From mboxrd@z Thu Jan 1 00:00:00 1970 From: Oliver Martin Subject: Re: [PATCH] Re: ep93xx_eth PHY problems Date: Sun, 14 Sep 2008 14:11:10 +0200 Message-ID: <20080914141110.18b42a42@ors.home> References: <47CD4857.2060406@student.tuwien.ac.at> <20080321202202.129c1239@ors> <48CC72D4.1020001@garzik.org> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Cc: netdev@vger.kernel.org, hvr@gnu.org, buytenh@wantstofly.org To: Jeff Garzik Return-path: Received: from ey-out-2122.google.com ([74.125.78.27]:1295 "EHLO ey-out-2122.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754500AbYINMLR (ORCPT ); Sun, 14 Sep 2008 08:11:17 -0400 Received: by ey-out-2122.google.com with SMTP id 6so709140eyi.37 for ; Sun, 14 Sep 2008 05:11:14 -0700 (PDT) In-Reply-To: <48CC72D4.1020001@garzik.org> Sender: netdev-owner@vger.kernel.org List-ID: Am Sat, 13 Sep 2008 22:11:32 -0400 schrieb Jeff Garzik: > Oliver Martin wrote: > > Am Tue, 04 Mar 2008 14:02:15 +0100 > > schrieb Oliver Martin : > >> Now I'm having different problems: When auto-negotiation is > >> enabled, it sometimes randomly changes the link state: > >> > >> PHY: 0:01 - Link is Down > >> PHY: 0:01 - Link is Up - 100/Full > >> > > > > I've found at least one part of the problem. The MDIO clock divisor > > wasn't set again after the reset in ep93xx_start_hw. It was left at > > the default, resulting in a 12.5 MHz clock, which the PHY (Micrel > > KSZ8721BL) didn't like too much. Calling ep93xx_mdio_reset after > > the reset fixed it, the link is now perfectly stable in > > auto-negotiation mode. > > > > There still seems to be another bug with auto-negotiation disabled: > > It doesn't switch to 1000 Mbps any more, but most times it doesn't > > stay in the intended configuration either. I haven't done any > > extensive testing of this yet, though. > > > > Signed-off-by: Oliver Martin > > Testing feedback? Lennert ack? We're currently using a variation of my original patch, which didn't handle ifdown correctly (there's a similar reset there). With the current setup, everything seems stable, but we still have some rather hackish workarounds for the original issue in our startup scripts. I don't have the hardware right now, but we'll specifically test without them both with and without auto-negotiation next week and report the results. Index: linux-2.6.26/drivers/net/arm/ep93xx_eth.c =================================================================== --- linux-2.6.26.orig/drivers/net/arm/ep93xx_eth.c 2008-07-25 03:13:33.000000000 +0200 +++ linux-2.6.26/drivers/net/arm/ep93xx_eth.c 2008-07-25 03:14:52.000000000 +0200 @@ -535,6 +535,22 @@ return 1; } +static int ep93xx_mdio_reset(struct mii_bus *bus) +{ + struct ep93xx_priv *ep = bus->priv; + + u32 selfctl = rdl(ep, REG_SELFCTL); + + selfctl &= ~(REG_SELFCTL_MDCDIV_MSK | REG_SELFCTL_PSPRS); + + selfctl |= (ep->mdc_divisor - 1) << REG_SELFCTL_MDCDIV_OFS; + selfctl |= REG_SELFCTL_PSPRS; + + wrl(ep, REG_SELFCTL, selfctl); + + return 0; +} + static int ep93xx_start_hw(struct net_device *dev) { struct ep93xx_priv *ep = netdev_priv(dev); @@ -553,6 +569,9 @@ return 1; } + /* The reset cleared REG_SELFCTL, so set the MDC divisor again */ + ep93xx_mdio_reset(&ep->mii_bus); + /* Receive descriptor ring. */ addr = ep->descs_dma_addr + offsetof(struct ep93xx_descs, rdesc); wrl(ep, REG_RXDQBADD, addr); @@ -625,6 +644,9 @@ if (i == 10) printk(KERN_CRIT DRV_MODULE_NAME ": hw failed to reset\n"); + + /* The reset cleared REG_SELFCTL, so set the MDC divisor again */ + ep93xx_mdio_reset(&ep->mii_bus); } static int ep93xx_open(struct net_device *dev) @@ -776,22 +798,6 @@ return 0; } -static int ep93xx_mdio_reset(struct mii_bus *bus) -{ - struct ep93xx_priv *ep = bus->priv; - - u32 selfctl = rdl(ep, REG_SELFCTL); - - selfctl &= ~(REG_SELFCTL_MDCDIV_MSK | REG_SELFCTL_PSPRS); - - selfctl |= (ep->mdc_divisor - 1) << REG_SELFCTL_MDCDIV_OFS; - selfctl |= REG_SELFCTL_PSPRS; - - wrl(ep, REG_SELFCTL, selfctl); - - return 0; -} - static void ep93xx_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info) {