[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <89525069-4fd5-2fd8-20eb-5af240a4bceb@arm.com>
Date: Thu, 12 May 2022 12:58:50 +0200
From: Dietmar Eggemann <dietmar.eggemann@....com>
To: Qing Wang <wangqing@...o.com>, Sudeep Holla <sudeep.holla@....com>,
Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
"Rafael J. Wysocki" <rafael@...nel.org>,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH V2] arch_topology: support parsing cluster_id from DT
On 11/05/2022 11:52, Qing Wang wrote:
> From: Wang Qing <wangqing@...o.com>
[...]
> @@ -582,7 +594,8 @@ static int __init parse_cluster(struct device_node *cluster, int depth)
> }
>
> if (leaf) {
> - ret = parse_core(c, package_id, core_id++);
> + ret = parse_core(c, package_id, (depth == 2)?cluster_id : -1,
> + core_id++);
> } else {
> pr_err("%pOF: Non-leaf cluster with core %s\n",
> cluster, name);
> @@ -599,9 +612,6 @@ static int __init parse_cluster(struct device_node *cluster, int depth)
> if (leaf && !has_cores)
> pr_warn("%pOF: empty cluster\n", cluster);
>
> - if (leaf)
> - package_id++;
> -
> return 0;
> }
The issue I mentioned under
https://lkml.kernel.org/r/bd746cf0-0fdd-1ee6-d394-67fffb5d9b58@arm.com
still exists.
Btw, I recommend the following test strategy.
(A) Create a set of dts files which represent today's topologies in DT:
(1) 8 CPUs flat (Arm DynamIQ single DSU)
(2) 2 groups of 4 CPUs (e.g. hikey 960) (which covers Phantom* domain)
(3) your QC SM8450 Armv9 tri-gear (4-3-1) DynamIQ single DSU w/ shared
L2 btwn CPU0-1 and CPU2-3.
...
* used in Android
(B) Compile dtb's
dtc -I dts -O dtb -o foo.dtb foo.dts
(C) Run them under qemu w/ and w/o CONFIG_SCHED_CLUSTER and check:
sudo qemu-system-aarch64 ... -dtb foo.dtb
(1) sched domains:
cat /sys/kernel/debug/sched/domains/cpu*/domain*/name
(2) sched flags:
cat /sys/kernel/debug/sched/domains/cpu*/domain*/flags
(3) cpumasks:
cat /proc/schedstat | awk '{print $1 " " $2 }' | grep ^[cd]
You can even mention the test results in your patch so that people see
that you already covered them. This will speed up the review-process
enormously.
Powered by blists - more mailing lists