1 files changed, 147 insertions, 0 deletions
diff --git a/Documentation/networking/xsk-tx-metadata.rst b/Documentation/networking/xsk-tx-metadata.rst
new file mode 100644
index 000000000000..df53a10ccac3
--- /dev/null
+++ b/Documentation/networking/xsk-tx-metadata.rst
@@ -0,0 +1,147 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+==================
+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`` and
+``XDP_UMEM_TX_METADATA_LEN`` flag 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)``.
+
+Note that in the original implementation the ``XDP_UMEM_TX_METADATA_LEN``
+flag was not required. Applications might attempt to create a umem
+with a flag first and if it fails, do another attempt without a flag.
+
+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.
+- ``XDP_TXMD_FLAGS_LAUNCH_TIME``: requests the device to schedule the
+  packet for transmission at a pre-determined time called launch time. The
+  value of launch time is indicated by ``launch_time`` field of
+  ``union xsk_tx_metadata``.
+
+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.
+
+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.
+
+Launch Time
+===========
+
+The value of the requested launch time should be based on the device's PTP
+Hardware Clock (PHC) to ensure accuracy. AF_XDP takes a different data path
+compared to the ETF queuing discipline, which organizes packets and delays
+their transmission. Instead, AF_XDP immediately hands off the packets to
+the device driver without rearranging their order or holding them prior to
+transmission. Since the driver maintains FIFO behavior and does not perform
+packet reordering, a packet with a launch time request will block other
+packets in the same Tx Queue until it is sent. Therefore, it is recommended
+to allocate separate queue for scheduling traffic that is intended for
+future transmission.
+
+In scenarios where the launch time offload feature is disabled, the device
+driver is expected to disregard the launch time request. For correct
+interpretation and meaningful operation, the launch time should never be
+set to a value larger than the farthest programmable time in the future
+(the horizon). Different devices have different hardware limitations on the
+launch time offload feature.
+
+stmmac driver
+-------------
+
+For stmmac, TSO and launch time (TBS) features are mutually exclusive for
+each individual Tx Queue. By default, the driver configures Tx Queue 0 to
+support TSO and the rest of the Tx Queues to support TBS. The launch time
+hardware offload feature can be enabled or disabled by using the tc-etf
+command to call the driver's ndo_setup_tc() callback.
+
+The value of the launch time that is programmed in the Enhanced Normal
+Transmit Descriptors is a 32-bit value, where the most significant 8 bits
+represent the time in seconds and the remaining 24 bits represent the time
+in 256 ns increments. The programmed launch time is compared against the
+PTP time (bits[39:8]) and rolls over after 256 seconds. Therefore, the
+horizon of the launch time for dwmac4 and dwxlgmac2 is 128 seconds in the
+future.
+
+igc driver
+----------
+
+For igc, all four Tx Queues support the launch time feature. The launch
+time hardware offload feature can be enabled or disabled by using the
+tc-etf command to call the driver's ndo_setup_tc() callback. When entering
+TSN mode, the igc driver will reset the device and create a default Qbv
+schedule with a 1-second cycle time, with all Tx Queues open at all times.
+
+The value of the launch time that is programmed in the Advanced Transmit
+Context Descriptor is a relative offset to the starting time of the Qbv
+transmission window of the queue. The Frst flag of the descriptor can be
+set to schedule the packet for the next Qbv cycle. Therefore, the horizon
+of the launch time for i225 and i226 is the ending time of the next cycle
+of the Qbv transmission window of the queue. For example, when the Qbv
+cycle time is set to 1 second, the horizon of the launch time ranges
+from 1 second to 2 seconds, depending on where the Qbv cycle is currently
+running.
+
+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``
+- ``tx-launch-time-fifo``: device supports ``XDP_TXMD_FLAGS_LAUNCH_TIME``
+
+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.