]>
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) { | |
6f0e8bf1 | 318 | unsigned long mfn = (val & XEN_PTE_MFN_MASK) >> PAGE_SHIFT; |
7e0563de VK |
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); | |
428 | set_64bit((u64 *)ptep, native_pte_val(pte)); | |
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); | |
434 | if (!xen_batched_set_pte(ptep, native_make_pte(0))) | |
435 | native_pte_clear(mm, addr, ptep); | |
436 | } | |
437 | ||
438 | static void xen_pmd_clear(pmd_t *pmdp) | |
439 | { | |
440 | trace_xen_mmu_pmd_clear(pmdp); | |
441 | set_pmd(pmdp, __pmd(0)); | |
442 | } | |
443 | #endif /* CONFIG_X86_PAE */ | |
444 | ||
445 | __visible pmd_t xen_make_pmd(pmdval_t pmd) | |
446 | { | |
447 | pmd = pte_pfn_to_mfn(pmd); | |
448 | return native_make_pmd(pmd); | |
449 | } | |
450 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd); | |
451 | ||
773dd2fc | 452 | #ifdef CONFIG_X86_64 |
7e0563de VK |
453 | __visible pudval_t xen_pud_val(pud_t pud) |
454 | { | |
455 | return pte_mfn_to_pfn(pud.pud); | |
456 | } | |
457 | PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val); | |
458 | ||
459 | __visible pud_t xen_make_pud(pudval_t pud) | |
460 | { | |
461 | pud = pte_pfn_to_mfn(pud); | |
462 | ||
463 | return native_make_pud(pud); | |
464 | } | |
465 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud); | |
466 | ||
467 | static pgd_t *xen_get_user_pgd(pgd_t *pgd) | |
468 | { | |
469 | pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK); | |
470 | unsigned offset = pgd - pgd_page; | |
471 | pgd_t *user_ptr = NULL; | |
472 | ||
473 | if (offset < pgd_index(USER_LIMIT)) { | |
474 | struct page *page = virt_to_page(pgd_page); | |
475 | user_ptr = (pgd_t *)page->private; | |
476 | if (user_ptr) | |
477 | user_ptr += offset; | |
478 | } | |
479 | ||
480 | return user_ptr; | |
481 | } | |
482 | ||
483 | static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val) | |
484 | { | |
485 | struct mmu_update u; | |
486 | ||
487 | u.ptr = virt_to_machine(ptr).maddr; | |
488 | u.val = p4d_val_ma(val); | |
489 | xen_extend_mmu_update(&u); | |
490 | } | |
491 | ||
492 | /* | |
493 | * Raw hypercall-based set_p4d, intended for in early boot before | |
494 | * there's a page structure. This implies: | |
495 | * 1. The only existing pagetable is the kernel's | |
496 | * 2. It is always pinned | |
497 | * 3. It has no user pagetable attached to it | |
498 | */ | |
499 | static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val) | |
500 | { | |
501 | preempt_disable(); | |
502 | ||
503 | xen_mc_batch(); | |
504 | ||
505 | __xen_set_p4d_hyper(ptr, val); | |
506 | ||
507 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
508 | ||
509 | preempt_enable(); | |
510 | } | |
511 | ||
512 | static void xen_set_p4d(p4d_t *ptr, p4d_t val) | |
513 | { | |
514 | pgd_t *user_ptr = xen_get_user_pgd((pgd_t *)ptr); | |
515 | pgd_t pgd_val; | |
516 | ||
517 | trace_xen_mmu_set_p4d(ptr, (p4d_t *)user_ptr, val); | |
518 | ||
519 | /* If page is not pinned, we can just update the entry | |
520 | directly */ | |
521 | if (!xen_page_pinned(ptr)) { | |
522 | *ptr = val; | |
523 | if (user_ptr) { | |
524 | WARN_ON(xen_page_pinned(user_ptr)); | |
525 | pgd_val.pgd = p4d_val_ma(val); | |
526 | *user_ptr = pgd_val; | |
527 | } | |
528 | return; | |
529 | } | |
530 | ||
531 | /* If it's pinned, then we can at least batch the kernel and | |
532 | user updates together. */ | |
533 | xen_mc_batch(); | |
534 | ||
535 | __xen_set_p4d_hyper(ptr, val); | |
536 | if (user_ptr) | |
537 | __xen_set_p4d_hyper((p4d_t *)user_ptr, val); | |
538 | ||
539 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
540 | } | |
773dd2fc | 541 | #endif /* CONFIG_X86_64 */ |
7e0563de VK |
542 | |
543 | static int xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd, | |
544 | int (*func)(struct mm_struct *mm, struct page *, enum pt_level), | |
545 | bool last, unsigned long limit) | |
546 | { | |
547 | int i, nr, flush = 0; | |
548 | ||
549 | nr = last ? pmd_index(limit) + 1 : PTRS_PER_PMD; | |
550 | for (i = 0; i < nr; i++) { | |
551 | if (!pmd_none(pmd[i])) | |
552 | flush |= (*func)(mm, pmd_page(pmd[i]), PT_PTE); | |
553 | } | |
554 | return flush; | |
555 | } | |
556 | ||
557 | static int xen_pud_walk(struct mm_struct *mm, pud_t *pud, | |
558 | int (*func)(struct mm_struct *mm, struct page *, enum pt_level), | |
559 | bool last, unsigned long limit) | |
560 | { | |
561 | int i, nr, flush = 0; | |
562 | ||
563 | nr = last ? pud_index(limit) + 1 : PTRS_PER_PUD; | |
564 | for (i = 0; i < nr; i++) { | |
565 | pmd_t *pmd; | |
566 | ||
567 | if (pud_none(pud[i])) | |
568 | continue; | |
569 | ||
570 | pmd = pmd_offset(&pud[i], 0); | |
571 | if (PTRS_PER_PMD > 1) | |
572 | flush |= (*func)(mm, virt_to_page(pmd), PT_PMD); | |
573 | flush |= xen_pmd_walk(mm, pmd, func, | |
574 | last && i == nr - 1, limit); | |
575 | } | |
576 | return flush; | |
577 | } | |
578 | ||
579 | static int xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d, | |
580 | int (*func)(struct mm_struct *mm, struct page *, enum pt_level), | |
581 | bool last, unsigned long limit) | |
582 | { | |
773dd2fc KS |
583 | int flush = 0; |
584 | pud_t *pud; | |
7e0563de | 585 | |
7e0563de | 586 | |
773dd2fc KS |
587 | if (p4d_none(*p4d)) |
588 | return flush; | |
7e0563de | 589 | |
773dd2fc KS |
590 | pud = pud_offset(p4d, 0); |
591 | if (PTRS_PER_PUD > 1) | |
592 | flush |= (*func)(mm, virt_to_page(pud), PT_PUD); | |
593 | flush |= xen_pud_walk(mm, pud, func, last, limit); | |
7e0563de VK |
594 | return flush; |
595 | } | |
596 | ||
597 | /* | |
598 | * (Yet another) pagetable walker. This one is intended for pinning a | |
599 | * pagetable. This means that it walks a pagetable and calls the | |
600 | * callback function on each page it finds making up the page table, | |
601 | * at every level. It walks the entire pagetable, but it only bothers | |
602 | * pinning pte pages which are below limit. In the normal case this | |
603 | * will be STACK_TOP_MAX, but at boot we need to pin up to | |
604 | * FIXADDR_TOP. | |
605 | * | |
606 | * For 32-bit the important bit is that we don't pin beyond there, | |
607 | * because then we start getting into Xen's ptes. | |
608 | * | |
609 | * For 64-bit, we must skip the Xen hole in the middle of the address | |
610 | * space, just after the big x86-64 virtual hole. | |
611 | */ | |
612 | static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd, | |
613 | int (*func)(struct mm_struct *mm, struct page *, | |
614 | enum pt_level), | |
615 | unsigned long limit) | |
616 | { | |
617 | int i, nr, flush = 0; | |
618 | unsigned hole_low, hole_high; | |
619 | ||
620 | /* The limit is the last byte to be touched */ | |
621 | limit--; | |
622 | BUG_ON(limit >= FIXADDR_TOP); | |
623 | ||
7e0563de VK |
624 | /* |
625 | * 64-bit has a great big hole in the middle of the address | |
626 | * space, which contains the Xen mappings. On 32-bit these | |
627 | * will end up making a zero-sized hole and so is a no-op. | |
628 | */ | |
629 | hole_low = pgd_index(USER_LIMIT); | |
630 | hole_high = pgd_index(PAGE_OFFSET); | |
631 | ||
632 | nr = pgd_index(limit) + 1; | |
633 | for (i = 0; i < nr; i++) { | |
634 | p4d_t *p4d; | |
635 | ||
636 | if (i >= hole_low && i < hole_high) | |
637 | continue; | |
638 | ||
639 | if (pgd_none(pgd[i])) | |
640 | continue; | |
641 | ||
642 | p4d = p4d_offset(&pgd[i], 0); | |
7e0563de VK |
643 | flush |= xen_p4d_walk(mm, p4d, func, i == nr - 1, limit); |
644 | } | |
645 | ||
646 | /* Do the top level last, so that the callbacks can use it as | |
647 | a cue to do final things like tlb flushes. */ | |
648 | flush |= (*func)(mm, virt_to_page(pgd), PT_PGD); | |
649 | ||
650 | return flush; | |
651 | } | |
652 | ||
653 | static int xen_pgd_walk(struct mm_struct *mm, | |
654 | int (*func)(struct mm_struct *mm, struct page *, | |
655 | enum pt_level), | |
656 | unsigned long limit) | |
657 | { | |
658 | return __xen_pgd_walk(mm, mm->pgd, func, limit); | |
659 | } | |
660 | ||
661 | /* If we're using split pte locks, then take the page's lock and | |
662 | return a pointer to it. Otherwise return NULL. */ | |
663 | static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm) | |
664 | { | |
665 | spinlock_t *ptl = NULL; | |
666 | ||
667 | #if USE_SPLIT_PTE_PTLOCKS | |
668 | ptl = ptlock_ptr(page); | |
669 | spin_lock_nest_lock(ptl, &mm->page_table_lock); | |
670 | #endif | |
671 | ||
672 | return ptl; | |
673 | } | |
674 | ||
675 | static void xen_pte_unlock(void *v) | |
676 | { | |
677 | spinlock_t *ptl = v; | |
678 | spin_unlock(ptl); | |
679 | } | |
680 | ||
681 | static void xen_do_pin(unsigned level, unsigned long pfn) | |
682 | { | |
683 | struct mmuext_op op; | |
684 | ||
685 | op.cmd = level; | |
686 | op.arg1.mfn = pfn_to_mfn(pfn); | |
687 | ||
688 | xen_extend_mmuext_op(&op); | |
689 | } | |
690 | ||
691 | static int xen_pin_page(struct mm_struct *mm, struct page *page, | |
692 | enum pt_level level) | |
693 | { | |
694 | unsigned pgfl = TestSetPagePinned(page); | |
695 | int flush; | |
696 | ||
697 | if (pgfl) | |
698 | flush = 0; /* already pinned */ | |
699 | else if (PageHighMem(page)) | |
700 | /* kmaps need flushing if we found an unpinned | |
701 | highpage */ | |
702 | flush = 1; | |
703 | else { | |
704 | void *pt = lowmem_page_address(page); | |
705 | unsigned long pfn = page_to_pfn(page); | |
706 | struct multicall_space mcs = __xen_mc_entry(0); | |
707 | spinlock_t *ptl; | |
708 | ||
709 | flush = 0; | |
710 | ||
711 | /* | |
712 | * We need to hold the pagetable lock between the time | |
713 | * we make the pagetable RO and when we actually pin | |
714 | * it. If we don't, then other users may come in and | |
715 | * attempt to update the pagetable by writing it, | |
716 | * which will fail because the memory is RO but not | |
717 | * pinned, so Xen won't do the trap'n'emulate. | |
718 | * | |
719 | * If we're using split pte locks, we can't hold the | |
720 | * entire pagetable's worth of locks during the | |
721 | * traverse, because we may wrap the preempt count (8 | |
722 | * bits). The solution is to mark RO and pin each PTE | |
723 | * page while holding the lock. This means the number | |
724 | * of locks we end up holding is never more than a | |
725 | * batch size (~32 entries, at present). | |
726 | * | |
727 | * If we're not using split pte locks, we needn't pin | |
728 | * the PTE pages independently, because we're | |
729 | * protected by the overall pagetable lock. | |
730 | */ | |
731 | ptl = NULL; | |
732 | if (level == PT_PTE) | |
733 | ptl = xen_pte_lock(page, mm); | |
734 | ||
735 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, | |
736 | pfn_pte(pfn, PAGE_KERNEL_RO), | |
737 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); | |
738 | ||
739 | if (ptl) { | |
740 | xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn); | |
741 | ||
742 | /* Queue a deferred unlock for when this batch | |
743 | is completed. */ | |
744 | xen_mc_callback(xen_pte_unlock, ptl); | |
745 | } | |
746 | } | |
747 | ||
748 | return flush; | |
749 | } | |
750 | ||
751 | /* This is called just after a mm has been created, but it has not | |
752 | been used yet. We need to make sure that its pagetable is all | |
753 | read-only, and can be pinned. */ | |
754 | static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd) | |
755 | { | |
756 | trace_xen_mmu_pgd_pin(mm, pgd); | |
757 | ||
758 | xen_mc_batch(); | |
759 | ||
760 | if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) { | |
761 | /* re-enable interrupts for flushing */ | |
762 | xen_mc_issue(0); | |
763 | ||
764 | kmap_flush_unused(); | |
765 | ||
766 | xen_mc_batch(); | |
767 | } | |
768 | ||
769 | #ifdef CONFIG_X86_64 | |
770 | { | |
771 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
772 | ||
773 | xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd))); | |
774 | ||
775 | if (user_pgd) { | |
776 | xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD); | |
777 | xen_do_pin(MMUEXT_PIN_L4_TABLE, | |
778 | PFN_DOWN(__pa(user_pgd))); | |
779 | } | |
780 | } | |
781 | #else /* CONFIG_X86_32 */ | |
782 | #ifdef CONFIG_X86_PAE | |
783 | /* Need to make sure unshared kernel PMD is pinnable */ | |
784 | xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), | |
785 | PT_PMD); | |
786 | #endif | |
787 | xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd))); | |
788 | #endif /* CONFIG_X86_64 */ | |
789 | xen_mc_issue(0); | |
790 | } | |
791 | ||
792 | static void xen_pgd_pin(struct mm_struct *mm) | |
793 | { | |
794 | __xen_pgd_pin(mm, mm->pgd); | |
795 | } | |
796 | ||
797 | /* | |
798 | * On save, we need to pin all pagetables to make sure they get their | |
799 | * mfns turned into pfns. Search the list for any unpinned pgds and pin | |
800 | * them (unpinned pgds are not currently in use, probably because the | |
801 | * process is under construction or destruction). | |
802 | * | |
803 | * Expected to be called in stop_machine() ("equivalent to taking | |
804 | * every spinlock in the system"), so the locking doesn't really | |
805 | * matter all that much. | |
806 | */ | |
807 | void xen_mm_pin_all(void) | |
808 | { | |
809 | struct page *page; | |
810 | ||
811 | spin_lock(&pgd_lock); | |
812 | ||
813 | list_for_each_entry(page, &pgd_list, lru) { | |
814 | if (!PagePinned(page)) { | |
815 | __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page)); | |
816 | SetPageSavePinned(page); | |
817 | } | |
818 | } | |
819 | ||
820 | spin_unlock(&pgd_lock); | |
821 | } | |
822 | ||
823 | /* | |
824 | * The init_mm pagetable is really pinned as soon as its created, but | |
825 | * that's before we have page structures to store the bits. So do all | |
826 | * the book-keeping now. | |
827 | */ | |
828 | static int __init xen_mark_pinned(struct mm_struct *mm, struct page *page, | |
829 | enum pt_level level) | |
830 | { | |
831 | SetPagePinned(page); | |
832 | return 0; | |
833 | } | |
834 | ||
835 | static void __init xen_mark_init_mm_pinned(void) | |
836 | { | |
837 | xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP); | |
838 | } | |
839 | ||
840 | static int xen_unpin_page(struct mm_struct *mm, struct page *page, | |
841 | enum pt_level level) | |
842 | { | |
843 | unsigned pgfl = TestClearPagePinned(page); | |
844 | ||
845 | if (pgfl && !PageHighMem(page)) { | |
846 | void *pt = lowmem_page_address(page); | |
847 | unsigned long pfn = page_to_pfn(page); | |
848 | spinlock_t *ptl = NULL; | |
849 | struct multicall_space mcs; | |
850 | ||
851 | /* | |
852 | * Do the converse to pin_page. If we're using split | |
853 | * pte locks, we must be holding the lock for while | |
854 | * the pte page is unpinned but still RO to prevent | |
855 | * concurrent updates from seeing it in this | |
856 | * partially-pinned state. | |
857 | */ | |
858 | if (level == PT_PTE) { | |
859 | ptl = xen_pte_lock(page, mm); | |
860 | ||
861 | if (ptl) | |
862 | xen_do_pin(MMUEXT_UNPIN_TABLE, pfn); | |
863 | } | |
864 | ||
865 | mcs = __xen_mc_entry(0); | |
866 | ||
867 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, | |
868 | pfn_pte(pfn, PAGE_KERNEL), | |
869 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); | |
870 | ||
871 | if (ptl) { | |
872 | /* unlock when batch completed */ | |
873 | xen_mc_callback(xen_pte_unlock, ptl); | |
874 | } | |
875 | } | |
876 | ||
877 | return 0; /* never need to flush on unpin */ | |
878 | } | |
879 | ||
880 | /* Release a pagetables pages back as normal RW */ | |
881 | static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd) | |
882 | { | |
883 | trace_xen_mmu_pgd_unpin(mm, pgd); | |
884 | ||
885 | xen_mc_batch(); | |
886 | ||
887 | xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
888 | ||
889 | #ifdef CONFIG_X86_64 | |
890 | { | |
891 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
892 | ||
893 | if (user_pgd) { | |
894 | xen_do_pin(MMUEXT_UNPIN_TABLE, | |
895 | PFN_DOWN(__pa(user_pgd))); | |
896 | xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD); | |
897 | } | |
898 | } | |
899 | #endif | |
900 | ||
901 | #ifdef CONFIG_X86_PAE | |
902 | /* Need to make sure unshared kernel PMD is unpinned */ | |
903 | xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), | |
904 | PT_PMD); | |
905 | #endif | |
906 | ||
907 | __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT); | |
908 | ||
909 | xen_mc_issue(0); | |
910 | } | |
911 | ||
912 | static void xen_pgd_unpin(struct mm_struct *mm) | |
913 | { | |
914 | __xen_pgd_unpin(mm, mm->pgd); | |
915 | } | |
916 | ||
917 | /* | |
918 | * On resume, undo any pinning done at save, so that the rest of the | |
919 | * kernel doesn't see any unexpected pinned pagetables. | |
920 | */ | |
921 | void xen_mm_unpin_all(void) | |
922 | { | |
923 | struct page *page; | |
924 | ||
925 | spin_lock(&pgd_lock); | |
926 | ||
927 | list_for_each_entry(page, &pgd_list, lru) { | |
928 | if (PageSavePinned(page)) { | |
929 | BUG_ON(!PagePinned(page)); | |
930 | __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page)); | |
931 | ClearPageSavePinned(page); | |
932 | } | |
933 | } | |
934 | ||
935 | spin_unlock(&pgd_lock); | |
936 | } | |
937 | ||
938 | static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next) | |
939 | { | |
940 | spin_lock(&next->page_table_lock); | |
941 | xen_pgd_pin(next); | |
942 | spin_unlock(&next->page_table_lock); | |
943 | } | |
944 | ||
945 | static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm) | |
946 | { | |
947 | spin_lock(&mm->page_table_lock); | |
948 | xen_pgd_pin(mm); | |
949 | spin_unlock(&mm->page_table_lock); | |
950 | } | |
951 | ||
3d28ebce | 952 | static void drop_mm_ref_this_cpu(void *info) |
7e0563de VK |
953 | { |
954 | struct mm_struct *mm = info; | |
7e0563de | 955 | |
3d28ebce | 956 | if (this_cpu_read(cpu_tlbstate.loaded_mm) == mm) |
7e0563de VK |
957 | leave_mm(smp_processor_id()); |
958 | ||
3d28ebce AL |
959 | /* |
960 | * If this cpu still has a stale cr3 reference, then make sure | |
961 | * it has been flushed. | |
962 | */ | |
7e0563de | 963 | if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd)) |
3d28ebce | 964 | xen_mc_flush(); |
7e0563de VK |
965 | } |
966 | ||
3d28ebce AL |
967 | #ifdef CONFIG_SMP |
968 | /* | |
969 | * Another cpu may still have their %cr3 pointing at the pagetable, so | |
970 | * we need to repoint it somewhere else before we can unpin it. | |
971 | */ | |
7e0563de VK |
972 | static void xen_drop_mm_ref(struct mm_struct *mm) |
973 | { | |
974 | cpumask_var_t mask; | |
975 | unsigned cpu; | |
976 | ||
3d28ebce | 977 | drop_mm_ref_this_cpu(mm); |
7e0563de VK |
978 | |
979 | /* Get the "official" set of cpus referring to our pagetable. */ | |
980 | if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) { | |
981 | for_each_online_cpu(cpu) { | |
94b1b03b | 982 | if (per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd)) |
7e0563de | 983 | continue; |
3d28ebce | 984 | smp_call_function_single(cpu, drop_mm_ref_this_cpu, mm, 1); |
7e0563de VK |
985 | } |
986 | return; | |
987 | } | |
7e0563de | 988 | |
3d28ebce AL |
989 | /* |
990 | * It's possible that a vcpu may have a stale reference to our | |
991 | * cr3, because its in lazy mode, and it hasn't yet flushed | |
992 | * its set of pending hypercalls yet. In this case, we can | |
993 | * look at its actual current cr3 value, and force it to flush | |
994 | * if needed. | |
995 | */ | |
94b1b03b | 996 | cpumask_clear(mask); |
7e0563de VK |
997 | for_each_online_cpu(cpu) { |
998 | if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd)) | |
999 | cpumask_set_cpu(cpu, mask); | |
1000 | } | |
1001 | ||
3d28ebce | 1002 | smp_call_function_many(mask, drop_mm_ref_this_cpu, mm, 1); |
7e0563de VK |
1003 | free_cpumask_var(mask); |
1004 | } | |
1005 | #else | |
1006 | static void xen_drop_mm_ref(struct mm_struct *mm) | |
1007 | { | |
3d28ebce | 1008 | drop_mm_ref_this_cpu(mm); |
7e0563de VK |
1009 | } |
1010 | #endif | |
1011 | ||
1012 | /* | |
1013 | * While a process runs, Xen pins its pagetables, which means that the | |
1014 | * hypervisor forces it to be read-only, and it controls all updates | |
1015 | * to it. This means that all pagetable updates have to go via the | |
1016 | * hypervisor, which is moderately expensive. | |
1017 | * | |
1018 | * Since we're pulling the pagetable down, we switch to use init_mm, | |
1019 | * unpin old process pagetable and mark it all read-write, which | |
1020 | * allows further operations on it to be simple memory accesses. | |
1021 | * | |
1022 | * The only subtle point is that another CPU may be still using the | |
1023 | * pagetable because of lazy tlb flushing. This means we need need to | |
1024 | * switch all CPUs off this pagetable before we can unpin it. | |
1025 | */ | |
1026 | static void xen_exit_mmap(struct mm_struct *mm) | |
1027 | { | |
1028 | get_cpu(); /* make sure we don't move around */ | |
1029 | xen_drop_mm_ref(mm); | |
1030 | put_cpu(); | |
1031 | ||
1032 | spin_lock(&mm->page_table_lock); | |
1033 | ||
1034 | /* pgd may not be pinned in the error exit path of execve */ | |
1035 | if (xen_page_pinned(mm->pgd)) | |
1036 | xen_pgd_unpin(mm); | |
1037 | ||
1038 | spin_unlock(&mm->page_table_lock); | |
1039 | } | |
1040 | ||
1041 | static void xen_post_allocator_init(void); | |
1042 | ||
1043 | static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn) | |
1044 | { | |
1045 | struct mmuext_op op; | |
1046 | ||
1047 | op.cmd = cmd; | |
1048 | op.arg1.mfn = pfn_to_mfn(pfn); | |
1049 | if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) | |
1050 | BUG(); | |
1051 | } | |
1052 | ||
1053 | #ifdef CONFIG_X86_64 | |
1054 | static void __init xen_cleanhighmap(unsigned long vaddr, | |
1055 | unsigned long vaddr_end) | |
1056 | { | |
1057 | unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1; | |
1058 | pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr); | |
1059 | ||
1060 | /* NOTE: The loop is more greedy than the cleanup_highmap variant. | |
1061 | * We include the PMD passed in on _both_ boundaries. */ | |
1062 | for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD)); | |
1063 | pmd++, vaddr += PMD_SIZE) { | |
1064 | if (pmd_none(*pmd)) | |
1065 | continue; | |
1066 | if (vaddr < (unsigned long) _text || vaddr > kernel_end) | |
1067 | set_pmd(pmd, __pmd(0)); | |
1068 | } | |
1069 | /* In case we did something silly, we should crash in this function | |
1070 | * instead of somewhere later and be confusing. */ | |
1071 | xen_mc_flush(); | |
1072 | } | |
1073 | ||
1074 | /* | |
1075 | * Make a page range writeable and free it. | |
1076 | */ | |
1077 | static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size) | |
1078 | { | |
1079 | void *vaddr = __va(paddr); | |
1080 | void *vaddr_end = vaddr + size; | |
1081 | ||
1082 | for (; vaddr < vaddr_end; vaddr += PAGE_SIZE) | |
1083 | make_lowmem_page_readwrite(vaddr); | |
1084 | ||
1085 | memblock_free(paddr, size); | |
1086 | } | |
1087 | ||
1088 | static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin) | |
1089 | { | |
1090 | unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK; | |
1091 | ||
1092 | if (unpin) | |
1093 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa)); | |
1094 | ClearPagePinned(virt_to_page(__va(pa))); | |
1095 | xen_free_ro_pages(pa, PAGE_SIZE); | |
1096 | } | |
1097 | ||
1098 | static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin) | |
1099 | { | |
1100 | unsigned long pa; | |
1101 | pte_t *pte_tbl; | |
1102 | int i; | |
1103 | ||
1104 | if (pmd_large(*pmd)) { | |
1105 | pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK; | |
1106 | xen_free_ro_pages(pa, PMD_SIZE); | |
1107 | return; | |
1108 | } | |
1109 | ||
1110 | pte_tbl = pte_offset_kernel(pmd, 0); | |
1111 | for (i = 0; i < PTRS_PER_PTE; i++) { | |
1112 | if (pte_none(pte_tbl[i])) | |
1113 | continue; | |
1114 | pa = pte_pfn(pte_tbl[i]) << PAGE_SHIFT; | |
1115 | xen_free_ro_pages(pa, PAGE_SIZE); | |
1116 | } | |
1117 | set_pmd(pmd, __pmd(0)); | |
1118 | xen_cleanmfnmap_free_pgtbl(pte_tbl, unpin); | |
1119 | } | |
1120 | ||
1121 | static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin) | |
1122 | { | |
1123 | unsigned long pa; | |
1124 | pmd_t *pmd_tbl; | |
1125 | int i; | |
1126 | ||
1127 | if (pud_large(*pud)) { | |
1128 | pa = pud_val(*pud) & PHYSICAL_PAGE_MASK; | |
1129 | xen_free_ro_pages(pa, PUD_SIZE); | |
1130 | return; | |
1131 | } | |
1132 | ||
1133 | pmd_tbl = pmd_offset(pud, 0); | |
1134 | for (i = 0; i < PTRS_PER_PMD; i++) { | |
1135 | if (pmd_none(pmd_tbl[i])) | |
1136 | continue; | |
1137 | xen_cleanmfnmap_pmd(pmd_tbl + i, unpin); | |
1138 | } | |
1139 | set_pud(pud, __pud(0)); | |
1140 | xen_cleanmfnmap_free_pgtbl(pmd_tbl, unpin); | |
1141 | } | |
1142 | ||
1143 | static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin) | |
1144 | { | |
1145 | unsigned long pa; | |
1146 | pud_t *pud_tbl; | |
1147 | int i; | |
1148 | ||
1149 | if (p4d_large(*p4d)) { | |
1150 | pa = p4d_val(*p4d) & PHYSICAL_PAGE_MASK; | |
1151 | xen_free_ro_pages(pa, P4D_SIZE); | |
1152 | return; | |
1153 | } | |
1154 | ||
1155 | pud_tbl = pud_offset(p4d, 0); | |
1156 | for (i = 0; i < PTRS_PER_PUD; i++) { | |
1157 | if (pud_none(pud_tbl[i])) | |
1158 | continue; | |
1159 | xen_cleanmfnmap_pud(pud_tbl + i, unpin); | |
1160 | } | |
1161 | set_p4d(p4d, __p4d(0)); | |
1162 | xen_cleanmfnmap_free_pgtbl(pud_tbl, unpin); | |
1163 | } | |
1164 | ||
1165 | /* | |
1166 | * Since it is well isolated we can (and since it is perhaps large we should) | |
1167 | * also free the page tables mapping the initial P->M table. | |
1168 | */ | |
1169 | static void __init xen_cleanmfnmap(unsigned long vaddr) | |
1170 | { | |
1171 | pgd_t *pgd; | |
1172 | p4d_t *p4d; | |
7e0563de VK |
1173 | bool unpin; |
1174 | ||
1175 | unpin = (vaddr == 2 * PGDIR_SIZE); | |
1176 | vaddr &= PMD_MASK; | |
1177 | pgd = pgd_offset_k(vaddr); | |
1178 | p4d = p4d_offset(pgd, 0); | |
773dd2fc KS |
1179 | if (!p4d_none(*p4d)) |
1180 | xen_cleanmfnmap_p4d(p4d, unpin); | |
7e0563de VK |
1181 | } |
1182 | ||
1183 | static void __init xen_pagetable_p2m_free(void) | |
1184 | { | |
1185 | unsigned long size; | |
1186 | unsigned long addr; | |
1187 | ||
1188 | size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long)); | |
1189 | ||
1190 | /* No memory or already called. */ | |
1191 | if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list) | |
1192 | return; | |
1193 | ||
1194 | /* using __ka address and sticking INVALID_P2M_ENTRY! */ | |
1195 | memset((void *)xen_start_info->mfn_list, 0xff, size); | |
1196 | ||
1197 | addr = xen_start_info->mfn_list; | |
1198 | /* | |
1199 | * We could be in __ka space. | |
1200 | * We roundup to the PMD, which means that if anybody at this stage is | |
1201 | * using the __ka address of xen_start_info or | |
1202 | * xen_start_info->shared_info they are in going to crash. Fortunatly | |
1203 | * we have already revectored in xen_setup_kernel_pagetable and in | |
1204 | * xen_setup_shared_info. | |
1205 | */ | |
1206 | size = roundup(size, PMD_SIZE); | |
1207 | ||
1208 | if (addr >= __START_KERNEL_map) { | |
1209 | xen_cleanhighmap(addr, addr + size); | |
1210 | size = PAGE_ALIGN(xen_start_info->nr_pages * | |
1211 | sizeof(unsigned long)); | |
1212 | memblock_free(__pa(addr), size); | |
1213 | } else { | |
1214 | xen_cleanmfnmap(addr); | |
1215 | } | |
1216 | } | |
1217 | ||
1218 | static void __init xen_pagetable_cleanhighmap(void) | |
1219 | { | |
1220 | unsigned long size; | |
1221 | unsigned long addr; | |
1222 | ||
1223 | /* At this stage, cleanup_highmap has already cleaned __ka space | |
1224 | * from _brk_limit way up to the max_pfn_mapped (which is the end of | |
1225 | * the ramdisk). We continue on, erasing PMD entries that point to page | |
1226 | * tables - do note that they are accessible at this stage via __va. | |
0d805ee7 ZD |
1227 | * As Xen is aligning the memory end to a 4MB boundary, for good |
1228 | * measure we also round up to PMD_SIZE * 2 - which means that if | |
7e0563de VK |
1229 | * anybody is using __ka address to the initial boot-stack - and try |
1230 | * to use it - they are going to crash. The xen_start_info has been | |
1231 | * taken care of already in xen_setup_kernel_pagetable. */ | |
1232 | addr = xen_start_info->pt_base; | |
0d805ee7 | 1233 | size = xen_start_info->nr_pt_frames * PAGE_SIZE; |
7e0563de | 1234 | |
0d805ee7 | 1235 | xen_cleanhighmap(addr, roundup(addr + size, PMD_SIZE * 2)); |
7e0563de | 1236 | xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base)); |
7e0563de VK |
1237 | } |
1238 | #endif | |
1239 | ||
1240 | static void __init xen_pagetable_p2m_setup(void) | |
1241 | { | |
7e0563de VK |
1242 | xen_vmalloc_p2m_tree(); |
1243 | ||
1244 | #ifdef CONFIG_X86_64 | |
1245 | xen_pagetable_p2m_free(); | |
1246 | ||
1247 | xen_pagetable_cleanhighmap(); | |
1248 | #endif | |
1249 | /* And revector! Bye bye old array */ | |
1250 | xen_start_info->mfn_list = (unsigned long)xen_p2m_addr; | |
1251 | } | |
1252 | ||
1253 | static void __init xen_pagetable_init(void) | |
1254 | { | |
1255 | paging_init(); | |
1256 | xen_post_allocator_init(); | |
1257 | ||
1258 | xen_pagetable_p2m_setup(); | |
1259 | ||
1260 | /* Allocate and initialize top and mid mfn levels for p2m structure */ | |
1261 | xen_build_mfn_list_list(); | |
1262 | ||
1263 | /* Remap memory freed due to conflicts with E820 map */ | |
989513a7 | 1264 | xen_remap_memory(); |
7e0563de VK |
1265 | |
1266 | xen_setup_shared_info(); | |
1267 | } | |
1268 | static void xen_write_cr2(unsigned long cr2) | |
1269 | { | |
1270 | this_cpu_read(xen_vcpu)->arch.cr2 = cr2; | |
1271 | } | |
1272 | ||
1273 | static unsigned long xen_read_cr2(void) | |
1274 | { | |
1275 | return this_cpu_read(xen_vcpu)->arch.cr2; | |
1276 | } | |
1277 | ||
1278 | unsigned long xen_read_cr2_direct(void) | |
1279 | { | |
1280 | return this_cpu_read(xen_vcpu_info.arch.cr2); | |
1281 | } | |
1282 | ||
1283 | static void xen_flush_tlb(void) | |
1284 | { | |
1285 | struct mmuext_op *op; | |
1286 | struct multicall_space mcs; | |
1287 | ||
1288 | trace_xen_mmu_flush_tlb(0); | |
1289 | ||
1290 | preempt_disable(); | |
1291 | ||
1292 | mcs = xen_mc_entry(sizeof(*op)); | |
1293 | ||
1294 | op = mcs.args; | |
1295 | op->cmd = MMUEXT_TLB_FLUSH_LOCAL; | |
1296 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1297 | ||
1298 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1299 | ||
1300 | preempt_enable(); | |
1301 | } | |
1302 | ||
1303 | static void xen_flush_tlb_single(unsigned long addr) | |
1304 | { | |
1305 | struct mmuext_op *op; | |
1306 | struct multicall_space mcs; | |
1307 | ||
1308 | trace_xen_mmu_flush_tlb_single(addr); | |
1309 | ||
1310 | preempt_disable(); | |
1311 | ||
1312 | mcs = xen_mc_entry(sizeof(*op)); | |
1313 | op = mcs.args; | |
1314 | op->cmd = MMUEXT_INVLPG_LOCAL; | |
1315 | op->arg1.linear_addr = addr & PAGE_MASK; | |
1316 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1317 | ||
1318 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1319 | ||
1320 | preempt_enable(); | |
1321 | } | |
1322 | ||
1323 | static void xen_flush_tlb_others(const struct cpumask *cpus, | |
a2055abe | 1324 | const struct flush_tlb_info *info) |
7e0563de VK |
1325 | { |
1326 | struct { | |
1327 | struct mmuext_op op; | |
1328 | #ifdef CONFIG_SMP | |
1329 | DECLARE_BITMAP(mask, num_processors); | |
1330 | #else | |
1331 | DECLARE_BITMAP(mask, NR_CPUS); | |
1332 | #endif | |
1333 | } *args; | |
1334 | struct multicall_space mcs; | |
1335 | ||
a2055abe | 1336 | trace_xen_mmu_flush_tlb_others(cpus, info->mm, info->start, info->end); |
7e0563de VK |
1337 | |
1338 | if (cpumask_empty(cpus)) | |
1339 | return; /* nothing to do */ | |
1340 | ||
1341 | mcs = xen_mc_entry(sizeof(*args)); | |
1342 | args = mcs.args; | |
1343 | args->op.arg2.vcpumask = to_cpumask(args->mask); | |
1344 | ||
1345 | /* Remove us, and any offline CPUS. */ | |
1346 | cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask); | |
1347 | cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask)); | |
1348 | ||
1349 | args->op.cmd = MMUEXT_TLB_FLUSH_MULTI; | |
a2055abe AL |
1350 | if (info->end != TLB_FLUSH_ALL && |
1351 | (info->end - info->start) <= PAGE_SIZE) { | |
7e0563de | 1352 | args->op.cmd = MMUEXT_INVLPG_MULTI; |
a2055abe | 1353 | args->op.arg1.linear_addr = info->start; |
7e0563de VK |
1354 | } |
1355 | ||
1356 | MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF); | |
1357 | ||
1358 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1359 | } | |
1360 | ||
1361 | static unsigned long xen_read_cr3(void) | |
1362 | { | |
1363 | return this_cpu_read(xen_cr3); | |
1364 | } | |
1365 | ||
1366 | static void set_current_cr3(void *v) | |
1367 | { | |
1368 | this_cpu_write(xen_current_cr3, (unsigned long)v); | |
1369 | } | |
1370 | ||
1371 | static void __xen_write_cr3(bool kernel, unsigned long cr3) | |
1372 | { | |
1373 | struct mmuext_op op; | |
1374 | unsigned long mfn; | |
1375 | ||
1376 | trace_xen_mmu_write_cr3(kernel, cr3); | |
1377 | ||
1378 | if (cr3) | |
1379 | mfn = pfn_to_mfn(PFN_DOWN(cr3)); | |
1380 | else | |
1381 | mfn = 0; | |
1382 | ||
1383 | WARN_ON(mfn == 0 && kernel); | |
1384 | ||
1385 | op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR; | |
1386 | op.arg1.mfn = mfn; | |
1387 | ||
1388 | xen_extend_mmuext_op(&op); | |
1389 | ||
1390 | if (kernel) { | |
1391 | this_cpu_write(xen_cr3, cr3); | |
1392 | ||
1393 | /* Update xen_current_cr3 once the batch has actually | |
1394 | been submitted. */ | |
1395 | xen_mc_callback(set_current_cr3, (void *)cr3); | |
1396 | } | |
1397 | } | |
1398 | static void xen_write_cr3(unsigned long cr3) | |
1399 | { | |
1400 | BUG_ON(preemptible()); | |
1401 | ||
1402 | xen_mc_batch(); /* disables interrupts */ | |
1403 | ||
1404 | /* Update while interrupts are disabled, so its atomic with | |
1405 | respect to ipis */ | |
1406 | this_cpu_write(xen_cr3, cr3); | |
1407 | ||
1408 | __xen_write_cr3(true, cr3); | |
1409 | ||
1410 | #ifdef CONFIG_X86_64 | |
1411 | { | |
1412 | pgd_t *user_pgd = xen_get_user_pgd(__va(cr3)); | |
1413 | if (user_pgd) | |
1414 | __xen_write_cr3(false, __pa(user_pgd)); | |
1415 | else | |
1416 | __xen_write_cr3(false, 0); | |
1417 | } | |
1418 | #endif | |
1419 | ||
1420 | xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ | |
1421 | } | |
1422 | ||
1423 | #ifdef CONFIG_X86_64 | |
1424 | /* | |
1425 | * At the start of the day - when Xen launches a guest, it has already | |
1426 | * built pagetables for the guest. We diligently look over them | |
1427 | * in xen_setup_kernel_pagetable and graft as appropriate them in the | |
65ade2f8 KS |
1428 | * init_top_pgt and its friends. Then when we are happy we load |
1429 | * the new init_top_pgt - and continue on. | |
7e0563de VK |
1430 | * |
1431 | * The generic code starts (start_kernel) and 'init_mem_mapping' sets | |
1432 | * up the rest of the pagetables. When it has completed it loads the cr3. | |
1433 | * N.B. that baremetal would start at 'start_kernel' (and the early | |
1434 | * #PF handler would create bootstrap pagetables) - so we are running | |
1435 | * with the same assumptions as what to do when write_cr3 is executed | |
1436 | * at this point. | |
1437 | * | |
1438 | * Since there are no user-page tables at all, we have two variants | |
1439 | * of xen_write_cr3 - the early bootup (this one), and the late one | |
1440 | * (xen_write_cr3). The reason we have to do that is that in 64-bit | |
1441 | * the Linux kernel and user-space are both in ring 3 while the | |
1442 | * hypervisor is in ring 0. | |
1443 | */ | |
1444 | static void __init xen_write_cr3_init(unsigned long cr3) | |
1445 | { | |
1446 | BUG_ON(preemptible()); | |
1447 | ||
1448 | xen_mc_batch(); /* disables interrupts */ | |
1449 | ||
1450 | /* Update while interrupts are disabled, so its atomic with | |
1451 | respect to ipis */ | |
1452 | this_cpu_write(xen_cr3, cr3); | |
1453 | ||
1454 | __xen_write_cr3(true, cr3); | |
1455 | ||
1456 | xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ | |
1457 | } | |
1458 | #endif | |
1459 | ||
1460 | static int xen_pgd_alloc(struct mm_struct *mm) | |
1461 | { | |
1462 | pgd_t *pgd = mm->pgd; | |
1463 | int ret = 0; | |
1464 | ||
1465 | BUG_ON(PagePinned(virt_to_page(pgd))); | |
1466 | ||
1467 | #ifdef CONFIG_X86_64 | |
1468 | { | |
1469 | struct page *page = virt_to_page(pgd); | |
1470 | pgd_t *user_pgd; | |
1471 | ||
1472 | BUG_ON(page->private != 0); | |
1473 | ||
1474 | ret = -ENOMEM; | |
1475 | ||
1476 | user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO); | |
1477 | page->private = (unsigned long)user_pgd; | |
1478 | ||
1479 | if (user_pgd != NULL) { | |
1480 | #ifdef CONFIG_X86_VSYSCALL_EMULATION | |
1481 | user_pgd[pgd_index(VSYSCALL_ADDR)] = | |
1482 | __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE); | |
1483 | #endif | |
1484 | ret = 0; | |
1485 | } | |
1486 | ||
1487 | BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd)))); | |
1488 | } | |
1489 | #endif | |
1490 | return ret; | |
1491 | } | |
1492 | ||
1493 | static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd) | |
1494 | { | |
1495 | #ifdef CONFIG_X86_64 | |
1496 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
1497 | ||
1498 | if (user_pgd) | |
1499 | free_page((unsigned long)user_pgd); | |
1500 | #endif | |
1501 | } | |
1502 | ||
1503 | /* | |
1504 | * Init-time set_pte while constructing initial pagetables, which | |
1505 | * doesn't allow RO page table pages to be remapped RW. | |
1506 | * | |
1507 | * If there is no MFN for this PFN then this page is initially | |
1508 | * ballooned out so clear the PTE (as in decrease_reservation() in | |
1509 | * drivers/xen/balloon.c). | |
1510 | * | |
1511 | * Many of these PTE updates are done on unpinned and writable pages | |
1512 | * and doing a hypercall for these is unnecessary and expensive. At | |
1513 | * this point it is not possible to tell if a page is pinned or not, | |
1514 | * so always write the PTE directly and rely on Xen trapping and | |
1515 | * emulating any updates as necessary. | |
1516 | */ | |
1517 | __visible pte_t xen_make_pte_init(pteval_t pte) | |
1518 | { | |
1519 | #ifdef CONFIG_X86_64 | |
1520 | unsigned long pfn; | |
1521 | ||
1522 | /* | |
1523 | * Pages belonging to the initial p2m list mapped outside the default | |
1524 | * address range must be mapped read-only. This region contains the | |
1525 | * page tables for mapping the p2m list, too, and page tables MUST be | |
1526 | * mapped read-only. | |
1527 | */ | |
1528 | pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT; | |
1529 | if (xen_start_info->mfn_list < __START_KERNEL_map && | |
1530 | pfn >= xen_start_info->first_p2m_pfn && | |
1531 | pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames) | |
1532 | pte &= ~_PAGE_RW; | |
1533 | #endif | |
1534 | pte = pte_pfn_to_mfn(pte); | |
1535 | return native_make_pte(pte); | |
1536 | } | |
1537 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init); | |
1538 | ||
1539 | static void __init xen_set_pte_init(pte_t *ptep, pte_t pte) | |
1540 | { | |
1541 | #ifdef CONFIG_X86_32 | |
1542 | /* If there's an existing pte, then don't allow _PAGE_RW to be set */ | |
1543 | if (pte_mfn(pte) != INVALID_P2M_ENTRY | |
1544 | && pte_val_ma(*ptep) & _PAGE_PRESENT) | |
1545 | pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) & | |
1546 | pte_val_ma(pte)); | |
1547 | #endif | |
1548 | native_set_pte(ptep, pte); | |
1549 | } | |
1550 | ||
1551 | /* Early in boot, while setting up the initial pagetable, assume | |
1552 | everything is pinned. */ | |
1553 | static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn) | |
1554 | { | |
1555 | #ifdef CONFIG_FLATMEM | |
1556 | BUG_ON(mem_map); /* should only be used early */ | |
1557 | #endif | |
1558 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1559 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
1560 | } | |
1561 | ||
1562 | /* Used for pmd and pud */ | |
1563 | static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn) | |
1564 | { | |
1565 | #ifdef CONFIG_FLATMEM | |
1566 | BUG_ON(mem_map); /* should only be used early */ | |
1567 | #endif | |
1568 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1569 | } | |
1570 | ||
1571 | /* Early release_pte assumes that all pts are pinned, since there's | |
1572 | only init_mm and anything attached to that is pinned. */ | |
1573 | static void __init xen_release_pte_init(unsigned long pfn) | |
1574 | { | |
1575 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); | |
1576 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
1577 | } | |
1578 | ||
1579 | static void __init xen_release_pmd_init(unsigned long pfn) | |
1580 | { | |
1581 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
1582 | } | |
1583 | ||
1584 | static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn) | |
1585 | { | |
1586 | struct multicall_space mcs; | |
1587 | struct mmuext_op *op; | |
1588 | ||
1589 | mcs = __xen_mc_entry(sizeof(*op)); | |
1590 | op = mcs.args; | |
1591 | op->cmd = cmd; | |
1592 | op->arg1.mfn = pfn_to_mfn(pfn); | |
1593 | ||
1594 | MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); | |
1595 | } | |
1596 | ||
1597 | static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot) | |
1598 | { | |
1599 | struct multicall_space mcs; | |
1600 | unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT); | |
1601 | ||
1602 | mcs = __xen_mc_entry(0); | |
1603 | MULTI_update_va_mapping(mcs.mc, (unsigned long)addr, | |
1604 | pfn_pte(pfn, prot), 0); | |
1605 | } | |
1606 | ||
1607 | /* This needs to make sure the new pte page is pinned iff its being | |
1608 | attached to a pinned pagetable. */ | |
1609 | static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, | |
1610 | unsigned level) | |
1611 | { | |
1612 | bool pinned = PagePinned(virt_to_page(mm->pgd)); | |
1613 | ||
1614 | trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned); | |
1615 | ||
1616 | if (pinned) { | |
1617 | struct page *page = pfn_to_page(pfn); | |
1618 | ||
1619 | SetPagePinned(page); | |
1620 | ||
1621 | if (!PageHighMem(page)) { | |
1622 | xen_mc_batch(); | |
1623 | ||
1624 | __set_pfn_prot(pfn, PAGE_KERNEL_RO); | |
1625 | ||
1626 | if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS) | |
1627 | __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
1628 | ||
1629 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1630 | } else { | |
1631 | /* make sure there are no stray mappings of | |
1632 | this page */ | |
1633 | kmap_flush_unused(); | |
1634 | } | |
1635 | } | |
1636 | } | |
1637 | ||
1638 | static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn) | |
1639 | { | |
1640 | xen_alloc_ptpage(mm, pfn, PT_PTE); | |
1641 | } | |
1642 | ||
1643 | static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn) | |
1644 | { | |
1645 | xen_alloc_ptpage(mm, pfn, PT_PMD); | |
1646 | } | |
1647 | ||
1648 | /* This should never happen until we're OK to use struct page */ | |
1649 | static inline void xen_release_ptpage(unsigned long pfn, unsigned level) | |
1650 | { | |
1651 | struct page *page = pfn_to_page(pfn); | |
1652 | bool pinned = PagePinned(page); | |
1653 | ||
1654 | trace_xen_mmu_release_ptpage(pfn, level, pinned); | |
1655 | ||
1656 | if (pinned) { | |
1657 | if (!PageHighMem(page)) { | |
1658 | xen_mc_batch(); | |
1659 | ||
1660 | if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS) | |
1661 | __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); | |
1662 | ||
1663 | __set_pfn_prot(pfn, PAGE_KERNEL); | |
1664 | ||
1665 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1666 | } | |
1667 | ClearPagePinned(page); | |
1668 | } | |
1669 | } | |
1670 | ||
1671 | static void xen_release_pte(unsigned long pfn) | |
1672 | { | |
1673 | xen_release_ptpage(pfn, PT_PTE); | |
1674 | } | |
1675 | ||
1676 | static void xen_release_pmd(unsigned long pfn) | |
1677 | { | |
1678 | xen_release_ptpage(pfn, PT_PMD); | |
1679 | } | |
1680 | ||
773dd2fc | 1681 | #ifdef CONFIG_X86_64 |
7e0563de VK |
1682 | static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn) |
1683 | { | |
1684 | xen_alloc_ptpage(mm, pfn, PT_PUD); | |
1685 | } | |
1686 | ||
1687 | static void xen_release_pud(unsigned long pfn) | |
1688 | { | |
1689 | xen_release_ptpage(pfn, PT_PUD); | |
1690 | } | |
1691 | #endif | |
1692 | ||
1693 | void __init xen_reserve_top(void) | |
1694 | { | |
1695 | #ifdef CONFIG_X86_32 | |
1696 | unsigned long top = HYPERVISOR_VIRT_START; | |
1697 | struct xen_platform_parameters pp; | |
1698 | ||
1699 | if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0) | |
1700 | top = pp.virt_start; | |
1701 | ||
1702 | reserve_top_address(-top); | |
1703 | #endif /* CONFIG_X86_32 */ | |
1704 | } | |
1705 | ||
1706 | /* | |
1707 | * Like __va(), but returns address in the kernel mapping (which is | |
1708 | * all we have until the physical memory mapping has been set up. | |
1709 | */ | |
1710 | static void * __init __ka(phys_addr_t paddr) | |
1711 | { | |
1712 | #ifdef CONFIG_X86_64 | |
1713 | return (void *)(paddr + __START_KERNEL_map); | |
1714 | #else | |
1715 | return __va(paddr); | |
1716 | #endif | |
1717 | } | |
1718 | ||
1719 | /* Convert a machine address to physical address */ | |
1720 | static unsigned long __init m2p(phys_addr_t maddr) | |
1721 | { | |
1722 | phys_addr_t paddr; | |
1723 | ||
6f0e8bf1 | 1724 | maddr &= XEN_PTE_MFN_MASK; |
7e0563de VK |
1725 | paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT; |
1726 | ||
1727 | return paddr; | |
1728 | } | |
1729 | ||
1730 | /* Convert a machine address to kernel virtual */ | |
1731 | static void * __init m2v(phys_addr_t maddr) | |
1732 | { | |
1733 | return __ka(m2p(maddr)); | |
1734 | } | |
1735 | ||
1736 | /* Set the page permissions on an identity-mapped pages */ | |
1737 | static void __init set_page_prot_flags(void *addr, pgprot_t prot, | |
1738 | unsigned long flags) | |
1739 | { | |
1740 | unsigned long pfn = __pa(addr) >> PAGE_SHIFT; | |
1741 | pte_t pte = pfn_pte(pfn, prot); | |
1742 | ||
1743 | if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags)) | |
1744 | BUG(); | |
1745 | } | |
1746 | static void __init set_page_prot(void *addr, pgprot_t prot) | |
1747 | { | |
1748 | return set_page_prot_flags(addr, prot, UVMF_NONE); | |
1749 | } | |
1750 | #ifdef CONFIG_X86_32 | |
1751 | static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn) | |
1752 | { | |
1753 | unsigned pmdidx, pteidx; | |
1754 | unsigned ident_pte; | |
1755 | unsigned long pfn; | |
1756 | ||
1757 | level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES, | |
1758 | PAGE_SIZE); | |
1759 | ||
1760 | ident_pte = 0; | |
1761 | pfn = 0; | |
1762 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) { | |
1763 | pte_t *pte_page; | |
1764 | ||
1765 | /* Reuse or allocate a page of ptes */ | |
1766 | if (pmd_present(pmd[pmdidx])) | |
1767 | pte_page = m2v(pmd[pmdidx].pmd); | |
1768 | else { | |
1769 | /* Check for free pte pages */ | |
1770 | if (ident_pte == LEVEL1_IDENT_ENTRIES) | |
1771 | break; | |
1772 | ||
1773 | pte_page = &level1_ident_pgt[ident_pte]; | |
1774 | ident_pte += PTRS_PER_PTE; | |
1775 | ||
1776 | pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE); | |
1777 | } | |
1778 | ||
1779 | /* Install mappings */ | |
1780 | for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) { | |
1781 | pte_t pte; | |
1782 | ||
1783 | if (pfn > max_pfn_mapped) | |
1784 | max_pfn_mapped = pfn; | |
1785 | ||
1786 | if (!pte_none(pte_page[pteidx])) | |
1787 | continue; | |
1788 | ||
1789 | pte = pfn_pte(pfn, PAGE_KERNEL_EXEC); | |
1790 | pte_page[pteidx] = pte; | |
1791 | } | |
1792 | } | |
1793 | ||
1794 | for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE) | |
1795 | set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO); | |
1796 | ||
1797 | set_page_prot(pmd, PAGE_KERNEL_RO); | |
1798 | } | |
1799 | #endif | |
1800 | void __init xen_setup_machphys_mapping(void) | |
1801 | { | |
1802 | struct xen_machphys_mapping mapping; | |
1803 | ||
1804 | if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) { | |
1805 | machine_to_phys_mapping = (unsigned long *)mapping.v_start; | |
1806 | machine_to_phys_nr = mapping.max_mfn + 1; | |
1807 | } else { | |
1808 | machine_to_phys_nr = MACH2PHYS_NR_ENTRIES; | |
1809 | } | |
1810 | #ifdef CONFIG_X86_32 | |
1811 | WARN_ON((machine_to_phys_mapping + (machine_to_phys_nr - 1)) | |
1812 | < machine_to_phys_mapping); | |
1813 | #endif | |
1814 | } | |
1815 | ||
1816 | #ifdef CONFIG_X86_64 | |
1817 | static void __init convert_pfn_mfn(void *v) | |
1818 | { | |
1819 | pte_t *pte = v; | |
1820 | int i; | |
1821 | ||
1822 | /* All levels are converted the same way, so just treat them | |
1823 | as ptes. */ | |
1824 | for (i = 0; i < PTRS_PER_PTE; i++) | |
1825 | pte[i] = xen_make_pte(pte[i].pte); | |
1826 | } | |
1827 | static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end, | |
1828 | unsigned long addr) | |
1829 | { | |
1830 | if (*pt_base == PFN_DOWN(__pa(addr))) { | |
1831 | set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG); | |
1832 | clear_page((void *)addr); | |
1833 | (*pt_base)++; | |
1834 | } | |
1835 | if (*pt_end == PFN_DOWN(__pa(addr))) { | |
1836 | set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG); | |
1837 | clear_page((void *)addr); | |
1838 | (*pt_end)--; | |
1839 | } | |
1840 | } | |
1841 | /* | |
1842 | * Set up the initial kernel pagetable. | |
1843 | * | |
1844 | * We can construct this by grafting the Xen provided pagetable into | |
1845 | * head_64.S's preconstructed pagetables. We copy the Xen L2's into | |
1846 | * level2_ident_pgt, and level2_kernel_pgt. This means that only the | |
1847 | * kernel has a physical mapping to start with - but that's enough to | |
1848 | * get __va working. We need to fill in the rest of the physical | |
1849 | * mapping once some sort of allocator has been set up. | |
1850 | */ | |
1851 | void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn) | |
1852 | { | |
1853 | pud_t *l3; | |
1854 | pmd_t *l2; | |
1855 | unsigned long addr[3]; | |
1856 | unsigned long pt_base, pt_end; | |
1857 | unsigned i; | |
1858 | ||
1859 | /* max_pfn_mapped is the last pfn mapped in the initial memory | |
1860 | * mappings. Considering that on Xen after the kernel mappings we | |
1861 | * have the mappings of some pages that don't exist in pfn space, we | |
1862 | * set max_pfn_mapped to the last real pfn mapped. */ | |
1863 | if (xen_start_info->mfn_list < __START_KERNEL_map) | |
1864 | max_pfn_mapped = xen_start_info->first_p2m_pfn; | |
1865 | else | |
1866 | max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list)); | |
1867 | ||
1868 | pt_base = PFN_DOWN(__pa(xen_start_info->pt_base)); | |
1869 | pt_end = pt_base + xen_start_info->nr_pt_frames; | |
1870 | ||
1871 | /* Zap identity mapping */ | |
65ade2f8 | 1872 | init_top_pgt[0] = __pgd(0); |
7e0563de | 1873 | |
989513a7 JG |
1874 | /* Pre-constructed entries are in pfn, so convert to mfn */ |
1875 | /* L4[272] -> level3_ident_pgt */ | |
1876 | /* L4[511] -> level3_kernel_pgt */ | |
65ade2f8 | 1877 | convert_pfn_mfn(init_top_pgt); |
7e0563de | 1878 | |
989513a7 JG |
1879 | /* L3_i[0] -> level2_ident_pgt */ |
1880 | convert_pfn_mfn(level3_ident_pgt); | |
1881 | /* L3_k[510] -> level2_kernel_pgt */ | |
1882 | /* L3_k[511] -> level2_fixmap_pgt */ | |
1883 | convert_pfn_mfn(level3_kernel_pgt); | |
1884 | ||
1885 | /* L3_k[511][506] -> level1_fixmap_pgt */ | |
1886 | convert_pfn_mfn(level2_fixmap_pgt); | |
7e0563de | 1887 | |
7e0563de VK |
1888 | /* We get [511][511] and have Xen's version of level2_kernel_pgt */ |
1889 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd); | |
1890 | l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud); | |
1891 | ||
1892 | addr[0] = (unsigned long)pgd; | |
1893 | addr[1] = (unsigned long)l3; | |
1894 | addr[2] = (unsigned long)l2; | |
1895 | /* Graft it onto L4[272][0]. Note that we creating an aliasing problem: | |
1896 | * Both L4[272][0] and L4[511][510] have entries that point to the same | |
1897 | * L2 (PMD) tables. Meaning that if you modify it in __va space | |
1898 | * it will be also modified in the __ka space! (But if you just | |
1899 | * modify the PMD table to point to other PTE's or none, then you | |
1900 | * are OK - which is what cleanup_highmap does) */ | |
1901 | copy_page(level2_ident_pgt, l2); | |
1902 | /* Graft it onto L4[511][510] */ | |
1903 | copy_page(level2_kernel_pgt, l2); | |
1904 | ||
2cc42bac JB |
1905 | /* |
1906 | * Zap execute permission from the ident map. Due to the sharing of | |
1907 | * L1 entries we need to do this in the L2. | |
1908 | */ | |
1909 | if (__supported_pte_mask & _PAGE_NX) { | |
1910 | for (i = 0; i < PTRS_PER_PMD; ++i) { | |
1911 | if (pmd_none(level2_ident_pgt[i])) | |
1912 | continue; | |
1913 | level2_ident_pgt[i] = pmd_set_flags(level2_ident_pgt[i], _PAGE_NX); | |
1914 | } | |
1915 | } | |
1916 | ||
7e0563de VK |
1917 | /* Copy the initial P->M table mappings if necessary. */ |
1918 | i = pgd_index(xen_start_info->mfn_list); | |
1919 | if (i && i < pgd_index(__START_KERNEL_map)) | |
65ade2f8 | 1920 | init_top_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i]; |
7e0563de | 1921 | |
989513a7 | 1922 | /* Make pagetable pieces RO */ |
65ade2f8 | 1923 | set_page_prot(init_top_pgt, PAGE_KERNEL_RO); |
989513a7 JG |
1924 | set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO); |
1925 | set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO); | |
1926 | set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO); | |
1927 | set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO); | |
1928 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); | |
1929 | set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO); | |
1930 | set_page_prot(level1_fixmap_pgt, PAGE_KERNEL_RO); | |
1931 | ||
1932 | /* Pin down new L4 */ | |
1933 | pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE, | |
65ade2f8 | 1934 | PFN_DOWN(__pa_symbol(init_top_pgt))); |
989513a7 JG |
1935 | |
1936 | /* Unpin Xen-provided one */ | |
1937 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
7e0563de | 1938 | |
989513a7 JG |
1939 | /* |
1940 | * At this stage there can be no user pgd, and no page structure to | |
1941 | * attach it to, so make sure we just set kernel pgd. | |
1942 | */ | |
1943 | xen_mc_batch(); | |
65ade2f8 | 1944 | __xen_write_cr3(true, __pa(init_top_pgt)); |
989513a7 | 1945 | xen_mc_issue(PARAVIRT_LAZY_CPU); |
7e0563de VK |
1946 | |
1947 | /* We can't that easily rip out L3 and L2, as the Xen pagetables are | |
1948 | * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ... for | |
1949 | * the initial domain. For guests using the toolstack, they are in: | |
1950 | * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only | |
1951 | * rip out the [L4] (pgd), but for guests we shave off three pages. | |
1952 | */ | |
1953 | for (i = 0; i < ARRAY_SIZE(addr); i++) | |
1954 | check_pt_base(&pt_base, &pt_end, addr[i]); | |
1955 | ||
1956 | /* Our (by three pages) smaller Xen pagetable that we are using */ | |
1957 | xen_pt_base = PFN_PHYS(pt_base); | |
1958 | xen_pt_size = (pt_end - pt_base) * PAGE_SIZE; | |
1959 | memblock_reserve(xen_pt_base, xen_pt_size); | |
1960 | ||
1961 | /* Revector the xen_start_info */ | |
1962 | xen_start_info = (struct start_info *)__va(__pa(xen_start_info)); | |
1963 | } | |
1964 | ||
1965 | /* | |
1966 | * Read a value from a physical address. | |
1967 | */ | |
1968 | static unsigned long __init xen_read_phys_ulong(phys_addr_t addr) | |
1969 | { | |
1970 | unsigned long *vaddr; | |
1971 | unsigned long val; | |
1972 | ||
1973 | vaddr = early_memremap_ro(addr, sizeof(val)); | |
1974 | val = *vaddr; | |
1975 | early_memunmap(vaddr, sizeof(val)); | |
1976 | return val; | |
1977 | } | |
1978 | ||
1979 | /* | |
1980 | * Translate a virtual address to a physical one without relying on mapped | |
69861e0a JG |
1981 | * page tables. Don't rely on big pages being aligned in (guest) physical |
1982 | * space! | |
7e0563de VK |
1983 | */ |
1984 | static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr) | |
1985 | { | |
1986 | phys_addr_t pa; | |
1987 | pgd_t pgd; | |
1988 | pud_t pud; | |
1989 | pmd_t pmd; | |
1990 | pte_t pte; | |
1991 | ||
6c690ee1 | 1992 | pa = read_cr3_pa(); |
7e0563de VK |
1993 | pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) * |
1994 | sizeof(pgd))); | |
1995 | if (!pgd_present(pgd)) | |
1996 | return 0; | |
1997 | ||
1998 | pa = pgd_val(pgd) & PTE_PFN_MASK; | |
1999 | pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) * | |
2000 | sizeof(pud))); | |
2001 | if (!pud_present(pud)) | |
2002 | return 0; | |
69861e0a | 2003 | pa = pud_val(pud) & PTE_PFN_MASK; |
7e0563de VK |
2004 | if (pud_large(pud)) |
2005 | return pa + (vaddr & ~PUD_MASK); | |
2006 | ||
2007 | pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) * | |
2008 | sizeof(pmd))); | |
2009 | if (!pmd_present(pmd)) | |
2010 | return 0; | |
69861e0a | 2011 | pa = pmd_val(pmd) & PTE_PFN_MASK; |
7e0563de VK |
2012 | if (pmd_large(pmd)) |
2013 | return pa + (vaddr & ~PMD_MASK); | |
2014 | ||
2015 | pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) * | |
2016 | sizeof(pte))); | |
2017 | if (!pte_present(pte)) | |
2018 | return 0; | |
2019 | pa = pte_pfn(pte) << PAGE_SHIFT; | |
2020 | ||
2021 | return pa | (vaddr & ~PAGE_MASK); | |
2022 | } | |
2023 | ||
2024 | /* | |
2025 | * Find a new area for the hypervisor supplied p2m list and relocate the p2m to | |
2026 | * this area. | |
2027 | */ | |
2028 | void __init xen_relocate_p2m(void) | |
2029 | { | |
773dd2fc | 2030 | phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys; |
7e0563de | 2031 | unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end; |
773dd2fc | 2032 | int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud; |
7e0563de VK |
2033 | pte_t *pt; |
2034 | pmd_t *pmd; | |
2035 | pud_t *pud; | |
7e0563de VK |
2036 | pgd_t *pgd; |
2037 | unsigned long *new_p2m; | |
2038 | int save_pud; | |
2039 | ||
2040 | size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long)); | |
2041 | n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT; | |
2042 | n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT; | |
2043 | n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT; | |
2044 | n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT; | |
773dd2fc | 2045 | n_frames = n_pte + n_pt + n_pmd + n_pud; |
7e0563de VK |
2046 | |
2047 | new_area = xen_find_free_area(PFN_PHYS(n_frames)); | |
2048 | if (!new_area) { | |
2049 | xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n"); | |
2050 | BUG(); | |
2051 | } | |
2052 | ||
2053 | /* | |
2054 | * Setup the page tables for addressing the new p2m list. | |
2055 | * We have asked the hypervisor to map the p2m list at the user address | |
2056 | * PUD_SIZE. It may have done so, or it may have used a kernel space | |
2057 | * address depending on the Xen version. | |
2058 | * To avoid any possible virtual address collision, just use | |
2059 | * 2 * PUD_SIZE for the new area. | |
2060 | */ | |
773dd2fc | 2061 | pud_phys = new_area; |
7e0563de VK |
2062 | pmd_phys = pud_phys + PFN_PHYS(n_pud); |
2063 | pt_phys = pmd_phys + PFN_PHYS(n_pmd); | |
2064 | p2m_pfn = PFN_DOWN(pt_phys) + n_pt; | |
2065 | ||
6c690ee1 | 2066 | pgd = __va(read_cr3_pa()); |
7e0563de | 2067 | new_p2m = (unsigned long *)(2 * PGDIR_SIZE); |
7e0563de | 2068 | save_pud = n_pud; |
773dd2fc KS |
2069 | for (idx_pud = 0; idx_pud < n_pud; idx_pud++) { |
2070 | pud = early_memremap(pud_phys, PAGE_SIZE); | |
2071 | clear_page(pud); | |
2072 | for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD); | |
2073 | idx_pmd++) { | |
2074 | pmd = early_memremap(pmd_phys, PAGE_SIZE); | |
2075 | clear_page(pmd); | |
2076 | for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD); | |
2077 | idx_pt++) { | |
2078 | pt = early_memremap(pt_phys, PAGE_SIZE); | |
2079 | clear_page(pt); | |
2080 | for (idx_pte = 0; | |
2081 | idx_pte < min(n_pte, PTRS_PER_PTE); | |
2082 | idx_pte++) { | |
2083 | set_pte(pt + idx_pte, | |
2084 | pfn_pte(p2m_pfn, PAGE_KERNEL)); | |
2085 | p2m_pfn++; | |
7e0563de | 2086 | } |
773dd2fc KS |
2087 | n_pte -= PTRS_PER_PTE; |
2088 | early_memunmap(pt, PAGE_SIZE); | |
2089 | make_lowmem_page_readonly(__va(pt_phys)); | |
2090 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, | |
2091 | PFN_DOWN(pt_phys)); | |
2092 | set_pmd(pmd + idx_pt, | |
2093 | __pmd(_PAGE_TABLE | pt_phys)); | |
2094 | pt_phys += PAGE_SIZE; | |
7e0563de | 2095 | } |
773dd2fc KS |
2096 | n_pt -= PTRS_PER_PMD; |
2097 | early_memunmap(pmd, PAGE_SIZE); | |
2098 | make_lowmem_page_readonly(__va(pmd_phys)); | |
2099 | pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE, | |
2100 | PFN_DOWN(pmd_phys)); | |
2101 | set_pud(pud + idx_pmd, __pud(_PAGE_TABLE | pmd_phys)); | |
2102 | pmd_phys += PAGE_SIZE; | |
7e0563de | 2103 | } |
773dd2fc KS |
2104 | n_pmd -= PTRS_PER_PUD; |
2105 | early_memunmap(pud, PAGE_SIZE); | |
2106 | make_lowmem_page_readonly(__va(pud_phys)); | |
2107 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys)); | |
2108 | set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys)); | |
2109 | pud_phys += PAGE_SIZE; | |
2110 | } | |
7e0563de VK |
2111 | |
2112 | /* Now copy the old p2m info to the new area. */ | |
2113 | memcpy(new_p2m, xen_p2m_addr, size); | |
2114 | xen_p2m_addr = new_p2m; | |
2115 | ||
2116 | /* Release the old p2m list and set new list info. */ | |
2117 | p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list)); | |
2118 | BUG_ON(!p2m_pfn); | |
2119 | p2m_pfn_end = p2m_pfn + PFN_DOWN(size); | |
2120 | ||
2121 | if (xen_start_info->mfn_list < __START_KERNEL_map) { | |
2122 | pfn = xen_start_info->first_p2m_pfn; | |
2123 | pfn_end = xen_start_info->first_p2m_pfn + | |
2124 | xen_start_info->nr_p2m_frames; | |
2125 | set_pgd(pgd + 1, __pgd(0)); | |
2126 | } else { | |
2127 | pfn = p2m_pfn; | |
2128 | pfn_end = p2m_pfn_end; | |
2129 | } | |
2130 | ||
2131 | memblock_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn)); | |
2132 | while (pfn < pfn_end) { | |
2133 | if (pfn == p2m_pfn) { | |
2134 | pfn = p2m_pfn_end; | |
2135 | continue; | |
2136 | } | |
2137 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
2138 | pfn++; | |
2139 | } | |
2140 | ||
2141 | xen_start_info->mfn_list = (unsigned long)xen_p2m_addr; | |
2142 | xen_start_info->first_p2m_pfn = PFN_DOWN(new_area); | |
2143 | xen_start_info->nr_p2m_frames = n_frames; | |
2144 | } | |
2145 | ||
2146 | #else /* !CONFIG_X86_64 */ | |
2147 | static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD); | |
2148 | static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD); | |
2149 | ||
2150 | static void __init xen_write_cr3_init(unsigned long cr3) | |
2151 | { | |
2152 | unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir)); | |
2153 | ||
6c690ee1 | 2154 | BUG_ON(read_cr3_pa() != __pa(initial_page_table)); |
7e0563de VK |
2155 | BUG_ON(cr3 != __pa(swapper_pg_dir)); |
2156 | ||
2157 | /* | |
2158 | * We are switching to swapper_pg_dir for the first time (from | |
2159 | * initial_page_table) and therefore need to mark that page | |
2160 | * read-only and then pin it. | |
2161 | * | |
2162 | * Xen disallows sharing of kernel PMDs for PAE | |
2163 | * guests. Therefore we must copy the kernel PMD from | |
2164 | * initial_page_table into a new kernel PMD to be used in | |
2165 | * swapper_pg_dir. | |
2166 | */ | |
2167 | swapper_kernel_pmd = | |
2168 | extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE); | |
2169 | copy_page(swapper_kernel_pmd, initial_kernel_pmd); | |
2170 | swapper_pg_dir[KERNEL_PGD_BOUNDARY] = | |
2171 | __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT); | |
2172 | set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO); | |
2173 | ||
2174 | set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO); | |
2175 | xen_write_cr3(cr3); | |
2176 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn); | |
2177 | ||
2178 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, | |
2179 | PFN_DOWN(__pa(initial_page_table))); | |
2180 | set_page_prot(initial_page_table, PAGE_KERNEL); | |
2181 | set_page_prot(initial_kernel_pmd, PAGE_KERNEL); | |
2182 | ||
2183 | pv_mmu_ops.write_cr3 = &xen_write_cr3; | |
2184 | } | |
2185 | ||
2186 | /* | |
2187 | * For 32 bit domains xen_start_info->pt_base is the pgd address which might be | |
2188 | * not the first page table in the page table pool. | |
2189 | * Iterate through the initial page tables to find the real page table base. | |
2190 | */ | |
51ae2538 | 2191 | static phys_addr_t __init xen_find_pt_base(pmd_t *pmd) |
7e0563de VK |
2192 | { |
2193 | phys_addr_t pt_base, paddr; | |
2194 | unsigned pmdidx; | |
2195 | ||
2196 | pt_base = min(__pa(xen_start_info->pt_base), __pa(pmd)); | |
2197 | ||
2198 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) | |
2199 | if (pmd_present(pmd[pmdidx]) && !pmd_large(pmd[pmdidx])) { | |
2200 | paddr = m2p(pmd[pmdidx].pmd); | |
2201 | pt_base = min(pt_base, paddr); | |
2202 | } | |
2203 | ||
2204 | return pt_base; | |
2205 | } | |
2206 | ||
2207 | void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn) | |
2208 | { | |
2209 | pmd_t *kernel_pmd; | |
2210 | ||
2211 | kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd); | |
2212 | ||
2213 | xen_pt_base = xen_find_pt_base(kernel_pmd); | |
2214 | xen_pt_size = xen_start_info->nr_pt_frames * PAGE_SIZE; | |
2215 | ||
2216 | initial_kernel_pmd = | |
2217 | extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE); | |
2218 | ||
2219 | max_pfn_mapped = PFN_DOWN(xen_pt_base + xen_pt_size + 512 * 1024); | |
2220 | ||
2221 | copy_page(initial_kernel_pmd, kernel_pmd); | |
2222 | ||
2223 | xen_map_identity_early(initial_kernel_pmd, max_pfn); | |
2224 | ||
2225 | copy_page(initial_page_table, pgd); | |
2226 | initial_page_table[KERNEL_PGD_BOUNDARY] = | |
2227 | __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT); | |
2228 | ||
2229 | set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO); | |
2230 | set_page_prot(initial_page_table, PAGE_KERNEL_RO); | |
2231 | set_page_prot(empty_zero_page, PAGE_KERNEL_RO); | |
2232 | ||
2233 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
2234 | ||
2235 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, | |
2236 | PFN_DOWN(__pa(initial_page_table))); | |
2237 | xen_write_cr3(__pa(initial_page_table)); | |
2238 | ||
2239 | memblock_reserve(xen_pt_base, xen_pt_size); | |
2240 | } | |
2241 | #endif /* CONFIG_X86_64 */ | |
2242 | ||
2243 | void __init xen_reserve_special_pages(void) | |
2244 | { | |
2245 | phys_addr_t paddr; | |
2246 | ||
2247 | memblock_reserve(__pa(xen_start_info), PAGE_SIZE); | |
2248 | if (xen_start_info->store_mfn) { | |
2249 | paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn)); | |
2250 | memblock_reserve(paddr, PAGE_SIZE); | |
2251 | } | |
2252 | if (!xen_initial_domain()) { | |
2253 | paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn)); | |
2254 | memblock_reserve(paddr, PAGE_SIZE); | |
2255 | } | |
2256 | } | |
2257 | ||
2258 | void __init xen_pt_check_e820(void) | |
2259 | { | |
2260 | if (xen_is_e820_reserved(xen_pt_base, xen_pt_size)) { | |
2261 | xen_raw_console_write("Xen hypervisor allocated page table memory conflicts with E820 map\n"); | |
2262 | BUG(); | |
2263 | } | |
2264 | } | |
2265 | ||
2266 | static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss; | |
2267 | ||
2268 | static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot) | |
2269 | { | |
2270 | pte_t pte; | |
2271 | ||
2272 | phys >>= PAGE_SHIFT; | |
2273 | ||
2274 | switch (idx) { | |
2275 | case FIX_BTMAP_END ... FIX_BTMAP_BEGIN: | |
7e0563de VK |
2276 | #ifdef CONFIG_X86_32 |
2277 | case FIX_WP_TEST: | |
2278 | # ifdef CONFIG_HIGHMEM | |
2279 | case FIX_KMAP_BEGIN ... FIX_KMAP_END: | |
2280 | # endif | |
2281 | #elif defined(CONFIG_X86_VSYSCALL_EMULATION) | |
2282 | case VSYSCALL_PAGE: | |
2283 | #endif | |
2284 | case FIX_TEXT_POKE0: | |
2285 | case FIX_TEXT_POKE1: | |
7e0563de VK |
2286 | /* All local page mappings */ |
2287 | pte = pfn_pte(phys, prot); | |
2288 | break; | |
2289 | ||
2290 | #ifdef CONFIG_X86_LOCAL_APIC | |
2291 | case FIX_APIC_BASE: /* maps dummy local APIC */ | |
2292 | pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL); | |
2293 | break; | |
2294 | #endif | |
2295 | ||
2296 | #ifdef CONFIG_X86_IO_APIC | |
2297 | case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END: | |
2298 | /* | |
2299 | * We just don't map the IO APIC - all access is via | |
2300 | * hypercalls. Keep the address in the pte for reference. | |
2301 | */ | |
2302 | pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL); | |
2303 | break; | |
2304 | #endif | |
2305 | ||
2306 | case FIX_PARAVIRT_BOOTMAP: | |
2307 | /* This is an MFN, but it isn't an IO mapping from the | |
2308 | IO domain */ | |
2309 | pte = mfn_pte(phys, prot); | |
2310 | break; | |
2311 | ||
2312 | default: | |
2313 | /* By default, set_fixmap is used for hardware mappings */ | |
2314 | pte = mfn_pte(phys, prot); | |
2315 | break; | |
2316 | } | |
2317 | ||
2318 | __native_set_fixmap(idx, pte); | |
2319 | ||
2320 | #ifdef CONFIG_X86_VSYSCALL_EMULATION | |
2321 | /* Replicate changes to map the vsyscall page into the user | |
2322 | pagetable vsyscall mapping. */ | |
2323 | if (idx == VSYSCALL_PAGE) { | |
2324 | unsigned long vaddr = __fix_to_virt(idx); | |
2325 | set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte); | |
2326 | } | |
2327 | #endif | |
2328 | } | |
2329 | ||
2330 | static void __init xen_post_allocator_init(void) | |
2331 | { | |
7e0563de VK |
2332 | pv_mmu_ops.set_pte = xen_set_pte; |
2333 | pv_mmu_ops.set_pmd = xen_set_pmd; | |
2334 | pv_mmu_ops.set_pud = xen_set_pud; | |
773dd2fc | 2335 | #ifdef CONFIG_X86_64 |
7e0563de VK |
2336 | pv_mmu_ops.set_p4d = xen_set_p4d; |
2337 | #endif | |
2338 | ||
2339 | /* This will work as long as patching hasn't happened yet | |
2340 | (which it hasn't) */ | |
2341 | pv_mmu_ops.alloc_pte = xen_alloc_pte; | |
2342 | pv_mmu_ops.alloc_pmd = xen_alloc_pmd; | |
2343 | pv_mmu_ops.release_pte = xen_release_pte; | |
2344 | pv_mmu_ops.release_pmd = xen_release_pmd; | |
773dd2fc | 2345 | #ifdef CONFIG_X86_64 |
7e0563de VK |
2346 | pv_mmu_ops.alloc_pud = xen_alloc_pud; |
2347 | pv_mmu_ops.release_pud = xen_release_pud; | |
2348 | #endif | |
2349 | pv_mmu_ops.make_pte = PV_CALLEE_SAVE(xen_make_pte); | |
2350 | ||
2351 | #ifdef CONFIG_X86_64 | |
2352 | pv_mmu_ops.write_cr3 = &xen_write_cr3; | |
2353 | SetPagePinned(virt_to_page(level3_user_vsyscall)); | |
2354 | #endif | |
2355 | xen_mark_init_mm_pinned(); | |
2356 | } | |
2357 | ||
2358 | static void xen_leave_lazy_mmu(void) | |
2359 | { | |
2360 | preempt_disable(); | |
2361 | xen_mc_flush(); | |
2362 | paravirt_leave_lazy_mmu(); | |
2363 | preempt_enable(); | |
2364 | } | |
2365 | ||
2366 | static const struct pv_mmu_ops xen_mmu_ops __initconst = { | |
2367 | .read_cr2 = xen_read_cr2, | |
2368 | .write_cr2 = xen_write_cr2, | |
2369 | ||
2370 | .read_cr3 = xen_read_cr3, | |
2371 | .write_cr3 = xen_write_cr3_init, | |
2372 | ||
2373 | .flush_tlb_user = xen_flush_tlb, | |
2374 | .flush_tlb_kernel = xen_flush_tlb, | |
2375 | .flush_tlb_single = xen_flush_tlb_single, | |
2376 | .flush_tlb_others = xen_flush_tlb_others, | |
2377 | ||
7e0563de VK |
2378 | .pgd_alloc = xen_pgd_alloc, |
2379 | .pgd_free = xen_pgd_free, | |
2380 | ||
2381 | .alloc_pte = xen_alloc_pte_init, | |
2382 | .release_pte = xen_release_pte_init, | |
2383 | .alloc_pmd = xen_alloc_pmd_init, | |
2384 | .release_pmd = xen_release_pmd_init, | |
2385 | ||
2386 | .set_pte = xen_set_pte_init, | |
2387 | .set_pte_at = xen_set_pte_at, | |
2388 | .set_pmd = xen_set_pmd_hyper, | |
2389 | ||
2390 | .ptep_modify_prot_start = __ptep_modify_prot_start, | |
2391 | .ptep_modify_prot_commit = __ptep_modify_prot_commit, | |
2392 | ||
2393 | .pte_val = PV_CALLEE_SAVE(xen_pte_val), | |
2394 | .pgd_val = PV_CALLEE_SAVE(xen_pgd_val), | |
2395 | ||
2396 | .make_pte = PV_CALLEE_SAVE(xen_make_pte_init), | |
2397 | .make_pgd = PV_CALLEE_SAVE(xen_make_pgd), | |
2398 | ||
2399 | #ifdef CONFIG_X86_PAE | |
2400 | .set_pte_atomic = xen_set_pte_atomic, | |
2401 | .pte_clear = xen_pte_clear, | |
2402 | .pmd_clear = xen_pmd_clear, | |
2403 | #endif /* CONFIG_X86_PAE */ | |
2404 | .set_pud = xen_set_pud_hyper, | |
2405 | ||
2406 | .make_pmd = PV_CALLEE_SAVE(xen_make_pmd), | |
2407 | .pmd_val = PV_CALLEE_SAVE(xen_pmd_val), | |
2408 | ||
773dd2fc | 2409 | #ifdef CONFIG_X86_64 |
7e0563de VK |
2410 | .pud_val = PV_CALLEE_SAVE(xen_pud_val), |
2411 | .make_pud = PV_CALLEE_SAVE(xen_make_pud), | |
2412 | .set_p4d = xen_set_p4d_hyper, | |
2413 | ||
2414 | .alloc_pud = xen_alloc_pmd_init, | |
2415 | .release_pud = xen_release_pmd_init, | |
773dd2fc | 2416 | #endif /* CONFIG_X86_64 */ |
7e0563de VK |
2417 | |
2418 | .activate_mm = xen_activate_mm, | |
2419 | .dup_mmap = xen_dup_mmap, | |
2420 | .exit_mmap = xen_exit_mmap, | |
2421 | ||
2422 | .lazy_mode = { | |
2423 | .enter = paravirt_enter_lazy_mmu, | |
2424 | .leave = xen_leave_lazy_mmu, | |
2425 | .flush = paravirt_flush_lazy_mmu, | |
2426 | }, | |
2427 | ||
2428 | .set_fixmap = xen_set_fixmap, | |
2429 | }; | |
2430 | ||
2431 | void __init xen_init_mmu_ops(void) | |
2432 | { | |
2433 | x86_init.paging.pagetable_init = xen_pagetable_init; | |
2434 | ||
7e0563de VK |
2435 | pv_mmu_ops = xen_mmu_ops; |
2436 | ||
2437 | memset(dummy_mapping, 0xff, PAGE_SIZE); | |
2438 | } | |
2439 | ||
2440 | /* Protected by xen_reservation_lock. */ | |
2441 | #define MAX_CONTIG_ORDER 9 /* 2MB */ | |
2442 | static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER]; | |
2443 | ||
2444 | #define VOID_PTE (mfn_pte(0, __pgprot(0))) | |
2445 | static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order, | |
2446 | unsigned long *in_frames, | |
2447 | unsigned long *out_frames) | |
2448 | { | |
2449 | int i; | |
2450 | struct multicall_space mcs; | |
2451 | ||
2452 | xen_mc_batch(); | |
2453 | for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) { | |
2454 | mcs = __xen_mc_entry(0); | |
2455 | ||
2456 | if (in_frames) | |
2457 | in_frames[i] = virt_to_mfn(vaddr); | |
2458 | ||
2459 | MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0); | |
2460 | __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY); | |
2461 | ||
2462 | if (out_frames) | |
2463 | out_frames[i] = virt_to_pfn(vaddr); | |
2464 | } | |
2465 | xen_mc_issue(0); | |
2466 | } | |
2467 | ||
2468 | /* | |
2469 | * Update the pfn-to-mfn mappings for a virtual address range, either to | |
2470 | * point to an array of mfns, or contiguously from a single starting | |
2471 | * mfn. | |
2472 | */ | |
2473 | static void xen_remap_exchanged_ptes(unsigned long vaddr, int order, | |
2474 | unsigned long *mfns, | |
2475 | unsigned long first_mfn) | |
2476 | { | |
2477 | unsigned i, limit; | |
2478 | unsigned long mfn; | |
2479 | ||
2480 | xen_mc_batch(); | |
2481 | ||
2482 | limit = 1u << order; | |
2483 | for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) { | |
2484 | struct multicall_space mcs; | |
2485 | unsigned flags; | |
2486 | ||
2487 | mcs = __xen_mc_entry(0); | |
2488 | if (mfns) | |
2489 | mfn = mfns[i]; | |
2490 | else | |
2491 | mfn = first_mfn + i; | |
2492 | ||
2493 | if (i < (limit - 1)) | |
2494 | flags = 0; | |
2495 | else { | |
2496 | if (order == 0) | |
2497 | flags = UVMF_INVLPG | UVMF_ALL; | |
2498 | else | |
2499 | flags = UVMF_TLB_FLUSH | UVMF_ALL; | |
2500 | } | |
2501 | ||
2502 | MULTI_update_va_mapping(mcs.mc, vaddr, | |
2503 | mfn_pte(mfn, PAGE_KERNEL), flags); | |
2504 | ||
2505 | set_phys_to_machine(virt_to_pfn(vaddr), mfn); | |
2506 | } | |
2507 | ||
2508 | xen_mc_issue(0); | |
2509 | } | |
2510 | ||
2511 | /* | |
2512 | * Perform the hypercall to exchange a region of our pfns to point to | |
2513 | * memory with the required contiguous alignment. Takes the pfns as | |
2514 | * input, and populates mfns as output. | |
2515 | * | |
2516 | * Returns a success code indicating whether the hypervisor was able to | |
2517 | * satisfy the request or not. | |
2518 | */ | |
2519 | static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in, | |
2520 | unsigned long *pfns_in, | |
2521 | unsigned long extents_out, | |
2522 | unsigned int order_out, | |
2523 | unsigned long *mfns_out, | |
2524 | unsigned int address_bits) | |
2525 | { | |
2526 | long rc; | |
2527 | int success; | |
2528 | ||
2529 | struct xen_memory_exchange exchange = { | |
2530 | .in = { | |
2531 | .nr_extents = extents_in, | |
2532 | .extent_order = order_in, | |
2533 | .extent_start = pfns_in, | |
2534 | .domid = DOMID_SELF | |
2535 | }, | |
2536 | .out = { | |
2537 | .nr_extents = extents_out, | |
2538 | .extent_order = order_out, | |
2539 | .extent_start = mfns_out, | |
2540 | .address_bits = address_bits, | |
2541 | .domid = DOMID_SELF | |
2542 | } | |
2543 | }; | |
2544 | ||
2545 | BUG_ON(extents_in << order_in != extents_out << order_out); | |
2546 | ||
2547 | rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange); | |
2548 | success = (exchange.nr_exchanged == extents_in); | |
2549 | ||
2550 | BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0))); | |
2551 | BUG_ON(success && (rc != 0)); | |
2552 | ||
2553 | return success; | |
2554 | } | |
2555 | ||
2556 | int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order, | |
2557 | unsigned int address_bits, | |
2558 | dma_addr_t *dma_handle) | |
2559 | { | |
2560 | unsigned long *in_frames = discontig_frames, out_frame; | |
2561 | unsigned long flags; | |
2562 | int success; | |
2563 | unsigned long vstart = (unsigned long)phys_to_virt(pstart); | |
2564 | ||
2565 | /* | |
2566 | * Currently an auto-translated guest will not perform I/O, nor will | |
2567 | * it require PAE page directories below 4GB. Therefore any calls to | |
2568 | * this function are redundant and can be ignored. | |
2569 | */ | |
2570 | ||
7e0563de VK |
2571 | if (unlikely(order > MAX_CONTIG_ORDER)) |
2572 | return -ENOMEM; | |
2573 | ||
2574 | memset((void *) vstart, 0, PAGE_SIZE << order); | |
2575 | ||
2576 | spin_lock_irqsave(&xen_reservation_lock, flags); | |
2577 | ||
2578 | /* 1. Zap current PTEs, remembering MFNs. */ | |
2579 | xen_zap_pfn_range(vstart, order, in_frames, NULL); | |
2580 | ||
2581 | /* 2. Get a new contiguous memory extent. */ | |
2582 | out_frame = virt_to_pfn(vstart); | |
2583 | success = xen_exchange_memory(1UL << order, 0, in_frames, | |
2584 | 1, order, &out_frame, | |
2585 | address_bits); | |
2586 | ||
2587 | /* 3. Map the new extent in place of old pages. */ | |
2588 | if (success) | |
2589 | xen_remap_exchanged_ptes(vstart, order, NULL, out_frame); | |
2590 | else | |
2591 | xen_remap_exchanged_ptes(vstart, order, in_frames, 0); | |
2592 | ||
2593 | spin_unlock_irqrestore(&xen_reservation_lock, flags); | |
2594 | ||
2595 | *dma_handle = virt_to_machine(vstart).maddr; | |
2596 | return success ? 0 : -ENOMEM; | |
2597 | } | |
2598 | EXPORT_SYMBOL_GPL(xen_create_contiguous_region); | |
2599 | ||
2600 | void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order) | |
2601 | { | |
2602 | unsigned long *out_frames = discontig_frames, in_frame; | |
2603 | unsigned long flags; | |
2604 | int success; | |
2605 | unsigned long vstart; | |
2606 | ||
7e0563de VK |
2607 | if (unlikely(order > MAX_CONTIG_ORDER)) |
2608 | return; | |
2609 | ||
2610 | vstart = (unsigned long)phys_to_virt(pstart); | |
2611 | memset((void *) vstart, 0, PAGE_SIZE << order); | |
2612 | ||
2613 | spin_lock_irqsave(&xen_reservation_lock, flags); | |
2614 | ||
2615 | /* 1. Find start MFN of contiguous extent. */ | |
2616 | in_frame = virt_to_mfn(vstart); | |
2617 | ||
2618 | /* 2. Zap current PTEs. */ | |
2619 | xen_zap_pfn_range(vstart, order, NULL, out_frames); | |
2620 | ||
2621 | /* 3. Do the exchange for non-contiguous MFNs. */ | |
2622 | success = xen_exchange_memory(1, order, &in_frame, 1UL << order, | |
2623 | 0, out_frames, 0); | |
2624 | ||
2625 | /* 4. Map new pages in place of old pages. */ | |
2626 | if (success) | |
2627 | xen_remap_exchanged_ptes(vstart, order, out_frames, 0); | |
2628 | else | |
2629 | xen_remap_exchanged_ptes(vstart, order, NULL, in_frame); | |
2630 | ||
2631 | spin_unlock_irqrestore(&xen_reservation_lock, flags); | |
2632 | } | |
2633 | EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region); | |
29985b09 JG |
2634 | |
2635 | #ifdef CONFIG_KEXEC_CORE | |
2636 | phys_addr_t paddr_vmcoreinfo_note(void) | |
2637 | { | |
2638 | if (xen_pv_domain()) | |
203e9e41 | 2639 | return virt_to_machine(vmcoreinfo_note).maddr; |
29985b09 | 2640 | else |
203e9e41 | 2641 | return __pa(vmcoreinfo_note); |
29985b09 JG |
2642 | } |
2643 | #endif /* CONFIG_KEXEC_CORE */ |