Re: [PATCH] Linux Kernel Markers 0.7 for 2.6.17 (with type checking!)

[Mathieu Desnoyers <compudj@krystal.dyndns.org>]

* Jeremy Fitzhardinge (jeremy@goop.org) wrote:
> Mathieu Desnoyers wrote:
> >* Jeremy Fitzhardinge (jeremy@goop.org) wrote:
> >  
> >>Mathieu Desnoyers wrote:
> >>    
> >>>#define MARK_SYM(name) \
> >>>       do { \
> >>>               __label__ here; \
> >>>               volatile static void *__mark_kprobe_##name \
> >>>                       asm (MARK_CALL_PREFIX#name) \
> >>>                       __attribute__((unused)) = &&here; \
> >>>here: \
> >>>               do { } while(0); \
> >>>       } while(0)
> >>>
> >>>Which fixes the problem. Some tests showed me that the compiler does not 
> >>>unroll
> >>>an otherwise unrolled loop when this specific macro is called. (test 
> >>>done with
> >>>-funroll-all-loops).
> >>>      
> >>Eh?  I thought you wanted to avoid changing the generated code?  
> >>Inhibiting loop unrolling could be a pretty large change...
> >>
> >>    
> >
> >Yes, if possible. But letting gcc duplicate those symbols brings many 
> >questions,
> >such as : how can we name each of them differently ? Is there any way to
> >automatically increment an "identifier" counter in assembly ?
> 
> Use a section instead:
> 
> struct marker {
> 	const char *name;
> 	const void *location;
> };
> 
> #define MARKER_SYM(name)
> 	do {
> 		__label__ here;
> 	here:	asm volatile(".section \".markers\"; .long %0, %1; 
> 	.previous" : : "m" (#name), "m" (*&&here));\
> 	} while(0);
> 
> Not a linker symbol, but it does let you find all the places containing 
> a particular mark.
> 

Very clever idea, as it lessens the impact on the compiler optimisations. Any
ideas about how we could fit in a list of "read" memory constraints based on a
vargs list in the macro ?

Mathieu

OpenPGP public key:              http://krystal.dyndns.org:8080/key/compudj.gpg
Key fingerprint:     8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68