Merge branch 'linus' into tracing/hw-breakpoints
[linux-2.6.git] / 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 #include <linux/io.h>
18
19 #include <asm/pgtable.h>
20 #include <asm/pgalloc.h>
21 #include <asm/tlbflush.h>
22 #include <asm/mmu_context.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 #include <asm/debugreg.h>
29
30 static void set_idt(void *newidt, __u16 limit)
31 {
32         struct desc_ptr curidt;
33
34         /* ia32 supports unaliged loads & stores */
35         curidt.size    = limit;
36         curidt.address = (unsigned long)newidt;
37
38         load_idt(&curidt);
39 }
40
41
42 static void set_gdt(void *newgdt, __u16 limit)
43 {
44         struct desc_ptr curgdt;
45
46         /* ia32 supports unaligned loads & stores */
47         curgdt.size    = limit;
48         curgdt.address = (unsigned long)newgdt;
49
50         load_gdt(&curgdt);
51 }
52
53 static void load_segments(void)
54 {
55 #define __STR(X) #X
56 #define STR(X) __STR(X)
57
58         __asm__ __volatile__ (
59                 "\tljmp $"STR(__KERNEL_CS)",$1f\n"
60                 "\t1:\n"
61                 "\tmovl $"STR(__KERNEL_DS)",%%eax\n"
62                 "\tmovl %%eax,%%ds\n"
63                 "\tmovl %%eax,%%es\n"
64                 "\tmovl %%eax,%%fs\n"
65                 "\tmovl %%eax,%%gs\n"
66                 "\tmovl %%eax,%%ss\n"
67                 : : : "eax", "memory");
68 #undef STR
69 #undef __STR
70 }
71
72 static void machine_kexec_free_page_tables(struct kimage *image)
73 {
74         free_page((unsigned long)image->arch.pgd);
75 #ifdef CONFIG_X86_PAE
76         free_page((unsigned long)image->arch.pmd0);
77         free_page((unsigned long)image->arch.pmd1);
78 #endif
79         free_page((unsigned long)image->arch.pte0);
80         free_page((unsigned long)image->arch.pte1);
81 }
82
83 static int machine_kexec_alloc_page_tables(struct kimage *image)
84 {
85         image->arch.pgd = (pgd_t *)get_zeroed_page(GFP_KERNEL);
86 #ifdef CONFIG_X86_PAE
87         image->arch.pmd0 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
88         image->arch.pmd1 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
89 #endif
90         image->arch.pte0 = (pte_t *)get_zeroed_page(GFP_KERNEL);
91         image->arch.pte1 = (pte_t *)get_zeroed_page(GFP_KERNEL);
92         if (!image->arch.pgd ||
93 #ifdef CONFIG_X86_PAE
94             !image->arch.pmd0 || !image->arch.pmd1 ||
95 #endif
96             !image->arch.pte0 || !image->arch.pte1) {
97                 machine_kexec_free_page_tables(image);
98                 return -ENOMEM;
99         }
100         return 0;
101 }
102
103 static void machine_kexec_page_table_set_one(
104         pgd_t *pgd, pmd_t *pmd, pte_t *pte,
105         unsigned long vaddr, unsigned long paddr)
106 {
107         pud_t *pud;
108
109         pgd += pgd_index(vaddr);
110 #ifdef CONFIG_X86_PAE
111         if (!(pgd_val(*pgd) & _PAGE_PRESENT))
112                 set_pgd(pgd, __pgd(__pa(pmd) | _PAGE_PRESENT));
113 #endif
114         pud = pud_offset(pgd, vaddr);
115         pmd = pmd_offset(pud, vaddr);
116         if (!(pmd_val(*pmd) & _PAGE_PRESENT))
117                 set_pmd(pmd, __pmd(__pa(pte) | _PAGE_TABLE));
118         pte = pte_offset_kernel(pmd, vaddr);
119         set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
120 }
121
122 static void machine_kexec_prepare_page_tables(struct kimage *image)
123 {
124         void *control_page;
125         pmd_t *pmd = NULL;
126
127         control_page = page_address(image->control_code_page);
128 #ifdef CONFIG_X86_PAE
129         pmd = image->arch.pmd0;
130 #endif
131         machine_kexec_page_table_set_one(
132                 image->arch.pgd, pmd, image->arch.pte0,
133                 (unsigned long)control_page, __pa(control_page));
134 #ifdef CONFIG_X86_PAE
135         pmd = image->arch.pmd1;
136 #endif
137         machine_kexec_page_table_set_one(
138                 image->arch.pgd, pmd, image->arch.pte1,
139                 __pa(control_page), __pa(control_page));
140 }
141
142 /*
143  * A architecture hook called to validate the
144  * proposed image and prepare the control pages
145  * as needed.  The pages for KEXEC_CONTROL_PAGE_SIZE
146  * have been allocated, but the segments have yet
147  * been copied into the kernel.
148  *
149  * Do what every setup is needed on image and the
150  * reboot code buffer to allow us to avoid allocations
151  * later.
152  *
153  * - Make control page executable.
154  * - Allocate page tables
155  * - Setup page tables
156  */
157 int machine_kexec_prepare(struct kimage *image)
158 {
159         int error;
160
161         if (nx_enabled)
162                 set_pages_x(image->control_code_page, 1);
163         error = machine_kexec_alloc_page_tables(image);
164         if (error)
165                 return error;
166         machine_kexec_prepare_page_tables(image);
167         return 0;
168 }
169
170 /*
171  * Undo anything leftover by machine_kexec_prepare
172  * when an image is freed.
173  */
174 void machine_kexec_cleanup(struct kimage *image)
175 {
176         if (nx_enabled)
177                 set_pages_nx(image->control_code_page, 1);
178         machine_kexec_free_page_tables(image);
179 }
180
181 /*
182  * Do not allocate memory (or fail in any way) in machine_kexec().
183  * We are past the point of no return, committed to rebooting now.
184  */
185 void machine_kexec(struct kimage *image)
186 {
187         unsigned long page_list[PAGES_NR];
188         void *control_page;
189         int save_ftrace_enabled;
190         asmlinkage unsigned long
191                 (*relocate_kernel_ptr)(unsigned long indirection_page,
192                                        unsigned long control_page,
193                                        unsigned long start_address,
194                                        unsigned int has_pae,
195                                        unsigned int preserve_context);
196
197 #ifdef CONFIG_KEXEC_JUMP
198         if (image->preserve_context)
199                 save_processor_state();
200 #endif
201
202         save_ftrace_enabled = __ftrace_enabled_save();
203
204         /* Interrupts aren't acceptable while we reboot */
205         local_irq_disable();
206         hw_breakpoint_disable();
207
208         if (image->preserve_context) {
209 #ifdef CONFIG_X86_IO_APIC
210                 /*
211                  * We need to put APICs in legacy mode so that we can
212                  * get timer interrupts in second kernel. kexec/kdump
213                  * paths already have calls to disable_IO_APIC() in
214                  * one form or other. kexec jump path also need
215                  * one.
216                  */
217                 disable_IO_APIC();
218 #endif
219         }
220
221         control_page = page_address(image->control_code_page);
222         memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
223
224         relocate_kernel_ptr = control_page;
225         page_list[PA_CONTROL_PAGE] = __pa(control_page);
226         page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
227         page_list[PA_PGD] = __pa(image->arch.pgd);
228
229         if (image->type == KEXEC_TYPE_DEFAULT)
230                 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
231                                                 << PAGE_SHIFT);
232
233         /*
234          * The segment registers are funny things, they have both a
235          * visible and an invisible part.  Whenever the visible part is
236          * set to a specific selector, the invisible part is loaded
237          * with from a table in memory.  At no other time is the
238          * descriptor table in memory accessed.
239          *
240          * I take advantage of this here by force loading the
241          * segments, before I zap the gdt with an invalid value.
242          */
243         load_segments();
244         /*
245          * The gdt & idt are now invalid.
246          * If you want to load them you must set up your own idt & gdt.
247          */
248         set_gdt(phys_to_virt(0), 0);
249         set_idt(phys_to_virt(0), 0);
250
251         /* now call it */
252         image->start = relocate_kernel_ptr((unsigned long)image->head,
253                                            (unsigned long)page_list,
254                                            image->start, cpu_has_pae,
255                                            image->preserve_context);
256
257 #ifdef CONFIG_KEXEC_JUMP
258         if (image->preserve_context)
259                 restore_processor_state();
260 #endif
261
262         __ftrace_enabled_restore(save_ftrace_enabled);
263 }
264
265 void arch_crash_save_vmcoreinfo(void)
266 {
267 #ifdef CONFIG_NUMA
268         VMCOREINFO_SYMBOL(node_data);
269         VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
270 #endif
271 #ifdef CONFIG_X86_PAE
272         VMCOREINFO_CONFIG(X86_PAE);
273 #endif
274 }
275