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
29static 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
41static 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
52static 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
71static void machine_kexec_free_page_tables(struct kimage *image)
72{
73	free_page((unsigned long)image->arch.pgd);
74#ifdef CONFIG_X86_PAE
75	free_page((unsigned long)image->arch.pmd0);
76	free_page((unsigned long)image->arch.pmd1);
77#endif
78	free_page((unsigned long)image->arch.pte0);
79	free_page((unsigned long)image->arch.pte1);
80}
81
82static int machine_kexec_alloc_page_tables(struct kimage *image)
83{
84	image->arch.pgd = (pgd_t *)get_zeroed_page(GFP_KERNEL);
85#ifdef CONFIG_X86_PAE
86	image->arch.pmd0 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
87	image->arch.pmd1 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
88#endif
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 ||
92#ifdef CONFIG_X86_PAE
93	    !image->arch.pmd0 || !image->arch.pmd1 ||
94#endif
95	    !image->arch.pte0 || !image->arch.pte1) {
96		machine_kexec_free_page_tables(image);
97		return -ENOMEM;
98	}
99	return 0;
100}
101
102static void machine_kexec_page_table_set_one(
103	pgd_t *pgd, pmd_t *pmd, pte_t *pte,
104	unsigned long vaddr, unsigned long paddr)
105{
106	pud_t *pud;
107
108	pgd += pgd_index(vaddr);
109#ifdef CONFIG_X86_PAE
110	if (!(pgd_val(*pgd) & _PAGE_PRESENT))
111		set_pgd(pgd, __pgd(__pa(pmd) | _PAGE_PRESENT));
112#endif
113	pud = pud_offset(pgd, vaddr);
114	pmd = pmd_offset(pud, vaddr);
115	if (!(pmd_val(*pmd) & _PAGE_PRESENT))
116		set_pmd(pmd, __pmd(__pa(pte) | _PAGE_TABLE));
117	pte = pte_offset_kernel(pmd, vaddr);
118	set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
119}
120
121static void machine_kexec_prepare_page_tables(struct kimage *image)
122{
123	void *control_page;
124	pmd_t *pmd = NULL;
125
126	control_page = page_address(image->control_code_page);
127#ifdef CONFIG_X86_PAE
128	pmd = image->arch.pmd0;
129#endif
130	machine_kexec_page_table_set_one(
131		image->arch.pgd, pmd, image->arch.pte0,
132		(unsigned long)control_page, __pa(control_page));
133#ifdef CONFIG_X86_PAE
134	pmd = image->arch.pmd1;
135#endif
136	machine_kexec_page_table_set_one(
137		image->arch.pgd, pmd, image->arch.pte1,
138		__pa(control_page), __pa(control_page));
139}
140
141/*
142 * A architecture hook called to validate the
143 * proposed image and prepare the control pages
144 * as needed.  The pages for KEXEC_CONTROL_PAGE_SIZE
145 * have been allocated, but the segments have yet
146 * been copied into the kernel.
147 *
148 * Do what every setup is needed on image and the
149 * reboot code buffer to allow us to avoid allocations
150 * later.
151 *
152 * - Make control page executable.
153 * - Allocate page tables
154 * - Setup page tables
155 */
156int machine_kexec_prepare(struct kimage *image)
157{
158	int error;
159
160	set_pages_x(image->control_code_page, 1);
161	error = machine_kexec_alloc_page_tables(image);
162	if (error)
163		return error;
164	machine_kexec_prepare_page_tables(image);
165	return 0;
166}
167
168/*
169 * Undo anything leftover by machine_kexec_prepare
170 * when an image is freed.
171 */
172void machine_kexec_cleanup(struct kimage *image)
173{
174	set_pages_nx(image->control_code_page, 1);
175	machine_kexec_free_page_tables(image);
176}
177
178/*
179 * Do not allocate memory (or fail in any way) in machine_kexec().
180 * We are past the point of no return, committed to rebooting now.
181 */
182void machine_kexec(struct kimage *image)
183{
184	unsigned long page_list[PAGES_NR];
185	void *control_page;
186	int save_ftrace_enabled;
187	asmlinkage unsigned long
188		(*relocate_kernel_ptr)(unsigned long indirection_page,
189				       unsigned long control_page,
190				       unsigned long start_address,
191				       unsigned int has_pae,
192				       unsigned int preserve_context);
193
194#ifdef CONFIG_KEXEC_JUMP
195	if (image->preserve_context)
196		save_processor_state();
197#endif
198
199	save_ftrace_enabled = __ftrace_enabled_save();
200
201	/* Interrupts aren't acceptable while we reboot */
202	local_irq_disable();
203	hw_breakpoint_disable();
204
205	if (image->preserve_context) {
206#ifdef CONFIG_X86_IO_APIC
207		/*
208		 * We need to put APICs in legacy mode so that we can
209		 * get timer interrupts in second kernel. kexec/kdump
210		 * paths already have calls to disable_IO_APIC() in
211		 * one form or other. kexec jump path also need
212		 * one.
213		 */
214		disable_IO_APIC();
215#endif
216	}
217
218	control_page = page_address(image->control_code_page);
219	memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
220
221	relocate_kernel_ptr = control_page;
222	page_list[PA_CONTROL_PAGE] = __pa(control_page);
223	page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
224	page_list[PA_PGD] = __pa(image->arch.pgd);
225
226	if (image->type == KEXEC_TYPE_DEFAULT)
227		page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
228						<< PAGE_SHIFT);
229
230	/*
231	 * The segment registers are funny things, they have both a
232	 * visible and an invisible part.  Whenever the visible part is
233	 * set to a specific selector, the invisible part is loaded
234	 * with from a table in memory.  At no other time is the
235	 * descriptor table in memory accessed.
236	 *
237	 * I take advantage of this here by force loading the
238	 * segments, before I zap the gdt with an invalid value.
239	 */
240	load_segments();
241	/*
242	 * The gdt & idt are now invalid.
243	 * If you want to load them you must set up your own idt & gdt.
244	 */
245	set_gdt(phys_to_virt(0), 0);
246	set_idt(phys_to_virt(0), 0);
247
248	/* now call it */
249	image->start = relocate_kernel_ptr((unsigned long)image->head,
250					   (unsigned long)page_list,
251					   image->start,
252					   boot_cpu_has(X86_FEATURE_PAE),
253					   image->preserve_context);
254
255#ifdef CONFIG_KEXEC_JUMP
256	if (image->preserve_context)
257		restore_processor_state();
258#endif
259
260	__ftrace_enabled_restore(save_ftrace_enabled);
261}
262
263void arch_crash_save_vmcoreinfo(void)
264{
265#ifdef CONFIG_NUMA
266	VMCOREINFO_SYMBOL(node_data);
267	VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
268#endif
269#ifdef CONFIG_X86_PAE
270	VMCOREINFO_CONFIG(X86_PAE);
271#endif
272}
273
274