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 #include <asm/sections.h>
25 #include <asm/trans_pgd.h>
26 
27 /**
28  * kexec_image_info - For debugging output.
29  */
30 #define kexec_image_info(_i) _kexec_image_info(__func__, __LINE__, _i)
31 static void _kexec_image_info(const char *func, int line,
32 	const struct kimage *kimage)
33 {
34 	unsigned long i;
35 
36 	pr_debug("%s:%d:\n", func, line);
37 	pr_debug("  kexec kimage info:\n");
38 	pr_debug("    type:        %d\n", kimage->type);
39 	pr_debug("    start:       %lx\n", kimage->start);
40 	pr_debug("    head:        %lx\n", kimage->head);
41 	pr_debug("    nr_segments: %lu\n", kimage->nr_segments);
42 	pr_debug("    dtb_mem: %pa\n", &kimage->arch.dtb_mem);
43 	pr_debug("    kern_reloc: %pa\n", &kimage->arch.kern_reloc);
44 	pr_debug("    el2_vectors: %pa\n", &kimage->arch.el2_vectors);
45 
46 	for (i = 0; i < kimage->nr_segments; i++) {
47 		pr_debug("      segment[%lu]: %016lx - %016lx, 0x%lx bytes, %lu pages\n",
48 			i,
49 			kimage->segment[i].mem,
50 			kimage->segment[i].mem + kimage->segment[i].memsz,
51 			kimage->segment[i].memsz,
52 			kimage->segment[i].memsz /  PAGE_SIZE);
53 	}
54 }
55 
56 void machine_kexec_cleanup(struct kimage *kimage)
57 {
58 	/* Empty routine needed to avoid build errors. */
59 }
60 
61 /**
62  * machine_kexec_prepare - Prepare for a kexec reboot.
63  *
64  * Called from the core kexec code when a kernel image is loaded.
65  * Forbid loading a kexec kernel if we have no way of hotplugging cpus or cpus
66  * are stuck in the kernel. This avoids a panic once we hit machine_kexec().
67  */
68 int machine_kexec_prepare(struct kimage *kimage)
69 {
70 	if (kimage->type != KEXEC_TYPE_CRASH && cpus_are_stuck_in_kernel()) {
71 		pr_err("Can't kexec: CPUs are stuck in the kernel.\n");
72 		return -EBUSY;
73 	}
74 
75 	return 0;
76 }
77 
78 /**
79  * kexec_segment_flush - Helper to flush the kimage segments to PoC.
80  */
81 static void kexec_segment_flush(const struct kimage *kimage)
82 {
83 	unsigned long i;
84 
85 	pr_debug("%s:\n", __func__);
86 
87 	for (i = 0; i < kimage->nr_segments; i++) {
88 		pr_debug("  segment[%lu]: %016lx - %016lx, 0x%lx bytes, %lu pages\n",
89 			i,
90 			kimage->segment[i].mem,
91 			kimage->segment[i].mem + kimage->segment[i].memsz,
92 			kimage->segment[i].memsz,
93 			kimage->segment[i].memsz /  PAGE_SIZE);
94 
95 		dcache_clean_inval_poc(
96 			(unsigned long)phys_to_virt(kimage->segment[i].mem),
97 			(unsigned long)phys_to_virt(kimage->segment[i].mem) +
98 				kimage->segment[i].memsz);
99 	}
100 }
101 
102 /* Allocates pages for kexec page table */
103 static void *kexec_page_alloc(void *arg)
104 {
105 	struct kimage *kimage = (struct kimage *)arg;
106 	struct page *page = kimage_alloc_control_pages(kimage, 0);
107 	void *vaddr = NULL;
108 
109 	if (!page)
110 		return NULL;
111 
112 	vaddr = page_address(page);
113 	memset(vaddr, 0, PAGE_SIZE);
114 
115 	return vaddr;
116 }
117 
118 int machine_kexec_post_load(struct kimage *kimage)
119 {
120 	int rc;
121 	pgd_t *trans_pgd;
122 	void *reloc_code = page_to_virt(kimage->control_code_page);
123 	long reloc_size;
124 	struct trans_pgd_info info = {
125 		.trans_alloc_page	= kexec_page_alloc,
126 		.trans_alloc_arg	= kimage,
127 	};
128 
129 	/* If in place, relocation is not used, only flush next kernel */
130 	if (kimage->head & IND_DONE) {
131 		kexec_segment_flush(kimage);
132 		kexec_image_info(kimage);
133 		return 0;
134 	}
135 
136 	kimage->arch.el2_vectors = 0;
137 	if (is_hyp_nvhe()) {
138 		rc = trans_pgd_copy_el2_vectors(&info,
139 						&kimage->arch.el2_vectors);
140 		if (rc)
141 			return rc;
142 	}
143 
144 	/* Create a copy of the linear map */
145 	trans_pgd = kexec_page_alloc(kimage);
146 	if (!trans_pgd)
147 		return -ENOMEM;
148 	rc = trans_pgd_create_copy(&info, &trans_pgd, PAGE_OFFSET, PAGE_END);
149 	if (rc)
150 		return rc;
151 	kimage->arch.ttbr1 = __pa(trans_pgd);
152 	kimage->arch.zero_page = __pa_symbol(empty_zero_page);
153 
154 	reloc_size = __relocate_new_kernel_end - __relocate_new_kernel_start;
155 	memcpy(reloc_code, __relocate_new_kernel_start, reloc_size);
156 	kimage->arch.kern_reloc = __pa(reloc_code);
157 	rc = trans_pgd_idmap_page(&info, &kimage->arch.ttbr0,
158 				  &kimage->arch.t0sz, reloc_code);
159 	if (rc)
160 		return rc;
161 	kimage->arch.phys_offset = virt_to_phys(kimage) - (long)kimage;
162 
163 	/* Flush the reloc_code in preparation for its execution. */
164 	dcache_clean_inval_poc((unsigned long)reloc_code,
165 			       (unsigned long)reloc_code + reloc_size);
166 	icache_inval_pou((uintptr_t)reloc_code,
167 			 (uintptr_t)reloc_code + reloc_size);
168 	kexec_image_info(kimage);
169 
170 	return 0;
171 }
172 
173 /**
174  * machine_kexec - Do the kexec reboot.
175  *
176  * Called from the core kexec code for a sys_reboot with LINUX_REBOOT_CMD_KEXEC.
177  */
178 void machine_kexec(struct kimage *kimage)
179 {
180 	bool in_kexec_crash = (kimage == kexec_crash_image);
181 	bool stuck_cpus = cpus_are_stuck_in_kernel();
182 
183 	/*
184 	 * New cpus may have become stuck_in_kernel after we loaded the image.
185 	 */
186 	BUG_ON(!in_kexec_crash && (stuck_cpus || (num_online_cpus() > 1)));
187 	WARN(in_kexec_crash && (stuck_cpus || smp_crash_stop_failed()),
188 		"Some CPUs may be stale, kdump will be unreliable.\n");
189 
190 	pr_info("Bye!\n");
191 
192 	local_daif_mask();
193 
194 	/*
195 	 * Both restart and kernel_reloc will shutdown the MMU, disable data
196 	 * caches. However, restart will start new kernel or purgatory directly,
197 	 * kernel_reloc contains the body of arm64_relocate_new_kernel
198 	 * In kexec case, kimage->start points to purgatory assuming that
199 	 * kernel entry and dtb address are embedded in purgatory by
200 	 * userspace (kexec-tools).
201 	 * In kexec_file case, the kernel starts directly without purgatory.
202 	 */
203 	if (kimage->head & IND_DONE) {
204 		typeof(cpu_soft_restart) *restart;
205 
206 		cpu_install_idmap();
207 		restart = (void *)__pa_symbol(function_nocfi(cpu_soft_restart));
208 		restart(is_hyp_nvhe(), kimage->start, kimage->arch.dtb_mem,
209 			0, 0);
210 	} else {
211 		void (*kernel_reloc)(struct kimage *kimage);
212 
213 		if (is_hyp_nvhe())
214 			__hyp_set_vectors(kimage->arch.el2_vectors);
215 		cpu_install_ttbr0(kimage->arch.ttbr0, kimage->arch.t0sz);
216 		kernel_reloc = (void *)kimage->arch.kern_reloc;
217 		kernel_reloc(kimage);
218 	}
219 
220 	BUG(); /* Should never get here. */
221 }
222 
223 static void machine_kexec_mask_interrupts(void)
224 {
225 	unsigned int i;
226 	struct irq_desc *desc;
227 
228 	for_each_irq_desc(i, desc) {
229 		struct irq_chip *chip;
230 		int ret;
231 
232 		chip = irq_desc_get_chip(desc);
233 		if (!chip)
234 			continue;
235 
236 		/*
237 		 * First try to remove the active state. If this
238 		 * fails, try to EOI the interrupt.
239 		 */
240 		ret = irq_set_irqchip_state(i, IRQCHIP_STATE_ACTIVE, false);
241 
242 		if (ret && irqd_irq_inprogress(&desc->irq_data) &&
243 		    chip->irq_eoi)
244 			chip->irq_eoi(&desc->irq_data);
245 
246 		if (chip->irq_mask)
247 			chip->irq_mask(&desc->irq_data);
248 
249 		if (chip->irq_disable && !irqd_irq_disabled(&desc->irq_data))
250 			chip->irq_disable(&desc->irq_data);
251 	}
252 }
253 
254 /**
255  * machine_crash_shutdown - shutdown non-crashing cpus and save registers
256  */
257 void machine_crash_shutdown(struct pt_regs *regs)
258 {
259 	local_irq_disable();
260 
261 	/* shutdown non-crashing cpus */
262 	crash_smp_send_stop();
263 
264 	/* for crashing cpu */
265 	crash_save_cpu(regs, smp_processor_id());
266 	machine_kexec_mask_interrupts();
267 
268 	pr_info("Starting crashdump kernel...\n");
269 }
270 
271 void arch_kexec_protect_crashkres(void)
272 {
273 	int i;
274 
275 	for (i = 0; i < kexec_crash_image->nr_segments; i++)
276 		set_memory_valid(
277 			__phys_to_virt(kexec_crash_image->segment[i].mem),
278 			kexec_crash_image->segment[i].memsz >> PAGE_SHIFT, 0);
279 }
280 
281 void arch_kexec_unprotect_crashkres(void)
282 {
283 	int i;
284 
285 	for (i = 0; i < kexec_crash_image->nr_segments; i++)
286 		set_memory_valid(
287 			__phys_to_virt(kexec_crash_image->segment[i].mem),
288 			kexec_crash_image->segment[i].memsz >> PAGE_SHIFT, 1);
289 }
290 
291 #ifdef CONFIG_HIBERNATION
292 /*
293  * To preserve the crash dump kernel image, the relevant memory segments
294  * should be mapped again around the hibernation.
295  */
296 void crash_prepare_suspend(void)
297 {
298 	if (kexec_crash_image)
299 		arch_kexec_unprotect_crashkres();
300 }
301 
302 void crash_post_resume(void)
303 {
304 	if (kexec_crash_image)
305 		arch_kexec_protect_crashkres();
306 }
307 
308 /*
309  * crash_is_nosave
310  *
311  * Return true only if a page is part of reserved memory for crash dump kernel,
312  * but does not hold any data of loaded kernel image.
313  *
314  * Note that all the pages in crash dump kernel memory have been initially
315  * marked as Reserved as memory was allocated via memblock_reserve().
316  *
317  * In hibernation, the pages which are Reserved and yet "nosave" are excluded
318  * from the hibernation iamge. crash_is_nosave() does thich check for crash
319  * dump kernel and will reduce the total size of hibernation image.
320  */
321 
322 bool crash_is_nosave(unsigned long pfn)
323 {
324 	int i;
325 	phys_addr_t addr;
326 
327 	if (!crashk_res.end)
328 		return false;
329 
330 	/* in reserved memory? */
331 	addr = __pfn_to_phys(pfn);
332 	if ((addr < crashk_res.start) || (crashk_res.end < addr)) {
333 		if (!crashk_low_res.end)
334 			return false;
335 
336 		if ((addr < crashk_low_res.start) || (crashk_low_res.end < addr))
337 			return false;
338 	}
339 
340 	if (!kexec_crash_image)
341 		return true;
342 
343 	/* not part of loaded kernel image? */
344 	for (i = 0; i < kexec_crash_image->nr_segments; i++)
345 		if (addr >= kexec_crash_image->segment[i].mem &&
346 				addr < (kexec_crash_image->segment[i].mem +
347 					kexec_crash_image->segment[i].memsz))
348 			return false;
349 
350 	return true;
351 }
352 
353 void crash_free_reserved_phys_range(unsigned long begin, unsigned long end)
354 {
355 	unsigned long addr;
356 	struct page *page;
357 
358 	for (addr = begin; addr < end; addr += PAGE_SIZE) {
359 		page = phys_to_page(addr);
360 		free_reserved_page(page);
361 	}
362 }
363 #endif /* CONFIG_HIBERNATION */
364