Messages in this thread Patch in this message | | | From | Denys Vlasenko <> | Subject | [PATCH 2/2] x86/asm/entry/32: Remove unnecessary optimization in stub32_clone | Date | Wed, 3 Jun 2015 15:58:50 +0200 |
| |
Really swap arguments #4 and #5 in stub32_clone instead of "optimizing" it into a move.
Yes, tls_val is currently unused. Yes, on some CPUs XCHG is a little bit more expensive than MOV. But a cycle or two on an expensive syscall like clone() is way below noise floor, and this optimization is simply not worth the obfuscation of logic.
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com> CC: Josh Triplett <josh@joshtriplett.org> CC: Linus Torvalds <torvalds@linux-foundation.org> CC: Steven Rostedt <rostedt@goodmis.org> CC: Ingo Molnar <mingo@kernel.org> CC: Borislav Petkov <bp@alien8.de> CC: "H. Peter Anvin" <hpa@zytor.com> CC: Andy Lutomirski <luto@amacapital.net> CC: Oleg Nesterov <oleg@redhat.com> CC: Frederic Weisbecker <fweisbec@gmail.com> CC: Alexei Starovoitov <ast@plumgrid.com> CC: Will Drewry <wad@chromium.org> CC: Kees Cook <keescook@chromium.org> CC: x86@kernel.org CC: linux-kernel@vger.kernel.org ---
This is a resend.
There was a patch by Josh Triplett "x86: Opt into HAVE_COPY_THREAD_TLS, for both 32-bit and 64-bit" sent on May 11, which does the same thing as part of a bigger cleanup. He was supportive of this patch because of comments. He will simply have to drop one hunk from his patch.
arch/x86/ia32/ia32entry.S | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-)
diff --git a/arch/x86/ia32/ia32entry.S b/arch/x86/ia32/ia32entry.S index 8e72256..0c302d0 100644 --- a/arch/x86/ia32/ia32entry.S +++ b/arch/x86/ia32/ia32entry.S @@ -567,11 +567,9 @@ GLOBAL(stub32_clone) * 32-bit clone API is clone(..., int tls_val, int *child_tidptr). * 64-bit clone API is clone(..., int *child_tidptr, int tls_val). * Native 64-bit kernel's sys_clone() implements the latter. - * We need to swap args here. But since tls_val is in fact ignored - * by sys_clone(), we can get away with an assignment - * (arg4 = arg5) instead of a full swap: + * We need to swap args here: */ - mov %r8, %rcx + xchg %r8, %rcx jmp ia32_ptregs_common ALIGN -- 1.8.1.4
| |