igc: Add support for AF_XDP zero-copy
Subsystems (3)
cover igc: Add support for AF_XDP zero-copy
Hi,
This series adds AF_XDP zero-copy feature to igc driver.
The initial patches do some code refactoring, preparing the code base to land the AF_XDP zero-copy feature, avoiding code duplications. The last patches of the series are the ones implementing the feature.
The last patch which indeed implements AF_XDP zero-copy support was originally way too lengthy so, for the sake of code review, I broke it up into two patches: one adding support for the RX functionality and the other one adding TX support. Feel free to squash those two patches when pushing the series to net-next if you find appropriate.
Best,
Andre
Andre Guedes (10):
igc: Move igc_xdp_is_enabled()
igc: Refactor igc_xdp_run_prog()
igc: Refactor igc_clean_rx_ring()
igc: Refactor XDP rxq info registration
igc: Introduce igc_update_rx_stats()
igc: Introduce igc_update_tx_stats()
igc: Introduce igc_unmap_tx_buffer() helper
igc: Replace IGC_TX_FLAGS_XDP flag by an enum
igc: Enable RX via AF_XDP zero-copy
igc: Enable TX via AF_XDP zero-copy
drivers/net/ethernet/intel/igc/igc.h | 33 +- drivers/net/ethernet/intel/igc/igc_base.h | 2 + drivers/net/ethernet/intel/igc/igc_main.c | 650 ++++++++++++++++++---- drivers/net/ethernet/intel/igc/igc_xdp.c | 107 +++- drivers/net/ethernet/intel/igc/igc_xdp.h | 8 +- 5 files changed, 672 insertions(+), 128 deletions(-)
Signature · 1 line
2.29.2
PATCH 1/10 igc: Move igc_xdp_is_enabled()
This patch moves the helper igc_xdp_is_enabled() to igc_xdp.h so it can be reused in igc_xdp.c by upcoming patches that will introduce AF_XDP zero-copy support to the driver.
--- drivers/net/ethernet/intel/igc/igc_main.c | 5 ----- drivers/net/ethernet/intel/igc/igc_xdp.h | 5 +++++ 2 files changed, 5 insertions(+), 5 deletions(-)
drivers/net/ethernet/intel/igc/igc_main.c +0 -5
@@ -515,11 +515,6 @@ static int igc_setup_all_rx_resources(struct igc_adapter *adapter)return err; } -static bool igc_xdp_is_enabled (struct igc_adapter *adapter) -{ -return !!adapter->xdp_prog; -} -/** * igc_configure_rx_ring - Configure a receive ring after Reset * @adapter: board private structure
drivers/net/ethernet/intel/igc/igc_xdp.h +5 -0
@@ -10,4 +10,9 @@ int igc_xdp_set_prog(struct igc_adapter *adapter, struct bpf_prog *prog,int igc_xdp_register_rxq_info (struct igc_ring *ring);void igc_xdp_unregister_rxq_info (struct igc_ring *ring); +static inline bool igc_xdp_is_enabled (struct igc_adapter *adapter) +{ +return !!adapter->xdp_prog; +} +#endif /* _IGC_XDP_H_ */
Signature · 1 line
2.29.2
PATCH 2/10 igc: Refactor igc_xdp_run_prog()
In reply to Andre GuedesThis patch refactors igc_xdp_run_prog() helper, preparing the code for AF_XDP zero-copy support which is added by upcoming patches.
With AF_XDP zero-copy support, igc_poll() will have a dedicated path when the rx ring has zero-copy is enabled. To avoid code duplication as much as possible, this patch encapsulates the code specific to handling XDP program actions into a local helper so it can be reused by the zero-copy path.
--- drivers/net/ethernet/intel/igc/igc_main.c | 56 +++++++++++------------ 1 file changed, 28 insertions(+), 28 deletions(-)
drivers/net/ethernet/intel/igc/igc_main.c +28 -28
@@ -2029,38 +2029,22 @@ static int igc_xdp_xmit_back(struct igc_adapter *adapter, struct xdp_buff *xdp)return res; } -static struct sk_buff *igc_xdp_run_prog (struct igc_adapter *adapter, -struct xdp_buff *xdp) +/* This function assumes rcu_read_lock() is held by the caller. */ +static int igc_xdp_do_run_prog (struct igc_adapter *adapter, +struct bpf_prog *prog, +struct xdp_buff *xdp) { -struct bpf_prog *prog; -int res; -u32 act; - -rcu_read_lock (); - - prog =READ_ONCE (adapter->xdp_prog); -if (!prog) { - res =IGC_XDP_PASS ; - goto unlock; - } +u32 act =bpf_prog_run_xdp (prog, xdp); - act =bpf_prog_run_xdp (prog, xdp);switch (act) {case XDP_PASS : - res =IGC_XDP_PASS ; -break ; +return IGC_XDP_PASS ;case XDP_TX : -if (igc_xdp_xmit_back (adapter, xdp) <0 ) - res =IGC_XDP_CONSUMED ; -else - res =IGC_XDP_TX ; -break ; +return igc_xdp_xmit_back (adapter, xdp) <0 ? +IGC_XDP_CONSUMED :IGC_XDP_TX ;case XDP_REDIRECT : -if (xdp_do_redirect (adapter->netdev, xdp, prog) <0 ) - res =IGC_XDP_CONSUMED ; -else - res =IGC_XDP_REDIRECT ; -break ; +return xdp_do_redirect (adapter->netdev, xdp, prog) <0 ? +IGC_XDP_CONSUMED :IGC_XDP_REDIRECT ;default :bpf_warn_invalid_xdp_action (act); fallthrough;
@@ -2068,9 +2052,25 @@ static struct sk_buff *igc_xdp_run_prog(struct igc_adapter *adapter,trace_xdp_exception (adapter->netdev, prog, act); fallthrough;case XDP_DROP : - res =IGC_XDP_CONSUMED ; -break ; +return IGC_XDP_CONSUMED ; } +} + +static struct sk_buff *igc_xdp_run_prog (struct igc_adapter *adapter, +struct xdp_buff *xdp) +{ +struct bpf_prog *prog; +int res; + +rcu_read_lock (); + + prog =READ_ONCE (adapter->xdp_prog); +if (!prog) { + res =IGC_XDP_PASS ; + goto unlock; + } + + res =igc_xdp_do_run_prog (adapter, prog, xdp); unlock:rcu_read_unlock ();
Signature · 1 line
2.29.2
PATCH 3/10 igc: Refactor igc_clean_rx_ring()
In reply to Andre GuedesThis patch refactors igc_clean_rx_ring() helper, preparing the code for AF_XDP zero-copy support which is added by upcoming patches.
The refactor consists of encapsulating page-shared specific code into its own helper and leave common code that will be shared by both page-shared and xsk pool in igc_clean_rx_ring().
--- drivers/net/ethernet/intel/igc/igc_main.c | 25 ++++++++++++++--------- 1 file changed, 15 insertions(+), 10 deletions(-)
drivers/net/ethernet/intel/igc/igc_main.c +15 -10
@@ -346,16 +346,10 @@ static int igc_setup_all_tx_resources(struct igc_adapter *adapter)return err; } -/** -* igc_clean_rx_ring - Free Rx Buffers per Queue -* @rx_ring: ring to free buffers from -*/ -static void igc_clean_rx_ring (struct igc_ring *rx_ring) +static void igc_clean_rx_ring_page_shared (struct igc_ring *rx_ring) {u16 i = rx_ring->next_to_clean; -clear_ring_uses_large_buffer (rx_ring); -dev_kfree_skb (rx_ring->skb); rx_ring->skb =NULL ;
@@ -385,10 +379,21 @@ static void igc_clean_rx_ring(struct igc_ring *rx_ring)if (i == rx_ring->count) i =0 ; } +} - rx_ring->next_to_alloc =0 ; - rx_ring->next_to_clean =0 ; - rx_ring->next_to_use =0 ; +/** +* igc_clean_rx_ring - Free Rx Buffers per Queue +* @ring: ring to free buffers from +*/ +static void igc_clean_rx_ring (struct igc_ring *ring) +{ +clear_ring_uses_large_buffer (ring); + +igc_clean_rx_ring_page_shared (ring); + + ring->next_to_alloc =0 ; + ring->next_to_clean =0 ; + ring->next_to_use =0 ; }/**
Signature · 1 line
2.29.2
PATCH 4/10 igc: Refactor XDP rxq info registration
In reply to Andre GuedesThis patch does a refactoring on the XDP rxq info registration code, preparing the driver for AF_XDP zero-copy support which is added by upcoming patches.
Currently, xdp_rxq and memory model are both registered during RX resource setup time by igc_xdp_register_rxq_info() helper. With AF_XDP, we want to register the memory model later on while configuring the ring because we will know which memory model type to register (MEM_TYPE_PAGE_SHARED or MEM_TYPE_XSK_BUFF_POOL).
The helpers igc_xdp_register_rxq_info() and igc_xdp_unregister_rxq_ info() are not useful anymore so they are removed.
--- drivers/net/ethernet/intel/igc/igc_main.c | 14 ++++++++---- drivers/net/ethernet/intel/igc/igc_xdp.c | 27 ----------------------- drivers/net/ethernet/intel/igc/igc_xdp.h | 3 --- 3 files changed, 10 insertions(+), 34 deletions(-)
drivers/net/ethernet/intel/igc/igc_main.c +10 -4
@@ -419,7 +419,7 @@ void igc_free_rx_resources(struct igc_ring *rx_ring) {igc_clean_rx_ring (rx_ring); -igc_xdp_unregister_rxq_info (rx_ring); +xdp_rxq_info_unreg (&rx_ring->xdp_rxq);vfree (rx_ring->rx_buffer_info); rx_ring->rx_buffer_info =NULL ;
@@ -460,9 +460,12 @@ int igc_setup_rx_resources(struct igc_ring *rx_ring)struct device *dev = rx_ring->dev;int size, desc_len, res; - res =igc_xdp_register_rxq_info (rx_ring); -if (res <0 ) + res =xdp_rxq_info_reg (&rx_ring->xdp_rxq, ndev, rx_ring->queue_index, +0 ); +if (res <0 ) { +netdev_err (ndev,"Failed to register xdp rxq info\n" );return res; + } size =sizeof (struct igc_rx_buffer ) * rx_ring->count; rx_ring->rx_buffer_info =vzalloc (size);
@@ -488,7 +491,7 @@ int igc_setup_rx_resources(struct igc_ring *rx_ring)return 0 ; err: -igc_xdp_unregister_rxq_info (rx_ring); +xdp_rxq_info_unreg (&rx_ring->xdp_rxq);vfree (rx_ring->rx_buffer_info); rx_ring->rx_buffer_info =NULL ;netdev_err (ndev,"Unable to allocate memory for Rx descriptor ring\n" );
@@ -536,6 +539,9 @@ static void igc_configure_rx_ring(struct igc_adapter *adapter,u32 srrctl =0 , rxdctl =0 ;u64 rdba = ring->dma; +WARN_ON (xdp_rxq_info_reg_mem_model (&ring->xdp_rxq, +MEM_TYPE_PAGE_SHARED ,NULL )); +if (igc_xdp_is_enabled (adapter))set_ring_uses_large_buffer (ring);
drivers/net/ethernet/intel/igc/igc_xdp.c +0 -27
@@ -31,30 +31,3 @@ int igc_xdp_set_prog(struct igc_adapter *adapter, struct bpf_prog *prog,return 0 ; } - -int igc_xdp_register_rxq_info (struct igc_ring *ring) -{ -struct net_device *dev = ring->netdev; -int err; - - err =xdp_rxq_info_reg (&ring->xdp_rxq, dev, ring->queue_index,0 ); -if (err) { -netdev_err (dev,"Failed to register xdp rxq info\n" ); -return err; - } - - err =xdp_rxq_info_reg_mem_model (&ring->xdp_rxq,MEM_TYPE_PAGE_SHARED , -NULL ); -if (err) { -netdev_err (dev,"Failed to register xdp rxq mem model\n" ); -xdp_rxq_info_unreg (&ring->xdp_rxq); -return err; - } - -return 0 ; -} - -void igc_xdp_unregister_rxq_info (struct igc_ring *ring) -{ -xdp_rxq_info_unreg (&ring->xdp_rxq); -}
drivers/net/ethernet/intel/igc/igc_xdp.h +0 -3
@@ -7,9 +7,6 @@int igc_xdp_set_prog (struct igc_adapter *adapter,struct bpf_prog *prog,struct netlink_ext_ack *extack); -int igc_xdp_register_rxq_info (struct igc_ring *ring); -void igc_xdp_unregister_rxq_info (struct igc_ring *ring); -static inline bool igc_xdp_is_enabled (struct igc_adapter *adapter) {return !!adapter->xdp_prog;
Signature · 1 line
2.29.2
PATCH 5/10 igc: Introduce igc_update_rx_stats()
In reply to Andre GuedesIn preparation for AF_XDP zero-copy support, this patch encapsulates the code that updates the driver RX stats so it can be reused in the zero-copy path.
--- drivers/net/ethernet/intel/igc/igc_main.c | 21 +++++++++++++++------ 1 file changed, 15 insertions(+), 6 deletions(-)
drivers/net/ethernet/intel/igc/igc_main.c +15 -6
@@ -2118,6 +2118,20 @@ static void igc_finalize_xdp(struct igc_adapter *adapter, int status)xdp_do_flush (); } +static void igc_update_rx_stats (struct igc_q_vector *q_vector, +unsigned int packets,unsigned int bytes) +{ +struct igc_ring *ring = q_vector->rx.ring; + +u64_stats_update_begin (&ring->rx_syncp); + ring->rx_stats.packets += packets; + ring->rx_stats.bytes += bytes; +u64_stats_update_end (&ring->rx_syncp); + + q_vector->rx.total_packets += packets; + q_vector->rx.total_bytes += bytes; +} +static int igc_clean_rx_irq (struct igc_q_vector *q_vector,const int budget) {unsigned int total_bytes =0 , total_packets =0 ;
@@ -2241,12 +2255,7 @@ static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget)/* place incomplete frames back on ring for completion */ rx_ring->skb = skb; -u64_stats_update_begin (&rx_ring->rx_syncp); - rx_ring->rx_stats.packets += total_packets; - rx_ring->rx_stats.bytes += total_bytes; -u64_stats_update_end (&rx_ring->rx_syncp); - q_vector->rx.total_packets += total_packets; - q_vector->rx.total_bytes += total_bytes; +igc_update_rx_stats (q_vector, total_packets, total_bytes);if (cleaned_count)igc_alloc_rx_buffers (rx_ring, cleaned_count);
Signature · 1 line
2.29.2
PATCH 6/10 igc: Introduce igc_update_tx_stats()
In reply to Andre GuedesLikewise we do with the RX stats, this patch encapsulates the code that updates the driver TX stats in its own local helper.
--- drivers/net/ethernet/intel/igc/igc_main.c | 22 ++++++++++++++++------ 1 file changed, 16 insertions(+), 6 deletions(-)
drivers/net/ethernet/intel/igc/igc_main.c +16 -6
@@ -2263,6 +2263,20 @@ static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget)return total_packets; } +static void igc_update_tx_stats (struct igc_q_vector *q_vector, +unsigned int packets,unsigned int bytes) +{ +struct igc_ring *ring = q_vector->tx.ring; + +u64_stats_update_begin (&ring->tx_syncp); + ring->tx_stats.bytes += bytes; + ring->tx_stats.packets += packets; +u64_stats_update_end (&ring->tx_syncp); + + q_vector->tx.total_bytes += bytes; + q_vector->tx.total_packets += packets; +} +/** * igc_clean_tx_irq - Reclaim resources after transmit completes * @q_vector: pointer to q_vector containing needed info
@@ -2365,12 +2379,8 @@ static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget) i += tx_ring->count; tx_ring->next_to_clean = i; -u64_stats_update_begin (&tx_ring->tx_syncp); - tx_ring->tx_stats.bytes += total_bytes; - tx_ring->tx_stats.packets += total_packets; -u64_stats_update_end (&tx_ring->tx_syncp); - q_vector->tx.total_bytes += total_bytes; - q_vector->tx.total_packets += total_packets; + +igc_update_tx_stats (q_vector, total_packets, total_bytes);if (test_bit (IGC_RING_FLAG_TX_DETECT_HANG , &tx_ring->flags)) {struct igc_hw *hw = &adapter->hw;
Signature · 1 line
2.29.2
PATCH 7/10 igc: Introduce igc_unmap_tx_buffer() helper
In reply to Andre GuedesIn preparation for AF_XDP zero-copy support, this patch encapsulates the code that unmaps tx buffers into its own local helper so we can reuse it when adding zero-copy support, avoiding code duplication.
--- drivers/net/ethernet/intel/igc/igc_main.c | 49 +++++++---------------- 1 file changed, 15 insertions(+), 34 deletions(-)
drivers/net/ethernet/intel/igc/igc_main.c +15 -34
@@ -171,6 +171,14 @@ static void igc_get_hw_control(struct igc_adapter *adapter) ctrl_ext | IGC_CTRL_EXT_DRV_LOAD); } +static void igc_unmap_tx_buffer (struct device *dev,struct igc_tx_buffer *buf) +{ +dma_unmap_single (dev,dma_unmap_addr (buf, dma), +dma_unmap_len (buf, len),DMA_TO_DEVICE ); + +dma_unmap_len_set (buf, len,0 ); +} +/** * igc_clean_tx_ring - Free Tx Buffers * @tx_ring: ring to be cleaned
@@ -188,11 +196,7 @@ static void igc_clean_tx_ring(struct igc_ring *tx_ring)else dev_kfree_skb_any (tx_buffer->skb); -/* unmap skb header data */ -dma_unmap_single (tx_ring->dev, -dma_unmap_addr (tx_buffer, dma), -dma_unmap_len (tx_buffer, len), -DMA_TO_DEVICE ); +igc_unmap_tx_buffer (tx_ring->dev, tx_buffer);/* check for eop_desc to determine the end of the packet */ eop_desc = tx_buffer->next_to_watch;
@@ -211,10 +215,7 @@ static void igc_clean_tx_ring(struct igc_ring *tx_ring)/* unmap any remaining paged data */ if (dma_unmap_len (tx_buffer, len)) -dma_unmap_page (tx_ring->dev, -dma_unmap_addr (tx_buffer, dma), -dma_unmap_len (tx_buffer, len), -DMA_TO_DEVICE ); +igc_unmap_tx_buffer (tx_ring->dev, tx_buffer); }/* move us one more past the eop_desc for start of next pkt */
@@ -1229,11 +1230,7 @@ static int igc_tx_map(struct igc_ring *tx_ring,/* clear dma mappings for failed tx_buffer_info map */ while (tx_buffer != first) {if (dma_unmap_len (tx_buffer, len)) -dma_unmap_page (tx_ring->dev, -dma_unmap_addr (tx_buffer, dma), -dma_unmap_len (tx_buffer, len), -DMA_TO_DEVICE ); -dma_unmap_len_set (tx_buffer, len,0 ); +igc_unmap_tx_buffer (tx_ring->dev, tx_buffer);if (i-- ==0 ) i += tx_ring->count;
@@ -1241,11 +1238,7 @@ static int igc_tx_map(struct igc_ring *tx_ring, }if (dma_unmap_len (tx_buffer, len)) -dma_unmap_single (tx_ring->dev, -dma_unmap_addr (tx_buffer, dma), -dma_unmap_len (tx_buffer, len), -DMA_TO_DEVICE ); -dma_unmap_len_set (tx_buffer, len,0 ); +igc_unmap_tx_buffer (tx_ring->dev, tx_buffer);dev_kfree_skb_any (tx_buffer->skb); tx_buffer->skb =NULL ;
@@ -2327,14 +2320,7 @@ static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget)else napi_consume_skb (tx_buffer->skb, napi_budget); -/* unmap skb header data */ -dma_unmap_single (tx_ring->dev, -dma_unmap_addr (tx_buffer, dma), -dma_unmap_len (tx_buffer, len), -DMA_TO_DEVICE ); - -/* clear tx_buffer data */ -dma_unmap_len_set (tx_buffer, len,0 ); +igc_unmap_tx_buffer (tx_ring->dev, tx_buffer);/* clear last DMA location and unmap remaining buffers */ while (tx_desc != eop_desc) {
@@ -2348,13 +2334,8 @@ static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget) }/* unmap any remaining paged data */ -if (dma_unmap_len (tx_buffer, len)) { -dma_unmap_page (tx_ring->dev, -dma_unmap_addr (tx_buffer, dma), -dma_unmap_len (tx_buffer, len), -DMA_TO_DEVICE ); -dma_unmap_len_set (tx_buffer, len,0 ); - } +if (dma_unmap_len (tx_buffer, len)) +igc_unmap_tx_buffer (tx_ring->dev, tx_buffer); }/* move us one more past the eop_desc for start of next pkt */
Signature · 1 line
2.29.2
PATCH 8/10 igc: Replace IGC_TX_FLAGS_XDP flag by an enum
In reply to Andre GuedesUp to this point, tx buffers are associated with either a skb or a xdpf, and the IGC_TX_FLAGS_XDP flag was enough to distinguish between these two case. However, with upcoming patches that will add AF_XDP zero-copy support, a third case will be introduced so this flag-based approach won't fit well.
In preparation to land AF_XDP zero-copy support, this patch replaces the IGC_TX_FLAGS_XDP flag by an enum which will be extended once zero-copy support is introduced to the driver.
--- drivers/net/ethernet/intel/igc/igc.h | 8 +++++-- drivers/net/ethernet/intel/igc/igc_main.c | 27 ++++++++++++++++++----- 2 files changed, 28 insertions(+), 7 deletions(-)
drivers/net/ethernet/intel/igc/igc.h +6 -2
@@ -376,8 +376,6 @@ enum igc_tx_flags {/* olinfo flags */ IGC_TX_FLAGS_IPV4 =0x10 , IGC_TX_FLAGS_CSUM = 0x20, - - IGC_TX_FLAGS_XDP = 0x100, }; enum igc_boards {
@@ -394,12 +392,18 @@ enum igc_boards {#define TXD_USE_COUNT (S ) DIV_ROUND_UP((S), IGC_MAX_DATA_PER_TXD)#define DESC_NEEDED (MAX_SKB_FRAGS + 4) +enum igc_tx_buffer_type { +IGC_TX_BUFFER_TYPE_SKB , +IGC_TX_BUFFER_TYPE_XDP , +}; +/* wrapper around a pointer to a socket buffer, * so a DMA handle can be stored along with the buffer */ struct igc_tx_buffer {union igc_adv_tx_desc *next_to_watch;unsigned long time_stamp; +enum igc_tx_buffer_type type;union {struct sk_buff *skb;struct xdp_frame *xdpf;
drivers/net/ethernet/intel/igc/igc_main.c +22 -5
@@ -191,10 +191,18 @@ static void igc_clean_tx_ring(struct igc_ring *tx_ring)while (i != tx_ring->next_to_use) {union igc_adv_tx_desc *eop_desc, *tx_desc; -if (tx_buffer->tx_flags &IGC_TX_FLAGS_XDP ) +switch (tx_buffer->type) { +case IGC_TX_BUFFER_TYPE_XDP :xdp_return_frame (tx_buffer->xdpf); -else +break ; +case IGC_TX_BUFFER_TYPE_SKB :dev_kfree_skb_any (tx_buffer->skb); +break ; +default : +netdev_warn_once (tx_ring->netdev, +"Unknown tx buffer type\n" ); +break ; + }igc_unmap_tx_buffer (tx_ring->dev, tx_buffer);
@@ -1371,6 +1379,7 @@ static netdev_tx_t igc_xmit_frame_ring(struct sk_buff *skb,/* record the location of the first descriptor for this packet */ first = &tx_ring->tx_buffer_info[tx_ring->next_to_use]; + first->type =IGC_TX_BUFFER_TYPE_SKB ; first->skb = skb; first->bytecount = skb->len; first->gso_segs =1 ;
@@ -1950,8 +1959,8 @@ static int igc_xdp_init_tx_buffer(struct igc_tx_buffer *buffer,return -ENOMEM ; } + buffer->type =IGC_TX_BUFFER_TYPE_XDP ; buffer->xdpf = xdpf; - buffer->tx_flags =IGC_TX_FLAGS_XDP ; buffer->protocol =0 ; buffer->bytecount = xdpf->len; buffer->gso_segs =1 ;
@@ -2315,10 +2324,18 @@ static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget) total_bytes += tx_buffer->bytecount; total_packets += tx_buffer->gso_segs; -if (tx_buffer->tx_flags &IGC_TX_FLAGS_XDP ) +switch (tx_buffer->type) { +case IGC_TX_BUFFER_TYPE_XDP :xdp_return_frame (tx_buffer->xdpf); -else +break ; +case IGC_TX_BUFFER_TYPE_SKB :napi_consume_skb (tx_buffer->skb, napi_budget); +break ; +default : +netdev_warn_once (tx_ring->netdev, +"Unknown tx buffer type\n" ); +break ; + }igc_unmap_tx_buffer (tx_ring->dev, tx_buffer);
Signature · 1 line
2.29.2
PATCH 9/10 igc: Enable RX via AF_XDP zero-copy
In reply to Andre GuedesThis patch adds support for receiving packets via AF_XDP zero-copy mechanism.
A new flag is added to 'enum igc_ring_flags_t' to indicate the ring has AF_XDP zero-copy enabled so proper ring setup is carried out during ring configuration in igc_configure_rx_ring().
RX buffers can now be allocated via the shared pages mechanism (default behavior of the driver) or via xsk pool (when AF_XDP zero-copy is enabled) so a union is added to the 'struct igc_rx_buffer' to cover both cases.
When AF_XDP zero-copy is enabled, rx buffers are allocated from the xsk pool using the new helper igc_alloc_rx_buffers_zc() which is the counterpart of igc_alloc_rx_buffers().
Likewise other Intel drivers that support AF_XDP zero-copy, in igc we have a dedicated path for cleaning up rx irqs when zero-copy is enabled. This avoids adding too many checks within igc_clean_rx_irq(), resulting in a more readable and efficient code since this function is called from the hot-path of the driver.
--- drivers/net/ethernet/intel/igc/igc.h | 22 +- drivers/net/ethernet/intel/igc/igc_base.h | 1 + drivers/net/ethernet/intel/igc/igc_main.c | 334 +++++++++++++++++++++- drivers/net/ethernet/intel/igc/igc_xdp.c | 98 +++++++ drivers/net/ethernet/intel/igc/igc_xdp.h | 2 + 5 files changed, 438 insertions(+), 19 deletions(-)
drivers/net/ethernet/intel/igc/igc.h +16 -6
@@ -113,6 +113,7 @@ struct igc_ring { };struct xdp_rxq_info xdp_rxq; +struct xsk_buff_pool *xsk_pool; }____cacheline_internodealigned_in_smp ;/* Board specific private data structure */
@@ -241,6 +242,9 @@ bool igc_has_link(struct igc_adapter *adapter);void igc_reset (struct igc_adapter *adapter);int igc_set_spd_dplx (struct igc_adapter *adapter,u32 spd,u8 dplx);void igc_update_stats (struct igc_adapter *adapter); +void igc_disable_rx_ring (struct igc_ring *ring); +void igc_enable_rx_ring (struct igc_ring *ring); +int igc_xsk_wakeup (struct net_device *dev,u32 queue_id,u32 flags);/* igc_dump declarations */ void igc_rings_dump (struct igc_adapter *adapter);
@@ -418,14 +422,19 @@ struct igc_tx_buffer { };struct igc_rx_buffer { -dma_addr_t dma; -struct page *page; +union { +struct { +dma_addr_t dma; +struct page *page;#if (BITS_PER_LONG >32 ) || (PAGE_SIZE >=65536 ) -__u32 page_offset; +__u32 page_offset;#else -__u16 page_offset; +__u16 page_offset;#endif -__u16 pagecnt_bias; +__u16 pagecnt_bias; + }; +struct xdp_buff *xdp; + }; };struct igc_q_vector {
@@ -511,7 +520,8 @@ enum igc_ring_flags_t {IGC_RING_FLAG_RX_SCTP_CSUM ,IGC_RING_FLAG_RX_LB_VLAN_BSWAP ,IGC_RING_FLAG_TX_CTX_IDX , - IGC_RING_FLAG_TX_DETECT_HANG +IGC_RING_FLAG_TX_DETECT_HANG , + IGC_RING_FLAG_AF_XDP_ZC, };#define ring_uses_large_buffer (ring) \
drivers/net/ethernet/intel/igc/igc_base.h +1 -0
@@ -81,6 +81,7 @@ union igc_adv_rx_desc {/* Additional Receive Descriptor Control definitions */ #define IGC_RXDCTL_QUEUE_ENABLE 0x02000000/* Ena specific Rx Queue */ +#define IGC_RXDCTL_SWFLUSH 0x04000000/* Receive Software Flush */ /* SRRCTL bit definitions */ #define IGC_SRRCTL_BSIZEPKT_SHIFT 10/* Shift _right_ */
drivers/net/ethernet/intel/igc/igc_main.c +321 -13
425 lines not on this page · raw message
drivers/net/ethernet/intel/igc/igc_xdp.c +98 -0
107 lines not on this page · raw message
drivers/net/ethernet/intel/igc/igc_xdp.h +2 -0
@@ -6,6 +6,8 @@int igc_xdp_set_prog (struct igc_adapter *adapter,struct bpf_prog *prog,struct netlink_ext_ack *extack); +int igc_xdp_setup_pool (struct igc_adapter *adapter,struct xsk_buff_pool *pool, +u16 queue_id);static inline bool igc_xdp_is_enabled (struct igc_adapter *adapter) {
Signature · 1 line
2.29.2
PATCH 10/10 igc: Enable TX via AF_XDP zero-copy
In reply to Andre GuedesThis patch adds support for transmitting packets via AF_XDP zero-copy mechanism.
The packet transmission itself is implemented by igc_xdp_xmit_zc() which is called from igc_clean_tx_irq() when the ring has AF_XDP zero-copy enabled. Likewise i40e and ice drivers, the transmission budget used is the number of descriptors available on the ring.
A new tx buffer type is introduced to 'enum igc_tx_buffer_type' to indicate the tx buffer uses memory from xsk pool so it can be properly cleaned after transmission or when the ring is cleaned.
The I225 controller has only 4 Tx hardware queues so the main difference between igc and other Intel drivers that support AF_XDP zero-copy is that there is no tx ring dedicated exclusively to XDP. Instead, tx rings are shared between the network stack and XDP, and netdev queue lock is used to ensure mutual exclusion. This is the same approach implemented to support XDP_TX and XDP_REDIRECT actions.
--- drivers/net/ethernet/intel/igc/igc.h | 3 + drivers/net/ethernet/intel/igc/igc_base.h | 1 + drivers/net/ethernet/intel/igc/igc_main.c | 115 +++++++++++++++++++++- drivers/net/ethernet/intel/igc/igc_xdp.c | 20 +++- 4 files changed, 131 insertions(+), 8 deletions(-)
drivers/net/ethernet/intel/igc/igc.h +3 -0
@@ -244,6 +244,8 @@ int igc_set_spd_dplx(struct igc_adapter *adapter, u32 spd, u8 dplx);void igc_update_stats (struct igc_adapter *adapter);void igc_disable_rx_ring (struct igc_ring *ring);void igc_enable_rx_ring (struct igc_ring *ring); +void igc_disable_tx_ring (struct igc_ring *ring); +void igc_enable_tx_ring (struct igc_ring *ring);int igc_xsk_wakeup (struct net_device *dev,u32 queue_id,u32 flags);/* igc_dump declarations */
@@ -399,6 +401,7 @@ enum igc_boards {enum igc_tx_buffer_type {IGC_TX_BUFFER_TYPE_SKB ,IGC_TX_BUFFER_TYPE_XDP , +IGC_TX_BUFFER_TYPE_XSK , };/* wrapper around a pointer to a socket buffer,
drivers/net/ethernet/intel/igc/igc_base.h +1 -0
@@ -78,6 +78,7 @@ union igc_adv_rx_desc {/* Additional Transmit Descriptor Control definitions */ #define IGC_TXDCTL_QUEUE_ENABLE 0x02000000/* Ena specific Tx Queue */ +#define IGC_TXDCTL_SWFLUSH 0x04000000/* Transmit Software Flush */ /* Additional Receive Descriptor Control definitions */ #define IGC_RXDCTL_QUEUE_ENABLE 0x02000000/* Ena specific Rx Queue */
drivers/net/ethernet/intel/igc/igc_main.c +111 -4
190 lines not on this page · raw message
drivers/net/ethernet/intel/igc/igc_xdp.c +16 -4
@@ -39,13 +39,14 @@ static int igc_xdp_enable_pool(struct igc_adapter *adapter, {struct net_device *ndev = adapter->netdev;struct device *dev = &adapter->pdev->dev; -struct igc_ring *rx_ring; +struct igc_ring *rx_ring, *tx_ring;struct napi_struct *napi;bool needs_reset;u32 frame_size;int err; -if (queue_id >= adapter->num_rx_queues) +if (queue_id >= adapter->num_rx_queues || + queue_id >= adapter->num_tx_queues)return -EINVAL ; frame_size =xsk_pool_get_rx_frame_size (pool);
@@ -67,18 +68,23 @@ static int igc_xdp_enable_pool(struct igc_adapter *adapter, needs_reset =netif_running (adapter->netdev) &&igc_xdp_is_enabled (adapter); rx_ring = adapter->rx_ring[queue_id]; + tx_ring = adapter->tx_ring[queue_id]; +/* Rx and Tx rings share the same napi context. */ napi = &rx_ring->q_vector->napi;if (needs_reset) {igc_disable_rx_ring (rx_ring); +igc_disable_tx_ring (tx_ring);napi_disable (napi); }set_bit (IGC_RING_FLAG_AF_XDP_ZC , &rx_ring->flags); +set_bit (IGC_RING_FLAG_AF_XDP_ZC , &tx_ring->flags);if (needs_reset) {napi_enable (napi);igc_enable_rx_ring (rx_ring); +igc_enable_tx_ring (tx_ring); err =igc_xsk_wakeup (ndev, queue_id,XDP_WAKEUP_RX );if (err)
@@ -90,12 +96,13 @@ static int igc_xdp_enable_pool(struct igc_adapter *adapter,static int igc_xdp_disable_pool (struct igc_adapter *adapter,u16 queue_id) { +struct igc_ring *rx_ring, *tx_ring;struct xsk_buff_pool *pool; -struct igc_ring *rx_ring;struct napi_struct *napi;bool needs_reset; -if (queue_id >= adapter->num_rx_queues) +if (queue_id >= adapter->num_rx_queues || + queue_id >= adapter->num_tx_queues)return -EINVAL ; pool =xsk_get_pool_from_qid (adapter->netdev, queue_id);
@@ -105,19 +112,24 @@ static int igc_xdp_disable_pool(struct igc_adapter *adapter, u16 queue_id) needs_reset =netif_running (adapter->netdev) &&igc_xdp_is_enabled (adapter); rx_ring = adapter->rx_ring[queue_id]; + tx_ring = adapter->tx_ring[queue_id]; +/* Rx and Tx rings share the same napi context. */ napi = &rx_ring->q_vector->napi;if (needs_reset) {igc_disable_rx_ring (rx_ring); +igc_disable_tx_ring (tx_ring);napi_disable (napi); }xsk_pool_dma_unmap (pool,IGC_RX_DMA_ATTR );clear_bit (IGC_RING_FLAG_AF_XDP_ZC , &rx_ring->flags); +clear_bit (IGC_RING_FLAG_AF_XDP_ZC , &tx_ring->flags);if (needs_reset) {napi_enable (napi);igc_enable_rx_ring (rx_ring); +igc_enable_tx_ring (tx_ring); }return 0 ;
Signature · 1 line
2.29.2
PATCH 2/10 igc: Refactor igc_xdp_run_prog()
In reply to Andre GuedesQuoted 1 line · Andre Guedes
This patch refactors
igc_xdp_run_prog()helper, preparing the code for
s/This patch refactors/Refactor
to follow the guideline of using the imperative mood in commit msgs.
Quoted 4 lines
AF_XDPzero-copy support which is added by upcoming patches.With
AF_XDPzero-copy support,igc_poll()will have a dedicated path when the rx ring has zero-copy is enabled. To avoid code duplication as
The end of this sentence is a bit bogus.
What about:
With AF_XDP zero-copy support, igc_poll() will have a dedicated path when rx ring's memory model is MEM_TYPE_XSK_BUFF_POOL.
?
Quoted 1 line
much as possible, this patch encapsulates the code specific to handling
s/this patch encapsulates/encapsulate
git grep "This patch" Documentation/process/submitting-patches.rst
as a refresher :)
Quoted 92 lines
XDP program actions into a local helper so it can be reused by the zero-copy path.
Signed-off-by Andre Guedes <[email protected]>--- drivers/net/ethernet/intel/igc/igc_main.c | 56 +++++++++++------------ 1 file changed, 28 insertions(+), 28 deletions(-)drivers/net/ethernet/intel/igc/igc_main.c +28
-28@@ -2029,38 +2029,22 @@ static int igc_xdp_xmit_back(struct igc_adapter *adapter, struct xdp_buff *xdp)return res; } -static struct sk_buff *igc_xdp_run_prog (struct igc_adapter *adapter, -struct xdp_buff *xdp) +/* This function assumes rcu_read_lock() is held by the caller. */ +static int igc_xdp_do_run_prog (struct igc_adapter *adapter, +struct bpf_prog *prog, +struct xdp_buff *xdp) { -struct bpf_prog *prog; -int res; -u32 act; - -rcu_read_lock (); - - prog =READ_ONCE (adapter->xdp_prog); -if (!prog) { - res =IGC_XDP_PASS ; - goto unlock; - } +u32 act =bpf_prog_run_xdp (prog, xdp); - act =bpf_prog_run_xdp (prog, xdp);switch (act) {case XDP_PASS : - res =IGC_XDP_PASS ; -break ; +return IGC_XDP_PASS ;case XDP_TX : -if (igc_xdp_xmit_back (adapter, xdp) <0 ) - res =IGC_XDP_CONSUMED ; -else - res =IGC_XDP_TX ; -break ; +return igc_xdp_xmit_back (adapter, xdp) <0 ? +IGC_XDP_CONSUMED :IGC_XDP_TX ;case XDP_REDIRECT : -if (xdp_do_redirect (adapter->netdev, xdp, prog) <0 ) - res =IGC_XDP_CONSUMED ; -else - res =IGC_XDP_REDIRECT ; -break ; +return xdp_do_redirect (adapter->netdev, xdp, prog) <0 ? +IGC_XDP_CONSUMED :IGC_XDP_REDIRECT ;default :bpf_warn_invalid_xdp_action (act); fallthrough;@@ -2068,9 +2052,25 @@ static struct sk_buff *igc_xdp_run_prog(struct igc_adapter *adapter,trace_xdp_exception (adapter->netdev, prog, act); fallthrough;case XDP_DROP : - res =IGC_XDP_CONSUMED ; -break ; +return IGC_XDP_CONSUMED ; } +} + +static struct sk_buff *igc_xdp_run_prog (struct igc_adapter *adapter, +struct xdp_buff *xdp) +{ +struct bpf_prog *prog; +int res; + +rcu_read_lock (); + + prog =READ_ONCE (adapter->xdp_prog); +if (!prog) { + res =IGC_XDP_PASS ; + goto unlock; + } + + res =igc_xdp_do_run_prog (adapter, prog, xdp); unlock:rcu_read_unlock ();Signature · 1 line
2.29.2
PATCH 4/10 igc: Refactor XDP rxq info registration
In reply to Andre GuedesQuoted 34 lines · Andre Guedes
This patch does a refactoring on the XDP rxq info registration code, preparing the driver for
AF_XDPzero-copy support which is added by upcoming patches.Currently,
xdp_rxqand memory model are both registered during RX resource setup time byigc_xdp_register_rxq_info()helper. WithAF_XDP, we want to register the memory model later on while configuring the ring because we will know which memory model type to register (MEM_TYPE_PAGE_SHAREDorMEM_TYPE_XSK_BUFF_POOL).The helpers
igc_xdp_register_rxq_info()andigc_xdp_unregister_rxq_info()are not useful anymore so they are removed.Signed-off-by Andre Guedes <[email protected]>--- drivers/net/ethernet/intel/igc/igc_main.c | 14 ++++++++---- drivers/net/ethernet/intel/igc/igc_xdp.c | 27 ----------------------- drivers/net/ethernet/intel/igc/igc_xdp.h | 3 --- 3 files changed, 10 insertions(+), 34 deletions(-)drivers/net/ethernet/intel/igc/igc_main.c +1
-1@@ -419,7 +419,7 @@ void igc_free_rx_resources(struct igc_ring *rx_ring) {igc_clean_rx_ring (rx_ring); -igc_xdp_unregister_rxq_info (rx_ring); +xdp_rxq_info_unreg (&rx_ring->xdp_rxq);vfree (rx_ring->rx_buffer_info); rx_ring->rx_buffer_info =NULL ;
drivers/net/ethernet/intel/igc/igc_main.c int igc_setup_rx_resources(struct igc_ring *rx_ring) · line 466
Quoted 9 lines
drivers/net/ethernet/intel/igc/igc_main.c +4
-2@@ -460,9 +460,12 @@ int igc_setup_rx_resources(struct igc_ring *rx_ring)struct device *dev = rx_ring->dev;int size, desc_len, res; - res =igc_xdp_register_rxq_info (rx_ring); -if (res <0 ) + res =xdp_rxq_info_reg (&rx_ring->xdp_rxq, ndev, rx_ring->queue_index, +0 ); +if (res <0 ) { +netdev_err (ndev,"Failed to register xdp rxq info\n" );
Maybe print out q idx?
Quoted 15 lines
drivers/net/ethernet/intel/igc/igc_main.c +2
-1return res; + } size =sizeof (struct igc_rx_buffer ) * rx_ring->count; rx_ring->rx_buffer_info =vzalloc (size);@@ -488,7 +491,7 @@ int igc_setup_rx_resources(struct igc_ring *rx_ring)return 0 ; err: -igc_xdp_unregister_rxq_info (rx_ring); +xdp_rxq_info_unreg (&rx_ring->xdp_rxq);vfree (rx_ring->rx_buffer_info); rx_ring->rx_buffer_info =NULL ;netdev_err (ndev,"Unable to allocate memory for Rx descriptor ring\n" );
drivers/net/ethernet/intel/igc/igc_main.c static void igc_configure_rx_ring(struct igc_adapter *adapter, · line 543
Quoted 5 lines
drivers/net/ethernet/intel/igc/igc_main.c +2
-0@@ -536,6 +539,9 @@ static void igc_configure_rx_ring(struct igc_adapter *adapter,u32 srrctl =0 , rxdctl =0 ;u64 rdba = ring->dma; +WARN_ON (xdp_rxq_info_reg_mem_model (&ring->xdp_rxq, +MEM_TYPE_PAGE_SHARED ,NULL ));
You should do the unroll in case it fails just like it was done in igc_xdp_register_rxq_info.
Quoted 56 lines
drivers/net/ethernet/intel/igc/igc_main.c +1
-0+if (igc_xdp_is_enabled (adapter))set_ring_uses_large_buffer (ring);drivers/net/ethernet/intel/igc/igc_xdp.c +0
-27@@ -31,30 +31,3 @@ int igc_xdp_set_prog(struct igc_adapter *adapter, struct bpf_prog *prog,return 0 ; } - -int igc_xdp_register_rxq_info (struct igc_ring *ring) -{ -struct net_device *dev = ring->netdev; -int err; - - err =xdp_rxq_info_reg (&ring->xdp_rxq, dev, ring->queue_index,0 ); -if (err) { -netdev_err (dev,"Failed to register xdp rxq info\n" ); -return err; - } - - err =xdp_rxq_info_reg_mem_model (&ring->xdp_rxq,MEM_TYPE_PAGE_SHARED , -NULL ); -if (err) { -netdev_err (dev,"Failed to register xdp rxq mem model\n" ); -xdp_rxq_info_unreg (&ring->xdp_rxq); -return err; - } - -return 0 ; -} - -void igc_xdp_unregister_rxq_info (struct igc_ring *ring) -{ -xdp_rxq_info_unreg (&ring->xdp_rxq); -}drivers/net/ethernet/intel/igc/igc_xdp.h +0
-3@@ -7,9 +7,6 @@int igc_xdp_set_prog (struct igc_adapter *adapter,struct bpf_prog *prog,struct netlink_ext_ack *extack); -int igc_xdp_register_rxq_info (struct igc_ring *ring); -void igc_xdp_unregister_rxq_info (struct igc_ring *ring); -static inline bool igc_xdp_is_enabled (struct igc_adapter *adapter) {return !!adapter->xdp_prog;Signature · 1 line
2.29.2
PATCH 6/10 igc: Introduce igc_update_tx_stats()
In reply to Andre GuedesQuoted 2 lines · Andre Guedes
Likewise we do with the RX stats, this patch encapsulates the code that updates the driver TX stats in its own local helper.
I think you can squash this onto previous patch.
Quoted 49 lines
Signed-off-by Andre Guedes <[email protected]>--- drivers/net/ethernet/intel/igc/igc_main.c | 22 ++++++++++++++++------ 1 file changed, 16 insertions(+), 6 deletions(-)drivers/net/ethernet/intel/igc/igc_main.c +16
-6@@ -2263,6 +2263,20 @@ static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget)return total_packets; } +static void igc_update_tx_stats (struct igc_q_vector *q_vector, +unsigned int packets,unsigned int bytes) +{ +struct igc_ring *ring = q_vector->tx.ring; + +u64_stats_update_begin (&ring->tx_syncp); + ring->tx_stats.bytes += bytes; + ring->tx_stats.packets += packets; +u64_stats_update_end (&ring->tx_syncp); + + q_vector->tx.total_bytes += bytes; + q_vector->tx.total_packets += packets; +} +/** * igc_clean_tx_irq - Reclaim resources after transmit completes * @q_vector: pointer to q_vector containing needed info @@ -2365,12 +2379,8 @@ static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget) i += tx_ring->count; tx_ring->next_to_clean = i; -u64_stats_update_begin (&tx_ring->tx_syncp); - tx_ring->tx_stats.bytes += total_bytes; - tx_ring->tx_stats.packets += total_packets; -u64_stats_update_end (&tx_ring->tx_syncp); - q_vector->tx.total_bytes += total_bytes; - q_vector->tx.total_packets += total_packets; + +igc_update_tx_stats (q_vector, total_packets, total_bytes);if (test_bit (IGC_RING_FLAG_TX_DETECT_HANG , &tx_ring->flags)) {struct igc_hw *hw = &adapter->hw;Signature · 1 line
2.29.2
PATCH 8/10 igc: Replace IGC_TX_FLAGS_XDP flag by an enum
In reply to Andre GuedesQuoted 60 lines · Andre Guedes
Up to this point, tx buffers are associated with either a skb or a xdpf, and the
IGC_TX_FLAGS_XDPflag was enough to distinguish between these two case. However, with upcoming patches that will addAF_XDPzero-copy support, a third case will be introduced so this flag-based approach won't fit well.In preparation to land
AF_XDPzero-copy support, this patch replaces theIGC_TX_FLAGS_XDPflag by an enum which will be extended once zero-copy support is introduced to the driver.Signed-off-by Andre Guedes <[email protected]>--- drivers/net/ethernet/intel/igc/igc.h | 8 +++++-- drivers/net/ethernet/intel/igc/igc_main.c | 27 ++++++++++++++++++----- 2 files changed, 28 insertions(+), 7 deletions(-)drivers/net/ethernet/intel/igc/igc.h +6
-2@@ -376,8 +376,6 @@ enum igc_tx_flags {/* olinfo flags */ IGC_TX_FLAGS_IPV4 =0x10 , IGC_TX_FLAGS_CSUM = 0x20, - - IGC_TX_FLAGS_XDP = 0x100, }; enum igc_boards {@@ -394,12 +392,18 @@ enum igc_boards {#define TXD_USE_COUNT (S ) DIV_ROUND_UP((S), IGC_MAX_DATA_PER_TXD)#define DESC_NEEDED (MAX_SKB_FRAGS + 4) +enum igc_tx_buffer_type { +IGC_TX_BUFFER_TYPE_SKB , +IGC_TX_BUFFER_TYPE_XDP , +}; +/* wrapper around a pointer to a socket buffer, * so a DMA handle can be stored along with the buffer */ struct igc_tx_buffer {union igc_adv_tx_desc *next_to_watch;unsigned long time_stamp; +enum igc_tx_buffer_type type;union {struct sk_buff *skb;struct xdp_frame *xdpf;drivers/net/ethernet/intel/igc/igc_main.c +2
-1@@ -191,10 +191,18 @@ static void igc_clean_tx_ring(struct igc_ring *tx_ring)while (i != tx_ring->next_to_use) {union igc_adv_tx_desc *eop_desc, *tx_desc; -if (tx_buffer->tx_flags &IGC_TX_FLAGS_XDP ) +switch (tx_buffer->type) { +case IGC_TX_BUFFER_TYPE_XDP :xdp_return_frame (tx_buffer->xdpf);
drivers/net/ethernet/intel/igc/igc_main.c static void igc_clean_tx_ring(struct igc_ring *tx_ring) · line 205
Quoted 10 lines
drivers/net/ethernet/intel/igc/igc_main.c +8
-1-else +break ; +case IGC_TX_BUFFER_TYPE_SKB :dev_kfree_skb_any (tx_buffer->skb); +break ; +default : +netdev_warn_once (tx_ring->netdev, +"Unknown tx buffer type\n" ); +break ; + }
nit: you've been doing some effort in order to reduce the code duplication as much as it's possible, yet here you introduce duplicated code for that desc cleanup :p maybe add a little helper for that as well?
Quoted 45 lines
drivers/net/ethernet/intel/igc/igc_main.c +12
-3igc_unmap_tx_buffer (tx_ring->dev, tx_buffer);@@ -1371,6 +1379,7 @@ static netdev_tx_t igc_xmit_frame_ring(struct sk_buff *skb,/* record the location of the first descriptor for this packet */ first = &tx_ring->tx_buffer_info[tx_ring->next_to_use]; + first->type =IGC_TX_BUFFER_TYPE_SKB ; first->skb = skb; first->bytecount = skb->len; first->gso_segs =1 ;@@ -1950,8 +1959,8 @@ static int igc_xdp_init_tx_buffer(struct igc_tx_buffer *buffer,return -ENOMEM ; } + buffer->type =IGC_TX_BUFFER_TYPE_XDP ; buffer->xdpf = xdpf; - buffer->tx_flags =IGC_TX_FLAGS_XDP ; buffer->protocol =0 ; buffer->bytecount = xdpf->len; buffer->gso_segs =1 ;@@ -2315,10 +2324,18 @@ static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget) total_bytes += tx_buffer->bytecount; total_packets += tx_buffer->gso_segs; -if (tx_buffer->tx_flags &IGC_TX_FLAGS_XDP ) +switch (tx_buffer->type) { +case IGC_TX_BUFFER_TYPE_XDP :xdp_return_frame (tx_buffer->xdpf); -else +break ; +case IGC_TX_BUFFER_TYPE_SKB :napi_consume_skb (tx_buffer->skb, napi_budget); +break ; +default : +netdev_warn_once (tx_ring->netdev, +"Unknown tx buffer type\n" ); +break ; + }igc_unmap_tx_buffer (tx_ring->dev, tx_buffer);Signature · 1 line
2.29.2
PATCH 2/10 igc: Refactor igc_xdp_run_prog()
In reply to Maciej FijalkowskiQuoting Maciej Fijalkowski (2020-12-21 14:45:12)
Quoted 21 lines · Andre Guedes
Quoted 1 line
This patch refactors
igc_xdp_run_prog()helper, preparing the code fors/This patch refactors/Refactor
to follow the guideline of using the imperative mood in commit msgs.
Quoted 4 lines
AF_XDPzero-copy support which is added by upcoming patches.With
AF_XDPzero-copy support,igc_poll()will have a dedicated path when the rx ring has zero-copy is enabled. To avoid code duplication asThe end of this sentence is a bit bogus.
What about:
With
AF_XDPzero-copy support,igc_poll()will have a dedicated path when rx ring's memory model isMEM_TYPE_XSK_BUFF_POOL.?
Quoted 8 lines
Quoted 1 line
much as possible, this patch encapsulates the code specific to handling
s/this patch encapsulates/encapsulate
git grep "This patch"
Documentation/process/submitting-patches.rstas a refresher :)
Thanks for the suggestions above. I'll fix all commit messages in the v2.
PATCH 4/10 igc: Refactor XDP rxq info registration
In reply to Maciej FijalkowskiQuoting Maciej Fijalkowski (2020-12-21 14:53:45)
Quoted 4 lines
Quoted 4 lines
drivers/net/ethernet/intel/igc/igc_main.c +0
-0@@ -460,9 +460,12 @@ int igc_setup_rx_resources(struct igc_ring *rx_ring)struct device *dev = rx_ring->dev;int size, desc_len, res;
Quoted 8 lines
drivers/net/ethernet/intel/igc/igc_main.c int igc_setup_rx_resources(struct igc_ring *rx_ring) · line 466
Quoted 6 lines
drivers/net/ethernet/intel/igc/igc_main.c +4
-2- res =igc_xdp_register_rxq_info (rx_ring); -if (res <0 ) + res =xdp_rxq_info_reg (&rx_ring->xdp_rxq, ndev, rx_ring->queue_index, +0 ); +if (res <0 ) { +netdev_err (ndev,"Failed to register xdp rxq info\n" );Maybe print out q idx?
I'll add that.
Quoted 9 lines
drivers/net/ethernet/intel/igc/igc_main.c static void igc_configure_rx_ring(struct igc_adapter *adapter, · line 543
Quoted 6 lines
drivers/net/ethernet/intel/igc/igc_main.c +2
-0@@ -536,6 +539,9 @@ static void igc_configure_rx_ring(struct igc_adapter *adapter,u32 srrctl =0 , rxdctl =0 ;u64 rdba = ring->dma; +WARN_ON (xdp_rxq_info_reg_mem_model (&ring->xdp_rxq, +MEM_TYPE_PAGE_SHARED ,NULL ));You should do the unroll in case it fails just like it was done in
igc_xdp_register_rxq_info.
This was inspired in ixgbe driver.
The only reason xdp_rxq_info_reg_mem_model() could fail here is if xdp_rxq wasn't registered. However, this is very unlikely to happen since it is registered in igc_setup_rx_resources() which is always called before igc_configure_rx_ring(). The WARN_ON() macro is used just in case.
If we really want to unroll, we should propagate the error back in the call chain, changing the returning type of igc_configure_rx_ring() as well as the other functions in the call chain, so the unrolling is done in the proper place.
IMO, such change isn't worth it. It seems like a lot of change to cover a case that is never expected. WARN_ON() sound more suitable in those cases. Also, ixgbe is around for quite some time and this doesn't seem to be an issue.
PATCH 6/10 igc: Introduce igc_update_tx_stats()
In reply to Maciej FijalkowskiQuoting Maciej Fijalkowski (2020-12-21 14:58:11)
Quoted 5 lines · Andre Guedes
Quoted 2 lines
Likewise we do with the RX stats, this patch encapsulates the code that updates the driver TX stats in its own local helper.
I think you can squash this onto previous patch.
I can do that.
PATCH 8/10 igc: Replace IGC_TX_FLAGS_XDP flag by an enum
In reply to Maciej FijalkowskiQuoting Maciej Fijalkowski (2020-12-21 15:09:05)
Quoted 14 lines
Quoted 14 lines
drivers/net/ethernet/intel/igc/igc_main.c +6
-2@@ -191,10 +191,18 @@ static void igc_clean_tx_ring(struct igc_ring *tx_ring)while (i != tx_ring->next_to_use) {union igc_adv_tx_desc *eop_desc, *tx_desc; -if (tx_buffer->tx_flags &IGC_TX_FLAGS_XDP ) +switch (tx_buffer->type) { +case IGC_TX_BUFFER_TYPE_XDP :xdp_return_frame (tx_buffer->xdpf); -else +break ; +case IGC_TX_BUFFER_TYPE_SKB :dev_kfree_skb_any (tx_buffer->skb); +break ; +default :
Quoted 8 lines
drivers/net/ethernet/intel/igc/igc_main.c static void igc_clean_tx_ring(struct igc_ring *tx_ring) · line 205
Quoted 4 lines
drivers/net/ethernet/intel/igc/igc_main.c +4
-0+netdev_warn_once (tx_ring->netdev, +"Unknown tx buffer type\n" ); +break ; + }nit: you've been doing some effort in order to reduce the code duplication as much as it's possible, yet here you introduce duplicated code for that desc cleanup :p maybe add a little helper for that as well?
Note that the handling in IGC_TX_BUFFER_TYPE_SKB case is different in igc_clean_tx_irq(): dev_kfree_skb_any() vs napi_consume_skb().
PATCH 4/10 igc: Refactor XDP rxq info registration
In reply to Andre GuedesQuoted 26 lines · Andre Guedes
Quoting Maciej Fijalkowski (2020-12-21 14:53:45)
Quoted 4 lines
Quoted 4 lines
drivers/net/ethernet/intel/igc/igc_main.c +0
-0@@ -460,9 +460,12 @@ int igc_setup_rx_resources(struct igc_ring *rx_ring)struct device *dev = rx_ring->dev;int size, desc_len, res;Quoted 8 lines
drivers/net/ethernet/intel/igc/igc_main.c int igc_setup_rx_resources(struct igc_ring *rx_ring) · line 466
Quoted 6 lines
drivers/net/ethernet/intel/igc/igc_main.c +4
-2- res =igc_xdp_register_rxq_info (rx_ring); -if (res <0 ) + res =xdp_rxq_info_reg (&rx_ring->xdp_rxq, ndev, rx_ring->queue_index, +0 ); +if (res <0 ) { +netdev_err (ndev,"Failed to register xdp rxq info\n" );Maybe print out q idx?
I'll add that.
Quoted 9 lines
drivers/net/ethernet/intel/igc/igc_main.c static void igc_configure_rx_ring(struct igc_adapter *adapter, · line 543
Quoted 6 lines
drivers/net/ethernet/intel/igc/igc_main.c +2
-0@@ -536,6 +539,9 @@ static void igc_configure_rx_ring(struct igc_adapter *adapter,u32 srrctl =0 , rxdctl =0 ;u64 rdba = ring->dma; +WARN_ON (xdp_rxq_info_reg_mem_model (&ring->xdp_rxq, +MEM_TYPE_PAGE_SHARED ,NULL ));You should do the unroll in case it fails just like it was done in
igc_xdp_register_rxq_info.
Quoted 6 lines
This was inspired in ixgbe driver.
The only reason
xdp_rxq_info_reg_mem_model()could fail here is ifxdp_rxqwasn't registered. However, this is very unlikely to happen since it is registered inigc_setup_rx_resources()which is always called beforeigc_configure_rx_ring(). TheWARN_ON()macro is used just in case.
Agreed on that but let's not disregard the other failure cases that can happen by saying that it can only fail when xdp_rxq wasn't registered.
I believe that functions returning statuses have been written in such way for some reason, so I feel that ignoring error statuses is a wrong attitude.
For example, igc_setup_all_rx_resources return value is ignored in igc_request_irq, but in __igc_open it is checked. Why?
One last thing is that all other drivers besides igb/ixgbe do the error handling.
Quoted 9 lines
If we really want to unroll, we should propagate the error back in the call chain, changing the returning type of
igc_configure_rx_ring()as well as the other functions in the call chain, so the unrolling is done in the proper place.IMO, such change isn't worth it. It seems like a lot of change to cover a case that is never expected.
WARN_ON()sound more suitable in those cases. Also, ixgbe is around for quite some time and this doesn't seem to be an issue.
Well, although I don't like it, I agree :P
The uncomfortable question would be what will happen if we fail to register that mem model but due to the fact that driver is written in a way that it is not profitable to do unrolling?
PATCH 8/10 igc: Replace IGC_TX_FLAGS_XDP flag by an enum
In reply to Andre GuedesQuoted 24 lines · Andre Guedes
Quoting Maciej Fijalkowski (2020-12-21 15:09:05)
Quoted 14 lines
Quoted 14 lines
drivers/net/ethernet/intel/igc/igc_main.c +6
-2@@ -191,10 +191,18 @@ static void igc_clean_tx_ring(struct igc_ring *tx_ring)while (i != tx_ring->next_to_use) {union igc_adv_tx_desc *eop_desc, *tx_desc; -if (tx_buffer->tx_flags &IGC_TX_FLAGS_XDP ) +switch (tx_buffer->type) { +case IGC_TX_BUFFER_TYPE_XDP :xdp_return_frame (tx_buffer->xdpf); -else +break ; +case IGC_TX_BUFFER_TYPE_SKB :dev_kfree_skb_any (tx_buffer->skb); +break ; +default :Quoted 8 lines
drivers/net/ethernet/intel/igc/igc_main.c static void igc_clean_tx_ring(struct igc_ring *tx_ring) · line 205
Quoted 4 lines
drivers/net/ethernet/intel/igc/igc_main.c +4
-0+netdev_warn_once (tx_ring->netdev, +"Unknown tx buffer type\n" ); +break ; + }nit: you've been doing some effort in order to reduce the code duplication as much as it's possible, yet here you introduce duplicated code for that desc cleanup :p maybe add a little helper for that as well?
Quoted 2 lines
Note that the handling in
IGC_TX_BUFFER_TYPE_SKBcase is different inigc_clean_tx_irq():dev_kfree_skb_any()vsnapi_consume_skb().
Yikes! Sorry, it was late.
PATCH 9/10 igc: Enable RX via AF_XDP zero-copy
In reply to Andre GuedesQuoted 31 lines · Andre Guedes
This patch adds support for receiving packets via
AF_XDPzero-copy mechanism.A new flag is added to 'enum
igc_ring_flags_t' to indicate the ring hasAF_XDPzero-copy enabled so proper ring setup is carried out during ring configuration inigc_configure_rx_ring().RX buffers can now be allocated via the shared pages mechanism (default behavior of the driver) or via xsk pool (when
AF_XDPzero-copy is enabled) so a union is added to the 'structigc_rx_buffer' to cover both cases.When
AF_XDPzero-copy is enabled, rx buffers are allocated from the xsk pool using the new helperigc_alloc_rx_buffers_zc()which is the counterpart ofigc_alloc_rx_buffers().Likewise other Intel drivers that support
AF_XDPzero-copy, in igc we have a dedicated path for cleaning up rx irqs when zero-copy is enabled. This avoids adding too many checks withinigc_clean_rx_irq(), resulting in a more readable and efficient code since this function is called from the hot-path of the driver.Signed-off-by Andre Guedes <[email protected]>--- drivers/net/ethernet/intel/igc/igc.h | 22 +- drivers/net/ethernet/intel/igc/igc_base.h | 1 + drivers/net/ethernet/intel/igc/igc_main.c | 334 +++++++++++++++++++++- drivers/net/ethernet/intel/igc/igc_xdp.c | 98 +++++++ drivers/net/ethernet/intel/igc/igc_xdp.h | 2 + 5 files changed, 438 insertions(+), 19 deletions(-)
[...]
Quoted 27 lines
drivers/net/ethernet/intel/igc/igc_main.c +27
-0+ +static int igc_clean_rx_irq_zc (struct igc_q_vector *q_vector,const int budget) +{ +struct igc_adapter *adapter = q_vector->adapter; +struct igc_ring *ring = q_vector->rx.ring; +u16 cleaned_count =igc_desc_unused (ring); +int total_bytes =0 , total_packets =0 ; +struct bpf_prog *prog; +bool failure = false; +int xdp_status =0 ; + +rcu_read_lock (); + + prog =READ_ONCE (adapter->xdp_prog); + +while (likely (total_packets < budget)) { +union igc_adv_rx_desc *desc; +struct igc_rx_buffer *bi; +unsigned int size; +int res; + + desc =IGC_RX_DESC (ring, ring->next_to_clean); + size =le16_to_cpu (desc->wb.upper.length); +if (!size) +break ; + +/* This memory barrier is needed to keep us from reading
drivers/net/ethernet/intel/igc/igc_main.c static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget) · line 2443
Quoted 10 lines
drivers/net/ethernet/intel/igc/igc_main.c +10
-0+* any other fields out of the rx_desc until we know the +* descriptor has been written back +*/ +dma_rmb (); + + bi = &ring->rx_buffer_info[ring->next_to_clean]; + +if (igc_test_staterr (desc,IGC_RXDADV_STAT_TSIP )) { +/* FIXME: For now, if packet buffer contains timestamp +* information, we discard it. Once XDP infrastructe
nit: s/infrastructe/infrastructure
Quoted 123 lines
drivers/net/ethernet/intel/igc/igc_main.c +86
-3+* provides a way to report hardware timestamps, the +* code below should be fixed. +*/ + bi->xdp->data +=IGC_TS_HDR_LEN ; + size -=IGC_TS_HDR_LEN ; + } + + bi->xdp->data_end = bi->xdp->data + size; +xsk_buff_dma_sync_for_cpu (bi->xdp, ring->xsk_pool); + + res =igc_xdp_do_run_prog (adapter, prog, bi->xdp); +switch (res) { +case IGC_XDP_PASS : +igc_dispatch_skb_zc (q_vector, desc, bi->xdp); + fallthrough; +case IGC_XDP_CONSUMED : +xsk_buff_free (bi->xdp); +break ; +case IGC_XDP_TX : +case IGC_XDP_REDIRECT : + xdp_status |= res; +break ; + } + + bi->xdp =NULL ; + total_bytes += size; + total_packets++; + cleaned_count++; + ring->next_to_clean++; +if (ring->next_to_clean == ring->count) + ring->next_to_clean =0 ; + } + +rcu_read_unlock (); + +if (cleaned_count >=IGC_RX_BUFFER_WRITE ) + failure = !igc_alloc_rx_buffers_zc (ring, cleaned_count); + +if (xdp_status) +igc_finalize_xdp (adapter, xdp_status); + +igc_update_rx_stats (q_vector, total_packets, total_bytes); + +if (xsk_uses_need_wakeup (ring->xsk_pool)) { +if (failure || ring->next_to_clean == ring->next_to_use) +xsk_set_rx_need_wakeup (ring->xsk_pool); +else +xsk_clear_rx_need_wakeup (ring->xsk_pool); +return total_packets; + } + +return failure ? budget : total_packets; +} +static void igc_update_tx_stats (struct igc_q_vector *q_vector,unsigned int packets,unsigned int bytes) {@@ -2958,7 +3188,10 @@ static void igc_configure(struct igc_adapter *adapter)for (i =0 ; i < adapter->num_rx_queues; i++) {struct igc_ring *ring = adapter->rx_ring[i]; -igc_alloc_rx_buffers (ring,igc_desc_unused (ring)); +if (ring->xsk_pool) +igc_alloc_rx_buffers_zc (ring,igc_desc_unused (ring)); +else +igc_alloc_rx_buffers (ring,igc_desc_unused (ring)); } }@@ -3573,14 +3806,18 @@ static int igc_poll(struct napi_struct *napi, int budget)struct igc_q_vector *q_vector =container_of (napi, struct igc_q_vector, napi); +struct igc_ring *rx_ring = q_vector->rx.ring; +bool clean_complete = true;int work_done =0 ;if (q_vector->tx.ring) clean_complete =igc_clean_tx_irq (q_vector, budget); -if (q_vector->rx.ring) { -int cleaned =igc_clean_rx_irq (q_vector, budget); +if (rx_ring) { +int cleaned = rx_ring->xsk_pool ? +igc_clean_rx_irq_zc (q_vector, budget) : +igc_clean_rx_irq (q_vector, budget); work_done += cleaned;if (cleaned >= budget)@@ -5161,6 +5398,9 @@ static int igc_bpf(struct net_device *dev, struct netdev_bpf *bpf)switch (bpf->command) {case XDP_SETUP_PROG :return igc_xdp_set_prog (adapter, bpf->prog, bpf->extack); +case XDP_SETUP_XSK_POOL : +return igc_xdp_setup_pool (adapter, bpf->xsk.pool, + bpf->xsk.queue_id);default :return -EOPNOTSUPP ; }@@ -5206,6 +5446,43 @@ static int igc_xdp_xmit(struct net_device *dev, int num_frames,return num_frames - drops; } +static void igc_trigger_rxtxq_interrupt (struct igc_adapter *adapter, +struct igc_q_vector *q_vector) +{ +struct igc_hw *hw = &adapter->hw; +u32 eics =0 ; + + eics |= q_vector->eims_value; +wr32 (IGC_EICS , eics); +} + +int igc_xsk_wakeup (struct net_device *dev,u32 queue_id,u32 flags) +{ +struct igc_adapter *adapter =netdev_priv (dev); +struct igc_q_vector *q_vector; +struct igc_ring *ring; + +if (test_bit (__IGC_DOWN, &adapter->state)) +return -ENETDOWN ; +
drivers/net/ethernet/intel/igc/igc_main.c static int igc_xdp_xmit(struct net_device *dev, int num_frames, · line 5477
Quoted 10 lines
drivers/net/ethernet/intel/igc/igc_main.c +10
-0+if (!igc_xdp_is_enabled (adapter)) +return -ENXIO ; + +if (queue_id >= adapter->num_rx_queues) +return -EINVAL ; + + ring = adapter->rx_ring[queue_id]; + +if (!ring->xsk_pool) +return -ENXIO ;
ring local variable feels a bit unnecessary.
Quoted 85 lines
drivers/net/ethernet/intel/igc/igc_main.c +39
-0+ + q_vector = adapter->q_vector[queue_id]; +if (!napi_if_scheduled_mark_missed (&q_vector->napi)) +igc_trigger_rxtxq_interrupt (adapter, q_vector); + +return 0 ; +} +static const struct net_device_ops igc_netdev_ops = { .ndo_open = igc_open, .ndo_stop = igc_close,@@ -5221,6 +5498,7 @@ static const struct net_device_ops igc_netdev_ops = { .ndo_setup_tc = igc_setup_tc, .ndo_bpf = igc_bpf, .ndo_xdp_xmit = igc_xdp_xmit, + .ndo_xsk_wakeup = igc_xsk_wakeup, };/* PCIe configuration access */ @@ -5974,6 +6252,36 @@ struct net_device *igc_get_hw_dev(struct igc_hw *hw)return adapter->netdev; } +static void igc_disable_rx_ring_hw (struct igc_ring *ring) +{ +struct igc_hw *hw = &ring->q_vector->adapter->hw; +u8 idx = ring->reg_idx; +u32 rxdctl; + + rxdctl =rd32 (IGC_RXDCTL (idx)); + rxdctl &= ~IGC_RXDCTL_QUEUE_ENABLE ; + rxdctl |=IGC_RXDCTL_SWFLUSH ; +wr32 (IGC_RXDCTL (idx), rxdctl); +} + +void igc_disable_rx_ring (struct igc_ring *ring) +{ +igc_disable_rx_ring_hw (ring); +igc_clean_rx_ring (ring); +} + +void igc_enable_rx_ring (struct igc_ring *ring) +{ +struct igc_adapter *adapter = ring->q_vector->adapter; + +igc_configure_rx_ring (adapter, ring); + +if (ring->xsk_pool) +igc_alloc_rx_buffers_zc (ring,igc_desc_unused (ring)); +else +igc_alloc_rx_buffers (ring,igc_desc_unused (ring)); +} +/** * igc_init_module - Driver Registration Routine * drivers/net/ethernet/intel/igc/igc_xdp.c +14
-0@@ -1,6 +1,8 @@// SPDX-License-Identifier: GPL-2.0 /* Copyright (c) 2020, Intel Corporation. */ +#include <net/xdp_sock_drv.h> +#include "igc.h" #include "igc_xdp.h" @@ -31,3 +33,99 @@ int igc_xdp_set_prog(struct igc_adapter *adapter, struct bpf_prog *prog,return 0 ; } + +static int igc_xdp_enable_pool (struct igc_adapter *adapter, +struct xsk_buff_pool *pool,u16 queue_id) +{ +struct net_device *ndev = adapter->netdev; +struct device *dev = &adapter->pdev->dev; +struct igc_ring *rx_ring; +struct napi_struct *napi; +bool needs_reset; +u32 frame_size; +int err; +
drivers/net/ethernet/intel/igc/igc_xdp.c int igc_xdp_set_prog(struct igc_adapter *adapter, struct bpf_prog *prog, · line 57
Quoted 10 lines
drivers/net/ethernet/intel/igc/igc_xdp.c +10
-0+if (queue_id >= adapter->num_rx_queues) +return -EINVAL ; + + frame_size =xsk_pool_get_rx_frame_size (pool); +if (frame_size <ETH_FRAME_LEN +VLAN_HLEN *2 ) { +/* For now, when XDP is enabled, the driver doesn't support +* frames that span over multiple buffers. The max frame size +* considered here is the ethernet frame size + vlan double +* tagging. +*/
I don't really follow that. You check if chunk size is less than ETH_FRAME_LEN + VLAN_HLEN * 2. chunk size will be at least 2k if I recall correctly. How is that related to fragmented buffers?
Quoted 32 lines
drivers/net/ethernet/intel/igc/igc_xdp.c +32
-0+return -EOPNOTSUPP ; + } + + err =xsk_pool_dma_map (pool, dev,IGC_RX_DMA_ATTR ); +if (err) { +netdev_err (ndev,"Failed to map xsk pool\n" ); +return err; + } + + needs_reset =netif_running (adapter->netdev) &&igc_xdp_is_enabled (adapter); + + rx_ring = adapter->rx_ring[queue_id]; + napi = &rx_ring->q_vector->napi; + +if (needs_reset) { +igc_disable_rx_ring (rx_ring); +napi_disable (napi); + } + +set_bit (IGC_RING_FLAG_AF_XDP_ZC , &rx_ring->flags); + +if (needs_reset) { +napi_enable (napi); +igc_enable_rx_ring (rx_ring); + + err =igc_xsk_wakeup (ndev, queue_id,XDP_WAKEUP_RX ); +if (err) +return err; + } + +return 0 ; +}
PATCH 10/10 igc: Enable TX via AF_XDP zero-copy
In reply to Andre GuedesQuoted 27 lines · Andre Guedes
This patch adds support for transmitting packets via
AF_XDPzero-copy mechanism.The packet transmission itself is implemented by
igc_xdp_xmit_zc()which is called fromigc_clean_tx_irq()when the ring hasAF_XDPzero-copy enabled. Likewise i40e and ice drivers, the transmission budget used is the number of descriptors available on the ring.A new tx buffer type is introduced to 'enum
igc_tx_buffer_type' to indicate the tx buffer uses memory from xsk pool so it can be properly cleaned after transmission or when the ring is cleaned.The I225 controller has only 4 Tx hardware queues so the main difference between igc and other Intel drivers that support
AF_XDPzero-copy is that there is no tx ring dedicated exclusively to XDP. Instead, tx rings are shared between the network stack and XDP, and netdev queue lock is used to ensure mutual exclusion. This is the same approach implemented to supportXDP_TXandXDP_REDIRECTactions.Signed-off-by Andre Guedes <[email protected]>--- drivers/net/ethernet/intel/igc/igc.h | 3 + drivers/net/ethernet/intel/igc/igc_base.h | 1 + drivers/net/ethernet/intel/igc/igc_main.c | 115 +++++++++++++++++++++- drivers/net/ethernet/intel/igc/igc_xdp.c | 20 +++- 4 files changed, 131 insertions(+), 8 deletions(-)
[...]
Quoted 42 lines
drivers/net/ethernet/intel/igc/igc_main.c +41
-0+static void igc_xdp_xmit_zc (struct igc_ring *ring) +{ +struct xsk_buff_pool *pool = ring->xsk_pool; +struct netdev_queue *nq =txring_txq (ring); +int cpu =smp_processor_id (); +struct xdp_desc xdp_desc; +bool work_done; +u16 budget; + +if (!netif_carrier_ok (ring->netdev)) +return ; + +__netif_tx_lock (nq, cpu); + + budget =igc_desc_unused (ring); + work_done = false; + +while (xsk_tx_peek_desc (pool, &xdp_desc) && budget--) { +u32 cmd_type, olinfo_status; +union igc_adv_tx_desc *desc; +struct igc_tx_buffer *bi; +dma_addr_t dma; + + cmd_type =IGC_ADVTXD_DTYP_DATA |IGC_ADVTXD_DCMD_DEXT | +IGC_ADVTXD_DCMD_IFCS |IGC_TXD_DCMD | + xdp_desc.len; + olinfo_status = xdp_desc.len <<IGC_ADVTXD_PAYLEN_SHIFT ; + + dma =xsk_buff_raw_get_dma (pool, xdp_desc.addr); +xsk_buff_raw_dma_sync_for_device (pool, dma, xdp_desc.len); + + desc =IGC_TX_DESC (ring, ring->next_to_use); + desc->read.cmd_type_len =cpu_to_le32 (cmd_type); + desc->read.olinfo_status =cpu_to_le32 (olinfo_status); + desc->read.buffer_addr =cpu_to_le64 (dma); + + bi = &ring->tx_buffer_info[ring->next_to_use]; + bi->type =IGC_TX_BUFFER_TYPE_XSK ; + bi->protocol =0 ; + bi->bytecount = xdp_desc.len; + bi->gso_segs =1 ;
drivers/net/ethernet/intel/igc/igc_main.c static void igc_update_tx_stats(struct igc_q_vector *q_vector, · line 2571
Quoted 10 lines
drivers/net/ethernet/intel/igc/igc_main.c +10
-0+ bi->time_stamp = jiffies; + bi->next_to_watch = desc; + +netdev_tx_sent_queue (txring_txq (ring), xdp_desc.len); + + ring->next_to_use++; +if (ring->next_to_use == ring->count) + ring->next_to_use =0 ; + + work_done = true;
nit: setting it on each iteration feels semi-optimal.
Quoted 9 lines
drivers/net/ethernet/intel/igc/igc_main.c +9
-0+ } + +if (work_done) { +igc_flush_tx_descriptors (ring); +xsk_tx_release (pool); + } + +__netif_tx_unlock (nq); +}
cover igc: Add support for AF_XDP zero-copy
In reply to Andre GuedesQuoted 8 lines · Andre Guedes
Hi,
This series adds
AF_XDPzero-copy feature to igc driver.The initial patches do some code refactoring, preparing the code base to land the
AF_XDPzero-copy feature, avoiding code duplications. The last patches of the series are the ones implementing the feature.
Quoted 5 lines
The last patch which indeed implements
AF_XDPzero-copy support was originally way too lengthy so, for the sake of code review, I broke it up into two patches: one adding support for the RX functionality and the other one adding TX support. Feel free to squash those two patches when pushing the series to net-next if you find appropriate.
Generally this looks good! I had only minor comments.
I'm interested how are you going to utilize the AF_XDP ZC, probably something related to TSN ;)
Quoted 26 lines
Best,
AndreAndre Guedes (10):
igc: Moveigc_xdp_is_enabled()
igc: Refactorigc_xdp_run_prog()
igc: Refactorigc_clean_rx_ring()
igc: Refactor XDP rxq info registration
igc: Introduceigc_update_rx_stats()
igc: Introduceigc_update_tx_stats()
igc: Introduceigc_unmap_tx_buffer()helper
igc: ReplaceIGC_TX_FLAGS_XDPflag by an enum
igc: Enable RX viaAF_XDPzero-copy
igc: Enable TX viaAF_XDPzero-copydrivers/net/ethernet/intel/igc/igc.h | 33 +- drivers/net/ethernet/intel/igc/igc_base.h | 2 + drivers/net/ethernet/intel/igc/igc_main.c | 650 ++++++++++++++++++---- drivers/net/ethernet/intel/igc/igc_xdp.c | 107 +++- drivers/net/ethernet/intel/igc/igc_xdp.h | 8 +- 5 files changed, 672 insertions(+), 128 deletions(-)Signature · 1 line
2.29.2
PATCH 4/10 igc: Refactor XDP rxq info registration
In reply to Maciej FijalkowskiQuoting Maciej Fijalkowski (2020-12-22 04:32:05)
Quoted 24 lines
Quoted 10 lines
Quoted 9 lines
drivers/net/ethernet/intel/igc/igc_main.c static void igc_configure_rx_ring(struct igc_adapter *adapter, · line 543
Quoted 6 lines
drivers/net/ethernet/intel/igc/igc_main.c +2
-0@@ -536,6 +539,9 @@ static void igc_configure_rx_ring(struct igc_adapter *adapter,u32 srrctl =0 , rxdctl =0 ;u64 rdba = ring->dma; +WARN_ON (xdp_rxq_info_reg_mem_model (&ring->xdp_rxq, +MEM_TYPE_PAGE_SHARED ,NULL ));You should do the unroll in case it fails just like it was done in
igc_xdp_register_rxq_info.Quoted 6 lines
This was inspired in ixgbe driver.
The only reason
xdp_rxq_info_reg_mem_model()could fail here is ifxdp_rxqwasn't registered. However, this is very unlikely to happen since it is registered inigc_setup_rx_resources()which is always called beforeigc_configure_rx_ring(). TheWARN_ON()macro is used just in case.Agreed on that but let's not disregard the other failure cases that can happen by saying that it can only fail when
xdp_rxqwasn't registered.I believe that functions returning statuses have been written in such way for some reason, so I feel that ignoring error statuses is a wrong attitude.
Quoted 5 lines
For example,
igc_setup_all_rx_resourcesreturn value is ignored inigc_request_irq, but in__igc_openit is checked. Why?One last thing is that all other drivers besides igb/ixgbe do the error handling.
Agreed. I have another series with some fixes to the driver that I'm working on. I can fix the issue with igc_request_irq() on that series.
Quoted 9 lines
Quoted 8 lines
If we really want to unroll, we should propagate the error back in the call chain, changing the returning type of
igc_configure_rx_ring()as well as the other functions in the call chain, so the unrolling is done in the proper place.IMO, such change isn't worth it. It seems like a lot of change to cover a case that is never expected.
WARN_ON()sound more suitable in those cases. Also, ixgbe is around for quite some time and this doesn't seem to be an issue.
Quoted 5 lines
Well, although I don't like it, I agree :P
The uncomfortable question would be what will happen if we fail to register that mem model but due to the fact that driver is written in a way that it is not profitable to do unrolling?
I see your point. In that case, I think we can tackle that in a separate series changing all three drivers (igc, ixgbe, and igb). Sounds good?
PATCH 9/10 igc: Enable RX via AF_XDP zero-copy
In reply to Maciej FijalkowskiQuoting Maciej Fijalkowski (2020-12-22 05:14:28)
Quoted 30 lines
Quoted 30 lines
drivers/net/ethernet/intel/igc/igc_main.c +30
-0+static int igc_clean_rx_irq_zc (struct igc_q_vector *q_vector,const int budget) +{ +struct igc_adapter *adapter = q_vector->adapter; +struct igc_ring *ring = q_vector->rx.ring; +u16 cleaned_count =igc_desc_unused (ring); +int total_bytes =0 , total_packets =0 ; +struct bpf_prog *prog; +bool failure = false; +int xdp_status =0 ; + +rcu_read_lock (); + + prog =READ_ONCE (adapter->xdp_prog); + +while (likely (total_packets < budget)) { +union igc_adv_rx_desc *desc; +struct igc_rx_buffer *bi; +unsigned int size; +int res; + + desc =IGC_RX_DESC (ring, ring->next_to_clean); + size =le16_to_cpu (desc->wb.upper.length); +if (!size) +break ; + +/* This memory barrier is needed to keep us from reading +* any other fields out of the rx_desc until we know the +* descriptor has been written back +*/ +dma_rmb ();
Quoted 8 lines
drivers/net/ethernet/intel/igc/igc_main.c static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget) · line 2443
Quoted 6 lines
drivers/net/ethernet/intel/igc/igc_main.c +6
-0+ + bi = &ring->rx_buffer_info[ring->next_to_clean]; + +if (igc_test_staterr (desc,IGC_RXDADV_STAT_TSIP )) { +/* FIXME: For now, if packet buffer contains timestamp +* information, we discard it. Once XDP infrastructe nit: s/infrastructe/infrastructure
Thanks, I'll fix it in the v2.
Quoted 13 lines
Quoted 13 lines
drivers/net/ethernet/intel/igc/igc_main.c +13
-0+int igc_xsk_wakeup (struct net_device *dev,u32 queue_id,u32 flags) +{ +struct igc_adapter *adapter =netdev_priv (dev); +struct igc_q_vector *q_vector; +struct igc_ring *ring; + +if (test_bit (__IGC_DOWN, &adapter->state)) +return -ENETDOWN ; + +if (!igc_xdp_is_enabled (adapter)) +return -ENXIO ; + +if (queue_id >= adapter->num_rx_queues)
Quoted 8 lines
drivers/net/ethernet/intel/igc/igc_main.c static int igc_xdp_xmit(struct net_device *dev, int num_frames, · line 5477
Quoted 6 lines
drivers/net/ethernet/intel/igc/igc_main.c +6
-0+return -EINVAL ; + + ring = adapter->rx_ring[queue_id]; + +if (!ring->xsk_pool) +return -ENXIO ;ring local variable feels a bit unnecessary.
It's just for the sake of code readability. I thought it reads better than:
if (!adapter->rx_ring[queue_id]->xsk_pool) return -ENXIO;
Quoted 17 lines
Quoted 17 lines
drivers/net/ethernet/intel/igc/igc_xdp.c +17
-0+static int igc_xdp_enable_pool (struct igc_adapter *adapter, +struct xsk_buff_pool *pool,u16 queue_id) +{ +struct net_device *ndev = adapter->netdev; +struct device *dev = &adapter->pdev->dev; +struct igc_ring *rx_ring; +struct napi_struct *napi; +bool needs_reset; +u32 frame_size; +int err; + +if (queue_id >= adapter->num_rx_queues) +return -EINVAL ; + + frame_size =xsk_pool_get_rx_frame_size (pool); +if (frame_size <ETH_FRAME_LEN +VLAN_HLEN *2 ) { +/* For now, when XDP is enabled, the driver doesn't support
Quoted 8 lines
drivers/net/ethernet/intel/igc/igc_xdp.c int igc_xdp_set_prog(struct igc_adapter *adapter, struct bpf_prog *prog, · line 57
Quoted 4 lines
drivers/net/ethernet/intel/igc/igc_xdp.c +4
-0+* frames that span over multiple buffers. The max frame size +* considered here is the ethernet frame size + vlan double +* tagging. +*/ I don't really follow that. You check if chunk size is less than
ETH_FRAME_LEN+VLAN_HLEN* 2. chunk size will be at least 2k if I recall correctly. How is that related to fragmented buffers?
Note that the code above checks the xsk frame size, not chunk size.
Yes, the chunk mim size is 2k indeed. However, if I'm reading the code correctly, the umem headroom can have an arbitrary value defined by the user (see xdp_umem_reg()). In this case, the xsk frame size could potentially be less than the ethernet frame size.
cover igc: Add support for AF_XDP zero-copy
In reply to Maciej FijalkowskiQuoting Maciej Fijalkowski (2020-12-22 06:20:45)
Quoted 9 lines · Andre Guedes
Quoted 8 lines
Hi,
This series adds
AF_XDPzero-copy feature to igc driver.The initial patches do some code refactoring, preparing the code base to land the
AF_XDPzero-copy feature, avoiding code duplications. The last patches of the series are the ones implementing the feature.
Quoted 7 lines
Quoted 5 lines
The last patch which indeed implements
AF_XDPzero-copy support was originally way too lengthy so, for the sake of code review, I broke it up into two patches: one adding support for the RX functionality and the other one adding TX support. Feel free to squash those two patches when pushing the series to net-next if you find appropriate.Generally this looks good! I had only minor comments.
Thanks for your review, Maciej. I'm working on the v2 and should submit it soon.
Quoted 2 lines
I'm interested how are you going to utilize the
AF_XDPZC, probably something related to TSN ;)
Yes, that's right. In TSN use cases, tx and rx latency within the system is relevant. AF_XDP ZC infra provides us a lower latency alternative to AF_PACKET.
PATCH 9/10 igc: Enable RX via AF_XDP zero-copy
In reply to Andre GuedesQuoted 64 lines · Andre Guedes
Quoting Maciej Fijalkowski (2020-12-22 05:14:28)
Quoted 30 lines
Quoted 30 lines
drivers/net/ethernet/intel/igc/igc_main.c +30
-0+static int igc_clean_rx_irq_zc (struct igc_q_vector *q_vector,const int budget) +{ +struct igc_adapter *adapter = q_vector->adapter; +struct igc_ring *ring = q_vector->rx.ring; +u16 cleaned_count =igc_desc_unused (ring); +int total_bytes =0 , total_packets =0 ; +struct bpf_prog *prog; +bool failure = false; +int xdp_status =0 ; + +rcu_read_lock (); + + prog =READ_ONCE (adapter->xdp_prog); + +while (likely (total_packets < budget)) { +union igc_adv_rx_desc *desc; +struct igc_rx_buffer *bi; +unsigned int size; +int res; + + desc =IGC_RX_DESC (ring, ring->next_to_clean); + size =le16_to_cpu (desc->wb.upper.length); +if (!size) +break ; + +/* This memory barrier is needed to keep us from reading +* any other fields out of the rx_desc until we know the +* descriptor has been written back +*/ +dma_rmb ();Quoted 8 lines
drivers/net/ethernet/intel/igc/igc_main.c static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget) · line 2443
Quoted 6 lines
drivers/net/ethernet/intel/igc/igc_main.c +6
-0+ + bi = &ring->rx_buffer_info[ring->next_to_clean]; + +if (igc_test_staterr (desc,IGC_RXDADV_STAT_TSIP )) { +/* FIXME: For now, if packet buffer contains timestamp +* information, we discard it. Once XDP infrastructe nit: s/infrastructe/infrastructure
Thanks, I'll fix it in the v2.
Quoted 13 lines
Quoted 13 lines
drivers/net/ethernet/intel/igc/igc_main.c +13
-0+int igc_xsk_wakeup (struct net_device *dev,u32 queue_id,u32 flags) +{ +struct igc_adapter *adapter =netdev_priv (dev); +struct igc_q_vector *q_vector; +struct igc_ring *ring; + +if (test_bit (__IGC_DOWN, &adapter->state)) +return -ENETDOWN ; + +if (!igc_xdp_is_enabled (adapter)) +return -ENXIO ; + +if (queue_id >= adapter->num_rx_queues)Quoted 8 lines
drivers/net/ethernet/intel/igc/igc_main.c static int igc_xdp_xmit(struct net_device *dev, int num_frames, · line 5477
Quoted 6 lines
drivers/net/ethernet/intel/igc/igc_main.c +6
-0+return -EINVAL ; + + ring = adapter->rx_ring[queue_id]; + +if (!ring->xsk_pool) +return -ENXIO ;ring local variable feels a bit unnecessary.
Quoted 4 lines
It's just for the sake of code readability. I thought it reads better than:
if (!adapter->rx_ring[queue_id]->xsk_pool) return -ENXIO;
Yeah, okay then.
Quoted 27 lines
Quoted 18 lines
Quoted 18 lines
drivers/net/ethernet/intel/igc/igc_xdp.c +17
-0+static int igc_xdp_enable_pool (struct igc_adapter *adapter, +struct xsk_buff_pool *pool,u16 queue_id) +{ +struct net_device *ndev = adapter->netdev; +struct device *dev = &adapter->pdev->dev; +struct igc_ring *rx_ring; +struct napi_struct *napi; +bool needs_reset; +u32 frame_size; +int err; + +if (queue_id >= adapter->num_rx_queues) +return -EINVAL ; + + frame_size =xsk_pool_get_rx_frame_size (pool); +if (frame_size <ETH_FRAME_LEN +VLAN_HLEN *2 ) { +/* For now, when XDP is enabled, the driver doesn't support Quoted 8 lines
drivers/net/ethernet/intel/igc/igc_xdp.c int igc_xdp_set_prog(struct igc_adapter *adapter, struct bpf_prog *prog, · line 57
Quoted 4 lines
drivers/net/ethernet/intel/igc/igc_xdp.c +4
-0+* frames that span over multiple buffers. The max frame size +* considered here is the ethernet frame size + vlan double +* tagging. +*/ I don't really follow that. You check if chunk size is less than
ETH_FRAME_LEN+VLAN_HLEN* 2. chunk size will be at least 2k if I recall correctly. How is that related to fragmented buffers?
Quoted 6 lines
Note that the code above checks the xsk frame size, not chunk size.
Yes, the chunk mim size is 2k indeed. However, if I'm reading the code correctly, the umem headroom can have an arbitrary value defined by the user (see
xdp_umem_reg()). In this case, the xsk frame size could potentially be less than the ethernet frame size.
Ok I see. So you actually meant that for lower frame size given to HW we would start to fragment the frames?
Comment was a bit unclear, hence my confusion.
Anyway, I feel like this is something generic that should be checked earlier at xsk code? Meaning, when we actually create the umem with user provided config at setsockopt?
PATCH 4/10 igc: Refactor XDP rxq info registration
In reply to Andre GuedesQuoted 43 lines · Andre Guedes
Quoting Maciej Fijalkowski (2020-12-22 04:32:05)
Quoted 24 lines
Quoted 10 lines
Quoted 9 lines
drivers/net/ethernet/intel/igc/igc_main.c static void igc_configure_rx_ring(struct igc_adapter *adapter, · line 543
Quoted 6 lines
drivers/net/ethernet/intel/igc/igc_main.c +2
-0@@ -536,6 +539,9 @@ static void igc_configure_rx_ring(struct igc_adapter *adapter,u32 srrctl =0 , rxdctl =0 ;u64 rdba = ring->dma; +WARN_ON (xdp_rxq_info_reg_mem_model (&ring->xdp_rxq, +MEM_TYPE_PAGE_SHARED ,NULL ));You should do the unroll in case it fails just like it was done in
igc_xdp_register_rxq_info.Quoted 6 lines
This was inspired in ixgbe driver.
The only reason
xdp_rxq_info_reg_mem_model()could fail here is ifxdp_rxqwasn't registered. However, this is very unlikely to happen since it is registered inigc_setup_rx_resources()which is always called beforeigc_configure_rx_ring(). TheWARN_ON()macro is used just in case.Agreed on that but let's not disregard the other failure cases that can happen by saying that it can only fail when
xdp_rxqwasn't registered.I believe that functions returning statuses have been written in such way for some reason, so I feel that ignoring error statuses is a wrong attitude.
Quoted 5 lines
For example,
igc_setup_all_rx_resourcesreturn value is ignored inigc_request_irq, but in__igc_openit is checked. Why?One last thing is that all other drivers besides igb/ixgbe do the error handling.
Agreed. I have another series with some fixes to the driver that I'm working on. I can fix the issue with
igc_request_irq()on that series.Quoted 9 lines
Quoted 8 lines
If we really want to unroll, we should propagate the error back in the call chain, changing the returning type of
igc_configure_rx_ring()as well as the other functions in the call chain, so the unrolling is done in the proper place.IMO, such change isn't worth it. It seems like a lot of change to cover a case that is never expected.
WARN_ON()sound more suitable in those cases. Also, ixgbe is around for quite some time and this doesn't seem to be an issue.
Quoted 8 lines
Quoted 5 lines
Well, although I don't like it, I agree :P
The uncomfortable question would be what will happen if we fail to register that mem model but due to the fact that driver is written in a way that it is not profitable to do unrolling?
I see your point. In that case, I think we can tackle that in a separate series changing all three drivers (igc, ixgbe, and igb). Sounds good?
Sure :)
PATCH 9/10 igc: Enable RX via AF_XDP zero-copy
In reply to Maciej FijalkowskiQuoting Maciej Fijalkowski (2020-12-22 12:59:25)
Quoted 28 lines
Quoted 28 lines
Quoted 17 lines
Quoted 17 lines
drivers/net/ethernet/intel/igc/igc_xdp.c +17
-0+static int igc_xdp_enable_pool (struct igc_adapter *adapter, +struct xsk_buff_pool *pool,u16 queue_id) +{ +struct net_device *ndev = adapter->netdev; +struct device *dev = &adapter->pdev->dev; +struct igc_ring *rx_ring; +struct napi_struct *napi; +bool needs_reset; +u32 frame_size; +int err; + +if (queue_id >= adapter->num_rx_queues) +return -EINVAL ; + + frame_size =xsk_pool_get_rx_frame_size (pool); +if (frame_size <ETH_FRAME_LEN +VLAN_HLEN *2 ) { +/* For now, when XDP is enabled, the driver doesn't support Quoted 8 lines
drivers/net/ethernet/intel/igc/igc_xdp.c int igc_xdp_set_prog(struct igc_adapter *adapter, struct bpf_prog *prog, · line 57
Quoted 4 lines
drivers/net/ethernet/intel/igc/igc_xdp.c +4
-0+* frames that span over multiple buffers. The max frame size +* considered here is the ethernet frame size + vlan double +* tagging. +*/ I don't really follow that. You check if chunk size is less than
ETH_FRAME_LEN+VLAN_HLEN* 2. chunk size will be at least 2k if I recall correctly. How is that related to fragmented buffers?Note that the code above checks the xsk frame size, not chunk size.
Quoted 7 lines
Quoted 4 lines
Yes, the chunk mim size is 2k indeed. However, if I'm reading the code correctly, the umem headroom can have an arbitrary value defined by the user (see
xdp_umem_reg()). In this case, the xsk frame size could potentially be less than the ethernet frame size.Ok I see. So you actually meant that for lower frame size given to HW we would start to fragment the frames?
Yes.
Quoted 1 line
Comment was a bit unclear, hence my confusion.
I'll try to make that comment more clear in the v2.
Quoted 3 lines
Anyway, I feel like this is something generic that should be checked earlier at xsk code? Meaning, when we actually create the umem with user provided config at setsockopt?
Yeah, that might be a good idea. For now, I'll keep the check in the driver since it doesn't work properly with smaller sizes. In the future, if we have that check in xsk layer, we can remove it from the driver.