[Crash-utility] [ANNOUNCE] crash-7.3.0 is available
HAGIO KAZUHITO(萩尾 一仁)
k-hagio-ab at nec.com
Tue Apr 27 08:19:28 UTC 2021
Download from: https://crash-utility.github.io/
or
https://github.com/crash-utility/crash/releases
The github master branch serves as a development branch that will contain
all patches that are queued for the next release:
$ git clone git://github.com/crash-utility/crash.git
Changelog:
- Add support for new lockless righbuffer that Linux 5.10 introduced.
Without the two patches, crash fails during session initialization
or "log" command fails with the error message:
crash: cannot determine length of symbol: log_end
(john.ogness at linutronix.de, nborisov at suse.com, k-hagio-ab at nec.com)
- Add support for VC exception stack on x86_64 Linux 5.10 and later
kernels that contain commit 02772fb9b68e ("x86/sev-es: Allocate and
map an IST stack for #VC handler").
(amakhalov at vmware.com)
- Fix regression for raw RAM dumpfiles. Commit f42db6a33f0e ("Support
core files with "unusual" layout") increased the minimal file size
from MIN_NETDUMP_ELF_HEADER_SIZE to SAFE_NETDUMP_ELF_HEADER_SIZE
which can lead to crash rejecting raw RAM dumpfiles. Without the
patch, the crash fails to start a session with the error message:
/var/tmp/ramdump_elf_XXXXXX: ELF header read: No such file or directory
crash: malformed ELF file: /var/tmp/ramdump_elf_XXXXXX
(zhaoqianli at xiaomi.com)
- Update mapping symbol filter in arm64_verify_symbol() to support the
long form of mapping symbols, e.g. "$x.<any...>". Without the
patch, the "dis" command cannot completely parse out the disassembly
of a function that has mapping symbols in the long form and misses
the tail part of the function.
(zhaoqianli at xiaomi.com)
- Move extensins/Makefile's ping check to recipe script. Without this
patch, in an environment where ping to github.com does not work,
"make clean" at the top-level crash directory always takes about 10
seconds unnecessarily.
(k-hagio-ab at nec.com)
- Fix for a segmentation fault when analyzing arm64 kernels that are
configured with CONFIG_IKCONFIG and have a strange entry that does
not contain the delimiter "=", such as "CONFIG_SECU+[some hex data]".
Without the patch, in the add_ikconfig_entry() function, strtok_r()
interprets it as consisting of a single token and the val variable
is set to NULL, and then strdup() crashes.
(liuyun01 at kylinos.cn)
- Fix a couple of issues that were detected by valgrind.
(d.hatayama at fujitsu.com)
- Add ability to un-set scope. The ability can come in very useful
when running automated pykdump scripts and needing scope to be
cleared between script runs.
(jpittman at redhat.com)
- Fix "sys [-t]|mod -S" after "mod -t" when crash runs with -s option.
Without the patch, the "sys [-t]" and "mod -S" options after "mod -t"
option fail with the error message:
sys: invalid structure member offset: tnt_false
FILE: kernel.c LINE: 11203 FUNCTION: show_kernel_taints_v4_10()
(k-hagio-ab at nec.com)
- Fix for "dev -d" option on Linux 5.11-rc1 and later kernels that
contains commit 0d02129e76edf91cf04fabf1efbc3a9a1f1d729a ("block:
merge struct block_device and struct hd_struct"). Without the patch,
the option fails with the error message:
dev: invalid structure member offset: hd_struct_dev
(k-hagio-ab at nec.com)
- Fix for "kmem -v" option on Linux 5.11-rc1 and later kernels that
contain commit 96e2db456135db0cf2476b6890f1e8b2fdcf21eb ("mm/vmalloc:
rework the drain logic"). Without the patch, the option will display
nothing or fail with the error message:
kmem: invalid kernel virtual address: <address> type: "vmlist addr"
(k-hagio-ab at nec.com)
- Add the base address of module to "mod" command output. Currently
the command shows the address of the module struct, but it is
inconvenient to know the address range of the module, so extend to
show the base adddress.
(yeyunfeng at huawei.com, k-hagio-ab at nec.com)
- Increase the value of __PHYSICAL_MASK_SHIFT_XEN to 52. The former
value of __PHYSICAL_MASK_SHIFT_XEN in crash (40) is smaller than the
kernel (52) since kernel commit 6f0e8bf167 (xen: support 52 bit
physical addresses in pv guests). This can cause x86_64_pud_offset()
to lose the most significant bits of pgd_pte, leading to a failed
xen_m2p() translation, resulting in crash failing with an error
message like this:
crash: read error: physical address: ffffffffffffffff type: "pud page"
(jbohac at suse.cz)
- Change log level print in older kernels. In older kernels that have
the variable-length-record log_buf, the log level and the log
flags/facility are not separated. Since the log level is only the
last three bits, and the flags/facility and level are separated in
5.10 and later kernels, only print those last three bits when using
'log -m'.
(jpittman at redhat.com)
- Reduce crash build log. The verbose output of tar command when
extracting the GDB source files occupies more than the half of crash
build log. It is not so helpful and makes the build log longer
needlessly especially on CI build test without the patch.
(k-hagio-ab at nec.com)
- Fix for "bt" command on Linux 5.12-rc1 and later x86_64 kernels that
contain commit 951c2a51ae75 ("x86/irq/64: Adjust the per CPU irq
stack pointer by 8"). Without the patch, the "bt" command and some
of its options that read irq stack fail with the error message:
bt: read of stack at <address> failed".
(k-hagio-ab at nec.com)
- Add valgrind support for the crash's custom memory allocator. This
helps detecting various memory errors on the crash's custom memory
allocator.
(d.hatayama at fujitsu.com)
- Fix for a couple of invalid read/write issues detected by valgrind.
(d.hatayama at fujitsu.com)
- Fix "struct" command to print member array of list_heads correctly.
Without the patch, due to the way that an array of list_head entries
are printed, parsing of them fails and the command does not print
anything:
crash> struct blk_mq_ctx.rq_completed ffffc447ffc0f740
crash>
(jpittman at redhat.com)
- Do not pass through 'sy' command to GDB. The GDB 'symbol-file'
command is prohibited in the crash utility, but an abbreviation of
it, the 'sy' is not prohibited. This can discard symbol table from
the current symbol file, and eventually caused the failure of crash
utility after executing the 'sys' command as below:
crash> sy
Discard symbol table from `/path/to/vmlinux'? (y or n) Please answer y or n.
Discard symbol table from `/path/to/vmlinux'? (y or n) No symbol file now.
crash> sys
double free or corruption (!prev)
Aborted (core dumped)
(lijiang at redhat.com)
- Refine zram related code for crash gcore command to support it.
(d.hatayama at fujitsu.com)
- Fix for the failure of 'set scope' command. Without the patch,
some commands such as 'sys' may cause subsequent 'set scope'
commands to fail.
(lijiang at redhat.com)
- Fix for offset print for function pointers that return pointers.
In the show_member_offset() function, when trying to handle function
pointers, the case for "(*" is handled. However, if the function
pointer returns a pointer or a pointer to a pointer, then the
condition is unhandled. This results in the offset not being
printed without the patch, for example:
crash> struct -o offload_callbacks
struct offload_callbacks {
struct sk_buff *(*gso_segment)(struct sk_buff *, netdev_features_t);
struct sk_buff **(*gro_receive)(struct sk_buff **, struct sk_buff *);
[16] int (*gro_complete)(struct sk_buff *, int);
}
(jpittman at redhat.com)
- Change functions within extensions/echo.c to be static and document
the issue in code comments, for extensions developers who takes
echo.c as reference, to avoid the issue that symbols in extension
modules are overwritten by former loaded one if it's the same name.
(ltao at redhat.com)
- Fix for 'bt' command and options on Linux 5.8-rc1 and later x86_64
kernels that contain merge commit 076f14be7fc9. The merged patches
changed the name of exception functions that have been used by the
crash utility to check the exception frame. Without the patch, the
command and options cannot display it.
(k-hagio-ab at nec.com)
- Fix for xen kernels that contain commit edcb5cf84f05
("x86/paravirt/xen: Remove xen_patch()"). Withouth the patch,
crash fails with an error message like this:
crash: seek error: physical address: 83640e000 type: "pud page"
(john.p.donnelly at oracle.com, k-hagio-ab at nec.com)
- Remove extensions/trace.c file, as the extension module moved to
the separate repository from the crash repository.
(k-hagio-ab at nec.com)
- Fix for uvtop conversion on ARM with LPAE. Without the patch,
arm_uvtop() calls arm_lpae_vtop() with the LPAE and it can use
LPAE_VTOP() also for a user virtual address. As a result, commands
that use uvtop conversion such as "ps -a", "gcore" fail as readmem()
for a uvaddr returns a seek error:
ps: cannot access user stack address: <address>
(k-hagio-ab at nec.com)
- Handle 1GB block for VM_L3_4K on arm64 architecture. Without the
patch, "vtop" command cannot display the block as a 1GB hugepage.
(johan.erlandsson at sony.com)
- Implement initial support for the MIP64 architecture.
(tangyouling at loongson.cn, chenhuacai at loongson.cn)
- Fix for HZ calculation using cfq_slice_async on Linux 4.8 and later
kernels that contain commit 9a7f38c42c2b ("cfq-iosched: Convert from
jiffies to nanoseconds"). Without the patch, the HZ calculation
results in a wrong and big value for machde->hz and crash can shows
a wrong uptime and timestamps in "log -T".
(martin.moore at hpe.com)
- Fix for HZ calculation on Linux 4.8 and later kernels that contain
commit 9a7f38c42c2b ("cfq-iosched: Convert from jiffies to
nanoseconds"). Without the patch, the HZ value can be set to a
hardcorded wrong value.
(k-hagio-ab at nec.com)
More information about the Crash-utility
mailing list