Hi Ingo,
Here is the patchset you requested. I did not port the marker reintegration
to your sched-devel tree though, because many changes happened since you have
done the original work.
It applies on top of sched-devel.git latest.
You will notice that I implemented what we discussed yesterday : using nops and
jump for the heavily optimized version of markers. Comments are welcome. Running
this with my ~120 LTTng markers on x86_32 detects 97% of the sites. 4 out of 120
did had to fall back on the standard immediate values because they have been
manipulated by gcc optimizations. The sched-devel.git port has been tested
on x86_32. Patches before the port are tested on x86_32 and x86_64.
Note that some folding of the immediate values patches could eventually be
required. At that point, add-all-cpus-option-to-stop-machine-run.patch would
become useless.
The series order is the following :
make-marker_debug-static.patch # in -mm
x86-nmi-safe-int3-and-page-fault.patch
check-for-breakpoint-in-text-poke-to-eliminate-bug-on.patch
#Kprobes mutex cleanup
kprobes-use-mutex-for-insn-pages.patch
kprobes-dont-use-kprobes-mutex-in-arch-code.patch
kprobes-declare-kprobes-mutex-static.patch
#Text Edit Lock (depends on Enhance DEBUG_RODATA and kprobes mutex cleanup)
text-edit-lock-architecture-independent-code.patch
text-edit-lock-kprobes-architecture-independent-support.patch
#
#Immediate Values
add-all-cpus-option-to-stop-machine-run.patch
immediate-values-architecture-independent-code.patch
immediate-values-kconfig-menu-in-embedded.patch
immediate-values-x86-optimization.patch
add-text-poke-and-sync-core-to-powerpc.patch
immediate-values-powerpc-optimization.patch
immediate-values-documentation.patch
immediate-values-support-init.patch
#
scheduler-profiling-use-immediate-values.patch
#
markers-remove-extra-format-argument.patch
markers-define-non-optimized-marker.patch
#
immediate-values-move-kprobes-x86-restore-interrupt-to-kdebug-h.patch
add-discard-section-to-x86.patch
im...
