From mboxrd@z Thu Jan 1 00:00:00 1970 From: Roger Quadros Subject: Re: [PATCH 3/6] xhci: probe phy before add usb_hcd Date: Tue, 2 Jun 2015 07:00:46 +0300 Message-ID: <556D2A6E.4070802@ti.com> References: <1432727283-20303-1-git-send-email-chunfeng.yun@mediatek.com> <1432727283-20303-4-git-send-email-chunfeng.yun@mediatek.com> <20150527121843.GA8611@saruman.tx.rr.com> <1433170337.12128.18.camel@mtksdaap41> <556C7770.70506@intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii"; Format="flowed" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <556C7770.70506-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: "Linux-mediatek" Errors-To: linux-mediatek-bounces+glpam-linux-mediatek=m.gmane.org-IAPFreCvJWM7uuMidbF8XUB+6BGkLq7r@public.gmane.org To: Mathias Nyman , Yingjoe Chen , balbi-l0cyMroinI0@public.gmane.org Cc: Mark Rutland , devicetree-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, Sascha Hauer , linux-usb-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, Matthias Brugger , Rob Herring , linux-mediatek-IAPFreCvJWM7uuMidbF8XUB+6BGkLq7r@public.gmane.org, chunfeng.yun-NuS5LvNUpcJWk0Htik3J/w@public.gmane.org, Maxime Ripard , linux-arm-kernel-IAPFreCvJWM7uuMidbF8XUB+6BGkLq7r@public.gmane.org List-Id: linux-mediatek@lists.infradead.org Mathias. On 01/06/15 18:17, Mathias Nyman wrote: > On 01.06.2015 17:52, Yingjoe Chen wrote: >> On Wed, 2015-05-27 at 07:18 -0500, Felipe Balbi wrote: >>> On Wed, May 27, 2015 at 07:48:00PM +0800, chunfeng.yun-NuS5LvNUpcJWk0Htik3J/w@public.gmane.org wrote: >>>> From: Chunfeng Yun >>>> >>>> find the phy driver before add primary usb_hcd to avoid acessing >>>> xHCI register which may hangup the system when the phy is not loaded >>>> yet and the related powers or clocks put in phy driver are not >>>> enabled. >>> >>> it seems like the same clock is needed by PHY and XHCI. This patch looks >>> incorrect. >> >> >> Hi, >> >> I agree that the driver should enable clock it used by itself and not >> depends on init order. This should be fixed. >> >> But in general, I think it make sense to only add hcd after all required >> resource are ready. At least it remove unnecessary calls to >> usb_add_hcd/usb_remove_hcd. Is it better if the commit message is >> changed to something like the below? >> >> Currently xhci_plat_probe() call usb_add_hcd before trying to init the >> phy. However if the phy is not ready at the moment, it have to remove >> the hcd and probe again later >> >> Change the init order so we only add hcd when all required resource are >> ready. >> > Hi > > Rogers Quadros patchseries that changes how HCDs are created and added for > xhci-plat was just sent forward. It first creates both HCDs, then adds them. > > Doesn't usb_add_hcd() as it's first task take care of the PHYs? > I'm hoping to remove that part completely from xhci-plat.c but haven't > looked into it properly yet. > usb_add_hcd() does if (IS_ENABLED(CONFIG_USB_PHY) && !hcd->usb_phy) { struct usb_phy *phy = usb_get_phy_dev(hcd->self.controller, 0); ... if (IS_ENABLED(CONFIG_GENERIC_PHY) && !hcd->phy) { struct phy *phy = phy_get(hcd->self.controller, "usb"); ... but xhci-plat.c does hcd->usb_phy = devm_usb_get_phy_by_phandle(&pdev->dev, "usb-phy", 0); So there needs to be some alignment. cheers, -roger