STRUCT MSM_FENCE_CONTEXT(9)

✨ struct msm_fence_context(9) ✨

okay so this is a page from the Kernel API Manual and it's literally all about struct msm_fence_context — the fence context for gpu 🎀 like yes bestie we're getting into the GPU's little bookkeeping era and it's giving main character energy fr fr.

---

💖 NAME 💖

struct msm_fence_context — fence context for gpu 🦋

---

🌸 SYNOPSIS 🌸

struct msm_fence_context {
    struct drm_device *dev;
    char name[32];
    unsigned context;
    unsigned index;
    uint32_t last_fence;
    uint32_t completed_fence;
    volatile uint32_t *fenceptr;
    spinlock_t spinlock;
    ktime_t next_deadline;
    uint32_t next_deadline_fence;
    struct hrtimer deadline_timer;
    struct kthread_work deadline_work;
 };

no cap, that's the whole vibe right there ⭐

---

👑 Members 👑

okay bestie, here's the cast of characters — every single one of them, no skips, we're not doing that 🙈

• dev — the drm device 💅
• name — the human readable name for this fence timeline, like a little name tag but for a timeline 🎀
• context — see dma_fence_context_alloc (that's where it comes from, iykyk)
• index — similar to context, but it's local to the msm_fence_context's, so it stays in its own lane 🦋
• last_fence — the Last assigned fence, incremented each time a fence is created on this fence context. And hear me out: if last_fence == completed_fence, there is no remaining pending work — like the queue is empty, she's done, periodt ✨
• completed_fence — The last completed fence, updated from the CPU after interrupt from GPU 🔥 so the CPU is the one writing this one, remember that
• fenceptr — The address that the GPU directly writes with completed fence seqno. This can be ahead of completed_fence. The way this works is we can peek at this to see if a fence has already signaled but the CPU hasn't gotten around to handling the irq and updating completed_fence 😭 so it's like the GPU already sent the text and the CPU hasn't opened it yet, fr fr
• spinlock — fence context spinlock 🔒 that's it, that's the whole member
• next_deadline — Time of next deadline ⏰
• next_deadline_fence — Fence value for next pending deadline. The deadline timer is canceled when this fence is signaled 🎀
• deadline_timer — tracks nearest deadline of a fence timeline and expires just before it — literally just before, not a moment later 👑
• deadline_work — work to do after deadline_timer expires 🫶 the little follow-up task, the afterparty if you will

---

💜 Description 💜

Each ringbuffer has a single fence context, with the GPU writing an incrementing fence seqno at the end of each submit 🌈 one context per ringbuffer, GPU does its little counter thing at the end of every submit — that's the whole flow bestie, slay 💅

---

⭐ SEE ALSO ⭐

Kernel file ./drivers/gpu/drm/msm/msm_fence.h 🦋 go read the header, she's iconic, she has the receipts

---

💀 real talk though: fenceptr pointing ahead of completed_fence is not a bug, it's the design — the GPU is just faster than the CPU's irq handling. don't panic when you see it 🙈

---

and that's the page!! 🎀 brought to you from ./drivers/gpu/drm/msm in October 2026 — you're welcome bestie, mwah 💖✨