Re: [PATCH kernel] powerpc/pseries/iommu: Add cond_resched() for huge updates

[Michael Ellerman <mpe@ellerman.id.au>]

Alexey Kardashevskiy <aik@ozlabs.ru> writes:
> Mapping ~5.000.000 TCEs currently takes about 40s; this is the amount
> required for a 300GB VM with 64k IOMMU page size. Anything bigger than
> this produces RCU stall warnings.

OK. Are we sure we're not doing anything stupid in that code to make it
go that slowly?

> This adds cond_resched() to allow the scheduler to do context switching
> when it decides to.
>
> This loop is called from dma_set_mask() which is a sleepable context.
>
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> ---
>  arch/powerpc/platforms/pseries/iommu.c | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/arch/powerpc/platforms/pseries/iommu.c b/arch/powerpc/platforms/pseries/iommu.c
> index 889dc2e44b89..2b8de822272f 100644
> --- a/arch/powerpc/platforms/pseries/iommu.c
> +++ b/arch/powerpc/platforms/pseries/iommu.c
> @@ -459,6 +459,7 @@ static int tce_setrange_multi_pSeriesLP(unsigned long start_pfn,
>  static int tce_setrange_multi_pSeriesLP_walk(unsigned long start_pfn,
>  		unsigned long num_pfn, void *arg)
>  {
> +	cond_resched();
>  	return tce_setrange_multi_pSeriesLP(start_pfn, num_pfn, arg);
>  }

Why there and not in tce_setrange_multi_pSeriesLP() ?

I'm not sure what the maximum granularity walk_system_ram_range() will
ever call us with is.

cheers