Re: [LTP] [PATCH v1] sched_rr_get_interval01.c: Put test process into absolute root cgroup (0::/)

Re: [LTP] [PATCH v1] sched_rr_get_interval01.c: Put test process into absolute root cgroup (0::/)

[Petr Vorel]

Hi Wei, all,

> When the CONFIG_RT_GROUP_SCHED=y config is set, test cases like sched_rr_get_interval01
> will failed since limitation of RT processes with cgroup v2 cpu controller.
> The limitation is RT processes have to be in the root cgroup before enabling cpu controller.
> By default the shell will not running in root cgroup "0::/" since systemd will put shell
> into 0::/user.slice/user-xx.slice/session-xx.scope, so ltp case run within shell will failed.
> We can use this patch to workaround above limitation. If we agree on this patch, i will
> continue do same patch to following cases:
> sched_rr_get_interval02
> sched_rr_get_interval03
> sched_setparam02
> sched_getscheduler01

Acked-by: Petr Vorel <pvorel@suse.cz>

LGTM.

@Michal @Li WDYT?

Kind regards,
Petr

> Fixes: https://github.com/linux-test-project/ltp/issues/1245
> Signed-off-by: Wei Gao <wegao@suse.com>
> ---
>  .../sched_rr_get_interval/sched_rr_get_interval01.c         | 6 ++++++
>  1 file changed, 6 insertions(+)

> diff --git a/testcases/kernel/syscalls/sched_rr_get_interval/sched_rr_get_interval01.c b/testcases/kernel/syscalls/sched_rr_get_interval/sched_rr_get_interval01.c
> index b4d75bdcc..55516ec89 100644
> --- a/testcases/kernel/syscalls/sched_rr_get_interval/sched_rr_get_interval01.c
> +++ b/testcases/kernel/syscalls/sched_rr_get_interval/sched_rr_get_interval01.c
> @@ -43,6 +43,12 @@ static void setup(void)

>  	tp.type = tv->ts_type;

> +	if (access("/sys/fs/cgroup/cgroup.controllers", F_OK) == 0) {
> +		int pid = getpid();
> +
> +		SAFE_FILE_PRINTF("/sys/fs/cgroup/cgroup.procs", "%d", pid);
> +	}
> +
>  	if ((sys_sched_setscheduler(0, SCHED_RR, &p)) == -1)
>  		tst_res(TFAIL | TERRNO, "sched_setscheduler() failed");

Re: [LTP] [PATCH v1] sched_rr_get_interval01.c: Put test process into absolute root cgroup (0::/)

[Andrea Cervesato]

Merged, thanks.

- Andrea

Re: [LTP] [PATCH v1] sched_rr_get_interval01.c: Put test process into absolute root cgroup (0::/)

[Michal Koutný]

[-- Attachment #1: Type: text/plain, Size: 345 bytes --]

On Thu, Jun 05, 2025 at 11:40:19AM +0200, Petr Vorel <pvorel@suse.cz> wrote:
> @Michal @Li WDYT?

RT_GROUP scheduling is v1 feature as of now.

Testing cgroup v2 makes only sense with 
CONFIG_RT_GROUP_SCHED=y and CONFIG_RT_GROUP_SCHED_DEFAULT_DISABLED=y
(this combination is equivalent to CONFIG_RT_GROUP_SCHED=n on v2).

HTH,
Michal

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]

Re: [LTP] [PATCH v1] sched_rr_get_interval01.c: Put test process into absolute root cgroup (0::/)

[Wei Gao]

On Thu, Jun 05, 2025 at 05:56:11PM +0200, Michal Koutný wrote:
> On Thu, Jun 05, 2025 at 11:40:19AM +0200, Petr Vorel <pvorel@suse.cz> wrote:
> > @Michal @Li WDYT?
> 
> RT_GROUP scheduling is v1 feature as of now.
> 
> Testing cgroup v2 makes only sense with 
> CONFIG_RT_GROUP_SCHED=y and CONFIG_RT_GROUP_SCHED_DEFAULT_DISABLED=y
> (this combination is equivalent to CONFIG_RT_GROUP_SCHED=n on v2).

@Michal Koutný  So we should skip test cgroupv2 with CONFIG_RT_GROUP_SCHED=yes, correct? Like following change?

diff --git a/testcases/kernel/syscalls/sched_rr_get_interval/sched_rr_get_interval01.c b/testcases/kernel/syscalls/sched_rr_get_interval/sched_rr_get_interval01.c
index 55516ec89..b12bd7857 100644
--- a/testcases/kernel/syscalls/sched_rr_get_interval/sched_rr_get_interval01.c
+++ b/testcases/kernel/syscalls/sched_rr_get_interval/sched_rr_get_interval01.c
@@ -43,10 +43,9 @@ static void setup(void)

        tp.type = tv->ts_type;

+       static const char * const kconf[] = {"CONFIG_RT_GROUP_SCHED=y", NULL};
+       if ((access("/sys/fs/cgroup/cgroup.controllers", F_OK) == 0) && !tst_kconfig_check(kconf_strict)) {
+               tst_brk(TCONF, "CONFIG_RT_GROUP_SCHED not support on cgroupv2");
+       }

> 
> HTH,
> Michal

Re: [LTP] [PATCH v1] sched_rr_get_interval01.c: Put test process into absolute root cgroup (0::/)

[Michal Koutný]

[-- Attachment #1: Type: text/plain, Size: 407 bytes --]

On Sat, Jun 07, 2025 at 02:42:22PM -0400, Wei Gao <wegao@suse.com> wrote:
> @Michal Koutný  So we should skip test cgroupv2 with CONFIG_RT_GROUP_SCHED=yes, correct? 

Ideally, no one should run v2 with CONFIG_RT_GROUP_SCHED=y, so this
would never fail :-p

> Like following change?

But if there are such poor souls, that skip should make the test not
obstruct the rest of LTP.

Thanks,
Michal

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]

Re: [LTP] [PATCH v1] sched_rr_get_interval01.c: Put test process into absolute root cgroup (0::/)

[Petr Vorel]

Hi all,

> On Sat, Jun 07, 2025 at 02:42:22PM -0400, Wei Gao <wegao@suse.com> wrote:
> > @Michal Koutný  So we should skip test cgroupv2 with CONFIG_RT_GROUP_SCHED=yes, correct? 

> Ideally, no one should run v2 with CONFIG_RT_GROUP_SCHED=y, so this
> would never fail :-p

@Michal thanks for a clarification. It's default n and
Documentation/scheduler/sched-rt-group.rst which mentions CONFIG_RT_GROUP_SCHED
also mentions Documentation/admin-guide/cgroup-v1/cgroups.rst, that should be a
good hint for people changing the defaults :).

Also, there are 3 old mips and sh boards which set CONFIG_RT_GROUP_SCHED=y.
Probably nobody cares much nowadays, but maybe it should be deleted.

> > Like following change?

> But if there are such poor souls, that skip should make the test not
> obstruct the rest of LTP.

@Andrea, you claimed it's merged, but maybe you reply to a wrong thread
because it's not merged.
https://lore.kernel.org/ltp/b15fa0cb-e893-4642-9aa0-05f732b75131@suse.com/

I agree with Cyril to have a custom function in lib/tst_cgroup.c, which would be
called by affected tests.

Kind regards,
Petr

> Thanks,
> Michal