Re: mmotm 2020-03-03-22-28 uploaded (warning: objtool:)

[Randy Dunlap <rdunlap@infradead.org>]

On 3/5/20 1:54 AM, Peter Zijlstra wrote:
> On Thu, Mar 05, 2020 at 05:16:22PM +0800, Walter Wu wrote:
>> On Thu, 2020-03-05 at 09:18 +0100, Peter Zijlstra wrote:
>>> On Thu, Mar 05, 2020 at 09:17:17AM +0100, Peter Zijlstra wrote:
>>>> On Wed, Mar 04, 2020 at 09:34:49AM -0800, Randy Dunlap wrote:
>>>
>>>>> mm/kasan/common.o: warning: objtool: kasan_report()+0x13: call to report_enabled() with UACCESS enabled
>>>>
>>>> I used next/master instead, and found the below broken commit
>>>> responsible for this.

Yes, I see that same warning in linux-next of 20200305.

>>>> @@ -634,12 +637,20 @@ void kasan_free_shadow(const struct vm_struct *vm)
>>>>  #endif
>>>>  
>>>>  extern void __kasan_report(unsigned long addr, size_t size, bool is_write, unsigned long ip);
>>>> +extern bool report_enabled(void);
>>>>  
>>>> -void kasan_report(unsigned long addr, size_t size, bool is_write, unsigned long ip)
>>>> +bool kasan_report(unsigned long addr, size_t size, bool is_write, unsigned long ip)
>>>>  {
>>>> -	unsigned long flags = user_access_save();
>>>> +	unsigned long flags;
>>>> +
>>>> +	if (likely(!report_enabled()))
>>>> +		return false;
>>>
>>> This adds an explicit call before the user_access_save() and that is a
>>> straight on bug.
>>>
>> Hi Peter,
>>
>> Thanks for your help. Unfortunately, I don't reproduce it in our
>> environment, so I have asked Stephen, if I can reproduce it, then we
>> will send new patch.
> 
> The patch is trivial; and all you need is an x86_64 (cross) compiler to
> reproduce.
> 
> 
> diff --git a/mm/kasan/common.c b/mm/kasan/common.c
> index ad2dc0c9cc17..2906358e42f0 100644
> --- a/mm/kasan/common.c
> +++ b/mm/kasan/common.c
> @@ -618,16 +618,17 @@ extern bool report_enabled(void);
>  
>  bool kasan_report(unsigned long addr, size_t size, bool is_write, unsigned long ip)
>  {
> -	unsigned long flags;
> +	unsigned long flags = user_access_save();
> +	bool ret = false;
>  
> -	if (likely(!report_enabled()))
> -		return false;
> +	if (likely(report_enabled())) {
> +		__kasan_report(addr, size, is_write, ip);
> +		ret = true;
> +	}
>  
> -	flags = user_access_save();
> -	__kasan_report(addr, size, is_write, ip);
>  	user_access_restore(flags);
>  
> -	return true;
> +	return ret;
>  }
>  
>  #ifdef CONFIG_MEMORY_HOTPLUG
> 

and that fixes the warning.  Thanks.

Acked-by: Randy Dunlap <rdunlap@infradead.org> # build-tested

-- 
~Randy