linux-kernel - [PATCH 0/2] Add 'tx_memcpy_offload' option to ntb

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <20251023072105.901707-1-den@valinux.co.jp>
Date: Thu, 23 Oct 2025 16:21:03 +0900
From: Koichiro Den <den@...inux.co.jp>
To: ntb@...ts.linux.dev,
	linux-kernel@...r.kernel.org
Cc: jdmason@...zu.us,
	dave.jiang@...el.com,
	allenbh@...il.com
Subject: [PATCH 0/2] Add 'tx_memcpy_offload' option to ntb_transport

This patch set adds an optional TX CPU memcpy offload path to ntb_transport
and does a small cleanup. The new module parameter:

    tx_memcpy_offload=<bool> (default: N)

When enabled, ntb_transport creates a per-QP kernel thread and schedules TX
memcpy_toio() work onto it.

On some platforms (e.g. Renesas R-Car S4), using a DMA engine in the TX
path does not always improve throughput, and memcpy_toio() becomes the
bottleneck on the CPU. Offloading the memcpy_toio() and the subsequent
descriptor writes to a per-QP kthread can raise the achievable throughput
by keeping the data path hot while letting the worker thread perform the
copy.

Koichiro Den (2):
  NTB: ntb_transport: Remove unused 'retries' field from ntb_queue_entry
  NTB: ntb_transport: Add 'tx_memcpy_offload' module option

 drivers/ntb/ntb_transport.c | 115 ++++++++++++++++++++++++++++++++----
 1 file changed, 102 insertions(+), 13 deletions(-)

-- 
2.48.1