selftests/bpf: Test pinning bpf timer to a core
authorDavid Vernet <void@manifault.com>
Wed, 4 Oct 2023 16:23:39 +0000 (11:23 -0500)
committerDaniel Borkmann <daniel@iogearbox.net>
Mon, 9 Oct 2023 14:29:06 +0000 (16:29 +0200)
Now that we support pinning a BPF timer to the current core, we should
test it with some selftests. This patch adds two new testcases to the
timer suite, which verifies that a BPF timer both with and without
BPF_F_TIMER_ABS, can be pinned to the calling core with BPF_F_TIMER_CPU_PIN.

Signed-off-by: David Vernet <void@manifault.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <song@kernel.org>
Acked-by: Hou Tao <houtao1@huawei.com>
Link: https://lore.kernel.org/bpf/20231004162339.200702-3-void@manifault.com
tools/testing/selftests/bpf/prog_tests/timer.c
tools/testing/selftests/bpf/progs/timer.c

index 290c21dbe65ab5546efdc17a50782515f0a20a43..d8bc838445ecd974b1087949799ef9a8e5a363b9 100644 (file)
@@ -14,6 +14,7 @@ static int timer(struct timer *timer_skel)
 
        ASSERT_EQ(timer_skel->data->callback_check, 52, "callback_check1");
        ASSERT_EQ(timer_skel->data->callback2_check, 52, "callback2_check1");
+       ASSERT_EQ(timer_skel->bss->pinned_callback_check, 0, "pinned_callback_check1");
 
        prog_fd = bpf_program__fd(timer_skel->progs.test1);
        err = bpf_prog_test_run_opts(prog_fd, &topts);
@@ -32,6 +33,9 @@ static int timer(struct timer *timer_skel)
        /* check that timer_cb3() was executed twice */
        ASSERT_EQ(timer_skel->bss->abs_data, 12, "abs_data");
 
+       /* check that timer_cb_pinned() was executed twice */
+       ASSERT_EQ(timer_skel->bss->pinned_callback_check, 2, "pinned_callback_check");
+
        /* check that there were no errors in timer execution */
        ASSERT_EQ(timer_skel->bss->err, 0, "err");
 
index 9a16d95213e11527c15cfb44b4fef8d9415f47fa..8b946c8188c65d10ed86de886ea9aead585de86d 100644 (file)
@@ -51,7 +51,7 @@ struct {
        __uint(max_entries, 1);
        __type(key, int);
        __type(value, struct elem);
-} abs_timer SEC(".maps");
+} abs_timer SEC(".maps"), soft_timer_pinned SEC(".maps"), abs_timer_pinned SEC(".maps");
 
 __u64 bss_data;
 __u64 abs_data;
@@ -59,6 +59,8 @@ __u64 err;
 __u64 ok;
 __u64 callback_check = 52;
 __u64 callback2_check = 52;
+__u64 pinned_callback_check;
+__s32 pinned_cpu;
 
 #define ARRAY 1
 #define HTAB 2
@@ -329,3 +331,62 @@ int BPF_PROG2(test3, int, a)
 
        return 0;
 }
+
+/* callback for pinned timer */
+static int timer_cb_pinned(void *map, int *key, struct bpf_timer *timer)
+{
+       __s32 cpu = bpf_get_smp_processor_id();
+
+       if (cpu != pinned_cpu)
+               err |= 16384;
+
+       pinned_callback_check++;
+       return 0;
+}
+
+static void test_pinned_timer(bool soft)
+{
+       int key = 0;
+       void *map;
+       struct bpf_timer *timer;
+       __u64 flags = BPF_F_TIMER_CPU_PIN;
+       __u64 start_time;
+
+       if (soft) {
+               map = &soft_timer_pinned;
+               start_time = 0;
+       } else {
+               map = &abs_timer_pinned;
+               start_time = bpf_ktime_get_boot_ns();
+               flags |= BPF_F_TIMER_ABS;
+       }
+
+       timer = bpf_map_lookup_elem(map, &key);
+       if (timer) {
+               if (bpf_timer_init(timer, map, CLOCK_BOOTTIME) != 0)
+                       err |= 4096;
+               bpf_timer_set_callback(timer, timer_cb_pinned);
+               pinned_cpu = bpf_get_smp_processor_id();
+               bpf_timer_start(timer, start_time + 1000, flags);
+       } else {
+               err |= 8192;
+       }
+}
+
+SEC("fentry/bpf_fentry_test4")
+int BPF_PROG2(test4, int, a)
+{
+       bpf_printk("test4");
+       test_pinned_timer(true);
+
+       return 0;
+}
+
+SEC("fentry/bpf_fentry_test5")
+int BPF_PROG2(test5, int, a)
+{
+       bpf_printk("test5");
+       test_pinned_timer(false);
+
+       return 0;
+}