2018-03-08 09:33:49

by Merlijn Wajer

[permalink] [raw]
Subject: [RFC PATCH] usb: musb: Fix external abort in musb_remove

This fixes an oops on unbind / module unload. The fix is similar to the
one in this commit: 0c3aae9bd59978fb8c3557d7883380bef0f2cfa1 (USB: musb:
fix late external abort on suspend), which only fixes the issue for
musb_suspend.

musb_remove function now also flushes any work and also calls
musb_platform_exit before disabling runtime pm.

Signed-off-by: Merlijn Wajer <[email protected]>
---

If I understand commit 0c3aae9bd59978fb8c3557d7883380bef0f2cfa1 correctly,
this patch should also be backported to 4.9 (and newer kernels).

I have verified that this works on the Nokia N900, RX-51. The problem did not
occur on the Motorola Droid 4, but I have yet to test if this patch causes
issues there.

drivers/usb/musb/musb_core.c | 8 +++++++-
1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/drivers/usb/musb/musb_core.c b/drivers/usb/musb/musb_core.c
index e2e95071328a..4fb0f345c04e 100644
--- a/drivers/usb/musb/musb_core.c
+++ b/drivers/usb/musb/musb_core.c
@@ -2472,13 +2472,19 @@ static int musb_remove(struct platform_device *pdev)
musb_platform_disable(musb);
spin_lock_irqsave(&musb->lock, flags);
musb_disable_interrupts(musb);
+
+ musb->flush_irq_work = true;
+ while (flush_delayed_work(&musb->irq_work))
+ ;
+ musb->flush_irq_work = false;
+
musb_writeb(musb->mregs, MUSB_DEVCTL, 0);
+ musb_platform_exit(musb);
spin_unlock_irqrestore(&musb->lock, flags);

pm_runtime_dont_use_autosuspend(musb->controller);
pm_runtime_put_sync(musb->controller);
pm_runtime_disable(musb->controller);
- musb_platform_exit(musb);
musb_phy_callback = NULL;
if (musb->dma_controller)
musb_dma_controller_destroy(musb->dma_controller);
--
2.16.2



2018-03-08 16:47:45

by Bin Liu

[permalink] [raw]
Subject: Re: [RFC PATCH] usb: musb: Fix external abort in musb_remove

Hi,

On Thu, Mar 08, 2018 at 10:31:21AM +0100, Merlijn Wajer wrote:
> This fixes an oops on unbind / module unload. The fix is similar to the
> one in this commit: 0c3aae9bd59978fb8c3557d7883380bef0f2cfa1 (USB: musb:
> fix late external abort on suspend), which only fixes the issue for
> musb_suspend.
>
> musb_remove function now also flushes any work and also calls
> musb_platform_exit before disabling runtime pm.
>
> Signed-off-by: Merlijn Wajer <[email protected]>
> ---
>
> If I understand commit 0c3aae9bd59978fb8c3557d7883380bef0f2cfa1 correctly,
> this patch should also be backported to 4.9 (and newer kernels).
>
> I have verified that this works on the Nokia N900, RX-51. The problem did not
> occur on the Motorola Droid 4, but I have yet to test if this patch causes
> issues there.
>
> drivers/usb/musb/musb_core.c | 8 +++++++-
> 1 file changed, 7 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/usb/musb/musb_core.c b/drivers/usb/musb/musb_core.c
> index e2e95071328a..4fb0f345c04e 100644
> --- a/drivers/usb/musb/musb_core.c
> +++ b/drivers/usb/musb/musb_core.c
> @@ -2472,13 +2472,19 @@ static int musb_remove(struct platform_device *pdev)
> musb_platform_disable(musb);
> spin_lock_irqsave(&musb->lock, flags);
> musb_disable_interrupts(musb);
> +
> + musb->flush_irq_work = true;
> + while (flush_delayed_work(&musb->irq_work))
> + ;
> + musb->flush_irq_work = false;

No need to flush, the work is already cancelled at this point.

> +
> musb_writeb(musb->mregs, MUSB_DEVCTL, 0);
> + musb_platform_exit(musb);

This can be move down to out side of holding the spinlock,

> spin_unlock_irqrestore(&musb->lock, flags);

to here.

>
> pm_runtime_dont_use_autosuspend(musb->controller);
> pm_runtime_put_sync(musb->controller);
> pm_runtime_disable(musb->controller);
> - musb_platform_exit(musb);
> musb_phy_callback = NULL;
> if (musb->dma_controller)
> musb_dma_controller_destroy(musb->dma_controller);
> --

Thanks for the patch.

Regards,
-Bin.

2018-03-08 18:02:27

by Merlijn Wajer

[permalink] [raw]
Subject: Re: [RFC PATCH] usb: musb: Fix external abort in musb_remove

Hi,

On 08/03/18 17:46, Bin Liu wrote:


> No need to flush, the work is already cancelled at this point.
>
>> +
>> musb_writeb(musb->mregs, MUSB_DEVCTL, 0);
>> + musb_platform_exit(musb);
>
> This can be move down to out side of holding the spinlock,
>
>> spin_unlock_irqrestore(&musb->lock, flags);
>
> to here.

Thank you - with I've verified that it works with these changes. I'm
surprised, because I believe that I had tried this already. I must have
missed a step in my testing-routine.

I will send a patch shortly.

Cheers,
Merlijn


Attachments:
signature.asc (235.00 B)
OpenPGP digital signature

2018-03-08 18:25:42

by Bin Liu

[permalink] [raw]
Subject: Re: [RFC PATCH] usb: musb: Fix external abort in musb_remove

On Thu, Mar 08, 2018 at 07:00:16PM +0100, Merlijn Wajer wrote:
> Hi,
>
> On 08/03/18 17:46, Bin Liu wrote:
>
>
> > No need to flush, the work is already cancelled at this point.
> >
> >> +
> >> musb_writeb(musb->mregs, MUSB_DEVCTL, 0);
> >> + musb_platform_exit(musb);
> >
> > This can be move down to out side of holding the spinlock,
> >
> >> spin_unlock_irqrestore(&musb->lock, flags);
> >
> > to here.
>
> Thank you - with I've verified that it works with these changes. I'm
> surprised, because I believe that I had tried this already. I must have
> missed a step in my testing-routine.
>
> I will send a patch shortly.

Thank you for fixing this up.

Regards,
-Bin.