Fixing a regression in e_to_tagname() for anonymous structs that are not
Add explicitly -std=c++11 flag to CXX as it is needed with older
Increasing PyKdump Python-API version before merging with master
Adding a warning about just a partially implemented support for
Remove sdef2ArtSU as we don't use it anymore
Fix some inconsitencies in purging caches on DLKM debuginfo reload
Reorganize code for TypeInfo and SUInfo to make it more readable.
Remove sdef2ArtSU definition as we don't use it anymore
Put exec_crash_command("struct nfs_server") in try/except block.
Made 'set_error' a weak symbol so that mpykdump.so can be loaded
Add version display for 'nvme' program
Reorganize code related to task state processing and
Fix routing table display for recent kernels (tested on 5.4)
Comment out a debugging print (left there by mistake)
Increase maximum number of tasks/threads to process to 1000000.
Improve SCSI symbolic info detection - some SLES kernels have
Adding the new 'mdadm' command to be included in mpykdump.so and
As 'dev' branch has been tested thoroughly, I merged its changes to 'master'
This is fixed now in 'dev' branch. After Redhat confirms that PyKdump RPMs works fine, the changes will be merged to 'master' branch too,
Hello Oscar, thank you for bugreport. The issue is due to the fact that crash-7.2.8 sources as retrieved from the new project WEB-site (github( are somewhat different from crash-7.2.8 as it was from old Dave Anderson's website. There is a new build process (Makefile etc.) for this 'new' version and as a result there are some problems. We plan to fix this shortly. Regards, Alex
Merge branch 'dev' into unstable
Merge branch 'unstable' of ssh://git.code.sf.net/p/pykdump/code into unstable
Fixing regression - by mistake, some needed imports were removed
Adding 'pointersize' computation to lowelevel module
Fixing regressions introduced by code refactoring
Printing CFS for some vmcores fails, possibly our code is incorrect
Add --nfsclient addr option to print info about nfs_client structures
Memoize conditionally 'bt -f' result for all threads
Add to crashinfo/hanginfo ad hoc tests - if there is LinuxDump/adHoc.py
Fixing a regression introduced while refactoring code: forgot to add
Avoid premature exit from print_xprt() when there is no ipsock yet.
A fix specific for Python-3.8.0. All regression tests passed, so
Redesign of epython.c to make it work with upcoming Python-3.8
A quick fix for idr_for_each for those kernels that use trees
Hi Rongwei, your solution is not very good as you will need to do this modification every time when you update PyKdump sources. You can achieve the same result like by using PYKDUMPPATH environment variable - it specifies all additional paths where to search for modules/programs. From epython.c sources: // We can specify extra path to search for modules and progs // by setting PYKDUMPPATH environment variable. // The order of search is like that: // 1. Current directory // 2. PYKDUMPPATH (if set,...
Hi Rongwei, by default, Python builds many modules as .so to be dynamically loaded. But to make PyKdump module as portable as possible, I link it against a specially built version of Python, with all needed modules built as 'static'. Look at https://sourceforge.net/p/pykdump/code/ci/master/tree/Extension/Setup.local-3.7 this is what I use for building Python. In particular, this compiles select module as 'static'. Regards, Alex
Check whether "struct fc_port" definition is available in vmcore before trying
Continue task-processing even when thread_info is corrupted,
Hello, I work for HPE and some time ago I hoped that we'll get access to ARM64 but so far we have none. If you want to add support for it, it would be great. PyKdump used to work with IA-64 many years ago, but then Redhat stopped supporting it and I have removed it from PyKdump. There are three groups of changes to support new architectures: 1. Possible changes in C-module (related to integer sizes and endianness) 2. Python parts of framework, once again possibly related to integer sizes 3. There...
In some cases, exec_crash_command raises crash.error exception
Copying a file logged erreneously to 'master'
Avoid traceback when some members in sunrpc_net are missing
Fixing a typo
Load modules even when there is no 'nfs' DLKM loaded, just 'nfsd'
With '-v', print sunrpc_net
Adding from Python Standard Library 'inspect' module and its
Merge branch 'master' into dev
Merge branch 'dev' of ssh://git.code.sf.net/p/pykdump/code into dev
Adding try/except in several places to make all command work at
Making cProfile module working again - this needs extra subroutines
Report coruption of runq CFQ when radix tree cannot be traversed
Addded '--cmd' option to run commands for each task
Merge branch 'dev' of ssh://git.code.sf.net/p/pykdump/code into dev
Fixing a trivial typo in the displayed message for SG panic
Merge branch 'dev' of ssh://git.code.sf.net/p/pykdump/code into dev
Adding DCache object that can be used for intermodule communication
Print RTO value when printing TCP sockets. Sort TCP connections
Print additional information about TCP retransmissions
Improve formatting of TCP summary
If vmcore was triggered by ServiceGuard deadman timer and there were
Add a new '--retrans' option to xportshow to print only connections
Merge branch 'dev' of ssh://git.code.sf.net/p/pykdump/code into dev
Add another test to decide whether 'struct request' is bad,
Adde a global option '--maxel=nnnn' to increase the maximum number of list
Merge branch 'dev' of ssh://git.code.sf.net/p/pykdump/code into dev
Print 'fib_nh' when using 'xportshow -r -vv'
Fixed a test for NFS-cache not being supported for specific kernels
Force UTF8 for stdout by doing sys.stdout.reconfigure(encoding='utf-8')
Sometimes we need to exit from PyKdump program without reaching its end.
Change how we display processes sorted by RSS only
Add a test for panics triggered by Oracle RAC agents
Add a test for pcc_cpufreq
Add a new LinuxDump.dlkm module for generic 'mod' related subroutines
Fix a bug in TCP listen option emulation for new kernels.
Check 'kmem -i' output for excessive overcommit, swapping and hugepages
Adding additional Python Standard Library modules by Redhat request
Fixed cache_detail test for unsupported kernels, add more output
Fixed algorithm for longChainOfPids()
Due to changes in 'kmem -s/S' output in the latest crash, parse
Merge branch 'dev' of ssh://git.code.sf.net/p/pykdump/code into dev
Continued processing of AF_UNIX sockets even when some pages in vmcore are
Hi Daniel, I was on vacation and just came back, sorry for delay with response. While we can discuss adding some specific modules, I think that adding all Python modules is not feasible, here is why. Python Standard Library includes both pure Python code and shared libraries that in turn might depend on other shared libraries. E.g. on my laptop: /usr/lib/python3.5/lib-dynload/ ls audioop.cpython-35m-x86_64-linux-gnu.so _bz2.cpython-35m-x86_64-linux-gnu.so cmath.cpython-35m-x86_64-linux-gnu.so _codecs_cn.cpython-35m-x86_64-linux-gnu.so...
A quick fix for getCache() not working:
Try loading two extra debuginfos - for 'nfsv3' and 'nfsv4'
If there is no 'idmap_wq' field in 'struct idmap', skip the code
Be more resilient while processing corrupted block devices and slabs
Fixing decode_waitq() for newer kernels. Still needs more testing
Do not add empty boxes when drawing a tree
Differeniate between cases when structure contains 'list head' vs
Add detection of threads waiting for NFSv4 ID-mapper
In PPC, enum_sdev_state.getnam() returning int
Adding files needed for build against Python-3.7
Bumping release version to 3.0.4
Increase maxel to 20000 in klistAll() as modern hosts can have
Fixed a regression in check_other_mutexes
Added MsgExtra class to be used for better diagnostic messages
Add detection of circular dependency in lists
Several interrelated changes to make tools work with the latest RHEL7.5
Improvements for decoding 'struct request' and redesigning code.
A partial fix for newer kernels (e.g. no enum rq_flag_bits anymore).