[Linux-ia64] CONFIG_USB_UHCI=y not possible?

[Linux-ia64] CONFIG_USB_UHCI=y not possible?

[John Hesterberg]

When I try to change CONFIG_USB_UHCI from a module to static (=m to =y),
I get:

    drivers/usb/usbdrv.o: In function `alloc_uhci':
    drivers/usb/usbdrv.o(.text.init+0x7e2): relocation truncated to fit: PCREL21B uhci_pci_remove

I don't see anything obvious (to me) with the disassembly:

     7e0:   11 60 01 40 00 21       [MIB]       mov r44=r32
     7e6:   00 00 00 02 00 00                   nop.i 0x0
     7ec:   08 00 00 50                         br.call.sptk.many b0~0 <alloc_uhci

It's see the same problem in a different kernel, with the same code
at a different offset (0x672).

Anyone have any idea what's going on with this?

Also, I have some recollection seeing something that said that USB
support has to be a module, maybe because it doesn't see the devices
otherwise?  Any truth to this?

Thanks,

John

Re: [Linux-ia64] CONFIG_USB_UHCI=y not possible?

[Stephane Eranian]

John,

On Wed, Feb 20, 2002 at 09:50:38AM -0600, John Hesterberg wrote:
> When I try to change CONFIG_USB_UHCI from a module to static (=m to =y),
> I get:
> 
>     drivers/usb/usbdrv.o: In function `alloc_uhci':
>     drivers/usb/usbdrv.o(.text.init+0x7e2): relocation truncated to fit: PCREL21B uhci_pci_remove
> 
> I don't see anything obvious (to me) with the disassembly:
> 
>      7e0:   11 60 01 40 00 21       [MIB]       mov r44=r32
>      7e6:   00 00 00 02 00 00                   nop.i 0x0
>      7ec:   08 00 00 50                         br.call.sptk.many b0~0 <alloc_uhci
> 
> It's see the same problem in a different kernel, with the same code
> at a different offset (0x672).
> 
> Anyone have any idea what's going on with this?
> 
Your kernel is too big (exceed range for relocation). You have to compile some subsystems as modules.

-- 

-Stephane

Re: [Linux-ia64] CONFIG_USB_UHCI=y not possible?

[John Hesterberg]

On Wed, Feb 20, 2002 at 08:09:55AM -0800, Stephane Eranian wrote:
> John,
> 
> On Wed, Feb 20, 2002 at 09:50:38AM -0600, John Hesterberg wrote:
> > When I try to change CONFIG_USB_UHCI from a module to static (=m to =y),
> > I get:
> > 
> >     drivers/usb/usbdrv.o: In function `alloc_uhci':
> >     drivers/usb/usbdrv.o(.text.init+0x7e2): relocation truncated to fit: PCREL21B uhci_pci_remove
> > 
> > I don't see anything obvious (to me) with the disassembly:
> > 
> >      7e0:   11 60 01 40 00 21       [MIB]       mov r44=r32
> >      7e6:   00 00 00 02 00 00                   nop.i 0x0
> >      7ec:   08 00 00 50                         br.call.sptk.many b0~0 <alloc_uhci
> > 
> > It's see the same problem in a different kernel, with the same code
> > at a different offset (0x672).
> > 
> > Anyone have any idea what's going on with this?
> > 
> Your kernel is too big (exceed range for relocation). You have to compile some subsystems as modules.
> 
> -- 
> 
> -Stephane

I don't think that's it.  I've hit this problem with very different
configurations, with other stuff turned on and off.  It fails at
this exact spot.

As Bill Nottingham noted, I think it's because the USB UHCI code is
calling a init function from an exit function, or maybe vice versa.
I'm guessing that some of the err handling code in alloc_uhci is
somehow ending up calling uhci_pci_remove, which is defined as
__devexit, which gets discarded by arch/ia64/vmlinux.lds.S.  So I
think I've almost connected the dots. :-)

John

Re: [Linux-ia64] CONFIG_USB_UHCI=y not possible?

[Andreas Schwab]

John Hesterberg <jh@sgi.com> writes:

|> As Bill Nottingham noted, I think it's because the USB UHCI code is
|> calling a init function from an exit function, or maybe vice versa.
|> I'm guessing that some of the err handling code in alloc_uhci is
|> somehow ending up calling uhci_pci_remove, which is defined as
|> __devexit, which gets discarded by arch/ia64/vmlinux.lds.S.  So I
|> think I've almost connected the dots. :-)

Yes, that has already been fixed in 2.4.18-rc1.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE GmbH, Deutschherrnstr. 15-19, D-90429 Nürnberg
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: [Linux-ia64] CONFIG_USB_UHCI=y not possible?

[John Hesterberg]

On Wed, Feb 20, 2002 at 09:50:38AM -0600, John Hesterberg wrote:
> When I try to change CONFIG_USB_UHCI from a module to static (=m to =y),
> I get:
> 
>     drivers/usb/usbdrv.o: In function `alloc_uhci':
>     drivers/usb/usbdrv.o(.text.init+0x7e2): relocation truncated to fit: PCREL21B uhci_pci_remove

OK, I think this is because alloc_uhci() calls uhci_pci_remove(),
which is defined as __devexit.  This gets dropped when it's not a module.

Below is a quick patch if anyone is interested.
I don't know if this is right or not, but it links.

I'll take this up with the USB maintainer(s).
Seems like if they allow CONFIG_USB_UHCI=y,
then this should be fixed.

John

diff -Naur 2.4.17-sn2/drivers/usb/usb-uhci.c linux/drivers/usb/usb-uhci.c
--- 2.4.17-sn2/drivers/usb/usb-uhci.c	Wed Feb 20 17:11:49 2002
+++ linux/drivers/usb/usb-uhci.c	Wed Feb 20 17:03:53 2002
@@ -3001,7 +3001,9 @@
 	s->irq = irq;
 
 	if(uhci_start_usb (s) < 0) {
+#if defined(MODULE) || defined(CONFIG_HOTPLUG)
 		uhci_pci_remove(dev);
+#endif
 		return -1;
 	}
 

Re: [Linux-ia64] CONFIG_USB_UHCI=y not possible?

[Greg KH]

On Wed, Feb 20, 2002 at 11:14:24AM -0600, John Hesterberg wrote:
> OK, I think this is because alloc_uhci() calls uhci_pci_remove(),
> which is defined as __devexit.  This gets dropped when it's not a module.
> 
> Below is a quick patch if anyone is interested.
> I don't know if this is right or not, but it links.

It's not the correct thing to do.

> I'll take this up with the USB maintainer(s).

I already know about it :)

> Seems like if they allow CONFIG_USB_UHCI=y,
> then this should be fixed.

It's been fixed for quite some time now in the 2.4.18-pre series.  Grab
2.4.18-rc2 and let me know if you still have this problem.

thanks,

greg k-h

Re: [Linux-ia64] CONFIG_USB_UHCI=y not possible?

[John Hesterberg]

This is fixed in 2.4.18-pre2 (or earlier) by removing the __devexit on
uhci_pci_remove.  I wish I'd noticed that sooner. :-)

I'm done replying to my own mails now.

John

On Wed, Feb 20, 2002 at 11:14:24AM -0600, John Hesterberg wrote:
> On Wed, Feb 20, 2002 at 09:50:38AM -0600, John Hesterberg wrote:
> > When I try to change CONFIG_USB_UHCI from a module to static (=m to =y),
> > I get:
> > 
> >     drivers/usb/usbdrv.o: In function `alloc_uhci':
> >     drivers/usb/usbdrv.o(.text.init+0x7e2): relocation truncated to fit: PCREL21B uhci_pci_remove
> 
> OK, I think this is because alloc_uhci() calls uhci_pci_remove(),
> which is defined as __devexit.  This gets dropped when it's not a module.
> 
> Below is a quick patch if anyone is interested.
> I don't know if this is right or not, but it links.
> 
> I'll take this up with the USB maintainer(s).
> Seems like if they allow CONFIG_USB_UHCI=y,
> then this should be fixed.
> 
> John
> 
> diff -Naur 2.4.17-sn2/drivers/usb/usb-uhci.c linux/drivers/usb/usb-uhci.c
> --- 2.4.17-sn2/drivers/usb/usb-uhci.c	Wed Feb 20 17:11:49 2002
> +++ linux/drivers/usb/usb-uhci.c	Wed Feb 20 17:03:53 2002
> @@ -3001,7 +3001,9 @@
>  	s->irq = irq;
>  
>  	if(uhci_start_usb (s) < 0) {
> +#if defined(MODULE) || defined(CONFIG_HOTPLUG)
>  		uhci_pci_remove(dev);
> +#endif
>  		return -1;
>  	}
>  
> 
> _______________________________________________
> Linux-IA64 mailing list
> Linux-IA64@linuxia64.org
> http://lists.linuxia64.org/lists/listinfo/linux-ia64