Re: [PATCH v8 06/14] media: platform: Improve the implementation of the system PM ops

[Tomasz Figa]

On Wed, May 27, 2020 at 3:58 AM Xia Jiang <xia.jiang@xxxxxxxxxxxx> wrote:
>
> On Thu, 2020-05-21 at 15:32 +0000, Tomasz Figa wrote:
> > 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@xxxxxxxxxxxx>
> > > ---
> > >  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?
> Dear Tomasz,
> I mean that if current image is processed in irq handler,suspend
> function can not get the lock(it was locked in irq handler).Should I
> move the spin_lock_irqsave(&jpeg->hw_lock, flags) to the start location
> of suspend function or

Do we have any guarantee that the interrupt handler would be executed
and acquire the spinlock before mtk_jpeg_suspend() is called?

> use wait_event_timeout() to handle the interrupt
> before suspend?

Yes, that would indeed work better. :)

However, please refer to the v4l2_m2m suspend/resume helpers [1] and
the MTK FD driver [2] for how to implement this nicely.

[1] https://patchwork.kernel.org/patch/11272917/
[2] https://patchwork.kernel.org/patch/11272903/

Best regards,
Tomasz