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;
MIME-Version: 1.0
References: <20191003191354.GA4481@Serenity>
In-Reply-To: <20191003191354.GA4481@Serenity>
From:   Tyler Ramer <tyaramer@gmail.com>
Date:   Fri, 4 Oct 2019 11:36:42 -0400
Message-ID: <CAKcoMVC2LdcmUx6j5JzuT-TsFGz=mwQ0MsprrKR2qeXoTmQ-TQ@mail.gmail.com>
Subject: Re: [PATCH] nvme-pci: Shutdown when removing dead controller
To:     Keith Busch <kbusch@kernel.org>, Jens Axboe <axboe@fb.com>,
        Christoph Hellwig <hch@lst.de>,
        Sagi Grimberg <sagi@grimberg.me>,
        linux-nvme@lists.infradead.org, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Here's a failure we had which represents the issue the patch is
intended to solve:

Aug 26 15:00:56 testhost kernel: nvme nvme4: async event result 00010300
Aug 26 15:01:27 testhost kernel: nvme nvme4: controller is down; will
reset: CSTS=0x3, PCI_STATUS=0x10
Aug 26 15:02:10 testhost kernel: nvme nvme4: Device not ready; aborting reset
Aug 26 15:02:10 testhost kernel: nvme nvme4: Removing after probe
failure status: -19

The CSTS warnings comes from nvme_timeout, and is printed by
nvme_warn_reset. A reset then occurs
Controller state should be NVME_CTRL_RESETTING

Now, in nvme_reset_work, controller is never marked "CONNECTING"  at:

     if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_CONNECTING))

because several lines above, we can determine that
nvme_pci_configure_admin_queues returns
a bad result, which triggers a goto out_unlock and prints "removing
after probe failure status: -19"

Because state is never changed to NVME_CTRL_CONNECTING or
NVME_CTRL_DELETING, the
logic added in https://github.com/torvalds/linux/commit/2036f7263d70e67d70a67899a468588cb7356bc9
should not apply. We can further validate that dev->ctrl.state ==
NVME_CTRL_RESETTING thanks to
the WARN_ON in nvme_reset_work.


On Thu, Oct 3, 2019 at 3:13 PM Tyler Ramer <tyaramer@gmail.com> wrote:
>
> Always shutdown the controller when nvme_remove_dead_controller is
> reached.
>
> It's possible for nvme_remove_dead_controller to be called as part of a
> failed reset, when there is a bad NVME_CSTS. The controller won't
> be comming back online, so we should shut it down rather than just
> disabling.
>
> Signed-off-by: Tyler Ramer <tyaramer@gmail.com>
> ---
>  drivers/nvme/host/pci.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
> index c0808f9eb8ab..c3f5ba22c625 100644
> --- a/drivers/nvme/host/pci.c
> +++ b/drivers/nvme/host/pci.c
> @@ -2509,7 +2509,7 @@ static void nvme_pci_free_ctrl(struct nvme_ctrl *ctrl)
>  static void nvme_remove_dead_ctrl(struct nvme_dev *dev)
>  {
>         nvme_get_ctrl(&dev->ctrl);
> -       nvme_dev_disable(dev, false);
> +       nvme_dev_disable(dev, true);
>         nvme_kill_queues(&dev->ctrl);
>         if (!queue_work(nvme_wq, &dev->remove_work))
>                 nvme_put_ctrl(&dev->ctrl);
> --
> 2.23.0
>