Hi Ali,
I want to record all memory reads/writes from a particular
application, running alongside others. I realized that even if I do get the
memory access info from physmem and check for user/kernel mode, it would be
missing accesses made to caches. (Is that right?). I hope its clear what I
am trying to do.
Is there another way where I can tap in into the cores itself, and dump
out any memory references made by a thread, on de-coding the relevant
instructions?
Thanks,
Sujay
----- Original Message -----
From: "Ali Saidi" <[email protected]>
To: "M5 users mailing list" <[email protected]>
Sent: Monday, August 17, 2009 9:57 PM
Subject: Re: [m5-users] memory trace and thread ID
> Well, it is probably the Threadinfo code mis-interpreting part of the
> structure. I don't know exactly what you want, but it's important to
> mention that the memory acccesses by the application are only a small
> part of a networking workload, there are also all the accesses the
> kernel does on behalf of the application.
>
> Ali
>
> On Aug 16, 2009, at 6:37 PM, Sujay Phadke wrote:
>
>> Ya I was surprised about the -ve start time too. Is is a bug? All of
>> them
>> appear that way. Thanks for clarifying about the processor mode. so
>> wrt your
>> last statement, I am confused. Will the check for user level mode
>> help me
>> with what I want? Do I want to dump all accesses between the start
>> and end
>> points of "m5-spec" thread, when the processor is in user level
>> mode? Or is
>> there a better way of doing this that you can think of?
>>
>> thanks,
>> Sujay
>>
>>
>> ----- Original Message -----
>> From: "Ali Saidi" <[email protected]>
>> To: "M5 users mailing list" <[email protected]>
>> Sent: Sunday, August 16, 2009 7:26 PM
>> Subject: Re: [m5-users] memory trace and thread ID
>>
>>
>>> The time started and pid don't appear to be correct. I'm pretty sure
>>> the maximum size a pid can be is 16 bits and the started time should
>>> be positive. At least you did get the thread names correctly out,
>>> which might be al that matters. The threadInfo event is printing
>>> every
>>> time the kernel changes the user application stack, however that
>>> doesn't mean that it's executing user level code. If you want a trace
>>> of user level memory operations you'll need to check if the processor
>>> is in user mode or kernel mode (there is a helper function in arch/
>>> alpha to do that). There isn't really any way to know which thread
>>> the
>>> kernel is doing a task for (e.g. if it receives a packet while the
>>> swapper is running, it's quite difficult to attribute that time to
>>> the
>>> application that was actually receiving that packet).
>>>
>>> Ali
>>>
>>>
>>>
>>> On Aug 16, 2009, at 6:08 PM, Sujay Phadke wrote:
>>>
>>>> Hi,
>>>> Thanks Ali. I am using a patched kernel now with the
>>>> modifications
>>>> you suggested. Running specweb in FS mode. What I want to do is
>>>> differentiate memory access of the m5-spec thread from all others
>>>> (like OS
>>>> ones). I get the following output:
>>>>
>>>> ***
>>>> 487306156500: threadinfo: Currently Executing Thread swapper, pid
>>>> 3211264,
>>>> started at: -4398043299840
>>>> 487306156500: testsys.physmem: Read of size 4 on address 0x3111a8
>>>> data
>>>> 0x6bfa8001
>>>> 487306157000: testsys.physmem: Read of size 4 on address 0x6b7c1c
>>>> data
>>>> 0x27ba001a
>>>> 487306157500: testsys.physmem: Read of size 4 on address 0x6b7c20
>>>> data
>>>> 0xa0680060
>>>> 487306157500: testsys.physmem: Read of size 4 on address 0x1f03c060
>>>> data 0x0
>>>> 487306158000: testsys.physmem: Read of size 4 on address 0x6b7c24
>>>> data
>>>> 0x23bd2124
>>>> 487306158500: testsys.physmem: Read of size 4 on address 0x6b7c28
>>>> data
>>>> 0x47e00409
>>>> 487306159000: testsys.physmem: Read of size 4 on address 0x6b7c2c
>>>> data
>>>> 0xa49da718
>>>> 487306159000: testsys.physmem: Read of size 8 on address 0x854458
>>>> data
>>>> 0xfffffc000085e340
>>>> ***
>>>> ***
>>>> 487306811000: threadinfo: Currently Executing Thread m5-spec, pid
>>>> 3211290,
>>>> started at: -4398043299840
>>>> 487306811000: testsys.physmem: Read of size 4 on address 0x3111a8
>>>> data
>>>> 0x6bfa8001
>>>> 487306811500: testsys.physmem: Read of size 4 on address 0x6b7c1c
>>>> data
>>>> 0x27ba001a
>>>> 487306812000: testsys.physmem: Read of size 4 on address 0x6b7c20
>>>> data
>>>> 0xa0680060
>>>> 487306812000: testsys.physmem: Read of size 4 on address 0x814060
>>>> data 0x0
>>>> 487306812500: testsys.physmem: Read of size 4 on address 0x6b7c24
>>>> data
>>>> 0x23bd2124
>>>> 487306813000: testsys.physmem: Read of size 4 on address 0x6b7c28
>>>> data
>>>> 0x47e00409
>>>> 487306813500: testsys.physmem: Read of size 4 on address 0x6b7c2c
>>>> data
>>>> 0xa49da718
>>>> 487306813500: testsys.physmem: Read of size 8 on address 0x854458
>>>> data
>>>> 0xfffffc000085e340
>>>> 487306814000: testsys.physmem: Read of size 4 on address 0x6b7c30
>>>> data 0x486
>>>>
>>>> --------
>>>>
>>>> So, if the OS thread swapper began at tick "487306156500", and the
>>>> m5-spec
>>>> begins at "487306811000" does it mean than all the memory accesses
>>>> between
>>>> these time points are that of "swapper"?. And all accesses after
>>>> 487306811000 are of m5-spec, till a new thread is executed? Or am I
>>>> wrong
>>>> here? If this is right, all I will have to do is right a script
>>>> which parses
>>>> this output inline and picks accesses which start when "m5-spec" is
>>>> started
>>>> and end at the next thread execution.
>>>>
>>>> Thanks,
>>>> Sujay
>>>>
>>>>
>>>> ----- Original Message -----
>>>> From: "Ali Saidi" <[email protected]>
>>>> To: "M5 users mailing list" <[email protected]>
>>>> Sent: Monday, August 10, 2009 5:20 PM
>>>> Subject: Re: [m5-users] memory trace and thread ID
>>>>
>>>>
>>>>>
>>>>> The image doesn't have a kernel in it. You get that separately. Why
>>>>> don't
>>>>> you sprinkle some DPRINTFN() calls around the Threading stuff
>>>>> (threadinfo.hh and the system.cc) and see where it's not working as
>>>>> it
>>>>> should? If it's from symbols not being found then no, otherwise the
>>>>> problem
>>>>> should be rather apparent.
>>>>>
>>>>> Ali
>>>>>
>>>>>
>>>>> On Mon, 10 Aug 2009 17:15:02 -0400, "Sujay Phadke"
>>>>> <[email protected]>
>>>>> wrote:
>>>>>> I am using the linux-latest image available from the M5 website.
>>>>>> Does
>>>>> that
>>>>>> need to be patched?
>>>>>>
>>>>>> ----- Original Message -----
>>>>>> From: "Ali Saidi" <[email protected]>
>>>>>> To: "M5 users mailing list" <[email protected]>
>>>>>> Sent: Monday, August 10, 2009 4:10 PM
>>>>>> Subject: Re: [m5-users] memory trace and thread ID
>>>>>>
>>>>>>
>>>>>>>
>>>>>>> The patches I'm referring to are in the linux-patches repository
>>>>>>> http://repo.m5sim.org/linux-patches. Depending on where you got
>>>>>>> your
>>>>>>> kernel
>>>>>>> from you might need to apply them.
>>>>>>>
>>>>>>> Ali
>>>>>>>
>>>>>>>
>>>>>>> On Mon, 10 Aug 2009 15:50:38 -0400, "Sujay Phadke"
>>>>>>> <[email protected]>
>>>>>>> wrote:
>>>>>>>> ----- Original Message -----
>>>>>>>> From: "Sujay Phadke" <[email protected]>
>>>>>>>> To: <[email protected]>
>>>>>>>> Sent: Monday, August 10, 2009 3:28 PM
>>>>>>>> Subject: Re: [m5-users] memory trace and thread ID
>>>>>>>>
>>>>>>>>
>>>>>>>>> I do have the traceflag on. But I dont have the patches
>>>>>>>>> applied. On the
>>>>>>> M5
>>>>>>>>>
>>>>>>>>> download page, the patches come under the section "Obsolete
>>>>>>>>> Revisions"
>>>>>>> and
>>>>>>>>>
>>>>>>>>> it says not required. I thought these were there in the m5-
>>>>>>>>> stable. Do I
>>>>>>>>> have to manually apply these?
>>>>>>>>>
>>>>>>>>> Thanks,
>>>>>>>>> Sujay
>>>>>>>>>
>>>>>>>>> ----- Original Message -----
>>>>>>>>> From: "Ali Saidi" <[email protected]>
>>>>>>>>> To: "Sujay Phadke" <[email protected]>
>>>>>>>>> Sent: Monday, August 10, 2009 2:04 PM
>>>>>>>>> Subject: Re: [m5-users] memory trace and thread ID
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Are you running with the Thread traceflag on? You must be
>>>>>>>>>> using a
>>>>>>> kernel
>>>>>>>>>> with the thread_info structure annotated correctly.
>>>>>>>>>> Specifically,
>>>>>>>>>> m5struct.diff from the linux-patches repository must have been
>>>>>>>>>> applied.
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Ali
>>>>>>>>>>
>>>>>>>>>> On Mon, 10 Aug 2009 13:40:05 -0400, "Sujay Phadke"
>>>>>>>>>> <[email protected]>
>>>>>>>>>> wrote:
>>>>>>>>>>> Thanks. I tried doing this, but this is all I get running
>>>>>>>>>>> specweb:
>>>>>>>>>>>
>>>>>>>>>>> 0: testsys.tsunami.io.rtc: Real-time clock set to Thu Jan 1
>>>>> 00:00:00
>>>>>>>>>> 2009
>>>>>>>>>>> 0: drivesys.tsunami.io.rtc: Real-time clock set to Thu Jan 1
>>>>>>>>>>> 00:00:00
>>>>>>>>>> 2009
>>>>>>>>>>> 4694261844000: drivesys.terminal: attach terminal 0
>>>>>>>>>>>
>>>>>>>>>>> nothing more in the file.
>>>>>>>>>>>
>>>>>>>>>>> Can any method in threadinfo.hh help?
>>>>>>>>>>>
>>>>>>>>>>> - Sujay
>>>>>>>>>>>
>>>>>>>>>>> ----- Original Message -----
>>>>>>>>>>> From: "Ali Saidi" <[email protected]>
>>>>>>>>>>> To: "M5 users mailing list" <[email protected]>
>>>>>>>>>>> Sent: Monday, August 10, 2009 12:48 PM
>>>>>>>>>>> Subject: Re: [m5-users] memory trace and thread ID
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> Remove the false condition from src/arch/alpha/linux/
>>>>>>>>>>>> system.cc:139
>>>>>>> and
>>>>>>>>>>>> Thread should work.
>>>>>>>>>>>>
>>>>>>>>>>>> Ali
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> On Mon, 10 Aug 2009 12:10:58 -0400, "Sujay Phadke"
>>>>>>>>>>>> <[email protected]>
>>>>>>>>>>>> wrote:
>>>>>>>>>>>>> Hello,
>>>>>>>>>>>>> I want to generate memory traces and associate them
>>>>>>>>>>>>> with the
>>>>>>>>>> thread
>>>>>>>>>>>> ID
>>>>>>>>>>>>> or process ID. Is there a way to do this? I am running
>>>>>>>>>>>>> m5 in
>>>>>>>>>>>>> FS
>>>>>>>>>>>>> mode
>>>>>>>>>>>>> and want to differentiate the memory accesses of a
>>>>>>>>>>>>> particular
>>>>>>>>>>>> process
>>>>>>>>>>>>> from those of others, including the OS. I looked into
>>>>>>>>>>>>> the
>>>>>>>>>>>>> "Thread"
>>>>>>>>>>>> and
>>>>>>>>>>>>> "ExecThread" traceflags but they dont help.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Thanks,
>>>>>>>>>>>>> Sujay
>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>> m5-users mailing list
>>>>>>>>>>>> [email protected]
>>>>>>>>>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> m5-users mailing list
>>>>>>>> [email protected]
>>>>>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>>>>>>> _______________________________________________
>>>>>>> m5-users mailing list
>>>>>>> [email protected]
>>>>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> m5-users mailing list
>>>>>> [email protected]
>>>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>>>>> _______________________________________________
>>>>> m5-users mailing list
>>>>> [email protected]
>>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>>>>>
>>>>
>>>> _______________________________________________
>>>> m5-users mailing list
>>>> [email protected]
>>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>>>>
>>>
>>> _______________________________________________
>>> m5-users mailing list
>>> [email protected]
>>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>>>
>>
>> _______________________________________________
>> m5-users mailing list
>> [email protected]
>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>>
>
> _______________________________________________
> m5-users mailing list
> [email protected]
> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
>
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
