DescriptionYuuichi Nagahama
2005-04-05 19:47:39 UTC
Description of problem:
When crashdump is executed via OS_INIT on IPF machine, backtrace command,
which is a subcommand of crash, does not work correctly.
Two problems were found in the OS_INIT code.
(1) OS_INIT handler has two stages.
stage1: handler written by assembler
stage2: handler written by C
The former is ia64_monarch_init_handler and ia64_slave_init_handler.
The latter is ia64_init_handler. ia64_init_handler is called only by
ia64_monarch_init_handler.
When INIT interrupt is asserted, one cpu calls
ia64_monarch_init_handler and the others call
ia64_slave_init_handler. It means that ia64_init_handler is called
by only one cpu. In that case, backtrace command fails. To make
backtrace succeed, all cpus need to call ia64_monarch_init_handler.
(2) The second problem occurs by correcting the first problem.
When OS_INIT handler is called, SAL hands handler some information
through register. The handler preserves this information in
ia64_sal_to_os_handoff_state. (Please see
SAL_TO_OS_MCA_HANDOFF_STATE_SAVE macro in the arch/ia64/kernel/
mca_asm.S.) If all cpus call ia64_monarch_init_handler at the same
time, they writes their own information to the
ia64_sal_to_os_handoff_state simultaneously and break its contents.
Version-Release number of selected component (if applicable):
kernel-2.4.21-31.EL
How reproducible:
always
Steps to Reproduce:
1. Enable Diskdump
2. Push OS_INIT switch
3. bt with crash command
Actual results:
Backtrace command does not work
Expected results:
Backtrace command works correctly
Additional info:
none
How to correct these problems:
(1) When OS_INIT handler is registered with SAL,
register ia64_monarch_init_handler as both monarch handler and slave
handler.
(2) Prepare ia64_sal_to_os_handoff_state of each cpu beforehand.
To verify the fix, we've done testing with kernel-2.4.21-32.9.EL. It works
correctly.
Regards,
Akira
Comment 15Red Hat Bugzilla
2005-09-28 14:53:08 UTC
An advisory has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on the solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.
http://rhn.redhat.com/errata/RHSA-2005-663.html
Description of problem: When crashdump is executed via OS_INIT on IPF machine, backtrace command, which is a subcommand of crash, does not work correctly. Two problems were found in the OS_INIT code. (1) OS_INIT handler has two stages. stage1: handler written by assembler stage2: handler written by C The former is ia64_monarch_init_handler and ia64_slave_init_handler. The latter is ia64_init_handler. ia64_init_handler is called only by ia64_monarch_init_handler. When INIT interrupt is asserted, one cpu calls ia64_monarch_init_handler and the others call ia64_slave_init_handler. It means that ia64_init_handler is called by only one cpu. In that case, backtrace command fails. To make backtrace succeed, all cpus need to call ia64_monarch_init_handler. (2) The second problem occurs by correcting the first problem. When OS_INIT handler is called, SAL hands handler some information through register. The handler preserves this information in ia64_sal_to_os_handoff_state. (Please see SAL_TO_OS_MCA_HANDOFF_STATE_SAVE macro in the arch/ia64/kernel/ mca_asm.S.) If all cpus call ia64_monarch_init_handler at the same time, they writes their own information to the ia64_sal_to_os_handoff_state simultaneously and break its contents. Version-Release number of selected component (if applicable): kernel-2.4.21-31.EL How reproducible: always Steps to Reproduce: 1. Enable Diskdump 2. Push OS_INIT switch 3. bt with crash command Actual results: Backtrace command does not work Expected results: Backtrace command works correctly Additional info: none