> > (with message queuing one can queue, say, a 16M network packet and not
> > expect the kernel to copy before sending... or transmitting to a different
> > process)
>
> Look harder. You can share areas using mmap() and the kernel transmit already
> does copy/checksum from user space.
<sigh> - but is it possible to mmap (as shared) a currently active
segment? Or will a 'message memory pool' have to be created?
(and how does one grow/shrink a mmap'd region? That's one thing I still
haven't figured out... apparently growing/shrinking mmap'd files uses
'ftruncate' but what about nonfile shared memory regions?)
> As to stuff like DMA from user space - yes we cannot do that right now.
> mmuio.c makes that possible for limited cases. That still has limits
> as the memory has to be locked down for the duration of the transfer
> and you can't pin arbitary amounts of memory down
No biggy - I'm not all that concerned with it (haven't the skills to
handle DMA coallesing <sp> code in any extent anyways)... I'll wait for
someone else to solve this problem <G>.
G'day, eh?
- Teunis