After being on a hiatus for more than one year and going through several rounds of Habana Labs driver maintainers due to Intel layoffs, there finally is some updated “habanalabs” AI accelerator kernel driver code slated to go into the upcoming Linux 6.18 kernel cycle. There is some new feature work but still no Gaudi 3 support for the upstream Linux kernel.
New Intel Habana Labs Linux driver co-maintainer Koby Elbaz announced in today’s pull request:
“We are excited to continue the upstreaming of the accel/habanalabs driver, based on the work previously done by Oded Gabbay and Ofir Bitton. From this cycle on, both me and Konstantin Sinyuk, will be maintaining the driver upstreaming, ensuring regular updates into drm-next.
This tag contains habanalabs driver changes for v6.18. It continues the previous upstream work from tags/drm-habanalabs-next-2024-06-23, Including improvements in debug and visibility, alongside general code cleanups, and new features such as vmalloc-backed coherent mmap, HLDIO infrastructure, etc.”
It’s been a year and a half since they last pushed updated Habana Labs driver feature code to the mainline kernel. There is now NVMe Direct I/O HLDIO, new passthrough APIs, improved hardware recovery, better error handling, and other changes with this pull request for Linux 6.18. The NVMe Direct I/O support appears to be the most exciting Habana Labs driver feature for this next kernel cycle.
Sadly though there still is no Gaudi 3 support for this mainline kernel driver. Intel for now is maintaining the Gaudi 3 AI accelerator support in their own downstream/out-of-tree driver code. One year ago I was told to expect Gaudi 3 driver code in October but that didn’t happen. With it now missing out on the Linux 6.18 merge window, we’ll be waiting until into 2026 before seeing any Gaudi 3 support in a mainline released kernel version.
In any event it’s nice seeing new feature code coming out of Intel for their Habana Labs Linux kernel driver.
