[parisc-linux] FW: Porting LWP

[Neulinger, Nathan]

Anyone have any thoughts on this? In particular the question about whether
setjmp/longjmp preserve the stack. 

When I try to have the code use ucontext (getcontext/setcontext) instead, it
gets complaints from glibc about setcontext/getcontext being unimplemented.

Is parisc linux more like ia64 or x86? I'd presume that I might just have to
use the ucontext stuff, but that's going to be difficult if libc doesn't
support it.

-- Nathan

------------------------------------------------------------
Nathan Neulinger                       EMail:  nneul@umr.edu
University of Missouri - Rolla         Phone: (573) 341-4841
Computing Services                       Fax: (573) 341-4216


-----Original Message-----
From: chas williams [mailto:chas@cmf.nrl.navy.mil] 
Sent: Monday, November 19, 2001 1:11 PM
To: Neulinger, Nathan
Subject: Re: Porting LWP 


>I know you've been involved with alot of the previous porting efforts in
>AFS. I was hoping you could give some tips on porting LWP.

just a few really, not really that many.

>I've been doing some work trying to get openafs functional on
>parisc_linux24, but LWP definately appears to be the problem at this point.
>(Note, openafs fully compiles, but anything using threads fails.)

lwp is probably the most difficult part of userland.  it truly is
a serious pain.  it would be nice if someone made lwp just use pthreads.
however, its just a gleam in someones eye for now.

>I did get the LWP_SP stuff corrected, had some trouble with that at first,
>but am not sure where else to concentrate on. I generally get segmentation
>faults on everything in savecontext.

the last time i worked on a parisc system i noted that that the 
stack grew UP and not down.  parisc might do the same.  the linux
swp code assumes that the stack grows down (atleast i am pretty sure
it does)

you usually can verify which way the stack goes with something like
the following

sub()
{
	int stack2;
	printf("%x\n", &stack2);
}

main()
{
	int stack1;
	printf("%x\n", &stack1);
	sub();
}

if stack grows up, stack2 is greater than stack1.  if stack grows down,
stack2 less than stack1.  (or so i think)

other than that, parisc might need more space saved on the stack.
you should also make sure you setjmp/longjmp preserve the stack info
(check any comments in the glibc source).  it doesnt on the ia64 and
i had to use ucontext() instead.