2 * Architecture specific (i386/x86_64) functions for kexec based crash dumps.
4 * Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
6 * Copyright (C) IBM Corporation, 2004. All rights reserved.
7 * Copyright (C) Red Hat Inc., 2014. All rights reserved.
9 * Vivek Goyal <vgoyal@redhat.com>
13 #define pr_fmt(fmt) "kexec: " fmt
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>
27 #include <asm/processor.h>
28 #include <asm/hardirq.h>
30 #include <asm/hw_irq.h>
32 #include <asm/e820/types.h>
33 #include <asm/io_apic.h>
35 #include <linux/kdebug.h>
37 #include <asm/reboot.h>
38 #include <asm/virtext.h>
39 #include <asm/intel_pt.h>
40 #include <asm/crash.h>
42 /* Used while preparing memory map entries for second kernel */
43 struct crash_memmap_data {
44 struct boot_params *params;
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.
56 crash_vmclear_fn __rcu *crash_vmclear_loaded_vmcss = NULL;
57 EXPORT_SYMBOL_GPL(crash_vmclear_loaded_vmcss);
58 unsigned long crash_zero_bytes;
60 static inline void cpu_crash_vmclear_loaded_vmcss(void)
62 crash_vmclear_fn *do_vmclear_operation = NULL;
65 do_vmclear_operation = rcu_dereference(crash_vmclear_loaded_vmcss);
66 if (do_vmclear_operation)
67 do_vmclear_operation();
71 #if defined(CONFIG_SMP) && defined(CONFIG_X86_LOCAL_APIC)
73 static void kdump_nmi_callback(int cpu, struct pt_regs *regs)
76 struct pt_regs fixed_regs;
78 if (!user_mode(regs)) {
79 crash_fixup_ss_esp(&fixed_regs, regs);
83 crash_save_cpu(regs, cpu);
86 * VMCLEAR VMCSs loaded on all cpus if needed.
88 cpu_crash_vmclear_loaded_vmcss();
90 /* Disable VMX or SVM if needed.
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.
96 cpu_emergency_vmxoff();
97 cpu_emergency_svm_disable();
100 * Disable Intel PT to stop its logging
102 cpu_emergency_stop_pt();
104 disable_local_APIC();
107 void kdump_nmi_shootdown_cpus(void)
109 nmi_shootdown_cpus(kdump_nmi_callback);
111 disable_local_APIC();
114 /* Override the weak function in kernel/panic.c */
115 void crash_smp_send_stop(void)
117 static int cpus_stopped;
122 if (smp_ops.crash_stop_other_cpus)
123 smp_ops.crash_stop_other_cpus();
131 void crash_smp_send_stop(void)
133 /* There are no cpus to shootdown */
137 void native_machine_crash_shutdown(struct pt_regs *regs)
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.
144 * In practice this means shooting down the other cpus in
147 /* The kernel is broken so disable interrupts */
150 crash_smp_send_stop();
153 * VMCLEAR VMCSs loaded on this cpu if needed.
155 cpu_crash_vmclear_loaded_vmcss();
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.
161 cpu_emergency_vmxoff();
162 cpu_emergency_svm_disable();
165 * Disable Intel PT to stop its logging
167 cpu_emergency_stop_pt();
169 #ifdef CONFIG_X86_IO_APIC
170 /* Prevent crash_kexec() from deadlocking on ioapic_lock. */
175 restore_boot_irq_mode();
176 #ifdef CONFIG_HPET_TIMER
179 crash_save_cpu(regs, safe_smp_processor_id());
182 #ifdef CONFIG_KEXEC_FILE
183 static int get_nr_ram_ranges_callback(struct resource *res, void *arg)
185 unsigned int *nr_ranges = arg;
191 /* Gather all the required information to prepare elf headers for ram regions */
192 static struct crash_mem *fill_up_crash_elf_data(void)
194 unsigned int nr_ranges = 0;
195 struct crash_mem *cmem;
197 walk_system_ram_res(0, -1, &nr_ranges,
198 get_nr_ram_ranges_callback);
203 * Exclusion of crash region and/or crashk_low_res may cause
204 * another range split. So add extra two slots here.
207 cmem = vzalloc(struct_size(cmem, ranges, nr_ranges));
211 cmem->max_nr_ranges = nr_ranges;
218 * Look for any unwanted ranges between mstart, mend and remove them. This
219 * might lead to split and split ranges are put in cmem->ranges[] array
221 static int elf_header_exclude_ranges(struct crash_mem *cmem)
225 /* Exclude crashkernel region */
226 ret = crash_exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
230 if (crashk_low_res.end) {
231 ret = crash_exclude_mem_range(cmem, crashk_low_res.start,
240 static int prepare_elf64_ram_headers_callback(struct resource *res, void *arg)
242 struct crash_mem *cmem = arg;
244 cmem->ranges[cmem->nr_ranges].start = res->start;
245 cmem->ranges[cmem->nr_ranges].end = res->end;
251 /* Prepare elf headers. Return addr and size */
252 static int prepare_elf_headers(struct kimage *image, void **addr,
255 struct crash_mem *cmem;
260 cmem = fill_up_crash_elf_data();
264 ret = walk_system_ram_res(0, -1, cmem,
265 prepare_elf64_ram_headers_callback);
269 /* Exclude unwanted mem ranges */
270 ret = elf_header_exclude_ranges(cmem);
274 /* By default prepare 64bit headers */
275 ret = crash_prepare_elf64_headers(cmem,
276 IS_ENABLED(CONFIG_X86_64), addr, sz);
281 * If a range matches backup region, adjust offset to backup
284 ehdr = (Elf64_Ehdr *)*addr;
285 phdr = (Elf64_Phdr *)(ehdr + 1);
286 for (i = 0; i < ehdr->e_phnum; phdr++, i++)
287 if (phdr->p_type == PT_LOAD &&
288 phdr->p_paddr == image->arch.backup_src_start &&
289 phdr->p_memsz == image->arch.backup_src_sz) {
290 phdr->p_offset = image->arch.backup_load_addr;
298 static int add_e820_entry(struct boot_params *params, struct e820_entry *entry)
300 unsigned int nr_e820_entries;
302 nr_e820_entries = params->e820_entries;
303 if (nr_e820_entries >= E820_MAX_ENTRIES_ZEROPAGE)
306 memcpy(¶ms->e820_table[nr_e820_entries], entry,
307 sizeof(struct e820_entry));
308 params->e820_entries++;
312 static int memmap_entry_callback(struct resource *res, void *arg)
314 struct crash_memmap_data *cmd = arg;
315 struct boot_params *params = cmd->params;
316 struct e820_entry ei;
318 ei.addr = res->start;
319 ei.size = resource_size(res);
321 add_e820_entry(params, &ei);
326 static int memmap_exclude_ranges(struct kimage *image, struct crash_mem *cmem,
327 unsigned long long mstart,
328 unsigned long long mend)
330 unsigned long start, end;
333 cmem->ranges[0].start = mstart;
334 cmem->ranges[0].end = mend;
337 /* Exclude Backup region */
338 start = image->arch.backup_load_addr;
339 end = start + image->arch.backup_src_sz - 1;
340 ret = crash_exclude_mem_range(cmem, start, end);
344 /* Exclude elf header region */
345 start = image->arch.elf_load_addr;
346 end = start + image->arch.elf_headers_sz - 1;
347 return crash_exclude_mem_range(cmem, start, end);
350 /* Prepare memory map for crash dump kernel */
351 int crash_setup_memmap_entries(struct kimage *image, struct boot_params *params)
355 struct e820_entry ei;
356 struct crash_memmap_data cmd;
357 struct crash_mem *cmem;
359 cmem = vzalloc(sizeof(struct crash_mem));
363 memset(&cmd, 0, sizeof(struct crash_memmap_data));
366 /* Add first 640K segment */
367 ei.addr = image->arch.backup_src_start;
368 ei.size = image->arch.backup_src_sz;
369 ei.type = E820_TYPE_RAM;
370 add_e820_entry(params, &ei);
372 /* Add ACPI tables */
373 cmd.type = E820_TYPE_ACPI;
374 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
375 walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1, &cmd,
376 memmap_entry_callback);
378 /* Add ACPI Non-volatile Storage */
379 cmd.type = E820_TYPE_NVS;
380 walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1, &cmd,
381 memmap_entry_callback);
383 /* Add crashk_low_res region */
384 if (crashk_low_res.end) {
385 ei.addr = crashk_low_res.start;
386 ei.size = crashk_low_res.end - crashk_low_res.start + 1;
387 ei.type = E820_TYPE_RAM;
388 add_e820_entry(params, &ei);
391 /* Exclude some ranges from crashk_res and add rest to memmap */
392 ret = memmap_exclude_ranges(image, cmem, crashk_res.start,
397 for (i = 0; i < cmem->nr_ranges; i++) {
398 ei.size = cmem->ranges[i].end - cmem->ranges[i].start + 1;
400 /* If entry is less than a page, skip it */
401 if (ei.size < PAGE_SIZE)
403 ei.addr = cmem->ranges[i].start;
404 ei.type = E820_TYPE_RAM;
405 add_e820_entry(params, &ei);
413 static int determine_backup_region(struct resource *res, void *arg)
415 struct kimage *image = arg;
417 image->arch.backup_src_start = res->start;
418 image->arch.backup_src_sz = resource_size(res);
420 /* Expecting only one range for backup region */
424 int crash_load_segments(struct kimage *image)
427 struct kexec_buf kbuf = { .image = image, .buf_min = 0,
428 .buf_max = ULONG_MAX, .top_down = false };
431 * Determine and load a segment for backup area. First 640K RAM
432 * region is backup source
435 ret = walk_system_ram_res(KEXEC_BACKUP_SRC_START, KEXEC_BACKUP_SRC_END,
436 image, determine_backup_region);
438 /* Zero or postive return values are ok */
442 /* Add backup segment. */
443 if (image->arch.backup_src_sz) {
444 kbuf.buffer = &crash_zero_bytes;
445 kbuf.bufsz = sizeof(crash_zero_bytes);
446 kbuf.memsz = image->arch.backup_src_sz;
447 kbuf.buf_align = PAGE_SIZE;
449 * Ideally there is no source for backup segment. This is
450 * copied in purgatory after crash. Just add a zero filled
451 * segment for now to make sure checksum logic works fine.
453 ret = kexec_add_buffer(&kbuf);
456 image->arch.backup_load_addr = kbuf.mem;
457 pr_debug("Loaded backup region at 0x%lx backup_start=0x%lx memsz=0x%lx\n",
458 image->arch.backup_load_addr,
459 image->arch.backup_src_start, kbuf.memsz);
462 /* Prepare elf headers and add a segment */
463 ret = prepare_elf_headers(image, &kbuf.buffer, &kbuf.bufsz);
467 image->arch.elf_headers = kbuf.buffer;
468 image->arch.elf_headers_sz = kbuf.bufsz;
470 kbuf.memsz = kbuf.bufsz;
471 kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
472 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
473 ret = kexec_add_buffer(&kbuf);
475 vfree((void *)image->arch.elf_headers);
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);
484 #endif /* CONFIG_KEXEC_FILE */