On Wed, Jun 25, 2014 at 4:04 PM, Russell King - ARM Linux
[off-list ref] wrote:

On Tue, Jun 24, 2014 at 10:29:48PM -0600, Dann Frazier wrote:

quoted

On Fri, Jun 20, 2014 at 5:18 PM, Iyappan Subramanian
[off-list ref] wrote:

quoted

+       ring->desc_addr = dma_zalloc_coherent(dev, size, &ring->dma,
+                                             GFP_KERNEL);

Iyappan,
When testing this driver on a 3.16-rc2 base, I'm finding that
desc_addr gets assigned to NULL here, which results in an oops later
on (see below).

Note that on failure here...

quoted

quoted

+       if (!ring->desc_addr)
+               goto err;

we jump to 'err'.

quoted

quoted

+err:
+       dma_free_coherent(dev, size, ring->desc_addr, ring->dma);

which then tries to call dma_free_coherent on a NULL pointer, and
possibly undefined ring->dma value.  That's not a nice thing to do,
and will probably lead to problems.  I know that none of the ARM
flavours of this function will handle this gracefully, and neither
does x86's version either.  So this is very probably illegal.

I will fix this.

quoted

quoted

+static int xgene_enet_create_desc_rings(struct net_device *ndev)
+{
+       struct xgene_enet_pdata *pdata = netdev_priv(ndev);
+       struct device *dev = &pdata->pdev->dev;
+       struct xgene_enet_desc_ring *rx_ring, *tx_ring, *cp_ring;
+       struct xgene_enet_desc_ring *buf_pool = NULL;
+       u8 cpu_bufnum = 0, eth_bufnum = 0;
+       u8 bp_bufnum = 0x20;
+       u16 ring_id, ring_num = 0;
+       int ret;
+
+       /* allocate rx descriptor ring */
+       ring_id = xgene_enet_get_ring_id(RING_OWNER_CPU, cpu_bufnum++);
+       rx_ring = xgene_enet_create_desc_ring(ndev, ring_num++,
+                                             RING_CFGSIZE_16KB, ring_id);
+       if (IS_ERR_OR_NULL(rx_ring)) {
+               ret = PTR_ERR(rx_ring);
+               goto err;
+       }

Here we test for IS_ERR_OR_NULL. In the oops I'm hitting, rx_ring is
NULL here - but PTR_ERR() apparently returns 0 in that case. So this
function ends up returning no error.

Yes, IS_ERR_OR_NULL is evil for this very reason and should be avoided
where possible.  There were discussions a while back about removing it,
or at least deprecating it because it causes more bugs (exactly of this
type) than it solves.

Thanks for pointing it out.  I will redesign IS_ERR_OR_NULL usage.

--
FTTC broadband for 0.8mile line: now at 9.7Mbps down 460kbps up... slowly
improving, and getting towards what was expected from it.