arch/x86_64/kernel/machine_kexec.c

   1 /*
   2  * machine_kexec.c - 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 <asm/pgtable.h>
  14 #include <asm/tlbflush.h>
  15 #include <asm/mmu_context.h>
  16 #include <asm/io.h>
  17
  18 #define PAGE_ALIGNED __attribute__ ((__aligned__(PAGE_SIZE)))
  19 static u64 kexec_pgd[512] PAGE_ALIGNED;
  20 static u64 kexec_pud0[512] PAGE_ALIGNED;
  21 static u64 kexec_pmd0[512] PAGE_ALIGNED;
  22 static u64 kexec_pte0[512] PAGE_ALIGNED;
  23 static u64 kexec_pud1[512] PAGE_ALIGNED;
  24 static u64 kexec_pmd1[512] PAGE_ALIGNED;
  25 static u64 kexec_pte1[512] PAGE_ALIGNED;
  26
  27 static void init_level2_page(pmd_t *level2p, unsigned long addr)
  28 {
  29         unsigned long end_addr;
  30
  31         addr &= PAGE_MASK;
  32         end_addr = addr + PUD_SIZE;
  33         while (addr < end_addr) {
  34                 set_pmd(level2p++, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
  35                 addr += PMD_SIZE;
  36         }
  37 }
  38
  39 static int init_level3_page(struct kimage *image, pud_t *level3p,
  40                                 unsigned long addr, unsigned long last_addr)
  41 {
  42         unsigned long end_addr;
  43         int result;
  44
  45         result = 0;
  46         addr &= PAGE_MASK;
  47         end_addr = addr + PGDIR_SIZE;
  48         while ((addr < last_addr) && (addr < end_addr)) {
  49                 struct page *page;
  50                 pmd_t *level2p;
  51
  52                 page = kimage_alloc_control_pages(image, 0);
  53                 if (!page) {
  54                         result = -ENOMEM;
  55                         goto out;
  56                 }
  57                 level2p = (pmd_t *)page_address(page);
  58                 init_level2_page(level2p, addr);
  59                 set_pud(level3p++, __pud(__pa(level2p) | _KERNPG_TABLE));
  60                 addr += PUD_SIZE;
  61         }
  62         /* clear the unused entries */
  63         while (addr < end_addr) {
  64                 pud_clear(level3p++);
  65                 addr += PUD_SIZE;
  66         }
  67 out:
  68         return result;
  69 }
  70
  71
  72 static int init_level4_page(struct kimage *image, pgd_t *level4p,
  73                                 unsigned long addr, unsigned long last_addr)
  74 {
  75         unsigned long end_addr;
  76         int result;
  77
  78         result = 0;
  79         addr &= PAGE_MASK;
  80         end_addr = addr + (PTRS_PER_PGD * PGDIR_SIZE);
  81         while ((addr < last_addr) && (addr < end_addr)) {
  82                 struct page *page;
  83                 pud_t *level3p;
  84
  85                 page = kimage_alloc_control_pages(image, 0);
  86                 if (!page) {
  87                         result = -ENOMEM;
  88                         goto out;
  89                 }
  90                 level3p = (pud_t *)page_address(page);
  91                 result = init_level3_page(image, level3p, addr, last_addr);
  92                 if (result) {
  93                         goto out;
  94                 }
  95                 set_pgd(level4p++, __pgd(__pa(level3p) | _KERNPG_TABLE));
  96                 addr += PGDIR_SIZE;
  97         }
  98         /* clear the unused entries */
  99         while (addr < end_addr) {
 100                 pgd_clear(level4p++);
 101                 addr += PGDIR_SIZE;
 102         }
 103 out:
 104         return result;
 105 }
 106
 107
 108 static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
 109 {
 110         pgd_t *level4p;
 111         level4p = (pgd_t *)__va(start_pgtable);
 112         return init_level4_page(image, level4p, 0, end_pfn << PAGE_SHIFT);
 113 }
 114
 115 static void set_idt(void *newidt, u16 limit)
 116 {
 117         struct desc_ptr curidt;
 118
 119         /* x86-64 supports unaliged loads & stores */
 120         curidt.size    = limit;
 121         curidt.address = (unsigned long)newidt;
 122
 123         __asm__ __volatile__ (
 124                 "lidtq %0\n"
 125                 : : "m" (curidt)
 126                 );
 127 };
 128
 129
 130 static void set_gdt(void *newgdt, u16 limit)
 131 {
 132         struct desc_ptr curgdt;
 133
 134         /* x86-64 supports unaligned loads & stores */
 135         curgdt.size    = limit;
 136         curgdt.address = (unsigned long)newgdt;
 137
 138         __asm__ __volatile__ (
 139                 "lgdtq %0\n"
 140                 : : "m" (curgdt)
 141                 );
 142 };
 143
 144 static void load_segments(void)
 145 {
 146         __asm__ __volatile__ (
 147                 "\tmovl %0,%%ds\n"
 148                 "\tmovl %0,%%es\n"
 149                 "\tmovl %0,%%ss\n"
 150                 "\tmovl %0,%%fs\n"
 151                 "\tmovl %0,%%gs\n"
 152                 : : "a" (__KERNEL_DS) : "memory"
 153                 );
 154 }
 155
 156 int machine_kexec_prepare(struct kimage *image)
 157 {
 158         unsigned long start_pgtable;
 159         int result;
 160
 161         /* Calculate the offsets */
 162         start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
 163
 164         /* Setup the identity mapped 64bit page table */
 165         result = init_pgtable(image, start_pgtable);
 166         if (result)
 167                 return result;
 168
 169         return 0;
 170 }
 171
 172 void machine_kexec_cleanup(struct kimage *image)
 173 {
 174         return;
 175 }
 176
 177 /*
 178  * Do not allocate memory (or fail in any way) in machine_kexec().
 179  * We are past the point of no return, committed to rebooting now.
 180  */
 181 NORET_TYPE void machine_kexec(struct kimage *image)
 182 {
 183         unsigned long page_list[PAGES_NR];
 184         void *control_page;
 185
 186         /* Interrupts aren't acceptable while we reboot */
 187         local_irq_disable();
 188
 189         control_page = page_address(image->control_code_page) + PAGE_SIZE;
 190         memcpy(control_page, relocate_kernel, PAGE_SIZE);
 191
 192         page_list[PA_CONTROL_PAGE] = __pa(control_page);
 193         page_list[VA_CONTROL_PAGE] = (unsigned long)relocate_kernel;
 194         page_list[PA_PGD] = __pa(kexec_pgd);
 195         page_list[VA_PGD] = (unsigned long)kexec_pgd;
 196         page_list[PA_PUD_0] = __pa(kexec_pud0);
 197         page_list[VA_PUD_0] = (unsigned long)kexec_pud0;
 198         page_list[PA_PMD_0] = __pa(kexec_pmd0);
 199         page_list[VA_PMD_0] = (unsigned long)kexec_pmd0;
 200         page_list[PA_PTE_0] = __pa(kexec_pte0);
 201         page_list[VA_PTE_0] = (unsigned long)kexec_pte0;
 202         page_list[PA_PUD_1] = __pa(kexec_pud1);
 203         page_list[VA_PUD_1] = (unsigned long)kexec_pud1;
 204         page_list[PA_PMD_1] = __pa(kexec_pmd1);
 205         page_list[VA_PMD_1] = (unsigned long)kexec_pmd1;
 206         page_list[PA_PTE_1] = __pa(kexec_pte1);
 207         page_list[VA_PTE_1] = (unsigned long)kexec_pte1;
 208
 209         page_list[PA_TABLE_PAGE] =
 210           (unsigned long)__pa(page_address(image->control_code_page));
 211
 212         /* The segment registers are funny things, they have both a
 213          * visible and an invisible part.  Whenever the visible part is
 214          * set to a specific selector, the invisible part is loaded
 215          * with from a table in memory.  At no other time is the
 216          * descriptor table in memory accessed.
 217          *
 218          * I take advantage of this here by force loading the
 219          * segments, before I zap the gdt with an invalid value.
 220          */
 221         load_segments();
 222         /* The gdt & idt are now invalid.
 223          * If you want to load them you must set up your own idt & gdt.
 224          */
 225         set_gdt(phys_to_virt(0),0);
 226         set_idt(phys_to_virt(0),0);
 227
 228         /* now call it */
 229         relocate_kernel((unsigned long)image->head, (unsigned long)page_list,
 230                         image->start);
 231 }
 232
 233 /* crashkernel=size@addr specifies the location to reserve for
 234  * a crash kernel.  By reserving this memory we guarantee
 235  * that linux never set's it up as a DMA target.
 236  * Useful for holding code to do something appropriate
 237  * after a kernel panic.
 238  */
 239 static int __init setup_crashkernel(char *arg)
 240 {
 241         unsigned long size, base;
 242         char *p;
 243         if (!arg)
 244                 return -EINVAL;
 245         size = memparse(arg, &p);
 246         if (arg == p)
 247                 return -EINVAL;
 248         if (*p == '@') {
 249                 base = memparse(p+1, &p);
 250                 /* FIXME: Do I want a sanity check to validate the
 251                  * memory range?  Yes you do, but it's too early for
 252                  * e820 -AK */
 253                 crashk_res.start = base;
 254                 crashk_res.end   = base + size - 1;
 255         }
 256         return 0;
 257 }
 258 early_param("crashkernel", setup_crashkernel);
 259