Re: Issue: stmmac reset by netdev watchdog due to tx queue timeout

From: Jochen Henneberg
Date: Wed Feb 22 2023 - 11:36:00 EST

Next message: Frank Jungclaus: "[PATCH] can: esd_usb: Improve code readability by means of replacing struct esd_usb_msg with a union"
Previous message: Rob Herring: "[GIT PULL] Devicetree updates for v6.3"
In reply to: Jochen Henneberg: "Issue: stmmac reset by netdev watchdog due to tx queue timeout"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Jochen Henneberg <jh@xxxxxxxxxxxxxxxxxxxxxxxxxx> writes:

> I have been debugging an issue with the stmmac network driver and the
> Intel Elkhart Lake SoC for quite some time now. The problem comes up
> when a port forwarding is configured with iptables NAT like this:
>
> iptables -t nat -A PREROUTING -p tcp --dport 222 \
> -j DNAT --to-destination 192.168.178.134:22
> iptables -t nat -A POSTROUTING -p tcp --dst 192.168.178.134 \
> --dport 22 j SNAT --to-source 192.168.178.138
>
> If I 'ssh -p 222 192.168.178.138' the sttmac is reset after some seconds
> with:
>
> [ 553.050018] NETDEV WATCHDOG: eno1 (intel-eth-pci): transmit queue 0
> timed out
> [ 553.050048] WARNING: CPU: 2 PID: 0 at net/sched/sch_generic.c:525
> dev_watchdog+0x23a/0x250
> [ 553.050059] Modules linked in: nft_chain_nat xt_nat nf_nat
> nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 xt_tcpudp nft_compat
> nf_tables libcrc32c nfnetlink snd_seq_dummy snd_hrtimer
> snd_hda_codec_hdmi snd_hda_codec_idt snd_hda_codec_generic
> ledtrig_audio snd_sof_pci_intel_tgl snd_sof_intel_hda_common
> soundwire_intel soundwire_generic_allocation soundwire_cadence
> snd_sof_intel_hda snd_sof_pci snd_sof_xtensa_dsp snd_sof snd_sof_utils
> snd_soc_hdac_hda snd_hda_ext_core snd_soc_acpi_intel_match binfmt_misc
> snd_soc_acpi soundwire_bus snd_soc_core snd_compress ac97_bus
> snd_pcm_dmaengine snd_hda_intel snd_intel_dspcfg snd_intel_sdw_acpi
> snd_hda_codec intel_rapl_msr snd_hda_core intel_rapl_common snd_hwdep
> x86_pkg_temp_thermal intel_powerclamp snd_pcm coretemp nls_iso8859_1
> snd_seq_midi kvm_intel snd_seq_midi_event snd_rawmidi mei_hdcp i915
> kvm mei_pxp snd_seq crct10dif_pclmul ghash_clmulni_intel
> snd_seq_device sha512_ssse3 drm_buddy ttm cmdlinepart aesni_intel
> snd_timer spi_nor crypto_simd
> [ 553.050187] drm_display_helper mtd cryptd snd intel_cstate cec
> rc_core joydev intel_wmi_thunderbolt drm_kms_helper soundcore
> i2c_algo_bit mei_me syscopyarea input_leds sysfillrect mei 8250_dw
> sysimgblt igen6_edac mac_hid intel_hid acpi_pad acpi_tad sparse_keymap
> msr parport_pc ppdev lp parport drm pstore_blk ramoops pstore_zone
> reed_solomon efi_pstore ip_tables x_tables autofs4 hid_logitech_hidpp
> hid_logitech_dj hid_generic usbhid uas hid usb_storage mxl_gpy
> polynomial mmc_block dwmac_intel spi_intel_pci i2c_i801 intel_ish_ipc
> gpio_kempld i2c_kempld crc32_pclmul spi_intel intel_ishtp i2c_smbus
> stmmac intel_lpss_pci ahci sdhci_pci xhci_pci intel_lpss
> xhci_pci_renesas pcs_xpcs cqhci libahci video phylink sdhci idma64
> kempld_core i2c_scmi wmi pinctrl_elkhartlake(+)
> [ 553.050354] CPU: 2 PID: 0 Comm: swapper/2 Tainted: G D 6.2.0-rc8+
> #22
> [ 553.050360] Hardware name: Default string Default string/COMe-mEL10
> E2, BIOS MEL1R904 11/02/2022
> [ 553.050363] RIP: 0010:dev_watchdog+0x23a/0x250
> [ 553.050370] Code: 00 e9 2b ff ff ff 48 89 df c6 05 de 01 83 01 01 e8
> fb 25 f8 ff 44 89 f1 48 89 de 48 c7 c7 a8 cd c0 a6 48 89 c2 e8 24 53
> 20 00 <0f> 0b e9 1c ff ff ff 66 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 40
> 00
> [ 553.050373] RSP: 0018:ffffbfc34018ce38 EFLAGS: 00010246
> [ 553.050378] RAX: 0000000000000000 RBX: ffffa0d392b68000 RCX:
> 0000000000000000
> [ 553.050381] RDX: 0000000000000000 RSI: 0000000000000000 RDI:
> 0000000000000000
> [ 553.050384] RBP: ffffbfc34018ce68 R08: 0000000000000000 R09:
> 0000000000000000
> [ 553.050386] R10: 0000000000000000 R11: 0000000000000000 R12:
> ffffa0d392b684c8
> [ 553.050389] R13: ffffa0d392b6841c R14: 0000000000000000 R15:
> 0000000000000000
> [ 553.050392] FS: 0000000000000000(0000) GS:ffffa0d4e4300000(0000)
> knlGS:0000000000000000
> [ 553.050395] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [ 553.050398] CR2: 000055da1a873000 CR3: 000000023fc10000 CR4:
> 0000000000350ee0
> [ 553.050402] Call Trace:
> [ 553.050406] <IRQ>
> [ 553.050413] ? __pfx_dev_watchdog+0x10/0x10
> [ 553.050419] call_timer_fn+0x29/0x160
> [ 553.050425] ? __pfx_dev_watchdog+0x10/0x10
> [ 553.050429] __run_timers+0x259/0x310
> [ 553.050434] run_timer_softirq+0x1d/0x40
> [ 553.050437] __do_softirq+0xd6/0x346
> [ 553.050444] ? hrtimer_interrupt+0x11f/0x230
> [ 553.050449] __irq_exit_rcu+0xa2/0xd0
> [ 553.050455] irq_exit_rcu+0xe/0x20
> [ 553.050459] sysvec_apic_timer_interrupt+0x92/0xd0
> [ 553.050465] </IRQ>
> [ 553.050467] <TASK>
> [ 553.050469] asm_sysvec_apic_timer_interrupt+0x1b/0x20
> [ 553.050475] RIP: 0010:cpuidle_enter_state+0xde/0x6f0
> [ 553.050481] Code: 7f 1b 5a e8 14 2f 4e ff 8b 53 04 49 89 c7 0f 1f 44
> 00 00 31 ff e8 22 40 4d ff 80 7d d0 00 0f 85 eb 00 00 00 fb 0f 1f 44
> 00 00 <45> 85 f6 0f 88 12 02 00 00 4d 63 ee 49 83 fd 09 0f 87 c7 04 00
> 00
> [ 553.050484] RSP: 0018:ffffbfc340113e38 EFLAGS: 00000246
> [ 553.050489] RAX: 0000000000000000 RBX: ffffa0d4e433c930 RCX:
> 0000000000000000
> [ 553.050492] RDX: 0000000000000002 RSI: 0000000000000000 RDI:
> 0000000000000000
> [ 553.050494] RBP: ffffbfc340113e88 R08: 0000000000000000 R09:
> 0000000000000000
> [ 553.050496] R10: 0000000000000000 R11: 0000000000000000 R12:
> ffffffffa76bb700
> [ 553.050499] R13: 0000000000000003 R14: 0000000000000003 R15:
> 00000080c45910fe
> [ 553.050504] ? cpuidle_enter_state+0xce/0x6f0
> [ 553.050508] cpuidle_enter+0x2e/0x50
> [ 553.050512] do_idle+0x216/0x2a0
> [ 553.050517] cpu_startup_entry+0x1d/0x20
> [ 553.050521] start_secondary+0x122/0x160
> [ 553.050527] secondary_startup_64_no_verify+0xe5/0xeb
> [ 553.050535] </TASK>
> [ 553.050537] ---[ end trace 0000000000000000 ]---
> [ 553.050584] intel-eth-pci 0000:00:1d.1 eno1: Reset adapter.
> [ 553.064639] intel-eth-pci 0000:00:1d.1 eno1: FPE workqueue stop
> [ 553.065416] intel-eth-pci 0000:00:1d.1 eno1: Register
> MEM_TYPE_PAGE_POOL RxQ-0
> [ 553.066003] intel-eth-pci 0000:00:1d.1 eno1: Register
> MEM_TYPE_PAGE_POOL RxQ-1
> [ 553.066469] intel-eth-pci 0000:00:1d.1 eno1: Register
> MEM_TYPE_PAGE_POOL RxQ-2
> [ 553.066933] intel-eth-pci 0000:00:1d.1 eno1: Register
> MEM_TYPE_PAGE_POOL RxQ-3
> [ 553.067428] intel-eth-pci 0000:00:1d.1 eno1: Register
> MEM_TYPE_PAGE_POOL RxQ-4
> [ 553.067895] intel-eth-pci 0000:00:1d.1 eno1: Register
> MEM_TYPE_PAGE_POOL RxQ-5
> [ 553.068466] intel-eth-pci 0000:00:1d.1 eno1: Register
> MEM_TYPE_PAGE_POOL RxQ-6
> [ 553.069025] intel-eth-pci 0000:00:1d.1 eno1: Register
> MEM_TYPE_PAGE_POOL RxQ-7
> [ 553.082011] dwmac4: Master AXI performs any burst length
> [ 553.082094] intel-eth-pci 0000:00:1d.1 eno1: Enabling Safety Features
> [ 553.082138] intel-eth-pci 0000:00:1d.1 eno1: IEEE 1588-2008 Advanced
> Timestamp supported
> [ 553.082438] intel-eth-pci 0000:00:1d.1 eno1: registered PTP clock
> [ 553.082641] intel-eth-pci 0000:00:1d.1 eno1: FPE workqueue start
> [ 553.082649] intel-eth-pci 0000:00:1d.1 eno1: configuring for
> inband/sgmii link mode
> [ 553.083700] intel-eth-pci 0000:00:1d.1 eno1: Link is Up - 1Gbps/Full
> - flow control off
>
> This does not happen during normal ssh to the EHL board, this does not
> happen if I do port forwarding through userspace, e. g. with socat. And
> this does not happen if two independent network interfaces are used in
> the iptables rules. I have not observed the issue with other network
> chips with the same kernel version so I think the issue must be in the
> stmmac driver.
>
> What happens is that the OWNED bit of a DMA descriptor is not reset
> (detected in dwmac4_wrback_get_tx_status()) and the queue's txtimer is
> looping endlessly until the timeout resets everything. Sometimes the
> situation is solved before the timeout, most of the times it is not but
> i always takes long until the OWNED bit is reset. This can be reproduced
> with 100% success.
>
> I have tried to understand where the issue comes from and I think it may
> have something to do with the calls to dma_wmb() and wmb() (or more
> precisely with missing barriers) but so far I had no success to solve it
> and provide a patch.
>
> The kernel that I am running is a quite recent linux-net
> (b60417a9f2b8). I have also tried with linux-net-next without
> success. The issue can already be observed with v5.15.39.
>
> Any help or suggestion how I can debug this further would be
> appreciated. Or if somebody else can reproduce or not reproduce the
> issue on the given platform may help as well.
>
> Regards
> -Jochen

Some more details about the devices involved:

intel-eth-pci 0000:00:1d.1: enabling device (0000 -> 0002)
intel-eth-pci 0000:00:1d.1: stmmac_config_multi_msi: multi MSI enablement successful
intel-eth-pci 0000:00:1d.1: User ID: 0x51, Synopsys ID: 0x52
intel-eth-pci 0000:00:1d.1: DWMAC4/5
intel-eth-pci 0000:00:1d.1: DMA HW capability register supported
intel-eth-pci 0000:00:1d.1: RX Checksum Offload Engine supported
intel-eth-pci 0000:00:1d.1: TX Checksum insertion supported
intel-eth-pci 0000:00:1d.1: TSO supported
intel-eth-pci 0000:00:1d.1: Enable RX Mitigation via HW Watchdog Timer
intel-eth-pci 0000:00:1d.1: device MAC address 00:e0:4b:74:93:db
intel-eth-pci 0000:00:1d.1: Enabled L3L4 Flow TC (entries=2)
intel-eth-pci 0000:00:1d.1: Enabled RFS Flow TC (entries=10)
intel-eth-pci 0000:00:1d.1: Enabling HW TC (entries=256, max_off=256)
intel-eth-pci 0000:00:1d.1: TSO feature enabled
intel-eth-pci 0000:00:1d.1: Using 32 bits DMA width
Maxlinear Ethernet GPY115B stmmac-2:01: Firmware Version: 7.110 (0x876E)
Maxlinear Ethernet GPY115B stmmac-2:01: attached PHY driver (mii_bus:phy_addr=stmmac-2:01>
intel-eth-pci 0000:00:1d.1 eno1: renamed from eth0
intel-eth-pci 0000:00:1d.1 eno1: Register MEM_TYPE_PAGE_POOL RxQ-0
intel-eth-pci 0000:00:1d.1 eno1: Register MEM_TYPE_PAGE_POOL RxQ-1
intel-eth-pci 0000:00:1d.1 eno1: Register MEM_TYPE_PAGE_POOL RxQ-2
intel-eth-pci 0000:00:1d.1 eno1: Register MEM_TYPE_PAGE_POOL RxQ-3
intel-eth-pci 0000:00:1d.1 eno1: Register MEM_TYPE_PAGE_POOL RxQ-4
intel-eth-pci 0000:00:1d.1 eno1: Register MEM_TYPE_PAGE_POOL RxQ-5
intel-eth-pci 0000:00:1d.1 eno1: Register MEM_TYPE_PAGE_POOL RxQ-6
intel-eth-pci 0000:00:1d.1 eno1: Register MEM_TYPE_PAGE_POOL RxQ-7
intel-eth-pci 0000:00:1d.1 eno1: Enabling Safety Features
intel-eth-pci 0000:00:1d.1 eno1: IEEE 1588-2008 Advanced Timestamp supported
intel-eth-pci 0000:00:1d.1 eno1: registered PTP clock
intel-eth-pci 0000:00:1d.1 eno1: FPE workqueue start
intel-eth-pci 0000:00:1d.1 eno1: configuring for inband/sgmii link mode
intel-eth-pci 0000:00:1d.1 eno1: Link is Up - 1Gbps/Full - flow control off

Regards
-Jochen

Next message: Frank Jungclaus: "[PATCH] can: esd_usb: Improve code readability by means of replacing struct esd_usb_msg with a union"
Previous message: Rob Herring: "[GIT PULL] Devicetree updates for v6.3"
In reply to: Jochen Henneberg: "Issue: stmmac reset by netdev watchdog due to tx queue timeout"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]