Skip to contentSkip to conversation

Introduce SoC sleep stats driver

PATCH Superseded Maulik Shah 3 participants 18 messages May 21, 2021 – Sep 5, 2021

Subsystems (4)

v8 of 12 Superseded Latest is v12 →

lore
Share & tools
lore
#2 lore raw

PATCH v8 2/5 soc: qcom: Add SoC sleep stats driver

In reply to Maulik Shah

From: Mahesh Sivasubramanian <[email protected]>

Let's add a driver to read the stats from remote processor and export to debugfs.

The driver creates "qcom_sleep_stats" directory in debugfs and adds files for various low power mode available. Below is sample output with command

cat /sys/kernel/debug/qcom_sleep_stats/ddr
count = 0
Last Entered At = 0
Last Exited At = 0
Accumulated Duration = 0

[mkshah: add subsystem sleep stats, create one file for each stat]
---
 drivers/soc/qcom/Kconfig           |  10 ++
 drivers/soc/qcom/Makefile          |   1 +
 drivers/soc/qcom/soc_sleep_stats.c | 255 +++++++++++++++++++++++++++++++++++++
 3 files changed, 266 insertions(+)
 create mode 100644 drivers/soc/qcom/soc_sleep_stats.c
drivers/soc/qcom/Kconfig +10 -0
index 79b568f..e80b63a 100644
@@ -190,6 +190,16 @@ config QCOM_SOCINFO
 	 Say yes here to support the Qualcomm socinfo driver, providing
 	 information about the SoC to user space.
 
+config QCOM_SOC_SLEEP_STATS
+	tristate "Qualcomm Technologies, Inc. (QTI) SoC sleep stats driver"
+	depends on ARCH_QCOM && DEBUG_FS || COMPILE_TEST
+	depends on QCOM_SMEM
+	help
+	  Qualcomm Technologies, Inc. (QTI) SoC sleep stats driver to read
+	  the shared memory exported by the remote processor related to
+	  various SoC level low power modes statistics and export to debugfs
+	  interface.
+
 config QCOM_WCNSS_CTRL
 	tristate "Qualcomm WCNSS control driver"
 	depends on ARCH_QCOM || COMPILE_TEST
drivers/soc/qcom/Makefile +1 -0
index ad675a6..5f30d74 100644
@@ -20,6 +20,7 @@ obj-$(CONFIG_QCOM_SMEM_STATE) += smem_state.o
 obj-$(CONFIG_QCOM_SMP2P)	+= smp2p.o
 obj-$(CONFIG_QCOM_SMSM)	+= smsm.o
 obj-$(CONFIG_QCOM_SOCINFO)	+= socinfo.o
+obj-$(CONFIG_QCOM_SOC_SLEEP_STATS)	+= soc_sleep_stats.o
 obj-$(CONFIG_QCOM_WCNSS_CTRL) += wcnss_ctrl.o
 obj-$(CONFIG_QCOM_APR) += apr.o
 obj-$(CONFIG_QCOM_LLCC) += llcc-qcom.o
drivers/soc/qcom/soc_sleep_stats.c new file +255 -0

255 lines not on this page · raw message

Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation
#0 lore raw

v8 cover Introduce SoC sleep stats driver

Changes in v8:
- Addressed bjorn's comments in driver from v7
- Update aoss_qmp device node reg size for sc7280

Changes in v7:
- Fix example in bindings documentation as per #address/size-cells = <1>.
- Add comment in driver from where 'ddr' subsystems name is read.
- Update comment in driver to s/beside/besides and others from v6.
- Rename debugfs_create_entries() from v6.
- Drop use of memcpy_fromio() to find the name.
- Use sizeof(*prv_data) in devm_kzalloc().
- Add change to define readq() if its not yet defined for compile support.
- Add wpss subsystem in the list of subsystems.
- Add module soft dependency on smem module.
- Add new change to add device node for sc7280.

Changes in v6:
- Address stephen's comments from v5 which includes below
- Pad 0 in documentation example to make address 8 digit
- define macro to calculate offset in driver
- Add appended_stats_avail to prv_data instead of using entire stats_config
- make array subsystems[] as const
- Add comment for SSR case
- Use memcpy_fromio() and devm_kcalloc() during probe
- Change file permission mode from 444 to 400

- Address guenter's comments to add depends on QCOM_SMEM

- Add adsp_island and cdsp_island subsystems
- Use strim() to remove whitespace in stat name

Changes in v5:
- Remove underscore from node name in Documentation and DTSI change
- Remove global config from driver change

Changes in v4:
- Address bjorn's comments from v3 on change 2.
- Add bjorn's Reviewed-by on change 3 and 4.

Changes in v3:
- Address stephen's comments from v2 in change 1 and 2.
- Address bjorn's comments from v2 in change 3 and 4.
- Add Rob and bjorn's Reviewed-by on YAML change.

Changes in v2:
- Convert Documentation to YAML.
- Address stephen's comments from v1.
- Use debugfs instead of sysfs.
- Add sc7180 dts changes for sleep stats
- Add defconfig changes to enable driver
- Include subsystem stats from [1] in this single stats driver.
- Address stephen's comments from [1]
- Update cover letter inline to mention [1]

Qualcomm Technologies, Inc. (QTI)'s chipsets support SoC level low power modes. SoCs Always On Processor/Resource Power Manager produces statistics of the SoC sleep modes involving lowering or powering down of the rails and the oscillator clock.

Additionally multiple subsystems present on SoC like modem, spss, adsp, cdsp maintains their low power mode statistics in shared memory (SMEM).

Statistics includes SoC sleep mode type, number of times LPM entered, time of last entry, exit, and accumulated sleep duration in seconds.

This series adds a driver to read the stats and export to debugfs.

[1] https://lore.kernel.org/patchwork/patch/1149381/

Mahesh Sivasubramanian (2):
dt-bindings: Introduce SoC sleep stats bindings
soc: qcom: Add SoC sleep stats driver

Maulik Shah (3):
arm64: dts: qcom: sc7180: Enable SoC sleep stats
arm64: defconfig: Enable SoC sleep stats driver
arm64: dts: qcom: sc7280: Enable SoC sleep stats

 .../bindings/soc/qcom/soc-sleep-stats.yaml         |  48 ++++
 arch/arm64/boot/dts/qcom/sc7180.dtsi               |   7 +-
 arch/arm64/boot/dts/qcom/sc7280.dtsi               |   7 +-
 arch/arm64/configs/defconfig                       |   1 +
 drivers/soc/qcom/Kconfig                           |  10 +
 drivers/soc/qcom/Makefile                          |   1 +
 drivers/soc/qcom/soc_sleep_stats.c                 | 255 +++++++++++++++++++++
 7 files changed, 327 insertions(+), 2 deletions(-)
 create mode 100644 Documentation/devicetree/bindings/soc/qcom/soc-sleep-stats.yaml
 create mode 100644 drivers/soc/qcom/soc_sleep_stats.c
Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation
#16 lore raw

PATCH v8 5/5 arm64: dts: qcom: sc7280: Enable SoC sleep stats

Add device node for SoC sleep stats driver which provides various low power mode stats.

Also update the reg size of aoss_qmp device to 0x400.

---
 arch/arm64/boot/dts/qcom/sc7280.dtsi | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
arch/arm64/boot/dts/qcom/sc7280.dtsi +6 -1
index 2cc4785..47afca8 100644
@@ -827,7 +827,7 @@
 
 		aoss_qmp: power-controller@c300000 {
 			compatible = "qcom,sc7280-aoss-qmp";
-			reg = <0 0x0c300000 0 0x100000>;
+			reg = <0 0x0c300000 0 0x400>;
 			interrupts-extended = <&ipcc IPCC_CLIENT_AOP
 						     IPCC_MPROC_SIGNAL_GLINK_QMP
 						     IRQ_TYPE_EDGE_RISING>;
@@ -838,6 +838,11 @@
 			#power-domain-cells = <1>;
 		};
 
+		rpmh-sleep-stats@c3f0000 {
+			compatible = "qcom,rpmh-sleep-stats";
+			reg = <0 0x0c3f0000 0 0x400>;
+		};
+
 		spmi_bus: spmi@c440000 {
 			compatible = "qcom,spmi-pmic-arb";
 			reg = <0 0x0c440000 0 0x1100>,
Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation
#1 lore raw

PATCH v8 1/5 dt-bindings: Introduce SoC sleep stats bindings

In reply to Maulik Shah

From: Mahesh Sivasubramanian <[email protected]>

Add device binding documentation for Qualcomm Technologies, Inc. (QTI) SoC sleep stats driver. The driver is used for displaying SoC sleep statistic maintained by Always On Processor or Resource Power Manager.

---
 .../bindings/soc/qcom/soc-sleep-stats.yaml         | 48 ++++++++++++++++++++++
 1 file changed, 48 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/soc/qcom/soc-sleep-stats.yaml
Documentation/devicetree/bindings/soc/qcom/soc-sleep-stats.yaml new file +48 -0
new file mode 100644
index 0000000..9078c4f
@@ -0,0 +1,48 @@
+# SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/soc/qcom/soc-sleep-stats.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: Qualcomm Technologies, Inc. (QTI) SoC sleep stats bindings
+
+maintainers:
+  - Maulik Shah <[email protected]>
+  - Lina Iyer <[email protected]>
+
+description:
+  Always On Processor/Resource Power Manager maintains statistics of the SoC
+  sleep modes involving powering down of the rails and oscillator clock.
+
+  Statistics includes SoC sleep mode type, number of times low power mode were
+  entered, time of last entry, time of last exit and accumulated sleep duration.
+
+properties:
+  compatible:
+    enum:
+      - qcom,rpmh-sleep-stats
+      - qcom,rpm-sleep-stats
+
+  reg:
+    maxItems: 1
+
+required:
+  - compatible
+  - reg
+
+additionalProperties: false
+
+examples:
+  # Example of rpmh sleep stats
+  - |
+    rpmh-sleep-stats@c3f0000 {
+      compatible = "qcom,rpmh-sleep-stats";
+      reg = <0x0c3f0000 0x400>;
+    };
+  # Example of rpm sleep stats
+  - |
+    rpm-sleep-stats@4690000 {
+      compatible = "qcom,rpm-sleep-stats";
+      reg = <0x04690000 0x400>;
+    };
+...
Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation
#6 lore raw

PATCH v8 3/5 arm64: dts: qcom: sc7180: Enable SoC sleep stats

In reply to Maulik Shah

Add device node for SoC sleep stats driver which provides various low power mode stats.

Also update the reg size of aoss_qmp device to 0x400.

---
 arch/arm64/boot/dts/qcom/sc7180.dtsi | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
arch/arm64/boot/dts/qcom/sc7180.dtsi +6 -1
index 6228ba2..889d04d 100644
@@ -3215,7 +3215,7 @@
 
 		aoss_qmp: power-controller@c300000 {
 			compatible = "qcom,sc7180-aoss-qmp";
-			reg = <0 0x0c300000 0 0x100000>;
+			reg = <0 0x0c300000 0 0x400>;
 			interrupts = <GIC_SPI 389 IRQ_TYPE_EDGE_RISING>;
 			mboxes = <&apss_shared 0>;
 
@@ -3223,6 +3223,11 @@
 			#power-domain-cells = <1>;
 		};
 
+		rpmh-sleep-stats@c3f0000 {
+			compatible = "qcom,rpmh-sleep-stats";
+			reg = <0 0x0c3f0000 0 0x400>;
+		};
+
 		spmi_bus: spmi@c440000 {
 			compatible = "qcom,spmi-pmic-arb";
 			reg = <0 0x0c440000 0 0x1100>,
Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation
#14 lore raw

PATCH v8 4/5 arm64: defconfig: Enable SoC sleep stats driver

In reply to Maulik Shah

Enable SoC sleep stats driver. The driver gives statistics for various low power modes on Qualcomm Technologies, Inc. (QTI) SoCs.

---
 arch/arm64/configs/defconfig | 1 +
 1 file changed, 1 insertion(+)
arch/arm64/configs/defconfig +1 -0
index 45db77d..0c1cad7 100644
@@ -1029,6 +1029,7 @@ CONFIG_QCOM_SMD_RPM=y
 CONFIG_QCOM_SMP2P=y
 CONFIG_QCOM_SMSM=y
 CONFIG_QCOM_SOCINFO=m
+CONFIG_QCOM_SOC_SLEEP_STATS=m
 CONFIG_QCOM_APR=m
 CONFIG_ARCH_R8A774A1=y
 CONFIG_ARCH_R8A774B1=y
Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation
#7 lore raw

Re: [PATCH v8 3/5] arm64: dts: qcom: sc7180: Enable SoC sleep stats

In reply to Maulik Shah

Quoting Maulik Shah (2021-05-21 04:26:09)

Quoted 26 lines

Add device node for SoC sleep stats driver which provides various low power mode stats.

Also update the reg size of aoss_qmp device to 0x400.

---
 arch/arm64/boot/dts/qcom/sc7180.dtsi | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
arch/arm64/boot/dts/qcom/sc7180.dtsi +1 -1
index 6228ba2..889d04d 100644
@@ -3215,7 +3215,7 @@
 
                aoss_qmp: power-controller@c300000 {
                        compatible = "qcom,sc7180-aoss-qmp";
-                       reg = <0 0x0c300000 0 0x100000>;
+                       reg = <0 0x0c300000 0 0x400>;
                        interrupts = <GIC_SPI 389 IRQ_TYPE_EDGE_RISING>;
                        mboxes = <&apss_shared 0>;
 

arch/arm64/boot/dts/qcom/sc7180.dtsi line 3230

Quoted 8 lines
arch/arm64/boot/dts/qcom/sc7180.dtsi +5 -0
@@ -3223,6 +3223,11 @@
                        #power-domain-cells = <1>;
                };
 
+               rpmh-sleep-stats@c3f0000 {
+                       compatible = "qcom,rpmh-sleep-stats";
+                       reg = <0 0x0c3f0000 0 0x400>;
+               };
+

Does this need to be in DT? Can the sc7180-aoss-qmp driver use the aux-bus and stick the sleep stats device on there?

Quoted 3 lines
arch/arm64/boot/dts/qcom/sc7180.dtsi +0 -0
                spmi_bus: spmi@c440000 {
                        compatible = "qcom,spmi-pmic-arb";
                        reg = <0 0x0c440000 0 0x1100>,
#17 lore raw

Re: [PATCH v8 5/5] arm64: dts: qcom: sc7280: Enable SoC sleep stats

In reply to Maulik Shah

Quoting Maulik Shah (2021-05-21 04:26:11)

Quoted 10 lines

Add device node for SoC sleep stats driver which provides various low power mode stats.

Also update the reg size of aoss_qmp device to 0x400.

---
 arch/arm64/boot/dts/qcom/sc7280.dtsi | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

Same question as in sc7180.

Quoted 30 lines
arch/arm64/boot/dts/qcom/sc7280.dtsi +6 -2
index 2cc4785..47afca8 100644
@@ -827,7 +827,7 @@
 
                aoss_qmp: power-controller@c300000 {
                        compatible = "qcom,sc7280-aoss-qmp";
-                       reg = <0 0x0c300000 0 0x100000>;
+                       reg = <0 0x0c300000 0 0x400>;
                        interrupts-extended = <&ipcc IPCC_CLIENT_AOP
                                                     IPCC_MPROC_SIGNAL_GLINK_QMP
                                                     IRQ_TYPE_EDGE_RISING>;
@@ -838,6 +838,11 @@
                        #power-domain-cells = <1>;
                };
 
+               rpmh-sleep-stats@c3f0000 {
+                       compatible = "qcom,rpmh-sleep-stats";
+                       reg = <0 0x0c3f0000 0 0x400>;
+               };
+
                spmi_bus: spmi@c440000 {
                        compatible = "qcom,spmi-pmic-arb";
                        reg = <0 0x0c440000 0 0x1100>,
--

QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation

#3 lore raw

Re: [PATCH v8 2/5] soc: qcom: Add SoC sleep stats driver

In reply to Maulik Shah

Quoting Maulik Shah (2021-05-21 04:26:08)

Quoted 54 lines
drivers/soc/qcom/soc_sleep_stats.c new file +48 -0
new file mode 100644
index 0000000..a8396f9
@@ -0,0 +1,255 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (c) 2011-2021, The Linux Foundation. All rights reserved.
+ */
+
+#include <linux/debugfs.h>
+#include <linux/device.h>
+#include <linux/io.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/platform_device.h>
+#include <linux/seq_file.h>
+
+#include <linux/soc/qcom/smem.h>
+#include <clocksource/arm_arch_timer.h>
+
+#define STAT_TYPE_OFFSET       0x0
+#define COUNT_OFFSET           0x4
+#define LAST_ENTERED_AT_OFFSET 0x8
+#define LAST_EXITED_AT_OFFSET  0x10
+#define ACCUMULATED_OFFSET     0x18
+#define CLIENT_VOTES_OFFSET    0x1c
+
+struct subsystem_data {
+       const char *name;
+       u32 smem_item;
+       u32 pid;
+};
+
+static const struct subsystem_data subsystems[] = {
+       { "modem", 605, 1 },
+       { "wpss", 605, 13 },
+       { "adsp", 606, 2 },
+       { "cdsp", 607, 5 },
+       { "slpi", 608, 3 },
+       { "gpu", 609, 0 },
+       { "display", 610, 0 },
+       { "adsp_island", 613, 2 },
+       { "slpi_island", 613, 3 },
+};
+
+struct stats_config {
+       size_t offset_addr;
+       size_t num_records;
+       bool appended_stats_avail;
+};
+
+struct stats_data {

drivers/soc/qcom/soc_sleep_stats.c line 58

Quoted 10 lines
drivers/soc/qcom/soc_sleep_stats.c new file +10 -0
+       bool appended_stats_avail;
+       void __iomem *base;
+};
+
+struct sleep_stats {
+       u32 stat_type;
+       u32 count;
+       u64 last_entered_at;
+       u64 last_exited_at;
+       u64 accumulated;

Is this coming from memory directly? So we need a bunch of __le32 and __le64 markings and then the endian swaps?

drivers/soc/qcom/soc_sleep_stats.c line 63

Quoted 5 lines
drivers/soc/qcom/soc_sleep_stats.c new file +5 -0
+};
+
+struct appended_stats {
+       u32 client_votes;
+       u32 reserved[3];

This one too? It's stored in smem?

drivers/soc/qcom/soc_sleep_stats.c line 74

Quoted 11 lines
drivers/soc/qcom/soc_sleep_stats.c new file +11 -0
+};
+
+static void qcom_print_stats(struct seq_file *s, const struct sleep_stats *stat)
+{
+       u64 accumulated = stat->accumulated;
+       /*
+        * If a subsystem is in sleep when reading the sleep stats adjust
+        * the accumulated sleep duration to show actual sleep time.
+        */
+       if (stat->last_entered_at > stat->last_exited_at)
+               accumulated += arch_timer_read_counter() - stat->last_entered_at;

Doesn't this read the virtual count? Which could be very different than the physical count that the subsystems see and then write into smem? I suppose we need some sort of "physical count" reading API to match up the timestamps that are there but that is sort of hard given that the physical count isn't always intended to be read by the kernel. Maybe we should split off an MMIO architected timer to be used by this code that uses the same clocksource that the other subsystems are using?

Quoted 6 lines
drivers/soc/qcom/soc_sleep_stats.c new file +6 -0
+
+       seq_printf(s, "Count: %u\n", stat->count);
+       seq_printf(s, "Last Entered At: %llu\n", stat->last_entered_at);
+       seq_printf(s, "Last Exited At: %llu\n", stat->last_exited_at);
+       seq_printf(s, "Accumulated Duration: %llu\n", accumulated);
+}

drivers/soc/qcom/soc_sleep_stats.c line 90

Quoted 10 lines
drivers/soc/qcom/soc_sleep_stats.c new file +10 -0
+
+static int qcom_subsystem_sleep_stats_show(struct seq_file *s, void *unused)
+{
+       struct subsystem_data *subsystem = s->private;
+       struct sleep_stats *stat;
+
+       /* Items are allocated lazily, so lookup pointer each time */
+       stat = qcom_smem_get(subsystem->pid, subsystem->smem_item, NULL);
+       if (IS_ERR(stat))
+               return PTR_ERR(stat);

Can this return EPROBE_DEFER to userspace? That would be pretty weird. Maybe we should convert the return value to -EIO regardless of the error.

Quoted 30 lines
drivers/soc/qcom/soc_sleep_stats.c new file +30 -0
+
+       qcom_print_stats(s, stat);
+
+       return 0;
+}
+
+static int qcom_soc_sleep_stats_show(struct seq_file *s, void *unused)
+{
+       struct stats_data *d = s->private;
+       void __iomem *reg = d->base;
+       struct sleep_stats stat;
+
+       memcpy_fromio(&stat, reg, sizeof(stat));
+       qcom_print_stats(s, &stat);
+
+       if (d->appended_stats_avail) {
+               struct appended_stats votes;
+
+               memcpy_fromio(&votes, reg + CLIENT_VOTES_OFFSET, sizeof(votes));
+               seq_printf(s, "Client Votes: %#x\n", votes.client_votes);
+       }
+
+       return 0;
+}
+
+DEFINE_SHOW_ATTRIBUTE(qcom_soc_sleep_stats);
+DEFINE_SHOW_ATTRIBUTE(qcom_subsystem_sleep_stats);
+
+static void qcom_create_soc_sleep_stat_files(struct dentry *root, void __iomem *reg,
+                                            struct stats_data *d, u32 num_records)

drivers/soc/qcom/soc_sleep_stats.c line 130

Quoted 10 lines
drivers/soc/qcom/soc_sleep_stats.c new file +10 -0
+{
+       char stat_type[sizeof(u32) + 1] = {0};
+       u32 offset = 0, type;
+       int i, j;
+
+       for (i = 0; i < num_records; i++) {
+               d[i].base = reg + offset;
+
+               /*
+                * Read the low power mode name and create debugfs file for it.

What do we do if there are two same named low power modes? Maybe they should be directories instead with pretty name and values inside of them.

Quoted 13 lines
drivers/soc/qcom/soc_sleep_stats.c new file +13 -0
+                * The names read could be of below,
+                * (may change depending on low power mode supported).
+                * For rpmh-sleep-stats: "aosd", "cxsd" and "ddr".
+                * For rpm-sleep-stats: "vmin" and "vlow".
+                */
+               type = readl(d[i].base);
+               for (j = 0; j < sizeof(u32); j++) {
+                       stat_type[j] = type & 0xff;
+                       type = type >> 8;
+               }
+               strim(stat_type);
+               debugfs_create_file(stat_type, 0400, root, &d[i],
+                                   &qcom_soc_sleep_stats_fops);

drivers/soc/qcom/soc_sleep_stats.c line 153

Quoted 10 lines
drivers/soc/qcom/soc_sleep_stats.c new file +10 -0
+
+               offset += sizeof(struct sleep_stats);
+               if (d[i].appended_stats_avail)
+                       offset += sizeof(struct appended_stats);
+       }
+}
+
+static void qcom_create_subsystem_stat_files(struct dentry *root)
+{
+       struct sleep_stats *stat;

const? The pointer isn't used though so is this even relevant?

drivers/soc/qcom/soc_sleep_stats.c line 161

Quoted 8 lines
drivers/soc/qcom/soc_sleep_stats.c new file +8 -0
+       int i;
+
+       for (i = 0; i < ARRAY_SIZE(subsystems); i++) {
+               stat = qcom_smem_get(subsystems[i].pid, subsystems[i].smem_item, NULL);
+               if (IS_ERR(stat))
+                       continue;
+
+               debugfs_create_file(subsystems[i].name, 0400, root, (void *)&subsystems[i],

Do we need to cast to void pointer? Is it unused? If it's unused then pass NULL?

Quoted 15 lines
drivers/soc/qcom/soc_sleep_stats.c new file +15 -0
+                                   &qcom_subsystem_sleep_stats_fops);
+       }
+}
+
+static int qcom_soc_sleep_stats_probe(struct platform_device *pdev)
+{
+       struct resource *res;
+       void __iomem *reg;
+       void __iomem *offset_addr;
+       phys_addr_t stats_base;
+       resource_size_t stats_size;
+       struct dentry *root;
+       const struct stats_config *config;
+       struct stats_data *d;
+       int i;

drivers/soc/qcom/soc_sleep_stats.c line 186

Quoted 10 lines
drivers/soc/qcom/soc_sleep_stats.c new file +10 -0
+
+       config = device_get_match_data(&pdev->dev);
+       if (!config)
+               return -ENODEV;
+
+       res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       if (!res)
+               return PTR_ERR(res);
+
+       offset_addr = ioremap(res->start + config->offset_addr, sizeof(u32));

Why can't we devm_platform_get_and_ioremap_resource()?

drivers/soc/qcom/soc_sleep_stats.c line 192

Quoted 6 lines
drivers/soc/qcom/soc_sleep_stats.c new file +6 -0
+       if (IS_ERR(offset_addr))
+               return PTR_ERR(offset_addr);
+
+       stats_base = res->start | readl_relaxed(offset_addr);
+       stats_size = resource_size(res);
+       iounmap(offset_addr);

Oh because of unmap? Can we add a resource for stats_base/size to the system? I'd really like to see what address that is at. I think it is actually within the same region of this "pointer" that is assigned in DT, but I honestly don't know why we wouldn't just hardcode the address in DT to begin with instead of doing this map, find the real address, and then map that. Is this really changing that often? Probably it changes a few times during development but then it settles down and we can just hardcode it in DT? At least please make a resource and reserve it so we know that the region isn't being used by other kernel drivers.

Quoted 41 lines
drivers/soc/qcom/soc_sleep_stats.c new file +41 -0
+
+       reg = devm_ioremap(&pdev->dev, stats_base, stats_size);
+       if (!reg)
+               return -ENOMEM;
+
+       d = devm_kcalloc(&pdev->dev, config->num_records,
+                        sizeof(*d), GFP_KERNEL);
+       if (!d)
+               return -ENOMEM;
+
+       for (i = 0; i < config->num_records; i++)
+               d[i].appended_stats_avail = config->appended_stats_avail;
+
+       root = debugfs_create_dir("qcom_sleep_stats", NULL);
+
+       qcom_create_subsystem_stat_files(root);
+       qcom_create_soc_sleep_stat_files(root, reg, d, config->num_records);
+
+       platform_set_drvdata(pdev, root);
+
+       return 0;
+}
+
+static int qcom_soc_sleep_stats_remove(struct platform_device *pdev)
+{
+       struct dentry *root = platform_get_drvdata(pdev);
+
+       debugfs_remove_recursive(root);
+
+       return 0;
+}
+
+static const struct stats_config rpm_data = {
+       .offset_addr = 0x14,
+       .num_records = 2,
+       .appended_stats_avail = true,
+};
+
+static const struct stats_config rpmh_data = {
+       .offset_addr = 0x4,
+       .num_records = 3,

drivers/soc/qcom/soc_sleep_stats.c line 243

Quoted 10 lines
drivers/soc/qcom/soc_sleep_stats.c new file +10 -0
+       .appended_stats_avail = false,
+};
+
+static const struct of_device_id qcom_soc_sleep_stats_table[] = {
+       { .compatible = "qcom,rpm-sleep-stats", .data = &rpm_data },
+       { .compatible = "qcom,rpmh-sleep-stats", .data = &rpmh_data },
+       { }
+};
+
+static struct platform_driver soc_sleep_stats_driver = {

A driver seems pretty heavyweight for a thing that sticks stuff into debugfs. It's not "driving" anything.

Quoted 12 lines
drivers/soc/qcom/soc_sleep_stats.c new file +12 -0
+       .probe = qcom_soc_sleep_stats_probe,
+       .remove = qcom_soc_sleep_stats_remove,
+       .driver = {
+               .name = "soc_sleep_stats",
+               .of_match_table = qcom_soc_sleep_stats_table,
+       },
+};
+module_platform_driver(soc_sleep_stats_driver);
+
+MODULE_DESCRIPTION("Qualcomm Technologies, Inc. (QTI) SoC Sleep Stats driver");
+MODULE_LICENSE("GPL v2");
+MODULE_SOFTDEP("pre: smem");
#5 lore raw

Re: [PATCH v8 2/5] soc: qcom: Add SoC sleep stats driver

In reply to Maulik Shah
Quoted 310 lines · Maulik Shah

From: Mahesh Sivasubramanian <[email protected]>

Let's add a driver to read the stats from remote processor and export to debugfs.

The driver creates "qcom_sleep_stats" directory in debugfs and adds files for various low power mode available. Below is sample output with command

cat /sys/kernel/debug/qcom_sleep_stats/ddr
count = 0
Last Entered At = 0
Last Exited At = 0
Accumulated Duration = 0

[mkshah: add subsystem sleep stats, create one file for each stat]
---
 drivers/soc/qcom/Kconfig           |  10 ++
 drivers/soc/qcom/Makefile          |   1 +
 drivers/soc/qcom/soc_sleep_stats.c | 255 +++++++++++++++++++++++++++++++++++++
 3 files changed, 266 insertions(+)
 create mode 100644 drivers/soc/qcom/soc_sleep_stats.c
drivers/soc/qcom/Kconfig +10 -0
index 79b568f..e80b63a 100644
@@ -190,6 +190,16 @@ config QCOM_SOCINFO
 	 Say yes here to support the Qualcomm socinfo driver, providing
 	 information about the SoC to user space.
 
+config QCOM_SOC_SLEEP_STATS
+	tristate "Qualcomm Technologies, Inc. (QTI) SoC sleep stats driver"
+	depends on ARCH_QCOM && DEBUG_FS || COMPILE_TEST
+	depends on QCOM_SMEM
+	help
+	  Qualcomm Technologies, Inc. (QTI) SoC sleep stats driver to read
+	  the shared memory exported by the remote processor related to
+	  various SoC level low power modes statistics and export to debugfs
+	  interface.
+
 config QCOM_WCNSS_CTRL
 	tristate "Qualcomm WCNSS control driver"
 	depends on ARCH_QCOM || COMPILE_TEST
drivers/soc/qcom/Makefile +1 -0
index ad675a6..5f30d74 100644
@@ -20,6 +20,7 @@ obj-$(CONFIG_QCOM_SMEM_STATE) += smem_state.o
 obj-$(CONFIG_QCOM_SMP2P)	+= smp2p.o
 obj-$(CONFIG_QCOM_SMSM)	+= smsm.o
 obj-$(CONFIG_QCOM_SOCINFO)	+= socinfo.o
+obj-$(CONFIG_QCOM_SOC_SLEEP_STATS)	+= soc_sleep_stats.o
 obj-$(CONFIG_QCOM_WCNSS_CTRL) += wcnss_ctrl.o
 obj-$(CONFIG_QCOM_APR) += apr.o
 obj-$(CONFIG_QCOM_LLCC) += llcc-qcom.o
drivers/soc/qcom/soc_sleep_stats.c new file +245 -0
new file mode 100644
index 0000000..a8396f9
@@ -0,0 +1,255 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (c) 2011-2021, The Linux Foundation. All rights reserved.
+ */
+
+#include <linux/debugfs.h>
+#include <linux/device.h>
+#include <linux/io.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/platform_device.h>
+#include <linux/seq_file.h>
+
+#include <linux/soc/qcom/smem.h>
+#include <clocksource/arm_arch_timer.h>
+
+#define STAT_TYPE_OFFSET	0x0
+#define COUNT_OFFSET		0x4
+#define LAST_ENTERED_AT_OFFSET	0x8
+#define LAST_EXITED_AT_OFFSET	0x10
+#define ACCUMULATED_OFFSET	0x18
+#define CLIENT_VOTES_OFFSET	0x1c
+
+struct subsystem_data {
+	const char *name;
+	u32 smem_item;
+	u32 pid;
+};
+
+static const struct subsystem_data subsystems[] = {
+	{ "modem", 605, 1 },
+	{ "wpss", 605, 13 },
+	{ "adsp", 606, 2 },
+	{ "cdsp", 607, 5 },
+	{ "slpi", 608, 3 },
+	{ "gpu", 609, 0 },
+	{ "display", 610, 0 },
+	{ "adsp_island", 613, 2 },
+	{ "slpi_island", 613, 3 },
+};
+
+struct stats_config {
+	size_t offset_addr;
+	size_t num_records;
+	bool appended_stats_avail;
+};
+
+struct stats_data {
+	bool appended_stats_avail;
+	void __iomem *base;
+};
+
+struct sleep_stats {
+	u32 stat_type;
+	u32 count;
+	u64 last_entered_at;
+	u64 last_exited_at;
+	u64 accumulated;
+};
+
+struct appended_stats {
+	u32 client_votes;
+	u32 reserved[3];
+};
+
+static void qcom_print_stats(struct seq_file *s, const struct sleep_stats *stat)
+{
+	u64 accumulated = stat->accumulated;
+	/*
+	 * If a subsystem is in sleep when reading the sleep stats adjust
+	 * the accumulated sleep duration to show actual sleep time.
+	 */
+	if (stat->last_entered_at > stat->last_exited_at)
+		accumulated += arch_timer_read_counter() - stat->last_entered_at;
+
+	seq_printf(s, "Count: %u\n", stat->count);
+	seq_printf(s, "Last Entered At: %llu\n", stat->last_entered_at);
+	seq_printf(s, "Last Exited At: %llu\n", stat->last_exited_at);
+	seq_printf(s, "Accumulated Duration: %llu\n", accumulated);
+}
+
+static int qcom_subsystem_sleep_stats_show(struct seq_file *s, void *unused)
+{
+	struct subsystem_data *subsystem = s->private;
+	struct sleep_stats *stat;
+
+	/* Items are allocated lazily, so lookup pointer each time */
+	stat = qcom_smem_get(subsystem->pid, subsystem->smem_item, NULL);
+	if (IS_ERR(stat))
+		return PTR_ERR(stat);
+
+	qcom_print_stats(s, stat);
+
+	return 0;
+}
+
+static int qcom_soc_sleep_stats_show(struct seq_file *s, void *unused)
+{
+	struct stats_data *d = s->private;
+	void __iomem *reg = d->base;
+	struct sleep_stats stat;
+
+	memcpy_fromio(&stat, reg, sizeof(stat));
+	qcom_print_stats(s, &stat);
+
+	if (d->appended_stats_avail) {
+		struct appended_stats votes;
+
+		memcpy_fromio(&votes, reg + CLIENT_VOTES_OFFSET, sizeof(votes));
+		seq_printf(s, "Client Votes: %#x\n", votes.client_votes);
+	}
+
+	return 0;
+}
+
+DEFINE_SHOW_ATTRIBUTE(qcom_soc_sleep_stats);
+DEFINE_SHOW_ATTRIBUTE(qcom_subsystem_sleep_stats);
+
+static void qcom_create_soc_sleep_stat_files(struct dentry *root, void __iomem *reg,
+					     struct stats_data *d, u32 num_records)
+{
+	char stat_type[sizeof(u32) + 1] = {0};
+	u32 offset = 0, type;
+	int i, j;
+
+	for (i = 0; i < num_records; i++) {
+		d[i].base = reg + offset;
+
+		/*
+		 * Read the low power mode name and create debugfs file for it.
+		 * The names read could be of below,
+		 * (may change depending on low power mode supported).
+		 * For rpmh-sleep-stats: "aosd", "cxsd" and "ddr".
+		 * For rpm-sleep-stats: "vmin" and "vlow".
+		 */
+		type = readl(d[i].base);
+		for (j = 0; j < sizeof(u32); j++) {
+			stat_type[j] = type & 0xff;
+			type = type >> 8;
+		}
+		strim(stat_type);
+		debugfs_create_file(stat_type, 0400, root, &d[i],
+				    &qcom_soc_sleep_stats_fops);
+
+		offset += sizeof(struct sleep_stats);
+		if (d[i].appended_stats_avail)
+			offset += sizeof(struct appended_stats);
+	}
+}
+
+static void qcom_create_subsystem_stat_files(struct dentry *root)
+{
+	struct sleep_stats *stat;
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(subsystems); i++) {
+		stat = qcom_smem_get(subsystems[i].pid, subsystems[i].smem_item, NULL);
+		if (IS_ERR(stat))
+			continue;
+
+		debugfs_create_file(subsystems[i].name, 0400, root, (void *)&subsystems[i],
+				    &qcom_subsystem_sleep_stats_fops);
+	}
+}
+
+static int qcom_soc_sleep_stats_probe(struct platform_device *pdev)
+{
+	struct resource *res;
+	void __iomem *reg;
+	void __iomem *offset_addr;
+	phys_addr_t stats_base;
+	resource_size_t stats_size;
+	struct dentry *root;
+	const struct stats_config *config;
+	struct stats_data *d;
+	int i;
+
+	config = device_get_match_data(&pdev->dev);
+	if (!config)
+		return -ENODEV;
+
+	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	if (!res)
+		return PTR_ERR(res);
+
+	offset_addr = ioremap(res->start + config->offset_addr, sizeof(u32));
+	if (IS_ERR(offset_addr))
+		return PTR_ERR(offset_addr);
+
+	stats_base = res->start | readl_relaxed(offset_addr);
+	stats_size = resource_size(res);
+	iounmap(offset_addr);
+
+	reg = devm_ioremap(&pdev->dev, stats_base, stats_size);
+	if (!reg)
+		return -ENOMEM;
+
+	d = devm_kcalloc(&pdev->dev, config->num_records,
+			 sizeof(*d), GFP_KERNEL);
+	if (!d)
+		return -ENOMEM;
+
+	for (i = 0; i < config->num_records; i++)
+		d[i].appended_stats_avail = config->appended_stats_avail;
+
+	root = debugfs_create_dir("qcom_sleep_stats", NULL);
+
+	qcom_create_subsystem_stat_files(root);
+	qcom_create_soc_sleep_stat_files(root, reg, d, config->num_records);
+
+	platform_set_drvdata(pdev, root);
+
+	return 0;
+}
+
+static int qcom_soc_sleep_stats_remove(struct platform_device *pdev)
+{
+	struct dentry *root = platform_get_drvdata(pdev);
+
+	debugfs_remove_recursive(root);
+
+	return 0;
+}
+
+static const struct stats_config rpm_data = {
+	.offset_addr = 0x14,
+	.num_records = 2,
+	.appended_stats_avail = true,
+};
+
+static const struct stats_config rpmh_data = {
+	.offset_addr = 0x4,
+	.num_records = 3,
+	.appended_stats_avail = false,
+};
+
+static const struct of_device_id qcom_soc_sleep_stats_table[] = {
+	{ .compatible = "qcom,rpm-sleep-stats", .data = &rpm_data },
+	{ .compatible = "qcom,rpmh-sleep-stats", .data = &rpmh_data },
+	{ }
+};
+
+static struct platform_driver soc_sleep_stats_driver = {
+	.probe = qcom_soc_sleep_stats_probe,
+	.remove = qcom_soc_sleep_stats_remove,

drivers/soc/qcom/soc_sleep_stats.c line 255

Quoted 10 lines
drivers/soc/qcom/soc_sleep_stats.c new file +10 -0
+	.driver = {
+		.name = "soc_sleep_stats",
+		.of_match_table = qcom_soc_sleep_stats_table,
+	},
+};
+module_platform_driver(soc_sleep_stats_driver);
+
+MODULE_DESCRIPTION("Qualcomm Technologies, Inc. (QTI) SoC Sleep Stats driver");
+MODULE_LICENSE("GPL v2");
+MODULE_SOFTDEP("pre: smem");

I still think the qcom_smem_* function calls results in a harder dependency than this softdep.

Regards,
Bjorn

#8 lore raw

Re: [PATCH v8 3/5] arm64: dts: qcom: sc7180: Enable SoC sleep stats

In reply to Stephen Boyd
Quoted 30 lines · Stephen Boyd

Quoting Maulik Shah (2021-05-21 04:26:09)

Quoted 29 lines

Add device node for SoC sleep stats driver which provides various low power mode stats.

Also update the reg size of aoss_qmp device to 0x400.

---
 arch/arm64/boot/dts/qcom/sc7180.dtsi | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
arch/arm64/boot/dts/qcom/sc7180.dtsi +1 -1
index 6228ba2..889d04d 100644
@@ -3215,7 +3215,7 @@
 
                aoss_qmp: power-controller@c300000 {
                        compatible = "qcom,sc7180-aoss-qmp";
-                       reg = <0 0x0c300000 0 0x100000>;
+                       reg = <0 0x0c300000 0 0x400>;
                        interrupts = <GIC_SPI 389 IRQ_TYPE_EDGE_RISING>;
                        mboxes = <&apss_shared 0>;
 
@@ -3223,6 +3223,11 @@
                        #power-domain-cells = <1>;
                };
 
Quoted 9 lines

arch/arm64/boot/dts/qcom/sc7180.dtsi line 3230

Quoted 5 lines
arch/arm64/boot/dts/qcom/sc7180.dtsi +5 -0
+               rpmh-sleep-stats@c3f0000 {
+                       compatible = "qcom,rpmh-sleep-stats";
+                       reg = <0 0x0c3f0000 0 0x400>;
+               };
+

Does this need to be in DT? Can the sc7180-aoss-qmp driver use the aux-bus and stick the sleep stats device on there?

The AOSS memory space has N chunks of "message ram", one is used for the QMP protocol (presumably the APSS specific one), a different one is used for the sleep stats.

I presume we could have come up with a binding for the entire AOSS/AOP and then describe (either implicit or explicitly) the QMP and debug-stats under that.

But we'd also have to come up with the same container-device for the RPM case.

Regards,
Bjorn

Quoted 3 lines
Quoted 3 lines
arch/arm64/boot/dts/qcom/sc7180.dtsi +0 -0
                spmi_bus: spmi@c440000 {
                        compatible = "qcom,spmi-pmic-arb";
                        reg = <0 0x0c440000 0 0x1100>,
#9 lore raw

Re: [PATCH v8 3/5] arm64: dts: qcom: sc7180: Enable SoC sleep stats

Quoting Bjorn Andersson (2021-05-31 10:57:03)

Quoted 46 lines · Stephen Boyd
Quoted 31 lines

Quoting Maulik Shah (2021-05-21 04:26:09)

Quoted 29 lines

Add device node for SoC sleep stats driver which provides various low power mode stats.

Also update the reg size of aoss_qmp device to 0x400.

---
 arch/arm64/boot/dts/qcom/sc7180.dtsi | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
arch/arm64/boot/dts/qcom/sc7180.dtsi +1 -1
index 6228ba2..889d04d 100644
@@ -3215,7 +3215,7 @@
 
                aoss_qmp: power-controller@c300000 {
                        compatible = "qcom,sc7180-aoss-qmp";
-                       reg = <0 0x0c300000 0 0x100000>;
+                       reg = <0 0x0c300000 0 0x400>;
                        interrupts = <GIC_SPI 389 IRQ_TYPE_EDGE_RISING>;
                        mboxes = <&apss_shared 0>;
 
@@ -3223,6 +3223,11 @@
                        #power-domain-cells = <1>;
                };
 
Quoted 9 lines

arch/arm64/boot/dts/qcom/sc7180.dtsi line 3230

Quoted 5 lines
arch/arm64/boot/dts/qcom/sc7180.dtsi +5 -0
+               rpmh-sleep-stats@c3f0000 {
+                       compatible = "qcom,rpmh-sleep-stats";
+                       reg = <0 0x0c3f0000 0 0x400>;
+               };
+

Does this need to be in DT? Can the sc7180-aoss-qmp driver use the aux-bus and stick the sleep stats device on there?

The AOSS memory space has N chunks of "message ram", one is used for the QMP protocol (presumably the APSS specific one), a different one is used for the sleep stats.

Quoted 6 lines

I presume we could have come up with a binding for the entire AOSS/AOP and then describe (either implicit or explicitly) the QMP and debug-stats under that.

But we'd also have to come up with the same container-device for the RPM case.

Because the rpm node doesn't include this region of memory today? I still fail to see why we're changing the existing binding and adding a DT node for this new region that is basically a debug feature.

#10 lore raw

Re: [PATCH v8 3/5] arm64: dts: qcom: sc7180: Enable SoC sleep stats

Quoted 51 lines · Stephen Boyd

Quoting Bjorn Andersson (2021-05-31 10:57:03)

Quoted 50 lines · Stephen Boyd
Quoted 31 lines

Quoting Maulik Shah (2021-05-21 04:26:09)

Quoted 29 lines

Add device node for SoC sleep stats driver which provides various low power mode stats.

Also update the reg size of aoss_qmp device to 0x400.

---
 arch/arm64/boot/dts/qcom/sc7180.dtsi | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
arch/arm64/boot/dts/qcom/sc7180.dtsi +1 -1
index 6228ba2..889d04d 100644
@@ -3215,7 +3215,7 @@
 
                aoss_qmp: power-controller@c300000 {
                        compatible = "qcom,sc7180-aoss-qmp";
-                       reg = <0 0x0c300000 0 0x100000>;
+                       reg = <0 0x0c300000 0 0x400>;
                        interrupts = <GIC_SPI 389 IRQ_TYPE_EDGE_RISING>;
                        mboxes = <&apss_shared 0>;
 
@@ -3223,6 +3223,11 @@
                        #power-domain-cells = <1>;
                };
 
Quoted 9 lines

arch/arm64/boot/dts/qcom/sc7180.dtsi line 3230

Quoted 5 lines
arch/arm64/boot/dts/qcom/sc7180.dtsi +5 -0
+               rpmh-sleep-stats@c3f0000 {
+                       compatible = "qcom,rpmh-sleep-stats";
+                       reg = <0 0x0c3f0000 0 0x400>;
+               };
+

Does this need to be in DT? Can the sc7180-aoss-qmp driver use the aux-bus and stick the sleep stats device on there?

The AOSS memory space has N chunks of "message ram", one is used for the QMP protocol (presumably the APSS specific one), a different one is used for the sleep stats.

I presume we could have come up with a binding for the entire AOSS/AOP and then describe (either implicit or explicitly) the QMP and debug-stats under that.

Quoted 6 lines
Quoted 2 lines

But we'd also have to come up with the same container-device for the RPM case.

Because the rpm node doesn't include this region of memory today? I still fail to see why we're changing the existing binding and adding a DT node for this new region that is basically a debug feature.

We're not changing the binding, the memory region for the "AOSS QMP" thing was never larger than 0x400.

0x100000 is the size of all the AOSS "msg_ram" regions. We don't have this whole thing described in a binding and we don't have an implementation for the whole thing.

If we're going for that we'd need to extend the binding to indicate which of the msg_ram regions are used for APSS QMP and for debug stats on particular platform (either by compatible, explicit properties or as some subnodes).

That said, as I looked into my other objection, for the RPM (non-hardened) case it seems that we're actually describing the RPM region. So there it would make sense to describe it as such in DT - but we don't have any other code (that I'm aware of) that would implement the "qcom,<platform>-rpm".

Regards,
Bjorn

#11 lore raw

Re: [PATCH v8 3/5] arm64: dts: qcom: sc7180: Enable SoC sleep stats

Quoting Bjorn Andersson (2021-06-02 19:44:40)

Quoted 38 lines · Stephen Boyd
Quoted 27 lines

Quoting Bjorn Andersson (2021-05-31 10:57:03)

Quoted 25 lines · Stephen Boyd
Quoted 6 lines

Quoting Maulik Shah (2021-05-21 04:26:09)

Quoted 4 lines
arch/arm64/boot/dts/qcom/sc7180.dtsi +0 -0
@@ -3223,6 +3223,11 @@
                        #power-domain-cells = <1>;
                };
 
Quoted 9 lines

arch/arm64/boot/dts/qcom/sc7180.dtsi line 3230

Quoted 5 lines
arch/arm64/boot/dts/qcom/sc7180.dtsi +5 -0
+               rpmh-sleep-stats@c3f0000 {
+                       compatible = "qcom,rpmh-sleep-stats";
+                       reg = <0 0x0c3f0000 0 0x400>;
+               };
+

Does this need to be in DT? Can the sc7180-aoss-qmp driver use the aux-bus and stick the sleep stats device on there?

The AOSS memory space has N chunks of "message ram", one is used for the QMP protocol (presumably the APSS specific one), a different one is used for the sleep stats.

I presume we could have come up with a binding for the entire AOSS/AOP and then describe (either implicit or explicitly) the QMP and debug-stats under that.

Quoted 6 lines
Quoted 2 lines

But we'd also have to come up with the same container-device for the RPM case.

Because the rpm node doesn't include this region of memory today? I still fail to see why we're changing the existing binding and adding a DT node for this new region that is basically a debug feature.

We're not changing the binding, the memory region for the "AOSS QMP" thing was never larger than 0x400.

Quoted 8 lines

0x100000 is the size of all the AOSS "msg_ram" regions. We don't have this whole thing described in a binding and we don't have an implementation for the whole thing.

If we're going for that we'd need to extend the binding to indicate which of the msg_ram regions are used for APSS QMP and for debug stats on particular platform (either by compatible, explicit properties or as some subnodes).

Fair enough. At the least, can we change the name of the node then to 'sram' or 'ram'? The 'rpmh-sleep-stats' node name is nonsense.

Quoted 8 lines

That said, as I looked into my other objection, for the RPM (non-hardened) case it seems that we're actually describing the RPM region. So there it would make sense to describe it as such in DT - but we don't have any other code (that I'm aware of) that would implement the "qcom,<platform>-rpm".

I only half parsed this part. Are you saying that because we don't have a driver for qcom,<platform>-rpm we shouldn't keep it all within the rpm node?

#12 lore raw

Re: [PATCH v8 3/5] arm64: dts: qcom: sc7180: Enable SoC sleep stats

Quoted 43 lines · Stephen Boyd

Quoting Bjorn Andersson (2021-06-02 19:44:40)

Quoted 42 lines · Stephen Boyd
Quoted 27 lines

Quoting Bjorn Andersson (2021-05-31 10:57:03)

Quoted 25 lines · Stephen Boyd
Quoted 6 lines

Quoting Maulik Shah (2021-05-21 04:26:09)

Quoted 4 lines
arch/arm64/boot/dts/qcom/sc7180.dtsi +0 -0
@@ -3223,6 +3223,11 @@
                        #power-domain-cells = <1>;
                };
 
Quoted 9 lines

arch/arm64/boot/dts/qcom/sc7180.dtsi line 3230

Quoted 5 lines
arch/arm64/boot/dts/qcom/sc7180.dtsi +5 -0
+               rpmh-sleep-stats@c3f0000 {
+                       compatible = "qcom,rpmh-sleep-stats";
+                       reg = <0 0x0c3f0000 0 0x400>;
+               };
+

Does this need to be in DT? Can the sc7180-aoss-qmp driver use the aux-bus and stick the sleep stats device on there?

The AOSS memory space has N chunks of "message ram", one is used for the QMP protocol (presumably the APSS specific one), a different one is used for the sleep stats.

I presume we could have come up with a binding for the entire AOSS/AOP and then describe (either implicit or explicitly) the QMP and debug-stats under that.

Quoted 6 lines
Quoted 2 lines

But we'd also have to come up with the same container-device for the RPM case.

Because the rpm node doesn't include this region of memory today? I still fail to see why we're changing the existing binding and adding a DT node for this new region that is basically a debug feature.

We're not changing the binding, the memory region for the "AOSS QMP" thing was never larger than 0x400.

0x100000 is the size of all the AOSS "msg_ram" regions. We don't have this whole thing described in a binding and we don't have an implementation for the whole thing.

Quoted 8 lines
Quoted 4 lines

If we're going for that we'd need to extend the binding to indicate which of the msg_ram regions are used for APSS QMP and for debug stats on particular platform (either by compatible, explicit properties or as some subnodes).

Fair enough. At the least, can we change the name of the node then to 'sram' or 'ram'? The 'rpmh-sleep-stats' node name is nonsense.

Yes, "ram" sounds like a better node name for both the qmp and sleep-stats region - in the RPMH case.

Quoted 9 lines
Quoted 8 lines

That said, as I looked into my other objection, for the RPM (non-hardened) case it seems that we're actually describing the RPM region. So there it would make sense to describe it as such in DT - but we don't have any other code (that I'm aware of) that would implement the "qcom,<platform>-rpm".

Quoted 3 lines

I only half parsed this part. Are you saying that because we don't have a driver for qcom,<platform>-rpm we shouldn't keep it all within the rpm node?

What I was trying to say is that in the RPM (non-H) case the described memory region is not a chunk of "ram" (or "sram"), but seems to rather be the RPM region. So there it seems more reasonable to have a non-debug compatible, but I don't think we have any other use for it than the debug-stats...

Regards,
Bjorn

#4 lore raw

Re: [PATCH v8 2/5] soc: qcom: Add SoC sleep stats driver

In reply to Stephen Boyd

Hi,

Quoted 59 lines · Stephen Boyd

Quoting Maulik Shah (2021-05-21 04:26:08)

Quoted 58 lines
drivers/soc/qcom/soc_sleep_stats.c new file +52 -0
new file mode 100644
index 0000000..a8396f9
@@ -0,0 +1,255 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (c) 2011-2021, The Linux Foundation. All rights reserved.
+ */
+
+#include <linux/debugfs.h>
+#include <linux/device.h>
+#include <linux/io.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/platform_device.h>
+#include <linux/seq_file.h>
+
+#include <linux/soc/qcom/smem.h>
+#include <clocksource/arm_arch_timer.h>
+
+#define STAT_TYPE_OFFSET       0x0
+#define COUNT_OFFSET           0x4
+#define LAST_ENTERED_AT_OFFSET 0x8
+#define LAST_EXITED_AT_OFFSET  0x10
+#define ACCUMULATED_OFFSET     0x18
+#define CLIENT_VOTES_OFFSET    0x1c
+
+struct subsystem_data {
+       const char *name;
+       u32 smem_item;
+       u32 pid;
+};
+
+static const struct subsystem_data subsystems[] = {
+       { "modem", 605, 1 },
+       { "wpss", 605, 13 },
+       { "adsp", 606, 2 },
+       { "cdsp", 607, 5 },
+       { "slpi", 608, 3 },
+       { "gpu", 609, 0 },
+       { "display", 610, 0 },
+       { "adsp_island", 613, 2 },
+       { "slpi_island", 613, 3 },
+};
+
+struct stats_config {
+       size_t offset_addr;
+       size_t num_records;
+       bool appended_stats_avail;
+};
+
+struct stats_data {
+       bool appended_stats_avail;
+       void __iomem *base;
+};
+
Quoted 8 lines

drivers/soc/qcom/soc_sleep_stats.c line 58

Quoted 6 lines
drivers/soc/qcom/soc_sleep_stats.c new file +6 -0
+struct sleep_stats {
+       u32 stat_type;
+       u32 count;
+       u64 last_entered_at;
+       u64 last_exited_at;
+       u64 accumulated;

Is this coming from memory directly? So we need a bunch of __le32 and __le64 markings and then the endian swaps?

Its stored in AOP (RPMH)/ RPM msgram for SoC sleep stats and individual subsystem stats stored in SMEM.

Quoted 7 lines

drivers/soc/qcom/soc_sleep_stats.c line 63

Quoted 6 lines
drivers/soc/qcom/soc_sleep_stats.c new file +5 -0
+};
+
+struct appended_stats {
+       u32 client_votes;
+       u32 reserved[3];

This one too? It's stored in smem?

Its only stored in RPM msgram. RPMH and SMEM do not have it.

Quoted 11 lines
Quoted 11 lines
drivers/soc/qcom/soc_sleep_stats.c new file +10 -0
+};
+
+static void qcom_print_stats(struct seq_file *s, const struct sleep_stats *stat)
+{
+       u64 accumulated = stat->accumulated;
+       /*
+        * If a subsystem is in sleep when reading the sleep stats adjust
+        * the accumulated sleep duration to show actual sleep time.
+        */
+       if (stat->last_entered_at > stat->last_exited_at)
Quoted 8 lines

drivers/soc/qcom/soc_sleep_stats.c line 74

Quoted 1 line
drivers/soc/qcom/soc_sleep_stats.c new file +1 -0
+               accumulated += arch_timer_read_counter() - stat->last_entered_at;

Doesn't this read the virtual count? Which could be very different than the physical count that the subsystems see and then write into smem? I suppose we need some sort of "physical count" reading API to match up the timestamps that are there but that is sort of hard given that the physical count isn't always intended to be read by the kernel. Maybe we should split off an MMIO architected timer to be used by this code that uses the same clocksource that the other subsystems are using?

The timer used is same across all subsystems. Beside it try to give more accurate information by modifying total sleep length if subsytem is staying in sleep.

Quoted 11 lines
Quoted 11 lines
drivers/soc/qcom/soc_sleep_stats.c new file +10 -0
+
+       seq_printf(s, "Count: %u\n", stat->count);
+       seq_printf(s, "Last Entered At: %llu\n", stat->last_entered_at);
+       seq_printf(s, "Last Exited At: %llu\n", stat->last_exited_at);
+       seq_printf(s, "Accumulated Duration: %llu\n", accumulated);
+}
+
+static int qcom_subsystem_sleep_stats_show(struct seq_file *s, void *unused)
+{
+       struct subsystem_data *subsystem = s->private;
Quoted 8 lines

drivers/soc/qcom/soc_sleep_stats.c line 90

Quoted 6 lines
drivers/soc/qcom/soc_sleep_stats.c new file +6 -0
+       struct sleep_stats *stat;
+
+       /* Items are allocated lazily, so lookup pointer each time */
+       stat = qcom_smem_get(subsystem->pid, subsystem->smem_item, NULL);
+       if (IS_ERR(stat))
+               return PTR_ERR(stat);

Can this return EPROBE_DEFER to userspace? That would be pretty weird. Maybe we should convert the return value to -EIO regardless of the error.

Ok. Updated in v9 to return -EIO.

Quoted 36 lines
Quoted 36 lines
drivers/soc/qcom/soc_sleep_stats.c new file +35 -0
+
+       qcom_print_stats(s, stat);
+
+       return 0;
+}
+
+static int qcom_soc_sleep_stats_show(struct seq_file *s, void *unused)
+{
+       struct stats_data *d = s->private;
+       void __iomem *reg = d->base;
+       struct sleep_stats stat;
+
+       memcpy_fromio(&stat, reg, sizeof(stat));
+       qcom_print_stats(s, &stat);
+
+       if (d->appended_stats_avail) {
+               struct appended_stats votes;
+
+               memcpy_fromio(&votes, reg + CLIENT_VOTES_OFFSET, sizeof(votes));
+               seq_printf(s, "Client Votes: %#x\n", votes.client_votes);
+       }
+
+       return 0;
+}
+
+DEFINE_SHOW_ATTRIBUTE(qcom_soc_sleep_stats);
+DEFINE_SHOW_ATTRIBUTE(qcom_subsystem_sleep_stats);
+
+static void qcom_create_soc_sleep_stat_files(struct dentry *root, void __iomem *reg,
+                                            struct stats_data *d, u32 num_records)
+{
+       char stat_type[sizeof(u32) + 1] = {0};
+       u32 offset = 0, type;
+       int i, j;
+
Quoted 8 lines

drivers/soc/qcom/soc_sleep_stats.c line 130

Quoted 5 lines
drivers/soc/qcom/soc_sleep_stats.c new file +5 -0
+       for (i = 0; i < num_records; i++) {
+               d[i].base = reg + offset;
+
+               /*
+                * Read the low power mode name and create debugfs file for it.

What do we do if there are two same named low power modes? Maybe they should be directories instead with pretty name and values inside of them.

This is agreed between AOP and Linux kernel so we would not have two or more same named low power modes.

This applies to many things like the structure defination which also be agreed and AOP/subsytems populates same

structure in MSGRAM / SMEM respectively which kernel can read in same format.

Quoted 17 lines
Quoted 17 lines
drivers/soc/qcom/soc_sleep_stats.c new file +16 -0
+                * The names read could be of below,
+                * (may change depending on low power mode supported).
+                * For rpmh-sleep-stats: "aosd", "cxsd" and "ddr".
+                * For rpm-sleep-stats: "vmin" and "vlow".
+                */
+               type = readl(d[i].base);
+               for (j = 0; j < sizeof(u32); j++) {
+                       stat_type[j] = type & 0xff;
+                       type = type >> 8;
+               }
+               strim(stat_type);
+               debugfs_create_file(stat_type, 0400, root, &d[i],
+                                   &qcom_soc_sleep_stats_fops);
+
+               offset += sizeof(struct sleep_stats);
+               if (d[i].appended_stats_avail)
Quoted 8 lines

drivers/soc/qcom/soc_sleep_stats.c line 153

Quoted 7 lines
drivers/soc/qcom/soc_sleep_stats.c new file +7 -0
+                       offset += sizeof(struct appended_stats);
+       }
+}
+
+static void qcom_create_subsystem_stat_files(struct dentry *root)
+{
+       struct sleep_stats *stat;

const? The pointer isn't used though so is this even relevant?

Changed to const.

Quoted 11 lines

drivers/soc/qcom/soc_sleep_stats.c line 161

Quoted 9 lines
drivers/soc/qcom/soc_sleep_stats.c new file +8 -0
+       int i;
+
+       for (i = 0; i < ARRAY_SIZE(subsystems); i++) {
+               stat = qcom_smem_get(subsystems[i].pid, subsystems[i].smem_item, NULL);
+               if (IS_ERR(stat))
+                       continue;
+
+               debugfs_create_file(subsystems[i].name, 0400, root, (void *)&subsystems[i],

Do we need to cast to void pointer? Is it unused? If it's unused then pass NULL?

It is used. It needs cast too since subsystems array is already a const, without cast will compiler will complain discarding const.

Quoted 19 lines
Quoted 19 lines
drivers/soc/qcom/soc_sleep_stats.c new file +18 -0
+                                   &qcom_subsystem_sleep_stats_fops);
+       }
+}
+
+static int qcom_soc_sleep_stats_probe(struct platform_device *pdev)
+{
+       struct resource *res;
+       void __iomem *reg;
+       void __iomem *offset_addr;
+       phys_addr_t stats_base;
+       resource_size_t stats_size;
+       struct dentry *root;
+       const struct stats_config *config;
+       struct stats_data *d;
+       int i;
+
+       config = device_get_match_data(&pdev->dev);
+       if (!config)
Quoted 8 lines

drivers/soc/qcom/soc_sleep_stats.c line 186

Quoted 7 lines
drivers/soc/qcom/soc_sleep_stats.c new file +7 -0
+               return -ENODEV;
+
+       res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       if (!res)
+               return PTR_ERR(res);
+
+       offset_addr = ioremap(res->start + config->offset_addr, sizeof(u32));

Why can't we devm_platform_get_and_ioremap_resource()?

Ok, updated in v9.

Quoted 9 lines

drivers/soc/qcom/soc_sleep_stats.c line 192

Quoted 7 lines
drivers/soc/qcom/soc_sleep_stats.c new file +6 -0
+       if (IS_ERR(offset_addr))
+               return PTR_ERR(offset_addr);
+
+       stats_base = res->start | readl_relaxed(offset_addr);
+       stats_size = resource_size(res);
+       iounmap(offset_addr);

Oh because of unmap? Can we add a resource for stats_base/size to the system? I'd really like to see what address that is at.

Ok it was developed like this since AOP can populate the stat at different address and it gives this address when reading the value at offset.

Quoted 5 lines

I think it is
actually within the same region of this "pointer" that is assigned in DT, but I honestly don't know why we wouldn't just hardcode the address in DT to begin with instead of doing this map, find the real address, and then map that. Is this really changing that often?

yes its in the same region. it generally don't change for single target but prone to change for other targets.

I have hardcoded that address it self in DT to start with and removed the offset part from driver.

Quoted 50 lines

Probably it
changes a few times during development but then it settles down and we can just hardcode it in DT? At least please make a resource and reserve it so we know that the region isn't being used by other kernel drivers.

Quoted 45 lines
drivers/soc/qcom/soc_sleep_stats.c new file +45 -0
+
+       reg = devm_ioremap(&pdev->dev, stats_base, stats_size);
+       if (!reg)
+               return -ENOMEM;
+
+       d = devm_kcalloc(&pdev->dev, config->num_records,
+                        sizeof(*d), GFP_KERNEL);
+       if (!d)
+               return -ENOMEM;
+
+       for (i = 0; i < config->num_records; i++)
+               d[i].appended_stats_avail = config->appended_stats_avail;
+
+       root = debugfs_create_dir("qcom_sleep_stats", NULL);
+
+       qcom_create_subsystem_stat_files(root);
+       qcom_create_soc_sleep_stat_files(root, reg, d, config->num_records);
+
+       platform_set_drvdata(pdev, root);
+
+       return 0;
+}
+
+static int qcom_soc_sleep_stats_remove(struct platform_device *pdev)
+{
+       struct dentry *root = platform_get_drvdata(pdev);
+
+       debugfs_remove_recursive(root);
+
+       return 0;
+}
+
+static const struct stats_config rpm_data = {
+       .offset_addr = 0x14,
+       .num_records = 2,
+       .appended_stats_avail = true,
+};
+
+static const struct stats_config rpmh_data = {
+       .offset_addr = 0x4,
+       .num_records = 3,
+       .appended_stats_avail = false,
+};
+
+static const struct of_device_id qcom_soc_sleep_stats_table[] = {
Quoted 8 lines

drivers/soc/qcom/soc_sleep_stats.c line 243

Quoted 6 lines
drivers/soc/qcom/soc_sleep_stats.c new file +6 -0
+       { .compatible = "qcom,rpm-sleep-stats", .data = &rpm_data },
+       { .compatible = "qcom,rpmh-sleep-stats", .data = &rpmh_data },
+       { }
+};
+
+static struct platform_driver soc_sleep_stats_driver = {

A driver seems pretty heavyweight for a thing that sticks stuff into debugfs. It's not "driving" anything.

Removed the driver from the end.

Thanks,
Maulik

Quoted 13 lines
Quoted 13 lines
drivers/soc/qcom/soc_sleep_stats.c new file +12 -0
+       .probe = qcom_soc_sleep_stats_probe,
+       .remove = qcom_soc_sleep_stats_remove,
+       .driver = {
+               .name = "soc_sleep_stats",
+               .of_match_table = qcom_soc_sleep_stats_table,
+       },
+};
+module_platform_driver(soc_sleep_stats_driver);
+
+MODULE_DESCRIPTION("Qualcomm Technologies, Inc. (QTI) SoC Sleep Stats driver");
+MODULE_LICENSE("GPL v2");
+MODULE_SOFTDEP("pre: smem");
Signature · 1 line
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation
#13 lore raw

Re: [PATCH v8 3/5] arm64: dts: qcom: sc7180: Enable SoC sleep stats

In reply to Bjorn Andersson

Hi,

Quoted 2 lines · Bjorn Andersson

Yes, "ram" sounds like a better node name for both the qmp and sleep-stats region - in the RPMH case.

Updated to use "aop_msgram" in v9.  I think saying only "ram" may further confuse with DDR.

Thanks,
Maulik

Signature · 1 line
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation