提交 · dad42c3038a59d27fced28ee4ec1d4a891b28155 · openeuler / Kernel

22 11月, 2016 8 次提交

mlx4: avoid unnecessary dirtying of critical fields · dad42c30

由 Eric Dumazet 提交于 11月 20, 2016

While stressing a 40Gbit mlx4 NIC with busy polling, I found false
sharing in mlx4 driver that can be easily avoided.

This patch brings an additional 7 % performance improvement in UDP_RR
workload.

1) If we received no frame during one mlx4_en_process_rx_cq()
   invocation, no need to call mlx4_cq_set_ci() and/or dirty ring->cons

2) Do not refill rx buffers if we have plenty of them.
   This avoids false sharing and allows some bulk/batch optimizations.
   Page allocator and its locks will thank us.

Finally, mlx4_en_poll_rx_cq() should not return 0 if it determined
cpu handling NIC IRQ should be changed. We should return budget-1
instead, to not fool net_rx_action() and its netdev_budget.

v2: keep AVG_PERF_COUNTER(... polled) even if polled is 0
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Tariq Toukan <tariqt@mellanox.com>
Reviewed-by: NTariq Toukan <tariqt@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

dad42c30

bnx2: use READ_ONCE() instead of barrier() · b668534c

由 Eric Dumazet 提交于 11月 19, 2016

barrier() is a big hammer compared to READ_ONCE(),
and requires comments explaining what is protected.

READ_ONCE() is more precise and compiler should generate
better overall code.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

b668534c

udp: avoid one cache line miss in recvmsg() · d21dbdfe

由 Eric Dumazet 提交于 11月 18, 2016

UDP_SKB_CB(skb)->partial_cov is located at offset 66 in skb,
requesting a cold cache line being read in cpu cache.

We can avoid this cache line miss for UDP sockets,
as partial_cov has a meaning only for UDPLite.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

d21dbdfe

Merge branch 'mlx5-bpf-refcnt-fixes' · ee9d5461

由 David S. Miller 提交于 11月 21, 2016

Daniel Borkmann says:

====================
Couple of BPF refcount fixes for mlx5

Various mlx5 bugs on eBPF refcount handling found during review.
Last patch in series adds a __must_check to BPF helpers to make
sure we won't run into it again w/o compiler complaining first.

v2 -> v3:

 - Just reworked patch 2/4 so we don't need bpf_prog_sub().
 - Rebased, rest as is.

v1 -> v2:

 - After discussion with Alexei, we agreed upon rebasing the
   patches against net-next.
 - Since net-next, I've also added the __must_check to enforce
   future users to check for errors.
 - Fixed up commit message #2.
 - Simplify assignment from patch #1 based on Saeed's feedback
   on previous set.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

ee9d5461

bpf: add __must_check attributes to refcount manipulating helpers · 6d67942d

由 Daniel Borkmann 提交于 11月 19, 2016

Helpers like bpf_prog_add(), bpf_prog_inc(), bpf_map_inc() can fail
with an error, so make sure the caller properly checks their return
value and not just ignores it, which could worst-case lead to use
after free.
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>
Acked-by: NAlexei Starovoitov <ast@kernel.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

6d67942d

bpf, mlx5: drop priv->xdp_prog reference on netdev cleanup · a055c19b

由 Daniel Borkmann 提交于 11月 19, 2016

mlx5e_xdp_set() is currently the only place where we drop reference on the
prog sitting in priv->xdp_prog when it's exchanged by a new one. We also
need to make sure that we eventually release that reference, for example,
in case the netdev is dismantled, otherwise we leak the program.

Fixes: 86994156 ("net/mlx5e: XDP fast RX drop bpf programs support")
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>
Acked-by: NSaeed Mahameed <saeedm@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

a055c19b

bpf, mlx5: fix various refcount issues in mlx5e_xdp_set · c54c0629

由 Daniel Borkmann 提交于 11月 19, 2016

There are multiple issues in mlx5e_xdp_set():

1) The batched bpf_prog_add() is currently not checked for errors. When
doing so, it should be done at an earlier point in time to makes sure
that we cannot fail anymore at the time we want to set the program for
each channel. The batched refs short-cut can only be performed when we
don't need to perform a reset for changing the rq type and the device
was in opened state. In case the device was not in opened state, then
the next mlx5e_open_locked() will aquire the refs from the control prog
via mlx5e_create_rq(), same when we need to perform a reset.

2) When swapping the priv->xdp_prog, then no extra reference count must be
taken since we got that from call path via dev_change_xdp_fd() already.
Otherwise, we'd never be able to release the program. Also, bpf_prog_add()
without checking the return code could fail.

c54c0629

bpf, mlx5: fix mlx5e_create_rq taking reference on prog · 97bc402d

由 Daniel Borkmann 提交于 11月 19, 2016

In mlx5e_create_rq(), when creating a new queue, we call bpf_prog_add() but
without checking the return value. bpf_prog_add() can fail since 92117d84
("bpf: fix refcnt overflow"), so we really must check it. Take the reference
right when we assign it to the rq from priv->xdp_prog, and just drop the
reference on error path. Destruction in mlx5e_destroy_rq() looks good, though.

Fixes: 86994156 ("net/mlx5e: XDP fast RX drop bpf programs support")
Signed-off-by: NDaniel Borkmann <daniel@iogearbox.net>
Acked-by: NSaeed Mahameed <saeedm@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

97bc402d

21 11月, 2016 7 次提交

Merge branch 'mV88e6xxx-interrupt-fixes' · f81a8a02

由 David S. Miller 提交于 11月 20, 2016

Andrew Lunn says:

====================
Fixes for the MV88e6xxx interrupt code

The interrupt code was never tested with a board who's probing
resulted in an -EPROBE_DEFFERED. So the clean up paths never got
tested. I now do have -EPROBE_DEFFERED, and things break badly during
cleanup. These are the fixes.

This is fixing code in net-next.

v2:
Fix typo pointed out by David Miller
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

f81a8a02

net: dsa: mv88e6xxx: Hold the mutex while freeing g1 interrupts · 61f7c3f8