Re: [PATCH v7] ethernet/arc/arc_emac - Add new driver

[Arnd Bergmann]

On Friday 21 June 2013, Joe Perches wrote:
> On Fri, 2013-06-21 at 10:53 +0000, Alexey Brodkin wrote:
> > On 06/21/2013 02:32 PM, Joe Perches wrote:
> > > On Fri, 2013-06-21 at 11:20 +0400, Alexey Brodkin wrote:
> > >> Driver for non-standard on-chip ethernet device ARC EMAC 10/100,
> > >> instantiated in some legacy ARC (Synopsys) FPGA Boards such as
> > >> ARCAngel4/ML50x.
> []
> > >> +          rxbd->data = (unsigned char *)cpu_to_le32(rx_buff->skb->data);
> > >
> > > 32 bit only.  Should the Kconfig block have some arch_arc dependency
> > > so it can't get compiled for 64 bit systems?
> []
> > So for now I may easily add dependency on ARC if it makes acceptance of 
> > driver easier.
> 
> I don't think it's a big problem.
> 
> Maybe add a Kconfig "depends on !64BIT".
> 
> Another thing to do would be to run it through sparse
> with make C=1

No, I think the driver should just be made 64-bit clean. Not because
anyone is going to need that of course, but to avoid having
to add silly dependencies.

The problem is that rxbd->data is the wrong type. It is declared
as a void*, but it is not a kernel pointer at all, it is a DMA
pointer. Look at this code in context:


+               addr = dma_map_single(&ndev->dev, (void *)rx_buff->skb->data,
+                                     buflen, DMA_FROM_DEVICE);
+               if (dma_mapping_error(&ndev->dev, addr)) {
+                       if (net_ratelimit())
+                               netdev_err(ndev, "cannot dma map\n");
+                       dev_kfree_skb(rx_buff->skb);
+                       stats->rx_errors++;
+                       continue;
+               }
+               dma_unmap_addr_set(&rx_buff, mapping, addr);
+               dma_unmap_len_set(&rx_buff, len, buflen);
+
+               rxbd->data = (unsigned char *)cpu_to_le32(rx_buff->skb->data);

the 'addr' returned by dma_map_single is what the device really
needs, although it is the same as rx_buff->skb->data with the
trivial definition of dma_map_single.

The last line here needs to be

		rxbd->data = cpu_to_le32(addr);

which fixes the bug, and has no dependency on a 32 bit CPU.

	Arnd
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/