From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Grygorii.Strashko@linaro.org" Subject: Re: [PATCH] ARM: dra7xx: hwmod: drop .modulemode from pcie1/2 hwmods Date: Mon, 09 Feb 2015 22:52:51 +0800 Message-ID: <54D8C9C3.5090500@linaro.org> References: <1422978684-4826-1-git-send-email-grygorii.strashko@linaro.org> <54D874E9.1040302@ti.com> <54D88BB3.4060104@linaro.org> <54D8B515.201@ti.com> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8; format=flowed Content-Transfer-Encoding: QUOTED-PRINTABLE Return-path: Received: from mail-pa0-f45.google.com ([209.85.220.45]:55711 "EHLO mail-pa0-f45.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1759582AbbBIOw5 (ORCPT ); Mon, 9 Feb 2015 09:52:57 -0500 Received: by mail-pa0-f45.google.com with SMTP id et14so34328028pad.4 for ; Mon, 09 Feb 2015 06:52:57 -0800 (PST) In-Reply-To: <54D8B515.201@ti.com> Sender: linux-omap-owner@vger.kernel.org List-Id: linux-omap@vger.kernel.org To: Kishon Vijay Abraham I , Tony Lindgren , Paul Walmsley , linux-omap@vger.kernel.org Cc: sumit.semwal@linaro.org, linux-arm-kernel@lists.infradead.org, Nishanth Menon On 02/09/2015 09:24 PM, Kishon Vijay Abraham I wrote: > Hi, > > On Monday 09 February 2015 03:58 PM, Grygorii.Strashko@linaro.org wro= te: >> Hi Kishon, >> On 02/09/2015 04:50 PM, Kishon Vijay Abraham I wrote: >>> On Tuesday 03 February 2015 09:21 PM, grygorii.strashko@linaro.org = wrote: >>>> From: Grygorii Strashko >>>> >>>> Now DRA7xx pcie1/2 hwmods define PRCM configuration as following: >>>> .clkctrl_offs =3D DRA7XX_CM_PCIE_CLKSTCTRL_OFFSET, >>>> .rstctrl_offs =3D DRA7XX_RM_L3INIT_RSTCTRL_OFFSET, >>>> .modulemode =3D MODULEMODE_SWCTRL, >>>> which is completely wrong because DRA7XX_CM_PCIE_CLKSTCTRL_OFFSET >>>> is clockdomain ctrl register and NOT module ctrl register. >>>> And they have diffrent allowed values for bits[0,1]: >>>> CLKTRCTRL=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2= =80=89=E2=80=89=E2=80=89MODULEMODE=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2= =80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2= =80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2= =80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2= =80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89 >>>> 0x0:=E2=80=89NO_SLEEP=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89= 0x0:=E2=80=89Module=E2=80=89is=E2=80=89disabled=E2=80=89by=E2=80=89SW.=E2= =80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2= =80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89 >>>> 0x1:=E2=80=89SW_SLEEP=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89= 0x1:=E2=80=89Module=E2=80=89is=E2=80=89managed=E2=80=89automatically=E2= =80=89by=E2=80=89HW=E2=80=89=E2=80=89 >>>> 0x2:=E2=80=89SW_WKUP=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2= =80=890x2:=E2=80=89Module=E2=80=89is=E2=80=89explicitly=E2=80=89enabled= =2E=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80= =89=E2=80=89=E2=80=89 >>>> 0x3:=E2=80=89HW_AUTO=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2=80=89=E2= =80=890x3:=E2=80=89Reserved=E2=80=89=E2=80=89 >>>> >>>> As result, following message can be seen during suspend: >>>> "omap_hwmod: pcie1: _wait_target_disable failed" >>>> >>>> Fix it by removing .modulemode from pcie1/2 hwmods and, in that >>>> way, prevent clockdomain ctrl register writing from HWMOD core. >>> >>> Looks correct except for one change. >>> >>> Acked-by: Kishon Vijay Abraham I >>>> >>>> Signed-off-by: Grygorii Strashko >>>> --- >>>> >>>> More over, it looks like pcie1/2 hwmods are fake and have to be dr= opped at all. >>>> The real HWMODs are PCIESS1/2. >>> >>> Not sure I get this. You mean "dra7xx_pcie1_hwmod" should be replac= ed with >>> "dra7xx_pciess1_hwmod"? Or you mean an entire new hwmod is missing? >>> >>> Please note we still have to enable the clock domain and main clock= =2E We've also >>> purposefully omitted sysconfig from hwmod data since pcie reset >>> (RM_PCIESS_RSTCTRL) should be done before accessing the syconfig re= gister and >>> the infrastructure for that is currently not present. >> >> What I'm trying to say is that now PM control data mixed between "pc= ieX" and "pcieX-phy" hwmods. >> After this patch "pcieX" hwmods will actually do nothing (I assume t= hat "pciex-phy" will be >> enabled before "pcieX"), and probably can be removed if "pcie_clkdm"= could be attached to "pcieX-phy" hwmod >> instead. >> >> More over, now, "pcie_clkdm" is connected to "pcieX" hwmod while MOD= ULEMODE register is controlled >> by "pciex-phy" hwmod, so when pciess is going to be enabled the "l3i= nit_clkdm" will be waken-up by >> hwmode core and not "pcie_clkdm" - as I can remember this is not goo= d (we should alway wake-up clockdomain >> and keep it in SWSUP mode while changing MODULEMODE and SYSC regis= ters). >> >> static struct omap_hwmod dra7xx_pcie1_hwmod =3D { >> .name =3D "pcie1", >> .class =3D &dra7xx_pcie_hwmod_class, >> .clkdm_name =3D "pcie_clkdm", >> .main_clk =3D "l4_root_clk_div", >> >> static struct omap_hwmod dra7xx_pcie1_phy_hwmod =3D { >> .name =3D "pcie1-phy", >> .class =3D &dra7xx_pcie_phy_hwmod_class, >> .clkdm_name =3D "l3init_clkdm", >> .main_clk =3D "l4_root_clk_div", >> >> So, in my opinion, some rework may be needed here. >> Am I right? > > you are right. We should have a single hwmod like dra7xx_pciess1_hwmo= d whose > clkdm should be "pcie_clkdm" and whose clkctrl_offs should be > DRA7XX_CM_L3INIT_PCIESS1_CLKCTRL_OFFSET (for controlling MODULEMODE).= PCIE PHY > shouldn't have a hwmod entry at all. > > Thanks > Kishon > Could this patch be applied any way? It fixes real issue for me. --=20 regards, -grygorii -- To unsubscribe from this list: send the line "unsubscribe linux-omap" i= n the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html From mboxrd@z Thu Jan 1 00:00:00 1970 From: grygorii.strashko@linaro.org (Grygorii.Strashko@linaro.org) Date: Mon, 09 Feb 2015 22:52:51 +0800 Subject: [PATCH] ARM: dra7xx: hwmod: drop .modulemode from pcie1/2 hwmods In-Reply-To: <54D8B515.201@ti.com> References: <1422978684-4826-1-git-send-email-grygorii.strashko@linaro.org> <54D874E9.1040302@ti.com> <54D88BB3.4060104@linaro.org> <54D8B515.201@ti.com> Message-ID: <54D8C9C3.5090500@linaro.org> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org On 02/09/2015 09:24 PM, Kishon Vijay Abraham I wrote: > Hi, > > On Monday 09 February 2015 03:58 PM, Grygorii.Strashko at linaro.org wrote: >> Hi Kishon, >> On 02/09/2015 04:50 PM, Kishon Vijay Abraham I wrote: >>> On Tuesday 03 February 2015 09:21 PM, grygorii.strashko at linaro.org wrote: >>>> From: Grygorii Strashko >>>> >>>> Now DRA7xx pcie1/2 hwmods define PRCM configuration as following: >>>> .clkctrl_offs = DRA7XX_CM_PCIE_CLKSTCTRL_OFFSET, >>>> .rstctrl_offs = DRA7XX_RM_L3INIT_RSTCTRL_OFFSET, >>>> .modulemode = MODULEMODE_SWCTRL, >>>> which is completely wrong because DRA7XX_CM_PCIE_CLKSTCTRL_OFFSET >>>> is clockdomain ctrl register and NOT module ctrl register. >>>> And they have diffrent allowed values for bits[0,1]: >>>> CLKTRCTRL?????????MODULEMODE?????????????????????????????????? >>>> 0x0:?NO_SLEEP?????0x0:?Module?is?disabled?by?SW.?????????????? >>>> 0x1:?SW_SLEEP?????0x1:?Module?is?managed?automatically?by?HW?? >>>> 0x2:?SW_WKUP??????0x2:?Module?is?explicitly?enabled.?????????? >>>> 0x3:?HW_AUTO??????0x3:?Reserved?? >>>> >>>> As result, following message can be seen during suspend: >>>> "omap_hwmod: pcie1: _wait_target_disable failed" >>>> >>>> Fix it by removing .modulemode from pcie1/2 hwmods and, in that >>>> way, prevent clockdomain ctrl register writing from HWMOD core. >>> >>> Looks correct except for one change. >>> >>> Acked-by: Kishon Vijay Abraham I >>>> >>>> Signed-off-by: Grygorii Strashko >>>> --- >>>> >>>> More over, it looks like pcie1/2 hwmods are fake and have to be dropped at all. >>>> The real HWMODs are PCIESS1/2. >>> >>> Not sure I get this. You mean "dra7xx_pcie1_hwmod" should be replaced with >>> "dra7xx_pciess1_hwmod"? Or you mean an entire new hwmod is missing? >>> >>> Please note we still have to enable the clock domain and main clock. We've also >>> purposefully omitted sysconfig from hwmod data since pcie reset >>> (RM_PCIESS_RSTCTRL) should be done before accessing the syconfig register and >>> the infrastructure for that is currently not present. >> >> What I'm trying to say is that now PM control data mixed between "pcieX" and "pcieX-phy" hwmods. >> After this patch "pcieX" hwmods will actually do nothing (I assume that "pciex-phy" will be >> enabled before "pcieX"), and probably can be removed if "pcie_clkdm" could be attached to "pcieX-phy" hwmod >> instead. >> >> More over, now, "pcie_clkdm" is connected to "pcieX" hwmod while MODULEMODE register is controlled >> by "pciex-phy" hwmod, so when pciess is going to be enabled the "l3init_clkdm" will be waken-up by >> hwmode core and not "pcie_clkdm" - as I can remember this is not good (we should alway wake-up clockdomain >> and keep it in SWSUP mode while changing MODULEMODE and SYSC registers). >> >> static struct omap_hwmod dra7xx_pcie1_hwmod = { >> .name = "pcie1", >> .class = &dra7xx_pcie_hwmod_class, >> .clkdm_name = "pcie_clkdm", >> .main_clk = "l4_root_clk_div", >> >> static struct omap_hwmod dra7xx_pcie1_phy_hwmod = { >> .name = "pcie1-phy", >> .class = &dra7xx_pcie_phy_hwmod_class, >> .clkdm_name = "l3init_clkdm", >> .main_clk = "l4_root_clk_div", >> >> So, in my opinion, some rework may be needed here. >> Am I right? > > you are right. We should have a single hwmod like dra7xx_pciess1_hwmod whose > clkdm should be "pcie_clkdm" and whose clkctrl_offs should be > DRA7XX_CM_L3INIT_PCIESS1_CLKCTRL_OFFSET (for controlling MODULEMODE). PCIE PHY > shouldn't have a hwmod entry at all. > > Thanks > Kishon > Could this patch be applied any way? It fixes real issue for me. -- regards, -grygorii