DragonFly kernel List (threaded) for 2007-02
DragonFly BSD
DragonFly kernel List (threaded) for 2007-02
[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]

Re: dynamic smp aware modules (was: Re: Multiple kernel building/installing)

From: Matthew Dillon <dillon@xxxxxxxxxxxxxxxxxxxx>
Date: Fri, 23 Feb 2007 10:59:36 -0800 (PST)

:I think we should change this.  And actually it is quite easy.  Replace
:#ifdef SMP
:	code
:	if (is_smp) {
:		code
:	}
:and have some header doing
:const static int is_smp =3D
:# ifdef SMP
:	1;
:# else
:	0;
:# endif
:extern const int is_smp;
:and initialize is_smp appropriately in some kernel source.
:The kernel will be compiled optimized, because the compiler knows that is=
:_smp is a const value and thus can remove dead code.  Modules will take t=
:he penalty of checking this variable, but that's hardly an impact, becaus=
:e it will be in the cache anyways.  Just make sure that the variable is i=
:n its own cache line, which never changes.
:  simon

    This works to some degree.  The only issue is that #ifdef'd SMP code
    is not parsed at all if SMP is not defined, and so it may reference
    variables and procedures that otherwise do not exist (and are undeclared)
    for non-SMP kernels.

    if (IS_SMP) { ... }  code *IS* parsed by the compiler, so even
    though the compiler may optimize it out you could still end up with
    warnings and errors pertaining to variables that are not declared
    for SMP builds.

    My only other comment on the matter is that constant declarations
    replacing #ifdef's should probably be all-caps.  e.g.  IS_SMP rather
    then is_smp, to differentiate those variables from normal variables
    and make the code more readable.

    I use the const trick for the KTR logging code.  It works quite well
    given the caveats.


[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]