mt76: Fix file loading

[System64fumo]

The previous method used hard coded firmware paths to load the firmware from disk. That works if the driver is built as a module but does not work if baked in.

The proper way to load firmware is with request_firmware which handles both file loading and blobs.

I also noticed that the mt76_get_eeprom_file function wasn't getting called at all due to a seemingly unused variable?

struct device_node *np = dev->dev->of_node;
if (!np)
return -ENOENT;

Likely a leftover from previous code.
Either way commenting it out seems to work.

I've tested firmware blob loading (baked into the kernel) and that seems to work, Regular file loading should also still work but i've not tested it.

[frank-w]

afaik mt76_get_eeprom_file should not be called from mt76_get_of_eeprom and leave the check. the node must be checked when loading from devicetree, but file does not need the mt76 node...so the logic has to be changed.

the changes above are because changing kernel_read to request_firmware? patch works for you without crashes, also when file does not exist?

[System64fumo]

I've not tested it when the file does not exist, I could test that but it should not crash, It would just default back to the missing firmware behavior (The behavior i was experiencing before this)

[System64fumo]

Alright, Just tested it without firmware and it behaves as expected:

[ 2.657673] mt7615e 0000:01:00.0: enabling device (0000 -> 0002)
[ 2.657814] mt7615e 0000:01:00.0: Requesting EEPROM: mediatek/mt7615e_rf.bin
[ 2.657849] mt7615e 0000:01:00.0: Direct firmware load for mediatek/mt7615e_rf.bin failed with error -2
[ 2.657854] mt7615e 0000:01:00.0: Failed to load EEPROM mediatek/mt7615e_rf.bin: -2
[ 2.659113] mt7615e 0000:01:00.0: Invalid MAC address, using random address ee:31:e9:49:7f:1a

The interface works but is using safe fallback defaults, In other words the same behavior as without this patch.

@frank-w Anything else you need me to test/change about this?

[frank-w]

Have not yet found time to look deeper into it. Imho the return for non-existent dt node must stay and the firmware load via file must be not called from there.

[System64fumo]

Can you elaborate on your points?
The DT check is commented out because otherwise the rest of the firmware loading never happens.
As for where mt76_get_eeprom_file is called from, I didn't change that so not sure where you would suggest it be called from.

[frank-w]

Imho the call to mt76_get_of_eeprom in mt76_eeprom_init should be changed to call file loading not in mt76_get_of_eeprom

[System64fumo]

So move it to mt76_eeprom_init?

mt76_eeprom_init would call the file loader first then fall back to the other loaders if that fails?

[frank-w]

Yes makes more sense. And does not break dt function.

[System64fumo]

Looks good?

[frank-w]

looks good so far, it is tested with latest change? could you squash the 2 commits into 1?

[System64fumo]

Yup tested this for basic functionality (Baked in firmware)
Didn't test it without a firmware file but it should still work as expected. (LMK if you want that tested too)

As for merging, Isn't there a squash button when pulling on your end?

[frank-w]

there is only "rebase and merge" with no squash option and it is easier to review when having this as 1 commit :)

[System64fumo]

Should be good now, Sorry for the hold up.
Any chance of this ending upstream someday btw?

[frank-w]

You can squash with my commit, rebase on openwrt/mt76 and send a pull-request there too. Currently i have no way to test it against all circumstances