[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <1363449458.25967.76.camel@gandalf.local.home>
Date: Sat, 16 Mar 2013 11:57:38 -0400
From: Steven Rostedt <rostedt@...dmis.org>
To: Joe Perches <joe@...ches.com>
Cc: Alexander Viro <viro@...iv.linux.org.uk>,
Andrew Morton <akpm@...ux-foundation.org>,
LKML <linux-kernel@...r.kernel.org>
Subject: Re: [RFC PATCH] seq_file: Use seq_puts when seq_printf has only a
format with no args
My macro nastiness is contagious ;-)
On Sat, 2013-03-16 at 06:50 -0700, Joe Perches wrote:
> Instead of converting the 800 or so uses of seq_printf with
> a constant format (without a % substitution) to seq_puts,
> maybe there's another way to slightly speed up these outputs.
>
> Taking a similar approach to commit abd84d60eb
> ("tracing: Optimize trace_printk() with one arg to use trace_puts()")
> use the preprocessor to convert seq_printf(seq, "string constant")
> to seq_puts(seq, "string constant")
>
> By stringifying __VA_ARGS__, we can, at compile time, determine
> the number of args that are being passed to seq_printf() and
> call seq_puts or seq_printf appropriately.
>
> The actual function definition for seq_printf must now
> be enclosed in parenthesis to avoid further macro expansion.
>
> Signed-off-by: Joe Perches <joe@...ches.com>
> ---
> fs/seq_file.c | 7 ++++++-
> include/linux/seq_file.h | 24 ++++++++++++++++++++++++
> 2 files changed, 30 insertions(+), 1 deletion(-)
>
> diff --git a/fs/seq_file.c b/fs/seq_file.c
> index 38bb59f..d3a957d 100644
> --- a/fs/seq_file.c
> +++ b/fs/seq_file.c
> @@ -405,7 +405,12 @@ int seq_vprintf(struct seq_file *m, const char *f, va_list args)
> }
> EXPORT_SYMBOL(seq_vprintf);
>
> -int seq_printf(struct seq_file *m, const char *f, ...)
> +/*
> + * seq_printf is also a macro that expands to seq_printf or seq_puts.
> + * This means that the actual function definition must be in parenthesis
> + * to prevent the preprocessor from expanding this function name again.
> + */
> +int (seq_printf)(struct seq_file *m, const char *f, ...)
That's rather ugly. Why not just #undef seq_printf before defining it?
> {
> int ret;
> va_list args;
> diff --git a/include/linux/seq_file.h b/include/linux/seq_file.h
> index 68a04a3..7255f01 100644
> --- a/include/linux/seq_file.h
> +++ b/include/linux/seq_file.h
> @@ -7,6 +7,7 @@
> #include <linux/mutex.h>
> #include <linux/cpumask.h>
> #include <linux/nodemask.h>
> +#include <linux/stringify.h>
>
> struct seq_operations;
> struct file;
> @@ -92,6 +93,29 @@ int seq_write(struct seq_file *seq, const void *data, size_t len);
> __printf(2, 3) int seq_printf(struct seq_file *, const char *, ...);
> __printf(2, 0) int seq_vprintf(struct seq_file *, const char *, va_list args);
>
> +/*
> + * A little optimization trick is done here. If there's only one
> + * argument, there's no need to scan the string for printf formats.
> + * seq_puts() will suffice. But how can we take advantage of
> + * using seq_puts() when seq_printf() has only one argument?
> + * By stringifying the args and checking the size we can tell
> + * whether or not there are args. __stringify(__VA_ARGS__) will
> + * turn into "" with a size of 1 when there are no args, anything
> + * else will be bigger. All we need to do is define a string to this,
> + * and then take its size and compare to 1. If it's bigger, use
> + * seq_printf() otherwise, optimize it to seq_puts(). Then just
> + * let gcc optimize the rest. The actual function definition of
> + * seq_printf must be (seq_printf) to prevent further macro expansion.
> + */
> +#define seq_printf(seq, fmt, ...) \
> +do { \
> + char va_args[] = __stringify(__VA_ARGS__); \
Interesting, how did you not get errors with multiple args? Although, I
did this macro testing in a normal C file and not in the kernel. Maybe I
screwed something up there and it would have worked for me too :-/
Anyway, not making va_args a whacky name is dangerous. This is why I add
those crazy underscores. If someone does:
var = 1;
va_args[] = "abc";
seq_printf(m, "%d %s", var, va_args);
What will be printed is:
1 var, va_args
That will be very confusing to people.
> + if (sizeof(va_args) > 1) \
> + seq_printf(seq, fmt, ##__VA_ARGS__); \
> + else \
> + seq_puts(seq, fmt); \
> +} while (0)
BTW, you need to return a value.
> +
> int seq_path(struct seq_file *, const struct path *, const char *);
> int seq_dentry(struct seq_file *, struct dentry *, const char *);
> int seq_path_root(struct seq_file *m, const struct path *path,
>
Here's an update to your patch, although I didn't change va_args.
Signed-off-by: Steven Rostedt <rostedt@...dmis.org>
Index: linux-trace.git/fs/seq_file.c
===================================================================
--- linux-trace.git.orig/fs/seq_file.c
+++ linux-trace.git/fs/seq_file.c
@@ -407,10 +407,11 @@ EXPORT_SYMBOL(seq_vprintf);
/*
* seq_printf is also a macro that expands to seq_printf or seq_puts.
- * This means that the actual function definition must be in parenthesis
- * to prevent the preprocessor from expanding this function name again.
+ * Undefine the macro before defining the actual function.
*/
-int (seq_printf)(struct seq_file *m, const char *f, ...)
+#undef seq_printf
+
+int seq_printf(struct seq_file *m, const char *f, ...)
{
int ret;
va_list args;
Index: linux-trace.git/include/linux/seq_file.h
===================================================================
--- linux-trace.git.orig/include/linux/seq_file.h
+++ linux-trace.git/include/linux/seq_file.h
@@ -108,13 +108,15 @@ __printf(2, 0) int seq_vprintf(struct se
* seq_printf must be (seq_printf) to prevent further macro expansion.
*/
#define seq_printf(seq, fmt, ...) \
-do { \
+({ \
char va_args[] = __stringify(__VA_ARGS__); \
+ int _____ret; \
if (sizeof(va_args) > 1) \
- seq_printf(seq, fmt, ##__VA_ARGS__); \
+ _____ret = seq_printf(seq, fmt, ##__VA_ARGS__); \
else \
- seq_puts(seq, fmt); \
-} while (0)
+ _____ret = seq_puts(seq, fmt); \
+ _____ret; \
+})
int seq_path(struct seq_file *, const struct path *, const char *);
int seq_dentry(struct seq_file *, struct dentry *, const char *);
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists