[PATCH] net: mvneta: remove useless RX descriptor prefetch

[PATCH] net: mvneta: remove useless RX descriptor prefetch

[Jisheng Zhang]

The rx descriptors are allocated using dma_alloc_coherent, so prefetch
doesn't really happen at all.

Signed-off-by: Jisheng Zhang <jszhang@marvell.com>
---
 drivers/net/ethernet/marvell/mvneta.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/drivers/net/ethernet/marvell/mvneta.c b/drivers/net/ethernet/marvell/mvneta.c
index 5880871..6c09a27 100644
--- a/drivers/net/ethernet/marvell/mvneta.c
+++ b/drivers/net/ethernet/marvell/mvneta.c
@@ -757,7 +757,6 @@ mvneta_rxq_next_desc_get(struct mvneta_rx_queue *rxq)
 	int rx_desc = rxq->next_desc_to_proc;
 
 	rxq->next_desc_to_proc = MVNETA_QUEUE_NEXT_DESC(rxq, rx_desc);
-	prefetch(rxq->descs + rxq->next_desc_to_proc);
 	return rxq->descs + rx_desc;
 }
 
-- 
2.8.0.rc3

Re: [PATCH] net: mvneta: remove useless RX descriptor prefetch

[Jisheng Zhang]

Hi,

+ linux arm kernel

On Thu, 31 Mar 2016 14:36:30 +0800 Jisheng Zhang wrote:

> The rx descriptors are allocated using dma_alloc_coherent, so prefetch
> doesn't really happen at all.

This is for RFC, I'm sorry to send it without changing its title -- s/PATCH/RFC.

I'm not sure whether there's any benefit to prefetch on space allocated from
dma_alloc_coherent.

Thanks

> 
> Signed-off-by: Jisheng Zhang <jszhang@marvell.com>
> ---
>  drivers/net/ethernet/marvell/mvneta.c | 1 -
>  1 file changed, 1 deletion(-)
> 
> diff --git a/drivers/net/ethernet/marvell/mvneta.c b/drivers/net/ethernet/marvell/mvneta.c
> index 5880871..6c09a27 100644
> --- a/drivers/net/ethernet/marvell/mvneta.c
> +++ b/drivers/net/ethernet/marvell/mvneta.c
> @@ -757,7 +757,6 @@ mvneta_rxq_next_desc_get(struct mvneta_rx_queue *rxq)
>  	int rx_desc = rxq->next_desc_to_proc;
>  
>  	rxq->next_desc_to_proc = MVNETA_QUEUE_NEXT_DESC(rxq, rx_desc);
> -	prefetch(rxq->descs + rxq->next_desc_to_proc);
>  	return rxq->descs + rx_desc;
>  }
>  

Re: [PATCH] net: mvneta: remove useless RX descriptor prefetch

[Jisheng Zhang]

Hi all,

On Thu, 31 Mar 2016 14:45:37 +0800 Jisheng Zhang wrote:

> Hi,
> 
> + linux arm kernel
> 
> On Thu, 31 Mar 2016 14:36:30 +0800 Jisheng Zhang wrote:
> 
> > The rx descriptors are allocated using dma_alloc_coherent, so prefetch
> > doesn't really happen at all.  
> 
> This is for RFC, I'm sorry to send it without changing its title -- s/PATCH/RFC.
> 
> I'm not sure whether there's any benefit to prefetch on space allocated from
> dma_alloc_coherent.

After more consideration, I think my patch is wrong.

As for coherent platforms, the space allocated from dma_alloc_coherent is
cacheable, so prefetch would definitely benefit us.

As for noncoherent platforms, the space allocated from dma_alloc_coherent is
uncacheable, but prefetch on arm/arm64 is implemented via. pld/prfm, the op
would be nop if target address is uncacheable.

So let's drop this patch.

Thanks,
Jisheng


> 
> Thanks
> 
> > 
> > Signed-off-by: Jisheng Zhang <jszhang@marvell.com>
> > ---
> >  drivers/net/ethernet/marvell/mvneta.c | 1 -
> >  1 file changed, 1 deletion(-)
> > 
> > diff --git a/drivers/net/ethernet/marvell/mvneta.c b/drivers/net/ethernet/marvell/mvneta.c
> > index 5880871..6c09a27 100644
> > --- a/drivers/net/ethernet/marvell/mvneta.c
> > +++ b/drivers/net/ethernet/marvell/mvneta.c
> > @@ -757,7 +757,6 @@ mvneta_rxq_next_desc_get(struct mvneta_rx_queue *rxq)
> >  	int rx_desc = rxq->next_desc_to_proc;
> >  
> >  	rxq->next_desc_to_proc = MVNETA_QUEUE_NEXT_DESC(rxq, rx_desc);
> > -	prefetch(rxq->descs + rxq->next_desc_to_proc);
> >  	return rxq->descs + rx_desc;
> >  }
> >    
> 
> 
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel