[PATCH] clk: imx: lpcg-scu: SW workaround for errata (e10858)

[PATCH] clk: imx: lpcg-scu: SW workaround for errata (e10858)

[Peng Fan (OSS)]

From: Peng Fan <peng.fan@nxp.com>

Back-to-back LPCG writes can be ignored by the LPCG register due to
a HW bug. The writes need to be separated by at least 4 cycles of
the gated clock.

The workaround is implemented as follows:
1. For clocks running greater than or equal to 24MHz, a read
followed by the write will provide sufficient delay.
2. For clocks running below 24MHz, add a delay of 4 clock cylces
after the write to the LPCG register.

Signed-off-by: Peng Fan <peng.fan@nxp.com>
---

 This is to upstream a downstream patch that goes through many releases
 in several years. It maybe hard to trigger, but to address potential
 issues, let's add this workaround.

 drivers/clk/imx/clk-lpcg-scu.c | 33 ++++++++++++++++++++++++++++++---
 1 file changed, 30 insertions(+), 3 deletions(-)

diff --git a/drivers/clk/imx/clk-lpcg-scu.c b/drivers/clk/imx/clk-lpcg-scu.c
index dd5abd09f3e2..b30d0f8b5bca 100644
--- a/drivers/clk/imx/clk-lpcg-scu.c
+++ b/drivers/clk/imx/clk-lpcg-scu.c
@@ -6,6 +6,7 @@
 
 #include <linux/bits.h>
 #include <linux/clk-provider.h>
+#include <linux/delay.h>
 #include <linux/err.h>
 #include <linux/io.h>
 #include <linux/slab.h>
@@ -41,6 +42,31 @@ struct clk_lpcg_scu {
 
 #define to_clk_lpcg_scu(_hw) container_of(_hw, struct clk_lpcg_scu, hw)
 
+/* e10858 -LPCG clock gating register synchronization errata */
+static void do_lpcg_workaround(u32 rate, void __iomem *reg, u32 val)
+{
+	writel(val, reg);
+
+	if (rate >= 24000000 || rate == 0) {
+		u32 reg1;
+
+		/*
+		 * The time taken to access the LPCG registers from the AP core
+		 * through the interconnect is longer than the minimum delay
+		 * of 4 clock cycles required by the errata.
+		 * Adding a readl will provide sufficient delay to prevent
+		 * back-to-back writes.
+		 */
+		reg1 = readl(reg);
+	} else {
+		/*
+		 * For clocks running below 24MHz, wait a minimum of
+		 * 4 clock cycles.
+		 */
+		ndelay(4 * (DIV_ROUND_UP(1000000000, rate)));
+	}
+}
+
 static int clk_lpcg_scu_enable(struct clk_hw *hw)
 {
 	struct clk_lpcg_scu *clk = to_clk_lpcg_scu(hw);
@@ -57,7 +83,8 @@ static int clk_lpcg_scu_enable(struct clk_hw *hw)
 		val |= CLK_GATE_SCU_LPCG_HW_SEL;
 
 	reg |= val << clk->bit_idx;
-	writel(reg, clk->reg);
+
+	do_lpcg_workaround(clk_hw_get_rate(hw), clk->reg, reg);
 
 	spin_unlock_irqrestore(&imx_lpcg_scu_lock, flags);
 
@@ -74,7 +101,7 @@ static void clk_lpcg_scu_disable(struct clk_hw *hw)
 
 	reg = readl_relaxed(clk->reg);
 	reg &= ~(CLK_GATE_SCU_LPCG_MASK << clk->bit_idx);
-	writel(reg, clk->reg);
+	do_lpcg_workaround(clk_hw_get_rate(hw), clk->reg, reg);
 
 	spin_unlock_irqrestore(&imx_lpcg_scu_lock, flags);
 }
@@ -151,7 +178,7 @@ static int __maybe_unused imx_clk_lpcg_scu_resume(struct device *dev)
 	 */
 
 	writel(clk->state, clk->reg);
-	writel(clk->state, clk->reg);
+	do_lpcg_workaround(0, clk->reg, clk->state);
 	dev_dbg(dev, "restore lpcg state 0x%x\n", clk->state);
 
 	return 0;
-- 
2.37.1

Re: [PATCH] clk: imx: lpcg-scu: SW workaround for errata (e10858)

[Frank Li]

On Wed, Feb 28, 2024 at 04:26:49PM +0800, Peng Fan (OSS) wrote:
> From: Peng Fan <peng.fan@nxp.com>
> 
> Back-to-back LPCG writes can be ignored by the LPCG register due to
> a HW bug. The writes need to be separated by at least 4 cycles of
> the gated clock.
> 
> The workaround is implemented as follows:
> 1. For clocks running greater than or equal to 24MHz, a read
> followed by the write will provide sufficient delay.
> 2. For clocks running below 24MHz, add a delay of 4 clock cylces
> after the write to the LPCG register.
> 
> Signed-off-by: Peng Fan <peng.fan@nxp.com>
> ---

Fixes tag? 

Frank
 
> 
>  This is to upstream a downstream patch that goes through many releases
>  in several years. It maybe hard to trigger, but to address potential
>  issues, let's add this workaround.
> 
>  drivers/clk/imx/clk-lpcg-scu.c | 33 ++++++++++++++++++++++++++++++---
>  1 file changed, 30 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/clk/imx/clk-lpcg-scu.c b/drivers/clk/imx/clk-lpcg-scu.c
> index dd5abd09f3e2..b30d0f8b5bca 100644
> --- a/drivers/clk/imx/clk-lpcg-scu.c
> +++ b/drivers/clk/imx/clk-lpcg-scu.c
> @@ -6,6 +6,7 @@
>  
>  #include <linux/bits.h>
>  #include <linux/clk-provider.h>
> +#include <linux/delay.h>
>  #include <linux/err.h>
>  #include <linux/io.h>
>  #include <linux/slab.h>
> @@ -41,6 +42,31 @@ struct clk_lpcg_scu {
>  
>  #define to_clk_lpcg_scu(_hw) container_of(_hw, struct clk_lpcg_scu, hw)
>  
> +/* e10858 -LPCG clock gating register synchronization errata */
> +static void do_lpcg_workaround(u32 rate, void __iomem *reg, u32 val)
> +{
> +	writel(val, reg);
> +
> +	if (rate >= 24000000 || rate == 0) {
> +		u32 reg1;
> +
> +		/*
> +		 * The time taken to access the LPCG registers from the AP core
> +		 * through the interconnect is longer than the minimum delay
> +		 * of 4 clock cycles required by the errata.
> +		 * Adding a readl will provide sufficient delay to prevent
> +		 * back-to-back writes.
> +		 */
> +		reg1 = readl(reg);
> +	} else {
> +		/*
> +		 * For clocks running below 24MHz, wait a minimum of
> +		 * 4 clock cycles.
> +		 */
> +		ndelay(4 * (DIV_ROUND_UP(1000000000, rate)));
> +	}
> +}
> +
>  static int clk_lpcg_scu_enable(struct clk_hw *hw)
>  {
>  	struct clk_lpcg_scu *clk = to_clk_lpcg_scu(hw);
> @@ -57,7 +83,8 @@ static int clk_lpcg_scu_enable(struct clk_hw *hw)
>  		val |= CLK_GATE_SCU_LPCG_HW_SEL;
>  
>  	reg |= val << clk->bit_idx;
> -	writel(reg, clk->reg);
> +
> +	do_lpcg_workaround(clk_hw_get_rate(hw), clk->reg, reg);
>  
>  	spin_unlock_irqrestore(&imx_lpcg_scu_lock, flags);
>  
> @@ -74,7 +101,7 @@ static void clk_lpcg_scu_disable(struct clk_hw *hw)
>  
>  	reg = readl_relaxed(clk->reg);
>  	reg &= ~(CLK_GATE_SCU_LPCG_MASK << clk->bit_idx);
> -	writel(reg, clk->reg);
> +	do_lpcg_workaround(clk_hw_get_rate(hw), clk->reg, reg);
>  
>  	spin_unlock_irqrestore(&imx_lpcg_scu_lock, flags);
>  }
> @@ -151,7 +178,7 @@ static int __maybe_unused imx_clk_lpcg_scu_resume(struct device *dev)
>  	 */
>  
>  	writel(clk->state, clk->reg);
> -	writel(clk->state, clk->reg);
> +	do_lpcg_workaround(0, clk->reg, clk->state);
>  	dev_dbg(dev, "restore lpcg state 0x%x\n", clk->state);
>  
>  	return 0;
> -- 
> 2.37.1
> 

Re: [PATCH] clk: imx: lpcg-scu: SW workaround for errata (e10858)

[Stephen Boyd]

Quoting Peng Fan (OSS) (2024-02-28 00:26:49)
> diff --git a/drivers/clk/imx/clk-lpcg-scu.c b/drivers/clk/imx/clk-lpcg-scu.c
> index dd5abd09f3e2..b30d0f8b5bca 100644
> --- a/drivers/clk/imx/clk-lpcg-scu.c
> +++ b/drivers/clk/imx/clk-lpcg-scu.c
> @@ -6,6 +6,7 @@
>  
>  #include <linux/bits.h>
>  #include <linux/clk-provider.h>
> +#include <linux/delay.h>
>  #include <linux/err.h>
>  #include <linux/io.h>
>  #include <linux/slab.h>
> @@ -41,6 +42,31 @@ struct clk_lpcg_scu {
>  
>  #define to_clk_lpcg_scu(_hw) container_of(_hw, struct clk_lpcg_scu, hw)
>  
> +/* e10858 -LPCG clock gating register synchronization errata */
> +static void do_lpcg_workaround(u32 rate, void __iomem *reg, u32 val)

unsigned long rate

> +{
> +       writel(val, reg);
> +
> +       if (rate >= 24000000 || rate == 0) {
> +               u32 reg1;

Please declare this variable at the start of the function.

> +
> +               /*
> +                * The time taken to access the LPCG registers from the AP core
> +                * through the interconnect is longer than the minimum delay
> +                * of 4 clock cycles required by the errata.
> +                * Adding a readl will provide sufficient delay to prevent
> +                * back-to-back writes.
> +                */
> +               reg1 = readl(reg);
> +       } else {
> +               /*
> +                * For clocks running below 24MHz, wait a minimum of
> +                * 4 clock cycles.
> +                */
> +               ndelay(4 * (DIV_ROUND_UP(1000000000, rate)));
> +       }
> +}
> +
>  static int clk_lpcg_scu_enable(struct clk_hw *hw)
>  {
>         struct clk_lpcg_scu *clk = to_clk_lpcg_scu(hw);
> @@ -57,7 +83,8 @@ static int clk_lpcg_scu_enable(struct clk_hw *hw)
>                 val |= CLK_GATE_SCU_LPCG_HW_SEL;
>  
>         reg |= val << clk->bit_idx;
> -       writel(reg, clk->reg);
> +
> +       do_lpcg_workaround(clk_hw_get_rate(hw), clk->reg, reg);

I'd prefer the name had 'writel' in it somewhere.
>  
>         spin_unlock_irqrestore(&imx_lpcg_scu_lock, flags);
>

[PATCH] clk: imx: lpcg-scu: SW workaround for errata (e10858)

[peng.fan]

From: Peng Fan <peng.fan@nxp.com>

Back-to-back LPCG writes can be ignored by the LPCG register due to
a HW bug. The writes need to be separated by at least 4 cycles of
the gated clock.

The workaround is implemented as follows:
1. For clocks running greater than or equal to 24MHz, a read
followed by the write will provide sufficient delay.
2. For clocks running below 24MHz, add a delay of 4 clock cylces
after the write to the LPCG register.

Signed-off-by: Peng Fan <peng.fan@nxp.com>
---
 drivers/clk/imx/clk-lpcg-scu.c | 31 +++++++++++++++++++++++++++++--
 1 file changed, 29 insertions(+), 2 deletions(-)

diff --git a/drivers/clk/imx/clk-lpcg-scu.c b/drivers/clk/imx/clk-lpcg-scu.c
index 1f0e44f921ae..6ee9d2caedf2 100644
--- a/drivers/clk/imx/clk-lpcg-scu.c
+++ b/drivers/clk/imx/clk-lpcg-scu.c
@@ -6,6 +6,7 @@
 
 #include <linux/bits.h>
 #include <linux/clk-provider.h>
+#include <linux/delay.h>
 #include <linux/err.h>
 #include <linux/io.h>
 #include <linux/slab.h>
@@ -38,6 +39,31 @@ struct clk_lpcg_scu {
 
 #define to_clk_lpcg_scu(_hw) container_of(_hw, struct clk_lpcg_scu, hw)
 
+/* e10858 -LPCG clock gating register synchronization errata */
+static void do_lpcg_workaround(u32 rate, void __iomem *reg, u32 val)
+{
+	writel(val, reg);
+
+	if (rate >= 24000000 || rate == 0) {
+		u32 reg1;
+
+		/*
+		 * The time taken to access the LPCG registers from the AP core
+		 * through the interconnect is longer than the minimum delay
+		 * of 4 clock cycles required by the errata.
+		 * Adding a readl will provide sufficient delay to prevent
+		 * back-to-back writes.
+		 */
+		reg1 = readl(reg);
+	} else {
+		/*
+		 * For clocks running below 24MHz, wait a minimum of
+		 * 4 clock cycles.
+		 */
+		ndelay(4 * (DIV_ROUND_UP(1000000000, rate)));
+	}
+}
+
 static int clk_lpcg_scu_enable(struct clk_hw *hw)
 {
 	struct clk_lpcg_scu *clk = to_clk_lpcg_scu(hw);
@@ -54,7 +80,8 @@ static int clk_lpcg_scu_enable(struct clk_hw *hw)
 		val |= CLK_GATE_SCU_LPCG_HW_SEL;
 
 	reg |= val << clk->bit_idx;
-	writel(reg, clk->reg);
+
+	do_lpcg_workaround(clk_hw_get_rate(hw), clk->reg, reg);
 
 	spin_unlock_irqrestore(&imx_lpcg_scu_lock, flags);
 
@@ -71,7 +98,7 @@ static void clk_lpcg_scu_disable(struct clk_hw *hw)
 
 	reg = readl_relaxed(clk->reg);
 	reg &= ~(CLK_GATE_SCU_LPCG_MASK << clk->bit_idx);
-	writel(reg, clk->reg);
+	do_lpcg_workaround(clk_hw_get_rate(hw), clk->reg, reg);
 
 	spin_unlock_irqrestore(&imx_lpcg_scu_lock, flags);
 }
-- 
2.28.0

Re: [PATCH] clk: imx: lpcg-scu: SW workaround for errata (e10858)

[Abel Vesa]

On 20-09-08 17:14:47, peng.fan@nxp.com wrote:
> From: Peng Fan <peng.fan@nxp.com>
> 
> Back-to-back LPCG writes can be ignored by the LPCG register due to
> a HW bug. The writes need to be separated by at least 4 cycles of
> the gated clock.
> 
> The workaround is implemented as follows:
> 1. For clocks running greater than or equal to 24MHz, a read
> followed by the write will provide sufficient delay.
> 2. For clocks running below 24MHz, add a delay of 4 clock cylces
> after the write to the LPCG register.
> 
> Signed-off-by: Peng Fan <peng.fan@nxp.com>
> ---
>  drivers/clk/imx/clk-lpcg-scu.c | 31 +++++++++++++++++++++++++++++--
>  1 file changed, 29 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/clk/imx/clk-lpcg-scu.c b/drivers/clk/imx/clk-lpcg-scu.c
> index 1f0e44f921ae..6ee9d2caedf2 100644
> --- a/drivers/clk/imx/clk-lpcg-scu.c
> +++ b/drivers/clk/imx/clk-lpcg-scu.c
> @@ -6,6 +6,7 @@
>  
>  #include <linux/bits.h>
>  #include <linux/clk-provider.h>
> +#include <linux/delay.h>
>  #include <linux/err.h>
>  #include <linux/io.h>
>  #include <linux/slab.h>
> @@ -38,6 +39,31 @@ struct clk_lpcg_scu {
>  
>  #define to_clk_lpcg_scu(_hw) container_of(_hw, struct clk_lpcg_scu, hw)
>  
> +/* e10858 -LPCG clock gating register synchronization errata */
> +static void do_lpcg_workaround(u32 rate, void __iomem *reg, u32 val)
> +{
> +	writel(val, reg);
> +
> +	if (rate >= 24000000 || rate == 0) {
> +		u32 reg1;
> +
> +		/*
> +		 * The time taken to access the LPCG registers from the AP core
> +		 * through the interconnect is longer than the minimum delay
> +		 * of 4 clock cycles required by the errata.
> +		 * Adding a readl will provide sufficient delay to prevent
> +		 * back-to-back writes.
> +		 */
> +		reg1 = readl(reg);
> +	} else {
> +		/*
> +		 * For clocks running below 24MHz, wait a minimum of
> +		 * 4 clock cycles.
> +		 */
> +		ndelay(4 * (DIV_ROUND_UP(1000000000, rate)));
> +	}

Just to make sure this is totally understood, if the lpcg consumer
needs to two enables/disables in less than 4 multiplied by the clock
period, the second enable/disable will be delayed.

If we're fine with this, then here is my R-b.

Reviewed-by: Abel Vesa <abel.vesa@nxp.com>

> +}
> +
>  static int clk_lpcg_scu_enable(struct clk_hw *hw)
>  {
>  	struct clk_lpcg_scu *clk = to_clk_lpcg_scu(hw);
> @@ -54,7 +80,8 @@ static int clk_lpcg_scu_enable(struct clk_hw *hw)
>  		val |= CLK_GATE_SCU_LPCG_HW_SEL;
>  
>  	reg |= val << clk->bit_idx;
> -	writel(reg, clk->reg);
> +
> +	do_lpcg_workaround(clk_hw_get_rate(hw), clk->reg, reg);
>  
>  	spin_unlock_irqrestore(&imx_lpcg_scu_lock, flags);
>  
> @@ -71,7 +98,7 @@ static void clk_lpcg_scu_disable(struct clk_hw *hw)
>  
>  	reg = readl_relaxed(clk->reg);
>  	reg &= ~(CLK_GATE_SCU_LPCG_MASK << clk->bit_idx);
> -	writel(reg, clk->reg);
> +	do_lpcg_workaround(clk_hw_get_rate(hw), clk->reg, reg);
>  
>  	spin_unlock_irqrestore(&imx_lpcg_scu_lock, flags);
>  }
> -- 
> 2.28.0
> 

RE: [PATCH] clk: imx: lpcg-scu: SW workaround for errata (e10858)

[Peng Fan]

> Subject: Re: [PATCH] clk: imx: lpcg-scu: SW workaround for errata (e10858)
> 
> On 20-09-08 17:14:47, peng.fan@nxp.com wrote:
> > From: Peng Fan <peng.fan@nxp.com>
> >
> > Back-to-back LPCG writes can be ignored by the LPCG register due to a
> > HW bug. The writes need to be separated by at least 4 cycles of the
> > gated clock.
> >
> > The workaround is implemented as follows:
> > 1. For clocks running greater than or equal to 24MHz, a read followed
> > by the write will provide sufficient delay.
> > 2. For clocks running below 24MHz, add a delay of 4 clock cylces after
> > the write to the LPCG register.
> >
> > Signed-off-by: Peng Fan <peng.fan@nxp.com>
> > ---
> >  drivers/clk/imx/clk-lpcg-scu.c | 31 +++++++++++++++++++++++++++++--
> >  1 file changed, 29 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/clk/imx/clk-lpcg-scu.c
> > b/drivers/clk/imx/clk-lpcg-scu.c index 1f0e44f921ae..6ee9d2caedf2
> > 100644
> > --- a/drivers/clk/imx/clk-lpcg-scu.c
> > +++ b/drivers/clk/imx/clk-lpcg-scu.c
> > @@ -6,6 +6,7 @@
> >
> >  #include <linux/bits.h>
> >  #include <linux/clk-provider.h>
> > +#include <linux/delay.h>
> >  #include <linux/err.h>
> >  #include <linux/io.h>
> >  #include <linux/slab.h>
> > @@ -38,6 +39,31 @@ struct clk_lpcg_scu {
> >
> >  #define to_clk_lpcg_scu(_hw) container_of(_hw, struct clk_lpcg_scu,
> > hw)
> >
> > +/* e10858 -LPCG clock gating register synchronization errata */
> > +static void do_lpcg_workaround(u32 rate, void __iomem *reg, u32 val)
> > +{
> > +	writel(val, reg);
> > +
> > +	if (rate >= 24000000 || rate == 0) {
> > +		u32 reg1;
> > +
> > +		/*
> > +		 * The time taken to access the LPCG registers from the AP core
> > +		 * through the interconnect is longer than the minimum delay
> > +		 * of 4 clock cycles required by the errata.
> > +		 * Adding a readl will provide sufficient delay to prevent
> > +		 * back-to-back writes.
> > +		 */
> > +		reg1 = readl(reg);
> > +	} else {
> > +		/*
> > +		 * For clocks running below 24MHz, wait a minimum of
> > +		 * 4 clock cycles.
> > +		 */
> > +		ndelay(4 * (DIV_ROUND_UP(1000000000, rate)));
> > +	}
> 
> Just to make sure this is totally understood, if the lpcg consumer needs to two
> enables/disables in less than 4 multiplied by the clock period, the second
> enable/disable will be delayed.

You surely wanna this following errata description.

Thanks,
Peng.

> 
> If we're fine with this, then here is my R-b.
> 
> Reviewed-by: Abel Vesa <abel.vesa@nxp.com>
> 
> > +}
> > +
> >  static int clk_lpcg_scu_enable(struct clk_hw *hw)  {
> >  	struct clk_lpcg_scu *clk = to_clk_lpcg_scu(hw); @@ -54,7 +80,8 @@
> > static int clk_lpcg_scu_enable(struct clk_hw *hw)
> >  		val |= CLK_GATE_SCU_LPCG_HW_SEL;
> >
> >  	reg |= val << clk->bit_idx;
> > -	writel(reg, clk->reg);
> > +
> > +	do_lpcg_workaround(clk_hw_get_rate(hw), clk->reg, reg);
> >
> >  	spin_unlock_irqrestore(&imx_lpcg_scu_lock, flags);
> >
> > @@ -71,7 +98,7 @@ static void clk_lpcg_scu_disable(struct clk_hw *hw)
> >
> >  	reg = readl_relaxed(clk->reg);
> >  	reg &= ~(CLK_GATE_SCU_LPCG_MASK << clk->bit_idx);
> > -	writel(reg, clk->reg);
> > +	do_lpcg_workaround(clk_hw_get_rate(hw), clk->reg, reg);
> >
> >  	spin_unlock_irqrestore(&imx_lpcg_scu_lock, flags);  }
> > --
> > 2.28.0
> >