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/pgtable.h>
17 #include <asm/pgalloc.h>
18 #include <asm/tlbflush.h>
19 #include <asm/mmu_context.h>
20 #include <asm/apic.h>
21 #include <asm/io_apic.h>
22 #include <asm/cpufeature.h>
23 #include <asm/desc.h>
24 #include <asm/set_memory.h>
25 #include <asm/debugreg.h>
26 
27 static void set_gdt(void *newgdt, __u16 limit)
28 {
29 	struct desc_ptr curgdt;
30 
31 	/* ia32 supports unaligned loads & stores */
32 	curgdt.size    = limit;
33 	curgdt.address = (unsigned long)newgdt;
34 
35 	load_gdt(&curgdt);
36 }
37 
38 static void load_segments(void)
39 {
40 #define __STR(X) #X
41 #define STR(X) __STR(X)
42 
43 	__asm__ __volatile__ (
44 		"\tljmp $"STR(__KERNEL_CS)",$1f\n"
45 		"\t1:\n"
46 		"\tmovl $"STR(__KERNEL_DS)",%%eax\n"
47 		"\tmovl %%eax,%%ds\n"
48 		"\tmovl %%eax,%%es\n"
49 		"\tmovl %%eax,%%ss\n"
50 		: : : "eax", "memory");
51 #undef STR
52 #undef __STR
53 }
54 
55 static void machine_kexec_free_page_tables(struct kimage *image)
56 {
57 	free_pages((unsigned long)image->arch.pgd, PGD_ALLOCATION_ORDER);
58 	image->arch.pgd = NULL;
59 #ifdef CONFIG_X86_PAE
60 	free_page((unsigned long)image->arch.pmd0);
61 	image->arch.pmd0 = NULL;
62 	free_page((unsigned long)image->arch.pmd1);
63 	image->arch.pmd1 = NULL;
64 #endif
65 	free_page((unsigned long)image->arch.pte0);
66 	image->arch.pte0 = NULL;
67 	free_page((unsigned long)image->arch.pte1);
68 	image->arch.pte1 = NULL;
69 }
70 
71 static int machine_kexec_alloc_page_tables(struct kimage *image)
72 {
73 	image->arch.pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
74 						    PGD_ALLOCATION_ORDER);
75 #ifdef CONFIG_X86_PAE
76 	image->arch.pmd0 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
77 	image->arch.pmd1 = (pmd_t *)get_zeroed_page(GFP_KERNEL);
78 #endif
79 	image->arch.pte0 = (pte_t *)get_zeroed_page(GFP_KERNEL);
80 	image->arch.pte1 = (pte_t *)get_zeroed_page(GFP_KERNEL);
81 	if (!image->arch.pgd ||
82 #ifdef CONFIG_X86_PAE
83 	    !image->arch.pmd0 || !image->arch.pmd1 ||
84 #endif
85 	    !image->arch.pte0 || !image->arch.pte1) {
86 		return -ENOMEM;
87 	}
88 	return 0;
89 }
90 
91 static void machine_kexec_page_table_set_one(
92 	pgd_t *pgd, pmd_t *pmd, pte_t *pte,
93 	unsigned long vaddr, unsigned long paddr)
94 {
95 	p4d_t *p4d;
96 	pud_t *pud;
97 
98 	pgd += pgd_index(vaddr);
99 #ifdef CONFIG_X86_PAE
100 	if (!(pgd_val(*pgd) & _PAGE_PRESENT))
101 		set_pgd(pgd, __pgd(__pa(pmd) | _PAGE_PRESENT));
102 #endif
103 	p4d = p4d_offset(pgd, vaddr);
104 	pud = pud_offset(p4d, vaddr);
105 	pmd = pmd_offset(pud, vaddr);
106 	if (!(pmd_val(*pmd) & _PAGE_PRESENT))
107 		set_pmd(pmd, __pmd(__pa(pte) | _PAGE_TABLE));
108 	pte = pte_offset_kernel(pmd, vaddr);
109 	set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
110 }
111 
112 static void machine_kexec_prepare_page_tables(struct kimage *image)
113 {
114 	void *control_page;
115 	pmd_t *pmd = NULL;
116 
117 	control_page = page_address(image->control_code_page);
118 #ifdef CONFIG_X86_PAE
119 	pmd = image->arch.pmd0;
120 #endif
121 	machine_kexec_page_table_set_one(
122 		image->arch.pgd, pmd, image->arch.pte0,
123 		(unsigned long)control_page, __pa(control_page));
124 #ifdef CONFIG_X86_PAE
125 	pmd = image->arch.pmd1;
126 #endif
127 	machine_kexec_page_table_set_one(
128 		image->arch.pgd, pmd, image->arch.pte1,
129 		__pa(control_page), __pa(control_page));
130 }
131 
132 /*
133  * A architecture hook called to validate the
134  * proposed image and prepare the control pages
135  * as needed.  The pages for KEXEC_CONTROL_PAGE_SIZE
136  * have been allocated, but the segments have yet
137  * been copied into the kernel.
138  *
139  * Do what every setup is needed on image and the
140  * reboot code buffer to allow us to avoid allocations
141  * later.
142  *
143  * - Make control page executable.
144  * - Allocate page tables
145  * - Setup page tables
146  */
147 int machine_kexec_prepare(struct kimage *image)
148 {
149 	int error;
150 
151 	set_memory_x((unsigned long)page_address(image->control_code_page), 1);
152 	error = machine_kexec_alloc_page_tables(image);
153 	if (error)
154 		return error;
155 	machine_kexec_prepare_page_tables(image);
156 	return 0;
157 }
158 
159 /*
160  * Undo anything leftover by machine_kexec_prepare
161  * when an image is freed.
162  */
163 void machine_kexec_cleanup(struct kimage *image)
164 {
165 	set_memory_nx((unsigned long)page_address(image->control_code_page), 1);
166 	machine_kexec_free_page_tables(image);
167 }
168 
169 /*
170  * Do not allocate memory (or fail in any way) in machine_kexec().
171  * We are past the point of no return, committed to rebooting now.
172  */
173 void machine_kexec(struct kimage *image)
174 {
175 	unsigned long page_list[PAGES_NR];
176 	void *control_page;
177 	int save_ftrace_enabled;
178 	asmlinkage unsigned long
179 		(*relocate_kernel_ptr)(unsigned long indirection_page,
180 				       unsigned long control_page,
181 				       unsigned long start_address,
182 				       unsigned int has_pae,
183 				       unsigned int preserve_context);
184 
185 #ifdef CONFIG_KEXEC_JUMP
186 	if (image->preserve_context)
187 		save_processor_state();
188 #endif
189 
190 	save_ftrace_enabled = __ftrace_enabled_save();
191 
192 	/* Interrupts aren't acceptable while we reboot */
193 	local_irq_disable();
194 	hw_breakpoint_disable();
195 
196 	if (image->preserve_context) {
197 #ifdef CONFIG_X86_IO_APIC
198 		/*
199 		 * We need to put APICs in legacy mode so that we can
200 		 * get timer interrupts in second kernel. kexec/kdump
201 		 * paths already have calls to restore_boot_irq_mode()
202 		 * in one form or other. kexec jump path also need one.
203 		 */
204 		clear_IO_APIC();
205 		restore_boot_irq_mode();
206 #endif
207 	}
208 
209 	control_page = page_address(image->control_code_page);
210 	memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
211 
212 	relocate_kernel_ptr = control_page;
213 	page_list[PA_CONTROL_PAGE] = __pa(control_page);
214 	page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
215 	page_list[PA_PGD] = __pa(image->arch.pgd);
216 
217 	if (image->type == KEXEC_TYPE_DEFAULT)
218 		page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
219 						<< PAGE_SHIFT);
220 
221 	/*
222 	 * The segment registers are funny things, they have both a
223 	 * visible and an invisible part.  Whenever the visible part is
224 	 * set to a specific selector, the invisible part is loaded
225 	 * with from a table in memory.  At no other time is the
226 	 * descriptor table in memory accessed.
227 	 *
228 	 * I take advantage of this here by force loading the
229 	 * segments, before I zap the gdt with an invalid value.
230 	 */
231 	load_segments();
232 	/*
233 	 * The gdt & idt are now invalid.
234 	 * If you want to load them you must set up your own idt & gdt.
235 	 */
236 	idt_invalidate(phys_to_virt(0));
237 	set_gdt(phys_to_virt(0), 0);
238 
239 	/* now call it */
240 	image->start = relocate_kernel_ptr((unsigned long)image->head,
241 					   (unsigned long)page_list,
242 					   image->start,
243 					   boot_cpu_has(X86_FEATURE_PAE),
244 					   image->preserve_context);
245 
246 #ifdef CONFIG_KEXEC_JUMP
247 	if (image->preserve_context)
248 		restore_processor_state();
249 #endif
250 
251 	__ftrace_enabled_restore(save_ftrace_enabled);
252 }
253