Re: [PATCH v2] usb: hcd: Add a usb_device argument to hc_driver.endpoint_reset()

[Greg Kroah-Hartman]

On Wed, Apr 16, 2025 at 08:29:58AM +0200, Michał Pecio wrote:
> On Tue, 15 Apr 2025 14:26:26 +0200, Greg Kroah-Hartman wrote:
> > > This fixes a 6.15-rc1 regression reported by Paul, which I was able
> > > to reproduce, where xhci_hcd doesn't handle endpoint_reset() after
> > > endpoint_disable() not followed by add_endpoint(). If a configured
> > > device is reset, stalling endpoints start to get stuck permanently.
> >
> > As this fixes a bug, can you add a Fixes: tag with the needed
> > information?
> 
> Hi Greg,
> 
> Sorry for bothering you, the real bug is that I forgot to carry over
> the RFC tag from v1.
> 
> The 6.15 regression is currently solved by reverts Mathias sent you.

Oh good!

> The underlying bug is much older, I would have to research where it
> went wrong exactly. It was very obscure; a class driver would need to:
> 
> 1. call usb_set_interface(), usb_reset_device() or something like that
> 2. submit some URBs to make the toggle/sequence state non-zero
> 3. call usb_clear_halt() on a not yet halted endpoint
> 
> Then the host endpoint wouldn't be reset, but the device would.
> 
> I know of drivers which do 1 and 2 or even 2 and 3, but I have not
> yet encountered a driver doing all three in this order.

Ick, I don't think we want the individual drivers to have to do this,
the host controller _should_ handle it as you are trying to do here.

Anyway, I'll let this one be on the list for now and wait for others to
review.

thanks,

greg k-h