[2.5.72] Oops on x86_64 running 32-bit code

[2.5.72] Oops on x86_64 running 32-bit code

[Bryan O'Sullivan]

Tried running BitKeeper under vanilla 2.5.72:

PML4 7ac6c067 PGD 7a872067 PMD 0
Oops: 0010 [1Ü
CPU 1
Pid: 3236, comm: bk Not tainted
RIP: 0010:[<0000000000000000>Ü [<0000000000000000>Ü
RSP: 0018:0000010037f88368  EFLAGS: 00010216
RAX: 0000000000004000 RBX: 000001007d4f2500 RCX: 0000000000000000
RDX: 000001007d4f2500 RSI: 0000000000000000 RDI: 0000000000000000
RBP: 0000000000000000 R08: 000001007a8e6000 R09: 0000000000000000
R10: 00000000009a0004 R11: 0000000000000000 R12: ffffffff80132f80
R13: 0000010037f881a8 R14: 0000010037f881a8 R15: 000001007e8e99c0
FS:  000000000054c480(0000) GS:ffffffff8041fd80(005b) knlGS:00000000a001b280
CS:  0010 DS: 002b ES: 002b CR0: 000000008005003b
CR2: 0000000000000000 CR3: 000000007ff14000 CR4: 00000000000007a0
Process bk (pid: 3236, stackpage=1007cd31a90)
Stack: 0000000000000000 0000000000000000 0000000000000441 00000000ffffbb00
       0000000000000246 0000000000000000 00000000ffffb698 00000000ffffb600
       0000000000000004 00000000ffffe405
Call Trace:<ffffffff80113b38>ädefault_do_nmi+56ü <ffffffff8011c787>ädo_nmi+87ü
       <ffffffff801126dc>änmi+112ü <ffffffff801320f2>äload_balance+34ü
                                                                                
                                                                                
Code:  Bad RIP value.

Re: [2.5.72] Oops on x86_64 running 32-bit code

[Andi Kleen]

On Wed, Jun 18, 2003 at 03:40:17PM -0700, Bryan O'Sullivan wrote:
> Call Trace:<ffffffff80113b38>??default_do_nmi+56?? <ffffffff8011c787>??do_nmi+87??
>        <ffffffff801126dc>??nmi+112?? <ffffffff801320f2>??load_balance+34??

Looks quite garbled. Is it reproducible?

-Andi

Re: [2.5.72] Oops on x86_64 running 32-bit code

[Bryan O'Sullivan]

On Wed, 2003-06-18 at 15:45, Andi Kleen wrote:

> Looks quite garbled. Is it reproducible?

Not yet.  I'll let you know if I see anything.

On an unrelated note, the trouble with running BK in 32-bit mode seems
to be back:

~ $ bk help
system: No child processes
system: No child processes
less: Bad file descriptor
system: No child processes

This time, there are no syslog errors or other reports to indicate
what's up.  I can't strace it to see where it's dying, because 64-bit
strace won't handle 32-bit code.

	<b

Re: [2.5.72] Oops on x86_64 running 32-bit code

[Andi Kleen]

On Wed, Jun 18, 2003 at 03:51:11PM -0700, Bryan O'Sullivan wrote:
> This time, there are no syslog errors or other reports to indicate
> what's up.  I can't strace it to see where it's dying, because 64-bit
> strace won't handle 32-bit code.

If it's a SuSE system you can use strace32, otherwise just copy
over a 32bit strace binary from some i386 box.

-Andi

Re: [2.5.72] Oops on x86_64 running 32-bit code

[Bryan O'Sullivan]

On Wed, 2003-06-18 at 16:06, Andi Kleen wrote:

> If it's a SuSE system you can use strace32, otherwise just copy
> over a 32bit strace binary from some i386 box.

Ugh - strace won't work, either.

I've slightly narrowed the problem down.  I'm using glibc 2.3.2 with TLS
and vsyscall support, but if I copy over an older 32-bit glibc,
everything is happy.

My guess is that 32-bit vsyscalls are still b0rked in some way under
2.5.72.

	<b