On Fri, Feb 10, 2006 at 12:34:10PM +0700, Mulyadi Santosa wrote: > Hi Jim > > It's me again. here I send you another patch for the vde_pcap_inject.c. > Nothing fancy here, I only did:
> - move out "open" and "close" from relay() and put them on main() > before entering for() loop. IMHO, this will make file operation a bit > faster This one I've thought about myself. Your way makes vde_pcap_inject more efficient. My way makes it usable if vde_pcap_inject is run before the kernel module is loaded. The idea is that vde_pcap_inject can be used everywhere vde_pcap can be, but in addition if you have vde-inject then it can and will make use of that as well. > > - I use __builtin_expect magic to optimize the "if" condition before > writing to /proc/vdeinject. here, inside relay() we assume that > vdeinject is likely exist. > What is __buiultin_expect() ? I've never seen it before. Your tests to make sure the fd is valid are unnecessary. perror() causes vde_pcap_inject to abort with an error, so any code called later on is guarranteed that fd is valid. Of course this means that vde_pcap_inject can only be used in combo with vde-inject, which I'm not sure is a good thing. E.g. a freeBSD host lacks vde-inject, so this would make vde_pcap_inject linux-specific. > PS: I can't check it thoroughly since my libpcap version (0.7.2, default > on Redhat 9. Yes I know it is old) doesn't have > pcap_get_selectable_fd(), so I can't compile it. > Change it to pcap_fileno(), and it should compile and work identically. (The difference is that pcap_get_selectable_fd() adds some basic sanity checks to make sure the fd is selectable before returning it. See http://www.tcpdump.org/lists/workers/2004/06/msg00093.html for more information.) I'll probably add an #ifdef for this. > regards > > Mulyadi -- Infinite complexity begets infinite beauty. Infinite precision begets infinite perfection. _______________________________________________ Qemu-devel mailing list Qemu-devel@nongnu.org http://lists.nongnu.org/mailman/listinfo/qemu-devel
