Thread (11 messages) flat view 11 messages, 4 authors, 2017-09-20

Re: [PATCH 1/2] mwifiex: resubmit failed to submit RX URBs in main thread

From: Kalle Valo <hidden>
Date: 2017-09-20 12:13:10

Ganapathi Bhat [off-list ref] writes:
From: James Cao <redacted>

Current driver has 6 Rx data URBs. Once any packet received
kernel calls our callback, in which the same URB will be
resubmitted after Rx indication. In URB submission function a new
skb will be allocated since the previous one is passed to upper
layer (freed later). Since the skb is from a special pool (not
regular memory), skb allocation may fail when kernel holds a lot
of Rx packets on some low resource platforms.
The special pool being GFP_ATOMIC allocations or what?
The URB will not be resubmitted in this no free skb case. If driver
fails to resubmit all 6 URBs, Rx will stop. To cover this scenario
check and resubmit Rx URBs in main thread.

Signed-off-by: James Cao <redacted>
Signed-off-by: Cathy Luo <redacted>
Signed-off-by: Ganapathi Bhat <redacted>
[...]
quoted hunk ↗ jump to hunk
@@ -278,6 +279,16 @@ int mwifiex_main_process(struct mwifiex_adapter *adapter)
 			break;
 		}
 
+		/* Try to resubmit RX URB if sunmission failed earlier */
+		if (!atomic_read(&adapter->rx_pending) &&
+		    adapter->iface_type == MWIFIEX_USB) {
+			usb_card = adapter->card;
+			if (atomic_read(&usb_card->rx_data_urb_pending) <
+			    MWIFIEX_RX_DATA_URB &&
+			    adapter->if_ops.submit_rem_rx_urbs)
+				adapter->if_ops.submit_rem_rx_urbs(adapter);
+		}
To me this just feels wrong. Normally the proceduce is to drop the frame
if allocations fail, not try to reallocate. I need more convincing that
this really is the right approach.

-- 
Kalle Valo
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help