[PATCH 00/15] Introduce HabanaLabs network drivers
Alexander Lobakin
aleksander.lobakin at intel.com
Mon Jun 17 12:34:52 UTC 2024
From: Omer Shpigelman <oshpigelman at habana.ai>
Date: Thu, 13 Jun 2024 11:21:53 +0300
> This patch set implements the HabanaLabs network drivers for Gaudi2 ASIC
> which is designed for scaling of AI neural networks training.
> The patch set includes the common code which is shared by all Gaudi ASICs
> and the Gaudi2 ASIC specific code. Newer ASICs code will be followed.
> All of these network drivers are modeled as an auxiliary devices to the
> parent driver.
>
> The newly added drivers are Core Network (CN), Ethernet and InfiniBand.
> All of these drivers are based on the existing habanalabs driver which
> serves as the compute driver and the entire platform.
> The habanalabs driver probes the network drivers which configure the
> relevant NIC HW of the device. In addition, it continuously communicates
> with the CN driver for providing some services which are not NIC specific
> e.g. PCI, MMU, FW communication etc.
>
> See the drivers scheme at:
> Documentation/networking/device_drivers/ethernet/intel/hbl.rst
>
> The CN driver is both a parent and a son driver. It serves as the common
> layer of many shared operations that are required by both EN and IB
> drivers.
>
> The Gaudi2 NIC HW is composed of 48 physical lanes, 56Gbps each. Each pair
> of lanes represent a 100Gbps logical port.
>
> The NIC HW was designed specifically for scaling AI training.
> Hence it basically functions as a regular NIC device but it is tuned for
> its dedicated purpose. As a result, the NIC HW supports Ethernet traffic
> and RDMA over modified ROCEv2 protocol.
> For example, with respect to the IB driver, the HW supports a single
> context and a single PD. The reason for this is that the operational use
> case of AI training for Gaudi2 consists of a single user
> application/process.
> Another example related to the IB driver is the lack of MR since a single
> application/process can share the entire MMU with the compute device.
> Moreover, the memory allocation of user data buffers which are used for
> RDMA communication is done via the habanalabs compute driver uAPI.
> With respect to the Ethernet driver, since the Ethernet flow is used
> mainly for control, the HW is not performance tuned e.g. it assumes a
> contiguous memory for the Rx buffers. Thus the EN driver needs to copy the
> Rx packets from the Rx buffer into the skb memory.
>
> The first 8 patches implement the CN driver.
> The next 2 patches implement the EN driver.
> The next 2 patches implement the IB driver.
> The last 3 patches modify the compute driver to support the CN driver.
>
> The patches are rebased on v6.10-rc3 tag:
> https://github.com/torvalds/linux/releases/tag/v6.10-rc3
>
> The patches are also available at:
> https://github.com/HabanaAI/drivers.gpu.linux-nic.kernel/tree/hbl_next
>
> The user-mode of the driver is being reviewed at:
> https://github.com/linux-rdma/rdma-core/pull/1472
>
> Any feedback, comment or question is welcome.
>
> Thanks,
> Omer
>
> Omer Shpigelman (15):
> net: hbl_cn: add habanalabs Core Network driver
> net: hbl_cn: memory manager component
> net: hbl_cn: physical layer support
> net: hbl_cn: QP state machine
> net: hbl_cn: memory trace events
> net: hbl_cn: debugfs support
> net: hbl_cn: gaudi2: ASIC register header files
> net: hbl_cn: gaudi2: ASIC specific support
> net: hbl_en: add habanalabs Ethernet driver
> net: hbl_en: gaudi2: ASIC specific support
> RDMA/hbl: add habanalabs RDMA driver
> RDMA/hbl: direct verbs support
> accel/habanalabs: network scaling support
> accel/habanalabs/gaudi2: CN registers header files
> accel/habanalabs/gaudi2: network scaling support
>
> .../ABI/testing/debugfs-driver-habanalabs_cn | 195 +
> .../device_drivers/ethernet/index.rst | 1 +
> .../device_drivers/ethernet/intel/hbl.rst | 82 +
> MAINTAINERS | 33 +
> drivers/accel/habanalabs/Kconfig | 1 +
> drivers/accel/habanalabs/Makefile | 3 +
> drivers/accel/habanalabs/cn/Makefile | 2 +
> drivers/accel/habanalabs/cn/cn.c | 815 +
> drivers/accel/habanalabs/cn/cn.h | 133 +
> .../habanalabs/common/command_submission.c | 2 +-
> drivers/accel/habanalabs/common/device.c | 23 +
> drivers/accel/habanalabs/common/firmware_if.c | 20 +
> drivers/accel/habanalabs/common/habanalabs.h | 43 +-
> .../accel/habanalabs/common/habanalabs_drv.c | 37 +-
> .../habanalabs/common/habanalabs_ioctl.c | 2 +
> drivers/accel/habanalabs/common/memory.c | 123 +
> drivers/accel/habanalabs/gaudi/gaudi.c | 14 +-
> drivers/accel/habanalabs/gaudi2/Makefile | 2 +-
> drivers/accel/habanalabs/gaudi2/gaudi2.c | 440 +-
> drivers/accel/habanalabs/gaudi2/gaudi2P.h | 41 +-
> drivers/accel/habanalabs/gaudi2/gaudi2_cn.c | 424 +
> drivers/accel/habanalabs/gaudi2/gaudi2_cn.h | 42 +
> .../habanalabs/gaudi2/gaudi2_coresight.c | 145 +-
> .../accel/habanalabs/gaudi2/gaudi2_security.c | 16 +-
> drivers/accel/habanalabs/goya/goya.c | 6 +
> .../include/gaudi2/asic_reg/gaudi2_regs.h | 10 +-
> .../include/gaudi2/asic_reg/nic0_phy_regs.h | 59 +
> .../nic0_qm0_axuser_nonsecured_regs.h | 61 +
> .../include/gaudi2/asic_reg/nic0_qpc1_regs.h | 905 +
> .../include/gaudi2/asic_reg/nic0_rxe0_regs.h | 725 +
> .../include/gaudi2/asic_reg/nic0_rxe1_regs.h | 725 +
> .../include/gaudi2/asic_reg/nic0_txe0_regs.h | 529 +
> .../include/gaudi2/asic_reg/nic0_txs0_regs.h | 289 +
> .../include/hw_ip/nic/nic_general.h | 15 +
> drivers/infiniband/Kconfig | 1 +
> drivers/infiniband/hw/Makefile | 1 +
> drivers/infiniband/hw/hbl/Kconfig | 18 +
> drivers/infiniband/hw/hbl/Makefile | 12 +
> drivers/infiniband/hw/hbl/hbl.h | 326 +
> drivers/infiniband/hw/hbl/hbl_encap.c | 216 +
> drivers/infiniband/hw/hbl/hbl_main.c | 493 +
> drivers/infiniband/hw/hbl/hbl_query_port.c | 96 +
> drivers/infiniband/hw/hbl/hbl_set_port_ex.c | 96 +
> drivers/infiniband/hw/hbl/hbl_usr_fifo.c | 252 +
> drivers/infiniband/hw/hbl/hbl_verbs.c | 2686 +
> drivers/net/ethernet/intel/Kconfig | 38 +
> drivers/net/ethernet/intel/Makefile | 2 +
> drivers/net/ethernet/intel/hbl_cn/Makefile | 14 +
> .../net/ethernet/intel/hbl_cn/common/Makefile | 3 +
> .../net/ethernet/intel/hbl_cn/common/hbl_cn.c | 5984 ++
> .../net/ethernet/intel/hbl_cn/common/hbl_cn.h | 1666 +
> .../intel/hbl_cn/common/hbl_cn_debugfs.c | 1457 +
> .../ethernet/intel/hbl_cn/common/hbl_cn_drv.c | 240 +
> .../intel/hbl_cn/common/hbl_cn_memory.c | 368 +
> .../ethernet/intel/hbl_cn/common/hbl_cn_phy.c | 234 +
> .../ethernet/intel/hbl_cn/common/hbl_cn_qp.c | 491 +
> .../net/ethernet/intel/hbl_cn/gaudi2/Makefile | 3 +
> .../asic_reg/arc_farm_kdma_ctx_axuser_masks.h | 135 +
> .../asic_reg/dcore0_sync_mngr_objs_regs.h | 43543 +++++++++++++++
> .../asic_reg/gaudi2_blocks_linux_driver.h | 45068 ++++++++++++++++
I don't think adding generated register defs etc. is a good idea.
You just bloat the kernel code while most of the values are not used.
When I work with HW and need to use some register defs, I add them
manually one-by-one only when needed. I know it takes more time than to
just add a whole generated reg file, but we don't need tens of thousand
unused locs in the kernel.
Please add only the actually used definitions. This applies to every
file from the series.
> .../hbl_cn/gaudi2/asic_reg/gaudi2_regs.h | 77 +
> .../asic_reg/nic0_mac_ch0_mac_128_masks.h | 339 +
> .../asic_reg/nic0_mac_ch0_mac_128_regs.h | 101 +
> .../asic_reg/nic0_mac_ch0_mac_pcs_masks.h | 713 +
> .../asic_reg/nic0_mac_ch0_mac_pcs_regs.h | 271 +
> .../asic_reg/nic0_mac_ch1_mac_pcs_regs.h | 271 +
> .../asic_reg/nic0_mac_ch2_mac_pcs_regs.h | 271 +
> .../asic_reg/nic0_mac_ch3_mac_pcs_regs.h | 271 +
> .../nic0_mac_glob_stat_control_reg_masks.h | 67 +
> .../nic0_mac_glob_stat_control_reg_regs.h | 37 +
> .../asic_reg/nic0_mac_glob_stat_rx0_regs.h | 93 +
> .../asic_reg/nic0_mac_glob_stat_rx2_regs.h | 93 +
> .../asic_reg/nic0_mac_glob_stat_tx0_regs.h | 75 +
> .../asic_reg/nic0_mac_glob_stat_tx2_regs.h | 75 +
> .../gaudi2/asic_reg/nic0_mac_rs_fec_regs.h | 157 +
> .../hbl_cn/gaudi2/asic_reg/nic0_phy_masks.h | 77 +
> .../hbl_cn/gaudi2/asic_reg/nic0_phy_regs.h | 59 +
> .../nic0_qm0_axuser_nonsecured_regs.h | 61 +
> .../asic_reg/nic0_qpc0_axuser_cong_que_regs.h | 61 +
> .../asic_reg/nic0_qpc0_axuser_db_fifo_regs.h | 61 +
> .../asic_reg/nic0_qpc0_axuser_err_fifo_regs.h | 61 +
> .../nic0_qpc0_axuser_ev_que_lbw_intr_regs.h | 61 +
> .../asic_reg/nic0_qpc0_axuser_qpc_req_regs.h | 61 +
> .../asic_reg/nic0_qpc0_axuser_qpc_resp_regs.h | 61 +
> .../asic_reg/nic0_qpc0_axuser_rxwqe_regs.h | 61 +
> .../nic0_qpc0_axuser_txwqe_lbw_qman_bp_regs.h | 61 +
> .../nic0_qpc0_dbfifo0_ci_upd_addr_regs.h | 27 +
> .../nic0_qpc0_dbfifosecur_ci_upd_addr_regs.h | 27 +
> .../hbl_cn/gaudi2/asic_reg/nic0_qpc0_masks.h | 963 +
> .../hbl_cn/gaudi2/asic_reg/nic0_qpc0_regs.h | 905 +
> .../hbl_cn/gaudi2/asic_reg/nic0_qpc1_regs.h | 905 +
> .../gaudi2/asic_reg/nic0_rxb_core_masks.h | 459 +
> .../gaudi2/asic_reg/nic0_rxb_core_regs.h | 665 +
> .../nic0_rxe0_axuser_axuser_cq0_regs.h | 61 +
> .../nic0_rxe0_axuser_axuser_cq1_regs.h | 61 +
> .../hbl_cn/gaudi2/asic_reg/nic0_rxe0_masks.h | 705 +
> .../hbl_cn/gaudi2/asic_reg/nic0_rxe0_regs.h | 725 +
> .../asic_reg/nic0_rxe0_wqe_aruser_regs.h | 61 +
> .../hbl_cn/gaudi2/asic_reg/nic0_rxe1_regs.h | 725 +
> .../gaudi2/asic_reg/nic0_serdes0_masks.h | 7163 +++
> .../gaudi2/asic_reg/nic0_serdes0_regs.h | 1679 +
> .../gaudi2/asic_reg/nic0_serdes1_regs.h | 1679 +
> .../asic_reg/nic0_tmr_axuser_tmr_fifo_regs.h | 61 +
> .../nic0_tmr_axuser_tmr_free_list_regs.h | 61 +
> .../asic_reg/nic0_tmr_axuser_tmr_fsm_regs.h | 61 +
> .../hbl_cn/gaudi2/asic_reg/nic0_tmr_masks.h | 361 +
> .../hbl_cn/gaudi2/asic_reg/nic0_tmr_regs.h | 183 +
> .../hbl_cn/gaudi2/asic_reg/nic0_txb_regs.h | 167 +
> .../hbl_cn/gaudi2/asic_reg/nic0_txe0_masks.h | 759 +
> .../hbl_cn/gaudi2/asic_reg/nic0_txe0_regs.h | 529 +
> .../hbl_cn/gaudi2/asic_reg/nic0_txs0_masks.h | 555 +
> .../hbl_cn/gaudi2/asic_reg/nic0_txs0_regs.h | 289 +
> .../nic0_umr0_0_completion_queue_ci_1_regs.h | 27 +
> .../nic0_umr0_0_unsecure_doorbell0_regs.h | 31 +
> .../nic0_umr0_0_unsecure_doorbell1_regs.h | 31 +
> .../gaudi2/asic_reg/prt0_mac_core_masks.h | 137 +
> .../gaudi2/asic_reg/prt0_mac_core_regs.h | 67 +
> .../ethernet/intel/hbl_cn/gaudi2/gaudi2_cn.c | 5689 ++
> .../ethernet/intel/hbl_cn/gaudi2/gaudi2_cn.h | 427 +
> .../intel/hbl_cn/gaudi2/gaudi2_cn_debugfs.c | 319 +
> .../intel/hbl_cn/gaudi2/gaudi2_cn_eq.c | 732 +
> .../intel/hbl_cn/gaudi2/gaudi2_cn_phy.c | 2743 +
> drivers/net/ethernet/intel/hbl_en/Makefile | 12 +
> .../net/ethernet/intel/hbl_en/common/Makefile | 3 +
> .../net/ethernet/intel/hbl_en/common/hbl_en.c | 1170 +
> .../net/ethernet/intel/hbl_en/common/hbl_en.h | 208 +
> .../intel/hbl_en/common/hbl_en_dcbnl.c | 101 +
> .../ethernet/intel/hbl_en/common/hbl_en_drv.c | 211 +
> .../intel/hbl_en/common/hbl_en_ethtool.c | 452 +
> .../net/ethernet/intel/hbl_en/gaudi2/Makefile | 2 +
> .../ethernet/intel/hbl_en/gaudi2/gaudi2_en.c | 728 +
> .../ethernet/intel/hbl_en/gaudi2/gaudi2_en.h | 53 +
> .../intel/hbl_en/gaudi2/gaudi2_en_dcbnl.c | 32 +
> include/linux/habanalabs/cpucp_if.h | 125 +-
> include/linux/habanalabs/hl_boot_if.h | 9 +-
> include/linux/net/intel/cn.h | 474 +
> include/linux/net/intel/cn_aux.h | 298 +
> include/linux/net/intel/cni.h | 636 +
> include/linux/net/intel/gaudi2.h | 432 +
> include/linux/net/intel/gaudi2_aux.h | 94 +
> include/trace/events/habanalabs_cn.h | 116 +
> include/uapi/drm/habanalabs_accel.h | 10 +-
> include/uapi/rdma/hbl-abi.h | 204 +
> include/uapi/rdma/hbl_user_ioctl_cmds.h | 66 +
> include/uapi/rdma/hbl_user_ioctl_verbs.h | 106 +
> include/uapi/rdma/ib_user_ioctl_verbs.h | 1 +
> 146 files changed, 148514 insertions(+), 70 deletions(-)
So most of these new lines are generated register definitions. The
series can be several times smaller if you follow my advice.
Thanks,
Olek
More information about the dri-devel
mailing list