Received-SPF: pass (google.com: domain of linux-kernel+bounces-125411-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
From: Andrew Halaney <ahalaney@redhat.com>
To: Andy Gross <agross@kernel.org>,
	Bjorn Andersson <andersson@kernel.org>,
	Konrad Dybcio <konrad.dybcio@linaro.org>,
	Manivannan Sadhasivam <mani@kernel.org>,
	"James E.J. Bottomley" <jejb@linux.ibm.com>,
	"Martin K. Petersen" <martin.petersen@oracle.com>,
	Hannes Reinecke <hare@suse.de>,
	Janek Kotas <jank@cadence.com>,
	Alim Akhtar <alim.akhtar@samsung.com>,
	Avri Altman <avri.altman@wdc.com>,
	Bart Van Assche <bvanassche@acm.org>,
	Can Guo <quic_cang@quicinc.com>,
	Anjana Hari <quic_ahari@quicinc.com>
Cc: Andrew Halaney <ahalaney@redhat.com>,
	Will Deacon <will@kernel.org>,
	linux-arm-msm@vger.kernel.org,
	linux-scsi@vger.kernel.org,
	linux-kernel@vger.kernel.org
Subject: [PATCH v5 10/11] scsi: ufs: core: Remove unnecessary wmb() after
 ringing doorbell
Date: Fri, 29 Mar 2024 15:46:52 -0500
Message-ID: <20240329-ufs-reset-ensure-effect-before-delay-v5-10-181252004586@redhat.com>
In-Reply-To: <20240329-ufs-reset-ensure-effect-before-delay-v5-0-181252004586@redhat.com>
References: <20240329-ufs-reset-ensure-effect-before-delay-v5-0-181252004586@redhat.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: quoted-printable

Currently, the doorbell is written to and a wmb() is used to commit it=0D
immediately.=0D
=0D
wmb() ensures that the write completes before following writes occur,=0D
but completion doesn't mean that it isn't stored in a buffer somewhere.=0D
The recommendation for ensuring this bit has taken effect on the device=0D
is to perform a read back to force it to make it all the way to the=0D
device. This is documented in device-io.rst and a talk by Will Deacon on=0D
this can be seen over here:=0D
=0D
    https://youtu.be/i6DayghhA8Q?si=3DMiyxB5cKJXSaoc01&t=3D1678=0D
=0D
But, completion and taking effect aren't necessary to guarantee here.=0D
=0D
There's already other examples of the doorbell being rung that don't do=0D
this. The writel() of the doorbell guarantees prior writes by this=0D
thread (to the request being setup for example) complete prior to the=0D
ringing of the doorbell, and the following=0D
wait_for_completion_io_timeout() doesn't require any special memory=0D
barriers either.=0D
=0D
With that in mind, just remove the wmb() altogether here.=0D
=0D
Fixes: ad1a1b9cd67a ("scsi: ufs: commit descriptors before setting the door=
bell")=0D
Signed-off-by: Andrew Halaney <ahalaney@redhat.com>=0D
---=0D
 drivers/ufs/core/ufshcd.c | 3 ---=0D
 1 file changed, 3 deletions(-)=0D
=0D
diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c=0D
index dfa4f827766a..a2f2941450fd 100644=0D
--- a/drivers/ufs/core/ufshcd.c=0D
+++ b/drivers/ufs/core/ufshcd.c=0D
@@ -7090,10 +7090,7 @@ static int __ufshcd_issue_tm_cmd(struct ufs_hba *hba=
,=0D
 =0D
 	/* send command to the controller */=0D
 	__set_bit(task_tag, &hba->outstanding_tasks);=0D
-=0D
 	ufshcd_writel(hba, 1 << task_tag, REG_UTP_TASK_REQ_DOOR_BELL);=0D
-	/* Make sure that doorbell is committed immediately */=0D
-	wmb();=0D
 =0D
 	spin_unlock_irqrestore(host->host_lock, flags);=0D
 =0D
=0D
-- =0D
2.44.0=0D
=0D