Hello Steffen,

On 28.11.23 17:29, Steffen Trumtrar wrote:
> rx_buffer gets dma_alloc'ed but is never dma_map'ed and therefor not
> flushed before it is initially used.
> 
> Map the rx_buffer when the macb is initialized and unmap it on ether_halt.
> 
> While at it, cleanup the dma_alloc_coherent rx_ring/tx_ring, too.
> 
> Signed-off-by: Steffen Trumtrar <[email protected]>
> ---
>  drivers/net/macb.c | 37 ++++++++++++++++++++++++++++---------
>  1 file changed, 28 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/net/macb.c b/drivers/net/macb.c
> index 260c1e806a..92f78f7253 100644
> --- a/drivers/net/macb.c
> +++ b/drivers/net/macb.c
> @@ -63,10 +63,13 @@ struct macb_device {
>       unsigned int            tx_head;
>  
>       void                    *rx_buffer;
> +     dma_addr_t              rx_buffer_phys;
>       void                    *tx_buffer;
>       void                    *rx_packet_buf;
>       struct macb_dma_desc    *rx_ring;
> +     dma_addr_t              rx_ring_phys;
>       struct macb_dma_desc    *tx_ring;
> +     dma_addr_t              tx_ring_phys;
>       struct macb_dma_desc    *gem_q1_descs;
>  
>       int                     rx_buffer_size;
> @@ -181,7 +184,7 @@ static int gem_recv(struct eth_device *edev)
>               barrier();
>               status = macb->rx_ring[macb->rx_tail].ctrl;
>               length = MACB_BFEXT(RX_FRMLEN, status);
> -             buffer = macb->rx_buffer + macb->rx_buffer_size * macb->rx_tail;
> +             buffer = (void *)macb->rx_buffer_phys + macb->rx_buffer_size * 
> macb->rx_tail;

For GEM-type NICs, rx_buffer_size is PKTSIZE (1518 bytes currently), which is 
not a multiple
of the cache line size of the 64 bytes cache line on the ZynqMP's Cortex-A53 
the driver
is supposed to support.

>               dma_sync_single_for_cpu(macb->dev, (unsigned long)buffer, 
> length,
>                                       DMA_FROM_DEVICE);

This means this could potentially invalidate adjacent buffer contents.


>       }
> @@ -891,8 +908,8 @@ static int macb_probe(struct device *dev)
>  
>       macb_init_rx_buffer_size(macb, PKTSIZE);

^ Here's where PKTSIZE comes from. I'd be in favor of changing the global 
PKTSIZE
definition to be a multiple of 64 bytes (or use DMA_ALIGNMENT, but this isn't
correctly set for ARM yet.  I just sent out a patch for that).

Cheers,
Ahmad

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |


Reply via email to