Re: [PATCH v2 1/2] mce: acpi/apei: Honour Firmware First for MCA banks listed in APEI HEST CMC

["Naveen N. Rao" <naveen.n.rao@linux.vnet.ibm.com>]

On 06/21/2013 01:04 PM, Borislav Petkov wrote:
> On Fri, Jun 21, 2013 at 02:52:25AM +0530, Naveen N. Rao wrote:
>> Exactly, but mce_poll_banks also doesn't have bits set for banks on
>> which CMCI is enabled.
>>
>> Let's say we have a cpu with 2 banks (not shared), none of which work
>> in FF mode. Both these banks support CMCI, so mce_poll_banks won't
>> have these bits set.
>>
>> On cpu offline, we call cmci_clear() which disables CMCI on these two
>> banks before offlining it. When this cpu is brought online again, we
>> call cmci_discover() which sees that mce_poll_banks doesn't have these
>> two banks enabled and will skip enabling CMCI thinking these are in
>> FF.
>
> Hmm, mce_intel has yet another bitfield - mce_banks_owned. (Btw, this is
> why I have a problem with adding yet another bitfield).
>
> The way I understand it is, if a bit is set in the owned bitfield, those
> banks belong to CMCI and are not polled.
>
> Now, can we use both mce_banks_owned and mce_poll_banks? If a bit in
> both bifields is cleared, the corresponding bank is not polled *and* is
> not owned by CMCI => it is in FF mode.
>
> Makes sense?
>

Yes, but I'm afraid this won't work either - mce_banks_owned is cleared 
during cpu offline. This is necessary since a cmci rediscover is 
triggered on cpu offline, so that if this bank is shared across cores, a 
different cpu can claim ownership of this bank.

The difference between the new bitfield and the existing bitfields is 
that the new one is not per-cpu. This is a global list of banks across 
cpus that we do not want enabled.


Thanks,
Naveen