[PATCH] x86: Use is_kprobe_fault to better match usage

[PATCH] x86: Use is_kprobe_fault to better match usage

[Harvey Harrison]

Currently the notify_page_fault helper is used to test it the page
fault was caused by a kprobe causing an early return from do_page_fault.

Change the name of the helper to is_kprobe_fault to match the usage and
remove the preempt_disable/enable pair around kprobe_running() with an
explicit test for preemption.  The idea for this comes from a patch
by Quentin Barnes to kprobes.c

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
---
Ingo, this may not be functionally equivalent, feel free to yank it out
if there is any trouble, but from what I've seen it should be OK.

Did you ever find a good kprobes test?

 arch/x86/mm/fault_32.c |   30 ++++++++++++++----------------
 arch/x86/mm/fault_64.c |   30 ++++++++++++++----------------
 2 files changed, 28 insertions(+), 32 deletions(-)

diff --git a/arch/x86/mm/fault_32.c b/arch/x86/mm/fault_32.c
index 051a4ec..5c48cc2 100644
--- a/arch/x86/mm/fault_32.c
+++ b/arch/x86/mm/fault_32.c
@@ -12,6 +12,7 @@
 #include <linux/mman.h>
 #include <linux/mm.h>
 #include <linux/smp.h>
+#include <linux/hardirq.h>
 #include <linux/interrupt.h>
 #include <linux/init.h>
 #include <linux/tty.h>
@@ -42,23 +43,20 @@
 #define PF_RSVD	(1<<3)
 #define PF_INSTR	(1<<4)
 
-static inline int notify_page_fault(struct pt_regs *regs)
+static inline int is_kprobe_fault(struct pt_regs *regs)
 {
-#ifdef CONFIG_KPROBES
 	int ret = 0;
-
-	/* kprobe_running() needs smp_processor_id() */
-	if (!user_mode_vm(regs)) {
-		preempt_disable();
-		if (kprobe_running() && kprobe_fault_handler(regs, 14))
-			ret = 1;
-		preempt_enable();
-	}
-
-	return ret;
-#else
-	return 0;
+#ifdef CONFIG_KPROBES
+	/*
+	 * If it is a kprobe fault we can not be premptible so return before
+	 * calling kprobe_running() as it will assert on smp_processor_id if
+	 * preemption is enabled.
+	 */
+	if (!user_mode_vm(regs) && !preemptible() && kprobe_running() &&
+	    kprobe_fault_handler(regs, 14))
+		ret = 1;
 #endif
+	return ret;
 }
 
 #ifdef CONFIG_X86_32
@@ -428,7 +426,7 @@ void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code)
 				return;
 		}
 #endif
-		if (notify_page_fault(regs))
+		if (is_kprobe_fault(regs))
 			return;
 		/*
 		 * Don't take the mm semaphore here. If we fixup a prefetch
@@ -437,7 +435,7 @@ void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code)
 		goto bad_area_nosemaphore;
 	}
 
-	if (notify_page_fault(regs))
+	if (is_kprobe_fault(regs))
 		return;
 
 	/* It's safe to allow irq's after cr2 has been saved and the vmalloc
diff --git a/arch/x86/mm/fault_64.c b/arch/x86/mm/fault_64.c
index 97b92b6..09008e5 100644
--- a/arch/x86/mm/fault_64.c
+++ b/arch/x86/mm/fault_64.c
@@ -13,6 +13,7 @@
 #include <linux/mman.h>
 #include <linux/mm.h>
 #include <linux/smp.h>
+#include <linux/hardirq.h>
 #include <linux/interrupt.h>
 #include <linux/init.h>
 #include <linux/tty.h>
@@ -45,23 +46,20 @@
 #define PF_RSVD	(1<<3)
 #define PF_INSTR	(1<<4)
 
-static inline int notify_page_fault(struct pt_regs *regs)
+static inline int is_kprobe_fault(struct pt_regs *regs)
 {
-#ifdef CONFIG_KPROBES
 	int ret = 0;
-
-	/* kprobe_running() needs smp_processor_id() */
-	if (!user_mode(regs)) {
-		preempt_disable();
-		if (kprobe_running() && kprobe_fault_handler(regs, 14))
-			ret = 1;
-		preempt_enable();
-	}
-
-	return ret;
-#else
-	return 0;
+#ifdef CONFIG_KPROBES
+	/*
+	 * If it is a kprobe fault we can not be premptible so return before
+	 * calling kprobe_running() as it will assert on smp_processor_id if
+	 * preemption is enabled.
+	 */
+	if (!user_mode_vm(regs) && !preemptible() && kprobe_running() &&
+	    kprobe_fault_handler(regs, 14))
+		ret = 1;
 #endif
+	return ret;
 }
 
 #ifdef CONFIG_X86_32
@@ -478,7 +476,7 @@ asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
 				return;
 		}
 #endif
-		if (notify_page_fault(regs))
+		if (is_kprobe_fault(regs))
 			return;
 		/*
 		 * Don't take the mm semaphore here. If we fixup a prefetch
@@ -487,7 +485,7 @@ asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
 		goto bad_area_nosemaphore;
 	}
 
-	if (notify_page_fault(regs))
+	if (is_kprobe_fault(regs))
 		return;
 
 	if (likely(regs->flags & X86_EFLAGS_IF))
-- 
1.5.4.rc2.1097.gb6e0d

Re: [PATCH] x86: Use is_kprobe_fault to better match usage

[Masami Hiramatsu]

Hi Harvey,

Harvey Harrison wrote:
> Currently the notify_page_fault helper is used to test it the page
> fault was caused by a kprobe causing an early return from do_page_fault.
> 
> Change the name of the helper to is_kprobe_fault to match the usage and
> remove the preempt_disable/enable pair around kprobe_running() with an
> explicit test for preemption.  The idea for this comes from a patch
> by Quentin Barnes to kprobes.c

Sure, that's right.
However, since other architectures also have notify_page_fault(),
I think all of those code might better be changed same time for
maintainability.

Thanks,

> Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
> ---
> Ingo, this may not be functionally equivalent, feel free to yank it out
> if there is any trouble, but from what I've seen it should be OK.
> 
> Did you ever find a good kprobes test?
> 
>  arch/x86/mm/fault_32.c |   30 ++++++++++++++----------------
>  arch/x86/mm/fault_64.c |   30 ++++++++++++++----------------
>  2 files changed, 28 insertions(+), 32 deletions(-)
> 
> diff --git a/arch/x86/mm/fault_32.c b/arch/x86/mm/fault_32.c
> index 051a4ec..5c48cc2 100644
> --- a/arch/x86/mm/fault_32.c
> +++ b/arch/x86/mm/fault_32.c
> @@ -12,6 +12,7 @@
>  #include <linux/mman.h>
>  #include <linux/mm.h>
>  #include <linux/smp.h>
> +#include <linux/hardirq.h>
>  #include <linux/interrupt.h>
>  #include <linux/init.h>
>  #include <linux/tty.h>
> @@ -42,23 +43,20 @@
>  #define PF_RSVD	(1<<3)
>  #define PF_INSTR	(1<<4)
>  
> -static inline int notify_page_fault(struct pt_regs *regs)
> +static inline int is_kprobe_fault(struct pt_regs *regs)
>  {
> -#ifdef CONFIG_KPROBES
>  	int ret = 0;
> -
> -	/* kprobe_running() needs smp_processor_id() */
> -	if (!user_mode_vm(regs)) {
> -		preempt_disable();
> -		if (kprobe_running() && kprobe_fault_handler(regs, 14))
> -			ret = 1;
> -		preempt_enable();
> -	}
> -
> -	return ret;
> -#else
> -	return 0;
> +#ifdef CONFIG_KPROBES
> +	/*
> +	 * If it is a kprobe fault we can not be premptible so return before
> +	 * calling kprobe_running() as it will assert on smp_processor_id if
> +	 * preemption is enabled.
> +	 */
> +	if (!user_mode_vm(regs) && !preemptible() && kprobe_running() &&
> +	    kprobe_fault_handler(regs, 14))
> +		ret = 1;
>  #endif
> +	return ret;
>  }
>  
>  #ifdef CONFIG_X86_32
> @@ -428,7 +426,7 @@ void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code)
>  				return;
>  		}
>  #endif
> -		if (notify_page_fault(regs))
> +		if (is_kprobe_fault(regs))
>  			return;
>  		/*
>  		 * Don't take the mm semaphore here. If we fixup a prefetch
> @@ -437,7 +435,7 @@ void __kprobes do_page_fault(struct pt_regs *regs, unsigned long error_code)
>  		goto bad_area_nosemaphore;
>  	}
>  
> -	if (notify_page_fault(regs))
> +	if (is_kprobe_fault(regs))
>  		return;
>  
>  	/* It's safe to allow irq's after cr2 has been saved and the vmalloc
> diff --git a/arch/x86/mm/fault_64.c b/arch/x86/mm/fault_64.c
> index 97b92b6..09008e5 100644
> --- a/arch/x86/mm/fault_64.c
> +++ b/arch/x86/mm/fault_64.c
> @@ -13,6 +13,7 @@
>  #include <linux/mman.h>
>  #include <linux/mm.h>
>  #include <linux/smp.h>
> +#include <linux/hardirq.h>
>  #include <linux/interrupt.h>
>  #include <linux/init.h>
>  #include <linux/tty.h>
> @@ -45,23 +46,20 @@
>  #define PF_RSVD	(1<<3)
>  #define PF_INSTR	(1<<4)
>  
> -static inline int notify_page_fault(struct pt_regs *regs)
> +static inline int is_kprobe_fault(struct pt_regs *regs)
>  {
> -#ifdef CONFIG_KPROBES
>  	int ret = 0;
> -
> -	/* kprobe_running() needs smp_processor_id() */
> -	if (!user_mode(regs)) {
> -		preempt_disable();
> -		if (kprobe_running() && kprobe_fault_handler(regs, 14))
> -			ret = 1;
> -		preempt_enable();
> -	}
> -
> -	return ret;
> -#else
> -	return 0;
> +#ifdef CONFIG_KPROBES
> +	/*
> +	 * If it is a kprobe fault we can not be premptible so return before
> +	 * calling kprobe_running() as it will assert on smp_processor_id if
> +	 * preemption is enabled.
> +	 */
> +	if (!user_mode_vm(regs) && !preemptible() && kprobe_running() &&
> +	    kprobe_fault_handler(regs, 14))
> +		ret = 1;
>  #endif
> +	return ret;
>  }
>  
>  #ifdef CONFIG_X86_32
> @@ -478,7 +476,7 @@ asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
>  				return;
>  		}
>  #endif
> -		if (notify_page_fault(regs))
> +		if (is_kprobe_fault(regs))
>  			return;
>  		/*
>  		 * Don't take the mm semaphore here. If we fixup a prefetch
> @@ -487,7 +485,7 @@ asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
>  		goto bad_area_nosemaphore;
>  	}
>  
> -	if (notify_page_fault(regs))
> +	if (is_kprobe_fault(regs))
>  		return;
>  
>  	if (likely(regs->flags & X86_EFLAGS_IF))

-- 
Masami Hiramatsu

Software Engineer
Hitachi Computer Products (America) Inc.
Software Solutions Division

e-mail: mhiramat@redhat.com, masami.hiramatsu.pt@hitachi.com

Re: [PATCH] x86: Use is_kprobe_fault to better match usage

[Harvey Harrison]

On Wed, 2008-01-02 at 21:36 -0500, Masami Hiramatsu wrote:
> Hi Harvey,
> 
> Harvey Harrison wrote:
> > Currently the notify_page_fault helper is used to test it the page
> > fault was caused by a kprobe causing an early return from do_page_fault.
> > 
> > Change the name of the helper to is_kprobe_fault to match the usage and
> > remove the preempt_disable/enable pair around kprobe_running() with an
> > explicit test for preemption.  The idea for this comes from a patch
> > by Quentin Barnes to kprobes.c
> 
> Sure, that's right.
> However, since other architectures also have notify_page_fault(),
> I think all of those code might better be changed same time for
> maintainability.
> 

How about a static inline in linux/kprobes.h with a big comment above
about when/why the !preemptible() check is sufficient?

Harvey

Re: [PATCH] x86: Use is_kprobe_fault to better match usage

[Masami Hiramatsu]

Harvey Harrison wrote:
> On Wed, 2008-01-02 at 21:36 -0500, Masami Hiramatsu wrote:
>> Hi Harvey,
>>
>> Harvey Harrison wrote:
>>> Currently the notify_page_fault helper is used to test it the page
>>> fault was caused by a kprobe causing an early return from do_page_fault.
>>>
>>> Change the name of the helper to is_kprobe_fault to match the usage and
>>> remove the preempt_disable/enable pair around kprobe_running() with an
>>> explicit test for preemption.  The idea for this comes from a patch
>>> by Quentin Barnes to kprobes.c
>> Sure, that's right.
>> However, since other architectures also have notify_page_fault(),
>> I think all of those code might better be changed same time for
>> maintainability.
>>
> 
> How about a static inline in linux/kprobes.h with a big comment above
> about when/why the !preemptible() check is sufficient?

Hmm, fault handling depends on the architecture. But current
notify_page_fault()s are very similar. I think unifying it is good idea.
We will be happy to review that if you send it.

Many thanks!

> 
> Harvey
> 
> 
> 

-- 
Masami Hiramatsu

Software Engineer
Hitachi Computer Products (America) Inc.
Software Solutions Division

e-mail: mhiramat@redhat.com, masami.hiramatsu.pt@hitachi.com

Re: [PATCH] x86: Use is_kprobe_fault to better match usage

[Ingo Molnar]

* Harvey Harrison <harvey.harrison@gmail.com> wrote:

> Currently the notify_page_fault helper is used to test it the page 
> fault was caused by a kprobe causing an early return from 
> do_page_fault.
> 
> Change the name of the helper to is_kprobe_fault to match the usage 
> and remove the preempt_disable/enable pair around kprobe_running() 
> with an explicit test for preemption.  The idea for this comes from a 
> patch by Quentin Barnes to kprobes.c
> 
> Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>

hm, this doesnt apply:

 Hunk #3 FAILED at 426.
 Hunk #4 succeeded at 413 (offset -22 lines).
 1 out of 4 hunks FAILED -- rejects in file arch/x86/mm/fault_32.c
 patching file arch/x86/mm/fault_64.c
 Hunk #3 FAILED at 476.
 Hunk #4 succeeded at 475 (offset -10 lines).
 1 out of 4 hunks FAILED -- rejects in file arch/x86/mm/fault_64.c

could you double-check x86.git#mm, perhaps we are out of sync with a 
patch somewhere?

	Ingo