Skip to contentSkip to conversation

lockdep: annotate i386 apm

PATCH In review Peter Zijlstra 3 participants 9 messages Oct 11, 2006 – Oct 12, 2006

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 lockdep: annotate i386 apm

Lockdep doesn't like to enable interrupts when they are enabled already.

BUG: warning at kernel/lockdep.c:1814/trace_hardirqs_on() (Not tainted)
 [<c04051ed>] show_trace_log_lvl+0x58/0x16a
 [<c04057fa>] show_trace+0xd/0x10
 [<c0405913>] dump_stack+0x19/0x1b
 [<c043abfb>] trace_hardirqs_on+0xa2/0x11e
 [<c041463c>] apm_bios_call_simple+0xcd/0xfd
 [<c0415242>] apm+0x92/0x5b1
 [<c0402005>] kernel_thread_helper+0x5/0xb
DWARF2 unwinder stuck at kernel_thread_helper+0x5/0xb
Leftover inexact backtrace:
 [<c04057fa>] show_trace+0xd/0x10
 [<c0405913>] dump_stack+0x19/0x1b
 [<c043abfb>] trace_hardirqs_on+0xa2/0x11e
 [<c041463c>] apm_bios_call_simple+0xcd/0xfd
 [<c0415242>] apm+0x92/0x5b1
 [<c0402005>] kernel_thread_helper+0x5/0xb
---
 arch/i386/kernel/apm.c |   25 ++++++++++++++++++-------
 1 file changed, 18 insertions(+), 7 deletions(-)
arch/i386/kernel/apm.c +18 -7
Index: linux-2.6.18.noarch/arch/i386/kernel/apm.c
===================================================================
@@ -539,11 +539,22 @@ static inline void apm_restore_cpus(cpum
  * Also, we KNOW that for the non error case of apm_bios_call, there
  * is no useful data returned in the low order 8 bits of eax.
  */
-#define APM_DO_CLI	\
-	if (apm_info.allow_ints) \
-		local_irq_enable(); \
-	else \
-		local_irq_disable();
+#define APM_DO_CLI \
+	do { \
+		if (apm_info.allow_ints) { \
+			if (irqs_disabled_flags(flags)) \
+				local_irq_enable(); \
+		} else \
+			local_irq_disable(); \
+	} while (0)
+
+#define APM_DO_STI \
+	do { \
+		if (irqs_disabled_flags(flags)) \
+			local_irq_disable(); \
+		else if (irqs_disabled()) \
+			local_irq_enable(); \
+	} while (0)
 
 #ifdef APM_ZERO_SEGS
 #	define APM_DECL_SEGS \
@@ -600,7 +611,7 @@ static u8 apm_bios_call(u32 func, u32 eb
 	APM_DO_SAVE_SEGS;
 	apm_bios_call_asm(func, ebx_in, ecx_in, eax, ebx, ecx, edx, esi);
 	APM_DO_RESTORE_SEGS;
-	local_irq_restore(flags);
+	APM_DO_STI;
 	gdt[0x40 / 8] = save_desc_40;
 	put_cpu();
 	apm_restore_cpus(cpus);
@@ -644,7 +655,7 @@ static u8 apm_bios_call_simple(u32 func,
 	APM_DO_SAVE_SEGS;
 	error = apm_bios_call_simple_asm(func, ebx_in, ecx_in, eax);
 	APM_DO_RESTORE_SEGS;
-	local_irq_restore(flags);
+	APM_DO_STI;
 	gdt[0x40 / 8] = save_desc_40;
 	put_cpu();
 	apm_restore_cpus(cpus);
#1 lore raw
Quoted 45 lines · On Wed, 11 Oct 2006 15:40:22 +0200 Peter Zijlstra <[email protected]>…

Lockdep doesn't like to enable interrupts when they are enabled already.

...

arch/i386/kernel/apm.c +17 -6
@@ -539,11 +539,22 @@ static inline void apm_restore_cpus(cpum
  * Also, we KNOW that for the non error case of apm_bios_call, there
  * is no useful data returned in the low order 8 bits of eax.
  */
-#define APM_DO_CLI	\
-	if (apm_info.allow_ints) \
-		local_irq_enable(); \
-	else \
-		local_irq_disable();
+#define APM_DO_CLI \
+	do { \
+		if (apm_info.allow_ints) { \
+			if (irqs_disabled_flags(flags)) \
+				local_irq_enable(); \
+		} else \
+			local_irq_disable(); \
+	} while (0)
+
+#define APM_DO_STI \
+	do { \
+		if (irqs_disabled_flags(flags)) \
+			local_irq_disable(); \
+		else if (irqs_disabled()) \
+			local_irq_enable(); \
+	} while (0)
 
 #ifdef APM_ZERO_SEGS
 #	define APM_DECL_SEGS \
@@ -600,7 +611,7 @@ static u8 apm_bios_call(u32 func, u32 eb
 	APM_DO_SAVE_SEGS;
 	apm_bios_call_asm(func, ebx_in, ecx_in, eax, ebx, ecx, edx, esi);
 	APM_DO_RESTORE_SEGS;
-	local_irq_restore(flags);
+	APM_DO_STI;
 	gdt[0x40 / 8] = save_desc_40;
 	put_cpu();
 	apm_restore_cpus(cpus);

arch/i386/kernel/apm.c static u8 apm_bios_call_simple(u32 func, · line 661

Quoted 8 lines
arch/i386/kernel/apm.c +1 -1
@@ -644,7 +655,7 @@ static u8 apm_bios_call_simple(u32 func,
 	APM_DO_SAVE_SEGS;
 	error = apm_bios_call_simple_asm(func, ebx_in, ecx_in, eax);
 	APM_DO_RESTORE_SEGS;
-	local_irq_restore(flags);
+	APM_DO_STI;
 	gdt[0x40 / 8] = save_desc_40;
 	put_cpu();
 	apm_restore_cpus(cpus);

ick.

Does this work?

arch/i386/kernel/apm.c renamed from arch/i386/kernel/apm.c~lockdep-annotate-i386-apm +6 -10
@@ -540,12 +540,6 @@ static inline void apm_restore_cpus(cpum
  * Also, we KNOW that for the non error case of apm_bios_call, there
  * is no useful data returned in the low order 8 bits of eax.
  */
-#define APM_DO_CLI	\
-	if (apm_info.allow_ints) \
-		local_irq_enable(); \
-	else \
-		local_irq_disable();
-
 #ifdef APM_ZERO_SEGS
 #	define APM_DECL_SEGS \
 		unsigned int saved_fs; unsigned int saved_gs;
@@ -596,8 +590,9 @@ static u8 apm_bios_call(u32 func, u32 eb
 	save_desc_40 = gdt[0x40 / 8];
 	gdt[0x40 / 8] = bad_bios_desc;
 
-	local_save_flags(flags);
-	APM_DO_CLI;
+	local_irq_save(flags);
+	if (apm_info.allow_ints)
+		local_irq_enable();
 	APM_DO_SAVE_SEGS;
 	apm_bios_call_asm(func, ebx_in, ecx_in, eax, ebx, ecx, edx, esi);
 	APM_DO_RESTORE_SEGS;
@@ -640,8 +635,9 @@ static u8 apm_bios_call_simple(u32 func,
 	save_desc_40 = gdt[0x40 / 8];
 	gdt[0x40 / 8] = bad_bios_desc;
 
-	local_save_flags(flags);
-	APM_DO_CLI;
+	local_irq_save(flags);
+	if (apm_info.allow_ints)
+		local_irq_enable();
 	APM_DO_SAVE_SEGS;
 	error = apm_bios_call_simple_asm(func, ebx_in, ecx_in, eax);
 	APM_DO_RESTORE_SEGS;

_

#2 lore raw
Quoted 100 lines · Andrew Morton
Quoted 45 lines · On Wed, 11 Oct 2006 15:40:22 +0200 Peter Zijlstra <[email protected]>…

Lockdep doesn't like to enable interrupts when they are enabled already.

...

arch/i386/kernel/apm.c +17 -6
@@ -539,11 +539,22 @@ static inline void apm_restore_cpus(cpum
  * Also, we KNOW that for the non error case of apm_bios_call, there
  * is no useful data returned in the low order 8 bits of eax.
  */
-#define APM_DO_CLI	\
-	if (apm_info.allow_ints) \
-		local_irq_enable(); \
-	else \
-		local_irq_disable();
+#define APM_DO_CLI \
+	do { \
+		if (apm_info.allow_ints) { \
+			if (irqs_disabled_flags(flags)) \
+				local_irq_enable(); \
+		} else \
+			local_irq_disable(); \
+	} while (0)
+
+#define APM_DO_STI \
+	do { \
+		if (irqs_disabled_flags(flags)) \
+			local_irq_disable(); \
+		else if (irqs_disabled()) \
+			local_irq_enable(); \
+	} while (0)
 
 #ifdef APM_ZERO_SEGS
 #	define APM_DECL_SEGS \
@@ -600,7 +611,7 @@ static u8 apm_bios_call(u32 func, u32 eb
 	APM_DO_SAVE_SEGS;
 	apm_bios_call_asm(func, ebx_in, ecx_in, eax, ebx, ecx, edx, esi);
 	APM_DO_RESTORE_SEGS;
-	local_irq_restore(flags);
+	APM_DO_STI;
 	gdt[0x40 / 8] = save_desc_40;
 	put_cpu();
 	apm_restore_cpus(cpus);

arch/i386/kernel/apm.c static u8 apm_bios_call_simple(u32 func, · line 661

Quoted 8 lines
arch/i386/kernel/apm.c +1 -1
@@ -644,7 +655,7 @@ static u8 apm_bios_call_simple(u32 func,
 	APM_DO_SAVE_SEGS;
 	error = apm_bios_call_simple_asm(func, ebx_in, ecx_in, eax);
 	APM_DO_RESTORE_SEGS;
-	local_irq_restore(flags);
+	APM_DO_STI;
 	gdt[0x40 / 8] = save_desc_40;
 	put_cpu();
 	apm_restore_cpus(cpus);

ick.

Does this work?

arch/i386/kernel/apm.c renamed from arch/i386/kernel/apm.c~lockdep-annotate-i386-apm +6 -10
@@ -540,12 +540,6 @@ static inline void apm_restore_cpus(cpum
  * Also, we KNOW that for the non error case of apm_bios_call, there
  * is no useful data returned in the low order 8 bits of eax.
  */
-#define APM_DO_CLI	\
-	if (apm_info.allow_ints) \
-		local_irq_enable(); \
-	else \
-		local_irq_disable();
-
 #ifdef APM_ZERO_SEGS
 #	define APM_DECL_SEGS \
 		unsigned int saved_fs; unsigned int saved_gs;
@@ -596,8 +590,9 @@ static u8 apm_bios_call(u32 func, u32 eb
 	save_desc_40 = gdt[0x40 / 8];
 	gdt[0x40 / 8] = bad_bios_desc;
 
-	local_save_flags(flags);
-	APM_DO_CLI;
+	local_irq_save(flags);
+	if (apm_info.allow_ints)
+		local_irq_enable();
 	APM_DO_SAVE_SEGS;
 	apm_bios_call_asm(func, ebx_in, ecx_in, eax, ebx, ecx, edx, esi);
 	APM_DO_RESTORE_SEGS;
@@ -640,8 +635,9 @@ static u8 apm_bios_call_simple(u32 func,
 	save_desc_40 = gdt[0x40 / 8];
 	gdt[0x40 / 8] = bad_bios_desc;
 
-	local_save_flags(flags);
-	APM_DO_CLI;
+	local_irq_save(flags);
+	if (apm_info.allow_ints)
+		local_irq_enable();
 	APM_DO_SAVE_SEGS;
 	error = apm_bios_call_simple_asm(func, ebx_in, ecx_in, eax);
 	APM_DO_RESTORE_SEGS;
Quoted 1 line

_

#define local_irq_restore(flags)                                \
        do {                                                    \
                if (raw_irqs_disabled_flags(flags)) {           \
                        raw_local_irq_restore(flags);           \
                        trace_hardirqs_off();                   \
                } else {                                        \
                        trace_hardirqs_on();                    \
                        raw_local_irq_restore(flags);           \
                }                                               \
        } while (0)

So, say interrupts were enabled when entering apm_bios_call*(); you now save that in flags, disable interrupts, and enable them again. Upon reaching local_irq_restore(), we'll hit the else branch with irq's enabled and call trace_hardirqs_on(), which goes EEEK!

#3 lore raw
Quoted 11 lines · On Thu, 12 Oct 2006 08:06:20 +0200 Peter Zijlstra <[email protected]>…
#define local_irq_restore(flags)                                \
        do {                                                    \
                if (raw_irqs_disabled_flags(flags)) {           \
                        raw_local_irq_restore(flags);           \
                        trace_hardirqs_off();                   \
                } else {                                        \
                        trace_hardirqs_on();                    \
                        raw_local_irq_restore(flags);           \
                }                                               \
        } while (0)
Quoted 4 lines

So, say interrupts were enabled when entering apm_bios_call*(); you now save that in flags, disable interrupts, and enable them again. Upon reaching local_irq_restore(), we'll hit the else branch with irq's enabled and call trace_hardirqs_on(), which goes EEEK!

I'd assumed lockdep was less stupid than that ;) This? Seems a bit overdone..

arch/i386/kernel/apm.c renamed from arch/i386/kernel/apm.c~lockdep-annotate-i386-apm +37 -14
@@ -540,12 +540,6 @@ static inline void apm_restore_cpus(cpum
  * Also, we KNOW that for the non error case of apm_bios_call, there
  * is no useful data returned in the low order 8 bits of eax.
  */
-#define APM_DO_CLI	\
-	if (apm_info.allow_ints) \
-		local_irq_enable(); \
-	else \
-		local_irq_disable();
-
 #ifdef APM_ZERO_SEGS
 #	define APM_DECL_SEGS \
 		unsigned int saved_fs; unsigned int saved_gs;
@@ -583,11 +577,12 @@ static u8 apm_bios_call(u32 func, u32 eb
 	u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, u32 *esi)
 {
 	APM_DECL_SEGS
-	unsigned long		flags;
 	cpumask_t		cpus;
 	int			cpu;
 	struct desc_struct	save_desc_40;
 	struct desc_struct	*gdt;
+	int			enable_irqs = 0;
+	int			disable_irqs = 0;
 
 	cpus = apm_save_cpus();
 	
@@ -596,12 +591,26 @@ static u8 apm_bios_call(u32 func, u32 eb
 	save_desc_40 = gdt[0x40 / 8];
 	gdt[0x40 / 8] = bad_bios_desc;
 
-	local_save_flags(flags);
-	APM_DO_CLI;
+	if (apm_info.allow_ints) {
+		if (irqs_disabled()) {
+			local_irq_enable();
+			disable_irqs = 1;
+		}
+	} else {
+		if (!irqs_disabled()) {
+			local_irq_disable();
+			enable_irqs = 1;
+		}
+	}
+
+
 	APM_DO_SAVE_SEGS;
 	apm_bios_call_asm(func, ebx_in, ecx_in, eax, ebx, ecx, edx, esi);
 	APM_DO_RESTORE_SEGS;
-	local_irq_restore(flags);
+	if (disable_irqs)
+		local_irq_disable();
+	if (enable_irqs)
+		local_irq_enable();
 	gdt[0x40 / 8] = save_desc_40;
 	put_cpu();
 	apm_restore_cpus(cpus);
@@ -627,11 +636,12 @@ static u8 apm_bios_call_simple(u32 func,
 {
 	u8			error;
 	APM_DECL_SEGS
-	unsigned long		flags;
 	cpumask_t		cpus;
 	int			cpu;
 	struct desc_struct	save_desc_40;
 	struct desc_struct	*gdt;
+	int			enable_irqs = 0;
+	int			disable_irqs = 0;
 
 	cpus = apm_save_cpus();
 	
@@ -640,12 +650,25 @@ static u8 apm_bios_call_simple(u32 func,
 	save_desc_40 = gdt[0x40 / 8];
 	gdt[0x40 / 8] = bad_bios_desc;
 
-	local_save_flags(flags);
-	APM_DO_CLI;
+	if (apm_info.allow_ints) {
+		if (irqs_disabled()) {
+			local_irq_enable();
+			disable_irqs = 1;
+		}
+	} else {
+		if (!irqs_disabled()) {
+			local_irq_disable();
+			enable_irqs = 1;
+		}
+	}
+
 	APM_DO_SAVE_SEGS;
 	error = apm_bios_call_simple_asm(func, ebx_in, ecx_in, eax);
 	APM_DO_RESTORE_SEGS;
-	local_irq_restore(flags);
+	if (disable_irqs)
+		local_irq_disable();
+	if (enable_irqs)
+		local_irq_enable();
 	gdt[0x40 / 8] = save_desc_40;
 	put_cpu();
 	apm_restore_cpus(cpus);

_

#4 lore raw
Quoted 14 lines · Andrew Morton
Quoted 11 lines · On Thu, 12 Oct 2006 08:06:20 +0200 Peter Zijlstra <[email protected]>…
#define local_irq_restore(flags)                                \
        do {                                                    \
                if (raw_irqs_disabled_flags(flags)) {           \
                        raw_local_irq_restore(flags);           \
                        trace_hardirqs_off();                   \
                } else {                                        \
                        trace_hardirqs_on();                    \
                        raw_local_irq_restore(flags);           \
                }                                               \
        } while (0)
Quoted 7 lines
Quoted 4 lines

So, say interrupts were enabled when entering apm_bios_call*(); you now save that in flags, disable interrupts, and enable them again. Upon reaching local_irq_restore(), we'll hit the else branch with irq's enabled and call trace_hardirqs_on(), which goes EEEK!

I'd assumed lockdep was less stupid than that ;) This? Seems a bit overdone..

Seems simple enough, bit on the verbose side, _however_ it depends on the BIOS call not changing the irq state. I don't know about BIOSs but I thought the general consensus was to distrust them.

Was my original _that_ hard to read?

Quoted 104 lines
arch/i386/kernel/apm.c renamed from arch/i386/kernel/apm.c~lockdep-annotate-i386-apm +37 -14
@@ -540,12 +540,6 @@ static inline void apm_restore_cpus(cpum
  * Also, we KNOW that for the non error case of apm_bios_call, there
  * is no useful data returned in the low order 8 bits of eax.
  */
-#define APM_DO_CLI	\
-	if (apm_info.allow_ints) \
-		local_irq_enable(); \
-	else \
-		local_irq_disable();
-
 #ifdef APM_ZERO_SEGS
 #	define APM_DECL_SEGS \
 		unsigned int saved_fs; unsigned int saved_gs;
@@ -583,11 +577,12 @@ static u8 apm_bios_call(u32 func, u32 eb
 	u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, u32 *esi)
 {
 	APM_DECL_SEGS
-	unsigned long		flags;
 	cpumask_t		cpus;
 	int			cpu;
 	struct desc_struct	save_desc_40;
 	struct desc_struct	*gdt;
+	int			enable_irqs = 0;
+	int			disable_irqs = 0;
 
 	cpus = apm_save_cpus();
 
@@ -596,12 +591,26 @@ static u8 apm_bios_call(u32 func, u32 eb
 	save_desc_40 = gdt[0x40 / 8];
 	gdt[0x40 / 8] = bad_bios_desc;
 
-	local_save_flags(flags);
-	APM_DO_CLI;
+	if (apm_info.allow_ints) {
+		if (irqs_disabled()) {
+			local_irq_enable();
+			disable_irqs = 1;
+		}
+	} else {
+		if (!irqs_disabled()) {
+			local_irq_disable();
+			enable_irqs = 1;
+		}
+	}
+
+
 	APM_DO_SAVE_SEGS;
 	apm_bios_call_asm(func, ebx_in, ecx_in, eax, ebx, ecx, edx, esi);
 	APM_DO_RESTORE_SEGS;
-	local_irq_restore(flags);
+	if (disable_irqs)
+		local_irq_disable();
+	if (enable_irqs)
+		local_irq_enable();
 	gdt[0x40 / 8] = save_desc_40;
 	put_cpu();
 	apm_restore_cpus(cpus);
@@ -627,11 +636,12 @@ static u8 apm_bios_call_simple(u32 func,
 {
 	u8			error;
 	APM_DECL_SEGS
-	unsigned long		flags;
 	cpumask_t		cpus;
 	int			cpu;
 	struct desc_struct	save_desc_40;
 	struct desc_struct	*gdt;
+	int			enable_irqs = 0;
+	int			disable_irqs = 0;
 
 	cpus = apm_save_cpus();
 
@@ -640,12 +650,25 @@ static u8 apm_bios_call_simple(u32 func,
 	save_desc_40 = gdt[0x40 / 8];
 	gdt[0x40 / 8] = bad_bios_desc;
 
-	local_save_flags(flags);
-	APM_DO_CLI;
+	if (apm_info.allow_ints) {
+		if (irqs_disabled()) {
+			local_irq_enable();
+			disable_irqs = 1;
+		}
+	} else {
+		if (!irqs_disabled()) {
+			local_irq_disable();
+			enable_irqs = 1;
+		}
+	}
+
 	APM_DO_SAVE_SEGS;
 	error = apm_bios_call_simple_asm(func, ebx_in, ecx_in, eax);
 	APM_DO_RESTORE_SEGS;
-	local_irq_restore(flags);
+	if (disable_irqs)
+		local_irq_disable();
+	if (enable_irqs)
+		local_irq_enable();
 	gdt[0x40 / 8] = save_desc_40;
 	put_cpu();
 	apm_restore_cpus(cpus);

_

#5 lore raw
Quoted 1 line · On Thu, 12 Oct 2006 08:50:50 +0200 Peter Zijlstra <[email protected]>…

Was my original _that_ hard to read?

The macro was bad enough. Then you want an added another one and made it reference a caller's local variable. Another dummy-shaped dent in my wall.

I'd appreciate a fixed-up patch which changes the code's niceness in a positive direction please - I'm obviously not cut out to work on apm.c.

#7 lore raw
In reply to Andrew Morton
Quoted 7 lines · Andrew Morton
Quoted 4 lines

So, say interrupts were enabled when entering apm_bios_call*(); you now save that in flags, disable interrupts, and enable them again. Upon reaching local_irq_restore(), we'll hit the else branch with irq's enabled and call trace_hardirqs_on(), which goes EEEK!

I'd assumed lockdep was less stupid than that ;) This? Seems a bit overdone..

the problem is not lockdep but that the BIOS enables IRQs behind the back of the kernel. Lockdep needs to be taught about that - if this happens unconditionally then i'd suggest to insert an unconditional trace_hardirqs_on() call to after the local_irq_save() that we do prior calling the BIOS. (that will be a NOP if lockdep is not enabled)

Ingo

#6 lore raw

PATCH 1/1 lockdep: annotate i386 apm

In reply to Andrew Morton

Ok, so how about this:

---
Lockdep doesn't like to enable interrupts when they are enabled already.

BUG: warning at kernel/lockdep.c:1814/trace_hardirqs_on() (Not tainted)
 [<c04051ed>] show_trace_log_lvl+0x58/0x16a
 [<c04057fa>] show_trace+0xd/0x10
 [<c0405913>] dump_stack+0x19/0x1b
 [<c043abfb>] trace_hardirqs_on+0xa2/0x11e
 [<c041463c>] apm_bios_call_simple+0xcd/0xfd
 [<c0415242>] apm+0x92/0x5b1
 [<c0402005>] kernel_thread_helper+0x5/0xb
DWARF2 unwinder stuck at kernel_thread_helper+0x5/0xb
Leftover inexact backtrace:
 [<c04057fa>] show_trace+0xd/0x10
 [<c0405913>] dump_stack+0x19/0x1b
 [<c043abfb>] trace_hardirqs_on+0xa2/0x11e
 [<c041463c>] apm_bios_call_simple+0xcd/0xfd
 [<c0415242>] apm+0x92/0x5b1
 [<c0402005>] kernel_thread_helper+0x5/0xb
---
 arch/i386/kernel/apm.c |   37 +++++++++++++++++++++++++++----------
 1 file changed, 27 insertions(+), 10 deletions(-)
arch/i386/kernel/apm.c +27 -10
Index: linux-2.6.18.noarch/arch/i386/kernel/apm.c
===================================================================
@@ -539,11 +539,30 @@ static inline void apm_restore_cpus(cpum
  * Also, we KNOW that for the non error case of apm_bios_call, there
  * is no useful data returned in the low order 8 bits of eax.
  */
-#define APM_DO_CLI	\
-	if (apm_info.allow_ints) \
-		local_irq_enable(); \
-	else \
+
+static inline unsigned long __apm_irq_save(void)
+{
+	unsigned long flags;
+	local_save_flags(flags);
+	if (apm_info.allow_ints) {
+		if (irqs_disabled_flags(flags))
+			local_irq_enable();
+	} else
+		local_irq_disable();
+
+	return flags;
+}
+
+#define apm_irq_save(flags) \
+	do { flags = __apm_irq_save(); } while (0)
+
+static inline void apm_irq_restore(unsigned long flags)
+{
+	if (irqs_disabled_flags(flags))
 		local_irq_disable();
+	else if (irqs_disabled())
+		local_irq_enable();
+}
 
 #ifdef APM_ZERO_SEGS
 #	define APM_DECL_SEGS \
@@ -595,12 +614,11 @@ static u8 apm_bios_call(u32 func, u32 eb
 	save_desc_40 = gdt[0x40 / 8];
 	gdt[0x40 / 8] = bad_bios_desc;
 
-	local_save_flags(flags);
-	APM_DO_CLI;
+	apm_irq_save(flags);
 	APM_DO_SAVE_SEGS;
 	apm_bios_call_asm(func, ebx_in, ecx_in, eax, ebx, ecx, edx, esi);
 	APM_DO_RESTORE_SEGS;
-	local_irq_restore(flags);
+	apm_irq_restore(flags);
 	gdt[0x40 / 8] = save_desc_40;
 	put_cpu();
 	apm_restore_cpus(cpus);
@@ -639,12 +657,11 @@ static u8 apm_bios_call_simple(u32 func,
 	save_desc_40 = gdt[0x40 / 8];
 	gdt[0x40 / 8] = bad_bios_desc;
 
-	local_save_flags(flags);
-	APM_DO_CLI;
+	apm_irq_save(flags);
 	APM_DO_SAVE_SEGS;
 	error = apm_bios_call_simple_asm(func, ebx_in, ecx_in, eax);
 	APM_DO_RESTORE_SEGS;
-	local_irq_restore(flags);
+	apm_irq_restore(flags);
 	gdt[0x40 / 8] = save_desc_40;
 	put_cpu();
 	apm_restore_cpus(cpus);
#8 lore raw
In reply to Ingo Molnar
Quoted 7 lines · Ingo Molnar
Quoted 5 lines · Andrew Morton
Quoted 4 lines

So, say interrupts were enabled when entering apm_bios_call*(); you now save that in flags, disable interrupts, and enable them again. Upon reaching local_irq_restore(), we'll hit the else branch with irq's enabled and call trace_hardirqs_on(), which goes EEEK!

Quoted 8 lines
Quoted 2 lines

I'd assumed lockdep was less stupid than that ;) This? Seems a bit overdone..

the problem is not lockdep but that the BIOS enables IRQs behind the back of the kernel. Lockdep needs to be taught about that - if this happens unconditionally then i'd suggest to insert an unconditional trace_hardirqs_on() call to after the local_irq_save() that we do prior calling the BIOS. (that will be a NOP if lockdep is not enabled)

Well, its not quite like that, the irq mess in apm does an explicit unbalanced irq action, and lockdep rightly triggers on that. And adding to that comes the fact that we do indeed call a BIOS routine which can do god knows what.

So we want to save irq state and force it into a known state; normally it will only be disable - however in this case (when apm_info.allow_ints) we force enable it. All end scope routines expect irqs disabled.

logically it looks something like this

local_irq_save(flags)
local_irq_enable()
...
local_irq_disable() <--- missing
local_irq_restore(flags)

except that the first two statements are blended together to avoid the unneeded disable.

The logic is sound because local_irq_restore will just set whatever was saved, except lockdep gets confused by the unbalanced operations - a good thing IMHO.