rdma_umem_fo...ch_dma_block(9)

✨ rdma_umem_for_each_dma_block ✨

okay so this is a page from the Kernel API Manual (yes, the serious kernel docs, but we're making it cute) — it's literally a macro that walks you through memory. Let's get into it bestie 💅

---

🌸 NAME 🌸

rdma_umem_for_each_dma_block — iterate over contiguous DMA blocks of the umem

---

🎀 SYNOPSIS 🎀

rdma_umem_for_each_dma_block (umem , biter , pgsz );

---

🦋 ARGUMENTS 🦋

• umem — the umem you're iterating over, bestie ✨
• biter — -- undescribed -- 🙈 (yeah the original page literally says that, no cap, i'm not making it up)
• pgsz — the page size you wanna split the list into 🎀

---

💖 DESCRIPTION 💖

pgsz must be <= PAGE_SIZE or computed by ib_umem_find_best_pgsz — that's a hard requirement, not a suggestion, okay? 👑

The returned DMA blocks are gonna be aligned to pgsz and they span this range:

ALIGN_DOWN(umem->address, pgsz) to ALIGN(umem->address + umem->length, pgsz)

It performs exactly ib_umem_num_dma_blocks iterations. Exactly. Not one more, not one less. periodt 💀

---

🔥 SEE ALSO 🔥

• Kernel file ./include/rdma/iter.h
• rdma_block_iter_dma_address(9)
• rdma_for_each_block(9)

---

ok and that's the whole vibe — go read ./include/rdma/iter.h if you wanna see the real thing, iykyk 🧚

mwah 💋 this page lives in the kernel source tree at ./include/rdma, dated October 2026, and yes the header up top literally shows the manual name twice because that's just how these pages do it. slay 💖