https://bugs.kde.org/show_bug.cgi?id=126348
--- Comment #5 from Vince Weaver <vincent.wea...@maine.edu> --- (In reply to Julian Seward from comment #4) > IIUC (and correct me if I am wrong): rdpmc support is something that > is either enabled or disabled at a process granularity (or larger), but > doesn't change whilst a process runs. It's a bit more complicated than that :( If you try to do a rdpmc and rdpmc is disabled in the CR4 register, you actually get a GPF: [419335.853268] traps: rdpmc_invalid[31138] general protection fault ip:56259dd2d056 sp:7fffed404ac0 error:0 in rdpmc_invalid[56259dd2d000+1000] The CR4 register status can change while the process is running. There's actually 3 states (I think) that happen on Linux, and this is configurable via /sys/devices/cpu/rdpmc 0. rdpmc disabled for everyone 1. rdpmc only enabled if a perf_event_open() enabled hardware event is active 2. rdpmc globally enabled for everyone I think "1" is currently the default on modern Linux kernels. With this setting, the CR4 bit value might change while the program is running, it is enabled at perf_event_open() and disabled after close() of that event. -- You are receiving this mail because: You are watching all bug changes.
