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

Re: [Xen-devel] Xen4.2 S3 regression?





On Mon, Sep 24, 2012 at 8:05 AM, Jan Beulich <JBeulich@xxxxxxxx> wrote:
>>> On 24.09.12 at 13:54, Ben Guthro <ben@xxxxxxxxxx> wrote:
> I see - Konrad - do you have a changeset I can look for with the features
> Jan describes?

Alternatively you could also stick a wbinvd() in the two possibly
relevant places...


I tried doing this right before disable_nonboot_cpus() in power.c, to no effect.

The behavior that I'm currently failing to understand seems to be a heisenbug, exhibited by the 2 attached patches, which are essentially the same, with a cpumask_copy in 2 different places.

When I have the debug copy just before do_suspend_lowlevel() - (debug1.patch) the output changes to what I would expect:

(XEN) XXX: CACHE issue? before=1 after=1
That is, only CPU0 is initialized both before, and after the S3 cycle.

However, when that same copy is moved up to before disable_nonboot_cpus() - as in debug2.patch - it seems to have an effect on the state afterward. I would have expected this to be something like 

(XEN) XXX: CACHE issue? before=3 after=1
But that is not what is actually printed.

I get
(XEN) XXX: CACHE issue? before=3 after=3
That is, the initialized bit is set for CPU1, when it shouldn't be.


It should be noted that even with the debug in that causes the value to be what I expect, this does not in itself is not sufficient to get things working entirely, as I then get stalls in the kernel rcu_sched


 
Jan

> On Mon, Sep 24, 2012 at 7:45 AM, Jan Beulich <JBeulich@xxxxxxxx> wrote:
>
>> >>> On 24.09.12 at 13:25, Ben Guthro <ben@xxxxxxxxxx> wrote:
>> > It is an older system (Core2) - so it may be using the default mechanism,
>> > but I'd have to go dig up some processor docs to be sure.
>>
>> This is not just a matter of what the CPU supports, but also what
>> info gets passed down from Dom0 - if e.g. your kernel doesn't
>> have Konrad's P-/C-state patches, then there's no way for Xen to
>> use any of the advanced methods.
>>
>> Jan
>>
>>




Attachment: debug1.patch
Description: Binary data

Attachment: debug2.patch
Description: Binary data

_______________________________________________
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®.