| Message ID | 20210128130625.54076-2-ardb@kernel.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | arm64: rework NEON yielding to avoid scheduling from asm code | expand |
On Thu, Jan 28, 2021 at 02:06:17PM +0100, Ard Biesheuvel wrote: > Add a macro cond_yield that branches to a specified label when called if > the TIF_NEED_RESCHED flag is set and decreasing the preempt count would > make the task preemptible again, resulting in a schedule to occur. This > can be used by kernel mode SIMD code that keeps a lot of state in SIMD > registers, which would make chunking the input in order to perform the > cond_resched() check from C code disproportionately costly. > > Signed-off-by: Ard Biesheuvel <ardb@kernel.org> > --- > arch/arm64/include/asm/assembler.h | 16 ++++++++++++++++ > 1 file changed, 16 insertions(+) > > diff --git a/arch/arm64/include/asm/assembler.h b/arch/arm64/include/asm/assembler.h > index bf125c591116..5f977a7c6b43 100644 > --- a/arch/arm64/include/asm/assembler.h > +++ b/arch/arm64/include/asm/assembler.h > @@ -745,6 +745,22 @@ USER(\label, ic ivau, \tmp2) // invalidate I line PoU > .Lyield_out_\@ : > .endm > > + /* > + * Check whether preempt-disabled code should yield as soon as it > + * is able. This is the case if re-enabling preemption a single > + * time results in a preempt count of zero, and the TIF_NEED_RESCHED > + * flag is set. (Note that the latter is stored negated in the > + * top word of the thread_info::preempt_count field) > + */ > + .macro cond_yield, lbl:req, tmp:req > +#ifdef CONFIG_PREEMPTION > + get_current_task \tmp > + ldr \tmp, [\tmp, #TSK_TI_PREEMPT] > + cmp \tmp, #PREEMPT_DISABLE_OFFSET > + beq \lbl Fancy that, I didn't know the '.' was optional in "b.eq"! Anyway, a very similar code sequence exists inside if_will_cond_yield_neon, only it doesn't touch the flags. Can we use that sequence instead, and then use the new macro from there? Will
On Thu, Jan 28, 2021 at 08:24:01PM +0000, Will Deacon wrote: > On Thu, Jan 28, 2021 at 02:06:17PM +0100, Ard Biesheuvel wrote: > > Add a macro cond_yield that branches to a specified label when called if > > the TIF_NEED_RESCHED flag is set and decreasing the preempt count would > > make the task preemptible again, resulting in a schedule to occur. This > > can be used by kernel mode SIMD code that keeps a lot of state in SIMD > > registers, which would make chunking the input in order to perform the > > cond_resched() check from C code disproportionately costly. > > > > Signed-off-by: Ard Biesheuvel <ardb@kernel.org> > > --- > > arch/arm64/include/asm/assembler.h | 16 ++++++++++++++++ > > 1 file changed, 16 insertions(+) > > > > diff --git a/arch/arm64/include/asm/assembler.h b/arch/arm64/include/asm/assembler.h > > index bf125c591116..5f977a7c6b43 100644 > > --- a/arch/arm64/include/asm/assembler.h > > +++ b/arch/arm64/include/asm/assembler.h > > @@ -745,6 +745,22 @@ USER(\label, ic ivau, \tmp2) // invalidate I line PoU > > .Lyield_out_\@ : > > .endm > > > > + /* > > + * Check whether preempt-disabled code should yield as soon as it > > + * is able. This is the case if re-enabling preemption a single > > + * time results in a preempt count of zero, and the TIF_NEED_RESCHED > > + * flag is set. (Note that the latter is stored negated in the > > + * top word of the thread_info::preempt_count field) > > + */ > > + .macro cond_yield, lbl:req, tmp:req > > +#ifdef CONFIG_PREEMPTION > > + get_current_task \tmp > > + ldr \tmp, [\tmp, #TSK_TI_PREEMPT] > > + cmp \tmp, #PREEMPT_DISABLE_OFFSET > > + beq \lbl > > Fancy that, I didn't know the '.' was optional in "b.eq"! > > Anyway, a very similar code sequence exists inside if_will_cond_yield_neon, > only it doesn't touch the flags. Can we use that sequence instead, and then > use the new macro from there? ... and now I noticed the last patch :) But it would still be nice not to clobber the flags inside the macro. Will
On Thu, 28 Jan 2021 at 21:25, Will Deacon <will@kernel.org> wrote: > > On Thu, Jan 28, 2021 at 08:24:01PM +0000, Will Deacon wrote: > > On Thu, Jan 28, 2021 at 02:06:17PM +0100, Ard Biesheuvel wrote: > > > Add a macro cond_yield that branches to a specified label when called if > > > the TIF_NEED_RESCHED flag is set and decreasing the preempt count would > > > make the task preemptible again, resulting in a schedule to occur. This > > > can be used by kernel mode SIMD code that keeps a lot of state in SIMD > > > registers, which would make chunking the input in order to perform the > > > cond_resched() check from C code disproportionately costly. > > > > > > Signed-off-by: Ard Biesheuvel <ardb@kernel.org> > > > --- > > > arch/arm64/include/asm/assembler.h | 16 ++++++++++++++++ > > > 1 file changed, 16 insertions(+) > > > > > > diff --git a/arch/arm64/include/asm/assembler.h b/arch/arm64/include/asm/assembler.h > > > index bf125c591116..5f977a7c6b43 100644 > > > --- a/arch/arm64/include/asm/assembler.h > > > +++ b/arch/arm64/include/asm/assembler.h > > > @@ -745,6 +745,22 @@ USER(\label, ic ivau, \tmp2) // invalidate I line PoU > > > .Lyield_out_\@ : > > > .endm > > > > > > + /* > > > + * Check whether preempt-disabled code should yield as soon as it > > > + * is able. This is the case if re-enabling preemption a single > > > + * time results in a preempt count of zero, and the TIF_NEED_RESCHED > > > + * flag is set. (Note that the latter is stored negated in the > > > + * top word of the thread_info::preempt_count field) > > > + */ > > > + .macro cond_yield, lbl:req, tmp:req > > > +#ifdef CONFIG_PREEMPTION > > > + get_current_task \tmp > > > + ldr \tmp, [\tmp, #TSK_TI_PREEMPT] > > > + cmp \tmp, #PREEMPT_DISABLE_OFFSET > > > + beq \lbl > > > > Fancy that, I didn't know the '.' was optional in "b.eq"! > > > > Anyway, a very similar code sequence exists inside if_will_cond_yield_neon, > > only it doesn't touch the flags. Can we use that sequence instead, and then > > use the new macro from there? > > ... and now I noticed the last patch :) > > But it would still be nice not to clobber the flags inside the macro. > Yeah, that's a good point - I did not consider that. I'll fix that for v2.
diff --git a/arch/arm64/include/asm/assembler.h b/arch/arm64/include/asm/assembler.h index bf125c591116..5f977a7c6b43 100644 --- a/arch/arm64/include/asm/assembler.h +++ b/arch/arm64/include/asm/assembler.h @@ -745,6 +745,22 @@ USER(\label, ic ivau, \tmp2) // invalidate I line PoU .Lyield_out_\@ : .endm + /* + * Check whether preempt-disabled code should yield as soon as it + * is able. This is the case if re-enabling preemption a single + * time results in a preempt count of zero, and the TIF_NEED_RESCHED + * flag is set. (Note that the latter is stored negated in the + * top word of the thread_info::preempt_count field) + */ + .macro cond_yield, lbl:req, tmp:req +#ifdef CONFIG_PREEMPTION + get_current_task \tmp + ldr \tmp, [\tmp, #TSK_TI_PREEMPT] + cmp \tmp, #PREEMPT_DISABLE_OFFSET + beq \lbl +#endif + .endm + /* * This macro emits a program property note section identifying * architecture features which require special handling, mainly for
Add a macro cond_yield that branches to a specified label when called if the TIF_NEED_RESCHED flag is set and decreasing the preempt count would make the task preemptible again, resulting in a schedule to occur. This can be used by kernel mode SIMD code that keeps a lot of state in SIMD registers, which would make chunking the input in order to perform the cond_resched() check from C code disproportionately costly. Signed-off-by: Ard Biesheuvel <ardb@kernel.org> --- arch/arm64/include/asm/assembler.h | 16 ++++++++++++++++ 1 file changed, 16 insertions(+)