Bug 6262 – fcpci produces unresolved symbol __udivdi3

• Product: Gentoo Linux
• Status: RESOLVED FIXED

[Description Andreas Koehler 2002-08-09 14:03:25 UTC]

on my gentoo 1.3b-r1 (gcc 3.1) the fcpci-ebuild produces us __udivdi3.
the definite solution is simply to replace the tools.c.diff with mine.
it really works :)

[Comment 1 Andreas Koehler 2002-08-09 14:05:24 UTC]

Created attachment 2949 [details, diff]
replaces glibc-routine (imho)

please dont patch anything with this, just replace

[Comment 2 Bart Verwilst 2002-08-09 15:27:28 UTC]

euhm.. what's that? should i save your attachment in place of the current tools.c.diff?  

[Comment 3 Andreas Koehler 2002-08-09 22:56:31 UTC]

lot of stress here... eehm. yeah, just replace the current files/tools.c.diff
with my attachment.
i hope i did not confuse you too much (was it the name of the attachment :-?).

[Comment 4 Alexander Holler 2002-08-10 05:43:24 UTC]

I've just had a look on the new patch. I think this includes a risk of an 
overflow. And I can't believe that gcc 3.1 does not include support for type 
long long. 
 

[Comment 5 Bart Verwilst 2002-08-10 06:15:35 UTC]

Hm... Want to adjust the patch so it works pretty please? We have +- 4 days until i'll make the iso's, and this prog should go on them :o) Thanks! 

[Comment 6 Alexander Holler 2002-08-10 06:20:52 UTC]

Just as explaining what I mean: in the new patch j*1000/HZ is replaced with  
j*(1000/HZ)+(j*(1000%HZ))/HZ. I don't know the range for jiffies (j) or HZ but  
for example if HZ is 600 than jiffies would be multiplied by 400 in the second  
term. 
 
Couldn't that __udivdi3 problem solved on another way? I think I've read 
somthing about that problem with gcc 3.x in another (totally different) 
thread. But I can't test it, because I'm only running 2.9x. 
 

[Comment 7 Andreas Koehler 2002-08-11 04:06:36 UTC]

sorry, but i dont have the time to work further on this the next four weeks.
so holler, plz arbeite dran.
btw i think its the kernel that does not support long long, not gcc3.1 (imho).

[Comment 8 Alexander Holler 2002-08-14 03:37:56 UTC]

I can't work on that problem, because I don't have it. So if the missing 
routine should be part of the kernel, please tell us what kernel you are 
using. 
 
I've send an e-mail to AVM telling them your problem and patch. That's all I 
can do. 
 
 

[Comment 9 Alexander Holler 2002-08-15 15:03:10 UTC]

I've got response from AVM,

they say the following should work for 3.x:

> unsigned long os_msec (void) {
>
>      return jiffies * (1000 / HZ);
> } /* os_msec */

But they didn't say if that do it with 2.9x too. Maybe a #ifdef GCC_3 (I don't 
know the correct name) would be the best.

Alexander

[Comment 10 Alexander Holler 2002-08-15 18:08:16 UTC]

BTW: I'm not good as maintainer, I've no access to cvs. 

[Comment 11 Mark Guertin 2002-10-05 11:15:28 UTC]

I think this has been adjusted in glibc AFAIK.  Azarah, can you confirm this?

[Comment 12 Martin Schlemmer (RETIRED) 2002-10-12 12:03:32 UTC]

Should be for x86 and ppc (sys-libs/glibc-2.2.5-r7 as of current for ppc).

[Comment 13 Andreas Koehler 2002-10-12 14:25:55 UTC]

never change a running system, especially if it is the only way to connect to
the internet. it works for me with the new glibc, and because i was the bad guy
opening the bug i will close it now.

[Comment 14 Seemant Kulleen (RETIRED) 2002-12-16 23:25:22 UTC]

closed

[Comment 15 Martin Schlemmer (RETIRED) 2002-12-17 13:22:34 UTC]

Its an issue with gcc3, but we did patch glibc a while back already.