[PATCH 1/2] scsi: ufs: ufs-qcom: add basic interconnect support

[PATCH 1/2] scsi: ufs: ufs-qcom: add basic interconnect support

[Brian Masney]

The firmware on the Qualcomm platforms expects the interconnect votes to
be present. Let's add very basic support where the maximum throughput is
requested to match what's done in a few other drivers.

This will not break boot on systems where the interconnects and
interconnect-names properties are not specified in device tree for UFS
since the interconnect framework will silently return.

Signed-off-by: Brian Masney <bmasney@redhat.com>
---
 drivers/ufs/host/ufs-qcom.c | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/drivers/ufs/host/ufs-qcom.c b/drivers/ufs/host/ufs-qcom.c
index 8ad1415e10b6..55bf8dd88985 100644
--- a/drivers/ufs/host/ufs-qcom.c
+++ b/drivers/ufs/host/ufs-qcom.c
@@ -7,6 +7,7 @@
 #include <linux/time.h>
 #include <linux/clk.h>
 #include <linux/delay.h>
+#include <linux/interconnect.h>
 #include <linux/module.h>
 #include <linux/of.h>
 #include <linux/platform_device.h>
@@ -936,6 +937,22 @@ static const struct reset_control_ops ufs_qcom_reset_ops = {
 	.deassert = ufs_qcom_reset_deassert,
 };
 
+static int ufs_qcom_icc_init(struct device *dev, char *pathname)
+{
+	struct icc_path *path;
+	int ret;
+
+	path = devm_of_icc_get(dev, pathname);
+	if (IS_ERR(path))
+		return dev_err_probe(dev, PTR_ERR(path), "failed to acquire interconnect path\n");
+
+	ret = icc_set_bw(path, 0, UINT_MAX);
+	if (ret < 0)
+		return dev_err_probe(dev, ret, "failed to set bandwidth request\n");
+
+	return 0;
+}
+
 /**
  * ufs_qcom_init - bind phy with controller
  * @hba: host controller instance
@@ -991,6 +1008,14 @@ static int ufs_qcom_init(struct ufs_hba *hba)
 			err = dev_err_probe(dev, PTR_ERR(host->generic_phy), "Failed to get PHY\n");
 			goto out_variant_clear;
 		}
+
+		err = ufs_qcom_icc_init(dev, "ufs-ddr");
+		if (err)
+			goto out_variant_clear;
+
+		err = ufs_qcom_icc_init(dev, "cpu-ufs");
+		if (err)
+			goto out_variant_clear;
 	}
 
 	host->device_reset = devm_gpiod_get_optional(dev, "reset",
-- 
2.38.1

Re: [PATCH 1/2] scsi: ufs: ufs-qcom: add basic interconnect support

[Manivannan Sadhasivam]

On Thu, Nov 17, 2022 at 05:49:56AM -0500, Brian Masney wrote:
> The firmware on the Qualcomm platforms expects the interconnect votes to
> be present. Let's add very basic support where the maximum throughput is
> requested to match what's done in a few other drivers.
> 
> This will not break boot on systems where the interconnects and
> interconnect-names properties are not specified in device tree for UFS
> since the interconnect framework will silently return.
> 
> Signed-off-by: Brian Masney <bmasney@redhat.com>
> ---
>  drivers/ufs/host/ufs-qcom.c | 25 +++++++++++++++++++++++++
>  1 file changed, 25 insertions(+)
> 
> diff --git a/drivers/ufs/host/ufs-qcom.c b/drivers/ufs/host/ufs-qcom.c
> index 8ad1415e10b6..55bf8dd88985 100644
> --- a/drivers/ufs/host/ufs-qcom.c
> +++ b/drivers/ufs/host/ufs-qcom.c
> @@ -7,6 +7,7 @@
>  #include <linux/time.h>
>  #include <linux/clk.h>
>  #include <linux/delay.h>
> +#include <linux/interconnect.h>
>  #include <linux/module.h>
>  #include <linux/of.h>
>  #include <linux/platform_device.h>
> @@ -936,6 +937,22 @@ static const struct reset_control_ops ufs_qcom_reset_ops = {
>  	.deassert = ufs_qcom_reset_deassert,
>  };
>  
> +static int ufs_qcom_icc_init(struct device *dev, char *pathname)
> +{
> +	struct icc_path *path;
> +	int ret;
> +
> +	path = devm_of_icc_get(dev, pathname);
> +	if (IS_ERR(path))
> +		return dev_err_probe(dev, PTR_ERR(path), "failed to acquire interconnect path\n");
> +
> +	ret = icc_set_bw(path, 0, UINT_MAX);

Please use icc macros for setting the bandwidth. Like, GBps_to_icc(),
MBps_to_icc() etc...

Also, during the init stage you can set a minimum bandwidth that will allow the
controller to get probed successfully. Then, you should update the bandwidth
based on the gear in pwr_change_notify() callback.

> +	if (ret < 0)
> +		return dev_err_probe(dev, ret, "failed to set bandwidth request\n");
> +
> +	return 0;
> +}
> +
>  /**
>   * ufs_qcom_init - bind phy with controller
>   * @hba: host controller instance
> @@ -991,6 +1008,14 @@ static int ufs_qcom_init(struct ufs_hba *hba)
>  			err = dev_err_probe(dev, PTR_ERR(host->generic_phy), "Failed to get PHY\n");
>  			goto out_variant_clear;
>  		}
> +
> +		err = ufs_qcom_icc_init(dev, "ufs-ddr");
> +		if (err)
> +			goto out_variant_clear;
> +
> +		err = ufs_qcom_icc_init(dev, "cpu-ufs");
> +		if (err)
> +			goto out_variant_clear;

It'd be nice to have a single function that initializes both paths.

Thanks,
Mani

>  	}
>  
>  	host->device_reset = devm_gpiod_get_optional(dev, "reset",
> -- 
> 2.38.1
> 

-- 
மணிவண்ணன் சதாசிவம்

Re: [PATCH 1/2] scsi: ufs: ufs-qcom: add basic interconnect support

[Dmitry Baryshkov]

On 17/11/2022 12:49, Brian Masney wrote:
> The firmware on the Qualcomm platforms expects the interconnect votes to
> be present. Let's add very basic support where the maximum throughput is
> requested to match what's done in a few other drivers.
> 
> This will not break boot on systems where the interconnects and
> interconnect-names properties are not specified in device tree for UFS
> since the interconnect framework will silently return.
> 
> Signed-off-by: Brian Masney <bmasney@redhat.com>
> ---
>   drivers/ufs/host/ufs-qcom.c | 25 +++++++++++++++++++++++++
>   1 file changed, 25 insertions(+)
> 
> diff --git a/drivers/ufs/host/ufs-qcom.c b/drivers/ufs/host/ufs-qcom.c
> index 8ad1415e10b6..55bf8dd88985 100644
> --- a/drivers/ufs/host/ufs-qcom.c
> +++ b/drivers/ufs/host/ufs-qcom.c
> @@ -7,6 +7,7 @@
>   #include <linux/time.h>
>   #include <linux/clk.h>
>   #include <linux/delay.h>
> +#include <linux/interconnect.h>
>   #include <linux/module.h>
>   #include <linux/of.h>
>   #include <linux/platform_device.h>
> @@ -936,6 +937,22 @@ static const struct reset_control_ops ufs_qcom_reset_ops = {
>   	.deassert = ufs_qcom_reset_deassert,
>   };
>   
> +static int ufs_qcom_icc_init(struct device *dev, char *pathname)
> +{
> +	struct icc_path *path;
> +	int ret;
> +
> +	path = devm_of_icc_get(dev, pathname);
> +	if (IS_ERR(path))
> +		return dev_err_probe(dev, PTR_ERR(path), "failed to acquire interconnect path\n");
> +
> +	ret = icc_set_bw(path, 0, UINT_MAX);

I noticed that this patch bumps peak_bw (and leaves average_bw as 0), 
while vendor kernels bump average_bw, but ib (peak_bw) is set to 0.

> +	if (ret < 0)
> +		return dev_err_probe(dev, ret, "failed to set bandwidth request\n");
> +
> +	return 0;
> +}
> +
>   /**
>    * ufs_qcom_init - bind phy with controller
>    * @hba: host controller instance
> @@ -991,6 +1008,14 @@ static int ufs_qcom_init(struct ufs_hba *hba)
>   			err = dev_err_probe(dev, PTR_ERR(host->generic_phy), "Failed to get PHY\n");
>   			goto out_variant_clear;
>   		}
> +
> +		err = ufs_qcom_icc_init(dev, "ufs-ddr");
> +		if (err)
> +			goto out_variant_clear;
> +
> +		err = ufs_qcom_icc_init(dev, "cpu-ufs");
> +		if (err)
> +			goto out_variant_clear;
>   	}
>   
>   	host->device_reset = devm_gpiod_get_optional(dev, "reset",

-- 
With best wishes
Dmitry

[PATCH 0/2] arm64/msm8996: enable UFS interconnect

[Dmitry Baryshkov]

MSM8996 requires a vote on UFS interconnects to work in a stable manner.
The first patch is a rework of older patch from Brian, see [1]

[1] https://lore.kernel.org/all/20221117104957.254648-2-bmasney@redhat.com/

Brian Masney (1):
  scsi: ufs: ufs-qcom: add basic interconnect support

Dmitry Baryshkov (1):
  arm64: dts: qcom: msm8996: enable UFS interconnects

 arch/arm64/boot/dts/qcom/msm8996.dtsi |  4 ++++
 drivers/ufs/host/ufs-qcom.c           | 26 ++++++++++++++++++++++++++
 2 files changed, 30 insertions(+)

-- 
2.39.0

[PATCH 1/2] scsi: ufs: ufs-qcom: add basic interconnect support

[Dmitry Baryshkov]

From: Brian Masney <bmasney@redhat.com>

The firmware on the Qualcomm platforms expects the interconnect votes to
be present. Let's add very basic support where the maximum throughput is
requested to match what's done in a few other drivers.

This will not break boot on systems where the interconnects and
interconnect-names properties are not specified in device tree for UFS
since the interconnect framework will silently return.

Signed-off-by: Brian Masney <bmasney@redhat.com>
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
---
 drivers/ufs/host/ufs-qcom.c | 26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)

diff --git a/drivers/ufs/host/ufs-qcom.c b/drivers/ufs/host/ufs-qcom.c
index 8ad1415e10b6..8267a4056120 100644
--- a/drivers/ufs/host/ufs-qcom.c
+++ b/drivers/ufs/host/ufs-qcom.c
@@ -14,6 +14,7 @@
 #include <linux/gpio/consumer.h>
 #include <linux/reset-controller.h>
 #include <linux/devfreq.h>
+#include <linux/interconnect.h>
 
 #include <ufs/ufshcd.h>
 #include "ufshcd-pltfrm.h"
@@ -936,6 +937,23 @@ static const struct reset_control_ops ufs_qcom_reset_ops = {
 	.deassert = ufs_qcom_reset_deassert,
 };
 
+static int ufs_qcom_icc_init(struct device *dev, char *pathname,
+		u32 avg_bw, u32 peak_bw)
+{
+	struct icc_path *path;
+	int ret;
+
+	path = devm_of_icc_get(dev, pathname);
+	if (IS_ERR(path))
+		return dev_err_probe(dev, PTR_ERR(path), "failed to acquire interconnect path\n");
+
+	ret = icc_set_bw(path, avg_bw, peak_bw);
+	if (ret < 0)
+		return dev_err_probe(dev, ret, "failed to set bandwidth request\n");
+
+	return 0;
+}
+
 /**
  * ufs_qcom_init - bind phy with controller
  * @hba: host controller instance
@@ -1005,6 +1023,14 @@ static int ufs_qcom_init(struct ufs_hba *hba)
 	ufs_qcom_get_controller_revision(hba, &host->hw_ver.major,
 		&host->hw_ver.minor, &host->hw_ver.step);
 
+	err = ufs_qcom_icc_init(dev, "ufs-ddr", 4096000, 0);
+	if (err)
+		goto out_variant_clear;
+
+	err = ufs_qcom_icc_init(dev, "cpu-ufs", 1000, 0);
+	if (err)
+		goto out_variant_clear;
+
 	/*
 	 * for newer controllers, device reference clock control bit has
 	 * moved inside UFS controller register address space itself.
-- 
2.39.0

[PATCH 2/2] arm64: dts: qcom: msm8996: enable UFS interconnects

[Dmitry Baryshkov]

Specify interconnects to be used by the UFS host controller.

Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
---
 arch/arm64/boot/dts/qcom/msm8996.dtsi | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/arch/arm64/boot/dts/qcom/msm8996.dtsi b/arch/arm64/boot/dts/qcom/msm8996.dtsi
index 55180586f7b6..0c2f7be9f205 100644
--- a/arch/arm64/boot/dts/qcom/msm8996.dtsi
+++ b/arch/arm64/boot/dts/qcom/msm8996.dtsi
@@ -2037,6 +2037,10 @@ ufshc: ufshc@624000 {
 				<0 0>,
 				<0 0>;
 
+			interconnects = <&a2noc MASTER_UFS &bimc SLAVE_EBI_CH0>,
+					<&bimc MASTER_AMPSS_M0 &cnoc SLAVE_UFS_CFG>;
+			interconnect-names = "ufs-ddr", "cpu-ufs";
+
 			lanes-per-direction = <1>;
 			#reset-cells = <1>;
 			status = "disabled";
-- 
2.39.0

Re: (subset) [PATCH 0/2] arm64/msm8996: enable UFS interconnect

[Bjorn Andersson]

On Thu, 19 Jan 2023 16:43:24 +0200, Dmitry Baryshkov wrote:
> MSM8996 requires a vote on UFS interconnects to work in a stable manner.
> The first patch is a rework of older patch from Brian, see [1]
> 
> [1] https://lore.kernel.org/all/20221117104957.254648-2-bmasney@redhat.com/
> 
> Brian Masney (1):
>   scsi: ufs: ufs-qcom: add basic interconnect support
> 
> [...]

Applied, thanks!

[2/2] arm64: dts: qcom: msm8996: enable UFS interconnects
      commit: bc72f13e4456afa34ccbd1dfc61aaea18f877b88

Best regards,
-- 
Bjorn Andersson <andersson@kernel.org>

Re: [PATCH 1/2] scsi: ufs: ufs-qcom: add basic interconnect support

[Konrad Dybcio]

On 17.11.2022 11:49, Brian Masney wrote:
> The firmware on the Qualcomm platforms expects the interconnect votes to
> be present. Let's add very basic support where the maximum throughput is
> requested to match what's done in a few other drivers.
> 
> This will not break boot on systems where the interconnects and
> interconnect-names properties are not specified in device tree for UFS
> since the interconnect framework will silently return.
> 
> Signed-off-by: Brian Masney <bmasney@redhat.com>
> ---
Hi everyone!

This was never merged, but it's actually strictly necessary!

For example UFS dies on SM8450 if we add sync_state to its interconnect
driver, as there's no votes cast.

Can we look into this again?

Konrad
>  drivers/ufs/host/ufs-qcom.c | 25 +++++++++++++++++++++++++
>  1 file changed, 25 insertions(+)
> 
> diff --git a/drivers/ufs/host/ufs-qcom.c b/drivers/ufs/host/ufs-qcom.c
> index 8ad1415e10b6..55bf8dd88985 100644
> --- a/drivers/ufs/host/ufs-qcom.c
> +++ b/drivers/ufs/host/ufs-qcom.c
> @@ -7,6 +7,7 @@
>  #include <linux/time.h>
>  #include <linux/clk.h>
>  #include <linux/delay.h>
> +#include <linux/interconnect.h>
>  #include <linux/module.h>
>  #include <linux/of.h>
>  #include <linux/platform_device.h>
> @@ -936,6 +937,22 @@ static const struct reset_control_ops ufs_qcom_reset_ops = {
>  	.deassert = ufs_qcom_reset_deassert,
>  };
>  
> +static int ufs_qcom_icc_init(struct device *dev, char *pathname)
> +{
> +	struct icc_path *path;
> +	int ret;
> +
> +	path = devm_of_icc_get(dev, pathname);
> +	if (IS_ERR(path))
> +		return dev_err_probe(dev, PTR_ERR(path), "failed to acquire interconnect path\n");
> +
> +	ret = icc_set_bw(path, 0, UINT_MAX);
> +	if (ret < 0)
> +		return dev_err_probe(dev, ret, "failed to set bandwidth request\n");
> +
> +	return 0;
> +}
> +
>  /**
>   * ufs_qcom_init - bind phy with controller
>   * @hba: host controller instance
> @@ -991,6 +1008,14 @@ static int ufs_qcom_init(struct ufs_hba *hba)
>  			err = dev_err_probe(dev, PTR_ERR(host->generic_phy), "Failed to get PHY\n");
>  			goto out_variant_clear;
>  		}
> +
> +		err = ufs_qcom_icc_init(dev, "ufs-ddr");
> +		if (err)
> +			goto out_variant_clear;
> +
> +		err = ufs_qcom_icc_init(dev, "cpu-ufs");
> +		if (err)
> +			goto out_variant_clear;
>  	}
>  
>  	host->device_reset = devm_gpiod_get_optional(dev, "reset",