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

[Steven Rostedt <rostedt@goodmis.org>]

On Mon, 16 Nov 2015 09:35:53 +0800
yalin wang <yalin.wang2010@gmail.com> wrote:

> > On Nov 13, 2015, at 22:01, Steven Rostedt <rostedt@goodmis.org> wrote:
> > 
> > On Fri, 13 Nov 2015 19:54:11 +0800
> > yalin wang <yalin.wang2010@gmail.com> wrote:
> >   
> >>>>> 	TP_fast_assign(
> >>>>> 		__entry->mm = mm;
> >>>>> -		__entry->pfn = pfn;
> >>>>> +		__entry->pfn = page_to_pfn(page);    
> >>>> 
> >>>> Instead of the condition, we could have:
> >>>> 
> >>>> 	__entry->pfn = page ? page_to_pfn(page) : -1;    
> >>> 
> >>> I agree. Please do it like this.    
> > 
> > hmm, pfn is defined as an unsigned long, would -1 be the best.
> > Or should it be (-1UL).
> > 
> > Then we could also have:
> > 
> >        TP_printk("mm=%p, scan_pfn=0x%lx%s, writable=%d, referenced=%d, none_or_zero=%d, status=%s, unmapped=%d",
> >                __entry->mm,
> >                __entry->pfn == (-1UL) ? 0 : __entry->pfn,
> > 		__entry->pfn == (-1UL) ? "(null)" : "",
> > 
> > Note the added %s after %lx I have in the print format.
> > 
> > -- Steve  
> it is not easy to print for perf tools in userspace ,
> if you use this format ,
> for user space perf tool, it print the entry by look up the member in entry struct by offset ,
> you print a dynamic string which user space perf tool don’t know how to print this string .

Have you tried it? It should work. If not, I'll fix it. The string
"null" is exported in the trace output file, and perf should have
enough information to know how to handle that. If it fails to parse, I
can easily fix it.

Remember, I'm the author of the parsing of events in userspace.

-- Steve