[PATCH v5 03/11] soc: renesas: rcar-sysc: Add DT support for SYSC PM domains

[PATCH v5 03/11] soc: renesas: rcar-sysc: Add DT support for SYSC PM domains

[Geert Uytterhoeven]

Populate the SYSC PM domains from DT, based on the presence of a device
node for the System Controller. The actual power area hiearchy, and
features of specific areas are obtained from tables in the C code.

The SYSCIER and SYSCIMR register values are derived from the power areas
present, which will help to get rid of the hardcoded values in R-Car H1
and R-Car Gen2 platform code later.

Initialization is done from an early_initcall(), to make sure the PM
Domains are initialized before secondary CPU bringup.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
---
v5:
  - Mask SYSC interrupts sources before enabling them (doesn't matter
    much as they're disabled at the GIC level anyway),
  - Re-add explicit "always-on" power area instead of aliasing the SoC's
    Clock Domain,
  - Merge the two initialization phases again,

v4:
  - Make sure not to clear reserved SYSCIMR bits that were set before,
  - Make the always-on power area implicit and always present, and an
    alias of the existing SoC's Clock Domain. This makes the number of
    power areas a compile-time constant, and allows to drop PD_ALWAYS_ON
    and some checks.
  - Split initialization in two phases,
  - Document that ARM cores are controlled by PSCI on R-Car Gen3
    (although the underlying CPG/APMU hardware is the same as on Gen2),
  - Minor improvements (double evaluation, unused parameter, debug
    message consolidation),

v3:
  - Drop check for CONFIG_PM_GENERIC_DOMAINS, which is now always
    enabled on R-Car SoCs,
  - Create PM Domains from hierarchy in C data instead of DT,
  - Initialize SYSCIER early, as SYSC needs the interrupt sources to be
    enabled to control power,
  - Mask all SYSC interrupt sources for the CPU,
  - Add support for an "always-on" domain,
  - Use early_initcall() instead of core_initcall(),
  - Do not power up CPU power areas during initialization, as this is
    handled later (directly or indirectly) by the SMP code,

v2:
  - Add missing definitions for SYSC_PWR_CA15_CPU and SYSC_PWR_CA7_CPU,
  - Add R-Car H3 (r8a7795) support,
  - Drop tests for CONFIG_ARCH_SHMOBILE_LEGACY,
  - Add missing break statements in rcar_sysc_pwr_on_off(),
  - Add missing calls to of_node_put() in error paths,
  - Fix build if CONFIG_PM=n,
  - Update compatible values,
  - Update copyright.
---
 drivers/soc/renesas/rcar-sysc.c | 206 +++++++++++++++++++++++++++++++++++++++-
 drivers/soc/renesas/rcar-sysc.h |  53 +++++++++++
 2 files changed, 258 insertions(+), 1 deletion(-)
 create mode 100644 drivers/soc/renesas/rcar-sysc.h

diff --git a/drivers/soc/renesas/rcar-sysc.c b/drivers/soc/renesas/rcar-sysc.c
index 9ba5fd15c53bf9b9..674a46ce46aac04d 100644
--- a/drivers/soc/renesas/rcar-sysc.c
+++ b/drivers/soc/renesas/rcar-sysc.c
@@ -2,6 +2,7 @@
  * R-Car SYSC Power management support
  *
  * Copyright (C) 2014  Magnus Damm
+ * Copyright (C) 2015-2016 Glider bvba
  *
  * This file is subject to the terms and conditions of the GNU General Public
  * License.  See the file "COPYING" in the main directory of this archive
@@ -11,10 +12,15 @@
 #include <linux/delay.h>
 #include <linux/err.h>
 #include <linux/mm.h>
+#include <linux/of_address.h>
+#include <linux/pm_domain.h>
+#include <linux/slab.h>
 #include <linux/spinlock.h>
 #include <linux/io.h>
 #include <linux/soc/renesas/rcar-sysc.h>
 
+#include "rcar-sysc.h"
+
 /* SYSC Common */
 #define SYSCSR			0x00	/* SYSC Status Register */
 #define SYSCISR			0x04	/* Interrupt Status Register */
@@ -29,7 +35,8 @@
 /*
  * Power Control Register Offsets inside the register block for each domain
  * Note: The "CR" registers for ARM cores exist on H1 only
- *       Use WFI to power off, CPG/APMU to resume ARM cores on R-Car Gen2
+ *	 Use WFI to power off, CPG/APMU to resume ARM cores on R-Car Gen2
+ *	 Use PSCI on R-Car Gen3
  */
 #define PWRSR_OFFS		0x00	/* Power Status Register */
 #define PWROFFCR_OFFS		0x04	/* Power Shutoff Control Register */
@@ -48,6 +55,8 @@
 #define SYSCISR_RETRIES		1000
 #define SYSCISR_DELAY_US	1
 
+#define RCAR_PD_ALWAYS_ON	32	/* Always-on power area */
+
 static void __iomem *rcar_sysc_base;
 static DEFINE_SPINLOCK(rcar_sysc_lock); /* SMP CPUs + I/O devices */
 
@@ -162,3 +171,198 @@ void __iomem *rcar_sysc_init(phys_addr_t base)
 
 	return rcar_sysc_base;
 }
+
+struct rcar_sysc_pd {
+	struct generic_pm_domain genpd;
+	struct rcar_sysc_ch ch;
+	unsigned int flags;
+	char name[0];
+};
+
+static inline struct rcar_sysc_pd *to_rcar_pd(struct generic_pm_domain *d)
+{
+	return container_of(d, struct rcar_sysc_pd, genpd);
+}
+
+static bool rcar_sysc_active_wakeup(struct device *dev)
+{
+	return true;
+}
+
+static int rcar_sysc_pd_power_off(struct generic_pm_domain *genpd)
+{
+	struct rcar_sysc_pd *pd = to_rcar_pd(genpd);
+
+	pr_debug("%s: %s\n", __func__, genpd->name);
+
+	if (pd->flags & PD_NO_CR) {
+		pr_debug("%s: Cannot control %s\n", __func__, genpd->name);
+		return -EBUSY;
+	}
+
+	if (pd->flags & PD_BUSY) {
+		pr_debug("%s: %s busy\n", __func__, genpd->name);
+		return -EBUSY;
+	}
+
+	return rcar_sysc_power_down(&pd->ch);
+}
+
+static int rcar_sysc_pd_power_on(struct generic_pm_domain *genpd)
+{
+	struct rcar_sysc_pd *pd = to_rcar_pd(genpd);
+
+	pr_debug("%s: %s\n", __func__, genpd->name);
+
+	if (pd->flags & PD_NO_CR) {
+		pr_debug("%s: Cannot control %s\n", __func__, genpd->name);
+		return 0;
+	}
+
+	return rcar_sysc_power_up(&pd->ch);
+}
+
+static void __init rcar_sysc_pd_setup(struct rcar_sysc_pd *pd)
+{
+	struct generic_pm_domain *genpd = &pd->genpd;
+	const char *name = pd->genpd.name;
+	struct dev_power_governor *gov = &simple_qos_governor;
+
+	if (pd->flags & PD_CPU) {
+		/*
+		 * This domain contains a CPU core and therefore it should
+		 * only be turned off if the CPU is not in use.
+		 */
+		pr_debug("PM domain %s contains %s\n", name, "CPU");
+		pd->flags |= PD_BUSY;
+		gov = &pm_domain_always_on_gov;
+	} else if (pd->flags & PD_SCU) {
+		/*
+		 * This domain contains an SCU and cache-controller, and
+		 * therefore it should only be turned off if the CPU cores are
+		 * not in use.
+		 */
+		pr_debug("PM domain %s contains %s\n", name, "SCU");
+		pd->flags |= PD_BUSY;
+		gov = &pm_domain_always_on_gov;
+	} else if (pd->flags & PD_NO_CR) {
+		/*
+		 * This domain cannot be turned off.
+		 */
+		pd->flags |= PD_BUSY;
+		gov = &pm_domain_always_on_gov;
+	}
+
+	pm_genpd_init(genpd, gov, false);
+	genpd->dev_ops.active_wakeup = rcar_sysc_active_wakeup;
+	genpd->power_off = rcar_sysc_pd_power_off;
+	genpd->power_on = rcar_sysc_pd_power_on;
+
+	if (pd->flags & (PD_CPU | PD_NO_CR)) {
+		/* Skip CPUs (handled by SMP code) and areas without control */
+		pr_debug("%s: Not touching %s\n", __func__, genpd->name);
+		return;
+	}
+
+	if (!rcar_sysc_power_is_off(&pd->ch)) {
+		pr_debug("%s: %s is already powered\n", __func__, genpd->name);
+		return;
+	}
+
+	rcar_sysc_power_up(&pd->ch);
+}
+
+static const struct of_device_id rcar_sysc_matches[] = {
+	{ /* sentinel */ }
+};
+
+struct rcar_pm_domains {
+	struct genpd_onecell_data onecell_data;
+	struct generic_pm_domain *domains[RCAR_PD_ALWAYS_ON + 1];
+};
+
+static int __init rcar_sysc_pd_init(void)
+{
+	const struct rcar_sysc_info *info;
+	const struct of_device_id *match;
+	struct rcar_pm_domains *domains;
+	struct device_node *np;
+	u32 syscier, syscimr;
+	void __iomem *base;
+	unsigned int i;
+	int error;
+
+	np = of_find_matching_node_and_match(NULL, rcar_sysc_matches, &match);
+	if (!np)
+		return -ENODEV;
+
+	info = match->data;
+
+	base = of_iomap(np, 0);
+	if (!base) {
+		pr_warn("%s: Cannot map regs\n", np->full_name);
+		error = -ENOMEM;
+		goto out_put;
+	}
+
+	rcar_sysc_base = base;
+
+	domains = kzalloc(sizeof(*domains), GFP_KERNEL);
+	if (!domains) {
+		error = -ENOMEM;
+		goto out_put;
+	}
+
+	domains->onecell_data.domains = domains->domains;
+	domains->onecell_data.num_domains = ARRAY_SIZE(domains->domains);
+
+	for (i = 0, syscier = 0; i < info->num_areas; i++)
+		syscier |= BIT(info->areas[i].isr_bit);
+
+	/*
+	 * Mask all interrupt sources to prevent the CPU from receiving them.
+	 * Make sure not to clear reserved bits that were set before.
+	 */
+	syscimr = ioread32(base + SYSCIMR);
+	syscimr |= syscier;
+	pr_debug("%s: syscimr = 0x%08x\n", np->full_name, syscimr);
+	iowrite32(syscimr, base + SYSCIMR);
+
+	/*
+	 * SYSC needs all interrupt sources enabled to control power.
+	 */
+	pr_debug("%s: syscier = 0x%08x\n", np->full_name, syscier);
+	iowrite32(syscier, base + SYSCIER);
+
+	for (i = 0; i < info->num_areas; i++) {
+		const struct rcar_sysc_area *area = &info->areas[i];
+		struct rcar_sysc_pd *pd;
+
+		pd = kzalloc(sizeof(*pd) + strlen(area->name) + 1, GFP_KERNEL);
+		if (!pd) {
+			error = -ENOMEM;
+			goto out_put;
+		}
+
+		strcpy(pd->name, area->name);
+		pd->genpd.name = pd->name;
+		pd->ch.chan_offs = area->chan_offs;
+		pd->ch.chan_bit = area->chan_bit;
+		pd->ch.isr_bit = area->isr_bit;
+		pd->flags = area->flags;
+
+		rcar_sysc_pd_setup(pd);
+		if (area->parent >= 0)
+			pm_genpd_add_subdomain(domains->domains[area->parent],
+					       &pd->genpd);
+
+		domains->domains[area->isr_bit] = &pd->genpd;
+	}
+
+	of_genpd_add_provider_onecell(np, &domains->onecell_data);
+
+out_put:
+	of_node_put(np);
+	return error;
+}
+early_initcall(rcar_sysc_pd_init);
diff --git a/drivers/soc/renesas/rcar-sysc.h b/drivers/soc/renesas/rcar-sysc.h
new file mode 100644
index 0000000000000000..7bb48b4f7334f960
--- /dev/null
+++ b/drivers/soc/renesas/rcar-sysc.h
@@ -0,0 +1,53 @@
+/*
+ * Renesas R-Car System Controller
+ *
+ * Copyright (C) 2016 Glider bvba
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ */
+#ifndef __SOC_RENESAS_RCAR_SYSC_H__
+#define __SOC_RENESAS_RCAR_SYSC_H__
+
+#include <linux/types.h>
+
+
+/*
+ * Power Domain flags
+ */
+#define PD_CPU		BIT(0)	/* Area contains main CPU core */
+#define PD_SCU		BIT(1)	/* Area contains SCU and L2 cache */
+#define PD_NO_CR	BIT(2)	/* Area lacks PWR{ON,OFF}CR registers */
+
+#define PD_BUSY		BIT(3)	/* Busy, for internal use only */
+
+#define PD_CPU_CR	PD_CPU		  /* CPU area has CR (R-Car H1) */
+#define PD_CPU_NOCR	PD_CPU | PD_NO_CR /* CPU area lacks CR (R-Car Gen2/3) */
+#define PD_ALWAYS_ON	PD_NO_CR	  /* Always-on area */
+
+
+/*
+ * Description of a Power Area
+ */
+
+struct rcar_sysc_area {
+	const char *name;
+	u16 chan_offs;		/* Offset of PWRSR register for this area */
+	u8 chan_bit;		/* Bit in PWR* (except for PWRUP in PWRSR) */
+	u8 isr_bit;		/* Bit in SYSCI*R */
+	int parent;		/* -1 if none */
+	unsigned int flags;	/* See PD_* */
+};
+
+
+/*
+ * SoC-specific Power Area Description
+ */
+
+struct rcar_sysc_info {
+	const struct rcar_sysc_area *areas;
+	unsigned int num_areas;
+};
+
+#endif /* __SOC_RENESAS_RCAR_SYSC_H__ */
-- 
1.9.1

[PATCH v5 09/11] soc: renesas: rcar-sysc: Add support for R-Car M2-N power areas

[Geert Uytterhoeven]

R-Car M2-N is identical to R-Car M2-W w.r.t. power domains, so reuse the
definitions from the latter.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
---
v5:
  - Add Reviewed-by,
  - Reference r8a7791_sysc_info directly for R-Car M2-N in rcar-sysc.c,

v4:
  - No changes,

v3:
  - New (converted from DT to C).
---
 drivers/soc/renesas/Makefile    | 3 ++-
 drivers/soc/renesas/rcar-sysc.c | 4 ++++
 2 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/drivers/soc/renesas/Makefile b/drivers/soc/renesas/Makefile
index 96463c05ee594335..c6c4ce7ef8a145ea 100644
--- a/drivers/soc/renesas/Makefile
+++ b/drivers/soc/renesas/Makefile
@@ -1,5 +1,6 @@
 obj-$(CONFIG_ARCH_R8A7779)	+= rcar-sysc.o r8a7779-sysc.o
 obj-$(CONFIG_ARCH_R8A7790)	+= rcar-sysc.o r8a7790-sysc.o
 obj-$(CONFIG_ARCH_R8A7791)	+= rcar-sysc.o r8a7791-sysc.o
-obj-$(CONFIG_ARCH_R8A7793)	+= rcar-sysc.o
+# R-Car M2-N is identical to R-Car M2-W w.r.t. power domains.
+obj-$(CONFIG_ARCH_R8A7793)	+= rcar-sysc.o r8a7791-sysc.o
 obj-$(CONFIG_ARCH_R8A7794)	+= rcar-sysc.o
diff --git a/drivers/soc/renesas/rcar-sysc.c b/drivers/soc/renesas/rcar-sysc.c
index b90bdd8d7269ffbe..d472d8b3fa591675 100644
--- a/drivers/soc/renesas/rcar-sysc.c
+++ b/drivers/soc/renesas/rcar-sysc.c
@@ -297,6 +297,10 @@ static const struct of_device_id rcar_sysc_matches[] = {
 #ifdef CONFIG_ARCH_R8A7791
 	{ .compatible = "renesas,r8a7791-sysc", .data = &r8a7791_sysc_info },
 #endif
+#ifdef CONFIG_ARCH_R8A7793
+	/* R-Car M2-N is identical to R-Car M2-W w.r.t. power domains. */
+	{ .compatible = "renesas,r8a7793-sysc", .data = &r8a7791_sysc_info },
+#endif
 	{ /* sentinel */ }
 };
 
-- 
1.9.1

[PATCH v5 05/11] soc: renesas: rcar-sysc: Enable Clock Domain for I/O devices

[Laurent Pinchart]

Hi Geert,

Thank you for the patch.

On Wednesday 13 Apr 2016 16:44:26 Geert Uytterhoeven wrote:
> On R-Car H3, some power areas (e.g. A3VP) contain I/O devices, which are
> also part of the CPG/MSSR Clock Domain.
> On all R-Car SoCs, devices in the "always-on" PM Domain are part of the
> Clock Domain served by the CPG/MSSR or CPG/MSTP driver.
> 
> Hook up the CPG/MSTP or CPG/MSSR Clock Domain attach/detach callbacks to
> enable power management using module clocks. Which callback to hook up
> depends on the presence of device nodes compatible with
> "renesas,cpg-mstp-clocks". This clears the path for a future migration
> from the CPG/MSTP to the CPG/MSSR driver on R-Car H1 and
> Gen2.
> 
> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>

Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>

> ---
> v5:
>   - Revert v4 changes,
>   - Use either the cpg_mssr_*() or cpg_mstp_*() callbacks,
>   - Drop dependency on r8a7795, as this is used for the "always-on" PM
>     Domain on R-Car H1 and Gen2, too,
> 
> v4:
>   - Remove the explicit dependency on the CPG/MSSR driver by forwarding
>     the attach/detach callbacks to the parent PM Domain.
>     If deemed reusable, rcar_sysc_{at,de}tach_dev() can be moved to
>     common genpd code later.
> 
> v3:
>   - Hook up the CPG/MSSR Clock Domain attach/detach callbacks instead of
>     using our own copies,
> 
> v2:
>   - New.
> ---
>  drivers/soc/renesas/rcar-sysc.c | 18 ++++++++++++++++++
>  1 file changed, 18 insertions(+)
> 
> diff --git a/drivers/soc/renesas/rcar-sysc.c
> b/drivers/soc/renesas/rcar-sysc.c index a333ef4152428440..be83636ef2647947
> 100644
> --- a/drivers/soc/renesas/rcar-sysc.c
> +++ b/drivers/soc/renesas/rcar-sysc.c
> @@ -9,6 +9,7 @@
>   * for more details.
>   */
> 
> +#include <linux/clk/renesas.h>
>  #include <linux/delay.h>
>  #include <linux/err.h>
>  #include <linux/mm.h>
> @@ -222,6 +223,8 @@ static int rcar_sysc_pd_power_on(struct
> generic_pm_domain *genpd) return rcar_sysc_power_up(&pd->ch);
>  }
> 
> +static bool has_cpg_mstp;
> +
>  static void __init rcar_sysc_pd_setup(struct rcar_sysc_pd *pd)
>  {
>  	struct generic_pm_domain *genpd = &pd->genpd;
> @@ -253,6 +256,18 @@ static void __init rcar_sysc_pd_setup(struct
> rcar_sysc_pd *pd) gov = &pm_domain_always_on_gov;
>  	}
> 
> +	if (!(pd->flags & (PD_CPU | PD_SCU))) {
> +		/* Enable Clock Domain for I/O devices */
> +		genpd->flags = GENPD_FLAG_PM_CLK;
> +		if (has_cpg_mstp) {
> +			genpd->attach_dev = cpg_mstp_attach_dev;
> +			genpd->detach_dev = cpg_mstp_detach_dev;
> +		} else {
> +			genpd->attach_dev = cpg_mssr_attach_dev;
> +			genpd->detach_dev = cpg_mssr_detach_dev;
> +		}
> +	}
> +
>  	pm_genpd_init(genpd, gov, false);
>  	genpd->dev_ops.active_wakeup = rcar_sysc_active_wakeup;
>  	genpd->power_off = rcar_sysc_pd_power_off;
> @@ -298,6 +313,9 @@ static int __init rcar_sysc_pd_init(void)
> 
>  	info = match->data;
> 
> +	has_cpg_mstp = of_find_compatible_node(NULL, NULL,
> +					       "renesas,cpg-mstp-clocks");
> +
>  	base = of_iomap(np, 0);
>  	if (!base) {
>  		pr_warn("%s: Cannot map regs\n", np->full_name);

-- 
Regards,

Laurent Pinchart

[PATCH v5 03/11] soc: renesas: rcar-sysc: Add DT support for SYSC PM domains

[Ulf Hansson]

[...]

> +
> +static bool rcar_sysc_active_wakeup(struct device *dev)
> +{
> +       return true;

I am interested to know why this is always returning true. Perhaps you
can elaborate a bit on that?

> +}
> +
> +static int rcar_sysc_pd_power_off(struct generic_pm_domain *genpd)
> +{
> +       struct rcar_sysc_pd *pd = to_rcar_pd(genpd);
> +
> +       pr_debug("%s: %s\n", __func__, genpd->name);
> +
> +       if (pd->flags & PD_NO_CR) {
> +               pr_debug("%s: Cannot control %s\n", __func__, genpd->name);
> +               return -EBUSY;
> +       }
> +
> +       if (pd->flags & PD_BUSY) {
> +               pr_debug("%s: %s busy\n", __func__, genpd->name);
> +               return -EBUSY;
> +       }
> +
> +       return rcar_sysc_power_down(&pd->ch);
> +}
> +
> +static int rcar_sysc_pd_power_on(struct generic_pm_domain *genpd)
> +{
> +       struct rcar_sysc_pd *pd = to_rcar_pd(genpd);
> +
> +       pr_debug("%s: %s\n", __func__, genpd->name);
> +
> +       if (pd->flags & PD_NO_CR) {
> +               pr_debug("%s: Cannot control %s\n", __func__, genpd->name);
> +               return 0;
> +       }
> +
> +       return rcar_sysc_power_up(&pd->ch);
> +}
> +
> +static void __init rcar_sysc_pd_setup(struct rcar_sysc_pd *pd)
> +{
> +       struct generic_pm_domain *genpd = &pd->genpd;
> +       const char *name = pd->genpd.name;
> +       struct dev_power_governor *gov = &simple_qos_governor;
> +
> +       if (pd->flags & PD_CPU) {
> +               /*
> +                * This domain contains a CPU core and therefore it should
> +                * only be turned off if the CPU is not in use.
> +                */
> +               pr_debug("PM domain %s contains %s\n", name, "CPU");
> +               pd->flags |= PD_BUSY;
> +               gov = &pm_domain_always_on_gov;
> +       } else if (pd->flags & PD_SCU) {
> +               /*
> +                * This domain contains an SCU and cache-controller, and
> +                * therefore it should only be turned off if the CPU cores are
> +                * not in use.
> +                */
> +               pr_debug("PM domain %s contains %s\n", name, "SCU");
> +               pd->flags |= PD_BUSY;
> +               gov = &pm_domain_always_on_gov;
> +       } else if (pd->flags & PD_NO_CR) {
> +               /*
> +                * This domain cannot be turned off.
> +                */
> +               pd->flags |= PD_BUSY;
> +               gov = &pm_domain_always_on_gov;
> +       }
> +
> +       pm_genpd_init(genpd, gov, false);

This seems weird. I don't think it correct to initialize genpd and
then continue with the below changes (assigning callbacks and power up
the domain).

I would recommend doing this in the reverse order.

> +       genpd->dev_ops.active_wakeup = rcar_sysc_active_wakeup;
> +       genpd->power_off = rcar_sysc_pd_power_off;
> +       genpd->power_on = rcar_sysc_pd_power_on;
> +
> +       if (pd->flags & (PD_CPU | PD_NO_CR)) {
> +               /* Skip CPUs (handled by SMP code) and areas without control */
> +               pr_debug("%s: Not touching %s\n", __func__, genpd->name);
> +               return;
> +       }
> +
> +       if (!rcar_sysc_power_is_off(&pd->ch)) {
> +               pr_debug("%s: %s is already powered\n", __func__, genpd->name);
> +               return;
> +       }
> +
> +       rcar_sysc_power_up(&pd->ch);
> +}

Kind regards
Uffe

[PATCH v5 03/11] soc: renesas: rcar-sysc: Add DT support for SYSC PM domains

[Geert Uytterhoeven]

Hi Ulf,

On Mon, Apr 18, 2016 at 2:21 PM, Ulf Hansson <ulf.hansson@linaro.org> wrote:
> [...]
>
>> +
>> +static bool rcar_sysc_active_wakeup(struct device *dev)
>> +{
>> +       return true;
>
> I am interested to know why this is always returning true. Perhaps you
> can elaborate a bit on that?

Too many copying from old shmobile PM Domain code?
Honestly, I don't know...

Perhaps Rafael still remembers the original rationale, as git history for
commit e3e0109138376bb2 ("ARM / shmobile: Support for I/O power domains for
SH7372 (v9)") doesn't have it.

Google did find: https://lkml.org/lkml/2011/6/30/471

Do we still need this at all? I.e. aren't PM Domains containing wake-up
devices kept powered automatically during system suspend?

>> +static void __init rcar_sysc_pd_setup(struct rcar_sysc_pd *pd)
>> +{
>> +       struct generic_pm_domain *genpd = &pd->genpd;
>> +       const char *name = pd->genpd.name;
>> +       struct dev_power_governor *gov = &simple_qos_governor;
>> +
>> +       if (pd->flags & PD_CPU) {
>> +               /*
>> +                * This domain contains a CPU core and therefore it should
>> +                * only be turned off if the CPU is not in use.
>> +                */
>> +               pr_debug("PM domain %s contains %s\n", name, "CPU");
>> +               pd->flags |= PD_BUSY;
>> +               gov = &pm_domain_always_on_gov;
>> +       } else if (pd->flags & PD_SCU) {
>> +               /*
>> +                * This domain contains an SCU and cache-controller, and
>> +                * therefore it should only be turned off if the CPU cores are
>> +                * not in use.
>> +                */
>> +               pr_debug("PM domain %s contains %s\n", name, "SCU");
>> +               pd->flags |= PD_BUSY;
>> +               gov = &pm_domain_always_on_gov;
>> +       } else if (pd->flags & PD_NO_CR) {
>> +               /*
>> +                * This domain cannot be turned off.
>> +                */
>> +               pd->flags |= PD_BUSY;
>> +               gov = &pm_domain_always_on_gov;
>> +       }
>> +
>> +       pm_genpd_init(genpd, gov, false);
>
> This seems weird. I don't think it correct to initialize genpd and
> then continue with the below changes (assigning callbacks and power up
> the domain).

I'm quite sure I wondered the same (for pm-rmobile), but discovered that
pm_genpd_init() overwrote something. Unfortunately I can't find the commit
that changed that...

> I would recommend doing this in the reverse order.
>
>> +       genpd->dev_ops.active_wakeup = rcar_sysc_active_wakeup;
>> +       genpd->power_off = rcar_sysc_pd_power_off;
>> +       genpd->power_on = rcar_sysc_pd_power_on;

OK, will give that a try...

Thanks for your comments!

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert at linux-m68k.org

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

[PATCH v5 03/11] soc: renesas: rcar-sysc: Add DT support for SYSC PM domains

[Geert Uytterhoeven]

Hi Ulf,

On Mon, Apr 18, 2016 at 2:59 PM, Geert Uytterhoeven
<geert@linux-m68k.org> wrote:
> On Mon, Apr 18, 2016 at 2:21 PM, Ulf Hansson <ulf.hansson@linaro.org> wrote:
>> [...]
>>
>>> +
>>> +static bool rcar_sysc_active_wakeup(struct device *dev)
>>> +{
>>> +       return true;
>>
>> I am interested to know why this is always returning true. Perhaps you
>> can elaborate a bit on that?
>
> Too many copying from old shmobile PM Domain code?
> Honestly, I don't know...
>
> Perhaps Rafael still remembers the original rationale, as git history for
> commit e3e0109138376bb2 ("ARM / shmobile: Support for I/O power domains for
> SH7372 (v9)") doesn't have it.
>
> Google did find: https://lkml.org/lkml/2011/6/30/471
>
> Do we still need this at all? I.e. aren't PM Domains containing wake-up
> devices kept powered automatically during system suspend?

No they aren't. So for pm-rmobile we do need it.

For rcar-sysc it's different: as no PM Domain contains wake-up devices
(all I/O devices are in the always-on power area), we don't need the callback.
Will drop it in v6.

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert at linux-m68k.org

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

[PATCH v5 03/11] soc: renesas: rcar-sysc: Add DT support for SYSC PM domains

[Ulf Hansson]

On 18 April 2016 at 15:39, Geert Uytterhoeven <geert@linux-m68k.org> wrote:
> Hi Ulf,
>
> On Mon, Apr 18, 2016 at 2:59 PM, Geert Uytterhoeven
> <geert@linux-m68k.org> wrote:
>> On Mon, Apr 18, 2016 at 2:21 PM, Ulf Hansson <ulf.hansson@linaro.org> wrote:
>>> [...]
>>>
>>>> +
>>>> +static bool rcar_sysc_active_wakeup(struct device *dev)
>>>> +{
>>>> +       return true;
>>>
>>> I am interested to know why this is always returning true. Perhaps you
>>> can elaborate a bit on that?
>>
>> Too many copying from old shmobile PM Domain code?
>> Honestly, I don't know...
>>
>> Perhaps Rafael still remembers the original rationale, as git history for
>> commit e3e0109138376bb2 ("ARM / shmobile: Support for I/O power domains for
>> SH7372 (v9)") doesn't have it.
>>
>> Google did find: https://lkml.org/lkml/2011/6/30/471
>>
>> Do we still need this at all? I.e. aren't PM Domains containing wake-up
>> devices kept powered automatically during system suspend?
>
> No they aren't. So for pm-rmobile we do need it.

I don't quite understand why genpd should need to treat all devices
within the same domain exactly the same, it seems suboptimal.

I guess it would be more clever to allow this to be controlled on per
device basis instead, so let's say from each driver.

>
> For rcar-sysc it's different: as no PM Domain contains wake-up devices
> (all I/O devices are in the always-on power area), we don't need the callback.
> Will drop it in v6.

Okay, great!

Kind regards
Uffe

[PATCH v5 03/11] soc: renesas: rcar-sysc: Add DT support for SYSC PM domains

[Geert Uytterhoeven]

Hi Ulf,

On Mon, Apr 18, 2016 at 4:02 PM, Ulf Hansson <ulf.hansson@linaro.org> wrote:
> On 18 April 2016 at 15:39, Geert Uytterhoeven <geert@linux-m68k.org> wrote:
>> On Mon, Apr 18, 2016 at 2:59 PM, Geert Uytterhoeven
>> <geert@linux-m68k.org> wrote:
>>> On Mon, Apr 18, 2016 at 2:21 PM, Ulf Hansson <ulf.hansson@linaro.org> wrote:
>>>> [...]
>>>>
>>>>> +
>>>>> +static bool rcar_sysc_active_wakeup(struct device *dev)
>>>>> +{
>>>>> +       return true;
>>>>
>>>> I am interested to know why this is always returning true. Perhaps you
>>>> can elaborate a bit on that?
>>>
>>> Too many copying from old shmobile PM Domain code?
>>> Honestly, I don't know...
>>>
>>> Perhaps Rafael still remembers the original rationale, as git history for
>>> commit e3e0109138376bb2 ("ARM / shmobile: Support for I/O power domains for
>>> SH7372 (v9)") doesn't have it.
>>>
>>> Google did find: https://lkml.org/lkml/2011/6/30/471
>>>
>>> Do we still need this at all? I.e. aren't PM Domains containing wake-up
>>> devices kept powered automatically during system suspend?
>>
>> No they aren't. So for pm-rmobile we do need it.
>
> I don't quite understand why genpd should need to treat all devices
> within the same domain exactly the same, it seems suboptimal.
>
> I guess it would be more clever to allow this to be controlled on per
> device basis instead, so let's say from each driver.

Perhaps this can be handled through device_set_wakeup_enable()?
Unfortunately this doesn't seem to be called from e.g. gpio-keys.

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert at linux-m68k.org

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

[PATCH v5 03/11] soc: renesas: rcar-sysc: Add DT support for SYSC PM domains

[Ulf Hansson]

On 20 April 2016 at 10:24, Geert Uytterhoeven <geert@linux-m68k.org> wrote:
> Hi Ulf,
>
> On Mon, Apr 18, 2016 at 4:02 PM, Ulf Hansson <ulf.hansson@linaro.org> wrote:
>> On 18 April 2016 at 15:39, Geert Uytterhoeven <geert@linux-m68k.org> wrote:
>>> On Mon, Apr 18, 2016 at 2:59 PM, Geert Uytterhoeven
>>> <geert@linux-m68k.org> wrote:
>>>> On Mon, Apr 18, 2016 at 2:21 PM, Ulf Hansson <ulf.hansson@linaro.org> wrote:
>>>>> [...]
>>>>>
>>>>>> +
>>>>>> +static bool rcar_sysc_active_wakeup(struct device *dev)
>>>>>> +{
>>>>>> +       return true;
>>>>>
>>>>> I am interested to know why this is always returning true. Perhaps you
>>>>> can elaborate a bit on that?
>>>>
>>>> Too many copying from old shmobile PM Domain code?
>>>> Honestly, I don't know...
>>>>
>>>> Perhaps Rafael still remembers the original rationale, as git history for
>>>> commit e3e0109138376bb2 ("ARM / shmobile: Support for I/O power domains for
>>>> SH7372 (v9)") doesn't have it.
>>>>
>>>> Google did find: https://lkml.org/lkml/2011/6/30/471
>>>>
>>>> Do we still need this at all? I.e. aren't PM Domains containing wake-up
>>>> devices kept powered automatically during system suspend?
>>>
>>> No they aren't. So for pm-rmobile we do need it.
>>
>> I don't quite understand why genpd should need to treat all devices
>> within the same domain exactly the same, it seems suboptimal.
>>
>> I guess it would be more clever to allow this to be controlled on per
>> device basis instead, so let's say from each driver.
>
> Perhaps this can be handled through device_set_wakeup_enable()?

Yes, the pm_wakeirq API should help with all what is needed.

> Unfortunately this doesn't seem to be called from e.g. gpio-keys.
>

Okay, so it's a matter of deployment for these devices/drivers.

A list of such drivers/devices that needs to be fixed would be great
to have. :-)

Kind regards
Uffe