Skip to contentSkip to conversation

clean up ret_from_{irq,exception}

PATCH In review Franck Bui-Huu 3 participants 7 messages Feb 6, 2007 – Feb 12, 2007 MIPS

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 clean up ret_from_{irq,exception}

From: Franck Bui-Huu <[email protected]>

This patch makes these routines a lot more readable whatever the value of CONFIG_PREEMPT.

It also moves one branch instruction from ret_from_irq() to ret_from_exception(). Therefore we favour the return from irq path which should be more common than the other one.

---
 arch/mips/kernel/entry.S |   15 +++++----------
 1 files changed, 5 insertions(+), 10 deletions(-)
arch/mips/kernel/entry.S +5 -10
index f10b6a1..571029b 100644
@@ -21,23 +21,18 @@
 #endif
 
 #ifndef CONFIG_PREEMPT
-	.macro	preempt_stop
-	local_irq_disable
-	.endm
 #define resume_kernel	restore_all
 #endif
 
 	.text
 	.align	5
-FEXPORT(ret_from_irq)
-	LONG_S	s0, TI_REGS($28)
-#ifdef CONFIG_PREEMPT
-FEXPORT(ret_from_exception)
-#else
-	b	_ret_from_irq
 FEXPORT(ret_from_exception)
-	preempt_stop
+#ifndef CONFIG_PREEMPT
+	local_irq_disable			# preempt stop
 #endif
+	b	_ret_from_irq
+FEXPORT(ret_from_irq)
+	LONG_S	s0, TI_REGS($28)
 FEXPORT(_ret_from_irq)
 	LONG_L	t0, PT_STATUS(sp)		# returning to kernel mode?
 	andi	t0, t0, KU_USER
Signature · 1 line
1.4.4.3.ge6d4
#1 lore raw
Quoted 7 lines · Franck Bui-Huu

This patch makes these routines a lot more readable whatever the value of CONFIG_PREEMPT.

It also moves one branch instruction from ret_from_irq() to ret_from_exception(). Therefore we favour the return from irq path which should be more common than the other one.

After this patch, entry.S becomes:

FEXPORT(ret_from_exception)
#ifndef CONFIG_PREEMPT
local_irq_disable # preempt stop

#endif
	b	_ret_from_irq
FEXPORT(ret_from_irq)
	LONG_S	s0, TI_REGS($28)

FEXPORT(_ret_from_irq)

Apparently your patch add an additional branch in critical path in CONFIG_PREEMPT=y case.

Maybe this would be better?

#ifdef CONFIG_PREEMPT
FEXPORT(ret_from_irq)

	LONG_S	s0, TI_REGS($28)
FEXPORT(ret_from_exception)
#else
FEXPORT(ret_from_exception)
	local_irq_disable			# preempt stop
	b	_ret_from_irq
FEXPORT(ret_from_irq)
	LONG_S	s0, TI_REGS($28)

#endif
FEXPORT(_ret_from_irq)

---
Atsushi Nemoto

#2 lore raw

Hi Atsushi,

Quoted 11 lines · Atsushi Nemoto

Maybe this would be better?

#ifdef CONFIG_PREEMPT
FEXPORT(ret_from_irq)

        LONG_S  s0, TI_REGS($28)

FEXPORT(ret_from_exception)
#else
FEXPORT(ret_from_exception)
local_irq_disable # preempt stop
b _ret_from_irq
FEXPORT(ret_from_irq)

Quoted 4 lines
        LONG_S  s0, TI_REGS($28)

#endif
FEXPORT(_ret_from_irq)

well maybe this one would be more readable:

-- >8 --

arch/mips/kernel/entry.S +0 -0
index f10b6a1..b5d27d5 100644
@@ -21,23 +21,20 @@

#endif

#ifndef CONFIG_PREEMPT
- .macro preempt_stop
- local_irq_disable
- .endm
#define resume_kernel restore_all
+#else
+#define _ret_from_irq ret_from_exception
#endif

	.text
	.align	5

-FEXPORT(ret_from_irq)
- LONG_S s0, TI_REGS($28)
-#ifdef CONFIG_PREEMPT
+#ifndef CONFIG_PREEMPT
FEXPORT(ret_from_exception)
-#else
+ local_irq_disable # preempt stop

	b	_ret_from_irq

-FEXPORT(ret_from_exception)
- preempt_stop
#endif
+FEXPORT(ret_from_irq)
+ LONG_S s0, TI_REGS($28)
FEXPORT(_ret_from_irq)

	LONG_L	t0, PT_STATUS(sp)		# returning to kernel mode?
	andi	t0, t0, KU_USER
Signature · 1 line
              Franck
#3 lore raw
Quoted 9 lines · Franck Bui-Huu

well maybe this one would be more readable:

-- >8 --

arch/mips/kernel/entry.S +0 -0
index f10b6a1..b5d27d5 100644

arch/mips/kernel/entry.S line 27

Quoted 10 lines
arch/mips/kernel/entry.S +2 -3
@@ -21,23 +21,20 @@
 #endif
 
 #ifndef CONFIG_PREEMPT
-	.macro	preempt_stop
-	local_irq_disable
-	.endm
 #define resume_kernel	restore_all
+#else
+#define _ret_from_irq	ret_from_exception
 #endif

_ret_from_irq is used by dec/int-handler.S directly, so you should not remove it (though decstation_defconfig disables CONFIG_PREEMPT).

But looking at dec/int-handler.S again, I can not see why it uses _ret_from_irq, and why it manipulates TI_REGS($28) in handle_it ...

It seems dec/int-handler.S has been broken for a while. I'll send a patch to fix it. If it was ACKed, I ACK your patch.

---
Atsushi Nemoto

#4 lore raw
Quoted 4 lines · Atsushi Nemoto

_ret_from_irq is used by dec/int-handler.S directly, so you should not remove it (though decstation_defconfig disables CONFIG_PREEMPT).

woah, you're right. the name '_ret_from_irq' looks so private to me that I didn't think that some other code could use it...

Quoted 6 lines

But looking at dec/int-handler.S again, I can not see why it uses _ret_from_irq, and why it manipulates TI_REGS($28) in handle_it ...

It seems dec/int-handler.S has been broken for a while. I'll send a patch to fix it. If it was ACKed, I ACK your patch.

ok I'm waiting...

Signature · 1 line
              Franck
#5 lore raw
In reply to Atsushi Nemoto
Quoted 8 lines · Atsushi Nemoto

_ret_from_irq is used by dec/int-handler.S directly, so you should not remove it (though decstation_defconfig disables CONFIG_PREEMPT).

But looking at dec/int-handler.S again, I can not see why it uses _ret_from_irq, and why it manipulates TI_REGS($28) in handle_it ...

It seems dec/int-handler.S has been broken for a while. I'll send a patch to fix it. If it was ACKed, I ACK your patch.

It works for me with 2.6.18 -- if it needs an update due to more recent
changes, then I have not got there yet. ;-) I plan to rewrite this bit in C if feasible like other platforms did too.

Maciej

#6 lore raw
Quoted 6 lines · Maciej W. Rozycki
Quoted 2 lines

It seems dec/int-handler.S has been broken for a while. I'll send a patch to fix it. If it was ACKed, I ACK your patch.

It works for me with 2.6.18 -- if it needs an update due to more recent
changes, then I have not got there yet. ;-) I plan to rewrite this bit in C if feasible like other platforms did too.

The breakage happened during 2.6.19 development cycle (large pt_regs argument removal).

---
Atsushi Nemoto