[PATCH v3 0/2] cpufreq/amd-pstate: Fix null pointer dereference when frequency invariance is disabled

[PATCH v3 0/2] cpufreq/amd-pstate: Fix null pointer dereference when frequency invariance is disabled

[Wyes Karny]

Currently, amd_pstate only uses `adjust_perf` and `target` callback
functions to get the frequency/effective utilization data from scaling
governors. But the governor can switch back to fast_switch callback when
frequency invariance is disabled. In that case, if the driver didn't set
fast_switch function pointer, null pointer dereference can happen and
kernel will crash.

Currently, frequency invariance is turned off when any anomaly is
detected with aperf/mperf readings. Which triggers this problem.

Fix this by adding `fast_switch` function in the amd_pstate driver. The
same problem could arise with intel_pstate driver when the driver is in
passive mode and the governor is schedutil.

Additionally, display a warning message if the scaling driver sets
fast_swich_possible flag but not fast_switch function pointer.

Changelog:
v2 -> v3
- Add fast_switch callback in amd_pstate and warn if any driver
  sets fast_swich_possible flag but not fast_switch function pointer.

v1 -> v2
- Remove fast_switch_enabled flag dependency for adjust_perf callback

v1: https://lore.kernel.org/linux-pm/20230410095045.14872-1-wyes.karny@amd.com/
v2: https://lore.kernel.org/linux-pm/20230509180503.739208-1-wyes.karny@amd.com/

Gautham R. Shenoy (1):
  amd_pstate: Add ->fast_switch() callback

Wyes Karny (1):
  cpufreq: Warn if fast_switch is not set

 drivers/cpufreq/amd-pstate.c | 35 +++++++++++++++++++++++++++++------
 drivers/cpufreq/cpufreq.c    | 18 ++++++++++++++++++
 include/linux/cpufreq.h      |  5 ++++-
 3 files changed, 51 insertions(+), 7 deletions(-)

-- 
2.34.1

[PATCH v3 1/2] amd_pstate: Add ->fast_switch() callback

[Wyes Karny]

From: "Gautham R. Shenoy" <gautham.shenoy@amd.com>

Schedutil normally calls the adjust_perf callback for drivers with
adjust_perf callback available and fast_switch_possible flag set.
However, when frequency invariance is disabled, schedutil tries to
invoke fast_switch. So, there is a chance of kernel crash if this
function pointer is not set. To protect against this scenario add
fast_switch callback in amd_pstate driver.

Fixes: 1d215f0319c2 ("cpufreq: amd-pstate: Add fast switch function for AMD P-State")

Signed-off-by: Wyes Karny <wyes.karny@amd.com>
Signed-off-by: Gautham R. Shenoy <gautham.shenoy@amd.com>
---
 drivers/cpufreq/amd-pstate.c | 35 +++++++++++++++++++++++++++++------
 1 file changed, 29 insertions(+), 6 deletions(-)

diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c
index 5a3d4aa0f45a..6cf875d94c41 100644
--- a/drivers/cpufreq/amd-pstate.c
+++ b/drivers/cpufreq/amd-pstate.c
@@ -444,9 +444,8 @@ static int amd_pstate_verify(struct cpufreq_policy_data *policy)
 	return 0;
 }
 
-static int amd_pstate_target(struct cpufreq_policy *policy,
-			     unsigned int target_freq,
-			     unsigned int relation)
+static int amd_pstate_update_freq(struct cpufreq_policy *policy,
+				  unsigned int target_freq, bool fast_switch)
 {
 	struct cpufreq_freqs freqs;
 	struct amd_cpudata *cpudata = policy->driver_data;
@@ -465,14 +464,37 @@ static int amd_pstate_target(struct cpufreq_policy *policy,
 	des_perf = DIV_ROUND_CLOSEST(target_freq * cap_perf,
 				     cpudata->max_freq);
 
-	cpufreq_freq_transition_begin(policy, &freqs);
+	WARN_ON(fast_switch && !policy->fast_switch_enabled);
+	/*
+	 * If fast_switch is desired, then there aren't any registered
+	 * transition notifiers. See comment for
+	 * cpufreq_enable_fast_switch().
+	 */
+	if (!fast_switch)
+		cpufreq_freq_transition_begin(policy, &freqs);
+
 	amd_pstate_update(cpudata, min_perf, des_perf,
-			  max_perf, false, policy->governor->flags);
-	cpufreq_freq_transition_end(policy, &freqs, false);
+			max_perf, fast_switch, policy->governor->flags);
+
+	if (!fast_switch)
+		cpufreq_freq_transition_end(policy, &freqs, false);
 
 	return 0;
 }
 
+static int amd_pstate_target(struct cpufreq_policy *policy,
+			     unsigned int target_freq,
+			     unsigned int relation)
+{
+	return amd_pstate_update_freq(policy, target_freq, false);
+}
+
+static unsigned int amd_pstate_fast_switch(struct cpufreq_policy *policy,
+				  unsigned int target_freq)
+{
+	return amd_pstate_update_freq(policy, target_freq, true);
+}
+
 static void amd_pstate_adjust_perf(unsigned int cpu,
 				   unsigned long _min_perf,
 				   unsigned long target_perf,
@@ -1309,6 +1331,7 @@ static struct cpufreq_driver amd_pstate_driver = {
 	.flags		= CPUFREQ_CONST_LOOPS | CPUFREQ_NEED_UPDATE_LIMITS,
 	.verify		= amd_pstate_verify,
 	.target		= amd_pstate_target,
+	.fast_switch    = amd_pstate_fast_switch,
 	.init		= amd_pstate_cpu_init,
 	.exit		= amd_pstate_cpu_exit,
 	.suspend	= amd_pstate_cpu_suspend,
-- 
2.34.1

[PATCH v3 2/2] cpufreq: Warn if fast_switch is not set

[Wyes Karny]

If fast_switch_possible flag is set by the scaling driver, the governor
is free to select fast_switch function even if adjust_perf is set. When
the frequency invariance is disabled due to some reason governor
fallbacks to fast_switch if fast_switch_possible is set. This could
crash the kernel if the driver didn't set the fast_switch function
pointer.

This issue becomes apparent when aperf/mperf overflow occurs with
amd_pstate (passive) + schedutil.  When this happens, kernel disables
frequency invariance calculation which causes schedutil to fallback to
sugov_update_single_freq which currently relies on the fast_switch
callback.

Normal flow:
  sugov_update_single_perf
    cpufreq_driver_adjust_perf
      cpufreq_driver->adjust_perf

Error case flow:
  sugov_update_single_perf
    sugov_update_single_freq  <-- This is chosen because the freq invariant is disabled due to aperf/mperf overflow
      cpufreq_driver_fast_switch
         cpufreq_driver->fast_switch <-- Here NULL pointer dereference is happening, because fast_switch is not set

Put up a warning message if the driver sets fast_switch_possible flag
but not fast_switch.

Signed-off-by: Wyes Karny <wyes.karny@amd.com>
---
 drivers/cpufreq/cpufreq.c | 18 ++++++++++++++++++
 include/linux/cpufreq.h   |  5 ++++-
 2 files changed, 22 insertions(+), 1 deletion(-)

diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
index 6b52ebe5a890..180be9235b08 100644
--- a/drivers/cpufreq/cpufreq.c
+++ b/drivers/cpufreq/cpufreq.c
@@ -501,6 +501,13 @@ void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
 	if (!policy->fast_switch_possible)
 		return;
 
+	/**
+	 * It's not expected driver's fast_switch callback is not set
+	 * even fast_switch_possible is true.
+	 */
+	if (!cpufreq_driver_has_fast_switch())
+		pr_alert_once("fast_switch callback is not set\n");
+
 	mutex_lock(&cpufreq_fast_switch_lock);
 	if (cpufreq_fast_switch_count >= 0) {
 		cpufreq_fast_switch_count++;
@@ -2143,6 +2150,17 @@ unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
 }
 EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
 
+/**
+ * cpufreq_driver_has_fast_switch - Check "fast switch" callback.
+ *
+ * Return 'true' if the ->fast_switch callback is present for the
+ * current driver or 'false' otherwise.
+ */
+bool cpufreq_driver_has_fast_switch(void)
+{
+	return !!cpufreq_driver->fast_switch;
+}
+
 /**
  * cpufreq_driver_adjust_perf - Adjust CPU performance level in one go.
  * @cpu: Target CPU.
diff --git a/include/linux/cpufreq.h b/include/linux/cpufreq.h
index 26e2eb399484..4277d6f8e50c 100644
--- a/include/linux/cpufreq.h
+++ b/include/linux/cpufreq.h
@@ -340,7 +340,9 @@ struct cpufreq_driver {
 	/*
 	 * ->fast_switch() replacement for drivers that use an internal
 	 * representation of performance levels and can pass hints other than
-	 * the target performance level to the hardware.
+	 * the target performance level to the hardware. If driver is setting this,
+	 * then it needs to set fast_switch also. Because in certain scenario scale
+	 * invariance could be disabled and governor can switch back to fast_switch.
 	 */
 	void		(*adjust_perf)(unsigned int cpu,
 				       unsigned long min_perf,
@@ -604,6 +606,7 @@ struct cpufreq_governor {
 /* Pass a target to the cpufreq driver */
 unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
 					unsigned int target_freq);
+bool cpufreq_driver_has_fast_switch(void);
 void cpufreq_driver_adjust_perf(unsigned int cpu,
 				unsigned long min_perf,
 				unsigned long target_perf,
-- 
2.34.1

Re: [PATCH v3 2/2] cpufreq: Warn if fast_switch is not set

[Viresh Kumar]

On 15-05-23, 11:34, Wyes Karny wrote:
> If fast_switch_possible flag is set by the scaling driver, the governor
> is free to select fast_switch function even if adjust_perf is set. When
> the frequency invariance is disabled due to some reason governor
> fallbacks to fast_switch if fast_switch_possible is set. This could
> crash the kernel if the driver didn't set the fast_switch function
> pointer.
> 
> This issue becomes apparent when aperf/mperf overflow occurs with
> amd_pstate (passive) + schedutil.  When this happens, kernel disables
> frequency invariance calculation which causes schedutil to fallback to
> sugov_update_single_freq which currently relies on the fast_switch
> callback.
> 
> Normal flow:
>   sugov_update_single_perf
>     cpufreq_driver_adjust_perf
>       cpufreq_driver->adjust_perf
> 
> Error case flow:
>   sugov_update_single_perf
>     sugov_update_single_freq  <-- This is chosen because the freq invariant is disabled due to aperf/mperf overflow
>       cpufreq_driver_fast_switch
>          cpufreq_driver->fast_switch <-- Here NULL pointer dereference is happening, because fast_switch is not set

Not sure if all these details are required for this patch or not. It
is logically incorrect to set fast_switch_possible, while fast_switch
isn't set. That's a reason enough.

> Put up a warning message if the driver sets fast_switch_possible flag
> but not fast_switch.
> 
> Signed-off-by: Wyes Karny <wyes.karny@amd.com>
> ---
>  drivers/cpufreq/cpufreq.c | 18 ++++++++++++++++++
>  include/linux/cpufreq.h   |  5 ++++-
>  2 files changed, 22 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> index 6b52ebe5a890..180be9235b08 100644
> --- a/drivers/cpufreq/cpufreq.c
> +++ b/drivers/cpufreq/cpufreq.c
> @@ -501,6 +501,13 @@ void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
>  	if (!policy->fast_switch_possible)
>  		return;
>  
> +	/**

Doc style comments aren't required here I guess.

> +	 * It's not expected driver's fast_switch callback is not set
> +	 * even fast_switch_possible is true.
> +	 */
> +	if (!cpufreq_driver_has_fast_switch())
> +		pr_alert_once("fast_switch callback is not set\n");
> +
>  	mutex_lock(&cpufreq_fast_switch_lock);
>  	if (cpufreq_fast_switch_count >= 0) {
>  		cpufreq_fast_switch_count++;
> @@ -2143,6 +2150,17 @@ unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
>  }
>  EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
>  
> +/**
> + * cpufreq_driver_has_fast_switch - Check "fast switch" callback.
> + *
> + * Return 'true' if the ->fast_switch callback is present for the
> + * current driver or 'false' otherwise.
> + */
> +bool cpufreq_driver_has_fast_switch(void)

Why create a routine for this, when no one else is going to use it ?

> +{
> +	return !!cpufreq_driver->fast_switch;
> +}

I think you should add the required check in cpufreq_online(), after
cpufreq_driver->init() is called, and return failure if fast_switch
isn't set and fast_switch_possible is.

-- 
viresh

Re: [PATCH v3 2/2] cpufreq: Warn if fast_switch is not set

[Wyes Karny]

Hi Viresh,

Thanks for looking into this patch.

On 16 May 06:40, Viresh Kumar wrote:
> On 15-05-23, 11:34, Wyes Karny wrote:
> > If fast_switch_possible flag is set by the scaling driver, the governor
> > is free to select fast_switch function even if adjust_perf is set. When
> > the frequency invariance is disabled due to some reason governor
> > fallbacks to fast_switch if fast_switch_possible is set. This could
> > crash the kernel if the driver didn't set the fast_switch function
> > pointer.
> > 
> > This issue becomes apparent when aperf/mperf overflow occurs with
> > amd_pstate (passive) + schedutil.  When this happens, kernel disables
> > frequency invariance calculation which causes schedutil to fallback to
> > sugov_update_single_freq which currently relies on the fast_switch
> > callback.
> > 
> > Normal flow:
> >   sugov_update_single_perf
> >     cpufreq_driver_adjust_perf
> >       cpufreq_driver->adjust_perf
> > 
> > Error case flow:
> >   sugov_update_single_perf
> >     sugov_update_single_freq  <-- This is chosen because the freq invariant is disabled due to aperf/mperf overflow
> >       cpufreq_driver_fast_switch
> >          cpufreq_driver->fast_switch <-- Here NULL pointer dereference is happening, because fast_switch is not set
> 
> Not sure if all these details are required for this patch or not. It
> is logically incorrect to set fast_switch_possible, while fast_switch
> isn't set. That's a reason enough.

Sure, I'll remove this if it's unnecessary.

> 
> > Put up a warning message if the driver sets fast_switch_possible flag
> > but not fast_switch.
> > 
> > Signed-off-by: Wyes Karny <wyes.karny@amd.com>
> > ---
> >  drivers/cpufreq/cpufreq.c | 18 ++++++++++++++++++
> >  include/linux/cpufreq.h   |  5 ++++-
> >  2 files changed, 22 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> > index 6b52ebe5a890..180be9235b08 100644
> > --- a/drivers/cpufreq/cpufreq.c
> > +++ b/drivers/cpufreq/cpufreq.c
> > @@ -501,6 +501,13 @@ void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
> >  	if (!policy->fast_switch_possible)
> >  		return;
> >  
> > +	/**
> 
> Doc style comments aren't required here I guess.

Okay.
> 
> > +	 * It's not expected driver's fast_switch callback is not set
> > +	 * even fast_switch_possible is true.
> > +	 */
> > +	if (!cpufreq_driver_has_fast_switch())
> > +		pr_alert_once("fast_switch callback is not set\n");
> > +
> >  	mutex_lock(&cpufreq_fast_switch_lock);
> >  	if (cpufreq_fast_switch_count >= 0) {
> >  		cpufreq_fast_switch_count++;
> > @@ -2143,6 +2150,17 @@ unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
> >  }
> >  EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
> >  
> > +/**
> > + * cpufreq_driver_has_fast_switch - Check "fast switch" callback.
> > + *
> > + * Return 'true' if the ->fast_switch callback is present for the
> > + * current driver or 'false' otherwise.
> > + */
> > +bool cpufreq_driver_has_fast_switch(void)
> 
> Why create a routine for this, when no one else is going to use it ?
> 
> > +{
> > +	return !!cpufreq_driver->fast_switch;
> > +}
> 
> I think you should add the required check in cpufreq_online(), after
> cpufreq_driver->init() is called, and return failure if fast_switch
> isn't set and fast_switch_possible is.

Sure, I'll do that. But one caution if we return failure, the drivers which
has the above conditions won't load.

Thanks,
Wyes
> 
> -- 
> viresh

Re: [PATCH v3 2/2] cpufreq: Warn if fast_switch is not set

[Viresh Kumar]

On 16-05-23, 11:19, Wyes Karny wrote:
> Sure, I'll do that. But one caution if we return failure, the drivers which
> has the above conditions won't load.

They shouldn't :)

-- 
viresh