Skip to contentSkip to conversation

RISC-V: ACPI: Add LPI support

PATCH Superseded Sunil V L 4 participants 11 messages Jan 11, 2024 – Jan 15, 2024

Subsystems (4)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Sunil V L

author
#0 lore raw

cover RISC-V: ACPI: Add LPI support

This series adds support for Low Power Idle (LPI) on ACPI based platforms.

LPI is described in the ACPI spec [1]. RISC-V FFH spec required to enable this is available at [2].

[1] - https://uefi.org/specs/ACPI/6.5/08_Processor_Configuration_and_Control.html#lpi-low-power-idle-states
[2] - https://github.com/riscv-non-isa/riscv-acpi-ffh/releases/download/v/riscv-ffh.pdf

Sunil V L (2):
ACPI: Enable ACPI_PROCESSOR for RISC-V
cpuidle: RISC-V: Add ACPI LPI support

 drivers/acpi/Kconfig                |  2 +-
 drivers/cpuidle/cpuidle-riscv-sbi.c | 78 +++++++++++++++++++++++++++++
 2 files changed, 79 insertions(+), 1 deletion(-)
Signature · 1 line
2.34.1
author
#1 lore raw

PATCH 1/2 ACPI: Enable ACPI_PROCESSOR for RISC-V

The ACPI processor driver is not currently enabled for RISC-V. This is required to enable CPU related functionalities like LPI and CPPC. Hence, enable ACPI_PROCESSOR for RISC-V.

---
 drivers/acpi/Kconfig | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/acpi/Kconfig +1 -1
index f819e760ff19..9a920752171c 100644
@@ -282,7 +282,7 @@ config ACPI_CPPC_LIB
 
 config ACPI_PROCESSOR
 	tristate "Processor"
-	depends on X86 || ARM64 || LOONGARCH
+	depends on X86 || ARM64 || LOONGARCH || RISCV
 	select ACPI_PROCESSOR_IDLE
 	select ACPI_CPU_FREQ_PSS if X86 || LOONGARCH
 	select THERMAL
Signature · 1 line
2.34.1
author
#6 lore raw

PATCH 2/2 cpuidle: RISC-V: Add ACPI LPI support

In reply to Sunil V L

Add required callbacks to support Low Power Idle (LPI) on ACPI based RISC-V platforms.

---
 drivers/cpuidle/cpuidle-riscv-sbi.c | 78 +++++++++++++++++++++++++++++
 1 file changed, 78 insertions(+)
drivers/cpuidle/cpuidle-riscv-sbi.c +78 -0
index e8094fc92491..cea67a54ab39 100644
@@ -632,3 +632,81 @@ static int __init sbi_cpuidle_init(void)
 	return 0;
 }
 device_initcall(sbi_cpuidle_init);
+
+#ifdef CONFIG_ACPI_PROCESSOR_IDLE
+
+#include <linux/acpi.h>
+#include <acpi/processor.h>
+
+#define RISCV_FFH_LPI_TYPE_MASK		0x1000000000000000ULL
+#define RISCV_FFH_LPI_RSVD_MASK		0x0FFFFFFF00000000ULL
+
+static int acpi_cpu_init_idle(unsigned int cpu)
+{
+	int i;
+	struct acpi_lpi_state *lpi;
+	struct acpi_processor *pr = per_cpu(processors, cpu);
+
+	if (unlikely(!pr || !pr->flags.has_lpi))
+		return -EINVAL;
+
+	/*
+	 * The SBI HSM suspend function is only available when:
+	 * 1) SBI version is 0.3 or higher
+	 * 2) SBI HSM extension is available
+	 */
+	if (sbi_spec_version < sbi_mk_version(0, 3) ||
+	    !sbi_probe_extension(SBI_EXT_HSM)) {
+		pr_warn("HSM suspend not available\n");
+		return -EINVAL;
+	}
+
+	if (pr->power.count <= 1)
+		return -ENODEV;
+
+	for (i = 1; i < pr->power.count; i++) {
+		u32 state;
+
+		lpi = &pr->power.lpi_states[i];
+
+		/* Validate Entry Method as per FFH spec.
+		 * bits[63:60] should be 0x1
+		 * bits[59:32] should be 0x0
+		 * bits[31:0] represent a SBI power_state
+		 */
+		if (!(lpi->address & RISCV_FFH_LPI_TYPE_MASK) ||
+		    (lpi->address & RISCV_FFH_LPI_RSVD_MASK)) {
+			pr_warn("Invalid LPI entry method %#llx\n", lpi->address);
+			return -EINVAL;
+		}
+
+		state = lpi->address;
+		if (!sbi_suspend_state_is_valid(state)) {
+			pr_warn("Invalid SBI power state %#x\n", state);
+			return -EINVAL;
+		}
+	}
+
+	return 0;
+}
+
+int acpi_processor_ffh_lpi_probe(unsigned int cpu)
+{
+	return acpi_cpu_init_idle(cpu);
+}
+
+int acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
+{
+	u32 state = lpi->address;
+
+	if (state & SBI_HSM_SUSP_NON_RET_BIT)
+		return CPU_PM_CPU_IDLE_ENTER_PARAM(sbi_suspend,
+						   lpi->index,
+						   state);
+	else
+		return CPU_PM_CPU_IDLE_ENTER_RETENTION_PARAM(sbi_suspend,
+							     lpi->index,
+							     state);
+}
+
+#endif
Signature · 1 line
2.34.1
Reviewed-by on v1 1/2
#2 lore raw

Re: [PATCH -next 1/2] ACPI: Enable ACPI_PROCESSOR for RISC-V

In reply to Sunil V L
Quoted 25 lines · Sunil V L

The ACPI processor driver is not currently enabled for RISC-V. This is required to enable CPU related functionalities like LPI and CPPC. Hence, enable ACPI_PROCESSOR for RISC-V.

---
 drivers/acpi/Kconfig | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/acpi/Kconfig +1 -1
index f819e760ff19..9a920752171c 100644
@@ -282,7 +282,7 @@ config ACPI_CPPC_LIB
 
 config ACPI_PROCESSOR
 	tristate "Processor"
-	depends on X86 || ARM64 || LOONGARCH
+	depends on X86 || ARM64 || LOONGARCH || RISCV
 	select ACPI_PROCESSOR_IDLE
 	select ACPI_CPU_FREQ_PSS if X86 || LOONGARCH
 	select THERMAL
Signature · 1 line
2.34.1

Hi Sunil,

Typically we'll want the Kconfig changes to come at the end of a series, or squashed into the patch that adds support for it, otherwise there's risk of build breakage during bisection. In this case, we're safe because the two new functions (I looked ahead) have __weak versions when they're not present.

Also, interestingly, it looks like this ancient line

obj-$(CONFIG_ACPI_PROCESSOR) += processor.o

in drivers/acpi/Makefile should be removed, since there's no drivers/acpi/processor.c file. I guess the make process silently filters object files which don't have corresponding source files? Maybe we should write a Makefile analyzer to see what other lines can be removed...

Anyway, for this patch, which I'd prefer to be swapped in order with the other patch, or just squashed into the other patch,

Thanks,
drew

#7 lore raw

Re: [PATCH -next 2/2] cpuidle: RISC-V: Add ACPI LPI support

In reply to Sunil V L
Quoted 14 lines · Sunil V L

Add required callbacks to support Low Power Idle (LPI) on ACPI based RISC-V platforms.

---
 drivers/cpuidle/cpuidle-riscv-sbi.c | 78 +++++++++++++++++++++++++++++
 1 file changed, 78 insertions(+)
drivers/cpuidle/cpuidle-riscv-sbi.c +0 -0
index e8094fc92491..cea67a54ab39 100644
@@ -632,3 +632,81 @@ static int __init sbi_cpuidle_init(void)
 	return 0;

drivers/cpuidle/cpuidle-riscv-sbi.c static int __init sbi_cpuidle_init(void) · line 642

Quoted 10 lines
drivers/cpuidle/cpuidle-riscv-sbi.c +8 -0
 }
 device_initcall(sbi_cpuidle_init);
+
+#ifdef CONFIG_ACPI_PROCESSOR_IDLE
+
+#include <linux/acpi.h>
+#include <acpi/processor.h>
+
+#define RISCV_FFH_LPI_TYPE_MASK		0x1000000000000000ULL
+#define RISCV_FFH_LPI_RSVD_MASK		0x0FFFFFFF00000000ULL

GENMASK might look nicer and the type mask is 0xF000000000000000ULL, where 0x1000000000000000ULL means that the type is an SBI identifier. We need both defined

#define RISCV_FFH_LPI_TYPE_MASK 0xF000000000000000ULL
#define RISCV_FFH_LPI_TYPE_SBI 0x1000000000000000ULL

as I point out below.

Quoted 8 lines
drivers/cpuidle/cpuidle-riscv-sbi.c +8 -0
+
+static int acpi_cpu_init_idle(unsigned int cpu)
+{
+	int i;
+	struct acpi_lpi_state *lpi;
+	struct acpi_processor *pr = per_cpu(processors, cpu);
+
+	if (unlikely(!pr || !pr->flags.has_lpi))

drivers/cpuidle/cpuidle-riscv-sbi.c static int __init sbi_cpuidle_init(void) · line 660

Quoted 10 lines
drivers/cpuidle/cpuidle-riscv-sbi.c +10 -0
+		return -EINVAL;
+
+	/*
+	 * The SBI HSM suspend function is only available when:
+	 * 1) SBI version is 0.3 or higher
+	 * 2) SBI HSM extension is available
+	 */
+	if (sbi_spec_version < sbi_mk_version(0, 3) ||
+	    !sbi_probe_extension(SBI_EXT_HSM)) {
+		pr_warn("HSM suspend not available\n");

The comment and these lines match what's done in sbi_cpuidle_init(). How about a static helper function to avoid duplication?

Quoted 5 lines
drivers/cpuidle/cpuidle-riscv-sbi.c +5 -0
+		return -EINVAL;
+	}
+
+	if (pr->power.count <= 1)
+		return -ENODEV;

drivers/cpuidle/cpuidle-riscv-sbi.c static int __init sbi_cpuidle_init(void) · line 675

Quoted 10 lines
drivers/cpuidle/cpuidle-riscv-sbi.c +10 -0
+
+	for (i = 1; i < pr->power.count; i++) {
+		u32 state;
+
+		lpi = &pr->power.lpi_states[i];
+
+		/* Validate Entry Method as per FFH spec.
+		 * bits[63:60] should be 0x1
+		 * bits[59:32] should be 0x0
+		 * bits[31:0] represent a SBI power_state

^ an

drivers/cpuidle/cpuidle-riscv-sbi.c static int __init sbi_cpuidle_init(void) · line 676

Quoted 1 line
drivers/cpuidle/cpuidle-riscv-sbi.c +1 -0
+		 */

Comment block needs opening wing (/*)

drivers/cpuidle/cpuidle-riscv-sbi.c static int __init sbi_cpuidle_init(void) · line 677

Quoted 1 line
drivers/cpuidle/cpuidle-riscv-sbi.c +1 -0
+		if (!(lpi->address & RISCV_FFH_LPI_TYPE_MASK) ||

This should be (lpi->address & RISCV_FFH_LPI_TYPE_MASK) != RISCV_FFH_LPI_TYPE_SBI

Quoted 38 lines
drivers/cpuidle/cpuidle-riscv-sbi.c +35 -0
+		    (lpi->address & RISCV_FFH_LPI_RSVD_MASK)) {
+			pr_warn("Invalid LPI entry method %#llx\n", lpi->address);
+			return -EINVAL;
+		}
+
+		state = lpi->address;
+		if (!sbi_suspend_state_is_valid(state)) {
+			pr_warn("Invalid SBI power state %#x\n", state);
+			return -EINVAL;
+		}
+	}
+
+	return 0;
+}
+
+int acpi_processor_ffh_lpi_probe(unsigned int cpu)
+{
+	return acpi_cpu_init_idle(cpu);
+}
+
+int acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
+{
+	u32 state = lpi->address;
+
+	if (state & SBI_HSM_SUSP_NON_RET_BIT)
+		return CPU_PM_CPU_IDLE_ENTER_PARAM(sbi_suspend,
+						   lpi->index,
+						   state);
+	else
+		return CPU_PM_CPU_IDLE_ENTER_RETENTION_PARAM(sbi_suspend,
+							     lpi->index,
+							     state);
+}
+
+#endif
Signature · 1 line
2.34.1

Thanks,
drew

author
#3 lore raw

Re: [PATCH -next 1/2] ACPI: Enable ACPI_PROCESSOR for RISC-V

In reply to Andrew Jones
Quoted 29 lines · Andrew Jones
Quoted 25 lines · Sunil V L

The ACPI processor driver is not currently enabled for RISC-V. This is required to enable CPU related functionalities like LPI and CPPC. Hence, enable ACPI_PROCESSOR for RISC-V.

---
 drivers/acpi/Kconfig | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/acpi/Kconfig +1 -1
index f819e760ff19..9a920752171c 100644
@@ -282,7 +282,7 @@ config ACPI_CPPC_LIB
 
 config ACPI_PROCESSOR
 	tristate "Processor"
-	depends on X86 || ARM64 || LOONGARCH
+	depends on X86 || ARM64 || LOONGARCH || RISCV
 	select ACPI_PROCESSOR_IDLE
 	select ACPI_CPU_FREQ_PSS if X86 || LOONGARCH
 	select THERMAL
Signature · 1 line
2.34.1

Hi Sunil,

Quoted 6 lines

Typically we'll want the Kconfig changes to come at the end of a series, or squashed into the patch that adds support for it, otherwise there's risk of build breakage during bisection. In this case, we're safe because the two new functions (I looked ahead) have __weak versions when they're not present.

Sure. Let me swap the order of the patches.

Quoted 10 lines

Also, interestingly, it looks like this ancient line

obj-$(CONFIG_ACPI_PROCESSOR) += processor.o

in drivers/acpi/Makefile should be removed, since there's no drivers/acpi/processor.c file. I guess the make process silently filters object files which don't have corresponding source files? Maybe we should write a Makefile analyzer to see what other lines can be removed...

Interesting.

Hi Rafael, any thoughts?

Quoted 3 lines

Anyway, for this patch, which I'd prefer to be swapped in order with the other patch, or just squashed into the other patch,

I prefer to keep as 2 separate patches. I will swap the order.

Quoted 2 lines

Thanks!
Sunil

author
#8 lore raw

Re: [PATCH -next 2/2] cpuidle: RISC-V: Add ACPI LPI support

In reply to Andrew Jones
Quoted 30 lines · Andrew Jones
Quoted 14 lines · Sunil V L

Add required callbacks to support Low Power Idle (LPI) on ACPI based RISC-V platforms.

---
 drivers/cpuidle/cpuidle-riscv-sbi.c | 78 +++++++++++++++++++++++++++++
 1 file changed, 78 insertions(+)
drivers/cpuidle/cpuidle-riscv-sbi.c +0 -0
index e8094fc92491..cea67a54ab39 100644
@@ -632,3 +632,81 @@ static int __init sbi_cpuidle_init(void)
 	return 0;

drivers/cpuidle/cpuidle-riscv-sbi.c static int __init sbi_cpuidle_init(void) · line 642

Quoted 10 lines
drivers/cpuidle/cpuidle-riscv-sbi.c +8 -0
 }
 device_initcall(sbi_cpuidle_init);
+
+#ifdef CONFIG_ACPI_PROCESSOR_IDLE
+
+#include <linux/acpi.h>
+#include <acpi/processor.h>
+
+#define RISCV_FFH_LPI_TYPE_MASK		0x1000000000000000ULL
+#define RISCV_FFH_LPI_RSVD_MASK		0x0FFFFFFF00000000ULL

GENMASK might look nicer and the type mask is 0xF000000000000000ULL, where 0x1000000000000000ULL means that the type is an SBI identifier. We need both defined

Quoted 3 lines

#define RISCV_FFH_LPI_TYPE_MASK 0xF000000000000000ULL
#define RISCV_FFH_LPI_TYPE_SBI 0x1000000000000000ULL

Sure. Let me use GENMASK and define both MASK and SBI type.

Quoted 15 lines

as I point out below.

Quoted 13 lines
drivers/cpuidle/cpuidle-riscv-sbi.c +13 -0
+
+static int acpi_cpu_init_idle(unsigned int cpu)
+{
+	int i;
+	struct acpi_lpi_state *lpi;
+	struct acpi_processor *pr = per_cpu(processors, cpu);
+
+	if (unlikely(!pr || !pr->flags.has_lpi))
+		return -EINVAL;
+
+	/*
+	 * The SBI HSM suspend function is only available when:
+	 * 1) SBI version is 0.3 or higher
Quoted 9 lines

drivers/cpuidle/cpuidle-riscv-sbi.c static int __init sbi_cpuidle_init(void) · line 660

Quoted 5 lines
drivers/cpuidle/cpuidle-riscv-sbi.c +5 -0
+	 * 2) SBI HSM extension is available
+	 */
+	if (sbi_spec_version < sbi_mk_version(0, 3) ||
+	    !sbi_probe_extension(SBI_EXT_HSM)) {
+		pr_warn("HSM suspend not available\n");

The comment and these lines match what's done in sbi_cpuidle_init(). How about a static helper function to avoid duplication?

Sure.

Quoted 17 lines
Quoted 5 lines
drivers/cpuidle/cpuidle-riscv-sbi.c +5 -0
+		return -EINVAL;
+	}
+
+	if (pr->power.count <= 1)
+		return -ENODEV;

drivers/cpuidle/cpuidle-riscv-sbi.c static int __init sbi_cpuidle_init(void) · line 675

Quoted 10 lines
drivers/cpuidle/cpuidle-riscv-sbi.c +10 -0
+
+	for (i = 1; i < pr->power.count; i++) {
+		u32 state;
+
+		lpi = &pr->power.lpi_states[i];
+
+		/* Validate Entry Method as per FFH spec.
+		 * bits[63:60] should be 0x1
+		 * bits[59:32] should be 0x0
+		 * bits[31:0] represent a SBI power_state

^ an

Quoted 4 lines

drivers/cpuidle/cpuidle-riscv-sbi.c static int __init sbi_cpuidle_init(void) · line 676

Quoted 1 line
drivers/cpuidle/cpuidle-riscv-sbi.c +1 -0
+		 */

Comment block needs opening wing (/*)

Okay.

Quoted 4 lines

drivers/cpuidle/cpuidle-riscv-sbi.c static int __init sbi_cpuidle_init(void) · line 677

Quoted 1 line
drivers/cpuidle/cpuidle-riscv-sbi.c +1 -0
+		if (!(lpi->address & RISCV_FFH_LPI_TYPE_MASK) ||

This should be (lpi->address & RISCV_FFH_LPI_TYPE_MASK) != RISCV_FFH_LPI_TYPE_SBI

Sure.

Let me send v2 in couple of days with these changes.

Thanks!
Sunil

Quoted 41 lines
Quoted 38 lines
drivers/cpuidle/cpuidle-riscv-sbi.c +35 -0
+		    (lpi->address & RISCV_FFH_LPI_RSVD_MASK)) {
+			pr_warn("Invalid LPI entry method %#llx\n", lpi->address);
+			return -EINVAL;
+		}
+
+		state = lpi->address;
+		if (!sbi_suspend_state_is_valid(state)) {
+			pr_warn("Invalid SBI power state %#x\n", state);
+			return -EINVAL;
+		}
+	}
+
+	return 0;
+}
+
+int acpi_processor_ffh_lpi_probe(unsigned int cpu)
+{
+	return acpi_cpu_init_idle(cpu);
+}
+
+int acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
+{
+	u32 state = lpi->address;
+
+	if (state & SBI_HSM_SUSP_NON_RET_BIT)
+		return CPU_PM_CPU_IDLE_ENTER_PARAM(sbi_suspend,
+						   lpi->index,
+						   state);
+	else
+		return CPU_PM_CPU_IDLE_ENTER_RETENTION_PARAM(sbi_suspend,
+							     lpi->index,
+							     state);
+}
+
+#endif
Signature · 1 line
2.34.1

Thanks,
drew

#4 lore raw

Re: [PATCH -next 1/2] ACPI: Enable ACPI_PROCESSOR for RISC-V

In reply to Andrew Jones

On Thu, Jan 11, 2024 at 11:00:12AM +0100, Andrew Jones wrote:

[...]

Quoted 5 lines

Also, interestingly, it looks like this ancient line

obj-$(CONFIG_ACPI_PROCESSOR) += processor.o

in drivers/acpi/Makefile should be removed,

No

Quoted 1 line

since there's no drivers/acpi/processor.c file.

Correct, but ..

Quoted 2 lines

I guess the make process silently filters object files which don't have corresponding source files?

May be, but I doubt if that is the case here.

processor.o is just aggregation of all processor_*.o and this will be the processor.ko when built as a module.

--
Regards,
Sudeep

#5 lore raw

Re: Re: [PATCH -next 1/2] ACPI: Enable ACPI_PROCESSOR for RISC-V

Quoted 16 lines · Sudeep Holla

On Thu, Jan 11, 2024 at 11:00:12AM +0100, Andrew Jones wrote:

[...]

Quoted 5 lines

Also, interestingly, it looks like this ancient line

obj-$(CONFIG_ACPI_PROCESSOR) += processor.o

in drivers/acpi/Makefile should be removed,

No

Quoted 1 line

since there's no drivers/acpi/processor.c file.

Correct, but ..

Quoted 7 lines
Quoted 2 lines

I guess the make process silently filters object files which don't have corresponding source files?

May be, but I doubt if that is the case here.

processor.o is just aggregation of all processor_*.o and this will be the processor.ko when built as a module.

Oh, I see. I had tried looking for a processor.o after building, to see if it was something like that, but it still didn't appear. It didn't occur to me to also try ACPI_PROCESSOR as a module.

I'll go put my head in some sand now.

Thanks,
drew

maintainer
#9 lore raw

Re: [PATCH -next 2/2] cpuidle: RISC-V: Add ACPI LPI support

In reply to Sunil V L
Quoted 85 lines · Sunil V L

Add required callbacks to support Low Power Idle (LPI) on ACPI based RISC-V platforms.

---
 drivers/cpuidle/cpuidle-riscv-sbi.c | 78 +++++++++++++++++++++++++++++
 1 file changed, 78 insertions(+)
drivers/cpuidle/cpuidle-riscv-sbi.c +68 -0
index e8094fc92491..cea67a54ab39 100644
@@ -632,3 +632,81 @@ static int __init sbi_cpuidle_init(void)
        return 0;
 }
 device_initcall(sbi_cpuidle_init);
+
+#ifdef CONFIG_ACPI_PROCESSOR_IDLE
+
+#include <linux/acpi.h>
+#include <acpi/processor.h>
+
+#define RISCV_FFH_LPI_TYPE_MASK                0x1000000000000000ULL
+#define RISCV_FFH_LPI_RSVD_MASK                0x0FFFFFFF00000000ULL
+
+static int acpi_cpu_init_idle(unsigned int cpu)
+{
+       int i;
+       struct acpi_lpi_state *lpi;
+       struct acpi_processor *pr = per_cpu(processors, cpu);
+
+       if (unlikely(!pr || !pr->flags.has_lpi))
+               return -EINVAL;
+
+       /*
+        * The SBI HSM suspend function is only available when:
+        * 1) SBI version is 0.3 or higher
+        * 2) SBI HSM extension is available
+        */
+       if (sbi_spec_version < sbi_mk_version(0, 3) ||
+           !sbi_probe_extension(SBI_EXT_HSM)) {
+               pr_warn("HSM suspend not available\n");
+               return -EINVAL;
+       }
+
+       if (pr->power.count <= 1)
+               return -ENODEV;
+
+       for (i = 1; i < pr->power.count; i++) {
+               u32 state;
+
+               lpi = &pr->power.lpi_states[i];
+
+               /* Validate Entry Method as per FFH spec.
+                * bits[63:60] should be 0x1
+                * bits[59:32] should be 0x0
+                * bits[31:0] represent a SBI power_state
+                */
+               if (!(lpi->address & RISCV_FFH_LPI_TYPE_MASK) ||
+                   (lpi->address & RISCV_FFH_LPI_RSVD_MASK)) {
+                       pr_warn("Invalid LPI entry method %#llx\n", lpi->address);
+                       return -EINVAL;
+               }
+
+               state = lpi->address;
+               if (!sbi_suspend_state_is_valid(state)) {
+                       pr_warn("Invalid SBI power state %#x\n", state);
+                       return -EINVAL;
+               }
+       }
+
+       return 0;
+}
+
+int acpi_processor_ffh_lpi_probe(unsigned int cpu)
+{
+       return acpi_cpu_init_idle(cpu);
+}
+
+int acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
+{
+       u32 state = lpi->address;
+
+       if (state & SBI_HSM_SUSP_NON_RET_BIT)

drivers/cpuidle/cpuidle-riscv-sbi.c static int __init sbi_cpuidle_init(void) · line 712

Quoted 10 lines
drivers/cpuidle/cpuidle-riscv-sbi.c +10 -0
+               return CPU_PM_CPU_IDLE_ENTER_PARAM(sbi_suspend,
+                                                  lpi->index,
+                                                  state);
+       else
+               return CPU_PM_CPU_IDLE_ENTER_RETENTION_PARAM(sbi_suspend,
+                                                            lpi->index,
+                                                            state);
+}
+
+#endif

Lets keep the cpuidle-riscv-sbi.c driver focused on DT only. Instead, I would suggest moving the required function from cpuidle-riscv-sbi.c to arch/riscv and have a separate driver under driver/acpi/riscv for LPI states.

Regards,
Anup

author
#10 lore raw

Re: [PATCH -next 2/2] cpuidle: RISC-V: Add ACPI LPI support

Quoted 93 lines · Anup Patel
Quoted 92 lines · Sunil V L

Add required callbacks to support Low Power Idle (LPI) on ACPI based RISC-V platforms.

---
 drivers/cpuidle/cpuidle-riscv-sbi.c | 78 +++++++++++++++++++++++++++++
 1 file changed, 78 insertions(+)
drivers/cpuidle/cpuidle-riscv-sbi.c +75 -0
index e8094fc92491..cea67a54ab39 100644
@@ -632,3 +632,81 @@ static int __init sbi_cpuidle_init(void)
        return 0;
 }
 device_initcall(sbi_cpuidle_init);
+
+#ifdef CONFIG_ACPI_PROCESSOR_IDLE
+
+#include <linux/acpi.h>
+#include <acpi/processor.h>
+
+#define RISCV_FFH_LPI_TYPE_MASK                0x1000000000000000ULL
+#define RISCV_FFH_LPI_RSVD_MASK                0x0FFFFFFF00000000ULL
+
+static int acpi_cpu_init_idle(unsigned int cpu)
+{
+       int i;
+       struct acpi_lpi_state *lpi;
+       struct acpi_processor *pr = per_cpu(processors, cpu);
+
+       if (unlikely(!pr || !pr->flags.has_lpi))
+               return -EINVAL;
+
+       /*
+        * The SBI HSM suspend function is only available when:
+        * 1) SBI version is 0.3 or higher
+        * 2) SBI HSM extension is available
+        */
+       if (sbi_spec_version < sbi_mk_version(0, 3) ||
+           !sbi_probe_extension(SBI_EXT_HSM)) {
+               pr_warn("HSM suspend not available\n");
+               return -EINVAL;
+       }
+
+       if (pr->power.count <= 1)
+               return -ENODEV;
+
+       for (i = 1; i < pr->power.count; i++) {
+               u32 state;
+
+               lpi = &pr->power.lpi_states[i];
+
+               /* Validate Entry Method as per FFH spec.
+                * bits[63:60] should be 0x1
+                * bits[59:32] should be 0x0
+                * bits[31:0] represent a SBI power_state
+                */
+               if (!(lpi->address & RISCV_FFH_LPI_TYPE_MASK) ||
+                   (lpi->address & RISCV_FFH_LPI_RSVD_MASK)) {
+                       pr_warn("Invalid LPI entry method %#llx\n", lpi->address);
+                       return -EINVAL;
+               }
+
+               state = lpi->address;
+               if (!sbi_suspend_state_is_valid(state)) {
+                       pr_warn("Invalid SBI power state %#x\n", state);
+                       return -EINVAL;
+               }
+       }
+
+       return 0;
+}
+
+int acpi_processor_ffh_lpi_probe(unsigned int cpu)
+{
+       return acpi_cpu_init_idle(cpu);
+}
+
+int acpi_processor_ffh_lpi_enter(struct acpi_lpi_state *lpi)
+{
+       u32 state = lpi->address;
+
+       if (state & SBI_HSM_SUSP_NON_RET_BIT)
+               return CPU_PM_CPU_IDLE_ENTER_PARAM(sbi_suspend,
+                                                  lpi->index,
+                                                  state);
+       else
+               return CPU_PM_CPU_IDLE_ENTER_RETENTION_PARAM(sbi_suspend,
+                                                            lpi->index,
+                                                            state);
Quoted 9 lines

drivers/cpuidle/cpuidle-riscv-sbi.c static int __init sbi_cpuidle_init(void) · line 712

Quoted 3 lines
drivers/cpuidle/cpuidle-riscv-sbi.c +3 -0
+}
+
+#endif

Lets keep the cpuidle-riscv-sbi.c driver focused on DT only. Instead, I would suggest moving the required function from cpuidle-riscv-sbi.c to arch/riscv and have a separate driver under driver/acpi/riscv for LPI states.

Okay, sure. Let me send v2 with your suggestion.

Thanks,
Sunil