Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Message-ID: <5f6b14c0a86e419b05102599602307a4782d8c64.camel@gmail.com>
Subject: Re: [PATCH] scsi: ufs: add missing host_lock in setup_xfer_req
From:   Bean Huo <huobean@gmail.com>
To:     Bart Van Assche <bvanassche@acm.org>,
        Jaegeuk Kim <jaegeuk@kernel.org>, linux-kernel@vger.kernel.org,
        linux-scsi@vger.kernel.org
Cc:     Stanley Chu <stanley.chu@mediatek.com>,
        Can Guo <cang@codeaurora.org>, Bean Huo <beanhuo@micron.com>,
        Asutosh Das <asutoshd@codeaurora.org>
Date:   Wed, 14 Jul 2021 23:22:24 +0200
In-Reply-To: <b0cd26d0-6ebc-b633-8669-a558597cc91d@acm.org>
References: <20210701005117.3846179-1-jaegeuk@kernel.org>
         <38432018-e8bf-f9f3-00bf-cd4b81c95c88@acm.org>
         <69b367bc44084f901d0d71fb8f9633ea7e5df36b.camel@gmail.com>
         <b0cd26d0-6ebc-b633-8669-a558597cc91d@acm.org>
Content-Type: text/plain; charset="UTF-8"
User-Agent: Evolution 3.36.5-0ubuntu1 
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Wed, 2021-07-14 at 11:09 -0700, Bart Van Assche wrote:
> On 7/13/21 12:45 PM, Bean Huo wrote:
> 
> > This change only impacts on the Samsung exynos platform. and Can's
> > optimization patch is to optimise the host_lock,, and removed
> > host_lock, now add back in this function makes sense to me.
> > but I am thinking how about hba->host_sem?
> 
> 
> Hi Bean,
> 
> 
> 
> Calls of exynos_ufs_specify_nexus_t_xfer_req() must be serialized,
> hence 
> 
> Jaegeuks' patch. This function is called from the I/O submission path
> so 
> 
> performance matters. My understanding is that spinlocks have less 
> 
> overhead than semaphores. Hence the choice for a spinlock.
> 
> 
> 
> Thanks,

> 
Bart,

After adding spin_lock/unlock_irqsave()
in ufshcd_vops_setup_xfer_req(), there will be 4 times of call of
host_lock lock/unlock in ufshcd_send_command(). Reduce the code scope
of protection, but increase the number of calls to
spin_lock/unlock_irqsave().  Almost each sub-funciton
in ufshcd_send_command() calls spin_lock/unlock_irqsave(). why not
directly take spin_lock/unlock_irqsave() away from each sub-fun, and
increase the scope in ufshcd_send_command()?

Bean


> Bart