rdma_set_ack_timeout(9)

💅 rdma_set_ack_timeout(9) 💅

ok so you're literally reading a Kernel API Manual page, bestie, and it's about rdma_set_ack_timeout — a whole little function that sets the ack timeout of a QP that's associated with a connection identifier. yes that's a lot of acronyms, we'll get there 🫶

---

✨ NAME ✨

rdma_set_ack_timeout — Set the ack timeout of QP associated with a connection identifier.

🌸 SYNOPSIS 🌸

int rdma_set_ack_timeout (struct rdma_cm_id *id , u8 timeout );

like, that's the whole vibe. two args, one int coming back at you. periodt. ✨

🎀 ARGUMENTS 🎀

• id — the communication identifier to associated with service type. (their words, not mine 😭 — it's the rdma_cm_id you're tying this to.)
• timeout — the ack timeout you're setting on a QP, and it's expressed as 4.096 * 2^(timeout) usec. yes, that's the actual formula, no cap, write it on your hand 💜

🦋 DESCRIPTION 🦋

okay so timing matters SO much here, like main character energy for real:

This function should be called before rdma_connect on the active side, and on the passive side before rdma_accept. ⏰

It is applicable to primary path only — don't get cute with it elsewhere 🙈

The timeout will affect the local side of the QP. it is not negotiated with the remote side, and zero disables the timer. so yeah, 0 = timer off, iconic little detail 💅

And — this is the juicy part — in case it is set before rdma_resolve_route, the value will also be used to determine PacketLifeTime for RoCE. two birds, one syscall, we love efficiency 🐦✨

⭐ RETURN ⭐

0 for success. that's it. that's the whole return section, slay 💖

🌈 SEE ALSO 🌈

Kernel file ./drivers/infiniband/core/cma.c — and a whole friend group of functions you should also go read:

• rdma_is_consumer_reject(9)
• rdma_iw_cm_id(9)
• cma_acquire_dev_by_src_ip(9)
• cma_ib_acquire_dev(9)
• rdma_set_min_rnr_timer(9)
• rdma_connect_locked(9)
• rdma_connect(9)
• rdma_connect_ece(9)
• rdma_accept(9)

---

ngl bestie, if you're touching RoCE paths, double-check the ordering rules above before you ship. the "before rdma_connect / before rdma_accept" thing is not a suggestion, it's giving load-bearing 🫶

aaaaand that's the page!! 🌸 signed off from ./drivers/infiniband/core, dated October 2026 — go set some timeouts, you absolute icon 👑💜