Skip to contentSkip to conversation

lib: utils: Add fdt_add_cpu_idle_states() helper function

PATCH Superseded Samuel Holland 3 participants 7 messages Dec 28, 2022 – Jan 13, 2023 Arm/Allwinner sunXi SoC support

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Samuel Holland

author maintainer
#0 lore raw

PATCH 1/2 lib: utils: Add fdt_add_cpu_idle_states() helper function

Since the availability and latency properties of CPU idle states depend on the specific SBI HSM implementation, it is appropriate that the idle states are added to the devicetree at runtime by that implementation.

This helper function adds a platform-provided array of idle states to the devicetree, following the SBI idle state binding. It makes some assumptions for simplicity, but these could be relaxed if needed.

---

 include/sbi_utils/fdt/fdt_fixup.h | 23 +++++++++
 lib/utils/fdt/fdt_fixup.c         | 80 +++++++++++++++++++++++++++++++
 2 files changed, 103 insertions(+)
include/sbi_utils/fdt/fdt_fixup.h +23 -0
index fb076ba..cab3f0f 100644
@@ -9,6 +9,29 @@
 #ifndef __FDT_FIXUP_H__
 #define __FDT_FIXUP_H__
 
+struct sbi_cpu_idle_state {
+	const char *name;
+	uint32_t suspend_param;
+	bool local_timer_stop;
+	uint32_t entry_latency_us;
+	uint32_t exit_latency_us;
+	uint32_t min_residency_us;
+	uint32_t wakeup_latency_us;
+};
+
+/**
+ * Add CPU idle states to cpu nodes in the DT
+ *
+ * Add information about CPU idle states to the devicetree. This function
+ * assumes that CPU idle states are not already present in the devicetree, and
+ * that all CPU states are equally applicable to all CPUs.
+ *
+ * @param fdt: device tree blob
+ * @param states: array of idle state descriptions, ending with empty element
+ * @return zero on success and -ve on failure
+ */
+int fdt_add_cpu_idle_states(void *dtb, const struct sbi_cpu_idle_state *state);
+
 /**
  * Fix up the CPU node in the device tree
  *
lib/utils/fdt/fdt_fixup.c +80 -0

86 lines not on this page · raw message

Signature · 1 line
2.37.4
author maintainer
#1 lore raw

PATCH 2/2 platform: generic: allwinner: Advertise nonretentive suspend

Add D1's nonretentive suspend state to the devicetree so S-mode software knows about it and can use it.

Latency and power measurements were taken on an Allwinner Nezha board:
- Entry latency was measured from the beginning of sbi_ecall_handler() to before the call to wfi() in sun20i_d1_hart_suspend().
- Exit latency was measured from the beginning of sbi_init() to before the call to sbi_hart_switch_mode() in init_warmboot().
- There was a 17.5 mW benefit from non-retentive suspend compared to WFI, with a 170 mW cost during the 107 us entry/exit period. This provides a break-even point around 1040 us. Residency includes entry latency, so round this up to 1100 us.
- The hardware power sequence latency (after the WFI) is assumed to be negligible, so set the wakeup latency to the exit latency.

---

 platform/generic/allwinner/sun20i-d1.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)
platform/generic/allwinner/sun20i-d1.c +20 -0
index 1da9e5b..8ca556e 100644
@@ -12,6 +12,7 @@
 #include <sbi/sbi_error.h>
 #include <sbi/sbi_hsm.h>
 #include <sbi/sbi_pmu.h>
+#include <sbi_utils/fdt/fdt_fixup.h>
 #include <sbi_utils/fdt/fdt_helper.h>
 #include <sbi_utils/irqchip/fdt_irqchip_plic.h>
 
@@ -202,6 +203,24 @@ static int sun20i_d1_final_init(bool cold_boot, const struct fdt_match *match)
 	return 0;
 }
 
+static const struct sbi_cpu_idle_state sun20i_d1_cpu_idle_states[] = {
+	{
+		.name			= "cpu-nonretentive",
+		.suspend_param		= SBI_HSM_SUSPEND_NON_RET_DEFAULT,
+		.local_timer_stop	= true,
+		.entry_latency_us	= 40,
+		.exit_latency_us	= 67,
+		.min_residency_us	= 1100,
+		.wakeup_latency_us	= 67,
+	},
+	{ }
+};
+
+static int sun20i_d1_fdt_fixup(void *fdt, const struct fdt_match *match)
+{
+	return fdt_add_cpu_idle_states(fdt, sun20i_d1_cpu_idle_states);
+}
+
 static void thead_c9xx_pmu_ctr_enable_irq(uint32_t ctr_idx)
 {
 	unsigned long mip_val;
@@ -265,5 +284,6 @@ static const struct fdt_match sun20i_d1_match[] = {
 const struct platform_override sun20i_d1 = {
 	.match_table	= sun20i_d1_match,
 	.final_init	= sun20i_d1_final_init,
+	.fdt_fixup	= sun20i_d1_fdt_fixup,
 	.extensions_init = sun20i_d1_extensions_init,
 };
Signature · 1 line
2.37.4
#2 lore raw

PATCH 2/2 platform: generic: allwinner: Advertise nonretentive suspend

Quoted 16 lines · Samuel Holland

Add D1's nonretentive suspend state to the devicetree so S-mode software knows about it and can use it.

Latency and power measurements were taken on an Allwinner Nezha board:
- Entry latency was measured from the beginning of sbi_ecall_handler() to before the call to wfi() in sun20i_d1_hart_suspend().
- Exit latency was measured from the beginning of sbi_init() to before the call to sbi_hart_switch_mode() in init_warmboot().
- There was a 17.5 mW benefit from non-retentive suspend compared to WFI, with a 170 mW cost during the 107 us entry/exit period. This provides a break-even point around 1040 us. Residency includes entry latency, so round this up to 1100 us.
- The hardware power sequence latency (after the WFI) is assumed to be negligible, so set the wakeup latency to the exit latency.

Quoted 1 line

Looks good to me.

Regards,
Anup

Quoted 57 lines

---

 platform/generic/allwinner/sun20i-d1.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)
platform/generic/allwinner/sun20i-d1.c +20 -1
index 1da9e5b..8ca556e 100644
@@ -12,6 +12,7 @@
 #include <sbi/sbi_error.h>
 #include <sbi/sbi_hsm.h>
 #include <sbi/sbi_pmu.h>
+#include <sbi_utils/fdt/fdt_fixup.h>
 #include <sbi_utils/fdt/fdt_helper.h>
 #include <sbi_utils/irqchip/fdt_irqchip_plic.h>
 
@@ -202,6 +203,24 @@ static int sun20i_d1_final_init(bool cold_boot, const struct fdt_match *match)
        return 0;
 }
 
+static const struct sbi_cpu_idle_state sun20i_d1_cpu_idle_states[] = {
+       {
+               .name                   = "cpu-nonretentive",
+               .suspend_param          = SBI_HSM_SUSPEND_NON_RET_DEFAULT,
+               .local_timer_stop       = true,
+               .entry_latency_us       = 40,
+               .exit_latency_us        = 67,
+               .min_residency_us       = 1100,
+               .wakeup_latency_us      = 67,
+       },
+       { }
+};
+
+static int sun20i_d1_fdt_fixup(void *fdt, const struct fdt_match *match)
+{
+       return fdt_add_cpu_idle_states(fdt, sun20i_d1_cpu_idle_states);
+}
+
 static void thead_c9xx_pmu_ctr_enable_irq(uint32_t ctr_idx)
 {
        unsigned long mip_val;
@@ -265,5 +284,6 @@ static const struct fdt_match sun20i_d1_match[] = {
 const struct platform_override sun20i_d1 = {
        .match_table    = sun20i_d1_match,
        .final_init     = sun20i_d1_final_init,
+       .fdt_fixup      = sun20i_d1_fdt_fixup,
        .extensions_init = sun20i_d1_extensions_init,
 };
--

2.37.4

--
opensbi mailing list
opensbi at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/opensbi

#3 lore raw

PATCH 1/2 lib: utils: Add fdt_add_cpu_idle_states() helper function

In reply to Samuel Holland

Hi Samuel,

Quoted 69 lines · Samuel Holland

Since the availability and latency properties of CPU idle states depend on the specific SBI HSM implementation, it is appropriate that the idle states are added to the devicetree at runtime by that implementation.

This helper function adds a platform-provided array of idle states to the devicetree, following the SBI idle state binding. It makes some assumptions for simplicity, but these could be relaxed if needed.

---

 include/sbi_utils/fdt/fdt_fixup.h | 23 +++++++++
 lib/utils/fdt/fdt_fixup.c         | 80 +++++++++++++++++++++++++++++++
 2 files changed, 103 insertions(+)
include/sbi_utils/fdt/fdt_fixup.h +23 -0
index fb076ba..cab3f0f 100644
@@ -9,6 +9,29 @@
 #ifndef __FDT_FIXUP_H__
 #define __FDT_FIXUP_H__
 
+struct sbi_cpu_idle_state {
+	const char *name;
+	uint32_t suspend_param;
+	bool local_timer_stop;
+	uint32_t entry_latency_us;
+	uint32_t exit_latency_us;
+	uint32_t min_residency_us;
+	uint32_t wakeup_latency_us;
+};
+
+/**
+ * Add CPU idle states to cpu nodes in the DT
+ *
+ * Add information about CPU idle states to the devicetree. This function
+ * assumes that CPU idle states are not already present in the devicetree, and
+ * that all CPU states are equally applicable to all CPUs.
+ *
+ * @param fdt: device tree blob
+ * @param states: array of idle state descriptions, ending with empty element
+ * @return zero on success and -ve on failure
+ */
+int fdt_add_cpu_idle_states(void *dtb, const struct sbi_cpu_idle_state *state);
+
 /**
  * Fix up the CPU node in the device tree
  *
lib/utils/fdt/fdt_fixup.c +12 -0
index 41f6cbb..d9aa0b2 100644
@@ -18,6 +18,86 @@
 #include <sbi_utils/fdt/fdt_pmu.h>
 #include <sbi_utils/fdt/fdt_helper.h>
 
+int fdt_add_cpu_idle_states(void *fdt, const struct sbi_cpu_idle_state *state)
+{
+	int cpu_node, cpus_node, err, idle_states_node;
+	uint32_t count, phandle;
+
+	err = fdt_open_into(fdt, fdt, fdt_totalsize(fdt) + 1024);
+	if (err < 0)
+		return err;
+
+	err = fdt_find_max_phandle(fdt, &phandle);
+	phandle++;
+	if (err < 0)

lib/utils/fdt/fdt_fixup.c line 42

Quoted 10 lines
lib/utils/fdt/fdt_fixup.c +10 -0
+		return err;
+
+	cpus_node = fdt_path_offset(fdt, "/cpus");
+	if (cpus_node < 0)
+		return cpus_node;
+
+	/* Create the idle-states node and its child nodes. */
+	idle_states_node = fdt_add_subnode(fdt, cpus_node, "idle-states");
+	if (idle_states_node < 0)
+		return idle_states_node;

It breaks platform final init when the ?idle-states? already exists:

____ _____ ____ _____

  / __ \                  / ____|  _ \_   _|
 | |  | |_ __   ___ _ __ | (___ | |_) || |
 | |  | | '_ \ / _ \ '_ \ \___ \|  _ < | |
 | |__| | |_) |  __/ | | |____) | |_) || |_
  \____/| .__/ \___|_| |_|_____/|____/_____|
        | |
        |_|

init_coldboot: platform final init failed (error -2)

The error code comes from fdt_add_subnode_namelen():
https://github.com/riscv-software-src/opensbi/blob/v1.2/lib/utils/libfdt/fdt_rw.c#L347

should it return 0 here?

Best regards,
Peter Lin

author maintainer
#4 lore raw

PATCH 1/2 lib: utils: Add fdt_add_cpu_idle_states() helper function

Quoted 76 lines · Yu-Chien Peter Lin

Hi Samuel,

Quoted 73 lines · Samuel Holland

Since the availability and latency properties of CPU idle states depend on the specific SBI HSM implementation, it is appropriate that the idle states are added to the devicetree at runtime by that implementation.

This helper function adds a platform-provided array of idle states to the devicetree, following the SBI idle state binding. It makes some assumptions for simplicity, but these could be relaxed if needed.

---

 include/sbi_utils/fdt/fdt_fixup.h | 23 +++++++++
 lib/utils/fdt/fdt_fixup.c         | 80 +++++++++++++++++++++++++++++++
 2 files changed, 103 insertions(+)
include/sbi_utils/fdt/fdt_fixup.h +23 -0
index fb076ba..cab3f0f 100644
@@ -9,6 +9,29 @@
 #ifndef __FDT_FIXUP_H__
 #define __FDT_FIXUP_H__
 
+struct sbi_cpu_idle_state {
+	const char *name;
+	uint32_t suspend_param;
+	bool local_timer_stop;
+	uint32_t entry_latency_us;
+	uint32_t exit_latency_us;
+	uint32_t min_residency_us;
+	uint32_t wakeup_latency_us;
+};
+
+/**
+ * Add CPU idle states to cpu nodes in the DT
+ *
+ * Add information about CPU idle states to the devicetree. This function
+ * assumes that CPU idle states are not already present in the devicetree, and
+ * that all CPU states are equally applicable to all CPUs.
+ *
+ * @param fdt: device tree blob
+ * @param states: array of idle state descriptions, ending with empty element
+ * @return zero on success and -ve on failure
+ */
+int fdt_add_cpu_idle_states(void *dtb, const struct sbi_cpu_idle_state *state);
+
 /**
  * Fix up the CPU node in the device tree
  *
lib/utils/fdt/fdt_fixup.c +16 -0
index 41f6cbb..d9aa0b2 100644
@@ -18,6 +18,86 @@
 #include <sbi_utils/fdt/fdt_pmu.h>
 #include <sbi_utils/fdt/fdt_helper.h>
 
+int fdt_add_cpu_idle_states(void *fdt, const struct sbi_cpu_idle_state *state)
+{
+	int cpu_node, cpus_node, err, idle_states_node;
+	uint32_t count, phandle;
+
+	err = fdt_open_into(fdt, fdt, fdt_totalsize(fdt) + 1024);
+	if (err < 0)
+		return err;
+
+	err = fdt_find_max_phandle(fdt, &phandle);
+	phandle++;
+	if (err < 0)
+		return err;
+
+	cpus_node = fdt_path_offset(fdt, "/cpus");
+	if (cpus_node < 0)
Quoted 8 lines

lib/utils/fdt/fdt_fixup.c line 42

Quoted 6 lines
lib/utils/fdt/fdt_fixup.c +6 -0
+		return cpus_node;
+
+	/* Create the idle-states node and its child nodes. */
+	idle_states_node = fdt_add_subnode(fdt, cpus_node, "idle-states");
+	if (idle_states_node < 0)
+		return idle_states_node;

It breaks platform final init when the ?idle-states? already exists:

Right, that is a documented limitation. I would expect there to be a single source of idle state information -- wherever the HSM implementation is. So my suggestion would be to remove idle-states from the static devicetree if you plan to use this function. If that is not desired, could you explain your scenario in a bit more detail?

Regards,
Samuel

#5 lore raw

PATCH 1/2 lib: utils: Add fdt_add_cpu_idle_states() helper function

Quoted 86 lines · Samuel Holland
Quoted 76 lines · Yu-Chien Peter Lin

Hi Samuel,

Quoted 73 lines · Samuel Holland

Since the availability and latency properties of CPU idle states depend on the specific SBI HSM implementation, it is appropriate that the idle states are added to the devicetree at runtime by that implementation.

This helper function adds a platform-provided array of idle states to the devicetree, following the SBI idle state binding. It makes some assumptions for simplicity, but these could be relaxed if needed.

---

 include/sbi_utils/fdt/fdt_fixup.h | 23 +++++++++
 lib/utils/fdt/fdt_fixup.c         | 80 +++++++++++++++++++++++++++++++
 2 files changed, 103 insertions(+)
include/sbi_utils/fdt/fdt_fixup.h +23 -0
index fb076ba..cab3f0f 100644
@@ -9,6 +9,29 @@
 #ifndef __FDT_FIXUP_H__
 #define __FDT_FIXUP_H__
 
+struct sbi_cpu_idle_state {
+	const char *name;
+	uint32_t suspend_param;
+	bool local_timer_stop;
+	uint32_t entry_latency_us;
+	uint32_t exit_latency_us;
+	uint32_t min_residency_us;
+	uint32_t wakeup_latency_us;
+};
+
+/**
+ * Add CPU idle states to cpu nodes in the DT
+ *
+ * Add information about CPU idle states to the devicetree. This function
+ * assumes that CPU idle states are not already present in the devicetree, and
+ * that all CPU states are equally applicable to all CPUs.
+ *
+ * @param fdt: device tree blob
+ * @param states: array of idle state descriptions, ending with empty element
+ * @return zero on success and -ve on failure
+ */
+int fdt_add_cpu_idle_states(void *dtb, const struct sbi_cpu_idle_state *state);
+
 /**
  * Fix up the CPU node in the device tree
  *
lib/utils/fdt/fdt_fixup.c +16 -0
index 41f6cbb..d9aa0b2 100644
@@ -18,6 +18,86 @@
 #include <sbi_utils/fdt/fdt_pmu.h>
 #include <sbi_utils/fdt/fdt_helper.h>
 
+int fdt_add_cpu_idle_states(void *fdt, const struct sbi_cpu_idle_state *state)
+{
+	int cpu_node, cpus_node, err, idle_states_node;
+	uint32_t count, phandle;
+
+	err = fdt_open_into(fdt, fdt, fdt_totalsize(fdt) + 1024);
+	if (err < 0)
+		return err;
+
+	err = fdt_find_max_phandle(fdt, &phandle);
+	phandle++;
+	if (err < 0)
+		return err;
+
+	cpus_node = fdt_path_offset(fdt, "/cpus");
+	if (cpus_node < 0)
Quoted 8 lines

lib/utils/fdt/fdt_fixup.c line 42

Quoted 6 lines
lib/utils/fdt/fdt_fixup.c +6 -0
+		return cpus_node;
+
+	/* Create the idle-states node and its child nodes. */
+	idle_states_node = fdt_add_subnode(fdt, cpus_node, "idle-states");
+	if (idle_states_node < 0)
+		return idle_states_node;

It breaks platform final init when the ?idle-states? already exists:

Quoted 5 lines

Right, that is a documented limitation. I would expect there to be a single source of idle state information -- wherever the HSM implementation is. So my suggestion would be to remove idle-states from the static devicetree if you plan to use this function. If that is not desired, could you explain your scenario in a bit more detail?

Understood, then I have no more questions.

Thanks,
Peter Lin

Quoted 4 lines

Regards,
Samuel

#6 lore raw

PATCH 1/2 lib: utils: Add fdt_add_cpu_idle_states() helper function

In reply to Samuel Holland
Quoted 87 lines · Samuel Holland
Quoted 76 lines · Yu-Chien Peter Lin

Hi Samuel,

Quoted 73 lines · Samuel Holland

Since the availability and latency properties of CPU idle states depend on the specific SBI HSM implementation, it is appropriate that the idle states are added to the devicetree at runtime by that implementation.

This helper function adds a platform-provided array of idle states to the devicetree, following the SBI idle state binding. It makes some assumptions for simplicity, but these could be relaxed if needed.

---

 include/sbi_utils/fdt/fdt_fixup.h | 23 +++++++++
 lib/utils/fdt/fdt_fixup.c         | 80 +++++++++++++++++++++++++++++++
 2 files changed, 103 insertions(+)
include/sbi_utils/fdt/fdt_fixup.h +23 -0
index fb076ba..cab3f0f 100644
@@ -9,6 +9,29 @@
 #ifndef __FDT_FIXUP_H__
 #define __FDT_FIXUP_H__
 
+struct sbi_cpu_idle_state {
+    const char *name;
+    uint32_t suspend_param;
+    bool local_timer_stop;
+    uint32_t entry_latency_us;
+    uint32_t exit_latency_us;
+    uint32_t min_residency_us;
+    uint32_t wakeup_latency_us;
+};
+
+/**
+ * Add CPU idle states to cpu nodes in the DT
+ *
+ * Add information about CPU idle states to the devicetree. This function
+ * assumes that CPU idle states are not already present in the devicetree, and
+ * that all CPU states are equally applicable to all CPUs.
+ *
+ * @param fdt: device tree blob
+ * @param states: array of idle state descriptions, ending with empty element
+ * @return zero on success and -ve on failure
+ */
+int fdt_add_cpu_idle_states(void *dtb, const struct sbi_cpu_idle_state *state);
+
 /**
  * Fix up the CPU node in the device tree
  *
lib/utils/fdt/fdt_fixup.c +16 -0
index 41f6cbb..d9aa0b2 100644
@@ -18,6 +18,86 @@
 #include <sbi_utils/fdt/fdt_pmu.h>
 #include <sbi_utils/fdt/fdt_helper.h>
 
+int fdt_add_cpu_idle_states(void *fdt, const struct sbi_cpu_idle_state *state)
+{
+    int cpu_node, cpus_node, err, idle_states_node;
+    uint32_t count, phandle;
+
+    err = fdt_open_into(fdt, fdt, fdt_totalsize(fdt) + 1024);
+    if (err < 0)
+            return err;
+
+    err = fdt_find_max_phandle(fdt, &phandle);
+    phandle++;
+    if (err < 0)
+            return err;
+
+    cpus_node = fdt_path_offset(fdt, "/cpus");
+    if (cpus_node < 0)
Quoted 8 lines

lib/utils/fdt/fdt_fixup.c line 42

Quoted 6 lines
lib/utils/fdt/fdt_fixup.c +6 -0
+            return cpus_node;
+
+    /* Create the idle-states node and its child nodes. */
+    idle_states_node = fdt_add_subnode(fdt, cpus_node, "idle-states");
+    if (idle_states_node < 0)
+            return idle_states_node;

It breaks platform final init when the ?idle-states? already exists:

Quoted 5 lines

Right, that is a documented limitation. I would expect there to be a single source of idle state information -- wherever the HSM implementation is. So my suggestion would be to remove idle-states from the static devicetree if you plan to use this function. If that is not desired, could you explain your scenario in a bit more detail?

I assume there are existing boards with DTBs in flash containing "idle-states" DT node. Better to add a check if "idle-states" DT node already exists.

Otherwise, this patch looks good to me.

Regards,
Anup

Quoted 9 lines

Regards,
Samuel

--
opensbi mailing list
opensbi at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/opensbi