This change actually defines the (initial) metadata layout that should be used by AF_XDP userspace (xsk_tx_metadata). The first field is flags which requests appropriate offloads, followed by the offload-specific fields. The supported per-device offloads are exported via netlink (new xsk-flags). The offloads themselves are still implemented in a bit of a framework-y fashion that's left from my initial kfunc attempt. I'm introducing new xsk_tx_metadata_ops which drivers are supposed to implement. The drivers are also supposed to call xsk_tx_metadata_request/xsk_tx_metadata_complete in the right places. Since xsk_tx_metadata_{request,_complete} are static inline, we don't incur any extra overhead doing indirect calls. The benefit of this scheme is as follows: - keeps all metadata layout parsing away from driver code - makes it easy to grep and see which drivers implement what - don't need any extra flags to maintain to keep track of what offloads are implemented; if the callback is implemented - the offload is supported (used by netlink reporting code) Two offloads are defined right now: 1. XDP_TXMD_FLAGS_CHECKSUM: skb-style csum_start+csum_offset 2. XDP_TXMD_FLAGS_TIMESTAMP: writes TX timestamp back into metadata area upon completion (tx_timestamp field) XDP_TXMD_FLAGS_TIMESTAMP is also implemented for XDP_COPY mode: it writes SW timestamp from the skb destructor (note I'm reusing hwtstamps to pass metadata pointer). The struct is forward-compatible and can be extended in the future by appending more fields.
Showing the changes to include (7 of 15 files). Show all
Unified · Split
| @@ -1865,6 +1865,7 @@ enum netdev_stat_type { · 1864 unmodified lines | |||
| 1865 | * @netdev_ops: Includes several pointers to callbacks, | 1865 | * @netdev_ops: Includes several pointers to callbacks, |
| 1866 | * if one wants to override the ndo_*() functions | 1866 | * if one wants to override the ndo_*() functions |
| 1867 | * @xdp_metadata_ops: Includes pointers to XDP metadata callbacks. | 1867 | * @xdp_metadata_ops: Includes pointers to XDP metadata callbacks. |
| 1868 | * @xsk_tx_metadata_ops: Includes pointers to AF_XDP TX metadata callbacks. | ||
| 1868 | * @ethtool_ops: Management operations | 1869 | * @ethtool_ops: Management operations |
| 1869 | * @l3mdev_ops: Layer 3 master device operations | 1870 | * @l3mdev_ops: Layer 3 master device operations |
| 1870 | * @ndisc_ops: Includes callbacks for different IPv6 neighbour | 1871 | * @ndisc_ops: Includes callbacks for different IPv6 neighbour |
| @@ -2128,6 +2129,7 @@ struct net_device { · 257 unmodified lines | |||
| 2128 | unsigned long long priv_flags; | 2129 | unsigned long long priv_flags; |
| 2129 | const struct net_device_ops *netdev_ops; | 2130 | const struct net_device_ops *netdev_ops; |
| 2130 | const struct xdp_metadata_ops *xdp_metadata_ops; | 2131 | const struct xdp_metadata_ops *xdp_metadata_ops; |
| 2132 | const struct xsk_tx_metadata_ops *xsk_tx_metadata_ops; | ||
| 2131 | int ifindex; | 2133 | int ifindex; |
| 2132 | unsigned short gflags; | 2134 | unsigned short gflags; |
| 2133 | unsigned short hard_header_len; | 2135 | unsigned short hard_header_len; |
| @@ -566,6 +566,15 @@ struct ubuf_info_msgzc { · 565 unmodified lines | |||
| 566 | int mm_account_pinned_pages(struct mmpin *mmp, size_t size); | 566 | int mm_account_pinned_pages(struct mmpin *mmp, size_t size); |
| 567 | void mm_unaccount_pinned_pages(struct mmpin *mmp); | 567 | void mm_unaccount_pinned_pages(struct mmpin *mmp); |
| 568 | 568 | ||
| 569 | /* Preserve some data across TX submission and completion. | ||
| 570 | * | ||
| 571 | * Note, this state is stored in the driver. Extending the layout | ||
| 572 | * might need some special care. | ||
| 573 | */ | ||
| 574 | struct xsk_tx_metadata_compl { | ||
| 575 | __u64 *tx_timestamp; | ||
| 576 | }; | ||
| 577 | |||
| 569 | /* This data is invariant across clones and lives at | 578 | /* This data is invariant across clones and lives at |
| 570 | * the end of the header data, ie. at skb->end. | 579 | * the end of the header data, ie. at skb->end. |
| 571 | */ | 580 | */ |
| @@ -578,7 +587,10 @@ struct skb_shared_info { · 6 unmodified lines | |||
| 578 | /* Warning: this field is not always filled in (UFO)! */ | 587 | /* Warning: this field is not always filled in (UFO)! */ |
| 579 | unsigned short gso_segs; | 588 | unsigned short gso_segs; |
| 580 | struct sk_buff *frag_list; | 589 | struct sk_buff *frag_list; |
| 581 | struct skb_shared_hwtstamps hwtstamps; | 590 | union { |
| 591 | struct skb_shared_hwtstamps hwtstamps; | ||
| 592 | struct xsk_tx_metadata_compl xsk_meta; | ||
| 593 | }; | ||
| 582 | unsigned int gso_type; | 594 | unsigned int gso_type; |
| 583 | u32 tskey; | 595 | u32 tskey; |
| 584 | 596 | ||
| @@ -93,12 +93,105 @@ struct xdp_sock { · 92 unmodified lines | |||
| 93 | struct xsk_queue *cq_tmp; /* Only as tmp storage before bind */ | 93 | struct xsk_queue *cq_tmp; /* Only as tmp storage before bind */ |
| 94 | }; | 94 | }; |
| 95 | 95 | ||
| 96 | /* | ||
| 97 | * AF_XDP TX metadata hooks for network devices. | ||
| 98 | * The following hooks can be defined; unless noted otherwise, they are | ||
| 99 | * optional and can be filled with a null pointer. | ||
| 100 | * | ||
| 101 | * void (*tmo_request_timestamp)(void *priv) | ||
| 102 | * Called when AF_XDP frame requested egress timestamp. | ||
| 103 | * | ||
| 104 | * u64 (*tmo_fill_timestamp)(void *priv) | ||
| 105 | * Called when AF_XDP frame, that had requested egress timestamp, | ||
| 106 | * received a completion. The hook needs to return the actual HW timestamp. | ||
| 107 | * | ||
| 108 | * void (*tmo_request_checksum)(u16 csum_start, u16 csum_offset, void *priv) | ||
| 109 | * Called when AF_XDP frame requested HW checksum offload. csum_start | ||
| 110 | * indicates position where checksumming should start. | ||
| 111 | * csum_offset indicates position where checksum should be stored. | ||
| 112 | * | ||
| 113 | */ | ||
| 114 | struct xsk_tx_metadata_ops { | ||
| 115 | void (*tmo_request_timestamp)(void *priv); | ||
| 116 | u64 (*tmo_fill_timestamp)(void *priv); | ||
| 117 | void (*tmo_request_checksum)(u16 csum_start, u16 csum_offset, void *priv); | ||
| 118 | }; | ||
| 119 | |||
| 96 | #ifdef CONFIG_XDP_SOCKETS | 120 | #ifdef CONFIG_XDP_SOCKETS |
| 97 | 121 | ||
| 98 | int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp); | 122 | int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp); |
| 99 | int __xsk_map_redirect(struct xdp_sock *xs, struct xdp_buff *xdp); | 123 | int __xsk_map_redirect(struct xdp_sock *xs, struct xdp_buff *xdp); |
| 100 | void __xsk_map_flush(void); | 124 | void __xsk_map_flush(void); |
| 101 | 125 | ||
| 126 | /** | ||
| 127 | * xsk_tx_metadata_to_compl - Save enough relevant metadata information | ||
| 128 | * to perform tx completion in the future. | ||
| 129 | * @meta: pointer to AF_XDP metadata area | ||
| 130 | * @compl: pointer to output struct xsk_tx_metadata_to_compl | ||
| 131 | * | ||
| 132 | * This function should be called by the networking device when | ||
| 133 | * it prepares AF_XDP egress packet. The value of @compl should be stored | ||
| 134 | * and passed to xsk_tx_metadata_complete upon TX completion. | ||
| 135 | */ | ||
| 136 | static inline void xsk_tx_metadata_to_compl(struct xsk_tx_metadata *meta, | ||
| 137 | struct xsk_tx_metadata_compl *compl) | ||
| 138 | { | ||
| 139 | if (!meta) | ||
| 140 | return; | ||
| 141 | |||
| 142 | if (meta->flags & XDP_TXMD_FLAGS_TIMESTAMP) | ||
| 143 | compl->tx_timestamp = &meta->completion.tx_timestamp; | ||
| 144 | else | ||
| 145 | compl->tx_timestamp = NULL; | ||
| 146 | } | ||
| 147 | |||
| 148 | /** | ||
| 149 | * xsk_tx_metadata_request - Evaluate AF_XDP TX metadata at submission | ||
| 150 | * and call appropriate xsk_tx_metadata_ops operation. | ||
| 151 | * @meta: pointer to AF_XDP metadata area | ||
| 152 | * @ops: pointer to struct xsk_tx_metadata_ops | ||
| 153 | * @priv: pointer to driver-private aread | ||
| 154 | * | ||
| 155 | * This function should be called by the networking device when | ||
| 156 | * it prepares AF_XDP egress packet. | ||
| 157 | */ | ||
| 158 | static inline void xsk_tx_metadata_request(const struct xsk_tx_metadata *meta, | ||
| 159 | const struct xsk_tx_metadata_ops *ops, | ||
| 160 | void *priv) | ||
| 161 | { | ||
| 162 | if (!meta) | ||
| 163 | return; | ||
| 164 | |||
| 165 | if (ops->tmo_request_timestamp) | ||
| 166 | if (meta->flags & XDP_TXMD_FLAGS_TIMESTAMP) | ||
| 167 | ops->tmo_request_timestamp(priv); | ||
| 168 | |||
| 169 | if (ops->tmo_request_checksum) | ||
| 170 | if (meta->flags & XDP_TXMD_FLAGS_CHECKSUM) | ||
| 171 | ops->tmo_request_checksum(meta->request.csum_start, | ||
| 172 | meta->request.csum_offset, priv); | ||
| 173 | } | ||
| 174 | |||
| 175 | /** | ||
| 176 | * xsk_tx_metadata_complete - Evaluate AF_XDP TX metadata at completion | ||
| 177 | * and call appropriate xsk_tx_metadata_ops operation. | ||
| 178 | * @compl: pointer to completion metadata produced from xsk_tx_metadata_to_compl | ||
| 179 | * @ops: pointer to struct xsk_tx_metadata_ops | ||
| 180 | * @priv: pointer to driver-private aread | ||
| 181 | * | ||
| 182 | * This function should be called by the networking device upon | ||
| 183 | * AF_XDP egress completion. | ||
| 184 | */ | ||
| 185 | static inline void xsk_tx_metadata_complete(struct xsk_tx_metadata_compl *compl, | ||
| 186 | const struct xsk_tx_metadata_ops *ops, | ||
| 187 | void *priv) | ||
| 188 | { | ||
| 189 | if (!compl) | ||
| 190 | return; | ||
| 191 | |||
| 192 | *compl->tx_timestamp = ops->tmo_fill_timestamp(priv); | ||
| 193 | } | ||
| 194 | |||
| 102 | #else | 195 | #else |
| 103 | 196 | ||
| 104 | static inline int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp) | 197 | static inline int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp) |
| @@ -115,6 +208,23 @@ static inline void __xsk_map_flush(void) · 10 unmodified lines | |||
| 115 | { | 208 | { |
| 116 | } | 209 | } |
| 117 | 210 | ||
| 211 | static inline void xsk_tx_metadata_to_compl(struct xsk_tx_metadata *meta, | ||
| 212 | struct xsk_tx_metadata_compl *compl) | ||
| 213 | { | ||
| 214 | } | ||
| 215 | |||
| 216 | static inline void xsk_tx_metadata_request(struct xsk_tx_metadata *meta, | ||
| 217 | const struct xsk_tx_metadata_ops *ops, | ||
| 218 | void *priv) | ||
| 219 | { | ||
| 220 | } | ||
| 221 | |||
| 222 | static inline void xsk_tx_metadata_complete(struct xsk_tx_metadata_compl *compl, | ||
| 223 | const struct xsk_tx_metadata_ops *ops, | ||
| 224 | void *priv) | ||
| 225 | { | ||
| 226 | } | ||
| 227 | |||
| 118 | #endif /* CONFIG_XDP_SOCKETS */ | 228 | #endif /* CONFIG_XDP_SOCKETS */ |
| 119 | 229 | ||
| 120 | #if defined(CONFIG_XDP_SOCKETS) && defined(CONFIG_DEBUG_NET) | 230 | #if defined(CONFIG_XDP_SOCKETS) && defined(CONFIG_DEBUG_NET) |
| @@ -165,6 +165,14 @@ static inline void *xsk_buff_raw_get_data(struct xsk_buff_pool *pool, u64 addr) · 164 unmodified lines | |||
| 165 | return xp_raw_get_data(pool, addr); | 165 | return xp_raw_get_data(pool, addr); |
| 166 | } | 166 | } |
| 167 | 167 | ||
| 168 | static inline struct xsk_tx_metadata *xsk_buff_get_metadata(struct xsk_buff_pool *pool, u64 addr) | ||
| 169 | { | ||
| 170 | if (!pool->tx_metadata_len) | ||
| 171 | return NULL; | ||
| 172 | |||
| 173 | return xp_raw_get_data(pool, addr) - pool->tx_metadata_len; | ||
| 174 | } | ||
| 175 | |||
| 168 | static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp, struct xsk_buff_pool *pool) | 176 | static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp, struct xsk_buff_pool *pool) |
| 169 | { | 177 | { |
| 170 | struct xdp_buff_xsk *xskb = container_of(xdp, struct xdp_buff_xsk, xdp); | 178 | struct xdp_buff_xsk *xskb = container_of(xdp, struct xdp_buff_xsk, xdp); |
| @@ -324,6 +332,11 @@ static inline void *xsk_buff_raw_get_data(struct xsk_buff_pool *pool, u64 addr) · 153 unmodified lines | |||
| 324 | return NULL; | 332 | return NULL; |
| 325 | } | 333 | } |
| 326 | 334 | ||
| 335 | static inline struct xsk_tx_metadata *xsk_buff_get_metadata(struct xsk_buff_pool *pool, u64 addr) | ||
| 336 | { | ||
| 337 | return NULL; | ||
| 338 | } | ||
| 339 | |||
| 327 | static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp, struct xsk_buff_pool *pool) | 340 | static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp, struct xsk_buff_pool *pool) |
| 328 | { | 341 | { |
| 329 | } | 342 | } |
| @@ -33,6 +33,7 @@ struct xdp_buff_xsk { · 32 unmodified lines | |||
| 33 | }; | 33 | }; |
| 34 | 34 | ||
| 35 | #define XSK_CHECK_PRIV_TYPE(t) BUILD_BUG_ON(sizeof(t) > offsetofend(struct xdp_buff_xsk, cb)) | 35 | #define XSK_CHECK_PRIV_TYPE(t) BUILD_BUG_ON(sizeof(t) > offsetofend(struct xdp_buff_xsk, cb)) |
| 36 | #define XSK_TX_COMPL_FITS(t) BUILD_BUG_ON(sizeof(struct xsk_tx_metadata_compl) > sizeof(t)) | ||
| 36 | 37 | ||
| 37 | struct xsk_dma_map { | 38 | struct xsk_dma_map { |
| 38 | dma_addr_t *dma_pages; | 39 | dma_addr_t *dma_pages; |
| @@ -234,4 +235,9 @@ static inline u64 xp_get_handle(struct xdp_buff_xsk *xskb) · 195 unmodified lines | |||
| 234 | return xskb->orig_addr + (offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT); | 235 | return xskb->orig_addr + (offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT); |
| 235 | } | 236 | } |
| 236 | 237 | ||
| 238 | static inline bool xp_tx_metadata_enabled(const struct xsk_buff_pool *pool) | ||
| 239 | { | ||
| 240 | return pool->tx_metadata_len > 0; | ||
| 241 | } | ||
| 242 | |||
| 237 | #endif /* XSK_BUFF_POOL_H_ */ | 243 | #endif /* XSK_BUFF_POOL_H_ */ |
| @@ -106,6 +106,41 @@ struct xdp_options { · 105 unmodified lines | |||
| 106 | #define XSK_UNALIGNED_BUF_ADDR_MASK \ | 106 | #define XSK_UNALIGNED_BUF_ADDR_MASK \ |
| 107 | ((1ULL << XSK_UNALIGNED_BUF_OFFSET_SHIFT) - 1) | 107 | ((1ULL << XSK_UNALIGNED_BUF_OFFSET_SHIFT) - 1) |
| 108 | 108 | ||
| 109 | /* Request transmit timestamp. Upon completion, put it into tx_timestamp | ||
| 110 | * field of struct xsk_tx_metadata. | ||
| 111 | */ | ||
| 112 | #define XDP_TXMD_FLAGS_TIMESTAMP (1 << 0) | ||
| 113 | |||
| 114 | /* Request transmit checksum offload. Checksum start position and offset | ||
| 115 | * are communicated via csum_start and csum_offset fields of struct | ||
| 116 | * xsk_tx_metadata. | ||
| 117 | */ | ||
| 118 | #define XDP_TXMD_FLAGS_CHECKSUM (1 << 1) | ||
| 119 | |||
| 120 | /* AF_XDP offloads request. 'request' union member is consumed by the driver | ||
| 121 | * when the packet is being transmitted. 'completion' union member is | ||
| 122 | * filled by the driver when the transmit completion arrives. | ||
| 123 | */ | ||
| 124 | struct xsk_tx_metadata { | ||
| 125 | __u64 flags; | ||
| 126 | |||
| 127 | union { | ||
| 128 | struct { | ||
| 129 | /* XDP_TXMD_FLAGS_CHECKSUM */ | ||
| 130 | |||
| 131 | /* Offset from desc->addr where checksumming should start. */ | ||
| 132 | __u16 csum_start; | ||
| 133 | /* Offset from csum_start where checksum should be stored. */ | ||
| 134 | __u16 csum_offset; | ||
| 135 | } request; | ||
| 136 | |||
| 137 | struct { | ||
| 138 | /* XDP_TXMD_FLAGS_TIMESTAMP */ | ||
| 139 | __u64 tx_timestamp; | ||
| 140 | } completion; | ||
| 141 | }; | ||
| 142 | }; | ||
| 143 | |||
| 109 | /* Rx/Tx descriptor */ | 144 | /* Rx/Tx descriptor */ |
| 110 | struct xdp_desc { | 145 | struct xdp_desc { |
| 111 | __u64 addr; | 146 | __u64 addr; |
| @@ -122,4 +157,7 @@ struct xdp_desc { · 10 unmodified lines | |||
| 122 | */ | 157 | */ |
| 123 | #define XDP_PKT_CONTD (1 << 0) | 158 | #define XDP_PKT_CONTD (1 << 0) |
| 124 | 159 | ||
| 160 | /* TX packet carries valid metadata. */ | ||
| 161 | #define XDP_TX_METADATA (1 << 1) | ||
| 162 | |||
| 125 | #endif /* _LINUX_IF_XDP_H */ | 163 | #endif /* _LINUX_IF_XDP_H */ |
| @@ -53,12 +53,28 @@ enum netdev_xdp_rx_metadata { · 52 unmodified lines | |||
| 53 | NETDEV_XDP_RX_METADATA_MASK = 3, | 53 | NETDEV_XDP_RX_METADATA_MASK = 3, |
| 54 | }; | 54 | }; |
| 55 | 55 | ||
| 56 | /** | ||
| 57 | * enum netdev_xsk_flags | ||
| 58 | * @NETDEV_XSK_FLAGS_TX_TIMESTAMP: HW timestamping egress packets is supported | ||
| 59 | * by the driver. | ||
| 60 | * @NETDEV_XSK_FLAGS_TX_CHECKSUM: L3 checksum HW offload is supported by the | ||
| 61 | * driver. | ||
| 62 | */ | ||
| 63 | enum netdev_xsk_flags { | ||
| 64 | NETDEV_XSK_FLAGS_TX_TIMESTAMP = 1, | ||
| 65 | NETDEV_XSK_FLAGS_TX_CHECKSUM = 2, | ||
| 66 | |||
| 67 | /* private: */ | ||
| 68 | NETDEV_XSK_FLAGS_MASK = 3, | ||
| 69 | }; | ||
| 70 | |||
| 56 | enum { | 71 | enum { |
| 57 | NETDEV_A_DEV_IFINDEX = 1, | 72 | NETDEV_A_DEV_IFINDEX = 1, |
| 58 | NETDEV_A_DEV_PAD, | 73 | NETDEV_A_DEV_PAD, |
| 59 | NETDEV_A_DEV_XDP_FEATURES, | 74 | NETDEV_A_DEV_XDP_FEATURES, |
| 60 | NETDEV_A_DEV_XDP_ZC_MAX_SEGS, | 75 | NETDEV_A_DEV_XDP_ZC_MAX_SEGS, |
| 61 | NETDEV_A_DEV_XDP_RX_METADATA_FEATURES, | 76 | NETDEV_A_DEV_XDP_RX_METADATA_FEATURES, |
| 77 | NETDEV_A_DEV_XSK_FEATURES, | ||
| 62 | 78 | ||
| 63 | __NETDEV_A_DEV_MAX, | 79 | __NETDEV_A_DEV_MAX, |
| 64 | NETDEV_A_DEV_MAX = (__NETDEV_A_DEV_MAX - 1) | 80 | NETDEV_A_DEV_MAX = (__NETDEV_A_DEV_MAX - 1) |