]>
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> |
5a0e3ad6 | 46 | #include <linux/gfp.h> |
3b827c1b JF |
47 | |
48 | #include <asm/pgtable.h> | |
49 | #include <asm/tlbflush.h> | |
5deb30d1 | 50 | #include <asm/fixmap.h> |
3b827c1b | 51 | #include <asm/mmu_context.h> |
319f3ba5 | 52 | #include <asm/setup.h> |
f4f97b3e | 53 | #include <asm/paravirt.h> |
cbcd79c2 | 54 | #include <asm/linkage.h> |
3b827c1b JF |
55 | |
56 | #include <asm/xen/hypercall.h> | |
f4f97b3e | 57 | #include <asm/xen/hypervisor.h> |
3b827c1b JF |
58 | |
59 | #include <xen/page.h> | |
60 | #include <xen/interface/xen.h> | |
59151001 | 61 | #include <xen/interface/hvm/hvm_op.h> |
319f3ba5 JF |
62 | #include <xen/interface/version.h> |
63 | #include <xen/hvc-console.h> | |
3b827c1b | 64 | |
f4f97b3e | 65 | #include "multicalls.h" |
3b827c1b | 66 | #include "mmu.h" |
994025ca JF |
67 | #include "debugfs.h" |
68 | ||
69 | #define MMU_UPDATE_HISTO 30 | |
70 | ||
71 | #ifdef CONFIG_XEN_DEBUG_FS | |
72 | ||
73 | static struct { | |
74 | u32 pgd_update; | |
75 | u32 pgd_update_pinned; | |
76 | u32 pgd_update_batched; | |
77 | ||
78 | u32 pud_update; | |
79 | u32 pud_update_pinned; | |
80 | u32 pud_update_batched; | |
81 | ||
82 | u32 pmd_update; | |
83 | u32 pmd_update_pinned; | |
84 | u32 pmd_update_batched; | |
85 | ||
86 | u32 pte_update; | |
87 | u32 pte_update_pinned; | |
88 | u32 pte_update_batched; | |
89 | ||
90 | u32 mmu_update; | |
91 | u32 mmu_update_extended; | |
92 | u32 mmu_update_histo[MMU_UPDATE_HISTO]; | |
93 | ||
94 | u32 prot_commit; | |
95 | u32 prot_commit_batched; | |
96 | ||
97 | u32 set_pte_at; | |
98 | u32 set_pte_at_batched; | |
99 | u32 set_pte_at_pinned; | |
100 | u32 set_pte_at_current; | |
101 | u32 set_pte_at_kernel; | |
102 | } mmu_stats; | |
103 | ||
104 | static u8 zero_stats; | |
105 | ||
106 | static inline void check_zero(void) | |
107 | { | |
108 | if (unlikely(zero_stats)) { | |
109 | memset(&mmu_stats, 0, sizeof(mmu_stats)); | |
110 | zero_stats = 0; | |
111 | } | |
112 | } | |
113 | ||
114 | #define ADD_STATS(elem, val) \ | |
115 | do { check_zero(); mmu_stats.elem += (val); } while(0) | |
116 | ||
117 | #else /* !CONFIG_XEN_DEBUG_FS */ | |
118 | ||
119 | #define ADD_STATS(elem, val) do { (void)(val); } while(0) | |
120 | ||
121 | #endif /* CONFIG_XEN_DEBUG_FS */ | |
3b827c1b | 122 | |
319f3ba5 JF |
123 | |
124 | /* | |
125 | * Identity map, in addition to plain kernel map. This needs to be | |
126 | * large enough to allocate page table pages to allocate the rest. | |
127 | * Each page can map 2MB. | |
128 | */ | |
129 | static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss; | |
130 | ||
131 | #ifdef CONFIG_X86_64 | |
132 | /* l3 pud for userspace vsyscall mapping */ | |
133 | static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss; | |
134 | #endif /* CONFIG_X86_64 */ | |
135 | ||
136 | /* | |
137 | * Note about cr3 (pagetable base) values: | |
138 | * | |
139 | * xen_cr3 contains the current logical cr3 value; it contains the | |
140 | * last set cr3. This may not be the current effective cr3, because | |
141 | * its update may be being lazily deferred. However, a vcpu looking | |
142 | * at its own cr3 can use this value knowing that it everything will | |
143 | * be self-consistent. | |
144 | * | |
145 | * xen_current_cr3 contains the actual vcpu cr3; it is set once the | |
146 | * hypercall to set the vcpu cr3 is complete (so it may be a little | |
147 | * out of date, but it will never be set early). If one vcpu is | |
148 | * looking at another vcpu's cr3 value, it should use this variable. | |
149 | */ | |
150 | DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */ | |
151 | DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */ | |
152 | ||
153 | ||
d6182fbf JF |
154 | /* |
155 | * Just beyond the highest usermode address. STACK_TOP_MAX has a | |
156 | * redzone above it, so round it up to a PGD boundary. | |
157 | */ | |
158 | #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK) | |
159 | ||
160 | ||
d451bb7a | 161 | #define P2M_ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(unsigned long)) |
cf0923ea | 162 | #define TOP_ENTRIES (MAX_DOMAIN_PAGES / P2M_ENTRIES_PER_PAGE) |
d451bb7a | 163 | |
cf0923ea | 164 | /* Placeholder for holes in the address space */ |
cbcd79c2 | 165 | static unsigned long p2m_missing[P2M_ENTRIES_PER_PAGE] __page_aligned_data = |
cf0923ea JF |
166 | { [ 0 ... P2M_ENTRIES_PER_PAGE-1 ] = ~0UL }; |
167 | ||
168 | /* Array of pointers to pages containing p2m entries */ | |
cbcd79c2 | 169 | static unsigned long *p2m_top[TOP_ENTRIES] __page_aligned_data = |
cf0923ea | 170 | { [ 0 ... TOP_ENTRIES - 1] = &p2m_missing[0] }; |
d451bb7a | 171 | |
d5edbc1f | 172 | /* Arrays of p2m arrays expressed in mfns used for save/restore */ |
cbcd79c2 | 173 | static unsigned long p2m_top_mfn[TOP_ENTRIES] __page_aligned_bss; |
d5edbc1f | 174 | |
cbcd79c2 JF |
175 | static unsigned long p2m_top_mfn_list[TOP_ENTRIES / P2M_ENTRIES_PER_PAGE] |
176 | __page_aligned_bss; | |
d5edbc1f | 177 | |
d451bb7a JF |
178 | static inline unsigned p2m_top_index(unsigned long pfn) |
179 | { | |
8006ec3e | 180 | BUG_ON(pfn >= MAX_DOMAIN_PAGES); |
d451bb7a JF |
181 | return pfn / P2M_ENTRIES_PER_PAGE; |
182 | } | |
183 | ||
184 | static inline unsigned p2m_index(unsigned long pfn) | |
185 | { | |
186 | return pfn % P2M_ENTRIES_PER_PAGE; | |
187 | } | |
188 | ||
d5edbc1f | 189 | /* Build the parallel p2m_top_mfn structures */ |
fa24ba62 | 190 | void xen_build_mfn_list_list(void) |
d5edbc1f JF |
191 | { |
192 | unsigned pfn, idx; | |
193 | ||
f63c2f24 | 194 | for (pfn = 0; pfn < MAX_DOMAIN_PAGES; pfn += P2M_ENTRIES_PER_PAGE) { |
d5edbc1f JF |
195 | unsigned topidx = p2m_top_index(pfn); |
196 | ||
197 | p2m_top_mfn[topidx] = virt_to_mfn(p2m_top[topidx]); | |
198 | } | |
199 | ||
f63c2f24 | 200 | for (idx = 0; idx < ARRAY_SIZE(p2m_top_mfn_list); idx++) { |
d5edbc1f JF |
201 | unsigned topidx = idx * P2M_ENTRIES_PER_PAGE; |
202 | p2m_top_mfn_list[idx] = virt_to_mfn(&p2m_top_mfn[topidx]); | |
203 | } | |
cdaead6b | 204 | } |
d5edbc1f | 205 | |
cdaead6b JF |
206 | void xen_setup_mfn_list_list(void) |
207 | { | |
d5edbc1f JF |
208 | BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); |
209 | ||
210 | HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list = | |
211 | virt_to_mfn(p2m_top_mfn_list); | |
212 | HYPERVISOR_shared_info->arch.max_pfn = xen_start_info->nr_pages; | |
213 | } | |
214 | ||
215 | /* Set up p2m_top to point to the domain-builder provided p2m pages */ | |
d451bb7a JF |
216 | void __init xen_build_dynamic_phys_to_machine(void) |
217 | { | |
d451bb7a | 218 | unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list; |
8006ec3e | 219 | unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages); |
d5edbc1f | 220 | unsigned pfn; |
d451bb7a | 221 | |
f63c2f24 | 222 | for (pfn = 0; pfn < max_pfn; pfn += P2M_ENTRIES_PER_PAGE) { |
d451bb7a JF |
223 | unsigned topidx = p2m_top_index(pfn); |
224 | ||
225 | p2m_top[topidx] = &mfn_list[pfn]; | |
226 | } | |
cdaead6b JF |
227 | |
228 | xen_build_mfn_list_list(); | |
d451bb7a JF |
229 | } |
230 | ||
231 | unsigned long get_phys_to_machine(unsigned long pfn) | |
232 | { | |
233 | unsigned topidx, idx; | |
234 | ||
8006ec3e JF |
235 | if (unlikely(pfn >= MAX_DOMAIN_PAGES)) |
236 | return INVALID_P2M_ENTRY; | |
237 | ||
d451bb7a | 238 | topidx = p2m_top_index(pfn); |
d451bb7a JF |
239 | idx = p2m_index(pfn); |
240 | return p2m_top[topidx][idx]; | |
241 | } | |
15ce6005 | 242 | EXPORT_SYMBOL_GPL(get_phys_to_machine); |
d451bb7a | 243 | |
e791ca0f JF |
244 | /* install a new p2m_top page */ |
245 | bool install_p2mtop_page(unsigned long pfn, unsigned long *p) | |
d451bb7a | 246 | { |
e791ca0f JF |
247 | unsigned topidx = p2m_top_index(pfn); |
248 | unsigned long **pfnp, *mfnp; | |
d451bb7a JF |
249 | unsigned i; |
250 | ||
e791ca0f JF |
251 | pfnp = &p2m_top[topidx]; |
252 | mfnp = &p2m_top_mfn[topidx]; | |
d451bb7a | 253 | |
f63c2f24 | 254 | for (i = 0; i < P2M_ENTRIES_PER_PAGE; i++) |
d451bb7a JF |
255 | p[i] = INVALID_P2M_ENTRY; |
256 | ||
e791ca0f | 257 | if (cmpxchg(pfnp, p2m_missing, p) == p2m_missing) { |
d5edbc1f | 258 | *mfnp = virt_to_mfn(p); |
e791ca0f JF |
259 | return true; |
260 | } | |
261 | ||
262 | return false; | |
d451bb7a JF |
263 | } |
264 | ||
e791ca0f | 265 | static void alloc_p2m(unsigned long pfn) |
d451bb7a | 266 | { |
e791ca0f | 267 | unsigned long *p; |
d451bb7a | 268 | |
e791ca0f JF |
269 | p = (void *)__get_free_page(GFP_KERNEL | __GFP_NOFAIL); |
270 | BUG_ON(p == NULL); | |
271 | ||
272 | if (!install_p2mtop_page(pfn, p)) | |
273 | free_page((unsigned long)p); | |
274 | } | |
275 | ||
276 | /* Try to install p2m mapping; fail if intermediate bits missing */ | |
277 | bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn) | |
278 | { | |
279 | unsigned topidx, idx; | |
8006ec3e JF |
280 | |
281 | if (unlikely(pfn >= MAX_DOMAIN_PAGES)) { | |
282 | BUG_ON(mfn != INVALID_P2M_ENTRY); | |
e791ca0f | 283 | return true; |
d451bb7a JF |
284 | } |
285 | ||
286 | topidx = p2m_top_index(pfn); | |
cf0923ea | 287 | if (p2m_top[topidx] == p2m_missing) { |
d451bb7a | 288 | if (mfn == INVALID_P2M_ENTRY) |
e791ca0f JF |
289 | return true; |
290 | return false; | |
d451bb7a JF |
291 | } |
292 | ||
293 | idx = p2m_index(pfn); | |
294 | p2m_top[topidx][idx] = mfn; | |
e791ca0f JF |
295 | |
296 | return true; | |
297 | } | |
298 | ||
299 | void set_phys_to_machine(unsigned long pfn, unsigned long mfn) | |
300 | { | |
301 | if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) { | |
302 | BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY); | |
303 | return; | |
304 | } | |
305 | ||
306 | if (unlikely(!__set_phys_to_machine(pfn, mfn))) { | |
307 | alloc_p2m(pfn); | |
308 | ||
309 | if (!__set_phys_to_machine(pfn, mfn)) | |
310 | BUG(); | |
311 | } | |
d451bb7a JF |
312 | } |
313 | ||
9976b39b JF |
314 | unsigned long arbitrary_virt_to_mfn(void *vaddr) |
315 | { | |
316 | xmaddr_t maddr = arbitrary_virt_to_machine(vaddr); | |
317 | ||
318 | return PFN_DOWN(maddr.maddr); | |
319 | } | |
320 | ||
ce803e70 | 321 | xmaddr_t arbitrary_virt_to_machine(void *vaddr) |
3b827c1b | 322 | { |
ce803e70 | 323 | unsigned long address = (unsigned long)vaddr; |
da7bfc50 | 324 | unsigned int level; |
9f32d21c CL |
325 | pte_t *pte; |
326 | unsigned offset; | |
3b827c1b | 327 | |
9f32d21c CL |
328 | /* |
329 | * if the PFN is in the linear mapped vaddr range, we can just use | |
330 | * the (quick) virt_to_machine() p2m lookup | |
331 | */ | |
332 | if (virt_addr_valid(vaddr)) | |
333 | return virt_to_machine(vaddr); | |
334 | ||
335 | /* otherwise we have to do a (slower) full page-table walk */ | |
3b827c1b | 336 | |
9f32d21c CL |
337 | pte = lookup_address(address, &level); |
338 | BUG_ON(pte == NULL); | |
339 | offset = address & ~PAGE_MASK; | |
ebd879e3 | 340 | return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset); |
3b827c1b JF |
341 | } |
342 | ||
343 | void make_lowmem_page_readonly(void *vaddr) | |
344 | { | |
345 | pte_t *pte, ptev; | |
346 | unsigned long address = (unsigned long)vaddr; | |
da7bfc50 | 347 | unsigned int level; |
3b827c1b | 348 | |
f0646e43 | 349 | pte = lookup_address(address, &level); |
3b827c1b JF |
350 | BUG_ON(pte == NULL); |
351 | ||
352 | ptev = pte_wrprotect(*pte); | |
353 | ||
354 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
355 | BUG(); | |
356 | } | |
357 | ||
358 | void make_lowmem_page_readwrite(void *vaddr) | |
359 | { | |
360 | pte_t *pte, ptev; | |
361 | unsigned long address = (unsigned long)vaddr; | |
da7bfc50 | 362 | unsigned int level; |
3b827c1b | 363 | |
f0646e43 | 364 | pte = lookup_address(address, &level); |
3b827c1b JF |
365 | BUG_ON(pte == NULL); |
366 | ||
367 | ptev = pte_mkwrite(*pte); | |
368 | ||
369 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
370 | BUG(); | |
371 | } | |
372 | ||
373 | ||
7708ad64 | 374 | static bool xen_page_pinned(void *ptr) |
e2426cf8 JF |
375 | { |
376 | struct page *page = virt_to_page(ptr); | |
377 | ||
378 | return PagePinned(page); | |
379 | } | |
380 | ||
7708ad64 | 381 | static void xen_extend_mmu_update(const struct mmu_update *update) |
3b827c1b | 382 | { |
d66bf8fc JF |
383 | struct multicall_space mcs; |
384 | struct mmu_update *u; | |
3b827c1b | 385 | |
400d3494 JF |
386 | mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u)); |
387 | ||
994025ca JF |
388 | if (mcs.mc != NULL) { |
389 | ADD_STATS(mmu_update_extended, 1); | |
390 | ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1); | |
391 | ||
400d3494 | 392 | mcs.mc->args[1]++; |
994025ca JF |
393 | |
394 | if (mcs.mc->args[1] < MMU_UPDATE_HISTO) | |
395 | ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1); | |
396 | else | |
397 | ADD_STATS(mmu_update_histo[0], 1); | |
398 | } else { | |
399 | ADD_STATS(mmu_update, 1); | |
400d3494 JF |
400 | mcs = __xen_mc_entry(sizeof(*u)); |
401 | MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); | |
994025ca | 402 | ADD_STATS(mmu_update_histo[1], 1); |
400d3494 | 403 | } |
d66bf8fc | 404 | |
d66bf8fc | 405 | u = mcs.args; |
400d3494 JF |
406 | *u = *update; |
407 | } | |
408 | ||
409 | void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val) | |
410 | { | |
411 | struct mmu_update u; | |
412 | ||
413 | preempt_disable(); | |
414 | ||
415 | xen_mc_batch(); | |
416 | ||
ce803e70 JF |
417 | /* ptr may be ioremapped for 64-bit pagetable setup */ |
418 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
400d3494 | 419 | u.val = pmd_val_ma(val); |
7708ad64 | 420 | xen_extend_mmu_update(&u); |
d66bf8fc | 421 | |
994025ca JF |
422 | ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); |
423 | ||
d66bf8fc JF |
424 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
425 | ||
426 | preempt_enable(); | |
3b827c1b JF |
427 | } |
428 | ||
e2426cf8 JF |
429 | void xen_set_pmd(pmd_t *ptr, pmd_t val) |
430 | { | |
994025ca JF |
431 | ADD_STATS(pmd_update, 1); |
432 | ||
e2426cf8 JF |
433 | /* If page is not pinned, we can just update the entry |
434 | directly */ | |
7708ad64 | 435 | if (!xen_page_pinned(ptr)) { |
e2426cf8 JF |
436 | *ptr = val; |
437 | return; | |
438 | } | |
439 | ||
994025ca JF |
440 | ADD_STATS(pmd_update_pinned, 1); |
441 | ||
e2426cf8 JF |
442 | xen_set_pmd_hyper(ptr, val); |
443 | } | |
444 | ||
3b827c1b JF |
445 | /* |
446 | * Associate a virtual page frame with a given physical page frame | |
447 | * and protection flags for that frame. | |
448 | */ | |
449 | void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags) | |
450 | { | |
836fe2f2 | 451 | set_pte_vaddr(vaddr, mfn_pte(mfn, flags)); |
3b827c1b JF |
452 | } |
453 | ||
454 | void xen_set_pte_at(struct mm_struct *mm, unsigned long addr, | |
455 | pte_t *ptep, pte_t pteval) | |
456 | { | |
994025ca JF |
457 | ADD_STATS(set_pte_at, 1); |
458 | // ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep)); | |
459 | ADD_STATS(set_pte_at_current, mm == current->mm); | |
460 | ADD_STATS(set_pte_at_kernel, mm == &init_mm); | |
461 | ||
d66bf8fc | 462 | if (mm == current->mm || mm == &init_mm) { |
8965c1c0 | 463 | if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) { |
d66bf8fc JF |
464 | struct multicall_space mcs; |
465 | mcs = xen_mc_entry(0); | |
466 | ||
467 | MULTI_update_va_mapping(mcs.mc, addr, pteval, 0); | |
994025ca | 468 | ADD_STATS(set_pte_at_batched, 1); |
d66bf8fc | 469 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
2bd50036 | 470 | goto out; |
d66bf8fc JF |
471 | } else |
472 | if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0) | |
2bd50036 | 473 | goto out; |
d66bf8fc JF |
474 | } |
475 | xen_set_pte(ptep, pteval); | |
2bd50036 | 476 | |
2829b449 | 477 | out: return; |
3b827c1b JF |
478 | } |
479 | ||
f63c2f24 T |
480 | pte_t xen_ptep_modify_prot_start(struct mm_struct *mm, |
481 | unsigned long addr, pte_t *ptep) | |
947a69c9 | 482 | { |
e57778a1 JF |
483 | /* Just return the pte as-is. We preserve the bits on commit */ |
484 | return *ptep; | |
485 | } | |
486 | ||
487 | void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr, | |
488 | pte_t *ptep, pte_t pte) | |
489 | { | |
400d3494 | 490 | struct mmu_update u; |
e57778a1 | 491 | |
400d3494 | 492 | xen_mc_batch(); |
947a69c9 | 493 | |
9f32d21c | 494 | u.ptr = arbitrary_virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD; |
400d3494 | 495 | u.val = pte_val_ma(pte); |
7708ad64 | 496 | xen_extend_mmu_update(&u); |
947a69c9 | 497 | |
994025ca JF |
498 | ADD_STATS(prot_commit, 1); |
499 | ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); | |
500 | ||
e57778a1 | 501 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
947a69c9 JF |
502 | } |
503 | ||
ebb9cfe2 JF |
504 | /* Assume pteval_t is equivalent to all the other *val_t types. */ |
505 | static pteval_t pte_mfn_to_pfn(pteval_t val) | |
947a69c9 | 506 | { |
ebb9cfe2 | 507 | if (val & _PAGE_PRESENT) { |
59438c9f | 508 | unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; |
77be1fab | 509 | pteval_t flags = val & PTE_FLAGS_MASK; |
d8355aca | 510 | val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags; |
ebb9cfe2 | 511 | } |
947a69c9 | 512 | |
ebb9cfe2 | 513 | return val; |
947a69c9 JF |
514 | } |
515 | ||
ebb9cfe2 | 516 | static pteval_t pte_pfn_to_mfn(pteval_t val) |
947a69c9 | 517 | { |
ebb9cfe2 | 518 | if (val & _PAGE_PRESENT) { |
59438c9f | 519 | unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; |
77be1fab | 520 | pteval_t flags = val & PTE_FLAGS_MASK; |
d8355aca | 521 | val = ((pteval_t)pfn_to_mfn(pfn) << PAGE_SHIFT) | flags; |
947a69c9 JF |
522 | } |
523 | ||
ebb9cfe2 | 524 | return val; |
947a69c9 JF |
525 | } |
526 | ||
ebb9cfe2 | 527 | pteval_t xen_pte_val(pte_t pte) |
947a69c9 | 528 | { |
ebb9cfe2 | 529 | return pte_mfn_to_pfn(pte.pte); |
947a69c9 | 530 | } |
da5de7c2 | 531 | PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val); |
947a69c9 | 532 | |
947a69c9 JF |
533 | pgdval_t xen_pgd_val(pgd_t pgd) |
534 | { | |
ebb9cfe2 | 535 | return pte_mfn_to_pfn(pgd.pgd); |
947a69c9 | 536 | } |
da5de7c2 | 537 | PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val); |
947a69c9 JF |
538 | |
539 | pte_t xen_make_pte(pteval_t pte) | |
540 | { | |
ebb9cfe2 JF |
541 | pte = pte_pfn_to_mfn(pte); |
542 | return native_make_pte(pte); | |
947a69c9 | 543 | } |
da5de7c2 | 544 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte); |
947a69c9 JF |
545 | |
546 | pgd_t xen_make_pgd(pgdval_t pgd) | |
547 | { | |
ebb9cfe2 JF |
548 | pgd = pte_pfn_to_mfn(pgd); |
549 | return native_make_pgd(pgd); | |
947a69c9 | 550 | } |
da5de7c2 | 551 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd); |
947a69c9 JF |
552 | |
553 | pmdval_t xen_pmd_val(pmd_t pmd) | |
554 | { | |
ebb9cfe2 | 555 | return pte_mfn_to_pfn(pmd.pmd); |
947a69c9 | 556 | } |
da5de7c2 | 557 | PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val); |
28499143 | 558 | |
e2426cf8 | 559 | void xen_set_pud_hyper(pud_t *ptr, pud_t val) |
f4f97b3e | 560 | { |
400d3494 | 561 | struct mmu_update u; |
f4f97b3e | 562 | |
d66bf8fc JF |
563 | preempt_disable(); |
564 | ||
400d3494 JF |
565 | xen_mc_batch(); |
566 | ||
ce803e70 JF |
567 | /* ptr may be ioremapped for 64-bit pagetable setup */ |
568 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
400d3494 | 569 | u.val = pud_val_ma(val); |
7708ad64 | 570 | xen_extend_mmu_update(&u); |
d66bf8fc | 571 | |
994025ca JF |
572 | ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); |
573 | ||
d66bf8fc JF |
574 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
575 | ||
576 | preempt_enable(); | |
f4f97b3e JF |
577 | } |
578 | ||
e2426cf8 JF |
579 | void xen_set_pud(pud_t *ptr, pud_t val) |
580 | { | |
994025ca JF |
581 | ADD_STATS(pud_update, 1); |
582 | ||
e2426cf8 JF |
583 | /* If page is not pinned, we can just update the entry |
584 | directly */ | |
7708ad64 | 585 | if (!xen_page_pinned(ptr)) { |
e2426cf8 JF |
586 | *ptr = val; |
587 | return; | |
588 | } | |
589 | ||
994025ca JF |
590 | ADD_STATS(pud_update_pinned, 1); |
591 | ||
e2426cf8 JF |
592 | xen_set_pud_hyper(ptr, val); |
593 | } | |
594 | ||
f4f97b3e JF |
595 | void xen_set_pte(pte_t *ptep, pte_t pte) |
596 | { | |
994025ca JF |
597 | ADD_STATS(pte_update, 1); |
598 | // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep)); | |
599 | ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); | |
600 | ||
f6e58732 | 601 | #ifdef CONFIG_X86_PAE |
f4f97b3e JF |
602 | ptep->pte_high = pte.pte_high; |
603 | smp_wmb(); | |
604 | ptep->pte_low = pte.pte_low; | |
f6e58732 JF |
605 | #else |
606 | *ptep = pte; | |
607 | #endif | |
f4f97b3e JF |
608 | } |
609 | ||
f6e58732 | 610 | #ifdef CONFIG_X86_PAE |
3b827c1b JF |
611 | void xen_set_pte_atomic(pte_t *ptep, pte_t pte) |
612 | { | |
f6e58732 | 613 | set_64bit((u64 *)ptep, native_pte_val(pte)); |
3b827c1b JF |
614 | } |
615 | ||
616 | void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep) | |
617 | { | |
618 | ptep->pte_low = 0; | |
619 | smp_wmb(); /* make sure low gets written first */ | |
620 | ptep->pte_high = 0; | |
621 | } | |
622 | ||
623 | void xen_pmd_clear(pmd_t *pmdp) | |
624 | { | |
e2426cf8 | 625 | set_pmd(pmdp, __pmd(0)); |
3b827c1b | 626 | } |
f6e58732 | 627 | #endif /* CONFIG_X86_PAE */ |
3b827c1b | 628 | |
abf33038 | 629 | pmd_t xen_make_pmd(pmdval_t pmd) |
3b827c1b | 630 | { |
ebb9cfe2 | 631 | pmd = pte_pfn_to_mfn(pmd); |
947a69c9 | 632 | return native_make_pmd(pmd); |
3b827c1b | 633 | } |
da5de7c2 | 634 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd); |
3b827c1b | 635 | |
f6e58732 JF |
636 | #if PAGETABLE_LEVELS == 4 |
637 | pudval_t xen_pud_val(pud_t pud) | |
638 | { | |
639 | return pte_mfn_to_pfn(pud.pud); | |
640 | } | |
da5de7c2 | 641 | PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val); |
f6e58732 JF |
642 | |
643 | pud_t xen_make_pud(pudval_t pud) | |
644 | { | |
645 | pud = pte_pfn_to_mfn(pud); | |
646 | ||
647 | return native_make_pud(pud); | |
648 | } | |
da5de7c2 | 649 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud); |
f6e58732 | 650 | |
d6182fbf | 651 | pgd_t *xen_get_user_pgd(pgd_t *pgd) |
f6e58732 | 652 | { |
d6182fbf JF |
653 | pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK); |
654 | unsigned offset = pgd - pgd_page; | |
655 | pgd_t *user_ptr = NULL; | |
f6e58732 | 656 | |
d6182fbf JF |
657 | if (offset < pgd_index(USER_LIMIT)) { |
658 | struct page *page = virt_to_page(pgd_page); | |
659 | user_ptr = (pgd_t *)page->private; | |
660 | if (user_ptr) | |
661 | user_ptr += offset; | |
662 | } | |
f6e58732 | 663 | |
d6182fbf JF |
664 | return user_ptr; |
665 | } | |
666 | ||
667 | static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val) | |
668 | { | |
669 | struct mmu_update u; | |
f6e58732 JF |
670 | |
671 | u.ptr = virt_to_machine(ptr).maddr; | |
672 | u.val = pgd_val_ma(val); | |
7708ad64 | 673 | xen_extend_mmu_update(&u); |
d6182fbf JF |
674 | } |
675 | ||
676 | /* | |
677 | * Raw hypercall-based set_pgd, intended for in early boot before | |
678 | * there's a page structure. This implies: | |
679 | * 1. The only existing pagetable is the kernel's | |
680 | * 2. It is always pinned | |
681 | * 3. It has no user pagetable attached to it | |
682 | */ | |
683 | void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val) | |
684 | { | |
685 | preempt_disable(); | |
686 | ||
687 | xen_mc_batch(); | |
688 | ||
689 | __xen_set_pgd_hyper(ptr, val); | |
f6e58732 JF |
690 | |
691 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
692 | ||
693 | preempt_enable(); | |
694 | } | |
695 | ||
696 | void xen_set_pgd(pgd_t *ptr, pgd_t val) | |
697 | { | |
d6182fbf JF |
698 | pgd_t *user_ptr = xen_get_user_pgd(ptr); |
699 | ||
994025ca JF |
700 | ADD_STATS(pgd_update, 1); |
701 | ||
f6e58732 JF |
702 | /* If page is not pinned, we can just update the entry |
703 | directly */ | |
7708ad64 | 704 | if (!xen_page_pinned(ptr)) { |
f6e58732 | 705 | *ptr = val; |
d6182fbf | 706 | if (user_ptr) { |
7708ad64 | 707 | WARN_ON(xen_page_pinned(user_ptr)); |
d6182fbf JF |
708 | *user_ptr = val; |
709 | } | |
f6e58732 JF |
710 | return; |
711 | } | |
712 | ||
994025ca JF |
713 | ADD_STATS(pgd_update_pinned, 1); |
714 | ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); | |
715 | ||
d6182fbf JF |
716 | /* If it's pinned, then we can at least batch the kernel and |
717 | user updates together. */ | |
718 | xen_mc_batch(); | |
719 | ||
720 | __xen_set_pgd_hyper(ptr, val); | |
721 | if (user_ptr) | |
722 | __xen_set_pgd_hyper(user_ptr, val); | |
723 | ||
724 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
f6e58732 JF |
725 | } |
726 | #endif /* PAGETABLE_LEVELS == 4 */ | |
727 | ||
f4f97b3e | 728 | /* |
5deb30d1 JF |
729 | * (Yet another) pagetable walker. This one is intended for pinning a |
730 | * pagetable. This means that it walks a pagetable and calls the | |
731 | * callback function on each page it finds making up the page table, | |
732 | * at every level. It walks the entire pagetable, but it only bothers | |
733 | * pinning pte pages which are below limit. In the normal case this | |
734 | * will be STACK_TOP_MAX, but at boot we need to pin up to | |
735 | * FIXADDR_TOP. | |
736 | * | |
737 | * For 32-bit the important bit is that we don't pin beyond there, | |
738 | * because then we start getting into Xen's ptes. | |
739 | * | |
740 | * For 64-bit, we must skip the Xen hole in the middle of the address | |
741 | * space, just after the big x86-64 virtual hole. | |
742 | */ | |
86bbc2c2 IC |
743 | static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd, |
744 | int (*func)(struct mm_struct *mm, struct page *, | |
745 | enum pt_level), | |
746 | unsigned long limit) | |
3b827c1b | 747 | { |
f4f97b3e | 748 | int flush = 0; |
5deb30d1 JF |
749 | unsigned hole_low, hole_high; |
750 | unsigned pgdidx_limit, pudidx_limit, pmdidx_limit; | |
751 | unsigned pgdidx, pudidx, pmdidx; | |
f4f97b3e | 752 | |
5deb30d1 JF |
753 | /* The limit is the last byte to be touched */ |
754 | limit--; | |
755 | BUG_ON(limit >= FIXADDR_TOP); | |
3b827c1b JF |
756 | |
757 | if (xen_feature(XENFEAT_auto_translated_physmap)) | |
f4f97b3e JF |
758 | return 0; |
759 | ||
5deb30d1 JF |
760 | /* |
761 | * 64-bit has a great big hole in the middle of the address | |
762 | * space, which contains the Xen mappings. On 32-bit these | |
763 | * will end up making a zero-sized hole and so is a no-op. | |
764 | */ | |
d6182fbf | 765 | hole_low = pgd_index(USER_LIMIT); |
5deb30d1 JF |
766 | hole_high = pgd_index(PAGE_OFFSET); |
767 | ||
768 | pgdidx_limit = pgd_index(limit); | |
769 | #if PTRS_PER_PUD > 1 | |
770 | pudidx_limit = pud_index(limit); | |
771 | #else | |
772 | pudidx_limit = 0; | |
773 | #endif | |
774 | #if PTRS_PER_PMD > 1 | |
775 | pmdidx_limit = pmd_index(limit); | |
776 | #else | |
777 | pmdidx_limit = 0; | |
778 | #endif | |
779 | ||
5deb30d1 | 780 | for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) { |
f4f97b3e | 781 | pud_t *pud; |
3b827c1b | 782 | |
5deb30d1 JF |
783 | if (pgdidx >= hole_low && pgdidx < hole_high) |
784 | continue; | |
f4f97b3e | 785 | |
5deb30d1 | 786 | if (!pgd_val(pgd[pgdidx])) |
3b827c1b | 787 | continue; |
f4f97b3e | 788 | |
5deb30d1 | 789 | pud = pud_offset(&pgd[pgdidx], 0); |
3b827c1b JF |
790 | |
791 | if (PTRS_PER_PUD > 1) /* not folded */ | |
eefb47f6 | 792 | flush |= (*func)(mm, virt_to_page(pud), PT_PUD); |
f4f97b3e | 793 | |
5deb30d1 | 794 | for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) { |
f4f97b3e | 795 | pmd_t *pmd; |
f4f97b3e | 796 | |
5deb30d1 JF |
797 | if (pgdidx == pgdidx_limit && |
798 | pudidx > pudidx_limit) | |
799 | goto out; | |
3b827c1b | 800 | |
5deb30d1 | 801 | if (pud_none(pud[pudidx])) |
3b827c1b | 802 | continue; |
f4f97b3e | 803 | |
5deb30d1 | 804 | pmd = pmd_offset(&pud[pudidx], 0); |
3b827c1b JF |
805 | |
806 | if (PTRS_PER_PMD > 1) /* not folded */ | |
eefb47f6 | 807 | flush |= (*func)(mm, virt_to_page(pmd), PT_PMD); |
f4f97b3e | 808 | |
5deb30d1 JF |
809 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) { |
810 | struct page *pte; | |
811 | ||
812 | if (pgdidx == pgdidx_limit && | |
813 | pudidx == pudidx_limit && | |
814 | pmdidx > pmdidx_limit) | |
815 | goto out; | |
3b827c1b | 816 | |
5deb30d1 | 817 | if (pmd_none(pmd[pmdidx])) |
3b827c1b JF |
818 | continue; |
819 | ||
5deb30d1 | 820 | pte = pmd_page(pmd[pmdidx]); |
eefb47f6 | 821 | flush |= (*func)(mm, pte, PT_PTE); |
3b827c1b JF |
822 | } |
823 | } | |
824 | } | |
11ad93e5 | 825 | |
5deb30d1 | 826 | out: |
11ad93e5 JF |
827 | /* Do the top level last, so that the callbacks can use it as |
828 | a cue to do final things like tlb flushes. */ | |
eefb47f6 | 829 | flush |= (*func)(mm, virt_to_page(pgd), PT_PGD); |
f4f97b3e JF |
830 | |
831 | return flush; | |
3b827c1b JF |
832 | } |
833 | ||
86bbc2c2 IC |
834 | static int xen_pgd_walk(struct mm_struct *mm, |
835 | int (*func)(struct mm_struct *mm, struct page *, | |
836 | enum pt_level), | |
837 | unsigned long limit) | |
838 | { | |
839 | return __xen_pgd_walk(mm, mm->pgd, func, limit); | |
840 | } | |
841 | ||
7708ad64 JF |
842 | /* If we're using split pte locks, then take the page's lock and |
843 | return a pointer to it. Otherwise return NULL. */ | |
eefb47f6 | 844 | static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm) |
74260714 JF |
845 | { |
846 | spinlock_t *ptl = NULL; | |
847 | ||
f7d0b926 | 848 | #if USE_SPLIT_PTLOCKS |
74260714 | 849 | ptl = __pte_lockptr(page); |
eefb47f6 | 850 | spin_lock_nest_lock(ptl, &mm->page_table_lock); |
74260714 JF |
851 | #endif |
852 | ||
853 | return ptl; | |
854 | } | |
855 | ||
7708ad64 | 856 | static void xen_pte_unlock(void *v) |
74260714 JF |
857 | { |
858 | spinlock_t *ptl = v; | |
859 | spin_unlock(ptl); | |
860 | } | |
861 | ||
862 | static void xen_do_pin(unsigned level, unsigned long pfn) | |
863 | { | |
864 | struct mmuext_op *op; | |
865 | struct multicall_space mcs; | |
866 | ||
867 | mcs = __xen_mc_entry(sizeof(*op)); | |
868 | op = mcs.args; | |
869 | op->cmd = level; | |
870 | op->arg1.mfn = pfn_to_mfn(pfn); | |
871 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
872 | } | |
873 | ||
eefb47f6 JF |
874 | static int xen_pin_page(struct mm_struct *mm, struct page *page, |
875 | enum pt_level level) | |
f4f97b3e | 876 | { |
d60cd46b | 877 | unsigned pgfl = TestSetPagePinned(page); |
f4f97b3e JF |
878 | int flush; |
879 | ||
880 | if (pgfl) | |
881 | flush = 0; /* already pinned */ | |
882 | else if (PageHighMem(page)) | |
883 | /* kmaps need flushing if we found an unpinned | |
884 | highpage */ | |
885 | flush = 1; | |
886 | else { | |
887 | void *pt = lowmem_page_address(page); | |
888 | unsigned long pfn = page_to_pfn(page); | |
889 | struct multicall_space mcs = __xen_mc_entry(0); | |
74260714 | 890 | spinlock_t *ptl; |
f4f97b3e JF |
891 | |
892 | flush = 0; | |
893 | ||
11ad93e5 JF |
894 | /* |
895 | * We need to hold the pagetable lock between the time | |
896 | * we make the pagetable RO and when we actually pin | |
897 | * it. If we don't, then other users may come in and | |
898 | * attempt to update the pagetable by writing it, | |
899 | * which will fail because the memory is RO but not | |
900 | * pinned, so Xen won't do the trap'n'emulate. | |
901 | * | |
902 | * If we're using split pte locks, we can't hold the | |
903 | * entire pagetable's worth of locks during the | |
904 | * traverse, because we may wrap the preempt count (8 | |
905 | * bits). The solution is to mark RO and pin each PTE | |
906 | * page while holding the lock. This means the number | |
907 | * of locks we end up holding is never more than a | |
908 | * batch size (~32 entries, at present). | |
909 | * | |
910 | * If we're not using split pte locks, we needn't pin | |
911 | * the PTE pages independently, because we're | |
912 | * protected by the overall pagetable lock. | |
913 | */ | |
74260714 JF |
914 | ptl = NULL; |
915 | if (level == PT_PTE) | |
eefb47f6 | 916 | ptl = xen_pte_lock(page, mm); |
74260714 | 917 | |
f4f97b3e JF |
918 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, |
919 | pfn_pte(pfn, PAGE_KERNEL_RO), | |
74260714 JF |
920 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
921 | ||
11ad93e5 | 922 | if (ptl) { |
74260714 JF |
923 | xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn); |
924 | ||
74260714 JF |
925 | /* Queue a deferred unlock for when this batch |
926 | is completed. */ | |
7708ad64 | 927 | xen_mc_callback(xen_pte_unlock, ptl); |
74260714 | 928 | } |
f4f97b3e JF |
929 | } |
930 | ||
931 | return flush; | |
932 | } | |
3b827c1b | 933 | |
f4f97b3e JF |
934 | /* This is called just after a mm has been created, but it has not |
935 | been used yet. We need to make sure that its pagetable is all | |
936 | read-only, and can be pinned. */ | |
eefb47f6 | 937 | static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd) |
3b827c1b | 938 | { |
d05fdf31 JF |
939 | vm_unmap_aliases(); |
940 | ||
f4f97b3e | 941 | xen_mc_batch(); |
3b827c1b | 942 | |
86bbc2c2 | 943 | if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) { |
d05fdf31 | 944 | /* re-enable interrupts for flushing */ |
f87e4cac | 945 | xen_mc_issue(0); |
d05fdf31 | 946 | |
f4f97b3e | 947 | kmap_flush_unused(); |
d05fdf31 | 948 | |
f87e4cac JF |
949 | xen_mc_batch(); |
950 | } | |
f4f97b3e | 951 | |
d6182fbf JF |
952 | #ifdef CONFIG_X86_64 |
953 | { | |
954 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
955 | ||
956 | xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd))); | |
957 | ||
958 | if (user_pgd) { | |
eefb47f6 | 959 | xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD); |
f63c2f24 T |
960 | xen_do_pin(MMUEXT_PIN_L4_TABLE, |
961 | PFN_DOWN(__pa(user_pgd))); | |
d6182fbf JF |
962 | } |
963 | } | |
964 | #else /* CONFIG_X86_32 */ | |
5deb30d1 JF |
965 | #ifdef CONFIG_X86_PAE |
966 | /* Need to make sure unshared kernel PMD is pinnable */ | |
47cb2ed9 | 967 | xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), |
eefb47f6 | 968 | PT_PMD); |
5deb30d1 | 969 | #endif |
28499143 | 970 | xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd))); |
d6182fbf | 971 | #endif /* CONFIG_X86_64 */ |
f4f97b3e | 972 | xen_mc_issue(0); |
3b827c1b JF |
973 | } |
974 | ||
eefb47f6 JF |
975 | static void xen_pgd_pin(struct mm_struct *mm) |
976 | { | |
977 | __xen_pgd_pin(mm, mm->pgd); | |
978 | } | |
979 | ||
0e91398f JF |
980 | /* |
981 | * On save, we need to pin all pagetables to make sure they get their | |
982 | * mfns turned into pfns. Search the list for any unpinned pgds and pin | |
983 | * them (unpinned pgds are not currently in use, probably because the | |
984 | * process is under construction or destruction). | |
eefb47f6 JF |
985 | * |
986 | * Expected to be called in stop_machine() ("equivalent to taking | |
987 | * every spinlock in the system"), so the locking doesn't really | |
988 | * matter all that much. | |
0e91398f JF |
989 | */ |
990 | void xen_mm_pin_all(void) | |
991 | { | |
992 | unsigned long flags; | |
993 | struct page *page; | |
74260714 | 994 | |
0e91398f | 995 | spin_lock_irqsave(&pgd_lock, flags); |
f4f97b3e | 996 | |
0e91398f JF |
997 | list_for_each_entry(page, &pgd_list, lru) { |
998 | if (!PagePinned(page)) { | |
eefb47f6 | 999 | __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page)); |
0e91398f JF |
1000 | SetPageSavePinned(page); |
1001 | } | |
1002 | } | |
1003 | ||
1004 | spin_unlock_irqrestore(&pgd_lock, flags); | |
3b827c1b JF |
1005 | } |
1006 | ||
c1f2f09e EH |
1007 | /* |
1008 | * The init_mm pagetable is really pinned as soon as its created, but | |
1009 | * that's before we have page structures to store the bits. So do all | |
1010 | * the book-keeping now. | |
1011 | */ | |
eefb47f6 JF |
1012 | static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page, |
1013 | enum pt_level level) | |
3b827c1b | 1014 | { |
f4f97b3e JF |
1015 | SetPagePinned(page); |
1016 | return 0; | |
1017 | } | |
3b827c1b | 1018 | |
b96229b5 | 1019 | static void __init xen_mark_init_mm_pinned(void) |
f4f97b3e | 1020 | { |
eefb47f6 | 1021 | xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP); |
f4f97b3e | 1022 | } |
3b827c1b | 1023 | |
eefb47f6 JF |
1024 | static int xen_unpin_page(struct mm_struct *mm, struct page *page, |
1025 | enum pt_level level) | |
f4f97b3e | 1026 | { |
d60cd46b | 1027 | unsigned pgfl = TestClearPagePinned(page); |
3b827c1b | 1028 | |
f4f97b3e JF |
1029 | if (pgfl && !PageHighMem(page)) { |
1030 | void *pt = lowmem_page_address(page); | |
1031 | unsigned long pfn = page_to_pfn(page); | |
74260714 JF |
1032 | spinlock_t *ptl = NULL; |
1033 | struct multicall_space mcs; | |
1034 | ||
11ad93e5 JF |
1035 | /* |
1036 | * Do the converse to pin_page. If we're using split | |
1037 | * pte locks, we must be holding the lock for while | |
1038 | * the pte page is unpinned but still RO to prevent | |
1039 | * concurrent updates from seeing it in this | |
1040 | * partially-pinned state. | |
1041 | */ | |
74260714 | 1042 | if (level == PT_PTE) { |
eefb47f6 | 1043 | ptl = xen_pte_lock(page, mm); |
74260714 | 1044 | |
11ad93e5 JF |
1045 | if (ptl) |
1046 | xen_do_pin(MMUEXT_UNPIN_TABLE, pfn); | |
74260714 JF |
1047 | } |
1048 | ||
1049 | mcs = __xen_mc_entry(0); | |
f4f97b3e JF |
1050 | |
1051 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, | |
1052 | pfn_pte(pfn, PAGE_KERNEL), | |
74260714 JF |
1053 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
1054 | ||
1055 | if (ptl) { | |
1056 | /* unlock when batch completed */ | |
7708ad64 | 1057 | xen_mc_callback(xen_pte_unlock, ptl); |
74260714 | 1058 | } |
f4f97b3e JF |
1059 | } |
1060 | ||
1061 | return 0; /* never need to flush on unpin */ | |
3b827c1b JF |
1062 | } |
1063 | ||
f4f97b3e | 1064 | /* Release a pagetables pages back as normal RW */ |
eefb47f6 | 1065 | static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd) |
f4f97b3e | 1066 | { |
f4f97b3e JF |
1067 | xen_mc_batch(); |
1068 | ||
74260714 | 1069 | xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); |
f4f97b3e | 1070 | |
d6182fbf JF |
1071 | #ifdef CONFIG_X86_64 |
1072 | { | |
1073 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
1074 | ||
1075 | if (user_pgd) { | |
f63c2f24 T |
1076 | xen_do_pin(MMUEXT_UNPIN_TABLE, |
1077 | PFN_DOWN(__pa(user_pgd))); | |
eefb47f6 | 1078 | xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD); |
d6182fbf JF |
1079 | } |
1080 | } | |
1081 | #endif | |
1082 | ||
5deb30d1 JF |
1083 | #ifdef CONFIG_X86_PAE |
1084 | /* Need to make sure unshared kernel PMD is unpinned */ | |
47cb2ed9 | 1085 | xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), |
eefb47f6 | 1086 | PT_PMD); |
5deb30d1 | 1087 | #endif |
d6182fbf | 1088 | |
86bbc2c2 | 1089 | __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT); |
f4f97b3e JF |
1090 | |
1091 | xen_mc_issue(0); | |
1092 | } | |
3b827c1b | 1093 | |
eefb47f6 JF |
1094 | static void xen_pgd_unpin(struct mm_struct *mm) |
1095 | { | |
1096 | __xen_pgd_unpin(mm, mm->pgd); | |
1097 | } | |
1098 | ||
0e91398f JF |
1099 | /* |
1100 | * On resume, undo any pinning done at save, so that the rest of the | |
1101 | * kernel doesn't see any unexpected pinned pagetables. | |
1102 | */ | |
1103 | void xen_mm_unpin_all(void) | |
1104 | { | |
1105 | unsigned long flags; | |
1106 | struct page *page; | |
1107 | ||
1108 | spin_lock_irqsave(&pgd_lock, flags); | |
1109 | ||
1110 | list_for_each_entry(page, &pgd_list, lru) { | |
1111 | if (PageSavePinned(page)) { | |
1112 | BUG_ON(!PagePinned(page)); | |
eefb47f6 | 1113 | __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page)); |
0e91398f JF |
1114 | ClearPageSavePinned(page); |
1115 | } | |
1116 | } | |
1117 | ||
1118 | spin_unlock_irqrestore(&pgd_lock, flags); | |
1119 | } | |
1120 | ||
3b827c1b JF |
1121 | void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next) |
1122 | { | |
f4f97b3e | 1123 | spin_lock(&next->page_table_lock); |
eefb47f6 | 1124 | xen_pgd_pin(next); |
f4f97b3e | 1125 | spin_unlock(&next->page_table_lock); |
3b827c1b JF |
1126 | } |
1127 | ||
1128 | void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm) | |
1129 | { | |
f4f97b3e | 1130 | spin_lock(&mm->page_table_lock); |
eefb47f6 | 1131 | xen_pgd_pin(mm); |
f4f97b3e | 1132 | spin_unlock(&mm->page_table_lock); |
3b827c1b JF |
1133 | } |
1134 | ||
3b827c1b | 1135 | |
f87e4cac JF |
1136 | #ifdef CONFIG_SMP |
1137 | /* Another cpu may still have their %cr3 pointing at the pagetable, so | |
1138 | we need to repoint it somewhere else before we can unpin it. */ | |
1139 | static void drop_other_mm_ref(void *info) | |
1140 | { | |
1141 | struct mm_struct *mm = info; | |
ce87b3d3 | 1142 | struct mm_struct *active_mm; |
3b827c1b | 1143 | |
9eb912d1 | 1144 | active_mm = percpu_read(cpu_tlbstate.active_mm); |
ce87b3d3 JF |
1145 | |
1146 | if (active_mm == mm) | |
f87e4cac | 1147 | leave_mm(smp_processor_id()); |
9f79991d JF |
1148 | |
1149 | /* If this cpu still has a stale cr3 reference, then make sure | |
1150 | it has been flushed. */ | |
7fd7d83d | 1151 | if (percpu_read(xen_current_cr3) == __pa(mm->pgd)) |
9f79991d | 1152 | load_cr3(swapper_pg_dir); |
f87e4cac | 1153 | } |
3b827c1b | 1154 | |
7708ad64 | 1155 | static void xen_drop_mm_ref(struct mm_struct *mm) |
f87e4cac | 1156 | { |
e4d98207 | 1157 | cpumask_var_t mask; |
9f79991d JF |
1158 | unsigned cpu; |
1159 | ||
f87e4cac JF |
1160 | if (current->active_mm == mm) { |
1161 | if (current->mm == mm) | |
1162 | load_cr3(swapper_pg_dir); | |
1163 | else | |
1164 | leave_mm(smp_processor_id()); | |
9f79991d JF |
1165 | } |
1166 | ||
1167 | /* Get the "official" set of cpus referring to our pagetable. */ | |
e4d98207 MT |
1168 | if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) { |
1169 | for_each_online_cpu(cpu) { | |
78f1c4d6 | 1170 | if (!cpumask_test_cpu(cpu, mm_cpumask(mm)) |
e4d98207 MT |
1171 | && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd)) |
1172 | continue; | |
1173 | smp_call_function_single(cpu, drop_other_mm_ref, mm, 1); | |
1174 | } | |
1175 | return; | |
1176 | } | |
78f1c4d6 | 1177 | cpumask_copy(mask, mm_cpumask(mm)); |
9f79991d JF |
1178 | |
1179 | /* It's possible that a vcpu may have a stale reference to our | |
1180 | cr3, because its in lazy mode, and it hasn't yet flushed | |
1181 | its set of pending hypercalls yet. In this case, we can | |
1182 | look at its actual current cr3 value, and force it to flush | |
1183 | if needed. */ | |
1184 | for_each_online_cpu(cpu) { | |
1185 | if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd)) | |
e4d98207 | 1186 | cpumask_set_cpu(cpu, mask); |
3b827c1b JF |
1187 | } |
1188 | ||
e4d98207 MT |
1189 | if (!cpumask_empty(mask)) |
1190 | smp_call_function_many(mask, drop_other_mm_ref, mm, 1); | |
1191 | free_cpumask_var(mask); | |
f87e4cac JF |
1192 | } |
1193 | #else | |
7708ad64 | 1194 | static void xen_drop_mm_ref(struct mm_struct *mm) |
f87e4cac JF |
1195 | { |
1196 | if (current->active_mm == mm) | |
1197 | load_cr3(swapper_pg_dir); | |
1198 | } | |
1199 | #endif | |
1200 | ||
1201 | /* | |
1202 | * While a process runs, Xen pins its pagetables, which means that the | |
1203 | * hypervisor forces it to be read-only, and it controls all updates | |
1204 | * to it. This means that all pagetable updates have to go via the | |
1205 | * hypervisor, which is moderately expensive. | |
1206 | * | |
1207 | * Since we're pulling the pagetable down, we switch to use init_mm, | |
1208 | * unpin old process pagetable and mark it all read-write, which | |
1209 | * allows further operations on it to be simple memory accesses. | |
1210 | * | |
1211 | * The only subtle point is that another CPU may be still using the | |
1212 | * pagetable because of lazy tlb flushing. This means we need need to | |
1213 | * switch all CPUs off this pagetable before we can unpin it. | |
1214 | */ | |
1215 | void xen_exit_mmap(struct mm_struct *mm) | |
1216 | { | |
1217 | get_cpu(); /* make sure we don't move around */ | |
7708ad64 | 1218 | xen_drop_mm_ref(mm); |
f87e4cac | 1219 | put_cpu(); |
3b827c1b | 1220 | |
f120f13e | 1221 | spin_lock(&mm->page_table_lock); |
df912ea4 JF |
1222 | |
1223 | /* pgd may not be pinned in the error exit path of execve */ | |
7708ad64 | 1224 | if (xen_page_pinned(mm->pgd)) |
eefb47f6 | 1225 | xen_pgd_unpin(mm); |
74260714 | 1226 | |
f120f13e | 1227 | spin_unlock(&mm->page_table_lock); |
3b827c1b | 1228 | } |
994025ca | 1229 | |
319f3ba5 JF |
1230 | static __init void xen_pagetable_setup_start(pgd_t *base) |
1231 | { | |
1232 | } | |
1233 | ||
f1d7062a TG |
1234 | static void xen_post_allocator_init(void); |
1235 | ||
319f3ba5 JF |
1236 | static __init void xen_pagetable_setup_done(pgd_t *base) |
1237 | { | |
1238 | xen_setup_shared_info(); | |
f1d7062a | 1239 | xen_post_allocator_init(); |
319f3ba5 JF |
1240 | } |
1241 | ||
1242 | static void xen_write_cr2(unsigned long cr2) | |
1243 | { | |
1244 | percpu_read(xen_vcpu)->arch.cr2 = cr2; | |
1245 | } | |
1246 | ||
1247 | static unsigned long xen_read_cr2(void) | |
1248 | { | |
1249 | return percpu_read(xen_vcpu)->arch.cr2; | |
1250 | } | |
1251 | ||
1252 | unsigned long xen_read_cr2_direct(void) | |
1253 | { | |
1254 | return percpu_read(xen_vcpu_info.arch.cr2); | |
1255 | } | |
1256 | ||
1257 | static void xen_flush_tlb(void) | |
1258 | { | |
1259 | struct mmuext_op *op; | |
1260 | struct multicall_space mcs; | |
1261 | ||
1262 | preempt_disable(); | |
1263 | ||
1264 | mcs = xen_mc_entry(sizeof(*op)); | |
1265 | ||
1266 | op = mcs.args; | |
1267 | op->cmd = MMUEXT_TLB_FLUSH_LOCAL; | |
1268 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1269 | ||
1270 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1271 | ||
1272 | preempt_enable(); | |
1273 | } | |
1274 | ||
1275 | static void xen_flush_tlb_single(unsigned long addr) | |
1276 | { | |
1277 | struct mmuext_op *op; | |
1278 | struct multicall_space mcs; | |
1279 | ||
1280 | preempt_disable(); | |
1281 | ||
1282 | mcs = xen_mc_entry(sizeof(*op)); | |
1283 | op = mcs.args; | |
1284 | op->cmd = MMUEXT_INVLPG_LOCAL; | |
1285 | op->arg1.linear_addr = addr & PAGE_MASK; | |
1286 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1287 | ||
1288 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1289 | ||
1290 | preempt_enable(); | |
1291 | } | |
1292 | ||
1293 | static void xen_flush_tlb_others(const struct cpumask *cpus, | |
1294 | struct mm_struct *mm, unsigned long va) | |
1295 | { | |
1296 | struct { | |
1297 | struct mmuext_op op; | |
1298 | DECLARE_BITMAP(mask, NR_CPUS); | |
1299 | } *args; | |
1300 | struct multicall_space mcs; | |
1301 | ||
e3f8a74e JF |
1302 | if (cpumask_empty(cpus)) |
1303 | return; /* nothing to do */ | |
319f3ba5 JF |
1304 | |
1305 | mcs = xen_mc_entry(sizeof(*args)); | |
1306 | args = mcs.args; | |
1307 | args->op.arg2.vcpumask = to_cpumask(args->mask); | |
1308 | ||
1309 | /* Remove us, and any offline CPUS. */ | |
1310 | cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask); | |
1311 | cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask)); | |
319f3ba5 JF |
1312 | |
1313 | if (va == TLB_FLUSH_ALL) { | |
1314 | args->op.cmd = MMUEXT_TLB_FLUSH_MULTI; | |
1315 | } else { | |
1316 | args->op.cmd = MMUEXT_INVLPG_MULTI; | |
1317 | args->op.arg1.linear_addr = va; | |
1318 | } | |
1319 | ||
1320 | MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF); | |
1321 | ||
319f3ba5 JF |
1322 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
1323 | } | |
1324 | ||
1325 | static unsigned long xen_read_cr3(void) | |
1326 | { | |
1327 | return percpu_read(xen_cr3); | |
1328 | } | |
1329 | ||
1330 | static void set_current_cr3(void *v) | |
1331 | { | |
1332 | percpu_write(xen_current_cr3, (unsigned long)v); | |
1333 | } | |
1334 | ||
1335 | static void __xen_write_cr3(bool kernel, unsigned long cr3) | |
1336 | { | |
1337 | struct mmuext_op *op; | |
1338 | struct multicall_space mcs; | |
1339 | unsigned long mfn; | |
1340 | ||
1341 | if (cr3) | |
1342 | mfn = pfn_to_mfn(PFN_DOWN(cr3)); | |
1343 | else | |
1344 | mfn = 0; | |
1345 | ||
1346 | WARN_ON(mfn == 0 && kernel); | |
1347 | ||
1348 | mcs = __xen_mc_entry(sizeof(*op)); | |
1349 | ||
1350 | op = mcs.args; | |
1351 | op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR; | |
1352 | op->arg1.mfn = mfn; | |
1353 | ||
1354 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1355 | ||
1356 | if (kernel) { | |
1357 | percpu_write(xen_cr3, cr3); | |
1358 | ||
1359 | /* Update xen_current_cr3 once the batch has actually | |
1360 | been submitted. */ | |
1361 | xen_mc_callback(set_current_cr3, (void *)cr3); | |
1362 | } | |
1363 | } | |
1364 | ||
1365 | static void xen_write_cr3(unsigned long cr3) | |
1366 | { | |
1367 | BUG_ON(preemptible()); | |
1368 | ||
1369 | xen_mc_batch(); /* disables interrupts */ | |
1370 | ||
1371 | /* Update while interrupts are disabled, so its atomic with | |
1372 | respect to ipis */ | |
1373 | percpu_write(xen_cr3, cr3); | |
1374 | ||
1375 | __xen_write_cr3(true, cr3); | |
1376 | ||
1377 | #ifdef CONFIG_X86_64 | |
1378 | { | |
1379 | pgd_t *user_pgd = xen_get_user_pgd(__va(cr3)); | |
1380 | if (user_pgd) | |
1381 | __xen_write_cr3(false, __pa(user_pgd)); | |
1382 | else | |
1383 | __xen_write_cr3(false, 0); | |
1384 | } | |
1385 | #endif | |
1386 | ||
1387 | xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ | |
1388 | } | |
1389 | ||
1390 | static int xen_pgd_alloc(struct mm_struct *mm) | |
1391 | { | |
1392 | pgd_t *pgd = mm->pgd; | |
1393 | int ret = 0; | |
1394 | ||
1395 | BUG_ON(PagePinned(virt_to_page(pgd))); | |
1396 | ||
1397 | #ifdef CONFIG_X86_64 | |
1398 | { | |
1399 | struct page *page = virt_to_page(pgd); | |
1400 | pgd_t *user_pgd; | |
1401 | ||
1402 | BUG_ON(page->private != 0); | |
1403 | ||
1404 | ret = -ENOMEM; | |
1405 | ||
1406 | user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO); | |
1407 | page->private = (unsigned long)user_pgd; | |
1408 | ||
1409 | if (user_pgd != NULL) { | |
1410 | user_pgd[pgd_index(VSYSCALL_START)] = | |
1411 | __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE); | |
1412 | ret = 0; | |
1413 | } | |
1414 | ||
1415 | BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd)))); | |
1416 | } | |
1417 | #endif | |
1418 | ||
1419 | return ret; | |
1420 | } | |
1421 | ||
1422 | static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd) | |
1423 | { | |
1424 | #ifdef CONFIG_X86_64 | |
1425 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
1426 | ||
1427 | if (user_pgd) | |
1428 | free_page((unsigned long)user_pgd); | |
1429 | #endif | |
1430 | } | |
1431 | ||
1f4f9315 JF |
1432 | #ifdef CONFIG_X86_32 |
1433 | static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte) | |
1434 | { | |
1435 | /* If there's an existing pte, then don't allow _PAGE_RW to be set */ | |
1436 | if (pte_val_ma(*ptep) & _PAGE_PRESENT) | |
1437 | pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) & | |
1438 | pte_val_ma(pte)); | |
1439 | ||
1440 | return pte; | |
1441 | } | |
1442 | ||
1443 | /* Init-time set_pte while constructing initial pagetables, which | |
1444 | doesn't allow RO pagetable pages to be remapped RW */ | |
1445 | static __init void xen_set_pte_init(pte_t *ptep, pte_t pte) | |
1446 | { | |
1447 | pte = mask_rw_pte(ptep, pte); | |
1448 | ||
1449 | xen_set_pte(ptep, pte); | |
1450 | } | |
1451 | #endif | |
319f3ba5 | 1452 | |
b96229b5 JF |
1453 | static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn) |
1454 | { | |
1455 | struct mmuext_op op; | |
1456 | op.cmd = cmd; | |
1457 | op.arg1.mfn = pfn_to_mfn(pfn); | |
1458 | if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) | |
1459 | BUG(); | |
1460 | } | |
1461 | ||
319f3ba5 JF |
1462 | /* Early in boot, while setting up the initial pagetable, assume |
1463 | everything is pinned. */ | |
1464 | static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn) | |
1465 | { | |
b96229b5 JF |
1466 | #ifdef CONFIG_FLATMEM |
1467 | BUG_ON(mem_map); /* should only be used early */ | |
1468 | #endif | |
1469 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1470 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
1471 | } | |
1472 | ||
1473 | /* Used for pmd and pud */ | |
1474 | static __init void xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn) | |
1475 | { | |
319f3ba5 JF |
1476 | #ifdef CONFIG_FLATMEM |
1477 | BUG_ON(mem_map); /* should only be used early */ | |
1478 | #endif | |
1479 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1480 | } | |
1481 | ||
1482 | /* Early release_pte assumes that all pts are pinned, since there's | |
1483 | only init_mm and anything attached to that is pinned. */ | |
b96229b5 | 1484 | static __init void xen_release_pte_init(unsigned long pfn) |
319f3ba5 | 1485 | { |
b96229b5 | 1486 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); |
319f3ba5 JF |
1487 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); |
1488 | } | |
1489 | ||
b96229b5 | 1490 | static __init void xen_release_pmd_init(unsigned long pfn) |
319f3ba5 | 1491 | { |
b96229b5 | 1492 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); |
319f3ba5 JF |
1493 | } |
1494 | ||
1495 | /* This needs to make sure the new pte page is pinned iff its being | |
1496 | attached to a pinned pagetable. */ | |
1497 | static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level) | |
1498 | { | |
1499 | struct page *page = pfn_to_page(pfn); | |
1500 | ||
1501 | if (PagePinned(virt_to_page(mm->pgd))) { | |
1502 | SetPagePinned(page); | |
1503 | ||
1504 | vm_unmap_aliases(); | |
1505 | if (!PageHighMem(page)) { | |
1506 | make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn))); | |
1507 | if (level == PT_PTE && USE_SPLIT_PTLOCKS) | |
1508 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
1509 | } else { | |
1510 | /* make sure there are no stray mappings of | |
1511 | this page */ | |
1512 | kmap_flush_unused(); | |
1513 | } | |
1514 | } | |
1515 | } | |
1516 | ||
1517 | static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn) | |
1518 | { | |
1519 | xen_alloc_ptpage(mm, pfn, PT_PTE); | |
1520 | } | |
1521 | ||
1522 | static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn) | |
1523 | { | |
1524 | xen_alloc_ptpage(mm, pfn, PT_PMD); | |
1525 | } | |
1526 | ||
1527 | /* This should never happen until we're OK to use struct page */ | |
1528 | static void xen_release_ptpage(unsigned long pfn, unsigned level) | |
1529 | { | |
1530 | struct page *page = pfn_to_page(pfn); | |
1531 | ||
1532 | if (PagePinned(page)) { | |
1533 | if (!PageHighMem(page)) { | |
1534 | if (level == PT_PTE && USE_SPLIT_PTLOCKS) | |
1535 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); | |
1536 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
1537 | } | |
1538 | ClearPagePinned(page); | |
1539 | } | |
1540 | } | |
1541 | ||
1542 | static void xen_release_pte(unsigned long pfn) | |
1543 | { | |
1544 | xen_release_ptpage(pfn, PT_PTE); | |
1545 | } | |
1546 | ||
1547 | static void xen_release_pmd(unsigned long pfn) | |
1548 | { | |
1549 | xen_release_ptpage(pfn, PT_PMD); | |
1550 | } | |
1551 | ||
1552 | #if PAGETABLE_LEVELS == 4 | |
1553 | static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn) | |
1554 | { | |
1555 | xen_alloc_ptpage(mm, pfn, PT_PUD); | |
1556 | } | |
1557 | ||
1558 | static void xen_release_pud(unsigned long pfn) | |
1559 | { | |
1560 | xen_release_ptpage(pfn, PT_PUD); | |
1561 | } | |
1562 | #endif | |
1563 | ||
1564 | void __init xen_reserve_top(void) | |
1565 | { | |
1566 | #ifdef CONFIG_X86_32 | |
1567 | unsigned long top = HYPERVISOR_VIRT_START; | |
1568 | struct xen_platform_parameters pp; | |
1569 | ||
1570 | if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0) | |
1571 | top = pp.virt_start; | |
1572 | ||
1573 | reserve_top_address(-top); | |
1574 | #endif /* CONFIG_X86_32 */ | |
1575 | } | |
1576 | ||
1577 | /* | |
1578 | * Like __va(), but returns address in the kernel mapping (which is | |
1579 | * all we have until the physical memory mapping has been set up. | |
1580 | */ | |
1581 | static void *__ka(phys_addr_t paddr) | |
1582 | { | |
1583 | #ifdef CONFIG_X86_64 | |
1584 | return (void *)(paddr + __START_KERNEL_map); | |
1585 | #else | |
1586 | return __va(paddr); | |
1587 | #endif | |
1588 | } | |
1589 | ||
1590 | /* Convert a machine address to physical address */ | |
1591 | static unsigned long m2p(phys_addr_t maddr) | |
1592 | { | |
1593 | phys_addr_t paddr; | |
1594 | ||
1595 | maddr &= PTE_PFN_MASK; | |
1596 | paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT; | |
1597 | ||
1598 | return paddr; | |
1599 | } | |
1600 | ||
1601 | /* Convert a machine address to kernel virtual */ | |
1602 | static void *m2v(phys_addr_t maddr) | |
1603 | { | |
1604 | return __ka(m2p(maddr)); | |
1605 | } | |
1606 | ||
1607 | static void set_page_prot(void *addr, pgprot_t prot) | |
1608 | { | |
1609 | unsigned long pfn = __pa(addr) >> PAGE_SHIFT; | |
1610 | pte_t pte = pfn_pte(pfn, prot); | |
1611 | ||
1612 | if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0)) | |
1613 | BUG(); | |
1614 | } | |
1615 | ||
1616 | static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn) | |
1617 | { | |
1618 | unsigned pmdidx, pteidx; | |
1619 | unsigned ident_pte; | |
1620 | unsigned long pfn; | |
1621 | ||
1622 | ident_pte = 0; | |
1623 | pfn = 0; | |
1624 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) { | |
1625 | pte_t *pte_page; | |
1626 | ||
1627 | /* Reuse or allocate a page of ptes */ | |
1628 | if (pmd_present(pmd[pmdidx])) | |
1629 | pte_page = m2v(pmd[pmdidx].pmd); | |
1630 | else { | |
1631 | /* Check for free pte pages */ | |
1632 | if (ident_pte == ARRAY_SIZE(level1_ident_pgt)) | |
1633 | break; | |
1634 | ||
1635 | pte_page = &level1_ident_pgt[ident_pte]; | |
1636 | ident_pte += PTRS_PER_PTE; | |
1637 | ||
1638 | pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE); | |
1639 | } | |
1640 | ||
1641 | /* Install mappings */ | |
1642 | for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) { | |
1643 | pte_t pte; | |
1644 | ||
1645 | if (pfn > max_pfn_mapped) | |
1646 | max_pfn_mapped = pfn; | |
1647 | ||
1648 | if (!pte_none(pte_page[pteidx])) | |
1649 | continue; | |
1650 | ||
1651 | pte = pfn_pte(pfn, PAGE_KERNEL_EXEC); | |
1652 | pte_page[pteidx] = pte; | |
1653 | } | |
1654 | } | |
1655 | ||
1656 | for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE) | |
1657 | set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO); | |
1658 | ||
1659 | set_page_prot(pmd, PAGE_KERNEL_RO); | |
1660 | } | |
1661 | ||
1662 | #ifdef CONFIG_X86_64 | |
1663 | static void convert_pfn_mfn(void *v) | |
1664 | { | |
1665 | pte_t *pte = v; | |
1666 | int i; | |
1667 | ||
1668 | /* All levels are converted the same way, so just treat them | |
1669 | as ptes. */ | |
1670 | for (i = 0; i < PTRS_PER_PTE; i++) | |
1671 | pte[i] = xen_make_pte(pte[i].pte); | |
1672 | } | |
1673 | ||
1674 | /* | |
1675 | * Set up the inital kernel pagetable. | |
1676 | * | |
1677 | * We can construct this by grafting the Xen provided pagetable into | |
1678 | * head_64.S's preconstructed pagetables. We copy the Xen L2's into | |
1679 | * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This | |
1680 | * means that only the kernel has a physical mapping to start with - | |
1681 | * but that's enough to get __va working. We need to fill in the rest | |
1682 | * of the physical mapping once some sort of allocator has been set | |
1683 | * up. | |
1684 | */ | |
1685 | __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, | |
1686 | unsigned long max_pfn) | |
1687 | { | |
1688 | pud_t *l3; | |
1689 | pmd_t *l2; | |
1690 | ||
1691 | /* Zap identity mapping */ | |
1692 | init_level4_pgt[0] = __pgd(0); | |
1693 | ||
1694 | /* Pre-constructed entries are in pfn, so convert to mfn */ | |
1695 | convert_pfn_mfn(init_level4_pgt); | |
1696 | convert_pfn_mfn(level3_ident_pgt); | |
1697 | convert_pfn_mfn(level3_kernel_pgt); | |
1698 | ||
1699 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd); | |
1700 | l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud); | |
1701 | ||
1702 | memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1703 | memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1704 | ||
1705 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd); | |
1706 | l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud); | |
1707 | memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1708 | ||
1709 | /* Set up identity map */ | |
1710 | xen_map_identity_early(level2_ident_pgt, max_pfn); | |
1711 | ||
1712 | /* Make pagetable pieces RO */ | |
1713 | set_page_prot(init_level4_pgt, PAGE_KERNEL_RO); | |
1714 | set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO); | |
1715 | set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO); | |
1716 | set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO); | |
1717 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); | |
1718 | set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO); | |
1719 | ||
1720 | /* Pin down new L4 */ | |
1721 | pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE, | |
1722 | PFN_DOWN(__pa_symbol(init_level4_pgt))); | |
1723 | ||
1724 | /* Unpin Xen-provided one */ | |
1725 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
1726 | ||
1727 | /* Switch over */ | |
1728 | pgd = init_level4_pgt; | |
1729 | ||
1730 | /* | |
1731 | * At this stage there can be no user pgd, and no page | |
1732 | * structure to attach it to, so make sure we just set kernel | |
1733 | * pgd. | |
1734 | */ | |
1735 | xen_mc_batch(); | |
1736 | __xen_write_cr3(true, __pa(pgd)); | |
1737 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
1738 | ||
1739 | reserve_early(__pa(xen_start_info->pt_base), | |
1740 | __pa(xen_start_info->pt_base + | |
1741 | xen_start_info->nr_pt_frames * PAGE_SIZE), | |
1742 | "XEN PAGETABLES"); | |
1743 | ||
1744 | return pgd; | |
1745 | } | |
1746 | #else /* !CONFIG_X86_64 */ | |
1747 | static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss; | |
1748 | ||
1749 | __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, | |
1750 | unsigned long max_pfn) | |
1751 | { | |
1752 | pmd_t *kernel_pmd; | |
1753 | ||
93dbda7c JF |
1754 | max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) + |
1755 | xen_start_info->nr_pt_frames * PAGE_SIZE + | |
1756 | 512*1024); | |
319f3ba5 JF |
1757 | |
1758 | kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd); | |
1759 | memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD); | |
1760 | ||
1761 | xen_map_identity_early(level2_kernel_pgt, max_pfn); | |
1762 | ||
1763 | memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD); | |
1764 | set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY], | |
1765 | __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT)); | |
1766 | ||
1767 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); | |
1768 | set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO); | |
1769 | set_page_prot(empty_zero_page, PAGE_KERNEL_RO); | |
1770 | ||
1771 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
1772 | ||
1773 | xen_write_cr3(__pa(swapper_pg_dir)); | |
1774 | ||
1775 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir))); | |
1776 | ||
33df4db0 JF |
1777 | reserve_early(__pa(xen_start_info->pt_base), |
1778 | __pa(xen_start_info->pt_base + | |
1779 | xen_start_info->nr_pt_frames * PAGE_SIZE), | |
1780 | "XEN PAGETABLES"); | |
1781 | ||
319f3ba5 JF |
1782 | return swapper_pg_dir; |
1783 | } | |
1784 | #endif /* CONFIG_X86_64 */ | |
1785 | ||
3b3809ac | 1786 | static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot) |
319f3ba5 JF |
1787 | { |
1788 | pte_t pte; | |
1789 | ||
1790 | phys >>= PAGE_SHIFT; | |
1791 | ||
1792 | switch (idx) { | |
1793 | case FIX_BTMAP_END ... FIX_BTMAP_BEGIN: | |
1794 | #ifdef CONFIG_X86_F00F_BUG | |
1795 | case FIX_F00F_IDT: | |
1796 | #endif | |
1797 | #ifdef CONFIG_X86_32 | |
1798 | case FIX_WP_TEST: | |
1799 | case FIX_VDSO: | |
1800 | # ifdef CONFIG_HIGHMEM | |
1801 | case FIX_KMAP_BEGIN ... FIX_KMAP_END: | |
1802 | # endif | |
1803 | #else | |
1804 | case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE: | |
1805 | #endif | |
1806 | #ifdef CONFIG_X86_LOCAL_APIC | |
1807 | case FIX_APIC_BASE: /* maps dummy local APIC */ | |
1808 | #endif | |
3ecb1b7d JF |
1809 | case FIX_TEXT_POKE0: |
1810 | case FIX_TEXT_POKE1: | |
1811 | /* All local page mappings */ | |
319f3ba5 JF |
1812 | pte = pfn_pte(phys, prot); |
1813 | break; | |
1814 | ||
1815 | default: | |
1816 | pte = mfn_pte(phys, prot); | |
1817 | break; | |
1818 | } | |
1819 | ||
1820 | __native_set_fixmap(idx, pte); | |
1821 | ||
1822 | #ifdef CONFIG_X86_64 | |
1823 | /* Replicate changes to map the vsyscall page into the user | |
1824 | pagetable vsyscall mapping. */ | |
1825 | if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) { | |
1826 | unsigned long vaddr = __fix_to_virt(idx); | |
1827 | set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte); | |
1828 | } | |
1829 | #endif | |
1830 | } | |
1831 | ||
f1d7062a | 1832 | static __init void xen_post_allocator_init(void) |
319f3ba5 JF |
1833 | { |
1834 | pv_mmu_ops.set_pte = xen_set_pte; | |
1835 | pv_mmu_ops.set_pmd = xen_set_pmd; | |
1836 | pv_mmu_ops.set_pud = xen_set_pud; | |
1837 | #if PAGETABLE_LEVELS == 4 | |
1838 | pv_mmu_ops.set_pgd = xen_set_pgd; | |
1839 | #endif | |
1840 | ||
1841 | /* This will work as long as patching hasn't happened yet | |
1842 | (which it hasn't) */ | |
1843 | pv_mmu_ops.alloc_pte = xen_alloc_pte; | |
1844 | pv_mmu_ops.alloc_pmd = xen_alloc_pmd; | |
1845 | pv_mmu_ops.release_pte = xen_release_pte; | |
1846 | pv_mmu_ops.release_pmd = xen_release_pmd; | |
1847 | #if PAGETABLE_LEVELS == 4 | |
1848 | pv_mmu_ops.alloc_pud = xen_alloc_pud; | |
1849 | pv_mmu_ops.release_pud = xen_release_pud; | |
1850 | #endif | |
1851 | ||
1852 | #ifdef CONFIG_X86_64 | |
1853 | SetPagePinned(virt_to_page(level3_user_vsyscall)); | |
1854 | #endif | |
1855 | xen_mark_init_mm_pinned(); | |
1856 | } | |
1857 | ||
b407fc57 JF |
1858 | static void xen_leave_lazy_mmu(void) |
1859 | { | |
5caecb94 | 1860 | preempt_disable(); |
b407fc57 JF |
1861 | xen_mc_flush(); |
1862 | paravirt_leave_lazy_mmu(); | |
5caecb94 | 1863 | preempt_enable(); |
b407fc57 | 1864 | } |
319f3ba5 | 1865 | |
030cb6c0 | 1866 | static const struct pv_mmu_ops xen_mmu_ops __initdata = { |
319f3ba5 JF |
1867 | .read_cr2 = xen_read_cr2, |
1868 | .write_cr2 = xen_write_cr2, | |
1869 | ||
1870 | .read_cr3 = xen_read_cr3, | |
1871 | .write_cr3 = xen_write_cr3, | |
1872 | ||
1873 | .flush_tlb_user = xen_flush_tlb, | |
1874 | .flush_tlb_kernel = xen_flush_tlb, | |
1875 | .flush_tlb_single = xen_flush_tlb_single, | |
1876 | .flush_tlb_others = xen_flush_tlb_others, | |
1877 | ||
1878 | .pte_update = paravirt_nop, | |
1879 | .pte_update_defer = paravirt_nop, | |
1880 | ||
1881 | .pgd_alloc = xen_pgd_alloc, | |
1882 | .pgd_free = xen_pgd_free, | |
1883 | ||
1884 | .alloc_pte = xen_alloc_pte_init, | |
1885 | .release_pte = xen_release_pte_init, | |
b96229b5 | 1886 | .alloc_pmd = xen_alloc_pmd_init, |
319f3ba5 | 1887 | .alloc_pmd_clone = paravirt_nop, |
b96229b5 | 1888 | .release_pmd = xen_release_pmd_init, |
319f3ba5 | 1889 | |
319f3ba5 JF |
1890 | #ifdef CONFIG_X86_64 |
1891 | .set_pte = xen_set_pte, | |
1892 | #else | |
1893 | .set_pte = xen_set_pte_init, | |
1894 | #endif | |
1895 | .set_pte_at = xen_set_pte_at, | |
1896 | .set_pmd = xen_set_pmd_hyper, | |
1897 | ||
1898 | .ptep_modify_prot_start = __ptep_modify_prot_start, | |
1899 | .ptep_modify_prot_commit = __ptep_modify_prot_commit, | |
1900 | ||
da5de7c2 JF |
1901 | .pte_val = PV_CALLEE_SAVE(xen_pte_val), |
1902 | .pgd_val = PV_CALLEE_SAVE(xen_pgd_val), | |
319f3ba5 | 1903 | |
da5de7c2 JF |
1904 | .make_pte = PV_CALLEE_SAVE(xen_make_pte), |
1905 | .make_pgd = PV_CALLEE_SAVE(xen_make_pgd), | |
319f3ba5 JF |
1906 | |
1907 | #ifdef CONFIG_X86_PAE | |
1908 | .set_pte_atomic = xen_set_pte_atomic, | |
319f3ba5 JF |
1909 | .pte_clear = xen_pte_clear, |
1910 | .pmd_clear = xen_pmd_clear, | |
1911 | #endif /* CONFIG_X86_PAE */ | |
1912 | .set_pud = xen_set_pud_hyper, | |
1913 | ||
da5de7c2 JF |
1914 | .make_pmd = PV_CALLEE_SAVE(xen_make_pmd), |
1915 | .pmd_val = PV_CALLEE_SAVE(xen_pmd_val), | |
319f3ba5 JF |
1916 | |
1917 | #if PAGETABLE_LEVELS == 4 | |
da5de7c2 JF |
1918 | .pud_val = PV_CALLEE_SAVE(xen_pud_val), |
1919 | .make_pud = PV_CALLEE_SAVE(xen_make_pud), | |
319f3ba5 JF |
1920 | .set_pgd = xen_set_pgd_hyper, |
1921 | ||
b96229b5 JF |
1922 | .alloc_pud = xen_alloc_pmd_init, |
1923 | .release_pud = xen_release_pmd_init, | |
319f3ba5 JF |
1924 | #endif /* PAGETABLE_LEVELS == 4 */ |
1925 | ||
1926 | .activate_mm = xen_activate_mm, | |
1927 | .dup_mmap = xen_dup_mmap, | |
1928 | .exit_mmap = xen_exit_mmap, | |
1929 | ||
1930 | .lazy_mode = { | |
1931 | .enter = paravirt_enter_lazy_mmu, | |
b407fc57 | 1932 | .leave = xen_leave_lazy_mmu, |
319f3ba5 JF |
1933 | }, |
1934 | ||
1935 | .set_fixmap = xen_set_fixmap, | |
1936 | }; | |
1937 | ||
030cb6c0 TG |
1938 | void __init xen_init_mmu_ops(void) |
1939 | { | |
1940 | x86_init.paging.pagetable_setup_start = xen_pagetable_setup_start; | |
1941 | x86_init.paging.pagetable_setup_done = xen_pagetable_setup_done; | |
1942 | pv_mmu_ops = xen_mmu_ops; | |
1943 | } | |
319f3ba5 | 1944 | |
59151001 SS |
1945 | static void xen_hvm_exit_mmap(struct mm_struct *mm) |
1946 | { | |
1947 | struct xen_hvm_pagetable_dying a; | |
1948 | int rc; | |
1949 | ||
1950 | a.domid = DOMID_SELF; | |
1951 | a.gpa = __pa(mm->pgd); | |
1952 | rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a); | |
1953 | WARN_ON_ONCE(rc < 0); | |
1954 | } | |
1955 | ||
1956 | static int is_pagetable_dying_supported(void) | |
1957 | { | |
1958 | struct xen_hvm_pagetable_dying a; | |
1959 | int rc = 0; | |
1960 | ||
1961 | a.domid = DOMID_SELF; | |
1962 | a.gpa = 0x00; | |
1963 | rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a); | |
1964 | if (rc < 0) { | |
1965 | printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n"); | |
1966 | return 0; | |
1967 | } | |
1968 | return 1; | |
1969 | } | |
1970 | ||
1971 | void __init xen_hvm_init_mmu_ops(void) | |
1972 | { | |
1973 | if (is_pagetable_dying_supported()) | |
1974 | pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap; | |
1975 | } | |
1976 | ||
994025ca JF |
1977 | #ifdef CONFIG_XEN_DEBUG_FS |
1978 | ||
1979 | static struct dentry *d_mmu_debug; | |
1980 | ||
1981 | static int __init xen_mmu_debugfs(void) | |
1982 | { | |
1983 | struct dentry *d_xen = xen_init_debugfs(); | |
1984 | ||
1985 | if (d_xen == NULL) | |
1986 | return -ENOMEM; | |
1987 | ||
1988 | d_mmu_debug = debugfs_create_dir("mmu", d_xen); | |
1989 | ||
1990 | debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats); | |
1991 | ||
1992 | debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update); | |
1993 | debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug, | |
1994 | &mmu_stats.pgd_update_pinned); | |
1995 | debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug, | |
1996 | &mmu_stats.pgd_update_pinned); | |
1997 | ||
1998 | debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update); | |
1999 | debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug, | |
2000 | &mmu_stats.pud_update_pinned); | |
2001 | debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug, | |
2002 | &mmu_stats.pud_update_pinned); | |
2003 | ||
2004 | debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update); | |
2005 | debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug, | |
2006 | &mmu_stats.pmd_update_pinned); | |
2007 | debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug, | |
2008 | &mmu_stats.pmd_update_pinned); | |
2009 | ||
2010 | debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update); | |
2011 | // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug, | |
2012 | // &mmu_stats.pte_update_pinned); | |
2013 | debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug, | |
2014 | &mmu_stats.pte_update_pinned); | |
2015 | ||
2016 | debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update); | |
2017 | debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug, | |
2018 | &mmu_stats.mmu_update_extended); | |
2019 | xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug, | |
2020 | mmu_stats.mmu_update_histo, 20); | |
2021 | ||
2022 | debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at); | |
2023 | debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug, | |
2024 | &mmu_stats.set_pte_at_batched); | |
2025 | debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug, | |
2026 | &mmu_stats.set_pte_at_current); | |
2027 | debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug, | |
2028 | &mmu_stats.set_pte_at_kernel); | |
2029 | ||
2030 | debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit); | |
2031 | debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug, | |
2032 | &mmu_stats.prot_commit_batched); | |
2033 | ||
2034 | return 0; | |
2035 | } | |
2036 | fs_initcall(xen_mmu_debugfs); | |
2037 | ||
2038 | #endif /* CONFIG_XEN_DEBUG_FS */ |