diff mbox series

[1/1] linux/io.h: Add cleanup defination for iounmap() and memunmap()

Message ID 20240731233145.2485874-1-Frank.Li@nxp.com (mailing list archive)
State Superseded
Headers show
Series [1/1] linux/io.h: Add cleanup defination for iounmap() and memunmap() | expand

Commit Message

Frank Li July 31, 2024, 11:31 p.m. UTC
Add DEFINE_FREE for iounmap() and memunmap() to support scope based
cleanup.

Signed-off-by: Frank Li <Frank.Li@nxp.com>
---
 include/linux/io.h | 3 +++
 1 file changed, 3 insertions(+)

Comments

Arnd Bergmann Aug. 1, 2024, 7:27 a.m. UTC | #1
On Thu, Aug 1, 2024, at 01:31, Frank Li wrote:
> Add DEFINE_FREE for iounmap() and memunmap() to support scope based
> cleanup.
>
> Signed-off-by: Frank Li <Frank.Li@nxp.com>
> ---
>  include/linux/io.h | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/include/linux/io.h b/include/linux/io.h
> index 59ec5eea696c4..7695d7973c8ff 100644
> --- a/include/linux/io.h
> +++ b/include/linux/io.h
> @@ -163,6 +163,9 @@ enum {
>  void *memremap(resource_size_t offset, size_t size, unsigned long flags);
>  void memunmap(void *addr);
> 
> +DEFINE_FREE(iounmap, void __iomem *, if (!IS_ERR_OR_NULL(_T)) iounmap(_T))
> +DEFINE_FREE(memunmap, void *, if (!IS_ERR_OR_NULL(_T)) memunmap(_T))

I don't like the use of IS_ERR_OR_NULL(), which tends
to indicate a problem in the interface design.

In which cases do you expect to need scope based cleanup
on an error pointer here? The only interfaces I see that
returns an __iomem error pointer are the devm_* ones, but
those have their own cleanup method.

    Arnd
Frank Li Aug. 1, 2024, 3:15 p.m. UTC | #2
On Thu, Aug 01, 2024 at 09:27:40AM +0200, Arnd Bergmann wrote:
> On Thu, Aug 1, 2024, at 01:31, Frank Li wrote:
> > Add DEFINE_FREE for iounmap() and memunmap() to support scope based
> > cleanup.
> >
> > Signed-off-by: Frank Li <Frank.Li@nxp.com>
> > ---
> >  include/linux/io.h | 3 +++
> >  1 file changed, 3 insertions(+)
> >
> > diff --git a/include/linux/io.h b/include/linux/io.h
> > index 59ec5eea696c4..7695d7973c8ff 100644
> > --- a/include/linux/io.h
> > +++ b/include/linux/io.h
> > @@ -163,6 +163,9 @@ enum {
> >  void *memremap(resource_size_t offset, size_t size, unsigned long flags);
> >  void memunmap(void *addr);
> >
> > +DEFINE_FREE(iounmap, void __iomem *, if (!IS_ERR_OR_NULL(_T)) iounmap(_T))
> > +DEFINE_FREE(memunmap, void *, if (!IS_ERR_OR_NULL(_T)) memunmap(_T))
>
> I don't like the use of IS_ERR_OR_NULL(), which tends
> to indicate a problem in the interface design.

Just !(_T) ? I just refer kfree()'s implementation.

>
> In which cases do you expect to need scope based cleanup
> on an error pointer here? The only interfaces I see that
> returns an __iomem error pointer are the devm_* ones, but
> those have their own cleanup method.


devm_* can help much especial in probe() function. but scope base cleanup
also useful.

Give a existed example:

drivers/clocksource/timer-fsl-ftm.c

static int __init ftm_timer_init(struct device_node *np) {
	...
	priv->clksrc_base = of_iomap(np, 1);
        if (!priv->clksrc_base) {
                pr_err("ftm: unable to map source timer registers\n");
                goto err_clksrc;
        }
	...

err_clksrc:
        iounmap(priv->clkevt_base);
}

If use scoped base cleanup it will be simple.
ftm_timer_init()
{
	base __free(ioumap) =  of_iomap(np, 1);

	...
	priv->clksrc_base = no_free_ptr(base);
	return 0;
}

drivers/clocksource/arm_arch_timer.c
Some example need map and unmap in function such as:
arch_timer_mem_frame_get_cntfrq()
{
	base = ioremap(frame->cntbase, frame->size);
	...
	iounmap(base);
}

of course this example is too simple. If there are many error branch
between ioremap() and iounmap(), scoped based clean up make code simple.

Frank

>
>     Arnd
Arnd Bergmann Aug. 2, 2024, 8:07 a.m. UTC | #3
On Thu, Aug 1, 2024, at 17:15, Frank Li wrote:
> On Thu, Aug 01, 2024 at 09:27:40AM +0200, Arnd Bergmann wrote:
>> On Thu, Aug 1, 2024, at 01:31, Frank Li wrote:
>> > diff --git a/include/linux/io.h b/include/linux/io.h
>> > index 59ec5eea696c4..7695d7973c8ff 100644
>> > --- a/include/linux/io.h
>> > +++ b/include/linux/io.h
>> > @@ -163,6 +163,9 @@ enum {
>> >  void *memremap(resource_size_t offset, size_t size, unsigned long flags);
>> >  void memunmap(void *addr);
>> >
>> > +DEFINE_FREE(iounmap, void __iomem *, if (!IS_ERR_OR_NULL(_T)) iounmap(_T))
>> > +DEFINE_FREE(memunmap, void *, if (!IS_ERR_OR_NULL(_T)) memunmap(_T))
>>
>> I don't like the use of IS_ERR_OR_NULL(), which tends
>> to indicate a problem in the interface design.
>
> Just !(_T) ? I just refer kfree()'s implementation.

Right, I think !(_T) is what we want here. I did see that kfree
has the same construct you are adding here, but did not read
up on how it got there.

>> In which cases do you expect to need scope based cleanup
>> on an error pointer here? The only interfaces I see that
>> returns an __iomem error pointer are the devm_* ones, but
>> those have their own cleanup method.
>
>
> devm_* can help much especial in probe() function. but scope base cleanup
> also useful.

Yes, that's fine. My only point here is that you can't combine
the cleanup function you added with devm_ioremap(), as that needs
to be paired with either implicit cleanup or with devm_iounmap() but
not plain iounmap().

    Arnd
diff mbox series

Patch

diff --git a/include/linux/io.h b/include/linux/io.h
index 59ec5eea696c4..7695d7973c8ff 100644
--- a/include/linux/io.h
+++ b/include/linux/io.h
@@ -163,6 +163,9 @@  enum {
 void *memremap(resource_size_t offset, size_t size, unsigned long flags);
 void memunmap(void *addr);
 
+DEFINE_FREE(iounmap, void __iomem *, if (!IS_ERR_OR_NULL(_T)) iounmap(_T))
+DEFINE_FREE(memunmap, void *, if (!IS_ERR_OR_NULL(_T)) memunmap(_T))
+
 /*
  * On x86 PAT systems we have memory tracking that keeps track of
  * the allowed mappings on memory ranges. This tracking works for