When you run the experiment, a program named er_archive runs at termination,
and, by default in er_kernel, copies all shared objects (kernel modules),
into the experiment, so that the experiment can be moved. It also
generates an index file, listing the address ranges of all functions in
the module. That is enough data so that the function data is correct
even when you move the experiment.
When you ask for disassembly, however, the Analyzer needs to find the
instruction stream to annotate. I'm not sure, but it may need to
find the .o's. I'm not sure about this warning, but it may arise
because er_archive copies the archives, but doesn't update the date
on the experiment, so you get that warning. If so, that's a bug. It
probably should be fixed by having er_kernel update the experiment
date one more time after er_archive runs.
But I don't know why you say the disassembly is wrong. What
are you looking at that has the "correct" disassembly? Can you
send us a pointer to the experiment that shows this problem, and
the name of the machine on which it shows up?
*Vladimir* -- can you please pursue this?
Thanks,
Marty
Roch Bourbonnais wrote On 07/14/07 08:30,:
>Never seen that. The archive is normally produce when you gather data
>and it contains
>a copy of all kernel modules. Sometimes if the archive is not present
>you can generate it
>using er_archive.
>
>If you move a .er directory between 2 system you might want to check
>the archive is present
>(each kernel module generates 2 files in there and I don't know much
>more than that about them).
>If the module is not in the archive I think the tool will use the
>module on the system running analyzer
>and that the disassembly might well be wrong then.
>
>I'll be off-net for 2 weeks, so analyzer-team can pick this up.
>
>-r
>
>Le 14 juil. 07 ? 16:00, Cathy Zhou a ?crit :
>
>
>
>>Hi Roch,
>>
>>[ You are the person I think who might know the answer. If not, do
>>you know anyone who would know? ]
>>
>>When I am using the tool analyzer (the one in /ws/onnv-tools/
>>SUNWspro/SS11/bin, and clicks the disaccembly tab to see the
>>disassembly code, it always pops up a warning:
>>
>>"Warning! Object file `.../archives/...` is newer than the
>>experiment data."
>>
>>If I ignore that warning, the assembly code being shown in that
>>window doesn't seem to "match" the really asm code of the function.
>>
>>Do you know what would be the problem? Would it affect other data
>>shown in the analyzer tool (e.g, the kcpu cycles statistics)?
>>
>>Thanks
>>- Cathy
>>
>
>
>
>
