Skip to contentSkip to conversation

mm/damon: trace: print address range in hex

PATCH In review Akinobu Mita 2 participants 7 messages Dec 30, 2024 – Jan 1, 2025

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mm/damon: trace: print address range in hex

Currently, the address ranges are displayed in decimal format in tracepoints for DAMON, but hexadecimal format is easier to compare with address ranges in /proc/<pid>/numa_maps.

Existing tools such as damo need to be able to handle changing the format of address ranges, so display address ranges in hexadecimal with a "0x" prefix.

---
 include/trace/events/damon.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
include/trace/events/damon.h +2 -2
index da4bd9fd1162..8df59ef18660 100644
@@ -41,7 +41,7 @@ TRACE_EVENT_CONDITION(damos_before_apply,
 		__entry->nr_regions = nr_regions;
 	),
 
-	TP_printk("ctx_idx=%u scheme_idx=%u target_idx=%lu nr_regions=%u %lu-%lu: %u %u",
+	TP_printk("ctx_idx=%u scheme_idx=%u target_idx=%lu nr_regions=%u 0x%lx-0x%lx: %u %u",
 			__entry->context_idx, __entry->scheme_idx,
 			__entry->target_idx, __entry->nr_regions,
 			__entry->start, __entry->end,
@@ -73,7 +73,7 @@ TRACE_EVENT(damon_aggregated,
 		__entry->age = r->age;
 	),
 
-	TP_printk("target_id=%lu nr_regions=%u %lu-%lu: %u %u",
+	TP_printk("target_id=%lu nr_regions=%u 0x%lx-0x%lx: %u %u",
 			__entry->target_id, __entry->nr_regions,
 			__entry->start, __entry->end,
 			__entry->nr_accesses, __entry->age)
Signature · 1 line
2.34.1
#1 lore raw

PATCH 1/1 _damo_records: support address range in hex format

In preparation for changing the display format of address ranges in tracepoints for DAMON, if the start and end addresses of an address range have a "0x" prefix, convert them as hexadecimal numbers.

---
 src/_damo_records.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
src/_damo_records.py +2 -2
index 6e541a0..55d864c 100644
@@ -251,7 +251,7 @@ def parse_damon_aggregated_perf_script_fields(fields):
     target_id = int(fields[5].split('=')[1])
     nr_regions = int(fields[6].split('=')[1])
 
-    start_addr, end_addr = [int(x) for x in fields[7][:-1].split('-')]
+    start_addr, end_addr = [int(x, 0) for x in fields[7][:-1].split('-')]
     nr_accesses = int(fields[8])
     if len(fields) == 10:
         age = int(fields[9])
@@ -288,7 +288,7 @@ def parse_damos_before_apply_perf_script_fields(fields):
     target_id = int(fields[7].split('=')[1])
     nr_regions = int(fields[8].split('=')[1])
 
-    start_addr, end_addr = [int(x) for x in fields[9][:-1].split('-')]
+    start_addr, end_addr = [int(x, 0) for x in fields[9][:-1].split('-')]
     nr_accesses = int(fields[10])
     age = int(fields[11])
     region = _damon.DamonRegion(start_addr, end_addr, nr_accesses,
Signature · 1 line
2.34.1
maintainer
#2 lore raw

Re: [PATCH damo] _damo_records: support address range in hex format

Quoted 3 lines · On Mon, 30 Dec 2024 22:42:21 +0900 Akinobu Mita <[email protected]> wrote:

In preparation for changing the display format of address ranges in tracepoints for DAMON, if the start and end addresses of an address range have a "0x" prefix, convert them as hexadecimal numbers.

Thank you for this patch. However, I'm not yet convinced to the DAMON side change. Let's discuss this change after a discussion on the DAMON change is finished.

Thanks,
SJ

[...]

maintainer
#3 lore raw
In reply to Akinobu Mita

Hello Akinobu,

Quoted 3 lines · On Mon, 30 Dec 2024 22:42:20 +0900 Akinobu Mita <[email protected]> wrote:

Currently, the address ranges are displayed in decimal format in tracepoints for DAMON, but hexadecimal format is easier to compare with address ranges in /proc/<pid>/numa_maps.

I agree that it could be easier for the use case. But, I think writing and using a script converting formats for DAMON tracepoint output or numa_maps could also be a way to resolve the issue. Do you have any problem at using such solution?

Quoted 3 lines

Existing tools such as damo need to be able to handle changing the format of address ranges,

Thank you for taking care of damo, too!

Quoted 2 lines

so display address ranges in hexadecimal with a "0x" prefix.

This is a user-visible behavioral change. I understand it is an improvement for your use case, but could be a degradation for some use cases. Also the change could confuse or break old use cases. I understand that 'damo' users will not get such problems thanks to your work, but not every DAMON users use 'damo'.

So I'd like to add this change only if there is some critical issue or benefit that we cannot fix or achieve without this change. As of now, I don't see such things since I think you could use some user-space tools for your use case. Please let me know if there is something that I'm missing.

Thanks,
SJ

Quoted 31 lines
---
 include/trace/events/damon.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
include/trace/events/damon.h +2 -2
index da4bd9fd1162..8df59ef18660 100644
@@ -41,7 +41,7 @@ TRACE_EVENT_CONDITION(damos_before_apply,
 		__entry->nr_regions = nr_regions;
 	),
 
-	TP_printk("ctx_idx=%u scheme_idx=%u target_idx=%lu nr_regions=%u %lu-%lu: %u %u",
+	TP_printk("ctx_idx=%u scheme_idx=%u target_idx=%lu nr_regions=%u 0x%lx-0x%lx: %u %u",
 			__entry->context_idx, __entry->scheme_idx,
 			__entry->target_idx, __entry->nr_regions,
 			__entry->start, __entry->end,
@@ -73,7 +73,7 @@ TRACE_EVENT(damon_aggregated,
 		__entry->age = r->age;
 	),
 
-	TP_printk("target_id=%lu nr_regions=%u %lu-%lu: %u %u",
+	TP_printk("target_id=%lu nr_regions=%u 0x%lx-0x%lx: %u %u",
 			__entry->target_id, __entry->nr_regions,
 			__entry->start, __entry->end,
 			__entry->nr_accesses, __entry->age)
Signature · 1 line
2.34.1
maintainer
#4 lore raw
Quoted 11 lines · On Mon, 30 Dec 2024 10:28:13 -0800 SeongJae Park <[email protected]> wrote:

Hello Akinobu,

Quoted 3 lines · On Mon, 30 Dec 2024 22:42:20 +0900 Akinobu Mita <[email protected]> wrote:

Currently, the address ranges are displayed in decimal format in tracepoints for DAMON, but hexadecimal format is easier to compare with address ranges in /proc/<pid>/numa_maps.

I agree that it could be easier for the use case. But, I think writing and using a script converting formats for DAMON tracepoint output or numa_maps could also be a way to resolve the issue.

For example, we can extend 'damo' to optionally do the conversion. Depending on your use case (if you also use 'damo' for your use case), it might be a way to move forward?

Thanks,
SJ

[...]

#5 lore raw

2024年12月31日(火) 3:39 SeongJae Park <[email protected]>:

Quoted 11 lines · On Mon, 30 Dec 2024 10:28:13 -0800 SeongJae Park <[email protected]> wrote:
Quoted 9 lines

Hello Akinobu,

Quoted 3 lines · On Mon, 30 Dec 2024 22:42:20 +0900 Akinobu Mita <[email protected]> wrote:

Currently, the address ranges are displayed in decimal format in tracepoints for DAMON, but hexadecimal format is easier to compare with address ranges in /proc/<pid>/numa_maps.

Quoted 7 lines
Quoted 3 lines

I agree that it could be easier for the use case. But, I think writing and using a script converting formats for DAMON tracepoint output or numa_maps could also be a way to resolve the issue.

For example, we can extend 'damo' to optionally do the conversion. Depending on your use case (if you also use 'damo' for your use case), it might be a way to move forward?

I'm currently using perf trace (-e damon:* --libtraceevent_print) to see which address range accesses were detected or not, so as you suggested, I'll create a script to convert the output of the perf trace.

maintainer
#6 lore raw
Quoted 16 lines · On Tue, 31 Dec 2024 21:56:43 +0900 Akinobu Mita <[email protected]> wrote:

2024年12月31日(火) 3:39 SeongJae Park <[email protected]>:

Quoted 15 lines · On Mon, 30 Dec 2024 10:28:13 -0800 SeongJae Park <[email protected]> wrote:
Quoted 12 lines

Hello Akinobu,

Quoted 3 lines · On Mon, 30 Dec 2024 22:42:20 +0900 Akinobu Mita <[email protected]> wrote:

Currently, the address ranges are displayed in decimal format in tracepoints for DAMON, but hexadecimal format is easier to compare with address ranges in /proc/<pid>/numa_maps.

I agree that it could be easier for the use case. But, I think writing and using a script converting formats for DAMON tracepoint output or numa_maps could also be a way to resolve the issue.

Quoted 6 lines
Quoted 3 lines

For example, we can extend 'damo' to optionally do the conversion. Depending on your use case (if you also use 'damo' for your use case), it might be a way to move forward?

I'm currently using perf trace (-e damon:* --libtraceevent_print) to see which address range accesses were detected or not,

Thank you for kindly sharing your use case. I'm not very sure why you need to use 'perf trace' instead of 'damo', though. 'damo' supports recording the whole events and getting access monitoring result snapshots in live, with a few results filtering criterias. I just assume you had to use 'perf trace' since you want to read full events in realtime, but please let me know if you want to know more details about 'damo' features that might help you.

Quoted 2 lines

so as you suggested, I'll create a
script to convert the output of the perf trace.

Thank you for accepting my suggestion. Please let me know if there is anything that I can help at writing the script. For example, if you really need to use 'perf trace', we're open to extend 'damo' for that.

Thanks,
SJ