Am Montag, den 16.01.2012, 09:14 +0100 schrieb Wolfgang Denk:
Dear Stephan Linz,
Hi Wolfgang,
you are right in all points. I'll go into the next round and handle all new request.
br, Stephan
In message 1326652166.18981.213.camel@keto you wrote:
come code already declares PktBuf ... can't you use that ?
Hm, what do you mean exactly here?
These are the two DMA transfer buffers. I have no idea if there are buffers in the upper layer (NET) and how I can use it for DMA transfers.
It's a bad thing not to look at common code, but thnings like that happen too easily.
However, now that this has been pointed out, you cannot simply ignore such a request.
Therfore I create my own rx/tx buffers and copy data. That reduce the performance a little bit, but it's OK. Furthermore I have to use DMA safe buffers here (no cache, 32 byte alignment).
I could imagine this to be a pretty common requirement.
this looks like a struct that should get allocated on the fly based on arguments given to the driver's registration func
OK, it wast a little bit RAM. We can optimize the code later. I want to see more testing results on differnet Microblaze and PPC platforms.
we can fix this now, right?
the memory is already zero-ed by the call to calloc, so this for loop is useless (and if it wasn't, you'd still use memset())
Sure, we can remove this part in one of the next code optimization.
No. Please fix this now, and then add a cleaned up driver. This is what the peer review process is all about.
Best regards,
Wolfgang Denk