Hailo 8 M2 on Ubuntu 25.10, PCIE 4.21 question

I’m running Ubuntu 25.10 server after an apt upgrade from 25.04, as of a few days ago, using secure boot. I understand this is not yet a upported version, but thought I’d ask this anyway. Kernel version in 25.10 is 6.17.0-6-generic. Also using Frigate in docker, current release of which requires the 4.21 version of the hailo driver. I built the 4.21 driver manually on Ubuntu server 25.04 and use the “install_dkms” command which worked fine. During the upgrade process to 25.10, DKMS seems to have worked on the hailo driver based on the upgrade log. But now at run time with Frigate up using a camera and Hailo for detecting, dmesg is getting lots of these message blocks and I don’t know what it indicates:

[256836.156632] hailo_vdma_buffer_map+0xac/0x560 [hailo_pci]

[256836.156640] hailo_vdma_buffer_map_ioctl+0xb3/0x1a0 [hailo_pci]

[256836.156643] hailo_vdma_ioctl+0x10a/0x340 [hailo_pci]

[256836.156646] hailo_pcie_fops_unlockedioctl+0x195/0x4b0 [hailo_pci]

[256836.156648] ? hailo_vdma_buffer_map+0x343/0x560 [hailo_pci]

[256836.156668] ? hailo_pcie_fops_unlockedioctl+0x24b/0x4b0 [hailo_pci]

This block repeats constantly, possibly every time Frigate is using /dev/hailo0? If someone knows what these indicate and if there is a way to turn them off or fixing the cause, I’d appreciate any info. Thanks in advance!

Hey @Steven_Olson,

This is usually a kernel mismatch with the driver. Sometimes it works and you’ll still get these errors, and sometimes it won’t work at all.

This is a known issue. When there’s a mismatch, we’re working on a fix/workaround to silence and suppress these errors so they don’t flood the dmesg.

The functions listed (hailo_vdma_buffer_map, hailo_vdma_buffer_map_ioctl, etc.) are responsible for mapping user-space buffers for DMA transfers between the host and the Hailo device [hailo_vdma_buffer_map_ioctl source].

Other possible but less likely causes:

  • Secure Boot: Secure Boot can interfere with out-of-tree kernel modules like the Hailo PCIe driver unless the module is properly signed and enrolled. However, since your driver appears to load, this may not be the direct cause.

Thanks, good info! Frigate seems to be doing detections fine with the driver, so spamming dmesg seems to be the only problem. During the upgrade dkms logged success messages, so I’m pretty sure the MOK got used to sign the driver correctly, cuz it loads fine with Secure Boot enabled. Later I may try a make all and make install_dkms just to see if anything changes with a manual rebuild/reinstall compared to what dkms did during the release upgrade.

Thanks again for the quick and thorough response!

1 Like