Roland Dreier wrote:
If you're going to use vmap(), then you might as well use vmalloc().

I might mislead Eli here, sorry for that.

the issue is with consuming address space, which is very limited on
32-bit systems (there is often less than 128 MB of vmalloc available
total).  However in this case it is probably OK.  I guess we don't
want to allocate these structures independently and take another
pointer deref for every send -- although I would be curious to know if
it actually costs much.

So what's your suggestion? maybe we go with vmalloc on 64-bit systems and on 32-bits system limit the user to what kmalloc can provide (which is about ~850 size rings with the tx_buf consuming 8*18 u64's + pointer.

Or

_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general

To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general

Reply via email to