]>
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> |
d2cb2145 | 45 | #include <linux/vmalloc.h> |
44408ad7 | 46 | #include <linux/module.h> |
5a0e3ad6 | 47 | #include <linux/gfp.h> |
3b827c1b JF |
48 | |
49 | #include <asm/pgtable.h> | |
50 | #include <asm/tlbflush.h> | |
5deb30d1 | 51 | #include <asm/fixmap.h> |
3b827c1b | 52 | #include <asm/mmu_context.h> |
319f3ba5 | 53 | #include <asm/setup.h> |
f4f97b3e | 54 | #include <asm/paravirt.h> |
7347b408 | 55 | #include <asm/e820.h> |
cbcd79c2 | 56 | #include <asm/linkage.h> |
08bbc9da | 57 | #include <asm/page.h> |
3b827c1b JF |
58 | |
59 | #include <asm/xen/hypercall.h> | |
f4f97b3e | 60 | #include <asm/xen/hypervisor.h> |
3b827c1b | 61 | |
c0011dbf | 62 | #include <xen/xen.h> |
3b827c1b JF |
63 | #include <xen/page.h> |
64 | #include <xen/interface/xen.h> | |
59151001 | 65 | #include <xen/interface/hvm/hvm_op.h> |
319f3ba5 | 66 | #include <xen/interface/version.h> |
c0011dbf | 67 | #include <xen/interface/memory.h> |
319f3ba5 | 68 | #include <xen/hvc-console.h> |
3b827c1b | 69 | |
f4f97b3e | 70 | #include "multicalls.h" |
3b827c1b | 71 | #include "mmu.h" |
994025ca JF |
72 | #include "debugfs.h" |
73 | ||
74 | #define MMU_UPDATE_HISTO 30 | |
75 | ||
19001c8c AN |
76 | /* |
77 | * Protects atomic reservation decrease/increase against concurrent increases. | |
78 | * Also protects non-atomic updates of current_pages and driver_pages, and | |
79 | * balloon lists. | |
80 | */ | |
81 | DEFINE_SPINLOCK(xen_reservation_lock); | |
82 | ||
994025ca JF |
83 | #ifdef CONFIG_XEN_DEBUG_FS |
84 | ||
85 | static struct { | |
86 | u32 pgd_update; | |
87 | u32 pgd_update_pinned; | |
88 | u32 pgd_update_batched; | |
89 | ||
90 | u32 pud_update; | |
91 | u32 pud_update_pinned; | |
92 | u32 pud_update_batched; | |
93 | ||
94 | u32 pmd_update; | |
95 | u32 pmd_update_pinned; | |
96 | u32 pmd_update_batched; | |
97 | ||
98 | u32 pte_update; | |
99 | u32 pte_update_pinned; | |
100 | u32 pte_update_batched; | |
101 | ||
102 | u32 mmu_update; | |
103 | u32 mmu_update_extended; | |
104 | u32 mmu_update_histo[MMU_UPDATE_HISTO]; | |
105 | ||
106 | u32 prot_commit; | |
107 | u32 prot_commit_batched; | |
108 | ||
109 | u32 set_pte_at; | |
110 | u32 set_pte_at_batched; | |
111 | u32 set_pte_at_pinned; | |
112 | u32 set_pte_at_current; | |
113 | u32 set_pte_at_kernel; | |
114 | } mmu_stats; | |
115 | ||
116 | static u8 zero_stats; | |
117 | ||
118 | static inline void check_zero(void) | |
119 | { | |
120 | if (unlikely(zero_stats)) { | |
121 | memset(&mmu_stats, 0, sizeof(mmu_stats)); | |
122 | zero_stats = 0; | |
123 | } | |
124 | } | |
125 | ||
126 | #define ADD_STATS(elem, val) \ | |
127 | do { check_zero(); mmu_stats.elem += (val); } while(0) | |
128 | ||
129 | #else /* !CONFIG_XEN_DEBUG_FS */ | |
130 | ||
131 | #define ADD_STATS(elem, val) do { (void)(val); } while(0) | |
132 | ||
133 | #endif /* CONFIG_XEN_DEBUG_FS */ | |
3b827c1b | 134 | |
319f3ba5 JF |
135 | |
136 | /* | |
137 | * Identity map, in addition to plain kernel map. This needs to be | |
138 | * large enough to allocate page table pages to allocate the rest. | |
139 | * Each page can map 2MB. | |
140 | */ | |
141 | static pte_t level1_ident_pgt[PTRS_PER_PTE * 4] __page_aligned_bss; | |
142 | ||
143 | #ifdef CONFIG_X86_64 | |
144 | /* l3 pud for userspace vsyscall mapping */ | |
145 | static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss; | |
146 | #endif /* CONFIG_X86_64 */ | |
147 | ||
148 | /* | |
149 | * Note about cr3 (pagetable base) values: | |
150 | * | |
151 | * xen_cr3 contains the current logical cr3 value; it contains the | |
152 | * last set cr3. This may not be the current effective cr3, because | |
153 | * its update may be being lazily deferred. However, a vcpu looking | |
154 | * at its own cr3 can use this value knowing that it everything will | |
155 | * be self-consistent. | |
156 | * | |
157 | * xen_current_cr3 contains the actual vcpu cr3; it is set once the | |
158 | * hypercall to set the vcpu cr3 is complete (so it may be a little | |
159 | * out of date, but it will never be set early). If one vcpu is | |
160 | * looking at another vcpu's cr3 value, it should use this variable. | |
161 | */ | |
162 | DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */ | |
163 | DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */ | |
164 | ||
165 | ||
d6182fbf JF |
166 | /* |
167 | * Just beyond the highest usermode address. STACK_TOP_MAX has a | |
168 | * redzone above it, so round it up to a PGD boundary. | |
169 | */ | |
170 | #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK) | |
171 | ||
a2e87529 | 172 | static unsigned long max_p2m_pfn __read_mostly = MAX_DOMAIN_PAGES; |
d6182fbf | 173 | |
a2e87529 JF |
174 | #define P2M_ENTRIES_PER_PAGE (PAGE_SIZE / sizeof(unsigned long)) |
175 | #define TOP_ENTRIES(pages) ((pages) / P2M_ENTRIES_PER_PAGE) | |
176 | #define MAX_TOP_ENTRIES TOP_ENTRIES(MAX_DOMAIN_PAGES) | |
d451bb7a | 177 | |
cf0923ea | 178 | /* Placeholder for holes in the address space */ |
a171ce6e | 179 | static RESERVE_BRK_ARRAY(unsigned long, p2m_missing, P2M_ENTRIES_PER_PAGE); |
cf0923ea JF |
180 | |
181 | /* Array of pointers to pages containing p2m entries */ | |
a2e87529 | 182 | static RESERVE_BRK_ARRAY(unsigned long *, p2m_top, MAX_TOP_ENTRIES); |
d451bb7a | 183 | |
d5edbc1f | 184 | /* Arrays of p2m arrays expressed in mfns used for save/restore */ |
a2e87529 | 185 | static RESERVE_BRK_ARRAY(unsigned long, p2m_top_mfn, MAX_TOP_ENTRIES); |
d5edbc1f | 186 | |
a171ce6e | 187 | static RESERVE_BRK_ARRAY(unsigned long, p2m_top_mfn_list, |
a2e87529 | 188 | (MAX_TOP_ENTRIES / P2M_ENTRIES_PER_PAGE)); |
d5edbc1f | 189 | |
d451bb7a JF |
190 | static inline unsigned p2m_top_index(unsigned long pfn) |
191 | { | |
a2e87529 | 192 | BUG_ON(pfn >= max_p2m_pfn); |
d451bb7a JF |
193 | return pfn / P2M_ENTRIES_PER_PAGE; |
194 | } | |
195 | ||
196 | static inline unsigned p2m_index(unsigned long pfn) | |
197 | { | |
198 | return pfn % P2M_ENTRIES_PER_PAGE; | |
199 | } | |
200 | ||
d5edbc1f | 201 | /* Build the parallel p2m_top_mfn structures */ |
fa24ba62 | 202 | void xen_build_mfn_list_list(void) |
d5edbc1f JF |
203 | { |
204 | unsigned pfn, idx; | |
205 | ||
a2e87529 | 206 | for (pfn = 0; pfn < max_p2m_pfn; pfn += P2M_ENTRIES_PER_PAGE) { |
d5edbc1f JF |
207 | unsigned topidx = p2m_top_index(pfn); |
208 | ||
209 | p2m_top_mfn[topidx] = virt_to_mfn(p2m_top[topidx]); | |
210 | } | |
211 | ||
a2e87529 JF |
212 | for (idx = 0; |
213 | idx < TOP_ENTRIES(max_p2m_pfn)/P2M_ENTRIES_PER_PAGE; | |
214 | idx++) { | |
d5edbc1f JF |
215 | unsigned topidx = idx * P2M_ENTRIES_PER_PAGE; |
216 | p2m_top_mfn_list[idx] = virt_to_mfn(&p2m_top_mfn[topidx]); | |
217 | } | |
cdaead6b | 218 | } |
d5edbc1f | 219 | |
cdaead6b JF |
220 | void xen_setup_mfn_list_list(void) |
221 | { | |
d5edbc1f JF |
222 | BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info); |
223 | ||
224 | HYPERVISOR_shared_info->arch.pfn_to_mfn_frame_list_list = | |
225 | virt_to_mfn(p2m_top_mfn_list); | |
226 | HYPERVISOR_shared_info->arch.max_pfn = xen_start_info->nr_pages; | |
227 | } | |
228 | ||
229 | /* Set up p2m_top to point to the domain-builder provided p2m pages */ | |
d451bb7a JF |
230 | void __init xen_build_dynamic_phys_to_machine(void) |
231 | { | |
d451bb7a | 232 | unsigned long *mfn_list = (unsigned long *)xen_start_info->mfn_list; |
8006ec3e | 233 | unsigned long max_pfn = min(MAX_DOMAIN_PAGES, xen_start_info->nr_pages); |
d5edbc1f | 234 | unsigned pfn; |
a171ce6e JF |
235 | unsigned i; |
236 | ||
a2e87529 JF |
237 | max_p2m_pfn = max_pfn; |
238 | ||
a171ce6e JF |
239 | p2m_missing = extend_brk(sizeof(*p2m_missing) * P2M_ENTRIES_PER_PAGE, |
240 | PAGE_SIZE); | |
241 | for (i = 0; i < P2M_ENTRIES_PER_PAGE; i++) | |
242 | p2m_missing[i] = ~0UL; | |
243 | ||
a2e87529 | 244 | p2m_top = extend_brk(sizeof(*p2m_top) * TOP_ENTRIES(max_pfn), |
a171ce6e | 245 | PAGE_SIZE); |
a2e87529 | 246 | for (i = 0; i < TOP_ENTRIES(max_pfn); i++) |
a171ce6e JF |
247 | p2m_top[i] = p2m_missing; |
248 | ||
a2e87529 JF |
249 | p2m_top_mfn = extend_brk(sizeof(*p2m_top_mfn) * TOP_ENTRIES(max_pfn), |
250 | PAGE_SIZE); | |
a171ce6e | 251 | p2m_top_mfn_list = extend_brk(sizeof(*p2m_top_mfn_list) * |
a2e87529 | 252 | (TOP_ENTRIES(max_pfn) / P2M_ENTRIES_PER_PAGE), |
a171ce6e | 253 | PAGE_SIZE); |
d451bb7a | 254 | |
f63c2f24 | 255 | for (pfn = 0; pfn < max_pfn; pfn += P2M_ENTRIES_PER_PAGE) { |
d451bb7a JF |
256 | unsigned topidx = p2m_top_index(pfn); |
257 | ||
258 | p2m_top[topidx] = &mfn_list[pfn]; | |
259 | } | |
cdaead6b JF |
260 | |
261 | xen_build_mfn_list_list(); | |
d451bb7a JF |
262 | } |
263 | ||
264 | unsigned long get_phys_to_machine(unsigned long pfn) | |
265 | { | |
266 | unsigned topidx, idx; | |
267 | ||
a2e87529 | 268 | if (unlikely(pfn >= max_p2m_pfn)) |
8006ec3e JF |
269 | return INVALID_P2M_ENTRY; |
270 | ||
d451bb7a | 271 | topidx = p2m_top_index(pfn); |
d451bb7a JF |
272 | idx = p2m_index(pfn); |
273 | return p2m_top[topidx][idx]; | |
274 | } | |
15ce6005 | 275 | EXPORT_SYMBOL_GPL(get_phys_to_machine); |
d451bb7a | 276 | |
e791ca0f JF |
277 | /* install a new p2m_top page */ |
278 | bool install_p2mtop_page(unsigned long pfn, unsigned long *p) | |
d451bb7a | 279 | { |
e791ca0f JF |
280 | unsigned topidx = p2m_top_index(pfn); |
281 | unsigned long **pfnp, *mfnp; | |
d451bb7a JF |
282 | unsigned i; |
283 | ||
e791ca0f JF |
284 | pfnp = &p2m_top[topidx]; |
285 | mfnp = &p2m_top_mfn[topidx]; | |
d451bb7a | 286 | |
f63c2f24 | 287 | for (i = 0; i < P2M_ENTRIES_PER_PAGE; i++) |
d451bb7a JF |
288 | p[i] = INVALID_P2M_ENTRY; |
289 | ||
e791ca0f | 290 | if (cmpxchg(pfnp, p2m_missing, p) == p2m_missing) { |
d5edbc1f | 291 | *mfnp = virt_to_mfn(p); |
e791ca0f JF |
292 | return true; |
293 | } | |
294 | ||
295 | return false; | |
d451bb7a JF |
296 | } |
297 | ||
e791ca0f | 298 | static void alloc_p2m(unsigned long pfn) |
d451bb7a | 299 | { |
e791ca0f | 300 | unsigned long *p; |
d451bb7a | 301 | |
e791ca0f JF |
302 | p = (void *)__get_free_page(GFP_KERNEL | __GFP_NOFAIL); |
303 | BUG_ON(p == NULL); | |
304 | ||
305 | if (!install_p2mtop_page(pfn, p)) | |
306 | free_page((unsigned long)p); | |
307 | } | |
308 | ||
309 | /* Try to install p2m mapping; fail if intermediate bits missing */ | |
310 | bool __set_phys_to_machine(unsigned long pfn, unsigned long mfn) | |
311 | { | |
312 | unsigned topidx, idx; | |
8006ec3e | 313 | |
a2e87529 | 314 | if (unlikely(pfn >= max_p2m_pfn)) { |
8006ec3e | 315 | BUG_ON(mfn != INVALID_P2M_ENTRY); |
e791ca0f | 316 | return true; |
d451bb7a JF |
317 | } |
318 | ||
319 | topidx = p2m_top_index(pfn); | |
cf0923ea | 320 | if (p2m_top[topidx] == p2m_missing) { |
d451bb7a | 321 | if (mfn == INVALID_P2M_ENTRY) |
e791ca0f JF |
322 | return true; |
323 | return false; | |
d451bb7a JF |
324 | } |
325 | ||
326 | idx = p2m_index(pfn); | |
327 | p2m_top[topidx][idx] = mfn; | |
e791ca0f JF |
328 | |
329 | return true; | |
330 | } | |
331 | ||
332 | void set_phys_to_machine(unsigned long pfn, unsigned long mfn) | |
333 | { | |
334 | if (unlikely(xen_feature(XENFEAT_auto_translated_physmap))) { | |
335 | BUG_ON(pfn != mfn && mfn != INVALID_P2M_ENTRY); | |
336 | return; | |
337 | } | |
338 | ||
339 | if (unlikely(!__set_phys_to_machine(pfn, mfn))) { | |
340 | alloc_p2m(pfn); | |
341 | ||
342 | if (!__set_phys_to_machine(pfn, mfn)) | |
343 | BUG(); | |
344 | } | |
d451bb7a JF |
345 | } |
346 | ||
9976b39b JF |
347 | unsigned long arbitrary_virt_to_mfn(void *vaddr) |
348 | { | |
349 | xmaddr_t maddr = arbitrary_virt_to_machine(vaddr); | |
350 | ||
351 | return PFN_DOWN(maddr.maddr); | |
352 | } | |
353 | ||
ce803e70 | 354 | xmaddr_t arbitrary_virt_to_machine(void *vaddr) |
3b827c1b | 355 | { |
ce803e70 | 356 | unsigned long address = (unsigned long)vaddr; |
da7bfc50 | 357 | unsigned int level; |
9f32d21c CL |
358 | pte_t *pte; |
359 | unsigned offset; | |
3b827c1b | 360 | |
9f32d21c CL |
361 | /* |
362 | * if the PFN is in the linear mapped vaddr range, we can just use | |
363 | * the (quick) virt_to_machine() p2m lookup | |
364 | */ | |
365 | if (virt_addr_valid(vaddr)) | |
366 | return virt_to_machine(vaddr); | |
367 | ||
368 | /* otherwise we have to do a (slower) full page-table walk */ | |
3b827c1b | 369 | |
9f32d21c CL |
370 | pte = lookup_address(address, &level); |
371 | BUG_ON(pte == NULL); | |
372 | offset = address & ~PAGE_MASK; | |
ebd879e3 | 373 | return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset); |
3b827c1b JF |
374 | } |
375 | ||
376 | void make_lowmem_page_readonly(void *vaddr) | |
377 | { | |
378 | pte_t *pte, ptev; | |
379 | unsigned long address = (unsigned long)vaddr; | |
da7bfc50 | 380 | unsigned int level; |
3b827c1b | 381 | |
f0646e43 | 382 | pte = lookup_address(address, &level); |
3b827c1b JF |
383 | BUG_ON(pte == NULL); |
384 | ||
385 | ptev = pte_wrprotect(*pte); | |
386 | ||
387 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
388 | BUG(); | |
389 | } | |
390 | ||
391 | void make_lowmem_page_readwrite(void *vaddr) | |
392 | { | |
393 | pte_t *pte, ptev; | |
394 | unsigned long address = (unsigned long)vaddr; | |
da7bfc50 | 395 | unsigned int level; |
3b827c1b | 396 | |
f0646e43 | 397 | pte = lookup_address(address, &level); |
3b827c1b JF |
398 | BUG_ON(pte == NULL); |
399 | ||
400 | ptev = pte_mkwrite(*pte); | |
401 | ||
402 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
403 | BUG(); | |
404 | } | |
405 | ||
406 | ||
7708ad64 | 407 | static bool xen_page_pinned(void *ptr) |
e2426cf8 JF |
408 | { |
409 | struct page *page = virt_to_page(ptr); | |
410 | ||
411 | return PagePinned(page); | |
412 | } | |
413 | ||
c0011dbf JF |
414 | static bool xen_iomap_pte(pte_t pte) |
415 | { | |
7347b408 | 416 | return pte_flags(pte) & _PAGE_IOMAP; |
c0011dbf JF |
417 | } |
418 | ||
419 | static void xen_set_iomap_pte(pte_t *ptep, pte_t pteval) | |
420 | { | |
421 | struct multicall_space mcs; | |
422 | struct mmu_update *u; | |
423 | ||
424 | mcs = xen_mc_entry(sizeof(*u)); | |
425 | u = mcs.args; | |
426 | ||
427 | /* ptep might be kmapped when using 32-bit HIGHPTE */ | |
428 | u->ptr = arbitrary_virt_to_machine(ptep).maddr; | |
429 | u->val = pte_val_ma(pteval); | |
430 | ||
431 | MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_IO); | |
432 | ||
433 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
434 | } | |
435 | ||
7708ad64 | 436 | static void xen_extend_mmu_update(const struct mmu_update *update) |
3b827c1b | 437 | { |
d66bf8fc JF |
438 | struct multicall_space mcs; |
439 | struct mmu_update *u; | |
3b827c1b | 440 | |
400d3494 JF |
441 | mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u)); |
442 | ||
994025ca JF |
443 | if (mcs.mc != NULL) { |
444 | ADD_STATS(mmu_update_extended, 1); | |
445 | ADD_STATS(mmu_update_histo[mcs.mc->args[1]], -1); | |
446 | ||
400d3494 | 447 | mcs.mc->args[1]++; |
994025ca JF |
448 | |
449 | if (mcs.mc->args[1] < MMU_UPDATE_HISTO) | |
450 | ADD_STATS(mmu_update_histo[mcs.mc->args[1]], 1); | |
451 | else | |
452 | ADD_STATS(mmu_update_histo[0], 1); | |
453 | } else { | |
454 | ADD_STATS(mmu_update, 1); | |
400d3494 JF |
455 | mcs = __xen_mc_entry(sizeof(*u)); |
456 | MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); | |
994025ca | 457 | ADD_STATS(mmu_update_histo[1], 1); |
400d3494 | 458 | } |
d66bf8fc | 459 | |
d66bf8fc | 460 | u = mcs.args; |
400d3494 JF |
461 | *u = *update; |
462 | } | |
463 | ||
464 | void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val) | |
465 | { | |
466 | struct mmu_update u; | |
467 | ||
468 | preempt_disable(); | |
469 | ||
470 | xen_mc_batch(); | |
471 | ||
ce803e70 JF |
472 | /* ptr may be ioremapped for 64-bit pagetable setup */ |
473 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
400d3494 | 474 | u.val = pmd_val_ma(val); |
7708ad64 | 475 | xen_extend_mmu_update(&u); |
d66bf8fc | 476 | |
994025ca JF |
477 | ADD_STATS(pmd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); |
478 | ||
d66bf8fc JF |
479 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
480 | ||
481 | preempt_enable(); | |
3b827c1b JF |
482 | } |
483 | ||
e2426cf8 JF |
484 | void xen_set_pmd(pmd_t *ptr, pmd_t val) |
485 | { | |
994025ca JF |
486 | ADD_STATS(pmd_update, 1); |
487 | ||
e2426cf8 JF |
488 | /* If page is not pinned, we can just update the entry |
489 | directly */ | |
7708ad64 | 490 | if (!xen_page_pinned(ptr)) { |
e2426cf8 JF |
491 | *ptr = val; |
492 | return; | |
493 | } | |
494 | ||
994025ca JF |
495 | ADD_STATS(pmd_update_pinned, 1); |
496 | ||
e2426cf8 JF |
497 | xen_set_pmd_hyper(ptr, val); |
498 | } | |
499 | ||
3b827c1b JF |
500 | /* |
501 | * Associate a virtual page frame with a given physical page frame | |
502 | * and protection flags for that frame. | |
503 | */ | |
504 | void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags) | |
505 | { | |
836fe2f2 | 506 | set_pte_vaddr(vaddr, mfn_pte(mfn, flags)); |
3b827c1b JF |
507 | } |
508 | ||
509 | void xen_set_pte_at(struct mm_struct *mm, unsigned long addr, | |
510 | pte_t *ptep, pte_t pteval) | |
511 | { | |
c0011dbf JF |
512 | if (xen_iomap_pte(pteval)) { |
513 | xen_set_iomap_pte(ptep, pteval); | |
514 | goto out; | |
515 | } | |
516 | ||
994025ca JF |
517 | ADD_STATS(set_pte_at, 1); |
518 | // ADD_STATS(set_pte_at_pinned, xen_page_pinned(ptep)); | |
519 | ADD_STATS(set_pte_at_current, mm == current->mm); | |
520 | ADD_STATS(set_pte_at_kernel, mm == &init_mm); | |
521 | ||
d66bf8fc | 522 | if (mm == current->mm || mm == &init_mm) { |
8965c1c0 | 523 | if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) { |
d66bf8fc JF |
524 | struct multicall_space mcs; |
525 | mcs = xen_mc_entry(0); | |
526 | ||
527 | MULTI_update_va_mapping(mcs.mc, addr, pteval, 0); | |
994025ca | 528 | ADD_STATS(set_pte_at_batched, 1); |
d66bf8fc | 529 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
2bd50036 | 530 | goto out; |
d66bf8fc JF |
531 | } else |
532 | if (HYPERVISOR_update_va_mapping(addr, pteval, 0) == 0) | |
2bd50036 | 533 | goto out; |
d66bf8fc JF |
534 | } |
535 | xen_set_pte(ptep, pteval); | |
2bd50036 | 536 | |
2829b449 | 537 | out: return; |
3b827c1b JF |
538 | } |
539 | ||
f63c2f24 T |
540 | pte_t xen_ptep_modify_prot_start(struct mm_struct *mm, |
541 | unsigned long addr, pte_t *ptep) | |
947a69c9 | 542 | { |
e57778a1 JF |
543 | /* Just return the pte as-is. We preserve the bits on commit */ |
544 | return *ptep; | |
545 | } | |
546 | ||
547 | void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr, | |
548 | pte_t *ptep, pte_t pte) | |
549 | { | |
400d3494 | 550 | struct mmu_update u; |
e57778a1 | 551 | |
400d3494 | 552 | xen_mc_batch(); |
947a69c9 | 553 | |
9f32d21c | 554 | u.ptr = arbitrary_virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD; |
400d3494 | 555 | u.val = pte_val_ma(pte); |
7708ad64 | 556 | xen_extend_mmu_update(&u); |
947a69c9 | 557 | |
994025ca JF |
558 | ADD_STATS(prot_commit, 1); |
559 | ADD_STATS(prot_commit_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); | |
560 | ||
e57778a1 | 561 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
947a69c9 JF |
562 | } |
563 | ||
ebb9cfe2 JF |
564 | /* Assume pteval_t is equivalent to all the other *val_t types. */ |
565 | static pteval_t pte_mfn_to_pfn(pteval_t val) | |
947a69c9 | 566 | { |
ebb9cfe2 | 567 | if (val & _PAGE_PRESENT) { |
59438c9f | 568 | unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; |
77be1fab | 569 | pteval_t flags = val & PTE_FLAGS_MASK; |
d8355aca | 570 | val = ((pteval_t)mfn_to_pfn(mfn) << PAGE_SHIFT) | flags; |
ebb9cfe2 | 571 | } |
947a69c9 | 572 | |
ebb9cfe2 | 573 | return val; |
947a69c9 JF |
574 | } |
575 | ||
ebb9cfe2 | 576 | static pteval_t pte_pfn_to_mfn(pteval_t val) |
947a69c9 | 577 | { |
ebb9cfe2 | 578 | if (val & _PAGE_PRESENT) { |
59438c9f | 579 | unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; |
77be1fab | 580 | pteval_t flags = val & PTE_FLAGS_MASK; |
d8355aca | 581 | val = ((pteval_t)pfn_to_mfn(pfn) << PAGE_SHIFT) | flags; |
947a69c9 JF |
582 | } |
583 | ||
ebb9cfe2 | 584 | return val; |
947a69c9 JF |
585 | } |
586 | ||
c0011dbf JF |
587 | static pteval_t iomap_pte(pteval_t val) |
588 | { | |
589 | if (val & _PAGE_PRESENT) { | |
590 | unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; | |
591 | pteval_t flags = val & PTE_FLAGS_MASK; | |
592 | ||
593 | /* We assume the pte frame number is a MFN, so | |
594 | just use it as-is. */ | |
595 | val = ((pteval_t)pfn << PAGE_SHIFT) | flags; | |
596 | } | |
597 | ||
598 | return val; | |
599 | } | |
600 | ||
ebb9cfe2 | 601 | pteval_t xen_pte_val(pte_t pte) |
947a69c9 | 602 | { |
c0011dbf JF |
603 | if (xen_initial_domain() && (pte.pte & _PAGE_IOMAP)) |
604 | return pte.pte; | |
605 | ||
ebb9cfe2 | 606 | return pte_mfn_to_pfn(pte.pte); |
947a69c9 | 607 | } |
da5de7c2 | 608 | PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val); |
947a69c9 | 609 | |
947a69c9 JF |
610 | pgdval_t xen_pgd_val(pgd_t pgd) |
611 | { | |
ebb9cfe2 | 612 | return pte_mfn_to_pfn(pgd.pgd); |
947a69c9 | 613 | } |
da5de7c2 | 614 | PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val); |
947a69c9 JF |
615 | |
616 | pte_t xen_make_pte(pteval_t pte) | |
617 | { | |
7347b408 AN |
618 | phys_addr_t addr = (pte & PTE_PFN_MASK); |
619 | ||
620 | /* | |
621 | * Unprivileged domains are allowed to do IOMAPpings for | |
622 | * PCI passthrough, but not map ISA space. The ISA | |
623 | * mappings are just dummy local mappings to keep other | |
624 | * parts of the kernel happy. | |
625 | */ | |
626 | if (unlikely(pte & _PAGE_IOMAP) && | |
627 | (xen_initial_domain() || addr >= ISA_END_ADDRESS)) { | |
c0011dbf | 628 | pte = iomap_pte(pte); |
7347b408 AN |
629 | } else { |
630 | pte &= ~_PAGE_IOMAP; | |
c0011dbf | 631 | pte = pte_pfn_to_mfn(pte); |
7347b408 | 632 | } |
c0011dbf | 633 | |
ebb9cfe2 | 634 | return native_make_pte(pte); |
947a69c9 | 635 | } |
da5de7c2 | 636 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte); |
947a69c9 JF |
637 | |
638 | pgd_t xen_make_pgd(pgdval_t pgd) | |
639 | { | |
ebb9cfe2 JF |
640 | pgd = pte_pfn_to_mfn(pgd); |
641 | return native_make_pgd(pgd); | |
947a69c9 | 642 | } |
da5de7c2 | 643 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd); |
947a69c9 JF |
644 | |
645 | pmdval_t xen_pmd_val(pmd_t pmd) | |
646 | { | |
ebb9cfe2 | 647 | return pte_mfn_to_pfn(pmd.pmd); |
947a69c9 | 648 | } |
da5de7c2 | 649 | PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val); |
28499143 | 650 | |
e2426cf8 | 651 | void xen_set_pud_hyper(pud_t *ptr, pud_t val) |
f4f97b3e | 652 | { |
400d3494 | 653 | struct mmu_update u; |
f4f97b3e | 654 | |
d66bf8fc JF |
655 | preempt_disable(); |
656 | ||
400d3494 JF |
657 | xen_mc_batch(); |
658 | ||
ce803e70 JF |
659 | /* ptr may be ioremapped for 64-bit pagetable setup */ |
660 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
400d3494 | 661 | u.val = pud_val_ma(val); |
7708ad64 | 662 | xen_extend_mmu_update(&u); |
d66bf8fc | 663 | |
994025ca JF |
664 | ADD_STATS(pud_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); |
665 | ||
d66bf8fc JF |
666 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
667 | ||
668 | preempt_enable(); | |
f4f97b3e JF |
669 | } |
670 | ||
e2426cf8 JF |
671 | void xen_set_pud(pud_t *ptr, pud_t val) |
672 | { | |
994025ca JF |
673 | ADD_STATS(pud_update, 1); |
674 | ||
e2426cf8 JF |
675 | /* If page is not pinned, we can just update the entry |
676 | directly */ | |
7708ad64 | 677 | if (!xen_page_pinned(ptr)) { |
e2426cf8 JF |
678 | *ptr = val; |
679 | return; | |
680 | } | |
681 | ||
994025ca JF |
682 | ADD_STATS(pud_update_pinned, 1); |
683 | ||
e2426cf8 JF |
684 | xen_set_pud_hyper(ptr, val); |
685 | } | |
686 | ||
f4f97b3e JF |
687 | void xen_set_pte(pte_t *ptep, pte_t pte) |
688 | { | |
c0011dbf JF |
689 | if (xen_iomap_pte(pte)) { |
690 | xen_set_iomap_pte(ptep, pte); | |
691 | return; | |
692 | } | |
693 | ||
994025ca JF |
694 | ADD_STATS(pte_update, 1); |
695 | // ADD_STATS(pte_update_pinned, xen_page_pinned(ptep)); | |
696 | ADD_STATS(pte_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); | |
697 | ||
f6e58732 | 698 | #ifdef CONFIG_X86_PAE |
f4f97b3e JF |
699 | ptep->pte_high = pte.pte_high; |
700 | smp_wmb(); | |
701 | ptep->pte_low = pte.pte_low; | |
f6e58732 JF |
702 | #else |
703 | *ptep = pte; | |
704 | #endif | |
f4f97b3e JF |
705 | } |
706 | ||
f6e58732 | 707 | #ifdef CONFIG_X86_PAE |
3b827c1b JF |
708 | void xen_set_pte_atomic(pte_t *ptep, pte_t pte) |
709 | { | |
c0011dbf JF |
710 | if (xen_iomap_pte(pte)) { |
711 | xen_set_iomap_pte(ptep, pte); | |
712 | return; | |
713 | } | |
714 | ||
f6e58732 | 715 | set_64bit((u64 *)ptep, native_pte_val(pte)); |
3b827c1b JF |
716 | } |
717 | ||
718 | void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep) | |
719 | { | |
720 | ptep->pte_low = 0; | |
721 | smp_wmb(); /* make sure low gets written first */ | |
722 | ptep->pte_high = 0; | |
723 | } | |
724 | ||
725 | void xen_pmd_clear(pmd_t *pmdp) | |
726 | { | |
e2426cf8 | 727 | set_pmd(pmdp, __pmd(0)); |
3b827c1b | 728 | } |
f6e58732 | 729 | #endif /* CONFIG_X86_PAE */ |
3b827c1b | 730 | |
abf33038 | 731 | pmd_t xen_make_pmd(pmdval_t pmd) |
3b827c1b | 732 | { |
ebb9cfe2 | 733 | pmd = pte_pfn_to_mfn(pmd); |
947a69c9 | 734 | return native_make_pmd(pmd); |
3b827c1b | 735 | } |
da5de7c2 | 736 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd); |
3b827c1b | 737 | |
f6e58732 JF |
738 | #if PAGETABLE_LEVELS == 4 |
739 | pudval_t xen_pud_val(pud_t pud) | |
740 | { | |
741 | return pte_mfn_to_pfn(pud.pud); | |
742 | } | |
da5de7c2 | 743 | PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val); |
f6e58732 JF |
744 | |
745 | pud_t xen_make_pud(pudval_t pud) | |
746 | { | |
747 | pud = pte_pfn_to_mfn(pud); | |
748 | ||
749 | return native_make_pud(pud); | |
750 | } | |
da5de7c2 | 751 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud); |
f6e58732 | 752 | |
d6182fbf | 753 | pgd_t *xen_get_user_pgd(pgd_t *pgd) |
f6e58732 | 754 | { |
d6182fbf JF |
755 | pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK); |
756 | unsigned offset = pgd - pgd_page; | |
757 | pgd_t *user_ptr = NULL; | |
f6e58732 | 758 | |
d6182fbf JF |
759 | if (offset < pgd_index(USER_LIMIT)) { |
760 | struct page *page = virt_to_page(pgd_page); | |
761 | user_ptr = (pgd_t *)page->private; | |
762 | if (user_ptr) | |
763 | user_ptr += offset; | |
764 | } | |
f6e58732 | 765 | |
d6182fbf JF |
766 | return user_ptr; |
767 | } | |
768 | ||
769 | static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val) | |
770 | { | |
771 | struct mmu_update u; | |
f6e58732 JF |
772 | |
773 | u.ptr = virt_to_machine(ptr).maddr; | |
774 | u.val = pgd_val_ma(val); | |
7708ad64 | 775 | xen_extend_mmu_update(&u); |
d6182fbf JF |
776 | } |
777 | ||
778 | /* | |
779 | * Raw hypercall-based set_pgd, intended for in early boot before | |
780 | * there's a page structure. This implies: | |
781 | * 1. The only existing pagetable is the kernel's | |
782 | * 2. It is always pinned | |
783 | * 3. It has no user pagetable attached to it | |
784 | */ | |
785 | void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val) | |
786 | { | |
787 | preempt_disable(); | |
788 | ||
789 | xen_mc_batch(); | |
790 | ||
791 | __xen_set_pgd_hyper(ptr, val); | |
f6e58732 JF |
792 | |
793 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
794 | ||
795 | preempt_enable(); | |
796 | } | |
797 | ||
798 | void xen_set_pgd(pgd_t *ptr, pgd_t val) | |
799 | { | |
d6182fbf JF |
800 | pgd_t *user_ptr = xen_get_user_pgd(ptr); |
801 | ||
994025ca JF |
802 | ADD_STATS(pgd_update, 1); |
803 | ||
f6e58732 JF |
804 | /* If page is not pinned, we can just update the entry |
805 | directly */ | |
7708ad64 | 806 | if (!xen_page_pinned(ptr)) { |
f6e58732 | 807 | *ptr = val; |
d6182fbf | 808 | if (user_ptr) { |
7708ad64 | 809 | WARN_ON(xen_page_pinned(user_ptr)); |
d6182fbf JF |
810 | *user_ptr = val; |
811 | } | |
f6e58732 JF |
812 | return; |
813 | } | |
814 | ||
994025ca JF |
815 | ADD_STATS(pgd_update_pinned, 1); |
816 | ADD_STATS(pgd_update_batched, paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU); | |
817 | ||
d6182fbf JF |
818 | /* If it's pinned, then we can at least batch the kernel and |
819 | user updates together. */ | |
820 | xen_mc_batch(); | |
821 | ||
822 | __xen_set_pgd_hyper(ptr, val); | |
823 | if (user_ptr) | |
824 | __xen_set_pgd_hyper(user_ptr, val); | |
825 | ||
826 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
f6e58732 JF |
827 | } |
828 | #endif /* PAGETABLE_LEVELS == 4 */ | |
829 | ||
f4f97b3e | 830 | /* |
5deb30d1 JF |
831 | * (Yet another) pagetable walker. This one is intended for pinning a |
832 | * pagetable. This means that it walks a pagetable and calls the | |
833 | * callback function on each page it finds making up the page table, | |
834 | * at every level. It walks the entire pagetable, but it only bothers | |
835 | * pinning pte pages which are below limit. In the normal case this | |
836 | * will be STACK_TOP_MAX, but at boot we need to pin up to | |
837 | * FIXADDR_TOP. | |
838 | * | |
839 | * For 32-bit the important bit is that we don't pin beyond there, | |
840 | * because then we start getting into Xen's ptes. | |
841 | * | |
842 | * For 64-bit, we must skip the Xen hole in the middle of the address | |
843 | * space, just after the big x86-64 virtual hole. | |
844 | */ | |
86bbc2c2 IC |
845 | static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd, |
846 | int (*func)(struct mm_struct *mm, struct page *, | |
847 | enum pt_level), | |
848 | unsigned long limit) | |
3b827c1b | 849 | { |
f4f97b3e | 850 | int flush = 0; |
5deb30d1 JF |
851 | unsigned hole_low, hole_high; |
852 | unsigned pgdidx_limit, pudidx_limit, pmdidx_limit; | |
853 | unsigned pgdidx, pudidx, pmdidx; | |
f4f97b3e | 854 | |
5deb30d1 JF |
855 | /* The limit is the last byte to be touched */ |
856 | limit--; | |
857 | BUG_ON(limit >= FIXADDR_TOP); | |
3b827c1b JF |
858 | |
859 | if (xen_feature(XENFEAT_auto_translated_physmap)) | |
f4f97b3e JF |
860 | return 0; |
861 | ||
5deb30d1 JF |
862 | /* |
863 | * 64-bit has a great big hole in the middle of the address | |
864 | * space, which contains the Xen mappings. On 32-bit these | |
865 | * will end up making a zero-sized hole and so is a no-op. | |
866 | */ | |
d6182fbf | 867 | hole_low = pgd_index(USER_LIMIT); |
5deb30d1 JF |
868 | hole_high = pgd_index(PAGE_OFFSET); |
869 | ||
870 | pgdidx_limit = pgd_index(limit); | |
871 | #if PTRS_PER_PUD > 1 | |
872 | pudidx_limit = pud_index(limit); | |
873 | #else | |
874 | pudidx_limit = 0; | |
875 | #endif | |
876 | #if PTRS_PER_PMD > 1 | |
877 | pmdidx_limit = pmd_index(limit); | |
878 | #else | |
879 | pmdidx_limit = 0; | |
880 | #endif | |
881 | ||
5deb30d1 | 882 | for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) { |
f4f97b3e | 883 | pud_t *pud; |
3b827c1b | 884 | |
5deb30d1 JF |
885 | if (pgdidx >= hole_low && pgdidx < hole_high) |
886 | continue; | |
f4f97b3e | 887 | |
5deb30d1 | 888 | if (!pgd_val(pgd[pgdidx])) |
3b827c1b | 889 | continue; |
f4f97b3e | 890 | |
5deb30d1 | 891 | pud = pud_offset(&pgd[pgdidx], 0); |
3b827c1b JF |
892 | |
893 | if (PTRS_PER_PUD > 1) /* not folded */ | |
eefb47f6 | 894 | flush |= (*func)(mm, virt_to_page(pud), PT_PUD); |
f4f97b3e | 895 | |
5deb30d1 | 896 | for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) { |
f4f97b3e | 897 | pmd_t *pmd; |
f4f97b3e | 898 | |
5deb30d1 JF |
899 | if (pgdidx == pgdidx_limit && |
900 | pudidx > pudidx_limit) | |
901 | goto out; | |
3b827c1b | 902 | |
5deb30d1 | 903 | if (pud_none(pud[pudidx])) |
3b827c1b | 904 | continue; |
f4f97b3e | 905 | |
5deb30d1 | 906 | pmd = pmd_offset(&pud[pudidx], 0); |
3b827c1b JF |
907 | |
908 | if (PTRS_PER_PMD > 1) /* not folded */ | |
eefb47f6 | 909 | flush |= (*func)(mm, virt_to_page(pmd), PT_PMD); |
f4f97b3e | 910 | |
5deb30d1 JF |
911 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) { |
912 | struct page *pte; | |
913 | ||
914 | if (pgdidx == pgdidx_limit && | |
915 | pudidx == pudidx_limit && | |
916 | pmdidx > pmdidx_limit) | |
917 | goto out; | |
3b827c1b | 918 | |
5deb30d1 | 919 | if (pmd_none(pmd[pmdidx])) |
3b827c1b JF |
920 | continue; |
921 | ||
5deb30d1 | 922 | pte = pmd_page(pmd[pmdidx]); |
eefb47f6 | 923 | flush |= (*func)(mm, pte, PT_PTE); |
3b827c1b JF |
924 | } |
925 | } | |
926 | } | |
11ad93e5 | 927 | |
5deb30d1 | 928 | out: |
11ad93e5 JF |
929 | /* Do the top level last, so that the callbacks can use it as |
930 | a cue to do final things like tlb flushes. */ | |
eefb47f6 | 931 | flush |= (*func)(mm, virt_to_page(pgd), PT_PGD); |
f4f97b3e JF |
932 | |
933 | return flush; | |
3b827c1b JF |
934 | } |
935 | ||
86bbc2c2 IC |
936 | static int xen_pgd_walk(struct mm_struct *mm, |
937 | int (*func)(struct mm_struct *mm, struct page *, | |
938 | enum pt_level), | |
939 | unsigned long limit) | |
940 | { | |
941 | return __xen_pgd_walk(mm, mm->pgd, func, limit); | |
942 | } | |
943 | ||
7708ad64 JF |
944 | /* If we're using split pte locks, then take the page's lock and |
945 | return a pointer to it. Otherwise return NULL. */ | |
eefb47f6 | 946 | static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm) |
74260714 JF |
947 | { |
948 | spinlock_t *ptl = NULL; | |
949 | ||
f7d0b926 | 950 | #if USE_SPLIT_PTLOCKS |
74260714 | 951 | ptl = __pte_lockptr(page); |
eefb47f6 | 952 | spin_lock_nest_lock(ptl, &mm->page_table_lock); |
74260714 JF |
953 | #endif |
954 | ||
955 | return ptl; | |
956 | } | |
957 | ||
7708ad64 | 958 | static void xen_pte_unlock(void *v) |
74260714 JF |
959 | { |
960 | spinlock_t *ptl = v; | |
961 | spin_unlock(ptl); | |
962 | } | |
963 | ||
964 | static void xen_do_pin(unsigned level, unsigned long pfn) | |
965 | { | |
966 | struct mmuext_op *op; | |
967 | struct multicall_space mcs; | |
968 | ||
969 | mcs = __xen_mc_entry(sizeof(*op)); | |
970 | op = mcs.args; | |
971 | op->cmd = level; | |
972 | op->arg1.mfn = pfn_to_mfn(pfn); | |
973 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
974 | } | |
975 | ||
eefb47f6 JF |
976 | static int xen_pin_page(struct mm_struct *mm, struct page *page, |
977 | enum pt_level level) | |
f4f97b3e | 978 | { |
d60cd46b | 979 | unsigned pgfl = TestSetPagePinned(page); |
f4f97b3e JF |
980 | int flush; |
981 | ||
982 | if (pgfl) | |
983 | flush = 0; /* already pinned */ | |
984 | else if (PageHighMem(page)) | |
985 | /* kmaps need flushing if we found an unpinned | |
986 | highpage */ | |
987 | flush = 1; | |
988 | else { | |
989 | void *pt = lowmem_page_address(page); | |
990 | unsigned long pfn = page_to_pfn(page); | |
991 | struct multicall_space mcs = __xen_mc_entry(0); | |
74260714 | 992 | spinlock_t *ptl; |
f4f97b3e JF |
993 | |
994 | flush = 0; | |
995 | ||
11ad93e5 JF |
996 | /* |
997 | * We need to hold the pagetable lock between the time | |
998 | * we make the pagetable RO and when we actually pin | |
999 | * it. If we don't, then other users may come in and | |
1000 | * attempt to update the pagetable by writing it, | |
1001 | * which will fail because the memory is RO but not | |
1002 | * pinned, so Xen won't do the trap'n'emulate. | |
1003 | * | |
1004 | * If we're using split pte locks, we can't hold the | |
1005 | * entire pagetable's worth of locks during the | |
1006 | * traverse, because we may wrap the preempt count (8 | |
1007 | * bits). The solution is to mark RO and pin each PTE | |
1008 | * page while holding the lock. This means the number | |
1009 | * of locks we end up holding is never more than a | |
1010 | * batch size (~32 entries, at present). | |
1011 | * | |
1012 | * If we're not using split pte locks, we needn't pin | |
1013 | * the PTE pages independently, because we're | |
1014 | * protected by the overall pagetable lock. | |
1015 | */ | |
74260714 JF |
1016 | ptl = NULL; |
1017 | if (level == PT_PTE) | |
eefb47f6 | 1018 | ptl = xen_pte_lock(page, mm); |
74260714 | 1019 | |
f4f97b3e JF |
1020 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, |
1021 | pfn_pte(pfn, PAGE_KERNEL_RO), | |
74260714 JF |
1022 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
1023 | ||
11ad93e5 | 1024 | if (ptl) { |
74260714 JF |
1025 | xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn); |
1026 | ||
74260714 JF |
1027 | /* Queue a deferred unlock for when this batch |
1028 | is completed. */ | |
7708ad64 | 1029 | xen_mc_callback(xen_pte_unlock, ptl); |
74260714 | 1030 | } |
f4f97b3e JF |
1031 | } |
1032 | ||
1033 | return flush; | |
1034 | } | |
3b827c1b | 1035 | |
f4f97b3e JF |
1036 | /* This is called just after a mm has been created, but it has not |
1037 | been used yet. We need to make sure that its pagetable is all | |
1038 | read-only, and can be pinned. */ | |
eefb47f6 | 1039 | static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd) |
3b827c1b | 1040 | { |
f4f97b3e | 1041 | xen_mc_batch(); |
3b827c1b | 1042 | |
86bbc2c2 | 1043 | if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) { |
d05fdf31 | 1044 | /* re-enable interrupts for flushing */ |
f87e4cac | 1045 | xen_mc_issue(0); |
d05fdf31 | 1046 | |
f4f97b3e | 1047 | kmap_flush_unused(); |
d05fdf31 | 1048 | |
f87e4cac JF |
1049 | xen_mc_batch(); |
1050 | } | |
f4f97b3e | 1051 | |
d6182fbf JF |
1052 | #ifdef CONFIG_X86_64 |
1053 | { | |
1054 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
1055 | ||
1056 | xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd))); | |
1057 | ||
1058 | if (user_pgd) { | |
eefb47f6 | 1059 | xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD); |
f63c2f24 T |
1060 | xen_do_pin(MMUEXT_PIN_L4_TABLE, |
1061 | PFN_DOWN(__pa(user_pgd))); | |
d6182fbf JF |
1062 | } |
1063 | } | |
1064 | #else /* CONFIG_X86_32 */ | |
5deb30d1 JF |
1065 | #ifdef CONFIG_X86_PAE |
1066 | /* Need to make sure unshared kernel PMD is pinnable */ | |
47cb2ed9 | 1067 | xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), |
eefb47f6 | 1068 | PT_PMD); |
5deb30d1 | 1069 | #endif |
28499143 | 1070 | xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd))); |
d6182fbf | 1071 | #endif /* CONFIG_X86_64 */ |
f4f97b3e | 1072 | xen_mc_issue(0); |
3b827c1b JF |
1073 | } |
1074 | ||
eefb47f6 JF |
1075 | static void xen_pgd_pin(struct mm_struct *mm) |
1076 | { | |
1077 | __xen_pgd_pin(mm, mm->pgd); | |
1078 | } | |
1079 | ||
0e91398f JF |
1080 | /* |
1081 | * On save, we need to pin all pagetables to make sure they get their | |
1082 | * mfns turned into pfns. Search the list for any unpinned pgds and pin | |
1083 | * them (unpinned pgds are not currently in use, probably because the | |
1084 | * process is under construction or destruction). | |
eefb47f6 JF |
1085 | * |
1086 | * Expected to be called in stop_machine() ("equivalent to taking | |
1087 | * every spinlock in the system"), so the locking doesn't really | |
1088 | * matter all that much. | |
0e91398f JF |
1089 | */ |
1090 | void xen_mm_pin_all(void) | |
1091 | { | |
1092 | unsigned long flags; | |
1093 | struct page *page; | |
74260714 | 1094 | |
0e91398f | 1095 | spin_lock_irqsave(&pgd_lock, flags); |
f4f97b3e | 1096 | |
0e91398f JF |
1097 | list_for_each_entry(page, &pgd_list, lru) { |
1098 | if (!PagePinned(page)) { | |
eefb47f6 | 1099 | __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page)); |
0e91398f JF |
1100 | SetPageSavePinned(page); |
1101 | } | |
1102 | } | |
1103 | ||
1104 | spin_unlock_irqrestore(&pgd_lock, flags); | |
3b827c1b JF |
1105 | } |
1106 | ||
c1f2f09e EH |
1107 | /* |
1108 | * The init_mm pagetable is really pinned as soon as its created, but | |
1109 | * that's before we have page structures to store the bits. So do all | |
1110 | * the book-keeping now. | |
1111 | */ | |
eefb47f6 JF |
1112 | static __init int xen_mark_pinned(struct mm_struct *mm, struct page *page, |
1113 | enum pt_level level) | |
3b827c1b | 1114 | { |
f4f97b3e JF |
1115 | SetPagePinned(page); |
1116 | return 0; | |
1117 | } | |
3b827c1b | 1118 | |
b96229b5 | 1119 | static void __init xen_mark_init_mm_pinned(void) |
f4f97b3e | 1120 | { |
eefb47f6 | 1121 | xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP); |
f4f97b3e | 1122 | } |
3b827c1b | 1123 | |
eefb47f6 JF |
1124 | static int xen_unpin_page(struct mm_struct *mm, struct page *page, |
1125 | enum pt_level level) | |
f4f97b3e | 1126 | { |
d60cd46b | 1127 | unsigned pgfl = TestClearPagePinned(page); |
3b827c1b | 1128 | |
f4f97b3e JF |
1129 | if (pgfl && !PageHighMem(page)) { |
1130 | void *pt = lowmem_page_address(page); | |
1131 | unsigned long pfn = page_to_pfn(page); | |
74260714 JF |
1132 | spinlock_t *ptl = NULL; |
1133 | struct multicall_space mcs; | |
1134 | ||
11ad93e5 JF |
1135 | /* |
1136 | * Do the converse to pin_page. If we're using split | |
1137 | * pte locks, we must be holding the lock for while | |
1138 | * the pte page is unpinned but still RO to prevent | |
1139 | * concurrent updates from seeing it in this | |
1140 | * partially-pinned state. | |
1141 | */ | |
74260714 | 1142 | if (level == PT_PTE) { |
eefb47f6 | 1143 | ptl = xen_pte_lock(page, mm); |
74260714 | 1144 | |
11ad93e5 JF |
1145 | if (ptl) |
1146 | xen_do_pin(MMUEXT_UNPIN_TABLE, pfn); | |
74260714 JF |
1147 | } |
1148 | ||
1149 | mcs = __xen_mc_entry(0); | |
f4f97b3e JF |
1150 | |
1151 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, | |
1152 | pfn_pte(pfn, PAGE_KERNEL), | |
74260714 JF |
1153 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
1154 | ||
1155 | if (ptl) { | |
1156 | /* unlock when batch completed */ | |
7708ad64 | 1157 | xen_mc_callback(xen_pte_unlock, ptl); |
74260714 | 1158 | } |
f4f97b3e JF |
1159 | } |
1160 | ||
1161 | return 0; /* never need to flush on unpin */ | |
3b827c1b JF |
1162 | } |
1163 | ||
f4f97b3e | 1164 | /* Release a pagetables pages back as normal RW */ |
eefb47f6 | 1165 | static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd) |
f4f97b3e | 1166 | { |
f4f97b3e JF |
1167 | xen_mc_batch(); |
1168 | ||
74260714 | 1169 | xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); |
f4f97b3e | 1170 | |
d6182fbf JF |
1171 | #ifdef CONFIG_X86_64 |
1172 | { | |
1173 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
1174 | ||
1175 | if (user_pgd) { | |
f63c2f24 T |
1176 | xen_do_pin(MMUEXT_UNPIN_TABLE, |
1177 | PFN_DOWN(__pa(user_pgd))); | |
eefb47f6 | 1178 | xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD); |
d6182fbf JF |
1179 | } |
1180 | } | |
1181 | #endif | |
1182 | ||
5deb30d1 JF |
1183 | #ifdef CONFIG_X86_PAE |
1184 | /* Need to make sure unshared kernel PMD is unpinned */ | |
47cb2ed9 | 1185 | xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), |
eefb47f6 | 1186 | PT_PMD); |
5deb30d1 | 1187 | #endif |
d6182fbf | 1188 | |
86bbc2c2 | 1189 | __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT); |
f4f97b3e JF |
1190 | |
1191 | xen_mc_issue(0); | |
1192 | } | |
3b827c1b | 1193 | |
eefb47f6 JF |
1194 | static void xen_pgd_unpin(struct mm_struct *mm) |
1195 | { | |
1196 | __xen_pgd_unpin(mm, mm->pgd); | |
1197 | } | |
1198 | ||
0e91398f JF |
1199 | /* |
1200 | * On resume, undo any pinning done at save, so that the rest of the | |
1201 | * kernel doesn't see any unexpected pinned pagetables. | |
1202 | */ | |
1203 | void xen_mm_unpin_all(void) | |
1204 | { | |
1205 | unsigned long flags; | |
1206 | struct page *page; | |
1207 | ||
1208 | spin_lock_irqsave(&pgd_lock, flags); | |
1209 | ||
1210 | list_for_each_entry(page, &pgd_list, lru) { | |
1211 | if (PageSavePinned(page)) { | |
1212 | BUG_ON(!PagePinned(page)); | |
eefb47f6 | 1213 | __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page)); |
0e91398f JF |
1214 | ClearPageSavePinned(page); |
1215 | } | |
1216 | } | |
1217 | ||
1218 | spin_unlock_irqrestore(&pgd_lock, flags); | |
1219 | } | |
1220 | ||
3b827c1b JF |
1221 | void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next) |
1222 | { | |
f4f97b3e | 1223 | spin_lock(&next->page_table_lock); |
eefb47f6 | 1224 | xen_pgd_pin(next); |
f4f97b3e | 1225 | spin_unlock(&next->page_table_lock); |
3b827c1b JF |
1226 | } |
1227 | ||
1228 | void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm) | |
1229 | { | |
f4f97b3e | 1230 | spin_lock(&mm->page_table_lock); |
eefb47f6 | 1231 | xen_pgd_pin(mm); |
f4f97b3e | 1232 | spin_unlock(&mm->page_table_lock); |
3b827c1b JF |
1233 | } |
1234 | ||
3b827c1b | 1235 | |
f87e4cac JF |
1236 | #ifdef CONFIG_SMP |
1237 | /* Another cpu may still have their %cr3 pointing at the pagetable, so | |
1238 | we need to repoint it somewhere else before we can unpin it. */ | |
1239 | static void drop_other_mm_ref(void *info) | |
1240 | { | |
1241 | struct mm_struct *mm = info; | |
ce87b3d3 | 1242 | struct mm_struct *active_mm; |
3b827c1b | 1243 | |
9eb912d1 | 1244 | active_mm = percpu_read(cpu_tlbstate.active_mm); |
ce87b3d3 JF |
1245 | |
1246 | if (active_mm == mm) | |
f87e4cac | 1247 | leave_mm(smp_processor_id()); |
9f79991d JF |
1248 | |
1249 | /* If this cpu still has a stale cr3 reference, then make sure | |
1250 | it has been flushed. */ | |
7fd7d83d | 1251 | if (percpu_read(xen_current_cr3) == __pa(mm->pgd)) |
9f79991d | 1252 | load_cr3(swapper_pg_dir); |
f87e4cac | 1253 | } |
3b827c1b | 1254 | |
7708ad64 | 1255 | static void xen_drop_mm_ref(struct mm_struct *mm) |
f87e4cac | 1256 | { |
e4d98207 | 1257 | cpumask_var_t mask; |
9f79991d JF |
1258 | unsigned cpu; |
1259 | ||
f87e4cac JF |
1260 | if (current->active_mm == mm) { |
1261 | if (current->mm == mm) | |
1262 | load_cr3(swapper_pg_dir); | |
1263 | else | |
1264 | leave_mm(smp_processor_id()); | |
9f79991d JF |
1265 | } |
1266 | ||
1267 | /* Get the "official" set of cpus referring to our pagetable. */ | |
e4d98207 MT |
1268 | if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) { |
1269 | for_each_online_cpu(cpu) { | |
78f1c4d6 | 1270 | if (!cpumask_test_cpu(cpu, mm_cpumask(mm)) |
e4d98207 MT |
1271 | && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd)) |
1272 | continue; | |
1273 | smp_call_function_single(cpu, drop_other_mm_ref, mm, 1); | |
1274 | } | |
1275 | return; | |
1276 | } | |
78f1c4d6 | 1277 | cpumask_copy(mask, mm_cpumask(mm)); |
9f79991d JF |
1278 | |
1279 | /* It's possible that a vcpu may have a stale reference to our | |
1280 | cr3, because its in lazy mode, and it hasn't yet flushed | |
1281 | its set of pending hypercalls yet. In this case, we can | |
1282 | look at its actual current cr3 value, and force it to flush | |
1283 | if needed. */ | |
1284 | for_each_online_cpu(cpu) { | |
1285 | if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd)) | |
e4d98207 | 1286 | cpumask_set_cpu(cpu, mask); |
3b827c1b JF |
1287 | } |
1288 | ||
e4d98207 MT |
1289 | if (!cpumask_empty(mask)) |
1290 | smp_call_function_many(mask, drop_other_mm_ref, mm, 1); | |
1291 | free_cpumask_var(mask); | |
f87e4cac JF |
1292 | } |
1293 | #else | |
7708ad64 | 1294 | static void xen_drop_mm_ref(struct mm_struct *mm) |
f87e4cac JF |
1295 | { |
1296 | if (current->active_mm == mm) | |
1297 | load_cr3(swapper_pg_dir); | |
1298 | } | |
1299 | #endif | |
1300 | ||
1301 | /* | |
1302 | * While a process runs, Xen pins its pagetables, which means that the | |
1303 | * hypervisor forces it to be read-only, and it controls all updates | |
1304 | * to it. This means that all pagetable updates have to go via the | |
1305 | * hypervisor, which is moderately expensive. | |
1306 | * | |
1307 | * Since we're pulling the pagetable down, we switch to use init_mm, | |
1308 | * unpin old process pagetable and mark it all read-write, which | |
1309 | * allows further operations on it to be simple memory accesses. | |
1310 | * | |
1311 | * The only subtle point is that another CPU may be still using the | |
1312 | * pagetable because of lazy tlb flushing. This means we need need to | |
1313 | * switch all CPUs off this pagetable before we can unpin it. | |
1314 | */ | |
1315 | void xen_exit_mmap(struct mm_struct *mm) | |
1316 | { | |
1317 | get_cpu(); /* make sure we don't move around */ | |
7708ad64 | 1318 | xen_drop_mm_ref(mm); |
f87e4cac | 1319 | put_cpu(); |
3b827c1b | 1320 | |
f120f13e | 1321 | spin_lock(&mm->page_table_lock); |
df912ea4 JF |
1322 | |
1323 | /* pgd may not be pinned in the error exit path of execve */ | |
7708ad64 | 1324 | if (xen_page_pinned(mm->pgd)) |
eefb47f6 | 1325 | xen_pgd_unpin(mm); |
74260714 | 1326 | |
f120f13e | 1327 | spin_unlock(&mm->page_table_lock); |
3b827c1b | 1328 | } |
994025ca | 1329 | |
319f3ba5 JF |
1330 | static __init void xen_pagetable_setup_start(pgd_t *base) |
1331 | { | |
1332 | } | |
1333 | ||
f1d7062a TG |
1334 | static void xen_post_allocator_init(void); |
1335 | ||
319f3ba5 JF |
1336 | static __init void xen_pagetable_setup_done(pgd_t *base) |
1337 | { | |
1338 | xen_setup_shared_info(); | |
f1d7062a | 1339 | xen_post_allocator_init(); |
319f3ba5 JF |
1340 | } |
1341 | ||
1342 | static void xen_write_cr2(unsigned long cr2) | |
1343 | { | |
1344 | percpu_read(xen_vcpu)->arch.cr2 = cr2; | |
1345 | } | |
1346 | ||
1347 | static unsigned long xen_read_cr2(void) | |
1348 | { | |
1349 | return percpu_read(xen_vcpu)->arch.cr2; | |
1350 | } | |
1351 | ||
1352 | unsigned long xen_read_cr2_direct(void) | |
1353 | { | |
1354 | return percpu_read(xen_vcpu_info.arch.cr2); | |
1355 | } | |
1356 | ||
1357 | static void xen_flush_tlb(void) | |
1358 | { | |
1359 | struct mmuext_op *op; | |
1360 | struct multicall_space mcs; | |
1361 | ||
1362 | preempt_disable(); | |
1363 | ||
1364 | mcs = xen_mc_entry(sizeof(*op)); | |
1365 | ||
1366 | op = mcs.args; | |
1367 | op->cmd = MMUEXT_TLB_FLUSH_LOCAL; | |
1368 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1369 | ||
1370 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1371 | ||
1372 | preempt_enable(); | |
1373 | } | |
1374 | ||
1375 | static void xen_flush_tlb_single(unsigned long addr) | |
1376 | { | |
1377 | struct mmuext_op *op; | |
1378 | struct multicall_space mcs; | |
1379 | ||
1380 | preempt_disable(); | |
1381 | ||
1382 | mcs = xen_mc_entry(sizeof(*op)); | |
1383 | op = mcs.args; | |
1384 | op->cmd = MMUEXT_INVLPG_LOCAL; | |
1385 | op->arg1.linear_addr = addr & PAGE_MASK; | |
1386 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1387 | ||
1388 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1389 | ||
1390 | preempt_enable(); | |
1391 | } | |
1392 | ||
1393 | static void xen_flush_tlb_others(const struct cpumask *cpus, | |
1394 | struct mm_struct *mm, unsigned long va) | |
1395 | { | |
1396 | struct { | |
1397 | struct mmuext_op op; | |
1398 | DECLARE_BITMAP(mask, NR_CPUS); | |
1399 | } *args; | |
1400 | struct multicall_space mcs; | |
1401 | ||
e3f8a74e JF |
1402 | if (cpumask_empty(cpus)) |
1403 | return; /* nothing to do */ | |
319f3ba5 JF |
1404 | |
1405 | mcs = xen_mc_entry(sizeof(*args)); | |
1406 | args = mcs.args; | |
1407 | args->op.arg2.vcpumask = to_cpumask(args->mask); | |
1408 | ||
1409 | /* Remove us, and any offline CPUS. */ | |
1410 | cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask); | |
1411 | cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask)); | |
319f3ba5 JF |
1412 | |
1413 | if (va == TLB_FLUSH_ALL) { | |
1414 | args->op.cmd = MMUEXT_TLB_FLUSH_MULTI; | |
1415 | } else { | |
1416 | args->op.cmd = MMUEXT_INVLPG_MULTI; | |
1417 | args->op.arg1.linear_addr = va; | |
1418 | } | |
1419 | ||
1420 | MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF); | |
1421 | ||
319f3ba5 JF |
1422 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
1423 | } | |
1424 | ||
1425 | static unsigned long xen_read_cr3(void) | |
1426 | { | |
1427 | return percpu_read(xen_cr3); | |
1428 | } | |
1429 | ||
1430 | static void set_current_cr3(void *v) | |
1431 | { | |
1432 | percpu_write(xen_current_cr3, (unsigned long)v); | |
1433 | } | |
1434 | ||
1435 | static void __xen_write_cr3(bool kernel, unsigned long cr3) | |
1436 | { | |
1437 | struct mmuext_op *op; | |
1438 | struct multicall_space mcs; | |
1439 | unsigned long mfn; | |
1440 | ||
1441 | if (cr3) | |
1442 | mfn = pfn_to_mfn(PFN_DOWN(cr3)); | |
1443 | else | |
1444 | mfn = 0; | |
1445 | ||
1446 | WARN_ON(mfn == 0 && kernel); | |
1447 | ||
1448 | mcs = __xen_mc_entry(sizeof(*op)); | |
1449 | ||
1450 | op = mcs.args; | |
1451 | op->cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR; | |
1452 | op->arg1.mfn = mfn; | |
1453 | ||
1454 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1455 | ||
1456 | if (kernel) { | |
1457 | percpu_write(xen_cr3, cr3); | |
1458 | ||
1459 | /* Update xen_current_cr3 once the batch has actually | |
1460 | been submitted. */ | |
1461 | xen_mc_callback(set_current_cr3, (void *)cr3); | |
1462 | } | |
1463 | } | |
1464 | ||
1465 | static void xen_write_cr3(unsigned long cr3) | |
1466 | { | |
1467 | BUG_ON(preemptible()); | |
1468 | ||
1469 | xen_mc_batch(); /* disables interrupts */ | |
1470 | ||
1471 | /* Update while interrupts are disabled, so its atomic with | |
1472 | respect to ipis */ | |
1473 | percpu_write(xen_cr3, cr3); | |
1474 | ||
1475 | __xen_write_cr3(true, cr3); | |
1476 | ||
1477 | #ifdef CONFIG_X86_64 | |
1478 | { | |
1479 | pgd_t *user_pgd = xen_get_user_pgd(__va(cr3)); | |
1480 | if (user_pgd) | |
1481 | __xen_write_cr3(false, __pa(user_pgd)); | |
1482 | else | |
1483 | __xen_write_cr3(false, 0); | |
1484 | } | |
1485 | #endif | |
1486 | ||
1487 | xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ | |
1488 | } | |
1489 | ||
1490 | static int xen_pgd_alloc(struct mm_struct *mm) | |
1491 | { | |
1492 | pgd_t *pgd = mm->pgd; | |
1493 | int ret = 0; | |
1494 | ||
1495 | BUG_ON(PagePinned(virt_to_page(pgd))); | |
1496 | ||
1497 | #ifdef CONFIG_X86_64 | |
1498 | { | |
1499 | struct page *page = virt_to_page(pgd); | |
1500 | pgd_t *user_pgd; | |
1501 | ||
1502 | BUG_ON(page->private != 0); | |
1503 | ||
1504 | ret = -ENOMEM; | |
1505 | ||
1506 | user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO); | |
1507 | page->private = (unsigned long)user_pgd; | |
1508 | ||
1509 | if (user_pgd != NULL) { | |
1510 | user_pgd[pgd_index(VSYSCALL_START)] = | |
1511 | __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE); | |
1512 | ret = 0; | |
1513 | } | |
1514 | ||
1515 | BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd)))); | |
1516 | } | |
1517 | #endif | |
1518 | ||
1519 | return ret; | |
1520 | } | |
1521 | ||
1522 | static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd) | |
1523 | { | |
1524 | #ifdef CONFIG_X86_64 | |
1525 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
1526 | ||
1527 | if (user_pgd) | |
1528 | free_page((unsigned long)user_pgd); | |
1529 | #endif | |
1530 | } | |
1531 | ||
1f4f9315 JF |
1532 | #ifdef CONFIG_X86_32 |
1533 | static __init pte_t mask_rw_pte(pte_t *ptep, pte_t pte) | |
1534 | { | |
1535 | /* If there's an existing pte, then don't allow _PAGE_RW to be set */ | |
1536 | if (pte_val_ma(*ptep) & _PAGE_PRESENT) | |
1537 | pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) & | |
1538 | pte_val_ma(pte)); | |
1539 | ||
1540 | return pte; | |
1541 | } | |
1542 | ||
1543 | /* Init-time set_pte while constructing initial pagetables, which | |
1544 | doesn't allow RO pagetable pages to be remapped RW */ | |
1545 | static __init void xen_set_pte_init(pte_t *ptep, pte_t pte) | |
1546 | { | |
1547 | pte = mask_rw_pte(ptep, pte); | |
1548 | ||
1549 | xen_set_pte(ptep, pte); | |
1550 | } | |
1551 | #endif | |
319f3ba5 | 1552 | |
b96229b5 JF |
1553 | static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn) |
1554 | { | |
1555 | struct mmuext_op op; | |
1556 | op.cmd = cmd; | |
1557 | op.arg1.mfn = pfn_to_mfn(pfn); | |
1558 | if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) | |
1559 | BUG(); | |
1560 | } | |
1561 | ||
319f3ba5 JF |
1562 | /* Early in boot, while setting up the initial pagetable, assume |
1563 | everything is pinned. */ | |
1564 | static __init void xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn) | |
1565 | { | |
b96229b5 JF |
1566 | #ifdef CONFIG_FLATMEM |
1567 | BUG_ON(mem_map); /* should only be used early */ | |
1568 | #endif | |
1569 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1570 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
1571 | } | |
1572 | ||
1573 | /* Used for pmd and pud */ | |
1574 | static __init void xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn) | |
1575 | { | |
319f3ba5 JF |
1576 | #ifdef CONFIG_FLATMEM |
1577 | BUG_ON(mem_map); /* should only be used early */ | |
1578 | #endif | |
1579 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1580 | } | |
1581 | ||
1582 | /* Early release_pte assumes that all pts are pinned, since there's | |
1583 | only init_mm and anything attached to that is pinned. */ | |
b96229b5 | 1584 | static __init void xen_release_pte_init(unsigned long pfn) |
319f3ba5 | 1585 | { |
b96229b5 | 1586 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); |
319f3ba5 JF |
1587 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); |
1588 | } | |
1589 | ||
b96229b5 | 1590 | static __init void xen_release_pmd_init(unsigned long pfn) |
319f3ba5 | 1591 | { |
b96229b5 | 1592 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); |
319f3ba5 JF |
1593 | } |
1594 | ||
1595 | /* This needs to make sure the new pte page is pinned iff its being | |
1596 | attached to a pinned pagetable. */ | |
1597 | static void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, unsigned level) | |
1598 | { | |
1599 | struct page *page = pfn_to_page(pfn); | |
1600 | ||
1601 | if (PagePinned(virt_to_page(mm->pgd))) { | |
1602 | SetPagePinned(page); | |
1603 | ||
319f3ba5 JF |
1604 | if (!PageHighMem(page)) { |
1605 | make_lowmem_page_readonly(__va(PFN_PHYS((unsigned long)pfn))); | |
1606 | if (level == PT_PTE && USE_SPLIT_PTLOCKS) | |
1607 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
1608 | } else { | |
1609 | /* make sure there are no stray mappings of | |
1610 | this page */ | |
1611 | kmap_flush_unused(); | |
1612 | } | |
1613 | } | |
1614 | } | |
1615 | ||
1616 | static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn) | |
1617 | { | |
1618 | xen_alloc_ptpage(mm, pfn, PT_PTE); | |
1619 | } | |
1620 | ||
1621 | static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn) | |
1622 | { | |
1623 | xen_alloc_ptpage(mm, pfn, PT_PMD); | |
1624 | } | |
1625 | ||
1626 | /* This should never happen until we're OK to use struct page */ | |
1627 | static void xen_release_ptpage(unsigned long pfn, unsigned level) | |
1628 | { | |
1629 | struct page *page = pfn_to_page(pfn); | |
1630 | ||
1631 | if (PagePinned(page)) { | |
1632 | if (!PageHighMem(page)) { | |
1633 | if (level == PT_PTE && USE_SPLIT_PTLOCKS) | |
1634 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); | |
1635 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
1636 | } | |
1637 | ClearPagePinned(page); | |
1638 | } | |
1639 | } | |
1640 | ||
1641 | static void xen_release_pte(unsigned long pfn) | |
1642 | { | |
1643 | xen_release_ptpage(pfn, PT_PTE); | |
1644 | } | |
1645 | ||
1646 | static void xen_release_pmd(unsigned long pfn) | |
1647 | { | |
1648 | xen_release_ptpage(pfn, PT_PMD); | |
1649 | } | |
1650 | ||
1651 | #if PAGETABLE_LEVELS == 4 | |
1652 | static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn) | |
1653 | { | |
1654 | xen_alloc_ptpage(mm, pfn, PT_PUD); | |
1655 | } | |
1656 | ||
1657 | static void xen_release_pud(unsigned long pfn) | |
1658 | { | |
1659 | xen_release_ptpage(pfn, PT_PUD); | |
1660 | } | |
1661 | #endif | |
1662 | ||
1663 | void __init xen_reserve_top(void) | |
1664 | { | |
1665 | #ifdef CONFIG_X86_32 | |
1666 | unsigned long top = HYPERVISOR_VIRT_START; | |
1667 | struct xen_platform_parameters pp; | |
1668 | ||
1669 | if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0) | |
1670 | top = pp.virt_start; | |
1671 | ||
1672 | reserve_top_address(-top); | |
1673 | #endif /* CONFIG_X86_32 */ | |
1674 | } | |
1675 | ||
1676 | /* | |
1677 | * Like __va(), but returns address in the kernel mapping (which is | |
1678 | * all we have until the physical memory mapping has been set up. | |
1679 | */ | |
1680 | static void *__ka(phys_addr_t paddr) | |
1681 | { | |
1682 | #ifdef CONFIG_X86_64 | |
1683 | return (void *)(paddr + __START_KERNEL_map); | |
1684 | #else | |
1685 | return __va(paddr); | |
1686 | #endif | |
1687 | } | |
1688 | ||
1689 | /* Convert a machine address to physical address */ | |
1690 | static unsigned long m2p(phys_addr_t maddr) | |
1691 | { | |
1692 | phys_addr_t paddr; | |
1693 | ||
1694 | maddr &= PTE_PFN_MASK; | |
1695 | paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT; | |
1696 | ||
1697 | return paddr; | |
1698 | } | |
1699 | ||
1700 | /* Convert a machine address to kernel virtual */ | |
1701 | static void *m2v(phys_addr_t maddr) | |
1702 | { | |
1703 | return __ka(m2p(maddr)); | |
1704 | } | |
1705 | ||
1706 | static void set_page_prot(void *addr, pgprot_t prot) | |
1707 | { | |
1708 | unsigned long pfn = __pa(addr) >> PAGE_SHIFT; | |
1709 | pte_t pte = pfn_pte(pfn, prot); | |
1710 | ||
1711 | if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, 0)) | |
1712 | BUG(); | |
1713 | } | |
1714 | ||
1715 | static __init void xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn) | |
1716 | { | |
1717 | unsigned pmdidx, pteidx; | |
1718 | unsigned ident_pte; | |
1719 | unsigned long pfn; | |
1720 | ||
1721 | ident_pte = 0; | |
1722 | pfn = 0; | |
1723 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) { | |
1724 | pte_t *pte_page; | |
1725 | ||
1726 | /* Reuse or allocate a page of ptes */ | |
1727 | if (pmd_present(pmd[pmdidx])) | |
1728 | pte_page = m2v(pmd[pmdidx].pmd); | |
1729 | else { | |
1730 | /* Check for free pte pages */ | |
1731 | if (ident_pte == ARRAY_SIZE(level1_ident_pgt)) | |
1732 | break; | |
1733 | ||
1734 | pte_page = &level1_ident_pgt[ident_pte]; | |
1735 | ident_pte += PTRS_PER_PTE; | |
1736 | ||
1737 | pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE); | |
1738 | } | |
1739 | ||
1740 | /* Install mappings */ | |
1741 | for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) { | |
1742 | pte_t pte; | |
1743 | ||
1744 | if (pfn > max_pfn_mapped) | |
1745 | max_pfn_mapped = pfn; | |
1746 | ||
1747 | if (!pte_none(pte_page[pteidx])) | |
1748 | continue; | |
1749 | ||
1750 | pte = pfn_pte(pfn, PAGE_KERNEL_EXEC); | |
1751 | pte_page[pteidx] = pte; | |
1752 | } | |
1753 | } | |
1754 | ||
1755 | for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE) | |
1756 | set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO); | |
1757 | ||
1758 | set_page_prot(pmd, PAGE_KERNEL_RO); | |
1759 | } | |
1760 | ||
1761 | #ifdef CONFIG_X86_64 | |
1762 | static void convert_pfn_mfn(void *v) | |
1763 | { | |
1764 | pte_t *pte = v; | |
1765 | int i; | |
1766 | ||
1767 | /* All levels are converted the same way, so just treat them | |
1768 | as ptes. */ | |
1769 | for (i = 0; i < PTRS_PER_PTE; i++) | |
1770 | pte[i] = xen_make_pte(pte[i].pte); | |
1771 | } | |
1772 | ||
1773 | /* | |
1774 | * Set up the inital kernel pagetable. | |
1775 | * | |
1776 | * We can construct this by grafting the Xen provided pagetable into | |
1777 | * head_64.S's preconstructed pagetables. We copy the Xen L2's into | |
1778 | * level2_ident_pgt, level2_kernel_pgt and level2_fixmap_pgt. This | |
1779 | * means that only the kernel has a physical mapping to start with - | |
1780 | * but that's enough to get __va working. We need to fill in the rest | |
1781 | * of the physical mapping once some sort of allocator has been set | |
1782 | * up. | |
1783 | */ | |
1784 | __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, | |
1785 | unsigned long max_pfn) | |
1786 | { | |
1787 | pud_t *l3; | |
1788 | pmd_t *l2; | |
1789 | ||
1790 | /* Zap identity mapping */ | |
1791 | init_level4_pgt[0] = __pgd(0); | |
1792 | ||
1793 | /* Pre-constructed entries are in pfn, so convert to mfn */ | |
1794 | convert_pfn_mfn(init_level4_pgt); | |
1795 | convert_pfn_mfn(level3_ident_pgt); | |
1796 | convert_pfn_mfn(level3_kernel_pgt); | |
1797 | ||
1798 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd); | |
1799 | l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud); | |
1800 | ||
1801 | memcpy(level2_ident_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1802 | memcpy(level2_kernel_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1803 | ||
1804 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map + PMD_SIZE)].pgd); | |
1805 | l2 = m2v(l3[pud_index(__START_KERNEL_map + PMD_SIZE)].pud); | |
1806 | memcpy(level2_fixmap_pgt, l2, sizeof(pmd_t) * PTRS_PER_PMD); | |
1807 | ||
1808 | /* Set up identity map */ | |
1809 | xen_map_identity_early(level2_ident_pgt, max_pfn); | |
1810 | ||
1811 | /* Make pagetable pieces RO */ | |
1812 | set_page_prot(init_level4_pgt, PAGE_KERNEL_RO); | |
1813 | set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO); | |
1814 | set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO); | |
1815 | set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO); | |
1816 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); | |
1817 | set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO); | |
1818 | ||
1819 | /* Pin down new L4 */ | |
1820 | pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE, | |
1821 | PFN_DOWN(__pa_symbol(init_level4_pgt))); | |
1822 | ||
1823 | /* Unpin Xen-provided one */ | |
1824 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
1825 | ||
1826 | /* Switch over */ | |
1827 | pgd = init_level4_pgt; | |
1828 | ||
1829 | /* | |
1830 | * At this stage there can be no user pgd, and no page | |
1831 | * structure to attach it to, so make sure we just set kernel | |
1832 | * pgd. | |
1833 | */ | |
1834 | xen_mc_batch(); | |
1835 | __xen_write_cr3(true, __pa(pgd)); | |
1836 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
1837 | ||
1838 | reserve_early(__pa(xen_start_info->pt_base), | |
1839 | __pa(xen_start_info->pt_base + | |
1840 | xen_start_info->nr_pt_frames * PAGE_SIZE), | |
1841 | "XEN PAGETABLES"); | |
1842 | ||
1843 | return pgd; | |
1844 | } | |
1845 | #else /* !CONFIG_X86_64 */ | |
1846 | static pmd_t level2_kernel_pgt[PTRS_PER_PMD] __page_aligned_bss; | |
1847 | ||
1848 | __init pgd_t *xen_setup_kernel_pagetable(pgd_t *pgd, | |
1849 | unsigned long max_pfn) | |
1850 | { | |
1851 | pmd_t *kernel_pmd; | |
1852 | ||
93dbda7c JF |
1853 | max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) + |
1854 | xen_start_info->nr_pt_frames * PAGE_SIZE + | |
1855 | 512*1024); | |
319f3ba5 JF |
1856 | |
1857 | kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd); | |
1858 | memcpy(level2_kernel_pgt, kernel_pmd, sizeof(pmd_t) * PTRS_PER_PMD); | |
1859 | ||
1860 | xen_map_identity_early(level2_kernel_pgt, max_pfn); | |
1861 | ||
1862 | memcpy(swapper_pg_dir, pgd, sizeof(pgd_t) * PTRS_PER_PGD); | |
1863 | set_pgd(&swapper_pg_dir[KERNEL_PGD_BOUNDARY], | |
1864 | __pgd(__pa(level2_kernel_pgt) | _PAGE_PRESENT)); | |
1865 | ||
1866 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); | |
1867 | set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO); | |
1868 | set_page_prot(empty_zero_page, PAGE_KERNEL_RO); | |
1869 | ||
1870 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
1871 | ||
1872 | xen_write_cr3(__pa(swapper_pg_dir)); | |
1873 | ||
1874 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(swapper_pg_dir))); | |
1875 | ||
33df4db0 JF |
1876 | reserve_early(__pa(xen_start_info->pt_base), |
1877 | __pa(xen_start_info->pt_base + | |
1878 | xen_start_info->nr_pt_frames * PAGE_SIZE), | |
1879 | "XEN PAGETABLES"); | |
1880 | ||
319f3ba5 JF |
1881 | return swapper_pg_dir; |
1882 | } | |
1883 | #endif /* CONFIG_X86_64 */ | |
1884 | ||
3b3809ac | 1885 | static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot) |
319f3ba5 JF |
1886 | { |
1887 | pte_t pte; | |
1888 | ||
1889 | phys >>= PAGE_SHIFT; | |
1890 | ||
1891 | switch (idx) { | |
1892 | case FIX_BTMAP_END ... FIX_BTMAP_BEGIN: | |
1893 | #ifdef CONFIG_X86_F00F_BUG | |
1894 | case FIX_F00F_IDT: | |
1895 | #endif | |
1896 | #ifdef CONFIG_X86_32 | |
1897 | case FIX_WP_TEST: | |
1898 | case FIX_VDSO: | |
1899 | # ifdef CONFIG_HIGHMEM | |
1900 | case FIX_KMAP_BEGIN ... FIX_KMAP_END: | |
1901 | # endif | |
1902 | #else | |
1903 | case VSYSCALL_LAST_PAGE ... VSYSCALL_FIRST_PAGE: | |
1904 | #endif | |
1905 | #ifdef CONFIG_X86_LOCAL_APIC | |
1906 | case FIX_APIC_BASE: /* maps dummy local APIC */ | |
1907 | #endif | |
3ecb1b7d JF |
1908 | case FIX_TEXT_POKE0: |
1909 | case FIX_TEXT_POKE1: | |
1910 | /* All local page mappings */ | |
319f3ba5 JF |
1911 | pte = pfn_pte(phys, prot); |
1912 | break; | |
1913 | ||
c0011dbf JF |
1914 | case FIX_PARAVIRT_BOOTMAP: |
1915 | /* This is an MFN, but it isn't an IO mapping from the | |
1916 | IO domain */ | |
319f3ba5 JF |
1917 | pte = mfn_pte(phys, prot); |
1918 | break; | |
c0011dbf JF |
1919 | |
1920 | default: | |
1921 | /* By default, set_fixmap is used for hardware mappings */ | |
1922 | pte = mfn_pte(phys, __pgprot(pgprot_val(prot) | _PAGE_IOMAP)); | |
1923 | break; | |
319f3ba5 JF |
1924 | } |
1925 | ||
1926 | __native_set_fixmap(idx, pte); | |
1927 | ||
1928 | #ifdef CONFIG_X86_64 | |
1929 | /* Replicate changes to map the vsyscall page into the user | |
1930 | pagetable vsyscall mapping. */ | |
1931 | if (idx >= VSYSCALL_LAST_PAGE && idx <= VSYSCALL_FIRST_PAGE) { | |
1932 | unsigned long vaddr = __fix_to_virt(idx); | |
1933 | set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte); | |
1934 | } | |
1935 | #endif | |
1936 | } | |
1937 | ||
f1d7062a | 1938 | static __init void xen_post_allocator_init(void) |
319f3ba5 JF |
1939 | { |
1940 | pv_mmu_ops.set_pte = xen_set_pte; | |
1941 | pv_mmu_ops.set_pmd = xen_set_pmd; | |
1942 | pv_mmu_ops.set_pud = xen_set_pud; | |
1943 | #if PAGETABLE_LEVELS == 4 | |
1944 | pv_mmu_ops.set_pgd = xen_set_pgd; | |
1945 | #endif | |
1946 | ||
1947 | /* This will work as long as patching hasn't happened yet | |
1948 | (which it hasn't) */ | |
1949 | pv_mmu_ops.alloc_pte = xen_alloc_pte; | |
1950 | pv_mmu_ops.alloc_pmd = xen_alloc_pmd; | |
1951 | pv_mmu_ops.release_pte = xen_release_pte; | |
1952 | pv_mmu_ops.release_pmd = xen_release_pmd; | |
1953 | #if PAGETABLE_LEVELS == 4 | |
1954 | pv_mmu_ops.alloc_pud = xen_alloc_pud; | |
1955 | pv_mmu_ops.release_pud = xen_release_pud; | |
1956 | #endif | |
1957 | ||
1958 | #ifdef CONFIG_X86_64 | |
1959 | SetPagePinned(virt_to_page(level3_user_vsyscall)); | |
1960 | #endif | |
1961 | xen_mark_init_mm_pinned(); | |
1962 | } | |
1963 | ||
b407fc57 JF |
1964 | static void xen_leave_lazy_mmu(void) |
1965 | { | |
5caecb94 | 1966 | preempt_disable(); |
b407fc57 JF |
1967 | xen_mc_flush(); |
1968 | paravirt_leave_lazy_mmu(); | |
5caecb94 | 1969 | preempt_enable(); |
b407fc57 | 1970 | } |
319f3ba5 | 1971 | |
030cb6c0 | 1972 | static const struct pv_mmu_ops xen_mmu_ops __initdata = { |
319f3ba5 JF |
1973 | .read_cr2 = xen_read_cr2, |
1974 | .write_cr2 = xen_write_cr2, | |
1975 | ||
1976 | .read_cr3 = xen_read_cr3, | |
1977 | .write_cr3 = xen_write_cr3, | |
1978 | ||
1979 | .flush_tlb_user = xen_flush_tlb, | |
1980 | .flush_tlb_kernel = xen_flush_tlb, | |
1981 | .flush_tlb_single = xen_flush_tlb_single, | |
1982 | .flush_tlb_others = xen_flush_tlb_others, | |
1983 | ||
1984 | .pte_update = paravirt_nop, | |
1985 | .pte_update_defer = paravirt_nop, | |
1986 | ||
1987 | .pgd_alloc = xen_pgd_alloc, | |
1988 | .pgd_free = xen_pgd_free, | |
1989 | ||
1990 | .alloc_pte = xen_alloc_pte_init, | |
1991 | .release_pte = xen_release_pte_init, | |
b96229b5 | 1992 | .alloc_pmd = xen_alloc_pmd_init, |
319f3ba5 | 1993 | .alloc_pmd_clone = paravirt_nop, |
b96229b5 | 1994 | .release_pmd = xen_release_pmd_init, |
319f3ba5 | 1995 | |
319f3ba5 JF |
1996 | #ifdef CONFIG_X86_64 |
1997 | .set_pte = xen_set_pte, | |
1998 | #else | |
1999 | .set_pte = xen_set_pte_init, | |
2000 | #endif | |
2001 | .set_pte_at = xen_set_pte_at, | |
2002 | .set_pmd = xen_set_pmd_hyper, | |
2003 | ||
2004 | .ptep_modify_prot_start = __ptep_modify_prot_start, | |
2005 | .ptep_modify_prot_commit = __ptep_modify_prot_commit, | |
2006 | ||
da5de7c2 JF |
2007 | .pte_val = PV_CALLEE_SAVE(xen_pte_val), |
2008 | .pgd_val = PV_CALLEE_SAVE(xen_pgd_val), | |
319f3ba5 | 2009 | |
da5de7c2 JF |
2010 | .make_pte = PV_CALLEE_SAVE(xen_make_pte), |
2011 | .make_pgd = PV_CALLEE_SAVE(xen_make_pgd), | |
319f3ba5 JF |
2012 | |
2013 | #ifdef CONFIG_X86_PAE | |
2014 | .set_pte_atomic = xen_set_pte_atomic, | |
319f3ba5 JF |
2015 | .pte_clear = xen_pte_clear, |
2016 | .pmd_clear = xen_pmd_clear, | |
2017 | #endif /* CONFIG_X86_PAE */ | |
2018 | .set_pud = xen_set_pud_hyper, | |
2019 | ||
da5de7c2 JF |
2020 | .make_pmd = PV_CALLEE_SAVE(xen_make_pmd), |
2021 | .pmd_val = PV_CALLEE_SAVE(xen_pmd_val), | |
319f3ba5 JF |
2022 | |
2023 | #if PAGETABLE_LEVELS == 4 | |
da5de7c2 JF |
2024 | .pud_val = PV_CALLEE_SAVE(xen_pud_val), |
2025 | .make_pud = PV_CALLEE_SAVE(xen_make_pud), | |
319f3ba5 JF |
2026 | .set_pgd = xen_set_pgd_hyper, |
2027 | ||
b96229b5 JF |
2028 | .alloc_pud = xen_alloc_pmd_init, |
2029 | .release_pud = xen_release_pmd_init, | |
319f3ba5 JF |
2030 | #endif /* PAGETABLE_LEVELS == 4 */ |
2031 | ||
2032 | .activate_mm = xen_activate_mm, | |
2033 | .dup_mmap = xen_dup_mmap, | |
2034 | .exit_mmap = xen_exit_mmap, | |
2035 | ||
2036 | .lazy_mode = { | |
2037 | .enter = paravirt_enter_lazy_mmu, | |
b407fc57 | 2038 | .leave = xen_leave_lazy_mmu, |
319f3ba5 JF |
2039 | }, |
2040 | ||
2041 | .set_fixmap = xen_set_fixmap, | |
2042 | }; | |
2043 | ||
030cb6c0 TG |
2044 | void __init xen_init_mmu_ops(void) |
2045 | { | |
2046 | x86_init.paging.pagetable_setup_start = xen_pagetable_setup_start; | |
2047 | x86_init.paging.pagetable_setup_done = xen_pagetable_setup_done; | |
2048 | pv_mmu_ops = xen_mmu_ops; | |
d2cb2145 JF |
2049 | |
2050 | vmap_lazy_unmap = false; | |
030cb6c0 | 2051 | } |
319f3ba5 | 2052 | |
08bbc9da AN |
2053 | /* Protected by xen_reservation_lock. */ |
2054 | #define MAX_CONTIG_ORDER 9 /* 2MB */ | |
2055 | static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER]; | |
2056 | ||
2057 | #define VOID_PTE (mfn_pte(0, __pgprot(0))) | |
2058 | static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order, | |
2059 | unsigned long *in_frames, | |
2060 | unsigned long *out_frames) | |
2061 | { | |
2062 | int i; | |
2063 | struct multicall_space mcs; | |
2064 | ||
2065 | xen_mc_batch(); | |
2066 | for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) { | |
2067 | mcs = __xen_mc_entry(0); | |
2068 | ||
2069 | if (in_frames) | |
2070 | in_frames[i] = virt_to_mfn(vaddr); | |
2071 | ||
2072 | MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0); | |
2073 | set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY); | |
2074 | ||
2075 | if (out_frames) | |
2076 | out_frames[i] = virt_to_pfn(vaddr); | |
2077 | } | |
2078 | xen_mc_issue(0); | |
2079 | } | |
2080 | ||
2081 | /* | |
2082 | * Update the pfn-to-mfn mappings for a virtual address range, either to | |
2083 | * point to an array of mfns, or contiguously from a single starting | |
2084 | * mfn. | |
2085 | */ | |
2086 | static void xen_remap_exchanged_ptes(unsigned long vaddr, int order, | |
2087 | unsigned long *mfns, | |
2088 | unsigned long first_mfn) | |
2089 | { | |
2090 | unsigned i, limit; | |
2091 | unsigned long mfn; | |
2092 | ||
2093 | xen_mc_batch(); | |
2094 | ||
2095 | limit = 1u << order; | |
2096 | for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) { | |
2097 | struct multicall_space mcs; | |
2098 | unsigned flags; | |
2099 | ||
2100 | mcs = __xen_mc_entry(0); | |
2101 | if (mfns) | |
2102 | mfn = mfns[i]; | |
2103 | else | |
2104 | mfn = first_mfn + i; | |
2105 | ||
2106 | if (i < (limit - 1)) | |
2107 | flags = 0; | |
2108 | else { | |
2109 | if (order == 0) | |
2110 | flags = UVMF_INVLPG | UVMF_ALL; | |
2111 | else | |
2112 | flags = UVMF_TLB_FLUSH | UVMF_ALL; | |
2113 | } | |
2114 | ||
2115 | MULTI_update_va_mapping(mcs.mc, vaddr, | |
2116 | mfn_pte(mfn, PAGE_KERNEL), flags); | |
2117 | ||
2118 | set_phys_to_machine(virt_to_pfn(vaddr), mfn); | |
2119 | } | |
2120 | ||
2121 | xen_mc_issue(0); | |
2122 | } | |
2123 | ||
2124 | /* | |
2125 | * Perform the hypercall to exchange a region of our pfns to point to | |
2126 | * memory with the required contiguous alignment. Takes the pfns as | |
2127 | * input, and populates mfns as output. | |
2128 | * | |
2129 | * Returns a success code indicating whether the hypervisor was able to | |
2130 | * satisfy the request or not. | |
2131 | */ | |
2132 | static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in, | |
2133 | unsigned long *pfns_in, | |
2134 | unsigned long extents_out, | |
2135 | unsigned int order_out, | |
2136 | unsigned long *mfns_out, | |
2137 | unsigned int address_bits) | |
2138 | { | |
2139 | long rc; | |
2140 | int success; | |
2141 | ||
2142 | struct xen_memory_exchange exchange = { | |
2143 | .in = { | |
2144 | .nr_extents = extents_in, | |
2145 | .extent_order = order_in, | |
2146 | .extent_start = pfns_in, | |
2147 | .domid = DOMID_SELF | |
2148 | }, | |
2149 | .out = { | |
2150 | .nr_extents = extents_out, | |
2151 | .extent_order = order_out, | |
2152 | .extent_start = mfns_out, | |
2153 | .address_bits = address_bits, | |
2154 | .domid = DOMID_SELF | |
2155 | } | |
2156 | }; | |
2157 | ||
2158 | BUG_ON(extents_in << order_in != extents_out << order_out); | |
2159 | ||
2160 | rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange); | |
2161 | success = (exchange.nr_exchanged == extents_in); | |
2162 | ||
2163 | BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0))); | |
2164 | BUG_ON(success && (rc != 0)); | |
2165 | ||
2166 | return success; | |
2167 | } | |
2168 | ||
2169 | int xen_create_contiguous_region(unsigned long vstart, unsigned int order, | |
2170 | unsigned int address_bits) | |
2171 | { | |
2172 | unsigned long *in_frames = discontig_frames, out_frame; | |
2173 | unsigned long flags; | |
2174 | int success; | |
2175 | ||
2176 | /* | |
2177 | * Currently an auto-translated guest will not perform I/O, nor will | |
2178 | * it require PAE page directories below 4GB. Therefore any calls to | |
2179 | * this function are redundant and can be ignored. | |
2180 | */ | |
2181 | ||
2182 | if (xen_feature(XENFEAT_auto_translated_physmap)) | |
2183 | return 0; | |
2184 | ||
2185 | if (unlikely(order > MAX_CONTIG_ORDER)) | |
2186 | return -ENOMEM; | |
2187 | ||
2188 | memset((void *) vstart, 0, PAGE_SIZE << order); | |
2189 | ||
08bbc9da AN |
2190 | spin_lock_irqsave(&xen_reservation_lock, flags); |
2191 | ||
2192 | /* 1. Zap current PTEs, remembering MFNs. */ | |
2193 | xen_zap_pfn_range(vstart, order, in_frames, NULL); | |
2194 | ||
2195 | /* 2. Get a new contiguous memory extent. */ | |
2196 | out_frame = virt_to_pfn(vstart); | |
2197 | success = xen_exchange_memory(1UL << order, 0, in_frames, | |
2198 | 1, order, &out_frame, | |
2199 | address_bits); | |
2200 | ||
2201 | /* 3. Map the new extent in place of old pages. */ | |
2202 | if (success) | |
2203 | xen_remap_exchanged_ptes(vstart, order, NULL, out_frame); | |
2204 | else | |
2205 | xen_remap_exchanged_ptes(vstart, order, in_frames, 0); | |
2206 | ||
2207 | spin_unlock_irqrestore(&xen_reservation_lock, flags); | |
2208 | ||
2209 | return success ? 0 : -ENOMEM; | |
2210 | } | |
2211 | EXPORT_SYMBOL_GPL(xen_create_contiguous_region); | |
2212 | ||
2213 | void xen_destroy_contiguous_region(unsigned long vstart, unsigned int order) | |
2214 | { | |
2215 | unsigned long *out_frames = discontig_frames, in_frame; | |
2216 | unsigned long flags; | |
2217 | int success; | |
2218 | ||
2219 | if (xen_feature(XENFEAT_auto_translated_physmap)) | |
2220 | return; | |
2221 | ||
2222 | if (unlikely(order > MAX_CONTIG_ORDER)) | |
2223 | return; | |
2224 | ||
2225 | memset((void *) vstart, 0, PAGE_SIZE << order); | |
2226 | ||
08bbc9da AN |
2227 | spin_lock_irqsave(&xen_reservation_lock, flags); |
2228 | ||
2229 | /* 1. Find start MFN of contiguous extent. */ | |
2230 | in_frame = virt_to_mfn(vstart); | |
2231 | ||
2232 | /* 2. Zap current PTEs. */ | |
2233 | xen_zap_pfn_range(vstart, order, NULL, out_frames); | |
2234 | ||
2235 | /* 3. Do the exchange for non-contiguous MFNs. */ | |
2236 | success = xen_exchange_memory(1, order, &in_frame, 1UL << order, | |
2237 | 0, out_frames, 0); | |
2238 | ||
2239 | /* 4. Map new pages in place of old pages. */ | |
2240 | if (success) | |
2241 | xen_remap_exchanged_ptes(vstart, order, out_frames, 0); | |
2242 | else | |
2243 | xen_remap_exchanged_ptes(vstart, order, NULL, in_frame); | |
2244 | ||
2245 | spin_unlock_irqrestore(&xen_reservation_lock, flags); | |
030cb6c0 | 2246 | } |
08bbc9da | 2247 | EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region); |
319f3ba5 | 2248 | |
ca65f9fc | 2249 | #ifdef CONFIG_XEN_PVHVM |
59151001 SS |
2250 | static void xen_hvm_exit_mmap(struct mm_struct *mm) |
2251 | { | |
2252 | struct xen_hvm_pagetable_dying a; | |
2253 | int rc; | |
2254 | ||
2255 | a.domid = DOMID_SELF; | |
2256 | a.gpa = __pa(mm->pgd); | |
2257 | rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a); | |
2258 | WARN_ON_ONCE(rc < 0); | |
2259 | } | |
2260 | ||
2261 | static int is_pagetable_dying_supported(void) | |
2262 | { | |
2263 | struct xen_hvm_pagetable_dying a; | |
2264 | int rc = 0; | |
2265 | ||
2266 | a.domid = DOMID_SELF; | |
2267 | a.gpa = 0x00; | |
2268 | rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a); | |
2269 | if (rc < 0) { | |
2270 | printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n"); | |
2271 | return 0; | |
2272 | } | |
2273 | return 1; | |
2274 | } | |
2275 | ||
2276 | void __init xen_hvm_init_mmu_ops(void) | |
2277 | { | |
2278 | if (is_pagetable_dying_supported()) | |
2279 | pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap; | |
2280 | } | |
ca65f9fc | 2281 | #endif |
59151001 | 2282 | |
994025ca JF |
2283 | #ifdef CONFIG_XEN_DEBUG_FS |
2284 | ||
2285 | static struct dentry *d_mmu_debug; | |
2286 | ||
2287 | static int __init xen_mmu_debugfs(void) | |
2288 | { | |
2289 | struct dentry *d_xen = xen_init_debugfs(); | |
2290 | ||
2291 | if (d_xen == NULL) | |
2292 | return -ENOMEM; | |
2293 | ||
2294 | d_mmu_debug = debugfs_create_dir("mmu", d_xen); | |
2295 | ||
2296 | debugfs_create_u8("zero_stats", 0644, d_mmu_debug, &zero_stats); | |
2297 | ||
2298 | debugfs_create_u32("pgd_update", 0444, d_mmu_debug, &mmu_stats.pgd_update); | |
2299 | debugfs_create_u32("pgd_update_pinned", 0444, d_mmu_debug, | |
2300 | &mmu_stats.pgd_update_pinned); | |
2301 | debugfs_create_u32("pgd_update_batched", 0444, d_mmu_debug, | |
2302 | &mmu_stats.pgd_update_pinned); | |
2303 | ||
2304 | debugfs_create_u32("pud_update", 0444, d_mmu_debug, &mmu_stats.pud_update); | |
2305 | debugfs_create_u32("pud_update_pinned", 0444, d_mmu_debug, | |
2306 | &mmu_stats.pud_update_pinned); | |
2307 | debugfs_create_u32("pud_update_batched", 0444, d_mmu_debug, | |
2308 | &mmu_stats.pud_update_pinned); | |
2309 | ||
2310 | debugfs_create_u32("pmd_update", 0444, d_mmu_debug, &mmu_stats.pmd_update); | |
2311 | debugfs_create_u32("pmd_update_pinned", 0444, d_mmu_debug, | |
2312 | &mmu_stats.pmd_update_pinned); | |
2313 | debugfs_create_u32("pmd_update_batched", 0444, d_mmu_debug, | |
2314 | &mmu_stats.pmd_update_pinned); | |
2315 | ||
2316 | debugfs_create_u32("pte_update", 0444, d_mmu_debug, &mmu_stats.pte_update); | |
2317 | // debugfs_create_u32("pte_update_pinned", 0444, d_mmu_debug, | |
2318 | // &mmu_stats.pte_update_pinned); | |
2319 | debugfs_create_u32("pte_update_batched", 0444, d_mmu_debug, | |
2320 | &mmu_stats.pte_update_pinned); | |
2321 | ||
2322 | debugfs_create_u32("mmu_update", 0444, d_mmu_debug, &mmu_stats.mmu_update); | |
2323 | debugfs_create_u32("mmu_update_extended", 0444, d_mmu_debug, | |
2324 | &mmu_stats.mmu_update_extended); | |
2325 | xen_debugfs_create_u32_array("mmu_update_histo", 0444, d_mmu_debug, | |
2326 | mmu_stats.mmu_update_histo, 20); | |
2327 | ||
2328 | debugfs_create_u32("set_pte_at", 0444, d_mmu_debug, &mmu_stats.set_pte_at); | |
2329 | debugfs_create_u32("set_pte_at_batched", 0444, d_mmu_debug, | |
2330 | &mmu_stats.set_pte_at_batched); | |
2331 | debugfs_create_u32("set_pte_at_current", 0444, d_mmu_debug, | |
2332 | &mmu_stats.set_pte_at_current); | |
2333 | debugfs_create_u32("set_pte_at_kernel", 0444, d_mmu_debug, | |
2334 | &mmu_stats.set_pte_at_kernel); | |
2335 | ||
2336 | debugfs_create_u32("prot_commit", 0444, d_mmu_debug, &mmu_stats.prot_commit); | |
2337 | debugfs_create_u32("prot_commit_batched", 0444, d_mmu_debug, | |
2338 | &mmu_stats.prot_commit_batched); | |
2339 | ||
2340 | return 0; | |
2341 | } | |
2342 | fs_initcall(xen_mmu_debugfs); | |
2343 | ||
2344 | #endif /* CONFIG_XEN_DEBUG_FS */ |