Add support for configuring and managing ETHOFLD hardware queues.
Keep the queue count and MSI-X allocation scheme same as NIC queues.
ETHOFLD hardware queues are dynamically allocated/destroyed as
TC-MQPRIO Qdisc offload is enabled/disabled on the corresponding
interface, respectively.
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add logic for validation and configuration of TC-MQPRIO Qdisc
offload. Also, add support to manage EOSW_TXQ, which have 1-to-1
mapping with EOTIDs, and expose them to network stack.
Move common skb validation in Tx path to a separate function and
add minimal Tx path for ETHOFLD. Update Tx queue selection to return
normal NIC Txq to send traffic pattern that can't go through ETHOFLD
Tx path.
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Simplify queue configuration and MSI-X allocation logic. Use a single
MSI-X information table for both NIC and ULDs. Remove hard-coded
MSI-X indices for firmware event queue and non data interrupts.
Instead, use the MSI-X bitmap to obtain a free MSI-X index
dynamically. Save each Rxq's index into the MSI-X information table,
within the Rxq structures themselves, for easier cleanup.
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
QoS offload needs Ethernet Offload (ETHOFLD) resources present in the
NIC. These resources are shared with other ULDs. So, query firmware
for the available number of traffic classes, as well as, start and
end indices (EOTID) of the ETHOFLD region.
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
phylink_set_port_modes will be compiled if CONFIG_PHYLINK enabled,
dpaa2_mac_validate will be compiled if CONFIG_FSL_DPAA2_ETH enabled,
it should select CONFIG_PHYLINK when dpaa2_mac_validate call
phylink_set_port_modes
drivers/net/ethernet/freescale/dpaa2/dpaa2-mac.o: In function `dpaa2_mac_validate':
dpaa2-mac.c:(.text+0x3a1): undefined reference to `phylink_set_port_modes'
drivers/net/ethernet/freescale/dpaa2/dpaa2-mac.o: In function `dpaa2_mac_connect':
dpaa2-mac.c:(.text+0x91a): undefined reference to `phylink_create'
dpaa2-mac.c:(.text+0x94e): undefined reference to `phylink_of_phy_connect'
dpaa2-mac.c:(.text+0x97f): undefined reference to `phylink_destroy'
drivers/net/ethernet/freescale/dpaa2/dpaa2-mac.o: In function `dpaa2_mac_disconnect':
dpaa2-mac.c:(.text+0xa9f): undefined reference to `phylink_disconnect_phy'
dpaa2-mac.c:(.text+0xab0): undefined reference to `phylink_destroy'
make: *** [vmlinux] Error 1
Fixes: 7194792308 ("dpaa2-eth: add MAC/PHY support through phylink")
Signed-off-by: Chenwandun <chenwandun@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jeff Kirsher says:
====================
100GbE Intel Wired LAN Driver Updates 2019-11-06
This series contains updates to ice driver only.
Scott adds ethtool -m support so that we can read eeprom data on SFP/OSFP
modules.
Anirudh updates the return value to properly reflect when SRIOV is not
supported.
Md Fahad updates the driver to handle a change in the NVM, where the
boot configuration section was moved to the Preserved Field Area (PFA)
of the NVM.
Paul resolves an issue when DCBx requests non-contiguous TCs, transmit
hangs could occur, so configure a default traffic class (TC0) in these
cases to prevent traffic hangs. Adds a print statement to notify the
user when unsupported modules are inserted.
Bruce fixes up the driver unload code flow to ensure we do not clear the
interrupt scheme until the reset is complete, otherwise a hardware error
may occur.
Dave updates the DCB initialization to set is_sw_lldp boolean when the
firmware has been detected to be in an untenable state. This will
ensure that the firmware is in a known state.
Michal saves off the PCI state and I/O BARs address after PCI bus reset
so that after the reset, device registers can be read. Also adds a NULL
pointer check to prevent a potential kernel panic.
Mitch resolves an issue where VF's on PF's other than 0 were not seeing
resets by using the per-PF VF ID instead of the absolute VF ID.
Krzysztof does some code cleanup to remove a unneeded wrapper and
reduces the code complexity.
Brett reduces confusion by changing the name of ice_vc_dis_vf() to
ice_vc_reset_vf() to better describe what the function is actually
doing.
v2: dropped patch 3 "ice: Add support for FW recovery mode detection"
from the origin al series, while Ani makes changes based on
community feedback to implement devlink into the changes.
v3: dropped patch 1 "ice: implement set_eeprom functionality" due to a
bug found and additional changes will be needed when Ani implements
devlink in the driver.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Convert the phy_mode() function to return the if_mode through an
argument, similar to the new form of of_get_phy_mode().
This will help with handling errors in a common manner and also will fix
an always true condition.
Fixes: 0c65b2b90d ("net: of_get_phy_mode: Change API to solve int/unit warnings")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Ioana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the DMA memory resource get failed case, the error is not
set and 0 will be returned. Fix it by removing redundant check
since devm_ioremap_resource() will handle it.
Fixes: 28ef9ebdb6 ("net: axienet: make use of axistream-connected attribute optional")
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Reviewed-by: Radhey Shyam Pandey <radhey.shyam.pandey@xilinx.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Function ptp_clock_register() returns ERR_PTR() and never returns
NULL. The NULL test should be removed.
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Acked-by: Igor Russkikh <irusskikh@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When the device is not capable of supporting SR-IOV -ENODEV is being
returned; -EOPNOTSUPP is more appropriate.
Signed-off-by: Anirudh Venkataramanan <anirudh.venkataramanan@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
ice_vc_dis_vf() tells iavf that it's going to perform a reset
and then performs a software reset. This is misleading based on
the function name because the VF does not get disabled. So fix
this by changing the name to ice_vc_reset_vf().
Signed-off-by: Brett Creeley <brett.creeley@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
ice_cleanup_hdrs() has been stripped of most of its content, it only serves
as a wrapper for eth_skb_pad(). We can get rid of it altogether and
simplify the codebase.
Signed-off-by: Krzysztof Kazimierczak <krzysztof.kazimierczak@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Print message to inform user of PCI link speed and width.
Signed-off-by: Paul Greenwalt <paul.greenwalt@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Print message to inform user if unsupported module is inserted, and
extend the topology / configuration detection.
Signed-off-by: Paul Greenwalt <paul.greenwalt@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The VF_MBX_ARQLEN register array is per-PF, not global, so we should not
use the absolute VF ID as an index. Instead, use the per-PF VF ID.
This fixes an issue with VFs on PFs other than 0 not seeing reset.
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Without this check rebuild vsi can lead to kernel panic.
Signed-off-by: Michal Swiatkowski <michal.swiatkowski@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Save state to correct recovery memory and I/O BARs address
after PCI bus reset. Without this after reset kernel can't
read device registers.
Signed-off-by: Michal Swiatkowski <michal.swiatkowski@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Adjust ice_init_dcb to set the is_sw_lldp boolean
in the case where the FW has been detected to be
in an untenable state such that the driver
should forcibly make sure it is off.
This will ensure that the FW is in a known state.
Signed-off-by: Dave Ertman <david.m.ertman@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
As part of the driver unload flow, a PF reset is issued which may still
cause an interrupt to be generated by the device. Do not clear the
interrupt scheme until the reset is complete and there are no pending
transactions otherwise a hardware error may occur.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
If DCBx request non-contiguous TCs, then the driver will configure default
traffic class (TC0). This is done to prevent Tx hang since the driver
currently does not support non-contiguous TC.
Signed-off-by: Paul Greenwalt <paul.greenwalt@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The Boot Configuration Section Block has been moved to the Preserved Field
Area (PFA) of NVM. Update the NVM reads that involves Boot Configuration
Section.
Signed-off-by: Md Fahad Iqbal Polash <md.fahad.iqbal.polash@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Implement ethtool -m support to read eeprom data from SFP/QSFP modules.
Signed-off-by: Scott W Taylor <scott.w.taylor@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Until now the size of a Rx buffer was artificially limited
to 1536B (which happens to be the default, after reset, hardware
value for a Rx buffer). This approach however leaves unused
memory space for Rx packets, since the driver uses a paged
allocation scheme that reserves half a page for each Rx skb.
There's also the inconvenience that frames around 1536 bytes
can get scattered if the limit is slightly exceeded. This limit
can be exceeded even for standard MTU of 1500B traffic, for common
cases like stacked VLANs, or DSA tags.
To address these issues, let's just compute the buffer size
starting from the upper limit of 2KB (half a page) and
subtract the skb overhead and alignment restrictions.
Signed-off-by: Claudiu Manoil <claudiu.manoil@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The function page_shift() is supported after the commit 94ad933810
("mm: introduce page_shift()").
So replace with page_shift() in ehea_is_hugepage() for readability.
Signed-off-by: Yunfeng Ye <yeyunfeng@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This change adds support for xmit_more based on the igb commit 6f19e12f62
("igb: flush when in xmit_more mode and under descriptor pressure") and
commit 6b16f9ee89 ("net: move skb->xmit_more hint to softnet data") that
were made to igb to support this feature. The function netif_xmit_stopped
is called to check whether transmit queue on device is currently unable to
send to determine whether we must write the tail because we can add no
further buffers.
When normal packets and/or xmit_more packets fill up tx_desc, it is
necessary to trigger NIC tx reg.
Following the advice from David Miller and Jakub Kicinski, after the
xmit_more feature is added, the following scenario will occur.
|
xmit_more packets
|
DMA_MAPPING
|
DMA_MAPPING error check
|
xmit_more packets already in HW xmit queue
|
In the above scenario, if DMA_MAPPING error occurrs, the xmit_more packets
already in HW xmit queue will also be dropped. This is different from the
behavior before xmit_more feature. So it is necessary to trigger NIC HW tx
reg in the above scenario.
To the non-xmit_more packets, the above scenario will not occur.
Tested:
- pktgen (xmit_more packets) SMP x86_64 ->
Test command:
./pktgen_sample03_burst_single_flow.sh ... -b 8 -n 1000000
Test results:
Params:
...
burst: 8
...
Result: OK: 12194004(c12188996+d5007) usec, 1000001 (1500byte,0frags)
82007pps 984Mb/sec (984084000bps) errors: 0
- iperf (normal packets) SMP x86_64 ->
Test command:
Server: iperf -s
Client: iperf -c serverip
Result:
TCP window size: 85.0 KByte (default)
------------------------------------------------------------
[ ID] Interval Transfer Bandwidth
[ 3] 0.0-10.0 sec 1.10 GBytes 942 Mbits/sec
CC: Joe Jin <joe.jin@oracle.com>
CC: JUNXIAO_BI <junxiao.bi@oracle.com>
Reported-and-tested-by: Nan san <nan.1986san@gmail.com>
Signed-off-by: Zhu Yanjun <yanjun.zhu@oracle.com>
Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch implements reset_prepare and reset_done, which are used
for handling FLR.
Signed-off-by: Vishal Kulkarni <vishal@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The new FW has added extra validation for HSI version to
make FW backward compatible with older VF drivers. Hence
set fp_hsi_ver to Fast Path HSI version of the FW in use.
Signed-off-by: Manish Rangankar <mrangankar@marvell.com>
Signed-off-by: Nilesh Javali <njavali@marvell.com>
Signed-off-by: Manish Chopra <manishc@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
PF driver doesn't enable tx-switching for all cos queues/clients,
which causes packets drop from PF to VF. Fix this by enabling
tx-switching on all cos queues/clients.
Signed-off-by: Manish Chopra <manishc@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
FW version 7.13.15 addresses the issue in Multi-cos implementation.
This patch re-enables the Multi-Cos support in the driver.
Fixes: d1f0b5dce8 ("bnx2x: Disable multi-cos feature.")
Signed-off-by: Sudarsana Reddy Kalluru <skalluru@marvell.com>
Signed-off-by: Ariel Elior <aelior@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 97a27d6d6e8d "bnx2x: Add FW 7.13.15.0" added said .bin FW to
linux-firmware tree. This FW addresses few important issues in the earlier
FW release.
This patch incorporates FW 7.13.15.0 in the bnx2x driver.
Signed-off-by: Sudarsana Reddy Kalluru <skalluru@marvell.com>
Signed-off-by: Ariel Elior <aelior@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pass a phy_interface_t to of_get_phy_mode(), by changing the type of
phy_mode in the device structure. This then requires that
zmii_attach() is also changes, since it takes a pointer to phy_mode.
Fixes: 0c65b2b90d ("net: of_get_phy_mode: Change API to solve int/unit warnings")
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
The macros HCLGE_MPF_ENBALE and HCLGEVF_MPF_ENBALE are defined but never
used. I was going to fix the spelling mistake "ENBALE" -> "ENABLE" but
found these macros are not used, so they can be removed.
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jeff Kirsher says:
====================
100GbE Intel Wired LAN Driver Updates 2019-11-04
This series contains updates to the ice driver only.
Anirudh refactors the code to reduce the kernel configuration flags and
introduces ice_base.c file.
Maciej does additional refactoring on the configuring of transmit
rings so that we are not configuring per each traffic class flow.
Added support for XDP in the ice driver. Provides additional
re-organizing of the code in preparation for adding build_skb() support
in the driver. Adjusted the computational padding logic for headroom
and tailroom to better support build_skb(), which also aligns with the
logic in other Intel LAN drivers. Added build_skb support and make use
of the XDP's data_meta.
Krzysztof refactors the driver to prepare for AF_XDP support in the
driver and then adds support for AF_XDP.
v2: Updated patch 3 of the series based on community feedback with the
following changes...
- return -EOPNOTSUPP instead of ENOTSUPP for too large MTU which makes
it impossible to attach XDP prog
- don't check for case when there's no XDP prog currently on interface
and ice_xdp() is called with NULL bpf_prog; this happens when user
does "ip link set eth0 xdp off" and no prog is present on VSI; no need
for that as it is handled by higher layer
- drop the extack message for unknown xdp->command
- use the smp_processor_id() for accessing the XDP Tx ring for XDP_TX
action
- don't leave the interface in downed state in case of any failure
during the XDP Tx resources handling
- undo rename of ice_build_ctob
The above changes caused a ripple effect in patches 4 & 5 to update
references to ice_build_ctob() which are now build_ctob()
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Jeff Kirsher says:
====================
10GbE Intel Wired LAN Driver Updates 2019-11-04
This series contains old Halloween candy updates, yet still sweet, to
fm10k, ixgbe and i40e.
Jake adds the missing initializers for a couple of the TLV attribute
macros. Added support for capturing and reporting statistics for all of
the VFs in a given PF. Lastly, bump the version of the fm10k driver to
reflect the recent changes.
Alex addresses locality issues in the ixgbe driver when it is loaded on
a system supporting multiple NUMA nodes.
Manjunath Patil provides changes to the ixgbe driver, similar to those
made to igb, to prevent transmit packets to request a hardware timestamp
when the NIC has not been setup via the SIOCSHWTSTAMP ioctl.
Alice adds support for x710 by adding the missing device id's in the
appropriate places to ensure all the features are enabled in i40e.
Jesse adds support for VF stats gathering in the i40e via the kernel
via ndo_get_vf_stats function.
v2: Fixed up commit id references in patch 5's description to align with
how commit id's should be referenced.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Implement the VF stats gathering via the kernel via ndo_get_vf_stats().
The driver will show per-VF stats in the output of the command:
ip -s link show dev <PF>
Testing Hints:
ip -s link show dev eth0
will return non-zero VF stats.
...
vf 0 MAC 00:55:aa:00:55:aa, spoof checking on, link-state enable, trust off
RX: bytes packets mcast bcast
128000 1000 104 104
TX: bytes packets
128000 1000
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>
The I40E_DEV_ID_10G_BASE_T_BC device id was added previously,
but was not enabled in all the appropriate places. Adding it
to enable it's use.
Signed-off-by: Alice Michael <alice.michael@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
HW timestamping can only be requested for a packet if the NIC is first
setup via ioctl(SIOCSHWTSTAMP). If this step was skipped, then the ixgbe
driver still allowed TX packets to request HW timestamping. In this
situation, we see 'clearing Tx Timestamp hang' noise in the log.
Fix this by checking that the NIC is configured for HW TX timestamping
before accepting a HW TX timestamping request.
Similar-to:
commit 26bd4e2db0 ("igb: protect TX timestamping from API misuse")
commit 0a6f2f05a2 ("igb: Fix a test with HWTSTAMP_TX_ON")
Signed-off-by: Manjunath Patil <manjunath.b.patil@oracle.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
An upcoming out-of-tree release will be occurring which will include the
recent functionality to support virtual function statistics. Update the
kernel driver version to match this.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
This patch is meant to address locality issues present in the ixgbe driver
when it is loaded on a system supporting multiple NUMA nodes and more CPUs
then the device can map in a 1:1 fashion. Instead of just arbitrarily
mapping itself to CPUs 0-62 it would make much more sense to map itself to
the local CPUs first, and then map itself to any remaining CPUs that might
be used.
The first effect of this is that queue 0 should always be allocated on the
local CPU/NUMA node. This is important as it is the default destination if
a packet doesn't match any existing flow director filter or RSS rule and as
such having it local should help to reduce QPI cross-talk in the event of
an unrecognized traffic type.
In addition this should increase the likelihood of the RSS queues being
allocated and used on CPUs local to the device while the ATR/Flow Director
queues would be able to route traffic directly to the CPU that is likely to
be processing it.
Signed-off-by: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Support capturing and reporting statistics for all of the VFs associated
with a given PF device via the ndo_get_vf_stats callback.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Add the missing field initializers for a couple of the TLV attribute
macros. This resolves the last few -Wmissing-field-initializers warnings
for the fm10k Linux driver.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
At this point ice driver is able to work on order 1 pages that are split
onto two 3k buffers. Let's reflect that when user is setting new MTU
size and XDP is present on interface.
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Driver is now prepared for building the skb around the existing Rx
buffer, so introduce the ice_build_skb responsible for it. Make use of
XDP's data_meta as well.
I've observed around 30% less CPU consumption with build_skb Rx path, in
comparison to legacy Rx. What stands behind such result is the avoidance
of flow_dissector (which we were diving into via eth_get_headlen) and no
memcpy calls.
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Take into account the underlying architecture specific settings and
based on that calculate the possible padding that can be supplied.
Typically, for x86 and standard MTU size we will end up with 192 bytes
of headroom. This is the same behavior as our other drivers have and we
can dedicate it for XDP purposes.
Furthermore, introduce the Rx ring flag for indicating whether build_skb
is used on particular. Based on that invoke the routines for padding
calculation.
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Add an ethtool "legacy-rx" priv flag for toggling the Rx path. This
control knob will be mainly used for build_skb usage as well as buffer
size/MTU manipulation.
In preparation for adding build_skb support in a way that it takes
care of how we set the values of max_frame and rx_buf_len fields of
struct ice_vsi. Specifically, in this patch mentioned fields are set to
values that will allow us to provide headroom and tailroom in-place.
This can be mostly broken down onto following:
- for legacy-rx "on" ethtool control knob, old behaviour is kept;
- for standard 1500 MTU size configure the buffer of size 1536, as
network stack is expecting the NET_SKB_PAD to be provided and
NET_IP_ALIGN can have a non-zero value (these can be typically equal
to 32 and 2, respectively);
- for larger MTUs go with max_frame set to 9k and configure the 3k
buffer in case when PAGE_SIZE of underlying arch is less than 8k; 3k
buffer is implying the need for order 1 page, so that our page
recycling scheme can still be applied;
With that said, substitute the hardcoded ICE_RXBUF_2048 and PAGE_SIZE
values in DMA API that we're making use of with rx_ring->rx_buf_len and
ice_rx_pg_size(rx_ring). The latter is an introduced helper for
determining the page size based on its order (which was figured out via
ice_rx_pg_order). Last but not least, take care of truesize calculation.
In the followup patch the headroom/tailroom computation logic will be
introduced.
This change aligns the buffer and frame configuration with other Intel
drivers, most importantly with iavf.
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Add zero copy AF_XDP support. This patch adds zero copy support for
Tx and Rx; code for zero copy is added to ice_xsk.h and ice_xsk.c.
For Tx, implement ndo_xsk_wakeup. As with other drivers, reuse
existing XDP Tx queues for this task, since XDP_REDIRECT guarantees
mutual exclusion between different NAPI contexts based on CPU ID. In
turn, a netdev can XDP_REDIRECT to another netdev with a different
NAPI context, since the operation is bound to a specific core and each
core has its own hardware ring.
For Rx, allocate frames as MEM_TYPE_ZERO_COPY on queues that AF_XDP is
enabled.
Signed-off-by: Krzysztof Kazimierczak <krzysztof.kazimierczak@intel.com>
Co-developed-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
In preparation of AF XDP, move functions that will be used both by skb and
zero-copy paths to a new file called ice_txrx_lib.c. This allows us to
avoid using ifdefs to control the staticness of said functions.
Move other functions (ice_rx_csum, ice_rx_hash and ice_ptype_to_htype)
called only by the moved ones to the new file as well.
Signed-off-by: Krzysztof Kazimierczak <krzysztof.kazimierczak@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Building with W=1 (cf.scripts/Makefile.extrawarn) outputs:
warning: variable ‘ret’ set but not used [-Wunused-but-set-variable]
Drop the unused 'ret' variable.
Signed-off-by: Christophe Roullier <christophe.roullier@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
IOC3 hardware needs a 16k aligned TX ring.
Signed-off-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Set NETIF_F_HIGHDMA together with the NETIF_F_IP_CSUM flag instead of
letting the second assignment overwrite it. Probably doesn't matter
in practice as none of the systems an IOC3 is usually found in has
highmem to start with.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
