[PATCH v5, 00/15] Using component framework to support multi hardware decode

yunfei.dong at mediatek.com yunfei.dong at mediatek.com
Fri Aug 20 07:59:38 UTC 2021


Hi Ezequiel,

Thanks for your detail feedback. 

On Thu, 2021-08-19 at 11:10 -0300, Ezequiel Garcia wrote:
> On Thu, 19 Aug 2021 at 04:13, yunfei.dong at mediatek.com
> <yunfei.dong at mediatek.com> wrote:
> >
> > Hi Ezequiel,
> >
> > Thanks for your suggestion.
> >
> > On Wed, 2021-08-18 at 11:11 -0300, Ezequiel Garcia wrote:
> > > +danvet
> > >
> > > Hi,
> > >
> > > On Tue, 10 Aug 2021 at 23:58, Yunfei Dong <yunfei.dong at mediatek.com>
> > > wrote:
> > > >
> > > > This series adds support for multi hardware decode into mtk-vcodec,
> > > > by first
> > > > adding component framework to manage each hardware information:
> > > > interrupt,
> > > > clock, register bases and power. Secondly add core thread to deal
> > > > with core
> > > > hardware message, at the same time, add msg queue for different
> > > > hardware
> > > > share messages. Lastly, the architecture of different specs are not
> > > > the same,
> > > > using specs type to separate them.
> > > >
> > >
> > > I don't think it's a good idea to introduce the component API in the
> > > media subsystem. It doesn't seem to be maintained, IRC there's not
> > > even
> > > a maintainer for it, and it has some issues that were never
> > > addressed.
> > >
> > > It would be really important to avoid it. Is it really needed in the
> > > first place?
> > >
> > > Thanks,
> > > Ezequiel
> >
> > For there are many hardware need to use, mt8192 is three and mt8195 is
> > five. Maybe need more to be used in the feature.
> >
> > Each hardware has independent clk/power/iommu port/irq.
> > Use component interface in prob to get each component's information.
> > Just enable the hardware when need to use it, very convenient and
> > simple.
> >
> > I found that there are many modules use component to manage hardware
> > information, such as iommu and drm etc.
> >
> 
> Many drivers support multiple hardware variants, where each variant
> has a different number of clocks or interrupts, see for instance
> struct hantro_variant which allows to expose different codec cores,
> some having both decoder/encoder, and some having just a decoder.
> 
> The component API is mostly used by DRM to aggregate independent
> subdevices (called components) into an aggregated driver.
> 
> For instance, a DRM driver needs to glue together the HDMI, MIPI,
> and plany controller, or any other hardware arrangement where
> devices can be described independently.
> 
The usage scenario is very similar with drm and iommu, So decide to use
component framework.
Decode has three/five or more hardwares, these hardware are independent.
For mt8183 just need core hardware to decode, but mt8192 has lat,soc and
core hardware to decode. When lat need to use, just enable lat hardware,
core is the same.And mt8195 will has two cores, each core can work well
independent.

For each component device just used to open their power/clk/iommu
port/irq when master need to enable it. The main logic is in master
device.

> The component API may look simple but has some issues, it's not easy
> to debug, and can cause troubles if not used as expected [1].
> It's worth making sure you actually need a framework
> to glue different devices together.
> 
Each hardware has its index, master can get hardware information
according these index, looks not complex. What do you mean about not
easy to debug?

> > Do you have any other suggestion for this architecture?
> >
> 
> Looking at the different patchsets that are posted, it's not clear
> to me what exactly are the different architectures that you intend
> to support, can you some documentation which clarifies that?
> 
Have five hardwares lat,soc,core0,core1 and main. Lat thread can use lat
soc and main, core thread can use soc,lat, core0 and core1. Core thread
can be used or not for different project. Also Need to use these
hardware dynamic at the same time. So I use component framework, just
need to know the used  hardware index according to different
project.Need not to do complex logic to manage these hardwares.


> Thanks,
> Ezequiel
> 
> [1] https://patchwork.kernel.org/project/linux-rockchip/cover/20200120170602.3832-1-ezequiel@collabora.com/


Thanks,
Yunfei Dong






More information about the dri-devel mailing list