[uClibc]Powerpc uclibc

[David Schleef]

On Wed, May 09, 2001 at 12:52:39PM +0300, Michael Shmulevich wrote:
> There are however several more serious problems, that I wonder how they 
> work for your PPC machine:
> 
> - mmap: realised through stack arguments passing, a problem which I have 
> asked earlier. This doesn't work on 2.2 at all -- neither for MIPS nor 
> for PPC. I had to use _syscal#() macros for them, in MIPS case it was 
> _syscall6() and for PPC it was _syscall0() (!!!) since this is how it 
> was in glibc.

Actually, this is _syscall6() on powerpc as well.  Or rather, it is
what _syscall6() would be if you extrapolated from _syscall5().  It
is now fixed.

> - ioctl: struct termios definitions goes in direct contradiction with 
> kerbel definitions: too many fields, wrong sizes, incorrect ioctl() 
> parameters. It  was seen when bbox's shell tried to ask isatty(0), which 
> returned 0. I had to rewrite the struct in bits/ioctls.h to represent 
> kernel's one.

<sarcasm>
yup.  This is one of glibc's attempts at being cute.  Of course,
we all need to make sure that an interface that hasn't been
extended in 20 years is rewritten to be extensible.
</sarcasm>

I see no reason to be binary compatible with glibc in this case,
so using the kernel header is preferable.  Then we don't have
to have that __stupid__ hack in ioctl() to test for TCGETS and
friends.


> So, David, I giess you have luck to work with 2.4 otherwise you'd pe 
> passing through the same hell I am passing right now.

Heh, no.  I just didn't test it as well as you did.  I was
mainly doing it in the gaps between when I had other things
to do, and those gaps disappeared for a while.  I mainly just
looked at the test programs.




dave...