From: Stanislaw Gruszka
> Sent: 30 July 2019 10:36
...
> > + len = pkt_stat.pkt_len + pkt_offset;
> > + skb = dev_alloc_skb(len);
> > + if (WARN_ONCE(!skb, "rx routine starvation\n"))
> > goto next_rp;
> >
> > /* put the DMA data including rx_desc from phy to new skb */
> > - skb_put_data(new, skb->data, new_len);
> > + skb_put_data(skb, rx_desc, len);
>
> Coping big packets it quite inefficient. What drivers usually do is
> copy only for small packets and for big ones allocate new rx buf
> (drop packet alloc if fail) and pas old buf to network stack via
> skb_add_rx_frag(). See iwlmvm as example.
If you have to do iommu setup/teardown then the breakeven point
for (not) copying may be surprisingly large.
You do need to do the measurements on a range of hardware.
Coping is also likely to affect the L1 cache - unless you can
copy quickly without polluting the cache.
It is all 'swings and roundabouts'.
David
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT,
UK
Registration No: 1397386 (Wales)
