openwrt/staging/blogic.git
8 years agobpf, x86: add support for constant blinding
Daniel Borkmann [Fri, 13 May 2016 17:08:33 +0000 (19:08 +0200)]
bpf, x86: add support for constant blinding

This patch adds recently added constant blinding helpers into the
x86 eBPF JIT. In the bpf_int_jit_compile() path, requirements are
to utilize bpf_jit_blind_constants()/bpf_jit_prog_release_other()
pair for rewriting the program into a blinded one, and to map the
BPF_REG_AX register to a CPU register. The mapping of BPF_REG_AX
is at non-callee saved register r10, and thus shared with cached
skb->data used for ld_abs/ind and not in every program type needed.
When blinding is not used, there's zero additional overhead in the
generated image.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobpf: add generic constant blinding for use in jits
Daniel Borkmann [Fri, 13 May 2016 17:08:32 +0000 (19:08 +0200)]
bpf: add generic constant blinding for use in jits

This work adds a generic facility for use from eBPF JIT compilers
that allows for further hardening of JIT generated images through
blinding constants. In response to the original work on BPF JIT
spraying published by Keegan McAllister [1], most BPF JITs were
changed to make images read-only and start at a randomized offset
in the page, where the rest was filled with trap instructions. We
have this nowadays in x86, arm, arm64 and s390 JIT compilers.
Additionally, later work also made eBPF interpreter images read
only for kernels supporting DEBUG_SET_MODULE_RONX, that is, x86,
arm, arm64 and s390 archs as well currently. This is done by
default for mentioned JITs when JITing is enabled. Furthermore,
we had a generic and configurable constant blinding facility on our
todo for quite some time now to further make spraying harder, and
first implementation since around netconf 2016.

We found that for systems where untrusted users can load cBPF/eBPF
code where JIT is enabled, start offset randomization helps a bit
to make jumps into crafted payload harder, but in case where larger
programs that cross page boundary are injected, we again have some
part of the program opcodes at a page start offset. With improved
guessing and more reliable payload injection, chances can increase
to jump into such payload. Elena Reshetova recently wrote a test
case for it [2, 3]. Moreover, eBPF comes with 64 bit constants, which
can leave some more room for payloads. Note that for all this,
additional bugs in the kernel are still required to make the jump
(and of course to guess right, to not jump into a trap) and naturally
the JIT must be enabled, which is disabled by default.

For helping mitigation, the general idea is to provide an option
bpf_jit_harden that admins can tweak along with bpf_jit_enable, so
that for cases where JIT should be enabled for performance reasons,
the generated image can be further hardened with blinding constants
for unpriviledged users (bpf_jit_harden == 1), with trading off
performance for these, but not for privileged ones. We also added
the option of blinding for all users (bpf_jit_harden == 2), which
is quite helpful for testing f.e. with test_bpf.ko. There are no
further e.g. hardening levels of bpf_jit_harden switch intended,
rationale is to have it dead simple to use as on/off. Since this
functionality would need to be duplicated over and over for JIT
compilers to use, which are already complex enough, we provide a
generic eBPF byte-code level based blinding implementation, which is
then just transparently JITed. JIT compilers need to make only a few
changes to integrate this facility and can be migrated one by one.

This option is for eBPF JITs and will be used in x86, arm64, s390
without too much effort, and soon ppc64 JITs, thus that native eBPF
can be blinded as well as cBPF to eBPF migrations, so that both can
be covered with a single implementation. The rule for JITs is that
bpf_jit_blind_constants() must be called from bpf_int_jit_compile(),
and in case blinding is disabled, we follow normally with JITing the
passed program. In case blinding is enabled and we fail during the
process of blinding itself, we must return with the interpreter.
Similarly, in case the JITing process after the blinding failed, we
return normally to the interpreter with the non-blinded code. Meaning,
interpreter doesn't change in any way and operates on eBPF code as
usual. For doing this pre-JIT blinding step, we need to make use of
a helper/auxiliary register, here BPF_REG_AX. This is strictly internal
to the JIT and not in any way part of the eBPF architecture. Just like
in the same way as JITs internally make use of some helper registers
when emitting code, only that here the helper register is one
abstraction level higher in eBPF bytecode, but nevertheless in JIT
phase. That helper register is needed since f.e. manually written
program can issue loads to all registers of eBPF architecture.

The core concept with the additional register is: blind out all 32
and 64 bit constants by converting BPF_K based instructions into a
small sequence from K_VAL into ((RND ^ K_VAL) ^ RND). Therefore, this
is transformed into: BPF_REG_AX := (RND ^ K_VAL), BPF_REG_AX ^= RND,
and REG <OP> BPF_REG_AX, so actual operation on the target register
is translated from BPF_K into BPF_X one that is operating on
BPF_REG_AX's content. During rewriting phase when blinding, RND is
newly generated via prandom_u32() for each processed instruction.
64 bit loads are split into two 32 bit loads to make translation and
patching not too complex. Only basic thing required by JITs is to
call the helper bpf_jit_blind_constants()/bpf_jit_prog_release_other()
pair, and to map BPF_REG_AX into an unused register.

Small bpf_jit_disasm extract from [2] when applied to x86 JIT:

echo 0 > /proc/sys/net/core/bpf_jit_harden

  ffffffffa034f5e9 + <x>:
  [...]
  39:   mov    $0xa8909090,%eax
  3e:   mov    $0xa8909090,%eax
  43:   mov    $0xa8ff3148,%eax
  48:   mov    $0xa89081b4,%eax
  4d:   mov    $0xa8900bb0,%eax
  52:   mov    $0xa810e0c1,%eax
  57:   mov    $0xa8908eb4,%eax
  5c:   mov    $0xa89020b0,%eax
  [...]

echo 1 > /proc/sys/net/core/bpf_jit_harden

  ffffffffa034f1e5 + <x>:
  [...]
  39:   mov    $0xe1192563,%r10d
  3f:   xor    $0x4989b5f3,%r10d
  46:   mov    %r10d,%eax
  49:   mov    $0xb8296d93,%r10d
  4f:   xor    $0x10b9fd03,%r10d
  56:   mov    %r10d,%eax
  59:   mov    $0x8c381146,%r10d
  5f:   xor    $0x24c7200e,%r10d
  66:   mov    %r10d,%eax
  69:   mov    $0xeb2a830e,%r10d
  6f:   xor    $0x43ba02ba,%r10d
  76:   mov    %r10d,%eax
  79:   mov    $0xd9730af,%r10d
  7f:   xor    $0xa5073b1f,%r10d
  86:   mov    %r10d,%eax
  89:   mov    $0x9a45662b,%r10d
  8f:   xor    $0x325586ea,%r10d
  96:   mov    %r10d,%eax
  [...]

As can be seen, original constants that carry payload are hidden
when enabled, actual operations are transformed from constant-based
to register-based ones, making jumps into constants ineffective.
Above extract/example uses single BPF load instruction over and
over, but of course all instructions with constants are blinded.

Performance wise, JIT with blinding performs a bit slower than just
JIT and faster than interpreter case. This is expected, since we
still get all the performance benefits from JITing and in normal
use-cases not every single instruction needs to be blinded. Summing
up all 296 test cases averaged over multiple runs from test_bpf.ko
suite, interpreter was 55% slower than JIT only and JIT with blinding
was 8% slower than JIT only. Since there are also some extremes in
the test suite, I expect for ordinary workloads that the performance
for the JIT with blinding case is even closer to JIT only case,
f.e. nmap test case from suite has averaged timings in ns 29 (JIT),
35 (+ blinding), and 151 (interpreter).

BPF test suite, seccomp test suite, eBPF sample code and various
bigger networking eBPF programs have been tested with this and were
running fine. For testing purposes, I also adapted interpreter and
redirected blinded eBPF image to interpreter and also here all tests
pass.

  [1] http://mainisusuallyafunction.blogspot.com/2012/11/attacking-hardened-linux-systems-with.html
  [2] https://github.com/01org/jit-spray-poc-for-ksp/
  [3] http://www.openwall.com/lists/kernel-hardening/2016/05/03/5

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Elena Reshetova <elena.reshetova@intel.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobpf: prepare bpf_int_jit_compile/bpf_prog_select_runtime apis
Daniel Borkmann [Fri, 13 May 2016 17:08:31 +0000 (19:08 +0200)]
bpf: prepare bpf_int_jit_compile/bpf_prog_select_runtime apis

Since the blinding is strictly only called from inside eBPF JITs,
we need to change signatures for bpf_int_jit_compile() and
bpf_prog_select_runtime() first in order to prepare that the
eBPF program we're dealing with can change underneath. Hence,
for call sites, we need to return the latest prog. No functional
change in this patch.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobpf: add bpf_patch_insn_single helper
Daniel Borkmann [Fri, 13 May 2016 17:08:30 +0000 (19:08 +0200)]
bpf: add bpf_patch_insn_single helper

Move the functionality to patch instructions out of the verifier
code and into the core as the new bpf_patch_insn_single() helper
will be needed later on for blinding as well. No changes in
functionality.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobpf, x86/arm64: remove useless checks on prog
Daniel Borkmann [Fri, 13 May 2016 17:08:29 +0000 (19:08 +0200)]
bpf, x86/arm64: remove useless checks on prog

There is never such a situation, where bpf_int_jit_compile() is
called with either prog as NULL or len as 0, so the tests are
unnecessary and confusing as people would just copy them. s390
doesn't have them, so no change is needed there.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobpf: split HAVE_BPF_JIT into cBPF and eBPF variant
Daniel Borkmann [Fri, 13 May 2016 17:08:28 +0000 (19:08 +0200)]
bpf: split HAVE_BPF_JIT into cBPF and eBPF variant

Split the HAVE_BPF_JIT into two for distinguishing cBPF and eBPF JITs.

Current cBPF ones:

  # git grep -n HAVE_CBPF_JIT arch/
  arch/arm/Kconfig:44:    select HAVE_CBPF_JIT
  arch/mips/Kconfig:18:   select HAVE_CBPF_JIT if !CPU_MICROMIPS
  arch/powerpc/Kconfig:129:       select HAVE_CBPF_JIT
  arch/sparc/Kconfig:35:  select HAVE_CBPF_JIT

Current eBPF ones:

  # git grep -n HAVE_EBPF_JIT arch/
  arch/arm64/Kconfig:61:  select HAVE_EBPF_JIT
  arch/s390/Kconfig:126:  select HAVE_EBPF_JIT if PACK_STACK && HAVE_MARCH_Z196_FEATURES
  arch/x86/Kconfig:94:    select HAVE_EBPF_JIT                    if X86_64

Later code also needs this facility to check for eBPF JITs.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobpf: move bpf_jit_enable declaration
Daniel Borkmann [Fri, 13 May 2016 17:08:27 +0000 (19:08 +0200)]
bpf: move bpf_jit_enable declaration

Move the bpf_jit_enable declaration to the filter.h file where
most other core code is declared, also since we're going to add
a second knob there.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobpf: minor cleanups in ebpf code
Daniel Borkmann [Fri, 13 May 2016 17:08:26 +0000 (19:08 +0200)]
bpf: minor cleanups in ebpf code

Besides others, remove redundant comments where the code is self
documenting enough, and properly indent various bpf_verifier_ops
and bpf_prog_type_list declarations. Moreover, remove two exports
that actually have no module user.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: dsa: mv88e6xxx: remove bridge work
Vivien Didelot [Sat, 14 May 2016 00:38:23 +0000 (20:38 -0400)]
net: dsa: mv88e6xxx: remove bridge work

Now that the bridge code defers the switchdev port state setting, there
is no need to defer the port STP state change within the mv88e6xxx code.
Thus get rid of the driver's bridge work code.

This also fixes a race condition where the DSA layer assumes that the
bridge code already set the unbridged port's STP state to Disabled
before restoring the Forwarding state.

As a consequence, this also fixes the FDB flush for the unbridged port
which now correctly occurs during the Forwarding to Disabled transition.

Fixes: 0bc05d585d38 ("switchdev: allow caller to explicitly request attr_set as deferred")
Reported-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: vrf: protect changes to private data with rcu
David Ahern [Fri, 13 May 2016 19:23:45 +0000 (12:23 -0700)]
net: vrf: protect changes to private data with rcu

One cpu can be processing packets which includes using the cached route
entries in the vrf device's private data and on another cpu the device
gets deleted which releases the routes and sets the pointers in net_vrf
to NULL. This results in datapath dereferencing a NULL pointer.

Fix by protecting access to dst's with rcu.

Fixes: 193125dbd8eb ("net: Introduce VRF device driver")
Fixes: 35402e313663 ("net: Add IPv6 support to VRF device")
Signed-off-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agotcp: minor optimizations around tcp_hdr() usage
Eric Dumazet [Fri, 13 May 2016 16:16:40 +0000 (09:16 -0700)]
tcp: minor optimizations around tcp_hdr() usage

tcp_hdr() is slightly more expensive than using skb->data in contexts
where we know they point to the same byte.

In receive path, tcp_v4_rcv() and tcp_v6_rcv() are in this situation,
as tcp header has not been pulled yet.

In output path, the same can be said when we just pushed the tcp header
in the skb, in tcp_transmit_skb() and tcp_make_synack()

Also factorize the two checks for tcb->tcp_flags & TCPHDR_SYN in
tcp_transmit_skb() and pass tcp header pointer to tcp_ecn_send(),
so that compiler can further optimize and avoid a reload.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonetlink: kill nla_put_u64()
Nicolas Dichtel [Fri, 13 May 2016 13:25:40 +0000 (15:25 +0200)]
netlink: kill nla_put_u64()

This function is not used anymore. nla_put_u64_64bit() should be used
instead.

Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agosock: propagate __sock_cmsg_send() error
Eric Dumazet [Fri, 13 May 2016 13:14:37 +0000 (06:14 -0700)]
sock: propagate __sock_cmsg_send() error

__sock_cmsg_send() might return different error codes, not only -EINVAL.

Fixes: 24025c465f77 ("ipv4: process socket-level control messages in IPv4")
Fixes: ad1e46a83716 ("ipv6: process socket-level control messages in IPv6")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: qrtr: fix build problems
Arnd Bergmann [Fri, 13 May 2016 13:09:58 +0000 (15:09 +0200)]
net: qrtr: fix build problems

Having multiple loadable modules with the same name cannot work
with modprobe, and having both net/qrtr/smd.ko and drivers/soc/qcom/smd.ko
results in a (somewhat cryptic) build error:

ERROR: "qcom_smd_driver_unregister" [net/qrtr/smd.ko] undefined!
ERROR: "qcom_smd_driver_register" [net/qrtr/smd.ko] undefined!
ERROR: "qcom_smd_set_drvdata" [net/qrtr/smd.ko] undefined!
ERROR: "qcom_smd_send" [net/qrtr/smd.ko] undefined!
ERROR: "qcom_smd_get_drvdata" [net/qrtr/smd.ko] undefined!
ERROR: "qcom_smd_driver_unregister" [drivers/soc/qcom/wcnss_ctrl.ko] undefined!
ERROR: "qcom_smd_driver_register" [drivers/soc/qcom/wcnss_ctrl.ko] undefined!
ERROR: "qcom_smd_set_drvdata" [drivers/soc/qcom/wcnss_ctrl.ko] undefined!
ERROR: "qcom_smd_send" [drivers/soc/qcom/wcnss_ctrl.ko] undefined!
ERROR: "qcom_smd_get_drvdata" [drivers/soc/qcom/wcnss_ctrl.ko] undefined!

Also, the qrtr driver uses the SMD interface and has a Kconfig dependency,
but also allows for compile-testing when SMD is disabled. However, if
with QCOM_SMD=m and COMPILE_TEST=y we can end up with QRTR_SMD=y and
that fails with a related link error.

The changes the dependency so we can still compile-test the driver but
not have it built-in if SMD is a module, to avoid running in the broken
configuration, and changes the Makefile to provide the driver under
a different module name.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Fixes: bdabad3e363d ("net: Add Qualcomm IPC router")
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'tc_flower_offload'
David S. Miller [Mon, 16 May 2016 17:43:52 +0000 (13:43 -0400)]
Merge branch 'tc_flower_offload'

Amir Vadai says:

====================
sched,mlx5: Offloaded TC flower filter statistics

This patchset introduces counters support for offloaded cls_flower filters.
When the user calls 'tc show -s ..', fl_dump is called.
Before fl_dump() returns the statistics, it calls the NIC driver (using a new
ndo_setup_tc() command - TC_CLSFLOWER_STATS) to read the hardware counters and
update the statistics accordingly. A new TC action op was added (stats_update())
to be used by the NIC driver to update the statistics.

Patchset was applied and tested over commit ed7cbbc ("udp: Resolve NULL pointer
dereference over flow-based vxlan device")
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5e: Hardware offloaded flower filter statistics support
Amir Vadai [Fri, 13 May 2016 12:55:42 +0000 (12:55 +0000)]
net/mlx5e: Hardware offloaded flower filter statistics support

Introduce support in updating statistics of offloaded TC flower
classifiers. Currently only the DROP action is supported.

Signed-off-by: Amir Vadai <amirva@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5_core: Flow counters infrastructure
Amir Vadai [Fri, 13 May 2016 12:55:41 +0000 (12:55 +0000)]
net/mlx5_core: Flow counters infrastructure

If a counter has the aging flag set when created, it is added to a list
of counters that will be queried periodically from a workqueue.  query
result and last use timestamp are cached.
add/del counter must be very efficient since thousands of such
operations might be issued in a second.
There is only a single reference to counters without aging, therefore
no need for locks.
But, counters with aging enabled are stored in a list. In order to make
code as lockless as possible, all the list manipulation and access to
hardware is done from a single context - the periodic counters query
thread.

The hardware supports multiple counters per FTE, however currently we
are using one counter for each FTE.

Signed-off-by: Amir Vadai <amirva@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5_core: Introduce flow steering destination of type counter
Amir Vadai [Fri, 13 May 2016 12:55:40 +0000 (12:55 +0000)]
net/mlx5_core: Introduce flow steering destination of type counter

When adding a flow steering rule with a counter, need to supply a
destination of type MLX5_FLOW_DESTINATION_TYPE_COUNTER, with a pointer
to a struct mlx5_fc.
Also, MLX5_FLOW_CONTEXT_ACTION_COUNT bit should be set in the action.

Signed-off-by: Amir Vadai <amirva@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5_core: Firmware commands to support flow counters
Amir Vadai [Fri, 13 May 2016 12:55:39 +0000 (12:55 +0000)]
net/mlx5_core: Firmware commands to support flow counters

Getting packet/byte statistics on flows is done through flow counters.
Implement the firmware commands to alloc, free and query flow counters.

Signed-off-by: Amir Vadai <amirva@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/mlx5_core: Use a macro in mlx5_command_str()
Amir Vadai [Fri, 13 May 2016 12:55:38 +0000 (12:55 +0000)]
net/mlx5_core: Use a macro in mlx5_command_str()

Use a macro instead of copying the OP name.

Signed-off-by: Amir Vadai <amirva@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/sched: cls_flower: Hardware offloaded filters statistics support
Amir Vadai [Fri, 13 May 2016 12:55:37 +0000 (12:55 +0000)]
net/sched: cls_flower: Hardware offloaded filters statistics support

Introduce a new command in ndo_setup_tc() for hardware offloaded
filters, to call the NIC driver, and make it update the statistics.
This will be done before dumping the filter and its statistics.

Signed-off-by: Amir Vadai <amirva@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/sched: act_gact: Update statistics when offloaded to hardware
Amir Vadai [Fri, 13 May 2016 12:55:36 +0000 (12:55 +0000)]
net/sched: act_gact: Update statistics when offloaded to hardware

Implement the stats_update callback that will be called by NIC drivers
for hardware offloaded filters.

Signed-off-by: Amir Vadai <amirva@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/sched: Enable netdev drivers to update statistics of offloaded actions
Amir Vadai [Fri, 13 May 2016 12:55:35 +0000 (12:55 +0000)]
net/sched: Enable netdev drivers to update statistics of offloaded actions

Introduce stats_update callback. netdev driver could call it for offloaded
actions to update the basic statistics (packets, bytes and last use).
Since bstats_update() and bstats_cpu_update() use skb as an argument to
get the counters, _bstats_update() and _bstats_cpu_update(), that get
bytes and packets as arguments, were added.

Signed-off-by: Amir Vadai <amirva@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'pxa168_eth-perf'
David S. Miller [Mon, 16 May 2016 17:39:50 +0000 (13:39 -0400)]
Merge branch 'pxa168_eth-perf'

Jisheng Zhang says:

====================
net: pxa168_eth: improve performance

This series is to improve the pxa168_eth driver performance by using
{readl|writel}_relaxed or appropriate memory barriers.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: pxa168_eth: Use dma_wmb/rmb where appropriate
Jisheng Zhang [Fri, 13 May 2016 11:57:30 +0000 (19:57 +0800)]
net: pxa168_eth: Use dma_wmb/rmb where appropriate

Update the pxa168_eth driver to use the dma_rmb/wmb calls instead of the
full barriers in order to improve performance: reduced 97ns/39ns on
average in tx/rx path on Marvell BG4CT platform.

Signed-off-by: Jisheng Zhang <jszhang@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: pxa168_eth: use {readl|writel}_relaxed instead of readl/writel
Jisheng Zhang [Fri, 13 May 2016 11:57:29 +0000 (19:57 +0800)]
net: pxa168_eth: use {readl|writel}_relaxed instead of readl/writel

Since appropriate memory barriers are already there, use the relaxed
version to improve performance a bit.

Signed-off-by: Jisheng Zhang <jszhang@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agovxlan: set mac_header correctly in GPE mode
Jiri Benc [Fri, 13 May 2016 08:48:42 +0000 (10:48 +0200)]
vxlan: set mac_header correctly in GPE mode

For VXLAN-GPE, the interface is ARPHRD_NONE, thus we need to reset
mac_header after pulling the outer header.

v2: Put the code to the existing conditional block as suggested by
    Shmulik Ladkani.

Fixes: e1e5314de08b ("vxlan: implement GPE")
Signed-off-by: Jiri Benc <jbenc@redhat.com>
Reviewed-by: Shmulik Ladkani <shmulik.ladkani@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'xen-netback-control-ring'
David S. Miller [Mon, 16 May 2016 17:35:57 +0000 (13:35 -0400)]
Merge branch 'xen-netback-control-ring'

Paul Durrant says:

====================
xen-netback: support for control ring

My recent patch to import an up-to-date include/xen/interface/io/netif.h
from the Xen Project brought in the necessary definitions to support the
new control shared ring and protocol. This patch series updates xen-netback
to support the new ring.

Patch #1 adds the necessary boilerplate to map the control ring and handle
messages. No implementation of the new protocol is included in this patch
so that it can be kept to a reasonable size.

Patch #2 adds the protocol implementation.

Patch #3 adds support for passing has values calculated by xen-netback to
capable frontends.

Patch #4 adds support for accepting hash values calculated by capable
frontends and using them the set the socket buffer hash.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoxen-netback: use hash value from the frontend
Paul Durrant [Fri, 13 May 2016 08:37:29 +0000 (09:37 +0100)]
xen-netback: use hash value from the frontend

My recent patch to include/xen/interface/io/netif.h defines a new extra
info type that can be used to pass hash values between backend and guest
frontend.

This patch adds code to xen-netback to use the value in a hash extra
info fragment passed from the guest frontend in a transmit-side
(i.e. netback receive side) packet to set the skb hash accordingly.

Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoxen-netback: pass hash value to the frontend
Paul Durrant [Fri, 13 May 2016 08:37:28 +0000 (09:37 +0100)]
xen-netback: pass hash value to the frontend

My recent patch to include/xen/interface/io/netif.h defines a new extra
info type that can be used to pass hash values between backend and guest
frontend.

This patch adds code to xen-netback to pass hash values calculated for
guest receive-side packets (i.e. netback transmit side) to the frontend.

Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoxen-netback: add control protocol implementation
Paul Durrant [Fri, 13 May 2016 08:37:27 +0000 (09:37 +0100)]
xen-netback: add control protocol implementation

My recent patch to include/xen/interface/io/netif.h defines a new shared
ring (in addition to the rx and tx rings) for passing control messages
from a VM frontend driver to a backend driver.

A previous patch added the necessary boilerplate for mapping the control
ring from the frontend, should it be created. This patch adds
implementations for each of the defined protocol messages.

Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Cc: Wei Liu <wei.liu2@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoxen-netback: add control ring boilerplate
Paul Durrant [Fri, 13 May 2016 08:37:26 +0000 (09:37 +0100)]
xen-netback: add control ring boilerplate

My recent patch to include/xen/interface/io/netif.h defines a new shared
ring (in addition to the rx and tx rings) for passing control messages
from a VM frontend driver to a backend driver.

This patch adds the necessary code to xen-netback to map this new shared
ring, should it be created by a frontend, but does not add implementations
for any of the defined protocol messages. These are added in a subsequent
patch for clarity.

Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Acked-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'cls_u32_hw_sw'
David S. Miller [Mon, 16 May 2016 17:30:57 +0000 (13:30 -0400)]
Merge branch 'cls_u32_hw_sw'

Sridhar Samudrala says:

====================
Enable SW only or HW only offloads with u32 classifier

This set of patches export TCA_CLS_FLAGS_SKIP_HW to userspace and also
introduces another flag TCA_CLS_FLAGS_SKIP_SW. These flags enable offloading
u32 filters to either SW or HW only.

The default semantics with no flags is to add the filter to HW if possible and
also into SW.
With SKIP_HW flag, the filter is only added to SW.
With SKIP_SW flag, the filter is added to HW and an error is returned
to user on failure.
These flags are mutually exclusive.
There was an earlier discussion on these semantics in the following email
thread.
http://thread.gmane.org/gmane.linux.network/401733
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: cls_u32: Add support for skip-sw flag to tc u32 classifier.
Samudrala, Sridhar [Fri, 13 May 2016 00:08:23 +0000 (17:08 -0700)]
net: cls_u32: Add support for skip-sw flag to tc u32 classifier.

On devices that support TC U32 offloads, this flag enables a filter to be
added only to HW. skip-sw and skip-hw are mutually exclusive flags. By
default without any flags, the filter is added to both HW and SW, but no
error checks are done in case of failure to add to HW. With skip-sw,
failure to add to HW is treated as an error.

Here is a sample script that adds 2 filters, one with skip-sw and the other
with skip-hw flag.

   # add ingress qdisc
   tc qdisc add dev p4p1 ingress

   # enable hw tc offload.
   ethtool -K p4p1 hw-tc-offload on

   # add u32 filter with skip-sw flag.
   tc filter add dev p4p1 parent ffff: protocol ip prio 99 \
      handle 800:0:1 u32 ht 800: flowid 800:1 \
      skip-sw \
      match ip src 192.168.1.0/24 \
      action drop

   # add u32 filter with skip-hw flag.
   tc filter add dev p4p1 parent ffff: protocol ip prio 99 \
      handle 800:0:2 u32 ht 800: flowid 800:2 \
      skip-hw \
      match ip src 192.168.2.0/24 \
      action drop

Signed-off-by: Sridhar Samudrala <sridhar.samudrala@intel.com>
Acked-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet: sched: Move TCA_CLS_FLAGS_SKIP_HW to uapi header file.
Samudrala, Sridhar [Fri, 13 May 2016 00:08:22 +0000 (17:08 -0700)]
net: sched: Move TCA_CLS_FLAGS_SKIP_HW to uapi header file.

Signed-off-by: Sridhar Samudrala <sridhar.samudrala@intel.com>
Acked-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'hv_netvsc-races'
David S. Miller [Mon, 16 May 2016 17:26:01 +0000 (13:26 -0400)]
Merge branch 'hv_netvsc-races'

Vitaly Kuznetsov says:

====================
hv_netvsc: avoid races on mtu change/set channels

Changes since v1:
- Rebased to net-next [Haiyang Zhang]

Original description:

MTU change and set channels operations are implemented as netvsc device
re-creation destroying internal structures (struct net_device stays). This
is really unfortunate but there is no support from Hyper-V host to do it
in a different way. Such re-creation is unsurprisingly racy, Haiyang
reported a crash when netvsc_change_mtu() is racing with
netvsc_link_change() but I was able to identify additional races upon
investigation. Both netvsc_set_channels() and netvsc_change_mtu() race
against:
1) netvsc_link_change()
2) netvsc_remove()
3) netvsc_send()

To solve these issues without introducing new locks some refactoring is
required. We need to get rid of very complex link graph in all the
internal structures and avoid traveling through structures which are being
removed.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agohv_netvsc: set nvdev link after populating chn_table
Vitaly Kuznetsov [Fri, 13 May 2016 11:55:25 +0000 (13:55 +0200)]
hv_netvsc: set nvdev link after populating chn_table

Crash in netvsc_send() is observed when netvsc device is re-created on
mtu change/set channels. The crash is caused by dereferencing of NULL
channel pointer which comes from chn_table. The root cause is a mixture
of two facts:
- we set nvdev pointer in net_device_context in alloc_net_device()
  before we populate chn_table.
- we populate chn_table[0] only.

The issue could be papered over by checking channel != NULL in
netvsc_send() but populating the whole chn_table and writing the
nvdev pointer afterwards seems more appropriate.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agohv_netvsc: synchronize netvsc_change_mtu()/netvsc_set_channels() with netvsc_remove()
Vitaly Kuznetsov [Fri, 13 May 2016 11:55:24 +0000 (13:55 +0200)]
hv_netvsc: synchronize netvsc_change_mtu()/netvsc_set_channels() with netvsc_remove()

When netvsc device is removed during mtu change or channels setup we get
into troubles as both paths are trying to remove the device. Synchronize
them with start_remove flag and rtnl lock.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agohv_netvsc: get rid of struct net_device pointer in struct netvsc_device
Vitaly Kuznetsov [Fri, 13 May 2016 11:55:23 +0000 (13:55 +0200)]
hv_netvsc: get rid of struct net_device pointer in struct netvsc_device

Simplify netvsvc pointer graph by getting rid of the redundant ndev
pointer. We can always get a pointer to struct net_device from somewhere
else.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agohv_netvsc: untangle the pointer mess
Vitaly Kuznetsov [Fri, 13 May 2016 11:55:22 +0000 (13:55 +0200)]
hv_netvsc: untangle the pointer mess

We have the following structures keeping netvsc adapter state:
- struct net_device
- struct net_device_context
- struct netvsc_device
- struct rndis_device
- struct hv_device
and there are pointers/dependencies between them:
- struct net_device_context is contained in struct net_device
- struct hv_device has driver_data pointer which points to
  'struct net_device' OR 'struct netvsc_device' depending on driver's
  state (!).
- struct net_device_context has a pointer to 'struct hv_device'.
- struct netvsc_device has pointers to 'struct hv_device' and
  'struct net_device_context'.
- struct rndis_device has a pointer to 'struct netvsc_device'.

Different functions get different structures as parameters and use these
pointers for traveling. The problem is (in addition to keeping in mind
this complex graph) that some of these structures (struct netvsc_device
and struct rndis_device) are being removed and re-created on mtu change
(as we implement it as re-creation of hyper-v device) so our travel using
these pointers is dangerous.

Simplify this to a the following:
- add struct netvsc_device pointer to struct net_device_context (which is
  a part of struct net_device and thus never disappears)
- remove struct hv_device and struct net_device_context pointers from
  struct netvsc_device
- replace pointer to 'struct netvsc_device' with pointer to
  'struct net_device'.
- always keep 'struct net_device' in hv_device driver_data.

We'll end up with the following 'circular' structure:

net_device:
 [net_device_context] -> netvsc_device -> rndis_device -> net_device
                      -> hv_device -> net_device

On MTU change we'll be removing the 'netvsc_device -> rndis_device'
branch and re-creating it making the synchronization easier.

There is one additional redundant pointer left, it is struct net_device
link in struct netvsc_device, it is going to be removed in a separate
commit.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agohv_netvsc: use start_remove flag to protect netvsc_link_change()
Vitaly Kuznetsov [Fri, 13 May 2016 11:55:21 +0000 (13:55 +0200)]
hv_netvsc: use start_remove flag to protect netvsc_link_change()

netvsc_link_change() can race with netvsc_change_mtu() or
netvsc_set_channels() as these functions destroy struct netvsc_device and
rndis filter. Use start_remove flag for syncronization. As
netvsc_change_mtu()/netvsc_set_channels() are called with rtnl lock held
we need to take it before checking start_remove value in
netvsc_link_change().

Reported-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agohv_netvsc: move start_remove flag to net_device_context
Vitaly Kuznetsov [Fri, 13 May 2016 11:55:20 +0000 (13:55 +0200)]
hv_netvsc: move start_remove flag to net_device_context

struct netvsc_device is destroyed on mtu change so keeping the
protection flag there is not a good idea. Move it to struct
net_device_context which is preserved.

Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agophy: add support for a reset-gpio specification
Uwe Kleine-König [Thu, 12 May 2016 10:00:33 +0000 (12:00 +0200)]
phy: add support for a reset-gpio specification

The framework only asserts (for now) that the reset gpio is not active.

Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Reviewed-by: Roger Quadros <rogerq@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Sun, 15 May 2016 17:47:27 +0000 (13:47 -0400)]
Merge branch '40GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
40GbE Intel Wired LAN Driver Updates 2016-05-14

This series contains updates to i40e and i40evf.

Kevin adds support to disable link on all ports and changes bits set
for telling firmware the PHY needs to be modified by the driver.

Anjali adds a feature to enable/disable all multicast for a trusted
VF.  Added priv-flag knob to configure global true promiscuous
support.

Shannon adds the support code for calling the admin queue API call
aq_set_switch_config().

Mitch modifies the VF, to log a message if an untrusted VF attempts to
configure promiscuous mode, but lies to it and returns everything is ok
instead of returning an error.  Corrects the logic for reporting the
receive packet hash.  Fixed the adding of a broadcast filter for VFs,
since that all VSIs are configured to receive broadcasts as default,
so do not need to add a filter.

Catherine refactors the ethtool get_settings to report the possible
supported link modes from what we know about the current PHY type and
that with the firmware supported PHY types.

Jacob changes the driver to use WARN_ONCE in order to highlight the
issue, but do not display a warning every time when receive hang
message is received.

Akeem corrects receive ptype payload layer for non_tunneled IPv6, when
it should be layer 4 for UDP, instead of layer 3.

Dan Carpenter fixes an uninitialized variable bug.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'bnxt_en-next'
David S. Miller [Sun, 15 May 2016 17:35:49 +0000 (13:35 -0400)]
Merge branch 'bnxt_en-next'

Michael Chan says:

====================
bnxt_en: updates for net-next.

Non-critical bug fixes, improvements, a new ethtool feature, and a new
device ID.

v2: Fixed a bug in bnxt_get_module_eeprom() found by Ben Hutchings.

Ajit Khaparde (2):
  bnxt_en: Add Support for ETHTOOL_GMODULEINFO and ETHTOOL_GMODULEEEPRO
  bnxt_en: Report PCIe link speed and width during driver load

Michael Chan (6):
  bnxt_en: Reduce maximum ring pages if page size is 64K.
  bnxt_en: Improve the delay logic for firmware response.
  bnxt_en: Fix length value in dmesg log firmware error message.
  bnxt_en: Simplify and improve unsupported SFP+ module reporting.
  bnxt_en: Add BCM57314 device ID.
  bnxt_en: Use dma_rmb() instead of rmb().

Satish Baddipadige (1):
  bnxt_en: Fix invalid max channel parameter in ethtool -l.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobnxt_en: Use dma_rmb() instead of rmb().
Michael Chan [Sun, 15 May 2016 07:04:51 +0000 (03:04 -0400)]
bnxt_en: Use dma_rmb() instead of rmb().

Use the weaker but more appropriate dma_rmb() to order the reading of
the completion ring.

Suggested-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobnxt_en: Add BCM57314 device ID.
Michael Chan [Sun, 15 May 2016 07:04:50 +0000 (03:04 -0400)]
bnxt_en: Add BCM57314 device ID.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobnxt_en: Simplify and improve unsupported SFP+ module reporting.
Michael Chan [Sun, 15 May 2016 07:04:49 +0000 (03:04 -0400)]
bnxt_en: Simplify and improve unsupported SFP+ module reporting.

The current code is more complicated than necessary and can only report
unsupported SFP+ module if it is plugged in after the device is up.

Rename bnxt_port_module_event() to bnxt_get_port_module_status().  We
already have the current module_status in the link_info structure, so
just check that and report any unsupported SFP+ module status.  Delete
the unnecessary last_port_module_event.  Call this function at the
end of bnxt_open to report unsupported module already plugged in.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobnxt_en: Fix length value in dmesg log firmware error message.
Michael Chan [Sun, 15 May 2016 07:04:48 +0000 (03:04 -0400)]
bnxt_en: Fix length value in dmesg log firmware error message.

The len value in the hwrm error message is wrong.  Use the properly adjusted
value in the variable len.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobnxt_en: Improve the delay logic for firmware response.
Michael Chan [Sun, 15 May 2016 07:04:47 +0000 (03:04 -0400)]
bnxt_en: Improve the delay logic for firmware response.

The current code has 2 problems:

1. The maximum wait time is not long enough.  It is about 60% of the
duration specified by the firmware.  It is calling usleep_range(600, 800)
for every 1 msec we are supposed to wait.

2. The granularity of the delay is too coarse.  Many simple firmware
commands finish in 25 usec or less.

We fix these 2 issues by multiplying the original 1 msec loop counter by
40 and calling usleep_range(25, 40) for each iteration.

There is also a second delay loop to wait for the last DMA word to
complete.  This delay loop should be a very short 5 usec wait.

This change results in much faster bring-up/down time:

Before the patch:

time ip link set p4p1 up

real    0m0.120s
user    0m0.001s
sys     0m0.009s

After the patch:

time ip link set p4p1 up

real    0m0.030s
user    0m0.000s
sys     0m0.010s

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobnxt_en: Reduce maximum ring pages if page size is 64K.
Michael Chan [Sun, 15 May 2016 07:04:46 +0000 (03:04 -0400)]
bnxt_en: Reduce maximum ring pages if page size is 64K.

The chip supports 4K/8K/64K page sizes for the rings and we try to
match it to the CPU PAGE_SIZE.  The current page size limits for the rings
are based on 4K/8K page size. If the page size is 64K, these limits are
too large.  Reduce them appropriately.

Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobnxt_en: Report PCIe link speed and width during driver load
Ajit Khaparde [Sun, 15 May 2016 07:04:45 +0000 (03:04 -0400)]
bnxt_en: Report PCIe link speed and width during driver load

Add code to log a message during driver load indicating PCIe link
speed and width.

The log message will look like this:
bnxt_en 0000:86:00.0 eth0: PCIe: Speed 8.0GT/s Width x8

Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobnxt_en: Add Support for ETHTOOL_GMODULEINFO and ETHTOOL_GMODULEEEPRO
Ajit Khaparde [Sun, 15 May 2016 07:04:44 +0000 (03:04 -0400)]
bnxt_en: Add Support for ETHTOOL_GMODULEINFO and ETHTOOL_GMODULEEEPRO

Add support to fetch the SFP EEPROM settings from the firmware
and display it via the ethtool -m command.  We support SFP+ and QSFP
modules.

v2: Fixed a bug in bnxt_get_module_eeprom() found by Ben Hutchings.

Signed-off-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobnxt_en: Fix invalid max channel parameter in ethtool -l.
Satish Baddipadige [Sun, 15 May 2016 07:04:43 +0000 (03:04 -0400)]
bnxt_en: Fix invalid max channel parameter in ethtool -l.

When there is only 1 MSI-X vector or in INTA mode, tx and rx pre-set
max channel parameters are shown incorrectly in ethtool -l.  With only 1
vector, bnxt_get_max_rings() will return -ENOMEM.  bnxt_get_channels
should check this return value, and set max_rx/max_tx to 0 if it is
non-zero.

Signed-off-by: Satish Baddipadige <sbaddipa@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
David S. Miller [Sun, 15 May 2016 17:32:12 +0000 (13:32 -0400)]
Merge git://git./linux/kernel/git/davem/net

The nf_conntrack_core.c fix in 'net' is not relevant in 'net-next'
because we no longer have a per-netns conntrack hash.

The ip_gre.c conflict as well as the iwlwifi ones were cases of
overlapping changes.

Conflicts:
drivers/net/wireless/intel/iwlwifi/mvm/tx.c
net/ipv4/ip_gre.c
net/netfilter/nf_conntrack_core.c

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Sat, 14 May 2016 21:15:06 +0000 (14:15 -0700)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Fix mvneta/bm dependencies, from Arnd Bergmann.

 2) RX completion hw bug workaround in bnxt_en, from Michael Chan.

 3) Kernel pointer leak in nf_conntrack, from Linus.

 4) Hoplimit route attribute limits not enforced properly, from Paolo
    Abeni.

 5) qlcnic driver NULL deref fix from Dan Carpenter.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net:
  arm64: bpf: jit JMP_JSET_{X,K}
  net/route: enforce hoplimit max value
  nf_conntrack: avoid kernel pointer value leak in slab name
  drivers: net: xgene: fix register offset
  drivers: net: xgene: fix statistics counters race condition
  drivers: net: xgene: fix ununiform latency across queues
  drivers: net: xgene: fix sharing of irqs
  drivers: net: xgene: fix IPv4 forward crash
  xen-netback: fix extra_info handling in xenvif_tx_err()
  net: mvneta: bm: fix dependencies again
  bnxt_en: Add workaround to detect bad opaque in rx completion (part 2)
  bnxt_en: Add workaround to detect bad opaque in rx completion (part 1)
  qlcnic: potential NULL dereference in qlcnic_83xx_get_minidump_template()

8 years agonet: switchdev: Drop EXPERIMENTAL from description
Florian Fainelli [Sat, 14 May 2016 19:49:54 +0000 (12:49 -0700)]
net: switchdev: Drop EXPERIMENTAL from description

Switchdev has been around for quite a while now, putting "EXPERIMENTAL"
in the description is no longer accurate, drop it.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoarm64: bpf: jit JMP_JSET_{X,K}
Zi Shen Lim [Fri, 13 May 2016 06:37:58 +0000 (23:37 -0700)]
arm64: bpf: jit JMP_JSET_{X,K}

Original implementation commit e54bcde3d69d ("arm64: eBPF JIT compiler")
had the relevant code paths, but due to an oversight always fail jiting.

As a result, we had been falling back to BPF interpreter whenever a BPF
program has JMP_JSET_{X,K} instructions.

With this fix, we confirm that the corresponding tests in lib/test_bpf
continue to pass, and also jited.

...
[    2.784553] test_bpf: #30 JSET jited:1 188 192 197 PASS
[    2.791373] test_bpf: #31 tcpdump port 22 jited:1 325 677 625 PASS
[    2.808800] test_bpf: #32 tcpdump complex jited:1 323 731 991 PASS
...
[    3.190759] test_bpf: #237 JMP_JSET_K: if (0x3 & 0x2) return 1 jited:1 110 PASS
[    3.192524] test_bpf: #238 JMP_JSET_K: if (0x3 & 0xffffffff) return 1 jited:1 98 PASS
[    3.211014] test_bpf: #249 JMP_JSET_X: if (0x3 & 0x2) return 1 jited:1 120 PASS
[    3.212973] test_bpf: #250 JMP_JSET_X: if (0x3 & 0xffffffff) return 1 jited:1 89 PASS
...

Fixes: e54bcde3d69d ("arm64: eBPF JIT compiler")
Signed-off-by: Zi Shen Lim <zlim.lnx@gmail.com>
Acked-by: Will Deacon <will.deacon@arm.com>
Acked-by: Yang Shi <yang.shi@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonet/route: enforce hoplimit max value
Paolo Abeni [Fri, 13 May 2016 16:33:41 +0000 (18:33 +0200)]
net/route: enforce hoplimit max value

Currently, when creating or updating a route, no check is performed
in both ipv4 and ipv6 code to the hoplimit value.

The caller can i.e. set hoplimit to 256, and when such route will
 be used, packets will be sent with hoplimit/ttl equal to 0.

This commit adds checks for the RTAX_HOPLIMIT value, in both ipv4
ipv6 route code, substituting any value greater than 255 with 255.

This is consistent with what is currently done for ADVMSS and MTU
in the ipv4 code.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonf_conntrack: avoid kernel pointer value leak in slab name
Linus Torvalds [Sat, 14 May 2016 18:11:44 +0000 (11:11 -0700)]
nf_conntrack: avoid kernel pointer value leak in slab name

The slab name ends up being visible in the directory structure under
/sys, and even if you don't have access rights to the file you can see
the filenames.

Just use a 64-bit counter instead of the pointer to the 'net' structure
to generate a unique name.

This code will go away in 4.7 when the conntrack code moves to a single
kmemcache, but this is the backportable simple solution to avoiding
leaking kernel pointers to user space.

Fixes: 5b3501faa874 ("netfilter: nf_conntrack: per netns nf_conntrack_cachep")
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Linus Torvalds [Sat, 14 May 2016 18:59:43 +0000 (11:59 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/viro/vfs

Pull vfs fixes from Al Viro:
 "Overlayfs fixes from Miklos, assorted fixes from me.

  Stable fodder of varying severity, all sat in -next for a while"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  ovl: ignore permissions on underlying lookup
  vfs: add lookup_hash() helper
  vfs: rename: check backing inode being equal
  vfs: add vfs_select_inode() helper
  get_rock_ridge_filename(): handle malformed NM entries
  ecryptfs: fix handling of directory opening
  atomic_open(): fix the handling of create_error
  fix the copy vs. map logics in blk_rq_map_user_iov()
  do_splice_to(): cap the size before passing to ->splice_read()

8 years agoi40e: fix an uninitialized variable bug
Dan Carpenter [Thu, 5 May 2016 13:18:02 +0000 (16:18 +0300)]
i40e: fix an uninitialized variable bug

We removed this initialization but it is required.  Let's put it back.

Fixes: 895106a577c4 ('i40e: trivial fixes')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoi40e: Bump version from 1.5.10 to 1.5.16
Bimmy Pujari [Tue, 3 May 2016 22:13:20 +0000 (15:13 -0700)]
i40e: Bump version from 1.5.10 to 1.5.16

Signed-off-by: Bimmy Pujari <bimmy.pujari@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoi40e: don't add broadcast filter for VFs
Mitch Williams [Tue, 3 May 2016 22:13:19 +0000 (15:13 -0700)]
i40e: don't add broadcast filter for VFs

Now that all VSIs are configured to receive broadcasts as default, we
don't need to add a filter. This eliminates an annoying but harmless
error message each time VFs are created or reset.

Change-ID: I4cd6339684df45b0d2722133eeb84c14fa93ea19
Signed-off-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>
8 years agoi40e/i40evf: properly report Rx packet hash
Mitch Williams [Tue, 3 May 2016 22:13:18 +0000 (15:13 -0700)]
i40e/i40evf: properly report Rx packet hash

This logic is inverted. If the RXHASH flag is set, then we should go
ahead and call skb_set_hash.

Change-ID: Ib2e30356dced1d3e939c8061ab6ad5bd94197e7c
Signed-off-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>
8 years agoi40e: set context to use VSI RSS LUT for SR-IOV
Ashish Shah [Tue, 3 May 2016 22:13:17 +0000 (15:13 -0700)]
i40e: set context to use VSI RSS LUT for SR-IOV

For the SR-IOV VSIs, when the queue filtering section is valid,
the RSS LUT needs to be set to use the VSI specific lookup table
(otherwise it will use the PF RSS LUT table).

Change-ID: Ia9377cc818078238a75c3bdeade1b593a91b3480
Signed-off-by: Ashish Shah <ashish.n.shah@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoi40e: Correct UDP packet header for non_tunnel-ipv6
Akeem G Abodunrin [Tue, 3 May 2016 22:13:16 +0000 (15:13 -0700)]
i40e: Correct UDP packet header for non_tunnel-ipv6

This patch corrects Rx ptype payload layer for non_tunneled ipv6. It
should be layer 4 for UDP, instead of layer 3.

Change-ID: I9382e4458ab3c4e58f6d2e9f195d5d4ee513805e
Signed-off-by: Akeem G Abodunrin <akeem.g.abodunrin@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoi40e: change Rx hang message into a WARN_ONCE
Jacob Keller [Tue, 3 May 2016 22:13:15 +0000 (15:13 -0700)]
i40e: change Rx hang message into a WARN_ONCE

Use WARN_ONCE in order to highlight the issue, but don't display
a warning every time. The user should be able to see the ethtool counter
we created if necessary to see how often it is occurring.

Change-ID: I40c4ea159819b64a7d33b7f5716749089791533a
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>
8 years agoi40e: Refactor ethtool get_settings
Catherine Sullivan [Tue, 3 May 2016 22:13:14 +0000 (15:13 -0700)]
i40e: Refactor ethtool get_settings

Previously we were only looking at the FW supported PHY types if link
was down, because we want to be more specific when link is up. This
refactor changes this. When link is down, we still rely on the FW
supported PHY types, but when link is up, we select the possible
supported link modes from what we know about the current PHY type, and
AND that with the FW supported PHY types.

Change-ID: Ice5dad83f2a17932b0b8b59f07439696ad6aa013
Signed-off-by: Catherine Sullivan <catherine.sullivan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoi40e: lie to the VF
Mitch Williams [Tue, 3 May 2016 22:13:13 +0000 (15:13 -0700)]
i40e: lie to the VF

If an untrusted VF attempts to configure promiscuous mode, log a message
pointing out its naughty behavior. But then, instead of returning an
error to the offender, just lie to it and say everything's OK. It will
continue on its way, thinking it's in promiscuous mode, but receiving no
packets except its own.

Change-ID: I63369215b1720f3c531eedfc06af86ff8c0e3dc8
Signed-off-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>
8 years agoi40e: Add vf-true-promisc-support priv flag
Anjali Singhai Jain [Tue, 3 May 2016 22:13:12 +0000 (15:13 -0700)]
i40e: Add vf-true-promisc-support priv flag

This patch adds priv-flag knob to configure global true promisc
support. With this patch the user can decide the flavor of
promiscuous that the VFs will see when promiscuous mode is enabled
on the interface. Since this a global setting for the whole device,
the priv-flag is exposed only on the first PF of the device.

The default is true promisc support is off, which means the promisc
mode for the VF will be limited/defport mode.

For the PF, we still will be in limited promisc unless in MFP mode
irrespective of the flavor picked through this knob.

Usage:
On PF0
ethtool --show-priv-flags p261p1
Private flags for p261p1:
MFP                    : off
LinkPolling            : off
flow-director-atr      : on
veb-stats              : off
hw-atr-eviction        : off
vf-true-promisc-support: off

to enable setting true promisc
ethtool --set-priv-flags p261p1 vf-true-promisc-support on

At this point if the VF is set to trust and promisc is enabled
on the VF through
ip link set ... promisc on
The VF/VFs will be able to see ALL ingress traffic

Change-Id: I8fac4b6eb1af9ca77b5376b79c50bdce5055bd94
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoi40e: Implement the API function for aq_set_switch_config
Shannon Nelson [Tue, 3 May 2016 22:13:11 +0000 (15:13 -0700)]
i40e: Implement the API function for aq_set_switch_config

Add the support code for calling the AdminQ API call aq_set_switch_config

Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoi40e: Add allmulti support for the VF
Anjali Singhai Jain [Tue, 3 May 2016 22:13:10 +0000 (15:13 -0700)]
i40e: Add allmulti support for the VF

This patch enables a feature to enable/disable all multicast
for a trusted VF.

Change-Id: I926eba7f8850c8d40f8ad7e08bbe4056bbd3985f
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoi40e: Add support for disabling all link and change bits needed for PHY interactions
Kevin Scott [Tue, 3 May 2016 22:13:09 +0000 (15:13 -0700)]
i40e: Add support for disabling all link and change bits needed for PHY interactions

Add flag to tell firmware to disable link on all ports.

This patch changes the bits set for telling firmware the PHY needs
to be modified by driver.  Without this patch, the setting will only
set that mode for the current port on the device.  Because the
MDIO interface is common for the copper device. The command needs to
set the mode for all ports.

Change-ID: I8baa7da91d384291ac95b41ae1a516604f8eb67f
Signed-off-by: Kevin Scott <kevin.c.scott@intel.com>
Signed-off-by: Carolyn Wyborny <carolyn.wyborny@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoMerge branch 'xgene-fixes'
David S. Miller [Sat, 14 May 2016 01:12:07 +0000 (21:12 -0400)]
Merge branch 'xgene-fixes'

Iyappan Subramanian says:

====================
drivers: net: xgene: Bug fixes

This patch set addresses the following bug fixes that were found during testing.

  1. IPv4 forward test crash
    - drivers: net: xgene: fix IPv4 forward crash

  2. Sharing of irqs
    - drivers: net: xgene: fix sharing of irqs

  3. Ununiform latency across queues
    - drivers: net: xgene: fix ununiform latency across queues

  4. Fix statistics counters race condition
    - drivers: net: xgene: fix statistics counters race condition

  5. Correcting register offset and field lengths
    - drivers: net: xgene: fix register offset

v2: Address review comments from v1
- Defer TSO fix, and reposting all other patches from v1

v1:
- Initial version
====================

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agodrivers: net: xgene: fix register offset
Iyappan Subramanian [Fri, 13 May 2016 23:53:01 +0000 (16:53 -0700)]
drivers: net: xgene: fix register offset

This patch fixes SG_RX_DV_GATE_REG_0_ADDR register offset
and ring state field lengths.

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Tested-by: Toan Le <toanle@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agodrivers: net: xgene: fix statistics counters race condition
Iyappan Subramanian [Fri, 13 May 2016 23:53:00 +0000 (16:53 -0700)]
drivers: net: xgene: fix statistics counters race condition

This patch fixes the race condition on updating the statistics
counters by moving the counters to the ring structure.

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Tested-by: Toan Le <toanle@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agodrivers: net: xgene: fix ununiform latency across queues
Iyappan Subramanian [Fri, 13 May 2016 23:52:59 +0000 (16:52 -0700)]
drivers: net: xgene: fix ununiform latency across queues

This patch addresses ununiform latency across queues by adding
more queues to match with, upto number of CPU cores.

Also, number of interrupts are increased and the channel numbers
are reordered.

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Tested-by: Toan Le <toanle@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agodrivers: net: xgene: fix sharing of irqs
Iyappan Subramanian [Fri, 13 May 2016 23:52:58 +0000 (16:52 -0700)]
drivers: net: xgene: fix sharing of irqs

Since hardware doesn't allow sharing of interrupts,
this patch fixes the same by removing IRQF_SHARED flag.

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Tested-by: Toan Le <toanle@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agodrivers: net: xgene: fix IPv4 forward crash
Iyappan Subramanian [Fri, 13 May 2016 23:52:57 +0000 (16:52 -0700)]
drivers: net: xgene: fix IPv4 forward crash

This patch fixes the crash observed during IPv4 forward test by
setting the drop field in the dbptr.

Signed-off-by: Iyappan Subramanian <isubramanian@apm.com>
Tested-by: Toan Le <toanle@apm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/next...
David S. Miller [Sat, 14 May 2016 01:08:24 +0000 (21:08 -0400)]
Merge branch '1GbE' of git://git./linux/kernel/git/jkirsher/next-queue

Jeff Kirsher says:

====================
1GbE Intel Wired LAN Driver Updates 2016-05-13

This series contains updates to e1000e, igb and igbvf.

Steve Shih fixes an issue for disabling auto-negotiation and forcing
speed and duplex settings for non-copper media.

Brian Walsh cleanups some inconsistency in the use of return variables
names to avoid confusion.

Jake cleans up the drivers to use the BIT() macro when it can, which will
future proof the drivers for GCC 6 when it gets released.  Cleaned up
dead code which was never being used.  Also fixed e1000e, where it was
incorrectly restting the SYSTIM registers every time the ioctl was being
run.

Denys Vlasenko fixes an oversight where incvalue variable holds a 32
bit value so we should declare it as such, instead of 64 bits.  Also
fixed an overflow check, where two reads are the same, then it is not
an overflow.

Nathan Sullivan fixes the PTP timestamps for transmit and receive
latency based on the current link speed.

Alexander Duyck adds support for partial GSO segmentation in the case
of tunnels for igb and igbvf.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge branch 'for-4.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj...
Linus Torvalds [Fri, 13 May 2016 23:26:46 +0000 (16:26 -0700)]
Merge branch 'for-4.6-fixes' of git://git./linux/kernel/git/tj/cgroup

Pull cgroup fixes from Tejun Heo:
 "During v4.6-rc1 cgroup namespace support was merged.  There is an
  issue where it's impossible to tell whether a given cgroup mount point
  is bind mounted or namespaced.  Serge has been working on the issue
  but it took longer than expected to resolve, so the late pull request.

  Given that it's a completely new feature and the patches don't touch
  anything else, the risk seems acceptable.  However, if this is too
  late, an alternative is plugging new cgroup ns creation for v4.6 and
  retrying for v4.7"

* 'for-4.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup:
  cgroup: fix compile warning
  kernfs: kernfs_sop_show_path: don't return 0 after seq_dentry call
  cgroup, kernfs: make mountinfo show properly scoped path for cgroup namespaces
  kernfs_path_from_node_locked: don't overwrite nlen

8 years agoMerge branch 'for-4.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq
Linus Torvalds [Fri, 13 May 2016 23:16:51 +0000 (16:16 -0700)]
Merge branch 'for-4.6-fixes' of git://git./linux/kernel/git/tj/wq

Pull workqueue fix from Tejun Heo:
 "CPU hotplug callbacks can invoke DOWN_FAILED w/o preceding
  DOWN_PREPARE which can trigger a WARN_ON() in workqueue.

  The bug has been there for a very long time.  It only triggers if CPU
  down fails at a specific point and I don't think it has adverse
  effects other than the warning messages.  The fix is very low impact"

* 'for-4.6-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq:
  workqueue: fix rebind bound workers warning

8 years agoe1000e: don't modify SYSTIM registers during SIOCSHWTSTAMP ioctl
Jacob Keller [Wed, 20 Apr 2016 18:36:42 +0000 (11:36 -0700)]
e1000e: don't modify SYSTIM registers during SIOCSHWTSTAMP ioctl

The e1000e_config_hwtstamp function was incorrectly resetting the SYSTIM
registers every time the ioctl was being run. If you happened to be
running ptp4l and lost the PTP connect (removing cable, or blocking the
UDP traffic for example), then ptp4l will eventually perform a restart
which involves re-requesting timestamp settings. In e1000e this has the
unfortunate and incorrect result of resetting SYSTIME to the kernel
time. Since kernel time is usually in UTC, and PTP time is in TAI, this
results in the leap second being re-applied.

Fix this by extracting the SYSTIME reset out into its own function,
e1000e_ptp_reset, which we call during reset to restore the hardware
registers. This function will (a) restart the timecounter based on the
new system time, (b) restore the previous PPB setting, and (c) restore
the previous hwtstamp settings.

In order to perform (b), I had to modify the adjfreq ptp function
pointer to store the old delta each time it is called. This also has the
side effect of restoring the correct base timinca register correctly.
The driver does not need to explicitly zero the ptp_delta variable since
the entire adapter structure comes zero-initialized.

Reported-by: Brian Walsh <brian@walsh.ws>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Brian Walsh <brian@walsh.ws>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoigb/igbvf: Add support for GSO partial
Alexander Duyck [Thu, 14 Apr 2016 21:19:38 +0000 (17:19 -0400)]
igb/igbvf: Add support for GSO partial

This patch adds support for partial GSO segmentation in the case of
tunnels.  Specifically with this change the driver an perform segmentation
as long as the frame either has IPv6 inner headers, or we are allowed to
mangle the IP IDs on the inner header.  This is needed because we will not
be modifying any fields from the start of the start of the outer transport
header to the start of the inner transport header as we are treating them
like they are just a block of IP options.

Signed-off-by: Alexander Duyck <aduyck@mirantis.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoe1000e: mark shifted values as unsigned
Jacob Keller [Wed, 13 Apr 2016 23:08:33 +0000 (16:08 -0700)]
e1000e: mark shifted values as unsigned

The E1000_ICH_NVM_SIG_MASK value is shifted, out to the 31st bit, which
is the signed bit for signed constants. Mark these values as unsigned to
prevent compiler warnings and issues on platforms which a different
signed bit implementation.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoe1000e: use BIT() macro for bit defines
Jacob Keller [Wed, 13 Apr 2016 23:08:32 +0000 (16:08 -0700)]
e1000e: use BIT() macro for bit defines

This prevents signed bitshift issues when the shift would overwrite the
signed bit, and prevents making this mistake in the future when copying
and modifying code.

Use GENMASK or the unsigned postfix for cases which aren't suitable for
BIT() macro.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoigbvf: use BIT() macro instead of shifts
Jacob Keller [Wed, 13 Apr 2016 23:08:31 +0000 (16:08 -0700)]
igbvf: use BIT() macro instead of shifts

To prevent signed bitshift issues, and improve code readability, use the
BIT() macro. Also make use of GENMASK or the unsigned postfix where this
is more appropriate than BIT()

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoigbvf: remove unused variable and dead code
Jacob Keller [Wed, 13 Apr 2016 23:08:30 +0000 (16:08 -0700)]
igbvf: remove unused variable and dead code

The variable rdlen is set but never used, and thus setting it is dead
code. Remove it.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoigb: adjust PTP timestamps for Tx/Rx latency
Nathan Sullivan [Tue, 3 May 2016 23:10:56 +0000 (18:10 -0500)]
igb: adjust PTP timestamps for Tx/Rx latency

Table 7-62 on page 338 of the i210 datasheet lists TX and RX latencies
for the various speeds the chip supports.  To give better PTP timestamp
accuracy, adjust the timestamps by the amounts Intel gives based on
current link speed.

Signed-off-by: Nathan Sullivan <nathan.sullivan@ni.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoe1000e: e1000e_cyclecounter_read(): do overflow check only if needed
Denys Vlasenko [Wed, 20 Apr 2016 15:45:56 +0000 (17:45 +0200)]
e1000e: e1000e_cyclecounter_read(): do overflow check only if needed

SYSTIMH:SYSTIML registers are incremented by 24-bit value TIMINCA[23..0]

er32(SYSTIML) are probably moderately expensive (they are pci bus reads).
Can we avoid one of them? Yes, we can.

If the SYSTIML value we see is smaller than 0xff000000, the overflow
into SYSTIMH would require at least two increments.

We do two reads, er32(SYSTIML) and er32(SYSTIMH), in this order.

Even if one increment happens between them, the overflow into SYSTIMH
is impossible, and we can avoid doing another er32(SYSTIML) read
and overflow check.

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoe1000e: e1000e_cyclecounter_read(): fix er32(SYSTIML) overflow check
Denys Vlasenko [Wed, 20 Apr 2016 15:45:55 +0000 (17:45 +0200)]
e1000e: e1000e_cyclecounter_read(): fix er32(SYSTIML) overflow check

If two consecutive reads of the counter are the same, it is also
not an overflow.  "systimel_1 < systimel_2" should be
"systimel_1 <= systimel_2".

Before the patch, we could perform an *erroneous* correction:

Let's say that systimel_1 == systimel_2 == 0xffffffff.
"systimel_1 < systimel_2" is false, we think it's an overflow,
we read "systimeh = er32(SYSTIMH)" which meanwhile had incremented,
and use "(systimeh << 32) + systimel_2" value which is 2^32 too large.

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
CC: intel-wired-lan@lists.osuosl.org
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoe1000e: e1000e_cyclecounter_read(): incvalue is 32 bits, not 64
Denys Vlasenko [Wed, 20 Apr 2016 15:45:54 +0000 (17:45 +0200)]
e1000e: e1000e_cyclecounter_read(): incvalue is 32 bits, not 64

"incvalue" variable holds a result of "er32(TIMINCA) &
E1000_TIMINCA_INCVALUE_MASK" and used in "do_div(temp, incvalue)"
as a divisor.

Thus, "u64 incvalue" declaration is probably a mistake.
Even though it seems to be a harmless one, let's fix it.

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoigb: make igb_update_pf_vlvf static
Jacob Keller [Wed, 13 Apr 2016 23:08:29 +0000 (16:08 -0700)]
igb: make igb_update_pf_vlvf static

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoigb: use BIT() macro or unsigned prefix
Jacob Keller [Wed, 13 Apr 2016 23:08:28 +0000 (16:08 -0700)]
igb: use BIT() macro or unsigned prefix

For bitshifts, we should make use of the BIT macro when possible, and
ensure that other bitshifts are marked as unsigned. This helps prevent
signed bitshift errors, and ensures similar style.

Make use of GENMASK and the unsigned postfix where BIT() isn't
appropriate.

Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoe1000e: Cleanup consistency in ret_val variable usage
Brian Walsh [Wed, 13 Apr 2016 03:22:30 +0000 (23:22 -0400)]
e1000e: Cleanup consistency in ret_val variable usage

Fixed the file to use a consistent ret_val for return value checking.

Signed-off-by: Brian Walsh <brian@walsh.ws>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoe1000e: fix ethtool autoneg off for non-copper
Steve Shih [Tue, 5 Apr 2016 18:30:03 +0000 (11:30 -0700)]
e1000e: fix ethtool autoneg off for non-copper

This patch fixes the issues for disabling auto-negotiation and forcing
speed and duplex settings for the non-copper media.

For non-copper media, e1000_get_settings should return ETH_TP_MDI_INVALID for
eth_tp_mdix_ctrl instead of ETH_TP_MDI_AUTO so subsequent e1000_set_settings
call would not fail with -EOPNOTSUPP.

e1000_set_spd_dplx should not automatically turn autoneg back on for forced
1000 Mbps full duplex settings for non-copper media.

Cc: xe-kernel@external.cisco.com
Cc: Daniel Walker <dwalker@fifo99.com>
Signed-off-by: Steve Shih <sshih@cisco.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
8 years agoMerge branch 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 13 May 2016 19:21:17 +0000 (12:21 -0700)]
Merge branch 'sched-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull scheduler fix from Ingo Molnar:
 "This is a revert to fix an interactivity problem.

  The proper fixes for the problems that the reverted commit exposed are
  now in sched/core (consisting of 3 patches), but were too risky for
  v4.6 and will arrive in the v4.7 merge window"

* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  Revert "sched/fair: Fix fairness issue on migration"

8 years agoMerge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 13 May 2016 18:54:02 +0000 (11:54 -0700)]
Merge branch 'perf-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull perf fixes from Ingo Molnar:
 "An uncharacteristically large number of bugs popped up in the last
  week:

   - various tooling fixes, two crashes and build problems
   - two Intel PT fixes
   - an KNL uncore driver fix
   - an Intel PMU driver fix"

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf stat: Fallback to user only counters when perf_event_paranoid > 1
  perf evsel: Handle EACCESS + perf_event_paranoid=2 in fallback()
  perf evsel: Improve EPERM error handling in open_strerror()
  tools lib traceevent: Do not reassign parg after collapse_tree()
  perf probe: Check if dwarf_getlocations() is available
  perf dwarf: Guard !x86_64 definitions under #ifdef else clause
  perf tools: Use readdir() instead of deprecated readdir_r()
  perf thread_map: Use readdir() instead of deprecated readdir_r()
  perf script: Use readdir() instead of deprecated readdir_r()
  perf tools: Use readdir() instead of deprecated readdir_r()
  perf/core: Disable the event on a truncated AUX record
  perf/x86/intel/pt: Generate PMI in the STOP region as well
  perf/x86: Fix undefined shift on 32-bit kernels
  perf/x86/msr: Fix SMI overflow
  perf/x86/intel/uncore: Fix CHA registers configuration procedure for Knights Landing platform
  perf diff: Fix duplicated output column

8 years agoMerge tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm...
Linus Torvalds [Fri, 13 May 2016 16:52:00 +0000 (09:52 -0700)]
Merge tag 'fixes-for-linus' of git://git./linux/kernel/git/arm/arm-soc

Pull ARM SoC fixes from Arnd Bergmann:
 "Three more bug fixes for ARM SoCs this week:

   - The Atmel sama5d2 was registering the wrong NFC device type

   - On Atmel sam9x5, the power management controller had an incorrect
     register area size

   - On ARM64 Allwinner machine was not secting the generic irqchip
     code, causing build errors in some configurations"

* tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc:
  ARM: dts: at91: sam9x5: Fix the memory range assigned to the PMC
  arm64/sunxi: 4.6-rc1: Add dependency on generic irq chip
  ARM: dts: at91: sama5d2: use "atmel,sama5d3-nfc" compatible for nfc