User-mode Linux kernel port

As an addition: The stage tarball is a x86 (i386) one. Trying to use for 
example an athlon-xp stage tarball crashes with "cannot set up LDT for 
thread-local storage"

   // Joel


On Tue, 27 Sep 2005, Joel Palmius wrote:

> Lengthy.. Summary is that on an amd64 host, uml hangs on skas0 mode if the 
> host kernel has been patched with -v9-pre7. Using an unpatched vanilla host 
> kernel, skas0 boots. Why this is so I have no clue, ask someone else. 
> : -)
>
> Further, 2.6.12-bs11 refuses to attempt running in skas0 on an unpatched 
> host.
>
> For getting a working gentoo uml on my amd64 host, I now use: unpatched 
> 2.6.13.2 host and 2.6.13.2-bs2 guest with TT mode disabled. The guest kernel 
> had to be compiled in a 32-bit chroot.
>
> Read on for my fumbling attempts at debugging:
>
> ---
>
> I've tried to get SKAS0 to work with a gentoo root, but with limited success. 
> TT mode works fine, but SKAS0 hangs at:
>
>   ..
>  NET:  Registered protocol family 1
>  NET:  Registered protocol family 17
>   Initialized stdio console driver
>   Console initialized on /dev/tty0
>   Initializing software serial port version 1
>    ubda: unknown partition table
>    ubdb: unknown partition table
>   VFS: Mounted root (ext2 filesystem).
>
> Top says:
>
>   ..
>   16686 root      17   0  128m 5396 5200 R 87.4  0.3   0:27.07 linux
>   16690 root      15   0    64   64   64 T 12.3  0.0   0:03.86 linux
>
> Strace says:
>
> ...
> waitpid(16690, [{WIFSTOPPED(s) && WSTOPSIG(s) == SIGSEGV}], WSTOPPED) = 16690
> ptrace(PTRACE_GETREGS, 16690, 0, 0xa0a6f884) = 0
> ptrace(PTRACE_GETFPXREGS, 16690, 0, 0xa0a6f934) = 0
> ptrace(0x34 /* PTRACE_??? */, 16690, 0, 0xa0a6fb34) = 0
> rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
> rt_sigprocmask(SIG_UNBLOCK, [ALRM VTALRM IO], NULL, 8) = 0
> rt_sigprocmask(SIG_SETMASK, NULL, [], 8) = 0
> rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
> rt_sigprocmask(SIG_UNBLOCK, [ALRM VTALRM IO], [ALRM VTALRM IO], 8) = 0
> rt_sigprocmask(SIG_BLOCK, [], NULL, 8)  = 0
> rt_sigprocmask(SIG_SETMASK, NULL, [], 8) = 0
> rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
> rt_sigprocmask(SIG_SETMASK, NULL, [ALRM VTALRM IO], 8) = 0
> rt_sigprocmask(SIG_SETMASK, NULL, [ALRM VTALRM IO], 8) = 0
> rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
> rt_sigprocmask(SIG_UNBLOCK, [], [ALRM VTALRM IO], 8) = 0
> rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
> rt_sigprocmask(SIG_SETMASK, NULL, [ALRM VTALRM IO], 8) = 0
> rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
> rt_sigprocmask(SIG_UNBLOCK, [], [ALRM VTALRM IO], 8) = 0
> rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
> rt_sigprocmask(SIG_UNBLOCK, [ALRM VTALRM IO], [ALRM VTALRM IO], 8) = 0
> rt_sigprocmask(SIG_BLOCK, [], NULL, 8)  = 0
> rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
> rt_sigprocmask(SIG_SETMASK, NULL, [ALRM VTALRM IO], 8) = 0
> rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
> rt_sigprocmask(SIG_UNBLOCK, [], [ALRM VTALRM IO], 8) = 0
> rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
> rt_sigprocmask(SIG_UNBLOCK, [ALRM VTALRM IO], NULL, 8) = 0
> ptrace(PTRACE_SETREGS, 16690, 0, 0xa0a6f884) = 0
> ptrace(PTRACE_SETFPXREGS, 16690, 0, 0xa0a6f934) = 0
> ptrace(PTRACE_SYSCALL, 16690, 0, SIG_0) = 0
> --- SIGCHLD (Child exited) @ 0 (0) ---
> ...(repeated)
>
> So far I've tried these things, all on 2.6.13.1-skas3-v9-pre7 amd64 host:
>
> 2.6.13.1+bs2, TT mode, 2005.1 gentoo root: works
> 2.6.13.1+bs2, TT mode, 2004.0 gentoo root: works
> 2.6.13.1+bs2, TT mode, slackware 8.1 root from uml homepage: works
> 2.6.13.1+bs2, SKAS0 mode, 2005.1 gentoo root: hangs
> 2.6.13.1+bs2, SKAS0 mode, 2004.0 gentoo root: hangs
> 2.6.13.1+bs2, SKAS0 mode, slackware 8.1 root from uml homepage: works
>
> (skas0 works on slackware, but hangs on gentoo, trying gentoo's 
> usermode-sources for guest instead)
>
> 2.6.12-r2-bs11, TT mode, 2005.1 gentoo root: works
> 2.6.12-r2-bs11, TT mode, 2004.0 gentoo root: works
> 2.6.12-r2-bs11, TT mode, slackware 8.1 root: works
> 2.6.12-r2-bs11, SKAS0 mode, 2005.1 gentoo root: hangs
> 2.6.12-r2-bs11, SKAS0 mode, 2004.0 gentoo root: hangs
> 2.6.12-r2-bs11, SKAS0 mode, slackware 8.1 root: works
>
> (same results as with patched vanilla kernel as guest)
>
> Having tried with different roots and different guest kernels, I now try an 
> unpatched host kernel, thinking maybe there's a problem with having applied 
> the host patch. Dropping 2004.0 test case since it didn't make any 
> difference. Host kernel is now unpatched 2.6.13.2:
>
> 2.6.12-r2-bs11, TT mode, 2005.1 gentoo root: works
> 2.6.12-r2-bs11, TT mode, slackware 8.1 root: works
>
> However, now the guest kernel refuses to switch to SKAS0 mode. It runs in TT 
> despite that I have set mode=skas0 on arg line.
>
> Compiling new kernel tree to be sure I didn't choose any strange options. 
> This time I simply exclude the TT mode (it isn't enabled by default).
>
> 2.6.13.2-bs2, SKAS0 mode, 2005.1 gentoo root: WORKS!!
>
> Okidoki.. It's possible to get skas0 and gentoo to work together after all. 
> Why gentoo needs an unpatched host kernel, while slackware dont, I have no 
> clue.
>
> Apart from the above, I also tested a few cases on a 32-bit host. This was a 
> bit unstructured, but my impression was that the results are the same, ie 
> gentoo guest won't boot in skas0 on a patched host.
>
> Please suggest more test cases.
>
>  // Joel
>
>
>
> -------------------------------------------------------
> SF.Net email is sponsored by:
> Tame your development challenges with Apache's Geronimo App Server. Download
> it for free - -and be entered to win a 42" plasma tv or your very own
> Sony(tm)PSP.  Click here to play: http://sourceforge.net/geronimo.php
> _______________________________________________
> User-mode-linux-user mailing list
> User-mode-linux-user@...
> https://lists.sourceforge.net/lists/listinfo/user-mode-linux-user
>

On Tuesday 27 September 2005 16:02, Joel Palmius wrote:
> Lengthy.. Summary is that on an amd64 host, uml hangs on skas0 mode if the
> host kernel has been patched with -v9-pre7.

> Using an unpatched vanilla 
> host kernel, skas0 boots. Why this is so I have no clue, ask someone else.
Have you correctly added "skas0" to the cmd line (not mode=skas0)? That could 
be an easy explaination. If this is not the cause, I'll try to debug the 
problem.
> :-)
>
> Further, 2.6.12-bs11 refuses to attempt running in skas0 on an unpatched
> host.
No skas0 support is included - if you want that try -bb12-skas0.

-- 
Inform me of my mistakes, so I can keep imitating Homer Simpson's "Doh!".
Paolo Giarrusso, aka Blaisorblade (Skype ID "PaoloGiarrusso", ICQ 215621894)
http://www.user-mode-linux.org/~blaisorblade

	

	
		
___________________________________ 
Yahoo! Mail: gratis 1GB per i messaggi e allegati da 10MB 
http://mail.yahoo.it

I always start the uml/skas0 test case via a script, which contains:

nice linux32 ./linux-2613-2 ubd0=root_fs ubd1=swap_fs eth0=tuntap,,,192.168.10.254 mode=skas0 mem=256m rw

(Isn't this the right way to ask for skas0 mode?)

   // Joel


On Tue, 27 Sep 2005, Blaisorblade wrote:

> On Tuesday 27 September 2005 16:02, Joel Palmius wrote:
>> Lengthy.. Summary is that on an amd64 host, uml hangs on skas0 mode if the
>> host kernel has been patched with -v9-pre7.
>
>> Using an unpatched vanilla
>> host kernel, skas0 boots. Why this is so I have no clue, ask someone else.
> Have you correctly added "skas0" to the cmd line (not mode=skas0)? That could
> be an easy explaination. If this is not the cause, I'll try to debug the
> problem.
>> :-)
>>
>> Further, 2.6.12-bs11 refuses to attempt running in skas0 on an unpatched
>> host.
> No skas0 support is included - if you want that try -bb12-skas0.
>
> -- 
> Inform me of my mistakes, so I can keep imitating Homer Simpson's "Doh!".
> Paolo Giarrusso, aka Blaisorblade (Skype ID "PaoloGiarrusso", ICQ 215621894)
> http://www.user-mode-linux.org/~blaisorblade
>
>
>
>
>
> ___________________________________
> Yahoo! Mail: gratis 1GB per i messaggi e allegati da 10MB
> http://mail.yahoo.it
>
>
>
> -------------------------------------------------------
> This SF.Net email is sponsored by:
> Power Architecture Resource Center: Free content, downloads, discussions,
> and more. http://solutions.newsforge.com/ibmarch.tmpl
> _______________________________________________
> User-mode-linux-user mailing list
> User-mode-linux-user@...
> https://lists.sourceforge.net/lists/listinfo/user-mode-linux-user
>

Sorry for spamming about gentoo.. :-)

I'm now stuck at getting a functional system up and running. The problem I 
have is a catch 22: I cannot run an athlon-xp tarball in skas0 because it 
complains with the dreaded "cannot set up LDT for thread-local storage", 
and I don't want to run a tarball compiled in i386 mode because then I 
might as well run in TT mode for speed issues.

Having read similar posts on the subject, I've tried:

- Adding LD_ASSUME_KERNEL=2.4.1 to command line (didn't make any 
difference)
- Inserting export LD_ASSUME_KERNEL=2.4.1 in start scripts (didn't make 
any difference)
- Move /lib/tls (can't since it doesn't exist)

Further I've dug into the glibc ebuild. It seems it automatically enables 
the configure flag "--with-tls" if the machine is detected as being 
anything higher than i386.

This leads me to conclude I have to compile a new system from scratch to 
get optimizations working under skas0.

Questions are thus:

- Which software packets cause this crash? Is it only glibc?
- How should glibc be compiled (it with which configure flags?)

I am guessing "--without-tls --without-nptl". Anything else?

Or is there a guest kernel patch which might enable me to run an 
unmodified gentoo tarball (compiled for athlon-xp) under skas0?

   // Joel


On Tue, 27 Sep 2005, Joel Palmius wrote:

> As an addition: The stage tarball is a x86 (i386) one. Trying to use for 
> example an athlon-xp stage tarball crashes with "cannot set up LDT for 
> thread-local storage"
>
>  // Joel
>
>
> On Tue, 27 Sep 2005, Joel Palmius wrote:
>
>>  Lengthy.. Summary is that on an amd64 host, uml hangs on skas0 mode if the
>>  host kernel has been patched with -v9-pre7. Using an unpatched vanilla
>>  host kernel, skas0 boots. Why this is so I have no clue, ask someone else. 
>> :  -)
>>
>>  Further, 2.6.12-bs11 refuses to attempt running in skas0 on an unpatched
>>  host.
>>
>>  For getting a working gentoo uml on my amd64 host, I now use: unpatched
>>  2.6.13.2 host and 2.6.13.2-bs2 guest with TT mode disabled. The guest
>>  kernel had to be compiled in a 32-bit chroot.
>>
>>  Read on for my fumbling attempts at debugging:
>>
>>  ---
>>
>>  I've tried to get SKAS0 to work with a gentoo root, but with limited
>>  success. TT mode works fine, but SKAS0 hangs at:
>>
>>   ..
>> NET:   Registered protocol family 1
>> NET:   Registered protocol family 17
>>    Initialized stdio console driver
>>    Console initialized on /dev/tty0
>>    Initializing software serial port version 1
>>     ubda: unknown partition table
>>     ubdb: unknown partition table
>>    VFS: Mounted root (ext2 filesystem).
>>
>>  Top says:
>>
>>    ..
>>    16686 root      17   0  128m 5396 5200 R 87.4  0.3   0:27.07 linux
>>    16690 root      15   0    64   64   64 T 12.3  0.0   0:03.86 linux
>>
>>  Strace says:
>>
>>  ...
>>  waitpid(16690, [{WIFSTOPPED(s) && WSTOPSIG(s) == SIGSEGV}], WSTOPPED) =
>>  16690
>>  ptrace(PTRACE_GETREGS, 16690, 0, 0xa0a6f884) = 0
>>  ptrace(PTRACE_GETFPXREGS, 16690, 0, 0xa0a6f934) = 0
>>  ptrace(0x34 /* PTRACE_??? */, 16690, 0, 0xa0a6fb34) = 0
>>  rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
>>  rt_sigprocmask(SIG_UNBLOCK, [ALRM VTALRM IO], NULL, 8) = 0
>>  rt_sigprocmask(SIG_SETMASK, NULL, [], 8) = 0
>>  rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
>>  rt_sigprocmask(SIG_UNBLOCK, [ALRM VTALRM IO], [ALRM VTALRM IO], 8) = 0
>>  rt_sigprocmask(SIG_BLOCK, [], NULL, 8)  = 0
>>  rt_sigprocmask(SIG_SETMASK, NULL, [], 8) = 0
>>  rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
>>  rt_sigprocmask(SIG_SETMASK, NULL, [ALRM VTALRM IO], 8) = 0
>>  rt_sigprocmask(SIG_SETMASK, NULL, [ALRM VTALRM IO], 8) = 0
>>  rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
>>  rt_sigprocmask(SIG_UNBLOCK, [], [ALRM VTALRM IO], 8) = 0
>>  rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
>>  rt_sigprocmask(SIG_SETMASK, NULL, [ALRM VTALRM IO], 8) = 0
>>  rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
>>  rt_sigprocmask(SIG_UNBLOCK, [], [ALRM VTALRM IO], 8) = 0
>>  rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
>>  rt_sigprocmask(SIG_UNBLOCK, [ALRM VTALRM IO], [ALRM VTALRM IO], 8) = 0
>>  rt_sigprocmask(SIG_BLOCK, [], NULL, 8)  = 0
>>  rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
>>  rt_sigprocmask(SIG_SETMASK, NULL, [ALRM VTALRM IO], 8) = 0
>>  rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
>>  rt_sigprocmask(SIG_UNBLOCK, [], [ALRM VTALRM IO], 8) = 0
>>  rt_sigprocmask(SIG_BLOCK, [ALRM VTALRM IO], NULL, 8) = 0
>>  rt_sigprocmask(SIG_UNBLOCK, [ALRM VTALRM IO], NULL, 8) = 0
>>  ptrace(PTRACE_SETREGS, 16690, 0, 0xa0a6f884) = 0
>>  ptrace(PTRACE_SETFPXREGS, 16690, 0, 0xa0a6f934) = 0
>>  ptrace(PTRACE_SYSCALL, 16690, 0, SIG_0) = 0
>>  --- SIGCHLD (Child exited) @ 0 (0) ---
>>  ...(repeated)
>>
>>  So far I've tried these things, all on 2.6.13.1-skas3-v9-pre7 amd64 host:
>>
>>  2.6.13.1+bs2, TT mode, 2005.1 gentoo root: works
>>  2.6.13.1+bs2, TT mode, 2004.0 gentoo root: works
>>  2.6.13.1+bs2, TT mode, slackware 8.1 root from uml homepage: works
>>  2.6.13.1+bs2, SKAS0 mode, 2005.1 gentoo root: hangs
>>  2.6.13.1+bs2, SKAS0 mode, 2004.0 gentoo root: hangs
>>  2.6.13.1+bs2, SKAS0 mode, slackware 8.1 root from uml homepage: works
>>
>>  (skas0 works on slackware, but hangs on gentoo, trying gentoo's
>>  usermode-sources for guest instead)
>>
>>  2.6.12-r2-bs11, TT mode, 2005.1 gentoo root: works
>>  2.6.12-r2-bs11, TT mode, 2004.0 gentoo root: works
>>  2.6.12-r2-bs11, TT mode, slackware 8.1 root: works
>>  2.6.12-r2-bs11, SKAS0 mode, 2005.1 gentoo root: hangs
>>  2.6.12-r2-bs11, SKAS0 mode, 2004.0 gentoo root: hangs
>>  2.6.12-r2-bs11, SKAS0 mode, slackware 8.1 root: works
>>
>>  (same results as with patched vanilla kernel as guest)
>>
>>  Having tried with different roots and different guest kernels, I now try
>>  an unpatched host kernel, thinking maybe there's a problem with having
>>  applied the host patch. Dropping 2004.0 test case since it didn't make any
>>  difference. Host kernel is now unpatched 2.6.13.2:
>>
>>  2.6.12-r2-bs11, TT mode, 2005.1 gentoo root: works
>>  2.6.12-r2-bs11, TT mode, slackware 8.1 root: works
>>
>>  However, now the guest kernel refuses to switch to SKAS0 mode. It runs in
>>  TT despite that I have set mode=skas0 on arg line.
>>
>>  Compiling new kernel tree to be sure I didn't choose any strange options.
>>  This time I simply exclude the TT mode (it isn't enabled by default).
>>
>>  2.6.13.2-bs2, SKAS0 mode, 2005.1 gentoo root: WORKS!!
>>
>>  Okidoki.. It's possible to get skas0 and gentoo to work together after
>>  all. Why gentoo needs an unpatched host kernel, while slackware dont, I
>>  have no clue.
>>
>>  Apart from the above, I also tested a few cases on a 32-bit host. This was
>>  a bit unstructured, but my impression was that the results are the same,
>>  ie gentoo guest won't boot in skas0 on a patched host.
>>
>>  Please suggest more test cases.
>>
>>   // Joel
>> 
>> 
>>
>>  -------------------------------------------------------
>>  SF.Net email is sponsored by:
>>  Tame your development challenges with Apache's Geronimo App Server.
>>  Download
>>  it for free - -and be entered to win a 42" plasma tv or your very own
>>  Sony(tm)PSP.  Click here to play: http://sourceforge.net/geronimo.php
>>  _______________________________________________
>>  User-mode-linux-user mailing list
>>  User-mode-linux-user@...
>>  https://lists.sourceforge.net/lists/listinfo/user-mode-linux-user
>> 
>
>
> -------------------------------------------------------
> This SF.Net email is sponsored by:
> Power Architecture Resource Center: Free content, downloads, discussions,
> and more. http://solutions.newsforge.com/ibmarch.tmpl
> _______________________________________________
> User-mode-linux-user mailing list
> User-mode-linux-user@...
> https://lists.sourceforge.net/lists/listinfo/user-mode-linux-user
>

On Tuesday 27 September 2005 19:45, Blaisorblade wrote:
> On Tuesday 27 September 2005 16:02, Joel Palmius wrote:
> > Lengthy.. Summary is that on an amd64 host, uml hangs on skas0 mode if
> > the host kernel has been patched with -v9-pre7.

> > Using an unpatched vanilla
> > host kernel, skas0 boots. Why this is so I have no clue, ask someone
> > else.

> Have you correctly added "skas0" to the cmd line (not mode=skas0)? That
> could be an easy explaination.
In fact, on a patched x86_64 host it would try to run in SKAS3 mode on x86_64, 
which doesn't work to my knowledge.
> If this is not the cause, I'll try to debug 
> the problem.

Ok, I see you mention explicitly trying mode=skas0. Yes, it's really 
confusing, but is explained in vmlinux --help. And UML 2.6.13 (and even 
2.6.12-bs) at the very beginning says "UML running in XXX mode".

But ok, ok, ok, I'm dumb. I'll add the mode=skas0 for consistency, since 
everybody is getting burned by this.

> > :-)
> >
> > Further, 2.6.12-bs11 refuses to attempt running in skas0 on an unpatched
> > host.

> No skas0 support is included - if you want that try -bb12-skas0.

> > Apart from the above, I also tested a few cases on a 32-bit host. This was 
> > a bit unstructured, but my impression was that the results are the same, 
> > ie gentoo guest won't boot in skas0 on a patched host.
This is a lot different. In the above cases, the problem was skas3 + 64bit 
host = kaboom.

Here it's different. But again, using mode=skas0 doesn't work, so you didn't 
test what you wanted.
-- 
Inform me of my mistakes, so I can keep imitating Homer Simpson's "Doh!".
Paolo Giarrusso, aka Blaisorblade (Skype ID "PaoloGiarrusso", ICQ 215621894)
http://www.user-mode-linux.org/~blaisorblade


		
___________________________________ 
Yahoo! Messenger: chiamate gratuite in tutto il mondo 
http://it.messenger.yahoo.com

On Tuesday 27 September 2005 21:09, Joel Palmius wrote:
> I always start the uml/skas0 test case via a script, which contains:
>
> nice linux32 ./linux-2613-2 ubd0=root_fs ubd1=swap_fs
> eth0=tuntap,,,192.168.10.254 mode=skas0 mem=256m rw

> (Isn't this the right way to ask for skas0 mode?)
No, it isn't. Use skas0.

I've added mode=skas0 will be added in 2.6.13-bs3 and 2.6.14, but skas0 will 
stay.

However, calling it "skas0" rather than "mode=skas0" was wrong, and so many 
people got it wrong I realized this.
-- 
Inform me of my mistakes, so I can keep imitating Homer Simpson's "Doh!".
Paolo Giarrusso, aka Blaisorblade (Skype ID "PaoloGiarrusso", ICQ 215621894)
http://www.user-mode-linux.org/~blaisorblade

	

	
		
___________________________________ 
Yahoo! Mail: gratis 1GB per i messaggi e allegati da 10MB 
http://mail.yahoo.it

On Wednesday 28 September 2005 10:22, Joel Palmius wrote:
> Sorry for spamming about gentoo.. :-)

Don't worry.

Short summary for Jeff: he gets in SKAS0 the message
"cannot set up LDT for thread-local storage" at startup (correct Joel?), but 
he isn't using NPTL, he's rather using LinuxThreads with TLS support.

On Gentoo, this is possible when using at least glibc-2.3.4 (2.3.3 ebuilds 
don't try to enable this).

And (Joel please confirm, gathered from your previous posts) he is running 
2.6.13.2-bs2, which includes the really neeeded modify_ldt() fix (missing 
break in switch) I recently sent.

This is not very different from Rob's problem with Ubuntu, or from problems on 
Debian Sarge, other than they have a fallback while Gentoo doesn't.

Is skas-hold-own-ldt a solution to that?
> I'm now stuck at getting a functional system up and running. The problem I
> have is a catch 22: I cannot run an athlon-xp tarball in skas0 because it
> complains with the dreaded "cannot set up LDT for thread-local storage",
> and I don't want to run a tarball compiled in i386 mode because then I
> might as well run in TT mode for speed issues.

Let's not exaggerate, I think the difference speed-wise between TT and SKAS0 
is much bigger. But expecting that root_fs to work is reasonable.

> Having read similar posts on the subject, I've tried:

> - Adding LD_ASSUME_KERNEL=2.4.1 to command line (didn't make any
> difference)
> - Inserting export LD_ASSUME_KERNEL=2.4.1 in start scripts (didn't make
> any difference)
> - Move /lib/tls (can't since it doesn't exist)
All these things help when the problem is a NPTL glibc, and make ld.so prefer 
the LinuxThreads version (i.e. the one in /lib rather than in /lib/tls). But 
your case is different - you have the problem with LinuxThreads (at least so 
I think - if the tarball is compiled with USE="nptl nptl-only" it won't work 
anyway until I fix NPTL support).

> Further I've dug into the glibc ebuild. It seems it automatically enables
> the configure flag "--with-tls" if the machine is detected as being
> anything higher than i386.

Hmm, possible, depending on the ebuild. In fact, there's some difference 
between enabling NPTL and enabling TLS, starting from 2.3.4 ebuilds.

I've checked glibc-2.3.5.ebuild (I run Gentoo, so I have the portage tree 
installed) and with LinuxThreads (no NPTL), want_tls is true for >=i486, but 
-tls is enabled starting from i686. But for your purposes, this is more or 
less the same.

> This leads me to conclude I have to compile a new system from scratch to
> get optimizations working under skas0.

> Questions are thus:

> - Which software packets cause this crash? Is it only glibc?

I think yes, but possibly add statically linked one which include the glibc 
code for this (I wouldn't expect a statically linked binary to use this 
stuff, but actually errno is a per-thread variable and can be implemented 
this way).

> - How should glibc be compiled (it with which configure flags?)
> I am guessing "--without-tls --without-nptl". Anything else?

At that point I'd add --without-__thread. But I'm not sure if other programs 
will be so happy... but maybe yes. A userspace program can use only __thread, 
and Gentoo allows disabling it, so it should work. But I'd use the below 
road.

> Or is there a guest kernel patch which might enable me to run an
> unmodified gentoo tarball (compiled for athlon-xp) under skas0?

This is possible, I think Jeff has some, but possibly they've not yet been 
merged. Will see with him.

>    // Joel

-- 
Inform me of my mistakes, so I can keep imitating Homer Simpson's "Doh!".
Paolo Giarrusso, aka Blaisorblade (Skype ID "PaoloGiarrusso", ICQ 215621894)
http://www.user-mode-linux.org/~blaisorblade


		
___________________________________ 
Yahoo! Messenger: chiamate gratuite in tutto il mondo 
http://it.messenger.yahoo.com