Two functions were always used to set the DMA addresses in Tx
descriptors, because this address is split into a base+offset in the
descriptors. A mask was used to come up with the base and offset
addresses and two functions were called, mvpp2_txdesc_dma_addr_set() and
mvpp2_txdesc_offset_set().
This patch moves the base+offset calculation logic to
mvpp2_txdesc_dma_addr_set(), and removes mvpp2_txdesc_offset_set() to
simplify things.
Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Cosmetic patch using the MVPP2_AGGR_TXQ_SIZE everywhere instead of the
size field of aggr_txq, as the size never change and is always equal to
the MVPP2_AGGR_TXQ_SIZE define.
Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Too many TSO descriptors can be required for the default queue size,
when using small MSS values for example. Prevent this by adding a
maximum number of allowed TSO segments (300). In addition set a stop and
a wake thresholds to stop the queue when there's no room for a 1 "worst
case scenario skb". Wake up the queue when the number of descriptors is
low enough.
Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch initialize the RSS tables to evenly (depending on the packets
RSS hashes) distribute the packets across port Rx queues. This helps to
handle packets on different CPUs to improve performances, as more queues
will be used in parallel.
Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
So far only the Rx FIFO size was initialized. For PPv2.2 the Tx FIFO
size can be set as well. This patch initializes the Tx FIFO size for
PPv2.2 controllers to 3K.
Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The Rx FIFO size was set to the same value for all ports. This patch
sets it depending on the maximum speed a given port can handle. This is
only working for PPv2.2.
Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Variable len is being assigned a value that is never read,
hence the assignment is redundant and can be removed. Cleans
up clang warning:
drivers/net/hamradio/mkiss.c:443:3: warning: Value stored to
'len' is never read
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The functions pci_alloc_consistent is obsolete. So it is replaced
with dma_alloc_coherent
Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Adding a FIB rule on a spectrum platform silently aborts FIB offload:
$ ip ru add pref 99 from all to 192.168.1.1 table 10
$ dmesg -c
[ 623.144736] mlxsw_spectrum 0000:03:00.0: FIB abort triggered. Note that FIB entries are no longer being offloaded to this device.
This patch reworks FIB rule handling to return a message to the user:
$ ip ru add pref 99 from all to 8.8.8.8 table 11
Error: spectrum: FIB rules not supported. Aborting offload.
spectrum currently only checks whether the fib rule is a default rule or
an l3mdev rule, both of which it knows how to handle. Any other it aborts
FIB offload. Move the processing to check the rule type inline with the
user request. If the rule is an unsupported one, then a work queue entry
is used to abort the offload. Change the rule delete handling to just
return since it does nothing at the moment.
Signed-off-by: David Ahern <dsahern@gmail.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
the verifier got progressively smarter over time and size of its internal
state grew as well. Time to reduce the memory consumption.
Before:
sizeof(struct bpf_verifier_state) = 6520
After:
sizeof(struct bpf_verifier_state) = 896
It's done by observing that majority of BPF programs use little to
no stack whereas verifier kept all of 512 stack slots ready always.
Instead dynamically reallocate struct verifier state when stack
access is detected.
Runtime difference before vs after is within a noise.
The number of processed instructions stays the same.
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Enable and disable switchdev on SRIOV capable LiquidIO NIC with devlink.
Create representor netdev for each SRIOV VF function on SRIOV enable and
and do the cleanup on SRIOV disable.
Signed-off-by: Vijaya Mohan Guvva <vijaya.guvva@cavium.com>
Signed-off-by: Satanand Burla <satananda.burla@cavium.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@cavium.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Enable switchdev for SRIOV capable LiquidIO NIC. It registers
a representor netdev (with switchdev_ops) for each SRIOV VF created.
It also has changes to send representor interface configurations like
admin state and MTU to LiquidIO firmware and to retrieve HW counted
VF stats for VF representor.
Signed-off-by: Vijaya Mohan Guvva <vijaya.guvva@cavium.com>
Signed-off-by: Satanand Burla <satananda.burla@cavium.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@cavium.com>
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Saeed Mahameed says:
====================
mlx5-updates-2017-10-31 mlx5e stats groups
This series from Kamal introduces an important refactoring for mlx5e stats
handling, which groups the stats into generic groups structure which allows
to control the behavior and stats reporting per group in a modular way.
In the first patch Kamal introduces a new data type "mlx5e_stats_grp" This change
defines a new API to create a group of stats and simplifies the way of handling them.
This struct will define the following behavior per group:
- get_num_stats() - return the number of counters in the group.
- fill_strings() - fill counters strings within the group.
- fill_stats() - fill counters values within the group.
All other patches will be straight forward refactoring per stats group,
where Kamal will move each mlx5e stats group to use the new API.
The idea is to have better flexibility and modularity to add new counters,
all ethtool logic was rendered generic and loops through the generic stats groups and
calls the groups callbacks to figure out how and what to report back to user space.
Introducing new file en_stats.c to hold all the new stat groups logic and implementation.
Static structures (counters descriptors) moved from en_stats.h to en_stats.c which reduces
the mlx5_core binary footprint, originally reported and addressed by Stephen Hemminger:
("mlx5: fix space waste from ethtool descriptions") which was waived due to this re-design.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Switch the channels counters to use the new stats group API.
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Switch the ipsec counters to use the new stats group API.
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Switch the pme counters to use the new stats group API.
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Switch the per prio pfc counters to use the new stats group API.
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Switch the per prio traffic counters to use the new stats group API.
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Switch the pcie counters to use the new stats group API.
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Switch the ethernet extended counters to use the new stats group API.
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Switch the physical statistical counters to use the new stats group API.
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Switch the RFC 2819 counters to use the new stats group API.
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Switch the RFC 2863 counters to use the new stats group API.
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Switch the IEEE 802.3 counters to use the new stats group API.
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Switch the vport counters to use the new stats group API.
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Switch the Q counters to use the new stats group API.
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Currently the mlx5e driver has multiple groups of stats, each group is
used for different purposes and it may depend on hardware capabilities
or not. The problem with the current implementation is that there is no
clear API to create a new group of stats.
This change define a new API to create a group of stats and simplifies
the way of handling them by defining a new struct "mlx5e_stats_grp" which
have the following three function pointers:
- get_num_stats() - return the number of counters in the group.
- fill_strings() - fill counters strings within the group.
- fill_stats() - fill counters values within the group.
The above function pointers are used within the ethtool callbaks while
calling "ethtool -S" from userspace. This change also switch the SW
group to use the new API.
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Reviewed-by: Gal Pressman <galp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
This patch enables tc-flower based hardware offloads. tc flower
filter provided by the kernel is configured as driver specific
cloud filter. The patch implements functions and admin queue
commands needed to support cloud filters in the driver and
adds cloud filters to configure these tc-flower filters.
The classification function of the filter is to direct matched
packets to a traffic class. The hardware traffic class is set
based on the the classid reserved in the range :ffe0 - :ffef.
Match Dst MAC and route to TC0:
prio 1 flower dst_mac 3c:fd:fe:a0:d6:70 skip_sw\
hw_tc 1
Match Dst IPv4,Dst Port and route to TC1:
prio 2 flower dst_ip 192.168.3.5/32\
ip_proto udp dst_port 25 skip_sw\
hw_tc 2
Match Dst IPv6,Dst Port and route to TC1:
prio 3 flower dst_ip fe8::200:1\
ip_proto udp dst_port 66 skip_sw\
hw_tc 2
Delete tc flower filter:
Example:
Flow Director Sideband is disabled while configuring cloud filters
via tc-flower and until any cloud filter exists.
Unsupported matches when cloud filters are added using enhanced
big buffer cloud filter mode of underlying switch include:
1. source port and source IP
2. Combined MAC address and IP fields.
3. Not specifying L4 port
These filter matches can however be used to redirect traffic to
the main VSI (tc 0) which does not require the enhanced big buffer
cloud filter support.
Signed-off-by: Amritha Nambiar <amritha.nambiar@intel.com>
Signed-off-by: Kiran Patil <kiran.patil@intel.com>
Signed-off-by: Anjali Singhai Jain <anjali.singhai@intel.com>
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Acked-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>
Introduce the cloud filter data structure and cleanup of cloud
filters associated with the device.
Signed-off-by: Amritha Nambiar <amritha.nambiar@intel.com>
Acked-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>
Add new admin queue definitions and extended fields for cloud
filter support. Define big buffer for extended general fields
in Add/Remove Cloud filters command.
Signed-off-by: Amritha Nambiar <amritha.nambiar@intel.com>
Signed-off-by: Kiran Patil <kiran.patil@intel.com>
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Acked-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>
Add definitions for L4 filters and switch modes based on cloud filters
modes and extend the set switch config command to include the
additional cloud filter mode.
Signed-off-by: Amritha Nambiar <amritha.nambiar@intel.com>
Signed-off-by: Kiran Patil <kiran.patil@intel.com>
Acked-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>
Add mapping of TCs with the seids of the channel VSIs. TC0
will be mapped to the main VSI seid and all other TCs are
mapped to the seid of the corresponding channel VSI.
Signed-off-by: Amritha Nambiar <amritha.nambiar@intel.com>
Acked-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>
This reverts commit 11f29003d6.
I am reverting this as I am fairly certain this can result in a memory leak
when combined with the current page recycling scheme. Specifically we end
up attempting to allocate fewer buffers than we recycled and this results
in us rewinding the next to alloc pointer which leads to leaks when we
overwrite the rx_buffer_info when processing the next frame.
Fixes: 11f29003d6 ("i40e/i40evf: bump tail only in multiples of 8")
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>
Whether or not there are vectors_left, we only need to redistribute
our vectors if we didn't get as many as we requested. With the current
check, the code will try to redistribute even if we did in fact get all
the vectors we requested - this can happen when we have more CPUs than
we do vectors. This restores an earlier check to be sure we only
redistribute if we didn't get the full count we requested.
Fixes: 4ce20abc64 (i40e: fix MSI-X vector redistribution if hw limit is reached)
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>
A cleanup of the PM code left an incorrect #ifdef in place, leading
to a harmless build warning:
drivers/net/ethernet/intel/i40e/i40e_main.c:12223:12: error: 'i40e_resume' defined but not used [-Werror=unused-function]
drivers/net/ethernet/intel/i40e/i40e_main.c:12185:12: error: 'i40e_suspend' defined but not used [-Werror=unused-function]
It's easier to use __maybe_unused attributes here, since you
can't pick the wrong one.
Fixes: 0e5d3da400 ("i40e: use newer generic PM support instead of legacy PM callbacks")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Acked-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Several conflicts here.
NFP driver bug fix adding nfp_netdev_is_nfp_repr() check to
nfp_fl_output() needed some adjustments because the code block is in
an else block now.
Parallel additions to net/pkt_cls.h and net/sch_generic.h
A bug fix in __tcp_retransmit_skb() conflicted with some of
the rbtree changes in net-next.
The tc action RCU callback fixes in 'net' had some overlap with some
of the recent tcf_block reworking.
Signed-off-by: David S. Miller <davem@davemloft.net>
This is very similar to the Macvlan VEPA mode, however, there is some
difference. IPvlan uses the mac-address of the lower device, so the VEPA
mode has implications of ICMP-redirects for packets destined for its
immediate neighbors sharing same master since the packets will have same
source and dest mac. The external switch/router will send redirect msg.
Having said that, this will be useful tool in terms of debugging
since IPvlan will not switch packets within its slaves and rely completely
on the external entity as intended in 802.1Qbg.
Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
IPvlan has always operated in bridge mode. However there are scenarios
where each slave should be able to talk through the master device but
not necessarily across each other. Think of an environment where each
of a namespace is a private and independant customer. In this scenario
the machine which is hosting these namespaces neither want to tell who
their neighbor is nor the individual namespaces care to talk to neighbor
on short-circuited network path.
This patch implements the mode that is very similar to the 'private' mode
in macvlan where individual slaves can send and receive traffic through
the master device, just that they can not talk among slave devices.
Signed-off-by: Mahesh Bandewar <maheshb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fixes the following sparse warnings:
drivers/net/ethernet/aquantia/atlantic/aq_ethtool.c:224:5: warning:
symbol 'aq_ethtool_get_coalesce' was not declared. Should it be static?
drivers/net/ethernet/aquantia/atlantic/aq_ethtool.c:245:5: warning:
symbol 'aq_ethtool_set_coalesce' was not declared. Should it be static?
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
bcm_sf2 and b53 replicate the same operations: clear all VLANs and set
their ports to the default VLAN tag (1 for these devices) so export the
b53 function doing just that.
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Creating a macvtap interface with the liquidio VF driver as lower device
causes this alarming message to show up in dmesg:
liquidio_link_ctrl_cmd_completion Unknown cmd 27
That's actually a false alarm because cmd 27 is the value of the macro
OCTNET_CMD_SET_UC_LIST which is known. It's a control command sent from
host to NIC firmware to set the unicast MAC address list of the macvtap
lower device.
Make the false alarm go away by adding a case for OCTNET_CMD_SET_UC_LIST
in liquidio_link_ctrl_cmd_completion().
Signed-off-by: Felix Manlunas <felix.manlunas@cavium.com>
Signed-off-by: Raghu Vatsavayi <raghu.vatsavayi@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In some cases, like internal vSwitch, the host doesn't provide
send indirection table updates. This patch sets the table to be
equal weight after subchannels are all open. Otherwise, all workload
will be on one TX channel.
As tested, this patch has largely increased the throughput over
internal vSwitch.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Check for CAP_NET_ADMIN with ns_capable() instead of capable()
to allow usage of ppp in user namespace other than the init one.
Signed-off-by: Matteo Croce <mcroce@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jeff Kirsher says:
====================
1GbE Intel Wired LAN Driver Updates 2017-10-27
This patchset is a proposal of how the Traffic Control subsystem can
be used to offload the configuration of the Credit Based Shaper
(defined in the IEEE 802.1Q-2014 Section 8.6.8.2) into supported
network devices.
As part of this work, we've assessed previous public discussions
related to TSN enabling: patches from Henrik Austad (Cisco), the
presentation from Eric Mann at Linux Plumbers 2012, patches from
Gangfeng Huang (National Instruments) and the current state of the
OpenAVNU project (https://github.com/AVnu/OpenAvnu/).
Overview
========
Time-sensitive Networking (TSN) is a set of standards that aim to
address resources availability for providing bandwidth reservation and
bounded latency on Ethernet based LANs. The proposal described here
aims to cover mainly what is needed to enable the following standards:
802.1Qat and 802.1Qav.
The initial target of this work is the Intel i210 NIC, but other
controllers' datasheet were also taken into account, like the Renesas
RZ/A1H RZ/A1M group and the Synopsis DesignWare Ethernet QoS
controller.
Proposal
========
Feature-wise, what is covered here is the configuration interfaces for
HW implementations of the Credit-Based shaper (CBS, 802.1Qav). CBS is
a per-queue shaper. Given that this feature is related to traffic
shaping, and that the traffic control subsystem already provides a
queueing discipline that offloads config into the device driver (i.e.
mqprio), designing a new qdisc for the specific purpose of offloading
the config for the CBS shaper seemed like a good fit.
For steering traffic into the correct queues, we use the socket option
SO_PRIORITY and then a mechanism to map priority to traffic classes /
Tx queues. The qdisc mqprio is currently used in our tests.
As for the CBS config interface, this patchset is proposing a new
qdisc called 'cbs'. Its 'tc' cmd line is:
$ tc qdisc add dev IFACE parent ID cbs locredit N hicredit M sendslope S \
idleslope I
Note that the parameters for this qdisc are the ones defined by the
802.1Q-2014 spec, so no hardware specific functionality is exposed here.
Per-stream shaping, as defined by IEEE 802.1Q-2014 Section 34.6.1, is
not yet covered by this proposal.
v2: Merged patch 6 of the original series into patch 4 based on feedback
from David Miller.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds checks at approprate places whether *dma_map*() call has
succeeded or not.
Original Work by: Santosh Rastapur <santosh@chelsio.com>
Signed-off-by: Arjun Vynipadath <arjun@chelsio.com>
Signed-off-by: Ganesh Goudar <ganeshgr@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Kirr: In particular with
ethtool -C <ifname> rx-usecs 0 rx-frames 0
now it is possible to disable RX delays when NIC usage requires low-latency.
See this thread for context:
https://www.spinics.net/lists/netdev/msg217665.html
My specific case is that:
We have many computers with gigabit Realtek NICs. For 2 such computers
connected to a gigabit store-and-forward switch the minimum round-trip
time for small pings (`ping -i 0 -w 3 -s 56 -q peer`) is ~ 30μs.
However it turned out that when Ethernet frame length transitions 127 ->
128 bytes (`ping -i 0 -w 3 -s {81 -> 82} -q peer`) the lowest RTT
transitions step-wise to ~ 270μs.
As David Light said this is RX interrupt mitigation done by NIC which creates
the latency. For workloads when low-latency is required with e.g. Intel,
BCM etc NIC drivers one just uses `ethtool -C rx-usecs ...` to reduce
the time NIC delays before interrupting CPU, but it turned out
`ethtool -C` is not supported by r8169 driver.
Like Stéphane ANCELOT I've traced the problem down to IntrMitigate being
hardcoded to != 0 for our chips (we have 8168 based NICs):
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/net/ethernet/realtek/r8169.c#n5460
static void rtl_hw_start_8169(struct net_device *dev) {
...
/*
* Undocumented corner. Supposedly:
* (TxTimer << 12) | (TxPackets << 8) | (RxTimer << 4) | RxPackets
*/
RTL_W16(IntrMitigate, 0x0000);
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/net/ethernet/realtek/r8169.c#n6346
static void rtl_hw_start_8168(struct net_device *dev) {
...
RTL_W16(IntrMitigate, 0x5151);
and then I've also found
https://www.spinics.net/lists/netdev/msg217665.html
and original Francois' patch:
https://www.spinics.net/lists/netdev/msg217984.htmlhttps://www.spinics.net/lists/netdev/msg218207.html
So could we please finally get support for tuning r8169 interrupt
coalescing in tree? (so that next poor soul who hits the problem does
not need to go all the way to dig into driver sources and internet
wildly and finally patch locally
-RTL_W16(IntrMitigate, 0x5151);
+RTL_W16(IntrMitigate, 0x5100);
guessing whether it is right or not and also having to care to deploy
the patch everywhere it needs to be used, etc...).
To do so I've took original Francois's patch from 2012 and reworked it a bit:
- updated to latest net-next.git;
- adjusted scaling setup based on feedback from Hayes to pick up scaling
vector depending not only on link speed but also on CPlusCmd[0:1] and to
adjust CPlusCmd[0:1] correspondingly when setting timings;
- improved a bit (I think so) error handling.
I've tested the patch on "RTL8168d/8111d" (XID 083000c0) and with it and
`ethtool -C rx-usecs 0 rx-frames 0` on both ends it improves:
- minimum RTT latency:
~270μs -> ~30μs (small packet),
~330μs -> ~110μs (full 1.5K ethernet frame)
- average RTT latency:
~480μs -> ~50μs (small packet),
~560μs -> ~125μs (full 1.5K ethernet frame)
( before:
root@neo1:# ping -i 0 -w 3 -s 82 -q neo2
PING neo2.kirr.nexedi.com (192.168.102.21) 82(110) bytes of data.
--- neo2.kirr.nexedi.com ping statistics ---
5906 packets transmitted, 5905 received, 0% packet loss, time 2999ms
rtt min/avg/max/mdev = 0.274/0.485/0.607/0.026 ms, ipg/ewma 0.508/0.489 ms
root@neo1:# ping -i 0 -w 3 -s 1472 -q neo2
PING neo2.kirr.nexedi.com (192.168.102.21) 1472(1500) bytes of data.
--- neo2.kirr.nexedi.com ping statistics ---
5073 packets transmitted, 5073 received, 0% packet loss, time 2999ms
rtt min/avg/max/mdev = 0.330/0.566/0.710/0.028 ms, ipg/ewma 0.591/0.544 ms
after:
root@neo1# ping -i 0 -w 3 -s 82 -q neo2
PING neo2.kirr.nexedi.com (192.168.102.21) 82(110) bytes of data.
--- neo2.kirr.nexedi.com ping statistics ---
45815 packets transmitted, 45815 received, 0% packet loss, time 3000ms
rtt min/avg/max/mdev = 0.036/0.051/0.368/0.010 ms, ipg/ewma 0.065/0.053 ms
root@neo1:# ping -i 0 -w 3 -s 1472 -q neo2
PING neo2.kirr.nexedi.com (192.168.102.21) 1472(1500) bytes of data.
--- neo2.kirr.nexedi.com ping statistics ---
21250 packets transmitted, 21250 received, 0% packet loss, time 3000ms
rtt min/avg/max/mdev = 0.112/0.125/0.390/0.007 ms, ipg/ewma 0.141/0.125 ms
the small -> 1.5K latency growth is understandable as it takes ~15μs
to transmit 1.5K on 1Gbps on the wire and with 2 hosts and 1 switch
and ICMP ECHO + ECHO reply the packet has to travel 4 ethernet
segments which is already 60μs;
probably something a bit else is also there as e.g. on Linux, even
with `cpupower frequency-set -g performance`, on some computers I've
noticed the kernel can be spending more time in software-only mode
when incoming packets go in less frequently. E.g. this program can
demonstrate the effect for ICMP ECHO processing:
https://lab.nexedi.com/kirr/bcc/blob/43cfc13b/tools/pinglat.py
(later this was found to be partly due to C-states exit latencies) )
We have this patch running in our testing setup for 1 months already
without any issues observed.
It remains to be clarified whether RX and TX timers use the same base.
For now I've set them equally, but Francois's original patch version
suggests it could be not the same.
I've got no feedback at all to my original posting of this patch and questions
https://www.spinics.net/lists/netdev/msg457173.html
neither from Francois, nor from any people from Realtek during one month.
So I suggest we simply apply it to net-next.git now.
Cc: Francois Romieu <romieu@fr.zoreil.com>
Cc: Hayes Wang <hayeswang@realtek.com>
Cc: Realtek linux nic maintainers <nic_swsd@realtek.com>
Cc: David Laight <David.Laight@ACULAB.COM>
Cc: Stéphane ANCELOT <sancelot@free.fr>
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Kirill Smelkov <kirr@nexedi.com>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The commit 9a393b5d59 ("tap: tap as an independent module") created a
separate tap module that implements tap functionality and exports
interfaces that will be used by macvtap and ipvtap modules to create
create respective tap devices.
However, that patch introduced a regression wherein the modules macvtap
and ipvtap can be removed (through modprobe -r) while there are
applications using the respective /dev/tapX devices. These applications
cause kernel to hold reference to /dev/tapX through 'struct cdev
macvtap_cdev' and 'struct cdev ipvtap_dev' defined in macvtap and ipvtap
modules respectively. So, when the application is later closed the
kernel panics because we are referencing KVA that is present in the
unloaded modules.
----------8<------- Example ----------8<----------
$ sudo ip li add name mv0 link enp7s0 type macvtap
$ sudo ip li show mv0 |grep mv0| awk -e '{print $1 $2}'
14:mv0@enp7s0:
$ cat /dev/tap14 &
$ lsmod |egrep -i 'tap|vlan'
macvtap 16384 0
macvlan 24576 1 macvtap
tap 24576 3 macvtap
$ sudo modprobe -r macvtap
$ fg
cat /dev/tap14
^C
<...system panics...>
BUG: unable to handle kernel paging request at ffffffffa038c500
IP: cdev_put+0xf/0x30
----------8<-----------------8<----------
The fix is to set cdev.owner to the module that creates the tap device
(either macvtap or ipvtap). With this set, the operations (in
fs/char_dev.c) on char device holds and releases the module through
cdev_get() and cdev_put() and will not allow the module to unload
prematurely.
Fixes: 9a393b5d59 (tap: tap as an independent module)
Signed-off-by: Girish Moodalbail <girish.moodalbail@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In preparation for unconditionally passing the struct timer_list pointer to
all timer callbacks, switch to using the new timer_setup() and from_timer()
to pass the timer pointer explicitly.
Cc: "David S. Miller" <davem@davemloft.net>
Cc: "yuval.shaia@oracle.com" <yuval.shaia@oracle.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Philippe Reynes <tremyfr@gmail.com>
Cc: Allen Pais <allen.lkml@gmail.com>
Cc: Tobias Klauser <tklauser@distanz.ch>
Cc: netdev@vger.kernel.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
In preparation for unconditionally passing the struct timer_list pointer to
all timer callbacks, switch to using the new timer_setup() and from_timer()
to pass the timer pointer explicitly.
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Allen Pais <allen.lkml@gmail.com>
Cc: yuan linyu <Linyu.Yuan@alcatel-sbell.com.cn>
Cc: Philippe Reynes <tremyfr@gmail.com>
Cc: netdev@vger.kernel.org
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
