2389144_en-US

[#1]

QorIQ T1024: Linux Ethernet driver stucks on high traffic

Hello,

I have a custom board with QorIQ T1024 processor running Linux Kernel version 5.4.3 (source https://github.com/nxp-qoriq/linux). I am now trying to migrate to the kernel version 6.6.52 (branch lf-6.6.y). The new kernel boots fine and all work as expected except Ethernet connection (100Mb, duplex full, autoneg off). Normal traffic using SSH or SFTP (transfer of large files) works. However, when I generate high traffic the connection stops working until reboot. Note that, this happens only in TX direction. RX direction works fine also on high traffic when running iperf test in direction PC -> target. I did not experience this issue on 5.4.3 kernel. When I run iperf test in direction target -> PC with limited bitrate 1Mb/s, the eth connection survives but full bitrate causes that functional Ethernet connection immediately stops working and stucks so I can't send data anymore nor ping the PC. The interface still appears as UP (ip link command). When I attempt to get interface statistics (ethtool -S ), ethtool hangs. I've learnt that the kernel driver is stuck and does not respond. When I attempt to turn off the interface, kernel messages are outputted - see attached ip_link_driver_busy.png

On high traffic, when it stucks, bit PBSY_DF is set in register FMQM_PnS and does not get cleared. Documentation says that it means response from Qman did not arrive. It appears like on high traffic, there is some problem with allocation/deallocation of resources in TX direction and it remains in locked state. The driver seems to be stucked while waiting for HW, a lock or a system resource. The only way to recover is reboot. Setting the interface down and up does not help.

I have upgraded Fman microcode to fsl_fman_ucode_t1024_r1.0_108_4_9.bin without any effect.

• Is QorIQ T1024 still supported in 6.6.y kernels? Are you aware of any changes incompatible with T1024?
• Are the FSL SDK DPAA and FSL SDK FMAN drivers still maintained in 6.6.y kernels? Are there any changes between version 5.4.3 and 6.6.52 that might break the functionality on T1024?

Kernel sources I am using:
Old: https://github.com/nxp-qoriq/linux/tree/134788b16485dd9fa81988681d2365ee38633fa2
New: https://github.com/nxp-qoriq/linux/tree/e0f9e2afd4cff3f02d71891244b4aa5899dfc786

Snippet of my kernel 6.6.52 config:

#
# Frame Manager support
#
CONFIG_FSL_SDK_FMAN=y
# CONFIG_FSL_SDK_FMAN_TEST is not set

#
# FMAN Processor support
#
# CONFIG_FMAN_P3040_P4080_P5020 is not set
# CONFIG_FMAN_P1023 is not set
# CONFIG_FMAN_V3H is not set
CONFIG_FMAN_V3L=y
# end of FMAN Processor support

# CONFIG_FSL_SDK_FMAN_RTC_API is not set
# CONFIG_FMAN_MIB_CNT_OVF_IRQ_EN is not set
CONFIG_FSL_FM_MAX_FRAME_SIZE=1522
CONFIG_FSL_FM_RX_EXTRA_HEADROOM=64
# CONFIG_FMAN_PFC is not set
# end of Frame Manager support

CONFIG_FSL_SDK_DPAA_ETH=y
# CONFIG_FSL_DPAA_HOOKS is not set
CONFIG_FSL_DPAA_OFFLINE_PORTS=y
CONFIG_FSL_DPAA_ADVANCED_DRIVERS=y
# CONFIG_FSL_DPAA_ETH_JUMBO_FRAME is not set
# CONFIG_FSL_DPAA_TS is not set
# CONFIG_FSL_DPAA_1588 is not set
CONFIG_FSL_DPAA_ETH_MAX_BUF_COUNT=1024
CONFIG_FSL_DPAA_ETH_REFILL_THRESHOLD=512
CONFIG_FSL_DPAA_CS_THRESHOLD_1G=0x06000000
CONFIG_FSL_DPAA_CS_THRESHOLD_10G=0x10000000
CONFIG_FSL_DPAA_INGRESS_CS_THRESHOLD=0x10000000
CONFIG_FSL_DPAA_ETH_DEBUGFS=y
CONFIG_FSL_DPAA_ETH_DEBUG=y
# CONFIG_FSL_DPAA_DBG_LOOP is not set

Appreciate any help.

Thank you,
Jiri

QorIQ T1 DevicesRe: QorIQ T1024: Linux Ethernet driver stucks on high traffic

Hello,

Based on the support status of the QorIQ Power Architecture T-series software, T1024 should not be considered officially supported on the newer lf-6.6.y kernel branch .

NXP’s official SDK support for T1024/T-series ended with the older QorIQ SDK line. The documented guidance says there is no SDK release for T1024 after SDK 2.0-1703 , and SDK 2.0-1703 used kernel 4.1.35 . NXP guidance also states that newer Linux BSPs may still contain T1024-related files, but they are not released to officially support T-series . Another NXP support statement says there is no official release after SDK 2.0-1703 for T-series products and no continuing SDK upgrade plan for that product line .

So the presence of T1024, DPAA, FMAN, QMan, or BMan code in lf-6.6.y should not be interpreted as NXP validation of T1024 operation on that kernel branch. The public nxp-qoriq/linux repository is a general QorIQ Linux tree, but the branch contents alone do not establish T1024 support or validation for 6.6.x

This is not a PHY/autoneg issue; the evidence points to a TX DPAA/QMan hang on an unvalidated T1024 kernel baseline, so the practical path is either stay on the working supported/validated baseline or debug the 6.6 port as custom maintenance.

Regards