Skip to contentSkip to conversation

cpumask: avoid playing with cpus_allowed in powernow-k8.c

PATCH In review Rusty Russell 3 participants 7 messages Jun 11, 2009 – Jun 15, 2009 x86

lore
Share & tools
lore
#0 lore raw

PATCH 3/6 cpumask: avoid playing with cpus_allowed in powernow-k8.c

From: Rusty Russell <[email protected]>

It's generally a very bad idea to mug some process's cpumask: it could legitimately and reasonably be changed by root, which could break us (if done before our code) or them (if we restore the wrong value).

I did not replace powernowk8_target; it needs fixing, but it grabs a mutex (so no smp_call_function_single here) but Mark points out it can be called multiple times per second, so work_on_cpu is too heavy.

To: [email protected]
To: [email protected]
Cc: [email protected]

---
 arch/x86/kernel/cpu/cpufreq/powernow-k8.c |  156 +++++++++++++++---------------
 1 file changed, 82 insertions(+), 74 deletions(-)
arch/x86/kernel/cpu/cpufreq/powernow-k8.c +60 -58
@@ -510,41 +510,34 @@ static int core_voltage_post_transition(
 	return 0;
 }
 
-static int check_supported_cpu(unsigned int cpu)
+static void check_supported_cpu(void *_rc)
 {
-	cpumask_t oldmask;
 	u32 eax, ebx, ecx, edx;
-	unsigned int rc = 0;
+	int *rc = _rc;
 
-	oldmask = current->cpus_allowed;
-	set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
-
-	if (smp_processor_id() != cpu) {
-		printk(KERN_ERR PFX "limiting to cpu %u failed\n", cpu);
-		goto out;
-	}
+	*rc = -ENODEV;
 
 	if (current_cpu_data.x86_vendor != X86_VENDOR_AMD)
-		goto out;
+		return;
 
 	eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
 	if (((eax & CPUID_XFAM) != CPUID_XFAM_K8) &&
 	    ((eax & CPUID_XFAM) < CPUID_XFAM_10H))
-		goto out;
+		return;
 
 	if ((eax & CPUID_XFAM) == CPUID_XFAM_K8) {
 		if (((eax & CPUID_USE_XFAM_XMOD) != CPUID_USE_XFAM_XMOD) ||
 		    ((eax & CPUID_XMOD) > CPUID_XMOD_REV_MASK)) {
 			printk(KERN_INFO PFX
 				"Processor cpuid %x not supported\n", eax);
-			goto out;
+			return;
 		}
 
 		eax = cpuid_eax(CPUID_GET_MAX_CAPABILITIES);
 		if (eax < CPUID_FREQ_VOLT_CAPABILITIES) {
 			printk(KERN_INFO PFX
 			       "No frequency change capabilities detected\n");
-			goto out;
+			return;
 		}
 
 		cpuid(CPUID_FREQ_VOLT_CAPABILITIES, &eax, &ebx, &ecx, &edx);
@@ -552,21 +545,17 @@ static int check_supported_cpu(unsigned 
 			!= P_STATE_TRANSITION_CAPABLE) {
 			printk(KERN_INFO PFX
 				"Power state transitions not supported\n");
-			goto out;
+			return;
 		}
 	} else { /* must be a HW Pstate capable processor */
 		cpuid(CPUID_FREQ_VOLT_CAPABILITIES, &eax, &ebx, &ecx, &edx);
 		if ((edx & USE_HW_PSTATE) == USE_HW_PSTATE)
 			cpu_family = CPU_HW_PSTATE;
 		else
-			goto out;
+			return;
 	}
 
-	rc = 1;
-
-out:
-	set_cpus_allowed_ptr(current, &oldmask);
-	return rc;
+	*rc = 0;
 }
 
 static int check_pst_table(struct powernow_k8_data *data, struct pst_s *pst,
@@ -1235,6 +1230,32 @@ static int powernowk8_verify(struct cpuf
 	return cpufreq_frequency_table_verify(pol, data->powernow_table);
 }
 
+struct init_on_cpu {
+	struct powernow_k8_data *data;
+	int rc;
+};
+
+static void __cpuinit powernowk8_cpu_init_on_cpu(void *_init_on_cpu)
+{
+	struct init_on_cpu *init_on_cpu = _init_on_cpu;
+
+	if (pending_bit_stuck()) {
+		printk(KERN_ERR PFX "failing init, change pending bit set\n");
+		init_on_cpu->rc = -ENODEV;
+		return;
+	}
+
+	if (query_current_values_with_pending_wait(init_on_cpu->data)) {
+		init_on_cpu->rc = -ENODEV;
+		return;
+	}
+
+	if (cpu_family == CPU_OPTERON)
+		fidvid_msr_init();
+
+	init_on_cpu->rc = 0;
+}
+
 static const char ACPI_PSS_BIOS_BUG_MSG[] =
 	KERN_ERR FW_BUG PFX "No compatible ACPI _PSS objects found.\n"
 	KERN_ERR FW_BUG PFX "Try again with latest BIOS.\n";
@@ -1243,13 +1264,14 @@ static const char ACPI_PSS_BIOS_BUG_MSG[
 static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
 {
 	struct powernow_k8_data *data;
-	cpumask_t oldmask;
+	struct init_on_cpu init_on_cpu;
 	int rc;
 
 	if (!cpu_online(pol->cpu))
 		return -ENODEV;
 
-	if (!check_supported_cpu(pol->cpu))
+	smp_call_function_single(pol->cpu, check_supported_cpu, &rc, 1);
+	if (rc)
 		return -ENODEV;
 
 	data = kzalloc(sizeof(struct powernow_k8_data), GFP_KERNEL);
@@ -1289,27 +1311,12 @@ static int __cpuinit powernowk8_cpu_init
 		pol->cpuinfo.transition_latency = get_transition_latency(data);
 
 	/* only run on specific CPU from here on */
-	oldmask = current->cpus_allowed;
-	set_cpus_allowed_ptr(current, &cpumask_of_cpu(pol->cpu));
-
-	if (smp_processor_id() != pol->cpu) {
-		printk(KERN_ERR PFX "limiting to cpu %u failed\n", pol->cpu);
-		goto err_out_unmask;
-	}
-
-	if (pending_bit_stuck()) {
-		printk(KERN_ERR PFX "failing init, change pending bit set\n");
-		goto err_out_unmask;
-	}
-
-	if (query_current_values_with_pending_wait(data))
-		goto err_out_unmask;
-
-	if (cpu_family == CPU_OPTERON)
-		fidvid_msr_init();
-
-	/* run on any CPU again */
-	set_cpus_allowed_ptr(current, &oldmask);
+	init_on_cpu.data = data;
+	smp_call_function_single(data->cpu, powernowk8_cpu_init_on_cpu,
+				 &init_on_cpu, 1);
+	rc = init_on_cpu.rc;
+	if (rc != 0)
+		goto err_out_exit_acpi;
 
 	if (cpu_family == CPU_HW_PSTATE)
 		cpumask_copy(pol->cpus, cpumask_of(pol->cpu));
@@ -1346,8 +1353,7 @@ static int __cpuinit powernowk8_cpu_init
 
 	return 0;
 
-err_out_unmask:
-	set_cpus_allowed_ptr(current, &oldmask);
+err_out_exit_acpi:
 	powernow_k8_cpu_exit_acpi(data);
 
 err_out:
@@ -1372,12 +1378,20 @@ static int __devexit powernowk8_cpu_exit
 	return 0;
 }
 
+static void query_values_on_cpu(void *_err)
+{
+	int *err = _err;
+	struct powernow_k8_data *data = __get_cpu_var(powernow_data);
+
+	*err = query_current_values_with_pending_wait(data);
+}
+
 static unsigned int powernowk8_get(unsigned int cpu)
 {
 	struct powernow_k8_data *data;
-	cpumask_t oldmask = current->cpus_allowed;
 	unsigned int khz = 0;
 	unsigned int first;
+	int err;
 
 	first = cpumask_first(cpu_core_mask(cpu));
 	data = per_cpu(powernow_data, first);
@@ -1385,15 +1399,8 @@ static unsigned int powernowk8_get(unsig
 	if (!data)
 		return -EINVAL;
 
-	set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
-	if (smp_processor_id() != cpu) {
-		printk(KERN_ERR PFX
-			"limiting to CPU %d failed in powernowk8_get\n", cpu);
-		set_cpus_allowed_ptr(current, &oldmask);
-		return 0;
-	}
-
-	if (query_current_values_with_pending_wait(data))
+	smp_call_function_single(first, query_values_on_cpu, &err, true);
+	if (err)
 		goto out;
 
 	if (cpu_family == CPU_HW_PSTATE)
@@ -1404,7 +1411,6 @@ static unsigned int powernowk8_get(unsig
 
 
 out:
-	set_cpus_allowed_ptr(current, &oldmask);
 	return khz;
 }
 
@@ -1430,7 +1436,9 @@ static int __cpuinit powernowk8_init(voi
 	unsigned int i, supported_cpus = 0;
 
 	for_each_online_cpu(i) {
-		if (check_supported_cpu(i))
+		int rc;
+		smp_call_function_single(i, check_supported_cpu, &rc, 1);
+		if (rc == 0)
 			supported_cpus++;
 	}
 
#1 lore raw
Quoted 9 lines · Rusty Russell

From: Rusty Russell <[email protected]>

It's generally a very bad idea to mug some process's cpumask: it could legitimately and reasonably be changed by root, which could break us (if done before our code) or them (if we restore the wrong value).

I did not replace powernowk8_target; it needs fixing, but it grabs a mutex (so no smp_call_function_single here) but Mark points out it can be called multiple times per second, so work_on_cpu is too heavy.

This one clashes with some diffs I merged earlier in cpufreq.git. Can you rebase on top of that please?
(The other two were fine).

Otherwise, they all look ok to me. As long as they survive testing I see no reason not to push them for .31.

Thanks,

Dave

#2 lore raw

PATCH 1/1 cpumask: avoid playing with cpus_allowed in powernow-k8.c

cpumask: avoid playing with cpus_allowed in powernow-k8.c

It's generally a very bad idea to mug some process's cpumask: it could legitimately and reasonably be changed by root, which could break us (if done before our code) or them (if we restore the wrong value).

I did not replace powernowk8_target; it needs fixing, but it grabs a mutex (so no smp_call_function_single here) but Mark points out it can be called multiple times per second, so work_on_cpu is too heavy.

To: [email protected]
To: [email protected]
Cc: [email protected]

arch/x86/kernel/cpu/cpufreq/powernow-k8.c +60 -58
index 20c7b99..1f55547 100644
@@ -508,41 +508,34 @@ static int core_voltage_post_transition(struct powernow_k8_data *data,
 	return 0;
 }
 
-static int check_supported_cpu(unsigned int cpu)
+static void check_supported_cpu(void *_rc)
 {
-	cpumask_t oldmask;
 	u32 eax, ebx, ecx, edx;
-	unsigned int rc = 0;
-
-	oldmask = current->cpus_allowed;
-	set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
+	int *rc = _rc;
 
-	if (smp_processor_id() != cpu) {
-		printk(KERN_ERR PFX "limiting to cpu %u failed\n", cpu);
-		goto out;
-	}
+	*rc = -ENODEV;
 
 	if (current_cpu_data.x86_vendor != X86_VENDOR_AMD)
-		goto out;
+		return;
 
 	eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
 	if (((eax & CPUID_XFAM) != CPUID_XFAM_K8) &&
 	    ((eax & CPUID_XFAM) < CPUID_XFAM_10H))
-		goto out;
+		return;
 
 	if ((eax & CPUID_XFAM) == CPUID_XFAM_K8) {
 		if (((eax & CPUID_USE_XFAM_XMOD) != CPUID_USE_XFAM_XMOD) ||
 		    ((eax & CPUID_XMOD) > CPUID_XMOD_REV_MASK)) {
 			printk(KERN_INFO PFX
 				"Processor cpuid %x not supported\n", eax);
-			goto out;
+			return;
 		}
 
 		eax = cpuid_eax(CPUID_GET_MAX_CAPABILITIES);
 		if (eax < CPUID_FREQ_VOLT_CAPABILITIES) {
 			printk(KERN_INFO PFX
 			       "No frequency change capabilities detected\n");
-			goto out;
+			return;
 		}
 
 		cpuid(CPUID_FREQ_VOLT_CAPABILITIES, &eax, &ebx, &ecx, &edx);
@@ -550,21 +543,17 @@ static int check_supported_cpu(unsigned int cpu)
 			!= P_STATE_TRANSITION_CAPABLE) {
 			printk(KERN_INFO PFX
 				"Power state transitions not supported\n");
-			goto out;
+			return;
 		}
 	} else { /* must be a HW Pstate capable processor */
 		cpuid(CPUID_FREQ_VOLT_CAPABILITIES, &eax, &ebx, &ecx, &edx);
 		if ((edx & USE_HW_PSTATE) == USE_HW_PSTATE)
 			cpu_family = CPU_HW_PSTATE;
 		else
-			goto out;
+			return;
 	}
 
-	rc = 1;
-
-out:
-	set_cpus_allowed_ptr(current, &oldmask);
-	return rc;
+	*rc = 0;
 }
 
 static int check_pst_table(struct powernow_k8_data *data, struct pst_s *pst,
@@ -1247,6 +1236,32 @@ static int powernowk8_verify(struct cpufreq_policy *pol)
 	return cpufreq_frequency_table_verify(pol, data->powernow_table);
 }
 
+struct init_on_cpu {
+	struct powernow_k8_data *data;
+	int rc;
+};
+
+static void __cpuinit powernowk8_cpu_init_on_cpu(void *_init_on_cpu)
+{
+	struct init_on_cpu *init_on_cpu = _init_on_cpu;
+
+	if (pending_bit_stuck()) {
+		printk(KERN_ERR PFX "failing init, change pending bit set\n");
+		init_on_cpu->rc = -ENODEV;
+		return;
+	}
+
+	if (query_current_values_with_pending_wait(init_on_cpu->data)) {
+		init_on_cpu->rc = -ENODEV;
+		return;
+	}
+
+	if (cpu_family == CPU_OPTERON)
+		fidvid_msr_init();
+
+	init_on_cpu->rc = 0;
+}
+
 /* per CPU init entry point to the driver */
 static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
 {
@@ -1254,13 +1269,14 @@ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
 		KERN_ERR FW_BUG PFX "No compatible ACPI _PSS objects found.\n"
 		KERN_ERR FW_BUG PFX "Try again with latest BIOS.\n";
 	struct powernow_k8_data *data;
-	cpumask_t oldmask;
+	struct init_on_cpu init_on_cpu;
 	int rc;
 
 	if (!cpu_online(pol->cpu))
 		return -ENODEV;
 
-	if (!check_supported_cpu(pol->cpu))
+	smp_call_function_single(pol->cpu, check_supported_cpu, &rc, 1);
+	if (rc)
 		return -ENODEV;
 
 	data = kzalloc(sizeof(struct powernow_k8_data), GFP_KERNEL);
@@ -1300,27 +1316,12 @@ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
 		pol->cpuinfo.transition_latency = get_transition_latency(data);
 
 	/* only run on specific CPU from here on */
-	oldmask = current->cpus_allowed;
-	set_cpus_allowed_ptr(current, &cpumask_of_cpu(pol->cpu));
-
-	if (smp_processor_id() != pol->cpu) {
-		printk(KERN_ERR PFX "limiting to cpu %u failed\n", pol->cpu);
-		goto err_out_unmask;
-	}
-
-	if (pending_bit_stuck()) {
-		printk(KERN_ERR PFX "failing init, change pending bit set\n");
-		goto err_out_unmask;
-	}
-
-	if (query_current_values_with_pending_wait(data))
-		goto err_out_unmask;
-
-	if (cpu_family == CPU_OPTERON)
-		fidvid_msr_init();
-
-	/* run on any CPU again */
-	set_cpus_allowed_ptr(current, &oldmask);
+	init_on_cpu.data = data;
+	smp_call_function_single(data->cpu, powernowk8_cpu_init_on_cpu,
+				 &init_on_cpu, 1);
+	rc = init_on_cpu.rc;
+	if (rc != 0)
+		goto err_out_exit_acpi;
 
 	if (cpu_family == CPU_HW_PSTATE)
 		cpumask_copy(pol->cpus, cpumask_of(pol->cpu));
@@ -1357,8 +1358,7 @@ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
 
 	return 0;
 
-err_out_unmask:
-	set_cpus_allowed_ptr(current, &oldmask);
+err_out_exit_acpi:
 	powernow_k8_cpu_exit_acpi(data);
 
 err_out:
@@ -1383,24 +1383,25 @@ static int __devexit powernowk8_cpu_exit(struct cpufreq_policy *pol)
 	return 0;
 }
 
+static void query_values_on_cpu(void *_err)
+{
+	int *err = _err;
+	struct powernow_k8_data *data = __get_cpu_var(powernow_data);
+
+	*err = query_current_values_with_pending_wait(data);
+}
+
 static unsigned int powernowk8_get(unsigned int cpu)
 {
 	struct powernow_k8_data *data = per_cpu(powernow_data, cpu);
-	cpumask_t oldmask = current->cpus_allowed;
 	unsigned int khz = 0;
+	int err;
 
 	if (!data)
 		return -EINVAL;
 
-	set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu));
-	if (smp_processor_id() != cpu) {
-		printk(KERN_ERR PFX
-			"limiting to CPU %d failed in powernowk8_get\n", cpu);
-		set_cpus_allowed_ptr(current, &oldmask);
-		return 0;
-	}
-
-	if (query_current_values_with_pending_wait(data))
+	smp_call_function_single(cpu, query_values_on_cpu, &err, true);
+	if (err)
 		goto out;
 
 	if (cpu_family == CPU_HW_PSTATE)
@@ -1411,7 +1412,6 @@ static unsigned int powernowk8_get(unsigned int cpu)
 
 
 out:
-	set_cpus_allowed_ptr(current, &oldmask);
 	return khz;
 }
 
@@ -1437,7 +1437,9 @@ static int __cpuinit powernowk8_init(void)
 	unsigned int i, supported_cpus = 0;
 
 	for_each_online_cpu(i) {
-		if (check_supported_cpu(i))
+		int rc;
+		smp_call_function_single(i, check_supported_cpu, &rc, 1);
+		if (rc == 0)
 			supported_cpus++;
 	}
 
Acked-by Tested-by on v1 1/1
#3 lore raw
Quoted 6 lines

cpumask: avoid playing with cpus_allowed in powernow-k8.c

It's generally a very bad idea to mug some process's cpumask: it could legitimately and reasonably be changed by root, which could break us (if done before our code) or them (if we restore the wrong value).

Quoted 8 lines

I did not replace powernowk8_target; it needs fixing, but it grabs a mutex (so no smp_call_function_single here) but Mark points out it can be called multiple times per second, so work_on_cpu is too heavy.

To: [email protected]
To: [email protected]
Cc: [email protected]

I ran this over the weekend on a 24 core box without any issues. Please commit.

#4 lore raw
In reply to Dave Jones
Quoted 3 lines · Dave Jones

This one clashes with some diffs I merged earlier in cpufreq.git. Can you rebase on top of that please?
(The other two were fine).

OK, done. Plus I have another (trivial) powernow-k8 patch (which still applies with offset).

Thanks!
Rusty.

#5 lore raw

PATCH 1/1 cpumask: new cpumask operators for arch/x86/kernel/cpu/cpufreq/powernow-k8.c

In reply to Dave Jones

Remove all old-style cpumask operators, and cpumask_t.

Also: get rid of the unused define_siblings function.

---
 arch/x86/kernel/cpu/cpufreq/powernow-k8.c |    8 ++++----
 arch/x86/kernel/cpu/cpufreq/powernow-k8.h |   11 -----------
 2 files changed, 4 insertions(+), 15 deletions(-)
arch/x86/kernel/cpu/cpufreq/powernow-k8.c +4 -4
@@ -1060,7 +1060,7 @@ static int transition_frequency_fidvid(s
 	freqs.old = find_khz_freq_from_fid(data->currfid);
 	freqs.new = find_khz_freq_from_fid(fid);
 
-	for_each_cpu_mask_nr(i, *(data->available_cores)) {
+	for_each_cpu(i, data->available_cores) {
 		freqs.cpu = i;
 		cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 	}
@@ -1068,7 +1068,7 @@ static int transition_frequency_fidvid(s
 	res = transition_fid_vid(data, fid, vid);
 	freqs.new = find_khz_freq_from_fid(data->currfid);
 
-	for_each_cpu_mask_nr(i, *(data->available_cores)) {
+	for_each_cpu(i, data->available_cores) {
 		freqs.cpu = i;
 		cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 	}
@@ -1093,7 +1093,7 @@ static int transition_frequency_pstate(s
 			data->currpstate);
 	freqs.new = find_khz_freq_from_pstate(data->powernow_table, pstate);
 
-	for_each_cpu_mask_nr(i, *(data->available_cores)) {
+	for_each_cpu(i, data->available_cores) {
 		freqs.cpu = i;
 		cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 	}
@@ -1101,7 +1101,7 @@ static int transition_frequency_pstate(s
 	res = transition_pstate(data, pstate);
 	freqs.new = find_khz_freq_from_pstate(data->powernow_table, pstate);
 
-	for_each_cpu_mask_nr(i, *(data->available_cores)) {
+	for_each_cpu(i, data->available_cores) {
 		freqs.cpu = i;
 		cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 	}
arch/x86/kernel/cpu/cpufreq/powernow-k8.h +0 -11
@@ -223,14 +223,3 @@ static void powernow_k8_acpi_pst_values(
 
 static int fill_powernow_table_pstate(struct powernow_k8_data *data, struct cpufreq_frequency_table *powernow_table);
 static int fill_powernow_table_fidvid(struct powernow_k8_data *data, struct cpufreq_frequency_table *powernow_table);
-
-#ifdef CONFIG_SMP
-static inline void define_siblings(int cpu, cpumask_t cpu_sharedcore_mask[])
-{
-}
-#else
-static inline void define_siblings(int cpu, cpumask_t cpu_sharedcore_mask[])
-{
-	cpu_set(0, cpu_sharedcore_mask[0]);
-}
-#endif
Acked-by Tested-by on v1 1/1
#6 lore raw

RE: [PATCH] cpumask: new cpumask operators for arch/x86/kernel/cpu/cpufreq/powernow-k8.c

Quoted 5 lines

Remove all old-style cpumask operators, and cpumask_t.

Also: get rid of the unused define_siblings function.

I ran this over the weekend on a 24 core box without any issues. Please commit.