xref: /linux/arch/x86/kernel/crash.c (revision 52338415)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Architecture specific (i386/x86_64) functions for kexec based crash dumps.
4  *
5  * Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
6  *
7  * Copyright (C) IBM Corporation, 2004. All rights reserved.
8  * Copyright (C) Red Hat Inc., 2014. All rights reserved.
9  * Authors:
10  *      Vivek Goyal <vgoyal@redhat.com>
11  *
12  */
13 
14 #define pr_fmt(fmt)	"kexec: " fmt
15 
16 #include <linux/types.h>
17 #include <linux/kernel.h>
18 #include <linux/smp.h>
19 #include <linux/reboot.h>
20 #include <linux/kexec.h>
21 #include <linux/delay.h>
22 #include <linux/elf.h>
23 #include <linux/elfcore.h>
24 #include <linux/export.h>
25 #include <linux/slab.h>
26 #include <linux/vmalloc.h>
27 
28 #include <asm/processor.h>
29 #include <asm/hardirq.h>
30 #include <asm/nmi.h>
31 #include <asm/hw_irq.h>
32 #include <asm/apic.h>
33 #include <asm/e820/types.h>
34 #include <asm/io_apic.h>
35 #include <asm/hpet.h>
36 #include <linux/kdebug.h>
37 #include <asm/cpu.h>
38 #include <asm/reboot.h>
39 #include <asm/virtext.h>
40 #include <asm/intel_pt.h>
41 #include <asm/crash.h>
42 
43 /* Used while preparing memory map entries for second kernel */
44 struct crash_memmap_data {
45 	struct boot_params *params;
46 	/* Type of memory */
47 	unsigned int type;
48 };
49 
50 /*
51  * This is used to VMCLEAR all VMCSs loaded on the
52  * processor. And when loading kvm_intel module, the
53  * callback function pointer will be assigned.
54  *
55  * protected by rcu.
56  */
57 crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss = NULL;
58 EXPORT_SYMBOL_GPL(crash_vmclear_loaded_vmcss);
59 
60 static inline void cpu_crash_vmclear_loaded_vmcss(void)
61 {
62 	crash_vmclear_fn *do_vmclear_operation = NULL;
63 
64 	rcu_read_lock();
65 	do_vmclear_operation = rcu_dereference(crash_vmclear_loaded_vmcss);
66 	if (do_vmclear_operation)
67 		do_vmclear_operation();
68 	rcu_read_unlock();
69 }
70 
71 #if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)
72 
73 static void kdump_nmi_callback(int cpu, struct pt_regs *regs)
74 {
75 	crash_save_cpu(regs, cpu);
76 
77 	/*
78 	 * VMCLEAR VMCSs loaded on all cpus if needed.
79 	 */
80 	cpu_crash_vmclear_loaded_vmcss();
81 
82 	/* Disable VMX or SVM if needed.
83 	 *
84 	 * We need to disable virtualization on all CPUs.
85 	 * Having VMX or SVM enabled on any CPU may break rebooting
86 	 * after the kdump kernel has finished its task.
87 	 */
88 	cpu_emergency_vmxoff();
89 	cpu_emergency_svm_disable();
90 
91 	/*
92 	 * Disable Intel PT to stop its logging
93 	 */
94 	cpu_emergency_stop_pt();
95 
96 	disable_local_APIC();
97 }
98 
99 void kdump_nmi_shootdown_cpus(void)
100 {
101 	nmi_shootdown_cpus(kdump_nmi_callback);
102 
103 	disable_local_APIC();
104 }
105 
106 /* Override the weak function in kernel/panic.c */
107 void crash_smp_send_stop(void)
108 {
109 	static int cpus_stopped;
110 
111 	if (cpus_stopped)
112 		return;
113 
114 	if (smp_ops.crash_stop_other_cpus)
115 		smp_ops.crash_stop_other_cpus();
116 	else
117 		smp_send_stop();
118 
119 	cpus_stopped = 1;
120 }
121 
122 #else
123 void crash_smp_send_stop(void)
124 {
125 	/* There are no cpus to shootdown */
126 }
127 #endif
128 
129 void native_machine_crash_shutdown(struct pt_regs *regs)
130 {
131 	/* This function is only called after the system
132 	 * has panicked or is otherwise in a critical state.
133 	 * The minimum amount of code to allow a kexec'd kernel
134 	 * to run successfully needs to happen here.
135 	 *
136 	 * In practice this means shooting down the other cpus in
137 	 * an SMP system.
138 	 */
139 	/* The kernel is broken so disable interrupts */
140 	local_irq_disable();
141 
142 	crash_smp_send_stop();
143 
144 	/*
145 	 * VMCLEAR VMCSs loaded on this cpu if needed.
146 	 */
147 	cpu_crash_vmclear_loaded_vmcss();
148 
149 	/* Booting kdump kernel with VMX or SVM enabled won't work,
150 	 * because (among other limitations) we can't disable paging
151 	 * with the virt flags.
152 	 */
153 	cpu_emergency_vmxoff();
154 	cpu_emergency_svm_disable();
155 
156 	/*
157 	 * Disable Intel PT to stop its logging
158 	 */
159 	cpu_emergency_stop_pt();
160 
161 #ifdef CONFIG_X86_IO_APIC
162 	/* Prevent crash_kexec() from deadlocking on ioapic_lock. */
163 	ioapic_zap_locks();
164 	clear_IO_APIC();
165 #endif
166 	lapic_shutdown();
167 	restore_boot_irq_mode();
168 #ifdef CONFIG_HPET_TIMER
169 	hpet_disable();
170 #endif
171 	crash_save_cpu(regs, safe_smp_processor_id());
172 }
173 
174 #ifdef CONFIG_KEXEC_FILE
175 
176 static unsigned long crash_zero_bytes;
177 
178 static int get_nr_ram_ranges_callback(struct resource *res, void *arg)
179 {
180 	unsigned int *nr_ranges = arg;
181 
182 	(*nr_ranges)++;
183 	return 0;
184 }
185 
186 /* Gather all the required information to prepare elf headers for ram regions */
187 static struct crash_mem *fill_up_crash_elf_data(void)
188 {
189 	unsigned int nr_ranges = 0;
190 	struct crash_mem *cmem;
191 
192 	walk_system_ram_res(0, -1, &nr_ranges,
193 				get_nr_ram_ranges_callback);
194 	if (!nr_ranges)
195 		return NULL;
196 
197 	/*
198 	 * Exclusion of crash region and/or crashk_low_res may cause
199 	 * another range split. So add extra two slots here.
200 	 */
201 	nr_ranges += 2;
202 	cmem = vzalloc(struct_size(cmem, ranges, nr_ranges));
203 	if (!cmem)
204 		return NULL;
205 
206 	cmem->max_nr_ranges = nr_ranges;
207 	cmem->nr_ranges = 0;
208 
209 	return cmem;
210 }
211 
212 /*
213  * Look for any unwanted ranges between mstart, mend and remove them. This
214  * might lead to split and split ranges are put in cmem->ranges[] array
215  */
216 static int elf_header_exclude_ranges(struct crash_mem *cmem)
217 {
218 	int ret = 0;
219 
220 	/* Exclude crashkernel region */
221 	ret = crash_exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
222 	if (ret)
223 		return ret;
224 
225 	if (crashk_low_res.end) {
226 		ret = crash_exclude_mem_range(cmem, crashk_low_res.start,
227 							crashk_low_res.end);
228 	}
229 
230 	return ret;
231 }
232 
233 static int prepare_elf64_ram_headers_callback(struct resource *res, void *arg)
234 {
235 	struct crash_mem *cmem = arg;
236 
237 	cmem->ranges[cmem->nr_ranges].start = res->start;
238 	cmem->ranges[cmem->nr_ranges].end = res->end;
239 	cmem->nr_ranges++;
240 
241 	return 0;
242 }
243 
244 /* Prepare elf headers. Return addr and size */
245 static int prepare_elf_headers(struct kimage *image, void **addr,
246 					unsigned long *sz)
247 {
248 	struct crash_mem *cmem;
249 	Elf64_Ehdr *ehdr;
250 	Elf64_Phdr *phdr;
251 	int ret, i;
252 
253 	cmem = fill_up_crash_elf_data();
254 	if (!cmem)
255 		return -ENOMEM;
256 
257 	ret = walk_system_ram_res(0, -1, cmem,
258 				prepare_elf64_ram_headers_callback);
259 	if (ret)
260 		goto out;
261 
262 	/* Exclude unwanted mem ranges */
263 	ret = elf_header_exclude_ranges(cmem);
264 	if (ret)
265 		goto out;
266 
267 	/* By default prepare 64bit headers */
268 	ret =  crash_prepare_elf64_headers(cmem,
269 				IS_ENABLED(CONFIG_X86_64), addr, sz);
270 	if (ret)
271 		goto out;
272 
273 	/*
274 	 * If a range matches backup region, adjust offset to backup
275 	 * segment.
276 	 */
277 	ehdr = (Elf64_Ehdr *)*addr;
278 	phdr = (Elf64_Phdr *)(ehdr + 1);
279 	for (i = 0; i < ehdr->e_phnum; phdr++, i++)
280 		if (phdr->p_type == PT_LOAD &&
281 				phdr->p_paddr == image->arch.backup_src_start &&
282 				phdr->p_memsz == image->arch.backup_src_sz) {
283 			phdr->p_offset = image->arch.backup_load_addr;
284 			break;
285 		}
286 out:
287 	vfree(cmem);
288 	return ret;
289 }
290 
291 static int add_e820_entry(struct boot_params *params, struct e820_entry *entry)
292 {
293 	unsigned int nr_e820_entries;
294 
295 	nr_e820_entries = params->e820_entries;
296 	if (nr_e820_entries >= E820_MAX_ENTRIES_ZEROPAGE)
297 		return 1;
298 
299 	memcpy(&params->e820_table[nr_e820_entries], entry,
300 			sizeof(struct e820_entry));
301 	params->e820_entries++;
302 	return 0;
303 }
304 
305 static int memmap_entry_callback(struct resource *res, void *arg)
306 {
307 	struct crash_memmap_data *cmd = arg;
308 	struct boot_params *params = cmd->params;
309 	struct e820_entry ei;
310 
311 	ei.addr = res->start;
312 	ei.size = resource_size(res);
313 	ei.type = cmd->type;
314 	add_e820_entry(params, &ei);
315 
316 	return 0;
317 }
318 
319 static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
320 				 unsigned long long mstart,
321 				 unsigned long long mend)
322 {
323 	unsigned long start, end;
324 	int ret = 0;
325 
326 	cmem->ranges[0].start = mstart;
327 	cmem->ranges[0].end = mend;
328 	cmem->nr_ranges = 1;
329 
330 	/* Exclude Backup region */
331 	start = image->arch.backup_load_addr;
332 	end = start + image->arch.backup_src_sz - 1;
333 	ret = crash_exclude_mem_range(cmem, start, end);
334 	if (ret)
335 		return ret;
336 
337 	/* Exclude elf header region */
338 	start = image->arch.elf_load_addr;
339 	end = start + image->arch.elf_headers_sz - 1;
340 	return crash_exclude_mem_range(cmem, start, end);
341 }
342 
343 /* Prepare memory map for crash dump kernel */
344 int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
345 {
346 	int i, ret = 0;
347 	unsigned long flags;
348 	struct e820_entry ei;
349 	struct crash_memmap_data cmd;
350 	struct crash_mem *cmem;
351 
352 	cmem = vzalloc(sizeof(struct crash_mem));
353 	if (!cmem)
354 		return -ENOMEM;
355 
356 	memset(&cmd, 0, sizeof(struct crash_memmap_data));
357 	cmd.params = params;
358 
359 	/* Add first 640K segment */
360 	ei.addr = image->arch.backup_src_start;
361 	ei.size = image->arch.backup_src_sz;
362 	ei.type = E820_TYPE_RAM;
363 	add_e820_entry(params, &ei);
364 
365 	/* Add ACPI tables */
366 	cmd.type = E820_TYPE_ACPI;
367 	flags = IORESOURCE_MEM | IORESOURCE_BUSY;
368 	walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd,
369 		       memmap_entry_callback);
370 
371 	/* Add ACPI Non-volatile Storage */
372 	cmd.type = E820_TYPE_NVS;
373 	walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, &cmd,
374 			memmap_entry_callback);
375 
376 	/* Add e820 reserved ranges */
377 	cmd.type = E820_TYPE_RESERVED;
378 	flags = IORESOURCE_MEM;
379 	walk_iomem_res_desc(IORES_DESC_RESERVED, flags, 0, -1, &cmd,
380 			   memmap_entry_callback);
381 
382 	/* Add crashk_low_res region */
383 	if (crashk_low_res.end) {
384 		ei.addr = crashk_low_res.start;
385 		ei.size = crashk_low_res.end - crashk_low_res.start + 1;
386 		ei.type = E820_TYPE_RAM;
387 		add_e820_entry(params, &ei);
388 	}
389 
390 	/* Exclude some ranges from crashk_res and add rest to memmap */
391 	ret = memmap_exclude_ranges(image, cmem, crashk_res.start,
392 						crashk_res.end);
393 	if (ret)
394 		goto out;
395 
396 	for (i = 0; i < cmem->nr_ranges; i++) {
397 		ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
398 
399 		/* If entry is less than a page, skip it */
400 		if (ei.size < PAGE_SIZE)
401 			continue;
402 		ei.addr = cmem->ranges[i].start;
403 		ei.type = E820_TYPE_RAM;
404 		add_e820_entry(params, &ei);
405 	}
406 
407 out:
408 	vfree(cmem);
409 	return ret;
410 }
411 
412 static int determine_backup_region(struct resource *res, void *arg)
413 {
414 	struct kimage *image = arg;
415 
416 	image->arch.backup_src_start = res->start;
417 	image->arch.backup_src_sz = resource_size(res);
418 
419 	/* Expecting only one range for backup region */
420 	return 1;
421 }
422 
423 int crash_load_segments(struct kimage *image)
424 {
425 	int ret;
426 	struct kexec_buf kbuf = { .image = image, .buf_min = 0,
427 				  .buf_max = ULONG_MAX, .top_down = false };
428 
429 	/*
430 	 * Determine and load a segment for backup area. First 640K RAM
431 	 * region is backup source
432 	 */
433 
434 	ret = walk_system_ram_res(KEXEC_BACKUP_SRC_START, KEXEC_BACKUP_SRC_END,
435 				image, determine_backup_region);
436 
437 	/* Zero or postive return values are ok */
438 	if (ret < 0)
439 		return ret;
440 
441 	/* Add backup segment. */
442 	if (image->arch.backup_src_sz) {
443 		kbuf.buffer = &crash_zero_bytes;
444 		kbuf.bufsz = sizeof(crash_zero_bytes);
445 		kbuf.memsz = image->arch.backup_src_sz;
446 		kbuf.buf_align = PAGE_SIZE;
447 		/*
448 		 * Ideally there is no source for backup segment. This is
449 		 * copied in purgatory after crash. Just add a zero filled
450 		 * segment for now to make sure checksum logic works fine.
451 		 */
452 		ret = kexec_add_buffer(&kbuf);
453 		if (ret)
454 			return ret;
455 		image->arch.backup_load_addr = kbuf.mem;
456 		pr_debug("Loaded backup region at 0x%lx backup_start=0x%lx memsz=0x%lx\n",
457 			 image->arch.backup_load_addr,
458 			 image->arch.backup_src_start, kbuf.memsz);
459 	}
460 
461 	/* Prepare elf headers and add a segment */
462 	ret = prepare_elf_headers(image, &kbuf.buffer, &kbuf.bufsz);
463 	if (ret)
464 		return ret;
465 
466 	image->arch.elf_headers = kbuf.buffer;
467 	image->arch.elf_headers_sz = kbuf.bufsz;
468 
469 	kbuf.memsz = kbuf.bufsz;
470 	kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
471 	kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
472 	ret = kexec_add_buffer(&kbuf);
473 	if (ret) {
474 		vfree((void *)image->arch.elf_headers);
475 		return ret;
476 	}
477 	image->arch.elf_load_addr = kbuf.mem;
478 	pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
479 		 image->arch.elf_load_addr, kbuf.bufsz, kbuf.bufsz);
480 
481 	return ret;
482 }
483 #endif /* CONFIG_KEXEC_FILE */
484