| Message ID | 20170928150520.27616-1-gregory.clement@free-electrons.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi, On jeu., sept. 28 2017, Gregory CLEMENT <gregory.clement@free-electrons.com> wrote: > diff --git a/drivers/mmc/host/sdhci-xenon.c b/drivers/mmc/host/sdhci-xenon.c > index 2eec2e652c53..845575593b59 100644 > --- a/drivers/mmc/host/sdhci-xenon.c > +++ b/drivers/mmc/host/sdhci-xenon.c > @@ -466,6 +466,7 @@ static int xenon_probe(struct platform_device *pdev) > { > struct sdhci_pltfm_host *pltfm_host; > struct sdhci_host *host; > + struct xenon_priv *priv; > int err; > > host = sdhci_pltfm_init(pdev, &sdhci_xenon_pdata, > @@ -474,6 +475,7 @@ static int xenon_probe(struct platform_device *pdev) > return PTR_ERR(host); > > pltfm_host = sdhci_priv(host); > + priv = sdhci_pltfm_priv(pltfm_host); > > /* > * Link Xenon specific mmc_host_ops function, > @@ -491,9 +493,20 @@ static int xenon_probe(struct platform_device *pdev) > if (err) > goto free_pltfm; > > + priv->axi_clk = devm_clk_get(&pdev->dev, "axi"); > + if (IS_ERR(priv->axi_clk)) { > + err = PTR_ERR(priv->axi_clk); > + dev_err(&pdev->dev, "Failed to setup axi clk: %d\n", err); I should manage the -EPROBE_DEFER case here. So a v2 of this patch is coming soon. Sorry to not have seen this before sending the patch. Gregory > + priv->axi_clk = NULL; > + } else { > + err = clk_prepare_enable(priv->axi_clk); > + if (err) > + goto err_clk; > + } > + > err = mmc_of_parse(host->mmc); > if (err) > - goto err_clk; > + goto err_clk_axi; > > sdhci_get_of_property(pdev); > > @@ -502,11 +515,11 @@ static int xenon_probe(struct platform_device *pdev) > /* Xenon specific dt parse */ > err = xenon_probe_dt(pdev); > if (err) > - goto err_clk; > + goto err_clk_axi; > > err = xenon_sdhc_prepare(host); > if (err) > - goto err_clk; > + goto err_clk_axi; > > pm_runtime_get_noresume(&pdev->dev); > pm_runtime_set_active(&pdev->dev); > @@ -527,6 +540,9 @@ static int xenon_probe(struct platform_device *pdev) > pm_runtime_disable(&pdev->dev); > pm_runtime_put_noidle(&pdev->dev); > xenon_sdhc_unprepare(host); > +err_clk_axi: > + if (priv->axi_clk) > + clk_disable_unprepare(priv->axi_clk); > err_clk: > clk_disable_unprepare(pltfm_host->clk); > free_pltfm: > @@ -538,6 +554,7 @@ static int xenon_remove(struct platform_device *pdev) > { > struct sdhci_host *host = platform_get_drvdata(pdev); > struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host); > + struct xenon_priv *priv = sdhci_pltfm_priv(pltfm_host); > > pm_runtime_get_sync(&pdev->dev); > pm_runtime_disable(&pdev->dev); > @@ -546,6 +563,8 @@ static int xenon_remove(struct platform_device *pdev) > sdhci_remove_host(host, 0); > > xenon_sdhc_unprepare(host); > + if (priv->axi_clk) > + clk_disable_unprepare(priv->axi_clk); > > clk_disable_unprepare(pltfm_host->clk); > > diff --git a/drivers/mmc/host/sdhci-xenon.h b/drivers/mmc/host/sdhci-xenon.h > index 2bc0510c0769..9994995c7c56 100644 > --- a/drivers/mmc/host/sdhci-xenon.h > +++ b/drivers/mmc/host/sdhci-xenon.h > @@ -83,6 +83,7 @@ struct xenon_priv { > unsigned char bus_width; > unsigned char timing; > unsigned int clock; > + struct clk *axi_clk; > > int phy_type; > /* > -- > 2.14.1 >
+ DT maintainers On Thu, 28 Sep 2017 17:05:20 +0200 Gregory CLEMENT wrote: > On Armada 7K/8K we need to explicitly enable the bus clock. The bus clock > is optional because not all the SoCs need them but at least for Armada > 7K/8K it is actually mandatory. > > The binding documentation is updating accordingly. > > Without this patch the kernel hand during boot if the mvpp2.2 network > driver was not present in the kernel. Indeed the clock needed by the > xenon controller was set by the network driver. > > Fixes: 3a3748dba881 ("mmc: sdhci-xenon: Add Marvell Xenon SDHC core > functionality)" > CC: Stable <stable@vger.kernel.org> > Signed-off-by: Gregory CLEMENT <gregory.clement@free-electrons.com> > --- > Hi Ulf and Adrian, > > This patch should be merged in 4.14-rc, as it fixes real issues. > > This patch maye looks like just as a nice clean-up but it is not. On > the earlier version of the series adding the support for the xenon > controller there was already this axi bus clock. But at this moment > the documentation for the clock on the Armada 7K/8K was missing and > there was no driver for it, so we just use the clock set by the > bootloader and never change them that's why it worked without any > visible problem. > > Then as explained in the commit log the network driver enabled the > clock for us, and it happened that it was setup before the xenon > driver. It is only thanks the new information we received on the > clocks of the Sov and with more exhaustive testes that we found this > issue and the fix for it. > > Thanks, > > Gregory > > > .../bindings/mmc/marvell,xenon-sdhci.txt | 12 ++++++----- > drivers/mmc/host/sdhci-xenon.c | 25 +++++++++++++++++++--- > drivers/mmc/host/sdhci-xenon.h | 1 + > 3 files changed, 30 insertions(+), 8 deletions(-) > > diff --git a/Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.txt b/Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.txt > index b878a1e305af..6f3db1f809ab 100644 > --- a/Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.txt > +++ b/Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.txt > @@ -16,11 +16,13 @@ Required Properties: > > - clocks: > Array of clocks required for SDHC. > - Require at least input clock for Xenon IP core. > + Require at least input clock for Xenon IP core. For Armada AP806 and > + CP110, the AXI clock is also mandatory. > > - clock-names: > Array of names corresponding to clocks property. > The input clock for Xenon IP core should be named as "core". > + The input clock for the AXI bus must be named as "axi". Is "bus" more suitable? > > - reg: > * For "marvell,armada-3700-sdhci", two register areas. > @@ -106,8 +108,8 @@ Example: > compatible = "marvell,armada-ap806-sdhci"; > reg = <0xaa0000 0x1000>; > interrupts = <GIC_SPI 13 IRQ_TYPE_LEVEL_HIGH> > - clocks = <&emmc_clk>; > - clock-names = "core"; > + clocks = <&emmc_clk>,<&axi_clk>; > + clock-names = "core","axi"; > bus-width = <4>; > marvell,xenon-phy-slow-mode; > marvell,xenon-tun-count = <11>; > @@ -126,8 +128,8 @@ Example: > interrupts = <GIC_SPI 55 IRQ_TYPE_LEVEL_HIGH> > vqmmc-supply = <&sd_vqmmc_regulator>; > vmmc-supply = <&sd_vmmc_regulator>; > - clocks = <&sdclk>; > - clock-names = "core"; > + clocks = <&sdclk>,<&axi_clk>; > + clock-names = "core","axi"; > bus-width = <4>; > marvell,xenon-tun-count = <9>; > }; > diff --git a/drivers/mmc/host/sdhci-xenon.c b/drivers/mmc/host/sdhci-xenon.c > index 2eec2e652c53..845575593b59 100644 > --- a/drivers/mmc/host/sdhci-xenon.c > +++ b/drivers/mmc/host/sdhci-xenon.c > @@ -466,6 +466,7 @@ static int xenon_probe(struct platform_device *pdev) > { > struct sdhci_pltfm_host *pltfm_host; > struct sdhci_host *host; > + struct xenon_priv *priv; > int err; > > host = sdhci_pltfm_init(pdev, &sdhci_xenon_pdata, > @@ -474,6 +475,7 @@ static int xenon_probe(struct platform_device *pdev) > return PTR_ERR(host); > > pltfm_host = sdhci_priv(host); > + priv = sdhci_pltfm_priv(pltfm_host); > > /* > * Link Xenon specific mmc_host_ops function, > @@ -491,9 +493,20 @@ static int xenon_probe(struct platform_device *pdev) > if (err) > goto free_pltfm; > > + priv->axi_clk = devm_clk_get(&pdev->dev, "axi"); > + if (IS_ERR(priv->axi_clk)) { > + err = PTR_ERR(priv->axi_clk); > + dev_err(&pdev->dev, "Failed to setup axi clk: %d\n", err); Since the bus clock is optional, can we remove this err msg? > + priv->axi_clk = NULL; > + } else { > + err = clk_prepare_enable(priv->axi_clk); > + if (err) > + goto err_clk; > + } > + > err = mmc_of_parse(host->mmc); > if (err) > - goto err_clk; > + goto err_clk_axi; > > sdhci_get_of_property(pdev); > > @@ -502,11 +515,11 @@ static int xenon_probe(struct platform_device *pdev) > /* Xenon specific dt parse */ > err = xenon_probe_dt(pdev); > if (err) > - goto err_clk; > + goto err_clk_axi; > > err = xenon_sdhc_prepare(host); > if (err) > - goto err_clk; > + goto err_clk_axi; > > pm_runtime_get_noresume(&pdev->dev); > pm_runtime_set_active(&pdev->dev); > @@ -527,6 +540,9 @@ static int xenon_probe(struct platform_device *pdev) > pm_runtime_disable(&pdev->dev); > pm_runtime_put_noidle(&pdev->dev); > xenon_sdhc_unprepare(host); > +err_clk_axi: > + if (priv->axi_clk) clk_disable_unprepare() will do the check. So this check could be removed to simplify the err path > + clk_disable_unprepare(priv->axi_clk); > err_clk: > clk_disable_unprepare(pltfm_host->clk); > free_pltfm: > @@ -538,6 +554,7 @@ static int xenon_remove(struct platform_device *pdev) > { > struct sdhci_host *host = platform_get_drvdata(pdev); > struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host); > + struct xenon_priv *priv = sdhci_pltfm_priv(pltfm_host); > > pm_runtime_get_sync(&pdev->dev); > pm_runtime_disable(&pdev->dev); > @@ -546,6 +563,8 @@ static int xenon_remove(struct platform_device *pdev) > sdhci_remove_host(host, 0); > > xenon_sdhc_unprepare(host); > + if (priv->axi_clk) > + clk_disable_unprepare(priv->axi_clk); ditto > > clk_disable_unprepare(pltfm_host->clk); > > diff --git a/drivers/mmc/host/sdhci-xenon.h b/drivers/mmc/host/sdhci-xenon.h > index 2bc0510c0769..9994995c7c56 100644 > --- a/drivers/mmc/host/sdhci-xenon.h > +++ b/drivers/mmc/host/sdhci-xenon.h > @@ -83,6 +83,7 @@ struct xenon_priv { > unsigned char bus_width; > unsigned char timing; > unsigned int clock; > + struct clk *axi_clk; > > int phy_type; > /*
Hi Jisheng, On ven., sept. 29 2017, Jisheng Zhang <Jisheng.Zhang@synaptics.com> wrote: > + DT maintainers > > On Thu, 28 Sep 2017 17:05:20 +0200 Gregory CLEMENT wrote: > >> On Armada 7K/8K we need to explicitly enable the bus clock. The bus clock >> is optional because not all the SoCs need them but at least for Armada >> 7K/8K it is actually mandatory. >> >> The binding documentation is updating accordingly. >> >> Without this patch the kernel hand during boot if the mvpp2.2 network >> driver was not present in the kernel. Indeed the clock needed by the >> xenon controller was set by the network driver. >> >> Fixes: 3a3748dba881 ("mmc: sdhci-xenon: Add Marvell Xenon SDHC core >> functionality)" >> CC: Stable <stable@vger.kernel.org> >> Signed-off-by: Gregory CLEMENT <gregory.clement@free-electrons.com> >> --- >> Hi Ulf and Adrian, >> >> This patch should be merged in 4.14-rc, as it fixes real issues. >> >> This patch maye looks like just as a nice clean-up but it is not. On >> the earlier version of the series adding the support for the xenon >> controller there was already this axi bus clock. But at this moment >> the documentation for the clock on the Armada 7K/8K was missing and >> there was no driver for it, so we just use the clock set by the >> bootloader and never change them that's why it worked without any >> visible problem. >> >> Then as explained in the commit log the network driver enabled the >> clock for us, and it happened that it was setup before the xenon >> driver. It is only thanks the new information we received on the >> clocks of the Sov and with more exhaustive testes that we found this >> issue and the fix for it. >> >> Thanks, >> >> Gregory >> >> >> .../bindings/mmc/marvell,xenon-sdhci.txt | 12 ++++++----- >> drivers/mmc/host/sdhci-xenon.c | 25 +++++++++++++++++++--- >> drivers/mmc/host/sdhci-xenon.h | 1 + >> 3 files changed, 30 insertions(+), 8 deletions(-) >> >> diff --git a/Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.txt b/Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.txt >> index b878a1e305af..6f3db1f809ab 100644 >> --- a/Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.txt >> +++ b/Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.txt >> @@ -16,11 +16,13 @@ Required Properties: >> >> - clocks: >> Array of clocks required for SDHC. >> - Require at least input clock for Xenon IP core. >> + Require at least input clock for Xenon IP core. For Armada AP806 and >> + CP110, the AXI clock is also mandatory. >> >> - clock-names: >> Array of names corresponding to clocks property. >> The input clock for Xenon IP core should be named as "core". >> + The input clock for the AXI bus must be named as "axi". > > Is "bus" more suitable? I find bus too vague. > >> >> - reg: >> * For "marvell,armada-3700-sdhci", two register areas. >> @@ -106,8 +108,8 @@ Example: >> compatible = "marvell,armada-ap806-sdhci"; >> reg = <0xaa0000 0x1000>; >> interrupts = <GIC_SPI 13 IRQ_TYPE_LEVEL_HIGH> >> - clocks = <&emmc_clk>; >> - clock-names = "core"; >> + clocks = <&emmc_clk>,<&axi_clk>; >> + clock-names = "core","axi"; >> bus-width = <4>; >> marvell,xenon-phy-slow-mode; >> marvell,xenon-tun-count = <11>; >> @@ -126,8 +128,8 @@ Example: >> interrupts = <GIC_SPI 55 IRQ_TYPE_LEVEL_HIGH> >> vqmmc-supply = <&sd_vqmmc_regulator>; >> vmmc-supply = <&sd_vmmc_regulator>; >> - clocks = <&sdclk>; >> - clock-names = "core"; >> + clocks = <&sdclk>,<&axi_clk>; >> + clock-names = "core","axi"; >> bus-width = <4>; >> marvell,xenon-tun-count = <9>; >> }; >> diff --git a/drivers/mmc/host/sdhci-xenon.c b/drivers/mmc/host/sdhci-xenon.c >> index 2eec2e652c53..845575593b59 100644 >> --- a/drivers/mmc/host/sdhci-xenon.c >> +++ b/drivers/mmc/host/sdhci-xenon.c >> @@ -466,6 +466,7 @@ static int xenon_probe(struct platform_device *pdev) >> { >> struct sdhci_pltfm_host *pltfm_host; >> struct sdhci_host *host; >> + struct xenon_priv *priv; >> int err; >> >> host = sdhci_pltfm_init(pdev, &sdhci_xenon_pdata, >> @@ -474,6 +475,7 @@ static int xenon_probe(struct platform_device *pdev) >> return PTR_ERR(host); >> >> pltfm_host = sdhci_priv(host); >> + priv = sdhci_pltfm_priv(pltfm_host); >> >> /* >> * Link Xenon specific mmc_host_ops function, >> @@ -491,9 +493,20 @@ static int xenon_probe(struct platform_device *pdev) >> if (err) >> goto free_pltfm; >> >> + priv->axi_clk = devm_clk_get(&pdev->dev, "axi"); >> + if (IS_ERR(priv->axi_clk)) { >> + err = PTR_ERR(priv->axi_clk); >> + dev_err(&pdev->dev, "Failed to setup axi clk: %d\n", err); > > Since the bus clock is optional, can we remove this err msg? I will change it in dev_info. > >> + priv->axi_clk = NULL; >> + } else { >> + err = clk_prepare_enable(priv->axi_clk); >> + if (err) >> + goto err_clk; >> + } >> + >> err = mmc_of_parse(host->mmc); >> if (err) >> - goto err_clk; >> + goto err_clk_axi; >> >> sdhci_get_of_property(pdev); >> >> @@ -502,11 +515,11 @@ static int xenon_probe(struct platform_device *pdev) >> /* Xenon specific dt parse */ >> err = xenon_probe_dt(pdev); >> if (err) >> - goto err_clk; >> + goto err_clk_axi; >> >> err = xenon_sdhc_prepare(host); >> if (err) >> - goto err_clk; >> + goto err_clk_axi; >> >> pm_runtime_get_noresume(&pdev->dev); >> pm_runtime_set_active(&pdev->dev); >> @@ -527,6 +540,9 @@ static int xenon_probe(struct platform_device *pdev) >> pm_runtime_disable(&pdev->dev); >> pm_runtime_put_noidle(&pdev->dev); >> xenon_sdhc_unprepare(host); >> +err_clk_axi: >> + if (priv->axi_clk) > > clk_disable_unprepare() will do the check. So this check could be removed > to simplify the err path Right I removed it > >> + clk_disable_unprepare(priv->axi_clk); >> err_clk: >> clk_disable_unprepare(pltfm_host->clk); >> free_pltfm: >> @@ -538,6 +554,7 @@ static int xenon_remove(struct platform_device *pdev) >> { >> struct sdhci_host *host = platform_get_drvdata(pdev); >> struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host); >> + struct xenon_priv *priv = sdhci_pltfm_priv(pltfm_host); >> >> pm_runtime_get_sync(&pdev->dev); >> pm_runtime_disable(&pdev->dev); >> @@ -546,6 +563,8 @@ static int xenon_remove(struct platform_device *pdev) >> sdhci_remove_host(host, 0); >> >> xenon_sdhc_unprepare(host); >> + if (priv->axi_clk) >> + clk_disable_unprepare(priv->axi_clk); > > ditto OK Thanks, Gregory
Hello, On Fri, 29 Sep 2017 11:13:00 +0200, Gregory CLEMENT wrote: > >> - clock-names: > >> Array of names corresponding to clocks property. > >> The input clock for Xenon IP core should be named as "core". > >> + The input clock for the AXI bus must be named as "axi". > > > > Is "bus" more suitable? > > I find bus too vague. Actually no, I believe "bus" is more common. Indeed, you could have the same IP, attached to a different interconnect I guess. > >> /* > >> * Link Xenon specific mmc_host_ops function, > >> @@ -491,9 +493,20 @@ static int xenon_probe(struct platform_device *pdev) > >> if (err) > >> goto free_pltfm; > >> > >> + priv->axi_clk = devm_clk_get(&pdev->dev, "axi"); > >> + if (IS_ERR(priv->axi_clk)) { -EPROBE_DEFER handling ? > >> + err = PTR_ERR(priv->axi_clk); > >> + dev_err(&pdev->dev, "Failed to setup axi clk: %d\n", err); > > > > Since the bus clock is optional, can we remove this err msg? > > I will change it in dev_info. This clock is optional (in order to keep DT backward compatibility), so there shouldn't be any error message, be it dev_err() or dev_info(). Best regards, Thomas
Hi Thomas, On ven., sept. 29 2017, Thomas Petazzoni <thomas.petazzoni@free-electrons.com> wrote: > Hello, > > On Fri, 29 Sep 2017 11:13:00 +0200, Gregory CLEMENT wrote: > >> >> - clock-names: >> >> Array of names corresponding to clocks property. >> >> The input clock for Xenon IP core should be named as "core". >> >> + The input clock for the AXI bus must be named as "axi". >> > >> > Is "bus" more suitable? >> >> I find bus too vague. > > Actually no, I believe "bus" is more common. Indeed, you could have the > same IP, attached to a different interconnect I guess. > >> >> /* >> >> * Link Xenon specific mmc_host_ops function, >> >> @@ -491,9 +493,20 @@ static int xenon_probe(struct platform_device *pdev) >> >> if (err) >> >> goto free_pltfm; >> >> >> >> + priv->axi_clk = devm_clk_get(&pdev->dev, "axi"); >> >> + if (IS_ERR(priv->axi_clk)) { > > -EPROBE_DEFER handling ? See http://lists.infradead.org/pipermail/linux-arm-kernel/2017-September/533510.html Gregory
Hi Gregory, 2017-09-28 23:05 GMT+08:00 Gregory CLEMENT <gregory.clement@free-electrons.com>: > On Armada 7K/8K we need to explicitly enable the bus clock. The bus clock > is optional because not all the SoCs need them but at least for Armada > 7K/8K it is actually mandatory. > > The binding documentation is updating accordingly. > > Without this patch the kernel hand during boot if the mvpp2.2 network > driver was not present in the kernel. Indeed the clock needed by the > xenon controller was set by the network driver. > > Fixes: 3a3748dba881 ("mmc: sdhci-xenon: Add Marvell Xenon SDHC core > functionality)" > CC: Stable <stable@vger.kernel.org> > Signed-off-by: Gregory CLEMENT <gregory.clement@free-electrons.com> > --- > Hi Ulf and Adrian, > > This patch should be merged in 4.14-rc, as it fixes real issues. > > This patch maye looks like just as a nice clean-up but it is not. On > the earlier version of the series adding the support for the xenon > controller there was already this axi bus clock. But at this moment > the documentation for the clock on the Armada 7K/8K was missing and > there was no driver for it, so we just use the clock set by the > bootloader and never change them that's why it worked without any > visible problem. > > Then as explained in the commit log the network driver enabled the > clock for us, and it happened that it was setup before the xenon > driver. It is only thanks the new information we received on the > clocks of the Sov and with more exhaustive testes that we found this > issue and the fix for it. I know very little about clk in Linux. But I just happened to review this very issue before I left Marvell. I had some concerns about this patch at that time. Please check the following. 1. If that network device is used again, so that clock will be setup *twice* since there are two duplicated clock setup in that network driver and Xenon driver. Right? 2. If that network device is used again and Xenon later exists and shutdown the clock, will that network device be corrupted? 3. In my very own opinion, such a shared AXI bus clock might be controlled in u-boot or a "SoC-level" init routine. Please correct me if there is any misunderstanding. Thank you. Best regards, Hu Ziji > > Thanks, > > Gregory > > > .../bindings/mmc/marvell,xenon-sdhci.txt | 12 ++++++----- > drivers/mmc/host/sdhci-xenon.c | 25 +++++++++++++++++++--- > drivers/mmc/host/sdhci-xenon.h | 1 + > 3 files changed, 30 insertions(+), 8 deletions(-) > > diff --git a/Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.txt b/Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.txt > index b878a1e305af..6f3db1f809ab 100644 > --- a/Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.txt > +++ b/Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.txt > @@ -16,11 +16,13 @@ Required Properties: > > - clocks: > Array of clocks required for SDHC. > - Require at least input clock for Xenon IP core. > + Require at least input clock for Xenon IP core. For Armada AP806 and > + CP110, the AXI clock is also mandatory. > > - clock-names: > Array of names corresponding to clocks property. > The input clock for Xenon IP core should be named as "core". > + The input clock for the AXI bus must be named as "axi". > > - reg: > * For "marvell,armada-3700-sdhci", two register areas. > @@ -106,8 +108,8 @@ Example: > compatible = "marvell,armada-ap806-sdhci"; > reg = <0xaa0000 0x1000>; > interrupts = <GIC_SPI 13 IRQ_TYPE_LEVEL_HIGH> > - clocks = <&emmc_clk>; > - clock-names = "core"; > + clocks = <&emmc_clk>,<&axi_clk>; > + clock-names = "core","axi"; > bus-width = <4>; > marvell,xenon-phy-slow-mode; > marvell,xenon-tun-count = <11>; > @@ -126,8 +128,8 @@ Example: > interrupts = <GIC_SPI 55 IRQ_TYPE_LEVEL_HIGH> > vqmmc-supply = <&sd_vqmmc_regulator>; > vmmc-supply = <&sd_vmmc_regulator>; > - clocks = <&sdclk>; > - clock-names = "core"; > + clocks = <&sdclk>,<&axi_clk>; > + clock-names = "core","axi"; > bus-width = <4>; > marvell,xenon-tun-count = <9>; > }; > diff --git a/drivers/mmc/host/sdhci-xenon.c b/drivers/mmc/host/sdhci-xenon.c > index 2eec2e652c53..845575593b59 100644 > --- a/drivers/mmc/host/sdhci-xenon.c > +++ b/drivers/mmc/host/sdhci-xenon.c > @@ -466,6 +466,7 @@ static int xenon_probe(struct platform_device *pdev) > { > struct sdhci_pltfm_host *pltfm_host; > struct sdhci_host *host; > + struct xenon_priv *priv; > int err; > > host = sdhci_pltfm_init(pdev, &sdhci_xenon_pdata, > @@ -474,6 +475,7 @@ static int xenon_probe(struct platform_device *pdev) > return PTR_ERR(host); > > pltfm_host = sdhci_priv(host); > + priv = sdhci_pltfm_priv(pltfm_host); > > /* > * Link Xenon specific mmc_host_ops function, > @@ -491,9 +493,20 @@ static int xenon_probe(struct platform_device *pdev) > if (err) > goto free_pltfm; > > + priv->axi_clk = devm_clk_get(&pdev->dev, "axi"); > + if (IS_ERR(priv->axi_clk)) { > + err = PTR_ERR(priv->axi_clk); > + dev_err(&pdev->dev, "Failed to setup axi clk: %d\n", err); > + priv->axi_clk = NULL; > + } else { > + err = clk_prepare_enable(priv->axi_clk); > + if (err) > + goto err_clk; > + } > + > err = mmc_of_parse(host->mmc); > if (err) > - goto err_clk; > + goto err_clk_axi; > > sdhci_get_of_property(pdev); > > @@ -502,11 +515,11 @@ static int xenon_probe(struct platform_device *pdev) > /* Xenon specific dt parse */ > err = xenon_probe_dt(pdev); > if (err) > - goto err_clk; > + goto err_clk_axi; > > err = xenon_sdhc_prepare(host); > if (err) > - goto err_clk; > + goto err_clk_axi; > > pm_runtime_get_noresume(&pdev->dev); > pm_runtime_set_active(&pdev->dev); > @@ -527,6 +540,9 @@ static int xenon_probe(struct platform_device *pdev) > pm_runtime_disable(&pdev->dev); > pm_runtime_put_noidle(&pdev->dev); > xenon_sdhc_unprepare(host); > +err_clk_axi: > + if (priv->axi_clk) > + clk_disable_unprepare(priv->axi_clk); > err_clk: > clk_disable_unprepare(pltfm_host->clk); > free_pltfm: > @@ -538,6 +554,7 @@ static int xenon_remove(struct platform_device *pdev) > { > struct sdhci_host *host = platform_get_drvdata(pdev); > struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host); > + struct xenon_priv *priv = sdhci_pltfm_priv(pltfm_host); > > pm_runtime_get_sync(&pdev->dev); > pm_runtime_disable(&pdev->dev); > @@ -546,6 +563,8 @@ static int xenon_remove(struct platform_device *pdev) > sdhci_remove_host(host, 0); > > xenon_sdhc_unprepare(host); > + if (priv->axi_clk) > + clk_disable_unprepare(priv->axi_clk); > > clk_disable_unprepare(pltfm_host->clk); > > diff --git a/drivers/mmc/host/sdhci-xenon.h b/drivers/mmc/host/sdhci-xenon.h > index 2bc0510c0769..9994995c7c56 100644 > --- a/drivers/mmc/host/sdhci-xenon.h > +++ b/drivers/mmc/host/sdhci-xenon.h > @@ -83,6 +83,7 @@ struct xenon_priv { > unsigned char bus_width; > unsigned char timing; > unsigned int clock; > + struct clk *axi_clk; > > int phy_type; > /* > -- > 2.14.1 > > -- > To unsubscribe from this list: send the line "unsubscribe linux-mmc" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Ziji, On Sat, 30 Sep 2017 10:47:39 +0800 Ziji Hu wrote: > Hi Gregory, > > 2017-09-28 23:05 GMT+08:00 Gregory CLEMENT <gregory.clement@free-electrons.com>: > > On Armada 7K/8K we need to explicitly enable the bus clock. The bus clock > > is optional because not all the SoCs need them but at least for Armada > > 7K/8K it is actually mandatory. > > > > The binding documentation is updating accordingly. > > > > Without this patch the kernel hand during boot if the mvpp2.2 network > > driver was not present in the kernel. Indeed the clock needed by the > > xenon controller was set by the network driver. > > > > Fixes: 3a3748dba881 ("mmc: sdhci-xenon: Add Marvell Xenon SDHC core > > functionality)" > > CC: Stable <stable@vger.kernel.org> > > Signed-off-by: Gregory CLEMENT <gregory.clement@free-electrons.com> > > --- > > Hi Ulf and Adrian, > > > > This patch should be merged in 4.14-rc, as it fixes real issues. > > > > This patch maye looks like just as a nice clean-up but it is not. On > > the earlier version of the series adding the support for the xenon > > controller there was already this axi bus clock. But at this moment > > the documentation for the clock on the Armada 7K/8K was missing and > > there was no driver for it, so we just use the clock set by the > > bootloader and never change them that's why it worked without any > > visible problem. > > > > Then as explained in the commit log the network driver enabled the > > clock for us, and it happened that it was setup before the xenon > > driver. It is only thanks the new information we received on the > > clocks of the Sov and with more exhaustive testes that we found this > > issue and the fix for it. > > I know very little about clk in Linux. > But I just happened to review this very issue before I left Marvell. > I had some concerns about this patch at that time. > Please check the following. > > 1. If that network device is used again, so that clock will be setup *twice* > since there are two duplicated clock setup in that network driver > and Xenon driver. Right? The first "setup" will be executed, the second one just increase the enable and prepare count. > 2. If that network device is used again and Xenon later exists and > shutdown the clock, > will that network device be corrupted? No. The enable and prepare count is reduced to 1, nothing else could happen. The clock won't be shutdown until there's no users. > 3. In my very own opinion, such a shared AXI bus clock might be > controlled in u-boot or > a "SoC-level" init routine. IMHO, it's the driver's responsibility to manage its own clocks. Thanks
diff --git a/Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.txt b/Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.txt index b878a1e305af..6f3db1f809ab 100644 --- a/Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.txt +++ b/Documentation/devicetree/bindings/mmc/marvell,xenon-sdhci.txt @@ -16,11 +16,13 @@ Required Properties: - clocks: Array of clocks required for SDHC. - Require at least input clock for Xenon IP core. + Require at least input clock for Xenon IP core. For Armada AP806 and + CP110, the AXI clock is also mandatory. - clock-names: Array of names corresponding to clocks property. The input clock for Xenon IP core should be named as "core". + The input clock for the AXI bus must be named as "axi". - reg: * For "marvell,armada-3700-sdhci", two register areas. @@ -106,8 +108,8 @@ Example: compatible = "marvell,armada-ap806-sdhci"; reg = <0xaa0000 0x1000>; interrupts = <GIC_SPI 13 IRQ_TYPE_LEVEL_HIGH> - clocks = <&emmc_clk>; - clock-names = "core"; + clocks = <&emmc_clk>,<&axi_clk>; + clock-names = "core","axi"; bus-width = <4>; marvell,xenon-phy-slow-mode; marvell,xenon-tun-count = <11>; @@ -126,8 +128,8 @@ Example: interrupts = <GIC_SPI 55 IRQ_TYPE_LEVEL_HIGH> vqmmc-supply = <&sd_vqmmc_regulator>; vmmc-supply = <&sd_vmmc_regulator>; - clocks = <&sdclk>; - clock-names = "core"; + clocks = <&sdclk>,<&axi_clk>; + clock-names = "core","axi"; bus-width = <4>; marvell,xenon-tun-count = <9>; }; diff --git a/drivers/mmc/host/sdhci-xenon.c b/drivers/mmc/host/sdhci-xenon.c index 2eec2e652c53..845575593b59 100644 --- a/drivers/mmc/host/sdhci-xenon.c +++ b/drivers/mmc/host/sdhci-xenon.c @@ -466,6 +466,7 @@ static int xenon_probe(struct platform_device *pdev) { struct sdhci_pltfm_host *pltfm_host; struct sdhci_host *host; + struct xenon_priv *priv; int err; host = sdhci_pltfm_init(pdev, &sdhci_xenon_pdata, @@ -474,6 +475,7 @@ static int xenon_probe(struct platform_device *pdev) return PTR_ERR(host); pltfm_host = sdhci_priv(host); + priv = sdhci_pltfm_priv(pltfm_host); /* * Link Xenon specific mmc_host_ops function, @@ -491,9 +493,20 @@ static int xenon_probe(struct platform_device *pdev) if (err) goto free_pltfm; + priv->axi_clk = devm_clk_get(&pdev->dev, "axi"); + if (IS_ERR(priv->axi_clk)) { + err = PTR_ERR(priv->axi_clk); + dev_err(&pdev->dev, "Failed to setup axi clk: %d\n", err); + priv->axi_clk = NULL; + } else { + err = clk_prepare_enable(priv->axi_clk); + if (err) + goto err_clk; + } + err = mmc_of_parse(host->mmc); if (err) - goto err_clk; + goto err_clk_axi; sdhci_get_of_property(pdev); @@ -502,11 +515,11 @@ static int xenon_probe(struct platform_device *pdev) /* Xenon specific dt parse */ err = xenon_probe_dt(pdev); if (err) - goto err_clk; + goto err_clk_axi; err = xenon_sdhc_prepare(host); if (err) - goto err_clk; + goto err_clk_axi; pm_runtime_get_noresume(&pdev->dev); pm_runtime_set_active(&pdev->dev); @@ -527,6 +540,9 @@ static int xenon_probe(struct platform_device *pdev) pm_runtime_disable(&pdev->dev); pm_runtime_put_noidle(&pdev->dev); xenon_sdhc_unprepare(host); +err_clk_axi: + if (priv->axi_clk) + clk_disable_unprepare(priv->axi_clk); err_clk: clk_disable_unprepare(pltfm_host->clk); free_pltfm: @@ -538,6 +554,7 @@ static int xenon_remove(struct platform_device *pdev) { struct sdhci_host *host = platform_get_drvdata(pdev); struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host); + struct xenon_priv *priv = sdhci_pltfm_priv(pltfm_host); pm_runtime_get_sync(&pdev->dev); pm_runtime_disable(&pdev->dev); @@ -546,6 +563,8 @@ static int xenon_remove(struct platform_device *pdev) sdhci_remove_host(host, 0); xenon_sdhc_unprepare(host); + if (priv->axi_clk) + clk_disable_unprepare(priv->axi_clk); clk_disable_unprepare(pltfm_host->clk); diff --git a/drivers/mmc/host/sdhci-xenon.h b/drivers/mmc/host/sdhci-xenon.h index 2bc0510c0769..9994995c7c56 100644 --- a/drivers/mmc/host/sdhci-xenon.h +++ b/drivers/mmc/host/sdhci-xenon.h @@ -83,6 +83,7 @@ struct xenon_priv { unsigned char bus_width; unsigned char timing; unsigned int clock; + struct clk *axi_clk; int phy_type; /*
On Armada 7K/8K we need to explicitly enable the bus clock. The bus clock is optional because not all the SoCs need them but at least for Armada 7K/8K it is actually mandatory. The binding documentation is updating accordingly. Without this patch the kernel hand during boot if the mvpp2.2 network driver was not present in the kernel. Indeed the clock needed by the xenon controller was set by the network driver. Fixes: 3a3748dba881 ("mmc: sdhci-xenon: Add Marvell Xenon SDHC core functionality)" CC: Stable <stable@vger.kernel.org> Signed-off-by: Gregory CLEMENT <gregory.clement@free-electrons.com> --- Hi Ulf and Adrian, This patch should be merged in 4.14-rc, as it fixes real issues. This patch maye looks like just as a nice clean-up but it is not. On the earlier version of the series adding the support for the xenon controller there was already this axi bus clock. But at this moment the documentation for the clock on the Armada 7K/8K was missing and there was no driver for it, so we just use the clock set by the bootloader and never change them that's why it worked without any visible problem. Then as explained in the commit log the network driver enabled the clock for us, and it happened that it was setup before the xenon driver. It is only thanks the new information we received on the clocks of the Sov and with more exhaustive testes that we found this issue and the fix for it. Thanks, Gregory .../bindings/mmc/marvell,xenon-sdhci.txt | 12 ++++++----- drivers/mmc/host/sdhci-xenon.c | 25 +++++++++++++++++++--- drivers/mmc/host/sdhci-xenon.h | 1 + 3 files changed, 30 insertions(+), 8 deletions(-)