Kristian Evensen says:
  In a project I am involved in, we are running ipsec (Strongswan) on
  different mt7621-based routers. Each router is configured as an
  initiator and has around ~30 tunnels to different responders (running
  on misc. devices). Before the flow cache was removed (kernel 4.9), we
  got a combined throughput of around 70Mbit/s for all tunnels on one
  router. However, we recently switched to kernel 4.14 (4.14.48), and
  the total throughput is somewhere around 57Mbit/s (best-case). I.e., a
  drop of around 20%. Reverting the flow cache removal restores, as
  expected, performance levels to that of kernel 4.9.

When pcpu xdst exists, it has to be validated first before it can be
used.

A negative hit thus increases cost vs. no-cache.

As number of tunnels increases, hit rate decreases so this pcpu caching
isn't a viable strategy.

Furthermore, the xdst cache also needs to run with BH off, so when
removing this the bh disable/enable pairs can be removed too.

Kristian tested a 4.14.y backport of this change and reported
increased performance:

  In our tests, the throughput reduction has been reduced from around -20%
  to -5%. We also see that the overall throughput is independent of the
  number of tunnels, while before the throughput was reduced as the number
  of tunnels increased.
Reported-by: Kristian Evensen <kristian.evensen@gmail.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>

This patch adds support for virtual xfrm interfaces.
Packets that are routed through such an interface
are guaranteed to be IPsec transformed or dropped.
It is a generic virtual interface that ensures IPsec
transformation, no need to know what happens behind
the interface. This means that we can tunnel IPv4 and
IPv6 through the same interface and support all xfrm
modes (tunnel, transport and beet) on it.
Co-developed-by: Lorenzo Colitti <lorenzo@google.com>
Co-developed-by: Benedict Wong <benedictwong@google.com>
Signed-off-by: Lorenzo Colitti <lorenzo@google.com>
Signed-off-by: Benedict Wong <benedictwong@google.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Acked-by: Shannon Nelson <shannon.nelson@oracle.com>
Tested-by: Benedict Wong <benedictwong@google.com>
Tested-by: Antony Antony <antony@phenome.org>
Reviewed-by: Eyal Birger <eyal.birger@gmail.com>

This patch adds the xfrm interface id as a lookup key
for xfrm states and policies. With this we can assign
states and policies to virtual xfrm interfaces.
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Acked-by: Shannon Nelson <shannon.nelson@oracle.com>
Acked-by: Benedict Wong <benedictwong@google.com>
Tested-by: Benedict Wong <benedictwong@google.com>
Tested-by: Antony Antony <antony@phenome.org>
Reviewed-by: Eyal Birger <eyal.birger@gmail.com>

Add a new flowi_xfrm structure with informations needed to do
a xfrm lookup. At the moment it keeps the informations about
the new xfrm interface id needed to lookup xfrm interfaces
that are introduced with a followup patch. We need this new
lookup key as other possible keys, like the ifindex is
already part of the xfrm selector and used as a key to
enforce the output device after the transformation in the
policy/state lookup.
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Acked-by: Shannon Nelson <shannon.nelson@oracle.com>
Acked-by: Benedict Wong <benedictwong@google.com>
Tested-by: Benedict Wong <benedictwong@google.com>
Tested-by: Antony Antony <antony@phenome.org>
Reviewed-by: Eyal Birger <eyal.birger@gmail.com>

We already support setting an output mark at the xfrm_state,
unfortunately this does not support the input direction and
masking the marks that will be applied to the skb. This change
adds support applying a masked value in both directions.

The existing XFRMA_OUTPUT_MARK number is reused for this purpose
and as it is now bi-directional, it is renamed to XFRMA_SET_MARK.

An additional XFRMA_SET_MARK_MASK attribute is added for setting the
mask. If the attribute mask not provided, it is set to 0xffffffff,
keeping the XFRMA_OUTPUT_MARK existing 'full mask' semantics.
Co-developed-by: Tobias Brunner <tobias@strongswan.org>
Co-developed-by: Eyal Birger <eyal.birger@gmail.com>
Co-developed-by: Lorenzo Colitti <lorenzo@google.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Tobias Brunner <tobias@strongswan.org>
Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
Signed-off-by: Lorenzo Colitti <lorenzo@google.com>

Heiner Kallweit says:

====================
r8169: enable ASPM on RTL8168E-VL

This patch series enables ASPM for the RTL8168E-VL and aligns ASPM entry
latency handling with the vendor driver before.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Let's enable ASPM also on the RTL8168E-VL (chip version 34).
Works fine on my Zotac Mini PC with this chip. Temperature when
being idle is significantly lower than before due to reaching
deeper PC states.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The r8168 vendor driver always uses value 0x27. In r8169 we have few
chips where 0x17 is used. So far this didn't matter because ASPM was
disabled anyway. Now that ASPM was re-enabled let's also use 0x27 only.

One of the chips affected by this change is RTL8168E-VL, on my system
with this chip value 0x27 works fine.

In addition rename rtl_csi_access_enable_2() to
rtl_set_def_aspm_entry_latency() to make clear that we set the default
ASPM entry latency.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Paul Burton says:

====================
net: pch_gbe: Cleanups

This series begins the process of cleaning up the pch_gbe network
driver. Whilst my ultimate goal is to add support for using this driver
on the MIPS Boston development board, this series sets that aside in
favor of making some more general cleanups. My hope is that this will
both make the driver a little more maleable & reduce the probability of
me gouging out my eyes.

Applies cleanly atop net-next as of 5424ea27 ("netns: get more
entropy from net_hash_mix()").
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Refactor pch_gbe_set_multi in order to avoid unnecessary indentation &
make it clearer what the code is doing.

The one behavioral change from this patch is that we'll no longer
configure the MAC address registers for multicast addresses when the
IFF_PROMISC or IFF_ALLMULTI flags are set. In these cases, just as when
we want to monitor more multicast addresses than we have MAC address
registers, we disable multicast filtering so the MAC address registers
are unused.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The pch_gbe driver sets up multicast address filters using a convoluted
mechanism by which pch_gbe_set_multi allocates an array to hold
multicast addresses, copies desired addresses into that array, calls a
pch_gbe_mac_mc_addr_list_update function which copies addresses out of
that array into MAC registers, then frees the array.

This patch simplifies this somewhat by inlining
pch_gbe_mac_mc_addr_list_update into pch_gbe_set_multi, and removing the
requirement for the MAC addresses to stored consecutively in a single
array.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Make use of the module_pci_driver() macro to remove some needless
boilerplate code from the pch_gbe driver. This does have the side effect
of removing the print of the driver's version during probe, but this is
pretty useless information anyway - the version has changed only once
whilst the driver has been in mainline, despite many changes being made
to it before and since.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The pch_gbe driver includes some code which appears to be an attempt to
work around a problem with the pch_gbe_free_rx_resources &
pch_gbe_free_tx_resources functions that no longer exists. Remove the
code guarded by the never-defined RINGFREE preprocessor macro.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The pch_gbe driver currently presumes that the PHY is connected using
RGMII, and would need further work to support other buses. It includes a
define which is always set that conditionalises some of the
RGMII-specific code regardless. Remove it. If we do ever support
different MII buses then preprocessor defines won't be the best way to
select between them anyway.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The pch_gbe driver calls a pch_gbe_hal_setup_init_funcs function which
ultimately sets the value of one field in struct pch_gbe_phy_info in a
convoluted way.

This patch removes pch_gbe_hal_setup_init_funcs in favor of inlining it,
and in turn its callee pch_gbe_plat_init_function_pointers, into the
single caller pch_gbe_sw_init.

With this pch_gbe_api.c & pch_gbe_api.h are essentially empty, so they
are removed & inclusions of the latter replaced with pch_gbe_phy.h which
was previously being included via pch_gbe_api.h.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

For some reason the pch_gbe driver contains a struct pch_gbe_functions
with pointers used by a HAL abstraction layer, even though there is only
one implementation of each function.

This patch removes the get_bus_info abstraction. Its single
implementation (pch_gbe_plat_get_bus_info) only sets values within a
struct pch_gbe_bus_info which is never used, so we simply remove the
call to it in pch_gbe_probe & remove struct pch_gbe_bus_info entirely.

Now that struct pch_gbe_functions is empty we remove it entirely too.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

For some reason the pch_gbe driver contains a struct pch_gbe_functions
with pointers used by a HAL abstraction layer, even though there is only
one implementation of each function.

This patch removes the init_hw abstraction in favor of inlining its
single implementation (pch_gbe_plat_init_hw) into its single caller
(pch_gbe_reset).
Signed-off-by: Paul Burton <paul.burton@mips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

For some reason the pch_gbe driver contains a struct pch_gbe_functions
with pointers used by a HAL abstraction layer, even though there is only
one implementation of each function.

This patch removes the read_phy_reg & write_phy_reg abstractions in
favor of calling pch_gbe_phy_read_reg_miic & pch_gbe_phy_write_reg_miic
directly.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

For some reason the pch_gbe driver contains a struct pch_gbe_functions
with pointers used by a HAL abstraction layer, even though there is only
one implementation of each function.

This patch removes the reset_phy abstraction in favor of calling
pch_gbe_phy_hw_reset directly.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

For some reason the pch_gbe driver contains a struct pch_gbe_functions
with pointers used by a HAL abstraction layer, even though there is only
one implementation of each function.

This patch removes the sw_reset_phy abstraction, which it turns out is
never even used. Its one implementation, which is already called
directly within the same translation unit, can therefore be made static
and removed from the pch_gbe_phy.h header.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

For some reason the pch_gbe driver contains a struct pch_gbe_functions
with pointers used by a HAL abstraction layer, even though there is only
one implementation of each function.

This patch removes the read_mac_addr abstraction in favor of calling
pch_gbe_mac_read_mac_addr directly. Since this is defined in the same
translation unit as all of its callers, we can make it static & remove
it from the pch_gbe.h header.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

For some reason the pch_gbe driver contains a struct pch_gbe_functions
with pointers used by a HAL abstraction layer, even though there is only
one implementation of each function.

This patch removes the power_up_phy & power_down_phy abstractions in
favor of calling pch_phy_power_up & pch_phy_power_down directly.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The pch_gbe driver includes a 'copybreak' parameter which appears to
have been copied from the e1000e driver but is entirely unused. Remove
the dead code.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

struct net are effectively allocated from order-1 pages on x86,
with one object per slab, meaning that the 13 low order bits
of their addresses are zero.

Once shifted by L1_CACHE_SHIFT, this leaves 7 zero-bits,
meaning that net_hash_mix() does not help spreading
objects on various hash tables.

For example, TCP listen table has 32 buckets, meaning that
all netns use the same bucket for port 80 or port 443.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Maciej Żenczykowski <maze@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

random_ether_addr is a #define for eth_random_addr which is
generally preferred in kernel code by ~3:1

Convert the uses of random_ether_addr to enable removing the #define

Miscellanea:

o Convert &vfmac[0] to equivalent vfmac and avoid unnecessary line wrap
Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The GPIO subsystem provides dummy GPIO consumer functions if GPIOLIB is
not enabled. Hence drivers that depend on GPIOLIB, but use GPIO consumer
functionality only, can still be compiled if GPIOLIB is not enabled.

Relax the dependency on GPIOLIB if COMPILE_TEST is enabled, where
appropriate.
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

On Intel platforms (Skylake and newer), ASPM support in r8169 is the
last missing puzzle to let CPU's Package C-State reaches PC8.  Without
ASPM support, the CPU cannot reach beyond PC3. PC8 can save additional
~3W in comparison with PC3 on a Coffee Lake platform, Dell G3 3779.

This is based on the work from Chunhao Lin <hau@realtek.com>.
Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Enable or disable ASPM should be done in PCI core instead of in the
device driver.

Commit ba04c7c9 ("r8169: disable ASPM") uses
pci_disable_link_state() to disable ASPM, but it's not the best way to
do it. If the device really wants to disable ASPM, we can use a quirk in
PCI core to prevent the PCI core from setting ASPM before probe.

Let's remove pci_disable_link_state() for now. Use PCI core quirks if
any regression happens.
Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This commit makes BBR use only the MSS (without any headers) to
calculate pacing rates when internal TCP-layer pacing is used.

This is necessary to achieve the correct pacing behavior in this case,
since tcp_internal_pacing() uses only the payload length to calculate
pacing delays.
Signed-off-by: Kevin Yang <yyd@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Sebastian Andrzej Siewior says:

====================
net/usb: Use irqsave in USB's complete callback

This is about using _irqsave() primitives in the completion callback in
order to get rid of local_irq_save() in __usb_hcd_giveback_urb().
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The USB completion callback does not disable interrupts while acquiring
the lock. We want to remove the local_irq_disable() invocation from
__usb_hcd_giveback_urb() and therefore it is required for the callback
handler to disable the interrupts while acquiring the lock.
The callback may be invoked either in IRQ or BH context depending on the
USB host controller.
Use the _irqsave() variant of the locking primitives.

Cc: Petko Manolov <petkan@nucleusys.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

The USB completion callback does not disable interrupts while acquiring
the lock. We want to remove the local_irq_disable() invocation from
__usb_hcd_giveback_urb() and therefore it is required for the callback
handler to disable the interrupts while acquiring the lock.
The callback may be invoked either in IRQ or BH context depending on the
USB host controller.
Use the _irqsave() variant of the locking primitives.

Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

The USB completion callback does not disable interrupts while acquiring
the lock. We want to remove the local_irq_disable() invocation from
__usb_hcd_giveback_urb() and therefore it is required for the callback
handler to disable the interrupts while acquiring the lock.
The callback may be invoked either in IRQ or BH context depending on the
USB host controller.
Use the _irqsave() variant of the locking primitives.

Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

The USB completion callback does not disable interrupts while acquiring
the lock. We want to remove the local_irq_disable() invocation from
__usb_hcd_giveback_urb() and therefore it is required for the callback
handler to disable the interrupts while acquiring the lock.
The callback may be invoked either in IRQ or BH context depending on the
USB host controller.
Use the _irqsave() variant of the locking primitives.

Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

The USB completion callback does not disable interrupts while acquiring
the lock. We want to remove the local_irq_disable() invocation from
__usb_hcd_giveback_urb() and therefore it is required for the callback
handler to disable the interrupts while acquiring the lock.
The callback may be invoked either in IRQ or BH context depending on the
USB host controller.
Use the _irqsave() variant of the locking primitives.

Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Joakim Tjernlund <joakim.tjernlund@infinera.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Sebastian Andrzej Siewior says:

====================
ISDN: use irqsave() in URB completion + usb_fill_int_urb

This series is mostly about using _irqsave() primitives in the
completion callback in order to get rid of local_irq_save() in
__usb_hcd_giveback_urb(). While at it, I also tried to move drivers to
use usb_fill_int_urb() otherwise it is hard find users of a certain API.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The USB completion callback does not disable interrupts while acquiring
the ->lock. We want to remove the local_irq_disable() invocation from
__usb_hcd_giveback_urb() and therefore it is required for the callback
handler to disable the interrupts while acquiring the lock.
The callback may be invoked either in IRQ or BH context depending on the
USB host controller.
Use the _irqsave() variant of the locking primitives.

Cc: Karsten Keil <isdn@linux-pingi.de>
Cc: netdev@vger.kernel.org
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

Using usb_fill_int_urb() helps to find code which initializes an
URB. A grep for members of the struct (like ->complete) reveal lots
of other things, too.

Cc: Karsten Keil <isdn@linux-pingi.de>
Cc: netdev@vger.kernel.org
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

Using usb_fill_int_urb() helps to find code which initializes an
URB. A grep for members of the struct (like ->complete) reveal lots
of other things, too.

The `interval' parameter is now set differently on HS and SS. The
argument is fed from bInterval so it should be the right thing to do.

Cc: Karsten Keil <isdn@linux-pingi.de>
Cc: netdev@vger.kernel.org
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>