inet_twsk_deschedule() calls are followed by inet_twsk_put().

Only particular case is in inet_twsk_purge() but there is no point
to defer the inet_twsk_put() after re-enabling BH.

Lets rename inet_twsk_deschedule() to inet_twsk_deschedule_put()
and move the inet_twsk_put() inside.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

timewait sockets have a complex refcounting logic.
Once we realize it should be similar to established and
syn_recv sockets, we can use sk_nulls_del_node_init_rcu()
and remove inet_twsk_unhash()

In particular, deferred inet_twsk_put() added in commit
13475a30 ("tcp: connect() race with timewait reuse")
looks unecessary : When removing a timewait socket from
ehash or bhash, caller must own a reference on the socket
anyway.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Kernel will crash the same if one of the pointer is NULL anyway.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Hop was always either 0 or sizeof(struct ipv6hdr).
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

MV88E6320 and MV88E6321 are largely compatible to MV886352,
but are members of a different chip family.
Signed-off-by: NAleksey S. Kazantsev <ioctl@yandex.ru>
Signed-off-by: NGuenter Roeck <linux@roeck-us.net>
Reviewed-by: NAndrew Lunn <andrew@lunn.ch>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Yuchung Cheng says:

====================
tcp: fixes some congestion control corner cases

This patch series fixes corner cases of TCP congestion control.
First issue is to avoid continuing slow start when cwnd reaches ssthresh.
Second issue is incorrectly processing order of congestion state and
cwnd update when entering fast recovery or undoing cwnd.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The congestion state and cwnd can be updated in the wrong order.
For example, upon receiving a dubious ACK, we incorrectly raise
the cwnd first (tcp_may_raise_cwnd()/tcp_cong_avoid()) because
the state is still Open, then enter recovery state to reduce cwnd.

For another example, if the ACK indicates spurious timeout or
retransmits, we first revert the cwnd reduction and congestion
state back to Open state.  But we don't raise the cwnd even though
the ACK does not indicate any congestion.

To fix this problem we should first call tcp_fastretrans_alert() to
process the dubious ACK and update the congestion state, then call
tcp_may_raise_cwnd() that raises cwnd based on the current state.
Signed-off-by: NYuchung Cheng <ycheng@google.com>
Signed-off-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NNandita Dukkipati <nanditad@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In the original design slow start is only used to raise cwnd
when cwnd is stricly below ssthresh. It makes little sense
to slow start when cwnd == ssthresh: especially
when hystart has set ssthresh in the initial ramp, or after
recovery when cwnd resets to ssthresh. Not doing so will
also help reduce the buffer bloat slightly.
Signed-off-by: NYuchung Cheng <ycheng@google.com>
Signed-off-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NNandita Dukkipati <nanditad@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add a helper to test the slow start condition in various congestion
control modules and other places. This is to prepare a slight improvement
in policy as to exactly when to slow start.
Signed-off-by: NYuchung Cheng <ycheng@google.com>
Signed-off-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NNandita Dukkipati <nanditad@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This change makes it so that the call skb_defer_rx_timestamp will first
check for a phydev before going in and manipulating the skb->data and
skb->len values.  By doing this we can avoid unnecessary work on network
devices that don't support phydev.  As a result we reduce the total
instruction count needed to process this on most devices.
Signed-off-by: NAlexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

V1 of this patch contains Eric Dumazet's suggestion to move the per
dst RTAX_QUICKACK check into tcp_in_quickack_mode(). Thanks Eric.

I ran some tests and after setting the "ip route change quickack 1"
knob there were still many delayed ACKs sent. This occured
because when icsk_ack.quick=0 the !icsk_ack.pingpong value is
subsequently ignored as tcp_in_quickack_mode() checks both these
values. The condition for a quick ack to trigger requires
that both icsk_ack.quick != 0 and icsk_ack.pingpong=0. Currently
only icsk_ack.pingpong is controlled by the knob. But the
icsk_ack.quick value changes dynamically depending on heuristics.
The crux of the matter is that delayed acks still cannot be entirely
disabled even with the RTAX_QUICKACK per dst knob enabled. This
patch ensures that a quick ack is always sent when the RTAX_QUICKACK
per dst knob is turned on.

The "ip route change quickack 1" knob was recently added to enable
quickacks. It was modeled around the TCP_QUICKACK setsockopt() option.
This issue is that even with "ip route change quickack 1" enabled
we still see delayed ACKs under some conditions. It would be nice
to be able to completely disable delayed ACKs.

Here is an example:

# netstat -s|grep dela
    3 delayed acks sent

For all routes enable the knob

# ip route change quickack 1

Generate some traffic across a slow link and we still see the delayed
acks.

# netstat -s|grep dela
    106 delayed acks sent
    1 delayed acks further delayed because of locked socket

The issue is that both the "ip route change quickack 1" knob and
the TCP_QUICKACK option set the icsk_ack.pingpong variable to 0.
However at the business end in the __tcp_ack_snd_check() routine,
tcp_in_quickack_mode() checks that both icsk_ack.quick != 0
and icsk_ack.pingpong=0 in order to trigger a quickack. As
icsk_ack.quick is determined by heuristics it can be 0. When
that occurs the icsk_ack.pingpong value is ignored and a delayed
ACK is sent regardless.

This patch moves the RTAX_QUICKACK per dst check into the
tcp_in_quickack_mode() routine which ensures that a quickack is
always sent when the quickack knob is enabled for that dst.
Signed-off-by: NJon Maxwell <jmaxwell37@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Implement ndo_change_mtu: on MTU change, reallocate Rx ring bufs and signal
HW of new port MTU value.
Signed-off-by: NScott Feldman <sfeldma@gmail.com>
Reviewed-by: NSimon Horman <simon.horman@netronome.com>
Tested-by: NSimon Horman <simon.horman@netronome.com>
Acked-by: NJiri Pirko <jiri@resnulli.us>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Use module_pci_driver for drivers whose init and exit functions
only register and unregister, respectively.

A simplified version of the Coccinelle semantic patch that performs
this transformation is as follows:

@a@
identifier f, x;
@@
-static f(...) { return pci_register_driver(&x); }

@b depends on a@
identifier e, a.x;
statement S;
@@
-static e(...) {
-pci_unregister_driver(&x);
-DBG_PRINT(INIT_DBG,"S");
- }

@c depends on a && b@
identifier a.f;
declarer name module_init;
@@
-module_init(f);

@d depends on a && b && c@
identifier b.e, a.x;
declarer name module_exit;
declarer name module_pci_driver;
@@
-module_exit(e);
+module_pci_driver(x);
Signed-off-by: NVaishali Thakkar <vthakkar1994@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If we don't have access to the new User GTS (T5+), use the old doorbell
mechanism; otherwise use the new BAR2 mechanism.
Signed-off-by: NHariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Currently "ALU_END_FROM_BE 32" and "ALU_END_FROM_LE 32" do not test if
the upper bits of the result are zeros (the arm64 JIT had such bugs).
Extend the two tests to catch this.
Acked-by: NDaniel Borkmann <daniel@iogearbox.net>
Acked-by: NAlexei Starovoitov <ast@plumgrid.com>
Signed-off-by: NXi Wang <xi.wang@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Hariprasad Shenai says:

====================
Cleanup, T6 changes and register range update

This patch series adds the following:
Don't use entire L2T table, update register ranges for T6 adapter,
read stats for only available channels for T6 and enable cim_la dump for
T6 adapter also.

This patch series has been created against net-next tree and includes
patches on cxgb4 driver.

We have included all the maintainers of respective drivers. Kindly review
the change and let us know in case of any review comments.
====================
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NHariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Updating the driver to read the stats of only available channels. T6 and
later has only 2 channels
Signed-off-by: NHariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NHariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The driver was retrieving the parameters for the bounds of its
slice of the L2T from the firmware and then throwing those away and
using the entire table. This corrects that problem.
Signed-off-by: NHariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Use module_pci_driver for drivers whose init and exit functions
only register and unregister, respectively.

A simplified version of the Coccinelle semantic patch that performs
this transformation is as follows:

@a@
identifier f, x;
@@
-static f(...) { return pci_register_driver(&x); }

@b depends on a@
identifier e, a.x;
@@
-static e(...) { pci_unregister_driver(&x); }

@c depends on a && b@
identifier a.f;
declarer name module_init;
@@
-module_init(f);

@d depends on a && b && c@
identifier b.e, a.x;
declarer name module_exit;
declarer name module_pci_driver;
@@
-module_exit(e);
+module_pci_driver(x);
Signed-off-by: NVaishali Thakkar <vthakkar1994@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When packet encapsulation is in use, the MTU needs to be reduced for
headroom reservation.
The existing code takes the updated MTU value only from the host side.
But vSwitch extensions, such as Open vSwitch, require the flexibility
to change the MTU to different values from within a guest during the
lifecycle of a vNIC, when the encapsulation protocol is changed. The
patch supports this kind of MTU changes.
Signed-off-by: NHaiyang Zhang <haiyangz@microsoft.com>
Reviewed-by: NK. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add multiqueue capabilities to ifb netdevice.

This removes last bottleneck for ingress when mq qdisc can be used
to shard load from multiple RX queues on physical device.

Tested:

# netem based setup, installed at receiver side
ETH=eth0
IFB=ifb10
EST="est 1sec 4sec" # Optional rate estimator
RTT_HALF=2ms
#REORDER=20us
#LOSS="loss 1"
TXQ=8

ip link add ifb10 numtxqueues $TXQ type ifb
ip link set dev $IFB up

tc qdisc add dev $ETH ingress 2>/dev/null

tc filter add dev $ETH parent ffff: \
   protocol ip u32 match u32 0 0 flowid 1:1 \
	action mirred egress redirect dev $IFB

tc qdisc del dev $IFB root 2>/dev/null

tc qdisc add dev $IFB root handle 1: mq
for i in `seq 1 $TXQ`
do
 slot=$( printf %x $(( i )) )
 tc qd add dev $IFB parent 1:$slot $EST netem \
	limit 100000 delay $RTT_HALF $REORDER $LOSS
done

lpaa24:~# tc -s -d qd sh dev ifb10
qdisc mq 1: root
 Sent 316544766 bytes 5265927 pkt (dropped 0, overlimits 0 requeues 0)
 backlog 98880b 1648p requeues 0
qdisc netem 8002: parent 1:1 limit 100000 delay 2.0ms
 Sent 39601416 bytes 658721 pkt (dropped 0, overlimits 0 requeues 0)
 rate 38235Kbit 79657pps backlog 12240b 204p requeues 0
qdisc netem 8003: parent 1:2 limit 100000 delay 2.0ms
 Sent 39472866 bytes 657227 pkt (dropped 0, overlimits 0 requeues 0)
 rate 38234Kbit 79655pps backlog 10620b 176p requeues 0
qdisc netem 8004: parent 1:3 limit 100000 delay 2.0ms
 Sent 39703417 bytes 659699 pkt (dropped 0, overlimits 0 requeues 0)
 rate 38320Kbit 79831pps backlog 12780b 213p requeues 0
qdisc netem 8005: parent 1:4 limit 100000 delay 2.0ms
 Sent 39565149 bytes 658011 pkt (dropped 0, overlimits 0 requeues 0)
 rate 38174Kbit 79530pps backlog 11880b 198p requeues 0
qdisc netem 8006: parent 1:5 limit 100000 delay 2.0ms
 Sent 39506078 bytes 657354 pkt (dropped 0, overlimits 0 requeues 0)
 rate 38195Kbit 79571pps backlog 12480b 208p requeues 0
qdisc netem 8007: parent 1:6 limit 100000 delay 2.0ms
 Sent 39675994 bytes 658849 pkt (dropped 0, overlimits 0 requeues 0)
 rate 38323Kbit 79838pps backlog 12600b 210p requeues 0
qdisc netem 8008: parent 1:7 limit 100000 delay 2.0ms
 Sent 39532042 bytes 658367 pkt (dropped 0, overlimits 0 requeues 0)
 rate 38177Kbit 79536pps backlog 13140b 219p requeues 0
qdisc netem 8009: parent 1:8 limit 100000 delay 2.0ms
 Sent 39488164 bytes 657705 pkt (dropped 0, overlimits 0 requeues 0)
 rate 38192Kbit 79568pps backlog 13Kb 222p requeues 0
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: John Fastabend <john.fastabend@gmail.com>
Acked-by: NAlexei Starovoitov <ast@plumgrid.com>
Acked-by: NJamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NHariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add extra check for total vfs for SRIOV to check if that value is
bigger than total vfs in pci SRIOV capabalities. Fix a check and
print of the number of maximum vfs that hw can handle. Fix a check
and print of the number of maximum vfs per port that driver can handle.
Signed-off-by: NCarol L Soto <clsoto@linux.vnet.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The trace bpf samples do not compile on s390x because they use x86
specific fields from the "pt_regs" structure.

Fix this and access the fields via new PT_REGS macros.
Signed-off-by: NMichael Holzheu <holzheu@linux.vnet.ibm.com>
Acked-by: NAlexei Starovoitov <ast@plumgrid.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Enable SG support for Zynq SOC family devices.
Signed-off-by: NPunnaiah Choudary Kalluri <punnaia@xilinx.com>
Acked-by: NNicolas Ferre <nicolas.ferre@atmel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

There are two duplicated xenvif_zerocopy_callback() definitions.
Remove one of them.
Signed-off-by: NLiang Li <liang.z.li@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Eric Dumazet says:

====================
net_sched: act: lockless operation

As mentioned by Alexei last week in Budapest, it is a bit weird
to take a spinlock in order to drop a packet in a tc filter...

Lets add percpu infra for tc actions and use it for gact & mirred.

Before changes, my host with 8 RX queues was handling 5 Mpps with gact,
and more than 11 Mpps after.

Mirred change is not yet visible if ifb+qdisc is used, as ifb is
not yet multi queue enabled, but is a step forward.
====================
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Like act_gact, act_mirred can be lockless in packet processing

1) Use percpu stats
2) update lastuse only every clock tick to avoid false sharing
3) use rcu to protect tcfm_dev
4) Remove spinlock usage, as it is no longer needed.

Next step : add multi queue capability to ifb device
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: John Fastabend <john.fastabend@gmail.com>
Acked-by: NJamal Hadi Salim <jhs@mojatatu.com>
Acked-by: NAlexei Starovoitov <ast@plumgrid.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Final step for gact RCU operation :

1) Use percpu stats
2) update lastuse only every clock tick to avoid false sharing
3) Remove spinlock acquisition, as it is no longer needed.

Since this is the last contended lock in packet RX when tc gact is used,
this gives impressive gain.

My host with 8 RX queues was handling 5 Mpps before the patch,
and more than 11 Mpps after patch.

Tested:

On receiver :

dev=eth0
tc qdisc del dev $dev ingress 2>/dev/null
tc qdisc add dev $dev ingress
tc filter del dev $dev root pref 10 2>/dev/null
tc filter del dev $dev pref 10 2>/dev/null
tc filter add dev $dev est 1sec 4sec parent ffff: protocol ip prio 1 \
	u32 match ip src 7.0.0.0/8 flowid 1:15 action drop

Sender sends packets flood from 7/8 network
Signed-off-by: NEric Dumazet <edumazet@google.com>
Acked-by: NAlexei Starovoitov <ast@plumgrid.com>
Acked-by: NJamal Hadi Salim <jhs@mojatatu.com>
Acked-by: NJohn Fastabend <john.fastabend@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Third step for gact RCU operation :

Following patch will get rid of spinlock protection,
so we need to read tcfg_ptype once.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: NJamal Hadi Salim <jhs@mojatatu.com>
Acked-by: NJohn Fastabend <john.fastabend@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Second step for gact RCU operation :

We want to get rid of the spinlock protecting gact operations.
Stats (packets/bytes) will soon be per cpu.

gact_determ() would not work without a central packet counter,
so lets add it for this mode.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: NJamal Hadi Salim <jhs@mojatatu.com>
Acked-by: NJohn Fastabend <john.fastabend@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

First step for gact RCU operation :

Instead of testing if tcfg_pval is zero or not, just make it 1.

No change in behavior, but slightly faster code.

The smp_rmb()/smp_wmb() barriers, while not strictly needed at this
stage are added for upcoming spinlock removal.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Acked-by: NAlexei Starovoitov <ast@plumgrid.com>
Acked-by: NJamal Hadi Salim <jhs@mojatatu.com>
Acked-by: NJohn Fastabend <john.fastabend@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Reuse existing percpu infrastructure John Fastabend added for qdisc.

This patch adds a new cpustats parameter to tcf_hash_create() and all
actions pass false, meaning this patch should have no effect yet.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: NJamal Hadi Salim <jhs@mojatatu.com>
Acked-by: NJohn Fastabend <john.fastabend@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

qdisc_bstats_update_cpu() and other helpers were added to support
percpu stats for qdisc.

We want to add percpu stats for tc action, so this patch add common
helpers.

qdisc_bstats_update_cpu() is renamed to qdisc_bstats_cpu_update()
qdisc_qstats_drop_cpu() is renamed to qdisc_qstats_cpu_drop()
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: NJamal Hadi Salim <jhs@mojatatu.com>
Acked-by: NJohn Fastabend <john.fastabend@gmail.com>
Acked-by: NAlexei Starovoitov <ast@plumgrid.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Mellanox driver has the knowledge if rxhash is a L4 hash,
if it receives a non fragmented TCP or UDP frame and
NETIF_F_RXCSUM is enabled on netdev.

ip_summed value is CHECKSUM_UNNECESSARY in this case.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Amir Vadai <amirv@mellanox.com>
Cc: Ido Shamay <idos@mellanox.com>
Acked-by: NIdo Shamay <idos@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Yuchung Cheng says:

====================
tcp: reducing lost retransmits in recovery

This patch series reduces lost retransmits in recovery, in particular
when dealing with traffic policers. The main problem is that
slow start in recovery under policing can cause massive lost and
retransmit storms: any excess sending rate turns into drops. The
solution is to avoid doing slow start when lost retransmit is
detected and use packet conservation instead.

On networks with traffic policers the patches have lowered the
TCP loss rates by ~20% from Google servers without latency regressions.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

PRR slow start is often too aggressive especially when drops are
caused by traffic policers. The policers mainly use token bucket
to enforce the rate so sending (twice) faster than the delivery
rate causes excessive drops.

This patch changes PRR to the conservative reduction bound
(CRB) mode in RFC 6937 by default. CRB follows the packet
conservation rule to send at most the delivery rate by default.

But if many packets are lost and the pipe is empty, CRB may take N
round trips to repair N losses. We conditionally turn on slow start
mode if all these conditions are made to speed up the recovery:

  1) on the second round or later in recovery
  2) retransmission sent in the previous round is delivered on this ACK
  3) no retransmission is marked lost on this ACK

By using packet conservation by default, this change reduces the loss
retransmits signicantly on networks that deploy traffic policers,
up to 20% reduction of overall loss rate.
Signed-off-by: NYuchung Cheng <ycheng@google.com>
Signed-off-by: NNandita Dukkipati <nanditad@google.com>
Signed-off-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If the retransmission in CA_Loss is lost again, we should not
continue to slow start or raise cwnd in congestion avoidance mode.
Instead we should enter fast recovery and use PRR to reduce cwnd,
following the principle in RFC5681:

"... or the loss of a retransmission, should be taken as two
 indications of congestion and, therefore, cwnd (and ssthresh) MUST
 be lowered twice in this case."

This is especially important to reduce loss when the CA_Loss
state was caused by a traffic policer dropping the entire inflight.
The CA_Loss state has a problem where a loss of L packets causes the
sender to send a burst of L packets. So a policer that's dropping
most packets in a given RTT can cause a huge retransmit storm. By
contrast, PRR includes logic to bound the number of outbound packets
that result from a given ACK. So switching to CA_Recovery on lost
retransmits in CA_Loss avoids this retransmit storm problem when
in CA_Loss.
Signed-off-by: NYuchung Cheng <ycheng@google.com>
Signed-off-by: NNandita Dukkipati <nanditad@google.com>
Signed-off-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>