| Message ID | 20250321151357.28540-1-michal.kubiak@intel.com |
|---|---|
| State | Superseded |
| Delegated to: | Anthony Nguyen |
| Headers | show |
| Series | [iwl-next] ice: add a separate Rx handler for flow director commands | expand |
> -----Original Message----- > From: Intel-wired-lan <intel-wired-lan-bounces@osuosl.org> On Behalf Of Michal > Kubiak > Sent: Friday, March 21, 2025 8:14 AM > To: intel-wired-lan@lists.osuosl.org > Cc: Fijalkowski, Maciej <maciej.fijalkowski@intel.com>; netdev@vger.kernel.org; > Kitszel, Przemyslaw <przemyslaw.kitszel@intel.com>; Kubiak, Michal > <michal.kubiak@intel.com>; Swiatkowski, Michal > <michal.swiatkowski@intel.com> > Subject: [Intel-wired-lan] [PATCH iwl-next] ice: add a separate Rx handler for flow > director commands > > The "ice" driver implementation uses the control VSI to handle > the flow director configuration for PFs and VFs. > > Unfortunately, although a separate VSI type was created to handle flow > director queues, the Rx queue handler was shared between the flow > director and a standard NAPI Rx handler. > > Such a design approach was not very flexible. First, it mixed hotpath > and slowpath code, blocking their further optimization. It also created > a huge overkill for the flow director command processing, which is > descriptor-based only, so there is no need to allocate Rx data buffers. > > For the above reasons, implement a separate Rx handler for the control > VSI. Also, remove from the NAPI handler the code dedicated to > configuring the flow director rules on VFs. > Do not allocate Rx data buffers to the flow director queues because > their processing is descriptor-based only. > Finally, allow Rx data queues to be allocated only for VSIs that have > netdev assigned to them. > > This handler splitting approach is the first step in converting the > driver to use the Page Pool (which can only be used for data queues). > > Test hints: > 1. Create a VF for any PF managed by the ice driver. > 2. In a loop, add and delete flow director rules for the VF, e.g.: > > for i in {1..128}; do > q=$(( i % 16 )) > ethtool -N ens802f0v0 flow-type tcp4 dst-port "$i" action "$q" > done > > for i in {0..127}; do > ethtool -N ens802f0v0 delete "$i" > done > > Suggested-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> > Suggested-by: Michal Swiatkowski <michal.swiatkowski@intel.com> > Acked-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> > Signed-off-by: Michal Kubiak <michal.kubiak@intel.com> This is a much cleaner approach! Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
On 3/21/25 16:13, Michal Kubiak wrote: > The "ice" driver implementation uses the control VSI to handle > the flow director configuration for PFs and VFs. > > Unfortunately, although a separate VSI type was created to handle flow > director queues, the Rx queue handler was shared between the flow > director and a standard NAPI Rx handler. > > Such a design approach was not very flexible. First, it mixed hotpath > and slowpath code, blocking their further optimization. It also created > a huge overkill for the flow director command processing, which is > descriptor-based only, so there is no need to allocate Rx data buffers. > > For the above reasons, implement a separate Rx handler for the control > VSI. Also, remove from the NAPI handler the code dedicated to > configuring the flow director rules on VFs. > Do not allocate Rx data buffers to the flow director queues because > their processing is descriptor-based only. > Finally, allow Rx data queues to be allocated only for VSIs that have > netdev assigned to them. > > This handler splitting approach is the first step in converting the > driver to use the Page Pool (which can only be used for data queues). > > Test hints: > 1. Create a VF for any PF managed by the ice driver. > 2. In a loop, add and delete flow director rules for the VF, e.g.: > > for i in {1..128}; do > q=$(( i % 16 )) > ethtool -N ens802f0v0 flow-type tcp4 dst-port "$i" action "$q" > done > > for i in {0..127}; do > ethtool -N ens802f0v0 delete "$i" > done > > Suggested-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> > Suggested-by: Michal Swiatkowski <michal.swiatkowski@intel.com> > Acked-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> > Signed-off-by: Michal Kubiak <michal.kubiak@intel.com> Thank you, a very nice improvement Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
On Fri, Mar 21, 2025 at 04:13:57PM +0100, Michal Kubiak wrote: > The "ice" driver implementation uses the control VSI to handle > the flow director configuration for PFs and VFs. > > Unfortunately, although a separate VSI type was created to handle flow > director queues, the Rx queue handler was shared between the flow > director and a standard NAPI Rx handler. > > Such a design approach was not very flexible. First, it mixed hotpath > and slowpath code, blocking their further optimization. It also created > a huge overkill for the flow director command processing, which is > descriptor-based only, so there is no need to allocate Rx data buffers. > > For the above reasons, implement a separate Rx handler for the control > VSI. Also, remove from the NAPI handler the code dedicated to > configuring the flow director rules on VFs. > Do not allocate Rx data buffers to the flow director queues because > their processing is descriptor-based only. > Finally, allow Rx data queues to be allocated only for VSIs that have > netdev assigned to them. > > This handler splitting approach is the first step in converting the > driver to use the Page Pool (which can only be used for data queues). > > Test hints: > 1. Create a VF for any PF managed by the ice driver. > 2. In a loop, add and delete flow director rules for the VF, e.g.: > > for i in {1..128}; do > q=$(( i % 16 )) > ethtool -N ens802f0v0 flow-type tcp4 dst-port "$i" action "$q" > done > > for i in {0..127}; do > ethtool -N ens802f0v0 delete "$i" > done > > Suggested-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> > Suggested-by: Michal Swiatkowski <michal.swiatkowski@intel.com> > Acked-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> > Signed-off-by: Michal Kubiak <michal.kubiak@intel.com> Reviewed-by: Simon Horman <horms@kernel.org>
On Fri, Mar 21, 2025 at 04:13:57PM +0100, Michal Kubiak wrote: > The "ice" driver implementation uses the control VSI to handle > the flow director configuration for PFs and VFs. > > Unfortunately, although a separate VSI type was created to handle flow > director queues, the Rx queue handler was shared between the flow > director and a standard NAPI Rx handler. > > Such a design approach was not very flexible. First, it mixed hotpath > and slowpath code, blocking their further optimization. It also created > a huge overkill for the flow director command processing, which is > descriptor-based only, so there is no need to allocate Rx data buffers. > > For the above reasons, implement a separate Rx handler for the control > VSI. Also, remove from the NAPI handler the code dedicated to > configuring the flow director rules on VFs. > Do not allocate Rx data buffers to the flow director queues because > their processing is descriptor-based only. > Finally, allow Rx data queues to be allocated only for VSIs that have > netdev assigned to them. > > This handler splitting approach is the first step in converting the > driver to use the Page Pool (which can only be used for data queues). > > Test hints: > 1. Create a VF for any PF managed by the ice driver. > 2. In a loop, add and delete flow director rules for the VF, e.g.: > > for i in {1..128}; do > q=$(( i % 16 )) > ethtool -N ens802f0v0 flow-type tcp4 dst-port "$i" action "$q" > done > > for i in {0..127}; do > ethtool -N ens802f0v0 delete "$i" > done > > Suggested-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> > Suggested-by: Michal Swiatkowski <michal.swiatkowski@intel.com> > Acked-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> > Signed-off-by: Michal Kubiak <michal.kubiak@intel.com> > --- (...) > diff --git a/drivers/net/ethernet/intel/ice/ice_txrx.h b/drivers/net/ethernet/intel/ice/ice_txrx.h > index 4b63081629d0..041768df0b23 100644 > --- a/drivers/net/ethernet/intel/ice/ice_txrx.h > +++ b/drivers/net/ethernet/intel/ice/ice_txrx.h > @@ -492,6 +492,7 @@ static inline unsigned int ice_rx_pg_order(struct ice_rx_ring *ring) > > union ice_32b_rx_flex_desc; > > +void ice_init_ctrl_rx_descs(struct ice_rx_ring *rx_ring, u32 num_descs); > bool ice_alloc_rx_bufs(struct ice_rx_ring *rxr, unsigned int cleaned_count); > netdev_tx_t ice_start_xmit(struct sk_buff *skb, struct net_device *netdev); > u16 > @@ -512,4 +513,5 @@ ice_prgm_fdir_fltr(struct ice_vsi *vsi, struct ice_fltr_desc *fdir_desc, > u8 *raw_packet); > int ice_clean_rx_irq(struct ice_rx_ring *rx_ring, int budget); this can now become static as it's not called from ice_lib.c anymore. can you address this in v2 if this patch got lost and has not been merged yet? > void ice_clean_ctrl_tx_irq(struct ice_tx_ring *tx_ring); > +void ice_clean_ctrl_rx_irq(struct ice_rx_ring *rx_ring); > #endif /* _ICE_TXRX_H_ */ > -- > 2.45.2 >
On Wed, May 14, 2025 at 12:26:40PM +0200, Maciej Fijalkowski wrote: > On Fri, Mar 21, 2025 at 04:13:57PM +0100, Michal Kubiak wrote: > > The "ice" driver implementation uses the control VSI to handle > > the flow director configuration for PFs and VFs. > > > > Unfortunately, although a separate VSI type was created to handle flow > > director queues, the Rx queue handler was shared between the flow > > director and a standard NAPI Rx handler. > > > > Such a design approach was not very flexible. First, it mixed hotpath > > and slowpath code, blocking their further optimization. It also created > > a huge overkill for the flow director command processing, which is > > descriptor-based only, so there is no need to allocate Rx data buffers. > > > > For the above reasons, implement a separate Rx handler for the control > > VSI. Also, remove from the NAPI handler the code dedicated to > > configuring the flow director rules on VFs. > > Do not allocate Rx data buffers to the flow director queues because > > their processing is descriptor-based only. > > Finally, allow Rx data queues to be allocated only for VSIs that have > > netdev assigned to them. > > > > This handler splitting approach is the first step in converting the > > driver to use the Page Pool (which can only be used for data queues). > > > > Test hints: > > 1. Create a VF for any PF managed by the ice driver. > > 2. In a loop, add and delete flow director rules for the VF, e.g.: > > > > for i in {1..128}; do > > q=$(( i % 16 )) > > ethtool -N ens802f0v0 flow-type tcp4 dst-port "$i" action "$q" > > done > > > > for i in {0..127}; do > > ethtool -N ens802f0v0 delete "$i" > > done > > > > Suggested-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> > > Suggested-by: Michal Swiatkowski <michal.swiatkowski@intel.com> > > Acked-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> > > Signed-off-by: Michal Kubiak <michal.kubiak@intel.com> > > --- > > (...) > > > diff --git a/drivers/net/ethernet/intel/ice/ice_txrx.h b/drivers/net/ethernet/intel/ice/ice_txrx.h > > index 4b63081629d0..041768df0b23 100644 > > --- a/drivers/net/ethernet/intel/ice/ice_txrx.h > > +++ b/drivers/net/ethernet/intel/ice/ice_txrx.h > > @@ -492,6 +492,7 @@ static inline unsigned int ice_rx_pg_order(struct ice_rx_ring *ring) > > > > union ice_32b_rx_flex_desc; > > > > +void ice_init_ctrl_rx_descs(struct ice_rx_ring *rx_ring, u32 num_descs); > > bool ice_alloc_rx_bufs(struct ice_rx_ring *rxr, unsigned int cleaned_count); > > netdev_tx_t ice_start_xmit(struct sk_buff *skb, struct net_device *netdev); > > u16 > > @@ -512,4 +513,5 @@ ice_prgm_fdir_fltr(struct ice_vsi *vsi, struct ice_fltr_desc *fdir_desc, > > u8 *raw_packet); > > int ice_clean_rx_irq(struct ice_rx_ring *rx_ring, int budget); > > this can now become static as it's not called from ice_lib.c anymore. > can you address this in v2 if this patch got lost and has not been merged > yet? > Makes sense. OK, I will prepare a quick v2 with this change only. Thanks, Michal > > void ice_clean_ctrl_tx_irq(struct ice_tx_ring *tx_ring); > > +void ice_clean_ctrl_rx_irq(struct ice_rx_ring *rx_ring); > > #endif /* _ICE_TXRX_H_ */ > > -- > > 2.45.2 > >
diff --git a/drivers/net/ethernet/intel/ice/ice_base.c b/drivers/net/ethernet/intel/ice/ice_base.c index b94c56a82cc5..8bc333b794eb 100644 --- a/drivers/net/ethernet/intel/ice/ice_base.c +++ b/drivers/net/ethernet/intel/ice/ice_base.c @@ -717,7 +717,10 @@ static int ice_vsi_cfg_rxq(struct ice_rx_ring *ring) return 0; } - ice_alloc_rx_bufs(ring, num_bufs); + if (ring->vsi->type == ICE_VSI_CTRL) + ice_init_ctrl_rx_descs(ring, num_bufs); + else + ice_alloc_rx_bufs(ring, num_bufs); return 0; } diff --git a/drivers/net/ethernet/intel/ice/ice_lib.c b/drivers/net/ethernet/intel/ice/ice_lib.c index 6392d27861e5..b857717441c3 100644 --- a/drivers/net/ethernet/intel/ice/ice_lib.c +++ b/drivers/net/ethernet/intel/ice/ice_lib.c @@ -497,8 +497,7 @@ static irqreturn_t ice_msix_clean_ctrl_vsi(int __always_unused irq, void *data) if (!q_vector->tx.tx_ring) return IRQ_HANDLED; -#define FDIR_RX_DESC_CLEAN_BUDGET 64 - ice_clean_rx_irq(q_vector->rx.rx_ring, FDIR_RX_DESC_CLEAN_BUDGET); + ice_clean_ctrl_rx_irq(q_vector->rx.rx_ring); ice_clean_ctrl_tx_irq(q_vector->tx.tx_ring); return IRQ_HANDLED; diff --git a/drivers/net/ethernet/intel/ice/ice_txrx.c b/drivers/net/ethernet/intel/ice/ice_txrx.c index 69cdbd2eda07..72980c504a92 100644 --- a/drivers/net/ethernet/intel/ice/ice_txrx.c +++ b/drivers/net/ethernet/intel/ice/ice_txrx.c @@ -20,7 +20,6 @@ #define ICE_RX_HDR_SIZE 256 -#define FDIR_DESC_RXDID 0x40 #define ICE_FDIR_CLEAN_DELAY 10 /** @@ -775,6 +774,37 @@ ice_alloc_mapped_page(struct ice_rx_ring *rx_ring, struct ice_rx_buf *bi) return true; } +/** + * ice_init_ctrl_rx_descs - Initialize Rx descriptors for control vsi. + * @rx_ring: ring to init descriptors on + * @count: number of descriptors to initialize + */ +void ice_init_ctrl_rx_descs(struct ice_rx_ring *rx_ring, u32 count) +{ + union ice_32b_rx_flex_desc *rx_desc; + u32 ntu = rx_ring->next_to_use; + + if (!count) + return; + + rx_desc = ICE_RX_DESC(rx_ring, ntu); + + do { + rx_desc++; + ntu++; + if (unlikely(ntu == rx_ring->count)) { + rx_desc = ICE_RX_DESC(rx_ring, 0); + ntu = 0; + } + + rx_desc->wb.status_error0 = 0; + count--; + } while (count); + + if (rx_ring->next_to_use != ntu) + ice_release_rx_desc(rx_ring, ntu); +} + /** * ice_alloc_rx_bufs - Replace used receive buffers * @rx_ring: ring to place buffers on @@ -795,8 +825,7 @@ bool ice_alloc_rx_bufs(struct ice_rx_ring *rx_ring, unsigned int cleaned_count) struct ice_rx_buf *bi; /* do nothing if no valid netdev defined */ - if ((!rx_ring->netdev && rx_ring->vsi->type != ICE_VSI_CTRL) || - !cleaned_count) + if (!rx_ring->netdev || !cleaned_count) return false; /* get the Rx descriptor and buffer based on next_to_use */ @@ -1252,6 +1281,45 @@ static void ice_put_rx_mbuf(struct ice_rx_ring *rx_ring, struct xdp_buff *xdp, rx_ring->nr_frags = 0; } +/** + * ice_clean_ctrl_rx_irq - Clean descriptors from flow director Rx ring + * @rx_ring: Rx descriptor ring for ctrl_vsi to transact packets on + * + * This function cleans Rx descriptors from the ctrl_vsi Rx ring used + * to set flow director rules on VFs. + */ +void ice_clean_ctrl_rx_irq(struct ice_rx_ring *rx_ring) +{ + u32 ntc = rx_ring->next_to_clean; + unsigned int total_rx_pkts = 0; + u32 cnt = rx_ring->count; + + while (likely(total_rx_pkts < ICE_DFLT_IRQ_WORK)) { + struct ice_vsi *ctrl_vsi = rx_ring->vsi; + union ice_32b_rx_flex_desc *rx_desc; + u16 stat_err_bits; + + rx_desc = ICE_RX_DESC(rx_ring, ntc); + + stat_err_bits = BIT(ICE_RX_FLEX_DESC_STATUS0_DD_S); + if (!ice_test_staterr(rx_desc->wb.status_error0, stat_err_bits)) + break; + + dma_rmb(); + + if (ctrl_vsi->vf) + ice_vc_fdir_irq_handler(ctrl_vsi, rx_desc); + + if (++ntc == cnt) + ntc = 0; + total_rx_pkts++; + } + + rx_ring->first_desc = ntc; + rx_ring->next_to_clean = ntc; + ice_init_ctrl_rx_descs(rx_ring, ICE_RX_DESC_UNUSED(rx_ring)); +} + /** * ice_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf * @rx_ring: Rx descriptor ring to transact packets on @@ -1311,17 +1379,6 @@ int ice_clean_rx_irq(struct ice_rx_ring *rx_ring, int budget) dma_rmb(); ice_trace(clean_rx_irq, rx_ring, rx_desc); - if (rx_desc->wb.rxdid == FDIR_DESC_RXDID || !rx_ring->netdev) { - struct ice_vsi *ctrl_vsi = rx_ring->vsi; - - if (rx_desc->wb.rxdid == FDIR_DESC_RXDID && - ctrl_vsi->vf) - ice_vc_fdir_irq_handler(ctrl_vsi, rx_desc); - if (++ntc == cnt) - ntc = 0; - rx_ring->first_desc = ntc; - continue; - } size = le16_to_cpu(rx_desc->wb.pkt_len) & ICE_RX_FLX_DESC_PKT_LEN_M; diff --git a/drivers/net/ethernet/intel/ice/ice_txrx.h b/drivers/net/ethernet/intel/ice/ice_txrx.h index 4b63081629d0..041768df0b23 100644 --- a/drivers/net/ethernet/intel/ice/ice_txrx.h +++ b/drivers/net/ethernet/intel/ice/ice_txrx.h @@ -492,6 +492,7 @@ static inline unsigned int ice_rx_pg_order(struct ice_rx_ring *ring) union ice_32b_rx_flex_desc; +void ice_init_ctrl_rx_descs(struct ice_rx_ring *rx_ring, u32 num_descs); bool ice_alloc_rx_bufs(struct ice_rx_ring *rxr, unsigned int cleaned_count); netdev_tx_t ice_start_xmit(struct sk_buff *skb, struct net_device *netdev); u16 @@ -512,4 +513,5 @@ ice_prgm_fdir_fltr(struct ice_vsi *vsi, struct ice_fltr_desc *fdir_desc, u8 *raw_packet); int ice_clean_rx_irq(struct ice_rx_ring *rx_ring, int budget); void ice_clean_ctrl_tx_irq(struct ice_tx_ring *tx_ring); +void ice_clean_ctrl_rx_irq(struct ice_rx_ring *rx_ring); #endif /* _ICE_TXRX_H_ */
The "ice" driver implementation uses the control VSI to handle the flow director configuration for PFs and VFs. Unfortunately, although a separate VSI type was created to handle flow director queues, the Rx queue handler was shared between the flow director and a standard NAPI Rx handler. Such a design approach was not very flexible. First, it mixed hotpath and slowpath code, blocking their further optimization. It also created a huge overkill for the flow director command processing, which is descriptor-based only, so there is no need to allocate Rx data buffers. For the above reasons, implement a separate Rx handler for the control VSI. Also, remove from the NAPI handler the code dedicated to configuring the flow director rules on VFs. Do not allocate Rx data buffers to the flow director queues because their processing is descriptor-based only. Finally, allow Rx data queues to be allocated only for VSIs that have netdev assigned to them. This handler splitting approach is the first step in converting the driver to use the Page Pool (which can only be used for data queues). Test hints: 1. Create a VF for any PF managed by the ice driver. 2. In a loop, add and delete flow director rules for the VF, e.g.: for i in {1..128}; do q=$(( i % 16 )) ethtool -N ens802f0v0 flow-type tcp4 dst-port "$i" action "$q" done for i in {0..127}; do ethtool -N ens802f0v0 delete "$i" done Suggested-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> Suggested-by: Michal Swiatkowski <michal.swiatkowski@intel.com> Acked-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> Signed-off-by: Michal Kubiak <michal.kubiak@intel.com> --- drivers/net/ethernet/intel/ice/ice_base.c | 5 +- drivers/net/ethernet/intel/ice/ice_lib.c | 3 +- drivers/net/ethernet/intel/ice/ice_txrx.c | 85 +++++++++++++++++++---- drivers/net/ethernet/intel/ice/ice_txrx.h | 2 + 4 files changed, 78 insertions(+), 17 deletions(-)