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/delay.h>
  12 #include <linux/string.h>
  13 #include <linux/reboot.h>
  14 #include <asm/pda.h>
  15 #include <asm/pgtable.h>
  16 #include <asm/pgalloc.h>
  17 #include <asm/tlbflush.h>
  18 #include <asm/mmu_context.h>
  19 #include <asm/io.h>
  20 #include <asm/apic.h>
  21 #include <asm/cpufeature.h>
  22 #include <asm/hw_irq.h>
  23
  24 #define LEVEL0_SIZE (1UL << 12UL)
  25 #define LEVEL1_SIZE (1UL << 21UL)
  26 #define LEVEL2_SIZE (1UL << 30UL)
  27 #define LEVEL3_SIZE (1UL << 39UL)
  28 #define LEVEL4_SIZE (1UL << 48UL)
  29
  30 #define L0_ATTR (_PAGE_PRESENT | _PAGE_RW | _PAGE_ACCESSED | _PAGE_DIRTY)
  31 #define L1_ATTR (_PAGE_PRESENT | _PAGE_RW | _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE)
  32 #define L2_ATTR (_PAGE_PRESENT | _PAGE_RW | _PAGE_ACCESSED | _PAGE_DIRTY)
  33 #define L3_ATTR (_PAGE_PRESENT | _PAGE_RW | _PAGE_ACCESSED | _PAGE_DIRTY)
  34
  35 static void init_level2_page(u64 *level2p, unsigned long addr)
  36 {
  37         unsigned long end_addr;
  38
  39         addr &= PAGE_MASK;
  40         end_addr = addr + LEVEL2_SIZE;
  41         while (addr < end_addr) {
  42                 *(level2p++) = addr | L1_ATTR;
  43                 addr += LEVEL1_SIZE;
  44         }
  45 }
  46
  47 static int init_level3_page(struct kimage *image, u64 *level3p,
  48                                 unsigned long addr, unsigned long last_addr)
  49 {
  50         unsigned long end_addr;
  51         int result;
  52
  53         result = 0;
  54         addr &= PAGE_MASK;
  55         end_addr = addr + LEVEL3_SIZE;
  56         while ((addr < last_addr) && (addr < end_addr)) {
  57                 struct page *page;
  58                 u64 *level2p;
  59
  60                 page = kimage_alloc_control_pages(image, 0);
  61                 if (!page) {
  62                         result = -ENOMEM;
  63                         goto out;
  64                 }
  65                 level2p = (u64 *)page_address(page);
  66                 init_level2_page(level2p, addr);
  67                 *(level3p++) = __pa(level2p) | L2_ATTR;
  68                 addr += LEVEL2_SIZE;
  69         }
  70         /* clear the unused entries */
  71         while (addr < end_addr) {
  72                 *(level3p++) = 0;
  73                 addr += LEVEL2_SIZE;
  74         }
  75 out:
  76         return result;
  77 }
  78
  79
  80 static int init_level4_page(struct kimage *image, u64 *level4p,
  81                                 unsigned long addr, unsigned long last_addr)
  82 {
  83         unsigned long end_addr;
  84         int result;
  85
  86         result = 0;
  87         addr &= PAGE_MASK;
  88         end_addr = addr + LEVEL4_SIZE;
  89         while ((addr < last_addr) && (addr < end_addr)) {
  90                 struct page *page;
  91                 u64 *level3p;
  92
  93                 page = kimage_alloc_control_pages(image, 0);
  94                 if (!page) {
  95                         result = -ENOMEM;
  96                         goto out;
  97                 }
  98                 level3p = (u64 *)page_address(page);
  99                 result = init_level3_page(image, level3p, addr, last_addr);
 100                 if (result) {
 101                         goto out;
 102                 }
 103                 *(level4p++) = __pa(level3p) | L3_ATTR;
 104                 addr += LEVEL3_SIZE;
 105         }
 106         /* clear the unused entries */
 107         while (addr < end_addr) {
 108                 *(level4p++) = 0;
 109                 addr += LEVEL3_SIZE;
 110         }
 111 out:
 112         return result;
 113 }
 114
 115
 116 static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
 117 {
 118         u64 *level4p;
 119         level4p = (u64 *)__va(start_pgtable);
 120         return init_level4_page(image, level4p, 0, end_pfn << PAGE_SHIFT);
 121 }
 122
 123 static void set_idt(void *newidt, u16 limit)
 124 {
 125         unsigned char curidt[10];
 126
 127         /* x86-64 supports unaliged loads & stores */
 128         (*(u16 *)(curidt)) = limit;
 129         (*(u64 *)(curidt +2)) = (unsigned long)(newidt);
 130
 131         __asm__ __volatile__ (
 132                 "lidt %0\n"
 133                 : "=m" (curidt)
 134                 );
 135 };
 136
 137
 138 static void set_gdt(void *newgdt, u16 limit)
 139 {
 140         unsigned char curgdt[10];
 141
 142         /* x86-64 supports unaligned loads & stores */
 143         (*(u16 *)(curgdt)) = limit;
 144         (*(u64 *)(curgdt +2)) = (unsigned long)(newgdt);
 145
 146         __asm__ __volatile__ (
 147                 "lgdt %0\n"
 148                 : "=m" (curgdt)
 149                 );
 150 };
 151
 152 static void load_segments(void)
 153 {
 154         __asm__ __volatile__ (
 155                 "\tmovl $"STR(__KERNEL_DS)",%eax\n"
 156                 "\tmovl %eax,%ds\n"
 157                 "\tmovl %eax,%es\n"
 158                 "\tmovl %eax,%ss\n"
 159                 "\tmovl %eax,%fs\n"
 160                 "\tmovl %eax,%gs\n"
 161                 );
 162 #undef STR
 163 #undef __STR
 164 }
 165
 166 typedef NORET_TYPE void (*relocate_new_kernel_t)(unsigned long indirection_page,
 167                                         unsigned long control_code_buffer,
 168                                         unsigned long start_address,
 169                                         unsigned long pgtable) ATTRIB_NORET;
 170
 171 const extern unsigned char relocate_new_kernel[];
 172 const extern unsigned long relocate_new_kernel_size;
 173
 174 int machine_kexec_prepare(struct kimage *image)
 175 {
 176         unsigned long start_pgtable, control_code_buffer;
 177         int result;
 178
 179         /* Calculate the offsets */
 180         start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
 181         control_code_buffer = start_pgtable + 4096UL;
 182
 183         /* Setup the identity mapped 64bit page table */
 184         result = init_pgtable(image, start_pgtable);
 185         if (result)
 186                 return result;
 187
 188         /* Place the code in the reboot code buffer */
 189         memcpy(__va(control_code_buffer), relocate_new_kernel,
 190                                                 relocate_new_kernel_size);
 191
 192         return 0;
 193 }
 194
 195 void machine_kexec_cleanup(struct kimage *image)
 196 {
 197         return;
 198 }
 199
 200 /*
 201  * Do not allocate memory (or fail in any way) in machine_kexec().
 202  * We are past the point of no return, committed to rebooting now.
 203  */
 204 NORET_TYPE void machine_kexec(struct kimage *image)
 205 {
 206         unsigned long page_list;
 207         unsigned long control_code_buffer;
 208         unsigned long start_pgtable;
 209         relocate_new_kernel_t rnk;
 210
 211         /* Interrupts aren't acceptable while we reboot */
 212         local_irq_disable();
 213
 214         /* Calculate the offsets */
 215         page_list = image->head;
 216         start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
 217         control_code_buffer = start_pgtable + 4096UL;
 218
 219         /* Set the low half of the page table to my identity mapped
 220          * page table for kexec.  Leave the high half pointing at the
 221          * kernel pages.   Don't bother to flush the global pages
 222          * as that will happen when I fully switch to my identity mapped
 223          * page table anyway.
 224          */
 225         memcpy(__va(read_cr3()), __va(start_pgtable), PAGE_SIZE/2);
 226         __flush_tlb();
 227
 228
 229         /* The segment registers are funny things, they are
 230          * automatically loaded from a table, in memory wherever you
 231          * set them to a specific selector, but this table is never
 232          * accessed again unless you set the segment to a different selector.
 233          *
 234          * The more common model are caches where the behide
 235          * the scenes work is done, but is also dropped at arbitrary
 236          * times.
 237          *
 238          * I take advantage of this here by force loading the
 239          * segments, before I zap the gdt with an invalid value.
 240          */
 241         load_segments();
 242         /* The gdt & idt are now invalid.
 243          * If you want to load them you must set up your own idt & gdt.
 244          */
 245         set_gdt(phys_to_virt(0),0);
 246         set_idt(phys_to_virt(0),0);
 247         /* now call it */
 248         rnk = (relocate_new_kernel_t) control_code_buffer;
 249         (*rnk)(page_list, control_code_buffer, image->start, start_pgtable);
 250 }