[PATCH] [v2] tracing: use %ps format string to print symbols

[PATCH] [v2] tracing: use %ps format string to print symbols

[Arnd Bergmann]

From: Arnd Bergmann <arnd@arndb.de>

clang started warning about excessive stack usage in
hist_trigger_print_key()

kernel/trace/trace_events_hist.c:4723:13: error: stack frame size (1336) exceeds limit (1024) in function 'hist_trigger_print_key' [-Werror,-Wframe-larger-than]

The problem is that there are two 512-byte arrays on the stack if
hist_trigger_stacktrace_print() gets inlined. I don't think this has
changed in the past five years, but something probably changed the
inlining decisions made by the compiler, so the problem is now made
more obvious.

Rather than printing the symbol names into separate buffers, it
seems we can simply use the special %ps format string modifier
to print the pointers symbolically and get rid of both buffers.

Marking hist_trigger_stacktrace_print() would be a simpler
way of avoiding the warning, but that would not address the
excessive stack usage.

Fixes: 69a0200c2e25 ("tracing: Add hist trigger support for stacktraces as keys")
Link: https://lore.kernel.org/all/20211015095704.49a99859@gandalf.local.home/
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
Changes in v2:
 - Use %pS instead of %ps to include offset in third string
 - add (void*) cast to avoid compile-time warnings

This is still only compile-tested to ensure that the warning
goes away, I have not validated what it does to the formatting.
---
 kernel/trace/trace_events_hist.c | 15 +++++----------
 1 file changed, 5 insertions(+), 10 deletions(-)

diff --git a/kernel/trace/trace_events_hist.c b/kernel/trace/trace_events_hist.c
index a6061a69aa84..d6d4362c273c 100644
--- a/kernel/trace/trace_events_hist.c
+++ b/kernel/trace/trace_events_hist.c
@@ -4706,7 +4706,6 @@ static void hist_trigger_stacktrace_print(struct seq_file *m,
 					  unsigned long *stacktrace_entries,
 					  unsigned int max_entries)
 {
-	char str[KSYM_SYMBOL_LEN];
 	unsigned int spaces = 8;
 	unsigned int i;
 
@@ -4715,8 +4714,7 @@ static void hist_trigger_stacktrace_print(struct seq_file *m,
 			return;
 
 		seq_printf(m, "%*c", 1 + spaces, ' ');
-		sprint_symbol(str, stacktrace_entries[i]);
-		seq_printf(m, "%s\n", str);
+		seq_printf(m, "%pS\n", (void*)stacktrace_entries[i]);
 	}
 }
 
@@ -4726,7 +4724,6 @@ static void hist_trigger_print_key(struct seq_file *m,
 				   struct tracing_map_elt *elt)
 {
 	struct hist_field *key_field;
-	char str[KSYM_SYMBOL_LEN];
 	bool multiline = false;
 	const char *field_name;
 	unsigned int i;
@@ -4747,14 +4744,12 @@ static void hist_trigger_print_key(struct seq_file *m,
 			seq_printf(m, "%s: %llx", field_name, uval);
 		} else if (key_field->flags & HIST_FIELD_FL_SYM) {
 			uval = *(u64 *)(key + key_field->offset);
-			sprint_symbol_no_offset(str, uval);
-			seq_printf(m, "%s: [%llx] %-45s", field_name,
-				   uval, str);
+			seq_printf(m, "%s: [%llx] %-45ps", field_name,
+				   uval, (void *)uval);
 		} else if (key_field->flags & HIST_FIELD_FL_SYM_OFFSET) {
 			uval = *(u64 *)(key + key_field->offset);
-			sprint_symbol(str, uval);
-			seq_printf(m, "%s: [%llx] %-55s", field_name,
-				   uval, str);
+			seq_printf(m, "%s: [%llx] %-55pS", field_name,
+				   uval, (void *)uval);
 		} else if (key_field->flags & HIST_FIELD_FL_EXECNAME) {
 			struct hist_elt_data *elt_data = elt->private_data;
 			char *comm;
-- 
2.29.2

Re: [PATCH] [v2] tracing: use %ps format string to print symbols

[Tom Zanussi]

Hi Arnd,

On Mon, 2021-10-18 at 15:25 +0200, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> clang started warning about excessive stack usage in
> hist_trigger_print_key()
> 
> kernel/trace/trace_events_hist.c:4723:13: error: stack frame size
> (1336) exceeds limit (1024) in function 'hist_trigger_print_key' [-
> Werror,-Wframe-larger-than]
> 
> The problem is that there are two 512-byte arrays on the stack if
> hist_trigger_stacktrace_print() gets inlined. I don't think this has
> changed in the past five years, but something probably changed the
> inlining decisions made by the compiler, so the problem is now made
> more obvious.
> 
> Rather than printing the symbol names into separate buffers, it
> seems we can simply use the special %ps format string modifier
> to print the pointers symbolically and get rid of both buffers.
> 
> Marking hist_trigger_stacktrace_print() would be a simpler
> way of avoiding the warning, but that would not address the
> excessive stack usage.
> 
> Fixes: 69a0200c2e25 ("tracing: Add hist trigger support for
> stacktraces as keys")
> Link: 
> https://lore.kernel.org/all/20211015095704.49a99859@gandalf.local.home/
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
> Changes in v2:
>  - Use %pS instead of %ps to include offset in third string
>  - add (void*) cast to avoid compile-time warnings
> 
> This is still only compile-tested to ensure that the warning
> goes away, I have not validated what it does to the formatting.


This looks fine to me, thanks for the patch!

Reviewed-by: Tom Zanussi <zanussi@kernel.org>
Tested-by: Tom Zanussi <zanussi@kernel.org>


> ---
>  kernel/trace/trace_events_hist.c | 15 +++++----------
>  1 file changed, 5 insertions(+), 10 deletions(-)
> 
> diff --git a/kernel/trace/trace_events_hist.c
> b/kernel/trace/trace_events_hist.c
> index a6061a69aa84..d6d4362c273c 100644
> --- a/kernel/trace/trace_events_hist.c
> +++ b/kernel/trace/trace_events_hist.c
> @@ -4706,7 +4706,6 @@ static void
> hist_trigger_stacktrace_print(struct seq_file *m,
>  					  unsigned long
> *stacktrace_entries,
>  					  unsigned int max_entries)
>  {
> -	char str[KSYM_SYMBOL_LEN];
>  	unsigned int spaces = 8;
>  	unsigned int i;
>  
> @@ -4715,8 +4714,7 @@ static void
> hist_trigger_stacktrace_print(struct seq_file *m,
>  			return;
>  
>  		seq_printf(m, "%*c", 1 + spaces, ' ');
> -		sprint_symbol(str, stacktrace_entries[i]);
> -		seq_printf(m, "%s\n", str);
> +		seq_printf(m, "%pS\n", (void*)stacktrace_entries[i]);
>  	}
>  }
>  
> @@ -4726,7 +4724,6 @@ static void hist_trigger_print_key(struct
> seq_file *m,
>  				   struct tracing_map_elt *elt)
>  {
>  	struct hist_field *key_field;
> -	char str[KSYM_SYMBOL_LEN];
>  	bool multiline = false;
>  	const char *field_name;
>  	unsigned int i;
> @@ -4747,14 +4744,12 @@ static void hist_trigger_print_key(struct
> seq_file *m,
>  			seq_printf(m, "%s: %llx", field_name, uval);
>  		} else if (key_field->flags & HIST_FIELD_FL_SYM) {
>  			uval = *(u64 *)(key + key_field->offset);
> -			sprint_symbol_no_offset(str, uval);
> -			seq_printf(m, "%s: [%llx] %-45s", field_name,
> -				   uval, str);
> +			seq_printf(m, "%s: [%llx] %-45ps", field_name,
> +				   uval, (void *)uval);
>  		} else if (key_field->flags & HIST_FIELD_FL_SYM_OFFSET)
> {
>  			uval = *(u64 *)(key + key_field->offset);
> -			sprint_symbol(str, uval);
> -			seq_printf(m, "%s: [%llx] %-55s", field_name,
> -				   uval, str);
> +			seq_printf(m, "%s: [%llx] %-55pS", field_name,
> +				   uval, (void *)uval);
>  		} else if (key_field->flags & HIST_FIELD_FL_EXECNAME) {
>  			struct hist_elt_data *elt_data = elt-
> >private_data;
>  			char *comm;

Re: [PATCH] [v2] tracing: use %ps format string to print symbols

[Arnd Bergmann]

On Mon, Oct 18, 2021 at 4:42 PM Tom Zanussi <zanussi@kernel.org> wrote:
> On Mon, 2021-10-18 at 15:25 +0200, Arnd Bergmann wrote:

> This looks fine to me, thanks for the patch!
>
> Reviewed-by: Tom Zanussi <zanussi@kernel.org>
> Tested-by: Tom Zanussi <zanussi@kernel.org>

Thanks for testing!

> > -                     seq_printf(m, "%s: [%llx] %-45s", field_name,
> > -                                uval, str);
> > +                     seq_printf(m, "%s: [%llx] %-45ps", field_name,
> > +                                uval, (void *)uval);

This turns out to be still wrong on 32-bit, as 'uval' is a u64 and I
can't cast that to a pointer without a uintptr_t cast first.

I was testing randconfig builds with clang here, and for some reason
that didn't catch the bug, but I found it after going back to gcc testing.

Sending v3 now.

       Arnd