Netdev List
 help / color / mirror / Atom feed
* [PATCH net 1/2] eth: bnxt_en: restore Rx coalescing after re-allocating the completion ring
@ 2026-11-20  3:46 Jakub Kicinski
  2026-11-20  3:46 ` [PATCH net 2/2] eth: bnxt_en: restore Tx " Jakub Kicinski
  2026-11-20  3:53 ` [PATCH net 1/2] eth: bnxt_en: restore Rx " netdev-bot+sinfo
  0 siblings, 2 replies; 3+ messages in thread
From: Jakub Kicinski @ 2026-11-20  3:46 UTC (permalink / raw)
  To: davem
  Cc: netdev, edumazet, pabeni, andrew+netdev, horms, michael.chan,
	pavan.chebbi, somnath.kotur, ajit.khaparde, michal.swiatkowski,
	Jakub Kicinski

With TPH enabled, queue restart frees and re-allocates the Rx
completion ring in the FW, to apply the new Steering Tag. The new
FW ring starts with default interrupt coalescing parameters. The
driver programs coalescing only at open, from ethtool and from DIM,
so after a restart the ring ignores the configured settings, while
ethtool keeps reporting them.

Re-program Rx coalescing of the ring after allocating it. Use DIM's
current profile if adaptive coalescing is enabled, the static
settings otherwise. The per-ring DIM values may be stale when DIM is
off, ethtool changes to static coalescing don't update them.

With rx-usecs 1000, re-allocating the ring on a BCM57508 lowers the
latency of a 1 byte ping-pong on that queue by ~250us. Latency goes
back up once ethtool re-programs coalescing.

Fixes: 6b6bf60fc95f ("bnxt_en: Reallocate RX completion ring for TPH support")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c | 24 +++++++++++++++++++++++
 1 file changed, 24 insertions(+)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index e78cf7ad3cbc..25d7c4cc3377 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -8618,6 +8618,25 @@ int bnxt_hwrm_set_coal(struct bnxt *bp)
 	return rc;
 }
 
+/* Program Rx coalescing of a single ring, e.g. after its completion ring
+ * was re-allocated in the FW, which resets the coalescing parameters.
+ */
+static int bnxt_hwrm_set_rx_ring_coal(struct bnxt *bp, struct bnxt_napi *bnapi)
+{
+	struct hwrm_ring_cmpl_ring_cfg_aggint_params_input *req;
+	int rc;
+
+	if (bp->flags & BNXT_FLAG_DIM)
+		return bnxt_hwrm_set_ring_coal(bp, bnapi);
+
+	rc = hwrm_req_init(bp, req, HWRM_RING_CMPL_RING_CFG_AGGINT_PARAMS);
+	if (rc)
+		return rc;
+
+	bnxt_hwrm_set_coal_params(bp, &bp->rx_coal, req);
+	return bnxt_hwrm_set_rx_coal(bp, bnapi, req);
+}
+
 static void bnxt_hwrm_stat_ctx_free(struct bnxt *bp)
 {
 	struct hwrm_stat_ctx_clr_stats_input *req0 = NULL;
@@ -16643,6 +16662,11 @@ static int bnxt_queue_start(struct net_device *dev,
 		rc = bnxt_hwrm_cp_ring_alloc_p5(bp, rxr->rx_cpr);
 		if (rc)
 			goto err_reset;
+
+		rc = bnxt_hwrm_set_rx_ring_coal(bp, bnapi);
+		if (rc)
+			netdev_warn(bp->dev, "HWRM set coalescing failure rc: %d\n",
+				    rc);
 	}
 
 	rc = bnxt_hwrm_rx_agg_ring_alloc(bp, rxr);
-- 
2.56.1


^ permalink raw reply related	[flat|nested] 3+ messages in thread

* [PATCH net 2/2] eth: bnxt_en: restore Tx coalescing after re-allocating the completion ring
  2026-11-20  3:46 [PATCH net 1/2] eth: bnxt_en: restore Rx coalescing after re-allocating the completion ring Jakub Kicinski
@ 2026-11-20  3:46 ` Jakub Kicinski
  2026-11-20  3:53 ` [PATCH net 1/2] eth: bnxt_en: restore Rx " netdev-bot+sinfo
  1 sibling, 0 replies; 3+ messages in thread
From: Jakub Kicinski @ 2026-11-20  3:46 UTC (permalink / raw)
  To: davem
  Cc: netdev, edumazet, pabeni, andrew+netdev, horms, michael.chan,
	pavan.chebbi, somnath.kotur, ajit.khaparde, michal.swiatkowski,
	Jakub Kicinski

With TPH enabled, queue restart also frees and re-allocates the Tx
completion rings which share the NAPI instance with the Rx queue.
Same as on the Rx side, the new FW rings start with default interrupt
coalescing parameters, and the configured settings are applied again
only at the next open or ethtool change.

On P5+ chips the Tx completion rings are separate from the Rx ones
even in combined channels, and the Tx settings do take effect. On a
BCM57508 the completion latency of sparse Tx traffic follows
tx-usecs, it's ~20us with tx-usecs 1 and ~220us with tx-usecs 200.

Re-program Tx coalescing of each completion ring after allocating it.
DIM only manages Rx, the static settings always apply to Tx.

Fixes: fe96d717d38e ("bnxt_en: Extend queue stop/start for TX rings")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 25d7c4cc3377..52606db3f7b4 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -8637,6 +8637,24 @@ static int bnxt_hwrm_set_rx_ring_coal(struct bnxt *bp, struct bnxt_napi *bnapi)
 	return bnxt_hwrm_set_rx_coal(bp, bnapi, req);
 }
 
+/* Program Tx coalescing of the completion ring of a single Tx ring, e.g.
+ * after it was re-allocated in the FW.
+ */
+static int bnxt_hwrm_set_tx_ring_coal(struct bnxt *bp,
+				      struct bnxt_tx_ring_info *txr)
+{
+	struct hwrm_ring_cmpl_ring_cfg_aggint_params_input *req;
+	int rc;
+
+	rc = hwrm_req_init(bp, req, HWRM_RING_CMPL_RING_CFG_AGGINT_PARAMS);
+	if (rc)
+		return rc;
+
+	bnxt_hwrm_set_coal_params(bp, &bp->tx_coal, req);
+	req->ring_id = cpu_to_le16(bnxt_cp_ring_for_tx(bp, txr));
+	return hwrm_req_send(bp, req);
+}
+
 static void bnxt_hwrm_stat_ctx_free(struct bnxt *bp)
 {
 	struct hwrm_stat_ctx_clr_stats_input *req0 = NULL;
@@ -11857,6 +11875,11 @@ static int bnxt_tx_queue_start(struct bnxt *bp, int idx)
 		if (rc)
 			return rc;
 
+		rc = bnxt_hwrm_set_tx_ring_coal(bp, txr);
+		if (rc)
+			netdev_warn(bp->dev, "HWRM set coalescing failure rc: %d\n",
+				    rc);
+
 		rc = bnxt_hwrm_tx_ring_alloc(bp, txr, false);
 		if (rc)
 			return rc;
-- 
2.56.1


^ permalink raw reply related	[flat|nested] 3+ messages in thread

* Re: [PATCH net 1/2] eth: bnxt_en: restore Rx coalescing after re-allocating the completion ring
  2026-11-20  3:46 [PATCH net 1/2] eth: bnxt_en: restore Rx coalescing after re-allocating the completion ring Jakub Kicinski
  2026-11-20  3:46 ` [PATCH net 2/2] eth: bnxt_en: restore Tx " Jakub Kicinski
@ 2026-11-20  3:53 ` netdev-bot+sinfo
  1 sibling, 0 replies; 3+ messages in thread
From: netdev-bot+sinfo @ 2026-11-20  3:53 UTC (permalink / raw)
  To: Jakub Kicinski
  Cc: davem, netdev, edumazet, pabeni, andrew+netdev, horms,
	michael.chan, pavan.chebbi, somnath.kotur, ajit.khaparde,
	michal.swiatkowski

Hi!

This is an automated message. This series looks like a fix, but its
commit messages seem to be missing some information:

 - How the issue was discovered, e.g. hit in production, hit during
   development, syzbot report, manual code inspection, LLM or static
   analysis tool scan.

Please do not repost the series just to address the above. Instead,
reply to this email with the missing information, so that reviewers
can take it into account.

*IMPORTANT*: If you do repost (for other reasons) - please include
the relevant info in the cover letter or commit messages.
-- 
The evaluation is done by an LLM so it may be wrong, if you think
that is the case please reply and explain.

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2026-11-20  3:53 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-20  3:46 [PATCH net 1/2] eth: bnxt_en: restore Rx coalescing after re-allocating the completion ring Jakub Kicinski
2026-11-20  3:46 ` [PATCH net 2/2] eth: bnxt_en: restore Tx " Jakub Kicinski
2026-11-20  3:53 ` [PATCH net 1/2] eth: bnxt_en: restore Rx " netdev-bot+sinfo

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox