On Tue, Sep 12, 2017 at 09:38:20AM -0600, Ross Zwisler wrote:

On Tue, Sep 12, 2017 at 04:44:11PM +1000, Dave Chinner wrote:

quoted

On Fri, Sep 08, 2017 at 03:21:53PM -0600, Ross Zwisler wrote:

quoted

This adds a regression test for the following kernel patch:

  xfs: always use DAX if mount option is used

This test will also pass with kernel v4.14-rc1 and beyond because the XFS
DAX I/O mount option has been disabled (but not removed), so the
"chattr -x" to turn off DAX doesn't actually do anything.

Signed-off-by: Ross Zwisler <redacted>
Suggested-by: Christoph Hellwig <hch@infradead.org>

.....

quoted

+pgsize=`$here/src/feature -s`
+
+# disable tracing, clear the existing trace buffer and turn on dax tracepoints
+echo 0 > /sys/kernel/debug/tracing/tracing_on
+echo > /sys/kernel/debug/tracing/trace
+echo 1 > /sys/kernel/debug/tracing/events/fs_dax/enable
+
+# enable tracing for our one mmap I/O, then see if dax was used
+echo 1 > /sys/kernel/debug/tracing/tracing_on
+xfs_io  -t -c "truncate $pgsize" \
+	-c "chattr -x" \
+	-c "mmap -r 0 $pgsize" -c "mread 0 $pgsize" -c "munmap" \
+	-f $SCRATCH_MNT/testfile >> $seqres.full
+echo 0 > /sys/kernel/debug/tracing/tracing_on

So what happens when the user is already tracing the test to
find a bug and the test turns all their tracing off?

Regardless of this screwing up developer bug triage, do we really
want to add a dependency on kernel tracing into the test harness?

Yep, these are both valid concerns.  I ended up trying to get around both of
them by using perf instead in my v2, as suggested by Dan:

https://www.spinics.net/lists/linux-xfs/msg10420.html

I think similar concerns exist with using perf, too....

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com