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;
Date:   Thu, 19 Sep 2019 08:13:01 -0600
From:   Keith Busch <kbusch@kernel.org>
To:     Bharat Kumar Gogada <bharatku@xilinx.com>
Cc:     "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        "linux-nvme@lists.infradead.org" <linux-nvme@lists.infradead.org>,
        Keith Busch <keith.busch@linux.intel.com>,
        "keith.busch@intel.com" <keith.busch@intel.com>
Subject: Re: NVMe Poll CQ on timeout
Message-ID: <20190919141301.GA61660@C02WT3WMHTD6>
References: <MN2PR02MB633689DBBA6DE9DD7A34043FA5890@MN2PR02MB6336.namprd02.prod.outlook.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <MN2PR02MB633689DBBA6DE9DD7A34043FA5890@MN2PR02MB6336.namprd02.prod.outlook.com>
User-Agent: Mutt/1.12.1 (2019-06-15)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, Sep 19, 2019 at 01:47:50PM +0000, Bharat Kumar Gogada wrote:
> Hi All,
> 
> We are testing NVMe cards on ARM64 platform, the card uses MSI-X interrupts.
> We are hitting following case in drivers/nvme/host/pci.c
> /*
>          * Did we miss an interrupt?
>          */
>         if (__nvme_poll(nvmeq, req->tag)) {
>                 dev_warn(dev->ctrl.device,
>                          "I/O %d QID %d timeout, completion polled\n",
>                          req->tag, nvmeq->qid);
>                 return BLK_EH_DONE;
>         }
> 
> Can anyone tell when does nvme_timeout gets invoked ?

Timeout is invoked when the driver didn't see a completion to a
submitted command.

> In what cases we see this interrupt miss ?

That usually happens for one of two reasons:

 1. The device didn't send any MSIx message for a CQE

 2. The device sent the MSIx message before posting the CQE

I've also seen h/w errata where the MSIx and CQE are re-ordered, which
can also lead to this.

A hardware trace would provide the most detailed view of what's
happening. You might be able to infer if you carefully account for
commands sent, interrupts received, and spurious interrupts detected.

> We are seeing this issue only for reads with following fio command 
> fio --name=randwrite --ioengine=libaio --iodepth=1 --rw=randread --bs=128k --direct=0 \
> --size=128M --numjobs=3 --group_reporting --filename=/dev/nvme0n1
> 
> We are not seeing issue with --rw=randwrite for same size.
> 
> Please let us know what can cause this issue.