* [PATCH] [RFC] fec: fix calculation of mii bus speed on mxs
@ 2012-09-06 13:50 Uwe Kleine-König
2012-09-10 7:32 ` Sascha Hauer
0 siblings, 1 reply; 2+ messages in thread
From: Uwe Kleine-König @ 2012-09-06 13:50 UTC (permalink / raw)
To: barebox; +Cc: kernel
According to a comment in Linux' fec driver, i.MX28 uses the same
formula for determination of the frequency divider as i.MX6, that is
(different from the i.MX28 manual):
parent clock / ((MII_SPEED + 1) * 2)
instead of
parent clock / (MII_SPEED * 2)
on the older i.MX SoCs. Fix the calculation accordingly.
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
--
But note that this doesn't fix accessing the phy on my machine. The
calculated value is 9 (with and without this patch btw) but accessing
the phy only gets reliable with a value of >=20 or alternatively don't
set the SPEED value before reading and writing a mii register but only
on probe.
The Freescale kernel 2.6.35_10.12.01 does the following instead (in
fec_switch.c):
fep->phy_speed = DIV_ROUND_UP(clk_get_rate(fep->clk), 5000000) << 1;
#ifdef CONFIG_ARCH_MXS
/* Can't get phy(8720) ID when set to 2.5M on MX28, lower it */
fep->phy_speed <<= 2;
#endif
which would result in 40 (and is unaware of the changed formula).
I didn't have an opportunity to check the signals with an oszilloscope, but
intend to fetch that later.
---
drivers/net/fec_imx.c | 48 ++++++++++++++++++++++++++++++++++++++----------
1 file changed, 38 insertions(+), 10 deletions(-)
diff --git a/drivers/net/fec_imx.c b/drivers/net/fec_imx.c
index 599a9b4..969c903 100644
--- a/drivers/net/fec_imx.c
+++ b/drivers/net/fec_imx.c
@@ -47,6 +47,41 @@ struct fec_frame {
uint8_t head[16]; /* MAC header(6 + 6 + 2) + 2(aligned) */
};
+static void fec_miidev_setspeed(struct fec_priv *fec)
+{
+ u32 mii_speed;
+
+ mii_speed = DIV_ROUND_UP(imx_get_fecclk(), 5000000);
+
+#ifdef CONFIG_ARCH_MXS
+ /*
+ * Compared to the other imx socs imx28 and imx6 have an additional "+1"
+ * in the formula for MII_SPEED. In MCIMX28RM Rev.1, 2010 this is not
+ * documented though.
+ */
+ mii_speed -= 1;
+#endif
+ if (!mii_speed)
+ /* Can this happen? */
+ mii_speed = 1;
+ if (mii_speed > 0x3f) {
+ static int once = 0;
+ if (!once)
+ printf("Warning: fec clk too high to get mii clk\n");
+ mii_speed = 0x3f;
+ }
+
+ /*
+ * XXX: some SoCs (at least i.MX28 and i.MX6, but not .iMX27) have an
+ * bitfield in ENET_MSCR (aka MII_SPEED) called HOLDTIME that needs to
+ * be >0 for imx_get_fecclk() > 100 MHz.
+ */
+
+ writel(mii_speed << 1, fec->regs + FEC_MII_SPEED);
+
+ return 0;
+}
+
/*
* MII-interface related functions
*/
@@ -59,8 +94,7 @@ static int fec_miidev_read(struct mii_device *mdev, int phyAddr, int regAddr)
uint32_t phy; /* convenient holder for the PHY */
uint64_t start;
- writel(((imx_get_fecclk() >> 20) / 5) << 1,
- fec->regs + FEC_MII_SPEED);
+ fec_miidev_setspeed(fec);
/*
* reading from any PHY's register is done by properly
* programming the FEC's MII data register.
@@ -103,8 +137,7 @@ static int fec_miidev_write(struct mii_device *mdev, int phyAddr,
uint32_t phy; /* convenient holder for the PHY */
uint64_t start;
- writel(((imx_get_fecclk() >> 20) / 5) << 1,
- fec->regs + FEC_MII_SPEED);
+ fec_miidev_setspeed(fec);
reg = regAddr << FEC_MII_DATA_RA_SHIFT;
phy = phyAddr << FEC_MII_DATA_PA_SHIFT;
@@ -292,12 +325,7 @@ static int fec_init(struct eth_device *dev)
rcntl = FEC_R_CNTRL_MAX_FL(1518);
if (fec->xcv_type != SEVENWIRE) {
rcntl |= FEC_R_CNTRL_MII_MODE;
- /*
- * Set MII_SPEED = (1/(mii_speed * 2)) * System Clock
- * and do not drop the Preamble.
- */
- writel(((imx_get_fecclk() >> 20) / 5) << 1,
- fec->regs + FEC_MII_SPEED);
+ fec_miidev_setspeed(fec);
}
if (fec->xcv_type == RMII) {
--
1.7.10.4
_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox
^ permalink raw reply [flat|nested] 2+ messages in thread* Re: [PATCH] [RFC] fec: fix calculation of mii bus speed on mxs
2012-09-06 13:50 [PATCH] [RFC] fec: fix calculation of mii bus speed on mxs Uwe Kleine-König
@ 2012-09-10 7:32 ` Sascha Hauer
0 siblings, 0 replies; 2+ messages in thread
From: Sascha Hauer @ 2012-09-10 7:32 UTC (permalink / raw)
To: Uwe Kleine-König; +Cc: barebox, kernel
Hi Uwe,
On Thu, Sep 06, 2012 at 03:50:01PM +0200, Uwe Kleine-König wrote:
> According to a comment in Linux' fec driver, i.MX28 uses the same
> formula for determination of the frequency divider as i.MX6, that is
> (different from the i.MX28 manual):
>
> parent clock / ((MII_SPEED + 1) * 2)
>
> instead of
>
> parent clock / (MII_SPEED * 2)
>
> on the older i.MX SoCs. Fix the calculation accordingly.
>
> Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
> --
> But note that this doesn't fix accessing the phy on my machine. The
> calculated value is 9 (with and without this patch btw) but accessing
> the phy only gets reliable with a value of >=20 or alternatively don't
> set the SPEED value before reading and writing a mii register but only
> on probe.
>
> The Freescale kernel 2.6.35_10.12.01 does the following instead (in
> fec_switch.c):
>
> fep->phy_speed = DIV_ROUND_UP(clk_get_rate(fep->clk), 5000000) << 1;
> #ifdef CONFIG_ARCH_MXS
> /* Can't get phy(8720) ID when set to 2.5M on MX28, lower it */
> fep->phy_speed <<= 2;
> #endif
>
> which would result in 40 (and is unaware of the changed formula).
>
> I didn't have an opportunity to check the signals with an oszilloscope, but
> intend to fetch that later.
> ---
> drivers/net/fec_imx.c | 48 ++++++++++++++++++++++++++++++++++++++----------
> 1 file changed, 38 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/net/fec_imx.c b/drivers/net/fec_imx.c
> index 599a9b4..969c903 100644
> --- a/drivers/net/fec_imx.c
> +++ b/drivers/net/fec_imx.c
> @@ -47,6 +47,41 @@ struct fec_frame {
> uint8_t head[16]; /* MAC header(6 + 6 + 2) + 2(aligned) */
> };
>
> +static void fec_miidev_setspeed(struct fec_priv *fec)
> +{
> + u32 mii_speed;
> +
> + mii_speed = DIV_ROUND_UP(imx_get_fecclk(), 5000000);
> +
> +#ifdef CONFIG_ARCH_MXS
> + /*
> + * Compared to the other imx socs imx28 and imx6 have an additional "+1"
> + * in the formula for MII_SPEED. In MCIMX28RM Rev.1, 2010 this is not
> + * documented though.
> + */
> + mii_speed -= 1;
> +#endif
Please use a if (cpu_is_mx28()) instead. While we are at it we should
also do this for i.MX6 and add a cpu_is_mx6() aswell.
Sascha
--
Pengutronix e.K. | |
Industrial Linux Solutions | http://www.pengutronix.de/ |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 |
Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2012-09-10 7:32 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2012-09-06 13:50 [PATCH] [RFC] fec: fix calculation of mii bus speed on mxs Uwe Kleine-König
2012-09-10 7:32 ` Sascha Hauer
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox