1 1
Bat

AllWinner H2+ integrated PHY

Recommended Posts

Note : feel free to move this topic "where it can be" or point location where to post.

I'm aware it's a cross way between hardware and software.

 

I'm currently using an Orange board PI R1, that have 2 ethernet, one came from internal Allwinner H2+ EMAC and EPHY, the other is an USB bridge

I have a full custom setup (based on armbian Linux version 4.13.9+ (root@bat-virtual-machine) (gcc version 7.1.1 20170707 (Linaro GCC 7.1-2017.08)) #113 SMP PREEMPT). I only used kernel, then have custom init and app.

I use ethernet in USB bridge like standard one (Linux kernel driver, TCPs sockets on it), I can reach 5Mo/s in both way (at same time), and that enough for my app.
Internal EMAC/EPHY is handled directly by my kernel module, has I'm using Ethercat protocol. Specific point, is that it heavy use full duplex, frame go out and will came back during emission (let's say 10 µs delay)

I solved all software issue, it's stable.

But I see some random packet corruption. I've tracked it down to PHY saying that it see incorrect symbol (avaible in packet descriptor), CRC is not good. Using known data, I clearly see that there is 1 or 2 bytes afftected on frame. For ex 0x00 => 0x40, or 0x00 => 0x0C.

PER is quite low but depend on product on other side (probably exact phy, or exact delay).

I have 1800 packet/s (for test packets are 1000 bytes)

With an Ethercat product, I have an error each 5mn (540k packets)
With an ethercat hub (it have a more stable, shorter delay),  I have an error each 2 hours (12M packet)
Error are not at specific packet byte, but random.

If a do a direct TX=>RX loopback at RJ45 connector, I have no error in 24h (that's my target).

Note :

-with other hardware, I reach this target with same Ethercat product and hub.

-I'm in quite envirnoment (no external EMC)

-Even if a "wiretapped" (put another pair in parallel of RX that go on a Wireshark on PC, frame is received OK, not corrupted).

-I can reach this 24h test OK, in very specific condition (replace onboard RJ45 + hardware force WiFi down - that is on top of ethernet wire + add delay on RX path), and with a specific delay configuration (modifying ethernet wiring will vary delay a bit, probably in range of 100 ps, same range that natural PHY jitter variation)

I've tried other various thing on hardware, trying to filter power, filter PHY power (1.1V), have a better power (it's 1.2V lowered by a serial resistor), filter general power. Tried also another RJ45 filtered jack. Tried on a second board. Tried to filter ethernet (10 pF on signal) in case of spike and so on.


I miss the exact schematic of Orange PI R1 board (one in web site is not the good one), and exact board layout.

I tried also to read/modify internal PHY registers. But I have not the documentation of internal MDIO registers.
The best I have is, I also found another driver in paste.bin :
https://github.com/allwinner-zh/linux-3.4-sunxi/blob/master/drivers/net/phy/sunxi-ephy.c
Specifically registers in extended pages are the one of more interest. On documented PHY, they often have "fine" tuning, that can be very helpful here.


The PHY is a Sunplus core IP (guess by OUI identifier). But I didn't find any documentation (even on another product of Sunplus or Allwinner) that can describe theses PHY registers.
Idea is to tweak some delay if possible ...

        phy_write(phydev, 0x1f, 0x0200);        /* Switch to Page 2 */

        phy_write(phydev, 0x18, 0x0000);        /* PHYAFE TRX optimization */

        phy_write(phydev, 0x1f, 0x0600);        /* Switch to Page 6 */

        phy_write(phydev, 0x14, 0x708f);        /* PHYAFE TX optimization */

        phy_write(phydev, 0x13, 0xF000);        /* PHYAFE RX optimization */

        phy_write(phydev, 0x15, 0x1530);

        phy_write(phydev, 0x1f, 0x0800);        /* Switch to Page 6 */

        phy_write(phydev, 0x18, 0x00bc);        /* PHYAFE TRX optimization */[tr]        [/tr]

 

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
1 1