[PATCH 1/2] net: num_dma_maps is not used

[PATCH 1/2] net: num_dma_maps is not used

[Eric Dumazet]

Get rid of num_dma_maps in struct skb_shared_info, as it seems unused.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
---
 include/linux/skbuff.h |    3 ---
 net/core/skb_dma_map.c |    1 -
 2 files changed, 4 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 7305da9..7485058 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -194,9 +194,6 @@ struct skb_shared_info {
 	unsigned short  gso_type;
 	__be32          ip6_frag_id;
 	union skb_shared_tx tx_flags;
-#ifdef CONFIG_HAS_DMA
-	unsigned int	num_dma_maps;
-#endif
 	struct sk_buff	*frag_list;
 	struct skb_shared_hwtstamps hwtstamps;
 	skb_frag_t	frags[MAX_SKB_FRAGS];
diff --git a/net/core/skb_dma_map.c b/net/core/skb_dma_map.c
index 8623492..7adb623 100644
--- a/net/core/skb_dma_map.c
+++ b/net/core/skb_dma_map.c
@@ -30,7 +30,6 @@ int skb_dma_map(struct device *dev, struct sk_buff *skb,
 			goto unwind;
 		sp->dma_maps[i + 1] = map;
 	}
-	sp->num_dma_maps = i + 1;
 
 	return 0;
 

Re: [PATCH 1/2] net: num_dma_maps is not used

[David Miller]

From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Fri, 05 Jun 2009 16:03:35 +0200

> Get rid of num_dma_maps in struct skb_shared_info, as it seems unused.
> 
> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>

It's going to be used when architectures implement optimized
versions of skb_dma_map(), which can thus virtually map
page vectors via IOMMU's and thus the resulting number of
DMA address/length pairs will be smaller than (num_frags + 1)

Similarly to how dma_map_sg() works.

Re: [PATCH 1/2] net: num_dma_maps is not used

[David Miller]

From: David Miller <davem@davemloft.net>
Date: Fri, 05 Jun 2009 18:09:49 -0700 (PDT)

> From: Eric Dumazet <eric.dumazet@gmail.com>
> Date: Fri, 05 Jun 2009 16:03:35 +0200
> 
>> Get rid of num_dma_maps in struct skb_shared_info, as it seems unused.
>> 
>> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
> 
> It's going to be used when architectures implement optimized
> versions of skb_dma_map(), which can thus virtually map
> page vectors via IOMMU's and thus the resulting number of
> DMA address/length pairs will be smaller than (num_frags + 1)
> 
> Similarly to how dma_map_sg() works.

But we aren't doing that yet and I have no idea when I'll get
to it, so I've applied your patch for now, thanks!