arch/x86/kernel/machine_kexec_32.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/delay.h>
  12 #include <linux/init.h>
  13 #include <linux/numa.h>
  14 #include <linux/ftrace.h>
  15 #include <linux/suspend.h>
  16 #include <linux/gfp.h>
  17
  18 #include <asm/pgtable.h>
  19 #include <asm/pgalloc.h>
  20 #include <asm/tlbflush.h>
  21 #include <asm/mmu_context.h>
  22 #include <asm/io.h>
  23 #include <asm/apic.h>
  24 #include <asm/cpufeature.h>
  25 #include <asm/desc.h>
  26 #include <asm/system.h>
  27 #include <asm/cacheflush.h>
  28
  29 static void set_idt(void *newidt, __u16 limit)
  30 {
  31         struct desc_ptr curidt;
  32
  33         /* ia32 supports unaliged loads & stores */
  34         curidt.size    = limit;
  35         curidt.address = (unsigned long)newidt;
  36
  37         load_idt(&curidt);
  38 }
  39
  40
  41 static void set_gdt(void *newgdt, __u16 limit)
  42 {
  43         struct desc_ptr curgdt;
  44
  45         /* ia32 supports unaligned loads & stores */
  46         curgdt.size    = limit;
  47         curgdt.address = (unsigned long)newgdt;
  48
  49         load_gdt(&curgdt);
  50 }
  51
  52 static void load_segments(void)
  53 {
  54 #define __STR(X) #X
  55 #define STR(X) __STR(X)
  56
  57         __asm__ __volatile__ (
  58                 "\tljmp $"STR(__KERNEL_CS)",$1f\n"
  59                 "\t1:\n"
  60                 "\tmovl $"STR(__KERNEL_DS)",%%eax\n"
  61                 "\tmovl %%eax,%%ds\n"
  62                 "\tmovl %%eax,%%es\n"
  63                 "\tmovl %%eax,%%fs\n"
  64                 "\tmovl %%eax,%%gs\n"
  65                 "\tmovl %%eax,%%ss\n"
  66                 ::: "eax", "memory");
  67 #undef STR
  68 #undef __STR
  69 }
  70
  71 static void machine_kexec_free_page_tables(struct kimage *image)
  72 {
  73         free_page((unsigned long)image->arch.pgd);
  74 #ifdef CONFIG_X86_PAE
  75         free_page((unsigned long)image->arch.pmd0);
  76         free_page((unsigned long)image->arch.pmd1);
  77 #endif
  78         free_page((unsigned long)image->arch.pte0);
  79         free_page((unsigned long)image->arch.pte1);
  80 }
  81
  82 static int machine_kexec_alloc_page_tables(struct kimage *image)
  83 {
  84         image->arch.pgd = (pgd_t *)get_zeroed_page(GFP_KERNEL);
  85 #ifdef CONFIG_X86_PAE
  86         image->arch.pmd0 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
  87         image->arch.pmd1 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
  88 #endif
  89         image->arch.pte0 = (pte_t *)get_zeroed_page(GFP_KERNEL);
  90         image->arch.pte1 = (pte_t *)get_zeroed_page(GFP_KERNEL);
  91         if (!image->arch.pgd ||
  92 #ifdef CONFIG_X86_PAE
  93             !image->arch.pmd0 || !image->arch.pmd1 ||
  94 #endif
  95             !image->arch.pte0 || !image->arch.pte1) {
  96                 machine_kexec_free_page_tables(image);
  97                 return -ENOMEM;
  98         }
  99         return 0;
 100 }
 101
 102 /*
 103  * A architecture hook called to validate the
 104  * proposed image and prepare the control pages
 105  * as needed.  The pages for KEXEC_CONTROL_PAGE_SIZE
 106  * have been allocated, but the segments have yet
 107  * been copied into the kernel.
 108  *
 109  * Do what every setup is needed on image and the
 110  * reboot code buffer to allow us to avoid allocations
 111  * later.
 112  *
 113  * - Make control page executable.
 114  * - Allocate page tables
 115  */
 116 int machine_kexec_prepare(struct kimage *image)
 117 {
 118         if (nx_enabled)
 119                 set_pages_x(image->control_code_page, 1);
 120         return machine_kexec_alloc_page_tables(image);
 121 }
 122
 123 /*
 124  * Undo anything leftover by machine_kexec_prepare
 125  * when an image is freed.
 126  */
 127 void machine_kexec_cleanup(struct kimage *image)
 128 {
 129         if (nx_enabled)
 130                 set_pages_nx(image->control_code_page, 1);
 131         machine_kexec_free_page_tables(image);
 132 }
 133
 134 /*
 135  * Do not allocate memory (or fail in any way) in machine_kexec().
 136  * We are past the point of no return, committed to rebooting now.
 137  */
 138 void machine_kexec(struct kimage *image)
 139 {
 140         unsigned long page_list[PAGES_NR];
 141         void *control_page;
 142         int save_ftrace_enabled;
 143         asmlinkage unsigned long
 144                 (*relocate_kernel_ptr)(unsigned long indirection_page,
 145                                        unsigned long control_page,
 146                                        unsigned long start_address,
 147                                        unsigned int has_pae,
 148                                        unsigned int preserve_context);
 149
 150 #ifdef CONFIG_KEXEC_JUMP
 151         if (kexec_image->preserve_context)
 152                 save_processor_state();
 153 #endif
 154
 155         save_ftrace_enabled = __ftrace_enabled_save();
 156
 157         /* Interrupts aren't acceptable while we reboot */
 158         local_irq_disable();
 159
 160         if (image->preserve_context) {
 161 #ifdef CONFIG_X86_IO_APIC
 162                 /* We need to put APICs in legacy mode so that we can
 163                  * get timer interrupts in second kernel. kexec/kdump
 164                  * paths already have calls to disable_IO_APIC() in
 165                  * one form or other. kexec jump path also need
 166                  * one.
 167                  */
 168                 disable_IO_APIC();
 169 #endif
 170         }
 171
 172         control_page = page_address(image->control_code_page);
 173         memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
 174
 175         relocate_kernel_ptr = control_page;
 176         page_list[PA_CONTROL_PAGE] = __pa(control_page);
 177         page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
 178         page_list[PA_PGD] = __pa(image->arch.pgd);
 179         page_list[VA_PGD] = (unsigned long)image->arch.pgd;
 180 #ifdef CONFIG_X86_PAE
 181         page_list[PA_PMD_0] = __pa(image->arch.pmd0);
 182         page_list[VA_PMD_0] = (unsigned long)image->arch.pmd0;
 183         page_list[PA_PMD_1] = __pa(image->arch.pmd1);
 184         page_list[VA_PMD_1] = (unsigned long)image->arch.pmd1;
 185 #endif
 186         page_list[PA_PTE_0] = __pa(image->arch.pte0);
 187         page_list[VA_PTE_0] = (unsigned long)image->arch.pte0;
 188         page_list[PA_PTE_1] = __pa(image->arch.pte1);
 189         page_list[VA_PTE_1] = (unsigned long)image->arch.pte1;
 190
 191         if (image->type == KEXEC_TYPE_DEFAULT)
 192                 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
 193                                                 << PAGE_SHIFT);
 194
 195         /* The segment registers are funny things, they have both a
 196          * visible and an invisible part.  Whenever the visible part is
 197          * set to a specific selector, the invisible part is loaded
 198          * with from a table in memory.  At no other time is the
 199          * descriptor table in memory accessed.
 200          *
 201          * I take advantage of this here by force loading the
 202          * segments, before I zap the gdt with an invalid value.
 203          */
 204         load_segments();
 205         /* The gdt & idt are now invalid.
 206          * If you want to load them you must set up your own idt & gdt.
 207          */
 208         set_gdt(phys_to_virt(0),0);
 209         set_idt(phys_to_virt(0),0);
 210
 211         /* now call it */
 212         image->start = relocate_kernel_ptr((unsigned long)image->head,
 213                                            (unsigned long)page_list,
 214                                            image->start, cpu_has_pae,
 215                                            image->preserve_context);
 216
 217 #ifdef CONFIG_KEXEC_JUMP
 218         if (kexec_image->preserve_context)
 219                 restore_processor_state();
 220 #endif
 221
 222         __ftrace_enabled_restore(save_ftrace_enabled);
 223 }
 224
 225 void arch_crash_save_vmcoreinfo(void)
 226 {
 227 #ifdef CONFIG_NUMA
 228         VMCOREINFO_SYMBOL(node_data);
 229         VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
 230 #endif
 231 #ifdef CONFIG_X86_PAE
 232         VMCOREINFO_CONFIG(X86_PAE);
 233 #endif
 234 }
 235