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;
Date:   Thu, 21 May 2020 15:32:57 +0000
From:   Tomasz Figa <tfiga@chromium.org>
To:     Xia Jiang <xia.jiang@mediatek.com>
Cc:     Hans Verkuil <hverkuil-cisco@xs4all.nl>,
        Mauro Carvalho Chehab <mchehab+samsung@kernel.org>,
        Rob Herring <robh+dt@kernel.org>,
        Matthias Brugger <matthias.bgg@gmail.com>,
        Rick Chang <rick.chang@mediatek.com>,
        linux-media@vger.kernel.org, devicetree@vger.kernel.org,
        linux-kernel@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
        linux-mediatek@lists.infradead.org,
        Marek Szyprowski <m.szyprowski@samsung.com>,
        srv_heupstream@mediatek.com, senozhatsky@chromium.org,
        mojahsu@chromium.org, drinkcat@chromium.org,
        maoguang.meng@mediatek.com, sj.huang@mediatek.com
Subject: Re: [PATCH v8 06/14] media: platform: Improve the implementation of
 the system PM ops
Message-ID: <20200521153257.GF209565@chromium.org>
References: <20200403094033.8288-1-xia.jiang@mediatek.com>
 <20200403094033.8288-7-xia.jiang@mediatek.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200403094033.8288-7-xia.jiang@mediatek.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi Xia,

On Fri, Apr 03, 2020 at 05:40:25PM +0800, Xia Jiang wrote:
> Cancel reset hw operation in suspend and resume function because this
> will be done in device_run().

This and...

> Add spin_lock and unlock operation in irq and resume function to make
> sure that the current frame is processed completely before suspend.

...this are two separate changes. Please split.

> 
> Signed-off-by: Xia Jiang <xia.jiang@mediatek.com>
> ---
>  drivers/media/platform/mtk-jpeg/mtk_jpeg_core.c | 11 +++++++++--
>  1 file changed, 9 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/media/platform/mtk-jpeg/mtk_jpeg_core.c b/drivers/media/platform/mtk-jpeg/mtk_jpeg_core.c
> index dd5cadd101ef..2fa3711fdc9b 100644
> --- a/drivers/media/platform/mtk-jpeg/mtk_jpeg_core.c
> +++ b/drivers/media/platform/mtk-jpeg/mtk_jpeg_core.c
> @@ -911,6 +911,8 @@ static irqreturn_t mtk_jpeg_dec_irq(int irq, void *priv)
>  	u32 dec_ret;
>  	int i;
>  
> +	spin_lock(&jpeg->hw_lock);
> +

nit: For consistency, it is recommended to always use the same, i.e. the
strongest, spin_(un)lock_ primitives when operating on the same spinlock.
In this case it would be the irqsave(restore) variants.

>  	dec_ret = mtk_jpeg_dec_get_int_status(jpeg->dec_reg_base);
>  	dec_irq_ret = mtk_jpeg_dec_enum_result(dec_ret);
>  	ctx = v4l2_m2m_get_curr_priv(jpeg->m2m_dev);
> @@ -941,6 +943,7 @@ static irqreturn_t mtk_jpeg_dec_irq(int irq, void *priv)
>  	v4l2_m2m_buf_done(src_buf, buf_state);
>  	v4l2_m2m_buf_done(dst_buf, buf_state);
>  	v4l2_m2m_job_finish(jpeg->m2m_dev, ctx->fh.m2m_ctx);
> +	spin_unlock(&jpeg->hw_lock);
>  	pm_runtime_put_sync(ctx->jpeg->dev);
>  	return IRQ_HANDLED;
>  }
> @@ -1191,7 +1194,6 @@ static __maybe_unused int mtk_jpeg_pm_suspend(struct device *dev)
>  {
>  	struct mtk_jpeg_dev *jpeg = dev_get_drvdata(dev);
>  
> -	mtk_jpeg_dec_reset(jpeg->dec_reg_base);
>  	mtk_jpeg_clk_off(jpeg);
>  
>  	return 0;
> @@ -1202,19 +1204,24 @@ static __maybe_unused int mtk_jpeg_pm_resume(struct device *dev)
>  	struct mtk_jpeg_dev *jpeg = dev_get_drvdata(dev);
>  
>  	mtk_jpeg_clk_on(jpeg);
> -	mtk_jpeg_dec_reset(jpeg->dec_reg_base);
>  
>  	return 0;
>  }
>  
>  static __maybe_unused int mtk_jpeg_suspend(struct device *dev)
>  {
> +	struct mtk_jpeg_dev *jpeg = dev_get_drvdata(dev);
> +	unsigned long flags;
>  	int ret;
>  
>  	if (pm_runtime_suspended(dev))
>  		return 0;
>  
> +	spin_lock_irqsave(&jpeg->hw_lock, flags);

What does this spinlock protect us from? I can see that it would prevent
the interrupt handler from being called, but is it okay to suspend the
system without handling the interrupt?

> +
>  	ret = mtk_jpeg_pm_suspend(dev);
> +

Looking at the implementation of mtk_jpeg_pm_suspend(), all it does is
disabling the clock. How do we make sure that there is no frame currently
being processed by the hardware?

Best regards,
Tomasz