From: Saeed Mahameed <saeed@kernel.org>
To: "David S. Miller" <davem@davemloft.net>,
Jakub Kicinski <kuba@kernel.org>, Paolo Abeni <pabeni@redhat.com>,
Eric Dumazet <edumazet@google.com>
Cc: Saeed Mahameed <saeedm@nvidia.com>,
netdev@vger.kernel.org, Tariq Toukan <tariqt@nvidia.com>,
Maxim Mikityanskiy <maximmi@nvidia.com>
Subject: [PATCH net-next 06/15] net/mlx5e: Improve MTT/KSM alignment
Date: Sat, 1 Oct 2022 21:56:23 -0700 [thread overview]
Message-ID: <20221002045632.291612-7-saeed@kernel.org> (raw)
In-Reply-To: <20221002045632.291612-1-saeed@kernel.org>
From: Maxim Mikityanskiy <maximmi@nvidia.com>
Make mlx5e_mpwrq_mtts_per_wqe take into account that KSM requires
smaller alignment than MTT.
Ensure that there is always an even amount of MTTs in a UMR WQE, so that
complete octwords are formed, and no garbage is mapped.
Drop extra alignment in MLX5_MTT_OCTW that may cause setting too big
ucseg->xlt_octowords, also leading to mapping garbage.
Generalize some calculations by introducing the MLX5_OCTWORD constant.
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 | 6 +-----
drivers/net/ethernet/mellanox/mlx5/core/en/params.c | 10 +++++++++-
drivers/net/ethernet/mellanox/mlx5/core/en/xsk/rx.c | 3 ++-
drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 13 +++++--------
drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 2 +-
5 files changed, 18 insertions(+), 16 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index a2d09f30acd1..93607db1dea4 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -109,12 +109,8 @@ struct page_pool;
#define MLX5_MPWRQ_MAX_PAGES_PER_WQE \
rounddown_pow_of_two(MLX5_UMR_MAX_MTT_SPACE / sizeof(struct mlx5_mtt))
-#define MLX5_ALIGN_MTTS(mtts) (ALIGN(mtts, 8))
-#define MLX5_ALIGNED_MTTS_OCTW(mtts) ((mtts) / 2)
-#define MLX5_MTT_OCTW(mtts) (MLX5_ALIGNED_MTTS_OCTW(MLX5_ALIGN_MTTS(mtts)))
-#define MLX5_KSM_OCTW(ksms) (ksms)
#define MLX5E_MAX_RQ_NUM_MTTS \
- (ALIGN_DOWN(U16_MAX, 4) * 2) /* So that MLX5_MTT_OCTW(num_mtts) fits into u16 */
+ (ALIGN_DOWN(U16_MAX, 4) * 2) /* Fits into u16 and aligned by WQEBB. */
#define MLX5E_MAX_RQ_NUM_KSMS (U16_MAX - 1) /* So that num_ksms fits into u16. */
#define MLX5E_ORDER2_MAX_PACKET_MTU (order_base_2(10 * 1024))
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/params.c b/drivers/net/ethernet/mellanox/mlx5/core/en/params.c
index b57855bf7629..e8c3b8abf941 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/params.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/params.c
@@ -91,6 +91,13 @@ u8 mlx5e_mpwrq_pages_per_wqe(struct mlx5_core_dev *mdev, u8 page_shift,
pages_per_wqe = log_wqe_sz > page_shift ? (1 << (log_wqe_sz - page_shift)) : 1;
+ /* Two MTTs are needed to form an octword. The number of MTTs is encoded
+ * in octwords in a UMR WQE, so we need at least two to avoid mapping
+ * garbage addresses.
+ */
+ if (WARN_ON_ONCE(pages_per_wqe < 2 && umr_mode == MLX5E_MPWRQ_UMR_MODE_ALIGNED))
+ pages_per_wqe = 2;
+
/* Sanity check for further calculations to succeed. */
BUILD_BUG_ON(MLX5_MPWRQ_MAX_PAGES_PER_WQE > 64);
if (WARN_ON_ONCE(pages_per_wqe > MLX5_MPWRQ_MAX_PAGES_PER_WQE))
@@ -131,7 +138,8 @@ u8 mlx5e_mpwrq_mtts_per_wqe(struct mlx5_core_dev *mdev, u8 page_shift,
* MTU. These oversize packets are dropped by the driver at a later
* stage.
*/
- return MLX5_ALIGN_MTTS(pages_per_wqe + 1);
+ return ALIGN(pages_per_wqe + 1,
+ MLX5_SEND_WQE_BB / mlx5e_mpwrq_umr_entry_size(umr_mode));
}
u32 mlx5e_mpwrq_max_num_entries(struct mlx5_core_dev *mdev,
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/xsk/rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en/xsk/rx.c
index e12a856331b8..4b2df2895505 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/xsk/rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/xsk/rx.c
@@ -66,9 +66,10 @@ int mlx5e_xsk_alloc_rx_mpwqe(struct mlx5e_rq *rq, u16 ix)
umr_wqe->ctrl.opmod_idx_opcode =
cpu_to_be32((icosq->pc << MLX5_WQE_CTRL_WQE_INDEX_SHIFT) | MLX5_OPCODE_UMR);
+ /* Optimized for speed: keep in sync with mlx5e_mpwrq_umr_entry_size. */
offset = ix * rq->mpwqe.mtts_per_wqe;
if (likely(rq->mpwqe.umr_mode == MLX5E_MPWRQ_UMR_MODE_ALIGNED))
- offset = MLX5_ALIGNED_MTTS_OCTW(offset);
+ offset = offset * sizeof(struct mlx5_mtt) / MLX5_OCTWORD;
umr_wqe->uctrl.xlt_offset = cpu_to_be16(offset);
icosq->db.wqe_info[pi] = (struct mlx5e_icosq_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 b5a416ff1603..2093b6cc6c7c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -205,14 +205,11 @@ static void mlx5e_disable_blocking_events(struct mlx5e_priv *priv)
static u16 mlx5e_mpwrq_umr_octowords(u32 entries, enum mlx5e_mpwrq_umr_mode umr_mode)
{
- switch (umr_mode) {
- case MLX5E_MPWRQ_UMR_MODE_ALIGNED:
- return MLX5_MTT_OCTW(entries);
- case MLX5E_MPWRQ_UMR_MODE_UNALIGNED:
- return MLX5_KSM_OCTW(entries);
- }
- WARN_ONCE(1, "MPWRQ UMR mode %d is not known\n", umr_mode);
- return 0;
+ u8 umr_entry_size = mlx5e_mpwrq_umr_entry_size(umr_mode);
+
+ WARN_ON_ONCE(entries * umr_entry_size % MLX5_OCTWORD);
+
+ return entries * umr_entry_size / MLX5_OCTWORD;
}
static inline void mlx5e_build_umr_wqe(struct mlx5e_rq *rq,
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index b61604d87701..58084650151f 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -682,7 +682,7 @@ static int mlx5e_alloc_rx_mpwqe(struct mlx5e_rq *rq, u16 ix)
cpu_to_be32((sq->pc << MLX5_WQE_CTRL_WQE_INDEX_SHIFT) |
MLX5_OPCODE_UMR);
- offset = MLX5_ALIGNED_MTTS_OCTW(ix * rq->mpwqe.mtts_per_wqe);
+ offset = (ix * rq->mpwqe.mtts_per_wqe) * sizeof(struct mlx5_mtt) / MLX5_OCTWORD;
umr_wqe->uctrl.xlt_offset = cpu_to_be16(offset);
sq->db.wqe_info[pi] = (struct mlx5e_icosq_wqe_info) {
--
2.37.3
next prev parent reply other threads:[~2022-10-02 5:14 UTC|newest]
Thread overview: 17+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-10-02 4:56 [PATCH net-next 00/15] ] mlx5 xsk updates part4 and more Saeed Mahameed
2022-10-02 4:56 ` [PATCH net-next 01/15] net/mlx5e: xsk: Flush RQ on XSK activation to save memory Saeed Mahameed
2022-10-02 4:56 ` [PATCH net-next 02/15] net/mlx5e: xsk: Set napi_id to support busy polling Saeed Mahameed
2022-10-02 4:56 ` [PATCH net-next 03/15] net/mlx5e: xsk: Include XSK skb_from_cqe callbacks in INDIRECT_CALL Saeed Mahameed
2022-10-02 4:56 ` [PATCH net-next 04/15] net/mlx5e: xsk: Improve need_wakeup logic Saeed Mahameed
2022-10-02 4:56 ` [PATCH net-next 05/15] net/mlx5e: xsk: Use umr_mode to calculate striding RQ parameters Saeed Mahameed
2022-10-02 4:56 ` Saeed Mahameed [this message]
2022-10-02 4:56 ` [PATCH net-next 07/15] net/mlx5e: xsk: Use KLM to protect frame overrun in unaligned mode Saeed Mahameed
2022-10-02 4:56 ` [PATCH net-next 08/15] net/mlx5e: xsk: Print a warning in slow configurations Saeed Mahameed
2022-10-02 4:56 ` [PATCH net-next 09/15] net/mlx5e: xsk: Optimize for unaligned mode with 3072-byte frames Saeed Mahameed
2022-10-02 4:56 ` [PATCH net-next 10/15] net/mlx5e: Expose rx_oversize_pkts_buffer counter Saeed Mahameed
2022-10-02 4:56 ` [PATCH net-next 11/15] net/mlx5: Start health poll at earlier stage of driver load Saeed Mahameed
2022-10-02 4:56 ` [PATCH net-next 12/15] net/mlx5: Set default grace period based on function type Saeed Mahameed
2022-10-02 4:56 ` [PATCH net-next 13/15] net/mlx5: E-Switch, Allow offloading fwd dest flow table with vport Saeed Mahameed
2022-10-02 4:56 ` [PATCH net-next 14/15] net/mlx5: E-switch, Don't update group if qos is not enabled Saeed Mahameed
2022-10-02 4:56 ` [PATCH net-next 15/15] net/mlx5: E-Switch, Return EBUSY if can't get mode lock Saeed Mahameed
2022-10-04 0:10 ` [PATCH net-next 00/15] ] mlx5 xsk updates part4 and more patchwork-bot+netdevbpf
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20221002045632.291612-7-saeed@kernel.org \
--to=saeed@kernel.org \
--cc=davem@davemloft.net \
--cc=edumazet@google.com \
--cc=kuba@kernel.org \
--cc=maximmi@nvidia.com \
--cc=netdev@vger.kernel.org \
--cc=pabeni@redhat.com \
--cc=saeedm@nvidia.com \
--cc=tariqt@nvidia.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).