Received-SPF: pass (google.com: domain of linux-kernel+bounces-98810-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Date: Mon, 11 Mar 2024 12:12:26 +0000
From: Mark Rutland <mark.rutland@arm.com>
To: "Christoph Lameter (Ampere)" <cl@linux.com>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>,
	Catalin Marinas <catalin.marinas@arm.com>,
	"linux-pm@vger.kernel.org" <linux-pm@vger.kernel.org>,
	"Rafael J. Wysocki" <rafael@kernel.org>,
	Viresh Kumar <vireshk@kernel.org>, Will Deacon <will@kernel.org>,
	Jonathan.Cameron@huawei.com, Matteo.Carlini@arm.com,
	Valentin.Schneider@arm.com, akpm@linux-foundation.org,
	anshuman.khandual@arm.com, Eric Mackay <eric.mackay@oracle.com>,
	dave.kleikamp@oracle.com, linux-arm-kernel@lists.infradead.org,
	linux-kernel@vger.kernel.org, linux-mm@kvack.org,
	linux@armlinux.org.uk, robin.murphy@arm.com,
	vanshikonda@os.amperecomputing.com, yang@os.amperecomputing.com,
	Nishanth Menon <nm@ti.com>, Stephen Boyd <sboyd@kernel.org>
Subject: Re: [PATCH v3] ARM64: Dynamically allocate cpumasks and increase
 supported CPUs to 512
Message-ID: <Ze71H4AdY786nSvn@FVFF77S0Q05N>
References: <37099a57-b655-3b3a-56d0-5f7fbd49d7db@gentwo.org>
 <CGME20240308140130eucas1p1259c805a0b6491ce2f69c6fca0264b1f@eucas1p1.samsung.com>
 <c1f2902d-cefc-4122-9b86-d1d32911f590@samsung.com>
 <ZesmAO1jJfEjTwxd@arm.com>
 <f160a532-1d31-41a7-b8ae-de8575c395e9@samsung.com>
 <e3952dc0-ec28-e7c7-e858-c47f146c93de@linux.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <e3952dc0-ec28-e7c7-e858-c47f146c93de@linux.com>

On Fri, Mar 08, 2024 at 09:08:59AM -0800, Christoph Lameter (Ampere) wrote:
> On Fri, 8 Mar 2024, Marek Szyprowski wrote:
> 
> > > > 
> > > > It looks that cpufreq-dt and/or opp drivers needs some adjustments
> > > > related with this change.
> > > That's strange. Is this with defconfig? I wonder whether NR_CPUS being
> > > larger caused the issue with this specific code. Otherwise
> > > CPUMASK_OFFSTACK may not work that well on arm64.
> 
> cpumask handling must use the accessor functions provided in
> include/linux/cpumask.h for declaring and accessing cpumasks. It is likely
> related to the driver opencoding one of the accessors.

I took a look at both the OPP code and the cpufreq-dt code and it looks like
those are doign the right thing w.r.t. cpumask manipulation (i.e. they only use
the cpumask accessors, and use the cpumask_var_*() functions to dynamically
allocate/free cpumasks). Maybe I've missed something, but superficially those
look right.

Marek, can you try reverting this commit and trying defconfig + NR_CPUS=512?
That'll have CPUMASK_OFFSTACK=n, and:

* If that blows up, we know the problem is independent of CPUMASK_OFFSTACK, and
  has something to do with large cpumasks (either a driver bug, or elsewhere).

* If that doesn't blow up, it suggests the problem is related to
  CPUMASK_OFFSTACK rather than with large cpumasks specifically.

Either way, we probably need to revert this patch for now, as this won't have
enough time to soak in linux-next in time for v6.9.

Mark.