Skip to contentSkip to conversation

mm/damon: Do little changes

PATCH In review Xin Hao 2 participants 6 messages Dec 9, 2021 – Dec 10, 2021

Subsystems (3)
lore
Share & tools
lore

Xin Hao

author
#0 lore raw

cover mm/damon: Do little changes

These patches are mainly used for fine adjustments.

Xin Hao (2):
mm/damon/dbgfs: Avoid target_ids display wrong pid value mm/damon: Modify the display form of damon tracepoint

 include/trace/events/damon.h | 6 +++---
 mm/damon/dbgfs.c             | 8 +++-----
 2 files changed, 6 insertions(+), 8 deletions(-)

--
2.31.0

author
#1 lore raw

PATCH 1/2 mm/damon/dbgfs: Avoid target_ids display wrong pid value

The return value of pid_vnr() is pid_t type which is defined by 'int', this may get a wrong pid value, if we set the return value type as 'unsigned long'. So there fix it.

---
 mm/damon/dbgfs.c | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)
mm/damon/dbgfs.c +3 -5
index 991b7fa0e858..97eebfb70ddf 100644
@@ -281,17 +281,15 @@ static inline bool targetid_is_pid(const struct damon_ctx *ctx)
 static ssize_t sprint_target_ids(struct damon_ctx *ctx, char *buf, ssize_t len)
 {
 	struct damon_target *t;
-	unsigned long id;
-	int written = 0;
+	int id, written = 0;
 	int rc;
 
 	damon_for_each_target(t, ctx) {
-		id = t->id;
 		if (targetid_is_pid(ctx))
 			/* Show pid numbers to debugfs users */
-			id = (unsigned long)pid_vnr((struct pid *)id);
+			id = (int)pid_vnr((struct pid *)t->id);
 
-		rc = scnprintf(&buf[written], len - written, "%lu ", id);
+		rc = scnprintf(&buf[written], len - written, "%d ", id);
 		if (!rc)
 			return -ENOMEM;
 		written += rc;
Signature · 1 line
2.31.0
author
#2 lore raw

PATCH 2/2 mm/damon: Modify the display form of damon tracepoint

In reply to Xin Hao

When I use the perf command to record damon monitor data, like below.

    # perf record -e damon:damon_aggregated
    # perf script
    ...target_id=18446462667479739520 nr_regions=13 281472805928960-281472942936064...
    ...target_id=18446462667479739520 nr_regions=13 281472942936064-281473080008704...
    ...target_id=18446462667479739520 nr_regions=13 281473080008704-281473216634880...

From a user's point of view, the 'target_id' and 'damon_region' which displays in decimal are not very friendly, So there do some changes, keep the 'target_id' display consistent with 'dbgfs/target_ids' interface and 'damon_region' is displayed in hexadecimal, just like below.

    # perf record -e damon:damon_aggregated
    # perf script
    ...target_id=5522 nr_regions=14 ffff716a3000-ffff79893000...
    ...target_id=5522 nr_regions=14 ffff79893000-ffff819dc000...
    ...target_id=5522 nr_regions=14 ffff819dc000-ffff89bd9000...
---
 include/trace/events/damon.h | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
include/trace/events/damon.h +3 -3
index 99ffa601e351..67de51814f4c 100644
@@ -17,7 +17,7 @@ TRACE_EVENT(damon_aggregated,
 	TP_ARGS(t, r, nr_regions),
 
 	TP_STRUCT__entry(
-		__field(unsigned long, target_id)
+		__field(int, target_id)
 		__field(unsigned int, nr_regions)
 		__field(unsigned long, start)
 		__field(unsigned long, end)
@@ -26,7 +26,7 @@ TRACE_EVENT(damon_aggregated,
 	),
 
 	TP_fast_assign(
-		__entry->target_id = t->id;
+		__entry->target_id = (int)pid_vnr((struct pid *)t->id);
 		__entry->nr_regions = nr_regions;
 		__entry->start = r->ar.start;
 		__entry->end = r->ar.end;
@@ -34,7 +34,7 @@ TRACE_EVENT(damon_aggregated,
 		__entry->age = r->age;
 	),
 
-	TP_printk("target_id=%lu nr_regions=%u %lu-%lu: %u %u",
+	TP_printk("target_id=%u nr_regions=%u %lx-%lx: %u %u",
 			__entry->target_id, __entry->nr_regions,
 			__entry->start, __entry->end,
 			__entry->nr_accesses, __entry->age)

--
2.31.0

maintainer
#3 lore raw

Re: [PATCH V1 2/2] mm/damon: Modify the display form of damon tracepoint

Quoted 36 lines · On Fri, 10 Dec 2021 00:33:17 +0800 Xin Hao <[email protected]> wrote:

When I use the perf command to record damon monitor data, like below.

    # perf record -e damon:damon_aggregated
    # perf script
    ...target_id=18446462667479739520 nr_regions=13 281472805928960-281472942936064...
    ...target_id=18446462667479739520 nr_regions=13 281472942936064-281473080008704...
    ...target_id=18446462667479739520 nr_regions=13 281473080008704-281473216634880...

From a user's point of view, the 'target_id' and 'damon_region' which displays in decimal are not very friendly, So there do some changes, keep the 'target_id' display consistent with 'dbgfs/target_ids' interface and 'damon_region' is displayed in hexadecimal, just like below.

    # perf record -e damon:damon_aggregated
    # perf script
    ...target_id=5522 nr_regions=14 ffff716a3000-ffff79893000...
    ...target_id=5522 nr_regions=14 ffff79893000-ffff819dc000...
    ...target_id=5522 nr_regions=14 ffff819dc000-ffff89bd9000...
---
 include/trace/events/damon.h | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
include/trace/events/damon.h +1 -1
index 99ffa601e351..67de51814f4c 100644
@@ -17,7 +17,7 @@ TRACE_EVENT(damon_aggregated,
 	TP_ARGS(t, r, nr_regions),
 
 	TP_STRUCT__entry(
-		__field(unsigned long, target_id)
+		__field(int, target_id)
 		__field(unsigned int, nr_regions)
 		__field(unsigned long, start)
 		__field(unsigned long, end)

include/trace/events/damon.h TRACE_EVENT(damon_aggregated, · line 29

Quoted 5 lines
include/trace/events/damon.h +1 -1
@@ -26,7 +26,7 @@ TRACE_EVENT(damon_aggregated,
 	),
 
 	TP_fast_assign(
-		__entry->target_id = t->id;
+		__entry->target_id = (int)pid_vnr((struct pid *)t->id);

I think this would break physical address space monitoring. Have you tested this change for that?

Thanks,
SJ

Quoted 14 lines
include/trace/events/damon.h +1 -2
 		__entry->nr_regions = nr_regions;
 		__entry->start = r->ar.start;
 		__entry->end = r->ar.end;
@@ -34,7 +34,7 @@ TRACE_EVENT(damon_aggregated,
 		__entry->age = r->age;
 	),
 
-	TP_printk("target_id=%lu nr_regions=%u %lu-%lu: %u %u",
+	TP_printk("target_id=%u nr_regions=%u %lx-%lx: %u %u",
 			__entry->target_id, __entry->nr_regions,
 			__entry->start, __entry->end,
 			__entry->nr_accesses, __entry->age)
--

2.31.0

author
#4 lore raw

Re: [PATCH V1 2/2] mm/damon: Modify the display form of damon tracepoint

Hi SeongJae:

Quoted 40 lines · SeongJae Park
Quoted 38 lines · On Fri, 10 Dec 2021 00:33:17 +0800 Xin Hao <[email protected]> wrote:

When I use the perf command to record damon monitor data, like below.

     # perf record -e damon:damon_aggregated
     # perf script
     ...target_id=18446462667479739520 nr_regions=13 281472805928960-281472942936064...
     ...target_id=18446462667479739520 nr_regions=13 281472942936064-281473080008704...
     ...target_id=18446462667479739520 nr_regions=13 281473080008704-281473216634880...

From a user's point of view, the 'target_id' and 'damon_region' which displays in decimal
are not very friendly, So there do some changes, keep the 'target_id' display consistent with 'dbgfs/target_ids' interface and 'damon_region' is displayed in hexadecimal, just like below.

     # perf record -e damon:damon_aggregated
     # perf script
     ...target_id=5522 nr_regions=14 ffff716a3000-ffff79893000...
     ...target_id=5522 nr_regions=14 ffff79893000-ffff819dc000...
     ...target_id=5522 nr_regions=14 ffff819dc000-ffff89bd9000...
---
  include/trace/events/damon.h | 6 +++---
  1 file changed, 3 insertions(+), 3 deletions(-)
include/trace/events/damon.h +1 -1
index 99ffa601e351..67de51814f4c 100644
@@ -17,7 +17,7 @@ TRACE_EVENT(damon_aggregated,
  	TP_ARGS(t, r, nr_regions),
 
  	TP_STRUCT__entry(
-		__field(unsigned long, target_id)
+		__field(int, target_id)
  		__field(unsigned int, nr_regions)
  		__field(unsigned long, start)
  		__field(unsigned long, end)
@@ -26,7 +26,7 @@ TRACE_EVENT(damon_aggregated,
  	),
 
Quoted 5 lines

include/trace/events/damon.h TRACE_EVENT(damon_aggregated, · line 29

Quoted 3 lines
include/trace/events/damon.h +1 -1
  	TP_fast_assign(
-		__entry->target_id = t->id;
+		__entry->target_id = (int)pid_vnr((struct pid *)t->id);

I think this would break physical address space monitoring. Have you tested this change for that?

Yes, you are right,  But I encountered some problems while testing physical address, it seems that my operation did not work

I did some test like this:

    # echo "42 0x0000000840000000 0x000000103fffffff" > init_regions
    # echo paddr > target_ids
    # echo on > monitor_on

i get the physical address from my kernel startup log.

15 [ 0.000000] Early memory node ranges
16 [ 0.000000] node 0: [mem 0x0000000040000000-0x000000083bc7ffff] 17 [ 0.000000] node 0: [mem 0x000000083bc80000-0x000000083bffffff] 18 [ 0.000000] node 0: [mem 0x000000083c000000-0x000000083c03ffff] 19 [ 0.000000] node 0: [mem 0x000000083c040000-0x000000083c0fffff] 20 [ 0.000000] node 0: [mem 0x000000083c100000-0x000000083f3dffff] 21 [ 0.000000] node 0: [mem 0x000000083f3e0000-0x000000083f46ffff] 22 [ 0.000000] node 0: [mem 0x000000083f470000-0x000000083f47ffff] 23 [ 0.000000] node 0: [mem 0x000000083f480000-0x000000083f59ffff] 24 [ 0.000000] node 0: [mem 0x000000083f5a0000-0x000000083fffffff] 25 [ 0.000000] node 1: [mem 0x0000000840000000-0x000000103fffffff] 26 [ 0.000000] node 2: [mem 0x0000001040000000-0x000000183fffffff] 27 [ 0.000000] node 3: [mem 0x0000001840000000-0x000000203fffffff] 28 [ 0.000000] Initmem setup node 0 [mem 0x0000000040000000-0x000000083fffffff] 29 [ 0.000000] On node 0 totalpages: 8388608

Is there anything wrong ?

Quoted 19 lines

Thanks,
SJ

Quoted 14 lines
include/trace/events/damon.h +1 -2
  		__entry->nr_regions = nr_regions;
  		__entry->start = r->ar.start;
  		__entry->end = r->ar.end;
@@ -34,7 +34,7 @@ TRACE_EVENT(damon_aggregated,
  		__entry->age = r->age;
  	),
 
-	TP_printk("target_id=%lu nr_regions=%u %lu-%lu: %u %u",
+	TP_printk("target_id=%u nr_regions=%u %lx-%lx: %u %u",
  			__entry->target_id, __entry->nr_regions,
  			__entry->start, __entry->end,
  			__entry->nr_accesses, __entry->age)
--

2.31.0

Signature · 2 lines
Best Regards!
Xin Hao
maintainer
#5 lore raw

Re: [PATCH V1 2/2] mm/damon: Modify the display form of damon tracepoint

Hi Xin,

Quoted 33 lines · On Fri, 10 Dec 2021 11:36:23 +0800 Xin Hao <[email protected]> wrote:

[-- Attachment #1: Type: text/plain, Size: 4038 bytes --]

Hi SeongJae:

Quoted 28 lines · SeongJae Park
Quoted 26 lines · On Fri, 10 Dec 2021 00:33:17 +0800 Xin Hao <[email protected]> wrote:

When I use the perf command to record damon monitor data, like below.

     # perf record -e damon:damon_aggregated
     # perf script
     ...target_id=18446462667479739520 nr_regions=13 281472805928960-281472942936064...
     ...target_id=18446462667479739520 nr_regions=13 281472942936064-281473080008704...
     ...target_id=18446462667479739520 nr_regions=13 281473080008704-281473216634880...

From a user's point of view, the 'target_id' and 'damon_region' which displays in decimal
are not very friendly, So there do some changes, keep the 'target_id' display consistent with 'dbgfs/target_ids' interface and 'damon_region' is displayed in hexadecimal, just like below.

     # perf record -e damon:damon_aggregated
     # perf script
     ...target_id=5522 nr_regions=14 ffff716a3000-ffff79893000...
     ...target_id=5522 nr_regions=14 ffff79893000-ffff819dc000...
     ...target_id=5522 nr_regions=14 ffff819dc000-ffff89bd9000...
---
  include/trace/events/damon.h | 6 +++---
  1 file changed, 3 insertions(+), 3 deletions(-)
include/trace/events/damon.h +0 -0
index 99ffa601e351..67de51814f4c 100644

[...]

Quoted 36 lines
Quoted 5 lines

include/trace/events/damon.h TRACE_EVENT(damon_aggregated, · line 29

Quoted 3 lines

TP_fast_assign(

include/trace/events/damon.h +1 -1
-		__entry->target_id = t->id;
+		__entry->target_id = (int)pid_vnr((struct pid *)t->id);

I think this would break physical address space monitoring. Have you tested this change for that?

Yes, you are right, But I encountered some problems while testing physical address, it seems that my operation did not work

I did some test like this:

     # echo "42 0x0000000840000000 0x000000103fffffff" > init_regions
     # echo paddr > target_ids
     # echo on > monitor_on

i get the physical address from my kernel startup log.

15 [ 0.000000] Early memory node ranges
16 [ 0.000000] node 0: [mem 0x0000000040000000-0x000000083bc7ffff] 17 [ 0.000000] node 0: [mem 0x000000083bc80000-0x000000083bffffff] 18 [ 0.000000] node 0: [mem 0x000000083c000000-0x000000083c03ffff] 19 [ 0.000000] node 0: [mem 0x000000083c040000-0x000000083c0fffff] 20 [ 0.000000] node 0: [mem 0x000000083c100000-0x000000083f3dffff] 21 [ 0.000000] node 0: [mem 0x000000083f3e0000-0x000000083f46ffff] 22 [ 0.000000] node 0: [mem 0x000000083f470000-0x000000083f47ffff] 23 [ 0.000000] node 0: [mem 0x000000083f480000-0x000000083f59ffff] 24 [ 0.000000] node 0: [mem 0x000000083f5a0000-0x000000083fffffff] 25 [ 0.000000] node 1: [mem 0x0000000840000000-0x000000103fffffff] 26 [ 0.000000] node 2: [mem 0x0000001040000000-0x000000183fffffff] 27 [ 0.000000] node 3: [mem 0x0000001840000000-0x000000203fffffff] 28 [ 0.000000] Initmem setup node 0 [mem 0x0000000040000000-0x000000083fffffff]
29 [ 0.000000] On node 0 totalpages: 8388608

Quoted 1 line

Is there anything wrong ?

"The target id should already in target_ids file"[1].

For proper use of DAMON, I'd like to recommend you to refer to, or use the official DAMON user space tool[2] instead of the debugfs interface. As the document[3] says "DAMON user space tool. This is for privileged people such as system administrators who want a just-working human-friendly interface".

Also, some of your patches including this break the user space tool. As it is an important part of DAMON echosystem, it would be great if you could consider taking care in keeping it unbroken, too.

[1] https://docs.kernel.org/admin-guide/mm/damon/usage.html#initial-monitoring-target-regions
[2] https://github.com/awslabs/damo
[3] https://docs.kernel.org/admin-guide/mm/damon/usage.html#detailed-usages

Thanks,
SJ

[...]