Skip to contentSkip to conversation

x86,cpu-hotplug: assign same CPU number to readded CPU

PATCH Superseded Yasuaki Ishimatsu 2 participants 4 messages Jul 17, 2014 – Jul 23, 2014 x86

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 x86,cpu-hotplug: assign same CPU number to readded CPU

llc_shared_map is not cleared even if CPU is offline or hot removed. So when hot-plugging CPU and assigning new CPU number to hot-added CPU, the mask has wrong value. The mask is used by CSF schduler to create sched_domain. So it breaks CFS scheduler.

Here is a example on my system.
My system has 4 sockets and each socket has 15 cores and HT is enabled. In this case, each core of sockes is numbered as follows:

         | CPU#
Socket#0 | 0-14 , 60-74
Socket#1 | 15-29, 75-89
Socket#2 | 30-44, 90-104
Socket#3 | 45-59, 105-119

Then llc_shared_mask of CPU#30 has 0x3fff80000001fffc0000000. It means that last level cache of Socket#2 is shared with CPU#30-44 and 90-104.

When hot-removing socket#2 and #3, each core of sockets is numbered as follows:

         | CPU#
Socket#0 | 0-14 , 60-74
Socket#1 | 15-29, 75-89

But llc_shared_mask is not cleared. So llc_shared_mask of CPU#30 remains having 0x3fff80000001fffc0000000.

After that, when hot-adding socket#2 and #3, each core of sockets is numbered as follows:

         | CPU#
Socket#0 | 0-14 , 60-74
Socket#1 | 15-29, 75-89
Socket#2 | 30-59
Socket#3 | 90-119

Then llc_shared_mask of CPU#30 becomes 0x3fff8000fffffffc0000000. It means that last level cache of Socket#2 is shared with CPU#30-59 and 90-104. So the mask has wrong value.

At first, I cleared hot-removed CPU number's bit from llc_shared_map when hot removing CPU. But Borislav suggested that the problem will disappear if readded CPU is assigned same CPU number. And llc_shared_map must not be changed.

So the patch assigns same CPU number to readded CPU by linking CPU number to APIC ID. And by the patch, the problem disappers.

---
v2: change cpuid to cpunum
v3: fix Borislav's email address of Suggested-by
fix typo (ACPI ID to APIC ID)

---
 arch/x86/kernel/apic/apic.c | 33 ++++++++++++++++++++++++++++++++-
 1 file changed, 32 insertions(+), 1 deletion(-)
arch/x86/kernel/apic/apic.c +32 -1
index ad28db7..5dc3e50 100644
@@ -220,6 +220,23 @@ static void apic_pm_activate(void);
 static unsigned long apic_phys;
 
 /*
+ * Bind APIC ID to Logical CPU number
+ * Logical CPU number to APIC ID does not change by this array
+ * even if CPU is hotplugged. So don't clear the array even if
+ * CPU is hot-removed
+ */
+static int apicid_to_cpunum[MAX_LOCAL_APIC] = {
+	[0 ... MAX_LOCAL_APIC-1] = -1,
+};
+
+/*
+ * Represent Logical CPU number bound to APIC ID
+ * Don't clear a bit even if CPU is hot-removed
+ */
+static DECLARE_BITMAP(cpu_used_bits, CONFIG_NR_CPUS);
+static struct cpumask *const cpu_used_mask = to_cpumask(cpu_used_bits);
+
+/*
  * Get the LAPIC version
  */
 static inline int lapic_get_version(void)
@@ -2122,6 +2139,17 @@ void disconnect_bsp_APIC(int virt_wire_setup)
 	apic_write(APIC_LVT1, value);
 }
 
+static int get_cpunum(int apicid)
+{
+	int cpu;
+
+	cpu = apicid_to_cpunum[apicid];
+	if (cpu < 0)
+		cpu = cpumask_next_zero(-1, cpu_used_mask);
+
+	return cpu;
+}
+
 int generic_processor_info(int apicid, int version)
 {
 	int cpu, max = nr_cpu_ids;
@@ -2199,7 +2227,9 @@ int generic_processor_info(int apicid, int version)
 		 */
 		cpu = 0;
 	} else
-		cpu = cpumask_next_zero(-1, cpu_present_mask);
+		cpu = get_cpunum(apicid);
+
+	apicid_to_cpunum[apicid] = cpu;
 
 	/*
 	 * Validate version
@@ -2228,6 +2258,7 @@ int generic_processor_info(int apicid, int version)
 	early_per_cpu(x86_cpu_to_logical_apicid, cpu) =
 		apic->x86_32_early_logical_apicid(cpu);
 #endif
+	cpumask_set_cpu(cpu, cpu_used_mask);
 	set_cpu_possible(cpu, true);
 	set_cpu_present(cpu, true);
 
Reviewed-by on v3 1/1
#1 lore raw
Quoted 74 lines · Yasuaki Ishimatsu

llc_shared_map is not cleared even if CPU is offline or hot removed. So when hot-plugging CPU and assigning new CPU number to hot-added CPU, the mask has wrong value. The mask is used by CSF schduler to create sched_domain. So it breaks CFS scheduler.

Here is a example on my system.
My system has 4 sockets and each socket has 15 cores and HT is enabled. In this case, each core of sockes is numbered as follows:

          | CPU#
Socket#0 | 0-14 , 60-74
Socket#1 | 15-29, 75-89
Socket#2 | 30-44, 90-104
Socket#3 | 45-59, 105-119

Then llc_shared_mask of CPU#30 has 0x3fff80000001fffc0000000. It means that last level cache of Socket#2 is shared with CPU#30-44 and 90-104.

When hot-removing socket#2 and #3, each core of sockets is numbered as follows:

          | CPU#

Socket#0 | 0-14 , 60-74
Socket#1 | 15-29, 75-89

But llc_shared_mask is not cleared. So llc_shared_mask of CPU#30 remains having 0x3fff80000001fffc0000000.

After that, when hot-adding socket#2 and #3, each core of sockets is numbered as follows:

          | CPU#
Socket#0 | 0-14 , 60-74
Socket#1 | 15-29, 75-89
Socket#2 | 30-59
Socket#3 | 90-119

Then llc_shared_mask of CPU#30 becomes 0x3fff8000fffffffc0000000. It means that last level cache of Socket#2 is shared with CPU#30-59 and 90-104. So the mask has wrong value.

At first, I cleared hot-removed CPU number's bit from llc_shared_map when hot removing CPU. But Borislav suggested that the problem will disappear if readded CPU is assigned same CPU number. And llc_shared_map must not be changed.

So the patch assigns same CPU number to readded CPU by linking CPU number to APIC ID. And by the patch, the problem disappers.

---
v2: change cpuid to cpunum
v3: fix Borislav's email address of Suggested-by
fix typo (ACPI ID to APIC ID)

---
  arch/x86/kernel/apic/apic.c | 33 ++++++++++++++++++++++++++++++++-
  1 file changed, 32 insertions(+), 1 deletion(-)
arch/x86/kernel/apic/apic.c +5 -0
index ad28db7..5dc3e50 100644
@@ -220,6 +220,23 @@ static void apic_pm_activate(void);
  static unsigned long apic_phys;
 
  /*
+ * Bind APIC ID to Logical CPU number
+ * Logical CPU number to APIC ID does not change by this array
+ * even if CPU is hotplugged. So don't clear the array even if
+ * CPU is hot-removed
+ */

arch/x86/kernel/apic/apic.c static void apic_pm_activate(void); · line 237

Quoted 10 lines
arch/x86/kernel/apic/apic.c +10 -0
+static int apicid_to_cpunum[MAX_LOCAL_APIC] = {
+	[0 ... MAX_LOCAL_APIC-1] = -1,
+};
+
+/*
+ * Represent Logical CPU number bound to APIC ID
+ * Don't clear a bit even if CPU is hot-removed
+ */
+static DECLARE_BITMAP(cpu_used_bits, CONFIG_NR_CPUS);
+static struct cpumask *const cpu_used_mask = to_cpumask(cpu_used_bits);

I think the name of cpu_used_mask is confusing since the term "used" sounds some state of CPU, which is not the case here. How about cpu_number_mask? This is more specific and avoids such confusion.

Otherwise, the change looks good.

Thanks,
-Toshi

arch/x86/kernel/apic/apic.c static void apic_pm_activate(void); · line 241

Quoted 4 lines
arch/x86/kernel/apic/apic.c +2 -0
+
+/*
   * Get the LAPIC version
   */

^^
ps. I had a problem in applying this patch since there was an extra blank character at the begging of unchanged lines. But this might be an issue in my mailer.

Reviewed-by on v3 1/1
#3 lore raw
In reply to Yasuaki Ishimatsu
Quoted 74 lines · Yasuaki Ishimatsu

llc_shared_map is not cleared even if CPU is offline or hot removed. So when hot-plugging CPU and assigning new CPU number to hot-added CPU, the mask has wrong value. The mask is used by CSF schduler to create sched_domain. So it breaks CFS scheduler.

Here is a example on my system.
My system has 4 sockets and each socket has 15 cores and HT is enabled. In this case, each core of sockes is numbered as follows:

          | CPU#
Socket#0 | 0-14 , 60-74
Socket#1 | 15-29, 75-89
Socket#2 | 30-44, 90-104
Socket#3 | 45-59, 105-119

Then llc_shared_mask of CPU#30 has 0x3fff80000001fffc0000000. It means that last level cache of Socket#2 is shared with CPU#30-44 and 90-104.

When hot-removing socket#2 and #3, each core of sockets is numbered as follows:

          | CPU#

Socket#0 | 0-14 , 60-74
Socket#1 | 15-29, 75-89

But llc_shared_mask is not cleared. So llc_shared_mask of CPU#30 remains having 0x3fff80000001fffc0000000.

After that, when hot-adding socket#2 and #3, each core of sockets is numbered as follows:

          | CPU#
Socket#0 | 0-14 , 60-74
Socket#1 | 15-29, 75-89
Socket#2 | 30-59
Socket#3 | 90-119

Then llc_shared_mask of CPU#30 becomes 0x3fff8000fffffffc0000000. It means that last level cache of Socket#2 is shared with CPU#30-59 and 90-104. So the mask has wrong value.

At first, I cleared hot-removed CPU number's bit from llc_shared_map when hot removing CPU. But Borislav suggested that the problem will disappear if readded CPU is assigned same CPU number. And llc_shared_map must not be changed.

So the patch assigns same CPU number to readded CPU by linking CPU number to APIC ID. And by the patch, the problem disappers.

---
v2: change cpuid to cpunum
v3: fix Borislav's email address of Suggested-by
fix typo (ACPI ID to APIC ID)

---
  arch/x86/kernel/apic/apic.c | 33 ++++++++++++++++++++++++++++++++-
  1 file changed, 32 insertions(+), 1 deletion(-)
arch/x86/kernel/apic/apic.c +5 -0
index ad28db7..5dc3e50 100644
@@ -220,6 +220,23 @@ static void apic_pm_activate(void);
  static unsigned long apic_phys;
 
  /*
+ * Bind APIC ID to Logical CPU number
+ * Logical CPU number to APIC ID does not change by this array
+ * even if CPU is hotplugged. So don't clear the array even if
+ * CPU is hot-removed
+ */

arch/x86/kernel/apic/apic.c static void apic_pm_activate(void); · line 237

Quoted 10 lines
arch/x86/kernel/apic/apic.c +10 -0
+static int apicid_to_cpunum[MAX_LOCAL_APIC] = {
+	[0 ... MAX_LOCAL_APIC-1] = -1,
+};
+
+/*
+ * Represent Logical CPU number bound to APIC ID
+ * Don't clear a bit even if CPU is hot-removed
+ */
+static DECLARE_BITMAP(cpu_used_bits, CONFIG_NR_CPUS);
+static struct cpumask *const cpu_used_mask = to_cpumask(cpu_used_bits);

I think the name of cpu_used_mask is confusing since the term "used" sounds representing some state of CPU, which is not the case here. How about cpu_number_mask? This would be more specific, and avoid such confusion.

Otherwise, the change looks good to me.

Thanks,
-Toshi

arch/x86/kernel/apic/apic.c static void apic_pm_activate(void); · line 241

Quoted 4 lines
arch/x86/kernel/apic/apic.c +2 -0
+
+/*
   * Get the LAPIC version
   */

^^
ps. I had a problem in applying this patch since there was an extra blank character at the begging of unchanged lines. But this might be an issue in my mailer.

#2 lore raw
In reply to Toshi Kani
Quoted 81 lines · Toshi Kani
Quoted 80 lines · Yasuaki Ishimatsu

llc_shared_map is not cleared even if CPU is offline or hot removed. So when hot-plugging CPU and assigning new CPU number to hot-added CPU, the mask has wrong value. The mask is used by CSF schduler to create sched_domain. So it breaks CFS scheduler.

Here is a example on my system.
My system has 4 sockets and each socket has 15 cores and HT is enabled. In this case, each core of sockes is numbered as follows:

           | CPU#
Socket#0 | 0-14 , 60-74
Socket#1 | 15-29, 75-89
Socket#2 | 30-44, 90-104
Socket#3 | 45-59, 105-119

Then llc_shared_mask of CPU#30 has 0x3fff80000001fffc0000000. It means that last level cache of Socket#2 is shared with CPU#30-44 and 90-104.

When hot-removing socket#2 and #3, each core of sockets is numbered as follows:

           | CPU#

Socket#0 | 0-14 , 60-74
Socket#1 | 15-29, 75-89

But llc_shared_mask is not cleared. So llc_shared_mask of CPU#30 remains having 0x3fff80000001fffc0000000.

After that, when hot-adding socket#2 and #3, each core of sockets is numbered as follows:

           | CPU#
Socket#0 | 0-14 , 60-74
Socket#1 | 15-29, 75-89
Socket#2 | 30-59
Socket#3 | 90-119

Then llc_shared_mask of CPU#30 becomes 0x3fff8000fffffffc0000000. It means that last level cache of Socket#2 is shared with CPU#30-59 and 90-104. So the mask has wrong value.

At first, I cleared hot-removed CPU number's bit from llc_shared_map when hot removing CPU. But Borislav suggested that the problem will disappear if readded CPU is assigned same CPU number. And llc_shared_map must not be changed.

So the patch assigns same CPU number to readded CPU by linking CPU number to APIC ID. And by the patch, the problem disappers.

---
v2: change cpuid to cpunum
v3: fix Borislav's email address of Suggested-by
fix typo (ACPI ID to APIC ID)

---
   arch/x86/kernel/apic/apic.c | 33 ++++++++++++++++++++++++++++++++-
   1 file changed, 32 insertions(+), 1 deletion(-)
arch/x86/kernel/apic/apic.c +11 -0
index ad28db7..5dc3e50 100644
@@ -220,6 +220,23 @@ static void apic_pm_activate(void);
   static unsigned long apic_phys;
 
   /*
+ * Bind APIC ID to Logical CPU number
+ * Logical CPU number to APIC ID does not change by this array
+ * even if CPU is hotplugged. So don't clear the array even if
+ * CPU is hot-removed
+ */
+static int apicid_to_cpunum[MAX_LOCAL_APIC] = {
+	[0 ... MAX_LOCAL_APIC-1] = -1,
+};
+
+/*
+ * Represent Logical CPU number bound to APIC ID
Quoted 8 lines

arch/x86/kernel/apic/apic.c static void apic_pm_activate(void); · line 237

Quoted 4 lines
arch/x86/kernel/apic/apic.c +4 -0
+ * Don't clear a bit even if CPU is hot-removed
+ */
+static DECLARE_BITMAP(cpu_used_bits, CONFIG_NR_CPUS);
+static struct cpumask *const cpu_used_mask = to_cpumask(cpu_used_bits);

I think the name of cpu_used_mask is confusing since the term "used" sounds some state of CPU, which is not the case here. How about cpu_number_mask? This is more specific and avoids such confusion.

I'll update it soon.

Quoted 4 lines

Otherwise, the change looks good.

Thank you for review.

Thanks,
Yasuaki Ishimatsu

Quoted 14 lines

Thanks,
-Toshi

arch/x86/kernel/apic/apic.c static void apic_pm_activate(void); · line 241

Quoted 4 lines
arch/x86/kernel/apic/apic.c +2 -0
+
+/*
    * Get the LAPIC version
    */

^^
ps. I had a problem in applying this patch since there was an extra blank character at the begging of unchanged lines. But this might be an issue in my mailer.