Skip to contentSkip to conversation

bpf: Add support for absolute value BPF timers

PATCH RESEND Superseded Tero Kristo 2 participants 3 messages Feb 21, 2023 – Feb 24, 2023 BPF

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH RESEND 1/1 bpf: Add support for absolute value BPF timers

Add a new flag BPF_F_TIMER_ABS that can be passed to bpf_timer_start() to start an absolute value timer instead of the default relative value. This makes the timer expire at an exact point in time, instead of a time with latencies and jitter induced by both the BPF and timer subsystems. This is useful e.g. in certain time sensitive profiling cases, where we need a timer to expire at an exact point in time.

---
 include/uapi/linux/bpf.h | 15 +++++++++++++++
 kernel/bpf/helpers.c     | 11 +++++++++--
 2 files changed, 24 insertions(+), 2 deletions(-)
include/uapi/linux/bpf.h +15 -0
index 464ca3f01fe7..7f5b71847984 100644
@@ -4951,6 +4951,12 @@ union bpf_attr {
  *		different maps if key/value layout matches across maps.
  *		Every bpf_timer_set_callback() can have different callback_fn.
  *
+ *		*flags* can be one of:
+ *
+ *		**BPF_F_TIMER_ABS**
+ *			Start the timer in absolute expire value instead of the
+ *			default relative one.
+ *
  *	Return
  *		0 on success.
  *		**-EINVAL** if *timer* was not initialized with bpf_timer_init() earlier
@@ -7050,4 +7056,13 @@ struct bpf_core_relo {
 	enum bpf_core_relo_kind kind;
 };
 
+/*
+ * Flags to control bpf_timer_start() behaviour.
+ *     - BPF_F_TIMER_ABS: Timeout passed is absolute time, by default it is
+ *       relative to current time.
+ */
+enum {
+	BPF_F_TIMER_ABS = (1ULL << 0),
+};
+
 #endif /* _UAPI__LINUX_BPF_H__ */
kernel/bpf/helpers.c +9 -2
index af30c6cbd65d..924849d89828 100644
@@ -1253,10 +1253,11 @@ BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, fla
 {
 	struct bpf_hrtimer *t;
 	int ret = 0;
+	enum hrtimer_mode mode;
 
 	if (in_nmi())
 		return -EOPNOTSUPP;
-	if (flags)
+	if (flags > BPF_F_TIMER_ABS)
 		return -EINVAL;
 	__bpf_spin_lock_irqsave(&timer->lock);
 	t = timer->timer;
@@ -1264,7 +1265,13 @@ BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, fla
 		ret = -EINVAL;
 		goto out;
 	}
-	hrtimer_start(&t->timer, ns_to_ktime(nsecs), HRTIMER_MODE_REL_SOFT);
+
+	if (flags & BPF_F_TIMER_ABS)
+		mode = HRTIMER_MODE_ABS_SOFT;
+	else
+		mode = HRTIMER_MODE_REL_SOFT;
+
+	hrtimer_start(&t->timer, ns_to_ktime(nsecs), mode);
 out:
 	__bpf_spin_unlock_irqrestore(&timer->lock);
 	return ret;
Signature · 1 line
2.25.1
#1 lore raw
Quoted 7 lines · Tero Kristo

Add a new flag BPF_F_TIMER_ABS that can be passed to bpf_timer_start() to start an absolute value timer instead of the default relative value. This makes the timer expire at an exact point in time, instead of a time with latencies and jitter induced by both the BPF and timer subsystems. This is useful e.g. in certain time sensitive profiling cases, where we need a timer to expire at an exact point in time.

"certain time sensitive profiling cases" is too vague.

Please precisely describe the use case.

Quoted 56 lines
---
 include/uapi/linux/bpf.h | 15 +++++++++++++++
 kernel/bpf/helpers.c     | 11 +++++++++--
 2 files changed, 24 insertions(+), 2 deletions(-)
include/uapi/linux/bpf.h +15 -0
index 464ca3f01fe7..7f5b71847984 100644
@@ -4951,6 +4951,12 @@ union bpf_attr {
  *             different maps if key/value layout matches across maps.
  *             Every bpf_timer_set_callback() can have different callback_fn.
  *
+ *             *flags* can be one of:
+ *
+ *             **BPF_F_TIMER_ABS**
+ *                     Start the timer in absolute expire value instead of the
+ *                     default relative one.
+ *
  *     Return
  *             0 on success.
  *             **-EINVAL** if *timer* was not initialized with bpf_timer_init() earlier
@@ -7050,4 +7056,13 @@ struct bpf_core_relo {
        enum bpf_core_relo_kind kind;
 };
 
+/*
+ * Flags to control bpf_timer_start() behaviour.
+ *     - BPF_F_TIMER_ABS: Timeout passed is absolute time, by default it is
+ *       relative to current time.
+ */
+enum {
+       BPF_F_TIMER_ABS = (1ULL << 0),
+};
+
 #endif /* _UAPI__LINUX_BPF_H__ */
kernel/bpf/helpers.c +2 -1
index af30c6cbd65d..924849d89828 100644
@@ -1253,10 +1253,11 @@ BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, fla
 {
        struct bpf_hrtimer *t;
        int ret = 0;
+       enum hrtimer_mode mode;
 
        if (in_nmi())
                return -EOPNOTSUPP;
-       if (flags)
+       if (flags > BPF_F_TIMER_ABS)
                return -EINVAL;
        __bpf_spin_lock_irqsave(&timer->lock);
        t = timer->timer;
@@ -1264,7 +1265,13 @@ BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, fla
                ret = -EINVAL;

kernel/bpf/helpers.c BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, fla · line 1274

Quoted 10 lines
kernel/bpf/helpers.c +7 -1
                goto out;
        }
-       hrtimer_start(&t->timer, ns_to_ktime(nsecs), HRTIMER_MODE_REL_SOFT);
+
+       if (flags & BPF_F_TIMER_ABS)
+               mode = HRTIMER_MODE_ABS_SOFT;
+       else
+               mode = HRTIMER_MODE_REL_SOFT;
+
+       hrtimer_start(&t->timer, ns_to_ktime(nsecs), mode);

The patch looks fine, but please add a selftest for new functionality in the 2nd patch and resend them together.

#2 lore raw
Quoted 10 lines · Alexei Starovoitov
Quoted 6 lines · Tero Kristo

Add a new flag BPF_F_TIMER_ABS that can be passed to bpf_timer_start() to start an absolute value timer instead of the default relative value. This makes the timer expire at an exact point in time, instead of a time with latencies and jitter induced by both the BPF and timer subsystems. This is useful e.g. in certain time sensitive profiling cases, where we need a timer to expire at an exact point in time.

"certain time sensitive profiling cases" is too vague.

Please precisely describe the use case.

Ok, I'll do my best to update the changelog on that. There is an open source tool available that will benefit from this change so I can refer to that.

Quoted 61 lines
Quoted 61 lines
---
  include/uapi/linux/bpf.h | 15 +++++++++++++++
  kernel/bpf/helpers.c     | 11 +++++++++--
  2 files changed, 24 insertions(+), 2 deletions(-)
include/uapi/linux/bpf.h +15 -0
index 464ca3f01fe7..7f5b71847984 100644
@@ -4951,6 +4951,12 @@ union bpf_attr {
   *             different maps if key/value layout matches across maps.
   *             Every bpf_timer_set_callback() can have different callback_fn.
   *
+ *             *flags* can be one of:
+ *
+ *             **BPF_F_TIMER_ABS**
+ *                     Start the timer in absolute expire value instead of the
+ *                     default relative one.
+ *
   *     Return
   *             0 on success.
   *             **-EINVAL** if *timer* was not initialized with bpf_timer_init() earlier
@@ -7050,4 +7056,13 @@ struct bpf_core_relo {
         enum bpf_core_relo_kind kind;
  };
 
+/*
+ * Flags to control bpf_timer_start() behaviour.
+ *     - BPF_F_TIMER_ABS: Timeout passed is absolute time, by default it is
+ *       relative to current time.
+ */
+enum {
+       BPF_F_TIMER_ABS = (1ULL << 0),
+};
+
  #endif /* _UAPI__LINUX_BPF_H__ */
kernel/bpf/helpers.c +3 -2
index af30c6cbd65d..924849d89828 100644
@@ -1253,10 +1253,11 @@ BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, fla
  {
         struct bpf_hrtimer *t;
         int ret = 0;
+       enum hrtimer_mode mode;
 
         if (in_nmi())
                 return -EOPNOTSUPP;
-       if (flags)
+       if (flags > BPF_F_TIMER_ABS)
                 return -EINVAL;
         __bpf_spin_lock_irqsave(&timer->lock);
         t = timer->timer;
@@ -1264,7 +1265,13 @@ BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, fla
                 ret = -EINVAL;
                 goto out;
         }
-       hrtimer_start(&t->timer, ns_to_ktime(nsecs), HRTIMER_MODE_REL_SOFT);
+
Quoted 8 lines

kernel/bpf/helpers.c BPF_CALL_3(bpf_timer_start, struct bpf_timer_kern *, timer, u64, nsecs, u64, fla · line 1274

Quoted 6 lines
kernel/bpf/helpers.c +6 -0
+       if (flags & BPF_F_TIMER_ABS)
+               mode = HRTIMER_MODE_ABS_SOFT;
+       else
+               mode = HRTIMER_MODE_REL_SOFT;
+
+       hrtimer_start(&t->timer, ns_to_ktime(nsecs), mode);

The patch looks fine, but please add a selftest for new functionality in the 2nd patch and resend them together.

Ok, will look at that, thanks!

-Tero