Re: [PATCH] media: venus: declare firmware files via MODULE_FIRMWARE

[Dmitry Baryshkov]

On Tue, Apr 21, 2026 at 10:10:14PM +0100, Christopher Obbard wrote:
> Hi Bryan,
> 
> On Tue, 2026-04-21 at 02:23 +0100, Bryan O'Donoghue wrote:
> > On 20/04/2026 20:57, Christopher Obbard wrote:
> > > Hi Bryan,
> > > 
> > > The background for this was to ensure the firmware is available in a
> > > system image
> > > or copied into the initrd if the driver is built-in. This is done in a
> > > CI environment without
> > > access to the hardware.
> > > 
> > > Debian's initramfs-tools looks at MODULE_FIRWMARE and copies those
> > > files into the
> > > initrd if the driver is built-in. For now, a workaround for me could
> > > be to just manually copy
> > > the firmware into the initrd and drop this patch.
> > > 
> > > I didn't get to the bottom of modifying the firmware loading code to
> > > make the driver
> > > defer probe to when the rootfs was ready and instead just threw the
> > > firmware into
> > > the initrd, but that's another story.
> > > 
> > > On Mon, 20 Apr 2026 at 11:20, Bryan O'Donoghue <bod@xxxxxxxxxx> wrote:
> > > > 
> > > > On 19/04/2026 23:39, Christopher Obbard wrote:
> > > > > The driver loads firmware blobs at runtime via request_firmware()
> > > > > but does not currently advertise the possible filenames. Add
> > > > > MODULE_FIRMWARE() entries for all known firmware variants so they are
> > > > > visible via modinfo and can be picked up by user space tooling.
> > > > 
> > > > What about the board dtsi files ?
> > > 
> > > I really am not keen on adding all of the firmware from the board dts files as
> > > MODULE_FIRMWARE entries into the driver, it then becomes two lists to
> > > maintain. What do you think?
> > > 
> > > BUT talking of two lists to maintain, I am essentially duplicating fwname in
> > > MODULE_FIRMWARE, so perhaps my argument is moot.
> > > 
> > > For my use-case, it may make sense to "teach" the initrd generator to look at
> > > the firmware-name entries from the dts files I am building for. But that is also
> > > not ideal.
> > > 
> > > 
> > > Cheers!
> > > 
> > > Chris
> > 
> > I think its reasonable to keep a list of firmware that is supplied in 
> > linux-firmware in the MODULE_FIRMWARE() list.
> > 
> > AFAIK paths such as "qcom/sdm845/SHIFT/axolotl/venus.mbn"; are not in 
> > linux-firmware so .. we shoudln't expect tooling to populate them into 
> > initrd.
> 
> In latest linux-firmware we have:
> 
> $ find -type f | grep -i venus
> ./qcom/venus-6.0/venus.mbn
> ./qcom/venus-5.2/venus.mbn
> ./qcom/venus-1.8/venus.mbn
> ./qcom/venus-4.2/venus.mbn
> ./qcom/venus-5.4/venus.mbn
> 
> So I wonder if I should send v2 adding MODULE_FIRMWARE entries for just those files in linux-firmware?

I'd suggest doing a simpler thing. Add firmware-name strings to
corresponding DT files (updating those when necessary) and add a warning
to iris / venus drivers when there is no firmware in the DT.

Then we won't need to add MODULE_FIRMWARE to the module. Instead you'd
have the standard code which fetches firmware files from the DT.

> 
> 
> Cheers!
> 
> Chris

-- 
With best wishes
Dmitry