On Tue, 07 Dec 2004 13:45:44 +0100, Leif Lindholm leif.lindholm@i3micro.com wrote:
On Mon, 2004-12-06 at 14:05 -0600, Andrew Dyer wrote:
A while back (Sep 23, 2004) someone named Leif posted a note about a patch to add support for the AMD DbAU1550 board to u-boot here https://sourceforge.net/mailarchive/forum.php?thread_id=5624545&forum_id...
Will send this to you privately, and later try to clean it up a bit and fix a few bugs this caused with the db1000 board, so Wolfgang might accept it into the CVS.
I've been working with your patch and expanding it a bit for stuff we wanted to use on our custom au1550 boards.
I noticed a problem with the cpu/mips/au1x00_ether.c net driver - it certainly causes problems for me on the au1550.
next_tx and next_rx are set to zero in the init routine - the au1550 docs (section 9.4.3.2 of au1550 databook) says the dmas may not start at buffer 0 coming out of reset, so those values are not guaranteed to be zero. The correct value can be init'd based on the dma address register CB field.
I changed the lines setting these up in au1x00_init() to:
next_tx = TX_GET_DMA_BUFFER(fifo_tx[0].addr); next_rx = RX_GET_DMA_BUFFER(fifo_rx[0].addr);
this issue was causing me problems with rx and tx timeouts, and changing these values fixes my problem (at least in limited testing). Can someone else try out this change on au1000/1100/1500 based systems?
Also, what is txbuf in au1x00_ether.c used for? It doesn't look like that storage ever gets used - every time through the au1x00_send routine the buffer gets passed in from the net code.