xe_exec_queu...st_fence_set(9)

πŸ“¦ version 7.3.1, in nixpkgs unstable

πŸ’… xe_exec_queu...st_fence_set β€” Kernel API Manual ✨

okay so bestie, welcome to the kernel API manual for xe_exec_queu...st_fence_set(9) πŸ’– yes the title is truncated, no that's not a typo, that's just how the manpage gods roll, iykyk πŸŽ€


🌸 NAME 🌸

xe_exec_queue_tlb_inval_last_fence_set β€” literally sets the last fence for TLB invalidation, that's the whole vibe πŸ¦‹


✨ SYNOPSIS ✨

void xe_exec_queue_tlb_inval_last_fence_set (struct xe_exec_queue *q ,
   struct xe_vm *vm , struct dma_fence *fence , unsigned int type );

return type is void, so like… she doesn't return anything, she just does it, main character energy πŸ‘‘


πŸŽ€ ARGUMENTS πŸŽ€

  • q β€” the exec queue πŸ’« like, the queue we're doing this to
  • vm β€” the VM that the engine does a bind for 🌈
  • fence β€” the fence 🫢 yes. that one.
  • type β€” either primary or media GT ⭐ pick one, bestie, those are the options

πŸ’œ DESCRIPTION πŸ’œ

so what she does is: set the last fence for the tlb invalidation type on the queue πŸ”₯ that's it, that's the job.

important part, don't skip this πŸ™ˆ β€” it increases the reference count for fence, which means when you're closing the queue you gotta call xe_exec_queue_tlb_inval_last_fence_put. not optional, not a suggestion, that's the cleanup rule, periodt πŸ’…

fr fr: set it β†’ refcount goes up β†’ put it when the queue closes. that's the contract 🧚

🌟 SEE ALSO 🌟


and that's the whole page, bestie 🌸 no cap, you now know more about TLB invalidation last fences than most people ever will πŸ†

πŸ’– from the ./drivers/gpu/drm/xe girlies, dated October 2026, signed off as xe_exec_queu...st_fence_set(9) πŸŽ€πŸ’…βœ¨

πŸ“¦ where this page comes from