Messages in this thread | | | Date | Wed, 26 Nov 2008 14:38:37 +0100 | From | "stephane eranian" <> | Subject | Re: [patch 00/24] perfmon: introduction |
| |
Hello Paul,
On Wed, Nov 26, 2008 at 11:05 AM, Paul Mackerras <paulus@samba.org> wrote: > eranian@googlemail.com writes: > >> This new version, named perfmon3, uses only 5 system calls (instead of 12). >> Each call was carefully designed to allow for future extensions. > > I notice that the userspace stuff in CVS assumes perfmon3 has 7 system > calls, the 5 added here plus pfm_create_sets and pfm_getinfo_sets. > Are you planning to add those two later (in which case we should > reserve the numbers now), or are you going to implement the > functionality of those two in pfm_write and pfm_read? >
Yes, the LKML patchset represents the very minimal functionality of perfmon3. Libpfm/pfmon are coded to support the fully-featured version to help with testing and evaluations.
The 2 systems calls you are referring to go with event set and multiplexing, an advanced feature which I intend to submit to LKML later. That's why I did not add those two calls immediately. People would have argued that the calls are not used.
Furthermore, there is still some discussions as to how best to add the set functionality to the syscall API. We need to be able to: - create new sets - retrieve runtime execution about sets, e.g., number of activations, timeout left - and possibility delete sets
I don't think we can reserve syscall numbers in advance. OTOH, the perfmon3 syscalls don't necessarily need to be contiguous, unless people tell me this is a rule used by the kernel developers.
| |