* [PATCH net 1/2] net: stmmac: Track the TX DMA threshold per queue
2026-10-13 11:45 [PATCH net 0/2] net: stmmac: Track and report the TX DMA threshold per queue Ovidiu Panait
@ 2026-10-13 11:45 ` Ovidiu Panait
2026-10-13 12:45 ` Maxime Chevallier
2026-10-13 11:45 ` [PATCH net 2/2] net: stmmac: Report " Ovidiu Panait
2026-10-13 12:44 ` [PATCH net 0/2] net: stmmac: Track and report " Maxime Chevallier
2 siblings, 1 reply; 5+ messages in thread
From: Ovidiu Panait @ 2026-10-13 11:45 UTC (permalink / raw)
To: maxime.chevallier, andrew+netdev, davem, edumazet, kuba, pabeni,
mcoquelin.stm32, alexandre.torgue, xiaoliang.yang_1
Cc: netdev, linux-stm32, linux-arm-kernel, linux-kernel,
Ovidiu Panait
Commit 3a6c12a0c6c3 ("net: stmmac: bump tc when get underflow error from
DMA descriptor") made dwmac4 raise the DMA threshold of the queues that
report TX underflows. This is done by incrementing a module global
parameter 'tc' by 64 on each underflow.
This leads to inconsistent TX DMA thresholds between queues and causes
issues on platforms with multiple stmmac controllers, as 'tc' is shared
by all of them. Once 'tc' goes over 256 the bumping stops, and queues
that did not underflow before are left at the initial threshold.
On the Renesas RZ/V2H platform, which has two dwmac4 controllers, we
observed sporadic iperf3 TCP TX failures:
iperf3 -c <server> -t5 --bind-dev end1
...
[ 5] 0.00-5.00 sec 1.00 MBytes 1.68 Mbits/sec 87 sender
[ 5] 0.00-5.00 sec 768 KBytes 1.26 Mbits/sec receiver
This happens because the TX queues of end1 were underflowing on a large
fraction of the packets, but the DMA TX threshold was never raised from
the default of 64 bytes. The threshold was not raised because end0
had underflowed first and had already pushed 'tc' past 256. Which
controller is affected depends on which one underflows first.
To fix this, make the TX DMA threshold per queue instead of global.
Each time an underflow happens on a queue, its TX DMA threshold gets
incremented, without affecting other queues or controllers. With this
change, 'tc' is only used as the initial value, and the RX threshold
is no longer bumped together with the TX one, as TX underflows have
nothing to do with the RX side.
Fixes: 3a6c12a0c6c3 ("net: stmmac: bump tc when get underflow error from DMA descriptor")
Signed-off-by: Ovidiu Panait <ovidiu.panait.rb@renesas.com>
---
drivers/net/ethernet/stmicro/stmmac/stmmac.h | 1 +
.../net/ethernet/stmicro/stmmac/stmmac_main.c | 33 +++++++++++--------
2 files changed, 21 insertions(+), 13 deletions(-)
diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac.h b/drivers/net/ethernet/stmicro/stmmac/stmmac.h
index 9278378407e8..85d63450693e 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac.h
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac.h
@@ -270,6 +270,7 @@ struct stmmac_priv {
bool sph_capable;
u32 sarc_type;
u32 rx_riwt[MTL_MAX_RX_QUEUES];
+ u32 tx_threshold[MTL_MAX_TX_QUEUES];
int hwts_rx_en;
bool tsfupdt_coarse;
diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
index b8b4de4c5b1a..b5ee0f297df6 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
@@ -2622,7 +2622,6 @@ static void stmmac_dma_operation_mode(struct stmmac_priv *priv)
}
if (priv->plat->force_thresh_dma_mode) {
- txmode = tc;
rxmode = tc;
} else if (priv->plat->force_sf_dma_mode || priv->plat->tx_coe) {
/*
@@ -2636,7 +2635,6 @@ static void stmmac_dma_operation_mode(struct stmmac_priv *priv)
rxmode = SF_DMA_MODE;
priv->xstats.threshold = SF_DMA_MODE;
} else {
- txmode = tc;
rxmode = SF_DMA_MODE;
}
@@ -2655,8 +2653,11 @@ static void stmmac_dma_operation_mode(struct stmmac_priv *priv)
for (chan = 0; chan < tx_channels_count; chan++) {
qmode = priv->plat->tx_queues_cfg[chan].mode_to_use;
- stmmac_dma_tx_mode(priv, priv->ioaddr, txmode, chan,
- txfifosz, qmode);
+ if (txmode == SF_DMA_MODE)
+ priv->tx_threshold[chan] = SF_DMA_MODE;
+
+ stmmac_dma_tx_mode(priv, priv->ioaddr, priv->tx_threshold[chan],
+ chan, txfifosz, qmode);
}
}
@@ -2822,17 +2823,20 @@ static bool stmmac_xdp_xmit_zc(struct stmmac_priv *priv, u32 queue, u32 budget)
static void stmmac_bump_dma_threshold(struct stmmac_priv *priv, u32 chan)
{
- if (unlikely(priv->xstats.threshold != SF_DMA_MODE) && tc <= 256) {
- tc += 64;
+ u32 thresh = priv->tx_threshold[chan];
- if (priv->plat->force_thresh_dma_mode)
- stmmac_set_dma_operation_mode(priv, tc, tc, chan);
- else
- stmmac_set_dma_operation_mode(priv, tc, SF_DMA_MODE,
- chan);
+ if (unlikely(thresh == SF_DMA_MODE) || thresh > 256)
+ return;
- priv->xstats.threshold = tc;
- }
+ thresh += 64;
+
+ if (priv->plat->force_thresh_dma_mode)
+ stmmac_set_dma_operation_mode(priv, thresh, tc, chan);
+ else
+ stmmac_set_dma_operation_mode(priv, thresh, SF_DMA_MODE, chan);
+
+ priv->tx_threshold[chan] = thresh;
+ priv->xstats.threshold = thresh;
}
/**
@@ -8133,6 +8137,9 @@ static int __stmmac_dvr_probe(struct device *device,
priv->xstats.threshold = tc;
+ for (i = 0; i < MTL_MAX_TX_QUEUES; i++)
+ priv->tx_threshold[i] = tc;
+
/* Initialize RSS */
rxq = priv->plat->rx_queues_to_use;
netdev_rss_key_fill(priv->rss.key, sizeof(priv->rss.key));
--
2.34.1
^ permalink raw reply related [flat|nested] 5+ messages in thread* Re: [PATCH net 1/2] net: stmmac: Track the TX DMA threshold per queue
2026-10-13 11:45 ` [PATCH net 1/2] net: stmmac: Track " Ovidiu Panait
@ 2026-10-13 12:45 ` Maxime Chevallier
0 siblings, 0 replies; 5+ messages in thread
From: Maxime Chevallier @ 2026-10-13 12:45 UTC (permalink / raw)
To: Ovidiu Panait, andrew+netdev, davem, edumazet, kuba, pabeni,
mcoquelin.stm32, alexandre.torgue, xiaoliang.yang_1
Cc: netdev, linux-stm32, linux-arm-kernel, linux-kernel
Hi,
On 10/13/26 13:45, Ovidiu Panait wrote:
> Commit 3a6c12a0c6c3 ("net: stmmac: bump tc when get underflow error from
> DMA descriptor") made dwmac4 raise the DMA threshold of the queues that
> report TX underflows. This is done by incrementing a module global
> parameter 'tc' by 64 on each underflow.
>
> This leads to inconsistent TX DMA thresholds between queues and causes
> issues on platforms with multiple stmmac controllers, as 'tc' is shared
> by all of them. Once 'tc' goes over 256 the bumping stops, and queues
> that did not underflow before are left at the initial threshold.
>
> On the Renesas RZ/V2H platform, which has two dwmac4 controllers, we
> observed sporadic iperf3 TCP TX failures:
>
> iperf3 -c <server> -t5 --bind-dev end1
> ...
> [ 5] 0.00-5.00 sec 1.00 MBytes 1.68 Mbits/sec 87 sender
> [ 5] 0.00-5.00 sec 768 KBytes 1.26 Mbits/sec receiver
>
> This happens because the TX queues of end1 were underflowing on a large
> fraction of the packets, but the DMA TX threshold was never raised from
> the default of 64 bytes. The threshold was not raised because end0
> had underflowed first and had already pushed 'tc' past 256. Which
> controller is affected depends on which one underflows first.
>
> To fix this, make the TX DMA threshold per queue instead of global.
> Each time an underflow happens on a queue, its TX DMA threshold gets
> incremented, without affecting other queues or controllers. With this
> change, 'tc' is only used as the initial value, and the RX threshold
> is no longer bumped together with the TX one, as TX underflows have
> nothing to do with the RX side.
>
> Fixes: 3a6c12a0c6c3 ("net: stmmac: bump tc when get underflow error from DMA descriptor")
> Signed-off-by: Ovidiu Panait <ovidiu.panait.rb@renesas.com>
Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Maxime
> ---
> drivers/net/ethernet/stmicro/stmmac/stmmac.h | 1 +
> .../net/ethernet/stmicro/stmmac/stmmac_main.c | 33 +++++++++++--------
> 2 files changed, 21 insertions(+), 13 deletions(-)
>
> diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac.h b/drivers/net/ethernet/stmicro/stmmac/stmmac.h
> index 9278378407e8..85d63450693e 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/stmmac.h
> +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac.h
> @@ -270,6 +270,7 @@ struct stmmac_priv {
> bool sph_capable;
> u32 sarc_type;
> u32 rx_riwt[MTL_MAX_RX_QUEUES];
> + u32 tx_threshold[MTL_MAX_TX_QUEUES];
> int hwts_rx_en;
> bool tsfupdt_coarse;
>
> diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> index b8b4de4c5b1a..b5ee0f297df6 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> @@ -2622,7 +2622,6 @@ static void stmmac_dma_operation_mode(struct stmmac_priv *priv)
> }
>
> if (priv->plat->force_thresh_dma_mode) {
> - txmode = tc;
> rxmode = tc;
> } else if (priv->plat->force_sf_dma_mode || priv->plat->tx_coe) {
> /*
> @@ -2636,7 +2635,6 @@ static void stmmac_dma_operation_mode(struct stmmac_priv *priv)
> rxmode = SF_DMA_MODE;
> priv->xstats.threshold = SF_DMA_MODE;
> } else {
> - txmode = tc;
> rxmode = SF_DMA_MODE;
> }
>
> @@ -2655,8 +2653,11 @@ static void stmmac_dma_operation_mode(struct stmmac_priv *priv)
> for (chan = 0; chan < tx_channels_count; chan++) {
> qmode = priv->plat->tx_queues_cfg[chan].mode_to_use;
>
> - stmmac_dma_tx_mode(priv, priv->ioaddr, txmode, chan,
> - txfifosz, qmode);
> + if (txmode == SF_DMA_MODE)
> + priv->tx_threshold[chan] = SF_DMA_MODE;
> +
> + stmmac_dma_tx_mode(priv, priv->ioaddr, priv->tx_threshold[chan],
> + chan, txfifosz, qmode);
> }
> }
>
> @@ -2822,17 +2823,20 @@ static bool stmmac_xdp_xmit_zc(struct stmmac_priv *priv, u32 queue, u32 budget)
>
> static void stmmac_bump_dma_threshold(struct stmmac_priv *priv, u32 chan)
> {
> - if (unlikely(priv->xstats.threshold != SF_DMA_MODE) && tc <= 256) {
> - tc += 64;
> + u32 thresh = priv->tx_threshold[chan];
>
> - if (priv->plat->force_thresh_dma_mode)
> - stmmac_set_dma_operation_mode(priv, tc, tc, chan);
> - else
> - stmmac_set_dma_operation_mode(priv, tc, SF_DMA_MODE,
> - chan);
> + if (unlikely(thresh == SF_DMA_MODE) || thresh > 256)
> + return;
>
> - priv->xstats.threshold = tc;
> - }
> + thresh += 64;
> +
> + if (priv->plat->force_thresh_dma_mode)
> + stmmac_set_dma_operation_mode(priv, thresh, tc, chan);
> + else
> + stmmac_set_dma_operation_mode(priv, thresh, SF_DMA_MODE, chan);
> +
> + priv->tx_threshold[chan] = thresh;
> + priv->xstats.threshold = thresh;
> }
>
> /**
> @@ -8133,6 +8137,9 @@ static int __stmmac_dvr_probe(struct device *device,
>
> priv->xstats.threshold = tc;
>
> + for (i = 0; i < MTL_MAX_TX_QUEUES; i++)
> + priv->tx_threshold[i] = tc;
> +
> /* Initialize RSS */
> rxq = priv->plat->rx_queues_to_use;
> netdev_rss_key_fill(priv->rss.key, sizeof(priv->rss.key));
^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH net 2/2] net: stmmac: Report the TX DMA threshold per queue
2026-10-13 11:45 [PATCH net 0/2] net: stmmac: Track and report the TX DMA threshold per queue Ovidiu Panait
2026-10-13 11:45 ` [PATCH net 1/2] net: stmmac: Track " Ovidiu Panait
@ 2026-10-13 11:45 ` Ovidiu Panait
2026-10-13 12:44 ` [PATCH net 0/2] net: stmmac: Track and report " Maxime Chevallier
2 siblings, 0 replies; 5+ messages in thread
From: Ovidiu Panait @ 2026-10-13 11:45 UTC (permalink / raw)
To: maxime.chevallier, andrew+netdev, davem, edumazet, kuba, pabeni,
mcoquelin.stm32, alexandre.torgue, xiaoliang.yang_1
Cc: netdev, linux-stm32, linux-arm-kernel, linux-kernel,
Ovidiu Panait
The "threshold" ethtool statistic reports a single TX DMA threshold for
the whole device. However, since commit d62a107a4f15 ("net: stmmac:
prepare dma interrupt treatment for multiple queues"), a single TX DMA
threshold per device doesn't make sense anymore, as each individual queue
that underflows gets its threshold increased, so the queues are out of
sync.
To make the reporting accurate, remove it and report the threshold of
each TX queue instead, as "qN_tx_threshold".
Signed-off-by: Ovidiu Panait <ovidiu.panait.rb@renesas.com>
---
drivers/net/ethernet/stmicro/stmmac/common.h | 1 -
drivers/net/ethernet/stmicro/stmmac/stmmac_ethtool.c | 3 ++-
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 4 ----
3 files changed, 2 insertions(+), 6 deletions(-)
diff --git a/drivers/net/ethernet/stmicro/stmmac/common.h b/drivers/net/ethernet/stmicro/stmmac/common.h
index 1729fb29ece3..c74bc7c9750e 100644
--- a/drivers/net/ethernet/stmicro/stmmac/common.h
+++ b/drivers/net/ethernet/stmicro/stmmac/common.h
@@ -158,7 +158,6 @@ struct stmmac_extra_stats {
unsigned long fatal_bus_error_irq;
/* Tx/Rx IRQ Events */
unsigned long rx_early_irq;
- unsigned long threshold;
unsigned long irq_receive_pmt_irq_n;
/* MMC info */
unsigned long mmc_tx_irq_n;
diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_ethtool.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_ethtool.c
index 30e0d499326d..c616a1c95e5d 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_ethtool.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_ethtool.c
@@ -89,7 +89,6 @@ static const struct stmmac_stats stmmac_gstrings_stats[] = {
STMMAC_STAT(fatal_bus_error_irq),
/* Tx/Rx IRQ Events */
STMMAC_STAT(rx_early_irq),
- STMMAC_STAT(threshold),
STMMAC_STAT(irq_receive_pmt_irq_n),
/* MMC info */
STMMAC_STAT(mmc_tx_irq_n),
@@ -289,6 +288,7 @@ static const struct stmmac_stats stmmac_mmc[] = {
static const char stmmac_qstats_tx_string[][ETH_GSTRING_LEN] = {
"tx_pkt_n",
"tx_irq_n",
+ "tx_threshold",
#define STMMAC_TXQ_STATS ARRAY_SIZE(stmmac_qstats_tx_string)
};
@@ -496,6 +496,7 @@ static void stmmac_get_per_qstats(struct stmmac_priv *priv, u64 *data)
*data++ = pkt_n;
*data++ = stmmac_get_tx_normal_irq_n(priv, q);
+ *data++ = priv->tx_threshold[q];
}
for (q = 0; q < rx_cnt; q++) {
diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
index b5ee0f297df6..2fed7713e376 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
@@ -2633,7 +2633,6 @@ static void stmmac_dma_operation_mode(struct stmmac_priv *priv)
*/
txmode = SF_DMA_MODE;
rxmode = SF_DMA_MODE;
- priv->xstats.threshold = SF_DMA_MODE;
} else {
rxmode = SF_DMA_MODE;
}
@@ -2836,7 +2835,6 @@ static void stmmac_bump_dma_threshold(struct stmmac_priv *priv, u32 chan)
stmmac_set_dma_operation_mode(priv, thresh, SF_DMA_MODE, chan);
priv->tx_threshold[chan] = thresh;
- priv->xstats.threshold = thresh;
}
/**
@@ -8135,8 +8133,6 @@ static int __stmmac_dvr_probe(struct device *device,
#endif
priv->msg_enable = netif_msg_init(debug, default_msg_level);
- priv->xstats.threshold = tc;
-
for (i = 0; i < MTL_MAX_TX_QUEUES; i++)
priv->tx_threshold[i] = tc;
--
2.34.1
^ permalink raw reply related [flat|nested] 5+ messages in thread* Re: [PATCH net 0/2] net: stmmac: Track and report the TX DMA threshold per queue
2026-10-13 11:45 [PATCH net 0/2] net: stmmac: Track and report the TX DMA threshold per queue Ovidiu Panait
2026-10-13 11:45 ` [PATCH net 1/2] net: stmmac: Track " Ovidiu Panait
2026-10-13 11:45 ` [PATCH net 2/2] net: stmmac: Report " Ovidiu Panait
@ 2026-10-13 12:44 ` Maxime Chevallier
2 siblings, 0 replies; 5+ messages in thread
From: Maxime Chevallier @ 2026-10-13 12:44 UTC (permalink / raw)
To: Ovidiu Panait, andrew+netdev, davem, edumazet, kuba, pabeni,
mcoquelin.stm32, alexandre.torgue, xiaoliang.yang_1
Cc: netdev, linux-stm32, linux-arm-kernel, linux-kernel
Hi Ovidiu,
On 10/13/26 13:45, Ovidiu Panait wrote:
> On the Renesas RZ/V2H platform, which has two dwmac4 controllers, we
> observed sporadic iperf3 TCP TX failures on one of the controllers, with
> throughput dropping from >900Mbit/s to ~1Mbit/s due to TX underflows and
> the resulting retransmissions.
>
> Commit 3a6c12a0c6c3 ("net: stmmac: bump tc when get underflow error from
> DMA descriptor") made dwmac4 raise the DMA threshold of the queues that
> report TX underflows, by increasing a global module parameter 'tc'.
> However, this logic does not work correctly across multiple queues and
> controllers, and can leave the TX DMA threshold stuck at 64 bytes.
>
> This series fixes that logic to track and report the TX DMA threshold per
> queue for each controller.
That's a nice cleanup, thanks for addressing it :)
Patch 2 belongs to net-next though, it's not a fix and it brings some user-visible
changes in the stats.
Thanks,
Maxime
^ permalink raw reply [flat|nested] 5+ messages in thread