On Monday, February 28, 2011 13:50:55 Michal Simek wrote:
Mike Frysinger wrote:
On Monday, February 28, 2011 04:40:33 Michal Simek wrote:
- return 1;
a bunch of these funcs return 1 when i'm pretty sure they should be 0
init function is called from eth.c:eth_init(). From the code below you see that return can be >=0.
funny enough, that func in your patch is returning 0 when it should be 1. it doesnt explain why your recv/send are returning 1 when it should be 0.
the init func is a special beast -- it returns the # of devices registered, not "1 is success".
+int xilinx_axiemac_initialize(bd_t *bis, int base_addr, int dma_addr) +{
- struct eth_device *dev;
- struct axidma_priv *dma;
- dev = malloc(sizeof(*dev));
- if (dev == NULL)
hang();- memset(dev, 0, sizeof(*dev));
- sprintf(dev->name, "Xilinx_AxiEmac");
- dev->iobase = base_addr;
- dma = dev->priv;
- dma->dmatx = dma_addr;
- dma->dmarx = (u32)dma->dmatx + 0x30; /* rx channel offset */
hmm, this is weird. you just memset(dev) to 0, and then used dev->priv without assigning it storage. so you're scribbling on top of address 0 with your dma struct here arent you ?
dev contains: iobase - axi emac baseaddr init, halt, send, recv pointers to functions and pointer priv
- others
I need to clear it that's why memset.
dma controller is special IP that's why I use priv for storing pointer to axidma_priv structure which contains two pointers to dma for master to slave and slave to master directions (rx and tx if you like). As I wrote dma controller is different IP that's why there are different addresses. axi_dma has offset between channels which is 0x30. I used structures for hw access.
ok, but i think you missed my point. let me use this example: dev->priv = 0; /* the memset */ dma = dev->priv; dma->dmatx = dma_addr;
you're doing a NULL pointer deref here. -mike