xref: /linux/arch/x86/kernel/crash.c (revision 856e7c4b619af622d56b3b454f7bec32a170ac99)
1 /*
2  * Architecture specific (i386/x86_64) functions for kexec based crash dumps.
3  *
4  * Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
5  *
6  * Copyright (C) IBM Corporation, 2004. All rights reserved.
7  * Copyright (C) Red Hat Inc., 2014. All rights reserved.
8  * Authors:
9  *      Vivek Goyal <vgoyal@redhat.com>
10  *
11  */
12 
13 #define pr_fmt(fmt)	"kexec: " fmt
14 
15 #include <linux/types.h>
16 #include <linux/kernel.h>
17 #include <linux/smp.h>
18 #include <linux/reboot.h>
19 #include <linux/kexec.h>
20 #include <linux/delay.h>
21 #include <linux/elf.h>
22 #include <linux/elfcore.h>
23 #include <linux/export.h>
24 #include <linux/slab.h>
25 #include <linux/vmalloc.h>
26 
27 #include <asm/processor.h>
28 #include <asm/hardirq.h>
29 #include <asm/nmi.h>
30 #include <asm/hw_irq.h>
31 #include <asm/apic.h>
32 #include <asm/e820/types.h>
33 #include <asm/io_apic.h>
34 #include <asm/hpet.h>
35 #include <linux/kdebug.h>
36 #include <asm/cpu.h>
37 #include <asm/reboot.h>
38 #include <asm/virtext.h>
39 #include <asm/intel_pt.h>
40 
41 /* Used while preparing memory map entries for second kernel */
42 struct crash_memmap_data {
43 	struct boot_params *params;
44 	/* Type of memory */
45 	unsigned int type;
46 };
47 
48 /*
49  * This is used to VMCLEAR all VMCSs loaded on the
50  * processor. And when loading kvm_intel module, the
51  * callback function pointer will be assigned.
52  *
53  * protected by rcu.
54  */
55 crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss = NULL;
56 EXPORT_SYMBOL_GPL(crash_vmclear_loaded_vmcss);
57 unsigned long crash_zero_bytes;
58 
59 static inline void cpu_crash_vmclear_loaded_vmcss(void)
60 {
61 	crash_vmclear_fn *do_vmclear_operation = NULL;
62 
63 	rcu_read_lock();
64 	do_vmclear_operation = rcu_dereference(crash_vmclear_loaded_vmcss);
65 	if (do_vmclear_operation)
66 		do_vmclear_operation();
67 	rcu_read_unlock();
68 }
69 
70 #if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)
71 
72 static void kdump_nmi_callback(int cpu, struct pt_regs *regs)
73 {
74 #ifdef CONFIG_X86_32
75 	struct pt_regs fixed_regs;
76 
77 	if (!user_mode(regs)) {
78 		crash_fixup_ss_esp(&fixed_regs, regs);
79 		regs = &fixed_regs;
80 	}
81 #endif
82 	crash_save_cpu(regs, cpu);
83 
84 	/*
85 	 * VMCLEAR VMCSs loaded on all cpus if needed.
86 	 */
87 	cpu_crash_vmclear_loaded_vmcss();
88 
89 	/* Disable VMX or SVM if needed.
90 	 *
91 	 * We need to disable virtualization on all CPUs.
92 	 * Having VMX or SVM enabled on any CPU may break rebooting
93 	 * after the kdump kernel has finished its task.
94 	 */
95 	cpu_emergency_vmxoff();
96 	cpu_emergency_svm_disable();
97 
98 	/*
99 	 * Disable Intel PT to stop its logging
100 	 */
101 	cpu_emergency_stop_pt();
102 
103 	disable_local_APIC();
104 }
105 
106 void kdump_nmi_shootdown_cpus(void)
107 {
108 	nmi_shootdown_cpus(kdump_nmi_callback);
109 
110 	disable_local_APIC();
111 }
112 
113 /* Override the weak function in kernel/panic.c */
114 void crash_smp_send_stop(void)
115 {
116 	static int cpus_stopped;
117 
118 	if (cpus_stopped)
119 		return;
120 
121 	if (smp_ops.crash_stop_other_cpus)
122 		smp_ops.crash_stop_other_cpus();
123 	else
124 		smp_send_stop();
125 
126 	cpus_stopped = 1;
127 }
128 
129 #else
130 void crash_smp_send_stop(void)
131 {
132 	/* There are no cpus to shootdown */
133 }
134 #endif
135 
136 void native_machine_crash_shutdown(struct pt_regs *regs)
137 {
138 	/* This function is only called after the system
139 	 * has panicked or is otherwise in a critical state.
140 	 * The minimum amount of code to allow a kexec'd kernel
141 	 * to run successfully needs to happen here.
142 	 *
143 	 * In practice this means shooting down the other cpus in
144 	 * an SMP system.
145 	 */
146 	/* The kernel is broken so disable interrupts */
147 	local_irq_disable();
148 
149 	crash_smp_send_stop();
150 
151 	/*
152 	 * VMCLEAR VMCSs loaded on this cpu if needed.
153 	 */
154 	cpu_crash_vmclear_loaded_vmcss();
155 
156 	/* Booting kdump kernel with VMX or SVM enabled won't work,
157 	 * because (among other limitations) we can't disable paging
158 	 * with the virt flags.
159 	 */
160 	cpu_emergency_vmxoff();
161 	cpu_emergency_svm_disable();
162 
163 	/*
164 	 * Disable Intel PT to stop its logging
165 	 */
166 	cpu_emergency_stop_pt();
167 
168 #ifdef CONFIG_X86_IO_APIC
169 	/* Prevent crash_kexec() from deadlocking on ioapic_lock. */
170 	ioapic_zap_locks();
171 	clear_IO_APIC();
172 #endif
173 	lapic_shutdown();
174 	restore_boot_irq_mode();
175 #ifdef CONFIG_HPET_TIMER
176 	hpet_disable();
177 #endif
178 	crash_save_cpu(regs, safe_smp_processor_id());
179 }
180 
181 #ifdef CONFIG_KEXEC_FILE
182 static int get_nr_ram_ranges_callback(struct resource *res, void *arg)
183 {
184 	unsigned int *nr_ranges = arg;
185 
186 	(*nr_ranges)++;
187 	return 0;
188 }
189 
190 /* Gather all the required information to prepare elf headers for ram regions */
191 static struct crash_mem *fill_up_crash_elf_data(void)
192 {
193 	unsigned int nr_ranges = 0;
194 	struct crash_mem *cmem;
195 
196 	walk_system_ram_res(0, -1, &nr_ranges,
197 				get_nr_ram_ranges_callback);
198 	if (!nr_ranges)
199 		return NULL;
200 
201 	/*
202 	 * Exclusion of crash region and/or crashk_low_res may cause
203 	 * another range split. So add extra two slots here.
204 	 */
205 	nr_ranges += 2;
206 	cmem = vzalloc(sizeof(struct crash_mem) +
207 			sizeof(struct crash_mem_range) * nr_ranges);
208 	if (!cmem)
209 		return NULL;
210 
211 	cmem->max_nr_ranges = nr_ranges;
212 	cmem->nr_ranges = 0;
213 
214 	return cmem;
215 }
216 
217 /*
218  * Look for any unwanted ranges between mstart, mend and remove them. This
219  * might lead to split and split ranges are put in cmem->ranges[] array
220  */
221 static int elf_header_exclude_ranges(struct crash_mem *cmem)
222 {
223 	int ret = 0;
224 
225 	/* Exclude crashkernel region */
226 	ret = crash_exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
227 	if (ret)
228 		return ret;
229 
230 	if (crashk_low_res.end) {
231 		ret = crash_exclude_mem_range(cmem, crashk_low_res.start,
232 							crashk_low_res.end);
233 		if (ret)
234 			return ret;
235 	}
236 
237 	return ret;
238 }
239 
240 static int prepare_elf64_ram_headers_callback(struct resource *res, void *arg)
241 {
242 	struct crash_mem *cmem = arg;
243 
244 	cmem->ranges[cmem->nr_ranges].start = res->start;
245 	cmem->ranges[cmem->nr_ranges].end = res->end;
246 	cmem->nr_ranges++;
247 
248 	return 0;
249 }
250 
251 /* Prepare elf headers. Return addr and size */
252 static int prepare_elf_headers(struct kimage *image, void **addr,
253 					unsigned long *sz)
254 {
255 	struct crash_mem *cmem;
256 	Elf64_Ehdr *ehdr;
257 	Elf64_Phdr *phdr;
258 	int ret, i;
259 
260 	cmem = fill_up_crash_elf_data();
261 	if (!cmem)
262 		return -ENOMEM;
263 
264 	ret = walk_system_ram_res(0, -1, cmem,
265 				prepare_elf64_ram_headers_callback);
266 	if (ret)
267 		goto out;
268 
269 	/* Exclude unwanted mem ranges */
270 	ret = elf_header_exclude_ranges(cmem);
271 	if (ret)
272 		goto out;
273 
274 	/* By default prepare 64bit headers */
275 	ret =  crash_prepare_elf64_headers(cmem,
276 				IS_ENABLED(CONFIG_X86_64), addr, sz);
277 	if (ret)
278 		goto out;
279 
280 	/*
281 	 * If a range matches backup region, adjust offset to backup
282 	 * segment.
283 	 */
284 	ehdr = (Elf64_Ehdr *)*addr;
285 	phdr = (Elf64_Phdr *)(ehdr + 1);
286 	for (i = 0; i < ehdr->e_phnum; phdr++, i++)
287 		if (phdr->p_type == PT_LOAD &&
288 				phdr->p_paddr == image->arch.backup_src_start &&
289 				phdr->p_memsz == image->arch.backup_src_sz) {
290 			phdr->p_offset = image->arch.backup_load_addr;
291 			break;
292 		}
293 out:
294 	vfree(cmem);
295 	return ret;
296 }
297 
298 static int add_e820_entry(struct boot_params *params, struct e820_entry *entry)
299 {
300 	unsigned int nr_e820_entries;
301 
302 	nr_e820_entries = params->e820_entries;
303 	if (nr_e820_entries >= E820_MAX_ENTRIES_ZEROPAGE)
304 		return 1;
305 
306 	memcpy(&params->e820_table[nr_e820_entries], entry,
307 			sizeof(struct e820_entry));
308 	params->e820_entries++;
309 	return 0;
310 }
311 
312 static int memmap_entry_callback(struct resource *res, void *arg)
313 {
314 	struct crash_memmap_data *cmd = arg;
315 	struct boot_params *params = cmd->params;
316 	struct e820_entry ei;
317 
318 	ei.addr = res->start;
319 	ei.size = resource_size(res);
320 	ei.type = cmd->type;
321 	add_e820_entry(params, &ei);
322 
323 	return 0;
324 }
325 
326 static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
327 				 unsigned long long mstart,
328 				 unsigned long long mend)
329 {
330 	unsigned long start, end;
331 	int ret = 0;
332 
333 	cmem->ranges[0].start = mstart;
334 	cmem->ranges[0].end = mend;
335 	cmem->nr_ranges = 1;
336 
337 	/* Exclude Backup region */
338 	start = image->arch.backup_load_addr;
339 	end = start + image->arch.backup_src_sz - 1;
340 	ret = crash_exclude_mem_range(cmem, start, end);
341 	if (ret)
342 		return ret;
343 
344 	/* Exclude elf header region */
345 	start = image->arch.elf_load_addr;
346 	end = start + image->arch.elf_headers_sz - 1;
347 	return crash_exclude_mem_range(cmem, start, end);
348 }
349 
350 /* Prepare memory map for crash dump kernel */
351 int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
352 {
353 	int i, ret = 0;
354 	unsigned long flags;
355 	struct e820_entry ei;
356 	struct crash_memmap_data cmd;
357 	struct crash_mem *cmem;
358 
359 	cmem = vzalloc(sizeof(struct crash_mem));
360 	if (!cmem)
361 		return -ENOMEM;
362 
363 	memset(&cmd, 0, sizeof(struct crash_memmap_data));
364 	cmd.params = params;
365 
366 	/* Add first 640K segment */
367 	ei.addr = image->arch.backup_src_start;
368 	ei.size = image->arch.backup_src_sz;
369 	ei.type = E820_TYPE_RAM;
370 	add_e820_entry(params, &ei);
371 
372 	/* Add ACPI tables */
373 	cmd.type = E820_TYPE_ACPI;
374 	flags = IORESOURCE_MEM | IORESOURCE_BUSY;
375 	walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd,
376 		       memmap_entry_callback);
377 
378 	/* Add ACPI Non-volatile Storage */
379 	cmd.type = E820_TYPE_NVS;
380 	walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, &cmd,
381 			memmap_entry_callback);
382 
383 	/* Add crashk_low_res region */
384 	if (crashk_low_res.end) {
385 		ei.addr = crashk_low_res.start;
386 		ei.size = crashk_low_res.end - crashk_low_res.start + 1;
387 		ei.type = E820_TYPE_RAM;
388 		add_e820_entry(params, &ei);
389 	}
390 
391 	/* Exclude some ranges from crashk_res and add rest to memmap */
392 	ret = memmap_exclude_ranges(image, cmem, crashk_res.start,
393 						crashk_res.end);
394 	if (ret)
395 		goto out;
396 
397 	for (i = 0; i < cmem->nr_ranges; i++) {
398 		ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
399 
400 		/* If entry is less than a page, skip it */
401 		if (ei.size < PAGE_SIZE)
402 			continue;
403 		ei.addr = cmem->ranges[i].start;
404 		ei.type = E820_TYPE_RAM;
405 		add_e820_entry(params, &ei);
406 	}
407 
408 out:
409 	vfree(cmem);
410 	return ret;
411 }
412 
413 static int determine_backup_region(struct resource *res, void *arg)
414 {
415 	struct kimage *image = arg;
416 
417 	image->arch.backup_src_start = res->start;
418 	image->arch.backup_src_sz = resource_size(res);
419 
420 	/* Expecting only one range for backup region */
421 	return 1;
422 }
423 
424 int crash_load_segments(struct kimage *image)
425 {
426 	int ret;
427 	struct kexec_buf kbuf = { .image = image, .buf_min = 0,
428 				  .buf_max = ULONG_MAX, .top_down = false };
429 
430 	/*
431 	 * Determine and load a segment for backup area. First 640K RAM
432 	 * region is backup source
433 	 */
434 
435 	ret = walk_system_ram_res(KEXEC_BACKUP_SRC_START, KEXEC_BACKUP_SRC_END,
436 				image, determine_backup_region);
437 
438 	/* Zero or postive return values are ok */
439 	if (ret < 0)
440 		return ret;
441 
442 	/* Add backup segment. */
443 	if (image->arch.backup_src_sz) {
444 		kbuf.buffer = &crash_zero_bytes;
445 		kbuf.bufsz = sizeof(crash_zero_bytes);
446 		kbuf.memsz = image->arch.backup_src_sz;
447 		kbuf.buf_align = PAGE_SIZE;
448 		/*
449 		 * Ideally there is no source for backup segment. This is
450 		 * copied in purgatory after crash. Just add a zero filled
451 		 * segment for now to make sure checksum logic works fine.
452 		 */
453 		ret = kexec_add_buffer(&kbuf);
454 		if (ret)
455 			return ret;
456 		image->arch.backup_load_addr = kbuf.mem;
457 		pr_debug("Loaded backup region at 0x%lx backup_start=0x%lx memsz=0x%lx\n",
458 			 image->arch.backup_load_addr,
459 			 image->arch.backup_src_start, kbuf.memsz);
460 	}
461 
462 	/* Prepare elf headers and add a segment */
463 	ret = prepare_elf_headers(image, &kbuf.buffer, &kbuf.bufsz);
464 	if (ret)
465 		return ret;
466 
467 	image->arch.elf_headers = kbuf.buffer;
468 	image->arch.elf_headers_sz = kbuf.bufsz;
469 
470 	kbuf.memsz = kbuf.bufsz;
471 	kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
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