sysprof in 2.6.27 kernel

sysprof in 2.6.27 kernel

[Lukas Hejtmanek]

Hello,

is there any port of sysprof gui that cooperates with the current 2.6.27 kernel
sysprof tracer?

-- 
Lukáš Hejtmánek

Re: sysprof in 2.6.27 kernel

[Soeren Sandmann]

Lukas Hejtmanek <xhejtman@ics.muni.cz> writes:

> Hello,
> 
> is there any port of sysprof gui that cooperates with the current 2.6.27 kernel
> sysprof tracer?

Yes, the port is accessible from here:

    svn checkout http://svn.gnome.org/svn/sysprof/branches/ftrace-branch sysprof

It's somewhat rough still. If you try it out, I will be interested in
any feedback you have.


Thanks,
Soren

Re: sysprof in 2.6.27 kernel

[Lukas Hejtmanek]

On Thu, Aug 14, 2008 at 01:23:16AM +0200, Soeren Sandmann wrote:
> It's somewhat rough still. If you try it out, I will be interested in
> any feedback you have.

another feedback. I got No map at all processes even at those I got debug info
installed.

-- 
Lukáš Hejtmánek

Re: sysprof in 2.6.27 kernel

[Soeren Sandmann]

Lukas Hejtmanek <xhejtman@ics.muni.cz> writes:

> On Thu, Aug 14, 2008 at 01:23:16AM +0200, Soeren Sandmann wrote:
> > It's somewhat rough still. If you try it out, I will be interested in
> > any feedback you have.
> 
> another feedback. I got No map at all processes even at those I got debug info
> installed.

This may be because sysprof user space is not running as root, or it
may be because the kernel is generating bogus addresses. 

Generally, "No Map" means sysprof is seeing samples in memory regions
that are not in /proc/<pid>/maps



Soren

Re: sysprof in 2.6.27 kernel

[Soeren Sandmann]

Soeren Sandmann <sandmann@daimi.au.dk> writes:

> Lukas Hejtmanek <xhejtman@ics.muni.cz> writes:
> 
> > On Thu, Aug 14, 2008 at 01:23:16AM +0200, Soeren Sandmann wrote:
> > > It's somewhat rough still. If you try it out, I will be interested in
> > > any feedback you have.
> > 
> > another feedback. I got No map at all processes even at those I got debug info
> > installed.
> 
> This may be because sysprof user space is not running as root, or it
> may be because the kernel is generating bogus addresses. 
> 
> Generally, "No Map" means sysprof is seeing samples in memory regions
> that are not in /proc/<pid>/maps

Actually, I seem to remember one cause of this: the vdso code creates
an unusual stackframe where the pointer to the previous frame is
stored in an unusual place. I think the sysprof kernel code needs to
take this into account, or alternatively send sufficient information
to userspace that it can figure it out itself.


Soren

Re: sysprof in 2.6.27 kernel

[Lukas Hejtmanek]

On Sat, Aug 16, 2008 at 04:50:30AM +0200, Soeren Sandmann wrote:
> This may be because sysprof user space is not running as root, or it
> may be because the kernel is generating bogus addresses. 

if I do not run it as root, the collecor does not start at all.
 
> Generally, "No Map" means sysprof is seeing samples in memory regions
> that are not in /proc/<pid>/maps

yes, that could be because only null addresses are received.

-- 
Lukáš Hejtmánek

Re: sysprof in 2.6.27 kernel

[Lukas Hejtmanek]

On Thu, Aug 14, 2008 at 01:23:16AM +0200, Soeren Sandmann wrote:
> It's somewhat rough still. If you try it out, I will be interested in
> any feedback you have.

and another one. sysproc gui reads user addresses from the sysprof pipe with
null pointers. is this ok or is this a bug in sysprof kernel module?

-- 
Lukáš Hejtmánek

Re: sysprof in 2.6.27 kernel

[Soeren Sandmann]

Lukas Hejtmanek <xhejtman@ics.muni.cz> writes:

> On Thu, Aug 14, 2008 at 01:23:16AM +0200, Soeren Sandmann wrote:
> > It's somewhat rough still. If you try it out, I will be interested in
> > any feedback you have.
> 
> and another one. sysproc gui reads user addresses from the sysprof pipe with
> null pointers. is this ok or is this a bug in sysprof kernel module?

That is not ok, but it could happen if the process in question was not
compiled with framepointers. Or it could be a bug in the kernel code.


Soren

Re: sysprof in 2.6.27 kernel

[Lukas Hejtmanek]

On Sat, Aug 16, 2008 at 04:52:54AM +0200, Soeren Sandmann wrote:
> > and another one. sysproc gui reads user addresses from the sysprof pipe with
> > null pointers. is this ok or is this a bug in sysprof kernel module?
> 
> That is not ok, but it could happen if the process in question was not
> compiled with framepointers. Or it could be a bug in the kernel code.

I tried to put printf in case that sysprof gui receives non null address from
the kernel. None print was displayed, so it seems ti receives only null
addresses.

-- 
Lukáš Hejtmánek