_copy_mc_to_iter(9)

💅 _copy_mc_to_iter(9) — like, the Kernel API Manual one

okay bestie, buckle up, we're reading a man page from the Kernel API Manual section 9, and honestly? this one is kind of a baddie. it's giving "low-level memory safety with main character energy" and i'm obsessed. 🦋

---

✨ NAME ✨

_copy_mc_to_iter — copy to iter, but make it safe, because it handles source memory error exceptions 💀 (yes, machine checks, we'll get there, don't panic)

---

🌸 SYNOPSIS 🌸

size_t _copy_mc_to_iter (const void *addr , size_t bytes , struct
       iov_iter *i );

that's it. three args. she's sleek. 🎀

---

🫶 ARGUMENTS 🫶

• addr — the source kernel address, like, where the data lives right now 🏠
• bytes — total transfer length, aka how much we're moving, fr fr
• i — the destination iterator, the glow-up destination, the final form ✨

---

🔥 DESCRIPTION 🔥

so here's the tea ☕ — the pmem driver deploys this function for the dax operation (dax_copy_to_iter) for dax reads (which bypass the page-cache and the block-layer, no cap, they just go straight there 💨).

and then? upon #MC (machine check, the scary one 😭), read(2) aborts and returns EIO, or returns the bytes successfully copied. so worst case she still tells you what she managed to do. iconic behavior honestly 👑

now, the main differences between this and your typical _copy_to_iter — let me list them because there are two and both matter:

• typical tail/residue handling after a fault retries the copy byte-by-byte until the fault happens again 😬 but re-triggering machine checks is potentially fatal, so this implementation instead uses source alignment and poison alignment assumptions to avoid re-triggering hardware exceptions. like, she does NOT want to poke the bear again, and honestly? valid. 🙈
• ITER_KVEC and ITER_BVEC can return short copies. compare that to copy_to_iter, where only ITER_IOVEC attempts might return a short copy. so yeah, she's built different 💅

⚠️ real talk: the byte-by-byte retry thing isn't just slow, re-triggering machine checks is potentially fatal — that's why the alignment assumptions exist. do not mess with that logic, bestie. 🚨

---

🌈 RETURN 🌈

number of bytes copied (may be 0) — yes, zero is a possible answer, and no, that's not a vibe fail, that's just the contract 🍓

---

⭐ SEE ALSO ⭐

all my faves from the kernel file ./lib/iov_iter.c, go give them a read, iykyk 💖

• _copy_from_iter_flushcache(9)
• iov_iter_folio_queue(9)
• iov_iter_xarray(9)
• iov_iter_discard(9)
• import_iovec(9)
• iov_iter_restore(9)
• iov_iter_extract_pages(9)
• iov_iter_extract_bvecs(9)

---

and that's the whole page, bestie 🎀 sourced straight from ./lib, dated October 2026, and no i did not make any of that up. mwah 💋 slay ✨