mi...@me... said:
> I've made the Linux application runner utility for Windows that I was
> working on over the summer available on the web.
Very cool. I'll fiddle with my site some to add pointers to your stuff.
> The basic difference between them is that 'Version1' executes the
> syscalls in the Linux application's process space and 'Version2'
> executes the syscalls in the monitor's process space.
With UML, it has to be in the app. You had only one thread, so you could do
syscalls wherever you wanted. The UML tracing thread has to serve all threads
in the kernel, so when a syscall sleeps, it has to be in the app.
> Only a very small subset of the Linux syscalls have actually been
> implemented and the ones that have been implemented are buggy!
That's OK. I've already got the system calls covered :-)
I'm curious as to why, once you had decided to do this, you didn't just link
the kernel in to get the real system calls. That would have resulted in a
completely independent implementation of UML... On Windows, no less...
> It's
> somewhat embarrassing releasing code in such a pre-alpha condition :P
You should have seen some early versions of UML...
In retrospect, I'm glad I didn't get slasdotted until last week. Much
earlier, and a lot of people could have been turned off by bugs, things that
didn't work, and things that weren't there. Now, the bugs aren't that
noticable, there aren't that many missing things, and I can claim that those
are on the way. The one downside of having it happen last week was that was
the middle of the stack overflow thing, but most slashdotters are probably
running 2.2, and didn't notice that.
Jeff
|
