On Mon, Jun 26, 2017 at 08:25:36AM -0600, Jens Axboe wrote:

On 06/26/2017 08:06 AM, Johannes Thumshirn wrote:

quoted

On Fri, Jun 23, 2017 at 09:36:14AM -0600, Jens Axboe wrote:

quoted

On 06/23/2017 08:29 AM, Johannes Thumshirn wrote:

quoted

From: Omar Sandoval <redacted>

This test-case performs I/O with fio while doing PCI disable/enable
cycles.

In the results we don't care for I/O errors but for hiccups in dmesg only.

Let's get this in, that would be a very useful test. A few comments -
not necessarily on this patch in particular, but for future cleanups
and improvements.

quoted

+	if _test_dev_is_rotational; then
+		size="32m"
+	else
+		size="1g"
+	fi

I introduced this idea in one of my previous patches. I wonder if we
should turn that into a helper. Pass in the dev, get returned a
suitable fio size, instead of hard coding this in each job that
needs it.

What I wanted to have here eventually is a helper that you can run when
you just want arbitrary I/O. Haven't gotten around to it.

quoted

Sure.

quoted

quoted

+	# start fio job
+	_run_fio --bs=4k --rw=randread --norandommap \
+		--name=reads --filename="$TEST_DEV" --size="$size" \
+		--numjobs=8 --direct=1 2>/dev/null &

I don't believe we check for fio errors right now, but we probably
should in the future. So I think you'd want to add something ala:

--ignore_error=EIO,ENXIO,ENODEV

to your options to make it explicit that you don't care about IO
errors for this test.

Yup, we redirect fio errors to /dev/null everywhere, we should fix that.

quoted

Oh nice, didn't know about the option. Btw as we're currently all have
arbitrary values for the numjobs parameter, how about a wrapper over getconf
_NPROCESSORS_ONLN?

Yes that's a good idea, then we can at least size the jobs based on
how many cores we have.

We can just use nproc for this.