Received-SPF: pass (google.com: domain of linux-kernel+bounces-31683-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
From: Yury Norov <yury.norov@gmail.com>
To: Andrew Morton <akpm@linux-foundation.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Ming Lei <ming.lei@redhat.com>,
	linux-kernel@vger.kernel.org
Cc: Yury Norov <yury.norov@gmail.com>,
	Andy Shevchenko <andriy.shevchenko@linux.intel.com>,
	Breno Leitao <leitao@debian.org>,
	Nathan Chancellor <nathan@kernel.org>,
	Rasmus Villemoes <linux@rasmusvillemoes.dk>,
	Zi Yan <ziy@nvidia.com>
Subject: [PATCH v5 0/9] lib/group_cpus: rework grp_spread_init_one() and make it O(1)
Date: Fri, 19 Jan 2024 18:50:44 -0800
Message-Id: <20240120025053.684838-1-yury.norov@gmail.com>
Precedence: bulk
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit

grp_spread_init_one() implementation is sub-optimal because it
traverses bitmaps from the beginning, instead of picking from the
previous iteration.

Fix it and use find_bit API where appropriate. While here, optimize
cpumasks allocation and drop unneeded cpumask_empty() call.

---
v1: https://lore.kernel.org/all/ZW5MI3rKQueLM0Bz@yury-ThinkPad/T/
v2: https://lore.kernel.org/lkml/ZXKNVRu3AfvjaFhK@fedora/T/
v3: https://lore.kernel.org/lkml/20231212042108.682072-7-yury.norov@gmail.com/T/
v4: https://lore.kernel.org/lkml/20231228200936.2475595-1-yury.norov@gmail.com/T/
v5: add CPUMASK_NULL macro and use it to initialize cpumask_var_t
    variables properly.

On cpumask_var_t initialization issue:

The idea of having different types behind the same typedef has been
considered nasty for quite a while. See a comment in include/linux/cpumask.h
for example.

Now that I'm trying to adopt kernel cleanup machinery to cpumasks, it
reveals another disadvantage of this approach - there's no way to assign
a cpumask_var_t variable at declaration time, which is required by
cleanup implementation.

To fix that, in v5 I added a CPUMASK_NULL macro as a workaround. This
CPUMASK_NULL would be also useful for those converting existing codebase
to enable cleanup variables.

On a long term, it's better to drop CPUMASK_OFFSTACK entirely. Moreover,
it's used only on Power and x86 machines if NR_CPUS >= 8K (unless people
enable it explicitly, and nobody bothers doing that in a real life). But
it requires some more discussions with Power and x64 people...

Meanwhile, I'm going to submit a patchset that deprecates cpumask_var_t,
and adds a new set of allocators which would support initialization at
declaration time.

Yury Norov (9):
  cpumask: introduce for_each_cpu_and_from()
  lib/group_cpus: optimize inner loop in grp_spread_init_one()
  lib/group_cpus: relax atomicity requirement in grp_spread_init_one()
  lib/group_cpus: optimize outer loop in grp_spread_init_one()
  lib/group_cpus: don't zero cpumasks in group_cpus_evenly() on
    allocation
  lib/group_cpus: drop unneeded cpumask_empty() call in
    __group_cpus_evenly()
  cpumask: define cleanup function for cpumasks
  lib/group_cpus: rework group_cpus_evenly()
  lib/group_cpus: simplify group_cpus_evenly() for more

 include/linux/cpumask.h |  16 ++++++
 include/linux/find.h    |   3 ++
 lib/group_cpus.c        | 110 ++++++++++++++++------------------------
 3 files changed, 62 insertions(+), 67 deletions(-)

-- 
2.40.1