Guest Post @ hexblog
Author: David Zimmer
Date: 12.20.13 - 5:06pm
Just a heads up I did a guest post over at hexblog. (local pdf)
I also did a some bench marking and played around with some optimization ideas a bit. Getting the number of functions in the idb using a two way CopyData calls was 7794 ticks. CopyData one way and sendmessage return value 5660, and just using SendMessage with numeric args alone 2905. (On my system 1 tick = 1/3579545 seconds).
SendMessage using a custom message and numeric args only was the best performance, but the prototype only allows for passing a function call index and one numeric arg. (Apparently you cant get cute and multiplex a function call as the low byte of the uMsg or it wont be delivered to the remote process and I am to lazy to register that many custom messages just to gain a second arg). As is, I was able to implement 13 of the 36 api over to fast call. If I start using memory mapped files, its possible to switch them all away from copydata, but it would take more benchmarking to see if that had value. Also the text based command structure is easier to debug than a binary format that would be used with memmapped files, so thats a consideration too.
I still stand by my decision to move away from sockets. The code is simplier and its rare to need to control it from another machine. I also already have this same api implemented at the end of a socket interface in my IDAVBScript plugin. One project that uses it is the OllySync plugin which will sync your ida disasm view as you step through code or hit break points in Olly in a VM. (Great for malware analysis)