tests: added check for kernel version

With older kernels this test can't work reliable,
since 'suspend' used to clean all dirty blocks
(taking even several seconds).

test/shell/lvconvert-cache-abort.sh

@@ -50,14 +50,15 @@ test "$(get lv_field $vg/$lv1 cache_dirty_blocks)" -gt 0 || {
 	skip "Cannot make a dirty writeback cache LV."
 }
 
-LVM_TEST_TAG="kill_me_$PREFIX" lvconvert -v --splitcache $vg/$lv1 >logconvert 2>&1 &
+LVM_TEST_TAG="kill_me_$PREFIX" lvconvert -vvvv --splitcache $vg/$lv1 >logconvert 2>&1 &
 PID_CONVERT=$!
 for i in {1..50}; do
-	dmsetup table "$vg-$lv1" | grep cleaner && break
-	test "$i" -ge 100 && die "Waited for cleaner policy on $vg/$lv1 too long!"
-	echo "Waiting for cleaner policy on $vg/$lv1"
+	dmsetup table "$vg-$lv1" |& tee out
+	grep cleaner out && break
+	echo "$i: Waiting for cleaner policy on $vg/$lv1"
 	sleep .05
 done
+test "$i" -ge 49 && die "Waited for cleaner policy on $vg/$lv1 too long!"
 
 # While lvconvert updated table to 'cleaner' policy now it 
 # should be running in 'Flushing' loop and just 1 KILL should
@@ -68,9 +69,16 @@ wait
 
 #cat logconvert || true
 
+# Problem of this test is, in older kernels, even the initial change to cleaner
+# policy table line causes long suspend which in practice is cleaning all the
+# dirty blocks - so the test can't really break the cache clearing.
+#
+# So the failure of test is reported only for recent kernels > 5.6
+# ans skipped otherwise - as those can't be fixed anyway
 grep -E "Flushing.*aborted" logconvert || {
 	cat logconvert || true
 	vgremove -f $vg
+	aux kernel_at_least 5 6 || skip "Cache missed to abort flushing with older kernel"
 	die "Flushing of $vg/$lv1 not aborted ?"
 }