Re: [PATCH] firmware_class: Add firmware filename overrides

[Charles Mooney]

On Thu, Mar 5, 2015 at 5:27 PM, Ming Lei <ming.lei@xxxxxxxxxxxxx> wrote:
> On Fri, Mar 6, 2015 at 6:36 AM, Dmitry Torokhov <dtor@xxxxxxxxxxxx> wrote:
>> On Thu, Mar 5, 2015 at 2:12 PM, Marcel Holtmann <marcel@xxxxxxxxxxxx> wrote:
>>> Hi Dmitry,
>>>
>>>>>>>>>> Specifically this was motivated by a situation where we have one
>>>>>>>>>> device with a dual-sourced touchscreen. Both use the same driver but
>>>>>>>>>> have different hardware & fw. Our FW updating software therefore,
>>>>>>>>>> needs to be able to update with the correct FW and detect all this at
>>>>>>>>>> runtime due to a read-only partition (so moving the firmware binaries
>>>>>>>>>> around isn't really an option)
>>>>>>>>>> Here the device has only one touchscreen at a time, but it isn't known
>>>>>>>>>> until run-time which will be present.
>>>>>>>>>>
>>>>>>>>>> So in this case the driver is serving the same function in each
>>>>>>>>>> situation (running a touchscreen) but may be working with different
>>>>>>>>>> hardware.
>>>>>>>>>>
>>>>>>>>>> Another situation where I've personally wanted this functionality is
>>>>>>>>>> on a device that uses the same touch driver for both a touchscreen and
>>>>>>>>>> a touchpad on the same device. If the driver only grabs a copy of FW
>>>>>>>>>> from, say, /lib/firmware/touch_fw.bin then you either need to move the
>>>>>>>>>> firmware binaries around on disk to update either device, or have a
>>>>>>>>>> change like this that allows you to override which filename it loads.
>>>>>>>>>> The moving option is not viable if you're using a RO filesystem.
>>>>>>>>>
>>>>>>>>> what is the actual problem here? We have drivers that load multiple firmware files and we have drivers that pick a different firmware depending on some parameters it reads from the device.
>>>>>>>>>
>>>>>>>>> Seems this is all possible already at the moment with the existing framework. You just need to update the drivers to operate properly.
>>>>>>>>>
>>>>>>>>
>>>>>>>> I totally agree, this functionality is not novel.  We could have added
>>>>>>>> this feature into the specific driver in question, but then we will
>>>>>>>> have to do the same thing on all the other drivers we might want to do
>>>>>>>> this on.  I guess the real problem that this solves is by adding the
>>>>>>>> change here, it allows you to override firmware names for *any* driver
>>>>>>>> without having to duplicate the functionality in each one as they come
>>>>>>>> up.
>>>>>>>>
>>>>>>>> For a specific instance, here at ChromiumOS we have devices that use
>>>>>>>> Atmel, Cypress, Synaptics, and Elan touchpads and touchscreens that
>>>>>>>> all can encounter this issue.  The Atmel driver has a similar version
>>>>>>>> of this feature baked into it but the others don't.  We could add a
>>>>>>>> fw_filename attribute to each of these drivers, but then it would have
>>>>>>>> to be maintained across (at least) four drivers.
>>>>>>>
>>>>>>> what I am hearing here is that you can not query the hardware and figure out which manufacturer it is and with that don't know which firmware you need.
>>>>>>
>>>>>> Right, the drivers in question (bunch of input drivers such as
>>>>>> elan_ts, atmel_mxt_ts, etc) might not be able to determine the
>>>>>> "proper" configuration to load. Factories quite often swap
>>>>>> pin-compatible parts and want to use the same image. Also the parts
>>>>>> can be swapped out during RMA while keeping the same image. Userspace
>>>>>> is able to query magnitude of sources and make an informed decision
>>>>>> that it then communicates to the kernel.
>>>>>>
>>>>>>> However if that is the case, then this seems to be something that should be solved with device tree.
>>>>>>
>>>>>> Why are we making device tree a hard dependency here?
>>>>>
>>>>> device tree is suppose to describe the hardware in your devices. If you can not determinate your hardware by enumeration or other means, then it should be done via device tree. Seems the perfect fit here.
>>>>
>>>> And if I do not have device tree?
>>>
>>> so if you do not have an enumeration method for your hardware, then I assume you most likely have device tree in one form or another.
>>>
>>> And even if you really don't, nothing is stopping you from adding device tree.
>>
>> We have ACPI (for example) and no, it is not 5.0.
>
> It depends if the driver can determinate what the device is from
> ACPI. If yes, you can just load the corresponding fw image
> for the current device. Otherwise the ACPI can't help your problem.
>
> Thanks,
> Ming Lei

We run into the situation where to very similar devices (all the same
HW models) need to ship with the same OS image.  One device may have a
pin-for-pin compatable, 2nd source version of some piece of hardware.
The device tree/etc is all the same because the two slighty different
parts are connected the same way (same i2c address, or similar).

The only way to tell them apart is by talking to the driver once the
device is already up and running.  In our motivating case it's reading
a sysfs attribute to get a manufacturer ID, but it could be anything
similar.

If you want to be able to put a different FW on these two very similar
devices (that can only be differentiated once they're up and running)
I think this might be the best way to do it -- apart from altering the
driver for every part that needs this kind of special treatment.

-Charlie
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/