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/numa.h>
13 #include <linux/ftrace.h>
14 #include <linux/suspend.h>
15 #include <linux/gfp.h>
16 #include <linux/io.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/apic.h>
23 #include <asm/io_apic.h>
24 #include <asm/cpufeature.h>
25 #include <asm/desc.h>
26 #include <asm/cacheflush.h>
27 #include <asm/debugreg.h>
28
set_idt(void * newidt,__u16 limit)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
set_gdt(void * newgdt,__u16 limit)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
load_segments(void)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
machine_kexec_free_page_tables(struct kimage * image)71 static void machine_kexec_free_page_tables(struct kimage *image)
72 {
73 free_page((unsigned long)image->arch.pgd);
74 image->arch.pgd = NULL;
75 #ifdef CONFIG_X86_PAE
76 free_page((unsigned long)image->arch.pmd0);
77 image->arch.pmd0 = NULL;
78 free_page((unsigned long)image->arch.pmd1);
79 image->arch.pmd1 = NULL;
80 #endif
81 free_page((unsigned long)image->arch.pte0);
82 image->arch.pte0 = NULL;
83 free_page((unsigned long)image->arch.pte1);
84 image->arch.pte1 = NULL;
85 }
86
machine_kexec_alloc_page_tables(struct kimage * image)87 static int machine_kexec_alloc_page_tables(struct kimage *image)
88 {
89 image->arch.pgd = (pgd_t *)get_zeroed_page(GFP_KERNEL);
90 #ifdef CONFIG_X86_PAE
91 image->arch.pmd0 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
92 image->arch.pmd1 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
93 #endif
94 image->arch.pte0 = (pte_t *)get_zeroed_page(GFP_KERNEL);
95 image->arch.pte1 = (pte_t *)get_zeroed_page(GFP_KERNEL);
96 if (!image->arch.pgd ||
97 #ifdef CONFIG_X86_PAE
98 !image->arch.pmd0 || !image->arch.pmd1 ||
99 #endif
100 !image->arch.pte0 || !image->arch.pte1) {
101 return -ENOMEM;
102 }
103 return 0;
104 }
105
machine_kexec_page_table_set_one(pgd_t * pgd,pmd_t * pmd,pte_t * pte,unsigned long vaddr,unsigned long paddr)106 static void machine_kexec_page_table_set_one(
107 pgd_t *pgd, pmd_t *pmd, pte_t *pte,
108 unsigned long vaddr, unsigned long paddr)
109 {
110 pud_t *pud;
111
112 pgd += pgd_index(vaddr);
113 #ifdef CONFIG_X86_PAE
114 if (!(pgd_val(*pgd) & _PAGE_PRESENT))
115 set_pgd(pgd, __pgd(__pa(pmd) | _PAGE_PRESENT));
116 #endif
117 pud = pud_offset(pgd, vaddr);
118 pmd = pmd_offset(pud, vaddr);
119 if (!(pmd_val(*pmd) & _PAGE_PRESENT))
120 set_pmd(pmd, __pmd(__pa(pte) | _PAGE_TABLE));
121 pte = pte_offset_kernel(pmd, vaddr);
122 set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
123 }
124
machine_kexec_prepare_page_tables(struct kimage * image)125 static void machine_kexec_prepare_page_tables(struct kimage *image)
126 {
127 void *control_page;
128 pmd_t *pmd = NULL;
129
130 control_page = page_address(image->control_code_page);
131 #ifdef CONFIG_X86_PAE
132 pmd = image->arch.pmd0;
133 #endif
134 machine_kexec_page_table_set_one(
135 image->arch.pgd, pmd, image->arch.pte0,
136 (unsigned long)control_page, __pa(control_page));
137 #ifdef CONFIG_X86_PAE
138 pmd = image->arch.pmd1;
139 #endif
140 machine_kexec_page_table_set_one(
141 image->arch.pgd, pmd, image->arch.pte1,
142 __pa(control_page), __pa(control_page));
143 }
144
145 /*
146 * A architecture hook called to validate the
147 * proposed image and prepare the control pages
148 * as needed. The pages for KEXEC_CONTROL_PAGE_SIZE
149 * have been allocated, but the segments have yet
150 * been copied into the kernel.
151 *
152 * Do what every setup is needed on image and the
153 * reboot code buffer to allow us to avoid allocations
154 * later.
155 *
156 * - Make control page executable.
157 * - Allocate page tables
158 * - Setup page tables
159 */
machine_kexec_prepare(struct kimage * image)160 int machine_kexec_prepare(struct kimage *image)
161 {
162 int error;
163
164 set_pages_x(image->control_code_page, 1);
165 error = machine_kexec_alloc_page_tables(image);
166 if (error)
167 return error;
168 machine_kexec_prepare_page_tables(image);
169 return 0;
170 }
171
172 /*
173 * Undo anything leftover by machine_kexec_prepare
174 * when an image is freed.
175 */
machine_kexec_cleanup(struct kimage * image)176 void machine_kexec_cleanup(struct kimage *image)
177 {
178 set_pages_nx(image->control_code_page, 1);
179 machine_kexec_free_page_tables(image);
180 }
181
182 /*
183 * Do not allocate memory (or fail in any way) in machine_kexec().
184 * We are past the point of no return, committed to rebooting now.
185 */
machine_kexec(struct kimage * image)186 void machine_kexec(struct kimage *image)
187 {
188 unsigned long page_list[PAGES_NR];
189 void *control_page;
190 int save_ftrace_enabled;
191 asmlinkage unsigned long
192 (*relocate_kernel_ptr)(unsigned long indirection_page,
193 unsigned long control_page,
194 unsigned long start_address,
195 unsigned int has_pae,
196 unsigned int preserve_context);
197
198 #ifdef CONFIG_KEXEC_JUMP
199 if (image->preserve_context)
200 save_processor_state();
201 #endif
202
203 save_ftrace_enabled = __ftrace_enabled_save();
204
205 /* Interrupts aren't acceptable while we reboot */
206 local_irq_disable();
207 hw_breakpoint_disable();
208
209 if (image->preserve_context) {
210 #ifdef CONFIG_X86_IO_APIC
211 /*
212 * We need to put APICs in legacy mode so that we can
213 * get timer interrupts in second kernel. kexec/kdump
214 * paths already have calls to disable_IO_APIC() in
215 * one form or other. kexec jump path also need
216 * one.
217 */
218 disable_IO_APIC();
219 #endif
220 }
221
222 control_page = page_address(image->control_code_page);
223 memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
224
225 relocate_kernel_ptr = control_page;
226 page_list[PA_CONTROL_PAGE] = __pa(control_page);
227 page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
228 page_list[PA_PGD] = __pa(image->arch.pgd);
229
230 if (image->type == KEXEC_TYPE_DEFAULT)
231 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
232 << PAGE_SHIFT);
233
234 /*
235 * The segment registers are funny things, they have both a
236 * visible and an invisible part. Whenever the visible part is
237 * set to a specific selector, the invisible part is loaded
238 * with from a table in memory. At no other time is the
239 * descriptor table in memory accessed.
240 *
241 * I take advantage of this here by force loading the
242 * segments, before I zap the gdt with an invalid value.
243 */
244 load_segments();
245 /*
246 * The gdt & idt are now invalid.
247 * If you want to load them you must set up your own idt & gdt.
248 */
249 set_gdt(phys_to_virt(0), 0);
250 set_idt(phys_to_virt(0), 0);
251
252 /* now call it */
253 image->start = relocate_kernel_ptr((unsigned long)image->head,
254 (unsigned long)page_list,
255 image->start,
256 boot_cpu_has(X86_FEATURE_PAE),
257 image->preserve_context);
258
259 #ifdef CONFIG_KEXEC_JUMP
260 if (image->preserve_context)
261 restore_processor_state();
262 #endif
263
264 __ftrace_enabled_restore(save_ftrace_enabled);
265 }
266
arch_crash_save_vmcoreinfo(void)267 void arch_crash_save_vmcoreinfo(void)
268 {
269 #ifdef CONFIG_NUMA
270 VMCOREINFO_SYMBOL(node_data);
271 VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
272 #endif
273 #ifdef CONFIG_X86_PAE
274 VMCOREINFO_CONFIG(X86_PAE);
275 #endif
276 }
277
278