Skip to contentSkip to conversation

perf/x86/amd: AMD Family 16h Data Breakpoint Extensions

PATCH In review Jacob Shin 5 participants 9 messages Apr 28, 2013 – Oct 1, 2013

Subsystems (2)
lore
Share & tools
lore

Jacob Shin

author
#0 lore raw

v4 cover perf/x86/amd: AMD Family 16h Data Breakpoint Extensions

Oleg, Ingo, this is my final push for 3.10. I understand that it might be too late for that .. if so I'll try again later for 3.11.

The following patchset enables hardware breakpoint bp_len greater than HW_BREAKPOINT_LEN_8 on AMD Family 16h and later.

  $ perf stat -e mem:0x1000/16:w a.out
                            ^^
                            bp_len

Will count writes to [0x1000 ~ 0x1010)

V4:
Even more per Oleg's suggestion:
* Further simplify info->len and info->mask setting switch statement

V3:
More per Oleg's suggestion:
* Use already existing bp_len instead of changing userland API and in kernel turn bp_len into proper AMD hardware breakpoint address mask.

V2:
Per Oleg's suggestions:
* Moved testing of bp_addr_mask to validate_hw_breakpoint()
* Changed perf tool syntax to mem:<addr>[/addr_mask][:access]

Jacob Shin (3):
perf/x86/amd: AMD support for bp_len > HW_BREAKPOINT_LEN_8 perf tools: allow user to specify hardware breakpoint bp_len perf tools: add hardware breakpoint bp_len test cases

 arch/x86/include/asm/cpufeature.h        |    2 ++
 arch/x86/include/asm/debugreg.h          |    5 +++
 arch/x86/include/asm/hw_breakpoint.h     |    1 +
 arch/x86/include/uapi/asm/msr-index.h    |    4 +++
 arch/x86/kernel/cpu/amd.c                |   19 +++++++++++
 arch/x86/kernel/hw_breakpoint.c          |   47 ++++++++++---------------
 tools/perf/Documentation/perf-record.txt |    7 ++--
 tools/perf/tests/parse-events.c          |   55 ++++++++++++++++++++++++++++++
 tools/perf/util/parse-events.c           |   17 ++++-----
 tools/perf/util/parse-events.h           |    2 +-
 tools/perf/util/parse-events.l           |    1 +
 tools/perf/util/parse-events.y           |   24 +++++++++++--
 12 files changed, 140 insertions(+), 44 deletions(-)
Signature · 1 line
1.7.9.5
author
#1 lore raw

PATCH v4 1/3 perf/x86/amd: AMD support for bp_len > HW_BREAKPOINT_LEN_8

Implement hardware breakpoint address mask for AMD Family 16h and above processors. CPUID feature bit indicates hardware support for DRn_ADDR_MASK MSRs. These masks further qualify DRn/DR7 hardware breakpoint addresses to allow matching of larger addresses ranges.

Valuable advice and pseudo code from Oleg Nesterov <[email protected]>

---
 arch/x86/include/asm/cpufeature.h     |    2 ++
 arch/x86/include/asm/debugreg.h       |    5 ++++
 arch/x86/include/asm/hw_breakpoint.h  |    1 +
 arch/x86/include/uapi/asm/msr-index.h |    4 +++
 arch/x86/kernel/cpu/amd.c             |   19 +++++++++++++
 arch/x86/kernel/hw_breakpoint.c       |   47 +++++++++++++--------------------
 6 files changed, 49 insertions(+), 29 deletions(-)
arch/x86/include/asm/cpufeature.h +2 -0
index ac10df7..bee6994 100644
@@ -168,6 +168,7 @@
 #define X86_FEATURE_TOPOEXT	(6*32+22) /* topology extensions CPUID leafs */
 #define X86_FEATURE_PERFCTR_CORE (6*32+23) /* core performance counter extensions */
 #define X86_FEATURE_PERFCTR_NB  (6*32+24) /* NB performance counter extensions */
+#define X86_FEATURE_BPEXT	(6*32+26) /* data breakpoint extension */
 #define X86_FEATURE_PERFCTR_L2	(6*32+28) /* L2 performance counter extensions */
 
 /*
@@ -317,6 +318,7 @@ extern const char * const x86_power_flags[32];
 #define cpu_has_cx16		boot_cpu_has(X86_FEATURE_CX16)
 #define cpu_has_eager_fpu	boot_cpu_has(X86_FEATURE_EAGER_FPU)
 #define cpu_has_topoext		boot_cpu_has(X86_FEATURE_TOPOEXT)
+#define cpu_has_bpext		boot_cpu_has(X86_FEATURE_BPEXT)
 
 #ifdef CONFIG_X86_64
 
arch/x86/include/asm/debugreg.h +5 -0
index 4b528a9..145b009 100644
@@ -114,5 +114,10 @@ static inline void debug_stack_usage_inc(void) { }
 static inline void debug_stack_usage_dec(void) { }
 #endif /* X86_64 */
 
+#ifdef CONFIG_CPU_SUP_AMD
+extern void set_dr_addr_mask(unsigned long mask, int dr);
+#else
+static inline void set_dr_addr_mask(unsigned long mask, int dr) { }
+#endif
 
 #endif /* _ASM_X86_DEBUGREG_H */
arch/x86/include/asm/hw_breakpoint.h +1 -0
index ef1c4d2..6c98be8 100644
@@ -12,6 +12,7 @@
  */
 struct arch_hw_breakpoint {
 	unsigned long	address;
+	unsigned long	mask;
 	u8		len;
 	u8		type;
 };
arch/x86/include/uapi/asm/msr-index.h +4 -0
index b575788..dac1d3b 100644
@@ -199,6 +199,10 @@
 /* Fam 16h MSRs */
 #define MSR_F16H_L2I_PERF_CTL		0xc0010230
 #define MSR_F16H_L2I_PERF_CTR		0xc0010231
+#define MSR_F16H_DR1_ADDR_MASK		0xc0011019
+#define MSR_F16H_DR2_ADDR_MASK		0xc001101a
+#define MSR_F16H_DR3_ADDR_MASK		0xc001101b
+#define MSR_F16H_DR0_ADDR_MASK		0xc0011027
 
 /* Fam 15h MSRs */
 #define MSR_F15H_PERF_CTL		0xc0010200
arch/x86/kernel/cpu/amd.c +19 -0
index fa96eb0..2c3f12e 100644
@@ -910,3 +910,22 @@ bool cpu_has_amd_erratum(const int *erratum)
 }
 
 EXPORT_SYMBOL_GPL(cpu_has_amd_erratum);
+
+void set_dr_addr_mask(unsigned long mask, int dr)
+{
+	if (!cpu_has_bpext)
+		return;
+
+	switch (dr) {
+	case 0:
+		wrmsr(MSR_F16H_DR0_ADDR_MASK, mask, 0);
+		break;
+	case 1:
+	case 2:
+	case 3:
+		wrmsr(MSR_F16H_DR1_ADDR_MASK - 1 + dr, mask, 0);
+		break;
+	default:
+		break;
+	}
+}
arch/x86/kernel/hw_breakpoint.c +18 -29

97 lines not on this page · raw message

Signature · 1 line
1.7.9.5
author
#2 lore raw

PATCH v4 2/3 perf tools: allow user to specify hardware breakpoint bp_len

In reply to Jacob Shin

Currently bp_len is given a default value of 4. Allow user to override it:

  $ perf stat -e mem:0x1000/8
                            ^
                            bp_len

If no value is given, it will default to 4 as it did before.

---
 tools/perf/Documentation/perf-record.txt |    7 +++++--
 tools/perf/util/parse-events.c           |   17 +++++++----------
 tools/perf/util/parse-events.h           |    2 +-
 tools/perf/util/parse-events.l           |    1 +
 tools/perf/util/parse-events.y           |   24 ++++++++++++++++++++++--
 5 files changed, 36 insertions(+), 15 deletions(-)
tools/perf/Documentation/perf-record.txt +5 -2
index d4da111..d05973b 100644
@@ -33,12 +33,15 @@ OPTIONS
         - a raw PMU event (eventsel+umask) in the form of rNNN where NNN is a
 	  hexadecimal event descriptor.
 
-        - a hardware breakpoint event in the form of '\mem:addr[:access]'
+        - a hardware breakpoint event in the form of '\mem:addr[/len][:access]'
           where addr is the address in memory you want to break in.
           Access is the memory access type (read, write, execute) it can
-          be passed as follows: '\mem:addr[:[r][w][x]]'.
+          be passed as follows: '\mem:addr[:[r][w][x]]'. len is the range,
+          number of bytes from specified addr, which the breakpoint will cover.
           If you want to profile read-write accesses in 0x1000, just set
           'mem:0x1000:rw'.
+          If you want to profile write accesses in [0x1000~1008), just set
+          'mem:0x1000/8:w'.
 
 --filter=<filter>::
         Event filter.
tools/perf/util/parse-events.c +7 -10
index 6c8bb0f..f101820 100644
@@ -508,7 +508,7 @@ do {					\
 }
 
 int parse_events_add_breakpoint(struct list_head **list, int *idx,
-				void *ptr, char *type)
+				void *ptr, char *type, u64 len)
 {
 	struct perf_event_attr attr;
 
@@ -518,14 +518,11 @@ int parse_events_add_breakpoint(struct list_head **list, int *idx,
 	if (parse_breakpoint_type(type, &attr))
 		return -EINVAL;
 
-	/*
-	 * We should find a nice way to override the access length
-	 * Provide some defaults for now
-	 */
-	if (attr.bp_type == HW_BREAKPOINT_X)
-		attr.bp_len = sizeof(long);
-	else
-		attr.bp_len = HW_BREAKPOINT_LEN_4;
+	/* Provide some defaults if len is not specified */
+	if (!len)
+		len = attr.bp_type == HW_BREAKPOINT_X ? sizeof(long) :
+							HW_BREAKPOINT_LEN_4;
+	attr.bp_len = len;
 
 	attr.type = PERF_TYPE_BREAKPOINT;
 	attr.sample_period = 1;
@@ -1147,7 +1144,7 @@ void print_events(const char *event_glob, bool name_only)
 		printf("\n");
 
 		printf("  %-50s [%s]\n",
-		       "mem:<addr>[:access]",
+		       "mem:<addr>[/len][:access]",
 			event_type_descriptors[PERF_TYPE_BREAKPOINT]);
 		printf("\n");
 	}
tools/perf/util/parse-events.h +1 -1
index 8a48593..8655b83 100644
@@ -92,7 +92,7 @@ int parse_events_add_numeric(struct list_head **list, int *idx,
 int parse_events_add_cache(struct list_head **list, int *idx,
 			   char *type, char *op_result1, char *op_result2);
 int parse_events_add_breakpoint(struct list_head **list, int *idx,
-				void *ptr, char *type);
+				void *ptr, char *type, u64 len);
 int parse_events_add_pmu(struct list_head **list, int *idx,
 			 char *pmu , struct list_head *head_config);
 void parse_events__set_leader(char *name, struct list_head *list);
tools/perf/util/parse-events.l +1 -0
index e9d1134..15b3efc 100644
@@ -191,6 +191,7 @@ r{num_raw_hex}		{ return raw(yyscanner); }
 <mem>{
 {modifier_bp}		{ return str(yyscanner, PE_MODIFIER_BP); }
 :			{ return ':'; }
+"/"			{ return '/'; }
 {num_dec}		{ return value(yyscanner, 10); }
 {num_hex}		{ return value(yyscanner, 16); }
 	/*
tools/perf/util/parse-events.y +22 -2
index afc44c1..1751fb1 100644
@@ -263,13 +263,33 @@ PE_NAME_CACHE_TYPE
 }
 
 event_legacy_mem:
+PE_PREFIX_MEM PE_VALUE '/' PE_VALUE ':' PE_MODIFIER_BP sep_dc
+{
+	struct parse_events_evlist *data = _data;
+	struct list_head *list = NULL;
+
+	ABORT_ON(parse_events_add_breakpoint(&list, &data->idx,
+					     (void *) $2, $6, $4));
+	$$ = list;
+}
+|
+PE_PREFIX_MEM PE_VALUE '/' PE_VALUE sep_dc
+{
+	struct parse_events_evlist *data = _data;
+	struct list_head *list = NULL;
+
+	ABORT_ON(parse_events_add_breakpoint(&list, &data->idx,
+					     (void *) $2, NULL, $4));
+	$$ = list;
+}
+|
 PE_PREFIX_MEM PE_VALUE ':' PE_MODIFIER_BP sep_dc
 {
 	struct parse_events_evlist *data = _data;
 	struct list_head *list = NULL;
 
 	ABORT_ON(parse_events_add_breakpoint(&list, &data->idx,
-					     (void *) $2, $4));
+					     (void *) $2, $4, 0));
 	$$ = list;
 }
 |
@@ -279,7 +299,7 @@ PE_PREFIX_MEM PE_VALUE sep_dc
 	struct list_head *list = NULL;
 
 	ABORT_ON(parse_events_add_breakpoint(&list, &data->idx,
-					     (void *) $2, NULL));
+					     (void *) $2, NULL, 0));
 	$$ = list;
 }
 
Signature · 1 line
1.7.9.5
author
#3 lore raw

PATCH v4 3/3 perf tools: add hardware breakpoint bp_len test cases

In reply to Jacob Shin
---
 tools/perf/tests/parse-events.c |   55 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 55 insertions(+)
tools/perf/tests/parse-events.c +55 -0
index 88e2f44..4fd0f96 100644
@@ -971,6 +971,49 @@ static int test__group_gh4(struct perf_evlist *evlist)
 	return 0;
 }
 
+static int test__checkevent_breakpoint_len(struct perf_evlist *evlist)
+{
+	struct perf_evsel *evsel = perf_evlist__first(evlist);
+
+	TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries);
+	TEST_ASSERT_VAL("wrong type", PERF_TYPE_BREAKPOINT == evsel->attr.type);
+	TEST_ASSERT_VAL("wrong config", 0 == evsel->attr.config);
+	TEST_ASSERT_VAL("wrong bp_type", (HW_BREAKPOINT_R | HW_BREAKPOINT_W) ==
+					 evsel->attr.bp_type);
+	TEST_ASSERT_VAL("wrong bp_len", HW_BREAKPOINT_LEN_1 ==
+					evsel->attr.bp_len);
+
+	return 0;
+}
+
+static int test__checkevent_breakpoint_len_w(struct perf_evlist *evlist)
+{
+	struct perf_evsel *evsel = perf_evlist__first(evlist);
+
+	TEST_ASSERT_VAL("wrong number of entries", 1 == evlist->nr_entries);
+	TEST_ASSERT_VAL("wrong type", PERF_TYPE_BREAKPOINT == evsel->attr.type);
+	TEST_ASSERT_VAL("wrong config", 0 == evsel->attr.config);
+	TEST_ASSERT_VAL("wrong bp_type", HW_BREAKPOINT_W ==
+					 evsel->attr.bp_type);
+	TEST_ASSERT_VAL("wrong bp_len", HW_BREAKPOINT_LEN_2 ==
+					evsel->attr.bp_len);
+
+	return 0;
+}
+
+static int
+test__checkevent_breakpoint_len_rw_modifier(struct perf_evlist *evlist)
+{
+	struct perf_evsel *evsel = perf_evlist__first(evlist);
+
+	TEST_ASSERT_VAL("wrong exclude_user", !evsel->attr.exclude_user);
+	TEST_ASSERT_VAL("wrong exclude_kernel", evsel->attr.exclude_kernel);
+	TEST_ASSERT_VAL("wrong exclude_hv", evsel->attr.exclude_hv);
+	TEST_ASSERT_VAL("wrong precise_ip", !evsel->attr.precise_ip);
+
+	return test__checkevent_breakpoint_rw(evlist);
+}
+
 static int count_tracepoints(void)
 {
 	char events_path[PATH_MAX];
@@ -1187,6 +1230,18 @@ static struct evlist_test test__events[] = {
 		.name  = "{cycles:G,cache-misses:H}:uG",
 		.check = test__group_gh4,
 	},
+	[38] = {
+		.name  = "mem:0/1",
+		.check = test__checkevent_breakpoint_len,
+	},
+	[39] = {
+		.name  = "mem:0/2:w",
+		.check = test__checkevent_breakpoint_len_w,
+	},
+	[40] = {
+		.name  = "mem:0/4:rw:u",
+		.check = test__checkevent_breakpoint_len_rw_modifier,
+	},
 };
 
 static struct evlist_test test__events_pmu[] = {
Signature · 1 line
1.7.9.5
#4 lore raw
In reply to Jacob Shin
Quoted 9 lines · Jacob Shin

The following patchset enables hardware breakpoint bp_len greater than HW_BREAKPOINT_LEN_8 on AMD Family 16h and later.

  $ perf stat -e mem:0x1000/16:w a.out
                            ^^
                            bp_len

Will count writes to [0x1000 ~ 0x1010)

Obviously I can't ack the changes in this area, but to me the whole series looks fine.

Afaics this needs more work to actually support a range on executable breakpoint, but I believe this needs another series.

Oleg.

#5 lore raw
Quoted 5 lines · Oleg Nesterov
Quoted 4 lines · Jacob Shin

The following patchset enables hardware breakpoint bp_len greater than HW_BREAKPOINT_LEN_8 on AMD Family 16h and later.

Quoted 8 lines
Quoted 5 lines
  $ perf stat -e mem:0x1000/16:w a.out
                            ^^
                            bp_len

Will count writes to [0x1000 ~ 0x1010)

Obviously I can't ack the changes in this area, but to me the whole series looks fine.

Thanks Oleg - can I add your Reviewed-by tags?

Ingo

Reviewed-by
#6 lore raw
Quoted 7 lines · Ingo Molnar
Quoted 3 lines · Oleg Nesterov

Obviously I can't ack the changes in this area, but to me the whole series looks fine.

Thanks Oleg - can I add your Reviewed-by tags?

Yes, sure, thanks,

#7 lore raw
Quoted 9 lines · Oleg Nesterov
Quoted 4 lines · Ingo Molnar
Quoted 2 lines · Oleg Nesterov

Obviously I can't ack the changes in this area, but to me the whole series looks fine.

Thanks Oleg - can I add your Reviewed-by tags?

Yes, sure, thanks,

Hi All,

I am following up with the status of this patch as I have not seen it merge upstream yet. I was working with Jacob, and I can help following up with any pending issues.

Thank you,

Suravee

#8 lore raw
Quoted 8 lines · Suravee Suthikulpanit
Quoted 7 lines · Oleg Nesterov
Quoted 4 lines · Ingo Molnar
Quoted 2 lines · Oleg Nesterov

Obviously I can't ack the changes in this area, but to me the whole series looks fine.

Thanks Oleg - can I add your Reviewed-by tags?

Yes, sure, thanks,

Quoted 8 lines
Quoted 2 lines

Hi All,

I am following up with the status of this patch as I have not seen it merge upstream yet.
I was working with Jacob, and I can help following up with any pending issues.

Hi Suravee,

I think the best would be to rebase this patchset against the latest version of perf:

git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git
perf/core

And repost it, so we can comment on a fresh and testable basis.

Thanks!