We have observed complete lock up of broadcast-link transmission due to
unacknowledged packets never being removed from the 'transmq' queue. This
is traced to nodes having their ack field set beyond the sequence number
of packets that have actually been transmitted to them.
Consider an example where node 1 has sent 10 packets to node 2 on a
link and node 3 has sent 20 packets to node 2 on another link. We
see examples of an ack from node 2 destined for node 3 being treated as
an ack from node 2 at node 1. This leads to the ack on the node 1 to node
2 link being increased to 20 even though we have only sent 10 packets.
When node 1 does get around to sending further packets, none of the
packets with sequence numbers less than 21 are actually removed from the
transmq.
To resolve this we reinstate some code lost in commit d999297c ("tipc:
reduce locking scope during packet reception") which ensures that only
messages destined for the receiving node are processed by that node. This
prevents the sequence numbers from getting out of sync and resolves the
packet leakage, thereby resolving the broadcast-link transmission
lock-ups we observed.

While we are aware that this change only patches over a root problem that
we still haven't identified, this is a sanity test that it is always
legitimate to do. It will remain in the code even after we identify and
fix the real problem.
Reviewed-by: Chris Packham <chris.packham@alliedtelesis.co.nz>
Reviewed-by: John Thompson <john.thompson@alliedtelesis.co.nz>
Signed-off-by: Hamish Martin <hamish.martin@alliedtelesis.co.nz>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

An out of bounds read of 2 bytes was discovered in cxgb3 with KASAN.

t3_config_rss() expects both arrays it gets as parameters to have
terminators. setup_rss(), the caller, forgets to add a terminator to
one of the arrays. Thankfully the iteration in t3_config_rss() stops
anyway, but in the last iteration the check for the terminator
is an out of bounds read.

Add the missing terminator to rspq_map[].
Reported-by: Jan Stancek <jstancek@redhat.com>
Signed-off-by: Michal Schmidt <mschmidt@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This takes the MAC address for smsc75xx/smsc95xx USB network devices
from a the device tree. This is required to get a usable persistent
address on the popular beagleboard, whose hardware designers
accidentally forgot that an ethernet device really requires an a
MAC address to be functional.

The Raspberry Pi also ships smsc9514 without a serial EEPROM, stores
the MAC address in ROM accessible via VC4 firmware.

The smsc75xx and smsc95xx drivers are just two copies of the
same code, so better fix both.

[lkundrak@v3.sk: updated to use of_get_property() as per suggestion from
Arnd, reworded the message and comments a bit]
Tested-by: Lubomir Rintel <lkundrak@v3.sk>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Lubomir Rintel <lkundrak@v3.sk>
Signed-off-by: David S. Miller <davem@davemloft.net>

I forgot to include a check for listener port equality when deciding
if two sockets should belong to the same reuseport group.  This was
not caught previously because it's only necessary when two listening
sockets for the same user happen to hash to the same listener bucket.
The same error does not exist in the UDP path.

Fixes: c125e80b("soreuseport: fast reuseport TCP socket selection")
Signed-off-by: Craig Gallek <kraig@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch fixes a bug which causes the behavior of whether to ignore
udp6 checksum of udp6 encapsulated l2tp tunnel contrary to what
userspace program requests.

When the flag `L2TP_ATTR_UDP_ZERO_CSUM6_RX` is set by userspace, it is
expected that udp6 checksums of received packets of the l2tp tunnel
to create should be ignored. In `l2tp_netlink.c`:
`l2tp_nl_cmd_tunnel_create()`, `cfg.udp6_zero_rx_checksums` is set
according to the flag, and then passed to `l2tp_core.c`:
`l2tp_tunnel_create()` and then `l2tp_tunnel_sock_create()`. In
`l2tp_tunnel_sock_create()`, `udp_conf.use_udp6_rx_checksums` is set
the same to `cfg.udp6_zero_rx_checksums`. However, if we want the
checksum to be ignored, `udp_conf.use_udp6_rx_checksums` should be set
to `false`, i.e. be set to the contrary. Similarly, the same should be
done to `udp_conf.use_udp6_tx_checksums`.
Signed-off-by: Miao Wang <shankerwangmiao@gmail.com>
Acked-by: James Chapman <jchapman@katalix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

After the commit e09acddf ("ip_tunnel: replace dst_cache with generic
implementation"), a preemption debug warning is triggered on ip4
tunnels updating; the dst cache helper needs to be invoked in unpreemptible
context.

We don't need to load the cache on tunnel update, so this commit fixes
the warning replacing the load with a dst cache reset, which is
preempt safe.

Fixes: e09acddf ("ip_tunnel: replace dst_cache with generic implementation")
Reported-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Alexei Starovoitov says:

====================
bpf: fix several bugs

First two patches address bugs found by Jann Horn.
Last patch is a minor samples fix spotted during the testing.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

llvm cannot always recognize memset as builtin function and optimize
it away, so just delete it. It was a leftover from testing
of bpf_perf_event_output() with large data structures.

Fixes: 39111695 ("samples: bpf: add bpf_perf_event_output example")
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

The commit 35578d79 ("bpf: Implement function bpf_perf_event_read() that get the selected hardware PMU conuter")
introduced clever way to check bpf_helper<->map_type compatibility.
Later on commit a43eec30 ("bpf: introduce bpf_perf_event_output() helper") adjusted
the logic and inadvertently broke it.
Get rid of the clever bool compare and go back to two-way check
from map and from helper perspective.

Fixes: a43eec30 ("bpf: introduce bpf_perf_event_output() helper")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

On a system with >32Gbyte of phyiscal memory and infinite RLIMIT_MEMLOCK,
the malicious application may overflow 32-bit bpf program refcnt.
It's also possible to overflow map refcnt on 1Tb system.
Impose 32k hard limit which means that the same bpf program or
map cannot be shared by more than 32k processes.

Fixes: 1be7f75d ("bpf: enable non-root eBPF programs")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

David Rivshin says:

====================
drivers: net: cpsw: phy-handle fixes

This series fixes a number of related issues around using phy-handle
properties in cpsw emac nodes.

Patch 1 fixes a bug if more than one slave is used, and either
slave uses the phy-handle property in the devicetree.

Patch 2 fixes a NULL pointer dereference which can occur if a
phy-handle property is used and of_phy_connect() return NULL,
such as with a bad devicetree.

Patch 3 fixes an issue where the phy-mode property would be ignored
if a phy-handle property was used. This also fixes a bogus error
message that would be emitted.

Patch 4 fixes makes the binding documentation more explicit that
exactly one PHY property should be used, and also marks phy_id as
deprecated.

Patch 5 cleans up the fixed-link case to work like the now-fixed
phy-handle case.

I have tested on the following hardware configurations:
 - (EVMSK) dual emac, phy_id property in both slaves
 - (EVMSK) dual emac, phy-handle property in both slaves
 - (EVMSK) a bad phy-handle property pointing to &mmc1
 - (EVMSK) phy_id property with incorrect PHY address
 - (BeagleBoneBlack) single emac, phy_id property
 - (custom) single emac, fixed-link subnode

Andrew Goodbody reported testing v2 on a board that doesn't use
dual_emac mode, but with 2 PHYs using phy-handle properties [1].

Nicolas Chauvet reported testing v2 on an HP t410 (dm8148).

Markus Brunner reported testing v1 on the following [2]:
 - emac0 with phy_id and emac1 with fixed phy
 - emac0 with phy-handle and emac1 with fixed phy
 - emac0 with fixed phy and emac1 with fixed phy

[1] https://lkml.org/lkml/2016/4/22/537
[2] http://www.spinics.net/lists/netdev/msg357890.html
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

If a fixed-link DT subnode is used, the phy_device was looked up so
that a PHY ID string could be constructed and passed to phy_connect().
This is not necessary, as the device_node can be passed directly to
of_phy_connect() instead. This reuses the same codepath as if the
phy-handle DT property was used.
Signed-off-by: David Rivshin <drivshin@allworx.com>
Tested-by: Nicolas Chauvet <kwizart@gmail.com>
Tested-by: Andrew Goodbody <andrew.goodbody@cambrionix.com>
Reviewed-by: Mugunthan V N <mugunthanvnm@ti.com>
Reviewed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The phy-handle, phy_id, and fixed-link properties are mutually exclusive,
and only one need be specified. Make this clear in the binding doc.

Also mark the phy_id property as deprecated, as phy-handle should be
used instead.
Signed-off-by: David Rivshin <drivshin@allworx.com>
Reviewed-by: Mugunthan V N <mugunthanvnm@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The phy-mode emac property was only being processed in the phy_id
or fixed-link cases. However if phy-handle was specified instead,
an error message would complain about the lack of phy_id or
fixed-link, and then jump past the of_get_phy_mode(). This would
result in the PHY mode defaulting to MII, regardless of what the
devicetree specified.

Fixes: 9e42f715 ("drivers: net: cpsw: add phy-handle parsing")
Signed-off-by: David Rivshin <drivshin@allworx.com>
Tested-by: Nicolas Chauvet <kwizart@gmail.com>
Tested-by: Andrew Goodbody <andrew.goodbody@cambrionix.com>
Reviewed-by: Mugunthan V N <mugunthanvnm@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

If an emac node has a phy-handle property that points to something
which is not a phy, then a segmentation fault will occur when the
interface is brought up. This is because while phy_connect() will
return ERR_PTR() on failure, of_phy_connect() will return NULL.
The common error check uses IS_ERR(), and so missed when
of_phy_connect() fails. The NULL pointer is then dereferenced.

Also, the common error message referenced slave->data->phy_id,
which would be empty in the case of phy-handle. Instead, use the
name of the device_node as a useful identifier. And in the phy_id
case add the error code for completeness.

Fixes: 9e42f715 ("drivers: net: cpsw: add phy-handle parsing")
Signed-off-by: David Rivshin <drivshin@allworx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Commit 9e42f715 ("drivers: net: cpsw: add
phy-handle parsing") saved the "phy-handle" phandle into a new cpsw_priv
field. However, phy connections are per-slave, so the phy_node field should
be in cpsw_slave_data rather than cpsw_priv.

This would go unnoticed in a single emac configuration. But in dual_emac
mode, the last "phy-handle" property parsed for either slave would be used
by both of them, causing them both to refer to the same phy_device.

Fixes: 9e42f715 ("drivers: net: cpsw: add phy-handle parsing")
Signed-off-by: David Rivshin <drivshin@allworx.com>
Tested-by: Nicolas Chauvet <kwizart@gmail.com>
Tested-by: Andrew Goodbody <andrew.goodbody@cambrionix.com>
Reviewed-by: Mugunthan V N <mugunthanvnm@ti.com>
Reviewed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Andreas Larsson <andreas@gaisler.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The collect metadata mode does not support GUE nor FOU. This might be
implemented later; until then, we should reject such config.

I think this is okay to be changed. It's unlikely anyone has such
configuration (as it doesn't work anyway) and we may need a way to
distinguish whether it's supported or not by the kernel later.

For backwards compatibility with iproute2, it's not possible to just check
the attribute presence (iproute2 always includes the attribute), the actual
value has to be checked, too.

Fixes: 2e15ea39 ("ip_gre: Add support to collect tunnel metadata.")
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Petko Manolov says:

====================
pegasus: correct buffer & packet sizes

As noticed by Lincoln Ramsay <a1291762@gmail.com> some old (usb 1.1) Pegasus
based devices may actually return more bytes than the specified in the datasheet
amount.  That would not be a problem if the allocated space for the SKB was
equal to the parameter passed to usb_fill_bulk_urb().  Some poor bugger (i
really hope it was not me, but 'git blame' is useless in this case, so anyway)
decided to add '+ 8' to the buffer length parameter.  Sometimes the usb transfer
overflows and corrupts the socket structure, leading to kernel panic.

The above doesn't seem to happen for newer (Pegasus2 based) devices which did
help this bug to hide for so long.

The new default is to not include the CRC at the end of each received package.
So far CRC has been ignored which makes no sense to do it in a first place.

The patch is against v4.6-rc5 and was tested on ADM8515 device by transferring
multiple gigabytes of data over a couple of days without any complaints from the
kernel.  Please apply it to whatever net tree you deem fit.

Changes since v1:

 - split the patch in two parts;
 - corrected the subject lines;

Changes since v2:

 - do not append CRC by default (based on a discussion with Johannes Berg);
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The default Pegasus setup was to append the status and CRC at the end of each
received packet.  The status bits are used to update various stats, but CRC has
been ignored.  The new default is to not append CRC at the end of RX packets.
Signed-off-by: Petko Manolov <petkan@mip-labs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

usb_fill_bulk_urb() receives buffer length parameter 8 bytes larger
than what's allocated by alloc_skb(); This seems to be a problem with
older (pegasus usb-1.1) devices, which may silently return more data
than the maximal packet length.
Reported-by: Lincoln Ramsay <a1291762@gmail.com>
Signed-off-by: Petko Manolov <petkan@mip-labs.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Jiri Benc says:

====================
gre: fix lwtunnel support

This patchset fixes a few bugs in ipgre metadata mode implementation.

As an example, in this setup:

ip a a 192.168.1.1/24 dev eth0
ip l a gre1 type gre external
ip l s gre1 up
ip a a 192.168.99.1/24 dev gre1
ip r a 192.168.99.2/32 encap ip dst 192.168.1.2 ttl 10 dev gre1
ping 192.168.99.2

the traffic does not go through before this patchset and does as expected
with it applied.

v3: Back to v1 in order not to break existing users. Dropped patch 3, will
    be fixed in iproute2 instead.
v2: Rejecting invalid configuration, added patch 3, dropped patch for
    ETH_P_TEB (will target net-next).
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

In ipgre (i.e. not gretap) + collect metadata mode, the skb was assumed to
contain Ethernet header and was encapsulated as ETH_P_TEB. This is not the
case, the interface is ARPHRD_IPGRE and the protocol to be used for
encapsulation is skb->protocol.

Fixes: 2e15ea39 ("ip_gre: Add support to collect tunnel metadata.")
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In ipgre mode (i.e. not gretap) with collect metadata flag set, the tunnel
is incorrectly assumed to be mGRE in NBMA mode (see commit 6a5f44d7).
This is not the case, we're controlling the encapsulation addresses by
lwtunnel metadata. And anyway, assigning dev->header_ops in collect metadata
mode does not make sense.

Although it would be more user firendly to reject requests that specify
both the collect metadata flag and a remote/local IP address, this would
break current users of gretap or introduce ugly code and differences in
handling ipgre and gretap configuration. Keep the current behavior of
remote/local IP address being ignored in such case.

v3: Back to v1, added explanation paragraph.
v2: Reject configuration specifying both remote/local address and collect
    metadata flag.

Fixes: 2e15ea39 ("ip_gre: Add support to collect tunnel metadata.")
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge tag 'mac80211-for-davem-2016-04-27' of git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211

Johannes Berg says:

====================
Just a single fix, for a per-CPU memory leak in a
(root user triggerable) error case.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Commit 13a56b44 ("at803x: Add support for hardware reset") added a
work-around for a hardware bug on the AT8030.  However, the work-around
was being called for all 803x PHYs, even those that don't need it.
Function at803x_link_change_notify() checks to make sure that it only
resets the PHY on the 8030, but it makes more sense to not call that
function at all if it isn't needed.
Signed-off-by: Timur Tabi <timur@codeaurora.org>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Antonio Quartulli says:

====================
In this patchset you can find the following fixes:

1) check skb size to avoid reading beyond its border when delivering
   payloads, by Sven Eckelmann
2) initialize last_seen time in neigh_node object to prevent cleanup
   routine from accidentally purge it, by Marek Lindner
3) release "recently added" slave interfaces upon virtual/batman
   interface shutdown, by Sven Eckelmann
4) properly decrease router object reference counter upon routing table
   update, by Sven Eckelmann
5) release queue slots when purging OGM packets of deactivating slave
   interface, by Linus Lüssing

Patch 2 and 3 have no "Fixes:" tag because the offending commits date
back to when batman-adv was not yet officially in the net tree.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The size allocated for target->hwinfo and the number of bytes copied in it
should be consistent.
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>

At forced 100 Full & Half duplex mode, chip may fail to set mode correctly
when cable is switched between long(~50+m) and short one.
As workaround, set to 10 before setting to 100 at forced 100 F/H mode.
Signed-off-by: Woojung Huh <woojung.huh@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Fix rx_bytes, tx_bytes and tx_frames error in netdev.stats.
- rx_bytes counted bytes excluding size of struct ethhdr.
- tx_packets didn't count multiple packets in a single urb
- tx_bytes included 8 bytes of extra commands.
Signed-off-by: Woojung Huh <woojung.huh@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The error return err is not initialized and there is a possibility
that err is not assigned causing mv88e6xxx_port_bridge_join to
return a garbage error return status. Fix this by initializing err
to 0.
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Fix casting in net_gso_ok. Otherwise the shift on
gso_type << NETIF_F_GSO_SHIFT may hit the 32th bit and make it look like
a INT_MIN, which is then promoted from signed to uint64 which is
0xffffffff80000000, resulting in wrong behavior when it is and'ed with
the feature itself, as in:

This test app:
#include <stdio.h>
#include <stdint.h>

int main(int argc, char **argv)
{
	uint64_t feature1;
	uint64_t feature2;
	int gso_type = 1 << 15;

	feature1 = gso_type << 16;
	feature2 = (uint64_t)gso_type << 16;
	printf("%lx %lx\n", feature1, feature2);

	return 0;
}

Gives:
ffffffff80000000 80000000

So that this:
   return (features & feature) == feature;
Actually works on more bits than expected and invalid ones.

Fix is to promote it earlier.

Issue noted while rebasing SCTP GSO patch but posting separetely as
someone else may experience this meanwhile.
Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When configured in fixed link, the DaVinci emac driver sets the
priv->phydev to NULL and further ioctl calls to the phy_mii_ioctl()
causes the kernel to crash.

Cc: Brian Hutchinson <b.hutchman@gmail.com>
Fixes: 1bb6aa56 ("net: davinci_emac: Add support for fixed-link PHY")
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Reviewed-by: Mugunthan V N <mugunthanvnm@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge tag 'wireless-drivers-for-davem-2016-04-25' of git://git.kernel.org/pub/scm/linux/kernel/git/kvalo/wireless-drivers

Kalle Valo says:

====================
wireless-drivers fixes for 4.6

ath9k

* fix a couple release old throughput regression on ar9281

iwlwifi

* add new device IDs for 8265
* fix a NULL pointer dereference when paging firmware asserts
* remove a WARNING on gscan capabilities
* fix MODULE_FIRMWARE for 8260
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Add myself and Edward Cree as maintainers.
Remove Shradha Shah, who is on extended leave.

Cc: David S. Miller <davem@davemloft.net>
Cc: Edward Cree <ecree@solarflare.com>
Cc: Shradha Shah <sshah@solarflare.com>
Signed-off-by: Bert Kenward <bkenward@solarflare.com>
Acked-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When certain firmware variants are selected (via the sfboot utility) the
SFC7000 and SFC8000 series NICs don't support RSS. The driver still
tries (and fails) to insert filters with the RSS flag, and the NIC fails
to pass traffic.

When the firmware reports RSS_LIMITED suppress allocating a default RSS
context. The absence of an RSS context is picked up in filter insertion
and RSS flags are discarded.
Signed-off-by: Bert Kenward <bkenward@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

napi_disable() can not be called with bh disabled, move locking just
around myri10ge_ss_lock_napi() .

Patches fixes following bug:

[  114.278378] BUG: sleeping function called from invalid context at net/core/dev.c:4383
<snip>
[  114.313712] Call Trace:
[  114.314943]  [<ffffffff817010ce>] dump_stack+0x19/0x1b
[  114.317673]  [<ffffffff810ce7f3>] __might_sleep+0x173/0x230
[  114.320566]  [<ffffffff815b3117>] napi_disable+0x27/0x90
[  114.323254]  [<ffffffffa01e437f>] myri10ge_close+0xbf/0x3f0 [myri10ge]
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Acked-by: Hyong-Youb Kim <hykim@myri.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

Michael Chan says:

====================
bnxt_en: Bug fixes for net.

Only use MSIX on VF, and fix rx page buffers on architectures with
PAGE_SIZE >= 64K.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

If PAGE_SIZE is bigger than BNXT_RX_PAGE_SIZE, that means the native CPU
page is bigger than the maximum length of the RX BD.  Divide the page
into multiple 32K buffers for the aggregation ring.

Add an offset field in the bnxt_sw_rx_agg_bd struct to keep track of the
page offset of each buffer.  Since each page can be referenced by multiple
buffer entries, call get_page() as needed to get the proper reference
count.
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>