Skip to contentSkip to conversation

x86: use new exception_enter()/exception_exit() API in do_device_not_available()

RFC PATCH In review Sergey Senozhatsky 2 participants 5 messages Aug 28, 2012 – Aug 29, 2012 x86

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 x86: use new exception_enter()/exception_exit() API in do_device_not_available()

x86: use new exception_enter()/exception_exit() API in do_device_not_available()

RCU can be in extended quiescent state when do_device_not_available() called.

[  676.400274] RCU used illegally from idle CPU!
[  676.400274] rcu_scheduler_active = 1, debug_locks = 0
[  676.400278] RCU used illegally from extended quiescent state!
[  676.400282] 2 locks held by sed/9468:
[  676.400285]  #0:  (&rq->lock){-.-.-.}, at: [<ffffffff814c67cf>] __schedule+0x1c0/0x992
[  676.400304]  #1:  (rcu_read_lock){.+.+..}, at: [<ffffffff8106a880>] cpuacct_charge+0x27/0x1e0
[  676.400321] 
[  676.400329] Call Trace:
[  676.400340]  [<ffffffff8108933a>] lockdep_rcu_suspicious+0x109/0x112
[  676.400347]  [<ffffffff8106a8f7>] cpuacct_charge+0x9e/0x1e0
[  676.400353]  [<ffffffff8106a880>] ? cpuacct_charge+0x27/0x1e0
[  676.400359]  [<ffffffff814c67cf>] ? __schedule+0x1c0/0x992
[  676.400366]  [<ffffffff8106d525>] update_curr+0x159/0x200
[  676.400375]  [<ffffffff8100a04f>] ? init_fpu+0x4e/0x8f
[  676.400382]  [<ffffffff8106e0da>] put_prev_task_fair+0x2a/0xac
[  676.400388]  [<ffffffff814c69a7>] __schedule+0x398/0x992
[  676.400397]  [<ffffffff810b4f0a>] ? rcu_irq_exit+0x83/0x99
[  676.400404]  [<ffffffff814c8bf0>] ? retint_restore_args+0x13/0x13
[  676.400411]  [<ffffffff810889e9>] ? lock_is_held+0x2d/0xa5
[  676.400414]  [<ffffffff8100a04f>] ? init_fpu+0x4e/0x8f
[  676.400417]  [<ffffffff81065d0a>] __cond_resched+0x28/0x34
[  676.400420]  [<ffffffff814c701a>] _cond_resched+0x2e/0x35
[  676.400425]  [<ffffffff8111cad4>] kmem_cache_alloc+0x4c/0x1c5
[  676.400430]  [<ffffffff810021ff>] ? math_state_restore+0x1f/0xdc
[  676.400433]  [<ffffffff8100a04f>] init_fpu+0x4e/0x8f
[  676.400437]  [<ffffffff8100220e>] math_state_restore+0x2e/0xdc
[  676.400440]  [<ffffffff814c982e>] do_device_not_available+0xe/0x10
[  676.400445]  [<ffffffff814d161b>] device_not_available+0x1b/0x20

---

 arch/x86/kernel/traps.c | 2 ++
 1 file changed, 2 insertions(+)
arch/x86/kernel/traps.c +2 -0
index ab82cbd..dff3e72 100644
@@ -654,7 +654,9 @@ do_device_not_available(struct pt_regs *regs, long error_code)
 		return;
 	}
 #endif
+	exception_enter(regs);
 	math_state_restore(); /* interrupts still off */
+	exception_enter(regs);
 #ifdef CONFIG_X86_32
 	conditional_sti(regs);
 #endif
maintainer
#1 lore raw
Quoted 46 lines · Sergey Senozhatsky

x86: use new exception_enter()/exception_exit() API in do_device_not_available()

RCU can be in extended quiescent state when do_device_not_available() called.

[  676.400274] RCU used illegally from idle CPU!
[  676.400274] rcu_scheduler_active = 1, debug_locks = 0
[  676.400278] RCU used illegally from extended quiescent state!
[  676.400282] 2 locks held by sed/9468:
[  676.400285]  #0:  (&rq->lock){-.-.-.}, at: [<ffffffff814c67cf>] __schedule+0x1c0/0x992
[  676.400304]  #1:  (rcu_read_lock){.+.+..}, at: [<ffffffff8106a880>] cpuacct_charge+0x27/0x1e0
[  676.400321] 
[  676.400329] Call Trace:
[  676.400340]  [<ffffffff8108933a>] lockdep_rcu_suspicious+0x109/0x112
[  676.400347]  [<ffffffff8106a8f7>] cpuacct_charge+0x9e/0x1e0
[  676.400353]  [<ffffffff8106a880>] ? cpuacct_charge+0x27/0x1e0
[  676.400359]  [<ffffffff814c67cf>] ? __schedule+0x1c0/0x992
[  676.400366]  [<ffffffff8106d525>] update_curr+0x159/0x200
[  676.400375]  [<ffffffff8100a04f>] ? init_fpu+0x4e/0x8f
[  676.400382]  [<ffffffff8106e0da>] put_prev_task_fair+0x2a/0xac
[  676.400388]  [<ffffffff814c69a7>] __schedule+0x398/0x992
[  676.400397]  [<ffffffff810b4f0a>] ? rcu_irq_exit+0x83/0x99
[  676.400404]  [<ffffffff814c8bf0>] ? retint_restore_args+0x13/0x13
[  676.400411]  [<ffffffff810889e9>] ? lock_is_held+0x2d/0xa5
[  676.400414]  [<ffffffff8100a04f>] ? init_fpu+0x4e/0x8f
[  676.400417]  [<ffffffff81065d0a>] __cond_resched+0x28/0x34
[  676.400420]  [<ffffffff814c701a>] _cond_resched+0x2e/0x35
[  676.400425]  [<ffffffff8111cad4>] kmem_cache_alloc+0x4c/0x1c5
[  676.400430]  [<ffffffff810021ff>] ? math_state_restore+0x1f/0xdc
[  676.400433]  [<ffffffff8100a04f>] init_fpu+0x4e/0x8f
[  676.400437]  [<ffffffff8100220e>] math_state_restore+0x2e/0xdc
[  676.400440]  [<ffffffff814c982e>] do_device_not_available+0xe/0x10
[  676.400445]  [<ffffffff814d161b>] device_not_available+0x1b/0x20

---

 arch/x86/kernel/traps.c | 2 ++
 1 file changed, 2 insertions(+)
arch/x86/kernel/traps.c +0 -0
index ab82cbd..dff3e72 100644

arch/x86/kernel/traps.c do_device_not_available(struct pt_regs *regs, long error_code) · line 659

Quoted 6 lines
arch/x86/kernel/traps.c +2 -0
@@ -654,7 +654,9 @@ do_device_not_available(struct pt_regs *regs, long error_code)
 		return;
 	}
 #endif
+	exception_enter(regs);
 	math_state_restore(); /* interrupts still off */
+	exception_enter(regs);

Shouldn't that be exception_exit actually?

Signature · 2 lines
Regards/Gruss,
    Boris.
#2 lore raw
Quoted 54 lines · Borislav Petkov

Date: Wed, 29 Aug 2012 00:11:13 +0200
From: Borislav Petkov <[email protected]>
To: Sergey Senozhatsky <[email protected]>
Cc: Thomas Gleixner <[email protected]>, Ingo Molnar <[email protected]>,
"H. Peter Anvin" <[email protected]>, [email protected], Steven Rostedt <[email protected]>, Srikar Dronamraju <[email protected]>, Peter Zijlstra <[email protected]>, "Paul E. McKenney" <[email protected]>, [email protected]
Subject: Re: [RFC PATCH] x86: use new exception_enter()/exception_exit()
API in do_device_not_available()
User-Agent: Mutt/1.5.21 (2010-09-15)

Quoted 41 lines · Sergey Senozhatsky

x86: use new exception_enter()/exception_exit() API in do_device_not_available()

RCU can be in extended quiescent state when do_device_not_available() called.

[  676.400274] RCU used illegally from idle CPU!
[  676.400274] rcu_scheduler_active = 1, debug_locks = 0
[  676.400278] RCU used illegally from extended quiescent state!
[  676.400282] 2 locks held by sed/9468:
[  676.400285]  #0:  (&rq->lock){-.-.-.}, at: [<ffffffff814c67cf>] __schedule+0x1c0/0x992
[  676.400304]  #1:  (rcu_read_lock){.+.+..}, at: [<ffffffff8106a880>] cpuacct_charge+0x27/0x1e0
[  676.400321] 
[  676.400329] Call Trace:
[  676.400340]  [<ffffffff8108933a>] lockdep_rcu_suspicious+0x109/0x112
[  676.400347]  [<ffffffff8106a8f7>] cpuacct_charge+0x9e/0x1e0
[  676.400353]  [<ffffffff8106a880>] ? cpuacct_charge+0x27/0x1e0
[  676.400359]  [<ffffffff814c67cf>] ? __schedule+0x1c0/0x992
[  676.400366]  [<ffffffff8106d525>] update_curr+0x159/0x200
[  676.400375]  [<ffffffff8100a04f>] ? init_fpu+0x4e/0x8f
[  676.400382]  [<ffffffff8106e0da>] put_prev_task_fair+0x2a/0xac
[  676.400388]  [<ffffffff814c69a7>] __schedule+0x398/0x992
[  676.400397]  [<ffffffff810b4f0a>] ? rcu_irq_exit+0x83/0x99
[  676.400404]  [<ffffffff814c8bf0>] ? retint_restore_args+0x13/0x13
[  676.400411]  [<ffffffff810889e9>] ? lock_is_held+0x2d/0xa5
[  676.400414]  [<ffffffff8100a04f>] ? init_fpu+0x4e/0x8f
[  676.400417]  [<ffffffff81065d0a>] __cond_resched+0x28/0x34
[  676.400420]  [<ffffffff814c701a>] _cond_resched+0x2e/0x35
[  676.400425]  [<ffffffff8111cad4>] kmem_cache_alloc+0x4c/0x1c5
[  676.400430]  [<ffffffff810021ff>] ? math_state_restore+0x1f/0xdc
[  676.400433]  [<ffffffff8100a04f>] init_fpu+0x4e/0x8f
[  676.400437]  [<ffffffff8100220e>] math_state_restore+0x2e/0xdc
[  676.400440]  [<ffffffff814c982e>] do_device_not_available+0xe/0x10
[  676.400445]  [<ffffffff814d161b>] device_not_available+0x1b/0x20

---

 arch/x86/kernel/traps.c | 2 ++
 1 file changed, 2 insertions(+)
Quoted 14 lines

arch/x86/kernel/traps.c do_device_not_available(struct pt_regs *regs, long error_code) · line 659

Quoted 11 lines
arch/x86/kernel/traps.c +2 -0
index ab82cbd..dff3e72 100644
@@ -654,7 +654,9 @@ do_device_not_available(struct pt_regs *regs, long error_code)
 		return;
 	}
 #endif
+	exception_enter(regs);
 	math_state_restore(); /* interrupts still off */
+	exception_enter(regs);

Shouldn't that be exception_exit actually?

OH! NO WAY I did that mistake! That was some last minute in-text-editor changes and somehow I screwed this up. Will re-send in a second.

Sorry, and thank you.

	-ss
maintainer
#3 lore raw
Quoted 5 lines · Sergey Senozhatsky

OH! NO WAY I did that mistake! That was some last minute in-text-editor changes and somehow I screwed this up. Will re-send in a second.

Sorry, and thank you.

I wouldn't worry - it can happen to anyone (and I'm talking from experience here) :-).

Signature · 2 lines
Regards/Gruss,
Boris.
#4 lore raw

Re: [RFC PATCH] x86: use new exception_enter()/exception_exit() API in do_device_not_available() (v2)

In reply to Borislav Petkov

x86: use new exception_enter()/exception_exit() API in do_device_not_available()

RCU can be in extended quiescent state when do_device_not_available() called.

[  676.400274] RCU used illegally from idle CPU!
[  676.400274] rcu_scheduler_active = 1, debug_locks = 0
[  676.400278] RCU used illegally from extended quiescent state!
[  676.400282] 2 locks held by sed/9468:
[  676.400285]  #0:  (&rq->lock){-.-.-.}, at: [<ffffffff814c67cf>] __schedule+0x1c0/0x992
[  676.400304]  #1:  (rcu_read_lock){.+.+..}, at: [<ffffffff8106a880>] cpuacct_charge+0x27/0x1e0
[  676.400321] 
[  676.400329] Call Trace:
[  676.400340]  [<ffffffff8108933a>] lockdep_rcu_suspicious+0x109/0x112
[  676.400347]  [<ffffffff8106a8f7>] cpuacct_charge+0x9e/0x1e0
[  676.400353]  [<ffffffff8106a880>] ? cpuacct_charge+0x27/0x1e0
[  676.400359]  [<ffffffff814c67cf>] ? __schedule+0x1c0/0x992
[  676.400366]  [<ffffffff8106d525>] update_curr+0x159/0x200
[  676.400375]  [<ffffffff8100a04f>] ? init_fpu+0x4e/0x8f
[  676.400382]  [<ffffffff8106e0da>] put_prev_task_fair+0x2a/0xac
[  676.400388]  [<ffffffff814c69a7>] __schedule+0x398/0x992
[  676.400397]  [<ffffffff810b4f0a>] ? rcu_irq_exit+0x83/0x99
[  676.400404]  [<ffffffff814c8bf0>] ? retint_restore_args+0x13/0x13
[  676.400411]  [<ffffffff810889e9>] ? lock_is_held+0x2d/0xa5
[  676.400414]  [<ffffffff8100a04f>] ? init_fpu+0x4e/0x8f
[  676.400417]  [<ffffffff81065d0a>] __cond_resched+0x28/0x34
[  676.400420]  [<ffffffff814c701a>] _cond_resched+0x2e/0x35
[  676.400425]  [<ffffffff8111cad4>] kmem_cache_alloc+0x4c/0x1c5
[  676.400430]  [<ffffffff810021ff>] ? math_state_restore+0x1f/0xdc
[  676.400433]  [<ffffffff8100a04f>] init_fpu+0x4e/0x8f
[  676.400437]  [<ffffffff8100220e>] math_state_restore+0x2e/0xdc
[  676.400440]  [<ffffffff814c982e>] do_device_not_available+0xe/0x10
[  676.400445]  [<ffffffff814d161b>] device_not_available+0x1b/0x20

---

 arch/x86/kernel/traps.c | 2 ++
 1 file changed, 2 insertions(+)
arch/x86/kernel/traps.c +2 -0
index ab82cbd..fa226ef 100644
@@ -654,7 +654,9 @@ do_device_not_available(struct pt_regs *regs, long error_code)
 		return;
 	}
 #endif
+	exception_enter(regs);
 	math_state_restore(); /* interrupts still off */
+	exception_exit(regs);
 #ifdef CONFIG_X86_32
 	conditional_sti(regs);
 #endif