Date:   Fri, 12 Nov 2021 12:03:35 +0000
From:   Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Cc:     linuxarm@huawei.com, mauro.chehab@huawei.com,
        Bhaskar Chowdhury <unixbhaskar@gmail.com>,
        Dan Carpenter <dan.carpenter@oracle.com>,
        Deepak R Varma <drv@mailo.com>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Hans Verkuil <hverkuil-cisco@xs4all.nl>,
        Hans de Goede <hdegoede@redhat.com>,
        Ingo Molnar <mingo@kernel.org>,
        Kaixu Xia <kaixuxia@tencent.com>,
        Mauro Carvalho Chehab <mchehab@kernel.org>,
        Peter Zijlstra <peterz@infradead.org>,
        Randy Dunlap <rdunlap@infradead.org>,
        Sakari Ailus <sakari.ailus@linux.intel.com>,
        Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>,
        Tsuchiya Yuto <kitakar@gmail.com>,
        linux-kernel@vger.kernel.org, linux-media@vger.kernel.org,
        linux-staging@lists.linux.dev
Subject: Re: [PATCH 0/3] atomisp: add support for enum frame rate and sizes
Message-ID: <20211112120335.74e437db@sal.lan>
In-Reply-To: <cover.1636651027.git.mchehab+huawei@kernel.org>
References: <cover.1636651027.git.mchehab+huawei@kernel.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
To:     unlisted-recipients:; (no To-header on input)
Precedence: bulk

Em Thu, 11 Nov 2021 17:27:55 +0000
Mauro Carvalho Chehab <mchehab+huawei@kernel.org> escreveu:

> Adding support for those two ioctls are trivial, and allow testing
> different resolutions with the driver.
> 
> Together with some improvements I made at camorama, it is now
> possible to change the atomisp sensor's resolution at the GUI.
> 
> Talking about camorama, I also added there an option to disable
> the Gtk cairo_scale() calls. On my tests here, placing ov2680 on
> its maximum resolution, I'm getting a framerate of 26 fps, which
> sounds reasonable, as the maximum would be 30fps, and I'm not
> using daylight.
> 
> When letting cairo_scale() to run, the rate reduces to 8 fps, meaning 
> that Gtk is not using GPU acceleration.
> 
> Funny enough, when resolution is lower, atomisp is giving a very
> bad framerate (around 3fps, even with the scaler disabled).
> 
> This is a very weird result, probably indicating some problems inside
> the driver. That requires further investigation.

Ok, when the sensor's resolution is not the highest one, what happens is
that  the frame is retrieved via some timeout logic:

[10447.865064] ov2680 i2c-OVTI2680:00: ov2680_s_stream one
[10447.865491] atomisp-isp2 0000:00:03.0: timeout recovery handling done
[10450.029956] atomisp-isp2 0000:00:03.0: [WARNING]asd 0 pipe ATOMISP ISP PREVIEW output ISP timeout 1!
[10450.030099] atomisp-isp2 0000:00:03.0: pipe on asd0 timeout cnt: (0, 0, 1, 0) of 2, recover = 1
[10450.030137] atomisp-isp2 0000:00:03.0: pipe on asd1 timeout cnt: (0, 0, 0, 0) of 2, recover = 1

It sounds that it can't properly detect the end of frame when the
sensor is not on its highest resolution.

Basically atomisp_wdt_work() is triggered by a 
IA_CSS_EVENT_TYPE_FW_ASSERT event. Identifying why this is happening 
is not trivial, though.

Regards,
Mauro