Re: 2.6.18 is problematic in VMware

[Zachary Amsden <zach@vmware.com>]

Jan Engelhardt wrote:
> > My experience with VMware on several recent processors (mostly P-M family)
> > is that it crawls unless I force this first:
> > echo 1 > /sys/module/processor/parameters/max_cstate
> >     
>
> My host processor is non-throttable
> Uni-processor.
>
> I tried with some kernels (hey, vmware had serial ports?)
> For whatever reason, the timecounting is not accurate.
> As you can see, the time difference between "WP bit" and the calibration 
> thing is less than half a second inside the guest, but on the host, 
> the delay is several seconds.
> The Capture Movie feature gets it right 
> http://jengelh.hopto.org/f/2618delay.avi
>   

I see a large delay _before_:

Calibrating delay using timer specific routine.. (lpj==

The lpj value is 38x the host lpj host value.  Booting with lpj="some 
random number" removes the stall for me.  And finally, breaking into 
kernel in this stall period consistently shows EIPs in calibrate_delay 
in backtrace.

> One can see that it pauses before calibration (already mentioned that) 
> and once again after NET: ... during IP init!? What's going on :(
>   

The NET: pause has always been there, I think it just becomes much more 
noticeable if the lpj value is 38x normal.

Conclusion: you are getting bad lpj computation during beginning.  
Doesn't appear to be a kernel bug, so let's take it off list.  You can 
bring it up on VMTN 
http://www.vmware.com/community/index.jspa?categoryID=1 for more 
on-topic support.  In fact, interesting test you can try - suspend / 
resume during the hang.  Does the discontinuity during calibration make 
it go away?

Zach
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/