Re: [PATCH 2/3] dmaengine: usb-dmac: Fix PM reference leak in usb_dmac_probe()

[Vinod Koul <vkoul@kernel.org>]

On 31-05-21, 11:19, Johan Hovold wrote:
> On Mon, May 31, 2021 at 02:27:34PM +0530, Vinod Koul wrote:
> > On 31-05-21, 14:11, yukuai (C) wrote:
> > > On 2021/05/31 12:00, Vinod Koul wrote:
> > > > On 17-05-21, 16:18, Yu Kuai wrote:
> > > > > pm_runtime_get_sync will increment pm usage counter even it failed.
> > > > > Forgetting to putting operation will result in reference leak here.
> > > > > Fix it by replacing it with pm_runtime_resume_and_get to keep usage
> > > > > counter balanced.
> > > > > 
> > > > > Reported-by: Hulk Robot <hulkci@huawei.com>
> > > > > Signed-off-by: Yu Kuai <yukuai3@huawei.com>
> > > > > ---
> > > > >   drivers/dma/sh/usb-dmac.c | 2 +-
> > > > >   1 file changed, 1 insertion(+), 1 deletion(-)
> > > > > 
> > > > > diff --git a/drivers/dma/sh/usb-dmac.c b/drivers/dma/sh/usb-dmac.c
> > > > > index 8f7ceb698226..2a6c8fd8854e 100644
> > > > > --- a/drivers/dma/sh/usb-dmac.c
> > > > > +++ b/drivers/dma/sh/usb-dmac.c
> > > > > @@ -796,7 +796,7 @@ static int usb_dmac_probe(struct platform_device *pdev)
> > > > >   	/* Enable runtime PM and initialize the device. */
> > > > >   	pm_runtime_enable(&pdev->dev);
> > > > > -	ret = pm_runtime_get_sync(&pdev->dev);
> > > > > +	ret = pm_runtime_resume_and_get(&pdev->dev);
> > > > 
> > > > This does not seem to fix anything.. the below goto goes and disables
> > > > the runtime_pm for this device and thus there wont be any leak
> > > Hi,
> > > 
> > > If pm_runtime_get_sync() fails and increments the pm.usage_count
> > > variable, pm_runtime_disable() does not reset the counter, and
> > > we still need to decrement the usage count when pm_runtime_get_sync()
> > > fails. Do I miss anthing?
> > 
> > Yes the rumtime_pm is disabled on failure here and the count would have
> > no consequence...
> 
> You should still balance the PM usage counter as it isn't reset for
> example when reloading the driver.

Should I driver trust that on load PM usage counter is balanced and not
to be reset..?

> Using pm_runtime_resume_and_get() is one way of handling this, but
> alternatively you could also move the error_pm label above the
> pm_runtime_put() in the error path.

That would be a better way I think

-- 
~Vinod