Re: [RFC PATCH v2 03/10] arm_mpam: resctrl: Do not fail CPU online when domain setup fails

[Jinjie Ruan]

在 2026/10/20 6:07, Reinette Chatre 写道:
> When a CPU comes online, mpam_resctrl_online_cpu() adds it to the resctrl
> domains of each resource in use, and sets up a new domain when the CPU
> is the first of its domain to come online. A failure to set up a domain
> is returned as the failure of MPAM's CPU hotplug startup callback.

Indeed the code works this way.

> 
> The CPU hotplug core does not run the teardown callback for a CPU whose
> startup callback failed, so everything done for the CPU before the
> failure stays in place: the CPU remains in the domains of the resources
> that were handled before the failing one, and MPAM's per-MSC state for
> the CPU is not undone. If the failure happens when MPAM installs its
> CPU hotplug callbacks during enabling, the hotplug core discards the
> callbacks, leaving MPAM enabled and resctrl initialized but MPAM no

By the way, I previously ran into a case during testing where adding the
"maxcpu" boot parameter caused MPAM enabling to fail.

Link:
https://lore.kernel.org/all/20261020134239.3921377-6-ruanjinjie@xxxxxxxxxx/

> longer notified of CPU hotplug. If it happens when a CPU comes online
> later, the CPU fails to come online.
> 
> Setting up a domain only fails if a memory allocation fails or if a
> class unexpectedly has no component for the CPU. For such rare failures
> losing a resctrl domain is a smaller cost than failing CPU online or
> leaving MPAM in an inconsistent state.
> 
> Do not report domain setup failures from mpam_resctrl_online_cpu() and
> leave the CPU without a domain for that resource. Make
> mpam_resctrl_alloc_domain() return nothing since its failures are no
> longer reported. Keep adding the CPU to resctrl's default group even

Yes, We now always call resctrl_online_cpu() to add them to the default
group.

> when a domain could not be set up, since the CPU is online and runs
> tasks regardless, and resctrl expects every online CPU to belong to a
> resource group. Drop the warning when a CPU going offline has no domain,
> since a missing domain is now expected.

It is now valid for the domain to be empty when a CPU goes offline.

> 
> Only domains that were set up are exposed through resctrl and torn down
> when their CPUs go offline. When another CPU of the same domain comes
> online, it sets up the domain again, and if that succeeds the domain
> does not include the CPU for which setup failed.
> 
> Signed-off-by: Reinette Chatre <reinette.chatre@xxxxxxxxx>
> ---
>  drivers/resctrl/mpam_devices.c  |  2 +-
>  drivers/resctrl/mpam_internal.h |  4 ++--
>  drivers/resctrl/mpam_resctrl.c  | 31 +++++++++++++++----------------
>  3 files changed, 18 insertions(+), 19 deletions(-)
> 
> diff --git a/drivers/resctrl/mpam_devices.c b/drivers/resctrl/mpam_devices.c
> index 48adc49b833c..d17d7a72dba1 100644
> --- a/drivers/resctrl/mpam_devices.c
> +++ b/drivers/resctrl/mpam_devices.c
> @@ -2249,7 +2249,7 @@ static int mpam_cpu_online(unsigned int cpu)
>  	}
>  
>  	if (mpam_resctrl_enabled)
> -		return mpam_resctrl_online_cpu(cpu);
> +		mpam_resctrl_online_cpu(cpu);
>  
>  	return 0;
>  }
> diff --git a/drivers/resctrl/mpam_internal.h b/drivers/resctrl/mpam_internal.h
> index 1f86cc158100..84c7adbb1a40 100644
> --- a/drivers/resctrl/mpam_internal.h
> +++ b/drivers/resctrl/mpam_internal.h
> @@ -525,13 +525,13 @@ int mpam_get_cpumask_from_cache_id(unsigned long cache_id, u32 cache_level,
>  #ifdef CONFIG_RESCTRL_FS
>  int mpam_resctrl_setup(void);
>  void mpam_resctrl_exit(void);
> -int mpam_resctrl_online_cpu(unsigned int cpu);
> +void mpam_resctrl_online_cpu(unsigned int cpu);
>  void mpam_resctrl_offline_cpu(unsigned int cpu);
>  void mpam_resctrl_teardown_class(struct mpam_class *class);
>  #else
>  static inline int mpam_resctrl_setup(void) { return 0; }
>  static inline void mpam_resctrl_exit(void) { }
> -static inline int mpam_resctrl_online_cpu(unsigned int cpu) { return 0; }
> +static inline void mpam_resctrl_online_cpu(unsigned int cpu) { }
>  static inline void mpam_resctrl_offline_cpu(unsigned int cpu) { }
>  static inline void mpam_resctrl_teardown_class(struct mpam_class *class) { }
>  #endif /* CONFIG_RESCTRL_FS */
> diff --git a/drivers/resctrl/mpam_resctrl.c b/drivers/resctrl/mpam_resctrl.c
> index b07e5dd895dd..dc442851fd7e 100644
> --- a/drivers/resctrl/mpam_resctrl.c
> +++ b/drivers/resctrl/mpam_resctrl.c
> @@ -1615,8 +1615,8 @@ static struct mpam_component *find_component(struct mpam_class *class, int cpu)
>  	return NULL;
>  }
>  
> -static struct mpam_resctrl_dom *
> -mpam_resctrl_alloc_domain(unsigned int cpu, struct mpam_resctrl_res *res)
> +static void mpam_resctrl_alloc_domain(unsigned int cpu,
> +				      struct mpam_resctrl_res *res)
>  {
>  	int err;
>  	struct mpam_resctrl_dom *dom;
> @@ -1631,11 +1631,11 @@ mpam_resctrl_alloc_domain(unsigned int cpu, struct mpam_resctrl_res *res)
>  	comp = find_component(class, cpu);
>  	/* class has no component for this CPU */
>  	if (WARN_ON_ONCE(!comp))
> -		return ERR_PTR(-EINVAL);
> +		return;
>  
>  	dom = kzalloc_node(sizeof(*dom), GFP_KERNEL, cpu_to_node(cpu));
>  	if (!dom)
> -		return ERR_PTR(-ENOMEM);
> +		return;
>  
>  	if (r->alloc_capable) {
>  		dom->ctrl_comp = comp;
> @@ -1678,7 +1678,6 @@ mpam_resctrl_alloc_domain(unsigned int cpu, struct mpam_resctrl_res *res)
>  		}
>  		if (!any_mon_comp) {
>  			WARN_ON_ONCE(0);
> -			err = -EFAULT;
>  			goto offline_ctrl_domain;
>  		}
>  
> @@ -1694,7 +1693,7 @@ mpam_resctrl_alloc_domain(unsigned int cpu, struct mpam_resctrl_res *res)
>  		pr_debug("Skipped monitor domain online - no monitors\n");
>  	}
>  
> -	return dom;
> +	return;
>  
>  offline_ctrl_domain:
>  	if (r->alloc_capable) {
> @@ -1703,9 +1702,6 @@ mpam_resctrl_alloc_domain(unsigned int cpu, struct mpam_resctrl_res *res)
>  	}
>  free_domain:
>  	kfree(dom);
> -	dom = ERR_PTR(err);
> -
> -	return dom;
>  }
>  
>  /*
> @@ -1757,7 +1753,7 @@ mpam_resctrl_get_domain_from_cpu(int cpu, struct mpam_resctrl_res *res)
>  	return mpam_resctrl_get_mon_domain_from_cpu(cpu);
>  }
>  
> -int mpam_resctrl_online_cpu(unsigned int cpu)
> +void mpam_resctrl_online_cpu(unsigned int cpu)
>  {
>  	struct mpam_resctrl_res *res;
>  	enum resctrl_res_level rid;
> @@ -1772,9 +1768,13 @@ int mpam_resctrl_online_cpu(unsigned int cpu)
>  
>  		dom = mpam_resctrl_get_domain_from_cpu(cpu, res);
>  		if (!dom) {
> -			dom = mpam_resctrl_alloc_domain(cpu, res);
> -			if (IS_ERR(dom))
> -				return PTR_ERR(dom);
> +			/*
> +			 * On failure the CPU is left without a domain for this
> +			 * resource, which mpam_resctrl_offline_cpu() tolerates.
> +			 * resctrl_online_cpu() still adds the CPU to the default
> +			 * group because the CPU is online and runs tasks.
> +			 */
> +			mpam_resctrl_alloc_domain(cpu, res);
>  		} else {
>  			if (r->alloc_capable) {
>  				struct rdt_ctrl_domain *ctrl_d = &dom->resctrl_ctrl_dom;
> @@ -1790,8 +1790,6 @@ int mpam_resctrl_online_cpu(unsigned int cpu)
>  	}
>  
>  	resctrl_online_cpu(cpu);
> -
> -	return 0;
>  }
>  
>  void mpam_resctrl_offline_cpu(unsigned int cpu)
> @@ -1812,8 +1810,9 @@ void mpam_resctrl_offline_cpu(unsigned int cpu)
>  		if (!res->class)
>  			continue;	// dummy resource
>  
> +		/* The domain may be missing if its setup failed. */
>  		dom = mpam_resctrl_get_domain_from_cpu(cpu, res);
> -		if (WARN_ON_ONCE(!dom))
> +		if (!dom)
>  			continue;

Reviewed-by: Jinjie Ruan <ruanjinjie@xxxxxxxxxx>

>  
>  		if (r->alloc_capable) {

-- 
Best regards,
Jinjie