[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCHv1] x86: only check for one watchdog NMI



>>> On 23.06.15 at 11:34, <andrew.cooper3@xxxxxxxxxx> wrote:
> On 23/06/15 10:09, Jan Beulich wrote:
>>>>> On 22.06.15 at 18:21, <david.vrabel@xxxxxxxxxx> wrote:
>>> Since the NMI handler can now recognize watchdog NMIs, make
>>> check_nmi_watchdog() only check for at least one watchdog NMI.  This
>>> prevents false negatives caused by other processors (which may be
>>> being power managed by the BIOS) running at reduced clock frequencies.
>> Doesn't this go a little too far, in that it allows through the case where
>> an NMI works exactly once (a behavior not unheard of)? Lowering the
>> count to e.g. 3 would seem acceptable, but not any further. Raising
>> the wait time might then need to be the way to go if the (approximate)
>> 1:3 ratio still isn't enough to cope with BIOS power managed CPUs.
>>
>> Btw., can an OS know of the power state CPUs come up in? I.e. can
>> the wait time be adjusted dynamically? Or is this (perhaps intentionally)
>> completely transparent to the OS?
> 
> With the mwait driver, probably, as it is specifically designed to
> completely bypass any firmware settings which might be in place.

The mwait driver deals with C states, while reduced frequency
means P states (and we specifically busy wait here to prevent
entering and C state).

> Anything else is dependent on how much information can be gleaned from
> the ACPI tables, but most firmware deliberately has a "BIOS controlled"
> mode which is designed to restrict what the OS is capable of doing.

None of which would help in identifying what state a CPU is in.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.