OSDN Git Service

uclinux-h8/linux.git
6 years agonet: dsa: Remove prepare phase for FDB
Arkadi Sharshevsky [Sun, 6 Aug 2017 13:15:40 +0000 (16:15 +0300)]
net: dsa: Remove prepare phase for FDB

The prepare phase for FDB add is unneeded because most of DSA devices
can have failures during bus transactions (SPI, I2C, etc.), thus, the
prepare phase cannot guarantee success of the commit stage.

The support for learning FDB through notification chain, which will be
introduced in the following patches, will provide the ability to notify
back the bridge about successful offload.

Signed-off-by: Arkadi Sharshevsky <arkadis@mellanox.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: Change DSA slave FDB API to be switchdev independent
Arkadi Sharshevsky [Sun, 6 Aug 2017 13:15:39 +0000 (16:15 +0300)]
net: dsa: Change DSA slave FDB API to be switchdev independent

In order to support FDB add/del to be on a notifier chain the slave
API need to be changed to be switchdev independent.

Signed-off-by: Arkadi Sharshevsky <arkadis@mellanox.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agohamradio: baycom: make hdlcdrv_ops const
Bhumika Goyal [Sun, 6 Aug 2017 08:51:45 +0000 (14:21 +0530)]
hamradio: baycom: make hdlcdrv_ops const

Make hdlcdrv_ops structures const as they are only passed to
hdlcdrv_register function. The corresponding argument is of type const,
so make the structures const.

Signed-off-by: Bhumika Goyal <bhumirks@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoxfrm: check that cached bundle is still valid
Florian Westphal [Sun, 6 Aug 2017 08:19:07 +0000 (10:19 +0200)]
xfrm: check that cached bundle is still valid

Quoting Ilan Tayari:
  1. Set up a host-to-host IPSec tunnel (or transport, doesn't matter)
  2. Ping over IPSec, or do something to populate the pcpu cache
  3. Join a MC group, then leave MC group
  4. Try to ping again using same CPU as before -> traffic
     doesn't egress the machine at all

Ilan debugged the problem down to the fact that one of the path dsts
devices point to lo due to earlier dst_dev_put().
In this case, dst is marked as DEAD and we cannot reuse the bundle.

The cache only asserted that the requested policy and that of the cached
bundle match, but its not enough - also verify the path is still valid.

Fixes: ec30d78c14a813 ("xfrm: add xdst pcpu cache")
Reported-by: Ayham Masood <ayhamm@mellanox.com>
Tested-by: Ilan Tayari <ilant@mellanox.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'net-dsa-remove-useless-arguments'
David S. Miller [Mon, 7 Aug 2017 21:24:21 +0000 (14:24 -0700)]
Merge branch 'net-dsa-remove-useless-arguments'

Vivien Didelot says:

====================
net: dsa: remove useless arguments

Several DSA core setup functions take many arguments, mostly because of
the legacy code. This patch series removes the useless args of these
functions, where either the dsa_switch or dsa_port argument is enough.

Changes in v2:
  - ds->dev is already assigned by dsa_switch_alloc
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: remove useless args of dsa_slave_create
Vivien Didelot [Sat, 5 Aug 2017 20:20:19 +0000 (16:20 -0400)]
net: dsa: remove useless args of dsa_slave_create

dsa_slave_create currently takes 4 arguments while it only needs the
related dsa_port and its name. Remove all other arguments.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: remove useless args of dsa_cpu_dsa_setup
Vivien Didelot [Sat, 5 Aug 2017 20:20:18 +0000 (16:20 -0400)]
net: dsa: remove useless args of dsa_cpu_dsa_setup

dsa_cpu_dsa_setup currently takes 4 arguments but they are all available
from the dsa_port argument. Remove all others.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: remove useless argument in legacy setup
Vivien Didelot [Sat, 5 Aug 2017 20:20:17 +0000 (16:20 -0400)]
net: dsa: remove useless argument in legacy setup

dsa_switch_alloc() already assigns ds-dev, which can be used in
dsa_switch_setup_one and dsa_cpu_dsa_setups instead of requiring an
additional struct device argument.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: hns3: fix spelling mistake: "capabilty" -> "capability"
Colin Ian King [Sat, 5 Aug 2017 13:46:35 +0000 (14:46 +0100)]
net: hns3: fix spelling mistake: "capabilty" -> "capability"

Trivial fix to spelling mistake in dev_err error message and also
split overly long line to avoid a checkpatch warning.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'Refactor-lan9303_xxx_packet_processing'
David S. Miller [Mon, 7 Aug 2017 21:18:01 +0000 (14:18 -0700)]
Merge branch 'Refactor-lan9303_xxx_packet_processing'

Egil Hjelmeland says:

====================
Refactor lan9303_xxx_packet_processing

This series is purely non functional.

It changes the lan9303_enable_packet_processing,
lan9303_disable_packet_processing() to pass port number (0,1,2) as
parameter instead of port offset. This aligns them with
other functions in the module, and makes it possible to simplify the code.

The lan9303_enable_packet_processing, lan9303_disable_packet_processing
functions operate on port. Therefore rename the functions to reflect that
as well.

Reviewer pointed out lan9303_get_ethtool_stats would be better off with
the use of a lan9303_read_switch_port(). So that was added to the series.

Changes v3 -> v4:
 - Whitespace adjustments.

Changes v2 -> v3:
 - Patch 1: Removed the change in lan9303_get_ethtool_stats
 - Added patch 4: rename lan9303_xxx_packet_processing
 - Added patch 5: refactor lan9303_get_ethtool_stats

Changes v1 -> v2:
 - introduced lan9303_write_switch_port() in first patch
 - inserted LAN9303_NUM_PORTS patch
 - Use LAN9303_NUM_PORTS in last patch. Plus whitespace change.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: lan9303: refactor lan9303_get_ethtool_stats
Egil Hjelmeland [Sat, 5 Aug 2017 11:05:50 +0000 (13:05 +0200)]
net: dsa: lan9303: refactor lan9303_get_ethtool_stats

In lan9303_get_ethtool_stats: Get rid of 0x400 constant magic
by using new lan9303_read_switch_reg() inside loop.
Reduced scope of two variables.

Signed-off-by: Egil Hjelmeland <privat@egil-hjelmeland.no>
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 agonet: dsa: lan9303: Rename lan9303_xxx_packet_processing()
Egil Hjelmeland [Sat, 5 Aug 2017 11:05:49 +0000 (13:05 +0200)]
net: dsa: lan9303: Rename lan9303_xxx_packet_processing()

The lan9303_enable_packet_processing, lan9303_disable_packet_processing
functions operate on port, so the names should reflect that.
And to align with lan9303_disable_processing(), rename:

lan9303_enable_packet_processing -> lan9303_enable_processing_port
lan9303_disable_packet_processing -> lan9303_disable_processing_port

Signed-off-by: Egil Hjelmeland <privat@egil-hjelmeland.no>
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 agonet: dsa: lan9303: Simplify lan9303_xxx_packet_processing() usage
Egil Hjelmeland [Sat, 5 Aug 2017 11:05:48 +0000 (13:05 +0200)]
net: dsa: lan9303: Simplify lan9303_xxx_packet_processing() usage

Simplify usage of lan9303_enable_packet_processing,
lan9303_disable_packet_processing()

Signed-off-by: Egil Hjelmeland <privat@egil-hjelmeland.no>
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 agonet: dsa: lan9303: define LAN9303_NUM_PORTS 3
Egil Hjelmeland [Sat, 5 Aug 2017 11:05:47 +0000 (13:05 +0200)]
net: dsa: lan9303: define LAN9303_NUM_PORTS 3

Will be used instead of '3' in upcomming patches.

Signed-off-by: Egil Hjelmeland <privat@egil-hjelmeland.no>
Signed-off-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 agonet: dsa: lan9303: Change lan9303_xxx_packet_processing() port param.
Egil Hjelmeland [Sat, 5 Aug 2017 11:05:46 +0000 (13:05 +0200)]
net: dsa: lan9303: Change lan9303_xxx_packet_processing() port param.

lan9303_enable_packet_processing, lan9303_disable_packet_processing()
Pass port number (0,1,2) as parameter instead of port offset.
Because other functions in the module pass port numbers.
And to enable simplifications in following patch.

Introduce lan9303_write_switch_port().

Signed-off-by: Egil Hjelmeland <privat@egil-hjelmeland.no>
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 'ipv6-sr-add-support-for-advanced-local-segment-processing'
David S. Miller [Mon, 7 Aug 2017 21:16:22 +0000 (14:16 -0700)]
Merge branch 'ipv6-sr-add-support-for-advanced-local-segment-processing'

David Lebrun says:

====================
ipv6: sr: add support for advanced local segment processing

v2: use EXPORT_SYMBOL_GPL

The current implementation of IPv6 SR supports SRH insertion/encapsulation
and basic segment endpoint behavior (i.e., processing of an SRH contained in
a packet whose active segment (IPv6 DA) is routed to the local node). This
behavior simply consists of updating the DA to the next segment and forwarding
the packet accordingly. This processing is realised for all such packets,
regardless of the active segment.

The most recent specifications of IPv6 SR [1] [2] extend the SRH processing
features as follows. Each segment endpoint defines a MyLocalSID table.
This table maps segments to operations to perform. For each ingress IPv6
packet whose DA is part of a given prefix, the segment endpoint looks
up the active segment (i.e., the IPv6 DA) in the MyLocalSID table and
applies the corresponding operation. Such specifications enable to specify
arbitrary operations besides the basic SRH processing and allow for a more
fine-grained classification.

This patch series implements those extended specifications by leveraging
a new type of lightweight tunnel, seg6local. The MyLocalSID table is
simply an arbitrary routing table (using CONFIG_IPV6_MULTIPLE_TABLES). The
following commands would assign the prefix fc00::/64 to the MyLocalSID
table, map the segment fc00::42 to the regular SRH processing function
(named "End"), and drop all packets received with an undefined active
segment:

ip -6 rule add fc00::/64 lookup 100
ip -6 route add fc00::42 encap seg6local action End dev eth0 table 100
ip -6 route add blackhole default table 100

As another example, the following command would assign the segment
fc00::1234 to the regular SRH processing function, except that the
processed packet must be forwarded to the next-hop fc42::1 (this operation
is named "End.X"):

ip -6 route add fc00::1234 encap seg6local action End.X nh6 fc42::1 dev eth0 table 100

Those two basic operations (End and End.X) are defined in [1]. A more
extensive list of advanced operations is defined in [2].

The first two patches of the series are preliminary work that remove an
assumption about initial SRH format, and export the two functions used to
insert and encapsulate an SRH onto packets. The third patch defines the
new seg6local lightweight tunnel and implement the core functions. The
fourth patch implements the operations needed to handle the newly defined
rtnetlink attributes. The fifth patch implements a few SRH processing
operations, including End and End.X.

[1] https://tools.ietf.org/html/draft-ietf-6man-segment-routing-header-07
[2] https://tools.ietf.org/html/draft-filsfils-spring-srv6-network-programming-01
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv6: sr: implement several seg6local actions
David Lebrun [Sat, 5 Aug 2017 10:39:48 +0000 (12:39 +0200)]
ipv6: sr: implement several seg6local actions

This patch implements the following seg6local actions.

- SEG6_LOCAL_ACTION_END: regular SRH processing. The DA of the packet
  is updated to the next segment and forwarded accordingly.

- SEG6_LOCAL_ACTION_END_X: same as above, except that the packet is
  forwarded to the specified IPv6 next-hop.

- SEG6_LOCAL_ACTION_END_DX6: decapsulate the packet and forward to
  inner IPv6 packet to the specified IPv6 next-hop.

- SEG6_LOCAL_ACTION_END_B6: insert the specified SRH directly after
  the IPv6 header of the packet.

- SEG6_LOCAL_ACTION_END_B6_ENCAP: encapsulate the packet within
  an outer IPv6 header, containing the specified SRH.

Signed-off-by: David Lebrun <david.lebrun@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv6: sr: add rtnetlink functions for seg6local action parameters
David Lebrun [Sat, 5 Aug 2017 10:38:27 +0000 (12:38 +0200)]
ipv6: sr: add rtnetlink functions for seg6local action parameters

This patch adds the necessary functions to parse, fill, and compare
seg6local rtnetlink attributes, for all defined action parameters.

- The SRH parameter defines an SRH to be inserted or encapsulated.
- The TABLE parameter defines the table to use for the route lookup of
  the next segment or the inner decapsulated packet.
- The NH4 parameter defines the IPv4 next-hop for an inner decapsulated
  IPv4 packet.
- The NH6 parameter defines the IPv6 next-hop for the next segment or
  for an inner decapsulated IPv6 packet
- The IIF parameter defines an ingress interface index.
- The OIF parameter defines an egress interface index.

Signed-off-by: David Lebrun <david.lebrun@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv6: sr: define core operations for seg6local lightweight tunnel
David Lebrun [Sat, 5 Aug 2017 10:38:26 +0000 (12:38 +0200)]
ipv6: sr: define core operations for seg6local lightweight tunnel

This patch implements a new type of lightweight tunnel named seg6local.
A seg6local lwt is defined by a type of action and a set of parameters.
The action represents the operation to perform on the packets matching the
lwt's route, and is not necessarily an encapsulation. The set of parameters
are arguments for the processing function.

Each action is defined in a struct seg6_action_desc within
seg6_action_table[]. This structure contains the action, mandatory
attributes, the processing function, and a static headroom size required by
the action. The mandatory attributes are encoded as a bitmask field. The
static headroom is set to a non-zero value when the processing function
always add a constant number of bytes to the skb (e.g. the header size for
encapsulations).

To facilitate rtnetlink-related operations such as parsing, fill_encap,
and cmp_encap, each type of action parameter is associated to three
function pointers, in seg6_action_params[].

All actions defined in seg6_local.h are detailed in [1].

[1] https://tools.ietf.org/html/draft-filsfils-spring-srv6-network-programming-01

Signed-off-by: David Lebrun <david.lebrun@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv6: sr: export SRH insertion functions
David Lebrun [Sat, 5 Aug 2017 10:38:25 +0000 (12:38 +0200)]
ipv6: sr: export SRH insertion functions

This patch exports the seg6_do_srh_encap() and seg6_do_srh_inline()
functions. It also removes the CONFIG_IPV6_SEG6_INLINE knob
that enabled the compilation of seg6_do_srh_inline(). This function
is now built-in.

Signed-off-by: David Lebrun <david.lebrun@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoipv6: sr: allow SRH insertion with arbitrary segments_left value
David Lebrun [Sat, 5 Aug 2017 10:38:24 +0000 (12:38 +0200)]
ipv6: sr: allow SRH insertion with arbitrary segments_left value

The seg6_validate_srh() function only allows SRHs whose active segment is
the first segment of the path. However, an application may insert an SRH
whose active segment is not the first one. Such an application might be
for example an SR-aware Virtual Network Function.

This patch enables to insert SRHs with an arbitrary active segment.

Signed-off-by: David Lebrun <david.lebrun@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobpf: devmap fix mutex in rcu critical section
John Fastabend [Sat, 5 Aug 2017 05:02:19 +0000 (22:02 -0700)]
bpf: devmap fix mutex in rcu critical section

Originally we used a mutex to protect concurrent devmap update
and delete operations from racing with netdev unregister notifier
callbacks.

The notifier hook is needed because we increment the netdev ref
count when a dev is added to the devmap. This ensures the netdev
reference is valid in the datapath. However, we don't want to block
unregister events, hence the initial mutex and notifier handler.

The concern was in the notifier hook we search the map for dev
entries that hold a refcnt on the net device being torn down. But,
in order to do this we require two steps,

  (i) dereference the netdev:  dev = rcu_dereference(map[i])
 (ii) test ifindex:   dev->ifindex == removing_ifindex

and then finally we can swap in the NULL dev in the map via an
xchg operation,

  xchg(map[i], NULL)

The danger here is a concurrent update could run a different
xchg op concurrently leading us to replace the new dev with a
NULL dev incorrectly.

      CPU 1                        CPU 2

   notifier hook                   bpf devmap update

   dev = rcu_dereference(map[i])
                                   dev = rcu_dereference(map[i])
                                   xchg(map[i]), new_dev);
                                   rcu_call(dev,...)
   xchg(map[i], NULL)

The above flow would create the incorrect state with the dev
reference in the update path being lost. To resolve this the
original code used a mutex around the above block. However,
updates, deletes, and lookups occur inside rcu critical sections
so we can't use a mutex in this context safely.

Fortunately, by writing slightly better code we can avoid the
mutex altogether. If CPU 1 in the above example uses a cmpxchg
and _only_ replaces the dev reference in the map when it is in
fact the expected dev the race is removed completely. The two
cases being illustrated here, first the race condition,

      CPU 1                          CPU 2

   notifier hook                     bpf devmap update

   dev = rcu_dereference(map[i])
                                     dev = rcu_dereference(map[i])
                                     xchg(map[i]), new_dev);
                                     rcu_call(dev,...)
   odev = cmpxchg(map[i], dev, NULL)

Now we can test the cmpxchg return value, detect odev != dev and
abort. Or in the good case,

      CPU 1                          CPU 2

   notifier hook                     bpf devmap update
   dev = rcu_dereference(map[i])
   odev = cmpxchg(map[i], dev, NULL)
                                     [...]

Now 'odev == dev' and we can do proper cleanup.

And viola the original race we tried to solve with a mutex is
corrected and the trace noted by Sasha below is resolved due
to removal of the mutex.

Note: When walking the devmap and removing dev references as needed
we depend on the core to fail any calls to dev_get_by_index() using
the ifindex of the device being removed. This way we do not race with
the user while searching the devmap.

Additionally, the mutex was also protecting list add/del/read on
the list of maps in-use. This patch converts this to an RCU list
and spinlock implementation. This protects the list from concurrent
alloc/free operations. The notifier hook walks this list so it uses
RCU read semantics.

BUG: sleeping function called from invalid context at kernel/locking/mutex.c:747
in_atomic(): 1, irqs_disabled(): 0, pid: 16315, name: syz-executor1
1 lock held by syz-executor1/16315:
 #0:  (rcu_read_lock){......}, at: [<ffffffff8c363bc2>] map_delete_elem kernel/bpf/syscall.c:577 [inline]
 #0:  (rcu_read_lock){......}, at: [<ffffffff8c363bc2>] SYSC_bpf kernel/bpf/syscall.c:1427 [inline]
 #0:  (rcu_read_lock){......}, at: [<ffffffff8c363bc2>] SyS_bpf+0x1d32/0x4ba0 kernel/bpf/syscall.c:1388

Fixes: 2ddf71e23cc2 ("net: add notifier hooks for devmap bpf map")
Reported-by: Sasha Levin <alexander.levin@verizon.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'net_sched-clean-up-filter-handle'
David S. Miller [Mon, 7 Aug 2017 21:12:18 +0000 (14:12 -0700)]
Merge branch 'net_sched-clean-up-filter-handle'

Cong Wang says:

====================
net_sched: clean up filter handle

This patchset sits in my local branch for a long time, it is time to
send it out. It cleans up the ambiguous use of 'unsigned long fh',
please see each of them for details.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet_sched: use void pointer for filter handle
WANG Cong [Sat, 5 Aug 2017 04:31:43 +0000 (21:31 -0700)]
net_sched: use void pointer for filter handle

Now we use 'unsigned long fh' as a pointer in every place,
it is safe to convert it to a void pointer now. This gets
rid of many casts to pointer.

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet_sched: refactor notification code for RTM_DELTFILTER
WANG Cong [Sat, 5 Aug 2017 04:31:42 +0000 (21:31 -0700)]
net_sched: refactor notification code for RTM_DELTFILTER

It is confusing to use 'unsigned long fh' as both a handle
and a pointer, especially commit 9ee7837449b3
("net sched filters: fix notification of filter delete with proper handle").

This patch introduces tfilter_del_notify() so that we can
pass it as a pointer as before, and we don't need to check
RTM_DELTFILTER in tcf_fill_node() any more.

This prepares for the next patch.

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agolwtunnel: replace EXPORT_SYMBOL with EXPORT_SYMBOL_GPL
Roopa Prabhu [Sat, 5 Aug 2017 01:19:18 +0000 (18:19 -0700)]
lwtunnel: replace EXPORT_SYMBOL with EXPORT_SYMBOL_GPL

Signed-off-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'bpf-add-support-for-sys-enter-exit-tracepoints'
David S. Miller [Mon, 7 Aug 2017 21:09:48 +0000 (14:09 -0700)]
Merge branch 'bpf-add-support-for-sys-enter-exit-tracepoints'

Yonghong Song says:

====================
bpf: add support for sys_{enter|exit}_* tracepoints

Currently, bpf programs cannot be attached to sys_enter_* and sys_exit_*
style tracepoints. The main reason is that syscalls/sys_enter_* and syscalls/sys_exit_*
tracepoints are treated differently from other tracepoints and there
is no bpf hook to it.

This patch set adds bpf support for these syscalls tracepoints and also
adds a test case for it.

Changelogs:
v3 -> v4:
 - Check the legality of ctx offset access for syscall tracepoint as well.
   trace_event_get_offsets will return correct max offset for each
   specific syscall tracepoint.
 - Use variable length array to avoid hardcode 6 as the maximum
   arguments beyond syscall_nr.
v2 -> v3:
 - Fix a build issue
v1 -> v2:
 - Do not use TRACE_EVENT_FL_CAP_ANY to identify syscall tracepoint.
   Instead use trace_event_call->class.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobpf: add a test case for syscalls/sys_{enter|exit}_* tracepoints
Yonghong Song [Fri, 4 Aug 2017 23:00:10 +0000 (16:00 -0700)]
bpf: add a test case for syscalls/sys_{enter|exit}_* tracepoints

Signed-off-by: Yonghong Song <yhs@fb.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agobpf: add support for sys_enter_* and sys_exit_* tracepoints
Yonghong Song [Fri, 4 Aug 2017 23:00:09 +0000 (16:00 -0700)]
bpf: add support for sys_enter_* and sys_exit_* tracepoints

Currently, bpf programs cannot be attached to sys_enter_* and sys_exit_*
style tracepoints. The iovisor/bcc issue #748
(https://github.com/iovisor/bcc/issues/748) documents this issue.
For example, if you try to attach a bpf program to tracepoints
syscalls/sys_enter_newfstat, you will get the following error:
   # ./tools/trace.py t:syscalls:sys_enter_newfstat
   Ioctl(PERF_EVENT_IOC_SET_BPF): Invalid argument
   Failed to attach BPF to tracepoint

The main reason is that syscalls/sys_enter_* and syscalls/sys_exit_*
tracepoints are treated differently from other tracepoints and there
is no bpf hook to it.

This patch adds bpf support for these syscalls tracepoints by
  . permitting bpf attachment in ioctl PERF_EVENT_IOC_SET_BPF
  . calling bpf programs in perf_syscall_enter and perf_syscall_exit

The legality of bpf program ctx access is also checked.
Function trace_event_get_offsets returns correct max offset for each
specific syscall tracepoint, which is compared against the maximum offset
access in bpf program.

Signed-off-by: Yonghong Song <yhs@fb.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoof_mdio: use of_property_read_u32_array()
Sergei Shtylyov [Fri, 4 Aug 2017 21:43:43 +0000 (00:43 +0300)]
of_mdio: use of_property_read_u32_array()

The "fixed-link" prop support predated of_property_read_u32_array(), so
basically had to open-code it. Using the modern API saves 24 bytes of the
object code (ARM gcc 4.8.5); the only behavior change would be that the
prop length check is now less strict (however the strict pre-check done
in of_phy_is_fixed_link() is left intact anyway)...

Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoibmvnic: Report rx buffer return codes as netdev_dbg
John Allen [Mon, 7 Aug 2017 20:42:30 +0000 (15:42 -0500)]
ibmvnic: Report rx buffer return codes as netdev_dbg

Reporting any return code for a receive buffer as an "rx error" only
produces alarming noise and the only values that have been observed to be
used in this field are not error conditions. Change this to a netdev_dbg
with a more descriptive message.

Signed-off-by: John Allen <jallen@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'net-l3mdev-Support-for-sockets-bound-to-enslaved-device'
David S. Miller [Mon, 7 Aug 2017 18:39:22 +0000 (11:39 -0700)]
Merge branch 'net-l3mdev-Support-for-sockets-bound-to-enslaved-device'

David Ahern says:

====================
net: l3mdev: Support for sockets bound to enslaved device

A missing piece to the VRF puzzle is the ability to bind sockets to
devices enslaved to a VRF. This patch set adds the enslaved device
index, sdif, to IPv4 and IPv6 socket lookups. The end result for users
is the following scope options for services:

1. "global" services - sockets not bound to any device

   Allows 1 service to work across all network interfaces with
   connected sockets bound to the VRF the connection originates
   (Requires net.ipv4.tcp_l3mdev_accept=1 for TCP and
    net.ipv4.udp_l3mdev_accept=1 for UDP)

2. "VRF" local services - sockets bound to a VRF

   Sockets work across all network interfaces enslaved to a VRF but
   are limited to just the one VRF.

3. "device" services - sockets bound to a specific network interface

   Service works only through the one specific interface.

v3
- convert __inet_lookup_established in dccp_v4_err; missed in v2

v2
- remove sk_lookup struct and add sdif as an argument to existing
  functions

Changes since RFC:
- no significant logic changes; mainly whitespace cleanups
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ipv6: add second dif to raw socket lookups
David Ahern [Mon, 7 Aug 2017 15:44:22 +0000 (08:44 -0700)]
net: ipv6: add second dif to raw socket lookups

Add a second device index, sdif, to raw socket lookups. sdif is the
index for ingress devices enslaved to an l3mdev. It allows the lookups
to consider the enslaved device as well as the L3 domain when searching
for a socket.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ipv6: add second dif to inet6 socket lookups
David Ahern [Mon, 7 Aug 2017 15:44:21 +0000 (08:44 -0700)]
net: ipv6: add second dif to inet6 socket lookups

Add a second device index, sdif, to inet6 socket lookups. sdif is the
index for ingress devices enslaved to an l3mdev. It allows the lookups
to consider the enslaved device as well as the L3 domain when searching
for a socket.

TCP moves the data in the cb. Prior to tcp_v4_rcv (e.g., early demux) the
ingress index is obtained from IPCB using inet_sdif and after tcp_v4_rcv
tcp_v4_sdif is used.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ipv6: add second dif to udp socket lookups
David Ahern [Mon, 7 Aug 2017 15:44:20 +0000 (08:44 -0700)]
net: ipv6: add second dif to udp socket lookups

Add a second device index, sdif, to udp socket lookups. sdif is the
index for ingress devices enslaved to an l3mdev. It allows the lookups
to consider the enslaved device as well as the L3 domain when searching
for a socket.

Early demux lookups are handled in the next patch as part of INET_MATCH
changes.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ipv4: add second dif to multicast source filter
David Ahern [Mon, 7 Aug 2017 15:44:19 +0000 (08:44 -0700)]
net: ipv4: add second dif to multicast source filter

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ipv4: add second dif to raw socket lookups
David Ahern [Mon, 7 Aug 2017 15:44:18 +0000 (08:44 -0700)]
net: ipv4: add second dif to raw socket lookups

Add a second device index, sdif, to raw socket lookups. sdif is the
index for ingress devices enslaved to an l3mdev. It allows the lookups
to consider the enslaved device as well as the L3 domain when searching
for a socket.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ipv4: add second dif to inet socket lookups
David Ahern [Mon, 7 Aug 2017 15:44:17 +0000 (08:44 -0700)]
net: ipv4: add second dif to inet socket lookups

Add a second device index, sdif, to inet socket lookups. sdif is the
index for ingress devices enslaved to an l3mdev. It allows the lookups
to consider the enslaved device as well as the L3 domain when searching
for a socket.

TCP moves the data in the cb. Prior to tcp_v4_rcv (e.g., early demux) the
ingress index is obtained from IPCB using inet_sdif and after the cb move
in  tcp_v4_rcv the tcp_v4_sdif helper is used.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: ipv4: add second dif to udp socket lookups
David Ahern [Mon, 7 Aug 2017 15:44:16 +0000 (08:44 -0700)]
net: ipv4: add second dif to udp socket lookups

Add a second device index, sdif, to udp socket lookups. sdif is the
index for ingress devices enslaved to an l3mdev. It allows the lookups
to consider the enslaved device as well as the L3 domain when searching
for a socket.

Early demux lookups are handled in the next patch as part of INET_MATCH
changes.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'wireless-drivers-next-for-davem-2017-08-07' of git://git.kernel.org/pub...
David S. Miller [Mon, 7 Aug 2017 18:34:41 +0000 (11:34 -0700)]
Merge tag 'wireless-drivers-next-for-davem-2017-08-07' of git://git./linux/kernel/git/kvalo/wireless-drivers-next

Kalle Valo says:

====================
wireless-drivers-next patches for 4.14

The first wireless-drivers-next pull request for 4.14. I'm submitting
this unusally late in the cycle as my vacation postponed this. But
even if this is late there's not still that much new features, mostly
cleanup or fixes.

Major changes:

ath10k

* preparation for wcn3990 support

iwlwifi

* Reorganization of the code into separate directories continues

qtnfmac

* regulatory support updates

* add get_channel, dump_survey and channel_switch cfg80211 handlers
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agohns3: fix unused function warning
Arnd Bergmann [Mon, 7 Aug 2017 10:41:53 +0000 (12:41 +0200)]
hns3: fix unused function warning

Without CONFIG_PCI_IOV, we get a harmless warning about an
unused function:

drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_main.c:2273:13: error: 'hclge_disable_sriov' defined but not used [-Werror=unused-function]

The #ifdefs in this driver are obviously wrong, so this just
removes them and uses an IS_ENABLED() check that does the same
thing correctly in a more readable way.

Fixes: 46a3df9f9718 ("net: hns3: Add HNS3 Acceleration Engine & Compatibility Layer Support")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge tag 'mlx5-shared-2017-08-07' of git://git.kernel.org/pub/scm/linux/kernel/git...
David S. Miller [Mon, 7 Aug 2017 17:42:09 +0000 (10:42 -0700)]
Merge tag 'mlx5-shared-2017-08-07' of git://git./linux/kernel/git/mellanox/linux

Saeed Mahameed says:

====================
mlx5-shared-2017-08-07

This series includes some mlx5 updates for both net-next and rdma trees.

From Saeed,
Core driver updates to allow selectively building the driver with
or without some large driver components, such as
- E-Switch (Ethernet SRIOV support).
- Multi-Physical Function Switch (MPFs) support.
For that we split E-Switch and MPFs functionalities into separate files.

From Erez,
Delay mlx5_core events when mlx5 interfaces, namely mlx5_ib, registration
is taking place and until it completes.

From Rabie,
Increase the maximum supported flow counters.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'net-sched-summer-cleanup-part-2-ndo_setup_tc'
David S. Miller [Mon, 7 Aug 2017 16:42:37 +0000 (09:42 -0700)]
Merge branch 'net-sched-summer-cleanup-part-2-ndo_setup_tc'

Jiri Pirko says:

====================
net: sched: summer cleanup part 2, ndo_setup_tc

This patchset focuses on ndo_setup_tc and its args.
Currently there are couple of things that do not make much sense.
The type is passed in struct tc_to_netdev, but as it is always
required, should be arg of the ndo. Other things are passed as args
but they are only relevant for cls offloads and not mqprio. Therefore,
they should be pushed to struct. As the tc_to_netdev struct in the end
is just a container of single pointer, we get rid of it and pass the
struct according to type. So in the end, we have:
ndo_setup_tc(dev, type, type_data_struct)

There are couple of cosmetics done on the way to make things smooth.
Also, reported error is consolidated to eopnotsupp in case the
asked offload is not supported.

v1->v2:
- added forgotten hns3pf bits
====================

Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: sched: get rid of struct tc_to_netdev
Jiri Pirko [Mon, 7 Aug 2017 08:15:32 +0000 (10:15 +0200)]
net: sched: get rid of struct tc_to_netdev

Get rid of struct tc_to_netdev which is now just unnecessary container
and rather pass per-type structures down to drivers directly.
Along with that, consolidate the naming of per-type structure variables
in cls_*.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: sched: change return value of ndo_setup_tc for driver supporting mqprio only
Jiri Pirko [Mon, 7 Aug 2017 08:15:31 +0000 (10:15 +0200)]
net: sched: change return value of ndo_setup_tc for driver supporting mqprio only

Change the return value from -EINVAL to -EOPNOTSUPP. The rest of the
drivers have it like that, so be aligned.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: sched: move prio into cls_common
Jiri Pirko [Mon, 7 Aug 2017 08:15:30 +0000 (10:15 +0200)]
net: sched: move prio into cls_common

prio is not cls_flower specific, but it is meaningful for all
classifiers. Seems that only mlxsw cares about the value. Obviously,
cls offload in other drivers is broken.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: sched: push cls related args into cls_common structure
Jiri Pirko [Mon, 7 Aug 2017 08:15:29 +0000 (10:15 +0200)]
net: sched: push cls related args into cls_common structure

As ndo_setup_tc is generic offload op for whole tc subsystem, does not
really make sense to have cls-specific args. So move them under
cls_common structurure which is embedded in all cls structs.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agohns3pf: don't check handle during mqprio offload
Jiri Pirko [Mon, 7 Aug 2017 08:15:28 +0000 (10:15 +0200)]
hns3pf: don't check handle during mqprio offload

Similar to the rest offloaders of mqprio, no need to check handle.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonfp: change flows in apps that offload ndo_setup_tc
Jiri Pirko [Mon, 7 Aug 2017 08:15:27 +0000 (10:15 +0200)]
nfp: change flows in apps that offload ndo_setup_tc

Change the flows a bit in preparation of follow-up changes in
ndo_setup_tc args. Also, change the error code to align with the rest of
the drivers.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodsa: push cls_matchall setup_tc processing into a separate function
Jiri Pirko [Mon, 7 Aug 2017 08:15:26 +0000 (10:15 +0200)]
dsa: push cls_matchall setup_tc processing into a separate function

Let dsa_slave_setup_tc be a splitter for specific setup_tc types and
push out cls_matchall specific code into a separate function.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: spectrum: rename cls arg in matchall processing
Jiri Pirko [Mon, 7 Aug 2017 08:15:25 +0000 (10:15 +0200)]
mlxsw: spectrum: rename cls arg in matchall processing

To sync-up with the naming in the rest of the driver, rename the cls arg.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlxsw: spectrum: push cls_flower and cls_matchall setup_tc processing into separate...
Jiri Pirko [Mon, 7 Aug 2017 08:15:24 +0000 (10:15 +0200)]
mlxsw: spectrum: push cls_flower and cls_matchall setup_tc processing into separate functions

Let mlxsw_sp_setup_tc be a splitter for specific setup_tc types and push
out cls_flower and cls_matchall specific codes into separate functions.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlx5e_rep: push cls_flower setup_tc processing into a separate function
Jiri Pirko [Mon, 7 Aug 2017 08:15:23 +0000 (10:15 +0200)]
mlx5e_rep: push cls_flower setup_tc processing into a separate function

Let mlx5e_rep_setup_tc (former mlx5e_rep_ndo_setup_tc) be a splitter for
specific setup_tc types and push out cls_flower specific code into
a separate function.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agomlx5e: push cls_flower and mqprio setup_tc processing into separate functions
Jiri Pirko [Mon, 7 Aug 2017 08:15:22 +0000 (10:15 +0200)]
mlx5e: push cls_flower and mqprio setup_tc processing into separate functions

Let mlx5e_setup_tc (former mlx5e_ndo_setup_tc) be a splitter for specific
setup_tc types and push out cls_flower and mqprio specific codes into
separate functions. Also change the return values so they are the same
as in the rest of the drivers.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoixgbe: push cls_u32 and mqprio setup_tc processing into separate functions
Jiri Pirko [Mon, 7 Aug 2017 08:15:21 +0000 (10:15 +0200)]
ixgbe: push cls_u32 and mqprio setup_tc processing into separate functions

Let __ixgbe_setup_tc be a splitter for specific setup_tc types and push out
cls_u32 and mqprio specific codes into separate functions. Also change
the return values so they are the same as in the rest of the drivers.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agocxgb4: push cls_u32 setup_tc processing into a separate function
Jiri Pirko [Mon, 7 Aug 2017 08:15:20 +0000 (10:15 +0200)]
cxgb4: push cls_u32 setup_tc processing into a separate function

Let cxgb_setup_tc be a splitter for specific setup_tc types and push out
cls_u32 specific code into a separate function.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: sched: make egress_dev flag part of flower offload struct
Jiri Pirko [Mon, 7 Aug 2017 08:15:19 +0000 (10:15 +0200)]
net: sched: make egress_dev flag part of flower offload struct

Since this is specific to flower now, make it part of the flower offload
struct.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: sched: rename TC_SETUP_MATCHALL to TC_SETUP_CLSMATCHALL
Jiri Pirko [Mon, 7 Aug 2017 08:15:18 +0000 (10:15 +0200)]
net: sched: rename TC_SETUP_MATCHALL to TC_SETUP_CLSMATCHALL

In order to be aligned with the rest of the types, rename
TC_SETUP_MATCHALL to TC_SETUP_CLSMATCHALL.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: sched: make type an argument for ndo_setup_tc
Jiri Pirko [Mon, 7 Aug 2017 08:15:17 +0000 (10:15 +0200)]
net: sched: make type an argument for ndo_setup_tc

Since the type is always present, push it to be a separate argument to
ndo_setup_tc. On the way, name the type enum and use it for arg type.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet/mlx5: Increase the maximum flow counters supported
Rabie Loulou [Sun, 9 Jul 2017 10:39:30 +0000 (13:39 +0300)]
net/mlx5: Increase the maximum flow counters supported

Read new NIC capability field which represnts 16 MSBs of the max flow
counters number supported (max_flow_counter_31_16).

Backward compatibility with older firmware is preserved, the modified
driver reads max_flow_counter_31_16 as 0 from the older firmware and
uses up to 64K counters.

Changed flow counter id from 16 bits to 32 bits. Backward compatibility
with older firmware is preserved as we kept the 16 LSBs of the counter
id in place and added 16 MSBs from reserved field.

Changed the background bulk reading of flow counters to work in chunks
of at most 32K counters, to make sure we don't attempt to allocate very
large buffers.

Signed-off-by: Rabie Loulou <rabiel@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5: Fix counter list hardware structure
Rabie Loulou [Mon, 10 Jul 2017 11:35:10 +0000 (14:35 +0300)]
net/mlx5: Fix counter list hardware structure

The counter list hardware structure doesn't contain a clear and
num_of_counters fields, remove them.

These wrong fields were never used by the driver hence no other driver
changes.

Fixes: a351a1b03bf1 ("net/mlx5: Introduce bulk reading of flow counters")
Signed-off-by: Rabie Loulou <rabiel@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5: Delay events till ib registration ends
Erez Shitrit [Wed, 7 Jun 2017 09:14:24 +0000 (12:14 +0300)]
net/mlx5: Delay events till ib registration ends

When mlx5_ib registers itself to mlx5_core as an interface, it will
call mlx5_add_device which will call mlx5_ib interface add callback,
in case the latter successfully returns, only then mlx5_core will add
it to the interface list and async events will be forwarded to mlx5_ib.
Between mlx5_ib interface add callback and mlx5_core adding the mlx5_ib
interface to its devices list, arriving mlx5_core events can be missed
by the new mlx5_ib registering interface.

In other words:
thread 1: mlx5_ib: mlx5_register_interface(dev)
thread 1: mlx5_core: mlx5_add_device(dev)
thread 1: mlx5_core: ctx = dev->add => (mlx5_ib)->mlx5_ib_add
thread 2: mlx5_core_event: **new event arrives, forward to dev_list
thread 1: mlx5_core: add_ctx_to_dev_list(ctx)
/* previous event was missed by the new interface.*/
It is ok to miss events before dev->add (mlx5_ib)->mlx5_ib_add_device
but not after.

We fix this race by accumulating the events that come between the
ib_register_device (inside mlx5_add_device->(dev->add)) till the adding
to the list completes and fire them to the new registering interface
after that.

Fixes: f1ee87fe55c8 ("net/mlx5: Organize device list API in one place")
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5: Add CONFIG_MLX5_ESWITCH Kconfig
Saeed Mahameed [Mon, 5 Jun 2017 12:17:12 +0000 (15:17 +0300)]
net/mlx5: Add CONFIG_MLX5_ESWITCH Kconfig

Allow to selectively build the driver with or without sriov eswitch, VF
representors and TC offloads.

Also remove the need of two ndo ops structures (sriov & basic)
and keep only one unified ndo ops, compile out VF SRIOV ndos when not
needed (MLX5_ESWITCH=n), and for VF netdev calling those ndos will result
in returning -EPERM.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Cc: Jes Sorensen <jsorensen@fb.com>
Cc: kernel-team@fb.com
6 years agonet/mlx5: Separate between E-Switch and MPFS
Saeed Mahameed [Sun, 4 Jun 2017 20:11:55 +0000 (23:11 +0300)]
net/mlx5: Separate between E-Switch and MPFS

Multi-Physical Function Switch (MPFs) is required for when multi-PF
configuration is enabled to allow passing user configured unicast MAC
addresses to the requesting PF.

Before this patch eswitch.c used to manage the HW MPFS l2 table,
E-Switch always (regardless of sriov) enabled vport(0) (NIC PF) vport's
contexts update on unicast mac address list changes, to populate the PF's
MPFS L2 table accordingly.

In downstream patch we would like to allow compiling the driver without
E-Switch functionalities, for that we move MPFS l2 table logic out
of eswitch.c into its own file, and provide Kconfig flag (MLX5_MPFS) to
allow compiling out MPFS for those who don't want Multi-PF support.

NIC PF netdevice will now directly update MPFS l2 table via the new MPFS
API. VF netdevice has no access to MPFS L2 table, so E-Switch will remain
responsible of updating its MPFS l2 table on behalf of its VFs.

Due to this change we also don't require enabling vport(0) (PF vport)
unicast mac changes events anymore, for when SRIOV is not enabled.
Which means E-Switch is now activated only on SRIOV activation, and not
required otherwise.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Cc: Jes Sorensen <jsorensen@fb.com>
Cc: kernel-team@fb.com
6 years agonet/mlx5: Unify vport manager capability check
Saeed Mahameed [Sun, 11 Jun 2017 16:05:10 +0000 (19:05 +0300)]
net/mlx5: Unify vport manager capability check

Expose MLX5_VPORT_MANAGER macro to check for strict vport manager
E-switch and MPFS (Multi Physical Function Switch) abilities.

VPORT manager must be a PF with an ethernet link and with FW advertised
vport group manager capability

Replace older checks with the new macro and use it where needed in
eswitch.c and mlx5e netdev eswitch related flows.

The same macro will be reused in MPFS separation downstream patch.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
6 years agonet/mlx5e: NIC netdev init flow cleanup
Saeed Mahameed [Tue, 6 Jun 2017 06:12:04 +0000 (09:12 +0300)]
net/mlx5e: NIC netdev init flow cleanup

Remove redundant call to unregister vport representor in mlx5e_add error
flow.

Hide the representor priv and eswitch internal structures from en_main.c
as preparation step for downstream patches which would allow building
the driver without support for  representors and eswitch.

Fixes: 6f08a22c5fb2 ("net/mlx5e: Register/unregister vport representors on interface attach/detach")
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
6 years agonet/mlx5e: Rearrange netdevice ops structures
Saeed Mahameed [Tue, 6 Jun 2017 14:46:49 +0000 (17:46 +0300)]
net/mlx5e: Rearrange netdevice ops structures

Since we are going to allow building the driver without eswitch support,
it would be possible to compile out the sriov netdevice ops struct such
that the basic ops instance will be used for non VF devices too.

Add missing udp tunnel ndos into mlx5e_netdev_ops_basic.

While here, rearrange some ndos in the sriov ops struct and put
vf/eswitch related ndos towards the end of it.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
6 years agoMerge branch 'sctp-remove-typedefs-from-structures-part-5'
David S. Miller [Mon, 7 Aug 2017 04:33:43 +0000 (21:33 -0700)]
Merge branch 'sctp-remove-typedefs-from-structures-part-5'

Xin Long says:

====================
sctp: remove typedefs from structures part 5

As we know, typedef is suggested not to use in kernel, even checkpatch.pl
also gives warnings about it. Now sctp is using it for many structures.

All this kind of typedef's using should be removed. This patchset is the
part 5 to remove all typedefs in include/net/sctp/constants.h.

Just as the part 1-4, No any code's logic would be changed in these patches,
only cleaning up.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: remove the typedef sctp_subtype_t
Xin Long [Sat, 5 Aug 2017 12:00:04 +0000 (20:00 +0800)]
sctp: remove the typedef sctp_subtype_t

This patch is to remove the typedef sctp_subtype_t, and
replace with union sctp_subtype in the places where it's
using this typedef.

Note that it doesn't fix many indents although it should,
as sctp_disposition_t's removal would mess them up again.
So better to fix them when removing sctp_disposition_t in
later patch.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: remove the typedef sctp_event_t
Xin Long [Sat, 5 Aug 2017 12:00:03 +0000 (20:00 +0800)]
sctp: remove the typedef sctp_event_t

This patch is to remove the typedef sctp_event_t, and
replace with enum sctp_event in the places where it's
using this typedef.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: remove the typedef sctp_event_timeout_t
Xin Long [Sat, 5 Aug 2017 12:00:02 +0000 (20:00 +0800)]
sctp: remove the typedef sctp_event_timeout_t

This patch is to remove the typedef sctp_event_timeout_t, and
replace with enum sctp_event_timeout in the places where it's
using this typedef.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: remove the typedef sctp_event_other_t
Xin Long [Sat, 5 Aug 2017 12:00:01 +0000 (20:00 +0800)]
sctp: remove the typedef sctp_event_other_t

This patch is to remove the typedef sctp_event_other_t, and
replace with enum sctp_event_other in the places where it's
using this typedef.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: remove the typedef sctp_event_primitive_t
Xin Long [Sat, 5 Aug 2017 12:00:00 +0000 (20:00 +0800)]
sctp: remove the typedef sctp_event_primitive_t

This patch is to remove the typedef sctp_event_primitive_t, and
replace with enum sctp_event_primitive in the places where it's
using this typedef.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: remove the typedef sctp_state_t
Xin Long [Sat, 5 Aug 2017 11:59:59 +0000 (19:59 +0800)]
sctp: remove the typedef sctp_state_t

This patch is to remove the typedef sctp_state_t, and
replace with enum sctp_state in the places where it's
using this typedef.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: remove the typedef sctp_ierror_t
Xin Long [Sat, 5 Aug 2017 11:59:58 +0000 (19:59 +0800)]
sctp: remove the typedef sctp_ierror_t

This patch is to remove the typedef sctp_ierror_t, and
replace with enum sctp_ierror in the places where it's
using this typedef.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: remove the typedef sctp_xmit_t
Xin Long [Sat, 5 Aug 2017 11:59:57 +0000 (19:59 +0800)]
sctp: remove the typedef sctp_xmit_t

This patch is to remove the typedef sctp_xmit_t, and
replace with enum sctp_xmit in the places where it's
using this typedef.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: remove the typedef sctp_sock_state_t
Xin Long [Sat, 5 Aug 2017 11:59:56 +0000 (19:59 +0800)]
sctp: remove the typedef sctp_sock_state_t

This patch is to remove the typedef sctp_sock_state_t, and
replace with enum sctp_sock_state in the places where it's
using this typedef.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: remove the typedef sctp_transport_cmd_t
Xin Long [Sat, 5 Aug 2017 11:59:55 +0000 (19:59 +0800)]
sctp: remove the typedef sctp_transport_cmd_t

This patch is to remove the typedef sctp_transport_cmd_t, and
replace with enum sctp_transport_cmd in the places where it's
using this typedef.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: remove the typedef sctp_scope_t
Xin Long [Sat, 5 Aug 2017 11:59:54 +0000 (19:59 +0800)]
sctp: remove the typedef sctp_scope_t

This patch is to remove the typedef sctp_scope_t, and
replace with enum sctp_scope in the places where it's
using this typedef.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: remove the typedef sctp_scope_policy_t
Xin Long [Sat, 5 Aug 2017 11:59:53 +0000 (19:59 +0800)]
sctp: remove the typedef sctp_scope_policy_t

This patch is to remove the typedef sctp_scope_policy_t and keep
it's members as an anonymous enum.

It is also to define SCTP_SCOPE_POLICY_MAX to replace the num 3
in sysctl.c to make codes clear.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: remove the typedef sctp_retransmit_reason_t
Xin Long [Sat, 5 Aug 2017 11:59:52 +0000 (19:59 +0800)]
sctp: remove the typedef sctp_retransmit_reason_t

This patch is to remove the typedef sctp_retransmit_reason_t, and
replace with enum sctp_retransmit_reason in the places where it's
using this typedef.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosctp: remove the typedef sctp_lower_cwnd_t
Xin Long [Sat, 5 Aug 2017 11:59:51 +0000 (19:59 +0800)]
sctp: remove the typedef sctp_lower_cwnd_t

This patch is to remove the typedef sctp_lower_cwnd_t, and
replace with enum sctp_lower_cwnd in the places where it's
using this typedef.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agodt-bindings: net: Document bindings for anarion-gmac
Alexandru Gagniuc [Fri, 4 Aug 2017 20:08:52 +0000 (13:08 -0700)]
dt-bindings: net: Document bindings for anarion-gmac

Signed-off-by: Alexandru Gagniuc <alex.g@adaptrum.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: stmmac: Add Adaptrum Anarion GMAC glue layer
Alexandru Gagniuc [Fri, 4 Aug 2017 20:08:51 +0000 (13:08 -0700)]
net: stmmac: Add Adaptrum Anarion GMAC glue layer

Before the GMAC on the Anarion chip can be used, the PHY interface
selection must be configured with the DWMAC block in reset.

This layer covers a block containing only two registers. Although it
is possible to model this as a reset controller and use the "resets"
property of stmmac, it's much more intuitive to include this in the
glue layer instead.

At this time only RGMII is supported, because it is the only mode
which has been validated hardware-wise.

Signed-off-by: Alexandru Gagniuc <alex.g@adaptrum.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonetvsc: fix rtnl deadlock on unregister of vf
stephen hemminger [Fri, 4 Aug 2017 19:14:00 +0000 (12:14 -0700)]
netvsc: fix rtnl deadlock on unregister of vf

With new transparent VF support, it is possible to get a deadlock
when some of the deferred work is running and the unregister_vf
is trying to cancel the work element. The solution is to use
trylock and reschedule (similar to bonding and team device).

Reported-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Fixes: 0c195567a8f6 ("netvsc: transparent VF management")
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: dsa: User per-cpu 64-bit statistics
Florian Fainelli [Fri, 4 Aug 2017 04:33:27 +0000 (21:33 -0700)]
net: dsa: User per-cpu 64-bit statistics

During testing with a background iperf pushing 1Gbit/sec worth of
traffic and having both ifconfig and ethtool collect statistics, we
could see quite frequent deadlocks. Convert the often accessed DSA slave
network devices statistics to per-cpu 64-bit statistics to remove these
deadlocks and provide fast efficient statistics updates.

Fixes: f613ed665bb3 ("net: dsa: Add support for 64-bit statistics")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoMerge branch 'tcp-cwnd-undo-refactor'
David S. Miller [Mon, 7 Aug 2017 04:25:10 +0000 (21:25 -0700)]
Merge branch 'tcp-cwnd-undo-refactor'

Yuchung Cheng says:

====================
tcp cwnd undo refactor

This patch series consolidate similar cwnd undo functions
implemented by various congestion control by using existing
tcp socket state variable. The first patch fixes a corner
case in of cwnd undo in Reno and HTCP. Since the bug has
existed for many years and is very minor, we consider this
patch set more suitable for net-next as the major change
is the refactor itself.

- v1->v2
  Fix trivial compile errors
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: consolidate congestion control undo functions
Yuchung Cheng [Fri, 4 Aug 2017 03:38:52 +0000 (20:38 -0700)]
tcp: consolidate congestion control undo functions

Most TCP congestion controls are using identical logic to undo
cwnd except BBR. This patch consolidates these similar functions
to the one used currently by Reno and others.

Suggested-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agotcp: fix cwnd undo in Reno and HTCP congestion controls
Yuchung Cheng [Fri, 4 Aug 2017 03:38:51 +0000 (20:38 -0700)]
tcp: fix cwnd undo in Reno and HTCP congestion controls

Using ssthresh to revert cwnd is less reliable when ssthresh is
bounded to 2 packets. This patch uses an existing variable in TCP
"prior_cwnd" that snapshots the cwnd right before entering fast
recovery and RTO recovery in Reno.  This fixes the issue discussed
in netdev thread: "A buggy behavior for Linux TCP Reno and HTCP"
https://www.spinics.net/lists/netdev/msg444955.html

Suggested-by: Neal Cardwell <ncardwell@google.com>
Reported-by: Wei Sun <unlcsewsun@gmail.com>
Signed-off-by: Yuchung Cheng <ncardwell@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: systemport: Support 64bit statistics
kiki good [Thu, 3 Aug 2017 23:07:45 +0000 (00:07 +0100)]
net: systemport: Support 64bit statistics

When using Broadcom Systemport device in 32bit Platform, ifconfig can
only report up to 4G tx,rx status, which will be wrapped to 0 when the
number of incoming or outgoing packets exceeds 4G, only taking
around 2 hours in busy network environment (such as streaming).
Therefore, it makes hard for network diagnostic tool to get reliable
statistical result, so the patch is used to add 64bit support for
Broadcom Systemport device in 32bit Platform.

This patch provides 64bit statistics capability on both ethtool and ifconfig.

Signed-off-by: Jianming.qiao <kiki-good@hotmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agoliquidio: moved console_bitmask module param to lio_main.c
Intiyaz Basha [Thu, 3 Aug 2017 22:10:17 +0000 (15:10 -0700)]
liquidio: moved console_bitmask module param to lio_main.c

Moving PF module param console_bitmask to lio_main.c for consistency.

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 agoliquidio: add missing strings in oct_dev_state_str array
Intiyaz Basha [Thu, 3 Aug 2017 20:08:24 +0000 (13:08 -0700)]
liquidio: add missing strings in oct_dev_state_str array

There's supposed to be a one-to-one correspondence between the 18 macros
that #define the OCT_DEV states (in octeon_device.h) and the strings in the
oct_dev_state_str array, but there are only 14 strings in the array.

Add the missing strings (so they become 18 in total), and also revise some
incorrect/outdated text of existing strings.

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 'phylink-and-sfp-support'
David S. Miller [Mon, 7 Aug 2017 03:55:29 +0000 (20:55 -0700)]
Merge branch 'phylink-and-sfp-support'

Russell King says:

====================
phylink and sfp support

This patch series introduces generic support for SFP sockets found on
various Marvell based platforms.  The idea here is to provide common
SFP socket support which can be re-used by network drivers as
appropriate, rather than each network driver having to re-implement
SFP socket support.

SFP sockets typically use other system resources, eg, I2C buses to read
identifying information, and GPIOs to monitor socket state and control
the socket.  Meanwhile, some network drivers drive multiple ethernet
ports from one instantiation of the driver.

It is not desirable to block the initialisation of a network driver
(thus denying other ports from being operational) if the resources
for the SFP socket are not yet available.  This means that an element
of independence between the SFP support code and the driver is
required.

More than that, SFP modules effectively bring hotplug PHYs to
networking - SFP copper modules normally contain a standard PHY
accessed over the I2C bus, and it is desirable to read their state
so network drivers can be appropriately configured.

To add to the complexity, SFP modules can be connected in at least
two places:

1. Directly to the serdes output of a MAC with no intervening PHY.
   For example:

     mvneta ----> SFP socket

2. To a PHY, for example:

     mvpp2 ---> PHY ---> copper
                 |
                 `-----> SFP socket

This code supports both setups, although it's not fully implemented
with scenario (2).

Moreover, the link presented by the SFP module can be one of the
10Gbase-R family (for SFP+ sockets), SGMII or 1000base-X (for SFP
sockets) depending on the module, and network drivers need to
reconfigure themselves accordingly for the link to come up.

For example, if the MAC is configured for SGMII and a fibre module
is plugged in, the link won't come up until the MAC is reconfigured
for 1000base-X mode.

The SFP code manages the SFP socket - detecting the module, reading
the identifying information, and managing the control and status
signals.  Importantly, it disables the SFP module transmitter when
the MAC is down, so that the laser is turned off (but that is not
a guarantee.)

phylink provides the mechanisms necessary to manage the link modes,
based on the SFP module type, and supports hot-plugging of the PHY
without needing the MAC driver to be brought up and down on
transitions.  phylink also supports the classical static PHY and
fixed-link modes.

I currently (but not included in this series) have code to convert
mvneta to use phylink, and the out of tree mvpp2x driver.  I have
nothing for the mvpp2 driver at present as that driver is only
recently becoming functional on 10G hardware, and is missing a lot
of features that are necessary to make things work correctly.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosfp: add SFP module support
Russell King [Tue, 25 Jul 2017 14:03:39 +0000 (15:03 +0100)]
sfp: add SFP module support

Add support for SFP hotpluggable modules via sfp-bus and phylink.
This supports both copper and optical SFP modules, which require
different Serdes modes in order to properly negotiate the link.

Optical SFP modules typically require the Serdes link to be talking
1000BaseX mode - this is the gigabit ethernet mode defined by the
802.3 standard.

Copper SFP modules typically integrate a PHY in the module to convert
from Serdes to copper, and the PHY will be configured by the vendor
to either present a 1000BaseX Serdes link (for fixed 1000BaseT) or a
SGMII Serdes link.  However, this is vendor defined, so we instead
detect the PHY, switch the link to SGMII mode, and use traditional
PHY based negotiation.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agophylink: add in-band autonegotiation support for 10GBase-KR mode.
Russell King [Tue, 25 Jul 2017 14:03:34 +0000 (15:03 +0100)]
phylink: add in-band autonegotiation support for 10GBase-KR mode.

Add in-band autonegotation support for 10GBase-KR mode.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agophylink: add support for MII ioctl access to Clause 45 PHYs
Russell King [Tue, 25 Jul 2017 14:03:28 +0000 (15:03 +0100)]
phylink: add support for MII ioctl access to Clause 45 PHYs

Add support for reading and writing the clause 45 MII registers.

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agophylink: add module EEPROM support
Russell King [Tue, 25 Jul 2017 14:03:23 +0000 (15:03 +0100)]
phylink: add module EEPROM support

Add support for reading module EEPROMs through phylink.

Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agosfp: add sfp-bus to bridge between network devices and sfp cages
Russell King [Tue, 25 Jul 2017 14:03:18 +0000 (15:03 +0100)]
sfp: add sfp-bus to bridge between network devices and sfp cages

Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agophylink: add phylink infrastructure
Russell King [Tue, 25 Jul 2017 14:03:13 +0000 (15:03 +0100)]
phylink: add phylink infrastructure

The link between the ethernet MAC and its PHY has become more complex
as the interface evolves.  This is especially true with serdes links,
where the part of the PHY is effectively integrated into the MAC.

Serdes links can be connected to a variety of devices, including SFF
modules soldered down onto the board with the MAC, a SFP cage with
a hotpluggable SFP module which may contain a PHY or directly modulate
the serdes signals onto optical media with or without a PHY, or even
a classical PHY connection.

Moreover, the negotiation information on serdes links comes in two
varieties - SGMII mode, where the PHY provides its speed/duplex/flow
control information to the MAC, and 1000base-X mode where both ends
exchange their abilities and each resolve the link capabilities.

This means we need a more flexible means to support these arrangements,
particularly with the hotpluggable nature of SFP, where the PHY can
be attached or detached after the network device has been brought up.

Ethtool information can come from multiple sources:
- we may have a PHY operating in either SGMII or 1000base-X mode, in
  which case we take ethtool/mii data directly from the PHY.
- we may have a optical SFP module without a PHY, with the MAC
  operating in 1000base-X mode - the ethtool/mii data needs to come
  from the MAC.
- we may have a copper SFP module with a PHY whic can't be accessed,
  which means we need to take ethtool/mii data from the MAC.

Phylink aims to solve this by providing an intermediary between the
MAC and PHY, providing a safe way for PHYs to be hotplugged, and
allowing a SFP driver to reconfigure the serdes connection.

Phylink also takes over support of fixed link connections, where the
speed/duplex/flow control are fixed, but link status may be controlled
by a GPIO signal.  By avoiding the fixed-phy implementation, phylink
can provide a faster response to link events: fixed-phy has to wait for
phylib to operate its state machine, which can take several seconds.
In comparison, phylink takes milliseconds.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
- remove sync status
- rework supported and advertisment handling
- add 1000base-x speed for fixed links
- use functionality exported from phy-core, reworking
  __phylink_ethtool_ksettings_set for it
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
6 years agonet: phy: add I2C mdio bus
Russell King [Tue, 25 Jul 2017 14:03:08 +0000 (15:03 +0100)]
net: phy: add I2C mdio bus

Add an I2C MDIO bus bridge library, to allow phylib to access PHYs which
are connected to an I2C bus instead of the more conventional MDIO bus.
Such PHYs can be found in SFP adapters and SFF modules.

Since PHYs appear at I2C bus address 0x40..0x5f, and 0x50/0x51 are
reserved for SFP EEPROMs/diagnostics, we must not allow the MDIO bus
to access these I2C addresses.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>