svm_migrate_copy_done(9)

πŸ“¦ version 7.3.1, in nixpkgs unstable

πŸ’– NAME πŸ’–

svm_migrate_copy_done β€” bestie, this one waits for the memory copy sdma to be done ✨

So this page lives in the Kernel API Manual (yes, that's the manual name, we're fancy now πŸ‘‘), and it's all about one little function that waits on a fence. Slay.


🌸 SYNOPSIS 🌸

int svm_migrate_copy_done (struct amdgpu_device *adev , struct dma_fence
*mfence );

That's the whole vibe. Two args, one int, zero drama πŸ’…


πŸŽ€ ARGUMENTS πŸŽ€

  • adev β€” the amdgpu device that the sdma memory copy is executing on. Like, this is the GPU doing the actual work, okay? 🫢
  • mfence β€” the migrate fence. The fence we're waiting on. Iconic little synchronization object πŸ”₯

πŸ¦‹ DESCRIPTION πŸ¦‹

Okay so the whole point: this function waits for the dma fence to be signaled. That's it. That's the job πŸ’€

And if the copy got split into multiple sdma operations, then the fence you're passing in here is the last sdma operation's fence β€” like the final boss of the copy chain πŸ‘‘


🌈 CONTEXT 🌈

Called after svm_migrate_copy_memory. That's the context, bestie. Don't call it before, that's not the vibe πŸ™ˆ


⭐ RETURN ⭐

  • 0 β€” success ✨
  • otherwise β€” an error code from dma fence signal 😭

So yeah, zero means we ate, anything else means the fence said no πŸ’”


πŸ’œ SEE ALSO πŸ’œ

Kernel file ./drivers/gpu/drm/amd/amdkfd/kfd_migrate.c

And the whole friend group of related functions, no cap:


Go read those too, iykyk 🧚

Alright bestie, that's the page πŸ“ This little icon was brought to you by ./drivers/gpu/drm/amd/amdkfd, dated October 2026, and yes β€” the page is literally called svm_migrate_copy_done(9). Periodt πŸ’…βœ¨

πŸ“¦ where this page comes from