On 07/07/2025 22:01, Simon Horman wrote:
On Mon, Jul 07, 2025 at 11:09:49AM +0200, Thomas Fourier wrote:
quoted
The DMA map functions can fail and should be tested for errors.
If the mapping fails, unmap and return an error.
Fixes: 38df6492eb51 ("et131x: Add PCIe gigabit ethernet driver et131x to drivers/net")
Signed-off-by: Thomas Fourier <redacted>
nits:
1) There are two spaces after "et131x:" in the subject.
One is enough.
2) I think you can drop "ethernet: " from the subject.
"et131x: " seems to be an appropriate prefix based on git history.
...
Ok, I'll make sure to fix that.
quoted
@@ -2578,6 +2593,28 @@ static int nic_send_packet(struct et131x_adapter *adapter, struct tcb *tcb)
&adapter->regs->global.watchdog_timer);
}
return 0;
+
+unmap_out:
+ // Unmap everything from i-1 to 1
+ while (--i) {
+ frag--;
+ dma_addr = desc[frag].addr_lo;
+ dma_addr |= (u64)desc[frag].addr_hi << 32;
+ dma_unmap_page(&adapter->pdev->dev, dma_addr,
+ desc[frag].len_vlan, DMA_TO_DEVICE);
+ }
I'm probably missing something obvious. But it seems to me that frag is
incremented iff a mapping is successful. So I think only the loop below is
needed.
Yes, frag is incremented only after a successful mapping.
The first loop is to unmap the body of the packet (i >= 1) which is mapped
with skb_frag_dma_map() (and unmapped with dma_unmap_page). The
second is to unmap the header which is either one or two mappings of
map_single. Should I make the comments more explicit?
quoted
+
+unmap_first_out:
+ // unmap header
+ while (frag--) {
+ frag--;
I don't think you want to decrement frag twice here.
quoted
+ dma_addr = desc[frag].addr_lo;
+ dma_addr |= (u64)desc[frag].addr_hi << 32;
+ dma_unmap_single(&adapter->pdev->dev, dma_addr,
+ desc[frag].len_vlan, DMA_TO_DEVICE);
+ }
+
+ return -ENOMEM;
}
static int send_packet(struct sk_buff *skb, struct et131x_adapter *adapter)