Ah yes, I didn't think of SE vs FS.  You're right, SyscallVerbose only
works in SE right now.

Steve

On Tue, Feb 23, 2010 at 7:09 PM, Gabriel Michael Black
<[email protected]> wrote:
> SyscallVerbose only works in SE mode unless I'm thuroughly confused
> about something. It would be really cool and useful to make it work in
> FS mode, but then you'd have to figure out when a syscall was
> happening and scrape out the right arguments using the right
> translations (which may or may not be in the TLB), figure out what
> process you're dealing with, etc. I assume you're using FS because you
> say you're spending 50% of your time in the kernel.
>
> There are less ambitious ways to get at some of that information. You
> might want to look at the tracer classes like ExecTracer in
> exetrace.hh and exetrace.hh. You could also put a DPRINTF into the
> execute function of the call_pal instruction (line 852+ in
> arch/alpha/isa/decoder.isa).
>
> Gabe
>
> Quoting ef <[email protected]>:
>
>> I tried that previously, Maybe I didnt run things long enough but it never
>> displayed anything. Might be an error on my part
>>
>> On Tue, Feb 23, 2010 at 8:03 PM, Steve Reinhardt <[email protected]> wrote:
>>
>>> You can also try enabling the SyscallVerbose trace flag... that may or
>>> may not be more convenient depending on what you're doing.
>>>
>>> Steve
>>>
>>> On Tue, Feb 23, 2010 at 5:08 PM, ef <[email protected]> wrote:
>>> > systems call can be tracked using the stat.txt output
>>> > search for things such as:
>>> > system.cpu0.kern.syscall::73                        1      0.00%
>>>  0.01%
>>> > # number of syscalls executed
>>> >
>>> > these numbers can be traced searching from system call tables. using an
>>> > alpha cross compiler  or the linux kernel source code (or google you can
>>> > trace these calls.)
>>> > system call macros usually have the prefix "__NR_"
>>> >
>>> >
>>> > On Wed, Feb 10, 2010 at 6:07 PM, ef <[email protected]> wrote:
>>> >>
>>> >> Hello,
>>> >> I am running a program where I noticed that 50% of M5 execution time is
>>> >> spent in the kernel. Looking at the program, I cannot find reasons why
>>> this
>>> >> would be the case (4 threads, where there is very little communcation
>>> >> between threads). Anyone have any idea on how to trace the instruction
>>> >> callsys,and see what system calls are made?
>>> >>
>>> >> I see a huge amount of callsys, swpipl, rti instructions being executed!
>>> >>
>>> >>
>>> >> Thanks,
>>> >> EF
>>> >
>>> > _______________________________________________
>>> > 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

Reply via email to