Commit 5fabb01207 ("net: stmmac: Add initial XDP support") sets
PP_FLAG_DMA_SYNC_DEV flag for page_pool unconditionally,
page_pool_recycle_direct() will call page_pool_dma_sync_for_device()
on every page even the page is not going to be reused by XDP program.
When XDP is not enabled, the page which holds the received buffer
will be recycled once the buffer is copied into new SKB by
skb_copy_to_linear_data(), then the MAC core will never reuse this
page any longer. Always setting PP_FLAG_DMA_SYNC_DEV wastes CPU cycles
on unnecessary calling of page_pool_dma_sync_for_device().
After this patch, up to 9% noticeable performance improvement was observed
on certain platforms.
Fixes: 5fabb01207 ("net: stmmac: Add initial XDP support")
Signed-off-by: Furong Xu <0x1207@gmail.com>
Link: https://patch.msgid.link/20240919121028.1348023-1-0x1207@gmail.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The RX frame size limit should not be based on the current MTU setting.
Instead it should be based on the hardware capabilities.
While we're here, improve the description of the receive frame length
setting as suggested by Niklas.
Fixes: c156633f13 ("Renesas Ethernet AVB driver proper")
Reviewed-by: Sergey Shtylyov <s.shtylyov@omp.ru>
Reviewed-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
Signed-off-by: Paul Barker <paul.barker.ct@bp.renesas.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The datasheets for all SoCs using the GbEth IP specify a maximum
transmission frame size of 1.5 kByte. I've confirmed through internal
discussions that support for 1522 byte frames has been validated, which
allows us to support the default MTU of 1500 bytes after reserving space
for the Ethernet header, frame checksums and an optional VLAN tag.
Fixes: 2e95e08ac0 ("ravb: Add rx_max_buf_size to struct ravb_hw_info")
Reviewed-by: Niklas Söderlund <niklas.soderlund+renesas@ragnatech.se>
Reviewed-by: Sergey Shtylyov <s.shtylyov@omp.ru>
Signed-off-by: Paul Barker <paul.barker.ct@bp.renesas.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The commit b8c43360f6 ("net: stmmac: No need to calculate speed divider
when offload is disabled") allows the "port_transmit_rate_kbps" to be
set to a value of 0, which is then passed to the "div_s64" function when
tc-cbs is disabled. This leads to a zero-division error.
When tc-cbs is disabled, the idleslope, sendslope, and credit values the
credit values are not required to be configured. Therefore, adding a return
statement after setting the txQ mode to DCB when tc-cbs is disabled would
prevent a zero-division error.
Fixes: b8c43360f6 ("net: stmmac: No need to calculate speed divider when offload is disabled")
Cc: <stable@vger.kernel.org>
Co-developed-by: Choong Yong Liang <yong.liang.choong@linux.intel.com>
Signed-off-by: Choong Yong Liang <yong.liang.choong@linux.intel.com>
Signed-off-by: KhaiWenTan <khai.wen.tan@linux.intel.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20240918061422.1589662-1-khai.wen.tan@linux.intel.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
In the ether3_probe function, a timer is initialized with a callback
function ether3_ledoff, bound to &prev(dev)->timer. Once the timer is
started, there is a risk of a race condition if the module or device
is removed, triggering the ether3_remove function to perform cleanup.
The sequence of operations that may lead to a UAF bug is as follows:
CPU0 CPU1
| ether3_ledoff
ether3_remove |
free_netdev(dev); |
put_devic |
kfree(dev); |
| ether3_outw(priv(dev)->regs.config2 |= CFG2_CTRLO, REG_CONFIG2);
| // use dev
Fix it by ensuring that the timer is canceled before proceeding with
the cleanup in ether3_remove.
Fixes: 6fd9c53f71 ("net: seeq: Convert timers to use timer_setup()")
Signed-off-by: Kaixin Wang <kxwang23@m.fudan.edu.cn>
Link: https://patch.msgid.link/20240915144045.451-1-kxwang23@m.fudan.edu.cn
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
axienet_free_tx_chain returns the number of DMA descriptors it's
handled. However, axienet_tx_poll treats the return as the number of
packets. When scatter-gather SKBs are enabled, a single packet may use
multiple DMA descriptors, which causes incorrect packet counts. Fix this
by explicitly keepting track of the number of packets processed as
separate from the DMA descriptors.
Budget does not affect the number of Tx completions we can process for
NAPI, so we use the ring size as the limit instead of budget. As we no
longer return the number of descriptors processed to axienet_tx_poll, we
now update tx_bd_ci in axienet_free_tx_chain.
Fixes: 8a3b7a252d ("drivers/net/ethernet/xilinx: added Xilinx AXI Ethernet driver")
Signed-off-by: Sean Anderson <sean.anderson@linux.dev>
Link: https://patch.msgid.link/20240913145156.2283067-1-sean.anderson@linux.dev
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The HSR stack allows to offload its Tx packet duplication functionality to
the hardware. Enable this offloading feature for ICSSG driver. Add support
to offload HSR Tx Tag Insertion and Rx Tag Removal and duplicate discard.
hsr tag insertion offload and hsr dup offload are tightly coupled in
firmware implementation. Both these features need to be enabled / disabled
together.
Duplicate discard is done as part of RX tag removal and it is
done by the firmware. When driver sends the r30 command
ICSSG_EMAC_HSR_RX_OFFLOAD_ENABLE, firmware does RX tag removal as well as
duplicate discard.
Signed-off-by: Ravi Gunasekaran <r-gunasekaran@ti.com>
Signed-off-by: MD Danish Anwar <danishanwar@ti.com>
Reviewed-by: Roger Quadros <rogerq@kernel.org>
Link: https://patch.msgid.link/20240911081603.2521729-5-danishanwar@ti.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add support for offloading HSR port-to-port frame forward to hardware.
When the slave interfaces are added to the HSR interface, the PRU cores
will be stopped and ICSSG HSR firmwares will be loaded to them.
Similarly, when HSR interface is deleted, the PRU cores will be
restarted and the last used firmwares will be reloaded. PRUeth
interfaces will be back to the last used mode.
This commit also renames some APIs that are common between switch and
hsr mode with '_fw_offload' suffix.
Signed-off-by: MD Danish Anwar <danishanwar@ti.com>
Reviewed-by: Roger Quadros <rogerq@kernel.org>
Link: https://patch.msgid.link/20240911081603.2521729-4-danishanwar@ti.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Display channel info in the RPM map debugfs output.
With this, cat /sys/kernel/debug/cn10k/rvu_pf_rpm_map
would display channel number for each device in addition to
the existing data.
Sample output:
PCI dev RVU PF Func NIX block rpm LMAC CHAN
0002:02:00.0 0x400 NIX0 rpm0 LMAC0 256
Signed-off-by: Linu Cherian <lcherian@marvell.com>
Link: https://patch.msgid.link/20240912161450.164402-3-lcherian@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The enetc driver uses ifdefs when checking whether
CONFIG_FSL_ENETC_PTP_CLOCK is enabled in a number of places. This works
if the driver is built-in but fails if the driver is available as a
kernel module. Replace the instances of ifdef with use of the IS_ENABLED
macro, that will evaluate as true when this feature is built as a kernel
module and follows the kernel's coding style.
Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Signed-off-by: Martyn Welch <martyn.welch@collabora.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20240912173742.484549-1-martyn.welch@collabora.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The driver calls netif_napi_set_irq() and then calls netif_napi_add(),
which calls netif_napi_add_weight(). At the end of
netif_napi_add_weight() is a call to netif_napi_set_irq(napi, -1), which
clears the previously set napi->irq value. Fix this by calling
netif_napi_set_irq() after calling netif_napi_add().
This was found when reviewing another patch and I have no way to test
this, but the fix seemed relatively straight forward.
Fixes: bc6107771b ("eth: fbnic: Allocate a netdevice and napi vectors with queues")
Signed-off-by: Brett Creeley <brett.creeley@amd.com>
Reviewed-by: Joe Damato <jdamato@fastly.com>
Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Link: https://patch.msgid.link/20240912174922.10550-1-brett.creeley@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
En-Wei reported that traffic breaks if cable is unplugged for more
than 3s and then re-plugged. This was supposed to be fixed by
621735f590 ("r8169: fix rare issue with broken rx after link-down on
RTL8125"). But apparently this didn't fix the issue for everybody.
The 3s threshold rang a bell, as this is the delay after which ALDPS
kicks in. And indeed disabling ALDPS fixes the issue for this user.
Maybe this fixes the issue in general. In a follow-up step we could
remove the first fix attempt and see whether anybody complains.
Fixes: f1bce4ad2f ("r8169: add support for RTL8125")
Tested-by: En-Wei WU <en-wei.wu@canonical.com>
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Link: https://patch.msgid.link/778b9d86-05c4-4856-be59-cde4487b9e52@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Now that we support multiple RX queues, enable default priority
to flow mapping so that higher priority packets come on higher
channels (flows).
The Classifier checks for PCP/DSCP priority in the packet and
routes them to the appropriate flow.
Signed-off-by: Roger Quadros <rogerq@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The Policer registers in the ALE register space are just shadow registers
and use an index field in the policer table control register to read/write
to the actual Polier registers.
Add helper functions to Read and Write to Policer registers.
Also add a helper function to set the thread value to classifier/policer
mapping. Any packet that first matches the classifier will be sent to the
thread (flow) that is set in the classifier to thread mapping table.
If not set then it goes to the default flow.
Default behaviour is to have 8 classifiers to map 8 DSCP/PCP
priorities to N receive threads (flows). N depends on number of
RX channels enabled for the port.
As per the standard [1] User prioritie 1 (Background) and 2 (Spare) have
lower priority than the user priority 0 (default). User priority 1 being
of the lowest priority.
[1] IEEE802.1D-2004, IEEE Standard for Local and metropolitan area networks
Table G-2 - Traffic type acronyms
Table G-3 - Defining traffic types
Signed-off-by: Roger Quadros <rogerq@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Adds regfileds for Policer registers and Thread mapping/control registers.
Signed-off-by: Roger Quadros <rogerq@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use regfields for number of ALE Entries and Policers.
The variants that support Policers/Classifiers have the number
of policers encoded in the ALE_STATUS register.
Use that and show the number of Policers in the ALE info message.
Signed-off-by: Roger Quadros <rogerq@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Map the entire ALE registerspace using regmap.
Add regfields for Major and Minor Version fields.
Signed-off-by: Roger Quadros <rogerq@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
am65-cpsw can support up to 8 queues at Rx.
Use a macro AM65_CPSW_MAX_RX_QUEUES to indicate that.
As there is only one DMA channel for RX traffic, the
8 queues come as 8 flows in that channel.
By default, we will start with 1 flow as defined by the
macro AM65_CPSW_DEFAULT_RX_CHN_FLOWS.
User can change the number of flows by ethtool like so
'ethtool -L ethx rx <N>'
All traffic will still come on flow 0. To get traffic on
different flows the Classifiers will need to be set up.
Signed-off-by: Roger Quadros <rogerq@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
