net: migrate remaining drivers to dedicated _rxfh_context ops
Subsystems (7)
Jakub Kicinski
maintainerv3 cover net: migrate remaining drivers to dedicated _rxfh_context ops
Around a year ago Ed added dedicated ops for managing RSS contexts. This significantly improved the clarity of the driver facing API. Migrate the remaining 3 drivers and remove the old way of muxing the RSS context operations via .set_rxfh().
v3:
- [mlx5] fix compilation
- [patch 4] remove more branches
- [patch 5] reformat
v2: https://lore.kernel.org/[email protected]
- [mlx5] remove hfunc local var in mlx5e_rxfh_hfunc_check()
- [mlx5] make the get functions void and add WARN_ON_ONCE()
- [patch 4] remove rxfh struct in netdev_rss_contexts_free()
v1: https://lore.kernel.org/[email protected]
Jakub Kicinski (5):
eth: otx2: migrate to the *_rxfh_context ops
eth: ice: drop the dead code related to rss_contexts
eth: mlx5: migrate to the *_rxfh_context ops
net: ethtool: remove the compat code for _rxfh_context ops
net: ethtool: reduce indent for _rxfh_context ops
.../marvell/octeontx2/nic/otx2_common.h | 8 +- .../net/ethernet/mellanox/mlx5/core/en/rss.h | 3 +- .../ethernet/mellanox/mlx5/core/en/rx_res.h | 7 +- include/linux/ethtool.h | 4 - drivers/net/ethernet/intel/ice/ice_ethtool.c | 28 +--- .../marvell/octeontx2/nic/otx2_common.c | 27 ++-- .../marvell/octeontx2/nic/otx2_ethtool.c | 137 +++++++++-------- .../ethernet/marvell/octeontx2/nic/otx2_pf.c | 6 +- .../ethernet/marvell/octeontx2/nic/otx2_xsk.c | 4 +- .../net/ethernet/mellanox/mlx5/core/en/rss.c | 5 +- .../ethernet/mellanox/mlx5/core/en/rx_res.c | 30 ++-- .../ethernet/mellanox/mlx5/core/en_ethtool.c | 145 ++++++++++++++---- net/core/dev.c | 15 +- net/ethtool/ioctl.c | 93 +++-------- net/ethtool/rss.c | 3 +- 15 files changed, 250 insertions(+), 265 deletions(-)
Signature · 1 line
2.50.0
Jakub Kicinski
maintainerPATCH v3 1/5 eth: otx2: migrate to the *_rxfh_context ops
otx2 only supports additional indirection tables (no separate keys etc.) so the conversion to dedicated callbacks and core-allocated context is mostly removing the code which stores the extra tables in the driver. Core already stores the indirection tables for additional contexts, and doesn't call .get for them.
One subtle change here is that we'll now start with the table covering all queues, not directing all traffic to queue 0. This is what core expects if the user doesn't pass the initial indir table explicitly (there's a WARN_ON() in the core trying to make sure driver authors don't forget to populate ctx to defaults).
Drivers implementing .create_rxfh_context don't have to set cap_rss_ctx_supported, so remove it.
--- .../marvell/octeontx2/nic/otx2_common.h | 8 +- .../marvell/octeontx2/nic/otx2_common.c | 27 ++-- .../marvell/octeontx2/nic/otx2_ethtool.c | 137 ++++++++++-------- .../ethernet/marvell/octeontx2/nic/otx2_pf.c | 6 +- .../ethernet/marvell/octeontx2/nic/otx2_xsk.c | 4 +- 5 files changed, 90 insertions(+), 92 deletions(-)
drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.h +2 -6
@@ -93,10 +93,6 @@ struct otx2_lmt_info {u64 lmt_addr;u16 lmt_id; }; -/* RSS configuration */ -struct otx2_rss_ctx { -u8 ind_tbl[MAX_RSS_INDIR_TBL_SIZE ]; -};struct otx2_rss_info {u8 enable;
@@ -104,7 +100,7 @@ struct otx2_rss_info {u16 rss_size;#define RSS_HASH_KEY_SIZE 44/* 352 bit key */ u8 key[RSS_HASH_KEY_SIZE ]; -struct otx2_rss_ctx *rss_ctx[MAX_RSS_GROUPS ]; +u32 ind_tbl[MAX_RSS_INDIR_TBL_SIZE ]; };/* NIX (or NPC) RX errors */
@@ -1067,7 +1063,7 @@ int otx2_set_hw_capabilities(struct otx2_nic *pfvf);int otx2_rss_init (struct otx2_nic *pfvf);int otx2_set_flowkey_cfg (struct otx2_nic *pfvf);void otx2_set_rss_key (struct otx2_nic *pfvf); -int otx2_set_rss_table (struct otx2_nic *pfvf,int ctx_id); +int otx2_set_rss_table (struct otx2_nic *pfvf,int ctx_id,const u32 *ind_tbl);/* Mbox handlers */ void mbox_handler_msix_offset(struct otx2_nic *pfvf,
drivers/net/ethernet/marvell/octeontx2/nic/otx2_common.c +9 -18
@@ -318,21 +318,20 @@ int otx2_set_flowkey_cfg(struct otx2_nic *pfvf)return err; } -int otx2_set_rss_table (struct otx2_nic *pfvf,int ctx_id) +int otx2_set_rss_table (struct otx2_nic *pfvf,int ctx_id,const u32 *ind_tbl) {struct otx2_rss_info *rss = &pfvf->hw.rss_info;const int index = rss->rss_size * ctx_id;struct mbox *mbox = &pfvf->mbox; -struct otx2_rss_ctx *rss_ctx;struct nix_aq_enq_req *aq;int idx, err;mutex_lock (&mbox->lock); - rss_ctx = rss->rss_ctx[ctx_id]; + ind_tbl = ind_tbl ?: rss->ind_tbl;/* Get memory to put this msg */ for (idx =0 ; idx < rss->rss_size; idx++) {/* Ignore the queue if AF_XDP zero copy is enabled */ -if (test_bit (rss_ctx->ind_tbl[idx], pfvf->af_xdp_zc_qidx)) +if (test_bit (ind_tbl[idx], pfvf->af_xdp_zc_qidx))continue ; aq =otx2_mbox_alloc_msg_nix_aq_enq (mbox);
@@ -352,7 +351,7 @@ int otx2_set_rss_table(struct otx2_nic *pfvf, int ctx_id) } } - aq->rss.rq = rss_ctx->ind_tbl[idx]; + aq->rss.rq = ind_tbl[idx];/* Fill AQ info */ aq->qidx = index + idx;
@@ -390,30 +389,22 @@ void otx2_set_rss_key(struct otx2_nic *pfvf)int otx2_rss_init (struct otx2_nic *pfvf) {struct otx2_rss_info *rss = &pfvf->hw.rss_info; -struct otx2_rss_ctx *rss_ctx;int idx, ret =0 ; - rss->rss_size =sizeof (*rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP ]); + rss->rss_size =sizeof (*rss->ind_tbl);/* Init RSS key if it is not setup already */ if (!rss->enable)netdev_rss_key_fill (rss->key,sizeof (rss->key));otx2_set_rss_key (pfvf); -if (!netif_is_rxfh_configured (pfvf->netdev)) { -/* Set RSS group 0 as default indirection table */ - rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP ] =kzalloc (rss->rss_size, -GFP_KERNEL ); -if (!rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP ]) -return -ENOMEM ; - - rss_ctx = rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP ]; +if (!netif_is_rxfh_configured (pfvf->netdev))for (idx =0 ; idx < rss->rss_size; idx++) - rss_ctx->ind_tbl[idx] = + rss->ind_tbl[idx] =ethtool_rxfh_indir_default (idx, pfvf->hw.rx_queues); - } - ret =otx2_set_rss_table (pfvf,DEFAULT_RSS_CONTEXT_GROUP ); + + ret =otx2_set_rss_table (pfvf,DEFAULT_RSS_CONTEXT_GROUP ,NULL );if (ret)return ret;
drivers/net/ethernet/marvell/octeontx2/nic/otx2_ethtool.c +76 -61
232 lines not on this page · raw message
drivers/net/ethernet/marvell/octeontx2/nic/otx2_pf.c +1 -5
@@ -2158,7 +2158,6 @@ int otx2_stop(struct net_device *netdev)struct otx2_nic *pf =netdev_priv (netdev);struct otx2_cq_poll *cq_poll =NULL ;struct otx2_qset *qset = &pf->qset; -struct otx2_rss_info *rss;int qidx, vec, wrk;/* If the DOWN flag is set resources are already freed */
@@ -2176,10 +2175,7 @@ int otx2_stop(struct net_device *netdev)otx2_rxtx_enable (pf, false);/* Clear RSS enable flag */ - rss = &pf->hw.rss_info; - rss->enable = false; -if (!netif_is_rxfh_configured (netdev)) -kfree (rss->rss_ctx[DEFAULT_RSS_CONTEXT_GROUP ]); + pf->hw.rss_info.enable = false;/* Cleanup Queue IRQ */ vec = pci_irq_vector(pf->pdev,
drivers/net/ethernet/marvell/octeontx2/nic/otx2_xsk.c +2 -2
@@ -132,7 +132,7 @@ int otx2_xsk_pool_enable(struct otx2_nic *pf, struct xsk_buff_pool *pool, u16 qiset_bit (qidx, pf->af_xdp_zc_qidx);otx2_clean_up_rq (pf, qidx);/* Reconfigure RSS table as 'qidx' cannot be part of RSS now */ -otx2_set_rss_table (pf,DEFAULT_RSS_CONTEXT_GROUP ); +otx2_set_rss_table (pf,DEFAULT_RSS_CONTEXT_GROUP ,NULL );/* Kick start the NAPI context so that receiving will start */ return otx2_xsk_wakeup (pf->netdev, qidx,XDP_WAKEUP_RX ); }
@@ -153,7 +153,7 @@ int otx2_xsk_pool_disable(struct otx2_nic *pf, u16 qidx)clear_bit (qidx, pf->af_xdp_zc_qidx);xsk_pool_dma_unmap (pool,DMA_ATTR_SKIP_CPU_SYNC |DMA_ATTR_WEAK_ORDERING );/* Reconfigure RSS table as 'qidx' now need to be part of RSS now */ -otx2_set_rss_table (pf,DEFAULT_RSS_CONTEXT_GROUP ); +otx2_set_rss_table (pf,DEFAULT_RSS_CONTEXT_GROUP ,NULL );return 0 ; }
Signature · 1 line
2.50.0
Re: [PATCH net-next v3 1/5] eth: otx2: migrate to the *_rxfh_context ops
Quoted 13 lines · Jakub Kicinski
otx2 only supports additional indirection tables (no separate keys etc.) so the conversion to dedicated callbacks and core-allocated context is mostly removing the code which stores the extra tables in the driver. Core already stores the indirection tables for additional contexts, and doesn't call .get for them.
One subtle change here is that we'll now start with the table covering all queues, not directing all traffic to queue 0. This is what core expects if the user doesn't pass the initial indir table explicitly (there's a
WARN_ON()in the core trying to make sure driver authors don't forget to populate ctx to defaults).
Quoted 5 lines
Drivers implementing .create_rxfh_context don't have to set
cap_rss_ctx_supported, so remove it.Tested-by Geetha Sowjanya <[email protected]>Signed-off-by Jakub Kicinski <[email protected]>
Jakub Kicinski
maintainerPATCH v3 2/5 eth: ice: drop the dead code related to rss_contexts
In reply to Jakub KicinskiICE appears to have some odd form of rss_context use plumbed in for .get_rxfh. The .set_rxfh side does not support creating contexts, however, so this must be dead code. For at least a year now (since commit 7964e7884643 ("net: ethtool: use the tracking array for get_rxfh on custom RSS contexts")) we have not been calling .get_rxfh with a non-zero rss_context. We just get the info from the RSS XArray under dev->ethtool.
Remove what must be dead code in the driver, clear the support flags.
--- drivers/net/ethernet/intel/ice/ice_ethtool.c | 28 +++----------------- 1 file changed, 3 insertions(+), 25 deletions(-)
drivers/net/ethernet/intel/ice/ice_ethtool.c +3 -25
@@ -3591,11 +3591,10 @@ static int ice_get_rxfh(struct net_device *netdev, struct ethtool_rxfh_param *rxfh) { struct ice_netdev_priv *np = netdev_priv(netdev); -u32 rss_context = rxfh->rss_context;struct ice_vsi *vsi = np->vsi;struct ice_pf *pf = vsi->back;u16 qcount, offset; -int err, num_tc, i; +int err, i;u8 *lut;if (!test_bit (ICE_FLAG_RSS_ENA , pf->flags)) {
@@ -3603,24 +3602,8 @@ ice_get_rxfh(struct net_device *netdev, struct ethtool_rxfh_param *rxfh)return -EOPNOTSUPP ; } -if (rss_context && !ice_is_adq_active (pf)) { -netdev_err (netdev,"RSS context cannot be non-zero when ADQ is not configured.\n" ); -return -EINVAL ; - } - - qcount = vsi->mqprio_qopt.qopt.count[rss_context]; - offset = vsi->mqprio_qopt.qopt.offset[rss_context]; - -if (rss_context &&ice_is_adq_active (pf)) { - num_tc = vsi->mqprio_qopt.qopt.num_tc; -if (rss_context >= num_tc) { -netdev_err (netdev,"RSS context:%d > num_tc:%d\n" , - rss_context, num_tc); -return -EINVAL ; - } -/* Use channel VSI of given TC */ - vsi = vsi->tc_map_vsi[rss_context]; - } + qcount = vsi->mqprio_qopt.qopt.count[0 ]; + offset = vsi->mqprio_qopt.qopt.offset[0 ]; rxfh->hfunc =ETH_RSS_HASH_TOP ;if (vsi->rss_hfunc ==ICE_AQ_VSI_Q_OPT_RSS_HASH_SYM_TPLZ )
@@ -3680,9 +3663,6 @@ ice_set_rxfh(struct net_device *netdev, struct ethtool_rxfh_param *rxfh, rxfh->hfunc != ETH_RSS_HASH_TOP)return -EOPNOTSUPP ; -if (rxfh->rss_context) -return -EOPNOTSUPP ; -if (!test_bit (ICE_FLAG_RSS_ENA , pf->flags)) {/* RSS not supported return error here */ netdev_warn (netdev,"RSS is not configured on this VSI!\n" );
@@ -4750,12 +4730,10 @@ static int ice_repr_ethtool_reset(struct net_device *dev, u32 *flags) }static const struct ethtool_ops ice_ethtool_ops = { - .cap_rss_ctx_supported = true, .supported_coalesce_params =ETHTOOL_COALESCE_USECS |ETHTOOL_COALESCE_USE_ADAPTIVE |ETHTOOL_COALESCE_RX_USECS_HIGH , .supported_input_xfrm =RXH_XFRM_SYM_XOR , - .rxfh_per_ctx_key = true, .get_link_ksettings = ice_get_link_ksettings, .set_link_ksettings = ice_set_link_ksettings, .get_fec_stats = ice_get_fec_stats,
Signature · 1 line
2.50.0
Re: [PATCH net-next v3 2/5] eth: ice: drop the dead code related to rss_contexts
Quoted 11 lines · Jakub Kicinski
ICE appears to have some odd form of
rss_contextuse plumbed in for .get_rxfh. The .set_rxfh side does not support creating contexts, however, so this must be dead code. For at least a year now (since commit7964e7884643("net: ethtool: use the tracking array forget_rxfhon custom RSS contexts")) we have not been calling .get_rxfh with a non-zerorss_context. We just get the info from the RSS XArray underdev->ethtool.Remove what must be dead code in the driver, clear the support flags.
Signed-off-by Jakub Kicinski <[email protected]>
Jakub Kicinski
maintainerPATCH v3 3/5 eth: mlx5: migrate to the *_rxfh_context ops
In reply to Jakub KicinskiConvert mlx5 to dedicated RXFH ops. This is a fairly shallow conversion, TBH, most of the driver code stays as is, but we let the core allocate the context ID for the driver.
mlx5e_rx_res_rss_get_rxfh() and friends are made void, since core only calls the driver for context 0. The second call is right after context creation so it must exist (tm).
Tested with drivers/net/hw/rss_ctx.py on MCX6.
---
v3:
- remove the stale return
- move init
v2: https://lore.kernel.org/[email protected]
- remove hfunc local var in mlx5e_rxfh_hfunc_check()
- make the get functions void and add WARN_ON_ONCE()
v1: https://lore.kernel.org/[email protected]
--- .../net/ethernet/mellanox/mlx5/core/en/rss.h | 3 +- .../ethernet/mellanox/mlx5/core/en/rx_res.h | 7 +- .../net/ethernet/mellanox/mlx5/core/en/rss.c | 5 +- .../ethernet/mellanox/mlx5/core/en/rx_res.c | 30 ++-- .../ethernet/mellanox/mlx5/core/en_ethtool.c | 145 ++++++++++++++---- 5 files changed, 130 insertions(+), 60 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en/rss.h +2 -1
@@ -47,7 +47,8 @@ void mlx5e_rss_disable(struct mlx5e_rss *rss);int mlx5e_rss_packet_merge_set_param (struct mlx5e_rss *rss,struct mlx5e_packet_merge_param *pkt_merge_param); -int mlx5e_rss_get_rxfh (struct mlx5e_rss *rss,u32 *indir,u8 *key,u8 *hfunc,bool *symmetric); +void mlx5e_rss_get_rxfh (struct mlx5e_rss *rss,u32 *indir,u8 *key,u8 *hfunc, +bool *symmetric);int mlx5e_rss_set_rxfh (struct mlx5e_rss *rss,const u32 *indir,const u8 *key,const u8 *hfunc,const bool *symmetric,u32 *rqns,u32 *vhca_ids,unsigned int num_rqns);
drivers/net/ethernet/mellanox/mlx5/core/en/rx_res.h +4 -3
@@ -48,8 +48,9 @@ void mlx5e_rx_res_xsk_update(struct mlx5e_rx_res *res, struct mlx5e_channels *ch/* Configuration API */ void mlx5e_rx_res_rss_set_indir_uniform (struct mlx5e_rx_res *res,unsigned int nch); -int mlx5e_rx_res_rss_get_rxfh (struct mlx5e_rx_res *res,u32 rss_idx, -u32 *indir,u8 *key,u8 *hfunc,bool *symmetric); +void mlx5e_rx_res_rss_get_rxfh (struct mlx5e_rx_res *res,u32 rss_idx, +u32 *indir,u8 *key,u8 *hfunc, +bool *symmetric);int mlx5e_rx_res_rss_set_rxfh (struct mlx5e_rx_res *res,u32 rss_idx,const u32 *indir,const u8 *key,const u8 *hfunc,const bool *symmetric);
@@ -61,7 +62,7 @@ int mlx5e_rx_res_rss_set_hash_fields(struct mlx5e_rx_res *res, u32 rss_idx,int mlx5e_rx_res_packet_merge_set_param (struct mlx5e_rx_res *res,struct mlx5e_packet_merge_param *pkt_merge_param); -int mlx5e_rx_res_rss_init (struct mlx5e_rx_res *res,u32 *rss_idx,unsigned int init_nch); +int mlx5e_rx_res_rss_init (struct mlx5e_rx_res *res,u32 rss_idx,unsigned int init_nch);int mlx5e_rx_res_rss_destroy (struct mlx5e_rx_res *res,u32 rss_idx);int mlx5e_rx_res_rss_cnt (struct mlx5e_rx_res *res);int mlx5e_rx_res_rss_index (struct mlx5e_rx_res *res,struct mlx5e_rss *rss);
drivers/net/ethernet/mellanox/mlx5/core/en/rss.c +2 -3
@@ -567,7 +567,8 @@ int mlx5e_rss_packet_merge_set_param(struct mlx5e_rss *rss,return final_err; } -int mlx5e_rss_get_rxfh (struct mlx5e_rss *rss,u32 *indir,u8 *key,u8 *hfunc,bool *symmetric) +void mlx5e_rss_get_rxfh (struct mlx5e_rss *rss,u32 *indir,u8 *key,u8 *hfunc, +bool *symmetric) {if (indir) memcpy(indir, rss->indir.table,
@@ -582,8 +583,6 @@ int mlx5e_rss_get_rxfh(struct mlx5e_rss *rss, u32 *indir, u8 *key, u8 *hfunc, boif (symmetric) *symmetric = rss->hash.symmetric; - -return 0 ; } int mlx5e_rss_set_rxfh(struct mlx5e_rss *rss, const u32 *indir,
drivers/net/ethernet/mellanox/mlx5/core/en/rx_res.c +11 -19
@@ -71,17 +71,12 @@ static int mlx5e_rx_res_rss_init_def(struct mlx5e_rx_res *res,return 0 ; } -int mlx5e_rx_res_rss_init (struct mlx5e_rx_res *res,u32 *rss_idx,unsigned int init_nch) +int mlx5e_rx_res_rss_init (struct mlx5e_rx_res *res,u32 rss_idx,unsigned int init_nch) {bool inner_ft_support = res->features &MLX5E_RX_RES_FEATURE_INNER_FT ;struct mlx5e_rss *rss; -int i; -for (i =1 ; i <MLX5E_MAX_NUM_RSS ; i++) -if (!res->rss[i]) -break ; - -if (i ==MLX5E_MAX_NUM_RSS ) +if (WARN_ON_ONCE (res->rss[rss_idx]))return -ENOSPC ; rss = mlx5e_rss_init(res->mdev, inner_ft_support, res->drop_rqn,
@@ -97,8 +92,7 @@ int mlx5e_rx_res_rss_init(struct mlx5e_rx_res *res, u32 *rss_idx, unsigned int imlx5e_rss_enable (rss, res->rss_rqns, vhca_ids, res->rss_nch); } - res->rss[i] = rss; - *rss_idx = i; + res->rss[rss_idx] = rss;return 0 ; }
@@ -193,19 +187,17 @@ void mlx5e_rx_res_rss_set_indir_uniform(struct mlx5e_rx_res *res, unsigned int nmlx5e_rss_set_indir_uniform (res->rss[0 ], nch); } -int mlx5e_rx_res_rss_get_rxfh (struct mlx5e_rx_res *res,u32 rss_idx, -u32 *indir,u8 *key,u8 *hfunc,bool *symmetric) +void mlx5e_rx_res_rss_get_rxfh (struct mlx5e_rx_res *res,u32 rss_idx, +u32 *indir,u8 *key,u8 *hfunc,bool *symmetric) { -struct mlx5e_rss *rss; +struct mlx5e_rss *rss =NULL ; -if (rss_idx >=MLX5E_MAX_NUM_RSS ) -return -EINVAL ; +if (rss_idx <MLX5E_MAX_NUM_RSS ) + rss = res->rss[rss_idx]; +if (WARN_ON_ONCE (!rss)) +return ; - rss = res->rss[rss_idx]; -if (!rss) -return -ENOENT ; - -return mlx5e_rss_get_rxfh (rss, indir, key, hfunc, symmetric); +mlx5e_rss_get_rxfh (rss, indir, key, hfunc, symmetric); } int mlx5e_rx_res_rss_set_rxfh(struct mlx5e_rx_res *res, u32 rss_idx,
drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c +111 -34
193 lines not on this page · raw message
Signature · 1 line
2.50.0
Re: [PATCH net-next v3 3/5] eth: mlx5: migrate to the *_rxfh_context ops
Quoted 11 lines · Jakub Kicinski
Convert mlx5 to dedicated RXFH ops. This is a fairly shallow conversion, TBH, most of the driver code stays as is, but we let the core allocate the context ID for the driver.
mlx5e_rx_res_rss_get_rxfh()and friends are made void, since core only calls the driver for context 0. The second call is right after context creation so it must exist (tm).Tested with
drivers/net/hw/rss_ctx.pyon MCX6.Signed-off-by Jakub Kicinski <[email protected]>
Thanks!
Jakub Kicinski
maintainerPATCH v3 4/5 net: ethtool: remove the compat code for _rxfh_context ops
In reply to Jakub KicinskiAll drivers are now converted to dedicated _rxfh_context ops. Remove the use of >set_rxfh() to manage additional contexts.
---
v3:
- remove a couple more branches
- bring back a comment
v2: https://lore.kernel.org/[email protected]
- remove rxfh struct in netdev_rss_contexts_free()
v1: https://lore.kernel.org/[email protected]
--- include/linux/ethtool.h | 4 --- net/core/dev.c | 15 +--------- net/ethtool/ioctl.c | 65 +++++++++-------------------------------- net/ethtool/rss.c | 3 +- 4 files changed, 16 insertions(+), 71 deletions(-)
include/linux/ethtool.h +0 -4
@@ -865,9 +865,6 @@ struct kernel_ethtool_ts_info {* @supported_input_xfrm: supported types of input xfrm from %RXH_XFRM_*. * @cap_link_lanes_supported: indicates if the driver supports lanes * parameter. -* @cap_rss_ctx_supported: indicates if the driver supports RSS -* contexts via legacy API, drivers implementing @create_rxfh_context -* do not have to set this bit. * @rxfh_per_ctx_fields: device supports selecting different header fields * for Rx hash calculation and RSS for each additional context. * @rxfh_per_ctx_key: device supports setting different RSS key for each
@@ -1100,7 +1097,6 @@ struct kernel_ethtool_ts_info {struct ethtool_ops {u32 supported_input_xfrm:8 ;u32 cap_link_lanes_supported:1 ; -u32 cap_rss_ctx_supported:1 ;u32 rxfh_per_ctx_fields:1 ;u32 rxfh_per_ctx_key:1 ;u32 cap_rss_rxnfc_adds:1 ;
net/core/dev.c +1 -14
@@ -11976,21 +11976,8 @@ static void netdev_rss_contexts_free(struct net_device *dev)mutex_lock (&dev->ethtool->rss_lock);xa_for_each (&dev->ethtool->rss_ctx, context, ctx) { -struct ethtool_rxfh_param rxfh; - - rxfh.indir =ethtool_rxfh_context_indir (ctx); - rxfh.key =ethtool_rxfh_context_key (ctx); - rxfh.hfunc = ctx->hfunc; - rxfh.input_xfrm = ctx->input_xfrm; - rxfh.rss_context = context; - rxfh.rss_delete = true; -xa_erase (&dev->ethtool->rss_ctx, context); -if (dev->ethtool_ops->create_rxfh_context) - dev->ethtool_ops->remove_rxfh_context (dev, ctx, - context,NULL ); -else - dev->ethtool_ops->set_rxfh (dev, &rxfh,NULL ); + dev->ethtool_ops->remove_rxfh_context (dev, ctx, context,NULL );kfree (ctx); }xa_destroy (&dev->ethtool->rss_ctx);
net/ethtool/ioctl.c +14 -51
108 lines not on this page · raw message
net/ethtool/rss.c +1 -2
@@ -163,8 +163,7 @@ rss_prepare_data(const struct ethnl_req_info *req_base,return -EOPNOTSUPP ;/* Some drivers don't handle rss_context */ -if (request->rss_context && - !ops->cap_rss_ctx_supported && !ops->create_rxfh_context) +if (request->rss_context && !ops->create_rxfh_context)return -EOPNOTSUPP ;mutex_lock (&dev->ethtool->rss_lock);
Signature · 1 line
2.50.0
Re: [PATCH net-next v3 4/5] net: ethtool: remove the compat code for _rxfh_context ops
Quoted 5 lines · Jakub Kicinski
All drivers are now converted to dedicated
_rxfh_contextops. Remove the use of >set_rxfh() to manage additional contexts.Reviewed-by Gal Pressman <[email protected]>Signed-off-by Jakub Kicinski <[email protected]>
Jakub Kicinski
maintainerPATCH v3 5/5 net: ethtool: reduce indent for _rxfh_context ops
In reply to Jakub KicinskiNow that we don't have the compat code we can reduce the indent a little. No functional changes.
---
v3:
- unsplit one more line
v2: https://lore.kernel.org/[email protected]
--- net/ethtool/ioctl.c | 30 ++++++++++++------------------ 1 file changed, 12 insertions(+), 18 deletions(-)
net/ethtool/ioctl.c +12 -18
@@ -1664,25 +1664,19 @@ static noinline_for_stack int ethtool_set_rxfh(struct net_device *dev, rxfh_dev.rss_context = rxfh.rss_context; rxfh_dev.input_xfrm = rxfh.input_xfrm; -if (rxfh.rss_context) { -if (create) { - ret = ops->create_rxfh_context (dev, ctx, &rxfh_dev, - extack); -/* Make sure driver populates defaults */ -WARN_ON_ONCE (!ret && !rxfh_dev.key && - ops->rxfh_per_ctx_key && - !memchr_inv (ethtool_rxfh_context_key (ctx), -0 , ctx->key_size)); - }else if (rxfh_dev.rss_delete) { - ret = ops->remove_rxfh_context (dev, ctx, - rxfh.rss_context, - extack); - }else { - ret = ops->modify_rxfh_context (dev, ctx, &rxfh_dev, - extack); - } - }else { +if (!rxfh.rss_context) { ret = ops->set_rxfh (dev, &rxfh_dev, extack); + }else if (create) { + ret = ops->create_rxfh_context (dev, ctx, &rxfh_dev, extack); +/* Make sure driver populates defaults */ +WARN_ON_ONCE (!ret && !rxfh_dev.key && ops->rxfh_per_ctx_key && + !memchr_inv (ethtool_rxfh_context_key (ctx),0 , + ctx->key_size)); + }else if (rxfh_dev.rss_delete) { + ret = ops->remove_rxfh_context (dev, ctx, rxfh.rss_context, + extack); + }else { + ret = ops->modify_rxfh_context (dev, ctx, &rxfh_dev, extack); }if (ret) {if (create) {
Signature · 1 line
2.50.0
Re: [PATCH net-next v3 5/5] net: ethtool: reduce indent for _rxfh_context ops
Quoted 5 lines · Jakub Kicinski
Now that we don't have the compat code we can reduce the indent a little. No functional changes.
Reviewed-by Gal Pressman <[email protected]>Signed-off-by Jakub Kicinski <[email protected]>
This series was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:
Hello:
This series was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:
Quoted 15 lines · On Mon, 7 Jul 2025 11:41:10 -0700 you wrote:
Around a year ago Ed added dedicated ops for managing RSS contexts. This significantly improved the clarity of the driver facing API. Migrate the remaining 3 drivers and remove the old way of muxing the RSS context operations via .set_rxfh().
v3:
- [mlx5] fix compilation
- [patch 4] remove more branches
- [patch 5] reformat
v2: https://lore.kernel.org/[email protected]
- [mlx5] remove hfunc local var inmlx5e_rxfh_hfunc_check()
- [mlx5] make the get functions void and addWARN_ON_ONCE()
- [patch 4] remove rxfh struct innetdev_rss_contexts_free()
v1: https://lore.kernel.org/[email protected]
Quoted 1 line
[...]
Here is the summary with links:
- [net-next,v3,1/5] eth: otx2: migrate to the *_rxfh_context ops
https://git.kernel.org/netdev/net-next/c/62e01d8c4170 - [net-next,v3,2/5] eth: ice: drop the dead code related to rss_contexts https://git.kernel.org/netdev/net-next/c/be78c83a8bbb - [net-next,v3,3/5] eth: mlx5: migrate to the *_rxfh_context ops https://git.kernel.org/netdev/net-next/c/afc55a0659a6 - [net-next,v3,4/5] net: ethtool: remove the compat code for _rxfh_context ops https://git.kernel.org/netdev/net-next/c/4e655028c29f - [net-next,v3,5/5] net: ethtool: reduce indent for _rxfh_context ops https://git.kernel.org/netdev/net-next/c/cd7e8841b61f
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html