Probe on function entry rather than the end of the prolog (s390x only)
David Wilder
dwilder@us.ibm.com
Wed Feb 14 02:25:00 GMT 2007
Frank Ch. Eigler wrote:
>
>I'm game for going prologue-free by default, agan subject to more
>testing. I will check out RHEL4 and RHEL5 on i386 and x86-64. If
>others can produce reports of the "installcheck" testsuite (and
>perhaps some hand-run eye-observed tests) for their own favorite
>platforms, we can flip the default.
>
>
It looks like you already flipped it:) Default mode is now probing at
the function entry.
>
>- FChE
>
>
I finished testing the prologue-free tests on s390x. The results are
attached.
I ran into an unrelated problem with the "conversions" test crashing my
system, but thats another story, it has nothing to do with the
prologue-free change. I disabled the conversion test for this run.
One additional modification. The syscall.mmap2 probe is giving a
semantic error on s390x. The arguments to sys_mmap2 are all in
userspace, the kernel is passed a pointer to a user structure thus the
current implementation of syscall.mmap2 wont work on s390x. I will fix
that another day :) The results are that all the syscall test fail, so
I commented out the offending code for this test run.
The results of the prologue-free for the s390x show no regressions with
the change and it fixes several of the tests.
Tests were made with the RHEL5 kernel. The first results are for the
old behavior of probing at the prologue end, the second are for probing
at function entry.
--
David Wilder
IBM Linux Technology Center
Beaverton, Oregon, USA
dwilder@us.ibm.com
(503)578-3789
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: systemtap.sum.prolog-end
URL: <http://sourceware.org/pipermail/systemtap/attachments/20070214/6a392db2/attachment.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: systemtap.sum.func-start
URL: <http://sourceware.org/pipermail/systemtap/attachments/20070214/6a392db2/attachment-0001.ksh>
More information about the Systemtap
mailing list