On Fri, Aug 16, 2013 at 9:34 AM, Peter A. Bigot pab@pabigot.com wrote:
On 08/16/2013 08:38 AM, Tom Rini wrote:
On Wed, Aug 14, 2013 at 09:53:16PM -0500, Peter A. Bigot wrote:
On 07/09/2013 02:43 AM, Naumann Andreas wrote:
In chapter 'Advisory 2.1 USB Host Clock Drift Causes USB Spec Non-compliance in Certain Configurations' of the TI Errata it is recommended to use certain div/mult values for the DPLL5 clock setup. So far u-boot used the old 34xx values, so I added the errata recommended values specificly for 36xx init only. Also, the FSEL registers exist no longer, so removed them from init.
Tested this on a AM3703 board with 19.2MHz oscillator, which previously couldnt lock the dpll5 (kernel complained). As a consequence the EHCI USB port wasnt usable in U-Boot and kernel. With this patch, kernel panics disappear and USB working fine in u-boot and kernel.
Signed-off-by: Andreas Naumann anaumann@ultratronik.de
While this patch works with Linux that has been patched for this erratum, it will cause problems with some unpatched versions of Linux.
Right. So Linux also needs to be patched for the erratum.
Yes. My point was that if you update u-boot alone, then try to use it to boot an unpatched Linux that assumes CM_CLKSEL5_PLL has its power-on value, USB will not work.
I think it's dangerous to assume that the mixture of an unpatched Linux with a patched u-boot will never occur, and the cause of the failure that results is pretty subtle. So whatever gets merged would be safer if it restored the default setting of CM_CLKSEL5_PLL prior to handing off control to Linux.
Agree, we should not apply this, till we also have an 'approved' patch for mainline linux posted. Right now we have a set of kernel hacks, but no agreed on method as the kernel maintainer did not have a board that suffered from the errata..
Regards,