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