[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <aLb_SJ2MYefD4JuN@localhost.localdomain>
Date: Tue, 2 Sep 2025 16:29:28 +0200
From: Frederic Weisbecker <frederic@...nel.org>
To: Sehee Jeong <sehee1.jeong@...sung.com>
Cc: anna-maria@...utronix.de, tglx@...utronix.de, corbet@....net,
linux-kernel@...r.kernel.org, linux-doc@...r.kernel.org
Subject: Re: timers/migration: add 'notmigr' kernel parameter to disable
timer migration
Le Thu, Aug 07, 2025 at 03:48:49PM +0900, Sehee Jeong a écrit :
> On heterogeneous systems with big.LITTLE architectures, timer migration
> may cause timers from little cores to run on big cores, or vice versa,
> because core type differences are not taken into account in the current
> timer migration logic. This can be undesirable in systems that require
> strict power management, predictable latency, or core isolation.
>
> This patch does not attempt to solve the structural limitation,
> but provides a workaround by introducing an optional early boot parameter:
>
> notmigr
>
> When specified, timer migration initialization is skipped entirely.
It's a shame we couldn't solve your problem yet but at least such an
option can be nice for general testing beyond big.LITTLE situations.
>
> Signed-off-by: Sehee Jeong <sehee1.jeong@...sung.com>
> ---
> Documentation/admin-guide/kernel-parameters.txt | 2 ++
> kernel/time/timer_migration.c | 14 ++++++++++++++
> 2 files changed, 16 insertions(+)
>
> diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> index 85c27c97b826..4bb10ac574e5 100644
> --- a/Documentation/admin-guide/kernel-parameters.txt
> +++ b/Documentation/admin-guide/kernel-parameters.txt
> @@ -4132,6 +4132,8 @@
> no_timer_check [X86,APIC] Disables the code which tests for
> broken timer IRQ sources.
>
> + notmigr [KNL,EARLY] Disable timer migration.
> +
> no_uaccess_flush
> [PPC,EARLY] Don't flush the L1-D cache after accessing user data.
>
> diff --git a/kernel/time/timer_migration.c b/kernel/time/timer_migration.c
> index 72538baa7a1f..7636a1b3ae6b 100644
> --- a/kernel/time/timer_migration.c
> +++ b/kernel/time/timer_migration.c
> @@ -422,6 +422,8 @@ static unsigned int tmigr_crossnode_level __read_mostly;
>
> static DEFINE_PER_CPU(struct tmigr_cpu, tmigr_cpu);
>
> +static bool tmigr_enabled = true;
> +
> #define TMIGR_NONE 0xFF
> #define BIT_CNT 8
>
> @@ -1790,6 +1792,9 @@ static int __init tmigr_init(void)
>
> BUILD_BUG_ON_NOT_POWER_OF_2(TMIGR_CHILDREN_PER_GROUP);
>
> + if (!tmigr_enabled)
> + return 0;
> +
> /* Nothing to do if running on UP */
> if (ncpus == 1)
> return 0;
> @@ -1854,3 +1859,12 @@ static int __init tmigr_init(void)
> return ret;
> }
> early_initcall(tmigr_init);
> +
> +static int __init notmigr(char *str)
> +{
> + tmigr_enabled = false;
> +
> + return 0;
> +}
> +
> +early_param("notmigr", notmigr);
This immediately reads like "not migr", which doesn't ring a bell.
no_tmigr would be better. I would even prefer tmigr=off (with default on)
so that the arguments can be further extended in the future if needed.
Thanks
> --
> 2.49.0
>
--
Frederic Weisbecker
SUSE Labs
Powered by blists - more mailing lists