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/processor.h>
30 #include <asm/hardirq.h>
31 #include <asm/nmi.h>
32 #include <asm/hw_irq.h>
33 #include <asm/apic.h>
34 #include <asm/e820/types.h>
35 #include <asm/io_apic.h>
36 #include <asm/hpet.h>
37 #include <linux/kdebug.h>
38 #include <asm/cpu.h>
39 #include <asm/reboot.h>
40 #include <asm/virtext.h>
41 #include <asm/intel_pt.h>
42 #include <asm/crash.h>
43 #include <asm/cmdline.h>
44
45 /* Used while preparing memory map entries for second kernel */
46 struct crash_memmap_data {
47 struct boot_params *params;
48 /* Type of memory */
49 unsigned int type;
50 };
51
52 /*
53 * This is used to VMCLEAR all VMCSs loaded on the
54 * processor. And when loading kvm_intel module, the
55 * callback function pointer will be assigned.
56 *
57 * protected by rcu.
58 */
59 crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss = NULL;
60 EXPORT_SYMBOL_GPL(crash_vmclear_loaded_vmcss);
61
cpu_crash_vmclear_loaded_vmcss(void)62 static inline void cpu_crash_vmclear_loaded_vmcss(void)
63 {
64 crash_vmclear_fn *do_vmclear_operation = NULL;
65
66 rcu_read_lock();
67 do_vmclear_operation = rcu_dereference(crash_vmclear_loaded_vmcss);
68 if (do_vmclear_operation)
69 do_vmclear_operation();
70 rcu_read_unlock();
71 }
72
73 /*
74 * When the crashkernel option is specified, only use the low
75 * 1M for the real mode trampoline.
76 */
crash_reserve_low_1M(void)77 void __init crash_reserve_low_1M(void)
78 {
79 if (cmdline_find_option(boot_command_line, "crashkernel", NULL, 0) < 0)
80 return;
81
82 memblock_reserve(0, 1<<20);
83 pr_info("Reserving the low 1M of memory for crashkernel\n");
84 }
85
86 #if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)
87
kdump_nmi_callback(int cpu,struct pt_regs * regs)88 static void kdump_nmi_callback(int cpu, struct pt_regs *regs)
89 {
90 crash_save_cpu(regs, cpu);
91
92 /*
93 * VMCLEAR VMCSs loaded on all cpus if needed.
94 */
95 cpu_crash_vmclear_loaded_vmcss();
96
97 /* Disable VMX or SVM if needed.
98 *
99 * We need to disable virtualization on all CPUs.
100 * Having VMX or SVM enabled on any CPU may break rebooting
101 * after the kdump kernel has finished its task.
102 */
103 cpu_emergency_vmxoff();
104 cpu_emergency_svm_disable();
105
106 /*
107 * Disable Intel PT to stop its logging
108 */
109 cpu_emergency_stop_pt();
110
111 disable_local_APIC();
112 }
113
kdump_nmi_shootdown_cpus(void)114 void kdump_nmi_shootdown_cpus(void)
115 {
116 nmi_shootdown_cpus(kdump_nmi_callback);
117
118 disable_local_APIC();
119 }
120
121 /* Override the weak function in kernel/panic.c */
crash_smp_send_stop(void)122 void crash_smp_send_stop(void)
123 {
124 static int cpus_stopped;
125
126 if (cpus_stopped)
127 return;
128
129 if (smp_ops.crash_stop_other_cpus)
130 smp_ops.crash_stop_other_cpus();
131 else
132 smp_send_stop();
133
134 cpus_stopped = 1;
135 }
136
137 #else
crash_smp_send_stop(void)138 void crash_smp_send_stop(void)
139 {
140 /* There are no cpus to shootdown */
141 }
142 #endif
143
native_machine_crash_shutdown(struct pt_regs * regs)144 void native_machine_crash_shutdown(struct pt_regs *regs)
145 {
146 /* This function is only called after the system
147 * has panicked or is otherwise in a critical state.
148 * The minimum amount of code to allow a kexec'd kernel
149 * to run successfully needs to happen here.
150 *
151 * In practice this means shooting down the other cpus in
152 * an SMP system.
153 */
154 /* The kernel is broken so disable interrupts */
155 local_irq_disable();
156
157 crash_smp_send_stop();
158
159 /*
160 * VMCLEAR VMCSs loaded on this cpu if needed.
161 */
162 cpu_crash_vmclear_loaded_vmcss();
163
164 /* Booting kdump kernel with VMX or SVM enabled won't work,
165 * because (among other limitations) we can't disable paging
166 * with the virt flags.
167 */
168 cpu_emergency_vmxoff();
169 cpu_emergency_svm_disable();
170
171 /*
172 * Disable Intel PT to stop its logging
173 */
174 cpu_emergency_stop_pt();
175
176 #ifdef CONFIG_X86_IO_APIC
177 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */
178 ioapic_zap_locks();
179 clear_IO_APIC();
180 #endif
181 lapic_shutdown();
182 restore_boot_irq_mode();
183 #ifdef CONFIG_HPET_TIMER
184 hpet_disable();
185 #endif
186 crash_save_cpu(regs, safe_smp_processor_id());
187 }
188
189 #ifdef CONFIG_KEXEC_FILE
190
get_nr_ram_ranges_callback(struct resource * res,void * arg)191 static int get_nr_ram_ranges_callback(struct resource *res, void *arg)
192 {
193 unsigned int *nr_ranges = arg;
194
195 (*nr_ranges)++;
196 return 0;
197 }
198
199 /* Gather all the required information to prepare elf headers for ram regions */
fill_up_crash_elf_data(void)200 static struct crash_mem *fill_up_crash_elf_data(void)
201 {
202 unsigned int nr_ranges = 0;
203 struct crash_mem *cmem;
204
205 walk_system_ram_res(0, -1, &nr_ranges, get_nr_ram_ranges_callback);
206 if (!nr_ranges)
207 return NULL;
208
209 /*
210 * Exclusion of crash region and/or crashk_low_res may cause
211 * another range split. So add extra two slots here.
212 */
213 nr_ranges += 2;
214 cmem = vzalloc(struct_size(cmem, ranges, nr_ranges));
215 if (!cmem)
216 return NULL;
217
218 cmem->max_nr_ranges = nr_ranges;
219 cmem->nr_ranges = 0;
220
221 return cmem;
222 }
223
224 /*
225 * Look for any unwanted ranges between mstart, mend and remove them. This
226 * might lead to split and split ranges are put in cmem->ranges[] array
227 */
elf_header_exclude_ranges(struct crash_mem * cmem)228 static int elf_header_exclude_ranges(struct crash_mem *cmem)
229 {
230 int ret = 0;
231
232 /* Exclude the low 1M because it is always reserved */
233 ret = crash_exclude_mem_range(cmem, 0, (1<<20)-1);
234 if (ret)
235 return ret;
236
237 /* Exclude crashkernel region */
238 ret = crash_exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
239 if (ret)
240 return ret;
241
242 if (crashk_low_res.end)
243 ret = crash_exclude_mem_range(cmem, crashk_low_res.start,
244 crashk_low_res.end);
245
246 return ret;
247 }
248
prepare_elf64_ram_headers_callback(struct resource * res,void * arg)249 static int prepare_elf64_ram_headers_callback(struct resource *res, void *arg)
250 {
251 struct crash_mem *cmem = arg;
252
253 cmem->ranges[cmem->nr_ranges].start = res->start;
254 cmem->ranges[cmem->nr_ranges].end = res->end;
255 cmem->nr_ranges++;
256
257 return 0;
258 }
259
260 /* Prepare elf headers. Return addr and size */
prepare_elf_headers(struct kimage * image,void ** addr,unsigned long * sz)261 static int prepare_elf_headers(struct kimage *image, void **addr,
262 unsigned long *sz)
263 {
264 struct crash_mem *cmem;
265 int ret;
266
267 cmem = fill_up_crash_elf_data();
268 if (!cmem)
269 return -ENOMEM;
270
271 ret = walk_system_ram_res(0, -1, cmem, prepare_elf64_ram_headers_callback);
272 if (ret)
273 goto out;
274
275 /* Exclude unwanted mem ranges */
276 ret = elf_header_exclude_ranges(cmem);
277 if (ret)
278 goto out;
279
280 /* By default prepare 64bit headers */
281 ret = crash_prepare_elf64_headers(cmem, IS_ENABLED(CONFIG_X86_64), addr, sz);
282
283 out:
284 vfree(cmem);
285 return ret;
286 }
287
add_e820_entry(struct boot_params * params,struct e820_entry * entry)288 static int add_e820_entry(struct boot_params *params, struct e820_entry *entry)
289 {
290 unsigned int nr_e820_entries;
291
292 nr_e820_entries = params->e820_entries;
293 if (nr_e820_entries >= E820_MAX_ENTRIES_ZEROPAGE)
294 return 1;
295
296 memcpy(¶ms->e820_table[nr_e820_entries], entry, sizeof(struct e820_entry));
297 params->e820_entries++;
298 return 0;
299 }
300
memmap_entry_callback(struct resource * res,void * arg)301 static int memmap_entry_callback(struct resource *res, void *arg)
302 {
303 struct crash_memmap_data *cmd = arg;
304 struct boot_params *params = cmd->params;
305 struct e820_entry ei;
306
307 ei.addr = res->start;
308 ei.size = resource_size(res);
309 ei.type = cmd->type;
310 add_e820_entry(params, &ei);
311
312 return 0;
313 }
314
memmap_exclude_ranges(struct kimage * image,struct crash_mem * cmem,unsigned long long mstart,unsigned long long mend)315 static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
316 unsigned long long mstart,
317 unsigned long long mend)
318 {
319 unsigned long start, end;
320
321 cmem->ranges[0].start = mstart;
322 cmem->ranges[0].end = mend;
323 cmem->nr_ranges = 1;
324
325 /* Exclude elf header region */
326 start = image->arch.elf_load_addr;
327 end = start + image->arch.elf_headers_sz - 1;
328 return crash_exclude_mem_range(cmem, start, end);
329 }
330
331 /* Prepare memory map for crash dump kernel */
crash_setup_memmap_entries(struct kimage * image,struct boot_params * params)332 int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
333 {
334 int i, ret = 0;
335 unsigned long flags;
336 struct e820_entry ei;
337 struct crash_memmap_data cmd;
338 struct crash_mem *cmem;
339
340 cmem = vzalloc(struct_size(cmem, ranges, 1));
341 if (!cmem)
342 return -ENOMEM;
343
344 memset(&cmd, 0, sizeof(struct crash_memmap_data));
345 cmd.params = params;
346
347 /* Add the low 1M */
348 cmd.type = E820_TYPE_RAM;
349 flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
350 walk_iomem_res_desc(IORES_DESC_NONE, flags, 0, (1<<20)-1, &cmd,
351 memmap_entry_callback);
352
353 /* Add ACPI tables */
354 cmd.type = E820_TYPE_ACPI;
355 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
356 walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd,
357 memmap_entry_callback);
358
359 /* Add ACPI Non-volatile Storage */
360 cmd.type = E820_TYPE_NVS;
361 walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, &cmd,
362 memmap_entry_callback);
363
364 /* Add e820 reserved ranges */
365 cmd.type = E820_TYPE_RESERVED;
366 flags = IORESOURCE_MEM;
367 walk_iomem_res_desc(IORES_DESC_RESERVED, flags, 0, -1, &cmd,
368 memmap_entry_callback);
369
370 /* Add crashk_low_res region */
371 if (crashk_low_res.end) {
372 ei.addr = crashk_low_res.start;
373 ei.size = resource_size(&crashk_low_res);
374 ei.type = E820_TYPE_RAM;
375 add_e820_entry(params, &ei);
376 }
377
378 /* Exclude some ranges from crashk_res and add rest to memmap */
379 ret = memmap_exclude_ranges(image, cmem, crashk_res.start, crashk_res.end);
380 if (ret)
381 goto out;
382
383 for (i = 0; i < cmem->nr_ranges; i++) {
384 ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
385
386 /* If entry is less than a page, skip it */
387 if (ei.size < PAGE_SIZE)
388 continue;
389 ei.addr = cmem->ranges[i].start;
390 ei.type = E820_TYPE_RAM;
391 add_e820_entry(params, &ei);
392 }
393
394 out:
395 vfree(cmem);
396 return ret;
397 }
398
crash_load_segments(struct kimage * image)399 int crash_load_segments(struct kimage *image)
400 {
401 int ret;
402 struct kexec_buf kbuf = { .image = image, .buf_min = 0,
403 .buf_max = ULONG_MAX, .top_down = false };
404
405 /* Prepare elf headers and add a segment */
406 ret = prepare_elf_headers(image, &kbuf.buffer, &kbuf.bufsz);
407 if (ret)
408 return ret;
409
410 image->arch.elf_headers = kbuf.buffer;
411 image->arch.elf_headers_sz = kbuf.bufsz;
412
413 kbuf.memsz = kbuf.bufsz;
414 kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
415 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
416 ret = kexec_add_buffer(&kbuf);
417 if (ret) {
418 vfree((void *)image->arch.elf_headers);
419 return ret;
420 }
421 image->arch.elf_load_addr = kbuf.mem;
422 pr_debug("Loaded ELF headers at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
423 image->arch.elf_load_addr, kbuf.bufsz, kbuf.bufsz);
424
425 return ret;
426 }
427 #endif /* CONFIG_KEXEC_FILE */
428