PREP sym53c8xx sym53c8xx brokeness due to 2.6.9-rc1-bk1 introduced residual data patch ...

Hello,

I have been investigating the prep breakage of the sym53c8xx driver on my powerstack II in recent kernels, and traced it to the residual data patch, or more exactly to the whole prep_pci.c/prep_setup.c patches that where introduced between 2.6.9-rc1 and 2.6.9-rc1-bk1. I didn't have time to find more details as the patch is not so small, and seems to do many things, and linux.bkbits.net seems done right now.

Mmm, after a bit more of investigation, the changeset breaking it is :

# This is a BitKeeper generated diff -Nru style patch. #
# ChangeSet
# 2004/08/16 10:35:18-07:00 [email protected]
# ppc32: On PReP, use residual data for PCI dev -> IRQ, and use it.
#
# Signed-off-by: Leigh Brown <[email protected]>
# Signed-off-by: Tom Rini <[email protected]>
#

# include/asm-ppc/residual.h
#   2004/08/16 10:35:09-07:00 [email protected] +7 -0
#   This adds a function to use the residual data to determine the IRQ
#   for a given PCI device, and changes prep_pcibios_fixup() to use it.
#
# arch/ppc/platforms/residual.c
#   2004/08/16 10:35:09-07:00 [email protected] +60 -0
#   This adds a function to use the residual data to determine the IRQ
#   for a given PCI device, and changes prep_pcibios_fixup() to use it.
#
# arch/ppc/platforms/prep_pci.c
#   2004/08/16 10:35:09-07:00 [email protected] +37 -23
#   This adds a function to use the residual data to determine the IRQ
#   for a given PCI device, and changes prep_pcibios_fixup() to use it.

Mmm, i guess this makes sense, since it seems the sym53c8xx doesn'y find its irq anymore or something. Tom Rini, or Leigh Brown, do you have any comments on this one ?

Quoted 35 lines · Sven Luther

Hello,

I have been investigating the prep breakage of the sym53c8xx driver on my powerstack II in recent kernels, and traced it to the residual data patch, or more exactly to the whole prep_pci.c/prep_setup.c patches that where introduced between 2.6.9-rc1 and 2.6.9-rc1-bk1. I didn't have time to find more details as the patch is not so small, and seems to do many things, and linux.bkbits.net seems done right now.

Mmm, after a bit more of investigation, the changeset breaking it is :

# This is a BitKeeper generated diff -Nru style patch. #
# ChangeSet
# 2004/08/16 10:35:18-07:00 [email protected]
# ppc32: On PReP, use residual data for PCI dev -> IRQ, and use it.
#
# Signed-off-by: Leigh Brown <[email protected]>
# Signed-off-by: Tom Rini <[email protected]>
#

# include/asm-ppc/residual.h
#   2004/08/16 10:35:09-07:00 [email protected] +7 -0
#   This adds a function to use the residual data to determine the IRQ
#   for a given PCI device, and changes prep_pcibios_fixup() to use it.
#
# arch/ppc/platforms/residual.c
#   2004/08/16 10:35:09-07:00 [email protected] +60 -0
#   This adds a function to use the residual data to determine the IRQ
#   for a given PCI device, and changes prep_pcibios_fixup() to use it.
#
# arch/ppc/platforms/prep_pci.c
#   2004/08/16 10:35:09-07:00 [email protected] +37 -23
#   This adds a function to use the residual data to determine the IRQ
#   for a given PCI device, and changes prep_pcibios_fixup() to use it.

Quoted 3 lines

Mmm, i guess this makes sense, since it seems the sym53c8xx doesn'y find its irq anymore or something. Tom Rini, or Leigh Brown, do you have any comments on this one ?

Well, we'd need to be certain it's this patch. If it's this patch, I think Leigh was suggesting at the time that the fix would be to fixup the residual data with the correct information.

But what also went in were changes to specific irq maps, based on what fixed a given machine for someone else, so one of those changes could also be it.

Quoted 5 lines

Quoted 5 lines

# This is a BitKeeper generated diff -Nru style patch. #
# ChangeSet
# 2004/08/16 10:35:18-07:00 [email protected]
# ppc32: On PReP, use residual data for PCI dev -> IRQ, and use it.

I got through all the changesets one by one, and it is most definitively this one. i unapply it and it works, i apply it and it breaks.

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

Quoted 8 lines · Sven Luther

Quoted 5 lines

Quoted 5 lines

# This is a BitKeeper generated diff -Nru style patch. #
# ChangeSet
# 2004/08/16 10:35:18-07:00 [email protected]
# ppc32: On PReP, use residual data for PCI dev -> IRQ, and use it.

I got through all the changesets one by one, and it is most definitively this one. i unapply it and it works, i apply it and it breaks.

i can confirm this one.

the changesets are listed here:
http://linux.bkbits.net:8080/linux-2.6/[email protected]

...and because i had some troubles getting a GNU diff-style patch with /usr/bin/bk, i made one against 2.6.11.6:

http://nerdbynature.de/bits/hal/2.6.11.6/1.1803.125.6.diff

more info and dmesg under:

http://nerdbynature.de/bits/hal/2.6.11.6/

to summarize: i don't know exactly which changes, but *some* changes had to be made to arch/ppc/platforms/prep_pci.c to get the network card working [1] again (network code kept locking up the machine). after this issue was resolved, i too noticed the scsi errors [2] others were complaing about.

Quoted 25 lines · Christian

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

Quoted 8 lines · Sven Luther

Quoted 5 lines

Quoted 5 lines

# This is a BitKeeper generated diff -Nru style patch. #
# ChangeSet
# 2004/08/16 10:35:18-07:00 [email protected]
# ppc32: On PReP, use residual data for PCI dev -> IRQ, and use it.

I got through all the changesets one by one, and it is most definitively this one. i unapply it and it works, i apply it and it breaks.

i can confirm this one.

the changesets are listed here:
http://linux.bkbits.net:8080/linux-2.6/[email protected]

...and because i had some troubles getting a GNU diff-style patch with /usr/bin/bk, i made one against 2.6.11.6:

http://nerdbynature.de/bits/hal/2.6.11.6/1.1803.125.6.diff

more info and dmesg under:

Quoted 7 lines

http://nerdbynature.de/bits/hal/2.6.11.6/

to summarize: i don't know exactly which changes, but *some* changes had to be made to arch/ppc/platforms/prep_pci.c to get the network card working [1] again (network code kept locking up the machine). after this issue was resolved, i too noticed the scsi errors [2] others were complaing about.

The issue seems obvious, the patch adds some different way of setting the irqs, based on the residual data, which fails to do what it should on the powerstack, and thus the irqs are fully left uninitialized.

Quoted 34 lines · Sven Luther

Quoted 32 lines · Christian

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

Quoted 8 lines · Sven Luther

Quoted 5 lines

Quoted 5 lines

# This is a BitKeeper generated diff -Nru style patch. #
# ChangeSet
# 2004/08/16 10:35:18-07:00 [email protected]
# ppc32: On PReP, use residual data for PCI dev -> IRQ, and use it.

I got through all the changesets one by one, and it is most definitively this one. i unapply it and it works, i apply it and it breaks.

i can confirm this one.

the changesets are listed here:
http://linux.bkbits.net:8080/linux-2.6/[email protected]

...and because i had some troubles getting a GNU diff-style patch with /usr/bin/bk, i made one against 2.6.11.6:

http://nerdbynature.de/bits/hal/2.6.11.6/1.1803.125.6.diff

more info and dmesg under:

http://nerdbynature.de/bits/hal/2.6.11.6/

to summarize: i don't know exactly which changes, but *some* changes had to be made to arch/ppc/platforms/prep_pci.c to get the network card working [1] again (network code kept locking up the machine). after this issue was resolved, i too noticed the scsi errors [2] others were complaing about.

Quoted 3 lines

The issue seems obvious, the patch adds some different way of setting the irqs, based on the residual data, which fails to do what it should on the powerstack, and thus the irqs are fully left uninitialized.

Can either of you verify that say 2.6.11.6 w/ "noresidual" on the command-line works? Thanks.

Quoted 39 lines · Tom Rini

Quoted 37 lines · Sven Luther

Quoted 25 lines · Christian

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

Quoted 8 lines · Sven Luther

Quoted 5 lines

Quoted 5 lines

# This is a BitKeeper generated diff -Nru style patch. #
# ChangeSet
# 2004/08/16 10:35:18-07:00 [email protected]
# ppc32: On PReP, use residual data for PCI dev -> IRQ, and use it.

I got through all the changesets one by one, and it is most definitively this one. i unapply it and it works, i apply it and it breaks.

i can confirm this one.

the changesets are listed here:
http://linux.bkbits.net:8080/linux-2.6/[email protected]

...and because i had some troubles getting a GNU diff-style patch with /usr/bin/bk, i made one against 2.6.11.6:

http://nerdbynature.de/bits/hal/2.6.11.6/1.1803.125.6.diff

more info and dmesg under:

Quoted 7 lines

http://nerdbynature.de/bits/hal/2.6.11.6/

to summarize: i don't know exactly which changes, but *some* changes had to be made to arch/ppc/platforms/prep_pci.c to get the network card working [1] again (network code kept locking up the machine). after this issue was resolved, i too noticed the scsi errors [2] others were complaing about.

The issue seems obvious, the patch adds some different way of setting the irqs, based on the residual data, which fails to do what it should on the powerstack, and thus the irqs are fully left uninitialized.

Quoted 2 lines

Can either of you verify that say 2.6.11.6 w/ "noresidual" on the command-line works? Thanks.

Notice that last i checked, there was a typo which forced you to say "nopresidual", and i think i did add this when i was testing, and it didn't help.

Can either of you verify that say 2.6.11.6 w/ "noresidual" on the command-line works? Thanks.

i booted vanilla 2.6.11.6 with noresidual (and "nopresidual" too, as Sven sugggested), but the scsi errors did not went away :(

Quoted 2 lines · Tom Rini

Can either of you verify that say 2.6.11.6 w/ "noresidual" on the command-line works? Thanks.

i booted vanilla 2.6.11.6 with noresidual (and "nopresidual" too, as Sven sugggested), but the scsi errors did not went away :(

on a side note, and perhaps totally unrelated: i always have PROC_PREPRESIDUAL=y in my .config, but i never had /proc/residual as promised from the help text.

Quoted 8 lines · Christian

Quoted 2 lines · Tom Rini

Can either of you verify that say 2.6.11.6 w/ "noresidual" on the command-line works? Thanks.

i booted vanilla 2.6.11.6 with noresidual (and "nopresidual" too, as Sven
sugggested), but the scsi errors did not went away :(

Quoted 3 lines

on a side note, and perhaps totally unrelated: i always have PROC_PREPRESIDUAL=y in my .config, but i never had /proc/residual as promised from the help text.

Odd. I thought that only happened if you had no residual data at all (which can happen on Powerstacks, esp if netbooting the kernel). But in that case the new code shouldn't be hit at all. Leigh?

Quoted 9 lines · Tom Rini

Quoted 8 lines · Christian

Quoted 2 lines · Tom Rini

Can either of you verify that say 2.6.11.6 w/ "noresidual" on the command-line works? Thanks.

i booted vanilla 2.6.11.6 with noresidual (and "nopresidual" too, as Sven
sugggested), but the scsi errors did not went away :(

Quoted 7 lines

Quoted 3 lines

on a side note, and perhaps totally unrelated: i always have PROC_PREPRESIDUAL=y in my .config, but i never had /proc/residual as promised from the help text.

Odd. I thought that only happened if you had no residual data at all (which can happen on Powerstacks, esp if netbooting the kernel). But in that case the new code shouldn't be hit at all. Leigh?

Hi, I'm back from my holidays, and have had a look at this. As I spent lots of time understanding the horrendous mess that was prep_pcibios_fixup(), I'd be loath to back out the changes I made (especially as they work so well for me ;-) ).

It turns out that prep_pib_init() is the culprit. Although it would appear to be coded for the non-openpic case, it obviously doesn't work. The old version of prep_pcibios_fixup() only called it if there is an openpic on the machine. We can restore that behaviour with the following patch:

prep_pci.c renamed from prep_pci.c.orig +7 -2

@@ -1245,8 +1245,13 @@
 		pci_write_config_byte(dev, PCI_INTERRUPT_LINE, dev->irq);
 	}
 
-	/* Setup the Winbond or Via PIB */
-	prep_pib_init();
+	/* Setup the Winbond or Via PIB - prep_pib_init() is coded for
+	 * the non-openpic case, but it breaks (at least) the Utah
+	 * (Powerstack II Pro4000), so only call it if we have an
+	 * openpic.
+	 */
+	if (have_openpic)
+		prep_pib_init();
 }
 
 static void __init

I've no idea even what machines would be affected by this, but it fixes Sven's problem and restores the old behaviour, so that can't be bad.

I guess fixing prep_pib_init() would be the better solution but I wouldn't know where to start. If this band-aid is good enough I can submit a proper patch, if you like.

If this works, I'd like to see this get into 2.6.12. Please re-send to akpm / this list. Assuming it's just the above:

Acked-by Tom Rini <[email protected]>

Quoted 4 lines · Tom Rini

If this works, I'd like to see this get into 2.6.12. Please re-send to akpm / this list. Assuming it's just the above:

Acked-by Tom Rini <[email protected]>

yes! i just applied Leigh's patch (thanks!) to a pristine 2.6.11.6 and my PReP here booted fine:
http://www.nerdbynature.de/bits/hal/2.6.11.6/leigh/

Quoted 47 lines · Leigh Brown

Quoted 8 lines · Tom Rini

Quoted 7 lines · Christian

Quoted 2 lines · Tom Rini

Can either of you verify that say 2.6.11.6 w/ "noresidual" on the command-line works? Thanks.

i booted vanilla 2.6.11.6 with noresidual (and "nopresidual" too, as Sven sugggested), but the scsi errors did not went away :(

Quoted 7 lines

Quoted 3 lines

on a side note, and perhaps totally unrelated: i always have PROC_PREPRESIDUAL=y in my .config, but i never had /proc/residual as promised from the help text.

Odd. I thought that only happened if you had no residual data at all (which can happen on Powerstacks, esp if netbooting the kernel). But in that case the new code shouldn't be hit at all. Leigh?

Hi, I'm back from my holidays, and have had a look at this. As I spent lots of time understanding the horrendous mess that was prep_pcibios_fixup(), I'd be loath to back out the changes I made (especially as they work so well for me ;-) ).

It turns out that prep_pib_init() is the culprit. Although it would appear to be coded for the non-openpic case, it obviously doesn't work. The old version of prep_pcibios_fixup() only called it if there is an openpic on the machine. We can restore that behaviour with the following patch:

prep_pci.c renamed from prep_pci.c.orig +7 -2

@@ -1245,8 +1245,13 @@
 		pci_write_config_byte(dev, PCI_INTERRUPT_LINE, dev->irq);
 	}
 
-	/* Setup the Winbond or Via PIB */
-	prep_pib_init();
+	/* Setup the Winbond or Via PIB - prep_pib_init() is coded for
+	 * the non-openpic case, but it breaks (at least) the Utah
+	 * (Powerstack II Pro4000), so only call it if we have an
+	 * openpic.
+	 */
+	if (have_openpic)
+		prep_pib_init();
 }
 
 static void __init

Quoted 7 lines

I've no idea even what machines would be affected by this, but it fixes Sven's problem and restores the old behaviour, so that can't be bad.

I guess fixing prep_pib_init() would be the better solution but I wouldn't know where to start. If this band-aid is good enough I can submit a proper patch, if you like.

If this works, I'd like to see this get into 2.6.12. Please re-send to akpm / this list. Assuming it's just the above:

Acked-by Tom Rini <[email protected]>

Quoted 6 lines · Christian

Quoted 4 lines · Tom Rini

If this works, I'd like to see this get into 2.6.12. Please re-send to akpm / this list. Assuming it's just the above:

Acked-by Tom Rini <[email protected]>

Quoted 3 lines

yes! i just applied Leigh's patch (thanks!) to a pristine 2.6.11.6 and my PReP here booted fine:
http://www.nerdbynature.de/bits/hal/2.6.11.6/leigh/

Great. Leigh, can you also submit to gregkh for 2.6.11.7? Thanks.

Quoted 59 lines · On Fri, 8 Apr 2005 07:54:31 -0700 Tom Rini wrote:

Quoted 47 lines · Leigh Brown

Quoted 8 lines · Tom Rini

Quoted 7 lines · Christian

Quoted 2 lines · Tom Rini

Can either of you verify that say 2.6.11.6 w/ "noresidual" on the command-line works? Thanks.

i booted vanilla 2.6.11.6 with noresidual (and "nopresidual" too, as Sven sugggested), but the scsi errors did not went away :(

Quoted 7 lines

Quoted 3 lines

on a side note, and perhaps totally unrelated: i always have PROC_PREPRESIDUAL=y in my .config, but i never had /proc/residual as promised from the help text.

Odd. I thought that only happened if you had no residual data at all (which can happen on Powerstacks, esp if netbooting the kernel). But in that case the new code shouldn't be hit at all. Leigh?

Hi, I'm back from my holidays, and have had a look at this. As I spent lots of time understanding the horrendous mess that was prep_pcibios_fixup(), I'd be loath to back out the changes I made (especially as they work so well for me ;-) ).

It turns out that prep_pib_init() is the culprit. Although it would appear to be coded for the non-openpic case, it obviously doesn't work. The old version of prep_pcibios_fixup() only called it if there is an openpic on the machine. We can restore that behaviour with the following patch:

prep_pci.c renamed from prep_pci.c.orig +7 -2

@@ -1245,8 +1245,13 @@
 		pci_write_config_byte(dev, PCI_INTERRUPT_LINE, dev->irq);
 	}
 
-	/* Setup the Winbond or Via PIB */
-	prep_pib_init();
+	/* Setup the Winbond or Via PIB - prep_pib_init() is coded for
+	 * the non-openpic case, but it breaks (at least) the Utah
+	 * (Powerstack II Pro4000), so only call it if we have an
+	 * openpic.
+	 */
+	if (have_openpic)
+		prep_pib_init();
 }
 
 static void __init

Quoted 7 lines

I've no idea even what machines would be affected by this, but it fixes Sven's problem and restores the old behaviour, so that can't be bad.

I guess fixing prep_pib_init() would be the better solution but I wouldn't know where to start. If this band-aid is good enough I can submit a proper patch, if you like.

If this works, I'd like to see this get into 2.6.12. Please re-send to akpm / this list. Assuming it's just the above:

Acked-by Tom Rini <[email protected]>

Quoted 11 lines · Tom Rini

Quoted 9 lines · Christian

Quoted 4 lines · Tom Rini

If this works, I'd like to see this get into 2.6.12. Please re-send to akpm / this list. Assuming it's just the above:

Acked-by Tom Rini <[email protected]>

yes! i just applied Leigh's patch (thanks!) to a pristine 2.6.11.6 and my PReP here booted fine:
http://www.nerdbynature.de/bits/hal/2.6.11.6/leigh/

Quoted 1 line

Great. Leigh, can you also submit to gregkh for 2.6.11.7? Thanks.

This patch restores the original behaviour of prep_pcibios_fixup() to only call prep_pib_init() on machines with an openpic. This allows the Powerstack II Pro4000 to boot again.

Signed-off-by Leigh Brown <[email protected]>