net/mlx5: SD, prefer sd_group_size from vport context
Subsystems (5)
Tariq Toukan
maintainerPATCH 1/1 net/mlx5: SD, prefer sd_group_size from vport context
From: Shay Drory <[email protected]>
Newer FW reports the SD group size directly in the NIC vport context via the sd_group_size field, gated by the sd_group_size capability. Switch sd_init() to source the group size from there and fall back to the MPIR-based host_buses query only when the cap is absent. sd_group_size might return 1 in some FW configuration. Add explicit check to disable SD creation in this case.
While here, rename host_buses to group_size throughout sd.c to follow the new name on capable FW.
issue: 4680618
--- .../net/ethernet/mellanox/mlx5/core/lib/sd.c | 75 ++++++++++--------- .../net/ethernet/mellanox/mlx5/core/lib/sd.h | 1 + .../net/ethernet/mellanox/mlx5/core/vport.c | 6 +- include/linux/mlx5/vport.h | 3 +- 4 files changed, 47 insertions(+), 38 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/lib/sd.c +39 -36
@@ -19,7 +19,7 @@struct mlx5_sd {u32 group_id; -u8 host_buses; +u8 group_size;struct mlx5_devcom_comp_dev *devcom;struct dentry *dfs;u8 state;
@@ -46,14 +46,14 @@ enum mlx5_sd_state { MLX5_SD_STATE_UP, }; -static int mlx5_sd_get_host_buses (struct mlx5_core_dev *dev) +static int mlx5_sd_get_group_size (struct mlx5_core_dev *dev) {struct mlx5_sd *sd =mlx5_get_sd (dev);if (!sd)return 1 ; -return sd->host_buses; +return sd->group_size; }struct mlx5_core_dev *mlx5_sd_get_primary (struct mlx5_core_dev *dev)
@@ -107,7 +107,7 @@ int mlx5_sd_pf_num_get(struct mlx5_core_dev *dev)if (pos == dev)break ; -return pf_num * sd->host_buses + i; +return pf_num * sd->group_size + i; } struct mlx5_core_dev *
@@ -118,7 +118,7 @@ mlx5_sd_primary_get_peer(struct mlx5_core_dev *primary, int idx)if (idx ==0 )return primary; -if (idx >=mlx5_sd_get_host_buses (primary)) +if (idx >=mlx5_sd_get_group_size (primary))return NULL ; sd =mlx5_get_sd (primary);
@@ -130,7 +130,7 @@ int mlx5_sd_ch_ix_get_dev_ix(struct mlx5_core_dev *dev, int ch_ix)if (is_mdev_switchdev_mode (dev))return 0 ; -return ch_ix %mlx5_sd_get_host_buses (dev); +return ch_ix %mlx5_sd_get_group_size (dev); }int mlx5_sd_ch_ix_get_vec_ix (struct mlx5_core_dev *dev,int ch_ix)
@@ -138,7 +138,7 @@ int mlx5_sd_ch_ix_get_vec_ix(struct mlx5_core_dev *dev, int ch_ix)if (is_mdev_switchdev_mode (dev))return ch_ix; -return ch_ix /mlx5_sd_get_host_buses (dev); +return ch_ix /mlx5_sd_get_group_size (dev); }struct mlx5_core_dev *mlx5_sd_ch_ix_get_dev (struct mlx5_core_dev *primary,int ch_ix)
@@ -164,7 +164,7 @@ static bool ft_create_alias_supported(struct mlx5_core_dev *dev) }static int mlx5_query_sd (struct mlx5_core_dev *dev,bool *sdm, -u8 *host_buses) +u8 *group_size) {u32 out[MLX5_ST_SZ_DW (mpir_reg)];int err;
@@ -174,7 +174,7 @@ static int mlx5_query_sd(struct mlx5_core_dev *dev, bool *sdm,return err; *sdm =MLX5_GET (mpir_reg, out, sdm); - *host_buses =MLX5_GET (mpir_reg, out, host_buses); + *group_size =MLX5_GET (mpir_reg, out, host_buses);return 0 ; }
@@ -184,10 +184,10 @@ static u32 mlx5_sd_group_id(struct mlx5_core_dev *dev, u8 sd_group)return (u32)((MLX5_CAP_GEN (dev, native_port_num) <<8 ) | sd_group); } -static bool mlx5_sd_caps_supported (struct mlx5_core_dev *dev,u8 host_buses) +static bool mlx5_sd_caps_supported (struct mlx5_core_dev *dev,u8 group_size) {/* Honor the SW implementation limit */ -if (host_buses >MLX5_SD_MAX_GROUP_SZ ) +if (group_size >MLX5_SD_MAX_GROUP_SZ )return false;/* Disconnect secondaries from the network */
@@ -200,7 +200,7 @@ static bool mlx5_sd_caps_supported(struct mlx5_core_dev *dev, u8 host_buses)/* RX steering from primary to secondaries */ if (!MLX5_CAP_GEN (dev, cross_vhca_rqt))return false; -if (host_buses >MLX5_CAP_GEN_2 (dev, max_rqt_vhca_id)) +if (group_size >MLX5_CAP_GEN_2 (dev, max_rqt_vhca_id))return false;/* TX steering from secondaries to primary */
@@ -214,7 +214,7 @@ static bool mlx5_sd_caps_supported(struct mlx5_core_dev *dev, u8 host_buses)bool mlx5_sd_is_supported (struct mlx5_core_dev *dev) { -u8 host_buses, sd_group; +u8 group_size =U8_MAX , sd_group;bool sdm;int err;
@@ -222,23 +222,25 @@ bool mlx5_sd_is_supported(struct mlx5_core_dev *dev)if (!mlx5_core_is_pf (dev))return false; - err =mlx5_query_nic_vport_sd_group (dev, &sd_group); -if (err || !sd_group) + err =mlx5_query_nic_vport_sd_group (dev, &sd_group, &group_size); +if (err || !sd_group || group_size <MLX5_SD_MIN_GROUP_SZ )return false; -if (!MLX5_CAP_MCAM_REG (dev, mpir)) -return false; +if (group_size ==U8_MAX ) { +if (!MLX5_CAP_MCAM_REG (dev, mpir)) +return false; - err =mlx5_query_sd (dev, &sdm, &host_buses); -if (err || !sdm) -return false; + err =mlx5_query_sd (dev, &sdm, &group_size); +if (err || !sdm) +return false; + } -return mlx5_sd_caps_supported (dev, host_buses); +return mlx5_sd_caps_supported (dev, group_size); }static int sd_init (struct mlx5_core_dev *dev) { -u8 host_buses, sd_group; +u8 group_size =U8_MAX , sd_group;struct mlx5_sd *sd;u32 group_id;bool sdm;
@@ -248,26 +250,27 @@ static int sd_init(struct mlx5_core_dev *dev)if (!mlx5_core_is_pf (dev))return 0 ; - err =mlx5_query_nic_vport_sd_group (dev, &sd_group); + err =mlx5_query_nic_vport_sd_group (dev, &sd_group, &group_size);if (err)return err; -if (!sd_group) +if (!sd_group || group_size <MLX5_SD_MIN_GROUP_SZ )return 0 ; -if (!MLX5_CAP_MCAM_REG (dev, mpir)) -return 0 ; +if (group_size ==U8_MAX ) { +if (!MLX5_CAP_MCAM_REG (dev, mpir)) +return 0 ; - err =mlx5_query_sd (dev, &sdm, &host_buses); -if (err) -return err; - -if (!sdm) -return 0 ; + err =mlx5_query_sd (dev, &sdm, &group_size); +if (err) +return err; +if (!sdm) +return 0 ; + } group_id =mlx5_sd_group_id (dev, sd_group); -if (!mlx5_sd_caps_supported (dev, host_buses)) { +if (!mlx5_sd_caps_supported (dev, group_size)) {sd_warn (dev,"can't support requested netdev combining for group id 0x%x, skipping\n" , group_id);return 0 ;
@@ -277,7 +280,7 @@ static int sd_init(struct mlx5_core_dev *dev)if (!sd)return -ENOMEM ; - sd->host_buses = host_buses; + sd->group_size = group_size; sd->group_id = group_id;mlx5_set_sd (dev, sd);
@@ -540,7 +543,7 @@ static int sd_register(struct mlx5_core_dev *dev) sd->devcom = devcom;mlx5_devcom_comp_lock (devcom); -if (mlx5_devcom_comp_get_size (devcom) != sd->host_buses || +if (mlx5_devcom_comp_get_size (devcom) != sd->group_size ||mlx5_devcom_comp_is_ready (devcom)) goto out;
@@ -576,7 +579,7 @@ static int sd_register(struct mlx5_core_dev *dev) DEVCOM_CANT_FAIL, primary); primary_sd =mlx5_get_sd (primary); -if (primary_sd->next_secondary_idx +1 == sd->host_buses) +if (primary_sd->next_secondary_idx +1 == sd->group_size)mlx5_devcom_comp_set_ready (devcom, true); out:mlx5_devcom_comp_unlock (devcom);
drivers/net/ethernet/mellanox/mlx5/core/lib/sd.h +1 -0
@@ -6,6 +6,7 @@#include <linux/types.h> +#define MLX5_SD_MIN_GROUP_SZ 2#define MLX5_SD_MAX_GROUP_SZ 2struct mlx5_sd ;
drivers/net/ethernet/mellanox/mlx5/core/vport.c +5 -1
@@ -550,7 +550,8 @@ int mlx5_query_nic_vport_system_image_guid(struct mlx5_core_dev *mdev, }EXPORT_SYMBOL_GPL (mlx5_query_nic_vport_system_image_guid); -int mlx5_query_nic_vport_sd_group (struct mlx5_core_dev *mdev,u8 *sd_group) +int mlx5_query_nic_vport_sd_group (struct mlx5_core_dev *mdev,u8 *sd_group, +u8 *sd_group_size) {int outlen =MLX5_ST_SZ_BYTES (query_nic_vport_context_out);u32 *out;
@@ -566,6 +567,9 @@ int mlx5_query_nic_vport_sd_group(struct mlx5_core_dev *mdev, u8 *sd_group) *sd_group =MLX5_GET (query_nic_vport_context_out, out, nic_vport_context.sd_group); +if (MLX5_CAP_GEN (mdev, sd_group_size)) + *sd_group_size =MLX5_GET (query_nic_vport_context_out, out, + nic_vport_context.sd_group_size); out:kvfree (out);return err;
include/linux/mlx5/vport.h +2 -1
@@ -78,7 +78,8 @@ int mlx5_query_nic_vport_mtu(struct mlx5_core_dev *mdev, u16 *mtu);int mlx5_modify_nic_vport_mtu (struct mlx5_core_dev *mdev,u16 mtu);int mlx5_query_nic_vport_system_image_guid (struct mlx5_core_dev *mdev,u64 *system_image_guid); -int mlx5_query_nic_vport_sd_group (struct mlx5_core_dev *mdev,u8 *sd_group); +int mlx5_query_nic_vport_sd_group (struct mlx5_core_dev *mdev,u8 *sd_group, +u8 *sd_group_size);int mlx5_query_nic_vport_node_guid (struct mlx5_core_dev *mdev,u16 vport,bool other_vport,u64 *node_guid); int mlx5_modify_nic_vport_node_guid(struct mlx5_core_dev *mdev,
base-commit: 001b5d347d8ba39b2dccaefcc57967b18caec8fe
Signature · 1 line
2.44.0
Quoted 9 lines · Tariq Toukan
From: Shay Drory <[email protected]>
Newer FW reports the SD group size directly in the NIC vport context via the
sd_group_sizefield, gated by thesd_group_sizecapability. Switchsd_init()to source the group size from there and fall back to the MPIR-basedhost_busesquery only when the cap is absent.sd_group_sizemight return 1 in some FW configuration. Add explicit check to disable SD creation in this case.
Quoted 7 lines
While here, rename
host_busestogroup_sizethroughoutsd.cto follow the new name on capable FW.issue: 4680618
Signed-off-by Shay Drory <[email protected]>Reviewed-by Moshe Shemesh <[email protected]>Signed-off-by Tariq Toukan <[email protected]>
Hi,
I'd appreciate it if you could take a look over the AI-generated review
of this patch at https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20260810093037.3138197-1-tariqt%40nvidia.com
But my feeling is that this patch is good as-is.
Quoted 21 lines · Simon Horman
External email: Use caution opening links or attachments
Quoted 9 lines · Tariq Toukan
From: Shay Drory <[email protected]>
Newer FW reports the SD group size directly in the NIC vport context via the
sd_group_sizefield, gated by thesd_group_sizecapability. Switchsd_init()to source the group size from there and fall back to the MPIR-basedhost_busesquery only when the cap is absent.sd_group_sizemight return 1 in some FW configuration. Add explicit check to disable SD creation in this case.Quoted 7 lines
While here, rename
host_busestogroup_sizethroughoutsd.cto follow the new name on capable FW.issue: 4680618
Signed-off-by Shay Drory <[email protected]>Reviewed-by Moshe Shemesh <[email protected]>Signed-off-by Tariq Toukan <[email protected]>
Quoted 8 lines
Hi,
I'd appreciate it if you could take a look over the AI-generated review
of this patch at https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20260810093037.3138197-1-tariqt%40nvidia.comBut my feeling is that this patch is good as-is.
Reviewed-by Simon Horman <[email protected]>
Hi,
thanks for the review.
Regarding the comment:
"Can an MPIR-derived group_size of 0 or 1 therefore still reach sd->group_size?"
[SD] AFAIK, no.
MPIR can't be 1, and if sdm bit is set, it means group_size != 0.
Regarding the comment:
"Would it be clearer to always write a defined value in the callee, or to report capability validity explicitly (for example a bool *valid), or to have sd.c test MLX5_CAP_GEN(dev, sd_group_size) directly instead of
overloading a representable field encoding?
[SD] We considered it, and prefer keeping the cap check confined to mlx5_query_nic_vport_sd_group() so callers don't have to repeat it.
Jakub Kicinski
maintainerQuoted 1 line · On Mon, 10 Aug 2026 12:30:37 +0300 Tariq Toukan wrote:
issue: 4680618
Stripped when applying, please pay more attention.
This patch was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:
Hello:
This patch was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:
Quoted 10 lines · On Mon, 10 Aug 2026 12:30:37 +0300 you wrote:
From: Shay Drory <[email protected]>
Newer FW reports the SD group size directly in the NIC vport context via the
sd_group_sizefield, gated by thesd_group_sizecapability. Switchsd_init()to source the group size from there and fall back to the MPIR-basedhost_busesquery only when the cap is absent.sd_group_sizemight return 1 in some FW configuration. Add explicit check to disable SD creation in this case.[...]
Here is the summary with links:
- [net-next] net/mlx5: SD, prefer sd_group_size from vport context
https://git.kernel.org/netdev/net-next/c/486e5419b7ec
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html