This is only a tiny part of the issue.
> This re-assembly simplifies some code, but it not only has a bad impact
> on memory management, it also involves a "useless" copy operation.
That "useless" operation is somewhat temporary - its only useless because
we dont reassemble and checksum in one pass. Also ATM already and HIPPI
very soon will be giving us 65280 byte frames that are not fragmented.
> UDP packets on receive without reassembling them - that would get better
> NFS performance due to getting rid of the copy, and it would also result
> in better network buffer allocation.
Its a partial fix.
What is really needed is something like
buffer=vreserve(65536); /* Allocate 64K of address space */
err=vfill(buffer, len, GFP_..); /* Put pages in where needed */
if(err==-ENOMEM) /* No pages */
vfree(buffer)
and the following other operations
for(i=vhead(buffer);i!=NULL;i=vnext(i))
{
scatter_gather[x]->ptr=virt_to_bus(vptr(i));
scatter_gather[x]->len=virt_to_bus(vlen(i));
}
x=simple_vbuf(x) /* Test if buffer has a simple linear
translation to physical */
The big problem we hit is that we need to rewrite a pile of DMA driven network
drivers to do scatter gather or at least to spot problems in buffers they
intend to DMA. Generally speaking it will be fine - something like
skb=skb_dmalinear(skb)
can return the buffer for most cases (not overlapping a page where the
data is), or a copy for the rare case it isnt and NULL for awkward cases where
we have to copy and have no RAM. That will make fixing up most drivers simple.
Alan