From 9df58d2bb71abb35108d9133689c8cb5d18f91d8 Mon Sep 17 00:00:00 2001 From: malc Date: Tue, 13 Nov 2007 02:43:07 +0300 Subject: [PATCH] Update documentation to reflect recent findings --- README.macosx | 7 ++++++- README.solaris | 3 +++ README.windows | 2 ++ 3 files changed, 11 insertions(+), 1 deletion(-) diff --git a/README.macosx b/README.macosx index 67ae730..2d1fb13 100644 --- a/README.macosx +++ b/README.macosx @@ -1,4 +1,9 @@ Tested on Mac OS X Tiger (PPC) SMP - not tested Kernel accounting appears to be rather chaotic (... and wrong ...) -No kernel mode driver hence no workaround + According to cursory glance at XNU source code it also updates + accounting times during timer interrupts, in effect behavior + is mostly the same as the on in Linux (Documentation/cpu-load.txt) + +No kernel mode driver hence no workaround. Same RDTSC/RDPMC hackery can +be used on Intel versions. diff --git a/README.solaris b/README.solaris index 21b1d93..93869e8 100644 --- a/README.solaris +++ b/README.solaris @@ -1,3 +1,6 @@ Tested on Solaris 10 (X86_64 32bit mode) SMP - tested (AMD X2 - i.e. 2 CPUs) Kernel accounting _appears_ to be accurate + According to cursory glance at OpenSolaris source code the kernel + does it right (i.e. gethrtime is called by the scheduler to update + accounting) diff --git a/README.windows b/README.windows index 4570561..3bdfb8d 100644 --- a/README.windows +++ b/README.windows @@ -2,6 +2,8 @@ Tested on Windows XP SP2 (X86) SMP - tested (AMD X2 - i.e. 2 CPUs) Kernel accounting _IS_ inaccurate +http://blog.kalmbachnet.de/?postid=28 + Apparently RMClock (http://cpu.rightmark.org/products/rmclock.shtml) either somehow sets CR4.PCE or hooks to the #GP(0) handler so that RDPMC works from user-space. -- 2.11.4.GIT