perf/arm-cmn: Optimise DTM counter reads

When multiple nodes of the same type are connected to the same XP
(particularly in CAL configurations), it seems that they are likely
to be consecutive in logical ID. Therefore, we're likely to gain a
small benefit from an easy tweak to optimise out consecutive reads
of the same set of DTM counters for an aggregated event.

Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/7777d77c2df17693cd3dabb6e268906e15238d82.1638530442.git.robin.murphy@arm.com
Signed-off-by: Will Deacon <will@kernel.org>

diff --git a/drivers/perf/arm-cmn.c b/drivers/perf/arm-cmn.c
index 8b98ca9666d014d5a3ffaf7b4271db6d6dc152c6..005a0d83bcac4f5d6d8ddd33afb19ef3d0fa9b13 100644 (file)
--- a/drivers/perf/arm-cmn.c
+++ b/drivers/perf/arm-cmn.c
@@ -690,18 +690,19 @@ static void arm_cmn_pmu_disable(struct pmu *pmu)
 static u64 arm_cmn_read_dtm(struct arm_cmn *cmn, struct arm_cmn_hw_event *hw,
                            bool snapshot)
 {
+       struct arm_cmn_dtm *dtm = NULL;
        struct arm_cmn_node *dn;
-       unsigned int i, offset;
-       u64 count = 0;
+       unsigned int i, offset, dtm_idx;
+       u64 reg, count = 0;
 
        offset = snapshot ? CMN_DTM_PMEVCNTSR : CMN_DTM_PMEVCNT;
        for_each_hw_dn(hw, dn, i) {
-               struct arm_cmn_dtm *dtm = &cmn->dtms[dn->dtm];
-               int dtm_idx = arm_cmn_get_index(hw->dtm_idx, i);
-               u64 reg = readq_relaxed(dtm->base + offset);
-               u16 dtm_count = reg >> (dtm_idx * 16);
-
-               count += dtm_count;
+               if (dtm != &cmn->dtms[dn->dtm]) {
+                       dtm = &cmn->dtms[dn->dtm];
+                       reg = readq_relaxed(dtm->base + offset);
+               }
+               dtm_idx = arm_cmn_get_index(hw->dtm_idx, i);
+               count += (u16)(reg >> (dtm_idx * 16));
        }
        return count;
 }