2. Marvell cnxk platform guide

This document gives an overview of Marvell OCTEON CN9K and CN10K RVU H/W block, packet flow and procedure to build DPDK on OCTEON cnxk platform.

More information about CN9K and CN10K SoC can be found at Marvell Official Website.

2.1. Supported OCTEON cnxk SoCs

  • CN106xx
  • CNF105xx

2.2. Resource Virtualization Unit architecture

The Fig. 2.3 diagram depicts the RVU architecture and a resource provisioning example.

../_images/cnxk_resource_virtualization.svg

Fig. 2.3 cnxk Resource virtualization architecture and provisioning example

Resource Virtualization Unit (RVU) on Marvell’s OCTEON CN9K/CN10K SoC maps HW resources belonging to the network, crypto and other functional blocks onto PCI-compatible physical and virtual functions.

Each functional block has multiple local functions (LFs) for provisioning to different PCIe devices. RVU supports multiple PCIe SRIOV physical functions (PFs) and virtual functions (VFs).

The Table 2.3 shows the various local functions (LFs) provided by the RVU and its functional mapping to DPDK subsystem.

Table 2.3 RVU managed functional blocks and its mapping to DPDK subsystem
# LF DPDK subsystem mapping
1 NIX rte_ethdev, rte_tm, rte_event_eth_[rt]x_adapter, rte_security
2 NPA rte_mempool
3 NPC rte_flow
4 CPT rte_cryptodev, rte_event_crypto_adapter
5 SSO rte_eventdev
6 TIM rte_event_timer_adapter
7 LBK rte_ethdev
8 DPI rte_rawdev
9 SDP rte_ethdev
10 REE rte_regexdev

PF0 is called the administrative / admin function (AF) and has exclusive privileges to provision RVU functional block’s LFs to each of the PF/VF.

PF/VFs communicates with AF via a shared memory region (mailbox).Upon receiving requests from PF/VF, AF does resource provisioning and other HW configuration.

AF is always attached to host, but PF/VFs may be used by host kernel itself, or attached to VMs or to userspace applications like DPDK, etc. So, AF has to handle provisioning/configuration requests sent by any device from any domain.

The AF driver does not receive or process any data. It is only a configuration driver used in control path.

The Fig. 2.3 diagram also shows a resource provisioning example where,

  1. PFx and PFx-VF0 bound to Linux netdev driver.
  2. PFx-VF1 ethdev driver bound to the first DPDK application.
  3. PFy ethdev driver, PFy-VF0 ethdev driver, PFz eventdev driver, PFm-VF0 cryptodev driver bound to the second DPDK application.

2.3. LBK HW Access

Loopback HW Unit (LBK) receives packets from NIX-RX and sends packets back to NIX-TX. The loopback block has N channels and contains data buffering that is shared across all channels. The LBK HW Unit is abstracted using ethdev subsystem, Where PF0’s VFs are exposed as ethdev device and odd-even pairs of VFs are tied together, that is, packets sent on odd VF end up received on even VF and vice versa. This would enable HW accelerated means of communication between two domains where even VF bound to the first domain and odd VF bound to the second domain.

Typical application usage models are,

  1. Communication between the Linux kernel and DPDK application.
  2. Exception path to Linux kernel from DPDK application as SW KNI replacement.
  3. Communication between two different DPDK applications.

2.4. SDP interface

System DPI Packet Interface unit(SDP) provides PCIe endpoint support for remote host to DMA packets into and out of cnxk SoC. SDP interface comes in to live only when cnxk SoC is connected in PCIe endpoint mode. It can be used to send/receive packets to/from remote host machine using input/output queue pairs exposed to it. SDP interface receives input packets from remote host from NIX-RX and sends packets to remote host using NIX-TX. Remote host machine need to use corresponding driver (kernel/user mode) to communicate with SDP interface on cnxk SoC. SDP supports single PCIe SRIOV physical function(PF) and multiple virtual functions(VF’s). Users can bind PF or VF to use SDP interface and it will be enumerated as ethdev ports.

The primary use case for SDP is to enable the smart NIC use case. Typical usage models are,

  1. Communication channel between remote host and cnxk SoC over PCIe.
  2. Transfer packets received from network interface to remote host over PCIe and vice-versa.

2.5. cnxk packet flow

The Fig. 2.4 diagram depicts the packet flow on cnxk SoC in conjunction with use of various HW accelerators.

../_images/cnxk_packet_flow_hw_accelerators.svg

Fig. 2.4 cnxk packet flow in conjunction with use of HW accelerators

2.6. HW Offload Drivers

This section lists dataplane H/W block(s) available in cnxk SoC.

  1. Ethdev Driver See CNXK Poll Mode driver for NIX Ethdev driver information.
  2. Mempool Driver See cnxk NPA Mempool Driver for NPA mempool driver information.
  3. Baseband PHY Driver See Marvell CNXK BPHY Driver for Baseband PHY driver information.

2.7. Procedure to Setup Platform

There are three main prerequisites for setting up DPDK on cnxk compatible board:

  1. RVU AF Linux kernel driver

    The dependent kernel drivers can be obtained from the kernel.org.

    Alternatively, the Marvell SDK also provides the required kernel drivers.

    Linux kernel should be configured with the following features enabled:

# 64K pages enabled for better performance
CONFIG_ARM64_64K_PAGES=y
CONFIG_ARM64_VA_BITS_48=y
# huge pages support enabled
CONFIG_HUGETLBFS=y
CONFIG_HUGETLB_PAGE=y
# VFIO enabled with TYPE1 IOMMU at minimum
CONFIG_VFIO_IOMMU_TYPE1=y
CONFIG_VFIO_VIRQFD=y
CONFIG_VFIO=y
CONFIG_VFIO_NOIOMMU=y
CONFIG_VFIO_PCI=y
CONFIG_VFIO_PCI_MMAP=y
# SMMUv3 driver
CONFIG_ARM_SMMU_V3=y
# ARMv8.1 LSE atomics
CONFIG_ARM64_LSE_ATOMICS=y
# OCTEONTX2 drivers
CONFIG_OCTEONTX2_MBOX=y
CONFIG_OCTEONTX2_AF=y
# Enable if netdev PF driver required
CONFIG_OCTEONTX2_PF=y
# Enable if netdev VF driver required
CONFIG_OCTEONTX2_VF=y
CONFIG_CRYPTO_DEV_OCTEONTX2_CPT=y
# Enable if OCTEONTX2 DMA PF driver required
CONFIG_OCTEONTX2_DPI_PF=n

  1. ARM64 Linux Tool Chain

    For example, the aarch64 Linaro Toolchain, which can be obtained from here.

    Alternatively, the Marvell SDK also provides GNU GCC toolchain, which is optimized for cnxk CPU.

  2. Rootfile system

    Any aarch64 supporting filesystem may be used. For example, Ubuntu 15.10 (Wily) or 16.04 LTS (Xenial) userland which can be obtained from http://cdimage.ubuntu.com/ubuntu-base/releases/16.04/release/ubuntu-base-16.04.1-base-arm64.tar.gz.

    Alternatively, the Marvell SDK provides the buildroot based root filesystem. The SDK includes all the above prerequisites necessary to bring up the cnxk board.

2.8. Debugging Options

Table 2.4 cnxk common debug options
# Component EAL log command
1 Common –log-level=’pmd.cnxk.base,8’
2 Mailbox –log-level=’pmd.cnxk.mbox,8’

2.8.1. Debugfs support

The RVU AF Linux kernel driver provides support to dump RVU blocks context or stats using debugfs.

Enable debugfs by:

  1. Compile kernel with debugfs enabled, i.e CONFIG_DEBUGFS=y.
  2. Boot OCTEON CN9K/CN10K with debugfs supported kernel.
  3. Verify debugfs mounted by default “mount | grep -i debugfs” or mount it manually by using.
# mount -t debugfs none /sys/kernel/debug

Currently debugfs supports the following RVU blocks NIX, NPA, NPC, NDC, SSO & RPM.

The file structure under /sys/kernel/debug is as follows

octeontx2/
|
cn10k/
|-- rpm
|   |-- rpm0
|   |   '-- lmac0
|   |       '-- stats
|   |-- rpm1
|   |   |-- lmac0
|   |   |   '-- stats
|   |   '-- lmac1
|   |       '-- stats
|   '-- rpm2
|       '-- lmac0
|           '-- stats
|-- cpt
|   |-- cpt_engines_info
|   |-- cpt_engines_sts
|   |-- cpt_err_info
|   |-- cpt_lfs_info
|   '-- cpt_pc
|---- nix
|   |-- cq_ctx
|   |-- ndc_rx_cache
|   |-- ndc_rx_hits_miss
|   |-- ndc_tx_cache
|   |-- ndc_tx_hits_miss
|   |-- qsize
|   |-- rq_ctx
|   '-- sq_ctx
|-- npa
|   |-- aura_ctx
|   |-- ndc_cache
|   |-- ndc_hits_miss
|   |-- pool_ctx
|   '-- qsize
|-- npc
|    |-- mcam_info
|    |-- mcam_rules
|    '-- rx_miss_act_stats
|-- rsrc_alloc
'-- sso
     |-- hws
     |   '-- sso_hws_info
     '-- hwgrp
         |-- sso_hwgrp_aq_thresh
         |-- sso_hwgrp_iaq_walk
         |-- sso_hwgrp_pc
         |-- sso_hwgrp_free_list_walk
         |-- sso_hwgrp_ient_walk
         '-- sso_hwgrp_taq_walk

RVU block LF allocation:

cat /sys/kernel/debug/cn10k/rsrc_alloc

pcifunc    NPA    NIX    SSO GROUP    SSOWS    TIM    CPT
PF1         0       0
PF4                 1
PF13                          0, 1     0, 1      0

RPM example usage:

cat /sys/kernel/debug/cn10k/rpm/rpm0/lmac0/stats

=======Link Status======

Link is UP 25000 Mbps

=======NIX RX_STATS(rpm port level)======

rx_ucast_frames: 0
rx_mcast_frames: 0
rx_bcast_frames: 0
rx_frames: 0
rx_bytes: 0
rx_drops: 0
rx_errors: 0

=======NIX TX_STATS(rpm port level)======

tx_ucast_frames: 0
tx_mcast_frames: 0
tx_bcast_frames: 0
tx_frames: 0
tx_bytes: 0
tx_drops: 0

=======rpm RX_STATS======

Octets of received packets: 0
Octets of received packets with out error: 0
Received packets with alignment errors: 0
Control/PAUSE packets received: 0
Packets received with Frame too long Errors: 0
Packets received with a1nrange length Errors: 0
Received packets: 0
Packets received with FrameCheckSequenceErrors: 0
Packets received with VLAN header: 0
Error packets: 0
Packets received with unicast DMAC: 0
Packets received with multicast DMAC: 0
Packets received with broadcast DMAC: 0
Dropped packets: 0
Total frames received on interface: 0
Packets received with an octet count < 64: 0
Packets received with an octet count == 64: 0
Packets received with an octet count of 65–127: 0
Packets received with an octet count of 128-255: 0
Packets received with an octet count of 256-511: 0
Packets received with an octet count of 512-1023: 0
Packets received with an octet count of 1024-1518: 0
Packets received with an octet count of > 1518: 0
Oversized Packets: 0
Jabber Packets: 0
Fragmented Packets: 0
CBFC(class based flow control) pause frames received for class 0: 0
CBFC pause frames received for class 1: 0
CBFC pause frames received for class 2: 0
CBFC pause frames received for class 3: 0
CBFC pause frames received for class 4: 0
CBFC pause frames received for class 5: 0
CBFC pause frames received for class 6: 0
CBFC pause frames received for class 7: 0
CBFC pause frames received for class 8: 0
CBFC pause frames received for class 9: 0
CBFC pause frames received for class 10: 0
CBFC pause frames received for class 11: 0
CBFC pause frames received for class 12: 0
CBFC pause frames received for class 13: 0
CBFC pause frames received for class 14: 0
CBFC pause frames received for class 15: 0
MAC control packets received: 0

=======rpm TX_STATS======

Total octets sent on the interface: 0
Total octets transmitted OK: 0
Control/Pause frames sent: 0
Total frames transmitted OK: 0
Total frames sent with VLAN header: 0
Error Packets: 0
Packets sent to unicast DMAC: 0
Packets sent to the multicast DMAC: 0
Packets sent to a broadcast DMAC: 0
Packets sent with an octet count == 64: 0
Packets sent with an octet count of 65–127: 0
Packets sent with an octet count of 128-255: 0
Packets sent with an octet count of 256-511: 0
Packets sent with an octet count of 512-1023: 0
Packets sent with an octet count of 1024-1518: 0
Packets sent with an octet count of > 1518: 0
CBFC(class based flow control) pause frames transmitted for class 0: 0
CBFC pause frames transmitted for class 1: 0
CBFC pause frames transmitted for class 2: 0
CBFC pause frames transmitted for class 3: 0
CBFC pause frames transmitted for class 4: 0
CBFC pause frames transmitted for class 5: 0
CBFC pause frames transmitted for class 6: 0
CBFC pause frames transmitted for class 7: 0
CBFC pause frames transmitted for class 8: 0
CBFC pause frames transmitted for class 9: 0
CBFC pause frames transmitted for class 10: 0
CBFC pause frames transmitted for class 11: 0
CBFC pause frames transmitted for class 12: 0
CBFC pause frames transmitted for class 13: 0
CBFC pause frames transmitted for class 14: 0
CBFC pause frames transmitted for class 15: 0
MAC control packets sent: 0
Total frames sent on the interface: 0

CPT example usage:

cat /sys/kernel/debug/cn10k/cpt/cpt_pc

CPT instruction requests   0
CPT instruction latency    0
CPT NCB read requests      0
CPT NCB read latency       0
CPT read requests caused by UC fills   0
CPT active cycles pc       1395642
CPT clock count pc         5579867595493

NIX example usage:

Usage: echo <nixlf> [cq number/all] > /sys/kernel/debug/cn10k/nix/cq_ctx
       cat /sys/kernel/debug/cn10k/nix/cq_ctx
echo 0 0 > /sys/kernel/debug/cn10k/nix/cq_ctx
cat /sys/kernel/debug/cn10k/nix/cq_ctx

=====cq_ctx for nixlf:0 and qidx:0 is=====
W0: base                        158ef1a00

W1: wrptr                       0
W1: avg_con                     0
W1: cint_idx                    0
W1: cq_err                      0
W1: qint_idx                    0
W1: bpid                        0
W1: bp_ena                      0

W2: update_time                 31043
W2:avg_level                    255
W2: head                        0
W2:tail                         0

W3: cq_err_int_ena              5
W3:cq_err_int                   0
W3: qsize                       4
W3:caching                      1
W3: substream                   0x000
W3: ena                                 1
W3: drop_ena                    1
W3: drop                        64
W3: bp                          0

NPA example usage:

Usage: echo <npalf> [pool number/all] > /sys/kernel/debug/cn10k/npa/pool_ctx
       cat /sys/kernel/debug/cn10k/npa/pool_ctx
echo 0 0 > /sys/kernel/debug/cn10k/npa/pool_ctx
cat /sys/kernel/debug/cn10k/npa/pool_ctx

======POOL : 0=======
W0: Stack base          1375bff00
W1: ena                 1
W1: nat_align           1
W1: stack_caching       1
W1: stack_way_mask      0
W1: buf_offset          1
W1: buf_size            19
W2: stack_max_pages     24315
W2: stack_pages         24314
W3: op_pc               267456
W4: stack_offset        2
W4: shift               5
W4: avg_level           255
W4: avg_con             0
W4: fc_ena              0
W4: fc_stype            0
W4: fc_hyst_bits        0
W4: fc_up_crossing      0
W4: update_time         62993
W5: fc_addr             0
W6: ptr_start           1593adf00
W7: ptr_end             180000000
W8: err_int             0
W8: err_int_ena         7
W8: thresh_int          0
W8: thresh_int_ena      0
W8: thresh_up           0
W8: thresh_qint_idx     0
W8: err_qint_idx        0

NPC example usage:

cat /sys/kernel/debug/cn10k/npc/mcam_info

NPC MCAM info:
RX keywidth    : 224bits
TX keywidth    : 224bits

MCAM entries   : 2048
Reserved       : 158
Available      : 1890

MCAM counters  : 512
Reserved       : 1
Available      : 511

SSO example usage:

Usage: echo [<hws>/all] > /sys/kernel/debug/cn10k/sso/hws/sso_hws_info
echo 0 > /sys/kernel/debug/cn10k/sso/hws/sso_hws_info

==================================================
SSOW HWS[0] Arbitration State      0x0
SSOW HWS[0] Guest Machine Control  0x0
SSOW HWS[0] SET[0] Group Mask[0] 0xffffffffffffffff
SSOW HWS[0] SET[0] Group Mask[1] 0xffffffffffffffff
SSOW HWS[0] SET[0] Group Mask[2] 0xffffffffffffffff
SSOW HWS[0] SET[0] Group Mask[3] 0xffffffffffffffff
SSOW HWS[0] SET[1] Group Mask[0] 0xffffffffffffffff
SSOW HWS[0] SET[1] Group Mask[1] 0xffffffffffffffff
SSOW HWS[0] SET[1] Group Mask[2] 0xffffffffffffffff
SSOW HWS[0] SET[1] Group Mask[3] 0xffffffffffffffff
==================================================

2.9. Compile DPDK

DPDK may be compiled either natively on OCTEON CN9K/CN10K platform or cross-compiled on an x86 based platform.

2.9.1. Native Compilation

meson build
ninja -C build

2.9.2. Cross Compilation

Refer to Cross compiling DPDK for aarch64 and aarch32 for generic arm64 details.

meson build --cross-file config/arm/arm64_cn10k_linux_gcc
ninja -C build

Note

By default, meson cross compilation uses aarch64-linux-gnu-gcc toolchain, if Marvell toolchain is available then it can be used by overriding the c, cpp, ar, strip binaries attributes to respective Marvell toolchain binaries in config/arm/arm64_cn10k_linux_gcc file.