[net-next,35/37] rxrpc: Manage RTT per-call rather than per-peer

Message ID	20241202143057.378147-36-dhowells@redhat.com (mailing list archive)
State	New
Delegated to:	Netdev Maintainers
Headers	show Received: from us-smtp-delivery-124.mimecast.com (us-smtp-delivery-124.mimecast.com [170.10.129.124]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id B46F7209669 for <netdev@vger.kernel.org>; Mon, 2 Dec 2024 14:33:35 +0000 (UTC) From: David Howells <dhowells@redhat.com> To: netdev@vger.kernel.org Cc: David Howells <dhowells@redhat.com>, Marc Dionne <marc.dionne@auristor.com>, Yunsheng Lin <linyunsheng@huawei.com>, "David S. Miller" <davem@davemloft.net>, Eric Dumazet <edumazet@google.com>, Jakub Kicinski <kuba@kernel.org>, Paolo Abeni <pabeni@redhat.com>, linux-afs@lists.infradead.org, linux-kernel@vger.kernel.org Subject: [PATCH net-next 35/37] rxrpc: Manage RTT per-call rather than per-peer Date: Mon, 2 Dec 2024 14:30:53 +0000 Message-ID: <20241202143057.378147-36-dhowells@redhat.com> In-Reply-To: <20241202143057.378147-1-dhowells@redhat.com> References: <20241202143057.378147-1-dhowells@redhat.com> Precedence: bulk MIME-Version: 1.0 Content-Transfer-Encoding: 8bit
Series	rxrpc: Implement jumbo DATA transmission and RACK-TLP \| expand [net-next,00/37] rxrpc: Implement jumbo DATA transmission and RACK-TLP [net-next,01/37] rxrpc: Fix handling of received connection abort [net-next,02/37] rxrpc: Use umin() and umax() rather than min_t()/max_t() where possible [net-next,03/37] rxrpc: Clean up Tx header flags generation handling [net-next,04/37] rxrpc: Don't set the MORE-PACKETS rxrpc wire header flag [net-next,05/37] rxrpc: Show stats counter for received reason-0 ACKs [net-next,06/37] rxrpc: Request an ACK on impending Tx stall [net-next,07/37] rxrpc: Use a large kvec[] in rxrpc_local rather than every rxrpc_txbuf [net-next,08/37] rxrpc: Implement path-MTU probing using padded PING ACKs (RFC8899) [net-next,09/37] rxrpc: Separate the packet length from the data length in rxrpc_txbuf [net-next,10/37] rxrpc: Prepare to be able to send jumbo DATA packets [net-next,11/37] rxrpc: Add a tracepoint to show variables pertinent to jumbo packet size [net-next,12/37] rxrpc: Fix CPU time starvation in I/O thread [net-next,13/37] rxrpc: Fix injection of packet loss [net-next,14/37] rxrpc: Only set DF=1 on initial DATA transmission [net-next,15/37] rxrpc: Timestamp DATA packets before transmitting them [net-next,16/37] rxrpc: Implement progressive transmission queue struct [net-next,17/37] rxrpc: call->acks_hard_ack is now the same call->tx_bottom, so remove it [net-next,18/37] rxrpc: Replace call->acks_first_seq with tracking of the hard ACK point [net-next,19/37] rxrpc: Display stats about jumbo packets transmitted and received [net-next,20/37] rxrpc: Adjust names and types of congestion-related fields [net-next,21/37] rxrpc: Use the new rxrpc_tx_queue struct to more efficiently process ACKs [net-next,22/37] rxrpc: Store the DATA serial in the txqueue and use this in RTT calc [net-next,23/37] rxrpc: Don't use received skbuff timestamps [net-next,24/37] rxrpc: Generate rtt_min [net-next,25/37] rxrpc: Adjust the rxrpc_rtt_rx tracepoint [net-next,26/37] rxrpc: Display userStatus in rxrpc_rx_ack trace [net-next,27/37] rxrpc: Fix the calculation and use of RTO [net-next,28/37] rxrpc: Fix initial resend timeout [net-next,29/37] rxrpc: Send jumbo DATA packets [net-next,30/37] rxrpc: Don't allocate a txbuf for an ACK transmission [net-next,31/37] rxrpc: Use irq-disabling spinlocks between app and I/O thread [net-next,32/37] rxrpc: Tidy up the ACK parsing a bit [net-next,33/37] rxrpc: Add a reason indicator to the tx_data tracepoint [net-next,34/37] rxrpc: Add a reason indicator to the tx_ack tracepoint [net-next,35/37] rxrpc: Manage RTT per-call rather than per-peer [net-next,36/37] rxrpc: Fix request for an ACK when cwnd is minimum [net-next,37/37] rxrpc: Implement RACK/TLP to deal with transmission stalls [RFC8985]

diff --git a/include/trace/events/rxrpc.h b/include/trace/events/rxrpc.h index 0cfc8e1baf1f..71df5c48a413 100644 --- a/include/trace/events/rxrpc.h +++ b/include/trace/events/rxrpc.h @@ -1453,7 +1453,7 @@ TRACE_EVENT(rxrpc_rtt_rx, __entry->rtt = rtt; __entry->srtt = srtt; __entry->rto = rto; - __entry->min_rtt = minmax_get(&call->peer->min_rtt) + __entry->min_rtt = minmax_get(&call->min_rtt) ), TP_printk("c=%08x [%d] %s sr=%08x rr=%08x rtt=%u srtt=%u rto=%u min=%u", diff --git a/net/rxrpc/ar-internal.h b/net/rxrpc/ar-internal.h index cc02a3fcb6b7..a48df82ef4c5 100644 --- a/net/rxrpc/ar-internal.h +++ b/net/rxrpc/ar-internal.h @@ -366,20 +366,9 @@ struct rxrpc_peer { unsigned short hdrsize; /* header size (IP + UDP + RxRPC) */ unsigned short tx_seg_max; /* Maximum number of transmissable segments */ - /* calculated RTT cache */ -#define RXRPC_RTT_CACHE_SIZE 32 - spinlock_t rtt_input_lock; /* RTT lock for input routine */ - ktime_t rtt_last_req; /* Time of last RTT request */ - unsigned int rtt_count; /* Number of samples we've got */ - unsigned int rtt_taken; /* Number of samples taken (wrapping) */ - struct minmax min_rtt; /* Estimated minimum RTT */ - - u32 srtt_us; /* smoothed round trip time << 3 in usecs */ - u32 mdev_us; /* medium deviation */ - u32 mdev_max_us; /* maximal mdev for the last rtt period */ - u32 rttvar_us; /* smoothed mdev_max */ - u32 rto_us; /* Retransmission timeout in usec */ - u8 backoff; /* Backoff timeout (as shift) */ + /* Calculated RTT cache */ + unsigned int recent_srtt_us; + unsigned int recent_rto_us; u8 cong_ssthresh; /* Congestion slow-start threshold */ }; @@ -766,6 +755,18 @@ struct rxrpc_call { rxrpc_serial_t acks_highest_serial; /* Highest serial number ACK'd */ unsigned short acks_nr_sacks; /* Number of soft acks recorded */ unsigned short acks_nr_snacks; /* Number of soft nacks recorded */ + + /* Calculated RTT cache */ + ktime_t rtt_last_req; /* Time of last RTT request */ + unsigned int rtt_count; /* Number of samples we've got */ + unsigned int rtt_taken; /* Number of samples taken (wrapping) */ + struct minmax min_rtt; /* Estimated minimum RTT */ + u32 srtt_us; /* smoothed round trip time << 3 in usecs */ + u32 mdev_us; /* medium deviation */ + u32 mdev_max_us; /* maximal mdev for the last rtt period */ + u32 rttvar_us; /* smoothed mdev_max */ + u32 rto_us; /* Retransmission timeout in usec */ + u8 backoff; /* Backoff timeout (as shift) */ }; /* @@ -1288,10 +1289,12 @@ static inline int rxrpc_abort_eproto(struct rxrpc_call *call, /* * rtt.c */ -void rxrpc_peer_add_rtt(struct rxrpc_call *, enum rxrpc_rtt_rx_trace, int, - rxrpc_serial_t, rxrpc_serial_t, ktime_t, ktime_t); -ktime_t rxrpc_get_rto_backoff(struct rxrpc_peer *peer, bool retrans); -void rxrpc_peer_init_rtt(struct rxrpc_peer *); +void rxrpc_call_add_rtt(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why, + int rtt_slot, + rxrpc_serial_t send_serial, rxrpc_serial_t resp_serial, + ktime_t send_time, ktime_t resp_time); +ktime_t rxrpc_get_rto_backoff(struct rxrpc_call *call, bool retrans); +void rxrpc_call_init_rtt(struct rxrpc_call *call); /* * rxkad.c diff --git a/net/rxrpc/call_event.c b/net/rxrpc/call_event.c index d1e80aeb8ef6..82db8bc664ac 100644 --- a/net/rxrpc/call_event.c +++ b/net/rxrpc/call_event.c @@ -44,8 +44,8 @@ void rxrpc_propose_delay_ACK(struct rxrpc_call *call, rxrpc_serial_t serial, trace_rxrpc_propose_ack(call, why, RXRPC_ACK_DELAY, serial); - if (call->peer->srtt_us) - delay = (call->peer->srtt_us >> 3) * NSEC_PER_USEC; + if (call->srtt_us) + delay = (call->srtt_us >> 3) * NSEC_PER_USEC; else delay = ms_to_ktime(READ_ONCE(rxrpc_soft_ack_delay)); ktime_add_ms(delay, call->tx_backoff); @@ -105,7 +105,7 @@ void rxrpc_resend(struct rxrpc_call *call, rxrpc_serial_t ack_serial, bool ping_ }; struct rxrpc_txqueue *tq = call->tx_queue; ktime_t lowest_xmit_ts = KTIME_MAX; - ktime_t rto = rxrpc_get_rto_backoff(call->peer, false); + ktime_t rto = rxrpc_get_rto_backoff(call, false); bool unacked = false; _enter("{%d,%d}", call->tx_bottom, call->tx_top); @@ -194,7 +194,7 @@ void rxrpc_resend(struct rxrpc_call *call, rxrpc_serial_t ack_serial, bool ping_ } while ((tq = tq->next)); if (lowest_xmit_ts < KTIME_MAX) { - ktime_t delay = rxrpc_get_rto_backoff(call->peer, req.did_send); + ktime_t delay = rxrpc_get_rto_backoff(call, req.did_send); ktime_t resend_at = ktime_add(lowest_xmit_ts, delay); _debug("delay %llu %lld", delay, ktime_sub(resend_at, req.now)); @@ -215,7 +215,7 @@ void rxrpc_resend(struct rxrpc_call *call, rxrpc_serial_t ack_serial, bool ping_ */ if (!req.did_send) { ktime_t next_ping = ktime_add_us(call->acks_latest_ts, - call->peer->srtt_us >> 3); + call->srtt_us >> 3); if (ktime_sub(next_ping, req.now) <= 0) rxrpc_send_ACK(call, RXRPC_ACK_PING, 0, @@ -365,8 +365,8 @@ static void rxrpc_transmit_some_data(struct rxrpc_call *call, */ static void rxrpc_send_initial_ping(struct rxrpc_call *call) { - if (call->peer->rtt_count < 3 || - ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000), + if (call->rtt_count < 3 || + ktime_before(ktime_add_ms(call->rtt_last_req, 1000), ktime_get_real())) rxrpc_send_ACK(call, RXRPC_ACK_PING, 0, rxrpc_propose_ack_ping_for_params); @@ -498,10 +498,10 @@ bool rxrpc_input_call_event(struct rxrpc_call *call) rxrpc_propose_ack_rx_idle); if (call->ackr_nr_unacked > 2) { - if (call->peer->rtt_count < 3) + if (call->rtt_count < 3) rxrpc_send_ACK(call, RXRPC_ACK_PING, 0, rxrpc_propose_ack_ping_for_rtt); - else if (ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000), + else if (ktime_before(ktime_add_ms(call->rtt_last_req, 1000), ktime_get_real())) rxrpc_send_ACK(call, RXRPC_ACK_PING, 0, rxrpc_propose_ack_ping_for_old_rtt); diff --git a/net/rxrpc/call_object.c b/net/rxrpc/call_object.c index 015349445a79..2405b8a4d69d 100644 --- a/net/rxrpc/call_object.c +++ b/net/rxrpc/call_object.c @@ -177,6 +177,8 @@ struct rxrpc_call *rxrpc_alloc_call(struct rxrpc_sock *rx, gfp_t gfp, call->cong_cwnd = RXRPC_MIN_CWND; call->cong_ssthresh = RXRPC_TX_MAX_WINDOW; + rxrpc_call_init_rtt(call); + call->rxnet = rxnet; call->rtt_avail = RXRPC_CALL_RTT_AVAIL_MASK; atomic_inc(&rxnet->nr_calls); diff --git a/net/rxrpc/input.c b/net/rxrpc/input.c index 13d6b8e13ff4..cdfabfbc1a38 100644 --- a/net/rxrpc/input.c +++ b/net/rxrpc/input.c @@ -71,11 +71,11 @@ static void rxrpc_congestion_management(struct rxrpc_call *call, /* We analyse the number of packets that get ACK'd per RTT * period and increase the window if we managed to fill it. */ - if (call->peer->rtt_count == 0) + if (call->rtt_count == 0) goto out; if (ktime_before(call->acks_latest_ts, ktime_add_us(call->cong_tstamp, - call->peer->srtt_us >> 3))) + call->srtt_us >> 3))) goto out_no_clear_ca; summary->change = rxrpc_cong_rtt_window_end; call->cong_tstamp = call->acks_latest_ts; @@ -179,7 +179,7 @@ void rxrpc_congestion_degrade(struct rxrpc_call *call) if (__rxrpc_call_state(call) == RXRPC_CALL_CLIENT_AWAIT_REPLY) return; - rtt = ns_to_ktime(call->peer->srtt_us * (1000 / 8)); + rtt = ns_to_ktime(call->srtt_us * (NSEC_PER_USEC / 8)); now = ktime_get_real(); if (!ktime_before(ktime_add(call->tx_last_sent, rtt), now)) return; @@ -200,7 +200,7 @@ static void rxrpc_add_data_rtt_sample(struct rxrpc_call *call, struct rxrpc_txqueue *tq, int ix) { - rxrpc_peer_add_rtt(call, rxrpc_rtt_rx_data_ack, -1, + rxrpc_call_add_rtt(call, rxrpc_rtt_rx_data_ack, -1, summary->acked_serial, summary->ack_serial, ktime_add_us(tq->xmit_ts_base, tq->segment_xmit_ts[ix]), call->acks_latest_ts); @@ -725,7 +725,7 @@ static void rxrpc_complete_rtt_probe(struct rxrpc_call *call, clear_bit(i + RXRPC_CALL_RTT_PEND_SHIFT, &call->rtt_avail); smp_mb(); /* Read data before setting avail bit */ set_bit(i, &call->rtt_avail); - rxrpc_peer_add_rtt(call, type, i, acked_serial, ack_serial, + rxrpc_call_add_rtt(call, type, i, acked_serial, ack_serial, sent_at, resp_time); matched = true; } diff --git a/net/rxrpc/output.c b/net/rxrpc/output.c index 36c2597768d6..891b85b3b9e7 100644 --- a/net/rxrpc/output.c +++ b/net/rxrpc/output.c @@ -233,7 +233,7 @@ static int rxrpc_fill_out_ack(struct rxrpc_call *call, int nr_kv, u8 ack_reason, if (ack_reason == RXRPC_ACK_PING) rxrpc_begin_rtt_probe(call, *_ack_serial, now, rxrpc_rtt_tx_ping); if (whdr->flags & RXRPC_REQUEST_ACK) - call->peer->rtt_last_req = now; + call->rtt_last_req = now; rxrpc_set_keepalive(call, now); return nr_kv; } @@ -472,9 +472,9 @@ static size_t rxrpc_prepare_data_subpacket(struct rxrpc_call *call, why = rxrpc_reqack_slow_start; else if (call->tx_winsize <= 2) why = rxrpc_reqack_small_txwin; - else if (call->peer->rtt_count < 3 && txb->seq & 1) + else if (call->rtt_count < 3) why = rxrpc_reqack_more_rtt; - else if (ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000), ktime_get_real())) + else if (ktime_before(ktime_add_ms(call->rtt_last_req, 1000), ktime_get_real())) why = rxrpc_reqack_old_rtt; else if (!last && !after(READ_ONCE(call->send_top), txb->seq)) why = rxrpc_reqack_app_stall; @@ -486,7 +486,7 @@ static size_t rxrpc_prepare_data_subpacket(struct rxrpc_call *call, if (why != rxrpc_reqack_no_srv_last) { flags |= RXRPC_REQUEST_ACK; trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_data, -1, serial); - call->peer->rtt_last_req = req->now; + call->rtt_last_req = req->now; } dont_set_request_ack: @@ -575,8 +575,8 @@ static size_t rxrpc_prepare_data_packet(struct rxrpc_call *call, struct rxrpc_se } /* Set timeouts */ - if (call->peer->rtt_count > 1) { - ktime_t delay = rxrpc_get_rto_backoff(call->peer, false); + if (call->rtt_count > 1) { + ktime_t delay = rxrpc_get_rto_backoff(call, false); call->ack_lost_at = ktime_add(req->now, delay); trace_rxrpc_timer_set(call, delay, rxrpc_timer_trace_lost_ack); @@ -589,7 +589,7 @@ static size_t rxrpc_prepare_data_packet(struct rxrpc_call *call, struct rxrpc_se trace_rxrpc_timer_set(call, delay, rxrpc_timer_trace_expect_rx); } if (call->resend_at == KTIME_MAX) { - ktime_t delay = rxrpc_get_rto_backoff(call->peer, false); + ktime_t delay = rxrpc_get_rto_backoff(call, false); call->resend_at = ktime_add(req->now, delay); trace_rxrpc_timer_set(call, delay, rxrpc_timer_trace_resend); diff --git a/net/rxrpc/peer_object.c b/net/rxrpc/peer_object.c index 27b34ed4d76a..e1c63129586b 100644 --- a/net/rxrpc/peer_object.c +++ b/net/rxrpc/peer_object.c @@ -235,12 +235,9 @@ struct rxrpc_peer *rxrpc_alloc_peer(struct rxrpc_local *local, gfp_t gfp, peer->service_conns = RB_ROOT; seqlock_init(&peer->service_conn_lock); spin_lock_init(&peer->lock); - spin_lock_init(&peer->rtt_input_lock); seqcount_init(&peer->mtu_lock); peer->debug_id = atomic_inc_return(&rxrpc_debug_id); - - rxrpc_peer_init_rtt(peer); - + peer->recent_srtt_us = UINT_MAX; peer->cong_ssthresh = RXRPC_TX_MAX_WINDOW; trace_rxrpc_peer(peer->debug_id, 1, why); } @@ -283,8 +280,6 @@ static void rxrpc_init_peer(struct rxrpc_local *local, struct rxrpc_peer *peer, peer->max_data = peer->if_mtu - peer->hdrsize; rxrpc_assess_MTU_size(local, peer); - - peer->rtt_last_req = ktime_get_real(); } /* @@ -496,7 +491,7 @@ EXPORT_SYMBOL(rxrpc_kernel_get_call_peer); */ unsigned int rxrpc_kernel_get_srtt(const struct rxrpc_peer *peer) { - return peer->rtt_count > 0 ? peer->srtt_us >> 3 : UINT_MAX; + return READ_ONCE(peer->recent_srtt_us); } EXPORT_SYMBOL(rxrpc_kernel_get_srtt); diff --git a/net/rxrpc/proc.c b/net/rxrpc/proc.c index 79fd472dbcfc..809e30bfc812 100644 --- a/net/rxrpc/proc.c +++ b/net/rxrpc/proc.c @@ -299,15 +299,15 @@ static int rxrpc_peer_seq_show(struct seq_file *seq, void *v) now = ktime_get_seconds(); seq_printf(seq, "UDP %-47.47s %-47.47s %3u" - " %4u %5u %6llus %8u %8u\n", + " %4u %5u %6llus %8d %8d\n", lbuff, rbuff, refcount_read(&peer->ref), peer->cong_ssthresh, peer->max_data, now - peer->last_tx_at, - peer->srtt_us >> 3, - peer->rto_us); + READ_ONCE(peer->recent_srtt_us), + READ_ONCE(peer->recent_rto_us)); return 0; } diff --git a/net/rxrpc/rtt.c b/net/rxrpc/rtt.c index aff75e168de8..7474f88d7b18 100644 --- a/net/rxrpc/rtt.c +++ b/net/rxrpc/rtt.c @@ -15,14 +15,14 @@ #define RXRPC_TIMEOUT_INIT ((unsigned int)(1 * USEC_PER_SEC)) /* RFC6298 2.1 initial RTO value */ #define rxrpc_jiffies32 ((u32)jiffies) /* As rxrpc_jiffies32 */ -static u32 rxrpc_rto_min_us(struct rxrpc_peer *peer) +static u32 rxrpc_rto_min_us(struct rxrpc_call *call) { return 200; } -static u32 __rxrpc_set_rto(const struct rxrpc_peer *peer) +static u32 __rxrpc_set_rto(const struct rxrpc_call *call) { - return (peer->srtt_us >> 3) + peer->rttvar_us; + return (call->srtt_us >> 3) + call->rttvar_us; } static u32 rxrpc_bound_rto(u32 rto) @@ -40,10 +40,10 @@ static u32 rxrpc_bound_rto(u32 rto) * To save cycles in the RFC 1323 implementation it was better to break * it up into three procedures. -- erics */ -static void rxrpc_rtt_estimator(struct rxrpc_peer *peer, long sample_rtt_us) +static void rxrpc_rtt_estimator(struct rxrpc_call *call, long sample_rtt_us) { long m = sample_rtt_us; /* RTT */ - u32 srtt = peer->srtt_us; + u32 srtt = call->srtt_us; /* The following amusing code comes from Jacobson's * article in SIGCOMM '88. Note that rtt and mdev @@ -66,7 +66,7 @@ static void rxrpc_rtt_estimator(struct rxrpc_peer *peer, long sample_rtt_us) srtt += m; /* rtt = 7/8 rtt + 1/8 new */ if (m < 0) { m = -m; /* m is now abs(error) */ - m -= (peer->mdev_us >> 2); /* similar update on mdev */ + m -= (call->mdev_us >> 2); /* similar update on mdev */ /* This is similar to one of Eifel findings. * Eifel blocks mdev updates when rtt decreases. * This solution is a bit different: we use finer gain @@ -78,31 +78,31 @@ static void rxrpc_rtt_estimator(struct rxrpc_peer *peer, long sample_rtt_us) if (m > 0) m >>= 3; } else { - m -= (peer->mdev_us >> 2); /* similar update on mdev */ + m -= (call->mdev_us >> 2); /* similar update on mdev */ } - peer->mdev_us += m; /* mdev = 3/4 mdev + 1/4 new */ - if (peer->mdev_us > peer->mdev_max_us) { - peer->mdev_max_us = peer->mdev_us; - if (peer->mdev_max_us > peer->rttvar_us) - peer->rttvar_us = peer->mdev_max_us; + call->mdev_us += m; /* mdev = 3/4 mdev + 1/4 new */ + if (call->mdev_us > call->mdev_max_us) { + call->mdev_max_us = call->mdev_us; + if (call->mdev_max_us > call->rttvar_us) + call->rttvar_us = call->mdev_max_us; } } else { /* no previous measure. */ srtt = m << 3; /* take the measured time to be rtt */ - peer->mdev_us = m << 1; /* make sure rto = 3*rtt */ - peer->rttvar_us = umax(peer->mdev_us, rxrpc_rto_min_us(peer)); - peer->mdev_max_us = peer->rttvar_us; + call->mdev_us = m << 1; /* make sure rto = 3*rtt */ + call->rttvar_us = umax(call->mdev_us, rxrpc_rto_min_us(call)); + call->mdev_max_us = call->rttvar_us; } - peer->srtt_us = umax(srtt, 1); + call->srtt_us = umax(srtt, 1); } /* * Calculate rto without backoff. This is the second half of Van Jacobson's * routine referred to above. */ -static void rxrpc_set_rto(struct rxrpc_peer *peer) +static void rxrpc_set_rto(struct rxrpc_call *call) { u32 rto; @@ -113,7 +113,7 @@ static void rxrpc_set_rto(struct rxrpc_peer *peer) * is invisible. Actually, Linux-2.4 also generates erratic * ACKs in some circumstances. */ - rto = __rxrpc_set_rto(peer); + rto = __rxrpc_set_rto(call); /* 2. Fixups made earlier cannot be right. * If we do not estimate RTO correctly without them, @@ -124,73 +124,73 @@ static void rxrpc_set_rto(struct rxrpc_peer *peer) /* NOTE: clamping at RXRPC_RTO_MIN is not required, current algo * guarantees that rto is higher. */ - peer->rto_us = rxrpc_bound_rto(rto); + call->rto_us = rxrpc_bound_rto(rto); } -static void rxrpc_update_rtt_min(struct rxrpc_peer *peer, ktime_t resp_time, long rtt_us) +static void rxrpc_update_rtt_min(struct rxrpc_call *call, ktime_t resp_time, long rtt_us) { /* Window size 5mins in approx usec (ipv4.sysctl_tcp_min_rtt_wlen) */ u32 wlen_us = 5ULL * NSEC_PER_SEC / 1024; - minmax_running_min(&peer->min_rtt, wlen_us, resp_time / 1024, + minmax_running_min(&call->min_rtt, wlen_us, resp_time / 1024, (u32)rtt_us ? : jiffies_to_usecs(1)); } -static void rxrpc_ack_update_rtt(struct rxrpc_peer *peer, ktime_t resp_time, long rtt_us) +static void rxrpc_ack_update_rtt(struct rxrpc_call *call, ktime_t resp_time, long rtt_us) { if (rtt_us < 0) return; /* Update RACK min RTT [RFC8985 6.1 Step 1]. */ - rxrpc_update_rtt_min(peer, resp_time, rtt_us); + rxrpc_update_rtt_min(call, resp_time, rtt_us); - rxrpc_rtt_estimator(peer, rtt_us); - rxrpc_set_rto(peer); + rxrpc_rtt_estimator(call, rtt_us); + rxrpc_set_rto(call); /* Only reset backoff on valid RTT measurement [RFC6298]. */ - peer->backoff = 0; + call->backoff = 0; } /* * Add RTT information to cache. This is called in softirq mode and has - * exclusive access to the peer RTT data. + * exclusive access to the call RTT data. */ -void rxrpc_peer_add_rtt(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why, +void rxrpc_call_add_rtt(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why, int rtt_slot, rxrpc_serial_t send_serial, rxrpc_serial_t resp_serial, ktime_t send_time, ktime_t resp_time) { - struct rxrpc_peer *peer = call->peer; s64 rtt_us; rtt_us = ktime_to_us(ktime_sub(resp_time, send_time)); if (rtt_us < 0) return; - spin_lock(&peer->rtt_input_lock); - rxrpc_ack_update_rtt(peer, resp_time, rtt_us); - if (peer->rtt_count < 3) - peer->rtt_count++; - peer->rtt_taken++; - spin_unlock(&peer->rtt_input_lock); + rxrpc_ack_update_rtt(call, resp_time, rtt_us); + if (call->rtt_count < 3) + call->rtt_count++; + call->rtt_taken++; + + WRITE_ONCE(call->peer->recent_srtt_us, call->srtt_us / 8); + WRITE_ONCE(call->peer->recent_rto_us, call->rto_us); trace_rxrpc_rtt_rx(call, why, rtt_slot, send_serial, resp_serial, - rtt_us, peer->srtt_us, peer->rto_us); + rtt_us, call->srtt_us, call->rto_us); } /* * Get the retransmission timeout to set in nanoseconds, backing it off each * time we retransmit. */ -ktime_t rxrpc_get_rto_backoff(struct rxrpc_peer *peer, bool retrans) +ktime_t rxrpc_get_rto_backoff(struct rxrpc_call *call, bool retrans) { u64 timo_us; - u32 backoff = READ_ONCE(peer->backoff); + u32 backoff = READ_ONCE(call->backoff); - timo_us = peer->rto_us; + timo_us = call->rto_us; timo_us <<= backoff; if (retrans && timo_us * 2 <= RXRPC_RTO_MAX) - WRITE_ONCE(peer->backoff, backoff + 1); + WRITE_ONCE(call->backoff, backoff + 1); if (timo_us < 1) timo_us = 1; @@ -198,10 +198,11 @@ ktime_t rxrpc_get_rto_backoff(struct rxrpc_peer *peer, bool retrans) return ns_to_ktime(timo_us * NSEC_PER_USEC); } -void rxrpc_peer_init_rtt(struct rxrpc_peer *peer) +void rxrpc_call_init_rtt(struct rxrpc_call *call) { - peer->rto_us = RXRPC_TIMEOUT_INIT; - peer->mdev_us = RXRPC_TIMEOUT_INIT; - peer->backoff = 0; - //minmax_reset(&peer->rtt_min, rxrpc_jiffies32, ~0U); + call->rtt_last_req = KTIME_MIN; + call->rto_us = RXRPC_TIMEOUT_INIT; + call->mdev_us = RXRPC_TIMEOUT_INIT; + call->backoff = 0; + //minmax_reset(&call->rtt_min, rxrpc_jiffies32, ~0U); } diff --git a/net/rxrpc/sendmsg.c b/net/rxrpc/sendmsg.c index 018053b71084..57d85c02ca6b 100644 --- a/net/rxrpc/sendmsg.c +++ b/net/rxrpc/sendmsg.c @@ -132,7 +132,7 @@ static int rxrpc_wait_for_tx_window_waitall(struct rxrpc_sock *rx, rxrpc_seq_t tx_start, tx_win; signed long rtt, timeout; - rtt = READ_ONCE(call->peer->srtt_us) >> 3; + rtt = READ_ONCE(call->srtt_us) >> 3; rtt = usecs_to_jiffies(rtt) * 2; if (rtt < 2) rtt = 2;

[net-next,35/37] rxrpc: Manage RTT per-call rather than per-peer

Commit Message

Patch