2014-12-23 15:15:15

by Min-Hua Chen

Subject: [PATCH] arm64: setup return path for el1_undef

Setup return path for el1_undef since el1_undef may
be handled by handlers.

asmlinkage void __exception do_undefinstr(struct pt_regs *regs)
{
siginfo_t info;
void __user *pc = (void __user *)instruction_pointer(regs);

/* check for AArch32 breakpoint instructions */
if (!aarch32_break_handler(regs))
return;

if (call_undef_hook(regs) == 0)
return;

...
}

Signed-off-by: Min-Hua Chen <[email protected]>
---
arch/arm64/kernel/entry.S | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/arm64/kernel/entry.S b/arch/arm64/kernel/entry.S
index fd4fa37..86ebec5 100644
--- a/arch/arm64/kernel/entry.S
+++ b/arch/arm64/kernel/entry.S
@@ -313,7 +313,8 @@ el1_undef:
*/
enable_dbg
mov x0, sp
- b do_undefinstr
+ bl do_undefinstr
+ kernel_exit 1
el1_dbg:
/*
* Debug exception handling
--
1.7.10.4

2014-12-23 15:57:26

by Catalin Marinas

Subject: Re: [PATCH] arm64: setup return path for el1_undef

On Tue, Dec 23, 2014 at 03:15:10PM +0000, Min-Hua Chen wrote:
> Setup return path for el1_undef since el1_undef may
> be handled by handlers.

Did you find a real issue or it was just code inspection.

> asmlinkage void __exception do_undefinstr(struct pt_regs *regs)
> {
> siginfo_t info;
> void __user *pc = (void __user *)instruction_pointer(regs);
>
> /* check for AArch32 breakpoint instructions */
> if (!aarch32_break_handler(regs))
> return;
>
> if (call_undef_hook(regs) == 0)
> return;
>
> ...
> }
> Signed-off-by: Min-Hua Chen <[email protected]>
> ---
> arch/arm64/kernel/entry.S | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/arch/arm64/kernel/entry.S b/arch/arm64/kernel/entry.S
> index fd4fa37..86ebec5 100644
> --- a/arch/arm64/kernel/entry.S
> +++ b/arch/arm64/kernel/entry.S
> @@ -313,7 +313,8 @@ el1_undef:
> */
> enable_dbg
> mov x0, sp
> - b do_undefinstr
> + bl do_undefinstr
> + kernel_exit 1
> el1_dbg:
> /*
> * Debug exception handling

I don't think this is needed. The code is pretty convoluted but for an
EL1 undefined exception we should never return from do_undefinstr(). The
call_undef_hook() function returns 1 if !user_mode(regs) and this should
cause a kernel panic. Basically we do not allow any kind of undefined
instructions in the arm64 kernel.

--
Catalin

2014-12-23 16:41:22

by Min-Hua Chen

Subject: Re: [PATCH] arm64: setup return path for el1_undef

On Tue, Dec 23, 2014 at 11:57 PM, Catalin Marinas
<[email protected]> wrote:
> On Tue, Dec 23, 2014 at 03:15:10PM +0000, Min-Hua Chen wrote:
>> Setup return path for el1_undef since el1_undef may
>> be handled by handlers.
>
> Did you find a real issue or it was just code inspection.

Thanks for your reply. It was just a code inspection.

Min-Hua

>
>> asmlinkage void __exception do_undefinstr(struct pt_regs *regs)
>> {
>> siginfo_t info;
>> void __user *pc = (void __user *)instruction_pointer(regs);
>>
>> /* check for AArch32 breakpoint instructions */
>> if (!aarch32_break_handler(regs))
>> return;
>>
>> if (call_undef_hook(regs) == 0)
>> return;
>>
>> ...
>> }
>> Signed-off-by: Min-Hua Chen <[email protected]>
>> ---
>> arch/arm64/kernel/entry.S | 3 ++-
>> 1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/arm64/kernel/entry.S b/arch/arm64/kernel/entry.S
>> index fd4fa37..86ebec5 100644
>> --- a/arch/arm64/kernel/entry.S
>> +++ b/arch/arm64/kernel/entry.S
>> @@ -313,7 +313,8 @@ el1_undef:
>> */
>> enable_dbg
>> mov x0, sp
>> - b do_undefinstr
>> + bl do_undefinstr
>> + kernel_exit 1
>> el1_dbg:
>> /*
>> * Debug exception handling
>
> I don't think this is needed. The code is pretty convoluted but for an
> EL1 undefined exception we should never return from do_undefinstr(). The
> call_undef_hook() function returns 1 if !user_mode(regs) and this should
> cause a kernel panic. Basically we do not allow any kind of undefined
> instructions in the arm64 kernel.
>
> --
> Catalin