In commit 3a9b76fd ("tcp: allow drivers to tweak TSQ logic")
I gave a code sample to set sk->sk_pacing_shift that was not complete.

Better add a helper that can be used by drivers without worries,
and maybe amended in the future.

A wifi driver might use it from its ndo_start_xmit()

Following call would setup TCP to allow up to ~8ms of queued data per
flow.

sk_pacing_shift_update(skb->sk, 7);
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Before this patch the bridge used a fixed 256 element hash table which
was fine for small use cases (in my tests it starts to degrade
above 1000 entries), but it wasn't enough for medium or large
scale deployments. Modern setups have thousands of participants in a
single bridge, even only enabling vlans and adding a few thousand vlan
entries will cause a few thousand fdbs to be automatically inserted per
participating port. So we need to scale the fdb table considerably to
cope with modern workloads, and this patch converts it to use a
rhashtable for its operations thus improving the bridge scalability.
Tests show the following results (10 runs each), at up to 1000 entries
rhashtable is ~3% slower, at 2000 rhashtable is 30% faster, at 3000 it
is 2 times faster and at 30000 it is 50 times faster.
Obviously this happens because of the properties of the two constructs
and is expected, rhashtable keeps pretty much a constant time even with
10000000 entries (tested), while the fixed hash table struggles
considerably even above 10000.
As a side effect this also reduces the net_bridge struct size from 3248
bytes to 1344 bytes. Also note that the key struct is 8 bytes.
Signed-off-by: NNikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Remove XGMII as an option for the 88x3310 PHY driver, as the PHY doesn't
support XGMII's 32-bit data lanes.  It supports USXGMII, which is not
XGMII, but a single-lane serdes interface - see
https://developer.cisco.com/site/usgmii-usxgmii/Signed-off-by: NRussell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: NAndrew Lunn <andrew@lunn.ch>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Heiner Kallweit says:

====================
r8169: extend PCI core and switch to device-managed functions in probe

Probe error path and remove callback can be significantly simplified
by using device-managed functions. To be able to do this in the r8169
driver we need a device-managed version of pci_set_mwi first.

v2:
Change patch 1 based on Björn's review comments and add his Acked-by.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

netif_napi_del is called implicitely by free_netdev, therefore we
don't have to do it explicitely.

When the probe error path is reached, the net_device isn't
registered yet. Therefore reordering the call to netif_napi_del
shouldn't cause any issues.
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Simplify probe error path and remove callback by using device-managed
functions.

rtl_disable_msi isn't needed any longer because the release callback
of pcim_enable_device does this implicitely.
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add pcim_set_mwi(), a device-managed version of pci_set_mwi().
First user is the Realtek r8169 driver.
Signed-off-by: NHeiner Kallweit <hkallweit1@gmail.com>
Acked-by: NBjorn Helgaas <bhelgaas@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

First, rename __inet_twsk_hashdance() to inet_twsk_hashdance()

Then, remove one inet_twsk_put() by setting tw_refcnt to 3 instead
of 4, but adding a fat warning that we do not have the right to access
tw anymore after inet_twsk_hashdance()
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Subash Abhinov Kasiviswanathan says:

====================
net: qualcomm: rmnet: Configuration options

This series adds support for configuring features on rmnet devices.
The rmnet specific features to be configured here are aggregation and
control commands.

Patch 1 is a cleanup of return codes in the transmit path.
Patch 2 removes some redundant ingress and egress macros.
Patch 3 restricts the creation of rmnet dev to one dev per mux id for a
given real dev.
Patch 4 adds ethernet data path support.
Patches 5-6 add support for configuring features on new and existing
rmnet devices.

v1->v2:
The memory leak fixed as part of patch 1 is merged seperately as
a896d94abd2c ("net: qualcomm: rmnet: Fix leak on transmit failure").
Fix a use after free in patch 4 if a packet with headroom lesser than ethernet
header length is received.

v2->v3:
Fix formatting problem in patch 5 in the return statement.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add an option to configure the mux id, aggregation and commad feature
for existing rmnet devices. Implement the changelink netlink
operation for this.
Signed-off-by: NSubash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add an option to configure the rmnet aggregation and command features
on device creation. This is achieved by using the vlan flags option.
Signed-off-by: NSubash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add support to send and receive packets over ethernet.
An example of usage is testing the data path on UML. This can be
achieved by setting up two UML instances in multicast mode and
associating rmnet over the UML ethernet device.
Signed-off-by: NSubash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Upon de-multiplexing data from one real dev, the packets can be sent
to an unique rmnet device for a given mux id.
Signed-off-by: NSubash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Multiplexing is always enabled when transmiting from a rmnet device,
so remove the redundant egress macros. De-multiplexing is always
enabled when receiving packets from a rmnet device, so remove those
ingress macros.
Signed-off-by: NSubash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Only the success and consumed entries were actually in use.
Use standard error codes instead.
Signed-off-by: NSubash Abhinov Kasiviswanathan <subashab@codeaurora.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch enables tail loss probe in cwnd reduction (CWR) state
to detect potential losses. Prior to this patch, since the sender
uses PRR to determine the cwnd in CWR state, the combination of
CWR+PRR plus tcp_tso_should_defer() could cause unnecessary stalls
upon losses: PRR makes cwnd so gentle that tcp_tso_should_defer()
defers sending wait for more ACKs. The ACKs may not come due to
packet losses.

Disallowing TLP when there is unused cwnd had the primary effect
of disallowing TLP when there is TSO deferral, Nagle deferral,
or we hit the rwin limit. Because basically every application
write() or incoming ACK will cause us to run tcp_write_xmit()
to see if we can send more, and then if we sent something we call
tcp_schedule_loss_probe() to see if we should schedule a TLP. At
that point, there are a few common reasons why some cwnd budget
could still be unused:

(a) rwin limit
(b) nagle check
(c) TSO deferral
(d) TSQ

For (d), after the next packet tx completion the TSQ mechanism
will allow us to send more packets, so we don't really need a
TLP (in practice it shouldn't matter whether we schedule one
or not). But for (a), (b), (c) the sender won't send any more
packets until it gets another ACK. But if the whole flight was
lost, or all the ACKs were lost, then we won't get any more ACKs,
and ideally we should schedule and send a TLP to get more feedback.
In particular for a long time we have wanted some kind of timer for
TSO deferral, and at least this would give us some kind of timer
Reported-by: NSteve Ibanez <sibanez@stanford.edu>
Signed-off-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NYuchung Cheng <ycheng@google.com>
Reviewed-by: NNandita Dukkipati <nanditad@google.com>
Reviewed-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Since we now hold RTNL lock in tc_action_net_exit(), it is good to
batch them to speedup tc action dismantle.

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: NCong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Stephen Hemminger says:

====================
hv_netvsc: Fix default and limit of recv buffer

The default for receive buffer descriptors is not correct, it should
match the default receive buffer size and the upper limit of receive
buffer size is too low.  Also, for older versions of Window servers
hosts, different lower limit check is necessary, otherwise the buffer
request will be rejected by the host, resulting vNIC not come up.

This patch set corrects these problems.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The values were not computed correctly. There are no significant
visible impact, though.

The intended size of RX buffer is 16 MB, and the default slot size is 1728.
So, NETVSC_DEFAULT_RX should be 16*1024*1024 / 1728 = 9709.

The intended size of TX buffer is 1 MB, and the slot size is 6144.
So, NETVSC_DEFAULT_TX should be 1024*1024 / 6144 = 170.

The patch puts the formula directly into the macro, and moves them to
hyperv_net.h, together with related macros.

Fixes: 5023a6db ("netvsc: increase default receive buffer size")
Signed-off-by: NHaiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: NStephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The max should be 31 MB on host with NVSP version > 2.

On legacy hosts (NVSP version <=2) only 15 MB receive buffer is allowed,
otherwise the buffer request will be rejected by the host, resulting
vNIC not coming up.

The NVSP version is only available after negotiation. So, we add the
limit checking for legacy hosts in netvsc_init_buf().

Fixes: 5023a6db ("netvsc: increase default receive buffer size")
Signed-off-by: NHaiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: NStephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Richard Leitner says:

====================
net: fec: fix refclk enable for SMSC LAN8710/20

This patch series fixes the use of the SMSC LAN8710/20 with a Freescale ETH
when the refclk is generated by the FSL.

This patchset depends on the "phylib: Add device reset GPIO support" patch
submitted by Geert Uytterhoeven/Sergei Shtylyov, which was merged to
net-next as commit bafbdd52 ("phylib: Add device reset GPIO support").

Changes v5:
	- fix reset delay calculation (max_t instead of min_t)

Changes v4:
	- simplify dts parsing
	- simplify reset delay evaluation and execution
	- fec: ensure to only reset once during fec_enet_open()
	- remove dependency notes from commit message
	- add reviews and acks

Changes v3:
	- use phylib to hard-reset the PHY
	- implement reset delays in phylib
	- add new phylib API & flag (PHY_RST_AFTER_CLK_EN) to determine if
	  a PHY is affected

Changes v2:
	- simplify and fix fec_reset_phy function to support multiple calls
	- include: linux: phy: harmonize phy_id{,_mask} type
	- reset the phy instead of not turning the clock on and off
	  (which would have caused a power consumption regression)
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Some PHYs (for example the SMSC LAN8710/LAN8720) doesn't allow turning
the refclk on and off again during operation (according to their
datasheet). Nonetheless exactly this behaviour was introduced for power
saving reasons by commit e8fcfcd5 ("net: fec: optimize the clock management to save power").
Therefore add support for the phy_reset_after_clk_enable function from
phylib to mitigate this issue.

Generally speaking this issue is only relevant if the ref clk for the
PHY is generated by the SoC and therefore the PHY is configured to
"REF_CLK In Mode". In our specific case (PCB) this problem does occur at
about every 10th to 50th POR of an LAN8710 connected to an i.MX6SOLO
SoC. The typical symptom of this problem is a "swinging" ethernet link.
Similar issues were reported by users of the NXP forum:
	https://community.nxp.com/thread/389902
	https://community.nxp.com/message/309354
With this patch applied the issue didn't occur for at least a few
hundret PORs of our board.

Fixes: e8fcfcd5 ("net: fec: optimize the clock management to save power")
Signed-off-by: NRichard Leitner <richard.leitner@skidata.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The Microchip/SMSC LAN8710/LAN8720 PHYs need (according to their
datasheet [1]) a continuous REF_CLK when configured to "REF_CLK In Mode".
Therefore set the PHY_RST_AFTER_CLK_EN flag for those PHYs to let the
ETH driver reset them after the REF_CLK is enabled.

[1] http://ww1.microchip.com/downloads/en/DeviceDoc/00002165B.pdfSigned-off-by: NRichard Leitner <richard.leitner@skidata.com>
Reviewed-by: NAndrew Lunn <andrew@lunn.ch>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Some PHYs need the refclk to be a continuous clock. Therefore they don't
allow turning it off and on again during operation. Nonetheless such a
clock switching is performed by some ETH drivers (namely FEC [1]) for
power saving reasons. An example for an affected PHY is the
SMSC/Microchip LAN8720 in "REF_CLK In Mode".

In order to provide a uniform method to overcome this problem this patch
adds a new phy_driver flag (PHY_RST_AFTER_CLK_EN) and corresponding
function phy_reset_after_clk_enable() to the phylib. These should be
used to trigger reset of the PHY after the refclk is switched on again.

[1] commit e8fcfcd5 ("net: fec: optimize the clock management to save power")
Signed-off-by: NRichard Leitner <richard.leitner@skidata.com>
Reviewed-by: NAndrew Lunn <andrew@lunn.ch>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Some PHYs need a minimum time after the reset gpio was asserted and/or
deasserted. To ensure we meet these timing requirements add two new
optional devicetree parameters for the phy: reset-delay-us and
reset-post-delay-us.
Signed-off-by: NRichard Leitner <richard.leitner@skidata.com>
Reviewed-by: NGeert Uytterhoeven <geert+renesas@glider.be>
Reviewed-by: NAndrew Lunn <andrew@lunn.ch>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Antoine Tenart says:

====================
net: mvpp2: various improvements

These patches are sent as a series to avoid any possible conflict, even
though there're not entirely related. I can send them separately if
needed. The series applies on today's net-next tree.

Since v1:
  - Removed the patch disabling TSO on allocation errors.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch adjust the coalescing parameters to the vendor
recommendations for the PPv2 network controller.
Suggested-by: NYan Markman <ymarkman@marvell.com>
Signed-off-by: NAntoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch adds the tx-usec value to the informations reported to
ethtool by the get_coalesce function.
Suggested-by: NYan Markman <ymarkman@marvell.com>
Signed-off-by: NAntoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Cosmetic patch aligning values in the ethtool get_coalesce function.
This patch do not modify in anyway the driver's behaviour.
Signed-off-by: NAntoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The Rx/Tx ring sizes can be adjusted thanks to ethtool given specific
network needs. This commit splits the default ring size from its max
value to allow ethtool to vary the parameters in both ways.
Signed-off-by: NYan Markman <ymarkman@marvell.com>
[Antoine: commit message]
Signed-off-by: NAntoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch adds a check to only free the TSO header buffer when its
allocation previously succeeded.
Signed-off-by: NAntoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Eric Dumazet says:

====================
tcp: better receiver autotuning

Now TCP senders no longer backoff when a drop is detected,
it appears we are very often receive window limited.

This series makes tcp_rcv_space_adjust() slightly more robust
and responsive.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Back in linux-3.13 (commit b0983d3c ("tcp: fix dynamic right sizing"))
I addressed the pressing issues we had with receiver autotuning.

But DRS suffers from extra latencies caused by rcv_rtt_est.rtt_us
drifts. One common problem happens during slow start, since the
apparent RTT measured by the receiver can be inflated by ~50%,
at the end of one packet train.

Also, a single drop can delay read() calls by one RTT, meaning
tcp_rcv_space_adjust() can be called one RTT too late.

By replacing the tri-modal heuristic with a continuous function,
we can offset the effects of not growing 'at the optimal time'.

The curve of the function matches prior behavior if the space
increased by 25% and 50% exactly.

Cost of added multiply/divide is small, considering a TCP flow
typically would run this part of the code few times in its life.

I tested this patch with 100 ms RTT / 1% loss link, 100 runs
of (netperf -l 5), and got an average throughput of 4600 Mbit
instead of 1700 Mbit.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Acked-by: NSoheil Hassas Yeganeh <soheil@google.com>
Acked-by: NWei Wang <weiwan@google.com>
Acked-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When using large tcp_rmem[2] values (I did tests with 500 MB),
I noticed overflows while computing rcvwin.

Lets fix this before the following patch.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Acked-by: NSoheil Hassas Yeganeh <soheil@google.com>
Acked-by: NWei Wang <weiwan@google.com>
Acked-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

While rcvbuf is properly clamped by tcp_rmem[2], rcvwin
is left to a potentially too big value.

It has no serious effect, since :
1) tcp_grow_window() has very strict checks.
2) window_clamp can be mangled by user space to any value anyway.

tcp_init_buffer_space() and companions use tcp_full_space(),
we use tcp_win_from_space() to avoid reloading sk->sk_rcvbuf
Signed-off-by: NEric Dumazet <edumazet@google.com>
Acked-by: NSoheil Hassas Yeganeh <soheil@google.com>
Acked-by: NWei Wang <weiwan@google.com>
Acked-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Since both tx_ring and first_tx are the head of tx ring, it not
necessary to use two structure members to statically indicate
the head of tx ring. So first_tx is removed.

CC: Srinivas Eeda <srinivas.eeda@oracle.com>
CC: Joe Jin <joe.jin@oracle.com>
CC: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: NZhu Yanjun <yanjun.zhu@oracle.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Jakub Kicinski says:

====================
nfp: dead code, clean ups and slight improvements

This series contains small clean ups from John and Carl, and brings
no functional changes.

John's improvements target the flower code.  First he makes sure we don't
allocate space in FW request messages for MAC matches if the TC rule does
not contain any.  The remaining two patches remove some dead code and
unused defines.

Carl follows up with a slight optimization to his recent ethtool FW state
dumps, byte swapping input parameters once instead of the data for every
dumped item.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Convert the requested dump level parameter to big-endian at the start of
nfp_net_dump_calculate_size() and nfp_net_dump_populate_buffer(), then
compare and assign it directly where needed in the traversal and prolog
code. This decreases the total number of conversions used.
Signed-off-by: NCarl Heymann <carl.heymann@netronome.com>
Reviewed-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Delete match field defines that are not supported at this time.
Signed-off-by: NJohn Hurley <john.hurley@netronome.com>
Reviewed-by: NDirk van der Merwe <dirk.vandermerwe@netronome.com>
Reviewed-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Port matching is selected by default on every rule so remove check for it
and delete 'else' side of the statement. Remove nfp_flower_meta_one as now
it will not feature in the code. Rename nfp_flower_meta_two given that one
has been removed.

'Additional metadata' if statement can never be true so remove it as well.
Signed-off-by: NJohn Hurley <john.hurley@netronome.com>
Reviewed-by: NDirk van der Merwe <dirk.vandermerwe@netronome.com>
Reviewed-by: NJakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>