Skip to contentSkip to conversation

linux 2.6 still not working with PReP (ppc32)

FYI In review Christian Kujau 3 participants 8 messages Dec 4, 2004 – Dec 6, 2004

lore
Share & tools
lore

Christian Kujau

author
#0 lore raw

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

hi,

upon receiving some calls about the status on this issue, i decided to update http://bugme.osdl.org/show_bug.cgi?id=1494 , which i reported a while back. yes, this is "just a PReP" and i am fully aware that PReP is not used in recent setups any more and i too will not my PReP for HPC applications. but perhaps someone has an idea about what's goin on here.

the bugreport should give you th full details, in short: linux 2.5.30 is the last working kernel for ppc32/PReP with networking enabled. (net)booting a recent 2.6 kernels spit out lots of these:

(with tulip NIC)

Dec  3 23:56:39 192.168.10.9 irq 9: nobody cared!
Dec  3 23:56:39 192.168.10.9 Call trace:
Dec  3 23:56:39 192.168.10.9 [c002e99c] __report_bad_irq+0x34/0xac
Dec  3 23:56:39 192.168.10.9 [c002eb00] note_interrupt+0xd0/0x104
Dec  3 23:56:39 192.168.10.9 [c002e480] __do_IRQ+0x174/0x184
Dec  3 23:56:39 192.168.10.9 [c00053e0] do_IRQ+0x38/0x98
Dec  3 23:56:39 192.168.10.9 [c00044cc] ret_from_except+0x0/0x14
Dec  3 23:56:39 192.168.10.9 [c002e748] setup_irq+0xd8/0x138
Dec  3 23:56:39 192.168.10.9 [c002e938] request_irq+0x90/0xc0
Dec  3 23:56:39 192.168.10.9 [c0103504] tulip_open+0x30/0xb6c
Dec  3 23:56:39 192.168.10.9 [c0117274] dev_open+0xb0/0xd8
Dec  3 23:56:39 192.168.10.9 [c0118a7c] dev_change_flags+0x6c/0x144
Dec  3 23:56:39 192.168.10.9 [c0123db4] netpoll_setup+0x1c4/0x364
Dec  3 23:56:39 192.168.10.9 [c0106fe0] init_netconsole+0x3c/0x94
Dec  3 23:56:39 192.168.10.9 [c0003a64] init+0xb8/0x230
Dec  3 23:56:39 192.168.10.9 [c0006524] kernel_thread+0x44/0x60
Dec  3 23:56:39 192.168.10.9 handlers:
Dec  3 23:56:39 192.168.10.9 [<c010101c>] (tulip_interrupt+0x0/0xd3c)
Dec  3 23:56:39 192.168.10.9 Disabling IRQ #9

(with 3c59x NIC)
192.168.10.9 NETDEV WATCHDOG: eth0: transmit timed out 192.168.10.9 eth0: transmit timed out, tx_status 00 status e800. 192.168.10.9 diagnostics: net 0cc0 media 8802 dma 002000b3 fifo 0000 192.168.10.9 eth0: Interrupt posted but not delivered -- IRQ blocked by another device?
192.168.10.9 Flags; bus-master 1, dirty 67(3) current 68(4) 192.168.10.9 Transmit list 806fe480 vs. c06fe3e0.

192.168.10.9 0: @c06fe200  length 80000054 status 00000054
192.168.10.9 1: @c06fe2a0  length 8000005e status 0000005e
192.168.10.9 2: @c06fe340  length 80000064 status 00000064
192.168.10.9 3: @c06fe3e0  length 80000075 status 00000075
192.168.10.9 4: @c06fe480  length 8000005b status 0000005b
192.168.10.9 5: @c06fe520  length 80000051 status 00000051
192.168.10.9 6: @c06fe5c0  length 8000005a status 0000005a
192.168.10.9 7: @c06fe660  length 8000005a status 0000005a
192.168.10.9 8: @c06fe700  length 8000005a status 0000005a
192.168.10.9 9: @c06fe7a0  length 8000005a status 0000005a
192.168.10.9 10: @c06fe840  length 8000005a status 0000005a
192.168.10.9 11: @c06fe8e0  length 8000005a status 0000005a
192.168.10.9 12: @c06fe980  length 8000005a status 0000005a
192.168.10.9 13: @c06fea20  length 8000005a status 0000005a
192.168.10.9 14: @c06feac0  length 8000005a status 0000005a
192.168.10.9 13: @c06fea20  length 8000005a status 0000005a
192.168.10.9 14: @c06feac0  length 8000005a status 0000005a
192.168.10.9 15: @c06feb60  length 8000005a status 0000005a

is it possible to use another irq for the network-device driver? Documentation/kernel-parameters.txt gives some options about other drivers, but i can't tell how to specify "use this irq" for tulip/3c59x.

especially the "IRQ blocked by another device?" message makes me think, that manual irq config could help here. but maybe i'm wrong.

the results posted on http://nerdbynature.de/bits/sheep/2.6.10-rc2/ are from current 2.6 kernels: the offical tree and the linuxppc25 tree was used.

thank you for comments,
Christian.
- --
BOFH excuse #441:

Hash table has woodworm
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFBsj3y+A7rjkF8z0wRAldLAJ9QFTfr8UeOtXbnSCGWBVJr7IAAKgCgiHRV iB/ozetyCDIMSfAa+1MRxVc=
=b5BK
-----END PGP SIGNATURE-----

#1 lore raw

You got the powerstack booting from scsi (reading and interpreting the bug report). I had problems booting 2.6 kernels as well (never tested any 2.5 kernels). It turned out that the pci slot numbering has changed sometime and this wasn't reflected in arch/ppc/platforms/prep_pci.c. After having set up the slot0...slot8 using the values of slot10...slot18 (except for slot1 which got value 4 so IDE is usable out of the box), the machine booted a 2.6 kernel (2.6.8).

Sebastian

Quoted 91 lines · Christian Kujau

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

hi,

upon receiving some calls about the status on this issue, i decided to update http://bugme.osdl.org/show_bug.cgi?id=1494 , which i reported a while back. yes, this is "just a PReP" and i am fully aware that PReP is not used in recent setups any more and i too will not my PReP for HPC applications. but perhaps someone has an idea about what's goin on here.

the bugreport should give you th full details, in short: linux 2.5.30 is the last working kernel for ppc32/PReP with networking enabled. (net)booting a recent 2.6 kernels spit out lots of these:

(with tulip NIC)

Dec  3 23:56:39 192.168.10.9 irq 9: nobody cared!
Dec  3 23:56:39 192.168.10.9 Call trace:
Dec  3 23:56:39 192.168.10.9 [c002e99c] __report_bad_irq+0x34/0xac
Dec  3 23:56:39 192.168.10.9 [c002eb00] note_interrupt+0xd0/0x104
Dec  3 23:56:39 192.168.10.9 [c002e480] __do_IRQ+0x174/0x184
Dec  3 23:56:39 192.168.10.9 [c00053e0] do_IRQ+0x38/0x98
Dec  3 23:56:39 192.168.10.9 [c00044cc] ret_from_except+0x0/0x14
Dec  3 23:56:39 192.168.10.9 [c002e748] setup_irq+0xd8/0x138
Dec  3 23:56:39 192.168.10.9 [c002e938] request_irq+0x90/0xc0
Dec  3 23:56:39 192.168.10.9 [c0103504] tulip_open+0x30/0xb6c
Dec  3 23:56:39 192.168.10.9 [c0117274] dev_open+0xb0/0xd8
Dec  3 23:56:39 192.168.10.9 [c0118a7c] dev_change_flags+0x6c/0x144
Dec  3 23:56:39 192.168.10.9 [c0123db4] netpoll_setup+0x1c4/0x364
Dec  3 23:56:39 192.168.10.9 [c0106fe0] init_netconsole+0x3c/0x94
Dec  3 23:56:39 192.168.10.9 [c0003a64] init+0xb8/0x230
Dec  3 23:56:39 192.168.10.9 [c0006524] kernel_thread+0x44/0x60
Dec  3 23:56:39 192.168.10.9 handlers:
Dec  3 23:56:39 192.168.10.9 [<c010101c>] (tulip_interrupt+0x0/0xd3c)
Dec  3 23:56:39 192.168.10.9 Disabling IRQ #9

(with 3c59x NIC)
192.168.10.9 NETDEV WATCHDOG: eth0: transmit timed out 192.168.10.9 eth0: transmit timed out, tx_status 00 status e800. 192.168.10.9 diagnostics: net 0cc0 media 8802 dma 002000b3 fifo 0000 192.168.10.9 eth0: Interrupt posted but not delivered -- IRQ blocked by another device?
192.168.10.9 Flags; bus-master 1, dirty 67(3) current 68(4) 192.168.10.9 Transmit list 806fe480 vs. c06fe3e0.

192.168.10.9 0: @c06fe200  length 80000054 status 00000054
192.168.10.9 1: @c06fe2a0  length 8000005e status 0000005e
192.168.10.9 2: @c06fe340  length 80000064 status 00000064
192.168.10.9 3: @c06fe3e0  length 80000075 status 00000075
192.168.10.9 4: @c06fe480  length 8000005b status 0000005b
192.168.10.9 5: @c06fe520  length 80000051 status 00000051
192.168.10.9 6: @c06fe5c0  length 8000005a status 0000005a
192.168.10.9 7: @c06fe660  length 8000005a status 0000005a
192.168.10.9 8: @c06fe700  length 8000005a status 0000005a
192.168.10.9 9: @c06fe7a0  length 8000005a status 0000005a
192.168.10.9 10: @c06fe840  length 8000005a status 0000005a
192.168.10.9 11: @c06fe8e0  length 8000005a status 0000005a
192.168.10.9 12: @c06fe980  length 8000005a status 0000005a
192.168.10.9 13: @c06fea20  length 8000005a status 0000005a
192.168.10.9 14: @c06feac0  length 8000005a status 0000005a
192.168.10.9 13: @c06fea20  length 8000005a status 0000005a
192.168.10.9 14: @c06feac0  length 8000005a status 0000005a
192.168.10.9 15: @c06feb60  length 8000005a status 0000005a

is it possible to use another irq for the network-device driver? Documentation/kernel-parameters.txt gives some options about other drivers, but i can't tell how to specify "use this irq" for tulip/3c59x.

especially the "IRQ blocked by another device?" message makes me think, that manual irq config could help here. but maybe i'm wrong.

the results posted on http://nerdbynature.de/bits/sheep/2.6.10-rc2/ are from current 2.6 kernels: the offical tree and the linuxppc25 tree was used.

thank you for comments,
Christian.
- --
BOFH excuse #441:

Hash table has woodworm
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFBsj3y+A7rjkF8z0wRAldLAJ9QFTfr8UeOtXbnSCGWBVJr7IAAKgCgiHRV iB/ozetyCDIMSfAa+1MRxVc=
=b5BK
-----END PGP SIGNATURE-----
_______________________________________________
Linuxppc-dev mailing list
[email protected]
https://ozlabs.org/mailman/listinfo/linuxppc-dev

author
#2 lore raw

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Quoted 1 line · Sebastian Heutling

You got the powerstack booting from scsi (reading and interpreting the

yes, but now the disk is gone and i have to use nfsroot.

Quoted 6 lines

bug report). I had problems booting 2.6 kernels as well (never tested any 2.5 kernels). It turned out that the pci slot numbering has changed sometime and this wasn't reflected in arch/ppc/platforms/prep_pci.c. After having set up the slot0...slot8 using the values of slot10...slot18 (except for slot1 which got value 4 so IDE is usable out of the box), the machine booted a 2.6 kernel (2.6.8).

boy! this *is* an early christmas present, for sure! how did you come to this conclusion? something *must* have changed in the pci setup after 2.5.30, because the problem did not go away when i used a different NIC driver. but i could not fiddle out the changeset to blame here, because i failed to compile many kernels > 2.5.30.

changing one line in arch/ppc/platforms/prep_pci.c made my PReP booting a recent 2.6-BK, patch attached.

it's booting now, "init=/bin/bash" is working, but i still can't ping to the outside world, which is still a bit strange, but i hope i can work it out.

tausend dank (really),
Christian.
- --
BOFH excuse #304:

routing problems on the neural net
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFBs0hj+A7rjkF8z0wRAo3RAKDF+QYA6cosLH+kWtiAmiWtRzlQkQCeKcpx FXOiNRfl9/vOynE8d4hDJJg=
=sKOm
-----END PGP SIGNATURE-----

arch/ppc/platforms/prep_pci.c.edited renamed from arch/ppc/platforms/prep_pci.c +1 -1
@@ -61,7 +61,7 @@
         0,   /* Slot 11 - unused */
         5,   /* Slot 12 - SCSI - NCR825A */
         0,   /* Slot 13 - unused */
-        3,   /* Slot 14 - enet */
+        1,   /* Slot 14 - enet */
         0,   /* Slot 15 - unused */
         2,   /* Slot 16 - unused */
         3,   /* Slot 17 - unused */
#3 lore raw
Quoted 15 lines · Christian Kujau

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Quoted 1 line · Sebastian Heutling

You got the powerstack booting from scsi (reading and interpreting the

yes, but now the disk is gone and i have to use nfsroot.

Quoted 6 lines

bug report). I had problems booting 2.6 kernels as well (never tested any 2.5 kernels). It turned out that the pci slot numbering has changed sometime and this wasn't reflected in arch/ppc/platforms/prep_pci.c. After having set up the slot0...slot8 using the values of slot10...slot18 (except for slot1 which got value 4 so IDE is usable out of the box), the machine booted a 2.6 kernel (2.6.8).

Quoted 2 lines

boy! this *is* an early christmas present, for sure! how did you come to this conclusion? something *must* have changed in the pci setup after

:-)

Pure trial and error in the first time. When realising that something is wrong with the assigned interrupts (looking at the kernel output) we tried to set the ones that were working in a 2.4 kernel. When the machine finally booted a "lspci" showed that the slot number changed.

Quoted 9 lines

2.5.30, because the problem did not go away when i used a different NIC driver. but i could not fiddle out the changeset to blame here, because i failed to compile many kernels > 2.5.30.

changing one line in arch/ppc/platforms/prep_pci.c made my PReP booting a recent 2.6-BK, patch attached.

it's booting now, "init=/bin/bash" is working, but i still can't ping to the outside world, which is still a bit strange, but i hope i can work it out.

Is it that onboard network tulip kind of thingy? Do you have some other PCI cards installed?
I used to have an extra symbios logic scsi-host-adapter installed. On high network load and loading/writing to a HD (using onboard scsi), I noticed a lot of problems: either the scsi-host-adapter lost the interrupt or the network driver lost some packets or both. After having removed the extra scsi-host-adapter it works fine now.

Quoted 2 lines

tausend dank (really),

You are welcome!

Sebastian

author
#4 lore raw

Re: [FYI] linux 2.6 still not^Wnow working with PReP (ppc32)

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Quoted 5 lines · Sebastian Heutling

Pure trial and error in the first time. When realising that something is wrong with the assigned interrupts (looking at the kernel output) we tried to set the ones that were working in a 2.4 kernel. When the machine finally booted a "lspci" showed that the slot number changed.

lspci, sounds easy ;-)
but i never would have looked at slot numbering....

Quoted 3 lines

Is it that onboard network tulip kind of thingy? Do you have some other PCI cards installed?

the card is onboard. since i had some issues with this card in early 2.4 kernels too, i used to have another NIC. in fact, very often i only compiled with 3c59x enabled - OF cared about the onboard NIC (tulip) and once the kernel is loaded, the tulip is not needed any more and 3c59x takes over.

http://nerdbynature.de/bits/sheep/2.6.10-rc3/ has some details.

Quoted 5 lines

I used to have an extra symbios logic scsi-host-adapter installed. On high network load and loading/writing to a HD (using onboard scsi), I noticed a lot of problems: either the scsi-host-adapter lost the interrupt or the network driver lost some packets or both. After having removed the extra scsi-host-adapter it works fine now.

hm, when i had 3 scsi-disks in this machine (fine IBM ones, hmmmm..) i only used the on-board scsi-bus. there was no other scsi-host-adapter inside, but i noticed the very same problems: upon massive i/o, the machine got stuck for 30sec, sometimes longer, load increased to 30..40 or so, then the machine got back, with a syslog full of scsi errors.

it's still so incredible that a 3-liner solved this issue. the bug-report was opened over a year ago...

thanks again,
Christian.
- --
BOFH excuse #13:

we're waiting for [the phone company] to fix that line -----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFBs9zX+A7rjkF8z0wRArwBAKDZ2ywiukQl/b/uPTG1/1jWHrIsxQCeIcKb ltnS8407uSUM+J6jiJABxxc=
=Ts4X
-----END PGP SIGNATURE-----

author
#5 lore raw

PATCH 1/1 linux 2.6 not^Wnow working with PReP (ppc32)

In reply to Sebastian Heutling

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

whoops! a bit too early, i had to change 3 lines, not one.

Signed-off-by Christian Kujau <[email protected]>

(new patch attached; credits should go to Sebastian Heutling aka [email protected])

thanks,
Christian.
- --
BOFH excuse #304:

routing problems on the neural net
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFBs2hE+A7rjkF8z0wRAi+3AJ42GZlqMZen6lqMwtyQSMbS61tkbwCeMNET nKgGhTpgwAcCJJ0hWUKoFqg=
=xyi5
-----END PGP SIGNATURE-----

arch/ppc/platforms/prep_pci.c.edited renamed from arch/ppc/platforms/prep_pci.c +3 -3
@@ -49,10 +49,10 @@
         0,   /* Slot 1  - unused */
         5,   /* Slot 2  - SCSI - NCR825A  */
         0,   /* Slot 3  - unused */
-        1,   /* Slot 4  - Ethernet - DEC2114x */
+        3,   /* Slot 4  - Ethernet - DEC2114x */
         0,   /* Slot 5  - unused */
-        3,   /* Slot 6  - PCI Card slot #1 */
-        4,   /* Slot 7  - PCI Card slot #2 */
+        2,   /* Slot 6  - PCI Card slot #1 */
+        3,   /* Slot 7  - PCI Card slot #2 */
         5,   /* Slot 8  - PCI Card slot #3 */
         5,   /* Slot 9  - PCI Bridge */
              /* added here in case we ever support PCI bridges */
#6 lore raw

Re: [PATCH] linux 2.6 not^Wnow working with PReP (ppc32)

Quoted 11 lines · Christian Kujau

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

whoops! a bit too early, i had to change 3 lines, not one.

Signed-off-by Christian Kujau <[email protected]>

(new patch attached; credits should go to Sebastian Heutling aka [email protected])

This doesn't cause anything bad to happen on mine (and it shouldn't, being a Blackhawk and all :)), so I'll send this on. Thanks.

Signature · 2 lines
Tom Rini
http://gate.crashing.org/~trini/
author
#7 lore raw

Re: [PATCH] linux 2.6 not^Wnow working with PReP (ppc32)

Quoted 3 lines · Tom Rini

This doesn't cause anything bad to happen on mine (and it shouldn't, being a Blackhawk and all :)), so I'll send this on. Thanks.

oh, fine! sounds good!

thanks,
Christian.

Signature · 1 line
make bzImage, not war