geneve RFC (draft-ietf-nvo3-geneve-14) allows a geneve packet to carry
multiple geneve opts, so it's necessary for lwtunnel to support adding
multiple geneve opts in one lwtunnel route. But vxlan and erspan opts
are still only allowed to add one option.

With this patch, iproute2 could make it like:

  # ip r a 1.1.1.0/24 encap ip id 1 geneve_opts 0:0:12121212,1:2:12121212 \
    dst 10.1.0.2 dev geneve1

  # ip r a 1.1.1.0/24 encap ip id 1 vxlan_opts 456 \
    dst 10.1.0.2 dev erspan1

  # ip r a 1.1.1.0/24 encap ip id 1 erspan_opts 1:123:0:0 \
    dst 10.1.0.2 dev erspan1

Which are pretty much like cls_flower and act_tunnel_key.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Semicolon is not required at the end of switch block. So, remove it.

Addresses coccinelle warning:
drivers/net/ethernet/chelsio/cxgb4/sge.c:2260:2-3: Unneeded semicolon

Fixes: 4846d533 ("cxgb4: add Tx and Rx path for ETHOFLD traffic")
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

On the NXP LS1028A, there are 2 Ethernet links between the Felix switch
and the ENETC:
- eno2 <-> swp4, at 2.5G
- eno3 <-> swp5, at 1G

Only one of the above Ethernet port pairs can act as a DSA link for
tagging.

When adding initial support for the driver, it was tested only on the 1G
eno3 <-> swp5 interface, due to the necessity of using PHYLIB initially
(which treats fixed-link interfaces as emulated C22 PHYs, so it doesn't
support fixed-link speeds higher than 1G).

After making PHYLINK work, it appears that swp4 still can't act as CPU
port. So it looks like ocelot_set_cpu_port was being called for swp4,
but then it was called again for swp5, overwriting the CPU port assigned
in the DT.

It appears that when you call dsa_upstream_port for a port that is not
defined in the device tree (such as swp5 when using swp4 as CPU port),
its dp->cpu_dp pointer is not initialized by dsa_tree_setup_default_cpu,
and this trips up the following condition in dsa_upstream_port:

	if (!cpu_dp)
		return port;

So the moral of the story is: don't call dsa_upstream_port for a port
that is not defined in the device tree, and therefore its dsa_port
structure is not completely initialized (ds->num_ports is still 6).

Fixes: 56051948 ("net: dsa: ocelot: add driver for Felix switch family")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In the case where the call to phy_interface_is_rgmii returns zero
the variable ret is left uninitialized and this is returned at
the end of the function dp83869_configure_rgmii.  Fix this by
returning 0 instead of the uninitialized value in ret.

Addresses-Coverity: ("Uninitialized scalar variable")
Fixes: 01db923e ("net: phy: dp83869: Add TI dp83869 phy")
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>

LWTUNNEL_IP_OPT_ERSPAN_VER is u8 type, and nla_put_u8 should have
been used instead of nla_put_u32(). This is a copy-paste error.

Fixes: b0a21810 ("lwtunnel: add options setting and dumping for erspan")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Michael Chan says:

====================
bnxt_en: Updates.

This series has the firmware interface update that changes the aRFS/ntuple
interface on 57500 chips.  The 2nd patch adds a counter and improves
the hardware buffer error handling on the 57500 chips.  The rest of the
series is mainly enhancements on error recovery and firmware reset.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

This is especially beneficial during the NVRAM related firmware
commands that have longer timeouts.  If the BNXT_STATE_FW_FATAL_COND
flag gets set while waiting for firmware response, abort and return
error.
Signed-off-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

During loss of heartbeat, log this warning message.
Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

For NVM params that are not supported in the current NVM
configuration, return the error as -EOPNOTSUPP.
Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Report health status update to devlink health reporter, once
reset is completed.

Cc: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The Linux driver is capable of being the master function to handle
resets, so we set the flag to let firmware know.  Some other
drivers, such as DPDK, is not capable and will not set the flag.
Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

If firmware supports hot reset, extend ETHTOOL_RESET to support
hot reset driver which does not require a driver reload after
ETHTOOL_RESET.  The driver will go through the same coordinated
reset sequence as a firmware initiated fatal/non-fatal reset.
Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Use the larger HWRM_COREDUMP_TIMEOUT value for coredump related
data response from the firmware.  These commands take longer than
normal commands.
Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When hardware reports RX buffer errors, the latest 57500 chips do not
require reset.  The packet is discarded by the hardware and the
ring will continue to operate.

Also, add an rx_buf_errors counter for this type of error.  It can help
the user to identify if the aggregation ring is too small.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The aRFS ring table interface has changed for the 57500 chips.  Updating
it accordingly so it will work with the latest production firmware.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Ido Schimmel says:

====================
selftests: Add ethtool and scale tests

This patch set adds generic ethtool tests and a mlxsw-specific router
scale test for Spectrum-2.

Patches #1-#2 from Danielle add the router scale test for Spectrum-2. It
re-uses the same test as Spectrum-1, but it is invoked with a different
scale, according to what it is queried from devlink-resource.

Patches #3-#5 from Amit are a re-work of the ethtool tests that were
posted in the past [1]. Patches #3-#4 add the necessary library
routines, whereas patch #5 adds the test itself. The test checks both
good and bad flows with autoneg on and off. The test plan it detailed in
the commit message.

Last time Andrew and Florian (copied) provided very useful feedback that
is incorporated in this set. Namely:

* Parse the value of the different link modes from
  /usr/include/linux/ethtool.h
* Differentiate between supported and advertised speeds and use the
  latter in autoneg tests
* Make the test generic and move it to net/forwarding/ instead of being
  mlxsw-specific

[1] https://patchwork.ozlabs.org/cover/1112903/
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Check configurations and packets transference with different variations
of autoneg and speed.

Test plan:
1. Test force of same speed with autoneg off
2. Test force of different speeds with autoneg off (should fail)
3. One side is autoneg on and other side sets force of common speeds
4. One side is autoneg on and other side only advertises a subset of the
   common speeds (one speed of the subset)
5. One side is autoneg on and other side only advertises a subset of the
   common speeds. Check that highest speed is negotiated
6. Test autoneg on, but each side advertises different speeds (should
   fail)
Signed-off-by: Amit Cohen <amitc@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add a function that waits for device with maximum number of iterations.
It enables to limit the waiting and prevent infinite loop.

This will be used by the subsequent patch which will set two ports to
different speeds in order to make sure they cannot negotiate a link.

Waiting for all the setup is limited with 10 minutes for each device.
Signed-off-by: Amit Cohen <amitc@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Functions:
1. speeds_arr_get
	The function returns an array of speed values from
        /usr/include/linux/ethtool.h The array looks as follows:
	[10baseT/Half] = 0,
	[10baseT/Full] = 1,
	...

2. ethtool_set:
	params: cmd
	The function runs ethtool by cmd (ethtool -s cmd) and checks if
	there was an error in configuration

3. dev_speeds_get:
	params: dev, with_mode (0 or 1), adver (0 or 1)
	return value: Array of supported/Advertised link modes
	with/without mode

	* Example 1:
	speeds_get swp1 0 0
	return: 1000 10000 40000
	* Example 2:
	speeds_get swp1 1 1
	return: 1000baseKX/Full 10000baseKR/Full 40000baseCR4/Full

4. common_speeds_get:
	params: dev1, dev2, with_mode (0 or 1), adver (0 or 1)
	return value: Array of common speeds of dev1 and dev2

	* Example:
	common_speeds_get swp1 swp2 0 0
	return: 1000 10000
	Assuming that swp1 supports 1000 10000 40000 and swp2 supports
	1000 10000
Signed-off-by: Amit Cohen <amitc@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The scale test for Spectrum-2 should only be invoked for Spectrum-2.
Skip the test otherwise.
Signed-off-by: Danielle Ratson <danieller@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Same as for Spectrum-1, test the ability to add the maximum number of
routes possible to the switch.

Invoke the test from the 'resource_scale' wrapper script.
Signed-off-by: Danielle Ratson <danieller@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Jesper Dangaard says:

====================
page_pool: followup changes to restore tracepoint features

This patchset is a followup to Jonathan patch, that do not release
pool until inflight == 0. That changed page_pool to be responsible for
its own delayed destruction instead of relying on xdp memory model.

As the page_pool maintainer, I'm promoting the use of tracepoint to
troubleshoot and help driver developers verify correctness when
converting at driver to use page_pool. The role of xdp:mem_disconnect
have changed, which broke my bpftrace tools for shutdown verification.
With these changes, the same capabilities are regained.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The MM tracepoint for page free (called kmem:mm_page_free) doesn't provide
the page pointer directly, instead it provides the PFN (Page Frame Number).
This is annoying when writing a page_pool leak detector in BPF.

This patch change page_pool tracepoints to also provide the PFN.
The page pointer is still provided to allow other kinds of
troubleshooting from BPF.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When Jonathan change the page_pool to become responsible to its
own shutdown via deferred work queue, then the disconnect_cnt
counter was removed from xdp memory model tracepoint.

This patch change the page_pool_inflight tracepoint name to
page_pool_release, because it reflects the new responsability
better.  And it reintroduces a counter that reflect the number of
times page_pool_release have been tried.

The counter is also used by the code, to only empty the alloc
cache once.  With a stuck work queue running every second and
counter being 64-bit, it will overrun in approx 584 billion
years. For comparison, Earth lifetime expectancy is 7.5 billion
years, before the Sun will engulf, and destroy, the Earth.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When looking at the details I realised that the memory poison in
__xdp_mem_allocator_rcu_free doesn't make sense. This is because the
SLUB allocator uses the first 16 bytes (on 64 bit), for its freelist,
which overlap with members in struct xdp_mem_allocator, that were
updated.  Thus, SLUB already does the "poisoning" for us.

I still believe that poisoning memory make sense in other cases.
Kernel have gained different use-after-free detection mechanism, but
enabling those is associated with a huge overhead. Experience is that
debugging facilities can change the timing so much, that that a race
condition will not be provoked when enabled. Thus, I'm still in favour
of poisoning memory where it makes sense.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

We currently match clause 45 PHYs using any ID read from a MMD marked
as present in the "Devices in package" registers 5 and 6.  However,
this is incorrect.  45.2 says:

  "The definition of the term package is vendor specific and could be
   a chip, module, or other similar entity."

so a package could be more or less than the whole PHY - a PHY could be
made up of several modules instantiated onto a single chip such as the
Marvell 88x3310, or some of the MMDs could be disabled according to
chip configuration, such as the Broadcom 84881.

In the case of Broadcom 84881, the "Devices in package" registers
contain 0xc000009b, meaning that there is a PHYXS present in the
package, but all registers in MMD 4 return 0xffff.  This leads to our
matching code incorrectly binding this PHY to one of our generic PHY
drivers.

This patch changes the way we determine whether to attempt to match a
MMD identifier, or use it to request a module - if the identifier is
all-ones, then we skip over it. When reading the identifiers, we
initialise phydev->c45_ids.device_ids to all-ones, only reading the
device ID if the "Devices in package" registers indicates we should.

This avoids the generic drivers incorrectly matching on a PHY ID of
0xffffffff.
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

Russell King says:

====================
Add support for SFPs behind PHYs

This series adds partial support for SFP cages connected to PHYs,
specifically optical SFPs.

We add core infrastructure to phylib for this, and arrange for
minimal code in the PHY driver - currently, this is code to verify
that the module is one that we can support for Marvell 10G PHYs.

v2: add yaml binding patch
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Add support for SFP+ cages to the Marvell 10G PHY driver. This is
slightly complicated by the way phylib works in that we need to use
a multi-step process to attach the SFP bus, and we also need to track
the phylink state machine to know when the module's transmit disable
signal should change state.

With appropriate DT changes, this allows the SFP+ canges on the
Macchiatobin platform to be functional.
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add core phylib help for supporting SFP sockets on PHYs.  This provides
a mechanism to inform the SFP layer about PHY up/down events, and also
unregister the SFP bus when the PHY is going away.
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Document the missing sfp property for ethernet controllers (which
has existed for some time) which is being extended to ethernet PHYs.
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Rob Herring <robh@kernel.org>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Pablo Neira Ayuso says:

====================
Netfilter updates for net-next

The following patchset contains Netfilter updates for net-next:

1) Wildcard support for the net,iface set from Kristian Evensen.

2) Offload support for matching on the input interface.

3) Simplify matching on vlan header fields.

4) Add nft_payload_rebuild_vlan_hdr() function to rebuild the vlan
   header from the vlan sk_buff metadata.

5) Pass extack to nft_flow_cls_offload_setup().

6) Add C-VLAN matching support.

7) Use time64_t in xt_time to fix y2038 overflow, from Arnd Bergmann.

8) Use time_t in nft_meta to fix y2038 overflow, also from Arnd.

9) Add flow_action_entry_next() helper function to flowtable offload
   infrastructure.

10) Add IPv6 support to the flowtable offload infrastructure.

11) Support for input interface matching from postrouting,
    from Phil Sutter.

12) Missing check for ndo callback in flowtable offload, from wenxu.

13) Remove conntrack parameter from flow_offload_fill_dir(), from wenxu.

14) Do not pass flow_rule object for rule removal, cookie is sufficient
    to achieve this.

15) Release flow_rule object in case of error from the offload commit
    path.

16) Undo offload ruleset updates if transaction fails.

17) Check for error when binding flowtable callbacks, from wenxu.

18) Always unbind flowtable callbacks when unregistering hooks.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Lots of overlapping changes and parallel additions, stuff
like that.
Signed-off-by: David S. Miller <davem@davemloft.net>

Pull crypto fix from Herbert Xu:
 "This reverts a number of changes to the khwrng thread which feeds the
  kernel random number pool from hwrng drivers. They were trying to fix
  issues with suspend-and-resume but ended up causing regressions"

* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
  Revert "hwrng: core - Freeze khwrng thread during suspend"

This reverts commit 03a3bb7a ("hwrng: core - Freeze khwrng
thread during suspend"), ff296293 ("random: Support freezable
kthreads in add_hwgenerator_randomness()") and 59b56948 ("random:
Use wait_event_freezable() in add_hwgenerator_randomness()").

These patches introduced regressions and we need more time to
get them ready for mainline.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Pull x86 fixes from Ingo Molnar:
 "Two fixes: disable unreliable HPET on Intel Coffe Lake platforms, and
  fix a lockdep splat in the resctrl code"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/resctrl: Fix potential lockdep warning
  x86/quirks: Disable HPET on Intel Coffe Lake platforms

Pull timer fix from Ingo Molnar:
 "Fix integer truncation bug in __do_adjtimex()"

* 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  ntp/y2038: Remove incorrect time_t truncation

Pull perf fixes from Ingo Molnar:
 "Misc fixes: a handful of AUX event handling related fixes, a Sparse
  fix and two ABI fixes"

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf/core: Fix missing static inline on perf_cgroup_switch()
  perf/core: Consistently fail fork on allocation failures
  perf/aux: Disallow aux_output for kernel events
  perf/core: Reattach a misplaced comment
  perf/aux: Fix the aux_output group inheritance fix
  perf/core: Disallow uncore-cgroup events

Pull networking fixes from David Miller:

 1) Fix memory leak in xfrm_state code, from Steffen Klassert.

 2) Fix races between devlink reload operations and device
    setup/cleanup, from Jiri Pirko.

 3) Null deref in NFC code, from Stephan Gerhold.

 4) Refcount fixes in SMC, from Ursula Braun.

 5) Memory leak in slcan open error paths, from Jouni Hogander.

 6) Fix ETS bandwidth validation in hns3, from Yonglong Liu.

 7) Info leak on short USB request answers in ax88172a driver, from
    Oliver Neukum.

 8) Release mem region properly in ep93xx_eth, from Chuhong Yuan.

 9) PTP config timestamp flags validation, from Richard Cochran.

10) Dangling pointers after SKB data realloc in seg6, from Andrea Mayer.

11) Missing free_netdev() in gemini driver, from Chuhong Yuan.

* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (56 commits)
  ipmr: Fix skb headroom in ipmr_get_route().
  net: hns3: cleanup of stray struct hns3_link_mode_mapping
  net/smc: fix fastopen for non-blocking connect()
  rds: ib: update WR sizes when bringing up connection
  net: gemini: add missed free_netdev
  net: dsa: tag_8021q: Fix dsa_8021q_restore_pvid for an absent pvid
  seg6: fix skb transport_header after decap_and_validate()
  seg6: fix srh pointer in get_srh()
  net: stmmac: Use the correct style for SPDX License Identifier
  octeontx2-af: Use the correct style for SPDX License Identifier
  ptp: Extend the test program to check the external time stamp flags.
  mlx5: Reject requests to enable time stamping on both edges.
  igb: Reject requests that fail to enable time stamping on both edges.
  dp83640: Reject requests to enable time stamping on both edges.
  mv88e6xxx: Reject requests to enable time stamping on both edges.
  ptp: Introduce strict checking of external time stamp options.
  renesas: reject unsupported external timestamp flags
  mlx5: reject unsupported external timestamp flags
  igb: reject unsupported external timestamp flags
  dp83640: reject unsupported external timestamp flags
  ...

drivers/net/phy/mscc.c:1683:3-4: Unneeded semicolon

 Remove unneeded semicolon.

Generated by: scripts/coccinelle/misc/semicolon.cocci

Fixes: 75a1ccfe ("mscc.c: Add support for additional VSC PHYs")
CC: Bryan Whitehead <Bryan.Whitehead@microchip.com>
Signed-off-by: kbuild test robot <lkp@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

tcp_mmap is used as a reference program for TCP rx zerocopy,
so it is important to point out some potential issues.

If multiple threads are concurrently using getsockopt(...
TCP_ZEROCOPY_RECEIVE), there is a chance the low-level mm
functions compete on shared ptl lock, if vma are arbitrary placed.

Instead of letting the mm layer place the chunks back to back,
this patch enforces an alignment so that each thread uses
a different ptl lock.

Performance measured on a 100 Gbit NIC, with 8 tcp_mmap clients
launched at the same time :

$ for f in {1..8}; do ./tcp_mmap -H 2002:a05:6608:290:: & done

In the following run, we reproduce the old behavior by requesting no alignment :

$ tcp_mmap -sz -C $((128*1024)) -a 4096
received 32768 MB (100 % mmap'ed) in 9.69532 s, 28.3516 Gbit
  cpu usage user:0.08634 sys:3.86258, 120.511 usec per MB, 171839 c-switches
received 32768 MB (100 % mmap'ed) in 25.4719 s, 10.7914 Gbit
  cpu usage user:0.055268 sys:21.5633, 659.745 usec per MB, 9065 c-switches
received 32768 MB (100 % mmap'ed) in 28.5419 s, 9.63069 Gbit
  cpu usage user:0.057401 sys:23.8761, 730.392 usec per MB, 14987 c-switches
received 32768 MB (100 % mmap'ed) in 28.655 s, 9.59268 Gbit
  cpu usage user:0.059689 sys:23.8087, 728.406 usec per MB, 18509 c-switches
received 32768 MB (100 % mmap'ed) in 28.7808 s, 9.55074 Gbit
  cpu usage user:0.066042 sys:23.4632, 718.056 usec per MB, 24702 c-switches
received 32768 MB (100 % mmap'ed) in 28.8259 s, 9.5358 Gbit
  cpu usage user:0.056547 sys:23.6628, 723.858 usec per MB, 23518 c-switches
received 32768 MB (100 % mmap'ed) in 28.8808 s, 9.51767 Gbit
  cpu usage user:0.059357 sys:23.8515, 729.703 usec per MB, 14691 c-switches
received 32768 MB (100 % mmap'ed) in 28.8879 s, 9.51534 Gbit
  cpu usage user:0.047115 sys:23.7349, 725.769 usec per MB, 21773 c-switches

New behavior (automatic alignment based on Hugepagesize),
we can see the system overhead being dramatically reduced.

$ tcp_mmap -sz -C $((128*1024))
received 32768 MB (100 % mmap'ed) in 13.5339 s, 20.3103 Gbit
  cpu usage user:0.122644 sys:3.4125, 107.884 usec per MB, 168567 c-switches
received 32768 MB (100 % mmap'ed) in 16.0335 s, 17.1439 Gbit
  cpu usage user:0.132428 sys:3.55752, 112.608 usec per MB, 188557 c-switches
received 32768 MB (100 % mmap'ed) in 17.5506 s, 15.6621 Gbit
  cpu usage user:0.155405 sys:3.24889, 103.891 usec per MB, 226652 c-switches
received 32768 MB (100 % mmap'ed) in 19.1924 s, 14.3222 Gbit
  cpu usage user:0.135352 sys:3.35583, 106.542 usec per MB, 207404 c-switches
received 32768 MB (100 % mmap'ed) in 22.3649 s, 12.2906 Gbit
  cpu usage user:0.142429 sys:3.53187, 112.131 usec per MB, 250225 c-switches
received 32768 MB (100 % mmap'ed) in 22.5336 s, 12.1986 Gbit
  cpu usage user:0.140654 sys:3.61971, 114.757 usec per MB, 253754 c-switches
received 32768 MB (100 % mmap'ed) in 22.5483 s, 12.1906 Gbit
  cpu usage user:0.134035 sys:3.55952, 112.718 usec per MB, 252997 c-switches
received 32768 MB (100 % mmap'ed) in 22.6442 s, 12.139 Gbit
  cpu usage user:0.126173 sys:3.71251, 117.147 usec per MB, 253728 c-switches
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Cc: Arjun Roy <arjunroy@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>