GitHub/MotorolaMobilityLLC/kernel-slsi.git
7 years agoixgb: use new API ethtool_{get|set}_link_ksettings
Philippe Reynes [Sun, 5 Feb 2017 23:11:11 +0000 (00:11 +0100)]
ixgb: use new API ethtool_{get|set}_link_ksettings

The ethtool API {get|set}_settings is deprecated.
We move this driver to new API {get|set}_link_ksettings.

As I don't have the hardware, I'd be very pleased if
someone may test this patch.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigbvf: use new API ethtool_{get|set}_link_ksettings
Philippe Reynes [Sun, 5 Feb 2017 22:55:19 +0000 (23:55 +0100)]
igbvf: use new API ethtool_{get|set}_link_ksettings

The ethtool API {get|set}_settings is deprecated.
We move this driver to new API {get|set}_link_ksettings.

As I don't have the hardware, I'd be very pleased if
someone may test this patch.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: use new API ethtool_{get|set}_link_ksettings
Philippe Reynes [Sun, 5 Feb 2017 17:55:44 +0000 (18:55 +0100)]
igb: use new API ethtool_{get|set}_link_ksettings

The ethtool API {get|set}_settings is deprecated.
We move this driver to new API {get|set}_link_ksettings.

As I don't have the hardware, I'd be very pleased if
someone may test this patch.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoe1000e: use new API ethtool_{get|set}_link_ksettings
Philippe Reynes [Thu, 26 Jan 2017 21:19:53 +0000 (22:19 +0100)]
e1000e: use new API ethtool_{get|set}_link_ksettings

The ethtool API {get|set}_settings is deprecated.
We move this driver to new API {get|set}_link_ksettings.

As I don't have the hardware, I'd be very pleased if
someone may test this patch.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoe1000: use new API ethtool_{get|set}_link_ksettings
Philippe Reynes [Sat, 21 Jan 2017 15:06:03 +0000 (16:06 +0100)]
e1000: use new API ethtool_{get|set}_link_ksettings

The ethtool API {get|set}_settings is deprecated.
We move this driver to new API {get|set}_link_ksettings.

As I don't have the hardware, I'd be very pleased if
someone may test this patch.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoCleanup some warning from timestamping code.
Ezequiel Lara Gomez [Sat, 11 Mar 2017 20:06:01 +0000 (20:06 +0000)]
Cleanup some warning from timestamping code.

Following checkpatch.pl recommendations (which include
replacing with <linux/io.h> the <asm/io.h>, since linux/io.h includes
it).

Signed-off-by: Ezequiel Lara Gomez <ezegomez@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoEnable tx timestamping on loopback and dummy
Ezequiel Lara Gomez [Sat, 11 Mar 2017 20:06:54 +0000 (20:06 +0000)]
Enable tx timestamping on loopback and dummy

This enables developing code that uses SOF_TIMESTAMPING_TX_SOFTWARE
by using localhost addresses (without needing to send packets outside),
as well as enabling unit and functional testing of TX timestamping code
without needing hardware support or network access.

It also fulfills the expectation of software network devices supporting
software-based timestamping.

Tested on qemu using txtimestamping.c from the kernel selftests, and
ethtool -T.

Signed-off-by: Ezequiel Lara Gomez <ezegomez@amazon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Tue, 21 Mar 2017 21:35:21 +0000 (14:35 -0700)]
Merge branch '40GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
40GbE Intel Wired LAN Driver Updates 2017-03-20

This series contains updates to i40e and i40evf only.

Philippe Reynes updates i40e and i40evf to use the new ethtool API for
{get|set}_link_ksettings.

Jake provides the remaining patches in the series, starting with a fix
for i40e where the firmware expected the port numbers for the offloaded
UDP tunnels in Little Endian format and we were sending them in Big Endian
format which put the wrong port number to be put in the UDP tunnel list.
Changed the driver to use __be32 values instead of arrays for
(src|dst)_ip.  Refactored the exit flow of i40e_add_fdir_ethtool() which
removes the dependency on having a non-zero return value.  Fixed a memory
leak by running kfree() and returning immediately when we fail to add
flow director filter.  Fixed a potential issue where could update the
filter count without actually succeeding in adding a filter, by moving
the ATR exit check to after we have sent the TCP/IPv4 filter to the ring
successfully.  Ensures that the fd_tcp_rule count is reset to 0, before
we reprogram the filters so that we do not end up with a stale count
which does not correctly reflect the number of programmed filters.  Added
a check whether we have TCP/IPv4 filters before re-enabling ATR after
flushing and replaying FDIR filters.  Added counters for each filter
type in preparation for adding code to properly check the mask value.
Fixed potential issues by explicitly checking the flow type at the
start of i40e_add_fdir_ethtool().  To avoid possible memory leaks,
we now unconditionally delete the old filter, even if it is identical to
the new filter and ensures will always update the filters as expected.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf-next
David S. Miller [Tue, 21 Mar 2017 21:28:08 +0000 (14:28 -0700)]
Merge git://git./linux/kernel/git/pablo/nf-next

Pablo Neira Ayuso says:

====================
Netfilter/IPVS updates for net-next

The following patchset contains Netfilter/IPVS updates for your
net-next tree. A couple of new features for nf_tables, and unsorted
cleanups and incremental updates for the Netfilter tree. More
specifically, they are:

1) Allow to check for TCP option presence via nft_exthdr, patch
   from Phil Sutter.

2) Add symmetric hash support to nft_hash, from Laura Garcia Liebana.

3) Use pr_cont() in ebt_log, from Joe Perches.

4) Remove some dead code in arp_tables reported via static analysis
   tool, from Colin Ian King.

5) Consolidate nf_tables expression validation, from Liping Zhang.

6) Consolidate set lookup via nft_set_lookup().

7) Remove unnecessary rcu read lock side in bridge netfilter, from
   Florian Westphal.

8) Remove unused variable in nf_reject_ipv4, from Tahee Yoo.

9) Pass nft_ctx struct to object initialization indirections, from
   Florian Westphal.

10) Add code to integrate conntrack helper into nf_tables, also from
    Florian.

11) Allow to check if interface index or name exists via
    NFTA_FIB_F_PRESENT, from Phil Sutter.

12) Simplify resolve_normal_ct(), from Florian.

13) Use per-limit spinlock in nft_limit and xt_limit, from Liping Zhang.

14) Use rwlock in nft_set_rbtree set, also from Liping Zhang.

15) One patch to remove a useless printk at netns init path in ipvs,
    and several patches to document IPVS knobs.

16) Use refcount_t for reference counter in the Netfilter/IPVS code,
    from Elena Reshetova.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Tue, 21 Mar 2017 21:19:20 +0000 (14:19 -0700)]
Merge branch '1GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
1GbE Intel Wired LAN Driver Updates 2017-03-17

This series contains updates to mainly igb, with one fix for ixgbe.

Alex does all the changes in the series, starting with adding support
for DMA_ATTR_WEAK_ORDERING to improve performance on some platforms.
Modified igb to use the length of the packet instead of the DD status
bit to determine if a new descriptor is ready to be processed.  Modified
the driver to only go through the region in the receive ring that was
designated to be cleaned up, instead of going through the entire ring
on cleanup.  Cleaned up the transmit side, by clearing the transmit
buffer_info only when resetting the rings.  Added a new upper limit for
receive, which is based on the size of a 2K buffer minus padding, which
will allow us to support build_skb going forward.  Fixed ethtool testing
to only sync on the size of the frame that is being tested, instead of
the entire receive buffer.  Updated the handling of page addresses to
always use a void pointer with the consistent name of "va" to indicate
that we are working with a virtual address.  Added a "chicken bit" so
that we can turn off the new receive allocation feature, in the case
where we need to fallback to the legacy receive path.  Added support for
using 3K buffers in order 1 pages the same way we were using 2K buffers
in 4K pages.  Added support for padding packet, since we limit the size
of the frame, we are able to write to an offset within the buffer instead
of having to write at the very start of the buffer.  This allows us to
leaving padding room for things like supporting XDP in the future.
Refactored the receive buffer page management, since there are 2-3 paths
that can be taken depending on what receive modes are enabled, so to
improve maintainability, break out the common bits into their own
functions.  Add support for build_skb, again.  Lastly, fixed a typo in
igb and ixgbe code comments.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoi40e: always remove old filter when adding new FDir filter
Jacob Keller [Mon, 6 Feb 2017 22:39:13 +0000 (14:39 -0800)]
i40e: always remove old filter when adding new FDir filter

The previous code relied on i40e_match_fdir_input_set to determine when
determining whether to free the old filter. Change this code so that we
simply unconditionally delete the old filter, even if it's identical to
the new filter. This ensures that we don't leak any memory, and that we
always update the filters as expected.

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>
7 years agoi40e: explicitly fail on extended MAC field for ethtool_rx_flow_spec
Jacob Keller [Mon, 6 Feb 2017 22:39:12 +0000 (14:39 -0800)]
i40e: explicitly fail on extended MAC field for ethtool_rx_flow_spec

Although we will fail the filter later due to checking flow_type which
will have a bogus invalid type, it is possible future refactoring will
remove this hidden failure case. Avoid a possible issue in the future by
explicitly checking the flow type at the start.

Change-Id: Ia98eb26f7b93ccbe38c7141e8f203ef496fc6598
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>
7 years agoi40e: add counters for UDP/IPv4 and IPv4 filters
Jacob Keller [Mon, 6 Feb 2017 22:38:46 +0000 (14:38 -0800)]
i40e: add counters for UDP/IPv4 and IPv4 filters

In preparation for adding code to properly check the mask values, we
will need to know the number of active filters for each type. Add
counters for each filter type. Rename the already existing fd_tcp_rule
to fd_tcp4_filter_cnt to match the style of other names. To avoid style
warnings, avoid assigning multiple parameters at once, and fix up one
other case where we did so previously.

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>
7 years agoi40e: don't re-enable ATR when flushing filters if SB has TCP4/IPv4 rules
Jacob Keller [Mon, 6 Feb 2017 22:38:45 +0000 (14:38 -0800)]
i40e: don't re-enable ATR when flushing filters if SB has TCP4/IPv4 rules

When flushing and replaying FDIR filters, it is possible we would
disable ATR, and then re-enable it even though we should have kept
it disabled due to existing TCP/IPv4 filters. Fix this by checking
whether we have TCP4/IPv4 filters before re-enabling.

Alternatively, we could instead restore ATR and then replay filters,
however, this would cause us to rapidly enable and then disable ATR in
some cases.

Change-ID: I076e4cc1e4409bce7f98f3c213295433a4ff43d8
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Avinash Dayanand <avinash.dayanand@intel.com>
Reviewed-by: Alan Brady <alan.brady@intel.com>
Reviewed-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: reset fd_tcp_rule count when restoring filters
Jacob Keller [Mon, 6 Feb 2017 22:38:44 +0000 (14:38 -0800)]
i40e: reset fd_tcp_rule count when restoring filters

Since we're about to reprogram the filters, we need to ensure that the
fd_tcp_rule count is correctly reset to 0. Otherwise, we will keep
a stale count that does not accurately reflect the number of programmed
TCPv4 filters.

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>
7 years agoi40e: remove redundant check for fd_tcp_rule when restoring filters
Jacob Keller [Mon, 6 Feb 2017 22:38:43 +0000 (14:38 -0800)]
i40e: remove redundant check for fd_tcp_rule when restoring filters

i40e_fdir_filter_restore re-adds all existing filters, which already
checks when adding a TCPv4 filter to disable ATR. We don't need to make
the check twice, so remove this redundant code.

Change-ID: Ia0b0690e23523915199d601494557def135c9d7f
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>
7 years agoi40e: exit ATR mode only when adding TCP/IPv4 filter succeeds
Jacob Keller [Mon, 6 Feb 2017 22:38:42 +0000 (14:38 -0800)]
i40e: exit ATR mode only when adding TCP/IPv4 filter succeeds

Move ATR exit check after we have sent the TCP/IPv4 filter to the ring
successfully. This avoids an issue where we potentially update the
filter count without actually succeeding in adding the filter. Now, we
only increment the fd_tcp_rule after we've succeeded. Additionally, we
will re-enable ATR mode only after deletion of the filter is actually
posted to the FDIR ring.

Change-ID: If5c1dea422081cc5e2de65618b01b4c3bf6bd586
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: return immediately when failing to add fdir filter
Jacob Keller [Mon, 6 Feb 2017 22:38:41 +0000 (14:38 -0800)]
i40e: return immediately when failing to add fdir filter

Instead of setting err=true and checking this to determine when to free
the raw_packet near the end of the function, simply kfree and return
immediately. The resulting code is a bit cleaner and has one less
variable. This also resolves a subtle bug in the ipv4 case which could
fail to add the first filter and then never free the memory, resulting
in a small memory leak.

Change-ID: I7583aac033481dc794b4acaa14445059c8930ff1
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Avinash Dayanand <avinash.dayanand@intel.com>
Reviewed-by: Alan Brady <alan.brady@intel.com>
Reviewed-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40e: rework exit flow of i40e_add_fdir_ethtool
Jacob Keller [Mon, 6 Feb 2017 22:38:40 +0000 (14:38 -0800)]
i40e: rework exit flow of i40e_add_fdir_ethtool

Refactor the exit flow of the i40e_add_fdir_ethtool function. Move the
input_label to the end of the function, removing the dependency on
having a non-zero return value. Add a comment explaining why it is ok
not to free the fdir data structure, because the structure is now stored
in the fdir_filter_list.

Change-Id: I723342181d59cd0c9f3b31140c37961ba37bb242
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>
7 years agoi40e: don't use arrays for (src|dst)_ip
Jacob Keller [Mon, 6 Feb 2017 22:38:39 +0000 (14:38 -0800)]
i40e: don't use arrays for (src|dst)_ip

The code originally included src_ip and dst_ip with enough space to
support ipv6 filters. However, no actual support for ipv6 filters has
been implemented. Thus, remove the arrays and just use __be32 values.
Should ipv6 support be added in the future, we can replace these with
a union that has sizes for both values.

Change-Id: I1bc04032244a80eb6ebc8a4e6c723a4a665c1dd5
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>
7 years agoi40e: send correct port number to AdminQ when enabling UDP tunnels
Jacob Keller [Mon, 6 Feb 2017 22:38:38 +0000 (14:38 -0800)]
i40e: send correct port number to AdminQ when enabling UDP tunnels

The firmware expects the port numbers for offloaded UDP tunnels in
Little Endian format. We accidentally sent the value in Big Endian
format which obviously will cause the wrong port number to be put into
the UDP tunnels list. This results in VxLAN and Geneve tunnel Rx
offloads being essentially disabled, unless the port number happens to
be identical after byte swapping. Note that i40e_aq_add_udp_tunnel()
will byteswap the parameter from host order into Little Endian so we
don't need worry about passing strictly a __le16 value to the command.

This patch essentially reverts b3f5c7bc88ba ("i40e: Fix for extra byte
swap in tunnel setup", 2016-08-24), but in a way that makes the result
much more clear to the reader.

Fixes: b3f5c7bc88ba ("i40e: Fix for extra byte swap in tunnel setup", 2016-08-24)
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Williams, Mitch A <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoi40evf: use new api ethtool_{get|set}_link_ksettings
Philippe Reynes [Sat, 4 Feb 2017 22:49:03 +0000 (23:49 +0100)]
i40evf: use new api ethtool_{get|set}_link_ksettings

The ethtool api {get|set}_settings is deprecated.
We move this driver to new api {get|set}_link_ksettings.

As I don't have the hardware, I'd be very pleased if
someone may test this patch.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agonetfilter: fix the warning on unused refcount variable
Reshetova, Elena [Mon, 20 Mar 2017 09:43:28 +0000 (11:43 +0200)]
netfilter: fix the warning on unused refcount variable

net/netfilter/nfnetlink_acct.c: In function 'nfnl_acct_try_del':
net/netfilter/nfnetlink_acct.c:329:15: warning: unused variable 'refcount' [-Wunused-variable]
unsigned int refcount;
             ^

Fixes: b54ab92b84b6 ("netfilter: refcounter conversions")
Signed-off-by: Elena Reshetova <elena.reshetova@intel.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agoi40e: use new api ethtool_{get|set}_link_ksettings
Philippe Reynes [Sat, 4 Feb 2017 21:05:06 +0000 (22:05 +0100)]
i40e: use new api ethtool_{get|set}_link_ksettings

The ethtool api {get|set}_settings is deprecated.
We move this driver to new api {get|set}_link_ksettings.

As I don't have the hardware, I'd be very pleased if
someone may test this patch.

Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb/ixgbe: Fix typo in igb_build_skb and/or ixgbe_build_skb code comment
Alexander Duyck [Wed, 15 Feb 2017 17:15:59 +0000 (09:15 -0800)]
igb/ixgbe: Fix typo in igb_build_skb and/or ixgbe_build_skb code comment

There was a typo that I had left in the code comments for the igb and ixgbe
functions that enabled build_skb support.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: Re-add support for build_skb in igb
Alexander Duyck [Tue, 7 Feb 2017 02:27:36 +0000 (18:27 -0800)]
igb: Re-add support for build_skb in igb

This reverts commit f9d40f6a9921 ("igb: Revert support for build_skb in
igb") and adds a few changes to update it to work with the latest version
of igb. We are now able to revert the removal of this due to the fact
that with the recent changes to the page count and the use of
DMA_ATTR_SKIP_CPU_SYNC we can make the pages writable so we should not be
invalidating the additional data added when we call build_skb.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: Break out Rx buffer page management
Alexander Duyck [Tue, 7 Feb 2017 02:27:26 +0000 (18:27 -0800)]
igb: Break out Rx buffer page management

At this point we have 2 to 3 paths that can be taken depending on what Rx
modes are enabled.  In order to better support that and improve the
maintainability I am breaking out the common bits from those paths and
making them into their own functions.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: Add support for padding packet
Alexander Duyck [Tue, 7 Feb 2017 02:27:14 +0000 (18:27 -0800)]
igb: Add support for padding packet

With the size of the frame limited we can now write to an offset within the
buffer instead of having to write at the very start of the buffer.  The
advantage to this is that it allows us to leave padding room for things
like supporting XDP in the future.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: Add support for using order 1 pages to receive large frames
Alexander Duyck [Tue, 7 Feb 2017 02:27:03 +0000 (18:27 -0800)]
igb: Add support for using order 1 pages to receive large frames

This patch adds support for using 3K buffers in order 1 pages the same way
we were using 2K buffers in 4K pages.  We are reserving 1K of room for now
to have space available for future headroom and tailroom when we enable
build_skb support.

One side effect of this patch is that we can end up using a larger buffer
if jumbo frames is enabled.  The impact shouldn't be too great, but it
could hurt small packet performance for UDP workloads if jumbo frames is
enabled as the truesize of frames will be larger.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: Add support for ethtool private flag to allow use of legacy Rx
Alexander Duyck [Tue, 7 Feb 2017 02:26:52 +0000 (18:26 -0800)]
igb: Add support for ethtool private flag to allow use of legacy Rx

Since there are potential drawbacks to the new Rx allocation approach I
thought it best to add a "chicken bit" so that we can turn the feature off
if in the event that a problem is found.

It also provides a means of validating the legacy Rx path in the event that
we are forced to fall back.  At some point in the future when we are
convinced we don't need it anymore we might be able to drop the legacy-rx
flag.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: Use page_address offset from page instead of masking virtual address
Alexander Duyck [Tue, 7 Feb 2017 02:26:40 +0000 (18:26 -0800)]
igb: Use page_address offset from page instead of masking virtual address

Update the handling of page addresses so that we always refer to them using
a void pointer, and try to use the consistent name of va indicating we are
working with a virtual address.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: Only sync size of expected frame in ethtool testing
Alexander Duyck [Tue, 7 Feb 2017 02:26:26 +0000 (18:26 -0800)]
igb: Only sync size of expected frame in ethtool testing

We only need to sync the size of the frame that is read to test.  We don't
need to sync the entire Rx buffer.  This way the testing is more consistent
with how we handle things in the receive path.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: Limit maximum frame Rx based on MTU
Alexander Duyck [Tue, 7 Feb 2017 02:26:15 +0000 (18:26 -0800)]
igb: Limit maximum frame Rx based on MTU

In order to support the use of build_skb going forward it will be necessary
to place a maximum limit on the amount of data we can receive when jumbo
frames is not enabled.  In order to do this I am adding a new upper limit
for receive based on the size of a 2K buffer minus padding.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: Don't bother clearing Tx buffer_info in igb_clean_tx_ring
Alexander Duyck [Tue, 7 Feb 2017 02:26:02 +0000 (18:26 -0800)]
igb: Don't bother clearing Tx buffer_info in igb_clean_tx_ring

In the case of the Tx rings we need to only clear the Tx buffer_info when
we are resetting the rings.  Ideally we do this when we configure the ring
to bring it back up instead of when we are taking it down in order to avoid
dirtying pages we don't need to.

In addition we don't need to clear the Tx descriptor ring since we will
fully repopulate it when we begin transmitting frames and next_to_watch can
be cleared to prevent the ring from being cleaned beyond that point instead
of needing to touch anything in the Tx descriptor ring.

Finally with these changes we can avoid having to reset the skb member of
the Tx buffer_info structure in the cleanup path since the skb will always
be associated with the first buffer which has next_to_watch set.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: Clear Rx buffer_info in configure instead of clean
Alexander Duyck [Tue, 7 Feb 2017 02:25:50 +0000 (18:25 -0800)]
igb: Clear Rx buffer_info in configure instead of clean

This change makes it so that instead of going through the entire ring on Rx
cleanup we only go through the region that was designated to be cleaned up
and stop when we reach the region where new allocations should start.

In addition we can avoid having to perform a memset on the Rx buffer_info
structures until we are about to start using the ring again.  By deferring
this we can avoid dirtying the cache any more than we have to which can
help to improve the time needed to bring the interface down and then back
up again in a reset or suspend/resume cycle.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: Use length to determine if descriptor is done
Alexander Duyck [Tue, 7 Feb 2017 02:25:41 +0000 (18:25 -0800)]
igb: Use length to determine if descriptor is done

This change makes it so that we use the length of the packet instead of the
DD status bit to determine if a new descriptor is ready to be processed.
The obvious advantage is that it cuts down on reads as we don't really even
need the DD bit if going from a 0 to a non-zero value on size is enough to
inform us that the packet has been completed.

In addition I have updated the code so that we only reset the Rx descriptor
length for descriptor zero when resetting a ring instead of having to do a
memset with 0 over the entire ring.  By doing this we can save some time on
initialization.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agoigb: Add support for DMA_ATTR_WEAK_ORDERING
Alexander Duyck [Tue, 7 Feb 2017 02:25:26 +0000 (18:25 -0800)]
igb: Add support for DMA_ATTR_WEAK_ORDERING

Since we are already using DMA attributes in igb for Rx there is no reason
why we can't also apply DMA_ATTR_WEAK_ORDERING which is needed on some
platforms to improve performance.

Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
7 years agonetfilter: refcounter conversions
Reshetova, Elena [Thu, 16 Mar 2017 08:03:34 +0000 (10:03 +0200)]
netfilter: refcounter conversions

refcount_t type and corresponding API (see include/linux/refcount.h)
should be used instead of atomic_t when the variable is used as
a reference counter. This allows to avoid accidental
refcounter overflows that might lead to use-after-free
situations.

Signed-off-by: Elena Reshetova <elena.reshetova@intel.com>
Signed-off-by: Hans Liljestrand <ishkamiel@gmail.com>
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David Windsor <dwindsor@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
7 years agoliquidio: fix wrong information about link modes reported to ethtool
Manish Awasthi [Thu, 16 Mar 2017 23:16:17 +0000 (16:16 -0700)]
liquidio: fix wrong information about link modes reported to ethtool

Information reported to ethtool about link modes is wrong for 25G NIC.  Fix
it by checking for presence of 25G NIC, checking the link speed reported by
NIC firmware, and then assigning proper values to the
ethtool_link_ksettings struct.

Signed-off-by: Manish Awasthi <manish.awasthi@cavium.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'netvsc-small-changes'
David S. Miller [Fri, 17 Mar 2017 04:39:51 +0000 (21:39 -0700)]
Merge branch 'netvsc-small-changes'

Stephen Hemminger says:

====================
netvsc: small changes for net-next

One bugfix, and two non-code patches
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonetvsc: remove unused #define
stephen hemminger [Thu, 16 Mar 2017 23:12:39 +0000 (16:12 -0700)]
netvsc: remove unused #define

Not used anywhere.

Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonetvsc: add comments about callback's and NAPI
stephen hemminger [Thu, 16 Mar 2017 23:12:38 +0000 (16:12 -0700)]
netvsc: add comments about callback's and NAPI

Add some short description of how callback's and NAPI interoperate.

Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonetvsc: avoid race with callback
stephen hemminger [Thu, 16 Mar 2017 23:12:37 +0000 (16:12 -0700)]
netvsc: avoid race with callback

Change the argument to channel callback from the channel pointer
to the internal data structure containing per-channel info.
This avoids any possible races when callback happens during
initialization and makes IRQ code simpler.

Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'bpf-inline-lookups'
David S. Miller [Fri, 17 Mar 2017 03:44:12 +0000 (20:44 -0700)]
Merge branch 'bpf-inline-lookups'

Alexei Starovoitov says:

====================
bpf: inline bpf_map_lookup_elem()

bpf_map_lookup_elem() is one of the most frequently used helper functions.
Improve JITed program performance by inlining this helper.

bpf_map_type before  after
hash 58M 74M
array 174M 280M

The values are number of lookups per second in ideal conditions
measured by micro-benchmark in patch 6.

The 'perf report' for HASH map type:
before:
    54.23%  map_perf_test  [kernel.kallsyms]  [k] __htab_map_lookup_elem
    14.24%  map_perf_test  [kernel.kallsyms]  [k] lookup_elem_raw
     8.84%  map_perf_test  [kernel.kallsyms]  [k] htab_map_lookup_elem
     5.93%  map_perf_test  [kernel.kallsyms]  [k] bpf_map_lookup_elem
     2.30%  map_perf_test  [kernel.kallsyms]  [k] bpf_prog_da4fc6a3f41761a2
     1.49%  map_perf_test  [kernel.kallsyms]  [k] kprobe_ftrace_handler

after:
    60.03%  map_perf_test  [kernel.kallsyms]  [k] __htab_map_lookup_elem
    18.07%  map_perf_test  [kernel.kallsyms]  [k] lookup_elem_raw
     2.91%  map_perf_test  [kernel.kallsyms]  [k] bpf_prog_da4fc6a3f41761a2
     1.94%  map_perf_test  [kernel.kallsyms]  [k] _einittext
     1.90%  map_perf_test  [kernel.kallsyms]  [k] __audit_syscall_exit
     1.72%  map_perf_test  [kernel.kallsyms]  [k] kprobe_ftrace_handler

so the cost of htab_map_lookup_elem() and bpf_map_lookup_elem()
is gone after inlining.

'per-cpu' and 'lru' map types can be optimized similarly in the future.

Note the sparse will complain that bpf is addictive ;)
kernel/bpf/hashtab.c:438:19: sparse: subtraction of functions? Share your drugs
kernel/bpf/verifier.c:3342:38: sparse: subtraction of functions? Share your drugs
it's not a new warning, just in new places.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosamples/bpf: add map_lookup microbenchmark
Alexei Starovoitov [Thu, 16 Mar 2017 01:26:44 +0000 (18:26 -0700)]
samples/bpf: add map_lookup microbenchmark

$ map_perf_test 128
speed of HASH bpf_map_lookup_elem() in lookups per second
w/o JIT w/JIT
before 46M 58M
after 42M 74M

perf report
before:
    54.23%  map_perf_test  [kernel.kallsyms]  [k] __htab_map_lookup_elem
    14.24%  map_perf_test  [kernel.kallsyms]  [k] lookup_elem_raw
     8.84%  map_perf_test  [kernel.kallsyms]  [k] htab_map_lookup_elem
     5.93%  map_perf_test  [kernel.kallsyms]  [k] bpf_map_lookup_elem
     2.30%  map_perf_test  [kernel.kallsyms]  [k] bpf_prog_da4fc6a3f41761a2
     1.49%  map_perf_test  [kernel.kallsyms]  [k] kprobe_ftrace_handler

after:
    60.03%  map_perf_test  [kernel.kallsyms]  [k] __htab_map_lookup_elem
    18.07%  map_perf_test  [kernel.kallsyms]  [k] lookup_elem_raw
     2.91%  map_perf_test  [kernel.kallsyms]  [k] bpf_prog_da4fc6a3f41761a2
     1.94%  map_perf_test  [kernel.kallsyms]  [k] _einittext
     1.90%  map_perf_test  [kernel.kallsyms]  [k] __audit_syscall_exit
     1.72%  map_perf_test  [kernel.kallsyms]  [k] kprobe_ftrace_handler

Notice that bpf_map_lookup_elem() and htab_map_lookup_elem() are trivial
functions, yet they take sizeable amount of cpu time.
htab_map_gen_lookup() removes bpf_map_lookup_elem() and converts
htab_map_lookup_elem() into three BPF insns which causing cpu time
for bpf_prog_da4fc6a3f41761a2() slightly increase.

$ map_perf_test 256
speed of ARRAY bpf_map_lookup_elem() in lookups per second
w/o JIT w/JIT
before 97M 174M
after 64M 280M

before:
    37.33%  map_perf_test  [kernel.kallsyms]  [k] array_map_lookup_elem
    13.95%  map_perf_test  [kernel.kallsyms]  [k] bpf_map_lookup_elem
     6.54%  map_perf_test  [kernel.kallsyms]  [k] bpf_prog_da4fc6a3f41761a2
     4.57%  map_perf_test  [kernel.kallsyms]  [k] kprobe_ftrace_handler

after:
    32.86%  map_perf_test  [kernel.kallsyms]  [k] bpf_prog_da4fc6a3f41761a2
     6.54%  map_perf_test  [kernel.kallsyms]  [k] kprobe_ftrace_handler

array_map_gen_lookup() removes calls to array_map_lookup_elem()
and bpf_map_lookup_elem() and replaces them with 7 bpf insns.

The performance without JIT is slower, since executing extra insns
in the interpreter is slower than running native C code,
but with JIT the performance gains are obvious,
since native C->x86 code is replaced with fewer bpf->x86 instructions.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: inline htab_map_lookup_elem()
Alexei Starovoitov [Thu, 16 Mar 2017 01:26:43 +0000 (18:26 -0700)]
bpf: inline htab_map_lookup_elem()

Optimize:
bpf_call
  bpf_map_lookup_elem
    map->ops->map_lookup_elem
      htab_map_lookup_elem
        __htab_map_lookup_elem
into:
bpf_call
  __htab_map_lookup_elem

to improve performance of JITed programs.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: add helper inlining infra and optimize map_array lookup
Alexei Starovoitov [Thu, 16 Mar 2017 01:26:42 +0000 (18:26 -0700)]
bpf: add helper inlining infra and optimize map_array lookup

Optimize bpf_call -> bpf_map_lookup_elem() -> array_map_lookup_elem()
into a sequence of bpf instructions.
When JIT is on the sequence of bpf instructions is the sequence
of native cpu instructions with significantly faster performance
than indirect call and two function's prologue/epilogue.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: adjust insn_aux_data when patching insns
Alexei Starovoitov [Thu, 16 Mar 2017 01:26:41 +0000 (18:26 -0700)]
bpf: adjust insn_aux_data when patching insns

convert_ctx_accesses() replaces single bpf instruction with a set of
instructions. Adjust corresponding insn_aux_data while patching.
It's needed to make sure subsequent 'for(all insn)' loops
have matching insn and insn_aux_data.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: refactor fixup_bpf_calls()
Alexei Starovoitov [Thu, 16 Mar 2017 01:26:40 +0000 (18:26 -0700)]
bpf: refactor fixup_bpf_calls()

reduce indent and make it iterate over instructions similar to
convert_ctx_accesses(). Also convert hard BUG_ON into soft verifier error.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobpf: move fixup_bpf_calls() function
Alexei Starovoitov [Thu, 16 Mar 2017 01:26:39 +0000 (18:26 -0700)]
bpf: move fixup_bpf_calls() function

no functional change.
move fixup_bpf_calls() to verifier.c
it's being refactored in the next patch

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp: remove tcp_tw_recycle
Soheil Hassas Yeganeh [Wed, 15 Mar 2017 20:30:46 +0000 (16:30 -0400)]
tcp: remove tcp_tw_recycle

The tcp_tw_recycle was already broken for connections
behind NAT, since the per-destination timestamp is not
monotonically increasing for multiple machines behind
a single destination address.

After the randomization of TCP timestamp offsets
in commit 8a5bd45f6616 (tcp: randomize tcp timestamp offsets
for each connection), the tcp_tw_recycle is broken for all
types of connections for the same reason: the timestamps
received from a single machine is not monotonically increasing,
anymore.

Remove tcp_tw_recycle, since it is not functional. Also, remove
the PAWSPassive SNMP counter since it is only used for
tcp_tw_recycle, and simplify tcp_v4_route_req and tcp_v6_route_req
since the strict argument is only set when tcp_tw_recycle is
enabled.

Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Cc: Lutz Vieweg <lvml@5t9.de>
Cc: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp: remove per-destination timestamp cache
Soheil Hassas Yeganeh [Wed, 15 Mar 2017 20:30:45 +0000 (16:30 -0400)]
tcp: remove per-destination timestamp cache

Commit 8a5bd45f6616 (tcp: randomize tcp timestamp offsets for each connection)
randomizes TCP timestamps per connection. After this commit,
there is no guarantee that the timestamps received from the
same destination are monotonically increasing. As a result,
the per-destination timestamp cache in TCP metrics (i.e., tcpm_ts
in struct tcp_metrics_block) is broken and cannot be relied upon.

Remove the per-destination timestamp cache and all related code
paths.

Note that this cache was already broken for caching timestamps of
multiple machines behind a NAT sharing the same address.

Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Cc: Lutz Vieweg <lvml@5t9.de>
Cc: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'sunvnet-better-connection-management'
David S. Miller [Fri, 17 Mar 2017 03:29:55 +0000 (20:29 -0700)]
Merge branch 'sunvnet-better-connection-management'

Shannon Nelson says:

====================
sunvnet: better connection management

These patches remove some problems in handling of carrier state
with the ldmvsw vswitch, remove  an xoff misuse in sunvnet, and
add stats for debug and tracking of point-to-point connections
between the ldom VMs.

v2:
 - added ldmvsw ndo_open to reset the LDC channel
 - updated copyrights
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosunvnet: xoff not needed when removing port link
Shannon Nelson [Tue, 14 Mar 2017 17:24:43 +0000 (10:24 -0700)]
sunvnet: xoff not needed when removing port link

The sunvnet netdev is connected to the controlling ldom's vswitch
for network bridging.  However, for higher performance between ldoms,
there also is a channel between each client ldom.  These connections are
represented in the sunvnet driver by a queue for each ldom.  The driver
uses select_queue to tell the stack which queue to use by tracking the mac
addresses on the other end of each port.  When a connected ldom shuts down,
the driver receives an LDC_EVENT_RESET and the port is removed from the
driver, thus a queue with no ldom on the other end will never be selected
for Tx.

The driver was trying to reinforce the "don't use this queue" notion with
netif_tx_stop_queue() and netif_tx_wake_queue(), which really should only
be used to signal a Tx queue is full (aka XOFF).  This misuse of queue
state resulted in NETDEV WATCHDOG messages and lots of unnecessary calls
into the driver's tx_timeout handler.  Simply removing these takes care
of the problem.

Orabug: 25190537

Signed-off-by: Shannon Nelson <shannon.nelson@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosunvnet: count multicast packets
Shannon Nelson [Tue, 14 Mar 2017 17:24:42 +0000 (10:24 -0700)]
sunvnet: count multicast packets

Make sure multicast packets get counted in the device.

Orabug: 25190537

Signed-off-by: Shannon Nelson <shannon.nelson@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosunvnet: track port queues correctly
Shannon Nelson [Tue, 14 Mar 2017 17:24:41 +0000 (10:24 -0700)]
sunvnet: track port queues correctly

Track our used and unused queue indexies correctly.  Otherwise, as ports
dropped out and returned, they all eventually ended up with the same
queue index.

Orabug: 25190537

Signed-off-by: Shannon Nelson <shannon.nelson@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agosunvnet: add stats to track ldom to ldom packets and bytes
Shannon Nelson [Tue, 14 Mar 2017 17:24:40 +0000 (10:24 -0700)]
sunvnet: add stats to track ldom to ldom packets and bytes

In this driver, there is a "port" created for the connection to each of
the other ldoms; a netdev queue is mapped to each port, and they are
collected under a single netdev.  The generic netdev statistics show
us all the traffic in and out of our network device, but don't show
individual queue/port stats.  This patch breaks out the traffic counts
for the individual ports and gives us a little view into the state of
those connections.

Orabug: 25190537

Signed-off-by: Shannon Nelson <shannon.nelson@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoldmvsw: better use of link up and down on ldom vswitch
Shannon Nelson [Tue, 14 Mar 2017 17:24:39 +0000 (10:24 -0700)]
ldmvsw: better use of link up and down on ldom vswitch

When an ldom VM is bound, the network vswitch infrastructure is set up for
it, but was being forced 'UP' by the userland switch configuration script.
When 'UP' but not actually connected to a running VM, the ipv6 neighbor
probes fail (not a horrible thing) and start cluttering up the kernel logs.
Funny thing: these are debug messages that never actually show up, but
we do see the net_ratelimited messages that say N callbacks were
suppressed.

This patch defers the netif_carrier_on() until an actual link has been
established with the VM, as indicated by receiving an LDC_EVENT_UP from
the underlying LDC protocol.  Similarly, we take the link down when we
see the LDC_EVENT_RESET.  Now when we see the ndo_open(), we reset the
link to get things talking again.

Orabug: 25525312

Signed-off-by: Shannon Nelson <shannon.nelson@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agobonding: add 802.3ad support for 25G speeds
Jarod Wilson [Tue, 14 Mar 2017 15:48:32 +0000 (11:48 -0400)]
bonding: add 802.3ad support for 25G speeds

Cut-n-paste enablement of 802.3ad bonding on 25G NICs, which currently
report 0 as their bandwidth.

CC: Jay Vosburgh <j.vosburgh@gmail.com>
CC: Veaceslav Falico <vfalico@gmail.com>
CC: Andy Gospodarek <andy@greyhouse.net>
CC: netdev@vger.kernel.org
Signed-off-by: Jarod Wilson <jarod@redhat.com>
Acked-by: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agotcp_westwood: fix tcp_westwood_info() style mistakes
chun Long [Tue, 14 Mar 2017 07:26:24 +0000 (15:26 +0800)]
tcp_westwood: fix tcp_westwood_info() style mistakes

replace comma to semi colons in tcp_westwood_info().
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio: use meaningful names for IRQs
Rick Farrington [Mon, 13 Mar 2017 19:58:04 +0000 (12:58 -0700)]
liquidio: use meaningful names for IRQs

All IRQs owned by the PF and VF drivers share the same nondescript name
"octeon"; this makes it difficult to setup interrupt affinity.

Change the IRQ names to reflect their specific purpose:

    LiquidIO<id>-<func>-<type>-<queue pair num>

Examples:
    LiquidIO0-pf0-rxtx-3
    LiquidIO1-vf1-rxtx-0
    LiquidIO0-pf0-aux

We cannot use netdev->name for naming the IRQs because:

    1.  Early during init, the PF and VF drivers require interrupts to
        send/receive control data from the NIC firmware; so the PF and VF
        must request IRQs long before the netdev struct is registered.

    2.  The IRQ name can only be specified at the time it is requested.
        It cannot be changed after that.

Signed-off-by: Rick Farrington <ricardo.farrington@cavium.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: Satanand Burla <satananda.burla@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoliquidio: remove/replace invalid code
Rick Farrington [Mon, 13 Mar 2017 19:07:32 +0000 (12:07 -0700)]
liquidio: remove/replace invalid code

Remove invalid call to dma_sync_single_for_cpu() because previous DMA
allocation was coherent--not streaming.  Remove code that references fields
in struct list_head; replace it with calls to list_empty() and
list_first_entry().  Also, add comment to clarify complicated if statement.

Signed-off-by: Rick Farrington <ricardo.farrington@cavium.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: Derek Chickles <derek.chickles@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonetem: apply correct delay when rate throttling
Nik Unger [Mon, 13 Mar 2017 17:16:58 +0000 (10:16 -0700)]
netem: apply correct delay when rate throttling

I recently reported on the netem list that iperf network benchmarks
show unexpected results when a bandwidth throttling rate has been
configured for netem. Specifically:

1) The measured link bandwidth *increases* when a higher delay is added
2) The measured link bandwidth appears higher than the specified limit
3) The measured link bandwidth for the same very slow settings varies significantly across
  machines

The issue can be reproduced by using tc to configure netem with a
512kbit rate and various (none, 1us, 50ms, 100ms, 200ms) delays on a
veth pair between network namespaces, and then using iperf (or any
other network benchmarking tool) to test throughput. Complete detailed
instructions are in the original email chain here:
https://lists.linuxfoundation.org/pipermail/netem/2017-February/001672.html

There appear to be two underlying bugs causing these effects:

- The first issue causes long delays when the rate is slow and no
  delay is configured (e.g., "rate 512kbit"). This is because SKBs are
  not orphaned when no delay is configured, so orphaning does not
  occur until *after* the rate-induced delay has been applied. For
  this reason, adding a tiny delay (e.g., "rate 512kbit delay 1us")
  dramatically increases the measured bandwidth.

- The second issue is that rate-induced delays are not correctly
  applied, allowing SKB delays to occur in parallel. The indended
  approach is to compute the delay for an SKB and to add this delay to
  the end of the current queue. However, the code does not detect
  existing SKBs in the queue due to improperly testing sch->q.qlen,
  which is nonzero even when packets exist only in the
  rbtree. Consequently, new SKBs do not wait for the current queue to
  empty. When packet delays vary significantly (e.g., if packet sizes
  are different), then this also causes unintended reordering.

I modified the code to expect a delay (and orphan the SKB) when a rate
is configured. I also added some defensive tests that correctly find
the latest scheduled delivery time, even if it is (unexpectedly) for a
packet in sch->q. I have tested these changes on the latest kernel
(4.11.0-rc1+) and the iperf / ping test results are as expected.

Signed-off-by: Nik Unger <njunger@uwaterloo.ca>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'sched-cleanups'
David S. Miller [Thu, 16 Mar 2017 19:02:15 +0000 (12:02 -0700)]
Merge branch 'sched-cleanups'

Or Gerlitz says:

====================
small set of sched cleanups

Just two cleanups -- but for the 2nd one I think we need ack from
Cong Wang to make sure this isn't actually a bug report..

changes from V1:
  - addressed comment from Sergei to use 12 hex digits etc
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/sched: fq_codel: Avoid set-but-unused variable
Or Gerlitz [Thu, 16 Mar 2017 10:53:42 +0000 (12:53 +0200)]
net/sched: fq_codel: Avoid set-but-unused variable

The code introduced by commit 2ccccf5fb43f ("net_sched: update
hierarchical backlog too") only sets prev_backlog in fq_codel_dequeue()
but not using that anywhere, remove that setting.

Cc: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet/sched: act_ife: Staticfy find_decode_metaid()
Or Gerlitz [Thu, 16 Mar 2017 10:53:41 +0000 (12:53 +0200)]
net/sched: act_ife: Staticfy find_decode_metaid()

As it's used only on that file.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: bgmac: Allow MAC address to be specified in DTB
Steve Lin [Thu, 16 Mar 2017 15:48:58 +0000 (11:48 -0400)]
net: ethernet: bgmac: Allow MAC address to be specified in DTB

Allows the BCMA version of the bgmac driver to obtain MAC address
from the device tree.  If no MAC address is specified there, then
the previous behavior (obtaining MAC address from SPROM) is
used.

Signed-off-by: Steve Lin <steven.lin1@broadcom.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Acked-by: Jon Mason <jon.mason@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: ethernet: fs_enet: Remove useless includes
Christophe Leroy [Thu, 16 Mar 2017 09:18:04 +0000 (10:18 +0100)]
net: ethernet: fs_enet: Remove useless includes

CONFIG_8xx is being deprecated. Since the includes dependent on
CONFIG_8xx are useless, just drop them.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoisdn: hardware: mISDN: Remove reference to CONFIG_8xx
Christophe Leroy [Thu, 16 Mar 2017 09:18:02 +0000 (10:18 +0100)]
isdn: hardware: mISDN: Remove reference to CONFIG_8xx

CONFIG_8xx is deprecated and should soon be removed in favor
of CONFIG_PPC_8xx.
Anyway, hfc_multi_8xx.h only uses 8xx I/O ports which are
linked to the CPM1 communication processor included in the 8xx
rather than the 8xx itself.

This patch therefore makes it dependent on CONFIG_CPM1 instead,
like several other drivers.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: mvneta: support suspend and resume
Jane Li [Thu, 16 Mar 2017 08:22:28 +0000 (16:22 +0800)]
net: mvneta: support suspend and resume

Add basic support for handling suspend and resume.

Signed-off-by: Jane Li <jiel@marvell.com>
Reviewed-by: Jisheng Zhang <jszhang@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'mlxsw-vrf'
David S. Miller [Thu, 16 Mar 2017 17:18:35 +0000 (10:18 -0700)]
Merge branch 'mlxsw-vrf'

Jiri Pirko says:

====================
mlxsw: Enable VRF offload

Ido says:

Packets received from netdevs enslaved to different VRF devices are
forwarded using different FIB tables. In the Spectrum ASIC this is
achieved by binding different router interfaces (RIFs) to different
virtual routers (VRs). Each RIF represents an enslaved netdev and each
VR has its own FIB table according to which packets are forwarded.

The first three patches add an helper to check if a FIB rule is a
default rule and extend the FIB notification chain to include the rule's
info as part of the RULE_{ADD,DEL} events. This allows offloading
drivers to sanitize the rules they don't support and flush their tables.

The fourth patch introduces a small change in the VRF driver to allow
capable drivers to more easily offload VRFs.

Finally, the last patches gradually add support for VRFs in the mlxsw
driver. First, on top of port netdevs, stacked LAG and VLAN devices and
then on top of bridges.

Some limitations I would like to point out:

1) The old model where 'oif' / 'iif' rules were programmed for each L3
master device isn't supported. Upon insertion of these rules the driver
will flush its tables and forwarding will be done by the kernel instead.
It's inferior in every way to the single 'l3mdev' rule, so this shouldn't
be an issue.

2) Inter-VRF routes pointing to a VRF device aren't offloaded. Packets
hitting these routes will be forwarded by the kernel. Inter-VRF routes
pointing to netdevs enslaved to a different VRF are offloaded.

3) There's a small discrepancy between the kernel's datapath and the
device's. By default, packets forwarded by the kernel first do a lookup
in the local table and then in the VRF's table (assuming no match). In
the device, lookup is done only in the VRF's table, which is probably
the intended behavior. Changes in v2 allow user to properly re-order the
default rules without triggering the abort mechanism.

Changes in v3:
* Remove 'l3mdev' from the matchall list, as it's related to the action
  and not the selector (David Ahern).
* Use container_of() instead of typecasting (David Ahern).
* Add David's Acked-by to the second patch.
* Add an helper in IPv4 code to check if rule is a default rule (David
  Ahern).

Changes in v2:
* Drop default rule indication and allow re-ordering of default rules
  (David Ahern).
* Remove ifdef around 'struct fib_rule_notifier_info' and drop redundant
  dependency on IP_MULTIPLE_TABLES from rocker and mlxsw.
* Add David's Acked-by to the fourth patch.
* Remove netif_is_vrf_master() and use netif_is_l3_master() instead
  (David Ahern).
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: spectrum_router: Don't abort on l3mdev rules
Ido Schimmel [Thu, 16 Mar 2017 08:08:20 +0000 (09:08 +0100)]
mlxsw: spectrum_router: Don't abort on l3mdev rules

Now that port netdevs can be enslaved to a VRF master we need to make
sure the device's routing tables won't be flushed upon the insertion of
a l3mdev rule.

Note that we assume the notified l3mdev rule is a simple rule as used by
the VRF master. We don't check for the presence of other selectors such
as 'iif' and 'oif'.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: spectrum_router: Add support for VRFs on top of bridges
Ido Schimmel [Thu, 16 Mar 2017 08:08:19 +0000 (09:08 +0100)]
mlxsw: spectrum_router: Add support for VRFs on top of bridges

In a similar fashion to the previous patch, allow bridges and VLAN
devices on top of bridges to be enslaved to a VRF master device.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: spectrum_router: Add support for VRFs
Ido Schimmel [Thu, 16 Mar 2017 08:08:18 +0000 (09:08 +0100)]
mlxsw: spectrum_router: Add support for VRFs

Allow port netdevs, LAG and VLAN devices stacked on top of these to be
enslaved to a VRF master device.

Upon enslavement, create a router interface (RIF) for the enslaved
netdev and associate it with a virtual router (VR) based on the VRF's
table ID.

If a RIF already exists for the netdev (f.e., due to the existence of an
IP address), then it's deleted and a new one is created with the
appropriate VR binding.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: spectrum_router: Don't destroy RIF if L3 slave
Ido Schimmel [Thu, 16 Mar 2017 08:08:17 +0000 (09:08 +0100)]
mlxsw: spectrum_router: Don't destroy RIF if L3 slave

We usually destroy the netdev's router interface (RIF) when the last IP
address is removed from it.

However, we shouldn't do that if it's enslaved to an L3 master device.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomlxsw: spectrum_router: Associate RIFs with correct VR
Ido Schimmel [Thu, 16 Mar 2017 08:08:16 +0000 (09:08 +0100)]
mlxsw: spectrum_router: Associate RIFs with correct VR

When a router interface (RIF) is created due to a netdev being enslaved
to a VRF master, then it should be associated with the appropriate
virtual router (VR) and not the default one.

If netdev is a VRF slave, lookup the VR based on the VRF's table ID.
Otherwise default to the MAIN table.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: vrf: Set slave's private flag before linking
Ido Schimmel [Thu, 16 Mar 2017 08:08:15 +0000 (09:08 +0100)]
net: vrf: Set slave's private flag before linking

Allow listeners of the subsequent CHANGEUPPER notification to retrieve
the VRF's table ID by calling l3mdev_fib_table() with the slave netdev.
Without this change, the netdev won't be considered an L3 slave and the
function would return 0.

This is consistent with other master device such as bridge and bond that
set the slave's private flag before linking. It also makes
do_vrf_{add,del}_slave() symmetric.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoipv4: fib_rules: Dump FIB rules when registering FIB notifier
Ido Schimmel [Thu, 16 Mar 2017 08:08:14 +0000 (09:08 +0100)]
ipv4: fib_rules: Dump FIB rules when registering FIB notifier

In commit c3852ef7f2f8 ("ipv4: fib: Replay events when registering FIB
notifier") we dumped the FIB tables and replayed the events to the
passed notification block.

However, we merely sent a RULE_ADD notification in case custom rules
were in use. As explained in previous patches, this approach won't work
anymore. Instead, we should notify the caller about all the FIB rules
and let it act accordingly.

Upon registration to the FIB notification chain, replay a RULE_ADD
notification for each programmed FIB rule, custom or not. The integrity
of the dump is ensured by the mechanism introduced in the above
mentioned commit.

Prevent regressions by making sure current listeners correctly sanitize
the notified rules.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoipv4: fib_rules: Add notifier info to FIB rules notifications
Ido Schimmel [Thu, 16 Mar 2017 08:08:13 +0000 (09:08 +0100)]
ipv4: fib_rules: Add notifier info to FIB rules notifications

Whenever a FIB rule is added or removed, a notification is sent in the
FIB notification chain. However, listeners don't have a way to tell
which rule was added or removed.

This is problematic as we would like to give listeners the ability to
decide which action to execute based on the notified rule. Specifically,
offloading drivers should be able to determine if they support the
reflection of the notified FIB rule and flush their LPM tables in case
they don't.

Do that by adding a notifier info to these notifications and embed the
common FIB rule struct in it.

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Acked-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoipv4: fib_rules: Check if rule is a default rule
Ido Schimmel [Thu, 16 Mar 2017 08:08:12 +0000 (09:08 +0100)]
ipv4: fib_rules: Check if rule is a default rule

Currently, when non-default (custom) FIB rules are used, devices capable
of layer 3 offloading flush their tables and let the kernel do the
forwarding instead.

When these devices' drivers are loaded they register to the FIB
notification chain, which lets them know about the existence of any
custom FIB rules. This is done by sending a RULE_ADD notification based
on the value of 'net->ipv4.fib_has_custom_rules'.

This approach is problematic when VRF offload is taken into account, as
upon the creation of the first VRF netdev, a l3mdev rule is programmed
to direct skbs to the VRF's table.

Instead of merely reading the above value and sending a single RULE_ADD
notification, we should iterate over all the FIB rules and send a
detailed notification for each, thereby allowing offloading drivers to
sanitize the rules they don't support and potentially flush their
tables.

While l3mdev rules are uniquely marked, the default rules are not.
Therefore, when they are being notified they might invoke offloading
drivers to unnecessarily flush their tables.

Solve this by adding an helper to check if a FIB rule is a default rule.
Namely, its selector should match all packets and its action should
point to the local, main or default tables.

As noted by David Ahern, uniquely marking the default rules is
insufficient. When using VRFs, it's common to avoid false hits by moving
the rule for the local table to just before the main table:

Default configuration:
$ ip rule show
0:      from all lookup local
32766:  from all lookup main
32767:  from all lookup default

Common configuration with VRFs:
$ ip rule show
1000:   from all lookup [l3mdev-table]
32765:  from all lookup local
32766:  from all lookup main
32767:  from all lookup default

Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agor8152: simply the arguments
hayeswang [Thu, 16 Mar 2017 06:32:22 +0000 (14:32 +0800)]
r8152: simply the arguments

Replace &tp->napi with napi and tp->netdev with netdev.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoipvs: Document sysctl pmtu_disc
Hangbin Liu [Mon, 20 Feb 2017 08:31:38 +0000 (16:31 +0800)]
ipvs: Document sysctl pmtu_disc

Document sysctl pmtu_disc based on commit 3654e61137db ("ipvs: add
pmtu_disc option to disable IP DF for TUN packets").

Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: Simon Horman <horms@verge.net.au>
7 years agoipvs: Document sysctl sync_ports
Hangbin Liu [Mon, 20 Feb 2017 08:31:37 +0000 (16:31 +0800)]
ipvs: Document sysctl sync_ports

Document sysctl sync_ports based on commit f73181c8288f ("ipvs: add support
for sync threads").

Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: Simon Horman <horms@verge.net.au>
7 years agoipvs: Document sysctl sync_qlen_max and sync_sock_size
Hangbin Liu [Mon, 20 Feb 2017 08:31:36 +0000 (16:31 +0800)]
ipvs: Document sysctl sync_qlen_max and sync_sock_size

Document sysctl sync_qlen_max and sync_sock_size based on
commit 1c003b1580e2 ("ipvs: wakeup master thread").

Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: Simon Horman <horms@verge.net.au>
7 years agoipvs: fix sync_threshold description and add sync_refresh_period, sync_retries
Hangbin Liu [Mon, 20 Feb 2017 08:31:35 +0000 (16:31 +0800)]
ipvs: fix sync_threshold description and add sync_refresh_period, sync_retries

Fix sync_threshold description which should have two values. Also add
sync_refresh_period and sync_retries based on commit 749c42b620a9
("ipvs: reduce sync rate with time thresholds").

Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: Simon Horman <horms@verge.net.au>
7 years agoipvs: remove an annoying printk in netns init
Cong Wang [Sat, 10 Dec 2016 05:09:59 +0000 (21:09 -0800)]
ipvs: remove an annoying printk in netns init

At most it is used for debugging purpose, but I don't think
it is even useful for debugging, just remove it.

Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Simon Horman <horms@verge.net.au>
7 years agoMerge branch 'xgene-bug-fixes'
David S. Miller [Thu, 16 Mar 2017 04:52:53 +0000 (21:52 -0700)]
Merge branch 'xgene-bug-fixes'

Iyappan Subramanian says:

====================
drivers: net: xgene: Bug fixes and errata workarounds

This patch set addresses bug fixes and errata workarounds.
====================

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: Quan Nguyen <qnguyen@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMAINTAINERS: Update X-Gene SoC ethernet maintainer
Iyappan Subramanian [Wed, 15 Mar 2017 20:27:21 +0000 (13:27 -0700)]
MAINTAINERS: Update X-Gene SoC ethernet maintainer

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: Keyur Chudgar <kchudgar@apm.com>
Signed-off-by: Quan Nguyen <qnguyen@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodrivers: net: xgene: Add workaround for errata 10GE_8/ENET_11
Iyappan Subramanian [Wed, 15 Mar 2017 20:27:20 +0000 (13:27 -0700)]
drivers: net: xgene: Add workaround for errata 10GE_8/ENET_11

This patch implements workaround for errata 10GE_8 and ENET_11:
"HW reports length error for valid 64 byte frames with len <46 bytes"
by recovering them from error.

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: Quan Nguyen <qnguyen@apm.com>
Signed-off-by: Toan Le <toanle@apm.com>
Tested-by: Fushen Chen <fchen@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodrivers: net: xgene: Add workaround for errata 10GE_1
Quan Nguyen [Wed, 15 Mar 2017 20:27:19 +0000 (13:27 -0700)]
drivers: net: xgene: Add workaround for errata 10GE_1

This patch implements workaround for errata 10GE_1:
10Gb Ethernet port FIFO threshold default values are incorrect.

Signed-off-by: Quan Nguyen <qnguyen@apm.com>
Signed-off-by: Toan Le <toanle@apm.com>
Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Tested-by: Fushen Chen <fchen@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodrivers: net: xgene: Fix Rx checksum validation logic
Iyappan Subramanian [Wed, 15 Mar 2017 20:27:18 +0000 (13:27 -0700)]
drivers: net: xgene: Fix Rx checksum validation logic

This patch fixes Rx checksum validation logic and
adds NETIF_F_RXCSUM flag.

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodrivers: net: xgene: Fix wrong logical operation
Quan Nguyen [Wed, 15 Mar 2017 20:27:17 +0000 (13:27 -0700)]
drivers: net: xgene: Fix wrong logical operation

This patch fixes the wrong logical OR operation by changing it to
bit-wise OR operation.

Fixes: 3bb502f83080 ("drivers: net: xgene: fix statistics counters race condition")
Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: Quan Nguyen <qnguyen@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodrivers: net: xgene: Fix hardware checksum setting
Quan Nguyen [Wed, 15 Mar 2017 20:27:16 +0000 (13:27 -0700)]
drivers: net: xgene: Fix hardware checksum setting

This patch fixes the hardware checksum settings by properly program
the classifier. Otherwise, packet may be received with checksum error
on X-Gene1 SoC.

Signed-off-by: Quan Nguyen <qnguyen@apm.com>
Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agodrivers: net: phy: xgene: Fix mdio write
Quan Nguyen [Wed, 15 Mar 2017 20:27:15 +0000 (13:27 -0700)]
drivers: net: phy: xgene: Fix mdio write

This patches fixes a typo in the argument to xgene_enet_wr_mdio_csr().

Signed-off-by: Quan Nguyen <qnguyen@apm.com>
Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'dsa-check-out-of-range-ageing-time'
David S. Miller [Wed, 15 Mar 2017 22:34:14 +0000 (15:34 -0700)]
Merge branch 'dsa-check-out-of-range-ageing-time'

Vivien Didelot says:

====================
net: dsa: check out-of-range ageing time

The ageing time limits supported by DSA drivers vary depending on the
switch model. If a driver returns -ERANGE for out-of-range values, the
switchdev commit phase will fail with the following stacktrace:

    # brctl setageing br0 4
    [ 8530.082179] WARNING: CPU: 0 PID: 910 at net/switchdev/switchdev.c:291 switchdev_port_attr_set_now+0xbc/0xc0
    [ 8530.090679] br0: Commit of attribute (id=5) failed.
    [ 8530.094256] Modules linked in:
    [ 8530.096032] CPU: 0 PID: 910 Comm: kworker/0:4 Tainted: G        W       4.10.0 #361
    [ 8530.102412] Hardware name: Freescale Vybrid VF5xx/VF6xx (Device Tree)
    [ 8530.107571] Workqueue: events switchdev_deferred_process_work
    [ 8530.112039] Backtrace:
    [ 8530.113224] [<8010ca34>] (dump_backtrace) from [<8010cd3c>] (show_stack+0x20/0x24)
    [ 8530.119521]  r6:00000000 r5:80834da0 r4:80ca7e48 r3:8120ca3c
    [ 8530.123908] [<8010cd1c>] (show_stack) from [<8037ad40>] (dump_stack+0x24/0x28)
    [ 8530.129873] [<8037ad1c>] (dump_stack) from [<80118de4>] (__warn+0xf4/0x10c)
    [ 8530.135545] [<80118cf0>] (__warn) from [<80118e44>] (warn_slowpath_fmt+0x48/0x50)
    [ 8530.141760]  r9:00000000 r8:81252bec r7:80f19d90 r6:9dc3c000 r5:80ca7e7c r4:80834de8
    [ 8530.148235] [<80118e00>] (warn_slowpath_fmt) from [<80670b20>] (switchdev_port_attr_set_now+0xbc/0xc0)
    [ 8530.156240]  r3:9dc3c000 r2:80834de8
    [ 8530.158539]  r4:ffffffde
    [ 8530.159788] [<80670a64>] (switchdev_port_attr_set_now) from [<80670b44>] (switchdev_port_attr_set_deferred+0x20/0x6c)
    [ 8530.169118]  r7:806705a8 r6:9dc3c000 r5:80f19d90 r4:80f19d80
    [ 8530.173500] [<80670b24>] (switchdev_port_attr_set_deferred) from [<80670580>] (switchdev_deferred_process+0x50/0xe8)
    [ 8530.182742]  r6:80ca6000 r5:81252bec r4:80f19d80 r3:80670b24
    [ 8530.187115] [<80670530>] (switchdev_deferred_process) from [<80670930>] (switchdev_deferred_process_work+0x1c/0x24)
    [ 8530.196277]  r8:00000000 r7:9ffdc100 r6:8120ad6c r5:9ddefc00 r4:81252bf4 r3:9de343c0
    [ 8530.202756] [<80670914>] (switchdev_deferred_process_work) from [<8012f770>] (process_one_work+0x120/0x3b0)
    [ 8530.211231] [<8012f650>] (process_one_work) from [<8012fa70>] (worker_thread+0x70/0x534)
    [ 8530.218046]  r10:9ddefc00 r9:8120ad6c r8:80ca6038 r7:8120ad80 r6:81211f80 r5:9ddefc18
    [ 8530.224579]  r4:8120ad6c
    [ 8530.225830] [<8012fa00>] (worker_thread) from [<80135640>] (kthread+0x114/0x144)
    [ 8530.231955]  r10:9f4e9e94 r9:9de1fe58 r8:8012fa00 r7:9ddefc00 r6:9de1fdc0 r5:00000000
    [ 8530.238497]  r4:9de1fe40
    [ 8530.239750] [<8013552c>] (kthread) from [<80108cd8>] (ret_from_fork+0x14/0x3c)
    [ 8530.245679]  r10:00000000 r9:00000000 r8:00000000 r7:00000000 r6:00000000 r5:8013552c
    [ 8530.252234]  r4:9de1fdc0 r3:80ca6000
    [ 8530.254512] ---[ end trace 87475cc71b80ef73 ]---
    [ 8530.257852] br0: failed (err=-34) to set attribute (id=5)

This patchset fixes this by adding ageing_time_min and ageing_time_max
fields to the dsa_switch structure, which can optionally be set by a DSA
driver.

If provided, the DSA core will check for out-of-range values in the
SWITCHDEV_ATTR_ID_BRIDGE_AGEING_TIME prepare phase and return -ERANGE
accordingly.

Finally set these limits in the mv88e6xxx driver.
====================

Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: mv88e6xxx: specify ageing time limits
Vivien Didelot [Wed, 15 Mar 2017 19:53:50 +0000 (15:53 -0400)]
net: dsa: mv88e6xxx: specify ageing time limits

Now that DSA has ageing time limits, specify them when registering a
switch so that out-of-range values are handled correctly by the core.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reported-by: Jason Cobham <jcobham@questertangent.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: check out-of-range ageing time value
Vivien Didelot [Wed, 15 Mar 2017 19:53:49 +0000 (15:53 -0400)]
net: dsa: check out-of-range ageing time value

If a DSA switch driver cannot program an ageing time value due to it
being out-of-range, switchdev will raise a stack trace before failing.

To fix this, add ageing_time_min and ageing_time_max members to the
dsa_switch in order for the switch drivers to optionally specify their
supported ageing time limits.

The DSA core will now check for provided ageing time limits and return
-ERANGE from the switchdev prepare phase if the value is out-of-range.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agonet: dsa: dsa_fastest_ageing_time return unsigned
Vivien Didelot [Wed, 15 Mar 2017 19:53:48 +0000 (15:53 -0400)]
net: dsa: dsa_fastest_ageing_time return unsigned

The ageing time is defined as unsigned int, so make
dsa_fastest_ageing_time return an unsigned int instead of int.

Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agoMerge branch 'mqprio-offload-more-info'
David S. Miller [Wed, 15 Mar 2017 22:20:28 +0000 (15:20 -0700)]
Merge branch 'mqprio-offload-more-info'

Alexander Duyck says:

====================
Add support for passing more information in mqprio offload

This patch series lays the groundwork for future work to allow us to make
full use of the mqprio options when offloading them to hardware.

Currently when we specify the hardware offload for mqprio the queue
configuration is completely ignored and the hardware is only notified of
the total number of traffic classes.  The problem is this leads to multiple
issues, one specific issue being you can pass the queue configuration you
want and it is totally ignored by the hardware.

What I am planning to do is add support for "hw" values in the
configuration greater than 1.  So for example we might have one mode of
mqprio offload that uses 1 and only offloads the TC counts like we
currently do.  Then we might look at adding an option 2 which would factor
in the TCs and the queue count information. This way we can select between
the type of offload we actually want and existing drivers that don't
support this can just fall back to their legacy configuration.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
7 years agomqprio: Modify mqprio to pass user parameters via ndo_setup_tc.
Amritha Nambiar [Wed, 15 Mar 2017 17:39:25 +0000 (10:39 -0700)]
mqprio: Modify mqprio to pass user parameters via ndo_setup_tc.

The configurable priority to traffic class mapping and the user specified
queue ranges are used to configure the traffic class, overriding the
hardware defaults when the 'hw' option is set to 0. However, when the 'hw'
option is non-zero, the hardware QOS defaults are used.

This patch makes it so that we can pass the data the user provided to
ndo_setup_tc. This allows us to pull in the queue configuration if the
user requested it as well as any additional hardware offload type
requested by using a value other than 1 for the hw value.

Finally it also provides a means for the device driver to return the level
supported for the offload type via the qopt->hw value. Previously we were
just always assuming the value to be 1, in the future values beyond just 1
may be supported.

Signed-off-by: Amritha Nambiar <amritha.nambiar@intel.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>