drm_sched_resubmit_jobs(9)

πŸ“¦ version 7.3.1, in nixpkgs unstable

🌸 drm_sched_resubmit_jobs(9) 🌸

okay so this is the drm_sched_resubmit_jobs(9) page from the Kernel API Manual, and bestie… it’s giving deprecated relic energy. The whole function is literally called out as Deprecated, don't use in new code! so maybe don’t make it your main character. πŸ’€


🚨 NAME 🚨

drm_sched_resubmit_jobs β€” Deprecated, don't use in new code!


✨ SYNOPSIS ✨

void drm_sched_resubmit_jobs (struct drm_gpu_scheduler *sched );

πŸŽ€ ARGUMENTS πŸŽ€

  • sched β€” scheduler instance.

πŸ’… DESCRIPTION πŸ’…

Okay so the tea: re-submitting jobs was a concept AMD came up with as a cheap way to implement recovery after a job timeout. Cute idea, right? Except it turned out to be not working very well. 😭

First of all, there are many problems with the dma_fence implementation and requirements. Either the implementation is risking deadlocks with core memory management, or it’s violating documented implementation details of the dma_fence object. Both are… not the vibe. 🫠

Drivers can still save and restore their state for recovery operations, but we shouldn’t make this a general scheduler feature around the dma_fence interface. The suggested driver-side replacement is to use drm_sched_for_each_pending_job after stopping the scheduler and implement their own recovery operations. That’s the glow-up, iykyk. ✨

⚠️ Deprecated alert: don’t use this in new code! This page literally says Deprecated, don't use in new code! and the description is a whole cautionary tale about dma_fence pain. Not a β€œmaybe avoid” situation, bestie. πŸ’€

πŸ”₯ SEE ALSO πŸ”₯


mwah πŸ’‹ that’s the whole page, bestie. Source/version: ./drivers/gpu/drm/scheduler; date: October 2026. Byeee 🫢

πŸ“¦ where this page comes from