linux-kernel - Re: [PATCH] arm: Fix topology setup in case of CPU hotplug for CONFIG_SCHED

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Wed, 20 Nov 2019 11:09:11 +0000
From:   Sudeep Holla <sudeep.holla@....com>
To:     Dietmar Eggemann <dietmar.eggemann@....com>
Cc:     Atish Patra <atish.patra@....com>,
        Russell King <linux@...linux.org.uk>,
        Morten Rasmussen <morten.rasmussen@....com>,
        Lukasz Luba <lukasz.luba@....com>,
        linux-kernel@...r.kernel.org, linux-arm-kernel@...ts.infradead.org,
        Sudeep Holla <sudeep.holla@....com>
Subject: Re: [PATCH] arm: Fix topology setup in case of CPU hotplug for
 CONFIG_SCHED_MC

Hi Dietmar, Lukasz,

Thanks for digging this bug and fixing it.

On Wed, Nov 20, 2019 at 10:42:12AM +0000, Dietmar Eggemann wrote:
> Commit ca74b316df96 ("arm: Use common cpu_topology structure and
> functions.") changed cpu_coregroup_mask() from the ARM32 specific
> implementation in arch/arm/include/asm/topology.h to the one shared
> with ARM64 and RISCV in drivers/base/arch_topology.c.
>
> Currently on Arm32 (TC2 w/ CONFIG_SCHED_MC) the task scheduler setup
> code (w/ CONFIG_SCHED_DEBUG) shows this during CPU hotplug:
>
>   ERROR: groups don't span domain->span
>
> It happens to CPUs of the cluster of the CPU which gets hot-plugged
> out on scheduler domain MC.
>
> Turns out that the shared cpu_coregroup_mask() requires that the
> hot-plugged CPU is removed from the core_sibling mask via
> remove_cpu_topology(). Otherwise the 'is core_sibling subset of
> cpumask_of_node()' doesn't work. In this case the task scheduler has to
> deal with cpumask_of_node instead of core_sibling which is wrong on
> scheduler domain MC.
>
> e.g. CPU3 hot-plugged out on TC2 [cluster0: 0,3-4 cluster1: 1-2]:
>
>   cpu_coregroup_mask(): CPU3 cpumask_of_node=0-2,4 core_sibling=0,3-4
>                                                                   ^
> should be:
>
>   cpu_coregroup_mask(): CPU3 cpumask_of_node=0-2,4 core_sibling=0,4
>
> Add remove_cpu_topology() to __cpu_disable() to remove the CPU from the
> topology masks in case of a CPU hotplug out operation.
>
> At the same time tweak store_cpu_topology() slightly so it will call
> update_siblings_masks() in case of CPU hotplug in operation via
> secondary_start_kernel()->smp_store_cpu_info().
>
> This aligns the Arm32 implementation with the Arm64 one.
>
> Fixes: ca74b316df96 ("arm: Use common cpu_topology structure and functions")

Apart from the minor nit below,

Reviewed-by: Sudeep Holla <sudeep.holla@....com>

> Signed-off-by: Dietmar Eggemann <dietmar.eggemann@....com>
> ---
>  arch/arm/kernel/smp.c      |  2 ++
>  arch/arm/kernel/topology.c | 15 +++++++--------
>  2 files changed, 9 insertions(+), 8 deletions(-)
>
> diff --git a/arch/arm/kernel/smp.c b/arch/arm/kernel/smp.c
> index 4b0bab2607e4..139c0d98fa29 100644
> --- a/arch/arm/kernel/smp.c
> +++ b/arch/arm/kernel/smp.c
> @@ -240,6 +240,8 @@ int __cpu_disable(void)
>  	if (ret)
>  		return ret;
>
> +	remove_cpu_topology(cpu);
> +
>  	/*
>  	 * Take this CPU offline.  Once we clear this, we can't return,
>  	 * and we must not schedule until we're ready to give up the cpu.
> diff --git a/arch/arm/kernel/topology.c b/arch/arm/kernel/topology.c
> index 5b9faba03afb..b37b0a340991 100644
> --- a/arch/arm/kernel/topology.c
> +++ b/arch/arm/kernel/topology.c
> @@ -196,9 +196,8 @@ void store_cpu_topology(unsigned int cpuid)
>  	struct cpu_topology *cpuid_topo = &cpu_topology[cpuid];
>  	unsigned int mpidr;
>
> -	/* If the cpu topology has been already set, just return */
> -	if (cpuid_topo->core_id != -1)
> -		return;
> +	if (cpuid_topo->package_id != -1)
> +		goto topology_populated;
>
>  	mpidr = read_cpuid_mpidr();
>
> @@ -231,14 +230,14 @@ void store_cpu_topology(unsigned int cpuid)
>  		cpuid_topo->package_id = -1;
>  	}
>
> -	update_siblings_masks(cpuid);
> -
>  	update_cpu_capacity(cpuid);
>
>  	pr_info("CPU%u: thread %d, cpu %d, socket %d, mpidr %x\n",
> -		cpuid, cpu_topology[cpuid].thread_id,
> -		cpu_topology[cpuid].core_id,
> -		cpu_topology[cpuid].package_id, mpidr);
> +		cpuid, cpuid_topo->thread_id, cpuid_topo->core_id,
> +		cpuid_topo->package_id, mpidr);
> +

[nit] The above is a clearly cosmetic cleanup and shouldn't be part of
this fix as they will be backported automatically. So I prefer to drop
this or make it separate patch if required.

--
Regards,
Sudeep