Re: [PATCH] firmware: maintain a firmware list

Re: [PATCH] firmware: maintain a firmware list

[Danilo Krummrich]

(Cc: driver-core; please make sure to Cc all relevant mailing lists)

On Wed Jul 15, 2026 at 3:13 PM CEST, Corentin Labbe wrote:
> From: Corentin LABBE <clabbe@baylibre.com>
>
> Checking if some firmware is missing need to check dmesg for error message,
> and on some machine this information could be lost since dmesg uses a ring buffer.
> Having the list of all firmware requests is useful in many situations,
> like selecting the minimal list when doing some buildroot config
> or reducing the size of linux-firmware on gentoo via saveconfig.

The proposed implementation aside, I don't see why this needs a new kernel
interface.

  - Why can't you use 'modinfo -F firmware' for all compiled modules?

  - Don't you have a journal persisting the dmesg logs?

  - Why can't you use kprobe_event on the kernel command line if you really need
    a runtime trace?

Untested, but something like

	kprobe_event=p:fw_req,_request_firmware,name=+0(%si):string

should work I think.

Thanks,
Danilo

Re: [PATCH] firmware: maintain a firmware list

[Corentin LABBE]

Le Wed, Jul 15, 2026 at 07:23:25PM +0200, Danilo Krummrich a écrit :
> (Cc: driver-core; please make sure to Cc all relevant mailing lists)
> 
> On Wed Jul 15, 2026 at 3:13 PM CEST, Corentin Labbe wrote:
> > From: Corentin LABBE <clabbe@baylibre.com>
> >
> > Checking if some firmware is missing need to check dmesg for error message,
> > and on some machine this information could be lost since dmesg uses a ring buffer.
> > Having the list of all firmware requests is useful in many situations,
> > like selecting the minimal list when doing some buildroot config
> > or reducing the size of linux-firmware on gentoo via saveconfig.
> 
> The proposed implementation aside, I don't see why this needs a new kernel
> interface.

Hello

You miss the point that I propose an easy way to know.
All your proposals imply scripting, grepping in difference place.
I propose a unique easy way to know the list, without any requirement

> 
>   - Why can't you use 'modinfo -F firmware' for all compiled modules?

Because it display all possible firmware for each module, I want only required/tried module for the host system

>   - Don't you have a journal persisting the dmesg logs?
> 

I propose something that work with embeded devices without storage and/or without syslog/journald.

>   - Why can't you use kprobe_event on the kernel command line if you really need
>     a runtime trace?
> 
> Untested, but something like
> 
> 	kprobe_event=p:fw_req,_request_firmware,name=+0(%si):string
> 
> should work I think.
This imply command line hacking, something I prefer to avoid, it is better when it just works without change.

Thanks
Regards

Re: [PATCH] firmware: maintain a firmware list

[Danilo Krummrich]

On Thu Jul 16, 2026 at 11:29 AM CEST, Corentin LABBE wrote:
> Le Wed, Jul 15, 2026 at 07:23:25PM +0200, Danilo Krummrich a écrit :
>> (Cc: driver-core; please make sure to Cc all relevant mailing lists)
>> 
>> On Wed Jul 15, 2026 at 3:13 PM CEST, Corentin Labbe wrote:
>> > From: Corentin LABBE <clabbe@baylibre.com>
>> >
>> > Checking if some firmware is missing need to check dmesg for error message,
>> > and on some machine this information could be lost since dmesg uses a ring buffer.
>> > Having the list of all firmware requests is useful in many situations,
>> > like selecting the minimal list when doing some buildroot config
>> > or reducing the size of linux-firmware on gentoo via saveconfig.
>> 
>> The proposed implementation aside, I don't see why this needs a new kernel
>> interface.
>
> Hello
>
> You miss the point that I propose an easy way to know.
> All your proposals imply scripting, grepping in difference place.
> I propose a unique easy way to know the list, without any requirement

No, I get that, but I think that the overhead of the proposed patch is not
justified, given that there are reasonable alternatives.

>> Untested, but something like
>> 
>> 	kprobe_event=p:fw_req,_request_firmware,name=+0(%si):string
>> 
>> should work I think.
> This imply command line hacking, something I prefer to avoid, it is better when it just works without change.

Adding a kernel command line parameter seems perfectly reasonable for your
use-case. We could maybe consider a dedicated TRACE_EVENT for firmware being
loaded if this turns out to be a thing people commonly want to have, but for now
it seems a kprobe_event works for your case.