Hello,
Dan found a problem with the newly released Nehalem code in libpfm. There was an issue with pfm_get_event_counters() when passing the event index for UNC_CLK_UNHALTED. I fixed that in CVS and uploaded a new tarball. Because of SF.net policy, it is not possible to use the same name for the tarball. So I called it libpfm-3.7.1.tar.gz but it expands in libpfm-3.7 like before. Thanks Dan. Sorry for the inconvenience. On Thu, Jan 22, 2009 at 12:26 AM, stephane eranian <eran...@googlemail.com> wrote: > Hello, > > I have released the new perfmon2 kernel patch for 2.6.28. There > are also new version for libpfm, and pfmon. > > Once again a lot of work went into this release both at the kernel > and user levels. I spent quite some time on pfmon which needed > some refreshing and lots of bug fixing. > > The major new kernel features: > > * Intel Nehalem support, including uncore PMU, both counting and > sampling. Nehalem PEBS support is delayed until 2.6.29 due to > incoming kernel internal changes to the way the DS_AREA is managed. > > > * bug fixing in AMD64 Northbridge event restriction implementation. > Event codes 0xe0-0xf9 are NB events and can only be measured > once per socket (at least that's one way of doing it). > > * improved hotplug support for AMD64 to deal with sparse phys_proc_id. > > * includes a patch to implement read-modify-write on debugctlmsr useful > for LBR and uncore PMU management > > * Power6 updates by Corey Ashford > > Note that the same level of features has also been integrated into the > perfmon3 > GIT tree but no patch is provided. > > The major features in pfmon: > > * tons of bug fixes especially with the -follow-* and triggers > > * change data structures layout to minimize memory consumption. Was > consuming > several Gigabytes when sampling a kernel compile, now down to 100MB or so. > Structures were split and shared whenever possible. > > * sampling sessions can be close right when they finish using the > --smpl-eager-save > option. Default is to wait until ALL sessions have ended. > > * use epoll() to workaround the fd_set limitations of select() > > * simplified and improved symbol management > > * Intel Nehalem support for counting and sampling with core and uncore PMU. > Pfmon attempts to print helpful error mesage when dealing with NHM event > restrictions! > > * Powerpc updates by Chris Arges > > The major new features for libpfm: > > * Intel Nehalem support > > * added support for ANYTHR for fixed counters on Intel Atom and > architectural perfmon v3 > > As usual, you can grab everything from the Sourceforge web site at: > > http://perfmon2.sf.net > > Thanks. > ------------------------------------------------------------------------------ This SF.net email is sponsored by: SourcForge Community SourceForge wants to tell your story. http://p.sf.net/sfu/sf-spreadtheword _______________________________________________ perfmon2-devel mailing list perfmon2-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/perfmon2-devel
