openwrt/staging/blogic.git
5 years agoselftests/bpf: add bpf-gcc support
Ilya Leoshkevich [Thu, 12 Sep 2019 16:05:43 +0000 (18:05 +0200)]
selftests/bpf: add bpf-gcc support

Now that binutils and gcc support for BPF is upstream, make use of it in
BPF selftests using alu32-like approach. Share as much as possible of
CFLAGS calculation with clang.

Fixes only obvious issues, leaving more complex ones for later:
- Use gcc-provided bpf-helpers.h instead of manually defining the
  helpers, change bpf_helpers.h include guard to avoid conflict.
- Include <linux/stddef.h> for __always_inline.
- Add $(OUTPUT)/../usr/include to include path in order to use local
  kernel headers instead of system kernel headers when building with O=.

In order to activate the bpf-gcc support, one needs to configure
binutils and gcc with --target=bpf and make them available in $PATH. In
particular, gcc must be installed as `bpf-gcc`, which is the default.

Right now with binutils 25a2915e8dba and gcc r275589 only a handful of
tests work:

# ./test_progs_bpf_gcc
# Summary: 7/39 PASSED, 1 SKIPPED, 98 FAILED

The reason for those failures are as follows:

- Build errors:
  - `error: too many function arguments for eBPF` for __always_inline
    functions read_str_var and read_map_var - must be inlining issue,
    and for process_l3_headers_v6, which relies on optimizing away
    function arguments.
  - `error: indirect call in function, which are not supported by eBPF`
    where there are no obvious indirect calls in the source calls, e.g.
    in __encap_ipip_none.
  - `error: field 'lock' has incomplete type` for fields of `struct
    bpf_spin_lock` type - bpf_spin_lock is re#defined by bpf-helpers.h,
    so its usage is sensitive to order of #includes.
  - `error: eBPF stack limit exceeded` in sysctl_tcp_mem.
- Load errors:
  - Missing object files due to above build errors.
  - `libbpf: failed to create map (name: 'test_ver.bss')`.
  - `libbpf: object file doesn't contain bpf program`.
  - `libbpf: Program '.text' contains unrecognized relo data pointing to
    section 0`.
  - `libbpf: BTF is required, but is missing or corrupted` - no BTF
    support in gcc yet.

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Cc: Jose E. Marchesi <jose.marchesi@oracle.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agokcm: use BPF_PROG_RUN
Sami Tolvanen [Thu, 5 Sep 2019 21:15:28 +0000 (14:15 -0700)]
kcm: use BPF_PROG_RUN

Instead of invoking struct bpf_prog::bpf_func directly, use the
BPF_PROG_RUN macro.

Signed-off-by: Sami Tolvanen <samitolvanen@google.com>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agoMerge branch 'move-sockopt-tests'
Alexei Starovoitov [Fri, 6 Sep 2019 16:59:06 +0000 (09:59 -0700)]
Merge branch 'move-sockopt-tests'

Stanislav Fomichev says:

====================
Now that test_progs is shaping into more generic test framework,
let's convert sockopt tests to it. This requires adding
a helper to create and join a cgroup first (test__join_cgroup).
Since we already hijack stdout/stderr that shouldn't be
a problem (cgroup helpers log to stderr).

The rest of the patches just move sockopt tests files under prog_tests/
and do the required small adjustments.
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agoselftests/bpf: test_progs: convert test_tcp_rtt
Stanislav Fomichev [Wed, 4 Sep 2019 16:25:09 +0000 (09:25 -0700)]
selftests/bpf: test_progs: convert test_tcp_rtt

Move the files, adjust includes, remove entry from Makefile & .gitignore

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agoselftests/bpf: test_progs: convert test_sockopt_inherit
Stanislav Fomichev [Wed, 4 Sep 2019 16:25:08 +0000 (09:25 -0700)]
selftests/bpf: test_progs: convert test_sockopt_inherit

Move the files, adjust includes, remove entry from Makefile & .gitignore

I also added pthread_cond_wait for the server thread startup. We don't
want to connect to the server that's not yet up (for some reason
this existing race is now more prominent with test_progs).

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agoselftests/bpf: test_progs: convert test_sockopt_multi
Stanislav Fomichev [Wed, 4 Sep 2019 16:25:07 +0000 (09:25 -0700)]
selftests/bpf: test_progs: convert test_sockopt_multi

Move the files, adjust includes, remove entry from Makefile & .gitignore

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agoselftests/bpf: test_progs: convert test_sockopt_sk
Stanislav Fomichev [Wed, 4 Sep 2019 16:25:06 +0000 (09:25 -0700)]
selftests/bpf: test_progs: convert test_sockopt_sk

Move the files, adjust includes, remove entry from Makefile & .gitignore

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agoselftests/bpf: test_progs: convert test_sockopt
Stanislav Fomichev [Wed, 4 Sep 2019 16:25:05 +0000 (09:25 -0700)]
selftests/bpf: test_progs: convert test_sockopt

Move the files, adjust includes, remove entry from Makefile & .gitignore

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agoselftests/bpf: test_progs: add test__join_cgroup helper
Stanislav Fomichev [Wed, 4 Sep 2019 16:25:04 +0000 (09:25 -0700)]
selftests/bpf: test_progs: add test__join_cgroup helper

test__join_cgroup() combines the following operations that usually
go hand in hand and returns cgroup fd:

  * setup cgroup environment (make sure cgroupfs is mounted)
  * mkdir cgroup
  * join cgroup

It also marks a test as a "cgroup cleanup needed" and removes cgroup
state after the test is done.

Signed-off-by: Stanislav Fomichev <sdf@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agokbuild: replace BASH-specific ${@:2} with shift and ${@}
Andrii Nakryiko [Thu, 5 Sep 2019 17:59:38 +0000 (10:59 -0700)]
kbuild: replace BASH-specific ${@:2} with shift and ${@}

${@:2} is BASH-specific extension, which makes link-vmlinux.sh rely on
BASH. Use shift and ${@} instead to fix this issue.

Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Fixes: 341dfcf8d78e ("btf: expose BTF info through sysfs")
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Masahiro Yamada <yamada.masahiro@socionext.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Yonghong Song <yhs@fb.com>
Reviewed-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next
David S. Miller [Fri, 6 Sep 2019 14:49:17 +0000 (16:49 +0200)]
Merge git://git./linux/kernel/git/bpf/bpf-next

Daniel Borkmann says:

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

The main changes are:

1) Add the ability to use unaligned chunks in the AF_XDP umem. By
   relaxing where the chunks can be placed, it allows to use an
   arbitrary buffer size and place whenever there is a free
   address in the umem. Helps more seamless DPDK AF_XDP driver
   integration. Support for i40e, ixgbe and mlx5e, from Kevin and
   Maxim.

2) Addition of a wakeup flag for AF_XDP tx and fill rings so the
   application can wake up the kernel for rx/tx processing which
   avoids busy-spinning of the latter, useful when app and driver
   is located on the same core. Support for i40e, ixgbe and mlx5e,
   from Magnus and Maxim.

3) bpftool fixes for printf()-like functions so compiler can actually
   enforce checks, bpftool build system improvements for custom output
   directories, and addition of 'bpftool map freeze' command, from Quentin.

4) Support attaching/detaching XDP programs from 'bpftool net' command,
   from Daniel.

5) Automatic xskmap cleanup when AF_XDP socket is released, and several
   barrier/{read,write}_once fixes in AF_XDP code, from Björn.

6) Relicense of bpf_helpers.h/bpf_endian.h for future libbpf
   inclusion as well as libbpf versioning improvements, from Andrii.

7) Several new BPF kselftests for verifier precision tracking, from Alexei.

8) Several BPF kselftest fixes wrt endianess to run on s390x, from Ilya.

9) And more BPF kselftest improvements all over the place, from Stanislav.

10) Add simple BPF map op cache for nfp driver to batch dumps, from Jakub.

11) AF_XDP socket umem mapping improvements for 32bit archs, from Ivan.

12) Add BPF-to-BPF call and BTF line info support for s390x JIT, from Yauheni.

13) Small optimization in arm64 JIT to spare 1 insns for BPF_MOD, from Jerin.

14) Fix an error check in bpf_tcp_gen_syncookie() helper, from Petar.

15) Various minor fixes and cleanups, from Nathan, Masahiro, Masanari,
    Peter, Wei, Yue.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agolan743x: remove redundant assignment to variable rx_process_result
Colin Ian King [Thu, 5 Sep 2019 14:01:35 +0000 (15:01 +0100)]
lan743x: remove redundant assignment to variable rx_process_result

The variable rx_process_result is being initialized with a value that
is never read and is being re-assigned immediately afterwards. The
assignment is redundant, so replace it with the return from function
lan743x_rx_process_packet.

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'ravb-remove-use-of-undocumented-registers'
David S. Miller [Fri, 6 Sep 2019 14:46:10 +0000 (16:46 +0200)]
Merge branch 'ravb-remove-use-of-undocumented-registers'

Simon Horman says:

====================
ravb: remove use of undocumented registers

this short series cleans up the RAVB driver a little.

The first patch corrects the spelling of the FBP field of SFO register.
This register field is unused and should have no run-time effect.

The remaining patches remove the use of undocumented registers
after some consultation with the internal Renesas BSP team.

Changes in v2:
* Corrected mangled state of first patch
* Patches 2/4 and 3/4 split out of a large patch
* Accumulated acks
* Tweaked changelog
* Claimed authorship of all patches

v1 of this series was tested on the following platforms.
No behaviour change is expected in v2.
* E3 Ebisu
* H3 Salvator-XS (ES2.0)
* M3-W Salvator-XS
* M3-N Salvator-XS
* RZ/G1C iW-RainboW-G23S
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoravb: TROCR register is only present on R-Car Gen3
Simon Horman [Thu, 5 Sep 2019 15:10:59 +0000 (17:10 +0200)]
ravb: TROCR register is only present on R-Car Gen3

Only use the TROCR register on R-Car Gen3 as it is not present on other
SoCs.

Offsets used for the undocumented registers are considered reserved and
should not be written to. After some internal investigation with Renesas it
remains unclear why this driver accesses these fields on R-Car Gen2 but
regardless of what the historical reasons are the current code is
considered incorrect.

Signed-off-by: Simon Horman <horms+renesas@verge.net.au>
Reviewed-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Acked-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoravb: remove undocumented endianness selection
Simon Horman [Thu, 5 Sep 2019 15:10:58 +0000 (17:10 +0200)]
ravb: remove undocumented endianness selection

This patch removes the use of the undocumented BOC bit of the CCC register.

Current documentation for EtherAVB (ravb) describes the offset of what the
driver uses as the BOC bit as reserved and that only a value of 0 should be
written. After some internal investigation with Renesas it remains unclear
why this driver accesses these fields but regardless of what the historical
reasons are the current code is considered incorrect.

Based on work by Kazuya Mizuguchi <kazuya.mizuguchi.ks@renesas.com>

Signed-off-by: Simon Horman <horms+renesas@verge.net.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoravb: remove undocumented counter processing
Simon Horman [Thu, 5 Sep 2019 15:10:57 +0000 (17:10 +0200)]
ravb: remove undocumented counter processing

This patch removes the use of the undocumented counter registers
CDCR, LCCR, CERCR, CEECR.

Offsets used for undocumented registers are considered reserved and
should not be written to. After some internal investigation with Renesas
it remains unclear why this driver accesses these fields but regardless of
what the historical reasons are the current code is considered incorrect.

Based on work by Kazuya Mizuguchi <kazuya.mizuguchi.ks@renesas.com>

Signed-off-by: Simon Horman <horms+renesas@verge.net.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoravb: correct typo in FBP field of SFO register
Simon Horman [Thu, 5 Sep 2019 15:10:56 +0000 (17:10 +0200)]
ravb: correct typo in FBP field of SFO register

The field name is FBP rather than FPB.

This field is unused and could equally be removed from the driver entirely.
But there seems no harm in leaving as documentation of the presence of the
field.

Based on work by Kazuya Mizuguchi <kazuya.mizuguchi.ks@renesas.com>

Signed-off-by: Simon Horman <horms+renesas@verge.net.au>
Reviewed-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Acked-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-hns3-add-some-bugfixes-and-cleanups'
David S. Miller [Fri, 6 Sep 2019 13:20:35 +0000 (15:20 +0200)]
Merge branch 'net-hns3-add-some-bugfixes-and-cleanups'

Huazhong Tan says:

====================
net: hns3: add some bugfixes and cleanups

This patch-set includes bugfixes and cleanups for the HNS3
ethernet controller driver.

[patch 01/07] fixes an error when setting VLAN offload.

[patch 02/07] fixes an double free issue when setting ringparam.

[patch 03/07] fixes a mis-assignment of hdev->reset_level.

[patch 04/07] adds a checking for client's validity.

[patch 05/07] simplifies bool variable's assignment.

[patch 06/07] disables loopback when initializing.

[patch 07/07] makes internal function to static.

Change log:
V1->V2: fixes comment from Sergei Shtylyov.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: make hclge_dbg_get_m7_stats_info static
Guojia Liao [Thu, 5 Sep 2019 13:31:42 +0000 (21:31 +0800)]
net: hns3: make hclge_dbg_get_m7_stats_info static

hclge_dbg_get_m7_info is used only in the hclge_debugfs.c,
so it should be declared with static.

Signed-off-by: Guojia Liao <liaoguojia@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: disable loopback setting in hclge_mac_init
Yufeng Mo [Thu, 5 Sep 2019 13:31:41 +0000 (21:31 +0800)]
net: hns3: disable loopback setting in hclge_mac_init

If the selftest and reset are performed at the same time, the loopback
setting may be still in the enable state after the reset. As a result,
packets cannot be sent out.

This patch fixes this issue by disabling loopback in hclge_mac_init.

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: remove explicit conversion to bool
Guojia Liao [Thu, 5 Sep 2019 13:31:40 +0000 (21:31 +0800)]
net: hns3: remove explicit conversion to bool

Relational and logical operators evaluate to bool,
explicit conversion is overly verbose and unnecessary.

Signed-off-by: Guojia Liao <liaoguojia@huawei.com>
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: add client node validity judgment
Peng Li [Thu, 5 Sep 2019 13:31:39 +0000 (21:31 +0800)]
net: hns3: add client node validity judgment

HNS3 driver can only unregister client which included in hnae3_client_list.
This patch adds the client node validity judgment.

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 mis-assignment to hdev->reset_level in hclge_reset
Huazhong Tan [Thu, 5 Sep 2019 13:31:38 +0000 (21:31 +0800)]
net: hns3: fix mis-assignment to hdev->reset_level in hclge_reset

Since hclge_get_reset_level may return HNAE3_NONE_RESET,
so hdev->reset_level can not be assigned with the return
value in the hclge_reset(), otherwise, it will cause
the use of hdev->reset_level in hclge_reset_event get
into error.

Fixes: 012fcb52f67c ("net: hns3: activate reset timer when calling reset_event")
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: fix double free bug when setting ringparam
Huazhong Tan [Thu, 5 Sep 2019 13:31:37 +0000 (21:31 +0800)]
net: hns3: fix double free bug when setting ringparam

The system will panic when change the ringparam in HNS3 drivers:

[ 1459.627727] hns3 0000:bd:00.0 eth6: Changing Tx/Rx ring ds from 1024/1024 to 24/24
[ 1459.635766] hns3 0000:bd:00.0 eth6: link down
[ 1459.640788] BUG: Bad page state in process ethtool  pfn:203f75c18
[ 1459.646940] page:ffff7ee4ffd70600 refcount:0 mapcount:0 mapping:ffff993fff40f400 index:0x0 compound_mapcount: 0
[ 1459.656987] flags: 0x9fffe00000010200(slab|head)
[ 1459.661591] raw: 9fffe00000010200 dead000000000100 dead000000000122 ffff993fff40f400
[ 1459.669302] raw: 0000000000000000 0000000080100010 00000000ffffffff 0000000000000000
[ 1459.677016] page dumped because: PAGE_FLAGS_CHECK_AT_FREE flag(s) set
[ 1459.683432] bad because of flags: 0x200(slab)
[ 1459.687775] Modules linked in: ib_ipoib ib_umad rpcrdma ib_iser libiscsi scsi_transport_iscsi hns_roce_hw_v2 crct10dif_ce hns3 ses hclge hnae3 hisi_hpre hisi_zip qm uacce ip_tables x_tables hisi_sas_v3_hw hisi_sas_main libsas scsi_transport_sas
[ 1459.709329] CPU: 14 PID: 17244 Comm: ethtool Tainted: G           O      5.3.0-rc4-00415-gc86f057 #1
[ 1459.718419] Hardware name: Huawei TaiShan 2280 V2/BC82AMDC, BIOS 2280-V2 CS V3.B040.01 07/26/2019
[ 1459.727248] Call trace:
[ 1459.729688]  dump_backtrace+0x0/0x150
[ 1459.733335]  show_stack+0x24/0x30
[ 1459.736639]  dump_stack+0xa0/0xc4
[ 1459.739943]  bad_page+0xf0/0x158
[ 1459.743157]  free_pages_check_bad+0x84/0xa0
[ 1459.747322]  __free_pages_ok+0x348/0x378
[ 1459.751228]  page_frag_free+0x80/0x88
[ 1459.754877]  skb_free_head+0x38/0x48
[ 1459.758436]  skb_release_data+0x134/0x160
[ 1459.762427]  skb_release_all+0x30/0x40
[ 1459.766158]  consume_skb+0x38/0x108
[ 1459.769633]  __dev_kfree_skb_any+0x58/0x68
[ 1459.773718]  hns3_fini_ring+0x48/0x58 [hns3]
[ 1459.777970]  hns3_set_ringparam+0x2a8/0x418 [hns3]
[ 1459.782741]  dev_ethtool+0x5f4/0x2080
[ 1459.786390]  dev_ioctl+0x190/0x3d8
[ 1459.789777]  sock_do_ioctl+0xf8/0x220
[ 1459.793423]  sock_ioctl+0x3bc/0x490
[ 1459.796896]  do_vfs_ioctl+0xc4/0x868
[ 1459.800454]  ksys_ioctl+0x8c/0xa0
[ 1459.803752]  __arm64_sys_ioctl+0x28/0x38
[ 1459.807658]  el0_svc_common.constprop.0+0xe0/0x1e0
[ 1459.812426]  el0_svc_handler+0x34/0x90
[ 1459.816158]  el0_svc+0x10/0x14
[ 1459.819220] Disabling lock debugging due to kernel taint
[ 1459.825182] ------------[ cut here ]------------

Since ndo_stop will reclaim the RX's skb allocated by the driver,
so the backed up ring parameter should not keep this info.

Fixes: a723fb8efe29 ("net: hns3: refine for set ring parameters")
Signed-off-by: Huazhong Tan <tanhuazhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns3: fix error VF index when setting VLAN offload
Jian Shen [Thu, 5 Sep 2019 13:31:36 +0000 (21:31 +0800)]
net: hns3: fix error VF index when setting VLAN offload

In original codes, the VF index used incorrectly in function
hclge_set_vlan_rx_offload_cfg() and hclge_set_vlan_rx_offload_cfg().
When VF id is greater than 8, for example 9, it will set the
same bit with VF id 1.

This patch fixes it by using  vport->vport_id % HCLGE_VF_NUM_PER_CMD /
HCLGE_VF_NUM_PER_BYTE as the array index, instead of vport->vport_id /
HCLGE_VF_NUM_PER_CMD.

Fixes: 052ece6dc19c ("net: hns3: add ethtool related offload command")
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 agostmmac: platform: adjust messages and move to dev level
Andy Shevchenko [Thu, 5 Sep 2019 13:00:53 +0000 (16:00 +0300)]
stmmac: platform: adjust messages and move to dev level

This patch amends the error and warning messages across the platform driver.
It includes the following changes:
 - append \n to the end of messages
 - change pr_* macros to dev_*

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: Do not check Link status when loopback is enabled
Jose Abreu [Thu, 5 Sep 2019 11:43:10 +0000 (13:43 +0200)]
net: phy: Do not check Link status when loopback is enabled

While running stmmac selftests I found that in my 1G setup some tests
were failling when running with PHY loopback enabled.

It looks like when loopback is enabled the PHY will report that Link is
down even though there is a valid connection.

As in loopback mode the data will not be sent anywhere we can bypass the
logic of checking if Link is valid thus saving unecessary reads.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet_sched: act_police: add 2 new attributes to support police 64bit rate and peakrate
David Dai [Wed, 4 Sep 2019 15:03:43 +0000 (10:03 -0500)]
net_sched: act_police: add 2 new attributes to support police 64bit rate and peakrate

For high speed adapter like Mellanox CX-5 card, it can reach upto
100 Gbits per second bandwidth. Currently htb already supports 64bit rate
in tc utility. However police action rate and peakrate are still limited
to 32bit value (upto 32 Gbits per second). Add 2 new attributes
TCA_POLICE_RATE64 and TCA_POLICE_RATE64 in kernel for 64bit support
so that tc utility can use them for 64bit rate and peakrate value to
break the 32bit limit, and still keep the backward binary compatibility.

Tested-by: David Dai <zdai@linux.vnet.ibm.com>
Signed-off-by: David Dai <zdai@linux.vnet.ibm.com>
Acked-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: openvswitch: Set OvS recirc_id from tc chain index
Paul Blakey [Wed, 4 Sep 2019 13:56:37 +0000 (16:56 +0300)]
net: openvswitch: Set OvS recirc_id from tc chain index

Offloaded OvS datapath rules are translated one to one to tc rules,
for example the following simplified OvS rule:

recirc_id(0),in_port(dev1),eth_type(0x0800),ct_state(-trk) actions:ct(),recirc(2)

Will be translated to the following tc rule:

$ tc filter add dev dev1 ingress \
    prio 1 chain 0 proto ip \
flower tcp ct_state -trk \
action ct pipe \
action goto chain 2

Received packets will first travel though tc, and if they aren't stolen
by it, like in the above rule, they will continue to OvS datapath.
Since we already did some actions (action ct in this case) which might
modify the packets, and updated action stats, we would like to continue
the proccessing with the correct recirc_id in OvS (here recirc_id(2))
where we left off.

To support this, introduce a new skb extension for tc, which
will be used for translating tc chain to ovs recirc_id to
handle these miss cases. Last tc chain index will be set
by tc goto chain action and read by OvS datapath.

Signed-off-by: Paul Blakey <paulb@mellanox.com>
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonfp: Drop unnecessary continue in nfp_net_pf_alloc_vnics
zhong jiang [Wed, 4 Sep 2019 03:46:23 +0000 (11:46 +0800)]
nfp: Drop unnecessary continue in nfp_net_pf_alloc_vnics

Continue is not needed at the bottom of a loop.

Signed-off-by: zhong jiang <zhongjiang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'bpf-af-xdp-barrier-fixes'
Daniel Borkmann [Thu, 5 Sep 2019 12:11:52 +0000 (14:11 +0200)]
Merge branch 'bpf-af-xdp-barrier-fixes'

Björn Töpel says:

====================
This is a four patch series of various barrier, {READ, WRITE}_ONCE
cleanups in the AF_XDP socket code. More details can be found in the
corresponding commit message. Previous revisions: v1 [4] and v2 [5].

For an AF_XDP socket, most control plane operations are done under the
control mutex (struct xdp_sock, mutex), but there are some places
where members of the struct is read outside the control mutex. The
dev, queue_id members are set in bind() and cleared at cleanup. The
umem, fq, cq, tx, rx, and state member are all assigned in various
places, e.g. bind() and setsockopt(). When the members are assigned,
they are protected by the control mutex, but since they are read
outside the mutex, a WRITE_ONCE is required to avoid store-tearing on
the read-side.

Prior the state variable was introduced by Ilya, the dev member was
used to determine whether the socket was bound or not. However, when
dev was read, proper SMP barriers and READ_ONCE were missing. In order
to address the missing barriers and READ_ONCE, we start using the
state variable as a point of synchronization. The state member
read/write is paired with proper SMP barriers, and from this follows
that the members described above does not need READ_ONCE statements if
used in conjunction with state check.

To summarize: The members struct xdp_sock members dev, queue_id, umem,
fq, cq, tx, rx, and state were read lock-less, with incorrect barriers
and missing {READ, WRITE}_ONCE. After this series umem, fq, cq, tx,
rx, and state are read lock-less. When these members are updated,
WRITE_ONCE is used. When read, READ_ONCE are only used when read
outside the control mutex (e.g. mmap) or, not synchronized with the
state member (XSK_BOUND plus smp_rmb())

[1] https://lore.kernel.org/bpf/beef16bb-a09b-40f1-7dd0-c323b4b89b17@iogearbox.net/
[2] https://lwn.net/Articles/793253/
[3] https://github.com/google/ktsan/wiki/READ_ONCE-and-WRITE_ONCE
[4] https://lore.kernel.org/bpf/20190822091306.20581-1-bjorn.topel@gmail.com/
[5] https://lore.kernel.org/bpf/20190826061053.15996-1-bjorn.topel@gmail.com/

v2->v3:
  Minor restructure of commits.
  Improve cover and commit messages. (Daniel)
v1->v2:
  Removed redundant dev check. (Jonathan)
====================

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoxsk: lock the control mutex in sock_diag interface
Björn Töpel [Wed, 4 Sep 2019 11:49:13 +0000 (13:49 +0200)]
xsk: lock the control mutex in sock_diag interface

When accessing the members of an XDP socket, the control mutex should
be held. This commit fixes that.

Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Fixes: a36b38aa2af6 ("xsk: add sock_diag interface for AF_XDP")
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoxsk: use state member for socket synchronization
Björn Töpel [Wed, 4 Sep 2019 11:49:12 +0000 (13:49 +0200)]
xsk: use state member for socket synchronization

Prior the state variable was introduced by Ilya, the dev member was
used to determine whether the socket was bound or not. However, when
dev was read, proper SMP barriers and READ_ONCE were missing. In order
to address the missing barriers and READ_ONCE, we start using the
state variable as a point of synchronization. The state member
read/write is paired with proper SMP barriers, and from this follows
that the members described above does not need READ_ONCE if used in
conjunction with state check.

In all syscalls and the xsk_rcv path we check if state is
XSK_BOUND. If that is the case we do a SMP read barrier, and this
implies that the dev, umem and all rings are correctly setup. Note
that no READ_ONCE are needed for these variable if used when state is
XSK_BOUND (plus the read barrier).

To summarize: The members struct xdp_sock members dev, queue_id, umem,
fq, cq, tx, rx, and state were read lock-less, with incorrect barriers
and missing {READ, WRITE}_ONCE. Now, umem, fq, cq, tx, rx, and state
are read lock-less. When these members are updated, WRITE_ONCE is
used. When read, READ_ONCE are only used when read outside the control
mutex (e.g. mmap) or, not synchronized with the state member
(XSK_BOUND plus smp_rmb())

Note that dev and queue_id do not need a WRITE_ONCE or READ_ONCE, due
to the introduce state synchronization (XSK_BOUND plus smp_rmb()).

Introducing the state check also fixes a race, found by syzcaller, in
xsk_poll() where umem could be accessed when stale.

Suggested-by: Hillf Danton <hdanton@sina.com>
Reported-by: syzbot+c82697e3043781e08802@syzkaller.appspotmail.com
Fixes: 77cd0d7b3f25 ("xsk: add support for need_wakeup flag in AF_XDP rings")
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoxsk: avoid store-tearing when assigning umem
Björn Töpel [Wed, 4 Sep 2019 11:49:11 +0000 (13:49 +0200)]
xsk: avoid store-tearing when assigning umem

The umem member of struct xdp_sock is read outside of the control
mutex, in the mmap implementation, and needs a WRITE_ONCE to avoid
potential store-tearing.

Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Fixes: 423f38329d26 ("xsk: add umem fill queue support and mmap")
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoxsk: avoid store-tearing when assigning queues
Björn Töpel [Wed, 4 Sep 2019 11:49:10 +0000 (13:49 +0200)]
xsk: avoid store-tearing when assigning queues

Use WRITE_ONCE when doing the store of tx, rx, fq, and cq, to avoid
potential store-tearing. These members are read outside of the control
mutex in the mmap implementation.

Acked-by: Jonathan Lemon <jonathan.lemon@gmail.com>
Fixes: 37b076933a8e ("xsk: add missing write- and data-dependency barrier")
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoselftests/bpf: precision tracking tests
Alexei Starovoitov [Tue, 3 Sep 2019 22:51:33 +0000 (15:51 -0700)]
selftests/bpf: precision tracking tests

Add two tests to check that stack slot marking during backtracking
doesn't trigger 'spi > allocated_stack' warning.
One test is using BPF_ST insn. Another is using BPF_STX.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoixgbe: fix xdp handle calculations
Kevin Laatz [Thu, 5 Sep 2019 01:12:17 +0000 (01:12 +0000)]
ixgbe: fix xdp handle calculations

Currently, we don't add headroom to the handle in ixgbe_zca_free,
ixgbe_alloc_buffer_slow_zc and ixgbe_alloc_buffer_zc. The addition of the
headroom to the handle was removed in
commit d8c3061e5edd ("ixgbe: modify driver for handling offsets"), which
will break things when headroom isvnon-zero. This patch fixes this and uses
xsk_umem_adjust_offset to add it appropritely based on the mode being run.

Fixes: d8c3061e5edd ("ixgbe: modify driver for handling offsets")
Reported-by: Bjorn Topel <bjorn.topel@intel.com>
Signed-off-by: Kevin Laatz <kevin.laatz@intel.com>
Acked-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agoi40e: fix xdp handle calculations
Kevin Laatz [Thu, 5 Sep 2019 01:11:44 +0000 (01:11 +0000)]
i40e: fix xdp handle calculations

Currently, we don't add headroom to the handle in i40e_zca_free,
i40e_alloc_buffer_slow_zc and i40e_alloc_buffer_zc. The addition of the
headroom to the handle was removed in
commit 2f86c806a8a8 ("i40e: modify driver for handling offsets"), which
will break things when headroom is non-zero. This patch fixes this and uses
xsk_umem_adjust_offset to add it appropritely based on the mode being run.

Fixes: 2f86c806a8a8 ("i40e: modify driver for handling offsets")
Reported-by: Bjorn Topel <bjorn.topel@intel.com>
Signed-off-by: Kevin Laatz <kevin.laatz@intel.com>
Acked-by: Björn Töpel <bjorn.topel@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
5 years agor8152: adjust the settings of ups flags
Hayes Wang [Thu, 5 Sep 2019 02:46:20 +0000 (10:46 +0800)]
r8152: adjust the settings of ups flags

The UPS feature only works for runtime suspend, so UPS flags only
need to be set before enabling runtime suspend. Therefore, I create
a struct to record relative information, and use it before runtime
suspend.

All chips could record such information, even though not all of
them support the feature of UPS. Then, some functions could be
combined.

Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: hns: Move static keyword to the front of declaration
Krzysztof Wilczynski [Wed, 4 Sep 2019 14:21:16 +0000 (16:21 +0200)]
net: hns: Move static keyword to the front of declaration

Move the static keyword to the front of declaration of g_dsaf_mode_match,
and resolve the following compiler warning that can be seen when building
with warnings enabled (W=1):

drivers/net/ethernet/hisilicon/hns/hns_dsaf_main.c:27:1: warning:
  ‘static’ is not at beginning of declaration [-Wold-style-declaration]

Signed-off-by: Krzysztof Wilczynski <kw@linux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: qed: Move static keyword to the front of declaration
Krzysztof Wilczynski [Wed, 4 Sep 2019 14:17:30 +0000 (16:17 +0200)]
net: qed: Move static keyword to the front of declaration

Move the static keyword to the front of declaration of iwarp_state_names,
and resolve the following compiler warning that can be seen when building
with warnings enabled (W=1):

drivers/net/ethernet/qlogic/qed/qed_iwarp.c:385:1: warning:
  ‘static’ is not at beginning of declaration [-Wold-style-declaration]

Also, resolve checkpatch.pl script warning:

WARNING: static const char * array should probably be
  static const char * const

Signed-off-by: Krzysztof Wilczynski <kw@linux.com>
Acked-by: Michal Kalderon <michal.kalderon@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agopppoatm: use %*ph to print small buffer
Andy Shevchenko [Wed, 4 Sep 2019 17:44:59 +0000 (20:44 +0300)]
pppoatm: use %*ph to print small buffer

Use %*ph format to print small buffer as hex string.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'Fix-GMII2RGMII-private-field'
David S. Miller [Thu, 5 Sep 2019 10:32:06 +0000 (12:32 +0200)]
Merge branch 'Fix-GMII2RGMII-private-field'

Harini Katakam says:

====================
Fix GMII2RGMII private field

Fix the usage of external phy's priv field by gmii2rgmii driver.

Based on net-next.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: phy: gmii2rgmii: Dont use priv field in phy device
Harini Katakam [Wed, 4 Sep 2019 14:00:21 +0000 (19:30 +0530)]
net: phy: gmii2rgmii: Dont use priv field in phy device

Use set/get drv data in phydev's mdio device instead. Phy device priv
field maybe used by the external phy driver and should not be
overwritten.

Signed-off-by: Harini Katakam <harini.katakam@xilinx.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoinclude: mdio: Add driver data helpers
Harini Katakam [Wed, 4 Sep 2019 14:00:20 +0000 (19:30 +0530)]
include: mdio: Add driver data helpers

Add set/get drv_data helpers for mdio device.

Signed-off-by: Harini Katakam <harini.katakam@xilinx.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'gianfar-some-assorted-cleanup'
David S. Miller [Thu, 5 Sep 2019 10:28:15 +0000 (12:28 +0200)]
Merge branch 'gianfar-some-assorted-cleanup'

Arseny Solokha says:

====================
gianfar: some assorted cleanup

This is a cleanup series for the gianfar Ethernet driver, following up a
discussion in [1]. It is intended to precede a conversion of gianfar from
PHYLIB to PHYLINK API, which will be submitted later in its version 2.
However, it won't make a conversion cleaner, except for the last patch in
this series. Obviously this series is not intended for -stable.

The first patch looks super controversial to me, as it moves lots of code
around for the sole purpose of getting rid of static forward declarations
in two translation units. On the other hand, this change is purely
mechanical and cannot do any harm other than cluttering git blame output.
I can prepare an alternative patch for only swapping adjacent functions
around, if necessary.

The second patch is a trivial follow-up to the first one, making functions
that are only called from the same translation unit static.

The third patch removes some now unused macro and structure definitions
from gianfar.h, slipped away from various cleanups in the past.

The fourth patch, also suggested in [1], makes the driver consistently use
PHY connection type value obtained from a Device Tree node, instead of
ignoring it and using the one auto-detected by MAC, when connecting to PHY.
Obviously a value has to be specified correctly in DT source, or omitted
altogether, in which case the driver will fall back to auto-detection. When
querying a DT node, the driver will also take both applicable properties
into account by making a proper API call instead of open-coding the lookup
half-way correctly.

[1] https://lore.kernel.org/netdev/CA+h21hruqt6nGG5ksDSwrGH_w5GtGF4fjAMCWJne7QJrjusERQ@mail.gmail.com/
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agogianfar: use DT more consistently when selecting PHY connection type
Arseny Solokha [Wed, 4 Sep 2019 13:52:22 +0000 (20:52 +0700)]
gianfar: use DT more consistently when selecting PHY connection type

Historically, gianfar only used phy-connection-type DT property when
connected to PHY in the rgmii-id mode. It ignored the property otherwise,
relying on the connection type auto-detection carried out by MAC and
providing that reconstructed mode to of_phy_connect(). It also did not
consider alternative phy-mode property at all.

Make the driver properly query DT node for PHY connection type first and
use an obtained value if it was specified there. Otherwise, if a particular
DT relies on connection type auto-detection, fall back to reconstructing
the value from MAC registers, as before.

Signed-off-by: Arseny Solokha <asolokha@kb.kras.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agogianfar: cleanup gianfar.h
Arseny Solokha [Wed, 4 Sep 2019 13:52:21 +0000 (20:52 +0700)]
gianfar: cleanup gianfar.h

Remove now unused macro and structure definitions from gianfar.h that have
accumulated there over time.

Signed-off-by: Arseny Solokha <asolokha@kb.kras.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agogianfar: make five functions static
Arseny Solokha [Wed, 4 Sep 2019 13:52:20 +0000 (20:52 +0700)]
gianfar: make five functions static

Make functions that do not have callers outside the translation unit they
are defined in static.

Signed-off-by: Arseny Solokha <asolokha@kb.kras.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agogianfar: remove forward declarations
Arseny Solokha [Wed, 4 Sep 2019 13:52:19 +0000 (20:52 +0700)]
gianfar: remove forward declarations

Remove forward declarations of various static functions located in two
driver implementation files and rearrange the corresponding definitions
accordingly.

This patch only introduces mechanical changes, namely it removes forward
declarations and moves function definitions around; it does not change any
functionality.

Signed-off-by: Arseny Solokha <asolokha@kb.kras.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'stmmac-next'
David S. Miller [Thu, 5 Sep 2019 10:19:55 +0000 (12:19 +0200)]
Merge branch 'stmmac-next'

Jose Abreu says:

====================
net: stmmac: Improvements for -next

Couple of improvements for -next tree. More info in commit logs.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: selftests: Add Jumbo Frame tests
Jose Abreu [Wed, 4 Sep 2019 13:17:05 +0000 (15:17 +0200)]
net: stmmac: selftests: Add Jumbo Frame tests

Add a test to validate the Jumbo Frame support in stmmac in single
channel and multichannel mode.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: xgmac: Enable RX Jumbo frame support
Jose Abreu [Wed, 4 Sep 2019 13:17:04 +0000 (15:17 +0200)]
net: stmmac: xgmac: Enable RX Jumbo frame support

We are already doing it by default in the TX path so we can also enable
Jumbo Frame support in the RX path independently of MTU value.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: Correctly assing MAX MTU in XGMAC cores case
Jose Abreu [Wed, 4 Sep 2019 13:17:03 +0000 (15:17 +0200)]
net: stmmac: Correctly assing MAX MTU in XGMAC cores case

Maximum MTU for XGMAC cores is 16k thus the check for presence of XGMAC
shall be done first in order to assign correct value.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: xgmac: Correct RAVSEL field interpretation
Jose Abreu [Wed, 4 Sep 2019 13:17:02 +0000 (15:17 +0200)]
net: stmmac: xgmac: Correct RAVSEL field interpretation

RAVSEL means that only RX side is available for AVB features. As we use
both TX and RX features we need to check if RAVSEL is selected and
disable AVB if only RX side is available.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: ethtool: Let user configure TX coalesce without RIWT
Jose Abreu [Wed, 4 Sep 2019 13:17:01 +0000 (15:17 +0200)]
net: stmmac: ethtool: Let user configure TX coalesce without RIWT

When RX Watchdog is disabled its currently not possible to configure TX
coalesce settings. Let user configure it anyway.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: Only consider RX error when HW Timestamping is not enabled
Jose Abreu [Wed, 4 Sep 2019 13:17:00 +0000 (15:17 +0200)]
net: stmmac: Only consider RX error when HW Timestamping is not enabled

Only consider that we have an error when HW Timestamping is not enabled
as this can give false positives due to the fact the RX Timestamping in
XGMAC and GMAC cores comes from context descriptors.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: selftests: Implement the ARP Offload test
Jose Abreu [Wed, 4 Sep 2019 13:16:59 +0000 (15:16 +0200)]
net: stmmac: selftests: Implement the ARP Offload test

Implement a test for ARP Offload feature.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: xgmac: Implement ARP Offload
Jose Abreu [Wed, 4 Sep 2019 13:16:58 +0000 (15:16 +0200)]
net: stmmac: xgmac: Implement ARP Offload

Implement the ARP Offload feature in XGMAC cores.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: selftests: Add selftest for L3/L4 Filters
Jose Abreu [Wed, 4 Sep 2019 13:16:57 +0000 (15:16 +0200)]
net: stmmac: selftests: Add selftest for L3/L4 Filters

Adds the selftests for L3 and L4 filters with DA/SA/DP/SP support.

Changes from v1:
- Reduce stack usage (kbuild test robot)

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: Implement L3/L4 Filters using TC Flower
Jose Abreu [Wed, 4 Sep 2019 13:16:56 +0000 (15:16 +0200)]
net: stmmac: Implement L3/L4 Filters using TC Flower

Implement filters for Layer 3 and Layer 4 using TC Flower API. Add the
corresponding callbacks in XGMAC core.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: Do not return error code in TC Initialization
Jose Abreu [Wed, 4 Sep 2019 13:16:55 +0000 (15:16 +0200)]
net: stmmac: Do not return error code in TC Initialization

As we can still use the remaining TC callbacks, e.g. CBS. We should not
fail in the initialization only because RX Parser is not available.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: xgmac: Add RBU handling in DMA interrupt
Jose Abreu [Wed, 4 Sep 2019 13:16:54 +0000 (15:16 +0200)]
net: stmmac: xgmac: Add RBU handling in DMA interrupt

Add the handling of Receive Buffer Unavailable interrupt in the DMA
handler of XGMAC cores.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: stmmac: selftests: Return proper error code to userspace
Jose Abreu [Wed, 4 Sep 2019 13:16:53 +0000 (15:16 +0200)]
net: stmmac: selftests: Return proper error code to userspace

We can do better than just return 1 to userspace. Lets return a proper
Linux error code.

Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge tag 'linux-can-next-for-5.4-20190904' of git://git.kernel.org/pub/scm/linux...
David S. Miller [Thu, 5 Sep 2019 10:17:50 +0000 (12:17 +0200)]
Merge tag 'linux-can-next-for-5.4-20190904' of git://git./linux/kernel/git/mkl/linux-can-next

Marc Kleine-Budde says:

====================
pull-request: can-next 2019-09-04 j1939

this is a pull request for net-next/master consisting of 21 patches.

the first 12 patches are by me and target the CAN core infrastructure.
They clean up the names of variables , structs and struct members,
convert can_rx_register() to use max() instead of open coding it and
remove unneeded code from the can_pernet_exit() callback.

The next three patches are also by me and they introduce and make use of
the CAN midlayer private structure. It is used to hold protocol specific
per device data structures.

The next patch is by Oleksij Rempel, switches the
&net->can.rcvlists_lock from a spin_lock() to a spin_lock_bh(), so that
it can be used from NAPI (soft IRQ) context.

The next 4 patches are by Kurt Van Dijck, he first updates his email
address via mailmap and then extends sockaddr_can to include j1939
members.

The final patch is the collective effort of many entities (The j1939
authors: Oliver Hartkopp, Bastian Stender, Elenita Hinds, kbuild test
robot, Kurt Van Dijck, Maxime Jayat, Robin van der Gracht, Oleksij
Rempel, Marc Kleine-Budde). It adds support of SAE J1939 protocol to the
CAN networking stack.

SAE J1939 is the vehicle bus recommended practice used for communication
and diagnostics among vehicle components. Originating in the car and
heavy-duty truck industry in the United States, it is now widely used in
other parts of the world.

P.S.: This pull request doesn't invalidate my last pull request:
      "pull-request: can-next 2019-09-03".
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agorocker: add missing init_net check in FIB notifier
Jiri Pirko [Wed, 4 Sep 2019 07:40:47 +0000 (09:40 +0200)]
rocker: add missing init_net check in FIB notifier

Take only FIB events that are happening in init_net into account. No other
namespaces are supported.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-kzfree'
David S. Miller [Thu, 5 Sep 2019 10:06:04 +0000 (12:06 +0200)]
Merge branch 'net-kzfree'

zhong jiang says:

====================
net: Use kzfree() directly

With the help of Coccinelle. We find some place to replace.

@@
expression M, S;
@@

- memset(M, 0, S);
- kfree(M);
+ kzfree(M);
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet: mpoa: Use kzfree rather than its implementation.
zhong jiang [Wed, 4 Sep 2019 02:39:12 +0000 (10:39 +0800)]
net: mpoa: Use kzfree rather than its implementation.

Use kzfree instead of memset() + kfree().

Signed-off-by: zhong jiang <zhongjiang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agosunrpc: Use kzfree rather than its implementation.
zhong jiang [Wed, 4 Sep 2019 02:39:11 +0000 (10:39 +0800)]
sunrpc: Use kzfree rather than its implementation.

Use kzfree instead of memset() + kfree().

Signed-off-by: zhong jiang <zhongjiang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoixgbe: Use kzfree() rather than its implementation.
zhong jiang [Wed, 4 Sep 2019 02:39:10 +0000 (10:39 +0800)]
ixgbe: Use kzfree() rather than its implementation.

Use kzfree() instead of memset() + kfree().

Signed-off-by: zhong jiang <zhongjiang@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge tag 'linux-can-next-for-5.4-20190903' of git://git.kernel.org/pub/scm/linux...
David S. Miller [Thu, 5 Sep 2019 07:57:31 +0000 (09:57 +0200)]
Merge tag 'linux-can-next-for-5.4-20190903' of git://git./linux/kernel/git/mkl/linux-can-next

Marc Kleine-Budde says:

====================
pull-request: can-next 2019-09-03

this is a pull request for net-next/master consisting of 15 patches.

The first patch is by Christer Beskow, targets the kvaser_pciefd driver
and fixes the PWM generator's frequency.

The next three patches are by Dan Murphy, the tcan4x5x is updated to use
a proper interrupts/interrupt-parent DT binding to specify the devices
IRQ line. Further the unneeded wake ups of the device is removed from
the driver.

A patch by me for the mcp25xx driver removes the deprecated board file
setup example. Three patches by Andy Shevchenko simplify clock handling,
update the driver from OF to device property API and simplify the
mcp251x_can_suspend() function.

The remaining 7 patches are by me and clean up checkpatch warnings in
the generic CAN device infrastructure.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agovsock/virtio: a better comment on credit update
Michael S. Tsirkin [Tue, 3 Sep 2019 07:38:16 +0000 (03:38 -0400)]
vsock/virtio: a better comment on credit update

The comment we have is just repeating what the code does.
Include the *reason* for the condition instead.

Cc: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'net-tls-minor-cleanups'
David S. Miller [Thu, 5 Sep 2019 07:49:49 +0000 (09:49 +0200)]
Merge branch 'net-tls-minor-cleanups'

Jakub Kicinski says:

====================
net/tls: minor cleanups

This set is a grab bag of TLS cleanups accumulated in my tree
in an attempt to avoid merge problems with net. Nothing stands
out. First patch dedups context information. Next control path
locking is very slightly optimized. Fourth patch cleans up
ugly #ifdefs.
====================

Reviewed-by: Boris Pismenny <borisp@mellanox.com>
Reviewed-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: dedup the record cleanup
Jakub Kicinski [Tue, 3 Sep 2019 04:31:06 +0000 (21:31 -0700)]
net/tls: dedup the record cleanup

If retransmit record hint fall into the cleanup window we will
free it by just walking the list. No need to duplicate the code.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: clean up the number of #ifdefs for CONFIG_TLS_DEVICE
Jakub Kicinski [Tue, 3 Sep 2019 04:31:05 +0000 (21:31 -0700)]
net/tls: clean up the number of #ifdefs for CONFIG_TLS_DEVICE

TLS code has a number of #ifdefs which make the code a little
harder to follow. Recent fixes removed the ifdef around the
TLS_HW define, so we can switch to the often used pattern
of defining tls_device functions as empty static inlines
in the header when CONFIG_TLS_DEVICE=n.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: narrow down the critical area of device_offload_lock
Jakub Kicinski [Tue, 3 Sep 2019 04:31:04 +0000 (21:31 -0700)]
net/tls: narrow down the critical area of device_offload_lock

On setsockopt path we need to hold device_offload_lock from
the moment we check netdev is up until the context is fully
ready to be added to the tls_device_list.

No need to hold it around the get_netdev_for_sock().
Change the code and remove the confusing comment.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: don't jump to return
Jakub Kicinski [Tue, 3 Sep 2019 04:31:03 +0000 (21:31 -0700)]
net/tls: don't jump to return

Reusing parts of error path for normal exit will make
next commit harder to read, untangle the two.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/tls: use the full sk_proto pointer
Jakub Kicinski [Tue, 3 Sep 2019 04:31:02 +0000 (21:31 -0700)]
net/tls: use the full sk_proto pointer

Since we already have the pointer to the full original sk_proto
stored use that instead of storing all individual callback
pointers as well.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoConvert usage of IN_MULTICAST to ipv4_is_multicast
Dave Taht [Mon, 2 Sep 2019 23:29:36 +0000 (16:29 -0700)]
Convert usage of IN_MULTICAST to ipv4_is_multicast

IN_MULTICAST's primary intent is as a uapi macro.

Elsewhere in the kernel we use ipv4_is_multicast consistently.

This patch unifies linux's multicast checks to use that function
rather than this macro.

Signed-off-by: Dave Taht <dave.taht@gmail.com>
Reviewed-by: Toke Høiland-Jørgensen <toke@toke.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agonet/sched: cbs: remove redundant assignment to variable port_rate
Colin Ian King [Mon, 2 Sep 2019 18:26:37 +0000 (19:26 +0100)]
net/sched: cbs: remove redundant assignment to variable port_rate

Variable port_rate is being initialized with a value that is never read
and is being re-assigned a little later on. The assignment is redundant
and hence can be removed.

Addresses-Coverity: ("Unused value")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoMerge branch 'ionic-Add-ionic-driver'
David S. Miller [Thu, 5 Sep 2019 07:24:44 +0000 (09:24 +0200)]
Merge branch 'ionic-Add-ionic-driver'

Shannon Nelson says:

====================
ionic: Add ionic driver

This is a patch series that adds the ionic driver, supporting the Pensando
ethernet device.

In this initial patchset we implement basic transmit and receive.  Later
patchsets will add more advanced features.

Our thanks to Saeed Mahameed, David Miller, Andrew Lunn, Michal Kubecek,
Jacub Kicinski, Jiri Pirko, Yunsheng Lin, and the ever present kbuild
test robots for their comments and suggestions.

New in v7:
 - stop Tx queue if no descriptor space left after a Tx
 - return ETIMEDOUT if the module data can't be copied out safely
 - remove unnecessary synchronize_irq() before free_irq()
 - use eth_prepare_mac_addr_change() and eth_commit_mac_addr_change() helpers
 - propagate error out of ionic_dl_info_get()

New in v6:
 - added a new patch with devlink info tags for ASIC and general FW
 - use the new devlink info tags in the driver
 - fixed up TxRx cleanup on setup failure
 - allow for possible 0 address from dma mapping of Tx buffers
 - remove a few more unnecessary debugfs error checks
 - use innocuous hardcoded strings in the identify message
 - removed a couple of unused functions and definitions
 - fix a leak in the error handling of port_info setup
 - changed from BUILD_BUG_ON() to static_assert()

New in v5:
 - code reorganized for more sane layout, with a side benefit of getting
   rid of a "defined but not used" complaint after patch 5
 - added "ionic_" prefix to struct definitions and fixed up remaining
   reverse christmas tree formatting (I think I got them all...)
 - ndo_open and ndo_stop reworked for better error recovery
 - interrupt coalescing enabled at driver start
 - unnecessary log messaging removed from events
 - double copy added in the module prom read to assure a clean copy
 - added BQL counting
 - fixed a TSO unmap issue found in testing
 - generalize a bit-flag wait with timeout
 - added devlink into earlier code and dropped patch 19

New in v4:
 - use devlink struct alloc for ionic device specific struct
 - add support for devlink_port
 - fixup devlink fixed vs running version usage
 - use bitmap_copy() instead of memcpy() for link_ksettings
 - don't bother to zero out the advertising bits before copying
   in the support bits
 - drop unknown xcvr types (will be expanded on later)
 - flap the connection to force auto-negotiation
 - use is_power_of_2() rather than open code
 - simplify set/get_pauseparam use of pause->autoneg
 - add a couple comments about NIC status data updated in DMA spaces

New in v3:
 - use le32_to_cpu() on queue_count[] values in debugfs
 - dma_free_coherent() can handle NULL pointers
 - remove unused SS_TEST from ethtool handlers
 - one more case of stop the tx ring if there is no room
 - remove a couple of stray // comments

New in v2:
 - removed debugfs error checking and cut down on debugfs use
 - remove redundant bounds checking on incoming values for mtu and ethtool
 - don't alloc rx_filter memory until the match type has been checked
 - free the ionic struct on remove
 - simplified link_up and netif_carrier_ok comparison
 - put stats into ethtool -S, out of debugfs
 - moved dev_cmd and dev_info dumping to ethtool -d, out of debugfs
 - added devlink support
 - used kernel's rss init routines rather than open code
 - set the Kbuild dependant on 64BIT
 - cut down on some unnecessary log messaging
 - cleaned up ionic_get_link_ksettings
 - cleaned up other little code bits here and there
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add coalesce and other features
Shannon Nelson [Tue, 3 Sep 2019 22:28:21 +0000 (15:28 -0700)]
ionic: Add coalesce and other features

Interrupt coalescing, tunable copybreak value, and
tx timeout.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add RSS support
Shannon Nelson [Tue, 3 Sep 2019 22:28:20 +0000 (15:28 -0700)]
ionic: Add RSS support

Add code to manipulate through ethtool the RSS configuration
used by the NIC.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add driver stats
Shannon Nelson [Tue, 3 Sep 2019 22:28:19 +0000 (15:28 -0700)]
ionic: Add driver stats

Add in the detailed statistics for ethtool -S that the driver
keeps as it processes packets.  Display of the additional
debug statistics can be enabled through the ethtool priv-flags
feature.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add netdev-event handling
Shannon Nelson [Tue, 3 Sep 2019 22:28:18 +0000 (15:28 -0700)]
ionic: Add netdev-event handling

When the netdev gets a new name from userland, pass that name
down to the NIC for internal tracking.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add Tx and Rx handling
Shannon Nelson [Tue, 3 Sep 2019 22:28:17 +0000 (15:28 -0700)]
ionic: Add Tx and Rx handling

Add both the Tx and Rx queue setup and handling.  The related
stats display comes later.  Instead of using the generic napi
routines used by the slow-path commands, the Tx and Rx paths
are simplified and inlined in one file in order to get better
compiler optimizations.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add initial ethtool support
Shannon Nelson [Tue, 3 Sep 2019 22:28:16 +0000 (15:28 -0700)]
ionic: Add initial ethtool support

Add in the basic ethtool callbacks for device information
and control.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add async link status check and basic stats
Shannon Nelson [Tue, 3 Sep 2019 22:28:15 +0000 (15:28 -0700)]
ionic: Add async link status check and basic stats

Add code to handle the link status event, and wire up the
basic netdev hardware stats.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add Rx filter and rx_mode ndo support
Shannon Nelson [Tue, 3 Sep 2019 22:28:14 +0000 (15:28 -0700)]
ionic: Add Rx filter and rx_mode ndo support

Add the Rx filtering and rx_mode NDO callbacks.  Also add
the deferred work thread handling needed to manage the filter
requests outside of the netif_addr_lock spinlock.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add management of rx filters
Shannon Nelson [Tue, 3 Sep 2019 22:28:13 +0000 (15:28 -0700)]
ionic: Add management of rx filters

Set up the infrastructure for managing Rx filters.  We can't ask the
hardware for what filters it has, so we keep a local list of filters
that we've pushed into the HW.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add the basic NDO callbacks for netdev support
Shannon Nelson [Tue, 3 Sep 2019 22:28:12 +0000 (15:28 -0700)]
ionic: Add the basic NDO callbacks for netdev support

Set up the initial NDO structure and callbacks for netdev
to use, and register the netdev.  This will allow us to do
a few basic operations on the device, but no traffic yet.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add notifyq support
Shannon Nelson [Tue, 3 Sep 2019 22:28:11 +0000 (15:28 -0700)]
ionic: Add notifyq support

The AdminQ is fine for sending messages and requests to the NIC,
but we also need to have events published from the NIC to the
driver.  The NotifyQ handles this for us, using the same interrupt
as AdminQ.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add adminq action
Shannon Nelson [Tue, 3 Sep 2019 22:28:10 +0000 (15:28 -0700)]
ionic: Add adminq action

Add AdminQ specific message requests and completion handling.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add basic adminq support
Shannon Nelson [Tue, 3 Sep 2019 22:28:09 +0000 (15:28 -0700)]
ionic: Add basic adminq support

Most of the NIC configuration happens through the AdminQ message
queue.  NAPI is used for basic interrupt handling and message
queue management.  These routines are set up to be shared among
different types of queues when used in slow-path handling.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add interrupts and doorbells
Shannon Nelson [Tue, 3 Sep 2019 22:28:08 +0000 (15:28 -0700)]
ionic: Add interrupts and doorbells

The ionic interrupt model is based on interrupt control blocks
accessed through the PCI BAR.  Doorbell registers are used by
the driver to signal to the NIC that requests are waiting on
the message queues.  Interrupts are used by the NIC to signal
to the driver that answers are waiting on the completion queues.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add basic lif support
Shannon Nelson [Tue, 3 Sep 2019 22:28:07 +0000 (15:28 -0700)]
ionic: Add basic lif support

The LIF is the Logical Interface, which represents the external
connections.  The NIC can multiplex many LIFs to a single port,
but in most setups, LIF0 is the primary control for the port.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add port management commands
Shannon Nelson [Tue, 3 Sep 2019 22:28:06 +0000 (15:28 -0700)]
ionic: Add port management commands

The port management commands apply to the physical port
associated with the PCI device, which might be shared among
several logical interfaces.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add hardware init and device commands
Shannon Nelson [Tue, 3 Sep 2019 22:28:05 +0000 (15:28 -0700)]
ionic: Add hardware init and device commands

The ionic device has a small set of PCI registers, including a
device control and data space, and a large set of message
commands.

Also adds new DEVLINK_INFO_VERSION_GENERIC tags for
ASIC_ID, ASIC_REV, and FW.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agoionic: Add basic framework for IONIC Network device driver
Shannon Nelson [Tue, 3 Sep 2019 22:28:04 +0000 (15:28 -0700)]
ionic: Add basic framework for IONIC Network device driver

This patch adds a basic driver framework for the Pensando IONIC
network device.  There is no functionality right now other than
the ability to load and unload.

Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>
5 years agodevlink: Add new info version tags for ASIC and FW
Shannon Nelson [Tue, 3 Sep 2019 22:28:03 +0000 (15:28 -0700)]
devlink: Add new info version tags for ASIC and FW

The current tag set is still rather small and needs a couple
more tags to help with ASIC identification and to have a
more generic FW version.

Cc: Jiri Pirko <jiri@resnulli.us>
Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Shannon Nelson <snelson@pensando.io>
Signed-off-by: David S. Miller <davem@davemloft.net>