Product SiteDocumentation Site

1.27. crash

1.27.1. RHBA-2010:0230: bug fix update

Updated crash packages that fix various bugs and add enhancements are now available.
The crash package is a core analysis suite. It is a self-contained tool that can be used to investigate either live systems, kernel core dumps created from the netdump, diskdump, and kdump packages from Red Hat Linux, the mcore kernel patch offered by Mission Critical Linux, or the LKCD kernel patch.
* if a kdump NMI was issued and the task kernel stack was changed, the backtrace would in some cases fail and produce an error: "bt: cannot transition from exception stack to current process stack". The crash package was updated to report task inconsistencies and change the active task as appropriate. Additionally, a new set -a option was added to manually set tasks to be the active task on its CPU. (BZ#504952)
* if the kernel data structures in a non-matching vmlinux varied widely enough from the kernel that generated the vmcore, erroneous data could be read and consumed. Several new defensive mechanisms have been added and it now fails in a more reasonable manner. (BZ#508156)
* running the bt -a command against a Xen hypervisor resulted in a "cannot resolve stack trace" warning message if the CPU received its shutdown NMI while running in an interrupt handler. The bt command was changed and the error no longer occurs. (BZ#510505)
* added support for dumpfile format of virsh dump of KVM kernels. (BZ#510519)
* if a dump was collected when there were one or more cpus offline in the system, an initialization-time failure would occur and the crash would abort. A patch was backported from upstream and the failure no longer occurs. (BZ#520506)
* running the 64-bit bt command could potentially start the backtrace of an active non-crashing task on its per-cpu IRQ stack, cause a faulty transition back to the process stack, the dumping of a bogus exception frame and the message "bt: WARNING: possibly bogus exception frame". The bt command was changed and it now starts from the NMI exception stack, the error no longer occurs. (BZ#523512)
* when the cpu_possible_map contains more CPUs than the cpu_online_map, the set, bt, runq and ps commands would reflect the existing but unused swapper tasks on the non-existent CPUs. The 64-bit PowerPC CPU count determination was fixed and the commands now run as expected. (BZ#550419)
* when INIT-generated pseudo-tasks were running in user-space and the kernel was unable to modify the kernel stack, the backtrace would not identify the interrupted task and would display a "bt: unwind: failed to locate return link" error message. The Itanium backtraces were fixed, and the backtrace now offers information regarding the task that was interrupted. The error message is also suppressed. (BZ #553353)
* using dump to analyze very large xendump core files with ELF sections located beyond a file offset of 4GB resulted in errors. Changes were made to the xc_core_verify() initialization code and dump now works as expected. (BZ #561767)
* The crash utility was rebased. See the changelog linked to in the references section below for full details. (BZ#528184)
All users of crash are advised to upgrade to these updated packages, which resolve these issues.

Note: This documentation is provided {and copyrighted} by Red Hat®, Inc. and is released via the Open Publication License. The copyright holder has added the further requirement that Distribution of substantively modified versions of this document is prohibited without the explicit permission of the copyright holder. The CentOS project redistributes these original works (in their unmodified form) as a reference for CentOS-5 because CentOS-5 is built from publicly available, open source SRPMS. The documentation is unmodified to be compliant with upstream distribution policy. Neither CentOS-5 nor the CentOS Project are in any way affiliated with or sponsored by Red Hat®, Inc.