Re: [ros-dev] Re: [ros-diffs] [ion] 14047: Dispatching & Queue Rewrite II:

[Alex Ionescu]

Gunnar Dalsnes wrote:
> [e-mail] wrote:
> > Dispatching & Queue Rewrite II:
> > - Rewrote wait code. It is now cleaner, more optimized and 
> >
> > faster. All waiting
> >       functions are now clearly differentiated instead of sharing code.
>
> Imo, sharing the wait code is cleaner... (I remember the time i merged 
> the old wait code into one routine, heh)

I admit it's cleaner, but speed is more important, imo. Debate that can 
go on for ages however :)
> These functions
> >   are called up to a dozen times a second, so having dedicated 
> >
> > code for each of
> >       them is a real boost in speed.
> >     - Fixed several queue issues, made a dedicated queue wait/wake 
> > function (you are not
> >       supposed to use KeWaitFor on a queue,
>
> I used KeWaitXxx in the queue impl. because i thought it was good to 
> have the wait code in one place only. Thou others find speed more 
> important:-D

This introduces hacks into KeWaitXxx (remember != WrQueue?? :P) and 
also, if you look inside a checked build of NT, you will find 
ASSERT(Header->Type != QueueObject) inside KeWaitXxx so that really made 
me unshare the code :)

Now that these changes are in, I have rewritten thread termination 
(based on Windows Internals Chapter 11) and thread reaping to use 
hypercrtical work queues. The code for the latter is much 
simpler/smaller and also quite faster, since the OS manages everything. 
Termination also works MUCH better, gets rid of the "terminate type" 
hacks that were added, and also flushes/delivers all APCs, timers and 
mutants (we only flushed mutants!). Windows Internals described the 
process in line-by-line pseudo-code so it was a great help.

I should commit that tonight.

Best regards,
Alex Ionescu