Bug#651215: Kernel fails to boot on NSLU2

Bug#651215: Kernel fails to boot on NSLU2

[Ben Hutchings]

On Tue, 2011-12-06 at 20:40 +0100, Jochen Friedrich wrote:
> Package: linux-image-3.1.0-1-ixp4xx
> Severity: normal
> 
> While 3.0.0-6 booted OK on NSLU2 platform, 3.1.0-1 or -4 fails to do so.
> 
> Boot log:
[...]
> [    0.000000] Linux version 3.1.0-1-ixp4xx (Debian 3.1.4-1) (waldi at debian.org)
> (gcc version 4.6.2 (Debian 4.6.2-4) ) #1 Wed Nov 30 06:35:38 UTC 2011
> [    0.000000] CPU: XScale-IXP42x Family [690541f1] revision 1 (ARMv5TE),
> cr=0000397f
> [    0.000000] CPU: VIVT data cache, VIVT instruction cache
> [    0.000000] Machine: Linksys NSLU2
[...]
> [    3.269860] ehci_hcd: USB 2.0 'Enhanced' Host Controller (EHCI) Driver
> [    3.350596] PCI: enabling device 0000:00:01.2 (0140 -> 0142)
> [    3.356553] ehci_hcd 0000:00:01.2: EHCI Host Controller
> [    3.452053] ehci_hcd 0000:00:01.2: new USB bus registered, assigned bus
> number 1
> [    3.459646] ehci_hcd 0000:00:01.2: coherent DMA mask 0x3ffffff is smaller
> than system GFP_DMA mask 0xffffffff
> [    3.570085] ehci_hcd 0000:00:01.2: coherent DMA mask 0x3ffffff is smaller
> than system GFP_DMA mask 0xffffffff
> [    3.680189] ehci_hcd 0000:00:01.2: coherent DMA mask 0x3ffffff is smaller
> than system GFP_DMA mask 0xffffffff
> (repeated over and over)

I assume that this has something to do with:

commit 650320181a08b64d4421c65c639cf47ad8cc2cd6
Author: Nicolas Pitre <nico@fluxnic.net>
Date:   Mon Jul 18 15:05:10 2011 -0400

    ARM: change ARM_DMA_ZONE_SIZE into a variable

commit 7553ee777b513c3bc8f45bb9fc75fb1bbc584ba1
Author: Nicolas Pitre <nicolas.pitre@linaro.org>
Date:   Tue Jul 5 22:28:09 2011 -0400

    ARM: mach-ixp4xx: move from ARM_DMA_ZONE_SIZE to mdesc->dma_zone_size

It's clear that the DMA zone size is supposed to be 64 MB on this
machine and I don't see why the DMA mask hasn't propagated correctly.
Any idea, Nicholas?

Ben.

-- 
Ben Hutchings
Absolutum obsoletum. (If it works, it's out of date.) - Stafford Beer
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 828 bytes
Desc: This is a digitally signed message part
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20111207/44fa3d7e/attachment-0001.sig>

Bug#651215: Kernel fails to boot on NSLU2

[Arnaud Patard (Rtp)]

Ben Hutchings <ben@decadent.org.uk> writes:
Hi,

> On Tue, 2011-12-06 at 20:40 +0100, Jochen Friedrich wrote:
>> Package: linux-image-3.1.0-1-ixp4xx
>> Severity: normal
>> 
>> While 3.0.0-6 booted OK on NSLU2 platform, 3.1.0-1 or -4 fails to do so.
>> 
>> Boot log:
> [...]
>> [    0.000000] Linux version 3.1.0-1-ixp4xx (Debian 3.1.4-1) (waldi at debian.org)
>> (gcc version 4.6.2 (Debian 4.6.2-4) ) #1 Wed Nov 30 06:35:38 UTC 2011
>> [    0.000000] CPU: XScale-IXP42x Family [690541f1] revision 1 (ARMv5TE),
>> cr=0000397f
>> [    0.000000] CPU: VIVT data cache, VIVT instruction cache
>> [    0.000000] Machine: Linksys NSLU2
> [...]
>> [    3.269860] ehci_hcd: USB 2.0 'Enhanced' Host Controller (EHCI) Driver
>> [    3.350596] PCI: enabling device 0000:00:01.2 (0140 -> 0142)
>> [    3.356553] ehci_hcd 0000:00:01.2: EHCI Host Controller
>> [    3.452053] ehci_hcd 0000:00:01.2: new USB bus registered, assigned bus
>> number 1
>> [    3.459646] ehci_hcd 0000:00:01.2: coherent DMA mask 0x3ffffff is smaller
>> than system GFP_DMA mask 0xffffffff
>> [    3.570085] ehci_hcd 0000:00:01.2: coherent DMA mask 0x3ffffff is smaller
>> than system GFP_DMA mask 0xffffffff
>> [    3.680189] ehci_hcd 0000:00:01.2: coherent DMA mask 0x3ffffff is smaller
>> than system GFP_DMA mask 0xffffffff
>> (repeated over and over)
>
> I assume that this has something to do with:
>
> commit 650320181a08b64d4421c65c639cf47ad8cc2cd6
> Author: Nicolas Pitre <nico@fluxnic.net>
> Date:   Mon Jul 18 15:05:10 2011 -0400
>
>     ARM: change ARM_DMA_ZONE_SIZE into a variable
>
> commit 7553ee777b513c3bc8f45bb9fc75fb1bbc584ba1
> Author: Nicolas Pitre <nicolas.pitre@linaro.org>
> Date:   Tue Jul 5 22:28:09 2011 -0400
>
>     ARM: mach-ixp4xx: move from ARM_DMA_ZONE_SIZE to mdesc->dma_zone_size
>
> It's clear that the DMA zone size is supposed to be 64 MB on this
> machine and I don't see why the DMA mask hasn't propagated correctly.
> Any idea, Nicholas?

I may be wrong but it seems that arm_dma_zone_size is used before being
set. It would be interesting if someone can boot test a nslu2 kernel with
appended patch.


Thanks,
Arnaud

-------------- next part --------------
A non-text attachment was scrubbed...
Name: nslu2_debug.patch
Type: text/x-diff
Size: 943 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20111207/af7b9578/attachment.bin>

Bug#651215: Kernel fails to boot on NSLU2

[Russell King - ARM Linux]

On Wed, Dec 07, 2011 at 11:05:58PM +0100, Arnaud Patard wrote:
> I may be wrong but it seems that arm_dma_zone_size is used before being
> set. It would be interesting if someone can boot test a nslu2 kernel with
> appended patch.

It does look like that's the case - arm_dma_zone_size is used in
arm_bootmem_free(), which is called from bootmem_init() and in turn
paging_init().

So, arm_dma_zone_size needs to be set before paging_init() is called.

I'd actually suggest moving it before:

        if (mdesc->restart_mode)
                reboot_setup(&mdesc->restart_mode);

so that it's set real early, to avoid any future problems with this.
The earlier these kinds of things are set, the less likely this kind
of bug will happen.

Bug#651215: Kernel fails to boot on NSLU2

[Nicolas Pitre]

On Wed, 7 Dec 2011, Russell King - ARM Linux wrote:

> On Wed, Dec 07, 2011 at 11:05:58PM +0100, Arnaud Patard wrote:
> > I may be wrong but it seems that arm_dma_zone_size is used before being
> > set. It would be interesting if someone can boot test a nslu2 kernel with
> > appended patch.
> 
> It does look like that's the case - arm_dma_zone_size is used in
> arm_bootmem_free(), which is called from bootmem_init() and in turn
> paging_init().
> 
> So, arm_dma_zone_size needs to be set before paging_init() is called.

Oops, indeed.

> I'd actually suggest moving it before:
> 
>         if (mdesc->restart_mode)
>                 reboot_setup(&mdesc->restart_mode);
> 
> so that it's set real early, to avoid any future problems with this.
> The earlier these kinds of things are set, the less likely this kind
> of bug will happen.

Arnaud: please resubmit your patch moving the call as suggested by 
Russell and mark it with:

Acked-by: Nicolas Pitre <nico@linaro.org>

and

CC: stable at kernel.org

Thanks


Nicolas