Current systemtap testsuite results

Current systemtap testsuite results

[William Cohen]

I tried out running the testsuite we have checked into the systemtap cvs 
on a variety of platforms with the current snapshot of systemtap 
(9/7/2005) running on each. The results were the worst on the RHEL4 
x86-64 machine. Below are more summaries of passes and fails and some 
analsysis on why things failed.  The kmodule.stp test fails for 
everything because the kernels do not have the required debug 
information for the modules.

-Will



RHEL4 x86-64 (calfee)
2.6.9-15.ELsmp

                 === systemtap Summary ===

# of expected passes            67
# of unexpected failures        13

FAIL: ./systemtap.base/and.stp startup (eof)
FAIL: ./systemtap.base/kmodule.stp compilation
FAIL: ./systemtap.base/logical_not.stp startup (eof)
FAIL: ./systemtap.base/or.stp startup (eof)
FAIL: ./systemtap.base/simple.stp startup (timeout)
FAIL: ./systemtap.base/subtract.stp startup (timeout)
FAIL: ./systemtap.base/tri.stp startup (timeout)
FAIL: ./systemtap.base/xor.stp startup (timeout)
FAIL: arith
FAIL: primes
FAIL: profile
FAIL: syscalls-count (0)
FAIL: syscalls-run (0)


On RHEL4 x86-64 the and.stp, logical_not.stp, or.stp failed in the same 
manner when an attempt to install the module was made. It appears that 
the names are not as unique as desired andthat the x86-64 is speeds 
enough to end up with the same name for two different tests.

insmod: error inserting '/tmp/staphXBwpp/stap_0_1126103660.ko': -1 File 
exists

systemtap.samples/arith.stp and systemtap.samples/profile.stp fail due 
to "implicit declaration of function `task_pt_regs'". Looks like the 
runtime library is not providing that.

systemtap.samples/syscalls.stp can crash the x86-64 kernel.

It looks like sometime tripped up the dejagnu test harness, so tests 
after simple.stp were not installed/running.





RHEL4 i686 (slingshot)
2.6.9-15.ELsmp


                 === systemtap Summary ===

# of expected passes            92
# of unexpected failures        1

FAIL: ./systemtap.base/kmodule.stp compilation


FC4 i686 (montague)
2.6.12-1.1447_FC4

                 === systemtap Summary ===

# of expected passes            90
# of unexpected failures        3

FAIL: ./systemtap.base/kmodule.stp compilation
FAIL: profile
FAIL: syscalls-run (14)


rawhide (trek)
2.6.13-1.1530_FC5smp

                 === systemtap Summary ===

# of expected passes            86
# of unexpected failures        5

FAIL: profile (timeout)
FAIL: profile
FAIL: syscalls-count (13)
FAIL: syscalls-run (timeout)
FAIL: syscalls-run (0)

Re: Current systemtap testsuite results

[Frank Ch. Eigler]

wcohen wrote:

> [...]
> RHEL4 x86-64 (calfee)
> 2.6.9-15.ELsmp

FWIW this is two kernel revisions too old.

> [...]
> On RHEL4 x86-64 the and.stp, logical_not.stp, or.stp failed in the
> same manner when an attempt to install the module was made. [...]
> insmod: error inserting '/tmp/staphXBwpp/stap_0_1126103660.ko': -1

I'll switch to a time-insensitive algorithm for naming those modules.

> [...] systemtap.samples/arith.stp and systemtap.samples/profile.stp
> fail due to "implicit declaration of function `task_pt_regs'".

task_pt_regs is an i386-only macro on at least one kernel I see.
There are references to this macro in src/tapsets.cxx associated with
the kernel.jiffies probe point family.  I'll add an i386-sensitivity
test until a more general way is found.

> [...] systemtap.samples/syscalls.stp can crash the x86-64 kernel.

I guess all in all these imperfect runs are a good way to counteract
my euphoria last night when all those tests and more ran beautifully
on my RHEL4 i686 laptop.  Thanks, dude. :-)


- FChE