rt2x00usb: fix indexes ordering on RX queue kick
authorStanislaw Gruszka <sgruszka@redhat.com>
Wed, 4 Jul 2012 11:10:02 +0000 (13:10 +0200)
committerMarc Dietrich <marvin24@gmx.de>
Sun, 27 Oct 2013 13:15:08 +0000 (14:15 +0100)
commitaddf3915930a721f00fe6272c704b9f75fc65933
tree4ab81af59540c8aaa2a8a43786be79ede645416c
parent266d1cfeedbdb51b8ac5f5b5aa697df7d5ec159e
rt2x00usb: fix indexes ordering on RX queue kick

commit efd821182cec8c92babef6e00a95066d3252fda4 upstream.

On rt2x00_dmastart() we increase index specified by Q_INDEX and on
rt2x00_dmadone() we increase index specified by Q_INDEX_DONE. So entries
between Q_INDEX_DONE and Q_INDEX are those we currently process in the
hardware. Entries between Q_INDEX and Q_INDEX_DONE are those we can
submit to the hardware.

According to that fix rt2x00usb_kick_queue(), as we need to submit RX
entries that are not processed by the hardware. It worked before only
for empty queue, otherwise was broken.

Note that for TX queues indexes ordering are ok. We need to kick entries
that have filled skb, but was not submitted to the hardware, i.e.
started from Q_INDEX_DONE and have ENTRY_DATA_PENDING bit set.

From practical standpoint this fixes RX queue stall, usually reproducible
in AP mode, like for example reported here:
https://bugzilla.redhat.com/show_bug.cgi?id=828824

Reported-and-tested-by: Franco Miceli <fmiceli@plan.ceibal.edu.uy>
Reported-and-tested-by: Tom Horsley <horsley1953@gmail.com>
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
drivers/net/wireless/rt2x00/rt2x00usb.c