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/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>
18 #include <asm/pgtable.h>
19 #include <asm/pgalloc.h>
20 #include <asm/tlbflush.h>
21 #include <asm/mmu_context.h>
24 #include <asm/cpufeature.h>
26 #include <asm/system.h>
27 #include <asm/cacheflush.h>
29 static void set_idt(void *newidt, __u16 limit)
31 struct desc_ptr curidt;
33 /* ia32 supports unaliged loads & stores */
35 curidt.address = (unsigned long)newidt;
41 static void set_gdt(void *newgdt, __u16 limit)
43 struct desc_ptr curgdt;
45 /* ia32 supports unaligned loads & stores */
47 curgdt.address = (unsigned long)newgdt;
52 static void load_segments(void)
55 #define STR(X) __STR(X)
57 __asm__ __volatile__ (
58 "\tljmp $"STR(__KERNEL_CS)",$1f\n"
60 "\tmovl $"STR(__KERNEL_DS)",%%eax\n"
71 static void machine_kexec_free_page_tables(struct kimage *image)
73 free_page((unsigned long)image->arch.pgd);
75 free_page((unsigned long)image->arch.pmd0);
76 free_page((unsigned long)image->arch.pmd1);
78 free_page((unsigned long)image->arch.pte0);
79 free_page((unsigned long)image->arch.pte1);
82 static int machine_kexec_alloc_page_tables(struct kimage *image)
84 image->arch.pgd = (pgd_t *)get_zeroed_page(GFP_KERNEL);
86 image->arch.pmd0 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
87 image->arch.pmd1 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
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 ||
93 !image->arch.pmd0 || !image->arch.pmd1 ||
95 !image->arch.pte0 || !image->arch.pte1) {
96 machine_kexec_free_page_tables(image);
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.
109 * Do what every setup is needed on image and the
110 * reboot code buffer to allow us to avoid allocations
113 * - Make control page executable.
114 * - Allocate page tables
116 int machine_kexec_prepare(struct kimage *image)
119 set_pages_x(image->control_code_page, 1);
120 return machine_kexec_alloc_page_tables(image);
124 * Undo anything leftover by machine_kexec_prepare
125 * when an image is freed.
127 void machine_kexec_cleanup(struct kimage *image)
130 set_pages_nx(image->control_code_page, 1);
131 machine_kexec_free_page_tables(image);
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.
138 void machine_kexec(struct kimage *image)
140 unsigned long page_list[PAGES_NR];
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);
150 #ifdef CONFIG_KEXEC_JUMP
151 if (kexec_image->preserve_context)
152 save_processor_state();
155 save_ftrace_enabled = __ftrace_enabled_save();
157 /* Interrupts aren't acceptable while we reboot */
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
172 control_page = page_address(image->control_code_page);
173 memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
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;
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;
191 if (image->type == KEXEC_TYPE_DEFAULT)
192 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
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.
201 * I take advantage of this here by force loading the
202 * segments, before I zap the gdt with an invalid value.
205 /* The gdt & idt are now invalid.
206 * If you want to load them you must set up your own idt & gdt.
208 set_gdt(phys_to_virt(0),0);
209 set_idt(phys_to_virt(0),0);
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);
217 #ifdef CONFIG_KEXEC_JUMP
218 if (kexec_image->preserve_context)
219 restore_processor_state();
222 __ftrace_enabled_restore(save_ftrace_enabled);
225 void arch_crash_save_vmcoreinfo(void)
228 VMCOREINFO_SYMBOL(node_data);
229 VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
231 #ifdef CONFIG_X86_PAE
232 VMCOREINFO_CONFIG(X86_PAE);