get_xmit_wait_counters(9)

✨ get_xmit_wait_counters (9) ✨

okay so this is a page from the Kernel API Manual and bestie, it's giving "tiny function, big job" energy 💅 buckle up, we're doing counter math 🧮✨

---

🎀 NAME 🎀

get_xmit_wait_counters — literally converts the HFI's SendWaitCnt/SendWaitVlCnt counter from TXE cycle units into flit times 🦋 no cap, that's the whole vibe check.

---

🌸 SYNOPSIS 🌸

u64 get_xmit_wait_counters (struct hfi1_pportdata *ppd , u16 link_width
, u16 link_speed , int vl );

---

💖 ARGUMENTS 💖

• ppd — the info of the physical Hfi port 🫶 like, the main character struct here.
• link_width — width of the active link 📏
• link_speed — speed of the active link 🏃‍♀️💨
• vl — represents VL0-VL7, and also VL15 for PortVLXmitWait counters requests 👑 and if the vl value is C_VL_COUNT, then it represents a SendWaitCnt counter request. iykyk 🔥

---

⭐ RETURN ⭐

returns the SendWaitCnt/SendWaitVlCnt counter value per vl 💜 that's it, that's the tea.

---

🌈 DESCRIPTION 🌈

So what it does: it converts SendWaitCnt/SendWaitVlCnt counter from TXE cycle times to flit times ✨.

Call this function to take samples of these counters 🔍. And here's the part that's lowkey kinda iconic — the function will calculate for the previous state transition, and then update the current state at the end of the function, using ppd->prev_link_width and ppd->port_vl_xmit_wait_last into port_vl_xmit_wait_curr and link_width 🎀

ngl, that state-carried-over-from-last-time thing is the whole trick. it's literally doing the glow-up at the end of the function 🧚💫

---

🦋 SEE ALSO 🦋

Kernel file ./drivers/infiniband/hw/hfi1/mad.c — and while you're in there, go say hi to get_pkeys(9), set_pkeys(9), tx_link_width(9), hfi1_pkey_validation_pma(9), and hfi1_process_mad(9) 💖

---

bestie that's a wrap 🎬 this cutie lives at ./drivers/infiniband/hw/hfi1, dated October 2026 — go forth and sample those counters, you absolute icon 👑💅🍓