Eric Dumazet says:

====================
tcp: second round for EDT conversion

First round of EDT patches left TCP stack in a non optimal state.

- High speed flows suffered from loss of performance, addressed
  by the first patch of this series.

- Second patch brings pacing to the current state of networking,
  since we now reach ~100 Gbit on a single TCP flow.

- Third patch implements a mitigation for scheduling delays,
  like the one we did in sch_fq in the past.

- Fourth patch removes one special case in sch_fq for ACK packets.

- Fifth patch removes a serious perfomance cost for TCP internal
  pacing. We should setup the high resolution timer only if
  really needed.

- Sixth patch fixes a typo in BBR.

- Last patch is one minor change in cdg congestion control.

Neal Cardwell also has a patch series fixing BBR after
EDT adoption.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

We store in tcp socket a cache of most recent high resolution
clock, there is no need to call local_clock() again, since
this cache is good enough.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

There was a typo in this parameter name.
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When TCP implements its own pacing (when no fq packet scheduler is used),
it is arming high resolution timer after a packet is sent.

But in many cases (like TCP_RR kind of workloads), this high resolution
timer expires before the application attempts to write the following
packet. This overhead also happens when the flow is ACK clocked and
cwnd limited instead of being limited by the pacing rate.

This leads to extra overhead (high number of IRQ)

Now tcp_wstamp_ns is reserved for the pacing timer only
(after commit "tcp: do not change tcp_wstamp_ns in tcp_mstamp_refresh"),
we can setup the timer only when a packet is about to be sent,
and if tcp_wstamp_ns is in the future.

This leads to a ~10% performance increase in TCP_RR workloads.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

With the new EDT model, sch_fq no longer has to special
case TCP pure acks, since their skb->tstamp will allow them
being sent without pacing delay.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In commit fefa569a ("net_sched: sch_fq: account for schedule/timers
drifts") we added a mitigation for scheduling jitter in fq packet scheduler.

This patch does the same in TCP stack, now it is using EDT model.

Note that this mitigation is valid for both external (fq packet scheduler)
or internal TCP pacing.

This uses the same strategy than the above commit, allowing
a time credit of half the packet currently sent.

Consider following case :

An skb is sent, after an idle period of 300 usec.
The air-time (skb->len/pacing_rate) is 500 usec
Instead of setting the pacing timer to now+500 usec,
it will use now+min(500/2, 300) -> now+250usec

This is like having a token bucket with a depth of half
an skb.

Tested:

tc qdisc replace dev eth0 root pfifo_fast

Before
netperf -P0 -H remote -- -q 1000000000 # 8000Mbit
540000 262144 262144    10.00    7710.43

After :
netperf -P0 -H remote -- -q 1000000000 # 8000 Mbit
540000 262144 262144    10.00    7999.75   # Much closer to 8000Mbit target
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sk_pacing_rate has beed introduced as a u32 field in 2013,
effectively limiting per flow pacing to 34Gbit.

We believe it is time to allow TCP to pace high speed flows
on 64bit hosts, as we now can reach 100Gbit on one TCP flow.

This patch adds no cost for 32bit kernels.

The tcpi_pacing_rate and tcpi_max_pacing_rate were already
exported as 64bit, so iproute2/ss command require no changes.

Unfortunately the SO_MAX_PACING_RATE socket option will stay
32bit and we will need to add a new option to let applications
control high pacing rates.

State      Recv-Q Send-Q Local Address:Port             Peer Address:Port
ESTAB      0      1787144  10.246.9.76:49992             10.246.9.77:36741
                 timer:(on,003ms,0) ino:91863 sk:2 <->
 skmem:(r0,rb540000,t66440,tb2363904,f605944,w1822984,o0,bl0,d0)
 ts sack bbr wscale:8,8 rto:201 rtt:0.057/0.006 mss:1448
 rcvmss:536 advmss:1448
 cwnd:138 ssthresh:178 bytes_acked:256699822585 segs_out:177279177
 segs_in:3916318 data_segs_out:177279175
 bbr:(bw:31276.8Mbps,mrtt:0,pacing_gain:1.25,cwnd_gain:2)
 send 28045.5Mbps lastrcv:73333
 pacing_rate 38705.0Mbps delivery_rate 22997.6Mbps
 busy:73333ms unacked:135 retrans:0/157 rcv_space:14480
 notsent:2085120 minrtt:0.013
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In EDT design, I made the mistake of using tcp_wstamp_ns
to store the last tcp_clock_ns() sample and to store the
pacing virtual timer.

This causes major regressions at high speed flows.

Introduce tcp_clock_cache to store last tcp_clock_ns().
This is needed because some arches have slow high-resolution
kernel time service.

tcp_wstamp_ns is only updated when a packet is sent.

Note that we can remove tcp_mstamp in the future since
tcp_mstamp is essentially tcp_clock_cache/1000, so the
apparent socket size increase is temporary.

Fixes: 9799ccb0 ("tcp: add tcp_wstamp_ns socket field")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

After per-port vlan stats, vlan stats should be released
when fail to add vlan

Fixes: 9163a0fc ("net: bridge: add support for per-port vlan stats")
CC: bridge@lists.linux-foundation.org
cc: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
CC: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: Zhang Yu <zhangyu31@baidu.com>
Signed-off-by: Li RongQing <lirongqing@baidu.com>
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add /proc/net/rxrpc/peers to display the list of peers currently active.
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add the missing unlock before return from function bsq_audit()
in the error handling case.

Fixes: 1d9d8be9 ("fore200e: check for dma mapping failures")
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Michael Chan says:

====================
bnxt_en: Add support for new 57500 chips.

This patch-set is larger than normal because I wanted a complete series
to add basic support for the new 57500 chips.  The new chips have the
following main differences compared to legacy chips:

1. Requires the PF driver to allocate DMA context memory as a backing
store.
2. New NQ (notification queue) for interrupt events.
3. One or more CP rings can be associated with an NQ.
4. 64-bit doorbells.

Most other structures and firmware APIs are compatible with legacy
devices with some exceptions.  For example, ring groups are no longer
used and RSS table format has changed.

The patch-set includes the usual firmware spec. update, some refactoring
and restructuring, and adding the new code to add basic support for the
new class of devices.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add a new poll function that polls for NQ events.  If the NQ event is
a CQ notification, we locate the CP ring from the cq_handle and call
__bnxt_poll_work() to handle RX/TX events on the CP ring.

Add a new has_more_work field in struct bnxt_cp_ring_info to indicate
budget has been reached.  __bnxt_poll_cqs_done() is called to update or
ARM the CP rings if budget has not been reached or not.  If budget
has been reached, the next bnxt_poll_p5() call will continue to poll
from the CQ rings directly.  Otherwise, the NQ will be ARMed for the
next IRQ.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Separate the CP ring polling logic in bnxt_poll_work() into 2 separate
functions __bnxt_poll_work() and __bnxt_poll_work_done().  Since the logic
is separated, we need to add tx_pkts and events fields to struct bnxt_napi
to keep track of the events to handle between the 2 functions.  We also
add had_work_done field to struct bnxt_cp_ring_info to indicate whether
some work was performed on the CP ring.

This is needed to better support the 57500 chips.  We need to poll up to
2 separate CP rings before we update or ARM the CP rings on the 57500 chips.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

On legacy chips, the CP ring may be shared between RX and TX and so only
setup the RX coalescing parameters in such a case.  On 57500 chips, we
always have a dedicated CP ring for TX so we can always set up the
TX coalescing parameters in bnxt_hwrm_set_coal().

Also, the min_timer coalescing parameter applies to the NQ on the new
chips and a separate firmware call needs to be made to set it up.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In the RX code path, we current use the bnxt_napi struct pointer to
identify the associated RX/CP rings.  Change it to use the struct
bnxt_cp_ring_info pointer instead since there are now up to 2
CP rings per MSIX.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

RSS context allocation and RSS indirection table setup are very different
on the new chip.  Refactor bnxt_setup_vnic() to call 2 different functions
to set up RSS for the vnic based on chip type.  On the new chip, the
number of RSS contexts and the indirection table size depends on the
number of RX rings.  Each indirection table entry is also different
on the new chip since ring groups are no longer used.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

On the new 57500 chips, we need to allocate one RSS context for every
64 RX rings.  In previous chips, only one RSS context per vnic is
required regardless of the number of RX rings.  So increase the max
RSS context array count to 8.

Hardware ring groups are not used on the new chips.  Note that the
software ring group structure is still maintained in the driver to
keep track of the rings associated with the vnic.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

On the new 57500 chips, we allocate/free one CP ring for each RX ring or
TX ring separately.  Using separate CP rings for RX/TX is an improvement
as TX events will no longer be stuck behind RX events.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Firmware ring allocation semantics are slightly different for most
ring types on 57500 chips.  Allocation/deallocation for NQ rings are
also added for the new chips.

A CP ring handle is also added so that from the NQ interrupt event,
we can locate the CP ring.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

On the new 57500 chips, getting the associated CP ring ID associated with
an RX ring or TX ring is different than before.  On the legacy chips,
we find the associated ring group and look up the CP ring ID.  On the
57500 chips, each RX ring and TX ring has a dedicated CP ring even if
they share the MSIX.  Use these helper functions at appropriate places
to get the CP ring ID.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

On 57500 chips, the original bnxt_cp_ring_info struct now refers to the
NQ.  bp->cp_nr_rings refer to the number of NQs on 57500 chips.  There
are now 2 pointers for the CP rings associated with RX and TX rings.
Modify bnxt_alloc_cp_rings() and bnxt_free_cp_rings() accordingly.

With multiple CP rings per NAPI, we need to add a pointer in
bnxt_cp_ring_info struct to point back to the bnxt_napi struct.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The ring reservation functions have to be modified for P5 chips in the
following ways:

- bnxt_cp_ring_info structs map to internal NQs as well as CP rings.
- Ring groups are not used.
- 1 CP ring must be available for each RX or TX ring.
- number of RSS contexts to reserve is multiples of 64 RX rings.
- RFS currently not supported.

Also, RX AGG rings are only used for jumbo frames, so we need to
unconditionally call bnxt_reserve_rings() in __bnxt_open_nic()
to see if we need to reserve AGG rings in case MTU has changed.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Store the maximum MSIX capability in PCIe config. space earlier.  When
we call firmware to query capability, we need to compare the PCIe
MSIX max count with the firmware count and use the smaller one as
the MSIX count for 57500 (P5) chips.

The new chips don't use ring groups.  But previous chips do and
the existing logic limits the available rings based on resource
calculations including ring groups.  Setting the max ring groups to
the max rx rings will work on the new chips without changing the
existing logic.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The 57500 series chips have a new 64-bit doorbell format.  Use a new
bnxt_db_info structure to unify the new and the old 32-bit doorbells.
Add a new bnxt_set_db() function to set up the doorbell addreses and
doorbell keys ahead of time.  Modify and introduce new doorbell
helpers to help abstract and unify the old and new doorbells.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

57500 series is a new chip class (P5) that requires some driver changes
in the next several patches.  This adds basic chip ID, doorbells, and
the notification queue (NQ) structures.  Each MSIX is associated with an
NQ instead of a CP ring in legacy chips.  Each NQ has up to 2 associated
CP rings for RX and TX.  The same bnxt_cp_ring_info struct will be used
for the NQ.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Call firmware to configure the DMA addresses of all context memory
pages on new devices requiring context memory.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

New device requires host context memory as a backing store.  Call
firmware to check for context memory requirements and store the
parameters.  Allocate host pages accordingly.

We also need to move the call bnxt_hwrm_queue_qportcfg() earlier
so that all the supported hardware queues and the IDs are known
before checking and allocating context memory.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Newer chips require the PTU_PTE_VALID bit to be set for every page
table entry for context memory and rings.  Additional bits are also
required for page table entries for all rings.  Add a flags field to
bnxt_ring_mem_info struct to specify these additional bits to be used
when setting up the pages tables as needed.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Move the DMA page table and vmem fields in bnxt_ring_struct to a new
bnxt_ring_mem_info struct.  This will allow context memory management
for a new device to re-use some of the existing infrastructure.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

New firmware spec. allows interrupt coalescing parameters, such as
maximums, timer units, supported features to be queried.  Update
the driver to make use of the new call to query these parameters
and provide the legacy defaults if the call is not available.

Replace the hard-coded values with these parameters.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Support the max_ext_req_len field from the HWRM_VER_GET_RESPONSE.
If this field is valid and greater than the mailbox size, use the
short command format to send firmware messages greater than the
mailbox size.  Newer devices use this method to send larger messages
to the firmware.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Latest firmware spec. has some additional rx extended port stats and new
tx extended port stats added.  We now need to check the size of the
returned rx and tx extended stats and determine how many counters are
valid.  New counters added include CoS byte and packet counts for rx
and tx.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Among the new changes are trusted VF support, 200Gbps support, and new
API to dump ring information on the new chips.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Stefano Brivio says:

====================
selftests: pmtu: Add test choice and captures

This series adds a couple of features useful for debugging: 1/2
allows selecting single tests and 2/2 adds optional traffic
captures.

Semantics for current invocation of test script are preserved.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

If --trace is passed as an option and tcpdump is available,
capture traffic for all relevant interfaces to per-test pcap
files named <test>_<interface>.pcap.
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

As number of tests is growing, it's quite convenient to allow
single tests to be run.

Display usage when the script is run with any invalid argument,
keep existing semantics when no arguments are passed so that
automated runs won't break.

Instead of just looping on the list of requested tests, if any,
check first that they exist, and go through them in a nested
loop to keep the existing way to display test descriptions.
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: Sabrina Dubroca <sd@queasysnail.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

netif_device_detach() stops all tx queues already, so we don't need
this call.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Simplify this function, no functional change intended.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>