perf/x86: Avoid exposing wrong/stale data in intel_pmu_lbr_read_32()

When the perf_branch_entry::{in_tx,abort,cycles} fields were added,
intel_pmu_lbr_read_32() wasn't updated to initialize them.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Cc: <stable@vger.kernel.org>
Fixes: 135c561 ("perf/x86/intel: Support Haswell/v4 LBR format")
Signed-off-by: Ingo Molnar <mingo@kernel.org>

arch/x86/events/intel/lbr.c

@@ -507,6 +507,9 @@ static void intel_pmu_lbr_read_32(struct cpu_hw_events *cpuc)
 		cpuc->lbr_entries[i].to		= msr_lastbranch.to;
 		cpuc->lbr_entries[i].mispred	= 0;
 		cpuc->lbr_entries[i].predicted	= 0;
+		cpuc->lbr_entries[i].in_tx	= 0;
+		cpuc->lbr_entries[i].abort	= 0;
+		cpuc->lbr_entries[i].cycles	= 0;
 		cpuc->lbr_entries[i].reserved	= 0;
 	}
 	cpuc->lbr_stack.nr = i;