Re: [BUG] OHCI crash when handover fails

Re: [BUG] OHCI crash when handover fails

[Benjamin Herrenschmidt]

On Fri, 2011-05-27 at 09:53 +1000, Benjamin Herrenschmidt wrote:
> Hi folks !
> 
> Current upstream, POWER7 machine running bare metal, there's no BIOS at
> all, so the whole handover failing thing is an oddity in the first place
> which I haven't tracked down yet, but it should cause that crash:

BTW. The same handover failure didn't cause a crash yesterday, so this
is a regression that happened with whatever Linus pulled last night, I
probably won't have a chance to bisect it tho for various reasons.

Cheers,
Ben.

> ohci_hcd 0000:07:02.0: OHCI Host Controller
> ohci_hcd 0000:07:02.0: new USB bus registered, assigned bus number 4
> ohci_hcd 0000:07:02.0: USB HC takeover failed!  (BIOS/SMM bug)
> ohci_hcd 0000:07:02.0: can't setup
> ohci_hcd 0000:07:02.0: USB bus 4 deregistered
> ------------[ cut here ]------------
> kernel BUG at /home/benh/linux-impreza/fs/dcache.c:419!
> cpu 0x0: Vector: 700 (Program Check) at [c0000003dbb9b580]
>     pc: c00000000014848c: .dput+0x50/0x26c
>     lr: c00000000014847c: .dput+0x40/0x26c
>     sp: c0000003dbb9b800
>    msr: 9000000000029032
>   current = 0xc0000003dbc6cec0
>   paca    = 0xc00000000ff88000
>     pid   = 2185, comm = work_for_cpu
> kernel BUG at /home/benh/linux-impreza/fs/dcache.c:419!
> enter ? for help
> [c0000003dbb9b800] c000000000148674 .dput+0x238/0x26c (unreliable)
> [c0000003dbb9b8a0] c0000000004bfd44 .fs_remove_file+0x278/0x2dc
> [c0000003dbb9b970] c0000000004c0374 .usbfs_notify+0x258/0x2e0
> [c0000003dbb9ba20] c000000000602498 .notifier_call_chain+0x6c/0xe0
> [c0000003dbb9bac0] c00000000008ffd4 .__blocking_notifier_call_chain
> +0x60/0x90
> [c0000003dbb9bb70] c0000000004bde70 .usb_notify_remove_bus+0x24/0x3c
> [c0000003dbb9bbf0] c0000000004af760 .usb_deregister_bus+0x80/0xe0
> [c0000003dbb9bc80] c0000000004b1274 .usb_add_hcd+0x6ec/0x750
> [c0000003dbb9bd40] c0000000004bf284 .usb_hcd_pci_probe+0x1e0/0x2b0
> [c0000003dbb9bde0] c000000000304ccc .local_pci_probe+0x34/0x48
> [c0000003dbb9be50] c0000000000809ac .do_work_for_cpu+0x34/0x60
> [c0000003dbb9bed0] c00000000008975c .kthread+0xb0/0xbc
> [c0000003dbb9bf90] c00000000001ce98 .kernel_thread+0x54/0x70
> 0:mon> 
> 
> 
> Cheers,
> Ben.

Re: [BUG] OHCI crash when handover fails

[Sebastian Andrzej Siewior]

* Benjamin Herrenschmidt | 2011-05-27 11:02:48 [+1000]:

>BTW. The same handover failure didn't cause a crash yesterday, so this
>is a regression that happened with whatever Linus pulled last night, I
>probably won't have a chance to bisect it tho for various reasons.

Ben, this is probably fixed by
 http://marc.info/?l=linux-fsdevel&m=130676974227377&w=2

>Cheers,
>Ben.

Sebastian

Re: [BUG] OHCI crash when handover fails

[Benjamin Herrenschmidt]

On Mon, 2011-05-30 at 18:25 +0200, Sebastian Andrzej Siewior wrote:
> * Benjamin Herrenschmidt | 2011-05-27 11:02:48 [+1000]:
> 
> >BTW. The same handover failure didn't cause a crash yesterday, so this
> >is a regression that happened with whatever Linus pulled last night, I
> >probably won't have a chance to bisect it tho for various reasons.
> 
> Ben, this is probably fixed by
>  http://marc.info/?l=linux-fsdevel&m=130676974227377&w=2

Sorry for the delay testing, got bogged down by more 3.0-*
regressions :-(

The patch does indeed fix the problem for me. I noticed that it's still
not upstream tho :-) Can you send it to Linus asap ?

Cheers,
Ben.

Re: [BUG] OHCI crash when handover fails

[Sebastian Andrzej Siewior]

* Benjamin Herrenschmidt | 2011-06-02 17:58:26 [+1000]:

>The patch does indeed fix the problem for me. I noticed that it's still
>not upstream tho :-) 

There is another version of the patch pending. I hope Greg picks it but
he is in .jp as far as I can see on lwn so it may take a while.

>Can you send it to Linus asap ?

The problem should go away for you if you disable CONFIG_USB_DEVICEFS :)
Nobody in userland should actually use this. udev and friends migrated
to the /sys interface.

>Cheers,
>Ben.

Sebastian