Bootstrap 1.4RC2 Stage 1 problems on AMD K6 - with fix

mike808 · n00b Joined: 19 Feb 2003 Posts: 5

I typically got compile failures (segmentation fault) in iconvdata in GCC. But I've seen other reports of folks having issues with K6 CPUs. I saw the advice to set CHOST to i586 instead
of the default i686 value. I've also seen tips to set CFLAGS to use -march=k6 or -mcpu=k6
in the /etc/make.conf file.

So I did that. Started a clean re-install.
Still barfed in various places during the bootstrap.

So, I watched ... very ... carefully (using Ctrl-S and Ctrl-Q to start/stop the scrolling
display) since the console has no scrollback. :=(

And I saw, to my surprise, that there were still some programs magically thinking that I was using a i686 CPU. Like the part where GCC is detecting whether or not it can compile a program, which it decides it can't, since the i686 code it generated craps out. That pretty
much stops further progress on the bootstrap....

So where was the i686 information coming from? I tracked it down to /etc/make.globals.

So, the advice, IMHO, needs to be clearer - if you are changing CHOST or CFLAGS
in /etc/make.conf you *MUST* also change them in /etc/make.globals - for non i686 CPUs!

Even when the instructions and all of the other posts say not to ever, ever change /etc/make.globals unless you really know what you're doing.

Well, I didn't know what I was doing, but the critical piece of information I learned
was this:

If you have an AMD K6 class CPU (e.g. K6/200 or any non i686 CPU), you must set CHOST and CFLAGS in both /etc/make.conf and /etc/make.globals.

In my case I am using:

cyrillic · Posted: Sun Feb 23, 2003 4:52 am Post subject:

mike808 · n00b Joined: 19 Feb 2003 Posts: 5

That might be a good thing to put in the instructions for folks to send in bug reports.
It's kind of hard to send one in when the error message has scrolled off
the screen, and you don't know anything about this magical scrollback key sequence.

Thanks for the tip.

sindre · Guru Joined: 01 Nov 2002 Posts: 315 Location: Norway

I'm in the process of installing Gento 1.4rc2 on my k6-3 450mhz. I had similar problems during bootstrap.

What I noticed, was that when emerging portage in the bootstrap process, my make.conf and make.globals get overwritten.

I thought it might be the CONFIG_PROTECT env thing, so I unset it. This didn't help, maybe if I set it to something special?

Anyway, I suspended the bootstrap after it automatically emerged portage, and edited the make.* files again (also changed chost and mcpu in make.globals as they defaults to i686).

It's running now, don't know if it will work or not (I'll report when it's done). Hope this can be sorted out.

sindre · Guru Joined: 01 Nov 2002 Posts: 315 Location: Norway

It didn't work out, gave me a compilation error. I decided to try to use gcc-3.2.1-r7 rather than r6. Upon editing the build to change ~x86 to x86, I noticed something regarding k6 and k6-2 processors. It would automatically change -march to i586 for those. I have a k6-3 so I decided to add a similar line for -march=k6-3.

Now it seems to have passed gcc complation without problems as all I see is lines regarding glibc. I'm not sure exactly how many of the changes were needed to make it work, since I included them all in the last try.

sindre · Guru Joined: 01 Nov 2002 Posts: 315 Location: Norway

Got a kernel crash... don't know why. Anyway it seems the bootstrap.sh takes care of the CFLAGS/CHOST thing, it stores them so it doesn't matter that the files get overwritten.

I'm tring again, this time I'll try the gcc-3.2.1-r6 again with the added k6-3 line. Gives me another chance to diagnose the problem so I can make a bug-report if neccesary.

mike808 · n00b Joined: 19 Feb 2003 Posts: 5

sindre · Guru Joined: 01 Nov 2002 Posts: 315 Location: Norway

I might have to use -march=i586 too then. I'm trying different versions of gcc and glibc, and running bootstrap.sh with 1.5 argument. Makes it run the first half of the process. Afterwards you can run it with argument 2, to execute the other half. If I get nowhere doing this I'll just back down to -march=i586

RobertTheBruce · n00b Joined: 14 Feb 2003 Posts: 41

I thought about using the -mcpu=k6-2 and -march=k6-2 but then decided to go with -march=i586 and Ihaven't had any problems at all with that option.

sindre · Guru Joined: 01 Nov 2002 Posts: 315 Location: Norway

So do you need to use -march=i586 permanently or just for the bootstrap?

This seems very strange, it kind of fails at random places.. I'll try bootstrapping with -march=i586. If this doesn't work, I'll test my RAM bricks...

It seems unlikely to be my hardware, as other k6* owners have similar problems.

sindre · Guru Joined: 01 Nov 2002 Posts: 315 Location: Norway

OK, I got through bootstrap and emerge system. The following ebuilds needed modification to compile:

needles · Posted: Thu Feb 27, 2003 3:51 am Post subject: Fix for K6-x woes

I have a K6-2 500, and am using Gentoo 1.4_rc2. I've had a glibc compile problem as mentioned above.

I dug around in Bugzilla and found this to be the most common suggestion for fixing the problems I've been running into:

ravenq · Apprentice Joined: 19 Feb 2003 Posts: 161

sindre · Guru Joined: 01 Nov 2002 Posts: 315 Location: Norway

Seems my struggle to get through the install was a complete waste of time. I booted up in my new and shiny gentoo system... Noticed I had compiled in the wrong network card, so I started recompiling the kernel. Midway in the kernel compilation... SEGFAULT!
:cry:

Seems anything but -march=i586 will make the system unstable/compile errors. Wonder if -march=i586 with -mcpu=k6-3 will be ok? Or -march=i586 -m3dnow -mmmx. I don't want to compile it all once more just to get another segfault thrown at me...

@needles: Why would you install gcc-3.2.2_preanything, when gcc-3.2.2 is done and avialible?

SpamMonkey · Posted: Sun Mar 02, 2003 10:04 am Post subject:

I'm currently trying to install Gentoo on a K6-2 333, 112mb Ram and an 8.4gb HD (sounds easy so far)..

But, after reading this thread I'm concerned that I'll run into problems when compiling.

At the moment I've simply set -march="i586" in my make.conf file so I'll keep my fingers crossed and post back if it works (or not)..

l8rs
Gavin
_________________
Solve two of the worlds problems..
Feed the homeless to the hungry

sindre · Guru Joined: 01 Nov 2002 Posts: 315 Location: Norway

I compiled it all with -march=i586 -mmmx -m3dnow etc.
Everything's fine until I try to emerge qt. It fails with a segfault... Seems someone with athlon processors have experienced the same, looking at the bug reports. Think I'll just have to experiment with the CFLAGS and see if I can make it work...

If I don't find any solution I might remerge gcc with no optimisatioin at all. I ran memtest86 for 5 hours or so without errors, so I think my RAM bricks are fine. Maybe I should test the processor too...

sindre · Guru Joined: 01 Nov 2002 Posts: 315 Location: Norway

Seems my cpu is not entirely stable... I ran cpuburn, and I got a kernel panic a little while later. I'm not sure what to do as it's not overclocked or anything. Maybe I'll have to get some better cooling.