Thread (29 messages) read the whole thread 29 messages, 4 authors, 2012-01-24

Re: Compat-wireless-3.2-rc6-3 is broken for rt2860 device

From: Andreas Hartmann <hidden>
Date: 2012-01-10 14:44:39

Am Tue, 10 Jan 2012 14:43:21 +0100
schrieb Helmut Schaa [off-list ref]:
quoted hunk ↗ jump to hunk
Hi,

On Tue, Jan 10, 2012 at 2:01 PM, Andreas Hartmann
[off-list ref] wrote:
quoted
Helmut Schaa schrieb:
[...]
quoted
So this looks good to me although the broken tx status for BAR frames
is for sure a rt2800pci problem (or even a hw issue).
Sorry - What do you mean with "hw issue"? Is my hw broken?
Hehe, no :)
quoted
Or is it a specific "behavior" of exactly this device:
I think most rt2800pci will behave the same and just interpret a
BlockAck as response to a BAR in a different manner as a BlockAck
in response to a Data frame.

When we send the BAR we tell the hw that the frame needs to be
acknowledged and according to the spec the hw will only count
a normal ACK as successful transmission, the peer will however
answer with a BlockAck as long as a BA session is active.

It might be possible to advise the hw to allow the BAR to be acked
by a BlockAck by specifying AMDPU=1 in the TXWI. Not sure
if that breaks anything else.

However, a BAR seems to be "ack-able" by both an ACK and a
BlockAck according to 802.11-2007. If the recipient answers with
an ACK the BA session might already be terminated and we would
stop the BA session as soon as 3 BARs failed with the previous
patch. This should be fine I guess.

So you might just give the following a try (also fully untested and
maybe with some of your debugging output enabled).

Signed-off-by: Helmut Schaa <redacted>
---
diff --git a/drivers/net/wireless/rt2x00/rt2x00queue.c
b/drivers/net/wireless/rt2x00/rt2x00queue.c
index e4641da..5d0a3d3 100644
--- a/drivers/net/wireless/rt2x00/rt2x00queue.c
+++ b/drivers/net/wireless/rt2x00/rt2x00queue.c
@@ -393,6 +393,9 @@ static void
rt2x00queue_create_tx_descriptor_ht(struct rt2x00_dev *rt2x00dev,
 		txdesc->u.ht.txop = TXOP_SIFS;
 	else
 		txdesc->u.ht.txop = TXOP_HTTXOP;
+
+	if (ieee80211_is_back_req(hdr->frame_control))
+		__set_bit(ENTRY_TXD_HT_AMPDU, &txdesc->flags);
 }

 static void rt2x00queue_create_tx_descriptor(struct rt2x00_dev *rt2x00dev,
Unfortunately, I don't think it fixes anything (an equal looking
sequence could be seen just 10s before - ok other ssn numbers :-)):

[38517.363964] rt2x00lib_txdone - no success - low_level_stats.dot11ACKFailureCount: 19
[38517.363977] __set_bit ENTRY_TXD_HT_AMPDU
[38517.363989] rt2x00lib_txdone - no success - low_level_stats.dot11ACKFailureCount: 20
[38517.363995] __set_bit ENTRY_TXD_HT_AMPDU
[38517.364005] rt2x00lib_txdone - no success - low_level_stats.dot11ACKFailureCount: 21
[38517.364011] __set_bit ENTRY_TXD_HT_AMPDU
[38517.364020] rt2x00lib_txdone - no success - low_level_stats.dot11ACKFailureCount: 22
[38517.364026] __set_bit ENTRY_TXD_HT_AMPDU
[38517.364034] rt2x00lib_txdone - no success - low_level_stats.dot11ACKFailureCount: 23
[38517.364042] __set_bit ENTRY_TXD_HT_AMPDU
[38517.375021] rt2x00lib_txdone - no success - low_level_stats.dot11ACKFailureCount: 24
[38517.375028] status.c ieee80211_tx_status() calls ieee80211_set_bar_pending. tid: <0> control: <4> ssn: <27248>
[38517.389129] rt2x00lib_txdone - no success - low_level_stats.dot11ACKFailureCount: 25
[38517.389132] status.c ieee80211_tx_status() calls ieee80211_set_bar_pending. tid: <0> control: <4> ssn: <27264>
[38517.397279] rt2x00lib_txdone - no success - low_level_stats.dot11ACKFailureCount: 26
[38517.397283] status.c ieee80211_tx_status() calls ieee80211_set_bar_pending. tid: <0> control: <4> ssn: <27280>
[38517.407062] rt2x00lib_txdone - no success - low_level_stats.dot11ACKFailureCount: 27
[38517.407069] status.c ieee80211_tx_status() calls ieee80211_set_bar_pending. tid: <0> control: <4> ssn: <27296>
[38517.417684] rt2x00lib_txdone - no success - low_level_stats.dot11ACKFailureCount: 28
[38517.417691] status.c ieee80211_tx_status() calls ieee80211_set_bar_pending. tid: <0> control: <4> ssn: <27312>
[38517.417742] ieee80211_check_pending_bar -> ieee80211_send_bar() retry cnt: 2 tid: 0 failed_bar_ssn: 27312
[38517.417751] __set_bit ENTRY_TXD_HT_AMPDU
[38517.417763] ieee80211_check_pending_bar -> ieee80211_send_bar() retry cnt: 1 tid: 0 failed_bar_ssn: 27312
[38517.417768] __set_bit ENTRY_TXD_HT_AMPDU
[38517.417775] ieee80211_check_pending_bar -> ieee80211_stop_tx_ba_session() retry cnt: 0 tid: 0 failed_bar_ssn: 27312
[38517.417840] Tx BA session stop requested for 00:25:9c:11:22:33 tid 0
[38517.426147] Stopping Tx BA session for 00:25:9c:11:22:33 tid 0
[38517.432511] rt2x00lib_txdone - no success - low_level_stats.dot11ACKFailureCount: 29
[38517.432518] status.c ieee80211_tx_status() calls ieee80211_set_bar_pending. tid: <0> control: <4> ssn: <27312>
[38517.439356] rt2x00lib_txdone - no success - low_level_stats.dot11ACKFailureCount: 30
[38517.439362] status.c ieee80211_tx_status() calls ieee80211_set_bar_pending. tid: <0> control: <4> ssn: <27312>
[38517.491026] Open BA session requested for 00:25:9c:11:22:33 tid 0
[38517.499048] activated addBA response timer on tid 0
[38517.507633] switched off addBA timer for tid 0
[38517.507637] Aggregation is on for tid 0


Some additional information might be interesting: 
I'm getting these tx errors nearly always (I haven't seen them in the
other direction until now) while sending data from STA -> AP (not from
AP -> STA). The throughput from STA -> AP is mostly about 8 MBit/s
whereas the other direction reaches up to 15 MBit/s.



Kind regards,
Andreas
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help