Re: hid-core: fix lockdep in hid_input_report() v2

[Andrew Morton <akpm@linux-foundation.org>]

On Sun, 20 Sep 2009 08:41:40 +0200
Krzysztof Helt <krzysztof.h1@wp.pl> wrote:

> From: Krzysztof Helt <krzysztof.h1@wp.pl>
> 
> There is a lockdep warning in the hid_input_report() due to kmalloc(.,GFP_KERNEL) 
> allocation of a temporary buffer.
> Change hid_input_report() call inside the hid_ctrl() after Jiri Slaby's advice. This 
> converts kmalloc() call into kmalloc(.,GFP_ATOMIC).
> 
> The lockdep trace is:
> 
> WARNING: at kernel/lockdep.c:2457 lockdep_trace_alloc+0xca/0xd0()
> Hardware name: Product Name
> Modules linked in: usbhid(+) hid uhci_hcd ne2k_pci via_agp 8390 i2c_via
> Pid: 0, comm: swapper Not tainted 2.6.31 #74
> Call Trace:
>  [<c0243b8a>] ? lockdep_trace_alloc+0xca/0xd0
>  [<c0221d1c>] warn_slowpath_common+0x7c/0xa0
>  [<c0243b8a>] ? lockdep_trace_alloc+0xca/0xd0
>  [<c0221d55>] warn_slowpath_null+0x15/0x20
>  [<c0243b8a>] lockdep_trace_alloc+0xca/0xd0
>  [<c027d2c8>] kmem_cache_alloc+0x28/0x100
>  [<cc9359ef>] ? hid_input_report+0x6f/0x220 [hid]
>  [<cc9359ef>] hid_input_report+0x6f/0x220 [hid]
>  [<cc94cabf>] hid_ctrl+0x7f/0x180 [usbhid]
>  [<c03b28b1>] usb_hcd_giveback_urb+0x41/0xa0
>  [<cc91a3fc>] uhci_giveback_urb+0x8c/0x280 [uhci_hcd]
>  [<c0279853>] ? dma_pool_free+0xd3/0x120
>  [<cc919848>] ? uhci_free_td+0x58/0xa0 [uhci_hcd]
>  [<cc91ac01>] uhci_scan_schedule+0x301/0x910 [uhci_hcd]
>  [<cc91cac4>] uhci_irq+0x94/0x940 [uhci_hcd]
>  [<c0245c43>] ? __lock_acquire+0x433/0xab0
>  [<c0206246>] ? mask_and_ack_8259A+0x66/0x100
>  [<c03b494b>] usb_hcd_irq+0x6b/0x90
>  [<c0255e7d>] handle_IRQ_event+0x2d/0xc0
>  [<c0257e85>] handle_level_irq+0x65/0xe0
>  [<c0204c74>] handle_irq+0x34/0x60
>  [<c0204bc9>] do_IRQ+0x39/0xb0
>  [<c024365c>] ? trace_hardirqs_on_caller+0x12c/0x180
>  [<c020328e>] common_interrupt+0x2e/0x40
>  [<c0208d48>] ? default_idle+0x38/0x50
>  [<c02108df>] apm_cpu_idle+0x10f/0x290
>  [<c0201b11>] cpu_idle+0x21/0x40
>  [<c045892d>] rest_init+0x4d/0x60
>  [<c0575815>] start_kernel+0x235/0x280
>  [<c05751f0>] ? unknown_bootoption+0x0/0x210
>  [<c057503f>] __init_begin+0x3f/0x50
> 

In my 2.6.31 tree, kernel/lockdep.c:2457 is in the middle of a #define
and the changelog is missing lots of info so I have no hope of
understanding what's going on here.


> ---
> This lockdep is in the current git tree.
> 
> diff --git a/drivers/hid/usbhid/hid-core.c b/drivers/hid/usbhid/hid-core.c
> index 1b0e07a..ab2869d 100644
> --- a/drivers/hid/usbhid/hid-core.c
> +++ b/drivers/hid/usbhid/hid-core.c
> @@ -455,7 +455,7 @@ static void hid_ctrl(struct urb *urb)
>  		if (usbhid->ctrl[usbhid->ctrltail].dir == USB_DIR_IN)
>  			hid_input_report(urb->context,
>  				usbhid->ctrl[usbhid->ctrltail].report->type,
> -				urb->transfer_buffer, urb->actual_length, 0);
> +				urb->transfer_buffer, urb->actual_length, 1);
>  		break;
>  	case -ESHUTDOWN:	/* unplug */
>  		unplug = 1;

um, OK, we're clearly inside spinlock here.

<starts chasing through the HID code>

Why does hid_input_field() use hard-coded GFP_ATOMIC even when
interrupt=0?

OK, I got down to

	if (hid->claimed & HID_CLAIMED_HIDDEV && interrupt && hid->hiddev_hid_event)
		hid->hiddev_hid_event(hid, field, usage, value);

in hid_process_event() and gave up.  I can't actually find anywhere
where the `interrupt' arg to hid_input_report() would cause that code
to do something non-atomic when interrupt==0.  The code all seems a bit
smelly.

I don't understand this patch.  Can we have a better changelog please?