Skip to contentSkip to conversation

perf: disable mmap2 support

PATCH In review Stephane Eranian 4 participants 17 messages Oct 9, 2013 – Oct 15, 2013 perf events

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 perf: disable mmap2 support

This patch disables the mmap2 record format support for now. The support needs a bit more work to cover VM_CLONE cases.

The patch leaves attr->mmap2 defined, but returns an error if it is set. That ensures no PERF_RECORD_MMAP2 record can be generated. Yet it minimizes the reverts in the perf tool.

---
 include/uapi/linux/perf_event.h |   18 +-----------------
 kernel/events/core.c            |    4 ++++
 2 files changed, 5 insertions(+), 17 deletions(-)
include/uapi/linux/perf_event.h +1 -17
index da48837..8074b29 100644
@@ -682,23 +682,7 @@ enum perf_event_type {
 	PERF_RECORD_SAMPLE			= 9,
 
 	/*
-	 * The MMAP2 records are an augmented version of MMAP, they add
-	 * maj, min, ino numbers to be used to uniquely identify each mapping
-	 *
-	 * struct {
-	 *	struct perf_event_header	header;
-	 *
-	 *	u32				pid, tid;
-	 *	u64				addr;
-	 *	u64				len;
-	 *	u64				pgoff;
-	 *	u32				maj;
-	 *	u32				min;
-	 *	u64				ino;
-	 *	u64				ino_generation;
-	 *	char				filename[];
-	 * 	struct sample_id		sample_id;
-	 * };
+	 * reserved, do not use
 	 */
 	PERF_RECORD_MMAP2			= 10,
 
kernel/events/core.c +4 -0
index c716385..d7152a6 100644
@@ -6782,6 +6782,10 @@ static int perf_copy_attr(struct perf_event_attr __user *uattr,
 	if (attr->read_format & ~(PERF_FORMAT_MAX-1))
 		return -EINVAL;
 
+	/* disable for now */
+	if (attr->mmap2)
+		return -EINVAL;
+
 	if (attr->sample_type & PERF_SAMPLE_BRANCH_STACK) {
 		u64 mask = attr->branch_sample_type;
 
Signature · 1 line
1.7.10.4
#1 lore raw
Quoted 9 lines · Stephane Eranian

This patch disables the mmap2 record format support for now. The support needs a bit more work to cover VM_CLONE cases.

The patch leaves attr->mmap2 defined, but returns an error if it is set. That ensures no PERF_RECORD_MMAP2 record can be generated. Yet it minimizes the reverts in the perf tool.

The userspace code generates MMAP2 events for synthesized mmap events. That should be reverted to MMAP until the format of MMAP2 events is finalized.

David

#2 lore raw
Quoted 7 lines · David Ahern
Quoted 6 lines · Stephane Eranian

This patch disables the mmap2 record format support for now. The support needs a bit more work to cover VM_CLONE cases.

Quoted 9 lines
Quoted 4 lines

The patch leaves attr->mmap2 defined, but returns an error if it is set. That ensures no PERF_RECORD_MMAP2 record can be generated. Yet it minimizes the reverts in the perf tool.

The userspace code generates MMAP2 events for synthesized mmap events. That should be reverted to MMAP until the format of MMAP2 events is finalized.

Ah, yes, that is correct. I forgot about that ;-( Let me send a revised patch then that covers both.

Quoted 1 line

David

#3 lore raw
Quoted 14 lines · Stephane Eranian
Quoted 13 lines · David Ahern
Quoted 10 lines · Stephane Eranian

This patch disables the mmap2 record format support for now. The support needs a bit more work to cover VM_CLONE cases.

The patch leaves attr->mmap2 defined, but returns an error if it is set. That ensures no PERF_RECORD_MMAP2 record can be generated. Yet it minimizes the reverts in the perf tool.

Quoted 6 lines
Quoted 3 lines

The userspace code generates MMAP2 events for synthesized mmap events. That should be reverted to MMAP until the format of MMAP2 events is finalized.

Ah, yes, that is correct. I forgot about that ;-( Let me send a revised patch then that covers both.

Looks like the easiest thing to do for perf is to revert:

Revert "perf tools: Add default handler for mmap2 events" git revert 6adb0b0ae26fcc35cfec068d71f13863faac5b44

Revert "perf tools: Add attr->mmap2 support" git revert 5c5e854bc760a2e2c878df3cfcf2afa4febcd511

Do you agree?

#4 lore raw
Quoted 9 lines · Stephane Eranian

Looks like the easiest thing to do for perf is to revert:

    Revert "perf tools: Add default handler for mmap2 events"
    git revert 6adb0b0ae26fcc35cfec068d71f13863faac5b44

    Revert "perf tools: Add attr->mmap2 support"
    git revert 5c5e854bc760a2e2c878df3cfcf2afa4febcd511

Do you agree?

If you want to strip all of mmap2 you also need 5654a0257a86c093c2b38ea5d3ceaa17affd735b.

David

#5 lore raw
Quoted 7 lines · David Ahern
Quoted 6 lines · Stephane Eranian

Looks like the easiest thing to do for perf is to revert:

Revert "perf tools: Add default handler for mmap2 events" git revert 6adb0b0ae26fcc35cfec068d71f13863faac5b44

Quoted 9 lines
Quoted 4 lines
    Revert "perf tools: Add attr->mmap2 support"
    git revert 5c5e854bc760a2e2c878df3cfcf2afa4febcd511

Do you agree?

If you want to strip all of mmap2 you also need
5654a0257a86c093c2b38ea5d3ceaa17affd735b.

You mean:
commit 384c671e33a7ddf905f2c60b433b9883f0e5a605

Author: David Ahern <[email protected]>
Date:   Sun Sep 22 19:44:58 2013 -0600

perf trace: Add mmap2 handler

Right?

#6 lore raw
Quoted 17 lines · Stephane Eranian
Quoted 7 lines · David Ahern
Quoted 6 lines · Stephane Eranian

Looks like the easiest thing to do for perf is to revert:

Revert "perf tools: Add default handler for mmap2 events" git revert 6adb0b0ae26fcc35cfec068d71f13863faac5b44

Quoted 9 lines
Quoted 4 lines
     Revert "perf tools: Add attr->mmap2 support"
     git revert 5c5e854bc760a2e2c878df3cfcf2afa4febcd511

Do you agree?

If you want to strip all of mmap2 you also need
5654a0257a86c093c2b38ea5d3ceaa17affd735b.

Quoted 8 lines

You mean:
commit 384c671e33a7ddf905f2c60b433b9883f0e5a605

Author: David Ahern <[email protected]>
Date:   Sun Sep 22 19:44:58 2013 -0600

perf trace: Add mmap2 handler

Right?

Yes, a couple of those have 2 commit ids. I pulled the id's from acme's perf/core branch

David

#7 lore raw

Em Wed, Oct 09, 2013 at 12:03:58PM -0600, David Ahern escreveu:

Quoted 8 lines · Stephane Eranian
Quoted 7 lines · David Ahern
Quoted 1 line · Stephane Eranian

You mean:
commit 384c671e33a7ddf905f2c60b433b9883f0e5a605

Author: David Ahern <[email protected]>
Date:   Sun Sep 22 19:44:58 2013 -0600
Quoted 6 lines
Quoted 3 lines

perf trace: Add mmap2 handler

Right?

Yes, a couple of those have 2 commit ids. I pulled the id's from acme's perf/core branch

I'm merging my perf/core branch with Ingo's build speedup series, so I've removed that patch already, will be ok as soon as the disabling is done in perf/urgent and those branches get merged.

- Arnaldo

#8 lore raw
In reply to Stephane Eranian

Hi Stephane:

Quoted 17 lines · Stephane Eranian
Quoted 7 lines · David Ahern
Quoted 6 lines · Stephane Eranian

Looks like the easiest thing to do for perf is to revert:

Revert "perf tools: Add default handler for mmap2 events" git revert 6adb0b0ae26fcc35cfec068d71f13863faac5b44

Quoted 9 lines
Quoted 4 lines
     Revert "perf tools: Add attr->mmap2 support"
     git revert 5c5e854bc760a2e2c878df3cfcf2afa4febcd511

Do you agree?

If you want to strip all of mmap2 you also need
5654a0257a86c093c2b38ea5d3ceaa17affd735b.

Quoted 7 lines

You mean:
commit 384c671e33a7ddf905f2c60b433b9883f0e5a605

Author: David Ahern <[email protected]>
Date:   Sun Sep 22 19:44:58 2013 -0600

perf trace: Add mmap2 handler

Given that 3.12 rc5 will be out on Sunday and it takes time to go from Arnaldo to Ingo to Linus the revert patches should go out today - in Arnaldo's next perf/urgent request.

David

#9 lore raw

Arnaldo,

Could you revert all the mmap2 patches in your perf/urgent tree for now? Thanks.

Quoted 39 lines · David Ahern

Hi Stephane:

Quoted 21 lines · Stephane Eranian
Quoted 12 lines · David Ahern
Quoted 10 lines · Stephane Eranian

Looks like the easiest thing to do for perf is to revert:

     Revert "perf tools: Add default handler for mmap2 events"
     git revert 6adb0b0ae26fcc35cfec068d71f13863faac5b44

     Revert "perf tools: Add attr->mmap2 support"
     git revert 5c5e854bc760a2e2c878df3cfcf2afa4febcd511
Quoted 7 lines
Quoted 1 line

Do you agree?

If you want to strip all of mmap2 you also need
5654a0257a86c093c2b38ea5d3ceaa17affd735b.

Quoted 7 lines

You mean:
commit 384c671e33a7ddf905f2c60b433b9883f0e5a605

Author: David Ahern <[email protected]>
Date:   Sun Sep 22 19:44:58 2013 -0600

perf trace: Add mmap2 handler

Given that 3.12 rc5 will be out on Sunday and it takes time to go from Arnaldo to Ingo to Linus the revert patches should go out today - in Arnaldo's next perf/urgent request.

David

#10 lore raw
Quoted 4 lines · Stephane Eranian

Arnaldo,

Could you revert all the mmap2 patches in your perf/urgent tree for now? Thanks.

I think it would be helpful if you re-sent the final patch, with a full changelog, etc., in a separate thread, not inside an existing discussion and such.

Thanks,

Ingo

#11 lore raw
Quoted 12 lines · Ingo Molnar
Quoted 4 lines · Stephane Eranian

Arnaldo,

Could you revert all the mmap2 patches in your perf/urgent tree for now? Thanks.

I think it would be helpful if you re-sent the final patch, with a full changelog, etc., in a separate thread, not inside an existing discussion and such.

Are you talking about the perf tool or kernel patch? I did not send any perf-related mmap2 revert patches.

Quoted 3 lines

Thanks,

Ingo

#12 lore raw
Quoted 6 lines · Stephane Eranian
Quoted 5 lines · Ingo Molnar
Quoted 2 lines · Stephane Eranian

Arnaldo,

Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

Could you revert all the mmap2 patches in your perf/urgent tree for now? Thanks.

I think it would be helpful if you re-sent the final patch, with a full changelog, etc., in a separate thread, not inside an existing discussion and such.

Are you talking about the perf tool or kernel patch?

Both.

Quoted 1 line

I did not send any perf-related mmap2 revert patches.

Would be nice to have a single patch that reverts both.

Thanks,

Ingo

#13 lore raw
Quoted 16 lines · Ingo Molnar
Quoted 13 lines · Stephane Eranian
Quoted 11 lines · Ingo Molnar
Quoted 4 lines · Stephane Eranian

Arnaldo,

Could you revert all the mmap2 patches in your perf/urgent tree for now? Thanks.

I think it would be helpful if you re-sent the final patch, with a full changelog, etc., in a separate thread, not inside an existing discussion and such.

Quoted 8 lines
Quoted 1 line

Are you talking about the perf tool or kernel patch?

Both.

Quoted 1 line

I did not send any perf-related mmap2 revert patches.

Would be nice to have a single patch that reverts both.

I will send a single patch to remove everything (kernel + tool) then.

#14 lore raw
Quoted 19 lines · Stephane Eranian
Quoted 18 lines · Ingo Molnar
Quoted 14 lines · Stephane Eranian
Quoted 11 lines · Ingo Molnar
Quoted 4 lines · Stephane Eranian

Arnaldo,

Could you revert all the mmap2 patches in your perf/urgent tree for now? Thanks.

I think it would be helpful if you re-sent the final patch, with a full changelog, etc., in a separate thread, not inside an existing discussion and such.

Are you talking about the perf tool or kernel patch?

Quoted 7 lines
Quoted 6 lines

Both.

Quoted 1 line

I did not send any perf-related mmap2 revert patches.

Would be nice to have a single patch that reverts both.

I will send a single patch to remove everything (kernel + tool) then.

Thanks!

Ingo

#15 lore raw
In reply to Stephane Eranian

Em Wed, Oct 09, 2013 at 06:17:43PM +0200, Stephane Eranian escreveu:

Quoted 8 lines

This patch disables the mmap2 record format support for now. The support needs a bit more work to cover VM_CLONE cases.

The patch leaves attr->mmap2 defined, but returns an error if it is set. That ensures no PERF_RECORD_MMAP2 record can be generated. Yet it minimizes the reverts in the perf tool.

Would be interesting to have a simple new tool that would detect all the mmaps that are shared in a given perf.data file, using the PERF_RECORD_MMAP2 information, and that tool should come together with the reworked kernel ABI, showing how it is used.

A 'perf test' entry, based on the test programs you guys are using to test this functionality would be even better, showing that all cases you've been striving to cover are in fact covered by the new ABI.

- Arnaldo

Quoted 53 lines
---
 include/uapi/linux/perf_event.h |   18 +-----------------
 kernel/events/core.c            |    4 ++++
 2 files changed, 5 insertions(+), 17 deletions(-)
include/uapi/linux/perf_event.h +1 -17
index da48837..8074b29 100644
@@ -682,23 +682,7 @@ enum perf_event_type {
 	PERF_RECORD_SAMPLE			= 9,
 
 	/*
-	 * The MMAP2 records are an augmented version of MMAP, they add
-	 * maj, min, ino numbers to be used to uniquely identify each mapping
-	 *
-	 * struct {
-	 *	struct perf_event_header	header;
-	 *
-	 *	u32				pid, tid;
-	 *	u64				addr;
-	 *	u64				len;
-	 *	u64				pgoff;
-	 *	u32				maj;
-	 *	u32				min;
-	 *	u64				ino;
-	 *	u64				ino_generation;
-	 *	char				filename[];
-	 * 	struct sample_id		sample_id;
-	 * };
+	 * reserved, do not use
 	 */
 	PERF_RECORD_MMAP2			= 10,
 
kernel/events/core.c +4 -0
index c716385..d7152a6 100644
@@ -6782,6 +6782,10 @@ static int perf_copy_attr(struct perf_event_attr __user *uattr,
 	if (attr->read_format & ~(PERF_FORMAT_MAX-1))
 		return -EINVAL;
 
+	/* disable for now */
+	if (attr->mmap2)
+		return -EINVAL;
+
 	if (attr->sample_type & PERF_SAMPLE_BRANCH_STACK) {
 		u64 mask = attr->branch_sample_type;
 
Signature · 1 line
1.7.10.4
#16 lore raw
Quoted 5 lines · Arnaldo Carvalho de Melo

Em Wed, Oct 09, 2013 at 06:17:43PM +0200, Stephane Eranian escreveu:

Quoted 3 lines

This patch disables the mmap2 record format support for now. The support needs a bit more work to cover VM_CLONE cases.

Quoted 8 lines
Quoted 3 lines

The patch leaves attr->mmap2 defined, but returns an error if it is set. That ensures no PERF_RECORD_MMAP2 record can be generated. Yet it minimizes the reverts in the perf tool.

Would be interesting to have a simple new tool that would detect all the mmaps that are shared in a given perf.data file, using the PERF_RECORD_MMAP2 information, and that tool should come together with the reworked kernel ABI, showing how it is used.

Indeed. Let's make that a requirement for the next round of mmap2 patches.

Issues like this is really why I hate not getting a fully functional usecase together with a new ABI - it's _so_ easy to come up with something subtly (or not so subtly) incomplete.

Quoted 3 lines

A 'perf test' entry, based on the test programs you guys are using to test this functionality would be even better, showing that all cases you've been striving to cover are in fact covered by the new ABI.

That too, in addition to an actual tool that people can use.

Thanks,

Ingo