Skip to contentSkip to conversation

ath10k: request firmware flush in ath10k_flush.

PATCH In review [email protected] 3 participants 7 messages Sep 19, 2014 – Sep 29, 2014 Qualcomm Atheros ATH10K wireless driver

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ath10k: request firmware flush in ath10k_flush.

From: Ben Greear <[email protected]>

CT firmware now supports flushing all tids for all peers for all vdevs. This appears to help the ath10k_flush logic work faster and not cause timeouts.

---

This requires the previously-sent patch that enables the ATH10K_FW_FEATURE_WMI_10X_CT feature bit. This patch provide a nice improvement in stability and flush performance in my testing scenarios.

 drivers/net/wireless/ath/ath10k/mac.c | 7 +++++++
 1 file changed, 7 insertions(+)
drivers/net/wireless/ath/ath10k/mac.c +7 -0
index af24f6c..6014856 100644
@@ -3830,6 +3830,7 @@ static void ath10k_flush(struct ieee80211_hw *hw, struct ieee80211_vif *vif,
 	struct ath10k *ar = hw->priv;
 	bool skip;
 	int ret;
+	u8 peer_addr[ETH_ALEN] = {0};
 
 	/* mac80211 doesn't care if we really xmit queued frames or not
 	 * we'll collect those frames either way if we stop/delete vdevs */
@@ -3841,6 +3842,12 @@ static void ath10k_flush(struct ieee80211_hw *hw, struct ieee80211_vif *vif,
 	if (ar->state == ATH10K_STATE_WEDGED)
 		goto skip;
 
+	/* If we are CT firmware, ask it to flush all tids on all peers on
+	 * all vdevs.  Normal firmware will just crash if you do this.
+	 */
+	if (test_bit(ATH10K_FW_FEATURE_WMI_10X_CT, ar->fw_features))
+		ath10k_wmi_peer_flush(ar, 0xFFFFFFFF, peer_addr, 0xFFFFFFFF);
+
 	ret = wait_event_timeout(ar->htt.empty_tx_wq, ({
 			bool empty;
 
Signature · 1 line
1.7.11.7
#1 lore raw

On 19 September 2014 20:28, <[email protected]> wrote:
[...]

drivers/net/wireless/ath/ath10k/mac.c static void ath10k_flush(struct ieee80211_hw *hw, struct ieee80211_vif *vif, · line 3849

Quoted 5 lines
drivers/net/wireless/ath/ath10k/mac.c +5 -0
+       /* If we are CT firmware, ask it to flush all tids on all peers on
+        * all vdevs.  Normal firmware will just crash if you do this.
+        */
+       if (test_bit(ATH10K_FW_FEATURE_WMI_10X_CT, ar->fw_features))
+               ath10k_wmi_peer_flush(ar, 0xFFFFFFFF, peer_addr, 0xFFFFFFFF);

I recall you've explained this some time ago, but can you refresh my memory, please? Is this any different from iterating over all peers and flushing each? Or does your firmware do so extra magic that is impossible to do with normal firmware commands?

Michał

author
#2 lore raw
Quoted 8 lines · Michal Kazior

On 19 September 2014 20:28, <[email protected]> wrote:
[...]

drivers/net/wireless/ath/ath10k/mac.c static void ath10k_flush(struct ieee80211_hw *hw, struct ieee80211_vif *vif, · line 3849

Quoted 5 lines
drivers/net/wireless/ath/ath10k/mac.c +5 -0
+       /* If we are CT firmware, ask it to flush all tids on all peers on
+        * all vdevs.  Normal firmware will just crash if you do this.
+        */
+       if (test_bit(ATH10K_FW_FEATURE_WMI_10X_CT, ar->fw_features))
+               ath10k_wmi_peer_flush(ar, 0xFFFFFFFF, peer_addr, 0xFFFFFFFF);
Quoted 4 lines

I recall you've explained this some time ago, but can you refresh my memory, please? Is this any different from iterating over all peers and flushing each? Or does your firmware do so extra magic that is impossible to do with normal firmware commands?

My firmware does that iteration internally.

You could probably do that in the driver, but it would be a lot of messages (for all vdevs, all peers, all tids)...
I was not sure if there were limits to the number
of commands you should attempt during the flush...

Thanks,
Ben

Signature · 2 lines
Ben Greear <[email protected]>
Candela Technologies Inc  http://www.candelatech.com
#3 lore raw
Quoted 17 lines · Ben Greear
Quoted 11 lines · Michal Kazior

On 19 September 2014 20:28, <[email protected]> wrote:
[...]

Quoted 8 lines

+ /* If we are CT firmware, ask it to flush all tids on all peers
on

drivers/net/wireless/ath/ath10k/mac.c +4 -0
+        * all vdevs.  Normal firmware will just crash if you do this.
+        */
+       if (test_bit(ATH10K_FW_FEATURE_WMI_10X_CT, ar->fw_features))
+               ath10k_wmi_peer_flush(ar, 0xFFFFFFFF, peer_addr,

0xFFFFFFFF);

Quoted 4 lines

I recall you've explained this some time ago, but can you refresh my memory, please? Is this any different from iterating over all peers and flushing each? Or does your firmware do so extra magic that is impossible to do with normal firmware commands?

Quoted 6 lines

My firmware does that iteration internally.

You could probably do that in the driver, but it would be a lot of messages (for all vdevs, all peers, all tids)...
I was not sure if there were limits to the number
of commands you should attempt during the flush...

Thanks. I think ath10k should do this instead of having CT-specific flush eventually.

I recall I actually did per-peer flushing in RFC patches a long time ago but I didn't pursue getting them merged. I think Denton uses these patches in his setup. They don't flush vdev self-peers though (do you do that internally?).

Michał

author
#4 lore raw
Quoted 20 lines · Michal Kazior
Quoted 19 lines · Ben Greear
Quoted 11 lines · Michal Kazior

On 19 September 2014 20:28, <[email protected]> wrote:
[...]

Quoted 8 lines

+ /* If we are CT firmware, ask it to flush all tids on all peers
on

drivers/net/wireless/ath/ath10k/mac.c +4 -0
+        * all vdevs.  Normal firmware will just crash if you do this.
+        */
+       if (test_bit(ATH10K_FW_FEATURE_WMI_10X_CT, ar->fw_features))
+               ath10k_wmi_peer_flush(ar, 0xFFFFFFFF, peer_addr,

0xFFFFFFFF);

Quoted 4 lines

I recall you've explained this some time ago, but can you refresh my memory, please? Is this any different from iterating over all peers and flushing each? Or does your firmware do so extra magic that is impossible to do with normal firmware commands?

My firmware does that iteration internally.

Quoted 7 lines
Quoted 4 lines

You could probably do that in the driver, but it would be a lot of messages (for all vdevs, all peers, all tids)...
I was not sure if there were limits to the number
of commands you should attempt during the flush...

Thanks. I think ath10k should do this instead of having CT-specific flush eventually.

Yes, though I would still like the optimization enabled if my firmware is running...

Quoted 4 lines

I recall I actually did per-peer flushing in RFC patches a long time ago but I didn't pursue getting them merged. I think Denton uses these patches in his setup. They don't flush vdev self-peers though (do you do that internally?).

I think I do. Let me check, and send you some details privately.

Thanks,
Ben

Signature · 2 lines
Ben Greear <[email protected]>
Candela Technologies Inc  http://www.candelatech.com
#5 lore raw
In reply to Michal Kazior
Quoted 20 lines · Michal Kazior
Quoted 19 lines · Ben Greear
Quoted 11 lines · Michal Kazior

On 19 September 2014 20:28, <[email protected]> wrote:
[...]

Quoted 8 lines

+ /* If we are CT firmware, ask it to flush all tids on all peers
on

drivers/net/wireless/ath/ath10k/mac.c +4 -0
+        * all vdevs.  Normal firmware will just crash if you do this.
+        */
+       if (test_bit(ATH10K_FW_FEATURE_WMI_10X_CT, ar->fw_features))
+               ath10k_wmi_peer_flush(ar, 0xFFFFFFFF, peer_addr,

0xFFFFFFFF);

Quoted 4 lines

I recall you've explained this some time ago, but can you refresh my memory, please? Is this any different from iterating over all peers and flushing each? Or does your firmware do so extra magic that is impossible to do with normal firmware commands?

My firmware does that iteration internally.

Quoted 7 lines
Quoted 4 lines

You could probably do that in the driver, but it would be a lot of messages (for all vdevs, all peers, all tids)...
I was not sure if there were limits to the number
of commands you should attempt during the flush...

Thanks. I think ath10k should do this instead of having CT-specific flush eventually.

I agree. We should not be forking functionality unless absolutely necessary.

Signature · 1 line
Kalle Valo
author
#6 lore raw
Quoted 5 lines · Kalle Valo
Quoted 2 lines

Thanks. I think ath10k should do this instead of having CT-specific flush eventually.

I agree. We should not be forking functionality unless absolutely necessary.

Ok, this patch is easy to carry in my own tree.

But, for things upstream firmware just cannot support, like tx-rate-status and rx-software-crypt, how do you want it flagged in the driver? Can I get the CT firmware flag in for that, or does it have to be some other way?

If you just will not accept any such patch, let me know so I can quite bothering you about it.

Thanks,
Ben

Signature · 2 lines
Ben Greear <[email protected]>
Candela Technologies Inc  http://www.candelatech.com