On Mon, 23 Sep 2013, Robert Hodaszi wrote: > Hi, > > I had a lot of USB errors on i.mx28. I could reproduce it most easily > with USB-serial adapters, and USB modems (CDC and WWAN drivers). If I > sent some data to the TTY port with the echo command in a loop (so > basically open the port, send data, and close the port), all of the > devices were dropped down from the root hub. (Later I found, it is > enough to just periodically open and close the TTY port.) > > The reason was an mx28 errata: "ENGR119653 USB: ARM to USB register > error issue" > > Freescale issued a patch for 2.6.35 to workaround this problem last > year. I ported this patch. However, it is not totally "device tree > compatible". I mean, this patch is only needed for mx28, not for the > other SOCs using chipidea, and I used ifdefs. So you can't compile a > kernel both for mx28 and e.g. for mx23. If you check the patch, you see, > that it is modifying the lowest layer, the writel() function; it is > changing it to an SWP instruction. According the errata, I need to use > SWP instructions to write USB registers. Presumably you _could_ use the SWP instructions for other SOCs, but they would slow down the driver. Right? > So I'm curious, what other possibilities do I have to make this patch > device tree compatible? I mean, I can't check the machine's type on each > USB register writing. Actually you can. See the definition of ehci_writel for an example. > Then I could use function pointers for register > writing function, or weak aliases somehow. But that would causes an > overhead because of the additional function calling, instead of simple > inline assembly codes. > > What's the standard way in this case? Or should I just leave the patch > as it is? Let's see what the Freescale and ChipIdea maintainers suggest. Alan Stern -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
