From a2e9a95d2190ef55bf0724ecdf8a466d393a86b6 Mon Sep 17 00:00:00 2001 From: Lianbo Jiang Date: Tue, 4 Aug 2020 12:49:32 +0800 Subject: kexec: Improve & fix crash_exclude_mem_range() to handle overlapping ranges The crash_exclude_mem_range() function can only handle one memory region a time. It will fail in the case in which the passed in area covers several memory regions. In this case, it will only exclude the first region, then return, but leave the later regions unsolved. E.g in a NEC system with two usable RAM regions inside the low 1M: ... BIOS-e820: [mem 0x0000000000000000-0x000000000003efff] usable BIOS-e820: [mem 0x000000000003f000-0x000000000003ffff] reserved BIOS-e820: [mem 0x0000000000040000-0x000000000009ffff] usable It will only exclude the memory region [0, 0x3efff], the memory region [0x40000, 0x9ffff] will still be added into /proc/vmcore, which may cause the following failure when dumping vmcore: ioremap on RAM at 0x0000000000040000 - 0x0000000000040fff WARNING: CPU: 0 PID: 665 at arch/x86/mm/ioremap.c:186 __ioremap_caller+0x2c7/0x2e0 ... RIP: 0010:__ioremap_caller+0x2c7/0x2e0 ... cp: error reading '/proc/vmcore': Cannot allocate memory kdump: saving vmcore failed In order to fix this bug, let's extend the crash_exclude_mem_range() to handle the overlapping ranges. [ mingo: Amended the changelog. ] Signed-off-by: Lianbo Jiang Signed-off-by: Ingo Molnar Acked-by: Dave Young Link: https://lore.kernel.org/r/20200804044933.1973-3-lijiang@redhat.com --- kernel/kexec_file.c | 35 +++++++++++++++++++++++------------ 1 file changed, 23 insertions(+), 12 deletions(-) (limited to 'kernel/kexec_file.c') diff --git a/kernel/kexec_file.c b/kernel/kexec_file.c index 94661d2d13ad..97fa68267197 100644 --- a/kernel/kexec_file.c +++ b/kernel/kexec_file.c @@ -1157,24 +1157,26 @@ int crash_exclude_mem_range(struct crash_mem *mem, unsigned long long mstart, unsigned long long mend) { int i, j; - unsigned long long start, end; + unsigned long long start, end, p_start, p_end; struct crash_mem_range temp_range = {0, 0}; for (i = 0; i < mem->nr_ranges; i++) { start = mem->ranges[i].start; end = mem->ranges[i].end; + p_start = mstart; + p_end = mend; if (mstart > end || mend < start) continue; /* Truncate any area outside of range */ if (mstart < start) - mstart = start; + p_start = start; if (mend > end) - mend = end; + p_end = end; /* Found completely overlapping range */ - if (mstart == start && mend == end) { + if (p_start == start && p_end == end) { mem->ranges[i].start = 0; mem->ranges[i].end = 0; if (i < mem->nr_ranges - 1) { @@ -1185,20 +1187,29 @@ int crash_exclude_mem_range(struct crash_mem *mem, mem->ranges[j].end = mem->ranges[j+1].end; } + + /* + * Continue to check if there are another overlapping ranges + * from the current position because of shifting the above + * mem ranges. + */ + i--; + mem->nr_ranges--; + continue; } mem->nr_ranges--; return 0; } - if (mstart > start && mend < end) { + if (p_start > start && p_end < end) { /* Split original range */ - mem->ranges[i].end = mstart - 1; - temp_range.start = mend + 1; + mem->ranges[i].end = p_start - 1; + temp_range.start = p_end + 1; temp_range.end = end; - } else if (mstart != start) - mem->ranges[i].end = mstart - 1; + } else if (p_start != start) + mem->ranges[i].end = p_start - 1; else - mem->ranges[i].start = mend + 1; + mem->ranges[i].start = p_end + 1; break; } @@ -1243,7 +1254,7 @@ int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map, * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64). * I think this is required by tools like gdb. So same physical - * memory will be mapped in two elf headers. One will contain kernel + * memory will be mapped in two elf headers. One will contain kernel * text virtual addresses and other will have __va(physical) addresses. */ @@ -1270,7 +1281,7 @@ int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map, ehdr->e_ehsize = sizeof(Elf64_Ehdr); ehdr->e_phentsize = sizeof(Elf64_Phdr); - /* Prepare one phdr of type PT_NOTE for each present cpu */ + /* Prepare one phdr of type PT_NOTE for each present CPU */ for_each_present_cpu(cpu) { phdr->p_type = PT_NOTE; notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu)); -- cgit v1.2.3 From 475f63ae63b5102ae6423d1712333929d04d6ecc Mon Sep 17 00:00:00 2001 From: Lianbo Jiang Date: Tue, 4 Aug 2020 12:49:33 +0800 Subject: kexec_file: Correctly output debugging information for the PT_LOAD ELF header Currently, when we enable the debugging switch to debug kexec_file, we always get the following incorrect results: kexec_file: Crash PT_LOAD elf header. phdr=00000000c988639b vaddr=0x0, paddr=0x0, sz=0x0 e_phnum=51 p_offset=0x0 kexec_file: Crash PT_LOAD elf header. phdr=000000003cca69a0 vaddr=0x0, paddr=0x0, sz=0x0 e_phnum=52 p_offset=0x0 kexec_file: Crash PT_LOAD elf header. phdr=00000000c584cb9f vaddr=0x0, paddr=0x0, sz=0x0 e_phnum=53 p_offset=0x0 kexec_file: Crash PT_LOAD elf header. phdr=00000000cf85d57f vaddr=0x0, paddr=0x0, sz=0x0 e_phnum=54 p_offset=0x0 kexec_file: Crash PT_LOAD elf header. phdr=00000000a4a8f847 vaddr=0x0, paddr=0x0, sz=0x0 e_phnum=55 p_offset=0x0 kexec_file: Crash PT_LOAD elf header. phdr=00000000272ec49f vaddr=0x0, paddr=0x0, sz=0x0 e_phnum=56 p_offset=0x0 kexec_file: Crash PT_LOAD elf header. phdr=00000000ea0b65de vaddr=0x0, paddr=0x0, sz=0x0 e_phnum=57 p_offset=0x0 kexec_file: Crash PT_LOAD elf header. phdr=000000001f5e490c vaddr=0x0, paddr=0x0, sz=0x0 e_phnum=58 p_offset=0x0 kexec_file: Crash PT_LOAD elf header. phdr=00000000dfe4109e vaddr=0x0, paddr=0x0, sz=0x0 e_phnum=59 p_offset=0x0 kexec_file: Crash PT_LOAD elf header. phdr=00000000480ed2b6 vaddr=0x0, paddr=0x0, sz=0x0 e_phnum=60 p_offset=0x0 kexec_file: Crash PT_LOAD elf header. phdr=0000000080b65151 vaddr=0x0, paddr=0x0, sz=0x0 e_phnum=61 p_offset=0x0 kexec_file: Crash PT_LOAD elf header. phdr=0000000024e31c5e vaddr=0x0, paddr=0x0, sz=0x0 e_phnum=62 p_offset=0x0 kexec_file: Crash PT_LOAD elf header. phdr=00000000332e0385 vaddr=0x0, paddr=0x0, sz=0x0 e_phnum=63 p_offset=0x0 kexec_file: Crash PT_LOAD elf header. phdr=000000002754d5da vaddr=0x0, paddr=0x0, sz=0x0 e_phnum=64 p_offset=0x0 kexec_file: Crash PT_LOAD elf header. phdr=00000000783320dd vaddr=0x0, paddr=0x0, sz=0x0 e_phnum=65 p_offset=0x0 kexec_file: Crash PT_LOAD elf header. phdr=0000000076fe5b64 vaddr=0x0, paddr=0x0, sz=0x0 e_phnum=66 p_offset=0x0 The reason is that kernel always prints the values of the next PT_LOAD instead of the current PT_LOAD. Change it to ensure that we can get the correct debugging information. [ mingo: Amended changelog, capitalized "ELF". ] Signed-off-by: Lianbo Jiang Signed-off-by: Ingo Molnar Acked-by: Dave Young Link: https://lore.kernel.org/r/20200804044933.1973-4-lijiang@redhat.com --- kernel/kexec_file.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'kernel/kexec_file.c') diff --git a/kernel/kexec_file.c b/kernel/kexec_file.c index 97fa68267197..3f7867c1820f 100644 --- a/kernel/kexec_file.c +++ b/kernel/kexec_file.c @@ -1246,7 +1246,7 @@ int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map, unsigned long long notes_addr; unsigned long mstart, mend; - /* extra phdr for vmcoreinfo elf note */ + /* extra phdr for vmcoreinfo ELF note */ nr_phdr = nr_cpus + 1; nr_phdr += mem->nr_ranges; @@ -1254,7 +1254,7 @@ int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map, * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64). * I think this is required by tools like gdb. So same physical - * memory will be mapped in two elf headers. One will contain kernel + * memory will be mapped in two ELF headers. One will contain kernel * text virtual addresses and other will have __va(physical) addresses. */ @@ -1323,10 +1323,10 @@ int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map, phdr->p_filesz = phdr->p_memsz = mend - mstart + 1; phdr->p_align = 0; ehdr->e_phnum++; - phdr++; - pr_debug("Crash PT_LOAD elf header. phdr=%p vaddr=0x%llx, paddr=0x%llx, sz=0x%llx e_phnum=%d p_offset=0x%llx\n", + pr_debug("Crash PT_LOAD ELF header. phdr=%p vaddr=0x%llx, paddr=0x%llx, sz=0x%llx e_phnum=%d p_offset=0x%llx\n", phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz, ehdr->e_phnum, phdr->p_offset); + phdr++; } *addr = buf; -- cgit v1.2.3