Re: [RFC PATCH] preempt: Add a generic function to return the preemption string.

[Steven Rostedt]

On Thu, 9 Jan 2025 12:43:39 +0100
Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:

> Hurmph.. a bunch of those combinations are nonsense :/
> 
> Also, we have a string thing in sched_dynamic_show().
> 
> Can't we do something like:
> 
> const char *preempt_model_str(void)
> {
> 	static char buf[128];
> 	size_t off = 0, len = sizeof(buf), r;
> 	bool brace = IS_ENABLED(CONFIG_PREEMPT_RT) &&
> 		     (IS_ENABLED(CONFIG_PREEMPT_DYNAMIC) ||
> 		      IS_ENABLED(CONFIG_PREEMPT_LAZY));
> 
> 	if (IS_ENABLED(CONFIG_PREEMPT_BUILD)) {
> 		r = snprintf(buf + off, len, "PREEMPT");
> 		off += r;
> 		len -= r;
> 
> 		if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
> 			r = snprintf(buf + off, len, "%sRT%s", 
> 					brace ? "_{" : "_",
> 					brace ? "," : "");
> 			off += r;
> 			len -= r;
> 		}
> 
> 		if (IS_ENABLED(CONFIG_PREEMPT_DYNAMIC)) {
> 			snprintf(buf + off, len, "(%s)%s", 
> 				preempt_modes[preempt_dynamic_mode],
> 				brace ? "}" : "");
> 			retun buf;
> 		}
> 
> 		if (IS_ENABLED(CONFIG_PREEMPT_LAZY)) {
> 			snprintf(buf + off, len, "LAZY%s", 
> 				brace ? "}" : "");
> 			retun buf;
> 		}

I'm sorry, but I can't even tell what the above is doing without my brain
hurting. Why make code that was easy to read into a cryptic obfuscation? I
can't see this as an optimization as IS_ENABLED() is determined at compile
time.

-- Steve


> 
> 		return buf;
> 	}
> 
> 	if (IS_ENABLED(CONFIG_VOLUNTARY_BUILD))
> 		return "VOLUNTARY";
> 
> 	return "NONE";
> }