 | 554 | } |
da5de7c2 | 555 | PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val); |
28499143 | 556 | |
e2426cf8 | 557 | void xen_set_pud_hyper(pud_t *ptr, pud_t val) |
f4f97b3e | 558 | { |
400d3494 | 559 | struct mmu_update u; |
f4f97b3e | 560 | |
d66bf8fc JF |
561 | preempt_disable(); |
562 | ||
400d3494 JF |
563 | xen_mc_batch(); |
564 | ||
ce803e70 JF |
565 | /* ptr may be ioremapped for 64-bit pagetable setup */ |
566 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
400d3494 | 567 | u.val = pud_val_ma(val); |
7708ad64 | 568 | xen_extend_mmu_update(&u); |
d66bf8fc | 569 | |
994025ca JF |
570 | ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); |
571 | ||
d66bf8fc JF |
572 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
573 | ||
574 | preempt_enable(); | |
f4f97b3e JF |
575 | } |
576 | ||
e2426cf8 JF |
577 | void xen_set_pud(pud_t *ptr, pud_t val) |
578 | { | |
994025ca JF |
579 | ADD_STATS(pud_update, 1); |
580 | ||
e2426cf8 JF |
581 | /* If page is not pinned, we can just update the entry |
582 | directly */ | |
7708ad64 | 583 | if (!xen_page_pinned(ptr)) { |
e2426cf8 JF |
584 | *ptr = val; |
585 | return; | |
586 | } | |
587 | ||
994025ca JF |
588 | ADD_STATS(pud_update_pinned, 1); |
589 | ||
e2426cf8 JF |
590 | xen_set_pud_hyper(ptr, val); |
591 | } | |
592 | ||
f4f97b3e JF |
593 | void xen_set_pte(pte_t *ptep, pte_t pte) |
594 | { | |
994025ca JF |
595 | ADD_STATS(pte_update, 1); |
596 | // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep)); | |
597 | ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); | |
598 | ||
f6e58732 | 599 | #ifdef CONFIG_X86_PAE |
f4f97b3e JF |
600 | ptep->pte_high = pte.pte_high; |
601 | smp_wmb(); | |
602 | ptep->pte_low = pte.pte_low; | |
f6e58732 JF |
603 | #else |
604 | *ptep = pte; | |
605 | #endif | |
f4f97b3e JF |
606 | } |
607 | ||
f6e58732 | 608 | #ifdef CONFIG_X86_PAE |
3b827c1b JF |
609 | void xen_set_pte_atomic(pte_t *ptep, pte_t pte) |
610 | { | |
f6e58732 | 611 | set_64bit((u64 *)ptep, native_pte_val(pte)); |
3b827c1b JF |
612 | } |
613 | ||
614 | void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep) | |
615 | { | |
616 | ptep->pte_low = 0; | |
617 | smp_wmb(); /* make sure low gets written first */ | |
618 | ptep->pte_high = 0; | |
619 | } | |
620 | ||
621 | void xen_pmd_clear(pmd_t *pmdp) | |
622 | { | |
e2426cf8 | 623 | set_pmd(pmdp, __pmd(0)); |
3b827c1b | 624 | } |
f6e58732 | 625 | #endif /* CONFIG_X86_PAE */ |
3b827c1b | 626 | |
abf33038 | 627 | pmd_t xen_make_pmd(pmdval_t pmd) |
3b827c1b | 628 | { |
ebb9cfe2 | 629 | pmd = pte_pfn_to_mfn(pmd); |
947a69c9 | 630 | return native_make_pmd(pmd); |
3b827c1b | 631 | } |
da5de7c2 | 632 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd); |
3b827c1b | 633 | |
f6e58732 JF |
634 | #if PAGETABLE_LEVELS == 4 |
635 | pudval_t xen_pud_val(pud_t pud) | |
636 | { | |
637 | return pte_mfn_to_pfn(pud.pud); | |
638 | } | |
da5de7c2 | 639 | PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val); |
f6e58732 JF |
640 | |
641 | pud_t xen_make_pud(pudval_t pud) | |
642 | { | |
643 | pud = pte_pfn_to_mfn(pud); | |
644 | ||
645 | return native_make_pud(pud); | |
646 | } | |
da5de7c2 | 647 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud); |
f6e58732 | 648 | |
d6182fbf | 649 | pgd_t *xen_get_user_pgd(pgd_t *pgd) |
f6e58732 | 650 | { |
d6182fbf JF |
651 | pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK); |
652 | unsigned offset = pgd - pgd_page; | |
653 | pgd_t *user_ptr = NULL; | |
f6e58732 | 654 | |
d6182fbf JF |
655 | if (offset < pgd_index(USER_LIMIT)) { |
656 | struct page *page = virt_to_page(pgd_page); | |
657 | user_ptr = (pgd_t *)page->private; | |
658 | if (user_ptr) | |
659 | user_ptr += offset; | |
660 | } | |
f6e58732 | 661 | |
d6182fbf JF |
662 | return user_ptr; |
663 | } | |
664 | ||
665 | static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val) | |
666 | { | |
667 | struct mmu_update u; | |
f6e58732 JF |
668 | |
669 | u.ptr = virt_to_machine(ptr).maddr; | |
670 | u.val = pgd_val_ma(val); | |
7708ad64 | 671 | xen_extend_mmu_update(&u); |
d6182fbf JF |
672 | } |
673 | ||
674 | /* | |
675 | * Raw hypercall-based set_pgd, intended for in early boot before | |
676 | * there's a page structure. This implies: | |
677 | * 1. The only existing pagetable is the kernel's | |
678 | * 2. It is always pinned | |
679 | * 3. It has no user pagetable attached to it | |
680 | */ | |
681 | void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val) | |
682 | { | |
683 | preempt_disable(); | |
684 | ||
685 | xen_mc_batch(); | |
686 | ||
687 | __xen_set_pgd_hyper(ptr, val); | |
f6e58732 JF |
688 | |
689 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
690 | ||
691 | preempt_enable(); | |
692 | } | |
693 | ||
694 | void xen_set_pgd(pgd_t *ptr, pgd_t val) | |
695 | { | |
d6182fbf JF |
696 | pgd_t *user_ptr = xen_get_user_pgd(ptr); |
697 | ||
994025ca JF |
698 | ADD_STATS(pgd_update, 1); |
699 | ||
f6e58732 JF |
700 | /* If page is not pinned, we can just update the entry |
701 | directly */ | |
7708ad64 | 702 | if (!xen_page_pinned(ptr)) { |
f6e58732 | 703 | *ptr = val; |
d6182fbf | 704 | if (user_ptr) { |
7708ad64 | 705 | WARN_ON(xen_page_pinned(user_ptr)); |
d6182fbf JF |
706 | *user_ptr = val; |
707 | } | |
f6e58732 JF |
708 | return; |
709 | } | |
710 | ||
994025ca JF |
711 | ADD_STATS(pgd_update_pinned, 1); |
712 | ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); | |
713 | ||
d6182fbf JF |
714 | /* If it's pinned, then we can at least batch the kernel and |
715 | user updates together. */ | |
716 | xen_mc_batch(); | |
717 | ||
718 | __xen_set_pgd_hyper(ptr, val); | |
719 | if (user_ptr) | |
720 | __xen_set_pgd_hyper(user_ptr, val); | |
721 | ||
722 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
f6e58732 JF |
723 | } |
724 | #endif /* PAGETABLE_LEVELS == 4 */ | |
725 | ||
f4f97b3e | 726 | /* |
5deb30d1 JF |
727 | * (Yet another) pagetable walker. This one is intended for pinning a |
728 | * pagetable. This means that it walks a pagetable and calls the | |
729 | * callback function on each page it finds making up the page table, | |
730 | * at every level. It walks the entire pagetable, but it only bothers | |
731 | * pinning pte pages which are below limit. In the normal case this | |
732 | * will be STACK_TOP_MAX, but at boot we need to pin up to | |
733 | * FIXADDR_TOP. | |
734 | * | |
735 | * For 32-bit the important bit is that we don't pin beyond there, | |
736 | * because then we start getting into Xen's ptes. | |
737 | * | |
738 | * For 64-bit, we must skip the Xen hole in the middle of the address | |
739 | * space, just after the big x86-64 virtual hole. | |
740 | */ | |
86bbc2c2 IC |
741 | static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd, |
742 | int (*func)(struct mm_struct *mm, struct page *, | |
743 | enum pt_level), | |
744 | unsigned long limit) | |
3b827c1b | 745 | { |
f4f97b3e | 746 | int flush = 0; |
5deb30d1 JF |
747 | unsigned hole_low, hole_high; |
748 | unsigned pgdidx_limit, pudidx_limit, pmdidx_limit; | |
749 | unsigned pgdidx, pudidx, pmdidx; | |
f4f97b3e | 750 | |
5deb30d1 JF |
751 | /* The limit is the last byte to be touched */ |
752 | limit--; | |
753 | BUG_ON(limit >= FIXADDR_TOP); | |
3b827c1b JF |
754 | |
755 | if (xen_feature(XENFEAT_auto_translated_physmap)) | |
f4f97b3e JF |
756 | return 0; |
757 | ||
5deb30d1 JF |
758 | /* |
759 | * 64-bit has a great big hole in the middle of the address | |
760 | * space, which contains the Xen mappings. On 32-bit these | |
761 | * will end up making a zero-sized hole and so is a no-op. | |
762 | */ | |
d6182fbf | 763 | hole_low = pgd_index(USER_LIMIT); |
5deb30d1 JF |
764 | hole_high = pgd_index(PAGE_OFFSET); |
765 | ||
766 | pgdidx_limit = pgd_index(limit); | |
767 | #if PTRS_PER_PUD > 1 | |
768 | pudidx_limit = pud_index(limit); | |
769 | #else | |
770 | pudidx_limit = 0; | |
771 | #endif | |
772 | #if PTRS_PER_PMD > 1 | |
773 | pmdidx_limit = pmd_index(limit); | |
774 | #else | |
775 | pmdidx_limit = 0; | |
776 | #endif | |
777 | ||
5deb30d1 | 778 | for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) { |
f4f97b3e | 779 | pud_t *pud; |
3b827c1b | 780 | |
5deb30d1 JF |
781 | if (pgdidx >= hole_low && pgdidx < hole_high) |
782 | continue; | |
f4f97b3e | 783 | |
5deb30d1 | 784 | if (!pgd_val(pgd[pgdidx])) |
3b827c1b | 785 | continue; |
f4f97b3e | 786 | |
5deb30d1 | 787 | pud = pud_offset(&pgd[pgdidx], 0); |
3b827c1b JF |
788 | |
789 | if (PTRS_PER_PUD > 1) /* not folded */ | |
eefb47f6 | 790 | flush |= (*func)(mm, virt_to_page(pud), PT_PUD); |
f4f97b3e | 791 | |
5deb30d1 | 792 | for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) { |
f4f97b3e | 793 | pmd_t *pmd; |
f4f97b3e | 794 | |
5deb30d1 JF |
795 | if (pgdidx == pgdidx_limit && |
796 | pudidx > pudidx_limit) | |
797 | goto out; | |
3b827c1b | 798 | |
5deb30d1 | 799 | if (pud_none(pud[pudidx])) |
3b827c1b | 800 | continue; |
f4f97b3e | 801 | |
5deb30d1 | 802 | pmd = pmd_offset(&pud[pudidx], 0); |
3b827c1b JF |
803 | |
804 | if (PTRS_PER_PMD > 1) /* not folded */ | |
eefb47f6 | 805 | flush |= (*func)(mm, virt_to_page(pmd), PT_PMD); |
f4f97b3e | 806 | |
5deb30d1 JF |
807 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) { |
808 | struct page *pte; | |
809 | ||
810 | if (pgdidx == pgdidx_limit && | |
811 | pudidx == pudidx_limit && | |
812 | pmdidx > pmdidx_limit) | |
813 | goto out; | |
3b827c1b | 814 | |
5deb30d1 | 815 | if (pmd_none(pmd[pmdidx])) |
3b827c1b JF |
816 | continue; |
817 | ||
5deb30d1 | 818 | pte = pmd_page(pmd[pmdidx]); |
eefb47f6 | 819 | flush |= (*func)(mm, pte, PT_PTE); |
3b827c1b JF |
820 | } |
821 | } | |
822 | } | |
11ad93e5 | 823 | |
5deb30d1 | 824 | out: |
11ad93e5 JF |
825 | /* Do the top level last, so that the callbacks can use it as |
826 | a cue to do final things like tlb flushes. */ | |
eefb47f6 | 827 | flush |= (*func)(mm, virt_to_page(pgd), PT_PGD); |
f4f97b3e JF |
828 | |
829 | return flush; | |
3b827c1b JF |
830 | } |
831 | ||
86bbc2c2 IC |
832 | static int xen_pgd_walk(struct mm_struct *mm, |
833 | int (*func)(struct mm_struct *mm, struct page *, | |
834 | enum pt_level), | |
835 | unsigned long limit) | |
836 | { | |
837 | return __xen_pgd_walk(mm, mm->pgd, func, limit); | |
838 | } | |
839 | ||
7708ad64 JF |
840 | /* If we're using split pte locks, then take the page's lock and |
841 | return a pointer to it. Otherwise return NULL. */ | |
eefb47f6 | 842 | static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm) |
74260714 JF |
843 | { |
844 | spinlock_t *ptl = NULL; | |
845 | ||
f7d0b926 | 846 | #if USE_SPLIT_PTLOCKS |
74260714 | 847 | ptl = __pte_lockptr(page); |
eefb47f6 | 848 | spin_lock_nest_lock(ptl, &mm->page_table_lock); |
74260714 JF |
849 | #endif |
850 | ||
851 | return ptl; | |
852 | } | |
853 | ||
7708ad64 | 854 | static void xen_pte_unlock(void *v) |
74260714 JF |
855 | { |
856 | spinlock_t *ptl = v; | |
857 | spin_unlock(ptl); | |
858 | } | |
859 | ||
860 | static void xen_do_pin(unsigned level, unsigned long pfn) | |
861 | { | |
862 | struct mmuext_op *op; | |
863 | struct multicall_space mcs; | |
864 | ||
865 | mcs = __xen_mc_entry(sizeof(*op)); | |
866 | op = mcs.args; | |
867 | op->cmd = level; | |
868 | op->arg1.mfn = pfn_to_mfn(pfn); | |
869 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
870 | } | |
871 | ||
eefb47f6 JF |
872 | static int xen_pin_page(struct mm_struct *mm, struct page *page, |
873 | enum pt_level level) | |
f4f97b3e | 874 | { |
d60cd46b | 875 | unsigned pgfl = TestSetPagePinned(page); |
f4f97b3e JF |
876 | int flush; |
877 | ||
878 | if (pgfl) | |
879 | flush = 0; /* already pinned */ | |
880 | else if (PageHighMem(page)) | |
881 | /* kmaps need flushing if we found an unpinned | |
882 | highpage */ | |
883 | flush = 1; | |
884 | else { | |
885 | void *pt = lowmem_page_address(page); | |
886 | unsigned long pfn = page_to_pfn(page); | |
887 | struct multicall_space mcs = __xen_mc_entry(0); | |
74260714 | 888 | spinlock_t *ptl; |
f4f97b3e JF |
889 | |
890 | flush = 0; | |
891 | ||
11ad93e5 JF |
892 | /* |
893 | * We need to hold the pagetable lock between the time | |
894 | * we make the pagetable RO and when we actually pin | |
895 | * it. If we don't, then other users may come in and | |
896 | * attempt to update the pagetable by writing it, | |
897 | * which will fail because the memory is RO but not | |
898 | * pinned, so Xen won't do the trap'n'emulate. | |
899 | * | |
900 | * If we're using split pte locks, we can't hold the | |
901 | * entire pagetable's worth of locks during the | |
902 | * traverse, because we may wrap the preempt count (8 | |
903 | * bits). The solution is to mark RO and pin each PTE | |
904 | * page while holding the lock. This means the number | |
905 | * of locks we end up holding is never more than a | |
906 | * batch size (~32 entries, at present). | |
907 | * | |
908 | * If we're not using split pte locks, we needn't pin | |
909 | * the PTE pages independently, because we're | |
910 | * protected by the overall pagetable lock. | |
911 | */ | |
74260714 JF |
912 | ptl = NULL; |
913 | if (level == PT_PTE) | |
eefb47f6 | 914 | ptl = xen_pte_lock(page, mm); |
74260714 | 915 | |
f4f97b3e JF |
916 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, |
917 | pfn_pte(pfn, PAGE_KERNEL_RO), | |
74260714 JF |
918 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
919 | ||
11ad93e5 | 920 | if (ptl) { |
74260714 JF |
921 | xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn); |
922 | ||
74260714 JF |
923 | /* Queue a deferred unlock for when this batch |
924 | is completed. */ | |
7708ad64 | 925 | xen_mc_callback(xen_pte_unlock, ptl); |
74260714 | 926 | } |
f4f97b3e JF |
927 | } |
928 | ||
929 | return flush; | |
930 | } | |
3b827c1b | 931 | |
f4f97b3e JF |
932 | /* This is called just after a mm has been created, but it has not |
933 | been used yet. We need to make sure that its pagetable is all | |
934 | read-only, and can be pinned. */ | |
eefb47f6 | 935 | static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd) |
3b827c1b | 936 | { |
d05fdf31 JF |
937 | vm_unmap_aliases(); |
938 | ||
f4f97b3e | 939 | xen_mc_batch(); |
3b827c1b | 940 | |
86bbc2c2 | 941 | if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) { |
d05fdf31 | 942 | /* re-enable interrupts for flushing */ |
f87e4cac | 943 | xen_mc_issue(0); |
d05fdf31 | 944 | |
f4f97b3e | 945 | kmap_flush_unused(); |
d05fdf31 | 946 | |
f87e4cac JF |
947 | xen_mc_batch(); |
948 | } | |
f4f97b3e | 949 | |
d6182fbf JF |
950 | #ifdef CONFIG_X86_64 |
951 | { | |
952 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
953 | ||
954 | xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd))); | |
955 | ||
956 | if (user_pgd) { | |
eefb47f6 | 957 | xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD); |
f63c2f24 T |
958 | xen_do_pin(MMUEXT_PIN_L4_TABLE, |
959 | PFN_DOWN(__pa(user_pgd))); | |
d6182fbf JF |
960 | } |
961 | } | |
962 | #else /* CONFIG_X86_32 */ | |
5deb30d1 JF |
963 | #ifdef CONFIG_X86_PAE |
964 | /* Need to make sure unshared kernel PMD is pinnable */ | |
47cb2ed9 | 965 | xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), |
eefb47f6 | 966 | PT_PMD); |
5deb30d1 | 967 | #endif |
28499143 | 968 | xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd))); |
d6182fbf | 969 | #endif /* CONFIG_X86_64 */ |
f4f97b3e | 970 | xen_mc_issue(0); |
3b827c1b JF |
971 | } |
972 | ||
eefb47f6 JF |
973 | static void xen_pgd_pin(struct mm_struct *mm) |
974 | { | |
975 | __xen_pgd_pin(mm, mm->pgd); | |
976 | } | |
977 | ||
0e91398f JF |
978 | /* |
979 | * On save, we need to pin all pagetables to make sure they get their | |
980 | * mfns turned into pfns. Search the list for any unpinned pgds and pin | |
981 | * them (unpinned pgds are not currently in use, probably because the | |
982 | * process is under construction or destruction). | |
eefb47f6 JF |
983 | * |
984 | * Expected to be called in stop_machine() ("equivalent to taking | |
985 | * every spinlock in the system"), so the locking doesn't really | |
986 | * matter all that much. | |
0e91398f JF |
987 | */ |
988 | void xen_mm_pin_all(void) | |
989 | { | |
990 | unsigned long flags; | |
991 | struct page *page; | |
74260714 | 992 | |
0e91398f | 993 | spin_lock_irqsave(&pgd_lock, flags); |
f4f97b3e | 994 | |
0e91398f JF |
995 | list_for_each_entry(page, &pgd_list, lru) { |
996 | if (!PagePinned(page)) { | |
eefb47f6 | 997 | __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page)); |
0e91398f JF |
998 | SetPageSavePinned(page); |
999 | } | |
1000 | } | |
1001 | ||
1002 | spin_unlock_irqrestore(&pgd_lock, flags); | |
3b827c1b JF |
1003 | } |
1004 | ||
c1f2f09e EH |
1005 | /* |
1006 | * The init_mm pagetable is really pinned as soon as its created, but | |
1007 | * that's before we have page structures to store the bits. So do all | |
1008 | * the book-keeping now. | |
1009 | */ | |
eefb47f6 JF |
1010 | static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page, |
1011 | enum pt_level level) | |
3b827c1b | 1012 | { |
f4f97b3e JF |
1013 | SetPagePinned(page); |
1014 | return 0; | |
1015 | } | |
3b827c1b | 1016 | |
b96229b5 | 1017 | static void __init xen_mark_init_mm_pinned(void) |
f4f97b3e | 1018 | { |
eefb47f6 | 1019 | xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP); |
f4f97b3e | 1020 | } |
3b827c1b | 1021 | |
eefb47f6 JF |
1022 | static int xen_unpin_page(struct mm_struct *mm, struct page *page, |
1023 | enum pt_level level) | |
f4f97b3e | 1024 | { |
d60cd46b | 1025 | unsigned pgfl = TestClearPagePinned(page); |
3b827c1b | 1026 | |
f4f97b3e JF |
1027 | if (pgfl && !PageHighMem(page)) { |
1028 | void *pt = lowmem_page_address(page); | |
1029 | unsigned long pfn = page_to_pfn(page); | |
74260714 JF |
1030 | spinlock_t *ptl = NULL; |
1031 | struct multicall_space mcs; | |
1032 | ||
11ad93e5 JF |
1033 | /* |
1034 | * Do the converse to pin_page. If we're using split | |
1035 | * pte locks, we must be holding the lock for while | |
1036 | * the pte page is unpinned but still RO to prevent | |
1037 | * concurrent updates from seeing it in this | |
1038 | * partially-pinned state. | |
1039 | */ | |
74260714 | 1040 | if (level == PT_PTE) { |
eefb47f6 | 1041 | ptl = xen_pte_lock(page, mm); |
74260714 | 1042 | |
11ad93e5 JF |
1043 | if (ptl) |
1044 | xen_do_pin(MMUEXT_UNPIN_TABLE, pfn); | |
74260714 JF |
1045 | } |
1046 | ||
1047 | mcs = __xen_mc_entry(0); | |
f4f97b3e JF |
1048 | |
1049 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, | |
1050 | pfn_pte(pfn, PAGE_KERNEL), | |
74260714 JF |
1051 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
1052 | ||
1053 | if (ptl) { | |
1054 | /* unlock when batch completed */ | |
7708ad64 | 1055 | xen_mc_callback(xen_pte_unlock, ptl); |
74260714 | 1056 | } |
f4f97b3e JF |
1057 | } |
1058 | ||
1059 | return 0; /* never need to flush on unpin */ | |
3b827c1b JF |
1060 | } |
1061 | ||
f4f97b3e | 1062 | /* Release a pagetables pages back as normal RW */ |
eefb47f6 | 1063 | static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd) |
f4f97b3e | 1064 | { |
f4f97b3e JF |
1065 | xen_mc_batch(); |
1066 | ||
74260714 | 1067 | xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); |
f4f97b3e | 1068 | |
d6182fbf JF |
1069 | #ifdef CONFIG_X86_64 |
1070 | { | |
1071 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
1072 | ||
1073 | if (user_pgd) { | |
f63c2f24 T |
1074 | xen_do_pin(MMUEXT_UNPIN_TABLE, |
1075 | PFN_DOWN(__pa(user_pgd))); | |
eefb47f6 | 1076 | xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD); |
d6182fbf JF |
1077 | } |
1078 | } | |
1079 | #endif | |
1080 | ||
5deb30d1 JF |
1081 | #ifdef CONFIG_X86_PAE |
1082 | /* Need to make sure unshared kernel PMD is unpinned */ | |
47cb2ed9 | 1083 | xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), |
eefb47f6 | 1084 | PT_PMD); |
5deb30d1 | 1085 | #endif |
d6182fbf | 1086 | |
86bbc2c2 | 1087 | __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT); |
f4f97b3e JF |
1088 | |
1089 | xen_mc_issue(0); | |
1090 | } | |
3b827c1b | 1091 | |
eefb47f6 JF |
1092 | static void xen_pgd_unpin(struct mm_struct *mm) |
1093 | { | |
1094 | __xen_pgd_unpin(mm, mm->pgd); | |
1095 | } | |
1096 | ||
0e91398f JF |
1097 | /* |
1098 | * On resume, undo any pinning done at save, so that the rest of the | |
1099 | * kernel doesn't see any unexpected pinned pagetables. | |
1100 | */ | |
1101 | void xen_mm_unpin_all(void) | |
1102 | { | |
1103 | unsigned long flags; | |
1104 | struct page *page; | |
1105 | ||
1106 | spin_lock_irqsave(&pgd_lock, flags); | |
1107 | ||
1108 | list_for_each_entry(page, &pgd_list, lru) { | |
1109 | if (PageSavePinned(page)) { | |
1110 | BUG_ON(!PagePinned(page)); | |
eefb47f6 | 1111 | __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page)); |
0e91398f JF |
1112 | ClearPageSavePinned(page); |
1113 | } | |
1114 | } | |
1115 | ||
1116 | spin_unlock_irqrestore(&pgd_lock, flags); | |
1117 | } | |
1118 | ||
3b827c1b JF |
1119 | void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next) |
1120 | { | |
f4f97b3e | 1121 | spin_lock(&next->page_table_lock); |
eefb47f6 | 1122 | xen_pgd_pin(next); |
f4f97b3e | 1123 | spin_unlock(&next->page_table_lock); |
3b827c1b JF |
1124 | } |
1125 | ||
1126 | void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm) | |
1127 | { | |
f4f97b3e | 1128 | spin_lock(&mm->page_table_lock); |
eefb47f6 | 1129 | xen_pgd_pin(mm); |
f4f97b3e | 1130 | spin_unlock(&mm->page_table_lock); |
3b827c1b JF |
1131 | } |
1132 | ||
3b827c1b | 1133 | |
f87e4cac JF |
1134 | #ifdef CONFIG_SMP |
1135 | /* Another cpu may still have their %cr3 pointing at the pagetable, so | |
1136 | we need to repoint it somewhere else before we can unpin it. */ | |
1137 | static void drop_other_mm_ref(void *info) | |
1138 | { | |
1139 | struct mm_struct *mm = info; | |
ce87b3d3 | 1140 | struct mm_struct *active_mm; |
3b827c1b | 1141 | |
9eb912d1 | 1142 | active_mm = percpu_read(cpu_tlbstate.active_mm); |
ce87b3d3 JF |
1143 | |
1144 | if (active_mm == mm) | |
f87e4cac | 1145 | leave_mm(smp_processor_id()); |
9f79991d JF |
1146 | |
1147 | /* If this cpu still has a stale cr3 reference, then make sure | |
1148 | it has been flushed. */ | |
7fd7d83d | 1149 | if (percpu_read(xen_current_cr3) == __pa(mm->pgd)) |
9f79991d | 1150 | load_cr3(swapper_pg_dir); |
f87e4cac | 1151 | } |
3b827c1b | 1152 | |
7708ad64 | 1153 | static void xen_drop_mm_ref(struct mm_struct *mm) |
f87e4cac | 1154 | { |
e4d98207 | 1155 | cpumask_var_t mask; |
9f79991d JF |
1156 | unsigned cpu; |
1157 | ||
f87e4cac JF |
1158 | if (current->active_mm == mm) { |
1159 | if (current->mm == mm) | |
1160 | load_cr3(swapper_pg_dir); | |
1161 | else | |
1162 | leave_mm(smp_processor_id()); | |
9f79991d JF |
1163 | } |
1164 | ||
1165 | /* Get the "official" set of cpus referring to our pagetable. */ | |
e4d98207 MT |
1166 | if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) { |
1167 | for_each_online_cpu(cpu) { | |
1168 | if (!cpumask_test_cpu(cpu, &mm->cpu_vm_mask) | |
1169 | && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd)) | |
1170 | continue; | |
1171 | smp_call_function_single(cpu, drop_other_mm_ref, mm, 1); | |
1172 | } | |
1173 | return; | |
1174 | } | |
1175 | cpumask_copy(mask, &mm->cpu_vm_mask); | |
9f79991d JF |
1176 | |
1177 | /* It's possible that a vcpu may have a stale reference to our | |
1178 | cr3, because its in lazy mode, and it hasn't yet flushed | |
1179 | its set of pending hypercalls yet. In this case, we can | |
1180 | look at its actual current cr3 value, and force it to flush | |
1181 | if needed. */ | |
1182 | for_each_online_cpu(cpu) { | |
1183 | if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd)) | |
e4d98207 | 1184 | cpumask_set_cpu(cpu, mask); |
3b827c1b JF |
1185 | } |
1186 | ||
e4d98207 MT |
1187 | if (!cpumask_empty(mask)) |
1188 | smp_call_function_many(mask, drop_other_mm_ref, mm, 1); | |
1189 | free_cpumask_var(mask); | |
f87e4cac JF |
1190 | } |
1191 | #else | |
7708ad64 | 1192 | static void xen_drop_mm_ref(struct mm_struct *mm) |
f87e4cac JF |
1193 | { |
1194 | if (current->active_mm == mm) | |
1195 | load_cr3(swapper_pg_dir); | |
1196 | } | |
1197 | #endif | |
1198 | ||
1199 | /* | |
1200 | * While a process runs, Xen pins its pagetables, which means that the | |
1201 | * hypervisor forces it to be read-only, and it controls all updates | |
1202 | * to it. This means that all pagetable updates have to go via the | |
1203 | * hypervisor, which is moderately expensive. | |
1204 | * | |
1205 | * Since we're pulling the pagetable down, we switch to use init_mm, | |
1206 | * unpin old process pagetable and mark it all read-write, which | |
1207 | * allows further operations on it to be simple memory accesses. | |
1208 | * | |
1209 | * The only subtle point is that another CPU may be still using the | |
1210 | * pagetable because of lazy tlb flushing. This means we need need to | |
1211 | * switch all CPUs off this pagetable before we can unpin it. | |
1212 | */ | |
1213 | void xen_exit_mmap(struct mm_struct *mm) | |
1214 | { | |
1215 | get_cpu(); /* make sure we don't move around */ | |
7708ad64 | 1216 | xen_drop_mm_ref(mm); |
f87e4cac | 1217 | put_cpu(); |
3b827c1b | 1218 | |
f120f13e | 1219 | spin_lock(&mm->page_table_lock); |
df912ea4 JF |
1220 | |
1221 | /* pgd may not be pinned in the error exit path of execve */ | |
7708ad64 | 1222 | if (xen_page_pinned(mm->pgd)) |
eefb47f6 | 1223 | xen_pgd_unpin(mm); |
74260714 | 1224 | |
f120f13e | 1225 | spin_unlock(&mm->page_table_lock); |
3b827c1b | 1226 | } |
994025ca | 1227 | |
319f3ba5 JF |
1228 | static __init void xen_pagetable_setup_start(pgd_t *base) |
1229 | { | |
1230 | } | |
1231 | ||
1232 | static __init void xen_pagetable_setup_done(pgd_t *base) | |
1233 | { | |
1234 | xen_setup_shared_info(); | |
1235 | } | |
1236 | ||
1237 | static void xen_write_cr2(unsigned long cr2) | |
1238 | { | |
1239 | percpu_read(xen_vcpu)->arch.cr2 = cr2; | |
1240 | } | |
1241 | ||
1242 | static unsigned long xen_read_cr2(void) | |
1243 | { | |
1244 | return percpu_read(xen_vcpu)->arch.cr2; | |
1245 | } | |
1246 | ||
1247 | unsigned long xen_read_cr2_direct(void) | |
1248 | { | |
1249 | return percpu_read(xen_vcpu_info.arch.cr2); | |
1250 | } | |
1251 | ||
1252 | static void xen_flush_tlb(void) | |
1253 | { | |
1254 | struct mmuext_op *op; | |
1255 | struct multicall_space mcs; | |
1256 | ||
1257 | preempt_disable(); | |
1258 | ||
1259 | mcs = xen_mc_entry(sizeof(*op)); | |
1260 | ||
1261 | op = mcs.args; | |
1262 | op->cmd = MMUEXT_TLB_FLUSH_LOCAL; | |
1263 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1264 | ||
1265 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1266 | ||
1267 | preempt_enable(); | |
1268 | } | |
1269 | ||
1270 | static void xen_flush_tlb_single(unsigned long addr) | |
1271 | { | |
1272 | struct mmuext_op *op; | |
1273 | struct multicall_space mcs; | |
1274 | ||
1275 | preempt_disable(); | |
1276 | ||
1277 | mcs = xen_mc_entry(sizeof(*op)); | |
1278 | op = mcs.args; | |
1279 | op->cmd = MMUEXT_INVLPG_LOCAL; | |
1280 | op->arg1.linear_addr = addr & PAGE_MASK; | |
1281 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1282 | ||
1283 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1284 | ||
1285 | preempt_enable(); | |
1286 | } | |
1287 | ||
1288 | static void xen_flush_tlb_others(const struct cpumask *cpus, | |
1289 | struct mm_struct *mm, unsigned long va) | |
1290 | { | |
1291 | struct { | |
1292 | struct mmuext_op op; | |
1293 | DECLARE_BITMAP(mask, NR_CPUS); | |
1294 | } *args; | |
1295 | struct multicall_space mcs; | |
1296 | ||
e3f8a74e JF |
1297 | if (cpumask_empty(cpus)) |
1298 | return; /* nothing to do */ | |
319f3ba5 JF |
1299 | |
1300 | mcs = xen_mc_entry(sizeof(*args)); | |
1301 | args = mcs.args; | |
1302 | args->op.arg2.vcpumask = to_cpumask(args->mask); | |
1303 | ||
1304 | /* Remove us, and any offline CPUS. */ | |
1305 | cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask); | |
1306 | cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask)); | |
319f3ba5 JF |
1307 | |
1308 | if (va == TLB_FLUSH_ALL) { | |
1309 | args->op.cmd = MMUEXT_TLB_FLUSH_MULTI; | |
1310 | } else { | |
1311 | args->op.cmd = MMUEXT_INVLPG_MULTI; | |
1312 | args->op.arg1.linear_addr = va; | |
1313 | } | |
1314 | ||
1315 | MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF); | |
1316 | ||
319f3ba5 JF |
1317 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
1318 | } | |
1319 | ||
1320 | static unsigned long xen_read_cr3(void) | |
1321 | { | |
1322 | return percpu_read(xen_cr3); | |
1323 | } | |
1324 | ||
1325 | static void set_current_cr3(void *v) | |
1326 | { | |
1327 | percpu_write(xen_current_cr3, (unsigned long)v); | |
1328 | } | |
1329 | ||
1330 | static void __xen_write_cr3(bool kernel, unsigned long cr3) | |
1331 | { | |
1332 | struct mmuext_op *op; | |
1333 | struct multicall_space mcs; | |
1334 | unsigned long mfn; | |
1335 | ||
1336 | if (cr3) | |
1337 | mfn = pfn_to_mfn(PFN_DOWN(cr3)); | |
1338 | else | |
1339 | mfn = 0; | |
1340 | ||
1341 | WARN_ON(mfn == 0 && kernel); | |
1342 | ||
1343 | mcs = __xen_mc_entry(sizeof(*op)); | |
1344 | ||
1345 | op = mcs.args; | |
1346 | op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR; | |
1347 | op->arg1.mfn = mfn; | |
1348 | ||
1349 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1350 | ||
1351 | if (kernel) { | |
1352 | percpu_write(xen_cr3, cr3); | |
1353 | ||
1354 | /* Update xen_current_cr3 once the batch has actually | |
1355 | been submitted. */ | |
1356 | xen_mc_callback(set_current_cr3, (void *)cr3); | |
1357 | } | |
1358 | } | |
1359 | ||
1360 | static void xen_write_cr3(unsigned long cr3) | |
1361 | { | |
1362 | BUG_ON(preemptible()); | |
1363 | ||
1364 | xen_mc_batch(); /* disables interrupts */ | |
1365 | ||
1366 | /* Update while interrupts are disabled, so its atomic with | |
1367 | respect to ipis */ | |
1368 | percpu_write(xen_cr3, cr3); | |
1369 | ||
1370 | __xen_write_cr3(true, cr3); | |
1371 | ||
1372 | #ifdef CONFIG_X86_64 | |
1373 | { | |
1374 | pgd_t *user_pgd = xen_get_user_pgd(__va(cr3)); | |
1375 | if (user_pgd) | |
1376 | __xen_write_cr3(false, __pa(user_pgd)); | |
1377 | else | |
1378 | __xen_write_cr3(false, 0); | |
1379 | } | |
1380 | #endif | |
1381 | ||
1382 | xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ | |
1383 | } | |
1384 | ||
1385 | static int xen_pgd_alloc(struct mm_struct *mm) | |
1386 | { | |
1387 | pgd_t *pgd = mm->pgd; | |
1388 | int ret = 0; | |
1389 | ||
1390 | BUG_ON(PagePinned(virt_to_page(pgd))); | |
1391 | ||
1392 | #ifdef CONFIG_X86_64 | |
1393 | { | |
1394 | struct page *page = virt_to_page(pgd); | |
1395 | pgd_t *user_pgd; | |
1396 | ||
1397 | BUG_ON(page->private != 0); | |
1398 | ||
1399 | ret = -ENOMEM; | |
1400 | ||
1401 | user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO); | |
1402 | page->private = (unsigned long)user_pgd; | |
1403 | ||
1404 | if (user_pgd != NULL) { | |
1405 | user_pgd[pgd_index(VSYSCALL_START)] = | |
1406 | __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE); | |
1407 | ret = 0; | |
1408 | } | |
1409 | ||
1410 | BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd)))); | |
1411 | } | |
1412 | #endif | |
1413 | ||
1414 | return ret; | |
1415 | } | |
1416 | ||
1417 | static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd) | |
1418 | { | |
1419 | #ifdef CONFIG_X86_64 | |
1420 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
1421 | ||
1422 | if (user_pgd) | |
1423 | free_page((unsigned long)user_pgd); | |
1424 | #endif | |
1425 | } | |
1426 | ||
1f4f9315 JF |
1427 | #ifdef CONFIG_HIGHPTE |
1428 | static void *xen_kmap_atomic_pte(struct page *page, enum km_type type) | |
1429 | { | |
1430 | pgprot_t prot = PAGE_KERNEL; | |
1431 | ||
1432 | if (PagePinned(page)) | |
1433 | prot = PAGE_KERNEL_RO; | |
1434 | ||
1435 | if (0 && PageHighMem(page)) | |
1436 | printk("mapping highpte %lx type %d prot %s\n", | |
1437 | page_to_pfn(page), type, | |
1438 | (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ"); | |
1439 | ||
1440 | return kmap_atomic_prot(page, type, prot); | |
1441 | } | |
1442 | #endif | |
1443 | ||
1444 | #ifdef CONFIG_X86_32 | |
1445 | static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte) | |
1446 | { | |
1447 | /* If there's an existing pte, then don't allow _PAGE_RW to be set */ | |
1448 | if (pte_val_ma(*ptep) & _PAGE_PRESENT) | |
1449 | pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) & | |
1450 | pte_val_ma(pte)); | |
1451 | ||
1452 | return pte; | |
1453 | } | |
1454 | ||
1455 | /* Init-time set_pte while constructing initial pagetables, which | |
1456 | doesn't allow RO pagetable pages to be remapped RW */ | |
1457 | static __init void xen_set_pte_init(pte_t *ptep, pte_t pte) | |
1458 | { | |
1459 | pte = mask_rw_pte(ptep, pte); | |
1460 | ||
1461 | xen_set_pte(ptep, pte); | |
1462 | } | |
1463 | #endif | |
319f3ba5 | 1464 | |
b96229b5 JF |
1465 | static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn) |
1466 | { | |
1467 | struct mmuext_op op; | |
1468 | op.cmd = cmd; | |
1469 | op.arg1.mfn = pfn_to_mfn(pfn); | |
1470 | if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) | |
1471 | BUG(); | |
1472 | } | |
1473 | ||
319f3ba5 JF |
1474 | /* Early in boot, while setting up the initial pagetable, assume |
1475 | everything is pinned. */ | |
1476 | static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn) | |
1477 | { | |
b96229b5 JF |
1478 | #ifdef CONFIG_FLATMEM |
1479 | BUG_ON(mem_map); /* should only be used early */ | |
1480 | #endif | |
1481 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1482 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
1483 | } | |
1484 | ||
1485 | /* Used for pmd and pud */ | |
1486 | static __init void xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn) | |
1487 | { | |
319f3ba5 JF |
1488 | #ifdef CONFIG_FLATMEM |
1489 | BUG_ON(mem_map); /* should only be used early */ | |
1490 | #endif | |
1491 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1492 | } | |
1493 | ||
1494 | /* Early release_pte assumes that all pts are pinned, since there's | |
1495 | only init_mm and anything attached to that is pinned. */ | |
b96229b5 | 1496 | static __init void xen_release_pte_init(unsigned long pfn) |
319f3ba5 | 1497 | { |
b96229b5 | 1498 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); |
319f3ba5 JF |
1499 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); |
1500 | } | |
1501 | ||
b96229b5 | 1502 | static __init void xen_release_pmd_init(unsigned long pfn) |
319f3ba5 | 1503 | { |
b96229b5 | 1504 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); |
319f3ba5 JF |
1505 | } |
1506 | ||
1507 | /* This needs to make sure the new pte page is pinned iff its being | |
1508 | attached to a pinned pagetable. */ | |
1509 | static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level) | |
1510 | { | |
1511 | struct page *page = pfn_to_page(pfn); | |
1512 | ||
1513 | if (PagePinned(virt_to_page(mm->pgd))) { | |
1514 | SetPagePinned(page); | |
1515 | ||
1516 | vm_unmap_aliases(); | |
1517 | if (!PageHighMem(page)) { | |
1518 | make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn))); | |
1519 | if (level == PT_PTE && USE_SPLIT_PTLOCKS) | |
1520 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
1521 | } else { | |
1522 | /* make sure there are no stray mappings of | |
1523 | this page */ | |
1524 | kmap_flush_unused(); | |
1525 | } | |
1526 | } | |
1527 | } | |
1528 | ||
1529 | static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn) | |
1530 | { | |
1531 | xen_alloc_ptpage(mm, pfn, PT_PTE); | |
1532 | } | |
1533 | ||
1534 | static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn) | |
1535 | { | |
1536 | xen_alloc_ptpage(mm, pfn, PT_PMD); | |
1537 | } | |
1538 | ||
1539 | /* This should never happen until we're OK to use struct page */ | |
1540 | static void xen_release_ptpage(unsigned long pfn, unsigned level) | |
1541 | { | |
1542 | struct page *page = pfn_to_page(pfn); | |
1543 | ||
1544 | if (PagePinned(page)) { | |
1545 | if (!PageHighMem(page)) { | |
1546 | if (level == PT_PTE && USE_SPLIT_PTLOCKS) | |
1547 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); | |
1548 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
1549 | } | |
1550 | ClearPagePinned(page); | |
1551 | } | |
1552 | } | |
1553 | ||
1554 | static void xen_release_pte(unsigned long pfn) | |
1555 | { | |
1556 | xen_release_ptpage(pfn, PT_PTE); | |
1557 | } | |
1558 | ||
1559 | static void xen_release_pmd(unsigned long pfn) | |
1560 | { | |
1561 | xen_release_ptpage(pfn, PT_PMD); | |
1562 | } | |
1563 | ||
1564 | #if PAGETABLE_LEVELS == 4 | |
1565 | static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn) | |
1566 | { | |
1567 | xen_alloc_ptpage(mm, pfn, PT_PUD); | |
1568 | } | |
1569 | ||
1570 | static void xen_release_pud(unsigned long pfn) | |
1571 | { | |
1572 | xen_release_ptpage(pfn, PT_PUD); | |
1573 | } | |
1574 | #endif | |
1575 | ||
1576 | void __init xen_reserve_top(void) | |
1577 | { | |
1578 | #ifdef CONFIG_X86_32 | |
1579 | unsigned long top = HYPERVISOR_VIRT_START; | |
1580 | struct xen_platform_parameters pp; | |
1581 | ||
1582 | if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0) | |
1583 | top = pp.virt_start; | |
1584 | ||
1585 | reserve_top_address(-top); | |
1586 | #endif /* CONFIG_X86_32 */ | |
1587 | } | |
1588 | ||
1589 | /* | |
1590 | * Like __va(), but returns address in the kernel mapping (which is | |
1591 | * all we have until the physical memory mapping has been set up. | |
1592 | */ | |
1593 | static void *__ka(phys_addr_t paddr) | |
1594 | { | |
1595 | #ifdef CONFIG_X86_64 | |
1596 | return (void *)(paddr + __START_KERNEL_map); | |
1597 | #else | |
1598 | return __va(paddr); | |
1599 | #endif | |
1600 | } | |
1601 | ||
1602 | /* Convert a machine address to physical address */ | |
1603 | static unsigned long m2p(phys_addr_t maddr) | |
1604 | { | |
1605 | phys_addr_t paddr; | |
1606 | ||
1607 | maddr &= PTE_PFN_MASK; | |
1608 | paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT; | |
1609 | ||
1610 | return paddr; | |
1611 | } | |
1612 | ||
1613 | /* Convert a machine address to kernel virtual */ | |
1614 | static void *m2v(phys_addr_t maddr) | |
1615 | { | |
1616 | return __ka(m2p(maddr)); | |
1617 | } | |
1618 | ||
1619 | static void set_page_prot(void *addr, pgprot_t prot) | |
1620 | { | |
1621 | unsigned long pfn = __pa(addr) >> PAGE_SHIFT; | |
1622 | pte_t pte = pfn_pte(pfn, prot); | |
1623 | ||
1624 | if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0)) | |
1625 | BUG(); | |
1626 | } | |
1627 | ||
1628 | static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn) | |
1629 | { | |
1630 | unsigned pmdidx, pteidx; | |
1631 | unsigned ident_pte; | |
1632 | unsigned long pfn; | |
1633 | ||
1634 | ident_pte = 0; | |
1635 | pfn = 0; | |
1636 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) { | |
1637 | pte_t *pte_page; | |
1638 | ||
1639 | /* Reuse or allocate a page of ptes */ | |
1640 | if (pmd_present(pmd[pmdidx])) | |
1641 | pte_page = m2v(pmd[pmdidx].pmd); | |
1642 | else { | |
1643 | /* Check for free pte pages */ | |
1644 | if (ident_pte == ARRAY_SIZE(level1_ident_pgt)) | |
1645 | break; | |
1646 | ||
1647 | pte_page = &level1_ident_pgt[ident_pte]; | |
1648 | ident_pte += PTRS_PER_PTE; | |
1649 | ||
1650 | pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE); | |
1651 | } | |
1652 | ||
1653 | /* Install mappings */ | |
1654 | for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) { | |
1655 | pte_t pte; | |
1656 | ||
1657 | if (pfn > max_pfn_mapped) | |
1658 | max_pfn_mapped = pfn; | |
1659 | ||
1660 | if (!pte_none(pte_page[pteidx])) | |
1661 | continue; | |
1662 | ||
1663 | pte = pfn_pte(pfn, PAGE_KERNEL_EXEC); | |
1664 | pte_page[pteidx] = pte; | |
1665 | } | |
1666 | } | |
1667 | ||
1668 | for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE) | |
1669 | set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO); | |
1670 | ||
1671 | set_page_prot(pmd, PAGE_KERNEL_RO); | |
1672 | } | |
1673 | ||
1674 | #ifdef CONFIG_X86_64 | |
1675 | static void convert_pfn_mfn(void *v) | |
1676 | { | |
1677 | pte_t *pte = v; | |
1678 | int i; | |
1679 | ||
1680 | /* All levels are converted the same way, so just treat them | |
1681 | as ptes. */ | |
1682 | for (i = 0; i < PTRS_PER_PTE; i++) | |
1683 | pte[i] = xen_make_pte(pte[i].pte); | |
1684 | } | |
1685 | ||
1686 | /* | |
1687 | * Set up the inital kernel pagetable. | |
1688 | * | |
1689 | * We can construct this by grafting the Xen provided pagetable into | |
1690 | * head_64.S's preconstructed pagetables. We copy the Xen L2's into | |
1691 | * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This | |
1692 | * means that only the kernel has a physical mapping to start with - | |
1693 | * but that's enough to get __va working. We need to fill in the rest | |
1694 | * of the physical mapping once some sort of allocator has been set | |
1695 | * up. | |
1696 | */ | |
1697 | __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, | |
1698 | unsigned long max_pfn) | |
1699 | { | |
1700 | pud_t *l3; | |
1701 | pmd_t *l2; | |
1702 | ||
1703 | /* Zap identity mapping */ | |
1704 | init_level4_pgt[0] = __pgd(0); | |
1705 | ||
1706 | /* Pre-constructed entries are in pfn, so convert to mfn */ | |
1707 | convert_pfn_mfn(init_level4_pgt); | |
1708 | convert_pfn_mfn(level3_ident_pgt); | |
1709 | convert_pfn_mfn(level3_kernel_pgt); | |
1710 | ||
1711 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd); | |
1712 | l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud); | |
1713 | ||
1714 | memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1715 | memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1716 | ||
1717 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd); | |
1718 | l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud); | |
1719 | memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1720 | ||
1721 | /* Set up identity map */ | |
1722 | xen_map_identity_early(level2_ident_pgt, max_pfn); | |
1723 | ||
1724 | /* Make pagetable pieces RO */ | |
1725 | set_page_prot(init_level4_pgt, PAGE_KERNEL_RO); | |
1726 | set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO); | |
1727 | set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO); | |
1728 | set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO); | |
1729 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); | |
1730 | set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO); | |
1731 | ||
1732 | /* Pin down new L4 */ | |
1733 | pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE, | |
1734 | PFN_DOWN(__pa_symbol(init_level4_pgt))); | |
1735 | ||
1736 | /* Unpin Xen-provided one */ | |
1737 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
1738 | ||
1739 | /* Switch over */ | |
1740 | pgd = init_level4_pgt; | |
1741 | ||
1742 | /* | |
1743 | * At this stage there can be no user pgd, and no page | |
1744 | * structure to attach it to, so make sure we just set kernel | |
1745 | * pgd. | |
1746 | */ | |
1747 | xen_mc_batch(); | |
1748 | __xen_write_cr3(true, __pa(pgd)); | |
1749 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
1750 | ||
1751 | reserve_early(__pa(xen_start_info->pt_base), | |
1752 | __pa(xen_start_info->pt_base + | |
1753 | xen_start_info->nr_pt_frames * PAGE_SIZE), | |
1754 | "XEN PAGETABLES"); | |
1755 | ||
1756 | return pgd; | |
1757 | } | |
1758 | #else /* !CONFIG_X86_64 */ | |
1759 | static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss; | |
1760 | ||
1761 | __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, | |
1762 | unsigned long max_pfn) | |
1763 | { | |
1764 | pmd_t *kernel_pmd; | |
1765 | ||
93dbda7c JF |
1766 | max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) + |
1767 | xen_start_info->nr_pt_frames * PAGE_SIZE + | |
1768 | 512*1024); | |
319f3ba5 JF |
1769 | |
1770 | kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd); | |
1771 | memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD); | |
1772 | ||
1773 | xen_map_identity_early(level2_kernel_pgt, max_pfn); | |
1774 | ||
1775 | memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD); | |
1776 | set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY], | |
1777 | __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT)); | |
1778 | ||
1779 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); | |
1780 | set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO); | |
1781 | set_page_prot(empty_zero_page, PAGE_KERNEL_RO); | |
1782 | ||
1783 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
1784 | ||
1785 | xen_write_cr3(__pa(swapper_pg_dir)); | |
1786 | ||
1787 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir))); | |
1788 | ||
33df4db0 JF |
1789 | reserve_early(__pa(xen_start_info->pt_base), |
1790 | __pa(xen_start_info->pt_base + | |
1791 | xen_start_info->nr_pt_frames * PAGE_SIZE), | |
1792 | "XEN PAGETABLES"); | |
1793 | ||
319f3ba5 JF |
1794 | return swapper_pg_dir; |
1795 | } | |
1796 | #endif /* CONFIG_X86_64 */ | |
1797 | ||
3b3809ac | 1798 | static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot) |
319f3ba5 JF |
1799 | { |
1800 | pte_t pte; | |
1801 | ||
1802 | phys >>= PAGE_SHIFT; | |
1803 | ||
1804 | switch (idx) { | |
1805 | case FIX_BTMAP_END ... FIX_BTMAP_BEGIN: | |
1806 | #ifdef CONFIG_X86_F00F_BUG | |
1807 | case FIX_F00F_IDT: | |
1808 | #endif | |
1809 | #ifdef CONFIG_X86_32 | |
1810 | case FIX_WP_TEST: | |
1811 | case FIX_VDSO: | |
1812 | # ifdef CONFIG_HIGHMEM | |
1813 | case FIX_KMAP_BEGIN ... FIX_KMAP_END: | |
1814 | # endif | |
1815 | #else | |
1816 | case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE: | |
1817 | #endif | |
1818 | #ifdef CONFIG_X86_LOCAL_APIC | |
1819 | case FIX_APIC_BASE: /* maps dummy local APIC */ | |
1820 | #endif | |
3ecb1b7d JF |
1821 | case FIX_TEXT_POKE0: |
1822 | case FIX_TEXT_POKE1: | |
1823 | /* All local page mappings */ | |
319f3ba5 JF |
1824 | pte = pfn_pte(phys, prot); |
1825 | break; | |
1826 | ||
1827 | default: | |
1828 | pte = mfn_pte(phys, prot); | |
1829 | break; | |
1830 | } | |
1831 | ||
1832 | __native_set_fixmap(idx, pte); | |
1833 | ||
1834 | #ifdef CONFIG_X86_64 | |
1835 | /* Replicate changes to map the vsyscall page into the user | |
1836 | pagetable vsyscall mapping. */ | |
1837 | if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) { | |
1838 | unsigned long vaddr = __fix_to_virt(idx); | |
1839 | set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte); | |
1840 | } | |
1841 | #endif | |
1842 | } | |
1843 | ||
1844 | __init void xen_post_allocator_init(void) | |
1845 | { | |
1846 | pv_mmu_ops.set_pte = xen_set_pte; | |
1847 | pv_mmu_ops.set_pmd = xen_set_pmd; | |
1848 | pv_mmu_ops.set_pud = xen_set_pud; | |
1849 | #if PAGETABLE_LEVELS == 4 | |
1850 | pv_mmu_ops.set_pgd = xen_set_pgd; | |
1851 | #endif | |
1852 | ||
1853 | /* This will work as long as patching hasn't happened yet | |
1854 | (which it hasn't) */ | |
1855 | pv_mmu_ops.alloc_pte = xen_alloc_pte; | |
1856 | pv_mmu_ops.alloc_pmd = xen_alloc_pmd; | |
1857 | pv_mmu_ops.release_pte = xen_release_pte; | |
1858 | pv_mmu_ops.release_pmd = xen_release_pmd; | |
1859 | #if PAGETABLE_LEVELS == 4 | |
1860 | pv_mmu_ops.alloc_pud = xen_alloc_pud; | |
1861 | pv_mmu_ops.release_pud = xen_release_pud; | |
1862 | #endif | |
1863 | ||
1864 | #ifdef CONFIG_X86_64 | |
1865 | SetPagePinned(virt_to_page(level3_user_vsyscall)); | |
1866 | #endif | |
1867 | xen_mark_init_mm_pinned(); | |
1868 | } | |
1869 | ||
b407fc57 JF |
1870 | static void xen_leave_lazy_mmu(void) |
1871 | { | |
5caecb94 | 1872 | preempt_disable(); |
b407fc57 JF |
1873 | xen_mc_flush(); |
1874 | paravirt_leave_lazy_mmu(); | |
5caecb94 | 1875 | preempt_enable(); |
b407fc57 | 1876 | } |
319f3ba5 JF |
1877 | |
1878 | const struct pv_mmu_ops xen_mmu_ops __initdata = { | |
1879 | .pagetable_setup_start = xen_pagetable_setup_start, | |
1880 | .pagetable_setup_done = xen_pagetable_setup_done, | |
1881 | ||
1882 | .read_cr2 = xen_read_cr2, | |
1883 | .write_cr2 = xen_write_cr2, | |
1884 | ||
1885 | .read_cr3 = xen_read_cr3, | |
1886 | .write_cr3 = xen_write_cr3, | |
1887 | ||
1888 | .flush_tlb_user = xen_flush_tlb, | |
1889 | .flush_tlb_kernel = xen_flush_tlb, | |
1890 | .flush_tlb_single = xen_flush_tlb_single, | |
1891 | .flush_tlb_others = xen_flush_tlb_others, | |
1892 | ||
1893 | .pte_update = paravirt_nop, | |
1894 | .pte_update_defer = paravirt_nop, | |
1895 | ||
1896 | .pgd_alloc = xen_pgd_alloc, | |
1897 | .pgd_free = xen_pgd_free, | |
1898 | ||
1899 | .alloc_pte = xen_alloc_pte_init, | |
1900 | .release_pte = xen_release_pte_init, | |
b96229b5 | 1901 | .alloc_pmd = xen_alloc_pmd_init, |
319f3ba5 | 1902 | .alloc_pmd_clone = paravirt_nop, |
b96229b5 | 1903 | .release_pmd = xen_release_pmd_init, |
319f3ba5 JF |
1904 | |
1905 | #ifdef CONFIG_HIGHPTE | |
1906 | .kmap_atomic_pte = xen_kmap_atomic_pte, | |
1907 | #endif | |
1908 | ||
1909 | #ifdef CONFIG_X86_64 | |
1910 | .set_pte = xen_set_pte, | |
1911 | #else | |
1912 | .set_pte = xen_set_pte_init, | |
1913 | #endif | |
1914 | .set_pte_at = xen_set_pte_at, | |
1915 | .set_pmd = xen_set_pmd_hyper, | |
1916 | ||
1917 | .ptep_modify_prot_start = __ptep_modify_prot_start, | |
1918 | .ptep_modify_prot_commit = __ptep_modify_prot_commit, | |
1919 | ||
da5de7c2 JF |
1920 | .pte_val = PV_CALLEE_SAVE(xen_pte_val), |
1921 | .pgd_val = PV_CALLEE_SAVE(xen_pgd_val), | |
319f3ba5 | 1922 | |
da5de7c2 JF |
1923 | .make_pte = PV_CALLEE_SAVE(xen_make_pte), |
1924 | .make_pgd = PV_CALLEE_SAVE(xen_make_pgd), | |
319f3ba5 JF |
1925 | |
1926 | #ifdef CONFIG_X86_PAE | |
1927 | .set_pte_atomic = xen_set_pte_atomic, | |
319f3ba5 JF |
1928 | .pte_clear = xen_pte_clear, |
1929 | .pmd_clear = xen_pmd_clear, | |
1930 | #endif /* CONFIG_X86_PAE */ | |
1931 | .set_pud = xen_set_pud_hyper, | |
1932 | ||
da5de7c2 JF |
1933 | .make_pmd = PV_CALLEE_SAVE(xen_make_pmd), |
1934 | .pmd_val = PV_CALLEE_SAVE(xen_pmd_val), | |
319f3ba5 JF |
1935 | |
1936 | #if PAGETABLE_LEVELS == 4 | |
da5de7c2 JF |
1937 | .pud_val = PV_CALLEE_SAVE(xen_pud_val), |
1938 | .make_pud = PV_CALLEE_SAVE(xen_make_pud), | |
319f3ba5 JF |
1939 | .set_pgd = xen_set_pgd_hyper, |
1940 | ||
b96229b5 JF |
1941 | .alloc_pud = xen_alloc_pmd_init, |
1942 | .release_pud = xen_release_pmd_init, | |
319f3ba5 JF |
1943 | #endif /* PAGETABLE_LEVELS == 4 */ |
1944 | ||
1945 | .activate_mm = xen_activate_mm, | |
1946 | .dup_mmap = xen_dup_mmap, | |
1947 | .exit_mmap = xen_exit_mmap, | |
1948 | ||
1949 | .lazy_mode = { | |
1950 | .enter = paravirt_enter_lazy_mmu, | |
b407fc57 | 1951 | .leave = xen_leave_lazy_mmu, |
319f3ba5 JF |
1952 | }, |
1953 | ||
1954 | .set_fixmap = xen_set_fixmap, | |
1955 | }; | |
1956 | ||
1957 | ||
994025ca JF |
1958 | #ifdef CONFIG_XEN_DEBUG_FS |
1959 | ||
1960 | static struct dentry *d_mmu_debug; | |
1961 | ||
1962 | static int __init xen_mmu_debugfs(void) | |
1963 | { | |
1964 | struct dentry *d_xen = xen_init_debugfs(); | |
1965 | ||
1966 | if (d_xen == NULL) | |
1967 | return -ENOMEM; | |
1968 | ||
1969 | d_mmu_debug = debugfs_create_dir("mmu", d_xen); | |
1970 | ||
1971 | debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats); | |
1972 | ||
1973 | debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update); | |
1974 | debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug, | |
1975 | &mmu_stats.pgd_update_pinned); | |
1976 | debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug, | |
1977 | &mmu_stats.pgd_update_pinned); | |
1978 | ||
1979 | debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update); | |
1980 | debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug, | |
1981 | &mmu_stats.pud_update_pinned); | |
1982 | debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug, | |
1983 | &mmu_stats.pud_update_pinned); | |
1984 | ||
1985 | debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update); | |
1986 | debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug, | |
1987 | &mmu_stats.pmd_update_pinned); | |
1988 | debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug, | |
1989 | &mmu_stats.pmd_update_pinned); | |
1990 | ||
1991 | debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update); | |
1992 | // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug, | |
1993 | // &mmu_stats.pte_update_pinned); | |
1994 | debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug, | |
1995 | &mmu_stats.pte_update_pinned); | |
1996 | ||
1997 | debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update); | |
1998 | debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug, | |
1999 | &mmu_stats.mmu_update_extended); | |
2000 | xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug, | |
2001 | mmu_stats.mmu_update_histo, 20); | |
2002 | ||
2003 | debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at); | |
2004 | debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug, | |
2005 | &mmu_stats.set_pte_at_batched); | |
2006 | debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug, | |
2007 | &mmu_stats.set_pte_at_current); | |
2008 | debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug, | |
2009 | &mmu_stats.set_pte_at_kernel); | |
2010 | ||
2011 | debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit); | |
2012 | debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug, | |
2013 | &mmu_stats.prot_commit_batched); | |
2014 | ||
2015 | return 0; | |
2016 | } | |
2017 | fs_initcall(xen_mmu_debugfs); | |
2018 | ||
2019 | #endif /* CONFIG_XEN_DEBUG_FS */ |