Use a local variable to make the code a bit more readable.
Signed-off-by: Jan Sokolowski <jan.sokolowski@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

lockdep_assert_held() is better suited to checking locking requirements,
since it won't get confused when someone else holds the lock. This is
also a step towards possibly removing spin_is_locked().
Signed-off-by: Lance Roy <ldr709@gmail.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

Igor Russkikh says:

====================
net: aquantia: add rx-flow filter support

In this patchset the rx-flow filters functionality and vlan filter offloads
are implemented.

The rules in NIC hardware have fixed order and priorities.
To support this, the locations of filters from ethtool perspective are also fixed:

* Locations 0 - 15 for VLAN ID filters
* Locations 16 - 31 for L2 EtherType and PCP filters
* Locations 32 - 39 for L3/L4 5-tuple filters (locations 32, 36 for IPv6)
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Since it uses the same NIC table as rx flow vlan filter therefore
rx-flow vlan filter accepts only vlans that present on the interface
in case of rx-vlan-filter is on.
Signed-off-by: Dmitry Bogdanov <dmitry.bogdanov@aquantia.com>
Signed-off-by: Igor Russkikh <igor.russkikh@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

L2 EtherType filters allows to filter packet by EtherType field or
both EtherType and User Priority (PCP) field of 802.1Q.
UserPriority (vlan) parameter must be accompanied by mask 0x1FFF. That
is to distinguish VLAN filter from L2 Ethertype filter with
UserPriority since both User Priority and VLAN ID are passed in the
same 'vlan' parameter.

Example:
To add a filter that directs IP4 packess of priority 3 to queue 3:
ethtool -N <ethX> flow-type ether proto 0x800 vlan 0x600 m 0x1FFF \
action 3 loc 16
Signed-off-by: Dmitry Bogdanov <dmitry.bogdanov@aquantia.com>
Signed-off-by: Igor Russkikh <igor.russkikh@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The VLAN filter (VLAN id) is compared against 16 filters.
VLAN id must be accompanied by mask 0xF000. That is to distinguish
VLAN filter from L2 Ethertype filter with UserPriority since both
User Priority and VLAN ID are passed in the same 'vlan' parameter.
Flow type may be any as  it is not matched for VLAN filter.
Due to fixed order of the rules in the NIC, the location 0-15 are
reserved for vlan filters.

Example:
To add a rule that directs packets from VLAN 2001 to queue 5:
ethtool -N <ethX> flow-type ip4 vlan 2001 m 0xF000 action 5 loc 0
Signed-off-by: Dmitry Bogdanov <dmitry.bogdanov@aquantia.com>
Signed-off-by: Igor Russkikh <igor.russkikh@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add support of L3/L4 5-tuple {protocol, src-ip, dst-ip, src-port, dst-port}
filters. Mask is not supported. Src-port and dst-port are only compared for
TCP/UDP/SCTP packets. Both IPv4 and IPv6 are supported.
The supported actions are the drop and the queue assignment.
Due to fixed order of the rules in the NIC, the location 32-39 are
reserved for L3/L4 5-tuple filters. The locations 32 and 36 are
reserved for IPv6 filters.

Examples:
sudo ethtool -N eth0 flow-type ip6 src-ip 2001:db8:0:f101::2 \
dst-ip 2001:db8:0:f101::5 action -1 loc 36

sudo ethtool -N eth0 flow-type udp4 src-ip 10.0.0.4 \
dst-ip 10.0.0.7 src-port 2000 dst-port 2001 action 2 loc 32
Signed-off-by: Dmitry Bogdanov <dmitry.bogdanov@aquantia.com>
Signed-off-by: Igor Russkikh <igor.russkikh@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add infrastructure to support ntuple filter configuration.
Add rule, remove rule, reapply on interface up.
Signed-off-by: Dmitry Bogdanov <dmitry.bogdanov@aquantia.com>
Signed-off-by: Igor Russkikh <igor.russkikh@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add missing register definitions and the functions accessing them
related to rx-flow filters.
Signed-off-by: Dmitry Bogdanov <dmitry.bogdanov@aquantia.com>
Signed-off-by: Igor Russkikh <igor.russkikh@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Ivan Khoronzhuk says:

====================
net: ethernet: ti: cpsw: allow vlan h/w timestamping

The patchset adds several improvements and allows vlan h/w ts.

Based on net-next/master
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Allow vlan tagged packets to be timestamped, as no any restrictions
for this.
Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Each slave has it's own receive timestamp filter. But cpts rx/tx
timestamp enable flags are used to allow ts retrieve only for one
user. This limitation causes data path redundancy and setting overlap
if cpsw module is in dual-mac mode for instance.

If rx ts is enabled only for one port - the second interface must expect
every incoming packet to be PTP packet w/o absolutely any reason, and if
it's PTP - do unneeded stuff, as rx filter for second port is not set
and cpts fifo is not supposed to contain appropriate ts event.
That's not correct.

So, to fix control overlap and avoid redundant CPU cycles, the patch
splits rx/tx ts enable flags between network devices. After the patch,
PTP timestamping still should be used for only one port (or PTP id
counter has to be different for both ports as cpts IP is common).
Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

The overflow event is running with 1 jiffy in case if txq is not
empty, but it can be emptied completely only if next tx event
consumes skb or deletes staled skb from the txq. In case of staled
skb, that can happen for some unpredictable reason (the ts event was
lost or timed out), the overflow event can be generated quite long
time consuming CPU w/o reason before next tx event happens. To avoid
it, purge txq before increasing overflow event rate.
Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

The msgtype and seqid that is smth that belongs to event for
comparison but not for staled txq skb.
Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

A recent change modified variable advertising from a u32 to a link mode
array and left the u32 zero comparison, so essential we now have an array
being compared to null which is not the intention. Fix this by using the
call to linkmode_empty to check if advertising is all zero.

Detected by CoverityScan, CID#1475424 ("Array compared against 0")

Fixes: 3c1bcc86 ("net: ethernet: Convert phydev advertize and supported from u32 to link mode")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Xin Long says:

====================
sctp: add support for sk_reuseport

sctp sk_reuseport allows multiple socks to listen on the same port and
addresses, as long as these socks have the same uid. This works pretty
much as TCP/UDP does, the only difference is that sctp is multi-homing
and all the bind_addrs in these socks will have to completely matched,
otherwise listen() will return err.

The below is when 5 sockets are listening on 172.16.254.254:6400 on a
server, 26 sockets on a client connect to 172.16.254.254:6400 and each
may be processed by a different socket on the server which is selected
by hash(lport, pport, paddr) in reuseport_select_sock():

 # ss --sctp -nn
   State      Recv-Q Send-Q        Local Address:Port     Peer Address:Port
   LISTEN     0      10           172.16.254.254:6400                *:*
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.2.1:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.2.4:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.3.3:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.3.4:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.5.2:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.5.3:1234
   LISTEN     0      10           172.16.254.254:6400                *:*
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.1.3:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.1.4:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.3.2:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.4.1:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.4.2:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.4.3:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.4.4:1234
   LISTEN     0      10           172.16.254.254:6400                *:*
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.1.2:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.3.5:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.4.5:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400   172.16.253.253:1234
   LISTEN     0      10           172.16.254.254:6400                *:*
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.2.2:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.2.3:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.5.4:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.5.5:1234
   LISTEN     0      10           172.16.254.254:6400                *:*
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.1.1:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.1.5:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.2.5:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.3.1:1234
   `- ESTAB   0      0       172.16.254.254%eth1:6400       172.16.5.1:1234
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

When socks' sk_reuseport is set, the same port and address are allowed
to be bound into these socks who have the same uid.

Note that the difference from sk_reuse is that it allows multiple socks
to listen on the same port and address.
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This is a part of sk_reuseport support for sctp. It defines a helper
sctp_bind_addrs_check() to check if the bind_addrs in two socks are
matched. It will add sock_reuseport if they are completely matched,
and return err if they are partly matched, and alloc sock_reuseport
if all socks are not matched at all.

It will work until sk_reuseport support is added in
sctp_get_port_local() in the next patch.

v1->v2:
  - use 'laddr->valid && laddr2->valid' check instead as Marcelo
    pointed in sctp_bind_addrs_check().
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This is a part of sk_reuseport support for sctp, and it selects a
sock by the hashkey of lport, paddr and dport by default. It will
work until sk_reuseport support is added in sctp_get_port_local()
in the next patch.

v1->v2:
  - define lport as __be16 instead of __be32 as Marcelo pointed in
    __sctp_rcv_lookup_endpoint().
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Fixes gcc '-Wunused-but-set-variable' warning:

drivers/net/phy/marvell.c: In function 'm88e1510_config_init':
drivers/net/phy/marvell.c:850:7: warning:
 variable 'pause' set but not used [-Wunused-but-set-variable]

It not used any more after commit 3c1bcc86 ("net: ethernet: Convert phydev
advertize and supported from u32 to link mode")
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Pull networking fixes from David Miller:
 "One last pull request before heading to Vancouver for LPC, here we have:

   1) Don't forget to free VSI contexts during ice driver unload, from
      Victor Raj.

   2) Don't forget napi delete calls during device remove in ice driver,
      from Dave Ertman.

   3) Don't request VLAN tag insertion of ibmvnic device when SKB
      doesn't have VLAN tags at all.

   4) IPV4 frag handling code has to accomodate the situation where two
      threads try to insert the same fragment into the hash table at the
      same time. From Eric Dumazet.

   5) Relatedly, don't flow separate on protocol ports for fragmented
      frames, also from Eric Dumazet.

   6) Memory leaks in qed driver, from Denis Bolotin.

   7) Correct valid MTU range in smsc95xx driver, from Stefan Wahren.

   8) Validate cls_flower nested policies properly, from Jakub Kicinski.

   9) Clearing of stats counters in mc88e6xxx driver doesn't retain
      important bits in the G1_STATS_OP register causing the chip to
      hang. Fix from Andrew Lunn"

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (41 commits)
  act_mirred: clear skb->tstamp on redirect
  net: dsa: mv88e6xxx: Fix clearing of stats counters
  tipc: fix link re-establish failure
  net: sched: cls_flower: validate nested enc_opts_policy to avoid warning
  net: mvneta: correct typo
  flow_dissector: do not dissect l4 ports for fragments
  net: qualcomm: rmnet: Fix incorrect assignment of real_dev
  net: aquantia: allow rx checksum offload configuration
  net: aquantia: invalid checksumm offload implementation
  net: aquantia: fixed enable unicast on 32 macvlan
  net: aquantia: fix potential IOMMU fault after driver unbind
  net: aquantia: synchronized flow control between mac/phy
  net: smsc95xx: Fix MTU range
  net: stmmac: Fix RX packet size > 8191
  qed: Fix potential memory corruption
  qed: Fix SPQ entries not returned to pool in error flows
  qed: Fix blocking/unlimited SPQ entries leak
  qed: Fix memory/entry leak in qed_init_sp_request()
  inet: frags: better deal with smp races
  net: hns3: bugfix for not checking return value
  ...

Merge tag 'kbuild-fixes-v4.20' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild

Pull Kbuild fixes from Masahiro Yamada:

 - fix build errors in binrpm-pkg and bindeb-pkg targets

 - fix false positive matches in merge_config.sh

 - fix build version mismatch in deb-pkg target

 - fix dtbs_install handling in (bin)deb-pkg target

 - revert a commit that allows setlocalversion to write to source tree

* tag 'kbuild-fixes-v4.20' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild:
  builddeb: Fix inclusion of dtbs in debian package
  Revert "scripts/setlocalversion: git: Make -dirty check more robust"
  kbuild: deb-pkg: fix too low build version number
  kconfig: merge_config: avoid false positive matches from comment lines
  kbuild: deb-pkg: fix bindeb-pkg breakage when O= is used
  kbuild: rpm-pkg: fix binrpm-pkg breakage when O= is used

Pull btrfs fixes from David Sterba:
 "Several fixes to recent release (4.19, fixes tagged for stable) and
  other fixes"

* tag 'for-4.20-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  Btrfs: fix missing delayed iputs on unmount
  Btrfs: fix data corruption due to cloning of eof block
  Btrfs: fix infinite loop on inode eviction after deduplication of eof block
  Btrfs: fix deadlock on tree root leaf when finding free extent
  btrfs: avoid link error with CONFIG_NO_AUTO_INLINE
  btrfs: tree-checker: Fix misleading group system information
  Btrfs: fix missing data checksums after a ranged fsync (msync)
  btrfs: fix pinned underflow after transaction aborted
  Btrfs: fix cur_offset in the error case for nocow

Pull ext4 fixes from Ted Ts'o:
 "A large number of ext4 bug fixes, mostly buffer and memory leaks on
  error return cleanup paths"

* tag 'ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4:
  ext4: missing !bh check in ext4_xattr_inode_write()
  ext4: fix buffer leak in __ext4_read_dirblock() on error path
  ext4: fix buffer leak in ext4_expand_extra_isize_ea() on error path
  ext4: fix buffer leak in ext4_xattr_move_to_block() on error path
  ext4: release bs.bh before re-using in ext4_xattr_block_find()
  ext4: fix buffer leak in ext4_xattr_get_block() on error path
  ext4: fix possible leak of s_journal_flag_rwsem in error path
  ext4: fix possible leak of sbi->s_group_desc_leak in error path
  ext4: remove unneeded brelse call in ext4_xattr_inode_update_ref()
  ext4: avoid possible double brelse() in add_new_gdb() on error path
  ext4: avoid buffer leak in ext4_orphan_add() after prior errors
  ext4: avoid buffer leak on shutdown in ext4_mark_iloc_dirty()
  ext4: fix possible inode leak in the retry loop of ext4_resize_fs()
  ext4: fix missing cleanup if ext4_alloc_flex_bg_array() fails while resizing
  ext4: add missing brelse() update_backups()'s error path
  ext4: add missing brelse() add_new_gdb_meta_bg()'s error path
  ext4: add missing brelse() in set_flexbg_block_bitmap()'s error path
  ext4: avoid potential extra brelse in setup_new_flex_group_blocks()

Pull x86 fixes from Thomas Gleixner:
 "A set of x86 fixes:

   - Cure the LDT remapping to user space on 5 level paging which ended
     up in the KASLR space

   - Remove LDT mapping before freeing the LDT pages

   - Make NFIT MCE handling more robust

   - Unbreak the VSMP build by removing the dependency on paravirt ops

   - Support broken PIT emulation on Microsoft hyperV

   - Don't trace vmware_sched_clock() to avoid tracer recursion

   - Remove -pipe from KBUILD CFLAGS which breaks clang and is also
     slower on GCC

   - Trivial coding style and typo fixes"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/cpu/vmware: Do not trace vmware_sched_clock()
  x86/vsmp: Remove dependency on pv_irq_ops
  x86/ldt: Remove unused variable in map_ldt_struct()
  x86/ldt: Unmap PTEs for the slot before freeing LDT pages
  x86/mm: Move LDT remap out of KASLR region on 5-level paging
  acpi/nfit, x86/mce: Validate a MCE's address before using it
  acpi/nfit, x86/mce: Handle only uncorrectable machine checks
  x86/build: Remove -pipe from KBUILD_CFLAGS
  x86/hyper-v: Fix indentation in hv_do_fast_hypercall16()
  Documentation/x86: Fix typo in zero-page.txt
  x86/hyper-v: Enable PIT shutdown quirk
  clockevents/drivers/i8253: Add support for PIT shutdown quirk

Pull perf fixes from Thomas Gleixner:
 "A bunch of perf tooling fixes:

   - Make the Intel PT SQL viewer more robust

   - Make the Intel PT debug log more useful

   - Support weak groups in perf record so it's behaving the same way as
     perf stat

   - Display the LBR stats in callchain entries properly in perf top

   - Handle different PMu names with common prefix properlin in pert
     stat

   - Start syscall augmenting in perf trace. Preparation for
     architecture independent eBPF instrumentation of syscalls.

   - Fix build breakage in JVMTI perf lib

   - Fix arm64 tools build failure wrt smp_load_{acquire,release}"

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf tools: Do not zero sample_id_all for group members
  perf tools: Fix undefined symbol scnprintf in libperf-jvmti.so
  perf beauty: Use SRCARCH, ARCH=x86_64 must map to "x86" to find the headers
  perf intel-pt: Add MTC and CYC timestamps to debug log
  perf intel-pt: Add more event information to debug log
  perf scripts python: exported-sql-viewer.py: Fix table find when table re-ordered
  perf scripts python: exported-sql-viewer.py: Add help window
  perf scripts python: exported-sql-viewer.py: Add Selected branches report
  perf scripts python: exported-sql-viewer.py: Fall back to /usr/local/lib/libxed.so
  perf top: Display the LBR stats in callchain entry
  perf stat: Handle different PMU names with common prefix
  perf record: Support weak groups
  perf evlist: Move perf_evsel__reset_weak_group into evlist
  perf augmented_syscalls: Start collecting pathnames in the BPF program
  perf trace: Fix setting of augmented payload when using eBPF + raw_syscalls
  perf trace: When augmenting raw_syscalls plug raw_syscalls:sys_exit too
  perf examples bpf: Start augmenting raw_syscalls:sys_{start,exit}
  tools headers barrier: Fix arm64 tools build failure wrt smp_load_{acquire,release}

Pull timer fix from Thomas Gleixner:
 "Just the removal of a redundant call into the sched deadline overrun
  check"

* 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  posix-cpu-timers: Remove useless call to check_dl_overrun()

Pull scheduler fixes from Thomas Gleixner:
 "Two small scheduler fixes:

   - Take hotplug lock in sched_init_smp(). Technically not really
     required, but lockdep will complain other.

   - Trivial comment fix in sched/fair"

* 'sched/urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched/fair: Fix a comment in task_numa_fault()
  sched/core: Take the hotplug lock in sched_init_smp()

Pull locking build fix from Thomas Gleixner:
 "A single fix for a build fail with CONFIG_PROFILE_ALL_BRANCHES=y in
  the qspinlock code"

* 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/qspinlock: Fix compile error

Pull core fixes from Thomas Gleixner:
 "A couple of fixlets for the core:

   - Kernel doc function documentation fixes

   - Missing prototypes for weak watchdog functions"

* 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  resource/docs: Complete kernel-doc style function documentation
  watchdog/core: Add missing prototypes for weak functions
  resource/docs: Fix new kernel-doc warnings

The PCI vendor id of U.S. Robotics isn't defined in pci_ids.h so far,
only ISDN driver w6692 has a private definition. Move the definition
to pci_ids.h and use it in the r8169 driver too.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Similar to 80ba92fa ("codel: add ce_threshold attribute")

After EDT adoption, it became easier to implement DCTCP-like CE marking.

In many cases, queues are not building in the network fabric but on
the hosts themselves.

If packets leaving fq missed their Earliest Departure Time by XXX usec,
we mark them with ECN CE. This gives a feedback (after one RTT) to
the sender to slow down and find better operating mode.

Example :

tc qd replace dev eth0 root fq ce_threshold 2.5ms
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

FQ pacing guarantees that paced packets queued by one flow do not
add head-of-line blocking for other flows.

After TCP GSO conversion, increasing limit_output_bytes to 1 MB is safe,
since this maps to 16 skbs at most in qdisc or device queues.
(or slightly more if some drivers lower {gso_max_segs|size})

We still can queue at most 1 ms worth of traffic (this can be scaled
by wifi drivers if they need to)

Tested:

# ethtool -c eth0 | egrep "tx-usecs:|tx-frames:" # 40 Gbit mlx4 NIC
tx-usecs: 16
tx-frames: 16
# tc qdisc replace dev eth0 root fq
# for f in {1..10};do netperf -P0 -H lpaa24,6 -o THROUGHPUT;done

Before patch:
27711
26118
27107
27377
27712
27388
27340
27117
27278
27509

After patch:
37434
36949
36658
36998
37711
37291
37605
36659
36544
37349
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Eric Dumazet says:

====================
tcp: tso defer improvements

This series makes tcp_tso_should_defer() a bit smarter :

1) MSG_EOR gives a hint to TCP to not defer some skbs

2) Second patch takes into account that head tstamp
   can be in the future.

3) Third patch uses existing high resolution state variables
   to have a more precise heuristic.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp_tso_should_defer() first heuristic is to not defer
if last send is "old enough".

Its current implementation uses jiffies and its low granularity.

TSO autodefer performance should not rely on kernel HZ :/

After EDT conversion, we have state variables in nanoseconds that
can allow us to properly implement the heuristic.

This patch increases TSO chunk sizes on medium rate flows,
especially when receivers do not use GRO or similar aggregation.

It also reduces bursts for HZ=100 or HZ=250 kernels, making TCP
behavior more uniform.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp_tso_should_defer() last step tries to check if the probable
next ACK packet is coming in less than half rtt.

Problem is that the head->tstamp might be in the future,
so we need to use signed arithmetics to avoid overflows.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Applications using MSG_EOR are giving a strong hint to TCP stack :

Subsequent sendmsg() can not append more bytes to skbs having
the EOR mark.

Do not try to TSO defer suchs skbs, there is really no hope.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Bitwise operation is a little faster.
So I replace after() with using the flag FLAG_SND_UNA_ADVANCED as it is
already set before.

In addtion, there's another similar improvement in tcp_cwnd_reduction().

Cc: Joe Perches <joe@perches.com>
Suggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>