Skip to contentSkip to conversation

renesas: r8a779g0: Add support for boost mode

PATCH In review Geert Uytterhoeven 2 participants 6 messages Dec 8, 2022 – Jan 26, 2023

Subsystems (4)
lore
Share & tools
lore

Geert Uytterhoeven

author maintainer
#0 lore raw

cover renesas: r8a779g0: Add support for boost mode

Hi all,

This patch series adds support for running the Cortex-A76 CPU cores on R-Car V4H at 1.8 GHz (High Performance mode aka "boost mode").

The first patch adds boost support to the clock driver. The second patch contains an RFC improvement for the clock driver. The third patch describes the 1.8 GHz operating point in DT.

Thanks for your comments!

Geert Uytterhoeven (3):
clk: renesas: r8a779g0: Add custom clock for PLL2
clk: renesas: rcar-gen4: Add support for fractional multiplication
arm64: dts: renesas: r8a779g0: Add Cortex-A76 1.8 GHz opp

 arch/arm64/boot/dts/renesas/r8a779g0.dtsi |   6 +
 drivers/clk/renesas/r8a779g0-cpg-mssr.c   |  14 +-
 drivers/clk/renesas/rcar-gen4-cpg.c       | 195 ++++++++++++++++++++++
 drivers/clk/renesas/rcar-gen4-cpg.h       |   1 +
 4 files changed, 209 insertions(+), 7 deletions(-)
Signature · 12 lines
2.25.1

Gr{oetje,eeting}s,

						Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
							    -- Linus Torvalds
author maintainer
#1 lore raw

PATCH 1/3 clk: renesas: r8a779g0: Add custom clock for PLL2

Currently the PLLs are modeled as fixed factor clocks, based on initial settings. However, enabling CPU boost clock rates requires increasing the PLL clock rates.

Add a custom clock driver to model the PLL clocks on R-Car Gen4, and use it for PLL2 on R-Car V4H. This allows the Z clock (Cortex-A76 core clock) to request PLL rate changes, and enable boost mode for the High Performance mode. For now this is limited to integer multiplication modes.

Note that the definition for CPG_PLLxCR0_NI uses the value for R-Car V4H. On R-Car S4-8, the integer and fractional multiplication fields are one bit larger resp. smaller, but R-Car S4-8 does not support High Performance mode.

---
 drivers/clk/renesas/r8a779g0-cpg-mssr.c |  14 +--
 drivers/clk/renesas/rcar-gen4-cpg.c     | 156 ++++++++++++++++++++++++
 drivers/clk/renesas/rcar-gen4-cpg.h     |   1 +
 3 files changed, 164 insertions(+), 7 deletions(-)
drivers/clk/renesas/r8a779g0-cpg-mssr.c +7 -7
index 745ff9214fe8609d..48d9f34bbf37657a 100644
@@ -66,13 +66,13 @@ static const struct cpg_core_clk r8a779g0_core_clks[] __initconst = {
 	DEF_INPUT("extalr",	CLK_EXTALR),
 
 	/* Internal Core Clocks */
-	DEF_BASE(".main", CLK_MAIN,	CLK_TYPE_GEN4_MAIN, CLK_EXTAL),
-	DEF_BASE(".pll1", CLK_PLL1,	CLK_TYPE_GEN4_PLL1, CLK_MAIN),
-	DEF_BASE(".pll2", CLK_PLL2,	CLK_TYPE_GEN4_PLL2, CLK_MAIN),
-	DEF_BASE(".pll3", CLK_PLL3,	CLK_TYPE_GEN4_PLL3, CLK_MAIN),
-	DEF_BASE(".pll4", CLK_PLL4,	CLK_TYPE_GEN4_PLL4, CLK_MAIN),
-	DEF_BASE(".pll5", CLK_PLL5,	CLK_TYPE_GEN4_PLL5, CLK_MAIN),
-	DEF_BASE(".pll6", CLK_PLL6,	CLK_TYPE_GEN4_PLL6, CLK_MAIN),
+	DEF_BASE(".main", CLK_MAIN,	CLK_TYPE_GEN4_MAIN,     CLK_EXTAL),
+	DEF_BASE(".pll1", CLK_PLL1,	CLK_TYPE_GEN4_PLL1,     CLK_MAIN),
+	DEF_BASE(".pll2", CLK_PLL2,	CLK_TYPE_GEN4_PLL2_VAR, CLK_MAIN),
+	DEF_BASE(".pll3", CLK_PLL3,	CLK_TYPE_GEN4_PLL3,     CLK_MAIN),
+	DEF_BASE(".pll4", CLK_PLL4,	CLK_TYPE_GEN4_PLL4,     CLK_MAIN),
+	DEF_BASE(".pll5", CLK_PLL5,	CLK_TYPE_GEN4_PLL5,     CLK_MAIN),
+	DEF_BASE(".pll6", CLK_PLL6,	CLK_TYPE_GEN4_PLL6,     CLK_MAIN),
 
 	DEF_FIXED(".pll1_div2",	CLK_PLL1_DIV2,	CLK_PLL1,	2, 1),
 	DEF_FIXED(".pll2_div2",	CLK_PLL2_DIV2,	CLK_PLL2,	2, 1),
drivers/clk/renesas/rcar-gen4-cpg.c +156 -0

174 lines not on this page · raw message

drivers/clk/renesas/rcar-gen4-cpg.h +1 -0
index 0a0e3bdb3a664c10..006537e29e4eb10e 100644
@@ -13,6 +13,7 @@ enum rcar_gen4_clk_types {
 	CLK_TYPE_GEN4_MAIN = CLK_TYPE_CUSTOM,
 	CLK_TYPE_GEN4_PLL1,
 	CLK_TYPE_GEN4_PLL2,
+	CLK_TYPE_GEN4_PLL2_VAR,
 	CLK_TYPE_GEN4_PLL2X_3X,	/* r8a779a0 only */
 	CLK_TYPE_GEN4_PLL3,
 	CLK_TYPE_GEN4_PLL4,
Signature · 1 line
2.25.1
author maintainer
#2 lore raw

PATCH 2/3 clk: renesas: rcar-gen4: Add support for fractional multiplication

In reply to Geert Uytterhoeven

R-Car Gen4 PLLs support fractional multiplication, which can improve accuracy when configuring a specific frequency.

Add support for fractional multiplication to the custom clock driver for PLLs, which is currently used only for PLL2 on R-Car V4H.

---
I am not so sure it is worth supporting this.
On R-Car V4H, the following clock rates are seen for PLL2 and the Cortex-A76 CPU core clock, when using the Normal vs. the High-Performance mode:

					Multiplication Mode
					Integer         Frational
					----------	----------
    Normal (1.7 GHz):            PLL2   3399999864      3399999997
				 Z0     1699999932      1699999999
    High-Performance (1.8 GHz):  PLL2   3599999856      3599999997
				 Z0     1799999928      1799999999

The improvement is of a similar order of magnitude as the accuracy of the external crystal, hence insignificant...

With the current implementation, fractional multiplication does not have any impact on lower performance points, as those rely on changing the Z0 divider (which supports 32 steps only), instead of changing the PLL.

---
 drivers/clk/renesas/rcar-gen4-cpg.c | 69 ++++++++++++++++++++++-------
 1 file changed, 54 insertions(+), 15 deletions(-)
drivers/clk/renesas/rcar-gen4-cpg.c +54 -15
index c68d8b987054131b..54dc3aa82b499725 100644
@@ -57,10 +57,13 @@ static u32 cpg_mode __initdata;
 #define SSMODE_DITHER		BIT(1)	/* Frequency Dithering */
 #define SSMODE_CENTER		BIT(0)	/* Center (vs. Down) Spread Dithering */
 
+#define CPG_PLLxCR1_NF		GENMASK(24, 0)	/* Fractional mult. factor */
+
 /* PLL Clocks */
 struct cpg_pll_clk {
 	struct clk_hw hw;
 	void __iomem *pllcr0_reg;
+	void __iomem *pllcr1_reg;
 	void __iomem *pllecr_reg;
 	u32 pllecr_pllst_mask;
 };
@@ -71,17 +74,26 @@ static unsigned long cpg_pll_clk_recalc_rate(struct clk_hw *hw,
 					     unsigned long parent_rate)
 {
 	struct cpg_pll_clk *pll_clk = to_pll_clk(hw);
-	unsigned int mult;
-
-	mult = FIELD_GET(CPG_PLLxCR0_NI, readl(pll_clk->pllcr0_reg)) + 1;
+	u32 cr0 = readl(pll_clk->pllcr0_reg);
+	unsigned int ni, nf;
+	unsigned long rate;
+
+	ni = (FIELD_GET(CPG_PLLxCR0_NI, cr0) + 1) * 2;
+	rate = parent_rate * ni;
+	if (cr0 & CPG_PLLxCR0_SSMODE_FM) {
+		nf = FIELD_GET(CPG_PLLxCR1_NF, readl(pll_clk->pllcr1_reg));
+		rate += ((u64)parent_rate * nf) >> 24;
+	}
 
-	return parent_rate * mult * 2;
+	return rate;
 }
 
 static int cpg_pll_clk_determine_rate(struct clk_hw *hw,
 				      struct clk_rate_request *req)
 {
-	unsigned int min_mult, max_mult, mult;
+	struct cpg_pll_clk *pll_clk = to_pll_clk(hw);
+	unsigned int min_mult, max_mult, ni, nf;
+	u32 cr0 = readl(pll_clk->pllcr0_reg);
 	unsigned long prate;
 
 	prate = req->best_parent_rate * 2;
@@ -90,10 +102,23 @@ static int cpg_pll_clk_determine_rate(struct clk_hw *hw,
 	if (max_mult < min_mult)
 		return -EINVAL;
 
-	mult = DIV_ROUND_CLOSEST_ULL(req->rate, prate);
-	mult = clamp(mult, min_mult, max_mult);
+	if (cr0 & CPG_PLLxCR0_SSMODE_FM) {
+		ni = div64_ul(req->rate, prate);
+		if (ni < min_mult) {
+			ni = min_mult;
+			nf = 0;
+		} else {
+			ni = min(ni, max_mult);
+			nf = ((u64)(req->rate - prate * ni) << 24) /
+			     req->best_parent_rate;
+		}
+	} else {
+		ni = DIV_ROUND_CLOSEST_ULL(req->rate, prate);
+		ni = clamp(ni, min_mult, max_mult);
+		nf = 0;
+	}
+	req->rate = prate * ni + (((u64)req->best_parent_rate * nf) >> 24);
 
-	req->rate = prate * mult;
 	return 0;
 }
 
@@ -101,17 +126,31 @@ static int cpg_pll_clk_set_rate(struct clk_hw *hw, unsigned long rate,
 				unsigned long parent_rate)
 {
 	struct cpg_pll_clk *pll_clk = to_pll_clk(hw);
-	unsigned int mult;
+	unsigned long prate = parent_rate * 2;
+	u32 cr0 = readl(pll_clk->pllcr0_reg);
+	unsigned int ni, nf;
+	int error;
 	u32 val;
 
-	mult = DIV_ROUND_CLOSEST_ULL(rate, parent_rate * 2);
-	mult = clamp(mult, 1U, 256U);
+	if (cr0 & CPG_PLLxCR0_SSMODE_FM) {
+		ni = div64_ul(rate, prate);
+		if (ni < 1) {
+			ni = 1;
+			nf = 0;
+		} else {
+			ni = min(ni, 256U);
+			nf = ((u64)(rate - prate * ni) << 24) / parent_rate;
+		}
+	} else {
+		ni = DIV_ROUND_CLOSEST_ULL(rate, prate);
+		ni = clamp(ni, 1U, 256U);
+	}
 
 	if (readl(pll_clk->pllcr0_reg) & CPG_PLLxCR0_KICK)
 		return -EBUSY;
 
 	cpg_reg_modify(pll_clk->pllcr0_reg, CPG_PLLxCR0_NI,
-		       FIELD_PREP(CPG_PLLxCR0_NI, mult - 1));
+		       FIELD_PREP(CPG_PLLxCR0_NI, ni - 1));
 
 	/*
 	 * Set KICK bit in PLLxCR0 to update hardware setting and wait for
@@ -161,12 +200,12 @@ static struct clk * __init cpg_pll_clk_register(const char *name,
 
 	pll_clk->hw.init = &init;
 	pll_clk->pllcr0_reg = base + cr0_offset;
+	pll_clk->pllcr1_reg = base + cr1_offset;
 	pll_clk->pllecr_reg = base + CPG_PLLECR;
 	pll_clk->pllecr_pllst_mask = CPG_PLLECR_PLLST(index);
 
-	/* Disable Fractional Multiplication and Frequency Dithering */
-	writel(0, base + cr1_offset);
-	cpg_reg_modify(pll_clk->pllcr0_reg, CPG_PLLxCR0_SSMODE, 0);
+	/* Enable Fractional Multiplication */
+	cpg_reg_modify(pll_clk->pllcr0_reg, 0, CPG_PLLxCR0_SSMODE_FM);
 
 	clk = clk_register(NULL, &pll_clk->hw);
 	if (IS_ERR(clk))
Signature · 1 line
2.25.1
#3 lore raw

Re: [PATCH 2/3] clk: renesas: rcar-gen4: Add support for fractional multiplication

Quoted 12 lines · Geert Uytterhoeven

R-Car Gen4 PLLs support fractional multiplication, which can improve accuracy when configuring a specific frequency.

Add support for fractional multiplication to the custom clock driver for PLLs, which is currently used only for PLL2 on R-Car V4H.

drivers/clk/renesas/rcar-gen4-cpg.c static int cpg_pll_clk_set_rate(struct clk_hw *hw, unsigned long rate, · line 132

Quoted 8 lines
drivers/clk/renesas/rcar-gen4-cpg.c +4 -1
@@ -101,17 +126,31 @@ static int cpg_pll_clk_set_rate(struct clk_hw *hw, unsigned long rate,
                                unsigned long parent_rate)
 {
        struct cpg_pll_clk *pll_clk = to_pll_clk(hw);
-       unsigned int mult;
+       unsigned long prate = parent_rate * 2;
+       u32 cr0 = readl(pll_clk->pllcr0_reg);
+       unsigned int ni, nf;
+       int error;

"error" is unused, and should be dropped.

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

#4 lore raw

Re: [PATCH 2/3] clk: renesas: rcar-gen4: Add support for fractional multiplication

In reply to Geert Uytterhoeven
Quoted 21 lines · Geert Uytterhoeven

R-Car Gen4 PLLs support fractional multiplication, which can improve accuracy when configuring a specific frequency.

Add support for fractional multiplication to the custom clock driver for PLLs, which is currently used only for PLL2 on R-Car V4H.

---
I am not so sure it is worth supporting this.
On R-Car V4H, the following clock rates are seen for PLL2 and the Cortex-A76 CPU core clock, when using the Normal vs. the High-Performance mode:

                                        Multiplication Mode
                                        Integer         Frational
                                        ----------      ----------
    Normal (1.7 GHz):            PLL2   3399999864      3399999997
                                 Z0     1699999932      1699999999
    High-Performance (1.8 GHz):  PLL2   3599999856      3599999997
                                 Z0     1799999928      1799999999
Quoted 2 lines

The improvement is of a similar order of magnitude as the accuracy of the external crystal, hence insignificant...

Hence I'm ignoring this, unless someone has a good reason to push through...

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

author maintainer
#5 lore raw

PATCH 3/3 arm64: dts: renesas: r8a779g0: Add Cortex-A76 1.8 GHz opp

In reply to Geert Uytterhoeven

Add an operating point for running the Cortex-A76 CPU cores on R-Car V4H at 1.8 GHz (High Performance mode).

Based on a patch in the BSP by Tho Vu.

---
Tested on the Renesas White-Hawk development board by using the CPUfreq userspace governor, enabling boost mode, writing the desired CPU clock rate to the CPUfreq policy's "scaling_setspeed" file in sysfs, verifying the clock rate of the Z0φ clock in debugfs, and running the dhrystones benchmark on the various CPU cores.

$ echo 1 > /sys/devices/system/cpu/cpufreq/boost
$ echo 1800000 > /sys/devices/system/cpu/cpufreq/policy0/scaling_setspeed
---
 arch/arm64/boot/dts/renesas/r8a779g0.dtsi | 6 ++++++
 1 file changed, 6 insertions(+)
arch/arm64/boot/dts/renesas/r8a779g0.dtsi +6 -0
index 83d1666a2ea16de2..7a87a5dc1b6ad219 100644
@@ -39,6 +39,12 @@ opp-1700000000 {
 			clock-latency-ns = <500000>;
 			opp-suspend;
 		};
+		opp-1800000000 {
+			opp-hz = /bits/ 64 <1800000000>;
+			opp-microvolt = <880000>;
+			clock-latency-ns = <500000>;
+			turbo-mode;
+		};
 	};
 
 	cpus {
Signature · 1 line
2.25.1