Re: [PATCH] arm/mach-omap2/display: fix possible object reference leak

Re: [PATCH] arm/mach-omap2/display: fix possible object reference leak

[Tony Lindgren]

Hi,

Adding devicetree list, Julia, Rob and Tomi to Cc.

* Peng Hao <peng.hao2@zte.com.cn> [190212 23:11]:
> of_find_device_by_node() takes a reference to the struct device
> when it finds a match via get_device.When returning error we should
> call put_device.
> 
> Signed-off-by: Peng Hao <peng.hao2@zte.com.cn>
> ---
>  arch/arm/mach-omap2/display.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/arch/arm/mach-omap2/display.c b/arch/arm/mach-omap2/display.c
> index f86b72d..c6aa9ed 100644
> --- a/arch/arm/mach-omap2/display.c
> +++ b/arch/arm/mach-omap2/display.c
> @@ -258,6 +258,7 @@ static int __init omapdss_init_of(void)
>  	r = of_platform_populate(node, NULL, NULL, &pdev->dev);
>  	if (r) {
>  		pr_err("Unable to populate DSS submodule devices\n");
> +		put_device(&pdev->dev);
>  		return r;
>  	}

In general, if the device tree node is never used afterwards,
should this be just:

	r = of_platform_populate(node, NULL, NULL, &pdev->dev);
	of_node_put(dev_node);
	if (r) {
		...
	}

If so, Julia might have a Coccinelle recpipe for it?

Regards,

Tony

Re: [PATCH] arm/mach-omap2/display: fix possible object reference leak

[Julia Lawall]

On Tue, 19 Feb 2019, Tony Lindgren wrote:

> Hi,
>
> Adding devicetree list, Julia, Rob and Tomi to Cc.
>
> * Peng Hao <peng.hao2@zte.com.cn> [190212 23:11]:
> > of_find_device_by_node() takes a reference to the struct device
> > when it finds a match via get_device.When returning error we should
> > call put_device.
> >
> > Signed-off-by: Peng Hao <peng.hao2@zte.com.cn>
> > ---
> >  arch/arm/mach-omap2/display.c | 1 +
> >  1 file changed, 1 insertion(+)
> >
> > diff --git a/arch/arm/mach-omap2/display.c b/arch/arm/mach-omap2/display.c
> > index f86b72d..c6aa9ed 100644
> > --- a/arch/arm/mach-omap2/display.c
> > +++ b/arch/arm/mach-omap2/display.c
> > @@ -258,6 +258,7 @@ static int __init omapdss_init_of(void)
> >  	r = of_platform_populate(node, NULL, NULL, &pdev->dev);
> >  	if (r) {
> >  		pr_err("Unable to populate DSS submodule devices\n");
> > +		put_device(&pdev->dev);
> >  		return r;
> >  	}
>
> In general, if the device tree node is never used afterwards,
> should this be just:
>
> 	r = of_platform_populate(node, NULL, NULL, &pdev->dev);
> 	of_node_put(dev_node);

Are these solving the same problems?  The of_node_put looks clearly
necessary, whether there is a success or failure.  I'm not familiar with
put_device.  I see that it does a kobject_put, but I don't know what
happens because of that.  But it looks like an inconsistency that Peng's
patch only considers the failure case, while your suggestion happens
always.

I guess Peng's patch is motivated by a Coccinelle script that Wen Yang
(also from ZTE) has been working on.  Perhaps there is a need to adjust
what is suggested by that script.

[Wen Yang added to CC]

julia

> 	if (r) {
> 		...
> 	}
>
> If so, Julia might have a Coccinelle recpipe for it?
>
> Regards,
>
> Tony
>

Re: [PATCH] arm/mach-omap2/display: fix possible object reference leak

[Julia Lawall]

On Tue, 19 Feb 2019, Tony Lindgren wrote:

> Hi,
>
> Adding devicetree list, Julia, Rob and Tomi to Cc.
>
> * Peng Hao <peng.hao2@zte.com.cn> [190212 23:11]:
> > of_find_device_by_node() takes a reference to the struct device
> > when it finds a match via get_device.When returning error we should
> > call put_device.
> >
> > Signed-off-by: Peng Hao <peng.hao2@zte.com.cn>
> > ---
> >  arch/arm/mach-omap2/display.c | 1 +
> >  1 file changed, 1 insertion(+)
> >
> > diff --git a/arch/arm/mach-omap2/display.c b/arch/arm/mach-omap2/display.c
> > index f86b72d..c6aa9ed 100644
> > --- a/arch/arm/mach-omap2/display.c
> > +++ b/arch/arm/mach-omap2/display.c
> > @@ -258,6 +258,7 @@ static int __init omapdss_init_of(void)
> >  	r = of_platform_populate(node, NULL, NULL, &pdev->dev);
> >  	if (r) {
> >  		pr_err("Unable to populate DSS submodule devices\n");
> > +		put_device(&pdev->dev);
> >  		return r;
> >  	}
>
> In general, if the device tree node is never used afterwards,
> should this be just:
>
> 	r = of_platform_populate(node, NULL, NULL, &pdev->dev);
> 	of_node_put(dev_node);
> 	if (r) {
> 		...
> 	}
>
> If so, Julia might have a Coccinelle recpipe for it?

Unfortunately this is not really an ideal case for Coccinelle, because
node is the result of calling a local function and Coccinelle doesn't by
default do any interprocedural analysis.  It is possible to write a rule
that explicitly looks for one function that returns a device node and then
the pattern of its usage in the caller, though.

julia

Re: [PATCH] arm/mach-omap2/display: fix possible object reference leak

[Tony Lindgren]

* Julia Lawall <julia.lawall@lip6.fr> [190219 17:33]:
> On Tue, 19 Feb 2019, Tony Lindgren wrote:
> > In general, if the device tree node is never used afterwards,
> > should this be just:
> >
> > 	r = of_platform_populate(node, NULL, NULL, &pdev->dev);
> > 	of_node_put(dev_node);
> > 	if (r) {
> > 		...
> > 	}
> >
> > If so, Julia might have a Coccinelle recpipe for it?
> 
> Unfortunately this is not really an ideal case for Coccinelle, because
> node is the result of calling a local function and Coccinelle doesn't by
> default do any interprocedural analysis.  It is possible to write a rule
> that explicitly looks for one function that returns a device node and then
> the pattern of its usage in the caller, though.

OK thanks for the information.

Regards,

Tony

Re: [PATCH] arm/mach-omap2/display: fix possible object referenceleak

[wen.yang99]

[-- Attachment #1.1: Type: text/plain, Size: 2489 bytes --]

Hi,
> > Adding devicetree list, Julia, Rob and Tomi to Cc.
> >
> > * Peng Hao <peng.hao2@zte.com.cn> [190212 23:11]:
> > > of_find_device_by_node() takes a reference to the struct device
> > > when it finds a match via get_device.When returning error we should
> > > call put_device.
> > >
> > > Signed-off-by: Peng Hao <peng.hao2@zte.com.cn>
> > > ---
> > >  arch/arm/mach-omap2/display.c | 1 +
> > >  1 file changed, 1 insertion(+)
> > >
> > > diff --git a/arch/arm/mach-omap2/display.c b/arch/arm/mach-omap2/display.c
> > > index f86b72d..c6aa9ed 100644
> > > --- a/arch/arm/mach-omap2/display.c
> > > +++ b/arch/arm/mach-omap2/display.c
> > > @@ -258,6 +258,7 @@ static int __init omapdss_init_of(void)
> > >      r = of_platform_populate(node, NULL, NULL, &pdev->dev);
> > >      if (r) {
> > >          pr_err("Unable to populate DSS submodule devices\n");
> > > +        put_device(&pdev->dev);
> > >          return r;
> > >      }
> >
> > In general, if the device tree node is never used afterwards,
> > should this be just:
> >
> >     r = of_platform_populate(node, NULL, NULL, &pdev->dev);
> >     of_node_put(dev_node);
> 
> Are these solving the same problems?  The of_node_put looks clearly
> necessary, whether there is a success or failure.  I'm not familiar with
> put_device.  I see that it does a kobject_put, but I don't know what
> happens because of that.  But it looks like an inconsistency that Peng's
> patch only considers the failure case, while your suggestion happens
> always.
>
> I guess Peng's patch is motivated by a Coccinelle script that Wen Yang
> (also from ZTE) has been working on.  Perhaps there is a need to adjust
> what is suggested by that script.

Yes, Peng Hao and I belong to the same company.
I found some code issues with this Coccinelle script:
Https://lkml.org/lkml/2019/2/16/107
Our other colleagues are submitting some patches to fix these discovered problems.

For this code(arch/arm/mach-omap2/display.c), there may be two issues:
1, the of_find_device_by_node() takes a reference to the underlying device
structure, we should call put_device()  release that reference.
2, of_find_compatible_node/of_find_node_by_name/of_parse_phandle...  will 
increase the refcount of device_node,  we should call of_node_put() to release
the refcount.

The first issue can already be detected by the Coccinelle script described above;
The second issue, we are still looking for ways to develop the corresponding Coccinelle script.

Regards,
  Wen