lkml.org 
[lkml]   [2022]   [Jun]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: Commit 282d8998e997 (srcu: Prevent expedited GPs and blocking readers from consuming CPU) cause qemu boot slow
From


On 6/15/2022 4:20 PM, Paolo Bonzini wrote:
> On 6/15/22 12:40, Neeraj Upadhyay wrote:
>>
>> This is useful data, thanks! Did you get chance to check between 100
>> and 1000, to narrow down further, from which point (does need to be
>> exact value) between 100 and 1000,  you start seeing degradation at,
>> for ex. 250, 500 , ...?
>>
>> Is it also possible to try experiment 10 and 11 with below diff.
>> What I have done in below diff is, call srcu_get_delay() only once
>> in try_check_zero() (and not for every loop iteration); also
>> retry with a different delay for the extra iteration which is done
>> when srcu_get_delay(ssp) returns 0.
>>
>> Once we have this data, can you also try by changing
>> SRCU_RETRY_CHECK_LONG_DELAY   to 100, on top of below diff.
>>
>> #define SRCU_RETRY_CHECK_LONG_DELAY  100
>
> Is there any data that you would like me to gather from the KVM side,
> for example with respect to how much it takes to do synchronize_srcu()
> on an unpatched kernel, or the duration of the read-sides?
>

Hi Paolo,

Thanks! I think synchronize_srcu() time and read side duration will
help. Given that changing SRCU_MAX_NODELAY_PHASE value has impact
on the boot time (and SRCU_MAX_NODELAY_PHASE impact is dependent on
duration of SRCU read side duration), this information will be helpful
to get more understanding of the scenario.


Thanks
Neeraj

> Thanks,
>
> Paolo
>

\
 
 \ /
  Last update: 2022-06-15 13:05    [W:1.348 / U:0.008 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site