[Skiboot] The state of "Running pollers with lock held"

[Michael Neuling]

On Wed, 2015-03-25 at 17:00 +1100, Benjamin Herrenschmidt wrote:
> On Wed, 2015-03-25 at 16:23 +1100, Stewart Smith wrote:
> > Vasant Hegde <hegdevasant at linux.vnet.ibm.com> writes:
> > 
> > > On 03/25/2015 08:26 AM, Stewart Smith wrote:
> > >> So,
> > >> 
> > >> I merged my patch to rework fsp_fetch_data to support proper
> > >> asynchronous fetching. This gets rid of the major source of these
> > >> warnings.
> > >> 
> > >
> > > I have not gone through your patches yet.. Will take a look tomorrow.
> > >
> > >> We may still have something in an op_display() code path (I should
> > >> really save the backtrace next time I see it).
> > >> 
> > >> We also have:
> > >> [3742482870,3] Running pollers with lock held !
> > >> CPU 00a0 Backtrace:
> > >>  S: 0000000031c839e0 R: 0000000030013a98   .backtrace+0x40
> > >>  S: 0000000031c83a60 R: 0000000030019ec8   .opal_run_pollers+0x9c
> > >>  S: 0000000031c83af0 R: 0000000030027074   .wait_for_resource_loaded+0x48
> > >>  S: 0000000031c83b90 R: 0000000030047104   .capp_load_ucode+0x1e0
> > >>  S: 0000000031c83c80 R: 000000003004d9bc   .probe_phb3+0x1240
> > >>  S: 0000000031c83e30 R: 0000000030014be0   .main_cpu_entry+0x468
> > >>  S: 0000000031c83f00 R: 000000003000253c   boot_entry+0x194
> > >
> > > I had fixed this one along with poller patch.. Looks like you have not applied
> > > that..
> > >
> > > https://lists.ozlabs.org/pipermail/skiboot/2015-March/000692.html

Vasant: That has "cc: Michael Neuling <mikey at neuling.org>" but I wasn't
not actually CCed on the email.  Please don't do that.  

> > 
> > Instead of that patch, could we move the starting of loading the LID to
> > earlier in PHB setup, only taking capi_lock right at the very end to
> > dump things into it.
> > 
> > Mikey? Your thoughts?
> 
> Yes :)

Yep.  Something like that.

We could just start the lid load early in boot in anticipation of us
needing it later.  When it's does download it just sets
capp_ucode_info.lid.  The phb code could then just spin waiting for that
to be set by another thread.

That being said, I'm surprised it takes very long.  It's only a ~32KB
file.  I guess there is a lot of FSP overhead?

Mikey

> Cheers,
> Ben.
> 
> > 
> > >> as well as (when doing code update):
> > >> 
> > >> [32872663019955,3] Running pollers with lock held !
> > >> CPU 0020 Backtrace:
> > >>  S: 0000000031a83960 R: 0000000030013a98   .backtrace+0x40
> > >>  S: 0000000031a839e0 R: 0000000030019ec8   .opal_run_pollers+0x9c
> > >>  S: 0000000031a83a70 R: 00000000300548e8   .fsp_sync_msg+0xb8
> > >>  S: 0000000031a83b20 R: 000000003005d2d8   .fsp_flash_firmware+0xbc
> > >>  S: 0000000031a83c20 R: 000000003006b070   .ibm_fsp_cec_reboot+0x7c
> > >>  S: 0000000031a83cb0 R: 0000000030026d20   .opal_cec_reboot+0x6c
> > >>  S: 0000000031a83d40 R: 0000000030005138   opal_entry+0x78
> > >>  S: 0000000031a83f00 R: 00000000300025c4   secondary_wait+0x84
> > >
> > > Since its related to code update I will take a look :-)
> > 
> > Thanks.
> > 
> > _______________________________________________
> > Skiboot mailing list
> > Skiboot at lists.ozlabs.org
> > https://lists.ozlabs.org/listinfo/skiboot
> 
>