perf: add ability to sample interrupted machine state

Enable capture of interrupted machine state for each sample.

Registers to sample are passed per event in the sample_regs_intr bitmask.

To sample interrupt machine state, the PERF_SAMPLE_INTR_REGS must be passed in sample_type.

The list of available registers is arch dependent and provided by asm/perf_regs.h

Registers are laid out as u64 in the order of the bit order of sample_intr_regs.

This patch also adds a new ABI version PERF_ATTR_SIZE_VER4 because we extend the perf_event_attr struct with a new u64 field.

Reviewed-by Jiri Olsa <[email protected]>

Reviewed-by Andi Kleen <[email protected]>

Signed-off-by Stephane Eranian <[email protected]>

---
 include/linux/perf_event.h      |    7 ++++--
 include/uapi/linux/perf_event.h |   15 ++++++++++++-
 kernel/events/core.c            |   46 +++++++++++++++++++++++++++++++++++++--
 3 files changed, 63 insertions(+), 5 deletions(-)

include/linux/perf_event.h +5 -2

index 893a0d0..68d46d5 100644
@@ -79,7 +79,7 @@ struct perf_branch_stack {
 	struct perf_branch_entry	entries[0];
 };
 
-struct perf_regs_user {
+struct perf_regs {
 	__u64		abi;
 	struct pt_regs	*regs;
 };

@@ -600,7 +600,8 @@ struct perf_sample_data {
 	struct perf_callchain_entry	*callchain;
 	struct perf_raw_record		*raw;
 	struct perf_branch_stack	*br_stack;
-	struct perf_regs_user		regs_user;
+	struct perf_regs		regs_user;
+	struct perf_regs		regs_intr;
 	u64				stack_user_size;
 	u64				weight;
 	/*

@@ -630,6 +631,8 @@ static inline void perf_sample_data_init(struct perf_sample_data *data,
 	data->weight = 0;
 	data->data_src.val = PERF_MEM_NA;
 	data->txn = 0;
+	data->regs_intr.abi = PERF_SAMPLE_REGS_ABI_NONE;
+	data->regs_intr.regs = NULL;
 }
 
 extern void perf_output_sample(struct perf_output_handle *handle,

include/uapi/linux/perf_event.h +14 -1

index 9269de2..48d4a01 100644
@@ -137,8 +137,9 @@ enum perf_event_sample_format {
 	PERF_SAMPLE_DATA_SRC			= 1U << 15,
 	PERF_SAMPLE_IDENTIFIER			= 1U << 16,
 	PERF_SAMPLE_TRANSACTION			= 1U << 17,
+	PERF_SAMPLE_REGS_INTR			= 1U << 18,
 
-	PERF_SAMPLE_MAX = 1U << 18,		/* non-ABI */
+	PERF_SAMPLE_MAX = 1U << 19,		/* non-ABI */
 };
 
 /*

@@ -238,6 +239,7 @@ enum perf_event_read_format {
 #define PERF_ATTR_SIZE_VER2	80	/* add: branch_sample_type */
 #define PERF_ATTR_SIZE_VER3	96	/* add: sample_regs_user */
 					/* add: sample_stack_user */
+#define PERF_ATTR_SIZE_VER4	104	/* add: sample_regs_intr */
 
 /*
  * Hardware event_id to monitor via a performance monitoring event:

@@ -334,6 +336,15 @@ struct perf_event_attr {
 
 	/* Align to u64. */
 	__u32	__reserved_2;
+	/*
+	 * Defines set of regs to dump for each sample
+	 * state captured on:
+	 *  - precise = 0: PMU interrupt
+	 *  - precise > 0: sampled instruction
+	 *
+	 * See asm/perf_regs.h for details.
+	 */
+	__u64	sample_regs_intr;
 };
 
 #define perf_flags(attr)	(*(&(attr)->read_format + 1))

@@ -686,6 +697,8 @@ enum perf_event_type {
 	 *	{ u64			weight;   } && PERF_SAMPLE_WEIGHT
 	 *	{ u64			data_src; } && PERF_SAMPLE_DATA_SRC
 	 *	{ u64			transaction; } && PERF_SAMPLE_TRANSACTION
+	 *	{ u64			abi; # enum perf_sample_regs_abi
+	 *	  u64			regs[weight(mask)]; } && PERF_SAMPLE_REGS_INTR
 	 * };
 	 */
 	PERF_RECORD_SAMPLE			= 9,

kernel/events/core.c +44 -2

index eaa636e..7941343 100644
@@ -4430,7 +4430,7 @@ perf_output_sample_regs(struct perf_output_handle *handle,
 	}
 }
 
-static void perf_sample_regs_user(struct perf_regs_user *regs_user,
+static void perf_sample_regs_user(struct perf_regs *regs_user,
 				  struct pt_regs *regs)
 {
 	if (!user_mode(regs)) {

@@ -4446,6 +4446,14 @@ static void perf_sample_regs_user(struct perf_regs_user *regs_user,
 	}
 }
 
+static void perf_sample_regs_intr(struct perf_regs *regs_intr,
+				  struct pt_regs *regs)
+{
+	regs_intr->regs = regs;
+	regs_intr->abi  = perf_reg_abi(current);
+}
+
+
 /*
  * Get remaining task size from user stack pointer.
  *

@@ -4827,6 +4835,23 @@ void perf_output_sample(struct perf_output_handle *handle,
 	if (sample_type & PERF_SAMPLE_TRANSACTION)
 		perf_output_put(handle, data->txn);
 
+	if (sample_type & PERF_SAMPLE_REGS_INTR) {
+		u64 abi = data->regs_intr.abi;
+		/*
+		 * If there are no regs to dump, notice it through
+		 * first u64 being zero (PERF_SAMPLE_REGS_ABI_NONE).
+		 */
+		perf_output_put(handle, abi);
+
+		if (abi) {
+			u64 mask = event->attr.sample_regs_intr;
+
+			perf_output_sample_regs(handle,
+						data->regs_intr.regs,
+						mask);
+		}
+	}
+
 	if (!event->attr.watermark) {
 		int wakeup_events = event->attr.wakeup_events;
 

@@ -4913,7 +4938,7 @@ void perf_prepare_sample(struct perf_event_header *header,
 		 * in case new sample type is added, because we could eat
 		 * up the rest of the sample size.
 		 */
-		struct perf_regs_user *uregs = &data->regs_user;
+		struct perf_regs *uregs = &data->regs_user;
 		u16 stack_size = event->attr.sample_stack_user;
 		u16 size = sizeof(u64);
 

@@ -4934,6 +4959,21 @@ void perf_prepare_sample(struct perf_event_header *header,
 		data->stack_user_size = stack_size;
 		header->size += size;
 	}
+
+	if (sample_type & PERF_SAMPLE_REGS_INTR) {
+		/* regs dump ABI info */
+		int size = sizeof(u64);
+
+		perf_sample_regs_intr(&data->regs_intr, regs);
+
+		if (data->regs_intr.regs) {
+			u64 mask = event->attr.sample_regs_intr;
+
+			size += hweight64(mask) * sizeof(u64);
+		}
+
+		header->size += size;
+	}
 }
 
 static void perf_event_output(struct perf_event *event,

@@ -7134,6 +7174,8 @@ static int perf_copy_attr(struct perf_event_attr __user *uattr,
 			ret = -EINVAL;
 	}
 
+	if (attr->sample_type & PERF_SAMPLE_REGS_INTR)
+		ret = perf_reg_validate(attr->sample_regs_intr);
 out:
 	return ret;
 

Signature · 1 line

1.7.9.5

Em Wed, Sep 24, 2014 at 01:48:37PM +0200, Stephane Eranian escreveu:

Quoted 20 lines

Enable capture of interrupted machine state for each sample.

Registers to sample are passed per event in the sample_regs_intr bitmask.

To sample interrupt machine state, the PERF_SAMPLE_INTR_REGS must be passed in sample_type.

The list of available registers is arch dependent and provided by asm/perf_regs.h

Registers are laid out as u64 in the order of the bit order of sample_intr_regs.

This patch also adds a new ABI version PERF_ATTR_SIZE_VER4 because we extend the perf_event_attr struct with a new u64 field.

So, trying to bisect a problem with how the TUI hist_entries browser renders callchains I got stuck with:

[root@zoo acme]# perf report
incompatible file format (rerun with -v to learn more)

[root@zoo acme]# perf report -v
file uses a more recent and unsupported ABI (8 bytes extra)

Because the perf.data file was generated with HEAD of perf/core and probably the above warning comes from a simple check against the ABI version...

I wonder if we can't just check that there are no sample_types that we don't know and if not, just process the file anyway, i.e. the sample will be parseable, no?

For now, in this case, I'll just regenerate the perf.data file with an older tool...

- Arnaldo

Quoted 183 lines

Reviewed-by Jiri Olsa <[email protected]>

Reviewed-by Andi Kleen <[email protected]>

Signed-off-by Stephane Eranian <[email protected]>

---
 include/linux/perf_event.h      |    7 ++++--
 include/uapi/linux/perf_event.h |   15 ++++++++++++-
 kernel/events/core.c            |   46 +++++++++++++++++++++++++++++++++++++--
 3 files changed, 63 insertions(+), 5 deletions(-)

include/linux/perf_event.h +5 -2

index 893a0d0..68d46d5 100644
@@ -79,7 +79,7 @@ struct perf_branch_stack {
 	struct perf_branch_entry	entries[0];
 };
 
-struct perf_regs_user {
+struct perf_regs {
 	__u64		abi;
 	struct pt_regs	*regs;
 };

@@ -600,7 +600,8 @@ struct perf_sample_data {
 	struct perf_callchain_entry	*callchain;
 	struct perf_raw_record		*raw;
 	struct perf_branch_stack	*br_stack;
-	struct perf_regs_user		regs_user;
+	struct perf_regs		regs_user;
+	struct perf_regs		regs_intr;
 	u64				stack_user_size;
 	u64				weight;
 	/*

@@ -630,6 +631,8 @@ static inline void perf_sample_data_init(struct perf_sample_data *data,
 	data->weight = 0;
 	data->data_src.val = PERF_MEM_NA;
 	data->txn = 0;
+	data->regs_intr.abi = PERF_SAMPLE_REGS_ABI_NONE;
+	data->regs_intr.regs = NULL;
 }
 
 extern void perf_output_sample(struct perf_output_handle *handle,

include/uapi/linux/perf_event.h +14 -1

index 9269de2..48d4a01 100644
@@ -137,8 +137,9 @@ enum perf_event_sample_format {
 	PERF_SAMPLE_DATA_SRC			= 1U << 15,
 	PERF_SAMPLE_IDENTIFIER			= 1U << 16,
 	PERF_SAMPLE_TRANSACTION			= 1U << 17,
+	PERF_SAMPLE_REGS_INTR			= 1U << 18,
 
-	PERF_SAMPLE_MAX = 1U << 18,		/* non-ABI */
+	PERF_SAMPLE_MAX = 1U << 19,		/* non-ABI */
 };
 
 /*

@@ -238,6 +239,7 @@ enum perf_event_read_format {
 #define PERF_ATTR_SIZE_VER2	80	/* add: branch_sample_type */
 #define PERF_ATTR_SIZE_VER3	96	/* add: sample_regs_user */
 					/* add: sample_stack_user */
+#define PERF_ATTR_SIZE_VER4	104	/* add: sample_regs_intr */
 
 /*
  * Hardware event_id to monitor via a performance monitoring event:

@@ -334,6 +336,15 @@ struct perf_event_attr {
 
 	/* Align to u64. */
 	__u32	__reserved_2;
+	/*
+	 * Defines set of regs to dump for each sample
+	 * state captured on:
+	 *  - precise = 0: PMU interrupt
+	 *  - precise > 0: sampled instruction
+	 *
+	 * See asm/perf_regs.h for details.
+	 */
+	__u64	sample_regs_intr;
 };
 
 #define perf_flags(attr)	(*(&(attr)->read_format + 1))

@@ -686,6 +697,8 @@ enum perf_event_type {
 	 *	{ u64			weight;   } && PERF_SAMPLE_WEIGHT
 	 *	{ u64			data_src; } && PERF_SAMPLE_DATA_SRC
 	 *	{ u64			transaction; } && PERF_SAMPLE_TRANSACTION
+	 *	{ u64			abi; # enum perf_sample_regs_abi
+	 *	  u64			regs[weight(mask)]; } && PERF_SAMPLE_REGS_INTR
 	 * };
 	 */
 	PERF_RECORD_SAMPLE			= 9,

kernel/events/core.c +44 -2

index eaa636e..7941343 100644
@@ -4430,7 +4430,7 @@ perf_output_sample_regs(struct perf_output_handle *handle,
 	}
 }
 
-static void perf_sample_regs_user(struct perf_regs_user *regs_user,
+static void perf_sample_regs_user(struct perf_regs *regs_user,
 				  struct pt_regs *regs)
 {
 	if (!user_mode(regs)) {

@@ -4446,6 +4446,14 @@ static void perf_sample_regs_user(struct perf_regs_user *regs_user,
 	}
 }
 
+static void perf_sample_regs_intr(struct perf_regs *regs_intr,
+				  struct pt_regs *regs)
+{
+	regs_intr->regs = regs;
+	regs_intr->abi  = perf_reg_abi(current);
+}
+
+
 /*
  * Get remaining task size from user stack pointer.
  *

@@ -4827,6 +4835,23 @@ void perf_output_sample(struct perf_output_handle *handle,
 	if (sample_type & PERF_SAMPLE_TRANSACTION)
 		perf_output_put(handle, data->txn);
 
+	if (sample_type & PERF_SAMPLE_REGS_INTR) {
+		u64 abi = data->regs_intr.abi;
+		/*
+		 * If there are no regs to dump, notice it through
+		 * first u64 being zero (PERF_SAMPLE_REGS_ABI_NONE).
+		 */
+		perf_output_put(handle, abi);
+
+		if (abi) {
+			u64 mask = event->attr.sample_regs_intr;
+
+			perf_output_sample_regs(handle,
+						data->regs_intr.regs,
+						mask);
+		}
+	}
+
 	if (!event->attr.watermark) {
 		int wakeup_events = event->attr.wakeup_events;
 

@@ -4913,7 +4938,7 @@ void perf_prepare_sample(struct perf_event_header *header,
 		 * in case new sample type is added, because we could eat
 		 * up the rest of the sample size.
 		 */
-		struct perf_regs_user *uregs = &data->regs_user;
+		struct perf_regs *uregs = &data->regs_user;
 		u16 stack_size = event->attr.sample_stack_user;
 		u16 size = sizeof(u64);
 

@@ -4934,6 +4959,21 @@ void perf_prepare_sample(struct perf_event_header *header,
 		data->stack_user_size = stack_size;
 		header->size += size;
 	}
+
+	if (sample_type & PERF_SAMPLE_REGS_INTR) {
+		/* regs dump ABI info */
+		int size = sizeof(u64);
+
+		perf_sample_regs_intr(&data->regs_intr, regs);
+
+		if (data->regs_intr.regs) {
+			u64 mask = event->attr.sample_regs_intr;
+
+			size += hweight64(mask) * sizeof(u64);
+		}
+
+		header->size += size;
+	}
 }
 
 static void perf_event_output(struct perf_event *event,

@@ -7134,6 +7174,8 @@ static int perf_copy_attr(struct perf_event_attr __user *uattr,
 			ret = -EINVAL;
 	}
 
+	if (attr->sample_type & PERF_SAMPLE_REGS_INTR)
+		ret = perf_reg_validate(attr->sample_regs_intr);
 out:
 	return ret;
 

Signature · 1 line

1.7.9.5

Em Fri, Nov 21, 2014 at 07:26:31PM -0200, Arnaldo Carvalho de Melo escreveu:

Quoted 5 lines

Em Wed, Sep 24, 2014 at 01:48:37PM +0200, Stephane Eranian escreveu:

Quoted 4 lines

This patch also adds a new ABI version
PERF_ATTR_SIZE_VER4 because we extend
the perf_event_attr struct with a new u64
field.

I think this problem is also related to this changeset:

[root@zoo ~]# perf test -v 15
15: struct perf_event_attr setup :

running '/home/acme/libexec/perf-core/tests/attr/test-stat-default' expected size=96, got 104
FAILED '/home/acme/libexec/perf-core/tests/attr/test-stat-default' - match failure
test child finished with 0
---- end ----
struct perf_event_attr setup: Ok
[root@zoo ~]# uname -r
3.17.3-200.fc20.x86_64
[root@zoo ~]#

Checking if this is just a matter of updating the test entry.

Em Tue, Dec 09, 2014 at 11:30:31AM -0200, Arnaldo Carvalho de Melo escreveu:

Quoted 11 lines

Em Fri, Nov 21, 2014 at 07:26:31PM -0200, Arnaldo Carvalho de Melo escreveu:

Quoted 5 lines

Em Wed, Sep 24, 2014 at 01:48:37PM +0200, Stephane Eranian escreveu:

Quoted 4 lines

This patch also adds a new ABI version
PERF_ATTR_SIZE_VER4 because we extend
the perf_event_attr struct with a new u64
field.

I think this problem is also related to this changeset:

[root@zoo ~]# perf test -v 15
15: struct perf_event_attr setup :

<SNIP>

Quoted 11 lines

running '/home/acme/libexec/perf-core/tests/attr/test-stat-default' expected size=96, got 104
FAILED '/home/acme/libexec/perf-core/tests/attr/test-stat-default' - match failure
test child finished with 0
---- end ----
struct perf_event_attr setup: Ok
[root@zoo ~]# uname -r
3.17.3-200.fc20.x86_64
[root@zoo ~]#

Quoted 1 line

Checking if this is just a matter of updating the test entry.

Well, here I think that size variable needs to somehow be tested against the value of another field, the ABI one, so that for each ABI we test against the rightsize, that is, after this cset, 8 bytes (sizeof u64) bigger, Jiri?

Quoted 26 lines

Em Tue, Dec 09, 2014 at 11:30:31AM -0200, Arnaldo Carvalho de Melo escreveu:

Quoted 11 lines

Em Fri, Nov 21, 2014 at 07:26:31PM -0200, Arnaldo Carvalho de Melo escreveu:

Quoted 5 lines

Em Wed, Sep 24, 2014 at 01:48:37PM +0200, Stephane Eranian escreveu:

Quoted 4 lines

This patch also adds a new ABI version
PERF_ATTR_SIZE_VER4 because we extend
the perf_event_attr struct with a new u64
field.

I think this problem is also related to this changeset:

[root@zoo ~]# perf test -v 15
15: struct perf_event_attr setup :

<SNIP>

Quoted 12 lines

running '/home/acme/libexec/perf-core/tests/attr/test-stat-default' expected size=96, got 104
FAILED '/home/acme/libexec/perf-core/tests/attr/test-stat-default' - match failure
test child finished with 0
---- end ----
struct perf_event_attr setup: Ok
[root@zoo ~]# uname -r
3.17.3-200.fc20.x86_64
[root@zoo ~]#

Checking if this is just a matter of updating the test entry.

Quoted 4 lines

Well, here I think that size variable needs to somehow be tested against the value of another field, the ABI one, so that for each ABI we test against the rightsize, that is, after this cset, 8 bytes (sizeof u64) bigger, Jiri?

well this patch enlarged the perf_event_attr so the test needs to be adjusted like in patch below

[jolsa@krava perf]$ ./perf test attr -vv
15: struct perf_event_attr setup :
--- start ---
test child forked, pid 9719
running './tests/attr/test-stat-group1'
'PERF_TEST_ATTR=/tmp/tmp4drvul ./perf stat -o /tmp/tmp4drvul/perf.data -e '{cycles,instructions}' kill >/dev/null 2>&1' ret 1
expected size=96, got 104
FAILED './tests/attr/test-stat-group1' - match failure

jirka

---
Following change adjusted 'struct perf_event_attr', but let the attr test's sizes untouched:
60e2364e60e8 perf: Add ability to sample machine state on interrupt

Adjusting test size values for attr test.

Signed-off-by Jiri Olsa <[email protected]>

---

tools/perf/tests/attr/base-record +1 -1

index f710b92ccff6..d3095dafed36 100644
@@ -5,7 +5,7 @@ group_fd=-1
 flags=0|8
 cpu=*
 type=0|1
-size=96
+size=104
 config=0
 sample_period=4000
 sample_type=263

tools/perf/tests/attr/base-stat +1 -1

index dc3ada2470c0..872ed7e24c7c 100644
@@ -5,7 +5,7 @@ group_fd=-1
 flags=0|8
 cpu=*
 type=0
-size=96
+size=104
 config=0
 sample_period=0
 sample_type=0

These do indeed appear to compile just fine, thanks!

Arnaldo, may I gently prod you to pass judgement over the tools bits?

Peter,

Quoted 4 lines · Peter Zijlstra

These do indeed appear to compile just fine, thanks!

I posted v7 with the x86_32 fixes and the eflags changes. Is that what you compiled?

Quoted 5 lines · Stephane Eranian

Peter,

Quoted 2 lines · Peter Zijlstra

Quoted 4 lines

Quoted 2 lines

These do indeed appear to compile just fine, thanks!

I posted v7 with the x86_32 fixes and the eflags changes. Is that what you compiled?

Yep, I grabbed the latest. I've not yet tried the x86_32 compile, but I did check the 'offending' patch and it has the #ifdef guard in place so I'll assume it works for now.