[Skiboot] [PATCH 1/2] eSEL: Make sure PANIC logs are sent to BMC before calling assert

[Klaus Heinrich Kiwi]

On 3/9/2020 12:39 PM, Vasant Hegde wrote:
>> I think this works for msg-> because PANIC msg are never de-allocated, 
>> even if ipmi_sel_free_msg(msg) may have raced with the conditionals 
>> above it.
> 
> Yes. That's intentional. We want to avoid allocation in panic path. 
> Hence we pre-allocate message and use that in PANIC path.
> 
>>
>>
>> But is there a chance there could be a race between 
>> ipmi_sel_free_msg() setting ->busy to false and another caller/thread 
>> immediately setting it back to true on a new 
>> elog_commit()->ipmi_sel_alloc_msg()? That could also mean that this 
>> new PANIC is still not expired on the ipmi/bt stack (i.e., in theory 
>> could cause this loop never to finish if there's always a new caller 
>> with PANIC waiting to be scheduled, so classify that under "nitpicking")
> 
> Generally one one thread makes reboot2 call. So we are good.

Thanks for the clarification.

Reviewed-by: Klaus Heinrich Kiwi <klaus at linux.vnet.ibm.com>

-- 
Klaus Heinrich Kiwi <klaus at linux.vnet.ibm.com>