diff mbox

[2/3,media] s5p-jpeg: Fix compilation with COMPILE_TEST

Message ID b7343e6296b5d1d68b7229b8307442fd4141bcb3.1410273306.git.m.chehab@samsung.com (mailing list archive)
State New, archived
Headers show

Commit Message

Mauro Carvalho Chehab Sept. 9, 2014, 2:38 p.m. UTC
ERROR: "__bad_ndelay" [drivers/media/platform/s5p-jpeg/s5p-jpeg.ko] undefined!

Yet, it sounds a bad idea to use ndelay to wait for 100 us
for the device to reset.

Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>

Comments

On 09/09/14 16:38, Mauro Carvalho Chehab wrote:
> ERROR: "__bad_ndelay" [drivers/media/platform/s5p-jpeg/s5p-jpeg.ko] undefined!
> 
> Yet, it sounds a bad idea to use ndelay to wait for 100 us
> for the device to reset.
> 
> Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
> Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
> 
> diff --git a/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c b/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c
> index e51c078360f5..01eeacf28843 100644
> --- a/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c
> +++ b/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c
> @@ -23,7 +23,9 @@ void exynos4_jpeg_sw_reset(void __iomem *base)
>  	reg = readl(base + EXYNOS4_JPEG_CNTL_REG);
>  	writel(reg & ~EXYNOS4_SOFT_RESET_HI, base + EXYNOS4_JPEG_CNTL_REG);
>  
> +#ifndef CONFIG_COMPILE_TEST
>  	ndelay(100000);
> +#endif

Wouldn't be a better fix to replace ndelay(100000); with udelay(100),
rather than sticking in a not so pretty #ifndef ?

I guess usleep_range() couldn't simply be used, since
exynos4_jpeg_sw_reset() is called with a spinlock held.

--
Regards,
Sylwester
Mauro Carvalho Chehab Sept. 9, 2014, 3:09 p.m. UTC | #2
Em Tue, 09 Sep 2014 16:58:58 +0200
Sylwester Nawrocki <s.nawrocki@samsung.com> escreveu:

> On 09/09/14 16:38, Mauro Carvalho Chehab wrote:
> > ERROR: "__bad_ndelay" [drivers/media/platform/s5p-jpeg/s5p-jpeg.ko] undefined!
> > 
> > Yet, it sounds a bad idea to use ndelay to wait for 100 us
> > for the device to reset.
> > 
> > Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
> > Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
> > 
> > diff --git a/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c b/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c
> > index e51c078360f5..01eeacf28843 100644
> > --- a/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c
> > +++ b/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c
> > @@ -23,7 +23,9 @@ void exynos4_jpeg_sw_reset(void __iomem *base)
> >  	reg = readl(base + EXYNOS4_JPEG_CNTL_REG);
> >  	writel(reg & ~EXYNOS4_SOFT_RESET_HI, base + EXYNOS4_JPEG_CNTL_REG);
> >  
> > +#ifndef CONFIG_COMPILE_TEST
> >  	ndelay(100000);
> > +#endif
> 
> Wouldn't be a better fix to replace ndelay(100000); with udelay(100),
> rather than sticking in a not so pretty #ifndef ?

Works for me. I'll submit a new version.

> I guess usleep_range() couldn't simply be used, since
> exynos4_jpeg_sw_reset() is called with a spinlock held.

Ok.

Regards,
Mauro
Jacek Anaszewski Sept. 9, 2014, 3:30 p.m. UTC | #3
Hi Mauro, Sylwester,

On 09/09/2014 05:09 PM, Mauro Carvalho Chehab wrote:
> Em Tue, 09 Sep 2014 16:58:58 +0200
> Sylwester Nawrocki <s.nawrocki@samsung.com> escreveu:
>
>> On 09/09/14 16:38, Mauro Carvalho Chehab wrote:
>>> ERROR: "__bad_ndelay" [drivers/media/platform/s5p-jpeg/s5p-jpeg.ko] undefined!
>>>
>>> Yet, it sounds a bad idea to use ndelay to wait for 100 us
>>> for the device to reset.
>>>
>>> Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
>>> Signed-off-by: Mauro Carvalho Chehab <m.chehab@samsung.com>
>>>
>>> diff --git a/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c b/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c
>>> index e51c078360f5..01eeacf28843 100644
>>> --- a/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c
>>> +++ b/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c
>>> @@ -23,7 +23,9 @@ void exynos4_jpeg_sw_reset(void __iomem *base)
>>>   	reg = readl(base + EXYNOS4_JPEG_CNTL_REG);
>>>   	writel(reg & ~EXYNOS4_SOFT_RESET_HI, base + EXYNOS4_JPEG_CNTL_REG);
>>>
>>> +#ifndef CONFIG_COMPILE_TEST
>>>   	ndelay(100000);
>>> +#endif
>>
>> Wouldn't be a better fix to replace ndelay(100000); with udelay(100),
>> rather than sticking in a not so pretty #ifndef ?
>
> Works for me. I'll submit a new version.
>
>> I guess usleep_range() couldn't simply be used, since
>> exynos4_jpeg_sw_reset() is called with a spinlock held.
>
> Ok.

Within few days I will perform some hardware tests, to verify
if there is more room for improvement here.

Best Regards,
Jacek Anaszewski
Arnd Bergmann Sept. 9, 2014, 5:54 p.m. UTC | #4
On Tuesday 09 September 2014 12:09:36 Mauro Carvalho Chehab wrote:
> -exynos4.c
> > > index e51c078360f5..01eeacf28843 100644
> > > --- a/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c
> > > +++ b/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c
> > > @@ -23,7 +23,9 @@ void exynos4_jpeg_sw_reset(void __iomem *base)
> > >     reg = readl(base + EXYNOS4_JPEG_CNTL_REG);
> > >     writel(reg & ~EXYNOS4_SOFT_RESET_HI, base + EXYNOS4_JPEG_CNTL_REG);
> > >  
> > > +#ifndef CONFIG_COMPILE_TEST
> > >     ndelay(100000);
> > > +#endif
> > 
> > Wouldn't be a better fix to replace ndelay(100000); with udelay(100),
> > rather than sticking in a not so pretty #ifndef ?
> 
> Works for me. I'll submit a new version.

New version looks good to me. On a more general level, I would argue
that we should not disable code based on COMPILE_TEST. The typical
use of this symbol is to make it possible to compile more code, not
to change the behavior of code on machines that were able to build
it already.

	Arnd
Mauro Carvalho Chehab Sept. 9, 2014, 6:40 p.m. UTC | #5
Em Tue, 09 Sep 2014 19:54:19 +0200
Arnd Bergmann <arnd@arndb.de> escreveu:

> On Tuesday 09 September 2014 12:09:36 Mauro Carvalho Chehab wrote:
> > -exynos4.c
> > > > index e51c078360f5..01eeacf28843 100644
> > > > --- a/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c
> > > > +++ b/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c
> > > > @@ -23,7 +23,9 @@ void exynos4_jpeg_sw_reset(void __iomem *base)
> > > >     reg = readl(base + EXYNOS4_JPEG_CNTL_REG);
> > > >     writel(reg & ~EXYNOS4_SOFT_RESET_HI, base + EXYNOS4_JPEG_CNTL_REG);
> > > >  
> > > > +#ifndef CONFIG_COMPILE_TEST
> > > >     ndelay(100000);
> > > > +#endif
> > > 
> > > Wouldn't be a better fix to replace ndelay(100000); with udelay(100),
> > > rather than sticking in a not so pretty #ifndef ?
> > 
> > Works for me. I'll submit a new version.
> 
> New version looks good to me. On a more general level, I would argue
> that we should not disable code based on COMPILE_TEST. The typical
> use of this symbol is to make it possible to compile more code, not
> to change the behavior of code on machines that were able to build
> it already.

Yeah, agreed as a general concept. In this case, however, it were
causing a compilation breakage on X86 (as it generates a non-existing
_bad_ndelay() symbol, if the time is bigger than 20000). See
include/asm-generic/delay.h.

Btw, I suspect that the only reason why ndelay(100000) causes a
compilation breakage is to avoid a big number, as the maximum limit
check ndelay() code (20000) at asm-generic is identical to the one
for udelay(). So, for ndelay, it means 20us, while, for udelay,
it means 20ms. Even so, both calls the very same implementation code.

Perhaps we should fix it, for both to accept a maximum time of 20ms.

Regards,
Mauro
diff mbox

Patch

diff --git a/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c b/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c
index e51c078360f5..01eeacf28843 100644
--- a/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c
+++ b/drivers/media/platform/s5p-jpeg/jpeg-hw-exynos4.c
@@ -23,7 +23,9 @@  void exynos4_jpeg_sw_reset(void __iomem *base)
 	reg = readl(base + EXYNOS4_JPEG_CNTL_REG);
 	writel(reg & ~EXYNOS4_SOFT_RESET_HI, base + EXYNOS4_JPEG_CNTL_REG);
 
+#ifndef CONFIG_COMPILE_TEST
 	ndelay(100000);
+#endif
 
 	writel(reg | EXYNOS4_SOFT_RESET_HI, base + EXYNOS4_JPEG_CNTL_REG);
 }