rxe_recheck_mr(9)
π rxe_recheck_mr β a kernel API manual moment, bestie
okay so this is rxe_recheck_mr, straight out of the Kernel API Manual (section 9, the kernel-functions section, iykyk) and honestly she's a little niche but she ATE π
β¨ NAME β¨
rxe_recheck_mr β revalidate that MR from its rkey and grab a reference to it. that's literally the whole gig π«Ά
πΈ SYNOPSIS πΈ
struct rxe_mr * rxe_recheck_mr (struct rxe_qp *qp , u32 rkey );
like, yes, the spacing before that comma is a little feral, but we leave code exactly as we found it, no cap π
π¦ ARGUMENTS π¦
qpβ the qp. that's it, that's the description, she's an icon of brevity β¨rkeyβ the rkey. also extremely minimal. main character energy of saying the absolute minimum π
π DESCRIPTION π
so what this code actually does is: it lets the MR get invalidated or deregistered, or, if a MW was used, it lets that MW get invalidated or deallocated. π
the assumption it's running on is lowkey huge so listen β it assumes that the access permissions, if they were originally good, are still OK, and that the mappings are unchanged. like it's trusting you not to be messy. π
TODO, and this is the important part, bestie: if someone reregisters an MR to change its size or its access permissions while the processing of an RDMA read is happening, we should kill the responder resource and complete the operation with an error. so like, that's the plan, not the current vibe. fr fr note the difference. π
β RETURN β
mr on success, else NULL. yes bestie, check for the NULL, do not skip the vibe check π
π₯ SEE ALSO π₯
Kernel file ./drivers/infiniband/sw/rxe/rxe_resp.c β go read her, she's the whole context π
kiss kiss, this page is brought to you by the fine people over at ./drivers/infiniband/sw/rxe, stamped October 2026, and honestly? iconic of them. go touch some infiniband, bestie π§ββοΈπ
π¦ where this page comes from
- nixpkgs unstable: linux-manual 7.3.1
- π girlified by deepseek-ai/DeepSeek-V4.1-Flash-1017 on : a language model's rewrite of the original, under its terms