The driver used this construct:
#define KMB_PWM_LEADIN_MASK GENMASK(30, 0)
static inline void keembay_pwm_update_bits(struct keembay_pwm *priv, u32 mask,
u32 val, u32 offset)
{
u32 buff = readl(priv->base + offset);
buff = u32_replace_bits(buff, val, mask);
writel(buff, priv->base + offset);
}
...
keembay_pwm_update_bits(priv, KMB_PWM_LEADIN_MASK, 0,
KMB_PWM_LEADIN_OFFSET(pwm->hwpwm));
With CONFIG_CC_OPTIMIZE_FOR_SIZE the compiler (here: gcc 10.2.0) this
triggers:
In file included from /home/uwe/gsrc/linux/drivers/pwm/pwm-keembay.c:16:
In function ‘field_multiplier’,
inlined from ‘keembay_pwm_update_bits’ at /home/uwe/gsrc/linux/include/linux/bitfield.h:124:17:
/home/uwe/gsrc/linux/include/linux/bitfield.h:119:3: error: call to ‘__bad_mask’ declared with attribute error: bad bitfield mask
119 | __bad_mask();
| ^~~~~~~~~~~~
In function ‘field_multiplier’,
inlined from ‘keembay_pwm_update_bits’ at /home/uwe/gsrc/linux/include/linux/bitfield.h:154:1:
/home/uwe/gsrc/linux/include/linux/bitfield.h:119:3: error: call to ‘__bad_mask’ declared with attribute error: bad bitfield mask
119 | __bad_mask();
| ^~~~~~~~~~~~
The compiler doesn't seem to be able to notice that with field being
0x3ffffff the expression
if ((field | (field - 1)) & ((field | (field - 1)) + 1))
__bad_mask();
can be optimized away.
So use __always_inline and document the problem in a comment to fix
this.
Reported-by: kernel test robot <[email protected]>
Signed-off-by: Uwe Kleine-König <[email protected]>
---
Hello,
I'm not sure this is the right fix. Maybe the bitfield stuff can be
changed somehow to make this problem go away, too?
Best regards
Uwe
drivers/pwm/pwm-keembay.c | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/drivers/pwm/pwm-keembay.c b/drivers/pwm/pwm-keembay.c
index 2b6dd070daa4..cdfdef66ff8e 100644
--- a/drivers/pwm/pwm-keembay.c
+++ b/drivers/pwm/pwm-keembay.c
@@ -63,7 +63,12 @@ static int keembay_clk_enable(struct device *dev, struct clk *clk)
return devm_add_action_or_reset(dev, keembay_clk_unprepare, clk);
}
-static inline void keembay_pwm_update_bits(struct keembay_pwm *priv, u32 mask,
+/*
+ * With gcc 10, CONFIG_CC_OPTIMIZE_FOR_SIZE and only "inline" instead of
+ * "__always_inline" this fails to compile because the compiler doesn't notice
+ * for all valid masks (e.g. KMB_PWM_LEADIN_MASK) that they are ok.
+ */
+static __always_inline void keembay_pwm_update_bits(struct keembay_pwm *priv, u32 mask,
u32 val, u32 offset)
{
u32 buff = readl(priv->base + offset);
--
2.28.0
Hi Uwe,
> From: Uwe Kleine-König <[email protected]>
> Sent: Monday, 16 November, 2020 5:08 PM
> Subject: [PATCH RFC] pwm: keembay: Fix build failure with -Os
>
> The driver used this construct:
>
> #define KMB_PWM_LEADIN_MASK GENMASK(30, 0)
>
> static inline void keembay_pwm_update_bits(struct keembay_pwm
> *priv, u32 mask,
> u32 val, u32 offset)
> {
> u32 buff = readl(priv->base + offset);
>
> buff = u32_replace_bits(buff, val, mask);
> writel(buff, priv->base + offset);
> }
>
> ...
> keembay_pwm_update_bits(priv, KMB_PWM_LEADIN_MASK, 0,
> KMB_PWM_LEADIN_OFFSET(pwm-
> >hwpwm));
>
> With CONFIG_CC_OPTIMIZE_FOR_SIZE the compiler (here: gcc 10.2.0) this
> triggers:
>
> In file included from /home/uwe/gsrc/linux/drivers/pwm/pwm-
> keembay.c:16:
> In function ‘field_multiplier’,
> inlined from ‘keembay_pwm_update_bits’ at
> /home/uwe/gsrc/linux/include/linux/bitfield.h:124:17:
> /home/uwe/gsrc/linux/include/linux/bitfield.h:119:3: error: call to
> ‘__bad_mask’ declared with attribute error: bad bitfield mask
> 119 | __bad_mask();
> | ^~~~~~~~~~~~
> In function ‘field_multiplier’,
> inlined from ‘keembay_pwm_update_bits’ at
> /home/uwe/gsrc/linux/include/linux/bitfield.h:154:1:
> /home/uwe/gsrc/linux/include/linux/bitfield.h:119:3: error: call to
> ‘__bad_mask’ declared with attribute error: bad bitfield mask
> 119 | __bad_mask();
> | ^~~~~~~~~~~~
>
> The compiler doesn't seem to be able to notice that with field being
> 0x3ffffff the expression
>
> if ((field | (field - 1)) & ((field | (field - 1)) + 1))
> __bad_mask();
>
> can be optimized away.
>
> So use __always_inline and document the problem in a comment to fix
> this.
>
> Reported-by: kernel test robot <[email protected]>
> Signed-off-by: Uwe Kleine-König <[email protected]>
Thank you for spending time in resolving this build failure.
I shall prepare and share the next version of patch with your approach.
> ---
> Hello,
>
> I'm not sure this is the right fix. Maybe the bitfield stuff can be
> changed somehow to make this problem go away, too?
>
> Best regards
> Uwe
>
> drivers/pwm/pwm-keembay.c | 7 ++++++-
> 1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/pwm/pwm-keembay.c b/drivers/pwm/pwm-keembay.c
> index 2b6dd070daa4..cdfdef66ff8e 100644
> --- a/drivers/pwm/pwm-keembay.c
> +++ b/drivers/pwm/pwm-keembay.c
> @@ -63,7 +63,12 @@ static int keembay_clk_enable(struct device *dev,
> struct clk *clk)
> return devm_add_action_or_reset(dev, keembay_clk_unprepare, clk);
> }
>
> -static inline void keembay_pwm_update_bits(struct keembay_pwm *priv, u32
> mask,
> +/*
> + * With gcc 10, CONFIG_CC_OPTIMIZE_FOR_SIZE and only "inline" instead of
> + * "__always_inline" this fails to compile because the compiler doesn't notice
> + * for all valid masks (e.g. KMB_PWM_LEADIN_MASK) that they are ok.
> + */
> +static __always_inline void keembay_pwm_update_bits(struct
> keembay_pwm *priv, u32 mask,
> u32 val, u32 offset)
> {
> u32 buff = readl(priv->base + offset);
> --
> 2.28.0
Thanks,
Vijay
On Tue, Nov 17, 2020 at 05:29:01PM +0000, Ayyathurai, Vijayakannan wrote:
> Hi Uwe,
>
> > From: Uwe Kleine-König <[email protected]>
> > Sent: Monday, 16 November, 2020 5:08 PM
> > Subject: [PATCH RFC] pwm: keembay: Fix build failure with -Os
> >
> > The driver used this construct:
> >
> > #define KMB_PWM_LEADIN_MASK GENMASK(30, 0)
> >
> > static inline void keembay_pwm_update_bits(struct keembay_pwm
> > *priv, u32 mask,
> > u32 val, u32 offset)
> > {
> > u32 buff = readl(priv->base + offset);
> >
> > buff = u32_replace_bits(buff, val, mask);
> > writel(buff, priv->base + offset);
> > }
> >
> > ...
> > keembay_pwm_update_bits(priv, KMB_PWM_LEADIN_MASK, 0,
> > KMB_PWM_LEADIN_OFFSET(pwm-
> > >hwpwm));
> >
> > With CONFIG_CC_OPTIMIZE_FOR_SIZE the compiler (here: gcc 10.2.0) this
> > triggers:
> >
> > In file included from /home/uwe/gsrc/linux/drivers/pwm/pwm-
> > keembay.c:16:
> > In function ‘field_multiplier’,
> > inlined from ‘keembay_pwm_update_bits’ at
> > /home/uwe/gsrc/linux/include/linux/bitfield.h:124:17:
> > /home/uwe/gsrc/linux/include/linux/bitfield.h:119:3: error: call to
> > ‘__bad_mask’ declared with attribute error: bad bitfield mask
> > 119 | __bad_mask();
> > | ^~~~~~~~~~~~
> > In function ‘field_multiplier’,
> > inlined from ‘keembay_pwm_update_bits’ at
> > /home/uwe/gsrc/linux/include/linux/bitfield.h:154:1:
> > /home/uwe/gsrc/linux/include/linux/bitfield.h:119:3: error: call to
> > ‘__bad_mask’ declared with attribute error: bad bitfield mask
> > 119 | __bad_mask();
> > | ^~~~~~~~~~~~
> >
> > The compiler doesn't seem to be able to notice that with field being
> > 0x3ffffff the expression
> >
> > if ((field | (field - 1)) & ((field | (field - 1)) + 1))
> > __bad_mask();
> >
> > can be optimized away.
> >
> > So use __always_inline and document the problem in a comment to fix
> > this.
> >
> > Reported-by: kernel test robot <[email protected]>
> > Signed-off-by: Uwe Kleine-König <[email protected]>
>
> Thank you for spending time in resolving this build failure.
>
> I shall prepare and share the next version of patch with your approach.
I don't understand this last sentence. IMHO there is currently nothing
you have to do for this problem. You can send an Ack however if you want
to.
Best regards
Uwe
--
Pengutronix e.K. | Uwe Kleine-König |
Industrial Linux Solutions | https://www.pengutronix.de/ |
[Cc: += linux-pwm which I forgot for the initial submission]
Hello,
On Mon, Nov 16, 2020 at 10:08:04AM +0100, Uwe Kleine-König wrote:
> The driver used this construct:
>
> #define KMB_PWM_LEADIN_MASK GENMASK(30, 0)
>
> static inline void keembay_pwm_update_bits(struct keembay_pwm *priv, u32 mask,
> u32 val, u32 offset)
> {
> u32 buff = readl(priv->base + offset);
>
> buff = u32_replace_bits(buff, val, mask);
> writel(buff, priv->base + offset);
> }
>
> ...
> keembay_pwm_update_bits(priv, KMB_PWM_LEADIN_MASK, 0,
> KMB_PWM_LEADIN_OFFSET(pwm->hwpwm));
>
> With CONFIG_CC_OPTIMIZE_FOR_SIZE the compiler (here: gcc 10.2.0) this
> triggers:
>
> In file included from /home/uwe/gsrc/linux/drivers/pwm/pwm-keembay.c:16:
> In function ‘field_multiplier’,
> inlined from ‘keembay_pwm_update_bits’ at /home/uwe/gsrc/linux/include/linux/bitfield.h:124:17:
> /home/uwe/gsrc/linux/include/linux/bitfield.h:119:3: error: call to ‘__bad_mask’ declared with attribute error: bad bitfield mask
> 119 | __bad_mask();
> | ^~~~~~~~~~~~
> In function ‘field_multiplier’,
> inlined from ‘keembay_pwm_update_bits’ at /home/uwe/gsrc/linux/include/linux/bitfield.h:154:1:
> /home/uwe/gsrc/linux/include/linux/bitfield.h:119:3: error: call to ‘__bad_mask’ declared with attribute error: bad bitfield mask
> 119 | __bad_mask();
> | ^~~~~~~~~~~~
>
> The compiler doesn't seem to be able to notice that with field being
> 0x3ffffff the expression
>
> if ((field | (field - 1)) & ((field | (field - 1)) + 1))
> __bad_mask();
>
> can be optimized away.
>
> So use __always_inline and document the problem in a comment to fix
> this.
>
> Reported-by: kernel test robot <[email protected]>
> Signed-off-by: Uwe Kleine-König <[email protected]>
> ---
> Hello,
>
> I'm not sure this is the right fix. Maybe the bitfield stuff can be
> changed somehow to make this problem go away, too?
Note, this patch
Fixes: cdbea243f419 ("pwm: Add PWM driver for Intel Keem Bay")
so this isn't critical for v5.10.
@thierry: If this is ok for you and Vijayakannan, you can squash this
into the original commit.
Best regards
Uwe
--
Pengutronix e.K. | Uwe Kleine-König |
Industrial Linux Solutions | https://www.pengutronix.de/ |
Hi Thierry,
> From: Uwe Kleine-König <[email protected]>
> Subject: Re: [PATCH RFC] pwm: keembay: Fix build failure with -Os
>
> [Cc: += linux-pwm which I forgot for the initial submission]
>
> Hello,
>
> On Mon, Nov 16, 2020 at 10:08:04AM +0100, Uwe Kleine-König wrote:
> > The driver used this construct:
> >
> > #define KMB_PWM_LEADIN_MASK GENMASK(30, 0)
> >
> > static inline void keembay_pwm_update_bits(struct keembay_pwm
> *priv, u32 mask,
> > u32 val, u32 offset)
> > {
> > u32 buff = readl(priv->base + offset);
> >
> > buff = u32_replace_bits(buff, val, mask);
> > writel(buff, priv->base + offset);
> > }
> >
> > ...
> > keembay_pwm_update_bits(priv, KMB_PWM_LEADIN_MASK, 0,
> > KMB_PWM_LEADIN_OFFSET(pwm-
> >hwpwm));
> >
> > With CONFIG_CC_OPTIMIZE_FOR_SIZE the compiler (here: gcc 10.2.0) this
> > triggers:
> >
> > In file included from /home/uwe/gsrc/linux/drivers/pwm/pwm-
> keembay.c:16:
> > In function ‘field_multiplier’,
> > inlined from ‘keembay_pwm_update_bits’ at
> /home/uwe/gsrc/linux/include/linux/bitfield.h:124:17:
> > /home/uwe/gsrc/linux/include/linux/bitfield.h:119:3: error: call to
> ‘__bad_mask’ declared with attribute error: bad bitfield mask
> > 119 | __bad_mask();
> > | ^~~~~~~~~~~~
> > In function ‘field_multiplier’,
> > inlined from ‘keembay_pwm_update_bits’ at
> /home/uwe/gsrc/linux/include/linux/bitfield.h:154:1:
> > /home/uwe/gsrc/linux/include/linux/bitfield.h:119:3: error: call to
> ‘__bad_mask’ declared with attribute error: bad bitfield mask
> > 119 | __bad_mask();
> > | ^~~~~~~~~~~~
> >
> > The compiler doesn't seem to be able to notice that with field being
> > 0x3ffffff the expression
> >
> > if ((field | (field - 1)) & ((field | (field - 1)) + 1))
> > __bad_mask();
> >
> > can be optimized away.
> >
> > So use __always_inline and document the problem in a comment to fix
> > this.
> >
> > Reported-by: kernel test robot <[email protected]>
> > Signed-off-by: Uwe Kleine-König <[email protected]>
> > ---
> > Hello,
> >
> > I'm not sure this is the right fix. Maybe the bitfield stuff can be
> > changed somehow to make this problem go away, too?
>
> Note, this patch
>
> Fixes: cdbea243f419 ("pwm: Add PWM driver for Intel Keem Bay")
>
> so this isn't critical for v5.10.
>
> @thierry: If this is ok for you and Vijayakannan, you can squash this
> into the original commit.
>
I am ok with Uwe approach.
I have compiled the change and tested in Keembay board as well.
> Best regards
> Uwe
>
> --
> Pengutronix e.K. | Uwe Kleine-König |
> Industrial Linux Solutions | https://www.pengutronix.de/ |
Thanks,
Vijay
On Mon, Nov 16, 2020 at 10:08:04AM +0100, Uwe Kleine-König wrote:
> The driver used this construct:
>
> #define KMB_PWM_LEADIN_MASK GENMASK(30, 0)
>
> static inline void keembay_pwm_update_bits(struct keembay_pwm *priv, u32 mask,
> u32 val, u32 offset)
> {
> u32 buff = readl(priv->base + offset);
>
> buff = u32_replace_bits(buff, val, mask);
> writel(buff, priv->base + offset);
> }
>
> ...
> keembay_pwm_update_bits(priv, KMB_PWM_LEADIN_MASK, 0,
> KMB_PWM_LEADIN_OFFSET(pwm->hwpwm));
>
> With CONFIG_CC_OPTIMIZE_FOR_SIZE the compiler (here: gcc 10.2.0) this
> triggers:
>
> In file included from /home/uwe/gsrc/linux/drivers/pwm/pwm-keembay.c:16:
> In function ‘field_multiplier’,
> inlined from ‘keembay_pwm_update_bits’ at /home/uwe/gsrc/linux/include/linux/bitfield.h:124:17:
> /home/uwe/gsrc/linux/include/linux/bitfield.h:119:3: error: call to ‘__bad_mask’ declared with attribute error: bad bitfield mask
> 119 | __bad_mask();
> | ^~~~~~~~~~~~
> In function ‘field_multiplier’,
> inlined from ‘keembay_pwm_update_bits’ at /home/uwe/gsrc/linux/include/linux/bitfield.h:154:1:
> /home/uwe/gsrc/linux/include/linux/bitfield.h:119:3: error: call to ‘__bad_mask’ declared with attribute error: bad bitfield mask
> 119 | __bad_mask();
> | ^~~~~~~~~~~~
>
> The compiler doesn't seem to be able to notice that with field being
> 0x3ffffff the expression
>
> if ((field | (field - 1)) & ((field | (field - 1)) + 1))
> __bad_mask();
>
> can be optimized away.
>
> So use __always_inline and document the problem in a comment to fix
> this.
>
> Reported-by: kernel test robot <[email protected]>
> Signed-off-by: Uwe Kleine-König <[email protected]>
> ---
> Hello,
>
> I'm not sure this is the right fix. Maybe the bitfield stuff can be
> changed somehow to make this problem go away, too?
>
> Best regards
> Uwe
>
> drivers/pwm/pwm-keembay.c | 7 ++++++-
> 1 file changed, 6 insertions(+), 1 deletion(-)
Applied, thanks.
Thierry
On Wed, Nov 18, 2020 at 05:41:57PM +0000, Ayyathurai, Vijayakannan wrote:
> Hi Thierry,
>
> > From: Uwe Kleine-König <[email protected]>
> > Subject: Re: [PATCH RFC] pwm: keembay: Fix build failure with -Os
> >
> > [Cc: += linux-pwm which I forgot for the initial submission]
> >
> > Hello,
> >
> > On Mon, Nov 16, 2020 at 10:08:04AM +0100, Uwe Kleine-König wrote:
> > > The driver used this construct:
> > >
> > > #define KMB_PWM_LEADIN_MASK GENMASK(30, 0)
> > >
> > > static inline void keembay_pwm_update_bits(struct keembay_pwm
> > *priv, u32 mask,
> > > u32 val, u32 offset)
> > > {
> > > u32 buff = readl(priv->base + offset);
> > >
> > > buff = u32_replace_bits(buff, val, mask);
> > > writel(buff, priv->base + offset);
> > > }
> > >
> > > ...
> > > keembay_pwm_update_bits(priv, KMB_PWM_LEADIN_MASK, 0,
> > > KMB_PWM_LEADIN_OFFSET(pwm-
> > >hwpwm));
> > >
> > > With CONFIG_CC_OPTIMIZE_FOR_SIZE the compiler (here: gcc 10.2.0) this
> > > triggers:
> > >
> > > In file included from /home/uwe/gsrc/linux/drivers/pwm/pwm-
> > keembay.c:16:
> > > In function ‘field_multiplier’,
> > > inlined from ‘keembay_pwm_update_bits’ at
> > /home/uwe/gsrc/linux/include/linux/bitfield.h:124:17:
> > > /home/uwe/gsrc/linux/include/linux/bitfield.h:119:3: error: call to
> > ‘__bad_mask’ declared with attribute error: bad bitfield mask
> > > 119 | __bad_mask();
> > > | ^~~~~~~~~~~~
> > > In function ‘field_multiplier’,
> > > inlined from ‘keembay_pwm_update_bits’ at
> > /home/uwe/gsrc/linux/include/linux/bitfield.h:154:1:
> > > /home/uwe/gsrc/linux/include/linux/bitfield.h:119:3: error: call to
> > ‘__bad_mask’ declared with attribute error: bad bitfield mask
> > > 119 | __bad_mask();
> > > | ^~~~~~~~~~~~
> > >
> > > The compiler doesn't seem to be able to notice that with field being
> > > 0x3ffffff the expression
> > >
> > > if ((field | (field - 1)) & ((field | (field - 1)) + 1))
> > > __bad_mask();
> > >
> > > can be optimized away.
> > >
> > > So use __always_inline and document the problem in a comment to fix
> > > this.
> > >
> > > Reported-by: kernel test robot <[email protected]>
> > > Signed-off-by: Uwe Kleine-König <[email protected]>
> > > ---
> > > Hello,
> > >
> > > I'm not sure this is the right fix. Maybe the bitfield stuff can be
> > > changed somehow to make this problem go away, too?
> >
> > Note, this patch
> >
> > Fixes: cdbea243f419 ("pwm: Add PWM driver for Intel Keem Bay")
> >
> > so this isn't critical for v5.10.
> >
> > @thierry: If this is ok for you and Vijayakannan, you can squash this
> > into the original commit.
> >
>
> I am ok with Uwe approach.
> I have compiled the change and tested in Keembay board as well.
I'll take that as a Tested-by. Next time, if you go through the trouble
of testing a patch, make sure to reply with a:
Tested-by: Your Name <[email protected]>
So that patchwork can pick that up and credit you for it.
Thierry