Skip to contentSkip to conversation

ksym_tracer: Fix to make the tracer work

PATCH In review Li Zefan 3 participants 11 messages Dec 30, 2009 – Jan 10, 2010 Tracing

lore
Share & tools
lore

Li Zefan

author
#0 lore raw

PATCH 1/4 ksym_tracer: Fix to make the tracer work

ksym tracer doesn't work:

 # echo tasklist_lock:rw- > ksym_trace_filter
 -bash: echo: write error: No such device

It's because we pass to perf_event_create_kernel_counter() a cpu number which is not present.

---
 kernel/hw_breakpoint.c    |   10 +++++++---
 kernel/trace/trace_ksym.c |    1 -
 2 files changed, 7 insertions(+), 4 deletions(-)
kernel/hw_breakpoint.c +7 -3
index dbcbf6a..50dbd59 100644
@@ -40,6 +40,7 @@
 #include <linux/percpu.h>
 #include <linux/sched.h>
 #include <linux/init.h>
+#include <linux/cpu.h>
 #include <linux/smp.h>
 
 #include <linux/hw_breakpoint.h>
@@ -388,7 +389,8 @@ register_wide_hw_breakpoint(struct perf_event_attr *attr,
 	if (!cpu_events)
 		return ERR_PTR(-ENOMEM);
 
-	for_each_possible_cpu(cpu) {
+	get_online_cpus();
+	for_each_online_cpu(cpu) {
 		pevent = per_cpu_ptr(cpu_events, cpu);
 		bp = perf_event_create_kernel_counter(attr, cpu, -1, triggered);
 
@@ -399,18 +401,20 @@ register_wide_hw_breakpoint(struct perf_event_attr *attr,
 			goto fail;
 		}
 	}
+	put_online_cpus();
 
 	return cpu_events;
 
 fail:
-	for_each_possible_cpu(cpu) {
+	for_each_online_cpu(cpu) {
 		pevent = per_cpu_ptr(cpu_events, cpu);
 		if (IS_ERR(*pevent))
 			break;
 		unregister_hw_breakpoint(*pevent);
 	}
+	put_online_cpus();
+
 	free_percpu(cpu_events);
-	/* return the error if any */
 	return ERR_PTR(err);
 }
 EXPORT_SYMBOL_GPL(register_wide_hw_breakpoint);
kernel/trace/trace_ksym.c +0 -1
index faf37fa..340b6ff 100644
@@ -197,7 +197,6 @@ int process_new_ksym_entry(char *ksymname, int op, unsigned long addr)
 	entry->attr.bp_addr = addr;
 	entry->attr.bp_len = HW_BREAKPOINT_LEN_4;
 
-	ret = -EAGAIN;
 	entry->ksym_hbp = register_wide_hw_breakpoint(&entry->attr,
 					ksym_hbp_handler);
 
Signature · 1 line
1.6.3
author
#1 lore raw

PATCH 2/4 ksym_tracer: Fix to allow writing newline to ksym_trace_filter

It used to work, but now doesn't:

 # echo > ksym_filter
 bash: echo: write error: Invalid argument

It's caused by d954fbf0ff6b5fdfb32350e85a2f15d3db976506 ("tracing: Fix wrong usage of strstrip in trace_ksyms").

---
 kernel/trace/trace_ksym.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)
kernel/trace/trace_ksym.c +2 -2
index 340b6ff..160a8d8 100644
@@ -299,8 +299,8 @@ static ssize_t ksym_trace_filter_write(struct file *file,
 	 * 2: echo 0 > ksym_trace_filter
 	 * 3: echo "*:---" > ksym_trace_filter
 	 */
-	if (!buf[0] || !strcmp(buf, "0") ||
-	    !strcmp(buf, "*:---")) {
+	if (!input_string[0] || !strcmp(input_string, "0") ||
+	    !strcmp(input_string, "*:---")) {
 		__ksym_trace_reset();
 		ret = 0;
 		goto out;
Signature · 1 line
1.6.3
author
#3 lore raw

PATCH 3/4 ksym_tracer: Fix race when incrementing count

In reply to Li Zefan

We are under rcu read section but not holding the write lock, so count++ is not atomic. Use atomic64_t instead.

---
 kernel/trace/trace_ksym.c |   12 +++++++-----
 1 files changed, 7 insertions(+), 5 deletions(-)
kernel/trace/trace_ksym.c +7 -5
index 160a8d8..67d79f7 100644
@@ -32,6 +32,8 @@
 #include <linux/hw_breakpoint.h>
 #include <asm/hw_breakpoint.h>
 
+#include <asm/atomic.h>
+
 /*
  * For now, let us restrict the no. of symbols traced simultaneously to number
  * of available hardware breakpoint registers.
@@ -44,7 +46,7 @@ struct trace_ksym {
 	struct perf_event	**ksym_hbp;
 	struct perf_event_attr	attr;
 #ifdef CONFIG_PROFILE_KSYM_TRACER
-	unsigned long		counter;
+	atomic64_t		counter;
 #endif
 	struct hlist_node	ksym_hlist;
 };
@@ -69,9 +71,8 @@ void ksym_collect_stats(unsigned long hbp_hit_addr)
 
 	rcu_read_lock();
 	hlist_for_each_entry_rcu(entry, node, &ksym_filter_head, ksym_hlist) {
-		if ((entry->attr.bp_addr == hbp_hit_addr) &&
-		    (entry->counter <= MAX_UL_INT)) {
-			entry->counter++;
+		if (entry->attr.bp_addr == hbp_hit_addr) {
+			atomic64_inc(&entry->counter);
 			break;
 		}
 	}
@@ -501,7 +502,8 @@ static int ksym_tracer_stat_show(struct seq_file *m, void *v)
 		seq_printf(m, "  %-36s", fn_name);
 	else
 		seq_printf(m, "  %-36s", "<NA>");
-	seq_printf(m, " %15lu\n", entry->counter);
+	seq_printf(m, " %15llu\n",
+		   (unsigned long long)atomic64_read(&entry->counter));
 
 	return 0;
 }
Signature · 1 line
1.6.3
author
#5 lore raw

PATCH 4/4 ksym_tracer: Remove trace_stat

In reply to Li Zefan

trace_stat is problematic. Don't use it, use seqfile instead.

This fixes a race that reading the stat file is not protected by any lock, which can lead to use after free.

---
 kernel/trace/trace_ksym.c |  127 ++++++++++++++++++---------------------------
 1 files changed, 50 insertions(+), 77 deletions(-)
kernel/trace/trace_ksym.c +50 -77
index 67d79f7..94103cd 100644
@@ -26,7 +26,6 @@
 #include <linux/fs.h>
 
 #include "trace_output.h"
-#include "trace_stat.h"
 #include "trace.h"
 
 #include <linux/hw_breakpoint.h>
@@ -444,103 +443,77 @@ struct tracer ksym_tracer __read_mostly =
 	.print_line	= ksym_trace_output
 };
 
-__init static int init_ksym_trace(void)
-{
-	struct dentry *d_tracer;
-	struct dentry *entry;
-
-	d_tracer = tracing_init_dentry();
-	ksym_filter_entry_count = 0;
-
-	entry = debugfs_create_file("ksym_trace_filter", 0644, d_tracer,
-				    NULL, &ksym_tracing_fops);
-	if (!entry)
-		pr_warning("Could not create debugfs "
-			   "'ksym_trace_filter' file\n");
-
-	return register_tracer(&ksym_tracer);
-}
-device_initcall(init_ksym_trace);
-
-
 #ifdef CONFIG_PROFILE_KSYM_TRACER
-static int ksym_tracer_stat_headers(struct seq_file *m)
+static int ksym_profile_show(struct seq_file *m, void *v)
 {
+	struct hlist_node *node;
+	struct trace_ksym *entry;
+	int access_type = 0;
+	char fn_name[KSYM_NAME_LEN];
+
 	seq_puts(m, "  Access Type ");
 	seq_puts(m, "  Symbol                                       Counter\n");
 	seq_puts(m, "  ----------- ");
 	seq_puts(m, "  ------                                       -------\n");
-	return 0;
-}
 
-static int ksym_tracer_stat_show(struct seq_file *m, void *v)
-{
-	struct hlist_node *stat = v;
-	struct trace_ksym *entry;
-	int access_type = 0;
-	char fn_name[KSYM_NAME_LEN];
+	rcu_read_lock();
+	hlist_for_each_entry_rcu(entry, node, &ksym_filter_head, ksym_hlist) {
 
-	entry = hlist_entry(stat, struct trace_ksym, ksym_hlist);
+		access_type = entry->attr.bp_type;
 
-	access_type = entry->attr.bp_type;
+		switch (access_type) {
+		case HW_BREAKPOINT_R:
+			seq_puts(m, "  R           ");
+			break;
+		case HW_BREAKPOINT_W:
+			seq_puts(m, "  W           ");
+			break;
+		case HW_BREAKPOINT_R | HW_BREAKPOINT_W:
+			seq_puts(m, "  RW          ");
+			break;
+		default:
+			seq_puts(m, "  NA          ");
+		}
 
-	switch (access_type) {
-	case HW_BREAKPOINT_R:
-		seq_puts(m, "  R           ");
-		break;
-	case HW_BREAKPOINT_W:
-		seq_puts(m, "  W           ");
-		break;
-	case HW_BREAKPOINT_R | HW_BREAKPOINT_W:
-		seq_puts(m, "  RW          ");
-		break;
-	default:
-		seq_puts(m, "  NA          ");
+		if (lookup_symbol_name(entry->attr.bp_addr, fn_name) >= 0)
+			seq_printf(m, "  %-36s", fn_name);
+		else
+			seq_printf(m, "  %-36s", "<NA>");
+		seq_printf(m, " %15llu\n",
+			   (unsigned long long)atomic64_read(&entry->counter));
 	}
-
-	if (lookup_symbol_name(entry->attr.bp_addr, fn_name) >= 0)
-		seq_printf(m, "  %-36s", fn_name);
-	else
-		seq_printf(m, "  %-36s", "<NA>");
-	seq_printf(m, " %15llu\n",
-		   (unsigned long long)atomic64_read(&entry->counter));
+	rcu_read_unlock();
 
 	return 0;
 }
 
-static void *ksym_tracer_stat_start(struct tracer_stat *trace)
+static int ksym_profile_open(struct inode *node, struct file *file)
 {
-	return ksym_filter_head.first;
-}
-
-static void *
-ksym_tracer_stat_next(void *v, int idx)
-{
-	struct hlist_node *stat = v;
-
-	return stat->next;
+	return single_open(file, ksym_profile_show, NULL);
 }
 
-static struct tracer_stat ksym_tracer_stats = {
-	.name = "ksym_tracer",
-	.stat_start = ksym_tracer_stat_start,
-	.stat_next = ksym_tracer_stat_next,
-	.stat_headers = ksym_tracer_stat_headers,
-	.stat_show = ksym_tracer_stat_show
+static const struct file_operations ksym_profile_fops = {
+	.open		= ksym_profile_open,
+	.read		= seq_read,
+	.llseek		= seq_lseek,
+	.release	= single_release,
 };
+#endif /* CONFIG_PROFILE_KSYM_TRACER */
 
-__init static int ksym_tracer_stat_init(void)
+__init static int init_ksym_trace(void)
 {
-	int ret;
+	struct dentry *d_tracer;
 
-	ret = register_stat_tracer(&ksym_tracer_stats);
-	if (ret) {
-		printk(KERN_WARNING "Warning: could not register "
-				    "ksym tracer stats\n");
-		return 1;
-	}
+	d_tracer = tracing_init_dentry();
 
-	return 0;
+	trace_create_file("ksym_trace_filter", 0644, d_tracer,
+			  NULL, &ksym_tracing_fops);
+
+#ifdef CONFIG_PROFILE_KSYM_TRACER
+	trace_create_file("ksym_profile", 0444, d_tracer,
+			  NULL, &ksym_profile_fops);
+#endif
+
+	return register_tracer(&ksym_tracer);
 }
-fs_initcall(ksym_tracer_stat_init);
-#endif /* CONFIG_PROFILE_KSYM_TRACER */
+device_initcall(init_ksym_trace);

-- 1.6.3

#7 lore raw
In reply to Li Zefan
Commit-ID:  88f7a890d74137ab0d126a5d65679cd620f1a289
Gitweb:     http://git.kernel.org/tip/88f7a890d74137ab0d126a5d65679cd620f1a289
Author:     Li Zefan <[email protected]>
AuthorDate: Wed, 30 Dec 2009 14:22:22 +0800
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 30 Dec 2009 07:50:47 +0100

ksym_tracer: Fix to make the tracer work

ksym tracer doesn't work:

 # echo tasklist_lock:rw- > ksym_trace_filter
 -bash: echo: write error: No such device

It's because we pass to perf_event_create_kernel_counter() a cpu number which is not present.

Cc Steven Rostedt <[email protected]>
Cc K.Prasad <[email protected]>
Cc Frederic Weisbecker <[email protected]>

LKML-Reference: <[email protected]>

---
 kernel/hw_breakpoint.c    |   10 +++++++---
 kernel/trace/trace_ksym.c |    1 -
 2 files changed, 7 insertions(+), 4 deletions(-)
kernel/hw_breakpoint.c +7 -3
index 366eedf..48fb0bb 100644
@@ -40,6 +40,7 @@
 #include <linux/percpu.h>
 #include <linux/sched.h>
 #include <linux/init.h>
+#include <linux/cpu.h>
 #include <linux/smp.h>
 
 #include <linux/hw_breakpoint.h>
@@ -388,7 +389,8 @@ register_wide_hw_breakpoint(struct perf_event_attr *attr,
 	if (!cpu_events)
 		return ERR_PTR(-ENOMEM);
 
-	for_each_possible_cpu(cpu) {
+	get_online_cpus();
+	for_each_online_cpu(cpu) {
 		pevent = per_cpu_ptr(cpu_events, cpu);
 		bp = perf_event_create_kernel_counter(attr, cpu, -1, triggered);
 
@@ -399,18 +401,20 @@ register_wide_hw_breakpoint(struct perf_event_attr *attr,
 			goto fail;
 		}
 	}
+	put_online_cpus();
 
 	return cpu_events;
 
 fail:
-	for_each_possible_cpu(cpu) {
+	for_each_online_cpu(cpu) {
 		pevent = per_cpu_ptr(cpu_events, cpu);
 		if (IS_ERR(*pevent))
 			break;
 		unregister_hw_breakpoint(*pevent);
 	}
+	put_online_cpus();
+
 	free_percpu(cpu_events);
-	/* return the error if any */
 	return ERR_PTR(err);
 }
 EXPORT_SYMBOL_GPL(register_wide_hw_breakpoint);
kernel/trace/trace_ksym.c +0 -1
index faf37fa..340b6ff 100644
@@ -197,7 +197,6 @@ int process_new_ksym_entry(char *ksymname, int op, unsigned long addr)
 	entry->attr.bp_addr = addr;
 	entry->attr.bp_len = HW_BREAKPOINT_LEN_4;
 
-	ret = -EAGAIN;
 	entry->ksym_hbp = register_wide_hw_breakpoint(&entry->attr,
 					ksym_hbp_handler);
 
#2 lore raw

[tip:tracing/urgent] ksym_tracer: Fix to allow writing newline to ksym_trace_filter

In reply to Li Zefan
Commit-ID:  3d13ec2efdb5843ad91e57b60d50b44d922cf063
Gitweb:     http://git.kernel.org/tip/3d13ec2efdb5843ad91e57b60d50b44d922cf063
Author:     Li Zefan <[email protected]>
AuthorDate: Wed, 30 Dec 2009 14:23:19 +0800
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 30 Dec 2009 07:50:49 +0100

ksym_tracer: Fix to allow writing newline to ksym_trace_filter

It used to work, but now doesn't:

 # echo > ksym_filter
 bash: echo: write error: Invalid argument

It's caused by d954fbf0ff6b5fdfb32350e85a2f15d3db976506 ("tracing: Fix wrong usage of strstrip in trace_ksyms").

Cc Steven Rostedt <[email protected]>
Cc K.Prasad <[email protected]>
Cc Frederic Weisbecker <[email protected]>

LKML-Reference: <[email protected]>

---
 kernel/trace/trace_ksym.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)
kernel/trace/trace_ksym.c +2 -2
index 340b6ff..160a8d8 100644
@@ -299,8 +299,8 @@ static ssize_t ksym_trace_filter_write(struct file *file,
 	 * 2: echo 0 > ksym_trace_filter
 	 * 3: echo "*:---" > ksym_trace_filter
 	 */
-	if (!buf[0] || !strcmp(buf, "0") ||
-	    !strcmp(buf, "*:---")) {
+	if (!input_string[0] || !strcmp(input_string, "0") ||
+	    !strcmp(input_string, "*:---")) {
 		__ksym_trace_reset();
 		ret = 0;
 		goto out;
#4 lore raw

[tip:tracing/urgent] ksym_tracer: Fix race when incrementing count

In reply to Li Zefan
Commit-ID:  e6d9491bf8ba6728cc86aeabbc688d20ec0563b5
Gitweb:     http://git.kernel.org/tip/e6d9491bf8ba6728cc86aeabbc688d20ec0563b5
Author:     Li Zefan <[email protected]>
AuthorDate: Wed, 30 Dec 2009 14:23:40 +0800
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 30 Dec 2009 07:50:49 +0100

ksym_tracer: Fix race when incrementing count

We are under rcu read section but not holding the write lock, so count++ is not atomic. Use atomic64_t instead.

Cc Steven Rostedt <[email protected]>
Cc K.Prasad <[email protected]>
Cc Frederic Weisbecker <[email protected]>

LKML-Reference: <[email protected]>

---
 kernel/trace/trace_ksym.c |   12 +++++++-----
 1 files changed, 7 insertions(+), 5 deletions(-)
kernel/trace/trace_ksym.c +7 -5
index 160a8d8..67d79f7 100644
@@ -32,6 +32,8 @@
 #include <linux/hw_breakpoint.h>
 #include <asm/hw_breakpoint.h>
 
+#include <asm/atomic.h>
+
 /*
  * For now, let us restrict the no. of symbols traced simultaneously to number
  * of available hardware breakpoint registers.
@@ -44,7 +46,7 @@ struct trace_ksym {
 	struct perf_event	**ksym_hbp;
 	struct perf_event_attr	attr;
 #ifdef CONFIG_PROFILE_KSYM_TRACER
-	unsigned long		counter;
+	atomic64_t		counter;
 #endif
 	struct hlist_node	ksym_hlist;
 };
@@ -69,9 +71,8 @@ void ksym_collect_stats(unsigned long hbp_hit_addr)
 
 	rcu_read_lock();
 	hlist_for_each_entry_rcu(entry, node, &ksym_filter_head, ksym_hlist) {
-		if ((entry->attr.bp_addr == hbp_hit_addr) &&
-		    (entry->counter <= MAX_UL_INT)) {
-			entry->counter++;
+		if (entry->attr.bp_addr == hbp_hit_addr) {
+			atomic64_inc(&entry->counter);
 			break;
 		}
 	}
@@ -501,7 +502,8 @@ static int ksym_tracer_stat_show(struct seq_file *m, void *v)
 		seq_printf(m, "  %-36s", fn_name);
 	else
 		seq_printf(m, "  %-36s", "<NA>");
-	seq_printf(m, " %15lu\n", entry->counter);
+	seq_printf(m, " %15llu\n",
+		   (unsigned long long)atomic64_read(&entry->counter));
 
 	return 0;
 }
#6 lore raw

[tip:tracing/urgent] ksym_tracer: Remove trace_stat

In reply to Li Zefan
Commit-ID:  53ab668064edaeef99c0ee22799483d45f4c81f6
Gitweb:     http://git.kernel.org/tip/53ab668064edaeef99c0ee22799483d45f4c81f6
Author:     Li Zefan <[email protected]>
AuthorDate: Wed, 30 Dec 2009 14:24:03 +0800
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 30 Dec 2009 07:50:50 +0100

ksym_tracer: Remove trace_stat

trace_stat is problematic. Don't use it, use seqfile instead.

This fixes a race that reading the stat file is not protected by any lock, which can lead to use after free.

Cc Steven Rostedt <[email protected]>
Cc K.Prasad <[email protected]>
Cc Frederic Weisbecker <[email protected]>

LKML-Reference: <[email protected]>

---
 kernel/trace/trace_ksym.c |  127 ++++++++++++++++++---------------------------
 1 files changed, 50 insertions(+), 77 deletions(-)
kernel/trace/trace_ksym.c +50 -77
index 67d79f7..94103cd 100644
@@ -26,7 +26,6 @@
 #include <linux/fs.h>
 
 #include "trace_output.h"
-#include "trace_stat.h"
 #include "trace.h"
 
 #include <linux/hw_breakpoint.h>
@@ -444,103 +443,77 @@ struct tracer ksym_tracer __read_mostly =
 	.print_line	= ksym_trace_output
 };
 
-__init static int init_ksym_trace(void)
-{
-	struct dentry *d_tracer;
-	struct dentry *entry;
-
-	d_tracer = tracing_init_dentry();
-	ksym_filter_entry_count = 0;
-
-	entry = debugfs_create_file("ksym_trace_filter", 0644, d_tracer,
-				    NULL, &ksym_tracing_fops);
-	if (!entry)
-		pr_warning("Could not create debugfs "
-			   "'ksym_trace_filter' file\n");
-
-	return register_tracer(&ksym_tracer);
-}
-device_initcall(init_ksym_trace);
-
-
 #ifdef CONFIG_PROFILE_KSYM_TRACER
-static int ksym_tracer_stat_headers(struct seq_file *m)
+static int ksym_profile_show(struct seq_file *m, void *v)
 {
+	struct hlist_node *node;
+	struct trace_ksym *entry;
+	int access_type = 0;
+	char fn_name[KSYM_NAME_LEN];
+
 	seq_puts(m, "  Access Type ");
 	seq_puts(m, "  Symbol                                       Counter\n");
 	seq_puts(m, "  ----------- ");
 	seq_puts(m, "  ------                                       -------\n");
-	return 0;
-}
 
-static int ksym_tracer_stat_show(struct seq_file *m, void *v)
-{
-	struct hlist_node *stat = v;
-	struct trace_ksym *entry;
-	int access_type = 0;
-	char fn_name[KSYM_NAME_LEN];
+	rcu_read_lock();
+	hlist_for_each_entry_rcu(entry, node, &ksym_filter_head, ksym_hlist) {
 
-	entry = hlist_entry(stat, struct trace_ksym, ksym_hlist);
+		access_type = entry->attr.bp_type;
 
-	access_type = entry->attr.bp_type;
+		switch (access_type) {
+		case HW_BREAKPOINT_R:
+			seq_puts(m, "  R           ");
+			break;
+		case HW_BREAKPOINT_W:
+			seq_puts(m, "  W           ");
+			break;
+		case HW_BREAKPOINT_R | HW_BREAKPOINT_W:
+			seq_puts(m, "  RW          ");
+			break;
+		default:
+			seq_puts(m, "  NA          ");
+		}
 
-	switch (access_type) {
-	case HW_BREAKPOINT_R:
-		seq_puts(m, "  R           ");
-		break;
-	case HW_BREAKPOINT_W:
-		seq_puts(m, "  W           ");
-		break;
-	case HW_BREAKPOINT_R | HW_BREAKPOINT_W:
-		seq_puts(m, "  RW          ");
-		break;
-	default:
-		seq_puts(m, "  NA          ");
+		if (lookup_symbol_name(entry->attr.bp_addr, fn_name) >= 0)
+			seq_printf(m, "  %-36s", fn_name);
+		else
+			seq_printf(m, "  %-36s", "<NA>");
+		seq_printf(m, " %15llu\n",
+			   (unsigned long long)atomic64_read(&entry->counter));
 	}
-
-	if (lookup_symbol_name(entry->attr.bp_addr, fn_name) >= 0)
-		seq_printf(m, "  %-36s", fn_name);
-	else
-		seq_printf(m, "  %-36s", "<NA>");
-	seq_printf(m, " %15llu\n",
-		   (unsigned long long)atomic64_read(&entry->counter));
+	rcu_read_unlock();
 
 	return 0;
 }
 
-static void *ksym_tracer_stat_start(struct tracer_stat *trace)
+static int ksym_profile_open(struct inode *node, struct file *file)
 {
-	return ksym_filter_head.first;
-}
-
-static void *
-ksym_tracer_stat_next(void *v, int idx)
-{
-	struct hlist_node *stat = v;
-
-	return stat->next;
+	return single_open(file, ksym_profile_show, NULL);
 }
 
-static struct tracer_stat ksym_tracer_stats = {
-	.name = "ksym_tracer",
-	.stat_start = ksym_tracer_stat_start,
-	.stat_next = ksym_tracer_stat_next,
-	.stat_headers = ksym_tracer_stat_headers,
-	.stat_show = ksym_tracer_stat_show
+static const struct file_operations ksym_profile_fops = {
+	.open		= ksym_profile_open,
+	.read		= seq_read,
+	.llseek		= seq_lseek,
+	.release	= single_release,
 };
+#endif /* CONFIG_PROFILE_KSYM_TRACER */
 
-__init static int ksym_tracer_stat_init(void)
+__init static int init_ksym_trace(void)
 {
-	int ret;
+	struct dentry *d_tracer;
 
-	ret = register_stat_tracer(&ksym_tracer_stats);
-	if (ret) {
-		printk(KERN_WARNING "Warning: could not register "
-				    "ksym tracer stats\n");
-		return 1;
-	}
+	d_tracer = tracing_init_dentry();
 
-	return 0;
+	trace_create_file("ksym_trace_filter", 0644, d_tracer,
+			  NULL, &ksym_tracing_fops);
+
+#ifdef CONFIG_PROFILE_KSYM_TRACER
+	trace_create_file("ksym_profile", 0444, d_tracer,
+			  NULL, &ksym_profile_fops);
+#endif
+
+	return register_tracer(&ksym_tracer);
 }
-fs_initcall(ksym_tracer_stat_init);
-#endif /* CONFIG_PROFILE_KSYM_TRACER */
+device_initcall(init_ksym_trace);
#8 lore raw
In reply to Li Zefan
Quoted 27 lines · Li Zefan

ksym tracer doesn't work:

 # echo tasklist_lock:rw- > ksym_trace_filter
 -bash: echo: write error: No such device

It's because we pass to perf_event_create_kernel_counter() a cpu number which is not present.

---
 kernel/hw_breakpoint.c    |   10 +++++++---
 kernel/trace/trace_ksym.c |    1 -
 2 files changed, 7 insertions(+), 4 deletions(-)
kernel/hw_breakpoint.c +1 -0
index dbcbf6a..50dbd59 100644
@@ -40,6 +40,7 @@
 #include <linux/percpu.h>
 #include <linux/sched.h>
 #include <linux/init.h>
+#include <linux/cpu.h>
 #include <linux/smp.h>
 
 #include <linux/hw_breakpoint.h>

kernel/hw_breakpoint.c register_wide_hw_breakpoint(struct perf_event_attr *attr, · line 396

Quoted 9 lines
kernel/hw_breakpoint.c +2 -1
@@ -388,7 +389,8 @@ register_wide_hw_breakpoint(struct perf_event_attr *attr,
 	if (!cpu_events)
 		return ERR_PTR(-ENOMEM);
 
-	for_each_possible_cpu(cpu) {
+	get_online_cpus();
+	for_each_online_cpu(cpu) {
 		pevent = per_cpu_ptr(cpu_events, cpu);
 		bp = perf_event_create_kernel_counter(attr, cpu, -1, triggered);
 

Frederic must have used for_each_possible_cpu() to account for CPUs that are offline at the time of registration, but may eventually turn online. Since register_wide_hw_breakpoint() interface is designed to deliver system-wide breakpoints, the debug registers of a new online CPU will should have the breakpoints populated to comprehensively notify all memory accesses over target address.

I'd rather wait to hear from Frederic to know why perf_event_create_kernel_counter() returns an error when run for an offline cpu and how it can be solved.

Thanks,
K.Prasad

author
#9 lore raw
Quoted 45 lines · K.Prasad
Quoted 27 lines · Li Zefan

ksym tracer doesn't work:

 # echo tasklist_lock:rw- > ksym_trace_filter
 -bash: echo: write error: No such device

It's because we pass to perf_event_create_kernel_counter() a cpu number which is not present.

---
 kernel/hw_breakpoint.c    |   10 +++++++---
 kernel/trace/trace_ksym.c |    1 -
 2 files changed, 7 insertions(+), 4 deletions(-)
kernel/hw_breakpoint.c +1 -0
index dbcbf6a..50dbd59 100644
@@ -40,6 +40,7 @@
 #include <linux/percpu.h>
 #include <linux/sched.h>
 #include <linux/init.h>
+#include <linux/cpu.h>
 #include <linux/smp.h>
 
 #include <linux/hw_breakpoint.h>

kernel/hw_breakpoint.c register_wide_hw_breakpoint(struct perf_event_attr *attr, · line 396

Quoted 9 lines
kernel/hw_breakpoint.c +2 -1
@@ -388,7 +389,8 @@ register_wide_hw_breakpoint(struct perf_event_attr *attr,
 	if (!cpu_events)
 		return ERR_PTR(-ENOMEM);
 
-	for_each_possible_cpu(cpu) {
+	get_online_cpus();
+	for_each_online_cpu(cpu) {
 		pevent = per_cpu_ptr(cpu_events, cpu);
 		bp = perf_event_create_kernel_counter(attr, cpu, -1, triggered);
 

Frederic must have used for_each_possible_cpu() to account for CPUs that are offline at the time of registration, but may eventually turn online. Since register_wide_hw_breakpoint() interface is designed to deliver system-wide breakpoints, the debug registers of a new online CPU will should have the breakpoints populated to comprehensively notify all memory accesses over target address.

Quoted 4 lines

I'd rather wait to hear from Frederic to know why perf_event_create_kernel_counter() returns an error when run for an offline cpu and how it can be solved.

See the comment in find_get_context() in kernel/perf_event.c:

                /*
                 * We could be clever and allow to attach a event to an
                 * offline CPU and activate it when the CPU comes up, but
                 * that's for later.
                 */
                if (!cpu_online(cpu))
                        return ERR_PTR(-ENODEV);

So I think we can use for_each_possible_cpu() in the future, but not now.

#10 lore raw
Quoted 23 lines · Li Zefan
Quoted 11 lines · K.Prasad

Frederic must have used for_each_possible_cpu() to account for CPUs that are offline at the time of registration, but may eventually turn online. Since register_wide_hw_breakpoint() interface is designed to deliver system-wide breakpoints, the debug registers of a new online CPU will should have the breakpoints populated to comprehensively notify all memory accesses over target address.

I'd rather wait to hear from Frederic to know why perf_event_create_kernel_counter() returns an error when run for an offline cpu and how it can be solved.

See the comment in find_get_context() in kernel/perf_event.c:

                /*
                 * We could be clever and allow to attach a event to an
                 * offline CPU and activate it when the CPU comes up, but
                 * that's for later.
                 */
                if (!cpu_online(cpu))
                        return ERR_PTR(-ENODEV);
Quoted 2 lines

So I think we can use for_each_possible_cpu() in the future, but not now.

Ah, right I indeed missed that.