If transmissions are done outside of the DSA switch (directly from the CPU
port), the STP state must not block the transmission. Otherwise, STP frames
are not correctly submitted and the STP frames cannot correctly detect
loops before switching a port in the forwarding state.
The same applies for the LLDP frames. These must be submitted independent
of the STP state to identify neighbors or configure POE limits.
It is not necessary to filter specific destination mac addresses because
the transmission was done outside the bridge/switch in the first place. The
transmission is therefore forced.
Signed-off-by: Issam Hamdi <ih@simonwunderlich.de>
Co-developed-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Sven Eckelmann <sven@narfation.org>
Link: https://github.com/openwrt/openwrt/pull/20184
Signed-off-by: Robert Marko <robimarko@gmail.com>
The RTL931x has next to its SPI flash controller a SPI master interface. It
is connected to
* SPI_CS#[1,0]: AH22 , AK22 (aka: GPIO 12, 11)
* SPI_CLK: AL23 (aka: GPIO 8)
* SPI_MISO: AM23 (aka: GPIO 9)
* SPI_MOSI: AL22 (aka: GPIO 10)
It is not the same as the SPI flash controller which uses pins:
* SPI_CS#[1,0]: B24, A24
* SPI_SCLK: A23
* SPI_SDI/SIO0: B21
* SPO_SDO_SIO1: B21
* SPI_SIO2: A22
* SPI_SIO3: B22
* SPI_RSTN: B23
As shown above, the SPI master controller shares its pin with GPIO 8, 9,
10, 11, 12. In some upcoming devices (like the Plasma Cloud PSX28/ESX28),
they will be used for SFP cage signaling. These pins must therefore be
switched manually to the GPIO mode.
The SPI_CTRL0 register provides all necessary configuration to enforce the
GPIO mode of the pins. And until more requirements (and a correct driver)
for the SPI master controller arise, it is therefore possible to use
pinctrl-single to configure it using the devicetree.
Previously the ethernet driver did configure the SPI master controller for
31.25 MHz. It is unknown for which kind of device this was originally made
and what was actually connected there. But this manual write to the
register conflicts potentially with the write of the pinctrl driver to the
same register. Luckily, we don't need this SPI speed configuration in the
ethernet driver. Still, to allow this device an easy migration, the
`spi0-31mhz` configuration was already prepared.
Signed-off-by: Sven Eckelmann <se@simonwunderlich.de>
Link: https://github.com/openwrt/openwrt/pull/20263
Signed-off-by: Robert Marko <robimarko@gmail.com>
There is some open coding in the ethernet driver. Drop
that and use kernel helpers instead.
- Use napi_gro_receive() instead of local skb list
- Use skb_put_data() instead of skb_put() plus memcpy()
- Use netdev_alloc_skb_ip_align() instead of manual alignment
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/20030
Signed-off-by: Robert Marko <robimarko@gmail.com>
So much code was distributed between phy/ethernet/dsa drivers. A lot
was already cleand up before. With this step the mdio bus gets its
own space and is no longer hidden inside the ethernet driver.
This commit is mostly a copy/paste that includes only minor changes.
- define prefixes are renamed to RTMDIO
- The driver is totally self contained (does not rely on SoC include)
- The DTS structure (mdio node below ethernet node) was kept
- The driver is added to the kernel config of all subtargets.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19942
Signed-off-by: Robert Marko <robimarko@gmail.com>
At the end of RX NAPI polling the counter and mask registers are
cleaned up. Although this might run in parallel there is no
synchronization and the register modifications are some wild mix.
RTL83xx enables only the interrupt of a single ring while RTL93xx
just reactivates all interrupts (even for other NAPI threads).
Make use of the driver lock and only modify the interrupt bits that
the current thread owns.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19960
Signed-off-by: Robert Marko <robimarko@gmail.com>
Now that the counter registers work fine there is no need to
free buffers in software. Hardware will automatically block
input processing when software processing is too slow.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19960
Signed-off-by: Robert Marko <robimarko@gmail.com>
The receive path of the RTL93xx SoCs is currently discarding packets
in software. Analysis gives the following explanation:
- RX ring size registers are setup with the full software ring size
- When packets are received the packet counter registers are increased
- After RX processing the counter registers are changed the wrong way
- From then SOC is allowed to receive more packets than software allows
- Overflow interrupts are fired
- As a reaction to that the software drops packets
Change the processing as follows:
- Setup ring size registers with a headroom of 2 buffers
- Decrease the counter registers with the real work done
With this change no more overflow interrupts occur because the SoC
disables the queues before they can overflow or hit a buffer that is
still owned by the CPU.
Benchmark from single stream iperf3 run, with server process running
on ZyXEL XGS1210 (RTL930x).
iperf3 run before
-----------------------------------------------------------
Server listening on 5201 (test #1)
-----------------------------------------------------------
Accepted connection from 192.168.2.86, port 54412
[ 5] local 192.168.2.71 port 5201 connected to 192.168.2.86 port 54418
[ ID] Interval Transfer Bitrate
[ 5] 0.00-1.00 sec 384 KBytes 3.14 Mbits/sec
[ 5] 1.00-2.00 sec 0.00 Bytes 0.00 bits/sec
[ 5] 2.00-3.00 sec 0.00 Bytes 0.00 bits/sec
[ 5] 3.00-4.01 sec 5.12 MBytes 42.8 Mbits/sec
[ 5] 4.01-5.00 sec 11.4 MBytes 95.8 Mbits/sec
[ 5] 5.00-6.00 sec 0.00 Bytes 0.00 bits/sec
[ 5] 6.00-7.00 sec 0.00 Bytes 0.00 bits/sec
[ 5] 7.00-8.00 sec 0.00 Bytes 0.00 bits/sec
[ 5] 8.00-9.00 sec 0.00 Bytes 0.00 bits/sec
[ 5] 9.00-10.00 sec 0.00 Bytes 0.00 bits/sec
iperf3 run after
-----------------------------------------------------------
Server listening on 5201 (test #1)
-----------------------------------------------------------
Accepted connection from 192.168.2.86, port 55228
[ 5] local 192.168.2.71 port 5201 connected to 192.168.2.86 port 55232
[ ID] Interval Transfer Bitrate
[ 5] 0.00-1.00 sec 22.8 MBytes 191 Mbits/sec
[ 5] 1.00-2.01 sec 25.4 MBytes 211 Mbits/sec
[ 5] 2.01-3.00 sec 25.4 MBytes 215 Mbits/sec
[ 5] 3.00-4.01 sec 26.5 MBytes 220 Mbits/sec
[ 5] 4.01-5.00 sec 26.2 MBytes 222 Mbits/sec
[ 5] 5.00-6.00 sec 26.9 MBytes 225 Mbits/sec
[ 5] 6.00-7.00 sec 27.0 MBytes 226 Mbits/sec
[ 5] 7.00-8.01 sec 26.9 MBytes 224 Mbits/sec
[ 5] 8.01-9.00 sec 26.5 MBytes 223 Mbits/sec
[ 5] 9.00-10.00 sec 26.8 MBytes 225 Mbits/sec
[ 5] 10.00-10.02 sec 640 KBytes 224 Mbits/sec
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19960
Signed-off-by: Robert Marko <robimarko@gmail.com>
Registers must not be accessed in parallel by multiple drivers.
Read-modify-write operations are not atomic, and the result of parallel
access is undefined.
The MAC_L2_GLOBAL_CTRL2 register is essentially a pin configuration
register and is represented by a pinmux node in the devicetree. Operations
on this register by the realtek,rtl838x-eth driver must therefore also be
reflected in the devicetree.
Since the MDIO sets used are board-specific, the pins must be enabled in
the board’s devicetree. This can be achieved using the pinctrl properties
for the realtek,rtl83xx-switch.
&switch0 {
pinctrl-names = "default";
pinctrl-0 = <&pinmux_enable_mdc_mdio_0>,
<&pinmux_enable_mdc_mdio_1>;
....
};
Signed-off-by: Sven Eckelmann <sven@narfation.org>
Link: https://github.com/openwrt/openwrt/pull/19815
Signed-off-by: Robert Marko <robimarko@gmail.com>
Currently the calculation for the CMU (even) SerDes works similar
to this pseudo code.
analog_backend_serdes = get_analog_serdes(frontend_serdes);
even_backend_serdes = analog_backend_serdes & ~1;
write_to(even_backend_serdes);
Because of the SerDes layout and frontend/backend mapping this can
be swapped to the following order with the same resulting Serdes.
even_frontend_serdes = frontend_serdes ~1;
analog_backend_serdes = get_analog_serdes(even_frontend_serdes);
write_to(analog_backed_serdes);
In the later example the frontend/backend mapping code is already
in our new functions. So swap the calculation logic and use the
new access functions. This allows to finally drop the old access
functions without mapping.
From now on all RTL931x SerDes functions will use a consistent
frontend view.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19873
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
The more we step down into the SerDes deeps the more confusing it
gets. Nevertheless it is not to late to fix a wrong assumption.
Until now it seemed as if the frontend/backend SerDes mapping is
totally without intersection. This is not true.
The backend SerDes mapping is also dependent on the mode. Especially
the proprietary Realtek XSGMII mode stands out from all other
mappings. So fix the descriptions and the calculation of the third
page package (digital 2 aka XSGMII 2).
As it was not yet used it had no impact.
Fixes: a4cbb44c1b ("realtek: convert access to RTL931x analog serdes pages")
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19873
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
For some reason 3 of the 4 mdio access functions contain an
artifical delay of 10ms. While it might have been part of
older Realtek SDKs it can no longer be found in current ones.
Remove the delays.
While we are here remove the pre-access bus ready checks.
It is sufficient to run them after the command start. If
anything fails the caller will get an error. This is the
same behaviour as for the other targets.
Finally cleanup the error handling. Something like this makes
no sense at all.
err = rtmdio_838x_smi_wait_op(100000);
if (err)
goto errout;
err = 0;
errout:
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19901
Signed-off-by: Robert Marko <robimarko@gmail.com>
The function of_get_mac_address is not taking care of evaluation the nvmem
address before trying to read out the mac-address properties. The driver
must check whether the return code is -EPROBE_DEFER and stop the probing
process in that case. If the nvmem-cell related driver code finished, the
probe can be redone ad the correct mac-address will appear for the device.
Signed-off-by: Harshal Gohel <hg@simonwunderlich.de>
Signed-off-by: Sharadanand Karanjkar <sk@simonwunderlich.de>
Link: https://github.com/openwrt/openwrt/pull/19362
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
The RTL931x has 14 frontend and at least 26 backend serdes. Currently
the programming functions always need to determine the right backend
serdes from the given frontend serdes on their own.
We plan to provide a consistent serdes mapping to all callers as follows
Frontend SerDes | 0 1 2 3 4 5 6 7 8 9 10 11 12 13
-----------------+------------------------------------------
Backend SerDes 1 | 0 1 2 3 6 7 10 11 14 15 18 19 22 23
Backend SerDes 2 | 0 1 2 4 6 8 10 12 14 16 18 20 22 24
Backend SerDes 3 | 0 1 2 5 6 9 10 13 14 17 18 21 22 25
frontend page "even" frontend SerDes "odd" frontend SerDes
page 0x000-0x03f (analog): page 0x000-0x03f back SDS page 0x000-0x03f back SDS
page 0x100-0x13f (XSGMII1): page 0x000-0x03f back SDS page 0x000-0x03f back SDS+1
page 0x200-0x23f (XSGMII2): page 0x000-0x03f back SDS page 0x000-0x03f back SDS+2
As a first micro step provide some helpers that simply operate on
frontend serdes and will determine the backend serdes on their own.
So rtmdio_931x_read_sds_phy() and rtmdio_931x_write_sds_phy() operate
on backend serdes. While rtmdio_931x_read_sds_phy_**new**() and
rtmdio_931x_write_sds_phy_**new**() operate on frontend serdes.
This is only an intermediate naming convention and will be cleanup
afterwards.
In a first step make use of these new functions whenever we
want to access the analog page. As the pages stay unchanged
in the new functions conversion is as simple as this:
Old:
asds = rtl931x_get_analog_sds(...)
rtmdio_931x_read_sds_phy(asds, ...)
New:
rtmdio_931x_read_sds_phy_new(sds, ...)
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19818
Signed-off-by: Robert Marko <robimarko@gmail.com>
When testing LLDP and STP, we observed that locally generated multicast
packets (e.g. LLDP, STP) were not restricted to the designated output
port(s). For example, when transmitting on `lan1`, the same packet was also
forwarded to other ports such as `lan2`.
Steps to reproduce:
1. Configure lldpd to use `lan1` in UCI and restart the service
2. Connect devices to `lan1` and `lan2`
3. Observe that the device on `lan2` still receives LLDP packets
The issue was caused by an incorrect `FWD_TYPE` setting in the TX CPU TAG,
which failed to enforce the selected egress port(s).
Fix this by updating the TX CPU TAG to set `FWD_TYPE` correctly, ensuring
that locally generated packets are transmitted only on the intended
port(s).
Signed-off-by: Issam Hamdi <ih@simonwunderlich.de>
Link: https://github.com/openwrt/openwrt/pull/19802
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
This commit repeats the mdio function relocation from the other targets.
In short that means:
- phy read/write functions are moved away from the phy driver
- SerDes read/write functions are moved away from the dsa driver
- All gets consolidated into the mdio driver (inside the ethernet driver)
This is mostly a copy/paste to keep the changes small. The SerDes phy mapping
and the simplification of the central bus functions will come later.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19743
Signed-off-by: Robert Marko <robimarko@gmail.com>
While checking setup routines for stability and completeness,
random RTL838x SoC I/O areas were intentionally overwritten.
As soon as L2_CTRL_1->FAST_AGE_OUT is set to 1, the system
stalls during bootup. Analysis shows that it loops endlessly
in rtl838x_hw_stop()
/* Flush L2 address cache */
if (priv->family_id == RTL8380_FAMILY_ID) {
for (int i = 0; i <= priv->cpu_port; i++) {
sw_w32(1 << 26 | 1 << 23 | i << 5, priv->r->l2_tbl_flush_ctrl);
do { } while (sw_r32(priv->r->l2_tbl_flush_ctrl) & (1 << 26));
}
This is exactly the same logic as in the vendor GPL. There
are no hints about possible timeouts or issues. The reason is
still unclear. Nevertheless harden the function for further fuzzy
tests. Do this by resetting the configuration value to its SoC
default.
Additionally convert some shifts to BIT() for better readability.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19679
Signed-off-by: Robert Marko <robimarko@gmail.com>
The RTL930x mdio functions are scattered around the code. Relocate
them to the bus (still inside the ethernet driver). With this change
the phy identification looks into the proper registers. The SerDes
phy identifier (register 2/3) must be changed.
Additionally provide a consistent SerDes register access through the
mdio bus. Until now when a SerDes directly drives a SFP module there
is no clear rule of how to handle its register set that consists of
two parts:
- c22 phy registers 0-15 live in the fiber page (2) of the SerDes
- other SerDes specific registers exist in pages before and after
The mdio bus and other SerDes functions are a wild mix of directly
looking into page 2 or just using self defined methods to access
data.
Adapt the bus to the new consistent phy interface that mixes the
SerDes register set like classic Realtek phys do it.
- Use register 31 as page select (already in the bus)
- Always keep the common registers 0-15 in place and read fiber page
- Map the SerDes internal registers into the upper vendor specific
registers 16-23 according to the page select register (31).
That gives a register mapping as follows:
+-----------------------+-----------------------+---------------+-------------+
| reg 0x00-0x0f | reg 0x10-0x17 | reg 0x18-0x1e | reg 0x1f |
+-----------------------+-----------------------+---------------+-------------+
| SerDes fiber page (3) | real SerDes registers | zero | SerDes page |
| registers 0 - 15 | in packages of 8 | | select reg |
+-----------------------+-----------------------+---------------+-------------+
Example to make it as clear as possible.
SerDes registers on a RTL930x show
Page / Reg | 0x00 0x01 0x02 0x03 0x04 0x05 0x06 0x07 0x08 0x09 0x0A 0x0B ...
-------------+----------------------------------------------------------------
0 - SDS | 0C03 0F00 7060 7106 074D 0EBF 0F0F 0359 5248 0000 0F80 0000 ...
1 - SDS_EXT | 0000 0000 85FA 8C6D 5CCC 0000 20D8 0003 79AA 8C64 00C3 1482 ...
2 - FIB | 1140 6189 001C CA40 01A0 0000 0000 0004 0000 0000 0000 0000 ...
3 - FIB_EXT | 1140 6109 001C CA40 01A0 0000 0000 0004 0000 0000 0000 0000 ...
This translates to this phy layout
| SerDes fiber registers normal SerDes registers zero p.sel
Page / Reg | 0x00 0x01 0x02 0x03 ... 0x10 0x11 0x12 0x13 ... 0x18 ... 0x1f
-------------+---------------------------------------------------------------
0 | 1140 6189 001C CA40 ... 0C03 0F00 7060 7106 ... 0000 ... 0000
1 | 1140 6189 001C CA40 ... 5248 0000 0F80 0000 ... 0000 ... 0001
...
4 | 1140 6189 001C CA40 ... 0000 0000 85FA 8C6D ... 0000 ... 0004
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19692
Signed-off-by: Robert Marko <robimarko@gmail.com>
Just like rtl930x, rtl931x also requires two reads to fetch current link
status.
While at it, rename the function to a proper naming scheme.
Signed-off-by: Harshal Gohel <hg@simonwunderlich.de>
Co-developed-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Co-developed-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Sharadanand Karanjkar <sk@simonwunderlich.de>
Link: https://github.com/openwrt/openwrt/pull/19578
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Move the definitions to mach-rtl83xx.h, so they can be used during init
to read more detailed SoC information. Also rename the RTL931X register,
as it has the same address on all RTL93xx.
Signed-off-by: Jan Hoffmann <jan@3e8.eu>
Link: https://github.com/openwrt/openwrt/pull/19653
Signed-off-by: Robert Marko <robimarko@gmail.com>
The migration of the RTL930x mdio/serdes access functions over to the
mdio bus is a little more complicated than for RTL83xx. There are several
places where the serdes is accessed directly. So do it in two steps. With
this first step:
- use the rtmdio prefix for the serdes reader/writer functions
- move the functions over to the bus (inside the ethernet driver)
- Adapt all callers.
This is not only a copy/paste but the serdes access will be hardened too.
For this:
- put a mutex around the read/write functions because we have only
indirect register access through a mdio style bus.
- Verify input values to avoid data mess.
Tested-by: Bjørn Mork <bjorn@mork.no>
Tested-by: Jan Hoffmann <jan@3e8.eu>
Tested-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19662
Signed-off-by: Robert Marko <robimarko@gmail.com>
The RTL839x mdio functions are scattered around the code. Relocate
them to the bus (still inside the ethernet driver).
Additionally provide a consistent SerDes register access through the
mdio bus. Until now when a SerDes directly drives a SFP module there
is no clear rule of how to handle its register set that consists of
two parts:
- c22 phy registers 0-15 live in the fiber page (2) of the SerDes
- other SerDes specific registers exist in pages before and after
The mdio bus and other SerDes functions are a wild mix of directly
looking into page 2 or just using self defined methods to access
data.
Adapt the bus to the new consistent phy interface that mixes the
SerDes register set like classic Realtek phys do it.
- Use register 31 as page select (already in the bus)
- Always keep the common registers 0-15 in place and read fiber page
- Map the SerDes internal registers into the upper vendor specific
registers 16-23 according to the page select register (31).
That gives a register mapping as follows:
+-----------------------+-----------------------+---------------+-------------+
| reg 0x00-0x0f | reg 0x10-0x17 | reg 0x18-0x1e | reg 0x1f |
+-----------------------+-----------------------+---------------+-------------+
| SerDes fiber page (3) | real SerDes registers | zero | SerDes page |
| registers 0 - 15 | in packages of 8 | | select reg |
+-----------------------+-----------------------+---------------+-------------+
Example to make it as clear as possible.
SerDes registers on a RTL839x show
Page / Reg | 0x00 0x01 0x02 0x03 0x04 0x05 0x06 0x07 0x08 0x09 0x0A 0x0B ...
-------------+----------------------------------------------------------------
0 - SDS | 0C03 0F00 7060 7106 074D 0EBF 0F0F 0359 5248 0000 0F80 0000 ...
1 - SDS_EXT | 0000 0000 85FA 8C6D 5CCC 0000 20D8 0003 79AA 8C64 00C3 1482 ...
2 - FIB | 1140 6189 001C CA40 01A0 0000 0000 0004 0000 0000 0000 0000 ...
3 - FIB_EXT | 1140 6109 001C CA40 01A0 0000 0000 0004 0000 0000 0000 0000 ...
This translates to this phy layout
| SerDes fiber registers normal SerDes registers zero p.sel
Page / Reg | 0x00 0x01 0x02 0x03 ... 0x10 0x11 0x12 0x13 ... 0x18 ... 0x1f
-------------+---------------------------------------------------------------
0 | 1140 6189 001C CA40 ... 0C03 0F00 7060 7106 ... 0000 ... 0000
1 | 1140 6189 001C CA40 ... 5248 0000 0F80 0000 ... 0000 ... 0001
...
4 | 1140 6189 001C CA40 ... 0000 0000 85FA 8C6D ... 0000 ... 0004
...
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19634
Signed-off-by: Robert Marko <robimarko@gmail.com>
When a SerDes directly drives a SFP module there is no clear rule of
how to handle its register set that consists of two parts:
- c22 phy registers 0-15 live in the fiber page (2) of the SerDes
- other SerDes specific registers exist in pages before and after
The mdio bus and other SerDes functions are a wild mix of directly
looking into page 2 or just using self defined methods to access
data.
Provide a consistent phy interface that mixes the SerDes register
set like classic Realtek phys do it.
- Use register 31 as page select (already in the bus)
- Always keep the common registers 0-15 in place and read fiber page
- Map the SerDes internal registers into the upper vendor specific
registers 16-23 according to the page select register (31).
That gives a register mapping as follows:
+-----------------------+-----------------------+---------------+-------------+
| reg 0x00-0x0f | reg 0x10-0x17 | reg 0x18-0x1e | reg 0x1f |
+-----------------------+-----------------------+---------------+-------------+
| SerDes fiber page (3) | real SerDes registers | zero | SerDes page |
| registers 0 - 15 | in packages of 8 | | select reg |
+-----------------------+-----------------------+---------------+-------------+
Example to make it as clear as possible.
SerDes registers on a RTL838x show
Page / Reg | 0x00 0x01 0x02 0x03 0x04 0x05 0x06 0x07 0x08 0x09 0x0A 0x0B ...
-------------+----------------------------------------------------------------
0 - SDS | 0C03 0F00 7060 7106 074D 0EBF 0F0F 0359 5248 0000 0F80 0000 ...
1 - SDS_EXT | 0000 0000 85FA 8C6D 5CCC 0000 20D8 0003 79AA 8C64 00C3 1482 ...
2 - FIB | 1140 6189 001C CA40 01A0 0000 0000 0004 0000 0000 0000 0000 ...
3 - FIB_EXT | 1140 6109 001C CA40 01A0 0000 0000 0004 0000 0000 0000 0000 ...
This translates to this phy layout
| SerDes fiber registers normal SerDes registers zero p.sel
Page / Reg | 0x00 0x01 0x02 0x03 ... 0x10 0x11 0x12 0x13 ... 0x18 ... 0x1f
-------------+---------------------------------------------------------------
0 | 1140 6189 001C CA40 ... 0C03 0F00 7060 7106 ... 0000 ... 0000
1 | 1140 6189 001C CA40 ... 5248 0000 0F80 0000 ... 0000 ... 0001
...
4 | 1140 6189 001C CA40 ... 0000 0000 85FA 8C6D ... 0000 ... 0004
For now just do it for RTL838x devices.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19604
Signed-off-by: Robert Marko <robimarko@gmail.com>
The mdio bus functions are still split between ethernet and dsa driver.
Before moving everthing out to a separate mdio driver we decided to
collect everything in the ethernet driver with the rtmdio prefix.
Take over the remaining RTL838x functions.
Remark: This is more or less a copy/paste with function renaming. As
there are still some consumers in the DSA driver the definitions and
inclusions must be flipped.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19484
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
RTL83xx devices have two types of receive interrupts for each of its
8 rings. One for packet received and another for ring overflow. When
the switch is flooded with incoming packets the receive handler will
disable the packet receive notification but still keeps the overflow
notification enabled. While the receive path "slowly" processes the
received packets each new packet triggers the overflow IRQ again. The
device becomes unresponsive and eventually produces messages like:
[18441.709764] rcu: Stack dump where RCU GP kthread last ran:
[18441.727892] Sending NMI from CPU 1 to CPUs 0:
[18441.742300] NMI backtrace for cpu 0 skipped: idling at 0x8080e994
[18415.251700] rcu: INFO: rcu_sched detected stalls on CPUs/tasks:
[18415.271350] rcu: 0-...!: (0 ticks this GP) idle=d740/0/0x0 ...
[18415.303046] rcu: (detected by 1, t=6004 jiffies, g=230925, ...
[18415.326095] Sending NMI from CPU 1 to CPUs 0:
[18415.340540] NMI backtrace for cpu 0
Fix this issue by always disabling receive and overflow interrupts at
the same time.
Test with hping3 --udp -p 5021 -d 1400 --flood 192.168.2.72
Before (3sec run):
[183260.324846] rtl838x-eth 1b00a300.ethernet eth0: RX buffer overrun: status 0x101, mask: 0x7ffeff
[183260.340524] rtl838x-eth 1b00a300.ethernet eth0: RX buffer overrun: status 0x1, mask: 0x7ffeff
[183260.345799] net_ratelimit: 489997 callbacks suppressed
After (3 sec run):
[ 373.981479] rtl838x-eth 1b00a300.ethernet eth0: rx ring overrun: status 0x101, mask: 0x7fffff
[ 374.031118] rtl838x-eth 1b00a300.ethernet eth0: rx ring overrun: status 0x101, mask: 0x7fffff
[ 377.919996] net_ratelimit: 34 callbacks suppressed
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19365
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
During setup the mdio driver decides the polling mode of the 4 smi
busses depending on the DTS phy settings. This works as follows:
- set polling to c45 if at least one phy is ethernet-phy-ieee802.3-c45
- set polling to c22 if all phys are ethernet-phy-ieee802.3-c22
On RTL930x it is not possible to switch to c22 if uboot has set c45
before. Fix this by overwriting the bitfield properly. While we are
here:
- Sort variables according to kernel style (inverse christmas tree)
- Initialize fields properly with = { 0 }
- Use GENMASK() for better readability
- Make use of RTMDIO_MAX_SMI_BUS
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19161
Signed-off-by: Robert Marko <robimarko@gmail.com>
With the follow up EEE patches the mdio bus will run c22 and c45
accesses during initial scan. Especially when accessing addresses
beyond the CPU port phy requests might fail in a way that cannot
be handled gratefully. Do two things
- do not allow access to addresses starting from cpu port
- set the scan disable bitmask to ports starting from cpu port
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/18935
Signed-off-by: Robert Marko <robimarko@gmail.com>
The validate function no longer exists in phylink_mac_ops. Remove
it for the internal ethernet interface. Instead provide some
meaningful mac capabilities.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/18935
Signed-off-by: Robert Marko <robimarko@gmail.com>
Automatically generated commit.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/18935
Signed-off-by: Robert Marko <robimarko@gmail.com>
