| Commit message (Collapse) | Author | Age |
|
|
|
| |
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@cascardo.eti.br>
|
|
|
|
| |
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@cascardo.eti.br>
|
|
|
|
| |
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@cascardo.eti.br>
|
|
|
|
| |
[git-debrebase make-patches: export and commit patches]
|
| |
|
|
|
|
|
|
|
|
|
| |
When installing kdump-tools on a chroot, the kernel might not be
available until boot. initrd generation is already taken care of when
kdump-tools is first loaded, so doing it during postinst is no longer
necessary.
Closes: #856594
|
|\
| |
| |
| | |
[git-debrebase pseudomerge: quick]
|
| |
| |
| |
| | |
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@debian.org>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
There is no need for root to build the package, and all files belong to
root anyway. dh_builddeb will end up calling dpkg-deb with
--root-owner-group, which will do the right thing.
A test build resulted in the same package contents, the binary packages
were reproducible bit-by-bit by changing only that.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@debian.org>
|
| |
| |
| |
| |
| |
| |
| | |
We should not reload kdump unconditionally after a hotplug event, but only
reload it when it was loaded already, which is what try-reload does.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@debian.org>
|
| |
| |
| |
| |
| |
| |
| | |
Use a lock file to allow try-reload to be run concurrently, and verify that
kdump is loaded before trying to unload and load again.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@debian.org>
|
| | |
|
| |
| |
| |
| | |
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@debian.org>
|
| |
| |
| |
| |
| |
| |
| |
| | |
reset_devices will be used by some drivers to do a special reset during
kdump. This will allow some systems with some devices that use such
drivers to kdump instead of fail to probe them.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@debian.org>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The solution used to allow multiple reloads during a hotplug event ended
up not working because udev won't execute two commands under a shell.
So, using a single command that reloads and do not interact with systemd
should work here.
As an extra, this adds support for other init systems.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@debian.org>
|
| |
| |
| |
| |
| |
| |
| | |
kdump-config reload will unload the current kdump kernel, and load a new
one.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@debian.org>
|
| |
| |
| |
| | |
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@debian.org>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
When a CPU is hotplugged, multiple events will be issued for each CPU
thread getting online. On a POWER system, that usually means 8 threads.
Those 8 events will cause systemd to consider the multiple restarts as
failed.
One alternative fix would be setting StartLimitIntervalSec to 0, but
that would apply to all cases where those failures might happen, not
only on the hotplug case.
Instead, we use reset-failed before try-restart, which will allow those
multiple restarts to happen on a short interval.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@debian.org>
|
| |
| |
| |
| |
| |
| |
| | |
This is needed on ppc64el, as CPUs are not added or removed, but simply
brought online.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| | |
On a kdump kernel, nr_cpus is broken, and it will take some time to be
properly fixed. In the meantime, we can just use maxcpus. In the worst
case, we will get an OOM and reboot instead of panicing too early during
boot.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
During the dump itself, spurious CPU or memory hotplug events will cause
the dump to fail, because the service would be restarted, and an old
incomplete dump would prevent the dump from being collected.
This work also allows us to stop requiring network during kdump loading,
and sets the way to stop requiring network when the dump is not over the
network.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@debian.org>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Kdump currently try mounting NFS (or doing the SSH dump) only once, and
if it fails, it just gives-up. Since kdump may be essential to debug hard
to reproduce bugs, we should improve the resilience and retry a bit,
delaying at each attempt.
This patch introduces a retry/delay mechanism for both NFS and SSH dumps;
the delay time is the same but number of retries is different (since NFS
mounts takes a long time between failures and is inherently more resilient),
both being configurable parameters from /etc.
The original trigger of this issue is a long-term (bad) behavior of some
NICs, which present a "Link Up" status _before_ being ready to transmit
packets; hence network kdump will try and fail without this patch.
Signed-off-by: Guilherme G. Piccoli <gpiccoli@canonical.com>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@debian.org>
|
| |\
| | |
| | |
| | |
| | | |
makedumpfile release 1:1.6.5-1 for unstable (sid) [dgit]
[dgit distro=debian]
|
| | |\
| | | |
| | | |
| | | | |
[dgit --overwrite]
|
| | | |
| | | |
| | | |
| | | |
| | | | |
===================================================================
Gbp-Pq: Name 0002-adapt-makefile-to-debian.patch
|
| | | |
| | | |
| | | |
| | | | |
Description: Remove libebl from link options
Gbp-Pq: Name 0001-Remove-libebl-linkage.patch
|
| | | |\
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
* Reload kdump after memory/CPU hotplug. (LP: #1655280)
* Fix adding crashkernel to zipl.conf when no quotation mark is used.
(LP: #1790788)
[dgit import unpatched makedumpfile 1:1.6.4-3]
|
| | | |
| | | |
| | | |
| | | | |
[dgit import tarball makedumpfile 1:1.6.4-3 makedumpfile_1.6.4-3.debian.tar.xz]
|
| | |
| | |
| | |
| | | |
[dgit import orig makedumpfile_1.6.4.orig.tar.gz]
|
| | |
| | |
| | |
| | | |
[git-debrebase make-patches: export and commit patches]
|
| | |
| | |
| | |
| | | |
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@debian.org>
|
| | |
| | |
| | |
| | |
| | | |
===================================================================
Gbp-Pq: Name 0002-adapt-makefile-to-debian.patch
|
| | |
| | |
| | |
| | | |
Description: Remove libebl from link options
Gbp-Pq: Name 0001-Remove-libebl-linkage.patch
|
| | |
| | |
| | |
| | | |
[git-debrebase make-patches: export and commit patches]
|
| |\ \
| | | |
| | | |
| | | | |
[git-debrebase pseudomerge: quick]
|
| | | |
| | | |
| | | |
| | | |
| | | | |
===================================================================
Gbp-Pq: Name 0002-adapt-makefile-to-debian.patch
|
| | | |
| | | |
| | | |
| | | | |
Description: Remove libebl from link options
Gbp-Pq: Name 0001-Remove-libebl-linkage.patch
|
| | | |
| | | |
| | | |
| | | |
| | | | |
===================================================================
Gbp-Pq: Name 0002-adapt-makefile-to-debian.patch
|
| | | |
| | | |
| | | |
| | | | |
Description: Remove libebl from link options
Gbp-Pq: Name 0001-Remove-libebl-linkage.patch
|
| | | |
| | | |
| | | |
| | | |
| | | | |
===================================================================
Gbp-Pq: Name 0002-adapt-makefile-to-debian.patch
|
| | | |
| | | |
| | | |
| | | | |
[git-debrebase changelog: new upstream 1.6.7]
|
|\ \ \ \
| | | | |
| | | | |
| | | | | |
[git-debrebase anchor: new upstream 1.6.7, merge]
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
Update makedumpfile to version 1.6.7.
Signed-off-by: Kazuhito Hagio <k-hagio-ab@nec.com>
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
Support newer kernels as follows:
- 5.2, 5.3, 5.4 (x86_64 SPARSEMEM)
- 5.4 (x86 FLATMEM, SPARSEMEM)
Signed-off-by: Kazuhito Hagio <k-hagio-ab@nec.com>
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
Update README file as follows:
- Change the required version of elfutils to 0.144 because of
the PN_XNUM support recently added.
- Update the email address to which bug report should be sent.
Signed-off-by: Kazuhito Hagio <k-hagio-ab@nec.com>
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
Since kernel v5.2 KASLR is supported on s390. Use recently introduced
get_kaslr_offset_general() for s390x in order to derive kaslr offset
from vmcoreinfo when -x makedumpfile option specified.
Signed-off-by: Mikhail Zaslonko <zaslonko@linux.ibm.com>
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
Currently SYMBOL(_stext) is passed to get_kaslr_offset() in
find_kaslr_offsets(), but it is always zero, because it has not
been set yet at the time.
On the other hand, the vaddr argument of get_kaslr_offset() is
only used to decide whether to return a KASLR offset or not, but
the return value is not used in find_kaslr_offsets().
Therefore, passing SYMBOL(_stext) is meaningless and confusing,
so let's pass it 0 explicitly to avoid confusion.
Reported-by: Lianbo Jiang <lijiang@redhat.com>
Signed-off-by: Kazuhito Hagio <k-hagio-ab@nec.com>
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
Generalize the get_kaslr_offset_arm64() and rename it to
get_kaslr_offset_general() for use by other architectures
supporting KASLR.
Also, modify the confusing comment in the function and its
x86_64 version.
Signed-off-by: Kazuhito Hagio <k-hagio@ab.jp.nec.com>
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
Suppress the following warnings in get_elf64_phnum() on 32-bit system.
elf_info.c: In function 'get_elf64_phnum':
elf_info.c:1052: warning: format '%lx' expects type 'long unsigned int',
but argument 4 has type 'Elf64_Off'
elf_info.c:1057: warning: format '%lx' expects type 'long unsigned int',
but argument 4 has type 'Elf64_Off'
Signed-off-by: Kazuhito Hagio <k-hagio@ab.jp.nec.com>
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
In non-cyclic mode with the --split option, each subprocess inherits
bitmap1/2->fd from parent. Then they lseek()/read() on the same fd,
which means that they interfere with each other.
This breaks the purpose of SPLITTING_FD_BITMAP(i) for each subprocess.
Without this patch, makedumpfile can fail with error like the following
in refiltering, or can break the dumpfile silently by excluding pages
wrongly.
readpage_kdump_compressed: pfn(2fc1000) is excluded from vmcore.
readmem: type_addr: 1, addr:2fc1000000, size:4096
read_pfn: Can't get the page data.
Fix it by assigning a subprocess dedicated fd to bitmap1/2->fd.
Signed-off-by: Pingfan Liu <piliu@redhat.com>
Signed-off-by: Kazuhito Hagio <k-hagio@ab.jp.nec.com>
|