[lkml]   [2014]   [Jul]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH 2/6] x86-mce: Modify CMCI storm exit to reenable instead of rediscover banks.
On Wed, Jul 09, 2014 at 02:34:39PM -0700, Havard Skinnemoen wrote:
> On Wed, Jul 9, 2014 at 1:20 PM, Luck, Tony <> wrote:
> >> The CMCI storm handler previously called cmci_reenable() when exiting a
> >> CMCI storm. However, when entering a CMCI storm the bank ownership was
> >> not relinquished by the affected CPUs. The CMCIs were only disabled via
> >> cmci_storm_disable_banks(). The handler was updated to instead call a
> >> new function, cmci_storm_enable_banks(), to reenable CMCI on the already
> >> owned banks instead of rediscovering CMCI banks (which were still owned
> >> but disabled).
> >
> > Won't this cause problems if we online a cpu during the storm. We will
> > re-run the discovery algorithm and some other cpu that shares the bank
> > will see MCi_CTL2{30} is zero and claim ownership.
> Yes, I think you're right. We didn't test this with CPU hotplugging.
> I'm at loss about how to fix it though. We need the CMCI bits to
> detect shared banks, but they're not reflecting the actual state of
> things at that point. If the CPU gives up ownership of the banks, then
> we might just see the storm move from CPU to CPU, right?
> We could keep a separate bitmask somewhere to indicate ownership, but
> even if we can see that the bank is shared with some other CPU, we
> don't know if it will be shared with a new CPU which we've never seen
> before...
> Perhaps we need to temporarily disable the storm handling when we're
> bringing up a new CPU?

Looking at this more, maybe cmci_storm_disable_banks() was a bad idea
after all. There's __cmci_disable_bank() which properly drops ownership
after having disabled CMCI.

Maybe we should kill cmci_storm_disable_banks() and do
__cmci_disable_bank by iterating over them all...


Sent from a fat crate under my desk. Formatting is fine.

 \ /
  Last update: 2014-07-10 18:21    [W:0.093 / U:3.880 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site