Hi Lin Wang,
on the screen shot I see that only the count for raw event 0x461 is zero. Event 0x461 is UNC_DRAM_PAGE_CLOSE.CH2 and not UNC_DRAM_PAGE_CLOSE.CH0. Instead you have to use event 0x161 to monitor UNC_DRAM_PAGE_CLOSE.CH0.
Hello Roman Dementiev,
Thanks for your reply, I upload a new screen shot. I can't get 0x161,0x261,0x461,0x162,0x262,0x462.
I can get dram open event, but i can't get dram page close event and dram page miss event.
I don't know why？
I had no trouble reading these events on my Intel Xeon X5680 using utilities from msrtools-1.2 to read and write the uncore performance monitor MSRs. I did not think that "perf" had any provisions for accessing the MSR-based uncore counters?
Hi Lin Wang,
the -e rXXX format is to collect statistics for *on-core* raw PMU events.But you would like to monitor events from *uncore* PMU. I need to check on the status of perf uncore support in Linux. There was a recent development in that direction. In the meantime, could you send the Linux kernel version you are using ("uname -a" command).
The initial support for Xeon E5620 (and other processors with microarchitecture codenamed Nehalem-EP, Westmere-EP and SandyBridge-EP) uncore events is included into the Linux kernel mainline starting from version 3.5