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 #include <linux/suspend.h>
  17
  18 #include <asm/pgtable.h>
  19 #include <asm/tlbflush.h>
  20 #include <asm/mmu_context.h>
  21 #include <asm/debugreg.h>
  22
  23 static int init_one_level2_page(struct kimage *image, pgd_t *pgd,
  24                                 unsigned long addr)
  25 {
  26         pud_t *pud;
  27         pmd_t *pmd;
  28         struct page *page;
  29         int result = -ENOMEM;
  30
  31         addr &= PMD_MASK;
  32         pgd += pgd_index(addr);
  33         if (!pgd_present(*pgd)) {
  34                 page = kimage_alloc_control_pages(image, 0);
  35                 if (!page)
  36                         goto out;
  37                 pud = (pud_t *)page_address(page);
  38                 memset(pud, 0, PAGE_SIZE);
  39                 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE));
  40         }
  41         pud = pud_offset(pgd, addr);
  42         if (!pud_present(*pud)) {
  43                 page = kimage_alloc_control_pages(image, 0);
  44                 if (!page)
  45                         goto out;
  46                 pmd = (pmd_t *)page_address(page);
  47                 memset(pmd, 0, PAGE_SIZE);
  48                 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
  49         }
  50         pmd = pmd_offset(pud, addr);
  51         if (!pmd_present(*pmd))
  52                 set_pmd(pmd, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
  53         result = 0;
  54 out:
  55         return result;
  56 }
  57
  58 static void init_level2_page(pmd_t *level2p, unsigned long addr)
  59 {
  60         unsigned long end_addr;
  61
  62         addr &= PAGE_MASK;
  63         end_addr = addr + PUD_SIZE;
  64         while (addr < end_addr) {
  65                 set_pmd(level2p++, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
  66                 addr += PMD_SIZE;
  67         }
  68 }
  69
  70 static int init_level3_page(struct kimage *image, pud_t *level3p,
  71                                 unsigned long addr, unsigned long last_addr)
  72 {
  73         unsigned long end_addr;
  74         int result;
  75
  76         result = 0;
  77         addr &= PAGE_MASK;
  78         end_addr = addr + PGDIR_SIZE;
  79         while ((addr < last_addr) && (addr < end_addr)) {
  80                 struct page *page;
  81                 pmd_t *level2p;
  82
  83                 page = kimage_alloc_control_pages(image, 0);
  84                 if (!page) {
  85                         result = -ENOMEM;
  86                         goto out;
  87                 }
  88                 level2p = (pmd_t *)page_address(page);
  89                 init_level2_page(level2p, addr);
  90                 set_pud(level3p++, __pud(__pa(level2p) | _KERNPG_TABLE));
  91                 addr += PUD_SIZE;
  92         }
  93         /* clear the unused entries */
  94         while (addr < end_addr) {
  95                 pud_clear(level3p++);
  96                 addr += PUD_SIZE;
  97         }
  98 out:
  99         return result;
 100 }
 101
 102
 103 static int init_level4_page(struct kimage *image, pgd_t *level4p,
 104                                 unsigned long addr, unsigned long last_addr)
 105 {
 106         unsigned long end_addr;
 107         int result;
 108
 109         result = 0;
 110         addr &= PAGE_MASK;
 111         end_addr = addr + (PTRS_PER_PGD * PGDIR_SIZE);
 112         while ((addr < last_addr) && (addr < end_addr)) {
 113                 struct page *page;
 114                 pud_t *level3p;
 115
 116                 page = kimage_alloc_control_pages(image, 0);
 117                 if (!page) {
 118                         result = -ENOMEM;
 119                         goto out;
 120                 }
 121                 level3p = (pud_t *)page_address(page);
 122                 result = init_level3_page(image, level3p, addr, last_addr);
 123                 if (result)
 124                         goto out;
 125                 set_pgd(level4p++, __pgd(__pa(level3p) | _KERNPG_TABLE));
 126                 addr += PGDIR_SIZE;
 127         }
 128         /* clear the unused entries */
 129         while (addr < end_addr) {
 130                 pgd_clear(level4p++);
 131                 addr += PGDIR_SIZE;
 132         }
 133 out:
 134         return result;
 135 }
 136
 137 static void free_transition_pgtable(struct kimage *image)
 138 {
 139         free_page((unsigned long)image->arch.pud);
 140         free_page((unsigned long)image->arch.pmd);
 141         free_page((unsigned long)image->arch.pte);
 142 }
 143
 144 static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
 145 {
 146         pud_t *pud;
 147         pmd_t *pmd;
 148         pte_t *pte;
 149         unsigned long vaddr, paddr;
 150         int result = -ENOMEM;
 151
 152         vaddr = (unsigned long)relocate_kernel;
 153         paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE);
 154         pgd += pgd_index(vaddr);
 155         if (!pgd_present(*pgd)) {
 156                 pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
 157                 if (!pud)
 158                         goto err;
 159                 image->arch.pud = pud;
 160                 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE));
 161         }
 162         pud = pud_offset(pgd, vaddr);
 163         if (!pud_present(*pud)) {
 164                 pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
 165                 if (!pmd)
 166                         goto err;
 167                 image->arch.pmd = pmd;
 168                 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
 169         }
 170         pmd = pmd_offset(pud, vaddr);
 171         if (!pmd_present(*pmd)) {
 172                 pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
 173                 if (!pte)
 174                         goto err;
 175                 image->arch.pte = pte;
 176                 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
 177         }
 178         pte = pte_offset_kernel(pmd, vaddr);
 179         set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
 180         return 0;
 181 err:
 182         free_transition_pgtable(image);
 183         return result;
 184 }
 185
 186
 187 static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
 188 {
 189         pgd_t *level4p;
 190         int result;
 191         level4p = (pgd_t *)__va(start_pgtable);
 192         result = init_level4_page(image, level4p, 0, max_pfn << PAGE_SHIFT);
 193         if (result)
 194                 return result;
 195         /*
 196          * image->start may be outside 0 ~ max_pfn, for example when
 197          * jump back to original kernel from kexeced kernel
 198          */
 199         result = init_one_level2_page(image, level4p, image->start);
 200         if (result)
 201                 return result;
 202         return init_transition_pgtable(image, level4p);
 203 }
 204
 205 static void set_idt(void *newidt, u16 limit)
 206 {
 207         struct desc_ptr curidt;
 208
 209         /* x86-64 supports unaliged loads & stores */
 210         curidt.size    = limit;
 211         curidt.address = (unsigned long)newidt;
 212
 213         __asm__ __volatile__ (
 214                 "lidtq %0\n"
 215                 : : "m" (curidt)
 216                 );
 217 };
 218
 219
 220 static void set_gdt(void *newgdt, u16 limit)
 221 {
 222         struct desc_ptr curgdt;
 223
 224         /* x86-64 supports unaligned loads & stores */
 225         curgdt.size    = limit;
 226         curgdt.address = (unsigned long)newgdt;
 227
 228         __asm__ __volatile__ (
 229                 "lgdtq %0\n"
 230                 : : "m" (curgdt)
 231                 );
 232 };
 233
 234 static void load_segments(void)
 235 {
 236         __asm__ __volatile__ (
 237                 "\tmovl %0,%%ds\n"
 238                 "\tmovl %0,%%es\n"
 239                 "\tmovl %0,%%ss\n"
 240                 "\tmovl %0,%%fs\n"
 241                 "\tmovl %0,%%gs\n"
 242                 : : "a" (__KERNEL_DS) : "memory"
 243                 );
 244 }
 245
 246 int machine_kexec_prepare(struct kimage *image)
 247 {
 248         unsigned long start_pgtable;
 249         int result;
 250
 251         /* Calculate the offsets */
 252         start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
 253
 254         /* Setup the identity mapped 64bit page table */
 255         result = init_pgtable(image, start_pgtable);
 256         if (result)
 257                 return result;
 258
 259         return 0;
 260 }
 261
 262 void machine_kexec_cleanup(struct kimage *image)
 263 {
 264         free_transition_pgtable(image);
 265 }
 266
 267 /*
 268  * Do not allocate memory (or fail in any way) in machine_kexec().
 269  * We are past the point of no return, committed to rebooting now.
 270  */
 271 void machine_kexec(struct kimage *image)
 272 {
 273         unsigned long page_list[PAGES_NR];
 274         void *control_page;
 275         int save_ftrace_enabled;
 276
 277 #ifdef CONFIG_KEXEC_JUMP
 278         if (image->preserve_context)
 279                 save_processor_state();
 280 #endif
 281
 282         save_ftrace_enabled = __ftrace_enabled_save();
 283
 284         /* Interrupts aren't acceptable while we reboot */
 285         local_irq_disable();
 286         hw_breakpoint_disable();
 287
 288         if (image->preserve_context) {
 289 #ifdef CONFIG_X86_IO_APIC
 290                 /*
 291                  * We need to put APICs in legacy mode so that we can
 292                  * get timer interrupts in second kernel. kexec/kdump
 293                  * paths already have calls to disable_IO_APIC() in
 294                  * one form or other. kexec jump path also need
 295                  * one.
 296                  */
 297                 disable_IO_APIC();
 298 #endif
 299         }
 300
 301         control_page = page_address(image->control_code_page) + PAGE_SIZE;
 302         memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
 303
 304         page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
 305         page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
 306         page_list[PA_TABLE_PAGE] =
 307           (unsigned long)__pa(page_address(image->control_code_page));
 308
 309         if (image->type == KEXEC_TYPE_DEFAULT)
 310                 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
 311                                                 << PAGE_SHIFT);
 312
 313         /*
 314          * The segment registers are funny things, they have both a
 315          * visible and an invisible part.  Whenever the visible part is
 316          * set to a specific selector, the invisible part is loaded
 317          * with from a table in memory.  At no other time is the
 318          * descriptor table in memory accessed.
 319          *
 320          * I take advantage of this here by force loading the
 321          * segments, before I zap the gdt with an invalid value.
 322          */
 323         load_segments();
 324         /*
 325          * The gdt & idt are now invalid.
 326          * If you want to load them you must set up your own idt & gdt.
 327          */
 328         set_gdt(phys_to_virt(0), 0);
 329         set_idt(phys_to_virt(0), 0);
 330
 331         /* now call it */
 332         image->start = relocate_kernel((unsigned long)image->head,
 333                                        (unsigned long)page_list,
 334                                        image->start,
 335                                        image->preserve_context);
 336
 337 #ifdef CONFIG_KEXEC_JUMP
 338         if (image->preserve_context)
 339                 restore_processor_state();
 340 #endif
 341
 342         __ftrace_enabled_restore(save_ftrace_enabled);
 343 }
 344
 345 void arch_crash_save_vmcoreinfo(void)
 346 {
 347         VMCOREINFO_SYMBOL(phys_base);
 348         VMCOREINFO_SYMBOL(init_level4_pgt);
 349
 350 #ifdef CONFIG_NUMA
 351         VMCOREINFO_SYMBOL(node_data);
 352         VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
 353 #endif
 354 }
 355