sdma_progress(9)

💅 sdma_progress(9) — a lil' kernel API moment

okay so bestie, we're in the Kernel API Manual today and the star of the show is sdma_progress, which is lowkey just "use seq number of detect head progress" ✨ (that's the NAME line, verbatim, don't come for me).

---

🦋 SYNOPSIS 🦋

unsigned sdma_progress (struct sdma_engine *sde , unsigned seq , struct sdma_txreq *tx );

like yes she returns an unsigned, she takes three args, she's booked and busy 💅

---

🎀 ARGUMENTS 🎀

• sde — the sdma_engine to check. that's the engine we're doing a vibe check on, fr.
• seq — base seq count. the starting point, the blueprint, the origin story.
• tx — the txreq for which we need to check descriptor availability. we're literally asking "is there room for this one, bestie?"

---

✨ DESCRIPTION ✨

so the way this works: sdma_progress is used in the appropriate spot in the sleep routine to check for potential ring progress 👑

this routine gets the seqcount before queuing the iowait structure for progress. order matters, iykyk 🫶

and then — if the seqcount indicates that progress needs to be checked — re-submission is detected by checking whether the descriptor queue has enough descriptor for the txreq. that's the whole plot twist. that's the tea ☕

---

🌈 SEE ALSO 🌈

• Kernel file ./drivers/infiniband/hw/hfi1/sdma.h sdma exported routines(9)
• sdma PSM/verbs API(9)
• Infrastructure calls(9)
• struct sdma_engine(9)
• sdma_empty(9)
• sdma_running(9)
• sdma_txinit_ahg(9)
• sdma_txinit(9)
• sdma_txadd_page(9)
• sdma_txadd_daddr(9)
• sdma_txadd_kvaddr(9)
• struct sdma_map_elem(9)
• sdma_engine_progress_schedule(9)

like, read the whole friend group, bestie, they all slay in their own way 💖

---

ngl the page lives at ./drivers/infiniband/hw/hfi1, and it's dated October 2026 — yes, the future, main character energy only 😭⏰

that's the page!! go forth and check those descriptor queues, you absolute icon 🌸💅 periodt.