Linux MIPS Architecture development
 help / color / mirror / Atom feed
* gcc RPM missing crtbegin.o
@ 1998-04-26  0:01 Alex deVries
  1998-04-26  1:48 ` ralf
  0 siblings, 1 reply; 8+ messages in thread
From: Alex deVries @ 1998-04-26  0:01 UTC (permalink / raw)
  To: SGI Linux


Hm. I've re-installed gcc from the RPM that's in the mustang directory on
linus (2.7.2-1, build date is Dec 4 03:23:00).  "Yipee", says I, "I can
now become an productive member of society". 

But, the gcc RPM doesn't have crtbegin.o, which is a bit odd.  It's not
the end of the world, I snarfed it from the cross compiling RPM for i386.
Was there a reason it wasn't included? 

I'm having some really wonky behaviour on my filesystem with .72; files
are disappearing off my filesystem left right and center.  For instance,
crtbegin.o has been on the system since I first got the machine last fall.
e2fsck's always come back clean. Now that I got my compiler working, I'm
running .1.91 again.

I also upgraded to the glibc 2.0.7 RPMs, which seems to have messed up
tcsh.  It'd be real nice to get gdb and/or strace working to debug things
like this.

- Alex

-- 
Alex deVries
""romantic engsoc guy who runs marathons" - csilcock@chat.carleton.ca
http://www.engsoc.carleton.ca/~adevries/ .

^ permalink raw reply	[flat|nested] 8+ messages in thread

* Re: gcc RPM missing crtbegin.o
  1998-04-26  0:01 gcc RPM missing crtbegin.o Alex deVries
@ 1998-04-26  1:48 ` ralf
  1998-04-26  3:00   ` Alex deVries
  0 siblings, 1 reply; 8+ messages in thread
From: ralf @ 1998-04-26  1:48 UTC (permalink / raw)
  To: Alex deVries; +Cc: SGI Linux

On Sat, Apr 25, 1998 at 08:01:33PM -0400, Alex deVries wrote:

> Hm. I've re-installed gcc from the RPM that's in the mustang directory on
> linus (2.7.2-1, build date is Dec 4 03:23:00).  "Yipee", says I, "I can
> now become an productive member of society". 
> 
> But, the gcc RPM doesn't have crtbegin.o, which is a bit odd.  It's not
> the end of the world, I snarfed it from the cross compiling RPM for i386.
> Was there a reason it wasn't included? 

It's spelled ``bug''.  This has been reported before and I'm shure I've
uploaded the apropriate fixes in form of new patch sets, source and binary
rpms.

>Clickety<

[ralf@lappi ralf]$ ssh -C -l ralf linus.linux.sgi.com
Enter passphrase for RSA key 'root@rio': 
Last login: Thu Apr 23 08:28:04 1998 from vulcan.ko.ivm.net
Welcome to linus.linux.sgi.com !

This machine serves as:
        o A  repository for SGI/Linux source code          /src/cvs
        o A  web server for SGI/Linux (www.linux.sgi.com)  /src/web
        o An ftp server for SGI/Linux (ftp.linux.sgi.com)  /src/ftp

Please, use the /src and /work directories for storing big stuff
Home dirs are on / and it fills quite fast.

Shell access to core developers and contributors via ssh only.
Happy hacking!

                        ariel@sgi.com   Jul, 1997
ralf@linus:ralf# locate gcc-2.7.2 | fgrep -e -3 | fgrep .rpm
/src/ftp/pub/redhat/redhat-5.0/RPMS/mips/gcc-2.7.2-3.mips.rpm
/src/ftp/pub/redhat/redhat-5.0/RPMS/mipsel/gcc-2.7.2-3.mipsel.rpm
/src/ftp/pub/redhat/redhat-5.0/SRPMS/gcc-2.7.2-3.src.rpm

ralf@linus:ralf#

Easy, isn't it?

> I'm having some really wonky behaviour on my filesystem with .72; files
> are disappearing off my filesystem left right and center.  For instance,
> crtbegin.o has been on the system since I first got the machine last fall.
> e2fsck's always come back clean. Now that I got my compiler working, I'm
> running .1.91 again.

I've fixed *shitloads* of bugs since .72, not too mention a dramatic increase
in performance.  You're right, filesystems on Indys (not other MIPS boxes)
had the problem of a certain ``volatilibility'', but that's gone for me.
If somebody still sees fs corruption on Indys running 2.1.91, please report.

> I also upgraded to the glibc 2.0.7 RPMs, which seems to have messed up
> tcsh.  It'd be real nice to get gdb and/or strace working to debug things
> like this.

Strace is working.  See the CVS.

There is no glibc 2.0.7.

  Ralf

^ permalink raw reply	[flat|nested] 8+ messages in thread

* Re: gcc RPM missing crtbegin.o
  1998-04-26  1:48 ` ralf
@ 1998-04-26  3:00   ` Alex deVries
  1998-04-26  3:39     ` ralf
  0 siblings, 1 reply; 8+ messages in thread
From: Alex deVries @ 1998-04-26  3:00 UTC (permalink / raw)
  To: ralf; +Cc: SGI Linux


On Sun, 26 Apr 1998 ralf@uni-koblenz.de wrote:
> It's spelled ``bug''.  This has been reported before and I'm shure I've
> uploaded the apropriate fixes in form of new patch sets, source and binary
> rpms.
> 

Okay, I had missed the announcement.

> Easy, isn't it?

Always. :)

> I've fixed *shitloads* of bugs since .72, not too mention a dramatic increase
> in performance.  You're right, filesystems on Indys (not other MIPS boxes)
> had the problem of a certain ``volatilibility'', but that's gone for me.

Yup, just bootstrapping from .72 to .91 was a bit troubling.

> If somebody still sees fs corruption on Indys running 2.1.91, please report.

I do have a problem that just came up.  Here it is, hand typed.  I was
doing a lot of work on sdc, which is an external 3GB drive that I know
works well.  It hasn't reported any problems in e2fsck.  There was a lot
of activity when this happened:

EXT2-fs error (device 08:11): ext2_find_entry: bad entry in directory
#41117: rec_len is smaller than minimal - offset = 0, inode=0, rec_len=0,
name_len = 0
EXT2-fs error (device 08:11): ext2_find_entry: bad entry in directory
#41117: rec_len is smaller than minimal - offset = 0, inode=0, rec_len=0,
name_len = 0
page fault from irq handler: 0000
$0 : 00000000 1000fc00 00001000 ffffffe0
$4 : 00000020 00000000 1000fc00 00000001
$8 : 1000fc00 1000001f 00000000 00000007
$12: 40000000 8bf50020 1000fc00 00000001
$16: 00000000 00001000 abf56020 8bf53800
$20: 00000002 bfbc0003 1fffffff bfb90000
$24: 00000002 0fb6f710 
$28: 00008000 08009d28 8bf57e70 080f3b5c
epc   :88020fc0
Status: 1000fc02
Cause: 00000008
Aiee, killing interrupt handler
Kernel panic: Attempted to kill the idle task!
In swapper task - not syncing

this is with my own compiled version of .91.

> Strace is working.  See the CVS.

Alright. 

- A

^ permalink raw reply	[flat|nested] 8+ messages in thread

* Re: gcc RPM missing crtbegin.o
  1998-04-26  3:00   ` Alex deVries
@ 1998-04-26  3:39     ` ralf
  1998-04-26  4:34       ` Alex deVries
  0 siblings, 1 reply; 8+ messages in thread
From: ralf @ 1998-04-26  3:39 UTC (permalink / raw)
  To: Alex deVries; +Cc: SGI Linux

On Sat, Apr 25, 1998 at 11:00:38PM -0400, Alex deVries wrote:

> > I've fixed *shitloads* of bugs since .72, not too mention a dramatic increase
> > in performance.  You're right, filesystems on Indys (not other MIPS boxes)
> > had the problem of a certain ``volatilibility'', but that's gone for me.
> 
> Yup, just bootstrapping from .72 to .91 was a bit troubling.

Ok.

> > If somebody still sees fs corruption on Indys running 2.1.91, please report.
> 
> I do have a problem that just came up.  Here it is, hand typed.  I was
> doing a lot of work on sdc, which is an external 3GB drive that I know
> works well.  It hasn't reported any problems in e2fsck.  There was a lot
> of activity when this happened:

Did you already fsck the filesystems running .91?  The corruption in the
older kernels was happening silently, so the boot fsck may have been skipped
even though the fs was corrupted.

> EXT2-fs error (device 08:11): ext2_find_entry: bad entry in directory
> #41117: rec_len is smaller than minimal - offset = 0, inode=0, rec_len=0,
> name_len = 0
> EXT2-fs error (device 08:11): ext2_find_entry: bad entry in directory
> #41117: rec_len is smaller than minimal - offset = 0, inode=0, rec_len=0,
> name_len = 0
> page fault from irq handler: 0000
> $0 : 00000000 1000fc00 00001000 ffffffe0
> $4 : 00000020 00000000 1000fc00 00000001
> $8 : 1000fc00 1000001f 00000000 00000007
> $12: 40000000 8bf50020 1000fc00 00000001
> $16: 00000000 00001000 abf56020 8bf53800
> $20: 00000002 bfbc0003 1fffffff bfb90000
> $24: 00000002 0fb6f710 
> $28: 00008000 08009d28 8bf57e70 080f3b5c
                                  ^^^^^^^^
> epc   :88020fc0
         ^^^^^^^^

Can you disassemble (mips-linux-objdump -d vmlinux) the kernel image and
tell me where these two addresses are pointing to?  Just send me twenty
lines or so around the addresses.  (In general that's the right thing to
do when the machine bombs with a register dump.)

(Also I'm pretty shure that you misstyped $28 and $29 or something really
bad happend.)

When did the kernel bomb?  During boot or?  What activity was going on
at the time when it happend?

  Ralf

^ permalink raw reply	[flat|nested] 8+ messages in thread

* Re: gcc RPM missing crtbegin.o
  1998-04-26  3:39     ` ralf
@ 1998-04-26  4:34       ` Alex deVries
  1998-04-26 11:14         ` ralf
  1998-04-26 12:09         ` ralf
  0 siblings, 2 replies; 8+ messages in thread
From: Alex deVries @ 1998-04-26  4:34 UTC (permalink / raw)
  To: ralf; +Cc: SGI Linux


On Sun, 26 Apr 1998 ralf@uni-koblenz.de wrote:
> > I do have a problem that just came up.  Here it is, hand typed.  I was
> > doing a lot of work on sdc, which is an external 3GB drive that I know
> > works well.  It hasn't reported any problems in e2fsck.  There was a lot
> > of activity when this happened:
> 
> Did you already fsck the filesystems running .91?  The corruption in the
> older kernels was happening silently, so the boot fsck may have been skipped
> even though the fs was corrupted.

Yup, I've had this problem twice now, both with .91. I'm pretty sure I can
reproduce it (but not until I'm at my desk tomorrow, since I need to hit
the little thumbtack to reboot it.  I still think SGI should have shipped
the Indys with thumbtacks.)

> > $0 : 00000000 1000fc00 00001000 ffffffe0
> > $4 : 00000020 00000000 1000fc00 00000001
> > $8 : 1000fc00 1000001f 00000000 00000007
> > $12: 40000000 8bf50020 1000fc00 00000001
> > $16: 00000000 00001000 abf56020 8bf53800
> > $20: 00000002 bfbc0003 1fffffff bfb90000
> > $24: 00000002 0fb6f710 
> > $28: 00008000 08009d28 8bf57e70 080f3b5c
>                                   ^^^^^^^^
> > epc   :88020fc0
>          ^^^^^^^^
> tell me where these two addresses are pointing to?  Just send me twenty
> lines or so around the addresses.  (In general that's the right thing to
> do when the machine bombs with a register dump.)
> (Also I'm pretty shure that you misstyped $28 and $29 or something really
> bad happend.)

Uh, there was nothing at 080f3b5c or nearby, so I'm pretty sure that that
was misttyped as 880f3b5c.  Here are both:



88020f44 <r4k_dma_cache_wback_inv_pc>:
88020f44:	3c028814 	lui	$v0,0x8814
88020f48:	8c425564 	lw	$v0,21860($v0)
88020f4c:	27bdffe0 	addiu	$sp,$sp,-32
88020f50:	afb00010 	sw	$s0,16($sp)
88020f54:	00808021 	move	$s0,$a0
88020f58:	afb10014 	sw	$s1,20($sp)
88020f5c:	00a08821 	move	$s1,$a1
88020f60:	0222102b 	sltu	$v0,$s1,$v0
88020f64:	14400007 	bnez	$v0,88020f84 <r4k_dma_cache_wback_inv_pc+40>
88020f68:	afbf0018 	sw	$ra,24($sp)
88020f6c:	3c028817 	lui	$v0,0x8817
88020f70:	8c42fab4 	lw	$v0,-1356($v0)
88020f74:	0040f809 	jalr	$v0
88020f78:	00000000 	nop
88020f7c:	0a0083f7 	j	88020fdc <r4k_dma_cache_wback_inv_pc+98>
88020f80:	00000000 	nop
88020f84:	40066000 	mfc0	$a2,$12
88020f88:	34c10001 	ori	$at,$a2,0x1
88020f8c:	38210001 	xori	$at,$at,0x1
88020f90:	40816000 	mtc0	$at,$12
88020fa0:	3c02a000 	lui	$v0,0xa000
88020fa4:	3c048814 	lui	$a0,0x8814
88020fa8:	8c84556c 	lw	$a0,21868($a0)
88020fac:	8c420000 	lw	$v0,0($v0)
88020fb0:	02111021 	addu	$v0,$s0,$s1
88020fb4:	00041823 	negu	$v1,$a0
88020fb8:	02032824 	and	$a1,$s0,$v1
88020fbc:	00431024 	and	$v0,$v0,$v1
88020fc0:	bcb50000 	cache	0x15,0($a1)
88020fc4:	14a2fffe 	bne	$a1,$v0,88020fc0 <r4k_dma_cache_wback_inv_pc+7c>
88020fc8:	00a42821 	addu	$a1,$a1,$a0
88020fcc:	40866000 	mtc0	$a2,$12
88020fdc:	3c028813 	lui	$v0,0x8813
88020fe0:	8c423d90 	lw	$v0,15760($v0)
88020fe4:	02002021 	move	$a0,$s0
88020fe8:	8c420008 	lw	$v0,8($v0)
88020fec:	0040f809 	jalr	$v0
88020ff0:	02202821 	move	$a1,$s1
88020ff4:	8fbf0018 	lw	$ra,24($sp)
88020ff8:	8fb10014 	lw	$s1,20($sp)
88020ffc:	8fb00010 	lw	$s0,16($sp)




880f3b00:	afb00018 	sw	$s0,24($sp)
880f3b04:	afa50014 	sw	$a1,20($sp)
880f3b08:	8e620000 	lw	$v0,0($s3)
880f3b0c:	8c52008c 	lw	$s2,140($v0)
880f3b10:	8c550074 	lw	$s5,116($v0)
880f3b14:	ac45007c 	sw	$a1,124($v0)
880f3b18:	9663007e 	lhu	$v1,126($s3)
880f3b1c:	8c57004c 	lw	$s7,76($v0)
880f3b20:	10600027 	beqz	$v1,880f3bc0 <dma_setup+ec>
880f3b24:	245e0070 	addiu	$s8,$v0,112
880f3b28:	00008821 	move	$s1,$zero
880f3b2c:	8e6400ec 	lw	$a0,236($s3)
880f3b30:	2462ffff 	addiu	$v0,$v1,-1
880f3b34:	04400018 	bltz	$v0,880f3b98 <dma_setup+c4>
880f3b38:	0000a021 	move	$s4,$zero
880f3b3c:	3c161fff 	lui	$s6,0x1fff
880f3b40:	36d6ffff 	ori	$s6,$s6,0xffff
880f3b44:	00808021 	move	$s0,$a0
880f3b48:	8e040000 	lw	$a0,0($s0)
880f3b4c:	3c028817 	lui	$v0,0x8817
880f3b50:	8c42fac0 	lw	$v0,-1344($v0)
880f3b54:	0040f809 	jalr	$v0
880f3b58:	24051000 	li	$a1,4096
880f3b5c:	8e020000 	lw	$v0,0($s0)
880f3b60:	00561024 	and	$v0,$v0,$s6
880f3b64:	ae420000 	sw	$v0,0($s2)
880f3b68:	8e020008 	lw	$v0,8($s0)
880f3b6c:	26940001 	addiu	$s4,$s4,1
880f3b70:	30423fff 	andi	$v0,$v0,0x3fff
880f3b74:	ae420004 	sw	$v0,4($s2)
880f3b78:	8e030008 	lw	$v1,8($s0)
880f3b7c:	26100010 	addiu	$s0,$s0,16
880f3b80:	26520010 	addiu	$s2,$s2,16
880f3b84:	9662007e 	lhu	$v0,126($s3)
880f3b88:	2442ffff 	addiu	$v0,$v0,-1
880f3b8c:	0054102a 	slt	$v0,$v0,$s4
880f3b90:	1040ffed 	beqz	$v0,880f3b48 <dma_setup+74>
880f3b94:	02238821 	addu	$s1,$s1,$v1
880f3b98:	24020012 	li	$v0,18
880f3b9c:	afd10020 	sw	$s1,32($s8)

> When did the kernel bomb?  During boot or?  What activity was going on
> at the time when it happend?

This was well after the boot.  The first time it happened, I was
installing a lot of .src.rpms onto sdc while I was compiling on sdc.  The
second time (after e2fsck'ing sda1 (1GB) and scd (3GB), which takes like
30 mins), I was just compiling by building RPMs.

- Alex

^ permalink raw reply	[flat|nested] 8+ messages in thread

* Re: gcc RPM missing crtbegin.o
  1998-04-26  4:34       ` Alex deVries
@ 1998-04-26 11:14         ` ralf
  1998-04-26 12:09         ` ralf
  1 sibling, 0 replies; 8+ messages in thread
From: ralf @ 1998-04-26 11:14 UTC (permalink / raw)
  To: Alex deVries; +Cc: SGI Linux

On Sun, Apr 26, 1998 at 12:34:54AM -0400, Alex deVries wrote:

> On Sun, 26 Apr 1998 ralf@uni-koblenz.de wrote:

> Yup, I've had this problem twice now, both with .91. I'm pretty sure I can
> reproduce it (but not until I'm at my desk tomorrow, since I need to hit
> the little thumbtack to reboot it.  I still think SGI should have shipped
> the Indys with thumbtacks.)

If you think that was a missfeature, take a look at SNI's RM200C.  It's has
the reset button mounted directly on the motherboard.  Which is why my
machine has never been closed.  And really f*cked idea for something which
had NT as OS option ;-)

Ho humm...  If I remember right the Indy has some watchdog onboard.  Time
to use it.

> > > $0 : 00000000 1000fc00 00001000 ffffffe0
> > > $4 : 00000020 00000000 1000fc00 00000001
> > > $8 : 1000fc00 1000001f 00000000 00000007
> > > $12: 40000000 8bf50020 1000fc00 00000001
> > > $16: 00000000 00001000 abf56020 8bf53800
> > > $20: 00000002 bfbc0003 1fffffff bfb90000
> > > $24: 00000002 0fb6f710 
> > > $28: 00008000 08009d28 8bf57e70 080f3b5c
> >                                   ^^^^^^^^
> > > epc   :88020fc0
> >          ^^^^^^^^
> > tell me where these two addresses are pointing to?  Just send me twenty
> > lines or so around the addresses.  (In general that's the right thing to
> > do when the machine bombs with a register dump.)
> > (Also I'm pretty shure that you misstyped $28 and $29 or something really
> > bad happend.)
> 
> Uh, there was nothing at 080f3b5c or nearby, so I'm pretty sure that that
> was misttyped as 880f3b5c.  Here are both:

Thanks.  Both addresses are pointing to code which is important for the
SCSI driver's correctness.  I'll do some more philosophy about the register
dump later.

  Ralf

^ permalink raw reply	[flat|nested] 8+ messages in thread

* Re: gcc RPM missing crtbegin.o
  1998-04-26  4:34       ` Alex deVries
  1998-04-26 11:14         ` ralf
@ 1998-04-26 12:09         ` ralf
  1998-04-26 21:52           ` Various kernel issues Alex deVries
  1 sibling, 1 reply; 8+ messages in thread
From: ralf @ 1998-04-26 12:09 UTC (permalink / raw)
  To: Alex deVries; +Cc: SGI Linux

On Sun, Apr 26, 1998 at 12:34:54AM -0400, Alex deVries wrote:

> > tell me where these two addresses are pointing to?  Just send me twenty
> > lines or so around the addresses.  (In general that's the right thing to
> > do when the machine bombs with a register dump.)
> > (Also I'm pretty shure that you misstyped $28 and $29 or something really
> > bad happend.)
> 
> Uh, there was nothing at 080f3b5c or nearby, so I'm pretty sure that that
> was misttyped as 880f3b5c.  Here are both:

880f3b5c makes sense.  The bad news is that none of the two routines
r4k_dma_cache_wback_inv_pc or dma_setup is the culprit.  Something else
must either have assembled bad SCSI scatter gather lists or corrupted
them such that a NULL pointer got passed down to dma_cache_wback_inv.
So no fix obvious.

Grrrr,

  Ralf

^ permalink raw reply	[flat|nested] 8+ messages in thread

* Various kernel issues.
  1998-04-26 12:09         ` ralf
@ 1998-04-26 21:52           ` Alex deVries
  0 siblings, 0 replies; 8+ messages in thread
From: Alex deVries @ 1998-04-26 21:52 UTC (permalink / raw)
  To: ralf; +Cc: SGI Linux



On Sun, 26 Apr 1998 ralf@uni-koblenz.de wrote:
> On Sun, Apr 26, 1998 at 12:34:54AM -0400, Alex deVries wrote:
> > Uh, there was nothing at 080f3b5c or nearby, so I'm pretty sure that that
> > was misttyped as 880f3b5c.  Here are both:
> 880f3b5c makes sense.  The bad news is that none of the two routines
> r4k_dma_cache_wback_inv_pc or dma_setup is the culprit.  Something else
> must either have assembled bad SCSI scatter gather lists or corrupted
> them such that a NULL pointer got passed down to dma_cache_wback_inv.
> So no fix obvious.

I'll play around with it a bit and make sure that I get similiar results
when I crash it... Let me know if I can help debug this in any way.

In the meantime, I'm trying to churn out updated RPMs using 2.1.72.  This
is difficult because after being up for about 5 minutes, gcc complains
that it can't create executables because:
- /usr/lib/crt1.o is corrupt or disappeared or
- crtbegin.o is gone like beer at a technical university
- program cc1 got fatal signal 11

Other weird things:
- tcsh doesn't work with .1.91, but it does with .72.  I know, I wouldn't
believe it either.  Sure, go ahead, make fun of my shell.
- my boot sequence with 72 now consists of:
   - unmount sdb1 and sdc cleanly
   - three finger salute
   - boot kernel
   - e2fsck, always errors, wait 20 mins
   - reinstall binutils, gcc and all of glibc RPM, always end up with
     weird problems like:
ls -l /usr/share/zoneinfo/America/Indiana/Knox
crwSrwxr-t   3 19496   10694   77, 161 Dec 22  1956 Knox
     Sure, Knox, Indiana might actually be a character device, but I'm
     doubtful about the date.

I'm > < this close to just NFS mounting everything over from my PC, these
disk problems are driving me nuts. This is all so weird, my life was
peachy before I got started with glibc 2.0.6 and stopped running .72.
Still, it's more stable than NT.

> Grrrr,

Indeed.

And in unrelated news, I'm moving to Toronto in a month.

- alex

^ permalink raw reply	[flat|nested] 8+ messages in thread

end of thread, other threads:[~1998-04-26 21:53 UTC | newest]

Thread overview: 8+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
1998-04-26  0:01 gcc RPM missing crtbegin.o Alex deVries
1998-04-26  1:48 ` ralf
1998-04-26  3:00   ` Alex deVries
1998-04-26  3:39     ` ralf
1998-04-26  4:34       ` Alex deVries
1998-04-26 11:14         ` ralf
1998-04-26 12:09         ` ralf
1998-04-26 21:52           ` Various kernel issues Alex deVries

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox