From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S933049AbdCUQJG (ORCPT ); Tue, 21 Mar 2017 12:09:06 -0400 Received: from mail-qt0-f194.google.com ([209.85.216.194]:33269 "EHLO mail-qt0-f194.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932947AbdCUQIi (ORCPT ); Tue, 21 Mar 2017 12:08:38 -0400 Subject: Re: [RFC PATCH] net: phy: Don't miss phy_suspend() on PHY_HALTED for PHYs with interrupts To: Roger Quadros , Andrew Lunn References: <1489585887-8683-1-git-send-email-rogerq@ti.com> <20170315140811.GD21021@lunn.ch> <6ec42ad7-0be3-f6c5-2ded-27bf3adbab23@ti.com> <20170315154912.GE21021@lunn.ch> <6d3f2d3c-69e0-43a4-9862-126a7c173934@ti.com> <625a9ee1-ea65-3991-2f5c-df95ddf12700@ti.com> Cc: davem@davemloft.net, kyle.roeschley@ni.com, netdev@vger.kernel.org, linux-kernel@vger.kernel.org From: Florian Fainelli Message-ID: <3abf0390-b91f-46f7-968d-0844f1a245a8@gmail.com> Date: Tue, 21 Mar 2017 09:00:40 -0700 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101 Thunderbird/45.7.0 MIME-Version: 1.0 In-Reply-To: <625a9ee1-ea65-3991-2f5c-df95ddf12700@ti.com> Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 03/21/2017 03:09 AM, Roger Quadros wrote: > On 20/03/17 18:41, Florian Fainelli wrote: >> On 03/16/2017 12:46 AM, Roger Quadros wrote: >>> On 15/03/17 17:49, Andrew Lunn wrote: >>>> On Wed, Mar 15, 2017 at 05:00:08PM +0200, Roger Quadros wrote: >>>>> Andrew, >>>>> >>>>> On 15/03/17 16:08, Andrew Lunn wrote: >>>>>> On Wed, Mar 15, 2017 at 03:51:27PM +0200, Roger Quadros wrote: >>>>>>> Since commit 3c293f4e08b5 ("net: phy: Trigger state machine on state change and not polling.") >>>>>>> phy_suspend() doesn't get called as part of phy_stop() for PHYs using >>>>>>> interrupts because the phy state machine is never triggered after a phy_stop(). >>>>>>> >>>>>>> Explicitly trigger the PHY state machine so that it can >>>>>>> see the new PHY state (HALTED) and suspend the PHY. >>>>>>> >>>>>>> Signed-off-by: Roger Quadros >>>>>> >>>>>> Hi Roger >>>>>> >>>>>> This seems sensible. It mirrors what phy_start() does. >>>>>> >>>>>> Reviewed-by: Andrew Lunn >>>>> >>>>> The reason for this being an RFC was the following comment just before >>>>> where I add the phy_trigger_machine() >>>>> >>>>> /* Cannot call flush_scheduled_work() here as desired because >>>>> * of rtnl_lock(), but PHY_HALTED shall guarantee phy_change() >>>>> * will not reenable interrupts. >>>>> */ >>>>> >>>>> Is this comment still applicable? If yes, is it OK to call >>>>> phy_trigger_machine() there? >>>> >>>> Humm, good question. >>>> >>>> My _guess_ would be, calling it with sync=True could >>>> deadlock. sync=False is probably safe. But lets see what Florian says. >>> >>> I agree that we should use phy_trigger_machine() with sync=False. >>> >>>> >>>>> >>>>>> >>>>>> It does however lead to a follow up question. Are there other places >>>>>> phydev->state is changed and it is missing a phy_trigger_machine()? >>>>>> >>>>> >>>>> One other place I think we should add phy_trigger_machine() is phy_start_aneg(). >>>> >>>> Humm, that might get us into a tight loop. >>>> >>>> phy_start_aneg() kicks the phy driver to start autoneg and sets >>>> phydev->state = PHY_AN. >>>> >>>> phy_trigger_machine() triggers the state machine immediately. >>>> >>>> In state PHY_AN, we check if aneg is done. If not, it sets needs_aneg >>>> = true. At the end of the state machine, this then calls >>>> phy_start_aneg(), and it all starts again. >>>> >>>> We are missing the 1s delay we have with polling. So for >>>> phy_start_aneg(), we might need a phy_delayed_trigger_machine(), which >>>> waits a second before doing anything? >>> >>> I think that should do the trick. >>> >>> How about this? >> >> This sounds like a possible fix indeed, however I would like to better >> assess the impact on non interrupt driven PHYs before rolling such a change. > > Is it safer if I add a check to do this only for interrupt driven PHYs? Yes I think this is a good solution that would not impact polled PHYs. Can you submit a formal patch for that change? Thanks! > > e.g. > > diff --git a/drivers/net/phy/phy.c b/drivers/net/phy/phy.c > index 4b855f2..e0f5755 100644 > --- a/drivers/net/phy/phy.c > +++ b/drivers/net/phy/phy.c > @@ -630,6 +630,9 @@ int phy_start_aneg(struct phy_device *phydev) > > out_unlock: > mutex_unlock(&phydev->lock); > + if (!err && phy_interrupt_is_valid(phydev)) > + queue_delayed_work(system_power_efficient_wq, &phydev->state_queue, HZ); > + > return err; > } > EXPORT_SYMBOL(phy_start_aneg); > -- Florian