mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH] arm_mpam: resctrl: Annotate domain list walks as RCU under cpus lock
@ 2026-11-16  2:48 Jinjie Ruan
  2026-11-16 12:12 ` Ben Horgan
  0 siblings, 1 reply; 2+ messages in thread
From: Jinjie Ruan @ 2026-11-16  2:48 UTC (permalink / raw)
  To: james.morse, ben.horgan, reinette.chatre, fenghuay, jic23,
	tan.shaopeng, zengheng4, gshan, linux-arm-kernel, linux-kernel
  Cc: ruanjinjie

The resctrl domain lists are RCU lists which are only modified while the
CPU hotplug write lock is held, so traversing them is protected by
cpus_read_lock().  That is the convention the generic resctrl code
follows.

The MPAM implementation omits the condition at three places, so lockdep
reports a spurious violation with CONFIG_PROVE_RCU_LIST on whenever
they run:

  WARNING: suspicious RCU usage
  drivers/resctrl/mpam_resctrl.c:1762 RCU-list traversed in non-reader section!!
   mpam_resctrl_get_domain_from_cpu+0x200/0x298
   mpam_resctrl_online_cpu+0xa4/0x160
   mpam_cpu_online+0x2b4/0x300
   cpuhp_invoke_callback+0x304/0x700

Likewise for mpam_resctrl_get_mon_domain_from_cpu() and
resctrl_arch_update_domains().

The lock really is held: lockdep_assert_cpus_held() is the first thing
each of these functions does, and the warning's own lock list shows
cpu_hotplug_lock and cpuhp_state-up taken.  Only the annotation is
missing, so add lockdep_is_cpus_held() to all three walks.

Fixes: 9cd2b522be2c ("arm_mpam: resctrl: Implement helpers to update configuration")
Fixes: 264c285999fc ("arm_mpam: resctrl: Add monitor initialisation and domain boilerplate")
Fixes: 09e61daf8e96 ("arm_mpam: resctrl: Add boilerplate cpuhp and domain allocation")
Cc: stable@vger.kernel.org
Signed-off-by: Jinjie Ruan <ruanjinjie@huawei.com>
---
 drivers/resctrl/mpam_resctrl.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/drivers/resctrl/mpam_resctrl.c b/drivers/resctrl/mpam_resctrl.c
index 5468367776e2..d25c2bac9678 100644
--- a/drivers/resctrl/mpam_resctrl.c
+++ b/drivers/resctrl/mpam_resctrl.c
@@ -1516,7 +1516,7 @@ int resctrl_arch_update_domains(struct rdt_resource *r, u32 closid)
 	if (!mpam_is_enabled())
 		return -EINVAL;
 
-	list_for_each_entry_rcu(d, &r->ctrl_domains, hdr.list) {
+	list_for_each_entry_rcu(d, &r->ctrl_domains, hdr.list, lockdep_is_cpus_held()) {
 		for (enum resctrl_conf_type t = 0; t < CDP_NUM_TYPES; t++) {
 			struct resctrl_staged_config *cfg = &d->staged_config[t];
 
@@ -1743,7 +1743,8 @@ static struct mpam_resctrl_dom *mpam_resctrl_get_mon_domain_from_cpu(int cpu)
 	if (cache_id < 0)
 		return NULL;
 
-	list_for_each_entry_rcu(dom, &l3->resctrl_res.mon_domains, resctrl_mon_dom.hdr.list) {
+	list_for_each_entry_rcu(dom, &l3->resctrl_res.mon_domains, resctrl_mon_dom.hdr.list,
+				lockdep_is_cpus_held()) {
 		if (dom->resctrl_mon_dom.hdr.id == cache_id)
 			return dom;
 	}
@@ -1759,7 +1760,8 @@ mpam_resctrl_get_domain_from_cpu(int cpu, struct mpam_resctrl_res *res)
 
 	lockdep_assert_cpus_held();
 
-	list_for_each_entry_rcu(dom, &r->ctrl_domains, resctrl_ctrl_dom.hdr.list) {
+	list_for_each_entry_rcu(dom, &r->ctrl_domains, resctrl_ctrl_dom.hdr.list,
+				lockdep_is_cpus_held()) {
 		if (cpumask_test_cpu(cpu, &dom->ctrl_comp->affinity))
 			return dom;
 	}
-- 
2.34.1


^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [PATCH] arm_mpam: resctrl: Annotate domain list walks as RCU under cpus lock
  2026-11-16  2:48 [PATCH] arm_mpam: resctrl: Annotate domain list walks as RCU under cpus lock Jinjie Ruan
@ 2026-11-16 12:12 ` Ben Horgan
  0 siblings, 0 replies; 2+ messages in thread
From: Ben Horgan @ 2026-11-16 12:12 UTC (permalink / raw)
  To: Jinjie Ruan, james.morse, reinette.chatre, fenghuay, jic23,
	tan.shaopeng, zengheng4, gshan, linux-arm-kernel, linux-kernel

Hi Jinjie,

On 16/11/2026 02:48, Jinjie Ruan wrote:
> The resctrl domain lists are RCU lists which are only modified while the
> CPU hotplug write lock is held, so traversing them is protected by
> cpus_read_lock().  That is the convention the generic resctrl code
> follows.
> 
> The MPAM implementation omits the condition at three places, so lockdep
> reports a spurious violation with CONFIG_PROVE_RCU_LIST on whenever
> they run:
> 
>   WARNING: suspicious RCU usage
>   drivers/resctrl/mpam_resctrl.c:1762 RCU-list traversed in non-reader section!!
>    mpam_resctrl_get_domain_from_cpu+0x200/0x298
>    mpam_resctrl_online_cpu+0xa4/0x160
>    mpam_cpu_online+0x2b4/0x300
>    cpuhp_invoke_callback+0x304/0x700
> 
> Likewise for mpam_resctrl_get_mon_domain_from_cpu() and
> resctrl_arch_update_domains().
> 
> The lock really is held: lockdep_assert_cpus_held() is the first thing
> each of these functions does, and the warning's own lock list shows
> cpu_hotplug_lock and cpuhp_state-up taken.  Only the annotation is
> missing, so add lockdep_is_cpus_held() to all three walks.
> 
> Fixes: 9cd2b522be2c ("arm_mpam: resctrl: Implement helpers to update configuration")
> Fixes: 264c285999fc ("arm_mpam: resctrl: Add monitor initialisation and domain boilerplate")
> Fixes: 09e61daf8e96 ("arm_mpam: resctrl: Add boilerplate cpuhp and domain allocation")
> Cc: stable@vger.kernel.org

The patch looks good to me but the Fixes and stable Cc can be dropped. This is just providing extra
debug information to allow a debug feature under RCU_EXPERT to be used. The equivalent commit in
fs/resctrl, 11b57be8e406 ("x86,fs/resctrl: Document safe RCU list traversal"), which is needed to
make this useful was only added after MPAM support was merged and has no Fixes tag.

Thanks,

Ben

> Signed-off-by: Jinjie Ruan <ruanjinjie@huawei.com>
> ---
>  drivers/resctrl/mpam_resctrl.c | 8 +++++---
>  1 file changed, 5 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/resctrl/mpam_resctrl.c b/drivers/resctrl/mpam_resctrl.c
> index 5468367776e2..d25c2bac9678 100644
> --- a/drivers/resctrl/mpam_resctrl.c
> +++ b/drivers/resctrl/mpam_resctrl.c
> @@ -1516,7 +1516,7 @@ int resctrl_arch_update_domains(struct rdt_resource *r, u32 closid)
>  	if (!mpam_is_enabled())
>  		return -EINVAL;
>  
> -	list_for_each_entry_rcu(d, &r->ctrl_domains, hdr.list) {
> +	list_for_each_entry_rcu(d, &r->ctrl_domains, hdr.list, lockdep_is_cpus_held()) {
>  		for (enum resctrl_conf_type t = 0; t < CDP_NUM_TYPES; t++) {
>  			struct resctrl_staged_config *cfg = &d->staged_config[t];
>  
> @@ -1743,7 +1743,8 @@ static struct mpam_resctrl_dom *mpam_resctrl_get_mon_domain_from_cpu(int cpu)
>  	if (cache_id < 0)
>  		return NULL;
>  
> -	list_for_each_entry_rcu(dom, &l3->resctrl_res.mon_domains, resctrl_mon_dom.hdr.list) {
> +	list_for_each_entry_rcu(dom, &l3->resctrl_res.mon_domains, resctrl_mon_dom.hdr.list,
> +				lockdep_is_cpus_held()) {
>  		if (dom->resctrl_mon_dom.hdr.id == cache_id)
>  			return dom;
>  	}
> @@ -1759,7 +1760,8 @@ mpam_resctrl_get_domain_from_cpu(int cpu, struct mpam_resctrl_res *res)
>  
>  	lockdep_assert_cpus_held();
>  
> -	list_for_each_entry_rcu(dom, &r->ctrl_domains, resctrl_ctrl_dom.hdr.list) {
> +	list_for_each_entry_rcu(dom, &r->ctrl_domains, resctrl_ctrl_dom.hdr.list,
> +				lockdep_is_cpus_held()) {
>  		if (cpumask_test_cpu(cpu, &dom->ctrl_comp->affinity))
>  			return dom;
>  	}


^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-11-16 12:12 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-16  2:48 [PATCH] arm_mpam: resctrl: Annotate domain list walks as RCU under cpus lock Jinjie Ruan
2026-11-16 12:12 ` Ben Horgan

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®