hfi1_do_send(9)

✨ hfi1_do_send(9) ✨

okay so this is a page from the Kernel API Manual bestie, and it's literally about hfi1_do_send — the function that, well, does a send on a QP 💅

---

🌸 NAME 🌸

hfi1_do_send — perform a send on a QP. that's it, that's the whole vibe 🫶

---

🎀 SYNOPSIS 🎀

void hfi1_do_send (struct rvt_qp *qp , bool in_thread );

---

🦋 ARGUMENTS 🦋

• qp — a pointer to the QP 👑 like, the queue pair in question, that's your main character here
• in_thread — true if you're in a workqueue thread 🌈 (so basically, "are we in a thread rn or not")

---

💖 DESCRIPTION 💖

okay so what this function does is process entries in the send work queue — and it just keeps going until either credit runs out or the queue itself is exhausted. iconic behavior, no notes ✨

and here's the important part, bestie: it's only allowed to let one CPU send a packet per QP. like, one at a time, single file, no cutting the line 🚦

why?? because otherwise 😭 two threads could send packets out of order and that's a whole mess we do NOT want. order matters here fr fr 💀

so yeah — one CPU per QP, that's the rule, periodt 💅

---

🔥 SEE ALSO 🔥

• the kernel file ./drivers/infiniband/hw/hfi1/ruc.c (go read it, iykyk 📖)
• hfi1_make_grh(9)
• build_ahg(9)
• hfi1_make_ruc_header_16B(9)
• hfi1_make_ruc_header_9B(9)
• hfi1_schedule_send_yield(9)

---

and that's the whole page, bestie 🧚 this lil gem of a function page lives over in ./drivers/infiniband/hw/hfi1 and it's dated October 2026 — so yeah, slightly from the future, no cap ⭐ mwah, go send some packets responsibly 💜