[PATCH v2 5/9] media: staging: rkisp1: add a pointer to rkisp1_device from struct rkisp1_isp

[Tomasz Figa]

On Sat, Aug 01, 2020 at 06:08:06PM +0200, Dafna Hirschfeld wrote:
> 
> 
> On 21.07.20 17:32, Tomasz Figa wrote:
> > On Tue, Jul 21, 2020 at 5:30 PM Dafna Hirschfeld
> > <dafna.hirschfeld at collabora.com> wrote:
> > > 
> > > Hi,
> > > 
> > > On 21.07.20 14:36, Tomasz Figa wrote:
> > > > On Tue, Jul 21, 2020 at 2:26 PM Dafna Hirschfeld
> > > > <dafna.hirschfeld at collabora.com> wrote:
> > > > > 
> > > > > Hi,
> > > > > 
> > > > > On 20.07.20 21:25, Helen Koike wrote:
> > > > > > 
> > > > > > 
> > > > > > On 7/18/20 11:59 AM, Dafna Hirschfeld wrote:
> > > > > > > The code in rkisp1-isp.c requires access to struct 'rkisp1_device'
> > > > > > > in several places. It access it using the 'container_of' macro.
> > > > > > > This patch adds a pointer to 'rkisp1_device' in struct 'rkisp1_isp'
> > > > > > > to simplify the access.
> > > > > > 
> > > > > > What is wrong with container_of?
> > > > > 
> > > > > I remember Laurent suggested it a while ago.
> > > > > I also feel container_of is a bit cumbersome and other entities already have a pointer to rkisp1_device.
> > > > > 
> > > > 
> > > > Do we even need the rkisp1_isp struct? Could we just pass rkisp1_device instead?
> > > 
> > > pass to where ?  You mean to the function rkisp1_mipi_csi2_start ?
> > 
> > Yes, all around the driver, where rkisp1_isp is passed.
> 
> Most of the functions are part of subdevice callback implementation
> where the rkisp1_device is not needed, so I don't the see the point.

Okay, so then I'd just lean towards keeping it as is. container_of is
generally considered more efficient than a pointer, because it doesn't
require a load operation to obtain a reference to the parent struct.

Best regards,
Tomasz