Re: [PATCH 2/2] staging: et131x: Fix 64bit tx dma address handling

From: Jeffrey Ladouceur
Date: Wed Oct 17 2012 - 20:31:10 EST

On Wed, Oct 17, 2012 at 5:15 PM, Mark Einon <mark.einon@xxxxxxxxx> wrote:
> The driver checks that the device can handle 64bit DMA addressing in
> et131x_pci_setup(), but then assumes that the top dword of a tx dma
> address is always zero when creating a dma mapping in nic_send_packet().
> Fix the mapping to use the higher dword of the dma_addr_t returned by
> dma_map_single() and skb_frag_dma_map().
>
> Also remove incorrect comments stating that dma_map_single() only returns
> a 32 bit address.
>
> Signed-off-by: Mark Einon <mark.einon@xxxxxxxxx>
> ---
> drivers/staging/et131x/et131x.c | 102 +++++++++++++++------------------------
> 1 file changed, 39 insertions(+), 63 deletions(-)
>
> diff --git a/drivers/staging/et131x/et131x.c b/drivers/staging/et131x/et131x.c
> index f0fe4bd..83643be 100644
> --- a/drivers/staging/et131x/et131x.c
> +++ b/drivers/staging/et131x/et131x.c
> @@ -3285,6 +3285,7 @@ static int nic_send_packet(struct et131x_adapter *adapter, struct tcb *tcb)
> struct skb_frag_struct *frags = &skb_shinfo(skb)->frags[0];
> unsigned long flags;
> struct phy_device *phydev = adapter->phydev;
> + dma_addr_t dma_addr;
>
> /* Part of the optimizations of this send routine restrict us to
> * sending 24 fragments at a pass. In practice we should never see
> @@ -3314,77 +3315,46 @@ static int nic_send_packet(struct et131x_adapter *adapter, struct tcb *tcb)
> * doesn't seem to like large fragments.
> */
> if (skb_headlen(skb) <= 1514) {
> - desc[frag].addr_hi = 0;
> /* Low 16bits are length, high is vlan and
> unused currently so zero */
> desc[frag].len_vlan = skb_headlen(skb);
> -
> - /* NOTE: Here, the dma_addr_t returned from
> - * dma_map_single() is implicitly cast as a
> - * u32. Although dma_addr_t can be
> - * 64-bit, the address returned by
> - * dma_map_single() is always 32-bit
> - * addressable (as defined by the pci/dma
> - * subsystem)
> - */
> - desc[frag++].addr_lo =
> - dma_map_single(&adapter->pdev->dev,
> - skb->data,
> - skb_headlen(skb),
> - DMA_TO_DEVICE);
> + dma_addr = dma_map_single(&adapter->pdev->dev,
> + skb->data,
> + skb_headlen(skb),
> + DMA_TO_DEVICE);
> + desc[frag].addr_lo = dma_addr & 0xFFFFFFFF;
> + desc[frag].addr_hi = dma_addr >> 32;

Maybe use macros defined in kernel.h instead:

desc[frag].addr_lo = lower_32_bits(dma_addr);
desc[frag].addr_hi = upper_32_bits(dma_addr);

A few more instances below.


> + frag++;
> } else {
> - desc[frag].addr_hi = 0;
> desc[frag].len_vlan = skb_headlen(skb) / 2;
> -
> - /* NOTE: Here, the dma_addr_t returned from
> - * dma_map_single() is implicitly cast as a
> - * u32. Although dma_addr_t can be
> - * 64-bit, the address returned by
> - * dma_map_single() is always 32-bit
> - * addressable (as defined by the pci/dma
> - * subsystem)
> - */
> - desc[frag++].addr_lo =
> - dma_map_single(&adapter->pdev->dev,
> - skb->data,
> - (skb_headlen(skb) / 2),
> - DMA_TO_DEVICE);
> - desc[frag].addr_hi = 0;
> + dma_addr = dma_map_single(&adapter->pdev->dev,
> + skb->data,
> + (skb_headlen(skb) / 2),
> + DMA_TO_DEVICE);
> + desc[frag].addr_lo = dma_addr & 0xFFFFFFFF;
> + desc[frag].addr_hi = dma_addr >> 32;
> + frag++;
>
> desc[frag].len_vlan = skb_headlen(skb) / 2;
> -
> - /* NOTE: Here, the dma_addr_t returned from
> - * dma_map_single() is implicitly cast as a
> - * u32. Although dma_addr_t can be
> - * 64-bit, the address returned by
> - * dma_map_single() is always 32-bit
> - * addressable (as defined by the pci/dma
> - * subsystem)
> - */
> - desc[frag++].addr_lo =
> - dma_map_single(&adapter->pdev->dev,
> - skb->data +
> - (skb_headlen(skb) / 2),
> - (skb_headlen(skb) / 2),
> - DMA_TO_DEVICE);
> + dma_addr = dma_map_single(&adapter->pdev->dev,
> + skb->data +
> + (skb_headlen(skb) / 2),
> + (skb_headlen(skb) / 2),
> + DMA_TO_DEVICE);
> + desc[frag].addr_lo = dma_addr & 0xFFFFFFFF;
> + desc[frag].addr_hi = dma_addr >> 32;
> + frag++;
> }
> } else {
> - desc[frag].addr_hi = 0;
> - desc[frag].len_vlan =
> - frags[i - 1].size;
> -
> - /* NOTE: Here, the dma_addr_t returned from
> - * dma_map_page() is implicitly cast as a u32.
> - * Although dma_addr_t can be 64-bit, the address
> - * returned by dma_map_page() is always 32-bit
> - * addressable (as defined by the pci/dma subsystem)
> - */
> - desc[frag++].addr_lo = skb_frag_dma_map(
> - &adapter->pdev->dev,
> - &frags[i - 1],
> - 0,
> - frags[i - 1].size,
> - DMA_TO_DEVICE);
> + desc[frag].len_vlan = frags[i - 1].size;
> + dma_addr = skb_frag_dma_map(&adapter->pdev->dev,
> + &frags[i - 1],
> + 0,
> + frags[i - 1].size,
> + DMA_TO_DEVICE);
> + desc[frag].addr_lo = dma_addr & 0xFFFFFFFF;
> + desc[frag].addr_hi = dma_addr >> 32;
> + frag++;
> }
> }
>
> @@ -3611,6 +3581,7 @@ static inline void free_send_packet(struct et131x_adapter *adapter,
> unsigned long flags;
> struct tx_desc *desc = NULL;
> struct net_device_stats *stats = &adapter->net_stats;
> + dma_addr_t dma_addr;
>
> if (tcb->flags & fMP_DEST_BROAD)
> atomic_inc(&adapter->stats.broadcast_pkts_xmtd);
> @@ -3631,8 +3602,13 @@ static inline void free_send_packet(struct et131x_adapter *adapter,
> (adapter->tx_ring.tx_desc_ring +
> INDEX10(tcb->index_start));
>
> + dma_addr = desc->addr_lo;
> +
> + if (sizeof(dma_addr_t) == sizeof(u64))
> + dma_addr += ((dma_addr_t)desc->addr_hi) << 32;

Probably end result the same, but don't see + too often. Not usually
the way we think of it.

dma_addr |= (dma_addr_t)desc->addr_hi << 32;

> +
> dma_unmap_single(&adapter->pdev->dev,
> - desc->addr_lo,
> + dma_addr,
> desc->len_vlan, DMA_TO_DEVICE);
>
> add_10bit(&tcb->index_start, 1);
> --
> 1.7.9.5
>
> _______________________________________________
> devel mailing list
> devel@xxxxxxxxxxxxxxxxxxxxxx
> http://driverdev.linuxdriverproject.org/mailman/listinfo/devel
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/