[xdp-hints] [PATCH bpf-next v5 00/13] xsk: TX metadata

[xdp-hints] [PATCH bpf-next v5 00/13] xsk: TX metadata

[Stanislav Fomichev]

This series implements initial TX metadata (offloads) for AF_XDP.
See patch #2 for the main implementation and mlx5/stmmac ones for the
example on how to consume the metadata on the device side.

Starting with two types of offloads:
- request TX timestamp (and write it back into the metadata area)
- request TX checksum offload

Changes since v4:
- remove 'render-max: true' from spec (Jakub)
- move xsk_tx_metadata_ops into include/net/xdp_sock.h (Jakub)
- christmas tree in netdev_nl_dev_fill (Jakub)
- fix > vs >= when dumping masks in samples (Jakub)
- switch to 8-byte alignment for tx metadata length (Jakub)
- spelling fixes in the doc (Magnus)
- deny metadata length >= 256 (Magnus)
- validate metadata flags and deny unknown ones (Jakub)
- move XDP_TX_METADATA_CHECKSUM_SW into umem config flag (Jakub)
- don't print timestamps twice in xdp_hw_metadata (Song)
- rename anonymous xsk_tx_metadata member into request (Alexei)
- add comment to xsk_tx_metadata (Alexei)

I've separated new bits that need a closer review into separate patches:
- xsk_tx_metadata flags validation:
  - xsk: Validate xsk_tx_metadata flags
- new umem flag for sw tx csum calculation (instead of per-packet flag)
  - xsk: Add option to calculate TX checksum in SW

v4: https://lore.kernel.org/bpf/20231019174944.3376335-1-sdf@google.com/

Performance (mlx5):

I've implemented a small xskgen tool to try to saturate single tx queue:
https://github.com/fomichev/xskgen/tree/master

Here are the performance numbers with some analysis.

1. Baseline. Running with commit eb62e6aef940 ("Merge branch 'bpf:
Support bpf_get_func_ip helper in uprobes'"), nothing from this series:

- with 1400 bytes of payload: 98 gbps, 8 mpps
./xskgen -s 1400 -b eth3 10:70:fd:48:10:77 10:70:fd:48:10:87 fe80::1270:fdff:fe48:1077 fe80::1270:fdff:fe48:1087 1 1
sent 10000000 packets 116960000000 bits, took 1.189130 sec, 98.357623 gbps 8.409509 mpps

- with 200 bytes of payload: 49 gbps, 23 mpps
./xskgen -s 200 -b eth3 10:70:fd:48:10:77 10:70:fd:48:10:87 fe80::1270:fdff:fe48:1077 fe80::1270:fdff:fe48:1087 1 1
sent 10000064 packets 20960134144 bits, took 0.422235 sec, 49.640921 gbps 23.683645 mpps

2. Adding single commit that supports reserving tx_metadata_len
   changes nothing numbers-wise.

- baseline for 1400
./xskgen -s 1400 -b eth3 10:70:fd:48:10:77 10:70:fd:48:10:87 fe80::1270:fdff:fe48:1077 fe80::1270:fdff:fe48:1087 1 1
sent 10000000 packets 116960000000 bits, took 1.189247 sec, 98.347946 gbps 8.408682 mpps

- baseline for 200
./xskgen -s 200 -b eth3 10:70:fd:48:10:77 10:70:fd:48:10:87 fe80::1270:fdff:fe48:1077 fe80::1270:fdff:fe48:1087 1 1
sent 10000000 packets 20960000000 bits, took 0.421248 sec, 49.756913 gbps 23.738985 mpps

3. Adding -M flag causes xskgen to reserve the metadata and fill it, but
   doesn't set XDP_TX_METADATA descriptor option.

- new baseline for 1400 (with only filling the metadata)
./xskgen -M -s 1400 -b eth3 10:70:fd:48:10:77 10:70:fd:48:10:87 fe80::1270:fdff:fe48:1077 fe80::1270:fdff:fe48:1087 1 1
sent 10000000 packets 116960000000 bits, took 1.188767 sec, 98.387657 gbps 8.412077 mpps

- new baseline for 200 (with only filling the metadata)
./xskgen -M -s 200 -b eth3 10:70:fd:48:10:77 10:70:fd:48:10:87 fe80::1270:fdff:fe48:1077 fe80::1270:fdff:fe48:1087 1 1
sent 10000000 packets 20960000000 bits, took 0.410213 sec, 51.095407 gbps 24.377579 mpps
(the numbers go sligtly up here, not really sure why, maybe some cache-related
side-effects?

4. Next, I'm running the same test but with the commit that adds actual
   general infra to parse XDP_TX_METADATA (but no driver support).
   Essentially applying "xsk: add TX timestamp and TX checksum offload support"
   from this series. Numbers are the same.

- fill metadata for 1400
./xskgen -M -s 1400 -b eth3 10:70:fd:48:10:77 10:70:fd:48:10:87 fe80::1270:fdff:fe48:1077 fe80::1270:fdff:fe48:1087 1 1
sent 10000000 packets 116960000000 bits, took 1.188430 sec, 98.415557 gbps 8.414463 mpps

- fill metadata for 200
./xskgen -M -s 200 -b eth3 10:70:fd:48:10:77 10:70:fd:48:10:87 fe80::1270:fdff:fe48:1077 fe80::1270:fdff:fe48:1087 1 1
sent 10000000 packets 20960000000 bits, took 0.411559 sec, 50.928299 gbps 24.297853 mpps

- request metadata for 1400
./xskgen -m -s 1400 -b eth3 10:70:fd:48:10:77 10:70:fd:48:10:87 fe80::1270:fdff:fe48:1077 fe80::1270:fdff:fe48:1087 1 1
sent 10000000 packets 116960000000 bits, took 1.188723 sec, 98.391299 gbps 8.412389 mpps

- request metadata for 200
./xskgen -m -s 200 -b eth3 10:70:fd:48:10:77 10:70:fd:48:10:87 fe80::1270:fdff:fe48:1077 fe80::1270:fdff:fe48:1087 1 1
sent 10000064 packets 20960134144 bits, took 0.411240 sec, 50.968131 gbps 24.316856 mpps

5. Now, for the most interesting part, I'm adding mlx5 driver support.
   The mpps for 200 bytes case goes down from 23 mpps to 19 mpps, but
   _only_ when I enable the metadata. This looks like a side effect
   of me pushing extra metadata pointer via mlx5e_xdpi_fifo_push.
   Hence, this part is wrapped into 'if (xp_tx_metadata_enabled)'
   to not affect the existing non-metadata use-cases. Since this is not
   regressing existing workloads, I'm not spending any time trying to
   optimize it more (and leaving it up to mlx owners to purse if
   they see any good way to do it).

- same baseline
./xskgen -s 1400 -b eth3 10:70:fd:48:10:77 10:70:fd:48:10:87 fe80::1270:fdff:fe48:1077 fe80::1270:fdff:fe48:1087 1 1
sent 10000000 packets 116960000000 bits, took 1.189434 sec, 98.332484 gbps 8.407360 mpps

./xskgen -s 200 -b eth3 10:70:fd:48:10:77 10:70:fd:48:10:87 fe80::1270:fdff:fe48:1077 fe80::1270:fdff:fe48:1087 1 1
sent 10000128 packets 20960268288 bits, took 0.425254 sec, 49.288821 gbps 23.515659 mpps

- fill metadata for 1400
./xskgen -M -s 1400 -b eth3 10:70:fd:48:10:77 10:70:fd:48:10:87 fe80::1270:fdff:fe48:1077 fe80::1270:fdff:fe48:1087 1 1
sent 10000000 packets 116960000000 bits, took 1.189528 sec, 98.324714 gbps 8.406696 mpps

- fill metadata for 200
./xskgen -M -s 200 -b eth3 10:70:fd:48:10:77 10:70:fd:48:10:87 fe80::1270:fdff:fe48:1077 fe80::1270:fdff:fe48:1087 1 1
sent 10000128 packets 20960268288 bits, took 0.519085 sec, 40.379260 gbps 19.264914 mpps

- request metadata for 1400
./xskgen -m -s 1400 -b eth3 10:70:fd:48:10:77 10:70:fd:48:10:87 fe80::1270:fdff:fe48:1077 fe80::1270:fdff:fe48:1087 1 1
sent 10000000 packets 116960000000 bits, took 1.189329 sec, 98.341165 gbps 8.408102 mpps

- request metadata for 200
./xskgen -m -s 200 -b eth3 10:70:fd:48:10:77 10:70:fd:48:10:87 fe80::1270:fdff:fe48:1077 fe80::1270:fdff:fe48:1087 1 1
sent 10000128 packets 20960268288 bits, took 0.519929 sec, 40.313713 gbps 19.233642 mpps

Acked-by: Magnus Karlsson <magnus.karlsson@intel.com>

Song Yoong Siang (1):
  net: stmmac: Add Tx HWTS support to XDP ZC

Stanislav Fomichev (12):
  xsk: Support tx_metadata_len
  xsk: Add TX timestamp and TX checksum offload support
  tools: ynl: Print xsk-features from the sample
  net/mlx5e: Implement AF_XDP TX timestamp and checksum offload
  xsk: Document tx_metadata_len layout
  xsk: Validate xsk_tx_metadata flags
  xsk: Add option to calculate TX checksum in SW
  selftests/xsk: Support tx_metadata_len
  selftests/bpf: Add csum helpers
  selftests/bpf: Add TX side to xdp_metadata
  selftests/bpf: Convert xdp_hw_metadata to XDP_USE_NEED_WAKEUP
  selftests/bpf: Add TX side to xdp_hw_metadata

 Documentation/netlink/specs/netdev.yaml       |  19 +-
 Documentation/networking/index.rst            |   1 +
 Documentation/networking/xsk-tx-metadata.rst  |  79 ++++++
 drivers/net/ethernet/mellanox/mlx5/core/en.h  |   4 +-
 .../net/ethernet/mellanox/mlx5/core/en/xdp.c  |  72 +++++-
 .../net/ethernet/mellanox/mlx5/core/en/xdp.h  |  11 +-
 .../ethernet/mellanox/mlx5/core/en/xsk/tx.c   |  17 +-
 .../net/ethernet/mellanox/mlx5/core/en_main.c |   1 +
 drivers/net/ethernet/stmicro/stmmac/stmmac.h  |  12 +
 .../net/ethernet/stmicro/stmmac/stmmac_main.c |  64 ++++-
 include/linux/netdevice.h                     |   2 +
 include/linux/skbuff.h                        |  14 +-
 include/net/xdp_sock.h                        | 111 +++++++++
 include/net/xdp_sock_drv.h                    |  34 +++
 include/net/xsk_buff_pool.h                   |   8 +
 include/uapi/linux/if_xdp.h                   |  47 +++-
 include/uapi/linux/netdev.h                   |  16 ++
 net/core/netdev-genl.c                        |  13 +-
 net/xdp/xdp_umem.c                            |  11 +-
 net/xdp/xsk.c                                 |  56 ++++-
 net/xdp/xsk_buff_pool.c                       |   2 +
 net/xdp/xsk_queue.h                           |  19 +-
 tools/include/uapi/linux/if_xdp.h             |  61 ++++-
 tools/include/uapi/linux/netdev.h             |  16 ++
 tools/net/ynl/generated/netdev-user.c         |  19 ++
 tools/net/ynl/generated/netdev-user.h         |   3 +
 tools/net/ynl/samples/netdev.c                |  10 +-
 tools/testing/selftests/bpf/network_helpers.h |  43 ++++
 .../selftests/bpf/prog_tests/xdp_metadata.c   |  33 ++-
 tools/testing/selftests/bpf/xdp_hw_metadata.c | 235 ++++++++++++++++--
 tools/testing/selftests/bpf/xsk.c             |   3 +
 tools/testing/selftests/bpf/xsk.h             |   1 +
 32 files changed, 967 insertions(+), 70 deletions(-)
 create mode 100644 Documentation/networking/xsk-tx-metadata.rst

-- 
2.42.0.869.gea05f2083d-goog

[xdp-hints] [PATCH bpf-next v5 01/13] xsk: Support tx_metadata_len

[Stanislav Fomichev]

For zerocopy mode, tx_desc->addr can point to an arbitrary offset
and carry some TX metadata in the headroom. For copy mode, there
is no way currently to populate skb metadata.

Introduce new tx_metadata_len umem config option that indicates how many
bytes to treat as metadata. Metadata bytes come prior to tx_desc address
(same as in RX case).

The size of the metadata has mostly the same constraints as XDP:
- less than 256 bytes
- 8-byte aligned (compared to 4-byte alignment on xdp, due to 8-byte
  timestamp in the completion)
- non-zero

This data is not interpreted in any way right now.

Reviewed-by: Song Yoong Siang <yoong.siang.song@intel.com>
Signed-off-by: Stanislav Fomichev <sdf@google.com>
---
 include/net/xdp_sock.h            |  1 +
 include/net/xsk_buff_pool.h       |  1 +
 include/uapi/linux/if_xdp.h       |  1 +
 net/xdp/xdp_umem.c                |  4 ++++
 net/xdp/xsk.c                     | 12 +++++++++++-
 net/xdp/xsk_buff_pool.c           |  1 +
 net/xdp/xsk_queue.h               | 17 ++++++++++-------
 tools/include/uapi/linux/if_xdp.h |  1 +
 8 files changed, 30 insertions(+), 8 deletions(-)

diff --git a/include/net/xdp_sock.h b/include/net/xdp_sock.h
index f83128007fb0..bcf765124f72 100644
--- a/include/net/xdp_sock.h
+++ b/include/net/xdp_sock.h
@@ -30,6 +30,7 @@ struct xdp_umem {
 	struct user_struct *user;
 	refcount_t users;
 	u8 flags;
+	u8 tx_metadata_len;
 	bool zc;
 	struct page **pgs;
 	int id;
diff --git a/include/net/xsk_buff_pool.h b/include/net/xsk_buff_pool.h
index b0bdff26fc88..1985ffaf9b0c 100644
--- a/include/net/xsk_buff_pool.h
+++ b/include/net/xsk_buff_pool.h
@@ -77,6 +77,7 @@ struct xsk_buff_pool {
 	u32 chunk_size;
 	u32 chunk_shift;
 	u32 frame_len;
+	u8 tx_metadata_len; /* inherited from umem */
 	u8 cached_need_wakeup;
 	bool uses_need_wakeup;
 	bool dma_need_sync;
diff --git a/include/uapi/linux/if_xdp.h b/include/uapi/linux/if_xdp.h
index 8d48863472b9..2ecf79282c26 100644
--- a/include/uapi/linux/if_xdp.h
+++ b/include/uapi/linux/if_xdp.h
@@ -76,6 +76,7 @@ struct xdp_umem_reg {
 	__u32 chunk_size;
 	__u32 headroom;
 	__u32 flags;
+	__u32 tx_metadata_len;
 };
 
 struct xdp_statistics {
diff --git a/net/xdp/xdp_umem.c b/net/xdp/xdp_umem.c
index 06cead2b8e34..946a687fb8e8 100644
--- a/net/xdp/xdp_umem.c
+++ b/net/xdp/xdp_umem.c
@@ -199,6 +199,9 @@ static int xdp_umem_reg(struct xdp_umem *umem, struct xdp_umem_reg *mr)
 	if (headroom >= chunk_size - XDP_PACKET_HEADROOM)
 		return -EINVAL;
 
+	if (mr->tx_metadata_len >= 256 || mr->tx_metadata_len % 8)
+		return -EINVAL;
+
 	umem->size = size;
 	umem->headroom = headroom;
 	umem->chunk_size = chunk_size;
@@ -207,6 +210,7 @@ static int xdp_umem_reg(struct xdp_umem *umem, struct xdp_umem_reg *mr)
 	umem->pgs = NULL;
 	umem->user = NULL;
 	umem->flags = mr->flags;
+	umem->tx_metadata_len = mr->tx_metadata_len;
 
 	INIT_LIST_HEAD(&umem->xsk_dma_list);
 	refcount_set(&umem->users, 1);
diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c
index ae9f8cb611f6..c904356e2800 100644
--- a/net/xdp/xsk.c
+++ b/net/xdp/xsk.c
@@ -1283,6 +1283,14 @@ struct xdp_umem_reg_v1 {
 	__u32 headroom;
 };
 
+struct xdp_umem_reg_v2 {
+	__u64 addr; /* Start of packet data area */
+	__u64 len; /* Length of packet data area */
+	__u32 chunk_size;
+	__u32 headroom;
+	__u32 flags;
+};
+
 static int xsk_setsockopt(struct socket *sock, int level, int optname,
 			  sockptr_t optval, unsigned int optlen)
 {
@@ -1326,8 +1334,10 @@ static int xsk_setsockopt(struct socket *sock, int level, int optname,
 
 		if (optlen < sizeof(struct xdp_umem_reg_v1))
 			return -EINVAL;
-		else if (optlen < sizeof(mr))
+		else if (optlen < sizeof(struct xdp_umem_reg_v2))
 			mr_size = sizeof(struct xdp_umem_reg_v1);
+		else if (optlen < sizeof(mr))
+			mr_size = sizeof(struct xdp_umem_reg_v2);
 
 		if (copy_from_sockptr(&mr, optval, mr_size))
 			return -EFAULT;
diff --git a/net/xdp/xsk_buff_pool.c b/net/xdp/xsk_buff_pool.c
index 49cb9f9a09be..386eddcdf837 100644
--- a/net/xdp/xsk_buff_pool.c
+++ b/net/xdp/xsk_buff_pool.c
@@ -85,6 +85,7 @@ struct xsk_buff_pool *xp_create_and_assign_umem(struct xdp_sock *xs,
 		XDP_PACKET_HEADROOM;
 	pool->umem = umem;
 	pool->addrs = umem->addrs;
+	pool->tx_metadata_len = umem->tx_metadata_len;
 	INIT_LIST_HEAD(&pool->free_list);
 	INIT_LIST_HEAD(&pool->xskb_list);
 	INIT_LIST_HEAD(&pool->xsk_tx_list);
diff --git a/net/xdp/xsk_queue.h b/net/xdp/xsk_queue.h
index 13354a1e4280..c74a1372bcb9 100644
--- a/net/xdp/xsk_queue.h
+++ b/net/xdp/xsk_queue.h
@@ -143,15 +143,17 @@ static inline bool xp_unused_options_set(u32 options)
 static inline bool xp_aligned_validate_desc(struct xsk_buff_pool *pool,
 					    struct xdp_desc *desc)
 {
-	u64 offset = desc->addr & (pool->chunk_size - 1);
+	u64 addr = desc->addr - pool->tx_metadata_len;
+	u64 len = desc->len + pool->tx_metadata_len;
+	u64 offset = addr & (pool->chunk_size - 1);
 
 	if (!desc->len)
 		return false;
 
-	if (offset + desc->len > pool->chunk_size)
+	if (offset + len > pool->chunk_size)
 		return false;
 
-	if (desc->addr >= pool->addrs_cnt)
+	if (addr >= pool->addrs_cnt)
 		return false;
 
 	if (xp_unused_options_set(desc->options))
@@ -162,16 +164,17 @@ static inline bool xp_aligned_validate_desc(struct xsk_buff_pool *pool,
 static inline bool xp_unaligned_validate_desc(struct xsk_buff_pool *pool,
 					      struct xdp_desc *desc)
 {
-	u64 addr = xp_unaligned_add_offset_to_addr(desc->addr);
+	u64 addr = xp_unaligned_add_offset_to_addr(desc->addr) - pool->tx_metadata_len;
+	u64 len = desc->len + pool->tx_metadata_len;
 
 	if (!desc->len)
 		return false;
 
-	if (desc->len > pool->chunk_size)
+	if (len > pool->chunk_size)
 		return false;
 
-	if (addr >= pool->addrs_cnt || addr + desc->len > pool->addrs_cnt ||
-	    xp_desc_crosses_non_contig_pg(pool, addr, desc->len))
+	if (addr >= pool->addrs_cnt || addr + len > pool->addrs_cnt ||
+	    xp_desc_crosses_non_contig_pg(pool, addr, len))
 		return false;
 
 	if (xp_unused_options_set(desc->options))
diff --git a/tools/include/uapi/linux/if_xdp.h b/tools/include/uapi/linux/if_xdp.h
index 73a47da885dc..34411a2e5b6c 100644
--- a/tools/include/uapi/linux/if_xdp.h
+++ b/tools/include/uapi/linux/if_xdp.h
@@ -76,6 +76,7 @@ struct xdp_umem_reg {
 	__u32 chunk_size;
 	__u32 headroom;
 	__u32 flags;
+	__u32 tx_metadata_len;
 };
 
 struct xdp_statistics {
-- 
2.42.0.869.gea05f2083d-goog

[xdp-hints] [PATCH bpf-next v5 02/13] xsk: Add TX timestamp and TX checksum offload support

[Stanislav Fomichev]

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.

Reviewed-by: Song Yoong Siang <yoong.siang.song@intel.com>
Signed-off-by: Stanislav Fomichev <sdf@google.com>
---
 Documentation/netlink/specs/netdev.yaml |  19 +++-
 include/linux/netdevice.h               |   2 +
 include/linux/skbuff.h                  |  14 ++-
 include/net/xdp_sock.h                  | 110 ++++++++++++++++++++++++
 include/net/xdp_sock_drv.h              |  13 +++
 include/net/xsk_buff_pool.h             |   6 ++
 include/uapi/linux/if_xdp.h             |  38 ++++++++
 include/uapi/linux/netdev.h             |  16 ++++
 net/core/netdev-genl.c                  |  13 ++-
 net/xdp/xsk.c                           |  34 ++++++++
 net/xdp/xsk_queue.h                     |   2 +-
 tools/include/uapi/linux/if_xdp.h       |  52 +++++++++--
 tools/include/uapi/linux/netdev.h       |  16 ++++
 tools/net/ynl/generated/netdev-user.c   |  19 ++++
 tools/net/ynl/generated/netdev-user.h   |   3 +
 15 files changed, 348 insertions(+), 9 deletions(-)

diff --git a/Documentation/netlink/specs/netdev.yaml b/Documentation/netlink/specs/netdev.yaml
index 14511b13f305..00439bcbd2e3 100644
--- a/Documentation/netlink/specs/netdev.yaml
+++ b/Documentation/netlink/specs/netdev.yaml
@@ -45,7 +45,6 @@ name: netdev
   -
     type: flags
     name: xdp-rx-metadata
-    render-max: true
     entries:
       -
         name: timestamp
@@ -55,6 +54,18 @@ name: netdev
         name: hash
         doc:
           Device is capable of exposing receive packet hash via bpf_xdp_metadata_rx_hash().
+  -
+    type: flags
+    name: xsk-flags
+    entries:
+      -
+        name: tx-timestamp
+        doc:
+          HW timestamping egress packets is supported by the driver.
+      -
+        name: tx-checksum
+        doc:
+          L3 checksum HW offload is supported by the driver.
 
 attribute-sets:
   -
@@ -86,6 +97,11 @@ name: netdev
              See Documentation/networking/xdp-rx-metadata.rst for more details.
         type: u64
         enum: xdp-rx-metadata
+      -
+        name: xsk-features
+        doc: Bitmask of enabled AF_XDP features.
+        type: u64
+        enum: xsk-flags
 
 operations:
   list:
@@ -103,6 +119,7 @@ name: netdev
             - xdp-features
             - xdp-zc-max-segs
             - xdp-rx-metadata-features
+            - xsk-features
       dump:
         reply: *dev-all
     -
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index a16c9cc063fe..fe915fa334cc 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1858,6 +1858,7 @@ enum netdev_ml_priv_type {
  *	@netdev_ops:	Includes several pointers to callbacks,
  *			if one wants to override the ndo_*() functions
  *	@xdp_metadata_ops:	Includes pointers to XDP metadata callbacks.
+ *	@xsk_tx_metadata_ops:	Includes pointers to AF_XDP TX metadata callbacks.
  *	@ethtool_ops:	Management operations
  *	@l3mdev_ops:	Layer 3 master device operations
  *	@ndisc_ops:	Includes callbacks for different IPv6 neighbour
@@ -2117,6 +2118,7 @@ struct net_device {
 	unsigned long long	priv_flags;
 	const struct net_device_ops *netdev_ops;
 	const struct xdp_metadata_ops *xdp_metadata_ops;
+	const struct xsk_tx_metadata_ops *xsk_tx_metadata_ops;
 	int			ifindex;
 	unsigned short		gflags;
 	unsigned short		hard_header_len;
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 27998f73183e..b370eb8d70f7 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -566,6 +566,15 @@ struct ubuf_info_msgzc {
 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
 void mm_unaccount_pinned_pages(struct mmpin *mmp);
 
+/* Preserve some data across TX submission and completion.
+ *
+ * Note, this state is stored in the driver. Extending the layout
+ * might need some special care.
+ */
+struct xsk_tx_metadata_compl {
+	__u64 *tx_timestamp;
+};
+
 /* This data is invariant across clones and lives at
  * the end of the header data, ie. at skb->end.
  */
@@ -578,7 +587,10 @@ struct skb_shared_info {
 	/* Warning: this field is not always filled in (UFO)! */
 	unsigned short	gso_segs;
 	struct sk_buff	*frag_list;
-	struct skb_shared_hwtstamps hwtstamps;
+	union {
+		struct skb_shared_hwtstamps hwtstamps;
+		struct xsk_tx_metadata_compl xsk_meta;
+	};
 	unsigned int	gso_type;
 	u32		tskey;
 
diff --git a/include/net/xdp_sock.h b/include/net/xdp_sock.h
index bcf765124f72..52afde24596b 100644
--- a/include/net/xdp_sock.h
+++ b/include/net/xdp_sock.h
@@ -93,12 +93,105 @@ struct xdp_sock {
 	struct xsk_queue *cq_tmp; /* Only as tmp storage before bind */
 };
 
+/*
+ * AF_XDP TX metadata hooks for network devices.
+ * The following hooks can be defined; unless noted otherwise, they are
+ * optional and can be filled with a null pointer.
+ *
+ * void (*tmo_request_timestamp)(void *priv)
+ *     Called when AF_XDP frame requested egress timestamp.
+ *
+ * u64 (*tmo_fill_timestamp)(void *priv)
+ *     Called when AF_XDP frame, that had requested egress timestamp,
+ *     received a completion. The hook needs to return the actual HW timestamp.
+ *
+ * void (*tmo_request_checksum)(u16 csum_start, u16 csum_offset, void *priv)
+ *     Called when AF_XDP frame requested HW checksum offload. csum_start
+ *     indicates position where checksumming should start.
+ *     csum_offset indicates position where checksum should be stored.
+ *
+ */
+struct xsk_tx_metadata_ops {
+	void	(*tmo_request_timestamp)(void *priv);
+	u64	(*tmo_fill_timestamp)(void *priv);
+	void	(*tmo_request_checksum)(u16 csum_start, u16 csum_offset, void *priv);
+};
+
 #ifdef CONFIG_XDP_SOCKETS
 
 int xsk_generic_rcv(struct xdp_sock *xs, struct xdp_buff *xdp);
 int __xsk_map_redirect(struct xdp_sock *xs, struct xdp_buff *xdp);
 void __xsk_map_flush(void);
 
+/**
+ *  xsk_tx_metadata_to_compl - Save enough relevant metadata information
+ *  to perform tx completion in the future.
+ *  @meta: pointer to AF_XDP metadata area
+ *  @compl: pointer to output struct xsk_tx_metadata_to_compl
+ *
+ *  This function should be called by the networking device when
+ *  it prepares AF_XDP egress packet. The value of @compl should be stored
+ *  and passed to xsk_tx_metadata_complete upon TX completion.
+ */
+static inline void xsk_tx_metadata_to_compl(struct xsk_tx_metadata *meta,
+					    struct xsk_tx_metadata_compl *compl)
+{
+	if (!meta)
+		return;
+
+	if (meta->request.flags & XDP_TXMD_FLAGS_TIMESTAMP)
+		compl->tx_timestamp = &meta->completion.tx_timestamp;
+	else
+		compl->tx_timestamp = NULL;
+}
+
+/**
+ *  xsk_tx_metadata_request - Evaluate AF_XDP TX metadata at submission
+ *  and call appropriate xsk_tx_metadata_ops operation.
+ *  @meta: pointer to AF_XDP metadata area
+ *  @ops: pointer to struct xsk_tx_metadata_ops
+ *  @priv: pointer to driver-private aread
+ *
+ *  This function should be called by the networking device when
+ *  it prepares AF_XDP egress packet.
+ */
+static inline void xsk_tx_metadata_request(const struct xsk_tx_metadata *meta,
+					   const struct xsk_tx_metadata_ops *ops,
+					   void *priv)
+{
+	if (!meta)
+		return;
+
+	if (ops->tmo_request_timestamp)
+		if (meta->request.flags & XDP_TXMD_FLAGS_TIMESTAMP)
+			ops->tmo_request_timestamp(priv);
+
+	if (ops->tmo_request_checksum)
+		if (meta->request.flags & XDP_TXMD_FLAGS_CHECKSUM)
+			ops->tmo_request_checksum(meta->request.csum_start,
+						  meta->request.csum_offset, priv);
+}
+
+/**
+ *  xsk_tx_metadata_complete - Evaluate AF_XDP TX metadata at completion
+ *  and call appropriate xsk_tx_metadata_ops operation.
+ *  @compl: pointer to completion metadata produced from xsk_tx_metadata_to_compl
+ *  @ops: pointer to struct xsk_tx_metadata_ops
+ *  @priv: pointer to driver-private aread
+ *
+ *  This function should be called by the networking device upon
+ *  AF_XDP egress completion.
+ */
+static inline void xsk_tx_metadata_complete(struct xsk_tx_metadata_compl *compl,
+					    const struct xsk_tx_metadata_ops *ops,
+					    void *priv)
+{
+	if (!compl)
+		return;
+
+	*compl->tx_timestamp = ops->tmo_fill_timestamp(priv);
+}
+
 #else
 
 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)
 {
 }
 
+static inline void xsk_tx_metadata_to_compl(struct xsk_tx_metadata *meta,
+					    struct xsk_tx_metadata_compl *compl)
+{
+}
+
+static inline void xsk_tx_metadata_request(struct xsk_tx_metadata *meta,
+					   const struct xsk_tx_metadata_ops *ops,
+					   void *priv)
+{
+}
+
+static inline void xsk_tx_metadata_complete(struct xsk_tx_metadata_compl *compl,
+					    const struct xsk_tx_metadata_ops *ops,
+					    void *priv)
+{
+}
+
 #endif /* CONFIG_XDP_SOCKETS */
 
 #if defined(CONFIG_XDP_SOCKETS) && defined(CONFIG_DEBUG_NET)
diff --git a/include/net/xdp_sock_drv.h b/include/net/xdp_sock_drv.h
index 1f6fc8c7a84c..e2558ac3e195 100644
--- a/include/net/xdp_sock_drv.h
+++ b/include/net/xdp_sock_drv.h
@@ -165,6 +165,14 @@ static inline void *xsk_buff_raw_get_data(struct xsk_buff_pool *pool, u64 addr)
 	return xp_raw_get_data(pool, addr);
 }
 
+static inline struct xsk_tx_metadata *xsk_buff_get_metadata(struct xsk_buff_pool *pool, u64 addr)
+{
+	if (!pool->tx_metadata_len)
+		return NULL;
+
+	return xp_raw_get_data(pool, addr) - pool->tx_metadata_len;
+}
+
 static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp, struct xsk_buff_pool *pool)
 {
 	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)
 	return NULL;
 }
 
+static inline struct xsk_tx_metadata *xsk_buff_get_metadata(struct xsk_buff_pool *pool, u64 addr)
+{
+	return NULL;
+}
+
 static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp, struct xsk_buff_pool *pool)
 {
 }
diff --git a/include/net/xsk_buff_pool.h b/include/net/xsk_buff_pool.h
index 1985ffaf9b0c..97f5cc10d79e 100644
--- a/include/net/xsk_buff_pool.h
+++ b/include/net/xsk_buff_pool.h
@@ -33,6 +33,7 @@ struct xdp_buff_xsk {
 };
 
 #define XSK_CHECK_PRIV_TYPE(t) BUILD_BUG_ON(sizeof(t) > offsetofend(struct xdp_buff_xsk, cb))
+#define XSK_TX_COMPL_FITS(t) BUILD_BUG_ON(sizeof(struct xsk_tx_metadata_compl) > sizeof(t))
 
 struct xsk_dma_map {
 	dma_addr_t *dma_pages;
@@ -234,4 +235,9 @@ static inline u64 xp_get_handle(struct xdp_buff_xsk *xskb)
 	return xskb->orig_addr + (offset << XSK_UNALIGNED_BUF_OFFSET_SHIFT);
 }
 
+static inline bool xp_tx_metadata_enabled(const struct xsk_buff_pool *pool)
+{
+	return pool->tx_metadata_len > 0;
+}
+
 #endif /* XSK_BUFF_POOL_H_ */
diff --git a/include/uapi/linux/if_xdp.h b/include/uapi/linux/if_xdp.h
index 2ecf79282c26..b0ee7ad19b51 100644
--- a/include/uapi/linux/if_xdp.h
+++ b/include/uapi/linux/if_xdp.h
@@ -106,6 +106,41 @@ struct xdp_options {
 #define XSK_UNALIGNED_BUF_ADDR_MASK \
 	((1ULL << XSK_UNALIGNED_BUF_OFFSET_SHIFT) - 1)
 
+/* Request transmit timestamp. Upon completion, put it into tx_timestamp
+ * field of struct xsk_tx_metadata.
+ */
+#define XDP_TXMD_FLAGS_TIMESTAMP		(1 << 0)
+
+/* Request transmit checksum offload. Checksum start position and offset
+ * are communicated via csum_start and csum_offset fields of struct
+ * xsk_tx_metadata.
+ */
+#define XDP_TXMD_FLAGS_CHECKSUM			(1 << 1)
+
+/* AF_XDP offloads request. 'request' union member is consumed by the driver
+ * when the packet is being transmitted. 'completion' union member is
+ * filled by the driver when the transmit completion arrives.
+ */
+struct xsk_tx_metadata {
+	union {
+		struct {
+			__u32 flags;
+
+			/* XDP_TXMD_FLAGS_CHECKSUM */
+
+			/* Offset from desc->addr where checksumming should start. */
+			__u16 csum_start;
+			/* Offset from csum_start where checksum should be stored. */
+			__u16 csum_offset;
+		} request;
+
+		struct {
+			/* XDP_TXMD_FLAGS_TIMESTAMP */
+			__u64 tx_timestamp;
+		} completion;
+	};
+};
+
 /* Rx/Tx descriptor */
 struct xdp_desc {
 	__u64 addr;
@@ -122,4 +157,7 @@ struct xdp_desc {
  */
 #define XDP_PKT_CONTD (1 << 0)
 
+/* TX packet carries valid metadata. */
+#define XDP_TX_METADATA (1 << 1)
+
 #endif /* _LINUX_IF_XDP_H */
diff --git a/include/uapi/linux/netdev.h b/include/uapi/linux/netdev.h
index 2943a151d4f1..48d5477a668c 100644
--- a/include/uapi/linux/netdev.h
+++ b/include/uapi/linux/netdev.h
@@ -53,12 +53,28 @@ enum netdev_xdp_rx_metadata {
 	NETDEV_XDP_RX_METADATA_MASK = 3,
 };
 
+/**
+ * enum netdev_xsk_flags
+ * @NETDEV_XSK_FLAGS_TX_TIMESTAMP: HW timestamping egress packets is supported
+ *   by the driver.
+ * @NETDEV_XSK_FLAGS_TX_CHECKSUM: L3 checksum HW offload is supported by the
+ *   driver.
+ */
+enum netdev_xsk_flags {
+	NETDEV_XSK_FLAGS_TX_TIMESTAMP = 1,
+	NETDEV_XSK_FLAGS_TX_CHECKSUM = 2,
+
+	/* private: */
+	NETDEV_XSK_FLAGS_MASK = 3,
+};
+
 enum {
 	NETDEV_A_DEV_IFINDEX = 1,
 	NETDEV_A_DEV_PAD,
 	NETDEV_A_DEV_XDP_FEATURES,
 	NETDEV_A_DEV_XDP_ZC_MAX_SEGS,
 	NETDEV_A_DEV_XDP_RX_METADATA_FEATURES,
+	NETDEV_A_DEV_XSK_FEATURES,
 
 	__NETDEV_A_DEV_MAX,
 	NETDEV_A_DEV_MAX = (__NETDEV_A_DEV_MAX - 1)
diff --git a/net/core/netdev-genl.c b/net/core/netdev-genl.c
index fe61f85bcf33..10f2124e9e23 100644
--- a/net/core/netdev-genl.c
+++ b/net/core/netdev-genl.c
@@ -6,6 +6,7 @@
 #include <net/net_namespace.h>
 #include <net/sock.h>
 #include <net/xdp.h>
+#include <net/xdp_sock.h>
 
 #include "netdev-genl-gen.h"
 
@@ -13,6 +14,7 @@ static int
 netdev_nl_dev_fill(struct net_device *netdev, struct sk_buff *rsp,
 		   const struct genl_info *info)
 {
+	u64 xsk_features = 0;
 	u64 xdp_rx_meta = 0;
 	void *hdr;
 
@@ -26,11 +28,20 @@ netdev_nl_dev_fill(struct net_device *netdev, struct sk_buff *rsp,
 XDP_METADATA_KFUNC_xxx
 #undef XDP_METADATA_KFUNC
 
+	if (netdev->xsk_tx_metadata_ops) {
+		if (netdev->xsk_tx_metadata_ops->tmo_fill_timestamp)
+			xsk_features |= NETDEV_XSK_FLAGS_TX_TIMESTAMP;
+		if (netdev->xsk_tx_metadata_ops->tmo_request_checksum)
+			xsk_features |= NETDEV_XSK_FLAGS_TX_CHECKSUM;
+	}
+
 	if (nla_put_u32(rsp, NETDEV_A_DEV_IFINDEX, netdev->ifindex) ||
 	    nla_put_u64_64bit(rsp, NETDEV_A_DEV_XDP_FEATURES,
 			      netdev->xdp_features, NETDEV_A_DEV_PAD) ||
 	    nla_put_u64_64bit(rsp, NETDEV_A_DEV_XDP_RX_METADATA_FEATURES,
-			      xdp_rx_meta, NETDEV_A_DEV_PAD)) {
+			      xdp_rx_meta, NETDEV_A_DEV_PAD) ||
+	    nla_put_u64_64bit(rsp, NETDEV_A_DEV_XSK_FEATURES,
+			      xsk_features, NETDEV_A_DEV_PAD)) {
 		genlmsg_cancel(rsp, hdr);
 		return -EINVAL;
 	}
diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c
index c904356e2800..84fd10201f2a 100644
--- a/net/xdp/xsk.c
+++ b/net/xdp/xsk.c
@@ -571,6 +571,13 @@ static u32 xsk_get_num_desc(struct sk_buff *skb)
 
 static void xsk_destruct_skb(struct sk_buff *skb)
 {
+	struct xsk_tx_metadata_compl *compl = &skb_shinfo(skb)->xsk_meta;
+
+	if (compl->tx_timestamp) {
+		/* sw completion timestamp, not a real one */
+		*compl->tx_timestamp = ktime_get_tai_fast_ns();
+	}
+
 	xsk_cq_submit_locked(xdp_sk(skb->sk), xsk_get_num_desc(skb));
 	sock_wfree(skb);
 }
@@ -655,8 +662,10 @@ static struct sk_buff *xsk_build_skb_zerocopy(struct xdp_sock *xs,
 static struct sk_buff *xsk_build_skb(struct xdp_sock *xs,
 				     struct xdp_desc *desc)
 {
+	struct xsk_tx_metadata *meta = NULL;
 	struct net_device *dev = xs->dev;
 	struct sk_buff *skb = xs->skb;
+	bool first_frag = false;
 	int err;
 
 	if (dev->priv_flags & IFF_TX_SKB_NO_LINEAR) {
@@ -687,6 +696,8 @@ static struct sk_buff *xsk_build_skb(struct xdp_sock *xs,
 				kfree_skb(skb);
 				goto free_err;
 			}
+
+			first_frag = true;
 		} else {
 			int nr_frags = skb_shinfo(skb)->nr_frags;
 			struct page *page;
@@ -709,12 +720,35 @@ static struct sk_buff *xsk_build_skb(struct xdp_sock *xs,
 
 			skb_add_rx_frag(skb, nr_frags, page, 0, len, 0);
 		}
+
+		if (first_frag && desc->options & XDP_TX_METADATA) {
+			if (unlikely(xs->pool->tx_metadata_len == 0)) {
+				err = -EINVAL;
+				goto free_err;
+			}
+
+			meta = buffer - xs->pool->tx_metadata_len;
+
+			if (meta->request.flags & XDP_TXMD_FLAGS_CHECKSUM) {
+				if (unlikely(meta->request.csum_start +
+					     meta->request.csum_offset +
+					     sizeof(__sum16) > len)) {
+					err = -EINVAL;
+					goto free_err;
+				}
+
+				skb->csum_start = hr + meta->request.csum_start;
+				skb->csum_offset = meta->request.csum_offset;
+				skb->ip_summed = CHECKSUM_PARTIAL;
+			}
+		}
 	}
 
 	skb->dev = dev;
 	skb->priority = READ_ONCE(xs->sk.sk_priority);
 	skb->mark = READ_ONCE(xs->sk.sk_mark);
 	skb->destructor = xsk_destruct_skb;
+	xsk_tx_metadata_to_compl(meta, &skb_shinfo(skb)->xsk_meta);
 	xsk_set_destructor_arg(skb);
 
 	return skb;
diff --git a/net/xdp/xsk_queue.h b/net/xdp/xsk_queue.h
index c74a1372bcb9..6f2d1621c992 100644
--- a/net/xdp/xsk_queue.h
+++ b/net/xdp/xsk_queue.h
@@ -137,7 +137,7 @@ static inline bool xskq_cons_read_addr_unchecked(struct xsk_queue *q, u64 *addr)
 
 static inline bool xp_unused_options_set(u32 options)
 {
-	return options & ~XDP_PKT_CONTD;
+	return options & ~(XDP_PKT_CONTD | XDP_TX_METADATA);
 }
 
 static inline bool xp_aligned_validate_desc(struct xsk_buff_pool *pool,
diff --git a/tools/include/uapi/linux/if_xdp.h b/tools/include/uapi/linux/if_xdp.h
index 34411a2e5b6c..29942c2c32dc 100644
--- a/tools/include/uapi/linux/if_xdp.h
+++ b/tools/include/uapi/linux/if_xdp.h
@@ -26,11 +26,11 @@
  */
 #define XDP_USE_NEED_WAKEUP (1 << 3)
 /* By setting this option, userspace application indicates that it can
- * handle multiple descriptors per packet thus enabling xsk core to split
+ * handle multiple descriptors per packet thus enabling AF_XDP to split
  * multi-buffer XDP frames into multiple Rx descriptors. Without this set
- * such frames will be dropped by xsk.
+ * such frames will be dropped.
  */
-#define XDP_USE_SG     (1 << 4)
+#define XDP_USE_SG	(1 << 4)
 
 /* Flags for xsk_umem_config flags */
 #define XDP_UMEM_UNALIGNED_CHUNK_FLAG (1 << 0)
@@ -106,6 +106,41 @@ struct xdp_options {
 #define XSK_UNALIGNED_BUF_ADDR_MASK \
 	((1ULL << XSK_UNALIGNED_BUF_OFFSET_SHIFT) - 1)
 
+/* Request transmit timestamp. Upon completion, put it into tx_timestamp
+ * field of union xsk_tx_metadata.
+ */
+#define XDP_TXMD_FLAGS_TIMESTAMP		(1 << 0)
+
+/* Request transmit checksum offload. Checksum start position and offset
+ * are communicated via csum_start and csum_offset fields of union
+ * xsk_tx_metadata.
+ */
+#define XDP_TXMD_FLAGS_CHECKSUM			(1 << 1)
+
+/* AF_XDP offloads request. 'request' union member is consumed by the driver
+ * when the packet is being transmitted. 'completion' union member is
+ * filled by the driver when the transmit completion arrives.
+ */
+struct xsk_tx_metadata {
+	union {
+		struct {
+			__u32 flags;
+
+			/* XDP_TXMD_FLAGS_CHECKSUM */
+
+			/* Offset from desc->addr where checksumming should start. */
+			__u16 csum_start;
+			/* Offset from csum_start where checksum should be stored. */
+			__u16 csum_offset;
+		} request;
+
+		struct {
+			/* XDP_TXMD_FLAGS_TIMESTAMP */
+			__u64 tx_timestamp;
+		} completion;
+	};
+};
+
 /* Rx/Tx descriptor */
 struct xdp_desc {
 	__u64 addr;
@@ -113,9 +148,16 @@ struct xdp_desc {
 	__u32 options;
 };
 
-/* Flag indicating packet constitutes of multiple buffers*/
+/* UMEM descriptor is __u64 */
+
+/* Flag indicating that the packet continues with the buffer pointed out by the
+ * next frame in the ring. The end of the packet is signalled by setting this
+ * bit to zero. For single buffer packets, every descriptor has 'options' set
+ * to 0 and this maintains backward compatibility.
+ */
 #define XDP_PKT_CONTD (1 << 0)
 
-/* UMEM descriptor is __u64 */
+/* TX packet carries valid metadata. */
+#define XDP_TX_METADATA (1 << 1)
 
 #endif /* _LINUX_IF_XDP_H */
diff --git a/tools/include/uapi/linux/netdev.h b/tools/include/uapi/linux/netdev.h
index 2943a151d4f1..48d5477a668c 100644
--- a/tools/include/uapi/linux/netdev.h
+++ b/tools/include/uapi/linux/netdev.h
@@ -53,12 +53,28 @@ enum netdev_xdp_rx_metadata {
 	NETDEV_XDP_RX_METADATA_MASK = 3,
 };
 
+/**
+ * enum netdev_xsk_flags
+ * @NETDEV_XSK_FLAGS_TX_TIMESTAMP: HW timestamping egress packets is supported
+ *   by the driver.
+ * @NETDEV_XSK_FLAGS_TX_CHECKSUM: L3 checksum HW offload is supported by the
+ *   driver.
+ */
+enum netdev_xsk_flags {
+	NETDEV_XSK_FLAGS_TX_TIMESTAMP = 1,
+	NETDEV_XSK_FLAGS_TX_CHECKSUM = 2,
+
+	/* private: */
+	NETDEV_XSK_FLAGS_MASK = 3,
+};
+
 enum {
 	NETDEV_A_DEV_IFINDEX = 1,
 	NETDEV_A_DEV_PAD,
 	NETDEV_A_DEV_XDP_FEATURES,
 	NETDEV_A_DEV_XDP_ZC_MAX_SEGS,
 	NETDEV_A_DEV_XDP_RX_METADATA_FEATURES,
+	NETDEV_A_DEV_XSK_FEATURES,
 
 	__NETDEV_A_DEV_MAX,
 	NETDEV_A_DEV_MAX = (__NETDEV_A_DEV_MAX - 1)
diff --git a/tools/net/ynl/generated/netdev-user.c b/tools/net/ynl/generated/netdev-user.c
index b5ffe8cd1144..6283d87dad37 100644
--- a/tools/net/ynl/generated/netdev-user.c
+++ b/tools/net/ynl/generated/netdev-user.c
@@ -58,6 +58,19 @@ const char *netdev_xdp_rx_metadata_str(enum netdev_xdp_rx_metadata value)
 	return netdev_xdp_rx_metadata_strmap[value];
 }
 
+static const char * const netdev_xsk_flags_strmap[] = {
+	[0] = "tx-timestamp",
+	[1] = "tx-checksum",
+};
+
+const char *netdev_xsk_flags_str(enum netdev_xsk_flags value)
+{
+	value = ffs(value) - 1;
+	if (value < 0 || value >= (int)MNL_ARRAY_SIZE(netdev_xsk_flags_strmap))
+		return NULL;
+	return netdev_xsk_flags_strmap[value];
+}
+
 /* Policies */
 struct ynl_policy_attr netdev_dev_policy[NETDEV_A_DEV_MAX + 1] = {
 	[NETDEV_A_DEV_IFINDEX] = { .name = "ifindex", .type = YNL_PT_U32, },
@@ -65,6 +78,7 @@ struct ynl_policy_attr netdev_dev_policy[NETDEV_A_DEV_MAX + 1] = {
 	[NETDEV_A_DEV_XDP_FEATURES] = { .name = "xdp-features", .type = YNL_PT_U64, },
 	[NETDEV_A_DEV_XDP_ZC_MAX_SEGS] = { .name = "xdp-zc-max-segs", .type = YNL_PT_U32, },
 	[NETDEV_A_DEV_XDP_RX_METADATA_FEATURES] = { .name = "xdp-rx-metadata-features", .type = YNL_PT_U64, },
+	[NETDEV_A_DEV_XSK_FEATURES] = { .name = "xsk-features", .type = YNL_PT_U64, },
 };
 
 struct ynl_policy_nest netdev_dev_nest = {
@@ -116,6 +130,11 @@ int netdev_dev_get_rsp_parse(const struct nlmsghdr *nlh, void *data)
 				return MNL_CB_ERROR;
 			dst->_present.xdp_rx_metadata_features = 1;
 			dst->xdp_rx_metadata_features = mnl_attr_get_u64(attr);
+		} else if (type == NETDEV_A_DEV_XSK_FEATURES) {
+			if (ynl_attr_validate(yarg, attr))
+				return MNL_CB_ERROR;
+			dst->_present.xsk_features = 1;
+			dst->xsk_features = mnl_attr_get_u64(attr);
 		}
 	}
 
diff --git a/tools/net/ynl/generated/netdev-user.h b/tools/net/ynl/generated/netdev-user.h
index 4fafac879df3..39af1908444b 100644
--- a/tools/net/ynl/generated/netdev-user.h
+++ b/tools/net/ynl/generated/netdev-user.h
@@ -19,6 +19,7 @@ extern const struct ynl_family ynl_netdev_family;
 const char *netdev_op_str(int op);
 const char *netdev_xdp_act_str(enum netdev_xdp_act value);
 const char *netdev_xdp_rx_metadata_str(enum netdev_xdp_rx_metadata value);
+const char *netdev_xsk_flags_str(enum netdev_xsk_flags value);
 
 /* Common nested types */
 /* ============== NETDEV_CMD_DEV_GET ============== */
@@ -50,12 +51,14 @@ struct netdev_dev_get_rsp {
 		__u32 xdp_features:1;
 		__u32 xdp_zc_max_segs:1;
 		__u32 xdp_rx_metadata_features:1;
+		__u32 xsk_features:1;
 	} _present;
 
 	__u32 ifindex;
 	__u64 xdp_features;
 	__u32 xdp_zc_max_segs;
 	__u64 xdp_rx_metadata_features;
+	__u64 xsk_features;
 };
 
 void netdev_dev_get_rsp_free(struct netdev_dev_get_rsp *rsp);
-- 
2.42.0.869.gea05f2083d-goog

[xdp-hints] [PATCH bpf-next v5 03/13] tools: ynl: Print xsk-features from the sample

[Stanislav Fomichev]

In a similar fashion we do for the other bit masks.
Fix mask parsing (>= vs >) while we are it.

Signed-off-by: Stanislav Fomichev <sdf@google.com>
---
 tools/net/ynl/samples/netdev.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/tools/net/ynl/samples/netdev.c b/tools/net/ynl/samples/netdev.c
index b828225daad0..591b90e21890 100644
--- a/tools/net/ynl/samples/netdev.c
+++ b/tools/net/ynl/samples/netdev.c
@@ -33,17 +33,23 @@ static void netdev_print_device(struct netdev_dev_get_rsp *d, unsigned int op)
 		return;
 
 	printf("xdp-features (%llx):", d->xdp_features);
-	for (int i = 0; d->xdp_features > 1U << i; i++) {
+	for (int i = 0; d->xdp_features >= 1U << i; i++) {
 		if (d->xdp_features & (1U << i))
 			printf(" %s", netdev_xdp_act_str(1 << i));
 	}
 
 	printf(" xdp-rx-metadata-features (%llx):", d->xdp_rx_metadata_features);
-	for (int i = 0; d->xdp_rx_metadata_features > 1U << i; i++) {
+	for (int i = 0; d->xdp_rx_metadata_features >= 1U << i; i++) {
 		if (d->xdp_rx_metadata_features & (1U << i))
 			printf(" %s", netdev_xdp_rx_metadata_str(1 << i));
 	}
 
+	printf(" xsk-features (%llx):", d->xsk_features);
+	for (int i = 0; d->xsk_features >= 1U << i; i++) {
+		if (d->xsk_features & (1U << i))
+			printf(" %s", netdev_xsk_flags_str(1 << i));
+	}
+
 	printf(" xdp-zc-max-segs=%u", d->xdp_zc_max_segs);
 
 	name = netdev_op_str(op);
-- 
2.42.0.869.gea05f2083d-goog

[xdp-hints] [PATCH bpf-next v5 04/13] net/mlx5e: Implement AF_XDP TX timestamp and checksum offload

[Stanislav Fomichev]

TX timestamp:
- requires passing clock, not sure I'm passing the correct one (from
  cq->mdev), but the timestamp value looks convincing

TX checksum:
- looks like device does packet parsing (and doesn't accept custom
  start/offset), so I'm ignoring user offsets

Cc: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Stanislav Fomichev <sdf@google.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en.h  |  4 +-
 .../net/ethernet/mellanox/mlx5/core/en/xdp.c  | 72 ++++++++++++++++---
 .../net/ethernet/mellanox/mlx5/core/en/xdp.h  | 11 ++-
 .../ethernet/mellanox/mlx5/core/en/xsk/tx.c   | 17 ++++-
 .../net/ethernet/mellanox/mlx5/core/en_main.c |  1 +
 5 files changed, 89 insertions(+), 16 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index b2a5da9739d2..43f027bf2da3 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -484,10 +484,12 @@ struct mlx5e_xdp_info_fifo {
 
 struct mlx5e_xdpsq;
 struct mlx5e_xmit_data;
+struct xsk_tx_metadata;
 typedef int (*mlx5e_fp_xmit_xdp_frame_check)(struct mlx5e_xdpsq *);
 typedef bool (*mlx5e_fp_xmit_xdp_frame)(struct mlx5e_xdpsq *,
 					struct mlx5e_xmit_data *,
-					int);
+					int,
+					struct xsk_tx_metadata *);
 
 struct mlx5e_xdpsq {
 	/* data path */
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c b/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c
index 7decc81ed33a..e2e7d82cfca4 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.c
@@ -103,7 +103,7 @@ mlx5e_xmit_xdp_buff(struct mlx5e_xdpsq *sq, struct mlx5e_rq *rq,
 		xdptxd->dma_addr = dma_addr;
 
 		if (unlikely(!INDIRECT_CALL_2(sq->xmit_xdp_frame, mlx5e_xmit_xdp_frame_mpwqe,
-					      mlx5e_xmit_xdp_frame, sq, xdptxd, 0)))
+					      mlx5e_xmit_xdp_frame, sq, xdptxd, 0, NULL)))
 			return false;
 
 		/* xmit_mode == MLX5E_XDP_XMIT_MODE_FRAME */
@@ -145,7 +145,7 @@ mlx5e_xmit_xdp_buff(struct mlx5e_xdpsq *sq, struct mlx5e_rq *rq,
 	xdptxd->dma_addr = dma_addr;
 
 	if (unlikely(!INDIRECT_CALL_2(sq->xmit_xdp_frame, mlx5e_xmit_xdp_frame_mpwqe,
-				      mlx5e_xmit_xdp_frame, sq, xdptxd, 0)))
+				      mlx5e_xmit_xdp_frame, sq, xdptxd, 0, NULL)))
 		return false;
 
 	/* xmit_mode == MLX5E_XDP_XMIT_MODE_PAGE */
@@ -261,6 +261,37 @@ const struct xdp_metadata_ops mlx5e_xdp_metadata_ops = {
 	.xmo_rx_hash			= mlx5e_xdp_rx_hash,
 };
 
+struct mlx5e_xsk_tx_complete {
+	struct mlx5_cqe64 *cqe;
+	struct mlx5e_cq *cq;
+};
+
+static u64 mlx5e_xsk_fill_timestamp(void *_priv)
+{
+	struct mlx5e_xsk_tx_complete *priv = _priv;
+	u64 ts;
+
+	ts = get_cqe_ts(priv->cqe);
+
+	if (mlx5_is_real_time_rq(priv->cq->mdev) || mlx5_is_real_time_sq(priv->cq->mdev))
+		return mlx5_real_time_cyc2time(&priv->cq->mdev->clock, ts);
+
+	return  mlx5_timecounter_cyc2time(&priv->cq->mdev->clock, ts);
+}
+
+static void mlx5e_xsk_request_checksum(u16 csum_start, u16 csum_offset, void *priv)
+{
+	struct mlx5_wqe_eth_seg *eseg = priv;
+
+	/* HW/FW is doing parsing, so offsets are largely ignored. */
+	eseg->cs_flags |= MLX5_ETH_WQE_L3_CSUM | MLX5_ETH_WQE_L4_CSUM;
+}
+
+const struct xsk_tx_metadata_ops mlx5e_xsk_tx_metadata_ops = {
+	.tmo_fill_timestamp		= mlx5e_xsk_fill_timestamp,
+	.tmo_request_checksum		= mlx5e_xsk_request_checksum,
+};
+
 /* returns true if packet was consumed by xdp */
 bool mlx5e_xdp_handle(struct mlx5e_rq *rq,
 		      struct bpf_prog *prog, struct mlx5e_xdp_buff *mxbuf)
@@ -398,11 +429,11 @@ INDIRECT_CALLABLE_SCOPE int mlx5e_xmit_xdp_frame_check_mpwqe(struct mlx5e_xdpsq
 
 INDIRECT_CALLABLE_SCOPE bool
 mlx5e_xmit_xdp_frame(struct mlx5e_xdpsq *sq, struct mlx5e_xmit_data *xdptxd,
-		     int check_result);
+		     int check_result, struct xsk_tx_metadata *meta);
 
 INDIRECT_CALLABLE_SCOPE bool
 mlx5e_xmit_xdp_frame_mpwqe(struct mlx5e_xdpsq *sq, struct mlx5e_xmit_data *xdptxd,
-			   int check_result)
+			   int check_result, struct xsk_tx_metadata *meta)
 {
 	struct mlx5e_tx_mpwqe *session = &sq->mpwqe;
 	struct mlx5e_xdpsq_stats *stats = sq->stats;
@@ -420,7 +451,7 @@ mlx5e_xmit_xdp_frame_mpwqe(struct mlx5e_xdpsq *sq, struct mlx5e_xmit_data *xdptx
 			 */
 			if (unlikely(sq->mpwqe.wqe))
 				mlx5e_xdp_mpwqe_complete(sq);
-			return mlx5e_xmit_xdp_frame(sq, xdptxd, 0);
+			return mlx5e_xmit_xdp_frame(sq, xdptxd, 0, meta);
 		}
 		if (!xdptxd->len) {
 			skb_frag_t *frag = &xdptxdf->sinfo->frags[0];
@@ -450,6 +481,7 @@ mlx5e_xmit_xdp_frame_mpwqe(struct mlx5e_xdpsq *sq, struct mlx5e_xmit_data *xdptx
 		 * and it's safe to complete it at any time.
 		 */
 		mlx5e_xdp_mpwqe_session_start(sq);
+		xsk_tx_metadata_request(meta, &mlx5e_xsk_tx_metadata_ops, &session->wqe->eth);
 	}
 
 	mlx5e_xdp_mpwqe_add_dseg(sq, p, stats);
@@ -480,7 +512,7 @@ INDIRECT_CALLABLE_SCOPE int mlx5e_xmit_xdp_frame_check(struct mlx5e_xdpsq *sq)
 
 INDIRECT_CALLABLE_SCOPE bool
 mlx5e_xmit_xdp_frame(struct mlx5e_xdpsq *sq, struct mlx5e_xmit_data *xdptxd,
-		     int check_result)
+		     int check_result, struct xsk_tx_metadata *meta)
 {
 	struct mlx5e_xmit_data_frags *xdptxdf =
 		container_of(xdptxd, struct mlx5e_xmit_data_frags, xd);
@@ -599,6 +631,8 @@ mlx5e_xmit_xdp_frame(struct mlx5e_xdpsq *sq, struct mlx5e_xmit_data *xdptxd,
 		sq->pc++;
 	}
 
+	xsk_tx_metadata_request(meta, &mlx5e_xsk_tx_metadata_ops, eseg);
+
 	sq->doorbell_cseg = cseg;
 
 	stats->xmit++;
@@ -608,7 +642,9 @@ mlx5e_xmit_xdp_frame(struct mlx5e_xdpsq *sq, struct mlx5e_xmit_data *xdptxd,
 static void mlx5e_free_xdpsq_desc(struct mlx5e_xdpsq *sq,
 				  struct mlx5e_xdp_wqe_info *wi,
 				  u32 *xsk_frames,
-				  struct xdp_frame_bulk *bq)
+				  struct xdp_frame_bulk *bq,
+				  struct mlx5e_cq *cq,
+				  struct mlx5_cqe64 *cqe)
 {
 	struct mlx5e_xdp_info_fifo *xdpi_fifo = &sq->db.xdpi_fifo;
 	u16 i;
@@ -668,10 +704,24 @@ static void mlx5e_free_xdpsq_desc(struct mlx5e_xdpsq *sq,
 
 			break;
 		}
-		case MLX5E_XDP_XMIT_MODE_XSK:
+		case MLX5E_XDP_XMIT_MODE_XSK: {
 			/* AF_XDP send */
+			struct xsk_tx_metadata_compl *compl = NULL;
+			struct mlx5e_xsk_tx_complete priv = {
+				.cqe = cqe,
+				.cq = cq,
+			};
+
+			if (xp_tx_metadata_enabled(sq->xsk_pool)) {
+				xdpi = mlx5e_xdpi_fifo_pop(xdpi_fifo);
+				compl = &xdpi.xsk_meta;
+
+				xsk_tx_metadata_complete(compl, &mlx5e_xsk_tx_metadata_ops, &priv);
+			}
+
 			(*xsk_frames)++;
 			break;
+		}
 		default:
 			WARN_ON_ONCE(true);
 		}
@@ -720,7 +770,7 @@ bool mlx5e_poll_xdpsq_cq(struct mlx5e_cq *cq)
 
 			sqcc += wi->num_wqebbs;
 
-			mlx5e_free_xdpsq_desc(sq, wi, &xsk_frames, &bq);
+			mlx5e_free_xdpsq_desc(sq, wi, &xsk_frames, &bq, cq, cqe);
 		} while (!last_wqe);
 
 		if (unlikely(get_cqe_opcode(cqe) != MLX5_CQE_REQ)) {
@@ -767,7 +817,7 @@ void mlx5e_free_xdpsq_descs(struct mlx5e_xdpsq *sq)
 
 		sq->cc += wi->num_wqebbs;
 
-		mlx5e_free_xdpsq_desc(sq, wi, &xsk_frames, &bq);
+		mlx5e_free_xdpsq_desc(sq, wi, &xsk_frames, &bq, NULL, NULL);
 	}
 
 	xdp_flush_frame_bulk(&bq);
@@ -840,7 +890,7 @@ int mlx5e_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **frames,
 		}
 
 		ret = INDIRECT_CALL_2(sq->xmit_xdp_frame, mlx5e_xmit_xdp_frame_mpwqe,
-				      mlx5e_xmit_xdp_frame, sq, xdptxd, 0);
+				      mlx5e_xmit_xdp_frame, sq, xdptxd, 0, NULL);
 		if (unlikely(!ret)) {
 			int j;
 
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.h b/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.h
index ecfe93a479da..e054db1e10f8 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/xdp.h
@@ -33,6 +33,7 @@
 #define __MLX5_EN_XDP_H__
 
 #include <linux/indirect_call_wrapper.h>
+#include <net/xdp_sock.h>
 
 #include "en.h"
 #include "en/txrx.h"
@@ -82,7 +83,7 @@ enum mlx5e_xdp_xmit_mode {
  *    num, page_1, page_2, ... , page_num.
  *
  * MLX5E_XDP_XMIT_MODE_XSK:
- *    none.
+ *    frame.xsk_meta.
  */
 #define MLX5E_XDP_FIFO_ENTRIES2DS_MAX_RATIO 4
 
@@ -97,6 +98,7 @@ union mlx5e_xdp_info {
 		u8 num;
 		struct page *page;
 	} page;
+	struct xsk_tx_metadata_compl xsk_meta;
 };
 
 struct mlx5e_xsk_param;
@@ -112,13 +114,16 @@ int mlx5e_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **frames,
 		   u32 flags);
 
 extern const struct xdp_metadata_ops mlx5e_xdp_metadata_ops;
+extern const struct xsk_tx_metadata_ops mlx5e_xsk_tx_metadata_ops;
 
 INDIRECT_CALLABLE_DECLARE(bool mlx5e_xmit_xdp_frame_mpwqe(struct mlx5e_xdpsq *sq,
 							  struct mlx5e_xmit_data *xdptxd,
-							  int check_result));
+							  int check_result,
+							  struct xsk_tx_metadata *meta));
 INDIRECT_CALLABLE_DECLARE(bool mlx5e_xmit_xdp_frame(struct mlx5e_xdpsq *sq,
 						    struct mlx5e_xmit_data *xdptxd,
-						    int check_result));
+						    int check_result,
+						    struct xsk_tx_metadata *meta));
 INDIRECT_CALLABLE_DECLARE(int mlx5e_xmit_xdp_frame_check_mpwqe(struct mlx5e_xdpsq *sq));
 INDIRECT_CALLABLE_DECLARE(int mlx5e_xmit_xdp_frame_check(struct mlx5e_xdpsq *sq));
 
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/xsk/tx.c b/drivers/net/ethernet/mellanox/mlx5/core/en/xsk/tx.c
index 597f319d4770..a59199ed590d 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/xsk/tx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/xsk/tx.c
@@ -55,12 +55,16 @@ static void mlx5e_xsk_tx_post_err(struct mlx5e_xdpsq *sq,
 
 	nopwqe = mlx5e_post_nop(&sq->wq, sq->sqn, &sq->pc);
 	mlx5e_xdpi_fifo_push(&sq->db.xdpi_fifo, *xdpi);
+	if (xp_tx_metadata_enabled(sq->xsk_pool))
+		mlx5e_xdpi_fifo_push(&sq->db.xdpi_fifo,
+				     (union mlx5e_xdp_info) { .xsk_meta = {} });
 	sq->doorbell_cseg = &nopwqe->ctrl;
 }
 
 bool mlx5e_xsk_tx(struct mlx5e_xdpsq *sq, unsigned int budget)
 {
 	struct xsk_buff_pool *pool = sq->xsk_pool;
+	struct xsk_tx_metadata *meta = NULL;
 	union mlx5e_xdp_info xdpi;
 	bool work_done = true;
 	bool flush = false;
@@ -93,12 +97,13 @@ bool mlx5e_xsk_tx(struct mlx5e_xdpsq *sq, unsigned int budget)
 		xdptxd.dma_addr = xsk_buff_raw_get_dma(pool, desc.addr);
 		xdptxd.data = xsk_buff_raw_get_data(pool, desc.addr);
 		xdptxd.len = desc.len;
+		meta = xsk_buff_get_metadata(pool, desc.addr);
 
 		xsk_buff_raw_dma_sync_for_device(pool, xdptxd.dma_addr, xdptxd.len);
 
 		ret = INDIRECT_CALL_2(sq->xmit_xdp_frame, mlx5e_xmit_xdp_frame_mpwqe,
 				      mlx5e_xmit_xdp_frame, sq, &xdptxd,
-				      check_result);
+				      check_result, meta);
 		if (unlikely(!ret)) {
 			if (sq->mpwqe.wqe)
 				mlx5e_xdp_mpwqe_complete(sq);
@@ -106,6 +111,16 @@ bool mlx5e_xsk_tx(struct mlx5e_xdpsq *sq, unsigned int budget)
 			mlx5e_xsk_tx_post_err(sq, &xdpi);
 		} else {
 			mlx5e_xdpi_fifo_push(&sq->db.xdpi_fifo, xdpi);
+			if (xp_tx_metadata_enabled(sq->xsk_pool)) {
+				struct xsk_tx_metadata_compl compl;
+
+				xsk_tx_metadata_to_compl(meta, &compl);
+				XSK_TX_COMPL_FITS(void *);
+
+				mlx5e_xdpi_fifo_push(&sq->db.xdpi_fifo,
+						     (union mlx5e_xdp_info)
+						     { .xsk_meta = compl });
+			}
 		}
 
 		flush = true;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index ea58c6917433..9da668f2b10b 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -5164,6 +5164,7 @@ static void mlx5e_build_nic_netdev(struct net_device *netdev)
 
 	netdev->netdev_ops = &mlx5e_netdev_ops;
 	netdev->xdp_metadata_ops = &mlx5e_xdp_metadata_ops;
+	netdev->xsk_tx_metadata_ops = &mlx5e_xsk_tx_metadata_ops;
 
 	mlx5e_dcbnl_build_netdev(netdev);
 
-- 
2.42.0.869.gea05f2083d-goog

[xdp-hints] [PATCH bpf-next v5 05/13] net: stmmac: Add Tx HWTS support to XDP ZC

[Stanislav Fomichev]

From: Song Yoong Siang <yoong.siang.song@intel.com>

This patch enables transmit hardware timestamp support to XDP zero copy
via XDP Tx metadata framework.

This patchset is tested with tools/testing/selftests/bpf/xdp_hw_metadata
on Intel Tiger Lake platform. Below are the test steps and results.

Command on DUT:
sudo ./xdp_hw_metadata <interface name>
sudo hwstamp_ctl -i <interface name> -t 1 -r 1

Command on Link Partner:
echo -n xdp | nc -u -q1 <destination IPv4 addr> 9091

Result:
xsk_ring_cons__peek: 1
0x55bbbf08b6d0: rx_desc[2]->addr=8c100 addr=8c100 comp_addr=8c100 EoP
No rx_hash err=-95
rx_timestamp:  1677762688429141540 (sec:1677762688.4291)
HW RX-time:   1677762688429141540 (sec:1677762688.4291) delta to User RX-time sec:0.0003 (250.665 usec)
XDP RX-time:   1677762688429375597 (sec:1677762688.4294) delta to User RX-time sec:0.0000 (16.608 usec)
0x55bbbf08b6d0: ping-pong with csum=561c (want f488) csum_start=34 csum_offset=6
0x55bbbf08b6d0: complete tx idx=2 addr=2008
tx_timestamp:  1677762688431127273 (sec:1677762688.4311)
HW TX-complete-time:   1677762688431127273 (sec:1677762688.4311) delta to User TX-complete-time sec:0.0083 (8331.655 usec)
XDP RX-time:   1677762688429375597 (sec:1677762688.4294) delta to User TX-complete-time sec:0.0101 (10083.331 usec)
HW RX-time:   1677762688429141540 (sec:1677762688.4291) delta to HW TX-complete-time sec:0.0020 (1985.733 usec)
0x55bbbf08b6d0: complete rx idx=130 addr=8c100

Signed-off-by: Song Yoong Siang <yoong.siang.song@intel.com>
Signed-off-by: Stanislav Fomichev <sdf@google.com>
---
 drivers/net/ethernet/stmicro/stmmac/stmmac.h  | 12 ++++
 .../net/ethernet/stmicro/stmmac/stmmac_main.c | 64 ++++++++++++++++++-
 2 files changed, 75 insertions(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac.h b/drivers/net/ethernet/stmicro/stmmac/stmmac.h
index cd7a9768de5f..686c94c2e8a7 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac.h
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac.h
@@ -51,6 +51,7 @@ struct stmmac_tx_info {
 	bool last_segment;
 	bool is_jumbo;
 	enum stmmac_txbuf_type buf_type;
+	struct xsk_tx_metadata_compl xsk_meta;
 };
 
 #define STMMAC_TBS_AVAIL	BIT(0)
@@ -100,6 +101,17 @@ struct stmmac_xdp_buff {
 	struct dma_desc *ndesc;
 };
 
+struct stmmac_metadata_request {
+	struct stmmac_priv *priv;
+	struct dma_desc *tx_desc;
+	bool *set_ic;
+};
+
+struct stmmac_xsk_tx_complete {
+	struct stmmac_priv *priv;
+	struct dma_desc *desc;
+};
+
 struct stmmac_rx_queue {
 	u32 rx_count_frames;
 	u32 queue_index;
diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
index 3e50fd53a617..001a07a69539 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
@@ -2430,6 +2430,46 @@ static void stmmac_dma_operation_mode(struct stmmac_priv *priv)
 	}
 }
 
+static void stmmac_xsk_request_timestamp(void *_priv)
+{
+	struct stmmac_metadata_request *meta_req = _priv;
+
+	stmmac_enable_tx_timestamp(meta_req->priv, meta_req->tx_desc);
+	*meta_req->set_ic = true;
+}
+
+static u64 stmmac_xsk_fill_timestamp(void *_priv)
+{
+	struct stmmac_xsk_tx_complete *tx_compl = _priv;
+	struct stmmac_priv *priv = tx_compl->priv;
+	struct dma_desc *desc = tx_compl->desc;
+	bool found = false;
+	u64 ns = 0;
+
+	if (!priv->hwts_tx_en)
+		return 0;
+
+	/* check tx tstamp status */
+	if (stmmac_get_tx_timestamp_status(priv, desc)) {
+		stmmac_get_timestamp(priv, desc, priv->adv_ts, &ns);
+		found = true;
+	} else if (!stmmac_get_mac_tx_timestamp(priv, priv->hw, &ns)) {
+		found = true;
+	}
+
+	if (found) {
+		ns -= priv->plat->cdc_error_adj;
+		return ns_to_ktime(ns);
+	}
+
+	return 0;
+}
+
+static const struct xsk_tx_metadata_ops stmmac_xsk_tx_metadata_ops = {
+	.tmo_request_timestamp		= stmmac_xsk_request_timestamp,
+	.tmo_fill_timestamp		= stmmac_xsk_fill_timestamp,
+};
+
 static bool stmmac_xdp_xmit_zc(struct stmmac_priv *priv, u32 queue, u32 budget)
 {
 	struct netdev_queue *nq = netdev_get_tx_queue(priv->dev, queue);
@@ -2449,6 +2489,8 @@ static bool stmmac_xdp_xmit_zc(struct stmmac_priv *priv, u32 queue, u32 budget)
 	budget = min(budget, stmmac_tx_avail(priv, queue));
 
 	while (budget-- > 0) {
+		struct stmmac_metadata_request meta_req;
+		struct xsk_tx_metadata *meta = NULL;
 		dma_addr_t dma_addr;
 		bool set_ic;
 
@@ -2472,6 +2514,7 @@ static bool stmmac_xdp_xmit_zc(struct stmmac_priv *priv, u32 queue, u32 budget)
 			tx_desc = tx_q->dma_tx + entry;
 
 		dma_addr = xsk_buff_raw_get_dma(pool, xdp_desc.addr);
+		meta = xsk_buff_get_metadata(pool, xdp_desc.addr);
 		xsk_buff_raw_dma_sync_for_device(pool, dma_addr, xdp_desc.len);
 
 		tx_q->tx_skbuff_dma[entry].buf_type = STMMAC_TXBUF_T_XSK_TX;
@@ -2499,6 +2542,11 @@ static bool stmmac_xdp_xmit_zc(struct stmmac_priv *priv, u32 queue, u32 budget)
 		else
 			set_ic = false;
 
+		meta_req.priv = priv;
+		meta_req.tx_desc = tx_desc;
+		meta_req.set_ic = &set_ic;
+		xsk_tx_metadata_request(meta, &stmmac_xsk_tx_metadata_ops,
+					&meta_req);
 		if (set_ic) {
 			tx_q->tx_count_frames = 0;
 			stmmac_set_tx_ic(priv, tx_desc);
@@ -2511,6 +2559,9 @@ static bool stmmac_xdp_xmit_zc(struct stmmac_priv *priv, u32 queue, u32 budget)
 
 		stmmac_enable_dma_transmission(priv, priv->ioaddr);
 
+		xsk_tx_metadata_to_compl(meta,
+					 &tx_q->tx_skbuff_dma[entry].xsk_meta);
+
 		tx_q->cur_tx = STMMAC_GET_ENTRY(tx_q->cur_tx, priv->dma_conf.dma_tx_size);
 		entry = tx_q->cur_tx;
 	}
@@ -2620,8 +2671,18 @@ static int stmmac_tx_clean(struct stmmac_priv *priv, int budget, u32 queue,
 			} else {
 				tx_packets++;
 			}
-			if (skb)
+			if (skb) {
 				stmmac_get_tx_hwtstamp(priv, p, skb);
+			} else {
+				struct stmmac_xsk_tx_complete tx_compl = {
+					.priv = priv,
+					.desc = p,
+				};
+
+				xsk_tx_metadata_complete(&tx_q->tx_skbuff_dma[entry].xsk_meta,
+							 &stmmac_xsk_tx_metadata_ops,
+							 &tx_compl);
+			}
 		}
 
 		if (likely(tx_q->tx_skbuff_dma[entry].buf &&
@@ -7449,6 +7510,7 @@ int stmmac_dvr_probe(struct device *device,
 	ndev->netdev_ops = &stmmac_netdev_ops;
 
 	ndev->xdp_metadata_ops = &stmmac_xdp_metadata_ops;
+	ndev->xsk_tx_metadata_ops = &stmmac_xsk_tx_metadata_ops;
 
 	ndev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
 			    NETIF_F_RXCSUM;
-- 
2.42.0.869.gea05f2083d-goog

[xdp-hints] [PATCH bpf-next v5 06/13] xsk: Document tx_metadata_len layout

[Stanislav Fomichev]

- how to use
- how to query features
- pointers to the examples

Signed-off-by: Stanislav Fomichev <sdf@google.com>
---
 Documentation/networking/index.rst           |  1 +
 Documentation/networking/xsk-tx-metadata.rst | 70 ++++++++++++++++++++
 2 files changed, 71 insertions(+)
 create mode 100644 Documentation/networking/xsk-tx-metadata.rst

diff --git a/Documentation/networking/index.rst b/Documentation/networking/index.rst
index 683eb42309cc..a297a894b366 100644
--- a/Documentation/networking/index.rst
+++ b/Documentation/networking/index.rst
@@ -123,6 +123,7 @@ Refer to :ref:`netdev-FAQ` for a guide on netdev development process specifics.
    xfrm_sync
    xfrm_sysctl
    xdp-rx-metadata
+   xsk-tx-metadata
 
 .. only::  subproject and html
 
diff --git a/Documentation/networking/xsk-tx-metadata.rst b/Documentation/networking/xsk-tx-metadata.rst
new file mode 100644
index 000000000000..4f376560b23f
--- /dev/null
+++ b/Documentation/networking/xsk-tx-metadata.rst
@@ -0,0 +1,70 @@
+==================
+AF_XDP TX Metadata
+==================
+
+This document describes how to enable offloads when transmitting packets
+via :doc:`af_xdp`. Refer to :doc:`xdp-rx-metadata` on how to access similar
+metadata on the receive side.
+
+General Design
+==============
+
+The headroom for the metadata is reserved via ``tx_metadata_len`` in
+``struct xdp_umem_reg``. The metadata length is therefore the same for
+every socket that shares the same umem. The metadata layout is a fixed UAPI,
+refer to ``union xsk_tx_metadata`` in ``include/uapi/linux/if_xdp.h``.
+Thus, generally, the ``tx_metadata_len`` field above should contain
+``sizeof(union xsk_tx_metadata)``.
+
+The headroom and the metadata itself should be located right before
+``xdp_desc->addr`` in the umem frame. Within a frame, the metadata
+layout is as follows::
+
+           tx_metadata_len
+     /                         \
+    +-----------------+---------+----------------------------+
+    | xsk_tx_metadata | padding |          payload           |
+    +-----------------+---------+----------------------------+
+                                ^
+                                |
+                          xdp_desc->addr
+
+An AF_XDP application can request headrooms larger than ``sizeof(struct
+xsk_tx_metadata)``. The kernel will ignore the padding (and will still
+use ``xdp_desc->addr - tx_metadata_len`` to locate
+the ``xsk_tx_metadata``). For the frames that shouldn't carry
+any metadata (i.e., the ones that don't have ``XDP_TX_METADATA`` option),
+the metadata area is ignored by the kernel as well.
+
+The flags field enables the particular offload:
+
+- ``XDP_TXMD_FLAGS_TIMESTAMP``: requests the device to put transmission
+  timestamp into ``tx_timestamp`` field of ``union xsk_tx_metadata``.
+- ``XDP_TXMD_FLAGS_CHECKSUM``: requests the device to calculate L4
+  checksum. ``csum_start`` specifies byte offset of where the checksumming
+  should start and ``csum_offset`` specifies byte offset where the
+  device should store the computed checksum.
+
+Besides the flags above, in order to trigger the offloads, the first
+packet's ``struct xdp_desc`` descriptor should set ``XDP_TX_METADATA``
+bit in the ``options`` field. Also note that in a multi-buffer packet
+only the first chunk should carry the metadata.
+
+Querying Device Capabilities
+============================
+
+Every devices exports its offloads capabilities via netlink netdev family.
+Refer to ``xsk-flags`` features bitmask in
+``Documentation/netlink/specs/netdev.yaml``.
+
+- ``tx-timestamp``: device supports ``XDP_TXMD_FLAGS_TIMESTAMP``
+- ``tx-checksum``: device supports ``XDP_TXMD_FLAGS_CHECKSUM``
+
+See ``tools/net/ynl/samples/netdev.c`` on how to query this information.
+
+Example
+=======
+
+See ``tools/testing/selftests/bpf/xdp_hw_metadata.c`` for an example
+program that handles TX metadata. Also see https://github.com/fomichev/xskgen
+for a more bare-bones example.
-- 
2.42.0.869.gea05f2083d-goog

[xdp-hints] [PATCH bpf-next v5 07/13] xsk: Validate xsk_tx_metadata flags

[Stanislav Fomichev]

Accept only the flags that the kernel knows about to make
sure we can extend this field in the future. Note that only
in XDP_COPY mode we propagate the error signal back to the user
(via sendmsg). For zerocopy mode we silently skip the metadata
for the descriptors that have wrong flags (since we process
the descriptors deep in the driver).

Signed-off-by: Stanislav Fomichev <sdf@google.com>
---
 include/net/xdp_sock_drv.h | 23 ++++++++++++++++++++++-
 net/xdp/xsk.c              |  4 ++++
 2 files changed, 26 insertions(+), 1 deletion(-)

diff --git a/include/net/xdp_sock_drv.h b/include/net/xdp_sock_drv.h
index e2558ac3e195..5885176ea01e 100644
--- a/include/net/xdp_sock_drv.h
+++ b/include/net/xdp_sock_drv.h
@@ -165,12 +165,28 @@ static inline void *xsk_buff_raw_get_data(struct xsk_buff_pool *pool, u64 addr)
 	return xp_raw_get_data(pool, addr);
 }
 
+#define XDP_TXMD_FLAGS_VALID ( \
+		XDP_TXMD_FLAGS_TIMESTAMP | \
+		XDP_TXMD_FLAGS_CHECKSUM | \
+	0)
+
+static inline bool xsk_buff_valid_tx_metadata(struct xsk_tx_metadata *meta)
+{
+	return !(meta->request.flags & ~XDP_TXMD_FLAGS_VALID);
+}
+
 static inline struct xsk_tx_metadata *xsk_buff_get_metadata(struct xsk_buff_pool *pool, u64 addr)
 {
+	struct xsk_tx_metadata *meta;
+
 	if (!pool->tx_metadata_len)
 		return NULL;
 
-	return xp_raw_get_data(pool, addr) - pool->tx_metadata_len;
+	meta = xp_raw_get_data(pool, addr) - pool->tx_metadata_len;
+	if (unlikely(!xsk_buff_valid_tx_metadata(meta)))
+		return NULL; /* no way to signal the error to the user */
+
+	return meta;
 }
 
 static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp, struct xsk_buff_pool *pool)
@@ -332,6 +348,11 @@ static inline void *xsk_buff_raw_get_data(struct xsk_buff_pool *pool, u64 addr)
 	return NULL;
 }
 
+static inline bool xsk_buff_valid_tx_metadata(struct xsk_tx_metadata *meta)
+{
+	return false;
+}
+
 static inline struct xsk_tx_metadata *xsk_buff_get_metadata(struct xsk_buff_pool *pool, u64 addr)
 {
 	return NULL;
diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c
index 84fd10201f2a..0e81ae6bfff4 100644
--- a/net/xdp/xsk.c
+++ b/net/xdp/xsk.c
@@ -728,6 +728,10 @@ static struct sk_buff *xsk_build_skb(struct xdp_sock *xs,
 			}
 
 			meta = buffer - xs->pool->tx_metadata_len;
+			if (unlikely(!xsk_buff_valid_tx_metadata(meta))) {
+				err = -EINVAL;
+				goto free_err;
+			}
 
 			if (meta->request.flags & XDP_TXMD_FLAGS_CHECKSUM) {
 				if (unlikely(meta->request.csum_start +
-- 
2.42.0.869.gea05f2083d-goog

[xdp-hints] [PATCH bpf-next v5 08/13] xsk: Add option to calculate TX checksum in SW

[Stanislav Fomichev]

For XDP_COPY mode, add a UMEM option XDP_UMEM_TX_SW_CSUM
to call skb_checksum_help in transmit path. Might be useful
to debugging issues with real hardware. I also use this mode
in the selftests.

Signed-off-by: Stanislav Fomichev <sdf@google.com>
---
 Documentation/networking/xsk-tx-metadata.rst | 9 +++++++++
 include/net/xsk_buff_pool.h                  | 1 +
 include/uapi/linux/if_xdp.h                  | 8 +++++++-
 net/xdp/xdp_umem.c                           | 7 ++++++-
 net/xdp/xsk.c                                | 6 ++++++
 net/xdp/xsk_buff_pool.c                      | 1 +
 tools/include/uapi/linux/if_xdp.h            | 8 +++++++-
 7 files changed, 37 insertions(+), 3 deletions(-)

diff --git a/Documentation/networking/xsk-tx-metadata.rst b/Documentation/networking/xsk-tx-metadata.rst
index 4f376560b23f..97ecfa480d00 100644
--- a/Documentation/networking/xsk-tx-metadata.rst
+++ b/Documentation/networking/xsk-tx-metadata.rst
@@ -50,6 +50,15 @@ packet's ``struct xdp_desc`` descriptor should set ``XDP_TX_METADATA``
 bit in the ``options`` field. Also note that in a multi-buffer packet
 only the first chunk should carry the metadata.
 
+Software TX Checksum
+====================
+
+For development and testing purposes its possible to pass
+``XDP_UMEM_TX_SW_CSUM`` flag to ``XDP_UMEM_REG`` UMEM registration call.
+In this case, when running in ``XDK_COPY`` mode, the TX checksum
+is calculated on the CPU. Do not enable this option in production because
+it will negatively affect performance.
+
 Querying Device Capabilities
 ============================
 
diff --git a/include/net/xsk_buff_pool.h b/include/net/xsk_buff_pool.h
index 97f5cc10d79e..8d48d37ab7c0 100644
--- a/include/net/xsk_buff_pool.h
+++ b/include/net/xsk_buff_pool.h
@@ -83,6 +83,7 @@ struct xsk_buff_pool {
 	bool uses_need_wakeup;
 	bool dma_need_sync;
 	bool unaligned;
+	bool tx_sw_csum;
 	void *addrs;
 	/* Mutual exclusion of the completion ring in the SKB mode. Two cases to protect:
 	 * NAPI TX thread and sendmsg error paths in the SKB destructor callback and when
diff --git a/include/uapi/linux/if_xdp.h b/include/uapi/linux/if_xdp.h
index b0ee7ad19b51..bed6adc3e9b4 100644
--- a/include/uapi/linux/if_xdp.h
+++ b/include/uapi/linux/if_xdp.h
@@ -33,7 +33,13 @@
 #define XDP_USE_SG	(1 << 4)
 
 /* Flags for xsk_umem_config flags */
-#define XDP_UMEM_UNALIGNED_CHUNK_FLAG (1 << 0)
+#define XDP_UMEM_UNALIGNED_CHUNK_FLAG	(1 << 0)
+
+/* Force checksum calculation in software. Can be used for testing or
+ * working around potential HW issues. This option causes performance
+ * degradation and only works in XDP_COPY mode.
+ */
+#define XDP_UMEM_TX_SW_CSUM		(1 << 1)
 
 struct sockaddr_xdp {
 	__u16 sxdp_family;
diff --git a/net/xdp/xdp_umem.c b/net/xdp/xdp_umem.c
index 946a687fb8e8..caa340134b0e 100644
--- a/net/xdp/xdp_umem.c
+++ b/net/xdp/xdp_umem.c
@@ -148,6 +148,11 @@ static int xdp_umem_account_pages(struct xdp_umem *umem)
 	return 0;
 }
 
+#define XDP_UMEM_FLAGS_VALID ( \
+		XDP_UMEM_UNALIGNED_CHUNK_FLAG | \
+		XDP_UMEM_TX_SW_CSUM | \
+	0)
+
 static int xdp_umem_reg(struct xdp_umem *umem, struct xdp_umem_reg *mr)
 {
 	bool unaligned_chunks = mr->flags & XDP_UMEM_UNALIGNED_CHUNK_FLAG;
@@ -167,7 +172,7 @@ static int xdp_umem_reg(struct xdp_umem *umem, struct xdp_umem_reg *mr)
 		return -EINVAL;
 	}
 
-	if (mr->flags & ~XDP_UMEM_UNALIGNED_CHUNK_FLAG)
+	if (mr->flags & ~XDP_UMEM_FLAGS_VALID)
 		return -EINVAL;
 
 	if (!unaligned_chunks && !is_power_of_2(chunk_size))
diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c
index 0e81ae6bfff4..e109b2aaeb2a 100644
--- a/net/xdp/xsk.c
+++ b/net/xdp/xsk.c
@@ -744,6 +744,12 @@ static struct sk_buff *xsk_build_skb(struct xdp_sock *xs,
 				skb->csum_start = hr + meta->request.csum_start;
 				skb->csum_offset = meta->request.csum_offset;
 				skb->ip_summed = CHECKSUM_PARTIAL;
+
+				if (unlikely(xs->pool->tx_sw_csum)) {
+					err = skb_checksum_help(skb);
+					if (err)
+						goto free_err;
+				}
 			}
 		}
 	}
diff --git a/net/xdp/xsk_buff_pool.c b/net/xdp/xsk_buff_pool.c
index 386eddcdf837..4f6f538a5462 100644
--- a/net/xdp/xsk_buff_pool.c
+++ b/net/xdp/xsk_buff_pool.c
@@ -86,6 +86,7 @@ struct xsk_buff_pool *xp_create_and_assign_umem(struct xdp_sock *xs,
 	pool->umem = umem;
 	pool->addrs = umem->addrs;
 	pool->tx_metadata_len = umem->tx_metadata_len;
+	pool->tx_sw_csum = umem->flags & XDP_UMEM_TX_SW_CSUM;
 	INIT_LIST_HEAD(&pool->free_list);
 	INIT_LIST_HEAD(&pool->xskb_list);
 	INIT_LIST_HEAD(&pool->xsk_tx_list);
diff --git a/tools/include/uapi/linux/if_xdp.h b/tools/include/uapi/linux/if_xdp.h
index 29942c2c32dc..8ed2ed9d1b17 100644
--- a/tools/include/uapi/linux/if_xdp.h
+++ b/tools/include/uapi/linux/if_xdp.h
@@ -33,7 +33,13 @@
 #define XDP_USE_SG	(1 << 4)
 
 /* Flags for xsk_umem_config flags */
-#define XDP_UMEM_UNALIGNED_CHUNK_FLAG (1 << 0)
+#define XDP_UMEM_UNALIGNED_CHUNK_FLAG	(1 << 0)
+
+/* Force checksum calculation in software. Can be used for testing or
+ * working around potential HW issues. This option causes performance
+ * degradation and only works in XDP_COPY mode.
+ */
+#define XDP_UMEM_TX_SW_CSUM		(1 << 1)
 
 struct sockaddr_xdp {
 	__u16 sxdp_family;
-- 
2.42.0.869.gea05f2083d-goog

[xdp-hints] [PATCH bpf-next v5 09/13] selftests/xsk: Support tx_metadata_len

[Stanislav Fomichev]

Add new config field and propagate to UMEM registration setsockopt.

Signed-off-by: Stanislav Fomichev <sdf@google.com>
---
 tools/testing/selftests/bpf/xsk.c | 3 +++
 tools/testing/selftests/bpf/xsk.h | 1 +
 2 files changed, 4 insertions(+)

diff --git a/tools/testing/selftests/bpf/xsk.c b/tools/testing/selftests/bpf/xsk.c
index e574711eeb84..25d568abf0f2 100644
--- a/tools/testing/selftests/bpf/xsk.c
+++ b/tools/testing/selftests/bpf/xsk.c
@@ -115,6 +115,7 @@ static void xsk_set_umem_config(struct xsk_umem_config *cfg,
 		cfg->frame_size = XSK_UMEM__DEFAULT_FRAME_SIZE;
 		cfg->frame_headroom = XSK_UMEM__DEFAULT_FRAME_HEADROOM;
 		cfg->flags = XSK_UMEM__DEFAULT_FLAGS;
+		cfg->tx_metadata_len = 0;
 		return;
 	}
 
@@ -123,6 +124,7 @@ static void xsk_set_umem_config(struct xsk_umem_config *cfg,
 	cfg->frame_size = usr_cfg->frame_size;
 	cfg->frame_headroom = usr_cfg->frame_headroom;
 	cfg->flags = usr_cfg->flags;
+	cfg->tx_metadata_len = usr_cfg->tx_metadata_len;
 }
 
 static int xsk_set_xdp_socket_config(struct xsk_socket_config *cfg,
@@ -252,6 +254,7 @@ int xsk_umem__create(struct xsk_umem **umem_ptr, void *umem_area,
 	mr.chunk_size = umem->config.frame_size;
 	mr.headroom = umem->config.frame_headroom;
 	mr.flags = umem->config.flags;
+	mr.tx_metadata_len = umem->config.tx_metadata_len;
 
 	err = setsockopt(umem->fd, SOL_XDP, XDP_UMEM_REG, &mr, sizeof(mr));
 	if (err) {
diff --git a/tools/testing/selftests/bpf/xsk.h b/tools/testing/selftests/bpf/xsk.h
index 771570bc3731..93c2cc413cfc 100644
--- a/tools/testing/selftests/bpf/xsk.h
+++ b/tools/testing/selftests/bpf/xsk.h
@@ -200,6 +200,7 @@ struct xsk_umem_config {
 	__u32 frame_size;
 	__u32 frame_headroom;
 	__u32 flags;
+	__u32 tx_metadata_len;
 };
 
 int xsk_attach_xdp_program(struct bpf_program *prog, int ifindex, u32 xdp_flags);
-- 
2.42.0.869.gea05f2083d-goog

[xdp-hints] [PATCH bpf-next v5 10/13] selftests/bpf: Add csum helpers

[Stanislav Fomichev]

Checksum helpers will be used to calculate pseudo-header checksum in
AF_XDP metadata selftests.

The helpers are mirroring existing kernel ones:
- csum_tcpudp_magic : IPv4 pseudo header csum
- csum_ipv6_magic : IPv6 pseudo header csum
- csum_fold : fold csum and do one's complement

Signed-off-by: Stanislav Fomichev <sdf@google.com>
---
 tools/testing/selftests/bpf/network_helpers.h | 43 +++++++++++++++++++
 1 file changed, 43 insertions(+)

diff --git a/tools/testing/selftests/bpf/network_helpers.h b/tools/testing/selftests/bpf/network_helpers.h
index 34f1200a781b..94b9be24e39b 100644
--- a/tools/testing/selftests/bpf/network_helpers.h
+++ b/tools/testing/selftests/bpf/network_helpers.h
@@ -71,4 +71,47 @@ struct nstoken;
  */
 struct nstoken *open_netns(const char *name);
 void close_netns(struct nstoken *token);
+
+static __u16 csum_fold(__u32 csum)
+{
+	csum = (csum & 0xffff) + (csum >> 16);
+	csum = (csum & 0xffff) + (csum >> 16);
+
+	return (__u16)~csum;
+}
+
+static inline __sum16 csum_tcpudp_magic(__be32 saddr, __be32 daddr,
+					__u32 len, __u8 proto,
+					__wsum csum)
+{
+	__u64 s = csum;
+
+	s += (__u32)saddr;
+	s += (__u32)daddr;
+	s += htons(proto + len);
+	s = (s & 0xffffffff) + (s >> 32);
+	s = (s & 0xffffffff) + (s >> 32);
+
+	return csum_fold((__u32)s);
+}
+
+static inline __sum16 csum_ipv6_magic(const struct in6_addr *saddr,
+				      const struct in6_addr *daddr,
+					__u32 len, __u8 proto,
+					__wsum csum)
+{
+	__u64 s = csum;
+	int i;
+
+	for (i = 0; i < 4; i++)
+		s += (__u32)saddr->s6_addr32[i];
+	for (i = 0; i < 4; i++)
+		s += (__u32)daddr->s6_addr32[i];
+	s += htons(proto + len);
+	s = (s & 0xffffffff) + (s >> 32);
+	s = (s & 0xffffffff) + (s >> 32);
+
+	return csum_fold((__u32)s);
+}
+
 #endif
-- 
2.42.0.869.gea05f2083d-goog

[xdp-hints] [PATCH bpf-next v5 11/13] selftests/bpf: Add TX side to xdp_metadata

[Stanislav Fomichev]

Request TX timestamp and make sure it's not empty.
Request TX checksum offload (SW-only) and make sure it's resolved
to the correct one.

Signed-off-by: Stanislav Fomichev <sdf@google.com>
---
 .../selftests/bpf/prog_tests/xdp_metadata.c   | 33 ++++++++++++++++---
 1 file changed, 29 insertions(+), 4 deletions(-)

diff --git a/tools/testing/selftests/bpf/prog_tests/xdp_metadata.c b/tools/testing/selftests/bpf/prog_tests/xdp_metadata.c
index 4439ba9392f8..9eec39969b1e 100644
--- a/tools/testing/selftests/bpf/prog_tests/xdp_metadata.c
+++ b/tools/testing/selftests/bpf/prog_tests/xdp_metadata.c
@@ -56,7 +56,8 @@ static int open_xsk(int ifindex, struct xsk *xsk)
 		.fill_size = XSK_RING_PROD__DEFAULT_NUM_DESCS,
 		.comp_size = XSK_RING_CONS__DEFAULT_NUM_DESCS,
 		.frame_size = XSK_UMEM__DEFAULT_FRAME_SIZE,
-		.flags = XDP_UMEM_UNALIGNED_CHUNK_FLAG,
+		.flags = XDP_UMEM_UNALIGNED_CHUNK_FLAG | XDP_UMEM_TX_SW_CSUM,
+		.tx_metadata_len = sizeof(struct xsk_tx_metadata),
 	};
 	__u32 idx;
 	u64 addr;
@@ -138,6 +139,7 @@ static void ip_csum(struct iphdr *iph)
 
 static int generate_packet(struct xsk *xsk, __u16 dst_port)
 {
+	struct xsk_tx_metadata *meta;
 	struct xdp_desc *tx_desc;
 	struct udphdr *udph;
 	struct ethhdr *eth;
@@ -151,10 +153,14 @@ static int generate_packet(struct xsk *xsk, __u16 dst_port)
 		return -1;
 
 	tx_desc = xsk_ring_prod__tx_desc(&xsk->tx, idx);
-	tx_desc->addr = idx % (UMEM_NUM / 2) * UMEM_FRAME_SIZE;
+	tx_desc->addr = idx % (UMEM_NUM / 2) * UMEM_FRAME_SIZE + sizeof(struct xsk_tx_metadata);
 	printf("%p: tx_desc[%u]->addr=%llx\n", xsk, idx, tx_desc->addr);
 	data = xsk_umem__get_data(xsk->umem_area, tx_desc->addr);
 
+	meta = data - sizeof(struct xsk_tx_metadata);
+	memset(meta, 0, sizeof(*meta));
+	meta->request.flags = XDP_TXMD_FLAGS_TIMESTAMP;
+
 	eth = data;
 	iph = (void *)(eth + 1);
 	udph = (void *)(iph + 1);
@@ -178,11 +184,17 @@ static int generate_packet(struct xsk *xsk, __u16 dst_port)
 	udph->source = htons(AF_XDP_SOURCE_PORT);
 	udph->dest = htons(dst_port);
 	udph->len = htons(sizeof(*udph) + UDP_PAYLOAD_BYTES);
-	udph->check = 0;
+	udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
+					 ntohs(udph->len), IPPROTO_UDP, 0);
 
 	memset(udph + 1, 0xAA, UDP_PAYLOAD_BYTES);
 
+	meta->request.flags |= XDP_TXMD_FLAGS_CHECKSUM;
+	meta->request.csum_start = sizeof(*eth) + sizeof(*iph);
+	meta->request.csum_offset = offsetof(struct udphdr, check);
+
 	tx_desc->len = sizeof(*eth) + sizeof(*iph) + sizeof(*udph) + UDP_PAYLOAD_BYTES;
+	tx_desc->options |= XDP_TX_METADATA;
 	xsk_ring_prod__submit(&xsk->tx, 1);
 
 	ret = sendto(xsk_socket__fd(xsk->socket), NULL, 0, MSG_DONTWAIT, NULL, 0);
@@ -194,13 +206,21 @@ static int generate_packet(struct xsk *xsk, __u16 dst_port)
 
 static void complete_tx(struct xsk *xsk)
 {
-	__u32 idx;
+	struct xsk_tx_metadata *meta;
 	__u64 addr;
+	void *data;
+	__u32 idx;
 
 	if (ASSERT_EQ(xsk_ring_cons__peek(&xsk->comp, 1, &idx), 1, "xsk_ring_cons__peek")) {
 		addr = *xsk_ring_cons__comp_addr(&xsk->comp, idx);
 
 		printf("%p: complete tx idx=%u addr=%llx\n", xsk, idx, addr);
+
+		data = xsk_umem__get_data(xsk->umem_area, addr);
+		meta = data - sizeof(struct xsk_tx_metadata);
+
+		ASSERT_NEQ(meta->completion.tx_timestamp, 0, "tx_timestamp");
+
 		xsk_ring_cons__release(&xsk->comp, 1);
 	}
 }
@@ -221,6 +241,7 @@ static int verify_xsk_metadata(struct xsk *xsk)
 	const struct xdp_desc *rx_desc;
 	struct pollfd fds = {};
 	struct xdp_meta *meta;
+	struct udphdr *udph;
 	struct ethhdr *eth;
 	struct iphdr *iph;
 	__u64 comp_addr;
@@ -257,6 +278,7 @@ static int verify_xsk_metadata(struct xsk *xsk)
 	ASSERT_EQ(eth->h_proto, htons(ETH_P_IP), "eth->h_proto");
 	iph = (void *)(eth + 1);
 	ASSERT_EQ((int)iph->version, 4, "iph->version");
+	udph = (void *)(iph + 1);
 
 	/* custom metadata */
 
@@ -270,6 +292,9 @@ static int verify_xsk_metadata(struct xsk *xsk)
 
 	ASSERT_EQ(meta->rx_hash_type, 0, "rx_hash_type");
 
+	/* checksum offload */
+	ASSERT_EQ(udph->check, 0x1c72, "csum");
+
 	xsk_ring_cons__release(&xsk->rx, 1);
 	refill_rx(xsk, comp_addr);
 
-- 
2.42.0.869.gea05f2083d-goog

[xdp-hints] [PATCH bpf-next v5 12/13] selftests/bpf: Convert xdp_hw_metadata to XDP_USE_NEED_WAKEUP

[Stanislav Fomichev]

This is the recommended way to run AF_XDP, so let's use it in the test.

Also, some unrelated changes to now blow up the log too much:
- change default mode to zerocopy and add -c to use copy mode
- small fixes for the flags/sizes/prints
- add print_tstamp_delta to print timestamp + reference

Signed-off-by: Stanislav Fomichev <sdf@google.com>
---
 tools/testing/selftests/bpf/xdp_hw_metadata.c | 73 +++++++++++++------
 1 file changed, 49 insertions(+), 24 deletions(-)

diff --git a/tools/testing/selftests/bpf/xdp_hw_metadata.c b/tools/testing/selftests/bpf/xdp_hw_metadata.c
index 17c0f92ff160..057f7c145f62 100644
--- a/tools/testing/selftests/bpf/xdp_hw_metadata.c
+++ b/tools/testing/selftests/bpf/xdp_hw_metadata.c
@@ -32,7 +32,7 @@
 
 #include "xdp_metadata.h"
 
-#define UMEM_NUM 16
+#define UMEM_NUM 256
 #define UMEM_FRAME_SIZE XSK_UMEM__DEFAULT_FRAME_SIZE
 #define UMEM_SIZE (UMEM_FRAME_SIZE * UMEM_NUM)
 #define XDP_FLAGS (XDP_FLAGS_DRV_MODE | XDP_FLAGS_REPLACE)
@@ -48,7 +48,7 @@ struct xsk {
 };
 
 struct xdp_hw_metadata *bpf_obj;
-__u16 bind_flags = XDP_COPY;
+__u16 bind_flags = XDP_USE_NEED_WAKEUP | XDP_ZEROCOPY;
 struct xsk *rx_xsk;
 const char *ifname;
 int ifindex;
@@ -68,7 +68,7 @@ static int open_xsk(int ifindex, struct xsk *xsk, __u32 queue_id)
 		.fill_size = XSK_RING_PROD__DEFAULT_NUM_DESCS,
 		.comp_size = XSK_RING_CONS__DEFAULT_NUM_DESCS,
 		.frame_size = XSK_UMEM__DEFAULT_FRAME_SIZE,
-		.flags = XDP_UMEM_UNALIGNED_CHUNK_FLAG,
+		.flags = XSK_UMEM__DEFAULT_FLAGS,
 	};
 	__u32 idx;
 	u64 addr;
@@ -110,7 +110,7 @@ static int open_xsk(int ifindex, struct xsk *xsk, __u32 queue_id)
 	for (i = 0; i < UMEM_NUM / 2; i++) {
 		addr = (UMEM_NUM / 2 + i) * UMEM_FRAME_SIZE;
 		printf("%p: rx_desc[%d] -> %lx\n", xsk, i, addr);
-		*xsk_ring_prod__fill_addr(&xsk->fill, i) = addr;
+		*xsk_ring_prod__fill_addr(&xsk->fill, idx + i) = addr;
 	}
 	xsk_ring_prod__submit(&xsk->fill, ret);
 
@@ -131,12 +131,22 @@ static void refill_rx(struct xsk *xsk, __u64 addr)
 	__u32 idx;
 
 	if (xsk_ring_prod__reserve(&xsk->fill, 1, &idx) == 1) {
-		printf("%p: complete idx=%u addr=%llx\n", xsk, idx, addr);
+		printf("%p: complete rx idx=%u addr=%llx\n", xsk, idx, addr);
 		*xsk_ring_prod__fill_addr(&xsk->fill, idx) = addr;
 		xsk_ring_prod__submit(&xsk->fill, 1);
 	}
 }
 
+static int kick_tx(struct xsk *xsk)
+{
+	return sendto(xsk_socket__fd(xsk->socket), NULL, 0, MSG_DONTWAIT, NULL, 0);
+}
+
+static int kick_rx(struct xsk *xsk)
+{
+	return recvfrom(xsk_socket__fd(xsk->socket), NULL, 0, MSG_DONTWAIT, NULL, NULL);
+}
+
 #define NANOSEC_PER_SEC 1000000000 /* 10^9 */
 static __u64 gettime(clockid_t clock_id)
 {
@@ -152,6 +162,17 @@ static __u64 gettime(clockid_t clock_id)
 	return (__u64) t.tv_sec * NANOSEC_PER_SEC + t.tv_nsec;
 }
 
+static void print_tstamp_delta(const char *name, const char *refname,
+			       __u64 tstamp, __u64 reference)
+{
+	__s64 delta = (__s64)reference - (__s64)tstamp;
+
+	printf("%s:   %llu (sec:%0.4f) delta to %s sec:%0.4f (%0.3f usec)\n",
+	       name, tstamp, (double)tstamp / NANOSEC_PER_SEC, refname,
+	       (double)delta / NANOSEC_PER_SEC,
+	       (double)delta / 1000);
+}
+
 static void verify_xdp_metadata(void *data, clockid_t clock_id)
 {
 	struct xdp_meta *meta;
@@ -167,22 +188,13 @@ static void verify_xdp_metadata(void *data, clockid_t clock_id)
 	printf("rx_timestamp:  %llu (sec:%0.4f)\n", meta->rx_timestamp,
 	       (double)meta->rx_timestamp / NANOSEC_PER_SEC);
 	if (meta->rx_timestamp) {
-		__u64 usr_clock = gettime(clock_id);
-		__u64 xdp_clock = meta->xdp_timestamp;
-		__s64 delta_X = xdp_clock - meta->rx_timestamp;
-		__s64 delta_X2U = usr_clock - xdp_clock;
-
-		printf("XDP RX-time:   %llu (sec:%0.4f) delta sec:%0.4f (%0.3f usec)\n",
-		       xdp_clock, (double)xdp_clock / NANOSEC_PER_SEC,
-		       (double)delta_X / NANOSEC_PER_SEC,
-		       (double)delta_X / 1000);
-
-		printf("AF_XDP time:   %llu (sec:%0.4f) delta sec:%0.4f (%0.3f usec)\n",
-		       usr_clock, (double)usr_clock / NANOSEC_PER_SEC,
-		       (double)delta_X2U / NANOSEC_PER_SEC,
-		       (double)delta_X2U / 1000);
-	}
+		__u64 ref_tstamp = gettime(clock_id);
 
+		print_tstamp_delta("HW RX-time", "User RX-time",
+				   meta->rx_timestamp, ref_tstamp);
+		print_tstamp_delta("XDP RX-time", "User RX-time",
+				   meta->xdp_timestamp, ref_tstamp);
+	}
 }
 
 static void verify_skb_metadata(int fd)
@@ -252,6 +264,13 @@ static int verify_metadata(struct xsk *rx_xsk, int rxq, int server_fd, clockid_t
 
 	while (true) {
 		errno = 0;
+
+		for (i = 0; i < rxq; i++) {
+			ret = kick_rx(&rx_xsk[i]);
+			if (ret)
+				printf("kick_rx ret=%d\n", ret);
+		}
+
 		ret = poll(fds, rxq + 1, 1000);
 		printf("poll: %d (%d) skip=%llu fail=%llu redir=%llu\n",
 		       ret, errno, bpf_obj->bss->pkts_skip,
@@ -420,8 +439,9 @@ static void print_usage(void)
 {
 	const char *usage =
 		"Usage: xdp_hw_metadata [OPTIONS] [IFNAME]\n"
-		"  -m    Enable multi-buffer XDP for larger MTU\n"
+		"  -c    Run in copy mode (zerocopy is default)\n"
 		"  -h    Display this help and exit\n\n"
+		"  -m    Enable multi-buffer XDP for larger MTU\n"
 		"Generate test packets on the other machine with:\n"
 		"  echo -n xdp | nc -u -q1 <dst_ip> 9091\n";
 
@@ -432,14 +452,19 @@ static void read_args(int argc, char *argv[])
 {
 	char opt;
 
-	while ((opt = getopt(argc, argv, "mh")) != -1) {
+	while ((opt = getopt(argc, argv, "chm")) != -1) {
 		switch (opt) {
-		case 'm':
-			bind_flags |= XDP_USE_SG;
+		case 'c':
+			bind_flags &= ~XDP_USE_NEED_WAKEUP;
+			bind_flags &= ~XDP_ZEROCOPY;
+			bind_flags |= XDP_COPY;
 			break;
 		case 'h':
 			print_usage();
 			exit(0);
+		case 'm':
+			bind_flags |= XDP_USE_SG;
+			break;
 		case '?':
 			if (isprint(optopt))
 				fprintf(stderr, "Unknown option: -%c\n", optopt);
-- 
2.42.0.869.gea05f2083d-goog

[xdp-hints] [PATCH bpf-next v5 13/13] selftests/bpf: Add TX side to xdp_hw_metadata

[Stanislav Fomichev]

When we get a packet on port 9091, we swap src/dst and send it out.
At this point we also request the timestamp and checksum offloads.

Checksum offload is verified by looking at the tcpdump on the other side.
The tool prints pseudo-header csum and the final one it expects.
The final checksum actually matches the incoming packets checksum
because we only flip the src/dst and don't change the payload.

Some other related changes:
- switched to zerocopy mode by default; new flag can be used to force
  old behavior
- request fixed tx_metadata_len headroom
- some other small fixes (umem size, fill idx+i, etc)

mvbz3:~# ./xdp_hw_metadata eth3
...
xsk_ring_cons__peek: 1
0x19546f8: rx_desc[0]->addr=80100 addr=80100 comp_addr=80100
rx_hash: 0x80B7EA8B with RSS type:0x2A
rx_timestamp:  1697580171852147395 (sec:1697580171.8521)
HW RX-time:   1697580171852147395 (sec:1697580171.8521), delta to User RX-time sec:0.2797 (279673.082 usec)
XDP RX-time:   1697580172131699047 (sec:1697580172.1317), delta to User RX-time sec:0.0001 (121.430 usec)
0x19546f8: ping-pong with csum=3b8e (want d862) csum_start=54 csum_offset=6
0x19546f8: complete tx idx=0 addr=8
tx_timestamp:  1697580172056756493 (sec:1697580172.0568)
HW TX-complete-time:   1697580172056756493 (sec:1697580172.0568), delta to User TX-complete-time sec:0.0852 (85175.537 usec)
XDP RX-time:   1697580172131699047 (sec:1697580172.1317), delta to User TX-complete-time sec:0.0102 (10232.983 usec)
HW RX-time:   1697580171852147395 (sec:1697580171.8521), delta to HW TX-complete-time sec:0.2046 (204609.098 usec)
0x19546f8: complete rx idx=128 addr=80100

mvbz4:~# nc  -Nu -q1 ${MVBZ3_LINK_LOCAL_IP}%eth3 9091

mvbz4:~# tcpdump -vvx -i eth3 udp
        tcpdump: listening on eth3, link-type EN10MB (Ethernet), snapshot length 262144 bytes
12:26:09.301074 IP6 (flowlabel 0x35fa5, hlim 127, next-header UDP (17) payload length: 11) fe80::1270:fdff:fe48:1087.55807 > fe80::1270:fdff:fe48:1077.9091: [bad udp cksum 0x3b8e -> 0xde7e!] UDP, length 3
        0x0000:  6003 5fa5 000b 117f fe80 0000 0000 0000
        0x0010:  1270 fdff fe48 1087 fe80 0000 0000 0000
        0x0020:  1270 fdff fe48 1077 d9ff 2383 000b 3b8e
        0x0030:  7864 70
12:26:09.301976 IP6 (flowlabel 0x35fa5, hlim 127, next-header UDP (17) payload length: 11) fe80::1270:fdff:fe48:1077.9091 > fe80::1270:fdff:fe48:1087.55807: [udp sum ok] UDP, length 3
        0x0000:  6003 5fa5 000b 117f fe80 0000 0000 0000
        0x0010:  1270 fdff fe48 1077 fe80 0000 0000 0000
        0x0020:  1270 fdff fe48 1087 2383 d9ff 000b de7e
        0x0030:  7864 70

Signed-off-by: Stanislav Fomichev <sdf@google.com>
---
 tools/testing/selftests/bpf/xdp_hw_metadata.c | 164 +++++++++++++++++-
 1 file changed, 160 insertions(+), 4 deletions(-)

diff --git a/tools/testing/selftests/bpf/xdp_hw_metadata.c b/tools/testing/selftests/bpf/xdp_hw_metadata.c
index 057f7c145f62..efaa3243483a 100644
--- a/tools/testing/selftests/bpf/xdp_hw_metadata.c
+++ b/tools/testing/selftests/bpf/xdp_hw_metadata.c
@@ -10,7 +10,9 @@
  *   - rx_hash
  *
  * TX:
- * - TBD
+ * - UDP 9091 packets trigger TX reply
+ * - TX HW timestamp is requested and reported back upon completion
+ * - TX checksum is requested
  */
 
 #include <test_progs.h>
@@ -24,11 +26,14 @@
 #include <linux/net_tstamp.h>
 #include <linux/udp.h>
 #include <linux/sockios.h>
+#include <linux/if_xdp.h>
 #include <sys/mman.h>
 #include <net/if.h>
 #include <ctype.h>
 #include <poll.h>
 #include <time.h>
+#include <unistd.h>
+#include <libgen.h>
 
 #include "xdp_metadata.h"
 
@@ -53,6 +58,9 @@ struct xsk *rx_xsk;
 const char *ifname;
 int ifindex;
 int rxq;
+bool skip_tx;
+__u64 last_hw_rx_timestamp;
+__u64 last_xdp_rx_timestamp;
 
 void test__fail(void) { /* for network_helpers.c */ }
 
@@ -69,6 +77,7 @@ static int open_xsk(int ifindex, struct xsk *xsk, __u32 queue_id)
 		.comp_size = XSK_RING_CONS__DEFAULT_NUM_DESCS,
 		.frame_size = XSK_UMEM__DEFAULT_FRAME_SIZE,
 		.flags = XSK_UMEM__DEFAULT_FLAGS,
+		.tx_metadata_len = sizeof(struct xsk_tx_metadata),
 	};
 	__u32 idx;
 	u64 addr;
@@ -185,15 +194,19 @@ static void verify_xdp_metadata(void *data, clockid_t clock_id)
 		printf("rx_hash: 0x%X with RSS type:0x%X\n",
 		       meta->rx_hash, meta->rx_hash_type);
 
-	printf("rx_timestamp:  %llu (sec:%0.4f)\n", meta->rx_timestamp,
-	       (double)meta->rx_timestamp / NANOSEC_PER_SEC);
 	if (meta->rx_timestamp) {
 		__u64 ref_tstamp = gettime(clock_id);
 
+		/* store received timestamps to calculate a delta at tx */
+		last_hw_rx_timestamp = meta->rx_timestamp;
+		last_xdp_rx_timestamp = meta->xdp_timestamp;
+
 		print_tstamp_delta("HW RX-time", "User RX-time",
 				   meta->rx_timestamp, ref_tstamp);
 		print_tstamp_delta("XDP RX-time", "User RX-time",
 				   meta->xdp_timestamp, ref_tstamp);
+	} else {
+		printf("No rx_timestamp\n");
 	}
 }
 
@@ -242,6 +255,129 @@ static void verify_skb_metadata(int fd)
 	printf("skb hwtstamp is not found!\n");
 }
 
+static bool complete_tx(struct xsk *xsk, clockid_t clock_id)
+{
+	struct xsk_tx_metadata *meta;
+	__u64 addr;
+	void *data;
+	__u32 idx;
+
+	if (!xsk_ring_cons__peek(&xsk->comp, 1, &idx))
+		return false;
+
+	addr = *xsk_ring_cons__comp_addr(&xsk->comp, idx);
+	data = xsk_umem__get_data(xsk->umem_area, addr);
+	meta = data - sizeof(struct xsk_tx_metadata);
+
+	printf("%p: complete tx idx=%u addr=%llx\n", xsk, idx, addr);
+
+	if (meta->completion.tx_timestamp) {
+		__u64 ref_tstamp = gettime(clock_id);
+
+		print_tstamp_delta("HW TX-complete-time", "User TX-complete-time",
+				   meta->completion.tx_timestamp, ref_tstamp);
+		print_tstamp_delta("XDP RX-time", "User TX-complete-time",
+				   last_xdp_rx_timestamp, ref_tstamp);
+		print_tstamp_delta("HW RX-time", "HW TX-complete-time",
+				   last_hw_rx_timestamp, meta->completion.tx_timestamp);
+	} else {
+		printf("No tx_timestamp\n");
+	}
+
+	xsk_ring_cons__release(&xsk->comp, 1);
+
+	return true;
+}
+
+#define swap(a, b, len) do { \
+	for (int i = 0; i < len; i++) { \
+		__u8 tmp = ((__u8 *)a)[i]; \
+		((__u8 *)a)[i] = ((__u8 *)b)[i]; \
+		((__u8 *)b)[i] = tmp; \
+	} \
+} while (0)
+
+static void ping_pong(struct xsk *xsk, void *rx_packet, clockid_t clock_id)
+{
+	struct xsk_tx_metadata *meta;
+	struct ipv6hdr *ip6h = NULL;
+	struct iphdr *iph = NULL;
+	struct xdp_desc *tx_desc;
+	struct udphdr *udph;
+	struct ethhdr *eth;
+	__sum16 want_csum;
+	void *data;
+	__u32 idx;
+	int ret;
+	int len;
+
+	ret = xsk_ring_prod__reserve(&xsk->tx, 1, &idx);
+	if (ret != 1) {
+		printf("%p: failed to reserve tx slot\n", xsk);
+		return;
+	}
+
+	tx_desc = xsk_ring_prod__tx_desc(&xsk->tx, idx);
+	tx_desc->addr = idx % (UMEM_NUM / 2) * UMEM_FRAME_SIZE + sizeof(struct xsk_tx_metadata);
+	data = xsk_umem__get_data(xsk->umem_area, tx_desc->addr);
+
+	meta = data - sizeof(struct xsk_tx_metadata);
+	memset(meta, 0, sizeof(*meta));
+	meta->request.flags = XDP_TXMD_FLAGS_TIMESTAMP;
+
+	eth = rx_packet;
+
+	if (eth->h_proto == htons(ETH_P_IP)) {
+		iph = (void *)(eth + 1);
+		udph = (void *)(iph + 1);
+	} else if (eth->h_proto == htons(ETH_P_IPV6)) {
+		ip6h = (void *)(eth + 1);
+		udph = (void *)(ip6h + 1);
+	} else {
+		printf("%p: failed to detect IP version for ping pong %04x\n", xsk, eth->h_proto);
+		xsk_ring_prod__cancel(&xsk->tx, 1);
+		return;
+	}
+
+	len = ETH_HLEN;
+	if (ip6h)
+		len += sizeof(*ip6h) + ntohs(ip6h->payload_len);
+	if (iph)
+		len += ntohs(iph->tot_len);
+
+	swap(eth->h_dest, eth->h_source, ETH_ALEN);
+	if (iph)
+		swap(&iph->saddr, &iph->daddr, 4);
+	else
+		swap(&ip6h->saddr, &ip6h->daddr, 16);
+	swap(&udph->source, &udph->dest, 2);
+
+	want_csum = udph->check;
+	if (ip6h)
+		udph->check = ~csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
+					       ntohs(udph->len), IPPROTO_UDP, 0);
+	else
+		udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
+						 ntohs(udph->len), IPPROTO_UDP, 0);
+
+	meta->request.flags |= XDP_TXMD_FLAGS_CHECKSUM;
+	if (iph)
+		meta->request.csum_start = sizeof(*eth) + sizeof(*iph);
+	else
+		meta->request.csum_start = sizeof(*eth) + sizeof(*ip6h);
+	meta->request.csum_offset = offsetof(struct udphdr, check);
+
+	printf("%p: ping-pong with csum=%04x (want %04x) csum_start=%d csum_offset=%d\n",
+	       xsk, ntohs(udph->check), ntohs(want_csum),
+	       meta->request.csum_start, meta->request.csum_offset);
+
+	memcpy(data, rx_packet, len); /* don't share umem chunk for simplicity */
+	tx_desc->options |= XDP_TX_METADATA;
+	tx_desc->len = len;
+
+	xsk_ring_prod__submit(&xsk->tx, 1);
+}
+
 static int verify_metadata(struct xsk *rx_xsk, int rxq, int server_fd, clockid_t clock_id)
 {
 	const struct xdp_desc *rx_desc;
@@ -307,6 +443,22 @@ static int verify_metadata(struct xsk *rx_xsk, int rxq, int server_fd, clockid_t
 				verify_xdp_metadata(xsk_umem__get_data(xsk->umem_area, addr),
 						    clock_id);
 				first_seg = false;
+
+				if (!skip_tx) {
+					/* mirror first chunk back */
+					ping_pong(xsk, xsk_umem__get_data(xsk->umem_area, addr),
+						  clock_id);
+
+					ret = kick_tx(xsk);
+					if (ret)
+						printf("kick_tx ret=%d\n", ret);
+
+					for (int j = 0; j < 500; j++) {
+						if (complete_tx(xsk, clock_id))
+							break;
+						usleep(10*1000);
+					}
+				}
 			}
 
 			xsk_ring_cons__release(&xsk->rx, 1);
@@ -442,6 +594,7 @@ static void print_usage(void)
 		"  -c    Run in copy mode (zerocopy is default)\n"
 		"  -h    Display this help and exit\n\n"
 		"  -m    Enable multi-buffer XDP for larger MTU\n"
+		"  -r    Don't generate AF_XDP reply (rx metadata only)\n"
 		"Generate test packets on the other machine with:\n"
 		"  echo -n xdp | nc -u -q1 <dst_ip> 9091\n";
 
@@ -452,7 +605,7 @@ static void read_args(int argc, char *argv[])
 {
 	char opt;
 
-	while ((opt = getopt(argc, argv, "chm")) != -1) {
+	while ((opt = getopt(argc, argv, "chmr")) != -1) {
 		switch (opt) {
 		case 'c':
 			bind_flags &= ~XDP_USE_NEED_WAKEUP;
@@ -465,6 +618,9 @@ static void read_args(int argc, char *argv[])
 		case 'm':
 			bind_flags |= XDP_USE_SG;
 			break;
+		case 'r':
+			skip_tx = true;
+			break;
 		case '?':
 			if (isprint(optopt))
 				fprintf(stderr, "Unknown option: -%c\n", optopt);
-- 
2.42.0.869.gea05f2083d-goog

[xdp-hints] Re: [PATCH bpf-next v5 06/13] xsk: Document tx_metadata_len layout

[Simon Horman]

On Thu, Nov 02, 2023 at 03:58:30PM -0700, Stanislav Fomichev wrote:
> - how to use
> - how to query features
> - pointers to the examples
> 
> Signed-off-by: Stanislav Fomichev <sdf@google.com>

...

> diff --git a/Documentation/networking/xsk-tx-metadata.rst b/Documentation/networking/xsk-tx-metadata.rst
> new file mode 100644
> index 000000000000..4f376560b23f
> --- /dev/null
> +++ b/Documentation/networking/xsk-tx-metadata.rst
> @@ -0,0 +1,70 @@

Hi Stan,

a minor nit from my side: an SPDX licence identifier tag should probably go
here.

> +==================
> +AF_XDP TX Metadata
> +==================
> +
> +This document describes how to enable offloads when transmitting packets
> +via :doc:`af_xdp`. Refer to :doc:`xdp-rx-metadata` on how to access similar
> +metadata on the receive side.

...

[xdp-hints] Re: [PATCH bpf-next v5 06/13] xsk: Document tx_metadata_len layout

[Stanislav Fomichev]

On Sun, Nov 5, 2023 at 4:45 AM Simon Horman <horms@kernel.org> wrote:
>
> On Thu, Nov 02, 2023 at 03:58:30PM -0700, Stanislav Fomichev wrote:
> > - how to use
> > - how to query features
> > - pointers to the examples
> >
> > Signed-off-by: Stanislav Fomichev <sdf@google.com>
>
> ...
>
> > diff --git a/Documentation/networking/xsk-tx-metadata.rst b/Documentation/networking/xsk-tx-metadata.rst
> > new file mode 100644
> > index 000000000000..4f376560b23f
> > --- /dev/null
> > +++ b/Documentation/networking/xsk-tx-metadata.rst
> > @@ -0,0 +1,70 @@
>
> Hi Stan,
>
> a minor nit from my side: an SPDX licence identifier tag should probably go
> here.

Ugh, thanks, not sure how I missed it :-(

[xdp-hints] Re: [PATCH bpf-next v5 00/13] xsk: TX metadata

[Jakub Kicinski]

On Thu,  2 Nov 2023 15:58:24 -0700 Stanislav Fomichev wrote:
> This series implements initial TX metadata (offloads) for AF_XDP.
> See patch #2 for the main implementation and mlx5/stmmac ones for the
> example on how to consume the metadata on the device side.
> 
> Starting with two types of offloads:
> - request TX timestamp (and write it back into the metadata area)
> - request TX checksum offload

Acked-by: Jakub Kicinski <kuba@kernel.org>

[xdp-hints] Re: [PATCH bpf-next v5 00/13] xsk: TX metadata

[Alexei Starovoitov]

On Thu, Nov 2, 2023 at 3:58 PM Stanislav Fomichev <sdf@google.com> wrote:
>
> This series implements initial TX metadata (offloads) for AF_XDP.
> See patch #2 for the main implementation and mlx5/stmmac ones for the
> example on how to consume the metadata on the device side.
>
> Starting with two types of offloads:
> - request TX timestamp (and write it back into the metadata area)
> - request TX checksum offload
>
> Changes since v4:
> - remove 'render-max: true' from spec (Jakub)
> - move xsk_tx_metadata_ops into include/net/xdp_sock.h (Jakub)
> - christmas tree in netdev_nl_dev_fill (Jakub)
> - fix > vs >= when dumping masks in samples (Jakub)
> - switch to 8-byte alignment for tx metadata length (Jakub)
> - spelling fixes in the doc (Magnus)
> - deny metadata length >= 256 (Magnus)
> - validate metadata flags and deny unknown ones (Jakub)
> - move XDP_TX_METADATA_CHECKSUM_SW into umem config flag (Jakub)
> - don't print timestamps twice in xdp_hw_metadata (Song)
> - rename anonymous xsk_tx_metadata member into request (Alexei)
> - add comment to xsk_tx_metadata (Alexei)
>
> I've separated new bits that need a closer review into separate patches:
> - xsk_tx_metadata flags validation:
>   - xsk: Validate xsk_tx_metadata flags
> - new umem flag for sw tx csum calculation (instead of per-packet flag)
>   - xsk: Add option to calculate TX checksum in SW

Stan,

new xdp_metadata is failing on s390. See BPF CI:

verify_xsk_metadata:FAIL:csum unexpected csum: actual 29212 != expected 7282

Other than this I think the patchset is good to go.
Pls fix and respin.

[xdp-hints] Re: [PATCH bpf-next v5 00/13] xsk: TX metadata

[Stanislav Fomichev]

On Thu, Nov 9, 2023 at 10:03 AM Alexei Starovoitov
<alexei.starovoitov@gmail.com> wrote:
>
> On Thu, Nov 2, 2023 at 3:58 PM Stanislav Fomichev <sdf@google.com> wrote:
> >
> > This series implements initial TX metadata (offloads) for AF_XDP.
> > See patch #2 for the main implementation and mlx5/stmmac ones for the
> > example on how to consume the metadata on the device side.
> >
> > Starting with two types of offloads:
> > - request TX timestamp (and write it back into the metadata area)
> > - request TX checksum offload
> >
> > Changes since v4:
> > - remove 'render-max: true' from spec (Jakub)
> > - move xsk_tx_metadata_ops into include/net/xdp_sock.h (Jakub)
> > - christmas tree in netdev_nl_dev_fill (Jakub)
> > - fix > vs >= when dumping masks in samples (Jakub)
> > - switch to 8-byte alignment for tx metadata length (Jakub)
> > - spelling fixes in the doc (Magnus)
> > - deny metadata length >= 256 (Magnus)
> > - validate metadata flags and deny unknown ones (Jakub)
> > - move XDP_TX_METADATA_CHECKSUM_SW into umem config flag (Jakub)
> > - don't print timestamps twice in xdp_hw_metadata (Song)
> > - rename anonymous xsk_tx_metadata member into request (Alexei)
> > - add comment to xsk_tx_metadata (Alexei)
> >
> > I've separated new bits that need a closer review into separate patches:
> > - xsk_tx_metadata flags validation:
> >   - xsk: Validate xsk_tx_metadata flags
> > - new umem flag for sw tx csum calculation (instead of per-packet flag)
> >   - xsk: Add option to calculate TX checksum in SW
>
> Stan,
>
> new xdp_metadata is failing on s390. See BPF CI:
>
> verify_xsk_metadata:FAIL:csum unexpected csum: actual 29212 != expected 7282
>
> Other than this I think the patchset is good to go.
> Pls fix and respin.

Oh, thanks for catching this, probably some endianness issues? Will take a look!

[xdp-hints] Re: [PATCH bpf-next v5 02/13] xsk: Add TX timestamp and TX checksum offload support

[Jesper Dangaard Brouer]

On 11/2/23 23:58, Stanislav Fomichev wrote:
> diff --git a/include/uapi/linux/if_xdp.h b/include/uapi/linux/if_xdp.h
> index 2ecf79282c26..b0ee7ad19b51 100644
> --- a/include/uapi/linux/if_xdp.h
> +++ b/include/uapi/linux/if_xdp.h
> @@ -106,6 +106,41 @@ struct xdp_options {
>   #define XSK_UNALIGNED_BUF_ADDR_MASK \
>   	((1ULL << XSK_UNALIGNED_BUF_OFFSET_SHIFT) - 1)
>   
> +/* Request transmit timestamp. Upon completion, put it into tx_timestamp
> + * field of struct xsk_tx_metadata.
> + */
> +#define XDP_TXMD_FLAGS_TIMESTAMP		(1 << 0)
> +
> +/* Request transmit checksum offload. Checksum start position and offset
> + * are communicated via csum_start and csum_offset fields of struct
> + * xsk_tx_metadata.
> + */
> +#define XDP_TXMD_FLAGS_CHECKSUM			(1 << 1)
> +
> +/* AF_XDP offloads request. 'request' union member is consumed by the driver
> + * when the packet is being transmitted. 'completion' union member is
> + * filled by the driver when the transmit completion arrives.
> + */
> +struct xsk_tx_metadata {
> +	union {
> +		struct {
> +			__u32 flags;
> +
> +			/* XDP_TXMD_FLAGS_CHECKSUM */
> +
> +			/* Offset from desc->addr where checksumming should start. */
> +			__u16 csum_start;
> +			/* Offset from csum_start where checksum should be stored. */
> +			__u16 csum_offset;
> +		} request;
> +
> +		struct {
> +			/* XDP_TXMD_FLAGS_TIMESTAMP */
> +			__u64 tx_timestamp;
> +		} completion;
> +	};
> +};

This looks wrong to me. It looks like member @flags is not avail at
completion time.  At completion time, I assume we also want to know if
someone requested to get the timestamp for this packet (else we could
read garbage).

Another thing (I've raised this before): It would be really practical to
store an u64 opaque value at TX and then read it at Completion time.
One use-case is a forwarding application storing HW RX-time and
comparing this to TX completion time to deduce the time spend processing
the packet.

--Jesper

[xdp-hints] Re: [PATCH bpf-next v5 02/13] xsk: Add TX timestamp and TX checksum offload support

[Stanislav Fomichev]

On Mon, Nov 13, 2023 at 5:16 AM Jesper Dangaard Brouer <hawk@kernel.org> wrote:
>
>
>
> On 11/2/23 23:58, Stanislav Fomichev wrote:
> > diff --git a/include/uapi/linux/if_xdp.h b/include/uapi/linux/if_xdp.h
> > index 2ecf79282c26..b0ee7ad19b51 100644
> > --- a/include/uapi/linux/if_xdp.h
> > +++ b/include/uapi/linux/if_xdp.h
> > @@ -106,6 +106,41 @@ struct xdp_options {
> >   #define XSK_UNALIGNED_BUF_ADDR_MASK \
> >       ((1ULL << XSK_UNALIGNED_BUF_OFFSET_SHIFT) - 1)
> >
> > +/* Request transmit timestamp. Upon completion, put it into tx_timestamp
> > + * field of struct xsk_tx_metadata.
> > + */
> > +#define XDP_TXMD_FLAGS_TIMESTAMP             (1 << 0)
> > +
> > +/* Request transmit checksum offload. Checksum start position and offset
> > + * are communicated via csum_start and csum_offset fields of struct
> > + * xsk_tx_metadata.
> > + */
> > +#define XDP_TXMD_FLAGS_CHECKSUM                      (1 << 1)
> > +
> > +/* AF_XDP offloads request. 'request' union member is consumed by the driver
> > + * when the packet is being transmitted. 'completion' union member is
> > + * filled by the driver when the transmit completion arrives.
> > + */
> > +struct xsk_tx_metadata {
> > +     union {
> > +             struct {
> > +                     __u32 flags;
> > +
> > +                     /* XDP_TXMD_FLAGS_CHECKSUM */
> > +
> > +                     /* Offset from desc->addr where checksumming should start. */
> > +                     __u16 csum_start;
> > +                     /* Offset from csum_start where checksum should be stored. */
> > +                     __u16 csum_offset;
> > +             } request;
> > +
> > +             struct {
> > +                     /* XDP_TXMD_FLAGS_TIMESTAMP */
> > +                     __u64 tx_timestamp;
> > +             } completion;
> > +     };
> > +};
>
> This looks wrong to me. It looks like member @flags is not avail at
> completion time.  At completion time, I assume we also want to know if
> someone requested to get the timestamp for this packet (else we could
> read garbage).

I've moved the parts that are preserved across tx and tx completion
into xsk_tx_metadata_compl.
This is to address Magnus/Maciej feedback where userspace might race
with the kernel.
See: https://lore.kernel.org/bpf/ZNoJenzKXW5QSR3E@boxer/

> Another thing (I've raised this before): It would be really practical to
> store an u64 opaque value at TX and then read it at Completion time.
> One use-case is a forwarding application storing HW RX-time and
> comparing this to TX completion time to deduce the time spend processing
> the packet.

This can be another member, right? But note that extending
xsk_tx_metadata_compl might be a bit complicated because drivers have
to carry this info somewhere. So we have to balance the amount of
passed data between the tx and the completion.

[xdp-hints] Re: [PATCH bpf-next v5 02/13] xsk: Add TX timestamp and TX checksum offload support

[Jesper Dangaard Brouer]

On 11/13/23 15:10, Stanislav Fomichev wrote:
> On Mon, Nov 13, 2023 at 5:16 AM Jesper Dangaard Brouer <hawk@kernel.org> wrote:
>>
>>
>> On 11/2/23 23:58, Stanislav Fomichev wrote:
>>> diff --git a/include/uapi/linux/if_xdp.h b/include/uapi/linux/if_xdp.h
>>> index 2ecf79282c26..b0ee7ad19b51 100644
>>> --- a/include/uapi/linux/if_xdp.h
>>> +++ b/include/uapi/linux/if_xdp.h
>>> @@ -106,6 +106,41 @@ struct xdp_options {
>>>    #define XSK_UNALIGNED_BUF_ADDR_MASK \
>>>        ((1ULL << XSK_UNALIGNED_BUF_OFFSET_SHIFT) - 1)
>>>
>>> +/* Request transmit timestamp. Upon completion, put it into tx_timestamp
>>> + * field of struct xsk_tx_metadata.
>>> + */
>>> +#define XDP_TXMD_FLAGS_TIMESTAMP             (1 << 0)
>>> +
>>> +/* Request transmit checksum offload. Checksum start position and offset
>>> + * are communicated via csum_start and csum_offset fields of struct
>>> + * xsk_tx_metadata.
>>> + */
>>> +#define XDP_TXMD_FLAGS_CHECKSUM                      (1 << 1)
>>> +
>>> +/* AF_XDP offloads request. 'request' union member is consumed by the driver
>>> + * when the packet is being transmitted. 'completion' union member is
>>> + * filled by the driver when the transmit completion arrives.
>>> + */
>>> +struct xsk_tx_metadata {
>>> +     union {
>>> +             struct {
>>> +                     __u32 flags;
>>> +
>>> +                     /* XDP_TXMD_FLAGS_CHECKSUM */
>>> +
>>> +                     /* Offset from desc->addr where checksumming should start. */
>>> +                     __u16 csum_start;
>>> +                     /* Offset from csum_start where checksum should be stored. */
>>> +                     __u16 csum_offset;
>>> +             } request;
>>> +
>>> +             struct {
>>> +                     /* XDP_TXMD_FLAGS_TIMESTAMP */
>>> +                     __u64 tx_timestamp;
>>> +             } completion;
>>> +     };
>>> +};
>>
>> This looks wrong to me. It looks like member @flags is not avail at
>> completion time.  At completion time, I assume we also want to know if
>> someone requested to get the timestamp for this packet (else we could
>> read garbage).
> 
> I've moved the parts that are preserved across tx and tx completion
> into xsk_tx_metadata_compl.
> This is to address Magnus/Maciej feedback where userspace might race
> with the kernel.
> See: https://lore.kernel.org/bpf/ZNoJenzKXW5QSR3E@boxer/
> 

Does this mean that every driver have to extend their TX-desc ring with
sizeof(struct xsk_tx_metadata_compl)?
Won't this affect the performance of this V5?

  $ pahole -C xsk_tx_metadata_compl 
./drivers/net/ethernet/stmicro/stmmac/stmmac.ko
  struct xsk_tx_metadata_compl {
	__u64 *              tx_timestamp;         /*     0     8 */

	/* size: 8, cachelines: 1, members: 1 */
	/* last cacheline: 8 bytes */
  };

Guess, I must be misunderstanding, as I was expecting to see the @flags
member being preserved across, as I get the race there.

Looking at stmmac driver, it does look like this xsk_tx_metadata_compl
is part of the TX-ring for completion (tx_skbuff_dma) and the
tx_timestamp data is getting stored here.  How is userspace AF_XDP
application getting access to the tx_timestamp data?
I though this was suppose to get stored in metadata data area (umem)?

Also looking at the code, the kernel would not have a "crash" race on
the flags member (if we preserve in struct), because the code checks the
driver HW-TS config-state + TX-descriptor for the availability of a
HW-TS in the descriptor.


>> Another thing (I've raised this before): It would be really practical to
>> store an u64 opaque value at TX and then read it at Completion time.
>> One use-case is a forwarding application storing HW RX-time and
>> comparing this to TX completion time to deduce the time spend processing
>> the packet.
> 
> This can be another member, right? But note that extending
> xsk_tx_metadata_compl might be a bit complicated because drivers have
> to carry this info somewhere. So we have to balance the amount of
> passed data between the tx and the completion.

I don't think my opaque value proposal is subject to same race problem.
I think this can be stores in metadata area and across tx and tx
completion, because any race on a flags change is the userspace
programmers problem, as it cannot cause any kernel crash (given kernel
have no need to read this).

--Jesper

[xdp-hints] Re: [PATCH bpf-next v5 02/13] xsk: Add TX timestamp and TX checksum offload support

[Stanislav Fomichev]

On 11/13, Jesper Dangaard Brouer wrote:
> 
> 
> On 11/13/23 15:10, Stanislav Fomichev wrote:
> > On Mon, Nov 13, 2023 at 5:16 AM Jesper Dangaard Brouer <hawk@kernel.org> wrote:
> > > 
> > > 
> > > On 11/2/23 23:58, Stanislav Fomichev wrote:
> > > > diff --git a/include/uapi/linux/if_xdp.h b/include/uapi/linux/if_xdp.h
> > > > index 2ecf79282c26..b0ee7ad19b51 100644
> > > > --- a/include/uapi/linux/if_xdp.h
> > > > +++ b/include/uapi/linux/if_xdp.h
> > > > @@ -106,6 +106,41 @@ struct xdp_options {
> > > >    #define XSK_UNALIGNED_BUF_ADDR_MASK \
> > > >        ((1ULL << XSK_UNALIGNED_BUF_OFFSET_SHIFT) - 1)
> > > > 
> > > > +/* Request transmit timestamp. Upon completion, put it into tx_timestamp
> > > > + * field of struct xsk_tx_metadata.
> > > > + */
> > > > +#define XDP_TXMD_FLAGS_TIMESTAMP             (1 << 0)
> > > > +
> > > > +/* Request transmit checksum offload. Checksum start position and offset
> > > > + * are communicated via csum_start and csum_offset fields of struct
> > > > + * xsk_tx_metadata.
> > > > + */
> > > > +#define XDP_TXMD_FLAGS_CHECKSUM                      (1 << 1)
> > > > +
> > > > +/* AF_XDP offloads request. 'request' union member is consumed by the driver
> > > > + * when the packet is being transmitted. 'completion' union member is
> > > > + * filled by the driver when the transmit completion arrives.
> > > > + */
> > > > +struct xsk_tx_metadata {
> > > > +     union {
> > > > +             struct {
> > > > +                     __u32 flags;
> > > > +
> > > > +                     /* XDP_TXMD_FLAGS_CHECKSUM */
> > > > +
> > > > +                     /* Offset from desc->addr where checksumming should start. */
> > > > +                     __u16 csum_start;
> > > > +                     /* Offset from csum_start where checksum should be stored. */
> > > > +                     __u16 csum_offset;
> > > > +             } request;
> > > > +
> > > > +             struct {
> > > > +                     /* XDP_TXMD_FLAGS_TIMESTAMP */
> > > > +                     __u64 tx_timestamp;
> > > > +             } completion;
> > > > +     };
> > > > +};
> > > 
> > > This looks wrong to me. It looks like member @flags is not avail at
> > > completion time.  At completion time, I assume we also want to know if
> > > someone requested to get the timestamp for this packet (else we could
> > > read garbage).
> > 
> > I've moved the parts that are preserved across tx and tx completion
> > into xsk_tx_metadata_compl.
> > This is to address Magnus/Maciej feedback where userspace might race
> > with the kernel.
> > See: https://lore.kernel.org/bpf/ZNoJenzKXW5QSR3E@boxer/
> > 
> 
> Does this mean that every driver have to extend their TX-desc ring with
> sizeof(struct xsk_tx_metadata_compl)?
> Won't this affect the performance of this V5?

Yes, but it doesn't have to be a descriptor. Might be some internal
driver completion queue (as in the case of mlx5). And definitely does
affect performance :-( (see all the static branches to disable it)
 
>  $ pahole -C xsk_tx_metadata_compl
> ./drivers/net/ethernet/stmicro/stmmac/stmmac.ko
>  struct xsk_tx_metadata_compl {
> 	__u64 *              tx_timestamp;         /*     0     8 */
> 
> 	/* size: 8, cachelines: 1, members: 1 */
> 	/* last cacheline: 8 bytes */
>  };
> 
> Guess, I must be misunderstanding, as I was expecting to see the @flags
> member being preserved across, as I get the race there.
>
> Looking at stmmac driver, it does look like this xsk_tx_metadata_compl
> is part of the TX-ring for completion (tx_skbuff_dma) and the
> tx_timestamp data is getting stored here.  How is userspace AF_XDP
> application getting access to the tx_timestamp data?
> I though this was suppose to get stored in metadata data area (umem)?
>
> Also looking at the code, the kernel would not have a "crash" race on
> the flags member (if we preserve in struct), because the code checks the
> driver HW-TS config-state + TX-descriptor for the availability of a
> HW-TS in the descriptor.

xsk_tx_metadata_compl stores a pointer to the completion timestamp
in the umem, so everything still arrives via the metadata area.

We want to make sure the flags are not changing across tx and tx completion.
Instead of saving the flags, we just use that xsk_tx_metadata_compl to
signal to the completion that "I know that I've requested the tx
completion timestamp, please put it at this address in umem".

I store the pointer instead of flags to avoid doing pointer math again
at completion. But it's an implementation detail and somewhat abstracted
from the drivers (besides the fact that it's probably has to fit in 8
bytes).

> > > Another thing (I've raised this before): It would be really practical to
> > > store an u64 opaque value at TX and then read it at Completion time.
> > > One use-case is a forwarding application storing HW RX-time and
> > > comparing this to TX completion time to deduce the time spend processing
> > > the packet.
> > 
> > This can be another member, right? But note that extending
> > xsk_tx_metadata_compl might be a bit complicated because drivers have
> > to carry this info somewhere. So we have to balance the amount of
> > passed data between the tx and the completion.
> 
> I don't think my opaque value proposal is subject to same race problem.
> I think this can be stores in metadata area and across tx and tx
> completion, because any race on a flags change is the userspace
> programmers problem, as it cannot cause any kernel crash (given kernel
> have no need to read this).

Thinking about it, I don't think this needs any special handing?
You can request sizeof(struct xsk_tx_metadata) + sizeof(opaque data)
as metadata. The kernel won't touch the 'opaque data' part. Or am I missing
something?

[xdp-hints] Re: [PATCH bpf-next v5 02/13] xsk: Add TX timestamp and TX checksum offload support

[Jesper Dangaard Brouer]

On 11/13/23 18:02, Stanislav Fomichev wrote:
> On 11/13, Jesper Dangaard Brouer wrote:
>>
>>
>> On 11/13/23 15:10, Stanislav Fomichev wrote:
>>> On Mon, Nov 13, 2023 at 5:16 AM Jesper Dangaard Brouer <hawk@kernel.org> wrote:
>>>>
>>>>
>>>> On 11/2/23 23:58, Stanislav Fomichev wrote:
>>>>> diff --git a/include/uapi/linux/if_xdp.h b/include/uapi/linux/if_xdp.h
>>>>> index 2ecf79282c26..b0ee7ad19b51 100644
>>>>> --- a/include/uapi/linux/if_xdp.h
>>>>> +++ b/include/uapi/linux/if_xdp.h
>>>>> @@ -106,6 +106,41 @@ struct xdp_options {
>>>>>     #define XSK_UNALIGNED_BUF_ADDR_MASK \
>>>>>         ((1ULL << XSK_UNALIGNED_BUF_OFFSET_SHIFT) - 1)
>>>>>
>>>>> +/* Request transmit timestamp. Upon completion, put it into tx_timestamp
>>>>> + * field of struct xsk_tx_metadata.
>>>>> + */
>>>>> +#define XDP_TXMD_FLAGS_TIMESTAMP             (1 << 0)
>>>>> +
>>>>> +/* Request transmit checksum offload. Checksum start position and offset
>>>>> + * are communicated via csum_start and csum_offset fields of struct
>>>>> + * xsk_tx_metadata.
>>>>> + */
>>>>> +#define XDP_TXMD_FLAGS_CHECKSUM                      (1 << 1)
>>>>> +
>>>>> +/* AF_XDP offloads request. 'request' union member is consumed by the driver
>>>>> + * when the packet is being transmitted. 'completion' union member is
>>>>> + * filled by the driver when the transmit completion arrives.
>>>>> + */
>>>>> +struct xsk_tx_metadata {
>>>>> +     union {
>>>>> +             struct {
>>>>> +                     __u32 flags;
>>>>> +
>>>>> +                     /* XDP_TXMD_FLAGS_CHECKSUM */
>>>>> +
>>>>> +                     /* Offset from desc->addr where checksumming should start. */
>>>>> +                     __u16 csum_start;
>>>>> +                     /* Offset from csum_start where checksum should be stored. */
>>>>> +                     __u16 csum_offset;
>>>>> +             } request;
>>>>> +
>>>>> +             struct {
>>>>> +                     /* XDP_TXMD_FLAGS_TIMESTAMP */
>>>>> +                     __u64 tx_timestamp;
>>>>> +             } completion;
>>>>> +     };
>>>>> +};
>>>>
>>>> This looks wrong to me. It looks like member @flags is not avail at
>>>> completion time.  At completion time, I assume we also want to know if
>>>> someone requested to get the timestamp for this packet (else we could
>>>> read garbage).
>>>
>>> I've moved the parts that are preserved across tx and tx completion
>>> into xsk_tx_metadata_compl.
>>> This is to address Magnus/Maciej feedback where userspace might race
>>> with the kernel.
>>> See: https://lore.kernel.org/bpf/ZNoJenzKXW5QSR3E@boxer/
>>>
>>
>> Does this mean that every driver have to extend their TX-desc ring with
>> sizeof(struct xsk_tx_metadata_compl)?
>> Won't this affect the performance of this V5?
> 
> Yes, but it doesn't have to be a descriptor. Might be some internal
> driver completion queue (as in the case of mlx5). And definitely does
> affect performance :-( (see all the static branches to disable it)
>   
>>   $ pahole -C xsk_tx_metadata_compl
>> ./drivers/net/ethernet/stmicro/stmmac/stmmac.ko
>>   struct xsk_tx_metadata_compl {
>> 	__u64 *              tx_timestamp;         /*     0     8 */
>>
>> 	/* size: 8, cachelines: 1, members: 1 */
>> 	/* last cacheline: 8 bytes */
>>   };
>>
>> Guess, I must be misunderstanding, as I was expecting to see the @flags
>> member being preserved across, as I get the race there.
>>
>> Looking at stmmac driver, it does look like this xsk_tx_metadata_compl
>> is part of the TX-ring for completion (tx_skbuff_dma) and the
>> tx_timestamp data is getting stored here.  How is userspace AF_XDP
>> application getting access to the tx_timestamp data?
>> I though this was suppose to get stored in metadata data area (umem)?
>>
>> Also looking at the code, the kernel would not have a "crash" race on
>> the flags member (if we preserve in struct), because the code checks the
>> driver HW-TS config-state + TX-descriptor for the availability of a
>> HW-TS in the descriptor.
> 
> xsk_tx_metadata_compl stores a pointer to the completion timestamp
> in the umem, so everything still arrives via the metadata area.
> 
> We want to make sure the flags are not changing across tx and tx completion.
> Instead of saving the flags, we just use that xsk_tx_metadata_compl to
> signal to the completion that "I know that I've requested the tx
> completion timestamp, please put it at this address in umem".
> 
> I store the pointer instead of flags to avoid doing pointer math again
> at completion. But it's an implementation detail and somewhat abstracted
> from the drivers (besides the fact that it's probably has to fit in 8
> bytes).

I see it now (what I missed). At TX time you are storing a pointer where
to (later) write the TS at completion time.  It just seems overkill to
store 8 byte (pointer) to signal (via NULL) if the HWTS was requested.
Space in the drivers TX-ring is performance critical, and I think driver
developers would prefer to find a bit to indicate HWTS requested.

If HWTS was *NOT* requested, then the metadata area will not be updated
(right, correct?). Then memory area is basically garbage that survived.
How does the AF_XDP application know this packet contains a HWTS or not?

 From an UAPI PoV wouldn't it be easier to use (and extend) via keeping
the @flags member (in struct xsk_tx_metadata), but (as you already do)
not let kernel checks depend on it (to avoid the races).

> 
>>>> Another thing (I've raised this before): It would be really practical to
>>>> store an u64 opaque value at TX and then read it at Completion time.
>>>> One use-case is a forwarding application storing HW RX-time and
>>>> comparing this to TX completion time to deduce the time spend processing
>>>> the packet.
>>>
>>> This can be another member, right? But note that extending
>>> xsk_tx_metadata_compl might be a bit complicated because drivers have
>>> to carry this info somewhere. So we have to balance the amount of
>>> passed data between the tx and the completion.
>>
>> I don't think my opaque value proposal is subject to same race problem.
>> I think this can be stores in metadata area and across tx and tx
>> completion, because any race on a flags change is the userspace
>> programmers problem, as it cannot cause any kernel crash (given kernel
>> have no need to read this).
> 
> Thinking about it, I don't think this needs any special handing?
> You can request sizeof(struct xsk_tx_metadata) + sizeof(opaque data)
> as metadata. The kernel won't touch the 'opaque data' part. Or am I missing
> something?

Sure, I can just create some room after struct xsk_tx_metadata, via 
setting something larger than sizeof(struct xsk_tx_metadata).  I'm 
buying that.

--Jesper

[xdp-hints] Re: [PATCH bpf-next v5 02/13] xsk: Add TX timestamp and TX checksum offload support

[Stanislav Fomichev]

On 11/15, Jesper Dangaard Brouer wrote:
> 
> 
> On 11/13/23 18:02, Stanislav Fomichev wrote:
> > On 11/13, Jesper Dangaard Brouer wrote:
> > > 
> > > 
> > > On 11/13/23 15:10, Stanislav Fomichev wrote:
> > > > On Mon, Nov 13, 2023 at 5:16 AM Jesper Dangaard Brouer <hawk@kernel.org> wrote:
> > > > > 
> > > > > 
> > > > > On 11/2/23 23:58, Stanislav Fomichev wrote:
> > > > > > diff --git a/include/uapi/linux/if_xdp.h b/include/uapi/linux/if_xdp.h
> > > > > > index 2ecf79282c26..b0ee7ad19b51 100644
> > > > > > --- a/include/uapi/linux/if_xdp.h
> > > > > > +++ b/include/uapi/linux/if_xdp.h
> > > > > > @@ -106,6 +106,41 @@ struct xdp_options {
> > > > > >     #define XSK_UNALIGNED_BUF_ADDR_MASK \
> > > > > >         ((1ULL << XSK_UNALIGNED_BUF_OFFSET_SHIFT) - 1)
> > > > > > 
> > > > > > +/* Request transmit timestamp. Upon completion, put it into tx_timestamp
> > > > > > + * field of struct xsk_tx_metadata.
> > > > > > + */
> > > > > > +#define XDP_TXMD_FLAGS_TIMESTAMP             (1 << 0)
> > > > > > +
> > > > > > +/* Request transmit checksum offload. Checksum start position and offset
> > > > > > + * are communicated via csum_start and csum_offset fields of struct
> > > > > > + * xsk_tx_metadata.
> > > > > > + */
> > > > > > +#define XDP_TXMD_FLAGS_CHECKSUM                      (1 << 1)
> > > > > > +
> > > > > > +/* AF_XDP offloads request. 'request' union member is consumed by the driver
> > > > > > + * when the packet is being transmitted. 'completion' union member is
> > > > > > + * filled by the driver when the transmit completion arrives.
> > > > > > + */
> > > > > > +struct xsk_tx_metadata {
> > > > > > +     union {
> > > > > > +             struct {
> > > > > > +                     __u32 flags;
> > > > > > +
> > > > > > +                     /* XDP_TXMD_FLAGS_CHECKSUM */
> > > > > > +
> > > > > > +                     /* Offset from desc->addr where checksumming should start. */
> > > > > > +                     __u16 csum_start;
> > > > > > +                     /* Offset from csum_start where checksum should be stored. */
> > > > > > +                     __u16 csum_offset;
> > > > > > +             } request;
> > > > > > +
> > > > > > +             struct {
> > > > > > +                     /* XDP_TXMD_FLAGS_TIMESTAMP */
> > > > > > +                     __u64 tx_timestamp;
> > > > > > +             } completion;
> > > > > > +     };
> > > > > > +};
> > > > > 
> > > > > This looks wrong to me. It looks like member @flags is not avail at
> > > > > completion time.  At completion time, I assume we also want to know if
> > > > > someone requested to get the timestamp for this packet (else we could
> > > > > read garbage).
> > > > 
> > > > I've moved the parts that are preserved across tx and tx completion
> > > > into xsk_tx_metadata_compl.
> > > > This is to address Magnus/Maciej feedback where userspace might race
> > > > with the kernel.
> > > > See: https://lore.kernel.org/bpf/ZNoJenzKXW5QSR3E@boxer/
> > > > 
> > > 
> > > Does this mean that every driver have to extend their TX-desc ring with
> > > sizeof(struct xsk_tx_metadata_compl)?
> > > Won't this affect the performance of this V5?
> > 
> > Yes, but it doesn't have to be a descriptor. Might be some internal
> > driver completion queue (as in the case of mlx5). And definitely does
> > affect performance :-( (see all the static branches to disable it)
> > >   $ pahole -C xsk_tx_metadata_compl
> > > ./drivers/net/ethernet/stmicro/stmmac/stmmac.ko
> > >   struct xsk_tx_metadata_compl {
> > > 	__u64 *              tx_timestamp;         /*     0     8 */
> > > 
> > > 	/* size: 8, cachelines: 1, members: 1 */
> > > 	/* last cacheline: 8 bytes */
> > >   };
> > > 
> > > Guess, I must be misunderstanding, as I was expecting to see the @flags
> > > member being preserved across, as I get the race there.
> > > 
> > > Looking at stmmac driver, it does look like this xsk_tx_metadata_compl
> > > is part of the TX-ring for completion (tx_skbuff_dma) and the
> > > tx_timestamp data is getting stored here.  How is userspace AF_XDP
> > > application getting access to the tx_timestamp data?
> > > I though this was suppose to get stored in metadata data area (umem)?
> > > 
> > > Also looking at the code, the kernel would not have a "crash" race on
> > > the flags member (if we preserve in struct), because the code checks the
> > > driver HW-TS config-state + TX-descriptor for the availability of a
> > > HW-TS in the descriptor.
> > 
> > xsk_tx_metadata_compl stores a pointer to the completion timestamp
> > in the umem, so everything still arrives via the metadata area.
> > 
> > We want to make sure the flags are not changing across tx and tx completion.
> > Instead of saving the flags, we just use that xsk_tx_metadata_compl to
> > signal to the completion that "I know that I've requested the tx
> > completion timestamp, please put it at this address in umem".
> > 
> > I store the pointer instead of flags to avoid doing pointer math again
> > at completion. But it's an implementation detail and somewhat abstracted
> > from the drivers (besides the fact that it's probably has to fit in 8
> > bytes).
> 
> I see it now (what I missed). At TX time you are storing a pointer where
> to (later) write the TS at completion time.  It just seems overkill to
> store 8 byte (pointer) to signal (via NULL) if the HWTS was requested.
> Space in the drivers TX-ring is performance critical, and I think driver
> developers would prefer to find a bit to indicate HWTS requested.
> 
> If HWTS was *NOT* requested, then the metadata area will not be updated
> (right, correct?). Then memory area is basically garbage that survived.
> How does the AF_XDP application know this packet contains a HWTS or not?
> 
> From an UAPI PoV wouldn't it be easier to use (and extend) via keeping
> the @flags member (in struct xsk_tx_metadata), but (as you already do)
> not let kernel checks depend on it (to avoid the races).

I was assuming the userspace can keep this signal out of band or use
the same idea as suggested with padding struct xsk_tx_metadata to keep
some data around. But I see your point, it might be convenient to
keep the original flags around during completion on the uapi side.

I think I can just move flags from the request union member to the outer
struct. So the struct xsk_tx_metadata would look like:

struct xsk_tx_metadata {
	__u32 flags; /* maybe can even make this u64? */

	union {
		__u16 csum_start;
		__u16 csum_offset;
	} request;

	union {
		__u64 tx_timestamp;
	} completion;

	__u32 padding; /* to drop this padding */
};

But I'd also keep the existing xsk_tx_metadata_compl to carry the
pointer+signal around. As I mentioned previously, it's completely
opaque to the driver and we can change the internals in the future.

IOW, we won't override the flags from the kernel side and as long
as the userspace consumer doesn't mess them up it should receive
the original value at completion.

Would that work for you?

[xdp-hints] Re: [PATCH bpf-next v5 02/13] xsk: Add TX timestamp and TX checksum offload support

[Jesper Dangaard Brouer]

On 11/15/23 14:37, Stanislav Fomichev wrote:
> On 11/15, Jesper Dangaard Brouer wrote:
>>
>>
>> On 11/13/23 18:02, Stanislav Fomichev wrote:
>>> On 11/13, Jesper Dangaard Brouer wrote:
>>>>
>>>>
>>>> On 11/13/23 15:10, Stanislav Fomichev wrote:
>>>>> On Mon, Nov 13, 2023 at 5:16 AM Jesper Dangaard Brouer <hawk@kernel.org> wrote:
>>>>>>
>>>>>>
>>>>>> On 11/2/23 23:58, Stanislav Fomichev wrote:
>>>>>>> diff --git a/include/uapi/linux/if_xdp.h b/include/uapi/linux/if_xdp.h
>>>>>>> index 2ecf79282c26..b0ee7ad19b51 100644
>>>>>>> --- a/include/uapi/linux/if_xdp.h
>>>>>>> +++ b/include/uapi/linux/if_xdp.h
>>>>>>> @@ -106,6 +106,41 @@ struct xdp_options {
>>>>>>>      #define XSK_UNALIGNED_BUF_ADDR_MASK \
>>>>>>>          ((1ULL << XSK_UNALIGNED_BUF_OFFSET_SHIFT) - 1)
>>>>>>>
>>>>>>> +/* Request transmit timestamp. Upon completion, put it into tx_timestamp
>>>>>>> + * field of struct xsk_tx_metadata.
>>>>>>> + */
>>>>>>> +#define XDP_TXMD_FLAGS_TIMESTAMP             (1 << 0)
>>>>>>> +
>>>>>>> +/* Request transmit checksum offload. Checksum start position and offset
>>>>>>> + * are communicated via csum_start and csum_offset fields of struct
>>>>>>> + * xsk_tx_metadata.
>>>>>>> + */
>>>>>>> +#define XDP_TXMD_FLAGS_CHECKSUM                      (1 << 1)
>>>>>>> +
>>>>>>> +/* AF_XDP offloads request. 'request' union member is consumed by the driver
>>>>>>> + * when the packet is being transmitted. 'completion' union member is
>>>>>>> + * filled by the driver when the transmit completion arrives.
>>>>>>> + */
>>>>>>> +struct xsk_tx_metadata {
>>>>>>> +     union {
>>>>>>> +             struct {
>>>>>>> +                     __u32 flags;
>>>>>>> +
>>>>>>> +                     /* XDP_TXMD_FLAGS_CHECKSUM */
>>>>>>> +
>>>>>>> +                     /* Offset from desc->addr where checksumming should start. */
>>>>>>> +                     __u16 csum_start;
>>>>>>> +                     /* Offset from csum_start where checksum should be stored. */
>>>>>>> +                     __u16 csum_offset;
>>>>>>> +             } request;
>>>>>>> +
>>>>>>> +             struct {
>>>>>>> +                     /* XDP_TXMD_FLAGS_TIMESTAMP */
>>>>>>> +                     __u64 tx_timestamp;
>>>>>>> +             } completion;
>>>>>>> +     };
>>>>>>> +};
>>>>>>
>>>>>> This looks wrong to me. It looks like member @flags is not avail at
>>>>>> completion time.  At completion time, I assume we also want to know if
>>>>>> someone requested to get the timestamp for this packet (else we could
>>>>>> read garbage).
>>>>>
>>>>> I've moved the parts that are preserved across tx and tx completion
>>>>> into xsk_tx_metadata_compl.
>>>>> This is to address Magnus/Maciej feedback where userspace might race
>>>>> with the kernel.
>>>>> See: https://lore.kernel.org/bpf/ZNoJenzKXW5QSR3E@boxer/
>>>>>
>>>>
>>>> Does this mean that every driver have to extend their TX-desc ring with
>>>> sizeof(struct xsk_tx_metadata_compl)?
>>>> Won't this affect the performance of this V5?
>>>
>>> Yes, but it doesn't have to be a descriptor. Might be some internal
>>> driver completion queue (as in the case of mlx5). And definitely does
>>> affect performance :-( (see all the static branches to disable it)
>>>>    $ pahole -C xsk_tx_metadata_compl
>>>> ./drivers/net/ethernet/stmicro/stmmac/stmmac.ko
>>>>    struct xsk_tx_metadata_compl {
>>>> 	__u64 *              tx_timestamp;         /*     0     8 */
>>>>
>>>> 	/* size: 8, cachelines: 1, members: 1 */
>>>> 	/* last cacheline: 8 bytes */
>>>>    };
>>>>
>>>> Guess, I must be misunderstanding, as I was expecting to see the @flags
>>>> member being preserved across, as I get the race there.
>>>>
>>>> Looking at stmmac driver, it does look like this xsk_tx_metadata_compl
>>>> is part of the TX-ring for completion (tx_skbuff_dma) and the
>>>> tx_timestamp data is getting stored here.  How is userspace AF_XDP
>>>> application getting access to the tx_timestamp data?
>>>> I though this was suppose to get stored in metadata data area (umem)?
>>>>
>>>> Also looking at the code, the kernel would not have a "crash" race on
>>>> the flags member (if we preserve in struct), because the code checks the
>>>> driver HW-TS config-state + TX-descriptor for the availability of a
>>>> HW-TS in the descriptor.
>>>
>>> xsk_tx_metadata_compl stores a pointer to the completion timestamp
>>> in the umem, so everything still arrives via the metadata area.
>>>
>>> We want to make sure the flags are not changing across tx and tx completion.
>>> Instead of saving the flags, we just use that xsk_tx_metadata_compl to
>>> signal to the completion that "I know that I've requested the tx
>>> completion timestamp, please put it at this address in umem".
>>>
>>> I store the pointer instead of flags to avoid doing pointer math again
>>> at completion. But it's an implementation detail and somewhat abstracted
>>> from the drivers (besides the fact that it's probably has to fit in 8
>>> bytes).
>>
>> I see it now (what I missed). At TX time you are storing a pointer where
>> to (later) write the TS at completion time.  It just seems overkill to
>> store 8 byte (pointer) to signal (via NULL) if the HWTS was requested.
>> Space in the drivers TX-ring is performance critical, and I think driver
>> developers would prefer to find a bit to indicate HWTS requested.
>>
>> If HWTS was *NOT* requested, then the metadata area will not be updated
>> (right, correct?). Then memory area is basically garbage that survived.
>> How does the AF_XDP application know this packet contains a HWTS or not?
>>
>>  From an UAPI PoV wouldn't it be easier to use (and extend) via keeping
>> the @flags member (in struct xsk_tx_metadata), but (as you already do)
>> not let kernel checks depend on it (to avoid the races).
> 
> I was assuming the userspace can keep this signal out of band or use
> the same idea as suggested with padding struct xsk_tx_metadata to keep
> some data around. But I see your point, it might be convenient to
> keep the original flags around during completion on the uapi side.
> 
> I think I can just move flags from the request union member to the outer
> struct. So the struct xsk_tx_metadata would look like:
> 
> struct xsk_tx_metadata {
> 	__u32 flags; /* maybe can even make this u64? */
> 

Yes to u64 for two reasons (1) this becomes UAPI and
(2) better alignment for tx_timestamp.
But I'm open to keeping it u32.

> 	union {
> 		__u16 csum_start;
> 		__u16 csum_offset;
> 	} request;
> 
> 	union {
> 		__u64 tx_timestamp;
> 	} completion;
> 
> 	__u32 padding; /* to drop this padding */
> };
> 
> But I'd also keep the existing xsk_tx_metadata_compl to carry the
> pointer+signal around. As I mentioned previously, it's completely
> opaque to the driver and we can change the internals in the future.
> 

Sure, it is an implementation detail and my objections are mostly that I
don't find it as a pretty code approach that can be hard to follow.
Maybe driver developer will object and change this later if it cost too
much to increase the element size in their TX-ring queues.

> IOW, we won't override the flags from the kernel side and as long
> as the userspace consumer doesn't mess them up it should receive
> the original value at completion.
> 
> Would that work for you?

Yes, that will work for me, thanks!

--Jesper

[xdp-hints] Re: [PATCH bpf-next v5 02/13] xsk: Add TX timestamp and TX checksum offload support

[Stanislav Fomichev]

On 11/16, Jesper Dangaard Brouer wrote:
> 
> 
> On 11/15/23 14:37, Stanislav Fomichev wrote:
> > On 11/15, Jesper Dangaard Brouer wrote:
> > > 
> > > 
> > > On 11/13/23 18:02, Stanislav Fomichev wrote:
> > > > On 11/13, Jesper Dangaard Brouer wrote:
> > > > > 
> > > > > 
> > > > > On 11/13/23 15:10, Stanislav Fomichev wrote:
> > > > > > On Mon, Nov 13, 2023 at 5:16 AM Jesper Dangaard Brouer <hawk@kernel.org> wrote:
> > > > > > > 
> > > > > > > 
> > > > > > > On 11/2/23 23:58, Stanislav Fomichev wrote:
> > > > > > > > diff --git a/include/uapi/linux/if_xdp.h b/include/uapi/linux/if_xdp.h
> > > > > > > > index 2ecf79282c26..b0ee7ad19b51 100644
> > > > > > > > --- a/include/uapi/linux/if_xdp.h
> > > > > > > > +++ b/include/uapi/linux/if_xdp.h
> > > > > > > > @@ -106,6 +106,41 @@ struct xdp_options {
> > > > > > > >      #define XSK_UNALIGNED_BUF_ADDR_MASK \
> > > > > > > >          ((1ULL << XSK_UNALIGNED_BUF_OFFSET_SHIFT) - 1)
> > > > > > > > 
> > > > > > > > +/* Request transmit timestamp. Upon completion, put it into tx_timestamp
> > > > > > > > + * field of struct xsk_tx_metadata.
> > > > > > > > + */
> > > > > > > > +#define XDP_TXMD_FLAGS_TIMESTAMP             (1 << 0)
> > > > > > > > +
> > > > > > > > +/* Request transmit checksum offload. Checksum start position and offset
> > > > > > > > + * are communicated via csum_start and csum_offset fields of struct
> > > > > > > > + * xsk_tx_metadata.
> > > > > > > > + */
> > > > > > > > +#define XDP_TXMD_FLAGS_CHECKSUM                      (1 << 1)
> > > > > > > > +
> > > > > > > > +/* AF_XDP offloads request. 'request' union member is consumed by the driver
> > > > > > > > + * when the packet is being transmitted. 'completion' union member is
> > > > > > > > + * filled by the driver when the transmit completion arrives.
> > > > > > > > + */
> > > > > > > > +struct xsk_tx_metadata {
> > > > > > > > +     union {
> > > > > > > > +             struct {
> > > > > > > > +                     __u32 flags;
> > > > > > > > +
> > > > > > > > +                     /* XDP_TXMD_FLAGS_CHECKSUM */
> > > > > > > > +
> > > > > > > > +                     /* Offset from desc->addr where checksumming should start. */
> > > > > > > > +                     __u16 csum_start;
> > > > > > > > +                     /* Offset from csum_start where checksum should be stored. */
> > > > > > > > +                     __u16 csum_offset;
> > > > > > > > +             } request;
> > > > > > > > +
> > > > > > > > +             struct {
> > > > > > > > +                     /* XDP_TXMD_FLAGS_TIMESTAMP */
> > > > > > > > +                     __u64 tx_timestamp;
> > > > > > > > +             } completion;
> > > > > > > > +     };
> > > > > > > > +};
> > > > > > > 
> > > > > > > This looks wrong to me. It looks like member @flags is not avail at
> > > > > > > completion time.  At completion time, I assume we also want to know if
> > > > > > > someone requested to get the timestamp for this packet (else we could
> > > > > > > read garbage).
> > > > > > 
> > > > > > I've moved the parts that are preserved across tx and tx completion
> > > > > > into xsk_tx_metadata_compl.
> > > > > > This is to address Magnus/Maciej feedback where userspace might race
> > > > > > with the kernel.
> > > > > > See: https://lore.kernel.org/bpf/ZNoJenzKXW5QSR3E@boxer/
> > > > > > 
> > > > > 
> > > > > Does this mean that every driver have to extend their TX-desc ring with
> > > > > sizeof(struct xsk_tx_metadata_compl)?
> > > > > Won't this affect the performance of this V5?
> > > > 
> > > > Yes, but it doesn't have to be a descriptor. Might be some internal
> > > > driver completion queue (as in the case of mlx5). And definitely does
> > > > affect performance :-( (see all the static branches to disable it)
> > > > >    $ pahole -C xsk_tx_metadata_compl
> > > > > ./drivers/net/ethernet/stmicro/stmmac/stmmac.ko
> > > > >    struct xsk_tx_metadata_compl {
> > > > > 	__u64 *              tx_timestamp;         /*     0     8 */
> > > > > 
> > > > > 	/* size: 8, cachelines: 1, members: 1 */
> > > > > 	/* last cacheline: 8 bytes */
> > > > >    };
> > > > > 
> > > > > Guess, I must be misunderstanding, as I was expecting to see the @flags
> > > > > member being preserved across, as I get the race there.
> > > > > 
> > > > > Looking at stmmac driver, it does look like this xsk_tx_metadata_compl
> > > > > is part of the TX-ring for completion (tx_skbuff_dma) and the
> > > > > tx_timestamp data is getting stored here.  How is userspace AF_XDP
> > > > > application getting access to the tx_timestamp data?
> > > > > I though this was suppose to get stored in metadata data area (umem)?
> > > > > 
> > > > > Also looking at the code, the kernel would not have a "crash" race on
> > > > > the flags member (if we preserve in struct), because the code checks the
> > > > > driver HW-TS config-state + TX-descriptor for the availability of a
> > > > > HW-TS in the descriptor.
> > > > 
> > > > xsk_tx_metadata_compl stores a pointer to the completion timestamp
> > > > in the umem, so everything still arrives via the metadata area.
> > > > 
> > > > We want to make sure the flags are not changing across tx and tx completion.
> > > > Instead of saving the flags, we just use that xsk_tx_metadata_compl to
> > > > signal to the completion that "I know that I've requested the tx
> > > > completion timestamp, please put it at this address in umem".
> > > > 
> > > > I store the pointer instead of flags to avoid doing pointer math again
> > > > at completion. But it's an implementation detail and somewhat abstracted
> > > > from the drivers (besides the fact that it's probably has to fit in 8
> > > > bytes).
> > > 
> > > I see it now (what I missed). At TX time you are storing a pointer where
> > > to (later) write the TS at completion time.  It just seems overkill to
> > > store 8 byte (pointer) to signal (via NULL) if the HWTS was requested.
> > > Space in the drivers TX-ring is performance critical, and I think driver
> > > developers would prefer to find a bit to indicate HWTS requested.
> > > 
> > > If HWTS was *NOT* requested, then the metadata area will not be updated
> > > (right, correct?). Then memory area is basically garbage that survived.
> > > How does the AF_XDP application know this packet contains a HWTS or not?
> > > 
> > >  From an UAPI PoV wouldn't it be easier to use (and extend) via keeping
> > > the @flags member (in struct xsk_tx_metadata), but (as you already do)
> > > not let kernel checks depend on it (to avoid the races).
> > 
> > I was assuming the userspace can keep this signal out of band or use
> > the same idea as suggested with padding struct xsk_tx_metadata to keep
> > some data around. But I see your point, it might be convenient to
> > keep the original flags around during completion on the uapi side.
> > 
> > I think I can just move flags from the request union member to the outer
> > struct. So the struct xsk_tx_metadata would look like:
> > 
> > struct xsk_tx_metadata {
> > 	__u32 flags; /* maybe can even make this u64? */
> > 
> 
> Yes to u64 for two reasons (1) this becomes UAPI and
> (2) better alignment for tx_timestamp.
> But I'm open to keeping it u32.
> 
> > 	union {
> > 		__u16 csum_start;
> > 		__u16 csum_offset;
> > 	} request;
> > 
> > 	union {
> > 		__u64 tx_timestamp;
> > 	} completion;
> > 
> > 	__u32 padding; /* to drop this padding */
> > };
> > 
> > But I'd also keep the existing xsk_tx_metadata_compl to carry the
> > pointer+signal around. As I mentioned previously, it's completely
> > opaque to the driver and we can change the internals in the future.
> > 
> 
> Sure, it is an implementation detail and my objections are mostly that I
> don't find it as a pretty code approach that can be hard to follow.
> Maybe driver developer will object and change this later if it cost too
> much to increase the element size in their TX-ring queues.

To make sure I understand, your preference is to save the flags, right?
A potential problem with that approach might be that we'd also have to
carry the pointer to the original umem chunk (blowing the overhead by extra
8 bytes) or pulling it of the tx completion descriptors in the drivers (extra
complexity). Pulling it out of the tx completion also might be
problematic because because we store iova/dma addresses in the
descriptors?