2.5.5[01]]: Xircom Cardbus broken (PCI resource collisions)

2.5.5[01]]: Xircom Cardbus broken (PCI resource collisions)

[Alessandro Suardi]

As per subject - my RBEM56G-100BTX isn't usable due to the reasons
  printed in /var/log/messages. This has already been reported but
  I don't recall seeing patches or any debug stuff to be tried

[...]

NET4: Linux TCP/IP 1.0 for NET4.0
IP: routing cache hash table of 2048 buckets, 16Kbytes
TCP: Hash tables configured (established 16384 bind 32768)
NET4: Unix domain sockets 1.0/SMP for Linux NET4.0.
cs: cb_alloc(bus 2): vendor 0x115d, device 0x0003
PCI: Device 02:00.0 not available because of resource collisions
PCI: Device 02:00.1 not available because of resource collisions
kjournald starting.  Commit interval 5 seconds
EXT3-fs: mounted filesystem with ordered data mode.

[...]

--alessandro

  "Seems that you can't get any more than half free"
        (Bruce Springsteen, "Straight Time")

Re: 2.5.5[01]]: Xircom Cardbus broken (PCI resource collisions)

[Valdis.Kletnieks]

[-- Attachment #1.1: Type: text/plain, Size: 214 bytes --]

> PCI: Device 02:00.0 not available because of resource collisions
> PCI: Device 02:00.1 not available because of resource collisions

Been there. Done that. Does the attached patch help? It did for me.

/Valdis



[-- Attachment #1.2: pcmcia.patch --]
[-- Type: text/plain , Size: 492 bytes --]

--- drivers/pcmcia/cardbus.c.dist	2002-12-03 01:49:29.000000000 -0500
+++ drivers/pcmcia/cardbus.c	2002-12-03 01:50:23.000000000 -0500
@@ -283,8 +283,6 @@
 		dev->hdr_type = hdr & 0x7f;
 
 		pci_setup_device(dev);
-		if (pci_enable_device(dev))
-			continue;
 
 		strcpy(dev->dev.bus_id, dev->slot_name);
 
@@ -302,6 +300,8 @@
 			pci_writeb(dev, PCI_INTERRUPT_LINE, irq);
 		}
 
+		if (pci_enable_device(dev))
+			continue;
 		device_register(&dev->dev);
 		pci_insert_device(dev, bus);
 	}

[-- Attachment #2: Type: application/pgp-signature, Size: 226 bytes --]

Re: 2.5.5[01]]: Xircom Cardbus broken (PCI resource collisions)

[Greg KH]

On Thu, Dec 12, 2002 at 05:47:17PM -0500, Valdis.Kletnieks@vt.edu wrote:
> > PCI: Device 02:00.0 not available because of resource collisions
> > PCI: Device 02:00.1 not available because of resource collisions
> 
> Been there. Done that. Does the attached patch help? It did for me.

Fixes the problem for me :)

thanks for the patch,

greg k-h

Re: 2.5.5[01]]: Xircom Cardbus broken (PCI resource collisions)

[Valdis.Kletnieks]

[-- Attachment #1: Type: text/plain, Size: 548 bytes --]

On Thu, 12 Dec 2002 16:43:30 PST, Greg KH said:
> On Thu, Dec 12, 2002 at 05:47:17PM -0500, Valdis.Kletnieks@vt.edu wrote:
> > > PCI: Device 02:00.0 not available because of resource collisions
> > > PCI: Device 02:00.1 not available because of resource collisions
> > 
> > Been there. Done that. Does the attached patch help? It did for me.
> 
> Fixes the problem for me :)

Glad to hear it.  I'm pretty sure I merely backed out a broken change, but
two decades have left me for a healthy respect for "Mel the Real Programmer"
stories ;)

/Valdis

[-- Attachment #2: Type: application/pgp-signature, Size: 226 bytes --]

Re: 2.5.5[01]]: Xircom Cardbus broken (PCI resource collisions)

[Arjan van de Ven]

On Thu, 2002-12-12 at 23:47, Valdis.Kletnieks@vt.edu wrote:

> Been there. Done that. Does the attached patch help? It did for me.
> 
> /Valdis
> 
> 
> ----
> 

> --- drivers/pcmcia/cardbus.c.dist	2002-12-03 01:49:29.000000000 -0500
> +++ drivers/pcmcia/cardbus.c	2002-12-03 01:50:23.000000000 -0500
> @@ -283,8 +283,6 @@
>  		dev->hdr_type = hdr & 0x7f;
>  
>  		pci_setup_device(dev);
> -		if (pci_enable_device(dev))
> -			continue;
>  
>  		strcpy(dev->dev.bus_id, dev->slot_name);
>  
> @@ -302,6 +300,8 @@
>  			pci_writeb(dev, PCI_INTERRUPT_LINE, irq);
>  		}
>  
> +		if (pci_enable_device(dev))
> +			continue;
>  		device_register(&dev->dev);
>  		pci_insert_device(dev, bus);
>  	}

interesting. BUT aren't we writing to the device 3 lines before where
you add the pci_enable_device()? That sounds like a bad plan to me ;(

Re: 2.5.5[01]]: Xircom Cardbus broken (PCI resource collisions)

[Valdis.Kletnieks]

[-- Attachment #1: Type: text/plain, Size: 1887 bytes --]

On Fri, 13 Dec 2002 11:10:24 +0100, Arjan van de Ven said:
> On Thu, 2002-12-12 at 23:47, Valdis.Kletnieks@vt.edu wrote:
> > --- drivers/pcmcia/cardbus.c.dist	2002-12-03 01:49:29.000000000 -0500
> > +++ drivers/pcmcia/cardbus.c	2002-12-03 01:50:23.000000000 -0500
> > @@ -283,8 +283,6 @@
> >  		dev->hdr_type = hdr & 0x7f;
> >  
> >  		pci_setup_device(dev);
> > -		if (pci_enable_device(dev))
> > -			continue;
> >  
> >  		strcpy(dev->dev.bus_id, dev->slot_name);
> >  
> > @@ -302,6 +300,8 @@
> >  			pci_writeb(dev, PCI_INTERRUPT_LINE, irq);
> >  		}
> >  
> > +		if (pci_enable_device(dev))
> > +			continue;
> >  		device_register(&dev->dev);
> >  		pci_insert_device(dev, bus);
> >  	}
> 
> interesting. BUT aren't we writing to the device 3 lines before where
> you add the pci_enable_device()? That sounds like a bad plan to me ;(

That's where pci_enable_device() *USED* to be - this is backing out a change
made in 2.5.50 - it added the if/continue wrapper and moved it up about 15
lines in the code. The problem seems to be that for many devices,
pci_enable_device() fails unless we do the pci_writeb() magic first.  Or
perhaps it's the call to pci_assign_resource() - I don't know. ;) The 2.4.18
tree has the pci_enable_device() at the same place - just without the if/
continue around it.  It can't be THAT evil, as that's the way 2.4.0 to 2.4.18
and 2.5.0 to 2.5.49 did it.. ;)

I see why the if/continue was added - you don't want to be calling
device_register()/pci_insert_device() if pci_enable_device() loses.  I don't
see why 2.5.50 moved the code up after pci_setup_device(). There's an outside
chance that the concept of moving the call was correct, but that it should have
been moved to between the calls to pci_assign_resource() and pci_readb().
If that's the case, then you're correct as well....

/Valdis (who shouldn't be trying to think before caffeine.. ;)


[-- Attachment #2: Type: application/pgp-signature, Size: 226 bytes --]

Re: 2.5.5[01]]: Xircom Cardbus broken (PCI resource collisions)

[Petr Konecny]

 Arjan> interesting. BUT aren't we writing to the device 3 lines before
 Arjan> where you add the pci_enable_device()? That sounds like a bad
 Arjan> plan to me ;(

 Valdis> I see why the if/continue was added - you don't want to be
 Valdis> calling device_register()/pci_insert_device() if
 Valdis> pci_enable_device() loses.  I don't see why 2.5.50 moved the
 Valdis> code up after pci_setup_device(). There's an outside chance
 Valdis> that the concept of moving the call was correct, but that it
 Valdis> should have been moved to between the calls to
 Valdis> pci_assign_resource() and pci_readb().  If that's the case,
 Valdis> then you're correct as well....
I can confirm that this indeed works. I moved the two lines before
pci_readb and the card works (every character you now read went through
it). Who shall submit a patch to Linus ?

Now I have to figure out, how to make it load without manual modprobe
xircom_cb. Oh the joys of new module loader.

                                                Petr
-- 
Computers are like air conditioners.  Both stop working, if you open
windows.
        -- Adam Heath

Re: 2.5.5[01]]: Xircom Cardbus broken (PCI resource collisions)

[Valdis.Kletnieks]

[-- Attachment #1: Type: text/plain, Size: 1180 bytes --]

On Fri, 13 Dec 2002 17:33:00 +0100, Petr Konecny <pekon@informatics.muni.cz>  said:

>  Valdis> I see why the if/continue was added - you don't want to be
>  Valdis> calling device_register()/pci_insert_device() if
>  Valdis> pci_enable_device() loses.  I don't see why 2.5.50 moved the
>  Valdis> code up after pci_setup_device(). There's an outside chance
>  Valdis> that the concept of moving the call was correct, but that it
>  Valdis> should have been moved to between the calls to
>  Valdis> pci_assign_resource() and pci_readb().  If that's the case,
>  Valdis> then you're correct as well....
> I can confirm that this indeed works. I moved the two lines before
> pci_readb and the card works (every character you now read went through
> it). Who shall submit a patch to Linus ?

The problem is this from the 2.5.50 Changelog that Linus posted:

Dave Jones <davej@suse.de>:
...
  o make cardbus PCI enable earlier

I'm willing to submit a patch, but I think Dave has to make the call whether
it should be backed out entirely, or moved after pci_assign_resource().
I certainly don't understand the code *or* PCI well enough to decide between
those two option...

/Valdis


[-- Attachment #2: Type: application/pgp-signature, Size: 226 bytes --]

Re: 2.5.5[01]]: Xircom Cardbus broken (PCI resource collisions)

[Dave Jones]

On Fri, Dec 13, 2002 at 12:18:58PM -0500, Valdis.Kletnieks@vt.edu wrote:
 > On Fri, 13 Dec 2002 17:33:00 +0100, Petr Konecny <pekon@informatics.muni.cz>  said:
 > > > I see why the if/continue was added - you don't want to be
 > > > calling device_register()/pci_insert_device() if
 > > > pci_enable_device() loses.  I don't see why 2.5.50 moved the
 > > > code up after pci_setup_device(). There's an outside chance
 > > > that the concept of moving the call was correct, but that it
 > > > should have been moved to between the calls to
 > > > pci_assign_resource() and pci_readb().  If that's the case,
 > > > then you're correct as well....
 > > I can confirm that this indeed works. I moved the two lines before
 > > pci_readb and the card works (every character you now read went through
 > > it). Who shall submit a patch to Linus ?
 > 
 > The problem is this from the 2.5.50 Changelog that Linus posted:
 > 
 > Dave Jones <davej@suse.de>:
 > ...
 >   o make cardbus PCI enable earlier
 > 
 > I'm willing to submit a patch, but I think Dave has to make the call whether
 > it should be backed out entirely, or moved after pci_assign_resource().
 > I certainly don't understand the code *or* PCI well enough to decide between
 > those two option...

It's my understanding that pci_enable_device() *must* be called
before we fiddle with dev->resource, dev->irq and the like.

		Dave

-- 
| Dave Jones.        http://www.codemonkey.org.uk
| SuSE Labs

Re: 2.5.5[01]]: Xircom Cardbus broken (PCI resource collisions)

[Valdis.Kletnieks]

[-- Attachment #1: Type: text/plain, Size: 1560 bytes --]

On Fri, 13 Dec 2002 17:36:56 GMT, Dave Jones said:

> It's my understanding that pci_enable_device() *must* be called
> before we fiddle with dev->resource, dev->irq and the like.

OK.. it looks like the problem only hits if it's a PCMCIA card *with an
onboard ROM*.

The immediate problem cause is in pcibios_enable_resources():

                if (!r->start && r->end) {
                        printk(KERN_ERR "PCI: Device %s not available because of resource collisions\n", dev->slot_name);

If there's an onboard ROM, this is set up in pcibios_assign_resources():

                if (pci_probe & PCI_ASSIGN_ROMS) {
                        r = &dev->resource[PCI_ROM_RESOURCE];
                        r->end -= r->start;
                        r->start = 0;
                        if (r->end)
                                pci_assign_resource(dev, PCI_ROM_RESOURCE);
                }               

but this hasn't happened yet in cb_alloc():

                /* FIXME: Do we need to enable the expansion ROM? */
                for (r = 0; r < 7; r++) {
                        struct resource *res = dev->resource + r;
                        if (res->flags)
                                pci_assign_resource(dev, r);
                }

So I think right *AFTER* this code is the right place for pci_enable_device()
because otherwise we won't have allocated the ROMs, so we'll get conflicts.

/Valdis (who went nuts looking at <6 and <7 all over the place.  Should I
volunteer to clean these up to #defines rather than inline magic numbers? ;)

[-- Attachment #2: Type: application/pgp-signature, Size: 226 bytes --]

Re: 2.5.5[01]]: Xircom Cardbus broken (PCI resource collisions)

[Jeff Garzik]

Dave Jones wrote:
> It's my understanding that pci_enable_device() *must* be called
> before we fiddle with dev->resource, dev->irq and the like.


True and correct, but -- this particular case is inside the cardbus 
core, where it presumeably might have a better idea of when it is best 
to call pci_enable_device (or perhaps even not at all, and twiddle the 
bits itself).

Re: 2.5.5[01]]: Xircom Cardbus broken (PCI resource collisions)

[Zwane Mwaikambo]

On Fri, 13 Dec 2002 Valdis.Kletnieks@vt.edu wrote:

> On Fri, 13 Dec 2002 17:36:56 GMT, Dave Jones said:
>
> > It's my understanding that pci_enable_device() *must* be called
> > before we fiddle with dev->resource, dev->irq and the like.
>
> OK.. it looks like the problem only hits if it's a PCMCIA card *with an
> onboard ROM*.

Hmm i just saw this thread, which card is the non working one?;

The computer is still running 2.5.50

02:00.0 Ethernet controller: Digital Equipment Corporation DECchip 21142/43 (rev 41)
        Subsystem: CNet Technology Inc: Unknown device 401a
        Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B-
        Status: Cap+ 66Mhz- UDF- FastB2B+ ParErr- DEVSEL=medium >TAbort- <TAbort- <MAbort- >SERR- <PERR-
        Latency: 64 (5000ns min, 10000ns max)
        Interrupt: pin A routed to IRQ 9
        Region 0: I/O ports at 1c00 [size=128]
        Region 1: Memory at 10800000 (32-bit, non-prefetchable) [size=1K]
        Expansion ROM at 10400000 [size=256K]
        Capabilities: [dc] Power Management version 1
                Flags: PMEClk- DSI- D1+ D2+ AuxCurrent=0mA PME(D0-,D1+,D2+,D3hot+,D3cold-)
                Status: D0 PME-Enable- DSel=0 DScale=0 PME+

# Bus options (PCI, PCMCIA, EISA, MCA, ISA)
# PCMCIA/CardBus support
CONFIG_PCMCIA=y
CONFIG_PCMCIA_XIRCOM=y
# CONFIG_PCMCIA_XIRTULIP is not set
# PCMCIA network device support
CONFIG_NET_PCMCIA=y
# CONFIG_PCMCIA_XIRC2PS is not set

--
function.linuxpower.ca

Re: 2.5.5[01]]: Xircom Cardbus broken (PCI resource collisions)

[Alan Cox]

On Fri, 2002-12-13 at 19:03, Jeff Garzik wrote:
> Dave Jones wrote:
> > It's my understanding that pci_enable_device() *must* be called
> > before we fiddle with dev->resource, dev->irq and the like.
> 
> 
> True and correct, but -- this particular case is inside the cardbus 
> core, where it presumeably might have a better idea of when it is best 
> to call pci_enable_device (or perhaps even not at all, and twiddle the 
> bits itself).

We can use enable_device_bars() to skip the ROM but do the rest, then
fix the ROM up 8)

Re: 2.5.5[01]]: Xircom Cardbus broken (PCI resource collisions)

[ALESSANDRO.SUARDI]

> On Fri, 13 Dec 2002 Valdis.Kletnieks@vt.edu wrote:
> > On Fri, 13 Dec 2002 17:36:56 GMT, Dave Jones said:
> >
> > > It's my understanding that pci_enable_device() *must* be called
> > > before we fiddle with dev->resource, dev->irq and the like.
> >
> > OK.. it looks like the problem only hits if it's a PCMCIA card *with an
> > onboard ROM*.
> Hmm i just saw this thread, which card is the non working one?;

It's a RBEM56G-100.

Sorry it took me a while to reply - Valdis' patch does fix the problem for
 me, too. Awaiting for a final form of the fix in the upcoming series :)


Thanks,

--alessandro

Re: 2.5.5[01]]: Xircom Cardbus broken (PCI resource collisions)

[Valdis.Kletnieks]

[-- Attachment #1: Type: text/plain, Size: 1183 bytes --]

On Sat, 14 Dec 2002 09:28:19 PST, "ALESSANDRO.SUARDI" said:
> > On Fri, 13 Dec 2002 Valdis.Kletnieks@vt.edu wrote:
> > > On Fri, 13 Dec 2002 17:36:56 GMT, Dave Jones said:
> > >
> > > > It's my understanding that pci_enable_device() *must* be called
> > > > before we fiddle with dev->resource, dev->irq and the like.
> > >
> > > OK.. it looks like the problem only hits if it's a PCMCIA card *with an
> > > onboard ROM*.
> > Hmm i just saw this thread, which card is the non working one?;
> 
> It's a RBEM56G-100.
> 
> Sorry it took me a while to reply - Valdis' patch does fix the problem for
>  me, too. Awaiting for a final form of the fix in the upcoming series :)

Same here.  I was getting bit on a Xircom card:

03:00.0 Ethernet controller: Xircom Cardbus Ethernet 10/100 (rev 03)
03:00.1 Serial controller: Xircom Cardbus Ethernet + 56k Modem (rev 03)

Dave has a point about not poking IRQ's before it's initialized, so I think
I'll let him and Alan discuss the *right* way to fix it. (Though if there's
need to test a patch more elegant/correct than mine, I'm more than happy to
do so...)


-- 
				Valdis Kletnieks
				Computer Systems Senior Engineer
				Virginia Tech


[-- Attachment #2: Type: application/pgp-signature, Size: 226 bytes --]