linux-2.6-microblaze.git
5 years agonet: sched: flower: refactor reoffload for concurrent access
Vlad Buslov [Wed, 24 Apr 2019 06:53:31 +0000 (09:53 +0300)]
net: sched: flower: refactor reoffload for concurrent access

Recent changes that introduced unlocked flower did not properly account for
case when reoffload is initiated concurrently with filter updates. To fix
the issue, extend flower with 'hw_filters' list that is used to store
filters that don't have 'skip_hw' flag set. Filter is added to the list
when it is inserted to hardware and only removed from it after being
unoffloaded from all drivers that parent block is attached to. This ensures
that concurrent reoffload can still access filter that is being deleted and
prevents race condition when driver callback can be removed when filter is
no longer accessible trough idr, but is still present in hardware.

Refactor fl_change() to respect new filter reference counter and to release
filter reference with __fl_put() in case of error, instead of directly
deallocating filter memory. This allows for concurrent access to filter
from fl_reoffload() and protects it with reference counting. Refactor
fl_reoffload() to iterate over hw_filters list instead of idr. Implement
fl_get_next_hw_filter() helper function that is used to iterate over
hw_filters list with reference counting and skips filters that are being
concurrently deleted.

Fixes: 92149190067d ("net: sched: flower: set unlocked flag for flower proto ops")
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Reviewed-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: mvneta: Switch to using devm_alloc_etherdev_mqs
Rosen Penev [Tue, 23 Apr 2019 20:46:03 +0000 (13:46 -0700)]
net: mvneta: Switch to using devm_alloc_etherdev_mqs

It allows some of the code to be simplified.

Tested on Turris Omnia.

Signed-off-by: Rosen Penev <rosenp@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agotipc: tipc_udp_recv() cleanup vs rcu verbs
Eric Dumazet [Tue, 23 Apr 2019 16:24:46 +0000 (09:24 -0700)]
tipc: tipc_udp_recv() cleanup vs rcu verbs

First thing tipc_udp_recv() does is to use rcu_dereference_sk_user_data(),
and this is really hinting we already own rcu_read_lock() from the caller
(UDP stack).

No need to add another rcu_read_lock()/rcu_read_unlock() pair.

Also use rcu_dereference() instead of rcu_dereference_rtnl()
in the data path.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jon Maloy <jon.maloy@ericsson.com>
Cc: Ying Xue <ying.xue@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: Use result arg in fib_lookup_arg consistently
David Ahern [Wed, 24 Apr 2019 01:05:33 +0000 (18:05 -0700)]
ipv6: Use result arg in fib_lookup_arg consistently

arg.result is sometimes used as fib6_result and sometimes used to
hold the rt6_info. Add rt6_info to fib6_result and make the use
of arg.result consistent through ipv6 rules.

The rt6 entry is filled in for lookups returning a dst_entry, but not
for direct fib_lookups that just want a fib6_info.

Fixes: effda4dd97e8 ("ipv6: Pass fib6_result to fib lookups")
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: fib6_rule_action_alt needs to return -EAGAIN
David Ahern [Wed, 24 Apr 2019 01:06:30 +0000 (18:06 -0700)]
ipv6: fib6_rule_action_alt needs to return -EAGAIN

fib rule actions should return -EAGAIN for the rules to continue to the
next one. A recent change overwrote err with the lookup always returning
0 (future change will make it more like IPv4) which means the rules
stopped at the first (e.g., local table lookup only). Catch and reset err
to -EAGAIN.

Fixes: effda4dd97e87 ("ipv6: Pass fib6_result to fib lookups")
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_router: Prevent ipv6 gateway with v4 route via replace and append
David Ahern [Tue, 23 Apr 2019 23:03:29 +0000 (16:03 -0700)]
mlxsw: spectrum_router: Prevent ipv6 gateway with v4 route via replace and append

mlxsw currently does not support v6 gateways with v4 routes. Commit
19a9d136f198 ("ipv4: Flag fib_info with a fib_nh using IPv6 gateway")
prevents a route from being added, but nothing stops the replace or
append. Add a catch for them too.
    $ ip  ro add 172.16.2.0/24 via 10.99.1.2
    $ ip  ro replace 172.16.2.0/24 via inet6 fe80::202:ff:fe00:b dev swp1s0
    Error: mlxsw_spectrum: IPv6 gateway with IPv4 route is not supported.
    $ ip  ro append 172.16.2.0/24 via inet6 fe80::202:ff:fe00:b dev swp1s0
    Error: mlxsw_spectrum: IPv6 gateway with IPv4 route is not supported.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'Taprio-qdisc-fixes'
David S. Miller [Wed, 24 Apr 2019 02:52:33 +0000 (19:52 -0700)]
Merge branch 'Taprio-qdisc-fixes'

Andre Guedes says:

====================
Taprio qdisc fixes

I'm re-sending this series, now with the "net-next" prefix in the subject.

The only change from the previous version is in patch 3. As suggested by Cong
Wang, it removes the !entry check within should_restart_cycle() since it is
already checked by the caller. As a side effect, that function becomes a dummy
wrapper on list_is_last() so we simply remove it and call list_is_last()
instead.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: sched: taprio: Fix taprio_dequeue()
Andre Guedes [Tue, 23 Apr 2019 19:44:24 +0000 (12:44 -0700)]
net: sched: taprio: Fix taprio_dequeue()

In case we don't have 'guard' or 'budget' to transmit the skb, we should
continue traversing the qdisc list since the remaining guard/budget
might be enough to transmit a skb from other children qdiscs.

Fixes: 5a781ccbd19e (“tc: Add support for configuring the taprio scheduler”)
Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: sched: taprio: Fix taprio_peek()
Andre Guedes [Tue, 23 Apr 2019 19:44:23 +0000 (12:44 -0700)]
net: sched: taprio: Fix taprio_peek()

While traversing taprio's children qdisc list, if the gate is closed for
a given traffic class, we should continue traversing the list since the
remaining qdiscs may have skb ready for transmission.

This patch also takes this opportunity and changes the function to use
the TAPRIO_ALL_GATES_OPEN macro instead of the magic number '-1'.

Fixes: 5a781ccbd19e (“tc: Add support for configuring the taprio scheduler”)
Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: sched: taprio: Remove should_restart_cycle()
Andre Guedes [Tue, 23 Apr 2019 19:44:22 +0000 (12:44 -0700)]
net: sched: taprio: Remove should_restart_cycle()

The 'entry' argument from should_restart_cycle() cannot be NULL since it
is already checked by the caller so the WARN_ON() within should_
restart_cycle() could be removed.  By doing that, that function becomes
a dummy wrapper on list_is_last() so this patch simply gets rid of it
and call list_is_last() within advance_sched() instead.

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: sched: taprio: Refactor taprio_get_start_time()
Andre Guedes [Tue, 23 Apr 2019 19:44:21 +0000 (12:44 -0700)]
net: sched: taprio: Refactor taprio_get_start_time()

This patch does a code refactoring to taprio_get_start_time() function
to improve readability and report error properly.

If 'base' time is later than 'now', the start time is equal to 'base'
and taprio_get_start_time() is done. That's the natural case so we move
that code to the beginning of the function. Also, if 'cycle' calculation
is zero, something went really wrong with taprio and we should log that
internal error properly.

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: sched: taprio: Remove pointless variable assigment
Andre Guedes [Tue, 23 Apr 2019 19:44:20 +0000 (12:44 -0700)]
net: sched: taprio: Remove pointless variable assigment

This patch removes a pointless variable assigment in taprio_change().
The 'err' variable is not used from this assignment to the next one so
this patch removes it.

Signed-off-by: Andre Guedes <andre.guedes@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: Change nhc_flags to unsigned char
David Ahern [Tue, 23 Apr 2019 15:48:09 +0000 (08:48 -0700)]
net: Change nhc_flags to unsigned char

nhc_flags holds the RTNH_F flags for a given nexthop (fib{6}_nh).
All of the RTNH_F_ flags fit in an unsigned char, and since the API to
userspace (rtnh_flags and lower byte of rtm_flags) is 1 byte it can not
grow. Make nhc_flags in fib_nh_common an unsigned char and shrink the
size of the struct by 8, from 56 to 48 bytes.

Update the flags arguments for up netdevice events and fib_nexthop_info
which determines the RTNH_F flags to return on a dump/event. The RTNH_F
flags are passed in the lower byte of rtm_flags which is an unsigned int
so use a temp variable for the flags to fib_nexthop_info and combine
with rtm_flags in the caller.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agolwtunnel: Pass encap and encap type attributes to lwtunnel_fill_encap
David Ahern [Tue, 23 Apr 2019 15:23:41 +0000 (08:23 -0700)]
lwtunnel: Pass encap and encap type attributes to lwtunnel_fill_encap

Currently, lwtunnel_fill_encap hardcodes the encap and encap type
attributes as RTA_ENCAP and RTA_ENCAP_TYPE, respectively. The nexthop
objects want to re-use this code but the encap attributes passed to
userspace as NHA_ENCAP and NHA_ENCAP_TYPE. Since that is the only
difference, change lwtunnel_fill_encap to take the attribute type as
an input.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoravb: Avoid unsupported internal delay mode for R-Car E3/D3
Simon Horman [Tue, 23 Apr 2019 13:01:53 +0000 (15:01 +0200)]
ravb: Avoid unsupported internal delay mode for R-Car E3/D3

According to the R-Car Gen3 Hardware Manual Rev 1.50 of Nov 30, 2018, the
TX clock internal delay mode isn't supported on R-Car E3 (r8a77990) or D3
(r8a77995). And by extension it is also not supported by RZ/G2E (r9a774c0).

This matches all ES versions of the affected SoCs as it is
not clear if this problem will be resolved in newer chips.
This can be revisited, as necessary.

This patch does not error-out if PHY_INTERFACE_MODE_RGMII_ID or
PHY_INTERFACE_MODE_RGMII_TXID are used on SoCs where TX clock delay
mode is not supported as there is a risk of introducing a regression
when used in conjunction with older DT blobs present in the field.
Rather, a warning is logged in such cases.

Based on work by Kazuya Mizuguchi.

Signed-off-by: Simon Horman <horms+renesas@verge.net.au>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: fix sparc64 compilation of sock_gettstamp
Stephen Rothwell [Tue, 23 Apr 2019 07:25:24 +0000 (17:25 +1000)]
net: fix sparc64 compilation of sock_gettstamp

net/core/sock.c: In function 'sock_gettstamp':
net/core/sock.c:3007:23: error: expected '}' before ';' token
    .tv_sec = ts.tv_sec;
                       ^
net/core/sock.c:3011:4: error: expected ')' before 'return'
    return -EFAULT;
    ^~~~~~
net/core/sock.c:3013:2: error: expected expression before '}' token
  }
  ^

Fixes: c7cbdbf29f48 ("net: rework SIOCGSTAMP ioctl handling")
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoisdn:mISDN: fix misuse of %x in hfcpci.c
Fuqian Huang [Tue, 23 Apr 2019 02:56:23 +0000 (10:56 +0800)]
isdn:mISDN: fix misuse of %x in hfcpci.c

Pointers should be printed with %p or %px rather than
cast to (u_long) and printed with %lx.
Change %lx to %p to print the pointer.
Change %lx to %pad to print the dma_addr_t.

Signed-off-by: Fuqian Huang <huangfq.daxian@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoisdn: hisax: Fix misuse of %x in config.c
Fuqian Huang [Tue, 23 Apr 2019 02:25:28 +0000 (10:25 +0800)]
isdn: hisax: Fix misuse of %x in config.c

Pointers should be printed with %p or %px rather than
cast to (u_long) type and printed with %lX.
As the function seems to be for debug purpose.
Change %lX to %px to print the pointer value.

Signed-off-by: Fuqian Huang <huangfq.daxian@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'ipv6-fib6_ref-conversion-to-refcount_t'
David S. Miller [Wed, 24 Apr 2019 00:19:48 +0000 (17:19 -0700)]
Merge branch 'ipv6-fib6_ref-conversion-to-refcount_t'

Eric Dumazet says:

====================
ipv6: fib6_ref conversion to refcount_t

We are chasing use-after-free in IPv6 that could have their origin
in fib6_ref 0 -> 1 transitions.

This patch series should help finding the root causes if these
illegal transitions ever happen.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: convert fib6_ref to refcount_t
Eric Dumazet [Tue, 23 Apr 2019 01:35:03 +0000 (18:35 -0700)]
ipv6: convert fib6_ref to refcount_t

We suspect some issues involving fib6_ref 0 -> 1 transitions might
cause strange syzbot reports.

Lets convert fib6_ref to refcount_t to catch them earlier.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Wei Wang <weiwan@google.com>
Acked-by: Wei Wang <weiwan@google.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: broadly use fib6_info_hold() helper
Eric Dumazet [Tue, 23 Apr 2019 01:35:02 +0000 (18:35 -0700)]
ipv6: broadly use fib6_info_hold() helper

Instead of using atomic_inc(), prefer fib6_info_hold()
so that upcoming refcount_t conversion is simpler.

Only fib6_info_alloc() is using atomic_set() since we
just allocated a new object.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Wei Wang <weiwan@google.com>
Acked-by: Wei Wang <weiwan@google.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: fib6_info_destroy_rcu() cleanup
Eric Dumazet [Tue, 23 Apr 2019 01:35:01 +0000 (18:35 -0700)]
ipv6: fib6_info_destroy_rcu() cleanup

We do not need to clear f6i->rt6i_exception_bucket right before
freeing f6i.

Note that f6i->rt6i_exception_bucket is properly protected by
f6i->exception_bucket_flushed being set to one in rt6_flush_exceptions()
under the protection of rt6_exception_lock.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Wei Wang <weiwan@google.com>
Acked-by: Wei Wang <weiwan@google.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge tag 'mlx5-updates-2019-04-22' of git://git.kernel.org/pub/scm/linux/kernel...
David S. Miller [Wed, 24 Apr 2019 00:03:40 +0000 (17:03 -0700)]
Merge tag 'mlx5-updates-2019-04-22' of git://git./linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5-updates-2019-04-22

This series includes updates to mlx5e driver RX data path and some
significant XDP RX/TX improvements to overcome/mitigate HW and PCIE
bottlenecks.

From Tariq:
1) Some Enhancements in rq->flags
2) Stabilize RX packet rate (on Striding RQ) with
multiple outstanding UMR posts
In this patch, we add support for multiple outstanding UMR posts,
 to allow faster gap closure between consuming MPWQEs and reposting
them back into the WQ.

Performance test:
As expected, huge improvement in large-scale (48 cores).

xdp_redirect_map, 64B UDP multi-stream.
Redirect from ConnectX-5 100Gbps to ConnectX-6 100Gbps.
CPU: Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz.

Before: Unstable, 7 to 30 Mpps
After:  Stable,   at 70.5 Mpps

From Shay:
3) XDP, Inline small packets into the TX MPWQE in XDP xmit flow

Upon high packet rate with multiple CPUs TX workloads, much of the HCA's
resources are spent on prefetching TX descriptors, thus affecting
transmission rates.
This patch comes to mitigate this problem by moving some workload to the
CPU and reducing the HW data prefetch overhead for small packets (<= 256B).

When forwarding packets with XDP, a packet that is smaller
than a certain size (set to ~256 bytes) would be sent inline within
its WQE TX descrptor (mem-copied), when the hardware tx queue is congested
beyond a pre-defined water-mark.

Performance:
    Tested packet rate for UDP 64Byte multi-stream
    over two dual port ConnectX-5 100Gbps NICs.
    CPU: Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz

    * Tested with hyper-threading disabled

    XDP_TX:

    |          | before | after   |       |
    | 24 rings | 51Mpps | 116Mpps | +126% |
    | 1 ring   | 12Mpps | 12Mpps  | same  |

    XDP_REDIRECT:

    ** Below is the transmit rate, not the redirection rate
    which might be larger, and is not affected by this patch.

    |          | before  | after   |      |
    | 32 rings | 64Mpps  | 92Mpps  | +43% |
    | 1 ring   | 6.4Mpps | 6.4Mpps | same |

As we can see, feature significantly improves scaling, without
hurting single ring performance.

From Maxim:
4) Some trivial refactoring and code improvements prior to a larger series
to support AF_XDP.
====================

Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/mlx5e: Use #define for the WQE wait timeout constant
Maxim Mikityanskiy [Tue, 5 Mar 2019 12:46:16 +0000 (14:46 +0200)]
net/mlx5e: Use #define for the WQE wait timeout constant

Create a #define for the timeout of mlx5e_wait_for_min_rx_wqes to
clarify the meaning of a magic number.

Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5e: Remove unused rx_page_reuse stat
Maxim Mikityanskiy [Thu, 21 Mar 2019 13:22:57 +0000 (15:22 +0200)]
net/mlx5e: Remove unused rx_page_reuse stat

Remove the no longer used page_reuse stat of RQs.

Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5e: Take HW interrupt trigger into a function
Maxim Mikityanskiy [Fri, 1 Mar 2019 10:05:21 +0000 (12:05 +0200)]
net/mlx5e: Take HW interrupt trigger into a function

mlx5e_trigger_irq posts a NOP to the ICO SQ just to trigger an IRQ and
enter the NAPI poll on the right CPU according to the affinity. Use it
in mlx5e_activate_rq.

Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5e: Remove unused parameter
Maxim Mikityanskiy [Wed, 27 Mar 2019 11:46:50 +0000 (13:46 +0200)]
net/mlx5e: Remove unused parameter

mdev is unused in mlx5e_rx_is_linear_skb.

Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5e: Add an underflow warning comment
Maxim Mikityanskiy [Wed, 27 Mar 2019 11:39:21 +0000 (13:39 +0200)]
net/mlx5e: Add an underflow warning comment

mlx5e_mpwqe_get_log_rq_size calculates the number of WQEs (N) based on
the requested number of frames in the RQ (F) and the number of packets
per WQE (P). It ensures that N is not less than the minimum number of
WQEs in an RQ (N_min). Arithmetically, it means that F / P >= N_min
should be true. This function deals with logarithms, so it should check
that log(F) - log(P) >= log(N_min). However, if F < P, this expression
will cause an unsigned underflow. Check log(F) >= log(P) + log(N_min)
instead.

Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5e: Move parameter calculation functions to en/params.c
Maxim Mikityanskiy [Wed, 27 Mar 2019 11:09:27 +0000 (13:09 +0200)]
net/mlx5e: Move parameter calculation functions to en/params.c

This commit moves the parameter calculation functions to a separate file
for better modularity and code sharing with future features.

Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5e: Report mlx5e_xdp_set errors
Maxim Mikityanskiy [Tue, 19 Mar 2019 16:32:37 +0000 (18:32 +0200)]
net/mlx5e: Report mlx5e_xdp_set errors

If the channels fail to reopen after setting an XDP program, return the
error code instead of 0. A proper fix is still needed, as now any error
while reopening the channels brings the interface down. This patch only
adds error reporting.

Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5e: Remove unused parameter
Maxim Mikityanskiy [Thu, 7 Mar 2019 17:30:30 +0000 (19:30 +0200)]
net/mlx5e: Remove unused parameter

params is unused in mlx5e_init_di_list.

Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.com>
Reviewed-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5e: XDP, Inline small packets into the TX MPWQE in XDP xmit flow
Shay Agroskin [Thu, 14 Mar 2019 12:54:07 +0000 (14:54 +0200)]
net/mlx5e: XDP, Inline small packets into the TX MPWQE in XDP xmit flow

Upon high packet rate with multiple CPUs TX workloads, much of the HCA's
resources are spent on prefetching TX descriptors, thus affecting
transmission rates.
This patch comes to mitigate this problem by moving some workload to the
CPU and reducing the HW data prefetch overhead for small packets (<= 256B).

When forwarding packets with XDP, a packet that is smaller
than a certain size (set to ~256 bytes) would be sent inline within
its WQE TX descrptor (mem-copied), when the hardware tx queue is congested
beyond a pre-defined water-mark.

This is added to better utilize the HW resources (which now makes
one less packet data prefetch) and allow better scalability, on the
account of CPU usage (which now 'memcpy's the packet into the WQE).

To load balance between HW and CPU and get max packet rate, we use
watermarks to detect how much the HW is congested and move the work
loads back and forth between HW and CPU.

Performance:
Tested packet rate for UDP 64Byte multi-stream
over two dual port ConnectX-5 100Gbps NICs.
CPU: Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz

* Tested with hyper-threading disabled

XDP_TX:

|          | before | after   |       |
| 24 rings | 51Mpps | 116Mpps | +126% |
| 1 ring   | 12Mpps | 12Mpps  | same  |

XDP_REDIRECT:

** Below is the transmit rate, not the redirection rate
which might be larger, and is not affected by this patch.

|          | before  | after   |      |
| 32 rings | 64Mpps  | 92Mpps  | +43% |
| 1 ring   | 6.4Mpps | 6.4Mpps | same |

As we can see, feature significantly improves scaling, without
hurting single ring performance.

Signed-off-by: Shay Agroskin <shayag@mellanox.com>
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5e: XDP, Add TX MPWQE session counter
Shay Agroskin [Mon, 25 Feb 2019 16:02:09 +0000 (18:02 +0200)]
net/mlx5e: XDP, Add TX MPWQE session counter

This counter tracks how many TX MPWQE sessions are started in XDP SQ
in XDP TX/REDIRECT flow. It counts per-channel and global stats.

Signed-off-by: Shay Agroskin <shayag@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5e: XDP, Enhance RQ indication for XDP redirect flush
Tariq Toukan [Sun, 10 Mar 2019 13:35:58 +0000 (15:35 +0200)]
net/mlx5e: XDP, Enhance RQ indication for XDP redirect flush

The XDP redirect flush indication belongs to the receive queue,
not to its XDP send queue.

For this, use a new bit on rq->flags.

Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Shay Agroskin <shayag@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5e: XDP, Fix shifted flag index in RQ bitmap
Tariq Toukan [Sun, 10 Mar 2019 13:29:13 +0000 (15:29 +0200)]
net/mlx5e: XDP, Fix shifted flag index in RQ bitmap

Values in enum mlx5e_rq_flag are used as bit indixes.
Intention was to use them with no BIT(i) wrapping.

No functional bug fix here, as the same (shifted)flag bit
is used for all set, test, and clear operations.

Fixes: 121e89275471 ("net/mlx5e: Refactor RQ XDP_TX indication")
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: Shay Agroskin <shayag@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agonet/mlx5e: RX, Support multiple outstanding UMR posts
Tariq Toukan [Wed, 27 Feb 2019 10:06:08 +0000 (12:06 +0200)]
net/mlx5e: RX, Support multiple outstanding UMR posts

The buffers mapping of the Multi-Packet WQEs (of Striding RQ)
is done via UMR posts, one UMR WQE per an RX MPWQE.

A single MPWQE is capable of serving many incoming packets,
usually larger than the budget of a single napi cycle.
Hence, posting a single UMR WQE per napi cycle (and handling its
completion in the next cycle) works fine in many common cases,
but not always.

When an XDP program is loaded, every MPWQE is capable of serving less
packets, to satisfy the packet-per-page requirement.
Thus, for the same number of packets more MPWQEs (and UMR posts)
are needed (twice as much for the default MTU), giving less latency
room for the UMR completions.

In this patch, we add support for multiple outstanding UMR posts,
to allow faster gap closure between consuming MPWQEs and reposting
them back into the WQ.

For better SW and HW locality, we combine the UMR posts in bulks of
(at least) two.

This is expected to improve packet rate in high CPU scale.

Performance test:
As expected, huge improvement in large-scale (48 cores).

xdp_redirect_map, 64B UDP multi-stream.
Redirect from ConnectX-5 100Gbps to ConnectX-6 100Gbps.
CPU: Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz.

Before: Unstable, 7 to 30 Mpps
After:  Stable,   at 70.5 Mpps

No degradation in other tested scenarios.

Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agoMerge branch 'mlx5-next' of git://git.kernel.org/pub/scm/linux/kernel/git/mellanox...
Saeed Mahameed [Tue, 23 Apr 2019 18:57:33 +0000 (11:57 -0700)]
Merge branch 'mlx5-next' of git://git./linux/kernel/git/mellanox/linux

5 years agoMerge branch 'net-phy-mscc-Improvements-to-VSC8514-PHY-driver'
David S. Miller [Tue, 23 Apr 2019 17:47:58 +0000 (10:47 -0700)]
Merge branch 'net-phy-mscc-Improvements-to-VSC8514-PHY-driver'

Kavya Sree Kotagiri says:

====================
net: phy: mscc: Improvements to VSC8514 PHY driver.

    The VSC8514 PHY is a 4-ports PHY that is 10/100/1000BASE-T, 100BASE-FX,
    1000BASE-X, can communicate with the MAC via QSGMII.
    The MAC interface protocol for each port within QSGMII can
    be either 1000BASE-X or SGMII, if the QSGMII MAC that the VSC8514 is
    connecting to supports this functionality.
    VSC8514 also supports SGMII MAC-side autonegotiation on each individual
    port, downshifting, can set the blinking pattern of each of its 4 LEDs,
    SyncE, 1000BASE-T Ring Resiliency as well as HP Auto-MDIX detection.

    This patch series adds support for 10BASE-T, 100BASE-TX, and
    1000BASE-T, QSGMII link with the MAC, downshifting, HP Auto-MDIX
    detection and blinking pattern for its 4 LEDs.

    The GPIO register bank is a set of registers that are common to all
    PHYs in the package. So any modification in any register of this bank
    affects all PHYs of the package.

    If the PHYs haven't been reset before booting the Linux kernel and were
    configured to use interrupts for e.g. link status updates, it is
    required to clear the interrupts mask register of all PHYs before being
    able to use interrupts with any PHY. The first PHY of the package that
    will be init will take care of clearing all PHYs interrupts mask
    registers. Thus, we need to keep track of the init sequence in the
    package, if it's already been done or if it's to be done.

    Most of the init sequence of a PHY of the package is common to all PHYs
    in the package, thus we use the SMI broadcast feature which enables us
    to propagate a write in one register of one PHY to all PHYs in the same
    package.

    This patch series adds support for VSC8514 in Microsemi driver(mscc.c)
    and removes support from Vitesse driver(vitesse.c).

v8
- mscc: Added appropriate code using phy_modify() in vsc8514_config_init().

v7
- mscc: Handled return values in vsc8514_config_init().

v6
- mscc: Added proper return value in vsc85xx_csr_ctrl_phy_read().
- mscc: Replaced __mdiobus_write and__mdiobus_read with __phy_write and __phy_read resp.
- mscc: Replaced register addresses in 8514_config_init() with proper constants.

v5
- mscc: Added return error statements for few function calls.
- mscc: Added comments in vsc85xx_csr_ctrl_phy_read() and vsc85xx_csr_ctrl_phy_write()
v4
- mscc: Removed features settings
- mscc: Removed aneg_done settings.

v3
- mscc: Used BIT(x) for PHY_MCB_S6G_WRITE and PHY_MCB_S6G_READ
        instead of hex.
- mscc: Replaced magic numbers with proper constants.
- mscc: Handled delays and timeouts at appropriate points.
- mscc: Added comments/explanation where requested.

v2
- mscc: Sorted variable declarations in reverse christmas tree order.

v1
- Added 0/2 file.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: vitesse: Remove support for VSC8514.
Kavya Sree Kotagiri [Mon, 22 Apr 2019 11:52:04 +0000 (11:52 +0000)]
net: phy: vitesse: Remove support for VSC8514.

Add support for VSC8514 in Microsemi driver (mscc.c)
with more features.

Signed-off-by: Kavya Sree Kotagiri <kavyasree.kotagiri@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: mscc: add support for VSC8514 PHY.
Kavya Sree Kotagiri [Mon, 22 Apr 2019 11:51:35 +0000 (11:51 +0000)]
net: phy: mscc: add support for VSC8514 PHY.

The VSC8514 PHY is a 4-ports PHY that is 10/100/1000BASE-T, 100BASE-FX,
1000BASE-X, can communicate with the MAC via QSGMII.
The MAC interface protocol for each port within QSGMII can
be either 1000BASE-X or SGMII, if the QSGMII MAC that the VSC8514 is
connecting to supports this functionality.
VSC8514 also supports SGMII MAC-side autonegotiation on each individual
port, downshifting, can set the blinking pattern of each of its 4 LEDs,
SyncE, 1000BASE-T Ring Resiliency as well as HP Auto-MDIX detection.

This adds support for 10BASE-T, 100BASE-TX, and 1000BASE-T,
QSGMII link with the MAC, downshifting, HP Auto-MDIX detection
and blinking pattern for its 4 LEDs.

The GPIO register bank is a set of registers that are common to all PHYs
in the package. So any modification in any register of this bank affects
all PHYs of the package.

If the PHYs haven't been reset before booting the Linux kernel and were
configured to use interrupts for e.g. link status updates, it is
required to clear the interrupts mask register of all PHYs before being
able to use interrupts with any PHY. The first PHY of the package that
will be init will take care of clearing all PHYs interrupts mask
registers. Thus, we need to keep track of the init sequence in the
package, if it's already been done or if it's to be done.

Most of the init sequence of a PHY of the package is common to all PHYs
in the package, thus we use the SMI broadcast feature which enables us
to propagate a write in one register of one PHY to all PHYs in the same
package.

Signed-off-by: Kavya Sree Kotagiri <kavyasree.kotagiri@microchip.com>
Signed-off-by: Quentin Schulz <quentin.schulz@bootlin.com>
Co-developed-by: Quentin Schulz <quentin.schulz@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: marvell: add new default led configure for m88e151x
Jian Shen [Mon, 22 Apr 2019 13:52:23 +0000 (21:52 +0800)]
net: phy: marvell: add new default led configure for m88e151x

The default m88e151x LED configuration is 0x1177, used LED[0]
for 1000M link, LED[1] for 100M link, and LED[2] for active.
But for some boards, which use LED[0] for link, and LED[1] for
active, prefer to be 0x1040. To be compatible with this case,
this patch defines a new dev_flag, and set it before connect
phy in HNS3 driver. When phy initializing, using the new
LED configuration if this dev_flag is set.

Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: systemport: Remove need for DMA descriptor
Florian Fainelli [Mon, 22 Apr 2019 16:46:44 +0000 (09:46 -0700)]
net: systemport: Remove need for DMA descriptor

All we do is write the length/status and address bits to a DMA
descriptor only to write its contents into on-chip registers right
after, eliminate this unnecessary step.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agobridge: Fix possible use-after-free when deleting bridge port
Ido Schimmel [Mon, 22 Apr 2019 09:33:19 +0000 (09:33 +0000)]
bridge: Fix possible use-after-free when deleting bridge port

When a bridge port is being deleted, do not dereference it later in
br_vlan_port_event() as it can result in a use-after-free [1] if the RCU
callback was executed before invoking the function.

[1]
[  129.638551] ==================================================================
[  129.646904] BUG: KASAN: use-after-free in br_vlan_port_event+0x53c/0x5fd
[  129.654406] Read of size 8 at addr ffff8881e4aa1ae8 by task ip/483
[  129.663008] CPU: 0 PID: 483 Comm: ip Not tainted 5.1.0-rc5-custom-02265-ga946bd73daac #1383
[  129.672359] Hardware name: Mellanox Technologies Ltd. MSN2100-CB2FO/SA001017, BIOS 5.6.5 06/07/2016
[  129.682484] Call Trace:
[  129.685242]  dump_stack+0xa9/0x10e
[  129.689068]  print_address_description.cold.2+0x9/0x25e
[  129.694930]  kasan_report.cold.3+0x78/0x9d
[  129.704420]  br_vlan_port_event+0x53c/0x5fd
[  129.728300]  br_device_event+0x2c7/0x7a0
[  129.741505]  notifier_call_chain+0xb5/0x1c0
[  129.746202]  rollback_registered_many+0x895/0xe90
[  129.793119]  unregister_netdevice_many+0x48/0x210
[  129.803384]  rtnl_delete_link+0xe1/0x140
[  129.815906]  rtnl_dellink+0x2a3/0x820
[  129.844166]  rtnetlink_rcv_msg+0x397/0x910
[  129.868517]  netlink_rcv_skb+0x137/0x3a0
[  129.882013]  netlink_unicast+0x49b/0x660
[  129.900019]  netlink_sendmsg+0x755/0xc90
[  129.915758]  ___sys_sendmsg+0x761/0x8e0
[  129.966315]  __sys_sendmsg+0xf0/0x1c0
[  129.988918]  do_syscall_64+0xa4/0x470
[  129.993032]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
[  129.998696] RIP: 0033:0x7ff578104b58
...
[  130.073811] Allocated by task 479:
[  130.077633]  __kasan_kmalloc.constprop.5+0xc1/0xd0
[  130.083008]  kmem_cache_alloc_trace+0x152/0x320
[  130.088090]  br_add_if+0x39c/0x1580
[  130.092005]  do_set_master+0x1aa/0x210
[  130.096211]  do_setlink+0x985/0x3100
[  130.100224]  __rtnl_newlink+0xc52/0x1380
[  130.104625]  rtnl_newlink+0x6b/0xa0
[  130.108541]  rtnetlink_rcv_msg+0x397/0x910
[  130.113136]  netlink_rcv_skb+0x137/0x3a0
[  130.117538]  netlink_unicast+0x49b/0x660
[  130.121939]  netlink_sendmsg+0x755/0xc90
[  130.126340]  ___sys_sendmsg+0x761/0x8e0
[  130.130645]  __sys_sendmsg+0xf0/0x1c0
[  130.134753]  do_syscall_64+0xa4/0x470
[  130.138864]  entry_SYSCALL_64_after_hwframe+0x49/0xbe

[  130.146195] Freed by task 0:
[  130.149421]  __kasan_slab_free+0x125/0x170
[  130.154016]  kfree+0xf3/0x310
[  130.157349]  kobject_put+0x1a8/0x4c0
[  130.161363]  rcu_core+0x859/0x19b0
[  130.165175]  __do_softirq+0x250/0xa26
[  130.170956] The buggy address belongs to the object at ffff8881e4aa1ae8
                which belongs to the cache kmalloc-1k of size 1024
[  130.184972] The buggy address is located 0 bytes inside of
                1024-byte region [ffff8881e4aa1ae8ffff8881e4aa1ee8)

Fixes: 9c0ec2e7182a ("bridge: support binding vlan dev link state to vlan member bridge ports")
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Cc: Mike Manning <mmanning@vyatta.att-mail.com>
Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Acked-by: Mike Manning <mmanning@vyatta.att-mail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agor8152: sync sa_family with the media type of network device
Crag.Wang [Mon, 22 Apr 2019 05:03:43 +0000 (13:03 +0800)]
r8152: sync sa_family with the media type of network device

Without this patch the socket address family sporadically gets wrong
value ends up the dev_set_mac_address() fails to set the desired MAC
address.

Fixes: 25766271e42f ("r8152: Refresh MAC address during USBDEVFS_RESET")
Signed-off-by: Crag.Wang <crag.wang@dell.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-By: Mario Limonciello <mario.limonciello@dell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'mlxsw-Shared-buffer-improvements'
David S. Miller [Tue, 23 Apr 2019 05:09:33 +0000 (22:09 -0700)]
Merge branch 'mlxsw-Shared-buffer-improvements'

Ido Schimmel says:

====================
mlxsw: Shared buffer improvements

This patchset includes two improvements with regards to shared buffer
configuration in mlxsw.

The first part of this patchset forbids the user from performing illegal
shared buffer configuration that can result in unnecessary packet loss.
In order to better communicate these configuration failures to the user,
extack is propagated from devlink towards drivers. This is done in
patches #1-#8.

The second part of the patchset deals with the shared buffer
configuration of the CPU port. When a packet is trapped by the device,
it is sent across the PCI bus to the attached host CPU. From the
device's perspective, it is as if the packet is transmitted through the
CPU port.

While testing traffic directed at the CPU it became apparent that for
certain packet sizes and certain burst sizes, the current shared buffer
configuration of the CPU port is inadequate and results in packet drops.
The configuration is adjusted by patches #9-#14 that create two new pools
- ingress & egress - which are dedicated for CPU traffic.
====================

Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_buffers: Adjust CPU port shared buffer egress quotas
Ido Schimmel [Mon, 22 Apr 2019 12:08:56 +0000 (12:08 +0000)]
mlxsw: spectrum_buffers: Adjust CPU port shared buffer egress quotas

Switch the CPU port to use the new dedicated egress pool instead the
previously used egress pool which was shared with normal front panel
ports.

Add per-port quotas for the amount of traffic that can be buffered for
the CPU port and also adjust the per-{port, TC} quotas.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_buffers: Allow skipping ingress port quota configuration
Ido Schimmel [Mon, 22 Apr 2019 12:08:55 +0000 (12:08 +0000)]
mlxsw: spectrum_buffers: Allow skipping ingress port quota configuration

The CPU port is used to transmit traffic that is trapped to the host
CPU. It is therefore irrelevant to define ingress quota for it.

Add a 'skip_ingress' argument to the function tasked with configuring
per-port quotas, so that ingress quotas could be skipped in case the
passed local port is the CPU port.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_buffers: Split business logic from mlxsw_sp_port_sb_pms_init()
Ido Schimmel [Mon, 22 Apr 2019 12:08:54 +0000 (12:08 +0000)]
mlxsw: spectrum_buffers: Split business logic from mlxsw_sp_port_sb_pms_init()

The function is used to set the per-port shared buffer quotas.
Currently, these quotas are only set for front panel ports, but a
subsequent patch will configure these quotas for the CPU port as well.

The configuration required for the CPU port is a bit different than that
of the front panel ports, so split the business logic into a separate
function which will be called with different parameters for the CPU
port.

No functional changes intended.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_buffers: Use new CPU ingress pool for control packets
Ido Schimmel [Mon, 22 Apr 2019 12:08:52 +0000 (12:08 +0000)]
mlxsw: spectrum_buffers: Use new CPU ingress pool for control packets

Use the new ingress pool that was added in the previous patch for
control packets (e.g., STP, LACP) that are trapped to the CPU.

The previous management pool is no longer necessary and therefore its
size is set to 0.

The maximum quota for traffic towards the CPU is increased to 50% of the
free space in the new ingress pool and therefore the reserved space is
reduced by half, to 10KB - in both the shared and headroom buffer. This
allows for more efficient utilization of the shared buffer as reserved
space cannot be used for other purposes.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_buffers: Add pools for CPU traffic
Ido Schimmel [Mon, 22 Apr 2019 12:08:51 +0000 (12:08 +0000)]
mlxsw: spectrum_buffers: Add pools for CPU traffic

Packets that are trapped to the CPU are transmitted through the CPU port
to the attached host. The CPU port is therefore like any other port and
needs to have shared buffer configuration.

The maximum quotas configured for the CPU are provided using dynamic
threshold and cannot be changed by the user. In order to make sure that
these thresholds are always valid, the configuration of the threshold
type of these pools is forbidden.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_buffers: Remove assumption about pool order
Ido Schimmel [Mon, 22 Apr 2019 12:08:50 +0000 (12:08 +0000)]
mlxsw: spectrum_buffers: Remove assumption about pool order

The code currently assumes that ingress pools have lower indices than
egress pools. This makes it impossible to add more ingress pools
without breaking user configuration that relies on a certain pool index
to correspond to an egress pool.

Remove such assumptions from the code, so that more ingress pools could
be added by subsequent patches.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_buffers: Forbid changing multicast TCs' attributes
Ido Schimmel [Mon, 22 Apr 2019 12:08:49 +0000 (12:08 +0000)]
mlxsw: spectrum_buffers: Forbid changing multicast TCs' attributes

Commit e83c045e53d7 ("mlxsw: spectrum_buffers: Configure MC pool")
configured the threshold of the multicast TCs as infinite so that the
admission of multicast packets is only depended on per-switch priority
threshold.

Forbid the user from changing the thresholds of these multicast TCs and
their binding to a different pool.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_buffers: Forbid changing threshold type of first egress pool
Ido Schimmel [Mon, 22 Apr 2019 12:08:47 +0000 (12:08 +0000)]
mlxsw: spectrum_buffers: Forbid changing threshold type of first egress pool

Multicast packets have three egress quotas:
* Per egress port
* Per egress port and traffic class
* Per switch priority

The limits on the switch priority are not exposed to the user and
specified as dynamic threshold on the first egress pool.

Forbid changing the threshold type of the first egress pool so that
these limits are always valid.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_buffers: Forbid configuration of multicast pool
Ido Schimmel [Mon, 22 Apr 2019 12:08:46 +0000 (12:08 +0000)]
mlxsw: spectrum_buffers: Forbid configuration of multicast pool

Commit e83c045e53d7 ("mlxsw: spectrum_buffers: Configure MC pool") added
a dedicated pool for multicast traffic. The pool is visible to the user
so that it would be possible to monitor its occupancy, but its
configuration should be forbidden in order to maintain its intended
operation.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_buffers: Add ability to veto TC's configuration
Ido Schimmel [Mon, 22 Apr 2019 12:08:45 +0000 (12:08 +0000)]
mlxsw: spectrum_buffers: Add ability to veto TC's configuration

Subsequent patches are going to need to veto changes in certain TCs'
binding and threshold configurations.

Add fields to the TC's struct that indicate if the TC can be bound to a
different pool and whether its threshold can change and enforce that.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_buffers: Add ability to veto pool's configuration
Ido Schimmel [Mon, 22 Apr 2019 12:08:43 +0000 (12:08 +0000)]
mlxsw: spectrum_buffers: Add ability to veto pool's configuration

Subsequent patches are going to need to veto changes in certain pools'
size and / or threshold type (mode).

Add two fields to the pool's struct that indicate if either of these
attributes is allowed to change and enforce that.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_buffers: Use defines for pool indices
Ido Schimmel [Mon, 22 Apr 2019 12:08:42 +0000 (12:08 +0000)]
mlxsw: spectrum_buffers: Use defines for pool indices

The pool indices are currently hard coded throughout the code, which
makes the code hard to follow and extend.

Overcome this by using defines for the pool indices.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_buffers: Add extack messages for invalid configurations
Ido Schimmel [Mon, 22 Apr 2019 12:08:41 +0000 (12:08 +0000)]
mlxsw: spectrum_buffers: Add extack messages for invalid configurations

Add extack messages to better communicate invalid configuration to the
user.

Example:

# devlink sb pool set pci/0000:01:00.0 pool 0 size 104857600 thtype dynamic
Error: mlxsw_spectrum: Exceeded shared buffer size.
devlink answers: Invalid argument

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: devlink: Add extack to shared buffer operations
Ido Schimmel [Mon, 22 Apr 2019 12:08:39 +0000 (12:08 +0000)]
net: devlink: Add extack to shared buffer operations

Add extack to shared buffer set operations, so that meaningful error
messages could be propagated to the user.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Petr Machata <petrm@mellanox.com>
Cc: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-clean-up-needless-use-of-module-infrastructure'
David S. Miller [Tue, 23 Apr 2019 04:51:00 +0000 (21:51 -0700)]
Merge branch 'net-clean-up-needless-use-of-module-infrastructure'

Paul Gortmaker says:

====================
clean up needless use of module infrastructure

People can embed modular includes and modular exit functions into code
that never use any of it, and they won't get any errors or warnings.

Using modular infrastructure in non-modules might seem harmless, but some
of the downfalls this leads to are:

 (1) it is easy to accidentally write unused module_exit removal code
 (2) it can be misleading when reading the source, thinking a driver can
     be modular when the Makefile and/or Kconfig prohibit it
 (3) an unused include of the module.h header file will in turn
     include nearly everything else; adding a lot to CPP overhead.
 (4) it gets copied/replicated into other drivers and spreads quickly.

As a data point for #3 above, an empty C file that just includes the
module.h header generates over 750kB of CPP output.  Repeating the same
experiment with init.h and the result is less than 12kB; with export.h
it is only about 1/2kB; with both it still is less than 12kB.  One driver
in this series gets the module.h ---> init.h+export.h conversion.

Worse, are headers in include/linux that in turn include <linux/module.h>
as they can impact a whole fleet of drivers, or a whole subsystem, so
special care should be used in order to avoid that.  Such headers should
only include what they need to be stand-alone; they should not be trying
to anticipate the various header needs of their possible end users.

In this series, four include/linux headers have module.h removed from
them because they don't strictly need it.  Then three chunks of net
related code have modular infrastructure that isn't used, removed.

There are no runtime changes, so the biggest risk is a genuine consumer
of module.h content relying on implicitly getting it from one of the
include/linux instances removed here - thus resulting in a build fail.

With that in mind, allmodconfig build testing was done on x86-64, arm64,
x86-32, arm. powerpc, and mips on linux-next (and hence net-next).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: strparser: make it explicitly non-modular
Paul Gortmaker [Sun, 21 Apr 2019 03:29:48 +0000 (23:29 -0400)]
net: strparser: make it explicitly non-modular

The Kconfig currently controlling compilation of this code is:

net/strparser/Kconfig:config STREAM_PARSER
net/strparser/Kconfig:  def_bool n

...meaning that it currently is not being built as a module by anyone.

Lets remove the modular code that is essentially orphaned, so that
when reading the driver there is no doubt it is builtin-only.

Since module_init translates to device_initcall in the non-modular
case, the init ordering remains unchanged with this commit.  For
clarity, we change the fcn name mod_init to dev_init at the same time.

We replace module.h with init.h and export.h ; the latter since this
file exports some syms.

Cc: "David S. Miller" <davem@davemloft.net>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Martin KaFai Lau <kafai@fb.com>
Cc: Song Liu <songliubraving@fb.com>
Cc: Yonghong Song <yhs@fb.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: bpfilter: dont use module_init in non-modular code
Paul Gortmaker [Sun, 21 Apr 2019 03:29:47 +0000 (23:29 -0400)]
net: bpfilter: dont use module_init in non-modular code

The Kconfig controlling this code is:

bpfilter/Kconfig:menuconfig BPFILTER
bpfilter/Kconfig:   bool "BPF based packet filtering framework (BPFILTER)"

Since it isn't a module, we shouldn't use module_init().  Instead we
use device_initcall() - which is exactly what module_init() defaults
to for non-modular code/builds.

We don't remove <linux/module.h> from the includes since this file does
a request_module() and hence is a valid user of that header file, even
though it is not modular itself.

Cc: "David S. Miller" <davem@davemloft.net>
Cc: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Cc: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agocgroup: net: remove left over MODULE_LICENSE tag
Paul Gortmaker [Sun, 21 Apr 2019 03:29:46 +0000 (23:29 -0400)]
cgroup: net: remove left over MODULE_LICENSE tag

The Kconfig currently controlling compilation of this code is:

net/Kconfig:config CGROUP_NET_PRIO
net/Kconfig:    bool "Network priority cgroup"

...meaning that it currently is not being built as a module by anyone,
as module support was discontinued in 2014.

We delete the MODULE_LICENSE tag since all that information is already
contained at the top of the file in the comments.

We don't delete module.h from the includes since it was no longer there
to begin with.

Cc: "David S. Miller" <davem@davemloft.net>
Cc: Tejun Heo <tj@kernel.org>
Cc: "Rosen, Rami" <rami.rosen@intel.com>
Cc: Daniel Wagner <daniel.wagner@bmw-carit.de>
Cc: netdev@vger.kernel.org
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: tc_act: drop include of module.h from tc_ife.h
Paul Gortmaker [Sun, 21 Apr 2019 03:29:45 +0000 (23:29 -0400)]
net: tc_act: drop include of module.h from tc_ife.h

Ideally, header files under include/linux shouldn't be adding
includes of other headers, in anticipation of their consumers,
but just the headers needed for the header itself to pass
parsing with CPP.

The module.h is particularly bad in this sense, as it itself does
include a whole bunch of other headers, due to the complexity of
module support.

Since tc_ife.h is not going into a module struct looking for
specific fields, we can just let it know that module is a struct,
just like about 60 other include/linux headers already do.

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: fib: drop include of module.h from fib_notifier.h
Paul Gortmaker [Sun, 21 Apr 2019 03:29:44 +0000 (23:29 -0400)]
net: fib: drop include of module.h from fib_notifier.h

Ideally, header files under include/linux shouldn't be adding
includes of other headers, in anticipation of their consumers,
but just the headers needed for the header itself to pass
parsing with CPP.

The module.h is particularly bad in this sense, as it itself does
include a whole bunch of other headers, due to the complexity of
module support.

Since fib_notifier.h is not going into a module struct looking for
specific fields, we can just let it know that module is a struct,
just like about 60 other include/linux headers already do.

Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ife: drop include of module.h from net/ife.h
Paul Gortmaker [Sun, 21 Apr 2019 03:29:43 +0000 (23:29 -0400)]
net: ife: drop include of module.h from net/ife.h

Ideally, header files under include/linux shouldn't be adding
includes of other headers, in anticipation of their consumers,
but just the headers needed for the header itself to pass
parsing with CPP.

The module.h is particularly bad in this sense, as it itself does
include a whole bunch of other headers, due to the complexity of
module support.

There doesn't appear to be anything in net/ife.h that is module
related, and build coverage doesn't appear to show any other
files/drivers relying implicitly on getting it from here.

So it appears we are simply free to just remove it in this case.

Cc: Yotam Gigi <yotam.gi@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: psample: drop include of module.h from psample.h
Paul Gortmaker [Sun, 21 Apr 2019 03:29:42 +0000 (23:29 -0400)]
net: psample: drop include of module.h from psample.h

Ideally, header files under include/linux shouldn't be adding
includes of other headers, in anticipation of their consumers,
but just the headers needed for the header itself to pass
parsing with CPP.

The module.h is particularly bad in this sense, as it itself does
include a whole bunch of other headers, due to the complexity of
module support.

There doesn't appear to be anything in psample.h that is module
related, and build coverage doesn't appear to show any other
files/drivers relying implicitly on getting it from here.

So it appears we are simply free to just remove it in this case.

Cc: Yotam Gigi <yotam.gi@gmail.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: Rename net/nexthop.h net/rtnh.h
David Ahern [Sat, 20 Apr 2019 16:28:20 +0000 (09:28 -0700)]
net: Rename net/nexthop.h net/rtnh.h

The header contains rtnh_ macros so rename the file accordingly.
Allows a later patch to use the nexthop.h name for the new
nexthop code.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: Remove fib6_info_nh_lwt
David Ahern [Sat, 20 Apr 2019 16:27:27 +0000 (09:27 -0700)]
ipv6: Remove fib6_info_nh_lwt

fib6_info_nh_lwt is no longer used; remove it.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoinclude/net/tcp.h: whitespace cleanup at tcp_v4_check
Daniel T. Lee [Sat, 20 Apr 2019 15:50:42 +0000 (00:50 +0900)]
include/net/tcp.h: whitespace cleanup at tcp_v4_check

This patch makes trivial whitespace fix to the function
tcp_v4_check at include/net/tcp.h file.

It has stylistic issue, which is "space required after that ','"
and it can be confirmed with ./scripts/checkpatch.pl tool.

    ERROR: space required after that ',' (ctx:VxV)
    #29: FILE: include/net/tcp.h:1317:
    +         return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
                                        ^

Signed-off-by: Daniel T. Lee <danieltimlee@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
David S. Miller [Tue, 23 Apr 2019 04:35:55 +0000 (21:35 -0700)]
Merge git://git./linux/kernel/git/bpf/bpf-next

Alexei Starovoitov says:

====================
pull-request: bpf-next 2019-04-22

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

The main changes are:

1) allow stack/queue helpers from more bpf program types, from Alban.

2) allow parallel verification of root bpf programs, from Alexei.

3) introduce bpf sysctl hook for trusted root cases, from Andrey.

4) recognize var/datasec in btf deduplication, from Andrii.

5) cpumap performance optimizations, from Jesper.

6) verifier prep for alu32 optimization, from Jiong.

7) libbpf xsk cleanup, from Magnus.

8) other various fixes and cleanups.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'bpf-verifier-lock'
Daniel Borkmann [Mon, 22 Apr 2019 23:50:44 +0000 (01:50 +0200)]
Merge branch 'bpf-verifier-lock'

Alexei Starovoitov says:

====================
Allow the bpf verifier to run in parallel for root.
====================

Acked-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agobpf: drop bpf_verifier_lock
Alexei Starovoitov [Fri, 19 Apr 2019 14:44:55 +0000 (07:44 -0700)]
bpf: drop bpf_verifier_lock

Drop bpf_verifier_lock for root to avoid being DoS-ed by unprivileged.
The BPF verifier is now fully parallel.
All unpriv users are still serialized by bpf_verifier_lock to avoid
exhausting kernel memory by running N parallel verifications.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agobpf: remove global variables
Alexei Starovoitov [Fri, 19 Apr 2019 14:44:54 +0000 (07:44 -0700)]
bpf: remove global variables

Move three global variables protected by bpf_verifier_lock into
'struct bpf_verifier_env' to allow parallel verification.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agobpf: document the verifier limits
Alexei Starovoitov [Thu, 18 Apr 2019 01:27:01 +0000 (18:27 -0700)]
bpf: document the verifier limits

Document the verifier limits.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoMerge tag 'v5.1-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux...
Saeed Mahameed [Mon, 22 Apr 2019 22:25:39 +0000 (15:25 -0700)]
Merge tag 'v5.1-rc1' of git://git./linux/kernel/git/torvalds/linux into mlx5-next

Linux 5.1-rc1

We forgot to reset the branch last merge window thus mlx5-next is outdated
and still based on 5.0-rc2. This merge commit is needed to sync mlx5-next
branch with 5.1-rc1.

Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
5 years agoipv6: Restore RTF_ADDRCONF check in rt6_qualify_for_ecmp
David Ahern [Mon, 22 Apr 2019 00:39:18 +0000 (17:39 -0700)]
ipv6: Restore RTF_ADDRCONF check in rt6_qualify_for_ecmp

The RTF_ADDRCONF flag filters out routes added by RA's in determining
which routes can be appended to an existing one to create a multipath
route. Restore the flag check and add a comment to document the RA piece.

Fixes: 4e54507ab1a9 ("ipv6: Simplify rt6_qualify_for_ecmp")
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoipv6: Simplify rt6_qualify_for_ecmp
David Ahern [Sun, 21 Apr 2019 15:49:01 +0000 (08:49 -0700)]
ipv6: Simplify rt6_qualify_for_ecmp

After commit c7a1ce397ada ("ipv6: Change addrconf_f6i_alloc to use
ip6_route_info_create"), the gateway is no longer filled in for fib6_nh
structs in a prefix route. Accordingly, the RTF_ADDRCONF flag check can
be dropped from the 'rt6_qualify_for_ecmp'.

Further, RTF_DYNAMIC is only set in rt6_info instances, so it can be
removed from the check as well.

This reduces rt6_qualify_for_ecmp and the mlxsw version to just checking
if the nexthop has a gateway which is the real indication of whether
entries can be coalesced into a multipath route.

Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hippi:Fix misuse of %x in rrunner.c
Fuqian Huang [Sun, 21 Apr 2019 11:48:26 +0000 (19:48 +0800)]
net: hippi:Fix misuse of %x in rrunner.c

The pointer should be printed with %p or %px rather than
cast to unsigned long type and printed with %08lx.
Change %08lx to %p to print the pointer.

Signed-off-by: Fuqian Huang <huangfq.daxian@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: ax25: fix misuse of %x
Fuqian Huang [Sun, 21 Apr 2019 11:48:06 +0000 (19:48 +0800)]
net: ax25: fix misuse of %x

Pointers should be printed with %p or %px rather than
cast to long type and printed with %8.8lx.
Change %8.8lx to %p to print the pointer.

Signed-off-by: Fuqian Huang <huangfq.daxian@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoatm: iphase: fix misuse of %x
Fuqian Huang [Sun, 21 Apr 2019 11:46:54 +0000 (19:46 +0800)]
atm: iphase: fix misuse of %x

Pointers should be printed with %p or %px rather than
cast to long type and printed with %x.
Change %x to %p to print the pointers.

Signed-off-by: Fuqian Huang <huangfq.daxian@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'mlxsw-Small-routing-improvements'
David S. Miller [Sun, 21 Apr 2019 17:31:45 +0000 (10:31 -0700)]
Merge branch 'mlxsw-Small-routing-improvements'

Ido Schimmel says:

====================
mlxsw: Small routing improvements

Patch #1 switches the driver to use a unique and stable ECMP/LAG seed.
This allows for consistent behavior across reboots and avoids hash
polarization at the same time.

Patch #2 relaxes the FIB rule validation in the driver to allow the
installation of rules that direct locally generated traffic (iif=lo).
This does not result in a discrepancy between both data paths because
packets received by the device would never match such rules.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum_router: Relax FIB rule validation
Ido Schimmel [Sun, 21 Apr 2019 07:18:36 +0000 (07:18 +0000)]
mlxsw: spectrum_router: Relax FIB rule validation

Currently, mlxsw does not support policy-based routing (PBR) and
therefore forbids the installation of non-default FIB rules except for
the l3mdev rule which is used for VRFs.

Relax the check to allow the installation of FIB rules that would never
match packets received by the device. Specifically, if the iif is that
of the loopback netdev. This is useful for users that need to redirect
locally generated packets based on FIB rules.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Tested-by: Alexander Petrovskiy <alexpe@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agomlxsw: spectrum: Use a stable ECMP/LAG seed
Ido Schimmel [Sun, 21 Apr 2019 07:18:34 +0000 (07:18 +0000)]
mlxsw: spectrum: Use a stable ECMP/LAG seed

In order to get a consistent behavior of traffic flows across reboots /
module unload, we need to use the same ECMP/LAG seed.

Calculate the seed by hashing the base MAC of the device. This results
in a seed that is both unique (to avoid polarization) and consistent.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: add SR-IOV trusted VF support
Pablo Cascón [Sat, 20 Apr 2019 00:20:09 +0000 (17:20 -0700)]
nfp: add SR-IOV trusted VF support

By default VFs are not trusted. Add ndo_set_vf_trust support to toggle
a new per-VF bit. Coupled with FW with this capability allows a
trusted VF to change its MAC even after being administratively set by
the PF. Also populate the trusted field on ndo_get_vf_config. Add the
same ndo to the representors.

Signed-off-by: Pablo Cascón <pablo.cascon@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'hns3-next'
David S. Miller [Sat, 20 Apr 2019 01:43:16 +0000 (18:43 -0700)]
Merge branch 'hns3-next'

Huazhong Tan says:

====================
net: hns3: add some DFX info for HNS3 driver

This patch-set adds some DFX information to HNS3 driver, for easily
debug some problems, and fixes some related bugs.

[patch 1/12 - 4/12] adds debug info about reset & interrupt events

[patch 5/12 - 7/12] adds debug info about TX time out & fixes
related bugs

[patch 8/12] adds support for setting netif message level

[patch 9/12 - 10/12] adds debugfs command to dump NCL & MAC info

[patch 11/12] adds VF's queue statistics info updating

[patch 12/12] adds a check for debugfs help function to decide which
commands are supportable
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: add function type check for debugfs help information
Yufeng Mo [Fri, 19 Apr 2019 03:05:47 +0000 (11:05 +0800)]
net: hns3: add function type check for debugfs help information

PF supports all debugfs command, but VF only supports part of
debugfs command. So VF should not show unsupported help information.

This patch adds a check for PF and PF to show the supportable help
information.

Signed-off-by: Yufeng Mo <moyufeng@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: add queue's statistics update to service task
liuzhongzhu [Fri, 19 Apr 2019 03:05:46 +0000 (11:05 +0800)]
net: hns3: add queue's statistics update to service task

This patch updates VF's TQP statistic info in the service task,
and adds a limitation to prevent update too frequently.

Signed-off-by: liuzhongzhu <liuzhongzhu@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: Add handling of MAC tunnel interruption
Weihang Li [Fri, 19 Apr 2019 03:05:45 +0000 (11:05 +0800)]
net: hns3: Add handling of MAC tunnel interruption

MAC tnl interruptions are different from other type of RAS and MSI-X
errors, because some bits, such as OVF/LR/RF will occur during link up
and down.

The drivers should clear status of all MAC tnl interruption bits but
shouldn't print any message that would mislead the users.

In case that link down and re-up in a short time because of some reasons,
we record when they occurred, and users can query them by debugfs.

Signed-off-by: Weihang Li <liweihang@hisilicon.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: add support for dump ncl config by debugfs
Weihang Li [Fri, 19 Apr 2019 03:05:44 +0000 (11:05 +0800)]
net: hns3: add support for dump ncl config by debugfs

This patch allow users to dump content of NCL_CONFIG by using debugfs
command.
Command format:
echo dump ncl_config <offset> <length> > cmd
It will print as follows:
hns3 0000:7d:00.0: offset |    data
hns3 0000:7d:00.0: 0x0000 | 0x00000020
hns3 0000:7d:00.0: 0x0004 | 0x00000400
hns3 0000:7d:00.0: 0x0008 | 0x08020401

Signed-off-by: Weihang Li <liweihang@hisilicon.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: Add support for netif message level settings
Yonglong Liu [Fri, 19 Apr 2019 03:05:43 +0000 (11:05 +0800)]
net: hns3: Add support for netif message level settings

This patch adds support for network interface message level
settings. The message level can be changed by module parameter
or ethtool.

Signed-off-by: Yonglong Liu <liuyonglong@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: dump more information when tx timeout happens
Jian Shen [Fri, 19 Apr 2019 03:05:42 +0000 (11:05 +0800)]
net: hns3: dump more information when tx timeout happens

Currently we just print few information when tx timeout happens.
In order to find out the cause of timeout, this patch prints more
information about the packet statistics, tqp registers and
napi state.

Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: fix loop condition of hns3_get_tx_timeo_queue_info()
Jian Shen [Fri, 19 Apr 2019 03:05:41 +0000 (11:05 +0800)]
net: hns3: fix loop condition of hns3_get_tx_timeo_queue_info()

In function hns3_get_tx_timeo_queue_info(), it should use
netdev->num_tx_queues, instead of netdve->real_num_tx_queues
as the loop limitation.

Fixes: 424eb834a9be ("net: hns3: Unified HNS3 {VF|PF} Ethernet Driver for hip08 SoC")
Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: refine tx timeout count handle
Jian Shen [Fri, 19 Apr 2019 03:05:40 +0000 (11:05 +0800)]
net: hns3: refine tx timeout count handle

In current codes, tx_timeout_cnt is used before increased,
then we can see the tx_timeout_count is still 0 from the
print when tx timeout happens, e.g.
"hns3 0000:7d:00.3 eth3: tx_timeout count: 0, queue id: 0, SW_NTU:
 0xa6, SW_NTC: 0xa4, HW_HEAD: 0xa4, HW_TAIL: 0xa6, INT: 0x1"

The tx_timeout_cnt should be updated before used.

Signed-off-by: Jian Shen <shenjian15@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: add some debug info for hclgevf_get_mbx_resp()
Huazhong Tan [Fri, 19 Apr 2019 03:05:39 +0000 (11:05 +0800)]
net: hns3: add some debug info for hclgevf_get_mbx_resp()

When wait for response timeout or response code not match, there
should be more information for debugging.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: add some debug information for hclge_check_event_cause
Huazhong Tan [Fri, 19 Apr 2019 03:05:38 +0000 (11:05 +0800)]
net: hns3: add some debug information for hclge_check_event_cause

When received vector0 msix and other event interrupt, it should
print the value of the register for debugging.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: add reset statistics for VF
Huazhong Tan [Fri, 19 Apr 2019 03:05:37 +0000 (11:05 +0800)]
net: hns3: add reset statistics for VF

This patch adds some statistics for VF reset.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: add reset statistics info for PF
Huazhong Tan [Fri, 19 Apr 2019 03:05:36 +0000 (11:05 +0800)]
net: hns3: add reset statistics info for PF

This patch adds statistics for PF's reset information,
also, provides a debugfs command to dump these statistics.

Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: Peng Li <lipeng321@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agolibbpf: fix BPF_LOG_BUF_SIZE off-by-one error
McCabe, Robert J [Fri, 19 Apr 2019 23:37:20 +0000 (18:37 -0500)]
libbpf: fix BPF_LOG_BUF_SIZE off-by-one error

The BPF_PROG_LOAD condition for kernel version <= 5.1 is

   log->len_total > UINT_MAX >> 8 /* (16 * 1024 * 1024) - 1 */

Signed-off-by: McCabe, Robert J <robert.mccabe@rockwellcollins.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agotcp: properly reset skb->truesize for tx recycling
Eric Dumazet [Fri, 19 Apr 2019 23:02:03 +0000 (16:02 -0700)]
tcp: properly reset skb->truesize for tx recycling

tcp sendmsg() and sendpage() normally advance skb->data_len
and skb->truesize by the payload added to an skb.

But sendmsg(fd, ..., MSG_ZEROCOPY) has to account for whole pages,
even if a single byte of payload is used in the page.

This means that we can not assume skb->truesize can be adjusted
by skb->data_len. We must instead overwrite its value.

Otherwise skb->truesize is too big and can hit socket sndbuf limit,
especially if the skb is recycled multiple times :/

Fixes: 472c2e07eef0 ("tcp: add one skb cache for tx")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>