From 9536923d3f353873e54d96769fbd2e7f867a73e1 Mon Sep 17 00:00:00 2001
From: Tariq Toukan <tariqt@nvidia.com>
Date: Thu, 20 May 2021 12:31:44 +0300
Subject: net/mlx5e: Remove unused tstamp SQ field

Remove tstamp pointer in mlx5e_txqsq as it's no longer used after
commit 7c39afb394c7 ("net/mlx5: PTP code migration to driver core section").

Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Reviewed-by: Aya Levin <ayal@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en.h      | 1 -
 drivers/net/ethernet/mellanox/mlx5/core/en/ptp.c  | 1 -
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 1 -
 3 files changed, 3 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index c14e06ca64d8..1389fd91321b 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -432,7 +432,6 @@ struct mlx5e_txqsq {
 	__be32                     mkey_be;
 	unsigned long              state;
 	unsigned int               hw_mtu;
-	struct hwtstamp_config    *tstamp;
 	struct mlx5_clock         *clock;
 	struct net_device         *netdev;
 	struct mlx5_core_dev      *mdev;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/ptp.c b/drivers/net/ethernet/mellanox/mlx5/core/en/ptp.c
index 82baafd3c00c..ad4e5e759426 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/ptp.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/ptp.c
@@ -195,7 +195,6 @@ static int mlx5e_ptp_alloc_txqsq(struct mlx5e_ptp *c, int txq_ix,
 	int node;
 
 	sq->pdev      = c->pdev;
-	sq->tstamp    = c->tstamp;
 	sq->clock     = &mdev->clock;
 	sq->mkey_be   = c->mkey_be;
 	sq->netdev    = c->netdev;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index bf80fb612449..31eab1ef14b4 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -1313,7 +1313,6 @@ static int mlx5e_alloc_txqsq(struct mlx5e_channel *c,
 	int err;
 
 	sq->pdev      = c->pdev;
-	sq->tstamp    = c->tstamp;
 	sq->clock     = &mdev->clock;
 	sq->mkey_be   = c->mkey_be;
 	sq->netdev    = c->netdev;
-- 
cgit v1.2.3


From c27bd1718c06eb356a4ea0ab8588ad442d953947 Mon Sep 17 00:00:00 2001
From: Aya Levin <ayal@nvidia.com>
Date: Mon, 17 Jan 2022 11:46:35 +0200
Subject: net/mlx5e: Read max WQEBBs on the SQ from firmware

Prior to this patch the maximal value for max WQEBBs (WQE Basic Blocks,
where WQE is a Work Queue Element) on the TX side was assumed to be 16
(fixed value). All firmware versions till today comply to this. In order
to be more flexible and resilient, read from FW the corresponding:
max_wqe_sz_sq. This value describes the maximum WQE size given in bytes,
thus max WQEBBs is given by the division in WQEBB's byte size. The
driver uses the top between 16 and the division result. This ensures
synchronization between driver and firmware and avoids unexpected
behavior. Store this value on the different SQs (Send Queues) for easy
access.

Signed-off-by: Aya Levin <ayal@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en.h       | 14 +++++++++++
 .../net/ethernet/mellanox/mlx5/core/en/params.c    |  8 +++----
 drivers/net/ethernet/mellanox/mlx5/core/en/ptp.c   |  2 +-
 drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h  | 28 +++++++++++++++-------
 drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c   |  4 +---
 .../ethernet/mellanox/mlx5/core/en_accel/ktls_tx.c |  6 ++---
 .../mellanox/mlx5/core/en_accel/tls_rxtx.c         |  2 +-
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c  | 15 ++++++++----
 8 files changed, 53 insertions(+), 26 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index 1389fd91321b..a51814d9ffa9 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -221,6 +221,16 @@ static inline int mlx5e_get_max_num_channels(struct mlx5_core_dev *mdev)
 		min_t(int, mlx5_comp_vectors_count(mdev), MLX5E_MAX_NUM_CHANNELS);
 }
 
+/* The maximum WQE size can be retrieved by max_wqe_sz_sq in
+ * bytes units. Driver hardens the limitation to 1KB (16
+ * WQEBBs), unless firmware capability is stricter.
+ */
+static inline u16 mlx5e_get_max_sq_wqebbs(struct mlx5_core_dev *mdev)
+{
+	return min_t(u16, MLX5_SEND_WQE_MAX_WQEBBS,
+		     MLX5_CAP_GEN(mdev, max_wqe_sz_sq) / MLX5_SEND_WQE_BB);
+}
+
 struct mlx5e_tx_wqe {
 	struct mlx5_wqe_ctrl_seg ctrl;
 	struct mlx5_wqe_eth_seg  eth;
@@ -445,6 +455,7 @@ struct mlx5e_txqsq {
 	struct work_struct         recover_work;
 	struct mlx5e_ptpsq        *ptpsq;
 	cqe_ts_to_ns               ptp_cyc2time;
+	u16                        max_sq_wqebbs;
 } ____cacheline_aligned_in_smp;
 
 struct mlx5e_dma_info {
@@ -539,6 +550,7 @@ struct mlx5e_xdpsq {
 	u32                        sqn;
 	struct device             *pdev;
 	__be32                     mkey_be;
+	u16                        stop_room;
 	u8                         min_inline_mode;
 	unsigned long              state;
 	unsigned int               hw_mtu;
@@ -546,6 +558,7 @@ struct mlx5e_xdpsq {
 	/* control path */
 	struct mlx5_wq_ctrl        wq_ctrl;
 	struct mlx5e_channel      *channel;
+	u16                        max_sq_wqebbs;
 } ____cacheline_aligned_in_smp;
 
 struct mlx5e_ktls_resync_resp;
@@ -574,6 +587,7 @@ struct mlx5e_icosq {
 	/* control path */
 	struct mlx5_wq_ctrl        wq_ctrl;
 	struct mlx5e_channel      *channel;
+	u16                        max_sq_wqebbs;
 
 	struct work_struct         recover_work;
 } ____cacheline_aligned_in_smp;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/params.c b/drivers/net/ethernet/mellanox/mlx5/core/en/params.c
index 66180ffb4606..4ce720da1865 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/params.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/params.c
@@ -196,13 +196,13 @@ u16 mlx5e_calc_sq_stop_room(struct mlx5_core_dev *mdev, struct mlx5e_params *par
 	u16 stop_room;
 
 	stop_room  = mlx5e_tls_get_stop_room(mdev, params);
-	stop_room += mlx5e_stop_room_for_wqe(MLX5_SEND_WQE_MAX_WQEBBS);
+	stop_room += mlx5e_stop_room_for_max_wqe(mdev);
 	if (is_mpwqe)
 		/* A MPWQE can take up to the maximum-sized WQE + all the normal
 		 * stop room can be taken if a new packet breaks the active
 		 * MPWQE session and allocates its WQEs right away.
 		 */
-		stop_room += mlx5e_stop_room_for_wqe(MLX5_SEND_WQE_MAX_WQEBBS);
+		stop_room += mlx5e_stop_room_for_max_wqe(mdev);
 
 	return stop_room;
 }
@@ -774,10 +774,10 @@ static void mlx5e_build_async_icosq_param(struct mlx5_core_dev *mdev,
 	void *wq = MLX5_ADDR_OF(sqc, sqc, wq);
 
 	mlx5e_build_sq_param_common(mdev, param);
-	param->stop_room = mlx5e_stop_room_for_wqe(1); /* for XSK NOP */
+	param->stop_room = mlx5e_stop_room_for_wqe(mdev, 1); /* for XSK NOP */
 	param->is_tls = mlx5e_accel_is_ktls_rx(mdev);
 	if (param->is_tls)
-		param->stop_room += mlx5e_stop_room_for_wqe(1); /* for TLS RX resync NOP */
+		param->stop_room += mlx5e_stop_room_for_wqe(mdev, 1); /* for TLS RX resync NOP */
 	MLX5_SET(sqc, sqc, reg_umr, MLX5_CAP_ETH(mdev, reg_umr_sq));
 	MLX5_SET(wq, wq, log_wq_sz, log_wq_size);
 	mlx5e_build_ico_cq_param(mdev, log_wq_size, &param->cqp);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/ptp.c b/drivers/net/ethernet/mellanox/mlx5/core/en/ptp.c
index ad4e5e759426..335b20b6383b 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/ptp.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/ptp.c
@@ -448,7 +448,7 @@ static void mlx5e_ptp_build_sq_param(struct mlx5_core_dev *mdev,
 
 	wq = MLX5_ADDR_OF(sqc, sqc, wq);
 	MLX5_SET(wq, wq, log_wq_sz, params->log_sq_size);
-	param->stop_room = mlx5e_stop_room_for_wqe(MLX5_SEND_WQE_MAX_WQEBBS);
+	param->stop_room = mlx5e_stop_room_for_max_wqe(mdev);
 	mlx5e_build_tx_cq_param(mdev, params, &param->cqp);
 }
 
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h b/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h
index b789af07829c..67dd4f415b7a 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h
@@ -431,10 +431,10 @@ mlx5e_set_eseg_swp(struct sk_buff *skb, struct mlx5_wqe_eth_seg *eseg,
 	}
 }
 
-static inline u16 mlx5e_stop_room_for_wqe(u16 wqe_size)
-{
-	BUILD_BUG_ON(PAGE_SIZE / MLX5_SEND_WQE_BB < MLX5_SEND_WQE_MAX_WQEBBS);
+#define MLX5E_STOP_ROOM(wqebbs) ((wqebbs) * 2 - 1)
 
+static inline u16 mlx5e_stop_room_for_wqe(struct mlx5_core_dev *mdev, u16 wqe_size)
+{
 	/* A WQE must not cross the page boundary, hence two conditions:
 	 * 1. Its size must not exceed the page size.
 	 * 2. If the WQE size is X, and the space remaining in a page is less
@@ -443,18 +443,28 @@ static inline u16 mlx5e_stop_room_for_wqe(u16 wqe_size)
 	 *    stop room of X-1 + X.
 	 * WQE size is also limited by the hardware limit.
 	 */
+	WARN_ONCE(wqe_size > mlx5e_get_max_sq_wqebbs(mdev),
+		  "wqe_size %u is greater than max SQ WQEBBs %u",
+		  wqe_size, mlx5e_get_max_sq_wqebbs(mdev));
 
-	if (__builtin_constant_p(wqe_size))
-		BUILD_BUG_ON(wqe_size > MLX5_SEND_WQE_MAX_WQEBBS);
-	else
-		WARN_ON_ONCE(wqe_size > MLX5_SEND_WQE_MAX_WQEBBS);
 
-	return wqe_size * 2 - 1;
+	return MLX5E_STOP_ROOM(wqe_size);
+}
+
+static inline u16 mlx5e_stop_room_for_max_wqe(struct mlx5_core_dev *mdev)
+{
+	return MLX5E_STOP_ROOM(mlx5e_get_max_sq_wqebbs(mdev));
 }
 
 static inline bool mlx5e_icosq_can_post_wqe(struct mlx5e_icosq *sq, u16 wqe_size)
 {
-	u16 room = sq->reserved_room + mlx5e_stop_room_for_wqe(wqe_size);
+	u16 room = sq->reserved_room;
+
+	WARN_ONCE(wqe_size > sq->max_sq_wqebbs,
+		  "wqe_size %u is greater than max SQ WQEBBs %u",
+		  wqe_size, sq->max_sq_wqebbs);
+
+	room += MLX5E_STOP_ROOM(wqe_size);
 
 	return mlx5e_wqc_has_room_for(&sq->wq, sq->cc, sq->pc, room);
 }
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c b/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c
index 56e10c84a706..a5e71e60e981 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c
@@ -245,10 +245,8 @@ enum {
 INDIRECT_CALLABLE_SCOPE int mlx5e_xmit_xdp_frame_check_mpwqe(struct mlx5e_xdpsq *sq)
 {
 	if (unlikely(!sq->mpwqe.wqe)) {
-		const u16 stop_room = mlx5e_stop_room_for_wqe(MLX5_SEND_WQE_MAX_WQEBBS);
-
 		if (unlikely(!mlx5e_wqc_has_room_for(&sq->wq, sq->cc, sq->pc,
-						     stop_room))) {
+						     sq->stop_room))) {
 			/* SQ is full, ring doorbell */
 			mlx5e_xmit_xdp_doorbell(sq);
 			sq->stats->full++;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/ktls_tx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/ktls_tx.c
index 9ad3459fb63a..aaf11c66bf4c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/ktls_tx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/ktls_tx.c
@@ -32,9 +32,9 @@ u16 mlx5e_ktls_get_stop_room(struct mlx5_core_dev *mdev, struct mlx5e_params *pa
 
 	num_dumps = mlx5e_ktls_dumps_num_wqes(params, MAX_SKB_FRAGS, TLS_MAX_PAYLOAD_SIZE);
 
-	stop_room += mlx5e_stop_room_for_wqe(MLX5E_TLS_SET_STATIC_PARAMS_WQEBBS);
-	stop_room += mlx5e_stop_room_for_wqe(MLX5E_TLS_SET_PROGRESS_PARAMS_WQEBBS);
-	stop_room += num_dumps * mlx5e_stop_room_for_wqe(MLX5E_KTLS_DUMP_WQEBBS);
+	stop_room += mlx5e_stop_room_for_wqe(mdev, MLX5E_TLS_SET_STATIC_PARAMS_WQEBBS);
+	stop_room += mlx5e_stop_room_for_wqe(mdev, MLX5E_TLS_SET_PROGRESS_PARAMS_WQEBBS);
+	stop_room += num_dumps * mlx5e_stop_room_for_wqe(mdev, MLX5E_KTLS_DUMP_WQEBBS);
 
 	return stop_room;
 }
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/tls_rxtx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/tls_rxtx.c
index 7a700f913582..a05580cea481 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/tls_rxtx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/tls_rxtx.c
@@ -386,5 +386,5 @@ u16 mlx5e_tls_get_stop_room(struct mlx5_core_dev *mdev, struct mlx5e_params *par
 
 	/* FPGA */
 	/* Resync SKB. */
-	return mlx5e_stop_room_for_wqe(MLX5_SEND_WQE_MAX_WQEBBS);
+	return mlx5e_stop_room_for_max_wqe(mdev);
 }
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 31eab1ef14b4..0a0e5bd77384 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -72,12 +72,13 @@
 
 bool mlx5e_check_fragmented_striding_rq_cap(struct mlx5_core_dev *mdev)
 {
-	bool striding_rq_umr = MLX5_CAP_GEN(mdev, striding_rq) &&
-		MLX5_CAP_GEN(mdev, umr_ptr_rlky) &&
-		MLX5_CAP_ETH(mdev, reg_umr_sq);
-	u16 max_wqe_sz_cap = MLX5_CAP_GEN(mdev, max_wqe_sz_sq);
-	bool inline_umr = MLX5E_UMR_WQE_INLINE_SZ <= max_wqe_sz_cap;
+	bool striding_rq_umr, inline_umr;
+	u16 max_wqe_sz_cap;
 
+	striding_rq_umr = MLX5_CAP_GEN(mdev, striding_rq) && MLX5_CAP_GEN(mdev, umr_ptr_rlky) &&
+			  MLX5_CAP_ETH(mdev, reg_umr_sq);
+	max_wqe_sz_cap = mlx5e_get_max_sq_wqebbs(mdev) * MLX5_SEND_WQE_BB;
+	inline_umr = max_wqe_sz_cap >= MLX5E_UMR_WQE_INLINE_SZ;
 	if (!striding_rq_umr)
 		return false;
 	if (!inline_umr) {
@@ -1164,6 +1165,8 @@ static int mlx5e_alloc_xdpsq(struct mlx5e_channel *c,
 		is_redirect ?
 			&c->priv->channel_stats[c->ix]->xdpsq :
 			&c->priv->channel_stats[c->ix]->rq_xdpsq;
+	sq->max_sq_wqebbs = mlx5e_get_max_sq_wqebbs(mdev);
+	sq->stop_room = MLX5E_STOP_ROOM(sq->max_sq_wqebbs);
 
 	param->wq.db_numa_node = cpu_to_node(c->cpu);
 	err = mlx5_wq_cyc_create(mdev, &param->wq, sqc_wq, wq, &sq->wq_ctrl);
@@ -1238,6 +1241,7 @@ static int mlx5e_alloc_icosq(struct mlx5e_channel *c,
 	sq->channel   = c;
 	sq->uar_map   = mdev->mlx5e_res.hw_objs.bfreg.map;
 	sq->reserved_room = param->stop_room;
+	sq->max_sq_wqebbs = mlx5e_get_max_sq_wqebbs(mdev);
 
 	param->wq.db_numa_node = cpu_to_node(c->cpu);
 	err = mlx5_wq_cyc_create(mdev, &param->wq, sqc_wq, wq, &sq->wq_ctrl);
@@ -1323,6 +1327,7 @@ static int mlx5e_alloc_txqsq(struct mlx5e_channel *c,
 	sq->uar_map   = mdev->mlx5e_res.hw_objs.bfreg.map;
 	sq->min_inline_mode = params->tx_min_inline_mode;
 	sq->hw_mtu    = MLX5E_SW2HW_MTU(params, params->sw_mtu);
+	sq->max_sq_wqebbs = mlx5e_get_max_sq_wqebbs(mdev);
 	INIT_WORK(&sq->recover_work, mlx5e_tx_err_cqe_work);
 	if (!MLX5_CAP_ETH(mdev, wqe_vlan_insert))
 		set_bit(MLX5E_SQ_STATE_VLAN_NEED_L2_INLINE, &sq->state);
-- 
cgit v1.2.3


From 76c31e5f758596509fbab120b8d055bf927ed165 Mon Sep 17 00:00:00 2001
From: Aya Levin <ayal@nvidia.com>
Date: Mon, 10 May 2021 10:13:06 +0300
Subject: net/mlx5e: Use FW limitation for max MPW WQEBBs

Calculate maximal count of MPW WQEBBs on SQ's creation and store it
there. Remove MLX5E_TX_MPW_MAX_NUM_DS and MLX5E_TX_MPW_MAX_WQEBBS.
Update mlx5e_tx_mpwqe_is_full() and mlx5e_xdp_mpqwe_is_full() .

Signed-off-by: Aya Levin <ayal@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en.h       | 23 ++++++++++++++++++++--
 .../net/ethernet/mellanox/mlx5/core/en/params.c    |  2 +-
 drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h  | 17 ++--------------
 drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c   |  4 ++--
 drivers/net/ethernet/mellanox/mlx5/core/en/xdp.h   |  7 ++++---
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c  |  2 ++
 drivers/net/ethernet/mellanox/mlx5/core/en_rx.c    |  2 +-
 drivers/net/ethernet/mellanox/mlx5/core/en_tx.c    |  4 ++--
 8 files changed, 35 insertions(+), 26 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index a51814d9ffa9..99529e238fc4 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -172,8 +172,9 @@ struct page_pool;
 #define MLX5E_KLM_ENTRIES_PER_WQE(wqe_size)\
 	ALIGN_DOWN(MLX5E_KLM_MAX_ENTRIES_PER_WQE(wqe_size), MLX5_UMR_KLM_ALIGNMENT)
 
-#define MLX5E_MAX_KLM_PER_WQE \
-	MLX5E_KLM_ENTRIES_PER_WQE(MLX5E_TX_MPW_MAX_NUM_DS << MLX5_MKEY_BSF_OCTO_SIZE)
+#define MLX5E_MAX_KLM_PER_WQE(mdev) \
+	MLX5E_KLM_ENTRIES_PER_WQE(mlx5e_get_sw_max_sq_mpw_wqebbs(mlx5e_get_max_sq_wqebbs(mdev)) \
+				   << MLX5_MKEY_BSF_OCTO_SIZE)
 
 #define MLX5E_MSG_LEVEL			NETIF_MSG_LINK
 
@@ -231,6 +232,22 @@ static inline u16 mlx5e_get_max_sq_wqebbs(struct mlx5_core_dev *mdev)
 		     MLX5_CAP_GEN(mdev, max_wqe_sz_sq) / MLX5_SEND_WQE_BB);
 }
 
+static inline u16 mlx5e_get_sw_max_sq_mpw_wqebbs(u16 max_sq_wqebbs)
+{
+/* The return value will be multiplied by MLX5_SEND_WQEBB_NUM_DS.
+ * Since max_sq_wqebbs may be up to MLX5_SEND_WQE_MAX_WQEBBS == 16,
+ * see mlx5e_get_max_sq_wqebbs(), the multiplication (16 * 4 == 64)
+ * overflows the 6-bit DS field of Ctrl Segment. Use a bound lower
+ * than MLX5_SEND_WQE_MAX_WQEBBS to let a full-session WQE be
+ * cache-aligned.
+ */
+#if L1_CACHE_BYTES < 128
+	return min_t(u16, max_sq_wqebbs, MLX5_SEND_WQE_MAX_WQEBBS - 1);
+#else
+	return min_t(u16, max_sq_wqebbs, MLX5_SEND_WQE_MAX_WQEBBS - 2);
+#endif
+}
+
 struct mlx5e_tx_wqe {
 	struct mlx5_wqe_ctrl_seg ctrl;
 	struct mlx5_wqe_eth_seg  eth;
@@ -437,6 +454,7 @@ struct mlx5e_txqsq {
 	struct netdev_queue       *txq;
 	u32                        sqn;
 	u16                        stop_room;
+	u16                        max_sq_mpw_wqebbs;
 	u8                         min_inline_mode;
 	struct device             *pdev;
 	__be32                     mkey_be;
@@ -551,6 +569,7 @@ struct mlx5e_xdpsq {
 	struct device             *pdev;
 	__be32                     mkey_be;
 	u16                        stop_room;
+	u16                        max_sq_mpw_wqebbs;
 	u8                         min_inline_mode;
 	unsigned long              state;
 	unsigned int               hw_mtu;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/params.c b/drivers/net/ethernet/mellanox/mlx5/core/en/params.c
index 4ce720da1865..d41936d65483 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/params.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/params.c
@@ -717,7 +717,7 @@ static u32 mlx5e_shampo_icosq_sz(struct mlx5_core_dev *mdev,
 	int wq_size = BIT(MLX5_GET(wq, wqc, log_wq_sz));
 	u32 wqebbs;
 
-	max_klm_per_umr = MLX5E_MAX_KLM_PER_WQE;
+	max_klm_per_umr = MLX5E_MAX_KLM_PER_WQE(mdev);
 	max_hd_per_wqe = mlx5e_shampo_hd_per_wqe(mdev, params, rq_param);
 	max_num_of_umr_per_wqe = max_hd_per_wqe / max_klm_per_umr;
 	rest = max_hd_per_wqe % max_klm_per_umr;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h b/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h
index 67dd4f415b7a..1c48cfad9dd7 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h
@@ -9,19 +9,6 @@
 
 #define MLX5E_TX_WQE_EMPTY_DS_COUNT (sizeof(struct mlx5e_tx_wqe) / MLX5_SEND_WQE_DS)
 
-/* The mult of MLX5_SEND_WQE_MAX_WQEBBS * MLX5_SEND_WQEBB_NUM_DS
- * (16 * 4 == 64) does not fit in the 6-bit DS field of Ctrl Segment.
- * We use a bound lower that MLX5_SEND_WQE_MAX_WQEBBS to let a
- * full-session WQE be cache-aligned.
- */
-#if L1_CACHE_BYTES < 128
-#define MLX5E_TX_MPW_MAX_WQEBBS (MLX5_SEND_WQE_MAX_WQEBBS - 1)
-#else
-#define MLX5E_TX_MPW_MAX_WQEBBS (MLX5_SEND_WQE_MAX_WQEBBS - 2)
-#endif
-
-#define MLX5E_TX_MPW_MAX_NUM_DS (MLX5E_TX_MPW_MAX_WQEBBS * MLX5_SEND_WQEBB_NUM_DS)
-
 #define INL_HDR_START_SZ (sizeof(((struct mlx5_wqe_eth_seg *)NULL)->inline_hdr.start))
 
 #define MLX5E_RX_ERR_CQE(cqe) (get_cqe_opcode(cqe) != MLX5_CQE_RESP_SEND)
@@ -308,9 +295,9 @@ mlx5e_tx_dma_unmap(struct device *pdev, struct mlx5e_sq_dma *dma)
 void mlx5e_sq_xmit_simple(struct mlx5e_txqsq *sq, struct sk_buff *skb, bool xmit_more);
 void mlx5e_tx_mpwqe_ensure_complete(struct mlx5e_txqsq *sq);
 
-static inline bool mlx5e_tx_mpwqe_is_full(struct mlx5e_tx_mpwqe *session)
+static inline bool mlx5e_tx_mpwqe_is_full(struct mlx5e_tx_mpwqe *session, u8 max_sq_mpw_wqebbs)
 {
-	return session->ds_count == MLX5E_TX_MPW_MAX_NUM_DS;
+	return session->ds_count == max_sq_mpw_wqebbs * MLX5_SEND_WQEBB_NUM_DS;
 }
 
 static inline void mlx5e_rqwq_reset(struct mlx5e_rq *rq)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c b/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c
index a5e71e60e981..a7f020399370 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c
@@ -199,7 +199,7 @@ static void mlx5e_xdp_mpwqe_session_start(struct mlx5e_xdpsq *sq)
 	struct mlx5e_tx_wqe *wqe;
 	u16 pi;
 
-	pi = mlx5e_xdpsq_get_next_pi(sq, MLX5E_TX_MPW_MAX_WQEBBS);
+	pi = mlx5e_xdpsq_get_next_pi(sq, sq->max_sq_mpw_wqebbs);
 	wqe = MLX5E_TX_FETCH_WQE(sq, pi);
 	net_prefetchw(wqe->data);
 
@@ -286,7 +286,7 @@ mlx5e_xmit_xdp_frame_mpwqe(struct mlx5e_xdpsq *sq, struct mlx5e_xmit_data *xdptx
 
 	mlx5e_xdp_mpwqe_add_dseg(sq, xdptxd, stats);
 
-	if (unlikely(mlx5e_xdp_mpqwe_is_full(session)))
+	if (unlikely(mlx5e_xdp_mpqwe_is_full(session, sq->max_sq_mpw_wqebbs)))
 		mlx5e_xdp_mpwqe_complete(sq);
 
 	mlx5e_xdpi_fifo_push(&sq->db.xdpi_fifo, xdpi);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.h b/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.h
index 8d991c3b7a50..c62f11d7ef6a 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.h
@@ -123,12 +123,13 @@ static inline bool mlx5e_xdp_get_inline_state(struct mlx5e_xdpsq *sq, bool cur)
 	return cur;
 }
 
-static inline bool mlx5e_xdp_mpqwe_is_full(struct mlx5e_tx_mpwqe *session)
+static inline bool mlx5e_xdp_mpqwe_is_full(struct mlx5e_tx_mpwqe *session, u8 max_sq_mpw_wqebbs)
 {
 	if (session->inline_on)
 		return session->ds_count + MLX5E_XDP_INLINE_WQE_MAX_DS_CNT >
-		       MLX5E_TX_MPW_MAX_NUM_DS;
-	return mlx5e_tx_mpwqe_is_full(session);
+		       max_sq_mpw_wqebbs * MLX5_SEND_WQEBB_NUM_DS;
+
+	return mlx5e_tx_mpwqe_is_full(session, max_sq_mpw_wqebbs);
 }
 
 struct mlx5e_xdp_wqe_info {
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 0a0e5bd77384..59427c5f5622 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -1167,6 +1167,7 @@ static int mlx5e_alloc_xdpsq(struct mlx5e_channel *c,
 			&c->priv->channel_stats[c->ix]->rq_xdpsq;
 	sq->max_sq_wqebbs = mlx5e_get_max_sq_wqebbs(mdev);
 	sq->stop_room = MLX5E_STOP_ROOM(sq->max_sq_wqebbs);
+	sq->max_sq_mpw_wqebbs = mlx5e_get_sw_max_sq_mpw_wqebbs(sq->max_sq_wqebbs);
 
 	param->wq.db_numa_node = cpu_to_node(c->cpu);
 	err = mlx5_wq_cyc_create(mdev, &param->wq, sqc_wq, wq, &sq->wq_ctrl);
@@ -1328,6 +1329,7 @@ static int mlx5e_alloc_txqsq(struct mlx5e_channel *c,
 	sq->min_inline_mode = params->tx_min_inline_mode;
 	sq->hw_mtu    = MLX5E_SW2HW_MTU(params, params->sw_mtu);
 	sq->max_sq_wqebbs = mlx5e_get_max_sq_wqebbs(mdev);
+	sq->max_sq_mpw_wqebbs = mlx5e_get_sw_max_sq_mpw_wqebbs(sq->max_sq_wqebbs);
 	INIT_WORK(&sq->recover_work, mlx5e_tx_err_cqe_work);
 	if (!MLX5_CAP_ETH(mdev, wqe_vlan_insert))
 		set_bit(MLX5E_SQ_STATE_VLAN_NEED_L2_INLINE, &sq->state);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index ee0a8f5206e3..91fdf957cd7c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -620,7 +620,7 @@ static int mlx5e_alloc_rx_hd_mpwqe(struct mlx5e_rq *rq)
 	struct mlx5e_icosq *sq = rq->icosq;
 	int i, err, max_klm_entries, len;
 
-	max_klm_entries = MLX5E_MAX_KLM_PER_WQE;
+	max_klm_entries = MLX5E_MAX_KLM_PER_WQE(rq->mdev);
 	klm_entries = bitmap_find_window(shampo->bitmap,
 					 shampo->hd_per_wqe,
 					 shampo->hd_per_wq, shampo->pi);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
index ee7ecb88adc1..9c91ef0e1ed2 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
@@ -544,7 +544,7 @@ static void mlx5e_tx_mpwqe_session_start(struct mlx5e_txqsq *sq,
 	struct mlx5e_tx_wqe *wqe;
 	u16 pi;
 
-	pi = mlx5e_txqsq_get_next_pi(sq, MLX5E_TX_MPW_MAX_WQEBBS);
+	pi = mlx5e_txqsq_get_next_pi(sq, sq->max_sq_mpw_wqebbs);
 	wqe = MLX5E_TX_FETCH_WQE(sq, pi);
 	net_prefetchw(wqe->data);
 
@@ -645,7 +645,7 @@ mlx5e_sq_xmit_mpwqe(struct mlx5e_txqsq *sq, struct sk_buff *skb,
 
 	mlx5e_tx_skb_update_hwts_flags(skb);
 
-	if (unlikely(mlx5e_tx_mpwqe_is_full(&sq->mpwqe))) {
+	if (unlikely(mlx5e_tx_mpwqe_is_full(&sq->mpwqe, sq->max_sq_mpw_wqebbs))) {
 		/* Might stop the queue and affect the retval of __netdev_tx_sent_queue. */
 		cseg = mlx5e_tx_mpwqe_session_complete(sq);
 
-- 
cgit v1.2.3


From befa41771f9e07f0a95b4a4a0d6b77fd334f4e44 Mon Sep 17 00:00:00 2001
From: Maxim Mikityanskiy <maximmi@nvidia.com>
Date: Tue, 25 Jan 2022 12:52:43 +0200
Subject: net/mlx5e: Cleanup of start/stop all queues

mlx5e_activate_priv_channels() and mlx5e_deactivate_priv_channels()
start and stop all netdev TX queues. This commit removes the unneeded
call to netif_tx_stop_all_queues and adds explanatory comments why these
operations are needed.

netif_tx_disable() does the same thing that netif_tx_stop_all_queues(),
but taking the TX lock, thus guaranteeing that ndo_start_xmit is not
running after return. That means that the netif_tx_stop_all_queues()
call is not really necessary.

The comments are improved: the TX watchdog timeout explanation is moved
to the start stage where it really belongs (it used to be in both
places, but was lost during some old refactoring) and rephrased in more
details; the explanation for stopping all TX queues is added.

Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 59427c5f5622..8507ebec1266 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -2716,6 +2716,11 @@ void mlx5e_activate_priv_channels(struct mlx5e_priv *priv)
 	mlx5e_activate_channels(&priv->channels);
 	mlx5e_qos_activate_queues(priv);
 	mlx5e_xdp_tx_enable(priv);
+
+	/* dev_watchdog() wants all TX queues to be started when the carrier is
+	 * OK, including the ones in range real_num_tx_queues..num_tx_queues-1.
+	 * Make it happy to avoid TX timeout false alarms.
+	 */
 	netif_tx_start_all_queues(priv->netdev);
 
 	if (mlx5e_is_vport_rep(priv))
@@ -2735,11 +2740,13 @@ void mlx5e_deactivate_priv_channels(struct mlx5e_priv *priv)
 	if (mlx5e_is_vport_rep(priv))
 		mlx5e_remove_sqs_fwd_rules(priv);
 
-	/* FIXME: This is a W/A only for tx timeout watch dog false alarm when
-	 * polling for inactive tx queues.
+	/* The results of ndo_select_queue are unreliable, while netdev config
+	 * is being changed (real_num_tx_queues, num_tc). Stop all queues to
+	 * prevent ndo_start_xmit from being called, so that it can assume that
+	 * the selected queue is always valid.
 	 */
-	netif_tx_stop_all_queues(priv->netdev);
 	netif_tx_disable(priv->netdev);
+
 	mlx5e_xdp_tx_disable(priv);
 	mlx5e_deactivate_channels(&priv->channels);
 }
-- 
cgit v1.2.3


From d08c6e2a4d0308a7922d7ef3b1b3af45d4096aad Mon Sep 17 00:00:00 2001
From: Maxim Mikityanskiy <maximmi@nvidia.com>
Date: Tue, 25 Jan 2022 12:52:48 +0200
Subject: net/mlx5e: Disable TX queues before registering the netdev

Normally, the queues are disabled when the channels are deactivated, and
enabled when the channels are activated. However, on register, the
channels are not active, but the queues are enabled by default. This
change fixes it, preventing mlx5e_xmit from running when the channels
are deactivated in the beginning.

Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 8507ebec1266..d84d9cdbdbd4 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -5359,6 +5359,7 @@ mlx5e_create_netdev(struct mlx5_core_dev *mdev, const struct mlx5e_profile *prof
 	}
 
 	netif_carrier_off(netdev);
+	netif_tx_disable(netdev);
 	dev_net_set(netdev, mlx5_core_net(mdev));
 
 	return netdev;
-- 
cgit v1.2.3


From 6ce204eac38723d3a24c7bf148221de933bba54d Mon Sep 17 00:00:00 2001
From: Maxim Mikityanskiy <maximmi@nvidia.com>
Date: Tue, 25 Jan 2022 12:52:49 +0200
Subject: net/mlx5e: Use a barrier after updating txq2sq

mlx5e_build_txq_maps updates txq2sq while TX queues are stopped. Add a
barrier to ensure that these changes are visible before the queues are
started and mlx5e_xmit reads from txq2sq.

This commit handles regular TX queues. Synchronization between HTB TX
queues and mlx5e_xmit is handled in the following commit.

Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index d84d9cdbdbd4..e64c3cb15ef6 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -2688,10 +2688,10 @@ static void mlx5e_build_txq_maps(struct mlx5e_priv *priv)
 	}
 
 	if (!priv->channels.ptp)
-		return;
+		goto out;
 
 	if (!test_bit(MLX5E_PTP_STATE_TX, priv->channels.ptp->state))
-		return;
+		goto out;
 
 	for (tc = 0; tc < num_tc; tc++) {
 		struct mlx5e_ptp *c = priv->channels.ptp;
@@ -2700,6 +2700,13 @@ static void mlx5e_build_txq_maps(struct mlx5e_priv *priv)
 		priv->txq2sq[sq->txq_ix] = sq;
 		priv->port_ptp_tc2realtxq[tc] = priv->num_tc_x_num_ch + tc;
 	}
+
+out:
+	/* Make the change to txq2sq visible before the queue is started.
+	 * As mlx5e_xmit runs under a spinlock, there is an implicit ACQUIRE,
+	 * which pairs with this barrier.
+	 */
+	smp_wmb();
 }
 
 static void mlx5e_update_num_tc_x_num_ch(struct mlx5e_priv *priv)
-- 
cgit v1.2.3


From 17c84cb46e33fcd682d5d5d6180162374cb24a34 Mon Sep 17 00:00:00 2001
From: Maxim Mikityanskiy <maximmi@nvidia.com>
Date: Tue, 25 Jan 2022 12:52:50 +0200
Subject: net/mlx5e: Sync txq2sq updates with mlx5e_xmit for HTB queues

This commit makes necessary changes to guarantee that txq2sq remains
stable while mlx5e_xmit is running. Proper synchronization is added for
HTB TX queues.

All updates to txq2sq are performed while the corresponding queue is
disabled (i.e. mlx5e_xmit doesn't run on that queue). smp_wmb after each
change guarantees that mlx5e_xmit can see the updated value after the
queue is enabled. Comments explaining this mechanism are added to
mlx5e_xmit.

When an HTB SQ can be deleted (after deleting an HTB node), synchronize
with RCU to wait for mlx5e_select_queue to finish and stop selecting
that queue, before we re-enable it to avoid TX timeout watchdog alarms.

Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en/qos.c | 24 ++++++++++++++++++++----
 drivers/net/ethernet/mellanox/mlx5/core/en_tx.c  | 13 +++++++++++++
 2 files changed, 33 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/qos.c b/drivers/net/ethernet/mellanox/mlx5/core/en/qos.c
index c1e07496c89c..ff45840581e3 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/qos.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/qos.c
@@ -50,7 +50,6 @@ static int mlx5e_find_unused_qos_qid(struct mlx5e_priv *priv)
 
 struct mlx5e_qos_node {
 	struct hlist_node hnode;
-	struct rcu_head rcu;
 	struct mlx5e_qos_node *parent;
 	u64 rate;
 	u32 bw_share;
@@ -132,7 +131,11 @@ static void mlx5e_sw_node_delete(struct mlx5e_priv *priv, struct mlx5e_qos_node
 		__clear_bit(node->qid, priv->htb.qos_used_qids);
 		mlx5e_update_tx_netdev_queues(priv);
 	}
-	kfree_rcu(node, rcu);
+	/* Make sure this qid is no longer selected by mlx5e_select_queue, so
+	 * that mlx5e_reactivate_qos_sq can safely restart the netdev TX queue.
+	 */
+	synchronize_net();
+	kfree(node);
 }
 
 /* TX datapath API */
@@ -273,10 +276,18 @@ err_free_sq:
 static void mlx5e_activate_qos_sq(struct mlx5e_priv *priv, struct mlx5e_qos_node *node)
 {
 	struct mlx5e_txqsq *sq;
+	u16 qid;
 
 	sq = mlx5e_get_qos_sq(priv, node->qid);
 
-	WRITE_ONCE(priv->txq2sq[mlx5e_qid_from_qos(&priv->channels, node->qid)], sq);
+	qid = mlx5e_qid_from_qos(&priv->channels, node->qid);
+
+	/* If it's a new queue, it will be marked as started at this point.
+	 * Stop it before updating txq2sq.
+	 */
+	mlx5e_tx_disable_queue(netdev_get_tx_queue(priv->netdev, qid));
+
+	priv->txq2sq[qid] = sq;
 
 	/* Make the change to txq2sq visible before the queue is started.
 	 * As mlx5e_xmit runs under a spinlock, there is an implicit ACQUIRE,
@@ -299,8 +310,13 @@ static void mlx5e_deactivate_qos_sq(struct mlx5e_priv *priv, u16 qid)
 	qos_dbg(priv->mdev, "Deactivate QoS SQ qid %u\n", qid);
 	mlx5e_deactivate_txqsq(sq);
 
-	/* The queue is disabled, no synchronization with datapath is needed. */
 	priv->txq2sq[mlx5e_qid_from_qos(&priv->channels, qid)] = NULL;
+
+	/* Make the change to txq2sq visible before the queue is started again.
+	 * As mlx5e_xmit runs under a spinlock, there is an implicit ACQUIRE,
+	 * which pairs with this barrier.
+	 */
+	smp_wmb();
 }
 
 static void mlx5e_close_qos_sq(struct mlx5e_priv *priv, u16 qid)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
index 9c91ef0e1ed2..726661774979 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
@@ -691,8 +691,21 @@ netdev_tx_t mlx5e_xmit(struct sk_buff *skb, struct net_device *dev)
 	struct mlx5e_txqsq *sq;
 	u16 pi;
 
+	/* All changes to txq2sq are performed in sync with mlx5e_xmit, when the
+	 * queue being changed is disabled, and smp_wmb guarantees that the
+	 * changes are visible before mlx5e_xmit tries to read from txq2sq. It
+	 * guarantees that the value of txq2sq[qid] doesn't change while
+	 * mlx5e_xmit is running on queue number qid. smb_wmb is paired with
+	 * HARD_TX_LOCK around ndo_start_xmit, which serves as an ACQUIRE.
+	 */
 	sq = priv->txq2sq[skb_get_queue_mapping(skb)];
 	if (unlikely(!sq)) {
+		/* Two cases when sq can be NULL:
+		 * 1. The HTB node is registered, and mlx5e_select_queue
+		 * selected its queue ID, but the SQ itself is not yet created.
+		 * 2. HTB SQ creation failed. Similar to the previous case, but
+		 * the SQ won't be created.
+		 */
 		dev_kfree_skb_any(skb);
 		return NETDEV_TX_OK;
 	}
-- 
cgit v1.2.3


From 8bf30be75069d6080659de9a28565c048f6cef9b Mon Sep 17 00:00:00 2001
From: Maxim Mikityanskiy <maximmi@nvidia.com>
Date: Tue, 25 Jan 2022 12:52:51 +0200
Subject: net/mlx5e: Introduce select queue parameters

ndo_select_queue can be called at any time, and there is no way to stop
the kernel from calling it to synchronize with configuration changes
(real_num_tx_queues, num_tc). This commit introduces an internal way in
mlx5e to sync mlx5e_select_queue() with these changes. The configuration
needed by this function is stored in a struct mlx5e_selq_params, which
is modified and accessed in an atomic way using RCU methods. The whole
ndo_select_queue is called under an RCU lock, providing the necessary
guarantees.

The parameters stored in the new struct mlx5e_selq_params should only be
used from inside mlx5e_select_queue. It's the minimal set of parameters
needed for mlx5e_select_queue to do its job efficiently, derived from
parameters stored elsewhere. That means that when the configuration
change, mlx5e_selq_params may need to be updated. In such cases, the
mlx5e_selq_prepare/mlx5e_selq_apply API should be used.

struct mlx5e_selq contains two slots for the params: active and standby.
mlx5e_selq_prepare updates the standby slot, and mlx5e_selq_apply swaps
the slots in a safe atomic way using the RCU API. It integrates well
with the open/activate stages of the configuration change flow.

Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/Makefile  |  2 +-
 drivers/net/ethernet/mellanox/mlx5/core/en.h      |  2 +
 drivers/net/ethernet/mellanox/mlx5/core/en/qos.c  | 12 ++-
 drivers/net/ethernet/mellanox/mlx5/core/en/selq.c | 95 +++++++++++++++++++++++
 drivers/net/ethernet/mellanox/mlx5/core/en/selq.h | 26 +++++++
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 33 +++++++-
 6 files changed, 165 insertions(+), 5 deletions(-)
 create mode 100644 drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
 create mode 100644 drivers/net/ethernet/mellanox/mlx5/core/en/selq.h

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/Makefile b/drivers/net/ethernet/mellanox/mlx5/core/Makefile
index fcfd38fa9e6c..a7170ab3af97 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/Makefile
+++ b/drivers/net/ethernet/mellanox/mlx5/core/Makefile
@@ -28,7 +28,7 @@ mlx5_core-$(CONFIG_MLX5_CORE_EN) += en/rqt.o en/tir.o en/rss.o en/rx_res.o \
 		en_selftest.o en/port.o en/monitor_stats.o en/health.o \
 		en/reporter_tx.o en/reporter_rx.o en/params.o en/xsk/pool.o \
 		en/xsk/setup.o en/xsk/rx.o en/xsk/tx.o en/devlink.o en/ptp.o \
-		en/qos.o en/trap.o en/fs_tt_redirect.o
+		en/qos.o en/trap.o en/fs_tt_redirect.o en/selq.o
 
 #
 # Netdev extra
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index 99529e238fc4..e1c2f296867a 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -59,6 +59,7 @@
 #include "lib/hv_vhca.h"
 #include "lib/clock.h"
 #include "en/rx_res.h"
+#include "en/selq.h"
 
 extern const struct net_device_ops mlx5e_netdev_ops;
 struct page_pool;
@@ -908,6 +909,7 @@ struct mlx5e_trap;
 
 struct mlx5e_priv {
 	/* priv data path fields - start */
+	struct mlx5e_selq selq;
 	struct mlx5e_txqsq **txq2sq;
 	int **channel_tc2realtxq;
 	int port_ptp_tc2realtxq[MLX5E_MAX_NUM_TC];
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/qos.c b/drivers/net/ethernet/mellanox/mlx5/core/en/qos.c
index ff45840581e3..ccfc8ae2fa71 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/qos.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/qos.c
@@ -501,9 +501,11 @@ int mlx5e_htb_root_add(struct mlx5e_priv *priv, u16 htb_maj_id, u16 htb_defcls,
 
 	opened = test_bit(MLX5E_STATE_OPENED, &priv->state);
 	if (opened) {
+		mlx5e_selq_prepare(&priv->selq, &priv->channels.params, true);
+
 		err = mlx5e_qos_alloc_queues(priv, &priv->channels);
 		if (err)
-			return err;
+			goto err_cancel_selq;
 	}
 
 	root = mlx5e_sw_node_create_root(priv);
@@ -524,6 +526,9 @@ int mlx5e_htb_root_add(struct mlx5e_priv *priv, u16 htb_maj_id, u16 htb_defcls,
 	 */
 	smp_store_release(&priv->htb.maj_id, htb_maj_id);
 
+	if (opened)
+		mlx5e_selq_apply(&priv->selq);
+
 	return 0;
 
 err_sw_node_delete:
@@ -532,6 +537,8 @@ err_sw_node_delete:
 err_free_queues:
 	if (opened)
 		mlx5e_qos_close_all_queues(&priv->channels);
+err_cancel_selq:
+	mlx5e_selq_cancel(&priv->selq);
 	return err;
 }
 
@@ -542,6 +549,9 @@ int mlx5e_htb_root_del(struct mlx5e_priv *priv)
 
 	qos_dbg(priv->mdev, "TC_HTB_DESTROY\n");
 
+	mlx5e_selq_prepare(&priv->selq, &priv->channels.params, false);
+	mlx5e_selq_apply(&priv->selq);
+
 	WRITE_ONCE(priv->htb.maj_id, 0);
 	synchronize_rcu(); /* Sync with mlx5e_select_htb_queue and TX data path. */
 
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
new file mode 100644
index 000000000000..50ea58a3cc94
--- /dev/null
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
@@ -0,0 +1,95 @@
+// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
+/* Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. */
+
+#include "selq.h"
+#include <linux/slab.h>
+#include <linux/netdevice.h>
+#include "en.h"
+
+struct mlx5e_selq_params {
+	unsigned int num_regular_queues;
+	unsigned int num_channels;
+	unsigned int num_tcs;
+	bool is_htb;
+	bool is_ptp;
+};
+
+int mlx5e_selq_init(struct mlx5e_selq *selq, struct mutex *state_lock)
+{
+	struct mlx5e_selq_params *init_params;
+
+	selq->state_lock = state_lock;
+
+	selq->standby = kvzalloc(sizeof(*selq->standby), GFP_KERNEL);
+	if (!selq->standby)
+		return -ENOMEM;
+
+	init_params = kvzalloc(sizeof(*selq->active), GFP_KERNEL);
+	if (!init_params) {
+		kvfree(selq->standby);
+		selq->standby = NULL;
+		return -ENOMEM;
+	}
+	/* Assign dummy values, so that mlx5e_select_queue won't crash. */
+	*init_params = (struct mlx5e_selq_params) {
+		.num_regular_queues = 1,
+		.num_channels = 1,
+		.num_tcs = 1,
+		.is_htb = false,
+		.is_ptp = false,
+	};
+	rcu_assign_pointer(selq->active, init_params);
+
+	return 0;
+}
+
+void mlx5e_selq_cleanup(struct mlx5e_selq *selq)
+{
+	WARN_ON_ONCE(selq->is_prepared);
+
+	kvfree(selq->standby);
+	selq->standby = NULL;
+	selq->is_prepared = true;
+
+	mlx5e_selq_apply(selq);
+
+	kvfree(selq->standby);
+	selq->standby = NULL;
+}
+
+void mlx5e_selq_prepare(struct mlx5e_selq *selq, struct mlx5e_params *params, bool htb)
+{
+	lockdep_assert_held(selq->state_lock);
+	WARN_ON_ONCE(selq->is_prepared);
+
+	selq->is_prepared = true;
+
+	selq->standby->num_channels = params->num_channels;
+	selq->standby->num_tcs = mlx5e_get_dcb_num_tc(params);
+	selq->standby->num_regular_queues =
+		selq->standby->num_channels * selq->standby->num_tcs;
+	selq->standby->is_htb = htb;
+	selq->standby->is_ptp = MLX5E_GET_PFLAG(params, MLX5E_PFLAG_TX_PORT_TS);
+}
+
+void mlx5e_selq_apply(struct mlx5e_selq *selq)
+{
+	struct mlx5e_selq_params *old_params;
+
+	WARN_ON_ONCE(!selq->is_prepared);
+
+	selq->is_prepared = false;
+
+	old_params = rcu_replace_pointer(selq->active, selq->standby,
+					 lockdep_is_held(selq->state_lock));
+	synchronize_net(); /* Wait until ndo_select_queue starts emitting correct values. */
+	selq->standby = old_params;
+}
+
+void mlx5e_selq_cancel(struct mlx5e_selq *selq)
+{
+	lockdep_assert_held(selq->state_lock);
+	WARN_ON_ONCE(!selq->is_prepared);
+
+	selq->is_prepared = false;
+}
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.h b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.h
new file mode 100644
index 000000000000..2648c23e8238
--- /dev/null
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.h
@@ -0,0 +1,26 @@
+/* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
+/* Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved. */
+
+#ifndef __MLX5_EN_SELQ_H__
+#define __MLX5_EN_SELQ_H__
+
+#include <linux/kernel.h>
+
+struct mlx5e_selq_params;
+
+struct mlx5e_selq {
+	struct mlx5e_selq_params __rcu *active;
+	struct mlx5e_selq_params *standby;
+	struct mutex *state_lock; /* points to priv->state_lock */
+	bool is_prepared;
+};
+
+struct mlx5e_params;
+
+int mlx5e_selq_init(struct mlx5e_selq *selq, struct mutex *state_lock);
+void mlx5e_selq_cleanup(struct mlx5e_selq *selq);
+void mlx5e_selq_prepare(struct mlx5e_selq *selq, struct mlx5e_params *params, bool htb);
+void mlx5e_selq_apply(struct mlx5e_selq *selq);
+void mlx5e_selq_cancel(struct mlx5e_selq *selq);
+
+#endif /* __MLX5_EN_SELQ_H__ */
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index e64c3cb15ef6..02f0ad653ece 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -2813,6 +2813,7 @@ static int mlx5e_switch_priv_channels(struct mlx5e_priv *priv,
 	mlx5e_close_channels(&old_chs);
 	priv->profile->update_rx(priv);
 
+	mlx5e_selq_apply(&priv->selq);
 out:
 	mlx5e_activate_priv_channels(priv);
 
@@ -2836,13 +2837,24 @@ int mlx5e_safe_switch_params(struct mlx5e_priv *priv,
 		return mlx5e_switch_priv_params(priv, params, preactivate, context);
 
 	new_chs.params = *params;
+
+	mlx5e_selq_prepare(&priv->selq, &new_chs.params, !!priv->htb.maj_id);
+
 	err = mlx5e_open_channels(priv, &new_chs);
 	if (err)
-		return err;
+		goto err_cancel_selq;
+
 	err = mlx5e_switch_priv_channels(priv, &new_chs, preactivate, context);
 	if (err)
-		mlx5e_close_channels(&new_chs);
+		goto err_close;
 
+	return 0;
+
+err_close:
+	mlx5e_close_channels(&new_chs);
+
+err_cancel_selq:
+	mlx5e_selq_cancel(&priv->selq);
 	return err;
 }
 
@@ -2882,6 +2894,8 @@ int mlx5e_open_locked(struct net_device *netdev)
 	struct mlx5e_priv *priv = netdev_priv(netdev);
 	int err;
 
+	mlx5e_selq_prepare(&priv->selq, &priv->channels.params, !!priv->htb.maj_id);
+
 	set_bit(MLX5E_STATE_OPENED, &priv->state);
 
 	err = mlx5e_open_channels(priv, &priv->channels);
@@ -2889,6 +2903,7 @@ int mlx5e_open_locked(struct net_device *netdev)
 		goto err_clear_state_opened_flag;
 
 	priv->profile->update_rx(priv);
+	mlx5e_selq_apply(&priv->selq);
 	mlx5e_activate_priv_channels(priv);
 	mlx5e_apply_traps(priv, true);
 	if (priv->profile->update_carrier)
@@ -2899,6 +2914,7 @@ int mlx5e_open_locked(struct net_device *netdev)
 
 err_clear_state_opened_flag:
 	clear_bit(MLX5E_STATE_OPENED, &priv->state);
+	mlx5e_selq_cancel(&priv->selq);
 	return err;
 }
 
@@ -5215,6 +5231,7 @@ int mlx5e_priv_init(struct mlx5e_priv *priv,
 		    struct mlx5_core_dev *mdev)
 {
 	int nch, num_txqs, node, i;
+	int err;
 
 	num_txqs = netdev->num_tx_queues;
 	nch = mlx5e_calc_max_nch(mdev, netdev, profile);
@@ -5231,6 +5248,11 @@ int mlx5e_priv_init(struct mlx5e_priv *priv,
 		return -ENOMEM;
 
 	mutex_init(&priv->state_lock);
+
+	err = mlx5e_selq_init(&priv->selq, &priv->state_lock);
+	if (err)
+		goto err_free_cpumask;
+
 	hash_init(priv->htb.qos_tc2node);
 	INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work);
 	INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work);
@@ -5239,7 +5261,7 @@ int mlx5e_priv_init(struct mlx5e_priv *priv,
 
 	priv->wq = create_singlethread_workqueue("mlx5e");
 	if (!priv->wq)
-		goto err_free_cpumask;
+		goto err_free_selq;
 
 	priv->txq2sq = kcalloc_node(num_txqs, sizeof(*priv->txq2sq), GFP_KERNEL, node);
 	if (!priv->txq2sq)
@@ -5279,6 +5301,8 @@ err_free_txq2sq:
 	kfree(priv->txq2sq);
 err_destroy_workqueue:
 	destroy_workqueue(priv->wq);
+err_free_selq:
+	mlx5e_selq_cleanup(&priv->selq);
 err_free_cpumask:
 	free_cpumask_var(priv->scratchpad.cpumask);
 	return -ENOMEM;
@@ -5301,6 +5325,9 @@ void mlx5e_priv_cleanup(struct mlx5e_priv *priv)
 	kfree(priv->tx_rates);
 	kfree(priv->txq2sq);
 	destroy_workqueue(priv->wq);
+	mutex_lock(&priv->state_lock);
+	mlx5e_selq_cleanup(&priv->selq);
+	mutex_unlock(&priv->state_lock);
 	free_cpumask_var(priv->scratchpad.cpumask);
 
 	for (i = 0; i < priv->htb.max_qos_sqs; i++)
-- 
cgit v1.2.3


From 6b23f6ab86a4f3a608b9bfa98616e99a34bb650c Mon Sep 17 00:00:00 2001
From: Maxim Mikityanskiy <maximmi@nvidia.com>
Date: Tue, 25 Jan 2022 12:52:52 +0200
Subject: net/mlx5e: Move mlx5e_select_queue to en/selq.c

This commit moves mlx5e_select_queue and all stuff related to
ndo_select_queue to en/selq.c to put all stuff working with selq into a
separate file.

Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en/selq.c | 112 ++++++++++++++++++++++
 drivers/net/ethernet/mellanox/mlx5/core/en/selq.h |   5 +
 drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h |   2 -
 drivers/net/ethernet/mellanox/mlx5/core/en_tx.c   | 111 ---------------------
 4 files changed, 117 insertions(+), 113 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
index 50ea58a3cc94..297ba7946753 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
@@ -5,6 +5,7 @@
 #include <linux/slab.h>
 #include <linux/netdevice.h>
 #include "en.h"
+#include "en/ptp.h"
 
 struct mlx5e_selq_params {
 	unsigned int num_regular_queues;
@@ -93,3 +94,114 @@ void mlx5e_selq_cancel(struct mlx5e_selq *selq)
 
 	selq->is_prepared = false;
 }
+
+#ifdef CONFIG_MLX5_CORE_EN_DCB
+static int mlx5e_get_dscp_up(struct mlx5e_priv *priv, struct sk_buff *skb)
+{
+	int dscp_cp = 0;
+
+	if (skb->protocol == htons(ETH_P_IP))
+		dscp_cp = ipv4_get_dsfield(ip_hdr(skb)) >> 2;
+	else if (skb->protocol == htons(ETH_P_IPV6))
+		dscp_cp = ipv6_get_dsfield(ipv6_hdr(skb)) >> 2;
+
+	return priv->dcbx_dp.dscp2prio[dscp_cp];
+}
+#endif
+
+static u16 mlx5e_select_ptpsq(struct net_device *dev, struct sk_buff *skb)
+{
+	struct mlx5e_priv *priv = netdev_priv(dev);
+	int up = 0;
+
+	if (!netdev_get_num_tc(dev))
+		goto return_txq;
+
+#ifdef CONFIG_MLX5_CORE_EN_DCB
+	if (priv->dcbx_dp.trust_state == MLX5_QPTS_TRUST_DSCP)
+		up = mlx5e_get_dscp_up(priv, skb);
+	else
+#endif
+		if (skb_vlan_tag_present(skb))
+			up = skb_vlan_tag_get_prio(skb);
+
+return_txq:
+	return priv->port_ptp_tc2realtxq[up];
+}
+
+static int mlx5e_select_htb_queue(struct mlx5e_priv *priv, struct sk_buff *skb,
+				  u16 htb_maj_id)
+{
+	u16 classid;
+
+	if ((TC_H_MAJ(skb->priority) >> 16) == htb_maj_id)
+		classid = TC_H_MIN(skb->priority);
+	else
+		classid = READ_ONCE(priv->htb.defcls);
+
+	if (!classid)
+		return 0;
+
+	return mlx5e_get_txq_by_classid(priv, classid);
+}
+
+u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
+		       struct net_device *sb_dev)
+{
+	struct mlx5e_priv *priv = netdev_priv(dev);
+	int num_tc_x_num_ch;
+	int txq_ix;
+	int up = 0;
+	int ch_ix;
+
+	/* Sync with mlx5e_update_num_tc_x_num_ch - avoid refetching. */
+	num_tc_x_num_ch = READ_ONCE(priv->num_tc_x_num_ch);
+	if (unlikely(dev->real_num_tx_queues > num_tc_x_num_ch)) {
+		struct mlx5e_ptp *ptp_channel;
+
+		/* Order maj_id before defcls - pairs with mlx5e_htb_root_add. */
+		u16 htb_maj_id = smp_load_acquire(&priv->htb.maj_id);
+
+		if (unlikely(htb_maj_id)) {
+			txq_ix = mlx5e_select_htb_queue(priv, skb, htb_maj_id);
+			if (txq_ix > 0)
+				return txq_ix;
+		}
+
+		ptp_channel = READ_ONCE(priv->channels.ptp);
+		if (unlikely(ptp_channel &&
+			     test_bit(MLX5E_PTP_STATE_TX, ptp_channel->state) &&
+			     mlx5e_use_ptpsq(skb)))
+			return mlx5e_select_ptpsq(dev, skb);
+
+		txq_ix = netdev_pick_tx(dev, skb, NULL);
+		/* Fix netdev_pick_tx() not to choose ptp_channel and HTB txqs.
+		 * If they are selected, switch to regular queues.
+		 * Driver to select these queues only at mlx5e_select_ptpsq()
+		 * and mlx5e_select_htb_queue().
+		 */
+		if (unlikely(txq_ix >= num_tc_x_num_ch))
+			txq_ix %= num_tc_x_num_ch;
+	} else {
+		txq_ix = netdev_pick_tx(dev, skb, NULL);
+	}
+
+	if (!netdev_get_num_tc(dev))
+		return txq_ix;
+
+#ifdef CONFIG_MLX5_CORE_EN_DCB
+	if (priv->dcbx_dp.trust_state == MLX5_QPTS_TRUST_DSCP)
+		up = mlx5e_get_dscp_up(priv, skb);
+	else
+#endif
+		if (skb_vlan_tag_present(skb))
+			up = skb_vlan_tag_get_prio(skb);
+
+	/* Normalize any picked txq_ix to [0, num_channels),
+	 * So we can return a txq_ix that matches the channel and
+	 * packet UP.
+	 */
+	ch_ix = priv->txq2sq[txq_ix]->ch_ix;
+
+	return priv->channel_tc2realtxq[ch_ix][up];
+}
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.h b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.h
index 2648c23e8238..b1c73b509f6b 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.h
@@ -16,6 +16,8 @@ struct mlx5e_selq {
 };
 
 struct mlx5e_params;
+struct net_device;
+struct sk_buff;
 
 int mlx5e_selq_init(struct mlx5e_selq *selq, struct mutex *state_lock);
 void mlx5e_selq_cleanup(struct mlx5e_selq *selq);
@@ -23,4 +25,7 @@ void mlx5e_selq_prepare(struct mlx5e_selq *selq, struct mlx5e_params *params, bo
 void mlx5e_selq_apply(struct mlx5e_selq *selq);
 void mlx5e_selq_cancel(struct mlx5e_selq *selq);
 
+u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
+		       struct net_device *sb_dev);
+
 #endif /* __MLX5_EN_SELQ_H__ */
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h b/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h
index 1c48cfad9dd7..210d23bf3701 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h
@@ -55,8 +55,6 @@ void mlx5e_free_rx_descs(struct mlx5e_rq *rq);
 void mlx5e_free_rx_in_progress_descs(struct mlx5e_rq *rq);
 
 /* TX */
-u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
-		       struct net_device *sb_dev);
 netdev_tx_t mlx5e_xmit(struct sk_buff *skb, struct net_device *dev);
 bool mlx5e_poll_tx_cq(struct mlx5e_cq *cq, int napi_budget);
 void mlx5e_free_txqsq_descs(struct mlx5e_txqsq *sq);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
index 726661774979..2dc48406cd08 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
@@ -53,117 +53,6 @@ static void mlx5e_dma_unmap_wqe_err(struct mlx5e_txqsq *sq, u8 num_dma)
 	}
 }
 
-#ifdef CONFIG_MLX5_CORE_EN_DCB
-static inline int mlx5e_get_dscp_up(struct mlx5e_priv *priv, struct sk_buff *skb)
-{
-	int dscp_cp = 0;
-
-	if (skb->protocol == htons(ETH_P_IP))
-		dscp_cp = ipv4_get_dsfield(ip_hdr(skb)) >> 2;
-	else if (skb->protocol == htons(ETH_P_IPV6))
-		dscp_cp = ipv6_get_dsfield(ipv6_hdr(skb)) >> 2;
-
-	return priv->dcbx_dp.dscp2prio[dscp_cp];
-}
-#endif
-
-static u16 mlx5e_select_ptpsq(struct net_device *dev, struct sk_buff *skb)
-{
-	struct mlx5e_priv *priv = netdev_priv(dev);
-	int up = 0;
-
-	if (!netdev_get_num_tc(dev))
-		goto return_txq;
-
-#ifdef CONFIG_MLX5_CORE_EN_DCB
-	if (priv->dcbx_dp.trust_state == MLX5_QPTS_TRUST_DSCP)
-		up = mlx5e_get_dscp_up(priv, skb);
-	else
-#endif
-		if (skb_vlan_tag_present(skb))
-			up = skb_vlan_tag_get_prio(skb);
-
-return_txq:
-	return priv->port_ptp_tc2realtxq[up];
-}
-
-static int mlx5e_select_htb_queue(struct mlx5e_priv *priv, struct sk_buff *skb,
-				  u16 htb_maj_id)
-{
-	u16 classid;
-
-	if ((TC_H_MAJ(skb->priority) >> 16) == htb_maj_id)
-		classid = TC_H_MIN(skb->priority);
-	else
-		classid = READ_ONCE(priv->htb.defcls);
-
-	if (!classid)
-		return 0;
-
-	return mlx5e_get_txq_by_classid(priv, classid);
-}
-
-u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
-		       struct net_device *sb_dev)
-{
-	struct mlx5e_priv *priv = netdev_priv(dev);
-	int num_tc_x_num_ch;
-	int txq_ix;
-	int up = 0;
-	int ch_ix;
-
-	/* Sync with mlx5e_update_num_tc_x_num_ch - avoid refetching. */
-	num_tc_x_num_ch = READ_ONCE(priv->num_tc_x_num_ch);
-	if (unlikely(dev->real_num_tx_queues > num_tc_x_num_ch)) {
-		struct mlx5e_ptp *ptp_channel;
-
-		/* Order maj_id before defcls - pairs with mlx5e_htb_root_add. */
-		u16 htb_maj_id = smp_load_acquire(&priv->htb.maj_id);
-
-		if (unlikely(htb_maj_id)) {
-			txq_ix = mlx5e_select_htb_queue(priv, skb, htb_maj_id);
-			if (txq_ix > 0)
-				return txq_ix;
-		}
-
-		ptp_channel = READ_ONCE(priv->channels.ptp);
-		if (unlikely(ptp_channel &&
-			     test_bit(MLX5E_PTP_STATE_TX, ptp_channel->state) &&
-			     mlx5e_use_ptpsq(skb)))
-			return mlx5e_select_ptpsq(dev, skb);
-
-		txq_ix = netdev_pick_tx(dev, skb, NULL);
-		/* Fix netdev_pick_tx() not to choose ptp_channel and HTB txqs.
-		 * If they are selected, switch to regular queues.
-		 * Driver to select these queues only at mlx5e_select_ptpsq()
-		 * and mlx5e_select_htb_queue().
-		 */
-		if (unlikely(txq_ix >= num_tc_x_num_ch))
-			txq_ix %= num_tc_x_num_ch;
-	} else {
-		txq_ix = netdev_pick_tx(dev, skb, NULL);
-	}
-
-	if (!netdev_get_num_tc(dev))
-		return txq_ix;
-
-#ifdef CONFIG_MLX5_CORE_EN_DCB
-	if (priv->dcbx_dp.trust_state == MLX5_QPTS_TRUST_DSCP)
-		up = mlx5e_get_dscp_up(priv, skb);
-	else
-#endif
-		if (skb_vlan_tag_present(skb))
-			up = skb_vlan_tag_get_prio(skb);
-
-	/* Normalize any picked txq_ix to [0, num_channels),
-	 * So we can return a txq_ix that matches the channel and
-	 * packet UP.
-	 */
-	ch_ix = priv->txq2sq[txq_ix]->ch_ix;
-
-	return priv->channel_tc2realtxq[ch_ix][up];
-}
-
 static inline int mlx5e_skb_l2_header_offset(struct sk_buff *skb)
 {
 #define MLX5E_MIN_INLINE (ETH_HLEN + VLAN_HLEN)
-- 
cgit v1.2.3


From 3ab45777a27c5f53a55a2c353b70648449ca7a33 Mon Sep 17 00:00:00 2001
From: Maxim Mikityanskiy <maximmi@nvidia.com>
Date: Tue, 25 Jan 2022 12:52:54 +0200
Subject: net/mlx5e: Use select queue parameters to sync with control flow

Start using the select queue parameters introduced in the previous
commit to have proper synchronization with changing the configuration
(such as number of channels and queues). It ensures that the state that
mlx5e_select_queue() sees is always consistent and stays the same while
the function is running. Also it allows mlx5e_select_queue to stop using
data structures that weren't synchronized properly: txq2sq,
channel_tc2realtxq, port_ptp_tc2realtxq. The last two are removed
completely, as they were used only in mlx5e_select_queue.

Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en.h      |  3 --
 drivers/net/ethernet/mellanox/mlx5/core/en/qos.c  |  6 ++-
 drivers/net/ethernet/mellanox/mlx5/core/en/selq.c | 52 +++++++++++------------
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 39 +----------------
 drivers/net/ethernet/mellanox/mlx5/core/en_rep.c  |  5 ---
 5 files changed, 32 insertions(+), 73 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index e1c2f296867a..e29ac77e9bec 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -911,8 +911,6 @@ struct mlx5e_priv {
 	/* priv data path fields - start */
 	struct mlx5e_selq selq;
 	struct mlx5e_txqsq **txq2sq;
-	int **channel_tc2realtxq;
-	int port_ptp_tc2realtxq[MLX5E_MAX_NUM_TC];
 #ifdef CONFIG_MLX5_CORE_EN_DCB
 	struct mlx5e_dcbx_dp       dcbx_dp;
 #endif
@@ -955,7 +953,6 @@ struct mlx5e_priv {
 	u16                        drop_rq_q_counter;
 	struct notifier_block      events_nb;
 	struct notifier_block      blocking_events_nb;
-	int                        num_tc_x_num_ch;
 
 	struct udp_tunnel_nic_info nic_info;
 #ifdef CONFIG_MLX5_CORE_EN_DCB
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/qos.c b/drivers/net/ethernet/mellanox/mlx5/core/en/qos.c
index ccfc8ae2fa71..9db677e9ca9c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/qos.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/qos.c
@@ -549,11 +549,15 @@ int mlx5e_htb_root_del(struct mlx5e_priv *priv)
 
 	qos_dbg(priv->mdev, "TC_HTB_DESTROY\n");
 
+	/* Wait until real_num_tx_queues is updated for mlx5e_select_queue,
+	 * so that we can safely switch to its non-HTB non-PTP fastpath.
+	 */
+	synchronize_net();
+
 	mlx5e_selq_prepare(&priv->selq, &priv->channels.params, false);
 	mlx5e_selq_apply(&priv->selq);
 
 	WRITE_ONCE(priv->htb.maj_id, 0);
-	synchronize_rcu(); /* Sync with mlx5e_select_htb_queue and TX data path. */
 
 	root = mlx5e_sw_node_find(priv, MLX5E_HTB_CLASSID_ROOT);
 	if (!root) {
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
index 297ba7946753..a0bed47fd392 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
@@ -4,6 +4,7 @@
 #include "selq.h"
 #include <linux/slab.h>
 #include <linux/netdevice.h>
+#include <linux/rcupdate.h>
 #include "en.h"
 #include "en/ptp.h"
 
@@ -109,12 +110,13 @@ static int mlx5e_get_dscp_up(struct mlx5e_priv *priv, struct sk_buff *skb)
 }
 #endif
 
-static u16 mlx5e_select_ptpsq(struct net_device *dev, struct sk_buff *skb)
+static u16 mlx5e_select_ptpsq(struct net_device *dev, struct sk_buff *skb,
+			      struct mlx5e_selq_params *selq)
 {
 	struct mlx5e_priv *priv = netdev_priv(dev);
 	int up = 0;
 
-	if (!netdev_get_num_tc(dev))
+	if (selq->num_tcs <= 1)
 		goto return_txq;
 
 #ifdef CONFIG_MLX5_CORE_EN_DCB
@@ -126,15 +128,15 @@ static u16 mlx5e_select_ptpsq(struct net_device *dev, struct sk_buff *skb)
 			up = skb_vlan_tag_get_prio(skb);
 
 return_txq:
-	return priv->port_ptp_tc2realtxq[up];
+	return selq->num_regular_queues + up;
 }
 
-static int mlx5e_select_htb_queue(struct mlx5e_priv *priv, struct sk_buff *skb,
-				  u16 htb_maj_id)
+static int mlx5e_select_htb_queue(struct mlx5e_priv *priv, struct sk_buff *skb)
 {
 	u16 classid;
 
-	if ((TC_H_MAJ(skb->priority) >> 16) == htb_maj_id)
+	/* Order maj_id before defcls - pairs with mlx5e_htb_root_add. */
+	if ((TC_H_MAJ(skb->priority) >> 16) == smp_load_acquire(&priv->htb.maj_id))
 		classid = TC_H_MIN(skb->priority);
 	else
 		classid = READ_ONCE(priv->htb.defcls);
@@ -149,30 +151,28 @@ u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
 		       struct net_device *sb_dev)
 {
 	struct mlx5e_priv *priv = netdev_priv(dev);
-	int num_tc_x_num_ch;
+	struct mlx5e_selq_params *selq;
 	int txq_ix;
 	int up = 0;
-	int ch_ix;
 
-	/* Sync with mlx5e_update_num_tc_x_num_ch - avoid refetching. */
-	num_tc_x_num_ch = READ_ONCE(priv->num_tc_x_num_ch);
-	if (unlikely(dev->real_num_tx_queues > num_tc_x_num_ch)) {
-		struct mlx5e_ptp *ptp_channel;
+	selq = rcu_dereference_bh(priv->selq.active);
 
-		/* Order maj_id before defcls - pairs with mlx5e_htb_root_add. */
-		u16 htb_maj_id = smp_load_acquire(&priv->htb.maj_id);
+	/* This is a workaround needed only for the mlx5e_netdev_change_profile
+	 * flow that zeroes out the whole priv without unregistering the netdev
+	 * and without preventing ndo_select_queue from being called.
+	 */
+	if (unlikely(!selq))
+		return 0;
 
-		if (unlikely(htb_maj_id)) {
-			txq_ix = mlx5e_select_htb_queue(priv, skb, htb_maj_id);
+	if (unlikely(selq->is_ptp || selq->is_htb)) {
+		if (unlikely(selq->is_htb)) {
+			txq_ix = mlx5e_select_htb_queue(priv, skb);
 			if (txq_ix > 0)
 				return txq_ix;
 		}
 
-		ptp_channel = READ_ONCE(priv->channels.ptp);
-		if (unlikely(ptp_channel &&
-			     test_bit(MLX5E_PTP_STATE_TX, ptp_channel->state) &&
-			     mlx5e_use_ptpsq(skb)))
-			return mlx5e_select_ptpsq(dev, skb);
+		if (unlikely(selq->is_ptp && mlx5e_use_ptpsq(skb)))
+			return mlx5e_select_ptpsq(dev, skb, selq);
 
 		txq_ix = netdev_pick_tx(dev, skb, NULL);
 		/* Fix netdev_pick_tx() not to choose ptp_channel and HTB txqs.
@@ -180,13 +180,13 @@ u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
 		 * Driver to select these queues only at mlx5e_select_ptpsq()
 		 * and mlx5e_select_htb_queue().
 		 */
-		if (unlikely(txq_ix >= num_tc_x_num_ch))
-			txq_ix %= num_tc_x_num_ch;
+		if (unlikely(txq_ix >= selq->num_regular_queues))
+			txq_ix %= selq->num_regular_queues;
 	} else {
 		txq_ix = netdev_pick_tx(dev, skb, NULL);
 	}
 
-	if (!netdev_get_num_tc(dev))
+	if (selq->num_tcs <= 1)
 		return txq_ix;
 
 #ifdef CONFIG_MLX5_CORE_EN_DCB
@@ -201,7 +201,5 @@ u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
 	 * So we can return a txq_ix that matches the channel and
 	 * packet UP.
 	 */
-	ch_ix = priv->txq2sq[txq_ix]->ch_ix;
-
-	return priv->channel_tc2realtxq[ch_ix][up];
+	return txq_ix % selq->num_channels + up * selq->num_channels;
 }
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 02f0ad653ece..b157c7aac4ca 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -2683,7 +2683,6 @@ static void mlx5e_build_txq_maps(struct mlx5e_priv *priv)
 			struct mlx5e_txqsq *sq = &c->sq[tc];
 
 			priv->txq2sq[sq->txq_ix] = sq;
-			priv->channel_tc2realtxq[i][tc] = i + tc * ch;
 		}
 	}
 
@@ -2698,7 +2697,6 @@ static void mlx5e_build_txq_maps(struct mlx5e_priv *priv)
 		struct mlx5e_txqsq *sq = &c->ptpsq[tc].txqsq;
 
 		priv->txq2sq[sq->txq_ix] = sq;
-		priv->port_ptp_tc2realtxq[tc] = priv->num_tc_x_num_ch + tc;
 	}
 
 out:
@@ -2709,16 +2707,8 @@ out:
 	smp_wmb();
 }
 
-static void mlx5e_update_num_tc_x_num_ch(struct mlx5e_priv *priv)
-{
-	/* Sync with mlx5e_select_queue. */
-	WRITE_ONCE(priv->num_tc_x_num_ch,
-		   mlx5e_get_dcb_num_tc(&priv->channels.params) * priv->channels.num);
-}
-
 void mlx5e_activate_priv_channels(struct mlx5e_priv *priv)
 {
-	mlx5e_update_num_tc_x_num_ch(priv);
 	mlx5e_build_txq_maps(priv);
 	mlx5e_activate_channels(&priv->channels);
 	mlx5e_qos_activate_queues(priv);
@@ -4673,11 +4663,6 @@ void mlx5e_build_nic_params(struct mlx5e_priv *priv, struct mlx5e_xsk *xsk, u16
 				     priv->max_nch);
 	mlx5e_params_mqprio_reset(params);
 
-	/* Set an initial non-zero value, so that mlx5e_select_queue won't
-	 * divide by zero if called before first activating channels.
-	 */
-	priv->num_tc_x_num_ch = params->num_channels * params->mqprio.num_tc;
-
 	/* SQ */
 	params->log_sq_size = is_kdump_kernel() ?
 		MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE :
@@ -5230,7 +5215,7 @@ int mlx5e_priv_init(struct mlx5e_priv *priv,
 		    struct net_device *netdev,
 		    struct mlx5_core_dev *mdev)
 {
-	int nch, num_txqs, node, i;
+	int nch, num_txqs, node;
 	int err;
 
 	num_txqs = netdev->num_tx_queues;
@@ -5271,30 +5256,13 @@ int mlx5e_priv_init(struct mlx5e_priv *priv,
 	if (!priv->tx_rates)
 		goto err_free_txq2sq;
 
-	priv->channel_tc2realtxq =
-		kcalloc_node(nch, sizeof(*priv->channel_tc2realtxq), GFP_KERNEL, node);
-	if (!priv->channel_tc2realtxq)
-		goto err_free_tx_rates;
-
-	for (i = 0; i < nch; i++) {
-		priv->channel_tc2realtxq[i] =
-			kcalloc_node(profile->max_tc, sizeof(**priv->channel_tc2realtxq),
-				     GFP_KERNEL, node);
-		if (!priv->channel_tc2realtxq[i])
-			goto err_free_channel_tc2realtxq;
-	}
-
 	priv->channel_stats =
 		kcalloc_node(nch, sizeof(*priv->channel_stats), GFP_KERNEL, node);
 	if (!priv->channel_stats)
-		goto err_free_channel_tc2realtxq;
+		goto err_free_tx_rates;
 
 	return 0;
 
-err_free_channel_tc2realtxq:
-	while (--i >= 0)
-		kfree(priv->channel_tc2realtxq[i]);
-	kfree(priv->channel_tc2realtxq);
 err_free_tx_rates:
 	kfree(priv->tx_rates);
 err_free_txq2sq:
@@ -5319,9 +5287,6 @@ void mlx5e_priv_cleanup(struct mlx5e_priv *priv)
 	for (i = 0; i < priv->stats_nch; i++)
 		kvfree(priv->channel_stats[i]);
 	kfree(priv->channel_stats);
-	for (i = 0; i < priv->max_nch; i++)
-		kfree(priv->channel_tc2realtxq[i]);
-	kfree(priv->channel_tc2realtxq);
 	kfree(priv->tx_rates);
 	kfree(priv->txq2sq);
 	destroy_workqueue(priv->wq);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c
index 06d1f46f1688..a3f257623af4 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c
@@ -632,11 +632,6 @@ static void mlx5e_build_rep_params(struct net_device *netdev)
 	params->mqprio.num_tc       = 1;
 	params->tunneled_offload_en = false;
 
-	/* Set an initial non-zero value, so that mlx5e_select_queue won't
-	 * divide by zero if called before first activating channels.
-	 */
-	priv->num_tc_x_num_ch = params->num_channels * params->mqprio.num_tc;
-
 	mlx5_query_min_inline(mdev, &params->tx_min_inline_mode);
 }
 
-- 
cgit v1.2.3


From 62f7991feab69ed398ad6a03b911fa0ee0549516 Mon Sep 17 00:00:00 2001
From: Maxim Mikityanskiy <maximmi@nvidia.com>
Date: Tue, 25 Jan 2022 12:52:55 +0200
Subject: net/mlx5e: Move repeating code that gets TC prio into a function

Both mlx5e_select_queue and mlx5e_select_ptpsq contain the same logic to
get user priority of a packet, according to the current trust state
settings. This commit moves this repeating code to its own function.

Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en/selq.c | 36 ++++++++++-------------
 1 file changed, 15 insertions(+), 21 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
index a0bed47fd392..aab2046da45b 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
@@ -110,24 +110,25 @@ static int mlx5e_get_dscp_up(struct mlx5e_priv *priv, struct sk_buff *skb)
 }
 #endif
 
+static int mlx5e_get_up(struct mlx5e_priv *priv, struct sk_buff *skb)
+{
+#ifdef CONFIG_MLX5_CORE_EN_DCB
+	if (priv->dcbx_dp.trust_state == MLX5_QPTS_TRUST_DSCP)
+		return mlx5e_get_dscp_up(priv, skb);
+#endif
+	if (skb_vlan_tag_present(skb))
+		return skb_vlan_tag_get_prio(skb);
+	return 0;
+}
+
 static u16 mlx5e_select_ptpsq(struct net_device *dev, struct sk_buff *skb,
 			      struct mlx5e_selq_params *selq)
 {
 	struct mlx5e_priv *priv = netdev_priv(dev);
-	int up = 0;
+	int up;
 
-	if (selq->num_tcs <= 1)
-		goto return_txq;
+	up = selq->num_tcs > 1 ? mlx5e_get_up(priv, skb) : 0;
 
-#ifdef CONFIG_MLX5_CORE_EN_DCB
-	if (priv->dcbx_dp.trust_state == MLX5_QPTS_TRUST_DSCP)
-		up = mlx5e_get_dscp_up(priv, skb);
-	else
-#endif
-		if (skb_vlan_tag_present(skb))
-			up = skb_vlan_tag_get_prio(skb);
-
-return_txq:
 	return selq->num_regular_queues + up;
 }
 
@@ -152,8 +153,7 @@ u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
 {
 	struct mlx5e_priv *priv = netdev_priv(dev);
 	struct mlx5e_selq_params *selq;
-	int txq_ix;
-	int up = 0;
+	int txq_ix, up;
 
 	selq = rcu_dereference_bh(priv->selq.active);
 
@@ -189,13 +189,7 @@ u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
 	if (selq->num_tcs <= 1)
 		return txq_ix;
 
-#ifdef CONFIG_MLX5_CORE_EN_DCB
-	if (priv->dcbx_dp.trust_state == MLX5_QPTS_TRUST_DSCP)
-		up = mlx5e_get_dscp_up(priv, skb);
-	else
-#endif
-		if (skb_vlan_tag_present(skb))
-			up = skb_vlan_tag_get_prio(skb);
+	up = mlx5e_get_up(priv, skb);
 
 	/* Normalize any picked txq_ix to [0, num_channels),
 	 * So we can return a txq_ix that matches the channel and
-- 
cgit v1.2.3


From ed5f9cf06b20f74c1098d6d62313e3e9af217fcb Mon Sep 17 00:00:00 2001
From: Maxim Mikityanskiy <maximmi@nvidia.com>
Date: Tue, 25 Jan 2022 12:52:56 +0200
Subject: net/mlx5e: Use READ_ONCE/WRITE_ONCE for DCBX trust state

trust_state can be written while mlx5e_select_queue() is reading it. To
avoid inconsistencies, use READ_ONCE and WRITE_ONCE for access and
updates, and touch the variable only once per operation.

Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en/selq.c  |  2 +-
 drivers/net/ethernet/mellanox/mlx5/core/en_dcbnl.c | 12 +++++++-----
 2 files changed, 8 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
index aab2046da45b..b8f1a955944d 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
@@ -113,7 +113,7 @@ static int mlx5e_get_dscp_up(struct mlx5e_priv *priv, struct sk_buff *skb)
 static int mlx5e_get_up(struct mlx5e_priv *priv, struct sk_buff *skb)
 {
 #ifdef CONFIG_MLX5_CORE_EN_DCB
-	if (priv->dcbx_dp.trust_state == MLX5_QPTS_TRUST_DSCP)
+	if (READ_ONCE(priv->dcbx_dp.trust_state) == MLX5_QPTS_TRUST_DSCP)
 		return mlx5e_get_dscp_up(priv, skb);
 #endif
 	if (skb_vlan_tag_present(skb))
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_dcbnl.c b/drivers/net/ethernet/mellanox/mlx5/core/en_dcbnl.c
index a4c8d8d00d5a..d659fe07d464 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_dcbnl.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_dcbnl.c
@@ -1142,7 +1142,7 @@ static int mlx5e_update_trust_state_hw(struct mlx5e_priv *priv, void *context)
 	err = mlx5_set_trust_state(priv->mdev, *trust_state);
 	if (err)
 		return err;
-	priv->dcbx_dp.trust_state = *trust_state;
+	WRITE_ONCE(priv->dcbx_dp.trust_state, *trust_state);
 
 	return 0;
 }
@@ -1187,16 +1187,18 @@ static int mlx5e_set_dscp2prio(struct mlx5e_priv *priv, u8 dscp, u8 prio)
 static int mlx5e_trust_initialize(struct mlx5e_priv *priv)
 {
 	struct mlx5_core_dev *mdev = priv->mdev;
+	u8 trust_state;
 	int err;
 
-	priv->dcbx_dp.trust_state = MLX5_QPTS_TRUST_PCP;
-
-	if (!MLX5_DSCP_SUPPORTED(mdev))
+	if (!MLX5_DSCP_SUPPORTED(mdev)) {
+		WRITE_ONCE(priv->dcbx_dp.trust_state, MLX5_QPTS_TRUST_PCP);
 		return 0;
+	}
 
-	err = mlx5_query_trust_state(priv->mdev, &priv->dcbx_dp.trust_state);
+	err = mlx5_query_trust_state(priv->mdev, &trust_state);
 	if (err)
 		return err;
+	WRITE_ONCE(priv->dcbx_dp.trust_state, trust_state);
 
 	mlx5e_params_calc_trust_tx_min_inline_mode(priv->mdev, &priv->channels.params,
 						   priv->dcbx_dp.trust_state);
-- 
cgit v1.2.3


From 3c87aedd48997f9ab3afdeb2084f6b6163a21db8 Mon Sep 17 00:00:00 2001
From: Maxim Mikityanskiy <maximmi@nvidia.com>
Date: Tue, 25 Jan 2022 12:52:57 +0200
Subject: net/mlx5e: Optimize mlx5e_select_queue

This commit optimizes mlx5e_select_queue for HTB and PTP cases by
short-cutting some checks, without sacrificing performance of the common
non-HTB non-PTP flow.

1. The HTB flow uses the fact that num_tcs == 1 to drop these checks
(it's not possible to attach both mqprio and htb as the root qdisc).
It's also enough to calculate `txq_ix % num_channels` only once, instead
of twice.

2. The PTP flow drops the check for HTB and the second calculation of
`txq_ix % num_channels`.

Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en/selq.c | 58 ++++++++++++++++-------
 1 file changed, 42 insertions(+), 16 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
index b8f1a955944d..b3ed5262d2a1 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
@@ -164,36 +164,62 @@ u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
 	if (unlikely(!selq))
 		return 0;
 
-	if (unlikely(selq->is_ptp || selq->is_htb)) {
-		if (unlikely(selq->is_htb)) {
-			txq_ix = mlx5e_select_htb_queue(priv, skb);
-			if (txq_ix > 0)
-				return txq_ix;
-		}
+	if (likely(!selq->is_ptp && !selq->is_htb)) {
+		/* No special queues, netdev_pick_tx returns one of the regular ones. */
+
+		txq_ix = netdev_pick_tx(dev, skb, NULL);
+
+		if (selq->num_tcs <= 1)
+			return txq_ix;
+
+		up = mlx5e_get_up(priv, skb);
+
+		/* Normalize any picked txq_ix to [0, num_channels),
+		 * So we can return a txq_ix that matches the channel and
+		 * packet UP.
+		 */
+		return txq_ix % selq->num_channels + up * selq->num_channels;
+	}
+
+	if (unlikely(selq->is_htb)) {
+		/* num_tcs == 1, shortcut for PTP */
+
+		txq_ix = mlx5e_select_htb_queue(priv, skb);
+		if (txq_ix > 0)
+			return txq_ix;
 
 		if (unlikely(selq->is_ptp && mlx5e_use_ptpsq(skb)))
-			return mlx5e_select_ptpsq(dev, skb, selq);
+			return selq->num_channels;
 
 		txq_ix = netdev_pick_tx(dev, skb, NULL);
+
 		/* Fix netdev_pick_tx() not to choose ptp_channel and HTB txqs.
 		 * If they are selected, switch to regular queues.
 		 * Driver to select these queues only at mlx5e_select_ptpsq()
 		 * and mlx5e_select_htb_queue().
 		 */
-		if (unlikely(txq_ix >= selq->num_regular_queues))
-			txq_ix %= selq->num_regular_queues;
-	} else {
-		txq_ix = netdev_pick_tx(dev, skb, NULL);
+		return txq_ix % selq->num_channels;
 	}
 
+	/* PTP is enabled */
+
+	if (mlx5e_use_ptpsq(skb))
+		return mlx5e_select_ptpsq(dev, skb, selq);
+
+	txq_ix = netdev_pick_tx(dev, skb, NULL);
+
+	/* Normalize any picked txq_ix to [0, num_channels). Queues in range
+	 * [0, num_regular_queues) will be mapped to the corresponding channel
+	 * index, so that we can apply the packet's UP (if num_tcs > 1).
+	 * If netdev_pick_tx() picks ptp_channel, switch to a regular queue,
+	 * because driver should select the PTP only at mlx5e_select_ptpsq().
+	 */
+	txq_ix %= selq->num_channels;
+
 	if (selq->num_tcs <= 1)
 		return txq_ix;
 
 	up = mlx5e_get_up(priv, skb);
 
-	/* Normalize any picked txq_ix to [0, num_channels),
-	 * So we can return a txq_ix that matches the channel and
-	 * packet UP.
-	 */
-	return txq_ix % selq->num_channels + up * selq->num_channels;
+	return txq_ix + up * selq->num_channels;
 }
-- 
cgit v1.2.3


From 3a9e5fff2ab0d6f4af701757d35b9453dc563b78 Mon Sep 17 00:00:00 2001
From: Maxim Mikityanskiy <maximmi@nvidia.com>
Date: Tue, 25 Jan 2022 12:52:59 +0200
Subject: net/mlx5e: Optimize modulo in mlx5e_select_queue

To improve the performance of the modulo operation (%), it's replaced by
a subtracting the divisor in a loop. The modulo is used to fix up an
out-of-bounds value that might be returned by netdev_pick_tx or to
convert the queue number to the channel number when num_tcs > 1. Both
situations are unlikely, because XPS is configured not to pick higher
queues (qid >= num_channels) by default, so under normal circumstances
the flow won't go inside the loop, and it will be faster than %.

num_tcs == 8 adds at most 7 iterations to the loop. PTP adds at most 1
iteration to the loop. HTB would add at most 256 iterations (when
num_channels == 1), so there is an additional boundary check in the HTB
flow, which falls back to % if more than 7 iterations are expected.

Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en/selq.c |  7 ++++---
 drivers/net/ethernet/mellanox/mlx5/core/en/selq.h | 20 ++++++++++++++++++++
 2 files changed, 24 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
index b3ed5262d2a1..667bc95a0d44 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
@@ -178,7 +178,8 @@ u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
 		 * So we can return a txq_ix that matches the channel and
 		 * packet UP.
 		 */
-		return txq_ix % selq->num_channels + up * selq->num_channels;
+		return mlx5e_txq_to_ch_ix(txq_ix, selq->num_channels) +
+			up * selq->num_channels;
 	}
 
 	if (unlikely(selq->is_htb)) {
@@ -198,7 +199,7 @@ u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
 		 * Driver to select these queues only at mlx5e_select_ptpsq()
 		 * and mlx5e_select_htb_queue().
 		 */
-		return txq_ix % selq->num_channels;
+		return mlx5e_txq_to_ch_ix_htb(txq_ix, selq->num_channels);
 	}
 
 	/* PTP is enabled */
@@ -214,7 +215,7 @@ u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
 	 * If netdev_pick_tx() picks ptp_channel, switch to a regular queue,
 	 * because driver should select the PTP only at mlx5e_select_ptpsq().
 	 */
-	txq_ix %= selq->num_channels;
+	txq_ix = mlx5e_txq_to_ch_ix(txq_ix, selq->num_channels);
 
 	if (selq->num_tcs <= 1)
 		return txq_ix;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.h b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.h
index b1c73b509f6b..6c070141d8f1 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.h
@@ -25,6 +25,26 @@ void mlx5e_selq_prepare(struct mlx5e_selq *selq, struct mlx5e_params *params, bo
 void mlx5e_selq_apply(struct mlx5e_selq *selq);
 void mlx5e_selq_cancel(struct mlx5e_selq *selq);
 
+static inline u16 mlx5e_txq_to_ch_ix(u16 txq, u16 num_channels)
+{
+	while (unlikely(txq >= num_channels))
+		txq -= num_channels;
+	return txq;
+}
+
+static inline u16 mlx5e_txq_to_ch_ix_htb(u16 txq, u16 num_channels)
+{
+	if (unlikely(txq >= num_channels)) {
+		if (unlikely(txq >= num_channels << 3))
+			txq %= num_channels;
+		else
+			do
+				txq -= num_channels;
+			while (txq >= num_channels);
+	}
+	return txq;
+}
+
 u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
 		       struct net_device *sb_dev);
 
-- 
cgit v1.2.3


From 71753b8ec103fd71d6ee90e522d797ccf978e4ed Mon Sep 17 00:00:00 2001
From: Maxim Mikityanskiy <maximmi@nvidia.com>
Date: Tue, 25 Jan 2022 12:53:00 +0200
Subject: net/mlx5e: Optimize the common case condition in mlx5e_select_queue

Check all booleans for special queues at once, when deciding whether to
go to the fast path in mlx5e_select_queue. Pack them into bitfields to
have some room for extensibility.

Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en/selq.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
index 667bc95a0d44..d98a277eb7f8 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/selq.c
@@ -12,8 +12,13 @@ struct mlx5e_selq_params {
 	unsigned int num_regular_queues;
 	unsigned int num_channels;
 	unsigned int num_tcs;
-	bool is_htb;
-	bool is_ptp;
+	union {
+		u8 is_special_queues;
+		struct {
+			bool is_htb : 1;
+			bool is_ptp : 1;
+		};
+	};
 };
 
 int mlx5e_selq_init(struct mlx5e_selq *selq, struct mutex *state_lock)
@@ -164,7 +169,7 @@ u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
 	if (unlikely(!selq))
 		return 0;
 
-	if (likely(!selq->is_ptp && !selq->is_htb)) {
+	if (likely(!selq->is_special_queues)) {
 		/* No special queues, netdev_pick_tx returns one of the regular ones. */
 
 		txq_ix = netdev_pick_tx(dev, skb, NULL);
-- 
cgit v1.2.3