1 | // SPDX-License-Identifier: GPL-2.0 |
2 | |
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/debugfs.h> |
45 | #include <linux/bug.h> |
46 | #include <linux/vmalloc.h> |
47 | #include <linux/export.h> |
48 | #include <linux/init.h> |
49 | #include <linux/gfp.h> |
50 | #include <linux/memblock.h> |
51 | #include <linux/seq_file.h> |
52 | #include <linux/crash_dump.h> |
53 | #include <linux/pgtable.h> |
54 | #ifdef CONFIG_KEXEC_CORE |
55 | #include <linux/kexec.h> |
56 | #endif |
57 | |
58 | #include <trace/events/xen.h> |
59 | |
60 | #include <asm/tlbflush.h> |
61 | #include <asm/fixmap.h> |
62 | #include <asm/mmu_context.h> |
63 | #include <asm/setup.h> |
64 | #include <asm/paravirt.h> |
65 | #include <asm/e820/api.h> |
66 | #include <asm/linkage.h> |
67 | #include <asm/page.h> |
68 | #include <asm/init.h> |
69 | #include <asm/memtype.h> |
70 | #include <asm/smp.h> |
71 | #include <asm/tlb.h> |
72 | |
73 | #include <asm/xen/hypercall.h> |
74 | #include <asm/xen/hypervisor.h> |
75 | |
76 | #include <xen/xen.h> |
77 | #include <xen/page.h> |
78 | #include <xen/interface/xen.h> |
79 | #include <xen/interface/hvm/hvm_op.h> |
80 | #include <xen/interface/version.h> |
81 | #include <xen/interface/memory.h> |
82 | #include <xen/hvc-console.h> |
83 | #include <xen/swiotlb-xen.h> |
84 | |
85 | #include "multicalls.h" |
86 | #include "mmu.h" |
87 | #include "debugfs.h" |
88 | |
89 | /* |
90 | * Prototypes for functions called via PV_CALLEE_SAVE_REGS_THUNK() in order |
91 | * to avoid warnings with "-Wmissing-prototypes". |
92 | */ |
93 | pteval_t xen_pte_val(pte_t pte); |
94 | pgdval_t xen_pgd_val(pgd_t pgd); |
95 | pmdval_t xen_pmd_val(pmd_t pmd); |
96 | pudval_t xen_pud_val(pud_t pud); |
97 | p4dval_t xen_p4d_val(p4d_t p4d); |
98 | pte_t xen_make_pte(pteval_t pte); |
99 | pgd_t xen_make_pgd(pgdval_t pgd); |
100 | pmd_t xen_make_pmd(pmdval_t pmd); |
101 | pud_t xen_make_pud(pudval_t pud); |
102 | p4d_t xen_make_p4d(p4dval_t p4d); |
103 | pte_t xen_make_pte_init(pteval_t pte); |
104 | |
105 | #ifdef CONFIG_X86_VSYSCALL_EMULATION |
106 | /* l3 pud for userspace vsyscall mapping */ |
107 | static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss; |
108 | #endif |
109 | |
110 | /* |
111 | * Protects atomic reservation decrease/increase against concurrent increases. |
112 | * Also protects non-atomic updates of current_pages and balloon lists. |
113 | */ |
114 | static DEFINE_SPINLOCK(xen_reservation_lock); |
115 | |
116 | /* |
117 | * Note about cr3 (pagetable base) values: |
118 | * |
119 | * xen_cr3 contains the current logical cr3 value; it contains the |
120 | * last set cr3. This may not be the current effective cr3, because |
121 | * its update may be being lazily deferred. However, a vcpu looking |
122 | * at its own cr3 can use this value knowing that it everything will |
123 | * be self-consistent. |
124 | * |
125 | * xen_current_cr3 contains the actual vcpu cr3; it is set once the |
126 | * hypercall to set the vcpu cr3 is complete (so it may be a little |
127 | * out of date, but it will never be set early). If one vcpu is |
128 | * looking at another vcpu's cr3 value, it should use this variable. |
129 | */ |
130 | DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */ |
131 | DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */ |
132 | |
133 | static phys_addr_t xen_pt_base, xen_pt_size __initdata; |
134 | |
135 | static DEFINE_STATIC_KEY_FALSE(xen_struct_pages_ready); |
136 | |
137 | /* |
138 | * Just beyond the highest usermode address. STACK_TOP_MAX has a |
139 | * redzone above it, so round it up to a PGD boundary. |
140 | */ |
141 | #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK) |
142 | |
143 | void make_lowmem_page_readonly(void *vaddr) |
144 | { |
145 | pte_t *pte, ptev; |
146 | unsigned long address = (unsigned long)vaddr; |
147 | unsigned int level; |
148 | |
149 | pte = lookup_address(address, level: &level); |
150 | if (pte == NULL) |
151 | return; /* vaddr missing */ |
152 | |
153 | ptev = pte_wrprotect(pte: *pte); |
154 | |
155 | if (HYPERVISOR_update_va_mapping(va: address, new_val: ptev, flags: 0)) |
156 | BUG(); |
157 | } |
158 | |
159 | void make_lowmem_page_readwrite(void *vaddr) |
160 | { |
161 | pte_t *pte, ptev; |
162 | unsigned long address = (unsigned long)vaddr; |
163 | unsigned int level; |
164 | |
165 | pte = lookup_address(address, level: &level); |
166 | if (pte == NULL) |
167 | return; /* vaddr missing */ |
168 | |
169 | ptev = pte_mkwrite_novma(pte: *pte); |
170 | |
171 | if (HYPERVISOR_update_va_mapping(va: address, new_val: ptev, flags: 0)) |
172 | BUG(); |
173 | } |
174 | |
175 | |
176 | /* |
177 | * During early boot all page table pages are pinned, but we do not have struct |
178 | * pages, so return true until struct pages are ready. |
179 | */ |
180 | static bool xen_page_pinned(void *ptr) |
181 | { |
182 | if (static_branch_likely(&xen_struct_pages_ready)) { |
183 | struct page *page = virt_to_page(ptr); |
184 | |
185 | return PagePinned(page); |
186 | } |
187 | return true; |
188 | } |
189 | |
190 | static void xen_extend_mmu_update(const struct mmu_update *update) |
191 | { |
192 | struct multicall_space mcs; |
193 | struct mmu_update *u; |
194 | |
195 | mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, arg_size: sizeof(*u)); |
196 | |
197 | if (mcs.mc != NULL) { |
198 | mcs.mc->args[1]++; |
199 | } else { |
200 | mcs = __xen_mc_entry(args: sizeof(*u)); |
201 | MULTI_mmu_update(mcl: mcs.mc, req: mcs.args, count: 1, NULL, DOMID_SELF); |
202 | } |
203 | |
204 | u = mcs.args; |
205 | *u = *update; |
206 | } |
207 | |
208 | static void xen_extend_mmuext_op(const struct mmuext_op *op) |
209 | { |
210 | struct multicall_space mcs; |
211 | struct mmuext_op *u; |
212 | |
213 | mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, arg_size: sizeof(*u)); |
214 | |
215 | if (mcs.mc != NULL) { |
216 | mcs.mc->args[1]++; |
217 | } else { |
218 | mcs = __xen_mc_entry(args: sizeof(*u)); |
219 | MULTI_mmuext_op(mcl: mcs.mc, op: mcs.args, count: 1, NULL, DOMID_SELF); |
220 | } |
221 | |
222 | u = mcs.args; |
223 | *u = *op; |
224 | } |
225 | |
226 | static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val) |
227 | { |
228 | struct mmu_update u; |
229 | |
230 | preempt_disable(); |
231 | |
232 | xen_mc_batch(); |
233 | |
234 | /* ptr may be ioremapped for 64-bit pagetable setup */ |
235 | u.ptr = arbitrary_virt_to_machine(address: ptr).maddr; |
236 | u.val = pmd_val_ma(val); |
237 | xen_extend_mmu_update(update: &u); |
238 | |
239 | xen_mc_issue(mode: XEN_LAZY_MMU); |
240 | |
241 | preempt_enable(); |
242 | } |
243 | |
244 | static void xen_set_pmd(pmd_t *ptr, pmd_t val) |
245 | { |
246 | trace_xen_mmu_set_pmd(pmdp: ptr, pmdval: val); |
247 | |
248 | /* If page is not pinned, we can just update the entry |
249 | directly */ |
250 | if (!xen_page_pinned(ptr)) { |
251 | *ptr = val; |
252 | return; |
253 | } |
254 | |
255 | xen_set_pmd_hyper(ptr, val); |
256 | } |
257 | |
258 | /* |
259 | * Associate a virtual page frame with a given physical page frame |
260 | * and protection flags for that frame. |
261 | */ |
262 | void __init set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags) |
263 | { |
264 | if (HYPERVISOR_update_va_mapping(va: vaddr, new_val: mfn_pte(page_nr: mfn, pgprot: flags), |
265 | UVMF_INVLPG)) |
266 | BUG(); |
267 | } |
268 | |
269 | static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval) |
270 | { |
271 | struct mmu_update u; |
272 | |
273 | if (xen_get_lazy_mode() != XEN_LAZY_MMU) |
274 | return false; |
275 | |
276 | xen_mc_batch(); |
277 | |
278 | u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE; |
279 | u.val = pte_val_ma(pte: pteval); |
280 | xen_extend_mmu_update(update: &u); |
281 | |
282 | xen_mc_issue(mode: XEN_LAZY_MMU); |
283 | |
284 | return true; |
285 | } |
286 | |
287 | static inline void __xen_set_pte(pte_t *ptep, pte_t pteval) |
288 | { |
289 | if (!xen_batched_set_pte(ptep, pteval)) { |
290 | /* |
291 | * Could call native_set_pte() here and trap and |
292 | * emulate the PTE write, but a hypercall is much cheaper. |
293 | */ |
294 | struct mmu_update u; |
295 | |
296 | u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE; |
297 | u.val = pte_val_ma(pte: pteval); |
298 | HYPERVISOR_mmu_update(req: &u, count: 1, NULL, DOMID_SELF); |
299 | } |
300 | } |
301 | |
302 | static void xen_set_pte(pte_t *ptep, pte_t pteval) |
303 | { |
304 | trace_xen_mmu_set_pte(ptep, pteval); |
305 | __xen_set_pte(ptep, pteval); |
306 | } |
307 | |
308 | pte_t xen_ptep_modify_prot_start(struct vm_area_struct *vma, |
309 | unsigned long addr, pte_t *ptep) |
310 | { |
311 | /* Just return the pte as-is. We preserve the bits on commit */ |
312 | trace_xen_mmu_ptep_modify_prot_start(mm: vma->vm_mm, addr, ptep, pteval: *ptep); |
313 | return *ptep; |
314 | } |
315 | |
316 | void xen_ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr, |
317 | pte_t *ptep, pte_t pte) |
318 | { |
319 | struct mmu_update u; |
320 | |
321 | trace_xen_mmu_ptep_modify_prot_commit(mm: vma->vm_mm, addr, ptep, pteval: pte); |
322 | xen_mc_batch(); |
323 | |
324 | u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD; |
325 | u.val = pte_val_ma(pte); |
326 | xen_extend_mmu_update(update: &u); |
327 | |
328 | xen_mc_issue(mode: XEN_LAZY_MMU); |
329 | } |
330 | |
331 | /* Assume pteval_t is equivalent to all the other *val_t types. */ |
332 | static pteval_t pte_mfn_to_pfn(pteval_t val) |
333 | { |
334 | if (val & _PAGE_PRESENT) { |
335 | unsigned long mfn = (val & XEN_PTE_MFN_MASK) >> PAGE_SHIFT; |
336 | unsigned long pfn = mfn_to_pfn(mfn); |
337 | |
338 | pteval_t flags = val & PTE_FLAGS_MASK; |
339 | if (unlikely(pfn == ~0)) |
340 | val = flags & ~_PAGE_PRESENT; |
341 | else |
342 | val = ((pteval_t)pfn << PAGE_SHIFT) | flags; |
343 | } |
344 | |
345 | return val; |
346 | } |
347 | |
348 | static pteval_t pte_pfn_to_mfn(pteval_t val) |
349 | { |
350 | if (val & _PAGE_PRESENT) { |
351 | unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT; |
352 | pteval_t flags = val & PTE_FLAGS_MASK; |
353 | unsigned long mfn; |
354 | |
355 | mfn = __pfn_to_mfn(pfn); |
356 | |
357 | /* |
358 | * If there's no mfn for the pfn, then just create an |
359 | * empty non-present pte. Unfortunately this loses |
360 | * information about the original pfn, so |
361 | * pte_mfn_to_pfn is asymmetric. |
362 | */ |
363 | if (unlikely(mfn == INVALID_P2M_ENTRY)) { |
364 | mfn = 0; |
365 | flags = 0; |
366 | } else |
367 | mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT); |
368 | val = ((pteval_t)mfn << PAGE_SHIFT) | flags; |
369 | } |
370 | |
371 | return val; |
372 | } |
373 | |
374 | __visible pteval_t xen_pte_val(pte_t pte) |
375 | { |
376 | pteval_t pteval = pte.pte; |
377 | |
378 | return pte_mfn_to_pfn(val: pteval); |
379 | } |
380 | PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val); |
381 | |
382 | __visible pgdval_t xen_pgd_val(pgd_t pgd) |
383 | { |
384 | return pte_mfn_to_pfn(val: pgd.pgd); |
385 | } |
386 | PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val); |
387 | |
388 | __visible pte_t xen_make_pte(pteval_t pte) |
389 | { |
390 | pte = pte_pfn_to_mfn(val: pte); |
391 | |
392 | return native_make_pte(val: pte); |
393 | } |
394 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte); |
395 | |
396 | __visible pgd_t xen_make_pgd(pgdval_t pgd) |
397 | { |
398 | pgd = pte_pfn_to_mfn(val: pgd); |
399 | return native_make_pgd(val: pgd); |
400 | } |
401 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd); |
402 | |
403 | __visible pmdval_t xen_pmd_val(pmd_t pmd) |
404 | { |
405 | return pte_mfn_to_pfn(val: pmd.pmd); |
406 | } |
407 | PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val); |
408 | |
409 | static void xen_set_pud_hyper(pud_t *ptr, pud_t val) |
410 | { |
411 | struct mmu_update u; |
412 | |
413 | preempt_disable(); |
414 | |
415 | xen_mc_batch(); |
416 | |
417 | /* ptr may be ioremapped for 64-bit pagetable setup */ |
418 | u.ptr = arbitrary_virt_to_machine(address: ptr).maddr; |
419 | u.val = pud_val_ma(val); |
420 | xen_extend_mmu_update(update: &u); |
421 | |
422 | xen_mc_issue(mode: XEN_LAZY_MMU); |
423 | |
424 | preempt_enable(); |
425 | } |
426 | |
427 | static void xen_set_pud(pud_t *ptr, pud_t val) |
428 | { |
429 | trace_xen_mmu_set_pud(pudp: ptr, pudval: val); |
430 | |
431 | /* If page is not pinned, we can just update the entry |
432 | directly */ |
433 | if (!xen_page_pinned(ptr)) { |
434 | *ptr = val; |
435 | return; |
436 | } |
437 | |
438 | xen_set_pud_hyper(ptr, val); |
439 | } |
440 | |
441 | __visible pmd_t xen_make_pmd(pmdval_t pmd) |
442 | { |
443 | pmd = pte_pfn_to_mfn(val: pmd); |
444 | return native_make_pmd(val: pmd); |
445 | } |
446 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd); |
447 | |
448 | __visible pudval_t xen_pud_val(pud_t pud) |
449 | { |
450 | return pte_mfn_to_pfn(val: pud.pud); |
451 | } |
452 | PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val); |
453 | |
454 | __visible pud_t xen_make_pud(pudval_t pud) |
455 | { |
456 | pud = pte_pfn_to_mfn(val: pud); |
457 | |
458 | return native_make_pud(val: pud); |
459 | } |
460 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud); |
461 | |
462 | static pgd_t *xen_get_user_pgd(pgd_t *pgd) |
463 | { |
464 | pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK); |
465 | unsigned offset = pgd - pgd_page; |
466 | pgd_t *user_ptr = NULL; |
467 | |
468 | if (offset < pgd_index(USER_LIMIT)) { |
469 | struct page *page = virt_to_page(pgd_page); |
470 | user_ptr = (pgd_t *)page->private; |
471 | if (user_ptr) |
472 | user_ptr += offset; |
473 | } |
474 | |
475 | return user_ptr; |
476 | } |
477 | |
478 | static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val) |
479 | { |
480 | struct mmu_update u; |
481 | |
482 | u.ptr = virt_to_machine(ptr).maddr; |
483 | u.val = p4d_val_ma(val); |
484 | xen_extend_mmu_update(update: &u); |
485 | } |
486 | |
487 | /* |
488 | * Raw hypercall-based set_p4d, intended for in early boot before |
489 | * there's a page structure. This implies: |
490 | * 1. The only existing pagetable is the kernel's |
491 | * 2. It is always pinned |
492 | * 3. It has no user pagetable attached to it |
493 | */ |
494 | static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val) |
495 | { |
496 | preempt_disable(); |
497 | |
498 | xen_mc_batch(); |
499 | |
500 | __xen_set_p4d_hyper(ptr, val); |
501 | |
502 | xen_mc_issue(mode: XEN_LAZY_MMU); |
503 | |
504 | preempt_enable(); |
505 | } |
506 | |
507 | static void xen_set_p4d(p4d_t *ptr, p4d_t val) |
508 | { |
509 | pgd_t *user_ptr = xen_get_user_pgd(pgd: (pgd_t *)ptr); |
510 | pgd_t pgd_val; |
511 | |
512 | trace_xen_mmu_set_p4d(p4dp: ptr, user_p4dp: (p4d_t *)user_ptr, p4dval: val); |
513 | |
514 | /* If page is not pinned, we can just update the entry |
515 | directly */ |
516 | if (!xen_page_pinned(ptr)) { |
517 | *ptr = val; |
518 | if (user_ptr) { |
519 | WARN_ON(xen_page_pinned(user_ptr)); |
520 | pgd_val.pgd = p4d_val_ma(val); |
521 | *user_ptr = pgd_val; |
522 | } |
523 | return; |
524 | } |
525 | |
526 | /* If it's pinned, then we can at least batch the kernel and |
527 | user updates together. */ |
528 | xen_mc_batch(); |
529 | |
530 | __xen_set_p4d_hyper(ptr, val); |
531 | if (user_ptr) |
532 | __xen_set_p4d_hyper(ptr: (p4d_t *)user_ptr, val); |
533 | |
534 | xen_mc_issue(mode: XEN_LAZY_MMU); |
535 | } |
536 | |
537 | #if CONFIG_PGTABLE_LEVELS >= 5 |
538 | __visible p4dval_t xen_p4d_val(p4d_t p4d) |
539 | { |
540 | return pte_mfn_to_pfn(val: p4d.p4d); |
541 | } |
542 | PV_CALLEE_SAVE_REGS_THUNK(xen_p4d_val); |
543 | |
544 | __visible p4d_t xen_make_p4d(p4dval_t p4d) |
545 | { |
546 | p4d = pte_pfn_to_mfn(val: p4d); |
547 | |
548 | return native_make_p4d(val: p4d); |
549 | } |
550 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_p4d); |
551 | #endif /* CONFIG_PGTABLE_LEVELS >= 5 */ |
552 | |
553 | static void xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd, |
554 | void (*func)(struct mm_struct *mm, struct page *, |
555 | enum pt_level), |
556 | bool last, unsigned long limit) |
557 | { |
558 | int i, nr; |
559 | |
560 | nr = last ? pmd_index(address: limit) + 1 : PTRS_PER_PMD; |
561 | for (i = 0; i < nr; i++) { |
562 | if (!pmd_none(pmd: pmd[i])) |
563 | (*func)(mm, pmd_page(pmd[i]), PT_PTE); |
564 | } |
565 | } |
566 | |
567 | static void xen_pud_walk(struct mm_struct *mm, pud_t *pud, |
568 | void (*func)(struct mm_struct *mm, struct page *, |
569 | enum pt_level), |
570 | bool last, unsigned long limit) |
571 | { |
572 | int i, nr; |
573 | |
574 | nr = last ? pud_index(address: limit) + 1 : PTRS_PER_PUD; |
575 | for (i = 0; i < nr; i++) { |
576 | pmd_t *pmd; |
577 | |
578 | if (pud_none(pud: pud[i])) |
579 | continue; |
580 | |
581 | pmd = pmd_offset(pud: &pud[i], address: 0); |
582 | if (PTRS_PER_PMD > 1) |
583 | (*func)(mm, virt_to_page(pmd), PT_PMD); |
584 | xen_pmd_walk(mm, pmd, func, last: last && i == nr - 1, limit); |
585 | } |
586 | } |
587 | |
588 | static void xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d, |
589 | void (*func)(struct mm_struct *mm, struct page *, |
590 | enum pt_level), |
591 | bool last, unsigned long limit) |
592 | { |
593 | pud_t *pud; |
594 | |
595 | |
596 | if (p4d_none(p4d: *p4d)) |
597 | return; |
598 | |
599 | pud = pud_offset(p4d, address: 0); |
600 | if (PTRS_PER_PUD > 1) |
601 | (*func)(mm, virt_to_page(pud), PT_PUD); |
602 | xen_pud_walk(mm, pud, func, last, limit); |
603 | } |
604 | |
605 | /* |
606 | * (Yet another) pagetable walker. This one is intended for pinning a |
607 | * pagetable. This means that it walks a pagetable and calls the |
608 | * callback function on each page it finds making up the page table, |
609 | * at every level. It walks the entire pagetable, but it only bothers |
610 | * pinning pte pages which are below limit. In the normal case this |
611 | * will be STACK_TOP_MAX, but at boot we need to pin up to |
612 | * FIXADDR_TOP. |
613 | * |
614 | * We must skip the Xen hole in the middle of the address space, just after |
615 | * the big x86-64 virtual hole. |
616 | */ |
617 | static void __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd, |
618 | void (*func)(struct mm_struct *mm, struct page *, |
619 | enum pt_level), |
620 | unsigned long limit) |
621 | { |
622 | int i, nr; |
623 | unsigned hole_low = 0, hole_high = 0; |
624 | |
625 | /* The limit is the last byte to be touched */ |
626 | limit--; |
627 | BUG_ON(limit >= FIXADDR_TOP); |
628 | |
629 | /* |
630 | * 64-bit has a great big hole in the middle of the address |
631 | * space, which contains the Xen mappings. |
632 | */ |
633 | hole_low = pgd_index(GUARD_HOLE_BASE_ADDR); |
634 | hole_high = pgd_index(GUARD_HOLE_END_ADDR); |
635 | |
636 | nr = pgd_index(limit) + 1; |
637 | for (i = 0; i < nr; i++) { |
638 | p4d_t *p4d; |
639 | |
640 | if (i >= hole_low && i < hole_high) |
641 | continue; |
642 | |
643 | if (pgd_none(pgd: pgd[i])) |
644 | continue; |
645 | |
646 | p4d = p4d_offset(pgd: &pgd[i], address: 0); |
647 | xen_p4d_walk(mm, p4d, func, last: i == nr - 1, limit); |
648 | } |
649 | |
650 | /* Do the top level last, so that the callbacks can use it as |
651 | a cue to do final things like tlb flushes. */ |
652 | (*func)(mm, virt_to_page(pgd), PT_PGD); |
653 | } |
654 | |
655 | static void xen_pgd_walk(struct mm_struct *mm, |
656 | void (*func)(struct mm_struct *mm, struct page *, |
657 | enum pt_level), |
658 | unsigned long limit) |
659 | { |
660 | __xen_pgd_walk(mm, pgd: mm->pgd, func, limit); |
661 | } |
662 | |
663 | /* If we're using split pte locks, then take the page's lock and |
664 | return a pointer to it. Otherwise return NULL. */ |
665 | static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm) |
666 | { |
667 | spinlock_t *ptl = NULL; |
668 | |
669 | #if USE_SPLIT_PTE_PTLOCKS |
670 | ptl = ptlock_ptr(page_ptdesc(page)); |
671 | spin_lock_nest_lock(ptl, &mm->page_table_lock); |
672 | #endif |
673 | |
674 | return ptl; |
675 | } |
676 | |
677 | static void xen_pte_unlock(void *v) |
678 | { |
679 | spinlock_t *ptl = v; |
680 | spin_unlock(lock: ptl); |
681 | } |
682 | |
683 | static void xen_do_pin(unsigned level, unsigned long pfn) |
684 | { |
685 | struct mmuext_op op; |
686 | |
687 | op.cmd = level; |
688 | op.arg1.mfn = pfn_to_mfn(pfn); |
689 | |
690 | xen_extend_mmuext_op(op: &op); |
691 | } |
692 | |
693 | static void xen_pin_page(struct mm_struct *mm, struct page *page, |
694 | enum pt_level level) |
695 | { |
696 | unsigned pgfl = TestSetPagePinned(page); |
697 | |
698 | if (!pgfl) { |
699 | void *pt = lowmem_page_address(page); |
700 | unsigned long pfn = page_to_pfn(page); |
701 | struct multicall_space mcs = __xen_mc_entry(args: 0); |
702 | spinlock_t *ptl; |
703 | |
704 | /* |
705 | * We need to hold the pagetable lock between the time |
706 | * we make the pagetable RO and when we actually pin |
707 | * it. If we don't, then other users may come in and |
708 | * attempt to update the pagetable by writing it, |
709 | * which will fail because the memory is RO but not |
710 | * pinned, so Xen won't do the trap'n'emulate. |
711 | * |
712 | * If we're using split pte locks, we can't hold the |
713 | * entire pagetable's worth of locks during the |
714 | * traverse, because we may wrap the preempt count (8 |
715 | * bits). The solution is to mark RO and pin each PTE |
716 | * page while holding the lock. This means the number |
717 | * of locks we end up holding is never more than a |
718 | * batch size (~32 entries, at present). |
719 | * |
720 | * If we're not using split pte locks, we needn't pin |
721 | * the PTE pages independently, because we're |
722 | * protected by the overall pagetable lock. |
723 | */ |
724 | ptl = NULL; |
725 | if (level == PT_PTE) |
726 | ptl = xen_pte_lock(page, mm); |
727 | |
728 | MULTI_update_va_mapping(mcl: mcs.mc, va: (unsigned long)pt, |
729 | new_val: pfn_pte(page_nr: pfn, PAGE_KERNEL_RO), |
730 | flags: level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
731 | |
732 | if (ptl) { |
733 | xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn); |
734 | |
735 | /* Queue a deferred unlock for when this batch |
736 | is completed. */ |
737 | xen_mc_callback(fn: xen_pte_unlock, data: ptl); |
738 | } |
739 | } |
740 | } |
741 | |
742 | /* This is called just after a mm has been created, but it has not |
743 | been used yet. We need to make sure that its pagetable is all |
744 | read-only, and can be pinned. */ |
745 | static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd) |
746 | { |
747 | pgd_t *user_pgd = xen_get_user_pgd(pgd); |
748 | |
749 | trace_xen_mmu_pgd_pin(mm, pgd); |
750 | |
751 | xen_mc_batch(); |
752 | |
753 | __xen_pgd_walk(mm, pgd, func: xen_pin_page, USER_LIMIT); |
754 | |
755 | xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd))); |
756 | |
757 | if (user_pgd) { |
758 | xen_pin_page(mm, virt_to_page(user_pgd), level: PT_PGD); |
759 | xen_do_pin(MMUEXT_PIN_L4_TABLE, |
760 | PFN_DOWN(__pa(user_pgd))); |
761 | } |
762 | |
763 | xen_mc_issue(mode: 0); |
764 | } |
765 | |
766 | static void xen_pgd_pin(struct mm_struct *mm) |
767 | { |
768 | __xen_pgd_pin(mm, pgd: mm->pgd); |
769 | } |
770 | |
771 | /* |
772 | * On save, we need to pin all pagetables to make sure they get their |
773 | * mfns turned into pfns. Search the list for any unpinned pgds and pin |
774 | * them (unpinned pgds are not currently in use, probably because the |
775 | * process is under construction or destruction). |
776 | * |
777 | * Expected to be called in stop_machine() ("equivalent to taking |
778 | * every spinlock in the system"), so the locking doesn't really |
779 | * matter all that much. |
780 | */ |
781 | void xen_mm_pin_all(void) |
782 | { |
783 | struct page *page; |
784 | |
785 | spin_lock(lock: &pgd_lock); |
786 | |
787 | list_for_each_entry(page, &pgd_list, lru) { |
788 | if (!PagePinned(page)) { |
789 | __xen_pgd_pin(mm: &init_mm, pgd: (pgd_t *)page_address(page)); |
790 | SetPageSavePinned(page); |
791 | } |
792 | } |
793 | |
794 | spin_unlock(lock: &pgd_lock); |
795 | } |
796 | |
797 | static void __init xen_mark_pinned(struct mm_struct *mm, struct page *page, |
798 | enum pt_level level) |
799 | { |
800 | SetPagePinned(page); |
801 | } |
802 | |
803 | /* |
804 | * The init_mm pagetable is really pinned as soon as its created, but |
805 | * that's before we have page structures to store the bits. So do all |
806 | * the book-keeping now once struct pages for allocated pages are |
807 | * initialized. This happens only after memblock_free_all() is called. |
808 | */ |
809 | static void __init xen_after_bootmem(void) |
810 | { |
811 | static_branch_enable(&xen_struct_pages_ready); |
812 | #ifdef CONFIG_X86_VSYSCALL_EMULATION |
813 | SetPagePinned(virt_to_page(level3_user_vsyscall)); |
814 | #endif |
815 | xen_pgd_walk(mm: &init_mm, func: xen_mark_pinned, FIXADDR_TOP); |
816 | } |
817 | |
818 | static void xen_unpin_page(struct mm_struct *mm, struct page *page, |
819 | enum pt_level level) |
820 | { |
821 | unsigned pgfl = TestClearPagePinned(page); |
822 | |
823 | if (pgfl) { |
824 | void *pt = lowmem_page_address(page); |
825 | unsigned long pfn = page_to_pfn(page); |
826 | spinlock_t *ptl = NULL; |
827 | struct multicall_space mcs; |
828 | |
829 | /* |
830 | * Do the converse to pin_page. If we're using split |
831 | * pte locks, we must be holding the lock for while |
832 | * the pte page is unpinned but still RO to prevent |
833 | * concurrent updates from seeing it in this |
834 | * partially-pinned state. |
835 | */ |
836 | if (level == PT_PTE) { |
837 | ptl = xen_pte_lock(page, mm); |
838 | |
839 | if (ptl) |
840 | xen_do_pin(MMUEXT_UNPIN_TABLE, pfn); |
841 | } |
842 | |
843 | mcs = __xen_mc_entry(args: 0); |
844 | |
845 | MULTI_update_va_mapping(mcl: mcs.mc, va: (unsigned long)pt, |
846 | new_val: pfn_pte(page_nr: pfn, PAGE_KERNEL), |
847 | flags: level == PT_PGD ? UVMF_TLB_FLUSH : 0); |
848 | |
849 | if (ptl) { |
850 | /* unlock when batch completed */ |
851 | xen_mc_callback(fn: xen_pte_unlock, data: ptl); |
852 | } |
853 | } |
854 | } |
855 | |
856 | /* Release a pagetables pages back as normal RW */ |
857 | static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd) |
858 | { |
859 | pgd_t *user_pgd = xen_get_user_pgd(pgd); |
860 | |
861 | trace_xen_mmu_pgd_unpin(mm, pgd); |
862 | |
863 | xen_mc_batch(); |
864 | |
865 | xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); |
866 | |
867 | if (user_pgd) { |
868 | xen_do_pin(MMUEXT_UNPIN_TABLE, |
869 | PFN_DOWN(__pa(user_pgd))); |
870 | xen_unpin_page(mm, virt_to_page(user_pgd), level: PT_PGD); |
871 | } |
872 | |
873 | __xen_pgd_walk(mm, pgd, func: xen_unpin_page, USER_LIMIT); |
874 | |
875 | xen_mc_issue(mode: 0); |
876 | } |
877 | |
878 | static void xen_pgd_unpin(struct mm_struct *mm) |
879 | { |
880 | __xen_pgd_unpin(mm, pgd: mm->pgd); |
881 | } |
882 | |
883 | /* |
884 | * On resume, undo any pinning done at save, so that the rest of the |
885 | * kernel doesn't see any unexpected pinned pagetables. |
886 | */ |
887 | void xen_mm_unpin_all(void) |
888 | { |
889 | struct page *page; |
890 | |
891 | spin_lock(lock: &pgd_lock); |
892 | |
893 | list_for_each_entry(page, &pgd_list, lru) { |
894 | if (PageSavePinned(page)) { |
895 | BUG_ON(!PagePinned(page)); |
896 | __xen_pgd_unpin(mm: &init_mm, pgd: (pgd_t *)page_address(page)); |
897 | ClearPageSavePinned(page); |
898 | } |
899 | } |
900 | |
901 | spin_unlock(lock: &pgd_lock); |
902 | } |
903 | |
904 | static void xen_enter_mmap(struct mm_struct *mm) |
905 | { |
906 | spin_lock(lock: &mm->page_table_lock); |
907 | xen_pgd_pin(mm); |
908 | spin_unlock(lock: &mm->page_table_lock); |
909 | } |
910 | |
911 | static void drop_mm_ref_this_cpu(void *info) |
912 | { |
913 | struct mm_struct *mm = info; |
914 | |
915 | if (this_cpu_read(cpu_tlbstate.loaded_mm) == mm) |
916 | leave_mm(smp_processor_id()); |
917 | |
918 | /* |
919 | * If this cpu still has a stale cr3 reference, then make sure |
920 | * it has been flushed. |
921 | */ |
922 | if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd)) |
923 | xen_mc_flush(); |
924 | } |
925 | |
926 | #ifdef CONFIG_SMP |
927 | /* |
928 | * Another cpu may still have their %cr3 pointing at the pagetable, so |
929 | * we need to repoint it somewhere else before we can unpin it. |
930 | */ |
931 | static void xen_drop_mm_ref(struct mm_struct *mm) |
932 | { |
933 | cpumask_var_t mask; |
934 | unsigned cpu; |
935 | |
936 | drop_mm_ref_this_cpu(info: mm); |
937 | |
938 | /* Get the "official" set of cpus referring to our pagetable. */ |
939 | if (!alloc_cpumask_var(mask: &mask, GFP_ATOMIC)) { |
940 | for_each_online_cpu(cpu) { |
941 | if (per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd)) |
942 | continue; |
943 | smp_call_function_single(cpuid: cpu, func: drop_mm_ref_this_cpu, info: mm, wait: 1); |
944 | } |
945 | return; |
946 | } |
947 | |
948 | /* |
949 | * It's possible that a vcpu may have a stale reference to our |
950 | * cr3, because its in lazy mode, and it hasn't yet flushed |
951 | * its set of pending hypercalls yet. In this case, we can |
952 | * look at its actual current cr3 value, and force it to flush |
953 | * if needed. |
954 | */ |
955 | cpumask_clear(dstp: mask); |
956 | for_each_online_cpu(cpu) { |
957 | if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd)) |
958 | cpumask_set_cpu(cpu, dstp: mask); |
959 | } |
960 | |
961 | smp_call_function_many(mask, func: drop_mm_ref_this_cpu, info: mm, wait: 1); |
962 | free_cpumask_var(mask); |
963 | } |
964 | #else |
965 | static void xen_drop_mm_ref(struct mm_struct *mm) |
966 | { |
967 | drop_mm_ref_this_cpu(mm); |
968 | } |
969 | #endif |
970 | |
971 | /* |
972 | * While a process runs, Xen pins its pagetables, which means that the |
973 | * hypervisor forces it to be read-only, and it controls all updates |
974 | * to it. This means that all pagetable updates have to go via the |
975 | * hypervisor, which is moderately expensive. |
976 | * |
977 | * Since we're pulling the pagetable down, we switch to use init_mm, |
978 | * unpin old process pagetable and mark it all read-write, which |
979 | * allows further operations on it to be simple memory accesses. |
980 | * |
981 | * The only subtle point is that another CPU may be still using the |
982 | * pagetable because of lazy tlb flushing. This means we need need to |
983 | * switch all CPUs off this pagetable before we can unpin it. |
984 | */ |
985 | static void xen_exit_mmap(struct mm_struct *mm) |
986 | { |
987 | get_cpu(); /* make sure we don't move around */ |
988 | xen_drop_mm_ref(mm); |
989 | put_cpu(); |
990 | |
991 | spin_lock(lock: &mm->page_table_lock); |
992 | |
993 | /* pgd may not be pinned in the error exit path of execve */ |
994 | if (xen_page_pinned(ptr: mm->pgd)) |
995 | xen_pgd_unpin(mm); |
996 | |
997 | spin_unlock(lock: &mm->page_table_lock); |
998 | } |
999 | |
1000 | static void xen_post_allocator_init(void); |
1001 | |
1002 | static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn) |
1003 | { |
1004 | struct mmuext_op op; |
1005 | |
1006 | op.cmd = cmd; |
1007 | op.arg1.mfn = pfn_to_mfn(pfn); |
1008 | if (HYPERVISOR_mmuext_op(op: &op, count: 1, NULL, DOMID_SELF)) |
1009 | BUG(); |
1010 | } |
1011 | |
1012 | static void __init xen_cleanhighmap(unsigned long vaddr, |
1013 | unsigned long vaddr_end) |
1014 | { |
1015 | unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1; |
1016 | pmd_t *pmd = level2_kernel_pgt + pmd_index(address: vaddr); |
1017 | |
1018 | /* NOTE: The loop is more greedy than the cleanup_highmap variant. |
1019 | * We include the PMD passed in on _both_ boundaries. */ |
1020 | for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD)); |
1021 | pmd++, vaddr += PMD_SIZE) { |
1022 | if (pmd_none(pmd: *pmd)) |
1023 | continue; |
1024 | if (vaddr < (unsigned long) _text || vaddr > kernel_end) |
1025 | set_pmd(pmdp: pmd, pmd: __pmd(val: 0)); |
1026 | } |
1027 | /* In case we did something silly, we should crash in this function |
1028 | * instead of somewhere later and be confusing. */ |
1029 | xen_mc_flush(); |
1030 | } |
1031 | |
1032 | /* |
1033 | * Make a page range writeable and free it. |
1034 | */ |
1035 | static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size) |
1036 | { |
1037 | void *vaddr = __va(paddr); |
1038 | void *vaddr_end = vaddr + size; |
1039 | |
1040 | for (; vaddr < vaddr_end; vaddr += PAGE_SIZE) |
1041 | make_lowmem_page_readwrite(vaddr); |
1042 | |
1043 | memblock_phys_free(base: paddr, size); |
1044 | } |
1045 | |
1046 | static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin) |
1047 | { |
1048 | unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK; |
1049 | |
1050 | if (unpin) |
1051 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa)); |
1052 | ClearPagePinned(virt_to_page(__va(pa))); |
1053 | xen_free_ro_pages(paddr: pa, PAGE_SIZE); |
1054 | } |
1055 | |
1056 | static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin) |
1057 | { |
1058 | unsigned long pa; |
1059 | pte_t *pte_tbl; |
1060 | int i; |
1061 | |
1062 | if (pmd_large(pte: *pmd)) { |
1063 | pa = pmd_val(pmd: *pmd) & PHYSICAL_PAGE_MASK; |
1064 | xen_free_ro_pages(paddr: pa, PMD_SIZE); |
1065 | return; |
1066 | } |
1067 | |
1068 | pte_tbl = pte_offset_kernel(pmd, address: 0); |
1069 | for (i = 0; i < PTRS_PER_PTE; i++) { |
1070 | if (pte_none(pte: pte_tbl[i])) |
1071 | continue; |
1072 | pa = pte_pfn(pte: pte_tbl[i]) << PAGE_SHIFT; |
1073 | xen_free_ro_pages(paddr: pa, PAGE_SIZE); |
1074 | } |
1075 | set_pmd(pmdp: pmd, pmd: __pmd(val: 0)); |
1076 | xen_cleanmfnmap_free_pgtbl(pgtbl: pte_tbl, unpin); |
1077 | } |
1078 | |
1079 | static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin) |
1080 | { |
1081 | unsigned long pa; |
1082 | pmd_t *pmd_tbl; |
1083 | int i; |
1084 | |
1085 | if (pud_large(pud: *pud)) { |
1086 | pa = pud_val(pud: *pud) & PHYSICAL_PAGE_MASK; |
1087 | xen_free_ro_pages(paddr: pa, PUD_SIZE); |
1088 | return; |
1089 | } |
1090 | |
1091 | pmd_tbl = pmd_offset(pud, address: 0); |
1092 | for (i = 0; i < PTRS_PER_PMD; i++) { |
1093 | if (pmd_none(pmd: pmd_tbl[i])) |
1094 | continue; |
1095 | xen_cleanmfnmap_pmd(pmd: pmd_tbl + i, unpin); |
1096 | } |
1097 | set_pud(pudp: pud, pud: __pud(val: 0)); |
1098 | xen_cleanmfnmap_free_pgtbl(pgtbl: pmd_tbl, unpin); |
1099 | } |
1100 | |
1101 | static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin) |
1102 | { |
1103 | unsigned long pa; |
1104 | pud_t *pud_tbl; |
1105 | int i; |
1106 | |
1107 | if (p4d_large(p4d: *p4d)) { |
1108 | pa = p4d_val(p4d: *p4d) & PHYSICAL_PAGE_MASK; |
1109 | xen_free_ro_pages(paddr: pa, P4D_SIZE); |
1110 | return; |
1111 | } |
1112 | |
1113 | pud_tbl = pud_offset(p4d, address: 0); |
1114 | for (i = 0; i < PTRS_PER_PUD; i++) { |
1115 | if (pud_none(pud: pud_tbl[i])) |
1116 | continue; |
1117 | xen_cleanmfnmap_pud(pud: pud_tbl + i, unpin); |
1118 | } |
1119 | set_p4d(p4dp: p4d, p4d: __p4d(val: 0)); |
1120 | xen_cleanmfnmap_free_pgtbl(pgtbl: pud_tbl, unpin); |
1121 | } |
1122 | |
1123 | /* |
1124 | * Since it is well isolated we can (and since it is perhaps large we should) |
1125 | * also free the page tables mapping the initial P->M table. |
1126 | */ |
1127 | static void __init xen_cleanmfnmap(unsigned long vaddr) |
1128 | { |
1129 | pgd_t *pgd; |
1130 | p4d_t *p4d; |
1131 | bool unpin; |
1132 | |
1133 | unpin = (vaddr == 2 * PGDIR_SIZE); |
1134 | vaddr &= PMD_MASK; |
1135 | pgd = pgd_offset_k(vaddr); |
1136 | p4d = p4d_offset(pgd, address: 0); |
1137 | if (!p4d_none(p4d: *p4d)) |
1138 | xen_cleanmfnmap_p4d(p4d, unpin); |
1139 | } |
1140 | |
1141 | static void __init xen_pagetable_p2m_free(void) |
1142 | { |
1143 | unsigned long size; |
1144 | unsigned long addr; |
1145 | |
1146 | size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long)); |
1147 | |
1148 | /* No memory or already called. */ |
1149 | if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list) |
1150 | return; |
1151 | |
1152 | /* using __ka address and sticking INVALID_P2M_ENTRY! */ |
1153 | memset((void *)xen_start_info->mfn_list, 0xff, size); |
1154 | |
1155 | addr = xen_start_info->mfn_list; |
1156 | /* |
1157 | * We could be in __ka space. |
1158 | * We roundup to the PMD, which means that if anybody at this stage is |
1159 | * using the __ka address of xen_start_info or |
1160 | * xen_start_info->shared_info they are in going to crash. Fortunately |
1161 | * we have already revectored in xen_setup_kernel_pagetable. |
1162 | */ |
1163 | size = roundup(size, PMD_SIZE); |
1164 | |
1165 | if (addr >= __START_KERNEL_map) { |
1166 | xen_cleanhighmap(vaddr: addr, vaddr_end: addr + size); |
1167 | size = PAGE_ALIGN(xen_start_info->nr_pages * |
1168 | sizeof(unsigned long)); |
1169 | memblock_free(ptr: (void *)addr, size); |
1170 | } else { |
1171 | xen_cleanmfnmap(vaddr: addr); |
1172 | } |
1173 | } |
1174 | |
1175 | static void __init xen_pagetable_cleanhighmap(void) |
1176 | { |
1177 | unsigned long size; |
1178 | unsigned long addr; |
1179 | |
1180 | /* At this stage, cleanup_highmap has already cleaned __ka space |
1181 | * from _brk_limit way up to the max_pfn_mapped (which is the end of |
1182 | * the ramdisk). We continue on, erasing PMD entries that point to page |
1183 | * tables - do note that they are accessible at this stage via __va. |
1184 | * As Xen is aligning the memory end to a 4MB boundary, for good |
1185 | * measure we also round up to PMD_SIZE * 2 - which means that if |
1186 | * anybody is using __ka address to the initial boot-stack - and try |
1187 | * to use it - they are going to crash. The xen_start_info has been |
1188 | * taken care of already in xen_setup_kernel_pagetable. */ |
1189 | addr = xen_start_info->pt_base; |
1190 | size = xen_start_info->nr_pt_frames * PAGE_SIZE; |
1191 | |
1192 | xen_cleanhighmap(vaddr: addr, roundup(addr + size, PMD_SIZE * 2)); |
1193 | xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base)); |
1194 | } |
1195 | |
1196 | static void __init xen_pagetable_p2m_setup(void) |
1197 | { |
1198 | xen_vmalloc_p2m_tree(); |
1199 | |
1200 | xen_pagetable_p2m_free(); |
1201 | |
1202 | xen_pagetable_cleanhighmap(); |
1203 | |
1204 | /* And revector! Bye bye old array */ |
1205 | xen_start_info->mfn_list = (unsigned long)xen_p2m_addr; |
1206 | } |
1207 | |
1208 | static void __init xen_pagetable_init(void) |
1209 | { |
1210 | /* |
1211 | * The majority of further PTE writes is to pagetables already |
1212 | * announced as such to Xen. Hence it is more efficient to use |
1213 | * hypercalls for these updates. |
1214 | */ |
1215 | pv_ops.mmu.set_pte = __xen_set_pte; |
1216 | |
1217 | paging_init(); |
1218 | xen_post_allocator_init(); |
1219 | |
1220 | xen_pagetable_p2m_setup(); |
1221 | |
1222 | /* Allocate and initialize top and mid mfn levels for p2m structure */ |
1223 | xen_build_mfn_list_list(); |
1224 | |
1225 | /* Remap memory freed due to conflicts with E820 map */ |
1226 | xen_remap_memory(); |
1227 | xen_setup_mfn_list_list(); |
1228 | } |
1229 | |
1230 | static noinstr void xen_write_cr2(unsigned long cr2) |
1231 | { |
1232 | this_cpu_read(xen_vcpu)->arch.cr2 = cr2; |
1233 | } |
1234 | |
1235 | static noinline void xen_flush_tlb(void) |
1236 | { |
1237 | struct mmuext_op *op; |
1238 | struct multicall_space mcs; |
1239 | |
1240 | preempt_disable(); |
1241 | |
1242 | mcs = xen_mc_entry(args: sizeof(*op)); |
1243 | |
1244 | op = mcs.args; |
1245 | op->cmd = MMUEXT_TLB_FLUSH_LOCAL; |
1246 | MULTI_mmuext_op(mcl: mcs.mc, op, count: 1, NULL, DOMID_SELF); |
1247 | |
1248 | xen_mc_issue(mode: XEN_LAZY_MMU); |
1249 | |
1250 | preempt_enable(); |
1251 | } |
1252 | |
1253 | static void xen_flush_tlb_one_user(unsigned long addr) |
1254 | { |
1255 | struct mmuext_op *op; |
1256 | struct multicall_space mcs; |
1257 | |
1258 | trace_xen_mmu_flush_tlb_one_user(addr); |
1259 | |
1260 | preempt_disable(); |
1261 | |
1262 | mcs = xen_mc_entry(args: sizeof(*op)); |
1263 | op = mcs.args; |
1264 | op->cmd = MMUEXT_INVLPG_LOCAL; |
1265 | op->arg1.linear_addr = addr & PAGE_MASK; |
1266 | MULTI_mmuext_op(mcl: mcs.mc, op, count: 1, NULL, DOMID_SELF); |
1267 | |
1268 | xen_mc_issue(mode: XEN_LAZY_MMU); |
1269 | |
1270 | preempt_enable(); |
1271 | } |
1272 | |
1273 | static void xen_flush_tlb_multi(const struct cpumask *cpus, |
1274 | const struct flush_tlb_info *info) |
1275 | { |
1276 | struct { |
1277 | struct mmuext_op op; |
1278 | DECLARE_BITMAP(mask, NR_CPUS); |
1279 | } *args; |
1280 | struct multicall_space mcs; |
1281 | const size_t mc_entry_size = sizeof(args->op) + |
1282 | sizeof(args->mask[0]) * BITS_TO_LONGS(num_possible_cpus()); |
1283 | |
1284 | trace_xen_mmu_flush_tlb_multi(cpus, mm: info->mm, addr: info->start, end: info->end); |
1285 | |
1286 | if (cpumask_empty(srcp: cpus)) |
1287 | return; /* nothing to do */ |
1288 | |
1289 | mcs = xen_mc_entry(args: mc_entry_size); |
1290 | args = mcs.args; |
1291 | args->op.arg2.vcpumask = to_cpumask(args->mask); |
1292 | |
1293 | /* Remove any offline CPUs */ |
1294 | cpumask_and(to_cpumask(args->mask), src1p: cpus, cpu_online_mask); |
1295 | |
1296 | args->op.cmd = MMUEXT_TLB_FLUSH_MULTI; |
1297 | if (info->end != TLB_FLUSH_ALL && |
1298 | (info->end - info->start) <= PAGE_SIZE) { |
1299 | args->op.cmd = MMUEXT_INVLPG_MULTI; |
1300 | args->op.arg1.linear_addr = info->start; |
1301 | } |
1302 | |
1303 | MULTI_mmuext_op(mcl: mcs.mc, op: &args->op, count: 1, NULL, DOMID_SELF); |
1304 | |
1305 | xen_mc_issue(mode: XEN_LAZY_MMU); |
1306 | } |
1307 | |
1308 | static unsigned long xen_read_cr3(void) |
1309 | { |
1310 | return this_cpu_read(xen_cr3); |
1311 | } |
1312 | |
1313 | static void set_current_cr3(void *v) |
1314 | { |
1315 | this_cpu_write(xen_current_cr3, (unsigned long)v); |
1316 | } |
1317 | |
1318 | static void __xen_write_cr3(bool kernel, unsigned long cr3) |
1319 | { |
1320 | struct mmuext_op op; |
1321 | unsigned long mfn; |
1322 | |
1323 | trace_xen_mmu_write_cr3(kernel, cr3); |
1324 | |
1325 | if (cr3) |
1326 | mfn = pfn_to_mfn(PFN_DOWN(cr3)); |
1327 | else |
1328 | mfn = 0; |
1329 | |
1330 | WARN_ON(mfn == 0 && kernel); |
1331 | |
1332 | op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR; |
1333 | op.arg1.mfn = mfn; |
1334 | |
1335 | xen_extend_mmuext_op(op: &op); |
1336 | |
1337 | if (kernel) { |
1338 | this_cpu_write(xen_cr3, cr3); |
1339 | |
1340 | /* Update xen_current_cr3 once the batch has actually |
1341 | been submitted. */ |
1342 | xen_mc_callback(fn: set_current_cr3, data: (void *)cr3); |
1343 | } |
1344 | } |
1345 | static void xen_write_cr3(unsigned long cr3) |
1346 | { |
1347 | pgd_t *user_pgd = xen_get_user_pgd(__va(cr3)); |
1348 | |
1349 | BUG_ON(preemptible()); |
1350 | |
1351 | xen_mc_batch(); /* disables interrupts */ |
1352 | |
1353 | /* Update while interrupts are disabled, so its atomic with |
1354 | respect to ipis */ |
1355 | this_cpu_write(xen_cr3, cr3); |
1356 | |
1357 | __xen_write_cr3(kernel: true, cr3); |
1358 | |
1359 | if (user_pgd) |
1360 | __xen_write_cr3(kernel: false, __pa(user_pgd)); |
1361 | else |
1362 | __xen_write_cr3(kernel: false, cr3: 0); |
1363 | |
1364 | xen_mc_issue(mode: XEN_LAZY_CPU); /* interrupts restored */ |
1365 | } |
1366 | |
1367 | /* |
1368 | * At the start of the day - when Xen launches a guest, it has already |
1369 | * built pagetables for the guest. We diligently look over them |
1370 | * in xen_setup_kernel_pagetable and graft as appropriate them in the |
1371 | * init_top_pgt and its friends. Then when we are happy we load |
1372 | * the new init_top_pgt - and continue on. |
1373 | * |
1374 | * The generic code starts (start_kernel) and 'init_mem_mapping' sets |
1375 | * up the rest of the pagetables. When it has completed it loads the cr3. |
1376 | * N.B. that baremetal would start at 'start_kernel' (and the early |
1377 | * #PF handler would create bootstrap pagetables) - so we are running |
1378 | * with the same assumptions as what to do when write_cr3 is executed |
1379 | * at this point. |
1380 | * |
1381 | * Since there are no user-page tables at all, we have two variants |
1382 | * of xen_write_cr3 - the early bootup (this one), and the late one |
1383 | * (xen_write_cr3). The reason we have to do that is that in 64-bit |
1384 | * the Linux kernel and user-space are both in ring 3 while the |
1385 | * hypervisor is in ring 0. |
1386 | */ |
1387 | static void __init xen_write_cr3_init(unsigned long cr3) |
1388 | { |
1389 | BUG_ON(preemptible()); |
1390 | |
1391 | xen_mc_batch(); /* disables interrupts */ |
1392 | |
1393 | /* Update while interrupts are disabled, so its atomic with |
1394 | respect to ipis */ |
1395 | this_cpu_write(xen_cr3, cr3); |
1396 | |
1397 | __xen_write_cr3(kernel: true, cr3); |
1398 | |
1399 | xen_mc_issue(mode: XEN_LAZY_CPU); /* interrupts restored */ |
1400 | } |
1401 | |
1402 | static int xen_pgd_alloc(struct mm_struct *mm) |
1403 | { |
1404 | pgd_t *pgd = mm->pgd; |
1405 | struct page *page = virt_to_page(pgd); |
1406 | pgd_t *user_pgd; |
1407 | int ret = -ENOMEM; |
1408 | |
1409 | BUG_ON(PagePinned(virt_to_page(pgd))); |
1410 | BUG_ON(page->private != 0); |
1411 | |
1412 | user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO); |
1413 | page->private = (unsigned long)user_pgd; |
1414 | |
1415 | if (user_pgd != NULL) { |
1416 | #ifdef CONFIG_X86_VSYSCALL_EMULATION |
1417 | user_pgd[pgd_index(VSYSCALL_ADDR)] = |
1418 | __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE); |
1419 | #endif |
1420 | ret = 0; |
1421 | } |
1422 | |
1423 | BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd)))); |
1424 | |
1425 | return ret; |
1426 | } |
1427 | |
1428 | static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd) |
1429 | { |
1430 | pgd_t *user_pgd = xen_get_user_pgd(pgd); |
1431 | |
1432 | if (user_pgd) |
1433 | free_page((unsigned long)user_pgd); |
1434 | } |
1435 | |
1436 | /* |
1437 | * Init-time set_pte while constructing initial pagetables, which |
1438 | * doesn't allow RO page table pages to be remapped RW. |
1439 | * |
1440 | * If there is no MFN for this PFN then this page is initially |
1441 | * ballooned out so clear the PTE (as in decrease_reservation() in |
1442 | * drivers/xen/balloon.c). |
1443 | * |
1444 | * Many of these PTE updates are done on unpinned and writable pages |
1445 | * and doing a hypercall for these is unnecessary and expensive. At |
1446 | * this point it is rarely possible to tell if a page is pinned, so |
1447 | * mostly write the PTE directly and rely on Xen trapping and |
1448 | * emulating any updates as necessary. |
1449 | */ |
1450 | static void __init xen_set_pte_init(pte_t *ptep, pte_t pte) |
1451 | { |
1452 | if (unlikely(is_early_ioremap_ptep(ptep))) |
1453 | __xen_set_pte(ptep, pteval: pte); |
1454 | else |
1455 | native_set_pte(ptep, pte); |
1456 | } |
1457 | |
1458 | __visible pte_t xen_make_pte_init(pteval_t pte) |
1459 | { |
1460 | unsigned long pfn; |
1461 | |
1462 | /* |
1463 | * Pages belonging to the initial p2m list mapped outside the default |
1464 | * address range must be mapped read-only. This region contains the |
1465 | * page tables for mapping the p2m list, too, and page tables MUST be |
1466 | * mapped read-only. |
1467 | */ |
1468 | pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT; |
1469 | if (xen_start_info->mfn_list < __START_KERNEL_map && |
1470 | pfn >= xen_start_info->first_p2m_pfn && |
1471 | pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames) |
1472 | pte &= ~_PAGE_RW; |
1473 | |
1474 | pte = pte_pfn_to_mfn(val: pte); |
1475 | return native_make_pte(val: pte); |
1476 | } |
1477 | PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init); |
1478 | |
1479 | /* Early in boot, while setting up the initial pagetable, assume |
1480 | everything is pinned. */ |
1481 | static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn) |
1482 | { |
1483 | #ifdef CONFIG_FLATMEM |
1484 | BUG_ON(mem_map); /* should only be used early */ |
1485 | #endif |
1486 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); |
1487 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); |
1488 | } |
1489 | |
1490 | /* Used for pmd and pud */ |
1491 | static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn) |
1492 | { |
1493 | #ifdef CONFIG_FLATMEM |
1494 | BUG_ON(mem_map); /* should only be used early */ |
1495 | #endif |
1496 | make_lowmem_page_readonly(__va(PFN_PHYS(pfn))); |
1497 | } |
1498 | |
1499 | /* Early release_pte assumes that all pts are pinned, since there's |
1500 | only init_mm and anything attached to that is pinned. */ |
1501 | static void __init xen_release_pte_init(unsigned long pfn) |
1502 | { |
1503 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); |
1504 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); |
1505 | } |
1506 | |
1507 | static void __init xen_release_pmd_init(unsigned long pfn) |
1508 | { |
1509 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); |
1510 | } |
1511 | |
1512 | static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn) |
1513 | { |
1514 | struct multicall_space mcs; |
1515 | struct mmuext_op *op; |
1516 | |
1517 | mcs = __xen_mc_entry(args: sizeof(*op)); |
1518 | op = mcs.args; |
1519 | op->cmd = cmd; |
1520 | op->arg1.mfn = pfn_to_mfn(pfn); |
1521 | |
1522 | MULTI_mmuext_op(mcl: mcs.mc, op: mcs.args, count: 1, NULL, DOMID_SELF); |
1523 | } |
1524 | |
1525 | static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot) |
1526 | { |
1527 | struct multicall_space mcs; |
1528 | unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT); |
1529 | |
1530 | mcs = __xen_mc_entry(args: 0); |
1531 | MULTI_update_va_mapping(mcl: mcs.mc, va: (unsigned long)addr, |
1532 | new_val: pfn_pte(page_nr: pfn, pgprot: prot), flags: 0); |
1533 | } |
1534 | |
1535 | /* This needs to make sure the new pte page is pinned iff its being |
1536 | attached to a pinned pagetable. */ |
1537 | static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn, |
1538 | unsigned level) |
1539 | { |
1540 | bool pinned = xen_page_pinned(ptr: mm->pgd); |
1541 | |
1542 | trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned); |
1543 | |
1544 | if (pinned) { |
1545 | struct page *page = pfn_to_page(pfn); |
1546 | |
1547 | pinned = false; |
1548 | if (static_branch_likely(&xen_struct_pages_ready)) { |
1549 | pinned = PagePinned(page); |
1550 | SetPagePinned(page); |
1551 | } |
1552 | |
1553 | xen_mc_batch(); |
1554 | |
1555 | __set_pfn_prot(pfn, PAGE_KERNEL_RO); |
1556 | |
1557 | if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS && !pinned) |
1558 | __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn); |
1559 | |
1560 | xen_mc_issue(mode: XEN_LAZY_MMU); |
1561 | } |
1562 | } |
1563 | |
1564 | static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn) |
1565 | { |
1566 | xen_alloc_ptpage(mm, pfn, level: PT_PTE); |
1567 | } |
1568 | |
1569 | static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn) |
1570 | { |
1571 | xen_alloc_ptpage(mm, pfn, level: PT_PMD); |
1572 | } |
1573 | |
1574 | /* This should never happen until we're OK to use struct page */ |
1575 | static inline void xen_release_ptpage(unsigned long pfn, unsigned level) |
1576 | { |
1577 | struct page *page = pfn_to_page(pfn); |
1578 | bool pinned = PagePinned(page); |
1579 | |
1580 | trace_xen_mmu_release_ptpage(pfn, level, pinned); |
1581 | |
1582 | if (pinned) { |
1583 | xen_mc_batch(); |
1584 | |
1585 | if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS) |
1586 | __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn); |
1587 | |
1588 | __set_pfn_prot(pfn, PAGE_KERNEL); |
1589 | |
1590 | xen_mc_issue(mode: XEN_LAZY_MMU); |
1591 | |
1592 | ClearPagePinned(page); |
1593 | } |
1594 | } |
1595 | |
1596 | static void xen_release_pte(unsigned long pfn) |
1597 | { |
1598 | xen_release_ptpage(pfn, level: PT_PTE); |
1599 | } |
1600 | |
1601 | static void xen_release_pmd(unsigned long pfn) |
1602 | { |
1603 | xen_release_ptpage(pfn, level: PT_PMD); |
1604 | } |
1605 | |
1606 | static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn) |
1607 | { |
1608 | xen_alloc_ptpage(mm, pfn, level: PT_PUD); |
1609 | } |
1610 | |
1611 | static void xen_release_pud(unsigned long pfn) |
1612 | { |
1613 | xen_release_ptpage(pfn, level: PT_PUD); |
1614 | } |
1615 | |
1616 | /* |
1617 | * Like __va(), but returns address in the kernel mapping (which is |
1618 | * all we have until the physical memory mapping has been set up. |
1619 | */ |
1620 | static void * __init __ka(phys_addr_t paddr) |
1621 | { |
1622 | return (void *)(paddr + __START_KERNEL_map); |
1623 | } |
1624 | |
1625 | /* Convert a machine address to physical address */ |
1626 | static unsigned long __init m2p(phys_addr_t maddr) |
1627 | { |
1628 | phys_addr_t paddr; |
1629 | |
1630 | maddr &= XEN_PTE_MFN_MASK; |
1631 | paddr = mfn_to_pfn(mfn: maddr >> PAGE_SHIFT) << PAGE_SHIFT; |
1632 | |
1633 | return paddr; |
1634 | } |
1635 | |
1636 | /* Convert a machine address to kernel virtual */ |
1637 | static void * __init m2v(phys_addr_t maddr) |
1638 | { |
1639 | return __ka(paddr: m2p(maddr)); |
1640 | } |
1641 | |
1642 | /* Set the page permissions on an identity-mapped pages */ |
1643 | static void __init set_page_prot_flags(void *addr, pgprot_t prot, |
1644 | unsigned long flags) |
1645 | { |
1646 | unsigned long pfn = __pa(addr) >> PAGE_SHIFT; |
1647 | pte_t pte = pfn_pte(page_nr: pfn, pgprot: prot); |
1648 | |
1649 | if (HYPERVISOR_update_va_mapping(va: (unsigned long)addr, new_val: pte, flags)) |
1650 | BUG(); |
1651 | } |
1652 | static void __init set_page_prot(void *addr, pgprot_t prot) |
1653 | { |
1654 | return set_page_prot_flags(addr, prot, UVMF_NONE); |
1655 | } |
1656 | |
1657 | void __init xen_setup_machphys_mapping(void) |
1658 | { |
1659 | struct xen_machphys_mapping mapping; |
1660 | |
1661 | if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, arg: &mapping) == 0) { |
1662 | machine_to_phys_mapping = (unsigned long *)mapping.v_start; |
1663 | machine_to_phys_nr = mapping.max_mfn + 1; |
1664 | } else { |
1665 | machine_to_phys_nr = MACH2PHYS_NR_ENTRIES; |
1666 | } |
1667 | } |
1668 | |
1669 | static void __init convert_pfn_mfn(void *v) |
1670 | { |
1671 | pte_t *pte = v; |
1672 | int i; |
1673 | |
1674 | /* All levels are converted the same way, so just treat them |
1675 | as ptes. */ |
1676 | for (i = 0; i < PTRS_PER_PTE; i++) |
1677 | pte[i] = xen_make_pte(pte: pte[i].pte); |
1678 | } |
1679 | static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end, |
1680 | unsigned long addr) |
1681 | { |
1682 | if (*pt_base == PFN_DOWN(__pa(addr))) { |
1683 | set_page_prot_flags(addr: (void *)addr, PAGE_KERNEL, UVMF_INVLPG); |
1684 | clear_page(page: (void *)addr); |
1685 | (*pt_base)++; |
1686 | } |
1687 | if (*pt_end == PFN_DOWN(__pa(addr))) { |
1688 | set_page_prot_flags(addr: (void *)addr, PAGE_KERNEL, UVMF_INVLPG); |
1689 | clear_page(page: (void *)addr); |
1690 | (*pt_end)--; |
1691 | } |
1692 | } |
1693 | /* |
1694 | * Set up the initial kernel pagetable. |
1695 | * |
1696 | * We can construct this by grafting the Xen provided pagetable into |
1697 | * head_64.S's preconstructed pagetables. We copy the Xen L2's into |
1698 | * level2_ident_pgt, and level2_kernel_pgt. This means that only the |
1699 | * kernel has a physical mapping to start with - but that's enough to |
1700 | * get __va working. We need to fill in the rest of the physical |
1701 | * mapping once some sort of allocator has been set up. |
1702 | */ |
1703 | void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn) |
1704 | { |
1705 | pud_t *l3; |
1706 | pmd_t *l2; |
1707 | unsigned long addr[3]; |
1708 | unsigned long pt_base, pt_end; |
1709 | unsigned i; |
1710 | |
1711 | /* max_pfn_mapped is the last pfn mapped in the initial memory |
1712 | * mappings. Considering that on Xen after the kernel mappings we |
1713 | * have the mappings of some pages that don't exist in pfn space, we |
1714 | * set max_pfn_mapped to the last real pfn mapped. */ |
1715 | if (xen_start_info->mfn_list < __START_KERNEL_map) |
1716 | max_pfn_mapped = xen_start_info->first_p2m_pfn; |
1717 | else |
1718 | max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list)); |
1719 | |
1720 | pt_base = PFN_DOWN(__pa(xen_start_info->pt_base)); |
1721 | pt_end = pt_base + xen_start_info->nr_pt_frames; |
1722 | |
1723 | /* Zap identity mapping */ |
1724 | init_top_pgt[0] = __pgd(val: 0); |
1725 | |
1726 | /* Pre-constructed entries are in pfn, so convert to mfn */ |
1727 | /* L4[273] -> level3_ident_pgt */ |
1728 | /* L4[511] -> level3_kernel_pgt */ |
1729 | convert_pfn_mfn(v: init_top_pgt); |
1730 | |
1731 | /* L3_i[0] -> level2_ident_pgt */ |
1732 | convert_pfn_mfn(v: level3_ident_pgt); |
1733 | /* L3_k[510] -> level2_kernel_pgt */ |
1734 | /* L3_k[511] -> level2_fixmap_pgt */ |
1735 | convert_pfn_mfn(v: level3_kernel_pgt); |
1736 | |
1737 | /* L3_k[511][508-FIXMAP_PMD_NUM ... 507] -> level1_fixmap_pgt */ |
1738 | convert_pfn_mfn(v: level2_fixmap_pgt); |
1739 | |
1740 | /* We get [511][511] and have Xen's version of level2_kernel_pgt */ |
1741 | l3 = m2v(maddr: pgd[pgd_index(__START_KERNEL_map)].pgd); |
1742 | l2 = m2v(maddr: l3[pud_index(__START_KERNEL_map)].pud); |
1743 | |
1744 | addr[0] = (unsigned long)pgd; |
1745 | addr[1] = (unsigned long)l3; |
1746 | addr[2] = (unsigned long)l2; |
1747 | /* Graft it onto L4[273][0]. Note that we creating an aliasing problem: |
1748 | * Both L4[273][0] and L4[511][510] have entries that point to the same |
1749 | * L2 (PMD) tables. Meaning that if you modify it in __va space |
1750 | * it will be also modified in the __ka space! (But if you just |
1751 | * modify the PMD table to point to other PTE's or none, then you |
1752 | * are OK - which is what cleanup_highmap does) */ |
1753 | copy_page(to: level2_ident_pgt, from: l2); |
1754 | /* Graft it onto L4[511][510] */ |
1755 | copy_page(to: level2_kernel_pgt, from: l2); |
1756 | |
1757 | /* |
1758 | * Zap execute permission from the ident map. Due to the sharing of |
1759 | * L1 entries we need to do this in the L2. |
1760 | */ |
1761 | if (__supported_pte_mask & _PAGE_NX) { |
1762 | for (i = 0; i < PTRS_PER_PMD; ++i) { |
1763 | if (pmd_none(pmd: level2_ident_pgt[i])) |
1764 | continue; |
1765 | level2_ident_pgt[i] = pmd_set_flags(pmd: level2_ident_pgt[i], _PAGE_NX); |
1766 | } |
1767 | } |
1768 | |
1769 | /* Copy the initial P->M table mappings if necessary. */ |
1770 | i = pgd_index(xen_start_info->mfn_list); |
1771 | if (i && i < pgd_index(__START_KERNEL_map)) |
1772 | init_top_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i]; |
1773 | |
1774 | /* Make pagetable pieces RO */ |
1775 | set_page_prot(addr: init_top_pgt, PAGE_KERNEL_RO); |
1776 | set_page_prot(addr: level3_ident_pgt, PAGE_KERNEL_RO); |
1777 | set_page_prot(addr: level3_kernel_pgt, PAGE_KERNEL_RO); |
1778 | set_page_prot(addr: level2_ident_pgt, PAGE_KERNEL_RO); |
1779 | set_page_prot(addr: level2_kernel_pgt, PAGE_KERNEL_RO); |
1780 | set_page_prot(addr: level2_fixmap_pgt, PAGE_KERNEL_RO); |
1781 | |
1782 | for (i = 0; i < FIXMAP_PMD_NUM; i++) { |
1783 | set_page_prot(addr: level1_fixmap_pgt + i * PTRS_PER_PTE, |
1784 | PAGE_KERNEL_RO); |
1785 | } |
1786 | |
1787 | /* Pin down new L4 */ |
1788 | pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE, |
1789 | PFN_DOWN(__pa_symbol(init_top_pgt))); |
1790 | |
1791 | /* Unpin Xen-provided one */ |
1792 | pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd))); |
1793 | |
1794 | #ifdef CONFIG_X86_VSYSCALL_EMULATION |
1795 | /* Pin user vsyscall L3 */ |
1796 | set_page_prot(addr: level3_user_vsyscall, PAGE_KERNEL_RO); |
1797 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, |
1798 | PFN_DOWN(__pa_symbol(level3_user_vsyscall))); |
1799 | #endif |
1800 | |
1801 | /* |
1802 | * At this stage there can be no user pgd, and no page structure to |
1803 | * attach it to, so make sure we just set kernel pgd. |
1804 | */ |
1805 | xen_mc_batch(); |
1806 | __xen_write_cr3(kernel: true, __pa(init_top_pgt)); |
1807 | xen_mc_issue(mode: XEN_LAZY_CPU); |
1808 | |
1809 | /* We can't that easily rip out L3 and L2, as the Xen pagetables are |
1810 | * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ... for |
1811 | * the initial domain. For guests using the toolstack, they are in: |
1812 | * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only |
1813 | * rip out the [L4] (pgd), but for guests we shave off three pages. |
1814 | */ |
1815 | for (i = 0; i < ARRAY_SIZE(addr); i++) |
1816 | check_pt_base(pt_base: &pt_base, pt_end: &pt_end, addr: addr[i]); |
1817 | |
1818 | /* Our (by three pages) smaller Xen pagetable that we are using */ |
1819 | xen_pt_base = PFN_PHYS(pt_base); |
1820 | xen_pt_size = (pt_end - pt_base) * PAGE_SIZE; |
1821 | memblock_reserve(base: xen_pt_base, size: xen_pt_size); |
1822 | |
1823 | /* Revector the xen_start_info */ |
1824 | xen_start_info = (struct start_info *)__va(__pa(xen_start_info)); |
1825 | } |
1826 | |
1827 | /* |
1828 | * Read a value from a physical address. |
1829 | */ |
1830 | static unsigned long __init xen_read_phys_ulong(phys_addr_t addr) |
1831 | { |
1832 | unsigned long *vaddr; |
1833 | unsigned long val; |
1834 | |
1835 | vaddr = early_memremap_ro(phys_addr: addr, size: sizeof(val)); |
1836 | val = *vaddr; |
1837 | early_memunmap(addr: vaddr, size: sizeof(val)); |
1838 | return val; |
1839 | } |
1840 | |
1841 | /* |
1842 | * Translate a virtual address to a physical one without relying on mapped |
1843 | * page tables. Don't rely on big pages being aligned in (guest) physical |
1844 | * space! |
1845 | */ |
1846 | static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr) |
1847 | { |
1848 | phys_addr_t pa; |
1849 | pgd_t pgd; |
1850 | pud_t pud; |
1851 | pmd_t pmd; |
1852 | pte_t pte; |
1853 | |
1854 | pa = read_cr3_pa(); |
1855 | pgd = native_make_pgd(val: xen_read_phys_ulong(addr: pa + pgd_index(vaddr) * |
1856 | sizeof(pgd))); |
1857 | if (!pgd_present(pgd)) |
1858 | return 0; |
1859 | |
1860 | pa = pgd_val(pgd) & PTE_PFN_MASK; |
1861 | pud = native_make_pud(val: xen_read_phys_ulong(addr: pa + pud_index(address: vaddr) * |
1862 | sizeof(pud))); |
1863 | if (!pud_present(pud)) |
1864 | return 0; |
1865 | pa = pud_val(pud) & PTE_PFN_MASK; |
1866 | if (pud_large(pud)) |
1867 | return pa + (vaddr & ~PUD_MASK); |
1868 | |
1869 | pmd = native_make_pmd(val: xen_read_phys_ulong(addr: pa + pmd_index(address: vaddr) * |
1870 | sizeof(pmd))); |
1871 | if (!pmd_present(pmd)) |
1872 | return 0; |
1873 | pa = pmd_val(pmd) & PTE_PFN_MASK; |
1874 | if (pmd_large(pte: pmd)) |
1875 | return pa + (vaddr & ~PMD_MASK); |
1876 | |
1877 | pte = native_make_pte(val: xen_read_phys_ulong(addr: pa + pte_index(address: vaddr) * |
1878 | sizeof(pte))); |
1879 | if (!pte_present(a: pte)) |
1880 | return 0; |
1881 | pa = pte_pfn(pte) << PAGE_SHIFT; |
1882 | |
1883 | return pa | (vaddr & ~PAGE_MASK); |
1884 | } |
1885 | |
1886 | /* |
1887 | * Find a new area for the hypervisor supplied p2m list and relocate the p2m to |
1888 | * this area. |
1889 | */ |
1890 | void __init xen_relocate_p2m(void) |
1891 | { |
1892 | phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys; |
1893 | unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end; |
1894 | int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud; |
1895 | pte_t *pt; |
1896 | pmd_t *pmd; |
1897 | pud_t *pud; |
1898 | pgd_t *pgd; |
1899 | unsigned long *new_p2m; |
1900 | |
1901 | size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long)); |
1902 | n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT; |
1903 | n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT; |
1904 | n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT; |
1905 | n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT; |
1906 | n_frames = n_pte + n_pt + n_pmd + n_pud; |
1907 | |
1908 | new_area = xen_find_free_area(PFN_PHYS(n_frames)); |
1909 | if (!new_area) { |
1910 | xen_raw_console_write(str: "Can't find new memory area for p2m needed due to E820 map conflict\n" ); |
1911 | BUG(); |
1912 | } |
1913 | |
1914 | /* |
1915 | * Setup the page tables for addressing the new p2m list. |
1916 | * We have asked the hypervisor to map the p2m list at the user address |
1917 | * PUD_SIZE. It may have done so, or it may have used a kernel space |
1918 | * address depending on the Xen version. |
1919 | * To avoid any possible virtual address collision, just use |
1920 | * 2 * PUD_SIZE for the new area. |
1921 | */ |
1922 | pud_phys = new_area; |
1923 | pmd_phys = pud_phys + PFN_PHYS(n_pud); |
1924 | pt_phys = pmd_phys + PFN_PHYS(n_pmd); |
1925 | p2m_pfn = PFN_DOWN(pt_phys) + n_pt; |
1926 | |
1927 | pgd = __va(read_cr3_pa()); |
1928 | new_p2m = (unsigned long *)(2 * PGDIR_SIZE); |
1929 | for (idx_pud = 0; idx_pud < n_pud; idx_pud++) { |
1930 | pud = early_memremap(phys_addr: pud_phys, PAGE_SIZE); |
1931 | clear_page(page: pud); |
1932 | for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD); |
1933 | idx_pmd++) { |
1934 | pmd = early_memremap(phys_addr: pmd_phys, PAGE_SIZE); |
1935 | clear_page(page: pmd); |
1936 | for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD); |
1937 | idx_pt++) { |
1938 | pt = early_memremap(phys_addr: pt_phys, PAGE_SIZE); |
1939 | clear_page(page: pt); |
1940 | for (idx_pte = 0; |
1941 | idx_pte < min(n_pte, PTRS_PER_PTE); |
1942 | idx_pte++) { |
1943 | pt[idx_pte] = pfn_pte(page_nr: p2m_pfn, |
1944 | PAGE_KERNEL); |
1945 | p2m_pfn++; |
1946 | } |
1947 | n_pte -= PTRS_PER_PTE; |
1948 | early_memunmap(addr: pt, PAGE_SIZE); |
1949 | make_lowmem_page_readonly(__va(pt_phys)); |
1950 | pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, |
1951 | PFN_DOWN(pt_phys)); |
1952 | pmd[idx_pt] = __pmd(_PAGE_TABLE | pt_phys); |
1953 | pt_phys += PAGE_SIZE; |
1954 | } |
1955 | n_pt -= PTRS_PER_PMD; |
1956 | early_memunmap(addr: pmd, PAGE_SIZE); |
1957 | make_lowmem_page_readonly(__va(pmd_phys)); |
1958 | pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE, |
1959 | PFN_DOWN(pmd_phys)); |
1960 | pud[idx_pmd] = __pud(_PAGE_TABLE | pmd_phys); |
1961 | pmd_phys += PAGE_SIZE; |
1962 | } |
1963 | n_pmd -= PTRS_PER_PUD; |
1964 | early_memunmap(addr: pud, PAGE_SIZE); |
1965 | make_lowmem_page_readonly(__va(pud_phys)); |
1966 | pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys)); |
1967 | set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys)); |
1968 | pud_phys += PAGE_SIZE; |
1969 | } |
1970 | |
1971 | /* Now copy the old p2m info to the new area. */ |
1972 | memcpy(new_p2m, xen_p2m_addr, size); |
1973 | xen_p2m_addr = new_p2m; |
1974 | |
1975 | /* Release the old p2m list and set new list info. */ |
1976 | p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list)); |
1977 | BUG_ON(!p2m_pfn); |
1978 | p2m_pfn_end = p2m_pfn + PFN_DOWN(size); |
1979 | |
1980 | if (xen_start_info->mfn_list < __START_KERNEL_map) { |
1981 | pfn = xen_start_info->first_p2m_pfn; |
1982 | pfn_end = xen_start_info->first_p2m_pfn + |
1983 | xen_start_info->nr_p2m_frames; |
1984 | set_pgd(pgd + 1, __pgd(0)); |
1985 | } else { |
1986 | pfn = p2m_pfn; |
1987 | pfn_end = p2m_pfn_end; |
1988 | } |
1989 | |
1990 | memblock_phys_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn)); |
1991 | while (pfn < pfn_end) { |
1992 | if (pfn == p2m_pfn) { |
1993 | pfn = p2m_pfn_end; |
1994 | continue; |
1995 | } |
1996 | make_lowmem_page_readwrite(__va(PFN_PHYS(pfn))); |
1997 | pfn++; |
1998 | } |
1999 | |
2000 | xen_start_info->mfn_list = (unsigned long)xen_p2m_addr; |
2001 | xen_start_info->first_p2m_pfn = PFN_DOWN(new_area); |
2002 | xen_start_info->nr_p2m_frames = n_frames; |
2003 | } |
2004 | |
2005 | void __init xen_reserve_special_pages(void) |
2006 | { |
2007 | phys_addr_t paddr; |
2008 | |
2009 | memblock_reserve(__pa(xen_start_info), PAGE_SIZE); |
2010 | if (xen_start_info->store_mfn) { |
2011 | paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn)); |
2012 | memblock_reserve(base: paddr, PAGE_SIZE); |
2013 | } |
2014 | if (!xen_initial_domain()) { |
2015 | paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn)); |
2016 | memblock_reserve(base: paddr, PAGE_SIZE); |
2017 | } |
2018 | } |
2019 | |
2020 | void __init xen_pt_check_e820(void) |
2021 | { |
2022 | if (xen_is_e820_reserved(start: xen_pt_base, size: xen_pt_size)) { |
2023 | xen_raw_console_write(str: "Xen hypervisor allocated page table memory conflicts with E820 map\n" ); |
2024 | BUG(); |
2025 | } |
2026 | } |
2027 | |
2028 | static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss; |
2029 | |
2030 | static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot) |
2031 | { |
2032 | pte_t pte; |
2033 | unsigned long vaddr; |
2034 | |
2035 | phys >>= PAGE_SHIFT; |
2036 | |
2037 | switch (idx) { |
2038 | case FIX_BTMAP_END ... FIX_BTMAP_BEGIN: |
2039 | #ifdef CONFIG_X86_VSYSCALL_EMULATION |
2040 | case VSYSCALL_PAGE: |
2041 | #endif |
2042 | /* All local page mappings */ |
2043 | pte = pfn_pte(page_nr: phys, pgprot: prot); |
2044 | break; |
2045 | |
2046 | #ifdef CONFIG_X86_LOCAL_APIC |
2047 | case FIX_APIC_BASE: /* maps dummy local APIC */ |
2048 | pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL); |
2049 | break; |
2050 | #endif |
2051 | |
2052 | #ifdef CONFIG_X86_IO_APIC |
2053 | case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END: |
2054 | /* |
2055 | * We just don't map the IO APIC - all access is via |
2056 | * hypercalls. Keep the address in the pte for reference. |
2057 | */ |
2058 | pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL); |
2059 | break; |
2060 | #endif |
2061 | |
2062 | case FIX_PARAVIRT_BOOTMAP: |
2063 | /* This is an MFN, but it isn't an IO mapping from the |
2064 | IO domain */ |
2065 | pte = mfn_pte(page_nr: phys, pgprot: prot); |
2066 | break; |
2067 | |
2068 | default: |
2069 | /* By default, set_fixmap is used for hardware mappings */ |
2070 | pte = mfn_pte(page_nr: phys, pgprot: prot); |
2071 | break; |
2072 | } |
2073 | |
2074 | vaddr = __fix_to_virt(idx); |
2075 | if (HYPERVISOR_update_va_mapping(va: vaddr, new_val: pte, UVMF_INVLPG)) |
2076 | BUG(); |
2077 | |
2078 | #ifdef CONFIG_X86_VSYSCALL_EMULATION |
2079 | /* Replicate changes to map the vsyscall page into the user |
2080 | pagetable vsyscall mapping. */ |
2081 | if (idx == VSYSCALL_PAGE) |
2082 | set_pte_vaddr_pud(pud_page: level3_user_vsyscall, vaddr, new_pte: pte); |
2083 | #endif |
2084 | } |
2085 | |
2086 | static void xen_enter_lazy_mmu(void) |
2087 | { |
2088 | enter_lazy(mode: XEN_LAZY_MMU); |
2089 | } |
2090 | |
2091 | static void xen_flush_lazy_mmu(void) |
2092 | { |
2093 | preempt_disable(); |
2094 | |
2095 | if (xen_get_lazy_mode() == XEN_LAZY_MMU) { |
2096 | arch_leave_lazy_mmu_mode(); |
2097 | arch_enter_lazy_mmu_mode(); |
2098 | } |
2099 | |
2100 | preempt_enable(); |
2101 | } |
2102 | |
2103 | static void __init xen_post_allocator_init(void) |
2104 | { |
2105 | pv_ops.mmu.set_pte = xen_set_pte; |
2106 | pv_ops.mmu.set_pmd = xen_set_pmd; |
2107 | pv_ops.mmu.set_pud = xen_set_pud; |
2108 | pv_ops.mmu.set_p4d = xen_set_p4d; |
2109 | |
2110 | /* This will work as long as patching hasn't happened yet |
2111 | (which it hasn't) */ |
2112 | pv_ops.mmu.alloc_pte = xen_alloc_pte; |
2113 | pv_ops.mmu.alloc_pmd = xen_alloc_pmd; |
2114 | pv_ops.mmu.release_pte = xen_release_pte; |
2115 | pv_ops.mmu.release_pmd = xen_release_pmd; |
2116 | pv_ops.mmu.alloc_pud = xen_alloc_pud; |
2117 | pv_ops.mmu.release_pud = xen_release_pud; |
2118 | pv_ops.mmu.make_pte = PV_CALLEE_SAVE(xen_make_pte); |
2119 | |
2120 | pv_ops.mmu.write_cr3 = &xen_write_cr3; |
2121 | } |
2122 | |
2123 | static void xen_leave_lazy_mmu(void) |
2124 | { |
2125 | preempt_disable(); |
2126 | xen_mc_flush(); |
2127 | leave_lazy(mode: XEN_LAZY_MMU); |
2128 | preempt_enable(); |
2129 | } |
2130 | |
2131 | static const typeof(pv_ops) xen_mmu_ops __initconst = { |
2132 | .mmu = { |
2133 | .read_cr2 = __PV_IS_CALLEE_SAVE(xen_read_cr2), |
2134 | .write_cr2 = xen_write_cr2, |
2135 | |
2136 | .read_cr3 = xen_read_cr3, |
2137 | .write_cr3 = xen_write_cr3_init, |
2138 | |
2139 | .flush_tlb_user = xen_flush_tlb, |
2140 | .flush_tlb_kernel = xen_flush_tlb, |
2141 | .flush_tlb_one_user = xen_flush_tlb_one_user, |
2142 | .flush_tlb_multi = xen_flush_tlb_multi, |
2143 | .tlb_remove_table = tlb_remove_table, |
2144 | |
2145 | .pgd_alloc = xen_pgd_alloc, |
2146 | .pgd_free = xen_pgd_free, |
2147 | |
2148 | .alloc_pte = xen_alloc_pte_init, |
2149 | .release_pte = xen_release_pte_init, |
2150 | .alloc_pmd = xen_alloc_pmd_init, |
2151 | .release_pmd = xen_release_pmd_init, |
2152 | |
2153 | .set_pte = xen_set_pte_init, |
2154 | .set_pmd = xen_set_pmd_hyper, |
2155 | |
2156 | .ptep_modify_prot_start = xen_ptep_modify_prot_start, |
2157 | .ptep_modify_prot_commit = xen_ptep_modify_prot_commit, |
2158 | |
2159 | .pte_val = PV_CALLEE_SAVE(xen_pte_val), |
2160 | .pgd_val = PV_CALLEE_SAVE(xen_pgd_val), |
2161 | |
2162 | .make_pte = PV_CALLEE_SAVE(xen_make_pte_init), |
2163 | .make_pgd = PV_CALLEE_SAVE(xen_make_pgd), |
2164 | |
2165 | .set_pud = xen_set_pud_hyper, |
2166 | |
2167 | .make_pmd = PV_CALLEE_SAVE(xen_make_pmd), |
2168 | .pmd_val = PV_CALLEE_SAVE(xen_pmd_val), |
2169 | |
2170 | .pud_val = PV_CALLEE_SAVE(xen_pud_val), |
2171 | .make_pud = PV_CALLEE_SAVE(xen_make_pud), |
2172 | .set_p4d = xen_set_p4d_hyper, |
2173 | |
2174 | .alloc_pud = xen_alloc_pmd_init, |
2175 | .release_pud = xen_release_pmd_init, |
2176 | |
2177 | #if CONFIG_PGTABLE_LEVELS >= 5 |
2178 | .p4d_val = PV_CALLEE_SAVE(xen_p4d_val), |
2179 | .make_p4d = PV_CALLEE_SAVE(xen_make_p4d), |
2180 | #endif |
2181 | |
2182 | .enter_mmap = xen_enter_mmap, |
2183 | .exit_mmap = xen_exit_mmap, |
2184 | |
2185 | .lazy_mode = { |
2186 | .enter = xen_enter_lazy_mmu, |
2187 | .leave = xen_leave_lazy_mmu, |
2188 | .flush = xen_flush_lazy_mmu, |
2189 | }, |
2190 | |
2191 | .set_fixmap = xen_set_fixmap, |
2192 | }, |
2193 | }; |
2194 | |
2195 | void __init xen_init_mmu_ops(void) |
2196 | { |
2197 | x86_init.paging.pagetable_init = xen_pagetable_init; |
2198 | x86_init.hyper.init_after_bootmem = xen_after_bootmem; |
2199 | |
2200 | pv_ops.mmu = xen_mmu_ops.mmu; |
2201 | |
2202 | memset(dummy_mapping, 0xff, PAGE_SIZE); |
2203 | } |
2204 | |
2205 | /* Protected by xen_reservation_lock. */ |
2206 | #define MAX_CONTIG_ORDER 9 /* 2MB */ |
2207 | static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER]; |
2208 | |
2209 | #define VOID_PTE (mfn_pte(0, __pgprot(0))) |
2210 | static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order, |
2211 | unsigned long *in_frames, |
2212 | unsigned long *out_frames) |
2213 | { |
2214 | int i; |
2215 | struct multicall_space mcs; |
2216 | |
2217 | xen_mc_batch(); |
2218 | for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) { |
2219 | mcs = __xen_mc_entry(args: 0); |
2220 | |
2221 | if (in_frames) |
2222 | in_frames[i] = virt_to_mfn((void *)vaddr); |
2223 | |
2224 | MULTI_update_va_mapping(mcl: mcs.mc, va: vaddr, VOID_PTE, flags: 0); |
2225 | __set_phys_to_machine(pfn: virt_to_pfn(v: (void *)vaddr), INVALID_P2M_ENTRY); |
2226 | |
2227 | if (out_frames) |
2228 | out_frames[i] = virt_to_pfn(v: (void *)vaddr); |
2229 | } |
2230 | xen_mc_issue(mode: 0); |
2231 | } |
2232 | |
2233 | /* |
2234 | * Update the pfn-to-mfn mappings for a virtual address range, either to |
2235 | * point to an array of mfns, or contiguously from a single starting |
2236 | * mfn. |
2237 | */ |
2238 | static void xen_remap_exchanged_ptes(unsigned long vaddr, int order, |
2239 | unsigned long *mfns, |
2240 | unsigned long first_mfn) |
2241 | { |
2242 | unsigned i, limit; |
2243 | unsigned long mfn; |
2244 | |
2245 | xen_mc_batch(); |
2246 | |
2247 | limit = 1u << order; |
2248 | for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) { |
2249 | struct multicall_space mcs; |
2250 | unsigned flags; |
2251 | |
2252 | mcs = __xen_mc_entry(args: 0); |
2253 | if (mfns) |
2254 | mfn = mfns[i]; |
2255 | else |
2256 | mfn = first_mfn + i; |
2257 | |
2258 | if (i < (limit - 1)) |
2259 | flags = 0; |
2260 | else { |
2261 | if (order == 0) |
2262 | flags = UVMF_INVLPG | UVMF_ALL; |
2263 | else |
2264 | flags = UVMF_TLB_FLUSH | UVMF_ALL; |
2265 | } |
2266 | |
2267 | MULTI_update_va_mapping(mcl: mcs.mc, va: vaddr, |
2268 | new_val: mfn_pte(page_nr: mfn, PAGE_KERNEL), flags); |
2269 | |
2270 | set_phys_to_machine(pfn: virt_to_pfn(v: (void *)vaddr), mfn); |
2271 | } |
2272 | |
2273 | xen_mc_issue(mode: 0); |
2274 | } |
2275 | |
2276 | /* |
2277 | * Perform the hypercall to exchange a region of our pfns to point to |
2278 | * memory with the required contiguous alignment. Takes the pfns as |
2279 | * input, and populates mfns as output. |
2280 | * |
2281 | * Returns a success code indicating whether the hypervisor was able to |
2282 | * satisfy the request or not. |
2283 | */ |
2284 | static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in, |
2285 | unsigned long *pfns_in, |
2286 | unsigned long extents_out, |
2287 | unsigned int order_out, |
2288 | unsigned long *mfns_out, |
2289 | unsigned int address_bits) |
2290 | { |
2291 | long rc; |
2292 | int success; |
2293 | |
2294 | struct xen_memory_exchange exchange = { |
2295 | .in = { |
2296 | .nr_extents = extents_in, |
2297 | .extent_order = order_in, |
2298 | .extent_start = pfns_in, |
2299 | .domid = DOMID_SELF |
2300 | }, |
2301 | .out = { |
2302 | .nr_extents = extents_out, |
2303 | .extent_order = order_out, |
2304 | .extent_start = mfns_out, |
2305 | .address_bits = address_bits, |
2306 | .domid = DOMID_SELF |
2307 | } |
2308 | }; |
2309 | |
2310 | BUG_ON(extents_in << order_in != extents_out << order_out); |
2311 | |
2312 | rc = HYPERVISOR_memory_op(XENMEM_exchange, arg: &exchange); |
2313 | success = (exchange.nr_exchanged == extents_in); |
2314 | |
2315 | BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0))); |
2316 | BUG_ON(success && (rc != 0)); |
2317 | |
2318 | return success; |
2319 | } |
2320 | |
2321 | int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order, |
2322 | unsigned int address_bits, |
2323 | dma_addr_t *dma_handle) |
2324 | { |
2325 | unsigned long *in_frames = discontig_frames, out_frame; |
2326 | unsigned long flags; |
2327 | int success; |
2328 | unsigned long vstart = (unsigned long)phys_to_virt(address: pstart); |
2329 | |
2330 | if (unlikely(order > MAX_CONTIG_ORDER)) |
2331 | return -ENOMEM; |
2332 | |
2333 | memset((void *) vstart, 0, PAGE_SIZE << order); |
2334 | |
2335 | spin_lock_irqsave(&xen_reservation_lock, flags); |
2336 | |
2337 | /* 1. Zap current PTEs, remembering MFNs. */ |
2338 | xen_zap_pfn_range(vaddr: vstart, order, in_frames, NULL); |
2339 | |
2340 | /* 2. Get a new contiguous memory extent. */ |
2341 | out_frame = virt_to_pfn(v: (void *)vstart); |
2342 | success = xen_exchange_memory(extents_in: 1UL << order, order_in: 0, pfns_in: in_frames, |
2343 | extents_out: 1, order_out: order, mfns_out: &out_frame, |
2344 | address_bits); |
2345 | |
2346 | /* 3. Map the new extent in place of old pages. */ |
2347 | if (success) |
2348 | xen_remap_exchanged_ptes(vaddr: vstart, order, NULL, first_mfn: out_frame); |
2349 | else |
2350 | xen_remap_exchanged_ptes(vaddr: vstart, order, mfns: in_frames, first_mfn: 0); |
2351 | |
2352 | spin_unlock_irqrestore(lock: &xen_reservation_lock, flags); |
2353 | |
2354 | *dma_handle = virt_to_machine(vstart).maddr; |
2355 | return success ? 0 : -ENOMEM; |
2356 | } |
2357 | |
2358 | void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order) |
2359 | { |
2360 | unsigned long *out_frames = discontig_frames, in_frame; |
2361 | unsigned long flags; |
2362 | int success; |
2363 | unsigned long vstart; |
2364 | |
2365 | if (unlikely(order > MAX_CONTIG_ORDER)) |
2366 | return; |
2367 | |
2368 | vstart = (unsigned long)phys_to_virt(address: pstart); |
2369 | memset((void *) vstart, 0, PAGE_SIZE << order); |
2370 | |
2371 | spin_lock_irqsave(&xen_reservation_lock, flags); |
2372 | |
2373 | /* 1. Find start MFN of contiguous extent. */ |
2374 | in_frame = virt_to_mfn((void *)vstart); |
2375 | |
2376 | /* 2. Zap current PTEs. */ |
2377 | xen_zap_pfn_range(vaddr: vstart, order, NULL, out_frames); |
2378 | |
2379 | /* 3. Do the exchange for non-contiguous MFNs. */ |
2380 | success = xen_exchange_memory(extents_in: 1, order_in: order, pfns_in: &in_frame, extents_out: 1UL << order, |
2381 | order_out: 0, mfns_out: out_frames, address_bits: 0); |
2382 | |
2383 | /* 4. Map new pages in place of old pages. */ |
2384 | if (success) |
2385 | xen_remap_exchanged_ptes(vaddr: vstart, order, mfns: out_frames, first_mfn: 0); |
2386 | else |
2387 | xen_remap_exchanged_ptes(vaddr: vstart, order, NULL, first_mfn: in_frame); |
2388 | |
2389 | spin_unlock_irqrestore(lock: &xen_reservation_lock, flags); |
2390 | } |
2391 | |
2392 | static noinline void xen_flush_tlb_all(void) |
2393 | { |
2394 | struct mmuext_op *op; |
2395 | struct multicall_space mcs; |
2396 | |
2397 | preempt_disable(); |
2398 | |
2399 | mcs = xen_mc_entry(args: sizeof(*op)); |
2400 | |
2401 | op = mcs.args; |
2402 | op->cmd = MMUEXT_TLB_FLUSH_ALL; |
2403 | MULTI_mmuext_op(mcl: mcs.mc, op, count: 1, NULL, DOMID_SELF); |
2404 | |
2405 | xen_mc_issue(mode: XEN_LAZY_MMU); |
2406 | |
2407 | preempt_enable(); |
2408 | } |
2409 | |
2410 | #define REMAP_BATCH_SIZE 16 |
2411 | |
2412 | struct remap_data { |
2413 | xen_pfn_t *pfn; |
2414 | bool contiguous; |
2415 | bool no_translate; |
2416 | pgprot_t prot; |
2417 | struct mmu_update *mmu_update; |
2418 | }; |
2419 | |
2420 | static int remap_area_pfn_pte_fn(pte_t *ptep, unsigned long addr, void *data) |
2421 | { |
2422 | struct remap_data *rmd = data; |
2423 | pte_t pte = pte_mkspecial(pte: mfn_pte(page_nr: *rmd->pfn, pgprot: rmd->prot)); |
2424 | |
2425 | /* |
2426 | * If we have a contiguous range, just update the pfn itself, |
2427 | * else update pointer to be "next pfn". |
2428 | */ |
2429 | if (rmd->contiguous) |
2430 | (*rmd->pfn)++; |
2431 | else |
2432 | rmd->pfn++; |
2433 | |
2434 | rmd->mmu_update->ptr = virt_to_machine(ptep).maddr; |
2435 | rmd->mmu_update->ptr |= rmd->no_translate ? |
2436 | MMU_PT_UPDATE_NO_TRANSLATE : |
2437 | MMU_NORMAL_PT_UPDATE; |
2438 | rmd->mmu_update->val = pte_val_ma(pte); |
2439 | rmd->mmu_update++; |
2440 | |
2441 | return 0; |
2442 | } |
2443 | |
2444 | int xen_remap_pfn(struct vm_area_struct *vma, unsigned long addr, |
2445 | xen_pfn_t *pfn, int nr, int *err_ptr, pgprot_t prot, |
2446 | unsigned int domid, bool no_translate) |
2447 | { |
2448 | int err = 0; |
2449 | struct remap_data rmd; |
2450 | struct mmu_update mmu_update[REMAP_BATCH_SIZE]; |
2451 | unsigned long range; |
2452 | int mapped = 0; |
2453 | |
2454 | BUG_ON(!((vma->vm_flags & (VM_PFNMAP | VM_IO)) == (VM_PFNMAP | VM_IO))); |
2455 | |
2456 | rmd.pfn = pfn; |
2457 | rmd.prot = prot; |
2458 | /* |
2459 | * We use the err_ptr to indicate if there we are doing a contiguous |
2460 | * mapping or a discontiguous mapping. |
2461 | */ |
2462 | rmd.contiguous = !err_ptr; |
2463 | rmd.no_translate = no_translate; |
2464 | |
2465 | while (nr) { |
2466 | int index = 0; |
2467 | int done = 0; |
2468 | int batch = min(REMAP_BATCH_SIZE, nr); |
2469 | int batch_left = batch; |
2470 | |
2471 | range = (unsigned long)batch << PAGE_SHIFT; |
2472 | |
2473 | rmd.mmu_update = mmu_update; |
2474 | err = apply_to_page_range(mm: vma->vm_mm, address: addr, size: range, |
2475 | fn: remap_area_pfn_pte_fn, data: &rmd); |
2476 | if (err) |
2477 | goto out; |
2478 | |
2479 | /* |
2480 | * We record the error for each page that gives an error, but |
2481 | * continue mapping until the whole set is done |
2482 | */ |
2483 | do { |
2484 | int i; |
2485 | |
2486 | err = HYPERVISOR_mmu_update(req: &mmu_update[index], |
2487 | count: batch_left, success_count: &done, domid); |
2488 | |
2489 | /* |
2490 | * @err_ptr may be the same buffer as @gfn, so |
2491 | * only clear it after each chunk of @gfn is |
2492 | * used. |
2493 | */ |
2494 | if (err_ptr) { |
2495 | for (i = index; i < index + done; i++) |
2496 | err_ptr[i] = 0; |
2497 | } |
2498 | if (err < 0) { |
2499 | if (!err_ptr) |
2500 | goto out; |
2501 | err_ptr[i] = err; |
2502 | done++; /* Skip failed frame. */ |
2503 | } else |
2504 | mapped += done; |
2505 | batch_left -= done; |
2506 | index += done; |
2507 | } while (batch_left); |
2508 | |
2509 | nr -= batch; |
2510 | addr += range; |
2511 | if (err_ptr) |
2512 | err_ptr += batch; |
2513 | cond_resched(); |
2514 | } |
2515 | out: |
2516 | |
2517 | xen_flush_tlb_all(); |
2518 | |
2519 | return err < 0 ? err : mapped; |
2520 | } |
2521 | EXPORT_SYMBOL_GPL(xen_remap_pfn); |
2522 | |
2523 | #ifdef CONFIG_KEXEC_CORE |
2524 | phys_addr_t paddr_vmcoreinfo_note(void) |
2525 | { |
2526 | if (xen_pv_domain()) |
2527 | return virt_to_machine(vmcoreinfo_note).maddr; |
2528 | else |
2529 | return __pa(vmcoreinfo_note); |
2530 | } |
2531 | #endif /* CONFIG_KEXEC_CORE */ |
2532 | |