Re: [RFC PATCH] seq_file: Use seq_puts when seq_printf has only aformat with no args

[Steven Rostedt]

On Sat, 2013-03-16 at 18:01 +0000, Al Viro wrote:
> On Sat, Mar 16, 2013 at 10:51:18AM -0700, Joe Perches wrote:
> > > This is certainly a neat trick.
> > > 
> > > But I don't really like the fact that it complicates things for every
> > > future code reader, especially when a trivial change in the caller
> > > would accomplish the same thing.  Do you have any idea how much
> > > performance we would gain in exchange for the complication?
> > 
> > Nope.  I believe it's trivial in any case.
> > I just saw Steven's trace hack and thought of seq_printk.
> > 
> > Is there a real performance sensitive seq_printf anywhere?
> 
> ... and _that_ is the question that should've been asked first.

I totally agree with you. I've been avoiding the "performance
sensitivity" question because of my narcissistic enjoyment of my macro
cleverness ;-)
And also because I have no "F"'n life.

I just figured someone else will point out the lack of clothes the
Emperor has on.

> 
> > It's trivial to replace seq_printf("constant") with
> > seq_puts but there are over a thousand of them.
> > 
> > It may be better to just leave everything as-is.
> 
> Quite.  Note that it's not equivalent to gcc treatment of printf/puts -
> there we have cases when it *is* a real hotpath (and I seriously suspect
> that it's in part driven by desire to discourage people from uglifying
> source by manual equivalents of that micro-optimization).  Moreover,
> glibc printf at least used to be heavy; kernel-side we are nowhere near
> that bad.

It's also a very hot path in tracing. One reason I only implemented the
macro trick with trace_printk() and not printk() nor seq_printk() is
because I knew those were not hot paths. The reason I created
trace_puts() in the first place, is because I had a bug I was trying to
debug where a trace_printk() would actually make the bug go away. It
added too much of an impact to get the race to trigger. But the
trace_puts() was able to do the trace and still have the bug trigger,
and I was able to debug the problem.

But I enjoyed this conversation while it lasted. Sorry it took up your
time. But it did call to attention that these macros that create
variables should probably have a naming policy to avoid macro traps.

-- Steve


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/