
On Thu, Aug 18, 2011 at 11:13:19AM -0400, Ben Gardiner wrote:
I understand that both Laurence and Stefan (cc'd) have confirmed that there dcache issues with the EMA; I am assuming that the 'issue' results in a delay in tftp'ing...
No; the issue was that packets were being sent on to the wire with incorrect IP checksums. It twigged that the problem was cache coherency when I put a full hexdump of each outgoing packet in and suddenly the outbound packets started being valid.
Disabling the caches will slow down decompression which will slow down boot overall when booting from flash.
Rather than masking the issue by disabling caches and slowing down u-boot for it's users perhaps TI should be fixing the EMAC drivers' cache bugs instead?
Indeed, the correct solution is to properly manage the net buffers with respect to the cache, although I have not attempted that change myself. I have seen some patches start to flow that make changes in drivers to work correctly with caches enabled, such as http://lists.denx.de/pipermail/u-boot/2011-August/098484.html .
Bye for now,