Do not lock the register structure in IRQ context. It is not
necessary and leads to lockups under SMP load.
Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
Rename the SoC-specific rtl838x_reg structure in the Ethernet
driver to avoid confusion with the structure of the same name
in the DSA driver. New name is: rtl838x_eth_reg
Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
Setting bits 20 and 23 in a u16 is obviously wrong.
According to https://www.svanheule.net/realtek/cypress/cputag
cpu_tag[2] starts at bit 48 in the cpu-tag structure, so
bit 43 is bit 5 in cpu_tag[2] and bit 40 is bit 8 in
cpu_tag[2].
Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
Set CONFIG_FORCE_MAX_ZONEORDER setting to 13 to allow larger
contiguous memory allocation for the DMA of the Ethernet
driver. Increase the number of entries in the RX ring
to 300 making use of the larger DMA region now possible for
receiveing packets.
Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
Creates RTL83XX as a basic kernel config parameter for the
RTL838X, RTL839x, RTL930X and RTL931X platforms with respective
configurations for the SoCs, which are introduced in addition.
Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
Recent versions of Realtek's SDK reset both the ethernet NIC and queues
(SW_NIC_RST and SW_Q_RST bits) when initialising the hardware.
Furthermore, when issuing a CPU reset on the Zyxel GS1900-8 (not
supported by any current driver), the networking part of the SoC is not
reset. This leads to unresponsive network after the restart. By
resetting both the ethernet NIC and queues, networking always comes up
reliably.
Suggested-by: Birger Koblitz <mail@birger-koblitz.de>
Signed-off-by: Sander Vanheule <sander@svanheule.net>
Tested-by: Stijn Tintel <stijn@linux-ipv6.be>
Commit 03e1d93e07 ("realtek: add driver support for routing
offload") added routing offload for IPv4, but broke IPv6 routing
completely. The routing table is empty and cannot be updated:
root@gs1900-10hp:~# ip -6 route
root@gs1900-10hp:~# ip -6 route add unreachable default
RTNETLINK answers: Invalid argument
As a side effect, this breaks opkg on IPv4 only systems too,
since uclient-fetch fails when there are no IPv6 routes:
root@gs1900-10hp:~# uclient-fetch http://192.168.99.1
Downloading 'http://192.168.99.1'
Failed to send request: Operation not permitted
Fix by returning NOTIFY_DONE when offloading is unsupported, falling
back to default behaviour.
Fixes: 03e1d93e07 ("realtek: add driver support for routing offload")
Signed-off-by: Bjørn Mork <bjorn@mork.no>
RTL8393 SoCs older than Revision C hang on accesses to PHYs with PHY address
larger or equal to the CPU-port (52). This will make scanning the MDIO bus
hang forever. Since the RTL8390 platform does not support more than
52 PHYs, return -EIO for phy addresses >= 52. Note that the RTL8390 family
of SoCs has a fixed mapping between port number and PHY-address.
Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
Adds SoC specific routing offload implementations for
RTL8380/90 and RTL9300. RTL83xx supports merely nexthop
routing, RTL9300 full host and prefix routes.
Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
Add generic support for listening to FIB and Event notifier updates and
use this information to hook into the L3 hardware capabilities of the
RTL SoCs.
Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
The ingress filter registers use 2 bits for each port to define the filtering
state, whereas the egress filter uses 1 bit. So for for the ingress filter
the register offset for a given port is:
(port >> 4) << 4: since there are 16 entries in a register of 32 bits
and for the egress filter:
(port >> 5) << 4: since there are 32 entries in a register of 32 bits
Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
Configure a sane L2 learning configuration upon DSA driver load so that the
switch can start learning L2 addresses. Also configure the correct flood masks
for broadcast and unknown unicast traffice.
Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
This adds RTL93xx-specific MAC configuration routines that allow also configuration
of 10GBit links for phylink. There is support for the Realtek-specific HISGMI
protocol.
Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
This adds support for offloading TC flower by using the Packet Inspection Engine
of the RTL-SoCs. Basic infrastructure support is provide with callbacks to the
tc subsystem and support for HW packet counters.
Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
All RTL SoCs addresss PHYs via their port number, which is mapped to an
SMI address. Add support for configuring this mapping via the .dts on all
SoCs apart from the 839x, where the mapping to the 64 ports is fixed.
Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
On RTL83xx enable learning of the MAC source address of the CPU port
from outgoing packets. Add documentation on bit fields. On RTL93xx
enable port-mask usage and the use of internal priority, these
SoCs automatically learn the MAC.
Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
Remove the storm control and attack warnings from the IRQ handler
of the Ethernet driver. There was no consequence to the detection
and the kernel can also handle at least the attacks itself.
Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
This enlarges the size of the TX ring buffer, which prevents warnings
when the buffer runs out of space.
Signed-off-by: Birger Koblitz <git@birger-koblitz.de>
dsa_to_port function in 5.10 returns dsa_port from the port list in
dsa_switch_tree, but the tree is built when the switch is registered
by dsa_register_switch and it's null in rtl83xx_mdio_probe.
So, we need to use dsa_to_port after the registration of the switch.
Signed-off-by: INAGAKI Hiroshi <musashino.open@gmail.com>
this patch fixes the following errors when compiling:
- dsa_switch_alloc is removed[1]
- a parameter "enum dsa_tag_protocol mprot" is added to dsa_tag_protocol
in dsa_switch_ops (include/net/dsa.h)
- several paramters are added to "phylink_mac_link_up" in dsa_switch_ops
(include/net/dsa.h)
added:
- int speed
- int duplex
- bool tx_pause
- bool rx_pause
- a parameter "struct switchdev_trans *trans" is added to
port_vlan_filtering in dsa_switch_ops (include/net/dsa.h)
[1]: https://lore.kernel.org/lkml/20191020031941.3805884-17-vivien.didelot@gmail.com/
Signed-off-by: INAGAKI Hiroshi <musashino.open@gmail.com>
this patch fixes the following errors when compiling:
- "unsigned int txqueue" is added as an additional parameter of
ndo_tx_timeout in net_device_ops (include/linux/netdevice.h)
- "mac_link_state" in phylink_mac_ops (include/linux/phylink.h)
is renamed to "mac_pcs_get_state" and changed the return value
to void from int
- several parameters are added to "mac_link_up" in phylink_mac_ops
(include/linux/phylink.h) and the order of the parameters is
changed
added:
- int speed
- int duplex
- bool tx_pause
- bool rx_pause
- a parameter "phy_interface_t *interface" is added to of_get_phy_mode
(drivers/of/of_net.c) and returns the state instead of phy mode
Signed-off-by: INAGAKI Hiroshi <musashino.open@gmail.com>
this patch copies the following files from 5.4 to 5.10:
- config-5.4 -> config-5.10
- files-5.4/ -> files-5.10/
- patches-5.4/ -> patches-5.10/
Signed-off-by: INAGAKI Hiroshi <musashino.open@gmail.com>
[rebase on change in files-5.4]
Signed-off-by: Adrian Schmutzler <freifunk@adrianschmutzler.de>
