[PATCH 0/5] Qualcomm DWC3 USB support

[PATCH 0/5] Qualcomm DWC3 USB support

[Caleb Connolly]

This series enables support for Qualcomm platforms in the DWC3 driver,
adds support for arbitrary sector sizes to the USB mass storage gadget,
and fixes an issue with the CDC ACM driver where it wouldn't initialise
the USB device.

Additionally, it fixes a syntax bug in the Qualcomm SMMU driver, and
makes USB_DWC3_GADGET select DM_USB_GADGET to fix compilation with
gadget mode.

This is part of a larger series enabling DWC3 USB support on Qualcomm
platforms, a feature branch with all patches can be found at [1].

[1]: https://source.denx.de/u-boot/custodians/u-boot-snapdragon/-/tree/b4/dwc3-qcom

---
Caleb Connolly (5):
      usb: dwc3-generic: implement Qualcomm wrapper
      usb: dwc3: select DM_USB_GADGET
      usb: gadget: CDC ACM: call usb_gadget_initialize
      usb: gadget: UMS: support multiple sector sizes
      iommu: qcom-smmu: fix debugging

 cmd/usb_mass_storage.c              |   4 --
 drivers/iommu/qcom-hyp-smmu.c       |   2 +-
 drivers/usb/dwc3/Kconfig            |   1 +
 drivers/usb/dwc3/dwc3-generic.c     |  99 ++++++++++++++++++++++++++++++++++-
 drivers/usb/gadget/f_acm.c          |   9 ++++
 drivers/usb/gadget/f_mass_storage.c | 101 ++++++++++++++++++++----------------
 drivers/usb/gadget/storage_common.c |  12 +++--
 include/usb_mass_storage.h          |   1 -
 8 files changed, 174 insertions(+), 55 deletions(-)
---
base-commit: 16d331e8f1581de1ac6283365774a57dc1403cbd

// Caleb (they/them)

[PATCH 1/5] usb: dwc3-generic: implement Qualcomm wrapper

[Caleb Connolly]

The Qualcomm specific dwc3 wrapper isn't hugely complicated, implemented
the missing initialisation for host and gadget mode.

Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>
---
 drivers/usb/dwc3/dwc3-generic.c | 99 ++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 98 insertions(+), 1 deletion(-)

diff --git a/drivers/usb/dwc3/dwc3-generic.c b/drivers/usb/dwc3/dwc3-generic.c
index 48da621ba966..1119cdecd26d 100644
--- a/drivers/usb/dwc3/dwc3-generic.c
+++ b/drivers/usb/dwc3/dwc3-generic.c
@@ -419,6 +419,99 @@ struct dwc3_glue_ops ti_ops = {
 	.glue_configure = dwc3_ti_glue_configure,
 };
 
+/* USB QSCRATCH Hardware registers */
+#define QSCRATCH_HS_PHY_CTRL 0x10
+#define UTMI_OTG_VBUS_VALID BIT(20)
+#define SW_SESSVLD_SEL BIT(28)
+
+#define QSCRATCH_SS_PHY_CTRL 0x30
+#define LANE0_PWR_PRESENT BIT(24)
+
+#define QSCRATCH_GENERAL_CFG 0x08
+#define PIPE_UTMI_CLK_SEL BIT(0)
+#define PIPE3_PHYSTATUS_SW BIT(3)
+#define PIPE_UTMI_CLK_DIS BIT(8)
+
+#define PWR_EVNT_IRQ_STAT_REG 0x58
+#define PWR_EVNT_LPM_IN_L2_MASK BIT(4)
+#define PWR_EVNT_LPM_OUT_L2_MASK BIT(5)
+
+#define SDM845_QSCRATCH_BASE_OFFSET 0xf8800
+#define SDM845_QSCRATCH_SIZE 0x400
+#define SDM845_DWC3_CORE_SIZE 0xcd00
+static inline void dwc3_qcom_setbits(void __iomem *base, u32 offset, u32 val)
+{
+	u32 reg;
+
+	reg = readl(base + offset);
+	reg |= val;
+	writel(reg, base + offset);
+
+	/* ensure that above write is through */
+	readl(base + offset);
+}
+
+static inline void dwc3_qcom_clrbits(void __iomem *base, u32 offset, u32 val)
+{
+	u32 reg;
+
+	reg = readl(base + offset);
+	reg &= ~val;
+	writel(reg, base + offset);
+
+	/* ensure that above write is through */
+	readl(base + offset);
+}
+
+static void dwc3_qcom_vbus_override_enable(void __iomem *qscratch_base, bool enable)
+{
+	if (enable) {
+		dwc3_qcom_setbits(qscratch_base, QSCRATCH_SS_PHY_CTRL,
+				  LANE0_PWR_PRESENT);
+		dwc3_qcom_setbits(qscratch_base, QSCRATCH_HS_PHY_CTRL,
+				  UTMI_OTG_VBUS_VALID | SW_SESSVLD_SEL);
+	} else {
+		dwc3_qcom_clrbits(qscratch_base, QSCRATCH_SS_PHY_CTRL,
+				  LANE0_PWR_PRESENT);
+		dwc3_qcom_clrbits(qscratch_base, QSCRATCH_HS_PHY_CTRL,
+				  UTMI_OTG_VBUS_VALID | SW_SESSVLD_SEL);
+	}
+}
+
+/* For controllers running without superspeed PHYs */
+static void dwc3_qcom_select_utmi_clk(void __iomem *qscratch_base)
+{
+	/* Configure dwc3 to use UTMI clock as PIPE clock not present */
+	dwc3_qcom_setbits(qscratch_base, QSCRATCH_GENERAL_CFG,
+			  PIPE_UTMI_CLK_DIS);
+
+	udelay(500);
+
+	dwc3_qcom_setbits(qscratch_base, QSCRATCH_GENERAL_CFG,
+			  PIPE_UTMI_CLK_SEL | PIPE3_PHYSTATUS_SW);
+
+	udelay(500);
+
+	dwc3_qcom_clrbits(qscratch_base, QSCRATCH_GENERAL_CFG,
+			  PIPE_UTMI_CLK_DIS);
+}
+
+static void dwc3_qcom_glue_configure(struct udevice *dev, int index,
+				     enum usb_dr_mode mode)
+{
+	void __iomem *qscratch_base = (void __iomem *)dev_read_addr(dev);
+
+	if (dev_read_bool(dev, "qcom,select-utmi-as-pipe-clk"))
+		dwc3_qcom_select_utmi_clk(qscratch_base);
+
+	if (mode != USB_DR_MODE_HOST)
+		dwc3_qcom_vbus_override_enable(qscratch_base, true);
+}
+
+struct dwc3_glue_ops qcom_ops = {
+	.glue_configure = dwc3_qcom_glue_configure,
+};
+
 static int dwc3_rk_glue_get_ctrl_dev(struct udevice *dev, ofnode *node)
 {
 	*node = dev_ofnode(dev);
@@ -506,6 +599,10 @@ static int dwc3_glue_reset_init(struct udevice *dev,
 	else if (ret)
 		return ret;
 
+	if (device_is_compatible(dev, "qcom,dwc3")) {
+		reset_assert_bulk(&glue->resets);
+		udelay(500);
+	}
 	ret = reset_deassert_bulk(&glue->resets);
 	if (ret) {
 		reset_release_bulk(&glue->resets);
@@ -623,7 +720,7 @@ static const struct udevice_id dwc3_glue_ids[] = {
 	{ .compatible = "rockchip,rk3399-dwc3" },
 	{ .compatible = "rockchip,rk3568-dwc3", .data = (ulong)&rk_ops },
 	{ .compatible = "rockchip,rk3588-dwc3", .data = (ulong)&rk_ops },
-	{ .compatible = "qcom,dwc3" },
+	{ .compatible = "qcom,dwc3", .data = (ulong)&qcom_ops },
 	{ .compatible = "fsl,imx8mp-dwc3", .data = (ulong)&imx8mp_ops },
 	{ .compatible = "fsl,imx8mq-dwc3" },
 	{ .compatible = "intel,tangier-dwc3" },

-- 
2.43.0

Re: [PATCH 1/5] usb: dwc3-generic: implement Qualcomm wrapper

[Mattijs Korpershoek]

Hi Caleb,

Thank you for the patch.

On mer., janv. 31, 2024 at 14:57, Caleb Connolly <caleb.connolly@linaro.org> wrote:

> The Qualcomm specific dwc3 wrapper isn't hugely complicated, implemented
> the missing initialisation for host and gadget mode.
>
> Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>
> ---
>  drivers/usb/dwc3/dwc3-generic.c | 99 ++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 98 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/usb/dwc3/dwc3-generic.c b/drivers/usb/dwc3/dwc3-generic.c
> index 48da621ba966..1119cdecd26d 100644
> --- a/drivers/usb/dwc3/dwc3-generic.c
> +++ b/drivers/usb/dwc3/dwc3-generic.c
> @@ -419,6 +419,99 @@ struct dwc3_glue_ops ti_ops = {
>  	.glue_configure = dwc3_ti_glue_configure,
>  };
>  
> +/* USB QSCRATCH Hardware registers */
> +#define QSCRATCH_HS_PHY_CTRL 0x10
> +#define UTMI_OTG_VBUS_VALID BIT(20)
> +#define SW_SESSVLD_SEL BIT(28)
> +
> +#define QSCRATCH_SS_PHY_CTRL 0x30
> +#define LANE0_PWR_PRESENT BIT(24)
> +
> +#define QSCRATCH_GENERAL_CFG 0x08
> +#define PIPE_UTMI_CLK_SEL BIT(0)
> +#define PIPE3_PHYSTATUS_SW BIT(3)
> +#define PIPE_UTMI_CLK_DIS BIT(8)
> +
> +#define PWR_EVNT_IRQ_STAT_REG 0x58
> +#define PWR_EVNT_LPM_IN_L2_MASK BIT(4)
> +#define PWR_EVNT_LPM_OUT_L2_MASK BIT(5)
> +
> +#define SDM845_QSCRATCH_BASE_OFFSET 0xf8800
> +#define SDM845_QSCRATCH_SIZE 0x400
> +#define SDM845_DWC3_CORE_SIZE 0xcd00
> +static inline void dwc3_qcom_setbits(void __iomem *base, u32 offset, u32 val)
> +{
> +	u32 reg;
> +
> +	reg = readl(base + offset);
> +	reg |= val;
> +	writel(reg, base + offset);

Why can't we use the setbits() macro here?
see: arch/arm/include/asm/io.h

> +
> +	/* ensure that above write is through */
> +	readl(base + offset);
> +}
> +
> +static inline void dwc3_qcom_clrbits(void __iomem *base, u32 offset, u32 val)
> +{
> +	u32 reg;
> +
> +	reg = readl(base + offset);
> +	reg &= ~val;
> +	writel(reg, base + offset);

Same question for clrbits()

> +
> +	/* ensure that above write is through */
> +	readl(base + offset);
> +}
> +
> +static void dwc3_qcom_vbus_override_enable(void __iomem *qscratch_base, bool enable)
> +{
> +	if (enable) {
> +		dwc3_qcom_setbits(qscratch_base, QSCRATCH_SS_PHY_CTRL,
> +				  LANE0_PWR_PRESENT);
> +		dwc3_qcom_setbits(qscratch_base, QSCRATCH_HS_PHY_CTRL,
> +				  UTMI_OTG_VBUS_VALID | SW_SESSVLD_SEL);
> +	} else {
> +		dwc3_qcom_clrbits(qscratch_base, QSCRATCH_SS_PHY_CTRL,
> +				  LANE0_PWR_PRESENT);
> +		dwc3_qcom_clrbits(qscratch_base, QSCRATCH_HS_PHY_CTRL,
> +				  UTMI_OTG_VBUS_VALID | SW_SESSVLD_SEL);
> +	}
> +}
> +
> +/* For controllers running without superspeed PHYs */
> +static void dwc3_qcom_select_utmi_clk(void __iomem *qscratch_base)
> +{
> +	/* Configure dwc3 to use UTMI clock as PIPE clock not present */
> +	dwc3_qcom_setbits(qscratch_base, QSCRATCH_GENERAL_CFG,
> +			  PIPE_UTMI_CLK_DIS);
> +
> +	udelay(500);
> +
> +	dwc3_qcom_setbits(qscratch_base, QSCRATCH_GENERAL_CFG,
> +			  PIPE_UTMI_CLK_SEL | PIPE3_PHYSTATUS_SW);
> +
> +	udelay(500);
> +
> +	dwc3_qcom_clrbits(qscratch_base, QSCRATCH_GENERAL_CFG,
> +			  PIPE_UTMI_CLK_DIS);
> +}
> +
> +static void dwc3_qcom_glue_configure(struct udevice *dev, int index,
> +				     enum usb_dr_mode mode)
> +{
> +	void __iomem *qscratch_base = (void __iomem *)dev_read_addr(dev);
> +
> +	if (dev_read_bool(dev, "qcom,select-utmi-as-pipe-clk"))
> +		dwc3_qcom_select_utmi_clk(qscratch_base);
> +
> +	if (mode != USB_DR_MODE_HOST)
> +		dwc3_qcom_vbus_override_enable(qscratch_base, true);
> +}
> +
> +struct dwc3_glue_ops qcom_ops = {
> +	.glue_configure = dwc3_qcom_glue_configure,
> +};
> +
>  static int dwc3_rk_glue_get_ctrl_dev(struct udevice *dev, ofnode *node)
>  {
>  	*node = dev_ofnode(dev);
> @@ -506,6 +599,10 @@ static int dwc3_glue_reset_init(struct udevice *dev,
>  	else if (ret)
>  		return ret;
>  
> +	if (device_is_compatible(dev, "qcom,dwc3")) {
> +		reset_assert_bulk(&glue->resets);

Any reason for not doing error handling on reset_assert_bulk() like it's
done below ?

> +		udelay(500);
> +	}
>  	ret = reset_deassert_bulk(&glue->resets);
>  	if (ret) {
>  		reset_release_bulk(&glue->resets);
> @@ -623,7 +720,7 @@ static const struct udevice_id dwc3_glue_ids[] = {
>  	{ .compatible = "rockchip,rk3399-dwc3" },
>  	{ .compatible = "rockchip,rk3568-dwc3", .data = (ulong)&rk_ops },
>  	{ .compatible = "rockchip,rk3588-dwc3", .data = (ulong)&rk_ops },
> -	{ .compatible = "qcom,dwc3" },
> +	{ .compatible = "qcom,dwc3", .data = (ulong)&qcom_ops },
>  	{ .compatible = "fsl,imx8mp-dwc3", .data = (ulong)&imx8mp_ops },
>  	{ .compatible = "fsl,imx8mq-dwc3" },
>  	{ .compatible = "intel,tangier-dwc3" },
>
> -- 
> 2.43.0

Re: [PATCH 1/5] usb: dwc3-generic: implement Qualcomm wrapper

[Caleb Connolly]

On 01/02/2024 09:34, Mattijs Korpershoek wrote:
> Hi Caleb,
> 
> Thank you for the patch.
> 
> On mer., janv. 31, 2024 at 14:57, Caleb Connolly <caleb.connolly@linaro.org> wrote:
> 
>> The Qualcomm specific dwc3 wrapper isn't hugely complicated, implemented
>> the missing initialisation for host and gadget mode.
>>
>> Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>
>> ---
>>  drivers/usb/dwc3/dwc3-generic.c | 99 ++++++++++++++++++++++++++++++++++++++++-
>>  1 file changed, 98 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/usb/dwc3/dwc3-generic.c b/drivers/usb/dwc3/dwc3-generic.c
>> index 48da621ba966..1119cdecd26d 100644
>> --- a/drivers/usb/dwc3/dwc3-generic.c
>> +++ b/drivers/usb/dwc3/dwc3-generic.c
>> @@ -419,6 +419,99 @@ struct dwc3_glue_ops ti_ops = {
>>  	.glue_configure = dwc3_ti_glue_configure,
>>  };
>>  
>> +/* USB QSCRATCH Hardware registers */
>> +#define QSCRATCH_HS_PHY_CTRL 0x10
>> +#define UTMI_OTG_VBUS_VALID BIT(20)
>> +#define SW_SESSVLD_SEL BIT(28)
>> +
>> +#define QSCRATCH_SS_PHY_CTRL 0x30
>> +#define LANE0_PWR_PRESENT BIT(24)
>> +
>> +#define QSCRATCH_GENERAL_CFG 0x08
>> +#define PIPE_UTMI_CLK_SEL BIT(0)
>> +#define PIPE3_PHYSTATUS_SW BIT(3)
>> +#define PIPE_UTMI_CLK_DIS BIT(8)
>> +
>> +#define PWR_EVNT_IRQ_STAT_REG 0x58
>> +#define PWR_EVNT_LPM_IN_L2_MASK BIT(4)
>> +#define PWR_EVNT_LPM_OUT_L2_MASK BIT(5)
>> +
>> +#define SDM845_QSCRATCH_BASE_OFFSET 0xf8800
>> +#define SDM845_QSCRATCH_SIZE 0x400
>> +#define SDM845_DWC3_CORE_SIZE 0xcd00
>> +static inline void dwc3_qcom_setbits(void __iomem *base, u32 offset, u32 val)
>> +{
>> +	u32 reg;
>> +
>> +	reg = readl(base + offset);
>> +	reg |= val;
>> +	writel(reg, base + offset);
> 
> Why can't we use the setbits() macro here?
> see: arch/arm/include/asm/io.h

setbits doesn't give us the same cache coherency guarantees I think(?)
the readl/writel macros include wmb() calls.

That said, I won't pretend to really know what I'm talking about here,
possible setbits() would be suitable, do you have any idea?
> 
>> +
>> +	/* ensure that above write is through */
>> +	readl(base + offset);
>> +}
>> +
>> +static inline void dwc3_qcom_clrbits(void __iomem *base, u32 offset, u32 val)
>> +{
>> +	u32 reg;
>> +
>> +	reg = readl(base + offset);
>> +	reg &= ~val;
>> +	writel(reg, base + offset);
> 
> Same question for clrbits()
> 
>> +
>> +	/* ensure that above write is through */
>> +	readl(base + offset);
>> +}
>> +

[snip]

>> +	if (device_is_compatible(dev, "qcom,dwc3")) {
>> +		reset_assert_bulk(&glue->resets);
> 
> Any reason for not doing error handling on reset_assert_bulk() like it's
> done below ?

Well, the Qualcomm reset driver will never return an error, and if it
did, the assert failing (presumably because it's already asserted) is
not necessarily an error we'd need to care about - it's only an error if
we can't deassert the reset.

So I think this is fine.
> 
>> +		udelay(500);
>> +	}
>>  	ret = reset_deassert_bulk(&glue->resets);
>>  	if (ret) {
>>  		reset_release_bulk(&glue->resets);
>> @@ -623,7 +720,7 @@ static const struct udevice_id dwc3_glue_ids[] = {
>>  	{ .compatible = "rockchip,rk3399-dwc3" },
>>  	{ .compatible = "rockchip,rk3568-dwc3", .data = (ulong)&rk_ops },
>>  	{ .compatible = "rockchip,rk3588-dwc3", .data = (ulong)&rk_ops },
>> -	{ .compatible = "qcom,dwc3" },
>> +	{ .compatible = "qcom,dwc3", .data = (ulong)&qcom_ops },
>>  	{ .compatible = "fsl,imx8mp-dwc3", .data = (ulong)&imx8mp_ops },
>>  	{ .compatible = "fsl,imx8mq-dwc3" },
>>  	{ .compatible = "intel,tangier-dwc3" },
>>
>> -- 
>> 2.43.0

-- 
// Caleb (they/them)

Re: [PATCH 1/5] usb: dwc3-generic: implement Qualcomm wrapper

[Mattijs Korpershoek]

Hi Caleb,

On Thu, Feb 01, 2024 at 14:24, Caleb Connolly <caleb.connolly@linaro.org> wrote:

> On 01/02/2024 09:34, Mattijs Korpershoek wrote:
>> Hi Caleb,
>> 
>> Thank you for the patch.
>> 
>> On mer., janv. 31, 2024 at 14:57, Caleb Connolly <caleb.connolly@linaro.org> wrote:
>> 
>>> The Qualcomm specific dwc3 wrapper isn't hugely complicated, implemented
>>> the missing initialisation for host and gadget mode.
>>>
>>> Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>
>>> ---
>>>  drivers/usb/dwc3/dwc3-generic.c | 99 ++++++++++++++++++++++++++++++++++++++++-
>>>  1 file changed, 98 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/usb/dwc3/dwc3-generic.c b/drivers/usb/dwc3/dwc3-generic.c
>>> index 48da621ba966..1119cdecd26d 100644
>>> --- a/drivers/usb/dwc3/dwc3-generic.c
>>> +++ b/drivers/usb/dwc3/dwc3-generic.c
>>> @@ -419,6 +419,99 @@ struct dwc3_glue_ops ti_ops = {
>>>  	.glue_configure = dwc3_ti_glue_configure,
>>>  };
>>>  
>>> +/* USB QSCRATCH Hardware registers */
>>> +#define QSCRATCH_HS_PHY_CTRL 0x10
>>> +#define UTMI_OTG_VBUS_VALID BIT(20)
>>> +#define SW_SESSVLD_SEL BIT(28)
>>> +
>>> +#define QSCRATCH_SS_PHY_CTRL 0x30
>>> +#define LANE0_PWR_PRESENT BIT(24)
>>> +
>>> +#define QSCRATCH_GENERAL_CFG 0x08
>>> +#define PIPE_UTMI_CLK_SEL BIT(0)
>>> +#define PIPE3_PHYSTATUS_SW BIT(3)
>>> +#define PIPE_UTMI_CLK_DIS BIT(8)
>>> +
>>> +#define PWR_EVNT_IRQ_STAT_REG 0x58
>>> +#define PWR_EVNT_LPM_IN_L2_MASK BIT(4)
>>> +#define PWR_EVNT_LPM_OUT_L2_MASK BIT(5)
>>> +
>>> +#define SDM845_QSCRATCH_BASE_OFFSET 0xf8800
>>> +#define SDM845_QSCRATCH_SIZE 0x400
>>> +#define SDM845_DWC3_CORE_SIZE 0xcd00
>>> +static inline void dwc3_qcom_setbits(void __iomem *base, u32 offset, u32 val)
>>> +{
>>> +	u32 reg;
>>> +
>>> +	reg = readl(base + offset);
>>> +	reg |= val;
>>> +	writel(reg, base + offset);
>> 
>> Why can't we use the setbits() macro here?
>> see: arch/arm/include/asm/io.h
>
> setbits doesn't give us the same cache coherency guarantees I think(?)
> the readl/writel macros include wmb() calls.

Indeed, I did not pay attention to that difference.

>
> That said, I won't pretend to really know what I'm talking about here,
> possible setbits() would be suitable, do you have any idea?

I am sorry, I don't know.

It's fine to keep it as is.

>> 
>>> +
>>> +	/* ensure that above write is through */
>>> +	readl(base + offset);
>>> +}
>>> +
>>> +static inline void dwc3_qcom_clrbits(void __iomem *base, u32 offset, u32 val)
>>> +{
>>> +	u32 reg;
>>> +
>>> +	reg = readl(base + offset);
>>> +	reg &= ~val;
>>> +	writel(reg, base + offset);
>> 
>> Same question for clrbits()
>> 
>>> +
>>> +	/* ensure that above write is through */
>>> +	readl(base + offset);
>>> +}
>>> +
>
> [snip]
>
>>> +	if (device_is_compatible(dev, "qcom,dwc3")) {
>>> +		reset_assert_bulk(&glue->resets);
>> 
>> Any reason for not doing error handling on reset_assert_bulk() like it's
>> done below ?
>
> Well, the Qualcomm reset driver will never return an error, and if it
> did, the assert failing (presumably because it's already asserted) is
> not necessarily an error we'd need to care about - it's only an error if
> we can't deassert the reset.
>
> So I think this is fine.

Ok, agreed.

Reviewed-by: Mattijs Korpershoek <mkorpershoek@baylibre.com>

>> 
>>> +		udelay(500);
>>> +	}
>>>  	ret = reset_deassert_bulk(&glue->resets);
>>>  	if (ret) {
>>>  		reset_release_bulk(&glue->resets);
>>> @@ -623,7 +720,7 @@ static const struct udevice_id dwc3_glue_ids[] = {
>>>  	{ .compatible = "rockchip,rk3399-dwc3" },
>>>  	{ .compatible = "rockchip,rk3568-dwc3", .data = (ulong)&rk_ops },
>>>  	{ .compatible = "rockchip,rk3588-dwc3", .data = (ulong)&rk_ops },
>>> -	{ .compatible = "qcom,dwc3" },
>>> +	{ .compatible = "qcom,dwc3", .data = (ulong)&qcom_ops },
>>>  	{ .compatible = "fsl,imx8mp-dwc3", .data = (ulong)&imx8mp_ops },
>>>  	{ .compatible = "fsl,imx8mq-dwc3" },
>>>  	{ .compatible = "intel,tangier-dwc3" },
>>>
>>> -- 
>>> 2.43.0
>
> -- 
> // Caleb (they/them)

Re: [PATCH 1/5] usb: dwc3-generic: implement Qualcomm wrapper

[Marek Vasut]

On 1/31/24 15:57, Caleb Connolly wrote:
> The Qualcomm specific dwc3 wrapper isn't hugely complicated, implemented
> the missing initialisation for host and gadget mode.
> 
> Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>
> ---
>   drivers/usb/dwc3/dwc3-generic.c | 99 ++++++++++++++++++++++++++++++++++++++++-
>   1 file changed, 98 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/usb/dwc3/dwc3-generic.c b/drivers/usb/dwc3/dwc3-generic.c
> index 48da621ba966..1119cdecd26d 100644
> --- a/drivers/usb/dwc3/dwc3-generic.c
> +++ b/drivers/usb/dwc3/dwc3-generic.c
> @@ -419,6 +419,99 @@ struct dwc3_glue_ops ti_ops = {
>   	.glue_configure = dwc3_ti_glue_configure,
>   };
>   
> +/* USB QSCRATCH Hardware registers */
> +#define QSCRATCH_HS_PHY_CTRL 0x10
> +#define UTMI_OTG_VBUS_VALID BIT(20)
> +#define SW_SESSVLD_SEL BIT(28)
> +
> +#define QSCRATCH_SS_PHY_CTRL 0x30
> +#define LANE0_PWR_PRESENT BIT(24)
> +
> +#define QSCRATCH_GENERAL_CFG 0x08
> +#define PIPE_UTMI_CLK_SEL BIT(0)
> +#define PIPE3_PHYSTATUS_SW BIT(3)
> +#define PIPE_UTMI_CLK_DIS BIT(8)
> +
> +#define PWR_EVNT_IRQ_STAT_REG 0x58
> +#define PWR_EVNT_LPM_IN_L2_MASK BIT(4)
> +#define PWR_EVNT_LPM_OUT_L2_MASK BIT(5)
> +
> +#define SDM845_QSCRATCH_BASE_OFFSET 0xf8800
> +#define SDM845_QSCRATCH_SIZE 0x400
> +#define SDM845_DWC3_CORE_SIZE 0xcd00

Newline here please

> +static inline void dwc3_qcom_setbits(void __iomem *base, u32 offset, u32 val)
> +{
> +	u32 reg;
> +
> +	reg = readl(base + offset);
> +	reg |= val;
> +	writel(reg, base + offset);

Use setbits_le32() .

> +	/* ensure that above write is through */
> +	readl(base + offset);

Is this needed ?

> +}
> +
> +static inline void dwc3_qcom_clrbits(void __iomem *base, u32 offset, u32 val)
> +{
> +	u32 reg;
> +
> +	reg = readl(base + offset);
> +	reg &= ~val;
> +	writel(reg, base + offset);

clrbits_le32()

> +	/* ensure that above write is through */
> +	readl(base + offset);
> +}
> +
> +static void dwc3_qcom_vbus_override_enable(void __iomem *qscratch_base, bool enable)
> +{
> +	if (enable) {
> +		dwc3_qcom_setbits(qscratch_base, QSCRATCH_SS_PHY_CTRL,
> +				  LANE0_PWR_PRESENT);
> +		dwc3_qcom_setbits(qscratch_base, QSCRATCH_HS_PHY_CTRL,
> +				  UTMI_OTG_VBUS_VALID | SW_SESSVLD_SEL);
> +	} else {
> +		dwc3_qcom_clrbits(qscratch_base, QSCRATCH_SS_PHY_CTRL,
> +				  LANE0_PWR_PRESENT);
> +		dwc3_qcom_clrbits(qscratch_base, QSCRATCH_HS_PHY_CTRL,
> +				  UTMI_OTG_VBUS_VALID | SW_SESSVLD_SEL);
> +	}
> +}
> +
> +/* For controllers running without superspeed PHYs */
> +static void dwc3_qcom_select_utmi_clk(void __iomem *qscratch_base)
> +{
> +	/* Configure dwc3 to use UTMI clock as PIPE clock not present */
> +	dwc3_qcom_setbits(qscratch_base, QSCRATCH_GENERAL_CFG,
> +			  PIPE_UTMI_CLK_DIS);
> +
> +	udelay(500);

Isn't there some possibility to poll for completion instead of fixed 
delay ? If so, use wait_for_bit or some such .

> +	dwc3_qcom_setbits(qscratch_base, QSCRATCH_GENERAL_CFG,
> +			  PIPE_UTMI_CLK_SEL | PIPE3_PHYSTATUS_SW);
> +
> +	udelay(500);
> +
> +	dwc3_qcom_clrbits(qscratch_base, QSCRATCH_GENERAL_CFG,
> +			  PIPE_UTMI_CLK_DIS);
> +}
> +
> +static void dwc3_qcom_glue_configure(struct udevice *dev, int index,
> +				     enum usb_dr_mode mode)
> +{
> +	void __iomem *qscratch_base = (void __iomem *)dev_read_addr(dev);

Can this be NULL ? If yes, add check for != NULL .

> +	if (dev_read_bool(dev, "qcom,select-utmi-as-pipe-clk"))
> +		dwc3_qcom_select_utmi_clk(qscratch_base);
> +
> +	if (mode != USB_DR_MODE_HOST)
> +		dwc3_qcom_vbus_override_enable(qscratch_base, true);
> +}
> +
> +struct dwc3_glue_ops qcom_ops = {
> +	.glue_configure = dwc3_qcom_glue_configure,
> +};
> +
>   static int dwc3_rk_glue_get_ctrl_dev(struct udevice *dev, ofnode *node)
>   {
>   	*node = dev_ofnode(dev);
> @@ -506,6 +599,10 @@ static int dwc3_glue_reset_init(struct udevice *dev,
>   	else if (ret)
>   		return ret;
>   
> +	if (device_is_compatible(dev, "qcom,dwc3")) {
> +		reset_assert_bulk(&glue->resets);
> +		udelay(500);

Why this delay here ?

[...]

Re: [PATCH 1/5] usb: dwc3-generic: implement Qualcomm wrapper

[Caleb Connolly]

Hi Marek,

On 06/02/2024 20:36, Marek Vasut wrote:
> On 1/31/24 15:57, Caleb Connolly wrote:
>> The Qualcomm specific dwc3 wrapper isn't hugely complicated, implemented
>> the missing initialisation for host and gadget mode.
>>
>> Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>
>> ---
>>   drivers/usb/dwc3/dwc3-generic.c | 99
>> ++++++++++++++++++++++++++++++++++++++++-
>>   1 file changed, 98 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/usb/dwc3/dwc3-generic.c
>> b/drivers/usb/dwc3/dwc3-generic.c
>> index 48da621ba966..1119cdecd26d 100644
>> --- a/drivers/usb/dwc3/dwc3-generic.c
>> +++ b/drivers/usb/dwc3/dwc3-generic.c
>> @@ -419,6 +419,99 @@ struct dwc3_glue_ops ti_ops = {
>>       .glue_configure = dwc3_ti_glue_configure,
>>   };
>>   +/* USB QSCRATCH Hardware registers */
>> +#define QSCRATCH_HS_PHY_CTRL 0x10
>> +#define UTMI_OTG_VBUS_VALID BIT(20)
>> +#define SW_SESSVLD_SEL BIT(28)
>> +
>> +#define QSCRATCH_SS_PHY_CTRL 0x30
>> +#define LANE0_PWR_PRESENT BIT(24)
>> +
>> +#define QSCRATCH_GENERAL_CFG 0x08
>> +#define PIPE_UTMI_CLK_SEL BIT(0)
>> +#define PIPE3_PHYSTATUS_SW BIT(3)
>> +#define PIPE_UTMI_CLK_DIS BIT(8)
>> +
>> +#define PWR_EVNT_IRQ_STAT_REG 0x58
>> +#define PWR_EVNT_LPM_IN_L2_MASK BIT(4)
>> +#define PWR_EVNT_LPM_OUT_L2_MASK BIT(5)
>> +
>> +#define SDM845_QSCRATCH_BASE_OFFSET 0xf8800
>> +#define SDM845_QSCRATCH_SIZE 0x400
>> +#define SDM845_DWC3_CORE_SIZE 0xcd00
> 
> Newline here please
> 
>> +static inline void dwc3_qcom_setbits(void __iomem *base, u32 offset,
>> u32 val)
>> +{
>> +    u32 reg;
>> +
>> +    reg = readl(base + offset);
>> +    reg |= val;
>> +    writel(reg, base + offset);
> 
> Use setbits_le32() .
> 
>> +    /* ensure that above write is through */
>> +    readl(base + offset);
> 
> Is this needed ?

I honestly don't know, this is copied from the Linux driver and it seems
to be very defensively written. I doubt it's strictly necessary.
> 
>> +}
>> +
>> +static inline void dwc3_qcom_clrbits(void __iomem *base, u32 offset,
>> u32 val)
>> +{
>> +    u32 reg;
>> +
>> +    reg = readl(base + offset);
>> +    reg &= ~val;
>> +    writel(reg, base + offset);
> 
> clrbits_le32()
> 
>> +    /* ensure that above write is through */
>> +    readl(base + offset);
>> +}
>> +
>> +static void dwc3_qcom_vbus_override_enable(void __iomem
>> *qscratch_base, bool enable)
>> +{
>> +    if (enable) {
>> +        dwc3_qcom_setbits(qscratch_base, QSCRATCH_SS_PHY_CTRL,
>> +                  LANE0_PWR_PRESENT);
>> +        dwc3_qcom_setbits(qscratch_base, QSCRATCH_HS_PHY_CTRL,
>> +                  UTMI_OTG_VBUS_VALID | SW_SESSVLD_SEL);
>> +    } else {
>> +        dwc3_qcom_clrbits(qscratch_base, QSCRATCH_SS_PHY_CTRL,
>> +                  LANE0_PWR_PRESENT);
>> +        dwc3_qcom_clrbits(qscratch_base, QSCRATCH_HS_PHY_CTRL,
>> +                  UTMI_OTG_VBUS_VALID | SW_SESSVLD_SEL);
>> +    }
>> +}
>> +
>> +/* For controllers running without superspeed PHYs */
>> +static void dwc3_qcom_select_utmi_clk(void __iomem *qscratch_base)
>> +{
>> +    /* Configure dwc3 to use UTMI clock as PIPE clock not present */
>> +    dwc3_qcom_setbits(qscratch_base, QSCRATCH_GENERAL_CFG,
>> +              PIPE_UTMI_CLK_DIS);
>> +
>> +    udelay(500);
> 
> Isn't there some possibility to poll for completion instead of fixed
> delay ? If so, use wait_for_bit or some such .

Not that I'm aware of, no. I think this hardware just has a blanket
"writes take X bus cycles to complete" rule or something. It's totally
possible that this code was originally written this way to work around
some issues on an FPGA prototype or something. Everything seems to still
work if I remove the delays so I'll drop them...
> 
>> +    dwc3_qcom_setbits(qscratch_base, QSCRATCH_GENERAL_CFG,
>> +              PIPE_UTMI_CLK_SEL | PIPE3_PHYSTATUS_SW);
>> +
>> +    udelay(500);
>> +
>> +    dwc3_qcom_clrbits(qscratch_base, QSCRATCH_GENERAL_CFG,
>> +              PIPE_UTMI_CLK_DIS);
>> +}
>> +
>> +static void dwc3_qcom_glue_configure(struct udevice *dev, int index,
>> +                     enum usb_dr_mode mode)
>> +{
>> +    void __iomem *qscratch_base = (void __iomem *)dev_read_addr(dev);
> 
> Can this be NULL ? If yes, add check for != NULL .
Will do.
> 
>> +    if (dev_read_bool(dev, "qcom,select-utmi-as-pipe-clk"))
>> +        dwc3_qcom_select_utmi_clk(qscratch_base);
>> +
>> +    if (mode != USB_DR_MODE_HOST)
>> +        dwc3_qcom_vbus_override_enable(qscratch_base, true);
>> +}
>> +
>> +struct dwc3_glue_ops qcom_ops = {
>> +    .glue_configure = dwc3_qcom_glue_configure,
>> +};
>> +
>>   static int dwc3_rk_glue_get_ctrl_dev(struct udevice *dev, ofnode *node)
>>   {
>>       *node = dev_ofnode(dev);
>> @@ -506,6 +599,10 @@ static int dwc3_glue_reset_init(struct udevice *dev,
>>       else if (ret)
>>           return ret;
>>   +    if (device_is_compatible(dev, "qcom,dwc3")) {
>> +        reset_assert_bulk(&glue->resets);
>> +        udelay(500);
> 
> Why this delay here ?

According to the docs, the reset should be asserted for at least 6 sleep
clock cycles, that's ~200us on sdm845, but it can vary by platform.

> 
> [...]

-- 
// Caleb (they/them)

Re: [PATCH 1/5] usb: dwc3-generic: implement Qualcomm wrapper

[Marek Vasut]

On 3/13/24 7:22 PM, Caleb Connolly wrote:

[...]

>>> +static inline void dwc3_qcom_setbits(void __iomem *base, u32 offset,
>>> u32 val)
>>> +{
>>> +    u32 reg;
>>> +
>>> +    reg = readl(base + offset);
>>> +    reg |= val;
>>> +    writel(reg, base + offset);
>>
>> Use setbits_le32() .
>>
>>> +    /* ensure that above write is through */
>>> +    readl(base + offset);
>>
>> Is this needed ?
> 
> I honestly don't know, this is copied from the Linux driver and it seems
> to be very defensively written. I doubt it's strictly necessary.

Does git log indicate anything ?

I suspect this is some sort of barrier .

[...]

>>> +/* For controllers running without superspeed PHYs */
>>> +static void dwc3_qcom_select_utmi_clk(void __iomem *qscratch_base)
>>> +{
>>> +    /* Configure dwc3 to use UTMI clock as PIPE clock not present */
>>> +    dwc3_qcom_setbits(qscratch_base, QSCRATCH_GENERAL_CFG,
>>> +              PIPE_UTMI_CLK_DIS);
>>> +
>>> +    udelay(500);
>>
>> Isn't there some possibility to poll for completion instead of fixed
>> delay ? If so, use wait_for_bit or some such .
> 
> Not that I'm aware of, no. I think this hardware just has a blanket
> "writes take X bus cycles to complete" rule or something. It's totally
> possible that this code was originally written this way to work around
> some issues on an FPGA prototype or something. Everything seems to still
> work if I remove the delays so I'll drop them...

Could you possibly ask someone ?

[...]

>>>    static int dwc3_rk_glue_get_ctrl_dev(struct udevice *dev, ofnode *node)
>>>    {
>>>        *node = dev_ofnode(dev);
>>> @@ -506,6 +599,10 @@ static int dwc3_glue_reset_init(struct udevice *dev,
>>>        else if (ret)
>>>            return ret;
>>>    +    if (device_is_compatible(dev, "qcom,dwc3")) {
>>> +        reset_assert_bulk(&glue->resets);
>>> +        udelay(500);
>>
>> Why this delay here ?
> 
> According to the docs, the reset should be asserted for at least 6 sleep
> clock cycles, that's ~200us on sdm845, but it can vary by platform.

A comment in the code would be nice.

Sorry for the abysmal delay in replies.

btw. the new version of this series is still OK to go in, unless you 
want to fill in the comments. They can also go in in separate follow up 
patch.

Re: [PATCH 1/5] usb: dwc3-generic: implement Qualcomm wrapper

[Mattijs Korpershoek]

Hi Caleb, Marek,

On jeu., mars 21, 2024 at 06:34, Marek Vasut <marex@denx.de> wrote:

> On 3/13/24 7:22 PM, Caleb Connolly wrote:
>
> [...]
>
>>>> +static inline void dwc3_qcom_setbits(void __iomem *base, u32 offset,
>>>> u32 val)
>>>> +{
>>>> +    u32 reg;
>>>> +
>>>> +    reg = readl(base + offset);
>>>> +    reg |= val;
>>>> +    writel(reg, base + offset);
>>>
>>> Use setbits_le32() .
>>>
>>>> +    /* ensure that above write is through */
>>>> +    readl(base + offset);
>>>
>>> Is this needed ?
>> 
>> I honestly don't know, this is copied from the Linux driver and it seems
>> to be very defensively written. I doubt it's strictly necessary.
>
> Does git log indicate anything ?
>
> I suspect this is some sort of barrier .
>
> [...]
>
>>>> +/* For controllers running without superspeed PHYs */
>>>> +static void dwc3_qcom_select_utmi_clk(void __iomem *qscratch_base)
>>>> +{
>>>> +    /* Configure dwc3 to use UTMI clock as PIPE clock not present */
>>>> +    dwc3_qcom_setbits(qscratch_base, QSCRATCH_GENERAL_CFG,
>>>> +              PIPE_UTMI_CLK_DIS);
>>>> +
>>>> +    udelay(500);
>>>
>>> Isn't there some possibility to poll for completion instead of fixed
>>> delay ? If so, use wait_for_bit or some such .
>> 
>> Not that I'm aware of, no. I think this hardware just has a blanket
>> "writes take X bus cycles to complete" rule or something. It's totally
>> possible that this code was originally written this way to work around
>> some issues on an FPGA prototype or something. Everything seems to still
>> work if I remove the delays so I'll drop them...
>
> Could you possibly ask someone ?
>
> [...]
>
>>>>    static int dwc3_rk_glue_get_ctrl_dev(struct udevice *dev, ofnode *node)
>>>>    {
>>>>        *node = dev_ofnode(dev);
>>>> @@ -506,6 +599,10 @@ static int dwc3_glue_reset_init(struct udevice *dev,
>>>>        else if (ret)
>>>>            return ret;
>>>>    +    if (device_is_compatible(dev, "qcom,dwc3")) {
>>>> +        reset_assert_bulk(&glue->resets);
>>>> +        udelay(500);
>>>
>>> Why this delay here ?
>> 
>> According to the docs, the reset should be asserted for at least 6 sleep
>> clock cycles, that's ~200us on sdm845, but it can vary by platform.
>
> A comment in the code would be nice.
>
> Sorry for the abysmal delay in replies.
>
> btw. the new version of this series is still OK to go in, unless you 
> want to fill in the comments. They can also go in in separate follow up 
> patch.

I'm interested by the answers above as well. As I took in the series [1] (to
avoid delaying it too much), please consider a follow up patch to add a
comment.

[1] https://lore.kernel.org/r/all/171101299073.1017001.16411913317437946645.b4-ty@baylibre.com/

Re: [PATCH 1/5] usb: dwc3-generic: implement Qualcomm wrapper

[Caleb Connolly]

Hi,

On 21/03/2024 09:25, Mattijs Korpershoek wrote:
> Hi Caleb, Marek,
> 
> On jeu., mars 21, 2024 at 06:34, Marek Vasut <marex@denx.de> wrote:
> 
>> On 3/13/24 7:22 PM, Caleb Connolly wrote:
>>
>> [...]
>>
>>>>> +static inline void dwc3_qcom_setbits(void __iomem *base, u32 offset,
>>>>> u32 val)
>>>>> +{
>>>>> +    u32 reg;
>>>>> +
>>>>> +    reg = readl(base + offset);
>>>>> +    reg |= val;
>>>>> +    writel(reg, base + offset);
>>>>
>>>> Use setbits_le32() .
>>>>
>>>>> +    /* ensure that above write is through */
>>>>> +    readl(base + offset);
>>>>
>>>> Is this needed ?
>>>
>>> I honestly don't know, this is copied from the Linux driver and it seems
>>> to be very defensively written. I doubt it's strictly necessary.
>>
>> Does git log indicate anything ?

Nope :/ it's there from when the driver was first added.
>>
>> I suspect this is some sort of barrier .
>>
>> [...]
>>
>>>>> +/* For controllers running without superspeed PHYs */
>>>>> +static void dwc3_qcom_select_utmi_clk(void __iomem *qscratch_base)
>>>>> +{
>>>>> +    /* Configure dwc3 to use UTMI clock as PIPE clock not present */
>>>>> +    dwc3_qcom_setbits(qscratch_base, QSCRATCH_GENERAL_CFG,
>>>>> +              PIPE_UTMI_CLK_DIS);
>>>>> +
>>>>> +    udelay(500);
>>>>
>>>> Isn't there some possibility to poll for completion instead of fixed
>>>> delay ? If so, use wait_for_bit or some such .
>>>
>>> Not that I'm aware of, no. I think this hardware just has a blanket
>>> "writes take X bus cycles to complete" rule or something. It's totally
>>> possible that this code was originally written this way to work around
>>> some issues on an FPGA prototype or something. Everything seems to still
>>> work if I remove the delays so I'll drop them...
>>
>> Could you possibly ask someone ?

Yeah I'll ask around, I'm not confident I'll find an answer though.
>>
>> [...]
>>
>>>>>    static int dwc3_rk_glue_get_ctrl_dev(struct udevice *dev, ofnode *node)
>>>>>    {
>>>>>        *node = dev_ofnode(dev);
>>>>> @@ -506,6 +599,10 @@ static int dwc3_glue_reset_init(struct udevice *dev,
>>>>>        else if (ret)
>>>>>            return ret;
>>>>>    +    if (device_is_compatible(dev, "qcom,dwc3")) {
>>>>> +        reset_assert_bulk(&glue->resets);
>>>>> +        udelay(500);
>>>>
>>>> Why this delay here ?
>>>
>>> According to the docs, the reset should be asserted for at least 6 sleep
>>> clock cycles, that's ~200us on sdm845, but it can vary by platform.
>>
>> A comment in the code would be nice.
>>
>> Sorry for the abysmal delay in replies.
>>
>> btw. the new version of this series is still OK to go in, unless you 
>> want to fill in the comments. They can also go in in separate follow up 
>> patch.
> 
> I'm interested by the answers above as well. As I took in the series [1] (to
> avoid delaying it too much), please consider a follow up patch to add a
> comment.

The v4 you picked up has a comment explaining this.
> 
> [1] https://lore.kernel.org/r/all/171101299073.1017001.16411913317437946645.b4-ty@baylibre.com/
> 

-- 
// Caleb (they/them)

Re: [PATCH 1/5] usb: dwc3-generic: implement Qualcomm wrapper

[Mattijs Korpershoek]

On jeu., mars 21, 2024 at 11:34, Caleb Connolly <caleb.connolly@linaro.org> wrote:

> Hi,
>
> On 21/03/2024 09:25, Mattijs Korpershoek wrote:
>> Hi Caleb, Marek,
>> 
>> On jeu., mars 21, 2024 at 06:34, Marek Vasut <marex@denx.de> wrote:
>> 
>>> On 3/13/24 7:22 PM, Caleb Connolly wrote:
>>>
>>> [...]
>>>
>>>>>> +static inline void dwc3_qcom_setbits(void __iomem *base, u32 offset,
>>>>>> u32 val)
>>>>>> +{
>>>>>> +    u32 reg;
>>>>>> +
>>>>>> +    reg = readl(base + offset);
>>>>>> +    reg |= val;
>>>>>> +    writel(reg, base + offset);
>>>>>
>>>>> Use setbits_le32() .
>>>>>
>>>>>> +    /* ensure that above write is through */
>>>>>> +    readl(base + offset);
>>>>>
>>>>> Is this needed ?
>>>>
>>>> I honestly don't know, this is copied from the Linux driver and it seems
>>>> to be very defensively written. I doubt it's strictly necessary.
>>>
>>> Does git log indicate anything ?
>
> Nope :/ it's there from when the driver was first added.
>>>
>>> I suspect this is some sort of barrier .
>>>
>>> [...]
>>>
>>>>>> +/* For controllers running without superspeed PHYs */
>>>>>> +static void dwc3_qcom_select_utmi_clk(void __iomem *qscratch_base)
>>>>>> +{
>>>>>> +    /* Configure dwc3 to use UTMI clock as PIPE clock not present */
>>>>>> +    dwc3_qcom_setbits(qscratch_base, QSCRATCH_GENERAL_CFG,
>>>>>> +              PIPE_UTMI_CLK_DIS);
>>>>>> +
>>>>>> +    udelay(500);
>>>>>
>>>>> Isn't there some possibility to poll for completion instead of fixed
>>>>> delay ? If so, use wait_for_bit or some such .
>>>>
>>>> Not that I'm aware of, no. I think this hardware just has a blanket
>>>> "writes take X bus cycles to complete" rule or something. It's totally
>>>> possible that this code was originally written this way to work around
>>>> some issues on an FPGA prototype or something. Everything seems to still
>>>> work if I remove the delays so I'll drop them...
>>>
>>> Could you possibly ask someone ?
>
> Yeah I'll ask around, I'm not confident I'll find an answer though.
>>>
>>> [...]
>>>
>>>>>>    static int dwc3_rk_glue_get_ctrl_dev(struct udevice *dev, ofnode *node)
>>>>>>    {
>>>>>>        *node = dev_ofnode(dev);
>>>>>> @@ -506,6 +599,10 @@ static int dwc3_glue_reset_init(struct udevice *dev,
>>>>>>        else if (ret)
>>>>>>            return ret;
>>>>>>    +    if (device_is_compatible(dev, "qcom,dwc3")) {
>>>>>> +        reset_assert_bulk(&glue->resets);
>>>>>> +        udelay(500);
>>>>>
>>>>> Why this delay here ?
>>>>
>>>> According to the docs, the reset should be asserted for at least 6 sleep
>>>> clock cycles, that's ~200us on sdm845, but it can vary by platform.
>>>
>>> A comment in the code would be nice.
>>>
>>> Sorry for the abysmal delay in replies.
>>>
>>> btw. the new version of this series is still OK to go in, unless you 
>>> want to fill in the comments. They can also go in in separate follow up 
>>> patch.
>> 
>> I'm interested by the answers above as well. As I took in the series [1] (to
>> avoid delaying it too much), please consider a follow up patch to add a
>> comment.
>
> The v4 you picked up has a comment explaining this.

Right, sorry I missed that. Thanks for pointing it out to me!

>> 
>> [1] https://lore.kernel.org/r/all/171101299073.1017001.16411913317437946645.b4-ty@baylibre.com/
>> 
>
> -- 
> // Caleb (they/them)

[PATCH 2/5] usb: dwc3: select DM_USB_GADGET

[Caleb Connolly]

DWC3 platforms depend on DM_USB_GADGET for gadget drivers to work,
otherwise compilation fails due to no implementation of
dm_usb_gadget_handle_interrupts().

Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>
---
 drivers/usb/dwc3/Kconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/usb/dwc3/Kconfig b/drivers/usb/dwc3/Kconfig
index c0c8c16fd9c2..8a70bc682322 100644
--- a/drivers/usb/dwc3/Kconfig
+++ b/drivers/usb/dwc3/Kconfig
@@ -11,6 +11,7 @@ config USB_DWC3_GADGET
 	bool "USB Gadget support for DWC3"
 	default y
 	depends on USB_GADGET
+	select DM_USB_GADGET
 	select USB_GADGET_DUALSPEED
 
 comment "Platform Glue Driver Support"

-- 
2.43.0

Re: [PATCH 2/5] usb: dwc3: select DM_USB_GADGET

[Mattijs Korpershoek]

Hi Caleb,

Thank you for the patch.

On mer., janv. 31, 2024 at 14:57, Caleb Connolly <caleb.connolly@linaro.org> wrote:

> DWC3 platforms depend on DM_USB_GADGET for gadget drivers to work,
> otherwise compilation fails due to no implementation of
> dm_usb_gadget_handle_interrupts().
>
> Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>

Reviewed-by: Mattijs Korpershoek <mkorpershoek@baylibre.com>

> ---
>  drivers/usb/dwc3/Kconfig | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/drivers/usb/dwc3/Kconfig b/drivers/usb/dwc3/Kconfig
> index c0c8c16fd9c2..8a70bc682322 100644
> --- a/drivers/usb/dwc3/Kconfig
> +++ b/drivers/usb/dwc3/Kconfig
> @@ -11,6 +11,7 @@ config USB_DWC3_GADGET
>  	bool "USB Gadget support for DWC3"
>  	default y
>  	depends on USB_GADGET
> +	select DM_USB_GADGET
>  	select USB_GADGET_DUALSPEED
>  
>  comment "Platform Glue Driver Support"
>
> -- 
> 2.43.0

[PATCH 3/5] usb: gadget: CDC ACM: call usb_gadget_initialize

[Caleb Connolly]

To actually use the gadget the peripheral driver must be probed and we
must call g_dnl_clear_detach(). Otherwise acm_stdio_start() will always
fail to find a UDC on DT platforms.

Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>
---
 drivers/usb/gadget/f_acm.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/drivers/usb/gadget/f_acm.c b/drivers/usb/gadget/f_acm.c
index de42e0189e8d..ba216128ab27 100644
--- a/drivers/usb/gadget/f_acm.c
+++ b/drivers/usb/gadget/f_acm.c
@@ -623,12 +623,21 @@ static void acm_stdio_puts(struct stdio_dev *dev, const char *str)
 
 static int acm_stdio_start(struct stdio_dev *dev)
 {
+	struct udevice *udc;
 	int ret;
 
 	if (dev->priv) { /* function already exist */
 		return 0;
 	}
 
+	ret = udc_device_get_by_index(0, &udc);
+	if (ret) {
+		pr_err("USB init failed: %d\n", ret);
+		return ret;
+	}
+
+	g_dnl_clear_detach();
+
 	ret = g_dnl_register("usb_serial_acm");
 	if (ret)
 		return ret;

-- 
2.43.0

Re: [PATCH 3/5] usb: gadget: CDC ACM: call usb_gadget_initialize

[Mattijs Korpershoek]

Hi Caleb,

Thank you for the patch.

On mer., janv. 31, 2024 at 14:57, Caleb Connolly <caleb.connolly@linaro.org> wrote:

> To actually use the gadget the peripheral driver must be probed and we
> must call g_dnl_clear_detach(). Otherwise acm_stdio_start() will always
> fail to find a UDC on DT platforms.
>
> Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>

Reviewed-by: Mattijs Korpershoek <mkorpershoek@baylibre.com>

> ---
>  drivers/usb/gadget/f_acm.c | 9 +++++++++
>  1 file changed, 9 insertions(+)
>
> diff --git a/drivers/usb/gadget/f_acm.c b/drivers/usb/gadget/f_acm.c
> index de42e0189e8d..ba216128ab27 100644
> --- a/drivers/usb/gadget/f_acm.c
> +++ b/drivers/usb/gadget/f_acm.c
> @@ -623,12 +623,21 @@ static void acm_stdio_puts(struct stdio_dev *dev, const char *str)
>  
>  static int acm_stdio_start(struct stdio_dev *dev)
>  {
> +	struct udevice *udc;
>  	int ret;
>  
>  	if (dev->priv) { /* function already exist */
>  		return 0;
>  	}
>  
> +	ret = udc_device_get_by_index(0, &udc);
> +	if (ret) {
> +		pr_err("USB init failed: %d\n", ret);
> +		return ret;
> +	}
> +
> +	g_dnl_clear_detach();
> +
>  	ret = g_dnl_register("usb_serial_acm");
>  	if (ret)
>  		return ret;
>
> -- 
> 2.43.0

[PATCH 4/5] usb: gadget: UMS: support multiple sector sizes

[Caleb Connolly]

UFS storage often uses a 4096-byte sector size, add support for dynamic
sector sizes based loosely on the Linux implementation.

Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>
---
 cmd/usb_mass_storage.c              |   4 --
 drivers/usb/gadget/f_mass_storage.c | 101 ++++++++++++++++++++----------------
 drivers/usb/gadget/storage_common.c |  12 +++--
 include/usb_mass_storage.h          |   1 -
 4 files changed, 65 insertions(+), 53 deletions(-)

diff --git a/cmd/usb_mass_storage.c b/cmd/usb_mass_storage.c
index a8ddeb494628..751701fe73af 100644
--- a/cmd/usb_mass_storage.c
+++ b/cmd/usb_mass_storage.c
@@ -88,10 +88,6 @@ static int ums_init(const char *devtype, const char *devnums_part_str)
 		if (!strchr(devnum_part_str, ':'))
 			partnum = 0;
 
-		/* f_mass_storage.c assumes SECTOR_SIZE sectors */
-		if (block_dev->blksz != SECTOR_SIZE)
-			goto cleanup;
-
 		ums_new = realloc(ums, (ums_count + 1) * sizeof(*ums));
 		if (!ums_new)
 			goto cleanup;
diff --git a/drivers/usb/gadget/f_mass_storage.c b/drivers/usb/gadget/f_mass_storage.c
index c725aed3f626..d880928044f4 100644
--- a/drivers/usb/gadget/f_mass_storage.c
+++ b/drivers/usb/gadget/f_mass_storage.c
@@ -724,12 +724,13 @@ static int do_read(struct fsg_common *common)
 		curlun->sense_data = SS_LOGICAL_BLOCK_ADDRESS_OUT_OF_RANGE;
 		return -EINVAL;
 	}
-	file_offset = ((loff_t) lba) << 9;
+	file_offset = ((loff_t)lba) << curlun->blkbits;
 
 	/* Carry out the file reads */
 	amount_left = common->data_size_from_cmnd;
-	if (unlikely(amount_left == 0))
+	if (unlikely(amount_left == 0)) {
 		return -EIO;		/* No default reply */
+	}
 
 	for (;;) {
 
@@ -768,13 +769,13 @@ static int do_read(struct fsg_common *common)
 
 		/* Perform the read */
 		rc = ums[common->lun].read_sector(&ums[common->lun],
-				      file_offset / SECTOR_SIZE,
-				      amount / SECTOR_SIZE,
+				      file_offset / curlun->blksize,
+				      amount / curlun->blksize,
 				      (char __user *)bh->buf);
 		if (!rc)
 			return -EIO;
 
-		nread = rc * SECTOR_SIZE;
+		nread = rc * curlun->blksize;
 
 		VLDBG(curlun, "file read %u @ %llu -> %d\n", amount,
 				(unsigned long long) file_offset,
@@ -787,7 +788,7 @@ static int do_read(struct fsg_common *common)
 		} else if (nread < amount) {
 			LDBG(curlun, "partial file read: %d/%u\n",
 					(int) nread, amount);
-			nread -= (nread & 511);	/* Round down to a block */
+			nread -= (nread & (curlun->blksize - 1));	/* Round down to a block */
 		}
 		file_offset  += nread;
 		amount_left  -= nread;
@@ -861,7 +862,7 @@ static int do_write(struct fsg_common *common)
 
 	/* Carry out the file writes */
 	get_some_more = 1;
-	file_offset = usb_offset = ((loff_t) lba) << 9;
+	file_offset = usb_offset = ((loff_t)lba) << curlun->blkbits;
 	amount_left_to_req = common->data_size_from_cmnd;
 	amount_left_to_write = common->data_size_from_cmnd;
 
@@ -893,7 +894,7 @@ static int do_write(struct fsg_common *common)
 				curlun->info_valid = 1;
 				continue;
 			}
-			amount -= (amount & 511);
+			amount -= (amount & (curlun->blksize - 1));
 			if (amount == 0) {
 
 				/* Why were we were asked to transfer a
@@ -942,12 +943,12 @@ static int do_write(struct fsg_common *common)
 
 			/* Perform the write */
 			rc = ums[common->lun].write_sector(&ums[common->lun],
-					       file_offset / SECTOR_SIZE,
-					       amount / SECTOR_SIZE,
+					       file_offset / curlun->blksize,
+					       amount / curlun->blksize,
 					       (char __user *)bh->buf);
 			if (!rc)
 				return -EIO;
-			nwritten = rc * SECTOR_SIZE;
+			nwritten = rc * curlun->blksize;
 
 			VLDBG(curlun, "file write %u @ %llu -> %d\n", amount,
 					(unsigned long long) file_offset,
@@ -960,7 +961,7 @@ static int do_write(struct fsg_common *common)
 			} else if (nwritten < amount) {
 				LDBG(curlun, "partial file write: %d/%u\n",
 						(int) nwritten, amount);
-				nwritten -= (nwritten & 511);
+				nwritten -= (nwritten & (curlun->blksize - 1));
 				/* Round down to a block */
 			}
 			file_offset += nwritten;
@@ -1034,8 +1035,8 @@ static int do_verify(struct fsg_common *common)
 		return -EIO;		/* No default reply */
 
 	/* Prepare to carry out the file verify */
-	amount_left = verification_length << 9;
-	file_offset = ((loff_t) lba) << 9;
+	amount_left = verification_length << curlun->blkbits;
+	file_offset = ((loff_t) lba) << curlun->blkbits;
 
 	/* Write out all the dirty buffers before invalidating them */
 
@@ -1058,12 +1059,12 @@ static int do_verify(struct fsg_common *common)
 
 		/* Perform the read */
 		rc = ums[common->lun].read_sector(&ums[common->lun],
-				      file_offset / SECTOR_SIZE,
-				      amount / SECTOR_SIZE,
+				      file_offset / curlun->blksize,
+				      amount / curlun->blksize,
 				      (char __user *)bh->buf);
 		if (!rc)
 			return -EIO;
-		nread = rc * SECTOR_SIZE;
+		nread = rc * curlun->blksize;
 
 		VLDBG(curlun, "file read %u @ %llu -> %d\n", amount,
 				(unsigned long long) file_offset,
@@ -1075,7 +1076,7 @@ static int do_verify(struct fsg_common *common)
 		} else if (nread < amount) {
 			LDBG(curlun, "partial file verify: %d/%u\n",
 					(int) nread, amount);
-			nread -= (nread & 511);	/* Round down to a sector */
+			nread -= (nread & (curlun->blksize - 1));	/* Round down to a sector */
 		}
 		if (nread == 0) {
 			curlun->sense_data = SS_UNRECOVERED_READ_ERROR;
@@ -1183,7 +1184,7 @@ static int do_read_capacity(struct fsg_common *common, struct fsg_buffhd *bh)
 
 	put_unaligned_be32(curlun->num_sectors - 1, &buf[0]);
 						/* Max logical block */
-	put_unaligned_be32(512, &buf[4]);	/* Block length */
+	put_unaligned_be32(curlun->blksize, &buf[4]);	/* Block length */
 	return 8;
 }
 
@@ -1370,7 +1371,7 @@ static int do_read_format_capacities(struct fsg_common *common,
 
 	put_unaligned_be32(curlun->num_sectors, &buf[0]);
 						/* Number of blocks */
-	put_unaligned_be32(512, &buf[4]);	/* Block length */
+	put_unaligned_be32(curlun->blksize, &buf[4]);	/* Block length */
 	buf[4] = 0x02;				/* Current capacity */
 	return 12;
 }
@@ -1781,6 +1782,16 @@ static int check_command(struct fsg_common *common, int cmnd_size,
 	return 0;
 }
 
+/* wrapper of check_command for data size in blocks handling */
+static int check_command_size_in_blocks(struct fsg_common *common,
+		int cmnd_size, enum data_direction data_dir,
+		unsigned int mask, int needs_medium, const char *name)
+{
+	common->data_size_from_cmnd <<= common->luns[common->lun].blkbits;
+	return check_command(common, cmnd_size, data_dir,
+			mask, needs_medium, name);
+}
+
 
 static int do_scsi_command(struct fsg_common *common)
 {
@@ -1865,30 +1876,30 @@ static int do_scsi_command(struct fsg_common *common)
 
 	case SC_READ_6:
 		i = common->cmnd[4];
-		common->data_size_from_cmnd = (i == 0 ? 256 : i) << 9;
-		reply = check_command(common, 6, DATA_DIR_TO_HOST,
-				      (7<<1) | (1<<4), 1,
-				      "READ(6)");
+		common->data_size_from_cmnd = (i == 0 ? 256 : i);
+		reply = check_command_size_in_blocks(common, 6, DATA_DIR_TO_HOST,
+						     (7<<1) | (1<<4), 1,
+						     "READ(6)");
 		if (reply == 0)
 			reply = do_read(common);
 		break;
 
 	case SC_READ_10:
 		common->data_size_from_cmnd =
-				get_unaligned_be16(&common->cmnd[7]) << 9;
-		reply = check_command(common, 10, DATA_DIR_TO_HOST,
-				      (1<<1) | (0xf<<2) | (3<<7), 1,
-				      "READ(10)");
+				get_unaligned_be16(&common->cmnd[7]);
+		reply = check_command_size_in_blocks(common, 10, DATA_DIR_TO_HOST,
+						     (1<<1) | (0xf<<2) | (3<<7), 1,
+						     "READ(10)");
 		if (reply == 0)
 			reply = do_read(common);
 		break;
 
 	case SC_READ_12:
 		common->data_size_from_cmnd =
-				get_unaligned_be32(&common->cmnd[6]) << 9;
-		reply = check_command(common, 12, DATA_DIR_TO_HOST,
-				      (1<<1) | (0xf<<2) | (0xf<<6), 1,
-				      "READ(12)");
+				get_unaligned_be32(&common->cmnd[6]);
+		reply = check_command_size_in_blocks(common, 12, DATA_DIR_TO_HOST,
+						     (1<<1) | (0xf<<2) | (0xf<<6), 1,
+						     "READ(12)");
 		if (reply == 0)
 			reply = do_read(common);
 		break;
@@ -1983,30 +1994,30 @@ static int do_scsi_command(struct fsg_common *common)
 
 	case SC_WRITE_6:
 		i = common->cmnd[4];
-		common->data_size_from_cmnd = (i == 0 ? 256 : i) << 9;
-		reply = check_command(common, 6, DATA_DIR_FROM_HOST,
-				      (7<<1) | (1<<4), 1,
-				      "WRITE(6)");
+		common->data_size_from_cmnd = (i == 0 ? 256 : i);
+		reply = check_command_size_in_blocks(common, 6, DATA_DIR_FROM_HOST,
+						     (7<<1) | (1<<4), 1,
+						     "WRITE(6)");
 		if (reply == 0)
 			reply = do_write(common);
 		break;
 
 	case SC_WRITE_10:
 		common->data_size_from_cmnd =
-				get_unaligned_be16(&common->cmnd[7]) << 9;
-		reply = check_command(common, 10, DATA_DIR_FROM_HOST,
-				      (1<<1) | (0xf<<2) | (3<<7), 1,
-				      "WRITE(10)");
+				get_unaligned_be16(&common->cmnd[7]);
+		reply = check_command_size_in_blocks(common, 10, DATA_DIR_FROM_HOST,
+						     (1<<1) | (0xf<<2) | (3<<7), 1,
+						     "WRITE(10)");
 		if (reply == 0)
 			reply = do_write(common);
 		break;
 
 	case SC_WRITE_12:
 		common->data_size_from_cmnd =
-				get_unaligned_be32(&common->cmnd[6]) << 9;
-		reply = check_command(common, 12, DATA_DIR_FROM_HOST,
-				      (1<<1) | (0xf<<2) | (0xf<<6), 1,
-				      "WRITE(12)");
+				get_unaligned_be32(&common->cmnd[6]);
+		reply = check_command_size_in_blocks(common, 12, DATA_DIR_FROM_HOST,
+						     (1<<1) | (0xf<<2) | (0xf<<6), 1,
+						     "WRITE(12)");
 		if (reply == 0)
 			reply = do_write(common);
 		break;
@@ -2497,7 +2508,7 @@ static struct fsg_common *fsg_common_init(struct fsg_common *common,
 	for (i = 0; i < nluns; i++) {
 		common->luns[i].removable = 1;
 
-		rc = fsg_lun_open(&common->luns[i], ums[i].num_sectors, "");
+		rc = fsg_lun_open(&common->luns[i], ums[i].num_sectors, ums->block_dev.blksz, "");
 		if (rc)
 			goto error_luns;
 	}
diff --git a/drivers/usb/gadget/storage_common.c b/drivers/usb/gadget/storage_common.c
index 5674e8fe4940..97dc6b6f729c 100644
--- a/drivers/usb/gadget/storage_common.c
+++ b/drivers/usb/gadget/storage_common.c
@@ -269,6 +269,7 @@ struct device_attribute { int i; };
 #define ETOOSMALL	525
 
 #include <log.h>
+#include <linux/log2.h>
 #include <usb_mass_storage.h>
 #include <dm/device_compat.h>
 
@@ -290,6 +291,8 @@ struct fsg_lun {
 	u32		sense_data;
 	u32		sense_data_info;
 	u32		unit_attention_data;
+	unsigned int	blkbits;
+	unsigned int	blksize; /* logical block size of bound block device */
 
 	struct device	dev;
 };
@@ -566,7 +569,7 @@ static struct usb_gadget_strings	fsg_stringtab = {
  */
 
 static int fsg_lun_open(struct fsg_lun *curlun, unsigned int num_sectors,
-			const char *filename)
+			unsigned int sector_size, const char *filename)
 {
 	int				ro;
 
@@ -574,9 +577,12 @@ static int fsg_lun_open(struct fsg_lun *curlun, unsigned int num_sectors,
 	ro = curlun->initially_ro;
 
 	curlun->ro = ro;
-	curlun->file_length = num_sectors << 9;
+	curlun->file_length = num_sectors * sector_size;
 	curlun->num_sectors = num_sectors;
-	debug("open backing file: %s\n", filename);
+	curlun->blksize = sector_size;
+	curlun->blkbits = order_base_2(sector_size >> 9) + 9;
+	debug("blksize: %u\n", sector_size);
+	debug("open backing file: '%s'\n", filename);
 
 	return 0;
 }
diff --git a/include/usb_mass_storage.h b/include/usb_mass_storage.h
index 83ab93b530d7..6d83d93cad7f 100644
--- a/include/usb_mass_storage.h
+++ b/include/usb_mass_storage.h
@@ -7,7 +7,6 @@
 #ifndef __USB_MASS_STORAGE_H__
 #define __USB_MASS_STORAGE_H__
 
-#define SECTOR_SIZE		0x200
 #include <part.h>
 #include <linux/usb/composite.h>
 

-- 
2.43.0

Re: [PATCH 4/5] usb: gadget: UMS: support multiple sector sizes

[Mattijs Korpershoek]

Hi Caleb,

Thank you for the patch.

On mer., janv. 31, 2024 at 14:57, Caleb Connolly <caleb.connolly@linaro.org> wrote:

> UFS storage often uses a 4096-byte sector size, add support for dynamic
> sector sizes based loosely on the Linux implementation.
>
> Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>

Reviewed-by: Mattijs Korpershoek <mkorpershoek@baylibre.com>

> ---
>  cmd/usb_mass_storage.c              |   4 --
>  drivers/usb/gadget/f_mass_storage.c | 101 ++++++++++++++++++++----------------
>  drivers/usb/gadget/storage_common.c |  12 +++--
>  include/usb_mass_storage.h          |   1 -
>  4 files changed, 65 insertions(+), 53 deletions(-)
>
> diff --git a/cmd/usb_mass_storage.c b/cmd/usb_mass_storage.c
> index a8ddeb494628..751701fe73af 100644
> --- a/cmd/usb_mass_storage.c
> +++ b/cmd/usb_mass_storage.c
> @@ -88,10 +88,6 @@ static int ums_init(const char *devtype, const char *devnums_part_str)
>  		if (!strchr(devnum_part_str, ':'))
>  			partnum = 0;
>  
> -		/* f_mass_storage.c assumes SECTOR_SIZE sectors */
> -		if (block_dev->blksz != SECTOR_SIZE)
> -			goto cleanup;
> -
>  		ums_new = realloc(ums, (ums_count + 1) * sizeof(*ums));
>  		if (!ums_new)
>  			goto cleanup;
> diff --git a/drivers/usb/gadget/f_mass_storage.c b/drivers/usb/gadget/f_mass_storage.c
> index c725aed3f626..d880928044f4 100644
> --- a/drivers/usb/gadget/f_mass_storage.c
> +++ b/drivers/usb/gadget/f_mass_storage.c
> @@ -724,12 +724,13 @@ static int do_read(struct fsg_common *common)
>  		curlun->sense_data = SS_LOGICAL_BLOCK_ADDRESS_OUT_OF_RANGE;
>  		return -EINVAL;
>  	}
> -	file_offset = ((loff_t) lba) << 9;
> +	file_offset = ((loff_t)lba) << curlun->blkbits;
>  
>  	/* Carry out the file reads */
>  	amount_left = common->data_size_from_cmnd;
> -	if (unlikely(amount_left == 0))
> +	if (unlikely(amount_left == 0)) {
>  		return -EIO;		/* No default reply */
> +	}
>  
>  	for (;;) {
>  
> @@ -768,13 +769,13 @@ static int do_read(struct fsg_common *common)
>  
>  		/* Perform the read */
>  		rc = ums[common->lun].read_sector(&ums[common->lun],
> -				      file_offset / SECTOR_SIZE,
> -				      amount / SECTOR_SIZE,
> +				      file_offset / curlun->blksize,
> +				      amount / curlun->blksize,
>  				      (char __user *)bh->buf);
>  		if (!rc)
>  			return -EIO;
>  
> -		nread = rc * SECTOR_SIZE;
> +		nread = rc * curlun->blksize;
>  
>  		VLDBG(curlun, "file read %u @ %llu -> %d\n", amount,
>  				(unsigned long long) file_offset,
> @@ -787,7 +788,7 @@ static int do_read(struct fsg_common *common)
>  		} else if (nread < amount) {
>  			LDBG(curlun, "partial file read: %d/%u\n",
>  					(int) nread, amount);
> -			nread -= (nread & 511);	/* Round down to a block */
> +			nread -= (nread & (curlun->blksize - 1));	/* Round down to a block */
>  		}
>  		file_offset  += nread;
>  		amount_left  -= nread;
> @@ -861,7 +862,7 @@ static int do_write(struct fsg_common *common)
>  
>  	/* Carry out the file writes */
>  	get_some_more = 1;
> -	file_offset = usb_offset = ((loff_t) lba) << 9;
> +	file_offset = usb_offset = ((loff_t)lba) << curlun->blkbits;
>  	amount_left_to_req = common->data_size_from_cmnd;
>  	amount_left_to_write = common->data_size_from_cmnd;
>  
> @@ -893,7 +894,7 @@ static int do_write(struct fsg_common *common)
>  				curlun->info_valid = 1;
>  				continue;
>  			}
> -			amount -= (amount & 511);
> +			amount -= (amount & (curlun->blksize - 1));
>  			if (amount == 0) {
>  
>  				/* Why were we were asked to transfer a
> @@ -942,12 +943,12 @@ static int do_write(struct fsg_common *common)
>  
>  			/* Perform the write */
>  			rc = ums[common->lun].write_sector(&ums[common->lun],
> -					       file_offset / SECTOR_SIZE,
> -					       amount / SECTOR_SIZE,
> +					       file_offset / curlun->blksize,
> +					       amount / curlun->blksize,
>  					       (char __user *)bh->buf);
>  			if (!rc)
>  				return -EIO;
> -			nwritten = rc * SECTOR_SIZE;
> +			nwritten = rc * curlun->blksize;
>  
>  			VLDBG(curlun, "file write %u @ %llu -> %d\n", amount,
>  					(unsigned long long) file_offset,
> @@ -960,7 +961,7 @@ static int do_write(struct fsg_common *common)
>  			} else if (nwritten < amount) {
>  				LDBG(curlun, "partial file write: %d/%u\n",
>  						(int) nwritten, amount);
> -				nwritten -= (nwritten & 511);
> +				nwritten -= (nwritten & (curlun->blksize - 1));
>  				/* Round down to a block */
>  			}
>  			file_offset += nwritten;
> @@ -1034,8 +1035,8 @@ static int do_verify(struct fsg_common *common)
>  		return -EIO;		/* No default reply */
>  
>  	/* Prepare to carry out the file verify */
> -	amount_left = verification_length << 9;
> -	file_offset = ((loff_t) lba) << 9;
> +	amount_left = verification_length << curlun->blkbits;
> +	file_offset = ((loff_t) lba) << curlun->blkbits;
>  
>  	/* Write out all the dirty buffers before invalidating them */
>  
> @@ -1058,12 +1059,12 @@ static int do_verify(struct fsg_common *common)
>  
>  		/* Perform the read */
>  		rc = ums[common->lun].read_sector(&ums[common->lun],
> -				      file_offset / SECTOR_SIZE,
> -				      amount / SECTOR_SIZE,
> +				      file_offset / curlun->blksize,
> +				      amount / curlun->blksize,
>  				      (char __user *)bh->buf);
>  		if (!rc)
>  			return -EIO;
> -		nread = rc * SECTOR_SIZE;
> +		nread = rc * curlun->blksize;
>  
>  		VLDBG(curlun, "file read %u @ %llu -> %d\n", amount,
>  				(unsigned long long) file_offset,
> @@ -1075,7 +1076,7 @@ static int do_verify(struct fsg_common *common)
>  		} else if (nread < amount) {
>  			LDBG(curlun, "partial file verify: %d/%u\n",
>  					(int) nread, amount);
> -			nread -= (nread & 511);	/* Round down to a sector */
> +			nread -= (nread & (curlun->blksize - 1));	/* Round down to a sector */
>  		}
>  		if (nread == 0) {
>  			curlun->sense_data = SS_UNRECOVERED_READ_ERROR;
> @@ -1183,7 +1184,7 @@ static int do_read_capacity(struct fsg_common *common, struct fsg_buffhd *bh)
>  
>  	put_unaligned_be32(curlun->num_sectors - 1, &buf[0]);
>  						/* Max logical block */
> -	put_unaligned_be32(512, &buf[4]);	/* Block length */
> +	put_unaligned_be32(curlun->blksize, &buf[4]);	/* Block length */
>  	return 8;
>  }
>  
> @@ -1370,7 +1371,7 @@ static int do_read_format_capacities(struct fsg_common *common,
>  
>  	put_unaligned_be32(curlun->num_sectors, &buf[0]);
>  						/* Number of blocks */
> -	put_unaligned_be32(512, &buf[4]);	/* Block length */
> +	put_unaligned_be32(curlun->blksize, &buf[4]);	/* Block length */
>  	buf[4] = 0x02;				/* Current capacity */
>  	return 12;
>  }
> @@ -1781,6 +1782,16 @@ static int check_command(struct fsg_common *common, int cmnd_size,
>  	return 0;
>  }
>  
> +/* wrapper of check_command for data size in blocks handling */
> +static int check_command_size_in_blocks(struct fsg_common *common,
> +		int cmnd_size, enum data_direction data_dir,
> +		unsigned int mask, int needs_medium, const char *name)
> +{
> +	common->data_size_from_cmnd <<= common->luns[common->lun].blkbits;
> +	return check_command(common, cmnd_size, data_dir,
> +			mask, needs_medium, name);
> +}
> +
>  
>  static int do_scsi_command(struct fsg_common *common)
>  {
> @@ -1865,30 +1876,30 @@ static int do_scsi_command(struct fsg_common *common)
>  
>  	case SC_READ_6:
>  		i = common->cmnd[4];
> -		common->data_size_from_cmnd = (i == 0 ? 256 : i) << 9;
> -		reply = check_command(common, 6, DATA_DIR_TO_HOST,
> -				      (7<<1) | (1<<4), 1,
> -				      "READ(6)");
> +		common->data_size_from_cmnd = (i == 0 ? 256 : i);
> +		reply = check_command_size_in_blocks(common, 6, DATA_DIR_TO_HOST,
> +						     (7<<1) | (1<<4), 1,
> +						     "READ(6)");
>  		if (reply == 0)
>  			reply = do_read(common);
>  		break;
>  
>  	case SC_READ_10:
>  		common->data_size_from_cmnd =
> -				get_unaligned_be16(&common->cmnd[7]) << 9;
> -		reply = check_command(common, 10, DATA_DIR_TO_HOST,
> -				      (1<<1) | (0xf<<2) | (3<<7), 1,
> -				      "READ(10)");
> +				get_unaligned_be16(&common->cmnd[7]);
> +		reply = check_command_size_in_blocks(common, 10, DATA_DIR_TO_HOST,
> +						     (1<<1) | (0xf<<2) | (3<<7), 1,
> +						     "READ(10)");
>  		if (reply == 0)
>  			reply = do_read(common);
>  		break;
>  
>  	case SC_READ_12:
>  		common->data_size_from_cmnd =
> -				get_unaligned_be32(&common->cmnd[6]) << 9;
> -		reply = check_command(common, 12, DATA_DIR_TO_HOST,
> -				      (1<<1) | (0xf<<2) | (0xf<<6), 1,
> -				      "READ(12)");
> +				get_unaligned_be32(&common->cmnd[6]);
> +		reply = check_command_size_in_blocks(common, 12, DATA_DIR_TO_HOST,
> +						     (1<<1) | (0xf<<2) | (0xf<<6), 1,
> +						     "READ(12)");
>  		if (reply == 0)
>  			reply = do_read(common);
>  		break;
> @@ -1983,30 +1994,30 @@ static int do_scsi_command(struct fsg_common *common)
>  
>  	case SC_WRITE_6:
>  		i = common->cmnd[4];
> -		common->data_size_from_cmnd = (i == 0 ? 256 : i) << 9;
> -		reply = check_command(common, 6, DATA_DIR_FROM_HOST,
> -				      (7<<1) | (1<<4), 1,
> -				      "WRITE(6)");
> +		common->data_size_from_cmnd = (i == 0 ? 256 : i);
> +		reply = check_command_size_in_blocks(common, 6, DATA_DIR_FROM_HOST,
> +						     (7<<1) | (1<<4), 1,
> +						     "WRITE(6)");
>  		if (reply == 0)
>  			reply = do_write(common);
>  		break;
>  
>  	case SC_WRITE_10:
>  		common->data_size_from_cmnd =
> -				get_unaligned_be16(&common->cmnd[7]) << 9;
> -		reply = check_command(common, 10, DATA_DIR_FROM_HOST,
> -				      (1<<1) | (0xf<<2) | (3<<7), 1,
> -				      "WRITE(10)");
> +				get_unaligned_be16(&common->cmnd[7]);
> +		reply = check_command_size_in_blocks(common, 10, DATA_DIR_FROM_HOST,
> +						     (1<<1) | (0xf<<2) | (3<<7), 1,
> +						     "WRITE(10)");
>  		if (reply == 0)
>  			reply = do_write(common);
>  		break;
>  
>  	case SC_WRITE_12:
>  		common->data_size_from_cmnd =
> -				get_unaligned_be32(&common->cmnd[6]) << 9;
> -		reply = check_command(common, 12, DATA_DIR_FROM_HOST,
> -				      (1<<1) | (0xf<<2) | (0xf<<6), 1,
> -				      "WRITE(12)");
> +				get_unaligned_be32(&common->cmnd[6]);
> +		reply = check_command_size_in_blocks(common, 12, DATA_DIR_FROM_HOST,
> +						     (1<<1) | (0xf<<2) | (0xf<<6), 1,
> +						     "WRITE(12)");
>  		if (reply == 0)
>  			reply = do_write(common);
>  		break;
> @@ -2497,7 +2508,7 @@ static struct fsg_common *fsg_common_init(struct fsg_common *common,
>  	for (i = 0; i < nluns; i++) {
>  		common->luns[i].removable = 1;
>  
> -		rc = fsg_lun_open(&common->luns[i], ums[i].num_sectors, "");
> +		rc = fsg_lun_open(&common->luns[i], ums[i].num_sectors, ums->block_dev.blksz, "");
>  		if (rc)
>  			goto error_luns;
>  	}
> diff --git a/drivers/usb/gadget/storage_common.c b/drivers/usb/gadget/storage_common.c
> index 5674e8fe4940..97dc6b6f729c 100644
> --- a/drivers/usb/gadget/storage_common.c
> +++ b/drivers/usb/gadget/storage_common.c
> @@ -269,6 +269,7 @@ struct device_attribute { int i; };
>  #define ETOOSMALL	525
>  
>  #include <log.h>
> +#include <linux/log2.h>
>  #include <usb_mass_storage.h>
>  #include <dm/device_compat.h>
>  
> @@ -290,6 +291,8 @@ struct fsg_lun {
>  	u32		sense_data;
>  	u32		sense_data_info;
>  	u32		unit_attention_data;
> +	unsigned int	blkbits;
> +	unsigned int	blksize; /* logical block size of bound block device */
>  
>  	struct device	dev;
>  };
> @@ -566,7 +569,7 @@ static struct usb_gadget_strings	fsg_stringtab = {
>   */
>  
>  static int fsg_lun_open(struct fsg_lun *curlun, unsigned int num_sectors,
> -			const char *filename)
> +			unsigned int sector_size, const char *filename)
>  {
>  	int				ro;
>  
> @@ -574,9 +577,12 @@ static int fsg_lun_open(struct fsg_lun *curlun, unsigned int num_sectors,
>  	ro = curlun->initially_ro;
>  
>  	curlun->ro = ro;
> -	curlun->file_length = num_sectors << 9;
> +	curlun->file_length = num_sectors * sector_size;
>  	curlun->num_sectors = num_sectors;
> -	debug("open backing file: %s\n", filename);
> +	curlun->blksize = sector_size;
> +	curlun->blkbits = order_base_2(sector_size >> 9) + 9;
> +	debug("blksize: %u\n", sector_size);
> +	debug("open backing file: '%s'\n", filename);
>  
>  	return 0;
>  }
> diff --git a/include/usb_mass_storage.h b/include/usb_mass_storage.h
> index 83ab93b530d7..6d83d93cad7f 100644
> --- a/include/usb_mass_storage.h
> +++ b/include/usb_mass_storage.h
> @@ -7,7 +7,6 @@
>  #ifndef __USB_MASS_STORAGE_H__
>  #define __USB_MASS_STORAGE_H__
>  
> -#define SECTOR_SIZE		0x200
>  #include <part.h>
>  #include <linux/usb/composite.h>
>  
>
> -- 
> 2.43.0

[PATCH 5/5] iommu: qcom-smmu: fix debugging

[Caleb Connolly]

The priv struct was wrong in dump_boot_mappings(). Causing errors when
compiling with -DDEBUG. Fix this.

Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>
---
 drivers/iommu/qcom-hyp-smmu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/iommu/qcom-hyp-smmu.c b/drivers/iommu/qcom-hyp-smmu.c
index 8e5cdb581550..f2b39de56f4a 100644
--- a/drivers/iommu/qcom-hyp-smmu.c
+++ b/drivers/iommu/qcom-hyp-smmu.c
@@ -319,7 +319,7 @@ static int qcom_smmu_connect(struct udevice *dev)
 }
 
 #ifdef DEBUG
-static inline void dump_boot_mappings(struct arm_smmu_priv *priv)
+static inline void dump_boot_mappings(struct qcom_smmu_priv *priv)
 {
 	u32 val;
 	int i;

-- 
2.43.0

Re: [PATCH 5/5] iommu: qcom-smmu: fix debugging

[Mattijs Korpershoek]

Hi Caleb,

Thank you for the patch.

On mer., janv. 31, 2024 at 14:57, Caleb Connolly <caleb.connolly@linaro.org> wrote:

> The priv struct was wrong in dump_boot_mappings(). Causing errors when
> compiling with -DDEBUG. Fix this.
>
> Signed-off-by: Caleb Connolly <caleb.connolly@linaro.org>

Reviewed-by: Mattijs Korpershoek <mkorpershoek@baylibre.com>

> ---
>  drivers/iommu/qcom-hyp-smmu.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/iommu/qcom-hyp-smmu.c b/drivers/iommu/qcom-hyp-smmu.c
> index 8e5cdb581550..f2b39de56f4a 100644
> --- a/drivers/iommu/qcom-hyp-smmu.c
> +++ b/drivers/iommu/qcom-hyp-smmu.c
> @@ -319,7 +319,7 @@ static int qcom_smmu_connect(struct udevice *dev)
>  }
>  
>  #ifdef DEBUG
> -static inline void dump_boot_mappings(struct arm_smmu_priv *priv)
> +static inline void dump_boot_mappings(struct qcom_smmu_priv *priv)
>  {
>  	u32 val;
>  	int i;
>
> -- 
> 2.43.0