[Skiboot] [PATCH v10 08/17] core/pci: Option for serialized jobs

Stewart Smith stewart at linux.vnet.ibm.com
Fri May 13 14:31:21 AEST 2016


Gavin Shan <gwshan at linux.vnet.ibm.com> writes:
> On Fri, May 13, 2016 at 10:33:27AM +1000, Stewart Smith wrote:
>>Gavin Shan <gwshan at linux.vnet.ibm.com> writes:
>>> Currently, the PHB reset and PCI enumeration are done concurrently
>>> on multiple CPU cores. The output messages are interleaved and not
>>> readable enough. This adds a option to do the jobs in serialized
>>> fashion for debugging purpose only. The serialized mode should be
>>> always disabled in field.
>>>
>>> Signed-off-by: Gavin Shan <gwshan at linux.vnet.ibm.com>
>>
>>I think we could accomplish this in the generic code a lot neater, in
>>core/cpu.c __cpu_queue_job() we could just add:
>>
>>#ifdef DEBUG_SERIALIZE_CPU_JOBS
>>if (cpu == NULL)
>>   cpu = this_cpu();
>>#endif
>>
>>and then we'd be automatically able to pretty easily enable serializing
>>of things that would otherwise happen in parallel.
>>
>>(better than ifdef may be something in debug descriptor... although
>>setting that on boot before things start executing may be casually fun)
>>
>
> Good idea. Thanks, Stewart. It will be included into next revision.
> The side-effect is to affect all paths including PCI enumeration. As
> it's for debugging, I guess it's not a problem.

There's likely other uses for it too, so for debug it's fine. Will be
just like booting with GA1 skiboot :)

> Yeah, debug descriptor should be better. I'm not sure if it has
> been supported in skiboot and how it can be enabled?

Likely need a define to make it the default rather than something
selectable at runtime - otherwise you have to be able to change the
contents of host memory in between skiboot load and starting to execute
it, which isn't always possible.

-- 
Stewart Smith
OPAL Architect, IBM.



More information about the Skiboot mailing list