setjmp test case

setjmp test case

[Roland McGrath]

I've committed the test case for glibc setjmp/longjmp probes.  It succeeds
as "untested" if the probes are missing from your installed libraries.

On a Fedora 15ish system with the scratch glibc I posted about here earlier
installed, it looks like it's actually working right.  But setjmp.exp fails
with:

line 1: expected "hit longjmp_target"
Got "Certificate /root/systemtap/testsuite/.systemtap-root/ssl/server/stap.cert added to database ../../../../etc/systemtap/staprun"

That seems like it would break some other stuff too, and it's rather
disconcerting to see it doing anything like that when all I'm using
is pure local stap stuff and want no client/server anything at all.

Note that after today I won't be on this list or at this address.
After this afternoon, contact me at roland@hack.frob.com if necessary.


Thanks,
Roland

Re: setjmp test case

[Dave Brolley]

This comes from the runtime/uprobes/Makefile. It would appear that your 
test case required the building of uprobes.ko. You may have seen 
discussion about why we sign this module and that is why the certificate 
is generated.

Perhaps I should make these options run silently, except for errors.

Dave

On 03/11/2011 01:56 PM, Roland McGrath wrote:
> I've committed the test case for glibc setjmp/longjmp probes.  It succeeds
> as "untested" if the probes are missing from your installed libraries.
>
> On a Fedora 15ish system with the scratch glibc I posted about here earlier
> installed, it looks like it's actually working right.  But setjmp.exp fails
> with:
>
> line 1: expected "hit longjmp_target"
> Got "Certificate /root/systemtap/testsuite/.systemtap-root/ssl/server/stap.cert added to database ../../../../etc/systemtap/staprun"
>
> That seems like it would break some other stuff too, and it's rather
> disconcerting to see it doing anything like that when all I'm using
> is pure local stap stuff and want no client/server anything at all.
>
> Note that after today I won't be on this list or at this address.
> After this afternoon, contact me at roland@hack.frob.com if necessary.
>
>
> Thanks,
> Roland

Re: setjmp test case

[Roland McGrath]

> This comes from the runtime/uprobes/Makefile. It would appear that your 
> test case required the building of uprobes.ko. You may have seen 
> discussion about why we sign this module and that is why the certificate 
> is generated.

I see.  I was surprised when I got it again on every run.
It seemed like it would be cached after the first one.
But perhaps it's stored someplace that the testsuite wipes out,
which is probably the right thing for it to do all things considered.

> Perhaps I should make these options run silently, except for errors.

Probably.  Or at least the testsuite infrastructure should do something to
suppress them.  I still find it slightly suspect that all this certificate
nonsense happens when I'm never even going to consider using the server.


Thanks,
Roland

Re: setjmp test case

[Dave Brolley]

On 03/11/2011 02:32 PM, Roland McGrath wrote:
> I still find it slightly suspect that all this certificate
> nonsense happens when I'm never even going to consider using the server.
>
The rationale is that someone may give an unprivileged user a pre-built 
module which requires the already-built uprobes.ko. The unprivileged 
user cannot load uprobes.ko unless it is signed.

Dave