Re: [PATCH] media: rzg2l-cru: Replace usleep_range with udelay

[David Laight <david.laight.linux@gmail.com>]

On Wed, 3 Dec 2025 09:45:43 +0100
Tommaso Merciai <tommaso.merciai.xr@bp.renesas.com> wrote:

> Hi Laurent,
> 
> Thanks for your review.
> 
> On Wed, Dec 03, 2025 at 11:35:52AM +0900, Laurent Pinchart wrote:
> > Hi Tommaso,
> > 
> > Thank you for the patch.
> > 
> > On Tue, Dec 02, 2025 at 05:08:41PM +0100, Tommaso Merciai wrote:  
> > > `usleep_range()` should not be used in atomic contexts like between
> > > spin_lock_irqsave()/spin_lock_irqrestore() pair inside function
> > > rzg2l_cru_stop_image_processing(). That may cause scheduling while
> > > atomic bug.
> > > 
> > > Signed-off-by: Khai Nguyen <khai.nguyen.wx@renesas.com>
> > > Signed-off-by: Hao Bui <hao.bui.yg@renesas.com>
> > > Signed-off-by: Tommaso Merciai <tommaso.merciai.xr@bp.renesas.com>
> > > ---
> > >  drivers/media/platform/renesas/rzg2l-cru/rzg2l-video.c | 4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/drivers/media/platform/renesas/rzg2l-cru/rzg2l-video.c b/drivers/media/platform/renesas/rzg2l-cru/rzg2l-video.c
> > > index 162e2ace6931..1355bfd186d4 100644
> > > --- a/drivers/media/platform/renesas/rzg2l-cru/rzg2l-video.c
> > > +++ b/drivers/media/platform/renesas/rzg2l-cru/rzg2l-video.c
> > > @@ -369,7 +369,7 @@ void rzg2l_cru_stop_image_processing(struct rzg2l_cru_dev *cru)
> > >  		if (cru->info->fifo_empty(cru))
> > >  			break;
> > >  
> > > -		usleep_range(10, 20);
> > > +		udelay(10);  
> > 
> > There's an instance of msleep() earlier in this function, surrounded by
> > spin_unlock_irqrestore() and spin_lock_irqsave(). I wondered if we
> > should do the same here, but that lead to a second question: why does
> > the driver need to cover the whole stop procedure with a spinlock in the
> > first place ?  
> 
> Good point :)
> Mmm maybe the only critical section into the
> rzg2l_cru_stop_image_processing() that needs
> spin_unlock_irqrestore()/spin_lock_irqsave()
> is:
> 
> spin_lock_irqsave(&cru->qlock, flags);
> cru->state = RZG2L_CRU_DMA_STOPPED;
> spin_unlock_irqrestore(&cru->qlock, flags);

That doesn't look right.
You pretty much never need a lock for a single assignment.
The most you need is a WRITE_ONCE() - and they are pretty unlikely
to matter at all.

	David


> 
> Please correct me if I'm wrong.
> 
> Kind Regards,
> Tommaso
> 
> >   
> > >  	}
> > >  
> > >  	/* Notify that FIFO is not empty here */
> > > @@ -385,7 +385,7 @@ void rzg2l_cru_stop_image_processing(struct rzg2l_cru_dev *cru)
> > >  			AMnAXISTPACK_AXI_STOP_ACK)
> > >  			break;
> > >  
> > > -		usleep_range(10, 20);
> > > +		udelay(10);
> > >  	}
> > >  
> > >  	/* Notify that AXI bus can not stop here */  
> > 
> > -- 
> > Regards,
> > 
> > Laurent Pinchart  
>