Commit | Line | Data |
---|---|---|
7e0563de VK |
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 | */ | |
41 | #include <linux/sched/mm.h> | |
42 | #include <linux/highmem.h> | |
43 | #include <linux/debugfs.h> | |
44 | #include <linux/bug.h> | |
45 | #include <linux/vmalloc.h> | |
46 | #include <linux/export.h> | |
47 | #include <linux/init.h> | |
48 | #include <linux/gfp.h> | |
49 | #include <linux/memblock.h> | |
50 | #include <linux/seq_file.h> | |
51 | #include <linux/crash_dump.h> | |
29985b09 JG |
52 | #ifdef CONFIG_KEXEC_CORE |
53 | #include <linux/kexec.h> | |
54 | #endif | |
7e0563de VK |
55 | |
56 | #include <trace/events/xen.h> | |
57 | ||
58 | #include <asm/pgtable.h> | |
59 | #include <asm/tlbflush.h> | |
60 | #include <asm/fixmap.h> | |
61 | #include <asm/mmu_context.h> | |
62 | #include <asm/setup.h> | |
63 | #include <asm/paravirt.h> | |
64 | #include <asm/e820/api.h> | |
65 | #include <asm/linkage.h> | |
66 | #include <asm/page.h> | |
67 | #include <asm/init.h> | |
68 | #include <asm/pat.h> | |
69 | #include <asm/smp.h> | |
70 | ||
71 | #include <asm/xen/hypercall.h> | |
72 | #include <asm/xen/hypervisor.h> | |
73 | ||
74 | #include <xen/xen.h> | |
75 | #include <xen/page.h> | |
76 | #include <xen/interface/xen.h> | |
77 | #include <xen/interface/hvm/hvm_op.h> | |
78 | #include <xen/interface/version.h> | |
79 | #include <xen/interface/memory.h> | |
80 | #include <xen/hvc-console.h> | |
81 | ||
82 | #include "multicalls.h" | |
83 | #include "mmu.h" | |
84 | #include "debugfs.h" | |
85 | ||
86 | #ifdef CONFIG_X86_32 | |
87 | /* | |
88 | * Identity map, in addition to plain kernel map. This needs to be | |
89 | * large enough to allocate page table pages to allocate the rest. | |
90 | * Each page can map 2MB. | |
91 | */ | |
92 | #define LEVEL1_IDENT_ENTRIES (PTRS_PER_PTE * 4) | |
93 | static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES); | |
94 | #endif | |
95 | #ifdef CONFIG_X86_64 | |
96 | /* l3 pud for userspace vsyscall mapping */ | |
97 | static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss; | |
98 | #endif /* CONFIG_X86_64 */ | |
99 | ||
100 | /* | |
101 | * Note about cr3 (pagetable base) values: | |
102 | * | |
103 | * xen_cr3 contains the current logical cr3 value; it contains the | |
104 | * last set cr3. This may not be the current effective cr3, because | |
105 | * its update may be being lazily deferred. However, a vcpu looking | |
106 | * at its own cr3 can use this value knowing that it everything will | |
107 | * be self-consistent. | |
108 | * | |
109 | * xen_current_cr3 contains the actual vcpu cr3; it is set once the | |
110 | * hypercall to set the vcpu cr3 is complete (so it may be a little | |
111 | * out of date, but it will never be set early). If one vcpu is | |
112 | * looking at another vcpu's cr3 value, it should use this variable. | |
113 | */ | |
114 | DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */ | |
115 | DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */ | |
116 | ||
117 | static phys_addr_t xen_pt_base, xen_pt_size __initdata; | |
118 | ||
119 | /* | |
120 | * Just beyond the highest usermode address. STACK_TOP_MAX has a | |
121 | * redzone above it, so round it up to a PGD boundary. | |
122 | */ | |
123 | #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK) | |
124 | ||
125 | void make_lowmem_page_readonly(void *vaddr) | |
126 | { | |
127 | pte_t *pte, ptev; | |
128 | unsigned long address = (unsigned long)vaddr; | |
129 | unsigned int level; | |
130 | ||
131 | pte = lookup_address(address, &level); | |
132 | if (pte == NULL) | |
133 | return; /* vaddr missing */ | |
134 | ||
135 | ptev = pte_wrprotect(*pte); | |
136 | ||
137 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
138 | BUG(); | |
139 | } | |
140 | ||
141 | void make_lowmem_page_readwrite(void *vaddr) | |
142 | { | |
143 | pte_t *pte, ptev; | |
144 | unsigned long address = (unsigned long)vaddr; | |
145 | unsigned int level; | |
146 | ||
147 | pte = lookup_address(address, &level); | |
148 | if (pte == NULL) | |
149 | return; /* vaddr missing */ | |
150 | ||
151 | ptev = pte_mkwrite(*pte); | |
152 | ||
153 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
154 | BUG(); | |
155 | } | |
156 | ||
157 | ||
158 | static bool xen_page_pinned(void *ptr) | |
159 | { | |
160 | struct page *page = virt_to_page(ptr); | |
161 | ||
162 | return PagePinned(page); | |
163 | } | |
164 | ||
7e0563de VK |
165 | static void xen_extend_mmu_update(const struct mmu_update *update) |
166 | { | |
167 | struct multicall_space mcs; | |
168 | struct mmu_update *u; | |
169 | ||
170 | mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u)); | |
171 | ||
172 | if (mcs.mc != NULL) { | |
173 | mcs.mc->args[1]++; | |
174 | } else { | |
175 | mcs = __xen_mc_entry(sizeof(*u)); | |
176 | MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); | |
177 | } | |
178 | ||
179 | u = mcs.args; | |
180 | *u = *update; | |
181 | } | |
182 | ||
183 | static void xen_extend_mmuext_op(const struct mmuext_op *op) | |
184 | { | |
185 | struct multicall_space mcs; | |
186 | struct mmuext_op *u; | |
187 | ||
188 | mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u)); | |
189 | ||
190 | if (mcs.mc != NULL) { | |
191 | mcs.mc->args[1]++; | |
192 | } else { | |
193 | mcs = __xen_mc_entry(sizeof(*u)); | |
194 | MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); | |
195 | } | |
196 | ||
197 | u = mcs.args; | |
198 | *u = *op; | |
199 | } | |
200 | ||
201 | static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val) | |
202 | { | |
203 | struct mmu_update u; | |
204 | ||
205 | preempt_disable(); | |
206 | ||
207 | xen_mc_batch(); | |
208 | ||
209 | /* ptr may be ioremapped for 64-bit pagetable setup */ | |
210 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
211 | u.val = pmd_val_ma(val); | |
212 | xen_extend_mmu_update(&u); | |
213 | ||
214 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
215 | ||
216 | preempt_enable(); | |
217 | } | |
218 | ||
219 | static void xen_set_pmd(pmd_t *ptr, pmd_t val) | |
220 | { | |
221 | trace_xen_mmu_set_pmd(ptr, val); | |
222 | ||
223 | /* If page is not pinned, we can just update the entry | |
224 | directly */ | |
225 | if (!xen_page_pinned(ptr)) { | |
226 | *ptr = val; | |
227 | return; | |
228 | } | |
229 | ||
230 | xen_set_pmd_hyper(ptr, val); | |
231 | } | |
232 | ||
233 | /* | |
234 | * Associate a virtual page frame with a given physical page frame | |
235 | * and protection flags for that frame. | |
236 | */ | |
237 | void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags) | |
238 | { | |
239 | set_pte_vaddr(vaddr, mfn_pte(mfn, flags)); | |
240 | } | |
241 | ||
242 | static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval) | |
243 | { | |
244 | struct mmu_update u; | |
245 | ||
246 | if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU) | |
247 | return false; | |
248 | ||
249 | xen_mc_batch(); | |
250 | ||
251 | u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE; | |
252 | u.val = pte_val_ma(pteval); | |
253 | xen_extend_mmu_update(&u); | |
254 | ||
255 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
256 | ||
257 | return true; | |
258 | } | |
259 | ||
260 | static inline void __xen_set_pte(pte_t *ptep, pte_t pteval) | |
261 | { | |
262 | if (!xen_batched_set_pte(ptep, pteval)) { | |
263 | /* | |
264 | * Could call native_set_pte() here and trap and | |
265 | * emulate the PTE write but with 32-bit guests this | |
266 | * needs two traps (one for each of the two 32-bit | |
267 | * words in the PTE) so do one hypercall directly | |
268 | * instead. | |
269 | */ | |
270 | struct mmu_update u; | |
271 | ||
272 | u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE; | |
273 | u.val = pte_val_ma(pteval); | |
274 | HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF); | |
275 | } | |
276 | } | |
277 | ||
278 | static void xen_set_pte(pte_t *ptep, pte_t pteval) | |
279 | { | |
280 | trace_xen_mmu_set_pte(ptep, pteval); | |
281 | __xen_set_pte(ptep, pteval); | |
282 | } | |
283 | ||
284 | static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr, | |
285 | pte_t *ptep, pte_t pteval) | |
286 | { | |
287 | trace_xen_mmu_set_pte_at(mm, addr, ptep, pteval); | |
288 | __xen_set_pte(ptep, pteval); | |
289 | } | |
290 | ||
291 | pte_t xen_ptep_modify_prot_start(struct mm_struct *mm, | |
292 | unsigned long addr, pte_t *ptep) | |
293 | { | |
294 | /* Just return the pte as-is. We preserve the bits on commit */ | |
295 | trace_xen_mmu_ptep_modify_prot_start(mm, addr, ptep, *ptep); | |
296 | return *ptep; | |
297 | } | |
298 | ||
299 | void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr, | |
300 | pte_t *ptep, pte_t pte) | |
301 | { | |
302 | struct mmu_update u; | |
303 | ||
304 | trace_xen_mmu_ptep_modify_prot_commit(mm, addr, ptep, pte); | |
305 | xen_mc_batch(); | |
306 | ||
307 | u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD; | |
308 | u.val = pte_val_ma(pte); | |
309 | xen_extend_mmu_update(&u); | |
310 | ||
311 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
312 | } | |
313 | ||
314 | /* Assume pteval_t is equivalent to all the other *val_t types. */ | |
315 | static pteval_t pte_mfn_to_pfn(pteval_t val) | |
316 | { | |
317 | if (val & _PAGE_PRESENT) { | |
318 | unsigned long mfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; | |
319 | unsigned long pfn = mfn_to_pfn(mfn); | |
320 | ||
321 | pteval_t flags = val & PTE_FLAGS_MASK; | |
322 | if (unlikely(pfn == ~0)) | |
323 | val = flags & ~_PAGE_PRESENT; | |
324 | else | |
325 | val = ((pteval_t)pfn << PAGE_SHIFT) | flags; | |
326 | } | |
327 | ||
328 | return val; | |
329 | } | |
330 | ||
331 | static pteval_t pte_pfn_to_mfn(pteval_t val) | |
332 | { | |
333 | if (val & _PAGE_PRESENT) { | |
334 | unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; | |
335 | pteval_t flags = val & PTE_FLAGS_MASK; | |
336 | unsigned long mfn; | |
337 | ||
989513a7 JG |
338 | mfn = __pfn_to_mfn(pfn); |
339 | ||
7e0563de VK |
340 | /* |
341 | * If there's no mfn for the pfn, then just create an | |
342 | * empty non-present pte. Unfortunately this loses | |
343 | * information about the original pfn, so | |
344 | * pte_mfn_to_pfn is asymmetric. | |
345 | */ | |
346 | if (unlikely(mfn == INVALID_P2M_ENTRY)) { | |
347 | mfn = 0; | |
348 | flags = 0; | |
349 | } else | |
350 | mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT); | |
351 | val = ((pteval_t)mfn << PAGE_SHIFT) | flags; | |
352 | } | |
353 | ||
354 | return val; | |
355 | } | |
356 | ||
357 | __visible pteval_t xen_pte_val(pte_t pte) | |
358 | { | |
359 | pteval_t pteval = pte.pte; | |
360 | ||
361 | return pte_mfn_to_pfn(pteval); | |
362 | } | |
363 | PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val); | |
364 | ||
365 | __visible pgdval_t xen_pgd_val(pgd_t pgd) | |
366 | { | |
367 | return pte_mfn_to_pfn(pgd.pgd); | |
368 | } | |
369 | PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val); | |
370 | ||
371 | __visible pte_t xen_make_pte(pteval_t pte) | |
372 | { | |
373 | pte = pte_pfn_to_mfn(pte); | |
374 | ||
375 | return native_make_pte(pte); | |
376 | } | |
377 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte); | |
378 | ||
379 | __visible pgd_t xen_make_pgd(pgdval_t pgd) | |
380 | { | |
381 | pgd = pte_pfn_to_mfn(pgd); | |
382 | return native_make_pgd(pgd); | |
383 | } | |
384 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd); | |
385 | ||
386 | __visible pmdval_t xen_pmd_val(pmd_t pmd) | |
387 | { | |
388 | return pte_mfn_to_pfn(pmd.pmd); | |
389 | } | |
390 | PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val); | |
391 | ||
392 | static void xen_set_pud_hyper(pud_t *ptr, pud_t val) | |
393 | { | |
394 | struct mmu_update u; | |
395 | ||
396 | preempt_disable(); | |
397 | ||
398 | xen_mc_batch(); | |
399 | ||
400 | /* ptr may be ioremapped for 64-bit pagetable setup */ | |
401 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
402 | u.val = pud_val_ma(val); | |
403 | xen_extend_mmu_update(&u); | |
404 | ||
405 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
406 | ||
407 | preempt_enable(); | |
408 | } | |
409 | ||
410 | static void xen_set_pud(pud_t *ptr, pud_t val) | |
411 | { | |
412 | trace_xen_mmu_set_pud(ptr, val); | |
413 | ||
414 | /* If page is not pinned, we can just update the entry | |
415 | directly */ | |
416 | if (!xen_page_pinned(ptr)) { | |
417 | *ptr = val; | |
418 | return; | |
419 | } | |
420 | ||
421 | xen_set_pud_hyper(ptr, val); | |
422 | } | |
423 | ||
424 | #ifdef CONFIG_X86_PAE | |
425 | static void xen_set_pte_atomic(pte_t *ptep, pte_t pte) | |
426 | { | |
427 | trace_xen_mmu_set_pte_atomic(ptep, pte); | |
13b23ccf | 428 | __xen_set_pte(ptep, pte); |
7e0563de VK |
429 | } |
430 | ||
431 | static void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep) | |
432 | { | |
433 | trace_xen_mmu_pte_clear(mm, addr, ptep); | |
13b23ccf | 434 | __xen_set_pte(ptep, native_make_pte(0)); |
7e0563de VK |
435 | } |
436 | ||
437 | static void xen_pmd_clear(pmd_t *pmdp) | |
438 | { | |
439 | trace_xen_mmu_pmd_clear(pmdp); | |
440 | set_pmd(pmdp, __pmd(0)); | |
441 | } | |
442 | #endif /* CONFIG_X86_PAE */ | |
443 | ||
444 | __visible pmd_t xen_make_pmd(pmdval_t pmd) | |
445 | { | |
446 | pmd = pte_pfn_to_mfn(pmd); | |
447 | return native_make_pmd(pmd); | |
448 | } | |
449 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd); | |
450 | ||
af02cd97 | 451 | #ifdef CONFIG_X86_64 |
7e0563de VK |
452 | __visible pudval_t xen_pud_val(pud_t pud) |
453 | { | |
454 | return pte_mfn_to_pfn(pud.pud); | |
455 | } | |
456 | PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val); | |
457 | ||
458 | __visible pud_t xen_make_pud(pudval_t pud) | |
459 | { | |
460 | pud = pte_pfn_to_mfn(pud); | |
461 | ||
462 | return native_make_pud(pud); | |
463 | } | |
464 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud); | |
465 | ||
466 | static pgd_t *xen_get_user_pgd(pgd_t *pgd) | |
467 | { | |
468 | pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK); | |
469 | unsigned offset = pgd - pgd_page; | |
470 | pgd_t *user_ptr = NULL; | |
471 | ||
472 | if (offset < pgd_index(USER_LIMIT)) { | |
473 | struct page *page = virt_to_page(pgd_page); | |
474 | user_ptr = (pgd_t *)page->private; | |
475 | if (user_ptr) | |
476 | user_ptr += offset; | |
477 | } | |
478 | ||
479 | return user_ptr; | |
480 | } | |
481 | ||
482 | static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val) | |
483 | { | |
484 | struct mmu_update u; | |
485 | ||
486 | u.ptr = virt_to_machine(ptr).maddr; | |
487 | u.val = p4d_val_ma(val); | |
488 | xen_extend_mmu_update(&u); | |
489 | } | |
490 | ||
491 | /* | |
492 | * Raw hypercall-based set_p4d, intended for in early boot before | |
493 | * there's a page structure. This implies: | |
494 | * 1. The only existing pagetable is the kernel's | |
495 | * 2. It is always pinned | |
496 | * 3. It has no user pagetable attached to it | |
497 | */ | |
498 | static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val) | |
499 | { | |
500 | preempt_disable(); | |
501 | ||
502 | xen_mc_batch(); | |
503 | ||
504 | __xen_set_p4d_hyper(ptr, val); | |
505 | ||
506 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
507 | ||
508 | preempt_enable(); | |
509 | } | |
510 | ||
511 | static void xen_set_p4d(p4d_t *ptr, p4d_t val) | |
512 | { | |
513 | pgd_t *user_ptr = xen_get_user_pgd((pgd_t *)ptr); | |
514 | pgd_t pgd_val; | |
515 | ||
516 | trace_xen_mmu_set_p4d(ptr, (p4d_t *)user_ptr, val); | |
517 | ||
518 | /* If page is not pinned, we can just update the entry | |
519 | directly */ | |
520 | if (!xen_page_pinned(ptr)) { | |
521 | *ptr = val; | |
522 | if (user_ptr) { | |
523 | WARN_ON(xen_page_pinned(user_ptr)); | |
524 | pgd_val.pgd = p4d_val_ma(val); | |
525 | *user_ptr = pgd_val; | |
526 | } | |
527 | return; | |
528 | } | |
529 | ||
530 | /* If it's pinned, then we can at least batch the kernel and | |
531 | user updates together. */ | |
532 | xen_mc_batch(); | |
533 | ||
534 | __xen_set_p4d_hyper(ptr, val); | |
535 | if (user_ptr) | |
536 | __xen_set_p4d_hyper((p4d_t *)user_ptr, val); | |
537 | ||
538 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
539 | } | |
af02cd97 | 540 | #endif /* CONFIG_X86_64 */ |
7e0563de VK |
541 | |
542 | static int xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd, | |
543 | int (*func)(struct mm_struct *mm, struct page *, enum pt_level), | |
544 | bool last, unsigned long limit) | |
545 | { | |
546 | int i, nr, flush = 0; | |
547 | ||
548 | nr = last ? pmd_index(limit) + 1 : PTRS_PER_PMD; | |
549 | for (i = 0; i < nr; i++) { | |
550 | if (!pmd_none(pmd[i])) | |
551 | flush |= (*func)(mm, pmd_page(pmd[i]), PT_PTE); | |
552 | } | |
553 | return flush; | |
554 | } | |
555 | ||
556 | static int xen_pud_walk(struct mm_struct *mm, pud_t *pud, | |
557 | int (*func)(struct mm_struct *mm, struct page *, enum pt_level), | |
558 | bool last, unsigned long limit) | |
559 | { | |
560 | int i, nr, flush = 0; | |
561 | ||
562 | nr = last ? pud_index(limit) + 1 : PTRS_PER_PUD; | |
563 | for (i = 0; i < nr; i++) { | |
564 | pmd_t *pmd; | |
565 | ||
566 | if (pud_none(pud[i])) | |
567 | continue; | |
568 | ||
569 | pmd = pmd_offset(&pud[i], 0); | |
570 | if (PTRS_PER_PMD > 1) | |
571 | flush |= (*func)(mm, virt_to_page(pmd), PT_PMD); | |
572 | flush |= xen_pmd_walk(mm, pmd, func, | |
573 | last && i == nr - 1, limit); | |
574 | } | |
575 | return flush; | |
576 | } | |
577 | ||
578 | static int xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d, | |
579 | int (*func)(struct mm_struct *mm, struct page *, enum pt_level), | |
580 | bool last, unsigned long limit) | |
581 | { | |
af02cd97 KS |
582 | int flush = 0; |
583 | pud_t *pud; | |
7e0563de | 584 | |
7e0563de | 585 | |
af02cd97 KS |
586 | if (p4d_none(*p4d)) |
587 | return flush; | |
7e0563de | 588 | |
af02cd97 KS |
589 | pud = pud_offset(p4d, 0); |
590 | if (PTRS_PER_PUD > 1) | |
591 | flush |= (*func)(mm, virt_to_page(pud), PT_PUD); | |
592 | flush |= xen_pud_walk(mm, pud, func, last, limit); | |
7e0563de VK |
593 | return flush; |
594 | } | |
595 | ||
596 | /* | |
597 | * (Yet another) pagetable walker. This one is intended for pinning a | |
598 | * pagetable. This means that it walks a pagetable and calls the | |
599 | * callback function on each page it finds making up the page table, | |
600 | * at every level. It walks the entire pagetable, but it only bothers | |
601 | * pinning pte pages which are below limit. In the normal case this | |
602 | * will be STACK_TOP_MAX, but at boot we need to pin up to | |
603 | * FIXADDR_TOP. | |
604 | * | |
605 | * For 32-bit the important bit is that we don't pin beyond there, | |
606 | * because then we start getting into Xen's ptes. | |
607 | * | |
608 | * For 64-bit, we must skip the Xen hole in the middle of the address | |
609 | * space, just after the big x86-64 virtual hole. | |
610 | */ | |
611 | static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd, | |
612 | int (*func)(struct mm_struct *mm, struct page *, | |
613 | enum pt_level), | |
614 | unsigned long limit) | |
615 | { | |
616 | int i, nr, flush = 0; | |
617 | unsigned hole_low, hole_high; | |
618 | ||
619 | /* The limit is the last byte to be touched */ | |
620 | limit--; | |
621 | BUG_ON(limit >= FIXADDR_TOP); | |
622 | ||
7e0563de VK |
623 | /* |
624 | * 64-bit has a great big hole in the middle of the address | |
625 | * space, which contains the Xen mappings. On 32-bit these | |
626 | * will end up making a zero-sized hole and so is a no-op. | |
627 | */ | |
628 | hole_low = pgd_index(USER_LIMIT); | |
629 | hole_high = pgd_index(PAGE_OFFSET); | |
630 | ||
631 | nr = pgd_index(limit) + 1; | |
632 | for (i = 0; i < nr; i++) { | |
633 | p4d_t *p4d; | |
634 | ||
635 | if (i >= hole_low && i < hole_high) | |
636 | continue; | |
637 | ||
638 | if (pgd_none(pgd[i])) | |
639 | continue; | |
640 | ||
641 | p4d = p4d_offset(&pgd[i], 0); | |
7e0563de VK |
642 | flush |= xen_p4d_walk(mm, p4d, func, i == nr - 1, limit); |
643 | } | |
644 | ||
645 | /* Do the top level last, so that the callbacks can use it as | |
646 | a cue to do final things like tlb flushes. */ | |
647 | flush |= (*func)(mm, virt_to_page(pgd), PT_PGD); | |
648 | ||
649 | return flush; | |
650 | } | |
651 | ||
652 | static int xen_pgd_walk(struct mm_struct *mm, | |
653 | int (*func)(struct mm_struct *mm, struct page *, | |
654 | enum pt_level), | |
655 | unsigned long limit) | |
656 | { | |
657 | return __xen_pgd_walk(mm, mm->pgd, func, limit); | |
658 | } | |
659 | ||
660 | /* If we're using split pte locks, then take the page's lock and | |
661 | return a pointer to it. Otherwise return NULL. */ | |
662 | static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm) | |
663 | { | |
664 | spinlock_t *ptl = NULL; | |
665 | ||
666 | #if USE_SPLIT_PTE_PTLOCKS | |
667 | ptl = ptlock_ptr(page); | |
668 | spin_lock_nest_lock(ptl, &mm->page_table_lock); | |
669 | #endif | |
670 | ||
671 | return ptl; | |
672 | } | |
673 | ||
674 | static void xen_pte_unlock(void *v) | |
675 | { | |
676 | spinlock_t *ptl = v; | |
677 | spin_unlock(ptl); | |
678 | } | |
679 | ||
680 | static void xen_do_pin(unsigned level, unsigned long pfn) | |
681 | { | |
682 | struct mmuext_op op; | |
683 | ||
684 | op.cmd = level; | |
685 | op.arg1.mfn = pfn_to_mfn(pfn); | |
686 | ||
687 | xen_extend_mmuext_op(&op); | |
688 | } | |
689 | ||
690 | static int xen_pin_page(struct mm_struct *mm, struct page *page, | |
691 | enum pt_level level) | |
692 | { | |
693 | unsigned pgfl = TestSetPagePinned(page); | |
694 | int flush; | |
695 | ||
696 | if (pgfl) | |
697 | flush = 0; /* already pinned */ | |
698 | else if (PageHighMem(page)) | |
699 | /* kmaps need flushing if we found an unpinned | |
700 | highpage */ | |
701 | flush = 1; | |
702 | else { | |
703 | void *pt = lowmem_page_address(page); | |
704 | unsigned long pfn = page_to_pfn(page); | |
705 | struct multicall_space mcs = __xen_mc_entry(0); | |
706 | spinlock_t *ptl; | |
707 | ||
708 | flush = 0; | |
709 | ||
710 | /* | |
711 | * We need to hold the pagetable lock between the time | |
712 | * we make the pagetable RO and when we actually pin | |
713 | * it. If we don't, then other users may come in and | |
714 | * attempt to update the pagetable by writing it, | |
715 | * which will fail because the memory is RO but not | |
716 | * pinned, so Xen won't do the trap'n'emulate. | |
717 | * | |
718 | * If we're using split pte locks, we can't hold the | |
719 | * entire pagetable's worth of locks during the | |
720 | * traverse, because we may wrap the preempt count (8 | |
721 | * bits). The solution is to mark RO and pin each PTE | |
722 | * page while holding the lock. This means the number | |
723 | * of locks we end up holding is never more than a | |
724 | * batch size (~32 entries, at present). | |
725 | * | |
726 | * If we're not using split pte locks, we needn't pin | |
727 | * the PTE pages independently, because we're | |
728 | * protected by the overall pagetable lock. | |
729 | */ | |
730 | ptl = NULL; | |
731 | if (level == PT_PTE) | |
732 | ptl = xen_pte_lock(page, mm); | |
733 | ||
734 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, | |
735 | pfn_pte(pfn, PAGE_KERNEL_RO), | |
736 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); | |
737 | ||
738 | if (ptl) { | |
739 | xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn); | |
740 | ||
741 | /* Queue a deferred unlock for when this batch | |
742 | is completed. */ | |
743 | xen_mc_callback(xen_pte_unlock, ptl); | |
744 | } | |
745 | } | |
746 | ||
747 | return flush; | |
748 | } | |
749 | ||
750 | /* This is called just after a mm has been created, but it has not | |
751 | been used yet. We need to make sure that its pagetable is all | |
752 | read-only, and can be pinned. */ | |
753 | static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd) | |
754 | { | |
755 | trace_xen_mmu_pgd_pin(mm, pgd); | |
756 | ||
757 | xen_mc_batch(); | |
758 | ||
759 | if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) { | |
760 | /* re-enable interrupts for flushing */ | |
761 | xen_mc_issue(0); | |
762 | ||
763 | kmap_flush_unused(); | |
764 | ||
765 | xen_mc_batch(); | |
766 | } | |
767 | ||
768 | #ifdef CONFIG_X86_64 | |
769 | { | |
770 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
771 | ||
772 | xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd))); | |
773 | ||
774 | if (user_pgd) { | |
775 | xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD); | |
776 | xen_do_pin(MMUEXT_PIN_L4_TABLE, | |
777 | PFN_DOWN(__pa(user_pgd))); | |
778 | } | |
779 | } | |
780 | #else /* CONFIG_X86_32 */ | |
781 | #ifdef CONFIG_X86_PAE | |
782 | /* Need to make sure unshared kernel PMD is pinnable */ | |
783 | xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), | |
784 | PT_PMD); | |
785 | #endif | |
786 | xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd))); | |
787 | #endif /* CONFIG_X86_64 */ | |
788 | xen_mc_issue(0); | |
789 | } | |
790 | ||
791 | static void xen_pgd_pin(struct mm_struct *mm) | |
792 | { | |
793 | __xen_pgd_pin(mm, mm->pgd); | |
794 | } | |
795 | ||
796 | /* | |
797 | * On save, we need to pin all pagetables to make sure they get their | |
798 | * mfns turned into pfns. Search the list for any unpinned pgds and pin | |
799 | * them (unpinned pgds are not currently in use, probably because the | |
800 | * process is under construction or destruction). | |
801 | * | |
802 | * Expected to be called in stop_machine() ("equivalent to taking | |
803 | * every spinlock in the system"), so the locking doesn't really | |
804 | * matter all that much. | |
805 | */ | |
806 | void xen_mm_pin_all(void) | |
807 | { | |
808 | struct page *page; | |
809 | ||
810 | spin_lock(&pgd_lock); | |
811 | ||
812 | list_for_each_entry(page, &pgd_list, lru) { | |
813 | if (!PagePinned(page)) { | |
814 | __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page)); | |
815 | SetPageSavePinned(page); | |
816 | } | |
817 | } | |
818 | ||
819 | spin_unlock(&pgd_lock); | |
820 | } | |
821 | ||
822 | /* | |
823 | * The init_mm pagetable is really pinned as soon as its created, but | |
824 | * that's before we have page structures to store the bits. So do all | |
825 | * the book-keeping now. | |
826 | */ | |
827 | static int __init xen_mark_pinned(struct mm_struct *mm, struct page *page, | |
828 | enum pt_level level) | |
829 | { | |
830 | SetPagePinned(page); | |
831 | return 0; | |
832 | } | |
833 | ||
834 | static void __init xen_mark_init_mm_pinned(void) | |
835 | { | |
836 | xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP); | |
837 | } | |
838 | ||
839 | static int xen_unpin_page(struct mm_struct *mm, struct page *page, | |
840 | enum pt_level level) | |
841 | { | |
842 | unsigned pgfl = TestClearPagePinned(page); | |
843 | ||
844 | if (pgfl && !PageHighMem(page)) { | |
845 | void *pt = lowmem_page_address(page); | |
846 | unsigned long pfn = page_to_pfn(page); | |
847 | spinlock_t *ptl = NULL; | |
848 | struct multicall_space mcs; | |
849 | ||
850 | /* | |
851 | * Do the converse to pin_page. If we're using split | |
852 | * pte locks, we must be holding the lock for while | |
853 | * the pte page is unpinned but still RO to prevent | |
854 | * concurrent updates from seeing it in this | |
855 | * partially-pinned state. | |
856 | */ | |
857 | if (level == PT_PTE) { | |
858 | ptl = xen_pte_lock(page, mm); | |
859 | ||
860 | if (ptl) | |
861 | xen_do_pin(MMUEXT_UNPIN_TABLE, pfn); | |
862 | } | |
863 | ||
864 | mcs = __xen_mc_entry(0); | |
865 | ||
866 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, | |
867 | pfn_pte(pfn, PAGE_KERNEL), | |
868 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); | |
869 | ||
870 | if (ptl) { | |
871 | /* unlock when batch completed */ | |
872 | xen_mc_callback(xen_pte_unlock, ptl); | |
873 | } | |
874 | } | |
875 | ||
876 | return 0; /* never need to flush on unpin */ | |
877 | } | |
878 | ||
879 | /* Release a pagetables pages back as normal RW */ | |
880 | static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd) | |
881 | { | |
882 | trace_xen_mmu_pgd_unpin(mm, pgd); | |
883 | ||
884 | xen_mc_batch(); | |
885 | ||
886 | xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
887 | ||
888 | #ifdef CONFIG_X86_64 | |
889 | { | |
890 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
891 | ||
892 | if (user_pgd) { | |
893 | xen_do_pin(MMUEXT_UNPIN_TABLE, | |
894 | PFN_DOWN(__pa(user_pgd))); | |
895 | xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD); | |
896 | } | |
897 | } | |
898 | #endif | |
899 | ||
900 | #ifdef CONFIG_X86_PAE | |
901 | /* Need to make sure unshared kernel PMD is unpinned */ | |
902 | xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), | |
903 | PT_PMD); | |
904 | #endif | |
905 | ||
906 | __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT); | |
907 | ||
908 | xen_mc_issue(0); | |
909 | } | |
910 | ||
911 | static void xen_pgd_unpin(struct mm_struct *mm) | |
912 | { | |
913 | __xen_pgd_unpin(mm, mm->pgd); | |
914 | } | |
915 | ||
916 | /* | |
917 | * On resume, undo any pinning done at save, so that the rest of the | |
918 | * kernel doesn't see any unexpected pinned pagetables. | |
919 | */ | |
920 | void xen_mm_unpin_all(void) | |
921 | { | |
922 | struct page *page; | |
923 | ||
924 | spin_lock(&pgd_lock); | |
925 | ||
926 | list_for_each_entry(page, &pgd_list, lru) { | |
927 | if (PageSavePinned(page)) { | |
928 | BUG_ON(!PagePinned(page)); | |
929 | __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page)); | |
930 | ClearPageSavePinned(page); | |
931 | } | |
932 | } | |
933 | ||
934 | spin_unlock(&pgd_lock); | |
935 | } | |
936 | ||
937 | static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next) | |
938 | { | |
939 | spin_lock(&next->page_table_lock); | |
940 | xen_pgd_pin(next); | |
941 | spin_unlock(&next->page_table_lock); | |
942 | } | |
943 | ||
944 | static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm) | |
945 | { | |
946 | spin_lock(&mm->page_table_lock); | |
947 | xen_pgd_pin(mm); | |
948 | spin_unlock(&mm->page_table_lock); | |
949 | } | |
950 | ||
3d28ebce | 951 | static void drop_mm_ref_this_cpu(void *info) |
7e0563de VK |
952 | { |
953 | struct mm_struct *mm = info; | |
7e0563de | 954 | |
3d28ebce | 955 | if (this_cpu_read(cpu_tlbstate.loaded_mm) == mm) |
7e0563de VK |
956 | leave_mm(smp_processor_id()); |
957 | ||
3d28ebce AL |
958 | /* |
959 | * If this cpu still has a stale cr3 reference, then make sure | |
960 | * it has been flushed. | |
961 | */ | |
7e0563de | 962 | if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd)) |
3d28ebce | 963 | xen_mc_flush(); |
7e0563de VK |
964 | } |
965 | ||
3d28ebce AL |
966 | #ifdef CONFIG_SMP |
967 | /* | |
968 | * Another cpu may still have their %cr3 pointing at the pagetable, so | |
969 | * we need to repoint it somewhere else before we can unpin it. | |
970 | */ | |
7e0563de VK |
971 | static void xen_drop_mm_ref(struct mm_struct *mm) |
972 | { | |
973 | cpumask_var_t mask; | |
974 | unsigned cpu; | |
975 | ||
3d28ebce | 976 | drop_mm_ref_this_cpu(mm); |
7e0563de VK |
977 | |
978 | /* Get the "official" set of cpus referring to our pagetable. */ | |
979 | if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) { | |
980 | for_each_online_cpu(cpu) { | |
94b1b03b | 981 | if (per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd)) |
7e0563de | 982 | continue; |
3d28ebce | 983 | smp_call_function_single(cpu, drop_mm_ref_this_cpu, mm, 1); |
7e0563de VK |
984 | } |
985 | return; | |
986 | } | |
7e0563de | 987 | |
3d28ebce AL |
988 | /* |
989 | * It's possible that a vcpu may have a stale reference to our | |
990 | * cr3, because its in lazy mode, and it hasn't yet flushed | |
991 | * its set of pending hypercalls yet. In this case, we can | |
992 | * look at its actual current cr3 value, and force it to flush | |
993 | * if needed. | |
994 | */ | |
94b1b03b | 995 | cpumask_clear(mask); |
7e0563de VK |
996 | for_each_online_cpu(cpu) { |
997 | if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd)) | |
998 | cpumask_set_cpu(cpu, mask); | |
999 | } | |
1000 | ||
3d28ebce | 1001 | smp_call_function_many(mask, drop_mm_ref_this_cpu, mm, 1); |
7e0563de VK |
1002 | free_cpumask_var(mask); |
1003 | } | |
1004 | #else | |
1005 | static void xen_drop_mm_ref(struct mm_struct *mm) | |
1006 | { | |
3d28ebce | 1007 | drop_mm_ref_this_cpu(mm); |
7e0563de VK |
1008 | } |
1009 | #endif | |
1010 | ||
1011 | /* | |
1012 | * While a process runs, Xen pins its pagetables, which means that the | |
1013 | * hypervisor forces it to be read-only, and it controls all updates | |
1014 | * to it. This means that all pagetable updates have to go via the | |
1015 | * hypervisor, which is moderately expensive. | |
1016 | * | |
1017 | * Since we're pulling the pagetable down, we switch to use init_mm, | |
1018 | * unpin old process pagetable and mark it all read-write, which | |
1019 | * allows further operations on it to be simple memory accesses. | |
1020 | * | |
1021 | * The only subtle point is that another CPU may be still using the | |
1022 | * pagetable because of lazy tlb flushing. This means we need need to | |
1023 | * switch all CPUs off this pagetable before we can unpin it. | |
1024 | */ | |
1025 | static void xen_exit_mmap(struct mm_struct *mm) | |
1026 | { | |
1027 | get_cpu(); /* make sure we don't move around */ | |
1028 | xen_drop_mm_ref(mm); | |
1029 | put_cpu(); | |
1030 | ||
1031 | spin_lock(&mm->page_table_lock); | |
1032 | ||
1033 | /* pgd may not be pinned in the error exit path of execve */ | |
1034 | if (xen_page_pinned(mm->pgd)) | |
1035 | xen_pgd_unpin(mm); | |
1036 | ||
1037 | spin_unlock(&mm->page_table_lock); | |
1038 | } | |
1039 | ||
1040 | static void xen_post_allocator_init(void); | |
1041 | ||
1042 | static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn) | |
1043 | { | |
1044 | struct mmuext_op op; | |
1045 | ||
1046 | op.cmd = cmd; | |
1047 | op.arg1.mfn = pfn_to_mfn(pfn); | |
1048 | if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) | |
1049 | BUG(); | |
1050 | } | |
1051 | ||
1052 | #ifdef CONFIG_X86_64 | |
1053 | static void __init xen_cleanhighmap(unsigned long vaddr, | |
1054 | unsigned long vaddr_end) | |
1055 | { | |
1056 | unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1; | |
1057 | pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr); | |
1058 | ||
1059 | /* NOTE: The loop is more greedy than the cleanup_highmap variant. | |
1060 | * We include the PMD passed in on _both_ boundaries. */ | |
1061 | for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD)); | |
1062 | pmd++, vaddr += PMD_SIZE) { | |
1063 | if (pmd_none(*pmd)) | |
1064 | continue; | |
1065 | if (vaddr < (unsigned long) _text || vaddr > kernel_end) | |
1066 | set_pmd(pmd, __pmd(0)); | |
1067 | } | |
1068 | /* In case we did something silly, we should crash in this function | |
1069 | * instead of somewhere later and be confusing. */ | |
1070 | xen_mc_flush(); | |
1071 | } | |
1072 | ||
1073 | /* | |
1074 | * Make a page range writeable and free it. | |
1075 | */ | |
1076 | static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size) | |
1077 | { | |
1078 | void *vaddr = __va(paddr); | |
1079 | void *vaddr_end = vaddr + size; | |
1080 | ||
1081 | for (; vaddr < vaddr_end; vaddr += PAGE_SIZE) | |
1082 | make_lowmem_page_readwrite(vaddr); | |
1083 | ||
1084 | memblock_free(paddr, size); | |
1085 | } | |
1086 | ||
1087 | static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin) | |
1088 | { | |
1089 | unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK; | |
1090 | ||
1091 | if (unpin) | |
1092 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa)); | |
1093 | ClearPagePinned(virt_to_page(__va(pa))); | |
1094 | xen_free_ro_pages(pa, PAGE_SIZE); | |
1095 | } | |
1096 | ||
1097 | static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin) | |
1098 | { | |
1099 | unsigned long pa; | |
1100 | pte_t *pte_tbl; | |
1101 | int i; | |
1102 | ||
1103 | if (pmd_large(*pmd)) { | |
1104 | pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK; | |
1105 | xen_free_ro_pages(pa, PMD_SIZE); | |
1106 | return; | |
1107 | } | |
1108 | ||
1109 | pte_tbl = pte_offset_kernel(pmd, 0); | |
1110 | for (i = 0; i < PTRS_PER_PTE; i++) { | |
1111 | if (pte_none(pte_tbl[i])) | |
1112 | continue; | |
1113 | pa = pte_pfn(pte_tbl[i]) << PAGE_SHIFT; | |
1114 | xen_free_ro_pages(pa, PAGE_SIZE); | |
1115 | } | |
1116 | set_pmd(pmd, __pmd(0)); | |
1117 | xen_cleanmfnmap_free_pgtbl(pte_tbl, unpin); | |
1118 | } | |
1119 | ||
1120 | static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin) | |
1121 | { | |
1122 | unsigned long pa; | |
1123 | pmd_t *pmd_tbl; | |
1124 | int i; | |
1125 | ||
1126 | if (pud_large(*pud)) { | |
1127 | pa = pud_val(*pud) & PHYSICAL_PAGE_MASK; | |
1128 | xen_free_ro_pages(pa, PUD_SIZE); | |
1129 | return; | |
1130 | } | |
1131 | ||
1132 | pmd_tbl = pmd_offset(pud, 0); | |
1133 | for (i = 0; i < PTRS_PER_PMD; i++) { | |
1134 | if (pmd_none(pmd_tbl[i])) | |
1135 | continue; | |
1136 | xen_cleanmfnmap_pmd(pmd_tbl + i, unpin); | |
1137 | } | |
1138 | set_pud(pud, __pud(0)); | |
1139 | xen_cleanmfnmap_free_pgtbl(pmd_tbl, unpin); | |
1140 | } | |
1141 | ||
1142 | static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin) | |
1143 | { | |
1144 | unsigned long pa; | |
1145 | pud_t *pud_tbl; | |
1146 | int i; | |
1147 | ||
1148 | if (p4d_large(*p4d)) { | |
1149 | pa = p4d_val(*p4d) & PHYSICAL_PAGE_MASK; | |
1150 | xen_free_ro_pages(pa, P4D_SIZE); | |
1151 | return; | |
1152 | } | |
1153 | ||
1154 | pud_tbl = pud_offset(p4d, 0); | |
1155 | for (i = 0; i < PTRS_PER_PUD; i++) { | |
1156 | if (pud_none(pud_tbl[i])) | |
1157 | continue; | |
1158 | xen_cleanmfnmap_pud(pud_tbl + i, unpin); | |
1159 | } | |
1160 | set_p4d(p4d, __p4d(0)); | |
1161 | xen_cleanmfnmap_free_pgtbl(pud_tbl, unpin); | |
1162 | } | |
1163 | ||
1164 | /* | |
1165 | * Since it is well isolated we can (and since it is perhaps large we should) | |
1166 | * also free the page tables mapping the initial P->M table. | |
1167 | */ | |
1168 | static void __init xen_cleanmfnmap(unsigned long vaddr) | |
1169 | { | |
1170 | pgd_t *pgd; | |
1171 | p4d_t *p4d; | |
7e0563de VK |
1172 | bool unpin; |
1173 | ||
1174 | unpin = (vaddr == 2 * PGDIR_SIZE); | |
1175 | vaddr &= PMD_MASK; | |
1176 | pgd = pgd_offset_k(vaddr); | |
1177 | p4d = p4d_offset(pgd, 0); | |
af02cd97 KS |
1178 | if (!p4d_none(*p4d)) |
1179 | xen_cleanmfnmap_p4d(p4d, unpin); | |
7e0563de VK |
1180 | } |
1181 | ||
1182 | static void __init xen_pagetable_p2m_free(void) | |
1183 | { | |
1184 | unsigned long size; | |
1185 | unsigned long addr; | |
1186 | ||
1187 | size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long)); | |
1188 | ||
1189 | /* No memory or already called. */ | |
1190 | if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list) | |
1191 | return; | |
1192 | ||
1193 | /* using __ka address and sticking INVALID_P2M_ENTRY! */ | |
1194 | memset((void *)xen_start_info->mfn_list, 0xff, size); | |
1195 | ||
1196 | addr = xen_start_info->mfn_list; | |
1197 | /* | |
1198 | * We could be in __ka space. | |
1199 | * We roundup to the PMD, which means that if anybody at this stage is | |
1200 | * using the __ka address of xen_start_info or | |
1201 | * xen_start_info->shared_info they are in going to crash. Fortunatly | |
1202 | * we have already revectored in xen_setup_kernel_pagetable and in | |
1203 | * xen_setup_shared_info. | |
1204 | */ | |
1205 | size = roundup(size, PMD_SIZE); | |
1206 | ||
1207 | if (addr >= __START_KERNEL_map) { | |
1208 | xen_cleanhighmap(addr, addr + size); | |
1209 | size = PAGE_ALIGN(xen_start_info->nr_pages * | |
1210 | sizeof(unsigned long)); | |
1211 | memblock_free(__pa(addr), size); | |
1212 | } else { | |
1213 | xen_cleanmfnmap(addr); | |
1214 | } | |
1215 | } | |
1216 | ||
1217 | static void __init xen_pagetable_cleanhighmap(void) | |
1218 | { | |
1219 | unsigned long size; | |
1220 | unsigned long addr; | |
1221 | ||
1222 | /* At this stage, cleanup_highmap has already cleaned __ka space | |
1223 | * from _brk_limit way up to the max_pfn_mapped (which is the end of | |
1224 | * the ramdisk). We continue on, erasing PMD entries that point to page | |
1225 | * tables - do note that they are accessible at this stage via __va. | |
0d805ee7 ZD |
1226 | * As Xen is aligning the memory end to a 4MB boundary, for good |
1227 | * measure we also round up to PMD_SIZE * 2 - which means that if | |
7e0563de VK |
1228 | * anybody is using __ka address to the initial boot-stack - and try |
1229 | * to use it - they are going to crash. The xen_start_info has been | |
1230 | * taken care of already in xen_setup_kernel_pagetable. */ | |
1231 | addr = xen_start_info->pt_base; | |
0d805ee7 | 1232 | size = xen_start_info->nr_pt_frames * PAGE_SIZE; |
7e0563de | 1233 | |
0d805ee7 | 1234 | xen_cleanhighmap(addr, roundup(addr + size, PMD_SIZE * 2)); |
7e0563de | 1235 | xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base)); |
7e0563de VK |
1236 | } |
1237 | #endif | |
1238 | ||
1239 | static void __init xen_pagetable_p2m_setup(void) | |
1240 | { | |
7e0563de VK |
1241 | xen_vmalloc_p2m_tree(); |
1242 | ||
1243 | #ifdef CONFIG_X86_64 | |
1244 | xen_pagetable_p2m_free(); | |
1245 | ||
1246 | xen_pagetable_cleanhighmap(); | |
1247 | #endif | |
1248 | /* And revector! Bye bye old array */ | |
1249 | xen_start_info->mfn_list = (unsigned long)xen_p2m_addr; | |
1250 | } | |
1251 | ||
1252 | static void __init xen_pagetable_init(void) | |
1253 | { | |
1254 | paging_init(); | |
1255 | xen_post_allocator_init(); | |
1256 | ||
1257 | xen_pagetable_p2m_setup(); | |
1258 | ||
1259 | /* Allocate and initialize top and mid mfn levels for p2m structure */ | |
1260 | xen_build_mfn_list_list(); | |
1261 | ||
1262 | /* Remap memory freed due to conflicts with E820 map */ | |
989513a7 | 1263 | xen_remap_memory(); |
7e0563de VK |
1264 | |
1265 | xen_setup_shared_info(); | |
1266 | } | |
1267 | static void xen_write_cr2(unsigned long cr2) | |
1268 | { | |
1269 | this_cpu_read(xen_vcpu)->arch.cr2 = cr2; | |
1270 | } | |
1271 | ||
1272 | static unsigned long xen_read_cr2(void) | |
1273 | { | |
1274 | return this_cpu_read(xen_vcpu)->arch.cr2; | |
1275 | } | |
1276 | ||
1277 | unsigned long xen_read_cr2_direct(void) | |
1278 | { | |
1279 | return this_cpu_read(xen_vcpu_info.arch.cr2); | |
1280 | } | |
1281 | ||
9a19a93b | 1282 | static noinline void xen_flush_tlb(void) |
7e0563de VK |
1283 | { |
1284 | struct mmuext_op *op; | |
1285 | struct multicall_space mcs; | |
1286 | ||
7e0563de VK |
1287 | preempt_disable(); |
1288 | ||
1289 | mcs = xen_mc_entry(sizeof(*op)); | |
1290 | ||
1291 | op = mcs.args; | |
1292 | op->cmd = MMUEXT_TLB_FLUSH_LOCAL; | |
1293 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1294 | ||
1295 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1296 | ||
1297 | preempt_enable(); | |
1298 | } | |
1299 | ||
208beef6 | 1300 | static void xen_flush_tlb_one_user(unsigned long addr) |
7e0563de VK |
1301 | { |
1302 | struct mmuext_op *op; | |
1303 | struct multicall_space mcs; | |
1304 | ||
208beef6 | 1305 | trace_xen_mmu_flush_tlb_one_user(addr); |
7e0563de VK |
1306 | |
1307 | preempt_disable(); | |
1308 | ||
1309 | mcs = xen_mc_entry(sizeof(*op)); | |
1310 | op = mcs.args; | |
1311 | op->cmd = MMUEXT_INVLPG_LOCAL; | |
1312 | op->arg1.linear_addr = addr & PAGE_MASK; | |
1313 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1314 | ||
1315 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1316 | ||
1317 | preempt_enable(); | |
1318 | } | |
1319 | ||
1320 | static void xen_flush_tlb_others(const struct cpumask *cpus, | |
a2055abe | 1321 | const struct flush_tlb_info *info) |
7e0563de VK |
1322 | { |
1323 | struct { | |
1324 | struct mmuext_op op; | |
1325 | #ifdef CONFIG_SMP | |
1326 | DECLARE_BITMAP(mask, num_processors); | |
1327 | #else | |
1328 | DECLARE_BITMAP(mask, NR_CPUS); | |
1329 | #endif | |
1330 | } *args; | |
1331 | struct multicall_space mcs; | |
1332 | ||
a2055abe | 1333 | trace_xen_mmu_flush_tlb_others(cpus, info->mm, info->start, info->end); |
7e0563de VK |
1334 | |
1335 | if (cpumask_empty(cpus)) | |
1336 | return; /* nothing to do */ | |
1337 | ||
1338 | mcs = xen_mc_entry(sizeof(*args)); | |
1339 | args = mcs.args; | |
1340 | args->op.arg2.vcpumask = to_cpumask(args->mask); | |
1341 | ||
1342 | /* Remove us, and any offline CPUS. */ | |
1343 | cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask); | |
1344 | cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask)); | |
1345 | ||
1346 | args->op.cmd = MMUEXT_TLB_FLUSH_MULTI; | |
a2055abe AL |
1347 | if (info->end != TLB_FLUSH_ALL && |
1348 | (info->end - info->start) <= PAGE_SIZE) { | |
7e0563de | 1349 | args->op.cmd = MMUEXT_INVLPG_MULTI; |
a2055abe | 1350 | args->op.arg1.linear_addr = info->start; |
7e0563de VK |
1351 | } |
1352 | ||
1353 | MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF); | |
1354 | ||
1355 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1356 | } | |
1357 | ||
1358 | static unsigned long xen_read_cr3(void) | |
1359 | { | |
1360 | return this_cpu_read(xen_cr3); | |
1361 | } | |
1362 | ||
1363 | static void set_current_cr3(void *v) | |
1364 | { | |
1365 | this_cpu_write(xen_current_cr3, (unsigned long)v); | |
1366 | } | |
1367 | ||
1368 | static void __xen_write_cr3(bool kernel, unsigned long cr3) | |
1369 | { | |
1370 | struct mmuext_op op; | |
1371 | unsigned long mfn; | |
1372 | ||
1373 | trace_xen_mmu_write_cr3(kernel, cr3); | |
1374 | ||
1375 | if (cr3) | |
1376 | mfn = pfn_to_mfn(PFN_DOWN(cr3)); | |
1377 | else | |
1378 | mfn = 0; | |
1379 | ||
1380 | WARN_ON(mfn == 0 && kernel); | |
1381 | ||
1382 | op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR; | |
1383 | op.arg1.mfn = mfn; | |
1384 | ||
1385 | xen_extend_mmuext_op(&op); | |
1386 | ||
1387 | if (kernel) { | |
1388 | this_cpu_write(xen_cr3, cr3); | |
1389 | ||
1390 | /* Update xen_current_cr3 once the batch has actually | |
1391 | been submitted. */ | |
1392 | xen_mc_callback(set_current_cr3, (void *)cr3); | |
1393 | } | |
1394 | } | |
1395 | static void xen_write_cr3(unsigned long cr3) | |
1396 | { | |
1397 | BUG_ON(preemptible()); | |
1398 | ||
1399 | xen_mc_batch(); /* disables interrupts */ | |
1400 | ||
1401 | /* Update while interrupts are disabled, so its atomic with | |
1402 | respect to ipis */ | |
1403 | this_cpu_write(xen_cr3, cr3); | |
1404 | ||
1405 | __xen_write_cr3(true, cr3); | |
1406 | ||
1407 | #ifdef CONFIG_X86_64 | |
1408 | { | |
1409 | pgd_t *user_pgd = xen_get_user_pgd(__va(cr3)); | |
1410 | if (user_pgd) | |
1411 | __xen_write_cr3(false, __pa(user_pgd)); | |
1412 | else | |
1413 | __xen_write_cr3(false, 0); | |
1414 | } | |
1415 | #endif | |
1416 | ||
1417 | xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ | |
1418 | } | |
1419 | ||
1420 | #ifdef CONFIG_X86_64 | |
1421 | /* | |
1422 | * At the start of the day - when Xen launches a guest, it has already | |
1423 | * built pagetables for the guest. We diligently look over them | |
1424 | * in xen_setup_kernel_pagetable and graft as appropriate them in the | |
65ade2f8 KS |
1425 | * init_top_pgt and its friends. Then when we are happy we load |
1426 | * the new init_top_pgt - and continue on. | |
7e0563de VK |
1427 | * |
1428 | * The generic code starts (start_kernel) and 'init_mem_mapping' sets | |
1429 | * up the rest of the pagetables. When it has completed it loads the cr3. | |
1430 | * N.B. that baremetal would start at 'start_kernel' (and the early | |
1431 | * #PF handler would create bootstrap pagetables) - so we are running | |
1432 | * with the same assumptions as what to do when write_cr3 is executed | |
1433 | * at this point. | |
1434 | * | |
1435 | * Since there are no user-page tables at all, we have two variants | |
1436 | * of xen_write_cr3 - the early bootup (this one), and the late one | |
1437 | * (xen_write_cr3). The reason we have to do that is that in 64-bit | |
1438 | * the Linux kernel and user-space are both in ring 3 while the | |
1439 | * hypervisor is in ring 0. | |
1440 | */ | |
1441 | static void __init xen_write_cr3_init(unsigned long cr3) | |
1442 | { | |
1443 | BUG_ON(preemptible()); | |
1444 | ||
1445 | xen_mc_batch(); /* disables interrupts */ | |
1446 | ||
1447 | /* Update while interrupts are disabled, so its atomic with | |
1448 | respect to ipis */ | |
1449 | this_cpu_write(xen_cr3, cr3); | |
1450 | ||
1451 | __xen_write_cr3(true, cr3); | |
1452 | ||
1453 | xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ | |
1454 | } | |
1455 | #endif | |
1456 | ||
1457 | static int xen_pgd_alloc(struct mm_struct *mm) | |
1458 | { | |
1459 | pgd_t *pgd = mm->pgd; | |
1460 | int ret = 0; | |
1461 | ||
1462 | BUG_ON(PagePinned(virt_to_page(pgd))); | |
1463 | ||
1464 | #ifdef CONFIG_X86_64 | |
1465 | { | |
1466 | struct page *page = virt_to_page(pgd); | |
1467 | pgd_t *user_pgd; | |
1468 | ||
1469 | BUG_ON(page->private != 0); | |
1470 | ||
1471 | ret = -ENOMEM; | |
1472 | ||
1473 | user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO); | |
1474 | page->private = (unsigned long)user_pgd; | |
1475 | ||
1476 | if (user_pgd != NULL) { | |
1477 | #ifdef CONFIG_X86_VSYSCALL_EMULATION | |
1478 | user_pgd[pgd_index(VSYSCALL_ADDR)] = | |
1479 | __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE); | |
1480 | #endif | |
1481 | ret = 0; | |
1482 | } | |
1483 | ||
1484 | BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd)))); | |
1485 | } | |
1486 | #endif | |
1487 | return ret; | |
1488 | } | |
1489 | ||
1490 | static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd) | |
1491 | { | |
1492 | #ifdef CONFIG_X86_64 | |
1493 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
1494 | ||
1495 | if (user_pgd) | |
1496 | free_page((unsigned long)user_pgd); | |
1497 | #endif | |
1498 | } | |
1499 | ||
1500 | /* | |
1501 | * Init-time set_pte while constructing initial pagetables, which | |
1502 | * doesn't allow RO page table pages to be remapped RW. | |
1503 | * | |
1504 | * If there is no MFN for this PFN then this page is initially | |
1505 | * ballooned out so clear the PTE (as in decrease_reservation() in | |
1506 | * drivers/xen/balloon.c). | |
1507 | * | |
1508 | * Many of these PTE updates are done on unpinned and writable pages | |
1509 | * and doing a hypercall for these is unnecessary and expensive. At | |
1510 | * this point it is not possible to tell if a page is pinned or not, | |
1511 | * so always write the PTE directly and rely on Xen trapping and | |
1512 | * emulating any updates as necessary. | |
1513 | */ | |
1514 | __visible pte_t xen_make_pte_init(pteval_t pte) | |
1515 | { | |
1516 | #ifdef CONFIG_X86_64 | |
1517 | unsigned long pfn; | |
1518 | ||
1519 | /* | |
1520 | * Pages belonging to the initial p2m list mapped outside the default | |
1521 | * address range must be mapped read-only. This region contains the | |
1522 | * page tables for mapping the p2m list, too, and page tables MUST be | |
1523 | * mapped read-only. | |
1524 | */ | |
1525 | pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT; | |
1526 | if (xen_start_info->mfn_list < __START_KERNEL_map && | |
1527 | pfn >= xen_start_info->first_p2m_pfn && | |
1528 | pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames) | |
1529 | pte &= ~_PAGE_RW; | |
1530 | #endif | |
1531 | pte = pte_pfn_to_mfn(pte); | |
1532 | return native_make_pte(pte); | |
1533 | } | |
1534 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init); | |
1535 | ||
1536 | static void __init xen_set_pte_init(pte_t *ptep, pte_t pte) | |
1537 | { | |
1538 | #ifdef CONFIG_X86_32 | |
1539 | /* If there's an existing pte, then don't allow _PAGE_RW to be set */ | |
1540 | if (pte_mfn(pte) != INVALID_P2M_ENTRY | |
1541 | && pte_val_ma(*ptep) & _PAGE_PRESENT) | |
1542 | pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) & | |
1543 | pte_val_ma(pte)); | |
1544 | #endif | |
13b23ccf | 1545 | __xen_set_pte(ptep, pte); |
7e0563de VK |
1546 | } |
1547 | ||
1548 | /* Early in boot, while setting up the initial pagetable, assume | |
1549 | everything is pinned. */ | |
1550 | static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn) | |
1551 | { | |
1552 | #ifdef CONFIG_FLATMEM | |
1553 | BUG_ON(mem_map); /* should only be used early */ | |
1554 | #endif | |
1555 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1556 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
1557 | } | |
1558 | ||
1559 | /* Used for pmd and pud */ | |
1560 | static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn) | |
1561 | { | |
1562 | #ifdef CONFIG_FLATMEM | |
1563 | BUG_ON(mem_map); /* should only be used early */ | |
1564 | #endif | |
1565 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1566 | } | |
1567 | ||
1568 | /* Early release_pte assumes that all pts are pinned, since there's | |
1569 | only init_mm and anything attached to that is pinned. */ | |
1570 | static void __init xen_release_pte_init(unsigned long pfn) | |
1571 | { | |
1572 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); | |
1573 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
1574 | } | |
1575 | ||
1576 | static void __init xen_release_pmd_init(unsigned long pfn) | |
1577 | { | |
1578 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
1579 | } | |
1580 | ||
1581 | static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn) | |
1582 | { | |
1583 | struct multicall_space mcs; | |
1584 | struct mmuext_op *op; | |
1585 | ||
1586 | mcs = __xen_mc_entry(sizeof(*op)); | |
1587 | op = mcs.args; | |
1588 | op->cmd = cmd; | |
1589 | op->arg1.mfn = pfn_to_mfn(pfn); | |
1590 | ||
1591 | MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); | |
1592 | } | |
1593 | ||
1594 | static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot) | |
1595 | { | |
1596 | struct multicall_space mcs; | |
1597 | unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT); | |
1598 | ||
1599 | mcs = __xen_mc_entry(0); | |
1600 | MULTI_update_va_mapping(mcs.mc, (unsigned long)addr, | |
1601 | pfn_pte(pfn, prot), 0); | |
1602 | } | |
1603 | ||
1604 | /* This needs to make sure the new pte page is pinned iff its being | |
1605 | attached to a pinned pagetable. */ | |
1606 | static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, | |
1607 | unsigned level) | |
1608 | { | |
1609 | bool pinned = PagePinned(virt_to_page(mm->pgd)); | |
1610 | ||
1611 | trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned); | |
1612 | ||
1613 | if (pinned) { | |
1614 | struct page *page = pfn_to_page(pfn); | |
1615 | ||
1616 | SetPagePinned(page); | |
1617 | ||
1618 | if (!PageHighMem(page)) { | |
1619 | xen_mc_batch(); | |
1620 | ||
1621 | __set_pfn_prot(pfn, PAGE_KERNEL_RO); | |
1622 | ||
1623 | if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS) | |
1624 | __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
1625 | ||
1626 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1627 | } else { | |
1628 | /* make sure there are no stray mappings of | |
1629 | this page */ | |
1630 | kmap_flush_unused(); | |
1631 | } | |
1632 | } | |
1633 | } | |
1634 | ||
1635 | static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn) | |
1636 | { | |
1637 | xen_alloc_ptpage(mm, pfn, PT_PTE); | |
1638 | } | |
1639 | ||
1640 | static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn) | |
1641 | { | |
1642 | xen_alloc_ptpage(mm, pfn, PT_PMD); | |
1643 | } | |
1644 | ||
1645 | /* This should never happen until we're OK to use struct page */ | |
1646 | static inline void xen_release_ptpage(unsigned long pfn, unsigned level) | |
1647 | { | |
1648 | struct page *page = pfn_to_page(pfn); | |
1649 | bool pinned = PagePinned(page); | |
1650 | ||
1651 | trace_xen_mmu_release_ptpage(pfn, level, pinned); | |
1652 | ||
1653 | if (pinned) { | |
1654 | if (!PageHighMem(page)) { | |
1655 | xen_mc_batch(); | |
1656 | ||
1657 | if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS) | |
1658 | __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); | |
1659 | ||
1660 | __set_pfn_prot(pfn, PAGE_KERNEL); | |
1661 | ||
1662 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1663 | } | |
1664 | ClearPagePinned(page); | |
1665 | } | |
1666 | } | |
1667 | ||
1668 | static void xen_release_pte(unsigned long pfn) | |
1669 | { | |
1670 | xen_release_ptpage(pfn, PT_PTE); | |
1671 | } | |
1672 | ||
1673 | static void xen_release_pmd(unsigned long pfn) | |
1674 | { | |
1675 | xen_release_ptpage(pfn, PT_PMD); | |
1676 | } | |
1677 | ||
af02cd97 | 1678 | #ifdef CONFIG_X86_64 |
7e0563de VK |
1679 | static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn) |
1680 | { | |
1681 | xen_alloc_ptpage(mm, pfn, PT_PUD); | |
1682 | } | |
1683 | ||
1684 | static void xen_release_pud(unsigned long pfn) | |
1685 | { | |
1686 | xen_release_ptpage(pfn, PT_PUD); | |
1687 | } | |
1688 | #endif | |
1689 | ||
1690 | void __init xen_reserve_top(void) | |
1691 | { | |
1692 | #ifdef CONFIG_X86_32 | |
1693 | unsigned long top = HYPERVISOR_VIRT_START; | |
1694 | struct xen_platform_parameters pp; | |
1695 | ||
1696 | if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0) | |
1697 | top = pp.virt_start; | |
1698 | ||
1699 | reserve_top_address(-top); | |
1700 | #endif /* CONFIG_X86_32 */ | |
1701 | } | |
1702 | ||
1703 | /* | |
1704 | * Like __va(), but returns address in the kernel mapping (which is | |
1705 | * all we have until the physical memory mapping has been set up. | |
1706 | */ | |
1707 | static void * __init __ka(phys_addr_t paddr) | |
1708 | { | |
1709 | #ifdef CONFIG_X86_64 | |
1710 | return (void *)(paddr + __START_KERNEL_map); | |
1711 | #else | |
1712 | return __va(paddr); | |
1713 | #endif | |
1714 | } | |
1715 | ||
1716 | /* Convert a machine address to physical address */ | |
1717 | static unsigned long __init m2p(phys_addr_t maddr) | |
1718 | { | |
1719 | phys_addr_t paddr; | |
1720 | ||
1721 | maddr &= PTE_PFN_MASK; | |
1722 | paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT; | |
1723 | ||
1724 | return paddr; | |
1725 | } | |
1726 | ||
1727 | /* Convert a machine address to kernel virtual */ | |
1728 | static void * __init m2v(phys_addr_t maddr) | |
1729 | { | |
1730 | return __ka(m2p(maddr)); | |
1731 | } | |
1732 | ||
1733 | /* Set the page permissions on an identity-mapped pages */ | |
1734 | static void __init set_page_prot_flags(void *addr, pgprot_t prot, | |
1735 | unsigned long flags) | |
1736 | { | |
1737 | unsigned long pfn = __pa(addr) >> PAGE_SHIFT; | |
1738 | pte_t pte = pfn_pte(pfn, prot); | |
1739 | ||
1740 | if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags)) | |
1741 | BUG(); | |
1742 | } | |
1743 | static void __init set_page_prot(void *addr, pgprot_t prot) | |
1744 | { | |
1745 | return set_page_prot_flags(addr, prot, UVMF_NONE); | |
1746 | } | |
1747 | #ifdef CONFIG_X86_32 | |
1748 | static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn) | |
1749 | { | |
1750 | unsigned pmdidx, pteidx; | |
1751 | unsigned ident_pte; | |
1752 | unsigned long pfn; | |
1753 | ||
1754 | level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES, | |
1755 | PAGE_SIZE); | |
1756 | ||
1757 | ident_pte = 0; | |
1758 | pfn = 0; | |
1759 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) { | |
1760 | pte_t *pte_page; | |
1761 | ||
1762 | /* Reuse or allocate a page of ptes */ | |
1763 | if (pmd_present(pmd[pmdidx])) | |
1764 | pte_page = m2v(pmd[pmdidx].pmd); | |
1765 | else { | |
1766 | /* Check for free pte pages */ | |
1767 | if (ident_pte == LEVEL1_IDENT_ENTRIES) | |
1768 | break; | |
1769 | ||
1770 | pte_page = &level1_ident_pgt[ident_pte]; | |
1771 | ident_pte += PTRS_PER_PTE; | |
1772 | ||
1773 | pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE); | |
1774 | } | |
1775 | ||
1776 | /* Install mappings */ | |
1777 | for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) { | |
1778 | pte_t pte; | |
1779 | ||
1780 | if (pfn > max_pfn_mapped) | |
1781 | max_pfn_mapped = pfn; | |
1782 | ||
1783 | if (!pte_none(pte_page[pteidx])) | |
1784 | continue; | |
1785 | ||
1786 | pte = pfn_pte(pfn, PAGE_KERNEL_EXEC); | |
1787 | pte_page[pteidx] = pte; | |
1788 | } | |
1789 | } | |
1790 | ||
1791 | for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE) | |
1792 | set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO); | |
1793 | ||
1794 | set_page_prot(pmd, PAGE_KERNEL_RO); | |
1795 | } | |
1796 | #endif | |
1797 | void __init xen_setup_machphys_mapping(void) | |
1798 | { | |
1799 | struct xen_machphys_mapping mapping; | |
1800 | ||
1801 | if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) { | |
1802 | machine_to_phys_mapping = (unsigned long *)mapping.v_start; | |
1803 | machine_to_phys_nr = mapping.max_mfn + 1; | |
1804 | } else { | |
1805 | machine_to_phys_nr = MACH2PHYS_NR_ENTRIES; | |
1806 | } | |
1807 | #ifdef CONFIG_X86_32 | |
1808 | WARN_ON((machine_to_phys_mapping + (machine_to_phys_nr - 1)) | |
1809 | < machine_to_phys_mapping); | |
1810 | #endif | |
1811 | } | |
1812 | ||
1813 | #ifdef CONFIG_X86_64 | |
1814 | static void __init convert_pfn_mfn(void *v) | |
1815 | { | |
1816 | pte_t *pte = v; | |
1817 | int i; | |
1818 | ||
1819 | /* All levels are converted the same way, so just treat them | |
1820 | as ptes. */ | |
1821 | for (i = 0; i < PTRS_PER_PTE; i++) | |
1822 | pte[i] = xen_make_pte(pte[i].pte); | |
1823 | } | |
1824 | static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end, | |
1825 | unsigned long addr) | |
1826 | { | |
1827 | if (*pt_base == PFN_DOWN(__pa(addr))) { | |
1828 | set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG); | |
1829 | clear_page((void *)addr); | |
1830 | (*pt_base)++; | |
1831 | } | |
1832 | if (*pt_end == PFN_DOWN(__pa(addr))) { | |
1833 | set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG); | |
1834 | clear_page((void *)addr); | |
1835 | (*pt_end)--; | |
1836 | } | |
1837 | } | |
1838 | /* | |
1839 | * Set up the initial kernel pagetable. | |
1840 | * | |
1841 | * We can construct this by grafting the Xen provided pagetable into | |
1842 | * head_64.S's preconstructed pagetables. We copy the Xen L2's into | |
1843 | * level2_ident_pgt, and level2_kernel_pgt. This means that only the | |
1844 | * kernel has a physical mapping to start with - but that's enough to | |
1845 | * get __va working. We need to fill in the rest of the physical | |
1846 | * mapping once some sort of allocator has been set up. | |
1847 | */ | |
1848 | void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn) | |
1849 | { | |
1850 | pud_t *l3; | |
1851 | pmd_t *l2; | |
1852 | unsigned long addr[3]; | |
1853 | unsigned long pt_base, pt_end; | |
1854 | unsigned i; | |
1855 | ||
1856 | /* max_pfn_mapped is the last pfn mapped in the initial memory | |
1857 | * mappings. Considering that on Xen after the kernel mappings we | |
1858 | * have the mappings of some pages that don't exist in pfn space, we | |
1859 | * set max_pfn_mapped to the last real pfn mapped. */ | |
1860 | if (xen_start_info->mfn_list < __START_KERNEL_map) | |
1861 | max_pfn_mapped = xen_start_info->first_p2m_pfn; | |
1862 | else | |
1863 | max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list)); | |
1864 | ||
1865 | pt_base = PFN_DOWN(__pa(xen_start_info->pt_base)); | |
1866 | pt_end = pt_base + xen_start_info->nr_pt_frames; | |
1867 | ||
1868 | /* Zap identity mapping */ | |
65ade2f8 | 1869 | init_top_pgt[0] = __pgd(0); |
7e0563de | 1870 | |
989513a7 | 1871 | /* Pre-constructed entries are in pfn, so convert to mfn */ |
d412ab7c | 1872 | /* L4[273] -> level3_ident_pgt */ |
989513a7 | 1873 | /* L4[511] -> level3_kernel_pgt */ |
65ade2f8 | 1874 | convert_pfn_mfn(init_top_pgt); |
7e0563de | 1875 | |
989513a7 JG |
1876 | /* L3_i[0] -> level2_ident_pgt */ |
1877 | convert_pfn_mfn(level3_ident_pgt); | |
1878 | /* L3_k[510] -> level2_kernel_pgt */ | |
1879 | /* L3_k[511] -> level2_fixmap_pgt */ | |
1880 | convert_pfn_mfn(level3_kernel_pgt); | |
1881 | ||
4fe780c1 | 1882 | /* L3_k[511][508-FIXMAP_PMD_NUM ... 507] -> level1_fixmap_pgt */ |
989513a7 | 1883 | convert_pfn_mfn(level2_fixmap_pgt); |
7e0563de | 1884 | |
7e0563de VK |
1885 | /* We get [511][511] and have Xen's version of level2_kernel_pgt */ |
1886 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd); | |
1887 | l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud); | |
1888 | ||
1889 | addr[0] = (unsigned long)pgd; | |
1890 | addr[1] = (unsigned long)l3; | |
1891 | addr[2] = (unsigned long)l2; | |
d412ab7c KS |
1892 | /* Graft it onto L4[273][0]. Note that we creating an aliasing problem: |
1893 | * Both L4[273][0] and L4[511][510] have entries that point to the same | |
7e0563de VK |
1894 | * L2 (PMD) tables. Meaning that if you modify it in __va space |
1895 | * it will be also modified in the __ka space! (But if you just | |
1896 | * modify the PMD table to point to other PTE's or none, then you | |
1897 | * are OK - which is what cleanup_highmap does) */ | |
1898 | copy_page(level2_ident_pgt, l2); | |
1899 | /* Graft it onto L4[511][510] */ | |
1900 | copy_page(level2_kernel_pgt, l2); | |
1901 | ||
c7f40ff4 JB |
1902 | /* |
1903 | * Zap execute permission from the ident map. Due to the sharing of | |
1904 | * L1 entries we need to do this in the L2. | |
1905 | */ | |
1906 | if (__supported_pte_mask & _PAGE_NX) { | |
1907 | for (i = 0; i < PTRS_PER_PMD; ++i) { | |
1908 | if (pmd_none(level2_ident_pgt[i])) | |
1909 | continue; | |
1910 | level2_ident_pgt[i] = pmd_set_flags(level2_ident_pgt[i], _PAGE_NX); | |
1911 | } | |
1912 | } | |
1913 | ||
7e0563de VK |
1914 | /* Copy the initial P->M table mappings if necessary. */ |
1915 | i = pgd_index(xen_start_info->mfn_list); | |
1916 | if (i && i < pgd_index(__START_KERNEL_map)) | |
65ade2f8 | 1917 | init_top_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i]; |
7e0563de | 1918 | |
989513a7 | 1919 | /* Make pagetable pieces RO */ |
65ade2f8 | 1920 | set_page_prot(init_top_pgt, PAGE_KERNEL_RO); |
989513a7 JG |
1921 | set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO); |
1922 | set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO); | |
1923 | set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO); | |
1924 | set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO); | |
1925 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); | |
1926 | set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO); | |
4fe780c1 FT |
1927 | |
1928 | for (i = 0; i < FIXMAP_PMD_NUM; i++) { | |
1929 | set_page_prot(level1_fixmap_pgt + i * PTRS_PER_PTE, | |
1930 | PAGE_KERNEL_RO); | |
1931 | } | |
989513a7 JG |
1932 | |
1933 | /* Pin down new L4 */ | |
1934 | pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE, | |
65ade2f8 | 1935 | PFN_DOWN(__pa_symbol(init_top_pgt))); |
989513a7 JG |
1936 | |
1937 | /* Unpin Xen-provided one */ | |
1938 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
7e0563de | 1939 | |
989513a7 JG |
1940 | /* |
1941 | * At this stage there can be no user pgd, and no page structure to | |
1942 | * attach it to, so make sure we just set kernel pgd. | |
1943 | */ | |
1944 | xen_mc_batch(); | |
65ade2f8 | 1945 | __xen_write_cr3(true, __pa(init_top_pgt)); |
989513a7 | 1946 | xen_mc_issue(PARAVIRT_LAZY_CPU); |
7e0563de VK |
1947 | |
1948 | /* We can't that easily rip out L3 and L2, as the Xen pagetables are | |
1949 | * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ... for | |
1950 | * the initial domain. For guests using the toolstack, they are in: | |
1951 | * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only | |
1952 | * rip out the [L4] (pgd), but for guests we shave off three pages. | |
1953 | */ | |
1954 | for (i = 0; i < ARRAY_SIZE(addr); i++) | |
1955 | check_pt_base(&pt_base, &pt_end, addr[i]); | |
1956 | ||
1957 | /* Our (by three pages) smaller Xen pagetable that we are using */ | |
1958 | xen_pt_base = PFN_PHYS(pt_base); | |
1959 | xen_pt_size = (pt_end - pt_base) * PAGE_SIZE; | |
1960 | memblock_reserve(xen_pt_base, xen_pt_size); | |
1961 | ||
1962 | /* Revector the xen_start_info */ | |
1963 | xen_start_info = (struct start_info *)__va(__pa(xen_start_info)); | |
1964 | } | |
1965 | ||
1966 | /* | |
1967 | * Read a value from a physical address. | |
1968 | */ | |
1969 | static unsigned long __init xen_read_phys_ulong(phys_addr_t addr) | |
1970 | { | |
1971 | unsigned long *vaddr; | |
1972 | unsigned long val; | |
1973 | ||
1974 | vaddr = early_memremap_ro(addr, sizeof(val)); | |
1975 | val = *vaddr; | |
1976 | early_memunmap(vaddr, sizeof(val)); | |
1977 | return val; | |
1978 | } | |
1979 | ||
1980 | /* | |
1981 | * Translate a virtual address to a physical one without relying on mapped | |
69861e0a JG |
1982 | * page tables. Don't rely on big pages being aligned in (guest) physical |
1983 | * space! | |
7e0563de VK |
1984 | */ |
1985 | static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr) | |
1986 | { | |
1987 | phys_addr_t pa; | |
1988 | pgd_t pgd; | |
1989 | pud_t pud; | |
1990 | pmd_t pmd; | |
1991 | pte_t pte; | |
1992 | ||
6c690ee1 | 1993 | pa = read_cr3_pa(); |
7e0563de VK |
1994 | pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) * |
1995 | sizeof(pgd))); | |
1996 | if (!pgd_present(pgd)) | |
1997 | return 0; | |
1998 | ||
1999 | pa = pgd_val(pgd) & PTE_PFN_MASK; | |
2000 | pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) * | |
2001 | sizeof(pud))); | |
2002 | if (!pud_present(pud)) | |
2003 | return 0; | |
69861e0a | 2004 | pa = pud_val(pud) & PTE_PFN_MASK; |
7e0563de VK |
2005 | if (pud_large(pud)) |
2006 | return pa + (vaddr & ~PUD_MASK); | |
2007 | ||
2008 | pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) * | |
2009 | sizeof(pmd))); | |
2010 | if (!pmd_present(pmd)) | |
2011 | return 0; | |
69861e0a | 2012 | pa = pmd_val(pmd) & PTE_PFN_MASK; |
7e0563de VK |
2013 | if (pmd_large(pmd)) |
2014 | return pa + (vaddr & ~PMD_MASK); | |
2015 | ||
2016 | pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) * | |
2017 | sizeof(pte))); | |
2018 | if (!pte_present(pte)) | |
2019 | return 0; | |
2020 | pa = pte_pfn(pte) << PAGE_SHIFT; | |
2021 | ||
2022 | return pa | (vaddr & ~PAGE_MASK); | |
2023 | } | |
2024 | ||
2025 | /* | |
2026 | * Find a new area for the hypervisor supplied p2m list and relocate the p2m to | |
2027 | * this area. | |
2028 | */ | |
2029 | void __init xen_relocate_p2m(void) | |
2030 | { | |
af02cd97 | 2031 | phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys; |
7e0563de | 2032 | unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end; |
af02cd97 | 2033 | int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud; |
7e0563de VK |
2034 | pte_t *pt; |
2035 | pmd_t *pmd; | |
2036 | pud_t *pud; | |
7e0563de VK |
2037 | pgd_t *pgd; |
2038 | unsigned long *new_p2m; | |
2039 | int save_pud; | |
2040 | ||
2041 | size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long)); | |
2042 | n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT; | |
2043 | n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT; | |
2044 | n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT; | |
2045 | n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT; | |
af02cd97 | 2046 | n_frames = n_pte + n_pt + n_pmd + n_pud; |
7e0563de VK |
2047 | |
2048 | new_area = xen_find_free_area(PFN_PHYS(n_frames)); | |
2049 | if (!new_area) { | |
2050 | xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n"); | |
2051 | BUG(); | |
2052 | } | |
2053 | ||
2054 | /* | |
2055 | * Setup the page tables for addressing the new p2m list. | |
2056 | * We have asked the hypervisor to map the p2m list at the user address | |
2057 | * PUD_SIZE. It may have done so, or it may have used a kernel space | |
2058 | * address depending on the Xen version. | |
2059 | * To avoid any possible virtual address collision, just use | |
2060 | * 2 * PUD_SIZE for the new area. | |
2061 | */ | |
af02cd97 | 2062 | pud_phys = new_area; |
7e0563de VK |
2063 | pmd_phys = pud_phys + PFN_PHYS(n_pud); |
2064 | pt_phys = pmd_phys + PFN_PHYS(n_pmd); | |
2065 | p2m_pfn = PFN_DOWN(pt_phys) + n_pt; | |
2066 | ||
6c690ee1 | 2067 | pgd = __va(read_cr3_pa()); |
7e0563de | 2068 | new_p2m = (unsigned long *)(2 * PGDIR_SIZE); |
7e0563de | 2069 | save_pud = n_pud; |
af02cd97 KS |
2070 | for (idx_pud = 0; idx_pud < n_pud; idx_pud++) { |
2071 | pud = early_memremap(pud_phys, PAGE_SIZE); | |
2072 | clear_page(pud); | |
2073 | for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD); | |
2074 | idx_pmd++) { | |
2075 | pmd = early_memremap(pmd_phys, PAGE_SIZE); | |
2076 | clear_page(pmd); | |
2077 | for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD); | |
2078 | idx_pt++) { | |
2079 | pt = early_memremap(pt_phys, PAGE_SIZE); | |
2080 | clear_page(pt); | |
2081 | for (idx_pte = 0; | |
2082 | idx_pte < min(n_pte, PTRS_PER_PTE); | |
2083 | idx_pte++) { | |
2084 | set_pte(pt + idx_pte, | |
2085 | pfn_pte(p2m_pfn, PAGE_KERNEL)); | |
2086 | p2m_pfn++; | |
7e0563de | 2087 | } |
af02cd97 KS |
2088 | n_pte -= PTRS_PER_PTE; |
2089 | early_memunmap(pt, PAGE_SIZE); | |
2090 | make_lowmem_page_readonly(__va(pt_phys)); | |
2091 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, | |
2092 | PFN_DOWN(pt_phys)); | |
2093 | set_pmd(pmd + idx_pt, | |
2094 | __pmd(_PAGE_TABLE | pt_phys)); | |
2095 | pt_phys += PAGE_SIZE; | |
7e0563de | 2096 | } |
af02cd97 KS |
2097 | n_pt -= PTRS_PER_PMD; |
2098 | early_memunmap(pmd, PAGE_SIZE); | |
2099 | make_lowmem_page_readonly(__va(pmd_phys)); | |
2100 | pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE, | |
2101 | PFN_DOWN(pmd_phys)); | |
2102 | set_pud(pud + idx_pmd, __pud(_PAGE_TABLE | pmd_phys)); | |
2103 | pmd_phys += PAGE_SIZE; | |
7e0563de | 2104 | } |
af02cd97 KS |
2105 | n_pmd -= PTRS_PER_PUD; |
2106 | early_memunmap(pud, PAGE_SIZE); | |
2107 | make_lowmem_page_readonly(__va(pud_phys)); | |
2108 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys)); | |
2109 | set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys)); | |
2110 | pud_phys += PAGE_SIZE; | |
2111 | } | |
7e0563de VK |
2112 | |
2113 | /* Now copy the old p2m info to the new area. */ | |
2114 | memcpy(new_p2m, xen_p2m_addr, size); | |
2115 | xen_p2m_addr = new_p2m; | |
2116 | ||
2117 | /* Release the old p2m list and set new list info. */ | |
2118 | p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list)); | |
2119 | BUG_ON(!p2m_pfn); | |
2120 | p2m_pfn_end = p2m_pfn + PFN_DOWN(size); | |
2121 | ||
2122 | if (xen_start_info->mfn_list < __START_KERNEL_map) { | |
2123 | pfn = xen_start_info->first_p2m_pfn; | |
2124 | pfn_end = xen_start_info->first_p2m_pfn + | |
2125 | xen_start_info->nr_p2m_frames; | |
2126 | set_pgd(pgd + 1, __pgd(0)); | |
2127 | } else { | |
2128 | pfn = p2m_pfn; | |
2129 | pfn_end = p2m_pfn_end; | |
2130 | } | |
2131 | ||
2132 | memblock_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn)); | |
2133 | while (pfn < pfn_end) { | |
2134 | if (pfn == p2m_pfn) { | |
2135 | pfn = p2m_pfn_end; | |
2136 | continue; | |
2137 | } | |
2138 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
2139 | pfn++; | |
2140 | } | |
2141 | ||
2142 | xen_start_info->mfn_list = (unsigned long)xen_p2m_addr; | |
2143 | xen_start_info->first_p2m_pfn = PFN_DOWN(new_area); | |
2144 | xen_start_info->nr_p2m_frames = n_frames; | |
2145 | } | |
2146 | ||
2147 | #else /* !CONFIG_X86_64 */ | |
2148 | static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD); | |
2149 | static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD); | |
2150 | ||
2151 | static void __init xen_write_cr3_init(unsigned long cr3) | |
2152 | { | |
2153 | unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir)); | |
2154 | ||
6c690ee1 | 2155 | BUG_ON(read_cr3_pa() != __pa(initial_page_table)); |
7e0563de VK |
2156 | BUG_ON(cr3 != __pa(swapper_pg_dir)); |
2157 | ||
2158 | /* | |
2159 | * We are switching to swapper_pg_dir for the first time (from | |
2160 | * initial_page_table) and therefore need to mark that page | |
2161 | * read-only and then pin it. | |
2162 | * | |
2163 | * Xen disallows sharing of kernel PMDs for PAE | |
2164 | * guests. Therefore we must copy the kernel PMD from | |
2165 | * initial_page_table into a new kernel PMD to be used in | |
2166 | * swapper_pg_dir. | |
2167 | */ | |
2168 | swapper_kernel_pmd = | |
2169 | extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE); | |
2170 | copy_page(swapper_kernel_pmd, initial_kernel_pmd); | |
2171 | swapper_pg_dir[KERNEL_PGD_BOUNDARY] = | |
2172 | __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT); | |
2173 | set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO); | |
2174 | ||
2175 | set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO); | |
2176 | xen_write_cr3(cr3); | |
2177 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn); | |
2178 | ||
2179 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, | |
2180 | PFN_DOWN(__pa(initial_page_table))); | |
2181 | set_page_prot(initial_page_table, PAGE_KERNEL); | |
2182 | set_page_prot(initial_kernel_pmd, PAGE_KERNEL); | |
2183 | ||
2184 | pv_mmu_ops.write_cr3 = &xen_write_cr3; | |
2185 | } | |
2186 | ||
2187 | /* | |
2188 | * For 32 bit domains xen_start_info->pt_base is the pgd address which might be | |
2189 | * not the first page table in the page table pool. | |
2190 | * Iterate through the initial page tables to find the real page table base. | |
2191 | */ | |
51ae2538 | 2192 | static phys_addr_t __init xen_find_pt_base(pmd_t *pmd) |
7e0563de VK |
2193 | { |
2194 | phys_addr_t pt_base, paddr; | |
2195 | unsigned pmdidx; | |
2196 | ||
2197 | pt_base = min(__pa(xen_start_info->pt_base), __pa(pmd)); | |
2198 | ||
2199 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) | |
2200 | if (pmd_present(pmd[pmdidx]) && !pmd_large(pmd[pmdidx])) { | |
2201 | paddr = m2p(pmd[pmdidx].pmd); | |
2202 | pt_base = min(pt_base, paddr); | |
2203 | } | |
2204 | ||
2205 | return pt_base; | |
2206 | } | |
2207 | ||
2208 | void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn) | |
2209 | { | |
2210 | pmd_t *kernel_pmd; | |
2211 | ||
2212 | kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd); | |
2213 | ||
2214 | xen_pt_base = xen_find_pt_base(kernel_pmd); | |
2215 | xen_pt_size = xen_start_info->nr_pt_frames * PAGE_SIZE; | |
2216 | ||
2217 | initial_kernel_pmd = | |
2218 | extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE); | |
2219 | ||
2220 | max_pfn_mapped = PFN_DOWN(xen_pt_base + xen_pt_size + 512 * 1024); | |
2221 | ||
2222 | copy_page(initial_kernel_pmd, kernel_pmd); | |
2223 | ||
2224 | xen_map_identity_early(initial_kernel_pmd, max_pfn); | |
2225 | ||
2226 | copy_page(initial_page_table, pgd); | |
2227 | initial_page_table[KERNEL_PGD_BOUNDARY] = | |
2228 | __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT); | |
2229 | ||
2230 | set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO); | |
2231 | set_page_prot(initial_page_table, PAGE_KERNEL_RO); | |
2232 | set_page_prot(empty_zero_page, PAGE_KERNEL_RO); | |
2233 | ||
2234 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
2235 | ||
2236 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, | |
2237 | PFN_DOWN(__pa(initial_page_table))); | |
2238 | xen_write_cr3(__pa(initial_page_table)); | |
2239 | ||
2240 | memblock_reserve(xen_pt_base, xen_pt_size); | |
2241 | } | |
2242 | #endif /* CONFIG_X86_64 */ | |
2243 | ||
2244 | void __init xen_reserve_special_pages(void) | |
2245 | { | |
2246 | phys_addr_t paddr; | |
2247 | ||
2248 | memblock_reserve(__pa(xen_start_info), PAGE_SIZE); | |
2249 | if (xen_start_info->store_mfn) { | |
2250 | paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn)); | |
2251 | memblock_reserve(paddr, PAGE_SIZE); | |
2252 | } | |
2253 | if (!xen_initial_domain()) { | |
2254 | paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn)); | |
2255 | memblock_reserve(paddr, PAGE_SIZE); | |
2256 | } | |
2257 | } | |
2258 | ||
2259 | void __init xen_pt_check_e820(void) | |
2260 | { | |
2261 | if (xen_is_e820_reserved(xen_pt_base, xen_pt_size)) { | |
2262 | xen_raw_console_write("Xen hypervisor allocated page table memory conflicts with E820 map\n"); | |
2263 | BUG(); | |
2264 | } | |
2265 | } | |
2266 | ||
2267 | static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss; | |
2268 | ||
2269 | static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot) | |
2270 | { | |
2271 | pte_t pte; | |
2272 | ||
2273 | phys >>= PAGE_SHIFT; | |
2274 | ||
2275 | switch (idx) { | |
2276 | case FIX_BTMAP_END ... FIX_BTMAP_BEGIN: | |
7e0563de VK |
2277 | #ifdef CONFIG_X86_32 |
2278 | case FIX_WP_TEST: | |
2279 | # ifdef CONFIG_HIGHMEM | |
2280 | case FIX_KMAP_BEGIN ... FIX_KMAP_END: | |
2281 | # endif | |
2282 | #elif defined(CONFIG_X86_VSYSCALL_EMULATION) | |
2283 | case VSYSCALL_PAGE: | |
2284 | #endif | |
2285 | case FIX_TEXT_POKE0: | |
2286 | case FIX_TEXT_POKE1: | |
7e0563de VK |
2287 | /* All local page mappings */ |
2288 | pte = pfn_pte(phys, prot); | |
2289 | break; | |
2290 | ||
2291 | #ifdef CONFIG_X86_LOCAL_APIC | |
2292 | case FIX_APIC_BASE: /* maps dummy local APIC */ | |
2293 | pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL); | |
2294 | break; | |
2295 | #endif | |
2296 | ||
2297 | #ifdef CONFIG_X86_IO_APIC | |
2298 | case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END: | |
2299 | /* | |
2300 | * We just don't map the IO APIC - all access is via | |
2301 | * hypercalls. Keep the address in the pte for reference. | |
2302 | */ | |
2303 | pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL); | |
2304 | break; | |
2305 | #endif | |
2306 | ||
2307 | case FIX_PARAVIRT_BOOTMAP: | |
2308 | /* This is an MFN, but it isn't an IO mapping from the | |
2309 | IO domain */ | |
2310 | pte = mfn_pte(phys, prot); | |
2311 | break; | |
2312 | ||
2313 | default: | |
2314 | /* By default, set_fixmap is used for hardware mappings */ | |
2315 | pte = mfn_pte(phys, prot); | |
2316 | break; | |
2317 | } | |
2318 | ||
2319 | __native_set_fixmap(idx, pte); | |
2320 | ||
2321 | #ifdef CONFIG_X86_VSYSCALL_EMULATION | |
2322 | /* Replicate changes to map the vsyscall page into the user | |
2323 | pagetable vsyscall mapping. */ | |
2324 | if (idx == VSYSCALL_PAGE) { | |
2325 | unsigned long vaddr = __fix_to_virt(idx); | |
2326 | set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte); | |
2327 | } | |
2328 | #endif | |
2329 | } | |
2330 | ||
2331 | static void __init xen_post_allocator_init(void) | |
2332 | { | |
7e0563de VK |
2333 | pv_mmu_ops.set_pte = xen_set_pte; |
2334 | pv_mmu_ops.set_pmd = xen_set_pmd; | |
2335 | pv_mmu_ops.set_pud = xen_set_pud; | |
af02cd97 | 2336 | #ifdef CONFIG_X86_64 |
7e0563de VK |
2337 | pv_mmu_ops.set_p4d = xen_set_p4d; |
2338 | #endif | |
2339 | ||
2340 | /* This will work as long as patching hasn't happened yet | |
2341 | (which it hasn't) */ | |
2342 | pv_mmu_ops.alloc_pte = xen_alloc_pte; | |
2343 | pv_mmu_ops.alloc_pmd = xen_alloc_pmd; | |
2344 | pv_mmu_ops.release_pte = xen_release_pte; | |
2345 | pv_mmu_ops.release_pmd = xen_release_pmd; | |
af02cd97 | 2346 | #ifdef CONFIG_X86_64 |
7e0563de VK |
2347 | pv_mmu_ops.alloc_pud = xen_alloc_pud; |
2348 | pv_mmu_ops.release_pud = xen_release_pud; | |
2349 | #endif | |
2350 | pv_mmu_ops.make_pte = PV_CALLEE_SAVE(xen_make_pte); | |
2351 | ||
2352 | #ifdef CONFIG_X86_64 | |
2353 | pv_mmu_ops.write_cr3 = &xen_write_cr3; | |
2354 | SetPagePinned(virt_to_page(level3_user_vsyscall)); | |
2355 | #endif | |
2356 | xen_mark_init_mm_pinned(); | |
2357 | } | |
2358 | ||
2359 | static void xen_leave_lazy_mmu(void) | |
2360 | { | |
2361 | preempt_disable(); | |
2362 | xen_mc_flush(); | |
2363 | paravirt_leave_lazy_mmu(); | |
2364 | preempt_enable(); | |
2365 | } | |
2366 | ||
2367 | static const struct pv_mmu_ops xen_mmu_ops __initconst = { | |
2368 | .read_cr2 = xen_read_cr2, | |
2369 | .write_cr2 = xen_write_cr2, | |
2370 | ||
2371 | .read_cr3 = xen_read_cr3, | |
2372 | .write_cr3 = xen_write_cr3_init, | |
2373 | ||
2374 | .flush_tlb_user = xen_flush_tlb, | |
2375 | .flush_tlb_kernel = xen_flush_tlb, | |
208beef6 | 2376 | .flush_tlb_one_user = xen_flush_tlb_one_user, |
7e0563de VK |
2377 | .flush_tlb_others = xen_flush_tlb_others, |
2378 | ||
7e0563de VK |
2379 | .pgd_alloc = xen_pgd_alloc, |
2380 | .pgd_free = xen_pgd_free, | |
2381 | ||
2382 | .alloc_pte = xen_alloc_pte_init, | |
2383 | .release_pte = xen_release_pte_init, | |
2384 | .alloc_pmd = xen_alloc_pmd_init, | |
2385 | .release_pmd = xen_release_pmd_init, | |
2386 | ||
2387 | .set_pte = xen_set_pte_init, | |
2388 | .set_pte_at = xen_set_pte_at, | |
2389 | .set_pmd = xen_set_pmd_hyper, | |
2390 | ||
2391 | .ptep_modify_prot_start = __ptep_modify_prot_start, | |
2392 | .ptep_modify_prot_commit = __ptep_modify_prot_commit, | |
2393 | ||
2394 | .pte_val = PV_CALLEE_SAVE(xen_pte_val), | |
2395 | .pgd_val = PV_CALLEE_SAVE(xen_pgd_val), | |
2396 | ||
2397 | .make_pte = PV_CALLEE_SAVE(xen_make_pte_init), | |
2398 | .make_pgd = PV_CALLEE_SAVE(xen_make_pgd), | |
2399 | ||
2400 | #ifdef CONFIG_X86_PAE | |
2401 | .set_pte_atomic = xen_set_pte_atomic, | |
2402 | .pte_clear = xen_pte_clear, | |
2403 | .pmd_clear = xen_pmd_clear, | |
2404 | #endif /* CONFIG_X86_PAE */ | |
2405 | .set_pud = xen_set_pud_hyper, | |
2406 | ||
2407 | .make_pmd = PV_CALLEE_SAVE(xen_make_pmd), | |
2408 | .pmd_val = PV_CALLEE_SAVE(xen_pmd_val), | |
2409 | ||
af02cd97 | 2410 | #ifdef CONFIG_X86_64 |
7e0563de VK |
2411 | .pud_val = PV_CALLEE_SAVE(xen_pud_val), |
2412 | .make_pud = PV_CALLEE_SAVE(xen_make_pud), | |
2413 | .set_p4d = xen_set_p4d_hyper, | |
2414 | ||
2415 | .alloc_pud = xen_alloc_pmd_init, | |
2416 | .release_pud = xen_release_pmd_init, | |
af02cd97 | 2417 | #endif /* CONFIG_X86_64 */ |
7e0563de VK |
2418 | |
2419 | .activate_mm = xen_activate_mm, | |
2420 | .dup_mmap = xen_dup_mmap, | |
2421 | .exit_mmap = xen_exit_mmap, | |
2422 | ||
2423 | .lazy_mode = { | |
2424 | .enter = paravirt_enter_lazy_mmu, | |
2425 | .leave = xen_leave_lazy_mmu, | |
2426 | .flush = paravirt_flush_lazy_mmu, | |
2427 | }, | |
2428 | ||
2429 | .set_fixmap = xen_set_fixmap, | |
2430 | }; | |
2431 | ||
2432 | void __init xen_init_mmu_ops(void) | |
2433 | { | |
2434 | x86_init.paging.pagetable_init = xen_pagetable_init; | |
2435 | ||
7e0563de VK |
2436 | pv_mmu_ops = xen_mmu_ops; |
2437 | ||
2438 | memset(dummy_mapping, 0xff, PAGE_SIZE); | |
2439 | } | |
2440 | ||
2441 | /* Protected by xen_reservation_lock. */ | |
2442 | #define MAX_CONTIG_ORDER 9 /* 2MB */ | |
2443 | static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER]; | |
2444 | ||
2445 | #define VOID_PTE (mfn_pte(0, __pgprot(0))) | |
2446 | static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order, | |
2447 | unsigned long *in_frames, | |
2448 | unsigned long *out_frames) | |
2449 | { | |
2450 | int i; | |
2451 | struct multicall_space mcs; | |
2452 | ||
2453 | xen_mc_batch(); | |
2454 | for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) { | |
2455 | mcs = __xen_mc_entry(0); | |
2456 | ||
2457 | if (in_frames) | |
2458 | in_frames[i] = virt_to_mfn(vaddr); | |
2459 | ||
2460 | MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0); | |
2461 | __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY); | |
2462 | ||
2463 | if (out_frames) | |
2464 | out_frames[i] = virt_to_pfn(vaddr); | |
2465 | } | |
2466 | xen_mc_issue(0); | |
2467 | } | |
2468 | ||
2469 | /* | |
2470 | * Update the pfn-to-mfn mappings for a virtual address range, either to | |
2471 | * point to an array of mfns, or contiguously from a single starting | |
2472 | * mfn. | |
2473 | */ | |
2474 | static void xen_remap_exchanged_ptes(unsigned long vaddr, int order, | |
2475 | unsigned long *mfns, | |
2476 | unsigned long first_mfn) | |
2477 | { | |
2478 | unsigned i, limit; | |
2479 | unsigned long mfn; | |
2480 | ||
2481 | xen_mc_batch(); | |
2482 | ||
2483 | limit = 1u << order; | |
2484 | for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) { | |
2485 | struct multicall_space mcs; | |
2486 | unsigned flags; | |
2487 | ||
2488 | mcs = __xen_mc_entry(0); | |
2489 | if (mfns) | |
2490 | mfn = mfns[i]; | |
2491 | else | |
2492 | mfn = first_mfn + i; | |
2493 | ||
2494 | if (i < (limit - 1)) | |
2495 | flags = 0; | |
2496 | else { | |
2497 | if (order == 0) | |
2498 | flags = UVMF_INVLPG | UVMF_ALL; | |
2499 | else | |
2500 | flags = UVMF_TLB_FLUSH | UVMF_ALL; | |
2501 | } | |
2502 | ||
2503 | MULTI_update_va_mapping(mcs.mc, vaddr, | |
2504 | mfn_pte(mfn, PAGE_KERNEL), flags); | |
2505 | ||
2506 | set_phys_to_machine(virt_to_pfn(vaddr), mfn); | |
2507 | } | |
2508 | ||
2509 | xen_mc_issue(0); | |
2510 | } | |
2511 | ||
2512 | /* | |
2513 | * Perform the hypercall to exchange a region of our pfns to point to | |
2514 | * memory with the required contiguous alignment. Takes the pfns as | |
2515 | * input, and populates mfns as output. | |
2516 | * | |
2517 | * Returns a success code indicating whether the hypervisor was able to | |
2518 | * satisfy the request or not. | |
2519 | */ | |
2520 | static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in, | |
2521 | unsigned long *pfns_in, | |
2522 | unsigned long extents_out, | |
2523 | unsigned int order_out, | |
2524 | unsigned long *mfns_out, | |
2525 | unsigned int address_bits) | |
2526 | { | |
2527 | long rc; | |
2528 | int success; | |
2529 | ||
2530 | struct xen_memory_exchange exchange = { | |
2531 | .in = { | |
2532 | .nr_extents = extents_in, | |
2533 | .extent_order = order_in, | |
2534 | .extent_start = pfns_in, | |
2535 | .domid = DOMID_SELF | |
2536 | }, | |
2537 | .out = { | |
2538 | .nr_extents = extents_out, | |
2539 | .extent_order = order_out, | |
2540 | .extent_start = mfns_out, | |
2541 | .address_bits = address_bits, | |
2542 | .domid = DOMID_SELF | |
2543 | } | |
2544 | }; | |
2545 | ||
2546 | BUG_ON(extents_in << order_in != extents_out << order_out); | |
2547 | ||
2548 | rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange); | |
2549 | success = (exchange.nr_exchanged == extents_in); | |
2550 | ||
2551 | BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0))); | |
2552 | BUG_ON(success && (rc != 0)); | |
2553 | ||
2554 | return success; | |
2555 | } | |
2556 | ||
2557 | int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order, | |
2558 | unsigned int address_bits, | |
2559 | dma_addr_t *dma_handle) | |
2560 | { | |
2561 | unsigned long *in_frames = discontig_frames, out_frame; | |
2562 | unsigned long flags; | |
2563 | int success; | |
2564 | unsigned long vstart = (unsigned long)phys_to_virt(pstart); | |
2565 | ||
2566 | /* | |
2567 | * Currently an auto-translated guest will not perform I/O, nor will | |
2568 | * it require PAE page directories below 4GB. Therefore any calls to | |
2569 | * this function are redundant and can be ignored. | |
2570 | */ | |
2571 | ||
7e0563de VK |
2572 | if (unlikely(order > MAX_CONTIG_ORDER)) |
2573 | return -ENOMEM; | |
2574 | ||
2575 | memset((void *) vstart, 0, PAGE_SIZE << order); | |
2576 | ||
2577 | spin_lock_irqsave(&xen_reservation_lock, flags); | |
2578 | ||
2579 | /* 1. Zap current PTEs, remembering MFNs. */ | |
2580 | xen_zap_pfn_range(vstart, order, in_frames, NULL); | |
2581 | ||
2582 | /* 2. Get a new contiguous memory extent. */ | |
2583 | out_frame = virt_to_pfn(vstart); | |
2584 | success = xen_exchange_memory(1UL << order, 0, in_frames, | |
2585 | 1, order, &out_frame, | |
2586 | address_bits); | |
2587 | ||
2588 | /* 3. Map the new extent in place of old pages. */ | |
2589 | if (success) | |
2590 | xen_remap_exchanged_ptes(vstart, order, NULL, out_frame); | |
2591 | else | |
2592 | xen_remap_exchanged_ptes(vstart, order, in_frames, 0); | |
2593 | ||
2594 | spin_unlock_irqrestore(&xen_reservation_lock, flags); | |
2595 | ||
2596 | *dma_handle = virt_to_machine(vstart).maddr; | |
2597 | return success ? 0 : -ENOMEM; | |
2598 | } | |
2599 | EXPORT_SYMBOL_GPL(xen_create_contiguous_region); | |
2600 | ||
2601 | void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order) | |
2602 | { | |
2603 | unsigned long *out_frames = discontig_frames, in_frame; | |
2604 | unsigned long flags; | |
2605 | int success; | |
2606 | unsigned long vstart; | |
2607 | ||
7e0563de VK |
2608 | if (unlikely(order > MAX_CONTIG_ORDER)) |
2609 | return; | |
2610 | ||
2611 | vstart = (unsigned long)phys_to_virt(pstart); | |
2612 | memset((void *) vstart, 0, PAGE_SIZE << order); | |
2613 | ||
2614 | spin_lock_irqsave(&xen_reservation_lock, flags); | |
2615 | ||
2616 | /* 1. Find start MFN of contiguous extent. */ | |
2617 | in_frame = virt_to_mfn(vstart); | |
2618 | ||
2619 | /* 2. Zap current PTEs. */ | |
2620 | xen_zap_pfn_range(vstart, order, NULL, out_frames); | |
2621 | ||
2622 | /* 3. Do the exchange for non-contiguous MFNs. */ | |
2623 | success = xen_exchange_memory(1, order, &in_frame, 1UL << order, | |
2624 | 0, out_frames, 0); | |
2625 | ||
2626 | /* 4. Map new pages in place of old pages. */ | |
2627 | if (success) | |
2628 | xen_remap_exchanged_ptes(vstart, order, out_frames, 0); | |
2629 | else | |
2630 | xen_remap_exchanged_ptes(vstart, order, NULL, in_frame); | |
2631 | ||
2632 | spin_unlock_irqrestore(&xen_reservation_lock, flags); | |
2633 | } | |
2634 | EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region); | |
29985b09 JG |
2635 | |
2636 | #ifdef CONFIG_KEXEC_CORE | |
2637 | phys_addr_t paddr_vmcoreinfo_note(void) | |
2638 | { | |
2639 | if (xen_pv_domain()) | |
203e9e41 | 2640 | return virt_to_machine(vmcoreinfo_note).maddr; |
29985b09 | 2641 | else |
203e9e41 | 2642 | return __pa(vmcoreinfo_note); |
29985b09 JG |
2643 | } |
2644 | #endif /* CONFIG_KEXEC_CORE */ |