Hi,
On 10/28/20 10:31 AM, Mathias Nyman wrote:
> On 27.10.2020 23.51, Hans de Goede wrote:
>> Hi,
>>
>> I standard run my local kernel builds with lockdep enabled, booting 5.10-rc1 with lockdep enabled results in:
>>
>> [ 3.339982] =====================================================
>> [ 3.339984] WARNING: HARDIRQ-safe -> HARDIRQ-unsafe lock order detected
>> [ 3.339987] 5.10.0-rc1pdx86+ #8 Not tainted
>> [ 3.339988] -----------------------------------------------------
>> [ 3.339991] systemd-udevd/386 [HC0[0]:SC0[0]:HE0:SE1] is trying to acquire:
>> [ 3.339993] ffffffffb1a94038 (pin_fs_lock){+.+.}-{2:2}, at: simple_pin_fs+0x22/0xa0
>> [ 3.339999]
>> and this task is already holding:
>> [ 3.340002] ffff9e7b87fbc430 (&xhci->lock){-.-.}-{2:2}, at: xhci_alloc_streams+0x5f9/0x810
>> [ 3.340008] which would create a new lock dependency:
>> [ 3.340009] (&xhci->lock){-.-.}-{2:2} -> (pin_fs_lock){+.+.}-{2:2}
>> [ 3.340013]
>> but this new dependency connects a HARDIRQ-irq-safe lock:
>> [ 3.340016] (&xhci->lock){-.-.}-{2:2}
>
> Thanks, just read a similar report from Mike Galbraith <efault@xxxxxx>
>
> Looks like 673d74683627 ("usb: xhci: add debugfs support for ep with stream")
> creates the new xhci debugfs stream files while holding the xhci->lock,
> creating this dependency.
>
> Below code should help, but I need to find my UAS drive to test it.
>
> diff --git a/drivers/usb/host/xhci.c b/drivers/usb/host/xhci.c
> index 482fe8c5e3b4..d4a8d0efbbc4 100644
> --- a/drivers/usb/host/xhci.c
> +++ b/drivers/usb/host/xhci.c
> @@ -3533,11 +3533,14 @@ static int xhci_alloc_streams(struct usb_hcd *hcd, struct usb_device *udev,
> xhci_dbg(xhci, "Slot %u ep ctx %u now has streams.\n",
> udev->slot_id, ep_index);
> vdev->eps[ep_index].ep_state |= EP_HAS_STREAMS;
> - xhci_debugfs_create_stream_files(xhci, vdev, ep_index);
> }
> xhci_free_command(xhci, config_cmd);
> spin_unlock_irqrestore(&xhci->lock, flags);
>
> + for (i = 0; i < num_eps; i++) {
> + ep_index = xhci_get_endpoint_index(&eps[i]->desc);
> + xhci_debugfs_create_stream_files(xhci, vdev, ep_index);
> + }
> /* Subtract 1 for stream 0, which drivers can't use */
> return num_streams - 1;
>
I can confirm that this patch fixes the lockdep oops.
Regards,
Hans
