Re: DTrace userland project

On Wed, Sep 1, 2010 at 12:57, Rui Paulo <rpaulo_at_freebsd.org> wrote:
> On 1 Sep 2010, at 11:20, Marius Nünnerich wrote:
>
>> On Wed, Sep 1, 2010 at 00:21, Rui Paulo <rpaulo_at_freebsd.org> wrote:
>>> Hi,
>>> The DTrace userland project is near completion and you can start using parts of it right away (only on FreeBSD HEAD right now).
>>>
>>> For more information on how to use DTrace with userland, please read:
>>> http://wiki.freebsd.org/DTrace/userland
>>
>> Hi,
>>
>> I set this up in a VirtualBox with 9-current r212073 amd64, 4-cores,
>> amd-v and nested paging enabled.
>> I tried
>> dtruss /usr/bin/true
>> it works but is really slow, it takes ~10s to switch vt.
>
> This is not really using the new userland dtrace code, but the old one. I suspect DTrace has a problem running on your system. Can you try a different VM software or real hardware? I never experienced this problem on VMWare Fusion.

I have no other VM or real hardware available at the moment.

>> I tried the pid example as well, if I execute it on ttyv0 I can see lots of
>> kernel trap 12 with interrupts disabled
>> messages flying by and inbetween the desired _sleep:entry probes. When
>> I execute that script on ttyv1 I obviously don't see the kernel trap
>> messages and the machine kindof locks up (no output at all). I can do
>> a clean ACPI shutdown though.
>
> This looks like I race that I haven't fixed yet. Either way, if the VM is interfering, then it will be hard to find out how to fix this.

It is reproducible every time. But I too have no idea how to tackle this.