Re: CONFIG_SCSI_SYM53C8XX_IOMAPPED

Re: CONFIG_SCSI_SYM53C8XX_IOMAPPED

[Matthew Wilcox]

On Thu, Sep 16, 2004 at 11:51:31AM +0200, Jens Schmalzing wrote:
> I recently encountered a machine (a Thinkpad 860) with a NCR 53C810
> SCSI controller that would not work properly unless the config option
> CONFIG_SCSI_SYM53C8XX_IOMAPPED was set.  Under these circumstances,
> installation was a royal pain in the ass, and therefore, I am inclined
> to trade performance for wider support and enable this in the PowerPC
> kernel image packages.
> 
> Any comments?

I was hoping nobody needed to set this option.  It has worse performance,
plus it needs to be enabled on some platforms, like SGI's Visual
Workstation which has completely shafted io port accesses.  As upstream
maintainer, I'd like more information about it please ...

 - How does it "not work properly"?  Does it hang on first access to
   the chip, or does it suffer data corruption?
 - Are the io memory resources on this Thinkpad properly programmed?
 - What kernel were you trying, 2.4 or 2.6?

-- 
"Next the statesmen will invent cheap lies, putting the blame upon 
the nation that is attacked, and every man will be glad of those
conscience-soothing falsities, and will diligently study them, and refuse
to examine any refutations of them; and thus he will by and by convince 
himself that the war is just, and will thank God for the better sleep 
he enjoys after this process of grotesque self-deception." -- Mark Twain

Re: CONFIG_SCSI_SYM53C8XX_IOMAPPED

[Jens Schmalzing]

Hi,

Matthew Wilcox writes:

> As upstream maintainer, I'd like more information about it please ...

>  - How does it "not work properly"?  Does it hang on first access to
>    the chip, or does it suffer data corruption?

It outputs a "can't map PCI MMIO region" and does nothing at all.

>  - Are the io memory resources on this Thinkpad properly programmed?

Most likely they aren't.  Any idea how I could find out?

>  - What kernel were you trying, 2.4 or 2.6?

I was using the official Debian kernel, version 2.6.8, from the
installation CD, built with:

CONFIG_SCSI_SYM53C8XX_2=m
CONFIG_SCSI_SYM53C8XX_DMA_ADDRESSING_MODE=1
CONFIG_SCSI_SYM53C8XX_DEFAULT_TAGS=16
CONFIG_SCSI_SYM53C8XX_MAX_TAGS=64
# CONFIG_SCSI_SYM53C8XX_IOMAPPED is not set

Regards, Jens.

-- 
J'qbpbe, le m'en fquz pe j'qbpbe!
Le veux aimeb et mqubib panz je pézqbpbe je djuz tqtaj!
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: CONFIG_SCSI_SYM53C8XX_IOMAPPED

[Matthew Wilcox]

On Thu, Sep 16, 2004 at 04:22:54PM +0200, Jens Schmalzing wrote:
> > As upstream maintainer, I'd like more information about it please ...
> 
> >  - How does it "not work properly"?  Does it hang on first access to
> >    the chip, or does it suffer data corruption?
> 
> It outputs a "can't map PCI MMIO region" and does nothing at all.

Interesting.  This can only happen if ioremap() returns NULL.  That can
happen for a number of reasons ...

 - Requesting a size of 0.  This isn't happening because the driver
   requests either 128 or 256.
 - Wrap-around.  If the address is right at the top of io space, this
   could happen, but is very unlikely.
 - If the page involved is normal RAM.  Seems unlikely, but if the system
   is misconfigured, it's possible.
 - If we can't get_vm_area().  Possible, but unlikely.
 - If remap_area_pages() fails.  Again, possible but unlikely.

> >  - Are the io memory resources on this Thinkpad properly programmed?
> 
> Most likely they aren't.  Any idea how I could find out?

lspci -v of that device would be interesting to see.

> >  - What kernel were you trying, 2.4 or 2.6?
> 
> I was using the official Debian kernel, version 2.6.8, from the
> installation CD, built with:
> 
> CONFIG_SCSI_SYM53C8XX_2=m
> CONFIG_SCSI_SYM53C8XX_DMA_ADDRESSING_MODE=1
> CONFIG_SCSI_SYM53C8XX_DEFAULT_TAGS=16
> CONFIG_SCSI_SYM53C8XX_MAX_TAGS=64
> # CONFIG_SCSI_SYM53C8XX_IOMAPPED is not set

Those are good config options.

I think we need to do a little debugging... could you add the following
patch and let me know the outcome?

--- drivers/scsi/sym53c8xx_2/sym_glue.c 16 Sep 2004 01:54:44 -0000      1.60
+++ drivers/scsi/sym53c8xx_2/sym_glue.c 16 Sep 2004 15:30:38 -0000
@@ -1674,6 +1674,8 @@ static struct Scsi_Host * __devinit sym_
 
 #ifndef SYM_CONF_IOMAPPED
        np->s.mmio_va = ioremap(dev->s.base_c, np->s.io_ws);
+       printk(KERN_DEBUG "map 0x%lx length %d returned 0x%p\n", dev->s.base_c,
+                       np->s.io_ws, np->s.mmio_va);
        if (!np->s.mmio_va) {
                printf_err("%s: can't map PCI MMIO region\n", sym_name(np));
                goto attach_failed;


-- 
"Next the statesmen will invent cheap lies, putting the blame upon 
the nation that is attacked, and every man will be glad of those
conscience-soothing falsities, and will diligently study them, and refuse
to examine any refutations of them; and thus he will by and by convince 
himself that the war is just, and will thank God for the better sleep 
he enjoys after this process of grotesque self-deception." -- Mark Twain

Re: CONFIG_SCSI_SYM53C8XX_IOMAPPED

[Jens Schmalzing]

Hi,

Matthew Wilcox writes:

> > It outputs a "can't map PCI MMIO region" and does nothing at all.

> Interesting.  This can only happen if ioremap() returns NULL.

Sorry.  I got the message completely wrong.  I really was:

| PCI: Enabling device 0000:00:0c.0 (0000 -> 0003)
| sym.0.12.0: MMIO base address disabled.

> lspci -v of that device would be interesting to see.

| 0000:00:0c.0 SCSI storage controller: LSI Logic / Symbios Logic 53c810 (rev 02)
|         Flags: bus master, medium devsel, latency 128, IRQ 13
|         I/O ports at 1000 [size=256]
|         [virtual] Memory at c0000000 (32-bit, non-prefetchable) [size=256]

> > CONFIG_SCSI_SYM53C8XX_2=m
> > CONFIG_SCSI_SYM53C8XX_DMA_ADDRESSING_MODE=1
> > CONFIG_SCSI_SYM53C8XX_DEFAULT_TAGS=16
> > CONFIG_SCSI_SYM53C8XX_MAX_TAGS=64
> > # CONFIG_SCSI_SYM53C8XX_IOMAPPED is not set

> Those are good config options.

So I used them for a test kernel, built from kernel-source-2.6.8 and
your little debugging patch applied.  What happened is this:

| map 0xc4000000 length 128 returned 0xf4000000

and the kernel initialized the SCSI controller correctly - as I could
verify, this is due to the backported patch from 2.6.9-rc2 that Hollis
Blanchard sent here earlier in Bug#271852.  So that problem will be
solved in the next revision.

Thanks for your time and help.

Regards, Jens.

-- 
J'qbpbe, le m'en fquz pe j'qbpbe!
Le veux aimeb et mqubib panz je pézqbpbe je djuz tqtaj!
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: CONFIG_SCSI_SYM53C8XX_IOMAPPED

[Matthew Wilcox]

On Thu, Sep 16, 2004 at 08:34:38PM +0200, Jens Schmalzing wrote:
> Sorry.  I got the message completely wrong.  I really was:
> 
> | PCI: Enabling device 0000:00:0c.0 (0000 -> 0003)
> | sym.0.12.0: MMIO base address disabled.

OK, so that's entirely different.  That happens if we read 0 directly
from the PCI BAR.  Mmmm.  I don't know how that would happen.  Actually,
I don't see why we consider 0 to be an invalid PCI address ... we should
probably be checking the length instead.

> > lspci -v of that device would be interesting to see.
> 
> | 0000:00:0c.0 SCSI storage controller: LSI Logic / Symbios Logic 53c810 (rev 02)
> |         Flags: bus master, medium devsel, latency 128, IRQ 13
> |         I/O ports at 1000 [size=256]
> |         [virtual] Memory at c0000000 (32-bit, non-prefetchable) [size=256]

I just picked up one of those from a friend:

0000:00:0b.0 SCSI storage controller: LSI Logic / Symbios Logic 53c810 (rev 02)
        Flags: bus master, medium devsel, latency 64, IRQ 10
        I/O ports at b400 [size=256]
        Memory at e3000000 (32-bit, non-prefetchable) [size=256]

> So I used them for a test kernel, built from kernel-source-2.6.8 and
> your little debugging patch applied.  What happened is this:
> 
> | map 0xc4000000 length 128 returned 0xf4000000
> 
> and the kernel initialized the SCSI controller correctly - as I could
> verify, this is due to the backported patch from 2.6.9-rc2 that Hollis
> Blanchard sent here earlier in Bug#271852.  So that problem will be
> solved in the next revision.

Urm.  I don't understand.  You said that you encountered the problem
in a Thinkpad 860, but bug 271852 refers to a powerpc-specific problem.

-- 
"Next the statesmen will invent cheap lies, putting the blame upon 
the nation that is attacked, and every man will be glad of those
conscience-soothing falsities, and will diligently study them, and refuse
to examine any refutations of them; and thus he will by and by convince 
himself that the war is just, and will thank God for the better sleep 
he enjoys after this process of grotesque self-deception." -- Mark Twain

Re: CONFIG_SCSI_SYM53C8XX_IOMAPPED

[Jens Schmalzing]

Hi,

Matthew Wilcox writes:

> Urm.  I don't understand.  You said that you encountered the problem
> in a Thinkpad 860, but bug 271852 refers to a powerpc-specific problem.

That's exactly what the Thinkpad 8xx series are - RS/6000 systems with
PowerPC 603e processors.  See:

korgan:~# cat /proc/cpuinfo 
processor       : 0
cpu             : 603ev
clock           : 166MHz
revision        : 2.0 (pvr 0007 0200)
bogomips        : 110.33
machine         : PReP IBM Thinkpad 850/860
upgrade cpu     : not present
scsi fuse       : ok
simms           : 0:32MiB 
l2 cache        : 256KiB look-aside direct-mapped write-back
60x bus         : 66MHz
pci bus         : 33MHz

Regards, Jens.

-- 
J'qbpbe, le m'en fquz pe j'qbpbe!
Le veux aimeb et mqubib panz je pézqbpbe je djuz tqtaj!
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html