1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * handle transition of Linux booting another kernel
4  * Copyright (C) 2002-2005 Eric Biederman  <ebiederm@xmission.com>
5  */
6 
7 #include <linux/mm.h>
8 #include <linux/kexec.h>
9 #include <linux/delay.h>
10 #include <linux/numa.h>
11 #include <linux/ftrace.h>
12 #include <linux/suspend.h>
13 #include <linux/gfp.h>
14 #include <linux/io.h>
15 
16 #include <asm/pgalloc.h>
17 #include <asm/tlbflush.h>
18 #include <asm/mmu_context.h>
19 #include <asm/apic.h>
20 #include <asm/io_apic.h>
21 #include <asm/cpufeature.h>
22 #include <asm/desc.h>
23 #include <asm/set_memory.h>
24 #include <asm/debugreg.h>
25 
set_gdt(void * newgdt,__u16 limit)26 static void set_gdt(void *newgdt, __u16 limit)
27 {
28 	struct desc_ptr curgdt;
29 
30 	/* ia32 supports unaligned loads & stores */
31 	curgdt.size    = limit;
32 	curgdt.address = (unsigned long)newgdt;
33 
34 	load_gdt(&curgdt);
35 }
36 
load_segments(void)37 static void load_segments(void)
38 {
39 #define __STR(X) #X
40 #define STR(X) __STR(X)
41 
42 	__asm__ __volatile__ (
43 		"\tljmp $"STR(__KERNEL_CS)",$1f\n"
44 		"\t1:\n"
45 		"\tmovl $"STR(__KERNEL_DS)",%%eax\n"
46 		"\tmovl %%eax,%%ds\n"
47 		"\tmovl %%eax,%%es\n"
48 		"\tmovl %%eax,%%ss\n"
49 		: : : "eax", "memory");
50 #undef STR
51 #undef __STR
52 }
53 
machine_kexec_free_page_tables(struct kimage * image)54 static void machine_kexec_free_page_tables(struct kimage *image)
55 {
56 	free_pages((unsigned long)image->arch.pgd, PGD_ALLOCATION_ORDER);
57 	image->arch.pgd = NULL;
58 #ifdef CONFIG_X86_PAE
59 	free_page((unsigned long)image->arch.pmd0);
60 	image->arch.pmd0 = NULL;
61 	free_page((unsigned long)image->arch.pmd1);
62 	image->arch.pmd1 = NULL;
63 #endif
64 	free_page((unsigned long)image->arch.pte0);
65 	image->arch.pte0 = NULL;
66 	free_page((unsigned long)image->arch.pte1);
67 	image->arch.pte1 = NULL;
68 }
69 
machine_kexec_alloc_page_tables(struct kimage * image)70 static int machine_kexec_alloc_page_tables(struct kimage *image)
71 {
72 	image->arch.pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
73 						    PGD_ALLOCATION_ORDER);
74 #ifdef CONFIG_X86_PAE
75 	image->arch.pmd0 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
76 	image->arch.pmd1 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
77 #endif
78 	image->arch.pte0 = (pte_t *)get_zeroed_page(GFP_KERNEL);
79 	image->arch.pte1 = (pte_t *)get_zeroed_page(GFP_KERNEL);
80 	if (!image->arch.pgd ||
81 #ifdef CONFIG_X86_PAE
82 	    !image->arch.pmd0 || !image->arch.pmd1 ||
83 #endif
84 	    !image->arch.pte0 || !image->arch.pte1) {
85 		return -ENOMEM;
86 	}
87 	return 0;
88 }
89 
machine_kexec_page_table_set_one(pgd_t * pgd,pmd_t * pmd,pte_t * pte,unsigned long vaddr,unsigned long paddr)90 static void machine_kexec_page_table_set_one(
91 	pgd_t *pgd, pmd_t *pmd, pte_t *pte,
92 	unsigned long vaddr, unsigned long paddr)
93 {
94 	p4d_t *p4d;
95 	pud_t *pud;
96 
97 	pgd += pgd_index(vaddr);
98 #ifdef CONFIG_X86_PAE
99 	if (!(pgd_val(*pgd) & _PAGE_PRESENT))
100 		set_pgd(pgd, __pgd(__pa(pmd) | _PAGE_PRESENT));
101 #endif
102 	p4d = p4d_offset(pgd, vaddr);
103 	pud = pud_offset(p4d, vaddr);
104 	pmd = pmd_offset(pud, vaddr);
105 	if (!(pmd_val(*pmd) & _PAGE_PRESENT))
106 		set_pmd(pmd, __pmd(__pa(pte) | _PAGE_TABLE));
107 	pte = pte_offset_kernel(pmd, vaddr);
108 	set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
109 }
110 
machine_kexec_prepare_page_tables(struct kimage * image)111 static void machine_kexec_prepare_page_tables(struct kimage *image)
112 {
113 	void *control_page;
114 	pmd_t *pmd = NULL;
115 
116 	control_page = page_address(image->control_code_page);
117 #ifdef CONFIG_X86_PAE
118 	pmd = image->arch.pmd0;
119 #endif
120 	machine_kexec_page_table_set_one(
121 		image->arch.pgd, pmd, image->arch.pte0,
122 		(unsigned long)control_page, __pa(control_page));
123 #ifdef CONFIG_X86_PAE
124 	pmd = image->arch.pmd1;
125 #endif
126 	machine_kexec_page_table_set_one(
127 		image->arch.pgd, pmd, image->arch.pte1,
128 		__pa(control_page), __pa(control_page));
129 }
130 
131 /*
132  * A architecture hook called to validate the
133  * proposed image and prepare the control pages
134  * as needed.  The pages for KEXEC_CONTROL_PAGE_SIZE
135  * have been allocated, but the segments have yet
136  * been copied into the kernel.
137  *
138  * Do what every setup is needed on image and the
139  * reboot code buffer to allow us to avoid allocations
140  * later.
141  *
142  * - Make control page executable.
143  * - Allocate page tables
144  * - Setup page tables
145  */
machine_kexec_prepare(struct kimage * image)146 int machine_kexec_prepare(struct kimage *image)
147 {
148 	int error;
149 
150 	set_memory_x((unsigned long)page_address(image->control_code_page), 1);
151 	error = machine_kexec_alloc_page_tables(image);
152 	if (error)
153 		return error;
154 	machine_kexec_prepare_page_tables(image);
155 	return 0;
156 }
157 
158 /*
159  * Undo anything leftover by machine_kexec_prepare
160  * when an image is freed.
161  */
machine_kexec_cleanup(struct kimage * image)162 void machine_kexec_cleanup(struct kimage *image)
163 {
164 	set_memory_nx((unsigned long)page_address(image->control_code_page), 1);
165 	machine_kexec_free_page_tables(image);
166 }
167 
168 /*
169  * Do not allocate memory (or fail in any way) in machine_kexec().
170  * We are past the point of no return, committed to rebooting now.
171  */
machine_kexec(struct kimage * image)172 void machine_kexec(struct kimage *image)
173 {
174 	unsigned long page_list[PAGES_NR];
175 	void *control_page;
176 	int save_ftrace_enabled;
177 	asmlinkage unsigned long
178 		(*relocate_kernel_ptr)(unsigned long indirection_page,
179 				       unsigned long control_page,
180 				       unsigned long start_address,
181 				       unsigned int has_pae,
182 				       unsigned int preserve_context);
183 
184 #ifdef CONFIG_KEXEC_JUMP
185 	if (image->preserve_context)
186 		save_processor_state();
187 #endif
188 
189 	save_ftrace_enabled = __ftrace_enabled_save();
190 
191 	/* Interrupts aren't acceptable while we reboot */
192 	local_irq_disable();
193 	hw_breakpoint_disable();
194 
195 	if (image->preserve_context) {
196 #ifdef CONFIG_X86_IO_APIC
197 		/*
198 		 * We need to put APICs in legacy mode so that we can
199 		 * get timer interrupts in second kernel. kexec/kdump
200 		 * paths already have calls to restore_boot_irq_mode()
201 		 * in one form or other. kexec jump path also need one.
202 		 */
203 		clear_IO_APIC();
204 		restore_boot_irq_mode();
205 #endif
206 	}
207 
208 	control_page = page_address(image->control_code_page);
209 	memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
210 
211 	relocate_kernel_ptr = control_page;
212 	page_list[PA_CONTROL_PAGE] = __pa(control_page);
213 	page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
214 	page_list[PA_PGD] = __pa(image->arch.pgd);
215 
216 	if (image->type == KEXEC_TYPE_DEFAULT)
217 		page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
218 						<< PAGE_SHIFT);
219 
220 	/*
221 	 * The segment registers are funny things, they have both a
222 	 * visible and an invisible part.  Whenever the visible part is
223 	 * set to a specific selector, the invisible part is loaded
224 	 * with from a table in memory.  At no other time is the
225 	 * descriptor table in memory accessed.
226 	 *
227 	 * I take advantage of this here by force loading the
228 	 * segments, before I zap the gdt with an invalid value.
229 	 */
230 	load_segments();
231 	/*
232 	 * The gdt & idt are now invalid.
233 	 * If you want to load them you must set up your own idt & gdt.
234 	 */
235 	idt_invalidate(phys_to_virt(0));
236 	set_gdt(phys_to_virt(0), 0);
237 
238 	/* now call it */
239 	image->start = relocate_kernel_ptr((unsigned long)image->head,
240 					   (unsigned long)page_list,
241 					   image->start,
242 					   boot_cpu_has(X86_FEATURE_PAE),
243 					   image->preserve_context);
244 
245 #ifdef CONFIG_KEXEC_JUMP
246 	if (image->preserve_context)
247 		restore_processor_state();
248 #endif
249 
250 	__ftrace_enabled_restore(save_ftrace_enabled);
251 }
252