Revert "bcmgenet: constrain max DMA burst length"

[nbuchwitz]

While debugging the genet issue, I spotted another redundant downstream patch.

fd2e504 dropped DMA_MAX_BURST_LENGTH from 0x10 to 0x08 back in 2018 to work around BCM2711 TX DMA timeouts. @lategoodbye's upstream a50e3a9 made this obsolete a year later by adding bcmgenet_plat_data and pinning 0x08 for BCM2711 only.

So the global override has already been a no-op on Pi hardware for a while.

[nbuchwitz]

btw: 7dd290a looks messed up and changes genet too. rebase / merge went bad? Origin seems to be 6c7d38b

Looked up the history of this commit (raspberrypi/rpi-eeprom#417) and the initial issue should be fixed in firmware with raspberrypi/rpi-eeprom#424. So given the amount of potential energy savings, this is also worth considering a revert. I will do some testing.

[pelwell]

7dd290a seems to be a squash of 6c7d38b and the patched up reversion of 5bef4b3, which is itself a reversion of 5255660.

I think they need to be separated, but not otherwise changed. Do you agree?

[pelwell]

And I agree with this reversion of a downstream patch that is no longer required.

[nbuchwitz]

7dd290a seems to be a squash of 6c7d38b and the patched up reversion of 5bef4b3, which is itself a reversion of 5255660.

I think they need to be separated, but not otherwise changed. Do you agree?

AFAIU it can / should be reverted too, once this is split. The firmware should be changed for quite a few years and according to Florian the power savings are worth it. I will update the pr with a split + revert so we can test. OK?

[pelwell]

I will update the pr with a split + revert so we can test. OK?

Sure - I'll rope @P33M in for comment.

[P33M]

In #3030 I narrowed the reproducer to Pi 4 + while true; do lsusb -v -d 2109:3431; done - without the quirk, lsusb will break after several cycles.

[nbuchwitz]

Seems to be a misunderstanding, I was only talking about the genet patch which was accidentaly squashed into the xhci patch. So the revert would be only for the genet part

[pelwell]

I was fooled - your talk of firmware made me think of the VL805.

[nbuchwitz]

My bad. Firmware = bootloader firmware in spi flash (phy needed proper reset for netboot four years ago)

[nbuchwitz]

to be continued here #7403