]>
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 | 186 | /* Build the parallel p2m_top_mfn structures */ |
cdaead6b | 187 | static void __init xen_build_mfn_list_list(void) |
d5edbc1f JF |
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 | } | |
cdaead6b | 201 | } |
d5edbc1f | 202 | |
cdaead6b JF |
203 | void xen_setup_mfn_list_list(void) |
204 | { | |
d5edbc1f JF |
205 | BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); |
206 | ||
207 | HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list = | |
208 | virt_to_mfn(p2m_top_mfn_list); | |
209 | HYPERVISOR_shared_info->arch.max_pfn = xen_start_info->nr_pages; | |
210 | } | |
211 | ||
212 | /* Set up p2m_top to point to the domain-builder provided p2m pages */ | |
d451bb7a JF |
213 | void __init xen_build_dynamic_phys_to_machine(void) |
214 | { | |
d451bb7a | 215 | unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list; |
8006ec3e | 216 | unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages); |
d5edbc1f | 217 | unsigned pfn; |
d451bb7a | 218 | |
f63c2f24 | 219 | for (pfn = 0; pfn < max_pfn; pfn += P2M_ENTRIES_PER_PAGE) { |
d451bb7a JF |
220 | unsigned topidx = p2m_top_index(pfn); |
221 | ||
222 | p2m_top[topidx] = &mfn_list[pfn]; | |
223 | } | |
cdaead6b JF |
224 | |
225 | xen_build_mfn_list_list(); | |
d451bb7a JF |
226 | } |
227 | ||
228 | unsigned long get_phys_to_machine(unsigned long pfn) | |
229 | { | |
230 | unsigned topidx, idx; | |
231 | ||
8006ec3e JF |
232 | if (unlikely(pfn >= MAX_DOMAIN_PAGES)) |
233 | return INVALID_P2M_ENTRY; | |
234 | ||
d451bb7a | 235 | topidx = p2m_top_index(pfn); |
d451bb7a JF |
236 | idx = p2m_index(pfn); |
237 | return p2m_top[topidx][idx]; | |
238 | } | |
15ce6005 | 239 | EXPORT_SYMBOL_GPL(get_phys_to_machine); |
d451bb7a | 240 | |
e791ca0f JF |
241 | /* install a new p2m_top page */ |
242 | bool install_p2mtop_page(unsigned long pfn, unsigned long *p) | |
d451bb7a | 243 | { |
e791ca0f JF |
244 | unsigned topidx = p2m_top_index(pfn); |
245 | unsigned long **pfnp, *mfnp; | |
d451bb7a JF |
246 | unsigned i; |
247 | ||
e791ca0f JF |
248 | pfnp = &p2m_top[topidx]; |
249 | mfnp = &p2m_top_mfn[topidx]; | |
d451bb7a | 250 | |
f63c2f24 | 251 | for (i = 0; i < P2M_ENTRIES_PER_PAGE; i++) |
d451bb7a JF |
252 | p[i] = INVALID_P2M_ENTRY; |
253 | ||
e791ca0f | 254 | if (cmpxchg(pfnp, p2m_missing, p) == p2m_missing) { |
d5edbc1f | 255 | *mfnp = virt_to_mfn(p); |
e791ca0f JF |
256 | return true; |
257 | } | |
258 | ||
259 | return false; | |
d451bb7a JF |
260 | } |
261 | ||
e791ca0f | 262 | static void alloc_p2m(unsigned long pfn) |
d451bb7a | 263 | { |
e791ca0f | 264 | unsigned long *p; |
d451bb7a | 265 | |
e791ca0f JF |
266 | p = (void *)__get_free_page(GFP_KERNEL | __GFP_NOFAIL); |
267 | BUG_ON(p == NULL); | |
268 | ||
269 | if (!install_p2mtop_page(pfn, p)) | |
270 | free_page((unsigned long)p); | |
271 | } | |
272 | ||
273 | /* Try to install p2m mapping; fail if intermediate bits missing */ | |
274 | bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn) | |
275 | { | |
276 | unsigned topidx, idx; | |
8006ec3e JF |
277 | |
278 | if (unlikely(pfn >= MAX_DOMAIN_PAGES)) { | |
279 | BUG_ON(mfn != INVALID_P2M_ENTRY); | |
e791ca0f | 280 | return true; |
d451bb7a JF |
281 | } |
282 | ||
283 | topidx = p2m_top_index(pfn); | |
cf0923ea | 284 | if (p2m_top[topidx] == p2m_missing) { |
d451bb7a | 285 | if (mfn == INVALID_P2M_ENTRY) |
e791ca0f JF |
286 | return true; |
287 | return false; | |
d451bb7a JF |
288 | } |
289 | ||
290 | idx = p2m_index(pfn); | |
291 | p2m_top[topidx][idx] = mfn; | |
e791ca0f JF |
292 | |
293 | return true; | |
294 | } | |
295 | ||
296 | void set_phys_to_machine(unsigned long pfn, unsigned long mfn) | |
297 | { | |
298 | if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) { | |
299 | BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY); | |
300 | return; | |
301 | } | |
302 | ||
303 | if (unlikely(!__set_phys_to_machine(pfn, mfn))) { | |
304 | alloc_p2m(pfn); | |
305 | ||
306 | if (!__set_phys_to_machine(pfn, mfn)) | |
307 | BUG(); | |
308 | } | |
d451bb7a JF |
309 | } |
310 | ||
9976b39b JF |
311 | unsigned long arbitrary_virt_to_mfn(void *vaddr) |
312 | { | |
313 | xmaddr_t maddr = arbitrary_virt_to_machine(vaddr); | |
314 | ||
315 | return PFN_DOWN(maddr.maddr); | |
316 | } | |
317 | ||
ce803e70 | 318 | xmaddr_t arbitrary_virt_to_machine(void *vaddr) |
3b827c1b | 319 | { |
ce803e70 | 320 | unsigned long address = (unsigned long)vaddr; |
da7bfc50 | 321 | unsigned int level; |
9f32d21c CL |
322 | pte_t *pte; |
323 | unsigned offset; | |
3b827c1b | 324 | |
9f32d21c CL |
325 | /* |
326 | * if the PFN is in the linear mapped vaddr range, we can just use | |
327 | * the (quick) virt_to_machine() p2m lookup | |
328 | */ | |
329 | if (virt_addr_valid(vaddr)) | |
330 | return virt_to_machine(vaddr); | |
331 | ||
332 | /* otherwise we have to do a (slower) full page-table walk */ | |
3b827c1b | 333 | |
9f32d21c CL |
334 | pte = lookup_address(address, &level); |
335 | BUG_ON(pte == NULL); | |
336 | offset = address & ~PAGE_MASK; | |
ebd879e3 | 337 | return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset); |
3b827c1b JF |
338 | } |
339 | ||
340 | void make_lowmem_page_readonly(void *vaddr) | |
341 | { | |
342 | pte_t *pte, ptev; | |
343 | unsigned long address = (unsigned long)vaddr; | |
da7bfc50 | 344 | unsigned int level; |
3b827c1b | 345 | |
f0646e43 | 346 | pte = lookup_address(address, &level); |
3b827c1b JF |
347 | BUG_ON(pte == NULL); |
348 | ||
349 | ptev = pte_wrprotect(*pte); | |
350 | ||
351 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
352 | BUG(); | |
353 | } | |
354 | ||
355 | void make_lowmem_page_readwrite(void *vaddr) | |
356 | { | |
357 | pte_t *pte, ptev; | |
358 | unsigned long address = (unsigned long)vaddr; | |
da7bfc50 | 359 | unsigned int level; |
3b827c1b | 360 | |
f0646e43 | 361 | pte = lookup_address(address, &level); |
3b827c1b JF |
362 | BUG_ON(pte == NULL); |
363 | ||
364 | ptev = pte_mkwrite(*pte); | |
365 | ||
366 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
367 | BUG(); | |
368 | } | |
369 | ||
370 | ||
7708ad64 | 371 | static bool xen_page_pinned(void *ptr) |
e2426cf8 JF |
372 | { |
373 | struct page *page = virt_to_page(ptr); | |
374 | ||
375 | return PagePinned(page); | |
376 | } | |
377 | ||
7708ad64 | 378 | static void xen_extend_mmu_update(const struct mmu_update *update) |
3b827c1b | 379 | { |
d66bf8fc JF |
380 | struct multicall_space mcs; |
381 | struct mmu_update *u; | |
3b827c1b | 382 | |
400d3494 JF |
383 | mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u)); |
384 | ||
994025ca JF |
385 | if (mcs.mc != NULL) { |
386 | ADD_STATS(mmu_update_extended, 1); | |
387 | ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1); | |
388 | ||
400d3494 | 389 | mcs.mc->args[1]++; |
994025ca JF |
390 | |
391 | if (mcs.mc->args[1] < MMU_UPDATE_HISTO) | |
392 | ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1); | |
393 | else | |
394 | ADD_STATS(mmu_update_histo[0], 1); | |
395 | } else { | |
396 | ADD_STATS(mmu_update, 1); | |
400d3494 JF |
397 | mcs = __xen_mc_entry(sizeof(*u)); |
398 | MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); | |
994025ca | 399 | ADD_STATS(mmu_update_histo[1], 1); |
400d3494 | 400 | } |
d66bf8fc | 401 | |
d66bf8fc | 402 | u = mcs.args; |
400d3494 JF |
403 | *u = *update; |
404 | } | |
405 | ||
406 | void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val) | |
407 | { | |
408 | struct mmu_update u; | |
409 | ||
410 | preempt_disable(); | |
411 | ||
412 | xen_mc_batch(); | |
413 | ||
ce803e70 JF |
414 | /* ptr may be ioremapped for 64-bit pagetable setup */ |
415 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
400d3494 | 416 | u.val = pmd_val_ma(val); |
7708ad64 | 417 | xen_extend_mmu_update(&u); |
d66bf8fc | 418 | |
994025ca JF |
419 | ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); |
420 | ||
d66bf8fc JF |
421 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
422 | ||
423 | preempt_enable(); | |
3b827c1b JF |
424 | } |
425 | ||
e2426cf8 JF |
426 | void xen_set_pmd(pmd_t *ptr, pmd_t val) |
427 | { | |
994025ca JF |
428 | ADD_STATS(pmd_update, 1); |
429 | ||
e2426cf8 JF |
430 | /* If page is not pinned, we can just update the entry |
431 | directly */ | |
7708ad64 | 432 | if (!xen_page_pinned(ptr)) { |
e2426cf8 JF |
433 | *ptr = val; |
434 | return; | |
435 | } | |
436 | ||
994025ca JF |
437 | ADD_STATS(pmd_update_pinned, 1); |
438 | ||
e2426cf8 JF |
439 | xen_set_pmd_hyper(ptr, val); |
440 | } | |
441 | ||
3b827c1b JF |
442 | /* |
443 | * Associate a virtual page frame with a given physical page frame | |
444 | * and protection flags for that frame. | |
445 | */ | |
446 | void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags) | |
447 | { | |
836fe2f2 | 448 | set_pte_vaddr(vaddr, mfn_pte(mfn, flags)); |
3b827c1b JF |
449 | } |
450 | ||
451 | void xen_set_pte_at(struct mm_struct *mm, unsigned long addr, | |
452 | pte_t *ptep, pte_t pteval) | |
453 | { | |
2bd50036 JF |
454 | /* updates to init_mm may be done without lock */ |
455 | if (mm == &init_mm) | |
456 | preempt_disable(); | |
457 | ||
994025ca JF |
458 | ADD_STATS(set_pte_at, 1); |
459 | // ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep)); | |
460 | ADD_STATS(set_pte_at_current, mm == current->mm); | |
461 | ADD_STATS(set_pte_at_kernel, mm == &init_mm); | |
462 | ||
d66bf8fc | 463 | if (mm == current->mm || mm == &init_mm) { |
8965c1c0 | 464 | if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) { |
d66bf8fc JF |
465 | struct multicall_space mcs; |
466 | mcs = xen_mc_entry(0); | |
467 | ||
468 | MULTI_update_va_mapping(mcs.mc, addr, pteval, 0); | |
994025ca | 469 | ADD_STATS(set_pte_at_batched, 1); |
d66bf8fc | 470 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
2bd50036 | 471 | goto out; |
d66bf8fc JF |
472 | } else |
473 | if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0) | |
2bd50036 | 474 | goto out; |
d66bf8fc JF |
475 | } |
476 | xen_set_pte(ptep, pteval); | |
2bd50036 JF |
477 | |
478 | out: | |
479 | if (mm == &init_mm) | |
480 | preempt_enable(); | |
3b827c1b JF |
481 | } |
482 | ||
f63c2f24 T |
483 | pte_t xen_ptep_modify_prot_start(struct mm_struct *mm, |
484 | unsigned long addr, pte_t *ptep) | |
947a69c9 | 485 | { |
e57778a1 JF |
486 | /* Just return the pte as-is. We preserve the bits on commit */ |
487 | return *ptep; | |
488 | } | |
489 | ||
490 | void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr, | |
491 | pte_t *ptep, pte_t pte) | |
492 | { | |
400d3494 | 493 | struct mmu_update u; |
e57778a1 | 494 | |
400d3494 | 495 | xen_mc_batch(); |
947a69c9 | 496 | |
9f32d21c | 497 | u.ptr = arbitrary_virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD; |
400d3494 | 498 | u.val = pte_val_ma(pte); |
7708ad64 | 499 | xen_extend_mmu_update(&u); |
947a69c9 | 500 | |
994025ca JF |
501 | ADD_STATS(prot_commit, 1); |
502 | ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); | |
503 | ||
e57778a1 | 504 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
947a69c9 JF |
505 | } |
506 | ||
ebb9cfe2 JF |
507 | /* Assume pteval_t is equivalent to all the other *val_t types. */ |
508 | static pteval_t pte_mfn_to_pfn(pteval_t val) | |
947a69c9 | 509 | { |
ebb9cfe2 | 510 | if (val & _PAGE_PRESENT) { |
59438c9f | 511 | unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; |
77be1fab | 512 | pteval_t flags = val & PTE_FLAGS_MASK; |
d8355aca | 513 | val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags; |
ebb9cfe2 | 514 | } |
947a69c9 | 515 | |
ebb9cfe2 | 516 | return val; |
947a69c9 JF |
517 | } |
518 | ||
ebb9cfe2 | 519 | static pteval_t pte_pfn_to_mfn(pteval_t val) |
947a69c9 | 520 | { |
ebb9cfe2 | 521 | if (val & _PAGE_PRESENT) { |
59438c9f | 522 | unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; |
77be1fab | 523 | pteval_t flags = val & PTE_FLAGS_MASK; |
d8355aca | 524 | val = ((pteval_t)pfn_to_mfn(pfn) << PAGE_SHIFT) | flags; |
947a69c9 JF |
525 | } |
526 | ||
ebb9cfe2 | 527 | return val; |
947a69c9 JF |
528 | } |
529 | ||
ebb9cfe2 | 530 | pteval_t xen_pte_val(pte_t pte) |
947a69c9 | 531 | { |
ebb9cfe2 | 532 | return pte_mfn_to_pfn(pte.pte); |
947a69c9 | 533 | } |
da5de7c2 | 534 | PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val); |
947a69c9 | 535 | |
947a69c9 JF |
536 | pgdval_t xen_pgd_val(pgd_t pgd) |
537 | { | |
ebb9cfe2 | 538 | return pte_mfn_to_pfn(pgd.pgd); |
947a69c9 | 539 | } |
da5de7c2 | 540 | PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val); |
947a69c9 JF |
541 | |
542 | pte_t xen_make_pte(pteval_t pte) | |
543 | { | |
ebb9cfe2 JF |
544 | pte = pte_pfn_to_mfn(pte); |
545 | return native_make_pte(pte); | |
947a69c9 | 546 | } |
da5de7c2 | 547 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte); |
947a69c9 JF |
548 | |
549 | pgd_t xen_make_pgd(pgdval_t pgd) | |
550 | { | |
ebb9cfe2 JF |
551 | pgd = pte_pfn_to_mfn(pgd); |
552 | return native_make_pgd(pgd); | |
947a69c9 | 553 | } |
da5de7c2 | 554 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd); |
947a69c9 JF |
555 | |
556 | pmdval_t xen_pmd_val(pmd_t pmd) | |
557 | { | |
ebb9cfe2 | 558 | return pte_mfn_to_pfn(pmd.pmd); |
947a69c9 | 559 | } |
da5de7c2 | 560 | PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val); |
28499143 | 561 | |
e2426cf8 | 562 | void xen_set_pud_hyper(pud_t *ptr, pud_t val) |
f4f97b3e | 563 | { |
400d3494 | 564 | struct mmu_update u; |
f4f97b3e | 565 | |
d66bf8fc JF |
566 | preempt_disable(); |
567 | ||
400d3494 JF |
568 | xen_mc_batch(); |
569 | ||
ce803e70 JF |
570 | /* ptr may be ioremapped for 64-bit pagetable setup */ |
571 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
400d3494 | 572 | u.val = pud_val_ma(val); |
7708ad64 | 573 | xen_extend_mmu_update(&u); |
d66bf8fc | 574 | |
994025ca JF |
575 | ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); |
576 | ||
d66bf8fc JF |
577 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
578 | ||
579 | preempt_enable(); | |
f4f97b3e JF |
580 | } |
581 | ||
e2426cf8 JF |
582 | void xen_set_pud(pud_t *ptr, pud_t val) |
583 | { | |
994025ca JF |
584 | ADD_STATS(pud_update, 1); |
585 | ||
e2426cf8 JF |
586 | /* If page is not pinned, we can just update the entry |
587 | directly */ | |
7708ad64 | 588 | if (!xen_page_pinned(ptr)) { |
e2426cf8 JF |
589 | *ptr = val; |
590 | return; | |
591 | } | |
592 | ||
994025ca JF |
593 | ADD_STATS(pud_update_pinned, 1); |
594 | ||
e2426cf8 JF |
595 | xen_set_pud_hyper(ptr, val); |
596 | } | |
597 | ||
f4f97b3e JF |
598 | void xen_set_pte(pte_t *ptep, pte_t pte) |
599 | { | |
994025ca JF |
600 | ADD_STATS(pte_update, 1); |
601 | // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep)); | |
602 | ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); | |
603 | ||
f6e58732 | 604 | #ifdef CONFIG_X86_PAE |
f4f97b3e JF |
605 | ptep->pte_high = pte.pte_high; |
606 | smp_wmb(); | |
607 | ptep->pte_low = pte.pte_low; | |
f6e58732 JF |
608 | #else |
609 | *ptep = pte; | |
610 | #endif | |
f4f97b3e JF |
611 | } |
612 | ||
f6e58732 | 613 | #ifdef CONFIG_X86_PAE |
3b827c1b JF |
614 | void xen_set_pte_atomic(pte_t *ptep, pte_t pte) |
615 | { | |
f6e58732 | 616 | set_64bit((u64 *)ptep, native_pte_val(pte)); |
3b827c1b JF |
617 | } |
618 | ||
619 | void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep) | |
620 | { | |
621 | ptep->pte_low = 0; | |
622 | smp_wmb(); /* make sure low gets written first */ | |
623 | ptep->pte_high = 0; | |
624 | } | |
625 | ||
626 | void xen_pmd_clear(pmd_t *pmdp) | |
627 | { | |
e2426cf8 | 628 | set_pmd(pmdp, __pmd(0)); |
3b827c1b | 629 | } |
f6e58732 | 630 | #endif /* CONFIG_X86_PAE */ |
3b827c1b | 631 | |
abf33038 | 632 | pmd_t xen_make_pmd(pmdval_t pmd) |
3b827c1b | 633 | { |
ebb9cfe2 | 634 | pmd = pte_pfn_to_mfn(pmd); |
947a69c9 | 635 | return native_make_pmd(pmd); |
3b827c1b | 636 | } |
da5de7c2 | 637 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd); |
3b827c1b | 638 | |
f6e58732 JF |
639 | #if PAGETABLE_LEVELS == 4 |
640 | pudval_t xen_pud_val(pud_t pud) | |
641 | { | |
642 | return pte_mfn_to_pfn(pud.pud); | |
643 | } | |
da5de7c2 | 644 | PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val); |
f6e58732 JF |
645 | |
646 | pud_t xen_make_pud(pudval_t pud) | |
647 | { | |
648 | pud = pte_pfn_to_mfn(pud); | |
649 | ||
650 | return native_make_pud(pud); | |
651 | } | |
da5de7c2 | 652 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud); |
f6e58732 | 653 | |
d6182fbf | 654 | pgd_t *xen_get_user_pgd(pgd_t *pgd) |
f6e58732 | 655 | { |
d6182fbf JF |
656 | pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK); |
657 | unsigned offset = pgd - pgd_page; | |
658 | pgd_t *user_ptr = NULL; | |
f6e58732 | 659 | |
d6182fbf JF |
660 | if (offset < pgd_index(USER_LIMIT)) { |
661 | struct page *page = virt_to_page(pgd_page); | |
662 | user_ptr = (pgd_t *)page->private; | |
663 | if (user_ptr) | |
664 | user_ptr += offset; | |
665 | } | |
f6e58732 | 666 | |
d6182fbf JF |
667 | return user_ptr; |
668 | } | |
669 | ||
670 | static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val) | |
671 | { | |
672 | struct mmu_update u; | |
f6e58732 JF |
673 | |
674 | u.ptr = virt_to_machine(ptr).maddr; | |
675 | u.val = pgd_val_ma(val); | |
7708ad64 | 676 | xen_extend_mmu_update(&u); |
d6182fbf JF |
677 | } |
678 | ||
679 | /* | |
680 | * Raw hypercall-based set_pgd, intended for in early boot before | |
681 | * there's a page structure. This implies: | |
682 | * 1. The only existing pagetable is the kernel's | |
683 | * 2. It is always pinned | |
684 | * 3. It has no user pagetable attached to it | |
685 | */ | |
686 | void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val) | |
687 | { | |
688 | preempt_disable(); | |
689 | ||
690 | xen_mc_batch(); | |
691 | ||
692 | __xen_set_pgd_hyper(ptr, val); | |
f6e58732 JF |
693 | |
694 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
695 | ||
696 | preempt_enable(); | |
697 | } | |
698 | ||
699 | void xen_set_pgd(pgd_t *ptr, pgd_t val) | |
700 | { | |
d6182fbf JF |
701 | pgd_t *user_ptr = xen_get_user_pgd(ptr); |
702 | ||
994025ca JF |
703 | ADD_STATS(pgd_update, 1); |
704 | ||
f6e58732 JF |
705 | /* If page is not pinned, we can just update the entry |
706 | directly */ | |
7708ad64 | 707 | if (!xen_page_pinned(ptr)) { |
f6e58732 | 708 | *ptr = val; |
d6182fbf | 709 | if (user_ptr) { |
7708ad64 | 710 | WARN_ON(xen_page_pinned(user_ptr)); |
d6182fbf JF |
711 | *user_ptr = val; |
712 | } | |
f6e58732 JF |
713 | return; |
714 | } | |
715 | ||
994025ca JF |
716 | ADD_STATS(pgd_update_pinned, 1); |
717 | ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); | |
718 | ||
d6182fbf JF |
719 | /* If it's pinned, then we can at least batch the kernel and |
720 | user updates together. */ | |
721 | xen_mc_batch(); | |
722 | ||
723 | __xen_set_pgd_hyper(ptr, val); | |
724 | if (user_ptr) | |
725 | __xen_set_pgd_hyper(user_ptr, val); | |
726 | ||
727 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
f6e58732 JF |
728 | } |
729 | #endif /* PAGETABLE_LEVELS == 4 */ | |
730 | ||
f4f97b3e | 731 | /* |
5deb30d1 JF |
732 | * (Yet another) pagetable walker. This one is intended for pinning a |
733 | * pagetable. This means that it walks a pagetable and calls the | |
734 | * callback function on each page it finds making up the page table, | |
735 | * at every level. It walks the entire pagetable, but it only bothers | |
736 | * pinning pte pages which are below limit. In the normal case this | |
737 | * will be STACK_TOP_MAX, but at boot we need to pin up to | |
738 | * FIXADDR_TOP. | |
739 | * | |
740 | * For 32-bit the important bit is that we don't pin beyond there, | |
741 | * because then we start getting into Xen's ptes. | |
742 | * | |
743 | * For 64-bit, we must skip the Xen hole in the middle of the address | |
744 | * space, just after the big x86-64 virtual hole. | |
745 | */ | |
86bbc2c2 IC |
746 | static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd, |
747 | int (*func)(struct mm_struct *mm, struct page *, | |
748 | enum pt_level), | |
749 | unsigned long limit) | |
3b827c1b | 750 | { |
f4f97b3e | 751 | int flush = 0; |
5deb30d1 JF |
752 | unsigned hole_low, hole_high; |
753 | unsigned pgdidx_limit, pudidx_limit, pmdidx_limit; | |
754 | unsigned pgdidx, pudidx, pmdidx; | |
f4f97b3e | 755 | |
5deb30d1 JF |
756 | /* The limit is the last byte to be touched */ |
757 | limit--; | |
758 | BUG_ON(limit >= FIXADDR_TOP); | |
3b827c1b JF |
759 | |
760 | if (xen_feature(XENFEAT_auto_translated_physmap)) | |
f4f97b3e JF |
761 | return 0; |
762 | ||
5deb30d1 JF |
763 | /* |
764 | * 64-bit has a great big hole in the middle of the address | |
765 | * space, which contains the Xen mappings. On 32-bit these | |
766 | * will end up making a zero-sized hole and so is a no-op. | |
767 | */ | |
d6182fbf | 768 | hole_low = pgd_index(USER_LIMIT); |
5deb30d1 JF |
769 | hole_high = pgd_index(PAGE_OFFSET); |
770 | ||
771 | pgdidx_limit = pgd_index(limit); | |
772 | #if PTRS_PER_PUD > 1 | |
773 | pudidx_limit = pud_index(limit); | |
774 | #else | |
775 | pudidx_limit = 0; | |
776 | #endif | |
777 | #if PTRS_PER_PMD > 1 | |
778 | pmdidx_limit = pmd_index(limit); | |
779 | #else | |
780 | pmdidx_limit = 0; | |
781 | #endif | |
782 | ||
5deb30d1 | 783 | for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) { |
f4f97b3e | 784 | pud_t *pud; |
3b827c1b | 785 | |
5deb30d1 JF |
786 | if (pgdidx >= hole_low && pgdidx < hole_high) |
787 | continue; | |
f4f97b3e | 788 | |
5deb30d1 | 789 | if (!pgd_val(pgd[pgdidx])) |
3b827c1b | 790 | continue; |
f4f97b3e | 791 | |
5deb30d1 | 792 | pud = pud_offset(&pgd[pgdidx], 0); |
3b827c1b JF |
793 | |
794 | if (PTRS_PER_PUD > 1) /* not folded */ | |
eefb47f6 | 795 | flush |= (*func)(mm, virt_to_page(pud), PT_PUD); |
f4f97b3e | 796 | |
5deb30d1 | 797 | for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) { |
f4f97b3e | 798 | pmd_t *pmd; |
f4f97b3e | 799 | |
5deb30d1 JF |
800 | if (pgdidx == pgdidx_limit && |
801 | pudidx > pudidx_limit) | |
802 | goto out; | |
3b827c1b | 803 | |
5deb30d1 | 804 | if (pud_none(pud[pudidx])) |
3b827c1b | 805 | continue; |
f4f97b3e | 806 | |
5deb30d1 | 807 | pmd = pmd_offset(&pud[pudidx], 0); |
3b827c1b JF |
808 | |
809 | if (PTRS_PER_PMD > 1) /* not folded */ | |
eefb47f6 | 810 | flush |= (*func)(mm, virt_to_page(pmd), PT_PMD); |
f4f97b3e | 811 | |
5deb30d1 JF |
812 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) { |
813 | struct page *pte; | |
814 | ||
815 | if (pgdidx == pgdidx_limit && | |
816 | pudidx == pudidx_limit && | |
817 | pmdidx > pmdidx_limit) | |
818 | goto out; | |
3b827c1b | 819 | |
5deb30d1 | 820 | if (pmd_none(pmd[pmdidx])) |
3b827c1b JF |
821 | continue; |
822 | ||
5deb30d1 | 823 | pte = pmd_page(pmd[pmdidx]); |
eefb47f6 | 824 | flush |= (*func)(mm, pte, PT_PTE); |
3b827c1b JF |
825 | } |
826 | } | |
827 | } | |
11ad93e5 | 828 | |
5deb30d1 | 829 | out: |
11ad93e5 JF |
830 | /* Do the top level last, so that the callbacks can use it as |
831 | a cue to do final things like tlb flushes. */ | |
eefb47f6 | 832 | flush |= (*func)(mm, virt_to_page(pgd), PT_PGD); |
f4f97b3e JF |
833 | |
834 | return flush; | |
3b827c1b JF |
835 | } |
836 | ||
86bbc2c2 IC |
837 | static int xen_pgd_walk(struct mm_struct *mm, |
838 | int (*func)(struct mm_struct *mm, struct page *, | |
839 | enum pt_level), | |
840 | unsigned long limit) | |
841 | { | |
842 | return __xen_pgd_walk(mm, mm->pgd, func, limit); | |
843 | } | |
844 | ||
7708ad64 JF |
845 | /* If we're using split pte locks, then take the page's lock and |
846 | return a pointer to it. Otherwise return NULL. */ | |
eefb47f6 | 847 | static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm) |
74260714 JF |
848 | { |
849 | spinlock_t *ptl = NULL; | |
850 | ||
f7d0b926 | 851 | #if USE_SPLIT_PTLOCKS |
74260714 | 852 | ptl = __pte_lockptr(page); |
eefb47f6 | 853 | spin_lock_nest_lock(ptl, &mm->page_table_lock); |
74260714 JF |
854 | #endif |
855 | ||
856 | return ptl; | |
857 | } | |
858 | ||
7708ad64 | 859 | static void xen_pte_unlock(void *v) |
74260714 JF |
860 | { |
861 | spinlock_t *ptl = v; | |
862 | spin_unlock(ptl); | |
863 | } | |
864 | ||
865 | static void xen_do_pin(unsigned level, unsigned long pfn) | |
866 | { | |
867 | struct mmuext_op *op; | |
868 | struct multicall_space mcs; | |
869 | ||
870 | mcs = __xen_mc_entry(sizeof(*op)); | |
871 | op = mcs.args; | |
872 | op->cmd = level; | |
873 | op->arg1.mfn = pfn_to_mfn(pfn); | |
874 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
875 | } | |
876 | ||
eefb47f6 JF |
877 | static int xen_pin_page(struct mm_struct *mm, struct page *page, |
878 | enum pt_level level) | |
f4f97b3e | 879 | { |
d60cd46b | 880 | unsigned pgfl = TestSetPagePinned(page); |
f4f97b3e JF |
881 | int flush; |
882 | ||
883 | if (pgfl) | |
884 | flush = 0; /* already pinned */ | |
885 | else if (PageHighMem(page)) | |
886 | /* kmaps need flushing if we found an unpinned | |
887 | highpage */ | |
888 | flush = 1; | |
889 | else { | |
890 | void *pt = lowmem_page_address(page); | |
891 | unsigned long pfn = page_to_pfn(page); | |
892 | struct multicall_space mcs = __xen_mc_entry(0); | |
74260714 | 893 | spinlock_t *ptl; |
f4f97b3e JF |
894 | |
895 | flush = 0; | |
896 | ||
11ad93e5 JF |
897 | /* |
898 | * We need to hold the pagetable lock between the time | |
899 | * we make the pagetable RO and when we actually pin | |
900 | * it. If we don't, then other users may come in and | |
901 | * attempt to update the pagetable by writing it, | |
902 | * which will fail because the memory is RO but not | |
903 | * pinned, so Xen won't do the trap'n'emulate. | |
904 | * | |
905 | * If we're using split pte locks, we can't hold the | |
906 | * entire pagetable's worth of locks during the | |
907 | * traverse, because we may wrap the preempt count (8 | |
908 | * bits). The solution is to mark RO and pin each PTE | |
909 | * page while holding the lock. This means the number | |
910 | * of locks we end up holding is never more than a | |
911 | * batch size (~32 entries, at present). | |
912 | * | |
913 | * If we're not using split pte locks, we needn't pin | |
914 | * the PTE pages independently, because we're | |
915 | * protected by the overall pagetable lock. | |
916 | */ | |
74260714 JF |
917 | ptl = NULL; |
918 | if (level == PT_PTE) | |
eefb47f6 | 919 | ptl = xen_pte_lock(page, mm); |
74260714 | 920 | |
f4f97b3e JF |
921 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, |
922 | pfn_pte(pfn, PAGE_KERNEL_RO), | |
74260714 JF |
923 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
924 | ||
11ad93e5 | 925 | if (ptl) { |
74260714 JF |
926 | xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn); |
927 | ||
74260714 JF |
928 | /* Queue a deferred unlock for when this batch |
929 | is completed. */ | |
7708ad64 | 930 | xen_mc_callback(xen_pte_unlock, ptl); |
74260714 | 931 | } |
f4f97b3e JF |
932 | } |
933 | ||
934 | return flush; | |
935 | } | |
3b827c1b | 936 | |
f4f97b3e JF |
937 | /* This is called just after a mm has been created, but it has not |
938 | been used yet. We need to make sure that its pagetable is all | |
939 | read-only, and can be pinned. */ | |
eefb47f6 | 940 | static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd) |
3b827c1b | 941 | { |
d05fdf31 JF |
942 | vm_unmap_aliases(); |
943 | ||
f4f97b3e | 944 | xen_mc_batch(); |
3b827c1b | 945 | |
86bbc2c2 | 946 | if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) { |
d05fdf31 | 947 | /* re-enable interrupts for flushing */ |
f87e4cac | 948 | xen_mc_issue(0); |
d05fdf31 | 949 | |
f4f97b3e | 950 | kmap_flush_unused(); |
d05fdf31 | 951 | |
f87e4cac JF |
952 | xen_mc_batch(); |
953 | } | |
f4f97b3e | 954 | |
d6182fbf JF |
955 | #ifdef CONFIG_X86_64 |
956 | { | |
957 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
958 | ||
959 | xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd))); | |
960 | ||
961 | if (user_pgd) { | |
eefb47f6 | 962 | xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD); |
f63c2f24 T |
963 | xen_do_pin(MMUEXT_PIN_L4_TABLE, |
964 | PFN_DOWN(__pa(user_pgd))); | |
d6182fbf JF |
965 | } |
966 | } | |
967 | #else /* CONFIG_X86_32 */ | |
5deb30d1 JF |
968 | #ifdef CONFIG_X86_PAE |
969 | /* Need to make sure unshared kernel PMD is pinnable */ | |
47cb2ed9 | 970 | xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), |
eefb47f6 | 971 | PT_PMD); |
5deb30d1 | 972 | #endif |
28499143 | 973 | xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd))); |
d6182fbf | 974 | #endif /* CONFIG_X86_64 */ |
f4f97b3e | 975 | xen_mc_issue(0); |
3b827c1b JF |
976 | } |
977 | ||
eefb47f6 JF |
978 | static void xen_pgd_pin(struct mm_struct *mm) |
979 | { | |
980 | __xen_pgd_pin(mm, mm->pgd); | |
981 | } | |
982 | ||
0e91398f JF |
983 | /* |
984 | * On save, we need to pin all pagetables to make sure they get their | |
985 | * mfns turned into pfns. Search the list for any unpinned pgds and pin | |
986 | * them (unpinned pgds are not currently in use, probably because the | |
987 | * process is under construction or destruction). | |
eefb47f6 JF |
988 | * |
989 | * Expected to be called in stop_machine() ("equivalent to taking | |
990 | * every spinlock in the system"), so the locking doesn't really | |
991 | * matter all that much. | |
0e91398f JF |
992 | */ |
993 | void xen_mm_pin_all(void) | |
994 | { | |
995 | unsigned long flags; | |
996 | struct page *page; | |
74260714 | 997 | |
0e91398f | 998 | spin_lock_irqsave(&pgd_lock, flags); |
f4f97b3e | 999 | |
0e91398f JF |
1000 | list_for_each_entry(page, &pgd_list, lru) { |
1001 | if (!PagePinned(page)) { | |
eefb47f6 | 1002 | __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page)); |
0e91398f JF |
1003 | SetPageSavePinned(page); |
1004 | } | |
1005 | } | |
1006 | ||
1007 | spin_unlock_irqrestore(&pgd_lock, flags); | |
3b827c1b JF |
1008 | } |
1009 | ||
c1f2f09e EH |
1010 | /* |
1011 | * The init_mm pagetable is really pinned as soon as its created, but | |
1012 | * that's before we have page structures to store the bits. So do all | |
1013 | * the book-keeping now. | |
1014 | */ | |
eefb47f6 JF |
1015 | static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page, |
1016 | enum pt_level level) | |
3b827c1b | 1017 | { |
f4f97b3e JF |
1018 | SetPagePinned(page); |
1019 | return 0; | |
1020 | } | |
3b827c1b | 1021 | |
b96229b5 | 1022 | static void __init xen_mark_init_mm_pinned(void) |
f4f97b3e | 1023 | { |
eefb47f6 | 1024 | xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP); |
f4f97b3e | 1025 | } |
3b827c1b | 1026 | |
eefb47f6 JF |
1027 | static int xen_unpin_page(struct mm_struct *mm, struct page *page, |
1028 | enum pt_level level) | |
f4f97b3e | 1029 | { |
d60cd46b | 1030 | unsigned pgfl = TestClearPagePinned(page); |
3b827c1b | 1031 | |
f4f97b3e JF |
1032 | if (pgfl && !PageHighMem(page)) { |
1033 | void *pt = lowmem_page_address(page); | |
1034 | unsigned long pfn = page_to_pfn(page); | |
74260714 JF |
1035 | spinlock_t *ptl = NULL; |
1036 | struct multicall_space mcs; | |
1037 | ||
11ad93e5 JF |
1038 | /* |
1039 | * Do the converse to pin_page. If we're using split | |
1040 | * pte locks, we must be holding the lock for while | |
1041 | * the pte page is unpinned but still RO to prevent | |
1042 | * concurrent updates from seeing it in this | |
1043 | * partially-pinned state. | |
1044 | */ | |
74260714 | 1045 | if (level == PT_PTE) { |
eefb47f6 | 1046 | ptl = xen_pte_lock(page, mm); |
74260714 | 1047 | |
11ad93e5 JF |
1048 | if (ptl) |
1049 | xen_do_pin(MMUEXT_UNPIN_TABLE, pfn); | |
74260714 JF |
1050 | } |
1051 | ||
1052 | mcs = __xen_mc_entry(0); | |
f4f97b3e JF |
1053 | |
1054 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, | |
1055 | pfn_pte(pfn, PAGE_KERNEL), | |
74260714 JF |
1056 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
1057 | ||
1058 | if (ptl) { | |
1059 | /* unlock when batch completed */ | |
7708ad64 | 1060 | xen_mc_callback(xen_pte_unlock, ptl); |
74260714 | 1061 | } |
f4f97b3e JF |
1062 | } |
1063 | ||
1064 | return 0; /* never need to flush on unpin */ | |
3b827c1b JF |
1065 | } |
1066 | ||
f4f97b3e | 1067 | /* Release a pagetables pages back as normal RW */ |
eefb47f6 | 1068 | static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd) |
f4f97b3e | 1069 | { |
f4f97b3e JF |
1070 | xen_mc_batch(); |
1071 | ||
74260714 | 1072 | xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); |
f4f97b3e | 1073 | |
d6182fbf JF |
1074 | #ifdef CONFIG_X86_64 |
1075 | { | |
1076 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
1077 | ||
1078 | if (user_pgd) { | |
f63c2f24 T |
1079 | xen_do_pin(MMUEXT_UNPIN_TABLE, |
1080 | PFN_DOWN(__pa(user_pgd))); | |
eefb47f6 | 1081 | xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD); |
d6182fbf JF |
1082 | } |
1083 | } | |
1084 | #endif | |
1085 | ||
5deb30d1 JF |
1086 | #ifdef CONFIG_X86_PAE |
1087 | /* Need to make sure unshared kernel PMD is unpinned */ | |
47cb2ed9 | 1088 | xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), |
eefb47f6 | 1089 | PT_PMD); |
5deb30d1 | 1090 | #endif |
d6182fbf | 1091 | |
86bbc2c2 | 1092 | __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT); |
f4f97b3e JF |
1093 | |
1094 | xen_mc_issue(0); | |
1095 | } | |
3b827c1b | 1096 | |
eefb47f6 JF |
1097 | static void xen_pgd_unpin(struct mm_struct *mm) |
1098 | { | |
1099 | __xen_pgd_unpin(mm, mm->pgd); | |
1100 | } | |
1101 | ||
0e91398f JF |
1102 | /* |
1103 | * On resume, undo any pinning done at save, so that the rest of the | |
1104 | * kernel doesn't see any unexpected pinned pagetables. | |
1105 | */ | |
1106 | void xen_mm_unpin_all(void) | |
1107 | { | |
1108 | unsigned long flags; | |
1109 | struct page *page; | |
1110 | ||
1111 | spin_lock_irqsave(&pgd_lock, flags); | |
1112 | ||
1113 | list_for_each_entry(page, &pgd_list, lru) { | |
1114 | if (PageSavePinned(page)) { | |
1115 | BUG_ON(!PagePinned(page)); | |
eefb47f6 | 1116 | __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page)); |
0e91398f JF |
1117 | ClearPageSavePinned(page); |
1118 | } | |
1119 | } | |
1120 | ||
1121 | spin_unlock_irqrestore(&pgd_lock, flags); | |
1122 | } | |
1123 | ||
3b827c1b JF |
1124 | void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next) |
1125 | { | |
f4f97b3e | 1126 | spin_lock(&next->page_table_lock); |
eefb47f6 | 1127 | xen_pgd_pin(next); |
f4f97b3e | 1128 | spin_unlock(&next->page_table_lock); |
3b827c1b JF |
1129 | } |
1130 | ||
1131 | void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm) | |
1132 | { | |
f4f97b3e | 1133 | spin_lock(&mm->page_table_lock); |
eefb47f6 | 1134 | xen_pgd_pin(mm); |
f4f97b3e | 1135 | spin_unlock(&mm->page_table_lock); |
3b827c1b JF |
1136 | } |
1137 | ||
3b827c1b | 1138 | |
f87e4cac JF |
1139 | #ifdef CONFIG_SMP |
1140 | /* Another cpu may still have their %cr3 pointing at the pagetable, so | |
1141 | we need to repoint it somewhere else before we can unpin it. */ | |
1142 | static void drop_other_mm_ref(void *info) | |
1143 | { | |
1144 | struct mm_struct *mm = info; | |
ce87b3d3 | 1145 | struct mm_struct *active_mm; |
3b827c1b | 1146 | |
9eb912d1 | 1147 | active_mm = percpu_read(cpu_tlbstate.active_mm); |
ce87b3d3 JF |
1148 | |
1149 | if (active_mm == mm) | |
f87e4cac | 1150 | leave_mm(smp_processor_id()); |
9f79991d JF |
1151 | |
1152 | /* If this cpu still has a stale cr3 reference, then make sure | |
1153 | it has been flushed. */ | |
6dbde353 | 1154 | if (percpu_read(xen_current_cr3) == __pa(mm->pgd)) { |
9f79991d JF |
1155 | load_cr3(swapper_pg_dir); |
1156 | arch_flush_lazy_cpu_mode(); | |
1157 | } | |
f87e4cac | 1158 | } |
3b827c1b | 1159 | |
7708ad64 | 1160 | static void xen_drop_mm_ref(struct mm_struct *mm) |
f87e4cac | 1161 | { |
e4d98207 | 1162 | cpumask_var_t mask; |
9f79991d JF |
1163 | unsigned cpu; |
1164 | ||
f87e4cac JF |
1165 | if (current->active_mm == mm) { |
1166 | if (current->mm == mm) | |
1167 | load_cr3(swapper_pg_dir); | |
1168 | else | |
1169 | leave_mm(smp_processor_id()); | |
9f79991d JF |
1170 | arch_flush_lazy_cpu_mode(); |
1171 | } | |
1172 | ||
1173 | /* Get the "official" set of cpus referring to our pagetable. */ | |
e4d98207 MT |
1174 | if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) { |
1175 | for_each_online_cpu(cpu) { | |
1176 | if (!cpumask_test_cpu(cpu, &mm->cpu_vm_mask) | |
1177 | && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd)) | |
1178 | continue; | |
1179 | smp_call_function_single(cpu, drop_other_mm_ref, mm, 1); | |
1180 | } | |
1181 | return; | |
1182 | } | |
1183 | cpumask_copy(mask, &mm->cpu_vm_mask); | |
9f79991d JF |
1184 | |
1185 | /* It's possible that a vcpu may have a stale reference to our | |
1186 | cr3, because its in lazy mode, and it hasn't yet flushed | |
1187 | its set of pending hypercalls yet. In this case, we can | |
1188 | look at its actual current cr3 value, and force it to flush | |
1189 | if needed. */ | |
1190 | for_each_online_cpu(cpu) { | |
1191 | if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd)) | |
e4d98207 | 1192 | cpumask_set_cpu(cpu, mask); |
3b827c1b JF |
1193 | } |
1194 | ||
e4d98207 MT |
1195 | if (!cpumask_empty(mask)) |
1196 | smp_call_function_many(mask, drop_other_mm_ref, mm, 1); | |
1197 | free_cpumask_var(mask); | |
f87e4cac JF |
1198 | } |
1199 | #else | |
7708ad64 | 1200 | static void xen_drop_mm_ref(struct mm_struct *mm) |
f87e4cac JF |
1201 | { |
1202 | if (current->active_mm == mm) | |
1203 | load_cr3(swapper_pg_dir); | |
1204 | } | |
1205 | #endif | |
1206 | ||
1207 | /* | |
1208 | * While a process runs, Xen pins its pagetables, which means that the | |
1209 | * hypervisor forces it to be read-only, and it controls all updates | |
1210 | * to it. This means that all pagetable updates have to go via the | |
1211 | * hypervisor, which is moderately expensive. | |
1212 | * | |
1213 | * Since we're pulling the pagetable down, we switch to use init_mm, | |
1214 | * unpin old process pagetable and mark it all read-write, which | |
1215 | * allows further operations on it to be simple memory accesses. | |
1216 | * | |
1217 | * The only subtle point is that another CPU may be still using the | |
1218 | * pagetable because of lazy tlb flushing. This means we need need to | |
1219 | * switch all CPUs off this pagetable before we can unpin it. | |
1220 | */ | |
1221 | void xen_exit_mmap(struct mm_struct *mm) | |
1222 | { | |
1223 | get_cpu(); /* make sure we don't move around */ | |
7708ad64 | 1224 | xen_drop_mm_ref(mm); |
f87e4cac | 1225 | put_cpu(); |
3b827c1b | 1226 | |
f120f13e | 1227 | spin_lock(&mm->page_table_lock); |
df912ea4 JF |
1228 | |
1229 | /* pgd may not be pinned in the error exit path of execve */ | |
7708ad64 | 1230 | if (xen_page_pinned(mm->pgd)) |
eefb47f6 | 1231 | xen_pgd_unpin(mm); |
74260714 | 1232 | |
f120f13e | 1233 | spin_unlock(&mm->page_table_lock); |
3b827c1b | 1234 | } |
994025ca | 1235 | |
319f3ba5 JF |
1236 | static __init void xen_pagetable_setup_start(pgd_t *base) |
1237 | { | |
1238 | } | |
1239 | ||
1240 | static __init void xen_pagetable_setup_done(pgd_t *base) | |
1241 | { | |
1242 | xen_setup_shared_info(); | |
1243 | } | |
1244 | ||
1245 | static void xen_write_cr2(unsigned long cr2) | |
1246 | { | |
1247 | percpu_read(xen_vcpu)->arch.cr2 = cr2; | |
1248 | } | |
1249 | ||
1250 | static unsigned long xen_read_cr2(void) | |
1251 | { | |
1252 | return percpu_read(xen_vcpu)->arch.cr2; | |
1253 | } | |
1254 | ||
1255 | unsigned long xen_read_cr2_direct(void) | |
1256 | { | |
1257 | return percpu_read(xen_vcpu_info.arch.cr2); | |
1258 | } | |
1259 | ||
1260 | static void xen_flush_tlb(void) | |
1261 | { | |
1262 | struct mmuext_op *op; | |
1263 | struct multicall_space mcs; | |
1264 | ||
1265 | preempt_disable(); | |
1266 | ||
1267 | mcs = xen_mc_entry(sizeof(*op)); | |
1268 | ||
1269 | op = mcs.args; | |
1270 | op->cmd = MMUEXT_TLB_FLUSH_LOCAL; | |
1271 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1272 | ||
1273 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1274 | ||
1275 | preempt_enable(); | |
1276 | } | |
1277 | ||
1278 | static void xen_flush_tlb_single(unsigned long addr) | |
1279 | { | |
1280 | struct mmuext_op *op; | |
1281 | struct multicall_space mcs; | |
1282 | ||
1283 | preempt_disable(); | |
1284 | ||
1285 | mcs = xen_mc_entry(sizeof(*op)); | |
1286 | op = mcs.args; | |
1287 | op->cmd = MMUEXT_INVLPG_LOCAL; | |
1288 | op->arg1.linear_addr = addr & PAGE_MASK; | |
1289 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1290 | ||
1291 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1292 | ||
1293 | preempt_enable(); | |
1294 | } | |
1295 | ||
1296 | static void xen_flush_tlb_others(const struct cpumask *cpus, | |
1297 | struct mm_struct *mm, unsigned long va) | |
1298 | { | |
1299 | struct { | |
1300 | struct mmuext_op op; | |
1301 | DECLARE_BITMAP(mask, NR_CPUS); | |
1302 | } *args; | |
1303 | struct multicall_space mcs; | |
1304 | ||
e3f8a74e JF |
1305 | if (cpumask_empty(cpus)) |
1306 | return; /* nothing to do */ | |
319f3ba5 JF |
1307 | |
1308 | mcs = xen_mc_entry(sizeof(*args)); | |
1309 | args = mcs.args; | |
1310 | args->op.arg2.vcpumask = to_cpumask(args->mask); | |
1311 | ||
1312 | /* Remove us, and any offline CPUS. */ | |
1313 | cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask); | |
1314 | cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask)); | |
319f3ba5 JF |
1315 | |
1316 | if (va == TLB_FLUSH_ALL) { | |
1317 | args->op.cmd = MMUEXT_TLB_FLUSH_MULTI; | |
1318 | } else { | |
1319 | args->op.cmd = MMUEXT_INVLPG_MULTI; | |
1320 | args->op.arg1.linear_addr = va; | |
1321 | } | |
1322 | ||
1323 | MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF); | |
1324 | ||
319f3ba5 JF |
1325 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
1326 | } | |
1327 | ||
1328 | static unsigned long xen_read_cr3(void) | |
1329 | { | |
1330 | return percpu_read(xen_cr3); | |
1331 | } | |
1332 | ||
1333 | static void set_current_cr3(void *v) | |
1334 | { | |
1335 | percpu_write(xen_current_cr3, (unsigned long)v); | |
1336 | } | |
1337 | ||
1338 | static void __xen_write_cr3(bool kernel, unsigned long cr3) | |
1339 | { | |
1340 | struct mmuext_op *op; | |
1341 | struct multicall_space mcs; | |
1342 | unsigned long mfn; | |
1343 | ||
1344 | if (cr3) | |
1345 | mfn = pfn_to_mfn(PFN_DOWN(cr3)); | |
1346 | else | |
1347 | mfn = 0; | |
1348 | ||
1349 | WARN_ON(mfn == 0 && kernel); | |
1350 | ||
1351 | mcs = __xen_mc_entry(sizeof(*op)); | |
1352 | ||
1353 | op = mcs.args; | |
1354 | op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR; | |
1355 | op->arg1.mfn = mfn; | |
1356 | ||
1357 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1358 | ||
1359 | if (kernel) { | |
1360 | percpu_write(xen_cr3, cr3); | |
1361 | ||
1362 | /* Update xen_current_cr3 once the batch has actually | |
1363 | been submitted. */ | |
1364 | xen_mc_callback(set_current_cr3, (void *)cr3); | |
1365 | } | |
1366 | } | |
1367 | ||
1368 | static void xen_write_cr3(unsigned long cr3) | |
1369 | { | |
1370 | BUG_ON(preemptible()); | |
1371 | ||
1372 | xen_mc_batch(); /* disables interrupts */ | |
1373 | ||
1374 | /* Update while interrupts are disabled, so its atomic with | |
1375 | respect to ipis */ | |
1376 | percpu_write(xen_cr3, cr3); | |
1377 | ||
1378 | __xen_write_cr3(true, cr3); | |
1379 | ||
1380 | #ifdef CONFIG_X86_64 | |
1381 | { | |
1382 | pgd_t *user_pgd = xen_get_user_pgd(__va(cr3)); | |
1383 | if (user_pgd) | |
1384 | __xen_write_cr3(false, __pa(user_pgd)); | |
1385 | else | |
1386 | __xen_write_cr3(false, 0); | |
1387 | } | |
1388 | #endif | |
1389 | ||
1390 | xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ | |
1391 | } | |
1392 | ||
1393 | static int xen_pgd_alloc(struct mm_struct *mm) | |
1394 | { | |
1395 | pgd_t *pgd = mm->pgd; | |
1396 | int ret = 0; | |
1397 | ||
1398 | BUG_ON(PagePinned(virt_to_page(pgd))); | |
1399 | ||
1400 | #ifdef CONFIG_X86_64 | |
1401 | { | |
1402 | struct page *page = virt_to_page(pgd); | |
1403 | pgd_t *user_pgd; | |
1404 | ||
1405 | BUG_ON(page->private != 0); | |
1406 | ||
1407 | ret = -ENOMEM; | |
1408 | ||
1409 | user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO); | |
1410 | page->private = (unsigned long)user_pgd; | |
1411 | ||
1412 | if (user_pgd != NULL) { | |
1413 | user_pgd[pgd_index(VSYSCALL_START)] = | |
1414 | __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE); | |
1415 | ret = 0; | |
1416 | } | |
1417 | ||
1418 | BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd)))); | |
1419 | } | |
1420 | #endif | |
1421 | ||
1422 | return ret; | |
1423 | } | |
1424 | ||
1425 | static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd) | |
1426 | { | |
1427 | #ifdef CONFIG_X86_64 | |
1428 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
1429 | ||
1430 | if (user_pgd) | |
1431 | free_page((unsigned long)user_pgd); | |
1432 | #endif | |
1433 | } | |
1434 | ||
1f4f9315 JF |
1435 | #ifdef CONFIG_HIGHPTE |
1436 | static void *xen_kmap_atomic_pte(struct page *page, enum km_type type) | |
1437 | { | |
1438 | pgprot_t prot = PAGE_KERNEL; | |
1439 | ||
1440 | if (PagePinned(page)) | |
1441 | prot = PAGE_KERNEL_RO; | |
1442 | ||
1443 | if (0 && PageHighMem(page)) | |
1444 | printk("mapping highpte %lx type %d prot %s\n", | |
1445 | page_to_pfn(page), type, | |
1446 | (unsigned long)pgprot_val(prot) & _PAGE_RW ? "WRITE" : "READ"); | |
1447 | ||
1448 | return kmap_atomic_prot(page, type, prot); | |
1449 | } | |
1450 | #endif | |
1451 | ||
1452 | #ifdef CONFIG_X86_32 | |
1453 | static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte) | |
1454 | { | |
1455 | /* If there's an existing pte, then don't allow _PAGE_RW to be set */ | |
1456 | if (pte_val_ma(*ptep) & _PAGE_PRESENT) | |
1457 | pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) & | |
1458 | pte_val_ma(pte)); | |
1459 | ||
1460 | return pte; | |
1461 | } | |
1462 | ||
1463 | /* Init-time set_pte while constructing initial pagetables, which | |
1464 | doesn't allow RO pagetable pages to be remapped RW */ | |
1465 | static __init void xen_set_pte_init(pte_t *ptep, pte_t pte) | |
1466 | { | |
1467 | pte = mask_rw_pte(ptep, pte); | |
1468 | ||
1469 | xen_set_pte(ptep, pte); | |
1470 | } | |
1471 | #endif | |
319f3ba5 | 1472 | |
b96229b5 JF |
1473 | static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn) |
1474 | { | |
1475 | struct mmuext_op op; | |
1476 | op.cmd = cmd; | |
1477 | op.arg1.mfn = pfn_to_mfn(pfn); | |
1478 | if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) | |
1479 | BUG(); | |
1480 | } | |
1481 | ||
319f3ba5 JF |
1482 | /* Early in boot, while setting up the initial pagetable, assume |
1483 | everything is pinned. */ | |
1484 | static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn) | |
1485 | { | |
b96229b5 JF |
1486 | #ifdef CONFIG_FLATMEM |
1487 | BUG_ON(mem_map); /* should only be used early */ | |
1488 | #endif | |
1489 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1490 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
1491 | } | |
1492 | ||
1493 | /* Used for pmd and pud */ | |
1494 | static __init void xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn) | |
1495 | { | |
319f3ba5 JF |
1496 | #ifdef CONFIG_FLATMEM |
1497 | BUG_ON(mem_map); /* should only be used early */ | |
1498 | #endif | |
1499 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1500 | } | |
1501 | ||
1502 | /* Early release_pte assumes that all pts are pinned, since there's | |
1503 | only init_mm and anything attached to that is pinned. */ | |
b96229b5 | 1504 | static __init void xen_release_pte_init(unsigned long pfn) |
319f3ba5 | 1505 | { |
b96229b5 | 1506 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); |
319f3ba5 JF |
1507 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); |
1508 | } | |
1509 | ||
b96229b5 | 1510 | static __init void xen_release_pmd_init(unsigned long pfn) |
319f3ba5 | 1511 | { |
b96229b5 | 1512 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); |
319f3ba5 JF |
1513 | } |
1514 | ||
1515 | /* This needs to make sure the new pte page is pinned iff its being | |
1516 | attached to a pinned pagetable. */ | |
1517 | static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level) | |
1518 | { | |
1519 | struct page *page = pfn_to_page(pfn); | |
1520 | ||
1521 | if (PagePinned(virt_to_page(mm->pgd))) { | |
1522 | SetPagePinned(page); | |
1523 | ||
1524 | vm_unmap_aliases(); | |
1525 | if (!PageHighMem(page)) { | |
1526 | make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn))); | |
1527 | if (level == PT_PTE && USE_SPLIT_PTLOCKS) | |
1528 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
1529 | } else { | |
1530 | /* make sure there are no stray mappings of | |
1531 | this page */ | |
1532 | kmap_flush_unused(); | |
1533 | } | |
1534 | } | |
1535 | } | |
1536 | ||
1537 | static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn) | |
1538 | { | |
1539 | xen_alloc_ptpage(mm, pfn, PT_PTE); | |
1540 | } | |
1541 | ||
1542 | static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn) | |
1543 | { | |
1544 | xen_alloc_ptpage(mm, pfn, PT_PMD); | |
1545 | } | |
1546 | ||
1547 | /* This should never happen until we're OK to use struct page */ | |
1548 | static void xen_release_ptpage(unsigned long pfn, unsigned level) | |
1549 | { | |
1550 | struct page *page = pfn_to_page(pfn); | |
1551 | ||
1552 | if (PagePinned(page)) { | |
1553 | if (!PageHighMem(page)) { | |
1554 | if (level == PT_PTE && USE_SPLIT_PTLOCKS) | |
1555 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); | |
1556 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
1557 | } | |
1558 | ClearPagePinned(page); | |
1559 | } | |
1560 | } | |
1561 | ||
1562 | static void xen_release_pte(unsigned long pfn) | |
1563 | { | |
1564 | xen_release_ptpage(pfn, PT_PTE); | |
1565 | } | |
1566 | ||
1567 | static void xen_release_pmd(unsigned long pfn) | |
1568 | { | |
1569 | xen_release_ptpage(pfn, PT_PMD); | |
1570 | } | |
1571 | ||
1572 | #if PAGETABLE_LEVELS == 4 | |
1573 | static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn) | |
1574 | { | |
1575 | xen_alloc_ptpage(mm, pfn, PT_PUD); | |
1576 | } | |
1577 | ||
1578 | static void xen_release_pud(unsigned long pfn) | |
1579 | { | |
1580 | xen_release_ptpage(pfn, PT_PUD); | |
1581 | } | |
1582 | #endif | |
1583 | ||
1584 | void __init xen_reserve_top(void) | |
1585 | { | |
1586 | #ifdef CONFIG_X86_32 | |
1587 | unsigned long top = HYPERVISOR_VIRT_START; | |
1588 | struct xen_platform_parameters pp; | |
1589 | ||
1590 | if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0) | |
1591 | top = pp.virt_start; | |
1592 | ||
1593 | reserve_top_address(-top); | |
1594 | #endif /* CONFIG_X86_32 */ | |
1595 | } | |
1596 | ||
1597 | /* | |
1598 | * Like __va(), but returns address in the kernel mapping (which is | |
1599 | * all we have until the physical memory mapping has been set up. | |
1600 | */ | |
1601 | static void *__ka(phys_addr_t paddr) | |
1602 | { | |
1603 | #ifdef CONFIG_X86_64 | |
1604 | return (void *)(paddr + __START_KERNEL_map); | |
1605 | #else | |
1606 | return __va(paddr); | |
1607 | #endif | |
1608 | } | |
1609 | ||
1610 | /* Convert a machine address to physical address */ | |
1611 | static unsigned long m2p(phys_addr_t maddr) | |
1612 | { | |
1613 | phys_addr_t paddr; | |
1614 | ||
1615 | maddr &= PTE_PFN_MASK; | |
1616 | paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT; | |
1617 | ||
1618 | return paddr; | |
1619 | } | |
1620 | ||
1621 | /* Convert a machine address to kernel virtual */ | |
1622 | static void *m2v(phys_addr_t maddr) | |
1623 | { | |
1624 | return __ka(m2p(maddr)); | |
1625 | } | |
1626 | ||
1627 | static void set_page_prot(void *addr, pgprot_t prot) | |
1628 | { | |
1629 | unsigned long pfn = __pa(addr) >> PAGE_SHIFT; | |
1630 | pte_t pte = pfn_pte(pfn, prot); | |
1631 | ||
1632 | if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0)) | |
1633 | BUG(); | |
1634 | } | |
1635 | ||
1636 | static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn) | |
1637 | { | |
1638 | unsigned pmdidx, pteidx; | |
1639 | unsigned ident_pte; | |
1640 | unsigned long pfn; | |
1641 | ||
1642 | ident_pte = 0; | |
1643 | pfn = 0; | |
1644 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) { | |
1645 | pte_t *pte_page; | |
1646 | ||
1647 | /* Reuse or allocate a page of ptes */ | |
1648 | if (pmd_present(pmd[pmdidx])) | |
1649 | pte_page = m2v(pmd[pmdidx].pmd); | |
1650 | else { | |
1651 | /* Check for free pte pages */ | |
1652 | if (ident_pte == ARRAY_SIZE(level1_ident_pgt)) | |
1653 | break; | |
1654 | ||
1655 | pte_page = &level1_ident_pgt[ident_pte]; | |
1656 | ident_pte += PTRS_PER_PTE; | |
1657 | ||
1658 | pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE); | |
1659 | } | |
1660 | ||
1661 | /* Install mappings */ | |
1662 | for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) { | |
1663 | pte_t pte; | |
1664 | ||
1665 | if (pfn > max_pfn_mapped) | |
1666 | max_pfn_mapped = pfn; | |
1667 | ||
1668 | if (!pte_none(pte_page[pteidx])) | |
1669 | continue; | |
1670 | ||
1671 | pte = pfn_pte(pfn, PAGE_KERNEL_EXEC); | |
1672 | pte_page[pteidx] = pte; | |
1673 | } | |
1674 | } | |
1675 | ||
1676 | for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE) | |
1677 | set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO); | |
1678 | ||
1679 | set_page_prot(pmd, PAGE_KERNEL_RO); | |
1680 | } | |
1681 | ||
1682 | #ifdef CONFIG_X86_64 | |
1683 | static void convert_pfn_mfn(void *v) | |
1684 | { | |
1685 | pte_t *pte = v; | |
1686 | int i; | |
1687 | ||
1688 | /* All levels are converted the same way, so just treat them | |
1689 | as ptes. */ | |
1690 | for (i = 0; i < PTRS_PER_PTE; i++) | |
1691 | pte[i] = xen_make_pte(pte[i].pte); | |
1692 | } | |
1693 | ||
1694 | /* | |
1695 | * Set up the inital kernel pagetable. | |
1696 | * | |
1697 | * We can construct this by grafting the Xen provided pagetable into | |
1698 | * head_64.S's preconstructed pagetables. We copy the Xen L2's into | |
1699 | * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This | |
1700 | * means that only the kernel has a physical mapping to start with - | |
1701 | * but that's enough to get __va working. We need to fill in the rest | |
1702 | * of the physical mapping once some sort of allocator has been set | |
1703 | * up. | |
1704 | */ | |
1705 | __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, | |
1706 | unsigned long max_pfn) | |
1707 | { | |
1708 | pud_t *l3; | |
1709 | pmd_t *l2; | |
1710 | ||
1711 | /* Zap identity mapping */ | |
1712 | init_level4_pgt[0] = __pgd(0); | |
1713 | ||
1714 | /* Pre-constructed entries are in pfn, so convert to mfn */ | |
1715 | convert_pfn_mfn(init_level4_pgt); | |
1716 | convert_pfn_mfn(level3_ident_pgt); | |
1717 | convert_pfn_mfn(level3_kernel_pgt); | |
1718 | ||
1719 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd); | |
1720 | l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud); | |
1721 | ||
1722 | memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1723 | memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1724 | ||
1725 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd); | |
1726 | l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud); | |
1727 | memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1728 | ||
1729 | /* Set up identity map */ | |
1730 | xen_map_identity_early(level2_ident_pgt, max_pfn); | |
1731 | ||
1732 | /* Make pagetable pieces RO */ | |
1733 | set_page_prot(init_level4_pgt, PAGE_KERNEL_RO); | |
1734 | set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO); | |
1735 | set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO); | |
1736 | set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO); | |
1737 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); | |
1738 | set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO); | |
1739 | ||
1740 | /* Pin down new L4 */ | |
1741 | pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE, | |
1742 | PFN_DOWN(__pa_symbol(init_level4_pgt))); | |
1743 | ||
1744 | /* Unpin Xen-provided one */ | |
1745 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
1746 | ||
1747 | /* Switch over */ | |
1748 | pgd = init_level4_pgt; | |
1749 | ||
1750 | /* | |
1751 | * At this stage there can be no user pgd, and no page | |
1752 | * structure to attach it to, so make sure we just set kernel | |
1753 | * pgd. | |
1754 | */ | |
1755 | xen_mc_batch(); | |
1756 | __xen_write_cr3(true, __pa(pgd)); | |
1757 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
1758 | ||
1759 | reserve_early(__pa(xen_start_info->pt_base), | |
1760 | __pa(xen_start_info->pt_base + | |
1761 | xen_start_info->nr_pt_frames * PAGE_SIZE), | |
1762 | "XEN PAGETABLES"); | |
1763 | ||
1764 | return pgd; | |
1765 | } | |
1766 | #else /* !CONFIG_X86_64 */ | |
1767 | static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss; | |
1768 | ||
1769 | __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, | |
1770 | unsigned long max_pfn) | |
1771 | { | |
1772 | pmd_t *kernel_pmd; | |
1773 | ||
93dbda7c JF |
1774 | max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) + |
1775 | xen_start_info->nr_pt_frames * PAGE_SIZE + | |
1776 | 512*1024); | |
319f3ba5 JF |
1777 | |
1778 | kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd); | |
1779 | memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD); | |
1780 | ||
1781 | xen_map_identity_early(level2_kernel_pgt, max_pfn); | |
1782 | ||
1783 | memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD); | |
1784 | set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY], | |
1785 | __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT)); | |
1786 | ||
1787 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); | |
1788 | set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO); | |
1789 | set_page_prot(empty_zero_page, PAGE_KERNEL_RO); | |
1790 | ||
1791 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
1792 | ||
1793 | xen_write_cr3(__pa(swapper_pg_dir)); | |
1794 | ||
1795 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir))); | |
1796 | ||
1797 | return swapper_pg_dir; | |
1798 | } | |
1799 | #endif /* CONFIG_X86_64 */ | |
1800 | ||
3b3809ac | 1801 | static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot) |
319f3ba5 JF |
1802 | { |
1803 | pte_t pte; | |
1804 | ||
1805 | phys >>= PAGE_SHIFT; | |
1806 | ||
1807 | switch (idx) { | |
1808 | case FIX_BTMAP_END ... FIX_BTMAP_BEGIN: | |
1809 | #ifdef CONFIG_X86_F00F_BUG | |
1810 | case FIX_F00F_IDT: | |
1811 | #endif | |
1812 | #ifdef CONFIG_X86_32 | |
1813 | case FIX_WP_TEST: | |
1814 | case FIX_VDSO: | |
1815 | # ifdef CONFIG_HIGHMEM | |
1816 | case FIX_KMAP_BEGIN ... FIX_KMAP_END: | |
1817 | # endif | |
1818 | #else | |
1819 | case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE: | |
1820 | #endif | |
1821 | #ifdef CONFIG_X86_LOCAL_APIC | |
1822 | case FIX_APIC_BASE: /* maps dummy local APIC */ | |
1823 | #endif | |
3ecb1b7d JF |
1824 | case FIX_TEXT_POKE0: |
1825 | case FIX_TEXT_POKE1: | |
1826 | /* All local page mappings */ | |
319f3ba5 JF |
1827 | pte = pfn_pte(phys, prot); |
1828 | break; | |
1829 | ||
1830 | default: | |
1831 | pte = mfn_pte(phys, prot); | |
1832 | break; | |
1833 | } | |
1834 | ||
1835 | __native_set_fixmap(idx, pte); | |
1836 | ||
1837 | #ifdef CONFIG_X86_64 | |
1838 | /* Replicate changes to map the vsyscall page into the user | |
1839 | pagetable vsyscall mapping. */ | |
1840 | if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) { | |
1841 | unsigned long vaddr = __fix_to_virt(idx); | |
1842 | set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte); | |
1843 | } | |
1844 | #endif | |
1845 | } | |
1846 | ||
1847 | __init void xen_post_allocator_init(void) | |
1848 | { | |
1849 | pv_mmu_ops.set_pte = xen_set_pte; | |
1850 | pv_mmu_ops.set_pmd = xen_set_pmd; | |
1851 | pv_mmu_ops.set_pud = xen_set_pud; | |
1852 | #if PAGETABLE_LEVELS == 4 | |
1853 | pv_mmu_ops.set_pgd = xen_set_pgd; | |
1854 | #endif | |
1855 | ||
1856 | /* This will work as long as patching hasn't happened yet | |
1857 | (which it hasn't) */ | |
1858 | pv_mmu_ops.alloc_pte = xen_alloc_pte; | |
1859 | pv_mmu_ops.alloc_pmd = xen_alloc_pmd; | |
1860 | pv_mmu_ops.release_pte = xen_release_pte; | |
1861 | pv_mmu_ops.release_pmd = xen_release_pmd; | |
1862 | #if PAGETABLE_LEVELS == 4 | |
1863 | pv_mmu_ops.alloc_pud = xen_alloc_pud; | |
1864 | pv_mmu_ops.release_pud = xen_release_pud; | |
1865 | #endif | |
1866 | ||
1867 | #ifdef CONFIG_X86_64 | |
1868 | SetPagePinned(virt_to_page(level3_user_vsyscall)); | |
1869 | #endif | |
1870 | xen_mark_init_mm_pinned(); | |
1871 | } | |
1872 | ||
319f3ba5 JF |
1873 | const struct pv_mmu_ops xen_mmu_ops __initdata = { |
1874 | .pagetable_setup_start = xen_pagetable_setup_start, | |
1875 | .pagetable_setup_done = xen_pagetable_setup_done, | |
1876 | ||
1877 | .read_cr2 = xen_read_cr2, | |
1878 | .write_cr2 = xen_write_cr2, | |
1879 | ||
1880 | .read_cr3 = xen_read_cr3, | |
1881 | .write_cr3 = xen_write_cr3, | |
1882 | ||
1883 | .flush_tlb_user = xen_flush_tlb, | |
1884 | .flush_tlb_kernel = xen_flush_tlb, | |
1885 | .flush_tlb_single = xen_flush_tlb_single, | |
1886 | .flush_tlb_others = xen_flush_tlb_others, | |
1887 | ||
1888 | .pte_update = paravirt_nop, | |
1889 | .pte_update_defer = paravirt_nop, | |
1890 | ||
1891 | .pgd_alloc = xen_pgd_alloc, | |
1892 | .pgd_free = xen_pgd_free, | |
1893 | ||
1894 | .alloc_pte = xen_alloc_pte_init, | |
1895 | .release_pte = xen_release_pte_init, | |
b96229b5 | 1896 | .alloc_pmd = xen_alloc_pmd_init, |
319f3ba5 | 1897 | .alloc_pmd_clone = paravirt_nop, |
b96229b5 | 1898 | .release_pmd = xen_release_pmd_init, |
319f3ba5 JF |
1899 | |
1900 | #ifdef CONFIG_HIGHPTE | |
1901 | .kmap_atomic_pte = xen_kmap_atomic_pte, | |
1902 | #endif | |
1903 | ||
1904 | #ifdef CONFIG_X86_64 | |
1905 | .set_pte = xen_set_pte, | |
1906 | #else | |
1907 | .set_pte = xen_set_pte_init, | |
1908 | #endif | |
1909 | .set_pte_at = xen_set_pte_at, | |
1910 | .set_pmd = xen_set_pmd_hyper, | |
1911 | ||
1912 | .ptep_modify_prot_start = __ptep_modify_prot_start, | |
1913 | .ptep_modify_prot_commit = __ptep_modify_prot_commit, | |
1914 | ||
da5de7c2 JF |
1915 | .pte_val = PV_CALLEE_SAVE(xen_pte_val), |
1916 | .pgd_val = PV_CALLEE_SAVE(xen_pgd_val), | |
319f3ba5 | 1917 | |
da5de7c2 JF |
1918 | .make_pte = PV_CALLEE_SAVE(xen_make_pte), |
1919 | .make_pgd = PV_CALLEE_SAVE(xen_make_pgd), | |
319f3ba5 JF |
1920 | |
1921 | #ifdef CONFIG_X86_PAE | |
1922 | .set_pte_atomic = xen_set_pte_atomic, | |
319f3ba5 JF |
1923 | .pte_clear = xen_pte_clear, |
1924 | .pmd_clear = xen_pmd_clear, | |
1925 | #endif /* CONFIG_X86_PAE */ | |
1926 | .set_pud = xen_set_pud_hyper, | |
1927 | ||
da5de7c2 JF |
1928 | .make_pmd = PV_CALLEE_SAVE(xen_make_pmd), |
1929 | .pmd_val = PV_CALLEE_SAVE(xen_pmd_val), | |
319f3ba5 JF |
1930 | |
1931 | #if PAGETABLE_LEVELS == 4 | |
da5de7c2 JF |
1932 | .pud_val = PV_CALLEE_SAVE(xen_pud_val), |
1933 | .make_pud = PV_CALLEE_SAVE(xen_make_pud), | |
319f3ba5 JF |
1934 | .set_pgd = xen_set_pgd_hyper, |
1935 | ||
b96229b5 JF |
1936 | .alloc_pud = xen_alloc_pmd_init, |
1937 | .release_pud = xen_release_pmd_init, | |
319f3ba5 JF |
1938 | #endif /* PAGETABLE_LEVELS == 4 */ |
1939 | ||
1940 | .activate_mm = xen_activate_mm, | |
1941 | .dup_mmap = xen_dup_mmap, | |
1942 | .exit_mmap = xen_exit_mmap, | |
1943 | ||
1944 | .lazy_mode = { | |
1945 | .enter = paravirt_enter_lazy_mmu, | |
1946 | .leave = xen_leave_lazy, | |
1947 | }, | |
1948 | ||
1949 | .set_fixmap = xen_set_fixmap, | |
1950 | }; | |
1951 | ||
1952 | ||
994025ca JF |
1953 | #ifdef CONFIG_XEN_DEBUG_FS |
1954 | ||
1955 | static struct dentry *d_mmu_debug; | |
1956 | ||
1957 | static int __init xen_mmu_debugfs(void) | |
1958 | { | |
1959 | struct dentry *d_xen = xen_init_debugfs(); | |
1960 | ||
1961 | if (d_xen == NULL) | |
1962 | return -ENOMEM; | |
1963 | ||
1964 | d_mmu_debug = debugfs_create_dir("mmu", d_xen); | |
1965 | ||
1966 | debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats); | |
1967 | ||
1968 | debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update); | |
1969 | debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug, | |
1970 | &mmu_stats.pgd_update_pinned); | |
1971 | debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug, | |
1972 | &mmu_stats.pgd_update_pinned); | |
1973 | ||
1974 | debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update); | |
1975 | debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug, | |
1976 | &mmu_stats.pud_update_pinned); | |
1977 | debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug, | |
1978 | &mmu_stats.pud_update_pinned); | |
1979 | ||
1980 | debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update); | |
1981 | debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug, | |
1982 | &mmu_stats.pmd_update_pinned); | |
1983 | debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug, | |
1984 | &mmu_stats.pmd_update_pinned); | |
1985 | ||
1986 | debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update); | |
1987 | // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug, | |
1988 | // &mmu_stats.pte_update_pinned); | |
1989 | debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug, | |
1990 | &mmu_stats.pte_update_pinned); | |
1991 | ||
1992 | debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update); | |
1993 | debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug, | |
1994 | &mmu_stats.mmu_update_extended); | |
1995 | xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug, | |
1996 | mmu_stats.mmu_update_histo, 20); | |
1997 | ||
1998 | debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at); | |
1999 | debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug, | |
2000 | &mmu_stats.set_pte_at_batched); | |
2001 | debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug, | |
2002 | &mmu_stats.set_pte_at_current); | |
2003 | debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug, | |
2004 | &mmu_stats.set_pte_at_kernel); | |
2005 | ||
2006 | debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit); | |
2007 | debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug, | |
2008 | &mmu_stats.prot_commit_batched); | |
2009 | ||
2010 | return 0; | |
2011 | } | |
2012 | fs_initcall(xen_mmu_debugfs); | |
2013 | ||
2014 | #endif /* CONFIG_XEN_DEBUG_FS */ |