Message ID | 20120529212010.0152a083@kryten (mailing list archive) |
---|---|
State | Superseded, archived |
Headers | show |
On 05/29/2012 06:20 AM, Anton Blanchard wrote: > > In system call exit we currently clear RI and EE separately. An > mtmsrd is a slow operation and we can save cycles by doing it all > in one go. > > This does complicate things a bit - we have to be careful to restore > RI if we branch out before returning to userspace. > > On a POWER7 with virtual cputime disabled this patch improves the > null system call by 7%. > > Signed-off-by: Anton Blanchard <anton@samba.org> > --- > > Index: linux-build/arch/powerpc/kernel/entry_64.S > =================================================================== > --- linux-build.orig/arch/powerpc/kernel/entry_64.S 2012-05-28 18:23:33.374451416 +1000 > +++ linux-build/arch/powerpc/kernel/entry_64.S 2012-05-29 21:18:22.280934940 +1000 > @@ -197,7 +197,16 @@ syscall_exit: > wrteei 0 > #else > ld r10,PACAKMSR(r13) > - mtmsrd r10,1 > + /* > + * For performance reasons we clear RI the same time that we > + * clear EE. We only need to clear RI just before we restore r13 > + * below, but batching it with EE saves us one expensive mtmsrd call. > + * We have to be careful to restore RI if we branch anywhere from > + * here (eg syscall_exit_work). > + */ > + li r9,MSR_RI > + andc r11,r10,r9 > + mtmsrd r11,1 > #endif /* CONFIG_PPC_BOOK3E */ > > ld r9,TI_FLAGS(r12) > @@ -214,17 +223,6 @@ BEGIN_FTR_SECTION > END_FTR_SECTION_IFCLR(CPU_FTR_STCX_CHECKS_ADDRESS) > andi. r6,r8,MSR_PR > ld r4,_LINK(r1) > - /* > - * Clear RI before restoring r13. If we are returning to > - * userspace and we take an exception after restoring r13, > - * we end up corrupting the userspace r13 value. > - */ > -#ifdef CONFIG_PPC_BOOK3S > - /* No MSR:RI on BookE */ > - li r12,MSR_RI > - andc r11,r10,r12 > - mtmsrd r11,1 /* clear MSR.RI */ > -#endif /* CONFIG_PPC_BOOK3S */ > > beq- 1f > ACCOUNT_CPU_USER_EXIT(r11, r12) > @@ -271,6 +269,7 @@ syscall_enosys: > b syscall_exit > > syscall_exit_work: > + mtmsrd r10,1 /* Restore RI */ That mtmsrd needs an #ifdef CONFIG_PPC_BOOK3S -Scott
Index: linux-build/arch/powerpc/kernel/entry_64.S =================================================================== --- linux-build.orig/arch/powerpc/kernel/entry_64.S 2012-05-28 18:23:33.374451416 +1000 +++ linux-build/arch/powerpc/kernel/entry_64.S 2012-05-29 21:18:22.280934940 +1000 @@ -197,7 +197,16 @@ syscall_exit: wrteei 0 #else ld r10,PACAKMSR(r13) - mtmsrd r10,1 + /* + * For performance reasons we clear RI the same time that we + * clear EE. We only need to clear RI just before we restore r13 + * below, but batching it with EE saves us one expensive mtmsrd call. + * We have to be careful to restore RI if we branch anywhere from + * here (eg syscall_exit_work). + */ + li r9,MSR_RI + andc r11,r10,r9 + mtmsrd r11,1 #endif /* CONFIG_PPC_BOOK3E */ ld r9,TI_FLAGS(r12) @@ -214,17 +223,6 @@ BEGIN_FTR_SECTION END_FTR_SECTION_IFCLR(CPU_FTR_STCX_CHECKS_ADDRESS) andi. r6,r8,MSR_PR ld r4,_LINK(r1) - /* - * Clear RI before restoring r13. If we are returning to - * userspace and we take an exception after restoring r13, - * we end up corrupting the userspace r13 value. - */ -#ifdef CONFIG_PPC_BOOK3S - /* No MSR:RI on BookE */ - li r12,MSR_RI - andc r11,r10,r12 - mtmsrd r11,1 /* clear MSR.RI */ -#endif /* CONFIG_PPC_BOOK3S */ beq- 1f ACCOUNT_CPU_USER_EXIT(r11, r12) @@ -271,6 +269,7 @@ syscall_enosys: b syscall_exit syscall_exit_work: + mtmsrd r10,1 /* Restore RI */ /* If TIF_RESTOREALL is set, don't scribble on either r3 or ccr. If TIF_NOERROR is set, just save r3 as it is. */
In system call exit we currently clear RI and EE separately. An mtmsrd is a slow operation and we can save cycles by doing it all in one go. This does complicate things a bit - we have to be careful to restore RI if we branch out before returning to userspace. On a POWER7 with virtual cputime disabled this patch improves the null system call by 7%. Signed-off-by: Anton Blanchard <anton@samba.org> ---