Re: [PATCH] media: rkisp1: Reduce min_queued_buffers to 1

From: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
To: Sebastian Fricke <sebastian.fricke@collabora.com>
Cc: Jacopo Mondi <jacopo.mondi@ideasonboard.com>,
	Dafna Hirschfeld <dafna@fastmail.com>,
	"open list:ROCKCHIP ISP V1 DRIVER" <linux-media@vger.kernel.org>,
	Mauro Carvalho Chehab <mchehab@kernel.org>,
	Heiko Stuebner <heiko@sntech.de>,
	"open list:ROCKCHIP ISP V1 DRIVER"
	<linux-rockchip@lists.infradead.org>,
	Adam Pigg <adam@piggz.co.uk>
Subject: Re: [PATCH] media: rkisp1: Reduce min_queued_buffers to 1
Date: Mon, 7 Oct 2024 16:47:34 +0300	[thread overview]
Message-ID: <20241007134734.GA7439@pendragon.ideasonboard.com> (raw)
In-Reply-To: <20241007125730.yn6xinehhm73yqob@basti-XPS-13-9310>

On Mon, Oct 07, 2024 at 02:57:30PM +0200, Sebastian Fricke wrote:
> Hey Jacopo,
> 
> On 07.10.2024 14:42, Jacopo Mondi wrote:
> > There apparently is no reason to require 3 queued buffers to call
> > streamon() for the RkISP1 as the driver operates with a scratch buffer
> > where frames can be directed to if there's no available buffer provided
> > by userspace.
> > 
> > Reduce the number of required buffers to 1 to allow applications to
> > operate with a single queued buffer.
> > 
> > Tested with libcamera, by operating with a single capture a request. The
> > same request (and associated capture buffer) gets recycled once
> > completed. This of course causes a frame rate drop but doesn't hinder
> > operations.
> > 
> > Signed-off-by: Jacopo Mondi <jacopo.mondi@ideasonboard.com>
> > ---
> > 
> > Adam,
> >    a few months ago you were exercizing your pinhole app with a single capture
> > request for StillCapture operations and you got the video device to hang because
> > no enough buffers where provided.
> > 
> > This small change should be enough to unblock you. Could you maybe give it a
> > spin if you're still working on this ?
> > 
> > Thanks
> >    j
> > ---
> > 
> >  drivers/media/platform/rockchip/rkisp1/rkisp1-capture.c | 4 +---
> >  1 file changed, 1 insertion(+), 3 deletions(-)
> > 
> > diff --git a/drivers/media/platform/rockchip/rkisp1/rkisp1-capture.c b/drivers/media/platform/rockchip/rkisp1/rkisp1-capture.c
> > index 2bddb4fa8a5c..34adaecdee54 100644
> > --- a/drivers/media/platform/rockchip/rkisp1/rkisp1-capture.c
> > +++ b/drivers/media/platform/rockchip/rkisp1/rkisp1-capture.c
> > @@ -35,8 +35,6 @@
> >  #define RKISP1_SP_DEV_NAME	RKISP1_DRIVER_NAME "_selfpath"
> >  #define RKISP1_MP_DEV_NAME	RKISP1_DRIVER_NAME "_mainpath"
> > 
> > -#define RKISP1_MIN_BUFFERS_NEEDED 3
> > -
> >  enum rkisp1_plane {
> >  	RKISP1_PLANE_Y	= 0,
> >  	RKISP1_PLANE_CB	= 1,
> > @@ -1563,7 +1561,7 @@ static int rkisp1_register_capture(struct rkisp1_capture *cap)
> >  	q->ops = &rkisp1_vb2_ops;
> >  	q->mem_ops = &vb2_dma_contig_memops;
> >  	q->buf_struct_size = sizeof(struct rkisp1_buffer);
> > -	q->min_queued_buffers = RKISP1_MIN_BUFFERS_NEEDED;
> 
> It looks like RKISP1_MIN_BUFFERS_NEEDED is used only here, so can you
> remove the define as well?

Isn't that exactly what this patch is doing ?

> rg 'RKISP1_MIN_BUFFERS_NEEDED' 
> drivers/media/platform/rockchip/rkisp1/rkisp1-capture.c
> 38:#define RKISP1_MIN_BUFFERS_NEEDED 3
> 1566:	q->min_queued_buffers = RKISP1_MIN_BUFFERS_NEEDED;
> 
> Or maybe just change the value, but I am not sure whether this can be
> considered a magic value.
> 
> > +	q->min_queued_buffers = 1;
> >  	q->timestamp_flags = V4L2_BUF_FLAG_TIMESTAMP_MONOTONIC;
> >  	q->lock = &node->vlock;
> >  	q->dev = cap->rkisp1->dev;

-- 
Regards,

Laurent Pinchart

_______________________________________________
Linux-rockchip mailing list
Linux-rockchip@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-rockchip