Re: [REGRESSION][v4.14.y][v4.15] x86/intel_rdt/cqm: Improve limbo list processing

[Joseph Salisbury <joseph.salisbury@canonical.com>]

On 01/16/2018 01:59 PM, Thomas Gleixner wrote:
> On Tue, 16 Jan 2018, Yu, Fenghua wrote:
>>> From: Thomas Gleixner [mailto:tglx@linutronix.de]
>> Is this a Haswell specific issue?
>>
>> I run the following test forever without issue on Broadwell and 4.15.0-rc6 with rdt mounted:
>> for ((;;)) do
>>         for ((i=1;i<88;i++)) do
>>                 echo 0 >/sys/devices/system/cpu/cpu$i/online
>>         done
>>         echo "online cpus:"
>>         grep processor /proc/cpuinfo |wc
>>         for ((i=1;i<88;i++)) do
>>                 echo 1 >/sys/devices/system/cpu/cpu$i/online
>>         done
>>         echo "online cpus:"
>>         grep processor /proc/cpuinfo|wc
>> done
>>
>> I'm finding a Haswell to reproduce the issue.
> Come on. This is crystal clear from the KASAN trace. And the fix is simple enough.
>
> You simply do not run into it because on your machine
>
>     is_llc_occupancy_enabled() is false...
>
> Thanks,
>
> 	tglx
> 	
> 8<--------------------	
>
> diff --git a/arch/x86/kernel/cpu/intel_rdt.c b/arch/x86/kernel/cpu/intel_rdt.c
> index 88dcf8479013..99442370de40 100644
> --- a/arch/x86/kernel/cpu/intel_rdt.c
> +++ b/arch/x86/kernel/cpu/intel_rdt.c
> @@ -525,10 +525,6 @@ static void domain_remove_cpu(int cpu, struct rdt_resource *r)
>  		 */
>  		if (static_branch_unlikely(&rdt_mon_enable_key))
>  			rmdir_mondata_subdir_allrdtgrp(r, d->id);
> -		kfree(d->ctrl_val);
> -		kfree(d->rmid_busy_llc);
> -		kfree(d->mbm_total);
> -		kfree(d->mbm_local);
>  		list_del(&d->list);
>  		if (is_mbm_enabled())
>  			cancel_delayed_work(&d->mbm_over);
> @@ -545,6 +541,10 @@ static void domain_remove_cpu(int cpu, struct rdt_resource *r)
>  			cancel_delayed_work(&d->cqm_limbo);
>  		}
>  
> +		kfree(d->ctrl_val);
> +		kfree(d->rmid_busy_llc);
> +		kfree(d->mbm_total);
> +		kfree(d->mbm_local);
>  		kfree(d);
>  		return;
>  	}

Thanks, Thomas.  I'll build some test kernels and have your patch tested
out.


Thanks,


Joe