Re: xfs kdevops baseline for next-20230725

["Darrick J. Wong" <djwong@kernel.org>]

On Thu, Jul 27, 2023 at 04:01:44PM -0700, Luis Chamberlain wrote:
> On Thu, Jul 27, 2023 at 03:58:59PM -0700, Darrick J. Wong wrote:
> > On Thu, Jul 27, 2023 at 11:21:35AM -0700, Luis Chamberlain wrote:
> > > The sections tested for are:
> > > 
> > > xfs_crc
> > > xfs_reflink
> > > xfs_reflink-normapbt
> > > xfs_reflink_1024
> > > xfs_reflink_2k
> > > xfs_reflink_4k
> > > xfs_nocrc
> > > xfs_nocrc_512
> > > xfs_nocrc_1k
> > > xfs_nocrc_2k
> > > xfs_nocrc_4k
> > > xfs_logdev
> > > xfs_rtdev
> > > xfs_rtlogdev
> > 
> > Question: Have you turned on gcov to determine how much of fs/xfs/ and
> > fs/iomap/ are actually getting exercised by these configurations?
> 
> Not yet, and it would somehow have to be the aggregate sum of the
> different guests, is that easy to sum up these days, given the same
> kernel is used?

I think so?  I sent out that patch to fstests that automates fstests
pushing gcov data into the report dir.  If you take each machine's data
to the build machine and make the paths match, you can then run lcov on
each report dir to spit out an lcov file, and then tell genhtml to spit
out an html report given all the lcov output files.

(Or you can mount the kernel source in the same place on the test vm as
the build machine, and fstests will at least run lcov for you...)

> > I have for my fstests fleet; it's about ~90% for iomap, ~87% for
> > xfs/libxfs, ~84% for the pagecache, and ~80% for xfs/scrub.  Was
> > wondering what everyone else got on the test.
> 
> Neat, it's something we could automate provided if you already have
> a way to sum this up, dump it somehwere and we I can make a set of
> ansible tasks to do it.

That part gets a little gross, because gcc embeds canonical source code
pathnames into the kernel binary, which has proven to be a tripping
point for getting lcov/genhtml to find the .c and .h files.

--D

> 
>   Luis