Am Donnerstag, den 15.11.2018, 01:36 +0100 schrieb Ladislav Michl:
> Receive buffers are properly synchronized only if Cadence is
> GEM. Fix it for MACB as well.
> 
> Fixes: 86dc5259e25d (net: macb: no need for coherent memory for receive 
> buffer)
> > Signed-off-by: Ladislav Michl <[email protected]>
> ---
>  Hi Sascha,
> 
>  I'm unsure this is proper fix, but works for me (as well as
>  reverting 86dc5259e25d)
> 
>  drivers/net/macb.c | 10 +++++++++-
>  1 file changed, 9 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/net/macb.c b/drivers/net/macb.c
> index 1c5752d10..2bd5b62a4 100644
> --- a/drivers/net/macb.c
> +++ b/drivers/net/macb.c
> @@ -216,9 +216,11 @@ static int macb_recv(struct eth_device *edev)
> >     dev_dbg(macb->dev, "%s\n", __func__);
>  
> >     for (;;) {
> > +           barrier();
> >             if (!(macb->rx_ring[rx_tail].addr & MACB_BIT(RX_USED)))
> >                     return -1;
>  
> > +           barrier();
> >             status = macb->rx_ring[rx_tail].ctrl;
> >             if (status & MACB_BIT(RX_SOF)) {
> >                     if (rx_tail != macb->rx_tail)
> @@ -229,6 +231,8 @@ static int macb_recv(struct eth_device *edev)
> >             if (status & MACB_BIT(RX_EOF)) {
> >                     buffer = macb->rx_buffer + macb->rx_buffer_size * 
> > macb->rx_tail;
> >                     length = MACB_BFEXT(RX_FRMLEN, status);
> > +                   dma_sync_single_for_cpu((unsigned long)buffer, length,
> > +                                           DMA_FROM_DEVICE);
> >                     if (wrapped) {
> >                             unsigned int headlen, taillen;
>  
> @@ -237,12 +241,17 @@ static int macb_recv(struct eth_device *edev)
> >                             taillen = length - headlen;
> >                             memcpy((void *)NetRxPackets[0],
> >                                    buffer, headlen);
> > +                           dma_sync_single_for_cpu((unsigned 
> > long)macb->rx_buffer,
> > +                                                   taillen, 
> > DMA_FROM_DEVICE);
> >                             memcpy((void *)NetRxPackets[0] + headlen,
>                                      macb->rx_buffer, taillen);

This is missing a dma_sync_single_for_device() here, matching the
...for_cpu added above. Otherwise patch looks good.

Regards,
Lucas

>                               buffer = (void *)NetRxPackets[0];
> >                     }
>  
> >                     net_receive(edev, buffer, length);
> > +                   dma_sync_single_for_device((unsigned long)buffer, 
> > length,
> > +                                              DMA_FROM_DEVICE);
> > +                   barrier();
> >                     if (++rx_tail >= macb->rx_ring_size)
> >                             rx_tail = 0;
> >                     reclaim_rx_buffers(macb, rx_tail);
> @@ -252,7 +261,6 @@ static int macb_recv(struct eth_device *edev)
> >                             rx_tail = 0;
> >                     }
> >             }
> > -           barrier();
> >     }
>  
> >     return 0;

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox

Reply via email to