Skip to contentSkip to conversation

x86, perf_counter, bts: incorporate review feedback and fix bugs

PATCH In review [email protected] 1 participant 7 messages Sep 2, 2009 – Sep 4, 2009 x86

lore
Share & tools
lore
#0 lore raw

cover x86, perf_counter, bts: incorporate review feedback and fix bugs

Signature · 14 lines
---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.
#1 lore raw

PATCH 1/3 x86, perf_counter, bts: fail if BTS is not available

Reserve PERF_COUNT_HW_BRANCH_INSTRUCTIONS with sample_period == 1 for BTS tracing and fail, if BTS is not available.

---
arch/x86/kernel/cpu/perf_counter.c | 25 11 + 14 - 0 !

 1 files changed, 11 insertions(+), 14 deletions(-)
arch/x86/kernel/cpu/perf_counter.c +11 -14
Index: b/arch/x86/kernel/cpu/perf_counter.c
===================================================================
@@ -769,7 +769,7 @@ static int reserve_bts_hardware(void)
 	int cpu, err = 0;
 
 	if (!bts_available())
-		return -EOPNOTSUPP;
+		return 0;
 
 	get_online_cpus();
 
@@ -914,7 +914,7 @@ static int __hw_perf_counter_init(struct
 			if (!reserve_pmc_hardware())
 				err = -EBUSY;
 			else
-				reserve_bts_hardware();
+				err = reserve_bts_hardware();
 		}
 		if (!err)
 			atomic_inc(&active_counters);
@@ -979,6 +979,13 @@ static int __hw_perf_counter_init(struct
 	if (config == -1LL)
 		return -EINVAL;
 
+	/*
+	 * Branch tracing:
+	 */
+	if ((attr->config == PERF_COUNT_HW_BRANCH_INSTRUCTIONS) &&
+	    (hwc->sample_period == 1) && !bts_available())
+		return -EOPNOTSUPP;
+
 	hwc->config |= config;
 
 	return 0;
@@ -1355,19 +1362,9 @@ static int x86_pmu_enable(struct perf_co
 
 	idx = fixed_mode_idx(counter, hwc);
 	if (idx == X86_PMC_IDX_FIXED_BTS) {
-		/*
-		 * Try to use BTS for branch tracing. If that is not
-		 * available, try to get a generic counter.
-		 */
-		if (unlikely(!cpuc->ds))
-			goto try_generic;
-
-		/*
-		 * Try to get the fixed counter, if that is already taken
-		 * then try to get a generic counter:
-		 */
+		/* BTS is already occupied. */
 		if (test_and_set_bit(idx, cpuc->used_mask))
-			goto try_generic;
+			return -EAGAIN;
 
 		hwc->config_base	= 0;
 		hwc->counter_base	= 0;

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.

#3 lore raw

PATCH 2/3 x86, perf_counter, bts: correct pointer-to-u64 casts

In reply to [email protected]

On 32bit, pointers in the DS AREA configuration are cast to u64. The current (long) cast to avoid compiler warnings results in a signed 64bit address.

---
arch/x86/kernel/cpu/perf_counter.c | 24 12 + 12 - 0 !

 1 files changed, 12 insertions(+), 12 deletions(-)
arch/x86/kernel/cpu/perf_counter.c +12 -12
Index: b/arch/x86/kernel/cpu/perf_counter.c
===================================================================
@@ -726,7 +726,8 @@ static inline void init_debug_store_on_c
 		return;
 
 	wrmsr_on_cpu(cpu, MSR_IA32_DS_AREA,
-		     (u32)((u64)(long)ds), (u32)((u64)(long)ds >> 32));
+		     (u32)((u64)(unsigned long)ds),
+		     (u32)((u64)(unsigned long)ds >> 32));
 }
 
 static inline void fini_debug_store_on_cpu(int cpu)
@@ -757,7 +758,7 @@ static void release_bts_hardware(void)
 
 		per_cpu(cpu_hw_counters, cpu).ds = NULL;
 
-		kfree((void *)(long)ds->bts_buffer_base);
+		kfree((void *)(unsigned long)ds->bts_buffer_base);
 		kfree(ds);
 	}
 
@@ -788,7 +789,7 @@ static int reserve_bts_hardware(void)
 			break;
 		}
 
-		ds->bts_buffer_base = (u64)(long)buffer;
+		ds->bts_buffer_base = (u64)(unsigned long)buffer;
 		ds->bts_index = ds->bts_buffer_base;
 		ds->bts_absolute_maximum =
 			ds->bts_buffer_base + BTS_BUFFER_SIZE;
@@ -1491,7 +1492,7 @@ static void intel_pmu_drain_bts_buffer(s
 	};
 	struct perf_counter *counter = cpuc->counters[X86_PMC_IDX_FIXED_BTS];
 	unsigned long orig_ip = data->regs->ip;
-	u64 at;
+	struct bts_record *at, *top;
 
 	if (!counter)
 		return;
@@ -1499,19 +1500,18 @@ static void intel_pmu_drain_bts_buffer(s
 	if (!ds)
 		return;
 
-	for (at = ds->bts_buffer_base;
-	     at < ds->bts_index;
-	     at += sizeof(struct bts_record)) {
-		struct bts_record *rec = (struct bts_record *)(long)at;
+	at  = (struct bts_record *)(unsigned long)ds->bts_buffer_base;
+	top = (struct bts_record *)(unsigned long)ds->bts_index;
 
-		data->regs->ip	= rec->from;
-		data->addr	= rec->to;
+	ds->bts_index = ds->bts_buffer_base;
+
+	for (; at < top; at++) {
+		data->regs->ip	= at->from;
+		data->addr	= at->to;
 
 		perf_counter_output(counter, 1, data);
 	}
 
-	ds->bts_index = ds->bts_buffer_base;
-
 	data->regs->ip	= orig_ip;
 	data->addr	= 0;
 

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.

#5 lore raw

PATCH 3/3 x86, perf_counter, bts: do not allow kernel BTS tracing.

In reply to [email protected]

Kernel BTS tracing generates too much data too fast for us to handle, causing the kernel to hang.

Fail for BTS requests for kernel code.

---
arch/x86/kernel/cpu/perf_counter.c | 11 9 + 2 - 0 !

 1 files changed, 9 insertions(+), 2 deletions(-)
arch/x86/kernel/cpu/perf_counter.c +9 -2
Index: b/arch/x86/kernel/cpu/perf_counter.c
===================================================================
@@ -984,8 +984,15 @@ static int __hw_perf_counter_init(struct
 	 * Branch tracing:
 	 */
 	if ((attr->config == PERF_COUNT_HW_BRANCH_INSTRUCTIONS) &&
-	    (hwc->sample_period == 1) && !bts_available())
-		return -EOPNOTSUPP;
+	    (hwc->sample_period == 1)) {
+		/* BTS is not supported by this architecture. */
+		if (!bts_available())
+			return -EOPNOTSUPP;
+
+		/* BTS is currently only allowed for user-mode. */
+		if (hwc->config & ARCH_PERFMON_EVENTSEL_OS)
+			return -EOPNOTSUPP;
+	}
 
 	hwc->config |= config;
 

---------------------------------------------------------------------
Intel GmbH
Dornacher Strasse 1
85622 Feldkirchen/Muenchen Germany
Sitz der Gesellschaft: Feldkirchen bei Muenchen
Geschaeftsfuehrer: Douglas Lusk, Peter Gleissner, Hannes Schwaderer
Registergericht: Muenchen HRB 47456 Ust.-IdNr.
VAT Registration No.: DE129385895
Citibank Frankfurt (BLZ 502 109 00) 600119052

This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.

#2 lore raw

[tip:perfcounters/core] x86, perf_counter, bts: Fail if BTS is not available

In reply to [email protected]
Commit-ID:  747b50aaf728987732e6ff3ba10aba4acc4e0277
Gitweb:     http://git.kernel.org/tip/747b50aaf728987732e6ff3ba10aba4acc4e0277
Author:     [email protected] <[email protected]>
AuthorDate: Wed, 2 Sep 2009 16:04:46 +0200
Committer:  Ingo Molnar <[email protected]>
CommitDate: Fri, 4 Sep 2009 09:26:39 +0200

x86, perf_counter, bts: Fail if BTS is not available

Reserve PERF_COUNT_HW_BRANCH_INSTRUCTIONS with sample_period == 1 for BTS tracing and fail, if BTS is not available.

LKML-Reference: <[email protected]>

---
 arch/x86/kernel/cpu/perf_counter.c |   25 +++++++++++--------------
 1 files changed, 11 insertions(+), 14 deletions(-)
arch/x86/kernel/cpu/perf_counter.c +11 -14
index 396e35d..2f41874 100644
@@ -769,7 +769,7 @@ static int reserve_bts_hardware(void)
 	int cpu, err = 0;
 
 	if (!bts_available())
-		return -EOPNOTSUPP;
+		return 0;
 
 	get_online_cpus();
 
@@ -914,7 +914,7 @@ static int __hw_perf_counter_init(struct perf_counter *counter)
 			if (!reserve_pmc_hardware())
 				err = -EBUSY;
 			else
-				reserve_bts_hardware();
+				err = reserve_bts_hardware();
 		}
 		if (!err)
 			atomic_inc(&active_counters);
@@ -979,6 +979,13 @@ static int __hw_perf_counter_init(struct perf_counter *counter)
 	if (config == -1LL)
 		return -EINVAL;
 
+	/*
+	 * Branch tracing:
+	 */
+	if ((attr->config == PERF_COUNT_HW_BRANCH_INSTRUCTIONS) &&
+	    (hwc->sample_period == 1) && !bts_available())
+		return -EOPNOTSUPP;
+
 	hwc->config |= config;
 
 	return 0;
@@ -1355,19 +1362,9 @@ static int x86_pmu_enable(struct perf_counter *counter)
 
 	idx = fixed_mode_idx(counter, hwc);
 	if (idx == X86_PMC_IDX_FIXED_BTS) {
-		/*
-		 * Try to use BTS for branch tracing. If that is not
-		 * available, try to get a generic counter.
-		 */
-		if (unlikely(!cpuc->ds))
-			goto try_generic;
-
-		/*
-		 * Try to get the fixed counter, if that is already taken
-		 * then try to get a generic counter:
-		 */
+		/* BTS is already occupied. */
 		if (test_and_set_bit(idx, cpuc->used_mask))
-			goto try_generic;
+			return -EAGAIN;
 
 		hwc->config_base	= 0;
 		hwc->counter_base	= 0;
#4 lore raw

[tip:perfcounters/core] x86, perf_counter, bts: Correct pointer-to-u64 casts

In reply to [email protected]
Commit-ID:  596da17f94c103348ebe04129c00d536ea0e80e2
Gitweb:     http://git.kernel.org/tip/596da17f94c103348ebe04129c00d536ea0e80e2
Author:     [email protected] <[email protected]>
AuthorDate: Wed, 2 Sep 2009 16:04:47 +0200
Committer:  Ingo Molnar <[email protected]>
CommitDate: Fri, 4 Sep 2009 09:26:39 +0200

x86, perf_counter, bts: Correct pointer-to-u64 casts

On 32bit, pointers in the DS AREA configuration are cast to u64. The current (long) cast to avoid compiler warnings results in a signed 64bit address.

LKML-Reference: <[email protected]>

---
 arch/x86/kernel/cpu/perf_counter.c |   24 ++++++++++++------------
 1 files changed, 12 insertions(+), 12 deletions(-)
arch/x86/kernel/cpu/perf_counter.c +12 -12
index 2f41874..3776b0b 100644
@@ -726,7 +726,8 @@ static inline void init_debug_store_on_cpu(int cpu)
 		return;
 
 	wrmsr_on_cpu(cpu, MSR_IA32_DS_AREA,
-		     (u32)((u64)(long)ds), (u32)((u64)(long)ds >> 32));
+		     (u32)((u64)(unsigned long)ds),
+		     (u32)((u64)(unsigned long)ds >> 32));
 }
 
 static inline void fini_debug_store_on_cpu(int cpu)
@@ -757,7 +758,7 @@ static void release_bts_hardware(void)
 
 		per_cpu(cpu_hw_counters, cpu).ds = NULL;
 
-		kfree((void *)(long)ds->bts_buffer_base);
+		kfree((void *)(unsigned long)ds->bts_buffer_base);
 		kfree(ds);
 	}
 
@@ -788,7 +789,7 @@ static int reserve_bts_hardware(void)
 			break;
 		}
 
-		ds->bts_buffer_base = (u64)(long)buffer;
+		ds->bts_buffer_base = (u64)(unsigned long)buffer;
 		ds->bts_index = ds->bts_buffer_base;
 		ds->bts_absolute_maximum =
 			ds->bts_buffer_base + BTS_BUFFER_SIZE;
@@ -1491,7 +1492,7 @@ static void intel_pmu_drain_bts_buffer(struct cpu_hw_counters *cpuc,
 	};
 	struct perf_counter *counter = cpuc->counters[X86_PMC_IDX_FIXED_BTS];
 	unsigned long orig_ip = data->regs->ip;
-	u64 at;
+	struct bts_record *at, *top;
 
 	if (!counter)
 		return;
@@ -1499,19 +1500,18 @@ static void intel_pmu_drain_bts_buffer(struct cpu_hw_counters *cpuc,
 	if (!ds)
 		return;
 
-	for (at = ds->bts_buffer_base;
-	     at < ds->bts_index;
-	     at += sizeof(struct bts_record)) {
-		struct bts_record *rec = (struct bts_record *)(long)at;
+	at  = (struct bts_record *)(unsigned long)ds->bts_buffer_base;
+	top = (struct bts_record *)(unsigned long)ds->bts_index;
 
-		data->regs->ip	= rec->from;
-		data->addr	= rec->to;
+	ds->bts_index = ds->bts_buffer_base;
+
+	for (; at < top; at++) {
+		data->regs->ip	= at->from;
+		data->addr	= at->to;
 
 		perf_counter_output(counter, 1, data);
 	}
 
-	ds->bts_index = ds->bts_buffer_base;
-
 	data->regs->ip	= orig_ip;
 	data->addr	= 0;
 
#6 lore raw

[tip:perfcounters/core] x86, perf_counter, bts: Do not allow kernel BTS tracing for now

In reply to [email protected]
Commit-ID:  1653192f510bd8114b7b133d7289e6e5c3e95046
Gitweb:     http://git.kernel.org/tip/1653192f510bd8114b7b133d7289e6e5c3e95046
Author:     [email protected] <[email protected]>
AuthorDate: Wed, 2 Sep 2009 16:04:48 +0200
Committer:  Ingo Molnar <[email protected]>
CommitDate: Fri, 4 Sep 2009 09:26:40 +0200

x86, perf_counter, bts: Do not allow kernel BTS tracing for now

Kernel BTS tracing generates too much data too fast for us to handle, causing the kernel to hang.

Fail for BTS requests for kernel code.

LKML-Reference: <[email protected]> [ This is really a workaround - but we want BTS tracing in .32
so make sure we dont regress. The lockup should be fixed ASAP. ]

---
 arch/x86/kernel/cpu/perf_counter.c |   11 +++++++++--
 1 files changed, 9 insertions(+), 2 deletions(-)
arch/x86/kernel/cpu/perf_counter.c +9 -2
index 3776b0b..f9cd084 100644
@@ -984,8 +984,15 @@ static int __hw_perf_counter_init(struct perf_counter *counter)
 	 * Branch tracing:
 	 */
 	if ((attr->config == PERF_COUNT_HW_BRANCH_INSTRUCTIONS) &&
-	    (hwc->sample_period == 1) && !bts_available())
-		return -EOPNOTSUPP;
+	    (hwc->sample_period == 1)) {
+		/* BTS is not supported by this architecture. */
+		if (!bts_available())
+			return -EOPNOTSUPP;
+
+		/* BTS is currently only allowed for user-mode. */
+		if (hwc->config & ARCH_PERFMON_EVENTSEL_OS)
+			return -EOPNOTSUPP;
+	}
 
 	hwc->config |= config;