Re: [PATCH v4 1/3] crash: Exclude crash kernel memory in crash core

[Sourabh Jain <sourabhjain@linux.ibm.com>]

On 12/02/26 08:58, Jinjie Ruan wrote:
>
> On 2026/2/10 20:30, Sourabh Jain wrote:
>> Hello Jinjie,
>>
>> On 09/02/26 15:29, Jinjie Ruan wrote:
>>> The exclude of crashk_res, crashk_low_res and crashk_cma memory
>>> are almost identical across different architectures, handling them
>>> in the crash core would eliminate a lot of duplication, so do
>>> them in the common code.
>>>
>>> And move the size calculation (and the realloc if needed) into the
>>> generic crash core so that:
>>>
>>> - New CMA regions or future crash-memory types can automatically
>>>     accounted for in crash core;
>>>
>>> - Each architecture no longer has to play whack-a-mole with
>>>     its private array size.
>>>
>>> To achieve the above goal, 4 architecture-specific functions are
>>> introduced:
>>>
>>> - arch_get_system_nr_ranges() and arch_prepare_elf64_ram_headers().
>>>     The 1st function pre-counts the number of memory ranges, and
>>>     the 2st function fill the memory ranges into the cmem->ranges[] array,
>>>     and count the actual number of ranges filled. The default
>>> implementation
>>>     is consistent with arm64 and loongson.
>>>
>>> - arch_crash_exclude_mem_range(). Realloc for powerpc. The default
>>>     implementation is crash_exclude_mem_range(), and use
>>>     crash_exclude_mem_range_guarded() to implement the arch version
>>>     for powerpc.
>>>
>>> - arch_get_crash_memory_ranges(). Get crash memory ranges for arch and
>>>     the default implementation is generic across x86, arm64, riscv, and
>>>     loongson by using the first two arch functions above. powerpc has its
>>>     own implementation by calling get_crash_memory_ranges().
>>>
>>> Tested on x86, arm64 and riscv with QEMU.
>>>
>>> Signed-off-by: Jinjie Ruan <ruanjinjie@huawei.com>
>>> ---
>>>    arch/arm64/kernel/machine_kexec_file.c     |  47 +--------
>>>    arch/loongarch/kernel/machine_kexec_file.c |  45 +-------
>>>    arch/powerpc/include/asm/kexec.h           |  13 +++
>>>    arch/powerpc/kexec/crash.c                 |  52 ++++++----
>>>    arch/powerpc/kexec/file_load_64.c          |  17 ++-
>>>    arch/powerpc/kexec/ranges.c                |  18 +---
>>>    arch/riscv/include/asm/kexec.h             |  10 ++
>>>    arch/riscv/kernel/machine_kexec_file.c     |  37 ++-----
>>>    arch/x86/include/asm/kexec.h               |  10 ++
>>>    arch/x86/kernel/crash.c                    | 104 ++-----------------
>>>    include/linux/crash_core.h                 | 114 +++++++++++++++++++--
>>>    kernel/crash_core.c                        |  71 +++++++++++--
>>>    12 files changed, 269 insertions(+), 269 deletions(-)
>>>
> [...]
>
>>>    extern void crash_ipi_callback(struct pt_regs *regs);
>>> diff --git a/arch/powerpc/kexec/crash.c b/arch/powerpc/kexec/crash.c
>>> index a325c1c02f96..5ade9a853fb0 100644
>>> --- a/arch/powerpc/kexec/crash.c
>>> +++ b/arch/powerpc/kexec/crash.c
>>> @@ -419,30 +419,21 @@ unsigned int arch_crash_get_elfcorehdr_size(void)
>>>        return sizeof(struct elfhdr) + (phdr_cnt * sizeof(Elf64_Phdr));
>>>    }
>>>    -/**
>>> - * update_crash_elfcorehdr() - Recreate the elfcorehdr and replace it
>>> with old
>>> - *                   elfcorehdr in the kexec segment array.
>>> - * @image: the active struct kimage
>>> - * @mn: struct memory_notify data handler
>>> - */
>>> -static void update_crash_elfcorehdr(struct kimage *image, struct
>>> memory_notify *mn)
>>> +int arch_get_crash_memory_ranges(struct crash_mem **cmem, unsigned
>>> long *nr_mem_ranges,
>>> +                 struct kimage *image, struct memory_notify *mn)
>>>    {
>>> +    unsigned long base_addr, size;
>>>        int ret;
>>> -    struct crash_mem *cmem = NULL;
>>> -    struct kexec_segment *ksegment;
>>> -    void *ptr, *mem, *elfbuf = NULL;
>>> -    unsigned long elfsz, memsz, base_addr, size;
>>>    -    ksegment = &image->segment[image->elfcorehdr_index];
>>> -    mem = (void *) ksegment->mem;
>>> -    memsz = ksegment->memsz;
>>> -
>>> -    ret = get_crash_memory_ranges(&cmem);
>>> +    ret = get_crash_memory_ranges(cmem);
>>>        if (ret) {
>>>            pr_err("Failed to get crash mem range\n");
>>> -        return;
>>> +        return ret;
>>>        }
>>>    +    if (!image || !mn)
>>> +        return 0;
>>> +
>>>        /*
>>>         * The hot unplugged memory is part of crash memory ranges,
>>>         * remove it here.
>>> @@ -450,14 +441,34 @@ static void update_crash_elfcorehdr(struct
>>> kimage *image, struct memory_notify *
>>>        if (image->hp_action == KEXEC_CRASH_HP_REMOVE_MEMORY) {
>>>            base_addr = PFN_PHYS(mn->start_pfn);
>>>            size = mn->nr_pages * PAGE_SIZE;
>>> -        ret = remove_mem_range(&cmem, base_addr, size);
>>> +        ret = remove_mem_range(cmem, base_addr, size);
>> I like the overall design for handling crashkernel memory exclusion
>> in this patch series, especially the way you managed to free the
>> crash_mem object (mem) in the generic code (crash_prepare_elf64_headers()).
> Thanks for the review.
>
>> However, the way crash memory is prepared after a memory hotplug
>> event on powerpc by calling remove_mem_range(), can leave the crash
>> memory ranges unsorted. This can cause issues in the generic code
>> when excluding crashkernel memory, because crash_exclude_mem_range()
>> expects crash_mem to be sorted.
> You are absolutely correct.
>
>> So I wrote a simple patch to cover this scenario. Including the
>> patch below as the first patch in this series would be helpful.
>> https://lore.kernel.org/all/20260210120803.433978-1-sourabhjain@linux.ibm.com/
> Thanks for the additional patch. I'll add it as the first patch in the
> next revision to ensure crash_mem remains sorted after memory hotplug
> events on powerpc.

Thanks you.

Please use the latest version (v2) available here:
https://lore.kernel.org/all/20260212060159.733023-1-sourabhjain@linux.ibm.com/

Regards,
Sourabh Jain