[OE-core] [PATCH 7/9] linux-yocto: aufs and systemtap

Bruce Ashfield bruce.ashfield at windriver.com
Wed Mar 14 12:46:08 UTC 2018


On 2018-03-14 8:41 AM, Burton, Ross wrote:
> On 6 March 2018 at 18:11, Bruce Ashfield <bruce.ashfield at windriver.com 
> <mailto:bruce.ashfield at windriver.com>> wrote:
> 
>     Integrating a configuration change for systemtap:
> 
>        features/systemtap/systemtap.cfg: enable CONFIG_KERNEL_DEBUG
> 
> 
> I noticed that all the performance machines had a big spike in kernel 
> build time and build directory size, and annoyingly spent too long 
> bisecting builds before noticing this in the commit message for the 
> offending commit.  If I re-read the logs I might have spotted this 
> sooner. :)

Agreed. I should have realized this as well, we went through a process
about a year ago of moving all of the debug configs with the split into
a production/debug set of kernel configurations.

> 
> $ buildstats-diff 20180314122002-3a8cce8b5c2 20180314122610-a7c9bc7020a/
> Ignoring tasks less than 00:03.0 (3.0s)
> Ignoring differences less than 00:01.0 (1.0s)
> 
>    PKG            TASK                      ABSDIFF   RELDIFF  CPUTIME1 
> -> CPUTIME2
>    linux-yocto    do_package                  20.2s   +153.0%     13.2s 
> -> 33.3s
>    linux-yocto    do_deploy                   29.9s  +1502.8%      2.0s 
> -> 31.9s
>    linux-yocto    do_package_write_ipk       125.5s   +310.2%     40.5s 
> -> 166.0s
>    linux-yocto    do_compile_kernelmodules   142.6s    +17.8%    800.3s 
> -> 942.9s
>    linux-yocto    do_package_write_rpm       156.0s   +563.6%     27.7s 
> -> 183.7s
>    linux-yocto    do_compile                 174.8s    +18.7%    934.3s 
> -> 1109.1s
> 
> Cumulative cputime:
>    649.5s    +35.0%    30:56.0 (1856.0s) -> 41:45.5 (2505.5s)
> 
> $ buildhistory-diff | grep SIZE
> packages/qemux86-poky-linux/linux-yocto/kernel-vmlinux: PKGSIZE changed 
> from 19587044 to 300944172 (+1436%)
> 
> Is this really what we want out of the box?  Can the kernel have the 
> debugging bits stripped out into a separate package so even if we have 
> the build time hit, we don't have the kernel size change?

I'll revert this for now (or see if I can quickly move it to a debug
kernel build), and yes, there is a better way to do this. I'm just not
sure I can get it done in time for the release.

Bruce

> 
> Ross




More information about the Openembedded-core mailing list