xref: /openbmc/linux/arch/x86/kernel/crash.c (revision f32e5616)
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 #include <asm/crash.h>
41 
42 /* Used while preparing memory map entries for second kernel */
43 struct crash_memmap_data {
44 	struct boot_params *params;
45 	/* Type of memory */
46 	unsigned int type;
47 };
48 
49 /*
50  * This is used to VMCLEAR all VMCSs loaded on the
51  * processor. And when loading kvm_intel module, the
52  * callback function pointer will be assigned.
53  *
54  * protected by rcu.
55  */
56 crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss = NULL;
57 EXPORT_SYMBOL_GPL(crash_vmclear_loaded_vmcss);
58 unsigned long crash_zero_bytes;
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 #ifdef CONFIG_X86_32
76 	struct pt_regs fixed_regs;
77 
78 	if (!user_mode(regs)) {
79 		crash_fixup_ss_esp(&fixed_regs, regs);
80 		regs = &fixed_regs;
81 	}
82 #endif
83 	crash_save_cpu(regs, cpu);
84 
85 	/*
86 	 * VMCLEAR VMCSs loaded on all cpus if needed.
87 	 */
88 	cpu_crash_vmclear_loaded_vmcss();
89 
90 	/* Disable VMX or SVM if needed.
91 	 *
92 	 * We need to disable virtualization on all CPUs.
93 	 * Having VMX or SVM enabled on any CPU may break rebooting
94 	 * after the kdump kernel has finished its task.
95 	 */
96 	cpu_emergency_vmxoff();
97 	cpu_emergency_svm_disable();
98 
99 	/*
100 	 * Disable Intel PT to stop its logging
101 	 */
102 	cpu_emergency_stop_pt();
103 
104 	disable_local_APIC();
105 }
106 
107 void kdump_nmi_shootdown_cpus(void)
108 {
109 	nmi_shootdown_cpus(kdump_nmi_callback);
110 
111 	disable_local_APIC();
112 }
113 
114 /* Override the weak function in kernel/panic.c */
115 void crash_smp_send_stop(void)
116 {
117 	static int cpus_stopped;
118 
119 	if (cpus_stopped)
120 		return;
121 
122 	if (smp_ops.crash_stop_other_cpus)
123 		smp_ops.crash_stop_other_cpus();
124 	else
125 		smp_send_stop();
126 
127 	cpus_stopped = 1;
128 }
129 
130 #else
131 void crash_smp_send_stop(void)
132 {
133 	/* There are no cpus to shootdown */
134 }
135 #endif
136 
137 void native_machine_crash_shutdown(struct pt_regs *regs)
138 {
139 	/* This function is only called after the system
140 	 * has panicked or is otherwise in a critical state.
141 	 * The minimum amount of code to allow a kexec'd kernel
142 	 * to run successfully needs to happen here.
143 	 *
144 	 * In practice this means shooting down the other cpus in
145 	 * an SMP system.
146 	 */
147 	/* The kernel is broken so disable interrupts */
148 	local_irq_disable();
149 
150 	crash_smp_send_stop();
151 
152 	/*
153 	 * VMCLEAR VMCSs loaded on this cpu if needed.
154 	 */
155 	cpu_crash_vmclear_loaded_vmcss();
156 
157 	/* Booting kdump kernel with VMX or SVM enabled won't work,
158 	 * because (among other limitations) we can't disable paging
159 	 * with the virt flags.
160 	 */
161 	cpu_emergency_vmxoff();
162 	cpu_emergency_svm_disable();
163 
164 	/*
165 	 * Disable Intel PT to stop its logging
166 	 */
167 	cpu_emergency_stop_pt();
168 
169 #ifdef CONFIG_X86_IO_APIC
170 	/* Prevent crash_kexec() from deadlocking on ioapic_lock. */
171 	ioapic_zap_locks();
172 	clear_IO_APIC();
173 #endif
174 	lapic_shutdown();
175 	restore_boot_irq_mode();
176 #ifdef CONFIG_HPET_TIMER
177 	hpet_disable();
178 #endif
179 	crash_save_cpu(regs, safe_smp_processor_id());
180 }
181 
182 #ifdef CONFIG_KEXEC_FILE
183 static int get_nr_ram_ranges_callback(struct resource *res, void *arg)
184 {
185 	unsigned int *nr_ranges = arg;
186 
187 	(*nr_ranges)++;
188 	return 0;
189 }
190 
191 /* Gather all the required information to prepare elf headers for ram regions */
192 static struct crash_mem *fill_up_crash_elf_data(void)
193 {
194 	unsigned int nr_ranges = 0;
195 	struct crash_mem *cmem;
196 
197 	walk_system_ram_res(0, -1, &nr_ranges,
198 				get_nr_ram_ranges_callback);
199 	if (!nr_ranges)
200 		return NULL;
201 
202 	/*
203 	 * Exclusion of crash region and/or crashk_low_res may cause
204 	 * another range split. So add extra two slots here.
205 	 */
206 	nr_ranges += 2;
207 	cmem = vzalloc(sizeof(struct crash_mem) +
208 			sizeof(struct crash_mem_range) * nr_ranges);
209 	if (!cmem)
210 		return NULL;
211 
212 	cmem->max_nr_ranges = nr_ranges;
213 	cmem->nr_ranges = 0;
214 
215 	return cmem;
216 }
217 
218 /*
219  * Look for any unwanted ranges between mstart, mend and remove them. This
220  * might lead to split and split ranges are put in cmem->ranges[] array
221  */
222 static int elf_header_exclude_ranges(struct crash_mem *cmem)
223 {
224 	int ret = 0;
225 
226 	/* Exclude crashkernel region */
227 	ret = crash_exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
228 	if (ret)
229 		return ret;
230 
231 	if (crashk_low_res.end) {
232 		ret = crash_exclude_mem_range(cmem, crashk_low_res.start,
233 							crashk_low_res.end);
234 		if (ret)
235 			return ret;
236 	}
237 
238 	return ret;
239 }
240 
241 static int prepare_elf64_ram_headers_callback(struct resource *res, void *arg)
242 {
243 	struct crash_mem *cmem = arg;
244 
245 	cmem->ranges[cmem->nr_ranges].start = res->start;
246 	cmem->ranges[cmem->nr_ranges].end = res->end;
247 	cmem->nr_ranges++;
248 
249 	return 0;
250 }
251 
252 /* Prepare elf headers. Return addr and size */
253 static int prepare_elf_headers(struct kimage *image, void **addr,
254 					unsigned long *sz)
255 {
256 	struct crash_mem *cmem;
257 	Elf64_Ehdr *ehdr;
258 	Elf64_Phdr *phdr;
259 	int ret, i;
260 
261 	cmem = fill_up_crash_elf_data();
262 	if (!cmem)
263 		return -ENOMEM;
264 
265 	ret = walk_system_ram_res(0, -1, cmem,
266 				prepare_elf64_ram_headers_callback);
267 	if (ret)
268 		goto out;
269 
270 	/* Exclude unwanted mem ranges */
271 	ret = elf_header_exclude_ranges(cmem);
272 	if (ret)
273 		goto out;
274 
275 	/* By default prepare 64bit headers */
276 	ret =  crash_prepare_elf64_headers(cmem,
277 				IS_ENABLED(CONFIG_X86_64), addr, sz);
278 	if (ret)
279 		goto out;
280 
281 	/*
282 	 * If a range matches backup region, adjust offset to backup
283 	 * segment.
284 	 */
285 	ehdr = (Elf64_Ehdr *)*addr;
286 	phdr = (Elf64_Phdr *)(ehdr + 1);
287 	for (i = 0; i < ehdr->e_phnum; phdr++, i++)
288 		if (phdr->p_type == PT_LOAD &&
289 				phdr->p_paddr == image->arch.backup_src_start &&
290 				phdr->p_memsz == image->arch.backup_src_sz) {
291 			phdr->p_offset = image->arch.backup_load_addr;
292 			break;
293 		}
294 out:
295 	vfree(cmem);
296 	return ret;
297 }
298 
299 static int add_e820_entry(struct boot_params *params, struct e820_entry *entry)
300 {
301 	unsigned int nr_e820_entries;
302 
303 	nr_e820_entries = params->e820_entries;
304 	if (nr_e820_entries >= E820_MAX_ENTRIES_ZEROPAGE)
305 		return 1;
306 
307 	memcpy(&params->e820_table[nr_e820_entries], entry,
308 			sizeof(struct e820_entry));
309 	params->e820_entries++;
310 	return 0;
311 }
312 
313 static int memmap_entry_callback(struct resource *res, void *arg)
314 {
315 	struct crash_memmap_data *cmd = arg;
316 	struct boot_params *params = cmd->params;
317 	struct e820_entry ei;
318 
319 	ei.addr = res->start;
320 	ei.size = resource_size(res);
321 	ei.type = cmd->type;
322 	add_e820_entry(params, &ei);
323 
324 	return 0;
325 }
326 
327 static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
328 				 unsigned long long mstart,
329 				 unsigned long long mend)
330 {
331 	unsigned long start, end;
332 	int ret = 0;
333 
334 	cmem->ranges[0].start = mstart;
335 	cmem->ranges[0].end = mend;
336 	cmem->nr_ranges = 1;
337 
338 	/* Exclude Backup region */
339 	start = image->arch.backup_load_addr;
340 	end = start + image->arch.backup_src_sz - 1;
341 	ret = crash_exclude_mem_range(cmem, start, end);
342 	if (ret)
343 		return ret;
344 
345 	/* Exclude elf header region */
346 	start = image->arch.elf_load_addr;
347 	end = start + image->arch.elf_headers_sz - 1;
348 	return crash_exclude_mem_range(cmem, start, end);
349 }
350 
351 /* Prepare memory map for crash dump kernel */
352 int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
353 {
354 	int i, ret = 0;
355 	unsigned long flags;
356 	struct e820_entry ei;
357 	struct crash_memmap_data cmd;
358 	struct crash_mem *cmem;
359 
360 	cmem = vzalloc(sizeof(struct crash_mem));
361 	if (!cmem)
362 		return -ENOMEM;
363 
364 	memset(&cmd, 0, sizeof(struct crash_memmap_data));
365 	cmd.params = params;
366 
367 	/* Add first 640K segment */
368 	ei.addr = image->arch.backup_src_start;
369 	ei.size = image->arch.backup_src_sz;
370 	ei.type = E820_TYPE_RAM;
371 	add_e820_entry(params, &ei);
372 
373 	/* Add ACPI tables */
374 	cmd.type = E820_TYPE_ACPI;
375 	flags = IORESOURCE_MEM | IORESOURCE_BUSY;
376 	walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd,
377 		       memmap_entry_callback);
378 
379 	/* Add ACPI Non-volatile Storage */
380 	cmd.type = E820_TYPE_NVS;
381 	walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, 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 	ret = kexec_add_buffer(&kbuf);
474 	if (ret) {
475 		vfree((void *)image->arch.elf_headers);
476 		return ret;
477 	}
478 	image->arch.elf_load_addr = kbuf.mem;
479 	pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
480 		 image->arch.elf_load_addr, kbuf.bufsz, kbuf.bufsz);
481 
482 	return ret;
483 }
484 #endif /* CONFIG_KEXEC_FILE */
485