Re: [Syllable-kernel] upcoming Syllable patches..

[Jake H. <jh...@an...>]

Maritimus wrote:
> 
> Ok I don't really know this stuff, but will this have a fallback if i586
> time stamp code doesn't work with a particular x86 CPU?
> 
> This looked interesting:
> http://www.cs.helsinki.fi/linux/linux-kernel/2001-31/0542.html

Hmm..  That is interesting.  I know that the "rdtsc" instruction was 
introduced with the Pentium so it should work on all machines that 
Syllable currently runs on.  Whether it's _accurate_ is a completely 
different question.  In particular, it's definitely not accurate under 
VMware as it appears to be an emulated instruction, not the real TSC 
clock, and it slows down massively when the virtual machine manager has 
to do a lot of emulation of system mode functions.  The Linux code 
appears to fall back to the PIT timer with a warning if the TSC loses 
too many ticks.  The comment at the top of the Linux code for 
gettimeofday() says that it achieves microsecond accuracy even with the 
PIT timer so I'm hoping to be able to get the accuracy I need for 
lmbench one way or the other.

Anyway, I'm going to hold off on the timer code for a day or two while I 
get caught up and submit diffs for my already completed fixes.  I've 
also been writing doxygen comment blocks for all the kernel functions 
after I got sick of looking at all those empty comment blocks.  It's a 
bit tedious but the end result should be quite nice looking.  I'm 
tagging all of the private functions with "\ingroup Internal" so they 
show up in their own module, and I will make a group for each set of 
related kernel functions, e.g. Areas, Semaphores, Networking, VFS.

-Jake