Skip to contentSkip to conversation

entry: preempt_schedule_irq() callers scrub

PATCH Applied Valentin Schneider 10 participants 30 messages Mar 11, 2019 – May 10, 2019

Subsystems (13)
x86 PowerPC MIPS RISC-V +9 more subsystems

v1 of 2 Applied Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover entry: preempt_schedule_irq() callers scrub

Hi,

This is the continuation of [1] where I'm hunting down preempt_schedule_irq() callers because of [2].

I told myself the best way to get this moving forward wouldn't be to write doc about it, but to go write some fixes and get some discussions going, which is what this patch-set is about.

I've looked at users of preempt_schedule_irq(), and made sure they didn't have one of those useless loops. The list of offenders is:

$ grep -r -I "preempt_schedule_irq" arch/ | cut -d/ -f2 | sort | uniq

arc
arm
arm64
c6x
csky
h8300
ia64
m68k
microblaze
mips
nds32
nios2
parisc
powerpc
riscv
s390
sh
sparc
x86
xtensa

Regarding that loop, archs seem to fall in 3 categories:
A) Those that don't have the loop
B) Those that have a small need_resched() loop around the preempt_schedule_irq() callsite
C) Those that branch to some more generic code further up the entry code and eventually branch back to preempt_schedule_irq()

arc, m68k, nios2 fall in A)
sparc, ia64, s390 fall in C)
all the others fall in B)

I've written patches for B) and C) EXCEPT for ia64 and s390 because I haven't been able to tell if it's actually fine to kill that "long jump" (and maybe I'm wrong on sparc). Hopefully folks who understand what goes on in there might be able to shed some light.

Also, since I sent patches for arm & arm64 in [1] I'm not including them here.

Boot-tested on:
- x86

Build-tested on:
- h8300
- c6x
- powerpc
- mips
- nds32
- microblaze
- sparc
- xtensa

Thanks,
Valentin

[1]: https://lore.kernel.org/lkml/[email protected]/
[2]: https://lore.kernel.org/lkml/[email protected]/

Valentin Schneider (14):
sched/core: Fix preempt_schedule() interrupt return comment
c6x: entry: Remove unneeded need_resched() loop
csky: entry: Remove unneeded need_resched() loop
h8300: entry: Remove unneeded need_resched() loop
microblaze: entry: Remove unneeded need_resched() loop
MIPS: entry: Remove unneeded need_resched() loop
nds32: ex-exit: Remove unneeded need_resched() loop
powerpc: entry: Remove unneeded need_resched() loop
RISC-V: entry: Remove unneeded need_resched() loop
sh: entry: Remove unneeded need_resched() loop
sh64: entry: Remove unneeded need_resched() loop
sparc64: rtrap: Remove unneeded need_resched() loop x86/entry: Remove unneeded need_resched() loop
xtensa: entry: Remove unneeded need_resched() loop

 arch/c6x/kernel/entry.S        | 3 +--
 arch/csky/kernel/entry.S       | 4 ----
 arch/h8300/kernel/entry.S      | 3 +--
 arch/microblaze/kernel/entry.S | 5 -----
 arch/mips/kernel/entry.S       | 3 +--
 arch/nds32/kernel/ex-exit.S    | 4 ++--
 arch/powerpc/kernel/entry_32.S | 6 +-----
 arch/powerpc/kernel/entry_64.S | 8 +-------
 arch/riscv/kernel/entry.S      | 3 +--
 arch/sh/kernel/cpu/sh5/entry.S | 5 +----
 arch/sh/kernel/entry-common.S  | 4 +---
 arch/sparc/kernel/rtrap_64.S   | 1 -
 arch/x86/entry/entry_32.S      | 3 +--
 arch/x86/entry/entry_64.S      | 3 +--
 arch/xtensa/kernel/entry.S     | 2 +-
 kernel/sched/core.c            | 7 +++----
 16 files changed, 16 insertions(+), 48 deletions(-)

--
2.20.1

#1 lore raw

PATCH 1/14 sched/core: Fix preempt_schedule() interrupt return comment

preempt_schedule_irq() is the one that should be called on return from interrupt, clean up the comment to avoid any ambiguity.

Cc Ingo Molnar <[email protected]>
Cc Peter Zijlstra <[email protected]>
---
 kernel/sched/core.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
kernel/sched/core.c +3 -4
index ead464a0f2e5..78c5aaa28278 100644
@@ -3652,9 +3652,8 @@ static void __sched notrace preempt_schedule_common(void)
 
 #ifdef CONFIG_PREEMPT
 /*
- * this is the entry point to schedule() from in-kernel preemption
- * off of preempt_enable. Kernel preemptions off return from interrupt
- * occur there and call schedule directly.
+ * This is the entry point to schedule() from in-kernel preemption
+ * off of preempt_enable.
  */
 asmlinkage __visible void __sched notrace preempt_schedule(void)
 {
@@ -3725,7 +3724,7 @@ EXPORT_SYMBOL_GPL(preempt_schedule_notrace);
 #endif /* CONFIG_PREEMPT */
 
 /*
- * this is the entry point to schedule() from kernel preemption
+ * This is the entry point to schedule() from kernel preemption
  * off of irq context.
  * Note, that this is called and return with irqs disabled. This will
  * protect us against recursive calling from irq.

--
2.20.1

#2 lore raw

PATCH 2/14 c6x: entry: Remove unneeded need_resched() loop

In reply to Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

---
 arch/c6x/kernel/entry.S | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
arch/c6x/kernel/entry.S +1 -2
index 2721c90b0121..17926e942edb 100644
@@ -567,7 +567,6 @@ resume_kernel:
 	NOP	4
  [A1]	BNOP	.S2	restore_all,5
 
-preempt_schedule:
 	GET_THREAD_INFO A2
 	LDW	.D1T1	*+A2(THREAD_INFO_FLAGS),A1
 #ifdef CONFIG_C6X_BIG_KERNEL
@@ -584,7 +583,7 @@ preempt_schedule:
 #else
 	B	.S2	preempt_schedule_irq
 #endif
-	ADDKPC	.S2	preempt_schedule,B3,4
+	ADDKPC	.S2	restore_all,B3,4
 #endif /* CONFIG_PREEMPT */
 
 ENTRY(enable_exception)
Signature · 1 line
2.20.1
Acked-by on v1 2/14
#3 lore raw

Re: [PATCH 02/14] c6x: entry: Remove unneeded need_resched() loop

Quoted 25 lines · Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

---
 arch/c6x/kernel/entry.S | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
arch/c6x/kernel/entry.S +0 -1
index 2721c90b0121..17926e942edb 100644
@@ -567,7 +567,6 @@ resume_kernel:
 	NOP	4
  [A1]	BNOP	.S2	restore_all,5
 
-preempt_schedule:
 	GET_THREAD_INFO A2
 	LDW	.D1T1	*+A2(THREAD_INFO_FLAGS),A1
 #ifdef CONFIG_C6X_BIG_KERNEL

arch/c6x/kernel/entry.S preempt_schedule: · line 589

Quoted 8 lines
arch/c6x/kernel/entry.S +1 -1
@@ -584,7 +583,7 @@ preempt_schedule:
 #else
 	B	.S2	preempt_schedule_irq
 #endif
-	ADDKPC	.S2	preempt_schedule,B3,4
+	ADDKPC	.S2	restore_all,B3,4
 #endif /* CONFIG_PREEMPT */
 
 ENTRY(enable_exception)
#4 lore raw

PATCH 3/14 csky: entry: Remove unneeded need_resched() loop

In reply to Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

---
 arch/csky/kernel/entry.S | 4 ----
 1 file changed, 4 deletions(-)
arch/csky/kernel/entry.S +0 -4
index 5137ed9062bd..2d9e4776ba7a 100644
@@ -307,11 +307,7 @@ ENTRY(csky_irq)
 	ldw	r8, (r9, TINFO_FLAGS)
 	btsti	r8, TIF_NEED_RESCHED
 	bf	2f
-1:
 	jbsr	preempt_schedule_irq	/* irq en/disable is done inside */
-	ldw	r7, (r9, TINFO_FLAGS)	/* get new tasks TI_FLAGS */
-	btsti	r7, TIF_NEED_RESCHED
-	bt	1b			/* go again */
 #endif
 2:
 	jmpi	ret_from_exception
Signature · 1 line
2.20.1
#5 lore raw

PATCH 4/14 h8300: entry: Remove unneeded need_resched() loop

In reply to Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

---
 arch/h8300/kernel/entry.S | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
arch/h8300/kernel/entry.S +1 -2
index 4ade5f8299ba..6bde028e7d4a 100644
@@ -323,7 +323,6 @@ restore_all:
 resume_kernel:
 	mov.l	@(TI_PRE_COUNT:16,er4),er0
 	bne	restore_all:8
-need_resched:
 	mov.l	@(TI_FLAGS:16,er4),er0
 	btst	#TIF_NEED_RESCHED,r0l
 	beq	restore_all:8
@@ -332,7 +331,7 @@ need_resched:
 	mov.l	sp,er0
 	jsr	@set_esp0
 	jsr	@preempt_schedule_irq
-	bra	need_resched:8
+	bra	restore_all:8
 #endif
 
 ret_from_fork:
Signature · 1 line
2.20.1
#6 lore raw

PATCH 5/14 microblaze: entry: Remove unneeded need_resched() loop

In reply to Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

---
 arch/microblaze/kernel/entry.S | 5 -----
 1 file changed, 5 deletions(-)
arch/microblaze/kernel/entry.S +0 -5
index 4e1b567becd6..de7083bd1d24 100644
@@ -738,14 +738,9 @@ no_intr_resched:
 	andi	r5, r5, _TIF_NEED_RESCHED;
 	beqi	r5, restore /* if zero jump over */
 
-preempt:
 	/* interrupts are off that's why I am calling preempt_chedule_irq */
 	bralid	r15, preempt_schedule_irq
 	nop
-	lwi	r11, CURRENT_TASK, TS_THREAD_INFO;	/* get thread info */
-	lwi	r5, r11, TI_FLAGS;		/* get flags in thread info */
-	andi	r5, r5, _TIF_NEED_RESCHED;
-	bnei	r5, preempt /* if non zero jump to resched */
 restore:
 #endif
 	VM_OFF /* MS: turn off MMU */
Signature · 1 line
2.20.1
#7 lore raw

PATCH 6/14 MIPS: entry: Remove unneeded need_resched() loop

In reply to Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

Do note that commit a18815abcdfd ("Use preempt_schedule_irq.") initially removed the existing loop, but commit cdaed73afb61 ("Fix preemption bug.") reintroduced it. It is however not clear what the issue was and why such a loop was reintroduced, so I'm probably missing something.

---
 arch/mips/kernel/entry.S | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
arch/mips/kernel/entry.S +1 -2
index d7de8adcfcc8..2240faeda62a 100644
@@ -58,7 +58,6 @@ resume_kernel:
 	local_irq_disable
 	lw	t0, TI_PRE_COUNT($28)
 	bnez	t0, restore_all
-need_resched:
 	LONG_L	t0, TI_FLAGS($28)
 	andi	t1, t0, _TIF_NEED_RESCHED
 	beqz	t1, restore_all
@@ -66,7 +65,7 @@ need_resched:
 	andi	t0, 1
 	beqz	t0, restore_all
 	jal	preempt_schedule_irq
-	b	need_resched
+	j	restore_all
 #endif
 
 FEXPORT(ret_from_kernel_thread)
Signature · 1 line
2.20.1
#8 lore raw

Re: [PATCH 06/14] MIPS: entry: Remove unneeded need_resched() loop

Hi Valentin,

Quoted 9 lines · Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

Do note that commit a18815abcdfd ("Use preempt_schedule_irq.") initially removed the existing loop, but commit cdaed73afb61 ("Fix preemption bug.") reintroduced it. It is however not clear what the issue was and why such a loop was reintroduced, so I'm probably missing something.

It looks to me like commit a18815abcdfd ("Use preempt_schedule_irq.") forgot the branch to restore_all, so would have fallen through to ret_from_fork() & done weird things.

Adding the branch to restore_all as you're doing here would have been a better fix than commit cdaed73afb61 ("Fix preemption bug.").

Quoted 22 lines
---
 arch/mips/kernel/entry.S | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
arch/mips/kernel/entry.S +0 -1
index d7de8adcfcc8..2240faeda62a 100644
@@ -58,7 +58,6 @@ resume_kernel:
 	local_irq_disable
 	lw	t0, TI_PRE_COUNT($28)
 	bnez	t0, restore_all
-need_resched:
 	LONG_L	t0, TI_FLAGS($28)
 	andi	t1, t0, _TIF_NEED_RESCHED
 	beqz	t1, restore_all

arch/mips/kernel/entry.S need_resched: · line 68

Quoted 5 lines
arch/mips/kernel/entry.S +1 -1
@@ -66,7 +65,7 @@ need_resched:
 	andi	t0, 1
 	beqz	t0, restore_all
 	jal	preempt_schedule_irq
-	b	need_resched
+	j	restore_all

One nit - why change from branch to jump? It's not a big deal, but I'd prefer we stick with the branch ("b") instruction for a few reasons:

- restore_all is nearby so there's no issue with it being out of range of a branch in any variation of the MIPS ISA.

- It's more consistent with the future of the MIPS architecture, eg. nanoMIPS in which branch instructions all use PC-relative immediate offsets & jump instructions are always of the "register" variety where the destination is specified by a register rather than an immediate encoded in the instruction (the assembler will fix it up & emit a branch anyway, but I generally prefer to invoke less magic in these areas...).

- A PC-relative branch won't generate an extra reloc in a relocatable kernel, whereas a jump will.

Even better would be if we take advantage of this being a tail call & do this:

PTR_LA ra, restore_all
j preempt_schedule_irq

(Where I left the call to preempt_schedule_irq using a jump because it may be further away.)

Though I don't mind if you wanna just s/j/b/ & leave the tail call optimisation for someone else to do as a later change.

Thanks,
Paul

Quoted 6 lines
arch/mips/kernel/entry.S +0 -0
 #endif
 
 FEXPORT(ret_from_kernel_thread)
Signature · 1 line
2.20.1
#9 lore raw

Re: [PATCH 06/14] MIPS: entry: Remove unneeded need_resched() loop

Hi Paul,

On 14/03/2019 18:13, Paul Burton wrote:
[...]

Quoted 8 lines

It looks to me like commit a18815abcdfd ("Use preempt_schedule_irq.") forgot the branch to restore_all, so would have fallen through to ret_from_fork() & done weird things.

Adding the branch to restore_all as you're doing here would have been a better fix than commit cdaed73afb61 ("Fix preemption bug.").

I didn't notice the missing branch to restore_all in that first commit - that makes (more) sense now.

[...]

Quoted 8 lines

arch/mips/kernel/entry.S need_resched: · line 68

Quoted 6 lines
arch/mips/kernel/entry.S +1 -1
@@ -66,7 +65,7 @@ need_resched:
 	andi	t0, 1
 	beqz	t0, restore_all
 	jal	preempt_schedule_irq
-	b	need_resched
+	j	restore_all

One nit - why change from branch to jump?

No actual reason there, I most likely deleted the branch, looked around, saw the "j restore_all" in @resume_userspace and went for that (shoddy I know...)

Quoted 14 lines

It's not a big deal, but I'd
prefer we stick with the branch ("b") instruction for a few reasons:

- restore_all is nearby so there's no issue with it being out of range of a branch in any variation of the MIPS ISA.

- It's more consistent with the future of the MIPS architecture, eg. nanoMIPS in which branch instructions all use PC-relative immediate offsets & jump instructions are always of the "register" variety where the destination is specified by a register rather than an immediate encoded in the instruction (the assembler will fix it up & emit a branch anyway, but I generally prefer to invoke less magic in these areas...).

Quoted 3 lines

- A PC-relative branch won't generate an extra reloc in a relocatable kernel, whereas a jump will.

Makes total sense, thanks for the detailed reasoning!

Quoted 9 lines

Even better would be if we take advantage of this being a tail call & do this:

PTR_LA ra, restore_all
j preempt_schedule_irq

(Where I left the call to preempt_schedule_irq using a jump because it may be further away.)

Right, that's even better, I'll send a v2 with that.

Quoted 12 lines

Though I don't mind if you wanna just s/j/b/ & leave the tail call optimisation for someone else to do as a later change.

Thanks,
Paul

Quoted 6 lines
arch/mips/kernel/entry.S +0 -0
 #endif
 
 FEXPORT(ret_from_kernel_thread)
Signature · 1 line
2.20.1
#10 lore raw

Re: [PATCH 06/14] MIPS: entry: Remove unneeded need_resched() loop

In reply to Paul Burton
Quoted 10 lines · Paul Burton

arch/mips/kernel/entry.S need_resched: · line 68

Quoted 6 lines
arch/mips/kernel/entry.S +1 -1
@@ -66,7 +65,7 @@ need_resched:
 	andi	t0, 1
 	beqz	t0, restore_all
 	jal	preempt_schedule_irq
-	b	need_resched
+	j	restore_all

One nit - why change from branch to jump? It's not a big deal, but I'd prefer we stick with the branch ("b") instruction for a few reasons:

Quoted 2 lines

- restore_all is nearby so there's no issue with it being out of range of a branch in any variation of the MIPS ISA.

FYI, if it does go out of range for whatever reason, then for non-PIC
code GAS will relax it to a jump anyway (with a relocation attached); for the regular MIPS ISA that is, where it has been doing that since forever (I meant to implement this for the microMIPS ISA too, but IIRC there was a complication there, probably coming from the existing more complex branch relaxation code and/or slightly different use of relocations, and then it fell through the cracks).

Maciej

#11 lore raw

PATCH v2 1/1 MIPS: entry: Remove unneeded need_resched() loop

In reply to Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

Note that commit a18815abcdfd ("Use preempt_schedule_irq.") initially removed the existing loop, but missed the final branch to restore_all. Commit cdaed73afb61 ("Fix preemption bug.") missed that and reintroduced the loop.

---
 arch/mips/kernel/entry.S | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
arch/mips/kernel/entry.S +2 -3
index d7de8adcfcc8..5469d43b6966 100644
@@ -58,15 +58,14 @@ resume_kernel:
 	local_irq_disable
 	lw	t0, TI_PRE_COUNT($28)
 	bnez	t0, restore_all
-need_resched:
 	LONG_L	t0, TI_FLAGS($28)
 	andi	t1, t0, _TIF_NEED_RESCHED
 	beqz	t1, restore_all
 	LONG_L	t0, PT_STATUS(sp)		# Interrupts off?
 	andi	t0, 1
 	beqz	t0, restore_all
-	jal	preempt_schedule_irq
-	b	need_resched
+	PTR_LA	ra, restore_all
+	j	preempt_schedule_irq
 #endif
 
 FEXPORT(ret_from_kernel_thread)

--
2.20.1

Applied
#12 lore raw

Re: [PATCH v2] MIPS: entry: Remove unneeded need_resched() loop

Hello,

Quoted 9 lines · Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

Note that commit a18815abcdfd ("Use preempt_schedule_irq.") initially removed the existing loop, but missed the final branch to restore_all. Commit cdaed73afb61 ("Fix preemption bug.") missed that and reintroduced the loop.

Quoted 5 lines

Applied to mips-next.

Thanks,
Paul

[ This message was auto-generated; if you believe anything is incorrect
then please email [email protected] to report it. ]

#13 lore raw

PATCH 7/14 nds32: ex-exit: Remove unneeded need_resched() loop

In reply to Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

Cc Greentime Hu <[email protected]>
Cc Vincent Chen <[email protected]>
---
 arch/nds32/kernel/ex-exit.S | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/nds32/kernel/ex-exit.S +2 -2
index 97ba15cd4180..1df02a793364 100644
@@ -163,7 +163,7 @@ resume_kernel:
 	gie_disable
 	lwi	$t0, [tsk+#TSK_TI_PREEMPT]
 	bnez	$t0, no_work_pending
-need_resched:
+
 	lwi	$t0, [tsk+#TSK_TI_FLAGS]
 	andi	$p1, $t0, #_TIF_NEED_RESCHED
 	beqz	$p1, no_work_pending
@@ -173,7 +173,7 @@ need_resched:
 	beqz	$t0, no_work_pending
 
 	jal	preempt_schedule_irq
-	b	need_resched
+	b	no_work_pending
 #endif
 
 /*
Signature · 1 line
2.20.1
Acked-by on v1 7/14
#14 lore raw

Re: [PATCH 07/14] nds32: ex-exit: Remove unneeded need_resched() loop

Hi Valentin,

Valentin Schneider <[email protected]> 於 2019年3月12日 週二 上午6:48寫道:

Quoted 26 lines

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

Cc Greentime Hu <[email protected]>
Cc Vincent Chen <[email protected]>
---
 arch/nds32/kernel/ex-exit.S | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/nds32/kernel/ex-exit.S +1 -1
index 97ba15cd4180..1df02a793364 100644
@@ -163,7 +163,7 @@ resume_kernel:
        gie_disable
        lwi     $t0, [tsk+#TSK_TI_PREEMPT]
        bnez    $t0, no_work_pending
-need_resched:
+
        lwi     $t0, [tsk+#TSK_TI_FLAGS]
        andi    $p1, $t0, #_TIF_NEED_RESCHED
        beqz    $p1, no_work_pending

arch/nds32/kernel/ex-exit.S need_resched: · line 179

Quoted 8 lines
arch/nds32/kernel/ex-exit.S +1 -1
@@ -173,7 +173,7 @@ need_resched:
        beqz    $t0, no_work_pending
 
        jal     preempt_schedule_irq
-       b       need_resched
+       b       no_work_pending
 #endif
 
 /*

Thank you. :)

#15 lore raw

PATCH 8/14 powerpc: entry: Remove unneeded need_resched() loop

In reply to Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

Cc Benjamin Herrenschmidt <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Michael Ellerman <[email protected]>
---
 arch/powerpc/kernel/entry_32.S | 6 +-----
 arch/powerpc/kernel/entry_64.S | 8 +-------
 2 files changed, 2 insertions(+), 12 deletions(-)
arch/powerpc/kernel/entry_32.S +1 -5
index 0768dfd8a64e..ff3fe3824a4a 100644
@@ -896,11 +896,7 @@ resume_kernel:
 	 */
 	bl	trace_hardirqs_off
 #endif
-1:	bl	preempt_schedule_irq
-	CURRENT_THREAD_INFO(r9, r1)
-	lwz	r3,TI_FLAGS(r9)
-	andi.	r0,r3,_TIF_NEED_RESCHED
-	bne-	1b
+	bl	preempt_schedule_irq
 #ifdef CONFIG_TRACE_IRQFLAGS
 	/* And now, to properly rebalance the above, we tell lockdep they
 	 * are being turned back on, which will happen when we return
arch/powerpc/kernel/entry_64.S +1 -7
index 435927f549c4..9c86c6826856 100644
@@ -857,13 +857,7 @@ resume_kernel:
 	 * sure we are soft-disabled first and reconcile irq state.
 	 */
 	RECONCILE_IRQ_STATE(r3,r4)
-1:	bl	preempt_schedule_irq
-
-	/* Re-test flags and eventually loop */
-	CURRENT_THREAD_INFO(r9, r1)
-	ld	r4,TI_FLAGS(r9)
-	andi.	r0,r4,_TIF_NEED_RESCHED
-	bne	1b
+	bl	preempt_schedule_irq
 
 	/*
 	 * arch_local_irq_restore() from preempt_schedule_irq above may
Signature · 1 line
2.20.1
#16 lore raw

Re: [PATCH 08/14] powerpc: entry: Remove unneeded need_resched() loop

Quoted 9 lines · Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

Cc Benjamin Herrenschmidt <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Michael Ellerman <[email protected]>

Applied to powerpc next, thanks.

https://git.kernel.org/powerpc/c/90437bffa5f9b1440ba03e023f4875d1

cheers

#17 lore raw

PATCH 9/14 RISC-V: entry: Remove unneeded need_resched() loop

In reply to Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

---
 arch/riscv/kernel/entry.S | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
arch/riscv/kernel/entry.S +1 -2
index fd9b57c8b4ce..18e3337e9c30 100644
@@ -258,12 +258,11 @@ restore_all:
 resume_kernel:
 	REG_L s0, TASK_TI_PREEMPT_COUNT(tp)
 	bnez s0, restore_all
-need_resched:
 	REG_L s0, TASK_TI_FLAGS(tp)
 	andi s0, s0, _TIF_NEED_RESCHED
 	beqz s0, restore_all
 	call preempt_schedule_irq
-	j need_resched
+	j restore_all
 #endif
 
 work_pending:
Signature · 1 line
2.20.1
#18 lore raw

PATCH 10/14 sh: entry: Remove unneeded need_resched() loop

In reply to Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

---
 arch/sh/kernel/entry-common.S | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
arch/sh/kernel/entry-common.S +1 -3
index d31f66e82ce5..65a105de52a0 100644
@@ -93,7 +93,7 @@ ENTRY(resume_kernel)
 	mov.l	@(TI_PRE_COUNT,r8), r0	! current_thread_info->preempt_count
 	tst	r0, r0
 	bf	noresched
-need_resched:
+
 	mov.l	@(TI_FLAGS,r8), r0	! current_thread_info->flags
 	tst	#_TIF_NEED_RESCHED, r0	! need_resched set?
 	bt	noresched
@@ -107,8 +107,6 @@ need_resched:
 	mov.l	1f, r0
 	jsr	@r0			! call preempt_schedule_irq
 	 nop
-	bra	need_resched
-	 nop
 
 noresched:
 	bra	__restore_all
Signature · 1 line
2.20.1
#19 lore raw

PATCH 11/14 sh64: entry: Remove unneeded need_resched() loop

In reply to Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

---
 arch/sh/kernel/cpu/sh5/entry.S | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
arch/sh/kernel/cpu/sh5/entry.S +1 -4
index de68ffdfffbf..40e6d9a7a6a2 100644
@@ -897,7 +897,6 @@ resume_kernel:
 	ld.l	r6, TI_PRE_COUNT, r7
 	beq/u	r7, ZERO, tr0
 
-need_resched:
 	ld.l	r6, TI_FLAGS, r7
 	movi	(1 << TIF_NEED_RESCHED), r8
 	and	r8, r7, r8
@@ -911,9 +910,7 @@ need_resched:
 	ori	r7, 1, r7
 	ptabs	r7, tr1
 	blink	tr1, LINK
-
-	pta	need_resched, tr1
-	blink	tr1, ZERO
+	blink   tr0, ZERO
 #endif
 
 	.global ret_from_syscall
Signature · 1 line
2.20.1
#20 lore raw

PATCH 12/14 sparc64: rtrap: Remove unneeded need_resched() loop

In reply to Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

We seem to be looping back somewhere much higher than the usual preempt/need_resched checks, but AFAICT we would just branch to 'rtrap_no_irq_enable' and then back to 'to_kernel', which is a need_resched() loop with a few extra steps.

---
 arch/sparc/kernel/rtrap_64.S | 1 -
 1 file changed, 1 deletion(-)
arch/sparc/kernel/rtrap_64.S +0 -1
index 29aa34f11720..57fb75686957 100644
@@ -322,7 +322,6 @@ to_kernel:
 		 nop
 		call			preempt_schedule_irq
 		 nop
-		ba,pt			%xcc, rtrap
 #endif
 kern_fpucheck:	ldub			[%g6 + TI_FPDEPTH], %l5
 		brz,pt			%l5, rt_continue
Signature · 1 line
2.20.1
maintainer
#21 lore raw

Re: [PATCH 12/14] sparc64: rtrap: Remove unneeded need_resched() loop

From: Valentin Schneider <[email protected]>
Date: Mon, 11 Mar 2019 22:47:50 +0000

Quoted 31 lines

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

We seem to be looping back somewhere much higher than the usual preempt/need_resched checks, but AFAICT we would just branch to 'rtrap_no_irq_enable' and then back to 'to_kernel', which is a need_resched() loop with a few extra steps.

---
 arch/sparc/kernel/rtrap_64.S | 1 -
 1 file changed, 1 deletion(-)
arch/sparc/kernel/rtrap_64.S +0 -1
index 29aa34f11720..57fb75686957 100644
@@ -322,7 +322,6 @@ to_kernel:
 		 nop
 		call			preempt_schedule_irq
 		 nop
-		ba,pt			%xcc, rtrap
 #endif
 kern_fpucheck:	ldub			[%g6 + TI_FPDEPTH], %l5
 		brz,pt			%l5, rt_continue
Signature · 1 line
2.20.1

We must re-evaluate the %tstate value stored in ptregs, you cannot make this change.

#22 lore raw

Re: [PATCH 12/14] sparc64: rtrap: Remove unneeded need_resched() loop

Hi,

On 11/03/2019 23:13, David Miller wrote:
[...]

Quoted 4 lines

We must re-evaluate the %tstate value stored in ptregs, you cannot make this change.

That's the one I was the less sure about, thanks for clearing it up and sorry for the noise.

#23 lore raw

PATCH 13/14 x86/entry: Remove unneeded need_resched() loop

In reply to Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

Cc Andy Lutomirski <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Borislav Petkov <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
---
 arch/x86/entry/entry_32.S | 3 +--
 arch/x86/entry/entry_64.S | 3 +--
 2 files changed, 2 insertions(+), 4 deletions(-)
arch/x86/entry/entry_32.S +1 -2
index d309f30cf7af..b1856fe9decf 100644
@@ -766,13 +766,12 @@ END(ret_from_exception)
 #ifdef CONFIG_PREEMPT
 ENTRY(resume_kernel)
 	DISABLE_INTERRUPTS(CLBR_ANY)
-.Lneed_resched:
 	cmpl	$0, PER_CPU_VAR(__preempt_count)
 	jnz	restore_all_kernel
 	testl	$X86_EFLAGS_IF, PT_EFLAGS(%esp)	# interrupts off (exception path) ?
 	jz	restore_all_kernel
 	call	preempt_schedule_irq
-	jmp	.Lneed_resched
+	jmp	restore_all_kernel
 END(resume_kernel)
 #endif
 
arch/x86/entry/entry_64.S +1 -2
index 1f0efdb7b629..e7e270603fe7 100644
@@ -645,10 +645,9 @@ retint_kernel:
 	/* Check if we need preemption */
 	btl	$9, EFLAGS(%rsp)		/* were interrupts off? */
 	jnc	1f
-0:	cmpl	$0, PER_CPU_VAR(__preempt_count)
+	cmpl	$0, PER_CPU_VAR(__preempt_count)
 	jnz	1f
 	call	preempt_schedule_irq
-	jmp	0b
 1:
 #endif
 	/*
Signature · 1 line
2.20.1
#24 lore raw

[tip:x86/entry] x86/entry: Remove unneeded need_resched() loop

x86/entry: Remove unneeded need_resched() loop
Commit-ID:  b5b447b6b4e899e0978b95cb42d272978f8c7b4d
Gitweb:     https://git.kernel.org/tip/b5b447b6b4e899e0978b95cb42d272978f8c7b4d
Author:     Valentin Schneider <[email protected]>
AuthorDate: Mon, 11 Mar 2019 22:47:51 +0000
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Fri, 5 Apr 2019 15:08:03 +0200

x86/entry: Remove unneeded need_resched() loop

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, there is no need for the outer architecture specific loop.

---
 arch/x86/entry/entry_32.S | 3 +--
 arch/x86/entry/entry_64.S | 3 +--
 2 files changed, 2 insertions(+), 4 deletions(-)
arch/x86/entry/entry_32.S +1 -2
index d309f30cf7af..b1856fe9decf 100644
@@ -766,13 +766,12 @@ END(ret_from_exception)
 #ifdef CONFIG_PREEMPT
 ENTRY(resume_kernel)
 	DISABLE_INTERRUPTS(CLBR_ANY)
-.Lneed_resched:
 	cmpl	$0, PER_CPU_VAR(__preempt_count)
 	jnz	restore_all_kernel
 	testl	$X86_EFLAGS_IF, PT_EFLAGS(%esp)	# interrupts off (exception path) ?
 	jz	restore_all_kernel
 	call	preempt_schedule_irq
-	jmp	.Lneed_resched
+	jmp	restore_all_kernel
 END(resume_kernel)
 #endif
 
arch/x86/entry/entry_64.S +1 -2
index 1f0efdb7b629..e7e270603fe7 100644
@@ -645,10 +645,9 @@ retint_kernel:
 	/* Check if we need preemption */
 	btl	$9, EFLAGS(%rsp)		/* were interrupts off? */
 	jnc	1f
-0:	cmpl	$0, PER_CPU_VAR(__preempt_count)
+	cmpl	$0, PER_CPU_VAR(__preempt_count)
 	jnz	1f
 	call	preempt_schedule_irq
-	jmp	0b
 1:
 #endif
 	/*
#25 lore raw

PATCH 14/14 xtensa: entry: Remove unneeded need_resched() loop

In reply to Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

---
 arch/xtensa/kernel/entry.S | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/xtensa/kernel/entry.S +1 -1
index e50f5124dc6f..43ecaac14ae6 100644
@@ -529,7 +529,7 @@ common_exception_return:
 	l32i	a4, a2, TI_PRE_COUNT
 	bnez	a4, 4f
 	call4	preempt_schedule_irq
-	j	1b
+	j	4f
 #endif
 
 #if XTENSA_FAKE_NMI
Signature · 1 line
2.20.1
maintainer Acked-by on v1 14/14
#26 lore raw

Re: [PATCH 14/14] xtensa: entry: Remove unneeded need_resched() loop

Quoted 5 lines · Valentin Schneider

Since the enabling and disabling of IRQs within preempt_schedule_irq() is contained in a need_resched() loop, we don't need the outer arch code loop.

Quoted 7 lines
---
 arch/xtensa/kernel/entry.S | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
Signature · 2 lines
Thanks.
-- Max
#28 lore raw
In reply to Valentin Schneider
Quoted 14 lines · Valentin Schneider

Hi,

This is the continuation of [1] where I'm hunting down preempt_schedule_irq() callers because of [2].

I told myself the best way to get this moving forward wouldn't be to write doc about it, but to go write some fixes and get some discussions going, which is what this patch-set is about.

I've looked at users of preempt_schedule_irq(), and made sure they didn't have one of those useless loops. The list of offenders is:

$ grep -r -I "preempt_schedule_irq" arch/ | cut -d/ -f2 | sort | uniq

...

Quoted 3 lines

Regarding that loop, archs seem to fall in 3 categories:
A) Those that don't have the loop

Please clarify that this is the right thing to do (since core code already has the loop) hence no fixing is required for this "category"

Quoted 14 lines

B) Those that have a small need_resched() loop around the preempt_schedule_irq() callsite
C) Those that branch to some more generic code further up the entry code and eventually branch back to preempt_schedule_irq()

arc, m68k, nios2 fall in A)

sparc, ia64, s390 fall in C)
all the others fall in B)

I've written patches for B) and C) EXCEPT for ia64 and s390 because I haven't been able to tell if it's actually fine to kill that "long jump" (and maybe I'm wrong on sparc). Hopefully folks who understand what goes on in there might be able to shed some light.

#29 lore raw

On 12/03/2019 18:03, Vineet Gupta wrote:
[...]

Quoted 6 lines
Quoted 2 lines

Regarding that loop, archs seem to fall in 3 categories:
A) Those that don't have the loop

Please clarify that this is the right thing to do (since core code already has the loop) hence no fixing is required for this "category"

Right, those don't need any change. I had a brief look at them to double check they had the proper need_resched() gate before calling preempt_schedule_irq() (with no loop) and they all seem fine. Also...

Quoted 7 lines
Quoted 6 lines

B) Those that have a small need_resched() loop around the preempt_schedule_irq() callsite
C) Those that branch to some more generic code further up the entry code and eventually branch back to preempt_schedule_irq()

arc, m68k, nios2 fall in A)

I forgot to include parisc in here.

[...]