Skip to contentSkip to conversation

NS83820 2.6.0-test5 driver seems unstable on IA64

DISCUSSION Manfred Spraul 4 participants 5 messages Sep 26, 2003 – Sep 27, 2003

lore
Share & tools
lore
#0 lore raw
Quoted 4 lines · David

Fine, then we should have something like an rx_copybreak scheme in the ns83820 driver too.

Is that really the right solution? Add a full-packet copy to every driver?
IMHO the fastest solution would be to copy only the ip & tcp headers, and keep the rest as it is. And preferable in the network core, to avoid having to copy&paste that into every driver.

--
Manfred

#1 lore raw
Quoted 1 line · On Fri, 26 Sep 2003 08:16:36 +0200 Manfred Spraul <[email protected]>…

Is that really the right solution? Add a full-packet copy to every driver?

In the short term, yes it is.

Quoted 3 lines

IMHO the fastest solution would be to copy only the ip & tcp headers, and keep the rest as it is. And preferable in the network core, to avoid having to copy&paste that into every driver.

You're absolutely correct, but implementing this is far from trivial. There are assumptions all over the place that the protocol data immediately follows the protocol headers.

If you're willing to do all of the auditing and work, have at it. :-)

#2 lore raw
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · David S. Miller

David> On Fri, 26 Sep 2003 08:16:36 +0200 Manfred Spraul

Quoted 2 lines · David> <[email protected]> wrote:
Quoted 2 lines

Is that really the right solution? Add a full-packet copy to every driver?

David> In the short term, yes it is.

I know nothing about the ns83820, but page 22 of the data sheet
(http://www.national.com/pf/DP/DP83820.html) suggests that it would be possible to setup _two_ descriptors for each incoming packet: the first one would cover the Ethernet header (14 bytes), the second the rest. That way, IP-header would be 8-byte aligned (since the ns83820's DMA engine seems to require 8-byte alignment for incoming data).

If so, this would let you get the IP-header and on aligned at the cost of extra descriptor fetching. The cost of fetching the extra descriptor will be big enough that you'd only want to do this when unaligned accesses are expensive, but hopefully it would be cheaper than copying the entire packet.

	--david
#3 lore raw
In reply to David S. Miller
Quoted 6 lines · David S. Miller
Quoted 1 line · On Fri, 26 Sep 2003 08:16:36 +0200 Manfred Spraul <[email protected]>…

Is that really the right solution? Add a full-packet copy to every driver?

In the short term, yes it is.

What about aligning the packet directly in the rx buffer (by memmoving the entire packet to buf+2) instead of copying to another skb? This appears to be a) more than 2 times faster b) easy to implement.

Ivan.

#4 lore raw
Quoted 3 lines · On Fri, 26 Sep 2003 18:38:27 +0400 Ivan Kokshaysky <[email protected]>…

What about aligning the packet directly in the rx buffer (by memmoving the entire packet to buf+2) instead of copying to another skb? This appears to be a) more than 2 times faster b) easy to implement.

That's another possibility.