Memory leak on heavy read load.

[haraldnh]

I'm running latest release (linux-6.7.y-1) on a card here, and while it runs fine when idling along, once I start pushing more data towards the card it starts leaking memory. When copying in data to the machine at around 5Gbps (which is what the disk array can handle), it consumes kernel memory at about 25% of received volume. Stopping the transfer halts the leak, but nothing is reclaimed at any point. And once it runs all out of memory, the oom killer will of course take down everything.

It's been way too long since I worked with network drivers, so I don't even know where to start looking. I would think about a tail overwrite of a ring buffer or something similar, but that's just a wild guess.

[DatPat]

It would help to know if this leak also occurs in the send-path and which kernel version you're on.

[haraldnh]

It's in the receive path only.

$ uname -a
Linux vmh3 6.8.12-2-pve #1 SMP PREEMPT_DYNAMIC PMX 6.8.12-2 (2024-09-05T10:03Z) x86_64 GNU/Linux

$ dmesg | grep tn40xx
[ 17.471638] tn40xx: Tehuti Network Driver from https://github.com/acooks/tn40xx-driver, linux-6.7.y-1
[ 17.471649] tn40xx: Supported phys : QT2025 TLK10232 AQR105 MUSTANG
[ 17.472139] tn40xx 0000:0b:00.0: enabling device (0140 -> 0142)
[ 17.472468] tn40xx 0000:0b:00.0: srom 0x0 HWver 16 build 0 lane# 4 max_pl 0x1 mrrs 0x5
[ 17.603586] tn40xx 0000:0b:00.0: SVID PHY type 3; MDIO scan Found 2
[ 17.603594] tn40xx 0000:0b:00.0: PHY detected ID=43A400 - QT2025 10Gbps SFP+
[ 19.365221] tn40xx 0000:0b:00.0: QT2025 FW version 2.0.3.3 module type 0x2
[ 19.469121] tn40xx 0000:0b:00.0 eth0: fw 0xe
[ 19.469152] tn40xx 0000:0b:00.0 eth0: Port A
[ 19.469361] tn40xx 0000:0b:00.0: 1 1fc9:4024:1fc9:3015
[ 19.469706] tn40xx: detected 1 cards, 1 loaded

The machine is an old HPE DL380gen9 with dual E5-2680v4 and 320G ram, talking to a Mikrotik CRS 312 on a pair of Finisair SFPs.