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