On Wed, 2010-01-20 at 10:03 -0800, Stephen Hemminger wrote:

On Wed, 20 Jan 2010 09:41:03 +0000
Jarek Poplawski [off-list ref] wrote:

quoted

quoted

Seems like an underlying bug in the DMA api. Maybe it just can't
handle operations on partial mapping.

Other drivers with same problem:
  bnx2, cassini, pcnet32, r8169, rrunner, skge, sungem, tg3, 

It seems using the same length (even without pci_unmap_len()) is
crucial here, but I hope maintainers (added to CC) will take care.

I'm still unsure how to do dma_sync properly in bnx2.  In the current
code, we always dma_sync_for_cpu a small portion of the SKB because the
rx descriptor is at the beginning of the SKB.  We get the packet length,
for example, from the rx descriptor.

If it's a big packet, we'll simply unmap the entire SKB buffer (with the
beginning portion already dma_sync'ed).  If the packet is smaller than
what we dma_sync'ed, we'll just copy the data to a new SKB.  We'll then
dma_sync_for_device the portion of the original buffer and recycle the
whole buffer back to the device for new packets.

So, is it correct to just change the dma_sync length to the full length
of the buffer?  It doesn't sound right to me.

Thanks.