2
* handle transition of Linux booting another kernel
3
* Copyright (C) 2002-2005 Eric Biederman <ebiederm@xmission.com>
5
* This source code is licensed under the GNU General Public License,
6
* Version 2. See the file COPYING for more details.
10
#include <linux/kexec.h>
11
#include <linux/string.h>
12
#include <linux/gfp.h>
13
#include <linux/reboot.h>
14
#include <linux/numa.h>
15
#include <linux/ftrace.h>
17
#include <linux/suspend.h>
19
#include <asm/pgtable.h>
20
#include <asm/tlbflush.h>
21
#include <asm/mmu_context.h>
22
#include <asm/debugreg.h>
24
static int init_one_level2_page(struct kimage *image, pgd_t *pgd,
33
pgd += pgd_index(addr);
34
if (!pgd_present(*pgd)) {
35
page = kimage_alloc_control_pages(image, 0);
38
pud = (pud_t *)page_address(page);
40
set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE));
42
pud = pud_offset(pgd, addr);
43
if (!pud_present(*pud)) {
44
page = kimage_alloc_control_pages(image, 0);
47
pmd = (pmd_t *)page_address(page);
49
set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
51
pmd = pmd_offset(pud, addr);
52
if (!pmd_present(*pmd))
53
set_pmd(pmd, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
59
static void init_level2_page(pmd_t *level2p, unsigned long addr)
61
unsigned long end_addr;
64
end_addr = addr + PUD_SIZE;
65
while (addr < end_addr) {
66
set_pmd(level2p++, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
71
static int init_level3_page(struct kimage *image, pud_t *level3p,
72
unsigned long addr, unsigned long last_addr)
74
unsigned long end_addr;
79
end_addr = addr + PGDIR_SIZE;
80
while ((addr < last_addr) && (addr < end_addr)) {
84
page = kimage_alloc_control_pages(image, 0);
89
level2p = (pmd_t *)page_address(page);
90
init_level2_page(level2p, addr);
91
set_pud(level3p++, __pud(__pa(level2p) | _KERNPG_TABLE));
94
/* clear the unused entries */
95
while (addr < end_addr) {
104
static int init_level4_page(struct kimage *image, pgd_t *level4p,
105
unsigned long addr, unsigned long last_addr)
107
unsigned long end_addr;
112
end_addr = addr + (PTRS_PER_PGD * PGDIR_SIZE);
113
while ((addr < last_addr) && (addr < end_addr)) {
117
page = kimage_alloc_control_pages(image, 0);
122
level3p = (pud_t *)page_address(page);
123
result = init_level3_page(image, level3p, addr, last_addr);
126
set_pgd(level4p++, __pgd(__pa(level3p) | _KERNPG_TABLE));
129
/* clear the unused entries */
130
while (addr < end_addr) {
131
pgd_clear(level4p++);
138
static void free_transition_pgtable(struct kimage *image)
140
free_page((unsigned long)image->arch.pud);
141
free_page((unsigned long)image->arch.pmd);
142
free_page((unsigned long)image->arch.pte);
145
static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
150
unsigned long vaddr, paddr;
151
int result = -ENOMEM;
153
vaddr = (unsigned long)relocate_kernel;
154
paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE);
155
pgd += pgd_index(vaddr);
156
if (!pgd_present(*pgd)) {
157
pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
160
image->arch.pud = pud;
161
set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE));
163
pud = pud_offset(pgd, vaddr);
164
if (!pud_present(*pud)) {
165
pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
168
image->arch.pmd = pmd;
169
set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
171
pmd = pmd_offset(pud, vaddr);
172
if (!pmd_present(*pmd)) {
173
pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
176
image->arch.pte = pte;
177
set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
179
pte = pte_offset_kernel(pmd, vaddr);
180
set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
183
free_transition_pgtable(image);
188
static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
192
level4p = (pgd_t *)__va(start_pgtable);
193
result = init_level4_page(image, level4p, 0, max_pfn << PAGE_SHIFT);
197
* image->start may be outside 0 ~ max_pfn, for example when
198
* jump back to original kernel from kexeced kernel
200
result = init_one_level2_page(image, level4p, image->start);
203
return init_transition_pgtable(image, level4p);
206
static void set_idt(void *newidt, u16 limit)
208
struct desc_ptr curidt;
210
/* x86-64 supports unaliged loads & stores */
212
curidt.address = (unsigned long)newidt;
214
__asm__ __volatile__ (
221
static void set_gdt(void *newgdt, u16 limit)
223
struct desc_ptr curgdt;
225
/* x86-64 supports unaligned loads & stores */
227
curgdt.address = (unsigned long)newgdt;
229
__asm__ __volatile__ (
235
static void load_segments(void)
237
__asm__ __volatile__ (
243
: : "a" (__KERNEL_DS) : "memory"
247
int machine_kexec_prepare(struct kimage *image)
249
unsigned long start_pgtable;
252
/* Calculate the offsets */
253
start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
255
/* Setup the identity mapped 64bit page table */
256
result = init_pgtable(image, start_pgtable);
263
void machine_kexec_cleanup(struct kimage *image)
265
free_transition_pgtable(image);
269
* Do not allocate memory (or fail in any way) in machine_kexec().
270
* We are past the point of no return, committed to rebooting now.
272
void machine_kexec(struct kimage *image)
274
unsigned long page_list[PAGES_NR];
276
int save_ftrace_enabled;
278
#ifdef CONFIG_KEXEC_JUMP
279
if (image->preserve_context)
280
save_processor_state();
283
save_ftrace_enabled = __ftrace_enabled_save();
285
/* Interrupts aren't acceptable while we reboot */
287
hw_breakpoint_disable();
289
if (image->preserve_context) {
290
#ifdef CONFIG_X86_IO_APIC
292
* We need to put APICs in legacy mode so that we can
293
* get timer interrupts in second kernel. kexec/kdump
294
* paths already have calls to disable_IO_APIC() in
295
* one form or other. kexec jump path also need
302
control_page = page_address(image->control_code_page) + PAGE_SIZE;
303
memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
305
page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
306
page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
307
page_list[PA_TABLE_PAGE] =
308
(unsigned long)__pa(page_address(image->control_code_page));
310
if (image->type == KEXEC_TYPE_DEFAULT)
311
page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
315
* The segment registers are funny things, they have both a
316
* visible and an invisible part. Whenever the visible part is
317
* set to a specific selector, the invisible part is loaded
318
* with from a table in memory. At no other time is the
319
* descriptor table in memory accessed.
321
* I take advantage of this here by force loading the
322
* segments, before I zap the gdt with an invalid value.
326
* The gdt & idt are now invalid.
327
* If you want to load them you must set up your own idt & gdt.
329
set_gdt(phys_to_virt(0), 0);
330
set_idt(phys_to_virt(0), 0);
333
image->start = relocate_kernel((unsigned long)image->head,
334
(unsigned long)page_list,
336
image->preserve_context);
338
#ifdef CONFIG_KEXEC_JUMP
339
if (image->preserve_context)
340
restore_processor_state();
343
__ftrace_enabled_restore(save_ftrace_enabled);
346
void arch_crash_save_vmcoreinfo(void)
348
VMCOREINFO_SYMBOL(phys_base);
349
VMCOREINFO_SYMBOL(init_level4_pgt);
352
VMCOREINFO_SYMBOL(node_data);
353
VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);