On Wed, 26 Sep 2001, jamal wrote:
> Is there any reason you are not using dev_queue_xmit()? (side
> benefits, you could hack this into using scatter gather schemes etc)
the reason is latency of messages, and reliability of sends. If we are in
the middle of a *nasty* crash that takes down the system down 10
microseconds later, then i wanted to make 100% sure the message gets out
and the bug can be debugged. The VGA console and the serial logging
console have such properties - messages are instantaneous. Networking is
very asynchron in nature, so a new network driver extension was needed to
achieve this goal.
There are crashes where there are no new IRQs processed after the crash,
so there is no chance for dev_queue_xmit to do any useful send. We must be
able to send packets even if the output interface's hardware queues are
full - and dev_queue_xmit relies on future device interrupts to send
queued packets.
this is not just theory - about 10% of the crashes i experience (during
development...) are such, and the majority of the 'hard to fix' crashes
have such properties, especially on busy remote servers.
Ingo
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sun Sep 30 2001 - 21:00:54 EST