egcs vs gcc-2.95.x

egcs vs gcc-2.95.x

[Tom Rini]

Hello all.  I happened to be skimming asm/byteorder.h for some reason, and
noticed that we have ___arch__swab16/___arch__swab32 written, but #if 0'ed
because of an egcs bug.  A bit of grep'ing showed that set_bit (asm/bitops.h)
used to be #if 0'ed because of another egcs bug but isn't now.  So the
question is, is anyone still using egcs to compile a kernel?  If, so
what version (1.0.x or 1.1.x) and does anyone remember what version of
egcs produced the bug?

If nothing else, perhaps in 2.5 this should go away..

--
Tom Rini (TR1265)
http://gate.crashing.org/~trini/

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: egcs vs gcc-2.95.x

[Cort Dougan]

Is this a comment by me?  If so, that was a problem with
__builtin_return_address() not working in leaf functions which still exists
today, I believe.  The "fix" there is a hack to leave those outline in lieu
of the inline versions so that several functions were made non-leaf.

In some other projects I've used changed those around so that
__builtin_return_address() is redefined to call another stub function so
that it's always called in a non-leaf function.  It's cleaner and more
reliable, but not any prettier.

} Hello all.  I happened to be skimming asm/byteorder.h for some reason, and
} noticed that we have ___arch__swab16/___arch__swab32 written, but #if 0'ed
} because of an egcs bug.  A bit of grep'ing showed that set_bit (asm/bitops.h)
} used to be #if 0'ed because of another egcs bug but isn't now.  So the
} question is, is anyone still using egcs to compile a kernel?  If, so
} what version (1.0.x or 1.1.x) and does anyone remember what version of
} egcs produced the bug?
}
} If nothing else, perhaps in 2.5 this should go away..
}
} --
} Tom Rini (TR1265)
} http://gate.crashing.org/~trini/
}
} _______________________________________________
} linuxppc-commit mailing list
} linuxppc-commit@source.mvista.com
} http://source.mvista.com/cgi-bin/mailman/listinfo/linuxppc-commit

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: egcs vs gcc-2.95.x

[Tom Rini]

On Thu, Sep 06, 2001 at 12:30:22PM -0600, Cort Dougan wrote:

> Is this a comment by me?  If so, that was a problem with
> __builtin_return_address() not working in leaf functions which still exists
> today, I believe.  The "fix" there is a hack to leave those outline in lieu
> of the inline versions so that several functions were made non-leaf.

It's unclaimed.  The comment is:
/* alas, egcs sounds like it has a bug in this code that doesn't use the
   inline asm correctly, and can cause file corruption. Until I hear that
   it's fixed, I can live without the extra speed. I hope. */

--
Tom Rini (TR1265)
http://gate.crashing.org/~trini/

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/