[PATCH bpf-next 1/2] bpf: Add ability to pin bpf timer to calling CPU

[PATCH bpf-next 1/2] bpf: Add ability to pin bpf timer to calling CPU

[David Vernet]

BPF supports creating high resolution timers using bpf_timer_* helper
functions. Currently, only the BPF_F_TIMER_ABS flag is supported, which
specifies that the timeout should be interpreted as absolute time. It
would also be useful to be able to pin that timer to a core. For
example, if you wanted to make a subset of cores run without timer
interrupts, and only have the timer be invoked on a single core.

This patch adds support for this with a new BPF_F_TIMER_CPU_PIN flag.
When specified, the HRTIMER_MODE_PINNED flag is passed to
hrtimer_start(). A subsequent patch will update selftests to validate.

Signed-off-by: David Vernet <void@manifault.com>
---
 include/uapi/linux/bpf.h       | 4 ++++
 kernel/bpf/helpers.c           | 5 ++++-
 tools/include/uapi/linux/bpf.h | 4 ++++
 3 files changed, 12 insertions(+), 1 deletion(-)

diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index 70bfa997e896..a7d4a1a69f21 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h
@@ -5096,6 +5096,8 @@ union bpf_attr {
  *		**BPF_F_TIMER_ABS**
  *			Start the timer in absolute expire value instead of the
  *			default relative one.
+ *		**BPF_F_TIMER_CPU_PIN**
+ *			Timer will be pinned to the CPU of the caller.
  *
  *	Return
  *		0 on success.
@@ -7309,9 +7311,11 @@ struct bpf_core_relo {
  * Flags to control bpf_timer_start() behaviour.
  *     - BPF_F_TIMER_ABS: Timeout passed is absolute time, by default it is
  *       relative to current time.
+ *     - BPF_F_TIMER_CPU_PIN: Timer will be pinned to the CPU of the caller.
  */
 enum {
 	BPF_F_TIMER_ABS = (1ULL << 0),
+	BPF_F_TIMER_CPU_PIN = (1ULL << 1),
 };
 
 /* BPF numbers iterator state */
diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
index dd1c69ee3375..d2840dd5b00d 100644
--- a/kernel/bpf/helpers.c
+++ b/kernel/bpf/helpers.c
@@ -1272,7 +1272,7 @@ BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, fla
 
 	if (in_nmi())
 		return -EOPNOTSUPP;
-	if (flags > BPF_F_TIMER_ABS)
+	if (flags & ~(BPF_F_TIMER_ABS | BPF_F_TIMER_CPU_PIN))
 		return -EINVAL;
 	__bpf_spin_lock_irqsave(&timer->lock);
 	t = timer->timer;
@@ -1286,6 +1286,9 @@ BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, fla
 	else
 		mode = HRTIMER_MODE_REL_SOFT;
 
+	if (flags & BPF_F_TIMER_CPU_PIN)
+		mode |= HRTIMER_MODE_PINNED;
+
 	hrtimer_start(&t->timer, ns_to_ktime(nsecs), mode);
 out:
 	__bpf_spin_unlock_irqrestore(&timer->lock);
diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
index 70bfa997e896..a7d4a1a69f21 100644
--- a/tools/include/uapi/linux/bpf.h
+++ b/tools/include/uapi/linux/bpf.h
@@ -5096,6 +5096,8 @@ union bpf_attr {
  *		**BPF_F_TIMER_ABS**
  *			Start the timer in absolute expire value instead of the
  *			default relative one.
+ *		**BPF_F_TIMER_CPU_PIN**
+ *			Timer will be pinned to the CPU of the caller.
  *
  *	Return
  *		0 on success.
@@ -7309,9 +7311,11 @@ struct bpf_core_relo {
  * Flags to control bpf_timer_start() behaviour.
  *     - BPF_F_TIMER_ABS: Timeout passed is absolute time, by default it is
  *       relative to current time.
+ *     - BPF_F_TIMER_CPU_PIN: Timer will be pinned to the CPU of the caller.
  */
 enum {
 	BPF_F_TIMER_ABS = (1ULL << 0),
+	BPF_F_TIMER_CPU_PIN = (1ULL << 1),
 };
 
 /* BPF numbers iterator state */
-- 
2.41.0

[PATCH bpf-next 2/2] bpf/selftests: Test pinning bpf timer to a core

[David Vernet]

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>
---
 .../testing/selftests/bpf/prog_tests/timer.c  |  4 +
 tools/testing/selftests/bpf/progs/timer.c     | 75 +++++++++++++++++++
 2 files changed, 79 insertions(+)

diff --git a/tools/testing/selftests/bpf/prog_tests/timer.c b/tools/testing/selftests/bpf/prog_tests/timer.c
index 290c21dbe65a..d8bc838445ec 100644
--- a/tools/testing/selftests/bpf/prog_tests/timer.c
+++ b/tools/testing/selftests/bpf/prog_tests/timer.c
@@ -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");
 
diff --git a/tools/testing/selftests/bpf/progs/timer.c b/tools/testing/selftests/bpf/progs/timer.c
index 9a16d95213e1..0112b9c038b4 100644
--- a/tools/testing/selftests/bpf/progs/timer.c
+++ b/tools/testing/selftests/bpf/progs/timer.c
@@ -53,12 +53,28 @@ struct {
 	__type(value, struct elem);
 } abs_timer SEC(".maps");
 
+struct {
+	__uint(type, BPF_MAP_TYPE_ARRAY);
+	__uint(max_entries, 1);
+	__type(key, int);
+	__type(value, struct elem);
+} soft_timer_pinned SEC(".maps");
+
+struct {
+	__uint(type, BPF_MAP_TYPE_ARRAY);
+	__uint(max_entries, 1);
+	__type(key, int);
+	__type(value, struct elem);
+} abs_timer_pinned SEC(".maps");
+
 __u64 bss_data;
 __u64 abs_data;
 __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 +345,62 @@ 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;
+}
-- 
2.41.0

Re: [PATCH bpf-next 1/2] bpf: Add ability to pin bpf timer to calling CPU

[Song Liu]

> On Oct 2, 2023, at 4:47 PM, David Vernet <void@manifault.com> wrote:
> 
> BPF supports creating high resolution timers using bpf_timer_* helper
> functions. Currently, only the BPF_F_TIMER_ABS flag is supported, which
> specifies that the timeout should be interpreted as absolute time. It
> would also be useful to be able to pin that timer to a core. For
> example, if you wanted to make a subset of cores run without timer
> interrupts, and only have the timer be invoked on a single core.
> 
> This patch adds support for this with a new BPF_F_TIMER_CPU_PIN flag.
> When specified, the HRTIMER_MODE_PINNED flag is passed to
> hrtimer_start(). A subsequent patch will update selftests to validate.
> 
> Signed-off-by: David Vernet <void@manifault.com>

Acked-by: Song Liu <song@kernel.org>

Re: [PATCH bpf-next 2/2] bpf/selftests: Test pinning bpf timer to a core

[Song Liu]

> On Oct 2, 2023, at 4:47 PM, David Vernet <void@manifault.com> wrote:
> 
> 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>

Acked-by: Song Liu <song@kernel.org>

With one nit/question below. 

> ---
> .../testing/selftests/bpf/prog_tests/timer.c  |  4 +
> tools/testing/selftests/bpf/progs/timer.c     | 75 +++++++++++++++++++
> 2 files changed, 79 insertions(+)
> 
> diff --git a/tools/testing/selftests/bpf/prog_tests/timer.c b/tools/testing/selftests/bpf/prog_tests/timer.c
> index 290c21dbe65a..d8bc838445ec 100644
> --- a/tools/testing/selftests/bpf/prog_tests/timer.c
> +++ b/tools/testing/selftests/bpf/prog_tests/timer.c
> @@ -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");
> 
> diff --git a/tools/testing/selftests/bpf/progs/timer.c b/tools/testing/selftests/bpf/progs/timer.c
> index 9a16d95213e1..0112b9c038b4 100644
> --- a/tools/testing/selftests/bpf/progs/timer.c
> +++ b/tools/testing/selftests/bpf/progs/timer.c
> @@ -53,12 +53,28 @@ struct {
> __type(value, struct elem);
> } abs_timer SEC(".maps");
> 
> +struct {
> + __uint(type, BPF_MAP_TYPE_ARRAY);
> + __uint(max_entries, 1);
> + __type(key, int);
> + __type(value, struct elem);
> +} soft_timer_pinned SEC(".maps");
> +
> +struct {
> + __uint(type, BPF_MAP_TYPE_ARRAY);
> + __uint(max_entries, 1);
> + __type(key, int);
> + __type(value, struct elem);
> +} abs_timer_pinned SEC(".maps");

nit: I think we can also do something like the following, but I am not 
sure whether this style is not recommended. 

diff --git i/tools/testing/selftests/bpf/progs/timer.c w/tools/testing/selftests/bpf/progs/timer.c
index 9a16d95213e1..638eeebcd6c9 100644
--- i/tools/testing/selftests/bpf/progs/timer.c
+++ w/tools/testing/selftests/bpf/progs/timer.c
@@ -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;




[...]

Re: [PATCH bpf-next 2/2] bpf/selftests: Test pinning bpf timer to a core

[David Vernet]

On Tue, Oct 03, 2023 at 06:15:03PM +0000, Song Liu wrote:
> 
> 
> > On Oct 2, 2023, at 4:47 PM, David Vernet <void@manifault.com> wrote:
> > 
> > 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>
> 
> Acked-by: Song Liu <song@kernel.org>
> 
> With one nit/question below. 
> 
> > ---
> > .../testing/selftests/bpf/prog_tests/timer.c  |  4 +
> > tools/testing/selftests/bpf/progs/timer.c     | 75 +++++++++++++++++++
> > 2 files changed, 79 insertions(+)
> > 
> > diff --git a/tools/testing/selftests/bpf/prog_tests/timer.c b/tools/testing/selftests/bpf/prog_tests/timer.c
> > index 290c21dbe65a..d8bc838445ec 100644
> > --- a/tools/testing/selftests/bpf/prog_tests/timer.c
> > +++ b/tools/testing/selftests/bpf/prog_tests/timer.c
> > @@ -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");
> > 
> > diff --git a/tools/testing/selftests/bpf/progs/timer.c b/tools/testing/selftests/bpf/progs/timer.c
> > index 9a16d95213e1..0112b9c038b4 100644
> > --- a/tools/testing/selftests/bpf/progs/timer.c
> > +++ b/tools/testing/selftests/bpf/progs/timer.c
> > @@ -53,12 +53,28 @@ struct {
> > __type(value, struct elem);
> > } abs_timer SEC(".maps");
> > 
> > +struct {
> > + __uint(type, BPF_MAP_TYPE_ARRAY);
> > + __uint(max_entries, 1);
> > + __type(key, int);
> > + __type(value, struct elem);
> > +} soft_timer_pinned SEC(".maps");
> > +
> > +struct {
> > + __uint(type, BPF_MAP_TYPE_ARRAY);
> > + __uint(max_entries, 1);
> > + __type(key, int);
> > + __type(value, struct elem);
> > +} abs_timer_pinned SEC(".maps");
> 
> nit: I think we can also do something like the following, but I am not 
> sure whether this style is not recommended. 
> 
> diff --git i/tools/testing/selftests/bpf/progs/timer.c w/tools/testing/selftests/bpf/progs/timer.c
> index 9a16d95213e1..638eeebcd6c9 100644
> --- i/tools/testing/selftests/bpf/progs/timer.c
> +++ w/tools/testing/selftests/bpf/progs/timer.c
> @@ -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");

This looks like a nice readability improvement / cleanup to me. If
nobody objects, I'd say let's apply it.

Thanks,
David