Re: [PATCH v6 00/10] kdump: reduce vmcore size and capture time
From: Chen Wandun
Date: Fri Sep 04 2026 - 07:50:41 EST
On 9/3/26 5:38 PM, Baoquan He wrote:
On 09/03/26 at 03:43pm, Wandun wrote:
On 9/3/26 15:31, Baoquan He wrote:
On 09/03/26 at 03:05pm, Wandun wrote:
On 9/2/26 16:53, Baoquan He wrote:
Hi Wandun,
On 09/02/26 at 03:31pm, Wandun Chen wrote:
From: Wandun Chen <chenwandun@xxxxxxxxxxx>
On SoCs that carve out large firmware-owned reserved memory (GPU,
camera ISP, ...), kdump currently dumps those carveouts as part of
system RAM even though their contents are firmware state that is not
useful for kernel crash analysis.
This series introduces a MEMBLOCK_NODUMP flag in memblock to filter
reserved memory on DT-based architectures (arm64, riscv, loongarch).
Reserved regions default are marked MEMBLOCK_NODUMP so kdump omits them;
reusable CMA regions are different because their pages are handed back
to the buddy allocator and may carry crash-relevant data.
ACPI systems already filter reserved memory out of the vmcore through
their existing path; only DT-based systems currently fail to filter these
regions, which is what this series addresses. The flag lives in memblock
itself rather than in a DT-only structure, so the mechanism is generic and
both ACPI and DT systems can benefit from it (suggested by Rob, thanks) [1].
Thanks for the effort. I am not against this patchset, and I haven't
went through it carefully. Just from the cover letter, you mentioned
generic, I am wondering if this can be generic for excluding other
memory regions. Asking this because I try to find a way to exclude
unwanted memory regions too, please check below link where there's
the relevant discussion. We definitely don't like inventing wheels
time after time. Do you think this memblock region excluding can
be used in other places of kernel?
https://lore.kernel.org/all/aoLFai0gzzH2bGgy@MiWiFi-R3L-srv/T/#u
Thanks for taking a look, and for the pointer to the crash-wipe thread.
To answer your question directly: the exclusion mechanism in this series
is already generic, memblock_mark_nodump() is the entry point. Any code
that wants a region out of the vmcore just marks it MEMBLOCK_NODUMP, when
preparing vmcore elf header by walking memblock, these regionss are skipped;
This series is simply one caller, marking the DTS /reserved-memory and
/memreserve/ regions with MEMBLOCK_NODUMP.
Thanks, for memory not allocated from memblock, e.g buddy, is it still
good?
Yes, still good, only need to pass the physical range to memblock_mark_nodump().
That's interesting. If drivers or any component allocate memory from
buddy, then you will call memblock_mark_nodump() to mark the region
specifically (memblock_setclr_flag will split regions if it's a sub
region of existing memblock region), it won't impact anything?
IIUC, it won't impact anyting; memblock_mark_nodump() only set
the MEMBLOCK_NODUMP bit on the matching memblock.memory regions,
and from the buddy allocator's view, there is nothing changed.
The only effect is the region split you already pointed out: marking
a sub-range costs up to two extra entries in the memblock.memory
region array.
Best regards,
Wandun
x86 currently builds the vmcore elf header from iomem_resource, not
memblock; it could be switched to memblock, but that would bring no
benefit for the problem this series solves, so this series makes no
x86 change. I'd personally be in favor of switching to memblock for
a unified implementation across architectures.
Best regards
Wandun
Since the reserved memory regions are filtered out, the vmcore is
smaller in size and faster to produce.
The series is based on linux-next and is organized as follows:
Patches 1-4: Preparation and bugfixes: fix the missing HugeTLB
flagname, switch riscv crash_mem to memblock, fold the
duplicated per-arch memblock walks into the weak
defaults, and serialize crash header preparation against
memory hotplug.
Patches 5-9: NODUMP infrastructure: switch crash_core to
for_each_mem_region(), introduce the MEMBLOCK_NODUMP
flag, add a dumpable flag to struct reserved_mem, and
mark /reserved-memory and /memreserve/ entries with
MEMBLOCK_NODUMP flag.
Patch 10: Exclude MEMBLOCK_NODUMP regions from the vmcore ELF
header.
In v5, Sashiko found some pre-existing issues related to reserved-memory,
and has no dependency on this series, so these issues have been addressed
in a separate series [2].
v5 --> v6:
1. Serialize crash header preparation against memory hotplug to avoid
out-of-bounds or use-after-free issues.
2. MEMBLOCK_NODUMP marking is now done after memblock allows resizing,
avoiding a panic from too few regions before resize is permitted.
3. Reordered the patches, put pre-existing bugfixes earlier in the series.
v4 --> v5:
1. Rework the mechanism around a memblock-level MEMBLOCK_NODUMP flag
(suggested by Rob) instead of the v4 opt-in 'dumpable' flag on
DT-only struct reserved_mem.
2. Switch the riscv vmcore elf header preparation to use memblock
instead of the resource tree, aligning it with arm64 and loongarch,
so riscv also can exclude reserved memory from vmcore.
3. Deduplicate the vmcore elf header preparation: arm64, riscv and
loongarch open-coded the same logic, so fold it into shared
__weak defaults in crash_core.
4. Drop the v4 patch that saved /memreserve/ entries into the
reserved_mem array; /memreserve/ is now marked MEMBLOCK_NODUMP
directly.
v3 --> v4:
1. Rebase this series on v7.2-rc1.
2. Add two cleanup patches (patch 02/03).
3. Simplify patch 03 to avoid checking whether initial_boot_params is
NULL multiple times, suggested by Rob.
v2 --> v3:
1. Fix out-of-bounds issue if device tree lacks /reserved-memory node.
2. Fix UAF issue when alloc_reserved_mem_array() fails.
3. Add some prepare patches.
v1 --> v2:
1. v1 added an opt-out DT property ('linux,no-dump'). Per Rob's
feedback [3], v2 drop that property and exclude reserve memory
by default.
2. Split some prepared patches from the original patches.
3. Address coding-style comments on patch 5 from Rob.
[1] https://lore.kernel.org/lkml/20260723234126.GA3253409-robh@xxxxxxxxxx/
[2] https://lore.kernel.org/lkml/20260818092420.2859026-1-chenwandun1@xxxxxxxxx/
[3] https://lore.kernel.org/lkml/20260506144542.GA2072596-robh@xxxxxxxxxx/
Meijing Zhao (1):
mm: memblock: add missing HugeTLB flag name
Wandun Chen (9):
riscv: build crash_mem ranges from memblock instead of resource tree
crash_core: fold duplicated memblock arch hooks into the weak default
crash_core: serialize crash header preparation against hotplug
crash_core: replace for_each_mem_range() with for_each_mem_region()
memblock: introduce MEMBLOCK_NODUMP flag
of: reserved_mem: add dumpable flag to opt-in vmcore
of: reserved_mem: mark /reserved-memory entries with MEMBLOCK_NODUMP
of: reserved_mem: mark /memreserve/ entries as MEMBLOCK_NODUMP
crash_core: skip MEMBLOCK_NODUMP regions when building vmcore ELF
header
arch/arm64/kernel/machine_kexec_file.c | 29 ------------
arch/loongarch/kernel/machine_kexec_file.c | 27 ------------
arch/riscv/Kconfig | 2 +-
arch/riscv/kernel/machine_kexec_file.c | 33 --------------
arch/x86/kernel/crash.c | 2 +-
drivers/of/fdt.c | 2 +
drivers/of/of_private.h | 2 +
drivers/of/of_reserved_mem.c | 48 ++++++++++++++++++++
include/linux/crash_core.h | 2 +
include/linux/memblock.h | 9 ++++
include/linux/of_reserved_mem.h | 1 +
kernel/crash_core.c | 51 ++++++++++++++++++++--
kernel/dma/contiguous.c | 1 +
mm/memblock.c | 17 ++++++++
14 files changed, 131 insertions(+), 95 deletions(-)
--
2.43.0