On Fri, May 26, 2006 at 08:48:21PM +0530, Joseph Koshy wrote: > >But, I have a naive question. Should this be integrated in > >some way with the hwpmc work of Joseph Koshy since I think it > >could be useful for DTRace to get information from the CPU as > >well. > > Disclaimer: I've only just started reading about DTrace. > > There appear to be two ways to integrate hwpmc and DTrace: > > - Augment the D virtual machine with a primitive that can > read PMC values (e.g.- using RDPMC or RDMSR instructions > on x86 CPUs). Make this primitive available > to scripts for allocating and reading from PMCs (say > a "pmcread()" builtin function). > > This approach would work well with counting mode PMCs > (both process and system-mode counting PMCs) and would > allow PMCs to be read at arbitrary points of time. > > We'll need a way of allocating system-wide & process-mode > PMCs; this could be done in userland (in dtrace(8)). > > - hwpmc(4) can be augmented to be a 'DTrace provider' > allowing D scripts to be run, say when a PC sample is > recorded. Before modifying the virtual machine in FreeBSD's DTrace port, it would be best to discuss this in the "DTrace Community" <http://www.opensolaris.org/jive/forum.jspa?forumID=7> The Sun guys will guide us on which way they think is appropriate. -- John BirrellReceived on Fri May 26 2006 - 18:03:28 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:56 UTC