Re: [RFC] Align tegra-ehci DMA transfers to 32B

[Greg KH]

On Fri, Dec 17, 2010 at 02:44:29PM -0800, rmorell@xxxxxxxxxx wrote:
> On Fri, Dec 17, 2010 at 02:32:27PM -0800, Greg KH wrote:
> > On Fri, Dec 17, 2010 at 01:58:47PM -0800, Robert Morell wrote:
> > > This small set of patches fixes an issue where DMA from the tegra EHCI
> > > controller could be corrupted.  It was most commonly seen with USB network
> > > adapters, though in theory it could happen with any USB traffic.
> > > 
> > > (Note: An attempt was made to fix this with commit 367c3aab, which set
> > > NET_IP_ALIGN to 0 and NET_SKB_PAD to 32.  Unfortunately, not all network
> > > drivers honor them (presumably since these are intended as optimizations rather
> > > than hard rules).  This does mean that properly-written network drivers should
> > > fall through this code with very little overhead, however.)
> > 
> > We don't have many USB network drivers, why not just fix them up to
> > handle this properly, then you will not need to change any core USB
> > code, right?
> 
> The USB core code is used by devices other than USB adapters.  We've
> only seen this problem so far with usbnet devices, but I can't test
> every USB device ever to make sure that they always align their DMA to
> 32 bytes.

Then it might just be easier for your driver to throw up a huge
WARN_ON() if it detects such memory so that the device driver could be
fixed, right?

thanks,

greg k-h
--
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/