Re: [dm-devel] use regular gendisk registration in device mapper

From: Milan Broz <gmazyland@gmail.com>
To: Christoph Hellwig <hch@lst.de>
Cc: Mike Snitzer <snitzer@redhat.com>, Jens Axboe <axboe@kernel.dk>,
	linux-block@vger.kernel.org, dm-devel@redhat.com
Subject: Re: [dm-devel] use regular gendisk registration in device mapper
Date: Wed, 28 Jul 2021 10:37:41 +0200	[thread overview]
Message-ID: <9e668239-78cc-55ad-8998-b7e39f573c34@gmail.com> (raw)
In-Reply-To: <20210728070655.GA5086@lst.de>

On 28/07/2021 09:06, Christoph Hellwig wrote:
> On Tue, Jul 27, 2021 at 10:38:16PM +0200, Milan Broz wrote:
>> BTW it would be also nice to run cryptsetup testsuite as root - we do a lot
>> of DM operations there (and we depend on sysfs on some places).
> 
> It already doesn't seem very happy in current mainline for me:
> 
> =======================
> 13 of 17 tests failed
> (12 tests were not run)
> =======================
> 
> but this series doesn't seem to change anything.
> 
> A lot of the not run tests seem to be due to broken assumptions
> that some code must be modular.  E.g. my kernel has scsi_debug built
> in, but it complains like this:
> 
> modprobe: ERROR: ../libkmod/libkmod.c:586 kmod_search_moddep() could not open moddep file '/lib/module'
> modprobe: FATAL: Module scsi_debug not found in directory /lib/modules/5.14.0-rc3+

Hi,

there should not be many assumptions, but yes, we depend on modular scsi_debug in some tests because we simulate
very specific hw attributes. So you have one emulated device compiled-in?

Or there is another way how to configure scsi_debug if compiled-in? (we use module parameters, I think it is
the same was how util-linux testsute works with scsi_debug).

Anyway, this is a bug, tests should be skipped (the same way if scsi_debug is not available).

I forgot to say - there is a list of packages that should be installed for make check mentioned
in README.md - I guess this was the reason some other tests were skipped.

(BTW could you send me output of the failed test run? I run it over Linus' tree and ti works so it is perhaps another
assumption that should be fixed.)

Thanks,
Milan