openwrt/staging/blogic.git
5 years agonet: phy: add genphy_c45_an_config_aneg
Andrew Lunn [Sun, 17 Feb 2019 09:29:19 +0000 (10:29 +0100)]
net: phy: add genphy_c45_an_config_aneg

C45 configuration of 10/100 and multi-giga bit auto negotiation
advertisement is standardized. Configuration of 1000Base-T however
appears to be vendor specific. Move the generic code out of the
Marvell driver into the common phy-c45.c file.

v2:
- change function name to genphy_c45_an_config_aneg

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
[hkallweit1@gmail.com: use new helper linkmode_adv_to_mii_10gbt_adv_t and split patch]
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: add helper linkmode_adv_to_mii_10gbt_adv_t
Heiner Kallweit [Sun, 17 Feb 2019 09:28:33 +0000 (10:28 +0100)]
net: phy: add helper linkmode_adv_to_mii_10gbt_adv_t

Add a helper linkmode_adv_to_mii_10gbt_adv_t(), similar to
linkmode_adv_to_mii_adv_t.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
David S. Miller [Sun, 17 Feb 2019 06:56:34 +0000 (22:56 -0800)]
Merge git://git./linux/kernel/git/bpf/bpf-next

Alexei Starovoitov says:

====================
pull-request: bpf-next 2019-02-16

The following pull-request contains BPF updates for your *net-next* tree.

The main changes are:

1) numerous libbpf API improvements, from Andrii, Andrey, Yonghong.

2) test all bpf progs in alu32 mode, from Jiong.

3) skb->sk access and bpf_sk_fullsock(), bpf_tcp_sock() helpers, from Martin.

4) support for IP encap in lwt bpf progs, from Peter.

5) remove XDP_QUERY_XSK_UMEM dead code, from Jan.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotools/libbpf: support bigger BTF data sizes
Andrii Nakryiko [Sat, 16 Feb 2019 03:52:18 +0000 (19:52 -0800)]
tools/libbpf: support bigger BTF data sizes

While it's understandable why kernel limits number of BTF types to 65535
and size of string section to 64KB, in libbpf as user-space library it's
too restrictive. E.g., pahole converting DWARF to BTF type information
for Linux kernel generates more than 3 million BTF types and more than
3MB of strings, before deduplication. So to allow btf__dedup() to do its
work, we need to be able to load bigger BTF sections using btf__new().

Singed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agoselftests: bpf: test_lwt_ip_encap: add negative tests.
Peter Oskolkov [Fri, 15 Feb 2019 23:49:33 +0000 (15:49 -0800)]
selftests: bpf: test_lwt_ip_encap: add negative tests.

As requested by David Ahern:

- add negative tests (no routes, explicitly unreachable destinations)
  to exercize error handling code paths;
- do not exit on test failures, but instead print a summary of
  passed/failed tests at the end.

Future patches will add TSO and VRF tests.

Signed-off-by: Peter Oskolkov <posk@google.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agonet: stmmac: use correct define to get rx timestamp on GMAC4
Alexandre Torgue [Thu, 14 Feb 2019 16:03:44 +0000 (17:03 +0100)]
net: stmmac: use correct define to get rx timestamp on GMAC4

In dwmac4_wrback_get_rx_timestamp_status we looking for a RX timestamp.
For that receive descriptors are handled and so we should use defines
related to receive descriptors. It'll no change the functional behavior
as RDES3_RDES1_VALID=TDES3_RS1V=BIT(26) but it makes code easier to read.

Signed-off-by: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoatm: clean up vcc_seq_next()
Dan Carpenter [Thu, 14 Feb 2019 06:56:35 +0000 (09:56 +0300)]
atm: clean up vcc_seq_next()

It's confusing to call PTR_ERR(v).  The PTR_ERR() function is basically
a fancy cast to long so it makes you wonder, was IS_ERR() intended?  But
that doesn't make sense because vcc_walk() doesn't return error
pointers.

This patch doesn't affect runtime, it's just a cleanup.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosock: consistent handling of extreme SO_SNDBUF/SO_RCVBUF values
Guillaume Nault [Wed, 13 Feb 2019 03:30:34 +0000 (04:30 +0100)]
sock: consistent handling of extreme SO_SNDBUF/SO_RCVBUF values

SO_SNDBUF and SO_RCVBUF (and their *BUFFORCE version) may overflow or
underflow their input value. This patch aims at providing explicit
handling of these extreme cases, to get a clear behaviour even with
values bigger than INT_MAX / 2 or lower than INT_MIN / 2.

For simplicity, only SO_SNDBUF and SO_SNDBUFFORCE are described here,
but the same explanation and fix apply to SO_RCVBUF and SO_RCVBUFFORCE
(with 'SNDBUF' replaced by 'RCVBUF' and 'wmem_max' by 'rmem_max').

Overflow of positive values

===========================

When handling SO_SNDBUF or SO_SNDBUFFORCE, if 'val' exceeds
INT_MAX / 2, the buffer size is set to its minimum value because
'val * 2' overflows, and max_t() considers that it's smaller than
SOCK_MIN_SNDBUF. For SO_SNDBUF, this can only happen with
net.core.wmem_max > INT_MAX / 2.

SO_SNDBUF and SO_SNDBUFFORCE are actually designed to let users probe
for the maximum buffer size by setting an arbitrary large number that
gets capped to the maximum allowed/possible size. Having the upper
half of the positive integer space to potentially reduce the buffer
size to its minimum value defeats this purpose.

This patch caps the base value to INT_MAX / 2, so that bigger values
don't overflow and keep setting the buffer size to its maximum.

Underflow of negative values
============================

For negative numbers, SO_SNDBUF always considers them bigger than
net.core.wmem_max, which is bounded by [SOCK_MIN_SNDBUF, INT_MAX].
Therefore such values are set to net.core.wmem_max and we're back to
the behaviour of positive integers described above (return maximum
buffer size if wmem_max <= INT_MAX / 2, return SOCK_MIN_SNDBUF
otherwise).

However, SO_SNDBUFFORCE behaves differently. The user value is
directly multiplied by two and compared with SOCK_MIN_SNDBUF. If
'val * 2' doesn't underflow or if it underflows to a value smaller
than SOCK_MIN_SNDBUF then buffer size is set to its minimum value.
Otherwise the buffer size is set to the underflowed value.

This patch treats negative values passed to SO_SNDBUFFORCE as null, to
prevent underflows. Therefore negative values now always set the buffer
size to its minimum value.

Even though SO_SNDBUF behaves inconsistently by setting buffer size to
the maximum value when passed a negative number, no attempt is made to
modify this behaviour. There may exist some programs that rely on using
negative numbers to set the maximum buffer size. Avoiding overflows
because of extreme net.core.wmem_max values is the most we can do here.

Summary of altered behaviours
=============================

val      : user-space value passed to setsockopt()
val_uf   : the underflowed value resulting from doubling val when
           val < INT_MIN / 2
wmem_max : short for net.core.wmem_max
val_cap  : min(val, wmem_max)
min_len  : minimal buffer length (that is, SOCK_MIN_SNDBUF)
max_len  : maximal possible buffer length, regardless of wmem_max (that
           is, INT_MAX - 1)
^^^^     : altered behaviour

SO_SNDBUF:
+-------------------------+-------------+------------+----------------+
|       CONDITION         | OLD RESULT  | NEW RESULT |    COMMENT     |
+-------------------------+-------------+------------+----------------+
| val < 0 &&              |             |            | No overflow,   |
| wmem_max <= INT_MAX/2   | wmem_max*2  | wmem_max*2 | keep original  |
|                         |             |            | behaviour      |
+-------------------------+-------------+------------+----------------+
| val < 0 &&              |             |            | Cap wmem_max   |
| INT_MAX/2 < wmem_max    | min_len     | max_len    | to prevent     |
|                         |             | ^^^^^^^    | overflow       |
+-------------------------+-------------+------------+----------------+
| 0 <= val <= min_len/2   | min_len     | min_len    | Ordinary case  |
+-------------------------+-------------+------------+----------------+
| min_len/2 < val &&      | val_cap*2   | val_cap*2  | Ordinary case  |
| val_cap <= INT_MAX/2    |             |            |                |
+-------------------------+-------------+------------+----------------+
| min_len < val &&        |             |            | Cap val_cap    |
| INT_MAX/2 < val_cap     | min_len     | max_len    | again to       |
| (implies that           |             | ^^^^^^^    | prevent        |
| INT_MAX/2 < wmem_max)   |             |            | overflow       |
+-------------------------+-------------+------------+----------------+

SO_SNDBUFFORCE:
+------------------------------+---------+---------+------------------+
|          CONDITION           | BEFORE  | AFTER   |     COMMENT      |
|                              | PATCH   | PATCH   |                  |
+------------------------------+---------+---------+------------------+
| val < INT_MIN/2 &&           | min_len | min_len | Underflow with   |
| val_uf <= min_len            |         |         | no consequence   |
+------------------------------+---------+---------+------------------+
| val < INT_MIN/2 &&           | val_uf  | min_len | Set val to 0 to  |
| val_uf > min_len             |         | ^^^^^^^ | avoid underflow  |
+------------------------------+---------+---------+------------------+
| INT_MIN/2 <= val < 0         | min_len | min_len | No underflow     |
+------------------------------+---------+---------+------------------+
| 0 <= val <= min_len/2        | min_len | min_len | Ordinary case    |
+------------------------------+---------+---------+------------------+
| min_len/2 < val <= INT_MAX/2 | val*2   | val*2   | Ordinary case    |
+------------------------------+---------+---------+------------------+
| INT_MAX/2 < val              | min_len | max_len | Cap val to       |
|                              |         | ^^^^^^^ | prevent overflow |
+------------------------------+---------+---------+------------------+

Signed-off-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge tag 'mlx5-updates-2019-02-15' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Sat, 16 Feb 2019 20:11:17 +0000 (12:11 -0800)]
Merge tag 'mlx5-updates-2019-02-15' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
Support Mellanox BlueField SmartNIC (mlx5-updates-2019-02-15)

Bodong Wang says,

BlueField device is a multi-core ARM processor in a highly integrated
system on chip coupled with the ConnectX interconnect controller.
BlueField device can be presented in one out of two modes:

- SEPARATED_HOST: ARM processors as a separated and orthogonal host
  like any other external host in the multi-host virtualization model.
- EMBEDDED_CPU: ARM processors as Embedded CPU (EC) and part of the
  external hosts virtualization model.

While existing driver already supports the device on separated_host
mode, this patch series focus on the functionalities of embedded_cpu
mode.

On embedded_cpu mode, BlueField device exposes regular network
controller PCI function in the BlueField host(e.g, x86). However, a
separate PCI function called Embedded CPU Physical Function(ECPF) is
also added to the ARM host side, where standard Linux distributions is
able to run on the ARM cores. Depends on the NV configuration from
firmware, ECPF can be the e-switch manager and firmware pages supplier.
If ECPF is configured as e-switch manager and page supplier, it will
take over the responsibilities from the PF on BlueField host includes:
- Owns, controls and manages all e-switch parts, and takes e-switch
  traffic by default. It also should perform ENABLE_HCA for the host
  PF just like a PF does for its VFs.
- Provides and manages the ICM host memory required for the HCA to
  store various contexts for itself, the PF and VFs belong the
  e-switch it manages.

The PF on BlueField host side is still responsible for:
- Control its own permanent MAC.
- PCI and SRIOV configurations and perform ENABLE_HCA for its VFs.

The ECPF can also retrieve information about the external host it
controls, like host identifier, PCI BDF and number of virtual functions.
As these parameters may be changed dynamically, an event will be triggered
to the driver on ECPF side.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 's390-next'
David S. Miller [Sat, 16 Feb 2019 04:35:30 +0000 (20:35 -0800)]
Merge branch 's390-next'

Julian Wiedmann says:

====================
s390/qeth: updates 2019-02-15

please apply a few more qeth patches to net-next. Along with some smaller
improvements, this revamps our code for the SW statistics that are exposed
through ETHTOOL_GSTATS.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agos390/qeth: split out OSN netdev ops
Julian Wiedmann [Fri, 15 Feb 2019 18:22:31 +0000 (19:22 +0100)]
s390/qeth: split out OSN netdev ops

Rather than special-casing OSN in a number of places, just give this
device type its own netdev_ops structure.

When setting up the OSN net_device, also skip the handling of the
various HW offloads (eg TSO). The device shouldn't be advertising any of
them, and the OSN code paths in qeth don't have support for them.
In particular RX VLAN filtering is not supported, so don't hook up those
callbacks in the netdev_ops.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agos390/qeth: add support for ETHTOOL_GRINGPARAM
Julian Wiedmann [Fri, 15 Feb 2019 18:22:30 +0000 (19:22 +0100)]
s390/qeth: add support for ETHTOOL_GRINGPARAM

Implement a trivial callback that exposes the queue sizes.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agos390/qeth: overhaul ethtool statistics
Julian Wiedmann [Fri, 15 Feb 2019 18:22:29 +0000 (19:22 +0100)]
s390/qeth: overhaul ethtool statistics

Accumulate per-TX queue statistics, and increase their size to 64 bit.
Don't bother with enabling/disabling the statistics, the overhead is
negligible.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agos390/qeth: move ethtool code into its own file
Julian Wiedmann [Fri, 15 Feb 2019 18:22:28 +0000 (19:22 +0100)]
s390/qeth: move ethtool code into its own file

Most of this is self-contained code.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agos390/qeth: reduce ethtool statistics
Julian Wiedmann [Fri, 15 Feb 2019 18:22:27 +0000 (19:22 +0100)]
s390/qeth: reduce ethtool statistics

Counting the number of function calls and the time spent in functions
is best left to proper tracing facilities.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agos390/qeth: use a static Output Queue array
Julian Wiedmann [Fri, 15 Feb 2019 18:22:26 +0000 (19:22 +0100)]
s390/qeth: use a static Output Queue array

qeth dynamically allocates an array for storing pointers to its
Output Queue structures. Switch this to a static array - we are
currently limited to 4 Output Queues, so shrinking the qeth_qdio_info
struct by just a few bytes doesn't justify the additional complexity.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agos390/qeth: allow manual recovery when device is SOFTSETUP
Julian Wiedmann [Fri, 15 Feb 2019 18:22:25 +0000 (19:22 +0100)]
s390/qeth: allow manual recovery when device is SOFTSETUP

Once a qeth ccwgroup device is set online, it's also armed for internal
recovery. So allow for testing that code path via sysfs, regardless of
whether the interface is up or down.

Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: forwarding: Add some missing configuration symbols
Florian Fainelli [Fri, 15 Feb 2019 18:14:52 +0000 (10:14 -0800)]
selftests: forwarding: Add some missing configuration symbols

For the forwarding selftests to work, we need network namespaces when
using veth/vrf otherwise ping/ping6 commands like these:

ip vrf exec vveth0 /bin/ping 192.0.2.2 -c 10 -i 0.1 -w 5

will fail because network namespaces may not be enabled.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/ipv6: prefer rcu_access_pointer() over rcu_dereference()
Paolo Abeni [Fri, 15 Feb 2019 17:15:37 +0000 (18:15 +0100)]
net/ipv6: prefer rcu_access_pointer() over rcu_dereference()

rt6_cache_allowed_for_pmtu() checks for rt->from presence, but
it does not access the RCU protected pointer. We can use
rcu_access_pointer() and clean-up the code a bit. No functional
changes intended.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: core: fix spelling mistake "temprature" -> "temperature"
Colin Ian King [Fri, 15 Feb 2019 15:11:53 +0000 (15:11 +0000)]
mlxsw: core: fix spelling mistake "temprature" -> "temperature"

There is a spelling mistake in several dev_err messages, fix these.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/mlx5: E-Switch, Allow transition to offloads mode for ECPF
Bodong Wang [Thu, 20 Dec 2018 23:28:06 +0000 (17:28 -0600)]
net/mlx5: E-Switch, Allow transition to offloads mode for ECPF

Currently, the e-switch driver requires going to legacy mode before
changing to the offloads mode. This makes sense for regular case as
the legacy mode is done by creating VFs.

However, it's problematic when ECPF is the eswitch manager. In such
case, ECPF will control the vports on peer host including the peer
PF and VFs. But ECPF doesn't need and shall not create VFs as the
VFs are created in the peer PF host.

Grant ECPF the ability to change from none to the offloads mode. Note
that currently the only way to go back to none mode is by unloading
the ECPF driver.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: E-Switch, Load/unload VF reps according to event from host PF
Bodong Wang [Wed, 30 Jan 2019 05:13:13 +0000 (23:13 -0600)]
net/mlx5: E-Switch, Load/unload VF reps according to event from host PF

When host PF changes the number of VFs, the ECPF esw driver will get
a FW event. It should query the number of VFs enabled by host PF and
update the VF reps accordingly. Note that host PF can't change the
number of VFs dynamically, it has to reset the number of VFs to 0
before changing to a new positive number.

The host event is registered when driver is moving to switchdev mode,
and it's the last step to do in esw_offloads_init. It's unregistered
and the work queue is flushed when driver quits from switchdev mode.
In this way, the host event and devlink command are serialized.

When driver is enabling switchdev mode, pay attention to the following
two facts:
1. Host PF must not have VF initialized as the flow table in ECPF has
   ENCAP enabled as default. Such flow table can't be created with
   existing initialized VFs.
2. ECPF doesn't know how many VFs the host PF will enable, ECPF
   offloads flow steering shall create the flow table/groups based on
   the max number of VFs possibly supported by host PF.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: E-Switch, Consider ECPF vport depends on eswitch ownership
Bodong Wang [Mon, 10 Dec 2018 17:59:33 +0000 (11:59 -0600)]
net/mlx5: E-Switch, Consider ECPF vport depends on eswitch ownership

ECPF connects to the eswitch through vport 0xfffe. ECPF may or may
not be the eswitch manager depending on firmware configuration.

1. If ECPF is eswitch manager: ECPF will take over the eswitch manager
   responsibility. A rep of the host PF shall be created at the ECPF
   side for the eswitch manager to control.

2. If ECPF is not eswitch manager: host PF will be the eswitch manager,
   ECPF acts similar as a VF to the host PF. Host PF will be aware
   of the ECPF vport presence and control it's rep.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: E-Switch, Assign a different position for uplink rep and vport
Bodong Wang [Fri, 14 Dec 2018 15:33:22 +0000 (09:33 -0600)]
net/mlx5: E-Switch, Assign a different position for uplink rep and vport

In offloads mode, the current implementation puts the uplink
representor at index zero of the vport reps array. It is not "natural"
to place it at index 0 since we want to put the representor for vport
0 at index 0 with the introduction of SmartNIC. A separate patch will
handle the case whether a rep is needed for vport 0 (PF vport).

So, we want to have a different placeholder for uplink vport and
representor. It was placed at the end of vport and rep array. Since
vport number can no longer act as an index into the vport or
representors arrays, use functions to map vport numbers to indices
when accessing the vports or representors arrays, and vice versa.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Signed-off-by: Eli Cohen <eli@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: E-Switch, Centralize repersentor reg/unreg to eswitch driver
Bodong Wang [Thu, 31 Jan 2019 23:42:57 +0000 (17:42 -0600)]
net/mlx5: E-Switch, Centralize repersentor reg/unreg to eswitch driver

Eswitch has two users: IB and ETH. They both register repersentors
when mlx5 interface is added, and unregister the repersentors when
mlx5 interface is removed. Ideally, each driver should only deal with
the entities which are unique to itself. However, current IB and ETH
drivers have to perform the following eswitch operations:

1. When registering, specify how many vports to register. This number
   is the same for both drivers which is the total available vport
   numbers.
2. When unregistering, specify the number of registered vports to do
   unregister. Also, unload the repersentors which are already loaded.

It's unnecessary for eswitch driver to hands out the control of above
operations to individual driver users, as they're not unique to each
driver. Instead, such operations should be centralized to eswitch
driver. This consolidates eswitch control flow, and simplified IB and
ETH driver.

This patch doesn't change any functionality.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: E-Switch, Support load/unload reps of specific vport types
Bodong Wang [Wed, 30 Jan 2019 04:57:21 +0000 (22:57 -0600)]
net/mlx5: E-Switch, Support load/unload reps of specific vport types

Currently the driver loads and unloads all reps in an unbreakable
group. However, with ECPF, the reps of special vports such as uplink
and host PF should always be loaded in switchdev mode where the reps
for VFs will be loaded on-demand and unloaded on no-demand. This is
a pre-step for that change.

This patch doesn't change any functionality.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: E-Switch, Add state to eswitch vport representors
Bodong Wang [Wed, 30 Jan 2019 03:48:31 +0000 (21:48 -0600)]
net/mlx5: E-Switch, Add state to eswitch vport representors

Currently the eswitch vport reps have a valid indicator, which is
set on register and unset on unregister. However, a rep can be loaded
or not loaded when doing unregister, current driver checks if the
vport of that rep is enabled as a flag to imply the rep is loaded.
However, for ECPF, this is not valid as the host PF will enable the
vports for its VFs instead.

Add three states: {unregistered, registered, loaded}, with the
following state changes across different operations:

create: (none)       -> unregistered
reg:    unregistered -> registered
load:   registered   -> loaded
unload: loaded       -> registered
unreg:  registered   -> unregistered

Note that the state shall only be updated inside eswitch driver rather
than individual drivers such as ETH or IB.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Suggested-by: Mark Bloch <markb@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: E-Switch, Use getter and iterator to access vport/rep
Bodong Wang [Tue, 29 Jan 2019 04:12:45 +0000 (22:12 -0600)]
net/mlx5: E-Switch, Use getter and iterator to access vport/rep

With only PF and VF, it is sufficient to have the vport/rep array
index as the vport number. This is because PF and VF vports numbers
are consecutive serial numbers. In downstream patches with
introducing of ECPF and UPLINK vports, it's not consecutive any more.

Use getter to get specific vport/rep, and use iterator to traversal
a list of vport/rep. This hides the translation between array index
and vport number, and provides flexibility of using different
translation mechanism in the future.

This patch doesn't change any functionality.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Suggested-by: Saeed Mahameed <saeedm@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: E-Switch, Split VF and special vports for offloads mode
Bodong Wang [Thu, 31 Jan 2019 20:40:53 +0000 (14:40 -0600)]
net/mlx5: E-Switch, Split VF and special vports for offloads mode

When driver is entering offloads mode, there are two major tasks to
do: initialize flow steering and create representors. Flow steering
should make sure enough flow table/group spaces are reserved for all
reps. Representors will be created in a group, all or none.

With the introduction of ECPF, flow steering should still reserve the
same spaces. But, the representors are not always loaded/unloaded in a
single piece. Once ECPF is in offloads mode, it will get the number
of VF changing event from host PF. In such scenario, only the VF reps
should be loaded/unloaded, not the reps for special vports (such as
the uplink vport).

Thus, when entering offloads mode, driver should specify the total
number of reps, and the number of VF reps separately. When leaving
offloads mode, the cleanup should use the information self-contained
in eswitch such as number of VFs.

This patch doesn't change any functionality.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: E-Switch, Refactor offloads flow steering init/cleanup
Bodong Wang [Thu, 7 Feb 2019 16:40:58 +0000 (10:40 -0600)]
net/mlx5: E-Switch, Refactor offloads flow steering init/cleanup

E-switch offloads mode initialize/cleanup multiple steering related
entities (flow table/group). Refactor these operations to internal
helper functions for better block design.

This patch doesn't change any functionality.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: E-Switch, Properly refer to host PF vport as other vport
Bodong Wang [Fri, 1 Feb 2019 23:34:55 +0000 (17:34 -0600)]
net/mlx5: E-Switch, Properly refer to host PF vport as other vport

Commands referring to vports use the following scheme:

1. When referring to my own vport, put 0 in vport and 0 in other_vport.
2. When referring to another vport, put the vport number of the
   referred vport and put 1 in other_vport. It was assumed that driver
   is accessing other vport when vport number is greater than 0.

With the above scheme, the case that ECPF eswitch manager is trying
to access host PF vport will fall over with scheme 1 as the vport
number is 0. This is apparently wrong as driver is trying to refer
other vport.

As such usage can only happen in the eswitch context, change relevant
functions to provide other vport input properly.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: E-Switch, Properly refer to the esw manager vport
Bodong Wang [Thu, 8 Nov 2018 20:37:04 +0000 (22:37 +0200)]
net/mlx5: E-Switch, Properly refer to the esw manager vport

In SmartNIC mode, the eswitch manager is not necessarily the PF
(vport 0). Use a helper function to get the correct eswitch manager
vport number and cache on the eswitch instance for fast reference.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Signed-off-by: Eli Cohen <eli@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: Correctly set LAG mode for ECPF
Bodong Wang [Wed, 13 Feb 2019 16:52:34 +0000 (10:52 -0600)]
net/mlx5: Correctly set LAG mode for ECPF

When bonding is added, driver assumes that it's RoCE LAG if no VF is
enabled. This is not enough for ECPF as the VF is enabled in host PF
side. LAG should only choose RoCE mode when both slave devices meet
conditions below:
 1. E-Switch offloads mode is NONE.
 2. No VF is enabled.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agoMerge branch 'mlx5-next' of git://git.kernel.org/pub/scm/linux/kernel/git/mellanox...
Saeed Mahameed [Fri, 15 Feb 2019 23:16:36 +0000 (15:16 -0800)]
Merge branch 'mlx5-next' of git://git./linux/kernel/git/mellanox/linux

Merge mlx5-next shared branched into net-next,

From Bodong Wang:
1) Introduction of ECPF (Embedded CPU Physical Function), and low level
bits for mlx5 SmartNic capabilities support.
2) Vport enumeration refactoring that affect mlx5_ib and mlx5_core

From Aya Levin,
3) Add support for 50Gbps per lane link modes in the Port Type and Speed
register (PTYS)
4) Refactor low level query functions for PTYS register
5) Add support for 50Gbps per lane link modes to mlx5_ib

Note: due to a change in API in mlx5/core and a later patch from net-next,
a fixup was squashed with this merge commit that replaces FDB_UPLINK_VPORT
with MLX5_VPORT_UPLINK which exists only in upstream net-next.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agobpf: make LWTUNNEL_BPF dependent on INET
Peter Oskolkov [Fri, 15 Feb 2019 17:51:35 +0000 (09:51 -0800)]
bpf: make LWTUNNEL_BPF dependent on INET

Lightweight tunnels are L3 constructs that are used with IP/IP6.

For example, lwtunnel_xmit is called from ip_output.c and
ip6_output.c only.

Make the dependency explicit at least for LWT-BPF, as now they
call into IP routing.

V2: added "Reported-by" below.

Reported-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Peter Oskolkov <posk@google.com>
Acked-by: Randy Dunlap <rdunlap@infradead.org> # build-tested
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
David S. Miller [Fri, 15 Feb 2019 20:38:38 +0000 (12:38 -0800)]
Merge git://git./linux/kernel/git/davem/net

The netfilter conflicts were rather simple overlapping
changes.

However, the cls_tcindex.c stuff was a bit more complex.

On the 'net' side, Cong is fixing several races and memory
leaks.  Whilst on the 'net-next' side we have Vlad adding
the rtnl-ness support.

What I've decided to do, in order to resolve this, is revert the
conversion over to using a workqueue that Cong did, bringing us back
to pure RCU.  I did it this way because I believe that either Cong's
races don't apply with have Vlad did things, or Cong will have to
implement the race fix slightly differently.

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge tag 'for-linus-20190215' of git://git.kernel.dk/linux-block
Linus Torvalds [Fri, 15 Feb 2019 17:12:28 +0000 (09:12 -0800)]
Merge tag 'for-linus-20190215' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:

 - Ensure we insert into the hctx dispatch list, if a request is marked
   as DONTPREP (Jianchao)

 - NVMe pull request, single missing unlock on error fix (Keith)

 - MD pull request, single fix for a potentially data corrupting issue
   (Nate)

 - Floppy check_events regression fix (Yufen)

* tag 'for-linus-20190215' of git://git.kernel.dk/linux-block:
  md/raid1: don't clear bitmap bits on interrupted recovery.
  floppy: check_events callback should not return a negative number
  nvme-pci: add missing unlock for reset error
  blk-mq: insert rq with DONTPREP to hctx dispatch list when requeue

5 years agoMerge tag 'for-5.0/dm-fixes-3' of git://git.kernel.org/pub/scm/linux/kernel/git/devic...
Linus Torvalds [Fri, 15 Feb 2019 16:50:48 +0000 (08:50 -0800)]
Merge tag 'for-5.0/dm-fixes-3' of git://git./linux/kernel/git/device-mapper/linux-dm

Pull device mapper fixes from Mike Snitzer:

 - Fix bug in DM crypt's sizing of its block integrity tag space,
   resulting in less memory use when DM crypt layers on DM integrity.

 - Fix a long-standing DM thinp crash consistency bug that was due to
   improper handling of FUA. This issue is specific to writes that fill
   an entire thinp block which needs to be allocated.

* tag 'for-5.0/dm-fixes-3' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm thin: fix bug where bio that overwrites thin block ignores FUA
  dm crypt: don't overallocate the integrity tag space

5 years agoMerge tag 'mmc-v5.0-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc
Linus Torvalds [Fri, 15 Feb 2019 16:45:28 +0000 (08:45 -0800)]
Merge tag 'mmc-v5.0-rc5' of git://git./linux/kernel/git/ulfh/mmc

Pull MMC fixes from Ulf Hansson:
 "A couple of MMC fixes intended for v5.0-rc7.

  MMC core:
   - Fix deadlock bug for block I/O requests

  MMC host:
   - sunxi: Disable broken HS-DDR mode for H5 by default
   - sunxi: Avoid unsupported speed modes declared via DT
   - meson-gx: Restore interrupt name"

* tag 'mmc-v5.0-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc:
  mmc: meson-gx: fix interrupt name
  mmc: block: handle complete_work on separate workqueue
  mmc: sunxi: Filter out unsupported modes declared in the device tree
  mmc: sunxi: Disable HS-DDR mode for H5 eMMC controller by default

5 years agoMerge tag 'drm-fixes-2019-02-15-1' of git://anongit.freedesktop.org/drm/drm
Linus Torvalds [Fri, 15 Feb 2019 16:20:33 +0000 (08:20 -0800)]
Merge tag 'drm-fixes-2019-02-15-1' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "Usual pull request, little larger than I'd like but nothing too
  strange in it. Willy found an bug in the lease ioctl calculations, but
  it's a drm master only ioctl which makes it harder to mess with.

 i915:
   - combo phy programming fix
   - opregion version check fix for VBT RVDA lookup
   - gem mmap ioctl race fix
   - fbdev hpd during suspend fix
   - array size bounds check fix in pmu

  amdgpu:
   - Vega20 psp fix
   - Add vrr range to debugfs for freesync debugging

  sched:
   - Scheduler race fix

  vkms:
   - license header fixups

  imx:
   - Fix CSI register offsets for i.MX51 and i.MX53.
   - Fix delayed page flip completion events on i.MX6QP due to
     unexpected behaviour of the PRE when issuing NOP buffer updates to
     the same buffer address.
   - Stop throwing errors for plane updates on disabled CRTCs when a
     userspace process is killed while a plane update is pending.
   - Add missing of_node_put cleanup in imx_ldb_bind"

* tag 'drm-fixes-2019-02-15-1' of git://anongit.freedesktop.org/drm/drm:
  drm: Use array_size() when creating lease
  drm/amdgpu/psp11: TA firmware is optional (v3)
  drm/i915/opregion: rvda is relative from opregion base in opregion 2.1+
  drm/i915/opregion: fix version check
  drm/i915: Prevent a race during I915_GEM_MMAP ioctl with WC set
  drm/i915: Block fbdev HPD processing during suspend
  drm/i915/pmu: Fix enable count array size and bounds checking
  drm/i915/cnl: Fix CNL macros for Voltage Swing programming
  drm/i915/icl: combo port vswing programming changes per BSPEC
  drm/vkms: Fix license inconsistent
  drm/amd/display: Expose connector VRR range via debugfs
  drm/sched: Always trace the dependencies we wait on, to fix a race.
  gpu: ipu-v3: pre: don't trigger update if buffer address doesn't change
  gpu: ipu-v3: Fix CSI offsets for imx53
  drm/imx: imx-ldb: add missing of_node_puts
  gpu: ipu-v3: Fix i.MX51 CSI control registers offset
  drm/imx: ignore plane updates on disabled crtcs

5 years agoMerge branch 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Linus Torvalds [Fri, 15 Feb 2019 16:11:43 +0000 (08:11 -0800)]
Merge branch 'linus' of git://git./linux/kernel/git/herbert/crypto-2.6

Pull crypto fix from Herbert Xu:
 "This fixes a crash on resume in the ccree driver"

* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
  crypto: ccree - fix resume race condition on init

5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Fri, 15 Feb 2019 16:00:11 +0000 (08:00 -0800)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Fix MAC address setting in mac80211 pmsr code, from Johannes Berg.

 2) Probe SFP modules after being attached, from Russell King.

 3) Byte ordering bug in SMC rx_curs_confirmed code, from Ursula Braun.

 4) Revert some r8169 changes that are causing regressions, from Heiner
    Kallweit.

 5) Fix spurious connection timeouts in netfilter nat code, from Florian
    Westphal.

 6) SKB leak in tipc, from Hoang Le.

 7) Short packet checkum issue in mlx4, similar to a previous mlx5
    change, from Saeed Mahameed. The issue is that whilst padding bytes
    are usually zero, it is not guarateed and the hardware doesn't take
    the padding bytes into consideration when generating the checksum.

 8) Fix various races in cls_tcindex, from Cong Wang.

 9) Need to set stream ext to NULL before freeing in SCTP code, from Xin
    Long.

10) Fix locking in phy_is_started, from Heiner Kallweit.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (54 commits)
  net: ethernet: freescale: set FEC ethtool regs version
  net: hns: Fix object reference leaks in hns_dsaf_roce_reset()
  mm: page_alloc: fix ref bias in page_frag_alloc() for 1-byte allocs
  net: phy: fix potential race in the phylib state machine
  net: phy: don't use locking in phy_is_started
  selftests: fix timestamping Makefile
  net: dsa: bcm_sf2: potential array overflow in bcm_sf2_sw_suspend()
  net: fix possible overflow in __sk_mem_raise_allocated()
  dsa: mv88e6xxx: Ensure all pending interrupts are handled prior to exit
  net: phy: fix interrupt handling in non-started states
  sctp: set stream ext to NULL after freeing it in sctp_stream_outq_migrate
  sctp: call gso_reset_checksum when computing checksum in sctp_gso_segment
  net/mlx5e: XDP, fix redirect resources availability check
  net/mlx5: Fix a compilation warning in events.c
  net/mlx5: No command allowed when command interface is not ready
  net/mlx5e: Fix NULL pointer derefernce in set channels error flow
  netfilter: nft_compat: use-after-free when deleting targets
  team: avoid complex list operations in team_nl_cmd_options_set()
  net_sched: fix two more memory leaks in cls_tcindex
  net_sched: fix a memory leak in cls_tcindex
  ...

5 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm...
Linus Torvalds [Fri, 15 Feb 2019 15:56:24 +0000 (07:56 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/ebiederm/user-namespace

Pull signal fix from Eric Biederman:
 "Just a single patch that restores PTRACE_EVENT_EXIT functionality that
  was accidentally broken by last weeks fixes"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace:
  signal: Restore the stop PTRACE_EVENT_EXIT

5 years agolibbpf: Introduce bpf_object__btf
Andrey Ignatov [Thu, 14 Feb 2019 23:01:43 +0000 (15:01 -0800)]
libbpf: Introduce bpf_object__btf

Add new accessor for bpf_object to get opaque struct btf * from it.

struct btf * is needed for all operations with BTF and it's present in
bpf_object. The only thing missing is a way to get it.

Example use-case is to get BTF key_type_id and value_type_id for a map in
bpf_object. It can be done with btf__get_map_kv_tids() but that function
requires struct btf *.

Similar API can be added for struct btf_ext but no use-case for it yet.

Signed-off-by: Andrey Ignatov <rdna@fb.com>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agolibbpf: Introduce bpf_map__resize
Andrey Ignatov [Thu, 14 Feb 2019 23:01:42 +0000 (15:01 -0800)]
libbpf: Introduce bpf_map__resize

Add bpf_map__resize() to change max_entries for a map.

Quite often necessary map size is unknown at compile time and can be
calculated only at run time.

Currently the following approach is used to do so:
* bpf_object__open_buffer() to open Elf file from a buffer;
* bpf_object__find_map_by_name() to find relevant map;
* bpf_map__def() to get map attributes and create struct
  bpf_create_map_attr from them;
* update max_entries in bpf_create_map_attr;
* bpf_create_map_xattr() to create new map with updated max_entries;
* bpf_map__reuse_fd() to replace the map in bpf_object with newly
  created one.

And after all this bpf_object can finally be loaded. The map will have
new size.

It 1) is quite a lot of steps; 2) doesn't take BTF into account.

For "2)" even more steps should be made and some of them require changes
to libbpf (e.g. to get struct btf * from bpf_object).

Instead the whole problem can be solved by introducing simple
bpf_map__resize() API that checks the map and sets new max_entries if
the map is not loaded yet.

So the new steps are:
* bpf_object__open_buffer() to open Elf file from a buffer;
* bpf_object__find_map_by_name() to find relevant map;
* bpf_map__resize() to update max_entries.

That's much simpler and works with BTF.

Signed-off-by: Andrey Ignatov <rdna@fb.com>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agonet: bpf: remove XDP_QUERY_XSK_UMEM enumerator
Jan Sokolowski [Wed, 13 Feb 2019 17:07:29 +0000 (18:07 +0100)]
net: bpf: remove XDP_QUERY_XSK_UMEM enumerator

Commit c9b47cc1fabc ("xsk: fix bug when trying to use both copy and
zero-copy on one queue id") moved the umem query code to the AF_XDP
core, and therefore removed the need to query the netdevice for a
umem.

This patch removes XDP_QUERY_XSK_UMEM and all code that implement that
behavior, which is just dead code.

Signed-off-by: Jan Sokolowski <jan.sokolowski@intel.com>
Acked-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agodrm: Use array_size() when creating lease
Matthew Wilcox [Thu, 14 Feb 2019 19:03:48 +0000 (11:03 -0800)]
drm: Use array_size() when creating lease

Passing an object_count of sufficient size will make
object_count * 4 wrap around to be very small, then a later function
will happily iterate off the end of the object_ids array.  Using
array_size() will saturate at SIZE_MAX, the kmalloc() will fail and
we'll return an -ENOMEM to the norty userspace.

Fixes: 62884cd386b8 ("drm: Add four ioctls for managing drm mode object leases [v7]")
Signed-off-by: Matthew Wilcox <willy@infradead.org>
Acked-by: Kees Cook <keescook@chromium.org>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: <stable@vger.kernel.org> # v4.15+
Signed-off-by: Dave Airlie <airlied@redhat.com>
5 years agoMerge branch 'drm-fixes-5.0' of git://people.freedesktop.org/~agd5f/linux into drm...
Dave Airlie [Fri, 15 Feb 2019 01:46:40 +0000 (11:46 +1000)]
Merge branch 'drm-fixes-5.0' of git://people.freedesktop.org/~agd5f/linux into drm-fixes

amdgpu:
- Vega20 psp fix
- Add vrr range to debugfs for freesync debugging

sched:
- Scheduler race fix

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexdeucher@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190213202958.3336-1-alexander.deucher@amd.com
5 years agoMerge tag 'drm-intel-fixes-2019-02-13' of git://anongit.freedesktop.org/drm/drm-intel...
Dave Airlie [Fri, 15 Feb 2019 01:22:37 +0000 (11:22 +1000)]
Merge tag 'drm-intel-fixes-2019-02-13' of git://anongit.freedesktop.org/drm/drm-intel into drm-fixes

drm/i915 fixes for v5.0-rc7:
- combo phy programming fix
- opregion version check fix for VBT RVDA lookup
- gem mmap ioctl race fix
- fbdev hpd during suspend fix
- array size bounds check fix in pmu

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/877ee3504b.fsf@intel.com
5 years agoMerge tag 'drm-misc-fixes-2019-02-13' of git://anongit.freedesktop.org/drm/drm-misc...
Dave Airlie [Fri, 15 Feb 2019 01:21:46 +0000 (11:21 +1000)]
Merge tag 'drm-misc-fixes-2019-02-13' of git://anongit.freedesktop.org/drm/drm-misc into drm-fixes

drm-misc-fixes for v5.0:
- Fix license inconsistency in vkms.

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/812e2f53-d72a-8fba-6c8c-fde8f44cf141@linux.intel.com
5 years agodm thin: fix bug where bio that overwrites thin block ignores FUA
Nikos Tsironis [Thu, 14 Feb 2019 18:38:47 +0000 (20:38 +0200)]
dm thin: fix bug where bio that overwrites thin block ignores FUA

When provisioning a new data block for a virtual block, either because
the block was previously unallocated or because we are breaking sharing,
if the whole block of data is being overwritten the bio that triggered
the provisioning is issued immediately, skipping copying or zeroing of
the data block.

When this bio completes the new mapping is inserted in to the pool's
metadata by process_prepared_mapping(), where the bio completion is
signaled to the upper layers.

This completion is signaled without first committing the metadata.  If
the bio in question has the REQ_FUA flag set and the system crashes
right after its completion and before the next metadata commit, then the
write is lost despite the REQ_FUA flag requiring that I/O completion for
this request must only be signaled after the data has been committed to
non-volatile storage.

Fix this by deferring the completion of overwrite bios, with the REQ_FUA
flag set, until after the metadata has been committed.

Cc: stable@vger.kernel.org
Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com>
Acked-by: Joe Thornber <ejt@redhat.com>
Acked-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
5 years agoMerge branch 'libbpf-cleanup'
Alexei Starovoitov [Thu, 14 Feb 2019 23:31:39 +0000 (15:31 -0800)]
Merge branch 'libbpf-cleanup'

Andrii Nakryiko says:

====================
This patchset replaces bzero() with memset() and syncs if_link.h header
to suppress unsynchronized headers warning.
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agotools: sync uapi/linux/if_link.h header
Andrii Nakryiko [Wed, 13 Feb 2019 18:25:54 +0000 (10:25 -0800)]
tools: sync uapi/linux/if_link.h header

Syncing if_link.h that got out of sync.

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agotools/bpf: replace bzero with memset
Andrii Nakryiko [Wed, 13 Feb 2019 18:25:53 +0000 (10:25 -0800)]
tools/bpf: replace bzero with memset

bzero() call is deprecated and superseded by memset().

Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Reported-by: David Laight <david.laight@aculab.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agobpf: fix memory leak in bpf_lwt_xmit_reroute
Peter Oskolkov [Thu, 14 Feb 2019 18:39:31 +0000 (10:39 -0800)]
bpf: fix memory leak in bpf_lwt_xmit_reroute

On error the skb should be freed. Tested with diff/steps
provided by David Ahern.

v2: surface routing errors to the user instead of a generic EINVAL,
    as suggested by David Ahern.

Reported-by: David Ahern <dsahern@gmail.com>
Fixes: 3bd0b15281af ("bpf: add handling of BPF_LWT_REROUTE to lwt_bpf.c")
Signed-off-by: Peter Oskolkov <posk@google.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agoRevert "exec: load_script: don't blindly truncate shebang string"
Linus Torvalds [Thu, 14 Feb 2019 23:02:18 +0000 (15:02 -0800)]
Revert "exec: load_script: don't blindly truncate shebang string"

This reverts commit 8099b047ecc431518b9bb6bdbba3549bbecdc343.

It turns out that people do actually depend on the shebang string being
truncated, and on the fact that an interpreter (like perl) will often
just re-interpret it entirely to get the full argument list.

Reported-by: Samuel Dionne-Riel <samuel@dionne-riel.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agoIB/mlx5: Add support for 50Gbps per lane link modes
Aya Levin [Wed, 13 Feb 2019 06:55:46 +0000 (22:55 -0800)]
IB/mlx5: Add support for 50Gbps per lane link modes

Driver now supports new link modes: 50Gbps per lane support for
50G/100G/200G. This patch reads the correct field (legacy vs. extended)
based on a FW indication bit, and adds a translation function (link
modes to IB width and speed) to the new link modes.

Signed-off-by: Aya Levin <ayal@mellanox.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: Add support to ext_* fields introduced in Port Type and Speed register
Aya Levin [Wed, 13 Feb 2019 06:55:45 +0000 (22:55 -0800)]
net/mlx5: Add support to ext_* fields introduced in Port Type and Speed register

This patch exposes new link modes (including 50Gbps per lane), and ext_*
fields which describes the new link modes in Port Type and Speed
register (PTYS).
Access functions, translation functions (speed <-> HW bits) and
link max speed function were modified.

Signed-off-by: Aya Levin <ayal@mellanox.com>
Reviewed-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: Add new fields to Port Type and Speed register
Aya Levin [Wed, 13 Feb 2019 06:55:44 +0000 (22:55 -0800)]
net/mlx5: Add new fields to Port Type and Speed register

Register Port Type and Speed (PTYS) introduces three new fields
extending the speed/protocols the can be reported and configured.

Signed-off-by: Aya Levin <ayal@mellanox.com>
Reviewed-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: Refactor queries to speed fields in Port Type and Speed register
Aya Levin [Wed, 13 Feb 2019 06:55:43 +0000 (22:55 -0800)]
net/mlx5: Refactor queries to speed fields in Port Type and Speed register

This patch fascicles queries to speed related fields in Port Type and
Speed register (PTYS) into a single API. I addition, this patch
refactors functions which serves only Ethernet driver: remove the
protocol type as an input parameter, move code from 'core' directory
into 'en' directory and add 'eth' prefix to the function's name. The
patch also encapsulates functions that are not used outside the Ethernet
driver removes redundant include files.

Signed-off-by: Aya Levin <ayal@mellanox.com>
Reviewed-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: E-Switch, Avoid magic numbers when initializing offloads mode
Bodong Wang [Wed, 13 Feb 2019 06:55:42 +0000 (22:55 -0800)]
net/mlx5: E-Switch, Avoid magic numbers when initializing offloads mode

When dealing with the offloads mode initialization, driver refers to
the number of VFs and add magic number one (1) to take account of the
uplink. This is not clear and will make the code less readable after
adding other vports (e.g. host PF). As these are special vports
compared to VF vports, add a helper macro to denote such special
vports and eliminate the use of magic number.

Moreover, when creating offloads flow table and groups, the driver
reserves two more slots for UC and MC miss rules. Replace this magic
number with a helper macro as well.

This patch doesn't change any functionality.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: Relocate vport macros to the vport header file
Bodong Wang [Wed, 13 Feb 2019 06:55:41 +0000 (22:55 -0800)]
net/mlx5: Relocate vport macros to the vport header file

These are two macros in the driver general header which deal with the
number of total vports and if a vport is vport manager. Such macros
are vport entities, better to place them at the vport header file.

This patch doesn't change any functionality.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: E-Switch, Normalize the name of uplink vport number
Bodong Wang [Wed, 13 Feb 2019 06:55:40 +0000 (22:55 -0800)]
net/mlx5: E-Switch, Normalize the name of uplink vport number

Driver used to name uplink vport as FDB_UPLINK_VPORT, it's hard to
comply with the same naming convention along with the introduction of
other vports. Use MLX5_VPORT as the prefix for such vports and
relocate the uplink vport definition to public header file for the
benefits of both net and IB drivers.

This patch doesn't change any functionality.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: Provide an alternative VF upper bound for ECPF
Bodong Wang [Wed, 13 Feb 2019 06:55:39 +0000 (22:55 -0800)]
net/mlx5: Provide an alternative VF upper bound for ECPF

ECPF doesn't support SR-IOV, but an ECPF E-Switch manager shall know
the max VFs supported by its peer host PF in order to control those
VF vports.

The current driver implementation uses the total vfs quantity as
provided by the pci sub-system for an upper bound of the VF vports
the e-switch code needs to deal with. This obviously can't work as
is on ECPF e-switch manager. For now, we use a hard coded value of
128 on such systems.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: Add host params change event
Bodong Wang [Wed, 13 Feb 2019 06:55:38 +0000 (22:55 -0800)]
net/mlx5: Add host params change event

In Embedded CPU (EC) configurations, the EC driver needs to know when
the number of virtual functions change on the corresponding PF at the
host side. This is required so the EC driver can create or destroy
representor net devices that represent the VFs ports.

Whenever a change in the number of VFs occurs, firmware will generate an
event towards the EC which will trigger a work to complete the rest of
the handling. The specifics of the handling will be introduced in a
downstream patch.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Signed-off-by: Eli Cohen <eli@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: Add query host params command
Bodong Wang [Wed, 13 Feb 2019 06:55:37 +0000 (22:55 -0800)]
net/mlx5: Add query host params command

The QUERY_HOST_PARAMS command is used by an Embedded CPU Physical
Function (ECPF) driver to identify and retrieve information about the
PF on the host side. E.g, number of virtual functions and PCI BDF.

The number of VFs can be changed on the fly, a function is added to
query current number of VFs and will be used in downstream patches.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Signed-off-by: Eli Cohen <eli@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: Update enable HCA dependency
Bodong Wang [Wed, 13 Feb 2019 06:55:36 +0000 (22:55 -0800)]
net/mlx5: Update enable HCA dependency

With the introduction of ECPF, we require that the ECPF driver will
aways call enable/disable HCA for that PF in the same way a PF does
this for its VFs. The PF is still responsible for calling enable and
disable HCA for its VFs.

To distinguish between the ECPF executing enable/disable HCA for
itself or for the PF, it sets the embedded CPU function bit in the
input params struct of these commands. When the bit is cleared and
function ID is zero, it refers to the peer PF.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: Introduce Mellanox SmartNIC and modify page management logic
Bodong Wang [Wed, 13 Feb 2019 06:55:35 +0000 (22:55 -0800)]
net/mlx5: Introduce Mellanox SmartNIC and modify page management logic

Mellanox's SmartNIC combines embedded CPU(e.g, ARM) processing power
with advanced network offloads to accelerate a multitude of security,
networking and storage applications.

With the introduction of the SmartNIC, there is a new PCI function
called Embedded CPU Physical Function(ECPF). And it's possible for a
PF to get its ICM pages from the ECPF PCI function. Driver shall
identify if it is running on such a function by reading a bit in
the initialization segment.

When firmware asks for pages, it would issue a page request event
specifying how many pages it requests and for which function. That
driver responds with a manage_pages command providing the requested
pages along with an indication for which function it is providing these
pages.

The encoding before this patch was as follows:
    function_id == 0: pages are requested for the function receiving
                      the EQE.
    function_id != 0: pages are requested for VF identified by the
                      function_id value

A new one bit field in the EQE identifies that pages are requested for
the ECPF.

The notion of page_supplier can be introduced here and to support that,
manage pages and query pages were modified so firmware can distinguish
the following cases:

1. Function provides pages for itself
2. PF provides pages for its VF
3. ECPF provides pages to itself
4. ECPF provides pages for another function

This distinction is possible through the introduction of the bit
"embedded_cpu_function" in query_pages, manage_pages and page request
EQE.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Signed-off-by: Eli Cohen <eli@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agoIB/mlx5: Use unified register/load function for uplink and VF vports
Bodong Wang [Wed, 13 Feb 2019 06:55:34 +0000 (22:55 -0800)]
IB/mlx5: Use unified register/load function for uplink and VF vports

IB driver maintains different registration and load function calls
for uplink and VF vports. This is not necessary as they only differ
with each other on their profiles.

This patch doesn't change any functionality.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: Use consistent vport num argument type
Bodong Wang [Wed, 13 Feb 2019 06:55:33 +0000 (22:55 -0800)]
net/mlx5: Use consistent vport num argument type

Use u16 for vport number, which matches how hardware refers to this
argument throughout commands.

This patch doesn't change any functionality.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Signed-off-by: Eli Cohen <eli@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5: Use void pointer as the type in address_of macro
Bodong Wang [Wed, 13 Feb 2019 06:55:32 +0000 (22:55 -0800)]
net/mlx5: Use void pointer as the type in address_of macro

Better to use void * and avoid unnecessary casts.

This patch doesn't change any functionality.

Signed-off-by: Bodong Wang <bodong@mellanox.com>
Signed-off-by: Eli Cohen <eli@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Daniel Jurgens <danielj@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agoRevert "gfs2: read journal in large chunks to locate the head"
Bob Peterson [Wed, 13 Feb 2019 20:12:17 +0000 (15:12 -0500)]
Revert "gfs2: read journal in large chunks to locate the head"

This reverts commit 2a5f14f279f59143139bcd1606903f2f80a34241.

This patch causes xfstests generic/311 to fail. Reverting this for
now until we have a proper fix.

Signed-off-by: Abhi Das <adas@redhat.com>
Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agosfc: ensure recovery after allocation failures
Robert Stonehouse [Thu, 14 Feb 2019 17:27:43 +0000 (17:27 +0000)]
sfc: ensure recovery after allocation failures

After failing to allocate a receive buffer the driver may fail to ever
request additional allocations. EF10 NICs require new receive buffers to
be pushed in batches of eight or more. The test for whether a slow fill
should be scheduled failed to take account of this. There is little
downside to *always* requesting a slow fill if we failed to allocate a
buffer, so the condition has been removed completely. The timer that
triggers the request for a refill has also been shortened.

Signed-off-by: Robert Stonehouse <rstonehouse@solarflare.com>
Signed-off-by: Bert Kenward <bkenward@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ethernet: freescale: set FEC ethtool regs version
Vivien Didelot [Thu, 14 Feb 2019 16:15:35 +0000 (11:15 -0500)]
net: ethernet: freescale: set FEC ethtool regs version

Currently the ethtool_regs version is set to 0 for FEC devices.

Use this field to store the register dump version exposed by the
kernel. The choosen version 2 corresponds to the kernel compile test:

        #if defined(CONFIG_M523x) || defined(CONFIG_M527x)
        || defined(CONFIG_M528x) || defined(CONFIG_M520x)
        || defined(CONFIG_M532x) || defined(CONFIG_ARM)
        || defined(CONFIG_ARM64) || defined(CONFIG_COMPILE_TEST)

and version 1 corresponds to the opposite. Binaries of ethtool unaware
of this version will dump the whole set as usual.

Signed-off-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: adaptec: starfire: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop...
Yang Wei [Thu, 14 Feb 2019 15:06:40 +0000 (23:06 +0800)]
net: adaptec: starfire: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles

dev_consume_skb_irq() should be called in intr_handler() when skb
xmit done. It makes drop profiles(dropwatch, perf) more friendly.

Signed-off-by: Yang Wei <yang.wei9@zte.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: 3com: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
Yang Wei [Thu, 14 Feb 2019 14:55:14 +0000 (22:55 +0800)]
net: 3com: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles

dev_consume_skb_irq() should be called when skb xmit done. It makes
drop profiles(dropwatch, perf) more friendly.

Signed-off-by: Yang Wei <yang.wei9@zte.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: arc_emac: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
Yang Wei [Thu, 14 Feb 2019 14:53:30 +0000 (22:53 +0800)]
net: arc_emac: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles

dev_consume_skb_irq() should be called in arc_emac_tx_clean() when
skb xmit done. It makes drop profiles(dropwatch, perf) more friendly.

Signed-off-by: Yang Wei <yang.wei9@zte.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: packetengines: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
Yang Wei [Thu, 14 Feb 2019 14:52:28 +0000 (22:52 +0800)]
net: packetengines: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles

dev_consume_skb_irq() should be called when skb xmit done. It makes
drop profiles(dropwatch, perf) more friendly.

Signed-off-by: Yang Wei <yang.wei9@zte.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: xilinx: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
Yang Wei [Thu, 14 Feb 2019 14:50:58 +0000 (22:50 +0800)]
net: xilinx: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles

dev_consume_skb_irq() should be called when skb xmit done. It makes
drop profiles(dropwatch, perf) more friendly.

Signed-off-by: Yang Wei <yang.wei9@zte.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: i825xx: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
Yang Wei [Thu, 14 Feb 2019 14:45:38 +0000 (22:45 +0800)]
net: i825xx: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles

dev_consume_skb_irq() should be called in i596_interrupt() when skb
xmit done. It makes drop profiles(dropwatch, perf) more friendly.

Signed-off-by: Yang Wei <yang.wei9@zte.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agolib: objagg: fix handling of object with 0 users when assembling hints
Jiri Pirko [Thu, 14 Feb 2019 14:39:07 +0000 (15:39 +0100)]
lib: objagg: fix handling of object with 0 users when assembling hints

It is possible that there might be an originally parent object with 0
direct users that is in hints no longer considered as parent. Then the
weight of this object is 0 and current code ignores him. That's why the
total amount of hint objects might be lower than for the original
objagg and WARN_ON is hit. Fix this be considering 0 weight valid.

Fixes: 9069a3817d82 ("lib: objagg: implement optimization hints assembly and use hints for object creation")
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'cxgb4-SGE-doorbell-queue-timer'
David S. Miller [Thu, 14 Feb 2019 17:39:35 +0000 (12:39 -0500)]
Merge branch 'cxgb4-SGE-doorbell-queue-timer'

Vishal Kulkarni says:

====================
cxgb4/cxgb4vfSupport for SGE doorbell queue timer

This series of patchs add SGE doorbell queue timer for faster DMA completions.

Patch 1 Implements SGE doorbell queue timer

Patch 2 Adds ethtool capability to set/get SGE doorbell queue timer tick

v2
- Reverse christmas tree formatting for local variables.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agocxgb4: Add capability to get/set SGE Doorbell Queue Timer Tick
Vishal Kulkarni [Thu, 14 Feb 2019 12:49:16 +0000 (18:19 +0530)]
cxgb4: Add capability to get/set SGE Doorbell Queue Timer Tick

This patch gets/sets SGE Doorbell Queue timer ticks via ethtool

Original work by: Casey Leedom <leedom@chelsio.com>

Signed-off-by: Vishal Kulkarni <vishal@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agocxgb4/cxgb4vf: Add support for SGE doorbell queue timer
Vishal Kulkarni [Thu, 14 Feb 2019 12:49:15 +0000 (18:19 +0530)]
cxgb4/cxgb4vf: Add support for SGE doorbell queue timer

T6 introduced a Timer Mechanism in SGE called the
SGE Doorbell Queue Timer. With this we can now configure
TX Queues to get CIDX Updates when:

    Time(CIDX == PIDX) >= Timer

Previously we rely on TX Queue Status Page updates by hardware
for DMA completions. This will make Hardware/Firmware actually
deliver the CIDX Updates as Ingress Queue messages with
commensurate Interrupts.

So we now have a new RX Path component for processing CIDX Updates
and reclaiming TX Descriptors faster.

Original work by: Casey Leedom <leedom@chelsio.com>

Signed-off-by: Vishal Kulkarni <vishal@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns: Fix object reference leaks in hns_dsaf_roce_reset()
Huang Zijiang [Thu, 14 Feb 2019 06:41:45 +0000 (14:41 +0800)]
net: hns: Fix object reference leaks in hns_dsaf_roce_reset()

The of_find_device_by_node() takes a reference to the underlying device
structure, we should release that reference.

Signed-off-by: Huang Zijiang <huang.zijiang@zte.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosfc: Replace dev_kfree_skb_any by dev_consume_skb_any
Huang Zijiang [Thu, 14 Feb 2019 06:42:13 +0000 (14:42 +0800)]
sfc: Replace dev_kfree_skb_any by dev_consume_skb_any

The skb should be freed by dev_consume_skb_any() in efx_tx_tso_fallback()
when skb is still used. The skb will be replaced by segments, so the
original skb should be consumed(not drop).

Signed-off-by: Huang Zijiang <huang.zijiang@zte.com.cn>
Acked-by: Bert Kenward <bkenward@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet:ethernet:cadence: Replace dev_kfree_skb_any by dev_consume_skb_any
Huang Zijiang [Thu, 14 Feb 2019 06:41:18 +0000 (14:41 +0800)]
net:ethernet:cadence: Replace dev_kfree_skb_any by dev_consume_skb_any

The skb should be freed by dev_consume_skb_any() in macb_pad_and_fcs()
when *skb is still used. The *skb is be replaced by nskb, so the
original *skb should be consumed(not drop).

Signed-off-by: Huang Zijiang <huang.zijiang@zte.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet:dl2k: Replace dev_kfree_skb_irq by dev_consume_skb_irq
Huang Zijiang [Thu, 14 Feb 2019 06:40:56 +0000 (14:40 +0800)]
net:dl2k: Replace dev_kfree_skb_irq by dev_consume_skb_irq

dev_consume_skb_irq() should be called when skb xmit
done.It makes drop profiles more friendly.

Signed-off-by: Huang Zijiang <huang.zijiang@zte.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet:dl2k: Modify the code style escaping the warning
Huang Zijiang [Thu, 14 Feb 2019 06:40:31 +0000 (14:40 +0800)]
net:dl2k: Modify the code style escaping the warning

modify the code style in order to removing the following warning
when excute the script checkpatch.pl
WARNING: space prohibited between function name and open parenthesis '('

Signed-off-by: Huang Zijiang <huang.zijiang@zte.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoisdn:hisax: Replace dev_kfree_skb_any by dev_consume_skb_any
Huang Zijiang [Thu, 14 Feb 2019 06:39:59 +0000 (14:39 +0800)]
isdn:hisax: Replace dev_kfree_skb_any by dev_consume_skb_any

The skb should be freed by dev_consume_skb_any() in hfcpci_fill_fifo()
when bcs->tx_skb is still used. The bcs->tx_skb is be replaced by
skb_dequeue(&bcs->squeue), so the original bcs->tx_skb should
be consumed(not drop).

Signed-off-by: Huang Zijiang <huang.zijiang@zte.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomm: page_alloc: fix ref bias in page_frag_alloc() for 1-byte allocs
Jann Horn [Wed, 13 Feb 2019 21:45:59 +0000 (22:45 +0100)]
mm: page_alloc: fix ref bias in page_frag_alloc() for 1-byte allocs

The basic idea behind ->pagecnt_bias is: If we pre-allocate the maximum
number of references that we might need to create in the fastpath later,
the bump-allocation fastpath only has to modify the non-atomic bias value
that tracks the number of extra references we hold instead of the atomic
refcount. The maximum number of allocations we can serve (under the
assumption that no allocation is made with size 0) is nc->size, so that's
the bias used.

However, even when all memory in the allocation has been given away, a
reference to the page is still held; and in the `offset < 0` slowpath, the
page may be reused if everyone else has dropped their references.
This means that the necessary number of references is actually
`nc->size+1`.

Luckily, from a quick grep, it looks like the only path that can call
page_frag_alloc(fragsz=1) is TAP with the IFF_NAPI_FRAGS flag, which
requires CAP_NET_ADMIN in the init namespace and is only intended to be
used for kernel testing and fuzzing.

To test for this issue, put a `WARN_ON(page_ref_count(page) == 0)` in the
`offset < 0` path, below the virt_to_page() call, and then repeatedly call
writev() on a TAP device with IFF_TAP|IFF_NO_PI|IFF_NAPI_FRAGS|IFF_NAPI,
with a vector consisting of 15 elements containing 1 byte each.

Signed-off-by: Jann Horn <jannh@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-phy-fix-locking-issue'
David S. Miller [Thu, 14 Feb 2019 17:04:55 +0000 (12:04 -0500)]
Merge branch 'net-phy-fix-locking-issue'

Heiner Kallweit says:

====================
net: phy: fix locking issue

Russell pointed out that the locking used in phy_is_started() isn't
needed and misleading. This locking also contributes to a race fixed
with patch 2.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: fix potential race in the phylib state machine
Heiner Kallweit [Wed, 13 Feb 2019 19:12:54 +0000 (20:12 +0100)]
net: phy: fix potential race in the phylib state machine

Russell reported the following race in the phylib state machine
(quoting from his mail):

if (phy_polling_mode(phydev) && phy_is_started(phydev))
phy_queue_state_machine(phydev, PHY_STATE_TIME);

state = PHY_UP
thread 0 thread 1
phy_disconnect()
+-phy_is_started()
phy_is_started()                |
`-phy_stop()
  +-phydev->state = PHY_HALTED
  `-phy_stop_machine()
    `-cancel_delayed_work_sync()
phy_queue_state_machine()
`-mod_delayed_work()

At this point, the phydev->state_queue() has been added back onto the
system workqueue despite phy_stop_machine() having been called and
cancel_delayed_work_sync() called on it.

Fix this by protecting the complete operation in thread 0.

Fixes: 2b3e88ea6528 ("net: phy: improve phy state checking")
Reported-by: Russell King - ARM Linux admin <linux@armlinux.org.uk>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: don't use locking in phy_is_started
Heiner Kallweit [Wed, 13 Feb 2019 19:11:40 +0000 (20:11 +0100)]
net: phy: don't use locking in phy_is_started

Russell suggested to remove the locking from phy_is_started() because
the read is atomic anyway and actually the locking may be more
misleading.

Fixes: 2b3e88ea6528 ("net: phy: improve phy state checking")
Suggested-by: Russell King - ARM Linux admin <linux@armlinux.org.uk>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoselftests: fix timestamping Makefile
Deepa Dinamani [Wed, 13 Feb 2019 17:09:13 +0000 (09:09 -0800)]
selftests: fix timestamping Makefile

The clean target in the makefile conflicts with the generic
kselftests lib.mk, and fails to properly remove the compiled
test programs.

Remove the redundant rule, the TEST_GEN_FILES will be already
removed by the CLEAN macro in lib.mk.

Signed-off-by: Deepa Dinamani <deepa.kernel@gmail.com>
Acked-by: Shuah Khan <shuah@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ipvlan_l3s: fix kconfig dependency warning
Randy Dunlap [Wed, 13 Feb 2019 16:55:02 +0000 (08:55 -0800)]
net: ipvlan_l3s: fix kconfig dependency warning

Fix the kconfig warning in IPVLAN_L3S when neither INET nor IPV6
is enabled:

WARNING: unmet direct dependencies detected for NET_L3_MASTER_DEV
  Depends on [n]: NET [=y] && (INET [=n] || IPV6 [=n])
  Selected by [y]:
  - IPVLAN_L3S [=y] && NETDEVICES [=y] && NET_CORE [=y] && NETFILTER [=y]

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Mahesh Bandewar <maheshb@google.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: nuvoton: w90p910_ether: replace dev_kfree_skb_irq by dev_consume_skb_irq for...
Yang Wei [Wed, 13 Feb 2019 15:21:02 +0000 (23:21 +0800)]
net: nuvoton: w90p910_ether: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles

dev_consume_skb_irq() should be called in w90p910_ether_start_xmit()
when skb xmit done. It makes drop profiles(dropwatch, perf) more
friendly.

Signed-off-by: Yang Wei <yang.wei9@zte.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: natsemi: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
Yang Wei [Wed, 13 Feb 2019 15:19:14 +0000 (23:19 +0800)]
net: natsemi: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles

dev_consume_skb_irq() should be called when skb xmit done. It makes
drop profiles(dropwatch, perf) more friendly.

Signed-off-by: Yang Wei <yang.wei9@zte.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: micrel: ks8695net: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop...
Yang Wei [Wed, 13 Feb 2019 15:18:09 +0000 (23:18 +0800)]
net: micrel: ks8695net: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles

dev_consume_skb_irq() should be called in ks8695_tx_irq() when skb
xmit done. It makes drop profiles(dropwatch, perf) more friendly.

Signed-off-by: Yang Wei <yang.wei9@zte.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: sgi: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
Yang Wei [Wed, 13 Feb 2019 15:17:06 +0000 (23:17 +0800)]
net: sgi: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles

dev_consume_skb_irq() should be called when skb xmit done. It makes
drop profiles(dropwatch, perf) more friendly.

Signed-off-by: Yang Wei <yang.wei9@zte.com.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>