amdgpu_amdkf...ction_fences(9)

📦 version 7.3.1, in nixpkgs unstable

💅 amdgpu_amdkfd_remove_all_eviction_fences(9) ✨

okay bestie so this is a page straight out of the Kernel API Manual — yes, that Kernel API Manual — and it's about yeeting eviction fences off a buffer object. no cap, it's a short one. let's get into it 🫶

🌸 NAME 🌸

amdgpu_amdkfd_remove_all_eviction_fences — Remove all eviction fences</p>. that's the whole vibe. all of them. gone. 💅

⭐ SYNOPSIS ⭐

void amdgpu_amdkfd_remove_all_eviction_fences (struct amdgpu_bo *bo );

one function, one argument, returns nothing (void). extremely low-drama. love that for us. 🎀

🦋 ARGUMENTS 🦋

  • bo — the BO where to remove the evictions fences from. literally just hand it the buffer object and it'll strip the eviction fences off it. that's it. that's the job. 🧚

🔥 DESCRIPTION 🔥

okay so this is important, listen up 👑

this function should only be used on release, when all references to the BO are already dropped. like, all of them. do not come in here early. it's giving "final cleanup only" energy.

here's why: we're removing the eviction fence from the private copy of the dma_resv object. why that one specifically? because that's the copy that gets used during release to figure out whether the BO is idle or not. so if you're doing the release-time idle check, that private copy is the main character and it needs to be fence-free. periodt. 💀

⚠️ real talk: this is a release-time-only situation. all refs dropped, or nothing. don't be out here calling it mid-life and then acting surprised. 🙈

🌈 SEE ALSO 🌈

the kernel file lives right here, bestie:

./drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c

and here's the whole friend group 🫶:

that's the whole page, bestie, you ate it up 💖 this one comes from ./drivers/gpu/drm/amd/amdgpu, dated October 2026 — go forth and remove those fences responsibly 🎀✨

📦 where this page comes from