Re: [PATCH V4] mm: fix kernel crash in khugepaged thread

[Vlastimil Babka <vbabka@suse.cz>]

On 17.11.2015 4:58, yalin wang wrote:
> 
>> On Nov 17, 2015, at 10:43, Steven Rostedt <rostedt@goodmis.org> wrote:
>>
>> On Tue, 17 Nov 2015 10:21:47 +0800
>> yalin wang <yalin.wang2010@gmail.com> wrote:
>>
>>
>>
>> Because the print_fmt has nothing to do with the fields. You can have
>> as your print_fmt as:
>>
>> 	TP_printk("Message = %s", "hello dolly!")
>>
>> And both userspace and the kernel with process that correctly (if I got
>> string processing working in userspace, which I believe I do). The
>> string is processed, it's not dependent on TP_STRUCT__entry() unless it
>> references a field there. Which can also be used too:
>>
>> 	TP_printk("Message = %s", __entry->musical ? "Hello dolly!" :
>> 			"Death Trap!")
>>
>> userspace will see in the entry:
>>
>> print_fmt: "Message = %s", REC->musical ? "Hello dolly!" : "Death Trap!"
>>
>> as long as the field "musical" exists, all is well.
>>
>> -- Steve
> Aha,  i see.
> Thanks very much for your explanation.
> Better print fat is :   
> TP_printk("mm=%p, scan_pfn=%s, writable=%d, referenced=%d, none_or_zero=%d, status=%s, unmapped=%d",
>                __entry->mm,
> 		__entry->pfn == (-1UL) ? "(null)" :  itoa(buff,  __entry->pin, 10), a?|..)
> 
> is this possible ?

I doubt so.

Why don't we just do (with %lx):
 __entry->pfn != -1UL ? __entry->pfn : 0,

Status already tells us that it's not a real pfn 0 (which I doubt would be
userspace-mapped and thus reachable by khugepaged anyway?).
Also it's what some other tracepoints do, see e.g. mm_page class in
include/trace/events/kmem.h.

> Thanks
> 
> 
> 
> 
> 
> 
> 
> 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>