diff mbox

Fw: [PATCH 01/01] arm: fix a preempt_count() corruption for CONFIG_PREEMPT=n

Message ID 50D10224.30600@codeaurora.org (mailing list archive)
State New, archived
Headers show

Commit Message

Stephen Boyd Dec. 18, 2012, 11:54 p.m. UTC 
On 12/14/12 23:51, Patrik, Kluba wrote:
> Date: Fri, 14 Dec 2012 15:13:57 +0100
> From: "Patrik, Kluba" <pkluba@dension.com>
> To: linux-kernel@vger.kernel.org
> Subject: [PATCH 01/01] arm: fix a preempt_count() corruption for
> CONFIG_PREEMPT=n
>
>
> Hi All,
>
> After a hard days' work, I've managed to track down a ghost causing
> sporadic kernel warnings and system crashes. The patch applies for the
> HEAD of linux-stable. I don't know how to create and submit patches
> correctly, and hope that somebody will pick it up...
>
> Regards,
> Patrik
> ---
> From 4dd31e88a2715389e5ac198dcf00b48b4f148ea6 Mon Sep 17 00:00:00 2001
> From: Patrik Kluba <pkluba@dension.com>
> Date: Fri, 14 Dec 2012 14:36:27 +0100
> Subject: [PATCH] arm: fix a preempt_count() corruption for
> CONFIG_PREEMPT=n
>
> All the preempt-disabling assembly code is under CONFIG_PREEMPT,
> while VFP_bounce calls preempt_enable() uncoditionally, leading to
> a preempt_count() corruption (gets set to 0xffffffff). Fix the
> imbalance by adding #ifdef CONFIG_PREEMPT before preempt_enable().
> ---
>  arch/arm/vfp/vfpmodule.c |    3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/arch/arm/vfp/vfpmodule.c b/arch/arm/vfp/vfpmodule.c
> index 3b44e0d..2184f7e 100644
> --- a/arch/arm/vfp/vfpmodule.c
> +++ b/arch/arm/vfp/vfpmodule.c
> @@ -428,7 +428,10 @@ void VFP_bounce(u32 trigger, u32 fpexc, struct
> pt_regs *regs) if (exceptions)
>  		vfp_raise_exceptions(exceptions, trigger, orig_fpscr,
> regs); exit:
> +#ifdef CONFIG_PREEMPT
>  	preempt_enable();
> +#endif
> +	return;

From include/linux/preempt.h

#ifdef CONFIG_PREEMPT_COUNT

...

#define preempt_enable() \
do { \
        preempt_enable_no_resched(); \
        barrier(); \
        preempt_check_resched(); \
} while (0)

...

#else /* !CONFIG_PREEMPT_COUNT */

...

#define preempt_enable()                do { } while (0)

...

#endif

And CONFIG_PREEMPT_COUNT=y only if CONFIG_PREEMPT=y but your subject
says PREEMPT_COUNT=n. Do you by chance have DEBUG_ATOMIC_SLEEP=y
selected in your config? That seems to select PREEMPT_COUNT as well and
so I believe we need to move ARM assembly to use PREEMPT_COUNT instead
of plain PREEMPT in its preempt_count changing assembly code.

Can you try this patch? I'll probably send a follow up patch that
consolidates the duplicated preempt_decrement code into some assembly
macro. We may also want to fix it up so that assembly stubs call into
the preempt tracer. I'm not sure how that works right now with VFP.

---->8------

Comments

Russell King - ARM Linux Dec. 19, 2012, 12:01 a.m. UTC | #1 
On Tue, Dec 18, 2012 at 03:54:12PM -0800, Stephen Boyd wrote:
> Can you try this patch? I'll probably send a follow up patch that
> consolidates the duplicated preempt_decrement code into some assembly
> macro. We may also want to fix it up so that assembly stubs call into
> the preempt tracer. I'm not sure how that works right now with VFP.

This looks like a sensible fix to me - once tested of course.
Stephen Boyd Jan. 3, 2013, 12:41 a.m. UTC | #2 
On 12/18/12 16:01, Russell King - ARM Linux wrote:
> On Tue, Dec 18, 2012 at 03:54:12PM -0800, Stephen Boyd wrote:
>> Can you try this patch? I'll probably send a follow up patch that
>> consolidates the duplicated preempt_decrement code into some assembly
>> macro. We may also want to fix it up so that assembly stubs call into
>> the preempt tracer. I'm not sure how that works right now with VFP.
> This looks like a sensible fix to me - once tested of course.

Kluba, did the patch work for you?
Patrik, Kluba Jan. 4, 2013, 7:48 a.m. UTC | #3 
On Wed, 02 Jan 2013 16:41:00 -0800
Stephen Boyd <sboyd@codeaurora.org> wrote:

> On 12/18/12 16:01, Russell King - ARM Linux wrote:
> > On Tue, Dec 18, 2012 at 03:54:12PM -0800, Stephen Boyd wrote:
> >> Can you try this patch? I'll probably send a follow up patch that
> >> consolidates the duplicated preempt_decrement code into some
> >> assembly macro. We may also want to fix it up so that assembly
> >> stubs call into the preempt tracer. I'm not sure how that works
> >> right now with VFP.
> > This looks like a sensible fix to me - once tested of course.
> 
> Kluba, did the patch work for you?
> 

Sorry for being late, was on holiday. Yes, you were right,
CONFIG_DEBUG_PREEMPT (after breaking dependency on CONFIG_PREEMPT -
this is how I caught the exact place) and CONFIG_DEBUG_ATOMIC_SLEEP was
enabled due to bug hunting in own code (and found yet another at an
unexpected place). 
I have chosen #ifdef CONFIG_PREEMPT to match with assembly code, but
your solution seems more reasonable.

Regards,
Patrik
Stephen Boyd Jan. 4, 2013, 10:21 p.m. UTC | #4 
On 01/03/13 23:48, Patrik, Kluba wrote:
> Sorry for being late, was on holiday. Yes, you were right,
> CONFIG_DEBUG_PREEMPT (after breaking dependency on CONFIG_PREEMPT -
> this is how I caught the exact place) and CONFIG_DEBUG_ATOMIC_SLEEP was
> enabled due to bug hunting in own code (and found yet another at an
> unexpected place). 
> I have chosen #ifdef CONFIG_PREEMPT to match with assembly code, but
> your solution seems more reasonable.

Did you test my patch? If so, can I get your tested by please so I can
stick it in the patch tracker?

Thanks,
Stephen
Patrik, Kluba Jan. 8, 2013, 11:51 a.m. UTC | #5 
On Fri, 04 Jan 2013 14:21:46 -0800
Stephen Boyd <sboyd@codeaurora.org> wrote:

> On 01/03/13 23:48, Patrik, Kluba wrote:
> > Sorry for being late, was on holiday. Yes, you were right,
> > CONFIG_DEBUG_PREEMPT (after breaking dependency on CONFIG_PREEMPT -
> > this is how I caught the exact place) and CONFIG_DEBUG_ATOMIC_SLEEP
> > was enabled due to bug hunting in own code (and found yet another
> > at an unexpected place). 
> > I have chosen #ifdef CONFIG_PREEMPT to match with assembly code, but
> > your solution seems more reasonable.
> 
> Did you test my patch? If so, can I get your tested by please so I can
> stick it in the patch tracker?

Yes, I have tested it. Here you are:

Tested-by: Patrik Kluba <pkluba@dension.com>

Bye,
Patrik
diff mbox

Patch

diff --git a/arch/arm/vfp/entry.S b/arch/arm/vfp/entry.S
index cc926c9..323ce1a 100644
--- a/arch/arm/vfp/entry.S
+++ b/arch/arm/vfp/entry.S
@@ -22,7 +22,7 @@ 
 @  IRQs disabled.
 @
 ENTRY(do_vfp)
-#ifdef CONFIG_PREEMPT
+#ifdef CONFIG_PREEMPT_COUNT
        ldr     r4, [r10, #TI_PREEMPT]  @ get preempt count
        add     r11, r4, #1             @ increment it
        str     r11, [r10, #TI_PREEMPT]
@@ -35,7 +35,7 @@  ENTRY(do_vfp)
 ENDPROC(do_vfp)
 
 ENTRY(vfp_null_entry)
-#ifdef CONFIG_PREEMPT
+#ifdef CONFIG_PREEMPT_COUNT
        get_thread_info r10
        ldr     r4, [r10, #TI_PREEMPT]  @ get preempt count
        sub     r11, r4, #1             @ decrement it
@@ -53,7 +53,7 @@  ENDPROC(vfp_null_entry)
 
        __INIT
 ENTRY(vfp_testing_entry)
-#ifdef CONFIG_PREEMPT
+#ifdef CONFIG_PREEMPT_COUNT
        get_thread_info r10
        ldr     r4, [r10, #TI_PREEMPT]  @ get preempt count
        sub     r11, r4, #1             @ decrement it
diff --git a/arch/arm/vfp/vfphw.S b/arch/arm/vfp/vfphw.S
index ea0349f..dd5e56f 100644
--- a/arch/arm/vfp/vfphw.S
+++ b/arch/arm/vfp/vfphw.S
@@ -168,7 +168,7 @@  vfp_hw_state_valid:
                                        @ else it's one 32-bit instruction, so
                                        @ always subtract 4 from the following
                                        @ instruction address.
-#ifdef CONFIG_PREEMPT
+#ifdef CONFIG_PREEMPT_COUNT
        get_thread_info r10
        ldr     r4, [r10, #TI_PREEMPT]  @ get preempt count
        sub     r11, r4, #1             @ decrement it
@@ -192,7 +192,7 @@  look_for_VFP_exceptions:
        @ not recognised by VFP
 
        DBGSTR  "not VFP"
-#ifdef CONFIG_PREEMPT
+#ifdef CONFIG_PREEMPT_COUNT
        get_thread_info r10
        ldr     r4, [r10, #TI_PREEMPT]  @ get preempt count
        sub     r11, r4, #1             @ decrement it