brcmsmac: Remove internal tx queue

The brcmsmac internal tx buffering is problematic. The amount of
buffering is excessive (228 packets in addition to the 256 slots in each
DMA ring), and frames may be dropped due to a lack of flow control.

This patch reworks the transmit code path to remove the internal
buffering. Frames are immediately handed off to the DMA support rather
than passing through an intermediate queue. Non-aggregate frames are
queued immediately into the tx rings, and aggregate frames are queued
temporarily in an AMPDU session until ready for transmit.

Transmit flow control is also added to avoid dropping packets when the
tx rings are full. Conceptually this is a separate change, but it's
included in this commit because removing the tx queue without adding
flow control could cause significant problems.

Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
Reviewed-by: Pieter-Paul Giesberts <pieterpg@broadcom.com>
Reviewed-by: Arend van Spriel <arend@broadcom.com>
Tested-by: Daniel Wagner <wagi@monom.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

1 files changed, 0 insertions, 1 deletions

diff --git a/drivers/net/wireless/brcm80211/brcmsmac/types.h b/drivers/net/wireless/brcm80211/brcmsmac/types.h
index e11ae83111e4..e3abc0ed502e 100644
--- a/drivers/net/wireless/brcm80211/brcmsmac/types.h
+++ b/drivers/net/wireless/brcm80211/brcmsmac/types.h
@@ -281,7 +281,6 @@ struct ieee80211_tx_queue_params;
 struct brcms_info;
 struct brcms_c_info;
 struct brcms_hardware;
-struct brcms_txq_info;
 struct brcms_band;
 struct dma_pub;
 struct si_pub;