On 18-Nov-2021, at 12:31 AM, Marcel Holtmann [off-list ref] wrote:

Hi Orlando,

quoted

quoted

So if this just affects two macs, why can't the fix be realized as a
quirk that is only enabled on those two systems? Or are they impossible
to detect clearly via DMI data or something like that?

I think we should be able to quirk based off the acpi _CID "apple-uart-blth"
or _HID "BCM2E7C". Marcel suggested quirking based of the acpi table here
https://lore.kernel.org/linux-bluetooth/1D2217A9-EA73-4D93-8D0B-5BC2718D4788@holtmann.org/ (local)

This would catch some unaffected Macs, but they don't support the LE Read
Transmit Power command anyway (the affected macs were released after it
was added to the Bluetooth spec, while the unaffected Macs were released
before it was added to the spec, and thus don't support it).

I'm not sure how to go about applying a quirk based off this, there are
quirks in drivers/bluetooth/hci_bcm.c (no_early_set_baudrate and
drive_rts_on_open), but they don't seem to be based off acpi ids.

It might be simpler to make it ignore the Unknown Command error, like
in this patch https://lore.kernel.org/linux-bluetooth/CABBYNZLjSfcG_KqTEbL6NOSvHhA5-b1t_S=3FQP4=GwW21kuzg@mail.gmail.com/ (local)
however that only applies on bluetooth-next and needed the status it
checks for to be -56, not 0x01.

so we abstain from try-and-error sending of commands. The Bluetooth spec
has a list of supported commands that a host can query for a reason. This
is really broken behavior of the controller and needs to be pointed out as
such.

Well all I can do is provide you any logs or information I can. But we do really wish to get this regression fixed soon.

The question is just how we quirk it.

Regards

Marcel