Re: Arm Fedora 18 systemtap test results

[William Cohen <wcohen at redhat dot com>]

On 11/26/2012 12:39 PM, David Smith wrote:
> On 11/23/2012 06:25 AM, Frank Ch. Eigler wrote:
>>
>> Hi, Will -
>>
>> wcohen wrote:
>>
>>> I have a Fedora 18 arm environment set up and ran the systemtap tests
>>> on the git repo and posted them to dejazilla: [...]
>>
>> Thank you.
>>
>>
>>> probed functions not available or found in the kernel
>>> FAIL: ./systemtap.examples/network/tcpipstat build (ip_route_input inlined)
>>> FAIL: buildok/ipmib-all-probes.stp (ip_route_input inlined)
>>> FAIL: buildok/ipmib-detailed.stp (ip_route_input inlined)
>>> FAIL: buildok/nfs-all-probes.stp (nfs.proc3.commit_done)
>>> FAIL: buildok/syscalls-arch-detailed.stp (arm missing syscall.sigaltstack) **
>>> FAIL: buildok/eighteen.stp (can't find kernel.function(**audit_getname")
>>
>> Could you fix these, but changing the testsuite to (also?) look for
>> something that does exist on these ARM kernels?

The tcipstat and ipmib*.stp tests failures are due to the tapset probing the ip_route_input function which is inlined on arm.

I will take a look and see what fixes can be done to have the other above tests.

>>
>>
>>> flagged as needing
>>> user-space facilities not available without kernel CONFIG_UTRACE or CONFIG_TRACEPOINTS/CONFIG_ARCH_SUPPORTS_UPROBES/CONFIG_UPROBES
>>>
>>> FAIL: bad-code
>>> FAIL: poll_map (0)
>>> FAIL: compiling pr10854.stp: user-space facilities not available without kernel CONFIG_UTRACE or CONFIG_TRACEPOINTS/CONFIG_ARCH_SUPPORTS_UPROBES/CONFIG_UPROBES
>>> FAIL: probefunc:.statement.(0xaddr).absolute startup (eof)
>>> FAIL: probefunc:kernel.function("scheduler_tick") startup (eof)
>>> FAIL: probefunc:kernel.function("context_switch").inline startup (eof)
>>> FAIL: utf_pretty
>>> FAIL: utf_user_trunc
>>> FAIL: ./systemtap.examples/general/para-callgraph-verbose build
>>> FAIL: ./systemtap.examples/general/para-callgraph build
>>> FAIL: ./systemtap.examples/io/inodewatch build
>>> FAIL: ./systemtap.examples/io/inodewatch2 build
>>> FAIL: ./systemtap.examples/io/traceio2 build
>>> FAIL: ./systemtap.examples/network/socket-trace build
>>> FAIL: ./systemtap.examples/process/auditbt build
>>> FAIL: ./systemtap.examples/profiling/fntimes build
>>> FAIL: ./systemtap.examples/profiling/functioncallcount build
>>> FAIL: buildok/context-symbols-embedded.stp
>>> FAIL: buildok/memory.stp
>>> FAIL: buildok/nd_syscalls-all-probes.stp
>>> FAIL: buildok/nd_syscalls2-detailed.stp
>>> FAIL: buildok/scsi-detailed.stp
>>> FAIL: 32-bit access nd_syscall
>>> FAIL: 32-bit acct nd_syscall
>>> FAIL: 32-bit alarm nd_syscall
>>> FAIL: 32-bit chmod nd_syscall
>>> FAIL: 32-bit clock nd_syscall
>>> FAIL: 32-bit dir nd_syscall
>>> FAIL: 32-bit dup nd_syscall
>>> FAIL: 32-bit eventfd nd_syscall
>>> FAIL: 32-bit forkwait nd_syscall
>>> FAIL: 32-bit futimes nd_syscall
>>> FAIL: 32-bit inotify nd_syscall
>>> FAIL: 32-bit itimer nd_syscall
>>> FAIL: 32-bit link nd_syscall
>>> FAIL: 32-bit mmap nd_syscall
>>> FAIL: 32-bit mount nd_syscall
>>> FAIL: 32-bit net1 nd_syscall
>>> FAIL: 32-bit openclose nd_syscall
>>> FAIL: 32-bit pipe nd_syscall
>>> FAIL: 32-bit poll nd_syscall
>>> FAIL: 32-bit readwrite nd_syscall
>>> FAIL: 32-bit rt_signal nd_syscall
>>> FAIL: 32-bit select nd_syscall
>>> FAIL: 32-bit sendfile nd_syscall
>>> FAIL: 32-bit signal nd_syscall
>>> FAIL: 32-bit signalfd nd_syscall
>>> FAIL: 32-bit stat nd_syscall
>>> FAIL: 32-bit statfs nd_syscall
>>> FAIL: 32-bit swap nd_syscall
>>> FAIL: 32-bit sync nd_syscall
>>> FAIL: 32-bit timer nd_syscall
>>> FAIL: 32-bit trunc nd_syscall
>>> FAIL: 32-bit uid nd_syscall
>>> FAIL: 32-bit umask nd_syscall
>>> FAIL: 32-bit unlink nd_syscall
>>> FAIL: 32-bit alarm syscall
>>> FAIL: 32-bit stat syscall
>>
>> These should more or less work; could you dig a bit deeper as to why
>> they don't?
> 
> As far as the nd_syscall tests go, I'll bet a good number of those
> failures go back to the ARM version of  _stp_arg() only supporting args
> 1-4 (http://sourceware.org/bugzilla/show_bug.cgi?id=14325).

It doesn't even get to run the instrumentation generated from nd_sys.stp. It fails building instrumentation.  The test ends up with errors about task finder function missing for "stap -v nd_sys.stp":

Pass 3: translated to C into "/tmp/stapY57v9M/stap_f864ec247f7b57968d6b57f907c4d848_371100_src.c" using 112488virt/19284res/3440shr/16352data kb, in 880usr/20sys/902real ms.
/tmp/stapY57v9M/stap_f864ec247f7b57968d6b57f907c4d848_371100_src.c: In function ‘systemtap_module_init’:
/tmp/stapY57v9M/stap_f864ec247f7b57968d6b57f907c4d848_371100_src.c:99486:3: error: implicit declaration of function ‘_stp_vma_init’ [-Werror=implicit-function-declaration]
/tmp/stapY57v9M/stap_f864ec247f7b57968d6b57f907c4d848_371100_src.c:99490:5: error: implicit declaration of function ‘stap_start_task_finder’ [-Werror=implicit-function-declaration]
/tmp/stapY57v9M/stap_f864ec247f7b57968d6b57f907c4d848_371100_src.c:99492:7: error: implicit declaration of function ‘stap_stop_task_finder’ [-Werror=implicit-function-declaration]
/tmp/stapY57v9M/stap_f864ec247f7b57968d6b57f907c4d848_371100_src.c:99577:5: error: implicit declaration of function ‘stap_task_finder_post_init’ [-Werror=implicit-function-declaration]
cc1: all warnings being treated as errors

It looks like this is something related to a pragma:vma in tapsets.   The particular kernel for arm doesn't have a uprobes support in it.



> 
> The 2 syscall failures could use some investigation.
> 

The alarm syscall and stat syscall shouldn't be lumped in with the nd_syscall failures.  Some of the output for the tests did not match.

alarm syscall didn't match alarm, pause, and nanosleep syscalls. alarm.c special cases some matches for the ia64.  looks like need to do sometime similar for __arm__.  for alarm() calls.  I checked in a fix for this.

stat syscall fails for a similar string mismatch and a patch has been pushed to fix this.

-Will