Changes committed to git://oss.sgi.com/pcp/pcp.git dev
src/pmstat/pmstat.c | 4 ++--
src/python/pcp.py | 8 ++++----
2 files changed, 6 insertions(+), 6 deletions(-)
commit 187140f124c77f06154ff9f26ad433f91c550500
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date: Fri Nov 9 17:58:39 2012 +1100
Fix pmstat handling of odd corner case in CPU metrics
Came across a machine which was exhibiting a kernel bug,
possibly in the Radeon driver, such that all CPU time was
being consumed in the kernel. Its likely interrupts were
disabled for extended periods too, as on a 1sec sampling
interval, pmstat would often observe no change in the CPU
time metrics.
When this unusual situation arises, pmstat attempts to
divide by zero and takes sigfpe - guard against this, and
handle it as if no values were observed (which is fairly
close to true).
commit a50212140dea611cb92b8626549f1140ae80c393
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date: Fri Nov 9 16:10:46 2012 +1100
Correct the python ctype used for pmAtomValue 32bit integers
The Red Hat DTS testing folk have uncovered a bug using PCP
QA test 707, in the python PMAPI wrapper, which results in a
sigsegv in the test_pcp.python code.
It turns out we've been using the wrong sized ctype for the
32bit integer values, and they ended up being assigned into
64bits by accident. This was harmless on (LE) x86_64 but on
powerpc64 (BE), life became interesting. Small numbers like
32 (e.g. hinv.ncpus) became very big numbers, and we end up
walking off the end of pythons address space when iterating
over each CPU.
Red Hat bz #873751.
|