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/string.h>
12 #include <linux/gfp.h>
13 #include <linux/reboot.h>
14 #include <linux/numa.h>
15 #include <linux/ftrace.h>
16 #include <linux/io.h>
17 #include <linux/suspend.h>
18 
19 #include <asm/pgtable.h>
20 #include <asm/tlbflush.h>
21 #include <asm/mmu_context.h>
22 #include <asm/debugreg.h>
23 
24 static int init_one_level2_page(struct kimage *image, pgd_t *pgd,
25 				unsigned long addr)
26 {
27 	pud_t *pud;
28 	pmd_t *pmd;
29 	struct page *page;
30 	int result = -ENOMEM;
31 
32 	addr &= PMD_MASK;
33 	pgd += pgd_index(addr);
34 	if (!pgd_present(*pgd)) {
35 		page = kimage_alloc_control_pages(image, 0);
36 		if (!page)
37 			goto out;
38 		pud = (pud_t *)page_address(page);
39 		clear_page(pud);
40 		set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE));
41 	}
42 	pud = pud_offset(pgd, addr);
43 	if (!pud_present(*pud)) {
44 		page = kimage_alloc_control_pages(image, 0);
45 		if (!page)
46 			goto out;
47 		pmd = (pmd_t *)page_address(page);
48 		clear_page(pmd);
49 		set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
50 	}
51 	pmd = pmd_offset(pud, addr);
52 	if (!pmd_present(*pmd))
53 		set_pmd(pmd, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
54 	result = 0;
55 out:
56 	return result;
57 }
58 
59 static void init_level2_page(pmd_t *level2p, unsigned long addr)
60 {
61 	unsigned long end_addr;
62 
63 	addr &= PAGE_MASK;
64 	end_addr = addr + PUD_SIZE;
65 	while (addr < end_addr) {
66 		set_pmd(level2p++, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
67 		addr += PMD_SIZE;
68 	}
69 }
70 
71 static int init_level3_page(struct kimage *image, pud_t *level3p,
72 				unsigned long addr, unsigned long last_addr)
73 {
74 	unsigned long end_addr;
75 	int result;
76 
77 	result = 0;
78 	addr &= PAGE_MASK;
79 	end_addr = addr + PGDIR_SIZE;
80 	while ((addr < last_addr) && (addr < end_addr)) {
81 		struct page *page;
82 		pmd_t *level2p;
83 
84 		page = kimage_alloc_control_pages(image, 0);
85 		if (!page) {
86 			result = -ENOMEM;
87 			goto out;
88 		}
89 		level2p = (pmd_t *)page_address(page);
90 		init_level2_page(level2p, addr);
91 		set_pud(level3p++, __pud(__pa(level2p) | _KERNPG_TABLE));
92 		addr += PUD_SIZE;
93 	}
94 	/* clear the unused entries */
95 	while (addr < end_addr) {
96 		pud_clear(level3p++);
97 		addr += PUD_SIZE;
98 	}
99 out:
100 	return result;
101 }
102 
103 
104 static int init_level4_page(struct kimage *image, pgd_t *level4p,
105 				unsigned long addr, unsigned long last_addr)
106 {
107 	unsigned long end_addr;
108 	int result;
109 
110 	result = 0;
111 	addr &= PAGE_MASK;
112 	end_addr = addr + (PTRS_PER_PGD * PGDIR_SIZE);
113 	while ((addr < last_addr) && (addr < end_addr)) {
114 		struct page *page;
115 		pud_t *level3p;
116 
117 		page = kimage_alloc_control_pages(image, 0);
118 		if (!page) {
119 			result = -ENOMEM;
120 			goto out;
121 		}
122 		level3p = (pud_t *)page_address(page);
123 		result = init_level3_page(image, level3p, addr, last_addr);
124 		if (result)
125 			goto out;
126 		set_pgd(level4p++, __pgd(__pa(level3p) | _KERNPG_TABLE));
127 		addr += PGDIR_SIZE;
128 	}
129 	/* clear the unused entries */
130 	while (addr < end_addr) {
131 		pgd_clear(level4p++);
132 		addr += PGDIR_SIZE;
133 	}
134 out:
135 	return result;
136 }
137 
138 static void free_transition_pgtable(struct kimage *image)
139 {
140 	free_page((unsigned long)image->arch.pud);
141 	free_page((unsigned long)image->arch.pmd);
142 	free_page((unsigned long)image->arch.pte);
143 }
144 
145 static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
146 {
147 	pud_t *pud;
148 	pmd_t *pmd;
149 	pte_t *pte;
150 	unsigned long vaddr, paddr;
151 	int result = -ENOMEM;
152 
153 	vaddr = (unsigned long)relocate_kernel;
154 	paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE);
155 	pgd += pgd_index(vaddr);
156 	if (!pgd_present(*pgd)) {
157 		pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
158 		if (!pud)
159 			goto err;
160 		image->arch.pud = pud;
161 		set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE));
162 	}
163 	pud = pud_offset(pgd, vaddr);
164 	if (!pud_present(*pud)) {
165 		pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
166 		if (!pmd)
167 			goto err;
168 		image->arch.pmd = pmd;
169 		set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
170 	}
171 	pmd = pmd_offset(pud, vaddr);
172 	if (!pmd_present(*pmd)) {
173 		pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
174 		if (!pte)
175 			goto err;
176 		image->arch.pte = pte;
177 		set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
178 	}
179 	pte = pte_offset_kernel(pmd, vaddr);
180 	set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
181 	return 0;
182 err:
183 	free_transition_pgtable(image);
184 	return result;
185 }
186 
187 
188 static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
189 {
190 	pgd_t *level4p;
191 	int result;
192 	level4p = (pgd_t *)__va(start_pgtable);
193 	result = init_level4_page(image, level4p, 0, max_pfn << PAGE_SHIFT);
194 	if (result)
195 		return result;
196 	/*
197 	 * image->start may be outside 0 ~ max_pfn, for example when
198 	 * jump back to original kernel from kexeced kernel
199 	 */
200 	result = init_one_level2_page(image, level4p, image->start);
201 	if (result)
202 		return result;
203 	return init_transition_pgtable(image, level4p);
204 }
205 
206 static void set_idt(void *newidt, u16 limit)
207 {
208 	struct desc_ptr curidt;
209 
210 	/* x86-64 supports unaliged loads & stores */
211 	curidt.size    = limit;
212 	curidt.address = (unsigned long)newidt;
213 
214 	__asm__ __volatile__ (
215 		"lidtq %0\n"
216 		: : "m" (curidt)
217 		);
218 };
219 
220 
221 static void set_gdt(void *newgdt, u16 limit)
222 {
223 	struct desc_ptr curgdt;
224 
225 	/* x86-64 supports unaligned loads & stores */
226 	curgdt.size    = limit;
227 	curgdt.address = (unsigned long)newgdt;
228 
229 	__asm__ __volatile__ (
230 		"lgdtq %0\n"
231 		: : "m" (curgdt)
232 		);
233 };
234 
235 static void load_segments(void)
236 {
237 	__asm__ __volatile__ (
238 		"\tmovl %0,%%ds\n"
239 		"\tmovl %0,%%es\n"
240 		"\tmovl %0,%%ss\n"
241 		"\tmovl %0,%%fs\n"
242 		"\tmovl %0,%%gs\n"
243 		: : "a" (__KERNEL_DS) : "memory"
244 		);
245 }
246 
247 int machine_kexec_prepare(struct kimage *image)
248 {
249 	unsigned long start_pgtable;
250 	int result;
251 
252 	/* Calculate the offsets */
253 	start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
254 
255 	/* Setup the identity mapped 64bit page table */
256 	result = init_pgtable(image, start_pgtable);
257 	if (result)
258 		return result;
259 
260 	return 0;
261 }
262 
263 void machine_kexec_cleanup(struct kimage *image)
264 {
265 	free_transition_pgtable(image);
266 }
267 
268 /*
269  * Do not allocate memory (or fail in any way) in machine_kexec().
270  * We are past the point of no return, committed to rebooting now.
271  */
272 void machine_kexec(struct kimage *image)
273 {
274 	unsigned long page_list[PAGES_NR];
275 	void *control_page;
276 	int save_ftrace_enabled;
277 
278 #ifdef CONFIG_KEXEC_JUMP
279 	if (image->preserve_context)
280 		save_processor_state();
281 #endif
282 
283 	save_ftrace_enabled = __ftrace_enabled_save();
284 
285 	/* Interrupts aren't acceptable while we reboot */
286 	local_irq_disable();
287 	hw_breakpoint_disable();
288 
289 	if (image->preserve_context) {
290 #ifdef CONFIG_X86_IO_APIC
291 		/*
292 		 * We need to put APICs in legacy mode so that we can
293 		 * get timer interrupts in second kernel. kexec/kdump
294 		 * paths already have calls to disable_IO_APIC() in
295 		 * one form or other. kexec jump path also need
296 		 * one.
297 		 */
298 		disable_IO_APIC();
299 #endif
300 	}
301 
302 	control_page = page_address(image->control_code_page) + PAGE_SIZE;
303 	memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
304 
305 	page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
306 	page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
307 	page_list[PA_TABLE_PAGE] =
308 	  (unsigned long)__pa(page_address(image->control_code_page));
309 
310 	if (image->type == KEXEC_TYPE_DEFAULT)
311 		page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
312 						<< PAGE_SHIFT);
313 
314 	/*
315 	 * The segment registers are funny things, they have both a
316 	 * visible and an invisible part.  Whenever the visible part is
317 	 * set to a specific selector, the invisible part is loaded
318 	 * with from a table in memory.  At no other time is the
319 	 * descriptor table in memory accessed.
320 	 *
321 	 * I take advantage of this here by force loading the
322 	 * segments, before I zap the gdt with an invalid value.
323 	 */
324 	load_segments();
325 	/*
326 	 * The gdt & idt are now invalid.
327 	 * If you want to load them you must set up your own idt & gdt.
328 	 */
329 	set_gdt(phys_to_virt(0), 0);
330 	set_idt(phys_to_virt(0), 0);
331 
332 	/* now call it */
333 	image->start = relocate_kernel((unsigned long)image->head,
334 				       (unsigned long)page_list,
335 				       image->start,
336 				       image->preserve_context);
337 
338 #ifdef CONFIG_KEXEC_JUMP
339 	if (image->preserve_context)
340 		restore_processor_state();
341 #endif
342 
343 	__ftrace_enabled_restore(save_ftrace_enabled);
344 }
345 
346 void arch_crash_save_vmcoreinfo(void)
347 {
348 	VMCOREINFO_SYMBOL(phys_base);
349 	VMCOREINFO_SYMBOL(init_level4_pgt);
350 
351 #ifdef CONFIG_NUMA
352 	VMCOREINFO_SYMBOL(node_data);
353 	VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
354 #endif
355 }
356 
357