]>
Commit | Line | Data |
---|---|---|
901d209a JG |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | ||
7e0563de VK |
3 | /* |
4 | * Xen mmu operations | |
5 | * | |
6 | * This file contains the various mmu fetch and update operations. | |
7 | * The most important job they must perform is the mapping between the | |
8 | * domain's pfn and the overall machine mfns. | |
9 | * | |
10 | * Xen allows guests to directly update the pagetable, in a controlled | |
11 | * fashion. In other words, the guest modifies the same pagetable | |
12 | * that the CPU actually uses, which eliminates the overhead of having | |
13 | * a separate shadow pagetable. | |
14 | * | |
15 | * In order to allow this, it falls on the guest domain to map its | |
16 | * notion of a "physical" pfn - which is just a domain-local linear | |
17 | * address - into a real "machine address" which the CPU's MMU can | |
18 | * use. | |
19 | * | |
20 | * A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be | |
21 | * inserted directly into the pagetable. When creating a new | |
22 | * pte/pmd/pgd, it converts the passed pfn into an mfn. Conversely, | |
23 | * when reading the content back with __(pgd|pmd|pte)_val, it converts | |
24 | * the mfn back into a pfn. | |
25 | * | |
26 | * The other constraint is that all pages which make up a pagetable | |
27 | * must be mapped read-only in the guest. This prevents uncontrolled | |
28 | * guest updates to the pagetable. Xen strictly enforces this, and | |
29 | * will disallow any pagetable update which will end up mapping a | |
30 | * pagetable page RW, and will disallow using any writable page as a | |
31 | * pagetable. | |
32 | * | |
33 | * Naively, when loading %cr3 with the base of a new pagetable, Xen | |
34 | * would need to validate the whole pagetable before going on. | |
35 | * Naturally, this is quite slow. The solution is to "pin" a | |
36 | * pagetable, which enforces all the constraints on the pagetable even | |
37 | * when it is not actively in use. This menas that Xen can be assured | |
38 | * that it is still valid when you do load it into %cr3, and doesn't | |
39 | * need to revalidate it. | |
40 | * | |
41 | * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007 | |
42 | */ | |
43 | #include <linux/sched/mm.h> | |
44 | #include <linux/highmem.h> | |
45 | #include <linux/debugfs.h> | |
46 | #include <linux/bug.h> | |
47 | #include <linux/vmalloc.h> | |
48 | #include <linux/export.h> | |
49 | #include <linux/init.h> | |
50 | #include <linux/gfp.h> | |
51 | #include <linux/memblock.h> | |
52 | #include <linux/seq_file.h> | |
53 | #include <linux/crash_dump.h> | |
29985b09 JG |
54 | #ifdef CONFIG_KEXEC_CORE |
55 | #include <linux/kexec.h> | |
56 | #endif | |
7e0563de VK |
57 | |
58 | #include <trace/events/xen.h> | |
59 | ||
60 | #include <asm/pgtable.h> | |
61 | #include <asm/tlbflush.h> | |
62 | #include <asm/fixmap.h> | |
63 | #include <asm/mmu_context.h> | |
64 | #include <asm/setup.h> | |
65 | #include <asm/paravirt.h> | |
66 | #include <asm/e820/api.h> | |
67 | #include <asm/linkage.h> | |
68 | #include <asm/page.h> | |
69 | #include <asm/init.h> | |
70 | #include <asm/pat.h> | |
71 | #include <asm/smp.h> | |
48a8b97c | 72 | #include <asm/tlb.h> |
7e0563de VK |
73 | |
74 | #include <asm/xen/hypercall.h> | |
75 | #include <asm/xen/hypervisor.h> | |
76 | ||
77 | #include <xen/xen.h> | |
78 | #include <xen/page.h> | |
79 | #include <xen/interface/xen.h> | |
80 | #include <xen/interface/hvm/hvm_op.h> | |
81 | #include <xen/interface/version.h> | |
82 | #include <xen/interface/memory.h> | |
83 | #include <xen/hvc-console.h> | |
84 | ||
85 | #include "multicalls.h" | |
86 | #include "mmu.h" | |
87 | #include "debugfs.h" | |
88 | ||
89 | #ifdef CONFIG_X86_32 | |
90 | /* | |
91 | * Identity map, in addition to plain kernel map. This needs to be | |
92 | * large enough to allocate page table pages to allocate the rest. | |
93 | * Each page can map 2MB. | |
94 | */ | |
95 | #define LEVEL1_IDENT_ENTRIES (PTRS_PER_PTE * 4) | |
96 | static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES); | |
97 | #endif | |
98 | #ifdef CONFIG_X86_64 | |
99 | /* l3 pud for userspace vsyscall mapping */ | |
100 | static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss; | |
101 | #endif /* CONFIG_X86_64 */ | |
102 | ||
f030aade JG |
103 | /* |
104 | * Protects atomic reservation decrease/increase against concurrent increases. | |
105 | * Also protects non-atomic updates of current_pages and balloon lists. | |
106 | */ | |
4f2d7af7 | 107 | static DEFINE_SPINLOCK(xen_reservation_lock); |
f030aade | 108 | |
7e0563de VK |
109 | /* |
110 | * Note about cr3 (pagetable base) values: | |
111 | * | |
112 | * xen_cr3 contains the current logical cr3 value; it contains the | |
113 | * last set cr3. This may not be the current effective cr3, because | |
114 | * its update may be being lazily deferred. However, a vcpu looking | |
115 | * at its own cr3 can use this value knowing that it everything will | |
116 | * be self-consistent. | |
117 | * | |
118 | * xen_current_cr3 contains the actual vcpu cr3; it is set once the | |
119 | * hypercall to set the vcpu cr3 is complete (so it may be a little | |
120 | * out of date, but it will never be set early). If one vcpu is | |
121 | * looking at another vcpu's cr3 value, it should use this variable. | |
122 | */ | |
123 | DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */ | |
124 | DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */ | |
125 | ||
126 | static phys_addr_t xen_pt_base, xen_pt_size __initdata; | |
127 | ||
6f84f8d1 PT |
128 | static DEFINE_STATIC_KEY_FALSE(xen_struct_pages_ready); |
129 | ||
7e0563de VK |
130 | /* |
131 | * Just beyond the highest usermode address. STACK_TOP_MAX has a | |
132 | * redzone above it, so round it up to a PGD boundary. | |
133 | */ | |
134 | #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK) | |
135 | ||
136 | void make_lowmem_page_readonly(void *vaddr) | |
137 | { | |
138 | pte_t *pte, ptev; | |
139 | unsigned long address = (unsigned long)vaddr; | |
140 | unsigned int level; | |
141 | ||
142 | pte = lookup_address(address, &level); | |
143 | if (pte == NULL) | |
144 | return; /* vaddr missing */ | |
145 | ||
146 | ptev = pte_wrprotect(*pte); | |
147 | ||
148 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
149 | BUG(); | |
150 | } | |
151 | ||
152 | void make_lowmem_page_readwrite(void *vaddr) | |
153 | { | |
154 | pte_t *pte, ptev; | |
155 | unsigned long address = (unsigned long)vaddr; | |
156 | unsigned int level; | |
157 | ||
158 | pte = lookup_address(address, &level); | |
159 | if (pte == NULL) | |
160 | return; /* vaddr missing */ | |
161 | ||
162 | ptev = pte_mkwrite(*pte); | |
163 | ||
164 | if (HYPERVISOR_update_va_mapping(address, ptev, 0)) | |
165 | BUG(); | |
166 | } | |
167 | ||
168 | ||
6f84f8d1 PT |
169 | /* |
170 | * During early boot all page table pages are pinned, but we do not have struct | |
171 | * pages, so return true until struct pages are ready. | |
172 | */ | |
7e0563de VK |
173 | static bool xen_page_pinned(void *ptr) |
174 | { | |
6f84f8d1 PT |
175 | if (static_branch_likely(&xen_struct_pages_ready)) { |
176 | struct page *page = virt_to_page(ptr); | |
7e0563de | 177 | |
6f84f8d1 PT |
178 | return PagePinned(page); |
179 | } | |
180 | return true; | |
7e0563de VK |
181 | } |
182 | ||
7e0563de VK |
183 | static void xen_extend_mmu_update(const struct mmu_update *update) |
184 | { | |
185 | struct multicall_space mcs; | |
186 | struct mmu_update *u; | |
187 | ||
188 | mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u)); | |
189 | ||
190 | if (mcs.mc != NULL) { | |
191 | mcs.mc->args[1]++; | |
192 | } else { | |
193 | mcs = __xen_mc_entry(sizeof(*u)); | |
194 | MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); | |
195 | } | |
196 | ||
197 | u = mcs.args; | |
198 | *u = *update; | |
199 | } | |
200 | ||
201 | static void xen_extend_mmuext_op(const struct mmuext_op *op) | |
202 | { | |
203 | struct multicall_space mcs; | |
204 | struct mmuext_op *u; | |
205 | ||
206 | mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u)); | |
207 | ||
208 | if (mcs.mc != NULL) { | |
209 | mcs.mc->args[1]++; | |
210 | } else { | |
211 | mcs = __xen_mc_entry(sizeof(*u)); | |
212 | MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); | |
213 | } | |
214 | ||
215 | u = mcs.args; | |
216 | *u = *op; | |
217 | } | |
218 | ||
219 | static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val) | |
220 | { | |
221 | struct mmu_update u; | |
222 | ||
223 | preempt_disable(); | |
224 | ||
225 | xen_mc_batch(); | |
226 | ||
227 | /* ptr may be ioremapped for 64-bit pagetable setup */ | |
228 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
229 | u.val = pmd_val_ma(val); | |
230 | xen_extend_mmu_update(&u); | |
231 | ||
232 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
233 | ||
234 | preempt_enable(); | |
235 | } | |
236 | ||
237 | static void xen_set_pmd(pmd_t *ptr, pmd_t val) | |
238 | { | |
239 | trace_xen_mmu_set_pmd(ptr, val); | |
240 | ||
241 | /* If page is not pinned, we can just update the entry | |
242 | directly */ | |
243 | if (!xen_page_pinned(ptr)) { | |
244 | *ptr = val; | |
245 | return; | |
246 | } | |
247 | ||
248 | xen_set_pmd_hyper(ptr, val); | |
249 | } | |
250 | ||
251 | /* | |
252 | * Associate a virtual page frame with a given physical page frame | |
253 | * and protection flags for that frame. | |
254 | */ | |
255 | void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags) | |
256 | { | |
257 | set_pte_vaddr(vaddr, mfn_pte(mfn, flags)); | |
258 | } | |
259 | ||
260 | static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval) | |
261 | { | |
262 | struct mmu_update u; | |
263 | ||
264 | if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU) | |
265 | return false; | |
266 | ||
267 | xen_mc_batch(); | |
268 | ||
269 | u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE; | |
270 | u.val = pte_val_ma(pteval); | |
271 | xen_extend_mmu_update(&u); | |
272 | ||
273 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
274 | ||
275 | return true; | |
276 | } | |
277 | ||
278 | static inline void __xen_set_pte(pte_t *ptep, pte_t pteval) | |
279 | { | |
280 | if (!xen_batched_set_pte(ptep, pteval)) { | |
281 | /* | |
282 | * Could call native_set_pte() here and trap and | |
283 | * emulate the PTE write but with 32-bit guests this | |
284 | * needs two traps (one for each of the two 32-bit | |
285 | * words in the PTE) so do one hypercall directly | |
286 | * instead. | |
287 | */ | |
288 | struct mmu_update u; | |
289 | ||
290 | u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE; | |
291 | u.val = pte_val_ma(pteval); | |
292 | HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF); | |
293 | } | |
294 | } | |
295 | ||
296 | static void xen_set_pte(pte_t *ptep, pte_t pteval) | |
297 | { | |
298 | trace_xen_mmu_set_pte(ptep, pteval); | |
299 | __xen_set_pte(ptep, pteval); | |
300 | } | |
301 | ||
302 | static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr, | |
303 | pte_t *ptep, pte_t pteval) | |
304 | { | |
305 | trace_xen_mmu_set_pte_at(mm, addr, ptep, pteval); | |
306 | __xen_set_pte(ptep, pteval); | |
307 | } | |
308 | ||
309 | pte_t xen_ptep_modify_prot_start(struct mm_struct *mm, | |
310 | unsigned long addr, pte_t *ptep) | |
311 | { | |
312 | /* Just return the pte as-is. We preserve the bits on commit */ | |
313 | trace_xen_mmu_ptep_modify_prot_start(mm, addr, ptep, *ptep); | |
314 | return *ptep; | |
315 | } | |
316 | ||
317 | void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr, | |
318 | pte_t *ptep, pte_t pte) | |
319 | { | |
320 | struct mmu_update u; | |
321 | ||
322 | trace_xen_mmu_ptep_modify_prot_commit(mm, addr, ptep, pte); | |
323 | xen_mc_batch(); | |
324 | ||
325 | u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD; | |
326 | u.val = pte_val_ma(pte); | |
327 | xen_extend_mmu_update(&u); | |
328 | ||
329 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
330 | } | |
331 | ||
332 | /* Assume pteval_t is equivalent to all the other *val_t types. */ | |
333 | static pteval_t pte_mfn_to_pfn(pteval_t val) | |
334 | { | |
335 | if (val & _PAGE_PRESENT) { | |
6f0e8bf1 | 336 | unsigned long mfn = (val & XEN_PTE_MFN_MASK) >> PAGE_SHIFT; |
7e0563de VK |
337 | unsigned long pfn = mfn_to_pfn(mfn); |
338 | ||
339 | pteval_t flags = val & PTE_FLAGS_MASK; | |
340 | if (unlikely(pfn == ~0)) | |
341 | val = flags & ~_PAGE_PRESENT; | |
342 | else | |
343 | val = ((pteval_t)pfn << PAGE_SHIFT) | flags; | |
344 | } | |
345 | ||
346 | return val; | |
347 | } | |
348 | ||
349 | static pteval_t pte_pfn_to_mfn(pteval_t val) | |
350 | { | |
351 | if (val & _PAGE_PRESENT) { | |
352 | unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; | |
353 | pteval_t flags = val & PTE_FLAGS_MASK; | |
354 | unsigned long mfn; | |
355 | ||
989513a7 JG |
356 | mfn = __pfn_to_mfn(pfn); |
357 | ||
7e0563de VK |
358 | /* |
359 | * If there's no mfn for the pfn, then just create an | |
360 | * empty non-present pte. Unfortunately this loses | |
361 | * information about the original pfn, so | |
362 | * pte_mfn_to_pfn is asymmetric. | |
363 | */ | |
364 | if (unlikely(mfn == INVALID_P2M_ENTRY)) { | |
365 | mfn = 0; | |
366 | flags = 0; | |
367 | } else | |
368 | mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT); | |
369 | val = ((pteval_t)mfn << PAGE_SHIFT) | flags; | |
370 | } | |
371 | ||
372 | return val; | |
373 | } | |
374 | ||
375 | __visible pteval_t xen_pte_val(pte_t pte) | |
376 | { | |
377 | pteval_t pteval = pte.pte; | |
378 | ||
379 | return pte_mfn_to_pfn(pteval); | |
380 | } | |
381 | PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val); | |
382 | ||
383 | __visible pgdval_t xen_pgd_val(pgd_t pgd) | |
384 | { | |
385 | return pte_mfn_to_pfn(pgd.pgd); | |
386 | } | |
387 | PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val); | |
388 | ||
389 | __visible pte_t xen_make_pte(pteval_t pte) | |
390 | { | |
391 | pte = pte_pfn_to_mfn(pte); | |
392 | ||
393 | return native_make_pte(pte); | |
394 | } | |
395 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte); | |
396 | ||
397 | __visible pgd_t xen_make_pgd(pgdval_t pgd) | |
398 | { | |
399 | pgd = pte_pfn_to_mfn(pgd); | |
400 | return native_make_pgd(pgd); | |
401 | } | |
402 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd); | |
403 | ||
404 | __visible pmdval_t xen_pmd_val(pmd_t pmd) | |
405 | { | |
406 | return pte_mfn_to_pfn(pmd.pmd); | |
407 | } | |
408 | PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val); | |
409 | ||
410 | static void xen_set_pud_hyper(pud_t *ptr, pud_t val) | |
411 | { | |
412 | struct mmu_update u; | |
413 | ||
414 | preempt_disable(); | |
415 | ||
416 | xen_mc_batch(); | |
417 | ||
418 | /* ptr may be ioremapped for 64-bit pagetable setup */ | |
419 | u.ptr = arbitrary_virt_to_machine(ptr).maddr; | |
420 | u.val = pud_val_ma(val); | |
421 | xen_extend_mmu_update(&u); | |
422 | ||
423 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
424 | ||
425 | preempt_enable(); | |
426 | } | |
427 | ||
428 | static void xen_set_pud(pud_t *ptr, pud_t val) | |
429 | { | |
430 | trace_xen_mmu_set_pud(ptr, val); | |
431 | ||
432 | /* If page is not pinned, we can just update the entry | |
433 | directly */ | |
434 | if (!xen_page_pinned(ptr)) { | |
435 | *ptr = val; | |
436 | return; | |
437 | } | |
438 | ||
439 | xen_set_pud_hyper(ptr, val); | |
440 | } | |
441 | ||
442 | #ifdef CONFIG_X86_PAE | |
443 | static void xen_set_pte_atomic(pte_t *ptep, pte_t pte) | |
444 | { | |
445 | trace_xen_mmu_set_pte_atomic(ptep, pte); | |
f7c90c2a | 446 | __xen_set_pte(ptep, pte); |
7e0563de VK |
447 | } |
448 | ||
449 | static void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep) | |
450 | { | |
451 | trace_xen_mmu_pte_clear(mm, addr, ptep); | |
f7c90c2a | 452 | __xen_set_pte(ptep, native_make_pte(0)); |
7e0563de VK |
453 | } |
454 | ||
455 | static void xen_pmd_clear(pmd_t *pmdp) | |
456 | { | |
457 | trace_xen_mmu_pmd_clear(pmdp); | |
458 | set_pmd(pmdp, __pmd(0)); | |
459 | } | |
460 | #endif /* CONFIG_X86_PAE */ | |
461 | ||
462 | __visible pmd_t xen_make_pmd(pmdval_t pmd) | |
463 | { | |
464 | pmd = pte_pfn_to_mfn(pmd); | |
465 | return native_make_pmd(pmd); | |
466 | } | |
467 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd); | |
468 | ||
773dd2fc | 469 | #ifdef CONFIG_X86_64 |
7e0563de VK |
470 | __visible pudval_t xen_pud_val(pud_t pud) |
471 | { | |
472 | return pte_mfn_to_pfn(pud.pud); | |
473 | } | |
474 | PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val); | |
475 | ||
476 | __visible pud_t xen_make_pud(pudval_t pud) | |
477 | { | |
478 | pud = pte_pfn_to_mfn(pud); | |
479 | ||
480 | return native_make_pud(pud); | |
481 | } | |
482 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud); | |
483 | ||
484 | static pgd_t *xen_get_user_pgd(pgd_t *pgd) | |
485 | { | |
486 | pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK); | |
487 | unsigned offset = pgd - pgd_page; | |
488 | pgd_t *user_ptr = NULL; | |
489 | ||
490 | if (offset < pgd_index(USER_LIMIT)) { | |
491 | struct page *page = virt_to_page(pgd_page); | |
492 | user_ptr = (pgd_t *)page->private; | |
493 | if (user_ptr) | |
494 | user_ptr += offset; | |
495 | } | |
496 | ||
497 | return user_ptr; | |
498 | } | |
499 | ||
500 | static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val) | |
501 | { | |
502 | struct mmu_update u; | |
503 | ||
504 | u.ptr = virt_to_machine(ptr).maddr; | |
505 | u.val = p4d_val_ma(val); | |
506 | xen_extend_mmu_update(&u); | |
507 | } | |
508 | ||
509 | /* | |
510 | * Raw hypercall-based set_p4d, intended for in early boot before | |
511 | * there's a page structure. This implies: | |
512 | * 1. The only existing pagetable is the kernel's | |
513 | * 2. It is always pinned | |
514 | * 3. It has no user pagetable attached to it | |
515 | */ | |
516 | static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val) | |
517 | { | |
518 | preempt_disable(); | |
519 | ||
520 | xen_mc_batch(); | |
521 | ||
522 | __xen_set_p4d_hyper(ptr, val); | |
523 | ||
524 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
525 | ||
526 | preempt_enable(); | |
527 | } | |
528 | ||
529 | static void xen_set_p4d(p4d_t *ptr, p4d_t val) | |
530 | { | |
531 | pgd_t *user_ptr = xen_get_user_pgd((pgd_t *)ptr); | |
532 | pgd_t pgd_val; | |
533 | ||
534 | trace_xen_mmu_set_p4d(ptr, (p4d_t *)user_ptr, val); | |
535 | ||
536 | /* If page is not pinned, we can just update the entry | |
537 | directly */ | |
538 | if (!xen_page_pinned(ptr)) { | |
539 | *ptr = val; | |
540 | if (user_ptr) { | |
541 | WARN_ON(xen_page_pinned(user_ptr)); | |
542 | pgd_val.pgd = p4d_val_ma(val); | |
543 | *user_ptr = pgd_val; | |
544 | } | |
545 | return; | |
546 | } | |
547 | ||
548 | /* If it's pinned, then we can at least batch the kernel and | |
549 | user updates together. */ | |
550 | xen_mc_batch(); | |
551 | ||
552 | __xen_set_p4d_hyper(ptr, val); | |
553 | if (user_ptr) | |
554 | __xen_set_p4d_hyper((p4d_t *)user_ptr, val); | |
555 | ||
556 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
557 | } | |
b9952ec7 KS |
558 | |
559 | #if CONFIG_PGTABLE_LEVELS >= 5 | |
560 | __visible p4dval_t xen_p4d_val(p4d_t p4d) | |
561 | { | |
562 | return pte_mfn_to_pfn(p4d.p4d); | |
563 | } | |
564 | PV_CALLEE_SAVE_REGS_THUNK(xen_p4d_val); | |
565 | ||
566 | __visible p4d_t xen_make_p4d(p4dval_t p4d) | |
567 | { | |
568 | p4d = pte_pfn_to_mfn(p4d); | |
569 | ||
570 | return native_make_p4d(p4d); | |
571 | } | |
572 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_p4d); | |
573 | #endif /* CONFIG_PGTABLE_LEVELS >= 5 */ | |
773dd2fc | 574 | #endif /* CONFIG_X86_64 */ |
7e0563de VK |
575 | |
576 | static int xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd, | |
577 | int (*func)(struct mm_struct *mm, struct page *, enum pt_level), | |
578 | bool last, unsigned long limit) | |
579 | { | |
580 | int i, nr, flush = 0; | |
581 | ||
582 | nr = last ? pmd_index(limit) + 1 : PTRS_PER_PMD; | |
583 | for (i = 0; i < nr; i++) { | |
584 | if (!pmd_none(pmd[i])) | |
585 | flush |= (*func)(mm, pmd_page(pmd[i]), PT_PTE); | |
586 | } | |
587 | return flush; | |
588 | } | |
589 | ||
590 | static int xen_pud_walk(struct mm_struct *mm, pud_t *pud, | |
591 | int (*func)(struct mm_struct *mm, struct page *, enum pt_level), | |
592 | bool last, unsigned long limit) | |
593 | { | |
594 | int i, nr, flush = 0; | |
595 | ||
596 | nr = last ? pud_index(limit) + 1 : PTRS_PER_PUD; | |
597 | for (i = 0; i < nr; i++) { | |
598 | pmd_t *pmd; | |
599 | ||
600 | if (pud_none(pud[i])) | |
601 | continue; | |
602 | ||
603 | pmd = pmd_offset(&pud[i], 0); | |
604 | if (PTRS_PER_PMD > 1) | |
605 | flush |= (*func)(mm, virt_to_page(pmd), PT_PMD); | |
606 | flush |= xen_pmd_walk(mm, pmd, func, | |
607 | last && i == nr - 1, limit); | |
608 | } | |
609 | return flush; | |
610 | } | |
611 | ||
612 | static int xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d, | |
613 | int (*func)(struct mm_struct *mm, struct page *, enum pt_level), | |
614 | bool last, unsigned long limit) | |
615 | { | |
773dd2fc KS |
616 | int flush = 0; |
617 | pud_t *pud; | |
7e0563de | 618 | |
7e0563de | 619 | |
773dd2fc KS |
620 | if (p4d_none(*p4d)) |
621 | return flush; | |
7e0563de | 622 | |
773dd2fc KS |
623 | pud = pud_offset(p4d, 0); |
624 | if (PTRS_PER_PUD > 1) | |
625 | flush |= (*func)(mm, virt_to_page(pud), PT_PUD); | |
626 | flush |= xen_pud_walk(mm, pud, func, last, limit); | |
7e0563de VK |
627 | return flush; |
628 | } | |
629 | ||
630 | /* | |
631 | * (Yet another) pagetable walker. This one is intended for pinning a | |
632 | * pagetable. This means that it walks a pagetable and calls the | |
633 | * callback function on each page it finds making up the page table, | |
634 | * at every level. It walks the entire pagetable, but it only bothers | |
635 | * pinning pte pages which are below limit. In the normal case this | |
636 | * will be STACK_TOP_MAX, but at boot we need to pin up to | |
637 | * FIXADDR_TOP. | |
638 | * | |
639 | * For 32-bit the important bit is that we don't pin beyond there, | |
640 | * because then we start getting into Xen's ptes. | |
641 | * | |
642 | * For 64-bit, we must skip the Xen hole in the middle of the address | |
643 | * space, just after the big x86-64 virtual hole. | |
644 | */ | |
645 | static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd, | |
646 | int (*func)(struct mm_struct *mm, struct page *, | |
647 | enum pt_level), | |
648 | unsigned long limit) | |
649 | { | |
650 | int i, nr, flush = 0; | |
651 | unsigned hole_low, hole_high; | |
652 | ||
653 | /* The limit is the last byte to be touched */ | |
654 | limit--; | |
655 | BUG_ON(limit >= FIXADDR_TOP); | |
656 | ||
7e0563de VK |
657 | /* |
658 | * 64-bit has a great big hole in the middle of the address | |
659 | * space, which contains the Xen mappings. On 32-bit these | |
660 | * will end up making a zero-sized hole and so is a no-op. | |
661 | */ | |
662 | hole_low = pgd_index(USER_LIMIT); | |
663 | hole_high = pgd_index(PAGE_OFFSET); | |
664 | ||
665 | nr = pgd_index(limit) + 1; | |
666 | for (i = 0; i < nr; i++) { | |
667 | p4d_t *p4d; | |
668 | ||
669 | if (i >= hole_low && i < hole_high) | |
670 | continue; | |
671 | ||
672 | if (pgd_none(pgd[i])) | |
673 | continue; | |
674 | ||
675 | p4d = p4d_offset(&pgd[i], 0); | |
7e0563de VK |
676 | flush |= xen_p4d_walk(mm, p4d, func, i == nr - 1, limit); |
677 | } | |
678 | ||
679 | /* Do the top level last, so that the callbacks can use it as | |
680 | a cue to do final things like tlb flushes. */ | |
681 | flush |= (*func)(mm, virt_to_page(pgd), PT_PGD); | |
682 | ||
683 | return flush; | |
684 | } | |
685 | ||
686 | static int xen_pgd_walk(struct mm_struct *mm, | |
687 | int (*func)(struct mm_struct *mm, struct page *, | |
688 | enum pt_level), | |
689 | unsigned long limit) | |
690 | { | |
691 | return __xen_pgd_walk(mm, mm->pgd, func, limit); | |
692 | } | |
693 | ||
694 | /* If we're using split pte locks, then take the page's lock and | |
695 | return a pointer to it. Otherwise return NULL. */ | |
696 | static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm) | |
697 | { | |
698 | spinlock_t *ptl = NULL; | |
699 | ||
700 | #if USE_SPLIT_PTE_PTLOCKS | |
701 | ptl = ptlock_ptr(page); | |
702 | spin_lock_nest_lock(ptl, &mm->page_table_lock); | |
703 | #endif | |
704 | ||
705 | return ptl; | |
706 | } | |
707 | ||
708 | static void xen_pte_unlock(void *v) | |
709 | { | |
710 | spinlock_t *ptl = v; | |
711 | spin_unlock(ptl); | |
712 | } | |
713 | ||
714 | static void xen_do_pin(unsigned level, unsigned long pfn) | |
715 | { | |
716 | struct mmuext_op op; | |
717 | ||
718 | op.cmd = level; | |
719 | op.arg1.mfn = pfn_to_mfn(pfn); | |
720 | ||
721 | xen_extend_mmuext_op(&op); | |
722 | } | |
723 | ||
724 | static int xen_pin_page(struct mm_struct *mm, struct page *page, | |
725 | enum pt_level level) | |
726 | { | |
727 | unsigned pgfl = TestSetPagePinned(page); | |
728 | int flush; | |
729 | ||
730 | if (pgfl) | |
731 | flush = 0; /* already pinned */ | |
732 | else if (PageHighMem(page)) | |
733 | /* kmaps need flushing if we found an unpinned | |
734 | highpage */ | |
735 | flush = 1; | |
736 | else { | |
737 | void *pt = lowmem_page_address(page); | |
738 | unsigned long pfn = page_to_pfn(page); | |
739 | struct multicall_space mcs = __xen_mc_entry(0); | |
740 | spinlock_t *ptl; | |
741 | ||
742 | flush = 0; | |
743 | ||
744 | /* | |
745 | * We need to hold the pagetable lock between the time | |
746 | * we make the pagetable RO and when we actually pin | |
747 | * it. If we don't, then other users may come in and | |
748 | * attempt to update the pagetable by writing it, | |
749 | * which will fail because the memory is RO but not | |
750 | * pinned, so Xen won't do the trap'n'emulate. | |
751 | * | |
752 | * If we're using split pte locks, we can't hold the | |
753 | * entire pagetable's worth of locks during the | |
754 | * traverse, because we may wrap the preempt count (8 | |
755 | * bits). The solution is to mark RO and pin each PTE | |
756 | * page while holding the lock. This means the number | |
757 | * of locks we end up holding is never more than a | |
758 | * batch size (~32 entries, at present). | |
759 | * | |
760 | * If we're not using split pte locks, we needn't pin | |
761 | * the PTE pages independently, because we're | |
762 | * protected by the overall pagetable lock. | |
763 | */ | |
764 | ptl = NULL; | |
765 | if (level == PT_PTE) | |
766 | ptl = xen_pte_lock(page, mm); | |
767 | ||
768 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, | |
769 | pfn_pte(pfn, PAGE_KERNEL_RO), | |
770 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); | |
771 | ||
772 | if (ptl) { | |
773 | xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn); | |
774 | ||
775 | /* Queue a deferred unlock for when this batch | |
776 | is completed. */ | |
777 | xen_mc_callback(xen_pte_unlock, ptl); | |
778 | } | |
779 | } | |
780 | ||
781 | return flush; | |
782 | } | |
783 | ||
784 | /* This is called just after a mm has been created, but it has not | |
785 | been used yet. We need to make sure that its pagetable is all | |
786 | read-only, and can be pinned. */ | |
787 | static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd) | |
788 | { | |
789 | trace_xen_mmu_pgd_pin(mm, pgd); | |
790 | ||
791 | xen_mc_batch(); | |
792 | ||
793 | if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) { | |
794 | /* re-enable interrupts for flushing */ | |
795 | xen_mc_issue(0); | |
796 | ||
797 | kmap_flush_unused(); | |
798 | ||
799 | xen_mc_batch(); | |
800 | } | |
801 | ||
802 | #ifdef CONFIG_X86_64 | |
803 | { | |
804 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
805 | ||
806 | xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd))); | |
807 | ||
808 | if (user_pgd) { | |
809 | xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD); | |
810 | xen_do_pin(MMUEXT_PIN_L4_TABLE, | |
811 | PFN_DOWN(__pa(user_pgd))); | |
812 | } | |
813 | } | |
814 | #else /* CONFIG_X86_32 */ | |
815 | #ifdef CONFIG_X86_PAE | |
816 | /* Need to make sure unshared kernel PMD is pinnable */ | |
817 | xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), | |
818 | PT_PMD); | |
819 | #endif | |
820 | xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd))); | |
821 | #endif /* CONFIG_X86_64 */ | |
822 | xen_mc_issue(0); | |
823 | } | |
824 | ||
825 | static void xen_pgd_pin(struct mm_struct *mm) | |
826 | { | |
827 | __xen_pgd_pin(mm, mm->pgd); | |
828 | } | |
829 | ||
830 | /* | |
831 | * On save, we need to pin all pagetables to make sure they get their | |
832 | * mfns turned into pfns. Search the list for any unpinned pgds and pin | |
833 | * them (unpinned pgds are not currently in use, probably because the | |
834 | * process is under construction or destruction). | |
835 | * | |
836 | * Expected to be called in stop_machine() ("equivalent to taking | |
837 | * every spinlock in the system"), so the locking doesn't really | |
838 | * matter all that much. | |
839 | */ | |
840 | void xen_mm_pin_all(void) | |
841 | { | |
842 | struct page *page; | |
843 | ||
844 | spin_lock(&pgd_lock); | |
845 | ||
846 | list_for_each_entry(page, &pgd_list, lru) { | |
847 | if (!PagePinned(page)) { | |
848 | __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page)); | |
849 | SetPageSavePinned(page); | |
850 | } | |
851 | } | |
852 | ||
853 | spin_unlock(&pgd_lock); | |
854 | } | |
855 | ||
7e0563de VK |
856 | static int __init xen_mark_pinned(struct mm_struct *mm, struct page *page, |
857 | enum pt_level level) | |
858 | { | |
859 | SetPagePinned(page); | |
860 | return 0; | |
861 | } | |
862 | ||
6f84f8d1 PT |
863 | /* |
864 | * The init_mm pagetable is really pinned as soon as its created, but | |
865 | * that's before we have page structures to store the bits. So do all | |
866 | * the book-keeping now once struct pages for allocated pages are | |
c6ffc5ca | 867 | * initialized. This happens only after memblock_free_all() is called. |
6f84f8d1 PT |
868 | */ |
869 | static void __init xen_after_bootmem(void) | |
7e0563de | 870 | { |
6f84f8d1 PT |
871 | static_branch_enable(&xen_struct_pages_ready); |
872 | #ifdef CONFIG_X86_64 | |
873 | SetPagePinned(virt_to_page(level3_user_vsyscall)); | |
874 | #endif | |
7e0563de VK |
875 | xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP); |
876 | } | |
877 | ||
878 | static int xen_unpin_page(struct mm_struct *mm, struct page *page, | |
879 | enum pt_level level) | |
880 | { | |
881 | unsigned pgfl = TestClearPagePinned(page); | |
882 | ||
883 | if (pgfl && !PageHighMem(page)) { | |
884 | void *pt = lowmem_page_address(page); | |
885 | unsigned long pfn = page_to_pfn(page); | |
886 | spinlock_t *ptl = NULL; | |
887 | struct multicall_space mcs; | |
888 | ||
889 | /* | |
890 | * Do the converse to pin_page. If we're using split | |
891 | * pte locks, we must be holding the lock for while | |
892 | * the pte page is unpinned but still RO to prevent | |
893 | * concurrent updates from seeing it in this | |
894 | * partially-pinned state. | |
895 | */ | |
896 | if (level == PT_PTE) { | |
897 | ptl = xen_pte_lock(page, mm); | |
898 | ||
899 | if (ptl) | |
900 | xen_do_pin(MMUEXT_UNPIN_TABLE, pfn); | |
901 | } | |
902 | ||
903 | mcs = __xen_mc_entry(0); | |
904 | ||
905 | MULTI_update_va_mapping(mcs.mc, (unsigned long)pt, | |
906 | pfn_pte(pfn, PAGE_KERNEL), | |
907 | level == PT_PGD ? UVMF_TLB_FLUSH : 0); | |
908 | ||
909 | if (ptl) { | |
910 | /* unlock when batch completed */ | |
911 | xen_mc_callback(xen_pte_unlock, ptl); | |
912 | } | |
913 | } | |
914 | ||
915 | return 0; /* never need to flush on unpin */ | |
916 | } | |
917 | ||
918 | /* Release a pagetables pages back as normal RW */ | |
919 | static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd) | |
920 | { | |
921 | trace_xen_mmu_pgd_unpin(mm, pgd); | |
922 | ||
923 | xen_mc_batch(); | |
924 | ||
925 | xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
926 | ||
927 | #ifdef CONFIG_X86_64 | |
928 | { | |
929 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
930 | ||
931 | if (user_pgd) { | |
932 | xen_do_pin(MMUEXT_UNPIN_TABLE, | |
933 | PFN_DOWN(__pa(user_pgd))); | |
934 | xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD); | |
935 | } | |
936 | } | |
937 | #endif | |
938 | ||
939 | #ifdef CONFIG_X86_PAE | |
940 | /* Need to make sure unshared kernel PMD is unpinned */ | |
941 | xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]), | |
942 | PT_PMD); | |
943 | #endif | |
944 | ||
945 | __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT); | |
946 | ||
947 | xen_mc_issue(0); | |
948 | } | |
949 | ||
950 | static void xen_pgd_unpin(struct mm_struct *mm) | |
951 | { | |
952 | __xen_pgd_unpin(mm, mm->pgd); | |
953 | } | |
954 | ||
955 | /* | |
956 | * On resume, undo any pinning done at save, so that the rest of the | |
957 | * kernel doesn't see any unexpected pinned pagetables. | |
958 | */ | |
959 | void xen_mm_unpin_all(void) | |
960 | { | |
961 | struct page *page; | |
962 | ||
963 | spin_lock(&pgd_lock); | |
964 | ||
965 | list_for_each_entry(page, &pgd_list, lru) { | |
966 | if (PageSavePinned(page)) { | |
967 | BUG_ON(!PagePinned(page)); | |
968 | __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page)); | |
969 | ClearPageSavePinned(page); | |
970 | } | |
971 | } | |
972 | ||
973 | spin_unlock(&pgd_lock); | |
974 | } | |
975 | ||
976 | static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next) | |
977 | { | |
978 | spin_lock(&next->page_table_lock); | |
979 | xen_pgd_pin(next); | |
980 | spin_unlock(&next->page_table_lock); | |
981 | } | |
982 | ||
983 | static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm) | |
984 | { | |
985 | spin_lock(&mm->page_table_lock); | |
986 | xen_pgd_pin(mm); | |
987 | spin_unlock(&mm->page_table_lock); | |
988 | } | |
989 | ||
3d28ebce | 990 | static void drop_mm_ref_this_cpu(void *info) |
7e0563de VK |
991 | { |
992 | struct mm_struct *mm = info; | |
7e0563de | 993 | |
3d28ebce | 994 | if (this_cpu_read(cpu_tlbstate.loaded_mm) == mm) |
7e0563de VK |
995 | leave_mm(smp_processor_id()); |
996 | ||
3d28ebce AL |
997 | /* |
998 | * If this cpu still has a stale cr3 reference, then make sure | |
999 | * it has been flushed. | |
1000 | */ | |
7e0563de | 1001 | if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd)) |
3d28ebce | 1002 | xen_mc_flush(); |
7e0563de VK |
1003 | } |
1004 | ||
3d28ebce AL |
1005 | #ifdef CONFIG_SMP |
1006 | /* | |
1007 | * Another cpu may still have their %cr3 pointing at the pagetable, so | |
1008 | * we need to repoint it somewhere else before we can unpin it. | |
1009 | */ | |
7e0563de VK |
1010 | static void xen_drop_mm_ref(struct mm_struct *mm) |
1011 | { | |
1012 | cpumask_var_t mask; | |
1013 | unsigned cpu; | |
1014 | ||
3d28ebce | 1015 | drop_mm_ref_this_cpu(mm); |
7e0563de VK |
1016 | |
1017 | /* Get the "official" set of cpus referring to our pagetable. */ | |
1018 | if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) { | |
1019 | for_each_online_cpu(cpu) { | |
94b1b03b | 1020 | if (per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd)) |
7e0563de | 1021 | continue; |
3d28ebce | 1022 | smp_call_function_single(cpu, drop_mm_ref_this_cpu, mm, 1); |
7e0563de VK |
1023 | } |
1024 | return; | |
1025 | } | |
7e0563de | 1026 | |
3d28ebce AL |
1027 | /* |
1028 | * It's possible that a vcpu may have a stale reference to our | |
1029 | * cr3, because its in lazy mode, and it hasn't yet flushed | |
1030 | * its set of pending hypercalls yet. In this case, we can | |
1031 | * look at its actual current cr3 value, and force it to flush | |
1032 | * if needed. | |
1033 | */ | |
94b1b03b | 1034 | cpumask_clear(mask); |
7e0563de VK |
1035 | for_each_online_cpu(cpu) { |
1036 | if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd)) | |
1037 | cpumask_set_cpu(cpu, mask); | |
1038 | } | |
1039 | ||
3d28ebce | 1040 | smp_call_function_many(mask, drop_mm_ref_this_cpu, mm, 1); |
7e0563de VK |
1041 | free_cpumask_var(mask); |
1042 | } | |
1043 | #else | |
1044 | static void xen_drop_mm_ref(struct mm_struct *mm) | |
1045 | { | |
3d28ebce | 1046 | drop_mm_ref_this_cpu(mm); |
7e0563de VK |
1047 | } |
1048 | #endif | |
1049 | ||
1050 | /* | |
1051 | * While a process runs, Xen pins its pagetables, which means that the | |
1052 | * hypervisor forces it to be read-only, and it controls all updates | |
1053 | * to it. This means that all pagetable updates have to go via the | |
1054 | * hypervisor, which is moderately expensive. | |
1055 | * | |
1056 | * Since we're pulling the pagetable down, we switch to use init_mm, | |
1057 | * unpin old process pagetable and mark it all read-write, which | |
1058 | * allows further operations on it to be simple memory accesses. | |
1059 | * | |
1060 | * The only subtle point is that another CPU may be still using the | |
1061 | * pagetable because of lazy tlb flushing. This means we need need to | |
1062 | * switch all CPUs off this pagetable before we can unpin it. | |
1063 | */ | |
1064 | static void xen_exit_mmap(struct mm_struct *mm) | |
1065 | { | |
1066 | get_cpu(); /* make sure we don't move around */ | |
1067 | xen_drop_mm_ref(mm); | |
1068 | put_cpu(); | |
1069 | ||
1070 | spin_lock(&mm->page_table_lock); | |
1071 | ||
1072 | /* pgd may not be pinned in the error exit path of execve */ | |
1073 | if (xen_page_pinned(mm->pgd)) | |
1074 | xen_pgd_unpin(mm); | |
1075 | ||
1076 | spin_unlock(&mm->page_table_lock); | |
1077 | } | |
1078 | ||
1079 | static void xen_post_allocator_init(void); | |
1080 | ||
1081 | static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn) | |
1082 | { | |
1083 | struct mmuext_op op; | |
1084 | ||
1085 | op.cmd = cmd; | |
1086 | op.arg1.mfn = pfn_to_mfn(pfn); | |
1087 | if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF)) | |
1088 | BUG(); | |
1089 | } | |
1090 | ||
1091 | #ifdef CONFIG_X86_64 | |
1092 | static void __init xen_cleanhighmap(unsigned long vaddr, | |
1093 | unsigned long vaddr_end) | |
1094 | { | |
1095 | unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1; | |
1096 | pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr); | |
1097 | ||
1098 | /* NOTE: The loop is more greedy than the cleanup_highmap variant. | |
1099 | * We include the PMD passed in on _both_ boundaries. */ | |
1100 | for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD)); | |
1101 | pmd++, vaddr += PMD_SIZE) { | |
1102 | if (pmd_none(*pmd)) | |
1103 | continue; | |
1104 | if (vaddr < (unsigned long) _text || vaddr > kernel_end) | |
1105 | set_pmd(pmd, __pmd(0)); | |
1106 | } | |
1107 | /* In case we did something silly, we should crash in this function | |
1108 | * instead of somewhere later and be confusing. */ | |
1109 | xen_mc_flush(); | |
1110 | } | |
1111 | ||
1112 | /* | |
1113 | * Make a page range writeable and free it. | |
1114 | */ | |
1115 | static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size) | |
1116 | { | |
1117 | void *vaddr = __va(paddr); | |
1118 | void *vaddr_end = vaddr + size; | |
1119 | ||
1120 | for (; vaddr < vaddr_end; vaddr += PAGE_SIZE) | |
1121 | make_lowmem_page_readwrite(vaddr); | |
1122 | ||
1123 | memblock_free(paddr, size); | |
1124 | } | |
1125 | ||
1126 | static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin) | |
1127 | { | |
1128 | unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK; | |
1129 | ||
1130 | if (unpin) | |
1131 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa)); | |
1132 | ClearPagePinned(virt_to_page(__va(pa))); | |
1133 | xen_free_ro_pages(pa, PAGE_SIZE); | |
1134 | } | |
1135 | ||
1136 | static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin) | |
1137 | { | |
1138 | unsigned long pa; | |
1139 | pte_t *pte_tbl; | |
1140 | int i; | |
1141 | ||
1142 | if (pmd_large(*pmd)) { | |
1143 | pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK; | |
1144 | xen_free_ro_pages(pa, PMD_SIZE); | |
1145 | return; | |
1146 | } | |
1147 | ||
1148 | pte_tbl = pte_offset_kernel(pmd, 0); | |
1149 | for (i = 0; i < PTRS_PER_PTE; i++) { | |
1150 | if (pte_none(pte_tbl[i])) | |
1151 | continue; | |
1152 | pa = pte_pfn(pte_tbl[i]) << PAGE_SHIFT; | |
1153 | xen_free_ro_pages(pa, PAGE_SIZE); | |
1154 | } | |
1155 | set_pmd(pmd, __pmd(0)); | |
1156 | xen_cleanmfnmap_free_pgtbl(pte_tbl, unpin); | |
1157 | } | |
1158 | ||
1159 | static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin) | |
1160 | { | |
1161 | unsigned long pa; | |
1162 | pmd_t *pmd_tbl; | |
1163 | int i; | |
1164 | ||
1165 | if (pud_large(*pud)) { | |
1166 | pa = pud_val(*pud) & PHYSICAL_PAGE_MASK; | |
1167 | xen_free_ro_pages(pa, PUD_SIZE); | |
1168 | return; | |
1169 | } | |
1170 | ||
1171 | pmd_tbl = pmd_offset(pud, 0); | |
1172 | for (i = 0; i < PTRS_PER_PMD; i++) { | |
1173 | if (pmd_none(pmd_tbl[i])) | |
1174 | continue; | |
1175 | xen_cleanmfnmap_pmd(pmd_tbl + i, unpin); | |
1176 | } | |
1177 | set_pud(pud, __pud(0)); | |
1178 | xen_cleanmfnmap_free_pgtbl(pmd_tbl, unpin); | |
1179 | } | |
1180 | ||
1181 | static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin) | |
1182 | { | |
1183 | unsigned long pa; | |
1184 | pud_t *pud_tbl; | |
1185 | int i; | |
1186 | ||
1187 | if (p4d_large(*p4d)) { | |
1188 | pa = p4d_val(*p4d) & PHYSICAL_PAGE_MASK; | |
1189 | xen_free_ro_pages(pa, P4D_SIZE); | |
1190 | return; | |
1191 | } | |
1192 | ||
1193 | pud_tbl = pud_offset(p4d, 0); | |
1194 | for (i = 0; i < PTRS_PER_PUD; i++) { | |
1195 | if (pud_none(pud_tbl[i])) | |
1196 | continue; | |
1197 | xen_cleanmfnmap_pud(pud_tbl + i, unpin); | |
1198 | } | |
1199 | set_p4d(p4d, __p4d(0)); | |
1200 | xen_cleanmfnmap_free_pgtbl(pud_tbl, unpin); | |
1201 | } | |
1202 | ||
1203 | /* | |
1204 | * Since it is well isolated we can (and since it is perhaps large we should) | |
1205 | * also free the page tables mapping the initial P->M table. | |
1206 | */ | |
1207 | static void __init xen_cleanmfnmap(unsigned long vaddr) | |
1208 | { | |
1209 | pgd_t *pgd; | |
1210 | p4d_t *p4d; | |
7e0563de VK |
1211 | bool unpin; |
1212 | ||
1213 | unpin = (vaddr == 2 * PGDIR_SIZE); | |
1214 | vaddr &= PMD_MASK; | |
1215 | pgd = pgd_offset_k(vaddr); | |
1216 | p4d = p4d_offset(pgd, 0); | |
773dd2fc KS |
1217 | if (!p4d_none(*p4d)) |
1218 | xen_cleanmfnmap_p4d(p4d, unpin); | |
7e0563de VK |
1219 | } |
1220 | ||
1221 | static void __init xen_pagetable_p2m_free(void) | |
1222 | { | |
1223 | unsigned long size; | |
1224 | unsigned long addr; | |
1225 | ||
1226 | size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long)); | |
1227 | ||
1228 | /* No memory or already called. */ | |
1229 | if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list) | |
1230 | return; | |
1231 | ||
1232 | /* using __ka address and sticking INVALID_P2M_ENTRY! */ | |
1233 | memset((void *)xen_start_info->mfn_list, 0xff, size); | |
1234 | ||
1235 | addr = xen_start_info->mfn_list; | |
1236 | /* | |
1237 | * We could be in __ka space. | |
1238 | * We roundup to the PMD, which means that if anybody at this stage is | |
1239 | * using the __ka address of xen_start_info or | |
1240 | * xen_start_info->shared_info they are in going to crash. Fortunatly | |
7b25b9cb | 1241 | * we have already revectored in xen_setup_kernel_pagetable. |
7e0563de VK |
1242 | */ |
1243 | size = roundup(size, PMD_SIZE); | |
1244 | ||
1245 | if (addr >= __START_KERNEL_map) { | |
1246 | xen_cleanhighmap(addr, addr + size); | |
1247 | size = PAGE_ALIGN(xen_start_info->nr_pages * | |
1248 | sizeof(unsigned long)); | |
1249 | memblock_free(__pa(addr), size); | |
1250 | } else { | |
1251 | xen_cleanmfnmap(addr); | |
1252 | } | |
1253 | } | |
1254 | ||
1255 | static void __init xen_pagetable_cleanhighmap(void) | |
1256 | { | |
1257 | unsigned long size; | |
1258 | unsigned long addr; | |
1259 | ||
1260 | /* At this stage, cleanup_highmap has already cleaned __ka space | |
1261 | * from _brk_limit way up to the max_pfn_mapped (which is the end of | |
1262 | * the ramdisk). We continue on, erasing PMD entries that point to page | |
1263 | * tables - do note that they are accessible at this stage via __va. | |
0d805ee7 ZD |
1264 | * As Xen is aligning the memory end to a 4MB boundary, for good |
1265 | * measure we also round up to PMD_SIZE * 2 - which means that if | |
7e0563de VK |
1266 | * anybody is using __ka address to the initial boot-stack - and try |
1267 | * to use it - they are going to crash. The xen_start_info has been | |
1268 | * taken care of already in xen_setup_kernel_pagetable. */ | |
1269 | addr = xen_start_info->pt_base; | |
0d805ee7 | 1270 | size = xen_start_info->nr_pt_frames * PAGE_SIZE; |
7e0563de | 1271 | |
0d805ee7 | 1272 | xen_cleanhighmap(addr, roundup(addr + size, PMD_SIZE * 2)); |
7e0563de | 1273 | xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base)); |
7e0563de VK |
1274 | } |
1275 | #endif | |
1276 | ||
1277 | static void __init xen_pagetable_p2m_setup(void) | |
1278 | { | |
7e0563de VK |
1279 | xen_vmalloc_p2m_tree(); |
1280 | ||
1281 | #ifdef CONFIG_X86_64 | |
1282 | xen_pagetable_p2m_free(); | |
1283 | ||
1284 | xen_pagetable_cleanhighmap(); | |
1285 | #endif | |
1286 | /* And revector! Bye bye old array */ | |
1287 | xen_start_info->mfn_list = (unsigned long)xen_p2m_addr; | |
1288 | } | |
1289 | ||
1290 | static void __init xen_pagetable_init(void) | |
1291 | { | |
1292 | paging_init(); | |
1293 | xen_post_allocator_init(); | |
1294 | ||
1295 | xen_pagetable_p2m_setup(); | |
1296 | ||
1297 | /* Allocate and initialize top and mid mfn levels for p2m structure */ | |
1298 | xen_build_mfn_list_list(); | |
1299 | ||
1300 | /* Remap memory freed due to conflicts with E820 map */ | |
989513a7 | 1301 | xen_remap_memory(); |
7b25b9cb | 1302 | xen_setup_mfn_list_list(); |
7e0563de VK |
1303 | } |
1304 | static void xen_write_cr2(unsigned long cr2) | |
1305 | { | |
1306 | this_cpu_read(xen_vcpu)->arch.cr2 = cr2; | |
1307 | } | |
1308 | ||
1309 | static unsigned long xen_read_cr2(void) | |
1310 | { | |
1311 | return this_cpu_read(xen_vcpu)->arch.cr2; | |
1312 | } | |
1313 | ||
1314 | unsigned long xen_read_cr2_direct(void) | |
1315 | { | |
1316 | return this_cpu_read(xen_vcpu_info.arch.cr2); | |
1317 | } | |
1318 | ||
45dd9b06 | 1319 | static noinline void xen_flush_tlb(void) |
7e0563de VK |
1320 | { |
1321 | struct mmuext_op *op; | |
1322 | struct multicall_space mcs; | |
1323 | ||
7e0563de VK |
1324 | preempt_disable(); |
1325 | ||
1326 | mcs = xen_mc_entry(sizeof(*op)); | |
1327 | ||
1328 | op = mcs.args; | |
1329 | op->cmd = MMUEXT_TLB_FLUSH_LOCAL; | |
1330 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1331 | ||
1332 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1333 | ||
1334 | preempt_enable(); | |
1335 | } | |
1336 | ||
1299ef1d | 1337 | static void xen_flush_tlb_one_user(unsigned long addr) |
7e0563de VK |
1338 | { |
1339 | struct mmuext_op *op; | |
1340 | struct multicall_space mcs; | |
1341 | ||
1299ef1d | 1342 | trace_xen_mmu_flush_tlb_one_user(addr); |
7e0563de VK |
1343 | |
1344 | preempt_disable(); | |
1345 | ||
1346 | mcs = xen_mc_entry(sizeof(*op)); | |
1347 | op = mcs.args; | |
1348 | op->cmd = MMUEXT_INVLPG_LOCAL; | |
1349 | op->arg1.linear_addr = addr & PAGE_MASK; | |
1350 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
1351 | ||
1352 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1353 | ||
1354 | preempt_enable(); | |
1355 | } | |
1356 | ||
1357 | static void xen_flush_tlb_others(const struct cpumask *cpus, | |
a2055abe | 1358 | const struct flush_tlb_info *info) |
7e0563de VK |
1359 | { |
1360 | struct { | |
1361 | struct mmuext_op op; | |
7e0563de | 1362 | DECLARE_BITMAP(mask, NR_CPUS); |
7e0563de VK |
1363 | } *args; |
1364 | struct multicall_space mcs; | |
66a640e7 ND |
1365 | const size_t mc_entry_size = sizeof(args->op) + |
1366 | sizeof(args->mask[0]) * BITS_TO_LONGS(num_possible_cpus()); | |
7e0563de | 1367 | |
a2055abe | 1368 | trace_xen_mmu_flush_tlb_others(cpus, info->mm, info->start, info->end); |
7e0563de VK |
1369 | |
1370 | if (cpumask_empty(cpus)) | |
1371 | return; /* nothing to do */ | |
1372 | ||
66a640e7 | 1373 | mcs = xen_mc_entry(mc_entry_size); |
7e0563de VK |
1374 | args = mcs.args; |
1375 | args->op.arg2.vcpumask = to_cpumask(args->mask); | |
1376 | ||
1377 | /* Remove us, and any offline CPUS. */ | |
1378 | cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask); | |
1379 | cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask)); | |
1380 | ||
1381 | args->op.cmd = MMUEXT_TLB_FLUSH_MULTI; | |
a2055abe AL |
1382 | if (info->end != TLB_FLUSH_ALL && |
1383 | (info->end - info->start) <= PAGE_SIZE) { | |
7e0563de | 1384 | args->op.cmd = MMUEXT_INVLPG_MULTI; |
a2055abe | 1385 | args->op.arg1.linear_addr = info->start; |
7e0563de VK |
1386 | } |
1387 | ||
1388 | MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF); | |
1389 | ||
1390 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1391 | } | |
1392 | ||
1393 | static unsigned long xen_read_cr3(void) | |
1394 | { | |
1395 | return this_cpu_read(xen_cr3); | |
1396 | } | |
1397 | ||
1398 | static void set_current_cr3(void *v) | |
1399 | { | |
1400 | this_cpu_write(xen_current_cr3, (unsigned long)v); | |
1401 | } | |
1402 | ||
1403 | static void __xen_write_cr3(bool kernel, unsigned long cr3) | |
1404 | { | |
1405 | struct mmuext_op op; | |
1406 | unsigned long mfn; | |
1407 | ||
1408 | trace_xen_mmu_write_cr3(kernel, cr3); | |
1409 | ||
1410 | if (cr3) | |
1411 | mfn = pfn_to_mfn(PFN_DOWN(cr3)); | |
1412 | else | |
1413 | mfn = 0; | |
1414 | ||
1415 | WARN_ON(mfn == 0 && kernel); | |
1416 | ||
1417 | op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR; | |
1418 | op.arg1.mfn = mfn; | |
1419 | ||
1420 | xen_extend_mmuext_op(&op); | |
1421 | ||
1422 | if (kernel) { | |
1423 | this_cpu_write(xen_cr3, cr3); | |
1424 | ||
1425 | /* Update xen_current_cr3 once the batch has actually | |
1426 | been submitted. */ | |
1427 | xen_mc_callback(set_current_cr3, (void *)cr3); | |
1428 | } | |
1429 | } | |
1430 | static void xen_write_cr3(unsigned long cr3) | |
1431 | { | |
1432 | BUG_ON(preemptible()); | |
1433 | ||
1434 | xen_mc_batch(); /* disables interrupts */ | |
1435 | ||
1436 | /* Update while interrupts are disabled, so its atomic with | |
1437 | respect to ipis */ | |
1438 | this_cpu_write(xen_cr3, cr3); | |
1439 | ||
1440 | __xen_write_cr3(true, cr3); | |
1441 | ||
1442 | #ifdef CONFIG_X86_64 | |
1443 | { | |
1444 | pgd_t *user_pgd = xen_get_user_pgd(__va(cr3)); | |
1445 | if (user_pgd) | |
1446 | __xen_write_cr3(false, __pa(user_pgd)); | |
1447 | else | |
1448 | __xen_write_cr3(false, 0); | |
1449 | } | |
1450 | #endif | |
1451 | ||
1452 | xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ | |
1453 | } | |
1454 | ||
1455 | #ifdef CONFIG_X86_64 | |
1456 | /* | |
1457 | * At the start of the day - when Xen launches a guest, it has already | |
1458 | * built pagetables for the guest. We diligently look over them | |
1459 | * in xen_setup_kernel_pagetable and graft as appropriate them in the | |
65ade2f8 KS |
1460 | * init_top_pgt and its friends. Then when we are happy we load |
1461 | * the new init_top_pgt - and continue on. | |
7e0563de VK |
1462 | * |
1463 | * The generic code starts (start_kernel) and 'init_mem_mapping' sets | |
1464 | * up the rest of the pagetables. When it has completed it loads the cr3. | |
1465 | * N.B. that baremetal would start at 'start_kernel' (and the early | |
1466 | * #PF handler would create bootstrap pagetables) - so we are running | |
1467 | * with the same assumptions as what to do when write_cr3 is executed | |
1468 | * at this point. | |
1469 | * | |
1470 | * Since there are no user-page tables at all, we have two variants | |
1471 | * of xen_write_cr3 - the early bootup (this one), and the late one | |
1472 | * (xen_write_cr3). The reason we have to do that is that in 64-bit | |
1473 | * the Linux kernel and user-space are both in ring 3 while the | |
1474 | * hypervisor is in ring 0. | |
1475 | */ | |
1476 | static void __init xen_write_cr3_init(unsigned long cr3) | |
1477 | { | |
1478 | BUG_ON(preemptible()); | |
1479 | ||
1480 | xen_mc_batch(); /* disables interrupts */ | |
1481 | ||
1482 | /* Update while interrupts are disabled, so its atomic with | |
1483 | respect to ipis */ | |
1484 | this_cpu_write(xen_cr3, cr3); | |
1485 | ||
1486 | __xen_write_cr3(true, cr3); | |
1487 | ||
1488 | xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */ | |
1489 | } | |
1490 | #endif | |
1491 | ||
1492 | static int xen_pgd_alloc(struct mm_struct *mm) | |
1493 | { | |
1494 | pgd_t *pgd = mm->pgd; | |
1495 | int ret = 0; | |
1496 | ||
1497 | BUG_ON(PagePinned(virt_to_page(pgd))); | |
1498 | ||
1499 | #ifdef CONFIG_X86_64 | |
1500 | { | |
1501 | struct page *page = virt_to_page(pgd); | |
1502 | pgd_t *user_pgd; | |
1503 | ||
1504 | BUG_ON(page->private != 0); | |
1505 | ||
1506 | ret = -ENOMEM; | |
1507 | ||
1508 | user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO); | |
1509 | page->private = (unsigned long)user_pgd; | |
1510 | ||
1511 | if (user_pgd != NULL) { | |
1512 | #ifdef CONFIG_X86_VSYSCALL_EMULATION | |
1513 | user_pgd[pgd_index(VSYSCALL_ADDR)] = | |
1514 | __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE); | |
1515 | #endif | |
1516 | ret = 0; | |
1517 | } | |
1518 | ||
1519 | BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd)))); | |
1520 | } | |
1521 | #endif | |
1522 | return ret; | |
1523 | } | |
1524 | ||
1525 | static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd) | |
1526 | { | |
1527 | #ifdef CONFIG_X86_64 | |
1528 | pgd_t *user_pgd = xen_get_user_pgd(pgd); | |
1529 | ||
1530 | if (user_pgd) | |
1531 | free_page((unsigned long)user_pgd); | |
1532 | #endif | |
1533 | } | |
1534 | ||
1535 | /* | |
1536 | * Init-time set_pte while constructing initial pagetables, which | |
1537 | * doesn't allow RO page table pages to be remapped RW. | |
1538 | * | |
1539 | * If there is no MFN for this PFN then this page is initially | |
1540 | * ballooned out so clear the PTE (as in decrease_reservation() in | |
1541 | * drivers/xen/balloon.c). | |
1542 | * | |
1543 | * Many of these PTE updates are done on unpinned and writable pages | |
1544 | * and doing a hypercall for these is unnecessary and expensive. At | |
1545 | * this point it is not possible to tell if a page is pinned or not, | |
1546 | * so always write the PTE directly and rely on Xen trapping and | |
1547 | * emulating any updates as necessary. | |
1548 | */ | |
1549 | __visible pte_t xen_make_pte_init(pteval_t pte) | |
1550 | { | |
1551 | #ifdef CONFIG_X86_64 | |
1552 | unsigned long pfn; | |
1553 | ||
1554 | /* | |
1555 | * Pages belonging to the initial p2m list mapped outside the default | |
1556 | * address range must be mapped read-only. This region contains the | |
1557 | * page tables for mapping the p2m list, too, and page tables MUST be | |
1558 | * mapped read-only. | |
1559 | */ | |
1560 | pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT; | |
1561 | if (xen_start_info->mfn_list < __START_KERNEL_map && | |
1562 | pfn >= xen_start_info->first_p2m_pfn && | |
1563 | pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames) | |
1564 | pte &= ~_PAGE_RW; | |
1565 | #endif | |
1566 | pte = pte_pfn_to_mfn(pte); | |
1567 | return native_make_pte(pte); | |
1568 | } | |
1569 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init); | |
1570 | ||
1571 | static void __init xen_set_pte_init(pte_t *ptep, pte_t pte) | |
1572 | { | |
1573 | #ifdef CONFIG_X86_32 | |
1574 | /* If there's an existing pte, then don't allow _PAGE_RW to be set */ | |
1575 | if (pte_mfn(pte) != INVALID_P2M_ENTRY | |
1576 | && pte_val_ma(*ptep) & _PAGE_PRESENT) | |
1577 | pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) & | |
1578 | pte_val_ma(pte)); | |
1579 | #endif | |
f7c90c2a | 1580 | __xen_set_pte(ptep, pte); |
7e0563de VK |
1581 | } |
1582 | ||
1583 | /* Early in boot, while setting up the initial pagetable, assume | |
1584 | everything is pinned. */ | |
1585 | static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn) | |
1586 | { | |
1587 | #ifdef CONFIG_FLATMEM | |
1588 | BUG_ON(mem_map); /* should only be used early */ | |
1589 | #endif | |
1590 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1591 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
1592 | } | |
1593 | ||
1594 | /* Used for pmd and pud */ | |
1595 | static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn) | |
1596 | { | |
1597 | #ifdef CONFIG_FLATMEM | |
1598 | BUG_ON(mem_map); /* should only be used early */ | |
1599 | #endif | |
1600 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); | |
1601 | } | |
1602 | ||
1603 | /* Early release_pte assumes that all pts are pinned, since there's | |
1604 | only init_mm and anything attached to that is pinned. */ | |
1605 | static void __init xen_release_pte_init(unsigned long pfn) | |
1606 | { | |
1607 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); | |
1608 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
1609 | } | |
1610 | ||
1611 | static void __init xen_release_pmd_init(unsigned long pfn) | |
1612 | { | |
1613 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
1614 | } | |
1615 | ||
1616 | static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn) | |
1617 | { | |
1618 | struct multicall_space mcs; | |
1619 | struct mmuext_op *op; | |
1620 | ||
1621 | mcs = __xen_mc_entry(sizeof(*op)); | |
1622 | op = mcs.args; | |
1623 | op->cmd = cmd; | |
1624 | op->arg1.mfn = pfn_to_mfn(pfn); | |
1625 | ||
1626 | MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF); | |
1627 | } | |
1628 | ||
1629 | static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot) | |
1630 | { | |
1631 | struct multicall_space mcs; | |
1632 | unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT); | |
1633 | ||
1634 | mcs = __xen_mc_entry(0); | |
1635 | MULTI_update_va_mapping(mcs.mc, (unsigned long)addr, | |
1636 | pfn_pte(pfn, prot), 0); | |
1637 | } | |
1638 | ||
1639 | /* This needs to make sure the new pte page is pinned iff its being | |
1640 | attached to a pinned pagetable. */ | |
1641 | static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, | |
1642 | unsigned level) | |
1643 | { | |
6f84f8d1 | 1644 | bool pinned = xen_page_pinned(mm->pgd); |
7e0563de VK |
1645 | |
1646 | trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned); | |
1647 | ||
1648 | if (pinned) { | |
1649 | struct page *page = pfn_to_page(pfn); | |
1650 | ||
6f84f8d1 PT |
1651 | if (static_branch_likely(&xen_struct_pages_ready)) |
1652 | SetPagePinned(page); | |
7e0563de VK |
1653 | |
1654 | if (!PageHighMem(page)) { | |
1655 | xen_mc_batch(); | |
1656 | ||
1657 | __set_pfn_prot(pfn, PAGE_KERNEL_RO); | |
1658 | ||
1659 | if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS) | |
1660 | __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); | |
1661 | ||
1662 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1663 | } else { | |
1664 | /* make sure there are no stray mappings of | |
1665 | this page */ | |
1666 | kmap_flush_unused(); | |
1667 | } | |
1668 | } | |
1669 | } | |
1670 | ||
1671 | static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn) | |
1672 | { | |
1673 | xen_alloc_ptpage(mm, pfn, PT_PTE); | |
1674 | } | |
1675 | ||
1676 | static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn) | |
1677 | { | |
1678 | xen_alloc_ptpage(mm, pfn, PT_PMD); | |
1679 | } | |
1680 | ||
1681 | /* This should never happen until we're OK to use struct page */ | |
1682 | static inline void xen_release_ptpage(unsigned long pfn, unsigned level) | |
1683 | { | |
1684 | struct page *page = pfn_to_page(pfn); | |
1685 | bool pinned = PagePinned(page); | |
1686 | ||
1687 | trace_xen_mmu_release_ptpage(pfn, level, pinned); | |
1688 | ||
1689 | if (pinned) { | |
1690 | if (!PageHighMem(page)) { | |
1691 | xen_mc_batch(); | |
1692 | ||
1693 | if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS) | |
1694 | __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); | |
1695 | ||
1696 | __set_pfn_prot(pfn, PAGE_KERNEL); | |
1697 | ||
1698 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
1699 | } | |
1700 | ClearPagePinned(page); | |
1701 | } | |
1702 | } | |
1703 | ||
1704 | static void xen_release_pte(unsigned long pfn) | |
1705 | { | |
1706 | xen_release_ptpage(pfn, PT_PTE); | |
1707 | } | |
1708 | ||
1709 | static void xen_release_pmd(unsigned long pfn) | |
1710 | { | |
1711 | xen_release_ptpage(pfn, PT_PMD); | |
1712 | } | |
1713 | ||
773dd2fc | 1714 | #ifdef CONFIG_X86_64 |
7e0563de VK |
1715 | static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn) |
1716 | { | |
1717 | xen_alloc_ptpage(mm, pfn, PT_PUD); | |
1718 | } | |
1719 | ||
1720 | static void xen_release_pud(unsigned long pfn) | |
1721 | { | |
1722 | xen_release_ptpage(pfn, PT_PUD); | |
1723 | } | |
1724 | #endif | |
1725 | ||
1726 | void __init xen_reserve_top(void) | |
1727 | { | |
1728 | #ifdef CONFIG_X86_32 | |
1729 | unsigned long top = HYPERVISOR_VIRT_START; | |
1730 | struct xen_platform_parameters pp; | |
1731 | ||
1732 | if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0) | |
1733 | top = pp.virt_start; | |
1734 | ||
1735 | reserve_top_address(-top); | |
1736 | #endif /* CONFIG_X86_32 */ | |
1737 | } | |
1738 | ||
1739 | /* | |
1740 | * Like __va(), but returns address in the kernel mapping (which is | |
1741 | * all we have until the physical memory mapping has been set up. | |
1742 | */ | |
1743 | static void * __init __ka(phys_addr_t paddr) | |
1744 | { | |
1745 | #ifdef CONFIG_X86_64 | |
1746 | return (void *)(paddr + __START_KERNEL_map); | |
1747 | #else | |
1748 | return __va(paddr); | |
1749 | #endif | |
1750 | } | |
1751 | ||
1752 | /* Convert a machine address to physical address */ | |
1753 | static unsigned long __init m2p(phys_addr_t maddr) | |
1754 | { | |
1755 | phys_addr_t paddr; | |
1756 | ||
6f0e8bf1 | 1757 | maddr &= XEN_PTE_MFN_MASK; |
7e0563de VK |
1758 | paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT; |
1759 | ||
1760 | return paddr; | |
1761 | } | |
1762 | ||
1763 | /* Convert a machine address to kernel virtual */ | |
1764 | static void * __init m2v(phys_addr_t maddr) | |
1765 | { | |
1766 | return __ka(m2p(maddr)); | |
1767 | } | |
1768 | ||
1769 | /* Set the page permissions on an identity-mapped pages */ | |
1770 | static void __init set_page_prot_flags(void *addr, pgprot_t prot, | |
1771 | unsigned long flags) | |
1772 | { | |
1773 | unsigned long pfn = __pa(addr) >> PAGE_SHIFT; | |
1774 | pte_t pte = pfn_pte(pfn, prot); | |
1775 | ||
1776 | if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags)) | |
1777 | BUG(); | |
1778 | } | |
1779 | static void __init set_page_prot(void *addr, pgprot_t prot) | |
1780 | { | |
1781 | return set_page_prot_flags(addr, prot, UVMF_NONE); | |
1782 | } | |
1783 | #ifdef CONFIG_X86_32 | |
1784 | static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn) | |
1785 | { | |
1786 | unsigned pmdidx, pteidx; | |
1787 | unsigned ident_pte; | |
1788 | unsigned long pfn; | |
1789 | ||
1790 | level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES, | |
1791 | PAGE_SIZE); | |
1792 | ||
1793 | ident_pte = 0; | |
1794 | pfn = 0; | |
1795 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) { | |
1796 | pte_t *pte_page; | |
1797 | ||
1798 | /* Reuse or allocate a page of ptes */ | |
1799 | if (pmd_present(pmd[pmdidx])) | |
1800 | pte_page = m2v(pmd[pmdidx].pmd); | |
1801 | else { | |
1802 | /* Check for free pte pages */ | |
1803 | if (ident_pte == LEVEL1_IDENT_ENTRIES) | |
1804 | break; | |
1805 | ||
1806 | pte_page = &level1_ident_pgt[ident_pte]; | |
1807 | ident_pte += PTRS_PER_PTE; | |
1808 | ||
1809 | pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE); | |
1810 | } | |
1811 | ||
1812 | /* Install mappings */ | |
1813 | for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) { | |
1814 | pte_t pte; | |
1815 | ||
1816 | if (pfn > max_pfn_mapped) | |
1817 | max_pfn_mapped = pfn; | |
1818 | ||
1819 | if (!pte_none(pte_page[pteidx])) | |
1820 | continue; | |
1821 | ||
1822 | pte = pfn_pte(pfn, PAGE_KERNEL_EXEC); | |
1823 | pte_page[pteidx] = pte; | |
1824 | } | |
1825 | } | |
1826 | ||
1827 | for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE) | |
1828 | set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO); | |
1829 | ||
1830 | set_page_prot(pmd, PAGE_KERNEL_RO); | |
1831 | } | |
1832 | #endif | |
1833 | void __init xen_setup_machphys_mapping(void) | |
1834 | { | |
1835 | struct xen_machphys_mapping mapping; | |
1836 | ||
1837 | if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) { | |
1838 | machine_to_phys_mapping = (unsigned long *)mapping.v_start; | |
1839 | machine_to_phys_nr = mapping.max_mfn + 1; | |
1840 | } else { | |
1841 | machine_to_phys_nr = MACH2PHYS_NR_ENTRIES; | |
1842 | } | |
1843 | #ifdef CONFIG_X86_32 | |
1844 | WARN_ON((machine_to_phys_mapping + (machine_to_phys_nr - 1)) | |
1845 | < machine_to_phys_mapping); | |
1846 | #endif | |
1847 | } | |
1848 | ||
1849 | #ifdef CONFIG_X86_64 | |
1850 | static void __init convert_pfn_mfn(void *v) | |
1851 | { | |
1852 | pte_t *pte = v; | |
1853 | int i; | |
1854 | ||
1855 | /* All levels are converted the same way, so just treat them | |
1856 | as ptes. */ | |
1857 | for (i = 0; i < PTRS_PER_PTE; i++) | |
1858 | pte[i] = xen_make_pte(pte[i].pte); | |
1859 | } | |
1860 | static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end, | |
1861 | unsigned long addr) | |
1862 | { | |
1863 | if (*pt_base == PFN_DOWN(__pa(addr))) { | |
1864 | set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG); | |
1865 | clear_page((void *)addr); | |
1866 | (*pt_base)++; | |
1867 | } | |
1868 | if (*pt_end == PFN_DOWN(__pa(addr))) { | |
1869 | set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG); | |
1870 | clear_page((void *)addr); | |
1871 | (*pt_end)--; | |
1872 | } | |
1873 | } | |
1874 | /* | |
1875 | * Set up the initial kernel pagetable. | |
1876 | * | |
1877 | * We can construct this by grafting the Xen provided pagetable into | |
1878 | * head_64.S's preconstructed pagetables. We copy the Xen L2's into | |
1879 | * level2_ident_pgt, and level2_kernel_pgt. This means that only the | |
1880 | * kernel has a physical mapping to start with - but that's enough to | |
1881 | * get __va working. We need to fill in the rest of the physical | |
1882 | * mapping once some sort of allocator has been set up. | |
1883 | */ | |
1884 | void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn) | |
1885 | { | |
1886 | pud_t *l3; | |
1887 | pmd_t *l2; | |
1888 | unsigned long addr[3]; | |
1889 | unsigned long pt_base, pt_end; | |
1890 | unsigned i; | |
1891 | ||
1892 | /* max_pfn_mapped is the last pfn mapped in the initial memory | |
1893 | * mappings. Considering that on Xen after the kernel mappings we | |
1894 | * have the mappings of some pages that don't exist in pfn space, we | |
1895 | * set max_pfn_mapped to the last real pfn mapped. */ | |
1896 | if (xen_start_info->mfn_list < __START_KERNEL_map) | |
1897 | max_pfn_mapped = xen_start_info->first_p2m_pfn; | |
1898 | else | |
1899 | max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list)); | |
1900 | ||
1901 | pt_base = PFN_DOWN(__pa(xen_start_info->pt_base)); | |
1902 | pt_end = pt_base + xen_start_info->nr_pt_frames; | |
1903 | ||
1904 | /* Zap identity mapping */ | |
65ade2f8 | 1905 | init_top_pgt[0] = __pgd(0); |
7e0563de | 1906 | |
989513a7 | 1907 | /* Pre-constructed entries are in pfn, so convert to mfn */ |
d52888aa | 1908 | /* L4[273] -> level3_ident_pgt */ |
989513a7 | 1909 | /* L4[511] -> level3_kernel_pgt */ |
65ade2f8 | 1910 | convert_pfn_mfn(init_top_pgt); |
7e0563de | 1911 | |
989513a7 JG |
1912 | /* L3_i[0] -> level2_ident_pgt */ |
1913 | convert_pfn_mfn(level3_ident_pgt); | |
1914 | /* L3_k[510] -> level2_kernel_pgt */ | |
1915 | /* L3_k[511] -> level2_fixmap_pgt */ | |
1916 | convert_pfn_mfn(level3_kernel_pgt); | |
1917 | ||
05ab1d8a | 1918 | /* L3_k[511][508-FIXMAP_PMD_NUM ... 507] -> level1_fixmap_pgt */ |
989513a7 | 1919 | convert_pfn_mfn(level2_fixmap_pgt); |
7e0563de | 1920 | |
7e0563de VK |
1921 | /* We get [511][511] and have Xen's version of level2_kernel_pgt */ |
1922 | l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd); | |
1923 | l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud); | |
1924 | ||
1925 | addr[0] = (unsigned long)pgd; | |
1926 | addr[1] = (unsigned long)l3; | |
1927 | addr[2] = (unsigned long)l2; | |
d52888aa KS |
1928 | /* Graft it onto L4[273][0]. Note that we creating an aliasing problem: |
1929 | * Both L4[273][0] and L4[511][510] have entries that point to the same | |
7e0563de VK |
1930 | * L2 (PMD) tables. Meaning that if you modify it in __va space |
1931 | * it will be also modified in the __ka space! (But if you just | |
1932 | * modify the PMD table to point to other PTE's or none, then you | |
1933 | * are OK - which is what cleanup_highmap does) */ | |
1934 | copy_page(level2_ident_pgt, l2); | |
1935 | /* Graft it onto L4[511][510] */ | |
1936 | copy_page(level2_kernel_pgt, l2); | |
1937 | ||
2cc42bac JB |
1938 | /* |
1939 | * Zap execute permission from the ident map. Due to the sharing of | |
1940 | * L1 entries we need to do this in the L2. | |
1941 | */ | |
1942 | if (__supported_pte_mask & _PAGE_NX) { | |
1943 | for (i = 0; i < PTRS_PER_PMD; ++i) { | |
1944 | if (pmd_none(level2_ident_pgt[i])) | |
1945 | continue; | |
1946 | level2_ident_pgt[i] = pmd_set_flags(level2_ident_pgt[i], _PAGE_NX); | |
1947 | } | |
1948 | } | |
1949 | ||
7e0563de VK |
1950 | /* Copy the initial P->M table mappings if necessary. */ |
1951 | i = pgd_index(xen_start_info->mfn_list); | |
1952 | if (i && i < pgd_index(__START_KERNEL_map)) | |
65ade2f8 | 1953 | init_top_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i]; |
7e0563de | 1954 | |
989513a7 | 1955 | /* Make pagetable pieces RO */ |
65ade2f8 | 1956 | set_page_prot(init_top_pgt, PAGE_KERNEL_RO); |
989513a7 JG |
1957 | set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO); |
1958 | set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO); | |
1959 | set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO); | |
1960 | set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO); | |
1961 | set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO); | |
1962 | set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO); | |
05ab1d8a FT |
1963 | |
1964 | for (i = 0; i < FIXMAP_PMD_NUM; i++) { | |
1965 | set_page_prot(level1_fixmap_pgt + i * PTRS_PER_PTE, | |
1966 | PAGE_KERNEL_RO); | |
1967 | } | |
989513a7 JG |
1968 | |
1969 | /* Pin down new L4 */ | |
1970 | pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE, | |
65ade2f8 | 1971 | PFN_DOWN(__pa_symbol(init_top_pgt))); |
989513a7 JG |
1972 | |
1973 | /* Unpin Xen-provided one */ | |
1974 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
7e0563de | 1975 | |
989513a7 JG |
1976 | /* |
1977 | * At this stage there can be no user pgd, and no page structure to | |
1978 | * attach it to, so make sure we just set kernel pgd. | |
1979 | */ | |
1980 | xen_mc_batch(); | |
65ade2f8 | 1981 | __xen_write_cr3(true, __pa(init_top_pgt)); |
989513a7 | 1982 | xen_mc_issue(PARAVIRT_LAZY_CPU); |
7e0563de VK |
1983 | |
1984 | /* We can't that easily rip out L3 and L2, as the Xen pagetables are | |
1985 | * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ... for | |
1986 | * the initial domain. For guests using the toolstack, they are in: | |
1987 | * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only | |
1988 | * rip out the [L4] (pgd), but for guests we shave off three pages. | |
1989 | */ | |
1990 | for (i = 0; i < ARRAY_SIZE(addr); i++) | |
1991 | check_pt_base(&pt_base, &pt_end, addr[i]); | |
1992 | ||
1993 | /* Our (by three pages) smaller Xen pagetable that we are using */ | |
1994 | xen_pt_base = PFN_PHYS(pt_base); | |
1995 | xen_pt_size = (pt_end - pt_base) * PAGE_SIZE; | |
1996 | memblock_reserve(xen_pt_base, xen_pt_size); | |
1997 | ||
1998 | /* Revector the xen_start_info */ | |
1999 | xen_start_info = (struct start_info *)__va(__pa(xen_start_info)); | |
2000 | } | |
2001 | ||
2002 | /* | |
2003 | * Read a value from a physical address. | |
2004 | */ | |
2005 | static unsigned long __init xen_read_phys_ulong(phys_addr_t addr) | |
2006 | { | |
2007 | unsigned long *vaddr; | |
2008 | unsigned long val; | |
2009 | ||
2010 | vaddr = early_memremap_ro(addr, sizeof(val)); | |
2011 | val = *vaddr; | |
2012 | early_memunmap(vaddr, sizeof(val)); | |
2013 | return val; | |
2014 | } | |
2015 | ||
2016 | /* | |
2017 | * Translate a virtual address to a physical one without relying on mapped | |
69861e0a JG |
2018 | * page tables. Don't rely on big pages being aligned in (guest) physical |
2019 | * space! | |
7e0563de VK |
2020 | */ |
2021 | static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr) | |
2022 | { | |
2023 | phys_addr_t pa; | |
2024 | pgd_t pgd; | |
2025 | pud_t pud; | |
2026 | pmd_t pmd; | |
2027 | pte_t pte; | |
2028 | ||
6c690ee1 | 2029 | pa = read_cr3_pa(); |
7e0563de VK |
2030 | pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) * |
2031 | sizeof(pgd))); | |
2032 | if (!pgd_present(pgd)) | |
2033 | return 0; | |
2034 | ||
2035 | pa = pgd_val(pgd) & PTE_PFN_MASK; | |
2036 | pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) * | |
2037 | sizeof(pud))); | |
2038 | if (!pud_present(pud)) | |
2039 | return 0; | |
69861e0a | 2040 | pa = pud_val(pud) & PTE_PFN_MASK; |
7e0563de VK |
2041 | if (pud_large(pud)) |
2042 | return pa + (vaddr & ~PUD_MASK); | |
2043 | ||
2044 | pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) * | |
2045 | sizeof(pmd))); | |
2046 | if (!pmd_present(pmd)) | |
2047 | return 0; | |
69861e0a | 2048 | pa = pmd_val(pmd) & PTE_PFN_MASK; |
7e0563de VK |
2049 | if (pmd_large(pmd)) |
2050 | return pa + (vaddr & ~PMD_MASK); | |
2051 | ||
2052 | pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) * | |
2053 | sizeof(pte))); | |
2054 | if (!pte_present(pte)) | |
2055 | return 0; | |
2056 | pa = pte_pfn(pte) << PAGE_SHIFT; | |
2057 | ||
2058 | return pa | (vaddr & ~PAGE_MASK); | |
2059 | } | |
2060 | ||
2061 | /* | |
2062 | * Find a new area for the hypervisor supplied p2m list and relocate the p2m to | |
2063 | * this area. | |
2064 | */ | |
2065 | void __init xen_relocate_p2m(void) | |
2066 | { | |
773dd2fc | 2067 | phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys; |
7e0563de | 2068 | unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end; |
773dd2fc | 2069 | int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud; |
7e0563de VK |
2070 | pte_t *pt; |
2071 | pmd_t *pmd; | |
2072 | pud_t *pud; | |
7e0563de VK |
2073 | pgd_t *pgd; |
2074 | unsigned long *new_p2m; | |
7e0563de VK |
2075 | |
2076 | size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long)); | |
2077 | n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT; | |
2078 | n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT; | |
2079 | n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT; | |
2080 | n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT; | |
773dd2fc | 2081 | n_frames = n_pte + n_pt + n_pmd + n_pud; |
7e0563de VK |
2082 | |
2083 | new_area = xen_find_free_area(PFN_PHYS(n_frames)); | |
2084 | if (!new_area) { | |
2085 | xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n"); | |
2086 | BUG(); | |
2087 | } | |
2088 | ||
2089 | /* | |
2090 | * Setup the page tables for addressing the new p2m list. | |
2091 | * We have asked the hypervisor to map the p2m list at the user address | |
2092 | * PUD_SIZE. It may have done so, or it may have used a kernel space | |
2093 | * address depending on the Xen version. | |
2094 | * To avoid any possible virtual address collision, just use | |
2095 | * 2 * PUD_SIZE for the new area. | |
2096 | */ | |
773dd2fc | 2097 | pud_phys = new_area; |
7e0563de VK |
2098 | pmd_phys = pud_phys + PFN_PHYS(n_pud); |
2099 | pt_phys = pmd_phys + PFN_PHYS(n_pmd); | |
2100 | p2m_pfn = PFN_DOWN(pt_phys) + n_pt; | |
2101 | ||
6c690ee1 | 2102 | pgd = __va(read_cr3_pa()); |
7e0563de | 2103 | new_p2m = (unsigned long *)(2 * PGDIR_SIZE); |
773dd2fc KS |
2104 | for (idx_pud = 0; idx_pud < n_pud; idx_pud++) { |
2105 | pud = early_memremap(pud_phys, PAGE_SIZE); | |
2106 | clear_page(pud); | |
2107 | for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD); | |
2108 | idx_pmd++) { | |
2109 | pmd = early_memremap(pmd_phys, PAGE_SIZE); | |
2110 | clear_page(pmd); | |
2111 | for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD); | |
2112 | idx_pt++) { | |
2113 | pt = early_memremap(pt_phys, PAGE_SIZE); | |
2114 | clear_page(pt); | |
2115 | for (idx_pte = 0; | |
2116 | idx_pte < min(n_pte, PTRS_PER_PTE); | |
2117 | idx_pte++) { | |
2118 | set_pte(pt + idx_pte, | |
2119 | pfn_pte(p2m_pfn, PAGE_KERNEL)); | |
2120 | p2m_pfn++; | |
7e0563de | 2121 | } |
773dd2fc KS |
2122 | n_pte -= PTRS_PER_PTE; |
2123 | early_memunmap(pt, PAGE_SIZE); | |
2124 | make_lowmem_page_readonly(__va(pt_phys)); | |
2125 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, | |
2126 | PFN_DOWN(pt_phys)); | |
2127 | set_pmd(pmd + idx_pt, | |
2128 | __pmd(_PAGE_TABLE | pt_phys)); | |
2129 | pt_phys += PAGE_SIZE; | |
7e0563de | 2130 | } |
773dd2fc KS |
2131 | n_pt -= PTRS_PER_PMD; |
2132 | early_memunmap(pmd, PAGE_SIZE); | |
2133 | make_lowmem_page_readonly(__va(pmd_phys)); | |
2134 | pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE, | |
2135 | PFN_DOWN(pmd_phys)); | |
2136 | set_pud(pud + idx_pmd, __pud(_PAGE_TABLE | pmd_phys)); | |
2137 | pmd_phys += PAGE_SIZE; | |
7e0563de | 2138 | } |
773dd2fc KS |
2139 | n_pmd -= PTRS_PER_PUD; |
2140 | early_memunmap(pud, PAGE_SIZE); | |
2141 | make_lowmem_page_readonly(__va(pud_phys)); | |
2142 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys)); | |
2143 | set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys)); | |
2144 | pud_phys += PAGE_SIZE; | |
2145 | } | |
7e0563de VK |
2146 | |
2147 | /* Now copy the old p2m info to the new area. */ | |
2148 | memcpy(new_p2m, xen_p2m_addr, size); | |
2149 | xen_p2m_addr = new_p2m; | |
2150 | ||
2151 | /* Release the old p2m list and set new list info. */ | |
2152 | p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list)); | |
2153 | BUG_ON(!p2m_pfn); | |
2154 | p2m_pfn_end = p2m_pfn + PFN_DOWN(size); | |
2155 | ||
2156 | if (xen_start_info->mfn_list < __START_KERNEL_map) { | |
2157 | pfn = xen_start_info->first_p2m_pfn; | |
2158 | pfn_end = xen_start_info->first_p2m_pfn + | |
2159 | xen_start_info->nr_p2m_frames; | |
2160 | set_pgd(pgd + 1, __pgd(0)); | |
2161 | } else { | |
2162 | pfn = p2m_pfn; | |
2163 | pfn_end = p2m_pfn_end; | |
2164 | } | |
2165 | ||
2166 | memblock_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn)); | |
2167 | while (pfn < pfn_end) { | |
2168 | if (pfn == p2m_pfn) { | |
2169 | pfn = p2m_pfn_end; | |
2170 | continue; | |
2171 | } | |
2172 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); | |
2173 | pfn++; | |
2174 | } | |
2175 | ||
2176 | xen_start_info->mfn_list = (unsigned long)xen_p2m_addr; | |
2177 | xen_start_info->first_p2m_pfn = PFN_DOWN(new_area); | |
2178 | xen_start_info->nr_p2m_frames = n_frames; | |
2179 | } | |
2180 | ||
2181 | #else /* !CONFIG_X86_64 */ | |
2182 | static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD); | |
2183 | static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD); | |
75f2d3a0 JG |
2184 | RESERVE_BRK(fixup_kernel_pmd, PAGE_SIZE); |
2185 | RESERVE_BRK(fixup_kernel_pte, PAGE_SIZE); | |
7e0563de VK |
2186 | |
2187 | static void __init xen_write_cr3_init(unsigned long cr3) | |
2188 | { | |
2189 | unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir)); | |
2190 | ||
6c690ee1 | 2191 | BUG_ON(read_cr3_pa() != __pa(initial_page_table)); |
7e0563de VK |
2192 | BUG_ON(cr3 != __pa(swapper_pg_dir)); |
2193 | ||
2194 | /* | |
2195 | * We are switching to swapper_pg_dir for the first time (from | |
2196 | * initial_page_table) and therefore need to mark that page | |
2197 | * read-only and then pin it. | |
2198 | * | |
2199 | * Xen disallows sharing of kernel PMDs for PAE | |
2200 | * guests. Therefore we must copy the kernel PMD from | |
2201 | * initial_page_table into a new kernel PMD to be used in | |
2202 | * swapper_pg_dir. | |
2203 | */ | |
2204 | swapper_kernel_pmd = | |
2205 | extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE); | |
2206 | copy_page(swapper_kernel_pmd, initial_kernel_pmd); | |
2207 | swapper_pg_dir[KERNEL_PGD_BOUNDARY] = | |
2208 | __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT); | |
2209 | set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO); | |
2210 | ||
2211 | set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO); | |
2212 | xen_write_cr3(cr3); | |
2213 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn); | |
2214 | ||
2215 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, | |
2216 | PFN_DOWN(__pa(initial_page_table))); | |
2217 | set_page_prot(initial_page_table, PAGE_KERNEL); | |
2218 | set_page_prot(initial_kernel_pmd, PAGE_KERNEL); | |
2219 | ||
5c83511b | 2220 | pv_ops.mmu.write_cr3 = &xen_write_cr3; |
7e0563de VK |
2221 | } |
2222 | ||
2223 | /* | |
2224 | * For 32 bit domains xen_start_info->pt_base is the pgd address which might be | |
2225 | * not the first page table in the page table pool. | |
2226 | * Iterate through the initial page tables to find the real page table base. | |
2227 | */ | |
51ae2538 | 2228 | static phys_addr_t __init xen_find_pt_base(pmd_t *pmd) |
7e0563de VK |
2229 | { |
2230 | phys_addr_t pt_base, paddr; | |
2231 | unsigned pmdidx; | |
2232 | ||
2233 | pt_base = min(__pa(xen_start_info->pt_base), __pa(pmd)); | |
2234 | ||
2235 | for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++) | |
2236 | if (pmd_present(pmd[pmdidx]) && !pmd_large(pmd[pmdidx])) { | |
2237 | paddr = m2p(pmd[pmdidx].pmd); | |
2238 | pt_base = min(pt_base, paddr); | |
2239 | } | |
2240 | ||
2241 | return pt_base; | |
2242 | } | |
2243 | ||
2244 | void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn) | |
2245 | { | |
2246 | pmd_t *kernel_pmd; | |
2247 | ||
2248 | kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd); | |
2249 | ||
2250 | xen_pt_base = xen_find_pt_base(kernel_pmd); | |
2251 | xen_pt_size = xen_start_info->nr_pt_frames * PAGE_SIZE; | |
2252 | ||
2253 | initial_kernel_pmd = | |
2254 | extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE); | |
2255 | ||
2256 | max_pfn_mapped = PFN_DOWN(xen_pt_base + xen_pt_size + 512 * 1024); | |
2257 | ||
2258 | copy_page(initial_kernel_pmd, kernel_pmd); | |
2259 | ||
2260 | xen_map_identity_early(initial_kernel_pmd, max_pfn); | |
2261 | ||
2262 | copy_page(initial_page_table, pgd); | |
2263 | initial_page_table[KERNEL_PGD_BOUNDARY] = | |
2264 | __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT); | |
2265 | ||
2266 | set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO); | |
2267 | set_page_prot(initial_page_table, PAGE_KERNEL_RO); | |
2268 | set_page_prot(empty_zero_page, PAGE_KERNEL_RO); | |
2269 | ||
2270 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); | |
2271 | ||
2272 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, | |
2273 | PFN_DOWN(__pa(initial_page_table))); | |
2274 | xen_write_cr3(__pa(initial_page_table)); | |
2275 | ||
2276 | memblock_reserve(xen_pt_base, xen_pt_size); | |
2277 | } | |
2278 | #endif /* CONFIG_X86_64 */ | |
2279 | ||
2280 | void __init xen_reserve_special_pages(void) | |
2281 | { | |
2282 | phys_addr_t paddr; | |
2283 | ||
2284 | memblock_reserve(__pa(xen_start_info), PAGE_SIZE); | |
2285 | if (xen_start_info->store_mfn) { | |
2286 | paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn)); | |
2287 | memblock_reserve(paddr, PAGE_SIZE); | |
2288 | } | |
2289 | if (!xen_initial_domain()) { | |
2290 | paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn)); | |
2291 | memblock_reserve(paddr, PAGE_SIZE); | |
2292 | } | |
2293 | } | |
2294 | ||
2295 | void __init xen_pt_check_e820(void) | |
2296 | { | |
2297 | if (xen_is_e820_reserved(xen_pt_base, xen_pt_size)) { | |
2298 | xen_raw_console_write("Xen hypervisor allocated page table memory conflicts with E820 map\n"); | |
2299 | BUG(); | |
2300 | } | |
2301 | } | |
2302 | ||
2303 | static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss; | |
2304 | ||
2305 | static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot) | |
2306 | { | |
2307 | pte_t pte; | |
2308 | ||
2309 | phys >>= PAGE_SHIFT; | |
2310 | ||
2311 | switch (idx) { | |
2312 | case FIX_BTMAP_END ... FIX_BTMAP_BEGIN: | |
7e0563de VK |
2313 | #ifdef CONFIG_X86_32 |
2314 | case FIX_WP_TEST: | |
2315 | # ifdef CONFIG_HIGHMEM | |
2316 | case FIX_KMAP_BEGIN ... FIX_KMAP_END: | |
2317 | # endif | |
2318 | #elif defined(CONFIG_X86_VSYSCALL_EMULATION) | |
2319 | case VSYSCALL_PAGE: | |
2320 | #endif | |
2321 | case FIX_TEXT_POKE0: | |
2322 | case FIX_TEXT_POKE1: | |
7e0563de VK |
2323 | /* All local page mappings */ |
2324 | pte = pfn_pte(phys, prot); | |
2325 | break; | |
2326 | ||
2327 | #ifdef CONFIG_X86_LOCAL_APIC | |
2328 | case FIX_APIC_BASE: /* maps dummy local APIC */ | |
2329 | pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL); | |
2330 | break; | |
2331 | #endif | |
2332 | ||
2333 | #ifdef CONFIG_X86_IO_APIC | |
2334 | case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END: | |
2335 | /* | |
2336 | * We just don't map the IO APIC - all access is via | |
2337 | * hypercalls. Keep the address in the pte for reference. | |
2338 | */ | |
2339 | pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL); | |
2340 | break; | |
2341 | #endif | |
2342 | ||
2343 | case FIX_PARAVIRT_BOOTMAP: | |
2344 | /* This is an MFN, but it isn't an IO mapping from the | |
2345 | IO domain */ | |
2346 | pte = mfn_pte(phys, prot); | |
2347 | break; | |
2348 | ||
2349 | default: | |
2350 | /* By default, set_fixmap is used for hardware mappings */ | |
2351 | pte = mfn_pte(phys, prot); | |
2352 | break; | |
2353 | } | |
2354 | ||
2355 | __native_set_fixmap(idx, pte); | |
2356 | ||
2357 | #ifdef CONFIG_X86_VSYSCALL_EMULATION | |
2358 | /* Replicate changes to map the vsyscall page into the user | |
2359 | pagetable vsyscall mapping. */ | |
2360 | if (idx == VSYSCALL_PAGE) { | |
2361 | unsigned long vaddr = __fix_to_virt(idx); | |
2362 | set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte); | |
2363 | } | |
2364 | #endif | |
2365 | } | |
2366 | ||
2367 | static void __init xen_post_allocator_init(void) | |
2368 | { | |
5c83511b JG |
2369 | pv_ops.mmu.set_pte = xen_set_pte; |
2370 | pv_ops.mmu.set_pmd = xen_set_pmd; | |
2371 | pv_ops.mmu.set_pud = xen_set_pud; | |
773dd2fc | 2372 | #ifdef CONFIG_X86_64 |
5c83511b | 2373 | pv_ops.mmu.set_p4d = xen_set_p4d; |
7e0563de VK |
2374 | #endif |
2375 | ||
2376 | /* This will work as long as patching hasn't happened yet | |
2377 | (which it hasn't) */ | |
5c83511b JG |
2378 | pv_ops.mmu.alloc_pte = xen_alloc_pte; |
2379 | pv_ops.mmu.alloc_pmd = xen_alloc_pmd; | |
2380 | pv_ops.mmu.release_pte = xen_release_pte; | |
2381 | pv_ops.mmu.release_pmd = xen_release_pmd; | |
773dd2fc | 2382 | #ifdef CONFIG_X86_64 |
5c83511b JG |
2383 | pv_ops.mmu.alloc_pud = xen_alloc_pud; |
2384 | pv_ops.mmu.release_pud = xen_release_pud; | |
7e0563de | 2385 | #endif |
5c83511b | 2386 | pv_ops.mmu.make_pte = PV_CALLEE_SAVE(xen_make_pte); |
7e0563de VK |
2387 | |
2388 | #ifdef CONFIG_X86_64 | |
5c83511b | 2389 | pv_ops.mmu.write_cr3 = &xen_write_cr3; |
7e0563de | 2390 | #endif |
7e0563de VK |
2391 | } |
2392 | ||
2393 | static void xen_leave_lazy_mmu(void) | |
2394 | { | |
2395 | preempt_disable(); | |
2396 | xen_mc_flush(); | |
2397 | paravirt_leave_lazy_mmu(); | |
2398 | preempt_enable(); | |
2399 | } | |
2400 | ||
2401 | static const struct pv_mmu_ops xen_mmu_ops __initconst = { | |
2402 | .read_cr2 = xen_read_cr2, | |
2403 | .write_cr2 = xen_write_cr2, | |
2404 | ||
2405 | .read_cr3 = xen_read_cr3, | |
2406 | .write_cr3 = xen_write_cr3_init, | |
2407 | ||
2408 | .flush_tlb_user = xen_flush_tlb, | |
2409 | .flush_tlb_kernel = xen_flush_tlb, | |
1299ef1d | 2410 | .flush_tlb_one_user = xen_flush_tlb_one_user, |
7e0563de | 2411 | .flush_tlb_others = xen_flush_tlb_others, |
48a8b97c | 2412 | .tlb_remove_table = tlb_remove_table, |
7e0563de | 2413 | |
7e0563de VK |
2414 | .pgd_alloc = xen_pgd_alloc, |
2415 | .pgd_free = xen_pgd_free, | |
2416 | ||
2417 | .alloc_pte = xen_alloc_pte_init, | |
2418 | .release_pte = xen_release_pte_init, | |
2419 | .alloc_pmd = xen_alloc_pmd_init, | |
2420 | .release_pmd = xen_release_pmd_init, | |
2421 | ||
2422 | .set_pte = xen_set_pte_init, | |
2423 | .set_pte_at = xen_set_pte_at, | |
2424 | .set_pmd = xen_set_pmd_hyper, | |
2425 | ||
2426 | .ptep_modify_prot_start = __ptep_modify_prot_start, | |
2427 | .ptep_modify_prot_commit = __ptep_modify_prot_commit, | |
2428 | ||
2429 | .pte_val = PV_CALLEE_SAVE(xen_pte_val), | |
2430 | .pgd_val = PV_CALLEE_SAVE(xen_pgd_val), | |
2431 | ||
2432 | .make_pte = PV_CALLEE_SAVE(xen_make_pte_init), | |
2433 | .make_pgd = PV_CALLEE_SAVE(xen_make_pgd), | |
2434 | ||
2435 | #ifdef CONFIG_X86_PAE | |
2436 | .set_pte_atomic = xen_set_pte_atomic, | |
2437 | .pte_clear = xen_pte_clear, | |
2438 | .pmd_clear = xen_pmd_clear, | |
2439 | #endif /* CONFIG_X86_PAE */ | |
2440 | .set_pud = xen_set_pud_hyper, | |
2441 | ||
2442 | .make_pmd = PV_CALLEE_SAVE(xen_make_pmd), | |
2443 | .pmd_val = PV_CALLEE_SAVE(xen_pmd_val), | |
2444 | ||
773dd2fc | 2445 | #ifdef CONFIG_X86_64 |
7e0563de VK |
2446 | .pud_val = PV_CALLEE_SAVE(xen_pud_val), |
2447 | .make_pud = PV_CALLEE_SAVE(xen_make_pud), | |
2448 | .set_p4d = xen_set_p4d_hyper, | |
2449 | ||
2450 | .alloc_pud = xen_alloc_pmd_init, | |
2451 | .release_pud = xen_release_pmd_init, | |
b9952ec7 KS |
2452 | |
2453 | #if CONFIG_PGTABLE_LEVELS >= 5 | |
2454 | .p4d_val = PV_CALLEE_SAVE(xen_p4d_val), | |
2455 | .make_p4d = PV_CALLEE_SAVE(xen_make_p4d), | |
2456 | #endif | |
773dd2fc | 2457 | #endif /* CONFIG_X86_64 */ |
7e0563de VK |
2458 | |
2459 | .activate_mm = xen_activate_mm, | |
2460 | .dup_mmap = xen_dup_mmap, | |
2461 | .exit_mmap = xen_exit_mmap, | |
2462 | ||
2463 | .lazy_mode = { | |
2464 | .enter = paravirt_enter_lazy_mmu, | |
2465 | .leave = xen_leave_lazy_mmu, | |
2466 | .flush = paravirt_flush_lazy_mmu, | |
2467 | }, | |
2468 | ||
2469 | .set_fixmap = xen_set_fixmap, | |
2470 | }; | |
2471 | ||
2472 | void __init xen_init_mmu_ops(void) | |
2473 | { | |
2474 | x86_init.paging.pagetable_init = xen_pagetable_init; | |
6f84f8d1 | 2475 | x86_init.hyper.init_after_bootmem = xen_after_bootmem; |
7e0563de | 2476 | |
5c83511b | 2477 | pv_ops.mmu = xen_mmu_ops; |
7e0563de VK |
2478 | |
2479 | memset(dummy_mapping, 0xff, PAGE_SIZE); | |
2480 | } | |
2481 | ||
2482 | /* Protected by xen_reservation_lock. */ | |
2483 | #define MAX_CONTIG_ORDER 9 /* 2MB */ | |
2484 | static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER]; | |
2485 | ||
2486 | #define VOID_PTE (mfn_pte(0, __pgprot(0))) | |
2487 | static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order, | |
2488 | unsigned long *in_frames, | |
2489 | unsigned long *out_frames) | |
2490 | { | |
2491 | int i; | |
2492 | struct multicall_space mcs; | |
2493 | ||
2494 | xen_mc_batch(); | |
2495 | for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) { | |
2496 | mcs = __xen_mc_entry(0); | |
2497 | ||
2498 | if (in_frames) | |
2499 | in_frames[i] = virt_to_mfn(vaddr); | |
2500 | ||
2501 | MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0); | |
2502 | __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY); | |
2503 | ||
2504 | if (out_frames) | |
2505 | out_frames[i] = virt_to_pfn(vaddr); | |
2506 | } | |
2507 | xen_mc_issue(0); | |
2508 | } | |
2509 | ||
2510 | /* | |
2511 | * Update the pfn-to-mfn mappings for a virtual address range, either to | |
2512 | * point to an array of mfns, or contiguously from a single starting | |
2513 | * mfn. | |
2514 | */ | |
2515 | static void xen_remap_exchanged_ptes(unsigned long vaddr, int order, | |
2516 | unsigned long *mfns, | |
2517 | unsigned long first_mfn) | |
2518 | { | |
2519 | unsigned i, limit; | |
2520 | unsigned long mfn; | |
2521 | ||
2522 | xen_mc_batch(); | |
2523 | ||
2524 | limit = 1u << order; | |
2525 | for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) { | |
2526 | struct multicall_space mcs; | |
2527 | unsigned flags; | |
2528 | ||
2529 | mcs = __xen_mc_entry(0); | |
2530 | if (mfns) | |
2531 | mfn = mfns[i]; | |
2532 | else | |
2533 | mfn = first_mfn + i; | |
2534 | ||
2535 | if (i < (limit - 1)) | |
2536 | flags = 0; | |
2537 | else { | |
2538 | if (order == 0) | |
2539 | flags = UVMF_INVLPG | UVMF_ALL; | |
2540 | else | |
2541 | flags = UVMF_TLB_FLUSH | UVMF_ALL; | |
2542 | } | |
2543 | ||
2544 | MULTI_update_va_mapping(mcs.mc, vaddr, | |
2545 | mfn_pte(mfn, PAGE_KERNEL), flags); | |
2546 | ||
2547 | set_phys_to_machine(virt_to_pfn(vaddr), mfn); | |
2548 | } | |
2549 | ||
2550 | xen_mc_issue(0); | |
2551 | } | |
2552 | ||
2553 | /* | |
2554 | * Perform the hypercall to exchange a region of our pfns to point to | |
2555 | * memory with the required contiguous alignment. Takes the pfns as | |
2556 | * input, and populates mfns as output. | |
2557 | * | |
2558 | * Returns a success code indicating whether the hypervisor was able to | |
2559 | * satisfy the request or not. | |
2560 | */ | |
2561 | static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in, | |
2562 | unsigned long *pfns_in, | |
2563 | unsigned long extents_out, | |
2564 | unsigned int order_out, | |
2565 | unsigned long *mfns_out, | |
2566 | unsigned int address_bits) | |
2567 | { | |
2568 | long rc; | |
2569 | int success; | |
2570 | ||
2571 | struct xen_memory_exchange exchange = { | |
2572 | .in = { | |
2573 | .nr_extents = extents_in, | |
2574 | .extent_order = order_in, | |
2575 | .extent_start = pfns_in, | |
2576 | .domid = DOMID_SELF | |
2577 | }, | |
2578 | .out = { | |
2579 | .nr_extents = extents_out, | |
2580 | .extent_order = order_out, | |
2581 | .extent_start = mfns_out, | |
2582 | .address_bits = address_bits, | |
2583 | .domid = DOMID_SELF | |
2584 | } | |
2585 | }; | |
2586 | ||
2587 | BUG_ON(extents_in << order_in != extents_out << order_out); | |
2588 | ||
2589 | rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange); | |
2590 | success = (exchange.nr_exchanged == extents_in); | |
2591 | ||
2592 | BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0))); | |
2593 | BUG_ON(success && (rc != 0)); | |
2594 | ||
2595 | return success; | |
2596 | } | |
2597 | ||
2598 | int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order, | |
2599 | unsigned int address_bits, | |
2600 | dma_addr_t *dma_handle) | |
2601 | { | |
2602 | unsigned long *in_frames = discontig_frames, out_frame; | |
2603 | unsigned long flags; | |
2604 | int success; | |
2605 | unsigned long vstart = (unsigned long)phys_to_virt(pstart); | |
2606 | ||
2607 | /* | |
2608 | * Currently an auto-translated guest will not perform I/O, nor will | |
2609 | * it require PAE page directories below 4GB. Therefore any calls to | |
2610 | * this function are redundant and can be ignored. | |
2611 | */ | |
2612 | ||
7e0563de VK |
2613 | if (unlikely(order > MAX_CONTIG_ORDER)) |
2614 | return -ENOMEM; | |
2615 | ||
2616 | memset((void *) vstart, 0, PAGE_SIZE << order); | |
2617 | ||
2618 | spin_lock_irqsave(&xen_reservation_lock, flags); | |
2619 | ||
2620 | /* 1. Zap current PTEs, remembering MFNs. */ | |
2621 | xen_zap_pfn_range(vstart, order, in_frames, NULL); | |
2622 | ||
2623 | /* 2. Get a new contiguous memory extent. */ | |
2624 | out_frame = virt_to_pfn(vstart); | |
2625 | success = xen_exchange_memory(1UL << order, 0, in_frames, | |
2626 | 1, order, &out_frame, | |
2627 | address_bits); | |
2628 | ||
2629 | /* 3. Map the new extent in place of old pages. */ | |
2630 | if (success) | |
2631 | xen_remap_exchanged_ptes(vstart, order, NULL, out_frame); | |
2632 | else | |
2633 | xen_remap_exchanged_ptes(vstart, order, in_frames, 0); | |
2634 | ||
2635 | spin_unlock_irqrestore(&xen_reservation_lock, flags); | |
2636 | ||
2637 | *dma_handle = virt_to_machine(vstart).maddr; | |
2638 | return success ? 0 : -ENOMEM; | |
2639 | } | |
2640 | EXPORT_SYMBOL_GPL(xen_create_contiguous_region); | |
2641 | ||
2642 | void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order) | |
2643 | { | |
2644 | unsigned long *out_frames = discontig_frames, in_frame; | |
2645 | unsigned long flags; | |
2646 | int success; | |
2647 | unsigned long vstart; | |
2648 | ||
7e0563de VK |
2649 | if (unlikely(order > MAX_CONTIG_ORDER)) |
2650 | return; | |
2651 | ||
2652 | vstart = (unsigned long)phys_to_virt(pstart); | |
2653 | memset((void *) vstart, 0, PAGE_SIZE << order); | |
2654 | ||
2655 | spin_lock_irqsave(&xen_reservation_lock, flags); | |
2656 | ||
2657 | /* 1. Find start MFN of contiguous extent. */ | |
2658 | in_frame = virt_to_mfn(vstart); | |
2659 | ||
2660 | /* 2. Zap current PTEs. */ | |
2661 | xen_zap_pfn_range(vstart, order, NULL, out_frames); | |
2662 | ||
2663 | /* 3. Do the exchange for non-contiguous MFNs. */ | |
2664 | success = xen_exchange_memory(1, order, &in_frame, 1UL << order, | |
2665 | 0, out_frames, 0); | |
2666 | ||
2667 | /* 4. Map new pages in place of old pages. */ | |
2668 | if (success) | |
2669 | xen_remap_exchanged_ptes(vstart, order, out_frames, 0); | |
2670 | else | |
2671 | xen_remap_exchanged_ptes(vstart, order, NULL, in_frame); | |
2672 | ||
2673 | spin_unlock_irqrestore(&xen_reservation_lock, flags); | |
2674 | } | |
2675 | EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region); | |
29985b09 | 2676 | |
f030aade JG |
2677 | static noinline void xen_flush_tlb_all(void) |
2678 | { | |
2679 | struct mmuext_op *op; | |
2680 | struct multicall_space mcs; | |
2681 | ||
2682 | preempt_disable(); | |
2683 | ||
2684 | mcs = xen_mc_entry(sizeof(*op)); | |
2685 | ||
2686 | op = mcs.args; | |
2687 | op->cmd = MMUEXT_TLB_FLUSH_ALL; | |
2688 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
2689 | ||
2690 | xen_mc_issue(PARAVIRT_LAZY_MMU); | |
2691 | ||
2692 | preempt_enable(); | |
2693 | } | |
2694 | ||
2695 | #define REMAP_BATCH_SIZE 16 | |
2696 | ||
2697 | struct remap_data { | |
2698 | xen_pfn_t *pfn; | |
2699 | bool contiguous; | |
2700 | bool no_translate; | |
2701 | pgprot_t prot; | |
2702 | struct mmu_update *mmu_update; | |
2703 | }; | |
2704 | ||
2705 | static int remap_area_pfn_pte_fn(pte_t *ptep, pgtable_t token, | |
2706 | unsigned long addr, void *data) | |
2707 | { | |
2708 | struct remap_data *rmd = data; | |
2709 | pte_t pte = pte_mkspecial(mfn_pte(*rmd->pfn, rmd->prot)); | |
2710 | ||
2711 | /* | |
2712 | * If we have a contiguous range, just update the pfn itself, | |
2713 | * else update pointer to be "next pfn". | |
2714 | */ | |
2715 | if (rmd->contiguous) | |
2716 | (*rmd->pfn)++; | |
2717 | else | |
2718 | rmd->pfn++; | |
2719 | ||
2720 | rmd->mmu_update->ptr = virt_to_machine(ptep).maddr; | |
2721 | rmd->mmu_update->ptr |= rmd->no_translate ? | |
2722 | MMU_PT_UPDATE_NO_TRANSLATE : | |
2723 | MMU_NORMAL_PT_UPDATE; | |
2724 | rmd->mmu_update->val = pte_val_ma(pte); | |
2725 | rmd->mmu_update++; | |
2726 | ||
2727 | return 0; | |
2728 | } | |
2729 | ||
2730 | int xen_remap_pfn(struct vm_area_struct *vma, unsigned long addr, | |
2731 | xen_pfn_t *pfn, int nr, int *err_ptr, pgprot_t prot, | |
2732 | unsigned int domid, bool no_translate, struct page **pages) | |
2733 | { | |
2734 | int err = 0; | |
2735 | struct remap_data rmd; | |
2736 | struct mmu_update mmu_update[REMAP_BATCH_SIZE]; | |
2737 | unsigned long range; | |
2738 | int mapped = 0; | |
2739 | ||
2740 | BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_IO)) == (VM_PFNMAP | VM_IO))); | |
2741 | ||
2742 | rmd.pfn = pfn; | |
2743 | rmd.prot = prot; | |
2744 | /* | |
2745 | * We use the err_ptr to indicate if there we are doing a contiguous | |
2746 | * mapping or a discontigious mapping. | |
2747 | */ | |
2748 | rmd.contiguous = !err_ptr; | |
2749 | rmd.no_translate = no_translate; | |
2750 | ||
2751 | while (nr) { | |
2752 | int index = 0; | |
2753 | int done = 0; | |
2754 | int batch = min(REMAP_BATCH_SIZE, nr); | |
2755 | int batch_left = batch; | |
2756 | ||
2757 | range = (unsigned long)batch << PAGE_SHIFT; | |
2758 | ||
2759 | rmd.mmu_update = mmu_update; | |
2760 | err = apply_to_page_range(vma->vm_mm, addr, range, | |
2761 | remap_area_pfn_pte_fn, &rmd); | |
2762 | if (err) | |
2763 | goto out; | |
2764 | ||
2765 | /* | |
2766 | * We record the error for each page that gives an error, but | |
2767 | * continue mapping until the whole set is done | |
2768 | */ | |
2769 | do { | |
2770 | int i; | |
2771 | ||
2772 | err = HYPERVISOR_mmu_update(&mmu_update[index], | |
2773 | batch_left, &done, domid); | |
2774 | ||
2775 | /* | |
2776 | * @err_ptr may be the same buffer as @gfn, so | |
2777 | * only clear it after each chunk of @gfn is | |
2778 | * used. | |
2779 | */ | |
2780 | if (err_ptr) { | |
2781 | for (i = index; i < index + done; i++) | |
2782 | err_ptr[i] = 0; | |
2783 | } | |
2784 | if (err < 0) { | |
2785 | if (!err_ptr) | |
2786 | goto out; | |
2787 | err_ptr[i] = err; | |
2788 | done++; /* Skip failed frame. */ | |
2789 | } else | |
2790 | mapped += done; | |
2791 | batch_left -= done; | |
2792 | index += done; | |
2793 | } while (batch_left); | |
2794 | ||
2795 | nr -= batch; | |
2796 | addr += range; | |
2797 | if (err_ptr) | |
2798 | err_ptr += batch; | |
2799 | cond_resched(); | |
2800 | } | |
2801 | out: | |
2802 | ||
2803 | xen_flush_tlb_all(); | |
2804 | ||
2805 | return err < 0 ? err : mapped; | |
2806 | } | |
2807 | EXPORT_SYMBOL_GPL(xen_remap_pfn); | |
2808 | ||
29985b09 JG |
2809 | #ifdef CONFIG_KEXEC_CORE |
2810 | phys_addr_t paddr_vmcoreinfo_note(void) | |
2811 | { | |
2812 | if (xen_pv_domain()) | |
203e9e41 | 2813 | return virt_to_machine(vmcoreinfo_note).maddr; |
29985b09 | 2814 | else |
203e9e41 | 2815 | return __pa(vmcoreinfo_note); |
29985b09 JG |
2816 | } |
2817 | #endif /* CONFIG_KEXEC_CORE */ |