performance-metrics: Add QCOW2 async L2 cache miss micro benchmark

Add micro_bench_qcow_async_l2_cache_miss which reads one cluster
from each of num_ops distinct L2 tables through the QcowAsync
io_uring path, forcing L2 cache eviction on nearly every read.

Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>

performance-metrics/src/main.rs

@@ -378,7 +378,7 @@ mod adjuster {
     }
 }
 
-const TEST_LIST: [PerformanceTest; 96] = [
+const TEST_LIST: [PerformanceTest; 98] = [
     PerformanceTest {
         name: "boot_time_ms",
         func_ptr: performance_boot_time,
@@ -1661,6 +1661,30 @@ const TEST_LIST: [PerformanceTest; 96] = [
         },
         unit_adjuster: adjuster::s_to_us,
     },
+    PerformanceTest {
+        name: "micro_block_qcow_async_l2_cache_miss_128_us",
+        func_ptr: micro_bench_block::micro_bench_qcow_async_l2_cache_miss,
+        control: PerformanceTestControl {
+            test_timeout: 10,
+            test_iterations: 20,
+            warmup_iterations: 5,
+            num_ops: Some(128),
+            ..PerformanceTestControl::default()
+        },
+        unit_adjuster: adjuster::s_to_us,
+    },
+    PerformanceTest {
+        name: "micro_block_qcow_async_l2_cache_miss_256_us",
+        func_ptr: micro_bench_block::micro_bench_qcow_async_l2_cache_miss,
+        control: PerformanceTestControl {
+            test_timeout: 10,
+            test_iterations: 20,
+            warmup_iterations: 5,
+            num_ops: Some(256),
+            ..PerformanceTestControl::default()
+        },
+        unit_adjuster: adjuster::s_to_us,
+    },
 ];
 
 fn run_test_with_timeout(

performance-metrics/src/micro_bench_block.rs

@@ -527,3 +527,25 @@ pub fn micro_bench_qcow_async_write(control: &PerformanceTestControl) -> f64 {
     drain_async_completions(async_io.as_mut(), num_ops);
     start.elapsed().as_secs_f64()
 }
+
+/// Read one cluster from each of num_ops distinct L2 tables in a sparse
+/// QCOW2 image through the QcowAsync io_uring path.
+///
+/// Returns the total read wall clock time in seconds.
+pub fn micro_bench_qcow_async_l2_cache_miss(control: &PerformanceTestControl) -> f64 {
+    let num_ops = control.num_ops.expect("num_ops required") as usize;
+    let (_tmp, disk) = util::sparse_qcow_async_tempfile(num_ops);
+    let mut async_io = disk
+        .new_async_io(num_ops as u32)
+        .expect("new_async_io failed");
+
+    let mut buf = vec![0u8; QCOW_CLUSTER_SIZE as usize];
+    let iovec = read_iovec(&mut buf);
+
+    let stride = L2_ENTRIES_PER_TABLE as u64 * QCOW_CLUSTER_SIZE;
+    let start = Instant::now();
+    submit_reads(async_io.as_mut(), num_ops, stride, &[iovec]);
+
+    drain_async_completions(async_io.as_mut(), num_ops);
+    start.elapsed().as_secs_f64()
+}