platform/kernel/linux-rpi.git
5 years agonet: axienet: Use standard IO accessors
Robert Hancock [Thu, 6 Jun 2019 22:28:06 +0000 (16:28 -0600)]
net: axienet: Use standard IO accessors

This driver was using in_be32 and out_be32 IO accessors which do not
exist on most platforms. Also, the use of big-endian accessors does not
seem correct as this hardware is accessed over an AXI bus which, to the
extent it has an endian-ness, is little-endian. Switch to standard
ioread32/iowrite32 accessors.

Signed-off-by: Robert Hancock <hancock@sedsystems.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: axienet: Fix casting of pointers to u32
Robert Hancock [Thu, 6 Jun 2019 22:28:05 +0000 (16:28 -0600)]
net: axienet: Fix casting of pointers to u32

This driver was casting skb pointers to u32 and storing them as such in
the DMA buffer descriptor, which is obviously broken on 64-bit. The area
of the buffer descriptor being used is not accessed by the hardware and
has sufficient room for a 32 or 64-bit pointer, so just store the skb
pointer as such.

Signed-off-by: Robert Hancock <hancock@sedsystems.ca>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: socfpga: fix phy and ptp_ref setup for Arria10/Stratix10
Dinh Nguyen [Wed, 5 Jun 2019 15:05:51 +0000 (10:05 -0500)]
net: stmmac: socfpga: fix phy and ptp_ref setup for Arria10/Stratix10

On the Arria10, Agilex, and Stratix10 SoC, there are a few differences from
the Cyclone5 and Arria5:
 - The emac PHY setup bits are in separate registers.
 - The PTP reference clock select mask is different.
 - The register to enable the emac signal from FPGA is different.

Thus, this patch creates a separate function for setting the phy modes on
Arria10/Agilex/Stratix10. The separation is based a new DTS binding:
"altr,socfpga-stmmac-a10-s10".

Signed-off-by: Dinh Nguyen <dinguyen@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agodt-bindings: socfpga-dwmac: add "altr, socfpga-stmmac-a10-s10" binding
Dinh Nguyen [Wed, 5 Jun 2019 15:05:50 +0000 (10:05 -0500)]
dt-bindings: socfpga-dwmac: add "altr, socfpga-stmmac-a10-s10" binding

Add the "altr,socfpga-stmmac-a10-s10" binding for Arria10/Agilex/Stratix10
implementation of the stmmac ethernet controller.

On the Arria10, Agilex, and Stratix10 SoCs, there are a few differences from
the Cyclone5 and Arria5:
     - The emac PHY setup bits are in separate registers.
     - The PTP reference clock select mask is different.
     - The register to enable the emac signal from FPGA is different.

Because of these differences, the dwmac-socfpga glue logic driver will
use this new binding to set the appropriate bits for PHY, PTP reference
clock, and signal from FPGA.

Signed-off-by: Dinh Nguyen <dinguyen@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'nfp-tls-add-basic-TX-offload'
David S. Miller [Thu, 6 Jun 2019 21:13:40 +0000 (14:13 -0700)]
Merge branch 'nfp-tls-add-basic-TX-offload'

Jakub Kicinski says:

====================
nfp: tls: add basic TX offload

This series adds initial TLS offload support to the nfp driver.
Only TX side is added for now.  We need minor adjustments to
the core tls code:
 - expose the per-skb fallback helper;
 - grow the driver context slightly;
 - add a helper to get to the driver state more easily.
We only support TX offload for now, and only if all packets
keep coming in order.  For retransmissions we use the
aforementioned software fallback, and in case there are
local drops we completely give up on given TCP stream.

This will obviously be improved soon, this patch set is the
minimal, functional yet easily reviewable chunk.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: tls: add basic statistics
Jakub Kicinski [Wed, 5 Jun 2019 21:11:43 +0000 (14:11 -0700)]
nfp: tls: add basic statistics

Count TX TLS packets: successes, out of order, and dropped due to
missing record info.  Make sure the RX and TX completion statistics
don't share cache lines with TX ones as much as possible.  With TLS
stats they are no longer reasonably aligned.

Signed-off-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: tls: add/delete TLS TX connections
Dirk van der Merwe [Wed, 5 Jun 2019 21:11:42 +0000 (14:11 -0700)]
nfp: tls: add/delete TLS TX connections

This patch adds the functionality to add and delete TLS connections on
the NFP, received from the kernel TLS callbacks.

Make use of the common control message (CCM) infrastructure to propagate
the kernel state to firmware.

Signed-off-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: tls: add datapath support for TLS TX
Dirk van der Merwe [Wed, 5 Jun 2019 21:11:41 +0000 (14:11 -0700)]
nfp: tls: add datapath support for TLS TX

Prepend connection handle to each transmitted TLS packet.

For each connection, the driver tracks the next sequence number
expected. If an out of order packet is observed, the driver calls into
the TLS kernel code to reencrypt that particular skb.

Signed-off-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: export TLS per skb encryption
Dirk van der Merwe [Wed, 5 Jun 2019 21:11:40 +0000 (14:11 -0700)]
net/tls: export TLS per skb encryption

While offloading TLS connections, drivers need to handle the case where
out of order packets need to be transmitted.

Other drivers obtain the entire TLS record for the specific skb to
provide as context to hardware for encryption. However, other designs
may also want to keep the hardware state intact and perform the
out of order encryption entirely on the host.

To achieve this, export the already existing software encryption
fallback path so drivers could access this.

Signed-off-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: simplify driver context retrieval
Jakub Kicinski [Wed, 5 Jun 2019 21:11:39 +0000 (14:11 -0700)]
net/tls: simplify driver context retrieval

Currently drivers have to ensure the alignment of their tls state
structure, which leads to unnecessary layers of getters and
encapsulated structures in each driver.

Simplify all this by marking the driver state as aligned (driver_state
members are currently aligned, so no hole is added, besides ALIGN in
TLS_OFFLOAD_CONTEXT_SIZE_RX/TX would reserve this extra space, anyway.)
With that we can add a common accessor to the core.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: split the TLS_DRIVER_STATE_SIZE and bump TX to 16 bytes
Jakub Kicinski [Wed, 5 Jun 2019 21:11:38 +0000 (14:11 -0700)]
net/tls: split the TLS_DRIVER_STATE_SIZE and bump TX to 16 bytes

8 bytes of driver state has been enough so far, but for drivers
which have to store 8 byte handle it's no longer practical to
store the state directly in the context.

Drivers generally don't need much extra state on RX side, while
TX side has to be tracking TCP sequence numbers.  Split the
lengths of max driver state size on RX and TX.

The struct tls_offload_context_tx currently stands at 616 bytes and
struct tls_offload_context_rx stands at 368 bytes.  Upcoming work
will consume extra 8 bytes in both for kernel-driven resync.
This means that we can bump TX side to 16 bytes and still fit
into the same number of cache lines but on RX side we would be 8
bytes over.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: prepare for more TX metadata prepend
Jakub Kicinski [Wed, 5 Jun 2019 21:11:37 +0000 (14:11 -0700)]
nfp: prepare for more TX metadata prepend

Subsequent patches will add support for more TX metadata fields.
Prepare for this by handling an additional double word - firmware
handle as metadata type 7.

Signed-off-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: add tls init code
Jakub Kicinski [Wed, 5 Jun 2019 21:11:36 +0000 (14:11 -0700)]
nfp: add tls init code

Add FW ABI defines and code for basic init of TLS offload.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: parse crypto opcode TLV
Jakub Kicinski [Wed, 5 Jun 2019 21:11:35 +0000 (14:11 -0700)]
nfp: parse crypto opcode TLV

Parse TLV containing a bitmask of supported crypto operations.
The TLV contains a capability bitmask (supported operations)
and enabled bitmask.  Each operation describes the crypto
protocol quite exhaustively (protocol, AEAD, direction).

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: add support for sending control messages via mailbox
Jakub Kicinski [Wed, 5 Jun 2019 21:11:34 +0000 (14:11 -0700)]
nfp: add support for sending control messages via mailbox

FW may prefer to handle some communication via a mailbox
or the vNIC may simply not have a control queue (VFs).
Add a way of exchanging ccm-compatible messages via a
mailbox.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: parse the mailbox cmsg TLV
Jakub Kicinski [Wed, 5 Jun 2019 21:11:33 +0000 (14:11 -0700)]
nfp: parse the mailbox cmsg TLV

Parse the mailbox TLV.  When control message queue is not available
we can fall back to passing the control messages via the vNIC
mailbox.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: make bar_lock a semaphore
Jakub Kicinski [Wed, 5 Jun 2019 21:11:32 +0000 (14:11 -0700)]
nfp: make bar_lock a semaphore

We will need to release the bar lock from a workqueue
so move from a mutex to a semaphore.  This lock should
not be too hot.  Unfortunately semaphores don't have
lockdep support.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: count all failed TX attempts as errors
Jakub Kicinski [Wed, 5 Jun 2019 21:11:31 +0000 (14:11 -0700)]
nfp: count all failed TX attempts as errors

Currently if we need to modify the head of the skb and allocation
fails we would free the skb and not increment the error counter.
Make sure all errors are counted.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: Add detection of 1000BaseX link mode support
Robert Hancock [Tue, 4 Jun 2019 22:15:01 +0000 (16:15 -0600)]
net: phy: Add detection of 1000BaseX link mode support

Add 1000BaseX to the link modes which are detected based on the
MII_ESTATUS register as per 802.3 Clause 22. This allows PHYs which
support 1000BaseX to work properly with drivers using phylink.

Previously 1000BaseX support was not detected, and if that was the only
mode the PHY indicated support for, phylink would refuse to attach it
due to the list of supported modes being empty.

Signed-off-by: Robert Hancock <hancock@sedsystems.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: mscc: ocelot: remove unused variable 'vcap_data_t'
YueHaibing [Thu, 6 Jun 2019 14:46:49 +0000 (22:46 +0800)]
net: mscc: ocelot: remove unused variable 'vcap_data_t'

Fix sparse warning:

drivers/net/ethernet/mscc/ocelot_ace.c:96:3:
 warning: symbol 'vcap_data_t' was not declared. Should it be static?

'vcap_data_t' never used so can be removed

Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: fec: Do not use netdev messages too early
Fabio Estevam [Thu, 6 Jun 2019 12:40:33 +0000 (09:40 -0300)]
net: fec: Do not use netdev messages too early

When a valid MAC address is not found the current messages
are shown:

fec 2188000.ethernet (unnamed net_device) (uninitialized): Invalid MAC address: 00:00:00:00:00:00
fec 2188000.ethernet (unnamed net_device) (uninitialized): Using random MAC address: aa:9f:25:eb:7e:aa

Since the network device has not been registered at this point, it is better
to use dev_err()/dev_info() instead, which will provide cleaner log
messages like these:

fec 2188000.ethernet: Invalid MAC address: 00:00:00:00:00:00
fec 2188000.ethernet: Using random MAC address: aa:9f:25:eb:7e:aa

Tested on a imx6dl-pico-pi board.

Signed-off-by: Fabio Estevam <festevam@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agovxlan: Use FDB_HASH_SIZE hash_locks to reduce contention
Litao jiao [Thu, 6 Jun 2019 09:57:58 +0000 (17:57 +0800)]
vxlan: Use FDB_HASH_SIZE hash_locks to reduce contention

The monolithic hash_lock could cause huge contention when
inserting/deletiing vxlan_fdbs into the fdb_head.

Use FDB_HASH_SIZE hash_locks to protect insertions/deletions
of vxlan_fdbs into the fdb_head hash table.

Suggested-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Litao jiao <jiaolitao@raisecom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: fix spelling mistake: "wtih" -> "with"
Colin Ian King [Thu, 6 Jun 2019 08:40:39 +0000 (09:40 +0100)]
ipv6: fix spelling mistake: "wtih" -> "with"

There is a spelling mistake in a NL_SET_ERR_MSG message. Fix it.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agor8169: silence sparse warning in rtl8169_start_xmit
Heiner Kallweit [Thu, 6 Jun 2019 05:49:17 +0000 (07:49 +0200)]
r8169: silence sparse warning in rtl8169_start_xmit

The opts[] array is of type u32. Therefore remove the wrong
cpu_to_le32(). The opts[] array members are converted to little endian
later when being assigned to the respective descriptor fields.

This is not a new issue, it just popped up due to r8169.c having
been renamed and more thoroughly checked. Due to the renaming
this patch applies to net-next only.

Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'tcp-flowlabel'
David S. Miller [Thu, 6 Jun 2019 02:05:01 +0000 (19:05 -0700)]
Merge branch 'tcp-flowlabel'

Eric Dumazet says:

====================
ipv6: tcp: more control on RST flowlabels

First patch allows to reflect incoming IPv6 flowlabel
on RST packets sent when no socket could handle the packet.

Second patch makes sure we send the same flowlabel
for RST or ACK packets on behalf of TIME_WAIT sockets.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: tcp: send consistent flowlabel in TIME_WAIT state
Eric Dumazet [Wed, 5 Jun 2019 14:55:10 +0000 (07:55 -0700)]
ipv6: tcp: send consistent flowlabel in TIME_WAIT state

After commit 1d13a96c74fc ("ipv6: tcp: fix flowlabel value in ACK
messages"), we stored in tw_flowlabel the flowlabel, in the
case ACK packets needed to be sent on behalf of a TIME_WAIT socket.

We can use the same field so that RST packets sent from
TIME_WAIT state also use a consistent flowlabel.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Florent Fourcot <florent.fourcot@wifirst.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: tcp: enable flowlabel reflection in some RST packets
Eric Dumazet [Wed, 5 Jun 2019 14:55:09 +0000 (07:55 -0700)]
ipv6: tcp: enable flowlabel reflection in some RST packets

When RST packets are sent because no socket could be found,
it makes sense to use flowlabel_reflect sysctl to decide
if a reflection of the flowlabel is requested.

This extends commit 22b6722bfa59 ("ipv6: Add sysctl for per
namespace flow label reflection"), for some TCP RST packets.

In order to provide full control of this new feature,
flowlabel_reflect becomes a bitmask.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agolib: objagg: Use struct_size() in kzalloc()
Gustavo A. R. Silva [Wed, 5 Jun 2019 14:45:16 +0000 (09:45 -0500)]
lib: objagg: Use struct_size() in kzalloc()

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 objagg_stats {
...
        struct objagg_obj_stats_info stats_info[];
};

size = sizeof(*objagg_stats) + sizeof(objagg_stats->stats_info[0]) * count;
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, stats_info, 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>
5 years agoinet_connection_sock: remove unused parameter of reqsk_queue_unlink func
Zhiqiang Liu [Wed, 5 Jun 2019 10:49:49 +0000 (18:49 +0800)]
inet_connection_sock: remove unused parameter of reqsk_queue_unlink func

small cleanup: "struct request_sock_queue *queue" parameter of reqsk_queue_unlink
func is never used in the func, so we can remove it.

Signed-off-by: Zhiqiang Liu <liuzhiqiang26@huawei.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: remove state PHY_FORCING
Heiner Kallweit [Tue, 4 Jun 2019 21:02:34 +0000 (23:02 +0200)]
net: phy: remove state PHY_FORCING

In the early days of phylib we had a functionality that changed to the
next lower speed in fixed mode if no link was established after a
certain period of time. This functionality has been removed years ago,
and state PHY_FORCING isn't needed any longer. Instead we can go from
UP to RUNNING or NOLINK directly (same as in autoneg mode).

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: rds: add per rds connection cache statistics
Zhu Yanjun [Mon, 3 Jun 2019 04:28:01 +0000 (00:28 -0400)]
net: rds: add per rds connection cache statistics

The variable cache_allocs is to indicate how many frags (KiB) are in one
rds connection frag cache.
The command "rds-info -Iv" will output the rds connection cache
statistics as below:
"
RDS IB Connections:
      LocalAddr RemoteAddr Tos SL  LocalDev            RemoteDev
      1.1.1.14 1.1.1.14   58 255  fe80::2:c903:a:7a31 fe80::2:c903:a:7a31
      send_wr=256, recv_wr=1024, send_sge=8, rdma_mr_max=4096,
      rdma_mr_size=257, cache_allocs=12
"
This means that there are about 12KiB frag in this rds connection frag
cache.
Since rds.h in rds-tools is not related with the kernel rds.h, the change
in kernel rds.h does not affect rds-tools.
rds-info in rds-tools 2.0.5 and 2.0.6 is tested with this commit. It works
well.

Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'dwmac-mediatek'
David S. Miller [Thu, 6 Jun 2019 00:03:14 +0000 (17:03 -0700)]
Merge branch 'dwmac-mediatek'

Biao Huang says:

====================
complete dwmac-mediatek driver and fix flow control issue

Changes in v2:
        patch#1: there is no extra action in mediatek_dwmac_remove, remove it

v1:
This series mainly complete dwmac-mediatek driver:
        1. add power on/off operations for dwmac-mediatek.
        2. disable rx watchdog to reduce rx path reponding time.
        3. change the default value of tx-frames from 25 to 1, so
           ptp4l will test pass by default.

and also fix the issue that flow control won't be disabled any more
once being enabled.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: dwmac4: fix flow control issue
Biao Huang [Mon, 3 Jun 2019 01:58:06 +0000 (09:58 +0800)]
net: stmmac: dwmac4: fix flow control issue

Current dwmac4_flow_ctrl will not clear
GMAC_RX_FLOW_CTRL_RFE/GMAC_RX_FLOW_CTRL_RFE bits,
so MAC hw will keep flow control on although expecting
flow control off by ethtool. Add codes to fix it.

Fixes: 477286b53f55 ("stmmac: add GMAC4 core support")
Signed-off-by: Biao Huang <biao.huang@mediatek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: modify default value of tx-frames
Biao Huang [Mon, 3 Jun 2019 01:58:05 +0000 (09:58 +0800)]
net: stmmac: modify default value of tx-frames

the default value of tx-frames is 25, it's too late when
passing tstamp to stack, then the ptp4l will fail:

ptp4l -i eth0 -f gPTP.cfg -m
ptp4l: selected /dev/ptp0 as PTP clock
ptp4l: port 1: INITIALIZING to LISTENING on INITIALIZE
ptp4l: port 0: INITIALIZING to LISTENING on INITIALIZE
ptp4l: port 1: link up
ptp4l: timed out while polling for tx timestamp
ptp4l: increasing tx_timestamp_timeout may correct this issue,
       but it is likely caused by a driver bug
ptp4l: port 1: send peer delay response failed
ptp4l: port 1: LISTENING to FAULTY on FAULT_DETECTED (FT_UNSPECIFIED)

ptp4l tests pass when changing the tx-frames from 25 to 1 with
ethtool -C option.
It should be fine to set tx-frames default value to 1, so ptp4l will pass
by default.

Signed-off-by: Biao Huang <biao.huang@mediatek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: dwmac-mediatek: disable rx watchdog
Biao Huang [Mon, 3 Jun 2019 01:58:04 +0000 (09:58 +0800)]
net: stmmac: dwmac-mediatek: disable rx watchdog

disable rx watchdog for dwmac-mediatek, then the hw will
issue a rx interrupt once receiving a packet, so the responding time
for rx path will be reduced.

Signed-off-by: Biao Huang <biao.huang@mediatek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: dwmac-mediatek: enable Ethernet power domain
Biao Huang [Mon, 3 Jun 2019 01:58:03 +0000 (09:58 +0800)]
net: stmmac: dwmac-mediatek: enable Ethernet power domain

add Ethernet power on/off operations in init/exit flow.

Signed-off-by: Biao Huang <biao.huang@mediatek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agodrivers: net: vxlan: drop unneeded likely() call around IS_ERR()
Enrico Weigelt [Wed, 5 Jun 2019 21:19:13 +0000 (23:19 +0200)]
drivers: net: vxlan: drop unneeded likely() call around IS_ERR()

IS_ERR() already calls unlikely(), so this extra likely() call
around the !IS_ERR() is not needed.

Signed-off-by: Enrico Weigelt <info@metux.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ipv6: drop unneeded likely() call around IS_ERR()
Enrico Weigelt [Wed, 5 Jun 2019 21:11:34 +0000 (23:11 +0200)]
net: ipv6: drop unneeded likely() call around IS_ERR()

IS_ERR() already calls unlikely(), so this extra unlikely() call
around IS_ERR() is not needed.

Signed-off-by: Enrico Weigelt <info@metux.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ipv4: drop unneeded likely() call around IS_ERR()
Enrico Weigelt [Wed, 5 Jun 2019 21:09:05 +0000 (23:09 +0200)]
net: ipv4: drop unneeded likely() call around IS_ERR()

IS_ERR() already calls unlikely(), so this extra unlikely() call
around IS_ERR() is not needed.

Signed-off-by: Enrico Weigelt <info@metux.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: openvswitch: drop unneeded likely() call around IS_ERR()
Enrico Weigelt [Wed, 5 Jun 2019 21:06:40 +0000 (23:06 +0200)]
net: openvswitch: drop unneeded likely() call around IS_ERR()

IS_ERR() already calls unlikely(), so this extra likely() call
around the !IS_ERR() is not needed.

Signed-off-by: Enrico Weigelt <info@metux.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: socket: drop unneeded likely() call around IS_ERR()
Enrico Weigelt [Wed, 5 Jun 2019 20:58:50 +0000 (22:58 +0200)]
net: socket: drop unneeded likely() call around IS_ERR()

IS_ERR() already calls unlikely(), so this extra likely() call
around the !IS_ERR() is not needed.

Signed-off-by: Enrico Weigelt <info@metux.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: flower: use struct_size() helper
Gustavo A. R. Silva [Wed, 5 Jun 2019 20:38:27 +0000 (15:38 -0500)]
nfp: flower: use struct_size() helper

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 nfp_tun_active_tuns {
...
        struct route_ip_info {
                __be32 ipv4;
                __be32 egress_port;
                __be32 extra[2];
        } tun_info[];
};

Make use of the struct_size() helper instead of an open-coded version
in order to avoid any potential type mistakes.

So, replace the following form:

sizeof(struct nfp_tun_active_tuns) + sizeof(struct route_ip_info) * count

with:

struct_size(payload, tun_info, count)

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>
5 years agoi40e: Check and set the PF driver state first in i40e_ndo_set_vf_mac
Lihong Yang [Wed, 5 Jun 2019 19:45:16 +0000 (12:45 -0700)]
i40e: Check and set the PF driver state first in i40e_ndo_set_vf_mac

The PF driver state flag __I40E_VIRTCHNL_OP_PENDING needs to be
checked and set at the beginning of i40e_ndo_set_vf_mac. Otherwise,
if there are error conditions before it, the flag will be cleared
unexpectedly by this function to cause potential race conditions.
Hence move the check to the top of this function.

Signed-off-by: Lihong Yang <lihong.yang@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoi40e: Do not check VF state in i40e_ndo_get_vf_config
Lihong Yang [Wed, 5 Jun 2019 19:45:15 +0000 (12:45 -0700)]
i40e: Do not check VF state in i40e_ndo_get_vf_config

The VF configuration returned in i40e_ndo_get_vf_config is
already stored by the PF. There is no dependency on any
specific state of the VF to return the configuration.
Drop the check against I40E_VF_STATE_INIT since it is not
needed.

Signed-off-by: Lihong Yang <lihong.yang@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch '10GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Wed, 5 Jun 2019 23:44:14 +0000 (16:44 -0700)]
Merge branch '10GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
10GbE Intel Wired LAN Driver Updates 2019-06-05

This series contains updates to mainly ixgbe, with a few updates to
i40e, net, ice and hns2 driver.

Jan adds support for tracking each queue pair for whether or not AF_XDP
zero copy is enabled.  Also updated the ixgbe driver to use the
netdev-provided umems so that we do not need to contain these structures
in our own adapter structure.

William Tu provides two fixes for AF_XDP statistics which were causing
incorrect counts.

Jake reduces the PTP transmit timestamp timeout from 15 seconds to 1 second,
which is still well after the maximum expected delay.  Also fixes an
issues with the PTP SDP pin setup which was not properly aligning on a
full second, so updated the code to account for the cyclecounter
multiplier and simplify the code to make the intent of the calculations
more clear.  Updated the function header comments to help with the code
documentation.  Added support for SDP/PPS output for x550 devices, which
is slightly different than x540 devices that currently have this
support.

Anirudh adds a new define for Link Layer Discovery Protocol to the
networking core, so that drivers do not have to create and use their own
definitions.  In addition, update all the drivers currently defining
their own LLDP define to use the new networking core define.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ixgbevf: fix a missing check of ixgbevf_write_msg_read_ack
Kangjie Lu [Sat, 9 Mar 2019 03:31:23 +0000 (21:31 -0600)]
net: ixgbevf: fix a missing check of ixgbevf_write_msg_read_ack

If ixgbevf_write_msg_read_ack fails, return its error code upstream

Signed-off-by: Kangjie Lu <kjlu@umn.edu>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Reviewed-by: Mukesh Ojha <mojha@codeaurora.org>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoixgbe: implement support for SDP/PPS output on X550 hardware
Jacob Keller [Fri, 12 Apr 2019 15:33:19 +0000 (08:33 -0700)]
ixgbe: implement support for SDP/PPS output on X550 hardware

Similar to the X540 hardware, enable support for generating a 1pps
output signal on SDP0.

This support is slightly different to the X540 hardware, because of the
register layout changes. First, the system time register is now
represented in 'cycles' and 'billions of cycles'. Second, we need to
also program the TSSDP register, as well as the ESDP register. Third,
the clock output uses only FREQOUT, instead of a full 64bit value for
the output clock period. Finally, we have to use the ST0 bit instead of
the SYNCLK bit in the TSAUXC register.

This support should work even for the hardware with a higher frequency
clock, as it carefully takes into account the multiply and shift of the
cycle counter used.

We also set the pps configuration to 1, since we now support generating
a pulse per second output.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agonet: hns3: Use LLDP ethertype define ETH_P_LLDP
Anirudh Venkataramanan [Thu, 11 Apr 2019 16:11:36 +0000 (09:11 -0700)]
net: hns3: Use LLDP ethertype define ETH_P_LLDP

Remove references to HCLGE_MAC_ETHERTYPE_LLDP and use ETH_P_LLDP instead.

Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoice: Use LLDP ethertype define ETH_P_LLDP
Jeff Kirsher [Wed, 5 Jun 2019 18:14:02 +0000 (11:14 -0700)]
ice: Use LLDP ethertype define ETH_P_LLDP

Instead of using a local define for the LLDP ethertype, use the kernel
define ETH_P_LLDP.

Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoixgbe: Use LLDP ethertype define ETH_P_LLDP
Anirudh Venkataramanan [Thu, 11 Apr 2019 16:11:35 +0000 (09:11 -0700)]
ixgbe: Use LLDP ethertype define ETH_P_LLDP

Remove references to IXGBE_ETH_P_LLD and use ETH_P_LLDP instead.

Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoi40e: Use LLDP ethertype define ETH_P_LLDP
Anirudh Venkataramanan [Thu, 11 Apr 2019 16:11:34 +0000 (09:11 -0700)]
i40e: Use LLDP ethertype define ETH_P_LLDP

Remove references to I40E_ETH_P_LLDP and use ETH_P_LLDP instead.

Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agonet: Add a define for LLDP ethertype
Anirudh Venkataramanan [Thu, 11 Apr 2019 16:11:33 +0000 (09:11 -0700)]
net: Add a define for LLDP ethertype

Add a new define ETH_P_LLDP for Link Layer Discovery Protocol (LLDP)
ethertype.

Suggested-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoixgbe: add a kernel documentation comment for ixgbe_ptp_get_ts_config
Jacob Keller [Mon, 8 Apr 2019 23:52:02 +0000 (16:52 -0700)]
ixgbe: add a kernel documentation comment for ixgbe_ptp_get_ts_config

This function was missing a documentation comment. Add one now.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoixgbe: use 'cc' instead of 'hw_cc' for local variable
Jacob Keller [Mon, 8 Apr 2019 23:52:01 +0000 (16:52 -0700)]
ixgbe: use 'cc' instead of 'hw_cc' for local variable

The ixgbe_ptp.c file sometimes uses hw_cc as the local variable for the
cycle counter in ixgbe_ptp_read_X550. However, we use just 'cc' as
a local variable for this by convention else where in the file.

Convert this lone usage of 'hw_cc' into just the shorter 'cc' name to
match the other read functions in the file.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoixgbe: fix PTP SDP pin setup on X540 hardware
Jacob Keller [Mon, 8 Apr 2019 23:51:59 +0000 (16:51 -0700)]
ixgbe: fix PTP SDP pin setup on X540 hardware

The function ixgbe_ptp_setup_sdp_X540 attempts to program a software
defined pin, in order to generate a pulse-per-second output on SDP 0.

It does work to generate the output, but does not align the output on
the full second. Additionally, it does not take into account the
cyclecounter multiplier. This leads to somewhat confusing code which is
likely to be incorrect if blindly copied to another hardware type.

Update this code to account for the cyclecounter multiplier, and to
directly use timecounter_read.

This change ensures that the SDP output will align properly on a full
second, and makes the intent of the calculations a bit more clear.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoixgbe: reduce PTP Tx timestamp timeout to 1 second
Jacob Keller [Mon, 8 Apr 2019 23:51:58 +0000 (16:51 -0700)]
ixgbe: reduce PTP Tx timestamp timeout to 1 second

Previously we waited for a whole 15 seconds before we cleared the Tx
timestamp state. This is astronomically long compared to the worst case
timings expected by our devices. In addition, this is longer than the
wait in ptp4l when it detects a fault (caused by missing Tx timestamps).
Thus, reduce the timer to only 1 second, which is well after the maximum
expected delay. This should reduce user frustration when a timestamp
does get dropped for some reason.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoixgbe: fix AF_XDP tx packet count
William Tu [Thu, 4 Apr 2019 16:36:55 +0000 (09:36 -0700)]
ixgbe: fix AF_XDP tx packet count

The total_packets count at ixgbe_clean_xdp_tx_irq is
always zero when testing with xdpsock -t -N. Set the gso_segs
to 1 to make the tx packet count correct.

Signed-off-by: William Tu <u9012063@gmail.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoixgbe: fix AF_XDP tx byte count
William Tu [Thu, 4 Apr 2019 16:36:54 +0000 (09:36 -0700)]
ixgbe: fix AF_XDP tx byte count

The tx bytecount is done twice.  When running
'./xdpsock -t -N -i eth3' and 'ip -s link show dev eth3'
The avg packet size is 120 instead of 60. So remove the
extra one.

Signed-off-by: William Tu <u9012063@gmail.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoixgbe: remove umem from adapter
Jan Sokolowski [Fri, 22 Mar 2019 21:16:38 +0000 (14:16 -0700)]
ixgbe: remove umem from adapter

As current implementation of netdev already contains and provides
umems for us, we no longer have the need to contain these
structures in ixgbe_adapter.

Refactor the code to operate on netdev-provided umems.

Signed-off-by: Jan Sokolowski <jan.sokolowski@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agoixgbe: add tracking of AF_XDP zero-copy state for each queue pair
Jan Sokolowski [Fri, 22 Mar 2019 21:16:37 +0000 (14:16 -0700)]
ixgbe: add tracking of AF_XDP zero-copy state for each queue pair

Here, we add a bitmap to the ixgbe_adapter that tracks if a
certain queue pair has been "zero-copy enabled" via the ndo_bpf.
The bitmap is used in ixgbe_xsk_umem, and enables zero-copy if
and only if XDP is enabled, the corresponding qid in the bitmap
is set, and the umem is non-NULL;

Signed-off-by: Jan Sokolowski <jan.sokolowski@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
5 years agonet: fec_ptp: Use dev_err() instead of pr_err()
Fabio Estevam [Wed, 5 Jun 2019 13:10:35 +0000 (10:10 -0300)]
net: fec_ptp: Use dev_err() instead of pr_err()

dev_err() is more appropriate for printing error messages inside
drivers, so switch to dev_err().

Signed-off-by: Fabio Estevam <festevam@gmail.com>
Acked-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'r8169-factor-out-firmware-handling'
David S. Miller [Wed, 5 Jun 2019 18:41:59 +0000 (11:41 -0700)]
Merge branch 'r8169-factor-out-firmware-handling'

Heiner Kallweit says:

====================
r8169: factor out firmware handling

Let's factor out firmware handling into a separate source code file.
This simplifies reading the code and makes clearer what the interface
between driver and firmware handling is.

v2:
- fix small whitespace issue in patch 2
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agor8169: factor out firmware handling
Heiner Kallweit [Wed, 5 Jun 2019 06:02:31 +0000 (08:02 +0200)]
r8169: factor out firmware handling

Let's factor out firmware handling into a separate source code file.
This simplifies reading the code and makes clearer what the interface
between driver and firmware handling is.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agor8169: rename r8169.c to r8169_main.c
Heiner Kallweit [Wed, 5 Jun 2019 05:59:57 +0000 (07:59 +0200)]
r8169: rename r8169.c to r8169_main.c

In preparation of factoring out firmware handling rename r8169.c to
r8169_main.c.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ethernet: mediatek: fix mtk_eth_soc build errors & warnings
Randy Dunlap [Wed, 5 Jun 2019 05:52:10 +0000 (22:52 -0700)]
net: ethernet: mediatek: fix mtk_eth_soc build errors & warnings

Fix build errors in Mediatek mtk_eth_soc driver.

It looks like these 3 source files were meant to be linked together
since 2 of them are library-like functions,
but they are currently being built as 3 loadable modules.

Fixes these build errors:

  WARNING: modpost: missing MODULE_LICENSE() in drivers/net/ethernet/mediatek/mtk_eth_path.o
  WARNING: modpost: missing MODULE_LICENSE() in drivers/net/ethernet/mediatek/mtk_sgmii.o
  ERROR: "mtk_sgmii_init" [drivers/net/ethernet/mediatek/mtk_eth_soc.ko] undefined!
  ERROR: "mtk_setup_hw_path" [drivers/net/ethernet/mediatek/mtk_eth_soc.ko] undefined!
  ERROR: "mtk_sgmii_setup_mode_force" [drivers/net/ethernet/mediatek/mtk_eth_soc.ko] undefined!
  ERROR: "mtk_sgmii_setup_mode_an" [drivers/net/ethernet/mediatek/mtk_eth_soc.ko] undefined!
  ERROR: "mtk_w32" [drivers/net/ethernet/mediatek/mtk_eth_path.ko] undefined!
  ERROR: "mtk_r32" [drivers/net/ethernet/mediatek/mtk_eth_path.ko] undefined!

This changes the loadable module name from mtk_eth_soc to mtk_eth.
I didn't see a way to leave it as mtk_eth_soc.

Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Sean Wang <sean.wang@mediatek.com>
Cc: John Crispin <blogic@openwrt.org>
Cc: Felix Fietkau <nbd@openwrt.org>
Cc: Nelson Chang <nelson.chang@mediatek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-dsa-mv88e6xxx-support-for-mv88e6250'
David S. Miller [Wed, 5 Jun 2019 03:07:57 +0000 (20:07 -0700)]
Merge branch 'net-dsa-mv88e6xxx-support-for-mv88e6250'

Rasmus Villemoes says:

====================
net: dsa: mv88e6xxx: support for mv88e6250

This adds support for the mv88e6250 chip. Initially based on the
mv88e6240, this time around, I've been through each ->ops callback and
checked that it makes sense, either replacing with a 6250 specific
variant or dropping it if no equivalent functionality seems to exist
for the 6250. Along the way, I found a few oddities in the existing
code, mostly sent as separate patches/questions.

The one relevant to the 6250 is the ieee_pri_map callback, where the
existing mv88e6085_g1_ieee_pri_map() is actually wrong for many of the
existing users. I've put the mv88e6250_g1_ieee_pri_map() patch first
in case some of the existing chips get switched over to use that and
it is deemed important enough for -stable.

v4:
- fix style issue in 1/10
- add Andrew's reviewed-by to 1,6,7,8,9,10.

v3:
- rebase on top of net-next/master
- add reviewed-bys to patches unchanged from v2 (2,3,4,5)
- add 6250-specific ->ieee_pri_map, ->port_set_speed, ->port_link_state (1,6,7)
- in addition, use mv88e6065_phylink_validate for ->phylink_validate,
  and don't implement ->port_get_cmode, ->port_set_jumbo_size,
  ->port_disable_learn_limit, ->rmu_disable
- drop ptp support
- add patch adding the compatible string to the DT binding (9)
- add small refactoring patch (10)

v2:
- rebase on top of net-next/master
- add reviewed-by to two patches unchanged from v1 (2,3)
- add separate watchdog_ops
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: mv88e6xxx: refactor mv88e6352_g1_reset
Rasmus Villemoes [Tue, 4 Jun 2019 07:34:35 +0000 (07:34 +0000)]
net: dsa: mv88e6xxx: refactor mv88e6352_g1_reset

The new mv88e6250_g1_reset() is identical to mv88e6352_g1_reset() except
for the call of mv88e6352_g1_wait_ppu_polling(), so refactor the 6352
version in term of the 6250 one. No functional change.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agodt-bindings: net: dsa: marvell: add "marvell,mv88e6250" compatible string
Rasmus Villemoes [Tue, 4 Jun 2019 07:34:33 +0000 (07:34 +0000)]
dt-bindings: net: dsa: marvell: add "marvell,mv88e6250" compatible string

The mv88e6250 has port_base_addr 0x8 or 0x18 (depending on
configuration pins), so it constitutes a new family and hence needs
its own compatible string.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: mv88e6xxx: add support for mv88e6250
Rasmus Villemoes [Tue, 4 Jun 2019 07:34:32 +0000 (07:34 +0000)]
net: dsa: mv88e6xxx: add support for mv88e6250

This adds support for the Marvell 88E6250. I've checked that each
member in the ops-structure makes sense, and basic switchdev
functionality works fine.

It uses the new dual_chip option, and since its port registers start
at SMI address 0x08 or 0x18 (i.e., always sw_addr + 0x08), we need to
introduce a new compatible string in order for the auto-identification
in mv88e6xxx_detect() to work.

The chip has four per port 16-bits statistics registers, two of which
correspond to the existing "sw_in_filtered" and "sw_out_filtered" (but
at offsets 0x13 and 0x10 rather than 0x12 and 0x13, because why should
this be easy...). Wiring up those four statistics seems to require
introducing a STATS_TYPE_PORT_6250 bit or similar, which seems a tad
ugly, so for now this just allows access to the STATS_TYPE_BANK0 ones.

The chip does have ptp support, and the existing
mv88e6352_{gpio,avb,ptp}_ops at first glance seem like they would work
out-of-the-box, but for simplicity (and lack of testing) I'm eliding
this.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: mv88e6xxx: implement port_link_state for mv88e6250
Rasmus Villemoes [Tue, 4 Jun 2019 07:34:30 +0000 (07:34 +0000)]
net: dsa: mv88e6xxx: implement port_link_state for mv88e6250

The mv88e6250 has a rather different way of reporting the link, speed
and duplex status. A simple difference is that the link bit is bit 12
rather than bit 11 of the port status register.

It gets more complicated for speed and duplex, which do not have
separate fields. Instead, there's a four-bit PortMode field, and
decoding that depends on whether it's a phy or mii port. For the phy
ports, only four of the 16 values have defined meaning; the rest are
called "reserved", so returning {SPEED,DUPLEX}_UNKNOWN seems
reasonable.

For the mii ports, most possible values are documented (0x3 and 0x5
are reserved), but I'm unable to make sense of them all. Since the
bits simply reflect the Px_MODE[3:0] configuration pins, just support
the subset that I'm certain about. Support for other setups can be
added later.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: mv88e6xxx: implement port_set_speed for mv88e6250
Rasmus Villemoes [Tue, 4 Jun 2019 07:34:29 +0000 (07:34 +0000)]
net: dsa: mv88e6xxx: implement port_set_speed for mv88e6250

The data sheet also mentions the possibility of selecting 200 Mbps for
the MII ports (ports 5 and 6) by setting the ForceSpd field to
0x2 (aka MV88E6065_PORT_MAC_CTL_SPEED_200). However, there's a note
that "actual speed is determined by bit 8 above", and flipping back a
page, one finds that bits 13:8 are reserved...

So without further information on what bit 8 means, let's stick to
supporting just 10 and 100 Mbps on all ports.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: mv88e6xxx: implement watchdog_ops for mv88e6250
Rasmus Villemoes [Tue, 4 Jun 2019 07:34:28 +0000 (07:34 +0000)]
net: dsa: mv88e6xxx: implement watchdog_ops for mv88e6250

The MV88E6352_G2_WDOG_CTL_* bits almost, but not quite, describe the
watchdog control register on the mv88e6250. Among those actually
referenced in the code, only QC_ENABLE differs (bit 6 rather than bit
5).

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: mv88e6xxx: implement vtu_getnext and vtu_loadpurge for mv88e6250
Rasmus Villemoes [Tue, 4 Jun 2019 07:34:27 +0000 (07:34 +0000)]
net: dsa: mv88e6xxx: implement vtu_getnext and vtu_loadpurge for mv88e6250

These are almost identical to the 6185 variants, but have fewer bits
for the FID.

Bit 10 of the VTU_OP register (offset 0x05) is the VidPolicy bit,
which one should probably preserve in mv88e6xxx_g1_vtu_op(), instead
of always writing a 0. However, on the 6352 family, that bit is
located at bit 12 in the VTU FID register (offset 0x02), and is always
unconditionally cleared by the mv88e6xxx_g1_vtu_fid_write()
function.

Since nothing in the existing driver seems to know or care about that
bit, it seems reasonable to not add the boilerplate to preserve it for
the 6250 (which would require adding a chip-specific vtu_op function,
or adding chip-quirks to the existing one).

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: mv88e6xxx: prepare mv88e6xxx_g1_atu_op() for the mv88e6250
Rasmus Villemoes [Tue, 4 Jun 2019 07:34:25 +0000 (07:34 +0000)]
net: dsa: mv88e6xxx: prepare mv88e6xxx_g1_atu_op() for the mv88e6250

All the currently supported chips have .num_databases either 256 or
4096, so this patch does not change behaviour for any of those. The
mv88e6250, however, has .num_databases == 64, and it does not put the
upper two bits in ATU control 13:12, but rather in ATU Operation
9:8. So change the logic to prepare for supporting mv88e6250.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Vivien Didelot <vivien.didelot@gmail.com>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: mv88e6xxx: introduce support for two chips using direct smi addressing
Rasmus Villemoes [Tue, 4 Jun 2019 07:34:24 +0000 (07:34 +0000)]
net: dsa: mv88e6xxx: introduce support for two chips using direct smi addressing

The 88e6250 (as well as 6220, 6071, 6070, 6020) do not support
multi-chip (indirect) addressing. However, one can still have two of
them on the same mdio bus, since the device only uses 16 of the 32
possible addresses, either addresses 0x00-0x0F or 0x10-0x1F depending
on the ADDR4 pin at reset [since ADDR4 is internally pulled high, the
latter is the default].

In order to prepare for supporting the 88e6250 and friends, introduce
mv88e6xxx_info::dual_chip to allow having a non-zero sw_addr while
still using direct addressing.

Reviewed-by: Vivien Didelot <vivien.didelot@gmail.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: dsa: mv88e6xxx: add mv88e6250_g1_ieee_pri_map
Rasmus Villemoes [Tue, 4 Jun 2019 07:34:23 +0000 (07:34 +0000)]
net: dsa: mv88e6xxx: add mv88e6250_g1_ieee_pri_map

Quite a few of the existing supported chips that use
mv88e6085_g1_ieee_pri_map as ->ieee_pri_map (including, incidentally,
mv88e6085 itself) actually have a reset value of 0xfa50 in the
G1_IEEE_PRI register.

The data sheet for the mv88e6095, however, does describe a reset value
of 0xfa41.

So rather than changing the value in the existing callback, introduce
a new variant with the 0xfa50 value. That will be used by the upcoming
mv88e6250, and existing chips can be switched over one by one,
preferably double-checking both the data sheet and actual hardware in
each case - if anybody actually feels this is important enough to
care.

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agovmxnet3: turn off lro when rxcsum is disabled
Ronak Doshi [Tue, 4 Jun 2019 06:58:38 +0000 (23:58 -0700)]
vmxnet3: turn off lro when rxcsum is disabled

Currently, when rx csum is disabled, vmxnet3 driver does not turn
off lro, which can cause performance issues if user does not turn off
lro explicitly. This patch adds fix_features support which is used to
turn off LRO whenever RXCSUM is disabled.

Signed-off-by: Ronak Doshi <doshir@vmware.com>
Acked-by: Rishi Mehta <rmehta@vmware.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-add-struct-nexthop-to-fib-info'
David S. Miller [Wed, 5 Jun 2019 02:26:50 +0000 (19:26 -0700)]
Merge branch 'net-add-struct-nexthop-to-fib-info'

David Ahern says:

====================
net: add struct nexthop to fib{6}_info

Set 10 of 11 to improve route scalability via support for nexthops as
standalone objects for fib entries.
    https://lwn.net/Articles/763950/

This sets adds 'struct nexthop' to fib_info and fib6_info. IPv4
already handles multiple fib_nh entries in a single fib_info, so
the conversion to use a nexthop struct is fairly mechanical. IPv6
using a nexthop struct with a fib6_info impacts a lot of core logic
which is built around the assumption of a single, builtin fib6_nh
per fib6_info. To make this easier to review, this set adds
nexthop to fib6_info and adds checks in most places fib6_info is
used. The next set finishes the IPv6 conversion, walking through
the places that need to consider all fib6_nh within a nexthop struct.

Offload drivers - mlx5, mlxsw and rocker - are changed to fail FIB
entries using nexthop objects. That limitation can be removed once
the drivers are updated to properly support separate nexthops.

This set starts by adding accessors for fib_nh and fib_nhs in a
fib_info. This makes it easier to extract the number of nexthops
in the fib entry and a specific fib_nh once the entry references
a struct nexthop. Patch 2 converts more of IPv4 code to use
fib_nh_common allowing a struct nexthop to use a fib6_nh with an
IPv4 entry.

Patches 3 and 4 add 'struct nexthop' to fib{6}_info and update
references to both take a different path when it is set. New
exported functions are added to the nexthop code to validate a
nexthop struct when configured for use with a fib entry. IPv4
is allowed to use a nexthop with either v4 or v6 entries. IPv6
is limited to v6 entries only. In both cases list_heads track
the fib entries using a nexthop struct for fast correlation on
events (e.g., device events or nexthop events like delete or
replace).

The last 3 patches add hooks to drivers listening for FIB
notificationas. All 3 of them reject the routes as unsupported,
returning an error message to the user via extack. For mlxsw
at least this is a stop gap measure until the driver is updated for
proper support.

Functional tests for nexthops have already been committed. Those tests
will be active after the next patch set which makes the code paths
created by this set and the next one live.

Existing code paths moved to the else branch of 'if (f{6}i->nh)' checks
are covered by existing tests under selftests/net.

v3
- remove ip6_create_rt_rcu from ip6_pol_route in patch 4 and use pcpu
  routes for REJECT routes with the blackhole nexthop (request from Wei)

v2
- no code changes from v1
- commit messages for first 4 patches updated
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agorocker: Fail attempts to use routes with nexthop objects
David Ahern [Tue, 4 Jun 2019 03:19:55 +0000 (20:19 -0700)]
rocker: Fail attempts to use routes with nexthop objects

Fail attempts to use nexthop objects with routes until support can be
properly added.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlx5: Fail attempts to use routes with nexthop objects
David Ahern [Tue, 4 Jun 2019 03:19:54 +0000 (20:19 -0700)]
mlx5: Fail attempts to use routes with nexthop objects

Fail attempts to use nexthop objects with routes until support can be
properly added.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: Fail attempts to use routes with nexthop objects
David Ahern [Tue, 4 Jun 2019 03:19:53 +0000 (20:19 -0700)]
mlxsw: Fail attempts to use routes with nexthop objects

Fail attempts to use nexthop objects with routes until support can be
properly added.

Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: Plumb support for nexthop object in a fib6_info
David Ahern [Tue, 4 Jun 2019 03:19:52 +0000 (20:19 -0700)]
ipv6: Plumb support for nexthop object in a fib6_info

Add struct nexthop and nh_list list_head to fib6_info. nh_list is the
fib6_info side of the nexthop <-> fib_info relationship. Since a fib6_info
referencing a nexthop object can not have 'sibling' entries (the old way
of doing multipath routes), the nh_list is a union with fib6_siblings.

Add f6i_list list_head to 'struct nexthop' to track fib6_info entries
using a nexthop instance. Update __remove_nexthop_fib to walk f6_list
and delete fib entries using the nexthop.

Add a few nexthop helpers for use when a nexthop is added to fib6_info:
- nexthop_fib6_nh - return first fib6_nh in a nexthop object
- fib6_info_nh_dev moved to nexthop.h and updated to use nexthop_fib6_nh
  if the fib6_info references a nexthop object
- nexthop_path_fib6_result - similar to ipv4, select a path within a
  multipath nexthop object. If the nexthop is a blackhole, set
  fib6_result type to RTN_BLACKHOLE, and set the REJECT flag

Update the fib6_info references to check for nh and take a different path
as needed:
- rt6_qualify_for_ecmp - if a fib entry uses a nexthop object it can NOT
  be coalesced with other fib entries into a multipath route
- rt6_duplicate_nexthop - use nexthop_cmp if either fib6_info references
  a nexthop
- addrconf (host routes), RA's and info entries (anything configured via
  ndisc) does not use nexthop objects
- fib6_info_destroy_rcu - put reference to nexthop object
- fib6_purge_rt - drop fib6_info from f6i_list
- fib6_select_path - update to use the new nexthop_path_fib6_result when
  fib entry uses a nexthop object
- rt6_device_match - update to catch use of nexthop object as a blackhole
  and set fib6_type and flags.
- ip6_route_info_create - don't add space for fib6_nh if fib entry is
  going to reference a nexthop object, take a reference to nexthop object,
  disallow use of source routing
- rt6_nlmsg_size - add space for RTA_NH_ID
- add rt6_fill_node_nexthop to add nexthop data on a dump

As with ipv4, most of the changes push existing code into the else branch
of whether the fib entry uses a nexthop object.

Update the nexthop code to walk f6i_list on a nexthop deleted to remove
fib entries referencing it.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv4: Plumb support for nexthop object in a fib_info
David Ahern [Tue, 4 Jun 2019 03:19:51 +0000 (20:19 -0700)]
ipv4: Plumb support for nexthop object in a fib_info

Add 'struct nexthop' and nh_list list_head to fib_info. nh_list is the
fib_info side of the nexthop <-> fib_info relationship.

Add fi_list list_head to 'struct nexthop' to track fib_info entries
using a nexthop instance. Add __remove_nexthop_fib and add it to
__remove_nexthop to walk the new list_head and mark those fib entries
as dead when the nexthop is deleted.

Add a few nexthop helpers for use when a nexthop is added to fib_info:
- nexthop_cmp to determine if 2 nexthops are the same
- nexthop_path_fib_result to select a path for a multipath
  'struct nexthop'
- nexthop_fib_nhc to select a specific fib_nh_common within a
  multipath 'struct nexthop'

Update existing fib_info_nhc to use nexthop_fib_nhc if a fib_info uses
a 'struct nexthop', and mark fib_info_nh as only used for the non-nexthop
case.

Update the fib_info functions to check for fi->nh and take a different
path as needed:
- free_fib_info_rcu - put the nexthop object reference
- fib_release_info - remove the fib_info from the nexthop's fi_list
- nh_comp - use nexthop_cmp when either fib_info references a nexthop
  object
- fib_info_hashfn - use the nexthop id for the hashing vs the oif of
  each fib_nh in a fib_info
- fib_nlmsg_size - add space for the RTA_NH_ID attribute
- fib_create_info - verify nexthop reference can be taken, verify
  nexthop spec is valid for fib entry, and add fib_info to fi_list for
  a nexthop
- fib_select_multipath - use the new nexthop_path_fib_result to select a
  path when nexthop objects are used
- fib_table_lookup - if the 'struct nexthop' is a blackhole nexthop, treat
  it the same as a fib entry using 'blackhole'

The bulk of the changes are in fib_semantics.c and most of that is
moving the existing change_nexthops into an else branch.

Update the nexthop code to walk fi_list on a nexthop deleted to remove
fib entries referencing it.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv4: Prepare for fib6_nh from a nexthop object
David Ahern [Tue, 4 Jun 2019 03:19:50 +0000 (20:19 -0700)]
ipv4: Prepare for fib6_nh from a nexthop object

Convert more IPv4 code to use fib_nh_common over fib_nh to enable routes
to use a fib6_nh based nexthop. In the end, only code not using a
nexthop object in a fib_info should directly access fib_nh in a fib_info
without checking the famiy and going through fib_nh_common. Those
functions will be marked when it is not directly evident.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv4: Use accessors for fib_info nexthop data
David Ahern [Tue, 4 Jun 2019 03:19:49 +0000 (20:19 -0700)]
ipv4: Use accessors for fib_info nexthop data

Use helpers to access fib_nh and fib_nhs fields of a fib_info. Drop the
fib_dev macro which is an alias for the first nexthop. Replacements:

  fi->fib_dev    --> fib_info_nh(fi, 0)->fib_nh_dev
  fi->fib_nh     --> fib_info_nh(fi, 0)
  fi->fib_nh[i]  --> fib_info_nh(fi, i)
  fi->fib_nhs    --> fib_info_num_path(fi)

where fib_info_nh(fi, i) returns fi->fib_nh[nhsel] and fib_info_num_path
returns fi->fib_nhs.

Move the existing fib_info_nhc to nexthop.h and define the new ones
there. A later patch adds a check if a fib_info uses a nexthop object,
and defining the helpers in nexthop.h avoid circular header
dependencies.

After this all remaining open coded references to fi->fib_nhs and
fi->fib_nh are in:
- fib_create_info and helpers used to lookup an existing fib_info
  entry, and
- the netdev event functions fib_sync_down_dev and fib_sync_up.

The latter two will not be reused for nexthops, and the fib_create_info
will be updated to handle a nexthop in a fib_info.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: Always allocate pcpu memory in a fib6_nh
David Ahern [Tue, 4 Jun 2019 01:37:03 +0000 (18:37 -0700)]
ipv6: Always allocate pcpu memory in a fib6_nh

A recent commit had an unintended side effect with reject routes:
rt6i_pcpu is expected to always be initialized for all fib6_info except
the null entry. The commit mentioned below skips it for reject routes
and ends up leaking references to the loopback device. For example,

    ip netns add foo
    ip -netns foo li set lo up
    ip -netns foo -6 ro add blackhole 2001:db8:1::1
    ip netns exec foo ping6 2001:db8:1::1
    ip netns del foo

ends up spewing:
    unregister_netdevice: waiting for lo to become free. Usage count = 3

The fib_nh_common_init is not needed for reject routes (no ipv4 caching
or encaps), so move the alloc_percpu_gfp after it and adjust the goto label.

Fixes: f40b6ae2b612 ("ipv6: Move pcpu cached routes to fib6_nh")
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agohinic: add LRO support
Xue Chaojing [Tue, 4 Jun 2019 01:16:08 +0000 (01:16 +0000)]
hinic: add LRO support

This patch adds LRO support for the HiNIC driver.

Reported-by: kbuild test robot <lkp@intel.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Xue Chaojing <xuechaojing@huawei.com>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'bond-mpls'
David S. Miller [Tue, 4 Jun 2019 21:49:38 +0000 (14:49 -0700)]
Merge branch 'bond-mpls'

Ariel Levkovich says:

====================
Support MPLS features in bonding and vlan net devices

Netdevice HW MPLS features are not passed from device driver's netdevice to
upper netdevice, specifically VLAN and bonding netdevice which are created
by the kernel when needed.

This prevents enablement and usage of HW offloads, such as TSO and checksumming
for MPLS tagged traffic when running via VLAN or bonding interface.

The patches introduce changes to the initialization steps of the VLAN and bonding
netdevices to inherit the MPLS features from lower netdevices to allow the HW
offloads.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: vlan: Inherit MPLS features from parent device
Ariel Levkovich [Mon, 3 Jun 2019 22:36:47 +0000 (22:36 +0000)]
net: vlan: Inherit MPLS features from parent device

During the creation of the VLAN interface net device,
the various device features and offloads are being set based
on the parent device's features.
The code initiates the basic, vlan and encapsulation features
but doesn't address the MPLS features set and they remain blank.
As a result, all device offloads that have significant performance
effect are disabled for MPLS traffic going via this VLAN device such
as checksumming and TSO.

This patch makes sure that MPLS features are also set for the
VLAN device based on the parent which will allow HW offloads of
checksumming and TSO to be performed on MPLS tagged packets.

Signed-off-by: Ariel Levkovich <lariel@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: bonding: Inherit MPLS features from slave devices
Ariel Levkovich [Mon, 3 Jun 2019 22:36:46 +0000 (22:36 +0000)]
net: bonding: Inherit MPLS features from slave devices

When setting the bonding interface net device features,
the kernel code doesn't address the slaves' MPLS features
and doesn't inherit them.

Therefore, HW offloads that enhance performance such as
checksumming and TSO are disabled for MPLS tagged traffic
flowing via the bonding interface.

The patch add the inheritance of the MPLS features from the
slave devices with a similar logic to setting the bonding device's
VLAN and encapsulation features.

CC: Jay Vosburgh <j.vosburgh@gmail.com>
CC: Veaceslav Falico <vfalico@gmail.com>
CC: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Ariel Levkovich <lariel@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-tls-small-general-improvements'
David S. Miller [Tue, 4 Jun 2019 21:33:50 +0000 (14:33 -0700)]
Merge branch 'net-tls-small-general-improvements'

Jakub Kicinski says:

====================
net/tls: small general improvements

This series cleans up and improves the tls code, mostly the offload
parts.

First a slight performance optimization - avoiding unnecessary re-
-encryption of records in patch 1.  Next patch 2 makes the code
more resilient by checking for errors in skb_copy_bits().  Next
commit removes a warning which can be triggered in normal operation,
(especially for devices explicitly making use of the fallback path).
Next two paths change the condition checking around the call to
tls_device_decrypted() to make it easier to extend.  Remaining
commits are centered around reorganizing struct tls_context for
better cache utilization.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: don't pass version to tls_advance_record_sn()
Jakub Kicinski [Mon, 3 Jun 2019 22:17:05 +0000 (15:17 -0700)]
net/tls: don't pass version to tls_advance_record_sn()

All callers pass prot->version as the last parameter
of tls_advance_record_sn(), yet tls_advance_record_sn()
itself needs a pointer to prot.  Pass prot from callers.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: reorganize struct tls_context
Jakub Kicinski [Mon, 3 Jun 2019 22:17:04 +0000 (15:17 -0700)]
net/tls: reorganize struct tls_context

struct tls_context is slightly badly laid out.  If we reorder things
right we can save 16 bytes (320 -> 304) but also make all fast path
data fit into two cache lines (one read only and one read/write,
down from four cache lines).

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: use version from prot
Jakub Kicinski [Mon, 3 Jun 2019 22:17:03 +0000 (15:17 -0700)]
net/tls: use version from prot

ctx->prot holds the same information as per-direction contexts.
Almost all code gets TLS version from this structure, convert
the last two stragglers, this way we can improve the cache
utilization by moving the per-direction data into cold cache lines.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: don't re-check msg decrypted status in tls_device_decrypted()
Jakub Kicinski [Mon, 3 Jun 2019 22:17:02 +0000 (15:17 -0700)]
net/tls: don't re-check msg decrypted status in tls_device_decrypted()

tls_device_decrypted() is only called from decrypt_skb_update(),
when ctx->decrypted == false, there is no need to re-check the bit.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: don't look for decrypted frames on non-offloaded sockets
Jakub Kicinski [Mon, 3 Jun 2019 22:17:01 +0000 (15:17 -0700)]
net/tls: don't look for decrypted frames on non-offloaded sockets

If the RX config of a TLS socket is SW, there is no point iterating
over the fragments and checking if frame is decrypted.  It will
always be fully encrypted.  Note that in fully encrypted case
the function doesn't actually touch any offload-related state,
so it's safe to call for TLS_SW, today.  Soon we will introduce
code which can only be called for offloaded contexts.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: remove false positive warning
Jakub Kicinski [Mon, 3 Jun 2019 22:17:00 +0000 (15:17 -0700)]
net/tls: remove false positive warning

It's possible that TCP stack will decide to retransmit a packet
right when that packet's data gets acked, especially in presence
of packet reordering.  This means that packets may be in flight,
even though tls_device code has already freed their record state.
Make fill_sg_in() and in turn tls_sw_fallback() not generate a
warning in that case, and quietly proceed to drop such frames.

Make the exit path from tls_sw_fallback() drop monitor friendly,
for users to be able to troubleshoot dropped retransmissions.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: check return values from skb_copy_bits() and skb_store_bits()
Jakub Kicinski [Mon, 3 Jun 2019 22:16:59 +0000 (15:16 -0700)]
net/tls: check return values from skb_copy_bits() and skb_store_bits()

In light of recent bugs, we should make a better effort of
checking return values.  In theory none of the functions should
fail today.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: fully initialize the msg wrapper skb
Jakub Kicinski [Mon, 3 Jun 2019 22:16:58 +0000 (15:16 -0700)]
net/tls: fully initialize the msg wrapper skb

If strparser gets cornered into starting a new message from
an sk_buff which already has frags, it will allocate a new
skb to become the "wrapper" around the fragments of the
message.

This new skb does not inherit any metadata fields.  In case
of TLS offload this may lead to unnecessarily re-encrypting
the message, as skb->decrypted is not set for the wrapper skb.

Try to be conservative and copy all fields of old skb
strparser's user may reasonably need.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: mscc: ocelot: Fix some struct initializations
Nathan Chancellor [Mon, 3 Jun 2019 20:49:53 +0000 (13:49 -0700)]
net: mscc: ocelot: Fix some struct initializations

Clang warns:

drivers/net/ethernet/mscc/ocelot_ace.c:335:37: warning: suggest braces
around initialization of subobject [-Wmissing-braces]
        struct ocelot_vcap_u64 payload = { 0 };
                                           ^
                                           {}
drivers/net/ethernet/mscc/ocelot_ace.c:336:28: warning: suggest braces
around initialization of subobject [-Wmissing-braces]
        struct vcap_data data = { 0 };
                                  ^
                                  {}
drivers/net/ethernet/mscc/ocelot_ace.c:683:37: warning: suggest braces
around initialization of subobject [-Wmissing-braces]
        struct ocelot_ace_rule del_ace = { 0 };
                                           ^
                                           {}
drivers/net/ethernet/mscc/ocelot_ace.c:743:28: warning: suggest braces
around initialization of subobject [-Wmissing-braces]
        struct vcap_data data = { 0 };
                                  ^
                                  {}
4 warnings generated.

One way to fix these warnings is to add additional braces like Clang
suggests; however, there has been a bit of push back from some
maintainers[1][2], who just prefer memset as it is unambiguous, doesn't
depend on a particular compiler version[3], and properly initializes all
subobjects. Do that here so there are no more warnings.

[1]: https://lore.kernel.org/lkml/022e41c0-8465-dc7a-a45c-64187ecd9684@amd.com/
[2]: https://lore.kernel.org/lkml/20181128.215241.702406654469517539.davem@davemloft.net/
[3]: https://lore.kernel.org/lkml/20181116150432.2408a075@redhat.com/

Fixes: b596229448dd ("net: mscc: ocelot: Add support for tcam")
Link: https://github.com/ClangBuiltLinux/linux/issues/505
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>