performance-metrics: Add QCOW2 async multicluster read micro benchmark

Add micro_bench_qcow_async_multi_cluster_read which reads 8
contiguous clusters (512 KiB) per request through the QcowAsync
io_uring path. With coalesced mappings this can hit the io_uring
fast path for a single Readv SQE.

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

performance-metrics/src/main.rs

@@ -378,7 +378,7 @@ mod adjuster {
     }
 }
 
-const TEST_LIST: [PerformanceTest; 88] = [
+const TEST_LIST: [PerformanceTest; 90] = [
     PerformanceTest {
         name: "boot_time_ms",
         func_ptr: performance_boot_time,
@@ -1565,6 +1565,30 @@ const TEST_LIST: [PerformanceTest; 88] = [
         },
         unit_adjuster: adjuster::s_to_us,
     },
+    PerformanceTest {
+        name: "micro_block_qcow_async_multi_cluster_read_128_us",
+        func_ptr: micro_bench_block::micro_bench_qcow_async_multi_cluster_read,
+        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_multi_cluster_read_256_us",
+        func_ptr: micro_bench_block::micro_bench_qcow_async_multi_cluster_read,
+        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

@@ -428,3 +428,31 @@ pub fn micro_bench_qcow_async_random_read(control: &PerformanceTestControl) -> f
     drain_async_completions(async_io.as_mut(), num_ops);
     start.elapsed().as_secs_f64()
 }
+
+/// Issue large multi-cluster reads from a prepopulated QCOW2 image
+/// through the QcowAsync io_uring path.
+///
+/// Each read spans 8 contiguous clusters (512 KiB). With coalesced
+/// mappings, this can hit the io_uring fast path for a single Readv.
+///
+/// Returns the total read wall clock time in seconds.
+pub fn micro_bench_qcow_async_multi_cluster_read(control: &PerformanceTestControl) -> f64 {
+    const CLUSTERS_PER_READ: usize = 8;
+
+    let num_ops = control.num_ops.expect("num_ops required") as usize;
+    let (_tmp, disk) = util::qcow_async_tempfile(num_ops);
+    let mut async_io = disk
+        .new_async_io(num_ops as u32)
+        .expect("new_async_io failed");
+
+    let read_size = CLUSTERS_PER_READ * QCOW_CLUSTER_SIZE as usize;
+    let mut buf = vec![0u8; read_size];
+    let iovec = read_iovec(&mut buf);
+
+    let num_reads = num_ops / CLUSTERS_PER_READ;
+    let start = Instant::now();
+    submit_reads(async_io.as_mut(), num_reads, read_size as u64, &[iovec]);
+
+    drain_async_completions(async_io.as_mut(), num_reads);
+    start.elapsed().as_secs_f64()
+}