Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1 | /*P:700 |
| 2 | * The pagetable code, on the other hand, still shows the scars of |
Rusty Russell | f938d2c | 2007-07-26 10:41:02 -0700 | [diff] [blame] | 3 | * previous encounters. It's functional, and as neat as it can be in the |
| 4 | * circumstances, but be wary, for these things are subtle and break easily. |
| 5 | * The Guest provides a virtual to physical mapping, but we can neither trust |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 6 | * it nor use it: we verify and convert it here then point the CPU to the |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 7 | * converted Guest pages when running the Guest. |
| 8 | :*/ |
Rusty Russell | f938d2c | 2007-07-26 10:41:02 -0700 | [diff] [blame] | 9 | |
| 10 | /* Copyright (C) Rusty Russell IBM Corporation 2006. |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 11 | * GPL v2 and any later version */ |
| 12 | #include <linux/mm.h> |
| 13 | #include <linux/types.h> |
| 14 | #include <linux/spinlock.h> |
| 15 | #include <linux/random.h> |
| 16 | #include <linux/percpu.h> |
| 17 | #include <asm/tlbflush.h> |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 18 | #include <asm/uaccess.h> |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 19 | #include <asm/bootparam.h> |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 20 | #include "lg.h" |
| 21 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 22 | /*M:008 |
| 23 | * We hold reference to pages, which prevents them from being swapped. |
Rusty Russell | f56a384 | 2007-07-26 10:41:05 -0700 | [diff] [blame] | 24 | * It'd be nice to have a callback in the "struct mm_struct" when Linux wants |
| 25 | * to swap out. If we had this, and a shrinker callback to trim PTE pages, we |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 26 | * could probably consider launching Guests as non-root. |
| 27 | :*/ |
Rusty Russell | f56a384 | 2007-07-26 10:41:05 -0700 | [diff] [blame] | 28 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 29 | /*H:300 |
| 30 | * The Page Table Code |
| 31 | * |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 32 | * We use two-level page tables for the Guest, or three-level with PAE. If |
| 33 | * you're not entirely comfortable with virtual addresses, physical addresses |
| 34 | * and page tables then I recommend you review arch/x86/lguest/boot.c's "Page |
| 35 | * Table Handling" (with diagrams!). |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 36 | * |
| 37 | * The Guest keeps page tables, but we maintain the actual ones here: these are |
| 38 | * called "shadow" page tables. Which is a very Guest-centric name: these are |
| 39 | * the real page tables the CPU uses, although we keep them up to date to |
| 40 | * reflect the Guest's. (See what I mean about weird naming? Since when do |
| 41 | * shadows reflect anything?) |
| 42 | * |
| 43 | * Anyway, this is the most complicated part of the Host code. There are seven |
| 44 | * parts to this: |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 45 | * (i) Looking up a page table entry when the Guest faults, |
| 46 | * (ii) Making sure the Guest stack is mapped, |
| 47 | * (iii) Setting up a page table entry when the Guest tells us one has changed, |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 48 | * (iv) Switching page tables, |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 49 | * (v) Flushing (throwing away) page tables, |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 50 | * (vi) Mapping the Switcher when the Guest is about to run, |
| 51 | * (vii) Setting up the page tables initially. |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 52 | :*/ |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 53 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 54 | /* |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 55 | * The Switcher uses the complete top PTE page. That's 1024 PTE entries (4MB) |
| 56 | * or 512 PTE entries with PAE (2MB). |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 57 | */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 58 | #define SWITCHER_PGD_INDEX (PTRS_PER_PGD - 1) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 59 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 60 | /* |
| 61 | * For PAE we need the PMD index as well. We use the last 2MB, so we |
| 62 | * will need the last pmd entry of the last pmd page. |
| 63 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 64 | #ifdef CONFIG_X86_PAE |
| 65 | #define SWITCHER_PMD_INDEX (PTRS_PER_PMD - 1) |
| 66 | #define RESERVE_MEM 2U |
| 67 | #define CHECK_GPGD_MASK _PAGE_PRESENT |
| 68 | #else |
| 69 | #define RESERVE_MEM 4U |
| 70 | #define CHECK_GPGD_MASK _PAGE_TABLE |
| 71 | #endif |
| 72 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 73 | /* |
| 74 | * We actually need a separate PTE page for each CPU. Remember that after the |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 75 | * Switcher code itself comes two pages for each CPU, and we don't want this |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 76 | * CPU's guest to see the pages of any other CPU. |
| 77 | */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 78 | static DEFINE_PER_CPU(pte_t *, switcher_pte_pages); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 79 | #define switcher_pte_page(cpu) per_cpu(switcher_pte_pages, cpu) |
| 80 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 81 | /*H:320 |
| 82 | * The page table code is curly enough to need helper functions to keep it |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 83 | * clear and clean. The kernel itself provides many of them; one advantage |
| 84 | * of insisting that the Guest and Host use the same CONFIG_PAE setting. |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 85 | * |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 86 | * There are two functions which return pointers to the shadow (aka "real") |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 87 | * page tables. |
| 88 | * |
| 89 | * spgd_addr() takes the virtual address and returns a pointer to the top-level |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 90 | * page directory entry (PGD) for that address. Since we keep track of several |
| 91 | * page tables, the "i" argument tells us which one we're interested in (it's |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 92 | * usually the current one). |
| 93 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 94 | static pgd_t *spgd_addr(struct lg_cpu *cpu, u32 i, unsigned long vaddr) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 95 | { |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 96 | unsigned int index = pgd_index(vaddr); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 97 | |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 98 | #ifndef CONFIG_X86_PAE |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 99 | /* We kill any Guest trying to touch the Switcher addresses. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 100 | if (index >= SWITCHER_PGD_INDEX) { |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 101 | kill_guest(cpu, "attempt to access switcher pages"); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 102 | index = 0; |
| 103 | } |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 104 | #endif |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 105 | /* Return a pointer index'th pgd entry for the i'th page table. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 106 | return &cpu->lg->pgdirs[i].pgdir[index]; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 107 | } |
| 108 | |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 109 | #ifdef CONFIG_X86_PAE |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 110 | /* |
| 111 | * This routine then takes the PGD entry given above, which contains the |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 112 | * address of the PMD page. It then returns a pointer to the PMD entry for the |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 113 | * given address. |
| 114 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 115 | static pmd_t *spmd_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr) |
| 116 | { |
| 117 | unsigned int index = pmd_index(vaddr); |
| 118 | pmd_t *page; |
| 119 | |
| 120 | /* We kill any Guest trying to touch the Switcher addresses. */ |
| 121 | if (pgd_index(vaddr) == SWITCHER_PGD_INDEX && |
| 122 | index >= SWITCHER_PMD_INDEX) { |
| 123 | kill_guest(cpu, "attempt to access switcher pages"); |
| 124 | index = 0; |
| 125 | } |
| 126 | |
| 127 | /* You should never call this if the PGD entry wasn't valid */ |
| 128 | BUG_ON(!(pgd_flags(spgd) & _PAGE_PRESENT)); |
| 129 | page = __va(pgd_pfn(spgd) << PAGE_SHIFT); |
| 130 | |
| 131 | return &page[index]; |
| 132 | } |
| 133 | #endif |
| 134 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 135 | /* |
| 136 | * This routine then takes the page directory entry returned above, which |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 137 | * contains the address of the page table entry (PTE) page. It then returns a |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 138 | * pointer to the PTE entry for the given address. |
| 139 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 140 | static pte_t *spte_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 141 | { |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 142 | #ifdef CONFIG_X86_PAE |
| 143 | pmd_t *pmd = spmd_addr(cpu, spgd, vaddr); |
| 144 | pte_t *page = __va(pmd_pfn(*pmd) << PAGE_SHIFT); |
| 145 | |
| 146 | /* You should never call this if the PMD entry wasn't valid */ |
| 147 | BUG_ON(!(pmd_flags(*pmd) & _PAGE_PRESENT)); |
| 148 | #else |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 149 | pte_t *page = __va(pgd_pfn(spgd) << PAGE_SHIFT); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 150 | /* You should never call this if the PGD entry wasn't valid */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 151 | BUG_ON(!(pgd_flags(spgd) & _PAGE_PRESENT)); |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 152 | #endif |
| 153 | |
Matias Zabaljauregui | 90603d1 | 2009-06-12 22:27:06 -0600 | [diff] [blame] | 154 | return &page[pte_index(vaddr)]; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 155 | } |
| 156 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 157 | /* |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 158 | * These functions are just like the above two, except they access the Guest |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 159 | * page tables. Hence they return a Guest address. |
| 160 | */ |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 161 | static unsigned long gpgd_addr(struct lg_cpu *cpu, unsigned long vaddr) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 162 | { |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 163 | unsigned int index = vaddr >> (PGDIR_SHIFT); |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 164 | return cpu->lg->pgdirs[cpu->cpu_pgd].gpgdir + index * sizeof(pgd_t); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 165 | } |
| 166 | |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 167 | #ifdef CONFIG_X86_PAE |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 168 | /* Follow the PGD to the PMD. */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 169 | static unsigned long gpmd_addr(pgd_t gpgd, unsigned long vaddr) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 170 | { |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 171 | unsigned long gpage = pgd_pfn(gpgd) << PAGE_SHIFT; |
| 172 | BUG_ON(!(pgd_flags(gpgd) & _PAGE_PRESENT)); |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 173 | return gpage + pmd_index(vaddr) * sizeof(pmd_t); |
| 174 | } |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 175 | |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 176 | /* Follow the PMD to the PTE. */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 177 | static unsigned long gpte_addr(struct lg_cpu *cpu, |
Rusty Russell | 92b4d8d | 2009-06-12 22:27:08 -0600 | [diff] [blame] | 178 | pmd_t gpmd, unsigned long vaddr) |
| 179 | { |
| 180 | unsigned long gpage = pmd_pfn(gpmd) << PAGE_SHIFT; |
| 181 | |
| 182 | BUG_ON(!(pmd_flags(gpmd) & _PAGE_PRESENT)); |
| 183 | return gpage + pte_index(vaddr) * sizeof(pte_t); |
| 184 | } |
| 185 | #else |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 186 | /* Follow the PGD to the PTE (no mid-level for !PAE). */ |
Rusty Russell | 92b4d8d | 2009-06-12 22:27:08 -0600 | [diff] [blame] | 187 | static unsigned long gpte_addr(struct lg_cpu *cpu, |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 188 | pgd_t gpgd, unsigned long vaddr) |
| 189 | { |
Rusty Russell | 92b4d8d | 2009-06-12 22:27:08 -0600 | [diff] [blame] | 190 | unsigned long gpage = pgd_pfn(gpgd) << PAGE_SHIFT; |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 191 | |
| 192 | BUG_ON(!(pgd_flags(gpgd) & _PAGE_PRESENT)); |
Matias Zabaljauregui | 90603d1 | 2009-06-12 22:27:06 -0600 | [diff] [blame] | 193 | return gpage + pte_index(vaddr) * sizeof(pte_t); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 194 | } |
Rusty Russell | 92b4d8d | 2009-06-12 22:27:08 -0600 | [diff] [blame] | 195 | #endif |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 196 | /*:*/ |
| 197 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 198 | /*M:014 |
| 199 | * get_pfn is slow: we could probably try to grab batches of pages here as |
| 200 | * an optimization (ie. pre-faulting). |
| 201 | :*/ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 202 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 203 | /*H:350 |
| 204 | * This routine takes a page number given by the Guest and converts it to |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 205 | * an actual, physical page number. It can fail for several reasons: the |
| 206 | * virtual address might not be mapped by the Launcher, the write flag is set |
| 207 | * and the page is read-only, or the write flag was set and the page was |
| 208 | * shared so had to be copied, but we ran out of memory. |
| 209 | * |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 210 | * This holds a reference to the page, so release_pte() is careful to put that |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 211 | * back. |
| 212 | */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 213 | static unsigned long get_pfn(unsigned long virtpfn, int write) |
| 214 | { |
| 215 | struct page *page; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 216 | |
Rusty Russell | 71a3f4e | 2008-08-12 17:52:53 -0500 | [diff] [blame] | 217 | /* gup me one page at this address please! */ |
| 218 | if (get_user_pages_fast(virtpfn << PAGE_SHIFT, 1, write, &page) == 1) |
| 219 | return page_to_pfn(page); |
| 220 | |
| 221 | /* This value indicates failure. */ |
| 222 | return -1UL; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 223 | } |
| 224 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 225 | /*H:340 |
| 226 | * Converting a Guest page table entry to a shadow (ie. real) page table |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 227 | * entry can be a little tricky. The flags are (almost) the same, but the |
| 228 | * Guest PTE contains a virtual page number: the CPU needs the real page |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 229 | * number. |
| 230 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 231 | static pte_t gpte_to_spte(struct lg_cpu *cpu, pte_t gpte, int write) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 232 | { |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 233 | unsigned long pfn, base, flags; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 234 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 235 | /* |
| 236 | * The Guest sets the global flag, because it thinks that it is using |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 237 | * PGE. We only told it to use PGE so it would tell us whether it was |
| 238 | * flushing a kernel mapping or a userspace mapping. We don't actually |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 239 | * use the global bit, so throw it away. |
| 240 | */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 241 | flags = (pte_flags(gpte) & ~_PAGE_GLOBAL); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 242 | |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 243 | /* The Guest's pages are offset inside the Launcher. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 244 | base = (unsigned long)cpu->lg->mem_base / PAGE_SIZE; |
Rusty Russell | 3c6b5bf | 2007-10-22 11:03:26 +1000 | [diff] [blame] | 245 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 246 | /* |
| 247 | * We need a temporary "unsigned long" variable to hold the answer from |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 248 | * get_pfn(), because it returns 0xFFFFFFFF on failure, which wouldn't |
| 249 | * fit in spte.pfn. get_pfn() finds the real physical number of the |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 250 | * page, given the virtual number. |
| 251 | */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 252 | pfn = get_pfn(base + pte_pfn(gpte), write); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 253 | if (pfn == -1UL) { |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 254 | kill_guest(cpu, "failed to get page %lu", pte_pfn(gpte)); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 255 | /* |
| 256 | * When we destroy the Guest, we'll go through the shadow page |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 257 | * tables and release_pte() them. Make sure we don't think |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 258 | * this one is valid! |
| 259 | */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 260 | flags = 0; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 261 | } |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 262 | /* Now we assemble our shadow PTE from the page number and flags. */ |
| 263 | return pfn_pte(pfn, __pgprot(flags)); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 264 | } |
| 265 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 266 | /*H:460 And to complete the chain, release_pte() looks like this: */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 267 | static void release_pte(pte_t pte) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 268 | { |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 269 | /* |
| 270 | * Remember that get_user_pages_fast() took a reference to the page, in |
| 271 | * get_pfn()? We have to put it back now. |
| 272 | */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 273 | if (pte_flags(pte) & _PAGE_PRESENT) |
Matias Zabaljauregui | 90603d1 | 2009-06-12 22:27:06 -0600 | [diff] [blame] | 274 | put_page(pte_page(pte)); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 275 | } |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 276 | /*:*/ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 277 | |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 278 | static void check_gpte(struct lg_cpu *cpu, pte_t gpte) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 279 | { |
Ahmed S. Darwish | 31f4b46 | 2008-02-09 23:24:09 +0100 | [diff] [blame] | 280 | if ((pte_flags(gpte) & _PAGE_PSE) || |
| 281 | pte_pfn(gpte) >= cpu->lg->pfn_limit) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 282 | kill_guest(cpu, "bad page table entry"); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 283 | } |
| 284 | |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 285 | static void check_gpgd(struct lg_cpu *cpu, pgd_t gpgd) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 286 | { |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 287 | if ((pgd_flags(gpgd) & ~CHECK_GPGD_MASK) || |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 288 | (pgd_pfn(gpgd) >= cpu->lg->pfn_limit)) |
| 289 | kill_guest(cpu, "bad page directory entry"); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 290 | } |
| 291 | |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 292 | #ifdef CONFIG_X86_PAE |
| 293 | static void check_gpmd(struct lg_cpu *cpu, pmd_t gpmd) |
| 294 | { |
| 295 | if ((pmd_flags(gpmd) & ~_PAGE_TABLE) || |
| 296 | (pmd_pfn(gpmd) >= cpu->lg->pfn_limit)) |
| 297 | kill_guest(cpu, "bad page middle directory entry"); |
| 298 | } |
| 299 | #endif |
| 300 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 301 | /*H:330 |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 302 | * (i) Looking up a page table entry when the Guest faults. |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 303 | * |
| 304 | * We saw this call in run_guest(): when we see a page fault in the Guest, we |
| 305 | * come here. That's because we only set up the shadow page tables lazily as |
| 306 | * they're needed, so we get page faults all the time and quietly fix them up |
| 307 | * and return to the Guest without it knowing. |
| 308 | * |
| 309 | * If we fixed up the fault (ie. we mapped the address), this routine returns |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 310 | * true. Otherwise, it was a real fault and we need to tell the Guest. |
| 311 | */ |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 312 | bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 313 | { |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 314 | pgd_t gpgd; |
| 315 | pgd_t *spgd; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 316 | unsigned long gpte_ptr; |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 317 | pte_t gpte; |
| 318 | pte_t *spte; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 319 | |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 320 | /* Mid level for PAE. */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 321 | #ifdef CONFIG_X86_PAE |
| 322 | pmd_t *spmd; |
| 323 | pmd_t gpmd; |
| 324 | #endif |
| 325 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 326 | /* First step: get the top-level Guest page table entry. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 327 | gpgd = lgread(cpu, gpgd_addr(cpu, vaddr), pgd_t); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 328 | /* Toplevel not present? We can't map it in. */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 329 | if (!(pgd_flags(gpgd) & _PAGE_PRESENT)) |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 330 | return false; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 331 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 332 | /* Now look at the matching shadow entry. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 333 | spgd = spgd_addr(cpu, cpu->cpu_pgd, vaddr); |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 334 | if (!(pgd_flags(*spgd) & _PAGE_PRESENT)) { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 335 | /* No shadow entry: allocate a new shadow PTE page. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 336 | unsigned long ptepage = get_zeroed_page(GFP_KERNEL); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 337 | /* |
| 338 | * This is not really the Guest's fault, but killing it is |
| 339 | * simple for this corner case. |
| 340 | */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 341 | if (!ptepage) { |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 342 | kill_guest(cpu, "out of memory allocating pte page"); |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 343 | return false; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 344 | } |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 345 | /* We check that the Guest pgd is OK. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 346 | check_gpgd(cpu, gpgd); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 347 | /* |
| 348 | * And we copy the flags to the shadow PGD entry. The page |
| 349 | * number in the shadow PGD is the page we just allocated. |
| 350 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 351 | set_pgd(spgd, __pgd(__pa(ptepage) | pgd_flags(gpgd))); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 352 | } |
| 353 | |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 354 | #ifdef CONFIG_X86_PAE |
| 355 | gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 356 | /* Middle level not present? We can't map it in. */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 357 | if (!(pmd_flags(gpmd) & _PAGE_PRESENT)) |
| 358 | return false; |
| 359 | |
| 360 | /* Now look at the matching shadow entry. */ |
| 361 | spmd = spmd_addr(cpu, *spgd, vaddr); |
| 362 | |
| 363 | if (!(pmd_flags(*spmd) & _PAGE_PRESENT)) { |
| 364 | /* No shadow entry: allocate a new shadow PTE page. */ |
| 365 | unsigned long ptepage = get_zeroed_page(GFP_KERNEL); |
| 366 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 367 | /* |
| 368 | * This is not really the Guest's fault, but killing it is |
| 369 | * simple for this corner case. |
| 370 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 371 | if (!ptepage) { |
| 372 | kill_guest(cpu, "out of memory allocating pte page"); |
| 373 | return false; |
| 374 | } |
| 375 | |
| 376 | /* We check that the Guest pmd is OK. */ |
| 377 | check_gpmd(cpu, gpmd); |
| 378 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 379 | /* |
| 380 | * And we copy the flags to the shadow PMD entry. The page |
| 381 | * number in the shadow PMD is the page we just allocated. |
| 382 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 383 | native_set_pmd(spmd, __pmd(__pa(ptepage) | pmd_flags(gpmd))); |
| 384 | } |
Rusty Russell | 92b4d8d | 2009-06-12 22:27:08 -0600 | [diff] [blame] | 385 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 386 | /* |
| 387 | * OK, now we look at the lower level in the Guest page table: keep its |
| 388 | * address, because we might update it later. |
| 389 | */ |
Rusty Russell | 92b4d8d | 2009-06-12 22:27:08 -0600 | [diff] [blame] | 390 | gpte_ptr = gpte_addr(cpu, gpmd, vaddr); |
| 391 | #else |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 392 | /* |
| 393 | * OK, now we look at the lower level in the Guest page table: keep its |
| 394 | * address, because we might update it later. |
| 395 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 396 | gpte_ptr = gpte_addr(cpu, gpgd, vaddr); |
Rusty Russell | 92b4d8d | 2009-06-12 22:27:08 -0600 | [diff] [blame] | 397 | #endif |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 398 | |
| 399 | /* Read the actual PTE value. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 400 | gpte = lgread(cpu, gpte_ptr, pte_t); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 401 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 402 | /* If this page isn't in the Guest page tables, we can't page it in. */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 403 | if (!(pte_flags(gpte) & _PAGE_PRESENT)) |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 404 | return false; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 405 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 406 | /* |
| 407 | * Check they're not trying to write to a page the Guest wants |
| 408 | * read-only (bit 2 of errcode == write). |
| 409 | */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 410 | if ((errcode & 2) && !(pte_flags(gpte) & _PAGE_RW)) |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 411 | return false; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 412 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 413 | /* User access to a kernel-only page? (bit 3 == user access) */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 414 | if ((errcode & 4) && !(pte_flags(gpte) & _PAGE_USER)) |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 415 | return false; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 416 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 417 | /* |
| 418 | * Check that the Guest PTE flags are OK, and the page number is below |
| 419 | * the pfn_limit (ie. not mapping the Launcher binary). |
| 420 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 421 | check_gpte(cpu, gpte); |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 422 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 423 | /* Add the _PAGE_ACCESSED and (for a write) _PAGE_DIRTY flag */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 424 | gpte = pte_mkyoung(gpte); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 425 | if (errcode & 2) |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 426 | gpte = pte_mkdirty(gpte); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 427 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 428 | /* Get the pointer to the shadow PTE entry we're going to set. */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 429 | spte = spte_addr(cpu, *spgd, vaddr); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 430 | |
| 431 | /* |
| 432 | * If there was a valid shadow PTE entry here before, we release it. |
| 433 | * This can happen with a write to a previously read-only entry. |
| 434 | */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 435 | release_pte(*spte); |
| 436 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 437 | /* |
| 438 | * If this is a write, we insist that the Guest page is writable (the |
| 439 | * final arg to gpte_to_spte()). |
| 440 | */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 441 | if (pte_dirty(gpte)) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 442 | *spte = gpte_to_spte(cpu, gpte, 1); |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 443 | else |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 444 | /* |
| 445 | * If this is a read, don't set the "writable" bit in the page |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 446 | * table entry, even if the Guest says it's writable. That way |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 447 | * we will come back here when a write does actually occur, so |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 448 | * we can update the Guest's _PAGE_DIRTY flag. |
| 449 | */ |
Matias Zabaljauregui | 90603d1 | 2009-06-12 22:27:06 -0600 | [diff] [blame] | 450 | native_set_pte(spte, gpte_to_spte(cpu, pte_wrprotect(gpte), 0)); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 451 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 452 | /* |
| 453 | * Finally, we write the Guest PTE entry back: we've set the |
| 454 | * _PAGE_ACCESSED and maybe the _PAGE_DIRTY flags. |
| 455 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 456 | lgwrite(cpu, gpte_ptr, pte_t, gpte); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 457 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 458 | /* |
| 459 | * The fault is fixed, the page table is populated, the mapping |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 460 | * manipulated, the result returned and the code complete. A small |
| 461 | * delay and a trace of alliteration are the only indications the Guest |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 462 | * has that a page fault occurred at all. |
| 463 | */ |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 464 | return true; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 465 | } |
| 466 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 467 | /*H:360 |
| 468 | * (ii) Making sure the Guest stack is mapped. |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 469 | * |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 470 | * Remember that direct traps into the Guest need a mapped Guest kernel stack. |
| 471 | * pin_stack_pages() calls us here: we could simply call demand_page(), but as |
| 472 | * we've seen that logic is quite long, and usually the stack pages are already |
| 473 | * mapped, so it's overkill. |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 474 | * |
| 475 | * This is a quick version which answers the question: is this virtual address |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 476 | * mapped by the shadow page tables, and is it writable? |
| 477 | */ |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 478 | static bool page_writable(struct lg_cpu *cpu, unsigned long vaddr) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 479 | { |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 480 | pgd_t *spgd; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 481 | unsigned long flags; |
| 482 | |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 483 | #ifdef CONFIG_X86_PAE |
| 484 | pmd_t *spmd; |
| 485 | #endif |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 486 | /* Look at the current top level entry: is it present? */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 487 | spgd = spgd_addr(cpu, cpu->cpu_pgd, vaddr); |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 488 | if (!(pgd_flags(*spgd) & _PAGE_PRESENT)) |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 489 | return false; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 490 | |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 491 | #ifdef CONFIG_X86_PAE |
| 492 | spmd = spmd_addr(cpu, *spgd, vaddr); |
| 493 | if (!(pmd_flags(*spmd) & _PAGE_PRESENT)) |
| 494 | return false; |
| 495 | #endif |
| 496 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 497 | /* |
| 498 | * Check the flags on the pte entry itself: it must be present and |
| 499 | * writable. |
| 500 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 501 | flags = pte_flags(*(spte_addr(cpu, *spgd, vaddr))); |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 502 | |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 503 | return (flags & (_PAGE_PRESENT|_PAGE_RW)) == (_PAGE_PRESENT|_PAGE_RW); |
| 504 | } |
| 505 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 506 | /* |
| 507 | * So, when pin_stack_pages() asks us to pin a page, we check if it's already |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 508 | * in the page tables, and if not, we call demand_page() with error code 2 |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 509 | * (meaning "write"). |
| 510 | */ |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 511 | void pin_page(struct lg_cpu *cpu, unsigned long vaddr) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 512 | { |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 513 | if (!page_writable(cpu, vaddr) && !demand_page(cpu, vaddr, 2)) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 514 | kill_guest(cpu, "bad stack page %#lx", vaddr); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 515 | } |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 516 | /*:*/ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 517 | |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 518 | #ifdef CONFIG_X86_PAE |
| 519 | static void release_pmd(pmd_t *spmd) |
| 520 | { |
| 521 | /* If the entry's not present, there's nothing to release. */ |
| 522 | if (pmd_flags(*spmd) & _PAGE_PRESENT) { |
| 523 | unsigned int i; |
| 524 | pte_t *ptepage = __va(pmd_pfn(*spmd) << PAGE_SHIFT); |
| 525 | /* For each entry in the page, we might need to release it. */ |
| 526 | for (i = 0; i < PTRS_PER_PTE; i++) |
| 527 | release_pte(ptepage[i]); |
| 528 | /* Now we can free the page of PTEs */ |
| 529 | free_page((long)ptepage); |
| 530 | /* And zero out the PMD entry so we never release it twice. */ |
| 531 | native_set_pmd(spmd, __pmd(0)); |
| 532 | } |
| 533 | } |
| 534 | |
| 535 | static void release_pgd(pgd_t *spgd) |
| 536 | { |
| 537 | /* If the entry's not present, there's nothing to release. */ |
| 538 | if (pgd_flags(*spgd) & _PAGE_PRESENT) { |
| 539 | unsigned int i; |
| 540 | pmd_t *pmdpage = __va(pgd_pfn(*spgd) << PAGE_SHIFT); |
| 541 | |
| 542 | for (i = 0; i < PTRS_PER_PMD; i++) |
| 543 | release_pmd(&pmdpage[i]); |
| 544 | |
| 545 | /* Now we can free the page of PMDs */ |
| 546 | free_page((long)pmdpage); |
| 547 | /* And zero out the PGD entry so we never release it twice. */ |
| 548 | set_pgd(spgd, __pgd(0)); |
| 549 | } |
| 550 | } |
| 551 | |
| 552 | #else /* !CONFIG_X86_PAE */ |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 553 | /*H:450 |
| 554 | * If we chase down the release_pgd() code, the non-PAE version looks like |
| 555 | * this. The PAE version is almost identical, but instead of calling |
| 556 | * release_pte it calls release_pmd(), which looks much like this. |
| 557 | */ |
Matias Zabaljauregui | 90603d1 | 2009-06-12 22:27:06 -0600 | [diff] [blame] | 558 | static void release_pgd(pgd_t *spgd) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 559 | { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 560 | /* If the entry's not present, there's nothing to release. */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 561 | if (pgd_flags(*spgd) & _PAGE_PRESENT) { |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 562 | unsigned int i; |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 563 | /* |
| 564 | * Converting the pfn to find the actual PTE page is easy: turn |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 565 | * the page number into a physical address, then convert to a |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 566 | * virtual address (easy for kernel pages like this one). |
| 567 | */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 568 | pte_t *ptepage = __va(pgd_pfn(*spgd) << PAGE_SHIFT); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 569 | /* For each entry in the page, we might need to release it. */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 570 | for (i = 0; i < PTRS_PER_PTE; i++) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 571 | release_pte(ptepage[i]); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 572 | /* Now we can free the page of PTEs */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 573 | free_page((long)ptepage); |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 574 | /* And zero out the PGD entry so we never release it twice. */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 575 | *spgd = __pgd(0); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 576 | } |
| 577 | } |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 578 | #endif |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 579 | |
| 580 | /*H:445 |
| 581 | * We saw flush_user_mappings() twice: once from the flush_user_mappings() |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 582 | * hypercall and once in new_pgdir() when we re-used a top-level pgdir page. |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 583 | * It simply releases every PTE page from 0 up to the Guest's kernel address. |
| 584 | */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 585 | static void flush_user_mappings(struct lguest *lg, int idx) |
| 586 | { |
| 587 | unsigned int i; |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 588 | /* Release every pgd entry up to the kernel's address. */ |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 589 | for (i = 0; i < pgd_index(lg->kernel_address); i++) |
Matias Zabaljauregui | 90603d1 | 2009-06-12 22:27:06 -0600 | [diff] [blame] | 590 | release_pgd(lg->pgdirs[idx].pgdir + i); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 591 | } |
| 592 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 593 | /*H:440 |
| 594 | * (v) Flushing (throwing away) page tables, |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 595 | * |
| 596 | * The Guest has a hypercall to throw away the page tables: it's used when a |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 597 | * large number of mappings have been changed. |
| 598 | */ |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 599 | void guest_pagetable_flush_user(struct lg_cpu *cpu) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 600 | { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 601 | /* Drop the userspace part of the current page table. */ |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 602 | flush_user_mappings(cpu->lg, cpu->cpu_pgd); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 603 | } |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 604 | /*:*/ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 605 | |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 606 | /* We walk down the guest page tables to get a guest-physical address */ |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 607 | unsigned long guest_pa(struct lg_cpu *cpu, unsigned long vaddr) |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 608 | { |
| 609 | pgd_t gpgd; |
| 610 | pte_t gpte; |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 611 | #ifdef CONFIG_X86_PAE |
| 612 | pmd_t gpmd; |
| 613 | #endif |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 614 | /* First step: get the top-level Guest page table entry. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 615 | gpgd = lgread(cpu, gpgd_addr(cpu, vaddr), pgd_t); |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 616 | /* Toplevel not present? We can't map it in. */ |
Rusty Russell | 6afbdd0 | 2009-03-30 21:55:23 -0600 | [diff] [blame] | 617 | if (!(pgd_flags(gpgd) & _PAGE_PRESENT)) { |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 618 | kill_guest(cpu, "Bad address %#lx", vaddr); |
Rusty Russell | 6afbdd0 | 2009-03-30 21:55:23 -0600 | [diff] [blame] | 619 | return -1UL; |
| 620 | } |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 621 | |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 622 | #ifdef CONFIG_X86_PAE |
| 623 | gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t); |
| 624 | if (!(pmd_flags(gpmd) & _PAGE_PRESENT)) |
| 625 | kill_guest(cpu, "Bad address %#lx", vaddr); |
Rusty Russell | 92b4d8d | 2009-06-12 22:27:08 -0600 | [diff] [blame] | 626 | gpte = lgread(cpu, gpte_addr(cpu, gpmd, vaddr), pte_t); |
| 627 | #else |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 628 | gpte = lgread(cpu, gpte_addr(cpu, gpgd, vaddr), pte_t); |
Rusty Russell | 92b4d8d | 2009-06-12 22:27:08 -0600 | [diff] [blame] | 629 | #endif |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 630 | if (!(pte_flags(gpte) & _PAGE_PRESENT)) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 631 | kill_guest(cpu, "Bad address %#lx", vaddr); |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 632 | |
| 633 | return pte_pfn(gpte) * PAGE_SIZE | (vaddr & ~PAGE_MASK); |
| 634 | } |
| 635 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 636 | /* |
| 637 | * We keep several page tables. This is a simple routine to find the page |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 638 | * table (if any) corresponding to this top-level address the Guest has given |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 639 | * us. |
| 640 | */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 641 | static unsigned int find_pgdir(struct lguest *lg, unsigned long pgtable) |
| 642 | { |
| 643 | unsigned int i; |
| 644 | for (i = 0; i < ARRAY_SIZE(lg->pgdirs); i++) |
Rusty Russell | 4357bd9 | 2008-03-11 09:35:57 -0500 | [diff] [blame] | 645 | if (lg->pgdirs[i].pgdir && lg->pgdirs[i].gpgdir == pgtable) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 646 | break; |
| 647 | return i; |
| 648 | } |
| 649 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 650 | /*H:435 |
| 651 | * And this is us, creating the new page directory. If we really do |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 652 | * allocate a new one (and so the kernel parts are not there), we set |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 653 | * blank_pgdir. |
| 654 | */ |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 655 | static unsigned int new_pgdir(struct lg_cpu *cpu, |
Rusty Russell | ee3db0f | 2007-10-22 11:03:34 +1000 | [diff] [blame] | 656 | unsigned long gpgdir, |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 657 | int *blank_pgdir) |
| 658 | { |
| 659 | unsigned int next; |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 660 | #ifdef CONFIG_X86_PAE |
| 661 | pmd_t *pmd_table; |
| 662 | #endif |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 663 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 664 | /* |
| 665 | * We pick one entry at random to throw out. Choosing the Least |
| 666 | * Recently Used might be better, but this is easy. |
| 667 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 668 | next = random32() % ARRAY_SIZE(cpu->lg->pgdirs); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 669 | /* If it's never been allocated at all before, try now. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 670 | if (!cpu->lg->pgdirs[next].pgdir) { |
| 671 | cpu->lg->pgdirs[next].pgdir = |
| 672 | (pgd_t *)get_zeroed_page(GFP_KERNEL); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 673 | /* If the allocation fails, just keep using the one we have */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 674 | if (!cpu->lg->pgdirs[next].pgdir) |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 675 | next = cpu->cpu_pgd; |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 676 | else { |
| 677 | #ifdef CONFIG_X86_PAE |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 678 | /* |
| 679 | * In PAE mode, allocate a pmd page and populate the |
| 680 | * last pgd entry. |
| 681 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 682 | pmd_table = (pmd_t *)get_zeroed_page(GFP_KERNEL); |
| 683 | if (!pmd_table) { |
| 684 | free_page((long)cpu->lg->pgdirs[next].pgdir); |
| 685 | set_pgd(cpu->lg->pgdirs[next].pgdir, __pgd(0)); |
| 686 | next = cpu->cpu_pgd; |
| 687 | } else { |
| 688 | set_pgd(cpu->lg->pgdirs[next].pgdir + |
| 689 | SWITCHER_PGD_INDEX, |
| 690 | __pgd(__pa(pmd_table) | _PAGE_PRESENT)); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 691 | /* |
| 692 | * This is a blank page, so there are no kernel |
| 693 | * mappings: caller must map the stack! |
| 694 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 695 | *blank_pgdir = 1; |
| 696 | } |
| 697 | #else |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 698 | *blank_pgdir = 1; |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 699 | #endif |
| 700 | } |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 701 | } |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 702 | /* Record which Guest toplevel this shadows. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 703 | cpu->lg->pgdirs[next].gpgdir = gpgdir; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 704 | /* Release all the non-kernel mappings. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 705 | flush_user_mappings(cpu->lg, next); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 706 | |
| 707 | return next; |
| 708 | } |
| 709 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 710 | /*H:430 |
| 711 | * (iv) Switching page tables |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 712 | * |
Matias Zabaljauregui | 90603d1 | 2009-06-12 22:27:06 -0600 | [diff] [blame] | 713 | * Now we've seen all the page table setting and manipulation, let's see |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 714 | * what happens when the Guest changes page tables (ie. changes the top-level |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 715 | * pgdir). This occurs on almost every context switch. |
| 716 | */ |
Glauber de Oliveira Costa | 4665ac8e | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 717 | void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 718 | { |
| 719 | int newpgdir, repin = 0; |
| 720 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 721 | /* Look to see if we have this one already. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 722 | newpgdir = find_pgdir(cpu->lg, pgtable); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 723 | /* |
| 724 | * If not, we allocate or mug an existing one: if it's a fresh one, |
| 725 | * repin gets set to 1. |
| 726 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 727 | if (newpgdir == ARRAY_SIZE(cpu->lg->pgdirs)) |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 728 | newpgdir = new_pgdir(cpu, pgtable, &repin); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 729 | /* Change the current pgd index to the new one. */ |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 730 | cpu->cpu_pgd = newpgdir; |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 731 | /* If it was completely blank, we map in the Guest kernel stack */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 732 | if (repin) |
Glauber de Oliveira Costa | 4665ac8e | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 733 | pin_stack_pages(cpu); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 734 | } |
| 735 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 736 | /*H:470 |
| 737 | * Finally, a routine which throws away everything: all PGD entries in all |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 738 | * the shadow page tables, including the Guest's kernel mappings. This is used |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 739 | * when we destroy the Guest. |
| 740 | */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 741 | static void release_all_pagetables(struct lguest *lg) |
| 742 | { |
| 743 | unsigned int i, j; |
| 744 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 745 | /* Every shadow pagetable this Guest has */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 746 | for (i = 0; i < ARRAY_SIZE(lg->pgdirs); i++) |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 747 | if (lg->pgdirs[i].pgdir) { |
| 748 | #ifdef CONFIG_X86_PAE |
| 749 | pgd_t *spgd; |
| 750 | pmd_t *pmdpage; |
| 751 | unsigned int k; |
| 752 | |
| 753 | /* Get the last pmd page. */ |
| 754 | spgd = lg->pgdirs[i].pgdir + SWITCHER_PGD_INDEX; |
| 755 | pmdpage = __va(pgd_pfn(*spgd) << PAGE_SHIFT); |
| 756 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 757 | /* |
| 758 | * And release the pmd entries of that pmd page, |
| 759 | * except for the switcher pmd. |
| 760 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 761 | for (k = 0; k < SWITCHER_PMD_INDEX; k++) |
| 762 | release_pmd(&pmdpage[k]); |
| 763 | #endif |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 764 | /* Every PGD entry except the Switcher at the top */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 765 | for (j = 0; j < SWITCHER_PGD_INDEX; j++) |
Matias Zabaljauregui | 90603d1 | 2009-06-12 22:27:06 -0600 | [diff] [blame] | 766 | release_pgd(lg->pgdirs[i].pgdir + j); |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 767 | } |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 768 | } |
| 769 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 770 | /* |
| 771 | * We also throw away everything when a Guest tells us it's changed a kernel |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 772 | * mapping. Since kernel mappings are in every page table, it's easiest to |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 773 | * throw them all away. This traps the Guest in amber for a while as |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 774 | * everything faults back in, but it's rare. |
| 775 | */ |
Glauber de Oliveira Costa | 4665ac8e | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 776 | void guest_pagetable_clear_all(struct lg_cpu *cpu) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 777 | { |
Glauber de Oliveira Costa | 4665ac8e | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 778 | release_all_pagetables(cpu->lg); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 779 | /* We need the Guest kernel stack mapped again. */ |
Glauber de Oliveira Costa | 4665ac8e | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 780 | pin_stack_pages(cpu); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 781 | } |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 782 | /*:*/ |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 783 | |
| 784 | /*M:009 |
| 785 | * Since we throw away all mappings when a kernel mapping changes, our |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 786 | * performance sucks for guests using highmem. In fact, a guest with |
| 787 | * PAGE_OFFSET 0xc0000000 (the default) and more than about 700MB of RAM is |
| 788 | * usually slower than a Guest with less memory. |
| 789 | * |
| 790 | * This, of course, cannot be fixed. It would take some kind of... well, I |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 791 | * don't know, but the term "puissant code-fu" comes to mind. |
| 792 | :*/ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 793 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 794 | /*H:420 |
| 795 | * This is the routine which actually sets the page table entry for then |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 796 | * "idx"'th shadow page table. |
| 797 | * |
| 798 | * Normally, we can just throw out the old entry and replace it with 0: if they |
| 799 | * use it demand_page() will put the new entry in. We need to do this anyway: |
| 800 | * The Guest expects _PAGE_ACCESSED to be set on its PTE the first time a page |
| 801 | * is read from, and _PAGE_DIRTY when it's written to. |
| 802 | * |
| 803 | * But Avi Kivity pointed out that most Operating Systems (Linux included) set |
| 804 | * these bits on PTEs immediately anyway. This is done to save the CPU from |
| 805 | * having to update them, but it helps us the same way: if they set |
| 806 | * _PAGE_ACCESSED then we can put a read-only PTE entry in immediately, and if |
| 807 | * they set _PAGE_DIRTY then we can put a writable PTE entry in immediately. |
| 808 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 809 | static void do_set_pte(struct lg_cpu *cpu, int idx, |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 810 | unsigned long vaddr, pte_t gpte) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 811 | { |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 812 | /* Look up the matching shadow page directory entry. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 813 | pgd_t *spgd = spgd_addr(cpu, idx, vaddr); |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 814 | #ifdef CONFIG_X86_PAE |
| 815 | pmd_t *spmd; |
| 816 | #endif |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 817 | |
| 818 | /* If the top level isn't present, there's no entry to update. */ |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 819 | if (pgd_flags(*spgd) & _PAGE_PRESENT) { |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 820 | #ifdef CONFIG_X86_PAE |
| 821 | spmd = spmd_addr(cpu, *spgd, vaddr); |
| 822 | if (pmd_flags(*spmd) & _PAGE_PRESENT) { |
| 823 | #endif |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 824 | /* Otherwise, start by releasing the existing entry. */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 825 | pte_t *spte = spte_addr(cpu, *spgd, vaddr); |
| 826 | release_pte(*spte); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 827 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 828 | /* |
| 829 | * If they're setting this entry as dirty or accessed, |
| 830 | * we might as well put that entry they've given us in |
| 831 | * now. This shaves 10% off a copy-on-write |
| 832 | * micro-benchmark. |
| 833 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 834 | if (pte_flags(gpte) & (_PAGE_DIRTY | _PAGE_ACCESSED)) { |
| 835 | check_gpte(cpu, gpte); |
| 836 | native_set_pte(spte, |
| 837 | gpte_to_spte(cpu, gpte, |
| 838 | pte_flags(gpte) & _PAGE_DIRTY)); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 839 | } else { |
| 840 | /* |
| 841 | * Otherwise kill it and we can demand_page() |
| 842 | * it in later. |
| 843 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 844 | native_set_pte(spte, __pte(0)); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 845 | } |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 846 | #ifdef CONFIG_X86_PAE |
| 847 | } |
| 848 | #endif |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 849 | } |
| 850 | } |
| 851 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 852 | /*H:410 |
| 853 | * Updating a PTE entry is a little trickier. |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 854 | * |
| 855 | * We keep track of several different page tables (the Guest uses one for each |
| 856 | * process, so it makes sense to cache at least a few). Each of these have |
| 857 | * identical kernel parts: ie. every mapping above PAGE_OFFSET is the same for |
| 858 | * all processes. So when the page table above that address changes, we update |
| 859 | * all the page tables, not just the current one. This is rare. |
| 860 | * |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 861 | * The benefit is that when we have to track a new page table, we can keep all |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 862 | * the kernel mappings. This speeds up context switch immensely. |
| 863 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 864 | void guest_set_pte(struct lg_cpu *cpu, |
Rusty Russell | ee3db0f | 2007-10-22 11:03:34 +1000 | [diff] [blame] | 865 | unsigned long gpgdir, unsigned long vaddr, pte_t gpte) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 866 | { |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 867 | /* |
| 868 | * Kernel mappings must be changed on all top levels. Slow, but doesn't |
| 869 | * happen often. |
| 870 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 871 | if (vaddr >= cpu->lg->kernel_address) { |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 872 | unsigned int i; |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 873 | for (i = 0; i < ARRAY_SIZE(cpu->lg->pgdirs); i++) |
| 874 | if (cpu->lg->pgdirs[i].pgdir) |
| 875 | do_set_pte(cpu, i, vaddr, gpte); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 876 | } else { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 877 | /* Is this page table one we have a shadow for? */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 878 | int pgdir = find_pgdir(cpu->lg, gpgdir); |
| 879 | if (pgdir != ARRAY_SIZE(cpu->lg->pgdirs)) |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 880 | /* If so, do the update. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 881 | do_set_pte(cpu, pgdir, vaddr, gpte); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 882 | } |
| 883 | } |
| 884 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 885 | /*H:400 |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 886 | * (iii) Setting up a page table entry when the Guest tells us one has changed. |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 887 | * |
| 888 | * Just like we did in interrupts_and_traps.c, it makes sense for us to deal |
| 889 | * with the other side of page tables while we're here: what happens when the |
| 890 | * Guest asks for a page table to be updated? |
| 891 | * |
| 892 | * We already saw that demand_page() will fill in the shadow page tables when |
| 893 | * needed, so we can simply remove shadow page table entries whenever the Guest |
| 894 | * tells us they've changed. When the Guest tries to use the new entry it will |
| 895 | * fault and demand_page() will fix it up. |
| 896 | * |
| 897 | * So with that in mind here's our code to to update a (top-level) PGD entry: |
| 898 | */ |
Matias Zabaljauregui | ebe0ba8 | 2009-05-30 15:48:08 -0300 | [diff] [blame] | 899 | void guest_set_pgd(struct lguest *lg, unsigned long gpgdir, u32 idx) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 900 | { |
| 901 | int pgdir; |
| 902 | |
| 903 | if (idx >= SWITCHER_PGD_INDEX) |
| 904 | return; |
| 905 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 906 | /* If they're talking about a page table we have a shadow for... */ |
Rusty Russell | ee3db0f | 2007-10-22 11:03:34 +1000 | [diff] [blame] | 907 | pgdir = find_pgdir(lg, gpgdir); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 908 | if (pgdir < ARRAY_SIZE(lg->pgdirs)) |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 909 | /* ... throw it away. */ |
Matias Zabaljauregui | 90603d1 | 2009-06-12 22:27:06 -0600 | [diff] [blame] | 910 | release_pgd(lg->pgdirs[pgdir].pgdir + idx); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 911 | } |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 912 | |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 913 | #ifdef CONFIG_X86_PAE |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 914 | /* For setting a mid-level, we just throw everything away. It's easy. */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 915 | void guest_set_pmd(struct lguest *lg, unsigned long pmdp, u32 idx) |
| 916 | { |
| 917 | guest_pagetable_clear_all(&lg->cpus[0]); |
| 918 | } |
| 919 | #endif |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 920 | |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 921 | /*H:505 |
| 922 | * To get through boot, we construct simple identity page mappings (which |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 923 | * set virtual == physical) and linear mappings which will get the Guest far |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 924 | * enough into the boot to create its own. The linear mapping means we |
| 925 | * simplify the Guest boot, but it makes assumptions about their PAGE_OFFSET, |
| 926 | * as you'll see. |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 927 | * |
| 928 | * We lay them out of the way, just below the initrd (which is why we need to |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 929 | * know its size here). |
| 930 | */ |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 931 | static unsigned long setup_pagetables(struct lguest *lg, |
| 932 | unsigned long mem, |
| 933 | unsigned long initrd_size) |
| 934 | { |
| 935 | pgd_t __user *pgdir; |
| 936 | pte_t __user *linear; |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 937 | unsigned long mem_base = (unsigned long)lg->mem_base; |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 938 | unsigned int mapped_pages, i, linear_pages; |
| 939 | #ifdef CONFIG_X86_PAE |
| 940 | pmd_t __user *pmds; |
| 941 | unsigned int j; |
| 942 | pgd_t pgd; |
| 943 | pmd_t pmd; |
| 944 | #else |
| 945 | unsigned int phys_linear; |
| 946 | #endif |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 947 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 948 | /* |
| 949 | * We have mapped_pages frames to map, so we need linear_pages page |
| 950 | * tables to map them. |
| 951 | */ |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 952 | mapped_pages = mem / PAGE_SIZE; |
| 953 | linear_pages = (mapped_pages + PTRS_PER_PTE - 1) / PTRS_PER_PTE; |
| 954 | |
| 955 | /* We put the toplevel page directory page at the top of memory. */ |
| 956 | pgdir = (pgd_t *)(mem + mem_base - initrd_size - PAGE_SIZE); |
| 957 | |
| 958 | /* Now we use the next linear_pages pages as pte pages */ |
| 959 | linear = (void *)pgdir - linear_pages * PAGE_SIZE; |
| 960 | |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 961 | #ifdef CONFIG_X86_PAE |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 962 | /* |
| 963 | * And the single mid page goes below that. We only use one, but |
| 964 | * that's enough to map 1G, which definitely gets us through boot. |
| 965 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 966 | pmds = (void *)linear - PAGE_SIZE; |
| 967 | #endif |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 968 | /* |
| 969 | * Linear mapping is easy: put every page's address into the |
| 970 | * mapping in order. |
| 971 | */ |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 972 | for (i = 0; i < mapped_pages; i++) { |
| 973 | pte_t pte; |
| 974 | pte = pfn_pte(i, __pgprot(_PAGE_PRESENT|_PAGE_RW|_PAGE_USER)); |
| 975 | if (copy_to_user(&linear[i], &pte, sizeof(pte)) != 0) |
| 976 | return -EFAULT; |
| 977 | } |
| 978 | |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 979 | #ifdef CONFIG_X86_PAE |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 980 | /* |
| 981 | * Make the Guest PMD entries point to the corresponding place in the |
| 982 | * linear mapping (up to one page worth of PMD). |
| 983 | */ |
Rusty Russell | 92b4d8d | 2009-06-12 22:27:08 -0600 | [diff] [blame] | 984 | for (i = j = 0; i < mapped_pages && j < PTRS_PER_PMD; |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 985 | i += PTRS_PER_PTE, j++) { |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 986 | /* FIXME: native_set_pmd is overkill here. */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 987 | native_set_pmd(&pmd, __pmd(((unsigned long)(linear + i) |
| 988 | - mem_base) | _PAGE_PRESENT | _PAGE_RW | _PAGE_USER)); |
| 989 | |
| 990 | if (copy_to_user(&pmds[j], &pmd, sizeof(pmd)) != 0) |
| 991 | return -EFAULT; |
| 992 | } |
| 993 | |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 994 | /* One PGD entry, pointing to that PMD page. */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 995 | set_pgd(&pgd, __pgd(((u32)pmds - mem_base) | _PAGE_PRESENT)); |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 996 | /* Copy it in as the first PGD entry (ie. addresses 0-1G). */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 997 | if (copy_to_user(&pgdir[0], &pgd, sizeof(pgd)) != 0) |
| 998 | return -EFAULT; |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 999 | /* |
| 1000 | * And the third PGD entry (ie. addresses 3G-4G). |
| 1001 | * |
| 1002 | * FIXME: This assumes that PAGE_OFFSET for the Guest is 0xC0000000. |
| 1003 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 1004 | if (copy_to_user(&pgdir[3], &pgd, sizeof(pgd)) != 0) |
| 1005 | return -EFAULT; |
| 1006 | #else |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1007 | /* |
| 1008 | * The top level points to the linear page table pages above. |
| 1009 | * We setup the identity and linear mappings here. |
| 1010 | */ |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 1011 | phys_linear = (unsigned long)linear - mem_base; |
| 1012 | for (i = 0; i < mapped_pages; i += PTRS_PER_PTE) { |
| 1013 | pgd_t pgd; |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1014 | /* |
| 1015 | * Create a PGD entry which points to the right part of the |
| 1016 | * linear PTE pages. |
| 1017 | */ |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 1018 | pgd = __pgd((phys_linear + i * sizeof(pte_t)) | |
| 1019 | (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER)); |
| 1020 | |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1021 | /* |
| 1022 | * Copy it into the PGD page at 0 and PAGE_OFFSET. |
| 1023 | */ |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 1024 | if (copy_to_user(&pgdir[i / PTRS_PER_PTE], &pgd, sizeof(pgd)) |
| 1025 | || copy_to_user(&pgdir[pgd_index(PAGE_OFFSET) |
| 1026 | + i / PTRS_PER_PTE], |
| 1027 | &pgd, sizeof(pgd))) |
| 1028 | return -EFAULT; |
| 1029 | } |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 1030 | #endif |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 1031 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1032 | /* |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1033 | * We return the top level (guest-physical) address: we remember where |
| 1034 | * this is to write it into lguest_data when the Guest initializes. |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1035 | */ |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 1036 | return (unsigned long)pgdir - mem_base; |
| 1037 | } |
| 1038 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1039 | /*H:500 |
| 1040 | * (vii) Setting up the page tables initially. |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 1041 | * |
| 1042 | * When a Guest is first created, the Launcher tells us where the toplevel of |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1043 | * its first page table is. We set some things up here: |
| 1044 | */ |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 1045 | int init_guest_pagetable(struct lguest *lg) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1046 | { |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 1047 | u64 mem; |
| 1048 | u32 initrd_size; |
| 1049 | struct boot_params __user *boot = (struct boot_params *)lg->mem_base; |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 1050 | #ifdef CONFIG_X86_PAE |
| 1051 | pgd_t *pgd; |
| 1052 | pmd_t *pmd_table; |
| 1053 | #endif |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1054 | /* |
| 1055 | * Get the Guest memory size and the ramdisk size from the boot header |
| 1056 | * located at lg->mem_base (Guest address 0). |
| 1057 | */ |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 1058 | if (copy_from_user(&mem, &boot->e820_map[0].size, sizeof(mem)) |
| 1059 | || get_user(initrd_size, &boot->hdr.ramdisk_size)) |
| 1060 | return -EFAULT; |
| 1061 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1062 | /* |
| 1063 | * We start on the first shadow page table, and give it a blank PGD |
| 1064 | * page. |
| 1065 | */ |
Matias Zabaljauregui | 58a2456 | 2008-09-29 01:40:07 -0300 | [diff] [blame] | 1066 | lg->pgdirs[0].gpgdir = setup_pagetables(lg, mem, initrd_size); |
| 1067 | if (IS_ERR_VALUE(lg->pgdirs[0].gpgdir)) |
| 1068 | return lg->pgdirs[0].gpgdir; |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 1069 | lg->pgdirs[0].pgdir = (pgd_t *)get_zeroed_page(GFP_KERNEL); |
| 1070 | if (!lg->pgdirs[0].pgdir) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1071 | return -ENOMEM; |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1072 | |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 1073 | #ifdef CONFIG_X86_PAE |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1074 | /* For PAE, we also create the initial mid-level. */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 1075 | pgd = lg->pgdirs[0].pgdir; |
| 1076 | pmd_table = (pmd_t *) get_zeroed_page(GFP_KERNEL); |
| 1077 | if (!pmd_table) |
| 1078 | return -ENOMEM; |
| 1079 | |
| 1080 | set_pgd(pgd + SWITCHER_PGD_INDEX, |
| 1081 | __pgd(__pa(pmd_table) | _PAGE_PRESENT)); |
| 1082 | #endif |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1083 | |
| 1084 | /* This is the current page table. */ |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 1085 | lg->cpus[0].cpu_pgd = 0; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1086 | return 0; |
| 1087 | } |
| 1088 | |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1089 | /*H:508 When the Guest calls LHCALL_LGUEST_INIT we do more setup. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 1090 | void page_table_guest_data_init(struct lg_cpu *cpu) |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 1091 | { |
| 1092 | /* We get the kernel address: above this is all kernel memory. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 1093 | if (get_user(cpu->lg->kernel_address, |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 1094 | &cpu->lg->lguest_data->kernel_address) |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1095 | /* |
| 1096 | * We tell the Guest that it can't use the top 2 or 4 MB |
| 1097 | * of virtual addresses used by the Switcher. |
| 1098 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 1099 | || put_user(RESERVE_MEM * 1024 * 1024, |
| 1100 | &cpu->lg->lguest_data->reserve_mem) |
| 1101 | || put_user(cpu->lg->pgdirs[0].gpgdir, |
| 1102 | &cpu->lg->lguest_data->pgdir)) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 1103 | kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 1104 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1105 | /* |
| 1106 | * In flush_user_mappings() we loop from 0 to |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 1107 | * "pgd_index(lg->kernel_address)". This assumes it won't hit the |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1108 | * Switcher mappings, so check that now. |
| 1109 | */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 1110 | #ifdef CONFIG_X86_PAE |
| 1111 | if (pgd_index(cpu->lg->kernel_address) == SWITCHER_PGD_INDEX && |
| 1112 | pmd_index(cpu->lg->kernel_address) == SWITCHER_PMD_INDEX) |
| 1113 | #else |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 1114 | if (pgd_index(cpu->lg->kernel_address) >= SWITCHER_PGD_INDEX) |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 1115 | #endif |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 1116 | kill_guest(cpu, "bad kernel address %#lx", |
| 1117 | cpu->lg->kernel_address); |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 1118 | } |
| 1119 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 1120 | /* When a Guest dies, our cleanup is fairly simple. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1121 | void free_guest_pagetable(struct lguest *lg) |
| 1122 | { |
| 1123 | unsigned int i; |
| 1124 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 1125 | /* Throw away all page table pages. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1126 | release_all_pagetables(lg); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 1127 | /* Now free the top levels: free_page() can handle 0 just fine. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1128 | for (i = 0; i < ARRAY_SIZE(lg->pgdirs); i++) |
| 1129 | free_page((long)lg->pgdirs[i].pgdir); |
| 1130 | } |
| 1131 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1132 | /*H:480 |
| 1133 | * (vi) Mapping the Switcher when the Guest is about to run. |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 1134 | * |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1135 | * The Switcher and the two pages for this CPU need to be visible in the |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 1136 | * Guest (and not the pages for other CPUs). We have the appropriate PTE pages |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1137 | * for each CPU already set up, we just need to hook them in now we know which |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1138 | * Guest is about to run on this CPU. |
| 1139 | */ |
Glauber de Oliveira Costa | 0c78441 | 2008-01-07 11:05:30 -0200 | [diff] [blame] | 1140 | void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1141 | { |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 1142 | pte_t *switcher_pte_page = __get_cpu_var(switcher_pte_pages); |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 1143 | pte_t regs_pte; |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 1144 | unsigned long pfn; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1145 | |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 1146 | #ifdef CONFIG_X86_PAE |
| 1147 | pmd_t switcher_pmd; |
| 1148 | pmd_t *pmd_table; |
| 1149 | |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1150 | /* FIXME: native_set_pmd is overkill here. */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 1151 | native_set_pmd(&switcher_pmd, pfn_pmd(__pa(switcher_pte_page) >> |
| 1152 | PAGE_SHIFT, PAGE_KERNEL_EXEC)); |
| 1153 | |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1154 | /* Figure out where the pmd page is, by reading the PGD, and converting |
| 1155 | * it to a virtual address. */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 1156 | pmd_table = __va(pgd_pfn(cpu->lg-> |
| 1157 | pgdirs[cpu->cpu_pgd].pgdir[SWITCHER_PGD_INDEX]) |
| 1158 | << PAGE_SHIFT); |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1159 | /* Now write it into the shadow page table. */ |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 1160 | native_set_pmd(&pmd_table[SWITCHER_PMD_INDEX], switcher_pmd); |
| 1161 | #else |
| 1162 | pgd_t switcher_pgd; |
| 1163 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1164 | /* |
| 1165 | * Make the last PGD entry for this Guest point to the Switcher's PTE |
| 1166 | * page for this CPU (with appropriate flags). |
| 1167 | */ |
Matias Zabaljauregui | ed1dc77 | 2009-05-30 15:35:49 -0300 | [diff] [blame] | 1168 | switcher_pgd = __pgd(__pa(switcher_pte_page) | __PAGE_KERNEL_EXEC); |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 1169 | |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 1170 | cpu->lg->pgdirs[cpu->cpu_pgd].pgdir[SWITCHER_PGD_INDEX] = switcher_pgd; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1171 | |
Matias Zabaljauregui | acdd0b6 | 2009-06-12 22:27:07 -0600 | [diff] [blame] | 1172 | #endif |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1173 | /* |
| 1174 | * We also change the Switcher PTE page. When we're running the Guest, |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 1175 | * we want the Guest's "regs" page to appear where the first Switcher |
| 1176 | * page for this CPU is. This is an optimization: when the Switcher |
| 1177 | * saves the Guest registers, it saves them into the first page of this |
| 1178 | * CPU's "struct lguest_pages": if we make sure the Guest's register |
| 1179 | * page is already mapped there, we don't have to copy them out |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1180 | * again. |
| 1181 | */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 1182 | pfn = __pa(cpu->regs_page) >> PAGE_SHIFT; |
Matias Zabaljauregui | 90603d1 | 2009-06-12 22:27:06 -0600 | [diff] [blame] | 1183 | native_set_pte(®s_pte, pfn_pte(pfn, PAGE_KERNEL)); |
| 1184 | native_set_pte(&switcher_pte_page[pte_index((unsigned long)pages)], |
| 1185 | regs_pte); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1186 | } |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 1187 | /*:*/ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1188 | |
| 1189 | static void free_switcher_pte_pages(void) |
| 1190 | { |
| 1191 | unsigned int i; |
| 1192 | |
| 1193 | for_each_possible_cpu(i) |
| 1194 | free_page((long)switcher_pte_page(i)); |
| 1195 | } |
| 1196 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1197 | /*H:520 |
| 1198 | * Setting up the Switcher PTE page for given CPU is fairly easy, given |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 1199 | * the CPU number and the "struct page"s for the Switcher code itself. |
| 1200 | * |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1201 | * Currently the Switcher is less than a page long, so "pages" is always 1. |
| 1202 | */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1203 | static __init void populate_switcher_pte_page(unsigned int cpu, |
| 1204 | struct page *switcher_page[], |
| 1205 | unsigned int pages) |
| 1206 | { |
| 1207 | unsigned int i; |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 1208 | pte_t *pte = switcher_pte_page(cpu); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1209 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 1210 | /* The first entries are easy: they map the Switcher code. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1211 | for (i = 0; i < pages; i++) { |
Matias Zabaljauregui | 90603d1 | 2009-06-12 22:27:06 -0600 | [diff] [blame] | 1212 | native_set_pte(&pte[i], mk_pte(switcher_page[i], |
| 1213 | __pgprot(_PAGE_PRESENT|_PAGE_ACCESSED))); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1214 | } |
| 1215 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 1216 | /* The only other thing we map is this CPU's pair of pages. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1217 | i = pages + cpu*2; |
| 1218 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 1219 | /* First page (Guest registers) is writable from the Guest */ |
Matias Zabaljauregui | 90603d1 | 2009-06-12 22:27:06 -0600 | [diff] [blame] | 1220 | native_set_pte(&pte[i], pfn_pte(page_to_pfn(switcher_page[i]), |
| 1221 | __pgprot(_PAGE_PRESENT|_PAGE_ACCESSED|_PAGE_RW))); |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 1222 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1223 | /* |
| 1224 | * The second page contains the "struct lguest_ro_state", and is |
| 1225 | * read-only. |
| 1226 | */ |
Matias Zabaljauregui | 90603d1 | 2009-06-12 22:27:06 -0600 | [diff] [blame] | 1227 | native_set_pte(&pte[i+1], pfn_pte(page_to_pfn(switcher_page[i+1]), |
| 1228 | __pgprot(_PAGE_PRESENT|_PAGE_ACCESSED))); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1229 | } |
| 1230 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1231 | /* |
| 1232 | * We've made it through the page table code. Perhaps our tired brains are |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1233 | * still processing the details, or perhaps we're simply glad it's over. |
| 1234 | * |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 1235 | * If nothing else, note that all this complexity in juggling shadow page tables |
| 1236 | * in sync with the Guest's page tables is for one reason: for most Guests this |
| 1237 | * page table dance determines how bad performance will be. This is why Xen |
| 1238 | * uses exotic direct Guest pagetable manipulation, and why both Intel and AMD |
| 1239 | * have implemented shadow page table support directly into hardware. |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1240 | * |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1241 | * There is just one file remaining in the Host. |
| 1242 | */ |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 1243 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1244 | /*H:510 |
| 1245 | * At boot or module load time, init_pagetables() allocates and populates |
| 1246 | * the Switcher PTE page for each CPU. |
| 1247 | */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1248 | __init int init_pagetables(struct page **switcher_page, unsigned int pages) |
| 1249 | { |
| 1250 | unsigned int i; |
| 1251 | |
| 1252 | for_each_possible_cpu(i) { |
Matias Zabaljauregui | df29f43 | 2007-10-22 11:03:33 +1000 | [diff] [blame] | 1253 | switcher_pte_page(i) = (pte_t *)get_zeroed_page(GFP_KERNEL); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1254 | if (!switcher_pte_page(i)) { |
| 1255 | free_switcher_pte_pages(); |
| 1256 | return -ENOMEM; |
| 1257 | } |
| 1258 | populate_switcher_pte_page(i, switcher_page, pages); |
| 1259 | } |
| 1260 | return 0; |
| 1261 | } |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 1262 | /*:*/ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1263 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 1264 | /* Cleaning up simply involves freeing the PTE page for each CPU. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 1265 | void free_pagetables(void) |
| 1266 | { |
| 1267 | free_switcher_pte_pages(); |
| 1268 | } |