perf_counter, x86: Set global control MSR correctly
authorYong Wang <yong.y.wang@linux.intel.com>
Wed, 24 Jun 2009 02:13:24 +0000 (10:13 +0800)
committerIngo Molnar <mingo@elte.hu>
Wed, 24 Jun 2009 08:51:24 +0000 (10:51 +0200)
Previous code made an assumption that the power on value of global
control MSR has enabled all fixed and general purpose counters properly.

However, this is not the case for certain Intel processors, such as
Atom - and it might also be firmware dependent.

Each enable bit in IA32_PERF_GLOBAL_CTRL is AND'ed with the
enable bits for all privilege levels in the respective IA32_PERFEVTSELx
or IA32_PERF_FIXED_CTR_CTRL MSRs to start/stop the counting of
respective counters. Counting is enabled if the AND'ed results is true;
counting is disabled when the result is false.

The end result is that all fixed counters are always disabled on Atom
processors because the assumption is just invalid.

Fix this by not initializing the ctrl-mask out of the global MSR,
but setting it to perf_counter_mask.

Reported-by: Stephane Eranian <eranian@googlemail.com>
Signed-off-by: Yong Wang <yong.y.wang@intel.com>
Cc: Arjan van de Ven <arjan@infradead.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
LKML-Reference: <20090624021324.GA2788@ywang-moblin2.bj.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
arch/x86/kernel/cpu/perf_counter.c

index 22eb3a1d4f9cd1620f9e521531b518f8b438c8a6..a310d19faca36ae8ef0fd1f25ad92145ae7f3ea9 100644 (file)
@@ -969,13 +969,6 @@ fixed_mode_idx(struct perf_counter *counter, struct hw_perf_counter *hwc)
        if (!x86_pmu.num_counters_fixed)
                return -1;
 
-       /*
-        * Quirk, IA32_FIXED_CTRs do not work on current Atom processors:
-        */
-       if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
-                                       boot_cpu_data.x86_model == 28)
-               return -1;
-
        event = hwc->config & ARCH_PERFMON_EVENT_MASK;
 
        if (unlikely(event == x86_pmu.event_map(PERF_COUNT_HW_INSTRUCTIONS)))
@@ -1428,8 +1421,6 @@ static int intel_pmu_init(void)
         */
        x86_pmu.num_counters_fixed      = max((int)edx.split.num_counters_fixed, 3);
 
-       rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, x86_pmu.intel_ctrl);
-
        /*
         * Install the hw-cache-events table:
         */
@@ -1514,6 +1505,7 @@ void __init init_hw_perf_counters(void)
 
        perf_counter_mask |=
                ((1LL << x86_pmu.num_counters_fixed)-1) << X86_PMC_IDX_FIXED;
+       x86_pmu.intel_ctrl = perf_counter_mask;
 
        perf_counters_lapic_init();
        register_die_notifier(&perf_counter_nmi_notifier);