sh_eth_tsu_soft_swap() is called twice by the driver, remove *inline* and
move that function from the header to the driver itself to let gcc decide
whether to expand it inline or not...
Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
Browsing thru the driver disassembly, I noticed that ARM gcc generated
no code whatsoever for sh_eth_soft_swap() while building a little-endian
kernel -- apparently __LITTLE_ENDIAN__ was not being #define'd, however
it got implicitly #define'd when building with the SH gcc (I could only
find the explicit #define __LITTLE_ENDIAN that was #include'd when building
a little-endian kernel). Luckily, the Ether controller only doing big-
endian DMA is encountered on the early SH771x SoCs only and all ARM SoCs
implement EDMR.DE and thus set 'sh_eth_cpu_data::hw_swap'. But anyway, we
need to fix the #ifdef inside sh_eth_soft_swap() to something that would
work on all architectures...
Signed-off-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull dma-mapping updates from Christoph Hellwig:
- replace the force_dma flag with a dma_configure bus method. (Nipun
Gupta, although one patch is іncorrectly attributed to me due to a
git rebase bug)
- use GFP_DMA32 more agressively in dma-direct. (Takashi Iwai)
- remove PCI_DMA_BUS_IS_PHYS and rely on the dma-mapping API to do the
right thing for bounce buffering.
- move dma-debug initialization to common code, and apply a few
cleanups to the dma-debug code.
- cleanup the Kconfig mess around swiotlb selection
- swiotlb comment fixup (Yisheng Xie)
- a trivial swiotlb fix. (Dan Carpenter)
- support swiotlb on RISC-V. (based on a patch from Palmer Dabbelt)
- add a new generic dma-noncoherent dma_map_ops implementation and use
it for arc, c6x and nds32.
- improve scatterlist validity checking in dma-debug. (Robin Murphy)
- add a struct device quirk to limit the dma-mask to 32-bit due to
bridge/system issues, and switch x86 to use it instead of a local
hack for VIA bridges.
- handle devices without a dma_mask more gracefully in the dma-direct
code.
* tag 'dma-mapping-4.18' of git://git.infradead.org/users/hch/dma-mapping: (48 commits)
dma-direct: don't crash on device without dma_mask
nds32: use generic dma_noncoherent_ops
nds32: implement the unmap_sg DMA operation
nds32: consolidate DMA cache maintainance routines
x86/pci-dma: switch the VIA 32-bit DMA quirk to use the struct device flag
x86/pci-dma: remove the explicit nodac and allowdac option
x86/pci-dma: remove the experimental forcesac boot option
Documentation/x86: remove a stray reference to pci-nommu.c
core, dma-direct: add a flag 32-bit dma limits
dma-mapping: remove unused gfp_t parameter to arch_dma_alloc_attrs
dma-debug: check scatterlist segments
c6x: use generic dma_noncoherent_ops
arc: use generic dma_noncoherent_ops
arc: fix arc_dma_{map,unmap}_page
arc: fix arc_dma_sync_sg_for_{cpu,device}
arc: simplify arc_dma_sync_single_for_{cpu,device}
dma-mapping: provide a generic dma-noncoherent implementation
dma-mapping: simplify Kconfig dependencies
riscv: add swiotlb support
riscv: only enable ZONE_DMA32 for 64-bit
...
Fix up a cast problem introduced by a sparse cleanup patch. This fixes
a problem where the encrypted packets were not recognized on Rx and
subsequently dropped.
Fixes: 9cfbfa701b ("ixgbe: cleanup sparse warnings")
Signed-off-by: Shannon Nelson <shannon.nelson@oracle.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Make sure we don't try to offload the decryption of an incoming
packet that should get delivered to the management engine. This
is a corner case that will likely be very seldom seen, but could
really confuse someone if they were to hit it.
Suggested-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Shannon Nelson <shannon.nelson@oracle.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
When offloading mirroring to gretap or ip6gretap netdevices, an 802.1q
bridge is one of the soft devices permissible in the underlay when
resolving the packet path. After the packet path is resolved to a
particular bridge egress device, flags on packet VLAN determine whether
the egressed packet should be tagged.
The current logic however only ever sets the VLAN tag, never suppresses
it. Thus if there's a VLAN netdevice above the bridge that determines
the packet VLAN, that VLAN is never unset, and mirroring is configured
with VLAN tagging.
Fix by setting the packet VLAN on both branches: set to zero (for unset)
when BRIDGE_VLAN_INFO_UNTAGGED, copy the resolved VLAN (e.g. from bridge
PVID) otherwise.
Fixes: 946a11e740 ("mlxsw: spectrum_span: Allow bridge for gretap mirror")
Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
VLAN deletion notifications are emitted before the relevant change is
projected to bridge configuration. Thus, like with VLAN addition,
schedule SPAN respin for later.
Fixes: c520bc6986 ("mlxsw: Respin SPAN on switchdev events")
Signed-off-by: Petr Machata <petrm@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Similar to ixgbevf, the same possibility for race exists. Extend the RTNL
lock in ixgbe_reset_subtask() to protect the state bits; this is to make
sure that we get the most up-to-date values for the bits and avoid a
possible race when going down.
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Add support for XDP meta data when using build skb variant of
the i40e driver. Implementation is analogous to the existing
ixgbe and ixgbevf support for meta data from 366a88fe2f ("bpf,
ixgbe: add meta data support") and be8333322e ("ixgbevf: Add
support for meta data"). With the build skb variant we get
192 bytes of extra headroom which can be used for encaps or
meta data.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Tested-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
ixgbe_dbg_reg_ops_read and ixgbe_dbg_netdev_ops_read copy-pasting
the same code except for ixgbe_dbg_netdev_ops_buf/ixgbe_dbg_reg_ops_buf,
so introduce a helper ixgbe_dbg_common_ops_read to remove redundant code.
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Extend the RTNL lock in ixgbevf_reset_subtask() to protect the state bits
check in addition to the call to ixgbevf_reinit_locked().
This is to make sure that we get the most up-to-date values for the bits
and avoid a possible race when going down.
Suggested-by: Zhiping du <zhipingdu@tencent.com>
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
In the case of the VF driver it is supposed to provide a context descriptor
that allows us to provide information about the header offsets inside of
the frame. However in the case of XDP we don't really have any of that
information since the data is minimally processed. As a result we were
seeing malicious driver detection (MDD) events being triggered when the PF
had that functionality enabled.
To address this I have added a bit of new code that will "prime" the XDP
ring by providing one context descriptor that assumes the minimal setup of
an Ethernet frame which is an L2 header length of 14. With just that we can
provide enough information to make the hardware happy so that we don't
trigger MDD events.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Move 10ms sleep out of function resetting TX queue.
Reset all the TX queues in one turn and
wait for all of them just once.
Use usleep_range() instead of mdelay() in order not to
affect transmission on other interfaces.
Signed-off-by: Sergey Nemov <sergey.nemov@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Issuing "ip link set up/down" can block TSICR interrupts, what results in
missing PTP Tx timestamp and no PPS pulse generation.
Problem happens when the link is set up with the TSICR interrupts pending.
ICR is cleared before enabling interrupts, while TSICR is not. When all TSICR
interrupts are pending at this moment, time_sync interrupt will never
be generated. TSICR should be cleared as well.
In order to reproduce the issue:
1. Setup linux with IEEE 1588 grandmaster and PPS output enabled
2. Continue setting link up/down with random intervals between commands
3. Wait until PPS is not generated ( only one pulse is generated and PPS
dies), and ptp4l complains constantly about Tx timeout.
Signed-off-by: Joanna Yurdal <jyu@trackman.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
There have been multiple reports of crashes that look like
kernel: RIP: 0010:[<ffffffff8110303f>] timecounter_read+0xf/0x50
[...]
kernel: Call Trace:
kernel: [<ffffffffa0806b0f>] e1000e_phc_gettime+0x2f/0x60 [e1000e]
kernel: [<ffffffffa0806c5d>] e1000e_systim_overflow_work+0x1d/0x80 [e1000e]
kernel: [<ffffffff810992c5>] process_one_work+0x155/0x440
kernel: [<ffffffff81099e16>] worker_thread+0x116/0x4b0
kernel: [<ffffffff8109f422>] kthread+0xd2/0xf0
kernel: [<ffffffff8163184f>] ret_from_fork+0x3f/0x70
These can be traced back to the fact that e1000e_systim_reset() skips the
timecounter_init() call if e1000e_get_base_timinca() returns -EINVAL, which
leads to a null deref in timecounter_read().
Commit 83129b37ef ("e1000e: fix systim issues", v4.2-rc1) reworked
e1000e_get_base_timinca() in such a way that it can return -EINVAL for
e1000_pch_spt if the SYSCFI bit is not set in TSYNCRXCTL.
Some experimentation has shown that on I219 (e1000_pch_spt, "MAC: 12")
adapters, the E1000_TSYNCRXCTL_SYSCFI flag is unstable; TSYNCRXCTL reads
sometimes don't have the SYSCFI bit set. Retrying the read shortly after
finds the bit to be set. This was observed at boot (probe) but also link up
and link down.
Moreover, the phc (PTP Hardware Clock) seems to operate normally even after
reads where SYSCFI=0. Therefore, remove this register read and
unconditionally set the clock parameters.
Reported-by: Achim Mildenberger <admin@fph.physik.uni-karlsruhe.de>
Message-Id: <20180425065243.g5mqewg5irkwgwgv@f2>
Bugzilla: https://bugzilla.suse.com/show_bug.cgi?id=1075876
Fixes: 83129b37ef ("e1000e: fix systim issues")
Signed-off-by: Benjamin Poirier <bpoirier@suse.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The function aq_fw2x_get_mac_permanent is local to the source and does
not need to be in global scope, so make it static.
Cleans up sparse warning:
warning: symbol 'aq_fw2x_get_mac_permanent' was not declared. Should it
be static?
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull verbs counters series from Leon Romanovsky:
====================
Verbs flow counters support
This series comes to allow user space applications to monitor real time
traffic activity and events of the verbs objects it manages, e.g.: ibv_qp,
ibv_wq, ibv_flow.
The API enables generic counters creation and define mapping to
association with a verbs object, the current mlx5 driver is using this API
for flow counters.
With this API, an application can monitor the entire life cycle of object
activity, defined here as a static counters attachment. This API also
allows dynamic counters monitoring of measurement points for a partial
period in the verbs object life cycle.
In addition it presents the implementation of the generic counters
interface.
This will be achieved by extending flow creation by adding a new flow
count specification type which allows the user to associate a previously
created flow counters using the generic verbs counters interface to the
created flow, once associated the user could read statistics by using the
read function of the generic counters interface.
The API includes:
1. create and destroyed API of a new counters objects
2. read the counters values from HW
Note:
Attaching API to allow application to define the measurement points per
objects is a user space only API and this data is passed to kernel when
the counted object (e.g. flow) is created with the counters object.
===================
* tag 'verbs_flow_counters':
IB/mlx5: Add counters read support
IB/mlx5: Add flow counters read support
IB/mlx5: Add flow counters binding support
IB/mlx5: Add counters create and destroy support
IB/uverbs: Add support for flow counters
IB/core: Add support for flow counters
IB/core: Support passing uhw for create_flow
IB/uverbs: Add read counters support
IB/core: Introduce counters read verb
IB/uverbs: Add create/destroy counters support
IB/core: Introduce counters object and its create/destroy
IB/uverbs: Add an ib_uobject getter to ioctl() infrastructure
net/mlx5: Export flow counter related API
net/mlx5: Use flow counter pointer as input to the query function
This patch fixes the clang warning of extraneous parentheses, with the
following coccinelle script.
@@
identifier i;
expression e;
statement s;
@@
if (
-(i == e)
+i == e
)
s
Suggested-by: Lukas Bulwahn <lukas.bulwahn@gmail.com>
Signed-off-by: Varsha Rao <rvarsha016@gmail.com>
Acked-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This adds support for Flexible PPS output (which is equivalent
to per_out output of PTP subsystem).
Tested using an oscilloscope and the following commands:
1) Start PTP4L:
# ptp4l -A -4 -H -m -i eth0 &
2) Set Flexible PPS frequency:
# echo <idx> <ts> <tns> <ps> <pns> > /sys/class/ptp/ptpX/period
Where, ts/tns is start time and ps/pns is period time, and ptpX is ptp
of eth0.
Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Joao Pinto <jpinto@synopsys.com>
Cc: Vitor Soares <soares@synopsys.com>
Cc: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Cc: Alexandre Torgue <alexandre.torgue@st.com>
Cc: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Incorrect shared memory address is used while deriving the values
for tc and pri_type. Use shmem address corresponding to 'oem_cfg_func'
where the management firmare saves tc/pri_type values.
Fixes: cac6f691 ("qed: Add support for Unified Fabric Port")
Signed-off-by: Sudarsana Reddy Kalluru <Sudarsana.Kalluru@cavium.com>
Signed-off-by: Ariel Elior <ariel.elior@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The structure shared between driver and management firmware (MFW)
differ in sizes. The additional field defined by the MFW is not
relevant to the current driver. Add a dummy field to the structure.
Fixes: cac6f691 ("qed: Add support for Unified Fabric Port")
Signed-off-by: Sudarsana Reddy Kalluru <Sudarsana.Kalluru@cavium.com>
Signed-off-by: Ariel Elior <ariel.elior@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When passed the XDP_XMIT_FLUSH flag virtnet_xdp_xmit now performs the
same virtqueue_kick as virtnet_xdp_flush.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When passed the XDP_XMIT_FLUSH flag tun_xdp_xmit now performs the same
kind of socket wake up as in tun_xdp_flush(). The wake up code from
tun_xdp_flush is generalized and shared with tun_xdp_xmit.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When passed the XDP_XMIT_FLUSH flag ixgbe_xdp_xmit now performs the
same kind of ring tail update as in ixgbe_xdp_flush. The update tail
code in ixgbe_xdp_flush is generalized and shared with ixgbe_xdp_xmit.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
When passed the XDP_XMIT_FLUSH flag i40e_xdp_xmit now performs the
same kind of ring tail update as in i40e_xdp_flush. The advantage is
that all the necessary checks have been performed and xdp_ring can be
updated, instead of having to perform the exact same steps/checks in
i40e_xdp_flush
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
This patch only change the API and reject any use of flags. This is an
intermediate step that allows us to implement the flush flag operation
later, for each individual driver in a separate patch.
The plan is to implement flush operation via XDP_XMIT_FLUSH flag
and then remove XDP_XMIT_FLAGS_NONE when done.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Kalle Valo says:
====================
wireless-drivers-next patches for 4.18
Hopefully the last pull request to 4.18 before the merge window.
Nothing major here, we have smaller new features and of course a lots
of fixes.
Major changes:
ath10k
* add memory dump support for QCA9888 and QCA99X0
* add support to configure channel dwell time
* support new DFS host confirmation feature in the firmware
ath
* update various regulatory mappings
wcn36xx
* various fixes to improve reliability
* add Factory Test Mode support
brmfmac
* add debugfs file for reading firmware capabilities
mwifiex
* support sysfs initiated device coredump
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 1383cb8103 ("mlx4_core: allocate ICM memory in page size chunks")
brought two regressions caught in our regression suite.
The big one is an additional cost of 256 bytes of overhead per 4096 bytes,
or 6.25 % which is unacceptable since ICM can be pretty large.
This comes from having to allocate one struct mlx4_icm_chunk (256 bytes)
per MLX4_TABLE_CHUNK, which the buggy commit shrank to 4KB
(instead of prior 256KB)
Note that mlx4_alloc_icm() is already able to try high order allocations
and fallback to low-order allocations under high memory pressure.
Most of these allocations happen right after boot time, when we get
plenty of non fragmented memory, there is really no point being so
pessimistic and break huge pages into order-0 ones just for fun.
We only have to tweak gfp_mask a bit, to help falling back faster,
without risking OOM killings.
Second regression is an KASAN fault, that will need further investigations.
Fixes: 1383cb8103 ("mlx4_core: allocate ICM memory in page size chunks")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Tariq Toukan <tariqt@mellanox.com>
Cc: John Sperbeck <jsperbeck@google.com>
Cc: Tarick Bedeir <tarick@google.com>
Cc: Qing Huang <qing.huang@oracle.com>
Cc: Daniel Jurgens <danielj@mellanox.com>
Cc: Zhu Yanjun <yanjun.zhu@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
axienet_open no longer return -ENODEV when PHY cannot be connected to
since commit d7cc3163e0 ("net: axienet: Support phy-less mode of operation")
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fixes the following sparse warning:
drivers/net/ethernet/mellanox/mlx5/core/fpga/tls.c:199:6: warning:
symbol 'mlx5_fpga_tls_send_teardown_cmd' was not declared. Should it be static?
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix to return a negative error code from the failover register fail
error handling case instead of 0, as done elsewhere in this function.
Fixes: 1ff78076d8 ("netvsc: refactor notifier/event handling code to use the failover framework")
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Filling in the padding slot in the bpf structure as a bug fix in 'ne'
overlapped with actually using that padding area for something in
'net-next'.
Signed-off-by: David S. Miller <davem@davemloft.net>
We can bail out immediately also in case of PHY_IGNORE_INTERRUPT because
phy_mac_interupt() informs us once the link is up.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This allows to un-expose the details of struct mlx5_fc and keep it
internal to the core driver.
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Avoid false sharing of cachelines by separating the cachelines of
TX stats that are dertied in xmit flow and in completion flow.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Prefer the linear SKB configuration of Legacy RQ over the
non-linear one of Striding RQ.
This implies that ConnectX-4 LX now uses legacy RQ by default,
as it does not support the linear configuration of Striding RQ.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Enhance the memory scheme of the legacy RQ, such that
only order-0 pages are used.
Whenever possible, prefer using a linear SKB, and build it
wrapping the WQE buffer.
Otherwise (for example, jumbo frames on x86), use non-linear SKB,
with as many frags as needed. In this case, multiple WQE
scatter entries are used, up to a maximum of 4 frags and 10KB of MTU.
This implied to remove support of HW LRO in legacy RQ, as it would
require large number of page allocations and scatter entries per WQE
on archs with PAGE_SIZE = 4KB, yielding bad performance.
In earlier patches, we guaranteed that all completions are in-order,
and that we use a cyclic WQ.
This creates an oppurtunity for a performance optimization:
The mapping between a "struct mlx5e_dma_info", and the
WQEs (struct mlx5e_wqe_frag_info) pointing to it, is constant
across different cycles of a WQ. This allows initializing
the mapping in the time of RQ creation, and not handle it
in datapath.
A struct mlx5e_dma_info that is shared between different WQEs
is allocated by the first WQE, and freed by the last one.
This implies an important requirement: WQEs that share the same
struct mlx5e_dma_info must be posted within the same NAPI.
Otherwise, upon completion, struct mlx5e_wqe_frag_info would mistakenly
point to the new struct mlx5e_dma_info, not the one that was posted
(and the HW wrote to).
This bulking requirement is actually good also for performance reasons,
hence we extend the bulk beyong the minimal requirement above.
With this memory scheme, the RQs memory footprint is reduce by a
factor of 2 on x86, and by a factor of 32 on PowerPC.
Same factors apply for the number of pages in a GRO session.
Performance tests:
ConnectX-4, single core, single RX ring, default MTU.
x86:
CPU: Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz
Packet rate (early drop in TC): no degradation
TCP streams: ~5% improvement
PowerPC:
CPU: POWER8 (raw), altivec supported
Packet rate (early drop in TC): 20% gain
TCP streams: 25% gain
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Now that LRO is not supported for Legacy RQ, there is no source of
out-of-order completions in the WQ, and we can use a cyclic one.
This has multiple advantages:
- reduces the WQE size (smaller PCI transactions).
- lower overhead in datapath (no handling of 'next' pointers).
- no reserved WQE for the WQ head (was need in linked-list).
- allows using a constant map between frag and dma_info struct, in downstream patch.
Performance tests:
ConnectX-4, single core, single RX ring.
Major gain in packet rate of single ring XDP drop.
Bottleneck is shifted form HW (at 16Mpps) to SW (at 20Mpps).
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Replace the common RQ WQ object with two separate ones for the
different RQ types.
This is in preparation for switching to using a cyclic WQ type
in Legacy RQ.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Current LRO implementation in Legacy RQ uses high-order pages.
In downstream patches of this series we complete the transition
to using only order-0 pages in RX datapath (which was already done
in Striding RQ).
Unlike the more advanced Striding RQ, Legacy RQ does not make reuse
of any non-consumed buffers of non-full LRO sessions, and combining
it with order-0 pages has many performance drawbacks.
Hence, here we totally remove LRO support in Legacy RQ.
This guarantees having no out-of-order completions, which allows using
a cyclic work queue (instead of a linked-list) in a downstream patch.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Get the logic of copying the packet header into the SKB linear part
into a generic function. Function does copy length alignment
and dma buffer sync.
It is currently called only within the MPWQE flow.
In a downstream patch, it will be called within the legacy RQ flow
as well.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Rename it and pass truesize as an extra argument, as it will be used also
in Legacy RQ in a downstream patch.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Make name more generic by dropping MPWRQ from it, as it will be
used also in Legacy RQ in a downstream patch.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Instead of maintaining a local copy of skb->len/data and updating
it upon every copy to the WQE inline part, just calculate it once
when needed, using the ihs.
This obsoletes the function mlx5e_tx_skb_pull_inline.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Add handlers for this event to perform graceful teardown of the device.
Signed-off-by: Ilan Tayari <ilant@mellanox.com>
Signed-off-by: Adi Nissim <adin@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
