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/reboot.h>
15 #include <linux/set_memory.h>
16 #include <linux/smp.h>
17
18 #include <asm/cacheflush.h>
19 #include <asm/cpu_ops.h>
20 #include <asm/daifflags.h>
21 #include <asm/memory.h>
22 #include <asm/mmu.h>
23 #include <asm/mmu_context.h>
24 #include <asm/page.h>
25 #include <asm/sections.h>
26 #include <asm/trans_pgd.h>
27
28 /**
29 * kexec_image_info - For debugging output.
30 */
31 #define kexec_image_info(_i) _kexec_image_info(__func__, __LINE__, _i)
_kexec_image_info(const char * func,int line,const struct kimage * kimage)32 static void _kexec_image_info(const char *func, int line,
33 const struct kimage *kimage)
34 {
35 unsigned long i;
36
37 pr_debug("%s:%d:\n", func, line);
38 pr_debug(" kexec kimage info:\n");
39 pr_debug(" type: %d\n", kimage->type);
40 pr_debug(" start: %lx\n", kimage->start);
41 pr_debug(" head: %lx\n", kimage->head);
42 pr_debug(" nr_segments: %lu\n", kimage->nr_segments);
43 pr_debug(" dtb_mem: %pa\n", &kimage->arch.dtb_mem);
44 pr_debug(" kern_reloc: %pa\n", &kimage->arch.kern_reloc);
45 pr_debug(" el2_vectors: %pa\n", &kimage->arch.el2_vectors);
46
47 for (i = 0; i < kimage->nr_segments; i++) {
48 pr_debug(" segment[%lu]: %016lx - %016lx, 0x%lx bytes, %lu pages\n",
49 i,
50 kimage->segment[i].mem,
51 kimage->segment[i].mem + kimage->segment[i].memsz,
52 kimage->segment[i].memsz,
53 kimage->segment[i].memsz / PAGE_SIZE);
54 }
55 }
56
machine_kexec_cleanup(struct kimage * kimage)57 void machine_kexec_cleanup(struct kimage *kimage)
58 {
59 /* Empty routine needed to avoid build errors. */
60 }
61
62 /**
63 * machine_kexec_prepare - Prepare for a kexec reboot.
64 *
65 * Called from the core kexec code when a kernel image is loaded.
66 * Forbid loading a kexec kernel if we have no way of hotplugging cpus or cpus
67 * are stuck in the kernel. This avoids a panic once we hit machine_kexec().
68 */
machine_kexec_prepare(struct kimage * kimage)69 int machine_kexec_prepare(struct kimage *kimage)
70 {
71 if (kimage->type != KEXEC_TYPE_CRASH && cpus_are_stuck_in_kernel()) {
72 pr_err("Can't kexec: CPUs are stuck in the kernel.\n");
73 return -EBUSY;
74 }
75
76 return 0;
77 }
78
79 /**
80 * kexec_segment_flush - Helper to flush the kimage segments to PoC.
81 */
kexec_segment_flush(const struct kimage * kimage)82 static void kexec_segment_flush(const struct kimage *kimage)
83 {
84 unsigned long i;
85
86 pr_debug("%s:\n", __func__);
87
88 for (i = 0; i < kimage->nr_segments; i++) {
89 pr_debug(" segment[%lu]: %016lx - %016lx, 0x%lx bytes, %lu pages\n",
90 i,
91 kimage->segment[i].mem,
92 kimage->segment[i].mem + kimage->segment[i].memsz,
93 kimage->segment[i].memsz,
94 kimage->segment[i].memsz / PAGE_SIZE);
95
96 dcache_clean_inval_poc(
97 (unsigned long)phys_to_virt(kimage->segment[i].mem),
98 (unsigned long)phys_to_virt(kimage->segment[i].mem) +
99 kimage->segment[i].memsz);
100 }
101 }
102
103 /* Allocates pages for kexec page table */
kexec_page_alloc(void * arg)104 static void *kexec_page_alloc(void *arg)
105 {
106 struct kimage *kimage = arg;
107 struct page *page = kimage_alloc_control_pages(kimage, 0);
108 void *vaddr = NULL;
109
110 if (!page)
111 return NULL;
112
113 vaddr = page_address(page);
114 memset(vaddr, 0, PAGE_SIZE);
115
116 return vaddr;
117 }
118
machine_kexec_post_load(struct kimage * kimage)119 int machine_kexec_post_load(struct kimage *kimage)
120 {
121 int rc;
122 pgd_t *trans_pgd;
123 void *reloc_code = page_to_virt(kimage->control_code_page);
124 long reloc_size;
125 struct trans_pgd_info info = {
126 .trans_alloc_page = kexec_page_alloc,
127 .trans_alloc_arg = kimage,
128 };
129
130 /* If in place, relocation is not used, only flush next kernel */
131 if (kimage->head & IND_DONE) {
132 kexec_segment_flush(kimage);
133 kexec_image_info(kimage);
134 return 0;
135 }
136
137 kimage->arch.el2_vectors = 0;
138 if (is_hyp_nvhe()) {
139 rc = trans_pgd_copy_el2_vectors(&info,
140 &kimage->arch.el2_vectors);
141 if (rc)
142 return rc;
143 }
144
145 /* Create a copy of the linear map */
146 trans_pgd = kexec_page_alloc(kimage);
147 if (!trans_pgd)
148 return -ENOMEM;
149 rc = trans_pgd_create_copy(&info, &trans_pgd, PAGE_OFFSET, PAGE_END);
150 if (rc)
151 return rc;
152 kimage->arch.ttbr1 = __pa(trans_pgd);
153 kimage->arch.zero_page = __pa_symbol(empty_zero_page);
154
155 reloc_size = __relocate_new_kernel_end - __relocate_new_kernel_start;
156 memcpy(reloc_code, __relocate_new_kernel_start, reloc_size);
157 kimage->arch.kern_reloc = __pa(reloc_code);
158 rc = trans_pgd_idmap_page(&info, &kimage->arch.ttbr0,
159 &kimage->arch.t0sz, reloc_code);
160 if (rc)
161 return rc;
162 kimage->arch.phys_offset = virt_to_phys(kimage) - (long)kimage;
163
164 /* Flush the reloc_code in preparation for its execution. */
165 dcache_clean_inval_poc((unsigned long)reloc_code,
166 (unsigned long)reloc_code + reloc_size);
167 icache_inval_pou((uintptr_t)reloc_code,
168 (uintptr_t)reloc_code + reloc_size);
169 kexec_image_info(kimage);
170
171 return 0;
172 }
173
174 /**
175 * machine_kexec - Do the kexec reboot.
176 *
177 * Called from the core kexec code for a sys_reboot with LINUX_REBOOT_CMD_KEXEC.
178 */
machine_kexec(struct kimage * kimage)179 void machine_kexec(struct kimage *kimage)
180 {
181 bool in_kexec_crash = (kimage == kexec_crash_image);
182 bool stuck_cpus = cpus_are_stuck_in_kernel();
183
184 /*
185 * New cpus may have become stuck_in_kernel after we loaded the image.
186 */
187 BUG_ON(!in_kexec_crash && (stuck_cpus || (num_online_cpus() > 1)));
188 WARN(in_kexec_crash && (stuck_cpus || smp_crash_stop_failed()),
189 "Some CPUs may be stale, kdump will be unreliable.\n");
190
191 pr_info("Bye!\n");
192
193 local_daif_mask();
194
195 /*
196 * Both restart and kernel_reloc will shutdown the MMU, disable data
197 * caches. However, restart will start new kernel or purgatory directly,
198 * kernel_reloc contains the body of arm64_relocate_new_kernel
199 * In kexec case, kimage->start points to purgatory assuming that
200 * kernel entry and dtb address are embedded in purgatory by
201 * userspace (kexec-tools).
202 * In kexec_file case, the kernel starts directly without purgatory.
203 */
204 if (kimage->head & IND_DONE) {
205 typeof(cpu_soft_restart) *restart;
206
207 cpu_install_idmap();
208 restart = (void *)__pa_symbol(cpu_soft_restart);
209 restart(is_hyp_nvhe(), kimage->start, kimage->arch.dtb_mem,
210 0, 0);
211 } else {
212 void (*kernel_reloc)(struct kimage *kimage);
213
214 if (is_hyp_nvhe())
215 __hyp_set_vectors(kimage->arch.el2_vectors);
216 cpu_install_ttbr0(kimage->arch.ttbr0, kimage->arch.t0sz);
217 kernel_reloc = (void *)kimage->arch.kern_reloc;
218 kernel_reloc(kimage);
219 }
220
221 BUG(); /* Should never get here. */
222 }
223
machine_kexec_mask_interrupts(void)224 static void machine_kexec_mask_interrupts(void)
225 {
226 unsigned int i;
227 struct irq_desc *desc;
228
229 for_each_irq_desc(i, desc) {
230 struct irq_chip *chip;
231 int ret;
232
233 chip = irq_desc_get_chip(desc);
234 if (!chip)
235 continue;
236
237 /*
238 * First try to remove the active state. If this
239 * fails, try to EOI the interrupt.
240 */
241 ret = irq_set_irqchip_state(i, IRQCHIP_STATE_ACTIVE, false);
242
243 if (ret && irqd_irq_inprogress(&desc->irq_data) &&
244 chip->irq_eoi)
245 chip->irq_eoi(&desc->irq_data);
246
247 if (chip->irq_mask)
248 chip->irq_mask(&desc->irq_data);
249
250 if (chip->irq_disable && !irqd_irq_disabled(&desc->irq_data))
251 chip->irq_disable(&desc->irq_data);
252 }
253 }
254
255 /**
256 * machine_crash_shutdown - shutdown non-crashing cpus and save registers
257 */
machine_crash_shutdown(struct pt_regs * regs)258 void machine_crash_shutdown(struct pt_regs *regs)
259 {
260 local_irq_disable();
261
262 /* shutdown non-crashing cpus */
263 crash_smp_send_stop();
264
265 /* for crashing cpu */
266 crash_save_cpu(regs, smp_processor_id());
267 machine_kexec_mask_interrupts();
268
269 pr_info("Starting crashdump kernel...\n");
270 }
271
272 #ifdef CONFIG_HIBERNATION
273 /*
274 * To preserve the crash dump kernel image, the relevant memory segments
275 * should be mapped again around the hibernation.
276 */
crash_prepare_suspend(void)277 void crash_prepare_suspend(void)
278 {
279 if (kexec_crash_image)
280 arch_kexec_unprotect_crashkres();
281 }
282
crash_post_resume(void)283 void crash_post_resume(void)
284 {
285 if (kexec_crash_image)
286 arch_kexec_protect_crashkres();
287 }
288
289 /*
290 * crash_is_nosave
291 *
292 * Return true only if a page is part of reserved memory for crash dump kernel,
293 * but does not hold any data of loaded kernel image.
294 *
295 * Note that all the pages in crash dump kernel memory have been initially
296 * marked as Reserved as memory was allocated via memblock_reserve().
297 *
298 * In hibernation, the pages which are Reserved and yet "nosave" are excluded
299 * from the hibernation iamge. crash_is_nosave() does thich check for crash
300 * dump kernel and will reduce the total size of hibernation image.
301 */
302
crash_is_nosave(unsigned long pfn)303 bool crash_is_nosave(unsigned long pfn)
304 {
305 int i;
306 phys_addr_t addr;
307
308 if (!crashk_res.end)
309 return false;
310
311 /* in reserved memory? */
312 addr = __pfn_to_phys(pfn);
313 if ((addr < crashk_res.start) || (crashk_res.end < addr)) {
314 if (!crashk_low_res.end)
315 return false;
316
317 if ((addr < crashk_low_res.start) || (crashk_low_res.end < addr))
318 return false;
319 }
320
321 if (!kexec_crash_image)
322 return true;
323
324 /* not part of loaded kernel image? */
325 for (i = 0; i < kexec_crash_image->nr_segments; i++)
326 if (addr >= kexec_crash_image->segment[i].mem &&
327 addr < (kexec_crash_image->segment[i].mem +
328 kexec_crash_image->segment[i].memsz))
329 return false;
330
331 return true;
332 }
333
crash_free_reserved_phys_range(unsigned long begin,unsigned long end)334 void crash_free_reserved_phys_range(unsigned long begin, unsigned long end)
335 {
336 unsigned long addr;
337 struct page *page;
338
339 for (addr = begin; addr < end; addr += PAGE_SIZE) {
340 page = phys_to_page(addr);
341 free_reserved_page(page);
342 }
343 }
344 #endif /* CONFIG_HIBERNATION */
345