tracing: Add an option to show symbols in _text+offset for function profiler
Subsystems (3)
Masami Hiramatsu (Google)
maintainerv3 cover tracing: Add an option to show symbols in _text+offset for function profiler
Hi,
This series implements an option to show symbols in _text+OFFSET format instead of symbol name in the function profiler. This is the 3rd version, the 2nd one is here;
https://lore.kernel.org/all/175854516202.353182.1216978967046454932.stgit@devnote2/
This version fixes to declare TRACE_ITER_* in trace.h and define it in trace.c [1/2].
Thank you,
---
Masami Hiramatsu (Google) (2):
tracing: Allow tracer to add more than 32 options
tracing: Add an option to show symbols in _text+offset for function profiler
kernel/trace/ftrace.c | 26 ++++++++++ kernel/trace/trace.c | 29 +++++++----- kernel/trace/trace.h | 92 +++++++++++++++++++++---------------- kernel/trace/trace_irqsoff.c | 2 - kernel/trace/trace_sched_wakeup.c | 2 - 5 files changed, 96 insertions(+), 55 deletions(-)
--
Masami Hiramatsu (Google) <[email protected]>
Masami Hiramatsu (Google)
maintainerPATCH v3 1/2 tracing: Allow tracer to add more than 32 options
From: Masami Hiramatsu (Google) <[email protected]>
Since enum trace_iterator_flags is 32bit, the max number of the option flags is limited to 32 and it is fully used now. To add a new option, we need to expand it.
This replaces trace_iterator_flags with just a list of static const u64 so that we can add another 32 new options. Now all option masks are u64 instead of u32 or unsigned int.
---
Changes in v3:
- Make TRACE_ITER_* to global.
--- kernel/trace/trace.c | 24 +++++++---- kernel/trace/trace.h | 81 +++++++++++++++++++------------------ kernel/trace/trace_irqsoff.c | 2 - kernel/trace/trace_sched_wakeup.c | 2 - 4 files changed, 58 insertions(+), 51 deletions(-)
kernel/trace/trace.c +15 -9
@@ -86,6 +86,11 @@ void __init disable_tracing_selftest(const char *reason)#define tracing_selftest_disabled 0#endif +/* Define TRACE_ITER_* flags. */ +#undef C +#define C (a, b) const u64 TRACE_ITER_##a = (1ULL << TRACE_ITER_##a##_BIT); +TRACE_FLAGS +/* Pipe tracepoints to printk */ static struct trace_iterator *tracepoint_print_iter;int tracepoint_printk;
@@ -1736,7 +1741,7 @@ unsigned long nsecs_to_usecs(unsigned long nsecs)* of strings in the order that the evals (enum) were defined. */ #undef C -#define C (a, b) b +#define C (a, b) b,/* These must match the bit positions in trace_iterator_flags */ static const char *trace_options[] = {
@@ -5144,7 +5149,7 @@ static int tracing_trace_options_show(struct seq_file *m, void *v) trace_opts = tr->current_trace->flags->opts;for (i =0 ; trace_options[i]; i++) { -if (tr->trace_flags & (1 << i)) +if (tr->trace_flags & (1ULL << i))seq_printf (m,"%s\n" , trace_options[i]);else seq_printf (m,"no%s\n" , trace_options[i]);
@@ -5197,7 +5202,7 @@ static int set_tracer_option(struct trace_array *tr, char *cmp, int neg) }/* Some tracers require overwrite to stay enabled */ -int trace_keep_overwrite (struct tracer *tracer,u32 mask,int set) +int trace_keep_overwrite (struct tracer *tracer,u64 mask,int set) {if (tracer->enabled && (mask &TRACE_ITER_OVERWRITE ) && !set)return -1 ;
@@ -5205,7 +5210,7 @@ int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)return 0 ; } -int set_tracer_flag (struct trace_array *tr,unsigned int mask,int enabled) +int set_tracer_flag (struct trace_array *tr,u64 mask,int enabled) { if ((mask == TRACE_ITER_RECORD_TGID) || (mask == TRACE_ITER_RECORD_CMD) ||
@@ -5307,7 +5312,7 @@ int trace_set_options(struct trace_array *tr, char *option)if (ret <0 ) ret =set_tracer_option (tr, cmp, neg);else - ret =set_tracer_flag (tr,1 << ret, !neg); + ret =set_tracer_flag (tr,1ULL << ret, !neg);mutex_unlock (&trace_types_lock);mutex_unlock (&event_mutex);
@@ -9119,7 +9124,7 @@ trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,get_tr_index (tr_index, &tr, &index); -if (tr->trace_flags & (1 << index)) +if (tr->trace_flags & (1ULL << index)) buf ="1\n" ;else buf ="0\n" ;
@@ -9148,7 +9153,7 @@ trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,mutex_lock (&event_mutex);mutex_lock (&trace_types_lock); - ret =set_tracer_flag (tr,1 << index, val); + ret =set_tracer_flag (tr,1ULL << index, val);mutex_unlock (&trace_types_lock);mutex_unlock (&event_mutex);
@@ -9312,8 +9317,9 @@ static void create_trace_options_dir(struct trace_array *tr)for (i =0 ; trace_options[i]; i++) {if (top_level || - !((1 << i) &TOP_LEVEL_TRACE_FLAGS )) + !((1ULL << i) &TOP_LEVEL_TRACE_FLAGS )) {create_trace_option_core_file (tr, trace_options[i], i); + } } }
@@ -9997,7 +10003,7 @@ static int __remove_instance(struct trace_array *tr)/* Disable all the flags that were enabled coming in */ for (i =0 ; i <TRACE_FLAGS_MAX_SIZE ; i++) {if ((1 << i) &ZEROED_TRACE_FLAGS ) -set_tracer_flag (tr,1 << i,0 ); +set_tracer_flag (tr,1ULL << i,0 ); }if (printk_trace == tr)
kernel/trace/trace.h +41 -40
148 lines not on this page · raw message
kernel/trace/trace_irqsoff.c +1 -1
@@ -539,7 +539,7 @@ static inline int irqsoff_function_set(struct trace_array *tr, u32 mask, int set }#endif /* CONFIG_FUNCTION_TRACER */ -static int irqsoff_flag_changed (struct trace_array *tr,u32 mask,int set) +static int irqsoff_flag_changed (struct trace_array *tr,u64 mask,int set) {struct tracer *tracer = tr->current_trace;
kernel/trace/trace_sched_wakeup.c +1 -1
@@ -328,7 +328,7 @@ __trace_function(struct trace_array *tr,trace_function (tr, ip, parent_ip, trace_ctx,NULL ); } -static int wakeup_flag_changed (struct trace_array *tr,u32 mask,int set) +static int wakeup_flag_changed (struct trace_array *tr,u64 mask,int set) {struct tracer *tracer = tr->current_trace;
Masami Hiramatsu (Google)
maintainerPATCH v3 2/2 tracing: Add an option to show symbols in _text+offset for function profiler
In reply to Masami Hiramatsu (Google)From: Masami Hiramatsu (Google) <[email protected]>
Function profiler shows the hit count of each function using its symbol name. However, there are some same-name local symbols, which we can not distinguish.
To solve this issue, this introduces an option to show the symbols in "_text+OFFSET" format. This can avoid exposing the random shift of KASLR. The functions in modules are shown as "MODNAME+OFFSET" where the offset is from ".text".
E.g. for the kernel text symbols, specify vmlinux and the output to
addr2line, you can find the actual function and source info;
$ addr2line -fie vmlinux _text+3078208 __balance_callbacks kernel/sched/core.c:5064
for modules, specify the module file and .text+OFFSET;
$ addr2line -fie samples/trace_events/trace-events-sample.ko .text+8224 do_simple_thread_func samples/trace_events/trace-events-sample.c:23
---
Changes in v2:
- Define a dummy TRACE_ITER_PROF_TEXT_OFFSET if CONFIG_FUNCTION_PROFILER=n.
--- kernel/trace/ftrace.c | 26 +++++++++++++++++++++++++- kernel/trace/trace.c | 5 +++-- kernel/trace/trace.h | 11 ++++++++++- 3 files changed, 38 insertions(+), 4 deletions(-)
kernel/trace/ftrace.c +25 -1
@@ -534,7 +534,9 @@ static int function_stat_headers(struct seq_file *m)static int function_stat_show (struct seq_file *m,void *v) { +struct trace_array *tr =trace_get_global_array ();struct ftrace_profile *rec = v; +const char *refsymbol =NULL ;char str[KSYM_SYMBOL_LEN ];#ifdef CONFIG_FUNCTION_GRAPH_TRACER static struct trace_seq s;
@@ -554,7 +556,29 @@ static int function_stat_show(struct seq_file *m, void *v)return 0 ;#endif -kallsyms_lookup (rec->ip,NULL ,NULL ,NULL , str); +if (tr->trace_flags &TRACE_ITER_PROF_TEXT_OFFSET ) { +long offset; + +if (core_kernel_text (rec->ip)) { + refsymbol ="_text" ; + offset = rec->ip - (unsigned long )_text; + }else { +struct module *mod; + +guard (rcu)(); + mod =__module_text_address (rec->ip); +if (mod) { + refsymbol = mod->name; +/* Calculate offset from module's text entry address. */ + offset = rec->ip - (unsigned long )mod->mem[MOD_TEXT ].base; + } + } +if (refsymbol) +snprintf (str,sizeof (str)," %s%+ld" , refsymbol, offset); + } +if (!refsymbol) +kallsyms_lookup (rec->ip,NULL ,NULL ,NULL , str); +seq_printf (m," %-30.30s %10lu" , str, rec->counter);#ifdef CONFIG_FUNCTION_GRAPH_TRACER
kernel/trace/trace.c +3 -2
@@ -527,7 +527,8 @@ EXPORT_SYMBOL_GPL(unregister_ftrace_export);/* trace_options that are only supported by global_trace */ #define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \ - TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD) + TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD | \ + TRACE_ITER_PROF_TEXT_OFFSET)/* trace_flags that are default zero for instances */ #define ZEROED_TRACE_FLAGS \
@@ -11111,7 +11112,7 @@ __init static int tracer_alloc_buffers(void)#ifdef CONFIG_FUNCTION_TRACER /* Used to set module cached ftrace filtering at boot up */ -__init struct trace_array *trace_get_global_array (void ) +struct trace_array *trace_get_global_array (void ) {return &global_trace; }
kernel/trace/trace.h +10 -1
@@ -1353,6 +1353,14 @@ extern int trace_get_user(struct trace_parser *parser, const char __user *ubuf,# define STACK_FLAGS #endif +#ifdef CONFIG_FUNCTION_PROFILER +# define PROFILER_FLAGS \ + C(PROF_TEXT_OFFSET, "prof-text-offset") +#else +# define PROFILER_FLAGS +# define TRACE_ITER_PROF_TEXT_OFFSET 0UL +#endif +/* * trace_iterator_flags is an enumeration that defines bit * positions into trace_flags that controls the output.
@@ -1391,7 +1399,8 @@ extern int trace_get_user(struct trace_parser *parser, const char __user *ubuf,FUNCTION_FLAGS \ FGRAPH_FLAGS \STACK_FLAGS \ -BRANCH_FLAGS + BRANCH_FLAGS \ +PROFILER_FLAGS /* * By defining C, we can make TRACE_FLAGS a list of bit names
Steven Rostedt
maintainerRe: [PATCH v3 1/2] tracing: Allow tracer to add more than 32 options
In reply to Masami Hiramatsu (Google)Quoted 15 lines · On Tue, 30 Sep 2025 07:34:53 +0900 "Masami Hiramatsu (Google)"…
kernel/trace/trace.c +5
-0@@ -86,6 +86,11 @@ void __init disable_tracing_selftest(const char *reason)#define tracing_selftest_disabled 0#endif +/* Define TRACE_ITER_* flags. */ +#undef C +#define C (a, b) const u64 TRACE_ITER_##a = (1ULL << TRACE_ITER_##a##_BIT); +TRACE_FLAGS +#undef C
kernel/trace/trace.c
Quoted 6 lines
kernel/trace/trace.c +3
-2-#define C (a, b) TRACE_ITER_##a = (1 << TRACE_ITER_##a##_BIT) +#define C (a, b) extern const u64 TRACE_ITER_##a; -enum trace_iterator_flags {TRACE_FLAGS }; +TRACE_FLAGS +#undef C
Why all this work when this could have been simply fixed with a:
-enum trace_iterator_flags { TRACE_FLAGS };
+enum64 trace_iterator_flags { TRACE_FLAGS };
?
Not to mention, using const u64 requires saving these numbers in an address and referencing them, instead of doing it inlined in text. That is, using u64 instead of enum64 is both slower and wastes more memory.
-- Steve
Masami Hiramatsu (Google)
maintainerRe: [PATCH v3 1/2] tracing: Allow tracer to add more than 32 options
Quoted 27 lines · On Wed, 15 Oct 2025 17:20:20 -0400 Steven Rostedt <[email protected]> wrote:
Quoted 15 lines · On Tue, 30 Sep 2025 07:34:53 +0900 "Masami Hiramatsu (Google)"…
kernel/trace/trace.c +5
-0@@ -86,6 +86,11 @@ void __init disable_tracing_selftest(const char *reason)#define tracing_selftest_disabled 0#endif +/* Define TRACE_ITER_* flags. */ +#undef C +#define C (a, b) const u64 TRACE_ITER_##a = (1ULL << TRACE_ITER_##a##_BIT); +TRACE_FLAGS +#undef C
kernel/trace/trace.c
Quoted 6 lines
kernel/trace/trace.c +3
-2-#define C (a, b) TRACE_ITER_##a = (1 << TRACE_ITER_##a##_BIT) +#define C (a, b) extern const u64 TRACE_ITER_##a; -enum trace_iterator_flags {TRACE_FLAGS }; +TRACE_FLAGS +#undef CWhy all this work when this could have been simply fixed with a:
kernel/trace/trace.h enum trace_iterator_bits {
Quoted 4 lines
kernel/trace/trace.h +1
-1-enum trace_iterator_flags {TRACE_FLAGS }; +enum64 trace_iterator_flags { TRACE_FLAGS }; ?
I could not find any other enum64 usage, so I doubt it is available. (Does it depend on compiler?)
It seems C23 standard support it...
Quoted 4 lines
Not to mention, using const u64 requires saving these numbers in an address and referencing them, instead of doing it inlined in text. That is, using u64 instead of enum64 is both slower and wastes more memory.
Yeah, I expected that the compiler could easily optimize correctly, but maybe not?
Thank you,
Quoted 3 lines
-- Steve
Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
Masami Hiramatsu (Google)
maintainerRe: [PATCH v3 1/2] tracing: Allow tracer to add more than 32 options
In reply to Steven RostedtQuoted 27 lines · On Wed, 15 Oct 2025 17:20:20 -0400 Steven Rostedt <[email protected]> wrote:
Quoted 15 lines · On Tue, 30 Sep 2025 07:34:53 +0900 "Masami Hiramatsu (Google)"…
kernel/trace/trace.c +5
-0@@ -86,6 +86,11 @@ void __init disable_tracing_selftest(const char *reason)#define tracing_selftest_disabled 0#endif +/* Define TRACE_ITER_* flags. */ +#undef C +#define C (a, b) const u64 TRACE_ITER_##a = (1ULL << TRACE_ITER_##a##_BIT); +TRACE_FLAGS +#undef C
kernel/trace/trace.c
Quoted 6 lines
kernel/trace/trace.c +3
-2-#define C (a, b) TRACE_ITER_##a = (1 << TRACE_ITER_##a##_BIT) +#define C (a, b) extern const u64 TRACE_ITER_##a; -enum trace_iterator_flags {TRACE_FLAGS }; +TRACE_FLAGS +#undef CWhy all this work when this could have been simply fixed with a:
Review
Quoted 2 lines
(quoted diff) +1
-1-enum trace_iterator_flags { TRACE_FLAGS }; +enum64 trace_iterator_flags { TRACE_FLAGS };
With enum64, clang 18 caused this error.
In file included from /home/mhiramat/ksrc/linux/kernel/trace/ring_buffer.c:36:/home/mhiramat/ksrc/linux/kernel/trace/trace.h:1427:1: error: unknown type name 'enum64'; did you mean 'enum'?
1427 | enum64 trace_iterator_flags { TRACE_FLAGS };
| ^~~~~~
| enumBut below (C++11/C23 or clang/gcc extension) passed.
-enum trace_iterator_flags { TRACE_FLAGS }; +enum trace_iterator_flags : uint64_t { TRACE_FLAGS };
So let's use this.
Thanks,
Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
Steven Rostedt
maintainerRe: [PATCH v3 1/2] tracing: Allow tracer to add more than 32 options
In reply to Masami Hiramatsu (Google)Quoted 6 lines · On Sat, 18 Oct 2025 00:01:30 +0900 Masami Hiramatsu (Google)…
On Wed, 15 Oct 2025 17:20:20 -0400
Steven Rostedt <[email protected]> wrote:I could not find any other enum64 usage, so I doubt it is available. (Does it depend on compiler?)
It seems C23 standard support it...
Bah, I thought I saw it used, but it appears it's BPF that does something special.
Quoted 8 lines
Quoted 5 lines
Not to mention, using const u64 requires saving these numbers in an address and referencing them, instead of doing it inlined in text. That is, using u64 instead of enum64 is both slower and wastes more memory.
Yeah, I expected that the compiler could easily optimize correctly, but maybe not?
I doubt it. The values are exported to be allowed to be used in other files, so I doubt it can optimize it.
The only thing I can think of is to unravel the enum into a bunch of #defines, that have the bit shifts.
-- Steve