Intel I210 on some Intel Alder Lake platforms can only achieve ~750Mbps
Tx speed via iperf. The RR2DCDELAY shows around 0x2xxx DMA delay, which
will be significantly lower when 1) ASPM is disabled or 2) SoC package
c-state stays above PC3. When the RR2DCDELAY is around 0x1xxx the Tx
speed can reach to ~950Mbps.
According to the I210 datasheet "8.26.1 PCIe Misc. Register - PCIEMISC",
"DMA Idle Indication" doesn't seem to tie to DMA coalesce anymore, so
set it to 1b for "DMA is considered idle when there is no Rx or Tx AND
when there are no TLPs indicating that CPU is active detected on the
PCIe link (such as the host executes CSR or Configuration register read
or write operation)" and performing Tx should also fall under "active
CPU on PCIe link" case.
In addition to that, commit b6e0c419f0 ("igb: Move DMA Coalescing init
code to separate function.") seems to wrongly changed from enabling
E1000_PCIEMISC_LX_DECISION to disabling it, also fix that.
Fixes: b6e0c419f0 ("igb: Move DMA Coalescing init code to separate function.")
Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Tested-by: Gurucharan <gurucharanx.g@intel.com> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Link: https://lore.kernel.org/r/20220621221056.604304-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
igb_read_phy_reg() will silently return, leaving phy_data untouched, if
hw->ops.read_reg isn't set. Depending on the uninitialized value of
phy_data, this led to the phy status check either succeeding immediately
or looping continuously for 2 seconds before emitting a noisy err-level
timeout. This message went out to the console even though there was no
actual problem.
Instead, first check if there is read_reg function pointer. If not,
proceed without trying to check the phy status register.
Fixes: b72f3f7200 ("igb: When GbE link up, wait for Remote receiver status condition")
Signed-off-by: Kevin Mitchell <kevmitch@arista.com>
Tested-by: Gurucharan <gurucharanx.g@intel.com> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
On changing the RX ring parameters igb uses a hack to avoid a warning
when calling xdp_rxq_info_reg via igb_setup_rx_resources. It just
clears the struct xdp_rxq_info content.
Instead, change this to unregister if we're already registered. Align
code to the igc code.
Fixes: 9cbc948b5a ("igb: add XDP support")
Signed-off-by: Corinna Vinschen <vinschen@redhat.com>
Acked-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Tested-by: Sandeep Penigalapati <sandeep.penigalapati@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
As stated in [1], dma_set_mask() with a 64-bit mask never fails if
dev->dma_mask is non-NULL.
So, if it fails, the 32 bits case will also fail for the same reason.
So, if dma_set_mask_and_coherent() succeeds, 'pci_using_dac' is known to be
1.
Simplify code and remove some dead code accordingly.
[1]: https://lkml.org/lkml/2021/6/7/398
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Alexei Starovoitov says:
====================
pull-request: bpf-next 2021-12-30
The following pull-request contains BPF updates for your *net-next* tree.
We've added 72 non-merge commits during the last 20 day(s) which contain
a total of 223 files changed, 3510 insertions(+), 1591 deletions(-).
The main changes are:
1) Automatic setrlimit in libbpf when bpf is memcg's in the kernel, from Andrii.
2) Beautify and de-verbose verifier logs, from Christy.
3) Composable verifier types, from Hao.
4) bpf_strncmp helper, from Hou.
5) bpf.h header dependency cleanup, from Jakub.
6) get_func_[arg|ret|arg_cnt] helpers, from Jiri.
7) Sleepable local storage, from KP.
8) Extend kfunc with PTR_TO_CTX, PTR_TO_MEM argument support, from Kumar.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Support for the PTP pin function on 82580/i354/i350 based adapters.
Because the time registers of these adapters do not have the nice split in
second rollovers as the i210 has, the implementation is slightly more
complex compared to the i210 implementation.
Signed-off-by: Ruud Bos <kernel.hbk@gmail.com>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Support for the PEROUT PTP pin function on 82580/i354/i350 based adapters.
Because the time registers of these adapters do not have the nice split in
second rollovers as the i210 has, the implementation is slightly more
complex compared to the i210 implementation.
Signed-off-by: Ruud Bos <kernel.hbk@gmail.com>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Remove code duplication in the tsync interrupt handler function by moving
this logic to separate functions. This keeps the interrupt handler readable
and allows the new functions to be extended for adapter types other than
i210.
Signed-off-by: Ruud Bos <kernel.hbk@gmail.com>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
napi_build_skb() reuses per-cpu NAPI skbuff_head cache in order
to save some cycles on freeing/allocating skbuff_heads on every
new Rx or completed Tx.
igb driver runs Tx completion polling cycle right before the Rx
one and uses napi_consume_skb() to feed the cache with skbuff_heads
of completed entries, so it's never empty and always warm at that
moment. Switch to the napi_build_skb() to relax mm pressure on
heavy Rx.
Signed-off-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com>
Tested-by: Gurucharan G <gurucharanx.g@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Recent net core changes caused an issue with few Intel drivers
(reportedly igb), where taking RTNL in RPM resume path results in a
deadlock. See [0] for a bug report. I don't think the core changes
are wrong, but taking RTNL in RPM resume path isn't needed.
The Intel drivers are the only ones doing this. See [1] for a
discussion on the issue. Following patch changes the RPM resume path
to not take RTNL.
[0] https://bugzilla.kernel.org/show_bug.cgi?id=215129
[1] https://lore.kernel.org/netdev/20211125074949.5f897431@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com/t/
Fixes: bd869245a3 ("net: core: try to runtime-resume detached device in __dev_open")
Fixes: f32a213765 ("ethtool: runtime-resume netdev parent before ethtool ioctl ops")
Tested-by: Martin Stolpe <martin.stolpe@gmail.com>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Link: https://lore.kernel.org/r/20211220201844.2714498-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Move checking condition of VF MAC filter before clearing
or adding MAC filter to VF to prevent potential blackout caused
by removal of necessary and working VF's MAC filter.
Fixes: 1b8b062a99 ("igb: add VF trust infrastructure")
Signed-off-by: Karen Sornek <karen.sornek@intel.com>
Tested-by: Konrad Jankowski <konrad0.jankowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
In non trivial scenarios, the action id alone is not sufficient to
identify the program causing the warning. Before the previous patch,
the generated stack-trace pointed out at least the involved device
driver.
Let's additionally include the program name and id, and the relevant
device name.
If the user needs additional infos, he can fetch them via a kernel
probe, leveraging the arguments added here.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://lore.kernel.org/bpf/ddb96bb975cbfddb1546cf5da60e77d5100b533c.1638189075.git.pabeni@redhat.com
Oleksandr brought a bug report where netpoll causes trace
messages in the log on igb.
Danielle brought this back up as still occurring, so we'll try
again.
[22038.710800] ------------[ cut here ]------------
[22038.710801] igb_poll+0x0/0x1440 [igb] exceeded budget in poll
[22038.710802] WARNING: CPU: 12 PID: 40362 at net/core/netpoll.c:155 netpoll_poll_dev+0x18a/0x1a0
As Alex suggested, change the driver to return work_done at the
exit of napi_poll, which should be safe to do in this driver
because it is not polling multiple queues in this single napi
context (multiple queues attached to one MSI-X vector). Several
other drivers contain the same simple sequence, so I hope
this will not create new problems.
Fixes: 16eb8815c2 ("igb: Refactor clean_rx_irq to reduce overhead and improve performance")
Reported-by: Oleksandr Natalenko <oleksandr@natalenko.name>
Reported-by: Danielle Ratson <danieller@nvidia.com>
Suggested-by: Alexander Duyck <alexander.duyck@gmail.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name>
Tested-by: Danielle Ratson <danieller@nvidia.com>
Link: https://lore.kernel.org/r/20211123204000.1597971-1-jesse.brandeburg@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In following patches, dev_watchdog() will no longer stop all queues.
It will read queue->trans_start locklessly.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The driver tried to use Linux' native software I2C bus master
(i2c-algo-bits) for exporting the I2C interface that talks to the SFP
cage(s) towards userspace. As-is, however, the physical SCL/SDA pins
were not moving at all, staying at logical 1 all the time.
The main culprit was the I2CPARAMS register where igb was not setting
the I2CBB_EN bit. That meant that all the careful signal bit-banging was
actually not being propagated to the chip pads (I verified this with a
scope).
The bit-banging was not correct either, because I2C is supposed to be an
open-collector bus, and the code was driving both lines via a totem
pole. The code was also trying to do operations which did not make any
sense with the i2c-algo-bits, namely manipulating both SDA and SCL from
igb_set_i2c_data (which is only supposed to set SDA). I'm not sure if
that was meant as an optimization, or was just flat out wrong, but given
that the i2c-algo-bits is set up to work with a totally dumb GPIO-ish
implementation underneath, there's no need for this code to be smart.
The open-drain vs. totem-pole is fixed by the usual trick where the
logical zero is implemented via regular output mode and outputting a
logical 0, and the logical high is implemented via the IO pad configured
as an input (thus floating), and letting the mandatory pull-up resistors
do the rest. Anything else is actually wrong on I2C where all devices
are supposed to have open-drain connection to the bus.
The missing I2CBB_EN is set (along with a safe initial value of the
GPIOs) just before registering this software I2C bus.
The chip datasheet mentions HW-implemented I2C transactions (SFP EEPROM
reads and writes) as well, but I'm not touching these for simplicity.
Tested on a LR-Link LRES2203PF-2SFP (which is an almost-miniPCIe form
factor card, a cable, and a module with two SFP cages). There was one
casualty, an old broken SFP we had laying around, which was used to
solder some thin wires as a DIY I2C breakout. Thanks for your service.
With this patch in place, I can `i2cdump -y 3 0x51 c` and read back data
which make sense. Yay.
Signed-off-by: Jan Kundrát <jan.kundrat@cesnet.cz>
See-also: https://www.spinics.net/lists/netdev/msg490554.html
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Tony Brelinski <tony.brelinski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Convert all Ethernet drivers from memcpy(... dev->addr_len)
to eth_hw_addr_set():
@@
expression dev, np;
@@
- memcpy(dev->dev_addr, np, dev->addr_len)
+ eth_hw_addr_set(dev, np)
In theory addr_len may not be ETH_ALEN, but we don't expect
non-Ethernet devices to live under this directory, and only
the following cases of setting addr_len exist:
- cxgb4 for mgmt device,
and the drivers which set it to ETH_ALEN: s2io, mlx4, vxge.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Most users of ndo_do_ioctl are ethernet drivers that implement
the MII commands SIOCGMIIPHY/SIOCGMIIREG/SIOCSMIIREG, or hardware
timestamping with SIOCSHWTSTAMP/SIOCGHWTSTAMP.
Separate these from the few drivers that use ndo_do_ioctl to
implement SIOCBOND, SIOCBR and SIOCWANDEV commands.
This is a purely cosmetic change intended to help readers find
their way through the implementation.
Cc: Doug Ledford <dledford@redhat.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jay Vosburgh <j.vosburgh@gmail.com>
Cc: Veaceslav Falico <vfalico@gmail.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Cc: Andrew Lunn <andrew@lunn.ch>
Cc: Vivien Didelot <vivien.didelot@gmail.com>
Cc: Florian Fainelli <f.fainelli@gmail.com>
Cc: Vladimir Oltean <olteanv@gmail.com>
Cc: Leon Romanovsky <leon@kernel.org>
Cc: linux-rdma@vger.kernel.org
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Assignment to *ring should be done after correctness check of the
argument queue.
Fixes: 91db364236 ("igb: Refactor igb_configure_cbs()")
Signed-off-by: Jedrzej Jagielski <jedrzej.jagielski@intel.com>
Acked-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Tested-by: Tony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Ensure that the adapter->q_vector[MAX_Q_VECTORS] array isn't accessed
beyond its size. It was fixed by using a local variable num_q_vectors
as a limit for loop index, and ensure that num_q_vectors is not bigger
than MAX_Q_VECTORS.
Fixes: 047e0030f1 ("igb: add new data structure for handling interrupts and NAPI")
Signed-off-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com>
Reviewed-by: Grzegorz Siwik <grzegorz.siwik@intel.com>
Reviewed-by: Arkadiusz Kubalewski <arkadiusz.kubalewski@intel.com>
Reviewed-by: Slawomir Laba <slawomirx.laba@intel.com>
Reviewed-by: Sylwester Dziedziuch <sylwesterx.dziedziuch@intel.com>
Reviewed-by: Mateusz Palczewski <mateusz.placzewski@intel.com>
Tested-by: Tony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
If an error occurs after a 'pci_enable_pcie_error_reporting()' call, it
must be undone by a corresponding 'pci_disable_pcie_error_reporting()'
call, as already done in the remove function.
Fixes: 40a914fa72 ("igb: Add support for pci-e Advanced Error Reporting")
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Tested-by: Tony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Cleans the next descriptor to watch (next_to_watch) when cleaning the
TX ring.
Failure to do so can cause invalid memory accesses. If igb_poll() runs
while the controller is reset this can lead to the driver try to free
a skb that was already freed.
(The crash is harder to reproduce with the igb driver, but the same
potential problem exists as the code is identical to igc)
Fixes: 7cc6fd4c60 ("igb: Don't bother clearing Tx buffer_info in igb_clean_tx_ring")
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Reported-by: Erez Geva <erez.geva.ext@siemens.com>
Tested-by: Tony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
The Intel drivers all have rcu_read_lock()/rcu_read_unlock() pairs around
XDP program invocations. However, the actual lifetime of the objects
referred by the XDP program invocation is longer, all the way through to
the call to xdp_do_flush(), making the scope of the rcu_read_lock() too
small. This turns out to be harmless because it all happens in a single
NAPI poll cycle (and thus under local_bh_disable()), but it makes the
rcu_read_lock() misleading.
Rather than extend the scope of the rcu_read_lock(), just get rid of it
entirely. With the addition of RCU annotations to the XDP_REDIRECT map
types that take bh execution into account, lockdep even understands this to
be safe, so there's really no reason to keep it around.
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: Jesper Dangaard Brouer <brouer@redhat.com> # i40e
Cc: Jesse Brandeburg <jesse.brandeburg@intel.com>
Cc: Tony Nguyen <anthony.l.nguyen@intel.com>
Cc: intel-wired-lan@lists.osuosl.org
Link: https://lore.kernel.org/bpf/20210624160609.292325-12-toke@redhat.com
Add missing exception tracing to XDP when a number of different
errors can occur. The support was only partial. Several errors
where not logged which would confuse the user quite a lot not
knowing where and why the packets disappeared.
Fixes: 9cbc948b5a ("igb: add XDP support")
Reported-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Tested-by: Vishakha Jambekar <vishakha.jambekar@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
When using native XDP with the igb driver, the XDP frame data doesn't point to
the beginning of the packet. It's off by 16 bytes. Everything works as expected
with XDP skb mode.
Actually these 16 bytes are used to store the packet timestamps. Therefore, pull
the timestamp before executing any XDP operations and adjust all other code
accordingly. The igc driver does it like that as well.
Tested with Intel i210 card and AF_XDP sockets.
Fixes: 9cbc948b5a ("igb: add XDP support")
Signed-off-by: Kurt Kanzenbach <kurt@linutronix.de>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Tested-by: Sandeep Penigalapati <sandeep.penigalapati@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
The igb driver was trying hard to be sparse correct, but somehow
ended up converting a variable into little endian order and then
tries to OR something with it.
A much plainer way of doing things is to leave all variables and
OR operations in CPU (non-endian) mode, and then convert to
little endian only once, which is what this change does.
This probably fixes a bug that might have been seen only on
big endian systems.
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Dave Switzer <david.switzer@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
The sparse build (C=2) finds some issues with how the driver
dealt with the (very difficult) hardware that in some generations
uses little-endian, and in others uses big endian, for the VLAN
field. The code as written picks __le16 as a type and for some
hardware revisions we override it to __be16 as done in this
patch. This impacted the VF driver as well so fix it there too.
Also change the vlan_tci assignment to override the sparse
warning without changing functionality.
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Dave Switzer <david.switzer@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
The igb and igc driver both use a trick of creating a local type
pointer on the stack to ease dealing with a receive descriptor in
64 bit chunks for printing. Sparse however was not taken into
account and receive descriptors are always in little endian
order, so just make the unions use __le64 instead of u64.
No functional change.
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Dvora Fuxbrumer <dvorax.fuxbrumer@linux.intel.com>
Tested-by: Dave Switzer <david.switzer@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
i210 has a total of 24KB of transmit packet buffer. When in Qav mode,
this buffer is divided into four pieces, one for each Tx queue.
Currently, 8KB are given to each of the two SR queues and 4KB are given
to each of the two SP queues.
However, it was noticed that such distribution can make best effort
traffic (which would usually go to the SP queues when Qav is enabled, as
the SR queues would be used by ETF or CBS qdiscs for TSN-aware traffic)
perform poorly. Using iperf3 to measure, one could see the performance
of best effort traffic drop by nearly a third (from 935Mbps to 578Mbps),
with no TSN traffic competing.
This patch redistributes the 24KB to each queue equally: 6KB each. On
tests, there was no notable performance reduction of best effort traffic
performance when there was no TSN traffic competing.
Below, more details about the data collected:
All experiments were run using the following qdisc setup:
qdisc taprio 100: root refcnt 9 tc 4 map 3 3 3 2 3 0 0 3 3 3 3 3 3 3 3 3
queues offset 0 count 1 offset 1 count 1 offset 2 count 1 offset 3 count 1
clockid TAI base-time 0 cycle-time 10000000 cycle-time-extension 0
index 0 cmd S gatemask 0xf interval 10000000
qdisc etf 8045: parent 100:1 clockid TAI delta 1000000 offload on
deadline_mode off skip_sock_check off
TSN traffic, when enabled, had this characteristics:
Packet size: 1500 bytes
Transmission interval: 125us
----------------------------------
Without this patch:
----------------------------------
- TCP data:
- No TSN traffic:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-20.00 sec 1.35 GBytes 578 Mbits/sec 0
- With TSN traffic:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-20.00 sec 1.07 GBytes 460 Mbits/sec 1
- TCP data limiting iperf3 buffer size to 4K:
- No TSN traffic:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-20.00 sec 1.35 GBytes 579 Mbits/sec 0
- With TSN traffic:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-20.00 sec 1.08 GBytes 462 Mbits/sec 0
- TCP data limiting iperf3 buffer size to 192 bytes (smallest size without
serious performance degradation):
- No TSN traffic:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-20.00 sec 1.34 GBytes 577 Mbits/sec 0
- With TSN traffic:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-20.00 sec 1.07 GBytes 461 Mbits/sec 1
- UDP data at 1000Mbit/sec:
- No TSN traffic:
[ ID] Interval Transfer Bitrate Jitter Lost/Total Datagrams
[ 5] 0.00-20.00 sec 1.36 GBytes 586 Mbits/sec 0.000 ms 0/1011407 (0%)
- With TSN traffic:
[ ID] Interval Transfer Bitrate Jitter Lost/Total Datagrams
[ 5] 0.00-20.00 sec 1.05 GBytes 451 Mbits/sec 0.000 ms 0/778672 (0%)
----------------------------------
With this patch:
----------------------------------
- TCP data:
- No TSN traffic:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-20.00 sec 2.17 GBytes 932 Mbits/sec 0
- With TSN traffic:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-20.00 sec 1.50 GBytes 646 Mbits/sec 1
- TCP data limiting iperf3 buffer size to 4K:
- No TSN traffic:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-20.00 sec 2.17 GBytes 931 Mbits/sec 0
- With TSN traffic:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-20.00 sec 1.50 GBytes 645 Mbits/sec 0
- TCP data limiting iperf3 buffer size to 192 bytes (smallest size without
serious performance degradation):
- No TSN traffic:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-20.00 sec 2.17 GBytes 932 Mbits/sec 1
- With TSN traffic:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-20.00 sec 1.50 GBytes 645 Mbits/sec 0
- UDP data at 1000Mbit/sec:
- No TSN traffic:
[ ID] Interval Transfer Bitrate Jitter Lost/Total Datagrams
[ 5] 0.00-20.00 sec 2.23 GBytes 956 Mbits/sec 0.000 ms 0/1650226 (0%)
- With TSN traffic:
[ ID] Interval Transfer Bitrate Jitter Lost/Total Datagrams
[ 5] 0.00-20.00 sec 1.51 GBytes 649 Mbits/sec 0.000 ms 0/1120264 (0%)
Signed-off-by: Ederson de Souza <ederson.desouza@intel.com>
Tested-by: Tony Brelinski <tonyx.brelinski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Alexei Starovoitov says:
====================
pull-request: bpf-next 2021-03-24
The following pull-request contains BPF updates for your *net-next* tree.
We've added 37 non-merge commits during the last 15 day(s) which contain
a total of 65 files changed, 3200 insertions(+), 738 deletions(-).
The main changes are:
1) Static linking of multiple BPF ELF files, from Andrii.
2) Move drop error path to devmap for XDP_REDIRECT, from Lorenzo.
3) Spelling fixes from various folks.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
A bunch of header comments were showing warnings when compiling
with W=1. Fix them all at once. This changes only comments.
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Add a couple of checks to make sure timestamping is on and that the
timestamp value from DMA is valid. This avoids any functional issues
that could come from a misinterpreted time stamp.
One of the functions changed doesn't need a return value added because
there was no value in checking from the calling locations.
While here, fix a couple of reverse christmas tree issues next to
the code being changed.
Fixes: f56e7bba22 ("igb: Pull timestamp from fragment before adding it to skb")
Fixes: 9cbc948b5a ("igb: add XDP support")
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Dave Switzer <david.switzer@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
We want to change the current ndo_xdp_xmit drop semantics because it will
allow us to implement better queue overflow handling. This is working
towards the larger goal of a XDP TX queue-hook. Move XDP_REDIRECT error
path handling from each XDP ethernet driver to devmap code. According to
the new APIs, the driver running the ndo_xdp_xmit pointer, will break tx
loop whenever the hw reports a tx error and it will just return to devmap
caller the number of successfully transmitted frames. It will be devmap
responsibility to free dropped frames.
Move each XDP ndo_xdp_xmit capable driver to the new APIs:
- veth
- virtio-net
- mvneta
- mvpp2
- socionext
- amazon ena
- bnxt
- freescale (dpaa2, dpaa)
- xen-frontend
- qede
- ice
- igb
- ixgbe
- i40e
- mlx5
- ti (cpsw, cpsw-new)
- tun
- sfc
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Reviewed-by: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Reviewed-by: Camelia Groza <camelia.groza@nxp.com>
Acked-by: Edward Cree <ecree.xilinx@gmail.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Acked-by: Shay Agroskin <shayagr@amazon.com>
Link: https://lore.kernel.org/bpf/ed670de24f951cfd77590decf0229a0ad7fd12f6.1615201152.git.lorenzo@kernel.org
s/structue/structure/
Signed-off-by: Bhaskar Chowdhury <unixbhaskar@gmail.com>
Acked-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Igb needs a similar fix as commit 75aab4e10a ("i40e: avoid
premature Rx buffer reuse")
The page recycle code, incorrectly, relied on that a page fragment
could not be freed inside xdp_do_redirect(). This assumption leads to
that page fragments that are used by the stack/XDP redirect can be
reused and overwritten.
To avoid this, store the page count prior invoking xdp_do_redirect().
Longer explanation:
Intel NICs have a recycle mechanism. The main idea is that a page is
split into two parts. One part is owned by the driver, one part might
be owned by someone else, such as the stack.
t0: Page is allocated, and put on the Rx ring
+---------------
used by NIC ->| upper buffer
(rx_buffer) +---------------
| lower buffer
+---------------
page count == USHRT_MAX
rx_buffer->pagecnt_bias == USHRT_MAX
t1: Buffer is received, and passed to the stack (e.g.)
+---------------
| upper buff (skb)
+---------------
used by NIC ->| lower buffer
(rx_buffer) +---------------
page count == USHRT_MAX
rx_buffer->pagecnt_bias == USHRT_MAX - 1
t2: Buffer is received, and redirected
+---------------
| upper buff (skb)
+---------------
used by NIC ->| lower buffer
(rx_buffer) +---------------
Now, prior calling xdp_do_redirect():
page count == USHRT_MAX
rx_buffer->pagecnt_bias == USHRT_MAX - 2
This means that buffer *cannot* be flipped/reused, because the skb is
still using it.
The problem arises when xdp_do_redirect() actually frees the
segment. Then we get:
page count == USHRT_MAX - 1
rx_buffer->pagecnt_bias == USHRT_MAX - 2
From a recycle perspective, the buffer can be flipped and reused,
which means that the skb data area is passed to the Rx HW ring!
To work around this, the page count is stored prior calling
xdp_do_redirect().
Fixes: 9cbc948b5a ("igb: add XDP support")
Signed-off-by: Li RongQing <lirongqing@baidu.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Tested-by: Vishakha Jambekar <vishakha.jambekar@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
As explained in commit 29d98f54a4 ("net: enetc: allow hardware
timestamping on TX queues with tc-etf enabled"), hardware TX
timestamping requires an skb with skb->tstamp = 0. When a packet is sent
with SO_TXTIME, the skb->skb_mstamp_ns corrupts the value of skb->tstamp,
so the drivers need to explicitly reset skb->tstamp to zero after
consuming the TX time.
Create a helper named skb_txtime_consumed() which does just that. All
drivers which offload TC_SETUP_QDISC_ETF should implement it, and it
would make it easier to assess during review whether they do the right
thing in order to be compatible with hardware timestamping or not.
Suggested-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Acked-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Tony Nguyen says:
====================
1GbE Intel Wired LAN Driver Updates 2021-02-03
This series contains updates to igc, igb, e1000e, and e1000 drivers.
Sasha adds counting of good transmit packets and reporting of NVM version
and gPHY version in ethtool firmware version. Replaces the use of strlcpy
to the preferred strscpy. Fixes a typo that caused the wrong register to be
output. He also removes an unused function pointer, some unneeded defines,
and a non-applicable comment. All changes for igc.
Gal Hammer fixes a typo which caused the RDBAL register values to be
shown instead of TDBAL for igb.
Nick Lowe enables RSS support for i211 devices for igb.
Tom Rix fixes checkpatch warning by removing h from printk format
specifier for igb.
Kaixu Xia removes setting of a variable that is overwritten before next
use for e1000e.
Sudip Mukherjee removes an unneeded assignment for e1000.
* '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue:
e1000: drop unneeded assignment in e1000_set_itr()
e1000e: remove the redundant value assignment in e1000_update_nvm_checksum_spt
igb: remove h from printk format specifier
igb: Enable RSS for Intel I211 Ethernet Controller
igb: fix TDBAL register show incorrect value
igc: Fix TDBAL register show incorrect value
igc: Remove unused FUNC_1 mask
igc: Remove unused local receiver mask
igc: Prefer strscpy over strlcpy
igc: Expose the gPHY firmware version
igc: Expose the NVM version
igc: Add Host Good Packets Transmitted Count
igc: Remove MULR mask define
igc: Remove igc_set_fw_version comment
igc: Clean up nvm_operations structure
====================
Link: https://lore.kernel.org/r/20210204004259.3662059-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Now we can remove a bunch of identical functions from the drivers and
make them use common dev_page_is_reusable(). All {,un}likely() checks
are omitted since it's already present in this helper.
Also update some comments near the call sites.
Suggested-by: David Rientjes <rientjes@google.com>
Suggested-by: Jakub Kicinski <kuba@kernel.org>
Cc: John Hubbard <jhubbard@nvidia.com>
Signed-off-by: Alexander Lobakin <alobakin@pm.me>
Reviewed-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This change fixes the checkpatch warning described in this
commit cbacb5ab0a ("docs: printk-formats: Stop encouraging use of
unnecessary %h[xudi] and %hh[xudi]")
Standard integer promotion is already done and %hx and %hhx is useless
so do not encourage the use of %hh[xudi] or %h[xudi].
Signed-off-by: Tom Rix <trix@redhat.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
The Intel I211 Ethernet Controller supports 2 Receive Side Scaling (RSS)
queues. It should not be excluded from having this feature enabled.
Via commit c883de9fd7 ("igb: rename igb define to be more generic")
E1000_MRQC_ENABLE_RSS_4Q was renamed to E1000_MRQC_ENABLE_RSS_MQ to
indicate that this is a generic bit flag to enable queues and not
a flag that is specific to devices that support 4 queues
The bit flag enables 2, 4 or 8 queues appropriately depending on the part.
Tested with a multicore CPU and frames were then distributed as expected.
This issue appears to have been introduced because of confusion caused
by the prior name.
Signed-off-by: Nick Lowe <nick.lowe@gmail.com>
Tested-by: David Switzer <david.switzer@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Fixed a typo which caused the registers dump function to read the
RDBAL register when printing TDBAL register values.
Signed-off-by: Gal Hammer <ghammer@redhat.com>
Tested-by: David Switzer <david.switzer@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Using skb_csum_is_sctp is a easier way to validate it's a SCTP
CRC checksum offload packet, and there is no need to parse the
packet to check its proto field, especially when it's a UDP or
GRE encapped packet.
So this patch also makes igb support SCTP CRC checksum offload
for UDP and GRE encapped packets.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Reviewed-by: Alexander Duyck <alexanderduyck@fb.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
xdp_return_frame_bulk() needs to pass a xdp_buff
to __xdp_return().
strlcpy got converted to strscpy but here it makes no
functional difference, so just keep the right code.
Conflicts:
net/netfilter/nf_tables_api.c
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Kai-Heng Feng