Message ID | 20150720105417.24801.66440.stgit@thinktux.in.ibm.com (mailing list archive) |
---|---|
State | Rejected |
Delegated to: | Michael Ellerman |
Headers | show |
On Mon, Jul 20, 2015 at 04:24:22PM +0530, Ananth N Mavinakayanahalli wrote: > There is a potential gain to be had with a direct > jump instead of a breakpoint, but the caveats need to be traded off > with the complexity it brings in. > > For now, mark OPTPROBES na for powerpc. If we haven't yet investigated whether this feature is something beneficial to the powerpc kernel, it should just stay as TODO, not be marked as n/a. Segher
On Mon, 2015-07-20 at 08:41 -0500, Segher Boessenkool wrote: > On Mon, Jul 20, 2015 at 04:24:22PM +0530, Ananth N Mavinakayanahalli wrote: > > There is a potential gain to be had with a direct > > jump instead of a breakpoint, but the caveats need to be traded off > > with the complexity it brings in. > > > > For now, mark OPTPROBES na for powerpc. > > If we haven't yet investigated whether this feature is something beneficial > to the powerpc kernel, it should just stay as TODO, not be marked as n/a. Yep, agreed. As discussed in the other thread we'll leave this as TODO until we know it's either not a win or not worth the complexity. cheers
diff --git a/Documentation/features/debug/optprobes/arch-support.txt b/Documentation/features/debug/optprobes/arch-support.txt index b8999d8..73662f9 100644 --- a/Documentation/features/debug/optprobes/arch-support.txt +++ b/Documentation/features/debug/optprobes/arch-support.txt @@ -27,7 +27,7 @@ | nios2: | TODO | | openrisc: | TODO | | parisc: | TODO | - | powerpc: | TODO | + | powerpc: | na | | s390: | TODO | | score: | TODO | | sh: | TODO |
Kprobes uses a breakpoint instruction to trap into execution flow and the probed instruction is single-stepped from an alternate location. On some architectures like x86, under certain conditions, the OPTPROBES feature enables replacing the probed instruction with a jump instead, resulting in a significant perfomance boost (both the breakpoint and single-step exception is bypassed for each kprobe). Powerpc has an in-kernel instruction emulator. Kprobes on powerpc uses this emulator already and bypasses the single-step exception, with a lot less complexity. There is a potential gain to be had with a direct jump instead of a breakpoint, but the caveats need to be traded off with the complexity it brings in. For now, mark OPTPROBES na for powerpc. Signed-off-by: Ananth N Mavinakayanahalli <ananth@in.ibm.com> --- .../features/debug/optprobes/arch-support.txt | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)