[ltt-dev] "Kernel/LTTV event size differs for event udpv4_rcv_extended: kernel 32, LTTV 28" error when view trace

xufeng zhang xufeng.zhang at windriver.com
Wed Nov 10 02:14:39 EST 2010


On 11/09/2010 07:57 PM, Mathieu Desnoyers wrote:
> * Mathieu Desnoyers (compudj at krystal.dyndns.org) wrote:
>    
>> * xufeng zhang (xufeng.zhang at windriver.com) wrote:
>>      
>>> On 11/08/2010 05:55 PM, Mathieu Desnoyers wrote:
>>>        
>>>> * xufeng zhang (xufeng.zhang at windriver.com) wrote:
>>>>
>>>>          
>>>>> On 11/08/2010 09:45 AM, Benjamin Poirier wrote:
>>>>>
>>>>>            
>>>>>> On 05/11/10 02:24 AM, xufeng zhang wrote:
>>>>>>
>>>>>>
>>>>>>              
>>>>>>> Hi,
>>>>>>> I'm using ltt-control-0.86 and lttv-0.12.31 on ARM to trace the kernel,
>>>>>>> however,
>>>>>>>
>>>>>>>
>>>>>>>                
>>>>>> Hello,
>>>>>>
>>>>>> Which version of the kernel tracer patches are you using?
>>>>>>
>>>>>>
>>>>>>              
>>>>> I have query the developers who merged LTTng into kernel, they said
>>>>> there is no incompatibility
>>>>> between LTTV and LTTng, so does there any other reasons which lead to
>>>>> this error?
>>>>>
>>>>> Thanks,
>>>>> Xufeng Zhang
>>>>>
>>>>>            
>>>>>> -Ben
>>>>>>
>>>>>>
>>>>>>
>>>>>>              
>>>>>>> I cannot view the trace after I have successfully generated, here is the
>>>>>>> error
>>>>>>> when I run 'lttv-gui -t trace1' command:
>>>>>>> ** Message: statistics viewer : background computation data ready.
>>>>>>>
>>>>>>> ** (lttv.real:6559): WARNING **: Cannot find pin_in in schedchange 671
>>>>>>>
>>>>>>> ** ERROR **: Kernel/LTTV event size differs for event
>>>>>>> udpv4_rcv_extended: kernel 32, LTTV 28
>>>>>>>
>>>>>>>                
>>>> Please compare the /ltt/probes/net-extended-trace.c file from your
>>>> distro kernel tree to the:
>>>>
>>>> lttng-modules package, file: /probes/net-extended-trace.c
>>>>
>>>> Also compare the /ltt/ltt-type-serializer.h (or was it
>>>> include/linux/ltt-type-serializer.h ?) from your distro kernel with the
>>>> lttng-modules package:
>>>>
>>>> /ltt-type-serializer.h
>>>>
>>>> I remember having fixed this problem in the past months. Look closely at
>>>> the structure layout for "struct serialize_l214421224411111", "struct
>>>> serialize_l4412228" and "struct serialize_l4421224411111" vs the types
>>>> passed to the DEFINE_MARKER_TP() in net-extended-trace.c.
>>>>
>>>>          
>>> I have compare my kernel tree to lttng-modules-0.19.2, there is no
>>> different
>>> in ltt-type-serializer.h file, and below is the diff info for
>>> net-extended-trace.c:
>>>
>>> --- net-extended-trace.c.kernel 2010-11-09 15:09:24.000000000 +0800
>>> +++ net-extended-trace.c        2010-11-09 15:12:47.000000000 +0800
>>> @@ -20,14 +20,15 @@
>>>
>>>   #include<linux/in_route.h>
>>>   #include<linux/ip.h>
>>> -#include<linux/ltt-type-serializer.h>
>>>   #include<linux/module.h>
>>>   #include<linux/tcp.h>
>>>   #include<linux/udp.h>
>>>   #include<net/route.h>
>>>   #include<trace/net.h>
>>>
>>> -void probe_net_dev_xmit_extended(struct sk_buff *skb);
>>> +#include "../ltt-type-serializer.h"
>>> +
>>> +void probe_net_dev_xmit_extended(void *_data, struct sk_buff *skb);
>>>
>>>   DEFINE_MARKER_TP(net, dev_xmit_extended, net_dev_xmit,
>>>          probe_net_dev_xmit_extended, "skb 0x%lX network_protocol #n2u%hu "
>>> @@ -35,7 +36,7 @@
>>>          "tot_len #n2u%hu ihl #1u%u source #n2u%hu dest #n2u%hu seq
>>> #n4u%lu "
>>>          "ack_seq #n4u%lu doff #1u%u ack #1u%u rst #1u%u syn #1u%u fin
>>> #1u%u");
>>>
>>> -notrace void probe_net_dev_xmit_extended(struct sk_buff *skb)
>>> +notrace void probe_net_dev_xmit_extended(void *_data, struct sk_buff *skb)
>>>   {
>>>          struct marker *marker;
>>>          struct serialize_l214421224411111 data;
>>> @@ -70,14 +71,14 @@
>>> &data, serialize_sizeof(data), sizeof(long));
>>>   }
>>>
>>> -void probe_tcpv4_rcv_extended(struct sk_buff *skb);
>>> +void probe_tcpv4_rcv_extended(void *_data, struct sk_buff *skb);
>>>
>>>   DEFINE_MARKER_TP(net, tcpv4_rcv_extended, net_tcpv4_rcv,
>>>          probe_tcpv4_rcv_extended, "skb 0x%lX saddr #n4u%lu daddr #n4u%lu "
>>>          "tot_len #n2u%hu ihl #1u%u source #n2u%hu dest #n2u%hu seq
>>> #n4u%lu "
>>>          "ack_seq #n4u%lu doff #1u%u ack #1u%u rst #1u%u syn #1u%u fin
>>> #1u%u");
>>>
>>> -notrace void probe_tcpv4_rcv_extended(struct sk_buff *skb)
>>> +notrace void probe_tcpv4_rcv_extended(void *_data, struct sk_buff *skb)
>>>   {
>>>          struct marker *marker;
>>>          struct serialize_l4421224411111 data;
>>> @@ -104,14 +105,14 @@
>>> &data, serialize_sizeof(data), sizeof(long));
>>>   }
>>>
>>> -void probe_udpv4_rcv_extended(struct sk_buff *skb);
>>> +void probe_udpv4_rcv_extended(void *_data, struct sk_buff *skb);
>>>
>>>   DEFINE_MARKER_TP(net, udpv4_rcv_extended, net_udpv4_rcv,
>>>          probe_udpv4_rcv_extended, "skb 0x%lX saddr #n4u%lu daddr #n4u%lu "
>>>          "unicast #1u%u ulen #n2u%hu source #n2u%hu dest #n2u%hu "
>>>          "data_start #8u%lx");
>>>
>>> -notrace void probe_udpv4_rcv_extended(struct sk_buff *skb)
>>> +notrace void probe_udpv4_rcv_extended(void *_data, struct sk_buff *skb)
>>>   {
>>>          struct marker *marker;
>>>          struct serialize_l4412228 data;
>>>
>>>
>>> I don't think the difference in net-extended-trace.c will make any impact.
>>> This problem has troubled me for weeks.
>>>        
>> Can you try to reproduce with the latest LTTng version ? It's very hard
>> for us to help you if we don't know the LTTng version you are using,
>> especially for a bug I am aware to have fixed recently.
>>      
> Ah, I remember it now. It had something to do with alignment of 64-bit
> fields on 32-bit machines. Older LTTng versions did align these fields
> on the next 64-bit multiple. New LTTng versions do the same as gcc and
> align these on the next 32-bit boundary.
>    
It make good sense, if align to 4-bytes, then the size of struct 
serialize_l4412228
will be 28, same as LTTV.
> Check the diff of your include/linux/ltt-core.h ltt_align() function
> between a recent lttng and yours.
>    
I diff my kernel ltt-core.h with patch-2.6.36-lttng-0.239, and found no 
different in
ltt_align() function.

--- ltt-core.h.kernel   2010-11-10 15:08:20.000000000 +0800
+++ ltt-core.h  2010-11-10 15:07:19.000000000 +0800
@@ -1,5 +1,5 @@
  /*
- * Copyright (C) 2005,2006 Mathieu Desnoyers (mathieu.desnoyers at polymtl.ca)
+ * Copyright (C) 2005-2010 Mathieu Desnoyers 
(mathieu.desnoyers at efficios.com)
   *
   * This contains the core definitions for the Linux Trace Toolkit.
   *
@@ -9,44 +9,10 @@
  #ifndef LTT_CORE_H
  #define LTT_CORE_H

-#include <linux/list.h>
-#include <linux/percpu.h>
-
-/* ltt's root dir in debugfs */
-#define LTT_ROOT        "ltt"
-
-/*
- * All modifications of ltt_traces must be done by ltt-tracer.c, while 
holding
- * the semaphore. Only reading of this information can be done 
elsewhere, with
- * the RCU mechanism : the preemption must be disabled while reading the
- * list.
- */
-struct ltt_traces {
-       struct list_head setup_head;    /* Pre-allocated traces list */
-       struct list_head head;          /* Allocated Traces list */
-       unsigned int num_active_traces; /* Number of active traces */
-} ____cacheline_aligned;
-
-extern struct ltt_traces ltt_traces;
-
-/*
- * get dentry of ltt's root dir
- */
-struct dentry *get_ltt_root(void);
-
-void put_ltt_root(void);
-
  /* Keep track of trap nesting inside LTT */
  DECLARE_PER_CPU(unsigned int, ltt_nesting);

-typedef int (*ltt_run_filter_functor)(void *trace, uint16_t eID);
-
-extern ltt_run_filter_functor ltt_run_filter;
-
-extern void ltt_filter_register(ltt_run_filter_functor func);
-extern void ltt_filter_unregister(void);
-
-#if defined(CONFIG_LTT) && defined(CONFIG_LTT_ALIGNMENT)
+#ifndef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS

  /*
   * Calculate the offset needed to align the type.
@@ -87,6 +53,6 @@
         return 0;
  }

-#endif /* defined(CONFIG_LTT) && defined(CONFIG_LTT_ALIGNMENT) */
+#endif /* HAVE_EFFICIENT_UNALIGNED_ACCESS */

  #endif /* LTT_CORE_H *


> Thanks,
>
> Mathieu
>
>    
>> Also, you might simply try to unload the net-extended-trace module (or
>> disable its markers) to see if it helps getting a parseable trace.
>>
>> Thanks,
>>
>> Mathieu
>>
>>      
>>> Thanks,
>>> Xufeng Zhang
>>>        
>>>> Thanks,
>>>>
>>>> Mathieu
>>>>
>>>>
>>>>
>>>>          
>>>>>>> aborting...
>>>>>>> /usr/local/bin/lttv-gui: line 10:  6559 Aborted
>>>>>>> $LTTV_CMD.real -m lttvwindow -m guievents -m guifilter -m guicontrolflow
>>>>>>> -m resourceview -m guistat
>>>>>>>
>>>>>>>
>>>>>>> And I found another people met the same problem on MIPS32, a patch have
>>>>>>> also been provided for MIPS32.
>>>>>>> Here is the similar code in ARM trace-clock.c:
>>>>>>> linux/arch/arm/mach-omap2/trace-clock.c line 537
>>>>>>> void get_trace_clock(void)
>>>>>>> {
>>>>>>>            spin_lock(&trace_clock_lock);
>>>>>>>            if (trace_clock_refcount++)
>>>>>>>                    goto end;
>>>>>>>            _start_trace_clock();
>>>>>>> end:
>>>>>>>            spin_unlock(&trace_clock_lock);
>>>>>>> }
>>>>>>> EXPORT_SYMBOL_GPL(get_trace_clock);
>>>>>>>
>>>>>>> void put_trace_clock(void)
>>>>>>> {
>>>>>>>            spin_lock(&trace_clock_lock);
>>>>>>>            WARN_ON(trace_clock_refcount<= 0);
>>>>>>>            if (trace_clock_refcount != 1)
>>>>>>>                    goto end;
>>>>>>>            _stop_trace_clock();
>>>>>>> end:
>>>>>>>            trace_clock_refcount--;
>>>>>>>            spin_unlock(&trace_clock_lock);
>>>>>>> }
>>>>>>> EXPORT_SYMBOL_GPL(put_trace_clock);
>>>>>>>
>>>>>>> Could anybody figure out what's wrong with the kernel code? Any
>>>>>>> suggestion would be appreciated.
>>>>>>>
>>>>>>>
>>>>>>> Thanks,
>>>>>>> Xufeng Zhang
>>>>>>>
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> ltt-dev mailing list
>>>>>>> ltt-dev at lists.casi.polymtl.ca
>>>>>>> http://lists.casi.polymtl.ca/cgi-bin/mailman/listinfo/ltt-dev
>>>>>>>
>>>>>>>
>>>>>>>                
>>>>>>
>>>>>>              
>>>>> _______________________________________________
>>>>> ltt-dev mailing list
>>>>> ltt-dev at lists.casi.polymtl.ca
>>>>> http://lists.casi.polymtl.ca/cgi-bin/mailman/listinfo/ltt-dev
>>>>>
>>>>>
>>>>>            
>>>>
>>>>          
>>>        
>> -- 
>> Mathieu Desnoyers
>> Operating System Efficiency R&D Consultant
>> EfficiOS Inc.
>> http://www.efficios.com
>>
>> _______________________________________________
>> ltt-dev mailing list
>> ltt-dev at lists.casi.polymtl.ca
>> http://lists.casi.polymtl.ca/cgi-bin/mailman/listinfo/ltt-dev
>>
>>      
>    





More information about the lttng-dev mailing list