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  #include <linux/memblock.h>
28  
29  #include <asm/bootparam.h>
30  #include <asm/processor.h>
31  #include <asm/hardirq.h>
32  #include <asm/nmi.h>
33  #include <asm/hw_irq.h>
34  #include <asm/apic.h>
35  #include <asm/e820/types.h>
36  #include <asm/io_apic.h>
37  #include <asm/hpet.h>
38  #include <linux/kdebug.h>
39  #include <asm/cpu.h>
40  #include <asm/reboot.h>
41  #include <asm/intel_pt.h>
42  #include <asm/crash.h>
43  #include <asm/cmdline.h>
44  #include <asm/sev.h>
45  
46  /* Used while preparing memory map entries for second kernel */
47  struct crash_memmap_data {
48  	struct boot_params *params;
49  	/* Type of memory */
50  	unsigned int type;
51  };
52  
53  #if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)
54  
kdump_nmi_callback(int cpu,struct pt_regs * regs)55  static void kdump_nmi_callback(int cpu, struct pt_regs *regs)
56  {
57  	crash_save_cpu(regs, cpu);
58  
59  	/*
60  	 * Disable Intel PT to stop its logging
61  	 */
62  	cpu_emergency_stop_pt();
63  
64  	kdump_sev_callback();
65  
66  	disable_local_APIC();
67  }
68  
kdump_nmi_shootdown_cpus(void)69  void kdump_nmi_shootdown_cpus(void)
70  {
71  	nmi_shootdown_cpus(kdump_nmi_callback);
72  
73  	disable_local_APIC();
74  }
75  
76  /* Override the weak function in kernel/panic.c */
crash_smp_send_stop(void)77  void crash_smp_send_stop(void)
78  {
79  	static int cpus_stopped;
80  
81  	if (cpus_stopped)
82  		return;
83  
84  	if (smp_ops.crash_stop_other_cpus)
85  		smp_ops.crash_stop_other_cpus();
86  	else
87  		smp_send_stop();
88  
89  	cpus_stopped = 1;
90  }
91  
92  #else
crash_smp_send_stop(void)93  void crash_smp_send_stop(void)
94  {
95  	/* There are no cpus to shootdown */
96  }
97  #endif
98  
native_machine_crash_shutdown(struct pt_regs * regs)99  void native_machine_crash_shutdown(struct pt_regs *regs)
100  {
101  	/* This function is only called after the system
102  	 * has panicked or is otherwise in a critical state.
103  	 * The minimum amount of code to allow a kexec'd kernel
104  	 * to run successfully needs to happen here.
105  	 *
106  	 * In practice this means shooting down the other cpus in
107  	 * an SMP system.
108  	 */
109  	/* The kernel is broken so disable interrupts */
110  	local_irq_disable();
111  
112  	crash_smp_send_stop();
113  
114  	cpu_emergency_disable_virtualization();
115  
116  	/*
117  	 * Disable Intel PT to stop its logging
118  	 */
119  	cpu_emergency_stop_pt();
120  
121  #ifdef CONFIG_X86_IO_APIC
122  	/* Prevent crash_kexec() from deadlocking on ioapic_lock. */
123  	ioapic_zap_locks();
124  	clear_IO_APIC();
125  #endif
126  	lapic_shutdown();
127  	restore_boot_irq_mode();
128  #ifdef CONFIG_HPET_TIMER
129  	hpet_disable();
130  #endif
131  
132  	/*
133  	 * Non-crash kexec calls enc_kexec_begin() while scheduling is still
134  	 * active. This allows the callback to wait until all in-flight
135  	 * shared<->private conversions are complete. In a crash scenario,
136  	 * enc_kexec_begin() gets called after all but one CPU have been shut
137  	 * down and interrupts have been disabled. This allows the callback to
138  	 * detect a race with the conversion and report it.
139  	 */
140  	x86_platform.guest.enc_kexec_begin();
141  	x86_platform.guest.enc_kexec_finish();
142  
143  	crash_save_cpu(regs, safe_smp_processor_id());
144  }
145  
146  #if defined(CONFIG_KEXEC_FILE) || defined(CONFIG_CRASH_HOTPLUG)
get_nr_ram_ranges_callback(struct resource * res,void * arg)147  static int get_nr_ram_ranges_callback(struct resource *res, void *arg)
148  {
149  	unsigned int *nr_ranges = arg;
150  
151  	(*nr_ranges)++;
152  	return 0;
153  }
154  
155  /* Gather all the required information to prepare elf headers for ram regions */
fill_up_crash_elf_data(void)156  static struct crash_mem *fill_up_crash_elf_data(void)
157  {
158  	unsigned int nr_ranges = 0;
159  	struct crash_mem *cmem;
160  
161  	walk_system_ram_res(0, -1, &nr_ranges, get_nr_ram_ranges_callback);
162  	if (!nr_ranges)
163  		return NULL;
164  
165  	/*
166  	 * Exclusion of crash region and/or crashk_low_res may cause
167  	 * another range split. So add extra two slots here.
168  	 */
169  	nr_ranges += 2;
170  	cmem = vzalloc(struct_size(cmem, ranges, nr_ranges));
171  	if (!cmem)
172  		return NULL;
173  
174  	cmem->max_nr_ranges = nr_ranges;
175  	cmem->nr_ranges = 0;
176  
177  	return cmem;
178  }
179  
180  /*
181   * Look for any unwanted ranges between mstart, mend and remove them. This
182   * might lead to split and split ranges are put in cmem->ranges[] array
183   */
elf_header_exclude_ranges(struct crash_mem * cmem)184  static int elf_header_exclude_ranges(struct crash_mem *cmem)
185  {
186  	int ret = 0;
187  
188  	/* Exclude the low 1M because it is always reserved */
189  	ret = crash_exclude_mem_range(cmem, 0, SZ_1M - 1);
190  	if (ret)
191  		return ret;
192  
193  	/* Exclude crashkernel region */
194  	ret = crash_exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
195  	if (ret)
196  		return ret;
197  
198  	if (crashk_low_res.end)
199  		ret = crash_exclude_mem_range(cmem, crashk_low_res.start,
200  					      crashk_low_res.end);
201  
202  	return ret;
203  }
204  
prepare_elf64_ram_headers_callback(struct resource * res,void * arg)205  static int prepare_elf64_ram_headers_callback(struct resource *res, void *arg)
206  {
207  	struct crash_mem *cmem = arg;
208  
209  	cmem->ranges[cmem->nr_ranges].start = res->start;
210  	cmem->ranges[cmem->nr_ranges].end = res->end;
211  	cmem->nr_ranges++;
212  
213  	return 0;
214  }
215  
216  /* Prepare elf headers. Return addr and size */
prepare_elf_headers(void ** addr,unsigned long * sz,unsigned long * nr_mem_ranges)217  static int prepare_elf_headers(void **addr, unsigned long *sz,
218  			       unsigned long *nr_mem_ranges)
219  {
220  	struct crash_mem *cmem;
221  	int ret;
222  
223  	cmem = fill_up_crash_elf_data();
224  	if (!cmem)
225  		return -ENOMEM;
226  
227  	ret = walk_system_ram_res(0, -1, cmem, prepare_elf64_ram_headers_callback);
228  	if (ret)
229  		goto out;
230  
231  	/* Exclude unwanted mem ranges */
232  	ret = elf_header_exclude_ranges(cmem);
233  	if (ret)
234  		goto out;
235  
236  	/* Return the computed number of memory ranges, for hotplug usage */
237  	*nr_mem_ranges = cmem->nr_ranges;
238  
239  	/* By default prepare 64bit headers */
240  	ret = crash_prepare_elf64_headers(cmem, IS_ENABLED(CONFIG_X86_64), addr, sz);
241  
242  out:
243  	vfree(cmem);
244  	return ret;
245  }
246  #endif
247  
248  #ifdef CONFIG_KEXEC_FILE
add_e820_entry(struct boot_params * params,struct e820_entry * entry)249  static int add_e820_entry(struct boot_params *params, struct e820_entry *entry)
250  {
251  	unsigned int nr_e820_entries;
252  
253  	nr_e820_entries = params->e820_entries;
254  	if (nr_e820_entries >= E820_MAX_ENTRIES_ZEROPAGE)
255  		return 1;
256  
257  	memcpy(&params->e820_table[nr_e820_entries], entry, sizeof(struct e820_entry));
258  	params->e820_entries++;
259  	return 0;
260  }
261  
memmap_entry_callback(struct resource * res,void * arg)262  static int memmap_entry_callback(struct resource *res, void *arg)
263  {
264  	struct crash_memmap_data *cmd = arg;
265  	struct boot_params *params = cmd->params;
266  	struct e820_entry ei;
267  
268  	ei.addr = res->start;
269  	ei.size = resource_size(res);
270  	ei.type = cmd->type;
271  	add_e820_entry(params, &ei);
272  
273  	return 0;
274  }
275  
memmap_exclude_ranges(struct kimage * image,struct crash_mem * cmem,unsigned long long mstart,unsigned long long mend)276  static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
277  				 unsigned long long mstart,
278  				 unsigned long long mend)
279  {
280  	unsigned long start, end;
281  
282  	cmem->ranges[0].start = mstart;
283  	cmem->ranges[0].end = mend;
284  	cmem->nr_ranges = 1;
285  
286  	/* Exclude elf header region */
287  	start = image->elf_load_addr;
288  	end = start + image->elf_headers_sz - 1;
289  	return crash_exclude_mem_range(cmem, start, end);
290  }
291  
292  /* Prepare memory map for crash dump kernel */
crash_setup_memmap_entries(struct kimage * image,struct boot_params * params)293  int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
294  {
295  	int i, ret = 0;
296  	unsigned long flags;
297  	struct e820_entry ei;
298  	struct crash_memmap_data cmd;
299  	struct crash_mem *cmem;
300  
301  	cmem = vzalloc(struct_size(cmem, ranges, 1));
302  	if (!cmem)
303  		return -ENOMEM;
304  
305  	memset(&cmd, 0, sizeof(struct crash_memmap_data));
306  	cmd.params = params;
307  
308  	/* Add the low 1M */
309  	cmd.type = E820_TYPE_RAM;
310  	flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
311  	walk_iomem_res_desc(IORES_DESC_NONE, flags, 0, (1<<20)-1, &cmd,
312  			    memmap_entry_callback);
313  
314  	/* Add ACPI tables */
315  	cmd.type = E820_TYPE_ACPI;
316  	flags = IORESOURCE_MEM | IORESOURCE_BUSY;
317  	walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd,
318  			    memmap_entry_callback);
319  
320  	/* Add ACPI Non-volatile Storage */
321  	cmd.type = E820_TYPE_NVS;
322  	walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, &cmd,
323  			    memmap_entry_callback);
324  
325  	/* Add e820 reserved ranges */
326  	cmd.type = E820_TYPE_RESERVED;
327  	flags = IORESOURCE_MEM;
328  	walk_iomem_res_desc(IORES_DESC_RESERVED, flags, 0, -1, &cmd,
329  			    memmap_entry_callback);
330  
331  	/* Add crashk_low_res region */
332  	if (crashk_low_res.end) {
333  		ei.addr = crashk_low_res.start;
334  		ei.size = resource_size(&crashk_low_res);
335  		ei.type = E820_TYPE_RAM;
336  		add_e820_entry(params, &ei);
337  	}
338  
339  	/* Exclude some ranges from crashk_res and add rest to memmap */
340  	ret = memmap_exclude_ranges(image, cmem, crashk_res.start, crashk_res.end);
341  	if (ret)
342  		goto out;
343  
344  	for (i = 0; i < cmem->nr_ranges; i++) {
345  		ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
346  
347  		/* If entry is less than a page, skip it */
348  		if (ei.size < PAGE_SIZE)
349  			continue;
350  		ei.addr = cmem->ranges[i].start;
351  		ei.type = E820_TYPE_RAM;
352  		add_e820_entry(params, &ei);
353  	}
354  
355  out:
356  	vfree(cmem);
357  	return ret;
358  }
359  
crash_load_segments(struct kimage * image)360  int crash_load_segments(struct kimage *image)
361  {
362  	int ret;
363  	unsigned long pnum = 0;
364  	struct kexec_buf kbuf = { .image = image, .buf_min = 0,
365  				  .buf_max = ULONG_MAX, .top_down = false };
366  
367  	/* Prepare elf headers and add a segment */
368  	ret = prepare_elf_headers(&kbuf.buffer, &kbuf.bufsz, &pnum);
369  	if (ret)
370  		return ret;
371  
372  	image->elf_headers	= kbuf.buffer;
373  	image->elf_headers_sz	= kbuf.bufsz;
374  	kbuf.memsz		= kbuf.bufsz;
375  
376  #ifdef CONFIG_CRASH_HOTPLUG
377  	/*
378  	 * The elfcorehdr segment size accounts for VMCOREINFO, kernel_map,
379  	 * maximum CPUs and maximum memory ranges.
380  	 */
381  	if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
382  		pnum = 2 + CONFIG_NR_CPUS_DEFAULT + CONFIG_CRASH_MAX_MEMORY_RANGES;
383  	else
384  		pnum += 2 + CONFIG_NR_CPUS_DEFAULT;
385  
386  	if (pnum < (unsigned long)PN_XNUM) {
387  		kbuf.memsz = pnum * sizeof(Elf64_Phdr);
388  		kbuf.memsz += sizeof(Elf64_Ehdr);
389  
390  		image->elfcorehdr_index = image->nr_segments;
391  
392  		/* Mark as usable to crash kernel, else crash kernel fails on boot */
393  		image->elf_headers_sz = kbuf.memsz;
394  	} else {
395  		pr_err("number of Phdrs %lu exceeds max\n", pnum);
396  	}
397  #endif
398  
399  	kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
400  	kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
401  	ret = kexec_add_buffer(&kbuf);
402  	if (ret)
403  		return ret;
404  	image->elf_load_addr = kbuf.mem;
405  	kexec_dprintk("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
406  		      image->elf_load_addr, kbuf.bufsz, kbuf.memsz);
407  
408  	return ret;
409  }
410  #endif /* CONFIG_KEXEC_FILE */
411  
412  #ifdef CONFIG_CRASH_HOTPLUG
413  
414  #undef pr_fmt
415  #define pr_fmt(fmt) "crash hp: " fmt
416  
arch_crash_hotplug_support(struct kimage * image,unsigned long kexec_flags)417  int arch_crash_hotplug_support(struct kimage *image, unsigned long kexec_flags)
418  {
419  
420  #ifdef CONFIG_KEXEC_FILE
421  	if (image->file_mode)
422  		return 1;
423  #endif
424  	/*
425  	 * Initially, crash hotplug support for kexec_load was added
426  	 * with the KEXEC_UPDATE_ELFCOREHDR flag. Later, this
427  	 * functionality was expanded to accommodate multiple kexec
428  	 * segment updates, leading to the introduction of the
429  	 * KEXEC_CRASH_HOTPLUG_SUPPORT kexec flag bit. Consequently,
430  	 * when the kexec tool sends either of these flags, it indicates
431  	 * that the required kexec segment (elfcorehdr) is excluded from
432  	 * the SHA calculation.
433  	 */
434  	return (kexec_flags & KEXEC_UPDATE_ELFCOREHDR ||
435  		kexec_flags & KEXEC_CRASH_HOTPLUG_SUPPORT);
436  }
437  
arch_crash_get_elfcorehdr_size(void)438  unsigned int arch_crash_get_elfcorehdr_size(void)
439  {
440  	unsigned int sz;
441  
442  	/* kernel_map, VMCOREINFO and maximum CPUs */
443  	sz = 2 + CONFIG_NR_CPUS_DEFAULT;
444  	if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
445  		sz += CONFIG_CRASH_MAX_MEMORY_RANGES;
446  	sz *= sizeof(Elf64_Phdr);
447  	return sz;
448  }
449  
450  /**
451   * arch_crash_handle_hotplug_event() - Handle hotplug elfcorehdr changes
452   * @image: a pointer to kexec_crash_image
453   * @arg: struct memory_notify handler for memory hotplug case and
454   *       NULL for CPU hotplug case.
455   *
456   * Prepare the new elfcorehdr and replace the existing elfcorehdr.
457   */
arch_crash_handle_hotplug_event(struct kimage * image,void * arg)458  void arch_crash_handle_hotplug_event(struct kimage *image, void *arg)
459  {
460  	void *elfbuf = NULL, *old_elfcorehdr;
461  	unsigned long nr_mem_ranges;
462  	unsigned long mem, memsz;
463  	unsigned long elfsz = 0;
464  
465  	/*
466  	 * As crash_prepare_elf64_headers() has already described all
467  	 * possible CPUs, there is no need to update the elfcorehdr
468  	 * for additional CPU changes.
469  	 */
470  	if ((image->file_mode || image->elfcorehdr_updated) &&
471  		((image->hp_action == KEXEC_CRASH_HP_ADD_CPU) ||
472  		(image->hp_action == KEXEC_CRASH_HP_REMOVE_CPU)))
473  		return;
474  
475  	/*
476  	 * Create the new elfcorehdr reflecting the changes to CPU and/or
477  	 * memory resources.
478  	 */
479  	if (prepare_elf_headers(&elfbuf, &elfsz, &nr_mem_ranges)) {
480  		pr_err("unable to create new elfcorehdr");
481  		goto out;
482  	}
483  
484  	/*
485  	 * Obtain address and size of the elfcorehdr segment, and
486  	 * check it against the new elfcorehdr buffer.
487  	 */
488  	mem = image->segment[image->elfcorehdr_index].mem;
489  	memsz = image->segment[image->elfcorehdr_index].memsz;
490  	if (elfsz > memsz) {
491  		pr_err("update elfcorehdr elfsz %lu > memsz %lu",
492  			elfsz, memsz);
493  		goto out;
494  	}
495  
496  	/*
497  	 * Copy new elfcorehdr over the old elfcorehdr at destination.
498  	 */
499  	old_elfcorehdr = kmap_local_page(pfn_to_page(mem >> PAGE_SHIFT));
500  	if (!old_elfcorehdr) {
501  		pr_err("mapping elfcorehdr segment failed\n");
502  		goto out;
503  	}
504  
505  	/*
506  	 * Temporarily invalidate the crash image while the
507  	 * elfcorehdr is updated.
508  	 */
509  	xchg(&kexec_crash_image, NULL);
510  	memcpy_flushcache(old_elfcorehdr, elfbuf, elfsz);
511  	xchg(&kexec_crash_image, image);
512  	kunmap_local(old_elfcorehdr);
513  	pr_debug("updated elfcorehdr\n");
514  
515  out:
516  	vfree(elfbuf);
517  }
518  #endif
519