vm fault on OpenRD [was: Re: Allwinner A13: Fatal kernel mode data abort: 'Translation Fault (S)']

[Abdoulaye Walsimou Gaye]

On 05/02/2013 05:21 PM, Ian Lepore wrote:
[....]
> ehci0: <Marvell Integrated USB 2.0 controller> mem 0xf1050000-0xf1050fff irq 48,19 on simplebus0
> usbus0: EHCI version 1.0
> usbus0: stop timeout
> usbus0: set host controller mode
>
> vm_fault(0xc0cc6674, 0, 1, 0) -> 1
> Fatal kernel mode data abort: 'Translation Fault (S)'
> trapframe: 0xc0cfab68
> FSR=00000005, FAR=00000010, spsr=200000d3
> r0 =c36605c0, r1 =c0cc6390, r2 =600000d3, r3 =00000000
> r4 =c0a4e65c, r5 =80000003, r6 =c36640d0, r7 =c0c12278
> r8 =ffffffff, r9 =c350f280, r10=c3664080, r11=c0cfabf0
> r12=c0cc8aa0, ssp=c0cfabb4, slr=c0c26698, pc =c0a50964
>
> [ thread pid 0 tid 100000 ]
> Stopped at
> While the error messages are similar, it appears that the cause is
> completely different.

Ian,

Yes I realized that, sorry for the original thread hijack!


> Given that it appears to have crashed while
> initializing USB, I can think of two things to try...
>
> Add USB_HOST_ALIGN=32 to your kernel config file.  This is required for
> all armv4 and armv5 platforms, but isn't in most of our stock config
> files right now.


Making only this change on my KERNCONF file leads to kernel hang on, 'usbus0: set host controller mode', see here:

miibus0: <MII bus> on mge0
e1000phy0: <Marvell 88E1116R Gigabit PHY> PHY 8 on miibus0
e1000phy0:  none, 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, 1000baseT, 1000baseT-master, 1000baseT-FDX, 1000baseT-FDX-master, auto
uart0: <16550 or compatible> mem 0xf1012000-0xf101201f irq 33 on simplebus0
uart0: console (1056,n,8,1)
uart1: <16550 or compatible> mem 0xf1012100-0xf101211f irq 34 on simplebus0
cesa0: <Marvell Cryptographic Engine and Security Accelerator> mem 0xf1030000-0xf103ffff irq 22 on simplebus0
ehci0: <Marvell Integrated USB 2.0 controller> mem 0xf1050000-0xf1050fff irq 48,19 on simplebus0
usbus0: EHCI version 1.0
usbus0: stop timeout
usbus0: set host controller mode



> Also, try the attached patch.  It avoids unconditionally enabling the
> cache "allocate on write" feature in the hardware, and instead leaves it
> in whatever state the bootloader set it to (on the assumption that the
> vendor-supplied bootloader knows best about when this feature is
> problematic; I think it may depend somewhat on how the board is
> designed).
>


Applying only your patch still leads to vm fault, see here:

ehci0: <Marvell Integrated USB 2.0 controller> mem 0xf1050000-0xf1050fff irq 48,19 on simplebus0
usbus0: EHCI version 1.0
usbus0: stop timeout
usbus0: set host controller mode

vm_fault(0xc0cbe934, ff0000, 1, 0) -> 1
Fatal kernel mode data abort: 'Translation Fault (S)'
trapframe: 0xc0cf2b5c
FSR=00000005, FAR=00ff0000, spsr=200000d3
r0 =00000000, r1 =00000000, r2 =00000000, r3 =00000000
r4 =c34f0260, r5 =c365c080, r6 =00000000, r7 =00000001
r8 =c355ca80, r9 =00000000, r10=00000000, r11=c0cf2bd4
r12=00ff0000, ssp=c0cf2ba8, slr=c0c96acc, pc =c0a4b0c4


Making the two changes (KERNCONF USB_HOST_ALIGN=32 and your patch) lead also to the same vm fault as above.

The board's bootloader is a mainline u-boot, see here:
Marvell>> version

U-Boot 2013.04-00008-g500bccf (Apr 27 2013 - 01:08:22)
OpenRD-Base
armel-unknown-linux-gnueabi-gcc (embtoolkit-00143-g3687f38) 4.7.4 20130411 (prerelease)
GNU ld (GNU Binutils) 2.23.2

Note that you are right, it is USB initialization the issue. If I completely disable USB, the kernel boots
correctly and tries to mount nfs root filesystem.

I will try to figure out what's going on, any suggestions are welcome.

Thanks,
AWG