Skip to contentSkip to conversation

mt76: set dma-done flag for flushed descriptors

PATCH In review Felix Fietkau 3 participants 9 messages Jan 27, 2020 – Feb 20, 2020

Subsystems (2)
lore
Share & tools
lore

Felix Fietkau

author maintainer
#0 lore raw

PATCH 1/5 mt76: set dma-done flag for flushed descriptors

Avoids a theoretical corner case where the hardware could try to process a stale descriptor after a watchdog reset

---
 drivers/net/wireless/mediatek/mt76/dma.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/wireless/mediatek/mt76/dma.c +3 -1
index 9e03a7871ff4..e69329feed78 100644
@@ -246,7 +246,9 @@ mt76_dma_dequeue(struct mt76_dev *dev, struct mt76_queue *q, bool flush,
 	if (!q->queued)
 		return NULL;
 
-	if (!flush && !(q->desc[idx].ctrl & cpu_to_le32(MT_DMA_CTL_DMA_DONE)))
+	if (flush)
+		q->desc[idx].ctrl |= cpu_to_le32(MT_DMA_CTL_DMA_DONE);
+	else if (!(q->desc[idx].ctrl & cpu_to_le32(MT_DMA_CTL_DMA_DONE)))
 		return NULL;
 
 	q->tail = (q->tail + 1) % q->ndesc;
Signature · 1 line
2.24.0
author maintainer
#1 lore raw

PATCH 2/5 mt76: fix handling full tx queues in mt76_dma_tx_queue_skb_raw

Fixes a theoretical issue where it could potentially overwrite an existing descriptor entry (and leaking its skb)

---
 drivers/net/wireless/mediatek/mt76/dma.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
drivers/net/wireless/mediatek/mt76/dma.c +8 -1
index e69329feed78..bcb11bb9aeeb 100644
@@ -271,10 +271,13 @@ mt76_dma_tx_queue_skb_raw(struct mt76_dev *dev, enum mt76_txq_id qid,
 	struct mt76_queue_buf buf;
 	dma_addr_t addr;
 
+	if (q->queued + 1 >= q->ndesc - 1)
+		goto error;
+
 	addr = dma_map_single(dev->dev, skb->data, skb->len,
 			      DMA_TO_DEVICE);
 	if (unlikely(dma_mapping_error(dev->dev, addr)))
-		return -ENOMEM;
+		goto error;
 
 	buf.addr = addr;
 	buf.len = skb->len;
@@ -285,6 +288,10 @@ mt76_dma_tx_queue_skb_raw(struct mt76_dev *dev, enum mt76_txq_id qid,
 	spin_unlock_bh(&q->lock);
 
 	return 0;
+
+error:
+	dev_kfree_skb(skb);
+	return -ENOMEM;
 }
 
 static int
Signature · 1 line
2.24.0
author maintainer
#2 lore raw

PATCH 3/5 mt76: dma: do not write cpu_idx on rx queue reset until after refill

In reply to Felix Fietkau

The hardware should only start processing the ring after at least one buffer has been added

---
 drivers/net/wireless/mediatek/mt76/dma.c | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)
drivers/net/wireless/mediatek/mt76/dma.c +8 -7
index bcb11bb9aeeb..e5dd7080e88e 100644
@@ -132,6 +132,11 @@ mt76_dma_sync_idx(struct mt76_dev *dev, struct mt76_queue *q)
 	writel(q->ndesc, &q->regs->ring_size);
 	q->head = readl(&q->regs->dma_idx);
 	q->tail = q->head;
+}
+
+static void
+mt76_dma_kick_queue(struct mt76_dev *dev, struct mt76_queue *q)
+{
 	writel(q->head, &q->regs->cpu_idx);
 }
 
@@ -193,8 +198,10 @@ mt76_dma_tx_cleanup(struct mt76_dev *dev, enum mt76_txq_id qid, bool flush)
 		dev->q_tx[__MT_TXQ_MAX + i].swq_queued -= n_swq_queued[4 + i];
 	}
 
-	if (flush)
+	if (flush) {
 		mt76_dma_sync_idx(dev, q);
+		mt76_dma_kick_queue(dev, q);
+	}
 
 	wake = wake && q->stopped &&
 	       qid < IEEE80211_NUM_ACS && q->queued < q->ndesc - 8;
@@ -257,12 +264,6 @@ mt76_dma_dequeue(struct mt76_dev *dev, struct mt76_queue *q, bool flush,
 	return mt76_dma_get_buf(dev, q, idx, len, info, more);
 }
 
-static void
-mt76_dma_kick_queue(struct mt76_dev *dev, struct mt76_queue *q)
-{
-	writel(q->head, &q->regs->cpu_idx);
-}
-
 static int
 mt76_dma_tx_queue_skb_raw(struct mt76_dev *dev, enum mt76_txq_id qid,
 			  struct sk_buff *skb, u32 tx_info)
Signature · 1 line
2.24.0
author maintainer
#3 lore raw

PATCH 4/5 mt76: mt7603: increase dma mcu rx ring size

In reply to Felix Fietkau

The ring is used for looping back tx powersave filtered frames, so it could use some more room, in case more than one aggregate was queued

---
 drivers/net/wireless/mediatek/mt76/mt7603/dma.c    | 2 +-
 drivers/net/wireless/mediatek/mt76/mt7603/mt7603.h | 1 +
 2 files changed, 2 insertions(+), 1 deletion(-)
drivers/net/wireless/mediatek/mt76/mt7603/dma.c +1 -1
index 57428467fe96..a08b85281170 100644
@@ -221,7 +221,7 @@ int mt7603_dma_init(struct mt7603_dev *dev)
 		return ret;
 
 	ret = mt7603_init_rx_queue(dev, &dev->mt76.q_rx[MT_RXQ_MCU], 1,
-				   MT_MCU_RING_SIZE, MT_RX_BUF_SIZE);
+				   MT7603_MCU_RX_RING_SIZE, MT_RX_BUF_SIZE);
 	if (ret)
 		return ret;
 
drivers/net/wireless/mediatek/mt76/mt7603/mt7603.h +1 -0
index 12fd2db2df10..ef374641fe80 100644
@@ -15,6 +15,7 @@
 
 #define MT7603_RATE_RETRY	2
 
+#define MT7603_MCU_RX_RING_SIZE	64
 #define MT7603_RX_RING_SIZE     128
 
 #define MT7603_FIRMWARE_E1	"mt7603_e1.bin"
Signature · 1 line
2.24.0
author maintainer
#4 lore raw

PATCH 5/5 mt76: enable Airtime Queue Limit support

In reply to Felix Fietkau

It is supported by all drivers

---
 drivers/net/wireless/mediatek/mt76/mac80211.c | 1 +
 1 file changed, 1 insertion(+)
drivers/net/wireless/mediatek/mt76/mac80211.c +1 -0
index 9947e627318b..2a85e0cb5793 100644
@@ -283,6 +283,7 @@ mt76_phy_init(struct mt76_dev *dev, struct ieee80211_hw *hw)
 
 	wiphy_ext_feature_set(wiphy, NL80211_EXT_FEATURE_CQM_RSSI_LIST);
 	wiphy_ext_feature_set(wiphy, NL80211_EXT_FEATURE_AIRTIME_FAIRNESS);
+	wiphy_ext_feature_set(wiphy, NL80211_EXT_FEATURE_AQL);
 
 	wiphy->available_antennas_tx = dev->phy.antenna_mask;
 	wiphy->available_antennas_rx = dev->phy.antenna_mask;
Signature · 1 line
2.24.0
#6 lore raw

Re: [PATCH 5/5] mt76: enable Airtime Queue Limit support

nope (iperf3 with 32 parallel streams after 5 seconds of running)

STA 30:24:32💿f9:c4 AC 2 txq pending airtime underflow: 4294967208, 88

Modules linked in: fast_classifier mt7615e mt76 mac80211 compat
CPU: 3 PID: 1890 Comm: hostapd Not tainted 4.14.171 #187

Stack : 00000000 00000001 00000040 8007e454 80490000 8048bc8c 80610000 805aee78 8058850c 8770b864 870f289c 805f07a7 80582204 00000001 8770b808 805f60a8 00000000 00000000 80650000 00000000 81147bb8 0000011e 00000007 00000000 00000000 80660000 80660000 68203a6d 80000000 80610000 00000000 873079ec 873a4654 00000775 00000000 00000001 00000000 80230b88 0000000c 8065000c
        ...

Call Trace:
[<800153a0>] show_stack+0x58/0x100
[<80438c54>] dump_stack+0xa4/0xe0
[<80035050>] __warn+0xe4/0x144
[<80034c8c>] warn_slowpath_fmt+0x30/0x3c
[<873079ec>] _616+0x110/0x214 [mac80211]
[<87301ce0>] _535+0x15c/0x4f8 [mac80211]
[<804224f0>] br_handle_frame_finish+0x4e0/0x524
[<873d5160>] _49+0x50/0xa4 [mt76]
[<87606e30>] _61+0x1e8/0x224 [mt7615e]

Am 01.02.2020 um 13:58 schrieb Toke Høiland-Jørgensen:

Quoted 10 lines
Quoted 3 lines · Felix Fietkau

It is supported by all drivers

Nice!

#7 lore raw

Re: [PATCH 5/5] mt76: enable Airtime Queue Limit support

Quoted 16 lines · Sebastian Gottschall

nope (iperf3 with 32 parallel streams after 5 seconds of running)

STA 30:24:32💿f9:c4 AC 2 txq pending airtime underflow: 4294967208, 88

Modules linked in: fast_classifier mt7615e mt76 mac80211 compat
CPU: 3 PID: 1890 Comm: hostapd Not tainted 4.14.171 #187

Stack : 00000000 00000001 00000040 8007e454 80490000 8048bc8c 80610000 805aee78
        8058850c 8770b864 870f289c 805f07a7 80582204 00000001 8770b808
805f60a8
        00000000 00000000 80650000 00000000 81147bb8 0000011e 00000007
00000000
        00000000 80660000 80660000 68203a6d 80000000 80610000 00000000
873079ec
        873a4654 00000775 00000000 00000001 00000000 80230b88 0000000c
8065000c
        ...

Quoted 10 lines
Call Trace:
[<800153a0>] show_stack+0x58/0x100
[<80438c54>] dump_stack+0xa4/0xe0
[<80035050>] __warn+0xe4/0x144
[<80034c8c>] warn_slowpath_fmt+0x30/0x3c
[<873079ec>] _616+0x110/0x214 [mac80211]
[<87301ce0>] _535+0x15c/0x4f8 [mac80211]
[<804224f0>] br_handle_frame_finish+0x4e0/0x524
[<873d5160>] _49+0x50/0xa4 [mt76]
[<87606e30>] _61+0x1e8/0x224 [mt7615e]

Huh, why is there a bridge function name in there in the middle of the callstack?

-Toke

#8 lore raw

Re: [PATCH 5/5] mt76: enable Airtime Queue Limit support

the interface is bridged to a bridge named br0 and owns the local lan ip.
its a typical setup. lan switch is bridged together with ap interface
iperf was not running on the device itself needs to be mentioned. i was running iperf -s on a device which was  connected by ethernet to this device. my client was a simple lenovo p52 laptop with a iwlwifi 9560 card wich was using iperf -c x.x.x.x -P 32 -i 1

Am 19.02.2020 um 23:23 schrieb Toke Høiland-Jørgensen:

Quoted 34 lines
Quoted 16 lines · Sebastian Gottschall

nope (iperf3 with 32 parallel streams after 5 seconds of running)

STA 30:24:32💿f9:c4 AC 2 txq pending airtime underflow: 4294967208, 88

Modules linked in: fast_classifier mt7615e mt76 mac80211 compat
CPU: 3 PID: 1890 Comm: hostapd Not tainted 4.14.171 #187

Stack : 00000000 00000001 00000040 8007e454 80490000 8048bc8c 80610000 805aee78
        8058850c 8770b864 870f289c 805f07a7 80582204 00000001 8770b808
805f60a8
        00000000 00000000 80650000 00000000 81147bb8 0000011e 00000007
00000000
        00000000 80660000 80660000 68203a6d 80000000 80610000 00000000
873079ec
        873a4654 00000775 00000000 00000001 00000000 80230b88 0000000c
8065000c
        ...

Quoted 10 lines
Call Trace:
[<800153a0>] show_stack+0x58/0x100
[<80438c54>] dump_stack+0xa4/0xe0
[<80035050>] __warn+0xe4/0x144
[<80034c8c>] warn_slowpath_fmt+0x30/0x3c
[<873079ec>] _616+0x110/0x214 [mac80211]
[<87301ce0>] _535+0x15c/0x4f8 [mac80211]
[<804224f0>] br_handle_frame_finish+0x4e0/0x524
[<873d5160>] _49+0x50/0xa4 [mt76]
[<87606e30>] _61+0x1e8/0x224 [mt7615e]

Huh, why is there a bridge function name in there in the middle of the callstack?

-Toke