amdgpu_gfx_kfd_sch_ctrl(9)

💅 amdgpu_gfx_kfd_sch_ctrl — a whole kernel API manual, but make it cute

okay so bestie, welcome to amdgpu_gfx_kfd_sch_ctrl(9), a real page from the Kernel API Manual. yes it's a kernel function. yes it's about schedulers. no you don't have to be scared. let's go 🫶

---

✨ NAME ✨

amdgpu_gfx_kfd_sch_ctrl — no cap, this is the function that controls the KFD scheduler from the KGD (that's the Graphics Driver, iykyk) 💅

---

🦋 SYNOPSIS 🦋

void amdgpu_gfx_kfd_sch_ctrl (struct amdgpu_device *adev , u32 idx ,
       bool enable );

literally the whole signature. three args, one void return, zero drama (until the NOTE section, we'll get there) ✨

---

🎀 ARGUMENTS 🎀

• adev — an amdgpu_device pointer. your main character struct 🌸
• idx — the index of the scheduler you wanna control. which one? this one. 🔥
• enable — whether to enable or disable the KFD scheduler. yes it's a bool. yes it's exactly what it sounds like. 💜

---

🌈 DESCRIPTION 🌈

so what does this thing actually do? it controls the KFD (Kernel Fusion Driver) scheduler from the KGD, bestie. it's part of the cleaner shader feature, and it plays a key role in enforcing process isolation on the GPU — which is kind of a big deal 👑

the mechanism? a reference count called kfd_sch_req_count, tracking how many requests there are to enable the KFD scheduler. here's the tea:

• when a request to enable the KFD scheduler is made, the reference count is decremented. (yes, decremented. that's what the page says. keep your eyes on it 👀)
• when the reference count reaches zero, a delayed work is scheduled to enforce isolation after a delay of GFX_SLICE_PERIOD. iconic behaviour, honestly ⏳

and when a request to disable the KFD scheduler comes in, the function goes in this exact order:

1. first it checks if the reference count is zero 🕵️
2. if it is, it cancels the delayed work for enforcing isolation, and checks if the KFD scheduler is active
3. if the KFD scheduler is active, it sends a request to stop the KFD scheduler and sets the KFD scheduler state to inactive 💀
4. then it increments the reference count ⬆️

the whole thing is synchronized using kfd_sch_mutex, so the KFD scheduler state and the reference count get updated atomically — no racing, no messy vibes, periodt 🔒

---

🙈 NOTE 🙈

listen. if the reference count is already zero when a request to enable the KFD scheduler is made, that means there's an imbalance bug somewhere. the function triggers a warning in this case. so if you see that warning, it's not a vibe, it's a bug hunt 🚨

---

🧚 SEE ALSO 🧚

the kernel file that started it all:

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

and the whole friend group, in this exact order:

• amdgpu_gfx_parse_disable_cu(9)
• amdgpu_gfx_handle_priv_fault(9)
• amdgpu_gfx_set_run_cleaner_shader(9)
• amdgpu_gfx_get_enforce_isolation(9)
• amdgpu_gfx_set_enforce_isolation(9)
• amdgpu_gfx_enforce_isolation_handler(9)
• amdgpu_gfx_enforce_isolation_wait_for_kfd(9)
• amdgpu_gfx_enforce_isolation_ring_begin_use(9)
• amdgpu_gfx_enforce_isolation_ring_end_use(9)
• amdgpu_gfx_csb_preamble_start(9)
• amdgpu_gfx_csb_data_parser(9)
• amdgpu_gfx_csb_preamble_end(9)

go read them. they're all slaying in their own way 🦋

---

💖 and that's the page!! this cutie lives in ./drivers/gpu/drm/amd/amdgpu, dated October 2026, and it's filed under amdgpu_gfx_kfd_sch_ctrl(9) in the Kernel API Manual. mwah, go enforce some isolation bestie 🌸✨🎀