Re: [PATCH v4] ipc/msg: mitigate the lock contention with percpu counter

[Dennis Zhou]

Hello,

On Thu, Sep 08, 2022 at 08:38:59AM -0700, Andrew Morton wrote:
> On Thu, 8 Sep 2022 16:25:47 +0800 "Sun, Jiebin" <jiebin.sun@xxxxxxxxx> wrote:
> 
> > In our case, if the local 
> > percpu counter is near to INT_MAX and there comes a big msgsz, the 
> > overflow issue could happen.
> 
> percpu_counter_add_batch() handles this - your big message
> won't overflow an s64.
> 
> 
> Lookng at percpu_counter_add_batch(), is this tweak right?
> 
> - don't need to update *fbc->counters inside the lock
> - that __this_cpu_sub() is an obscure way of zeroing the thing
> 
> --- a/lib/percpu_counter.c~a
> +++ a/lib/percpu_counter.c
> @@ -89,8 +89,8 @@ void percpu_counter_add_batch(struct per
>  		unsigned long flags;
>  		raw_spin_lock_irqsave(&fbc->lock, flags);
>  		fbc->count += count;
> -		__this_cpu_sub(*fbc->counters, count - amount);
>  		raw_spin_unlock_irqrestore(&fbc->lock, flags);
> +		__this_cpu_write(*fbc->counters, 0);

I don't think this is irq safe. It'd be best to leave it inside the
spinlock as then we can use __this_cpu_write() to 0 in there.

>  	} else {
>  		this_cpu_add(*fbc->counters, amount);
>  	}
> _
> 

Thanks,
Dennis