Re: Re: [PATCH perf/core 1/2] [BUGFIX] perf-probe: Fix to handle optimized not-inlined but has no instance

From: Masami Hiramatsu
Date: Sun Feb 08 2015 - 03:37:26 EST


(2015/01/30 23:21), Arnaldo Carvalho de Melo wrote:
> Em Fri, Jan 30, 2015 at 06:37:44PM +0900, Masami Hiramatsu escreveu:
>> Fix to handle optimized no-inline functions which have
>> only function definition but no actual instance at
>> that point. To fix this problem, we need to find actual
>> instance of the function.
>>
>> Without this patch:
>> ----
>> # perf probe -a __up
>> Failed to get entry address of __up.
>> Error: Failed to add events.
>> # perf probe -L __up
>> Specified source line is not found.
>> Error: Failed to show lines.
>> ----
>>
>> With this patch:
>> ----
>> # perf probe -a __up
>> Added new event:
>> probe:__up (on __up)
>>
>> You can now use it in all perf tools, such as:
>>
>> perf record -e probe:__up -aR sleep 1
>>
>> # perf probe -L __up
>> <__up@/home/fedora/ksrc/linux-3/kernel/locking/semaphore.c:0>
>> 0 static noinline void __sched __up(struct semaphore *sem)
>> {
>> struct semaphore_waiter *waiter = list_first_entry(&sem->wait_
>> struct semaphore_waite
>> 4 list_del(&waiter->list);
>> 5 waiter->up = true;
>> 6 wake_up_process(waiter->task);
>> 7 }
>> ----
>
> Since __up here was built in some other way, I looked for another
> 'noinline' function to try, and it failed here:
>
> [root@zoo ~]# perf probe -L vmalloc_fault | head -5
> <vmalloc_fault@/home/git/linux/arch/x86/mm/fault.c:0>
> 0 static noinline int vmalloc_fault(unsigned long address)
> 1 {
> pgd_t *pgd, *pgd_ref;
> pud_t *pud, *pud_ref;
> [root@zoo ~]# perf probe vmalloc_fault
> Added new event:
> Failed to write event: Invalid argument
> Error: Failed to add events.
>
> ---->>>>>> APPLY THE PATCH, rebuild and try again:
>
> [root@zoo ~]# perf probe vmalloc_fault
> Added new event:
> Failed to write event: Invalid argument
> Error: Failed to add events.
> [root@zoo ~]# perf probe -v vmalloc_fault
> probe-definition(0): vmalloc_fault
> symbol:vmalloc_fault file:(null) line:0 offset:0 return:0 lazy:(null)
> 0 arguments
> Looking at the vmlinux_path (7 entries long)
> Using /lib/modules/3.19.0-rc6+/build/vmlinux for symbols
> Open Debuginfo file: /lib/modules/3.19.0-rc6+/build/vmlinux
> Try to find probe point from debuginfo.
> Probe point found: vmalloc_fault+0
> Found 1 probe_trace_events.
> Opening /sys/kernel/debug/tracing/kprobe_events write=1
> Added new event:
> Writing event: p:probe/vmalloc_fault _text+289600
> Failed to write event: Invalid argument
> Error: Failed to add events. Reason: Invalid argument (Code: -22)
> [root@zoo ~]# grep -w vmalloc_fault /proc/kallsyms
> ffffffff81046b40 t vmalloc_fault
> [root@zoo ~]# grep -w _text /proc/kallsyms
> ffffffff81000000 T _text
> [root@zoo ~]# python
> Python 2.7.5 (default, Nov 3 2014, 14:26:24)
> [GCC 4.8.3 20140911 (Red Hat 4.8.3-7)] on linux2
> Type "help", "copyright", "credits" or "license" for more information.
> >>> 0xffffffff81046b40 - 0xffffffff81000000
> 289600L
> >>>
> [root@zoo ~]# readelf -s /lib/modules/3.19.0-rc6+/build/vmlinux | egrep -w _text\|vmalloc_fault
> 3499: ffffffff81046b40 410 FUNC LOCAL DEFAULT 1 vmalloc_fault
> 48873: ffffffff81000000 0 NOTYPE GLOBAL DEFAULT 1 _text
> [root@zoo ~]#
>
> ---------------------------------------------------------------------------
>
> So perhaps I need a 'noinline' _and_ '__sched', so that I have a "optimized
> no-inline functions which have only function definition but no actual instance
> at that point"?

Ok, actually vmalloc_fault is marked as a nokprobe symbol.

$ grep vmalloc_fault -r arch/x86/
arch/x86/mm/fault.c:static noinline int vmalloc_fault(unsigned long address)
arch/x86/mm/fault.c:NOKPROBE_SYMBOL(vmalloc_fault);
arch/x86/mm/fault.c:static noinline int vmalloc_fault(unsigned long address)
arch/x86/mm/fault.c:NOKPROBE_SYMBOL(vmalloc_fault);
arch/x86/mm/fault.c: if (vmalloc_fault(address) >= 0)

All the symbols which marked by NOKPROBE_SYMBOL macro can not be
kprobed. And we now have <debugfs>/kprobes/blacklist special file
to get all nokprobe symbols.
I'll make another perf-probe patch which looks the blacklist before
trying define new events.

Thank you,

>
> Looking for one, will post here after objdump -dS finishes...
>
> - Arnaldo
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>


--
Masami HIRAMATSU
Software Platform Research Dept. Linux Technology Research Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt@xxxxxxxxxxx


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/