xref: /openbmc/linux/arch/x86/kernel/crash.c (revision 67bf4745)
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 		if (ret)
229 			return ret;
230 	}
231 
232 	return ret;
233 }
234 
235 static int prepare_elf64_ram_headers_callback(struct resource *res, void *arg)
236 {
237 	struct crash_mem *cmem = arg;
238 
239 	cmem->ranges[cmem->nr_ranges].start = res->start;
240 	cmem->ranges[cmem->nr_ranges].end = res->end;
241 	cmem->nr_ranges++;
242 
243 	return 0;
244 }
245 
246 /* Prepare elf headers. Return addr and size */
247 static int prepare_elf_headers(struct kimage *image, void **addr,
248 					unsigned long *sz)
249 {
250 	struct crash_mem *cmem;
251 	Elf64_Ehdr *ehdr;
252 	Elf64_Phdr *phdr;
253 	int ret, i;
254 
255 	cmem = fill_up_crash_elf_data();
256 	if (!cmem)
257 		return -ENOMEM;
258 
259 	ret = walk_system_ram_res(0, -1, cmem,
260 				prepare_elf64_ram_headers_callback);
261 	if (ret)
262 		goto out;
263 
264 	/* Exclude unwanted mem ranges */
265 	ret = elf_header_exclude_ranges(cmem);
266 	if (ret)
267 		goto out;
268 
269 	/* By default prepare 64bit headers */
270 	ret =  crash_prepare_elf64_headers(cmem,
271 				IS_ENABLED(CONFIG_X86_64), addr, sz);
272 	if (ret)
273 		goto out;
274 
275 	/*
276 	 * If a range matches backup region, adjust offset to backup
277 	 * segment.
278 	 */
279 	ehdr = (Elf64_Ehdr *)*addr;
280 	phdr = (Elf64_Phdr *)(ehdr + 1);
281 	for (i = 0; i < ehdr->e_phnum; phdr++, i++)
282 		if (phdr->p_type == PT_LOAD &&
283 				phdr->p_paddr == image->arch.backup_src_start &&
284 				phdr->p_memsz == image->arch.backup_src_sz) {
285 			phdr->p_offset = image->arch.backup_load_addr;
286 			break;
287 		}
288 out:
289 	vfree(cmem);
290 	return ret;
291 }
292 
293 static int add_e820_entry(struct boot_params *params, struct e820_entry *entry)
294 {
295 	unsigned int nr_e820_entries;
296 
297 	nr_e820_entries = params->e820_entries;
298 	if (nr_e820_entries >= E820_MAX_ENTRIES_ZEROPAGE)
299 		return 1;
300 
301 	memcpy(&params->e820_table[nr_e820_entries], entry,
302 			sizeof(struct e820_entry));
303 	params->e820_entries++;
304 	return 0;
305 }
306 
307 static int memmap_entry_callback(struct resource *res, void *arg)
308 {
309 	struct crash_memmap_data *cmd = arg;
310 	struct boot_params *params = cmd->params;
311 	struct e820_entry ei;
312 
313 	ei.addr = res->start;
314 	ei.size = resource_size(res);
315 	ei.type = cmd->type;
316 	add_e820_entry(params, &ei);
317 
318 	return 0;
319 }
320 
321 static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
322 				 unsigned long long mstart,
323 				 unsigned long long mend)
324 {
325 	unsigned long start, end;
326 	int ret = 0;
327 
328 	cmem->ranges[0].start = mstart;
329 	cmem->ranges[0].end = mend;
330 	cmem->nr_ranges = 1;
331 
332 	/* Exclude Backup region */
333 	start = image->arch.backup_load_addr;
334 	end = start + image->arch.backup_src_sz - 1;
335 	ret = crash_exclude_mem_range(cmem, start, end);
336 	if (ret)
337 		return ret;
338 
339 	/* Exclude elf header region */
340 	start = image->arch.elf_load_addr;
341 	end = start + image->arch.elf_headers_sz - 1;
342 	return crash_exclude_mem_range(cmem, start, end);
343 }
344 
345 /* Prepare memory map for crash dump kernel */
346 int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
347 {
348 	int i, ret = 0;
349 	unsigned long flags;
350 	struct e820_entry ei;
351 	struct crash_memmap_data cmd;
352 	struct crash_mem *cmem;
353 
354 	cmem = vzalloc(sizeof(struct crash_mem));
355 	if (!cmem)
356 		return -ENOMEM;
357 
358 	memset(&cmd, 0, sizeof(struct crash_memmap_data));
359 	cmd.params = params;
360 
361 	/* Add first 640K segment */
362 	ei.addr = image->arch.backup_src_start;
363 	ei.size = image->arch.backup_src_sz;
364 	ei.type = E820_TYPE_RAM;
365 	add_e820_entry(params, &ei);
366 
367 	/* Add ACPI tables */
368 	cmd.type = E820_TYPE_ACPI;
369 	flags = IORESOURCE_MEM | IORESOURCE_BUSY;
370 	walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd,
371 		       memmap_entry_callback);
372 
373 	/* Add ACPI Non-volatile Storage */
374 	cmd.type = E820_TYPE_NVS;
375 	walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, &cmd,
376 			memmap_entry_callback);
377 
378 	/* Add e820 reserved ranges */
379 	cmd.type = E820_TYPE_RESERVED;
380 	flags = IORESOURCE_MEM;
381 	walk_iomem_res_desc(IORES_DESC_RESERVED, flags, 0, -1, &cmd,
382 			   memmap_entry_callback);
383 
384 	/* Add crashk_low_res region */
385 	if (crashk_low_res.end) {
386 		ei.addr = crashk_low_res.start;
387 		ei.size = crashk_low_res.end - crashk_low_res.start + 1;
388 		ei.type = E820_TYPE_RAM;
389 		add_e820_entry(params, &ei);
390 	}
391 
392 	/* Exclude some ranges from crashk_res and add rest to memmap */
393 	ret = memmap_exclude_ranges(image, cmem, crashk_res.start,
394 						crashk_res.end);
395 	if (ret)
396 		goto out;
397 
398 	for (i = 0; i < cmem->nr_ranges; i++) {
399 		ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
400 
401 		/* If entry is less than a page, skip it */
402 		if (ei.size < PAGE_SIZE)
403 			continue;
404 		ei.addr = cmem->ranges[i].start;
405 		ei.type = E820_TYPE_RAM;
406 		add_e820_entry(params, &ei);
407 	}
408 
409 out:
410 	vfree(cmem);
411 	return ret;
412 }
413 
414 static int determine_backup_region(struct resource *res, void *arg)
415 {
416 	struct kimage *image = arg;
417 
418 	image->arch.backup_src_start = res->start;
419 	image->arch.backup_src_sz = resource_size(res);
420 
421 	/* Expecting only one range for backup region */
422 	return 1;
423 }
424 
425 int crash_load_segments(struct kimage *image)
426 {
427 	int ret;
428 	struct kexec_buf kbuf = { .image = image, .buf_min = 0,
429 				  .buf_max = ULONG_MAX, .top_down = false };
430 
431 	/*
432 	 * Determine and load a segment for backup area. First 640K RAM
433 	 * region is backup source
434 	 */
435 
436 	ret = walk_system_ram_res(KEXEC_BACKUP_SRC_START, KEXEC_BACKUP_SRC_END,
437 				image, determine_backup_region);
438 
439 	/* Zero or postive return values are ok */
440 	if (ret < 0)
441 		return ret;
442 
443 	/* Add backup segment. */
444 	if (image->arch.backup_src_sz) {
445 		kbuf.buffer = &crash_zero_bytes;
446 		kbuf.bufsz = sizeof(crash_zero_bytes);
447 		kbuf.memsz = image->arch.backup_src_sz;
448 		kbuf.buf_align = PAGE_SIZE;
449 		/*
450 		 * Ideally there is no source for backup segment. This is
451 		 * copied in purgatory after crash. Just add a zero filled
452 		 * segment for now to make sure checksum logic works fine.
453 		 */
454 		ret = kexec_add_buffer(&kbuf);
455 		if (ret)
456 			return ret;
457 		image->arch.backup_load_addr = kbuf.mem;
458 		pr_debug("Loaded backup region at 0x%lx backup_start=0x%lx memsz=0x%lx\n",
459 			 image->arch.backup_load_addr,
460 			 image->arch.backup_src_start, kbuf.memsz);
461 	}
462 
463 	/* Prepare elf headers and add a segment */
464 	ret = prepare_elf_headers(image, &kbuf.buffer, &kbuf.bufsz);
465 	if (ret)
466 		return ret;
467 
468 	image->arch.elf_headers = kbuf.buffer;
469 	image->arch.elf_headers_sz = kbuf.bufsz;
470 
471 	kbuf.memsz = kbuf.bufsz;
472 	kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
473 	kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
474 	ret = kexec_add_buffer(&kbuf);
475 	if (ret) {
476 		vfree((void *)image->arch.elf_headers);
477 		return ret;
478 	}
479 	image->arch.elf_load_addr = kbuf.mem;
480 	pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
481 		 image->arch.elf_load_addr, kbuf.bufsz, kbuf.bufsz);
482 
483 	return ret;
484 }
485 #endif /* CONFIG_KEXEC_FILE */
486