IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
This patch ensures ixgbe will not try to offload hash tables from the
u32 module. The device class does not currently support this so until
it is enabled just abort on these tables.
Interestingly the more flexible your hardware is the less code you
need to implement to guard against these cases.
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This adds initial support for offloading the u32 tc classifier. This
initial implementation only implements a few base matches and actions
to illustrate the use of the infrastructure patches.
However it is an interesting subset because it handles the u32 next
hdr logic to correctly map tcp packets from ip headers using the ihl
and protocol fields. After this is accepted we can extend the match
and action fields easily by updating the model header file.
Also only the drop action is supported initially.
Here is a short test script,
#tc qdisc add dev eth4 ingress
#tc filter add dev eth4 parent ffff: protocol ip \
u32 ht 800: order 1 \
match ip dst 15.0.0.1/32 match ip src 15.0.0.2/32 action drop
<-- hardware has dst/src ip match rule installed -->
#tc filter del dev eth4 parent ffff: prio 49152
#tc filter add dev eth4 parent ffff: protocol ip prio 99 \
handle 1: u32 divisor 1
#tc filter add dev eth4 protocol ip parent ffff: prio 99 \
u32 ht 800: order 1 link 1: \
offset at 0 mask 0f00 shift 6 plus 0 eat match ip protocol 6 ff
#tc filter add dev eth4 parent ffff: protocol ip \
u32 ht 1: order 3 match tcp src 23 ffff action drop
<-- hardware has tcp src port rule installed -->
#tc qdisc del dev eth4 parent ffff:
<-- hardware cleaned up -->
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch updates setup_tc so we can pass additional parameters into
the ndo op in a generic way. To do this we provide structured union
and type flag.
This lets each classifier and qdisc provide its own set of attributes
without having to add new ndo ops or grow the signature of the
callback.
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ndo_setup_tc() op was added to support drivers offloading tx
qdiscs however only support for mqprio was ever added. So we
only ever added support for passing the number of traffic classes
to the driver.
This patch generalizes the ndo_setup_tc op so that a handle can
be provided to indicate if the offload is for ingress or egress
or potentially even child qdiscs.
CC: Murali Karicheri <m-karicheri2@ti.com>
CC: Shradha Shah <sshah@solarflare.com>
CC: Or Gerlitz <ogerlitz@mellanox.com>
CC: Ariel Elior <ariel.elior@qlogic.com>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Bruce Allan <bruce.w.allan@intel.com>
CC: Jesse Brandeburg <jesse.brandeburg@intel.com>
CC: Don Skidmore <donald.c.skidmore@intel.com>
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is an opportunity to bulk free SKBs during reclaiming of
resources after DMA transmit completes in ixgbe_clean_tx_irq. Thus,
bulk freeing at this point does not introduce any added latency.
Simply use napi_consume_skb() which were recently introduced. The
napi_budget parameter is needed by napi_consume_skb() to detect if it
is called from netpoll.
Benchmarking IPv4-forwarding, on CPU i7-4790K @4.2GHz (no turbo boost)
Single CPU/flow numbers: before: 1982144 pps -> after : 2064446 pps
Improvement: +82302 pps, -20 nanosec, +4.1%
(SLUB and GCC version 5.1.1 20150618 (Red Hat 5.1.1-4))
Joint work with Alexander Duyck.
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Right now ATR is not handling IPv6 extended headers, so ATR is not
being performed on such packets. Fix that by skipping extended
headers when they are present. This also fixes a problem where
the ATR code was not checking that the inner protocol was actually
TCP before setting up the signature rules. Since the protocol check
is intimately involved with the extended header processing as well,
this all gets fixed together.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
When FCoE is enabled with SR-IOV on the X550 NIC the hardware
generates MDD events.
This patch fixes these by setting the expected values in the
Tx context descriptors for FCoE/FIP frames and adding a flush
after writing the RDLEN register.
Signed-off-by: Neerav Parikh <neerav.parikh@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
If an outer UDP checksum is set, pass the skb up with CHECKSUM_NONE
so that the stack will check the checksum. Do not increment an
error counter, because we don't know that there is an actual error.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Clean up minor redundancy in the setting of hw_enc_features that
makes it appears that X550 uniquely has more encapsulation features
than other devices. The driver only supports one more feature, so
make it look that way. No longer set NETIF_F_SG since that is set
by the register_netdev call. Thanks to Alex Duyck for noticing this
slight confusion.
Reported-by: Alexander Duyck <aduyck@mirantis.com>
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Conflicts:
drivers/net/geneve.c
Here we had an overlapping change, where in 'net' the extraneous stats
bump was being removed whilst in 'net-next' the final argument to
udp_tunnel6_xmit_skb() was being changed.
Signed-off-by: David S. Miller <davem@davemloft.net>
The name NETIF_F_ALL_CSUM is a misnomer. This does not correspond to the
set of features for offloading all checksums. This is a mask of the
checksum offload related features bits. It is incorrect to set both
NETIF_F_HW_CSUM and NETIF_F_IP_CSUM or NETIF_F_IPV6 at the same time for
features of a device.
This patch:
- Changes instances of NETIF_F_ALL_CSUM to NETIF_F_CSUM_MASK (where
NETIF_F_ALL_CSUM is being used as a mask).
- Changes bonding, sfc/efx, ipvlan, macvlan, vlan, and team drivers to
use NEITF_F_HW_CSUM in features list instead of NETIF_F_ALL_CSUM.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The SCTP checksum is really a CRC and is very different from the
standards 1's complement checksum that serves as the checksum
for IP protocols. This offload interface is also very different.
Rename NETIF_F_SCTP_CSUM to NETIF_F_SCTP_CRC to highlight these
differences. The term CSUM should be reserved in the stack to refer
to the standard 1's complement IP checksum.
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch is a follow-on for enabling VLAN promiscuous and allowing the PF
to add VLANs without adding a VLVF entry. What this patch does is go
through and free the VLVF registers if they are not needed as the VLAN
belongs only to the PF which is the default pool.
Signed-off-by: Alexander Duyck <aduyck@mirantis.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch adds support for VLAN promiscuous with SR-IOV enabled.
The code prior to this patch was only adding the PF to VLANs that the VF
had added. As such enabling promiscuous mode would actually not add any
additional VLAN filters so visibility was limited. This lead to a number
of issues as the bridge and OVS would expect us to accept all VLAN tagged
packets when promiscuous mode was enabled, and instead we would filter out
most if not all depending on the configuration of the PF.
With this patch what we do is set all the bits in the VFTA and all of the
VLVF bits associated with the pool belonging to the PF. By doing this the
PF is guaranteed to receive all VLAN tagged traffic associated with the RAR
filters assigned to the PF. In addition we will clean up those same bits
in the event of promiscuous mode being disabled.
Signed-off-by: Alexander Duyck <aduyck@mirantis.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch adds support for bypassing the VLVF entry creation when the PF
is adding a new VLAN. The advantage to doing this is that we can then save
the VLVF entries for the VFs which must have them in order to function,
versus the PF which can fall back on the default pool entry.
Signed-off-by: Alexander Duyck <aduyck@mirantis.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Add a check to make certain mac_table was actually allocated and is not
NULL. If it is NULL return -ENOMEM and allow the probe routine to fail
rather then causing a NULL pointer dereference further down the line.
Signed-off-by: Alexander Duyck <aduyck@mirantis.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Enabling SR-IOV and then bringing the interface up was resulting in the PF
MAC addresses getting into a bad state. Specifically the MAC address was
enabled for both VF 0 and the PF. This resulted in some odd behaviors such
as VF 0 receiving a copy of the PFs traffic, which in turn enables the
ability for VF 0 to spoof the PF.
A workaround for this issue appears to be to bring up the interface first
and then enable SR-IOV as this way the reset is then triggered in the
existing code.
In order to correct this I have added a change to ixgbe_setup_tc where if
the interface is down we still will at least call ixgbe_reset so that the
MAC addresses for the device are reset to the correct pools.
Steps to reproduce issue:
modprobe ixgbe
echo 7 > /sys/bus/pci/devices/0000\:01\:00.1/sriov_numvfs
ifconfig enp1s0f1 up
ethregs -s 1:00.1 | grep MPSAR | grep -v 00000000
Result:
MPSAR[0] 00000081
MPSAR[254] 00000001
Expected Result, behavior after patch:
MPSAR[0] 00000080
MPSAR[254] 00000080
Signed-off-by: Alexander Duyck <aduyck@mirantis.com>
Tested-by: Darin Miller <darin.j.miller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Check for and handle IPv6 extended headers so that Tx checksum
offload can be done. Also use skb_checksum_help for unexpected
cases. Thanks to Tom Herbert for noticing these problems. Thanks
to Alexander Duyck for recognizing problems with the first version
of this patch and recognizing how to coalesce error conditions
into a single location.
Reported-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Save VF device pointers and take references to speed accesses used
to monitor the device behavior to avoid slot resets. The saved
information avoids lock contention during the search used to access
each of the VFs.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Darin Miller <darin.j.miller@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The X550EM_x devices handle clocking differently, so update the
PTP implementation to accommodate them. This involves significant
changes to ixgbe's PTP code to accommodate the new range of
behaviors including things like non-power-of-2 clock wrapping.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Darin Miller <darin.j.miller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This change makes it so that we allow the PF to make use of all free RAR
entries for FDB use if needed.
Previously the code limited us to 16 unicast entries, however this was
shared between MACVLAN which wasn't limited and the FDB code which was. So
instead of treating the FDB code as a second class citizen I have updated
it so that it has access to just as many entries as the MACVLAN filters.
Signed-off-by: Alexander Duyck <aduyck@mirantis.com>
Tested-by: Darin Miller <darin.j.miller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This change replaces the ixgbe_write_uc_addr_list call in ixgbe_set_rx_mode
with a call to __dev_uc_sync instead. This works much better with the MAC
addr list code that was already in place and solves an issue in which you
couldn't remove an FDB address without having to reset the port.
Signed-off-by: Alexander Duyck <aduyck@mirantis.com>
Tested-by: Darin Miller <darin.j.miller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
In the process of tracking down a memory leak when adding/removing FDB
entries I had to go through the MAC address configuration code for ixgbe.
In the process of doing so I found a number of issues that impacted
readability and performance. This change updates the code in general to
clean it up so it becomes clear what each step is doing. From what I can
tell there a couple of bugs cleaned up in this code.
First is the fact that the MAC addresses were being double counted for the
PF. As a result once entries up to 63 had been used you could no longer
add additional filters.
A simple test case for this:
for i in `seq 0 96`
do
ip link add link ens8 name mv$i type macvlan
ip link set dev mv$i up
done
Test script:
ethregs -s 0:8.0 | grep -e "RAH" | grep 8000....$
When things are working correctly RAL/H registers 1 - 97 will be consumed.
In the failing case it will stop at 63 and prevent any further filters from
being added.
Signed-off-by: Alexander Duyck <aduyck@mirantis.com>
Tested-by: Darin Miller <darin.j.miller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Use a private workqueue to avoid hangs that were otherwise possible
when performing stress tests, such as creating and destroying many
VFS repeatedly.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Darin Miller <darin.j.miller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The ixgbe_intr and ixgbe/ixgbevf_msix_clean_rings functions run from hard
interrupt context or with interrupts already disabled in netpoll.
They can use napi_schedule_irqoff() instead of napi_schedule()
Signed-off-by: Alexander Duyck <aduyck@mirantis.com>
Tested-by: Darin Miller <darin.j.miller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch corrects an issue in which the polling routine would increase
the budget for Rx to at least 1 per queue if multiple queues were present.
This would result in Rx packets being processed when the budget was 0 which
is meant to indicate that no Rx can be handled.
Signed-off-by: Alexander Duyck <aduyck@mirantis.com>
Tested-by: Darin Miller <darin.j.miller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The commit dfaf891dd3e1 ("ixgbe: Refactor the RSS configuration code")
introduced a few kernel-doc errors:
1) The function name is missing;
2) The format is wrong;
3) The short description is redundant.
Fix all the above for the correct execution of the kernel doc.
Signed-off-by: Jean Sacren <sakiwit@gmail.com>
Tested-by: Darin Miller <darin.j.miller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Delete a redundant include of net/vxlan.h.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Darin Miller <darin.j.miller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
We would like to automatically provide busy polling support
to all NAPI drivers, without them having to implement anything.
skb_mark_napi_id() can be called from napi_gro_receive() and
napi_get_frags().
Few drivers are still calling skb_mark_napi_id() because
they use netif_receive_skb(). They should eventually call
napi_gro_receive() instead. I will leave this to drivers
maintainers.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Implements the new netdev op to trust VF in ixgbe.
The administrator can turn on and off VF trusted by ip command which
supports trust message.
# ip link set dev eth0 vf 1 trust on
or
# ip link set dev eth0 vf 1 trust off
Send a ping to reset VF on changing the status of trusting.
VF driver will reconfigure its features on reset.
Signed-off-by: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
Tested-by: Krishneil Singh <Krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
As per Eric Dumazet's previous patches:
(see commit (24d2e4a50737) - tg3: use napi_complete_done())
Quoting verbatim:
Using napi_complete_done() instead of napi_complete() allows
us to use /sys/class/net/ethX/gro_flush_timeout
GRO layer can aggregate more packets if the flush is delayed a bit,
without having to set too big coalescing parameters that impact
latencies.
</end quote>
Tested
configuration: low latency via ethtool -C ethx adaptive-rx off
rx-usecs 10 adaptive-tx off tx-usecs 15
workload: streaming rx using netperf TCP_MAERTS
igb:
MIGRATED TCP MAERTS TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.0.0.1 () port 0 AF_INET : demo
...
Interim result: 941.48 10^6bits/s over 1.000 seconds ending at 1440193171.589
Alignment Offset Bytes Bytes Recvs Bytes Sends
Local Remote Local Remote Xfered Per Per
Recv Send Recv Send Recv (avg) Send (avg)
8 8 0 0 1176930056 1475.36 797726 16384.00 71905
MIGRATED TCP MAERTS TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.0.0.1 () port 0 AF_INET : demo
...
Interim result: 941.49 10^6bits/s over 0.997 seconds ending at 1440193142.763
Alignment Offset Bytes Bytes Recvs Bytes Sends
Local Remote Local Remote Xfered Per Per
Recv Send Recv Send Recv (avg) Send (avg)
8 8 0 0 1175182320 50476.00 23282 16384.00 71816
i40e:
Hard to test because the traffic is incoming so fast (24Gb/s) that GRO
always receives 87kB, even at the highest interrupt rate.
Other drivers were only compile tested.
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch disables LRO by default in favor of GRO.
LRO is incompatible with forwarding and is disabled when forwarding
is turned on which makes the default offloads of the driver
inconsistent. LRO can still be enabled via ethtool.
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Darin Miller <darin.j.miller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch makes sure that flow control packets initiated by the VF are
dropped and reported as spoofed.
Flow control packets can be used to limit the throughput or as DOS
attack when generated from a VF. Flow control is not supported per VF
hence any pause frames generated from a VF are considered malicious.
Also cleaned up indentation and some redundant comments.
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
With the addition of X550em_x SFP+ support, the driver is now
functionally equivalent to what will be the 4.2.1 driver when
released, so change the version to match.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The X540 thermal interrupt (IXGBE_EIMS_TS) is not an SDP, so it
doesn't need to be enabled in ixgbe_setup_gpie(). In fact the
value is simply not for the GPIE register at all.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This change does two things. First, it makes it so that we always
set the relaxed ordering bits related to the DCA registers even if
DCA is not enabled. Second, it moves the configuration out of the
ixgbe_down function and into the ixgbe_configure function before
enabling the Rx and Tx rings. This ensures that DCA is configured
correctly before starting to process packets.
Thanks to Alex Duyck for this fix.
CC: Alex Duyck <aduyck@mirantis.com>
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Add new device ID for X550EM device with SFPs.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch sets RDRXCTL.PSP when the driver is in SRIOV mode which
enables padding of small packets.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Setting the X550* RDRXCTL register should fall through into X540
and 82599, not 82598.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Reduce the frequency of polling for SFP modules. Because the
service task sometimes runs at high rates, we can poll for
SFPs too often. When an SFP is not present, the I2C timeouts
that result are very costly. So, prevent SFP polling from
being done more than once every two seconds. To reduce latency,
the poll time is cleared in a couple of cases to permit the
next service task execution to poll the SFP module.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Add support for the SFP insertion interrupt on X550EM devices with
SFPs.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
When an SFP not present error is returned by the reset_hw method,
accept it and go on, since an SFP can still be inserted. Previously
it was only accepted for 82598 devices.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch is part of the future enablement of X550 SFP+ support. This
HW uses different SDP so the interrupts need to be set up accordingly.
Signed-off-by: Donald C Skidmore <donald.c.skidmore@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch updates the lowest limit for adaptive interrupt interrupt
moderation to roughly 12K interrupts per second.
The way I came about reaching 12K as the desired interrupt rate is by
testing with UDP flows. Specifically I had a simple test that ran a
netperf UDP_STREAM test at varying sizes. What I found was as the packet
sizes increased the performance fell steadily behind until we were only
able to receive at ~4Gb/s with a message size of 65507. A bit of digging
found that we were dropping packets for the socket in the network stack,
and looking at things further what I found was I could solve it by increasing
the interrupt rate, or increasing the rmem_default/rmem_max. What I found was
that when the interrupt coalescing resulted in more data being processed
per interrupt than could be stored in the socket buffer we started losing
packets and the performance dropped. So I reached 12K based on the
following math.
rmem_default = 212992
skb->truesize = 2994
212992 / 2994 = 71.14 packets to fill the buffer
packet rate at 1514 packet size is 812744pps
71.14 / 812744 = 87.9us to fill socket buffer
From there it was just a matter of choosing the interrupt rate and
providing a bit of wiggle room which is why I decided to go with 12K
interrupts per second as that uses a value of 84us.
The data below is based on VM to VM over a direct assigned ixgbe interface.
The test run was:
netperf -H <ip> -t UDP_STREAM"
Socket Message Elapsed Messages CPU Service
Size Size Time Okay Errors Throughput Util Demand
bytes bytes secs # # 10^6bits/sec % SS us/KB
Before:
212992 65507 60.00 1100662 0 9613.4 10.89 0.557
212992 60.00 473474 4135.4 11.27 0.576
After:
212992 65507 60.00 1100413 0 9611.2 10.73 0.549
212992 60.00 974132 8508.3 11.69 0.598
Using bare metal the data is similar but not as dramatic as the throughput
increases from about 8.5Gb/s to 9.5Gb/s.
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
When the .remove() callback for a PF is called, SR-IOV support for the
device is disabled, which requires unbinding and removing the VFs.
The VFs may be in-use either by the host kernel or userspace, such as
assigned to a VM through vfio-pci. In this latter case, the VFs may
be removed either by shutting down the VM or hot-unplugging the
devices from the VM. Unfortunately in the case of a Windows 2012 R2
guest, hot-unplug is broken due to the ordering of the PF driver
teardown. Disabling SR-IOV prior to unregister_netdev() avoids this
issue.
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Acked-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Krishneil Singh <krishneil.k.singh@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Add checks for systems that don't have SFP's to avoid incorrectly
acting on interrupts that are falsely interpreted as SFP events.
This also includes a modified check generating the EICR mask to be
more forward-looking.
Signed-off-by: Don Skidmore <donald.c.skidmore@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
When unbinding an SR-IOV device with VFs configured from ixgbe, the
driver behaves in one of two ways. If max_vfs was specified, the
SR-IOV state is disabled, removing the VFs. The occurs regardless of
whether the VF count was later modified through sysfs. If however
max_vfs is zero, such as by not specifying the module parameter, the
VFs persist after the PF is unbound from ixgbe. If the PF is then
bound to vfio-pci to be assigned to a VM, the PF is non-functional.
>From the comment, commit da36b64736cf ("ixgbe: Implement PCI SR-IOV
sysfs callback operation") clearly intended this alternate behavior,
but probably didn't realize the PF doesn't work in this mode.
This bimodal behavior is confusing to users and results in a state
where the PF is broken for other uses unless the user sets
sriov_numvfs to zero prior to unbinding the device. Remove this
behavior so that VFs are removed and the PF is functional for other
uses after unbind, regardless of the way VFs are enabled.
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Acked-by: Greg Rose <gregory.v.rose@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Now that we can do 2.5G link speed, we need to be able to report it.
Also change the nested triadic involved in creating the log message
to instead use a simpler switch statement to set a string pointer.
Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch resolves an issue where users were not able to dynamically
set number of queues for 82598 via ethtool -L
Reported-by: Tal Abudi <talabudi@gmail.com>
Signed-off-by: Emil Tantilov <emil.s.tantilov@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Allows to change the rxfh indirection table and/or key using
ethtool interface.
Signed-off-by: Tom Barbette <tom.barbette@ulg.ac.be>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
