Re: [PATCH blktests] create a test for direct io offsets

[Keith Busch <kbusch@kernel.org>]

On Mon, Oct 20, 2025 at 12:40:07PM +0000, Shinichiro Kawasaki wrote:
> > +static void test_full_size_aligned()
> > +{
> > +	int ret;
> > +
> > +	memset(in_buf, 0, buf_size);
> > +	ret = pwrite(test_fd, out_buf, buf_size, 0);
> 
> As I noted before, when I tried with QEMU SAS drive, the buf_size is large
> (192MiB) and pwrite returned the value smaller than that (128MiB). This
> caused the compare() failure below. How about to check the return values
> from pwrite() and pread(), and pass it to compare()?

Oh, short writes are not expected. I should check that the actual
written matched the requested total. I wonder why it's bailing out part
way through, though.

> > +        if (ret < 0) {
> > +		if (should_fail)
> > +			return;
> > +		err(errno, "%s: failed to write buf", __func__);
> 
> When I ran the test with QEMU SATA drive, it failed here. The drive
> had virt_boundary=1. I'm not sure if it means test side bug or kernel side bug.

Hm, I tested QEMU's AHCI and that seems okay. Is that what you're using
to attach the "SATA" drive? I'd like to recreate this and figure out
what's going on here.

If this test says virt_boundary is 1 for your device, that should mean
there is no virtual bounday to consider; this test should succeed with
those reported queue limits.