[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <20250530-amd-hsmp-v1-0-3222bffa4008@hillion.co.uk>
Date: Fri, 30 May 2025 17:15:20 +0100
From: Jake Hillion via B4 Relay <devnull+jake.hillion.co.uk@...nel.org>
To: Naveen Krishna Chatradhi <naveenkrishna.chatradhi@....com>,
Carlos Bilbao <carlos.bilbao@...nel.org>,
Hans de Goede <hdegoede@...hat.com>,
Ilpo Järvinen <ilpo.jarvinen@...ux.intel.com>
Cc: platform-driver-x86@...r.kernel.org, linux-kernel@...r.kernel.org,
sched-ext@...a.com, Jake Hillion <jake@...lion.co.uk>,
Blaise Sanouillet <linux@...ise.sanouillet.com>
Subject: [PATCH 0/2] x86/platform/amd: better handle scheduling delays in
hsmp_send_message
hsmp_send_message currently relies in 2 places on the assumption that
usleep_range will complete in well under 100ms. This is not guaranteed,
and is prevalent when running sched_ext schedulers or possible under
other high load conditions.
These patches alter the behaviour in two ways:
1. Checks the result of `mbox_status` a final time if the sleep took us
past the timeout. This gives a useful result under test when there
are significant scheduling delays, rather than -ETIMEDOUT.
2. Removes the 100ms limit on awaiting the semaphore. This allows a
second thread to compete even when the other suffers large scheduling
delays.
Signed-off-by: Jake Hillion <jake@...lion.co.uk>
---
Jake Hillion (2):
x86/platform/amd: move final timeout check to after final sleep
x86/platform/amd: replace down_timeout with down_interruptible
drivers/platform/x86/amd/hsmp/hsmp.c | 14 ++++++--------
1 file changed, 6 insertions(+), 8 deletions(-)
---
base-commit: 44ed0f35df343d00b8d38006854f96e333104a66
change-id: 20250527-amd-hsmp-8c358985b4fb
Best regards,
--
Jake Hillion <jake@...lion.co.uk>
Powered by blists - more mailing lists