clk: sifive: Fix chip hang when booting linux directly

This patch adds the release_reset hook interface to __prci_wrpll_data. During clock enablement, the function (if present) will be called after PLL registers are configured. It aligns the logic to the driver in u-boot. When there's a previous bootloader stage, such as u-boot, it usually enables the gemgxlpll clock when trying to PXE/network boot. The kernel boots fine, but we should not depend on it being our previous stage, and the logic within:

a. We (linux) can get directly invoked by firmware (OpenSBI). b. U-boot doesn't necessarily have to initialize ethernet and enable the
clock (when not enabled in CONFIG).

When the kernel is the first to initialize gemgxlpll, it must also release the corresponding reset. Otherwise the chip will just hang during macb initialization, and even external JTAG debugger will lose control over the risc-v debug module. (Observed with my Sifive Unmatched Rev.B board)

The patch took the dt-bindings and logics directly from u-boot with some additional modifications:
- Use __prci_writel after __prci_readl to have barrier semantic. U-boot has the strong version of readl/writel, but linux has the relaxed ones.
- Use pd->reset.rcdev.ops to access the reset regs.
- Split reset bindings for FU540/FU740 and use them directly, instead of looking it up through reset-names.

Signed-off-by Bo Gan <[email protected]>

---
 drivers/clk/sifive/fu540-prci.h  | 16 ++++++++++++++++
 drivers/clk/sifive/fu740-prci.h  | 31 +++++++++++++++++++++++++++++++
 drivers/clk/sifive/sifive-prci.c | 23 +++++++++++++++++++++++
 drivers/clk/sifive/sifive-prci.h |  8 ++++++++
 4 files changed, 78 insertions(+)

drivers/clk/sifive/fu540-prci.h +16 -0

index e0173324f3c5..9d2ca18f47a4 100644
@@ -23,9 +23,24 @@
 #include <linux/module.h>
 
 #include <dt-bindings/clock/sifive-fu540-prci.h>
+#include <dt-bindings/reset/sifive-fu540-prci.h>
 
 #include "sifive-prci.h"
 
+/**
+ * sifive_fu540_prci_ethernet_release_reset() - Release ethernet reset
+ * @pd: struct __prci_data * for the PRCI containing the Ethernet CLK mux reg
+ *
+ */
+static void sifive_fu540_prci_ethernet_release_reset(struct __prci_data *pd)
+{
+	/* Release GEMGXL reset */
+	pd->reset.rcdev.ops->deassert(&pd->reset.rcdev, FU540_PRCI_RST_GEMGXL_N);
+
+	/* Procmon => core clock */
+	sifive_prci_set_procmoncfg(pd, PRCI_PROCMONCFG_CORE_CLOCK_MASK);
+}
+
 /* PRCI integration data for each WRPLL instance */
 
 static struct __prci_wrpll_data sifive_fu540_prci_corepll_data = {

@@ -43,6 +58,7 @@ static struct __prci_wrpll_data sifive_fu540_prci_ddrpll_data = {
 static struct __prci_wrpll_data sifive_fu540_prci_gemgxlpll_data = {
 	.cfg0_offs = PRCI_GEMGXLPLLCFG0_OFFSET,
 	.cfg1_offs = PRCI_GEMGXLPLLCFG1_OFFSET,
+	.release_reset = sifive_fu540_prci_ethernet_release_reset,
 };
 
 /* Linux clock framework integration */

drivers/clk/sifive/fu740-prci.h +31 -0

index f31cd30fc395..dd0f54277a99 100644
@@ -10,9 +10,38 @@
 #include <linux/module.h>
 
 #include <dt-bindings/clock/sifive-fu740-prci.h>
+#include <dt-bindings/reset/sifive-fu740-prci.h>
 
 #include "sifive-prci.h"
 
+/**
+ * sifive_fu740_prci_ethernet_release_reset() - Release ethernet reset
+ * @pd: struct __prci_data * for the PRCI containing the Ethernet CLK mux reg
+ *
+ */
+static void sifive_fu740_prci_ethernet_release_reset(struct __prci_data *pd)
+{
+	/* Release GEMGXL reset */
+	pd->reset.rcdev.ops->deassert(&pd->reset.rcdev, FU740_PRCI_RST_GEMGXL_N);
+
+	/* Procmon => core clock */
+	sifive_prci_set_procmoncfg(pd, PRCI_PROCMONCFG_CORE_CLOCK_MASK);
+
+	/* Release Chiplink reset */
+	pd->reset.rcdev.ops->deassert(&pd->reset.rcdev, FU740_PRCI_RST_CLTX_N);
+}
+
+/**
+ * sifive_fu740_prci_cltx_release_reset() - Release cltx reset
+ * @pd: struct __prci_data * for the PRCI containing the Ethernet CLK mux reg
+ *
+ */
+static void sifive_fu740_prci_cltx_release_reset(struct __prci_data *pd)
+{
+	/* Release CLTX reset */
+	pd->reset.rcdev.ops->deassert(&pd->reset.rcdev, FU740_PRCI_RST_CLTX_N);
+}
+
 /* PRCI integration data for each WRPLL instance */
 
 static struct __prci_wrpll_data sifive_fu740_prci_corepll_data = {

@@ -30,6 +59,7 @@ static struct __prci_wrpll_data sifive_fu740_prci_ddrpll_data = {
 static struct __prci_wrpll_data sifive_fu740_prci_gemgxlpll_data = {
 	.cfg0_offs = PRCI_GEMGXLPLLCFG0_OFFSET,
 	.cfg1_offs = PRCI_GEMGXLPLLCFG1_OFFSET,
+	.release_reset = sifive_fu740_prci_ethernet_release_reset,
 };
 
 static struct __prci_wrpll_data sifive_fu740_prci_dvfscorepll_data = {

@@ -49,6 +79,7 @@ static struct __prci_wrpll_data sifive_fu740_prci_hfpclkpll_data = {
 static struct __prci_wrpll_data sifive_fu740_prci_cltxpll_data = {
 	.cfg0_offs = PRCI_CLTXPLLCFG0_OFFSET,
 	.cfg1_offs = PRCI_CLTXPLLCFG1_OFFSET,
+	.release_reset = sifive_fu740_prci_cltx_release_reset,
 };
 
 /* Linux clock framework integration */

drivers/clk/sifive/sifive-prci.c +23 -0

index caba0400f8a2..ae8055a84466 100644
@@ -249,6 +249,9 @@ int sifive_prci_clock_enable(struct clk_hw *hw)
 	if (pwd->disable_bypass)
 		pwd->disable_bypass(pd);
 
+	if (pwd->release_reset)
+		pwd->release_reset(pd);
+
 	return 0;
 }
 

@@ -448,6 +451,26 @@ void sifive_prci_hfpclkpllsel_use_hfpclkpll(struct __prci_data *pd)
 	r = __prci_readl(pd, PRCI_HFPCLKPLLSEL_OFFSET);	/* barrier */
 }
 
+/*
+ * PROCMONCFG
+ */
+
+/**
+ * sifive_prci_set_procmoncfg() - set PROCMONCFG
+ * @pd: struct __prci_data * PRCI context
+ * @val: u32 value to write to PROCMONCFG register
+ *
+ * Set the PROCMONCFG register to @val
+ *
+ * Context: Any context.  Caller must prevent concurrent changes to the
+ *          PROCMONCFG_OFFSET register.
+ */
+void sifive_prci_set_procmoncfg(struct __prci_data *pd, u32 val)
+{
+	__prci_writel(val, PRCI_PROCMONCFG_OFFSET, pd);
+	__prci_readl(pd, PRCI_PROCMONCFG_OFFSET);	/* barrier */
+}
+
 /* PCIE AUX clock APIs for enable, disable. */
 int sifive_prci_pcie_aux_clock_is_enabled(struct clk_hw *hw)
 {

drivers/clk/sifive/sifive-prci.h +8 -0

index 91658a88af4e..825a0aef9fd5 100644
@@ -210,6 +210,9 @@
 
 /* PROCMONCFG */
 #define PRCI_PROCMONCFG_OFFSET			0xf0
+#define PRCI_PROCMONCFG_CORE_CLOCK_SHIFT	24
+#define PRCI_PROCMONCFG_CORE_CLOCK_MASK					\
+		(0x1 << PRCI_PROCMONCFG_CORE_CLOCK_SHIFT)
 
 /*
  * Private structures

@@ -235,6 +238,7 @@ struct __prci_data {
  * @disable_bypass: fn ptr to code to not bypass the WRPLL (or NULL)
  * @cfg0_offs: WRPLL CFG0 register offset (in bytes) from the PRCI base address
  * @cfg1_offs: WRPLL CFG1 register offset (in bytes) from the PRCI base address
+ * @release_reset: fn ptr to code to release clock reset
  *
  * @enable_bypass and @disable_bypass are used for WRPLL instances
  * that contain a separate external glitchless clock mux downstream

@@ -246,6 +250,7 @@ struct __prci_wrpll_data {
 	void (*disable_bypass)(struct __prci_data *pd);
 	u8 cfg0_offs;
 	u8 cfg1_offs;
+	void (*release_reset)(struct __prci_data *pd);
 };
 
 /**

@@ -290,6 +295,9 @@ void sifive_prci_corepllsel_use_corepll(struct __prci_data *pd);
 void sifive_prci_hfpclkpllsel_use_hfclk(struct __prci_data *pd);
 void sifive_prci_hfpclkpllsel_use_hfpclkpll(struct __prci_data *pd);
 
+/* PROCMONCFG */
+void sifive_prci_set_procmoncfg(struct __prci_data *pd, u32 val);
+
 /* Linux clock framework integration */
 long sifive_prci_wrpll_round_rate(struct clk_hw *hw, unsigned long rate,
 				  unsigned long *parent_rate);

Signature · 1 line

2.34.1

Quoted 16 lines · Bo Gan

This patch adds the release_reset hook interface to __prci_wrpll_data. During clock enablement, the function (if present) will be called after PLL registers are configured. It aligns the logic to the driver in u-boot. When there's a previous bootloader stage, such as u-boot, it usually enables the gemgxlpll clock when trying to PXE/network boot. The kernel boots fine, but we should not depend on it being our previous stage, and the logic within:

a. We (linux) can get directly invoked by firmware (OpenSBI). b. U-boot doesn't necessarily have to initialize ethernet and enable the
clock (when not enabled in CONFIG).

When the kernel is the first to initialize gemgxlpll, it must also release the corresponding reset. Otherwise the chip will just hang during macb initialization, and even external JTAG debugger will lose control over the risc-v debug module. (Observed with my Sifive Unmatched Rev.B board)

Quoted 7 lines

The patch took the dt-bindings and logics directly from u-boot with some additional modifications:
- Use __prci_writel after __prci_readl to have barrier semantic. U-boot has the strong version of readl/writel, but linux has the relaxed ones.
- Use pd->reset.rcdev.ops to access the reset regs.
- Split reset bindings for FU540/FU740 and use them directly, instead of looking it up through reset-names.

The macb driver already supports using a reset at boot time (see zynq and mpfs) if hooked up in the devicetree, why doesn't that work for you in this situation?

Thanks,
Conor.

Quoted 189 lines

Signed-off-by Bo Gan <[email protected]>

---
  drivers/clk/sifive/fu540-prci.h  | 16 ++++++++++++++++
  drivers/clk/sifive/fu740-prci.h  | 31 +++++++++++++++++++++++++++++++
  drivers/clk/sifive/sifive-prci.c | 23 +++++++++++++++++++++++
  drivers/clk/sifive/sifive-prci.h |  8 ++++++++
  4 files changed, 78 insertions(+)

drivers/clk/sifive/fu540-prci.h +16 -0

index e0173324f3c5..9d2ca18f47a4 100644
@@ -23,9 +23,24 @@
  #include <linux/module.h>
 
  #include <dt-bindings/clock/sifive-fu540-prci.h>
+#include <dt-bindings/reset/sifive-fu540-prci.h>
 
  #include "sifive-prci.h"
 
+/**
+ * sifive_fu540_prci_ethernet_release_reset() - Release ethernet reset
+ * @pd: struct __prci_data * for the PRCI containing the Ethernet CLK mux reg
+ *
+ */
+static void sifive_fu540_prci_ethernet_release_reset(struct __prci_data *pd)
+{
+	/* Release GEMGXL reset */
+	pd->reset.rcdev.ops->deassert(&pd->reset.rcdev, FU540_PRCI_RST_GEMGXL_N);
+
+	/* Procmon => core clock */
+	sifive_prci_set_procmoncfg(pd, PRCI_PROCMONCFG_CORE_CLOCK_MASK);
+}
+
  /* PRCI integration data for each WRPLL instance */
 
  static struct __prci_wrpll_data sifive_fu540_prci_corepll_data = {

@@ -43,6 +58,7 @@ static struct __prci_wrpll_data sifive_fu540_prci_ddrpll_data = {
  static struct __prci_wrpll_data sifive_fu540_prci_gemgxlpll_data = {
  	.cfg0_offs = PRCI_GEMGXLPLLCFG0_OFFSET,
  	.cfg1_offs = PRCI_GEMGXLPLLCFG1_OFFSET,
+	.release_reset = sifive_fu540_prci_ethernet_release_reset,
  };
 
  /* Linux clock framework integration */

drivers/clk/sifive/fu740-prci.h +31 -0

index f31cd30fc395..dd0f54277a99 100644
@@ -10,9 +10,38 @@
  #include <linux/module.h>
 
  #include <dt-bindings/clock/sifive-fu740-prci.h>
+#include <dt-bindings/reset/sifive-fu740-prci.h>
 
  #include "sifive-prci.h"
 
+/**
+ * sifive_fu740_prci_ethernet_release_reset() - Release ethernet reset
+ * @pd: struct __prci_data * for the PRCI containing the Ethernet CLK mux reg
+ *
+ */
+static void sifive_fu740_prci_ethernet_release_reset(struct __prci_data *pd)
+{
+	/* Release GEMGXL reset */
+	pd->reset.rcdev.ops->deassert(&pd->reset.rcdev, FU740_PRCI_RST_GEMGXL_N);
+
+	/* Procmon => core clock */
+	sifive_prci_set_procmoncfg(pd, PRCI_PROCMONCFG_CORE_CLOCK_MASK);
+
+	/* Release Chiplink reset */
+	pd->reset.rcdev.ops->deassert(&pd->reset.rcdev, FU740_PRCI_RST_CLTX_N);
+}
+
+/**
+ * sifive_fu740_prci_cltx_release_reset() - Release cltx reset
+ * @pd: struct __prci_data * for the PRCI containing the Ethernet CLK mux reg
+ *
+ */
+static void sifive_fu740_prci_cltx_release_reset(struct __prci_data *pd)
+{
+	/* Release CLTX reset */
+	pd->reset.rcdev.ops->deassert(&pd->reset.rcdev, FU740_PRCI_RST_CLTX_N);
+}
+
  /* PRCI integration data for each WRPLL instance */
 
  static struct __prci_wrpll_data sifive_fu740_prci_corepll_data = {

@@ -30,6 +59,7 @@ static struct __prci_wrpll_data sifive_fu740_prci_ddrpll_data = {
  static struct __prci_wrpll_data sifive_fu740_prci_gemgxlpll_data = {
  	.cfg0_offs = PRCI_GEMGXLPLLCFG0_OFFSET,
  	.cfg1_offs = PRCI_GEMGXLPLLCFG1_OFFSET,
+	.release_reset = sifive_fu740_prci_ethernet_release_reset,
  };
 
  static struct __prci_wrpll_data sifive_fu740_prci_dvfscorepll_data = {

@@ -49,6 +79,7 @@ static struct __prci_wrpll_data sifive_fu740_prci_hfpclkpll_data = {
  static struct __prci_wrpll_data sifive_fu740_prci_cltxpll_data = {
  	.cfg0_offs = PRCI_CLTXPLLCFG0_OFFSET,
  	.cfg1_offs = PRCI_CLTXPLLCFG1_OFFSET,
+	.release_reset = sifive_fu740_prci_cltx_release_reset,
  };
 
  /* Linux clock framework integration */

drivers/clk/sifive/sifive-prci.c +23 -0

index caba0400f8a2..ae8055a84466 100644
@@ -249,6 +249,9 @@ int sifive_prci_clock_enable(struct clk_hw *hw)
  	if (pwd->disable_bypass)
  		pwd->disable_bypass(pd);
 
+	if (pwd->release_reset)
+		pwd->release_reset(pd);
+
  	return 0;
  }
 

@@ -448,6 +451,26 @@ void sifive_prci_hfpclkpllsel_use_hfpclkpll(struct __prci_data *pd)
  	r = __prci_readl(pd, PRCI_HFPCLKPLLSEL_OFFSET);	/* barrier */
  }
 
+/*
+ * PROCMONCFG
+ */
+
+/**
+ * sifive_prci_set_procmoncfg() - set PROCMONCFG
+ * @pd: struct __prci_data * PRCI context
+ * @val: u32 value to write to PROCMONCFG register
+ *
+ * Set the PROCMONCFG register to @val
+ *
+ * Context: Any context.  Caller must prevent concurrent changes to the
+ *          PROCMONCFG_OFFSET register.
+ */
+void sifive_prci_set_procmoncfg(struct __prci_data *pd, u32 val)
+{
+	__prci_writel(val, PRCI_PROCMONCFG_OFFSET, pd);
+	__prci_readl(pd, PRCI_PROCMONCFG_OFFSET);	/* barrier */
+}
+
  /* PCIE AUX clock APIs for enable, disable. */
  int sifive_prci_pcie_aux_clock_is_enabled(struct clk_hw *hw)
  {

drivers/clk/sifive/sifive-prci.h +8 -0

index 91658a88af4e..825a0aef9fd5 100644
@@ -210,6 +210,9 @@
 
  /* PROCMONCFG */
  #define PRCI_PROCMONCFG_OFFSET			0xf0
+#define PRCI_PROCMONCFG_CORE_CLOCK_SHIFT	24
+#define PRCI_PROCMONCFG_CORE_CLOCK_MASK					\
+		(0x1 << PRCI_PROCMONCFG_CORE_CLOCK_SHIFT)
 
  /*
   * Private structures

@@ -235,6 +238,7 @@ struct __prci_data {
   * @disable_bypass: fn ptr to code to not bypass the WRPLL (or NULL)
   * @cfg0_offs: WRPLL CFG0 register offset (in bytes) from the PRCI base address
   * @cfg1_offs: WRPLL CFG1 register offset (in bytes) from the PRCI base address
+ * @release_reset: fn ptr to code to release clock reset
   *
   * @enable_bypass and @disable_bypass are used for WRPLL instances
   * that contain a separate external glitchless clock mux downstream

@@ -246,6 +250,7 @@ struct __prci_wrpll_data {
  	void (*disable_bypass)(struct __prci_data *pd);
  	u8 cfg0_offs;
  	u8 cfg1_offs;
+	void (*release_reset)(struct __prci_data *pd);
  };
 
  /**

@@ -290,6 +295,9 @@ void sifive_prci_corepllsel_use_corepll(struct __prci_data *pd);
  void sifive_prci_hfpclkpllsel_use_hfclk(struct __prci_data *pd);
  void sifive_prci_hfpclkpllsel_use_hfpclkpll(struct __prci_data *pd);
 
+/* PROCMONCFG */
+void sifive_prci_set_procmoncfg(struct __prci_data *pd, u32 val);
+
  /* Linux clock framework integration */
  long sifive_prci_wrpll_round_rate(struct clk_hw *hw, unsigned long rate,
  				  unsigned long *parent_rate);

Signature · 1 line

2.34.1