[PATCH AUTOSEL 5.4 24/28] pm: cpupower: Fix the snapshot-order of tsc,mperf, clock in mperf_stop()

[Sasha Levin <sashal@kernel.org>]

From: "Gautham R. Shenoy" <gautham.shenoy@amd.com>

[ Upstream commit cda7ac8ce7de84cf32a3871ba5f318aa3b79381e ]

In the function mperf_start(), mperf_monitor snapshots the time, tsc
and finally the aperf,mperf MSRs. However, this order of snapshotting
in is reversed in mperf_stop(). As a result, the C0 residency (which
is computed as delta_mperf * 100 / delta_tsc) is under-reported on
CPUs that is 100% busy.

Fix this by snapshotting time, tsc and then aperf,mperf in
mperf_stop() in the same order as in mperf_start().

Link: https://lore.kernel.org/r/20250612122355.19629-2-gautham.shenoy@amd.com
Signed-off-by: Gautham R. Shenoy <gautham.shenoy@amd.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---

LLM Generated explanations, may be completely bogus:

**Backport Status: YES**

This commit should be backported to stable kernel trees for the
following reasons:

1. **Clear bug fix with user impact**: The commit fixes a calculation
   bug in the cpupower tool where C0 residency (processor active time)
   is under-reported on 100% busy CPUs. This directly affects users
   monitoring CPU performance and power states.

2. **Simple and contained fix**: The change is minimal - it only
   reorders three function calls in `mperf_stop()` to match the order in
   `mperf_start()`:
   - Before: `mperf_measure_stats()` → `mperf_get_tsc()` →
     `clock_gettime()`
   - After: `clock_gettime()` → `mperf_get_tsc()` →
     `mperf_measure_stats()`

3. **Low risk of regression**: The fix simply ensures consistent
   snapshot ordering between start and stop measurements. This is a
   logical correction that aligns the measurement methodology without
   introducing new functionality.

4. **Clear root cause**: The bug occurs because the reversed snapshot
   order causes timing skew in the C0 residency calculation
   (`delta_mperf * 100 / delta_tsc`). When snapshots are taken in
   different orders at start vs stop, the time windows don't align
   properly, leading to incorrect calculations.

5. **Affects monitoring tool accuracy**: While this is a userspace tool
   bug rather than a kernel bug, cpupower is an important diagnostic
   tool shipped with the kernel. Incorrect C0 residency reporting could
   mislead system administrators and developers analyzing CPU
   performance.

6. **No architectural changes**: The fix doesn't change any data
   structures, APIs, or fundamental behavior - it simply corrects the
   measurement sequence.

The commit meets stable tree criteria as it fixes a real bug affecting
users, is minimal in scope, and has very low risk of introducing new
issues.

 tools/power/cpupower/utils/idle_monitor/mperf_monitor.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tools/power/cpupower/utils/idle_monitor/mperf_monitor.c b/tools/power/cpupower/utils/idle_monitor/mperf_monitor.c
index 0e8c0b879d4b..09c606760464 100644
--- a/tools/power/cpupower/utils/idle_monitor/mperf_monitor.c
+++ b/tools/power/cpupower/utils/idle_monitor/mperf_monitor.c
@@ -201,9 +201,9 @@ static int mperf_stop(void)
 	int cpu;
 
 	for (cpu = 0; cpu < cpu_count; cpu++) {
-		mperf_measure_stats(cpu);
-		mperf_get_tsc(&tsc_at_measure_end[cpu]);
 		clock_gettime(CLOCK_REALTIME, &time_end[cpu]);
+		mperf_get_tsc(&tsc_at_measure_end[cpu]);
+		mperf_measure_stats(cpu);
 	}
 
 	return 0;
-- 
2.39.5