perf/core: Fix the mlock accounting, again

Commit:

  5e6c3c7 ("perf/aux: Fix tracking of auxiliary trace buffer allocation")

tried to guess the correct combination of arithmetic operations that would
undo the AUX buffer's mlock accounting, and failed, leaking the bottom part
when an allocation needs to be charged partially to both user->locked_vm
and mm->pinned_vm, eventually leaving the user with no locked bonus:

  $ perf record -e intel_pt//u -m1,128 uname
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.061 MB perf.data ]

  $ perf record -e intel_pt//u -m1,128 uname
  Permission error mapping pages.
  Consider increasing /proc/sys/kernel/perf_event_mlock_kb,
  or try again with a smaller value of -m/--mmap_pages.
  (current value: 1,128)

Fix this by subtracting both locked and pinned counts when AUX buffer is
unmapped.

Reported-by: Thomas Richter <tmricht@linux.ibm.com>
Tested-by: Thomas Richter <tmricht@linux.ibm.com>
Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Signed-off-by: Ingo Molnar <mingo@kernel.org>

kernel/events/core.c

@@ -5607,10 +5607,8 @@ static void perf_mmap_close(struct vm_area_struct *vma)
 		perf_pmu_output_stop(event);
 
 		/* now it's safe to free the pages */
-		if (!rb->aux_mmap_locked)
-			atomic_long_sub(rb->aux_nr_pages, &mmap_user->locked_vm);
-		else
-			atomic64_sub(rb->aux_mmap_locked, &vma->vm_mm->pinned_vm);
+		atomic_long_sub(rb->aux_nr_pages - rb->aux_mmap_locked, &mmap_user->locked_vm);
+		atomic64_sub(rb->aux_mmap_locked, &vma->vm_mm->pinned_vm);
 
 		/* this has to be the last one */
 		rb_free_aux(rb);