Re: [PATCH] powerpc/mm/cxl: Add barrier when setting mm cpumask

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Mon, 2017-08-28 at 13:23 +0530, Aneesh Kumar K.V wrote:
> Benjamin Herrenschmidt <benh@kernel.crashing.org> writes:
> 
> > On Mon, 2017-08-28 at 11:55 +0530, Aneesh Kumar K.V wrote:
> > > We need to add memory barrier so that the page table walk doesn't happen
> > > before the cpumask is set and made visible to the other cpus. We need
> > > to use a sync here instead of lwsync because lwsync is not sufficient for
> > > store/load ordering.
> > > 
> > > We also need to add an if (mm) check so that we do the right thing when called
> > > with a kernel context. For kernel context, we have mm = NULL. W.r.t kernel
> > > address we can skip setting the mm cpumask.
> > > 
> > > Fixes: 0f4bc0932e ("powerpc/mm/cxl: Add the fault handling cpu to mm cpumask")
> > > Cc: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
> > > Reported-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
> > > Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
> > > Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
> > > ---
> > >  drivers/misc/cxl/fault.c | 14 ++++++++++++--
> > >  1 file changed, 12 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/drivers/misc/cxl/fault.c b/drivers/misc/cxl/fault.c
> > > index ab507e4ed69b..caed2a523bee 100644
> > > --- a/drivers/misc/cxl/fault.c
> > > +++ b/drivers/misc/cxl/fault.c
> > > @@ -141,9 +141,19 @@ int cxl_handle_mm_fault(struct mm_struct *mm, u64 dsisr, u64 dar)
> > >  	/*
> > >  	 * Add the fault handling cpu to task mm cpumask so that we
> > >  	 * can do a safe lockless page table walk when inserting the
> > > -	 * hash page table entry.
> > > +	 * hash page table entry. This function get called with a
> > > +	 * valid mm for all user space applications. Hence using
> > > +	 * if (mm) check is sufficient here.
> > >  	 */
> > > -	cpumask_set_cpu(smp_processor_id(), mm_cpumask(mm));
> > > +	if (mm) {
> > > +		cpumask_set_cpu(smp_processor_id(), mm_cpumask(mm));
> > 
> > First test if it's already set as this should be quite common and the
> > cost of setting is a full atomic.
> > 
> 
> Something like below ?

No, don't use cpumask_test_and_set_cpu, that has extra barriers you
don't want.

Do an

	if (!cpumask_test(...)) {
		cpumask_set(...);
		smp_mb();
	}

Like we do in context switch.

> diff --git a/drivers/misc/cxl/fault.c b/drivers/misc/cxl/fault.c
> index caed2a523bee..ccf8568262e4 100644
> --- a/drivers/misc/cxl/fault.c
> +++ b/drivers/misc/cxl/fault.c
> @@ -146,13 +146,13 @@ int cxl_handle_mm_fault(struct mm_struct *mm, u64 dsisr, u64 dar)
>  	 * if (mm) check is sufficient here.
>  	 */
>  	if (mm) {
> -		cpumask_set_cpu(smp_processor_id(), mm_cpumask(mm));
> -		/*
> -		 * We need to make sure we walk the table only after
> -		 * we update the cpumask. The other side of the barrier is
> -		 * explained * in serialize_against_pte_lookup()
> -		 */
> -		smp_mb();
> +		if (!cpumask_test_and_set_cpu(smp_processor_id(), mm_cpumask(mm)))
> +			/*
> +			 * We need to make sure we walk the table only after
> +			 * we update the cpumask. The other side of the barrier
> +			 * is explained in serialize_against_pte_lookup()
> +			 */
> +			smp_mb();
>  	}
>  	if ((result = copro_handle_mm_fault(mm, dar, dsisr, &flt))) {
>  		pr_devel("copro_handle_mm_fault failed: %#x\n", result);