Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1622517 > unrolled thread

[patch 09/13] cpufreq/ia64: Replace racy task affinity logic

Started byThomas Gleixner <tglx@linutronix.de>
First post2017-04-12 22:30 +0200
Last post2017-04-13 04:50 +0200
Articles 4 — 3 participants

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  [patch 09/13] cpufreq/ia64: Replace racy task affinity logic Thomas Gleixner <tglx@linutronix.de> - 2017-04-12 22:30 +0200
    [patch V2 09/13] cpufreq/ia64: Replace racy task affinity logic Thomas Gleixner <tglx@linutronix.de> - 2017-04-12 23:00 +0200
      [tip:sched/core] cpufreq/ia64: Replace racy task affinity logic tip-bot for Thomas Gleixner <tipbot@zytor.com> - 2017-04-15 16:30 +0200
    Re: [patch 09/13] cpufreq/ia64: Replace racy task affinity logic Viresh Kumar <viresh.kumar@linaro.org> - 2017-04-13 04:50 +0200

#1622517 — [patch 09/13] cpufreq/ia64: Replace racy task affinity logic

FromThomas Gleixner <tglx@linutronix.de>
Date2017-04-12 22:30 +0200
Subject[patch 09/13] cpufreq/ia64: Replace racy task affinity logic
Message-ID<tvx3P-QV-1@gated-at.bofh.it>
The get() and target() callbacks must run on the affected cpu. This is
achieved by temporarily setting the affinity of the calling thread to the
requested CPU and reset it to the original affinity afterwards.

That's racy vs. concurrent affinity settings for that thread resulting in
code executing on the wrong CPU and overwriting the new affinity setting.

Replace it by work_on_cpu(). All call pathes which invoke the callbacks are
already protected against CPU hotplug.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: Viresh Kumar <viresh.kumar@linaro.org>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: linux-pm@vger.kernel.org
---
 drivers/cpufreq/ia64-acpi-cpufreq.c |   91 +++++++++++++++---------------------
 1 file changed, 38 insertions(+), 53 deletions(-)

--- a/drivers/cpufreq/ia64-acpi-cpufreq.c
+++ b/drivers/cpufreq/ia64-acpi-cpufreq.c
@@ -34,6 +34,11 @@ struct cpufreq_acpi_io {
 	unsigned int				resume;
 };
 
+struct cpufreq_acpi_req {
+	unsigned int		cpu;
+	unsigned int		state;
+};
+
 static struct cpufreq_acpi_io	*acpi_io_data[NR_CPUS];
 
 static struct cpufreq_driver acpi_cpufreq_driver;
@@ -83,8 +88,7 @@ processor_get_pstate (
 static unsigned
 extract_clock (
 	struct cpufreq_acpi_io *data,
-	unsigned value,
-	unsigned int cpu)
+	unsigned value)
 {
 	unsigned long i;
 
@@ -98,60 +102,42 @@ extract_clock (
 }
 
 
-static unsigned int
+static long
 processor_get_freq (
-	struct cpufreq_acpi_io	*data,
-	unsigned int		cpu)
+	void *arg)
 {
-	int			ret = 0;
-	u32			value = 0;
-	cpumask_t		saved_mask;
-	unsigned long 		clock_freq;
+	struct cpufreq_acpi_req *req = arg;
+	struct cpufreq_acpi_io	*data = acpi_io_data[req->cpu];
+	u32			value;
+	int			ret;
 
 	pr_debug("processor_get_freq\n");
-
-	saved_mask = current->cpus_allowed;
-	set_cpus_allowed_ptr(current, cpumask_of(cpu));
 	if (smp_processor_id() != cpu)
-		goto migrate_end;
+		return -EAGAIN;
 
 	/* processor_get_pstate gets the instantaneous frequency */
 	ret = processor_get_pstate(&value);
-
 	if (ret) {
-		set_cpus_allowed_ptr(current, &saved_mask);
 		pr_warn("get performance failed with error %d\n", ret);
-		ret = 0;
-		goto migrate_end;
+		return ret;
 	}
-	clock_freq = extract_clock(data, value, cpu);
-	ret = (clock_freq*1000);
-
-migrate_end:
-	set_cpus_allowed_ptr(current, &saved_mask);
-	return ret;
+	return 1000 * extract_clock(data, value);
 }
 
 
-static int
+static long
 processor_set_freq (
-	struct cpufreq_acpi_io	*data,
-	struct cpufreq_policy   *policy,
-	int			state)
+	void *arg)
 {
-	int			ret = 0;
-	u32			value = 0;
-	cpumask_t		saved_mask;
-	int			retval;
+	struct cpufreq_acpi_req *req = arg;
+	unsigned int		cpu = req->cpu;
+	struct cpufreq_acpi_io	*data = acpi_io_data[cpu];
+	int			ret, state = req->state;
+	u32			value;
 
 	pr_debug("processor_set_freq\n");
-
-	saved_mask = current->cpus_allowed;
-	set_cpus_allowed_ptr(current, cpumask_of(policy->cpu));
-	if (smp_processor_id() != policy->cpu) {
-		retval = -EAGAIN;
-		goto migrate_end;
-	}
+	if (smp_processor_id() != cpu)
+		return -EAGAIN;
 
 	if (state == data->acpi_data.state) {
 		if (unlikely(data->resume)) {
@@ -159,8 +145,7 @@ processor_set_freq (
 			data->resume = 0;
 		} else {
 			pr_debug("Already at target state (P%d)\n", state);
-			retval = 0;
-			goto migrate_end;
+			return 0;
 		}
 	}
 
@@ -171,7 +156,6 @@ processor_set_freq (
 	 * First we write the target state's 'control' value to the
 	 * control_register.
 	 */
-
 	value = (u32) data->acpi_data.states[state].control;
 
 	pr_debug("Transitioning to state: 0x%08x\n", value);
@@ -179,17 +163,11 @@ processor_set_freq (
 	ret = processor_set_pstate(value);
 	if (ret) {
 		pr_warn("Transition failed with error %d\n", ret);
-		retval = -ENODEV;
-		goto migrate_end;
+		return -ENODEV;
 	}
 
 	data->acpi_data.state = state;
-
-	retval = 0;
-
-migrate_end:
-	set_cpus_allowed_ptr(current, &saved_mask);
-	return (retval);
+	return 0;
 }
 
 
@@ -197,11 +175,13 @@ static unsigned int
 acpi_cpufreq_get (
 	unsigned int		cpu)
 {
-	struct cpufreq_acpi_io *data = acpi_io_data[cpu];
+	struct cpufreq_acpi_req req;
+	long ret;
 
-	pr_debug("acpi_cpufreq_get\n");
+	req.cpu = cpu;
+	ret = work_on_cpu(cpu, processor_get_freq, &req);
 
-	return processor_get_freq(data, cpu);
+	return ret > 0 ? (unsigned int) ret : 0;
 }
 
 
@@ -210,7 +190,12 @@ acpi_cpufreq_target (
 	struct cpufreq_policy   *policy,
 	unsigned int index)
 {
-	return processor_set_freq(acpi_io_data[policy->cpu], policy, index);
+	struct cpufreq_acpi_req req;
+
+	req.cpu = policy->cpu;
+	req.state = index;
+
+	return work_on_cpu(cpu, processor_set_freq, &req);
 }
 
 static int

[toc] | [next] | [standalone]

#1622549 — [patch V2 09/13] cpufreq/ia64: Replace racy task affinity logic

FromThomas Gleixner <tglx@linutronix.de>
Date2017-04-12 23:00 +0200
Subject[patch V2 09/13] cpufreq/ia64: Replace racy task affinity logic
Message-ID<tvxwS-15r-1@gated-at.bofh.it>
In reply to#1622517
The get() and target() callbacks must run on the affected cpu. This is
achieved by temporarily setting the affinity of the calling thread to the
requested CPU and reset it to the original affinity afterwards.

That's racy vs. concurrent affinity settings for that thread resulting in
code executing on the wrong CPU and overwriting the new affinity setting.

Replace it by work_on_cpu(). All call pathes which invoke the callbacks are
already protected against CPU hotplug.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: Viresh Kumar <viresh.kumar@linaro.org>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: linux-pm@vger.kernel.org
---

V2: Fixup kbuild-robot complaints

 drivers/cpufreq/ia64-acpi-cpufreq.c |   92 +++++++++++++++---------------------
 1 file changed, 39 insertions(+), 53 deletions(-)

Index: b/drivers/cpufreq/ia64-acpi-cpufreq.c
===================================================================
--- a/drivers/cpufreq/ia64-acpi-cpufreq.c
+++ b/drivers/cpufreq/ia64-acpi-cpufreq.c
@@ -34,6 +34,11 @@ struct cpufreq_acpi_io {
 	unsigned int				resume;
 };
 
+struct cpufreq_acpi_req {
+	unsigned int		cpu;
+	unsigned int		state;
+};
+
 static struct cpufreq_acpi_io	*acpi_io_data[NR_CPUS];
 
 static struct cpufreq_driver acpi_cpufreq_driver;
@@ -83,8 +88,7 @@ processor_get_pstate (
 static unsigned
 extract_clock (
 	struct cpufreq_acpi_io *data,
-	unsigned value,
-	unsigned int cpu)
+	unsigned value)
 {
 	unsigned long i;
 
@@ -98,60 +102,43 @@ extract_clock (
 }
 
 
-static unsigned int
+static long
 processor_get_freq (
-	struct cpufreq_acpi_io	*data,
-	unsigned int		cpu)
+	void *arg)
 {
-	int			ret = 0;
-	u32			value = 0;
-	cpumask_t		saved_mask;
-	unsigned long 		clock_freq;
+	struct cpufreq_acpi_req *req = arg;
+	unsigned int		cpu = req->cpu;
+	struct cpufreq_acpi_io	*data = acpi_io_data[cpu];
+	u32			value;
+	int			ret;
 
 	pr_debug("processor_get_freq\n");
-
-	saved_mask = current->cpus_allowed;
-	set_cpus_allowed_ptr(current, cpumask_of(cpu));
 	if (smp_processor_id() != cpu)
-		goto migrate_end;
+		return -EAGAIN;
 
 	/* processor_get_pstate gets the instantaneous frequency */
 	ret = processor_get_pstate(&value);
-
 	if (ret) {
-		set_cpus_allowed_ptr(current, &saved_mask);
 		pr_warn("get performance failed with error %d\n", ret);
-		ret = 0;
-		goto migrate_end;
+		return ret;
 	}
-	clock_freq = extract_clock(data, value, cpu);
-	ret = (clock_freq*1000);
-
-migrate_end:
-	set_cpus_allowed_ptr(current, &saved_mask);
-	return ret;
+	return 1000 * extract_clock(data, value);
 }
 
 
-static int
+static long
 processor_set_freq (
-	struct cpufreq_acpi_io	*data,
-	struct cpufreq_policy   *policy,
-	int			state)
+	void *arg)
 {
-	int			ret = 0;
-	u32			value = 0;
-	cpumask_t		saved_mask;
-	int			retval;
+	struct cpufreq_acpi_req *req = arg;
+	unsigned int		cpu = req->cpu;
+	struct cpufreq_acpi_io	*data = acpi_io_data[cpu];
+	int			ret, state = req->state;
+	u32			value;
 
 	pr_debug("processor_set_freq\n");
-
-	saved_mask = current->cpus_allowed;
-	set_cpus_allowed_ptr(current, cpumask_of(policy->cpu));
-	if (smp_processor_id() != policy->cpu) {
-		retval = -EAGAIN;
-		goto migrate_end;
-	}
+	if (smp_processor_id() != cpu)
+		return -EAGAIN;
 
 	if (state == data->acpi_data.state) {
 		if (unlikely(data->resume)) {
@@ -159,8 +146,7 @@ processor_set_freq (
 			data->resume = 0;
 		} else {
 			pr_debug("Already at target state (P%d)\n", state);
-			retval = 0;
-			goto migrate_end;
+			return 0;
 		}
 	}
 
@@ -171,7 +157,6 @@ processor_set_freq (
 	 * First we write the target state's 'control' value to the
 	 * control_register.
 	 */
-
 	value = (u32) data->acpi_data.states[state].control;
 
 	pr_debug("Transitioning to state: 0x%08x\n", value);
@@ -179,17 +164,11 @@ processor_set_freq (
 	ret = processor_set_pstate(value);
 	if (ret) {
 		pr_warn("Transition failed with error %d\n", ret);
-		retval = -ENODEV;
-		goto migrate_end;
+		return -ENODEV;
 	}
 
 	data->acpi_data.state = state;
-
-	retval = 0;
-
-migrate_end:
-	set_cpus_allowed_ptr(current, &saved_mask);
-	return (retval);
+	return 0;
 }
 
 
@@ -197,11 +176,13 @@ static unsigned int
 acpi_cpufreq_get (
 	unsigned int		cpu)
 {
-	struct cpufreq_acpi_io *data = acpi_io_data[cpu];
+	struct cpufreq_acpi_req req;
+	long ret;
 
-	pr_debug("acpi_cpufreq_get\n");
+	req.cpu = cpu;
+	ret = work_on_cpu(cpu, processor_get_freq, &req);
 
-	return processor_get_freq(data, cpu);
+	return ret > 0 ? (unsigned int) ret : 0;
 }
 
 
@@ -210,7 +191,12 @@ acpi_cpufreq_target (
 	struct cpufreq_policy   *policy,
 	unsigned int index)
 {
-	return processor_set_freq(acpi_io_data[policy->cpu], policy, index);
+	struct cpufreq_acpi_req req;
+
+	req.cpu = policy->cpu;
+	req.state = index;
+
+	return work_on_cpu(req.cpu, processor_set_freq, &req);
 }
 
 static int

[toc] | [prev] | [next] | [standalone]

#1624068 — [tip:sched/core] cpufreq/ia64: Replace racy task affinity logic

Fromtip-bot for Thomas Gleixner <tipbot@zytor.com>
Date2017-04-15 16:30 +0200
Subject[tip:sched/core] cpufreq/ia64: Replace racy task affinity logic
Message-ID<twwS6-7Dd-13@gated-at.bofh.it>
In reply to#1622549
Commit-ID:  38f05ed04beb276f780fcd2b5c0b78c76d0b3c0c
Gitweb:     http://git.kernel.org/tip/38f05ed04beb276f780fcd2b5c0b78c76d0b3c0c
Author:     Thomas Gleixner <tglx@linutronix.de>
AuthorDate: Wed, 12 Apr 2017 22:55:03 +0200
Committer:  Thomas Gleixner <tglx@linutronix.de>
CommitDate: Sat, 15 Apr 2017 12:20:55 +0200

cpufreq/ia64: Replace racy task affinity logic

The get() and target() callbacks must run on the affected cpu. This is
achieved by temporarily setting the affinity of the calling thread to the
requested CPU and reset it to the original affinity afterwards.

That's racy vs. concurrent affinity settings for that thread resulting in
code executing on the wrong CPU and overwriting the new affinity setting.

Replace it by work_on_cpu(). All call pathes which invoke the callbacks are
already protected against CPU hotplug.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Sebastian Siewior <bigeasy@linutronix.de>
Cc: linux-pm@vger.kernel.org
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Tejun Heo <tj@kernel.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Len Brown <lenb@kernel.org>
Link: http://lkml.kernel.org/r/alpine.DEB.2.20.1704122231100.2548@nanos
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

---
 drivers/cpufreq/ia64-acpi-cpufreq.c | 92 ++++++++++++++++---------------------
 1 file changed, 39 insertions(+), 53 deletions(-)

diff --git a/drivers/cpufreq/ia64-acpi-cpufreq.c b/drivers/cpufreq/ia64-acpi-cpufreq.c
index e28a31a..a757c0a 100644
--- a/drivers/cpufreq/ia64-acpi-cpufreq.c
+++ b/drivers/cpufreq/ia64-acpi-cpufreq.c
@@ -34,6 +34,11 @@ struct cpufreq_acpi_io {
 	unsigned int				resume;
 };
 
+struct cpufreq_acpi_req {
+	unsigned int		cpu;
+	unsigned int		state;
+};
+
 static struct cpufreq_acpi_io	*acpi_io_data[NR_CPUS];
 
 static struct cpufreq_driver acpi_cpufreq_driver;
@@ -83,8 +88,7 @@ processor_get_pstate (
 static unsigned
 extract_clock (
 	struct cpufreq_acpi_io *data,
-	unsigned value,
-	unsigned int cpu)
+	unsigned value)
 {
 	unsigned long i;
 
@@ -98,60 +102,43 @@ extract_clock (
 }
 
 
-static unsigned int
+static long
 processor_get_freq (
-	struct cpufreq_acpi_io	*data,
-	unsigned int		cpu)
+	void *arg)
 {
-	int			ret = 0;
-	u32			value = 0;
-	cpumask_t		saved_mask;
-	unsigned long 		clock_freq;
+	struct cpufreq_acpi_req *req = arg;
+	unsigned int		cpu = req->cpu;
+	struct cpufreq_acpi_io	*data = acpi_io_data[cpu];
+	u32			value;
+	int			ret;
 
 	pr_debug("processor_get_freq\n");
-
-	saved_mask = current->cpus_allowed;
-	set_cpus_allowed_ptr(current, cpumask_of(cpu));
 	if (smp_processor_id() != cpu)
-		goto migrate_end;
+		return -EAGAIN;
 
 	/* processor_get_pstate gets the instantaneous frequency */
 	ret = processor_get_pstate(&value);
-
 	if (ret) {
-		set_cpus_allowed_ptr(current, &saved_mask);
 		pr_warn("get performance failed with error %d\n", ret);
-		ret = 0;
-		goto migrate_end;
+		return ret;
 	}
-	clock_freq = extract_clock(data, value, cpu);
-	ret = (clock_freq*1000);
-
-migrate_end:
-	set_cpus_allowed_ptr(current, &saved_mask);
-	return ret;
+	return 1000 * extract_clock(data, value);
 }
 
 
-static int
+static long
 processor_set_freq (
-	struct cpufreq_acpi_io	*data,
-	struct cpufreq_policy   *policy,
-	int			state)
+	void *arg)
 {
-	int			ret = 0;
-	u32			value = 0;
-	cpumask_t		saved_mask;
-	int			retval;
+	struct cpufreq_acpi_req *req = arg;
+	unsigned int		cpu = req->cpu;
+	struct cpufreq_acpi_io	*data = acpi_io_data[cpu];
+	int			ret, state = req->state;
+	u32			value;
 
 	pr_debug("processor_set_freq\n");
-
-	saved_mask = current->cpus_allowed;
-	set_cpus_allowed_ptr(current, cpumask_of(policy->cpu));
-	if (smp_processor_id() != policy->cpu) {
-		retval = -EAGAIN;
-		goto migrate_end;
-	}
+	if (smp_processor_id() != cpu)
+		return -EAGAIN;
 
 	if (state == data->acpi_data.state) {
 		if (unlikely(data->resume)) {
@@ -159,8 +146,7 @@ processor_set_freq (
 			data->resume = 0;
 		} else {
 			pr_debug("Already at target state (P%d)\n", state);
-			retval = 0;
-			goto migrate_end;
+			return 0;
 		}
 	}
 
@@ -171,7 +157,6 @@ processor_set_freq (
 	 * First we write the target state's 'control' value to the
 	 * control_register.
 	 */
-
 	value = (u32) data->acpi_data.states[state].control;
 
 	pr_debug("Transitioning to state: 0x%08x\n", value);
@@ -179,17 +164,11 @@ processor_set_freq (
 	ret = processor_set_pstate(value);
 	if (ret) {
 		pr_warn("Transition failed with error %d\n", ret);
-		retval = -ENODEV;
-		goto migrate_end;
+		return -ENODEV;
 	}
 
 	data->acpi_data.state = state;
-
-	retval = 0;
-
-migrate_end:
-	set_cpus_allowed_ptr(current, &saved_mask);
-	return (retval);
+	return 0;
 }
 
 
@@ -197,11 +176,13 @@ static unsigned int
 acpi_cpufreq_get (
 	unsigned int		cpu)
 {
-	struct cpufreq_acpi_io *data = acpi_io_data[cpu];
+	struct cpufreq_acpi_req req;
+	long ret;
 
-	pr_debug("acpi_cpufreq_get\n");
+	req.cpu = cpu;
+	ret = work_on_cpu(cpu, processor_get_freq, &req);
 
-	return processor_get_freq(data, cpu);
+	return ret > 0 ? (unsigned int) ret : 0;
 }
 
 
@@ -210,7 +191,12 @@ acpi_cpufreq_target (
 	struct cpufreq_policy   *policy,
 	unsigned int index)
 {
-	return processor_set_freq(acpi_io_data[policy->cpu], policy, index);
+	struct cpufreq_acpi_req req;
+
+	req.cpu = policy->cpu;
+	req.state = index;
+
+	return work_on_cpu(req.cpu, processor_set_freq, &req);
 }
 
 static int

[toc] | [prev] | [next] | [standalone]

#1622675

FromViresh Kumar <viresh.kumar@linaro.org>
Date2017-04-13 04:50 +0200
Message-ID<tvCZz-4V5-7@gated-at.bofh.it>
In reply to#1622517
On 12-04-17, 22:07, Thomas Gleixner wrote:
> The get() and target() callbacks must run on the affected cpu. This is
> achieved by temporarily setting the affinity of the calling thread to the
> requested CPU and reset it to the original affinity afterwards.
> 
> That's racy vs. concurrent affinity settings for that thread resulting in
> code executing on the wrong CPU and overwriting the new affinity setting.
> 
> Replace it by work_on_cpu(). All call pathes which invoke the callbacks are
> already protected against CPU hotplug.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> Cc: "Rafael J. Wysocki" <rjw@rjwysocki.net>
> Cc: Viresh Kumar <viresh.kumar@linaro.org>
> Cc: Tony Luck <tony.luck@intel.com>
> Cc: Fenghua Yu <fenghua.yu@intel.com>
> Cc: linux-pm@vger.kernel.org
> ---
>  drivers/cpufreq/ia64-acpi-cpufreq.c |   91 +++++++++++++++---------------------
>  1 file changed, 38 insertions(+), 53 deletions(-)

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

-- 
viresh

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web