[PATCH] sched/core: Remove wakeup_gran with single call calc_delta_fair

[PATCH] sched/core: Remove wakeup_gran with single call calc_delta_fair

[Abel Vesa]

This patch removes function wakeup_gran and its call by calling directly
calc_delta_fair function with sysctl_sched_wakeup_granularity and se as arguments.

Signed-off-by: Abel Vesa <abelvesa@gmail.com>
---
 kernel/sched/fair.c | 38 ++++++++++++++++----------------------
 1 file changed, 16 insertions(+), 22 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 46855d0..63a4051 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -4927,27 +4927,6 @@ migrate_task_rq_fair(struct task_struct *p, int next_cpu)
 }
 #endif /* CONFIG_SMP */
 
-static unsigned long
-wakeup_gran(struct sched_entity *curr, struct sched_entity *se)
-{
-	unsigned long gran = sysctl_sched_wakeup_granularity;
-
-	/*
-	 * Since its curr running now, convert the gran from real-time
-	 * to virtual-time in his units.
-	 *
-	 * By using 'se' instead of 'curr' we penalize light tasks, so
-	 * they get preempted easier. That is, if 'se' < 'curr' then
-	 * the resulting gran will be larger, therefore penalizing the
-	 * lighter, if otoh 'se' > 'curr' then the resulting gran will
-	 * be smaller, again penalizing the lighter task.
-	 *
-	 * This is especially important for buddies when the leftmost
-	 * task is higher priority than the buddy.
-	 */
-	return calc_delta_fair(gran, se);
-}
-
 /*
  * Should 'se' preempt 'curr'.
  *
@@ -4970,7 +4949,22 @@ wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se)
 	if (vdiff <= 0)
 		return -1;
 
-	gran = wakeup_gran(curr, se);
+	/*
+	 * Since its curr running now, convert the gran from real-time
+	 * to virtual-time in his units.
+	 *
+	 * By using 'se' instead of 'curr' we penalize light tasks, so
+	 * they get preempted easier. That is, if 'se' < 'curr' then
+	 * the resulting gran will be larger, therefore penalizing the
+	 * lighter, if otoh 'se' > 'curr' then the resulting gran will
+	 * be smaller, again penalizing the lighter task.
+	 *
+	 * This is especially important for buddies when the leftmost
+	 * task is higher priority than the buddy.
+	 */
+
+	gran = calc_delta_fair(sysctl_sched_wakeup_granularity, se);
+
 	if (vdiff > gran)
 		return 1;
 
-- 
1.9.1

Re: [PATCH] sched/core: Remove wakeup_gran with single call calc_delta_fair

[Peter Zijlstra]

On Mon, Apr 06, 2015 at 09:43:43PM +0300, Abel Vesa wrote:
> This patch removes function wakeup_gran and its call by calling directly
> calc_delta_fair function with sysctl_sched_wakeup_granularity and se as arguments.

But why?

Re: [PATCH] sched/core: Remove wakeup_gran with single call calc_delta_fair

[Abel Vesa]

On Tue, Apr 7, 2015 at 1:30 PM, Peter Zijlstra <peterz@infradead.org> wrote:
> On Mon, Apr 06, 2015 at 09:43:43PM +0300, Abel Vesa wrote:
>> This patch removes function wakeup_gran and its call by calling directly
>> calc_delta_fair function with sysctl_sched_wakeup_granularity and se as arguments.
>
> But why?

I was thinking that maybe reducing another function call would be a better idea.
I don't see what's the point of calling a 'wakeup_gran' just to
declare 'gran' as local variable and then
call calc_delta_fair instead of calling directly calc_delta_fair.
Plus, I think it seems more
readable to me. Of course, I might be wrong.

Re: [PATCH] sched/core: Remove wakeup_gran with single call calc_delta_fair

[Mike Galbraith]

On Tue, 2015-04-07 at 16:37 +0300, Abel Vesa wrote:
> On Tue, Apr 7, 2015 at 1:30 PM, Peter Zijlstra <peterz@infradead.org
> > wrote:
> > On Mon, Apr 06, 2015 at 09:43:43PM +0300, Abel Vesa wrote:
> > > This patch removes function wakeup_gran and its call by calling 
> > > directly
> > > calc_delta_fair function with sysctl_sched_wakeup_granularity 
> > > and se as arguments.
> > 
> > But why?
> 
> I was thinking that maybe reducing another function call would be a 
> better idea.
> I don't see what's the point of calling a 'wakeup_gran' just to
> declare 'gran' as local variable and then
> call calc_delta_fair instead of calling directly calc_delta_fair.
> Plus, I think it seems more
> readable to me. Of course, I might be wrong.

I expected it be inlined by gcc, but that didn't happen with gcc-4.8.3.

        -Mike

Re: [PATCH] sched/core: Remove wakeup_gran with single call calc_delta_fair

[Abel Vesa]

On Tue, Apr 7, 2015 at 7:55 PM, Mike Galbraith <umgwanakikbuti@gmail.com> wrote:
> On Tue, 2015-04-07 at 16:37 +0300, Abel Vesa wrote:
>> On Tue, Apr 7, 2015 at 1:30 PM, Peter Zijlstra <peterz@infradead.org
>> > wrote:
>> > On Mon, Apr 06, 2015 at 09:43:43PM +0300, Abel Vesa wrote:
>> > > This patch removes function wakeup_gran and its call by calling
>> > > directly
>> > > calc_delta_fair function with sysctl_sched_wakeup_granularity
>> > > and se as arguments.
>> >
>> > But why?
>>
>> I was thinking that maybe reducing another function call would be a
>> better idea.
>> I don't see what's the point of calling a 'wakeup_gran' just to
>> declare 'gran' as local variable and then
>> call calc_delta_fair instead of calling directly calc_delta_fair.
>> Plus, I think it seems more
>> readable to me. Of course, I might be wrong.
>
> I expected it be inlined by gcc, but that didn't happen with gcc-4.8.3.
>
>         -Mike

well, I really don't know what to say but I got curious too
and I compiled it with:

gcc (Ubuntu/Linaro 4.6.4-6ubuntu2) 4.6.4

and then did a objdump and got this:

0000010c <wakeup_preempt_entity>:
     10c:       e92d4038        push    {r3, r4, r5, lr}
     110:       e1c123d0        ldrd    r2, [r1, #48]   ; 0x30
     114:       e1c043d0        ldrd    r4, [r0, #48]   ; 0x30
     118:       e0544002        subs    r4, r4, r2
     11c:       e0c55003        sbc     r5, r5, r3
     120:       e3540001        cmp     r4, #1
     124:       e2d53000        sbcs    r3, r5, #0
     128:       ba00000b        blt     15c <wakeup_preempt_entity+0x50>
     12c:       e5912000        ldr     r2, [r1]
     130:       e59f3040        ldr     r3, [pc, #64]   ; 178
<wakeup_preempt_entity+0x6c>
     134:       e3520b01        cmp     r2, #1024       ; 0x400
     138:       e5930000        ldr     r0, [r3]
     13c:       1a000008        bne     164 <wakeup_preempt_entity+0x58>
     140:       e1a02000        mov     r2, r0
     144:       e3a03000        mov     r3, #0
     148:       e1520004        cmp     r2, r4
     14c:       e0d31005        sbcs    r1, r3, r5
     150:       a3a00000        movge   r0, #0
     154:       b3a00001        movlt   r0, #1
     158:       e8bd8038        pop     {r3, r4, r5, pc}
     15c:       e3e00000        mvn     r0, #0
     160:       e8bd8038        pop     {r3, r4, r5, pc}
     164:       e1a03001        mov     r3, r1
     168:       e3a02b01        mov     r2, #1024       ; 0x400
     16c:       e3a01000        mov     r1, #0
     170:       ebffffa2        bl      0 <__calc_delta>
     174:       eafffff1        b       140 <wakeup_preempt_entity+0x34>
     178:       00000000        .word   0x00000000

0000017c <sched_slice>:
.....

Now ofcourse, I'm not an ARM assembly language expert but I don't
think there is
any kind of branch instruction to something called 'wakeup_gran' , so I guess
Peter was right, at least this version of gcc says so.