[Skiboot] [PATCH 0/2] direct-controls: p9 stop/cont fixes

Nicholas Piggin npiggin at gmail.com
Thu May 3 18:38:07 AEST 2018


This is unchanged since the RFC, passed some testing: `pdbg stop`
a thread, then enter xmon in the host which cases it to NMI IPI
all other CPUs. Result now is that it skips the stopped thread and
it simply doesn't make it into xmon. pdbg cont resumes as normal,
(but it looks like the thread doesn't receive its doorbell to get
into xmon). Sending it a sreset with pdbg kicks it into xmon, and
then resuming from xmon seems to result in the system recovering
fine.

Without this patch, the thread will be given the sreset, which
can have undefined consequences if we're currently ramming or
in some other stat. At best it will kick the thread out of quiesce
and we lose our pdbg debugging session, at worst probably checkstop
the box.

These fixes should not affect non-error paths, or fix the timeout
setting speial wakeup, or other errors we've often seen.

Thanks,
Nick

Nicholas Piggin (2):
  core/direct-controls: fix p9_cont_thread for stopped/inactive threads
  core/direct-controls: improve p9_stop_thread error handling

 core/direct-controls.c | 101 ++++++++++++++++++++++++++++++-----------
 1 file changed, 75 insertions(+), 26 deletions(-)

-- 
2.17.0



More information about the Skiboot mailing list