xref: /linux/arch/arm64/kernel/machine_kexec.c (revision f3956ebb3bf06ab2266ad5ee2214aed46405810c)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * kexec for arm64
4  *
5  * Copyright (C) Linaro.
6  * Copyright (C) Huawei Futurewei Technologies.
7  */
8 
9 #include <linux/interrupt.h>
10 #include <linux/irq.h>
11 #include <linux/kernel.h>
12 #include <linux/kexec.h>
13 #include <linux/page-flags.h>
14 #include <linux/set_memory.h>
15 #include <linux/smp.h>
16 
17 #include <asm/cacheflush.h>
18 #include <asm/cpu_ops.h>
19 #include <asm/daifflags.h>
20 #include <asm/memory.h>
21 #include <asm/mmu.h>
22 #include <asm/mmu_context.h>
23 #include <asm/page.h>
24 
25 #include "cpu-reset.h"
26 
27 /* Global variables for the arm64_relocate_new_kernel routine. */
28 extern const unsigned char arm64_relocate_new_kernel[];
29 extern const unsigned long arm64_relocate_new_kernel_size;
30 
31 /**
32  * kexec_image_info - For debugging output.
33  */
34 #define kexec_image_info(_i) _kexec_image_info(__func__, __LINE__, _i)
35 static void _kexec_image_info(const char *func, int line,
36 	const struct kimage *kimage)
37 {
38 	unsigned long i;
39 
40 	pr_debug("%s:%d:\n", func, line);
41 	pr_debug("  kexec kimage info:\n");
42 	pr_debug("    type:        %d\n", kimage->type);
43 	pr_debug("    start:       %lx\n", kimage->start);
44 	pr_debug("    head:        %lx\n", kimage->head);
45 	pr_debug("    nr_segments: %lu\n", kimage->nr_segments);
46 	pr_debug("    kern_reloc: %pa\n", &kimage->arch.kern_reloc);
47 
48 	for (i = 0; i < kimage->nr_segments; i++) {
49 		pr_debug("      segment[%lu]: %016lx - %016lx, 0x%lx bytes, %lu pages\n",
50 			i,
51 			kimage->segment[i].mem,
52 			kimage->segment[i].mem + kimage->segment[i].memsz,
53 			kimage->segment[i].memsz,
54 			kimage->segment[i].memsz /  PAGE_SIZE);
55 	}
56 }
57 
58 void machine_kexec_cleanup(struct kimage *kimage)
59 {
60 	/* Empty routine needed to avoid build errors. */
61 }
62 
63 int machine_kexec_post_load(struct kimage *kimage)
64 {
65 	void *reloc_code = page_to_virt(kimage->control_code_page);
66 
67 	memcpy(reloc_code, arm64_relocate_new_kernel,
68 	       arm64_relocate_new_kernel_size);
69 	kimage->arch.kern_reloc = __pa(reloc_code);
70 	kexec_image_info(kimage);
71 
72 	/*
73 	 * For execution with the MMU off, reloc_code needs to be cleaned to the
74 	 * PoC and invalidated from the I-cache.
75 	 */
76 	dcache_clean_inval_poc((unsigned long)reloc_code,
77 			    (unsigned long)reloc_code +
78 				    arm64_relocate_new_kernel_size);
79 	icache_inval_pou((uintptr_t)reloc_code,
80 				(uintptr_t)reloc_code +
81 					arm64_relocate_new_kernel_size);
82 
83 	return 0;
84 }
85 
86 /**
87  * machine_kexec_prepare - Prepare for a kexec reboot.
88  *
89  * Called from the core kexec code when a kernel image is loaded.
90  * Forbid loading a kexec kernel if we have no way of hotplugging cpus or cpus
91  * are stuck in the kernel. This avoids a panic once we hit machine_kexec().
92  */
93 int machine_kexec_prepare(struct kimage *kimage)
94 {
95 	if (kimage->type != KEXEC_TYPE_CRASH && cpus_are_stuck_in_kernel()) {
96 		pr_err("Can't kexec: CPUs are stuck in the kernel.\n");
97 		return -EBUSY;
98 	}
99 
100 	return 0;
101 }
102 
103 /**
104  * kexec_list_flush - Helper to flush the kimage list and source pages to PoC.
105  */
106 static void kexec_list_flush(struct kimage *kimage)
107 {
108 	kimage_entry_t *entry;
109 
110 	for (entry = &kimage->head; ; entry++) {
111 		unsigned int flag;
112 		unsigned long addr;
113 
114 		/* flush the list entries. */
115 		dcache_clean_inval_poc((unsigned long)entry,
116 				    (unsigned long)entry +
117 					    sizeof(kimage_entry_t));
118 
119 		flag = *entry & IND_FLAGS;
120 		if (flag == IND_DONE)
121 			break;
122 
123 		addr = (unsigned long)phys_to_virt(*entry & PAGE_MASK);
124 
125 		switch (flag) {
126 		case IND_INDIRECTION:
127 			/* Set entry point just before the new list page. */
128 			entry = (kimage_entry_t *)addr - 1;
129 			break;
130 		case IND_SOURCE:
131 			/* flush the source pages. */
132 			dcache_clean_inval_poc(addr, addr + PAGE_SIZE);
133 			break;
134 		case IND_DESTINATION:
135 			break;
136 		default:
137 			BUG();
138 		}
139 	}
140 }
141 
142 /**
143  * kexec_segment_flush - Helper to flush the kimage segments to PoC.
144  */
145 static void kexec_segment_flush(const struct kimage *kimage)
146 {
147 	unsigned long i;
148 
149 	pr_debug("%s:\n", __func__);
150 
151 	for (i = 0; i < kimage->nr_segments; i++) {
152 		pr_debug("  segment[%lu]: %016lx - %016lx, 0x%lx bytes, %lu pages\n",
153 			i,
154 			kimage->segment[i].mem,
155 			kimage->segment[i].mem + kimage->segment[i].memsz,
156 			kimage->segment[i].memsz,
157 			kimage->segment[i].memsz /  PAGE_SIZE);
158 
159 		dcache_clean_inval_poc(
160 			(unsigned long)phys_to_virt(kimage->segment[i].mem),
161 			(unsigned long)phys_to_virt(kimage->segment[i].mem) +
162 				kimage->segment[i].memsz);
163 	}
164 }
165 
166 /**
167  * machine_kexec - Do the kexec reboot.
168  *
169  * Called from the core kexec code for a sys_reboot with LINUX_REBOOT_CMD_KEXEC.
170  */
171 void machine_kexec(struct kimage *kimage)
172 {
173 	bool in_kexec_crash = (kimage == kexec_crash_image);
174 	bool stuck_cpus = cpus_are_stuck_in_kernel();
175 
176 	/*
177 	 * New cpus may have become stuck_in_kernel after we loaded the image.
178 	 */
179 	BUG_ON(!in_kexec_crash && (stuck_cpus || (num_online_cpus() > 1)));
180 	WARN(in_kexec_crash && (stuck_cpus || smp_crash_stop_failed()),
181 		"Some CPUs may be stale, kdump will be unreliable.\n");
182 
183 	/* Flush the kimage list and its buffers. */
184 	kexec_list_flush(kimage);
185 
186 	/* Flush the new image if already in place. */
187 	if ((kimage != kexec_crash_image) && (kimage->head & IND_DONE))
188 		kexec_segment_flush(kimage);
189 
190 	pr_info("Bye!\n");
191 
192 	local_daif_mask();
193 
194 	/*
195 	 * cpu_soft_restart will shutdown the MMU, disable data caches, then
196 	 * transfer control to the kern_reloc which contains a copy of
197 	 * the arm64_relocate_new_kernel routine.  arm64_relocate_new_kernel
198 	 * uses physical addressing to relocate the new image to its final
199 	 * position and transfers control to the image entry point when the
200 	 * relocation is complete.
201 	 * In kexec case, kimage->start points to purgatory assuming that
202 	 * kernel entry and dtb address are embedded in purgatory by
203 	 * userspace (kexec-tools).
204 	 * In kexec_file case, the kernel starts directly without purgatory.
205 	 */
206 	cpu_soft_restart(kimage->arch.kern_reloc, kimage->head, kimage->start,
207 			 kimage->arch.dtb_mem);
208 
209 	BUG(); /* Should never get here. */
210 }
211 
212 static void machine_kexec_mask_interrupts(void)
213 {
214 	unsigned int i;
215 	struct irq_desc *desc;
216 
217 	for_each_irq_desc(i, desc) {
218 		struct irq_chip *chip;
219 		int ret;
220 
221 		chip = irq_desc_get_chip(desc);
222 		if (!chip)
223 			continue;
224 
225 		/*
226 		 * First try to remove the active state. If this
227 		 * fails, try to EOI the interrupt.
228 		 */
229 		ret = irq_set_irqchip_state(i, IRQCHIP_STATE_ACTIVE, false);
230 
231 		if (ret && irqd_irq_inprogress(&desc->irq_data) &&
232 		    chip->irq_eoi)
233 			chip->irq_eoi(&desc->irq_data);
234 
235 		if (chip->irq_mask)
236 			chip->irq_mask(&desc->irq_data);
237 
238 		if (chip->irq_disable && !irqd_irq_disabled(&desc->irq_data))
239 			chip->irq_disable(&desc->irq_data);
240 	}
241 }
242 
243 /**
244  * machine_crash_shutdown - shutdown non-crashing cpus and save registers
245  */
246 void machine_crash_shutdown(struct pt_regs *regs)
247 {
248 	local_irq_disable();
249 
250 	/* shutdown non-crashing cpus */
251 	crash_smp_send_stop();
252 
253 	/* for crashing cpu */
254 	crash_save_cpu(regs, smp_processor_id());
255 	machine_kexec_mask_interrupts();
256 
257 	pr_info("Starting crashdump kernel...\n");
258 }
259 
260 void arch_kexec_protect_crashkres(void)
261 {
262 	int i;
263 
264 	kexec_segment_flush(kexec_crash_image);
265 
266 	for (i = 0; i < kexec_crash_image->nr_segments; i++)
267 		set_memory_valid(
268 			__phys_to_virt(kexec_crash_image->segment[i].mem),
269 			kexec_crash_image->segment[i].memsz >> PAGE_SHIFT, 0);
270 }
271 
272 void arch_kexec_unprotect_crashkres(void)
273 {
274 	int i;
275 
276 	for (i = 0; i < kexec_crash_image->nr_segments; i++)
277 		set_memory_valid(
278 			__phys_to_virt(kexec_crash_image->segment[i].mem),
279 			kexec_crash_image->segment[i].memsz >> PAGE_SHIFT, 1);
280 }
281 
282 #ifdef CONFIG_HIBERNATION
283 /*
284  * To preserve the crash dump kernel image, the relevant memory segments
285  * should be mapped again around the hibernation.
286  */
287 void crash_prepare_suspend(void)
288 {
289 	if (kexec_crash_image)
290 		arch_kexec_unprotect_crashkres();
291 }
292 
293 void crash_post_resume(void)
294 {
295 	if (kexec_crash_image)
296 		arch_kexec_protect_crashkres();
297 }
298 
299 /*
300  * crash_is_nosave
301  *
302  * Return true only if a page is part of reserved memory for crash dump kernel,
303  * but does not hold any data of loaded kernel image.
304  *
305  * Note that all the pages in crash dump kernel memory have been initially
306  * marked as Reserved as memory was allocated via memblock_reserve().
307  *
308  * In hibernation, the pages which are Reserved and yet "nosave" are excluded
309  * from the hibernation iamge. crash_is_nosave() does thich check for crash
310  * dump kernel and will reduce the total size of hibernation image.
311  */
312 
313 bool crash_is_nosave(unsigned long pfn)
314 {
315 	int i;
316 	phys_addr_t addr;
317 
318 	if (!crashk_res.end)
319 		return false;
320 
321 	/* in reserved memory? */
322 	addr = __pfn_to_phys(pfn);
323 	if ((addr < crashk_res.start) || (crashk_res.end < addr))
324 		return false;
325 
326 	if (!kexec_crash_image)
327 		return true;
328 
329 	/* not part of loaded kernel image? */
330 	for (i = 0; i < kexec_crash_image->nr_segments; i++)
331 		if (addr >= kexec_crash_image->segment[i].mem &&
332 				addr < (kexec_crash_image->segment[i].mem +
333 					kexec_crash_image->segment[i].memsz))
334 			return false;
335 
336 	return true;
337 }
338 
339 void crash_free_reserved_phys_range(unsigned long begin, unsigned long end)
340 {
341 	unsigned long addr;
342 	struct page *page;
343 
344 	for (addr = begin; addr < end; addr += PAGE_SIZE) {
345 		page = phys_to_page(addr);
346 		free_reserved_page(page);
347 	}
348 }
349 #endif /* CONFIG_HIBERNATION */
350