Re: [meta-freescale] Linux v5.10 for i.MX8 development

On Tue, Sep 21, 2021 at 1:36 AM Andrey Zhizhikin <[email protected]> wrote:
>
> Hello Ankur,
>
> On Mon, Sep 20, 2021 at 11:30 PM Ankur Tyagi <[email protected]> wrote:
> >
> > Hi,
> >
> > We are currently using Linux v5.4 for i.MX8 provided by recipe linux-fslc-imx_5.4.bb because as mentioned in recipe as well, branch "5.4-2.3.x-imx" from repo "linux-fslc" get changes merged from branch "imx_5.4.70_2.3.0" of linux-imx repo (https://source.codeaurora.org/external/imx/linux-imx/log/?h=imx_5.4.70_2.3.0)
> >
> > Now we are planning to upgrade to Linux v5.10 and here I have few questions:
> >
> > 1. Recipe linux-fslc-lts_5.10.bb doesn't support i.MX8 https://github.com/Freescale/meta-freescale/blob/325ad529187fb1309b0a889c5196c8df039aeea9/recipes-ke... Neither I can see any merge from NXP linux-imx branch "lf-5.10.y" into linux-fslc branch "5.10.x+fslc".
>
> When you say that i.MX8 is not supported - to which SoC you're
> referring to? Mainline kernel on the [master] branch of this layer
> does provide a descent support for all i.MX8M derivatives, but I'm not
> entirely sure of the state for i.MX8 family SoC support since I do not
> have that HW.

Andrey,

By decent support for i.MX8M derivatives you mean no USB, no PCI, no
DSI, no MIPI right? :)

I haven't looked at the yocto kernel recipes but I spend a lot of time
with the mainline kernel that I'm sure they are based on. While most
of the power-domain stuff landed in 5.13 we are still waiting for
blk-ctl which is needed to finish out the power-domain and reset
control for all of those blocks unfortunately.

>
> Can you specify which derivative you're talking about?
>
> > Is there any plan for that merge? If yes, any timeline by which it will happen?
>
> I've looked into updating NXP branch [lf-5.10.y], and it is not
> trivial... I'm now short on time to have a full swing of merging the
> latest stable update on top of it, but probably would get to this
> eventually. If there are any volunteers here that are willing to
> undertake this task - this would be really appreciated.

Could you explain the difference between NXP's 'lf-' and 'imx-'
branches? I can never understand their branch naming scheme nor their
commit log prefix scheme.

Best regards,

Tim

>
> > If no, then can I use "imx_v8_defconfig" from linux-imx repo to build kernel using linux-fslc repo?
>
> You can use the `defconfig` file, which is present in the layer to
> build [linux-fslc] for i.MX8 machine. It resides in
> recipes-kernel/linux/linux-fslc/armv8a/defconfig and is synchronized
> with stable kernel release 5.14.
>
> If you however opt-in for [linux-fslc-lts] kernel package, then the
> kernel config file would be taken from
> recipes-kernel/linux/linux-fslc-lts/armv8a/defconfig, which is also
> synchronized with 5.10 from stable.
>
> Are there any specific configuration options you need from
> imx_v8_defconfig, which should be present in the defconfig for
> mainline recipes?
>
> > i.MX8 is supported in linux-imx_5.10.bb recipe https://github.com/Freescale/meta-freescale/blob/325ad529187fb1309b0a889c5196c8df039aeea9/recipes-ke... which is using kernel from branch "lf-5.10.y" provided by linux-imx repo but this repo is NXP internal and doesn't get updated with stable Linux kernel. It is providing v5.10.35 (https://source.codeaurora.org/external/imx/linux-imx/log/?h=lf-5.10.y)
>
> Correct, but as I've already pointed out - the merge of stable updates
> to this kernel base is not trivial... :/
>
> >
> > So ideally, just like v5.4 would like to use linux-fslc repo for v5.10 development as well.
> >
> >
> > Will appreciate any guidance around appropriate repo for v5.10 development for i.MX8
>
> As one of the things to try out: you can switch the BSP flavor to use
> "mainline" instead of "nxp", which would take all upstream components
> instead of those provided by NXP. Be aware, that if you rely on any
> functionality which is present in NXP BSP only (e.g. GPU/VPU) - this
> would be unavailable in Mainline BSP.
>
> >
> > thanks
> > Ankur
> >
> >
>
>
> --
> Regards,
> Andrey.
>
> 
>