arch/x86/kernel/machine_kexec_64.c

   1 /*
   2  * handle transition of Linux booting another kernel
   3  * Copyright (C) 2002-2005 Eric Biederman  <ebiederm@xmission.com>
   4  *
   5  * This source code is licensed under the GNU General Public License,
   6  * Version 2.  See the file COPYING for more details.
   7  */
   8
   9 #include <linux/mm.h>
  10 #include <linux/kexec.h>
  11 #include <linux/string.h>
  12 #include <linux/reboot.h>
  13 #include <linux/numa.h>
  14 #include <linux/ftrace.h>
  15 #include <linux/io.h>
  16
  17 #include <asm/pgtable.h>
  18 #include <asm/tlbflush.h>
  19 #include <asm/mmu_context.h>
  20
  21 static void init_level2_page(pmd_t *level2p, unsigned long addr)
  22 {
  23         unsigned long end_addr;
  24
  25         addr &= PAGE_MASK;
  26         end_addr = addr + PUD_SIZE;
  27         while (addr < end_addr) {
  28                 set_pmd(level2p++, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
  29                 addr += PMD_SIZE;
  30         }
  31 }
  32
  33 static int init_level3_page(struct kimage *image, pud_t *level3p,
  34                                 unsigned long addr, unsigned long last_addr)
  35 {
  36         unsigned long end_addr;
  37         int result;
  38
  39         result = 0;
  40         addr &= PAGE_MASK;
  41         end_addr = addr + PGDIR_SIZE;
  42         while ((addr < last_addr) && (addr < end_addr)) {
  43                 struct page *page;
  44                 pmd_t *level2p;
  45
  46                 page = kimage_alloc_control_pages(image, 0);
  47                 if (!page) {
  48                         result = -ENOMEM;
  49                         goto out;
  50                 }
  51                 level2p = (pmd_t *)page_address(page);
  52                 init_level2_page(level2p, addr);
  53                 set_pud(level3p++, __pud(__pa(level2p) | _KERNPG_TABLE));
  54                 addr += PUD_SIZE;
  55         }
  56         /* clear the unused entries */
  57         while (addr < end_addr) {
  58                 pud_clear(level3p++);
  59                 addr += PUD_SIZE;
  60         }
  61 out:
  62         return result;
  63 }
  64
  65
  66 static int init_level4_page(struct kimage *image, pgd_t *level4p,
  67                                 unsigned long addr, unsigned long last_addr)
  68 {
  69         unsigned long end_addr;
  70         int result;
  71
  72         result = 0;
  73         addr &= PAGE_MASK;
  74         end_addr = addr + (PTRS_PER_PGD * PGDIR_SIZE);
  75         while ((addr < last_addr) && (addr < end_addr)) {
  76                 struct page *page;
  77                 pud_t *level3p;
  78
  79                 page = kimage_alloc_control_pages(image, 0);
  80                 if (!page) {
  81                         result = -ENOMEM;
  82                         goto out;
  83                 }
  84                 level3p = (pud_t *)page_address(page);
  85                 result = init_level3_page(image, level3p, addr, last_addr);
  86                 if (result)
  87                         goto out;
  88                 set_pgd(level4p++, __pgd(__pa(level3p) | _KERNPG_TABLE));
  89                 addr += PGDIR_SIZE;
  90         }
  91         /* clear the unused entries */
  92         while (addr < end_addr) {
  93                 pgd_clear(level4p++);
  94                 addr += PGDIR_SIZE;
  95         }
  96 out:
  97         return result;
  98 }
  99
 100 static void free_transition_pgtable(struct kimage *image)
 101 {
 102         free_page((unsigned long)image->arch.pud);
 103         free_page((unsigned long)image->arch.pmd);
 104         free_page((unsigned long)image->arch.pte);
 105 }
 106
 107 static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
 108 {
 109         pud_t *pud;
 110         pmd_t *pmd;
 111         pte_t *pte;
 112         unsigned long vaddr, paddr;
 113         int result = -ENOMEM;
 114
 115         vaddr = (unsigned long)relocate_kernel;
 116         paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE);
 117         pgd += pgd_index(vaddr);
 118         if (!pgd_present(*pgd)) {
 119                 pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
 120                 if (!pud)
 121                         goto err;
 122                 image->arch.pud = pud;
 123                 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE));
 124         }
 125         pud = pud_offset(pgd, vaddr);
 126         if (!pud_present(*pud)) {
 127                 pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
 128                 if (!pmd)
 129                         goto err;
 130                 image->arch.pmd = pmd;
 131                 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
 132         }
 133         pmd = pmd_offset(pud, vaddr);
 134         if (!pmd_present(*pmd)) {
 135                 pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
 136                 if (!pte)
 137                         goto err;
 138                 image->arch.pte = pte;
 139                 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
 140         }
 141         pte = pte_offset_kernel(pmd, vaddr);
 142         set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
 143         return 0;
 144 err:
 145         free_transition_pgtable(image);
 146         return result;
 147 }
 148
 149
 150 static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
 151 {
 152         pgd_t *level4p;
 153         int result;
 154         level4p = (pgd_t *)__va(start_pgtable);
 155         result = init_level4_page(image, level4p, 0, max_pfn << PAGE_SHIFT);
 156         if (result)
 157                 return result;
 158         return init_transition_pgtable(image, level4p);
 159 }
 160
 161 static void set_idt(void *newidt, u16 limit)
 162 {
 163         struct desc_ptr curidt;
 164
 165         /* x86-64 supports unaliged loads & stores */
 166         curidt.size    = limit;
 167         curidt.address = (unsigned long)newidt;
 168
 169         __asm__ __volatile__ (
 170                 "lidtq %0\n"
 171                 : : "m" (curidt)
 172                 );
 173 };
 174
 175
 176 static void set_gdt(void *newgdt, u16 limit)
 177 {
 178         struct desc_ptr curgdt;
 179
 180         /* x86-64 supports unaligned loads & stores */
 181         curgdt.size    = limit;
 182         curgdt.address = (unsigned long)newgdt;
 183
 184         __asm__ __volatile__ (
 185                 "lgdtq %0\n"
 186                 : : "m" (curgdt)
 187                 );
 188 };
 189
 190 static void load_segments(void)
 191 {
 192         __asm__ __volatile__ (
 193                 "\tmovl %0,%%ds\n"
 194                 "\tmovl %0,%%es\n"
 195                 "\tmovl %0,%%ss\n"
 196                 "\tmovl %0,%%fs\n"
 197                 "\tmovl %0,%%gs\n"
 198                 : : "a" (__KERNEL_DS) : "memory"
 199                 );
 200 }
 201
 202 int machine_kexec_prepare(struct kimage *image)
 203 {
 204         unsigned long start_pgtable;
 205         int result;
 206
 207         /* Calculate the offsets */
 208         start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
 209
 210         /* Setup the identity mapped 64bit page table */
 211         result = init_pgtable(image, start_pgtable);
 212         if (result)
 213                 return result;
 214
 215         return 0;
 216 }
 217
 218 void machine_kexec_cleanup(struct kimage *image)
 219 {
 220         free_transition_pgtable(image);
 221 }
 222
 223 /*
 224  * Do not allocate memory (or fail in any way) in machine_kexec().
 225  * We are past the point of no return, committed to rebooting now.
 226  */
 227 void machine_kexec(struct kimage *image)
 228 {
 229         unsigned long page_list[PAGES_NR];
 230         void *control_page;
 231
 232         tracer_disable();
 233
 234         /* Interrupts aren't acceptable while we reboot */
 235         local_irq_disable();
 236
 237         control_page = page_address(image->control_code_page) + PAGE_SIZE;
 238         memcpy(control_page, relocate_kernel, PAGE_SIZE);
 239
 240         page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
 241         page_list[PA_TABLE_PAGE] =
 242           (unsigned long)__pa(page_address(image->control_code_page));
 243
 244         /*
 245          * The segment registers are funny things, they have both a
 246          * visible and an invisible part.  Whenever the visible part is
 247          * set to a specific selector, the invisible part is loaded
 248          * with from a table in memory.  At no other time is the
 249          * descriptor table in memory accessed.
 250          *
 251          * I take advantage of this here by force loading the
 252          * segments, before I zap the gdt with an invalid value.
 253          */
 254         load_segments();
 255         /*
 256          * The gdt & idt are now invalid.
 257          * If you want to load them you must set up your own idt & gdt.
 258          */
 259         set_gdt(phys_to_virt(0), 0);
 260         set_idt(phys_to_virt(0), 0);
 261
 262         /* now call it */
 263         relocate_kernel((unsigned long)image->head, (unsigned long)page_list,
 264                         image->start);
 265 }
 266
 267 void arch_crash_save_vmcoreinfo(void)
 268 {
 269         VMCOREINFO_SYMBOL(phys_base);
 270         VMCOREINFO_SYMBOL(init_level4_pgt);
 271
 272 #ifdef CONFIG_NUMA
 273         VMCOREINFO_SYMBOL(node_data);
 274         VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
 275 #endif
 276 }
 277