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