After 610d2b601b ("rocker: Remove getting PORT_BRIDGE_FLAGS") we no
longer have a port_attr_bridge_flags_get member in the rocker_world_ops
structre, fix that.
Fixes: 610d2b601b ("rocker: Remove getting PORT_BRIDGE_FLAGS")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix race condition between ena_update_on_link_change() and
ena_restore_device().
This race can occur if link notification arrives while the driver
is performing a reset sequence. In this case link can be set up,
enabling the device, before it is fully restored. If packets are
sent at this time, the driver might access uninitialized data
structures, causing kernel crash.
Move the clearing of ENA_FLAG_ONGOING_RESET and netif_carrier_on()
after ena_up() to ensure the device is ready when link is set up.
Fixes: d18e4f6834 ("net: ena: fix race condition between device reset and link up setup")
Signed-off-by: Arthur Kiyanovski <akiyano@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When an ethernet frame is padded to meet the minimum ethernet frame
size, the padding octets are not covered by the hardware checksum.
Fortunately the padding octets are usually zero's, which don't affect
checksum. However, it is not guaranteed. For example, switches might
choose to make other use of these octets.
This repeatedly causes kernel hardware checksum fault.
Prior to the cited commit below, skb checksum was forced to be
CHECKSUM_NONE when padding is detected. After it, we need to keep
skb->csum updated. However, fixing up CHECKSUM_COMPLETE requires to
verify and parse IP headers, it does not worth the effort as the packets
are so small that CHECKSUM_COMPLETE has no significant advantage.
Future work: when reporting checksum complete is not an option for
IP non-TCP/UDP packets, we can actually fallback to report checksum
unnecessary, by looking at cqe IPOK bit.
Fixes: 88078d98d1 ("net: pskb_trim_rcsum() and CHECKSUM_COMPLETE are friends")
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch is to add PTP clock driver for ENETC.
The driver reused QorIQ PTP clock driver.
Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Strings containing "ptp_qoriq" or "qoriq_ptp" which were used for
structure/function names were complained by users. Let's just use
the unique "ptp_qoriq" to make these names more consistent.
This patch is just to unify the names using "ptp_qoriq". It hasn't
changed any functions.
Signed-off-by: Yangbo Lu <yangbo.lu@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is no code that attempts to get the
SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS attribute, remove support for that.
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is no code that will query the SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS
attribute remove support for that.
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There are several places which make the decision whether to access the
XLGMAC vs GMAC that only check for PHY_INTERFACE_MODE_10GKR and not its
XAUI variant. Switch these to use the new helper so that we have
consistency through the driver.
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When an ethernet frame is padded to meet the minimum ethernet frame
size, the padding octets are not covered by the hardware checksum.
Fortunately the padding octets are usually zero's, which don't affect
checksum. However, it is not guaranteed. For example, switches might
choose to make other use of these octets.
This repeatedly causes kernel hardware checksum fault.
Prior to the cited commit below, skb checksum was forced to be
CHECKSUM_NONE when padding is detected. After it, we need to keep
skb->csum updated. However, fixing up CHECKSUM_COMPLETE requires to
verify and parse IP headers, it does not worth the effort as the packets
are so small that CHECKSUM_COMPLETE has no significant advantage.
Future work: when reporting checksum complete is not an option for
IP non-TCP/UDP packets, we can actually fallback to report checksum
unnecessary, by looking at cqe IPOK bit.
Fixes: 88078d98d1 ("net: pskb_trim_rcsum() and CHECKSUM_COMPLETE are friends")
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commit b639583f9e.
As per discussion with Jakub Kicinski and Michal Kubecek,
this will be better addressed by soon-too-come ethtool netlink
API with additional indication that given configuration request
is supposed to be persisted.
Also, remove the parameter support from bnxt_en driver.
Cc: Jiri Pirko <jiri@mellanox.com>
Cc: Michael Chan <michael.chan@broadcom.com>
Cc: Michal Kubecek <mkubecek@suse.cz>
Suggested-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Vasundhara Volam <vasundhara-v.volam@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The DMA API generally relies on a struct device to work properly, and
only barely works without one for legacy reasons. Pass the easily
available struct device from the platform_device to remedy this.
Note that smc911x apparently is a PIO chip with an external DMA
handshake, and we probably use the wrong device here. But at least
it matches the mapping side, which apparently works or at least
worked in the not too distant past.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The DMA API generally relies on a struct device to work properly, and
only barely works without one for legacy reasons. Pass the easily
available struct device from the platform_device to remedy this.
Also use GFP_KERNEL instead of GFP_ATOMIC as the gfp_t for the memory
allocation, as we aren't in interrupt context or under a lock.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The DMA API generally relies on a struct device to work properly, and
only barely works without one for legacy reasons. Pass the easily
available struct device from the platform_device to remedy this.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The DMA API generally relies on a struct device to work properly, and
only barely works without one for legacy reasons. Pass the easily
available struct device from the platform_device to remedy this.
Note that this driver seems to entirely lack dma_map_single error
handling, but that is left for another time.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The DMA API generally relies on a struct device to work properly, and
only barely works without one for legacy reasons. Pass the easily
available struct device from the platform_device to remedy this.
Note this driver seems to lack dma_unmap_* calls entirely, but fixing
that is left for another time.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The DMA API generally relies on a struct device to work properly, and
only barely works without one for legacy reasons. Pass the easily
available struct device from the platform_device to remedy this.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The DMA API generally relies on a struct device to work properly, and
only barely works without one for legacy reasons. Pass the easily
available struct device from the platform_device to remedy this.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
In case the register access failed an error would be logged anyway, so
we can drop the warning.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The LAG port collecting (receive) function was mistakenly set when the
port was registered as a LAG member, while it should be set only when
the port collection state is set to true. Set LAG port to collecting
when it is set to distributing, as described in the IEEE link
aggregation standard coupled control mux machine state diagram.
Signed-off-by: Nir Dotan <nird@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The bitmap of found partitions in efx_ef10_mtd_probe was not
initialised, causing partitions to be suppressed based off whatever
value was in the bitmap at the start.
Fixes: 3366463513 ("sfc: suppress duplicate nvmem partition types in efx_ef10_mtd_probe")
Signed-off-by: Bert Kenward <bkenward@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently bpf_offload_dev does not have any priv pointer, forcing
the drivers to work backwards from the netdev in program metadata.
This is not great given programs are conceptually associated with
the offload device, and it means one or two unnecessary deferences.
Add a priv pointer to bpf_offload_dev.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Quentin Monnet <quentin.monnet@netronome.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
The manufacturing team requests we include vendor and product
in the serial number field, as the serial number itself is not
unique across manufacturing facilities and products.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
mlx5_eq_cq_get() is called in IRQ handler, the spinlock inside
gets a lot of contentions when we test some heavy workload
with 60 RX queues and 80 CPU's, and it is clearly shown in the
flame graph.
In fact, radix_tree_lookup() is perfectly fine with RCU read lock,
we don't have to take a spinlock on this hot path. This is pretty
much similar to commit 291c566a28
("net/mlx4_core: Fix racy CQ (Completion Queue) free"). Slow paths
are still serialized with the spinlock, and with synchronize_irq()
it should be safe to just move the fast path to RCU read lock.
This patch itself reduces the latency by about 50% for our memcached
workload on a 4.14 kernel we test. In upstream, as pointed out by Saeed,
this spinlock gets some rework in commit 02d92f7903
("net/mlx5: CQ Database per EQ"), so the difference could be smaller.
Cc: Saeed Mahameed <saeedm@mellanox.com>
Cc: Tariq Toukan <tariqt@mellanox.com>
Acked-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
This reverts commit bd7153bd83.
There doesn't seem to be anything wrong with this patch,
it's just reverted to get a stable baseline again.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When mvpp2 configures the flow control modes in mvpp2_xlg_config() for
10G mode, it only ever set the flow control enable bits. There is no
mechanism to clear these bits, which means that userspace is unable to
use standard APIs to disable flow control (the only way is to poke the
register directly.)
Fix the missing bit clearance to allow flow control to be disabled.
This means that, by default, as there is no negotiation in 10G modes
with mvpp2, flow control is now disabled rather than being rx-only.
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
phylink already limits which interface modes are able to call the
MACs AN restart function, but in any case, the commentry seems
incorrect: the AN restart bit does not automatically clear when
set. This has been found via manual setting using devmem2, and
we can observe that the AN does indeed restart and complete, yet
the AN restart bit remains set. Explicitly clear the AN restart
bit.
Tested-by: Sven Auhagen <sven.auhagen@voleatech.de>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
When reading the pause bits in mac_link_state, mvpp2 was reporting
the state of the "active pause" bits, which are set when the MAC is
in pause mode. This is not what phylink wants - we want the
negotiated pause state. Fix the definition so we read the correct
bits.
Tested-by: Sven Auhagen <sven.auhagen@voleatech.de>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
mac_config() can be called at any point, and the expected behaviour
from MAC drivers is to only reprogram when necessary - and certainly
avoid taking the link down on every call.
Unfortunately, mvpp2 does exactly that - it takes the link down, and
reprograms everything, and then releases the forced-link down.
This is bad, it can cause the link to bounce:
- SFP detects signal, disables LOS indication.
- SFP code calls into phylink, calling phylink_sfp_link_up() which
triggers a resolve.
- phylink_resolve() calls phylink_get_mac_state() and finds the MAC
reporting link up.
- phylink wants to configure the pause mode on the MAC, so calls
phylink_mac_config()
- mvpp2 takes the link down temporarily, generating a MAC link down
event followed by another MAC link event.
- phylink calls mac_link_up() and then processes the MAC link down
event.
- phylink_resolve() gets called again, registers the link down, and
calls mach_link_down() before re-running itself.
- phylink_resolve() starts again at step 3 above. This sequence
repeats.
GMAC versions prior to mvpp2 do not require the link to be taken down
except when certain link properties (eg, switching between SGMII and
1000base-X mode, or enabling/disabling in-band negotiation) are
changed. Implement this for mvpp2.
Tested-by: Sven Auhagen <sven.auhagen@voleatech.de>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
It appears that the mvpp22 can get stuck with SGMII negotiation. The
symptoms are that in-band negotiation never completes and the partner
(eg, PHY) never reports SGMII link up, or if it supports negotiation
bypass, goes into negotiation bypass mode (which will happen when the
PHY sees that the MAC is alive but gets no response.)
Triggering the PHY end of the link to re-negotiate results in the
bypass bit clearing on the PHY, and then re-setting - indicating that
the problem is at the mvpp22 GMAC end.
Asserting the GMAC reset and de-asserting it resolves the issue.
Arrange to assert the GMAC reset at probe time, and deassert it only
after we have configured the GMAC for the appropriate mode. This
resolves the issue.
Tested-by: Sven Auhagen <sven.auhagen@voleatech.de>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Sven Auhagen reported issues with negotiation on a couple of his
platforms using a mixture of SFP and PHYs in various different
modes. Debugging to root cause proved difficult, but essentially
the problem comes down to the mvpp2 phylink implementation being
slightly at odds with what is expected.
phylink operates in three modes: phy, fixed-link, and in-band mode.
In the first two modes, the expected behaviour from a MAC driver is
that phylink resolves the operating mode and passes the mode to the
MAC driver for it to program, including when the link should be
brought up or taken down. This is basically the same as the libphy
approach. This does not negate the requirement to advertise a correct
control word for interface modes that have control words where that
can be reasonably controlled.
The second mode is in-band mode, where the MAC is expected to use the
in-band control word to determine the operating mode.
The mvneta driver implements the correct pattern required to support
this: configure the port interface type separately from the in-band
mode(s). This is now specified in the phylink documentation patches.
mvpp2 was programming in-band mode for SGMII and the 802.3z modes no
what, and avoided forcing the link up in fixed/phy modes. This caused
a problem with some boards where the PHY is by default programmed to
enter AN bypass mode, the PHY would report that the link was up, but
the mvpp2 never completed the exchange of control word.
Another issue that mvpp2 has is it sets SGMII AN format control word
for both SGMII and 802.3z modes. The format of the control word is
defined by MVPP2_GMAC_INBAND_AN_MASK, which should be set for SGMII
and clear for 802.3z. Available Marvell documentation for earlier
GMAC implementations does not make this clear, but this has been
ascertained via extensive testing on earlier GMAC implementations,
and then confirmed with a Macchiatobin Single Shot connected to a
Clearfog: when MVPP2_GMAC_INBAND_AN_MASK is set, the clearfog does
not receive the advertised pause mode settings.
Lastly, there is no flow control in the in-band control word in Cisco
SGMII, setting the flow control autonegotiation bit even with a PHY
that has the Marvell extension to send this information does not result
in the flow control being enabled at the MAC. We need to do this
manually using the information provided via phylink.
Re-code mvpp2's mac_config() and mac_link_up() to follow this pattern.
This allows Sven Auhagen's board and Macchiatobin to reliably bring
the link up with the 88e1512 PHY with phylink operating in PHY mode
with COMPHY built as a module but the rest of the networking built-in,
and u-boot having brought up the interface. in-band mode requires an
additional patch to resolve another problem.
Tested-by: Sven Auhagen <sven.auhagen@voleatech.de>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
One of the more common cases of allocation size calculations is finding
the size of a structure that has a zero-sized array at the end, along
with memory for some number of elements for that array. For example:
struct foo {
int stuff;
struct boo entry[];
};
size = sizeof(struct foo) + count * sizeof(struct boo);
instance = kzalloc(size, GFP_KERNEL);
Instead of leaving these open-coded and prone to type mistakes, we can
now use the new struct_size() helper:
instance = kzalloc(struct_size(instance, entry, count), GFP_KERNEL);
Notice that, in this case, variable size is not necessary, hence
it is removed.
This code was detected with the help of Coccinelle.
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
One of the more common cases of allocation size calculations is finding
the size of a structure that has a zero-sized array at the end, along
with memory for some number of elements for that array. For example:
struct foo {
int stuff;
struct boo entry[];
};
size = sizeof(struct foo) + count * sizeof(struct boo);
instance = kzalloc(size, GFP_KERNEL)
Instead of leaving these open-coded and prone to type mistakes, we can
now use the new struct_size() helper:
instance = kzalloc(struct_size(instance, entry, count), GFP_KERNEL)
Notice that, in this case, variable size is not necessary, hence
it is removed.
This code was detected with the help of Coccinelle.
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
One of the more common cases of allocation size calculations is finding
the size of a structure that has a zero-sized array at the end, along
with memory for some number of elements for that array. For example:
struct foo {
int stuff;
struct boo entry[];
};
size = sizeof(struct foo) + count * sizeof(struct boo);
instance = alloc(size, GFP_KERNEL);
Instead of leaving these open-coded and prone to type mistakes, we can
now use the new struct_size() helper:
size = struct_size(instance, entry, count);
This code was detected with the help of Coccinelle.
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
One of the more common cases of allocation size calculations is finding
the size of a structure that has a zero-sized array at the end, along
with memory for some number of elements for that array. For example:
struct foo {
int stuff;
struct boo entry[];
};
size = sizeof(struct foo) + count * sizeof(struct boo);
instance = kzalloc(size, GFP_KERNEL);
Instead of leaving these open-coded and prone to type mistakes, we can
now use the new struct_size() helper:
instance = kzalloc(struct_size(instance, entry, count), GFP_KERNEL);
Notice that, in this case, variable size is not necessary, hence
it is removed.
This code was detected with the help of Coccinelle.
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
One of the more common cases of allocation size calculations is finding
the size of a structure that has a zero-sized array at the end, along
with memory for some number of elements for that array. For example:
struct foo {
int stuff;
void *entry[];
};
size = sizeof(struct foo) + count * sizeof(void *);
instance = alloc(size, GFP_KERNEL);
Instead of leaving these open-coded and prone to type mistakes, we can
now use the new struct_size() helper:
instance = alloc(struct_size(instance, entry, count), GFP_KERNEL);
Notice that, in this case, variable size is not necessary, hence
it is removed.
This code was detected with the help of Coccinelle.
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
One of the more common cases of allocation size calculations is finding
the size of a structure that has a zero-sized array at the end, along
with memory for some number of elements for that array. For example:
struct foo {
int stuff;
struct boo entry[];
};
size = sizeof(struct foo) + count * sizeof(struct boo);
instance = kzalloc(size, GFP_KERNEL)
Instead of leaving these open-coded and prone to type mistakes, we can
now use the new struct_size() helper:
instance = kzalloc(struct_size(instance, entry, count), GFP_KERNEL)
Notice that, in this case, variable alloc_size is not necessary, hence
it is removed.
This code was detected with the help of Coccinelle.
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
One of the more common cases of allocation size calculations is finding
the size of a structure that has a zero-sized array at the end, along
with memory for some number of elements for that array. For example:
struct foo {
int stuff;
struct boo entry[];
};
size = sizeof(struct foo) + count * sizeof(struct boo);
instance = kzalloc(size, GFP_KERNEL)
Instead of leaving these open-coded and prone to type mistakes, we can
now use the new struct_size() helper:
instance = kzalloc(struct_size(instance, entry, count), GFP_KERNEL)
Notice that, in this case, variable fsz is not necessary, hence
it is removed.
This code was detected with the help of Coccinelle.
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a new compatibility string for this product. It's using
at91sam9260-macb layout but has a newer hardware revision: it's safer
to use its own string.
Signed-off-by: Nicolas Ferre <nicolas.ferre@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As vregion rehash is happening in delayed work, add some visibility to
the process using a few tracepoints.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Expose new driver-specific "acl_region_rehash_interval" devlink param
which would allow user to alter default ACL region rehash interval.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the hints are returned, the migration should be started. For that to
happen, there is a need to create a second physical region in TCAM with
new ERP set by passing the hints and then move chunk by chunk,
entry by entry.
During the transition, two lookups will occur. One in old region and
another in new region. The highest priority rule will be chosen.
In an unlikely case that the migration will fail and also rollback to
original region will fail the vregion will become in bad state.
Everything will work, only no future rehash will be possible. In a
follow-up work, this can be resolved by trying to resume the rollback
in delayed work and repair the vregion.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
