2007/11/5, Thomas Petazzoni <thomas.petazzoni@xxxxxxxx>: > Hi, > > Le Sat, 3 Nov 2007 13:33:12 +0000, > ninjaboy <n0b0dyn1nj4@xxxxxxxxx> a écrit : > > > I've seen that there are two section (.altinstructions and > > .altinstr_replacement), there are some "polimorphic" code that is > > replaced at runtime, now in arch/x86_64/lib/copy_user.S there are > > copy_user_generic_unrolled and copy_user_generic_string, that replaces > > the function that is used, my question is: where and how the decision > > on what code is used (in this case > > copy_user_generic_unrolled or copy_user_generic_string) for replace > > code? > > Here: > http://lxr.free-electrons.com/source/arch/x86_64/lib/copy_user.S#046 > > with the following line: > ALTERNATIVE_JUMP > X86_FEATURE_REP_GOOD,copy_user_generic_unrolled,copy_user_generic_string > > with the ALTERNATIVE_JUMP macro being defined at the top of the > copy_user.S file. It adds: > > * code to the normal text section to call the "default" function, the > unrolled version ; > > * code to the altinstr_replacement section to call the > copy_user_generic_string function ; > > * informations to the altinstructions section, which is processed at > boot time by arch/i386/kernel/alternative.c in order to know if an > alternative must be applied or not, depending on the presence/absence > on some processor features (here X86_FEATURE_REP_GOOD) > Yes, that makes sense, but why on AMD64 X2 is better rep insn than unrolled? In my userspace test case rep is slower than unrolled. thanks, -- noone is alone. -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
