[PATCH net-next v4] net: af_packet: Use hrtimer to do the retire operation

[PATCH net-next v4] net: af_packet: Use hrtimer to do the retire operation

[Xin Zhao]

In a system with high real-time requirements, the timeout mechanism of
ordinary timers with jiffies granularity is insufficient to meet the
demands for real-time performance. Meanwhile, the optimization of CPU
usage with af_packet is quite significant. Use hrtimer instead of timer
to help compensate for the shortcomings in real-time performance.
In HZ=100 or HZ=250 system, the update of TP_STATUS_USER is not real-time
enough, with fluctuations reaching over 8ms (on a system with HZ=250).
This is unacceptable in some high real-time systems that require timely
processing of network packets. By replacing it with hrtimer, if a timeout
of 2ms is set, the update of TP_STATUS_USER can be stabilized to within
3 ms.

Signed-off-by: Xin Zhao <jackzxcui1989@163.com>
---
Changes in v4:
- Add 'bool start' to distinguish whether the call to _prb_refresh_rx_retire_blk_timer
  is for prb_open_block. When it is for prb_open_block, execute hrtimer_start to
  (re)start the hrtimer; otherwise, use hrtimer_set_expires to update the expiration
  time of the hrtimer
  as suggested by Willem de Bruijn;
Changes in v3:
- return HRTIMER_NORESTART when pkc->delete_blk_timer is true
  as suggested by Willem de Bruijn;
- Drop the retire_blk_tov field of tpacket_kbdq_core, add interval_ktime instead
  as suggested by Willem de Bruijn;
- Add comments to explain why hrtimer_set_expires(not hrtimer_forward_now) is used in
  _prb_refresh_rx_retire_blk_timer
  as suggested by Willem de Bruijn;

Changes in v2:
- Drop the tov_in_msecs field of tpacket_kbdq_core added by the patch
  as suggested by Willem de Bruijn;
- Link to v2: https://lore.kernel.org/all/20250815044141.1374446-1-jackzxcui1989@163.com/

Changes in v1:
- Do not add another config for the current changes
  as suggested by Eric Dumazet;
- Mention the beneficial cases 'HZ=100 or HZ=250' in the changelog
  as suggested by Eric Dumazet;
- Add some performance details to the changelog
  as suggested by Ferenc Fejes;
- Delete the 'pkc->tov_in_msecs == 0' bounds check which is not necessary
  as suggested by Willem de Bruijn;
- Use hrtimer_set_expires instead of hrtimer_start_range_ns when retire timer needs update
  as suggested by Willem de Bruijn. Start the hrtimer in prb_setup_retire_blk_timer;
- Just return HRTIMER_RESTART directly as all cases return the same value
  as suggested by Willem de Bruijn;
- Link to v1: https://lore.kernel.org/all/20250813165201.1492779-1-jackzxcui1989@163.com/
- Link to v0: https://lore.kernel.org/all/20250806055210.1530081-1-jackzxcui1989@163.com/

---
---
 net/packet/af_packet.c | 52 ++++++++++++++++++++++++++++--------------
 net/packet/diag.c      |  2 +-
 net/packet/internal.h  |  5 ++--
 3 files changed, 38 insertions(+), 21 deletions(-)

diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
index a7017d7f0..5a1e80185 100644
--- a/net/packet/af_packet.c
+++ b/net/packet/af_packet.c
@@ -203,8 +203,8 @@ static void prb_retire_current_block(struct tpacket_kbdq_core *,
 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 static void prb_open_block(struct tpacket_kbdq_core *,
 		struct tpacket_block_desc *);
-static void prb_retire_rx_blk_timer_expired(struct timer_list *);
-static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
+static enum hrtimer_restart prb_retire_rx_blk_timer_expired(struct hrtimer *);
+static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *, bool);
 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 		struct tpacket3_hdr *);
@@ -581,7 +581,7 @@ static __be16 vlan_get_protocol_dgram(const struct sk_buff *skb)
 
 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 {
-	timer_delete_sync(&pkc->retire_blk_timer);
+	hrtimer_cancel(&pkc->retire_blk_timer);
 }
 
 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
@@ -603,9 +603,8 @@ static void prb_setup_retire_blk_timer(struct packet_sock *po)
 	struct tpacket_kbdq_core *pkc;
 
 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
-	timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
-		    0);
-	pkc->retire_blk_timer.expires = jiffies;
+	hrtimer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
+		      CLOCK_MONOTONIC, HRTIMER_MODE_REL_SOFT);
 }
 
 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
@@ -672,11 +671,10 @@ static void init_prb_bdqc(struct packet_sock *po,
 	p1->last_kactive_blk_num = 0;
 	po->stats.stats3.tp_freeze_q_cnt = 0;
 	if (req_u->req3.tp_retire_blk_tov)
-		p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
+		p1->interval_ktime = ms_to_ktime(req_u->req3.tp_retire_blk_tov);
 	else
-		p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
-						req_u->req3.tp_block_size);
-	p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
+		p1->interval_ktime = ms_to_ktime(prb_calc_retire_blk_tmo(po,
+						req_u->req3.tp_block_size));
 	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 	rwlock_init(&p1->blk_fill_in_prog_lock);
 
@@ -688,11 +686,27 @@ static void init_prb_bdqc(struct packet_sock *po,
 
 /*  Do NOT update the last_blk_num first.
  *  Assumes sk_buff_head lock is held.
+ *  We only need to (re)start an hrtimer in prb_open_block.
+ *  Otherwise, we just need to update the expiration time of the hrtimer.
  */
-static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
+static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc,
+		bool start)
 {
-	mod_timer(&pkc->retire_blk_timer,
-			jiffies + pkc->tov_in_jiffies);
+	if (start)
+		hrtimer_start(&pkc->retire_blk_timer, pkc->interval_ktime,
+			      HRTIMER_MODE_REL_SOFT);
+	else
+		/* We cannot use hrtimer_forward_now here because the function
+		 * _prb_refresh_rx_retire_blk_timer can be called not only when
+		 * the retire timer expires, but also when the kernel logic for
+		 * receiving network packets detects that a network packet has
+		 * filled up a block and calls prb_open_block to use the next
+		 * block. This can lead to a WARN_ON being triggered in
+		 * hrtimer_forward_now when it checks if the timer has already
+		 * been enqueued.
+		 */
+		hrtimer_set_expires(&pkc->retire_blk_timer,
+				    ktime_add(ktime_get(), pkc->interval_ktime));
 	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 }
 
@@ -719,8 +733,9 @@ static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
  * prb_calc_retire_blk_tmo() calculates the tmo.
  *
  */
-static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
+static enum hrtimer_restart prb_retire_rx_blk_timer_expired(struct hrtimer *t)
 {
+	enum hrtimer_restart ret = HRTIMER_RESTART;
 	struct packet_sock *po =
 		timer_container_of(po, t, rx_ring.prb_bdqc.retire_blk_timer);
 	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
@@ -732,8 +747,10 @@ static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
 	frozen = prb_queue_frozen(pkc);
 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 
-	if (unlikely(pkc->delete_blk_timer))
+	if (unlikely(pkc->delete_blk_timer)) {
+		ret = HRTIMER_NORESTART;
 		goto out;
+	}
 
 	/* We only need to plug the race when the block is partially filled.
 	 * tpacket_rcv:
@@ -786,10 +803,11 @@ static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
 	}
 
 refresh_timer:
-	_prb_refresh_rx_retire_blk_timer(pkc);
+	_prb_refresh_rx_retire_blk_timer(pkc, false);
 
 out:
 	spin_unlock(&po->sk.sk_receive_queue.lock);
+	return ret;
 }
 
 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
@@ -921,7 +939,7 @@ static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 
 	prb_thaw_queue(pkc1);
-	_prb_refresh_rx_retire_blk_timer(pkc1);
+	_prb_refresh_rx_retire_blk_timer(pkc1, true);
 
 	smp_wmb();
 }
diff --git a/net/packet/diag.c b/net/packet/diag.c
index 6ce1dcc28..c8f43e0c1 100644
--- a/net/packet/diag.c
+++ b/net/packet/diag.c
@@ -83,7 +83,7 @@ static int pdiag_put_ring(struct packet_ring_buffer *ring, int ver, int nl_type,
 	pdr.pdr_frame_nr = ring->frame_max + 1;
 
 	if (ver > TPACKET_V2) {
-		pdr.pdr_retire_tmo = ring->prb_bdqc.retire_blk_tov;
+		pdr.pdr_retire_tmo = ktime_to_ms(ring->prb_bdqc.interval_ktime);
 		pdr.pdr_sizeof_priv = ring->prb_bdqc.blk_sizeof_priv;
 		pdr.pdr_features = ring->prb_bdqc.feature_req_word;
 	} else {
diff --git a/net/packet/internal.h b/net/packet/internal.h
index 1e743d031..19d4f0b73 100644
--- a/net/packet/internal.h
+++ b/net/packet/internal.h
@@ -45,12 +45,11 @@ struct tpacket_kbdq_core {
 	/* Default is set to 8ms */
 #define DEFAULT_PRB_RETIRE_TOV	(8)
 
-	unsigned short  retire_blk_tov;
+	ktime_t		interval_ktime;
 	unsigned short  version;
-	unsigned long	tov_in_jiffies;
 
 	/* timer to retire an outstanding block */
-	struct timer_list retire_blk_timer;
+	struct hrtimer  retire_blk_timer;
 };
 
 struct pgv {
-- 
2.34.1

[PATCH net-next v4] net: af_packet: Use hrtimer to do the retire operation

[Xin Zhao]

In a system with high real-time requirements, the timeout mechanism of
ordinary timers with jiffies granularity is insufficient to meet the
demands for real-time performance. Meanwhile, the optimization of CPU
usage with af_packet is quite significant. Use hrtimer instead of timer
to help compensate for the shortcomings in real-time performance.
In HZ=100 or HZ=250 system, the update of TP_STATUS_USER is not real-time
enough, with fluctuations reaching over 8ms (on a system with HZ=250).
This is unacceptable in some high real-time systems that require timely
processing of network packets. By replacing it with hrtimer, if a timeout
of 2ms is set, the update of TP_STATUS_USER can be stabilized to within
3 ms.

Signed-off-by: Xin Zhao <jackzxcui1989@163.com>
---
Changes in v4:
- Add 'bool start' to distinguish whether the call to _prb_refresh_rx_retire_blk_timer
  is for prb_open_block. When it is for prb_open_block, execute hrtimer_start to
  (re)start the hrtimer; otherwise, use hrtimer_set_expires to update the expiration
  time of the hrtimer
  as suggested by Willem de Bruijn;
Changes in v3:
- return HRTIMER_NORESTART when pkc->delete_blk_timer is true
  as suggested by Willem de Bruijn;
- Drop the retire_blk_tov field of tpacket_kbdq_core, add interval_ktime instead
  as suggested by Willem de Bruijn;
- Add comments to explain why hrtimer_set_expires(not hrtimer_forward_now) is used in
  _prb_refresh_rx_retire_blk_timer
  as suggested by Willem de Bruijn;
- Link to v3: https://lore.kernel.org/all/20250816170130.3969354-1-jackzxcui1989@163.com/

Changes in v2:
- Drop the tov_in_msecs field of tpacket_kbdq_core added by the patch
  as suggested by Willem de Bruijn;
- Link to v2: https://lore.kernel.org/all/20250815044141.1374446-1-jackzxcui1989@163.com/

Changes in v1:
- Do not add another config for the current changes
  as suggested by Eric Dumazet;
- Mention the beneficial cases 'HZ=100 or HZ=250' in the changelog
  as suggested by Eric Dumazet;
- Add some performance details to the changelog
  as suggested by Ferenc Fejes;
- Delete the 'pkc->tov_in_msecs == 0' bounds check which is not necessary
  as suggested by Willem de Bruijn;
- Use hrtimer_set_expires instead of hrtimer_start_range_ns when retire timer needs update
  as suggested by Willem de Bruijn. Start the hrtimer in prb_setup_retire_blk_timer;
- Just return HRTIMER_RESTART directly as all cases return the same value
  as suggested by Willem de Bruijn;
- Link to v1: https://lore.kernel.org/all/20250813165201.1492779-1-jackzxcui1989@163.com/
- Link to v0: https://lore.kernel.org/all/20250806055210.1530081-1-jackzxcui1989@163.com/

---
---
 net/packet/af_packet.c | 52 ++++++++++++++++++++++++++++--------------
 net/packet/diag.c      |  2 +-
 net/packet/internal.h  |  5 ++--
 3 files changed, 38 insertions(+), 21 deletions(-)

diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
index a7017d7f0..5a1e80185 100644
--- a/net/packet/af_packet.c
+++ b/net/packet/af_packet.c
@@ -203,8 +203,8 @@ static void prb_retire_current_block(struct tpacket_kbdq_core *,
 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 static void prb_open_block(struct tpacket_kbdq_core *,
 		struct tpacket_block_desc *);
-static void prb_retire_rx_blk_timer_expired(struct timer_list *);
-static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
+static enum hrtimer_restart prb_retire_rx_blk_timer_expired(struct hrtimer *);
+static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *, bool);
 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 		struct tpacket3_hdr *);
@@ -581,7 +581,7 @@ static __be16 vlan_get_protocol_dgram(const struct sk_buff *skb)
 
 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 {
-	timer_delete_sync(&pkc->retire_blk_timer);
+	hrtimer_cancel(&pkc->retire_blk_timer);
 }
 
 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
@@ -603,9 +603,8 @@ static void prb_setup_retire_blk_timer(struct packet_sock *po)
 	struct tpacket_kbdq_core *pkc;
 
 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
-	timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
-		    0);
-	pkc->retire_blk_timer.expires = jiffies;
+	hrtimer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
+		      CLOCK_MONOTONIC, HRTIMER_MODE_REL_SOFT);
 }
 
 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
@@ -672,11 +671,10 @@ static void init_prb_bdqc(struct packet_sock *po,
 	p1->last_kactive_blk_num = 0;
 	po->stats.stats3.tp_freeze_q_cnt = 0;
 	if (req_u->req3.tp_retire_blk_tov)
-		p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
+		p1->interval_ktime = ms_to_ktime(req_u->req3.tp_retire_blk_tov);
 	else
-		p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
-						req_u->req3.tp_block_size);
-	p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
+		p1->interval_ktime = ms_to_ktime(prb_calc_retire_blk_tmo(po,
+						req_u->req3.tp_block_size));
 	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 	rwlock_init(&p1->blk_fill_in_prog_lock);
 
@@ -688,11 +686,27 @@ static void init_prb_bdqc(struct packet_sock *po,
 
 /*  Do NOT update the last_blk_num first.
  *  Assumes sk_buff_head lock is held.
+ *  We only need to (re)start an hrtimer in prb_open_block.
+ *  Otherwise, we just need to update the expiration time of the hrtimer.
  */
-static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
+static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc,
+		bool start)
 {
-	mod_timer(&pkc->retire_blk_timer,
-			jiffies + pkc->tov_in_jiffies);
+	if (start)
+		hrtimer_start(&pkc->retire_blk_timer, pkc->interval_ktime,
+			      HRTIMER_MODE_REL_SOFT);
+	else
+		/* We cannot use hrtimer_forward_now here because the function
+		 * _prb_refresh_rx_retire_blk_timer can be called not only when
+		 * the retire timer expires, but also when the kernel logic for
+		 * receiving network packets detects that a network packet has
+		 * filled up a block and calls prb_open_block to use the next
+		 * block. This can lead to a WARN_ON being triggered in
+		 * hrtimer_forward_now when it checks if the timer has already
+		 * been enqueued.
+		 */
+		hrtimer_set_expires(&pkc->retire_blk_timer,
+				    ktime_add(ktime_get(), pkc->interval_ktime));
 	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 }
 
@@ -719,8 +733,9 @@ static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
  * prb_calc_retire_blk_tmo() calculates the tmo.
  *
  */
-static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
+static enum hrtimer_restart prb_retire_rx_blk_timer_expired(struct hrtimer *t)
 {
+	enum hrtimer_restart ret = HRTIMER_RESTART;
 	struct packet_sock *po =
 		timer_container_of(po, t, rx_ring.prb_bdqc.retire_blk_timer);
 	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
@@ -732,8 +747,10 @@ static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
 	frozen = prb_queue_frozen(pkc);
 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 
-	if (unlikely(pkc->delete_blk_timer))
+	if (unlikely(pkc->delete_blk_timer)) {
+		ret = HRTIMER_NORESTART;
 		goto out;
+	}
 
 	/* We only need to plug the race when the block is partially filled.
 	 * tpacket_rcv:
@@ -786,10 +803,11 @@ static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
 	}
 
 refresh_timer:
-	_prb_refresh_rx_retire_blk_timer(pkc);
+	_prb_refresh_rx_retire_blk_timer(pkc, false);
 
 out:
 	spin_unlock(&po->sk.sk_receive_queue.lock);
+	return ret;
 }
 
 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
@@ -921,7 +939,7 @@ static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 
 	prb_thaw_queue(pkc1);
-	_prb_refresh_rx_retire_blk_timer(pkc1);
+	_prb_refresh_rx_retire_blk_timer(pkc1, true);
 
 	smp_wmb();
 }
diff --git a/net/packet/diag.c b/net/packet/diag.c
index 6ce1dcc28..c8f43e0c1 100644
--- a/net/packet/diag.c
+++ b/net/packet/diag.c
@@ -83,7 +83,7 @@ static int pdiag_put_ring(struct packet_ring_buffer *ring, int ver, int nl_type,
 	pdr.pdr_frame_nr = ring->frame_max + 1;
 
 	if (ver > TPACKET_V2) {
-		pdr.pdr_retire_tmo = ring->prb_bdqc.retire_blk_tov;
+		pdr.pdr_retire_tmo = ktime_to_ms(ring->prb_bdqc.interval_ktime);
 		pdr.pdr_sizeof_priv = ring->prb_bdqc.blk_sizeof_priv;
 		pdr.pdr_features = ring->prb_bdqc.feature_req_word;
 	} else {
diff --git a/net/packet/internal.h b/net/packet/internal.h
index 1e743d031..19d4f0b73 100644
--- a/net/packet/internal.h
+++ b/net/packet/internal.h
@@ -45,12 +45,11 @@ struct tpacket_kbdq_core {
 	/* Default is set to 8ms */
 #define DEFAULT_PRB_RETIRE_TOV	(8)
 
-	unsigned short  retire_blk_tov;
+	ktime_t		interval_ktime;
 	unsigned short  version;
-	unsigned long	tov_in_jiffies;
 
 	/* timer to retire an outstanding block */
-	struct timer_list retire_blk_timer;
+	struct hrtimer  retire_blk_timer;
 };
 
 struct pgv {
-- 
2.34.1

[PATCH net-next v4] net: af_packet: Use hrtimer to do the retire operation

[Xin Zhao]

In a system with high real-time requirements, the timeout mechanism of
ordinary timers with jiffies granularity is insufficient to meet the
demands for real-time performance. Meanwhile, the optimization of CPU
usage with af_packet is quite significant. Use hrtimer instead of timer
to help compensate for the shortcomings in real-time performance.
In HZ=100 or HZ=250 system, the update of TP_STATUS_USER is not real-time
enough, with fluctuations reaching over 8ms (on a system with HZ=250).
This is unacceptable in some high real-time systems that require timely
processing of network packets. By replacing it with hrtimer, if a timeout
of 2ms is set, the update of TP_STATUS_USER can be stabilized to within
3 ms.

Signed-off-by: Xin Zhao <jackzxcui1989@163.com>

---
Changes in v4:
- Add 'bool start' to distinguish whether the call to _prb_refresh_rx_retire_blk_timer
  is for prb_open_block. When it is for prb_open_block, execute hrtimer_start to
  (re)start the hrtimer; otherwise, use hrtimer_set_expires to update the expiration
  time of the hrtimer
  as suggested by Willem de Bruijn;

Changes in v3:
- return HRTIMER_NORESTART when pkc->delete_blk_timer is true
  as suggested by Willem de Bruijn;
- Drop the retire_blk_tov field of tpacket_kbdq_core, add interval_ktime instead
  as suggested by Willem de Bruijn;
- Add comments to explain why hrtimer_set_expires(not hrtimer_forward_now) is used in
  _prb_refresh_rx_retire_blk_timer
  as suggested by Willem de Bruijn;
- Link to v3: https://lore.kernel.org/all/20250816170130.3969354-1-jackzxcui1989@163.com/

Changes in v2:
- Drop the tov_in_msecs field of tpacket_kbdq_core added by the patch
  as suggested by Willem de Bruijn;
- Link to v2: https://lore.kernel.org/all/20250815044141.1374446-1-jackzxcui1989@163.com/

Changes in v1:
- Do not add another config for the current changes
  as suggested by Eric Dumazet;
- Mention the beneficial cases 'HZ=100 or HZ=250' in the changelog
  as suggested by Eric Dumazet;
- Add some performance details to the changelog
  as suggested by Ferenc Fejes;
- Delete the 'pkc->tov_in_msecs == 0' bounds check which is not necessary
  as suggested by Willem de Bruijn;
- Use hrtimer_set_expires instead of hrtimer_start_range_ns when retire timer needs update
  as suggested by Willem de Bruijn. Start the hrtimer in prb_setup_retire_blk_timer;
- Just return HRTIMER_RESTART directly as all cases return the same value
  as suggested by Willem de Bruijn;
- Link to v1: https://lore.kernel.org/all/20250813165201.1492779-1-jackzxcui1989@163.com/
- Link to v0: https://lore.kernel.org/all/20250806055210.1530081-1-jackzxcui1989@163.com/
---
 net/packet/af_packet.c | 52 ++++++++++++++++++++++++++++--------------
 net/packet/diag.c      |  2 +-
 net/packet/internal.h  |  5 ++--
 3 files changed, 38 insertions(+), 21 deletions(-)

diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
index a7017d7f0..5a1e80185 100644
--- a/net/packet/af_packet.c
+++ b/net/packet/af_packet.c
@@ -203,8 +203,8 @@ static void prb_retire_current_block(struct tpacket_kbdq_core *,
 static int prb_queue_frozen(struct tpacket_kbdq_core *);
 static void prb_open_block(struct tpacket_kbdq_core *,
 		struct tpacket_block_desc *);
-static void prb_retire_rx_blk_timer_expired(struct timer_list *);
-static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
+static enum hrtimer_restart prb_retire_rx_blk_timer_expired(struct hrtimer *);
+static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *, bool);
 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
 		struct tpacket3_hdr *);
@@ -581,7 +581,7 @@ static __be16 vlan_get_protocol_dgram(const struct sk_buff *skb)
 
 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
 {
-	timer_delete_sync(&pkc->retire_blk_timer);
+	hrtimer_cancel(&pkc->retire_blk_timer);
 }
 
 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
@@ -603,9 +603,8 @@ static void prb_setup_retire_blk_timer(struct packet_sock *po)
 	struct tpacket_kbdq_core *pkc;
 
 	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
-	timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
-		    0);
-	pkc->retire_blk_timer.expires = jiffies;
+	hrtimer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
+		      CLOCK_MONOTONIC, HRTIMER_MODE_REL_SOFT);
 }
 
 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
@@ -672,11 +671,10 @@ static void init_prb_bdqc(struct packet_sock *po,
 	p1->last_kactive_blk_num = 0;
 	po->stats.stats3.tp_freeze_q_cnt = 0;
 	if (req_u->req3.tp_retire_blk_tov)
-		p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
+		p1->interval_ktime = ms_to_ktime(req_u->req3.tp_retire_blk_tov);
 	else
-		p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
-						req_u->req3.tp_block_size);
-	p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
+		p1->interval_ktime = ms_to_ktime(prb_calc_retire_blk_tmo(po,
+						req_u->req3.tp_block_size));
 	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
 	rwlock_init(&p1->blk_fill_in_prog_lock);
 
@@ -688,11 +686,27 @@ static void init_prb_bdqc(struct packet_sock *po,
 
 /*  Do NOT update the last_blk_num first.
  *  Assumes sk_buff_head lock is held.
+ *  We only need to (re)start an hrtimer in prb_open_block.
+ *  Otherwise, we just need to update the expiration time of the hrtimer.
  */
-static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
+static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc,
+		bool start)
 {
-	mod_timer(&pkc->retire_blk_timer,
-			jiffies + pkc->tov_in_jiffies);
+	if (start)
+		hrtimer_start(&pkc->retire_blk_timer, pkc->interval_ktime,
+			      HRTIMER_MODE_REL_SOFT);
+	else
+		/* We cannot use hrtimer_forward_now here because the function
+		 * _prb_refresh_rx_retire_blk_timer can be called not only when
+		 * the retire timer expires, but also when the kernel logic for
+		 * receiving network packets detects that a network packet has
+		 * filled up a block and calls prb_open_block to use the next
+		 * block. This can lead to a WARN_ON being triggered in
+		 * hrtimer_forward_now when it checks if the timer has already
+		 * been enqueued.
+		 */
+		hrtimer_set_expires(&pkc->retire_blk_timer,
+				    ktime_add(ktime_get(), pkc->interval_ktime));
 	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
 }
 
@@ -719,8 +733,9 @@ static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
  * prb_calc_retire_blk_tmo() calculates the tmo.
  *
  */
-static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
+static enum hrtimer_restart prb_retire_rx_blk_timer_expired(struct hrtimer *t)
 {
+	enum hrtimer_restart ret = HRTIMER_RESTART;
 	struct packet_sock *po =
 		timer_container_of(po, t, rx_ring.prb_bdqc.retire_blk_timer);
 	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
@@ -732,8 +747,10 @@ static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
 	frozen = prb_queue_frozen(pkc);
 	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
 
-	if (unlikely(pkc->delete_blk_timer))
+	if (unlikely(pkc->delete_blk_timer)) {
+		ret = HRTIMER_NORESTART;
 		goto out;
+	}
 
 	/* We only need to plug the race when the block is partially filled.
 	 * tpacket_rcv:
@@ -786,10 +803,11 @@ static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
 	}
 
 refresh_timer:
-	_prb_refresh_rx_retire_blk_timer(pkc);
+	_prb_refresh_rx_retire_blk_timer(pkc, false);
 
 out:
 	spin_unlock(&po->sk.sk_receive_queue.lock);
+	return ret;
 }
 
 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
@@ -921,7 +939,7 @@ static void prb_open_block(struct tpacket_kbdq_core *pkc1,
 	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
 
 	prb_thaw_queue(pkc1);
-	_prb_refresh_rx_retire_blk_timer(pkc1);
+	_prb_refresh_rx_retire_blk_timer(pkc1, true);
 
 	smp_wmb();
 }
diff --git a/net/packet/diag.c b/net/packet/diag.c
index 6ce1dcc28..c8f43e0c1 100644
--- a/net/packet/diag.c
+++ b/net/packet/diag.c
@@ -83,7 +83,7 @@ static int pdiag_put_ring(struct packet_ring_buffer *ring, int ver, int nl_type,
 	pdr.pdr_frame_nr = ring->frame_max + 1;
 
 	if (ver > TPACKET_V2) {
-		pdr.pdr_retire_tmo = ring->prb_bdqc.retire_blk_tov;
+		pdr.pdr_retire_tmo = ktime_to_ms(ring->prb_bdqc.interval_ktime);
 		pdr.pdr_sizeof_priv = ring->prb_bdqc.blk_sizeof_priv;
 		pdr.pdr_features = ring->prb_bdqc.feature_req_word;
 	} else {
diff --git a/net/packet/internal.h b/net/packet/internal.h
index 1e743d031..19d4f0b73 100644
--- a/net/packet/internal.h
+++ b/net/packet/internal.h
@@ -45,12 +45,11 @@ struct tpacket_kbdq_core {
 	/* Default is set to 8ms */
 #define DEFAULT_PRB_RETIRE_TOV	(8)
 
-	unsigned short  retire_blk_tov;
+	ktime_t		interval_ktime;
 	unsigned short  version;
-	unsigned long	tov_in_jiffies;
 
 	/* timer to retire an outstanding block */
-	struct timer_list retire_blk_timer;
+	struct hrtimer  retire_blk_timer;
 };
 
 struct pgv {
-- 
2.34.1

Re: [PATCH net-next v4] net: af_packet: Use hrtimer to do the retire operation

[Willem de Bruijn]

Xin Zhao wrote:
> In a system with high real-time requirements, the timeout mechanism of
> ordinary timers with jiffies granularity is insufficient to meet the
> demands for real-time performance. Meanwhile, the optimization of CPU
> usage with af_packet is quite significant. Use hrtimer instead of timer
> to help compensate for the shortcomings in real-time performance.
> In HZ=100 or HZ=250 system, the update of TP_STATUS_USER is not real-time
> enough, with fluctuations reaching over 8ms (on a system with HZ=250).
> This is unacceptable in some high real-time systems that require timely
> processing of network packets. By replacing it with hrtimer, if a timeout
> of 2ms is set, the update of TP_STATUS_USER can be stabilized to within
> 3 ms.
> 
> Signed-off-by: Xin Zhao <jackzxcui1989@163.com>
> 
> ---
> Changes in v4:
> - Add 'bool start' to distinguish whether the call to _prb_refresh_rx_retire_blk_timer
>   is for prb_open_block. When it is for prb_open_block, execute hrtimer_start to
>   (re)start the hrtimer; otherwise, use hrtimer_set_expires to update the expiration
>   time of the hrtimer
>   as suggested by Willem de Bruijn;
> 
> Changes in v3:
> - return HRTIMER_NORESTART when pkc->delete_blk_timer is true
>   as suggested by Willem de Bruijn;
> - Drop the retire_blk_tov field of tpacket_kbdq_core, add interval_ktime instead
>   as suggested by Willem de Bruijn;
> - Add comments to explain why hrtimer_set_expires(not hrtimer_forward_now) is used in
>   _prb_refresh_rx_retire_blk_timer
>   as suggested by Willem de Bruijn;
> - Link to v3: https://lore.kernel.org/all/20250816170130.3969354-1-jackzxcui1989@163.com/
> 
> Changes in v2:
> - Drop the tov_in_msecs field of tpacket_kbdq_core added by the patch
>   as suggested by Willem de Bruijn;
> - Link to v2: https://lore.kernel.org/all/20250815044141.1374446-1-jackzxcui1989@163.com/
> 
> Changes in v1:
> - Do not add another config for the current changes
>   as suggested by Eric Dumazet;
> - Mention the beneficial cases 'HZ=100 or HZ=250' in the changelog
>   as suggested by Eric Dumazet;
> - Add some performance details to the changelog
>   as suggested by Ferenc Fejes;
> - Delete the 'pkc->tov_in_msecs == 0' bounds check which is not necessary
>   as suggested by Willem de Bruijn;
> - Use hrtimer_set_expires instead of hrtimer_start_range_ns when retire timer needs update
>   as suggested by Willem de Bruijn. Start the hrtimer in prb_setup_retire_blk_timer;
> - Just return HRTIMER_RESTART directly as all cases return the same value
>   as suggested by Willem de Bruijn;
> - Link to v1: https://lore.kernel.org/all/20250813165201.1492779-1-jackzxcui1989@163.com/
> - Link to v0: https://lore.kernel.org/all/20250806055210.1530081-1-jackzxcui1989@163.com/
> ---
>  net/packet/af_packet.c | 52 ++++++++++++++++++++++++++++--------------
>  net/packet/diag.c      |  2 +-
>  net/packet/internal.h  |  5 ++--
>  3 files changed, 38 insertions(+), 21 deletions(-)
> 
> diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
> index a7017d7f0..5a1e80185 100644
> --- a/net/packet/af_packet.c
> +++ b/net/packet/af_packet.c
> @@ -203,8 +203,8 @@ static void prb_retire_current_block(struct tpacket_kbdq_core *,
>  static int prb_queue_frozen(struct tpacket_kbdq_core *);
>  static void prb_open_block(struct tpacket_kbdq_core *,
>  		struct tpacket_block_desc *);
> -static void prb_retire_rx_blk_timer_expired(struct timer_list *);
> -static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
> +static enum hrtimer_restart prb_retire_rx_blk_timer_expired(struct hrtimer *);
> +static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *, bool);
>  static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
>  static void prb_clear_rxhash(struct tpacket_kbdq_core *,
>  		struct tpacket3_hdr *);
> @@ -581,7 +581,7 @@ static __be16 vlan_get_protocol_dgram(const struct sk_buff *skb)
>  
>  static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
>  {
> -	timer_delete_sync(&pkc->retire_blk_timer);
> +	hrtimer_cancel(&pkc->retire_blk_timer);
>  }
>  
>  static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
> @@ -603,9 +603,8 @@ static void prb_setup_retire_blk_timer(struct packet_sock *po)
>  	struct tpacket_kbdq_core *pkc;
>  
>  	pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
> -	timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
> -		    0);
> -	pkc->retire_blk_timer.expires = jiffies;
> +	hrtimer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
> +		      CLOCK_MONOTONIC, HRTIMER_MODE_REL_SOFT);
>  }
>  
>  static int prb_calc_retire_blk_tmo(struct packet_sock *po,
> @@ -672,11 +671,10 @@ static void init_prb_bdqc(struct packet_sock *po,
>  	p1->last_kactive_blk_num = 0;
>  	po->stats.stats3.tp_freeze_q_cnt = 0;
>  	if (req_u->req3.tp_retire_blk_tov)
> -		p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
> +		p1->interval_ktime = ms_to_ktime(req_u->req3.tp_retire_blk_tov);
>  	else
> -		p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
> -						req_u->req3.tp_block_size);
> -	p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
> +		p1->interval_ktime = ms_to_ktime(prb_calc_retire_blk_tmo(po,
> +						req_u->req3.tp_block_size));
>  	p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
>  	rwlock_init(&p1->blk_fill_in_prog_lock);
>  
> @@ -688,11 +686,27 @@ static void init_prb_bdqc(struct packet_sock *po,
>  
>  /*  Do NOT update the last_blk_num first.
>   *  Assumes sk_buff_head lock is held.
> + *  We only need to (re)start an hrtimer in prb_open_block.
> + *  Otherwise, we just need to update the expiration time of the hrtimer.

"We" don't do anything in the middle of a computation. Anyway, branch is
self explanatory enough, can drop comment.

>   */
> -static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
> +static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc,
> +		bool start)

Indentation, align with first argument on previous line

>  {
> -	mod_timer(&pkc->retire_blk_timer,
> -			jiffies + pkc->tov_in_jiffies);
> +	if (start)
> +		hrtimer_start(&pkc->retire_blk_timer, pkc->interval_ktime,
> +			      HRTIMER_MODE_REL_SOFT);

It's okay to call this from inside a timer callback itself and return
HRTIMER_RESTART? I don't know off the top of my head.

> +	else
> +		/* We cannot use hrtimer_forward_now here because the function
> +		 * _prb_refresh_rx_retire_blk_timer can be called not only when
> +		 * the retire timer expires, but also when the kernel logic for
> +		 * receiving network packets detects that a network packet has
> +		 * filled up a block and calls prb_open_block to use the next
> +		 * block. This can lead to a WARN_ON being triggered in
> +		 * hrtimer_forward_now when it checks if the timer has already
> +		 * been enqueued.
> +		 */

As discussed, this will be changed in v5.

> +		hrtimer_set_expires(&pkc->retire_blk_timer,
> +				    ktime_add(ktime_get(), pkc->interval_ktime));
>  	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
>  }
>  
> @@ -719,8 +733,9 @@ static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
>   * prb_calc_retire_blk_tmo() calculates the tmo.
>   *
>   */
> -static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
> +static enum hrtimer_restart prb_retire_rx_blk_timer_expired(struct hrtimer *t)
>  {
> +	enum hrtimer_restart ret = HRTIMER_RESTART;
>  	struct packet_sock *po =
>  		timer_container_of(po, t, rx_ring.prb_bdqc.retire_blk_timer);
>  	struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
> @@ -732,8 +747,10 @@ static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
>  	frozen = prb_queue_frozen(pkc);
>  	pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
>  
> -	if (unlikely(pkc->delete_blk_timer))
> +	if (unlikely(pkc->delete_blk_timer)) {
> +		ret = HRTIMER_NORESTART;
>  		goto out;
> +	}
>  
>  	/* We only need to plug the race when the block is partially filled.
>  	 * tpacket_rcv:
> @@ -786,10 +803,11 @@ static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
>  	}
>  
>  refresh_timer:
> -	_prb_refresh_rx_retire_blk_timer(pkc);
> +	_prb_refresh_rx_retire_blk_timer(pkc, false);
>  
>  out:
>  	spin_unlock(&po->sk.sk_receive_queue.lock);
> +	return ret;
>  }
>  
>  static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
> @@ -921,7 +939,7 @@ static void prb_open_block(struct tpacket_kbdq_core *pkc1,
>  	pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
>  
>  	prb_thaw_queue(pkc1);
> -	_prb_refresh_rx_retire_blk_timer(pkc1);
> +	_prb_refresh_rx_retire_blk_timer(pkc1, true);
>  
>  	smp_wmb();
>  }
> diff --git a/net/packet/diag.c b/net/packet/diag.c
> index 6ce1dcc28..c8f43e0c1 100644
> --- a/net/packet/diag.c
> +++ b/net/packet/diag.c
> @@ -83,7 +83,7 @@ static int pdiag_put_ring(struct packet_ring_buffer *ring, int ver, int nl_type,
>  	pdr.pdr_frame_nr = ring->frame_max + 1;
>  
>  	if (ver > TPACKET_V2) {
> -		pdr.pdr_retire_tmo = ring->prb_bdqc.retire_blk_tov;
> +		pdr.pdr_retire_tmo = ktime_to_ms(ring->prb_bdqc.interval_ktime);
>  		pdr.pdr_sizeof_priv = ring->prb_bdqc.blk_sizeof_priv;
>  		pdr.pdr_features = ring->prb_bdqc.feature_req_word;
>  	} else {
> diff --git a/net/packet/internal.h b/net/packet/internal.h
> index 1e743d031..19d4f0b73 100644
> --- a/net/packet/internal.h
> +++ b/net/packet/internal.h
> @@ -45,12 +45,11 @@ struct tpacket_kbdq_core {
>  	/* Default is set to 8ms */
>  #define DEFAULT_PRB_RETIRE_TOV	(8)
>  
> -	unsigned short  retire_blk_tov;
> +	ktime_t		interval_ktime;
>  	unsigned short  version;
> -	unsigned long	tov_in_jiffies;
>  
>  	/* timer to retire an outstanding block */
> -	struct timer_list retire_blk_timer;
> +	struct hrtimer  retire_blk_timer;
>  };
>  
>  struct pgv {
> -- 
> 2.34.1
> 

Re: [PATCH net-next v4] net: af_packet: Use hrtimer to do the retire operation

[Xin Zhao]

On Mon, 2025-08-18 at 17:29 +0800, Willem wrote:

> "We" don't do anything in the middle of a computation. Anyway, branch is
> self explanatory enough, can drop comment.
> 
> >   */
> > -static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
> > +static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc,
> > +		bool start)
> 
> Indentation, align with first argument on previous line


> > +	else
> > +		/* We cannot use hrtimer_forward_now here because the function
> > +		 * _prb_refresh_rx_retire_blk_timer can be called not only when
> > +		 * the retire timer expires, but also when the kernel logic for
> > +		 * receiving network packets detects that a network packet has
> > +		 * filled up a block and calls prb_open_block to use the next
> > +		 * block. This can lead to a WARN_ON being triggered in
> > +		 * hrtimer_forward_now when it checks if the timer has already
> > +		 * been enqueued.
> > +		 */
> 
> As discussed, this will be changed in v5.


I will change them in v5. And I will ensure that there is a 24-hour send gap between
each patch.


> >  {
> > -	mod_timer(&pkc->retire_blk_timer,
> > -			jiffies + pkc->tov_in_jiffies);
> > +	if (start)
> > +		hrtimer_start(&pkc->retire_blk_timer, pkc->interval_ktime,
> > +			      HRTIMER_MODE_REL_SOFT);
> 
> It's okay to call this from inside a timer callback itself and return
> HRTIMER_RESTART? I don't know off the top of my head.

Although I have been using hrtimer_start to restart the timer within the callback in
our project and seem to work weill, I found that it seems no one does this in the
current mainline kernel code. Therefore, I will add a boolean parameter to the
callback in version 5 to indicate whether it is within the callback function. If it is
in the callback function, I will use hrtimer_forward_now instead of hrtimer_start.
Additionally, while looking at the historical Git logs of hrtimer, I noticed that it is
possible to call hrtimer_start to start the hrtimer outside of the hrtimer callback, but
it requires the protection of raw_spin_lock_irqsave. When entering the
_prb_refresh_rx_retire_blk_timer function, as noted in the comments, there is already
protection with the sk_buff_head lock, so I only need to add a set of irq save and restore
operations. The reason for this is based on the reference from link
https://lore.kernel.org/all/20150415113105.GT5029@twins.programming.kicks-ass.net/T/#u and
the implementation of the perf_mux_hrtimer_restart function.

The implementation of the _prb_refresh_rx_retire_blk_timer function in PATCH v5:

/*  Do NOT update the last_blk_num first.
 *  Assumes sk_buff_head lock is held.
 */
static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc,
					     bool start, bool callback)
{
	unsigned long flags;

	local_irq_save(flags);
	if (start && !callback)
		hrtimer_start(&pkc->retire_blk_timer, pkc->interval_ktime,
			      HRTIMER_MODE_REL_SOFT);
	else
		hrtimer_forward_now(&pkc->retire_blk_timer, pkc->interval_ktime);
	local_irq_restore(flags);
	pkc->last_kactive_blk_num = pkc->kactive_blk_num;
}


Thanks
Xin Zhao

Re: [PATCH net-next v4] net: af_packet: Use hrtimer to do the retire operation

[Xin Zhao]

On Tue, 2025-08-19 at 22:18 +0800, Willem wrote:

> > -static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
> > +static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc,
> > +					     bool start, bool callback)
> >  {
> > -	mod_timer(&pkc->retire_blk_timer,
> > -			jiffies + pkc->tov_in_jiffies);
> > +	unsigned long flags;
> > +
> > +	local_irq_save(flags);
> 
> The two environments that can race are the timer callback running in
> softirq context or the open_block from tpacket_rcv in process context.
> 
> So worst case the process context path needs to disable bh?
> 
> As you pointed out, the accesses to the hrtimer fields are already
> protected, by the caller holding sk.sk_receive_queue.lock.
> 
> So it should be sufficient to just test hrtimer_is_queued inside that
> critical section before calling hrtimer_start?
> 
> Side-note: tpacket_rcv calls spin_lock, not spin_lock_bh. But if the
> same lock can also be taken in softirq context, the process context
> caller should use the _bh variant. This is not new with your patch.
> Classical timers also run in softirq context. I may be overlooking
> something, will need to take a closer look at that.
> 
> In any case, I don't think local_irq_save is needed. 

Indeed, there is no need to use local_irq_save. The use case I referenced from
perf_mux_hrtimer_restart is different from ours. Our timer callback does not run in
hard interrupt context, so it is unnecessary to use local_irq_save. I will make this
change in PATCH v6.



On Wed, 2025-08-20 at 4:21 +0800, Willem wrote:
 
> > So worst case the process context path needs to disable bh?
> > 
> > As you pointed out, the accesses to the hrtimer fields are already
> > protected, by the caller holding sk.sk_receive_queue.lock.
> > 
> > So it should be sufficient to just test hrtimer_is_queued inside that
> > critical section before calling hrtimer_start?
> > 
> > Side-note: tpacket_rcv calls spin_lock, not spin_lock_bh. But if the
> > same lock can also be taken in softirq context, the process context
> > caller should use the _bh variant. This is not new with your patch.
> > Classical timers also run in softirq context. I may be overlooking
> > something, will need to take a closer look at that.
> > 
> > In any case, I don't think local_irq_save is needed. 
> 
> 
> 
> 
> I meant prb_open_block
> 
> tpacket_rcv runs in softirq context (from __netif_receive_skb_core)
> or with bottom halves disabled (from __dev_queue_xmit, or if rx uses
> napi_threaded).
> 
> That is likely why the spin_lock_bh variant is not explicitly needed.

Before I saw your reply, I was almost considering replacing spin_lock with
spin_lock_bh in our project before calling packet_current_rx_frame in
tpacket_rcv. I just couldn't understand why we haven't encountered any
deadlocks or RCU issues due to not properly adding _bh in our project until
I saw your reply.
I truly admire your ability to identify all the scenarios that use the
tpacket_rcv function in such a short amount of time. For me, finding all the
instances where tpacket_rcv is assigned to prot_hook.func for proxy calls is
a painful and lengthy task. Even if I manage to find them, I would still
worry about missing some.


Thanks
Xin Zhao

Re: [PATCH net-next v4] net: af_packet: Use hrtimer to do the retire operation

[Willem de Bruijn]

Xin Zhao wrote:
> On Tue, 2025-08-19 at 22:18 +0800, Willem wrote:
> 
> > > -static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
> > > +static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc,
> > > +					     bool start, bool callback)
> > >  {
> > > -	mod_timer(&pkc->retire_blk_timer,
> > > -			jiffies + pkc->tov_in_jiffies);
> > > +	unsigned long flags;
> > > +
> > > +	local_irq_save(flags);
> > 
> > The two environments that can race are the timer callback running in
> > softirq context or the open_block from tpacket_rcv in process context.
> > 
> > So worst case the process context path needs to disable bh?
> > 
> > As you pointed out, the accesses to the hrtimer fields are already
> > protected, by the caller holding sk.sk_receive_queue.lock.
> > 
> > So it should be sufficient to just test hrtimer_is_queued inside that
> > critical section before calling hrtimer_start?
> > 
> > Side-note: tpacket_rcv calls spin_lock, not spin_lock_bh. But if the
> > same lock can also be taken in softirq context, the process context
> > caller should use the _bh variant. This is not new with your patch.
> > Classical timers also run in softirq context. I may be overlooking
> > something, will need to take a closer look at that.
> > 
> > In any case, I don't think local_irq_save is needed. 
> 
> Indeed, there is no need to use local_irq_save. The use case I referenced from
> perf_mux_hrtimer_restart is different from ours. Our timer callback does not run in
> hard interrupt context, so it is unnecessary to use local_irq_save. I will make this
> change in PATCH v6.
> 
> 
> 
> On Wed, 2025-08-20 at 4:21 +0800, Willem wrote:
>  
> > > So worst case the process context path needs to disable bh?
> > > 
> > > As you pointed out, the accesses to the hrtimer fields are already
> > > protected, by the caller holding sk.sk_receive_queue.lock.
> > > 
> > > So it should be sufficient to just test hrtimer_is_queued inside that
> > > critical section before calling hrtimer_start?
> > > 
> > > Side-note: tpacket_rcv calls spin_lock, not spin_lock_bh. But if the
> > > same lock can also be taken in softirq context, the process context
> > > caller should use the _bh variant. This is not new with your patch.
> > > Classical timers also run in softirq context. I may be overlooking
> > > something, will need to take a closer look at that.
> > > 
> > > In any case, I don't think local_irq_save is needed. 
> > 
> > 
> > 
> > 
> > I meant prb_open_block
> > 
> > tpacket_rcv runs in softirq context (from __netif_receive_skb_core)
> > or with bottom halves disabled (from __dev_queue_xmit, or if rx uses
> > napi_threaded).
> > 
> > That is likely why the spin_lock_bh variant is not explicitly needed.
> 
> Before I saw your reply, I was almost considering replacing spin_lock with
> spin_lock_bh in our project before calling packet_current_rx_frame in
> tpacket_rcv. I just couldn't understand why we haven't encountered any
> deadlocks or RCU issues due to not properly adding _bh in our project until
> I saw your reply.
> I truly admire your ability to identify all the scenarios that use the
> tpacket_rcv function in such a short amount of time. For me, finding all the
> instances where tpacket_rcv is assigned to prot_hook.func for proxy calls is
> a painful and lengthy task. Even if I manage to find them, I would still
> worry about missing some.

Thanks. I also reasoned backwards. If there had been a problem,
lockdep would have reported it long ago.