[PATCH 1/3] cpufreq: ppc: Add missing #include <asm/smp.h>

[PATCH 1/3] cpufreq: ppc: Add missing #include <asm/smp.h>

[Geert Uytterhoeven]

If CONFIG_SMP=n, <linux/smp.h> does not include <asm/smp.h>, causing:

drivers/cpufreq/ppc-corenet-cpufreq.c: In function 'corenet_cpufreq_cpu_init':
drivers/cpufreq/ppc-corenet-cpufreq.c:173:3: error: implicit declaration of function 'get_hard_smp_processor_id' [-Werror=implicit-function-declaration]

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
---
 drivers/cpufreq/ppc-corenet-cpufreq.c |    2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/cpufreq/ppc-corenet-cpufreq.c b/drivers/cpufreq/ppc-corenet-cpufreq.c
index b7e677be1df0..e78f9c806de4 100644
--- a/drivers/cpufreq/ppc-corenet-cpufreq.c
+++ b/drivers/cpufreq/ppc-corenet-cpufreq.c
@@ -22,6 +22,8 @@
 #include <linux/smp.h>
 #include <sysdev/fsl_soc.h>
 
+#include <asm/smp.h>
+
 /**
  * struct cpu_data - per CPU data struct
  * @parent: the parent node of cpu clock
-- 
1.7.9.5

[PATCH 2/3] cpufreq: ppc: Fix integer overflow in expression

[Geert Uytterhoeven]

On 32-bit, "12 * NSEC_PER_SEC" doesn't fit in "unsigned long"
(NSEC_PER_SEC is a "long" constant), causing an integer overflow:

drivers/cpufreq/ppc-corenet-cpufreq.c: In function 'corenet_cpufreq_cpu_init':
drivers/cpufreq/ppc-corenet-cpufreq.c:211:9: warning: integer overflow in expression [-Woverflow]

Force the intermediate to be 64-bit by adding an "ULL" suffix to the
constant multiplier to fix this.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
---
 drivers/cpufreq/ppc-corenet-cpufreq.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/cpufreq/ppc-corenet-cpufreq.c b/drivers/cpufreq/ppc-corenet-cpufreq.c
index e78f9c806de4..53881d78a931 100644
--- a/drivers/cpufreq/ppc-corenet-cpufreq.c
+++ b/drivers/cpufreq/ppc-corenet-cpufreq.c
@@ -208,7 +208,7 @@ static int corenet_cpufreq_cpu_init(struct cpufreq_policy *policy)
 		per_cpu(cpu_data, i) = data;
 
 	policy->cpuinfo.transition_latency =
-				(12 * NSEC_PER_SEC) / fsl_get_sys_freq();
+				(12ULL * NSEC_PER_SEC) / fsl_get_sys_freq();
 	of_node_put(np);
 
 	return 0;
-- 
1.7.9.5

Re: [PATCH 2/3] cpufreq: ppc: Fix integer overflow in expression

[Viresh Kumar]

On 17 April 2014 15:23, Geert Uytterhoeven <geert+renesas@glider.be> wrote:
> On 32-bit, "12 * NSEC_PER_SEC" doesn't fit in "unsigned long"
> (NSEC_PER_SEC is a "long" constant), causing an integer overflow:
>
> drivers/cpufreq/ppc-corenet-cpufreq.c: In function 'corenet_cpufreq_cpu_init':
> drivers/cpufreq/ppc-corenet-cpufreq.c:211:9: warning: integer overflow in expression [-Woverflow]
>
> Force the intermediate to be 64-bit by adding an "ULL" suffix to the
> constant multiplier to fix this.
>
> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
> ---
>  drivers/cpufreq/ppc-corenet-cpufreq.c |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/cpufreq/ppc-corenet-cpufreq.c b/drivers/cpufreq/ppc-corenet-cpufreq.c
> index e78f9c806de4..53881d78a931 100644
> --- a/drivers/cpufreq/ppc-corenet-cpufreq.c
> +++ b/drivers/cpufreq/ppc-corenet-cpufreq.c
> @@ -208,7 +208,7 @@ static int corenet_cpufreq_cpu_init(struct cpufreq_policy *policy)
>                 per_cpu(cpu_data, i) = data;
>
>         policy->cpuinfo.transition_latency =
> -                               (12 * NSEC_PER_SEC) / fsl_get_sys_freq();
> +                               (12ULL * NSEC_PER_SEC) / fsl_get_sys_freq();
>         of_node_put(np);
>
>         return 0;

Acked-by: Viresh Kumar <viresh.kumar@linaro.org>

[PATCH 3/3] cpufreq: ppc: Fix handling of non-existent clocks

[Geert Uytterhoeven]

If the clock doesn't exist, clk_get_rate() returns -EINVAL, which becomes
a large number (freq is u32), failing the "freq < min_cpufreq" test.
Explicitly test for "(u32)-EINVAL" to fix this.

Update the comment, and fix a grammer issue while we're at it.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
---
 drivers/cpufreq/ppc-corenet-cpufreq.c |    7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/drivers/cpufreq/ppc-corenet-cpufreq.c b/drivers/cpufreq/ppc-corenet-cpufreq.c
index 53881d78a931..7027eab814ce 100644
--- a/drivers/cpufreq/ppc-corenet-cpufreq.c
+++ b/drivers/cpufreq/ppc-corenet-cpufreq.c
@@ -179,10 +179,11 @@ static int corenet_cpufreq_cpu_init(struct cpufreq_policy *policy)
 		clk = of_clk_get(data->parent, i);
 		freq = clk_get_rate(clk);
 		/*
-		 * the clock is valid if its frequency is not masked
-		 * and large than minimum allowed frequency.
+		 * the clock is valid if it exists, its frequency is not
+		 * masked, and larger than minimum allowed frequency.
 		 */
-		if (freq < min_cpufreq || (mask & (1 << i)))
+		if (freq == (u32)-EINVAL || freq < min_cpufreq ||
+		    (mask & (1 << i)))
 			table[i].frequency = CPUFREQ_ENTRY_INVALID;
 		else
 			table[i].frequency = freq / 1000;
-- 
1.7.9.5

Re: [PATCH 3/3] cpufreq: ppc: Fix handling of non-existent clocks

[Viresh Kumar]

On 17 April 2014 15:23, Geert Uytterhoeven <geert+renesas@glider.be> wrote:
> If the clock doesn't exist, clk_get_rate() returns -EINVAL

You clk_get_rate() isn't written well then, it should return zero.
@Mike: I didn't see this clearly mentioned in clk.h, should we fix
that?

>, which becomes
> a large number (freq is u32), failing the "freq < min_cpufreq" test.
> Explicitly test for "(u32)-EINVAL" to fix this.

That's a bad check. We should have done this instead:

(s32)freq < 0; but that would be true for high values of clock. And that's
why clk_get_rate() must return zero for errors.

> Update the comment, and fix a grammer issue while we're at it.
>
> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
> ---
>  drivers/cpufreq/ppc-corenet-cpufreq.c |    7 ++++---
>  1 file changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/cpufreq/ppc-corenet-cpufreq.c b/drivers/cpufreq/ppc-corenet-cpufreq.c
> index 53881d78a931..7027eab814ce 100644
> --- a/drivers/cpufreq/ppc-corenet-cpufreq.c
> +++ b/drivers/cpufreq/ppc-corenet-cpufreq.c
> @@ -179,10 +179,11 @@ static int corenet_cpufreq_cpu_init(struct cpufreq_policy *policy)
>                 clk = of_clk_get(data->parent, i);
>                 freq = clk_get_rate(clk);
>                 /*
> -                * the clock is valid if its frequency is not masked
> -                * and large than minimum allowed frequency.
> +                * the clock is valid if it exists, its frequency is not
> +                * masked, and larger than minimum allowed frequency.
>                  */
> -               if (freq < min_cpufreq || (mask & (1 << i)))
> +               if (freq == (u32)-EINVAL || freq < min_cpufreq ||
> +                   (mask & (1 << i)))
>                         table[i].frequency = CPUFREQ_ENTRY_INVALID;
>                 else
>                         table[i].frequency = freq / 1000;
> --
> 1.7.9.5
>

Re: [PATCH 3/3] cpufreq: ppc: Fix handling of non-existent clocks

[Geert Uytterhoeven]

Hi Viresh,

On Mon, Apr 21, 2014 at 8:01 AM, Viresh Kumar <viresh.kumar@linaro.org> wrote:
> On 17 April 2014 15:23, Geert Uytterhoeven <geert+renesas@glider.be> wrote:
>> If the clock doesn't exist, clk_get_rate() returns -EINVAL
>
> You clk_get_rate() isn't written well then, it should return zero.

You're right, thanks!
Once again I looked at the wrong clk_get_rate() implementation :-(
These non-CCF variants should die soon...

So the original code was correct: if the clock is not valid, freq == 0,
and thus "freq < min_cpufreq" is true.

> @Mike: I didn't see this clearly mentioned in clk.h, should we fix
> that?
>
>>, which becomes
>> a large number (freq is u32), failing the "freq < min_cpufreq" test.
>> Explicitly test for "(u32)-EINVAL" to fix this.
>
> That's a bad check. We should have done this instead:
>
> (s32)freq < 0; but that would be true for high values of clock. And that's
> why clk_get_rate() must return zero for errors.
>
>> Update the comment, and fix a grammer issue while we're at it.
>>
>> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
>> ---
>>  drivers/cpufreq/ppc-corenet-cpufreq.c |    7 ++++---
>>  1 file changed, 4 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/cpufreq/ppc-corenet-cpufreq.c b/drivers/cpufreq/ppc-corenet-cpufreq.c
>> index 53881d78a931..7027eab814ce 100644
>> --- a/drivers/cpufreq/ppc-corenet-cpufreq.c
>> +++ b/drivers/cpufreq/ppc-corenet-cpufreq.c
>> @@ -179,10 +179,11 @@ static int corenet_cpufreq_cpu_init(struct cpufreq_policy *policy)
>>                 clk = of_clk_get(data->parent, i);
>>                 freq = clk_get_rate(clk);
>>                 /*
>> -                * the clock is valid if its frequency is not masked
>> -                * and large than minimum allowed frequency.
>> +                * the clock is valid if it exists, its frequency is not
>> +                * masked, and larger than minimum allowed frequency.
>>                  */
>> -               if (freq < min_cpufreq || (mask & (1 << i)))
>> +               if (freq == (u32)-EINVAL || freq < min_cpufreq ||
>> +                   (mask & (1 << i)))
>>                         table[i].frequency = CPUFREQ_ENTRY_INVALID;
>>                 else
>>                         table[i].frequency = freq / 1000;

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Re: [PATCH 1/3] cpufreq: ppc: Add missing #include <asm/smp.h>

[Viresh Kumar]

On 17 April 2014 15:23, Geert Uytterhoeven <geert+renesas@glider.be> wrote:
> If CONFIG_SMP=n, <linux/smp.h> does not include <asm/smp.h>, causing:
>
> drivers/cpufreq/ppc-corenet-cpufreq.c: In function 'corenet_cpufreq_cpu_init':
> drivers/cpufreq/ppc-corenet-cpufreq.c:173:3: error: implicit declaration of function 'get_hard_smp_processor_id' [-Werror=implicit-function-declaration]
>
> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
> ---
>  drivers/cpufreq/ppc-corenet-cpufreq.c |    2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/drivers/cpufreq/ppc-corenet-cpufreq.c b/drivers/cpufreq/ppc-corenet-cpufreq.c
> index b7e677be1df0..e78f9c806de4 100644
> --- a/drivers/cpufreq/ppc-corenet-cpufreq.c
> +++ b/drivers/cpufreq/ppc-corenet-cpufreq.c
> @@ -22,6 +22,8 @@
>  #include <linux/smp.h>
>  #include <sysdev/fsl_soc.h>
>
> +#include <asm/smp.h>
> +

Probably similar to what Srivatsa did:

+#include <asm/smp.h> /* Required for cpu_sibling_mask() in UP configs */

??