I tested with the new official dtb and the following diff on the Nov 24th
snapshot and it works fine.
I have been using this code on my orange pi one (Allwinner h3) for a couple
of weeks now without any problems. The new dtb seems to have only required a
name change for the clock and reset, back to what was in the original code.
(It seems there is a bit of formatting that caused the diff to pick up the
name lines that did not change.)
# diff -u if_dwxe.c_orig if_dwxe.c
--- if_dwxe.c_orig Tue Nov 28 13:04:10 2017
+++ if_dwxe.c Thu Nov 30 16:19:25 2017
@@ -146,11 +146,11 @@
#define DWXE_MDIO_CMD_PHY_REG_SHIFT 4
#define DWXE_MDIO_CMD_PHY_ADDR_SHIFT 12
#define DWXE_MDIO_CMD_MDC_DIV_RATIO_M_SHIFT 20
-#define DWXE_MDIO_CMD_MDC_DIV_RATIO_M_MASK (0x7 << 20)
-#define DWXE_MDIO_CMD_MDC_DIV_RATIO_M_16 (0 << 20)
-#define DWXE_MDIO_CMD_MDC_DIV_RATIO_M_32 (1 << 20)
-#define DWXE_MDIO_CMD_MDC_DIV_RATIO_M_64 (2 << 20)
-#define DWXE_MDIO_CMD_MDC_DIV_RATIO_M_128 (3 << 20)
+#define DWXE_MDIO_CMD_MDC_DIV_RATIO_M_MASK 0x7
+#define DWXE_MDIO_CMD_MDC_DIV_RATIO_M_16 0
+#define DWXE_MDIO_CMD_MDC_DIV_RATIO_M_32 1
+#define DWXE_MDIO_CMD_MDC_DIV_RATIO_M_64 2
+#define DWXE_MDIO_CMD_MDC_DIV_RATIO_M_128 3
#define DWXE_MDIO_DATA 0x4C
#define DWXE_MACADDR_HI 0x50
#define DWXE_MACADDR_LO 0x54
@@ -236,6 +236,7 @@
#define SYSCON_H3_EPHY_LED_POL (1 << 17) /* 1: active low, 0:
active high */
#define SYSCON_H3_EPHY_CLK_SEL (1 << 18) /* 1: 24MHz, 0: 25MHz */
#define SYSCON_H3_EPHY_ADDR_SHIFT 20
+#define SYSCON_H3_EPHY_ADDR_MASK 0x1f
struct dwxe_buf {
bus_dmamap_t tb_map;
@@ -270,6 +271,7 @@
struct mii_data sc_mii;
#define sc_media sc_mii.mii_media
int sc_link;
+ int sc_phyloc;
struct dwxe_dmamem *sc_txring;
struct dwxe_buf *sc_txbuf;
@@ -357,7 +359,6 @@
struct fdt_attach_args *faa = aux;
struct ifnet *ifp;
int phy, phy_supply, node;
- int phyloc = MII_PHY_ANY;
sc->sc_node = faa->fa_node;
sc->sc_iot = faa->fa_iot;
@@ -372,13 +373,12 @@
phy = OF_getpropint(faa->fa_node, "phy-handle", 0);
node = OF_getnodebyphandle(phy);
if (node)
- phyloc = OF_getpropint(node, "reg", phyloc);
-
+ sc->sc_phyloc = OF_getpropint(node, "reg", MII_PHY_ANY);
pinctrl_byname(faa->fa_node, "default");
/* Enable clock. */
- clock_enable(faa->fa_node, "stmmaceth");
- reset_deassert(faa->fa_node, "stmmaceth");
+ clock_enable(faa->fa_node, "stmmaceth");
+ reset_deassert(faa->fa_node, "stmmaceth");
delay(5000);
/* Power up PHY. */
@@ -386,7 +386,7 @@
if (phy_supply)
regulator_enable(phy_supply);
- sc->sc_clk = clock_get_frequency(faa->fa_node, "stmmaceth");
+ sc->sc_clk = clock_get_frequency(faa->fa_node, "stmmaceth");
if (sc->sc_clk > 160000000)
sc->sc_clk = DWXE_MDIO_CMD_MDC_DIV_RATIO_M_128;
else if (sc->sc_clk > 80000000)
@@ -424,8 +424,8 @@
dwxe_reset(sc);
- mii_attach(self, &sc->sc_mii, 0xffffffff, phyloc,
- MII_OFFSET_ANY, 0);
+ mii_attach(self, &sc->sc_mii, 0xffffffff, sc->sc_phyloc,
+ MII_OFFSET_ANY, MIIF_NOISOLATE);
if (LIST_FIRST(&sc->sc_mii.mii_phys) == NULL) {
printf("%s: no PHY found!\n", sc->sc_dev.dv_xname);
ifmedia_add(&sc->sc_media, IFM_ETHER|IFM_MANUAL, 0, NULL);
@@ -466,8 +466,14 @@
syscon |= SYSCON_EPIT | SYSCON_ETCS_EXT_GMII;
else if (!strncmp(phy_mode, "mii", strlen("mii")) &&
OF_is_compatible(sc->sc_node, "allwinner,sun8i-h3-emac")) {
- panic("%s: setup internal phy", DEVNAME(sc));
- return;
+ syscon &= ~SYSCON_H3_EPHY_SHUTDOWN;
+ syscon |= SYSCON_H3_EPHY_SELECT|SYSCON_H3_EPHY_CLK_SEL;
+ if (OF_getproplen(sc->sc_node, "allwinner,leds-active-low")
>= 0)
+ syscon |= SYSCON_H3_EPHY_LED_POL;
+ else
+ syscon &= ~SYSCON_H3_EPHY_LED_POL;
+ syscon &= ~(SYSCON_H3_EPHY_ADDR_MASK <<
SYSCON_H3_EPHY_ADDR_SHIFT);
+ syscon |= sc->sc_phyloc << SYSCON_H3_EPHY_ADDR_SHIFT;
}
free(phy_mode, M_TEMP, len);
-----Original Message-----
From: owner-...@openbsd.org [mailto:owner-...@openbsd.org] On Behalf Of
Jonathan Gray
Sent: Monday, November 27, 2017 12:46 AM
To: Mark Kettenis <mark.kette...@xs4all.nl>
Cc: arm@openbsd.org
Subject: Re: FW: help with setting up dwxe on orange pi one (h3)
On Thu, Sep 28, 2017 at 04:09:36PM +0200, Mark Kettenis wrote:
> Jonathan Gray schreef op 2017-09-28 09:51:
> > On Thu, Sep 28, 2017 at 09:34:49AM +0200, Mark Kettenis wrote:
> > > Jonathan Gray schreef op 2017-09-28 05:02:
> > > > On Wed, Sep 27, 2017 at 10:56:44PM +0200, Patrick Wildt wrote:
> > > > > So, first of all copying the dtb entries is not a good idea.
> > > > > The reason is that the phandles are gonna be all wrong and
> > > > > overriden, because those are _generated_ on compile time. As
> > > > > you can see, the ethernet controller references phy handle
> > > > > 0x7, but the phy has a phandle of 0x47.
> > > > > Something is wrong there.
> > > > >
> > > > > You should try to apply this "revert" that was committed in linux.
> > > > > Hope
> > > > > they'll soon make up their minds and commit a newer version.
> > > >
> > > > The device tree that is included with
> > > > /usr/local/share/u-boot/orangepi_one/u-boot-sunxi-with-spl.bin
> > > > should still have emac nodes.
> > >
> > > But they use a different binding from the Linux kernel, which are
> > > the ones Patrick used for the implementation. And some of the
> > > regulator stuff is missing so things don't actually work even if
> > > we would change the driver to use the bindings currently used by
> > > U-Boot. It's a mess.
> > >
> > > So using the dtb from Linux with the revert applied is the way to
> > > go for now.
> > >
> >
> > So we patch the dtb port to revert the revert commits for now?
> > Doesn't appear to have been added back in 3.14 rcs yet.
>
> We could do that.
The port now builds 4.15-rc1 which has EMAC for A64/H5/H3 added back.
The mdio child nodes change but it looks like it should already be
compatible with what is in the tree as that is not used.
