4 * This file contains the various mmu fetch and update operations.
5 * The most important job they must perform is the mapping between the
6 * domain's pfn and the overall machine mfns.
8 * Xen allows guests to directly update the pagetable, in a controlled
9 * fashion. In other words, the guest modifies the same pagetable
10 * that the CPU actually uses, which eliminates the overhead of having
11 * a separate shadow pagetable.
13 * In order to allow this, it falls on the guest domain to map its
14 * notion of a "physical" pfn - which is just a domain-local linear
15 * address - into a real "machine address" which the CPU's MMU can
18 * A pgd_t/pmd_t/pte_t will typically contain an mfn, and so can be
19 * inserted directly into the pagetable. When creating a new
20 * pte/pmd/pgd, it converts the passed pfn into an mfn. Conversely,
21 * when reading the content back with __(pgd|pmd|pte)_val, it converts
22 * the mfn back into a pfn.
24 * The other constraint is that all pages which make up a pagetable
25 * must be mapped read-only in the guest. This prevents uncontrolled
26 * guest updates to the pagetable. Xen strictly enforces this, and
27 * will disallow any pagetable update which will end up mapping a
28 * pagetable page RW, and will disallow using any writable page as a
31 * Naively, when loading %cr3 with the base of a new pagetable, Xen
32 * would need to validate the whole pagetable before going on.
33 * Naturally, this is quite slow. The solution is to "pin" a
34 * pagetable, which enforces all the constraints on the pagetable even
35 * when it is not actively in use. This menas that Xen can be assured
36 * that it is still valid when you do load it into %cr3, and doesn't
37 * need to revalidate it.
39 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
41 #include <linux/sched/mm.h>
42 #include <linux/highmem.h>
43 #include <linux/debugfs.h>
44 #include <linux/bug.h>
45 #include <linux/vmalloc.h>
46 #include <linux/export.h>
47 #include <linux/init.h>
48 #include <linux/gfp.h>
49 #include <linux/memblock.h>
50 #include <linux/seq_file.h>
51 #include <linux/crash_dump.h>
52 #ifdef CONFIG_KEXEC_CORE
53 #include <linux/kexec.h>
56 #include <trace/events/xen.h>
58 #include <asm/pgtable.h>
59 #include <asm/tlbflush.h>
60 #include <asm/fixmap.h>
61 #include <asm/mmu_context.h>
62 #include <asm/setup.h>
63 #include <asm/paravirt.h>
64 #include <asm/e820/api.h>
65 #include <asm/linkage.h>
72 #include <asm/xen/hypercall.h>
73 #include <asm/xen/hypervisor.h>
77 #include <xen/interface/xen.h>
78 #include <xen/interface/hvm/hvm_op.h>
79 #include <xen/interface/version.h>
80 #include <xen/interface/memory.h>
81 #include <xen/hvc-console.h>
83 #include "multicalls.h"
89 * Identity map, in addition to plain kernel map. This needs to be
90 * large enough to allocate page table pages to allocate the rest.
91 * Each page can map 2MB.
93 #define LEVEL1_IDENT_ENTRIES (PTRS_PER_PTE * 4)
94 static RESERVE_BRK_ARRAY(pte_t, level1_ident_pgt, LEVEL1_IDENT_ENTRIES);
97 /* l3 pud for userspace vsyscall mapping */
98 static pud_t level3_user_vsyscall[PTRS_PER_PUD] __page_aligned_bss;
99 #endif /* CONFIG_X86_64 */
102 * Note about cr3 (pagetable base) values:
104 * xen_cr3 contains the current logical cr3 value; it contains the
105 * last set cr3. This may not be the current effective cr3, because
106 * its update may be being lazily deferred. However, a vcpu looking
107 * at its own cr3 can use this value knowing that it everything will
108 * be self-consistent.
110 * xen_current_cr3 contains the actual vcpu cr3; it is set once the
111 * hypercall to set the vcpu cr3 is complete (so it may be a little
112 * out of date, but it will never be set early). If one vcpu is
113 * looking at another vcpu's cr3 value, it should use this variable.
115 DEFINE_PER_CPU(unsigned long, xen_cr3); /* cr3 stored as physaddr */
116 DEFINE_PER_CPU(unsigned long, xen_current_cr3); /* actual vcpu cr3 */
118 static phys_addr_t xen_pt_base, xen_pt_size __initdata;
120 static DEFINE_STATIC_KEY_FALSE(xen_struct_pages_ready);
123 * Just beyond the highest usermode address. STACK_TOP_MAX has a
124 * redzone above it, so round it up to a PGD boundary.
126 #define USER_LIMIT ((STACK_TOP_MAX + PGDIR_SIZE - 1) & PGDIR_MASK)
128 void make_lowmem_page_readonly(void *vaddr)
131 unsigned long address = (unsigned long)vaddr;
134 pte = lookup_address(address, &level);
136 return; /* vaddr missing */
138 ptev = pte_wrprotect(*pte);
140 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
144 void make_lowmem_page_readwrite(void *vaddr)
147 unsigned long address = (unsigned long)vaddr;
150 pte = lookup_address(address, &level);
152 return; /* vaddr missing */
154 ptev = pte_mkwrite(*pte);
156 if (HYPERVISOR_update_va_mapping(address, ptev, 0))
162 * During early boot all page table pages are pinned, but we do not have struct
163 * pages, so return true until struct pages are ready.
165 static bool xen_page_pinned(void *ptr)
167 if (static_branch_likely(&xen_struct_pages_ready)) {
168 struct page *page = virt_to_page(ptr);
170 return PagePinned(page);
175 static void xen_extend_mmu_update(const struct mmu_update *update)
177 struct multicall_space mcs;
178 struct mmu_update *u;
180 mcs = xen_mc_extend_args(__HYPERVISOR_mmu_update, sizeof(*u));
182 if (mcs.mc != NULL) {
185 mcs = __xen_mc_entry(sizeof(*u));
186 MULTI_mmu_update(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
193 static void xen_extend_mmuext_op(const struct mmuext_op *op)
195 struct multicall_space mcs;
198 mcs = xen_mc_extend_args(__HYPERVISOR_mmuext_op, sizeof(*u));
200 if (mcs.mc != NULL) {
203 mcs = __xen_mc_entry(sizeof(*u));
204 MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
211 static void xen_set_pmd_hyper(pmd_t *ptr, pmd_t val)
219 /* ptr may be ioremapped for 64-bit pagetable setup */
220 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
221 u.val = pmd_val_ma(val);
222 xen_extend_mmu_update(&u);
224 xen_mc_issue(PARAVIRT_LAZY_MMU);
229 static void xen_set_pmd(pmd_t *ptr, pmd_t val)
231 trace_xen_mmu_set_pmd(ptr, val);
233 /* If page is not pinned, we can just update the entry
235 if (!xen_page_pinned(ptr)) {
240 xen_set_pmd_hyper(ptr, val);
244 * Associate a virtual page frame with a given physical page frame
245 * and protection flags for that frame.
247 void set_pte_mfn(unsigned long vaddr, unsigned long mfn, pgprot_t flags)
249 set_pte_vaddr(vaddr, mfn_pte(mfn, flags));
252 static bool xen_batched_set_pte(pte_t *ptep, pte_t pteval)
256 if (paravirt_get_lazy_mode() != PARAVIRT_LAZY_MMU)
261 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
262 u.val = pte_val_ma(pteval);
263 xen_extend_mmu_update(&u);
265 xen_mc_issue(PARAVIRT_LAZY_MMU);
270 static inline void __xen_set_pte(pte_t *ptep, pte_t pteval)
272 if (!xen_batched_set_pte(ptep, pteval)) {
274 * Could call native_set_pte() here and trap and
275 * emulate the PTE write but with 32-bit guests this
276 * needs two traps (one for each of the two 32-bit
277 * words in the PTE) so do one hypercall directly
282 u.ptr = virt_to_machine(ptep).maddr | MMU_NORMAL_PT_UPDATE;
283 u.val = pte_val_ma(pteval);
284 HYPERVISOR_mmu_update(&u, 1, NULL, DOMID_SELF);
288 static void xen_set_pte(pte_t *ptep, pte_t pteval)
290 trace_xen_mmu_set_pte(ptep, pteval);
291 __xen_set_pte(ptep, pteval);
294 static void xen_set_pte_at(struct mm_struct *mm, unsigned long addr,
295 pte_t *ptep, pte_t pteval)
297 trace_xen_mmu_set_pte_at(mm, addr, ptep, pteval);
298 __xen_set_pte(ptep, pteval);
301 pte_t xen_ptep_modify_prot_start(struct mm_struct *mm,
302 unsigned long addr, pte_t *ptep)
304 /* Just return the pte as-is. We preserve the bits on commit */
305 trace_xen_mmu_ptep_modify_prot_start(mm, addr, ptep, *ptep);
309 void xen_ptep_modify_prot_commit(struct mm_struct *mm, unsigned long addr,
310 pte_t *ptep, pte_t pte)
314 trace_xen_mmu_ptep_modify_prot_commit(mm, addr, ptep, pte);
317 u.ptr = virt_to_machine(ptep).maddr | MMU_PT_UPDATE_PRESERVE_AD;
318 u.val = pte_val_ma(pte);
319 xen_extend_mmu_update(&u);
321 xen_mc_issue(PARAVIRT_LAZY_MMU);
324 /* Assume pteval_t is equivalent to all the other *val_t types. */
325 static pteval_t pte_mfn_to_pfn(pteval_t val)
327 if (val & _PAGE_PRESENT) {
328 unsigned long mfn = (val & XEN_PTE_MFN_MASK) >> PAGE_SHIFT;
329 unsigned long pfn = mfn_to_pfn(mfn);
331 pteval_t flags = val & PTE_FLAGS_MASK;
332 if (unlikely(pfn == ~0))
333 val = flags & ~_PAGE_PRESENT;
335 val = ((pteval_t)pfn << PAGE_SHIFT) | flags;
341 static pteval_t pte_pfn_to_mfn(pteval_t val)
343 if (val & _PAGE_PRESENT) {
344 unsigned long pfn = (val & PTE_PFN_MASK) >> PAGE_SHIFT;
345 pteval_t flags = val & PTE_FLAGS_MASK;
348 mfn = __pfn_to_mfn(pfn);
351 * If there's no mfn for the pfn, then just create an
352 * empty non-present pte. Unfortunately this loses
353 * information about the original pfn, so
354 * pte_mfn_to_pfn is asymmetric.
356 if (unlikely(mfn == INVALID_P2M_ENTRY)) {
360 mfn &= ~(FOREIGN_FRAME_BIT | IDENTITY_FRAME_BIT);
361 val = ((pteval_t)mfn << PAGE_SHIFT) | flags;
367 __visible pteval_t xen_pte_val(pte_t pte)
369 pteval_t pteval = pte.pte;
371 return pte_mfn_to_pfn(pteval);
373 PV_CALLEE_SAVE_REGS_THUNK(xen_pte_val);
375 __visible pgdval_t xen_pgd_val(pgd_t pgd)
377 return pte_mfn_to_pfn(pgd.pgd);
379 PV_CALLEE_SAVE_REGS_THUNK(xen_pgd_val);
381 __visible pte_t xen_make_pte(pteval_t pte)
383 pte = pte_pfn_to_mfn(pte);
385 return native_make_pte(pte);
387 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte);
389 __visible pgd_t xen_make_pgd(pgdval_t pgd)
391 pgd = pte_pfn_to_mfn(pgd);
392 return native_make_pgd(pgd);
394 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pgd);
396 __visible pmdval_t xen_pmd_val(pmd_t pmd)
398 return pte_mfn_to_pfn(pmd.pmd);
400 PV_CALLEE_SAVE_REGS_THUNK(xen_pmd_val);
402 static void xen_set_pud_hyper(pud_t *ptr, pud_t val)
410 /* ptr may be ioremapped for 64-bit pagetable setup */
411 u.ptr = arbitrary_virt_to_machine(ptr).maddr;
412 u.val = pud_val_ma(val);
413 xen_extend_mmu_update(&u);
415 xen_mc_issue(PARAVIRT_LAZY_MMU);
420 static void xen_set_pud(pud_t *ptr, pud_t val)
422 trace_xen_mmu_set_pud(ptr, val);
424 /* If page is not pinned, we can just update the entry
426 if (!xen_page_pinned(ptr)) {
431 xen_set_pud_hyper(ptr, val);
434 #ifdef CONFIG_X86_PAE
435 static void xen_set_pte_atomic(pte_t *ptep, pte_t pte)
437 trace_xen_mmu_set_pte_atomic(ptep, pte);
438 set_64bit((u64 *)ptep, native_pte_val(pte));
441 static void xen_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
443 trace_xen_mmu_pte_clear(mm, addr, ptep);
444 if (!xen_batched_set_pte(ptep, native_make_pte(0)))
445 native_pte_clear(mm, addr, ptep);
448 static void xen_pmd_clear(pmd_t *pmdp)
450 trace_xen_mmu_pmd_clear(pmdp);
451 set_pmd(pmdp, __pmd(0));
453 #endif /* CONFIG_X86_PAE */
455 __visible pmd_t xen_make_pmd(pmdval_t pmd)
457 pmd = pte_pfn_to_mfn(pmd);
458 return native_make_pmd(pmd);
460 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pmd);
463 __visible pudval_t xen_pud_val(pud_t pud)
465 return pte_mfn_to_pfn(pud.pud);
467 PV_CALLEE_SAVE_REGS_THUNK(xen_pud_val);
469 __visible pud_t xen_make_pud(pudval_t pud)
471 pud = pte_pfn_to_mfn(pud);
473 return native_make_pud(pud);
475 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pud);
477 static pgd_t *xen_get_user_pgd(pgd_t *pgd)
479 pgd_t *pgd_page = (pgd_t *)(((unsigned long)pgd) & PAGE_MASK);
480 unsigned offset = pgd - pgd_page;
481 pgd_t *user_ptr = NULL;
483 if (offset < pgd_index(USER_LIMIT)) {
484 struct page *page = virt_to_page(pgd_page);
485 user_ptr = (pgd_t *)page->private;
493 static void __xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
497 u.ptr = virt_to_machine(ptr).maddr;
498 u.val = p4d_val_ma(val);
499 xen_extend_mmu_update(&u);
503 * Raw hypercall-based set_p4d, intended for in early boot before
504 * there's a page structure. This implies:
505 * 1. The only existing pagetable is the kernel's
506 * 2. It is always pinned
507 * 3. It has no user pagetable attached to it
509 static void __init xen_set_p4d_hyper(p4d_t *ptr, p4d_t val)
515 __xen_set_p4d_hyper(ptr, val);
517 xen_mc_issue(PARAVIRT_LAZY_MMU);
522 static void xen_set_p4d(p4d_t *ptr, p4d_t val)
524 pgd_t *user_ptr = xen_get_user_pgd((pgd_t *)ptr);
527 trace_xen_mmu_set_p4d(ptr, (p4d_t *)user_ptr, val);
529 /* If page is not pinned, we can just update the entry
531 if (!xen_page_pinned(ptr)) {
534 WARN_ON(xen_page_pinned(user_ptr));
535 pgd_val.pgd = p4d_val_ma(val);
541 /* If it's pinned, then we can at least batch the kernel and
542 user updates together. */
545 __xen_set_p4d_hyper(ptr, val);
547 __xen_set_p4d_hyper((p4d_t *)user_ptr, val);
549 xen_mc_issue(PARAVIRT_LAZY_MMU);
552 #if CONFIG_PGTABLE_LEVELS >= 5
553 __visible p4dval_t xen_p4d_val(p4d_t p4d)
555 return pte_mfn_to_pfn(p4d.p4d);
557 PV_CALLEE_SAVE_REGS_THUNK(xen_p4d_val);
559 __visible p4d_t xen_make_p4d(p4dval_t p4d)
561 p4d = pte_pfn_to_mfn(p4d);
563 return native_make_p4d(p4d);
565 PV_CALLEE_SAVE_REGS_THUNK(xen_make_p4d);
566 #endif /* CONFIG_PGTABLE_LEVELS >= 5 */
567 #endif /* CONFIG_X86_64 */
569 static int xen_pmd_walk(struct mm_struct *mm, pmd_t *pmd,
570 int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
571 bool last, unsigned long limit)
573 int i, nr, flush = 0;
575 nr = last ? pmd_index(limit) + 1 : PTRS_PER_PMD;
576 for (i = 0; i < nr; i++) {
577 if (!pmd_none(pmd[i]))
578 flush |= (*func)(mm, pmd_page(pmd[i]), PT_PTE);
583 static int xen_pud_walk(struct mm_struct *mm, pud_t *pud,
584 int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
585 bool last, unsigned long limit)
587 int i, nr, flush = 0;
589 nr = last ? pud_index(limit) + 1 : PTRS_PER_PUD;
590 for (i = 0; i < nr; i++) {
593 if (pud_none(pud[i]))
596 pmd = pmd_offset(&pud[i], 0);
597 if (PTRS_PER_PMD > 1)
598 flush |= (*func)(mm, virt_to_page(pmd), PT_PMD);
599 flush |= xen_pmd_walk(mm, pmd, func,
600 last && i == nr - 1, limit);
605 static int xen_p4d_walk(struct mm_struct *mm, p4d_t *p4d,
606 int (*func)(struct mm_struct *mm, struct page *, enum pt_level),
607 bool last, unsigned long limit)
616 pud = pud_offset(p4d, 0);
617 if (PTRS_PER_PUD > 1)
618 flush |= (*func)(mm, virt_to_page(pud), PT_PUD);
619 flush |= xen_pud_walk(mm, pud, func, last, limit);
624 * (Yet another) pagetable walker. This one is intended for pinning a
625 * pagetable. This means that it walks a pagetable and calls the
626 * callback function on each page it finds making up the page table,
627 * at every level. It walks the entire pagetable, but it only bothers
628 * pinning pte pages which are below limit. In the normal case this
629 * will be STACK_TOP_MAX, but at boot we need to pin up to
632 * For 32-bit the important bit is that we don't pin beyond there,
633 * because then we start getting into Xen's ptes.
635 * For 64-bit, we must skip the Xen hole in the middle of the address
636 * space, just after the big x86-64 virtual hole.
638 static int __xen_pgd_walk(struct mm_struct *mm, pgd_t *pgd,
639 int (*func)(struct mm_struct *mm, struct page *,
643 int i, nr, flush = 0;
644 unsigned hole_low, hole_high;
646 /* The limit is the last byte to be touched */
648 BUG_ON(limit >= FIXADDR_TOP);
651 * 64-bit has a great big hole in the middle of the address
652 * space, which contains the Xen mappings. On 32-bit these
653 * will end up making a zero-sized hole and so is a no-op.
655 hole_low = pgd_index(USER_LIMIT);
656 hole_high = pgd_index(PAGE_OFFSET);
658 nr = pgd_index(limit) + 1;
659 for (i = 0; i < nr; i++) {
662 if (i >= hole_low && i < hole_high)
665 if (pgd_none(pgd[i]))
668 p4d = p4d_offset(&pgd[i], 0);
669 flush |= xen_p4d_walk(mm, p4d, func, i == nr - 1, limit);
672 /* Do the top level last, so that the callbacks can use it as
673 a cue to do final things like tlb flushes. */
674 flush |= (*func)(mm, virt_to_page(pgd), PT_PGD);
679 static int xen_pgd_walk(struct mm_struct *mm,
680 int (*func)(struct mm_struct *mm, struct page *,
684 return __xen_pgd_walk(mm, mm->pgd, func, limit);
687 /* If we're using split pte locks, then take the page's lock and
688 return a pointer to it. Otherwise return NULL. */
689 static spinlock_t *xen_pte_lock(struct page *page, struct mm_struct *mm)
691 spinlock_t *ptl = NULL;
693 #if USE_SPLIT_PTE_PTLOCKS
694 ptl = ptlock_ptr(page);
695 spin_lock_nest_lock(ptl, &mm->page_table_lock);
701 static void xen_pte_unlock(void *v)
707 static void xen_do_pin(unsigned level, unsigned long pfn)
712 op.arg1.mfn = pfn_to_mfn(pfn);
714 xen_extend_mmuext_op(&op);
717 static int xen_pin_page(struct mm_struct *mm, struct page *page,
720 unsigned pgfl = TestSetPagePinned(page);
724 flush = 0; /* already pinned */
725 else if (PageHighMem(page))
726 /* kmaps need flushing if we found an unpinned
730 void *pt = lowmem_page_address(page);
731 unsigned long pfn = page_to_pfn(page);
732 struct multicall_space mcs = __xen_mc_entry(0);
738 * We need to hold the pagetable lock between the time
739 * we make the pagetable RO and when we actually pin
740 * it. If we don't, then other users may come in and
741 * attempt to update the pagetable by writing it,
742 * which will fail because the memory is RO but not
743 * pinned, so Xen won't do the trap'n'emulate.
745 * If we're using split pte locks, we can't hold the
746 * entire pagetable's worth of locks during the
747 * traverse, because we may wrap the preempt count (8
748 * bits). The solution is to mark RO and pin each PTE
749 * page while holding the lock. This means the number
750 * of locks we end up holding is never more than a
751 * batch size (~32 entries, at present).
753 * If we're not using split pte locks, we needn't pin
754 * the PTE pages independently, because we're
755 * protected by the overall pagetable lock.
759 ptl = xen_pte_lock(page, mm);
761 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
762 pfn_pte(pfn, PAGE_KERNEL_RO),
763 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
766 xen_do_pin(MMUEXT_PIN_L1_TABLE, pfn);
768 /* Queue a deferred unlock for when this batch
770 xen_mc_callback(xen_pte_unlock, ptl);
777 /* This is called just after a mm has been created, but it has not
778 been used yet. We need to make sure that its pagetable is all
779 read-only, and can be pinned. */
780 static void __xen_pgd_pin(struct mm_struct *mm, pgd_t *pgd)
782 trace_xen_mmu_pgd_pin(mm, pgd);
786 if (__xen_pgd_walk(mm, pgd, xen_pin_page, USER_LIMIT)) {
787 /* re-enable interrupts for flushing */
797 pgd_t *user_pgd = xen_get_user_pgd(pgd);
799 xen_do_pin(MMUEXT_PIN_L4_TABLE, PFN_DOWN(__pa(pgd)));
802 xen_pin_page(mm, virt_to_page(user_pgd), PT_PGD);
803 xen_do_pin(MMUEXT_PIN_L4_TABLE,
804 PFN_DOWN(__pa(user_pgd)));
807 #else /* CONFIG_X86_32 */
808 #ifdef CONFIG_X86_PAE
809 /* Need to make sure unshared kernel PMD is pinnable */
810 xen_pin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
813 xen_do_pin(MMUEXT_PIN_L3_TABLE, PFN_DOWN(__pa(pgd)));
814 #endif /* CONFIG_X86_64 */
818 static void xen_pgd_pin(struct mm_struct *mm)
820 __xen_pgd_pin(mm, mm->pgd);
824 * On save, we need to pin all pagetables to make sure they get their
825 * mfns turned into pfns. Search the list for any unpinned pgds and pin
826 * them (unpinned pgds are not currently in use, probably because the
827 * process is under construction or destruction).
829 * Expected to be called in stop_machine() ("equivalent to taking
830 * every spinlock in the system"), so the locking doesn't really
831 * matter all that much.
833 void xen_mm_pin_all(void)
837 spin_lock(&pgd_lock);
839 list_for_each_entry(page, &pgd_list, lru) {
840 if (!PagePinned(page)) {
841 __xen_pgd_pin(&init_mm, (pgd_t *)page_address(page));
842 SetPageSavePinned(page);
846 spin_unlock(&pgd_lock);
849 static int __init xen_mark_pinned(struct mm_struct *mm, struct page *page,
857 * The init_mm pagetable is really pinned as soon as its created, but
858 * that's before we have page structures to store the bits. So do all
859 * the book-keeping now once struct pages for allocated pages are
860 * initialized. This happens only after free_all_bootmem() is called.
862 static void __init xen_after_bootmem(void)
864 static_branch_enable(&xen_struct_pages_ready);
866 SetPagePinned(virt_to_page(level3_user_vsyscall));
868 xen_pgd_walk(&init_mm, xen_mark_pinned, FIXADDR_TOP);
871 static int xen_unpin_page(struct mm_struct *mm, struct page *page,
874 unsigned pgfl = TestClearPagePinned(page);
876 if (pgfl && !PageHighMem(page)) {
877 void *pt = lowmem_page_address(page);
878 unsigned long pfn = page_to_pfn(page);
879 spinlock_t *ptl = NULL;
880 struct multicall_space mcs;
883 * Do the converse to pin_page. If we're using split
884 * pte locks, we must be holding the lock for while
885 * the pte page is unpinned but still RO to prevent
886 * concurrent updates from seeing it in this
887 * partially-pinned state.
889 if (level == PT_PTE) {
890 ptl = xen_pte_lock(page, mm);
893 xen_do_pin(MMUEXT_UNPIN_TABLE, pfn);
896 mcs = __xen_mc_entry(0);
898 MULTI_update_va_mapping(mcs.mc, (unsigned long)pt,
899 pfn_pte(pfn, PAGE_KERNEL),
900 level == PT_PGD ? UVMF_TLB_FLUSH : 0);
903 /* unlock when batch completed */
904 xen_mc_callback(xen_pte_unlock, ptl);
908 return 0; /* never need to flush on unpin */
911 /* Release a pagetables pages back as normal RW */
912 static void __xen_pgd_unpin(struct mm_struct *mm, pgd_t *pgd)
914 trace_xen_mmu_pgd_unpin(mm, pgd);
918 xen_do_pin(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
922 pgd_t *user_pgd = xen_get_user_pgd(pgd);
925 xen_do_pin(MMUEXT_UNPIN_TABLE,
926 PFN_DOWN(__pa(user_pgd)));
927 xen_unpin_page(mm, virt_to_page(user_pgd), PT_PGD);
932 #ifdef CONFIG_X86_PAE
933 /* Need to make sure unshared kernel PMD is unpinned */
934 xen_unpin_page(mm, pgd_page(pgd[pgd_index(TASK_SIZE)]),
938 __xen_pgd_walk(mm, pgd, xen_unpin_page, USER_LIMIT);
943 static void xen_pgd_unpin(struct mm_struct *mm)
945 __xen_pgd_unpin(mm, mm->pgd);
949 * On resume, undo any pinning done at save, so that the rest of the
950 * kernel doesn't see any unexpected pinned pagetables.
952 void xen_mm_unpin_all(void)
956 spin_lock(&pgd_lock);
958 list_for_each_entry(page, &pgd_list, lru) {
959 if (PageSavePinned(page)) {
960 BUG_ON(!PagePinned(page));
961 __xen_pgd_unpin(&init_mm, (pgd_t *)page_address(page));
962 ClearPageSavePinned(page);
966 spin_unlock(&pgd_lock);
969 static void xen_activate_mm(struct mm_struct *prev, struct mm_struct *next)
971 spin_lock(&next->page_table_lock);
973 spin_unlock(&next->page_table_lock);
976 static void xen_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
978 spin_lock(&mm->page_table_lock);
980 spin_unlock(&mm->page_table_lock);
983 static void drop_mm_ref_this_cpu(void *info)
985 struct mm_struct *mm = info;
987 if (this_cpu_read(cpu_tlbstate.loaded_mm) == mm)
988 leave_mm(smp_processor_id());
991 * If this cpu still has a stale cr3 reference, then make sure
992 * it has been flushed.
994 if (this_cpu_read(xen_current_cr3) == __pa(mm->pgd))
1000 * Another cpu may still have their %cr3 pointing at the pagetable, so
1001 * we need to repoint it somewhere else before we can unpin it.
1003 static void xen_drop_mm_ref(struct mm_struct *mm)
1008 drop_mm_ref_this_cpu(mm);
1010 /* Get the "official" set of cpus referring to our pagetable. */
1011 if (!alloc_cpumask_var(&mask, GFP_ATOMIC)) {
1012 for_each_online_cpu(cpu) {
1013 if (per_cpu(xen_current_cr3, cpu) != __pa(mm->pgd))
1015 smp_call_function_single(cpu, drop_mm_ref_this_cpu, mm, 1);
1021 * It's possible that a vcpu may have a stale reference to our
1022 * cr3, because its in lazy mode, and it hasn't yet flushed
1023 * its set of pending hypercalls yet. In this case, we can
1024 * look at its actual current cr3 value, and force it to flush
1027 cpumask_clear(mask);
1028 for_each_online_cpu(cpu) {
1029 if (per_cpu(xen_current_cr3, cpu) == __pa(mm->pgd))
1030 cpumask_set_cpu(cpu, mask);
1033 smp_call_function_many(mask, drop_mm_ref_this_cpu, mm, 1);
1034 free_cpumask_var(mask);
1037 static void xen_drop_mm_ref(struct mm_struct *mm)
1039 drop_mm_ref_this_cpu(mm);
1044 * While a process runs, Xen pins its pagetables, which means that the
1045 * hypervisor forces it to be read-only, and it controls all updates
1046 * to it. This means that all pagetable updates have to go via the
1047 * hypervisor, which is moderately expensive.
1049 * Since we're pulling the pagetable down, we switch to use init_mm,
1050 * unpin old process pagetable and mark it all read-write, which
1051 * allows further operations on it to be simple memory accesses.
1053 * The only subtle point is that another CPU may be still using the
1054 * pagetable because of lazy tlb flushing. This means we need need to
1055 * switch all CPUs off this pagetable before we can unpin it.
1057 static void xen_exit_mmap(struct mm_struct *mm)
1059 get_cpu(); /* make sure we don't move around */
1060 xen_drop_mm_ref(mm);
1063 spin_lock(&mm->page_table_lock);
1065 /* pgd may not be pinned in the error exit path of execve */
1066 if (xen_page_pinned(mm->pgd))
1069 spin_unlock(&mm->page_table_lock);
1072 static void xen_post_allocator_init(void);
1074 static void __init pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1076 struct mmuext_op op;
1079 op.arg1.mfn = pfn_to_mfn(pfn);
1080 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
1084 #ifdef CONFIG_X86_64
1085 static void __init xen_cleanhighmap(unsigned long vaddr,
1086 unsigned long vaddr_end)
1088 unsigned long kernel_end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
1089 pmd_t *pmd = level2_kernel_pgt + pmd_index(vaddr);
1091 /* NOTE: The loop is more greedy than the cleanup_highmap variant.
1092 * We include the PMD passed in on _both_ boundaries. */
1093 for (; vaddr <= vaddr_end && (pmd < (level2_kernel_pgt + PTRS_PER_PMD));
1094 pmd++, vaddr += PMD_SIZE) {
1097 if (vaddr < (unsigned long) _text || vaddr > kernel_end)
1098 set_pmd(pmd, __pmd(0));
1100 /* In case we did something silly, we should crash in this function
1101 * instead of somewhere later and be confusing. */
1106 * Make a page range writeable and free it.
1108 static void __init xen_free_ro_pages(unsigned long paddr, unsigned long size)
1110 void *vaddr = __va(paddr);
1111 void *vaddr_end = vaddr + size;
1113 for (; vaddr < vaddr_end; vaddr += PAGE_SIZE)
1114 make_lowmem_page_readwrite(vaddr);
1116 memblock_free(paddr, size);
1119 static void __init xen_cleanmfnmap_free_pgtbl(void *pgtbl, bool unpin)
1121 unsigned long pa = __pa(pgtbl) & PHYSICAL_PAGE_MASK;
1124 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(pa));
1125 ClearPagePinned(virt_to_page(__va(pa)));
1126 xen_free_ro_pages(pa, PAGE_SIZE);
1129 static void __init xen_cleanmfnmap_pmd(pmd_t *pmd, bool unpin)
1135 if (pmd_large(*pmd)) {
1136 pa = pmd_val(*pmd) & PHYSICAL_PAGE_MASK;
1137 xen_free_ro_pages(pa, PMD_SIZE);
1141 pte_tbl = pte_offset_kernel(pmd, 0);
1142 for (i = 0; i < PTRS_PER_PTE; i++) {
1143 if (pte_none(pte_tbl[i]))
1145 pa = pte_pfn(pte_tbl[i]) << PAGE_SHIFT;
1146 xen_free_ro_pages(pa, PAGE_SIZE);
1148 set_pmd(pmd, __pmd(0));
1149 xen_cleanmfnmap_free_pgtbl(pte_tbl, unpin);
1152 static void __init xen_cleanmfnmap_pud(pud_t *pud, bool unpin)
1158 if (pud_large(*pud)) {
1159 pa = pud_val(*pud) & PHYSICAL_PAGE_MASK;
1160 xen_free_ro_pages(pa, PUD_SIZE);
1164 pmd_tbl = pmd_offset(pud, 0);
1165 for (i = 0; i < PTRS_PER_PMD; i++) {
1166 if (pmd_none(pmd_tbl[i]))
1168 xen_cleanmfnmap_pmd(pmd_tbl + i, unpin);
1170 set_pud(pud, __pud(0));
1171 xen_cleanmfnmap_free_pgtbl(pmd_tbl, unpin);
1174 static void __init xen_cleanmfnmap_p4d(p4d_t *p4d, bool unpin)
1180 if (p4d_large(*p4d)) {
1181 pa = p4d_val(*p4d) & PHYSICAL_PAGE_MASK;
1182 xen_free_ro_pages(pa, P4D_SIZE);
1186 pud_tbl = pud_offset(p4d, 0);
1187 for (i = 0; i < PTRS_PER_PUD; i++) {
1188 if (pud_none(pud_tbl[i]))
1190 xen_cleanmfnmap_pud(pud_tbl + i, unpin);
1192 set_p4d(p4d, __p4d(0));
1193 xen_cleanmfnmap_free_pgtbl(pud_tbl, unpin);
1197 * Since it is well isolated we can (and since it is perhaps large we should)
1198 * also free the page tables mapping the initial P->M table.
1200 static void __init xen_cleanmfnmap(unsigned long vaddr)
1206 unpin = (vaddr == 2 * PGDIR_SIZE);
1208 pgd = pgd_offset_k(vaddr);
1209 p4d = p4d_offset(pgd, 0);
1210 if (!p4d_none(*p4d))
1211 xen_cleanmfnmap_p4d(p4d, unpin);
1214 static void __init xen_pagetable_p2m_free(void)
1219 size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
1221 /* No memory or already called. */
1222 if ((unsigned long)xen_p2m_addr == xen_start_info->mfn_list)
1225 /* using __ka address and sticking INVALID_P2M_ENTRY! */
1226 memset((void *)xen_start_info->mfn_list, 0xff, size);
1228 addr = xen_start_info->mfn_list;
1230 * We could be in __ka space.
1231 * We roundup to the PMD, which means that if anybody at this stage is
1232 * using the __ka address of xen_start_info or
1233 * xen_start_info->shared_info they are in going to crash. Fortunatly
1234 * we have already revectored in xen_setup_kernel_pagetable.
1236 size = roundup(size, PMD_SIZE);
1238 if (addr >= __START_KERNEL_map) {
1239 xen_cleanhighmap(addr, addr + size);
1240 size = PAGE_ALIGN(xen_start_info->nr_pages *
1241 sizeof(unsigned long));
1242 memblock_free(__pa(addr), size);
1244 xen_cleanmfnmap(addr);
1248 static void __init xen_pagetable_cleanhighmap(void)
1253 /* At this stage, cleanup_highmap has already cleaned __ka space
1254 * from _brk_limit way up to the max_pfn_mapped (which is the end of
1255 * the ramdisk). We continue on, erasing PMD entries that point to page
1256 * tables - do note that they are accessible at this stage via __va.
1257 * As Xen is aligning the memory end to a 4MB boundary, for good
1258 * measure we also round up to PMD_SIZE * 2 - which means that if
1259 * anybody is using __ka address to the initial boot-stack - and try
1260 * to use it - they are going to crash. The xen_start_info has been
1261 * taken care of already in xen_setup_kernel_pagetable. */
1262 addr = xen_start_info->pt_base;
1263 size = xen_start_info->nr_pt_frames * PAGE_SIZE;
1265 xen_cleanhighmap(addr, roundup(addr + size, PMD_SIZE * 2));
1266 xen_start_info->pt_base = (unsigned long)__va(__pa(xen_start_info->pt_base));
1270 static void __init xen_pagetable_p2m_setup(void)
1272 xen_vmalloc_p2m_tree();
1274 #ifdef CONFIG_X86_64
1275 xen_pagetable_p2m_free();
1277 xen_pagetable_cleanhighmap();
1279 /* And revector! Bye bye old array */
1280 xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
1283 static void __init xen_pagetable_init(void)
1286 xen_post_allocator_init();
1288 xen_pagetable_p2m_setup();
1290 /* Allocate and initialize top and mid mfn levels for p2m structure */
1291 xen_build_mfn_list_list();
1293 /* Remap memory freed due to conflicts with E820 map */
1295 xen_setup_mfn_list_list();
1297 static void xen_write_cr2(unsigned long cr2)
1299 this_cpu_read(xen_vcpu)->arch.cr2 = cr2;
1302 static unsigned long xen_read_cr2(void)
1304 return this_cpu_read(xen_vcpu)->arch.cr2;
1307 unsigned long xen_read_cr2_direct(void)
1309 return this_cpu_read(xen_vcpu_info.arch.cr2);
1312 static noinline void xen_flush_tlb(void)
1314 struct mmuext_op *op;
1315 struct multicall_space mcs;
1319 mcs = xen_mc_entry(sizeof(*op));
1322 op->cmd = MMUEXT_TLB_FLUSH_LOCAL;
1323 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1325 xen_mc_issue(PARAVIRT_LAZY_MMU);
1330 static void xen_flush_tlb_one_user(unsigned long addr)
1332 struct mmuext_op *op;
1333 struct multicall_space mcs;
1335 trace_xen_mmu_flush_tlb_one_user(addr);
1339 mcs = xen_mc_entry(sizeof(*op));
1341 op->cmd = MMUEXT_INVLPG_LOCAL;
1342 op->arg1.linear_addr = addr & PAGE_MASK;
1343 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
1345 xen_mc_issue(PARAVIRT_LAZY_MMU);
1350 static void xen_flush_tlb_others(const struct cpumask *cpus,
1351 const struct flush_tlb_info *info)
1354 struct mmuext_op op;
1355 DECLARE_BITMAP(mask, NR_CPUS);
1357 struct multicall_space mcs;
1358 const size_t mc_entry_size = sizeof(args->op) +
1359 sizeof(args->mask[0]) * BITS_TO_LONGS(num_possible_cpus());
1361 trace_xen_mmu_flush_tlb_others(cpus, info->mm, info->start, info->end);
1363 if (cpumask_empty(cpus))
1364 return; /* nothing to do */
1366 mcs = xen_mc_entry(mc_entry_size);
1368 args->op.arg2.vcpumask = to_cpumask(args->mask);
1370 /* Remove us, and any offline CPUS. */
1371 cpumask_and(to_cpumask(args->mask), cpus, cpu_online_mask);
1372 cpumask_clear_cpu(smp_processor_id(), to_cpumask(args->mask));
1374 args->op.cmd = MMUEXT_TLB_FLUSH_MULTI;
1375 if (info->end != TLB_FLUSH_ALL &&
1376 (info->end - info->start) <= PAGE_SIZE) {
1377 args->op.cmd = MMUEXT_INVLPG_MULTI;
1378 args->op.arg1.linear_addr = info->start;
1381 MULTI_mmuext_op(mcs.mc, &args->op, 1, NULL, DOMID_SELF);
1383 xen_mc_issue(PARAVIRT_LAZY_MMU);
1386 static unsigned long xen_read_cr3(void)
1388 return this_cpu_read(xen_cr3);
1391 static void set_current_cr3(void *v)
1393 this_cpu_write(xen_current_cr3, (unsigned long)v);
1396 static void __xen_write_cr3(bool kernel, unsigned long cr3)
1398 struct mmuext_op op;
1401 trace_xen_mmu_write_cr3(kernel, cr3);
1404 mfn = pfn_to_mfn(PFN_DOWN(cr3));
1408 WARN_ON(mfn == 0 && kernel);
1410 op.cmd = kernel ? MMUEXT_NEW_BASEPTR : MMUEXT_NEW_USER_BASEPTR;
1413 xen_extend_mmuext_op(&op);
1416 this_cpu_write(xen_cr3, cr3);
1418 /* Update xen_current_cr3 once the batch has actually
1420 xen_mc_callback(set_current_cr3, (void *)cr3);
1423 static void xen_write_cr3(unsigned long cr3)
1425 BUG_ON(preemptible());
1427 xen_mc_batch(); /* disables interrupts */
1429 /* Update while interrupts are disabled, so its atomic with
1431 this_cpu_write(xen_cr3, cr3);
1433 __xen_write_cr3(true, cr3);
1435 #ifdef CONFIG_X86_64
1437 pgd_t *user_pgd = xen_get_user_pgd(__va(cr3));
1439 __xen_write_cr3(false, __pa(user_pgd));
1441 __xen_write_cr3(false, 0);
1445 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1448 #ifdef CONFIG_X86_64
1450 * At the start of the day - when Xen launches a guest, it has already
1451 * built pagetables for the guest. We diligently look over them
1452 * in xen_setup_kernel_pagetable and graft as appropriate them in the
1453 * init_top_pgt and its friends. Then when we are happy we load
1454 * the new init_top_pgt - and continue on.
1456 * The generic code starts (start_kernel) and 'init_mem_mapping' sets
1457 * up the rest of the pagetables. When it has completed it loads the cr3.
1458 * N.B. that baremetal would start at 'start_kernel' (and the early
1459 * #PF handler would create bootstrap pagetables) - so we are running
1460 * with the same assumptions as what to do when write_cr3 is executed
1463 * Since there are no user-page tables at all, we have two variants
1464 * of xen_write_cr3 - the early bootup (this one), and the late one
1465 * (xen_write_cr3). The reason we have to do that is that in 64-bit
1466 * the Linux kernel and user-space are both in ring 3 while the
1467 * hypervisor is in ring 0.
1469 static void __init xen_write_cr3_init(unsigned long cr3)
1471 BUG_ON(preemptible());
1473 xen_mc_batch(); /* disables interrupts */
1475 /* Update while interrupts are disabled, so its atomic with
1477 this_cpu_write(xen_cr3, cr3);
1479 __xen_write_cr3(true, cr3);
1481 xen_mc_issue(PARAVIRT_LAZY_CPU); /* interrupts restored */
1485 static int xen_pgd_alloc(struct mm_struct *mm)
1487 pgd_t *pgd = mm->pgd;
1490 BUG_ON(PagePinned(virt_to_page(pgd)));
1492 #ifdef CONFIG_X86_64
1494 struct page *page = virt_to_page(pgd);
1497 BUG_ON(page->private != 0);
1501 user_pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
1502 page->private = (unsigned long)user_pgd;
1504 if (user_pgd != NULL) {
1505 #ifdef CONFIG_X86_VSYSCALL_EMULATION
1506 user_pgd[pgd_index(VSYSCALL_ADDR)] =
1507 __pgd(__pa(level3_user_vsyscall) | _PAGE_TABLE);
1512 BUG_ON(PagePinned(virt_to_page(xen_get_user_pgd(pgd))));
1518 static void xen_pgd_free(struct mm_struct *mm, pgd_t *pgd)
1520 #ifdef CONFIG_X86_64
1521 pgd_t *user_pgd = xen_get_user_pgd(pgd);
1524 free_page((unsigned long)user_pgd);
1529 * Init-time set_pte while constructing initial pagetables, which
1530 * doesn't allow RO page table pages to be remapped RW.
1532 * If there is no MFN for this PFN then this page is initially
1533 * ballooned out so clear the PTE (as in decrease_reservation() in
1534 * drivers/xen/balloon.c).
1536 * Many of these PTE updates are done on unpinned and writable pages
1537 * and doing a hypercall for these is unnecessary and expensive. At
1538 * this point it is not possible to tell if a page is pinned or not,
1539 * so always write the PTE directly and rely on Xen trapping and
1540 * emulating any updates as necessary.
1542 __visible pte_t xen_make_pte_init(pteval_t pte)
1544 #ifdef CONFIG_X86_64
1548 * Pages belonging to the initial p2m list mapped outside the default
1549 * address range must be mapped read-only. This region contains the
1550 * page tables for mapping the p2m list, too, and page tables MUST be
1553 pfn = (pte & PTE_PFN_MASK) >> PAGE_SHIFT;
1554 if (xen_start_info->mfn_list < __START_KERNEL_map &&
1555 pfn >= xen_start_info->first_p2m_pfn &&
1556 pfn < xen_start_info->first_p2m_pfn + xen_start_info->nr_p2m_frames)
1559 pte = pte_pfn_to_mfn(pte);
1560 return native_make_pte(pte);
1562 PV_CALLEE_SAVE_REGS_THUNK(xen_make_pte_init);
1564 static void __init xen_set_pte_init(pte_t *ptep, pte_t pte)
1566 #ifdef CONFIG_X86_32
1567 /* If there's an existing pte, then don't allow _PAGE_RW to be set */
1568 if (pte_mfn(pte) != INVALID_P2M_ENTRY
1569 && pte_val_ma(*ptep) & _PAGE_PRESENT)
1570 pte = __pte_ma(((pte_val_ma(*ptep) & _PAGE_RW) | ~_PAGE_RW) &
1573 native_set_pte(ptep, pte);
1576 /* Early in boot, while setting up the initial pagetable, assume
1577 everything is pinned. */
1578 static void __init xen_alloc_pte_init(struct mm_struct *mm, unsigned long pfn)
1580 #ifdef CONFIG_FLATMEM
1581 BUG_ON(mem_map); /* should only be used early */
1583 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1584 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1587 /* Used for pmd and pud */
1588 static void __init xen_alloc_pmd_init(struct mm_struct *mm, unsigned long pfn)
1590 #ifdef CONFIG_FLATMEM
1591 BUG_ON(mem_map); /* should only be used early */
1593 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
1596 /* Early release_pte assumes that all pts are pinned, since there's
1597 only init_mm and anything attached to that is pinned. */
1598 static void __init xen_release_pte_init(unsigned long pfn)
1600 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1601 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1604 static void __init xen_release_pmd_init(unsigned long pfn)
1606 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
1609 static inline void __pin_pagetable_pfn(unsigned cmd, unsigned long pfn)
1611 struct multicall_space mcs;
1612 struct mmuext_op *op;
1614 mcs = __xen_mc_entry(sizeof(*op));
1617 op->arg1.mfn = pfn_to_mfn(pfn);
1619 MULTI_mmuext_op(mcs.mc, mcs.args, 1, NULL, DOMID_SELF);
1622 static inline void __set_pfn_prot(unsigned long pfn, pgprot_t prot)
1624 struct multicall_space mcs;
1625 unsigned long addr = (unsigned long)__va(pfn << PAGE_SHIFT);
1627 mcs = __xen_mc_entry(0);
1628 MULTI_update_va_mapping(mcs.mc, (unsigned long)addr,
1629 pfn_pte(pfn, prot), 0);
1632 /* This needs to make sure the new pte page is pinned iff its being
1633 attached to a pinned pagetable. */
1634 static inline void xen_alloc_ptpage(struct mm_struct *mm, unsigned long pfn,
1637 bool pinned = xen_page_pinned(mm->pgd);
1639 trace_xen_mmu_alloc_ptpage(mm, pfn, level, pinned);
1642 struct page *page = pfn_to_page(pfn);
1644 if (static_branch_likely(&xen_struct_pages_ready))
1645 SetPagePinned(page);
1647 if (!PageHighMem(page)) {
1650 __set_pfn_prot(pfn, PAGE_KERNEL_RO);
1652 if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1653 __pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE, pfn);
1655 xen_mc_issue(PARAVIRT_LAZY_MMU);
1657 /* make sure there are no stray mappings of
1659 kmap_flush_unused();
1664 static void xen_alloc_pte(struct mm_struct *mm, unsigned long pfn)
1666 xen_alloc_ptpage(mm, pfn, PT_PTE);
1669 static void xen_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
1671 xen_alloc_ptpage(mm, pfn, PT_PMD);
1674 /* This should never happen until we're OK to use struct page */
1675 static inline void xen_release_ptpage(unsigned long pfn, unsigned level)
1677 struct page *page = pfn_to_page(pfn);
1678 bool pinned = PagePinned(page);
1680 trace_xen_mmu_release_ptpage(pfn, level, pinned);
1683 if (!PageHighMem(page)) {
1686 if (level == PT_PTE && USE_SPLIT_PTE_PTLOCKS)
1687 __pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, pfn);
1689 __set_pfn_prot(pfn, PAGE_KERNEL);
1691 xen_mc_issue(PARAVIRT_LAZY_MMU);
1693 ClearPagePinned(page);
1697 static void xen_release_pte(unsigned long pfn)
1699 xen_release_ptpage(pfn, PT_PTE);
1702 static void xen_release_pmd(unsigned long pfn)
1704 xen_release_ptpage(pfn, PT_PMD);
1707 #ifdef CONFIG_X86_64
1708 static void xen_alloc_pud(struct mm_struct *mm, unsigned long pfn)
1710 xen_alloc_ptpage(mm, pfn, PT_PUD);
1713 static void xen_release_pud(unsigned long pfn)
1715 xen_release_ptpage(pfn, PT_PUD);
1719 void __init xen_reserve_top(void)
1721 #ifdef CONFIG_X86_32
1722 unsigned long top = HYPERVISOR_VIRT_START;
1723 struct xen_platform_parameters pp;
1725 if (HYPERVISOR_xen_version(XENVER_platform_parameters, &pp) == 0)
1726 top = pp.virt_start;
1728 reserve_top_address(-top);
1729 #endif /* CONFIG_X86_32 */
1733 * Like __va(), but returns address in the kernel mapping (which is
1734 * all we have until the physical memory mapping has been set up.
1736 static void * __init __ka(phys_addr_t paddr)
1738 #ifdef CONFIG_X86_64
1739 return (void *)(paddr + __START_KERNEL_map);
1745 /* Convert a machine address to physical address */
1746 static unsigned long __init m2p(phys_addr_t maddr)
1750 maddr &= XEN_PTE_MFN_MASK;
1751 paddr = mfn_to_pfn(maddr >> PAGE_SHIFT) << PAGE_SHIFT;
1756 /* Convert a machine address to kernel virtual */
1757 static void * __init m2v(phys_addr_t maddr)
1759 return __ka(m2p(maddr));
1762 /* Set the page permissions on an identity-mapped pages */
1763 static void __init set_page_prot_flags(void *addr, pgprot_t prot,
1764 unsigned long flags)
1766 unsigned long pfn = __pa(addr) >> PAGE_SHIFT;
1767 pte_t pte = pfn_pte(pfn, prot);
1769 if (HYPERVISOR_update_va_mapping((unsigned long)addr, pte, flags))
1772 static void __init set_page_prot(void *addr, pgprot_t prot)
1774 return set_page_prot_flags(addr, prot, UVMF_NONE);
1776 #ifdef CONFIG_X86_32
1777 static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
1779 unsigned pmdidx, pteidx;
1783 level1_ident_pgt = extend_brk(sizeof(pte_t) * LEVEL1_IDENT_ENTRIES,
1788 for (pmdidx = 0; pmdidx < PTRS_PER_PMD && pfn < max_pfn; pmdidx++) {
1791 /* Reuse or allocate a page of ptes */
1792 if (pmd_present(pmd[pmdidx]))
1793 pte_page = m2v(pmd[pmdidx].pmd);
1795 /* Check for free pte pages */
1796 if (ident_pte == LEVEL1_IDENT_ENTRIES)
1799 pte_page = &level1_ident_pgt[ident_pte];
1800 ident_pte += PTRS_PER_PTE;
1802 pmd[pmdidx] = __pmd(__pa(pte_page) | _PAGE_TABLE);
1805 /* Install mappings */
1806 for (pteidx = 0; pteidx < PTRS_PER_PTE; pteidx++, pfn++) {
1809 if (pfn > max_pfn_mapped)
1810 max_pfn_mapped = pfn;
1812 if (!pte_none(pte_page[pteidx]))
1815 pte = pfn_pte(pfn, PAGE_KERNEL_EXEC);
1816 pte_page[pteidx] = pte;
1820 for (pteidx = 0; pteidx < ident_pte; pteidx += PTRS_PER_PTE)
1821 set_page_prot(&level1_ident_pgt[pteidx], PAGE_KERNEL_RO);
1823 set_page_prot(pmd, PAGE_KERNEL_RO);
1826 void __init xen_setup_machphys_mapping(void)
1828 struct xen_machphys_mapping mapping;
1830 if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
1831 machine_to_phys_mapping = (unsigned long *)mapping.v_start;
1832 machine_to_phys_nr = mapping.max_mfn + 1;
1834 machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
1836 #ifdef CONFIG_X86_32
1837 WARN_ON((machine_to_phys_mapping + (machine_to_phys_nr - 1))
1838 < machine_to_phys_mapping);
1842 #ifdef CONFIG_X86_64
1843 static void __init convert_pfn_mfn(void *v)
1848 /* All levels are converted the same way, so just treat them
1850 for (i = 0; i < PTRS_PER_PTE; i++)
1851 pte[i] = xen_make_pte(pte[i].pte);
1853 static void __init check_pt_base(unsigned long *pt_base, unsigned long *pt_end,
1856 if (*pt_base == PFN_DOWN(__pa(addr))) {
1857 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1858 clear_page((void *)addr);
1861 if (*pt_end == PFN_DOWN(__pa(addr))) {
1862 set_page_prot_flags((void *)addr, PAGE_KERNEL, UVMF_INVLPG);
1863 clear_page((void *)addr);
1868 * Set up the initial kernel pagetable.
1870 * We can construct this by grafting the Xen provided pagetable into
1871 * head_64.S's preconstructed pagetables. We copy the Xen L2's into
1872 * level2_ident_pgt, and level2_kernel_pgt. This means that only the
1873 * kernel has a physical mapping to start with - but that's enough to
1874 * get __va working. We need to fill in the rest of the physical
1875 * mapping once some sort of allocator has been set up.
1877 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
1881 unsigned long addr[3];
1882 unsigned long pt_base, pt_end;
1885 /* max_pfn_mapped is the last pfn mapped in the initial memory
1886 * mappings. Considering that on Xen after the kernel mappings we
1887 * have the mappings of some pages that don't exist in pfn space, we
1888 * set max_pfn_mapped to the last real pfn mapped. */
1889 if (xen_start_info->mfn_list < __START_KERNEL_map)
1890 max_pfn_mapped = xen_start_info->first_p2m_pfn;
1892 max_pfn_mapped = PFN_DOWN(__pa(xen_start_info->mfn_list));
1894 pt_base = PFN_DOWN(__pa(xen_start_info->pt_base));
1895 pt_end = pt_base + xen_start_info->nr_pt_frames;
1897 /* Zap identity mapping */
1898 init_top_pgt[0] = __pgd(0);
1900 /* Pre-constructed entries are in pfn, so convert to mfn */
1901 /* L4[272] -> level3_ident_pgt */
1902 /* L4[511] -> level3_kernel_pgt */
1903 convert_pfn_mfn(init_top_pgt);
1905 /* L3_i[0] -> level2_ident_pgt */
1906 convert_pfn_mfn(level3_ident_pgt);
1907 /* L3_k[510] -> level2_kernel_pgt */
1908 /* L3_k[511] -> level2_fixmap_pgt */
1909 convert_pfn_mfn(level3_kernel_pgt);
1911 /* L3_k[511][506] -> level1_fixmap_pgt */
1912 convert_pfn_mfn(level2_fixmap_pgt);
1914 /* We get [511][511] and have Xen's version of level2_kernel_pgt */
1915 l3 = m2v(pgd[pgd_index(__START_KERNEL_map)].pgd);
1916 l2 = m2v(l3[pud_index(__START_KERNEL_map)].pud);
1918 addr[0] = (unsigned long)pgd;
1919 addr[1] = (unsigned long)l3;
1920 addr[2] = (unsigned long)l2;
1921 /* Graft it onto L4[272][0]. Note that we creating an aliasing problem:
1922 * Both L4[272][0] and L4[511][510] have entries that point to the same
1923 * L2 (PMD) tables. Meaning that if you modify it in __va space
1924 * it will be also modified in the __ka space! (But if you just
1925 * modify the PMD table to point to other PTE's or none, then you
1926 * are OK - which is what cleanup_highmap does) */
1927 copy_page(level2_ident_pgt, l2);
1928 /* Graft it onto L4[511][510] */
1929 copy_page(level2_kernel_pgt, l2);
1932 * Zap execute permission from the ident map. Due to the sharing of
1933 * L1 entries we need to do this in the L2.
1935 if (__supported_pte_mask & _PAGE_NX) {
1936 for (i = 0; i < PTRS_PER_PMD; ++i) {
1937 if (pmd_none(level2_ident_pgt[i]))
1939 level2_ident_pgt[i] = pmd_set_flags(level2_ident_pgt[i], _PAGE_NX);
1943 /* Copy the initial P->M table mappings if necessary. */
1944 i = pgd_index(xen_start_info->mfn_list);
1945 if (i && i < pgd_index(__START_KERNEL_map))
1946 init_top_pgt[i] = ((pgd_t *)xen_start_info->pt_base)[i];
1948 /* Make pagetable pieces RO */
1949 set_page_prot(init_top_pgt, PAGE_KERNEL_RO);
1950 set_page_prot(level3_ident_pgt, PAGE_KERNEL_RO);
1951 set_page_prot(level3_kernel_pgt, PAGE_KERNEL_RO);
1952 set_page_prot(level3_user_vsyscall, PAGE_KERNEL_RO);
1953 set_page_prot(level2_ident_pgt, PAGE_KERNEL_RO);
1954 set_page_prot(level2_kernel_pgt, PAGE_KERNEL_RO);
1955 set_page_prot(level2_fixmap_pgt, PAGE_KERNEL_RO);
1956 set_page_prot(level1_fixmap_pgt, PAGE_KERNEL_RO);
1958 /* Pin down new L4 */
1959 pin_pagetable_pfn(MMUEXT_PIN_L4_TABLE,
1960 PFN_DOWN(__pa_symbol(init_top_pgt)));
1962 /* Unpin Xen-provided one */
1963 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
1966 * At this stage there can be no user pgd, and no page structure to
1967 * attach it to, so make sure we just set kernel pgd.
1970 __xen_write_cr3(true, __pa(init_top_pgt));
1971 xen_mc_issue(PARAVIRT_LAZY_CPU);
1973 /* We can't that easily rip out L3 and L2, as the Xen pagetables are
1974 * set out this way: [L4], [L1], [L2], [L3], [L1], [L1] ... for
1975 * the initial domain. For guests using the toolstack, they are in:
1976 * [L4], [L3], [L2], [L1], [L1], order .. So for dom0 we can only
1977 * rip out the [L4] (pgd), but for guests we shave off three pages.
1979 for (i = 0; i < ARRAY_SIZE(addr); i++)
1980 check_pt_base(&pt_base, &pt_end, addr[i]);
1982 /* Our (by three pages) smaller Xen pagetable that we are using */
1983 xen_pt_base = PFN_PHYS(pt_base);
1984 xen_pt_size = (pt_end - pt_base) * PAGE_SIZE;
1985 memblock_reserve(xen_pt_base, xen_pt_size);
1987 /* Revector the xen_start_info */
1988 xen_start_info = (struct start_info *)__va(__pa(xen_start_info));
1992 * Read a value from a physical address.
1994 static unsigned long __init xen_read_phys_ulong(phys_addr_t addr)
1996 unsigned long *vaddr;
1999 vaddr = early_memremap_ro(addr, sizeof(val));
2001 early_memunmap(vaddr, sizeof(val));
2006 * Translate a virtual address to a physical one without relying on mapped
2007 * page tables. Don't rely on big pages being aligned in (guest) physical
2010 static phys_addr_t __init xen_early_virt_to_phys(unsigned long vaddr)
2019 pgd = native_make_pgd(xen_read_phys_ulong(pa + pgd_index(vaddr) *
2021 if (!pgd_present(pgd))
2024 pa = pgd_val(pgd) & PTE_PFN_MASK;
2025 pud = native_make_pud(xen_read_phys_ulong(pa + pud_index(vaddr) *
2027 if (!pud_present(pud))
2029 pa = pud_val(pud) & PTE_PFN_MASK;
2031 return pa + (vaddr & ~PUD_MASK);
2033 pmd = native_make_pmd(xen_read_phys_ulong(pa + pmd_index(vaddr) *
2035 if (!pmd_present(pmd))
2037 pa = pmd_val(pmd) & PTE_PFN_MASK;
2039 return pa + (vaddr & ~PMD_MASK);
2041 pte = native_make_pte(xen_read_phys_ulong(pa + pte_index(vaddr) *
2043 if (!pte_present(pte))
2045 pa = pte_pfn(pte) << PAGE_SHIFT;
2047 return pa | (vaddr & ~PAGE_MASK);
2051 * Find a new area for the hypervisor supplied p2m list and relocate the p2m to
2054 void __init xen_relocate_p2m(void)
2056 phys_addr_t size, new_area, pt_phys, pmd_phys, pud_phys;
2057 unsigned long p2m_pfn, p2m_pfn_end, n_frames, pfn, pfn_end;
2058 int n_pte, n_pt, n_pmd, n_pud, idx_pte, idx_pt, idx_pmd, idx_pud;
2063 unsigned long *new_p2m;
2066 size = PAGE_ALIGN(xen_start_info->nr_pages * sizeof(unsigned long));
2067 n_pte = roundup(size, PAGE_SIZE) >> PAGE_SHIFT;
2068 n_pt = roundup(size, PMD_SIZE) >> PMD_SHIFT;
2069 n_pmd = roundup(size, PUD_SIZE) >> PUD_SHIFT;
2070 n_pud = roundup(size, P4D_SIZE) >> P4D_SHIFT;
2071 n_frames = n_pte + n_pt + n_pmd + n_pud;
2073 new_area = xen_find_free_area(PFN_PHYS(n_frames));
2075 xen_raw_console_write("Can't find new memory area for p2m needed due to E820 map conflict\n");
2080 * Setup the page tables for addressing the new p2m list.
2081 * We have asked the hypervisor to map the p2m list at the user address
2082 * PUD_SIZE. It may have done so, or it may have used a kernel space
2083 * address depending on the Xen version.
2084 * To avoid any possible virtual address collision, just use
2085 * 2 * PUD_SIZE for the new area.
2087 pud_phys = new_area;
2088 pmd_phys = pud_phys + PFN_PHYS(n_pud);
2089 pt_phys = pmd_phys + PFN_PHYS(n_pmd);
2090 p2m_pfn = PFN_DOWN(pt_phys) + n_pt;
2092 pgd = __va(read_cr3_pa());
2093 new_p2m = (unsigned long *)(2 * PGDIR_SIZE);
2095 for (idx_pud = 0; idx_pud < n_pud; idx_pud++) {
2096 pud = early_memremap(pud_phys, PAGE_SIZE);
2098 for (idx_pmd = 0; idx_pmd < min(n_pmd, PTRS_PER_PUD);
2100 pmd = early_memremap(pmd_phys, PAGE_SIZE);
2102 for (idx_pt = 0; idx_pt < min(n_pt, PTRS_PER_PMD);
2104 pt = early_memremap(pt_phys, PAGE_SIZE);
2107 idx_pte < min(n_pte, PTRS_PER_PTE);
2109 set_pte(pt + idx_pte,
2110 pfn_pte(p2m_pfn, PAGE_KERNEL));
2113 n_pte -= PTRS_PER_PTE;
2114 early_memunmap(pt, PAGE_SIZE);
2115 make_lowmem_page_readonly(__va(pt_phys));
2116 pin_pagetable_pfn(MMUEXT_PIN_L1_TABLE,
2118 set_pmd(pmd + idx_pt,
2119 __pmd(_PAGE_TABLE | pt_phys));
2120 pt_phys += PAGE_SIZE;
2122 n_pt -= PTRS_PER_PMD;
2123 early_memunmap(pmd, PAGE_SIZE);
2124 make_lowmem_page_readonly(__va(pmd_phys));
2125 pin_pagetable_pfn(MMUEXT_PIN_L2_TABLE,
2126 PFN_DOWN(pmd_phys));
2127 set_pud(pud + idx_pmd, __pud(_PAGE_TABLE | pmd_phys));
2128 pmd_phys += PAGE_SIZE;
2130 n_pmd -= PTRS_PER_PUD;
2131 early_memunmap(pud, PAGE_SIZE);
2132 make_lowmem_page_readonly(__va(pud_phys));
2133 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, PFN_DOWN(pud_phys));
2134 set_pgd(pgd + 2 + idx_pud, __pgd(_PAGE_TABLE | pud_phys));
2135 pud_phys += PAGE_SIZE;
2138 /* Now copy the old p2m info to the new area. */
2139 memcpy(new_p2m, xen_p2m_addr, size);
2140 xen_p2m_addr = new_p2m;
2142 /* Release the old p2m list and set new list info. */
2143 p2m_pfn = PFN_DOWN(xen_early_virt_to_phys(xen_start_info->mfn_list));
2145 p2m_pfn_end = p2m_pfn + PFN_DOWN(size);
2147 if (xen_start_info->mfn_list < __START_KERNEL_map) {
2148 pfn = xen_start_info->first_p2m_pfn;
2149 pfn_end = xen_start_info->first_p2m_pfn +
2150 xen_start_info->nr_p2m_frames;
2151 set_pgd(pgd + 1, __pgd(0));
2154 pfn_end = p2m_pfn_end;
2157 memblock_free(PFN_PHYS(pfn), PAGE_SIZE * (pfn_end - pfn));
2158 while (pfn < pfn_end) {
2159 if (pfn == p2m_pfn) {
2163 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
2167 xen_start_info->mfn_list = (unsigned long)xen_p2m_addr;
2168 xen_start_info->first_p2m_pfn = PFN_DOWN(new_area);
2169 xen_start_info->nr_p2m_frames = n_frames;
2172 #else /* !CONFIG_X86_64 */
2173 static RESERVE_BRK_ARRAY(pmd_t, initial_kernel_pmd, PTRS_PER_PMD);
2174 static RESERVE_BRK_ARRAY(pmd_t, swapper_kernel_pmd, PTRS_PER_PMD);
2175 RESERVE_BRK(fixup_kernel_pmd, PAGE_SIZE);
2176 RESERVE_BRK(fixup_kernel_pte, PAGE_SIZE);
2178 static void __init xen_write_cr3_init(unsigned long cr3)
2180 unsigned long pfn = PFN_DOWN(__pa(swapper_pg_dir));
2182 BUG_ON(read_cr3_pa() != __pa(initial_page_table));
2183 BUG_ON(cr3 != __pa(swapper_pg_dir));
2186 * We are switching to swapper_pg_dir for the first time (from
2187 * initial_page_table) and therefore need to mark that page
2188 * read-only and then pin it.
2190 * Xen disallows sharing of kernel PMDs for PAE
2191 * guests. Therefore we must copy the kernel PMD from
2192 * initial_page_table into a new kernel PMD to be used in
2195 swapper_kernel_pmd =
2196 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
2197 copy_page(swapper_kernel_pmd, initial_kernel_pmd);
2198 swapper_pg_dir[KERNEL_PGD_BOUNDARY] =
2199 __pgd(__pa(swapper_kernel_pmd) | _PAGE_PRESENT);
2200 set_page_prot(swapper_kernel_pmd, PAGE_KERNEL_RO);
2202 set_page_prot(swapper_pg_dir, PAGE_KERNEL_RO);
2204 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE, pfn);
2206 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE,
2207 PFN_DOWN(__pa(initial_page_table)));
2208 set_page_prot(initial_page_table, PAGE_KERNEL);
2209 set_page_prot(initial_kernel_pmd, PAGE_KERNEL);
2211 pv_mmu_ops.write_cr3 = &xen_write_cr3;
2215 * For 32 bit domains xen_start_info->pt_base is the pgd address which might be
2216 * not the first page table in the page table pool.
2217 * Iterate through the initial page tables to find the real page table base.
2219 static phys_addr_t __init xen_find_pt_base(pmd_t *pmd)
2221 phys_addr_t pt_base, paddr;
2224 pt_base = min(__pa(xen_start_info->pt_base), __pa(pmd));
2226 for (pmdidx = 0; pmdidx < PTRS_PER_PMD; pmdidx++)
2227 if (pmd_present(pmd[pmdidx]) && !pmd_large(pmd[pmdidx])) {
2228 paddr = m2p(pmd[pmdidx].pmd);
2229 pt_base = min(pt_base, paddr);
2235 void __init xen_setup_kernel_pagetable(pgd_t *pgd, unsigned long max_pfn)
2239 kernel_pmd = m2v(pgd[KERNEL_PGD_BOUNDARY].pgd);
2241 xen_pt_base = xen_find_pt_base(kernel_pmd);
2242 xen_pt_size = xen_start_info->nr_pt_frames * PAGE_SIZE;
2244 initial_kernel_pmd =
2245 extend_brk(sizeof(pmd_t) * PTRS_PER_PMD, PAGE_SIZE);
2247 max_pfn_mapped = PFN_DOWN(xen_pt_base + xen_pt_size + 512 * 1024);
2249 copy_page(initial_kernel_pmd, kernel_pmd);
2251 xen_map_identity_early(initial_kernel_pmd, max_pfn);
2253 copy_page(initial_page_table, pgd);
2254 initial_page_table[KERNEL_PGD_BOUNDARY] =
2255 __pgd(__pa(initial_kernel_pmd) | _PAGE_PRESENT);
2257 set_page_prot(initial_kernel_pmd, PAGE_KERNEL_RO);
2258 set_page_prot(initial_page_table, PAGE_KERNEL_RO);
2259 set_page_prot(empty_zero_page, PAGE_KERNEL_RO);
2261 pin_pagetable_pfn(MMUEXT_UNPIN_TABLE, PFN_DOWN(__pa(pgd)));
2263 pin_pagetable_pfn(MMUEXT_PIN_L3_TABLE,
2264 PFN_DOWN(__pa(initial_page_table)));
2265 xen_write_cr3(__pa(initial_page_table));
2267 memblock_reserve(xen_pt_base, xen_pt_size);
2269 #endif /* CONFIG_X86_64 */
2271 void __init xen_reserve_special_pages(void)
2275 memblock_reserve(__pa(xen_start_info), PAGE_SIZE);
2276 if (xen_start_info->store_mfn) {
2277 paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->store_mfn));
2278 memblock_reserve(paddr, PAGE_SIZE);
2280 if (!xen_initial_domain()) {
2281 paddr = PFN_PHYS(mfn_to_pfn(xen_start_info->console.domU.mfn));
2282 memblock_reserve(paddr, PAGE_SIZE);
2286 void __init xen_pt_check_e820(void)
2288 if (xen_is_e820_reserved(xen_pt_base, xen_pt_size)) {
2289 xen_raw_console_write("Xen hypervisor allocated page table memory conflicts with E820 map\n");
2294 static unsigned char dummy_mapping[PAGE_SIZE] __page_aligned_bss;
2296 static void xen_set_fixmap(unsigned idx, phys_addr_t phys, pgprot_t prot)
2300 phys >>= PAGE_SHIFT;
2303 case FIX_BTMAP_END ... FIX_BTMAP_BEGIN:
2304 #ifdef CONFIG_X86_32
2306 # ifdef CONFIG_HIGHMEM
2307 case FIX_KMAP_BEGIN ... FIX_KMAP_END:
2309 #elif defined(CONFIG_X86_VSYSCALL_EMULATION)
2312 case FIX_TEXT_POKE0:
2313 case FIX_TEXT_POKE1:
2314 /* All local page mappings */
2315 pte = pfn_pte(phys, prot);
2318 #ifdef CONFIG_X86_LOCAL_APIC
2319 case FIX_APIC_BASE: /* maps dummy local APIC */
2320 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2324 #ifdef CONFIG_X86_IO_APIC
2325 case FIX_IO_APIC_BASE_0 ... FIX_IO_APIC_BASE_END:
2327 * We just don't map the IO APIC - all access is via
2328 * hypercalls. Keep the address in the pte for reference.
2330 pte = pfn_pte(PFN_DOWN(__pa(dummy_mapping)), PAGE_KERNEL);
2334 case FIX_PARAVIRT_BOOTMAP:
2335 /* This is an MFN, but it isn't an IO mapping from the
2337 pte = mfn_pte(phys, prot);
2341 /* By default, set_fixmap is used for hardware mappings */
2342 pte = mfn_pte(phys, prot);
2346 __native_set_fixmap(idx, pte);
2348 #ifdef CONFIG_X86_VSYSCALL_EMULATION
2349 /* Replicate changes to map the vsyscall page into the user
2350 pagetable vsyscall mapping. */
2351 if (idx == VSYSCALL_PAGE) {
2352 unsigned long vaddr = __fix_to_virt(idx);
2353 set_pte_vaddr_pud(level3_user_vsyscall, vaddr, pte);
2358 static void __init xen_post_allocator_init(void)
2360 pv_mmu_ops.set_pte = xen_set_pte;
2361 pv_mmu_ops.set_pmd = xen_set_pmd;
2362 pv_mmu_ops.set_pud = xen_set_pud;
2363 #ifdef CONFIG_X86_64
2364 pv_mmu_ops.set_p4d = xen_set_p4d;
2367 /* This will work as long as patching hasn't happened yet
2368 (which it hasn't) */
2369 pv_mmu_ops.alloc_pte = xen_alloc_pte;
2370 pv_mmu_ops.alloc_pmd = xen_alloc_pmd;
2371 pv_mmu_ops.release_pte = xen_release_pte;
2372 pv_mmu_ops.release_pmd = xen_release_pmd;
2373 #ifdef CONFIG_X86_64
2374 pv_mmu_ops.alloc_pud = xen_alloc_pud;
2375 pv_mmu_ops.release_pud = xen_release_pud;
2377 pv_mmu_ops.make_pte = PV_CALLEE_SAVE(xen_make_pte);
2379 #ifdef CONFIG_X86_64
2380 pv_mmu_ops.write_cr3 = &xen_write_cr3;
2384 static void xen_leave_lazy_mmu(void)
2388 paravirt_leave_lazy_mmu();
2392 static const struct pv_mmu_ops xen_mmu_ops __initconst = {
2393 .read_cr2 = xen_read_cr2,
2394 .write_cr2 = xen_write_cr2,
2396 .read_cr3 = xen_read_cr3,
2397 .write_cr3 = xen_write_cr3_init,
2399 .flush_tlb_user = xen_flush_tlb,
2400 .flush_tlb_kernel = xen_flush_tlb,
2401 .flush_tlb_one_user = xen_flush_tlb_one_user,
2402 .flush_tlb_others = xen_flush_tlb_others,
2403 .tlb_remove_table = tlb_remove_table,
2405 .pgd_alloc = xen_pgd_alloc,
2406 .pgd_free = xen_pgd_free,
2408 .alloc_pte = xen_alloc_pte_init,
2409 .release_pte = xen_release_pte_init,
2410 .alloc_pmd = xen_alloc_pmd_init,
2411 .release_pmd = xen_release_pmd_init,
2413 .set_pte = xen_set_pte_init,
2414 .set_pte_at = xen_set_pte_at,
2415 .set_pmd = xen_set_pmd_hyper,
2417 .ptep_modify_prot_start = __ptep_modify_prot_start,
2418 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
2420 .pte_val = PV_CALLEE_SAVE(xen_pte_val),
2421 .pgd_val = PV_CALLEE_SAVE(xen_pgd_val),
2423 .make_pte = PV_CALLEE_SAVE(xen_make_pte_init),
2424 .make_pgd = PV_CALLEE_SAVE(xen_make_pgd),
2426 #ifdef CONFIG_X86_PAE
2427 .set_pte_atomic = xen_set_pte_atomic,
2428 .pte_clear = xen_pte_clear,
2429 .pmd_clear = xen_pmd_clear,
2430 #endif /* CONFIG_X86_PAE */
2431 .set_pud = xen_set_pud_hyper,
2433 .make_pmd = PV_CALLEE_SAVE(xen_make_pmd),
2434 .pmd_val = PV_CALLEE_SAVE(xen_pmd_val),
2436 #ifdef CONFIG_X86_64
2437 .pud_val = PV_CALLEE_SAVE(xen_pud_val),
2438 .make_pud = PV_CALLEE_SAVE(xen_make_pud),
2439 .set_p4d = xen_set_p4d_hyper,
2441 .alloc_pud = xen_alloc_pmd_init,
2442 .release_pud = xen_release_pmd_init,
2444 #if CONFIG_PGTABLE_LEVELS >= 5
2445 .p4d_val = PV_CALLEE_SAVE(xen_p4d_val),
2446 .make_p4d = PV_CALLEE_SAVE(xen_make_p4d),
2448 #endif /* CONFIG_X86_64 */
2450 .activate_mm = xen_activate_mm,
2451 .dup_mmap = xen_dup_mmap,
2452 .exit_mmap = xen_exit_mmap,
2455 .enter = paravirt_enter_lazy_mmu,
2456 .leave = xen_leave_lazy_mmu,
2457 .flush = paravirt_flush_lazy_mmu,
2460 .set_fixmap = xen_set_fixmap,
2463 void __init xen_init_mmu_ops(void)
2465 x86_init.paging.pagetable_init = xen_pagetable_init;
2466 x86_init.hyper.init_after_bootmem = xen_after_bootmem;
2468 pv_mmu_ops = xen_mmu_ops;
2470 memset(dummy_mapping, 0xff, PAGE_SIZE);
2473 /* Protected by xen_reservation_lock. */
2474 #define MAX_CONTIG_ORDER 9 /* 2MB */
2475 static unsigned long discontig_frames[1<<MAX_CONTIG_ORDER];
2477 #define VOID_PTE (mfn_pte(0, __pgprot(0)))
2478 static void xen_zap_pfn_range(unsigned long vaddr, unsigned int order,
2479 unsigned long *in_frames,
2480 unsigned long *out_frames)
2483 struct multicall_space mcs;
2486 for (i = 0; i < (1UL<<order); i++, vaddr += PAGE_SIZE) {
2487 mcs = __xen_mc_entry(0);
2490 in_frames[i] = virt_to_mfn(vaddr);
2492 MULTI_update_va_mapping(mcs.mc, vaddr, VOID_PTE, 0);
2493 __set_phys_to_machine(virt_to_pfn(vaddr), INVALID_P2M_ENTRY);
2496 out_frames[i] = virt_to_pfn(vaddr);
2502 * Update the pfn-to-mfn mappings for a virtual address range, either to
2503 * point to an array of mfns, or contiguously from a single starting
2506 static void xen_remap_exchanged_ptes(unsigned long vaddr, int order,
2507 unsigned long *mfns,
2508 unsigned long first_mfn)
2515 limit = 1u << order;
2516 for (i = 0; i < limit; i++, vaddr += PAGE_SIZE) {
2517 struct multicall_space mcs;
2520 mcs = __xen_mc_entry(0);
2524 mfn = first_mfn + i;
2526 if (i < (limit - 1))
2530 flags = UVMF_INVLPG | UVMF_ALL;
2532 flags = UVMF_TLB_FLUSH | UVMF_ALL;
2535 MULTI_update_va_mapping(mcs.mc, vaddr,
2536 mfn_pte(mfn, PAGE_KERNEL), flags);
2538 set_phys_to_machine(virt_to_pfn(vaddr), mfn);
2545 * Perform the hypercall to exchange a region of our pfns to point to
2546 * memory with the required contiguous alignment. Takes the pfns as
2547 * input, and populates mfns as output.
2549 * Returns a success code indicating whether the hypervisor was able to
2550 * satisfy the request or not.
2552 static int xen_exchange_memory(unsigned long extents_in, unsigned int order_in,
2553 unsigned long *pfns_in,
2554 unsigned long extents_out,
2555 unsigned int order_out,
2556 unsigned long *mfns_out,
2557 unsigned int address_bits)
2562 struct xen_memory_exchange exchange = {
2564 .nr_extents = extents_in,
2565 .extent_order = order_in,
2566 .extent_start = pfns_in,
2570 .nr_extents = extents_out,
2571 .extent_order = order_out,
2572 .extent_start = mfns_out,
2573 .address_bits = address_bits,
2578 BUG_ON(extents_in << order_in != extents_out << order_out);
2580 rc = HYPERVISOR_memory_op(XENMEM_exchange, &exchange);
2581 success = (exchange.nr_exchanged == extents_in);
2583 BUG_ON(!success && ((exchange.nr_exchanged != 0) || (rc == 0)));
2584 BUG_ON(success && (rc != 0));
2589 int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
2590 unsigned int address_bits,
2591 dma_addr_t *dma_handle)
2593 unsigned long *in_frames = discontig_frames, out_frame;
2594 unsigned long flags;
2596 unsigned long vstart = (unsigned long)phys_to_virt(pstart);
2599 * Currently an auto-translated guest will not perform I/O, nor will
2600 * it require PAE page directories below 4GB. Therefore any calls to
2601 * this function are redundant and can be ignored.
2604 if (unlikely(order > MAX_CONTIG_ORDER))
2607 memset((void *) vstart, 0, PAGE_SIZE << order);
2609 spin_lock_irqsave(&xen_reservation_lock, flags);
2611 /* 1. Zap current PTEs, remembering MFNs. */
2612 xen_zap_pfn_range(vstart, order, in_frames, NULL);
2614 /* 2. Get a new contiguous memory extent. */
2615 out_frame = virt_to_pfn(vstart);
2616 success = xen_exchange_memory(1UL << order, 0, in_frames,
2617 1, order, &out_frame,
2620 /* 3. Map the new extent in place of old pages. */
2622 xen_remap_exchanged_ptes(vstart, order, NULL, out_frame);
2624 xen_remap_exchanged_ptes(vstart, order, in_frames, 0);
2626 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2628 *dma_handle = virt_to_machine(vstart).maddr;
2629 return success ? 0 : -ENOMEM;
2631 EXPORT_SYMBOL_GPL(xen_create_contiguous_region);
2633 void xen_destroy_contiguous_region(phys_addr_t pstart, unsigned int order)
2635 unsigned long *out_frames = discontig_frames, in_frame;
2636 unsigned long flags;
2638 unsigned long vstart;
2640 if (unlikely(order > MAX_CONTIG_ORDER))
2643 vstart = (unsigned long)phys_to_virt(pstart);
2644 memset((void *) vstart, 0, PAGE_SIZE << order);
2646 spin_lock_irqsave(&xen_reservation_lock, flags);
2648 /* 1. Find start MFN of contiguous extent. */
2649 in_frame = virt_to_mfn(vstart);
2651 /* 2. Zap current PTEs. */
2652 xen_zap_pfn_range(vstart, order, NULL, out_frames);
2654 /* 3. Do the exchange for non-contiguous MFNs. */
2655 success = xen_exchange_memory(1, order, &in_frame, 1UL << order,
2658 /* 4. Map new pages in place of old pages. */
2660 xen_remap_exchanged_ptes(vstart, order, out_frames, 0);
2662 xen_remap_exchanged_ptes(vstart, order, NULL, in_frame);
2664 spin_unlock_irqrestore(&xen_reservation_lock, flags);
2666 EXPORT_SYMBOL_GPL(xen_destroy_contiguous_region);
2668 #ifdef CONFIG_KEXEC_CORE
2669 phys_addr_t paddr_vmcoreinfo_note(void)
2671 if (xen_pv_domain())
2672 return virt_to_machine(vmcoreinfo_note).maddr;
2674 return __pa(vmcoreinfo_note);
2676 #endif /* CONFIG_KEXEC_CORE */