USB regressions in 5.15 and 6.1

[techflashYT]

In 5.15 and 6.1, USB on the Nintendo Wii has undergone several critical regressions.

In 5.15, USB descriptors have started experiencing corruption:

• Flash drive:
  • Manufacturer: USn\n (normally USB\n)
  • Product: SanDisk 3.2Gen\n (normally SanDisk 3.2Gen1\n)
• Keyboard:
  • Manufacturer: Logitece\n (normally Logitech\n)
  • Product: USB Receive\n (normally USB Receiver\n)
• Internal Bluetooth Radio (on the second OHCI bus):
  • Manufacturer: Broadcom Corp\n (correct)
  • Product: BCM2045A\n (correct)

Weirdly, I can't trace this down to any one specific change. Commit e7cb072eb988e46295512617c39d004f9e1c26f8 ("init/initramfs.c: do unpacking asynchronously") is what first started triggering it, but is absolutely not the cause. Something much deeper is going wrong there.

In 6.1, USB falls flat on it's face entirely, no devices can properly be identified.

[Wiibuntu]

This is deeper than initramfs as with my Ubuntu builds I use a straight root boot, no initramfs. And still have USB issues.

[techflashYT]

I specifically said that it's not initramfs related.
That commit is just what started triggering it, back in 5.13 or wherever it was. There's a deeper issue, that's triggered inconsistently, and initramfs affects it.

[Wiibuntu]

I hate to say I have a similar issue in v5.10 without initramfs. USB devices fail all together almost every time. Booting from /dev/mmcblk0p2 works fine still but anything USB even Ethernet fails.

[techflashYT]

USB is always super jank. It has been this way for ages. It doesn't matter what kernel version you're using.
That is not a new issue, and is unrelated to what I'm tracking here.

[superbeelavaman]

iirc the usb jank goes all the way back far enough it's easier to fix it then to find where it broke

[techflashYT]

Except that fixing it typically depends on knowing where/how it broke.

[Wiibuntu]

In v5.7 device class/interface descriptor checks where tightened I think, maybe related to that somehow?

[techflashYT]

But it works as well as it did in 5.4, in 5.10. There was no regression there.
If you think it doesn't, that's because it's completely random, and there is no concrete way to test it, so it may have appeared as though it was working "fine" in 5.4, even if it really wouldn't do so consistently.
This is why I list 4.19 as the latest stable - none of the 5.x builds are fully working yet.