RE: [RFC/PATCH v2 0/3] add gadget quirk to adapt f_fs for DWC3

[Alan Stern]

On Wed, 30 Oct 2013, David Laight wrote:

> > Wouldn't it be simpler and safer to just do this unconditionally? Sure,
> > you need it for DWC3 because the controller refuses to do an OUT transfer
> > at all if the transfer size is less than maxpacketsize. But it's possible
> > that other controllers allow the transfer, and it works in most cases,
> > but if an error occurs and the host sends too much data, they could
> > overrun the buffer and crash your device.
> > 
> > For example, the DWC2 databook says "For OUT transfers, the Transfer
> > Size field in the endpoint's Transfer Size register must be a multiple
> > of the maximum packet size of the endpoint". But I don't think the
> > controller enforces that, it is up to the programmer to do the right
> > thing. So that controller probably needs this quirk also. There could be
> > more like that which we don't know about.
> > 
> > So unless the buffer allocation code is in a real fast path, I would
> > suggest to just do the aligned buffer allocation always.
> 
> You wouldn't normally want to pad OUT transfers that way - if only
> because of the additional USB bandwidth use.

What additional bandwidth use?  Allocating more memory doesn't mean any
additional data will be transmitted over the USB bus.

> Also, if the controller can't do (I assume bulk) OUT (and IN?)
> transfers for less than maxpacketsize it seriously restricts
> the type of devices that can be attached - none of the USB
> ethernet devices would work.

The controllers _are_ capable of doing shorter transfers.  You are
missing the point: These are device controllers, not host controllers,
so they don't have any choice about the length of an OUT transfer.

Alan Stern

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/