[rh7,v2,3/3] ve/page_alloc, kstat: account allocation latencies per-task

Submitted by Andrey Ryabinin on Feb. 15, 2018, 2:48 p.m.

Details

Message ID 20180215144815.1341-3-aryabinin@virtuozzo.com
State New
Series "Series without cover letter"
Headers show

Commit Message

Andrey Ryabinin Feb. 15, 2018, 2:48 p.m.
Vstorage wants per-process allocation latencies:

 - total accumulated latency (total time spent inside the kernel allocator)
 - total alloc attempts (so that average latency can be calculated)

This adds /proc/<pid>/vz_latency file which outputs the numbers:

Type                   Total_lat                Calls
allocatomic:                    0                 1334
alloc:                    8000000                36643
allocmp:                        0                  919

https://jira.sw.ru/browse/PSBM-81395
Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Pavel Borzenkov <pborzenkov@virtuozzo.com>
---
 fs/proc/base.c        | 51 +++++++++++++++++++++++++++++++++++++++++++++++++++
 include/linux/sched.h |  4 ++++
 mm/page_alloc.c       |  5 +++++
 3 files changed, 60 insertions(+)

Patch hide | download patch | download mbox

diff --git a/fs/proc/base.c b/fs/proc/base.c
index 4c29a4f026ab..5646a351c076 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -55,6 +55,7 @@ 
 #include <linux/stat.h>
 #include <linux/task_io_accounting_ops.h>
 #include <linux/init.h>
+#include <linux/kstat.h>
 #include <linux/capability.h>
 #include <linux/file.h>
 #include <linux/fdtable.h>
@@ -574,6 +575,53 @@  static const struct file_operations proc_lstats_operations = {
 
 #endif
 
+#ifdef CONFIG_VE
+static void lastlat_seq_show(struct seq_file *m,
+		const char *name,
+		struct kstat_lat_snap_struct *snap)
+{
+	seq_printf(m, "%-12s %20Lu %20lu\n", name,
+			snap->totlat, snap->count);
+}
+
+static int vz_lat_show_proc(struct seq_file *m, void *v)
+{
+	int i;
+	struct inode *inode = m->private;
+	struct task_struct *task = get_proc_task(inode);
+	static const char *alloc_descr[] = {
+		"allocatomic:",
+		"alloc:",
+		"allocmp:",
+	};
+	static const int alloc_types[] = {
+		KSTAT_ALLOCSTAT_ATOMIC,
+		KSTAT_ALLOCSTAT_LOW,
+		KSTAT_ALLOCSTAT_LOW_MP,
+	};
+
+	seq_printf(m, "%-11s %20s %20s\n",
+			"Type", "Total_lat", "Calls");
+
+	for (i = 0; i < ARRAY_SIZE(alloc_types); i++)
+		lastlat_seq_show(m, alloc_descr[i],
+				&task->alloc_lat[alloc_types[i]]);
+	return 0;
+}
+
+static int vz_lat_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, vz_lat_show_proc, inode);
+}
+
+static const struct file_operations proc_vz_lat_operations = {
+	.open		= vz_lat_open,
+	.read		= seq_read,
+	.llseek		= seq_lseek,
+	.release	= single_release,
+};
+#endif
+
 #ifdef CONFIG_CGROUPS
 static int cgroup_open(struct inode *inode, struct file *file)
 {
@@ -3011,6 +3059,9 @@  static const struct pid_entry tgid_base_stuff[] = {
 	REG("timers",	  S_IRUGO, proc_timers_operations),
 	REG("aio",	  S_IRUGO|S_IWUSR, proc_aio_operations),
 #endif
+#ifdef CONFIG_VE
+	REG("vz_latency", S_IRUGO, proc_vz_lat_operations),
+#endif
 };
 
 static int proc_tgid_base_readdir(struct file * filp,
diff --git a/include/linux/sched.h b/include/linux/sched.h
index cc52094b4e97..1595be347b6d 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -30,6 +30,7 @@  struct sched_param {
 
 #include <linux/smp.h>
 #include <linux/sem.h>
+#include <linux/kstat.h>
 #include <linux/signal.h>
 #include <linux/compiler.h>
 #include <linux/completion.h>
@@ -1833,6 +1834,9 @@  struct task_struct {
 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
 	struct tlbflush_unmap_batch tlb_ubc;
 #endif
+#ifdef CONFIG_VE
+	struct kstat_lat_snap_struct alloc_lat[KSTAT_ALLOCSTAT_NR];
+#endif
 #if defined(CONFIG_FUNCTION_GRAPH_TRACER) && defined(CONFIG_S390)
 	/* Index of current stored address in ret_stack */
 	int curr_ret_stack;
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index f08ddeb77b82..fed5d879179f 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -3156,8 +3156,13 @@  static void __alloc_collect_stats(gfp_t gfp_mask, unsigned int order,
 	local_irq_save(flags);
 	cpu = smp_processor_id();
 	KSTAT_LAT_PCPU_ADD(&kstat_glob.alloc_lat[ind], time);
+
+	current->alloc_lat[ind].totlat += time;
+	current->alloc_lat[ind].count++;
+
 	if (!page)
 		kstat_glob.alloc_fails[cpu][ind]++;
+
 	local_irq_restore(flags);
 #endif
 }

Comments

Andrey Ryabinin Feb. 15, 2018, 2:49 p.m.
On 02/15/2018 05:48 PM, Andrey Ryabinin wrote:
> Vstorage wants per-process allocation latencies:
> 
>  - total accumulated latency (total time spent inside the kernel allocator)
>  - total alloc attempts (so that average latency can be calculated)
> 
> This adds /proc/<pid>/vz_latency file which outputs the numbers:
> 
> Type                   Total_lat                Calls
> allocatomic:                    0                 1334
> alloc:                    8000000                36643
> allocmp:                        0                  919
> 

Pavel, are you ok with that format?
Andrey Ryabinin Feb. 15, 2018, 3:07 p.m.
[pborzenkov@virtuozzo.com didn't work, retrying via Pavel.Borzenkov@acronis.com]

On 02/15/2018 05:48 PM, Andrey Ryabinin wrote:
> Vstorage wants per-process allocation latencies:
> 
>  - total accumulated latency (total time spent inside the kernel allocator)
>  - total alloc attempts (so that average latency can be calculated)
> 
> This adds /proc/<pid>/vz_latency file which outputs the numbers:
> 
> Type                   Total_lat                Calls
> allocatomic:                    0                 1334
> alloc:                    8000000                36643
> allocmp:                        0                  919
> 

Pavel, are you ok with that format? 


> https://jira.sw.ru/browse/PSBM-81395
> Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
> Cc: Pavel Borzenkov <pborzenkov@virtuozzo.com>
> ---
>  fs/proc/base.c        | 51 +++++++++++++++++++++++++++++++++++++++++++++++++++
>  include/linux/sched.h |  4 ++++
>  mm/page_alloc.c       |  5 +++++
>  3 files changed, 60 insertions(+)
> 
> diff --git a/fs/proc/base.c b/fs/proc/base.c
> index 4c29a4f026ab..5646a351c076 100644
> --- a/fs/proc/base.c
> +++ b/fs/proc/base.c
> @@ -55,6 +55,7 @@
>  #include <linux/stat.h>
>  #include <linux/task_io_accounting_ops.h>
>  #include <linux/init.h>
> +#include <linux/kstat.h>
>  #include <linux/capability.h>
>  #include <linux/file.h>
>  #include <linux/fdtable.h>
> @@ -574,6 +575,53 @@ static const struct file_operations proc_lstats_operations = {
>  
>  #endif
>  
> +#ifdef CONFIG_VE
> +static void lastlat_seq_show(struct seq_file *m,
> +		const char *name,
> +		struct kstat_lat_snap_struct *snap)
> +{
> +	seq_printf(m, "%-12s %20Lu %20lu\n", name,
> +			snap->totlat, snap->count);
> +}
> +
> +static int vz_lat_show_proc(struct seq_file *m, void *v)
> +{
> +	int i;
> +	struct inode *inode = m->private;
> +	struct task_struct *task = get_proc_task(inode);
> +	static const char *alloc_descr[] = {
> +		"allocatomic:",
> +		"alloc:",
> +		"allocmp:",
> +	};
> +	static const int alloc_types[] = {
> +		KSTAT_ALLOCSTAT_ATOMIC,
> +		KSTAT_ALLOCSTAT_LOW,
> +		KSTAT_ALLOCSTAT_LOW_MP,
> +	};
> +
> +	seq_printf(m, "%-11s %20s %20s\n",
> +			"Type", "Total_lat", "Calls");
> +
> +	for (i = 0; i < ARRAY_SIZE(alloc_types); i++)
> +		lastlat_seq_show(m, alloc_descr[i],
> +				&task->alloc_lat[alloc_types[i]]);
> +	return 0;
> +}
> +
> +static int vz_lat_open(struct inode *inode, struct file *file)
> +{
> +	return single_open(file, vz_lat_show_proc, inode);
> +}
> +
> +static const struct file_operations proc_vz_lat_operations = {
> +	.open		= vz_lat_open,
> +	.read		= seq_read,
> +	.llseek		= seq_lseek,
> +	.release	= single_release,
> +};
> +#endif
> +
>  #ifdef CONFIG_CGROUPS
>  static int cgroup_open(struct inode *inode, struct file *file)
>  {
> @@ -3011,6 +3059,9 @@ static const struct pid_entry tgid_base_stuff[] = {
>  	REG("timers",	  S_IRUGO, proc_timers_operations),
>  	REG("aio",	  S_IRUGO|S_IWUSR, proc_aio_operations),
>  #endif
> +#ifdef CONFIG_VE
> +	REG("vz_latency", S_IRUGO, proc_vz_lat_operations),
> +#endif
>  };
>  
>  static int proc_tgid_base_readdir(struct file * filp,
> diff --git a/include/linux/sched.h b/include/linux/sched.h
> index cc52094b4e97..1595be347b6d 100644
> --- a/include/linux/sched.h
> +++ b/include/linux/sched.h
> @@ -30,6 +30,7 @@ struct sched_param {
>  
>  #include <linux/smp.h>
>  #include <linux/sem.h>
> +#include <linux/kstat.h>
>  #include <linux/signal.h>
>  #include <linux/compiler.h>
>  #include <linux/completion.h>
> @@ -1833,6 +1834,9 @@ struct task_struct {
>  #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
>  	struct tlbflush_unmap_batch tlb_ubc;
>  #endif
> +#ifdef CONFIG_VE
> +	struct kstat_lat_snap_struct alloc_lat[KSTAT_ALLOCSTAT_NR];
> +#endif
>  #if defined(CONFIG_FUNCTION_GRAPH_TRACER) && defined(CONFIG_S390)
>  	/* Index of current stored address in ret_stack */
>  	int curr_ret_stack;
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index f08ddeb77b82..fed5d879179f 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -3156,8 +3156,13 @@ static void __alloc_collect_stats(gfp_t gfp_mask, unsigned int order,
>  	local_irq_save(flags);
>  	cpu = smp_processor_id();
>  	KSTAT_LAT_PCPU_ADD(&kstat_glob.alloc_lat[ind], time);
> +
> +	current->alloc_lat[ind].totlat += time;
> +	current->alloc_lat[ind].count++;
> +
>  	if (!page)
>  		kstat_glob.alloc_fails[cpu][ind]++;
> +
>  	local_irq_restore(flags);
>  #endif
>  }
>
Pavel Borzenkov Feb. 15, 2018, 3:13 p.m.
> On 15 Feb 2018, at 18:07, Andrey Ryabinin <aryabinin@virtuozzo.com> wrote:
> 
> [pborzenkov@virtuozzo.com didn't work, retrying via Pavel.Borzenkov@acronis.com]
> 
> On 02/15/2018 05:48 PM, Andrey Ryabinin wrote:
>> Vstorage wants per-process allocation latencies:
>> 
>> - total accumulated latency (total time spent inside the kernel allocator)
>> - total alloc attempts (so that average latency can be calculated)
>> 
>> This adds /proc/<pid>/vz_latency file which outputs the numbers:
>> 
>> Type                   Total_lat                Calls
>> allocatomic:                    0                 1334
>> alloc:                    8000000                36643
>> allocmp:                        0                  919
>> 
> 
> Pavel, are you ok with that format? 

Yes

> 
> 
>> https://jira.sw.ru/browse/PSBM-81395
>> Signed-off-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
>> Cc: Pavel Borzenkov <pborzenkov@virtuozzo.com>
>> ---
>> fs/proc/base.c        | 51 +++++++++++++++++++++++++++++++++++++++++++++++++++
>> include/linux/sched.h |  4 ++++
>> mm/page_alloc.c       |  5 +++++
>> 3 files changed, 60 insertions(+)
>> 
>> diff --git a/fs/proc/base.c b/fs/proc/base.c
>> index 4c29a4f026ab..5646a351c076 100644
>> --- a/fs/proc/base.c
>> +++ b/fs/proc/base.c
>> @@ -55,6 +55,7 @@
>> #include <linux/stat.h>
>> #include <linux/task_io_accounting_ops.h>
>> #include <linux/init.h>
>> +#include <linux/kstat.h>
>> #include <linux/capability.h>
>> #include <linux/file.h>
>> #include <linux/fdtable.h>
>> @@ -574,6 +575,53 @@ static const struct file_operations proc_lstats_operations = {
>> 
>> #endif
>> 
>> +#ifdef CONFIG_VE
>> +static void lastlat_seq_show(struct seq_file *m,
>> +		const char *name,
>> +		struct kstat_lat_snap_struct *snap)
>> +{
>> +	seq_printf(m, "%-12s %20Lu %20lu\n", name,
>> +			snap->totlat, snap->count);
>> +}
>> +
>> +static int vz_lat_show_proc(struct seq_file *m, void *v)
>> +{
>> +	int i;
>> +	struct inode *inode = m->private;
>> +	struct task_struct *task = get_proc_task(inode);
>> +	static const char *alloc_descr[] = {
>> +		"allocatomic:",
>> +		"alloc:",
>> +		"allocmp:",
>> +	};
>> +	static const int alloc_types[] = {
>> +		KSTAT_ALLOCSTAT_ATOMIC,
>> +		KSTAT_ALLOCSTAT_LOW,
>> +		KSTAT_ALLOCSTAT_LOW_MP,
>> +	};
>> +
>> +	seq_printf(m, "%-11s %20s %20s\n",
>> +			"Type", "Total_lat", "Calls");
>> +
>> +	for (i = 0; i < ARRAY_SIZE(alloc_types); i++)
>> +		lastlat_seq_show(m, alloc_descr[i],
>> +				&task->alloc_lat[alloc_types[i]]);
>> +	return 0;
>> +}
>> +
>> +static int vz_lat_open(struct inode *inode, struct file *file)
>> +{
>> +	return single_open(file, vz_lat_show_proc, inode);
>> +}
>> +
>> +static const struct file_operations proc_vz_lat_operations = {
>> +	.open		= vz_lat_open,
>> +	.read		= seq_read,
>> +	.llseek		= seq_lseek,
>> +	.release	= single_release,
>> +};
>> +#endif
>> +
>> #ifdef CONFIG_CGROUPS
>> static int cgroup_open(struct inode *inode, struct file *file)
>> {
>> @@ -3011,6 +3059,9 @@ static const struct pid_entry tgid_base_stuff[] = {
>> 	REG("timers",	  S_IRUGO, proc_timers_operations),
>> 	REG("aio",	  S_IRUGO|S_IWUSR, proc_aio_operations),
>> #endif
>> +#ifdef CONFIG_VE
>> +	REG("vz_latency", S_IRUGO, proc_vz_lat_operations),
>> +#endif
>> };
>> 
>> static int proc_tgid_base_readdir(struct file * filp,
>> diff --git a/include/linux/sched.h b/include/linux/sched.h
>> index cc52094b4e97..1595be347b6d 100644
>> --- a/include/linux/sched.h
>> +++ b/include/linux/sched.h
>> @@ -30,6 +30,7 @@ struct sched_param {
>> 
>> #include <linux/smp.h>
>> #include <linux/sem.h>
>> +#include <linux/kstat.h>
>> #include <linux/signal.h>
>> #include <linux/compiler.h>
>> #include <linux/completion.h>
>> @@ -1833,6 +1834,9 @@ struct task_struct {
>> #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
>> 	struct tlbflush_unmap_batch tlb_ubc;
>> #endif
>> +#ifdef CONFIG_VE
>> +	struct kstat_lat_snap_struct alloc_lat[KSTAT_ALLOCSTAT_NR];
>> +#endif
>> #if defined(CONFIG_FUNCTION_GRAPH_TRACER) && defined(CONFIG_S390)
>> 	/* Index of current stored address in ret_stack */
>> 	int curr_ret_stack;
>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>> index f08ddeb77b82..fed5d879179f 100644
>> --- a/mm/page_alloc.c
>> +++ b/mm/page_alloc.c
>> @@ -3156,8 +3156,13 @@ static void __alloc_collect_stats(gfp_t gfp_mask, unsigned int order,
>> 	local_irq_save(flags);
>> 	cpu = smp_processor_id();
>> 	KSTAT_LAT_PCPU_ADD(&kstat_glob.alloc_lat[ind], time);
>> +
>> +	current->alloc_lat[ind].totlat += time;
>> +	current->alloc_lat[ind].count++;
>> +
>> 	if (!page)
>> 		kstat_glob.alloc_fails[cpu][ind]++;
>> +
>> 	local_irq_restore(flags);
>> #endif
>> }
>>