linux-2.6-microblaze.git
6 years agonet/ipv6: Fix ip6_convert_metrics() bug
Eric Dumazet [Thu, 19 Apr 2018 16:14:53 +0000 (09:14 -0700)]
net/ipv6: Fix ip6_convert_metrics() bug

If ip6_convert_metrics() fails to allocate memory, it should not
overwrite rt->fib6_metrics or we risk a crash later as syzbot found.

BUG: KASAN: null-ptr-deref in atomic_read include/asm-generic/atomic-instrumented.h:21 [inline]
BUG: KASAN: null-ptr-deref in refcount_sub_and_test+0x92/0x330 lib/refcount.c:179
Read of size 4 at addr 0000000000000044 by task syzkaller832429/4487

CPU: 1 PID: 4487 Comm: syzkaller832429 Not tainted 4.16.0+ #6
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0x1b9/0x294 lib/dump_stack.c:113
 kasan_report_error mm/kasan/report.c:352 [inline]
 kasan_report.cold.7+0x6d/0x2fe mm/kasan/report.c:412
 check_memory_region_inline mm/kasan/kasan.c:260 [inline]
 check_memory_region+0x13e/0x1b0 mm/kasan/kasan.c:267
 kasan_check_read+0x11/0x20 mm/kasan/kasan.c:272
 atomic_read include/asm-generic/atomic-instrumented.h:21 [inline]
 refcount_sub_and_test+0x92/0x330 lib/refcount.c:179
 refcount_dec_and_test+0x1a/0x20 lib/refcount.c:212
 fib6_info_destroy+0x2d0/0x3c0 net/ipv6/ip6_fib.c:206
 fib6_info_release include/net/ip6_fib.h:304 [inline]
 ip6_route_info_create+0x677/0x3240 net/ipv6/route.c:3020
 ip6_route_add+0x23/0xb0 net/ipv6/route.c:3030
 inet6_rtm_newroute+0x142/0x160 net/ipv6/route.c:4406
 rtnetlink_rcv_msg+0x466/0xc10 net/core/rtnetlink.c:4648
 netlink_rcv_skb+0x172/0x440 net/netlink/af_netlink.c:2448
 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4666
 netlink_unicast_kernel net/netlink/af_netlink.c:1310 [inline]
 netlink_unicast+0x58b/0x740 net/netlink/af_netlink.c:1336
 netlink_sendmsg+0x9f0/0xfa0 net/netlink/af_netlink.c:1901
 sock_sendmsg_nosec net/socket.c:629 [inline]
 sock_sendmsg+0xd5/0x120 net/socket.c:639
 ___sys_sendmsg+0x805/0x940 net/socket.c:2117
 __sys_sendmsg+0x115/0x270 net/socket.c:2155
 SYSC_sendmsg net/socket.c:2164 [inline]
 SyS_sendmsg+0x29/0x30 net/socket.c:2162
 do_syscall_64+0x29e/0x9d0 arch/x86/entry/common.c:287
 entry_SYSCALL_64_after_hwframe+0x42/0xb7

Fixes: d4ead6b34b67 ("net/ipv6: move metrics from dst to rt6_info")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: David Ahern <dsa@cumulusnetworks.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Acked-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'qed-Use-trust-mode-to-override-forced-MAC'
David S. Miller [Fri, 20 Apr 2018 15:26:37 +0000 (11:26 -0400)]
Merge branch 'qed-Use-trust-mode-to-override-forced-MAC'

Shahed Shaikh says:

====================
qed* : Use trust mode to override forced MAC

This patchset adds a support to override forced MAC (MAC set by PF for a VF)
when trust mode is enabled using

First patch adds a real change to use .ndo_set_vf_trust to override forced MAC
and allow user to change VFs from VF interface itself.

Second patch takes care of a corner case, where MAC change from VF won't
take effect when VF interface is down, by introducing a new TLV
(a way to send message from VF to PF) to give a hint to PF to update
its bulletin board.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoqed* : Add new TLV to request PF to update MAC in bulletin board
Shahed Shaikh [Thu, 19 Apr 2018 12:50:12 +0000 (05:50 -0700)]
qed* : Add new TLV to request PF to update MAC in bulletin board

There may be a need for VF driver to request PF to explicitly update its
bulletin with a MAC address.
e.g. When user assigns a MAC address to VF while VF is still down,
and PF's bulletin board contains different MAC address, in this case,
when VF's interface is brought up, it gets loaded with MAC address from
bulletin board which is not desirable.

To handle this corner case, we need a new TLV to request PF to update
its bulletin board with suggested MAC.

This request will be honored only for trusted VFs.

Signed-off-by: Shahed Shaikh <shahed.shaikh@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoqed* : use trust mode to allow VF to override forced MAC
Shahed Shaikh [Thu, 19 Apr 2018 12:50:11 +0000 (05:50 -0700)]
qed* : use trust mode to allow VF to override forced MAC

As per existing behavior, when PF sets a MAC address for a VF
(also called as forced MAC), VF is not allowed to change its
MAC address afterwards.
This puts the limitation on few use cases such as bonding of VFs,
where bonding driver asks VF to change its MAC address.

This patch uses a VF trust mode to allow VF to change its MAC address
in spite PF has set a forced MAC for that VF.

Signed-off-by: Shahed Shaikh <shahed.shaikh@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'geneve-mtu'
David S. Miller [Fri, 20 Apr 2018 15:20:06 +0000 (11:20 -0400)]
Merge branch 'geneve-mtu'

Alexey Kodanev says:

====================
geneve: verify user specified MTU or adjust with a lower device

The first two patches don't introduce any functional changes and
contain minor cleanups for code readability.

The last one adds a new function geneve_link_config() similar to the
other tunnels. The function will be used on a new link creation or
when 'remote' parameter is changed. It adjusts a user specified MTU
or, if it finds a lower device, tunes the tunnel MTU using it.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agogeneve: configure MTU based on a lower device
Alexey Kodanev [Thu, 19 Apr 2018 12:42:32 +0000 (15:42 +0300)]
geneve: configure MTU based on a lower device

Currently, on a new link creation or when 'remote' address parameter
is updated, an MTU is not changed and always equals 1500. When a lower
device has a larger MTU, it might not be efficient, e.g. for UDP, and
requires the manual MTU adjustments to match the MTU of the lower
device.

This patch tries to automate this process, finds a lower device using
the 'remote' address parameter, then uses its MTU to tune GENEVE's MTU:
  * on a new link creation
  * when 'remote' parameter is changed

Also with this patch, the MTU from a user, on a new link creation, is
passed to geneve_change_mtu() where it is verified, and MTU adjustments
with a lower device is skipped in that case. Prior that change, it was
possible to set the invalid MTU values on a new link creation.

Signed-off-by: Alexey Kodanev <alexey.kodanev@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agogeneve: check MTU for a minimum in geneve_change_mtu()
Alexey Kodanev [Thu, 19 Apr 2018 12:42:31 +0000 (15:42 +0300)]
geneve: check MTU for a minimum in geneve_change_mtu()

geneve_change_mtu() will be used not only as ndo_change_mtu() callback,
but also to verify a user specified MTU on a new link creation in the
next patch.

Signed-off-by: Alexey Kodanev <alexey.kodanev@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agogeneve: cleanup hard coded value for Ethernet header length
Alexey Kodanev [Thu, 19 Apr 2018 12:42:30 +0000 (15:42 +0300)]
geneve: cleanup hard coded value for Ethernet header length

Use ETH_HLEN instead and introduce two new macros: GENEVE_IPV4_HLEN
and GENEVE_IPV6_HLEN that include Ethernet header length, corresponded
IP header length and GENEVE_BASE_HLEN.

Signed-off-by: Alexey Kodanev <alexey.kodanev@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agogeneve: remove white-space before '#if IS_ENABLED(CONFIG_IPV6)'
Alexey Kodanev [Thu, 19 Apr 2018 12:42:29 +0000 (15:42 +0300)]
geneve: remove white-space before '#if IS_ENABLED(CONFIG_IPV6)'

Signed-off-by: Alexey Kodanev <alexey.kodanev@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoteam: account for oper state
George Wilkie [Thu, 19 Apr 2018 10:34:14 +0000 (11:34 +0100)]
team: account for oper state

Account for operational state when determining port linkup state,
as per Documentation/networking/operstates.txt.

Signed-off-by: George Wilkie <gwilkie@vyatta.att-mail.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'tipc-Confgiuration-of-MTU-for-media-UDP'
David S. Miller [Fri, 20 Apr 2018 15:04:06 +0000 (11:04 -0400)]
Merge branch 'tipc-Confgiuration-of-MTU-for-media-UDP'

GhantaKrishnamurthy MohanKrishna says:

====================
tipc: Confgiuration of MTU for media UDP

Systematic measurements have shown that an emulated MTU of 14k for
UDP bearers is the optimal value for maximal throughput. Accordingly,
the default MTU of UDP bearers is changed to 14k.

We also provide users with a fallback option from this value,
by providing support to configure MTU for UDP bearers. The following
options are introduced which are symmetrical to the design of
confguring link tolerance.

- Configure media with new MTU value, which will take effect on
links going up after the moment it was configured. Alternatively,
the bearer has to be disabled and re-enabled, for existing links to
reflect the configured value.

- Configure bearer with new MTU value, which take effect on
running links dynamically.

Please note:
- User has to change MTU at both endpoints, otherwise the link
will fall back to smallest MTU after a reset.
- Failover from a link with higher MTU to a link with lower MTU
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotipc: confgiure and apply UDP bearer MTU on running links
GhantaKrishnamurthy MohanKrishna [Thu, 19 Apr 2018 09:06:20 +0000 (11:06 +0200)]
tipc: confgiure and apply UDP bearer MTU on running links

Currently, we have option to configure MTU of UDP media. The configured
MTU takes effect on the links going up after that moment. I.e, a user
has to reset bearer to have new value applied across its links. This is
confusing and disturbing on a running cluster.

We now introduce the functionality to change the default UDP bearer MTU
in struct tipc_bearer. Additionally, the links are updated dynamically,
without any need for a reset, when bearer value is changed. We leverage
the existing per-link functionality and the design being symetrical to
the confguration of link tolerance.

Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: GhantaKrishnamurthy MohanKrishna <mohan.krishna.ghanta.krishnamurthy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotipc: implement configuration of UDP media MTU
GhantaKrishnamurthy MohanKrishna [Thu, 19 Apr 2018 09:06:19 +0000 (11:06 +0200)]
tipc: implement configuration of UDP media MTU

In previous commit, we changed the default emulated MTU for UDP bearers
to 14k.

This commit adds the functionality to set/change the default value
by configuring new MTU for UDP media. UDP bearer(s) have to be disabled
and enabled back for the new MTU to take effect.

Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: GhantaKrishnamurthy MohanKrishna <mohan.krishna.ghanta.krishnamurthy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotipc: set default MTU for UDP media
GhantaKrishnamurthy MohanKrishna [Thu, 19 Apr 2018 09:06:18 +0000 (11:06 +0200)]
tipc: set default MTU for UDP media

Currently, all bearers are configured with MTU value same as the
underlying L2 device. However, in case of bearers with media type
UDP, higher throughput is possible with a fixed and higher emulated
MTU value than adapting to the underlying L2 MTU.

In this commit, we introduce a parameter mtu in struct tipc_media
and a default value is set for UDP. A default value of 14k
was determined by experimentation and found to have a higher throughput
than 16k. MTU for UDP bearers are assigned the above set value of
media MTU.

Acked-by: Ying Xue <ying.xue@windriver.com>
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: GhantaKrishnamurthy MohanKrishna <mohan.krishna.ghanta.krishnamurthy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoliquidio: Added ndo_get_vf_stats support
Intiyaz Basha [Thu, 19 Apr 2018 06:18:28 +0000 (23:18 -0700)]
liquidio: Added ndo_get_vf_stats support

Added the ndo to gather VF statistics through the PF.

Collect VF statistics via mailbox from VF.

Signed-off-by: Intiyaz Basha <intiyaz.basha@cavium.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'ave-fix-the-activation-issues-for-some-UniPhier-SoCs'
David S. Miller [Fri, 20 Apr 2018 14:49:41 +0000 (10:49 -0400)]
Merge branch 'ave-fix-the-activation-issues-for-some-UniPhier-SoCs'

Kunihiko Hayashi says:

====================
ave: fix the activation issues for some UniPhier SoCs

This add the following stuffs to fix the activation issues and satisfy
requirements for AVE ethernet driver implemented on some UniPhier SoCs.

- Add support for additional necessary clocks and resets, because the kernel
  is stalled on Pro4 due to lack of them.

- Check whether the SoC supports the specified phy-mode

- Add DT property support indicating system controller that has the feature
  for configurating phy-mode including built-in phy on LD11.

v1: https://www.spinics.net/lists/netdev/msg494904.html

Changes since v1:
- Add 'Reviewed-by' lines
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ethernet: ave: add support for phy-mode setting of system controller
Kunihiko Hayashi [Thu, 19 Apr 2018 07:24:55 +0000 (16:24 +0900)]
net: ethernet: ave: add support for phy-mode setting of system controller

This patch adds support for specifying system controller that configures
phy-mode setting.

According to the DT property "phy-mode", it's necessary to configure the
controller, which is used to choose the settings of the MAC suitable,
for example, mdio pin connections, internal clocks, and so on.

Supported phy-modes are SoC-dependent. The driver allows phy-mode to set
"internal" if the SoC has a built-in PHY, and {"mii", "rmii", "rgmii"}
if the SoC supports each mode. So we have to check whether the phy-mode
is valid or not.

This adds the following features for each SoC:
- check whether the SoC supports the specified phy-mode
- configure the controller accroding to phy-mode

The DT property accepts one argument to distinguish them for multiple MAC
instances.

ethernet@65000000 {
...
socionext,syscon-phy-mode = <&soc_glue 0>;
};

ethernet@65200000 {
...
socionext,syscon-phy-mode = <&soc_glue 1>;
};

Signed-off-by: Kunihiko Hayashi <hayashi.kunihiko@socionext.com>
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodt-bindings: net: ave: add syscon-phy-mode property to configure phy-mode setting
Kunihiko Hayashi [Thu, 19 Apr 2018 07:24:54 +0000 (16:24 +0900)]
dt-bindings: net: ave: add syscon-phy-mode property to configure phy-mode setting

Add "socionext,syscon-phy-mode" property to specify system controller that
configures the settings about phy-mode.

Signed-off-by: Kunihiko Hayashi <hayashi.kunihiko@socionext.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ethernet: ave: add multiple clocks and resets support as required property
Kunihiko Hayashi [Thu, 19 Apr 2018 07:24:53 +0000 (16:24 +0900)]
net: ethernet: ave: add multiple clocks and resets support as required property

When the link is becoming up for Pro4 SoC, the kernel is stalled
due to some missing clocks and resets.

The AVE block for Pro4 is connected to the GIO bus in the SoC.
Without its clock/reset, the access to the AVE register makes the
system stall.

In the same way, another MAC clock for Giga-bit Connection and
the PHY clock are also required for Pro4 to activate the Giga-bit feature
and to recognize the PHY.

To satisfy these requirements, this patch adds support for multiple clocks
and resets, and adds the clock-names and reset-names to the binding because
we need to distinguish clock/reset for the AVE main block and the others.

Also, make the resets a required property. Currently, "reset is
optional" relies on that the bootloader or firmware has deasserted
the reset before booting the kernel.  Drivers should work without
such expectation.

Fixes: 4c270b55a5af ("net: ethernet: socionext: add AVE ethernet driver")
Suggested-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Signed-off-by: Kunihiko Hayashi <hayashi.kunihiko@socionext.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: mdio-boardinfo: Allow recursive mdiobus_register()
Andrew Lunn [Thu, 19 Apr 2018 00:00:47 +0000 (02:00 +0200)]
net: phy: mdio-boardinfo: Allow recursive mdiobus_register()

mdiobus_register will search for any mdiobus board info registered for
the bus being registered. If found, it will probe devices on the bus.
That device, if for example it is an ethernet switch, may then try to
register an mdio bus. Thus we need to allow recursive calls to
mdiobus_register.

Holding the mdio_board_lock will cause a deadlock during this
recursion. Release the lock and use list_for_each_entry_safe.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'Amiga-xsurf100'
David S. Miller [Thu, 19 Apr 2018 20:11:12 +0000 (16:11 -0400)]
Merge branch 'Amiga-xsurf100'

Michael Schmitz says:

====================
New network driver for Amiga X-Surf 100 (m68k)

[This is a resend of my v3 series which was based on the wrong version and
tree. Only substantial change is to Asix AX99796B PHY driver.]

This patch series adds support for the Individual Computers X-Surf 100
network card for m68k Amiga, a network adapter based on the AX88796 chip set.

The driver was originally written for kernel version 3.19 by Michael Karcher
(see CC:), and adapted to 4.16+ for submission to netdev by me. Questions
regarding motivation for some of the changes are probably best directed at
Michael Karcher.

The driver has been tested by Adrian <glaubitz@physik.fu-berlin.de> who will
send his Tested-by tag separately.

A few changes to the ax88796 driver were required:
- to read the MAC address, some setup of the ax99796 chip must be done,
- attach to the MII bus only on device open to allow module unloading,
- allow to supersede ax_block_input/ax_block_output by card-specific
  optimized code,
- use an optional interrupt status callback to allow easier sharing of the
  card interrupt,
- set IRQF_SHARED if platform IRQ resource is marked shareable

The Asix Electronix PHY used on the X-Surf 100 is buggy, and causes the
software reset to hang if the previous command sent to the PHY was also
a soft reset. This bug requires addition of a PHY driver for Asix PHYs
to provide a fixed .soft_reset function, included in this series.

Some additional cleanup:
- do not attempt to free IRQ in ax_remove (complements 82533ad9a1c),
- clear platform drvdata on probe fail and module remove.

Changes since v1:

Raised in review by Andrew Lunn:
- move MII code around to avoid need for forward declaration,
- combine patches 2 and 7 to add cleanup in error path

Changes since v2:

- corrected authorship attribution to Michael Karcher

Suggested by Geert Uytterhoeven:
- use ei_local->reset_8390() instead of duplicating ax_reset_8390(),
- use %pR to format struct resource pointers,
- assign pdev and xs100 pointers in declaration,
- don't split error messages,
- change Kconfig logic to only require XSURF100 set on Amiga

Suggested by Andrew Lunn:
- add COMPILE_TEST to ax88796 Kconfig options,
- use new Asix PHY driver for X-Surf 100

Suggested by Andrew Lunn/Finn Thain:
- declare struct sk_buff in ax88796.h,
- correct whitespace error in ax88796.h

Changes since v3:

- various checkpatch cleanup

Andrew Lunn:
- don't duplicate genphy_soft_reset in Asix PHY driver, just call
  genphy_soft_reset after writing zero to control register
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet-next: New ax88796 platform driver for Amiga X-Surf 100 Zorro board (m68k)
Michael Karcher [Thu, 19 Apr 2018 02:05:26 +0000 (14:05 +1200)]
net-next: New ax88796 platform driver for Amiga X-Surf 100 Zorro board (m68k)

Add platform device driver to populate the ax88796 platform data from
information provided by the XSurf100 zorro device driver. The ax88796
module will be loaded through this module's probe function.

Signed-off-by: Michael Karcher <kernel@mkarcher.dialup.fu-berlin.de>
Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet-next: ax88796: release platform device drvdata on probe error and module remove
Michael Schmitz [Thu, 19 Apr 2018 02:05:25 +0000 (14:05 +1200)]
net-next: ax88796: release platform device drvdata on probe error and module remove

The net device struct pointer is stored as platform device drvdata on
module probe - clear the drvdata entry on probe fail there, as well as
when unloading the module.

Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet-next: ax88796: set IRQF_SHARED flag when IRQ resource is marked as shareable
Michael Karcher [Thu, 19 Apr 2018 02:05:24 +0000 (14:05 +1200)]
net-next: ax88796: set IRQF_SHARED flag when IRQ resource is marked as shareable

On the Amiga X-Surf100, the network card interrupt is shared with many
other interrupt sources, so requires the IRQF_SHARED flag to register.

Signed-off-by: Michael Karcher <kernel@mkarcher.dialup.fu-berlin.de>
Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet-next: ax88796: add interrupt status callback to platform data
Michael Karcher [Thu, 19 Apr 2018 02:05:23 +0000 (14:05 +1200)]
net-next: ax88796: add interrupt status callback to platform data

To be able to tell the ax88796 driver whether it is sensible to enter
the 8390 interrupt handler, an "is this interrupt caused by the 88796"
callback has been added to the ax_plat_data structure (with NULL being
compatible to the previous behaviour).

Signed-off-by: Michael Karcher <kernel@mkarcher.dialup.fu-berlin.de>
Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet-next: ax88796: Add block_input/output hooks to ax_plat_data
Michael Karcher [Thu, 19 Apr 2018 02:05:22 +0000 (14:05 +1200)]
net-next: ax88796: Add block_input/output hooks to ax_plat_data

Add platform specific hooks for block transfer reads/writes of packet
buffer data, superseding the default provided ax_block_input/output.
Currently used for m68k Amiga XSurf100.

Signed-off-by: Michael Karcher <kernel@mkarcher.dialup.fu-berlin.de>
Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet-next: ax88796: Do not free IRQ in ax_remove() (already freed in ax_close()).
Michael Karcher [Thu, 19 Apr 2018 02:05:21 +0000 (14:05 +1200)]
net-next: ax88796: Do not free IRQ in ax_remove() (already freed in ax_close()).

This complements the fix in 82533ad9a1c ("net: ethernet: ax88796:
don't call free_irq without request_irq first") that removed the
free_irq call in the error path of probe, to also not call free_irq
when remove is called to revert the effects of probe.

Fixes: 82533ad9a1c (net: ethernet: ax88796: don't call free_irq without request_irq first)
Signed-off-by: Michael Karcher <kernel@mkarcher.dialup.fu-berlin.de>
Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
Reviewed-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet-next: ax88796: Attach MII bus only when open
Michael Karcher [Thu, 19 Apr 2018 02:05:20 +0000 (14:05 +1200)]
net-next: ax88796: Attach MII bus only when open

Call ax_mii_init in ax_open(), and unregister/remove mdiobus resources
in ax_close().

This is needed to be able to unload the module, as the module is busy
while the MII bus is attached.

Signed-off-by: Michael Karcher <kernel@mkarcher.dialup.fu-berlin.de>
Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet-next: ax88796: Fix MAC address reading
Michael Karcher [Thu, 19 Apr 2018 02:05:19 +0000 (14:05 +1200)]
net-next: ax88796: Fix MAC address reading

To read the MAC address from the (virtual) SAprom, the remote DMA
unit needs to be set up like for every other process access to card-local
memory.

Signed-off-by: Michael Karcher <kernel@mkarcher.dialup.fu-berlin.de>
Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet-next: phy: new Asix Electronics PHY driver
Michael Schmitz [Thu, 19 Apr 2018 02:05:18 +0000 (14:05 +1200)]
net-next: phy: new Asix Electronics PHY driver

The Asix Electronics PHY found on the X-Surf 100 Amiga Zorro network
card by Individual Computers is buggy, and needs the reset bit toggled
as workaround to make a PHY soft reset succeed.

Add workaround driver just for this special case.

Suggested in xsurf100 patch series review by Andrew Lunn <andrew@lunn.ch>

Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'Modernize-mdio-gpio'
David S. Miller [Thu, 19 Apr 2018 19:59:12 +0000 (15:59 -0400)]
Merge branch 'Modernize-mdio-gpio'

Andrew Lunn says:

====================
Modernize mdio-gpio

This patchset is inspired by a previous version by Linus Walleij

It reworks the mdio-gpio code to make use of gpio descriptors instead
of gpio numbers. However compared to the previous version, it retains
support for platform devices. It does however remove the platform_data
header file. The needed GPIOs are now passed by making use of a gpiod
lookup table. e.g:

static struct gpiod_lookup_table zii_scu_mdio_gpiod_table = {
.dev_id = "mdio-gpio.0",
.table = {
GPIO_LOOKUP_IDX("gpio_ich", 17, NULL, MDIO_GPIO_MDC,
GPIO_ACTIVE_HIGH),
GPIO_LOOKUP_IDX("gpio_ich", 2, NULL, MDIO_GPIO_MDIO,
GPIO_ACTIVE_HIGH),
GPIO_LOOKUP_IDX("gpio_ich", 21, NULL, MDIO_GPIO_MDO,
GPIO_ACTIVE_LOW),
},
};
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: mdio-gpio: Remove redundant platform data header
Andrew Lunn [Wed, 18 Apr 2018 23:02:59 +0000 (01:02 +0200)]
net: phy: mdio-gpio: Remove redundant platform data header

The platform data header file is now unused. Remove it, but add
an extra include which it brought in.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: mdio-gpio: Add #defines for the GPIO index's
Andrew Lunn [Wed, 18 Apr 2018 23:02:58 +0000 (01:02 +0200)]
net: phy: mdio-gpio: Add #defines for the GPIO index's

The GPIOs are described in device tree using a list, without names.
Add defines to indicate what each index in the list means. These
defines should also be used by platform devices passing GPIOs via a
GPIO lookup table.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: mdio-gpio: Parse properties directly into bitbang structure
Andrew Lunn [Wed, 18 Apr 2018 23:02:57 +0000 (01:02 +0200)]
net: phy: mdio-gpio: Parse properties directly into bitbang structure

The same parsing code can be used for both OF and platform devices, if
the platform device uses a gpiod_lookup_table. Parse these properties
directly into the bitbang structure, rather than use an intermediate
platform data structure.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: mdio-gpio: Move allocation for bitbanging data
Andrew Lunn [Wed, 18 Apr 2018 23:02:56 +0000 (01:02 +0200)]
net: phy: mdio-gpio: Move allocation for bitbanging data

Moving the allocation of this structure to the probe function is a
step towards making it the core data structure of the driver.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: mdio-gpio: Swap to using gpio descriptors
Andrew Lunn [Wed, 18 Apr 2018 23:02:55 +0000 (01:02 +0200)]
net: phy: mdio-gpio: Swap to using gpio descriptors

This simplifies the code, removing the need to handle active low
flags, etc.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: mdio-gpio: Remove support for IRQs in platform data
Andrew Lunn [Wed, 18 Apr 2018 23:02:54 +0000 (01:02 +0200)]
net: phy: mdio-gpio: Remove support for IRQs in platform data

No current devices use IRQs in platform data, so remove support for
it. The MDIO core will also initialise the new bus such that all
addresses are polled, so remove the unneeded re-initialisation.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: mdio-gpio: remove support for phy mask
Andrew Lunn [Wed, 18 Apr 2018 23:02:53 +0000 (01:02 +0200)]
net: phy: mdio-gpio: remove support for phy mask

This is not needed any more by devices using platform data, so remove
it.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: mdio-gpio: remove support for ignoring turn around
Andrew Lunn [Wed, 18 Apr 2018 23:02:52 +0000 (01:02 +0200)]
net: phy: mdio-gpio: remove support for ignoring turn around

This is not needed any more by devices using platform data, so remove
it.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: mdio-bitbang: Remove reset support
Andrew Lunn [Wed, 18 Apr 2018 23:02:51 +0000 (01:02 +0200)]
net: phy: mdio-bitbang: Remove reset support

The mdio-gpio driver was the only user of the interface reset option.
Since it no longer uses it, remove it from the bit banging code.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: mdio-gpio: Remove reset function
Andrew Lunn [Wed, 18 Apr 2018 23:02:50 +0000 (01:02 +0200)]
net: phy: mdio-gpio: Remove reset function

The platform data can contain a function to call to reset
the bit banging interface. It is not used, so remove it.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy_ mdio-gpio: Fixup , which should be ;
Andrew Lunn [Wed, 18 Apr 2018 23:02:49 +0000 (01:02 +0200)]
net: phy_ mdio-gpio: Fixup , which should be ;

Seems like an old typ0.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'ipv6-followup-to-fib6_info-change'
David S. Miller [Thu, 19 Apr 2018 19:40:14 +0000 (15:40 -0400)]
Merge branch 'ipv6-followup-to-fib6_info-change'

David Ahern says:

====================
net/ipv6: followup to fib6_info change

Followup to fib change for IPv6.

First 2 patches rename fib6_info struct elements to match its name,
and rename addrconf_dst_alloc to match what it returns.

Patches 3-7 refactor the code to remove the need for fib6_idev reducing
fib6_info by another 8 bytes to 200 bytes.

Patch 8 fixes the gfp flags argument to addrconf_prefix_route in a
couple of places.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Fix gfp_flags arg to addrconf_prefix_route
David Ahern [Wed, 18 Apr 2018 22:39:06 +0000 (15:39 -0700)]
net/ipv6: Fix gfp_flags arg to addrconf_prefix_route

Eric noticed that __ipv6_ifa_notify is called under rcu_read_lock, so
the gfp argument to addrconf_prefix_route can not be GFP_KERNEL.

While scrubbing other calls I noticed addrconf_addr_gen has one
place with GFP_ATOMIC that can be GFP_KERNEL.

Fixes: acb54e3cba404 ("net/ipv6: Add gfp_flags to route add functions")
Reported-by: syzbot+2add39b05179b31f912f@syzkaller.appspotmail.com
Reported-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Remove fib6_idev
David Ahern [Wed, 18 Apr 2018 22:39:05 +0000 (15:39 -0700)]
net/ipv6: Remove fib6_idev

fib6_idev can be obtained from __in6_dev_get on the nexthop device
rather than caching it in the fib6_info. Remove it.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Remove compare of fib6_idev from rt6_duplicate_nexthop
David Ahern [Wed, 18 Apr 2018 22:39:04 +0000 (15:39 -0700)]
net/ipv6: Remove compare of fib6_idev from rt6_duplicate_nexthop

After 4832c30d5458 ("net: ipv6: put host and anycast routes on device
with address") the comparison of idev does not add value since it
correlates to the nexthop device which is already compared. Remove
the idev comparison.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Change ip6_route_get_saddr to get dev from route
David Ahern [Wed, 18 Apr 2018 22:39:03 +0000 (15:39 -0700)]
net/ipv6: Change ip6_route_get_saddr to get dev from route

Prior to 4832c30d5458 ("net: ipv6: put host and anycast routes on device
with address") host routes and anycast routes were installed with the
device set to loopback (or VRF device once that feature was added). In the
older code dst.dev was set to loopback (needed for packet tx) and rt6i_idev
was used to denote the actual interface.

Commit 4832c30d5458 changed the code to have dst.dev pointing to the real
device with the switch to lo or vrf device done on dst clones. As a
consequence of this change ip6_route_get_saddr can just pass the nexthop
device to ipv6_dev_get_saddr.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Remove unnecessary checks on fib6_idev
David Ahern [Wed, 18 Apr 2018 22:39:02 +0000 (15:39 -0700)]
net/ipv6: Remove unnecessary checks on fib6_idev

Prior to 4832c30d5458 ("net: ipv6: put host and anycast routes on device
with address") host routes and anycast routes were installed with the
device set to loopback (or VRF device once that feature was added). In the
older code dst.dev was set to loopback (needed for packet tx) and rt6i_idev
was used to denote the actual interface.

Commit 4832c30d5458 changed the code to have dst.dev pointing to the real
device with the switch to lo or vrf device done on dst clones. As a
consequence of this change a couple of device checks during route lookups
are no longer needed. Remove them.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Remove aca_idev
David Ahern [Wed, 18 Apr 2018 22:39:01 +0000 (15:39 -0700)]
net/ipv6: Remove aca_idev

aca_idev has only 1 user - inet6_fill_ifacaddr - and it only
wants the device index which can be extracted from the fib6_info
nexthop.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Rename addrconf_dst_alloc
David Ahern [Wed, 18 Apr 2018 22:39:00 +0000 (15:39 -0700)]
net/ipv6: Rename addrconf_dst_alloc

addrconf_dst_alloc now returns a fib6_info. Update the name
and its users to reflect the change.

Rename only; no functional change intended.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Rename fib6_info struct elements
David Ahern [Wed, 18 Apr 2018 22:38:59 +0000 (15:38 -0700)]
net/ipv6: Rename fib6_info struct elements

Change the prefix for fib6_info struct elements from rt6i_ to fib6_.
rt6i_pcpu and rt6i_exception_bucket are left as is given that they
point to rt6_info entries.

Rename only; not functional change intended.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: pskb_trim_rcsum() and CHECKSUM_COMPLETE are friends
Eric Dumazet [Wed, 18 Apr 2018 18:43:15 +0000 (11:43 -0700)]
net: pskb_trim_rcsum() and CHECKSUM_COMPLETE are friends

After working on IP defragmentation lately, I found that some large
packets defeat CHECKSUM_COMPLETE optimization because of NIC adding
zero paddings on the last (small) fragment.

While removing the padding with pskb_trim_rcsum(), we set skb->ip_summed
to CHECKSUM_NONE, forcing a full csum validation, even if all prior
fragments had CHECKSUM_COMPLETE set.

We can instead compute the checksum of the part we are trimming,
usually smaller than the part we keep.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet-next/hinic: add arm64 support
Zhao Chen [Wed, 18 Apr 2018 10:07:39 +0000 (06:07 -0400)]
net-next/hinic: add arm64 support

This patch enables arm64 platform support for the HINIC driver.

Signed-off-by: Zhao Chen <zhaochen6@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'TCP-data-delivery-and-ECN-stats-tracking'
David S. Miller [Thu, 19 Apr 2018 17:05:17 +0000 (13:05 -0400)]
Merge branch 'TCP-data-delivery-and-ECN-stats-tracking'

Yuchung Cheng says:

====================
tracking TCP data delivery and ECN stats

This patch series improve tracking the data delivery status
  1. minor improvement on SYN data
  2. accounting bytes delivered with CE marks
  3. exporting the delivery stats to applications

s.t. users can get better sense of TCP performance at per host,
per connection, and even per application message level.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: export packets delivery info
Yuchung Cheng [Wed, 18 Apr 2018 06:18:49 +0000 (23:18 -0700)]
tcp: export packets delivery info

Export data delivered and delivered with CE marks to
1) SNMP TCPDelivered and TCPDeliveredCE
2) getsockopt(TCP_INFO)
3) Timestamping API SOF_TIMESTAMPING_OPT_STATS

Note that for SCM_TSTAMP_ACK, the delivery info in
SOF_TIMESTAMPING_OPT_STATS is reported before the info
was fully updated on the ACK.

These stats help application monitor TCP delivery and ECN status
on per host, per connection, even per message level.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: track total bytes delivered with ECN CE marks
Yuchung Cheng [Wed, 18 Apr 2018 06:18:48 +0000 (23:18 -0700)]
tcp: track total bytes delivered with ECN CE marks

Introduce a new delivered_ce stat in tcp socket to estimate
number of packets being marked with CE bits. The estimation is
done via ACKs with ECE bit. Depending on the actual receiver
behavior, the estimation could have biases.

Since the TCP sender can't really see the CE bit in the data path,
so the sender is technically counting packets marked delivered with
the "ECE / ECN-Echo" flag set.

With RFC3168 ECN, because the ECE bit is sticky, this count can
drastically overestimate the nummber of CE-marked data packets

With DCTCP-style ECN this should be reasonably precise unless there
is loss in the ACK path, in which case it's not precise.

With AccECN proposal this can be made still more precise, even in
the case some degree of ACK loss.

However this is sender's best estimate of CE information.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: new helper to calculate newly delivered
Yuchung Cheng [Wed, 18 Apr 2018 06:18:47 +0000 (23:18 -0700)]
tcp: new helper to calculate newly delivered

Add new helper tcp_newly_delivered() to prepare the ECN accounting change.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: better delivery accounting for SYN-ACK and SYN-data
Yuchung Cheng [Wed, 18 Apr 2018 06:18:46 +0000 (23:18 -0700)]
tcp: better delivery accounting for SYN-ACK and SYN-data

the tcp_sock:delivered has inconsistent accounting for SYN and FIN.
1. it counts pure FIN
2. it counts pure SYN
3. it counts SYN-data twice
4. it does not count SYN-ACK

For congestion control perspective it does not matter much as C.C. only
cares about the difference not the aboslute value. But the next patch
would export this field to user-space so it's better to report the absolute
value w/o these caveats.

This patch counts SYN, SYN-ACK, or SYN-data delivery once always in
the "delivered" field.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv6: frags: fix a lockdep false positive
Eric Dumazet [Wed, 18 Apr 2018 01:11:44 +0000 (18:11 -0700)]
ipv6: frags: fix a lockdep false positive

lockdep does not know that the locks used by IPv4 defrag
and IPv6 reassembly units are of different classes.

It complains because of following chains :

1) sch_direct_xmit()        (lock txq->_xmit_lock)
    dev_hard_start_xmit()
     xmit_one()
      dev_queue_xmit_nit()
       packet_rcv_fanout()
        ip_check_defrag()
         ip_defrag()
          spin_lock()     (lock frag queue spinlock)

2) ip6_input_finish()
    ipv6_frag_rcv()       (lock frag queue spinlock)
     ip6_frag_queue()
      icmpv6_param_prob() (lock txq->_xmit_lock at some point)

We could add lockdep annotations, but we also can make sure IPv6
calls icmpv6_param_prob() only after the release of the frag queue spinlock,
since this naturally makes frag queue spinlock a leaf in lock hierarchy.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agohv_netvsc: Add NetVSP v6 and v6.1 into version negotiation
Haiyang Zhang [Tue, 17 Apr 2018 22:31:47 +0000 (15:31 -0700)]
hv_netvsc: Add NetVSP v6 and v6.1 into version negotiation

This patch adds the NetVSP v6 and 6.1 message structures, and includes
these versions into NetVSC/NetVSP version negotiation process.

Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agohv_netvsc: propogate Hyper-V friendly name into interface alias
Stephen Hemminger [Tue, 17 Apr 2018 21:25:30 +0000 (14:25 -0700)]
hv_netvsc: propogate Hyper-V friendly name into interface alias

This patch implement the 'Device Naming' feature of the Hyper-V
network device API. In Hyper-V on the host through the GUI or PowerShell
it is possible to enable the device naming feature which causes
the host to make available to the guest the name of the device.
This shows up in the RNDIS protocol as the friendly name.

The name has no particular meaning and is limited to 256 characters.
The value can only be set via PowerShell on the host, but could
be scripted for mass deployments. The default value is the
string 'Network Adapter' and since that is the same for all devices
and useless, the driver ignores it.

In Windows, the value goes into a registry key for use in SNMP
ifAlias. For Linux, this patch puts the value in the network
device alias property; where it is visible in ip tools and SNMP.

The host provided ifAlias is just a suggestion, and can be
overridden by later ip commands.

Also requires exporting dev_set_alias in netdev core.

Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'r8169-series-with-further-smaller-improvements'
David S. Miller [Thu, 19 Apr 2018 01:12:00 +0000 (21:12 -0400)]
Merge branch 'r8169-series-with-further-smaller-improvements'

Heiner Kallweit says:

====================
r8169: series with further smaller improvements

This series includes further smaller improvements.

Then I think the basic cleanup has been done and next step would be
preparing the switch to phylib.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: remove jumbo_tx_csum from chip config struct
Heiner Kallweit [Tue, 17 Apr 2018 21:36:12 +0000 (23:36 +0200)]
r8169: remove jumbo_tx_csum from chip config struct

According to the chip configuration entries only RTL8169 (ver <= 06)
supports tx checksumming for jumbo packets.
By the way: constant JUMBO_1K is a little misleading because it refers
to the standard packet size and not to a jumbo packet size.

By implementing this rule we can get rid of configuring tx checksumming
support per chip type.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: improve pci region handling
Heiner Kallweit [Tue, 17 Apr 2018 21:34:22 +0000 (23:34 +0200)]
r8169: improve pci region handling

The region to be used is always the first of type IORESOURCE_MEM.
We can implement this rule directly w/o having to specify which
region is the first one per configuration entry.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: drop member txd_version from struct rtl8169_private
Heiner Kallweit [Tue, 17 Apr 2018 21:33:03 +0000 (23:33 +0200)]
r8169: drop member txd_version from struct rtl8169_private

txd_version is used in rtl_init_one() only, so we can drop member
txd_version from struct rtl8169_private.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: improve rtl8169_get_mac_version
Heiner Kallweit [Tue, 17 Apr 2018 21:32:15 +0000 (23:32 +0200)]
r8169: improve rtl8169_get_mac_version

Certain entries in array mac_info[] are redundant, so remove them:
0x7cf, 0x2c200000 (VER 33): matched by entry 0x7c8, 0x2c000000
0x7cf, 0x28300000 (VER 26): matched by entry 0x7c8, 0x28000000
0x7cf, 0x3cb00000 (VER 24): matched by entry 0x7c8, 0x3c800000
0x7cf, 0x3c400000 (VER 22): matched by entry 0x7c8, 0x3c000000
0x7cf, 0x38500000 (VER 17): matched by entry 0x7c8, 0x38000000
0x7cf, 0x44900000 (VER 39): matched by entry 0x7c8, 0x44800000
0x7cf, 0x40b00000 (VER 30): matched by entry 0x7c8, 0x40800000
0x7cf, 0x40a00000 (VER 30): matched by entry 0x7c8, 0x40800000
0x7cf, 0x34a00000 (VER 09): matched by entry 0x7c8, 0x34800000
0x7cf, 0x24a00000 (VER 09): matched by entry 0x7c8, 0x24800000

In addition don't mask out bits 30 and 29 when printing the XID.
Most likely this is a relict from the times when the driver covered
RTL8169 chip version only.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: don't display tp->mmio_addr address
Heiner Kallweit [Tue, 17 Apr 2018 21:31:21 +0000 (23:31 +0200)]
r8169: don't display tp->mmio_addr address

For security reasons since commit ad67b74d2469 "printk: hash addresses
printed with %p" %p doesn't display the full address any longer.
We could switch to %px, but I think the pointer address doesn't
provide a real benefit, so remove printing the hashed address.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: drop member opts1_mask from struct rtl8169_private
Heiner Kallweit [Tue, 17 Apr 2018 21:30:29 +0000 (23:30 +0200)]
r8169: drop member opts1_mask from struct rtl8169_private

We can get rid of member opts1_mask and in addition save a few cpu
cycles in the hot path of rtl_rx().

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: change interrupt handler argument type
Heiner Kallweit [Tue, 17 Apr 2018 21:29:20 +0000 (23:29 +0200)]
r8169: change interrupt handler argument type

Code can be a little simplified by switching the interrupt handler
argument type to struct rtl8169_private *.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: change argument type of counters handling functions
Heiner Kallweit [Tue, 17 Apr 2018 21:28:28 +0000 (23:28 +0200)]
r8169: change argument type of counters handling functions

The counter handling functions don't deal with the net_device, so code
can be simplified by changing the argument type to
struct rtl8169_private *.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: change hw_start argument type
Heiner Kallweit [Tue, 17 Apr 2018 21:27:38 +0000 (23:27 +0200)]
r8169: change hw_start argument type

Code can be simplified by changing the argument type of hw_start
callbacks from struct net_device * to struct rtl8169_private *.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: remove rtl8169_map_to_asic
Heiner Kallweit [Tue, 17 Apr 2018 21:26:41 +0000 (23:26 +0200)]
r8169: remove rtl8169_map_to_asic

This function is very simple and used only once, so we can inline
the two statements.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: replace rx_buf_sz with a constant
Heiner Kallweit [Tue, 17 Apr 2018 21:25:46 +0000 (23:25 +0200)]
r8169: replace rx_buf_sz with a constant

rx_buf_sz is constant, so we don't have to pass it as parameter and
in general can replace it with a constant.

When working on this I noticed that also before in
rtl_set_rx_max_size() a value of 0x4000 is set, what is not in line
with the chip spec. According to the spec only bits 0..13 are used
and we set an effective value of zero therefore.
However, the driver still seems to work and due to potential side
effects I'm reluctant to make a change.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: remove unneeded check in rtl8169_rx_fill
Heiner Kallweit [Tue, 17 Apr 2018 21:24:43 +0000 (23:24 +0200)]
r8169: remove unneeded check in rtl8169_rx_fill

rtl8169_rx_fill() is called only once and directly before the call
array tp->Rx_databuff[] is filled with zero's. Therefore we don't
need this check.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: improve rtl8169_init_ring
Heiner Kallweit [Tue, 17 Apr 2018 21:23:35 +0000 (23:23 +0200)]
r8169: improve rtl8169_init_ring

This function doesn't use the net_device, therefore change the
parameter to type struct rtl8169_private * to simplify the code.
In addition we don't need the calculations in the memset
statements, we can use the size of the arrays directly.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: simplify rtl8169_alloc_rx_data
Heiner Kallweit [Tue, 17 Apr 2018 21:22:14 +0000 (23:22 +0200)]
r8169: simplify rtl8169_alloc_rx_data

dev->dev.parent has the same value as tp_to_dev(tp)
(set by SET_NETDEV_DEV() in rtl_init_one()) and we know it can't be NULL.
This allows us to simplify the code.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: switch to napi_schedule_irqoff
Heiner Kallweit [Tue, 17 Apr 2018 21:21:01 +0000 (23:21 +0200)]
r8169: switch to napi_schedule_irqoff

napi_schedule() is called from hard irq context, so we can switch to
napi_schedule_irqoff() and avoid some overhead.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: use constant NAPI_POLL_WAIT
Heiner Kallweit [Tue, 17 Apr 2018 21:20:03 +0000 (23:20 +0200)]
r8169: use constant NAPI_POLL_WAIT

We can use generic constant NAPI_POLL_WAIT instead of defining an own
constant for the same value.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: use skb_copy_to_linear_data in rtl8169_try_rx_copy
Heiner Kallweit [Tue, 17 Apr 2018 21:19:07 +0000 (23:19 +0200)]
r8169: use skb_copy_to_linear_data in rtl8169_try_rx_copy

Not a giant leap for mankind, but let's avoid the open-coded memcpy
and use standard helper skb_copy_to_linear_data instead.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: remove member align from struct rtl_cfg_info
Heiner Kallweit [Tue, 17 Apr 2018 21:18:05 +0000 (23:18 +0200)]
r8169: remove member align from struct rtl_cfg_info

Since commit 6f0333b8fde4 "r8169: use 50% less ram for RX ring" member
align isn't used any longer, so remove it.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agor8169: remove unused member features from struct
Heiner Kallweit [Tue, 17 Apr 2018 21:17:13 +0000 (23:17 +0200)]
r8169: remove unused member features from struct

Member features of struct rtl8169_private isn't used any longer since
commit 6c6aa15fdea5 "r8169: improve interrupt handling", so remove it.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'netcp-K2G-SoC-support'
David S. Miller [Thu, 19 Apr 2018 01:00:58 +0000 (21:00 -0400)]
Merge branch 'netcp-K2G-SoC-support'

Murali Karicheri says:

====================
Add support for netcp driver on K2G SoC

K2G SoC is another variant of Keystone family of SoCs. This patch
series add support for NetCP driver on this SoC. The QMSS found on
K2G SoC is a cut down version of the QMSS found on other keystone
devices with less number of queues, internal link ram etc. The patch
series has 2 patch sets that goes into the drivers/soc and the
rest has to be applied to net sub system. Please review and merge
if this looks good.

K2G TRM is located at http://www.ti.com/lit/ug/spruhy8g/spruhy8g.pdf
Thanks

The boot logs on K2G ICE board (tftp boot over Ethernet and from mmc)
https://pastebin.ubuntu.com/p/yvZ6drFhkW/

The boot logs on K2G GP board (tftp boot over Ethernet and from mmc)
https://pastebin.ubuntu.com/p/QTr6K7s4Zp/

Also regressed boot on K2HK and K2L EVMs as we have modified GBE
version detection logic (K2E uses same version of NetCP as in K2L.
So regression on one of them is needed).

Boot log on K2L and K2HK EVMs are at
https://pastebin.ubuntu.com/p/N9DBdPjbvR/

This series applies to net-next master branch.

Change history:
  v4 - ready for merge to net-next
       Folded the series "Add promiscous mode support in k2g network driver"
       into this.
       Fixed a typo in 5/11 (sgmii to rgmii) based on TI internal comment
       Reworked 4/11 and title changed to reflect additional changes to
       exclude sgmii configuration code for 2U cpsw. Use IS_SS_ID_2U()
       macro for customization.
       Added Reviewed-by from Rob Herring against 1/13
  v3 - Addressed comments from Andrew Lunn and Grygorii Strashko
       against v2.
  v2 - Addressed following comments on initial version
       - split patch 3/5 to multiple patches from Andrew Lunn
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: netcp: ethss: k2g: add promiscuous mode support
WingMan Kwok [Tue, 17 Apr 2018 21:30:42 +0000 (17:30 -0400)]
net: netcp: ethss: k2g: add promiscuous mode support

This patch adds support for promiscuous mode in k2g's network
driver. When upper layer instructs to transition from
non-promiscuous mode to promiscuous mode or vice versa
K2G network driver needs to configure ALE accordingly
so that in case of non-promiscuous mode, ALE will not flood
all unicast packets to host port, while in promiscuous
mode, it will pass all received unicast packets to
host port.

Signed-off-by: WingMan Kwok <w-kwok2@ti.com>
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: netcp: add api to support set rx mode in netcp modules
WingMan Kwok [Tue, 17 Apr 2018 21:30:41 +0000 (17:30 -0400)]
net: netcp: add api to support set rx mode in netcp modules

This patch adds an API to support setting rx mode in
netcp modules.  If a netcp module needs to be notified
when upper layer transitions from one rx mode to
another and react accordingly, such a module will implement
the new API set_rx_mode added in this patch.  Currently
rx modes supported are PROMISCUOUS and NON_PROMISCUOUS
modes.

Signed-off-by: WingMan Kwok <w-kwok2@ti.com>
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: netcp: support probe deferral
Murali Karicheri [Tue, 17 Apr 2018 21:30:40 +0000 (17:30 -0400)]
net: netcp: support probe deferral

The netcp driver shouldn't proceed until the knav qmss and dma
devices are ready. So return -EPROBE_DEFER if these devices are not
ready.

Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoRevert "net: netcp: remove dead code from the driver"
Murali Karicheri [Tue, 17 Apr 2018 21:30:39 +0000 (17:30 -0400)]
Revert "net: netcp: remove dead code from the driver"

As the probe sequence is not guaranteed contrary to the assumption
of the commit 2d8e276a9030, same has to be reverted.

commit 2d8e276a9030 ("net: netcp: remove dead code from the driver")

Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: netcp: ethss: use of_get_phy_mode() to support different RGMII modes
Murali Karicheri [Tue, 17 Apr 2018 21:30:38 +0000 (17:30 -0400)]
net: netcp: ethss: use of_get_phy_mode() to support different RGMII modes

The phy used for K2G allows for internal delays to be added optionally
to the clock circuitry based on board desing. To add this support,
enhance the driver to use of_get_phy_mode() to read the phy-mode from
the phy device and pass the same to phy through of_phy_connect().

Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: netcp: ethss: re-use stats handling code for 2u hardware
Murali Karicheri [Tue, 17 Apr 2018 21:30:37 +0000 (17:30 -0400)]
net: netcp: ethss: re-use stats handling code for 2u hardware

The stats block in 2u cpsw hardware is similar to the one on nu
and hence handle it in a similar way by using a macro that includes
2u hardware as well.

Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: netcp: ethss: map vlan priorities to zero flow
Murali Karicheri [Tue, 17 Apr 2018 21:30:36 +0000 (17:30 -0400)]
net: netcp: ethss: map vlan priorities to zero flow

The driver currently support only vlan priority zero. So map the
vlan priorities to zero flow in hardware.

Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: netcp: ethss: use rgmii link status for 2u cpsw hardware
Murali Karicheri [Tue, 17 Apr 2018 21:30:35 +0000 (17:30 -0400)]
net: netcp: ethss: use rgmii link status for 2u cpsw hardware

Introduce rgmii link status to handle link state events for 2u
cpsw hardware on K2G.

Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: netcp: ethss: add support for handling rgmii link interface
Murali Karicheri [Tue, 17 Apr 2018 21:30:34 +0000 (17:30 -0400)]
net: netcp: ethss: add support for handling rgmii link interface

2u cpsw hardware on K2G uses rgmii link to interface with Phy. So add
support for this interface in the code so that driver can be re-used
for 2u hardware.

Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: netcp: ethss: make sgmii configuration conditional
Murali Karicheri [Tue, 17 Apr 2018 21:30:33 +0000 (17:30 -0400)]
net: netcp: ethss: make sgmii configuration conditional

As a preparatory patch to add support for 2u cpsw hardware found on
K2G SoC, make sgmii configuration conditional. This is required
since 2u uses RGMII interface instead of SGMII and to allow for driver
re-use.

Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: netcp: ethss: use macro for checking ss_version consistently
Murali Karicheri [Tue, 17 Apr 2018 21:30:32 +0000 (17:30 -0400)]
net: netcp: ethss: use macro for checking ss_version consistently

Driver currently uses macro for NU and XBE hardwrae, while other
places for older hardware such as that on K2H/K SoC (version 1.4
of the cpsw hardware, it explicitly check for the ss_version
inline. Add a new macro for version 1.4 and use it to customize
code in the driver. While at it also fix similar issue with
checking XBE version by re-using existing macro IS_SS_ID_XGBE().

Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosoc: ti: K2G: provide APIs to support driver probe deferral
Murali Karicheri [Tue, 17 Apr 2018 21:30:31 +0000 (17:30 -0400)]
soc: ti: K2G: provide APIs to support driver probe deferral

This patch provide APIs to allow client drivers to support
probe deferral. On K2G SoC, devices can be probed only
after the ti_sci_pm_domains driver is probed and ready.
As drivers may get probed at different order, any driver
that depends on knav dma and qmss drivers, for example
netcp network driver, needs to defer probe until
knav devices are probed and ready to service. To do this,
add an API to query the device ready status from the knav
dma and qmss devices.

Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosoc: ti: K2G: enhancement to support QMSS in K2G NAVSS
Murali Karicheri [Tue, 17 Apr 2018 21:30:30 +0000 (17:30 -0400)]
soc: ti: K2G: enhancement to support QMSS in K2G NAVSS

Navigator Subsystem (NAVSS) available on K2G SoC has a cut down
version of QMSS with less number of queues, internal linking ram
with lesser number of buffers etc.  It doesn't have status and
explicit push register space as in QMSS available on other K2 SoCs.
So define reg indices specific to QMSS on K2G. This patch introduces
"ti,66ak2g-navss-qm" compatibility to identify QMSS on K2G NAVSS
and to customize the dts handling code. Per Device manual,
descriptors with index less than or equal to regions0_size is in region 0
in the case of K2 QMSS where as for QMSS on K2G, descriptors with index
less than regions0_size is in region 0. So update the size accordingly in
the regions0_size bits of the linking ram size 0 register.

Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: WingMan Kwok <w-kwok2@ti.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'ipv6-Separate-data-structures-for-FIB-and-data-path'
David S. Miller [Wed, 18 Apr 2018 03:41:18 +0000 (23:41 -0400)]
Merge branch 'ipv6-Separate-data-structures-for-FIB-and-data-path'

David Ahern says:

====================
net/ipv6: Separate data structures for FIB and data path

IPv6 uses the same data struct for both control plane (FIB entries) and
data path (dst entries). This struct has elements needed for both paths
adding memory overhead and complexity (taking a dst hold in most places
but an additional reference on rt6i_ref in a few). Furthermore, because
of the dst_alloc tie, all FIB entries are allocated with GFP_ATOMIC.

This patch set separates FIB entries from dst entries, better aligning
IPv6 code with IPv4, simplifying the reference counting and allowing
FIB entries added by userspace (not autoconf) to use GFP_KERNEL. It is
first step to a number of performance and scalability changes.

The end result of this patch set:
  - FIB entries (fib6_info):
        /* size: 208, cachelines: 4, members: 25 */
        /* sum members: 207, holes: 1, sum holes: 1 */

  - dst entries (rt6_info)
       /* size: 240, cachelines: 4, members: 11 */

Versus the the single rt6_info struct today for both paths:
      /* size: 320, cachelines: 5, members: 28 */

This amounts to a 35% reduction in memory use for FIB entries and a
25% reduction for dst entries.

With respect to locking FIB entries use RCU and a single atomic
counter with fib6_info_hold and fib6_info_release helpers to manage
the reference counting. dst entries use only the traditional dst
refcounts with dst_hold and dst_release.

FIB entries for host routes are referenced by inet6_ifaddr and
ifacaddr6. In both cases, additional holds are taken -- similar to
what is done for devices.

This set is the first of many changes to improve the scalability of the
IPv6 code. Follow on changes include:
- consolidating duplicate fib6_info references like IPv4 does with
  duplicate fib_info

- moving fib6_info into a slab cache to avoid allocation roundups to
  power of 2 (the 208 size becomes a 256 actual allocation)

- Allow FIB lookups without generating a dst (e.g., most rt6_lookup
  users just want to verify the egress device). Means moving dst
  allocation to the other side of fib6_rule_lookup which again aligns
  with IPv4 behavior

- using separate standalone nexthop objects which have performance
  benefits beyond fib_info consolidation

At this point I am not seeing any refcount leaks or underflows, no
oops or bug_ons, or warnings from kasan, so I think it is ready for
others to beat up on it finding errors in code paths I have missed.

v2 changes
- rebased to top of tree
- improved commit message on patch 7

v1 changes
- rebased to top of tree
- fix memory leak of metrics as noted by Ido
- MTU fixes based on pmtu tests (thanks Stefano Brivio for writing)

RFC v2 changes
- improved commit messages
- move common metrics code from dst.c to net/ipv4/metrics.c (comment
  from DaveM)
- address comments from Wei Wang and Martin KaFai Lau (let me know if
  I missed something)
- fixes detected by kernel test robots
  + added fib6_metric_set to change metric on a FIB entry which could
    be pointing to read-only dst_default_metrics
  + 0day testing found a problem with an intermediate patch; added
    dst_hold_safe on rt->from. Code is removed 3 patches later
- allow cacheinfo to handle NULL dst; means only expires is pushed to
  userspace
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Remove unused code and variables for rt6_info
David Ahern [Wed, 18 Apr 2018 00:33:27 +0000 (17:33 -0700)]
net/ipv6: Remove unused code and variables for rt6_info

Drop unneeded elements from rt6_info struct and rearrange layout to
something more relevant for the data path.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: Flip FIB entries to fib6_info
David Ahern [Wed, 18 Apr 2018 00:33:26 +0000 (17:33 -0700)]
net/ipv6: Flip FIB entries to fib6_info

Convert all code paths referencing a FIB entry from
rt6_info to fib6_info.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: separate handling of FIB entries from dst based routes
David Ahern [Wed, 18 Apr 2018 00:33:25 +0000 (17:33 -0700)]
net/ipv6: separate handling of FIB entries from dst based routes

Last step before flipping the data type for FIB entries:
- use fib6_info_alloc to create FIB entries in ip6_route_info_create
  and addrconf_dst_alloc
- use fib6_info_release in place of dst_release, ip6_rt_put and
  rt6_release
- remove the dst_hold before calling __ip6_ins_rt or ip6_del_rt
- when purging routes, drop per-cpu routes
- replace inc and dec of rt6i_ref with fib6_info_hold and fib6_info_release
- use rt->from since it points to the FIB entry
- drop references to exception bucket, fib6_metrics and per-cpu from
  dst entries (those are relevant for fib entries only)

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/ipv6: introduce fib6_info struct and helpers
David Ahern [Wed, 18 Apr 2018 00:33:24 +0000 (17:33 -0700)]
net/ipv6: introduce fib6_info struct and helpers

Add fib6_info struct and alloc, destroy, hold and release helpers.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>