Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Message-ID: <1520260903.2637.34.camel@gmail.com>
Subject: Re: regression: SCSI/SATA failure
From:   Artem Bityutskiy <dedekind1@gmail.com>
Reply-To: dedekind1@gmail.com
To:     Christoph Hellwig <hch@lst.de>
Cc:     linux-kernel@vger.kernel.org, Thomas Gleixner <tglx@linutronix.de>,
        Christian Borntraeger <borntraeger@de.ibm.com>,
        Stefan Haberland <sth@linux.vnet.ibm.com>,
        Jens Axboe <axboe@kernel.dk>,
        "Herring, Jan-kristian Augustin" 
        <jan-kristian.augustin.herring@intel.com>,
        Thorsten Leemhuis <regressions@leemhuis.info>
Date:   Mon, 05 Mar 2018 16:41:43 +0200
In-Reply-To: <1519311270.2535.53.camel@intel.com>
References: <1519311270.2535.53.camel@intel.com>
Content-Type: text/plain; charset="UTF-8"
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Linux-Regression-ID: lr#15a115

On Thu, 2018-02-22 at 16:54 +0200, Artem Bityutskiy wrote:
> Hi Christoph,
> 
> one of our test box Skylake servers does not boot with v4.16-rcX.
> Bisection lead us to this commit:
> 
> 84676c1f21e8 genirq/affinity: assign vectors to all possible CPUs
> 
> Reverting this single commit fixes the problem.
> 
> The server is a Dell R640 machine with the latest Dell BIOS. It has a
> single SATA SSD and we do not use raid, even though the system does
> have a megaraid controller.

Correction: we have Raid0 with this single disk.

> Are you aware of this issue? Below is the failure message and the
> full
> dmesg with some debugging boot parameters is here:
> 
> https://pastebin.com/raw/tTYrTAEQ

FYI, the regression still exists and reverting this single patch fixes
it. But today Dell server

I did not have time to really debug this, but I think people who are
working with this should quickly see what is going on.

I think the platform reports way too large possible CPU count. Indeed,
in dmesg I see this:

[    0.000000] smpboot: Allowing 328 CPUs, 224 hotplug CPUs

224 is way too large for this system. It only has 2 sockets, it but the
number looks like if the system had 4 sockets.

The commit changes IRQ affinity logic from being per-present CPU to
being per-possible CPU:

-       for_each_present_cpu(cpu)
+       for_each_possible_cpu(cpu)

And it looks like this has an unexpected side-effect on this Dell
platform.

Artem.