ipmi: various fixes for panic notifier robustness

This patche set fixes multiple bugs in IPMI driver especially for the panic notifier callback panic_event() and related functions. This callback is called before kdump if crash_kexec_post_notifiers boot option is specified, so its robustness is very important.

panic_event() first tries to flush queued messages before panic, and this can cause multiple trouble such as hang-up. PATCH 3 to 6 addresses this kind of problems.

PATCH 7 addresses a problem caused by malfunctioning BMC.

NOTE: I'm planning to post another patch set which includes a patch to drop all queued messages at the beginning of panic_event() if crash_kexec_post_notifiers boot option is specified. This will improve the robustness of the panic notifier callback.

---

Hidehiro Kawai (7):
ipmi: Remove unneeded set_run_to_completion call
ipmi: Factor out message flushing procedure
ipmi: Don't flush messages in sneder() in run-to-completion mode
ipmi: Avoid touching possible corrupted lists in the panic context
ipmi: Don't call receive handler in the panic context
ipmi: Handle queued messages more certainly on panic ipmi/kcs: Don't run the KCS state machine when it is KCS_IDLE

 drivers/char/ipmi/ipmi_kcs_sm.c     |    4 ++
 drivers/char/ipmi/ipmi_msghandler.c |   66 +++++++++++++++++++++++++++++++++--
 drivers/char/ipmi/ipmi_si_intf.c    |   45 ++++++++++++------------
 include/linux/ipmi_smi.h            |   10 +++++
 4 files changed, 100 insertions(+), 25 deletions(-)

Hidehiro Kawai
Hitachi, Ltd. Research & Development Group