]>
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> |
a9ce6bc1 | 48 | #include <linux/memblock.h> |
2222e71b | 49 | #include <linux/seq_file.h> |
34b6f01a | 50 | #include <linux/crash_dump.h> |
3b827c1b | 51 | |
84708807 JF |
52 | #include <trace/events/xen.h> |
53 | ||
3b827c1b JF |
54 | #include <asm/pgtable.h> |
55 | #include <asm/tlbflush.h> | |
5deb30d1 | 56 | #include <asm/fixmap.h> |
3b827c1b | 57 | #include <asm/mmu_context.h> |
319f3ba5 | 58 | #include <asm/setup.h> |
f4f97b3e | 59 | #include <asm/paravirt.h> |
7347b408 | 60 | #include <asm/e820.h> |
cbcd79c2 | 61 | #include <asm/linkage.h> |
08bbc9da | 62 | #include <asm/page.h> |
fef5ba79 | 63 | #include <asm/init.h> |
41f2e477 | 64 | #include <asm/pat.h> |
900cba88 | 65 | #include <asm/smp.h> |
3b827c1b JF |
66 | |
67 | #include <asm/xen/hypercall.h> | |
f4f97b3e | 68 | #include <asm/xen/hypervisor.h> |
3b827c1b | 69 | |
c0011dbf | 70 | #include <xen/xen.h> |
3b827c1b JF |
71 | #include <xen/page.h> |
72 | #include <xen/interface/xen.h> | |
59151001 | 73 | #include <xen/interface/hvm/hvm_op.h> |
319f3ba5 | 74 | #include <xen/interface/version.h> |
c0011dbf | 75 | #include <xen/interface/memory.h> |
319f3ba5 | 76 | #include <xen/hvc-console.h> |
3b827c1b | 77 | |
f4f97b3e | 78 | #include "multicalls.h" |
3b827c1b | 79 | #include "mmu.h" |
994025ca JF |
80 | #include "debugfs.h" |
81 | ||
19001c8c AN |
82 | /* |
83 | * Protects atomic reservation decrease/increase against concurrent increases. | |
06f521d5 | 84 | * Also protects non-atomic updates of current_pages and balloon lists. |
19001c8c AN |
85 | */ |
86 | DEFINE_SPINLOCK(xen_reservation_lock); | |
87 | ||
caaf9ecf | 88 | #ifdef CONFIG_X86_32 |
319f3ba5 JF |
89 | /* |
90 | * Identity map, in addition to plain kernel map. This needs to be | |
91 | * large enough to allocate page table pages to allocate the rest. | |
92 | * Each page can map 2MB. | |
93 | */ | |
764f0138 JF |
94 | #define LEVEL1_IDENT_ENTRIES (PTRS_PER_PTE * 4) |
95 | static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES); | |
caaf9ecf | 96 | #endif |
319f3ba5 JF |
97 | #ifdef CONFIG_X86_64 |
98 | /* l3 pud for userspace vsyscall mapping */ | |
99 | static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss; | |
100 | #endif /* CONFIG_X86_64 */ | |
101 | ||
102 | /* | |
103 | * Note about cr3 (pagetable base) values: | |
104 | * | |
105 | * xen_cr3 contains the current logical cr3 value; it contains the | |
106 | * last set cr3. This may not be the current effective cr3, because | |
107 | * its update may be being lazily deferred. However, a vcpu looking | |
108 | * at its own cr3 can use this value knowing that it everything will | |
109 | * be self-consistent. | |
110 | * | |
111 | * xen_current_cr3 contains the actual vcpu cr3; it is set once the | |
112 | * hypercall to set the vcpu cr3 is complete (so it may be a little | |
113 | * out of date, but it will never be set early). If one vcpu is | |
114 | * looking at another vcpu's cr3 value, it should use this variable. | |
115 | */ | |
116 | DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */ | |
117 | DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */ | |
118 | ||
119 | ||
d6182fbf JF |
120 | /* |
121 | * Just beyond the highest usermode address. STACK_TOP_MAX has a | |
122 | * redzone above it, so round it up to a PGD boundary. | |
123 | */ | |
124 | #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK) | |
125 | ||
9976b39b JF |
126 | unsigned long arbitrary_virt_to_mfn(void *vaddr) |
127 | { | |
128 | xmaddr_t maddr = arbitrary_virt_to_machine(vaddr); | |
129 | ||
130 | return PFN_DOWN(maddr.maddr); | |
131 | } | |
132 | ||
ce803e70 | 133 | xmaddr_t arbitrary_virt_to_machine(void *vaddr) |
3b827c1b | 134 | { |
ce803e70 | 135 | unsigned long address = (unsigned long)vaddr; |
da7bfc50 | 136 | unsigned int level; |
9f32d21c CL |
137 | pte_t *pte; |
138 | unsigned offset; | |
3b827c1b | 139 | |
9f32d21c CL |
140 | /* |
141 | * if the PFN is in the linear mapped vaddr range, we can just use | |
142 | * the (quick) virt_to_machine() p2m lookup | |
143 | */ | |
144 | if (virt_addr_valid(vaddr)) | |
145 | return virt_to_machine(vaddr); | |
146 | ||
147 | /* otherwise we have to do a (slower) full page-table walk */ | |
3b827c1b | 148 | |
9f32d21c CL |
149 | pte = lookup_address(address, &level); |
150 | BUG_ON(pte == NULL); | |
151 | offset = address & ~PAGE_MASK; | |
ebd879e3 | 152 | return XMADDR(((phys_addr_t)pte_mfn(*pte) << PAGE_SHIFT) + offset); |
3b827c1b | 153 | } |
de23be5f | 154 | EXPORT_SYMBOL_GPL(arbitrary_virt_to_machine); |
3b827c1b JF |
155 | |
156 | void make_lowmem_page_readonly(void *vaddr) | |
157 | { | |
158 | pte_t *pte, ptev; | |
159 | unsigned long address = (unsigned long)vaddr; | |
da7bfc50 | 160 | unsigned int level; |
3b827c1b | 161 | |
f0646e43 | 162 | pte = lookup_address(address, &level); |
fef5ba79 JF |
163 | if (pte == NULL) |
164 | return; /* vaddr missing */ | |
3b827c1b JF |
165 | |
166 | ptev = pte_wrprotect(*pte); | |
167 | ||
168 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
169 | BUG(); | |
170 | } | |
171 | ||
172 | void make_lowmem_page_readwrite(void *vaddr) | |
173 | { | |
174 | pte_t *pte, ptev; | |
175 | unsigned long address = (unsigned long)vaddr; | |
da7bfc50 | 176 | unsigned int level; |
3b827c1b | 177 | |
f0646e43 | 178 | pte = lookup_address(address, &level); |
fef5ba79 JF |
179 | if (pte == NULL) |
180 | return; /* vaddr missing */ | |
3b827c1b JF |
181 | |
182 | ptev = pte_mkwrite(*pte); | |
183 | ||
184 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
185 | BUG(); | |
186 | } | |
187 | ||
188 | ||
7708ad64 | 189 | static bool xen_page_pinned(void *ptr) |
e2426cf8 JF |
190 | { |
191 | struct page *page = virt_to_page(ptr); | |
192 | ||
193 | return PagePinned(page); | |
194 | } | |
195 | ||
eba3ff8b | 196 | void xen_set_domain_pte(pte_t *ptep, pte_t pteval, unsigned domid) |
c0011dbf JF |
197 | { |
198 | struct multicall_space mcs; | |
199 | struct mmu_update *u; | |
200 | ||
84708807 JF |
201 | trace_xen_mmu_set_domain_pte(ptep, pteval, domid); |
202 | ||
c0011dbf JF |
203 | mcs = xen_mc_entry(sizeof(*u)); |
204 | u = mcs.args; | |
205 | ||
206 | /* ptep might be kmapped when using 32-bit HIGHPTE */ | |
d5108316 | 207 | u->ptr = virt_to_machine(ptep).maddr; |
c0011dbf JF |
208 | u->val = pte_val_ma(pteval); |
209 | ||
eba3ff8b | 210 | MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, domid); |
c0011dbf JF |
211 | |
212 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
213 | } | |
eba3ff8b JF |
214 | EXPORT_SYMBOL_GPL(xen_set_domain_pte); |
215 | ||
7708ad64 | 216 | static void xen_extend_mmu_update(const struct mmu_update *update) |
3b827c1b | 217 | { |
d66bf8fc JF |
218 | struct multicall_space mcs; |
219 | struct mmu_update *u; | |
3b827c1b | 220 | |
400d3494 JF |
221 | mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u)); |
222 | ||
994025ca | 223 | if (mcs.mc != NULL) { |
400d3494 | 224 | mcs.mc->args[1]++; |
994025ca | 225 | } else { |
400d3494 JF |
226 | mcs = __xen_mc_entry(sizeof(*u)); |
227 | MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); | |
228 | } | |
d66bf8fc | 229 | |
d66bf8fc | 230 | u = mcs.args; |
400d3494 JF |
231 | *u = *update; |
232 | } | |
233 | ||
dcf7435c JF |
234 | static void xen_extend_mmuext_op(const struct mmuext_op *op) |
235 | { | |
236 | struct multicall_space mcs; | |
237 | struct mmuext_op *u; | |
238 | ||
239 | mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u)); | |
240 | ||
241 | if (mcs.mc != NULL) { | |
242 | mcs.mc->args[1]++; | |
243 | } else { | |
244 | mcs = __xen_mc_entry(sizeof(*u)); | |
245 | MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); | |
246 | } | |
247 | ||
248 | u = mcs.args; | |
249 | *u = *op; | |
250 | } | |
251 | ||
4c13629f | 252 | static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val) |
400d3494 JF |
253 | { |
254 | struct mmu_update u; | |
255 | ||
256 | preempt_disable(); | |
257 | ||
258 | xen_mc_batch(); | |
259 | ||
ce803e70 JF |
260 | /* ptr may be ioremapped for 64-bit pagetable setup */ |
261 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
400d3494 | 262 | u.val = pmd_val_ma(val); |
7708ad64 | 263 | xen_extend_mmu_update(&u); |
d66bf8fc JF |
264 | |
265 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
266 | ||
267 | preempt_enable(); | |
3b827c1b JF |
268 | } |
269 | ||
4c13629f | 270 | static void xen_set_pmd(pmd_t *ptr, pmd_t val) |
e2426cf8 | 271 | { |
84708807 JF |
272 | trace_xen_mmu_set_pmd(ptr, val); |
273 | ||
e2426cf8 JF |
274 | /* If page is not pinned, we can just update the entry |
275 | directly */ | |
7708ad64 | 276 | if (!xen_page_pinned(ptr)) { |
e2426cf8 JF |
277 | *ptr = val; |
278 | return; | |
279 | } | |
280 | ||
281 | xen_set_pmd_hyper(ptr, val); | |
282 | } | |
283 | ||
3b827c1b JF |
284 | /* |
285 | * Associate a virtual page frame with a given physical page frame | |
286 | * and protection flags for that frame. | |
287 | */ | |
288 | void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags) | |
289 | { | |
836fe2f2 | 290 | set_pte_vaddr(vaddr, mfn_pte(mfn, flags)); |
3b827c1b JF |
291 | } |
292 | ||
4a35c13c | 293 | static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval) |
3b827c1b | 294 | { |
4a35c13c | 295 | struct mmu_update u; |
c0011dbf | 296 | |
4a35c13c JF |
297 | if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU) |
298 | return false; | |
994025ca | 299 | |
4a35c13c | 300 | xen_mc_batch(); |
d66bf8fc | 301 | |
4a35c13c JF |
302 | u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE; |
303 | u.val = pte_val_ma(pteval); | |
304 | xen_extend_mmu_update(&u); | |
a99ac5e8 | 305 | |
4a35c13c | 306 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
2bd50036 | 307 | |
4a35c13c JF |
308 | return true; |
309 | } | |
310 | ||
84708807 | 311 | static inline void __xen_set_pte(pte_t *ptep, pte_t pteval) |
4a35c13c | 312 | { |
d095d43e DV |
313 | if (!xen_batched_set_pte(ptep, pteval)) { |
314 | /* | |
315 | * Could call native_set_pte() here and trap and | |
316 | * emulate the PTE write but with 32-bit guests this | |
317 | * needs two traps (one for each of the two 32-bit | |
318 | * words in the PTE) so do one hypercall directly | |
319 | * instead. | |
320 | */ | |
321 | struct mmu_update u; | |
322 | ||
323 | u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE; | |
324 | u.val = pte_val_ma(pteval); | |
325 | HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF); | |
326 | } | |
3b827c1b JF |
327 | } |
328 | ||
84708807 JF |
329 | static void xen_set_pte(pte_t *ptep, pte_t pteval) |
330 | { | |
331 | trace_xen_mmu_set_pte(ptep, pteval); | |
332 | __xen_set_pte(ptep, pteval); | |
333 | } | |
334 | ||
4c13629f | 335 | static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr, |
4a35c13c JF |
336 | pte_t *ptep, pte_t pteval) |
337 | { | |
84708807 JF |
338 | trace_xen_mmu_set_pte_at(mm, addr, ptep, pteval); |
339 | __xen_set_pte(ptep, pteval); | |
3b827c1b JF |
340 | } |
341 | ||
f63c2f24 T |
342 | pte_t xen_ptep_modify_prot_start(struct mm_struct *mm, |
343 | unsigned long addr, pte_t *ptep) | |
947a69c9 | 344 | { |
e57778a1 | 345 | /* Just return the pte as-is. We preserve the bits on commit */ |
84708807 | 346 | trace_xen_mmu_ptep_modify_prot_start(mm, addr, ptep, *ptep); |
e57778a1 JF |
347 | return *ptep; |
348 | } | |
349 | ||
350 | void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr, | |
351 | pte_t *ptep, pte_t pte) | |
352 | { | |
400d3494 | 353 | struct mmu_update u; |
e57778a1 | 354 | |
84708807 | 355 | trace_xen_mmu_ptep_modify_prot_commit(mm, addr, ptep, pte); |
400d3494 | 356 | xen_mc_batch(); |
947a69c9 | 357 | |
d5108316 | 358 | u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD; |
400d3494 | 359 | u.val = pte_val_ma(pte); |
7708ad64 | 360 | xen_extend_mmu_update(&u); |
947a69c9 | 361 | |
e57778a1 | 362 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
947a69c9 JF |
363 | } |
364 | ||
ebb9cfe2 JF |
365 | /* Assume pteval_t is equivalent to all the other *val_t types. */ |
366 | static pteval_t pte_mfn_to_pfn(pteval_t val) | |
947a69c9 | 367 | { |
5926f87f | 368 | if (val & _PAGE_PRESENT) { |
59438c9f | 369 | unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; |
b7e5ffe5 KRW |
370 | unsigned long pfn = mfn_to_pfn(mfn); |
371 | ||
77be1fab | 372 | pteval_t flags = val & PTE_FLAGS_MASK; |
b7e5ffe5 KRW |
373 | if (unlikely(pfn == ~0)) |
374 | val = flags & ~_PAGE_PRESENT; | |
375 | else | |
376 | val = ((pteval_t)pfn << PAGE_SHIFT) | flags; | |
ebb9cfe2 | 377 | } |
947a69c9 | 378 | |
ebb9cfe2 | 379 | return val; |
947a69c9 JF |
380 | } |
381 | ||
ebb9cfe2 | 382 | static pteval_t pte_pfn_to_mfn(pteval_t val) |
947a69c9 | 383 | { |
5926f87f | 384 | if (val & _PAGE_PRESENT) { |
59438c9f | 385 | unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; |
77be1fab | 386 | pteval_t flags = val & PTE_FLAGS_MASK; |
fb38923e | 387 | unsigned long mfn; |
cfd8951e | 388 | |
fb38923e KRW |
389 | if (!xen_feature(XENFEAT_auto_translated_physmap)) |
390 | mfn = get_phys_to_machine(pfn); | |
391 | else | |
392 | mfn = pfn; | |
cfd8951e JF |
393 | /* |
394 | * If there's no mfn for the pfn, then just create an | |
395 | * empty non-present pte. Unfortunately this loses | |
396 | * information about the original pfn, so | |
397 | * pte_mfn_to_pfn is asymmetric. | |
398 | */ | |
399 | if (unlikely(mfn == INVALID_P2M_ENTRY)) { | |
400 | mfn = 0; | |
401 | flags = 0; | |
7f2f8822 DV |
402 | } else |
403 | mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT); | |
cfd8951e | 404 | val = ((pteval_t)mfn << PAGE_SHIFT) | flags; |
947a69c9 JF |
405 | } |
406 | ||
ebb9cfe2 | 407 | return val; |
947a69c9 JF |
408 | } |
409 | ||
a2e7f0e3 | 410 | __visible pteval_t xen_pte_val(pte_t pte) |
947a69c9 | 411 | { |
41f2e477 | 412 | pteval_t pteval = pte.pte; |
47591df5 | 413 | |
41f2e477 | 414 | return pte_mfn_to_pfn(pteval); |
947a69c9 | 415 | } |
da5de7c2 | 416 | PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val); |
947a69c9 | 417 | |
a2e7f0e3 | 418 | __visible pgdval_t xen_pgd_val(pgd_t pgd) |
947a69c9 | 419 | { |
ebb9cfe2 | 420 | return pte_mfn_to_pfn(pgd.pgd); |
947a69c9 | 421 | } |
da5de7c2 | 422 | PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val); |
947a69c9 | 423 | |
a2e7f0e3 | 424 | __visible pte_t xen_make_pte(pteval_t pte) |
947a69c9 | 425 | { |
7f2f8822 | 426 | pte = pte_pfn_to_mfn(pte); |
c0011dbf | 427 | |
ebb9cfe2 | 428 | return native_make_pte(pte); |
947a69c9 | 429 | } |
da5de7c2 | 430 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte); |
947a69c9 | 431 | |
a2e7f0e3 | 432 | __visible pgd_t xen_make_pgd(pgdval_t pgd) |
947a69c9 | 433 | { |
ebb9cfe2 JF |
434 | pgd = pte_pfn_to_mfn(pgd); |
435 | return native_make_pgd(pgd); | |
947a69c9 | 436 | } |
da5de7c2 | 437 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd); |
947a69c9 | 438 | |
a2e7f0e3 | 439 | __visible pmdval_t xen_pmd_val(pmd_t pmd) |
947a69c9 | 440 | { |
ebb9cfe2 | 441 | return pte_mfn_to_pfn(pmd.pmd); |
947a69c9 | 442 | } |
da5de7c2 | 443 | PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val); |
28499143 | 444 | |
4c13629f | 445 | static void xen_set_pud_hyper(pud_t *ptr, pud_t val) |
f4f97b3e | 446 | { |
400d3494 | 447 | struct mmu_update u; |
f4f97b3e | 448 | |
d66bf8fc JF |
449 | preempt_disable(); |
450 | ||
400d3494 JF |
451 | xen_mc_batch(); |
452 | ||
ce803e70 JF |
453 | /* ptr may be ioremapped for 64-bit pagetable setup */ |
454 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
400d3494 | 455 | u.val = pud_val_ma(val); |
7708ad64 | 456 | xen_extend_mmu_update(&u); |
d66bf8fc JF |
457 | |
458 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
459 | ||
460 | preempt_enable(); | |
f4f97b3e JF |
461 | } |
462 | ||
4c13629f | 463 | static void xen_set_pud(pud_t *ptr, pud_t val) |
e2426cf8 | 464 | { |
84708807 JF |
465 | trace_xen_mmu_set_pud(ptr, val); |
466 | ||
e2426cf8 JF |
467 | /* If page is not pinned, we can just update the entry |
468 | directly */ | |
7708ad64 | 469 | if (!xen_page_pinned(ptr)) { |
e2426cf8 JF |
470 | *ptr = val; |
471 | return; | |
472 | } | |
473 | ||
474 | xen_set_pud_hyper(ptr, val); | |
475 | } | |
476 | ||
f6e58732 | 477 | #ifdef CONFIG_X86_PAE |
4c13629f | 478 | static void xen_set_pte_atomic(pte_t *ptep, pte_t pte) |
3b827c1b | 479 | { |
84708807 | 480 | trace_xen_mmu_set_pte_atomic(ptep, pte); |
f6e58732 | 481 | set_64bit((u64 *)ptep, native_pte_val(pte)); |
3b827c1b JF |
482 | } |
483 | ||
4c13629f | 484 | static void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep) |
3b827c1b | 485 | { |
84708807 | 486 | trace_xen_mmu_pte_clear(mm, addr, ptep); |
4a35c13c JF |
487 | if (!xen_batched_set_pte(ptep, native_make_pte(0))) |
488 | native_pte_clear(mm, addr, ptep); | |
3b827c1b JF |
489 | } |
490 | ||
4c13629f | 491 | static void xen_pmd_clear(pmd_t *pmdp) |
3b827c1b | 492 | { |
84708807 | 493 | trace_xen_mmu_pmd_clear(pmdp); |
e2426cf8 | 494 | set_pmd(pmdp, __pmd(0)); |
3b827c1b | 495 | } |
f6e58732 | 496 | #endif /* CONFIG_X86_PAE */ |
3b827c1b | 497 | |
a2e7f0e3 | 498 | __visible pmd_t xen_make_pmd(pmdval_t pmd) |
3b827c1b | 499 | { |
ebb9cfe2 | 500 | pmd = pte_pfn_to_mfn(pmd); |
947a69c9 | 501 | return native_make_pmd(pmd); |
3b827c1b | 502 | } |
da5de7c2 | 503 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd); |
3b827c1b | 504 | |
f6e58732 | 505 | #if PAGETABLE_LEVELS == 4 |
a2e7f0e3 | 506 | __visible pudval_t xen_pud_val(pud_t pud) |
f6e58732 JF |
507 | { |
508 | return pte_mfn_to_pfn(pud.pud); | |
509 | } | |
da5de7c2 | 510 | PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val); |
f6e58732 | 511 | |
a2e7f0e3 | 512 | __visible pud_t xen_make_pud(pudval_t pud) |
f6e58732 JF |
513 | { |
514 | pud = pte_pfn_to_mfn(pud); | |
515 | ||
516 | return native_make_pud(pud); | |
517 | } | |
da5de7c2 | 518 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud); |
f6e58732 | 519 | |
4c13629f | 520 | static pgd_t *xen_get_user_pgd(pgd_t *pgd) |
f6e58732 | 521 | { |
d6182fbf JF |
522 | pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK); |
523 | unsigned offset = pgd - pgd_page; | |
524 | pgd_t *user_ptr = NULL; | |
f6e58732 | 525 | |
d6182fbf JF |
526 | if (offset < pgd_index(USER_LIMIT)) { |
527 | struct page *page = virt_to_page(pgd_page); | |
528 | user_ptr = (pgd_t *)page->private; | |
529 | if (user_ptr) | |
530 | user_ptr += offset; | |
531 | } | |
f6e58732 | 532 | |
d6182fbf JF |
533 | return user_ptr; |
534 | } | |
535 | ||
536 | static void __xen_set_pgd_hyper(pgd_t *ptr, pgd_t val) | |
537 | { | |
538 | struct mmu_update u; | |
f6e58732 JF |
539 | |
540 | u.ptr = virt_to_machine(ptr).maddr; | |
541 | u.val = pgd_val_ma(val); | |
7708ad64 | 542 | xen_extend_mmu_update(&u); |
d6182fbf JF |
543 | } |
544 | ||
545 | /* | |
546 | * Raw hypercall-based set_pgd, intended for in early boot before | |
547 | * there's a page structure. This implies: | |
548 | * 1. The only existing pagetable is the kernel's | |
549 | * 2. It is always pinned | |
550 | * 3. It has no user pagetable attached to it | |
551 | */ | |
4c13629f | 552 | static void __init xen_set_pgd_hyper(pgd_t *ptr, pgd_t val) |
d6182fbf JF |
553 | { |
554 | preempt_disable(); | |
555 | ||
556 | xen_mc_batch(); | |
557 | ||
558 | __xen_set_pgd_hyper(ptr, val); | |
f6e58732 JF |
559 | |
560 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
561 | ||
562 | preempt_enable(); | |
563 | } | |
564 | ||
4c13629f | 565 | static void xen_set_pgd(pgd_t *ptr, pgd_t val) |
f6e58732 | 566 | { |
d6182fbf JF |
567 | pgd_t *user_ptr = xen_get_user_pgd(ptr); |
568 | ||
84708807 JF |
569 | trace_xen_mmu_set_pgd(ptr, user_ptr, val); |
570 | ||
f6e58732 JF |
571 | /* If page is not pinned, we can just update the entry |
572 | directly */ | |
7708ad64 | 573 | if (!xen_page_pinned(ptr)) { |
f6e58732 | 574 | *ptr = val; |
d6182fbf | 575 | if (user_ptr) { |
7708ad64 | 576 | WARN_ON(xen_page_pinned(user_ptr)); |
d6182fbf JF |
577 | *user_ptr = val; |
578 | } | |
f6e58732 JF |
579 | return; |
580 | } | |
581 | ||
d6182fbf JF |
582 | /* If it's pinned, then we can at least batch the kernel and |
583 | user updates together. */ | |
584 | xen_mc_batch(); | |
585 | ||
586 | __xen_set_pgd_hyper(ptr, val); | |
587 | if (user_ptr) | |
588 | __xen_set_pgd_hyper(user_ptr, val); | |
589 | ||
590 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
f6e58732 JF |
591 | } |
592 | #endif /* PAGETABLE_LEVELS == 4 */ | |
593 | ||
f4f97b3e | 594 | /* |
5deb30d1 JF |
595 | * (Yet another) pagetable walker. This one is intended for pinning a |
596 | * pagetable. This means that it walks a pagetable and calls the | |
597 | * callback function on each page it finds making up the page table, | |
598 | * at every level. It walks the entire pagetable, but it only bothers | |
599 | * pinning pte pages which are below limit. In the normal case this | |
600 | * will be STACK_TOP_MAX, but at boot we need to pin up to | |
601 | * FIXADDR_TOP. | |
602 | * | |
603 | * For 32-bit the important bit is that we don't pin beyond there, | |
604 | * because then we start getting into Xen's ptes. | |
605 | * | |
606 | * For 64-bit, we must skip the Xen hole in the middle of the address | |
607 | * space, just after the big x86-64 virtual hole. | |
608 | */ | |
86bbc2c2 IC |
609 | static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd, |
610 | int (*func)(struct mm_struct *mm, struct page *, | |
611 | enum pt_level), | |
612 | unsigned long limit) | |
3b827c1b | 613 | { |
f4f97b3e | 614 | int flush = 0; |
5deb30d1 JF |
615 | unsigned hole_low, hole_high; |
616 | unsigned pgdidx_limit, pudidx_limit, pmdidx_limit; | |
617 | unsigned pgdidx, pudidx, pmdidx; | |
f4f97b3e | 618 | |
5deb30d1 JF |
619 | /* The limit is the last byte to be touched */ |
620 | limit--; | |
621 | BUG_ON(limit >= FIXADDR_TOP); | |
3b827c1b JF |
622 | |
623 | if (xen_feature(XENFEAT_auto_translated_physmap)) | |
f4f97b3e JF |
624 | return 0; |
625 | ||
5deb30d1 JF |
626 | /* |
627 | * 64-bit has a great big hole in the middle of the address | |
628 | * space, which contains the Xen mappings. On 32-bit these | |
629 | * will end up making a zero-sized hole and so is a no-op. | |
630 | */ | |
d6182fbf | 631 | hole_low = pgd_index(USER_LIMIT); |
5deb30d1 JF |
632 | hole_high = pgd_index(PAGE_OFFSET); |
633 | ||
634 | pgdidx_limit = pgd_index(limit); | |
635 | #if PTRS_PER_PUD > 1 | |
636 | pudidx_limit = pud_index(limit); | |
637 | #else | |
638 | pudidx_limit = 0; | |
639 | #endif | |
640 | #if PTRS_PER_PMD > 1 | |
641 | pmdidx_limit = pmd_index(limit); | |
642 | #else | |
643 | pmdidx_limit = 0; | |
644 | #endif | |
645 | ||
5deb30d1 | 646 | for (pgdidx = 0; pgdidx <= pgdidx_limit; pgdidx++) { |
f4f97b3e | 647 | pud_t *pud; |
3b827c1b | 648 | |
5deb30d1 JF |
649 | if (pgdidx >= hole_low && pgdidx < hole_high) |
650 | continue; | |
f4f97b3e | 651 | |
5deb30d1 | 652 | if (!pgd_val(pgd[pgdidx])) |
3b827c1b | 653 | continue; |
f4f97b3e | 654 | |
5deb30d1 | 655 | pud = pud_offset(&pgd[pgdidx], 0); |
3b827c1b JF |
656 | |
657 | if (PTRS_PER_PUD > 1) /* not folded */ | |
eefb47f6 | 658 | flush |= (*func)(mm, virt_to_page(pud), PT_PUD); |
f4f97b3e | 659 | |
5deb30d1 | 660 | for (pudidx = 0; pudidx < PTRS_PER_PUD; pudidx++) { |
f4f97b3e | 661 | pmd_t *pmd; |
f4f97b3e | 662 | |
5deb30d1 JF |
663 | if (pgdidx == pgdidx_limit && |
664 | pudidx > pudidx_limit) | |
665 | goto out; | |
3b827c1b | 666 | |
5deb30d1 | 667 | if (pud_none(pud[pudidx])) |
3b827c1b | 668 | continue; |
f4f97b3e | 669 | |
5deb30d1 | 670 | pmd = pmd_offset(&pud[pudidx], 0); |
3b827c1b JF |
671 | |
672 | if (PTRS_PER_PMD > 1) /* not folded */ | |
eefb47f6 | 673 | flush |= (*func)(mm, virt_to_page(pmd), PT_PMD); |
f4f97b3e | 674 | |
5deb30d1 JF |
675 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) { |
676 | struct page *pte; | |
677 | ||
678 | if (pgdidx == pgdidx_limit && | |
679 | pudidx == pudidx_limit && | |
680 | pmdidx > pmdidx_limit) | |
681 | goto out; | |
3b827c1b | 682 | |
5deb30d1 | 683 | if (pmd_none(pmd[pmdidx])) |
3b827c1b JF |
684 | continue; |
685 | ||
5deb30d1 | 686 | pte = pmd_page(pmd[pmdidx]); |
eefb47f6 | 687 | flush |= (*func)(mm, pte, PT_PTE); |
3b827c1b JF |
688 | } |
689 | } | |
690 | } | |
11ad93e5 | 691 | |
5deb30d1 | 692 | out: |
11ad93e5 JF |
693 | /* Do the top level last, so that the callbacks can use it as |
694 | a cue to do final things like tlb flushes. */ | |
eefb47f6 | 695 | flush |= (*func)(mm, virt_to_page(pgd), PT_PGD); |
f4f97b3e JF |
696 | |
697 | return flush; | |
3b827c1b JF |
698 | } |
699 | ||
86bbc2c2 IC |
700 | static int xen_pgd_walk(struct mm_struct *mm, |
701 | int (*func)(struct mm_struct *mm, struct page *, | |
702 | enum pt_level), | |
703 | unsigned long limit) | |
704 | { | |
705 | return __xen_pgd_walk(mm, mm->pgd, func, limit); | |
706 | } | |
707 | ||
7708ad64 JF |
708 | /* If we're using split pte locks, then take the page's lock and |
709 | return a pointer to it. Otherwise return NULL. */ | |
eefb47f6 | 710 | static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm) |
74260714 JF |
711 | { |
712 | spinlock_t *ptl = NULL; | |
713 | ||
57c1ffce | 714 | #if USE_SPLIT_PTE_PTLOCKS |
49076ec2 | 715 | ptl = ptlock_ptr(page); |
eefb47f6 | 716 | spin_lock_nest_lock(ptl, &mm->page_table_lock); |
74260714 JF |
717 | #endif |
718 | ||
719 | return ptl; | |
720 | } | |
721 | ||
7708ad64 | 722 | static void xen_pte_unlock(void *v) |
74260714 JF |
723 | { |
724 | spinlock_t *ptl = v; | |
725 | spin_unlock(ptl); | |
726 | } | |
727 | ||
728 | static void xen_do_pin(unsigned level, unsigned long pfn) | |
729 | { | |
dcf7435c | 730 | struct mmuext_op op; |
74260714 | 731 | |
dcf7435c JF |
732 | op.cmd = level; |
733 | op.arg1.mfn = pfn_to_mfn(pfn); | |
734 | ||
735 | xen_extend_mmuext_op(&op); | |
74260714 JF |
736 | } |
737 | ||
eefb47f6 JF |
738 | static int xen_pin_page(struct mm_struct *mm, struct page *page, |
739 | enum pt_level level) | |
f4f97b3e | 740 | { |
d60cd46b | 741 | unsigned pgfl = TestSetPagePinned(page); |
f4f97b3e JF |
742 | int flush; |
743 | ||
744 | if (pgfl) | |
745 | flush = 0; /* already pinned */ | |
746 | else if (PageHighMem(page)) | |
747 | /* kmaps need flushing if we found an unpinned | |
748 | highpage */ | |
749 | flush = 1; | |
750 | else { | |
751 | void *pt = lowmem_page_address(page); | |
752 | unsigned long pfn = page_to_pfn(page); | |
753 | struct multicall_space mcs = __xen_mc_entry(0); | |
74260714 | 754 | spinlock_t *ptl; |
f4f97b3e JF |
755 | |
756 | flush = 0; | |
757 | ||
11ad93e5 JF |
758 | /* |
759 | * We need to hold the pagetable lock between the time | |
760 | * we make the pagetable RO and when we actually pin | |
761 | * it. If we don't, then other users may come in and | |
762 | * attempt to update the pagetable by writing it, | |
763 | * which will fail because the memory is RO but not | |
764 | * pinned, so Xen won't do the trap'n'emulate. | |
765 | * | |
766 | * If we're using split pte locks, we can't hold the | |
767 | * entire pagetable's worth of locks during the | |
768 | * traverse, because we may wrap the preempt count (8 | |
769 | * bits). The solution is to mark RO and pin each PTE | |
770 | * page while holding the lock. This means the number | |
771 | * of locks we end up holding is never more than a | |
772 | * batch size (~32 entries, at present). | |
773 | * | |
774 | * If we're not using split pte locks, we needn't pin | |
775 | * the PTE pages independently, because we're | |
776 | * protected by the overall pagetable lock. | |
777 | */ | |
74260714 JF |
778 | ptl = NULL; |
779 | if (level == PT_PTE) | |
eefb47f6 | 780 | ptl = xen_pte_lock(page, mm); |
74260714 | 781 | |
f4f97b3e JF |
782 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, |
783 | pfn_pte(pfn, PAGE_KERNEL_RO), | |
74260714 JF |
784 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
785 | ||
11ad93e5 | 786 | if (ptl) { |
74260714 JF |
787 | xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn); |
788 | ||
74260714 JF |
789 | /* Queue a deferred unlock for when this batch |
790 | is completed. */ | |
7708ad64 | 791 | xen_mc_callback(xen_pte_unlock, ptl); |
74260714 | 792 | } |
f4f97b3e JF |
793 | } |
794 | ||
795 | return flush; | |
796 | } | |
3b827c1b | 797 | |
f4f97b3e JF |
798 | /* This is called just after a mm has been created, but it has not |
799 | been used yet. We need to make sure that its pagetable is all | |
800 | read-only, and can be pinned. */ | |
eefb47f6 | 801 | static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd) |
3b827c1b | 802 | { |
5f94fb5b JF |
803 | trace_xen_mmu_pgd_pin(mm, pgd); |
804 | ||
f4f97b3e | 805 | xen_mc_batch(); |
3b827c1b | 806 | |
86bbc2c2 | 807 | if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) { |
d05fdf31 | 808 | /* re-enable interrupts for flushing */ |
f87e4cac | 809 | xen_mc_issue(0); |
d05fdf31 | 810 | |
f4f97b3e | 811 | kmap_flush_unused(); |
d05fdf31 | 812 | |
f87e4cac JF |
813 | xen_mc_batch(); |
814 | } | |
f4f97b3e | 815 | |
d6182fbf JF |
816 | #ifdef CONFIG_X86_64 |
817 | { | |
818 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
819 | ||
820 | xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd))); | |
821 | ||
822 | if (user_pgd) { | |
eefb47f6 | 823 | xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD); |
f63c2f24 T |
824 | xen_do_pin(MMUEXT_PIN_L4_TABLE, |
825 | PFN_DOWN(__pa(user_pgd))); | |
d6182fbf JF |
826 | } |
827 | } | |
828 | #else /* CONFIG_X86_32 */ | |
5deb30d1 JF |
829 | #ifdef CONFIG_X86_PAE |
830 | /* Need to make sure unshared kernel PMD is pinnable */ | |
47cb2ed9 | 831 | xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), |
eefb47f6 | 832 | PT_PMD); |
5deb30d1 | 833 | #endif |
28499143 | 834 | xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd))); |
d6182fbf | 835 | #endif /* CONFIG_X86_64 */ |
f4f97b3e | 836 | xen_mc_issue(0); |
3b827c1b JF |
837 | } |
838 | ||
eefb47f6 JF |
839 | static void xen_pgd_pin(struct mm_struct *mm) |
840 | { | |
841 | __xen_pgd_pin(mm, mm->pgd); | |
842 | } | |
843 | ||
0e91398f JF |
844 | /* |
845 | * On save, we need to pin all pagetables to make sure they get their | |
846 | * mfns turned into pfns. Search the list for any unpinned pgds and pin | |
847 | * them (unpinned pgds are not currently in use, probably because the | |
848 | * process is under construction or destruction). | |
eefb47f6 JF |
849 | * |
850 | * Expected to be called in stop_machine() ("equivalent to taking | |
851 | * every spinlock in the system"), so the locking doesn't really | |
852 | * matter all that much. | |
0e91398f JF |
853 | */ |
854 | void xen_mm_pin_all(void) | |
855 | { | |
0e91398f | 856 | struct page *page; |
74260714 | 857 | |
a79e53d8 | 858 | spin_lock(&pgd_lock); |
f4f97b3e | 859 | |
0e91398f JF |
860 | list_for_each_entry(page, &pgd_list, lru) { |
861 | if (!PagePinned(page)) { | |
eefb47f6 | 862 | __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page)); |
0e91398f JF |
863 | SetPageSavePinned(page); |
864 | } | |
865 | } | |
866 | ||
a79e53d8 | 867 | spin_unlock(&pgd_lock); |
3b827c1b JF |
868 | } |
869 | ||
c1f2f09e EH |
870 | /* |
871 | * The init_mm pagetable is really pinned as soon as its created, but | |
872 | * that's before we have page structures to store the bits. So do all | |
873 | * the book-keeping now. | |
874 | */ | |
3f508953 | 875 | static int __init xen_mark_pinned(struct mm_struct *mm, struct page *page, |
eefb47f6 | 876 | enum pt_level level) |
3b827c1b | 877 | { |
f4f97b3e JF |
878 | SetPagePinned(page); |
879 | return 0; | |
880 | } | |
3b827c1b | 881 | |
b96229b5 | 882 | static void __init xen_mark_init_mm_pinned(void) |
f4f97b3e | 883 | { |
eefb47f6 | 884 | xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP); |
f4f97b3e | 885 | } |
3b827c1b | 886 | |
eefb47f6 JF |
887 | static int xen_unpin_page(struct mm_struct *mm, struct page *page, |
888 | enum pt_level level) | |
f4f97b3e | 889 | { |
d60cd46b | 890 | unsigned pgfl = TestClearPagePinned(page); |
3b827c1b | 891 | |
f4f97b3e JF |
892 | if (pgfl && !PageHighMem(page)) { |
893 | void *pt = lowmem_page_address(page); | |
894 | unsigned long pfn = page_to_pfn(page); | |
74260714 JF |
895 | spinlock_t *ptl = NULL; |
896 | struct multicall_space mcs; | |
897 | ||
11ad93e5 JF |
898 | /* |
899 | * Do the converse to pin_page. If we're using split | |
900 | * pte locks, we must be holding the lock for while | |
901 | * the pte page is unpinned but still RO to prevent | |
902 | * concurrent updates from seeing it in this | |
903 | * partially-pinned state. | |
904 | */ | |
74260714 | 905 | if (level == PT_PTE) { |
eefb47f6 | 906 | ptl = xen_pte_lock(page, mm); |
74260714 | 907 | |
11ad93e5 JF |
908 | if (ptl) |
909 | xen_do_pin(MMUEXT_UNPIN_TABLE, pfn); | |
74260714 JF |
910 | } |
911 | ||
912 | mcs = __xen_mc_entry(0); | |
f4f97b3e JF |
913 | |
914 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, | |
915 | pfn_pte(pfn, PAGE_KERNEL), | |
74260714 JF |
916 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
917 | ||
918 | if (ptl) { | |
919 | /* unlock when batch completed */ | |
7708ad64 | 920 | xen_mc_callback(xen_pte_unlock, ptl); |
74260714 | 921 | } |
f4f97b3e JF |
922 | } |
923 | ||
924 | return 0; /* never need to flush on unpin */ | |
3b827c1b JF |
925 | } |
926 | ||
f4f97b3e | 927 | /* Release a pagetables pages back as normal RW */ |
eefb47f6 | 928 | static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd) |
f4f97b3e | 929 | { |
5f94fb5b JF |
930 | trace_xen_mmu_pgd_unpin(mm, pgd); |
931 | ||
f4f97b3e JF |
932 | xen_mc_batch(); |
933 | ||
74260714 | 934 | xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); |
f4f97b3e | 935 | |
d6182fbf JF |
936 | #ifdef CONFIG_X86_64 |
937 | { | |
938 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
939 | ||
940 | if (user_pgd) { | |
f63c2f24 T |
941 | xen_do_pin(MMUEXT_UNPIN_TABLE, |
942 | PFN_DOWN(__pa(user_pgd))); | |
eefb47f6 | 943 | xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD); |
d6182fbf JF |
944 | } |
945 | } | |
946 | #endif | |
947 | ||
5deb30d1 JF |
948 | #ifdef CONFIG_X86_PAE |
949 | /* Need to make sure unshared kernel PMD is unpinned */ | |
47cb2ed9 | 950 | xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), |
eefb47f6 | 951 | PT_PMD); |
5deb30d1 | 952 | #endif |
d6182fbf | 953 | |
86bbc2c2 | 954 | __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT); |
f4f97b3e JF |
955 | |
956 | xen_mc_issue(0); | |
957 | } | |
3b827c1b | 958 | |
eefb47f6 JF |
959 | static void xen_pgd_unpin(struct mm_struct *mm) |
960 | { | |
961 | __xen_pgd_unpin(mm, mm->pgd); | |
962 | } | |
963 | ||
0e91398f JF |
964 | /* |
965 | * On resume, undo any pinning done at save, so that the rest of the | |
966 | * kernel doesn't see any unexpected pinned pagetables. | |
967 | */ | |
968 | void xen_mm_unpin_all(void) | |
969 | { | |
0e91398f JF |
970 | struct page *page; |
971 | ||
a79e53d8 | 972 | spin_lock(&pgd_lock); |
0e91398f JF |
973 | |
974 | list_for_each_entry(page, &pgd_list, lru) { | |
975 | if (PageSavePinned(page)) { | |
976 | BUG_ON(!PagePinned(page)); | |
eefb47f6 | 977 | __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page)); |
0e91398f JF |
978 | ClearPageSavePinned(page); |
979 | } | |
980 | } | |
981 | ||
a79e53d8 | 982 | spin_unlock(&pgd_lock); |
0e91398f JF |
983 | } |
984 | ||
4c13629f | 985 | static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next) |
3b827c1b | 986 | { |
f4f97b3e | 987 | spin_lock(&next->page_table_lock); |
eefb47f6 | 988 | xen_pgd_pin(next); |
f4f97b3e | 989 | spin_unlock(&next->page_table_lock); |
3b827c1b JF |
990 | } |
991 | ||
4c13629f | 992 | static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm) |
3b827c1b | 993 | { |
f4f97b3e | 994 | spin_lock(&mm->page_table_lock); |
eefb47f6 | 995 | xen_pgd_pin(mm); |
f4f97b3e | 996 | spin_unlock(&mm->page_table_lock); |
3b827c1b JF |
997 | } |
998 | ||
3b827c1b | 999 | |
f87e4cac JF |
1000 | #ifdef CONFIG_SMP |
1001 | /* Another cpu may still have their %cr3 pointing at the pagetable, so | |
1002 | we need to repoint it somewhere else before we can unpin it. */ | |
1003 | static void drop_other_mm_ref(void *info) | |
1004 | { | |
1005 | struct mm_struct *mm = info; | |
ce87b3d3 | 1006 | struct mm_struct *active_mm; |
3b827c1b | 1007 | |
2113f469 | 1008 | active_mm = this_cpu_read(cpu_tlbstate.active_mm); |
ce87b3d3 | 1009 | |
2113f469 | 1010 | if (active_mm == mm && this_cpu_read(cpu_tlbstate.state) != TLBSTATE_OK) |
f87e4cac | 1011 | leave_mm(smp_processor_id()); |
9f79991d JF |
1012 | |
1013 | /* If this cpu still has a stale cr3 reference, then make sure | |
1014 | it has been flushed. */ | |
2113f469 | 1015 | if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd)) |
9f79991d | 1016 | load_cr3(swapper_pg_dir); |
f87e4cac | 1017 | } |
3b827c1b | 1018 | |
7708ad64 | 1019 | static void xen_drop_mm_ref(struct mm_struct *mm) |
f87e4cac | 1020 | { |
e4d98207 | 1021 | cpumask_var_t mask; |
9f79991d JF |
1022 | unsigned cpu; |
1023 | ||
f87e4cac JF |
1024 | if (current->active_mm == mm) { |
1025 | if (current->mm == mm) | |
1026 | load_cr3(swapper_pg_dir); | |
1027 | else | |
1028 | leave_mm(smp_processor_id()); | |
9f79991d JF |
1029 | } |
1030 | ||
1031 | /* Get the "official" set of cpus referring to our pagetable. */ | |
e4d98207 MT |
1032 | if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) { |
1033 | for_each_online_cpu(cpu) { | |
78f1c4d6 | 1034 | if (!cpumask_test_cpu(cpu, mm_cpumask(mm)) |
e4d98207 MT |
1035 | && per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd)) |
1036 | continue; | |
1037 | smp_call_function_single(cpu, drop_other_mm_ref, mm, 1); | |
1038 | } | |
1039 | return; | |
1040 | } | |
78f1c4d6 | 1041 | cpumask_copy(mask, mm_cpumask(mm)); |
9f79991d JF |
1042 | |
1043 | /* It's possible that a vcpu may have a stale reference to our | |
1044 | cr3, because its in lazy mode, and it hasn't yet flushed | |
1045 | its set of pending hypercalls yet. In this case, we can | |
1046 | look at its actual current cr3 value, and force it to flush | |
1047 | if needed. */ | |
1048 | for_each_online_cpu(cpu) { | |
1049 | if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd)) | |
e4d98207 | 1050 | cpumask_set_cpu(cpu, mask); |
3b827c1b JF |
1051 | } |
1052 | ||
e4d98207 MT |
1053 | if (!cpumask_empty(mask)) |
1054 | smp_call_function_many(mask, drop_other_mm_ref, mm, 1); | |
1055 | free_cpumask_var(mask); | |
f87e4cac JF |
1056 | } |
1057 | #else | |
7708ad64 | 1058 | static void xen_drop_mm_ref(struct mm_struct *mm) |
f87e4cac JF |
1059 | { |
1060 | if (current->active_mm == mm) | |
1061 | load_cr3(swapper_pg_dir); | |
1062 | } | |
1063 | #endif | |
1064 | ||
1065 | /* | |
1066 | * While a process runs, Xen pins its pagetables, which means that the | |
1067 | * hypervisor forces it to be read-only, and it controls all updates | |
1068 | * to it. This means that all pagetable updates have to go via the | |
1069 | * hypervisor, which is moderately expensive. | |
1070 | * | |
1071 | * Since we're pulling the pagetable down, we switch to use init_mm, | |
1072 | * unpin old process pagetable and mark it all read-write, which | |
1073 | * allows further operations on it to be simple memory accesses. | |
1074 | * | |
1075 | * The only subtle point is that another CPU may be still using the | |
1076 | * pagetable because of lazy tlb flushing. This means we need need to | |
1077 | * switch all CPUs off this pagetable before we can unpin it. | |
1078 | */ | |
4c13629f | 1079 | static void xen_exit_mmap(struct mm_struct *mm) |
f87e4cac JF |
1080 | { |
1081 | get_cpu(); /* make sure we don't move around */ | |
7708ad64 | 1082 | xen_drop_mm_ref(mm); |
f87e4cac | 1083 | put_cpu(); |
3b827c1b | 1084 | |
f120f13e | 1085 | spin_lock(&mm->page_table_lock); |
df912ea4 JF |
1086 | |
1087 | /* pgd may not be pinned in the error exit path of execve */ | |
7708ad64 | 1088 | if (xen_page_pinned(mm->pgd)) |
eefb47f6 | 1089 | xen_pgd_unpin(mm); |
74260714 | 1090 | |
f120f13e | 1091 | spin_unlock(&mm->page_table_lock); |
3b827c1b | 1092 | } |
994025ca | 1093 | |
c7112887 AR |
1094 | static void xen_post_allocator_init(void); |
1095 | ||
7f914062 KRW |
1096 | #ifdef CONFIG_X86_64 |
1097 | static void __init xen_cleanhighmap(unsigned long vaddr, | |
1098 | unsigned long vaddr_end) | |
1099 | { | |
1100 | unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1; | |
1101 | pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr); | |
1102 | ||
1103 | /* NOTE: The loop is more greedy than the cleanup_highmap variant. | |
1104 | * We include the PMD passed in on _both_ boundaries. */ | |
1105 | for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PAGE_SIZE)); | |
1106 | pmd++, vaddr += PMD_SIZE) { | |
1107 | if (pmd_none(*pmd)) | |
1108 | continue; | |
1109 | if (vaddr < (unsigned long) _text || vaddr > kernel_end) | |
1110 | set_pmd(pmd, __pmd(0)); | |
1111 | } | |
1112 | /* In case we did something silly, we should crash in this function | |
1113 | * instead of somewhere later and be confusing. */ | |
1114 | xen_mc_flush(); | |
1115 | } | |
32df75cd | 1116 | static void __init xen_pagetable_p2m_copy(void) |
319f3ba5 | 1117 | { |
7f914062 KRW |
1118 | unsigned long size; |
1119 | unsigned long addr; | |
32df75cd KRW |
1120 | unsigned long new_mfn_list; |
1121 | ||
1122 | if (xen_feature(XENFEAT_auto_translated_physmap)) | |
1123 | return; | |
1124 | ||
1125 | size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long)); | |
1126 | ||
32df75cd | 1127 | new_mfn_list = xen_revector_p2m_tree(); |
b621e157 KRW |
1128 | /* No memory or already called. */ |
1129 | if (!new_mfn_list || new_mfn_list == xen_start_info->mfn_list) | |
32df75cd | 1130 | return; |
7f914062 | 1131 | |
b621e157 KRW |
1132 | /* using __ka address and sticking INVALID_P2M_ENTRY! */ |
1133 | memset((void *)xen_start_info->mfn_list, 0xff, size); | |
1134 | ||
1135 | /* We should be in __ka space. */ | |
1136 | BUG_ON(xen_start_info->mfn_list < __START_KERNEL_map); | |
1137 | addr = xen_start_info->mfn_list; | |
1138 | /* We roundup to the PMD, which means that if anybody at this stage is | |
1139 | * using the __ka address of xen_start_info or xen_start_info->shared_info | |
1140 | * they are in going to crash. Fortunatly we have already revectored | |
1141 | * in xen_setup_kernel_pagetable and in xen_setup_shared_info. */ | |
1142 | size = roundup(size, PMD_SIZE); | |
1143 | xen_cleanhighmap(addr, addr + size); | |
1144 | ||
1145 | size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long)); | |
1146 | memblock_free(__pa(xen_start_info->mfn_list), size); | |
1147 | /* And revector! Bye bye old array */ | |
1148 | xen_start_info->mfn_list = new_mfn_list; | |
1149 | ||
3aca7fbc KRW |
1150 | /* At this stage, cleanup_highmap has already cleaned __ka space |
1151 | * from _brk_limit way up to the max_pfn_mapped (which is the end of | |
1152 | * the ramdisk). We continue on, erasing PMD entries that point to page | |
1153 | * tables - do note that they are accessible at this stage via __va. | |
1154 | * For good measure we also round up to the PMD - which means that if | |
1155 | * anybody is using __ka address to the initial boot-stack - and try | |
1156 | * to use it - they are going to crash. The xen_start_info has been | |
1157 | * taken care of already in xen_setup_kernel_pagetable. */ | |
1158 | addr = xen_start_info->pt_base; | |
1159 | size = roundup(xen_start_info->nr_pt_frames * PAGE_SIZE, PMD_SIZE); | |
1160 | ||
1161 | xen_cleanhighmap(addr, addr + size); | |
1162 | xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base)); | |
1163 | #ifdef DEBUG | |
1164 | /* This is superflous and is not neccessary, but you know what | |
1165 | * lets do it. The MODULES_VADDR -> MODULES_END should be clear of | |
1166 | * anything at this stage. */ | |
1167 | xen_cleanhighmap(MODULES_VADDR, roundup(MODULES_VADDR, PUD_SIZE) - 1); | |
1168 | #endif | |
32df75cd KRW |
1169 | } |
1170 | #endif | |
1171 | ||
1172 | static void __init xen_pagetable_init(void) | |
1173 | { | |
1174 | paging_init(); | |
32df75cd KRW |
1175 | #ifdef CONFIG_X86_64 |
1176 | xen_pagetable_p2m_copy(); | |
7f914062 | 1177 | #endif |
2c185687 JG |
1178 | /* Allocate and initialize top and mid mfn levels for p2m structure */ |
1179 | xen_build_mfn_list_list(); | |
1180 | ||
1181 | xen_setup_shared_info(); | |
f1d7062a | 1182 | xen_post_allocator_init(); |
319f3ba5 | 1183 | } |
319f3ba5 JF |
1184 | static void xen_write_cr2(unsigned long cr2) |
1185 | { | |
2113f469 | 1186 | this_cpu_read(xen_vcpu)->arch.cr2 = cr2; |
319f3ba5 JF |
1187 | } |
1188 | ||
1189 | static unsigned long xen_read_cr2(void) | |
1190 | { | |
2113f469 | 1191 | return this_cpu_read(xen_vcpu)->arch.cr2; |
319f3ba5 JF |
1192 | } |
1193 | ||
1194 | unsigned long xen_read_cr2_direct(void) | |
1195 | { | |
2113f469 | 1196 | return this_cpu_read(xen_vcpu_info.arch.cr2); |
319f3ba5 JF |
1197 | } |
1198 | ||
95a7d768 KRW |
1199 | void xen_flush_tlb_all(void) |
1200 | { | |
1201 | struct mmuext_op *op; | |
1202 | struct multicall_space mcs; | |
1203 | ||
1204 | trace_xen_mmu_flush_tlb_all(0); | |
1205 | ||
1206 | preempt_disable(); | |
1207 | ||
1208 | mcs = xen_mc_entry(sizeof(*op)); | |
1209 | ||
1210 | op = mcs.args; | |
1211 | op->cmd = MMUEXT_TLB_FLUSH_ALL; | |
1212 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1213 | ||
1214 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1215 | ||
1216 | preempt_enable(); | |
1217 | } | |
319f3ba5 JF |
1218 | static void xen_flush_tlb(void) |
1219 | { | |
1220 | struct mmuext_op *op; | |
1221 | struct multicall_space mcs; | |
1222 | ||
c8eed171 JF |
1223 | trace_xen_mmu_flush_tlb(0); |
1224 | ||
319f3ba5 JF |
1225 | preempt_disable(); |
1226 | ||
1227 | mcs = xen_mc_entry(sizeof(*op)); | |
1228 | ||
1229 | op = mcs.args; | |
1230 | op->cmd = MMUEXT_TLB_FLUSH_LOCAL; | |
1231 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1232 | ||
1233 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1234 | ||
1235 | preempt_enable(); | |
1236 | } | |
1237 | ||
1238 | static void xen_flush_tlb_single(unsigned long addr) | |
1239 | { | |
1240 | struct mmuext_op *op; | |
1241 | struct multicall_space mcs; | |
1242 | ||
c8eed171 JF |
1243 | trace_xen_mmu_flush_tlb_single(addr); |
1244 | ||
319f3ba5 JF |
1245 | preempt_disable(); |
1246 | ||
1247 | mcs = xen_mc_entry(sizeof(*op)); | |
1248 | op = mcs.args; | |
1249 | op->cmd = MMUEXT_INVLPG_LOCAL; | |
1250 | op->arg1.linear_addr = addr & PAGE_MASK; | |
1251 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1252 | ||
1253 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1254 | ||
1255 | preempt_enable(); | |
1256 | } | |
1257 | ||
1258 | static void xen_flush_tlb_others(const struct cpumask *cpus, | |
e7b52ffd AS |
1259 | struct mm_struct *mm, unsigned long start, |
1260 | unsigned long end) | |
319f3ba5 JF |
1261 | { |
1262 | struct { | |
1263 | struct mmuext_op op; | |
32dd1194 | 1264 | #ifdef CONFIG_SMP |
900cba88 | 1265 | DECLARE_BITMAP(mask, num_processors); |
32dd1194 KRW |
1266 | #else |
1267 | DECLARE_BITMAP(mask, NR_CPUS); | |
1268 | #endif | |
319f3ba5 JF |
1269 | } *args; |
1270 | struct multicall_space mcs; | |
1271 | ||
e7b52ffd | 1272 | trace_xen_mmu_flush_tlb_others(cpus, mm, start, end); |
c8eed171 | 1273 | |
e3f8a74e JF |
1274 | if (cpumask_empty(cpus)) |
1275 | return; /* nothing to do */ | |
319f3ba5 JF |
1276 | |
1277 | mcs = xen_mc_entry(sizeof(*args)); | |
1278 | args = mcs.args; | |
1279 | args->op.arg2.vcpumask = to_cpumask(args->mask); | |
1280 | ||
1281 | /* Remove us, and any offline CPUS. */ | |
1282 | cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask); | |
1283 | cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask)); | |
319f3ba5 | 1284 | |
e7b52ffd | 1285 | args->op.cmd = MMUEXT_TLB_FLUSH_MULTI; |
ce7184bd | 1286 | if (end != TLB_FLUSH_ALL && (end - start) <= PAGE_SIZE) { |
319f3ba5 | 1287 | args->op.cmd = MMUEXT_INVLPG_MULTI; |
e7b52ffd | 1288 | args->op.arg1.linear_addr = start; |
319f3ba5 JF |
1289 | } |
1290 | ||
1291 | MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF); | |
1292 | ||
319f3ba5 JF |
1293 | xen_mc_issue(PARAVIRT_LAZY_MMU); |
1294 | } | |
1295 | ||
1296 | static unsigned long xen_read_cr3(void) | |
1297 | { | |
2113f469 | 1298 | return this_cpu_read(xen_cr3); |
319f3ba5 JF |
1299 | } |
1300 | ||
1301 | static void set_current_cr3(void *v) | |
1302 | { | |
2113f469 | 1303 | this_cpu_write(xen_current_cr3, (unsigned long)v); |
319f3ba5 JF |
1304 | } |
1305 | ||
1306 | static void __xen_write_cr3(bool kernel, unsigned long cr3) | |
1307 | { | |
dcf7435c | 1308 | struct mmuext_op op; |
319f3ba5 JF |
1309 | unsigned long mfn; |
1310 | ||
c8eed171 JF |
1311 | trace_xen_mmu_write_cr3(kernel, cr3); |
1312 | ||
319f3ba5 JF |
1313 | if (cr3) |
1314 | mfn = pfn_to_mfn(PFN_DOWN(cr3)); | |
1315 | else | |
1316 | mfn = 0; | |
1317 | ||
1318 | WARN_ON(mfn == 0 && kernel); | |
1319 | ||
dcf7435c JF |
1320 | op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR; |
1321 | op.arg1.mfn = mfn; | |
319f3ba5 | 1322 | |
dcf7435c | 1323 | xen_extend_mmuext_op(&op); |
319f3ba5 JF |
1324 | |
1325 | if (kernel) { | |
2113f469 | 1326 | this_cpu_write(xen_cr3, cr3); |
319f3ba5 JF |
1327 | |
1328 | /* Update xen_current_cr3 once the batch has actually | |
1329 | been submitted. */ | |
1330 | xen_mc_callback(set_current_cr3, (void *)cr3); | |
1331 | } | |
1332 | } | |
319f3ba5 JF |
1333 | static void xen_write_cr3(unsigned long cr3) |
1334 | { | |
1335 | BUG_ON(preemptible()); | |
1336 | ||
1337 | xen_mc_batch(); /* disables interrupts */ | |
1338 | ||
1339 | /* Update while interrupts are disabled, so its atomic with | |
1340 | respect to ipis */ | |
2113f469 | 1341 | this_cpu_write(xen_cr3, cr3); |
319f3ba5 JF |
1342 | |
1343 | __xen_write_cr3(true, cr3); | |
1344 | ||
1345 | #ifdef CONFIG_X86_64 | |
1346 | { | |
1347 | pgd_t *user_pgd = xen_get_user_pgd(__va(cr3)); | |
1348 | if (user_pgd) | |
1349 | __xen_write_cr3(false, __pa(user_pgd)); | |
1350 | else | |
1351 | __xen_write_cr3(false, 0); | |
1352 | } | |
1353 | #endif | |
1354 | ||
1355 | xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ | |
1356 | } | |
1357 | ||
0cc9129d KRW |
1358 | #ifdef CONFIG_X86_64 |
1359 | /* | |
1360 | * At the start of the day - when Xen launches a guest, it has already | |
1361 | * built pagetables for the guest. We diligently look over them | |
1362 | * in xen_setup_kernel_pagetable and graft as appropiate them in the | |
1363 | * init_level4_pgt and its friends. Then when we are happy we load | |
1364 | * the new init_level4_pgt - and continue on. | |
1365 | * | |
1366 | * The generic code starts (start_kernel) and 'init_mem_mapping' sets | |
1367 | * up the rest of the pagetables. When it has completed it loads the cr3. | |
1368 | * N.B. that baremetal would start at 'start_kernel' (and the early | |
1369 | * #PF handler would create bootstrap pagetables) - so we are running | |
1370 | * with the same assumptions as what to do when write_cr3 is executed | |
1371 | * at this point. | |
1372 | * | |
1373 | * Since there are no user-page tables at all, we have two variants | |
1374 | * of xen_write_cr3 - the early bootup (this one), and the late one | |
1375 | * (xen_write_cr3). The reason we have to do that is that in 64-bit | |
1376 | * the Linux kernel and user-space are both in ring 3 while the | |
1377 | * hypervisor is in ring 0. | |
1378 | */ | |
1379 | static void __init xen_write_cr3_init(unsigned long cr3) | |
1380 | { | |
1381 | BUG_ON(preemptible()); | |
1382 | ||
1383 | xen_mc_batch(); /* disables interrupts */ | |
1384 | ||
1385 | /* Update while interrupts are disabled, so its atomic with | |
1386 | respect to ipis */ | |
1387 | this_cpu_write(xen_cr3, cr3); | |
1388 | ||
1389 | __xen_write_cr3(true, cr3); | |
1390 | ||
1391 | xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ | |
0cc9129d KRW |
1392 | } |
1393 | #endif | |
1394 | ||
319f3ba5 JF |
1395 | static int xen_pgd_alloc(struct mm_struct *mm) |
1396 | { | |
1397 | pgd_t *pgd = mm->pgd; | |
1398 | int ret = 0; | |
1399 | ||
1400 | BUG_ON(PagePinned(virt_to_page(pgd))); | |
1401 | ||
1402 | #ifdef CONFIG_X86_64 | |
1403 | { | |
1404 | struct page *page = virt_to_page(pgd); | |
1405 | pgd_t *user_pgd; | |
1406 | ||
1407 | BUG_ON(page->private != 0); | |
1408 | ||
1409 | ret = -ENOMEM; | |
1410 | ||
1411 | user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO); | |
1412 | page->private = (unsigned long)user_pgd; | |
1413 | ||
1414 | if (user_pgd != NULL) { | |
1ad83c85 | 1415 | #ifdef CONFIG_X86_VSYSCALL_EMULATION |
f40c3300 | 1416 | user_pgd[pgd_index(VSYSCALL_ADDR)] = |
319f3ba5 | 1417 | __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE); |
1ad83c85 | 1418 | #endif |
319f3ba5 JF |
1419 | ret = 0; |
1420 | } | |
1421 | ||
1422 | BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd)))); | |
1423 | } | |
1424 | #endif | |
1425 | ||
1426 | return ret; | |
1427 | } | |
1428 | ||
1429 | static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd) | |
1430 | { | |
1431 | #ifdef CONFIG_X86_64 | |
1432 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
1433 | ||
1434 | if (user_pgd) | |
1435 | free_page((unsigned long)user_pgd); | |
1436 | #endif | |
1437 | } | |
1438 | ||
ee176455 | 1439 | #ifdef CONFIG_X86_32 |
3f508953 | 1440 | static pte_t __init mask_rw_pte(pte_t *ptep, pte_t pte) |
1f4f9315 JF |
1441 | { |
1442 | /* If there's an existing pte, then don't allow _PAGE_RW to be set */ | |
1443 | if (pte_val_ma(*ptep) & _PAGE_PRESENT) | |
1444 | pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) & | |
1445 | pte_val_ma(pte)); | |
ee176455 SS |
1446 | |
1447 | return pte; | |
1448 | } | |
1449 | #else /* CONFIG_X86_64 */ | |
3f508953 | 1450 | static pte_t __init mask_rw_pte(pte_t *ptep, pte_t pte) |
ee176455 | 1451 | { |
1f4f9315 JF |
1452 | return pte; |
1453 | } | |
ee176455 | 1454 | #endif /* CONFIG_X86_64 */ |
1f4f9315 | 1455 | |
d095d43e DV |
1456 | /* |
1457 | * Init-time set_pte while constructing initial pagetables, which | |
1458 | * doesn't allow RO page table pages to be remapped RW. | |
1459 | * | |
66a27dde DV |
1460 | * If there is no MFN for this PFN then this page is initially |
1461 | * ballooned out so clear the PTE (as in decrease_reservation() in | |
1462 | * drivers/xen/balloon.c). | |
1463 | * | |
d095d43e DV |
1464 | * Many of these PTE updates are done on unpinned and writable pages |
1465 | * and doing a hypercall for these is unnecessary and expensive. At | |
1466 | * this point it is not possible to tell if a page is pinned or not, | |
1467 | * so always write the PTE directly and rely on Xen trapping and | |
1468 | * emulating any updates as necessary. | |
1469 | */ | |
3f508953 | 1470 | static void __init xen_set_pte_init(pte_t *ptep, pte_t pte) |
1f4f9315 | 1471 | { |
66a27dde DV |
1472 | if (pte_mfn(pte) != INVALID_P2M_ENTRY) |
1473 | pte = mask_rw_pte(ptep, pte); | |
1474 | else | |
1475 | pte = __pte_ma(0); | |
1f4f9315 | 1476 | |
d095d43e | 1477 | native_set_pte(ptep, pte); |
1f4f9315 | 1478 | } |
319f3ba5 | 1479 | |
b96229b5 JF |
1480 | static void pin_pagetable_pfn(unsigned cmd, unsigned long pfn) |
1481 | { | |
1482 | struct mmuext_op op; | |
1483 | op.cmd = cmd; | |
1484 | op.arg1.mfn = pfn_to_mfn(pfn); | |
1485 | if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) | |
1486 | BUG(); | |
1487 | } | |
1488 | ||
319f3ba5 JF |
1489 | /* Early in boot, while setting up the initial pagetable, assume |
1490 | everything is pinned. */ | |
3f508953 | 1491 | static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn) |
319f3ba5 | 1492 | { |
b96229b5 JF |
1493 | #ifdef CONFIG_FLATMEM |
1494 | BUG_ON(mem_map); /* should only be used early */ | |
1495 | #endif | |
1496 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1497 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
1498 | } | |
1499 | ||
1500 | /* Used for pmd and pud */ | |
3f508953 | 1501 | static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn) |
b96229b5 | 1502 | { |
319f3ba5 JF |
1503 | #ifdef CONFIG_FLATMEM |
1504 | BUG_ON(mem_map); /* should only be used early */ | |
1505 | #endif | |
1506 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1507 | } | |
1508 | ||
1509 | /* Early release_pte assumes that all pts are pinned, since there's | |
1510 | only init_mm and anything attached to that is pinned. */ | |
3f508953 | 1511 | static void __init xen_release_pte_init(unsigned long pfn) |
319f3ba5 | 1512 | { |
b96229b5 | 1513 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); |
319f3ba5 JF |
1514 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); |
1515 | } | |
1516 | ||
3f508953 | 1517 | static void __init xen_release_pmd_init(unsigned long pfn) |
319f3ba5 | 1518 | { |
b96229b5 | 1519 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); |
319f3ba5 JF |
1520 | } |
1521 | ||
bc7fe1d9 JF |
1522 | static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn) |
1523 | { | |
1524 | struct multicall_space mcs; | |
1525 | struct mmuext_op *op; | |
1526 | ||
1527 | mcs = __xen_mc_entry(sizeof(*op)); | |
1528 | op = mcs.args; | |
1529 | op->cmd = cmd; | |
1530 | op->arg1.mfn = pfn_to_mfn(pfn); | |
1531 | ||
1532 | MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); | |
1533 | } | |
1534 | ||
1535 | static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot) | |
1536 | { | |
1537 | struct multicall_space mcs; | |
1538 | unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT); | |
1539 | ||
1540 | mcs = __xen_mc_entry(0); | |
1541 | MULTI_update_va_mapping(mcs.mc, (unsigned long)addr, | |
1542 | pfn_pte(pfn, prot), 0); | |
1543 | } | |
1544 | ||
319f3ba5 JF |
1545 | /* This needs to make sure the new pte page is pinned iff its being |
1546 | attached to a pinned pagetable. */ | |
bc7fe1d9 JF |
1547 | static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, |
1548 | unsigned level) | |
319f3ba5 | 1549 | { |
bc7fe1d9 JF |
1550 | bool pinned = PagePinned(virt_to_page(mm->pgd)); |
1551 | ||
c2ba050d | 1552 | trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned); |
319f3ba5 | 1553 | |
c2ba050d | 1554 | if (pinned) { |
bc7fe1d9 | 1555 | struct page *page = pfn_to_page(pfn); |
319f3ba5 | 1556 | |
319f3ba5 JF |
1557 | SetPagePinned(page); |
1558 | ||
319f3ba5 | 1559 | if (!PageHighMem(page)) { |
bc7fe1d9 JF |
1560 | xen_mc_batch(); |
1561 | ||
1562 | __set_pfn_prot(pfn, PAGE_KERNEL_RO); | |
1563 | ||
57c1ffce | 1564 | if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS) |
bc7fe1d9 JF |
1565 | __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); |
1566 | ||
1567 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
319f3ba5 JF |
1568 | } else { |
1569 | /* make sure there are no stray mappings of | |
1570 | this page */ | |
1571 | kmap_flush_unused(); | |
1572 | } | |
1573 | } | |
1574 | } | |
1575 | ||
1576 | static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn) | |
1577 | { | |
1578 | xen_alloc_ptpage(mm, pfn, PT_PTE); | |
1579 | } | |
1580 | ||
1581 | static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn) | |
1582 | { | |
1583 | xen_alloc_ptpage(mm, pfn, PT_PMD); | |
1584 | } | |
1585 | ||
1586 | /* This should never happen until we're OK to use struct page */ | |
bc7fe1d9 | 1587 | static inline void xen_release_ptpage(unsigned long pfn, unsigned level) |
319f3ba5 JF |
1588 | { |
1589 | struct page *page = pfn_to_page(pfn); | |
c2ba050d | 1590 | bool pinned = PagePinned(page); |
319f3ba5 | 1591 | |
c2ba050d | 1592 | trace_xen_mmu_release_ptpage(pfn, level, pinned); |
319f3ba5 | 1593 | |
c2ba050d | 1594 | if (pinned) { |
319f3ba5 | 1595 | if (!PageHighMem(page)) { |
bc7fe1d9 JF |
1596 | xen_mc_batch(); |
1597 | ||
57c1ffce | 1598 | if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS) |
bc7fe1d9 JF |
1599 | __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); |
1600 | ||
1601 | __set_pfn_prot(pfn, PAGE_KERNEL); | |
1602 | ||
1603 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
319f3ba5 JF |
1604 | } |
1605 | ClearPagePinned(page); | |
1606 | } | |
1607 | } | |
1608 | ||
1609 | static void xen_release_pte(unsigned long pfn) | |
1610 | { | |
1611 | xen_release_ptpage(pfn, PT_PTE); | |
1612 | } | |
1613 | ||
1614 | static void xen_release_pmd(unsigned long pfn) | |
1615 | { | |
1616 | xen_release_ptpage(pfn, PT_PMD); | |
1617 | } | |
1618 | ||
1619 | #if PAGETABLE_LEVELS == 4 | |
1620 | static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn) | |
1621 | { | |
1622 | xen_alloc_ptpage(mm, pfn, PT_PUD); | |
1623 | } | |
1624 | ||
1625 | static void xen_release_pud(unsigned long pfn) | |
1626 | { | |
1627 | xen_release_ptpage(pfn, PT_PUD); | |
1628 | } | |
1629 | #endif | |
1630 | ||
1631 | void __init xen_reserve_top(void) | |
1632 | { | |
1633 | #ifdef CONFIG_X86_32 | |
1634 | unsigned long top = HYPERVISOR_VIRT_START; | |
1635 | struct xen_platform_parameters pp; | |
1636 | ||
1637 | if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0) | |
1638 | top = pp.virt_start; | |
1639 | ||
1640 | reserve_top_address(-top); | |
1641 | #endif /* CONFIG_X86_32 */ | |
1642 | } | |
1643 | ||
1644 | /* | |
1645 | * Like __va(), but returns address in the kernel mapping (which is | |
1646 | * all we have until the physical memory mapping has been set up. | |
1647 | */ | |
1648 | static void *__ka(phys_addr_t paddr) | |
1649 | { | |
1650 | #ifdef CONFIG_X86_64 | |
1651 | return (void *)(paddr + __START_KERNEL_map); | |
1652 | #else | |
1653 | return __va(paddr); | |
1654 | #endif | |
1655 | } | |
1656 | ||
1657 | /* Convert a machine address to physical address */ | |
1658 | static unsigned long m2p(phys_addr_t maddr) | |
1659 | { | |
1660 | phys_addr_t paddr; | |
1661 | ||
1662 | maddr &= PTE_PFN_MASK; | |
1663 | paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT; | |
1664 | ||
1665 | return paddr; | |
1666 | } | |
1667 | ||
1668 | /* Convert a machine address to kernel virtual */ | |
1669 | static void *m2v(phys_addr_t maddr) | |
1670 | { | |
1671 | return __ka(m2p(maddr)); | |
1672 | } | |
1673 | ||
4ec5387c | 1674 | /* Set the page permissions on an identity-mapped pages */ |
b2222794 | 1675 | static void set_page_prot_flags(void *addr, pgprot_t prot, unsigned long flags) |
319f3ba5 JF |
1676 | { |
1677 | unsigned long pfn = __pa(addr) >> PAGE_SHIFT; | |
1678 | pte_t pte = pfn_pte(pfn, prot); | |
1679 | ||
4e44e44b MR |
1680 | /* For PVH no need to set R/O or R/W to pin them or unpin them. */ |
1681 | if (xen_feature(XENFEAT_auto_translated_physmap)) | |
1682 | return; | |
1683 | ||
b2222794 | 1684 | if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags)) |
319f3ba5 JF |
1685 | BUG(); |
1686 | } | |
b2222794 KRW |
1687 | static void set_page_prot(void *addr, pgprot_t prot) |
1688 | { | |
1689 | return set_page_prot_flags(addr, prot, UVMF_NONE); | |
1690 | } | |
caaf9ecf | 1691 | #ifdef CONFIG_X86_32 |
3f508953 | 1692 | static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn) |
319f3ba5 JF |
1693 | { |
1694 | unsigned pmdidx, pteidx; | |
1695 | unsigned ident_pte; | |
1696 | unsigned long pfn; | |
1697 | ||
764f0138 JF |
1698 | level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES, |
1699 | PAGE_SIZE); | |
1700 | ||
319f3ba5 JF |
1701 | ident_pte = 0; |
1702 | pfn = 0; | |
1703 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) { | |
1704 | pte_t *pte_page; | |
1705 | ||
1706 | /* Reuse or allocate a page of ptes */ | |
1707 | if (pmd_present(pmd[pmdidx])) | |
1708 | pte_page = m2v(pmd[pmdidx].pmd); | |
1709 | else { | |
1710 | /* Check for free pte pages */ | |
764f0138 | 1711 | if (ident_pte == LEVEL1_IDENT_ENTRIES) |
319f3ba5 JF |
1712 | break; |
1713 | ||
1714 | pte_page = &level1_ident_pgt[ident_pte]; | |
1715 | ident_pte += PTRS_PER_PTE; | |
1716 | ||
1717 | pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE); | |
1718 | } | |
1719 | ||
1720 | /* Install mappings */ | |
1721 | for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) { | |
1722 | pte_t pte; | |
1723 | ||
a91d9287 SS |
1724 | #ifdef CONFIG_X86_32 |
1725 | if (pfn > max_pfn_mapped) | |
1726 | max_pfn_mapped = pfn; | |
1727 | #endif | |
1728 | ||
319f3ba5 JF |
1729 | if (!pte_none(pte_page[pteidx])) |
1730 | continue; | |
1731 | ||
1732 | pte = pfn_pte(pfn, PAGE_KERNEL_EXEC); | |
1733 | pte_page[pteidx] = pte; | |
1734 | } | |
1735 | } | |
1736 | ||
1737 | for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE) | |
1738 | set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO); | |
1739 | ||
1740 | set_page_prot(pmd, PAGE_KERNEL_RO); | |
1741 | } | |
caaf9ecf | 1742 | #endif |
7e77506a IC |
1743 | void __init xen_setup_machphys_mapping(void) |
1744 | { | |
1745 | struct xen_machphys_mapping mapping; | |
7e77506a IC |
1746 | |
1747 | if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) { | |
1748 | machine_to_phys_mapping = (unsigned long *)mapping.v_start; | |
ccbcdf7c | 1749 | machine_to_phys_nr = mapping.max_mfn + 1; |
7e77506a | 1750 | } else { |
ccbcdf7c | 1751 | machine_to_phys_nr = MACH2PHYS_NR_ENTRIES; |
7e77506a | 1752 | } |
ccbcdf7c | 1753 | #ifdef CONFIG_X86_32 |
61cca2fa JB |
1754 | WARN_ON((machine_to_phys_mapping + (machine_to_phys_nr - 1)) |
1755 | < machine_to_phys_mapping); | |
ccbcdf7c | 1756 | #endif |
7e77506a IC |
1757 | } |
1758 | ||
319f3ba5 JF |
1759 | #ifdef CONFIG_X86_64 |
1760 | static void convert_pfn_mfn(void *v) | |
1761 | { | |
1762 | pte_t *pte = v; | |
1763 | int i; | |
1764 | ||
1765 | /* All levels are converted the same way, so just treat them | |
1766 | as ptes. */ | |
1767 | for (i = 0; i < PTRS_PER_PTE; i++) | |
1768 | pte[i] = xen_make_pte(pte[i].pte); | |
1769 | } | |
488f046d KRW |
1770 | static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end, |
1771 | unsigned long addr) | |
1772 | { | |
1773 | if (*pt_base == PFN_DOWN(__pa(addr))) { | |
b2222794 | 1774 | set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG); |
488f046d KRW |
1775 | clear_page((void *)addr); |
1776 | (*pt_base)++; | |
1777 | } | |
1778 | if (*pt_end == PFN_DOWN(__pa(addr))) { | |
b2222794 | 1779 | set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG); |
488f046d KRW |
1780 | clear_page((void *)addr); |
1781 | (*pt_end)--; | |
1782 | } | |
1783 | } | |
319f3ba5 | 1784 | /* |
0d2eb44f | 1785 | * Set up the initial kernel pagetable. |
319f3ba5 JF |
1786 | * |
1787 | * We can construct this by grafting the Xen provided pagetable into | |
1788 | * head_64.S's preconstructed pagetables. We copy the Xen L2's into | |
0b5a5063 SB |
1789 | * level2_ident_pgt, and level2_kernel_pgt. This means that only the |
1790 | * kernel has a physical mapping to start with - but that's enough to | |
1791 | * get __va working. We need to fill in the rest of the physical | |
1792 | * mapping once some sort of allocator has been set up. NOTE: for | |
1793 | * PVH, the page tables are native. | |
319f3ba5 | 1794 | */ |
3699aad0 | 1795 | void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn) |
319f3ba5 JF |
1796 | { |
1797 | pud_t *l3; | |
1798 | pmd_t *l2; | |
488f046d KRW |
1799 | unsigned long addr[3]; |
1800 | unsigned long pt_base, pt_end; | |
1801 | unsigned i; | |
319f3ba5 | 1802 | |
14988a4d SS |
1803 | /* max_pfn_mapped is the last pfn mapped in the initial memory |
1804 | * mappings. Considering that on Xen after the kernel mappings we | |
1805 | * have the mappings of some pages that don't exist in pfn space, we | |
1806 | * set max_pfn_mapped to the last real pfn mapped. */ | |
1807 | max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list)); | |
1808 | ||
488f046d KRW |
1809 | pt_base = PFN_DOWN(__pa(xen_start_info->pt_base)); |
1810 | pt_end = pt_base + xen_start_info->nr_pt_frames; | |
1811 | ||
319f3ba5 JF |
1812 | /* Zap identity mapping */ |
1813 | init_level4_pgt[0] = __pgd(0); | |
1814 | ||
4e44e44b MR |
1815 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { |
1816 | /* Pre-constructed entries are in pfn, so convert to mfn */ | |
1817 | /* L4[272] -> level3_ident_pgt | |
1818 | * L4[511] -> level3_kernel_pgt */ | |
1819 | convert_pfn_mfn(init_level4_pgt); | |
1820 | ||
1821 | /* L3_i[0] -> level2_ident_pgt */ | |
1822 | convert_pfn_mfn(level3_ident_pgt); | |
1823 | /* L3_k[510] -> level2_kernel_pgt | |
0b5a5063 | 1824 | * L3_k[511] -> level2_fixmap_pgt */ |
4e44e44b | 1825 | convert_pfn_mfn(level3_kernel_pgt); |
0b5a5063 SB |
1826 | |
1827 | /* L3_k[511][506] -> level1_fixmap_pgt */ | |
1828 | convert_pfn_mfn(level2_fixmap_pgt); | |
4e44e44b | 1829 | } |
4fac153a | 1830 | /* We get [511][511] and have Xen's version of level2_kernel_pgt */ |
319f3ba5 JF |
1831 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd); |
1832 | l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud); | |
1833 | ||
488f046d KRW |
1834 | addr[0] = (unsigned long)pgd; |
1835 | addr[1] = (unsigned long)l3; | |
1836 | addr[2] = (unsigned long)l2; | |
4fac153a | 1837 | /* Graft it onto L4[272][0]. Note that we creating an aliasing problem: |
0b5a5063 | 1838 | * Both L4[272][0] and L4[511][510] have entries that point to the same |
4fac153a KRW |
1839 | * L2 (PMD) tables. Meaning that if you modify it in __va space |
1840 | * it will be also modified in the __ka space! (But if you just | |
1841 | * modify the PMD table to point to other PTE's or none, then you | |
1842 | * are OK - which is what cleanup_highmap does) */ | |
ae895ed7 | 1843 | copy_page(level2_ident_pgt, l2); |
0b5a5063 | 1844 | /* Graft it onto L4[511][510] */ |
ae895ed7 | 1845 | copy_page(level2_kernel_pgt, l2); |
319f3ba5 | 1846 | |
4e44e44b MR |
1847 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { |
1848 | /* Make pagetable pieces RO */ | |
1849 | set_page_prot(init_level4_pgt, PAGE_KERNEL_RO); | |
1850 | set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO); | |
1851 | set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO); | |
1852 | set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO); | |
1853 | set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO); | |
1854 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); | |
1855 | set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO); | |
0b5a5063 | 1856 | set_page_prot(level1_fixmap_pgt, PAGE_KERNEL_RO); |
4e44e44b MR |
1857 | |
1858 | /* Pin down new L4 */ | |
1859 | pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE, | |
1860 | PFN_DOWN(__pa_symbol(init_level4_pgt))); | |
1861 | ||
1862 | /* Unpin Xen-provided one */ | |
1863 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
319f3ba5 | 1864 | |
4e44e44b MR |
1865 | /* |
1866 | * At this stage there can be no user pgd, and no page | |
1867 | * structure to attach it to, so make sure we just set kernel | |
1868 | * pgd. | |
1869 | */ | |
1870 | xen_mc_batch(); | |
1871 | __xen_write_cr3(true, __pa(init_level4_pgt)); | |
1872 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
1873 | } else | |
1874 | native_write_cr3(__pa(init_level4_pgt)); | |
319f3ba5 | 1875 | |
488f046d KRW |
1876 | /* We can't that easily rip out L3 and L2, as the Xen pagetables are |
1877 | * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ... for | |
1878 | * the initial domain. For guests using the toolstack, they are in: | |
1879 | * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only | |
1880 | * rip out the [L4] (pgd), but for guests we shave off three pages. | |
1881 | */ | |
1882 | for (i = 0; i < ARRAY_SIZE(addr); i++) | |
1883 | check_pt_base(&pt_base, &pt_end, addr[i]); | |
319f3ba5 | 1884 | |
488f046d KRW |
1885 | /* Our (by three pages) smaller Xen pagetable that we are using */ |
1886 | memblock_reserve(PFN_PHYS(pt_base), (pt_end - pt_base) * PAGE_SIZE); | |
7f914062 KRW |
1887 | /* Revector the xen_start_info */ |
1888 | xen_start_info = (struct start_info *)__va(__pa(xen_start_info)); | |
319f3ba5 JF |
1889 | } |
1890 | #else /* !CONFIG_X86_64 */ | |
5b5c1af1 IC |
1891 | static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD); |
1892 | static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD); | |
1893 | ||
3f508953 | 1894 | static void __init xen_write_cr3_init(unsigned long cr3) |
5b5c1af1 IC |
1895 | { |
1896 | unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir)); | |
1897 | ||
1898 | BUG_ON(read_cr3() != __pa(initial_page_table)); | |
1899 | BUG_ON(cr3 != __pa(swapper_pg_dir)); | |
1900 | ||
1901 | /* | |
1902 | * We are switching to swapper_pg_dir for the first time (from | |
1903 | * initial_page_table) and therefore need to mark that page | |
1904 | * read-only and then pin it. | |
1905 | * | |
1906 | * Xen disallows sharing of kernel PMDs for PAE | |
1907 | * guests. Therefore we must copy the kernel PMD from | |
1908 | * initial_page_table into a new kernel PMD to be used in | |
1909 | * swapper_pg_dir. | |
1910 | */ | |
1911 | swapper_kernel_pmd = | |
1912 | extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE); | |
ae895ed7 | 1913 | copy_page(swapper_kernel_pmd, initial_kernel_pmd); |
5b5c1af1 IC |
1914 | swapper_pg_dir[KERNEL_PGD_BOUNDARY] = |
1915 | __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT); | |
1916 | set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO); | |
1917 | ||
1918 | set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO); | |
1919 | xen_write_cr3(cr3); | |
1920 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn); | |
1921 | ||
1922 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, | |
1923 | PFN_DOWN(__pa(initial_page_table))); | |
1924 | set_page_prot(initial_page_table, PAGE_KERNEL); | |
1925 | set_page_prot(initial_kernel_pmd, PAGE_KERNEL); | |
1926 | ||
1927 | pv_mmu_ops.write_cr3 = &xen_write_cr3; | |
1928 | } | |
319f3ba5 | 1929 | |
3699aad0 | 1930 | void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn) |
319f3ba5 JF |
1931 | { |
1932 | pmd_t *kernel_pmd; | |
1933 | ||
5b5c1af1 IC |
1934 | initial_kernel_pmd = |
1935 | extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE); | |
f0991802 | 1936 | |
a91d9287 SS |
1937 | max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->pt_base) + |
1938 | xen_start_info->nr_pt_frames * PAGE_SIZE + | |
1939 | 512*1024); | |
319f3ba5 JF |
1940 | |
1941 | kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd); | |
ae895ed7 | 1942 | copy_page(initial_kernel_pmd, kernel_pmd); |
319f3ba5 | 1943 | |
5b5c1af1 | 1944 | xen_map_identity_early(initial_kernel_pmd, max_pfn); |
319f3ba5 | 1945 | |
ae895ed7 | 1946 | copy_page(initial_page_table, pgd); |
5b5c1af1 IC |
1947 | initial_page_table[KERNEL_PGD_BOUNDARY] = |
1948 | __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT); | |
319f3ba5 | 1949 | |
5b5c1af1 IC |
1950 | set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO); |
1951 | set_page_prot(initial_page_table, PAGE_KERNEL_RO); | |
319f3ba5 JF |
1952 | set_page_prot(empty_zero_page, PAGE_KERNEL_RO); |
1953 | ||
1954 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
1955 | ||
5b5c1af1 IC |
1956 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, |
1957 | PFN_DOWN(__pa(initial_page_table))); | |
1958 | xen_write_cr3(__pa(initial_page_table)); | |
319f3ba5 | 1959 | |
24aa0788 | 1960 | memblock_reserve(__pa(xen_start_info->pt_base), |
dc6821e0 | 1961 | xen_start_info->nr_pt_frames * PAGE_SIZE); |
319f3ba5 JF |
1962 | } |
1963 | #endif /* CONFIG_X86_64 */ | |
1964 | ||
98511f35 JF |
1965 | static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss; |
1966 | ||
3b3809ac | 1967 | static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot) |
319f3ba5 JF |
1968 | { |
1969 | pte_t pte; | |
1970 | ||
1971 | phys >>= PAGE_SHIFT; | |
1972 | ||
1973 | switch (idx) { | |
1974 | case FIX_BTMAP_END ... FIX_BTMAP_BEGIN: | |
4eefbe79 | 1975 | case FIX_RO_IDT: |
319f3ba5 JF |
1976 | #ifdef CONFIG_X86_32 |
1977 | case FIX_WP_TEST: | |
319f3ba5 JF |
1978 | # ifdef CONFIG_HIGHMEM |
1979 | case FIX_KMAP_BEGIN ... FIX_KMAP_END: | |
1980 | # endif | |
1ad83c85 | 1981 | #elif defined(CONFIG_X86_VSYSCALL_EMULATION) |
f40c3300 | 1982 | case VSYSCALL_PAGE: |
319f3ba5 | 1983 | #endif |
3ecb1b7d JF |
1984 | case FIX_TEXT_POKE0: |
1985 | case FIX_TEXT_POKE1: | |
1986 | /* All local page mappings */ | |
319f3ba5 JF |
1987 | pte = pfn_pte(phys, prot); |
1988 | break; | |
1989 | ||
98511f35 JF |
1990 | #ifdef CONFIG_X86_LOCAL_APIC |
1991 | case FIX_APIC_BASE: /* maps dummy local APIC */ | |
1992 | pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL); | |
1993 | break; | |
1994 | #endif | |
1995 | ||
1996 | #ifdef CONFIG_X86_IO_APIC | |
1997 | case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END: | |
1998 | /* | |
1999 | * We just don't map the IO APIC - all access is via | |
2000 | * hypercalls. Keep the address in the pte for reference. | |
2001 | */ | |
27abd14b | 2002 | pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL); |
98511f35 JF |
2003 | break; |
2004 | #endif | |
2005 | ||
c0011dbf JF |
2006 | case FIX_PARAVIRT_BOOTMAP: |
2007 | /* This is an MFN, but it isn't an IO mapping from the | |
2008 | IO domain */ | |
319f3ba5 JF |
2009 | pte = mfn_pte(phys, prot); |
2010 | break; | |
c0011dbf JF |
2011 | |
2012 | default: | |
2013 | /* By default, set_fixmap is used for hardware mappings */ | |
7f2f8822 | 2014 | pte = mfn_pte(phys, prot); |
c0011dbf | 2015 | break; |
319f3ba5 JF |
2016 | } |
2017 | ||
2018 | __native_set_fixmap(idx, pte); | |
2019 | ||
1ad83c85 | 2020 | #ifdef CONFIG_X86_VSYSCALL_EMULATION |
319f3ba5 JF |
2021 | /* Replicate changes to map the vsyscall page into the user |
2022 | pagetable vsyscall mapping. */ | |
f40c3300 | 2023 | if (idx == VSYSCALL_PAGE) { |
319f3ba5 JF |
2024 | unsigned long vaddr = __fix_to_virt(idx); |
2025 | set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte); | |
2026 | } | |
2027 | #endif | |
2028 | } | |
2029 | ||
3f508953 | 2030 | static void __init xen_post_allocator_init(void) |
319f3ba5 | 2031 | { |
4e44e44b MR |
2032 | if (xen_feature(XENFEAT_auto_translated_physmap)) |
2033 | return; | |
2034 | ||
319f3ba5 JF |
2035 | pv_mmu_ops.set_pte = xen_set_pte; |
2036 | pv_mmu_ops.set_pmd = xen_set_pmd; | |
2037 | pv_mmu_ops.set_pud = xen_set_pud; | |
2038 | #if PAGETABLE_LEVELS == 4 | |
2039 | pv_mmu_ops.set_pgd = xen_set_pgd; | |
2040 | #endif | |
2041 | ||
2042 | /* This will work as long as patching hasn't happened yet | |
2043 | (which it hasn't) */ | |
2044 | pv_mmu_ops.alloc_pte = xen_alloc_pte; | |
2045 | pv_mmu_ops.alloc_pmd = xen_alloc_pmd; | |
2046 | pv_mmu_ops.release_pte = xen_release_pte; | |
2047 | pv_mmu_ops.release_pmd = xen_release_pmd; | |
2048 | #if PAGETABLE_LEVELS == 4 | |
2049 | pv_mmu_ops.alloc_pud = xen_alloc_pud; | |
2050 | pv_mmu_ops.release_pud = xen_release_pud; | |
2051 | #endif | |
2052 | ||
2053 | #ifdef CONFIG_X86_64 | |
d3eb2c89 | 2054 | pv_mmu_ops.write_cr3 = &xen_write_cr3; |
319f3ba5 JF |
2055 | SetPagePinned(virt_to_page(level3_user_vsyscall)); |
2056 | #endif | |
2057 | xen_mark_init_mm_pinned(); | |
2058 | } | |
2059 | ||
b407fc57 JF |
2060 | static void xen_leave_lazy_mmu(void) |
2061 | { | |
5caecb94 | 2062 | preempt_disable(); |
b407fc57 JF |
2063 | xen_mc_flush(); |
2064 | paravirt_leave_lazy_mmu(); | |
5caecb94 | 2065 | preempt_enable(); |
b407fc57 | 2066 | } |
319f3ba5 | 2067 | |
3f508953 | 2068 | static const struct pv_mmu_ops xen_mmu_ops __initconst = { |
319f3ba5 JF |
2069 | .read_cr2 = xen_read_cr2, |
2070 | .write_cr2 = xen_write_cr2, | |
2071 | ||
2072 | .read_cr3 = xen_read_cr3, | |
5b5c1af1 | 2073 | .write_cr3 = xen_write_cr3_init, |
319f3ba5 JF |
2074 | |
2075 | .flush_tlb_user = xen_flush_tlb, | |
2076 | .flush_tlb_kernel = xen_flush_tlb, | |
2077 | .flush_tlb_single = xen_flush_tlb_single, | |
2078 | .flush_tlb_others = xen_flush_tlb_others, | |
2079 | ||
2080 | .pte_update = paravirt_nop, | |
2081 | .pte_update_defer = paravirt_nop, | |
2082 | ||
2083 | .pgd_alloc = xen_pgd_alloc, | |
2084 | .pgd_free = xen_pgd_free, | |
2085 | ||
2086 | .alloc_pte = xen_alloc_pte_init, | |
2087 | .release_pte = xen_release_pte_init, | |
b96229b5 | 2088 | .alloc_pmd = xen_alloc_pmd_init, |
b96229b5 | 2089 | .release_pmd = xen_release_pmd_init, |
319f3ba5 | 2090 | |
319f3ba5 | 2091 | .set_pte = xen_set_pte_init, |
319f3ba5 JF |
2092 | .set_pte_at = xen_set_pte_at, |
2093 | .set_pmd = xen_set_pmd_hyper, | |
2094 | ||
2095 | .ptep_modify_prot_start = __ptep_modify_prot_start, | |
2096 | .ptep_modify_prot_commit = __ptep_modify_prot_commit, | |
2097 | ||
da5de7c2 JF |
2098 | .pte_val = PV_CALLEE_SAVE(xen_pte_val), |
2099 | .pgd_val = PV_CALLEE_SAVE(xen_pgd_val), | |
319f3ba5 | 2100 | |
da5de7c2 JF |
2101 | .make_pte = PV_CALLEE_SAVE(xen_make_pte), |
2102 | .make_pgd = PV_CALLEE_SAVE(xen_make_pgd), | |
319f3ba5 JF |
2103 | |
2104 | #ifdef CONFIG_X86_PAE | |
2105 | .set_pte_atomic = xen_set_pte_atomic, | |
319f3ba5 JF |
2106 | .pte_clear = xen_pte_clear, |
2107 | .pmd_clear = xen_pmd_clear, | |
2108 | #endif /* CONFIG_X86_PAE */ | |
2109 | .set_pud = xen_set_pud_hyper, | |
2110 | ||
da5de7c2 JF |
2111 | .make_pmd = PV_CALLEE_SAVE(xen_make_pmd), |
2112 | .pmd_val = PV_CALLEE_SAVE(xen_pmd_val), | |
319f3ba5 JF |
2113 | |
2114 | #if PAGETABLE_LEVELS == 4 | |
da5de7c2 JF |
2115 | .pud_val = PV_CALLEE_SAVE(xen_pud_val), |
2116 | .make_pud = PV_CALLEE_SAVE(xen_make_pud), | |
319f3ba5 JF |
2117 | .set_pgd = xen_set_pgd_hyper, |
2118 | ||
b96229b5 JF |
2119 | .alloc_pud = xen_alloc_pmd_init, |
2120 | .release_pud = xen_release_pmd_init, | |
319f3ba5 JF |
2121 | #endif /* PAGETABLE_LEVELS == 4 */ |
2122 | ||
2123 | .activate_mm = xen_activate_mm, | |
2124 | .dup_mmap = xen_dup_mmap, | |
2125 | .exit_mmap = xen_exit_mmap, | |
2126 | ||
2127 | .lazy_mode = { | |
2128 | .enter = paravirt_enter_lazy_mmu, | |
b407fc57 | 2129 | .leave = xen_leave_lazy_mmu, |
511ba86e | 2130 | .flush = paravirt_flush_lazy_mmu, |
319f3ba5 JF |
2131 | }, |
2132 | ||
2133 | .set_fixmap = xen_set_fixmap, | |
2134 | }; | |
2135 | ||
030cb6c0 TG |
2136 | void __init xen_init_mmu_ops(void) |
2137 | { | |
7737b215 | 2138 | x86_init.paging.pagetable_init = xen_pagetable_init; |
76bcceff MR |
2139 | |
2140 | /* Optimization - we can use the HVM one but it has no idea which | |
2141 | * VCPUs are descheduled - which means that it will needlessly IPI | |
2142 | * them. Xen knows so let it do the job. | |
2143 | */ | |
2144 | if (xen_feature(XENFEAT_auto_translated_physmap)) { | |
2145 | pv_mmu_ops.flush_tlb_others = xen_flush_tlb_others; | |
2146 | return; | |
2147 | } | |
030cb6c0 | 2148 | pv_mmu_ops = xen_mmu_ops; |
d2cb2145 | 2149 | |
98511f35 | 2150 | memset(dummy_mapping, 0xff, PAGE_SIZE); |
030cb6c0 | 2151 | } |
319f3ba5 | 2152 | |
08bbc9da AN |
2153 | /* Protected by xen_reservation_lock. */ |
2154 | #define MAX_CONTIG_ORDER 9 /* 2MB */ | |
2155 | static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER]; | |
2156 | ||
2157 | #define VOID_PTE (mfn_pte(0, __pgprot(0))) | |
2158 | static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order, | |
2159 | unsigned long *in_frames, | |
2160 | unsigned long *out_frames) | |
2161 | { | |
2162 | int i; | |
2163 | struct multicall_space mcs; | |
2164 | ||
2165 | xen_mc_batch(); | |
2166 | for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) { | |
2167 | mcs = __xen_mc_entry(0); | |
2168 | ||
2169 | if (in_frames) | |
2170 | in_frames[i] = virt_to_mfn(vaddr); | |
2171 | ||
2172 | MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0); | |
6eaa412f | 2173 | __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY); |
08bbc9da AN |
2174 | |
2175 | if (out_frames) | |
2176 | out_frames[i] = virt_to_pfn(vaddr); | |
2177 | } | |
2178 | xen_mc_issue(0); | |
2179 | } | |
2180 | ||
2181 | /* | |
2182 | * Update the pfn-to-mfn mappings for a virtual address range, either to | |
2183 | * point to an array of mfns, or contiguously from a single starting | |
2184 | * mfn. | |
2185 | */ | |
2186 | static void xen_remap_exchanged_ptes(unsigned long vaddr, int order, | |
2187 | unsigned long *mfns, | |
2188 | unsigned long first_mfn) | |
2189 | { | |
2190 | unsigned i, limit; | |
2191 | unsigned long mfn; | |
2192 | ||
2193 | xen_mc_batch(); | |
2194 | ||
2195 | limit = 1u << order; | |
2196 | for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) { | |
2197 | struct multicall_space mcs; | |
2198 | unsigned flags; | |
2199 | ||
2200 | mcs = __xen_mc_entry(0); | |
2201 | if (mfns) | |
2202 | mfn = mfns[i]; | |
2203 | else | |
2204 | mfn = first_mfn + i; | |
2205 | ||
2206 | if (i < (limit - 1)) | |
2207 | flags = 0; | |
2208 | else { | |
2209 | if (order == 0) | |
2210 | flags = UVMF_INVLPG | UVMF_ALL; | |
2211 | else | |
2212 | flags = UVMF_TLB_FLUSH | UVMF_ALL; | |
2213 | } | |
2214 | ||
2215 | MULTI_update_va_mapping(mcs.mc, vaddr, | |
2216 | mfn_pte(mfn, PAGE_KERNEL), flags); | |
2217 | ||
2218 | set_phys_to_machine(virt_to_pfn(vaddr), mfn); | |
2219 | } | |
2220 | ||
2221 | xen_mc_issue(0); | |
2222 | } | |
2223 | ||
2224 | /* | |
2225 | * Perform the hypercall to exchange a region of our pfns to point to | |
2226 | * memory with the required contiguous alignment. Takes the pfns as | |
2227 | * input, and populates mfns as output. | |
2228 | * | |
2229 | * Returns a success code indicating whether the hypervisor was able to | |
2230 | * satisfy the request or not. | |
2231 | */ | |
2232 | static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in, | |
2233 | unsigned long *pfns_in, | |
2234 | unsigned long extents_out, | |
2235 | unsigned int order_out, | |
2236 | unsigned long *mfns_out, | |
2237 | unsigned int address_bits) | |
2238 | { | |
2239 | long rc; | |
2240 | int success; | |
2241 | ||
2242 | struct xen_memory_exchange exchange = { | |
2243 | .in = { | |
2244 | .nr_extents = extents_in, | |
2245 | .extent_order = order_in, | |
2246 | .extent_start = pfns_in, | |
2247 | .domid = DOMID_SELF | |
2248 | }, | |
2249 | .out = { | |
2250 | .nr_extents = extents_out, | |
2251 | .extent_order = order_out, | |
2252 | .extent_start = mfns_out, | |
2253 | .address_bits = address_bits, | |
2254 | .domid = DOMID_SELF | |
2255 | } | |
2256 | }; | |
2257 | ||
2258 | BUG_ON(extents_in << order_in != extents_out << order_out); | |
2259 | ||
2260 | rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange); | |
2261 | success = (exchange.nr_exchanged == extents_in); | |
2262 | ||
2263 | BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0))); | |
2264 | BUG_ON(success && (rc != 0)); | |
2265 | ||
2266 | return success; | |
2267 | } | |
2268 | ||
1b65c4e5 | 2269 | int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order, |
69908907 SS |
2270 | unsigned int address_bits, |
2271 | dma_addr_t *dma_handle) | |
08bbc9da AN |
2272 | { |
2273 | unsigned long *in_frames = discontig_frames, out_frame; | |
2274 | unsigned long flags; | |
2275 | int success; | |
1b65c4e5 | 2276 | unsigned long vstart = (unsigned long)phys_to_virt(pstart); |
08bbc9da AN |
2277 | |
2278 | /* | |
2279 | * Currently an auto-translated guest will not perform I/O, nor will | |
2280 | * it require PAE page directories below 4GB. Therefore any calls to | |
2281 | * this function are redundant and can be ignored. | |
2282 | */ | |
2283 | ||
2284 | if (xen_feature(XENFEAT_auto_translated_physmap)) | |
2285 | return 0; | |
2286 | ||
2287 | if (unlikely(order > MAX_CONTIG_ORDER)) | |
2288 | return -ENOMEM; | |
2289 | ||
2290 | memset((void *) vstart, 0, PAGE_SIZE << order); | |
2291 | ||
08bbc9da AN |
2292 | spin_lock_irqsave(&xen_reservation_lock, flags); |
2293 | ||
2294 | /* 1. Zap current PTEs, remembering MFNs. */ | |
2295 | xen_zap_pfn_range(vstart, order, in_frames, NULL); | |
2296 | ||
2297 | /* 2. Get a new contiguous memory extent. */ | |
2298 | out_frame = virt_to_pfn(vstart); | |
2299 | success = xen_exchange_memory(1UL << order, 0, in_frames, | |
2300 | 1, order, &out_frame, | |
2301 | address_bits); | |
2302 | ||
2303 | /* 3. Map the new extent in place of old pages. */ | |
2304 | if (success) | |
2305 | xen_remap_exchanged_ptes(vstart, order, NULL, out_frame); | |
2306 | else | |
2307 | xen_remap_exchanged_ptes(vstart, order, in_frames, 0); | |
2308 | ||
2309 | spin_unlock_irqrestore(&xen_reservation_lock, flags); | |
2310 | ||
69908907 | 2311 | *dma_handle = virt_to_machine(vstart).maddr; |
08bbc9da AN |
2312 | return success ? 0 : -ENOMEM; |
2313 | } | |
2314 | EXPORT_SYMBOL_GPL(xen_create_contiguous_region); | |
2315 | ||
1b65c4e5 | 2316 | void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order) |
08bbc9da AN |
2317 | { |
2318 | unsigned long *out_frames = discontig_frames, in_frame; | |
2319 | unsigned long flags; | |
2320 | int success; | |
1b65c4e5 | 2321 | unsigned long vstart; |
08bbc9da AN |
2322 | |
2323 | if (xen_feature(XENFEAT_auto_translated_physmap)) | |
2324 | return; | |
2325 | ||
2326 | if (unlikely(order > MAX_CONTIG_ORDER)) | |
2327 | return; | |
2328 | ||
1b65c4e5 | 2329 | vstart = (unsigned long)phys_to_virt(pstart); |
08bbc9da AN |
2330 | memset((void *) vstart, 0, PAGE_SIZE << order); |
2331 | ||
08bbc9da AN |
2332 | spin_lock_irqsave(&xen_reservation_lock, flags); |
2333 | ||
2334 | /* 1. Find start MFN of contiguous extent. */ | |
2335 | in_frame = virt_to_mfn(vstart); | |
2336 | ||
2337 | /* 2. Zap current PTEs. */ | |
2338 | xen_zap_pfn_range(vstart, order, NULL, out_frames); | |
2339 | ||
2340 | /* 3. Do the exchange for non-contiguous MFNs. */ | |
2341 | success = xen_exchange_memory(1, order, &in_frame, 1UL << order, | |
2342 | 0, out_frames, 0); | |
2343 | ||
2344 | /* 4. Map new pages in place of old pages. */ | |
2345 | if (success) | |
2346 | xen_remap_exchanged_ptes(vstart, order, out_frames, 0); | |
2347 | else | |
2348 | xen_remap_exchanged_ptes(vstart, order, NULL, in_frame); | |
2349 | ||
2350 | spin_unlock_irqrestore(&xen_reservation_lock, flags); | |
030cb6c0 | 2351 | } |
08bbc9da | 2352 | EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region); |
319f3ba5 | 2353 | |
ca65f9fc | 2354 | #ifdef CONFIG_XEN_PVHVM |
34b6f01a OH |
2355 | #ifdef CONFIG_PROC_VMCORE |
2356 | /* | |
2357 | * This function is used in two contexts: | |
2358 | * - the kdump kernel has to check whether a pfn of the crashed kernel | |
2359 | * was a ballooned page. vmcore is using this function to decide | |
2360 | * whether to access a pfn of the crashed kernel. | |
2361 | * - the kexec kernel has to check whether a pfn was ballooned by the | |
2362 | * previous kernel. If the pfn is ballooned, handle it properly. | |
2363 | * Returns 0 if the pfn is not backed by a RAM page, the caller may | |
2364 | * handle the pfn special in this case. | |
2365 | */ | |
2366 | static int xen_oldmem_pfn_is_ram(unsigned long pfn) | |
2367 | { | |
2368 | struct xen_hvm_get_mem_type a = { | |
2369 | .domid = DOMID_SELF, | |
2370 | .pfn = pfn, | |
2371 | }; | |
2372 | int ram; | |
2373 | ||
2374 | if (HYPERVISOR_hvm_op(HVMOP_get_mem_type, &a)) | |
2375 | return -ENXIO; | |
2376 | ||
2377 | switch (a.mem_type) { | |
2378 | case HVMMEM_mmio_dm: | |
2379 | ram = 0; | |
2380 | break; | |
2381 | case HVMMEM_ram_rw: | |
2382 | case HVMMEM_ram_ro: | |
2383 | default: | |
2384 | ram = 1; | |
2385 | break; | |
2386 | } | |
2387 | ||
2388 | return ram; | |
2389 | } | |
2390 | #endif | |
2391 | ||
59151001 SS |
2392 | static void xen_hvm_exit_mmap(struct mm_struct *mm) |
2393 | { | |
2394 | struct xen_hvm_pagetable_dying a; | |
2395 | int rc; | |
2396 | ||
2397 | a.domid = DOMID_SELF; | |
2398 | a.gpa = __pa(mm->pgd); | |
2399 | rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a); | |
2400 | WARN_ON_ONCE(rc < 0); | |
2401 | } | |
2402 | ||
2403 | static int is_pagetable_dying_supported(void) | |
2404 | { | |
2405 | struct xen_hvm_pagetable_dying a; | |
2406 | int rc = 0; | |
2407 | ||
2408 | a.domid = DOMID_SELF; | |
2409 | a.gpa = 0x00; | |
2410 | rc = HYPERVISOR_hvm_op(HVMOP_pagetable_dying, &a); | |
2411 | if (rc < 0) { | |
2412 | printk(KERN_DEBUG "HVMOP_pagetable_dying not supported\n"); | |
2413 | return 0; | |
2414 | } | |
2415 | return 1; | |
2416 | } | |
2417 | ||
2418 | void __init xen_hvm_init_mmu_ops(void) | |
2419 | { | |
2420 | if (is_pagetable_dying_supported()) | |
2421 | pv_mmu_ops.exit_mmap = xen_hvm_exit_mmap; | |
34b6f01a OH |
2422 | #ifdef CONFIG_PROC_VMCORE |
2423 | register_oldmem_pfn_is_ram(&xen_oldmem_pfn_is_ram); | |
2424 | #endif | |
59151001 | 2425 | } |
ca65f9fc | 2426 | #endif |
59151001 | 2427 | |
77945ca7 MR |
2428 | #ifdef CONFIG_XEN_PVH |
2429 | /* | |
2430 | * Map foreign gfn (fgfn), to local pfn (lpfn). This for the user | |
2431 | * space creating new guest on pvh dom0 and needing to map domU pages. | |
2432 | */ | |
2433 | static int xlate_add_to_p2m(unsigned long lpfn, unsigned long fgfn, | |
2434 | unsigned int domid) | |
2435 | { | |
2436 | int rc, err = 0; | |
2437 | xen_pfn_t gpfn = lpfn; | |
2438 | xen_ulong_t idx = fgfn; | |
2439 | ||
2440 | struct xen_add_to_physmap_range xatp = { | |
2441 | .domid = DOMID_SELF, | |
2442 | .foreign_domid = domid, | |
2443 | .size = 1, | |
2444 | .space = XENMAPSPACE_gmfn_foreign, | |
2445 | }; | |
2446 | set_xen_guest_handle(xatp.idxs, &idx); | |
2447 | set_xen_guest_handle(xatp.gpfns, &gpfn); | |
2448 | set_xen_guest_handle(xatp.errs, &err); | |
2449 | ||
2450 | rc = HYPERVISOR_memory_op(XENMEM_add_to_physmap_range, &xatp); | |
2451 | if (rc < 0) | |
2452 | return rc; | |
2453 | return err; | |
2454 | } | |
2455 | ||
2456 | static int xlate_remove_from_p2m(unsigned long spfn, int count) | |
2457 | { | |
2458 | struct xen_remove_from_physmap xrp; | |
2459 | int i, rc; | |
2460 | ||
2461 | for (i = 0; i < count; i++) { | |
2462 | xrp.domid = DOMID_SELF; | |
2463 | xrp.gpfn = spfn+i; | |
2464 | rc = HYPERVISOR_memory_op(XENMEM_remove_from_physmap, &xrp); | |
2465 | if (rc) | |
2466 | break; | |
2467 | } | |
2468 | return rc; | |
2469 | } | |
2470 | ||
2471 | struct xlate_remap_data { | |
2472 | unsigned long fgfn; /* foreign domain's gfn */ | |
2473 | pgprot_t prot; | |
2474 | domid_t domid; | |
2475 | int index; | |
2476 | struct page **pages; | |
2477 | }; | |
2478 | ||
2479 | static int xlate_map_pte_fn(pte_t *ptep, pgtable_t token, unsigned long addr, | |
2480 | void *data) | |
2481 | { | |
2482 | int rc; | |
2483 | struct xlate_remap_data *remap = data; | |
2484 | unsigned long pfn = page_to_pfn(remap->pages[remap->index++]); | |
2485 | pte_t pteval = pte_mkspecial(pfn_pte(pfn, remap->prot)); | |
2486 | ||
2487 | rc = xlate_add_to_p2m(pfn, remap->fgfn, remap->domid); | |
2488 | if (rc) | |
2489 | return rc; | |
2490 | native_set_pte(ptep, pteval); | |
2491 | ||
2492 | return 0; | |
2493 | } | |
2494 | ||
2495 | static int xlate_remap_gfn_range(struct vm_area_struct *vma, | |
2496 | unsigned long addr, unsigned long mfn, | |
2497 | int nr, pgprot_t prot, unsigned domid, | |
2498 | struct page **pages) | |
2499 | { | |
2500 | int err; | |
2501 | struct xlate_remap_data pvhdata; | |
2502 | ||
2503 | BUG_ON(!pages); | |
2504 | ||
2505 | pvhdata.fgfn = mfn; | |
2506 | pvhdata.prot = prot; | |
2507 | pvhdata.domid = domid; | |
2508 | pvhdata.index = 0; | |
2509 | pvhdata.pages = pages; | |
2510 | err = apply_to_page_range(vma->vm_mm, addr, nr << PAGE_SHIFT, | |
2511 | xlate_map_pte_fn, &pvhdata); | |
2512 | flush_tlb_all(); | |
2513 | return err; | |
2514 | } | |
2515 | #endif | |
2516 | ||
de1ef206 IC |
2517 | #define REMAP_BATCH_SIZE 16 |
2518 | ||
2519 | struct remap_data { | |
2520 | unsigned long mfn; | |
2521 | pgprot_t prot; | |
2522 | struct mmu_update *mmu_update; | |
2523 | }; | |
2524 | ||
2525 | static int remap_area_mfn_pte_fn(pte_t *ptep, pgtable_t token, | |
2526 | unsigned long addr, void *data) | |
2527 | { | |
2528 | struct remap_data *rmd = data; | |
f59c5145 | 2529 | pte_t pte = pte_mkspecial(mfn_pte(rmd->mfn++, rmd->prot)); |
de1ef206 | 2530 | |
d5108316 | 2531 | rmd->mmu_update->ptr = virt_to_machine(ptep).maddr; |
de1ef206 IC |
2532 | rmd->mmu_update->val = pte_val_ma(pte); |
2533 | rmd->mmu_update++; | |
2534 | ||
2535 | return 0; | |
2536 | } | |
2537 | ||
2538 | int xen_remap_domain_mfn_range(struct vm_area_struct *vma, | |
2539 | unsigned long addr, | |
7892f692 | 2540 | xen_pfn_t mfn, int nr, |
9a032e39 IC |
2541 | pgprot_t prot, unsigned domid, |
2542 | struct page **pages) | |
2543 | ||
de1ef206 IC |
2544 | { |
2545 | struct remap_data rmd; | |
2546 | struct mmu_update mmu_update[REMAP_BATCH_SIZE]; | |
2547 | int batch; | |
2548 | unsigned long range; | |
2549 | int err = 0; | |
2550 | ||
314e51b9 | 2551 | BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_IO)) == (VM_PFNMAP | VM_IO))); |
de1ef206 | 2552 | |
77945ca7 MR |
2553 | if (xen_feature(XENFEAT_auto_translated_physmap)) { |
2554 | #ifdef CONFIG_XEN_PVH | |
2555 | /* We need to update the local page tables and the xen HAP */ | |
2556 | return xlate_remap_gfn_range(vma, addr, mfn, nr, prot, | |
2557 | domid, pages); | |
2558 | #else | |
2559 | return -EINVAL; | |
2560 | #endif | |
2561 | } | |
2562 | ||
de1ef206 IC |
2563 | rmd.mfn = mfn; |
2564 | rmd.prot = prot; | |
2565 | ||
2566 | while (nr) { | |
2567 | batch = min(REMAP_BATCH_SIZE, nr); | |
2568 | range = (unsigned long)batch << PAGE_SHIFT; | |
2569 | ||
2570 | rmd.mmu_update = mmu_update; | |
2571 | err = apply_to_page_range(vma->vm_mm, addr, range, | |
2572 | remap_area_mfn_pte_fn, &rmd); | |
2573 | if (err) | |
2574 | goto out; | |
2575 | ||
69870a84 DV |
2576 | err = HYPERVISOR_mmu_update(mmu_update, batch, NULL, domid); |
2577 | if (err < 0) | |
de1ef206 IC |
2578 | goto out; |
2579 | ||
2580 | nr -= batch; | |
2581 | addr += range; | |
2582 | } | |
2583 | ||
2584 | err = 0; | |
2585 | out: | |
2586 | ||
95a7d768 | 2587 | xen_flush_tlb_all(); |
de1ef206 IC |
2588 | |
2589 | return err; | |
2590 | } | |
2591 | EXPORT_SYMBOL_GPL(xen_remap_domain_mfn_range); | |
9a032e39 IC |
2592 | |
2593 | /* Returns: 0 success */ | |
2594 | int xen_unmap_domain_mfn_range(struct vm_area_struct *vma, | |
2595 | int numpgs, struct page **pages) | |
2596 | { | |
2597 | if (!pages || !xen_feature(XENFEAT_auto_translated_physmap)) | |
2598 | return 0; | |
2599 | ||
77945ca7 MR |
2600 | #ifdef CONFIG_XEN_PVH |
2601 | while (numpgs--) { | |
2602 | /* | |
2603 | * The mmu has already cleaned up the process mmu | |
2604 | * resources at this point (lookup_address will return | |
2605 | * NULL). | |
2606 | */ | |
2607 | unsigned long pfn = page_to_pfn(pages[numpgs]); | |
2608 | ||
2609 | xlate_remove_from_p2m(pfn, 1); | |
2610 | } | |
2611 | /* | |
2612 | * We don't need to flush tlbs because as part of | |
2613 | * xlate_remove_from_p2m, the hypervisor will do tlb flushes | |
2614 | * after removing the p2m entries from the EPT/NPT | |
2615 | */ | |
2616 | return 0; | |
2617 | #else | |
9a032e39 | 2618 | return -EINVAL; |
77945ca7 | 2619 | #endif |
9a032e39 IC |
2620 | } |
2621 | EXPORT_SYMBOL_GPL(xen_unmap_domain_mfn_range); |