1 /*
2  * kexec for arm64
3  *
4  * Copyright (C) Linaro.
5  * Copyright (C) Huawei Futurewei Technologies.
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  */
11 
12 #include <linux/interrupt.h>
13 #include <linux/irq.h>
14 #include <linux/kernel.h>
15 #include <linux/kexec.h>
16 #include <linux/page-flags.h>
17 #include <linux/smp.h>
18 
19 #include <asm/cacheflush.h>
20 #include <asm/cpu_ops.h>
21 #include <asm/memory.h>
22 #include <asm/mmu.h>
23 #include <asm/mmu_context.h>
24 #include <asm/page.h>
25 
26 #include "cpu-reset.h"
27 
28 /* Global variables for the arm64_relocate_new_kernel routine. */
29 extern const unsigned char arm64_relocate_new_kernel[];
30 extern const unsigned long arm64_relocate_new_kernel_size;
31 
32 /**
33  * kexec_image_info - For debugging output.
34  */
35 #define kexec_image_info(_i) _kexec_image_info(__func__, __LINE__, _i)
36 static void _kexec_image_info(const char *func, int line,
37 	const struct kimage *kimage)
38 {
39 	unsigned long i;
40 
41 	pr_debug("%s:%d:\n", func, line);
42 	pr_debug("  kexec kimage info:\n");
43 	pr_debug("    type:        %d\n", kimage->type);
44 	pr_debug("    start:       %lx\n", kimage->start);
45 	pr_debug("    head:        %lx\n", kimage->head);
46 	pr_debug("    nr_segments: %lu\n", kimage->nr_segments);
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 /**
64  * machine_kexec_prepare - Prepare for a kexec reboot.
65  *
66  * Called from the core kexec code when a kernel image is loaded.
67  * Forbid loading a kexec kernel if we have no way of hotplugging cpus or cpus
68  * are stuck in the kernel. This avoids a panic once we hit machine_kexec().
69  */
70 int machine_kexec_prepare(struct kimage *kimage)
71 {
72 	kexec_image_info(kimage);
73 
74 	if (kimage->type != KEXEC_TYPE_CRASH && cpus_are_stuck_in_kernel()) {
75 		pr_err("Can't kexec: CPUs are stuck in the kernel.\n");
76 		return -EBUSY;
77 	}
78 
79 	return 0;
80 }
81 
82 /**
83  * kexec_list_flush - Helper to flush the kimage list and source pages to PoC.
84  */
85 static void kexec_list_flush(struct kimage *kimage)
86 {
87 	kimage_entry_t *entry;
88 
89 	for (entry = &kimage->head; ; entry++) {
90 		unsigned int flag;
91 		void *addr;
92 
93 		/* flush the list entries. */
94 		__flush_dcache_area(entry, sizeof(kimage_entry_t));
95 
96 		flag = *entry & IND_FLAGS;
97 		if (flag == IND_DONE)
98 			break;
99 
100 		addr = phys_to_virt(*entry & PAGE_MASK);
101 
102 		switch (flag) {
103 		case IND_INDIRECTION:
104 			/* Set entry point just before the new list page. */
105 			entry = (kimage_entry_t *)addr - 1;
106 			break;
107 		case IND_SOURCE:
108 			/* flush the source pages. */
109 			__flush_dcache_area(addr, PAGE_SIZE);
110 			break;
111 		case IND_DESTINATION:
112 			break;
113 		default:
114 			BUG();
115 		}
116 	}
117 }
118 
119 /**
120  * kexec_segment_flush - Helper to flush the kimage segments to PoC.
121  */
122 static void kexec_segment_flush(const struct kimage *kimage)
123 {
124 	unsigned long i;
125 
126 	pr_debug("%s:\n", __func__);
127 
128 	for (i = 0; i < kimage->nr_segments; i++) {
129 		pr_debug("  segment[%lu]: %016lx - %016lx, 0x%lx bytes, %lu pages\n",
130 			i,
131 			kimage->segment[i].mem,
132 			kimage->segment[i].mem + kimage->segment[i].memsz,
133 			kimage->segment[i].memsz,
134 			kimage->segment[i].memsz /  PAGE_SIZE);
135 
136 		__flush_dcache_area(phys_to_virt(kimage->segment[i].mem),
137 			kimage->segment[i].memsz);
138 	}
139 }
140 
141 /**
142  * machine_kexec - Do the kexec reboot.
143  *
144  * Called from the core kexec code for a sys_reboot with LINUX_REBOOT_CMD_KEXEC.
145  */
146 void machine_kexec(struct kimage *kimage)
147 {
148 	phys_addr_t reboot_code_buffer_phys;
149 	void *reboot_code_buffer;
150 	bool in_kexec_crash = (kimage == kexec_crash_image);
151 	bool stuck_cpus = cpus_are_stuck_in_kernel();
152 
153 	/*
154 	 * New cpus may have become stuck_in_kernel after we loaded the image.
155 	 */
156 	BUG_ON(!in_kexec_crash && (stuck_cpus || (num_online_cpus() > 1)));
157 	WARN(in_kexec_crash && (stuck_cpus || smp_crash_stop_failed()),
158 		"Some CPUs may be stale, kdump will be unreliable.\n");
159 
160 	reboot_code_buffer_phys = page_to_phys(kimage->control_code_page);
161 	reboot_code_buffer = phys_to_virt(reboot_code_buffer_phys);
162 
163 	kexec_image_info(kimage);
164 
165 	pr_debug("%s:%d: control_code_page:        %p\n", __func__, __LINE__,
166 		kimage->control_code_page);
167 	pr_debug("%s:%d: reboot_code_buffer_phys:  %pa\n", __func__, __LINE__,
168 		&reboot_code_buffer_phys);
169 	pr_debug("%s:%d: reboot_code_buffer:       %p\n", __func__, __LINE__,
170 		reboot_code_buffer);
171 	pr_debug("%s:%d: relocate_new_kernel:      %p\n", __func__, __LINE__,
172 		arm64_relocate_new_kernel);
173 	pr_debug("%s:%d: relocate_new_kernel_size: 0x%lx(%lu) bytes\n",
174 		__func__, __LINE__, arm64_relocate_new_kernel_size,
175 		arm64_relocate_new_kernel_size);
176 
177 	/*
178 	 * Copy arm64_relocate_new_kernel to the reboot_code_buffer for use
179 	 * after the kernel is shut down.
180 	 */
181 	memcpy(reboot_code_buffer, arm64_relocate_new_kernel,
182 		arm64_relocate_new_kernel_size);
183 
184 	/* Flush the reboot_code_buffer in preparation for its execution. */
185 	__flush_dcache_area(reboot_code_buffer, arm64_relocate_new_kernel_size);
186 	flush_icache_range((uintptr_t)reboot_code_buffer,
187 		arm64_relocate_new_kernel_size);
188 
189 	/* Flush the kimage list and its buffers. */
190 	kexec_list_flush(kimage);
191 
192 	/* Flush the new image if already in place. */
193 	if ((kimage != kexec_crash_image) && (kimage->head & IND_DONE))
194 		kexec_segment_flush(kimage);
195 
196 	pr_info("Bye!\n");
197 
198 	/* Disable all DAIF exceptions. */
199 	asm volatile ("msr daifset, #0xf" : : : "memory");
200 
201 	/*
202 	 * cpu_soft_restart will shutdown the MMU, disable data caches, then
203 	 * transfer control to the reboot_code_buffer which contains a copy of
204 	 * the arm64_relocate_new_kernel routine.  arm64_relocate_new_kernel
205 	 * uses physical addressing to relocate the new image to its final
206 	 * position and transfers control to the image entry point when the
207 	 * relocation is complete.
208 	 */
209 
210 	cpu_soft_restart(kimage != kexec_crash_image,
211 		reboot_code_buffer_phys, kimage->head, kimage->start, 0);
212 
213 	BUG(); /* Should never get here. */
214 }
215 
216 static void machine_kexec_mask_interrupts(void)
217 {
218 	unsigned int i;
219 	struct irq_desc *desc;
220 
221 	for_each_irq_desc(i, desc) {
222 		struct irq_chip *chip;
223 		int ret;
224 
225 		chip = irq_desc_get_chip(desc);
226 		if (!chip)
227 			continue;
228 
229 		/*
230 		 * First try to remove the active state. If this
231 		 * fails, try to EOI the interrupt.
232 		 */
233 		ret = irq_set_irqchip_state(i, IRQCHIP_STATE_ACTIVE, false);
234 
235 		if (ret && irqd_irq_inprogress(&desc->irq_data) &&
236 		    chip->irq_eoi)
237 			chip->irq_eoi(&desc->irq_data);
238 
239 		if (chip->irq_mask)
240 			chip->irq_mask(&desc->irq_data);
241 
242 		if (chip->irq_disable && !irqd_irq_disabled(&desc->irq_data))
243 			chip->irq_disable(&desc->irq_data);
244 	}
245 }
246 
247 /**
248  * machine_crash_shutdown - shutdown non-crashing cpus and save registers
249  */
250 void machine_crash_shutdown(struct pt_regs *regs)
251 {
252 	local_irq_disable();
253 
254 	/* shutdown non-crashing cpus */
255 	smp_send_crash_stop();
256 
257 	/* for crashing cpu */
258 	crash_save_cpu(regs, smp_processor_id());
259 	machine_kexec_mask_interrupts();
260 
261 	pr_info("Starting crashdump kernel...\n");
262 }
263 
264 void arch_kexec_protect_crashkres(void)
265 {
266 	int i;
267 
268 	kexec_segment_flush(kexec_crash_image);
269 
270 	for (i = 0; i < kexec_crash_image->nr_segments; i++)
271 		set_memory_valid(
272 			__phys_to_virt(kexec_crash_image->segment[i].mem),
273 			kexec_crash_image->segment[i].memsz >> PAGE_SHIFT, 0);
274 }
275 
276 void arch_kexec_unprotect_crashkres(void)
277 {
278 	int i;
279 
280 	for (i = 0; i < kexec_crash_image->nr_segments; i++)
281 		set_memory_valid(
282 			__phys_to_virt(kexec_crash_image->segment[i].mem),
283 			kexec_crash_image->segment[i].memsz >> PAGE_SHIFT, 1);
284 }
285 
286 #ifdef CONFIG_HIBERNATION
287 /*
288  * To preserve the crash dump kernel image, the relevant memory segments
289  * should be mapped again around the hibernation.
290  */
291 void crash_prepare_suspend(void)
292 {
293 	if (kexec_crash_image)
294 		arch_kexec_unprotect_crashkres();
295 }
296 
297 void crash_post_resume(void)
298 {
299 	if (kexec_crash_image)
300 		arch_kexec_protect_crashkres();
301 }
302 
303 /*
304  * crash_is_nosave
305  *
306  * Return true only if a page is part of reserved memory for crash dump kernel,
307  * but does not hold any data of loaded kernel image.
308  *
309  * Note that all the pages in crash dump kernel memory have been initially
310  * marked as Reserved in kexec_reserve_crashkres_pages().
311  *
312  * In hibernation, the pages which are Reserved and yet "nosave" are excluded
313  * from the hibernation iamge. crash_is_nosave() does thich check for crash
314  * dump kernel and will reduce the total size of hibernation image.
315  */
316 
317 bool crash_is_nosave(unsigned long pfn)
318 {
319 	int i;
320 	phys_addr_t addr;
321 
322 	if (!crashk_res.end)
323 		return false;
324 
325 	/* in reserved memory? */
326 	addr = __pfn_to_phys(pfn);
327 	if ((addr < crashk_res.start) || (crashk_res.end < addr))
328 		return false;
329 
330 	if (!kexec_crash_image)
331 		return true;
332 
333 	/* not part of loaded kernel image? */
334 	for (i = 0; i < kexec_crash_image->nr_segments; i++)
335 		if (addr >= kexec_crash_image->segment[i].mem &&
336 				addr < (kexec_crash_image->segment[i].mem +
337 					kexec_crash_image->segment[i].memsz))
338 			return false;
339 
340 	return true;
341 }
342 
343 void crash_free_reserved_phys_range(unsigned long begin, unsigned long end)
344 {
345 	unsigned long addr;
346 	struct page *page;
347 
348 	for (addr = begin; addr < end; addr += PAGE_SIZE) {
349 		page = phys_to_page(addr);
350 		ClearPageReserved(page);
351 		free_reserved_page(page);
352 	}
353 }
354 #endif /* CONFIG_HIBERNATION */
355 
356 void arch_crash_save_vmcoreinfo(void)
357 {
358 	VMCOREINFO_NUMBER(VA_BITS);
359 	/* Please note VMCOREINFO_NUMBER() uses "%d", not "%x" */
360 	vmcoreinfo_append_str("NUMBER(kimage_voffset)=0x%llx\n",
361 						kimage_voffset);
362 	vmcoreinfo_append_str("NUMBER(PHYS_OFFSET)=0x%llx\n",
363 						PHYS_OFFSET);
364 }
365