* [PATCH v2 RESEND 01/11] platform/x86/amd/pmf: Add support for PMF core layer
2022-07-28 18:20 [PATCH v2 RESEND 00/11] platform/x86/amd/pmf: Introduce AMD PMF Driver Shyam Sundar S K
@ 2022-07-28 18:20 ` Shyam Sundar S K
2022-08-01 11:16 ` Hans de Goede
2022-07-28 18:20 ` [PATCH v2 RESEND 02/11] platform/x86/amd/pmf: Add support for PMF APCI layer Shyam Sundar S K
` (9 subsequent siblings)
10 siblings, 1 reply; 33+ messages in thread
From: Shyam Sundar S K @ 2022-07-28 18:20 UTC (permalink / raw)
To: hdegoede, markgross; +Cc: platform-driver-x86, Patil.Reddy, Shyam Sundar S K
PMF core layer is meant to abstract the common functionalities
across PMF features. This layer also does the plumbing work
like setting up the mailbox channel for the communication
between the PMF driver and the PMFW (Power Management Firmware)
running on the SMU.
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
---
drivers/platform/x86/amd/Kconfig | 2 +
drivers/platform/x86/amd/Makefile | 1 +
drivers/platform/x86/amd/pmf/Kconfig | 16 ++
drivers/platform/x86/amd/pmf/Makefile | 8 +
drivers/platform/x86/amd/pmf/core.c | 235 ++++++++++++++++++++++++++
drivers/platform/x86/amd/pmf/pmf.h | 46 +++++
6 files changed, 308 insertions(+)
create mode 100644 drivers/platform/x86/amd/pmf/Kconfig
create mode 100644 drivers/platform/x86/amd/pmf/Makefile
create mode 100644 drivers/platform/x86/amd/pmf/core.c
create mode 100644 drivers/platform/x86/amd/pmf/pmf.h
diff --git a/drivers/platform/x86/amd/Kconfig b/drivers/platform/x86/amd/Kconfig
index c0d0a3c5170c..a825af8126c8 100644
--- a/drivers/platform/x86/amd/Kconfig
+++ b/drivers/platform/x86/amd/Kconfig
@@ -3,6 +3,8 @@
# AMD x86 Platform Specific Drivers
#
+source "drivers/platform/x86/amd/pmf/Kconfig"
+
config AMD_PMC
tristate "AMD SoC PMC driver"
depends on ACPI && PCI && RTC_CLASS
diff --git a/drivers/platform/x86/amd/Makefile b/drivers/platform/x86/amd/Makefile
index a03fbb08e808..2c229198e24c 100644
--- a/drivers/platform/x86/amd/Makefile
+++ b/drivers/platform/x86/amd/Makefile
@@ -8,3 +8,4 @@ amd-pmc-y := pmc.o
obj-$(CONFIG_AMD_PMC) += amd-pmc.o
amd_hsmp-y := hsmp.o
obj-$(CONFIG_AMD_HSMP) += amd_hsmp.o
+obj-$(CONFIG_AMD_PMF) += pmf/
diff --git a/drivers/platform/x86/amd/pmf/Kconfig b/drivers/platform/x86/amd/pmf/Kconfig
new file mode 100644
index 000000000000..2a5f72419515
--- /dev/null
+++ b/drivers/platform/x86/amd/pmf/Kconfig
@@ -0,0 +1,16 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# AMD PMF Driver
+#
+
+config AMD_PMF
+ tristate "AMD Platform Management Framework"
+ depends on ACPI
+ help
+ This driver provides support for the AMD Platform Management Framework.
+ The goal is to enhance end user experience by making AMD PCs smarter,
+ quiter, power efficient by adapting to user behavior and environment.
+
+ To compile this driver as a module, choose M here: the module will
+ be called amd_pmf.
+
diff --git a/drivers/platform/x86/amd/pmf/Makefile b/drivers/platform/x86/amd/pmf/Makefile
new file mode 100644
index 000000000000..459005f659e5
--- /dev/null
+++ b/drivers/platform/x86/amd/pmf/Makefile
@@ -0,0 +1,8 @@
+# SPDX-License-Identifier: GPL-2.0
+#
+# Makefile for linux/drivers/platform/x86/amd/pmf
+# AMD Platform Management Framework
+#
+
+obj-$(CONFIG_AMD_PMF) += amd-pmf.o
+amd-pmf-objs := core.o
diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
new file mode 100644
index 000000000000..aef97965c181
--- /dev/null
+++ b/drivers/platform/x86/amd/pmf/core.c
@@ -0,0 +1,235 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * AMD Platform Management Framework Driver
+ *
+ * Copyright (c) 2022, Advanced Micro Devices, Inc.
+ * All Rights Reserved.
+ *
+ * Author: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
+ */
+
+#include <linux/iopoll.h>
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/platform_device.h>
+#include "pmf.h"
+
+/* PMF-SMU communication registers */
+#define AMD_PMF_REGISTER_MESSAGE 0xA18
+#define AMD_PMF_REGISTER_RESPONSE 0xA78
+#define AMD_PMF_REGISTER_ARGUMENT 0xA58
+
+/* Base address of SMU for mapping physical address to virtual address */
+#define AMD_PMF_SMU_INDEX_ADDRESS 0xB8
+#define AMD_PMF_SMU_INDEX_DATA 0xBC
+#define AMD_PMF_MAPPING_SIZE 0x01000
+#define AMD_PMF_BASE_ADDR_OFFSET 0x10000
+#define AMD_PMF_BASE_ADDR_LO 0x13B102E8
+#define AMD_PMF_BASE_ADDR_HI 0x13B102EC
+#define AMD_PMF_BASE_ADDR_LO_MASK GENMASK(15, 0)
+#define AMD_PMF_BASE_ADDR_HI_MASK GENMASK(31, 20)
+
+/* SMU Response Codes */
+#define AMD_PMF_RESULT_OK 0x01
+#define AMD_PMF_RESULT_CMD_REJECT_BUSY 0xFC
+#define AMD_PMF_RESULT_CMD_REJECT_PREREQ 0xFD
+#define AMD_PMF_RESULT_CMD_UNKNOWN 0xFE
+#define AMD_PMF_RESULT_FAILED 0xFF
+
+/* List of supported CPU ids */
+#define AMD_CPU_ID_PS 0x14e8
+
+#define PMF_MSG_DELAY_MIN_US 50
+#define RESPONSE_REGISTER_LOOP_MAX 20000
+
+#define DELAY_MIN_US 2000
+#define DELAY_MAX_US 3000
+
+static inline u32 amd_pmf_reg_read(struct amd_pmf_dev *dev, int reg_offset)
+{
+ return ioread32(dev->regbase + reg_offset);
+}
+
+static inline void amd_pmf_reg_write(struct amd_pmf_dev *dev, int reg_offset, u32 val)
+{
+ iowrite32(val, dev->regbase + reg_offset);
+}
+
+static void __maybe_unused amd_pmf_dump_registers(struct amd_pmf_dev *dev)
+{
+ u32 value;
+
+ value = amd_pmf_reg_read(dev, AMD_PMF_REGISTER_RESPONSE);
+ dev_dbg(dev->dev, "AMD_PMF_REGISTER_RESPONSE:%x\n", value);
+
+ value = amd_pmf_reg_read(dev, AMD_PMF_REGISTER_ARGUMENT);
+ dev_dbg(dev->dev, "AMD_PMF_REGISTER_ARGUMENT:%d\n", value);
+
+ value = amd_pmf_reg_read(dev, AMD_PMF_REGISTER_MESSAGE);
+ dev_dbg(dev->dev, "AMD_PMF_REGISTER_MESSAGE:%x\n", value);
+}
+
+int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32 *data)
+{
+ int rc;
+ u32 val;
+
+ mutex_lock(&dev->lock);
+
+ /* Wait until we get a valid response */
+ rc = readx_poll_timeout(ioread32, dev->regbase + AMD_PMF_REGISTER_RESPONSE,
+ val, val != 0, PMF_MSG_DELAY_MIN_US,
+ PMF_MSG_DELAY_MIN_US * RESPONSE_REGISTER_LOOP_MAX);
+ if (rc) {
+ dev_err(dev->dev, "failed to talk to SMU\n");
+ goto out_unlock;
+ }
+
+ /* Write zero to response register */
+ amd_pmf_reg_write(dev, AMD_PMF_REGISTER_RESPONSE, 0);
+
+ /* Write argument into argument register */
+ amd_pmf_reg_write(dev, AMD_PMF_REGISTER_ARGUMENT, arg);
+
+ /* Write message ID to message ID register */
+ amd_pmf_reg_write(dev, AMD_PMF_REGISTER_MESSAGE, message);
+
+ /* Wait until we get a valid response */
+ rc = readx_poll_timeout(ioread32, dev->regbase + AMD_PMF_REGISTER_RESPONSE,
+ val, val != 0, PMF_MSG_DELAY_MIN_US,
+ PMF_MSG_DELAY_MIN_US * RESPONSE_REGISTER_LOOP_MAX);
+ if (rc) {
+ dev_err(dev->dev, "SMU response timed out\n");
+ goto out_unlock;
+ }
+
+ switch (val) {
+ case AMD_PMF_RESULT_OK:
+ if (get) {
+ /* PMFW may take longer time to return back the data */
+ usleep_range(DELAY_MIN_US, 10 * DELAY_MAX_US);
+ *data = amd_pmf_reg_read(dev, AMD_PMF_REGISTER_ARGUMENT);
+ }
+ break;
+ case AMD_PMF_RESULT_CMD_REJECT_BUSY:
+ dev_err(dev->dev, "SMU not ready. err: 0x%x\n", val);
+ rc = -EBUSY;
+ goto out_unlock;
+ case AMD_PMF_RESULT_CMD_UNKNOWN:
+ dev_err(dev->dev, "SMU cmd unknown. err: 0x%x\n", val);
+ rc = -EINVAL;
+ goto out_unlock;
+ case AMD_PMF_RESULT_CMD_REJECT_PREREQ:
+ case AMD_PMF_RESULT_FAILED:
+ default:
+ dev_err(dev->dev, "SMU cmd failed. err: 0x%x\n", val);
+ rc = -EIO;
+ goto out_unlock;
+ }
+
+out_unlock:
+ mutex_unlock(&dev->lock);
+ amd_pmf_dump_registers(dev);
+ return rc;
+}
+
+static const struct pci_device_id pmf_pci_ids[] = {
+ { PCI_DEVICE(PCI_VENDOR_ID_AMD, AMD_CPU_ID_PS) },
+ { }
+};
+
+static const struct acpi_device_id amd_pmf_acpi_ids[] = {
+ {"AMDI0102", 0},
+ { }
+};
+MODULE_DEVICE_TABLE(acpi, amd_pmf_acpi_ids);
+
+static int amd_pmf_probe(struct platform_device *pdev)
+{
+ struct amd_pmf_dev *dev;
+ struct pci_dev *rdev;
+ u32 base_addr_lo;
+ u32 base_addr_hi;
+ u64 base_addr;
+ u32 val;
+ int err;
+
+ dev = devm_kzalloc(&pdev->dev, sizeof(*dev), GFP_KERNEL);
+ if (!dev)
+ return -ENOMEM;
+
+ dev->dev = &pdev->dev;
+
+ rdev = pci_get_domain_bus_and_slot(0, 0, PCI_DEVFN(0, 0));
+ if (!rdev || !pci_match_id(pmf_pci_ids, rdev)) {
+ pci_dev_put(rdev);
+ return -ENODEV;
+ }
+
+ dev->cpu_id = rdev->device;
+ err = pci_write_config_dword(rdev, AMD_PMF_SMU_INDEX_ADDRESS, AMD_PMF_BASE_ADDR_LO);
+ if (err) {
+ dev_err(dev->dev, "error writing to 0x%x\n", AMD_PMF_SMU_INDEX_ADDRESS);
+ pci_dev_put(rdev);
+ return pcibios_err_to_errno(err);
+ }
+
+ err = pci_read_config_dword(rdev, AMD_PMF_SMU_INDEX_DATA, &val);
+ if (err) {
+ pci_dev_put(rdev);
+ return pcibios_err_to_errno(err);
+ }
+
+ base_addr_lo = val & AMD_PMF_BASE_ADDR_HI_MASK;
+
+ err = pci_write_config_dword(rdev, AMD_PMF_SMU_INDEX_ADDRESS, AMD_PMF_BASE_ADDR_HI);
+ if (err) {
+ dev_err(dev->dev, "error writing to 0x%x\n", AMD_PMF_SMU_INDEX_ADDRESS);
+ pci_dev_put(rdev);
+ return pcibios_err_to_errno(err);
+ }
+
+ err = pci_read_config_dword(rdev, AMD_PMF_SMU_INDEX_DATA, &val);
+ if (err) {
+ pci_dev_put(rdev);
+ return pcibios_err_to_errno(err);
+ }
+
+ base_addr_hi = val & AMD_PMF_BASE_ADDR_LO_MASK;
+ pci_dev_put(rdev);
+ base_addr = ((u64)base_addr_hi << 32 | base_addr_lo);
+
+ dev->regbase = devm_ioremap(dev->dev, base_addr + AMD_PMF_BASE_ADDR_OFFSET,
+ AMD_PMF_MAPPING_SIZE);
+ if (!dev->regbase)
+ return -ENOMEM;
+
+ platform_set_drvdata(pdev, dev);
+
+ mutex_init(&dev->lock);
+ dev_info(dev->dev, "registered PMF device successfully\n");
+
+ return 0;
+}
+
+static int amd_pmf_remove(struct platform_device *pdev)
+{
+ struct amd_pmf_dev *dev = platform_get_drvdata(pdev);
+
+ mutex_destroy(&dev->lock);
+ kfree(dev->buf);
+ return 0;
+}
+
+static struct platform_driver amd_pmf_driver = {
+ .driver = {
+ .name = "amd-pmf",
+ .acpi_match_table = amd_pmf_acpi_ids,
+ },
+ .probe = amd_pmf_probe,
+ .remove = amd_pmf_remove,
+};
+module_platform_driver(amd_pmf_driver);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("AMD Platform Management Framework Driver");
diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
new file mode 100644
index 000000000000..1c2e942e5096
--- /dev/null
+++ b/drivers/platform/x86/amd/pmf/pmf.h
@@ -0,0 +1,46 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * AMD Platform Management Framework Driver
+ *
+ * Copyright (c) 2022, Advanced Micro Devices, Inc.
+ * All Rights Reserved.
+ *
+ * Author: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
+ */
+
+#ifndef PMF_H
+#define PMF_H
+
+/* Message Definitions */
+#define SET_SPL 0x03 /* SPL: Sustained Power Limit */
+#define SET_SPPT 0x05 /* SPPT: Slow Package Power Tracking */
+#define SET_FPPT 0x07 /* FPPT: Fast Package Power Tracking */
+#define GET_SPL 0x0B
+#define GET_SPPT 0x0D
+#define GET_FPPT 0x0F
+#define SET_DRAM_ADDR_HIGH 0x14
+#define SET_DRAM_ADDR_LOW 0x15
+#define SET_TRANSFER_TABLE 0x16
+#define SET_STT_MIN_LIMIT 0x18 /* STT: Skin Temperature Tracking */
+#define SET_STT_LIMIT_APU 0x19
+#define SET_STT_LIMIT_HS2 0x1A
+#define SET_SPPT_APU_ONLY 0x1D
+#define GET_SPPT_APU_ONLY 0x1E
+#define GET_STT_MIN_LIMIT 0x1F
+#define GET_STT_LIMIT_APU 0x20
+#define GET_STT_LIMIT_HS2 0x21
+
+struct amd_pmf_dev {
+ void __iomem *regbase;
+ void __iomem *smu_virt_addr;
+ void *buf;
+ u32 base_addr;
+ u32 cpu_id;
+ struct device *dev;
+ struct mutex lock; /* protects the PMF interface */
+};
+
+/* Core Layer */
+int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32 *data);
+
+#endif /* PMF_H */
--
2.25.1
^ permalink raw reply related [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 01/11] platform/x86/amd/pmf: Add support for PMF core layer
2022-07-28 18:20 ` [PATCH v2 RESEND 01/11] platform/x86/amd/pmf: Add support for PMF core layer Shyam Sundar S K
@ 2022-08-01 11:16 ` Hans de Goede
0 siblings, 0 replies; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 11:16 UTC (permalink / raw)
To: Shyam Sundar S K, markgross; +Cc: platform-driver-x86, Patil.Reddy
Hi,
On 7/28/22 20:20, Shyam Sundar S K wrote:
> PMF core layer is meant to abstract the common functionalities
> across PMF features. This layer also does the plumbing work
> like setting up the mailbox channel for the communication
> between the PMF driver and the PMFW (Power Management Firmware)
> running on the SMU.
>
> Reviewed-by: Hans de Goede <hdegoede@redhat.com>
> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
This patch still looks good to me :)
Regards,
Hans
> ---
> drivers/platform/x86/amd/Kconfig | 2 +
> drivers/platform/x86/amd/Makefile | 1 +
> drivers/platform/x86/amd/pmf/Kconfig | 16 ++
> drivers/platform/x86/amd/pmf/Makefile | 8 +
> drivers/platform/x86/amd/pmf/core.c | 235 ++++++++++++++++++++++++++
> drivers/platform/x86/amd/pmf/pmf.h | 46 +++++
> 6 files changed, 308 insertions(+)
> create mode 100644 drivers/platform/x86/amd/pmf/Kconfig
> create mode 100644 drivers/platform/x86/amd/pmf/Makefile
> create mode 100644 drivers/platform/x86/amd/pmf/core.c
> create mode 100644 drivers/platform/x86/amd/pmf/pmf.h
>
> diff --git a/drivers/platform/x86/amd/Kconfig b/drivers/platform/x86/amd/Kconfig
> index c0d0a3c5170c..a825af8126c8 100644
> --- a/drivers/platform/x86/amd/Kconfig
> +++ b/drivers/platform/x86/amd/Kconfig
> @@ -3,6 +3,8 @@
> # AMD x86 Platform Specific Drivers
> #
>
> +source "drivers/platform/x86/amd/pmf/Kconfig"
> +
> config AMD_PMC
> tristate "AMD SoC PMC driver"
> depends on ACPI && PCI && RTC_CLASS
> diff --git a/drivers/platform/x86/amd/Makefile b/drivers/platform/x86/amd/Makefile
> index a03fbb08e808..2c229198e24c 100644
> --- a/drivers/platform/x86/amd/Makefile
> +++ b/drivers/platform/x86/amd/Makefile
> @@ -8,3 +8,4 @@ amd-pmc-y := pmc.o
> obj-$(CONFIG_AMD_PMC) += amd-pmc.o
> amd_hsmp-y := hsmp.o
> obj-$(CONFIG_AMD_HSMP) += amd_hsmp.o
> +obj-$(CONFIG_AMD_PMF) += pmf/
> diff --git a/drivers/platform/x86/amd/pmf/Kconfig b/drivers/platform/x86/amd/pmf/Kconfig
> new file mode 100644
> index 000000000000..2a5f72419515
> --- /dev/null
> +++ b/drivers/platform/x86/amd/pmf/Kconfig
> @@ -0,0 +1,16 @@
> +# SPDX-License-Identifier: GPL-2.0-only
> +#
> +# AMD PMF Driver
> +#
> +
> +config AMD_PMF
> + tristate "AMD Platform Management Framework"
> + depends on ACPI
> + help
> + This driver provides support for the AMD Platform Management Framework.
> + The goal is to enhance end user experience by making AMD PCs smarter,
> + quiter, power efficient by adapting to user behavior and environment.
> +
> + To compile this driver as a module, choose M here: the module will
> + be called amd_pmf.
> +
> diff --git a/drivers/platform/x86/amd/pmf/Makefile b/drivers/platform/x86/amd/pmf/Makefile
> new file mode 100644
> index 000000000000..459005f659e5
> --- /dev/null
> +++ b/drivers/platform/x86/amd/pmf/Makefile
> @@ -0,0 +1,8 @@
> +# SPDX-License-Identifier: GPL-2.0
> +#
> +# Makefile for linux/drivers/platform/x86/amd/pmf
> +# AMD Platform Management Framework
> +#
> +
> +obj-$(CONFIG_AMD_PMF) += amd-pmf.o
> +amd-pmf-objs := core.o
> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
> new file mode 100644
> index 000000000000..aef97965c181
> --- /dev/null
> +++ b/drivers/platform/x86/amd/pmf/core.c
> @@ -0,0 +1,235 @@
> +// SPDX-License-Identifier: GPL-2.0-or-later
> +/*
> + * AMD Platform Management Framework Driver
> + *
> + * Copyright (c) 2022, Advanced Micro Devices, Inc.
> + * All Rights Reserved.
> + *
> + * Author: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> + */
> +
> +#include <linux/iopoll.h>
> +#include <linux/module.h>
> +#include <linux/pci.h>
> +#include <linux/platform_device.h>
> +#include "pmf.h"
> +
> +/* PMF-SMU communication registers */
> +#define AMD_PMF_REGISTER_MESSAGE 0xA18
> +#define AMD_PMF_REGISTER_RESPONSE 0xA78
> +#define AMD_PMF_REGISTER_ARGUMENT 0xA58
> +
> +/* Base address of SMU for mapping physical address to virtual address */
> +#define AMD_PMF_SMU_INDEX_ADDRESS 0xB8
> +#define AMD_PMF_SMU_INDEX_DATA 0xBC
> +#define AMD_PMF_MAPPING_SIZE 0x01000
> +#define AMD_PMF_BASE_ADDR_OFFSET 0x10000
> +#define AMD_PMF_BASE_ADDR_LO 0x13B102E8
> +#define AMD_PMF_BASE_ADDR_HI 0x13B102EC
> +#define AMD_PMF_BASE_ADDR_LO_MASK GENMASK(15, 0)
> +#define AMD_PMF_BASE_ADDR_HI_MASK GENMASK(31, 20)
> +
> +/* SMU Response Codes */
> +#define AMD_PMF_RESULT_OK 0x01
> +#define AMD_PMF_RESULT_CMD_REJECT_BUSY 0xFC
> +#define AMD_PMF_RESULT_CMD_REJECT_PREREQ 0xFD
> +#define AMD_PMF_RESULT_CMD_UNKNOWN 0xFE
> +#define AMD_PMF_RESULT_FAILED 0xFF
> +
> +/* List of supported CPU ids */
> +#define AMD_CPU_ID_PS 0x14e8
> +
> +#define PMF_MSG_DELAY_MIN_US 50
> +#define RESPONSE_REGISTER_LOOP_MAX 20000
> +
> +#define DELAY_MIN_US 2000
> +#define DELAY_MAX_US 3000
> +
> +static inline u32 amd_pmf_reg_read(struct amd_pmf_dev *dev, int reg_offset)
> +{
> + return ioread32(dev->regbase + reg_offset);
> +}
> +
> +static inline void amd_pmf_reg_write(struct amd_pmf_dev *dev, int reg_offset, u32 val)
> +{
> + iowrite32(val, dev->regbase + reg_offset);
> +}
> +
> +static void __maybe_unused amd_pmf_dump_registers(struct amd_pmf_dev *dev)
> +{
> + u32 value;
> +
> + value = amd_pmf_reg_read(dev, AMD_PMF_REGISTER_RESPONSE);
> + dev_dbg(dev->dev, "AMD_PMF_REGISTER_RESPONSE:%x\n", value);
> +
> + value = amd_pmf_reg_read(dev, AMD_PMF_REGISTER_ARGUMENT);
> + dev_dbg(dev->dev, "AMD_PMF_REGISTER_ARGUMENT:%d\n", value);
> +
> + value = amd_pmf_reg_read(dev, AMD_PMF_REGISTER_MESSAGE);
> + dev_dbg(dev->dev, "AMD_PMF_REGISTER_MESSAGE:%x\n", value);
> +}
> +
> +int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32 *data)
> +{
> + int rc;
> + u32 val;
> +
> + mutex_lock(&dev->lock);
> +
> + /* Wait until we get a valid response */
> + rc = readx_poll_timeout(ioread32, dev->regbase + AMD_PMF_REGISTER_RESPONSE,
> + val, val != 0, PMF_MSG_DELAY_MIN_US,
> + PMF_MSG_DELAY_MIN_US * RESPONSE_REGISTER_LOOP_MAX);
> + if (rc) {
> + dev_err(dev->dev, "failed to talk to SMU\n");
> + goto out_unlock;
> + }
> +
> + /* Write zero to response register */
> + amd_pmf_reg_write(dev, AMD_PMF_REGISTER_RESPONSE, 0);
> +
> + /* Write argument into argument register */
> + amd_pmf_reg_write(dev, AMD_PMF_REGISTER_ARGUMENT, arg);
> +
> + /* Write message ID to message ID register */
> + amd_pmf_reg_write(dev, AMD_PMF_REGISTER_MESSAGE, message);
> +
> + /* Wait until we get a valid response */
> + rc = readx_poll_timeout(ioread32, dev->regbase + AMD_PMF_REGISTER_RESPONSE,
> + val, val != 0, PMF_MSG_DELAY_MIN_US,
> + PMF_MSG_DELAY_MIN_US * RESPONSE_REGISTER_LOOP_MAX);
> + if (rc) {
> + dev_err(dev->dev, "SMU response timed out\n");
> + goto out_unlock;
> + }
> +
> + switch (val) {
> + case AMD_PMF_RESULT_OK:
> + if (get) {
> + /* PMFW may take longer time to return back the data */
> + usleep_range(DELAY_MIN_US, 10 * DELAY_MAX_US);
> + *data = amd_pmf_reg_read(dev, AMD_PMF_REGISTER_ARGUMENT);
> + }
> + break;
> + case AMD_PMF_RESULT_CMD_REJECT_BUSY:
> + dev_err(dev->dev, "SMU not ready. err: 0x%x\n", val);
> + rc = -EBUSY;
> + goto out_unlock;
> + case AMD_PMF_RESULT_CMD_UNKNOWN:
> + dev_err(dev->dev, "SMU cmd unknown. err: 0x%x\n", val);
> + rc = -EINVAL;
> + goto out_unlock;
> + case AMD_PMF_RESULT_CMD_REJECT_PREREQ:
> + case AMD_PMF_RESULT_FAILED:
> + default:
> + dev_err(dev->dev, "SMU cmd failed. err: 0x%x\n", val);
> + rc = -EIO;
> + goto out_unlock;
> + }
> +
> +out_unlock:
> + mutex_unlock(&dev->lock);
> + amd_pmf_dump_registers(dev);
> + return rc;
> +}
> +
> +static const struct pci_device_id pmf_pci_ids[] = {
> + { PCI_DEVICE(PCI_VENDOR_ID_AMD, AMD_CPU_ID_PS) },
> + { }
> +};
> +
> +static const struct acpi_device_id amd_pmf_acpi_ids[] = {
> + {"AMDI0102", 0},
> + { }
> +};
> +MODULE_DEVICE_TABLE(acpi, amd_pmf_acpi_ids);
> +
> +static int amd_pmf_probe(struct platform_device *pdev)
> +{
> + struct amd_pmf_dev *dev;
> + struct pci_dev *rdev;
> + u32 base_addr_lo;
> + u32 base_addr_hi;
> + u64 base_addr;
> + u32 val;
> + int err;
> +
> + dev = devm_kzalloc(&pdev->dev, sizeof(*dev), GFP_KERNEL);
> + if (!dev)
> + return -ENOMEM;
> +
> + dev->dev = &pdev->dev;
> +
> + rdev = pci_get_domain_bus_and_slot(0, 0, PCI_DEVFN(0, 0));
> + if (!rdev || !pci_match_id(pmf_pci_ids, rdev)) {
> + pci_dev_put(rdev);
> + return -ENODEV;
> + }
> +
> + dev->cpu_id = rdev->device;
> + err = pci_write_config_dword(rdev, AMD_PMF_SMU_INDEX_ADDRESS, AMD_PMF_BASE_ADDR_LO);
> + if (err) {
> + dev_err(dev->dev, "error writing to 0x%x\n", AMD_PMF_SMU_INDEX_ADDRESS);
> + pci_dev_put(rdev);
> + return pcibios_err_to_errno(err);
> + }
> +
> + err = pci_read_config_dword(rdev, AMD_PMF_SMU_INDEX_DATA, &val);
> + if (err) {
> + pci_dev_put(rdev);
> + return pcibios_err_to_errno(err);
> + }
> +
> + base_addr_lo = val & AMD_PMF_BASE_ADDR_HI_MASK;
> +
> + err = pci_write_config_dword(rdev, AMD_PMF_SMU_INDEX_ADDRESS, AMD_PMF_BASE_ADDR_HI);
> + if (err) {
> + dev_err(dev->dev, "error writing to 0x%x\n", AMD_PMF_SMU_INDEX_ADDRESS);
> + pci_dev_put(rdev);
> + return pcibios_err_to_errno(err);
> + }
> +
> + err = pci_read_config_dword(rdev, AMD_PMF_SMU_INDEX_DATA, &val);
> + if (err) {
> + pci_dev_put(rdev);
> + return pcibios_err_to_errno(err);
> + }
> +
> + base_addr_hi = val & AMD_PMF_BASE_ADDR_LO_MASK;
> + pci_dev_put(rdev);
> + base_addr = ((u64)base_addr_hi << 32 | base_addr_lo);
> +
> + dev->regbase = devm_ioremap(dev->dev, base_addr + AMD_PMF_BASE_ADDR_OFFSET,
> + AMD_PMF_MAPPING_SIZE);
> + if (!dev->regbase)
> + return -ENOMEM;
> +
> + platform_set_drvdata(pdev, dev);
> +
> + mutex_init(&dev->lock);
> + dev_info(dev->dev, "registered PMF device successfully\n");
> +
> + return 0;
> +}
> +
> +static int amd_pmf_remove(struct platform_device *pdev)
> +{
> + struct amd_pmf_dev *dev = platform_get_drvdata(pdev);
> +
> + mutex_destroy(&dev->lock);
> + kfree(dev->buf);
> + return 0;
> +}
> +
> +static struct platform_driver amd_pmf_driver = {
> + .driver = {
> + .name = "amd-pmf",
> + .acpi_match_table = amd_pmf_acpi_ids,
> + },
> + .probe = amd_pmf_probe,
> + .remove = amd_pmf_remove,
> +};
> +module_platform_driver(amd_pmf_driver);
> +
> +MODULE_LICENSE("GPL");
> +MODULE_DESCRIPTION("AMD Platform Management Framework Driver");
> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
> new file mode 100644
> index 000000000000..1c2e942e5096
> --- /dev/null
> +++ b/drivers/platform/x86/amd/pmf/pmf.h
> @@ -0,0 +1,46 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +/*
> + * AMD Platform Management Framework Driver
> + *
> + * Copyright (c) 2022, Advanced Micro Devices, Inc.
> + * All Rights Reserved.
> + *
> + * Author: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> + */
> +
> +#ifndef PMF_H
> +#define PMF_H
> +
> +/* Message Definitions */
> +#define SET_SPL 0x03 /* SPL: Sustained Power Limit */
> +#define SET_SPPT 0x05 /* SPPT: Slow Package Power Tracking */
> +#define SET_FPPT 0x07 /* FPPT: Fast Package Power Tracking */
> +#define GET_SPL 0x0B
> +#define GET_SPPT 0x0D
> +#define GET_FPPT 0x0F
> +#define SET_DRAM_ADDR_HIGH 0x14
> +#define SET_DRAM_ADDR_LOW 0x15
> +#define SET_TRANSFER_TABLE 0x16
> +#define SET_STT_MIN_LIMIT 0x18 /* STT: Skin Temperature Tracking */
> +#define SET_STT_LIMIT_APU 0x19
> +#define SET_STT_LIMIT_HS2 0x1A
> +#define SET_SPPT_APU_ONLY 0x1D
> +#define GET_SPPT_APU_ONLY 0x1E
> +#define GET_STT_MIN_LIMIT 0x1F
> +#define GET_STT_LIMIT_APU 0x20
> +#define GET_STT_LIMIT_HS2 0x21
> +
> +struct amd_pmf_dev {
> + void __iomem *regbase;
> + void __iomem *smu_virt_addr;
> + void *buf;
> + u32 base_addr;
> + u32 cpu_id;
> + struct device *dev;
> + struct mutex lock; /* protects the PMF interface */
> +};
> +
> +/* Core Layer */
> +int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32 *data);
> +
> +#endif /* PMF_H */
^ permalink raw reply [flat|nested] 33+ messages in thread
* [PATCH v2 RESEND 02/11] platform/x86/amd/pmf: Add support for PMF APCI layer
2022-07-28 18:20 [PATCH v2 RESEND 00/11] platform/x86/amd/pmf: Introduce AMD PMF Driver Shyam Sundar S K
2022-07-28 18:20 ` [PATCH v2 RESEND 01/11] platform/x86/amd/pmf: Add support for PMF core layer Shyam Sundar S K
@ 2022-07-28 18:20 ` Shyam Sundar S K
2022-08-01 11:24 ` Hans de Goede
2022-08-01 12:48 ` Hans de Goede
2022-07-28 18:20 ` [PATCH v2 RESEND 03/11] platform/x86/amd/pmf: Add support SPS PMF feature Shyam Sundar S K
` (8 subsequent siblings)
10 siblings, 2 replies; 33+ messages in thread
From: Shyam Sundar S K @ 2022-07-28 18:20 UTC (permalink / raw)
To: hdegoede, markgross; +Cc: platform-driver-x86, Patil.Reddy, Shyam Sundar S K
PMF driver implements the ACPI methods as defined by AMD for PMF Support.
The ACPI layer acts as a glue that helps in providing the infrastructure
for OEMs customization.
OEMs can refer to PMF support documentation to decide on the list of
functions to be supported on their specific platform model.
AMD mandates that PMF ACPI fn0 and fn1 to be implemented which
provides the set of functions, params and the notifications that
would be sent to PMF driver so that PMF driver can adapt and
react.
Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
---
drivers/platform/x86/amd/pmf/Makefile | 2 +-
drivers/platform/x86/amd/pmf/acpi.c | 182 ++++++++++++++++++++++++++
drivers/platform/x86/amd/pmf/core.c | 1 +
drivers/platform/x86/amd/pmf/pmf.h | 32 +++++
4 files changed, 216 insertions(+), 1 deletion(-)
create mode 100644 drivers/platform/x86/amd/pmf/acpi.c
diff --git a/drivers/platform/x86/amd/pmf/Makefile b/drivers/platform/x86/amd/pmf/Makefile
index 459005f659e5..2617eba773ce 100644
--- a/drivers/platform/x86/amd/pmf/Makefile
+++ b/drivers/platform/x86/amd/pmf/Makefile
@@ -5,4 +5,4 @@
#
obj-$(CONFIG_AMD_PMF) += amd-pmf.o
-amd-pmf-objs := core.o
+amd-pmf-objs := core.o acpi.o
diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
new file mode 100644
index 000000000000..bd796d7e1d96
--- /dev/null
+++ b/drivers/platform/x86/amd/pmf/acpi.c
@@ -0,0 +1,182 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * AMD Platform Management Framework Driver
+ *
+ * Copyright (c) 2022, Advanced Micro Devices, Inc.
+ * All Rights Reserved.
+ *
+ * Author: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
+ */
+
+#include <linux/acpi.h>
+#include "pmf.h"
+
+#define APMF_METHOD "\\_SB_.PMF_.APMF"
+
+static unsigned long supported_func;
+
+static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask);
+
+static union acpi_object *apmf_if_call(struct apmf_if *apmf_if, int func, struct acpi_buffer *param)
+{
+ struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
+ struct acpi_object_list apmf_if_arg_list;
+ union acpi_object apmf_if_args[2];
+ acpi_status status;
+
+ apmf_if_arg_list.count = 2;
+ apmf_if_arg_list.pointer = &apmf_if_args[0];
+
+ apmf_if_args[0].type = ACPI_TYPE_INTEGER;
+ apmf_if_args[0].integer.value = func;
+
+ if (param) {
+ apmf_if_args[1].type = ACPI_TYPE_BUFFER;
+ apmf_if_args[1].buffer.length = param->length;
+ apmf_if_args[1].buffer.pointer = param->pointer;
+ } else {
+ apmf_if_args[1].type = ACPI_TYPE_INTEGER;
+ apmf_if_args[1].integer.value = 0;
+ }
+
+ status = acpi_evaluate_object(apmf_if->handle, NULL, &apmf_if_arg_list, &buffer);
+ if (ACPI_FAILURE(status)) {
+ pr_err("PMF: APMF method call failed\n");
+ if (status != AE_NOT_FOUND)
+ kfree(buffer.pointer);
+
+ return NULL;
+ }
+
+ return buffer.pointer;
+}
+
+static int apmf_if_call_store_buffer(struct apmf_if *apmf_if, int func, void *dest, size_t out_sz)
+{
+ union acpi_object *info;
+ size_t size;
+ int err = 0;
+
+ info = apmf_if_call(apmf_if, func, NULL);
+ if (!info)
+ return -EIO;
+
+ if (info->type != ACPI_TYPE_BUFFER) {
+ pr_err("object is not a buffer\n");
+ err = -EINVAL;
+ goto out;
+ }
+
+ if (info->buffer.length < 2) {
+ pr_err("buffer too small\n");
+ err = -EINVAL;
+ goto out;
+ }
+
+ size = *(u16 *)info->buffer.pointer;
+ if (info->buffer.length < size) {
+ pr_err("buffer smaller then headersize %u < %zu\n",
+ info->buffer.length, size);
+ err = -EINVAL;
+ goto out;
+ }
+
+ if (size < out_sz) {
+ pr_err("buffer too small %zu\n", size);
+ err = -EINVAL;
+ goto out;
+ }
+
+ size = out_sz;
+ memcpy(dest, info->buffer.pointer, size);
+
+out:
+ kfree(info);
+ return err;
+}
+
+static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
+{
+ func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
+}
+
+int is_apmf_func_supported(unsigned long index)
+{
+ /* If bit-n is set, that indicates function n+1 is supported */
+ return ((supported_func & (1 << (index - 1))) != 0);
+}
+
+static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
+{
+ struct apmf_verify_interface output;
+ int err;
+
+ err = apmf_if_call_store_buffer(apmf_if, APMF_FUNC_VERIFY_INTERFACE,
+ &output, sizeof(output));
+ if (err)
+ return err;
+
+ apmf_if_parse_functions(&apmf_if->func, output.supported_functions);
+ supported_func = output.supported_functions;
+ dev_dbg(pdev->dev, "supported functions:0x%x notifications:0x%x\n",
+ output.supported_functions, output.notification_mask);
+
+ return 0;
+}
+
+static int apmf_get_system_params(struct apmf_if *apmf_if)
+{
+ struct apmf_system_params params;
+ int err;
+
+ if (apmf_if->func.system_params) {
+ err = apmf_if_call_store_buffer(apmf_if, APMF_FUNC_GET_SYS_PARAMS,
+ ¶ms, sizeof(params));
+ if (err)
+ return err;
+ }
+
+ pr_debug("PMF: system params mask:0x%x flags:0x%x cmd_code:0x%x\n",
+ params.valid_mask,
+ params.flags,
+ params.command_code);
+ params.flags = params.flags & params.valid_mask;
+
+ return 0;
+}
+
+int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
+{
+ acpi_handle apmf_if_handle;
+ struct apmf_if *apmf_if;
+ acpi_status status;
+ int ret;
+
+ status = acpi_get_handle(NULL, (acpi_string) APMF_METHOD, &apmf_if_handle);
+ if (ACPI_FAILURE(status))
+ return -ENODEV;
+
+ apmf_if = kzalloc(sizeof(*apmf_if), GFP_KERNEL);
+ if (!apmf_if)
+ goto out;
+
+ apmf_if->handle = apmf_if_handle;
+
+ ret = apmf_if_verify_interface(pmf_dev, apmf_if);
+ if (ret) {
+ dev_err(pmf_dev->dev, "APMF verify interface failed :%d\n", ret);
+ kfree(apmf_if);
+ goto out;
+ }
+ pmf_dev->apmf_if = apmf_if;
+
+ ret = apmf_get_system_params(apmf_if);
+ if (ret) {
+ dev_err(pmf_dev->dev, "APMF apmf_get_system_params failed :%d\n", ret);
+ kfree(apmf_if);
+ goto out;
+ }
+
+out:
+ return ret;
+}
diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
index aef97965c181..c5002b7bb904 100644
--- a/drivers/platform/x86/amd/pmf/core.c
+++ b/drivers/platform/x86/amd/pmf/core.c
@@ -204,6 +204,7 @@ static int amd_pmf_probe(struct platform_device *pdev)
if (!dev->regbase)
return -ENOMEM;
+ apmf_acpi_init(dev);
platform_set_drvdata(pdev, dev);
mutex_init(&dev->lock);
diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
index 1c2e942e5096..08c6bc0e2b42 100644
--- a/drivers/platform/x86/amd/pmf/pmf.h
+++ b/drivers/platform/x86/amd/pmf/pmf.h
@@ -11,6 +11,12 @@
#ifndef PMF_H
#define PMF_H
+#include <linux/acpi.h>
+
+/* APMF Functions */
+#define APMF_FUNC_VERIFY_INTERFACE 0
+#define APMF_FUNC_GET_SYS_PARAMS 1
+
/* Message Definitions */
#define SET_SPL 0x03 /* SPL: Sustained Power Limit */
#define SET_SPPT 0x05 /* SPPT: Slow Package Power Tracking */
@@ -30,6 +36,30 @@
#define GET_STT_LIMIT_APU 0x20
#define GET_STT_LIMIT_HS2 0x21
+/* AMD PMF BIOS interfaces */
+struct apmf_if_functions {
+ bool system_params;
+};
+
+struct apmf_if {
+ acpi_handle handle;
+ struct apmf_if_functions func;
+};
+
+struct apmf_verify_interface {
+ u16 size;
+ u16 version;
+ u32 notification_mask;
+ u32 supported_functions;
+} __packed;
+
+struct apmf_system_params {
+ u16 size;
+ u32 valid_mask;
+ u32 flags;
+ u8 command_code;
+} __packed;
+
struct amd_pmf_dev {
void __iomem *regbase;
void __iomem *smu_virt_addr;
@@ -37,10 +67,12 @@ struct amd_pmf_dev {
u32 base_addr;
u32 cpu_id;
struct device *dev;
+ struct apmf_if *apmf_if;
struct mutex lock; /* protects the PMF interface */
};
/* Core Layer */
+int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32 *data);
#endif /* PMF_H */
--
2.25.1
^ permalink raw reply related [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 02/11] platform/x86/amd/pmf: Add support for PMF APCI layer
2022-07-28 18:20 ` [PATCH v2 RESEND 02/11] platform/x86/amd/pmf: Add support for PMF APCI layer Shyam Sundar S K
@ 2022-08-01 11:24 ` Hans de Goede
2022-08-01 12:48 ` Hans de Goede
1 sibling, 0 replies; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 11:24 UTC (permalink / raw)
To: Shyam Sundar S K, markgross; +Cc: platform-driver-x86, Patil.Reddy
Hi,
On 7/28/22 20:20, Shyam Sundar S K wrote:
> PMF driver implements the ACPI methods as defined by AMD for PMF Support.
> The ACPI layer acts as a glue that helps in providing the infrastructure
> for OEMs customization.
>
> OEMs can refer to PMF support documentation to decide on the list of
> functions to be supported on their specific platform model.
>
> AMD mandates that PMF ACPI fn0 and fn1 to be implemented which
> provides the set of functions, params and the notifications that
> would be sent to PMF driver so that PMF driver can adapt and
> react.
>
> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> ---
> drivers/platform/x86/amd/pmf/Makefile | 2 +-
> drivers/platform/x86/amd/pmf/acpi.c | 182 ++++++++++++++++++++++++++
> drivers/platform/x86/amd/pmf/core.c | 1 +
> drivers/platform/x86/amd/pmf/pmf.h | 32 +++++
> 4 files changed, 216 insertions(+), 1 deletion(-)
> create mode 100644 drivers/platform/x86/amd/pmf/acpi.c
>
> diff --git a/drivers/platform/x86/amd/pmf/Makefile b/drivers/platform/x86/amd/pmf/Makefile
> index 459005f659e5..2617eba773ce 100644
> --- a/drivers/platform/x86/amd/pmf/Makefile
> +++ b/drivers/platform/x86/amd/pmf/Makefile
> @@ -5,4 +5,4 @@
> #
>
> obj-$(CONFIG_AMD_PMF) += amd-pmf.o
> -amd-pmf-objs := core.o
> +amd-pmf-objs := core.o acpi.o
> diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
> new file mode 100644
> index 000000000000..bd796d7e1d96
> --- /dev/null
> +++ b/drivers/platform/x86/amd/pmf/acpi.c
> @@ -0,0 +1,182 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * AMD Platform Management Framework Driver
> + *
> + * Copyright (c) 2022, Advanced Micro Devices, Inc.
> + * All Rights Reserved.
> + *
> + * Author: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> + */
> +
> +#include <linux/acpi.h>
> +#include "pmf.h"
> +
> +#define APMF_METHOD "\\_SB_.PMF_.APMF"
> +
> +static unsigned long supported_func;
> +
> +static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask);
> +
> +static union acpi_object *apmf_if_call(struct apmf_if *apmf_if, int func, struct acpi_buffer *param)
> +{
> + struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
> + struct acpi_object_list apmf_if_arg_list;
> + union acpi_object apmf_if_args[2];
> + acpi_status status;
> +
> + apmf_if_arg_list.count = 2;
> + apmf_if_arg_list.pointer = &apmf_if_args[0];
> +
> + apmf_if_args[0].type = ACPI_TYPE_INTEGER;
> + apmf_if_args[0].integer.value = func;
> +
> + if (param) {
> + apmf_if_args[1].type = ACPI_TYPE_BUFFER;
> + apmf_if_args[1].buffer.length = param->length;
> + apmf_if_args[1].buffer.pointer = param->pointer;
> + } else {
> + apmf_if_args[1].type = ACPI_TYPE_INTEGER;
> + apmf_if_args[1].integer.value = 0;
> + }
> +
> + status = acpi_evaluate_object(apmf_if->handle, NULL, &apmf_if_arg_list, &buffer);
> + if (ACPI_FAILURE(status)) {
> + pr_err("PMF: APMF method call failed\n");
> + if (status != AE_NOT_FOUND)
> + kfree(buffer.pointer);
> +
> + return NULL;
> + }
> +
> + return buffer.pointer;
> +}
> +
The below helper is much better then the open coding in the v1 series, thank you.
> +static int apmf_if_call_store_buffer(struct apmf_if *apmf_if, int func, void *dest, size_t out_sz)
> +{
> + union acpi_object *info;
> + size_t size;
> + int err = 0;
> +
> + info = apmf_if_call(apmf_if, func, NULL);
> + if (!info)
> + return -EIO;
> +
> + if (info->type != ACPI_TYPE_BUFFER) {
> + pr_err("object is not a buffer\n");
Since you use pr_err, you should put the following line:
#define pr_fmt(fmt) "AMD-PMF: " fmt
*above* the first #include in this .c file, so that the errors
get pre-fixed with "AMD-PMF: " when printed.
> + err = -EINVAL;
> + goto out;
> + }
> +
> + if (info->buffer.length < 2) {
> + pr_err("buffer too small\n");
> + err = -EINVAL;
> + goto out;
> + }
> +
> + size = *(u16 *)info->buffer.pointer;
> + if (info->buffer.length < size) {
> + pr_err("buffer smaller then headersize %u < %zu\n",
> + info->buffer.length, size);
> + err = -EINVAL;
> + goto out;
> + }
> +
> + if (size < out_sz) {
> + pr_err("buffer too small %zu\n", size);
> + err = -EINVAL;
> + goto out;
> + }
> +
> + size = out_sz;
Please drop this line
> + memcpy(dest, info->buffer.pointer, size);
and just use out_sz directly here instead of size.
> +
> +out:
> + kfree(info);
> + return err;
> +}
> +
> +static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
> +{
> + func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
> +}
> +
> +int is_apmf_func_supported(unsigned long index)
> +{
> + /* If bit-n is set, that indicates function n+1 is supported */
> + return ((supported_func & (1 << (index - 1))) != 0);
> +}
> +
> +static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
> +{
> + struct apmf_verify_interface output;
> + int err;
> +
> + err = apmf_if_call_store_buffer(apmf_if, APMF_FUNC_VERIFY_INTERFACE,
> + &output, sizeof(output));
> + if (err)
> + return err;
> +
> + apmf_if_parse_functions(&apmf_if->func, output.supported_functions);
> + supported_func = output.supported_functions;
> + dev_dbg(pdev->dev, "supported functions:0x%x notifications:0x%x\n",
> + output.supported_functions, output.notification_mask);
> +
> + return 0;
> +}
> +
> +static int apmf_get_system_params(struct apmf_if *apmf_if)
> +{
> + struct apmf_system_params params;
> + int err;
> +
> + if (apmf_if->func.system_params) {
> + err = apmf_if_call_store_buffer(apmf_if, APMF_FUNC_GET_SYS_PARAMS,
> + ¶ms, sizeof(params));
> + if (err)
> + return err;
> + }
> +
> + pr_debug("PMF: system params mask:0x%x flags:0x%x cmd_code:0x%x\n",
> + params.valid_mask,
> + params.flags,
> + params.command_code);
> + params.flags = params.flags & params.valid_mask;
> +
> + return 0;
> +}
> +
> +int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
> +{
> + acpi_handle apmf_if_handle;
> + struct apmf_if *apmf_if;
> + acpi_status status;
> + int ret;
> +
> + status = acpi_get_handle(NULL, (acpi_string) APMF_METHOD, &apmf_if_handle);
> + if (ACPI_FAILURE(status))
> + return -ENODEV;
> +
> + apmf_if = kzalloc(sizeof(*apmf_if), GFP_KERNEL);
> + if (!apmf_if)
> + goto out;
> +
> + apmf_if->handle = apmf_if_handle;
> +
> + ret = apmf_if_verify_interface(pmf_dev, apmf_if);
> + if (ret) {
> + dev_err(pmf_dev->dev, "APMF verify interface failed :%d\n", ret);
> + kfree(apmf_if);
> + goto out;
> + }
> + pmf_dev->apmf_if = apmf_if;
> +
> + ret = apmf_get_system_params(apmf_if);
> + if (ret) {
> + dev_err(pmf_dev->dev, "APMF apmf_get_system_params failed :%d\n", ret);
> + kfree(apmf_if);
> + goto out;
If you hit this error path you now have pmf_dev->ampf_of pointing to free-ed
memory. Please move the:
pmf_dev->apmf_if = apmf_if;
statement to below this "if (ret) {}" block.
> + }
> +
> +out:
> + return ret;
> +}
> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
> index aef97965c181..c5002b7bb904 100644
> --- a/drivers/platform/x86/amd/pmf/core.c
> +++ b/drivers/platform/x86/amd/pmf/core.c
> @@ -204,6 +204,7 @@ static int amd_pmf_probe(struct platform_device *pdev)
> if (!dev->regbase)
> return -ENOMEM;
>
> + apmf_acpi_init(dev);
> platform_set_drvdata(pdev, dev);
>
> mutex_init(&dev->lock);
> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
> index 1c2e942e5096..08c6bc0e2b42 100644
> --- a/drivers/platform/x86/amd/pmf/pmf.h
> +++ b/drivers/platform/x86/amd/pmf/pmf.h
> @@ -11,6 +11,12 @@
> #ifndef PMF_H
> #define PMF_H
>
> +#include <linux/acpi.h>
> +
> +/* APMF Functions */
> +#define APMF_FUNC_VERIFY_INTERFACE 0
> +#define APMF_FUNC_GET_SYS_PARAMS 1
> +
> /* Message Definitions */
> #define SET_SPL 0x03 /* SPL: Sustained Power Limit */
> #define SET_SPPT 0x05 /* SPPT: Slow Package Power Tracking */
> @@ -30,6 +36,30 @@
> #define GET_STT_LIMIT_APU 0x20
> #define GET_STT_LIMIT_HS2 0x21
>
> +/* AMD PMF BIOS interfaces */
> +struct apmf_if_functions {
> + bool system_params;
> +};
> +
> +struct apmf_if {
> + acpi_handle handle;
> + struct apmf_if_functions func;
> +};
> +
> +struct apmf_verify_interface {
> + u16 size;
> + u16 version;
> + u32 notification_mask;
> + u32 supported_functions;
> +} __packed;
> +
> +struct apmf_system_params {
> + u16 size;
> + u32 valid_mask;
> + u32 flags;
> + u8 command_code;
> +} __packed;
> +
> struct amd_pmf_dev {
> void __iomem *regbase;
> void __iomem *smu_virt_addr;
> @@ -37,10 +67,12 @@ struct amd_pmf_dev {
> u32 base_addr;
> u32 cpu_id;
> struct device *dev;
> + struct apmf_if *apmf_if;
> struct mutex lock; /* protects the PMF interface */
> };
>
> /* Core Layer */
> +int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
> int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32 *data);
>
> #endif /* PMF_H */
Except for the few small remarks this looks good to me now,
so with the few remarks fixed, yoy can add my:
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
To the next version.
Regards,
Hans
^ permalink raw reply [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 02/11] platform/x86/amd/pmf: Add support for PMF APCI layer
2022-07-28 18:20 ` [PATCH v2 RESEND 02/11] platform/x86/amd/pmf: Add support for PMF APCI layer Shyam Sundar S K
2022-08-01 11:24 ` Hans de Goede
@ 2022-08-01 12:48 ` Hans de Goede
2022-08-01 13:18 ` Hans de Goede
1 sibling, 1 reply; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 12:48 UTC (permalink / raw)
To: Shyam Sundar S K, markgross; +Cc: platform-driver-x86, Patil.Reddy
Hi,
On 7/28/22 20:20, Shyam Sundar S K wrote:
> PMF driver implements the ACPI methods as defined by AMD for PMF Support.
> The ACPI layer acts as a glue that helps in providing the infrastructure
> for OEMs customization.
>
> OEMs can refer to PMF support documentation to decide on the list of
> functions to be supported on their specific platform model.
>
> AMD mandates that PMF ACPI fn0 and fn1 to be implemented which
> provides the set of functions, params and the notifications that
> would be sent to PMF driver so that PMF driver can adapt and
> react.
>
> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> ---
> drivers/platform/x86/amd/pmf/Makefile | 2 +-
> drivers/platform/x86/amd/pmf/acpi.c | 182 ++++++++++++++++++++++++++
> drivers/platform/x86/amd/pmf/core.c | 1 +
> drivers/platform/x86/amd/pmf/pmf.h | 32 +++++
> 4 files changed, 216 insertions(+), 1 deletion(-)
> create mode 100644 drivers/platform/x86/amd/pmf/acpi.c
>
> diff --git a/drivers/platform/x86/amd/pmf/Makefile b/drivers/platform/x86/amd/pmf/Makefile
> index 459005f659e5..2617eba773ce 100644
> --- a/drivers/platform/x86/amd/pmf/Makefile
> +++ b/drivers/platform/x86/amd/pmf/Makefile
> @@ -5,4 +5,4 @@
> #
>
> obj-$(CONFIG_AMD_PMF) += amd-pmf.o
> -amd-pmf-objs := core.o
> +amd-pmf-objs := core.o acpi.o
> diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
> new file mode 100644
> index 000000000000..bd796d7e1d96
> --- /dev/null
> +++ b/drivers/platform/x86/amd/pmf/acpi.c
> @@ -0,0 +1,182 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * AMD Platform Management Framework Driver
> + *
> + * Copyright (c) 2022, Advanced Micro Devices, Inc.
> + * All Rights Reserved.
> + *
> + * Author: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> + */
> +
> +#include <linux/acpi.h>
> +#include "pmf.h"
> +
> +#define APMF_METHOD "\\_SB_.PMF_.APMF"
> +
> +static unsigned long supported_func;
> +
> +static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask);
> +
> +static union acpi_object *apmf_if_call(struct apmf_if *apmf_if, int func, struct acpi_buffer *param)
> +{
> + struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
> + struct acpi_object_list apmf_if_arg_list;
> + union acpi_object apmf_if_args[2];
> + acpi_status status;
> +
> + apmf_if_arg_list.count = 2;
> + apmf_if_arg_list.pointer = &apmf_if_args[0];
> +
> + apmf_if_args[0].type = ACPI_TYPE_INTEGER;
> + apmf_if_args[0].integer.value = func;
> +
> + if (param) {
> + apmf_if_args[1].type = ACPI_TYPE_BUFFER;
> + apmf_if_args[1].buffer.length = param->length;
> + apmf_if_args[1].buffer.pointer = param->pointer;
> + } else {
> + apmf_if_args[1].type = ACPI_TYPE_INTEGER;
> + apmf_if_args[1].integer.value = 0;
> + }
> +
> + status = acpi_evaluate_object(apmf_if->handle, NULL, &apmf_if_arg_list, &buffer);
> + if (ACPI_FAILURE(status)) {
> + pr_err("PMF: APMF method call failed\n");
> + if (status != AE_NOT_FOUND)
> + kfree(buffer.pointer);
> +
> + return NULL;
> + }
> +
> + return buffer.pointer;
> +}
> +
> +static int apmf_if_call_store_buffer(struct apmf_if *apmf_if, int func, void *dest, size_t out_sz)
> +{
> + union acpi_object *info;
> + size_t size;
> + int err = 0;
> +
> + info = apmf_if_call(apmf_if, func, NULL);
> + if (!info)
> + return -EIO;
> +
> + if (info->type != ACPI_TYPE_BUFFER) {
> + pr_err("object is not a buffer\n");
> + err = -EINVAL;
> + goto out;
> + }
> +
> + if (info->buffer.length < 2) {
> + pr_err("buffer too small\n");
> + err = -EINVAL;
> + goto out;
> + }
> +
> + size = *(u16 *)info->buffer.pointer;
> + if (info->buffer.length < size) {
> + pr_err("buffer smaller then headersize %u < %zu\n",
> + info->buffer.length, size);
> + err = -EINVAL;
> + goto out;
> + }
> +
> + if (size < out_sz) {
> + pr_err("buffer too small %zu\n", size);
> + err = -EINVAL;
> + goto out;
> + }
> +
> + size = out_sz;
> + memcpy(dest, info->buffer.pointer, size);
> +
> +out:
> + kfree(info);
> + return err;
> +}
> +
> +static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
> +{
> + func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
I just noticed that this is not correct, this should be:
func->system_params = mask & BIT(APMF_FUNC_GET_SYS_PARAMS - 1);
Which for APMF_FUNC_GET_SYS_PARAMS happens to be the same, but for
other AMPF_FUNC_* values the difference does matter!
Also this information is duplicated by:
is_apmf_func_supported(APMF_FUNC_GET_SYS_PARAMS) please drop the
apmf_if_functions struct and use is_apmf_func_supported() everywhere.
Note this means that you also will need to drop func->sbios_heartbeat
which does not come from the mask and which is initialized in
a different place rather then inside apmf_if_parse_functions() which
already shows that sbios_hearbeat does not really belong inside
this struct.
Instead of moving the sbios_heartbeat flag, please just directly
use notify_cfg.hb_interval inside apmf_acpi_deinit(). This will also
make the cancel_delayed_work_sync(&pmf_dev->heart_beat) condition
match the queue condition 1:1.
> +}
> +
> +int is_apmf_func_supported(unsigned long index)
> +{
> + /* If bit-n is set, that indicates function n+1 is supported */
> + return ((supported_func & (1 << (index - 1))) != 0);
Please write this as:
return (supported_func & BIT(index - 1)) != 0;
or as:
return !!(supported_func & BIT(index - 1));
> +}
> +
> +static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
> +{
> + struct apmf_verify_interface output;
> + int err;
> +
> + err = apmf_if_call_store_buffer(apmf_if, APMF_FUNC_VERIFY_INTERFACE,
> + &output, sizeof(output));
> + if (err)
> + return err;
> +
> + apmf_if_parse_functions(&apmf_if->func, output.supported_functions);
> + supported_func = output.supported_functions;
> + dev_dbg(pdev->dev, "supported functions:0x%x notifications:0x%x\n",
> + output.supported_functions, output.notification_mask);
> +
> + return 0;
> +}
> +
> +static int apmf_get_system_params(struct apmf_if *apmf_if)
> +{
> + struct apmf_system_params params;
> + int err;
> +
> + if (apmf_if->func.system_params) {
> + err = apmf_if_call_store_buffer(apmf_if, APMF_FUNC_GET_SYS_PARAMS,
> + ¶ms, sizeof(params));
> + if (err)
> + return err;
> + }
> +
> + pr_debug("PMF: system params mask:0x%x flags:0x%x cmd_code:0x%x\n",
> + params.valid_mask,
> + params.flags,
> + params.command_code);
> + params.flags = params.flags & params.valid_mask;
> +
> + return 0;
> +}
> +
> +int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
> +{
> + acpi_handle apmf_if_handle;
> + struct apmf_if *apmf_if;
> + acpi_status status;
> + int ret;
> +
> + status = acpi_get_handle(NULL, (acpi_string) APMF_METHOD, &apmf_if_handle);
> + if (ACPI_FAILURE(status))
> + return -ENODEV;
> +
> + apmf_if = kzalloc(sizeof(*apmf_if), GFP_KERNEL);
> + if (!apmf_if)
> + goto out;
> +
> + apmf_if->handle = apmf_if_handle;
> +
> + ret = apmf_if_verify_interface(pmf_dev, apmf_if);
> + if (ret) {
> + dev_err(pmf_dev->dev, "APMF verify interface failed :%d\n", ret);
> + kfree(apmf_if);
> + goto out;
> + }
> + pmf_dev->apmf_if = apmf_if;
After dropping the apmf_if_functions sub-struct from amp_if amp_if only contains
an acpi_handle. Please just store the handle (and later on also the struct
apmf_notification_cfg) directly inside struct amd_pmf_dev, this also removes
the need for doing the whole alloc + free + associated error handling dance here.
Regsards,
Hans
> +
> + ret = apmf_get_system_params(apmf_if);
> + if (ret) {
> + dev_err(pmf_dev->dev, "APMF apmf_get_system_params failed :%d\n", ret);
> + kfree(apmf_if);
> + goto out;
> + }
> +
> +out:
> + return ret;
> +}
> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
> index aef97965c181..c5002b7bb904 100644
> --- a/drivers/platform/x86/amd/pmf/core.c
> +++ b/drivers/platform/x86/amd/pmf/core.c
> @@ -204,6 +204,7 @@ static int amd_pmf_probe(struct platform_device *pdev)
> if (!dev->regbase)
> return -ENOMEM;
>
> + apmf_acpi_init(dev);
> platform_set_drvdata(pdev, dev);
>
> mutex_init(&dev->lock);
> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
> index 1c2e942e5096..08c6bc0e2b42 100644
> --- a/drivers/platform/x86/amd/pmf/pmf.h
> +++ b/drivers/platform/x86/amd/pmf/pmf.h
> @@ -11,6 +11,12 @@
> #ifndef PMF_H
> #define PMF_H
>
> +#include <linux/acpi.h>
> +
> +/* APMF Functions */
> +#define APMF_FUNC_VERIFY_INTERFACE 0
> +#define APMF_FUNC_GET_SYS_PARAMS 1
> +
> /* Message Definitions */
> #define SET_SPL 0x03 /* SPL: Sustained Power Limit */
> #define SET_SPPT 0x05 /* SPPT: Slow Package Power Tracking */
> @@ -30,6 +36,30 @@
> #define GET_STT_LIMIT_APU 0x20
> #define GET_STT_LIMIT_HS2 0x21
>
> +/* AMD PMF BIOS interfaces */
> +struct apmf_if_functions {
> + bool system_params;
> +};
> +
> +struct apmf_if {
> + acpi_handle handle;
> + struct apmf_if_functions func;
> +};
> +
> +struct apmf_verify_interface {
> + u16 size;
> + u16 version;
> + u32 notification_mask;
> + u32 supported_functions;
> +} __packed;
> +
> +struct apmf_system_params {
> + u16 size;
> + u32 valid_mask;
> + u32 flags;
> + u8 command_code;
> +} __packed;
> +
> struct amd_pmf_dev {
> void __iomem *regbase;
> void __iomem *smu_virt_addr;
> @@ -37,10 +67,12 @@ struct amd_pmf_dev {
> u32 base_addr;
> u32 cpu_id;
> struct device *dev;
> + struct apmf_if *apmf_if;
> struct mutex lock; /* protects the PMF interface */
> };
>
> /* Core Layer */
> +int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
> int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32 *data);
>
> #endif /* PMF_H */
^ permalink raw reply [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 02/11] platform/x86/amd/pmf: Add support for PMF APCI layer
2022-08-01 12:48 ` Hans de Goede
@ 2022-08-01 13:18 ` Hans de Goede
0 siblings, 0 replies; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 13:18 UTC (permalink / raw)
To: Shyam Sundar S K, markgross; +Cc: platform-driver-x86, Patil.Reddy
Hi,
On 8/1/22 14:48, Hans de Goede wrote:
> Hi,
>
> On 7/28/22 20:20, Shyam Sundar S K wrote:
>> PMF driver implements the ACPI methods as defined by AMD for PMF Support.
>> The ACPI layer acts as a glue that helps in providing the infrastructure
>> for OEMs customization.
>>
>> OEMs can refer to PMF support documentation to decide on the list of
>> functions to be supported on their specific platform model.
>>
>> AMD mandates that PMF ACPI fn0 and fn1 to be implemented which
>> provides the set of functions, params and the notifications that
>> would be sent to PMF driver so that PMF driver can adapt and
>> react.
>>
>> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
>> ---
>> drivers/platform/x86/amd/pmf/Makefile | 2 +-
>> drivers/platform/x86/amd/pmf/acpi.c | 182 ++++++++++++++++++++++++++
>> drivers/platform/x86/amd/pmf/core.c | 1 +
>> drivers/platform/x86/amd/pmf/pmf.h | 32 +++++
>> 4 files changed, 216 insertions(+), 1 deletion(-)
>> create mode 100644 drivers/platform/x86/amd/pmf/acpi.c
>>
>> diff --git a/drivers/platform/x86/amd/pmf/Makefile b/drivers/platform/x86/amd/pmf/Makefile
>> index 459005f659e5..2617eba773ce 100644
>> --- a/drivers/platform/x86/amd/pmf/Makefile
>> +++ b/drivers/platform/x86/amd/pmf/Makefile
>> @@ -5,4 +5,4 @@
>> #
>>
>> obj-$(CONFIG_AMD_PMF) += amd-pmf.o
>> -amd-pmf-objs := core.o
>> +amd-pmf-objs := core.o acpi.o
>> diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
>> new file mode 100644
>> index 000000000000..bd796d7e1d96
>> --- /dev/null
>> +++ b/drivers/platform/x86/amd/pmf/acpi.c
>> @@ -0,0 +1,182 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +/*
>> + * AMD Platform Management Framework Driver
>> + *
>> + * Copyright (c) 2022, Advanced Micro Devices, Inc.
>> + * All Rights Reserved.
>> + *
>> + * Author: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
>> + */
>> +
>> +#include <linux/acpi.h>
>> +#include "pmf.h"
>> +
>> +#define APMF_METHOD "\\_SB_.PMF_.APMF"
>> +
>> +static unsigned long supported_func;
>> +
>> +static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask);
>> +
>> +static union acpi_object *apmf_if_call(struct apmf_if *apmf_if, int func, struct acpi_buffer *param)
>> +{
>> + struct acpi_buffer buffer = { ACPI_ALLOCATE_BUFFER, NULL };
>> + struct acpi_object_list apmf_if_arg_list;
>> + union acpi_object apmf_if_args[2];
>> + acpi_status status;
>> +
>> + apmf_if_arg_list.count = 2;
>> + apmf_if_arg_list.pointer = &apmf_if_args[0];
>> +
>> + apmf_if_args[0].type = ACPI_TYPE_INTEGER;
>> + apmf_if_args[0].integer.value = func;
>> +
>> + if (param) {
>> + apmf_if_args[1].type = ACPI_TYPE_BUFFER;
>> + apmf_if_args[1].buffer.length = param->length;
>> + apmf_if_args[1].buffer.pointer = param->pointer;
>> + } else {
>> + apmf_if_args[1].type = ACPI_TYPE_INTEGER;
>> + apmf_if_args[1].integer.value = 0;
>> + }
>> +
>> + status = acpi_evaluate_object(apmf_if->handle, NULL, &apmf_if_arg_list, &buffer);
>> + if (ACPI_FAILURE(status)) {
>> + pr_err("PMF: APMF method call failed\n");
Since the plan is to drop struct apmf_if and to directly store the
handle in amd_pmf_dev we will now have an amd_pmf_dev here so
please use dev_err() instead of pr_err() here.
>> + if (status != AE_NOT_FOUND)
>> + kfree(buffer.pointer);
>> +
>> + return NULL;
>> + }
>> +
>> + return buffer.pointer;
>> +}
>> +
>> +static int apmf_if_call_store_buffer(struct apmf_if *apmf_if, int func, void *dest, size_t out_sz)
>> +{
>> + union acpi_object *info;
>> + size_t size;
>> + int err = 0;
>> +
>> + info = apmf_if_call(apmf_if, func, NULL);
>> + if (!info)
>> + return -EIO;
>> +
>> + if (info->type != ACPI_TYPE_BUFFER) {
>> + pr_err("object is not a buffer\n");
Since the plan is to drop struct apmf_if and to directly store the
handle in amd_pmf_dev we will now have an amd_pmf_dev here so
please use dev_err() instead of pr_err() here.
This also means that adding the:
#define pr_fmt(fmt) "AMD-PMF: " fmt
line will no longer be necessary.
>> + err = -EINVAL;
>> + goto out;
>> + }
>> +
>> + if (info->buffer.length < 2) {
>> + pr_err("buffer too small\n");
>> + err = -EINVAL;
>> + goto out;
>> + }
>> +
>> + size = *(u16 *)info->buffer.pointer;
>> + if (info->buffer.length < size) {
>> + pr_err("buffer smaller then headersize %u < %zu\n",
>> + info->buffer.length, size);
>> + err = -EINVAL;
>> + goto out;
>> + }
>> +
>> + if (size < out_sz) {
>> + pr_err("buffer too small %zu\n", size);
>> + err = -EINVAL;
>> + goto out;
>> + }
>> +
>> + size = out_sz;
>> + memcpy(dest, info->buffer.pointer, size);
>> +
>> +out:
>> + kfree(info);
>> + return err;
>> +}
>> +
>> +static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
>> +{
>> + func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
>
> I just noticed that this is not correct, this should be:
>
> func->system_params = mask & BIT(APMF_FUNC_GET_SYS_PARAMS - 1);
>
> Which for APMF_FUNC_GET_SYS_PARAMS happens to be the same, but for
> other AMPF_FUNC_* values the difference does matter!
>
> Also this information is duplicated by:
>
> is_apmf_func_supported(APMF_FUNC_GET_SYS_PARAMS) please drop the
> apmf_if_functions struct and use is_apmf_func_supported() everywhere.
>
> Note this means that you also will need to drop func->sbios_heartbeat
> which does not come from the mask and which is initialized in
> a different place rather then inside apmf_if_parse_functions() which
> already shows that sbios_hearbeat does not really belong inside
> this struct.
>
> Instead of moving the sbios_heartbeat flag, please just directly
> use notify_cfg.hb_interval inside apmf_acpi_deinit(). This will also
> make the cancel_delayed_work_sync(&pmf_dev->heart_beat) condition
> match the queue condition 1:1.
>
>
>> +}
>> +
>> +int is_apmf_func_supported(unsigned long index)
>> +{
>> + /* If bit-n is set, that indicates function n+1 is supported */
>> + return ((supported_func & (1 << (index - 1))) != 0);
>
> Please write this as:
>
> return (supported_func & BIT(index - 1)) != 0;
>
> or as:
>
> return !!(supported_func & BIT(index - 1));
>
>
Also please make this function take a "struct amd_pmf_dev *pdev" argument
and add an "u32 supported_func" to amd_pmf_dev instead of making it a global
variable.
Regards,
Hans
>> +}
>> +
>> +static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
>> +{
>> + struct apmf_verify_interface output;
>> + int err;
>> +
>> + err = apmf_if_call_store_buffer(apmf_if, APMF_FUNC_VERIFY_INTERFACE,
>> + &output, sizeof(output));
>> + if (err)
>> + return err;
>> +
>> + apmf_if_parse_functions(&apmf_if->func, output.supported_functions);
>> + supported_func = output.supported_functions;
>> + dev_dbg(pdev->dev, "supported functions:0x%x notifications:0x%x\n",
>> + output.supported_functions, output.notification_mask);
>> +
>> + return 0;
>> +}
>> +
>> +static int apmf_get_system_params(struct apmf_if *apmf_if)
>> +{
>> + struct apmf_system_params params;
>> + int err;
>> +
>> + if (apmf_if->func.system_params) {
>> + err = apmf_if_call_store_buffer(apmf_if, APMF_FUNC_GET_SYS_PARAMS,
>> + ¶ms, sizeof(params));
>> + if (err)
>> + return err;
>> + }
>> +
>> + pr_debug("PMF: system params mask:0x%x flags:0x%x cmd_code:0x%x\n",
>> + params.valid_mask,
>> + params.flags,
>> + params.command_code);
>> + params.flags = params.flags & params.valid_mask;
>> +
>> + return 0;
>> +}
>> +
>> +int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
>> +{
>> + acpi_handle apmf_if_handle;
>> + struct apmf_if *apmf_if;
>> + acpi_status status;
>> + int ret;
>> +
>> + status = acpi_get_handle(NULL, (acpi_string) APMF_METHOD, &apmf_if_handle);
>> + if (ACPI_FAILURE(status))
>> + return -ENODEV;
>> +
>> + apmf_if = kzalloc(sizeof(*apmf_if), GFP_KERNEL);
>> + if (!apmf_if)
>> + goto out;
>> +
>> + apmf_if->handle = apmf_if_handle;
>> +
>> + ret = apmf_if_verify_interface(pmf_dev, apmf_if);
>> + if (ret) {
>> + dev_err(pmf_dev->dev, "APMF verify interface failed :%d\n", ret);
>> + kfree(apmf_if);
>> + goto out;
>> + }
>> + pmf_dev->apmf_if = apmf_if;
>
> After dropping the apmf_if_functions sub-struct from amp_if amp_if only contains
> an acpi_handle. Please just store the handle (and later on also the struct
> apmf_notification_cfg) directly inside struct amd_pmf_dev, this also removes
> the need for doing the whole alloc + free + associated error handling dance here.
>
> Regsards,
>
> Hans
>
>
>> +
>> + ret = apmf_get_system_params(apmf_if);
>> + if (ret) {
>> + dev_err(pmf_dev->dev, "APMF apmf_get_system_params failed :%d\n", ret);
>> + kfree(apmf_if);
>> + goto out;
>> + }
>> +
>> +out:
>> + return ret;
>> +}
>> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
>> index aef97965c181..c5002b7bb904 100644
>> --- a/drivers/platform/x86/amd/pmf/core.c
>> +++ b/drivers/platform/x86/amd/pmf/core.c
>> @@ -204,6 +204,7 @@ static int amd_pmf_probe(struct platform_device *pdev)
>> if (!dev->regbase)
>> return -ENOMEM;
>>
>> + apmf_acpi_init(dev);
>> platform_set_drvdata(pdev, dev);
>>
>> mutex_init(&dev->lock);
>> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
>> index 1c2e942e5096..08c6bc0e2b42 100644
>> --- a/drivers/platform/x86/amd/pmf/pmf.h
>> +++ b/drivers/platform/x86/amd/pmf/pmf.h
>> @@ -11,6 +11,12 @@
>> #ifndef PMF_H
>> #define PMF_H
>>
>> +#include <linux/acpi.h>
>> +
>> +/* APMF Functions */
>> +#define APMF_FUNC_VERIFY_INTERFACE 0
>> +#define APMF_FUNC_GET_SYS_PARAMS 1
>> +
>> /* Message Definitions */
>> #define SET_SPL 0x03 /* SPL: Sustained Power Limit */
>> #define SET_SPPT 0x05 /* SPPT: Slow Package Power Tracking */
>> @@ -30,6 +36,30 @@
>> #define GET_STT_LIMIT_APU 0x20
>> #define GET_STT_LIMIT_HS2 0x21
>>
>> +/* AMD PMF BIOS interfaces */
>> +struct apmf_if_functions {
>> + bool system_params;
>> +};
>> +
>> +struct apmf_if {
>> + acpi_handle handle;
>> + struct apmf_if_functions func;
>> +};
>> +
>> +struct apmf_verify_interface {
>> + u16 size;
>> + u16 version;
>> + u32 notification_mask;
>> + u32 supported_functions;
>> +} __packed;
>> +
>> +struct apmf_system_params {
>> + u16 size;
>> + u32 valid_mask;
>> + u32 flags;
>> + u8 command_code;
>> +} __packed;
>> +
>> struct amd_pmf_dev {
>> void __iomem *regbase;
>> void __iomem *smu_virt_addr;
>> @@ -37,10 +67,12 @@ struct amd_pmf_dev {
>> u32 base_addr;
>> u32 cpu_id;
>> struct device *dev;
>> + struct apmf_if *apmf_if;
>> struct mutex lock; /* protects the PMF interface */
>> };
>>
>> /* Core Layer */
>> +int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
>> int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32 *data);
>>
>> #endif /* PMF_H */
^ permalink raw reply [flat|nested] 33+ messages in thread
* [PATCH v2 RESEND 03/11] platform/x86/amd/pmf: Add support SPS PMF feature
2022-07-28 18:20 [PATCH v2 RESEND 00/11] platform/x86/amd/pmf: Introduce AMD PMF Driver Shyam Sundar S K
2022-07-28 18:20 ` [PATCH v2 RESEND 01/11] platform/x86/amd/pmf: Add support for PMF core layer Shyam Sundar S K
2022-07-28 18:20 ` [PATCH v2 RESEND 02/11] platform/x86/amd/pmf: Add support for PMF APCI layer Shyam Sundar S K
@ 2022-07-28 18:20 ` Shyam Sundar S K
2022-08-01 11:30 ` Hans de Goede
2022-08-01 12:50 ` Hans de Goede
2022-07-28 18:20 ` [PATCH v2 RESEND 04/11] platform/x86/amd/pmf: Add debugfs information Shyam Sundar S K
` (7 subsequent siblings)
10 siblings, 2 replies; 33+ messages in thread
From: Shyam Sundar S K @ 2022-07-28 18:20 UTC (permalink / raw)
To: hdegoede, markgross; +Cc: platform-driver-x86, Patil.Reddy, Shyam Sundar S K
SPS (a.k.a. Static Power Slider) gives a feel of Windows performance
power slider for the Linux users, where the user selects a certain
mode (like "balanced", "low-power" or "performance") and the thermals
associated with each selected mode gets applied from the silicon
side via the mailboxes defined through PMFW.
PMF driver hooks to platform_profile by reading the PMF ACPI fn9 to
see if the support is being advertised by ACPI interface.
If supported, the PMF driver reacts to platform_profile selection choices
made by the user and adjust the system thermal behavior.
Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
---
drivers/platform/x86/amd/pmf/Makefile | 2 +-
drivers/platform/x86/amd/pmf/acpi.c | 8 ++
drivers/platform/x86/amd/pmf/core.c | 26 +++++
drivers/platform/x86/amd/pmf/pmf.h | 64 +++++++++++
drivers/platform/x86/amd/pmf/sps.c | 152 ++++++++++++++++++++++++++
5 files changed, 251 insertions(+), 1 deletion(-)
create mode 100644 drivers/platform/x86/amd/pmf/sps.c
diff --git a/drivers/platform/x86/amd/pmf/Makefile b/drivers/platform/x86/amd/pmf/Makefile
index 2617eba773ce..557521a80427 100644
--- a/drivers/platform/x86/amd/pmf/Makefile
+++ b/drivers/platform/x86/amd/pmf/Makefile
@@ -5,4 +5,4 @@
#
obj-$(CONFIG_AMD_PMF) += amd-pmf.o
-amd-pmf-objs := core.o acpi.o
+amd-pmf-objs := core.o acpi.o sps.o
diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
index bd796d7e1d96..41abd8680d8d 100644
--- a/drivers/platform/x86/amd/pmf/acpi.c
+++ b/drivers/platform/x86/amd/pmf/acpi.c
@@ -98,6 +98,7 @@ static int apmf_if_call_store_buffer(struct apmf_if *apmf_if, int func, void *de
static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
{
func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
+ func->static_slider_granular = mask & APMF_FUNC_STATIC_SLIDER_GRANULAR;
}
int is_apmf_func_supported(unsigned long index)
@@ -106,6 +107,13 @@ int is_apmf_func_supported(unsigned long index)
return ((supported_func & (1 << (index - 1))) != 0);
}
+int apmf_get_static_slider_granular(struct apmf_if *apmf_if,
+ struct apmf_static_slider_granular_output *data)
+{
+ return apmf_if_call_store_buffer(apmf_if, APMF_FUNC_STATIC_SLIDER_GRANULAR,
+ data, sizeof(*data));
+}
+
static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
{
struct apmf_verify_interface output;
diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
index c5002b7bb904..b6006e8ee1a1 100644
--- a/drivers/platform/x86/amd/pmf/core.c
+++ b/drivers/platform/x86/amd/pmf/core.c
@@ -12,6 +12,7 @@
#include <linux/module.h>
#include <linux/pci.h>
#include <linux/platform_device.h>
+#include <linux/power_supply.h>
#include "pmf.h"
/* PMF-SMU communication registers */
@@ -45,6 +46,14 @@
#define DELAY_MIN_US 2000
#define DELAY_MAX_US 3000
+int amd_pmf_get_power_source(void)
+{
+ if (power_supply_is_system_supplied() > 0)
+ return POWER_SOURCE_AC;
+ else
+ return POWER_SOURCE_DC;
+}
+
static inline u32 amd_pmf_reg_read(struct amd_pmf_dev *dev, int reg_offset)
{
return ioread32(dev->regbase + reg_offset);
@@ -138,6 +147,21 @@ static const struct pci_device_id pmf_pci_ids[] = {
{ }
};
+static void amd_pmf_init_features(struct amd_pmf_dev *dev)
+{
+ /* Enable Static Slider */
+ if (is_apmf_func_supported(APMF_FUNC_STATIC_SLIDER_GRANULAR)) {
+ amd_pmf_init_sps(dev);
+ dev_dbg(dev->dev, "SPS enabled and Platform Profiles registered\n");
+ }
+}
+
+static void amd_pmf_deinit_features(struct amd_pmf_dev *dev)
+{
+ if (is_apmf_func_supported(APMF_FUNC_STATIC_SLIDER_GRANULAR))
+ amd_pmf_deinit_sps(dev);
+}
+
static const struct acpi_device_id amd_pmf_acpi_ids[] = {
{"AMDI0102", 0},
{ }
@@ -206,6 +230,7 @@ static int amd_pmf_probe(struct platform_device *pdev)
apmf_acpi_init(dev);
platform_set_drvdata(pdev, dev);
+ amd_pmf_init_features(dev);
mutex_init(&dev->lock);
dev_info(dev->dev, "registered PMF device successfully\n");
@@ -218,6 +243,7 @@ static int amd_pmf_remove(struct platform_device *pdev)
struct amd_pmf_dev *dev = platform_get_drvdata(pdev);
mutex_destroy(&dev->lock);
+ amd_pmf_deinit_features(dev);
kfree(dev->buf);
return 0;
}
diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
index 08c6bc0e2b42..30740a5cd30d 100644
--- a/drivers/platform/x86/amd/pmf/pmf.h
+++ b/drivers/platform/x86/amd/pmf/pmf.h
@@ -12,10 +12,12 @@
#define PMF_H
#include <linux/acpi.h>
+#include <linux/platform_profile.h>
/* APMF Functions */
#define APMF_FUNC_VERIFY_INTERFACE 0
#define APMF_FUNC_GET_SYS_PARAMS 1
+#define APMF_FUNC_STATIC_SLIDER_GRANULAR 9
/* Message Definitions */
#define SET_SPL 0x03 /* SPL: Sustained Power Limit */
@@ -36,9 +38,12 @@
#define GET_STT_LIMIT_APU 0x20
#define GET_STT_LIMIT_HS2 0x21
+#define ARG_NONE 0
+
/* AMD PMF BIOS interfaces */
struct apmf_if_functions {
bool system_params;
+ bool static_slider_granular;
};
struct apmf_if {
@@ -60,6 +65,30 @@ struct apmf_system_params {
u8 command_code;
} __packed;
+enum amd_stt_skin_temp {
+ STT_TEMP_APU,
+ STT_TEMP_HS2,
+ STT_TEMP_COUNT,
+};
+
+enum amd_slider_op {
+ SLIDER_OP_GET,
+ SLIDER_OP_SET,
+};
+
+enum power_source {
+ POWER_SOURCE_AC,
+ POWER_SOURCE_DC,
+ POWER_SOURCE_MAX,
+};
+
+enum power_modes {
+ POWER_MODE_PERFORMANCE,
+ POWER_MODE_BALANCED_POWER,
+ POWER_MODE_POWER_SAVER,
+ POWER_MODE_MAX,
+};
+
struct amd_pmf_dev {
void __iomem *regbase;
void __iomem *smu_virt_addr;
@@ -68,11 +97,46 @@ struct amd_pmf_dev {
u32 cpu_id;
struct device *dev;
struct apmf_if *apmf_if;
+ enum platform_profile_option current_profile;
+ struct platform_profile_handler pprof;
struct mutex lock; /* protects the PMF interface */
};
+struct apmf_sps_prop_granular {
+ u32 fppt;
+ u32 sppt;
+ u32 sppt_apu_only;
+ u32 spl;
+ u32 stt_min;
+ u8 stt_skin_temp[STT_TEMP_COUNT];
+ u32 fan_id;
+} __packed;
+
+/* Static Slider */
+struct apmf_static_slider_granular_output {
+ u16 size;
+ struct apmf_sps_prop_granular prop[POWER_SOURCE_MAX * POWER_MODE_MAX];
+} __packed;
+
+struct amd_pmf_static_slider_granular {
+ u16 size;
+ struct apmf_sps_prop_granular prop[POWER_SOURCE_MAX][POWER_MODE_MAX];
+};
+
/* Core Layer */
int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
+int is_apmf_func_supported(unsigned long index);
int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32 *data);
+int amd_pmf_get_power_source(void);
+
+/* SPS Layer */
+u8 amd_pmf_get_pprof_modes(struct amd_pmf_dev *pmf);
+void amd_pmf_update_slider(struct amd_pmf_dev *dev, bool op, int idx,
+ struct amd_pmf_static_slider_granular *table);
+int amd_pmf_init_sps(struct amd_pmf_dev *dev);
+void amd_pmf_deinit_sps(struct amd_pmf_dev *dev);
+int apmf_get_static_slider_granular(struct apmf_if *ampf_if,
+ struct apmf_static_slider_granular_output *output);
+void amd_pmf_load_defaults_sps(struct amd_pmf_dev *dev);
#endif /* PMF_H */
diff --git a/drivers/platform/x86/amd/pmf/sps.c b/drivers/platform/x86/amd/pmf/sps.c
new file mode 100644
index 000000000000..2601adbfe14a
--- /dev/null
+++ b/drivers/platform/x86/amd/pmf/sps.c
@@ -0,0 +1,152 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * AMD Platform Management Framework (PMF) Driver
+ *
+ * Copyright (c) 2022, Advanced Micro Devices, Inc.
+ * All Rights Reserved.
+ *
+ * Author: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
+ */
+
+#include "pmf.h"
+
+static int amd_pmf_profile_get(struct platform_profile_handler *pprof,
+ enum platform_profile_option *profile);
+static int amd_pmf_profile_set(struct platform_profile_handler *pprof,
+ enum platform_profile_option profile);
+static struct amd_pmf_static_slider_granular config_store;
+
+void amd_pmf_load_defaults_sps(struct amd_pmf_dev *dev)
+{
+ struct apmf_static_slider_granular_output output;
+ int i, j, idx = 0;
+
+ memset(&config_store, 0, sizeof(config_store));
+
+ if (dev->apmf_if->func.static_slider_granular) {
+ apmf_get_static_slider_granular(dev->apmf_if, &output);
+
+ for (i = 0; i < POWER_SOURCE_MAX; i++) {
+ for (j = 0; j < POWER_MODE_MAX; j++) {
+ config_store.prop[i][j].spl = output.prop[idx].spl;
+ config_store.prop[i][j].sppt = output.prop[idx].sppt;
+ config_store.prop[i][j].sppt_apu_only =
+ output.prop[idx].sppt_apu_only;
+ config_store.prop[i][j].fppt = output.prop[idx].fppt;
+ config_store.prop[i][j].stt_min = output.prop[idx].stt_min;
+ config_store.prop[i][j].stt_skin_temp[STT_TEMP_APU] =
+ output.prop[idx].stt_skin_temp[STT_TEMP_APU];
+ config_store.prop[i][j].stt_skin_temp[STT_TEMP_HS2] =
+ output.prop[idx].stt_skin_temp[STT_TEMP_HS2];
+ config_store.prop[i][j].fan_id = output.prop[idx].fan_id;
+ idx++;
+ }
+ }
+ }
+}
+
+void amd_pmf_update_slider(struct amd_pmf_dev *dev, bool op, int idx,
+ struct amd_pmf_static_slider_granular *table)
+{
+ int src = amd_pmf_get_power_source();
+
+ if (op == SLIDER_OP_SET) {
+ amd_pmf_send_cmd(dev, SET_SPL, false, config_store.prop[src][idx].spl, NULL);
+ amd_pmf_send_cmd(dev, SET_FPPT, false, config_store.prop[src][idx].fppt, NULL);
+ amd_pmf_send_cmd(dev, SET_SPPT, false, config_store.prop[src][idx].sppt, NULL);
+ amd_pmf_send_cmd(dev, SET_SPPT_APU_ONLY, false,
+ config_store.prop[src][idx].sppt_apu_only, NULL);
+ amd_pmf_send_cmd(dev, SET_STT_MIN_LIMIT, false,
+ config_store.prop[src][idx].stt_min, NULL);
+ amd_pmf_send_cmd(dev, SET_STT_LIMIT_APU, false,
+ config_store.prop[src][idx].stt_skin_temp[STT_TEMP_APU], NULL);
+ amd_pmf_send_cmd(dev, SET_STT_LIMIT_HS2, false,
+ config_store.prop[src][idx].stt_skin_temp[STT_TEMP_HS2], NULL);
+ } else if (op == SLIDER_OP_GET) {
+ amd_pmf_send_cmd(dev, GET_SPL, true, ARG_NONE, &table->prop[src][idx].spl);
+ amd_pmf_send_cmd(dev, GET_FPPT, true, ARG_NONE, &table->prop[src][idx].fppt);
+ amd_pmf_send_cmd(dev, GET_SPPT, true, ARG_NONE, &table->prop[src][idx].sppt);
+ amd_pmf_send_cmd(dev, GET_SPPT_APU_ONLY, true, ARG_NONE,
+ &table->prop[src][idx].sppt_apu_only);
+ amd_pmf_send_cmd(dev, GET_STT_MIN_LIMIT, true, ARG_NONE,
+ &table->prop[src][idx].stt_min);
+ amd_pmf_send_cmd(dev, GET_STT_LIMIT_APU, true, ARG_NONE,
+ (u32 *)&table->prop[src][idx].stt_skin_temp[STT_TEMP_APU]);
+ amd_pmf_send_cmd(dev, GET_STT_LIMIT_HS2, true, ARG_NONE,
+ (u32 *)&table->prop[src][idx].stt_skin_temp[STT_TEMP_HS2]);
+ }
+}
+
+static int amd_pmf_profile_get(struct platform_profile_handler *pprof,
+ enum platform_profile_option *profile)
+{
+ struct amd_pmf_dev *pmf = container_of(pprof, struct amd_pmf_dev, pprof);
+
+ *profile = pmf->current_profile;
+ return 0;
+}
+
+u8 amd_pmf_get_pprof_modes(struct amd_pmf_dev *pmf)
+{
+ u8 mode;
+
+ switch (pmf->current_profile) {
+ case PLATFORM_PROFILE_PERFORMANCE:
+ mode = POWER_MODE_PERFORMANCE;
+ break;
+ case PLATFORM_PROFILE_BALANCED:
+ mode = POWER_MODE_BALANCED_POWER;
+ break;
+ case PLATFORM_PROFILE_LOW_POWER:
+ mode = POWER_MODE_POWER_SAVER;
+ break;
+ default:
+ dev_err(pmf->dev, "Unknown Platform Profile.\n");
+ break;
+ }
+
+ return mode;
+}
+
+int amd_pmf_profile_set(struct platform_profile_handler *pprof,
+ enum platform_profile_option profile)
+{
+ struct amd_pmf_dev *pmf = container_of(pprof, struct amd_pmf_dev, pprof);
+ u8 mode;
+
+ pmf->current_profile = profile;
+ mode = amd_pmf_get_pprof_modes(pmf);
+ amd_pmf_update_slider(pmf, SLIDER_OP_SET, mode, NULL);
+ return 0;
+}
+
+int amd_pmf_init_sps(struct amd_pmf_dev *dev)
+{
+ int err = 0;
+
+ dev->pprof.profile_get = amd_pmf_profile_get;
+ dev->pprof.profile_set = amd_pmf_profile_set;
+
+ /* Setup supported modes */
+ set_bit(PLATFORM_PROFILE_LOW_POWER, dev->pprof.choices);
+ set_bit(PLATFORM_PROFILE_BALANCED, dev->pprof.choices);
+ set_bit(PLATFORM_PROFILE_PERFORMANCE, dev->pprof.choices);
+
+ /* Create platform_profile structure and register */
+ err = platform_profile_register(&dev->pprof);
+ if (err) {
+ dev_err(dev->dev, "Failed to register SPS support, this is most likely an SBIOS bug: %d\n",
+ err);
+ return -EEXIST;
+ }
+
+ dev->current_profile = PLATFORM_PROFILE_BALANCED;
+ amd_pmf_load_defaults_sps(dev);
+
+ return err;
+}
+
+void amd_pmf_deinit_sps(struct amd_pmf_dev *dev)
+{
+ platform_profile_remove();
+}
--
2.25.1
^ permalink raw reply related [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 03/11] platform/x86/amd/pmf: Add support SPS PMF feature
2022-07-28 18:20 ` [PATCH v2 RESEND 03/11] platform/x86/amd/pmf: Add support SPS PMF feature Shyam Sundar S K
@ 2022-08-01 11:30 ` Hans de Goede
2022-08-01 12:50 ` Hans de Goede
1 sibling, 0 replies; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 11:30 UTC (permalink / raw)
To: Shyam Sundar S K, markgross; +Cc: platform-driver-x86, Patil.Reddy
Hi,
On 7/28/22 20:20, Shyam Sundar S K wrote:
> SPS (a.k.a. Static Power Slider) gives a feel of Windows performance
> power slider for the Linux users, where the user selects a certain
> mode (like "balanced", "low-power" or "performance") and the thermals
> associated with each selected mode gets applied from the silicon
> side via the mailboxes defined through PMFW.
>
> PMF driver hooks to platform_profile by reading the PMF ACPI fn9 to
> see if the support is being advertised by ACPI interface.
>
> If supported, the PMF driver reacts to platform_profile selection choices
> made by the user and adjust the system thermal behavior.
>
> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
Thanks, version 2 looks good to me:
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Regards,
Hans
> ---
> drivers/platform/x86/amd/pmf/Makefile | 2 +-
> drivers/platform/x86/amd/pmf/acpi.c | 8 ++
> drivers/platform/x86/amd/pmf/core.c | 26 +++++
> drivers/platform/x86/amd/pmf/pmf.h | 64 +++++++++++
> drivers/platform/x86/amd/pmf/sps.c | 152 ++++++++++++++++++++++++++
> 5 files changed, 251 insertions(+), 1 deletion(-)
> create mode 100644 drivers/platform/x86/amd/pmf/sps.c
>
> diff --git a/drivers/platform/x86/amd/pmf/Makefile b/drivers/platform/x86/amd/pmf/Makefile
> index 2617eba773ce..557521a80427 100644
> --- a/drivers/platform/x86/amd/pmf/Makefile
> +++ b/drivers/platform/x86/amd/pmf/Makefile
> @@ -5,4 +5,4 @@
> #
>
> obj-$(CONFIG_AMD_PMF) += amd-pmf.o
> -amd-pmf-objs := core.o acpi.o
> +amd-pmf-objs := core.o acpi.o sps.o
> diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
> index bd796d7e1d96..41abd8680d8d 100644
> --- a/drivers/platform/x86/amd/pmf/acpi.c
> +++ b/drivers/platform/x86/amd/pmf/acpi.c
> @@ -98,6 +98,7 @@ static int apmf_if_call_store_buffer(struct apmf_if *apmf_if, int func, void *de
> static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
> {
> func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
> + func->static_slider_granular = mask & APMF_FUNC_STATIC_SLIDER_GRANULAR;
> }
>
> int is_apmf_func_supported(unsigned long index)
> @@ -106,6 +107,13 @@ int is_apmf_func_supported(unsigned long index)
> return ((supported_func & (1 << (index - 1))) != 0);
> }
>
> +int apmf_get_static_slider_granular(struct apmf_if *apmf_if,
> + struct apmf_static_slider_granular_output *data)
> +{
> + return apmf_if_call_store_buffer(apmf_if, APMF_FUNC_STATIC_SLIDER_GRANULAR,
> + data, sizeof(*data));
> +}
> +
> static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
> {
> struct apmf_verify_interface output;
> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
> index c5002b7bb904..b6006e8ee1a1 100644
> --- a/drivers/platform/x86/amd/pmf/core.c
> +++ b/drivers/platform/x86/amd/pmf/core.c
> @@ -12,6 +12,7 @@
> #include <linux/module.h>
> #include <linux/pci.h>
> #include <linux/platform_device.h>
> +#include <linux/power_supply.h>
> #include "pmf.h"
>
> /* PMF-SMU communication registers */
> @@ -45,6 +46,14 @@
> #define DELAY_MIN_US 2000
> #define DELAY_MAX_US 3000
>
> +int amd_pmf_get_power_source(void)
> +{
> + if (power_supply_is_system_supplied() > 0)
> + return POWER_SOURCE_AC;
> + else
> + return POWER_SOURCE_DC;
> +}
> +
> static inline u32 amd_pmf_reg_read(struct amd_pmf_dev *dev, int reg_offset)
> {
> return ioread32(dev->regbase + reg_offset);
> @@ -138,6 +147,21 @@ static const struct pci_device_id pmf_pci_ids[] = {
> { }
> };
>
> +static void amd_pmf_init_features(struct amd_pmf_dev *dev)
> +{
> + /* Enable Static Slider */
> + if (is_apmf_func_supported(APMF_FUNC_STATIC_SLIDER_GRANULAR)) {
> + amd_pmf_init_sps(dev);
> + dev_dbg(dev->dev, "SPS enabled and Platform Profiles registered\n");
> + }
> +}
> +
> +static void amd_pmf_deinit_features(struct amd_pmf_dev *dev)
> +{
> + if (is_apmf_func_supported(APMF_FUNC_STATIC_SLIDER_GRANULAR))
> + amd_pmf_deinit_sps(dev);
> +}
> +
> static const struct acpi_device_id amd_pmf_acpi_ids[] = {
> {"AMDI0102", 0},
> { }
> @@ -206,6 +230,7 @@ static int amd_pmf_probe(struct platform_device *pdev)
>
> apmf_acpi_init(dev);
> platform_set_drvdata(pdev, dev);
> + amd_pmf_init_features(dev);
>
> mutex_init(&dev->lock);
> dev_info(dev->dev, "registered PMF device successfully\n");
> @@ -218,6 +243,7 @@ static int amd_pmf_remove(struct platform_device *pdev)
> struct amd_pmf_dev *dev = platform_get_drvdata(pdev);
>
> mutex_destroy(&dev->lock);
> + amd_pmf_deinit_features(dev);
> kfree(dev->buf);
> return 0;
> }
> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
> index 08c6bc0e2b42..30740a5cd30d 100644
> --- a/drivers/platform/x86/amd/pmf/pmf.h
> +++ b/drivers/platform/x86/amd/pmf/pmf.h
> @@ -12,10 +12,12 @@
> #define PMF_H
>
> #include <linux/acpi.h>
> +#include <linux/platform_profile.h>
>
> /* APMF Functions */
> #define APMF_FUNC_VERIFY_INTERFACE 0
> #define APMF_FUNC_GET_SYS_PARAMS 1
> +#define APMF_FUNC_STATIC_SLIDER_GRANULAR 9
>
> /* Message Definitions */
> #define SET_SPL 0x03 /* SPL: Sustained Power Limit */
> @@ -36,9 +38,12 @@
> #define GET_STT_LIMIT_APU 0x20
> #define GET_STT_LIMIT_HS2 0x21
>
> +#define ARG_NONE 0
> +
> /* AMD PMF BIOS interfaces */
> struct apmf_if_functions {
> bool system_params;
> + bool static_slider_granular;
> };
>
> struct apmf_if {
> @@ -60,6 +65,30 @@ struct apmf_system_params {
> u8 command_code;
> } __packed;
>
> +enum amd_stt_skin_temp {
> + STT_TEMP_APU,
> + STT_TEMP_HS2,
> + STT_TEMP_COUNT,
> +};
> +
> +enum amd_slider_op {
> + SLIDER_OP_GET,
> + SLIDER_OP_SET,
> +};
> +
> +enum power_source {
> + POWER_SOURCE_AC,
> + POWER_SOURCE_DC,
> + POWER_SOURCE_MAX,
> +};
> +
> +enum power_modes {
> + POWER_MODE_PERFORMANCE,
> + POWER_MODE_BALANCED_POWER,
> + POWER_MODE_POWER_SAVER,
> + POWER_MODE_MAX,
> +};
> +
> struct amd_pmf_dev {
> void __iomem *regbase;
> void __iomem *smu_virt_addr;
> @@ -68,11 +97,46 @@ struct amd_pmf_dev {
> u32 cpu_id;
> struct device *dev;
> struct apmf_if *apmf_if;
> + enum platform_profile_option current_profile;
> + struct platform_profile_handler pprof;
> struct mutex lock; /* protects the PMF interface */
> };
>
> +struct apmf_sps_prop_granular {
> + u32 fppt;
> + u32 sppt;
> + u32 sppt_apu_only;
> + u32 spl;
> + u32 stt_min;
> + u8 stt_skin_temp[STT_TEMP_COUNT];
> + u32 fan_id;
> +} __packed;
> +
> +/* Static Slider */
> +struct apmf_static_slider_granular_output {
> + u16 size;
> + struct apmf_sps_prop_granular prop[POWER_SOURCE_MAX * POWER_MODE_MAX];
> +} __packed;
> +
> +struct amd_pmf_static_slider_granular {
> + u16 size;
> + struct apmf_sps_prop_granular prop[POWER_SOURCE_MAX][POWER_MODE_MAX];
> +};
> +
> /* Core Layer */
> int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
> +int is_apmf_func_supported(unsigned long index);
> int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32 *data);
> +int amd_pmf_get_power_source(void);
> +
> +/* SPS Layer */
> +u8 amd_pmf_get_pprof_modes(struct amd_pmf_dev *pmf);
> +void amd_pmf_update_slider(struct amd_pmf_dev *dev, bool op, int idx,
> + struct amd_pmf_static_slider_granular *table);
> +int amd_pmf_init_sps(struct amd_pmf_dev *dev);
> +void amd_pmf_deinit_sps(struct amd_pmf_dev *dev);
> +int apmf_get_static_slider_granular(struct apmf_if *ampf_if,
> + struct apmf_static_slider_granular_output *output);
> +void amd_pmf_load_defaults_sps(struct amd_pmf_dev *dev);
>
> #endif /* PMF_H */
> diff --git a/drivers/platform/x86/amd/pmf/sps.c b/drivers/platform/x86/amd/pmf/sps.c
> new file mode 100644
> index 000000000000..2601adbfe14a
> --- /dev/null
> +++ b/drivers/platform/x86/amd/pmf/sps.c
> @@ -0,0 +1,152 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * AMD Platform Management Framework (PMF) Driver
> + *
> + * Copyright (c) 2022, Advanced Micro Devices, Inc.
> + * All Rights Reserved.
> + *
> + * Author: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> + */
> +
> +#include "pmf.h"
> +
> +static int amd_pmf_profile_get(struct platform_profile_handler *pprof,
> + enum platform_profile_option *profile);
> +static int amd_pmf_profile_set(struct platform_profile_handler *pprof,
> + enum platform_profile_option profile);
> +static struct amd_pmf_static_slider_granular config_store;
> +
> +void amd_pmf_load_defaults_sps(struct amd_pmf_dev *dev)
> +{
> + struct apmf_static_slider_granular_output output;
> + int i, j, idx = 0;
> +
> + memset(&config_store, 0, sizeof(config_store));
> +
> + if (dev->apmf_if->func.static_slider_granular) {
> + apmf_get_static_slider_granular(dev->apmf_if, &output);
> +
> + for (i = 0; i < POWER_SOURCE_MAX; i++) {
> + for (j = 0; j < POWER_MODE_MAX; j++) {
> + config_store.prop[i][j].spl = output.prop[idx].spl;
> + config_store.prop[i][j].sppt = output.prop[idx].sppt;
> + config_store.prop[i][j].sppt_apu_only =
> + output.prop[idx].sppt_apu_only;
> + config_store.prop[i][j].fppt = output.prop[idx].fppt;
> + config_store.prop[i][j].stt_min = output.prop[idx].stt_min;
> + config_store.prop[i][j].stt_skin_temp[STT_TEMP_APU] =
> + output.prop[idx].stt_skin_temp[STT_TEMP_APU];
> + config_store.prop[i][j].stt_skin_temp[STT_TEMP_HS2] =
> + output.prop[idx].stt_skin_temp[STT_TEMP_HS2];
> + config_store.prop[i][j].fan_id = output.prop[idx].fan_id;
> + idx++;
> + }
> + }
> + }
> +}
> +
> +void amd_pmf_update_slider(struct amd_pmf_dev *dev, bool op, int idx,
> + struct amd_pmf_static_slider_granular *table)
> +{
> + int src = amd_pmf_get_power_source();
> +
> + if (op == SLIDER_OP_SET) {
> + amd_pmf_send_cmd(dev, SET_SPL, false, config_store.prop[src][idx].spl, NULL);
> + amd_pmf_send_cmd(dev, SET_FPPT, false, config_store.prop[src][idx].fppt, NULL);
> + amd_pmf_send_cmd(dev, SET_SPPT, false, config_store.prop[src][idx].sppt, NULL);
> + amd_pmf_send_cmd(dev, SET_SPPT_APU_ONLY, false,
> + config_store.prop[src][idx].sppt_apu_only, NULL);
> + amd_pmf_send_cmd(dev, SET_STT_MIN_LIMIT, false,
> + config_store.prop[src][idx].stt_min, NULL);
> + amd_pmf_send_cmd(dev, SET_STT_LIMIT_APU, false,
> + config_store.prop[src][idx].stt_skin_temp[STT_TEMP_APU], NULL);
> + amd_pmf_send_cmd(dev, SET_STT_LIMIT_HS2, false,
> + config_store.prop[src][idx].stt_skin_temp[STT_TEMP_HS2], NULL);
> + } else if (op == SLIDER_OP_GET) {
> + amd_pmf_send_cmd(dev, GET_SPL, true, ARG_NONE, &table->prop[src][idx].spl);
> + amd_pmf_send_cmd(dev, GET_FPPT, true, ARG_NONE, &table->prop[src][idx].fppt);
> + amd_pmf_send_cmd(dev, GET_SPPT, true, ARG_NONE, &table->prop[src][idx].sppt);
> + amd_pmf_send_cmd(dev, GET_SPPT_APU_ONLY, true, ARG_NONE,
> + &table->prop[src][idx].sppt_apu_only);
> + amd_pmf_send_cmd(dev, GET_STT_MIN_LIMIT, true, ARG_NONE,
> + &table->prop[src][idx].stt_min);
> + amd_pmf_send_cmd(dev, GET_STT_LIMIT_APU, true, ARG_NONE,
> + (u32 *)&table->prop[src][idx].stt_skin_temp[STT_TEMP_APU]);
> + amd_pmf_send_cmd(dev, GET_STT_LIMIT_HS2, true, ARG_NONE,
> + (u32 *)&table->prop[src][idx].stt_skin_temp[STT_TEMP_HS2]);
> + }
> +}
> +
> +static int amd_pmf_profile_get(struct platform_profile_handler *pprof,
> + enum platform_profile_option *profile)
> +{
> + struct amd_pmf_dev *pmf = container_of(pprof, struct amd_pmf_dev, pprof);
> +
> + *profile = pmf->current_profile;
> + return 0;
> +}
> +
> +u8 amd_pmf_get_pprof_modes(struct amd_pmf_dev *pmf)
> +{
> + u8 mode;
> +
> + switch (pmf->current_profile) {
> + case PLATFORM_PROFILE_PERFORMANCE:
> + mode = POWER_MODE_PERFORMANCE;
> + break;
> + case PLATFORM_PROFILE_BALANCED:
> + mode = POWER_MODE_BALANCED_POWER;
> + break;
> + case PLATFORM_PROFILE_LOW_POWER:
> + mode = POWER_MODE_POWER_SAVER;
> + break;
> + default:
> + dev_err(pmf->dev, "Unknown Platform Profile.\n");
> + break;
> + }
> +
> + return mode;
> +}
> +
> +int amd_pmf_profile_set(struct platform_profile_handler *pprof,
> + enum platform_profile_option profile)
> +{
> + struct amd_pmf_dev *pmf = container_of(pprof, struct amd_pmf_dev, pprof);
> + u8 mode;
> +
> + pmf->current_profile = profile;
> + mode = amd_pmf_get_pprof_modes(pmf);
> + amd_pmf_update_slider(pmf, SLIDER_OP_SET, mode, NULL);
> + return 0;
> +}
> +
> +int amd_pmf_init_sps(struct amd_pmf_dev *dev)
> +{
> + int err = 0;
> +
> + dev->pprof.profile_get = amd_pmf_profile_get;
> + dev->pprof.profile_set = amd_pmf_profile_set;
> +
> + /* Setup supported modes */
> + set_bit(PLATFORM_PROFILE_LOW_POWER, dev->pprof.choices);
> + set_bit(PLATFORM_PROFILE_BALANCED, dev->pprof.choices);
> + set_bit(PLATFORM_PROFILE_PERFORMANCE, dev->pprof.choices);
> +
> + /* Create platform_profile structure and register */
> + err = platform_profile_register(&dev->pprof);
> + if (err) {
> + dev_err(dev->dev, "Failed to register SPS support, this is most likely an SBIOS bug: %d\n",
> + err);
> + return -EEXIST;
> + }
> +
> + dev->current_profile = PLATFORM_PROFILE_BALANCED;
> + amd_pmf_load_defaults_sps(dev);
> +
> + return err;
> +}
> +
> +void amd_pmf_deinit_sps(struct amd_pmf_dev *dev)
> +{
> + platform_profile_remove();
> +}
^ permalink raw reply [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 03/11] platform/x86/amd/pmf: Add support SPS PMF feature
2022-07-28 18:20 ` [PATCH v2 RESEND 03/11] platform/x86/amd/pmf: Add support SPS PMF feature Shyam Sundar S K
2022-08-01 11:30 ` Hans de Goede
@ 2022-08-01 12:50 ` Hans de Goede
1 sibling, 0 replies; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 12:50 UTC (permalink / raw)
To: Shyam Sundar S K, markgross; +Cc: platform-driver-x86, Patil.Reddy
Hi
On 7/28/22 20:20, Shyam Sundar S K wrote:
> SPS (a.k.a. Static Power Slider) gives a feel of Windows performance
> power slider for the Linux users, where the user selects a certain
> mode (like "balanced", "low-power" or "performance") and the thermals
> associated with each selected mode gets applied from the silicon
> side via the mailboxes defined through PMFW.
>
> PMF driver hooks to platform_profile by reading the PMF ACPI fn9 to
> see if the support is being advertised by ACPI interface.
>
> If supported, the PMF driver reacts to platform_profile selection choices
> made by the user and adjust the system thermal behavior.
>
> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> ---
> drivers/platform/x86/amd/pmf/Makefile | 2 +-
> drivers/platform/x86/amd/pmf/acpi.c | 8 ++
> drivers/platform/x86/amd/pmf/core.c | 26 +++++
> drivers/platform/x86/amd/pmf/pmf.h | 64 +++++++++++
> drivers/platform/x86/amd/pmf/sps.c | 152 ++++++++++++++++++++++++++
> 5 files changed, 251 insertions(+), 1 deletion(-)
> create mode 100644 drivers/platform/x86/amd/pmf/sps.c
>
> diff --git a/drivers/platform/x86/amd/pmf/Makefile b/drivers/platform/x86/amd/pmf/Makefile
> index 2617eba773ce..557521a80427 100644
> --- a/drivers/platform/x86/amd/pmf/Makefile
> +++ b/drivers/platform/x86/amd/pmf/Makefile
> @@ -5,4 +5,4 @@
> #
>
> obj-$(CONFIG_AMD_PMF) += amd-pmf.o
> -amd-pmf-objs := core.o acpi.o
> +amd-pmf-objs := core.o acpi.o sps.o
> diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
> index bd796d7e1d96..41abd8680d8d 100644
> --- a/drivers/platform/x86/amd/pmf/acpi.c
> +++ b/drivers/platform/x86/amd/pmf/acpi.c
> @@ -98,6 +98,7 @@ static int apmf_if_call_store_buffer(struct apmf_if *apmf_if, int func, void *de
> static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
> {
> func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
> + func->static_slider_granular = mask & APMF_FUNC_STATIC_SLIDER_GRANULAR;
I just noticed that this is not correct, this should be:
func->system_params = mask & BIT(APMF_FUNC_... - 1);
Which does something completely different!
Also this information is duplicated by:
is_apmf_func_supported(APMF_FUNC_...) so please drop the apmf_if_functions
struct completely and use is_apmf_func_supported() everywhere.
Regards,
Hans
> }
>
> int is_apmf_func_supported(unsigned long index)
> @@ -106,6 +107,13 @@ int is_apmf_func_supported(unsigned long index)
> return ((supported_func & (1 << (index - 1))) != 0);
> }
>
> +int apmf_get_static_slider_granular(struct apmf_if *apmf_if,
> + struct apmf_static_slider_granular_output *data)
> +{
> + return apmf_if_call_store_buffer(apmf_if, APMF_FUNC_STATIC_SLIDER_GRANULAR,
> + data, sizeof(*data));
> +}
> +
> static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
> {
> struct apmf_verify_interface output;
> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
> index c5002b7bb904..b6006e8ee1a1 100644
> --- a/drivers/platform/x86/amd/pmf/core.c
> +++ b/drivers/platform/x86/amd/pmf/core.c
> @@ -12,6 +12,7 @@
> #include <linux/module.h>
> #include <linux/pci.h>
> #include <linux/platform_device.h>
> +#include <linux/power_supply.h>
> #include "pmf.h"
>
> /* PMF-SMU communication registers */
> @@ -45,6 +46,14 @@
> #define DELAY_MIN_US 2000
> #define DELAY_MAX_US 3000
>
> +int amd_pmf_get_power_source(void)
> +{
> + if (power_supply_is_system_supplied() > 0)
> + return POWER_SOURCE_AC;
> + else
> + return POWER_SOURCE_DC;
> +}
> +
> static inline u32 amd_pmf_reg_read(struct amd_pmf_dev *dev, int reg_offset)
> {
> return ioread32(dev->regbase + reg_offset);
> @@ -138,6 +147,21 @@ static const struct pci_device_id pmf_pci_ids[] = {
> { }
> };
>
> +static void amd_pmf_init_features(struct amd_pmf_dev *dev)
> +{
> + /* Enable Static Slider */
> + if (is_apmf_func_supported(APMF_FUNC_STATIC_SLIDER_GRANULAR)) {
> + amd_pmf_init_sps(dev);
> + dev_dbg(dev->dev, "SPS enabled and Platform Profiles registered\n");
> + }
> +}
> +
> +static void amd_pmf_deinit_features(struct amd_pmf_dev *dev)
> +{
> + if (is_apmf_func_supported(APMF_FUNC_STATIC_SLIDER_GRANULAR))
> + amd_pmf_deinit_sps(dev);
> +}
> +
> static const struct acpi_device_id amd_pmf_acpi_ids[] = {
> {"AMDI0102", 0},
> { }
> @@ -206,6 +230,7 @@ static int amd_pmf_probe(struct platform_device *pdev)
>
> apmf_acpi_init(dev);
> platform_set_drvdata(pdev, dev);
> + amd_pmf_init_features(dev);
>
> mutex_init(&dev->lock);
> dev_info(dev->dev, "registered PMF device successfully\n");
> @@ -218,6 +243,7 @@ static int amd_pmf_remove(struct platform_device *pdev)
> struct amd_pmf_dev *dev = platform_get_drvdata(pdev);
>
> mutex_destroy(&dev->lock);
> + amd_pmf_deinit_features(dev);
> kfree(dev->buf);
> return 0;
> }
> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
> index 08c6bc0e2b42..30740a5cd30d 100644
> --- a/drivers/platform/x86/amd/pmf/pmf.h
> +++ b/drivers/platform/x86/amd/pmf/pmf.h
> @@ -12,10 +12,12 @@
> #define PMF_H
>
> #include <linux/acpi.h>
> +#include <linux/platform_profile.h>
>
> /* APMF Functions */
> #define APMF_FUNC_VERIFY_INTERFACE 0
> #define APMF_FUNC_GET_SYS_PARAMS 1
> +#define APMF_FUNC_STATIC_SLIDER_GRANULAR 9
>
> /* Message Definitions */
> #define SET_SPL 0x03 /* SPL: Sustained Power Limit */
> @@ -36,9 +38,12 @@
> #define GET_STT_LIMIT_APU 0x20
> #define GET_STT_LIMIT_HS2 0x21
>
> +#define ARG_NONE 0
> +
> /* AMD PMF BIOS interfaces */
> struct apmf_if_functions {
> bool system_params;
> + bool static_slider_granular;
> };
>
> struct apmf_if {
> @@ -60,6 +65,30 @@ struct apmf_system_params {
> u8 command_code;
> } __packed;
>
> +enum amd_stt_skin_temp {
> + STT_TEMP_APU,
> + STT_TEMP_HS2,
> + STT_TEMP_COUNT,
> +};
> +
> +enum amd_slider_op {
> + SLIDER_OP_GET,
> + SLIDER_OP_SET,
> +};
> +
> +enum power_source {
> + POWER_SOURCE_AC,
> + POWER_SOURCE_DC,
> + POWER_SOURCE_MAX,
> +};
> +
> +enum power_modes {
> + POWER_MODE_PERFORMANCE,
> + POWER_MODE_BALANCED_POWER,
> + POWER_MODE_POWER_SAVER,
> + POWER_MODE_MAX,
> +};
> +
> struct amd_pmf_dev {
> void __iomem *regbase;
> void __iomem *smu_virt_addr;
> @@ -68,11 +97,46 @@ struct amd_pmf_dev {
> u32 cpu_id;
> struct device *dev;
> struct apmf_if *apmf_if;
> + enum platform_profile_option current_profile;
> + struct platform_profile_handler pprof;
> struct mutex lock; /* protects the PMF interface */
> };
>
> +struct apmf_sps_prop_granular {
> + u32 fppt;
> + u32 sppt;
> + u32 sppt_apu_only;
> + u32 spl;
> + u32 stt_min;
> + u8 stt_skin_temp[STT_TEMP_COUNT];
> + u32 fan_id;
> +} __packed;
> +
> +/* Static Slider */
> +struct apmf_static_slider_granular_output {
> + u16 size;
> + struct apmf_sps_prop_granular prop[POWER_SOURCE_MAX * POWER_MODE_MAX];
> +} __packed;
> +
> +struct amd_pmf_static_slider_granular {
> + u16 size;
> + struct apmf_sps_prop_granular prop[POWER_SOURCE_MAX][POWER_MODE_MAX];
> +};
> +
> /* Core Layer */
> int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
> +int is_apmf_func_supported(unsigned long index);
> int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32 *data);
> +int amd_pmf_get_power_source(void);
> +
> +/* SPS Layer */
> +u8 amd_pmf_get_pprof_modes(struct amd_pmf_dev *pmf);
> +void amd_pmf_update_slider(struct amd_pmf_dev *dev, bool op, int idx,
> + struct amd_pmf_static_slider_granular *table);
> +int amd_pmf_init_sps(struct amd_pmf_dev *dev);
> +void amd_pmf_deinit_sps(struct amd_pmf_dev *dev);
> +int apmf_get_static_slider_granular(struct apmf_if *ampf_if,
> + struct apmf_static_slider_granular_output *output);
> +void amd_pmf_load_defaults_sps(struct amd_pmf_dev *dev);
>
> #endif /* PMF_H */
> diff --git a/drivers/platform/x86/amd/pmf/sps.c b/drivers/platform/x86/amd/pmf/sps.c
> new file mode 100644
> index 000000000000..2601adbfe14a
> --- /dev/null
> +++ b/drivers/platform/x86/amd/pmf/sps.c
> @@ -0,0 +1,152 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * AMD Platform Management Framework (PMF) Driver
> + *
> + * Copyright (c) 2022, Advanced Micro Devices, Inc.
> + * All Rights Reserved.
> + *
> + * Author: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> + */
> +
> +#include "pmf.h"
> +
> +static int amd_pmf_profile_get(struct platform_profile_handler *pprof,
> + enum platform_profile_option *profile);
> +static int amd_pmf_profile_set(struct platform_profile_handler *pprof,
> + enum platform_profile_option profile);
> +static struct amd_pmf_static_slider_granular config_store;
> +
> +void amd_pmf_load_defaults_sps(struct amd_pmf_dev *dev)
> +{
> + struct apmf_static_slider_granular_output output;
> + int i, j, idx = 0;
> +
> + memset(&config_store, 0, sizeof(config_store));
> +
> + if (dev->apmf_if->func.static_slider_granular) {
> + apmf_get_static_slider_granular(dev->apmf_if, &output);
> +
> + for (i = 0; i < POWER_SOURCE_MAX; i++) {
> + for (j = 0; j < POWER_MODE_MAX; j++) {
> + config_store.prop[i][j].spl = output.prop[idx].spl;
> + config_store.prop[i][j].sppt = output.prop[idx].sppt;
> + config_store.prop[i][j].sppt_apu_only =
> + output.prop[idx].sppt_apu_only;
> + config_store.prop[i][j].fppt = output.prop[idx].fppt;
> + config_store.prop[i][j].stt_min = output.prop[idx].stt_min;
> + config_store.prop[i][j].stt_skin_temp[STT_TEMP_APU] =
> + output.prop[idx].stt_skin_temp[STT_TEMP_APU];
> + config_store.prop[i][j].stt_skin_temp[STT_TEMP_HS2] =
> + output.prop[idx].stt_skin_temp[STT_TEMP_HS2];
> + config_store.prop[i][j].fan_id = output.prop[idx].fan_id;
> + idx++;
> + }
> + }
> + }
> +}
> +
> +void amd_pmf_update_slider(struct amd_pmf_dev *dev, bool op, int idx,
> + struct amd_pmf_static_slider_granular *table)
> +{
> + int src = amd_pmf_get_power_source();
> +
> + if (op == SLIDER_OP_SET) {
> + amd_pmf_send_cmd(dev, SET_SPL, false, config_store.prop[src][idx].spl, NULL);
> + amd_pmf_send_cmd(dev, SET_FPPT, false, config_store.prop[src][idx].fppt, NULL);
> + amd_pmf_send_cmd(dev, SET_SPPT, false, config_store.prop[src][idx].sppt, NULL);
> + amd_pmf_send_cmd(dev, SET_SPPT_APU_ONLY, false,
> + config_store.prop[src][idx].sppt_apu_only, NULL);
> + amd_pmf_send_cmd(dev, SET_STT_MIN_LIMIT, false,
> + config_store.prop[src][idx].stt_min, NULL);
> + amd_pmf_send_cmd(dev, SET_STT_LIMIT_APU, false,
> + config_store.prop[src][idx].stt_skin_temp[STT_TEMP_APU], NULL);
> + amd_pmf_send_cmd(dev, SET_STT_LIMIT_HS2, false,
> + config_store.prop[src][idx].stt_skin_temp[STT_TEMP_HS2], NULL);
> + } else if (op == SLIDER_OP_GET) {
> + amd_pmf_send_cmd(dev, GET_SPL, true, ARG_NONE, &table->prop[src][idx].spl);
> + amd_pmf_send_cmd(dev, GET_FPPT, true, ARG_NONE, &table->prop[src][idx].fppt);
> + amd_pmf_send_cmd(dev, GET_SPPT, true, ARG_NONE, &table->prop[src][idx].sppt);
> + amd_pmf_send_cmd(dev, GET_SPPT_APU_ONLY, true, ARG_NONE,
> + &table->prop[src][idx].sppt_apu_only);
> + amd_pmf_send_cmd(dev, GET_STT_MIN_LIMIT, true, ARG_NONE,
> + &table->prop[src][idx].stt_min);
> + amd_pmf_send_cmd(dev, GET_STT_LIMIT_APU, true, ARG_NONE,
> + (u32 *)&table->prop[src][idx].stt_skin_temp[STT_TEMP_APU]);
> + amd_pmf_send_cmd(dev, GET_STT_LIMIT_HS2, true, ARG_NONE,
> + (u32 *)&table->prop[src][idx].stt_skin_temp[STT_TEMP_HS2]);
> + }
> +}
> +
> +static int amd_pmf_profile_get(struct platform_profile_handler *pprof,
> + enum platform_profile_option *profile)
> +{
> + struct amd_pmf_dev *pmf = container_of(pprof, struct amd_pmf_dev, pprof);
> +
> + *profile = pmf->current_profile;
> + return 0;
> +}
> +
> +u8 amd_pmf_get_pprof_modes(struct amd_pmf_dev *pmf)
> +{
> + u8 mode;
> +
> + switch (pmf->current_profile) {
> + case PLATFORM_PROFILE_PERFORMANCE:
> + mode = POWER_MODE_PERFORMANCE;
> + break;
> + case PLATFORM_PROFILE_BALANCED:
> + mode = POWER_MODE_BALANCED_POWER;
> + break;
> + case PLATFORM_PROFILE_LOW_POWER:
> + mode = POWER_MODE_POWER_SAVER;
> + break;
> + default:
> + dev_err(pmf->dev, "Unknown Platform Profile.\n");
> + break;
> + }
> +
> + return mode;
> +}
> +
> +int amd_pmf_profile_set(struct platform_profile_handler *pprof,
> + enum platform_profile_option profile)
> +{
> + struct amd_pmf_dev *pmf = container_of(pprof, struct amd_pmf_dev, pprof);
> + u8 mode;
> +
> + pmf->current_profile = profile;
> + mode = amd_pmf_get_pprof_modes(pmf);
> + amd_pmf_update_slider(pmf, SLIDER_OP_SET, mode, NULL);
> + return 0;
> +}
> +
> +int amd_pmf_init_sps(struct amd_pmf_dev *dev)
> +{
> + int err = 0;
> +
> + dev->pprof.profile_get = amd_pmf_profile_get;
> + dev->pprof.profile_set = amd_pmf_profile_set;
> +
> + /* Setup supported modes */
> + set_bit(PLATFORM_PROFILE_LOW_POWER, dev->pprof.choices);
> + set_bit(PLATFORM_PROFILE_BALANCED, dev->pprof.choices);
> + set_bit(PLATFORM_PROFILE_PERFORMANCE, dev->pprof.choices);
> +
> + /* Create platform_profile structure and register */
> + err = platform_profile_register(&dev->pprof);
> + if (err) {
> + dev_err(dev->dev, "Failed to register SPS support, this is most likely an SBIOS bug: %d\n",
> + err);
> + return -EEXIST;
> + }
> +
> + dev->current_profile = PLATFORM_PROFILE_BALANCED;
> + amd_pmf_load_defaults_sps(dev);
> +
> + return err;
> +}
> +
> +void amd_pmf_deinit_sps(struct amd_pmf_dev *dev)
> +{
> + platform_profile_remove();
> +}
^ permalink raw reply [flat|nested] 33+ messages in thread
* [PATCH v2 RESEND 04/11] platform/x86/amd/pmf: Add debugfs information
2022-07-28 18:20 [PATCH v2 RESEND 00/11] platform/x86/amd/pmf: Introduce AMD PMF Driver Shyam Sundar S K
` (2 preceding siblings ...)
2022-07-28 18:20 ` [PATCH v2 RESEND 03/11] platform/x86/amd/pmf: Add support SPS PMF feature Shyam Sundar S K
@ 2022-07-28 18:20 ` Shyam Sundar S K
2022-08-01 11:35 ` Hans de Goede
2022-07-28 18:20 ` [PATCH v2 RESEND 05/11] platform/x86/amd/pmf: Add heartbeat signal support Shyam Sundar S K
` (6 subsequent siblings)
10 siblings, 1 reply; 33+ messages in thread
From: Shyam Sundar S K @ 2022-07-28 18:20 UTC (permalink / raw)
To: hdegoede, markgross; +Cc: platform-driver-x86, Patil.Reddy, Shyam Sundar S K
Add debugfs support to the PMF driver so that using this interface the
live counters from the PMFW can be queried to see if the power parameters
are getting set properly when a certain power mode change happens.
Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
---
drivers/platform/x86/amd/pmf/core.c | 36 +++++++++++++++++++++++++++++
drivers/platform/x86/amd/pmf/pmf.h | 1 +
2 files changed, 37 insertions(+)
diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
index b6006e8ee1a1..4ce69864879a 100644
--- a/drivers/platform/x86/amd/pmf/core.c
+++ b/drivers/platform/x86/amd/pmf/core.c
@@ -8,6 +8,7 @@
* Author: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
*/
+#include <linux/debugfs.h>
#include <linux/iopoll.h>
#include <linux/module.h>
#include <linux/pci.h>
@@ -46,6 +47,39 @@
#define DELAY_MIN_US 2000
#define DELAY_MAX_US 3000
+static int current_power_limits_show(struct seq_file *seq, void *unused)
+{
+ struct amd_pmf_dev *dev = seq->private;
+ struct amd_pmf_static_slider_granular table;
+ int mode, src = 0;
+
+ mode = amd_pmf_get_pprof_modes(dev);
+ src = amd_pmf_get_power_source();
+ amd_pmf_update_slider(dev, SLIDER_OP_GET, mode, &table);
+ seq_printf(seq, "spl:%u fppt:%u sppt:%u sppt_apu_only:%u stt_min:%u stt[APU]:%u stt[HS2]: %u\n",
+ table.prop[src][mode].spl,
+ table.prop[src][mode].fppt,
+ table.prop[src][mode].sppt,
+ table.prop[src][mode].sppt_apu_only,
+ table.prop[src][mode].stt_min,
+ table.prop[src][mode].stt_skin_temp[STT_TEMP_APU],
+ table.prop[src][mode].stt_skin_temp[STT_TEMP_HS2]);
+ return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(current_power_limits);
+
+static void amd_pmf_dbgfs_unregister(struct amd_pmf_dev *dev)
+{
+ debugfs_remove_recursive(dev->dbgfs_dir);
+}
+
+static void amd_pmf_dbgfs_register(struct amd_pmf_dev *dev)
+{
+ dev->dbgfs_dir = debugfs_create_dir("amd_pmf", NULL);
+ debugfs_create_file("current_power_limits", 0644, dev->dbgfs_dir, dev,
+ ¤t_power_limits_fops);
+}
+
int amd_pmf_get_power_source(void)
{
if (power_supply_is_system_supplied() > 0)
@@ -231,6 +265,7 @@ static int amd_pmf_probe(struct platform_device *pdev)
apmf_acpi_init(dev);
platform_set_drvdata(pdev, dev);
amd_pmf_init_features(dev);
+ amd_pmf_dbgfs_register(dev);
mutex_init(&dev->lock);
dev_info(dev->dev, "registered PMF device successfully\n");
@@ -244,6 +279,7 @@ static int amd_pmf_remove(struct platform_device *pdev)
mutex_destroy(&dev->lock);
amd_pmf_deinit_features(dev);
+ amd_pmf_dbgfs_unregister(dev);
kfree(dev->buf);
return 0;
}
diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
index 30740a5cd30d..de8dbd5e04e8 100644
--- a/drivers/platform/x86/amd/pmf/pmf.h
+++ b/drivers/platform/x86/amd/pmf/pmf.h
@@ -100,6 +100,7 @@ struct amd_pmf_dev {
enum platform_profile_option current_profile;
struct platform_profile_handler pprof;
struct mutex lock; /* protects the PMF interface */
+ struct dentry *dbgfs_dir;
};
struct apmf_sps_prop_granular {
--
2.25.1
^ permalink raw reply related [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 04/11] platform/x86/amd/pmf: Add debugfs information
2022-07-28 18:20 ` [PATCH v2 RESEND 04/11] platform/x86/amd/pmf: Add debugfs information Shyam Sundar S K
@ 2022-08-01 11:35 ` Hans de Goede
0 siblings, 0 replies; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 11:35 UTC (permalink / raw)
To: Shyam Sundar S K, markgross; +Cc: platform-driver-x86, Patil.Reddy
Hi,
On 7/28/22 20:20, Shyam Sundar S K wrote:
> Add debugfs support to the PMF driver so that using this interface the
> live counters from the PMFW can be queried to see if the power parameters
> are getting set properly when a certain power mode change happens.
>
> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
Thanks, version 2 looks good to me:
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Regards,
Hans
> ---
> drivers/platform/x86/amd/pmf/core.c | 36 +++++++++++++++++++++++++++++
> drivers/platform/x86/amd/pmf/pmf.h | 1 +
> 2 files changed, 37 insertions(+)
>
> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
> index b6006e8ee1a1..4ce69864879a 100644
> --- a/drivers/platform/x86/amd/pmf/core.c
> +++ b/drivers/platform/x86/amd/pmf/core.c
> @@ -8,6 +8,7 @@
> * Author: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> */
>
> +#include <linux/debugfs.h>
> #include <linux/iopoll.h>
> #include <linux/module.h>
> #include <linux/pci.h>
> @@ -46,6 +47,39 @@
> #define DELAY_MIN_US 2000
> #define DELAY_MAX_US 3000
>
> +static int current_power_limits_show(struct seq_file *seq, void *unused)
> +{
> + struct amd_pmf_dev *dev = seq->private;
> + struct amd_pmf_static_slider_granular table;
> + int mode, src = 0;
> +
> + mode = amd_pmf_get_pprof_modes(dev);
> + src = amd_pmf_get_power_source();
> + amd_pmf_update_slider(dev, SLIDER_OP_GET, mode, &table);
> + seq_printf(seq, "spl:%u fppt:%u sppt:%u sppt_apu_only:%u stt_min:%u stt[APU]:%u stt[HS2]: %u\n",
> + table.prop[src][mode].spl,
> + table.prop[src][mode].fppt,
> + table.prop[src][mode].sppt,
> + table.prop[src][mode].sppt_apu_only,
> + table.prop[src][mode].stt_min,
> + table.prop[src][mode].stt_skin_temp[STT_TEMP_APU],
> + table.prop[src][mode].stt_skin_temp[STT_TEMP_HS2]);
> + return 0;
> +}
> +DEFINE_SHOW_ATTRIBUTE(current_power_limits);
> +
> +static void amd_pmf_dbgfs_unregister(struct amd_pmf_dev *dev)
> +{
> + debugfs_remove_recursive(dev->dbgfs_dir);
> +}
> +
> +static void amd_pmf_dbgfs_register(struct amd_pmf_dev *dev)
> +{
> + dev->dbgfs_dir = debugfs_create_dir("amd_pmf", NULL);
> + debugfs_create_file("current_power_limits", 0644, dev->dbgfs_dir, dev,
> + ¤t_power_limits_fops);
> +}
> +
> int amd_pmf_get_power_source(void)
> {
> if (power_supply_is_system_supplied() > 0)
> @@ -231,6 +265,7 @@ static int amd_pmf_probe(struct platform_device *pdev)
> apmf_acpi_init(dev);
> platform_set_drvdata(pdev, dev);
> amd_pmf_init_features(dev);
> + amd_pmf_dbgfs_register(dev);
>
> mutex_init(&dev->lock);
> dev_info(dev->dev, "registered PMF device successfully\n");
> @@ -244,6 +279,7 @@ static int amd_pmf_remove(struct platform_device *pdev)
>
> mutex_destroy(&dev->lock);
> amd_pmf_deinit_features(dev);
> + amd_pmf_dbgfs_unregister(dev);
> kfree(dev->buf);
> return 0;
> }
> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
> index 30740a5cd30d..de8dbd5e04e8 100644
> --- a/drivers/platform/x86/amd/pmf/pmf.h
> +++ b/drivers/platform/x86/amd/pmf/pmf.h
> @@ -100,6 +100,7 @@ struct amd_pmf_dev {
> enum platform_profile_option current_profile;
> struct platform_profile_handler pprof;
> struct mutex lock; /* protects the PMF interface */
> + struct dentry *dbgfs_dir;
> };
>
> struct apmf_sps_prop_granular {
^ permalink raw reply [flat|nested] 33+ messages in thread
* [PATCH v2 RESEND 05/11] platform/x86/amd/pmf: Add heartbeat signal support
2022-07-28 18:20 [PATCH v2 RESEND 00/11] platform/x86/amd/pmf: Introduce AMD PMF Driver Shyam Sundar S K
` (3 preceding siblings ...)
2022-07-28 18:20 ` [PATCH v2 RESEND 04/11] platform/x86/amd/pmf: Add debugfs information Shyam Sundar S K
@ 2022-07-28 18:20 ` Shyam Sundar S K
2022-08-01 11:38 ` Hans de Goede
2022-07-28 18:20 ` [PATCH v2 RESEND 06/11] platform/x86/amd/pmf: Add fan control support Shyam Sundar S K
` (5 subsequent siblings)
10 siblings, 1 reply; 33+ messages in thread
From: Shyam Sundar S K @ 2022-07-28 18:20 UTC (permalink / raw)
To: hdegoede, markgross; +Cc: platform-driver-x86, Patil.Reddy, Shyam Sundar S K
PMF driver can send periodic heartbeat signals to OEM BIOS. When BIOS does
not receive the signal after a period of time, it can infer that AMDPMF
has hung or failed to load.
In this situation, BIOS can fallback to legacy operations. OEM can modify
the time interval of the signal or completely disable signals through
ACPI Method.
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
---
drivers/platform/x86/amd/pmf/acpi.c | 42 +++++++++++++++++++++++++++--
drivers/platform/x86/amd/pmf/core.c | 1 +
drivers/platform/x86/amd/pmf/pmf.h | 10 +++++++
3 files changed, 51 insertions(+), 2 deletions(-)
diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
index 41abd8680d8d..58e4893edea2 100644
--- a/drivers/platform/x86/amd/pmf/acpi.c
+++ b/drivers/platform/x86/amd/pmf/acpi.c
@@ -114,6 +114,26 @@ int apmf_get_static_slider_granular(struct apmf_if *apmf_if,
data, sizeof(*data));
}
+static void apmf_sbios_heartbeat_notify(struct work_struct *work)
+{
+ struct amd_pmf_dev *dev = container_of(work, struct amd_pmf_dev, heart_beat.work);
+ struct apmf_notification_cfg *n = &dev->apmf_if->notify_cfg;
+ union acpi_object *info;
+ int err = 0;
+
+ dev_dbg(dev->dev, "Sending heartbeat to SBIOS\n");
+ info = apmf_if_call(dev->apmf_if, APMF_FUNC_SBIOS_HEARTBEAT, NULL);
+ if (!info) {
+ err = -EIO;
+ goto out;
+ }
+
+ schedule_delayed_work(&dev->heart_beat, msecs_to_jiffies(n->hb_interval * 1000));
+
+out:
+ kfree(info);
+}
+
static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
{
struct apmf_verify_interface output;
@@ -134,6 +154,7 @@ static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *ap
static int apmf_get_system_params(struct apmf_if *apmf_if)
{
+ struct apmf_notification_cfg *n = &apmf_if->notify_cfg;
struct apmf_system_params params;
int err;
@@ -144,17 +165,26 @@ static int apmf_get_system_params(struct apmf_if *apmf_if)
return err;
}
- pr_debug("PMF: system params mask:0x%x flags:0x%x cmd_code:0x%x\n",
+ pr_debug("PMF: system params mask:0x%x flags:0x%x cmd_code:0x%x heartbeat:%d\n",
params.valid_mask,
params.flags,
- params.command_code);
+ params.command_code,
+ params.heartbeat_int);
params.flags = params.flags & params.valid_mask;
+ n->hb_interval = params.heartbeat_int;
return 0;
}
+void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev)
+{
+ if (pmf_dev->apmf_if->func.sbios_heartbeat)
+ cancel_delayed_work_sync(&pmf_dev->heart_beat);
+}
+
int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
{
+ struct apmf_notification_cfg *n;
acpi_handle apmf_if_handle;
struct apmf_if *apmf_if;
acpi_status status;
@@ -177,6 +207,7 @@ int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
goto out;
}
pmf_dev->apmf_if = apmf_if;
+ n = &apmf_if->notify_cfg;
ret = apmf_get_system_params(apmf_if);
if (ret) {
@@ -185,6 +216,13 @@ int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
goto out;
}
+ if (n->hb_interval) {
+ apmf_if->func.sbios_heartbeat = true;
+ /* send heartbeats only if the interval is not zero */
+ INIT_DELAYED_WORK(&pmf_dev->heart_beat, apmf_sbios_heartbeat_notify);
+ schedule_delayed_work(&pmf_dev->heart_beat, 0);
+ }
+
out:
return ret;
}
diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
index 4ce69864879a..6c1c5a89fe71 100644
--- a/drivers/platform/x86/amd/pmf/core.c
+++ b/drivers/platform/x86/amd/pmf/core.c
@@ -279,6 +279,7 @@ static int amd_pmf_remove(struct platform_device *pdev)
mutex_destroy(&dev->lock);
amd_pmf_deinit_features(dev);
+ apmf_acpi_deinit(dev);
amd_pmf_dbgfs_unregister(dev);
kfree(dev->buf);
return 0;
diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
index de8dbd5e04e8..f546062a10a7 100644
--- a/drivers/platform/x86/amd/pmf/pmf.h
+++ b/drivers/platform/x86/amd/pmf/pmf.h
@@ -17,6 +17,7 @@
/* APMF Functions */
#define APMF_FUNC_VERIFY_INTERFACE 0
#define APMF_FUNC_GET_SYS_PARAMS 1
+#define APMF_FUNC_SBIOS_HEARTBEAT 4
#define APMF_FUNC_STATIC_SLIDER_GRANULAR 9
/* Message Definitions */
@@ -43,12 +44,18 @@
/* AMD PMF BIOS interfaces */
struct apmf_if_functions {
bool system_params;
+ bool sbios_heartbeat;
bool static_slider_granular;
};
+struct apmf_notification_cfg {
+ int hb_interval; /* SBIOS heartbeat interval */
+};
+
struct apmf_if {
acpi_handle handle;
struct apmf_if_functions func;
+ struct apmf_notification_cfg notify_cfg;
};
struct apmf_verify_interface {
@@ -63,6 +70,7 @@ struct apmf_system_params {
u32 valid_mask;
u32 flags;
u8 command_code;
+ u32 heartbeat_int;
} __packed;
enum amd_stt_skin_temp {
@@ -99,6 +107,7 @@ struct amd_pmf_dev {
struct apmf_if *apmf_if;
enum platform_profile_option current_profile;
struct platform_profile_handler pprof;
+ struct delayed_work heart_beat;
struct mutex lock; /* protects the PMF interface */
struct dentry *dbgfs_dir;
};
@@ -126,6 +135,7 @@ struct amd_pmf_static_slider_granular {
/* Core Layer */
int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
+void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev);
int is_apmf_func_supported(unsigned long index);
int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32 *data);
int amd_pmf_get_power_source(void);
--
2.25.1
^ permalink raw reply related [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 05/11] platform/x86/amd/pmf: Add heartbeat signal support
2022-07-28 18:20 ` [PATCH v2 RESEND 05/11] platform/x86/amd/pmf: Add heartbeat signal support Shyam Sundar S K
@ 2022-08-01 11:38 ` Hans de Goede
0 siblings, 0 replies; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 11:38 UTC (permalink / raw)
To: Shyam Sundar S K, markgross; +Cc: platform-driver-x86, Patil.Reddy
Hi,
On 7/28/22 20:20, Shyam Sundar S K wrote:
> PMF driver can send periodic heartbeat signals to OEM BIOS. When BIOS does
> not receive the signal after a period of time, it can infer that AMDPMF
> has hung or failed to load.
>
> In this situation, BIOS can fallback to legacy operations. OEM can modify
> the time interval of the signal or completely disable signals through
> ACPI Method.
>
> Reviewed-by: Hans de Goede <hdegoede@redhat.com>
> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
Still looks good to me.
Regards,
Hans
> ---
> drivers/platform/x86/amd/pmf/acpi.c | 42 +++++++++++++++++++++++++++--
> drivers/platform/x86/amd/pmf/core.c | 1 +
> drivers/platform/x86/amd/pmf/pmf.h | 10 +++++++
> 3 files changed, 51 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
> index 41abd8680d8d..58e4893edea2 100644
> --- a/drivers/platform/x86/amd/pmf/acpi.c
> +++ b/drivers/platform/x86/amd/pmf/acpi.c
> @@ -114,6 +114,26 @@ int apmf_get_static_slider_granular(struct apmf_if *apmf_if,
> data, sizeof(*data));
> }
>
> +static void apmf_sbios_heartbeat_notify(struct work_struct *work)
> +{
> + struct amd_pmf_dev *dev = container_of(work, struct amd_pmf_dev, heart_beat.work);
> + struct apmf_notification_cfg *n = &dev->apmf_if->notify_cfg;
> + union acpi_object *info;
> + int err = 0;
> +
> + dev_dbg(dev->dev, "Sending heartbeat to SBIOS\n");
> + info = apmf_if_call(dev->apmf_if, APMF_FUNC_SBIOS_HEARTBEAT, NULL);
> + if (!info) {
> + err = -EIO;
> + goto out;
> + }
> +
> + schedule_delayed_work(&dev->heart_beat, msecs_to_jiffies(n->hb_interval * 1000));
> +
> +out:
> + kfree(info);
> +}
> +
> static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
> {
> struct apmf_verify_interface output;
> @@ -134,6 +154,7 @@ static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *ap
>
> static int apmf_get_system_params(struct apmf_if *apmf_if)
> {
> + struct apmf_notification_cfg *n = &apmf_if->notify_cfg;
> struct apmf_system_params params;
> int err;
>
> @@ -144,17 +165,26 @@ static int apmf_get_system_params(struct apmf_if *apmf_if)
> return err;
> }
>
> - pr_debug("PMF: system params mask:0x%x flags:0x%x cmd_code:0x%x\n",
> + pr_debug("PMF: system params mask:0x%x flags:0x%x cmd_code:0x%x heartbeat:%d\n",
> params.valid_mask,
> params.flags,
> - params.command_code);
> + params.command_code,
> + params.heartbeat_int);
> params.flags = params.flags & params.valid_mask;
> + n->hb_interval = params.heartbeat_int;
>
> return 0;
> }
>
> +void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev)
> +{
> + if (pmf_dev->apmf_if->func.sbios_heartbeat)
> + cancel_delayed_work_sync(&pmf_dev->heart_beat);
> +}
> +
> int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
> {
> + struct apmf_notification_cfg *n;
> acpi_handle apmf_if_handle;
> struct apmf_if *apmf_if;
> acpi_status status;
> @@ -177,6 +207,7 @@ int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
> goto out;
> }
> pmf_dev->apmf_if = apmf_if;
> + n = &apmf_if->notify_cfg;
>
> ret = apmf_get_system_params(apmf_if);
> if (ret) {
> @@ -185,6 +216,13 @@ int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
> goto out;
> }
>
> + if (n->hb_interval) {
> + apmf_if->func.sbios_heartbeat = true;
> + /* send heartbeats only if the interval is not zero */
> + INIT_DELAYED_WORK(&pmf_dev->heart_beat, apmf_sbios_heartbeat_notify);
> + schedule_delayed_work(&pmf_dev->heart_beat, 0);
> + }
> +
> out:
> return ret;
> }
> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
> index 4ce69864879a..6c1c5a89fe71 100644
> --- a/drivers/platform/x86/amd/pmf/core.c
> +++ b/drivers/platform/x86/amd/pmf/core.c
> @@ -279,6 +279,7 @@ static int amd_pmf_remove(struct platform_device *pdev)
>
> mutex_destroy(&dev->lock);
> amd_pmf_deinit_features(dev);
> + apmf_acpi_deinit(dev);
> amd_pmf_dbgfs_unregister(dev);
> kfree(dev->buf);
> return 0;
> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
> index de8dbd5e04e8..f546062a10a7 100644
> --- a/drivers/platform/x86/amd/pmf/pmf.h
> +++ b/drivers/platform/x86/amd/pmf/pmf.h
> @@ -17,6 +17,7 @@
> /* APMF Functions */
> #define APMF_FUNC_VERIFY_INTERFACE 0
> #define APMF_FUNC_GET_SYS_PARAMS 1
> +#define APMF_FUNC_SBIOS_HEARTBEAT 4
> #define APMF_FUNC_STATIC_SLIDER_GRANULAR 9
>
> /* Message Definitions */
> @@ -43,12 +44,18 @@
> /* AMD PMF BIOS interfaces */
> struct apmf_if_functions {
> bool system_params;
> + bool sbios_heartbeat;
> bool static_slider_granular;
> };
>
> +struct apmf_notification_cfg {
> + int hb_interval; /* SBIOS heartbeat interval */
> +};
> +
> struct apmf_if {
> acpi_handle handle;
> struct apmf_if_functions func;
> + struct apmf_notification_cfg notify_cfg;
> };
>
> struct apmf_verify_interface {
> @@ -63,6 +70,7 @@ struct apmf_system_params {
> u32 valid_mask;
> u32 flags;
> u8 command_code;
> + u32 heartbeat_int;
> } __packed;
>
> enum amd_stt_skin_temp {
> @@ -99,6 +107,7 @@ struct amd_pmf_dev {
> struct apmf_if *apmf_if;
> enum platform_profile_option current_profile;
> struct platform_profile_handler pprof;
> + struct delayed_work heart_beat;
> struct mutex lock; /* protects the PMF interface */
> struct dentry *dbgfs_dir;
> };
> @@ -126,6 +135,7 @@ struct amd_pmf_static_slider_granular {
>
> /* Core Layer */
> int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
> +void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev);
> int is_apmf_func_supported(unsigned long index);
> int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32 *data);
> int amd_pmf_get_power_source(void);
^ permalink raw reply [flat|nested] 33+ messages in thread
* [PATCH v2 RESEND 06/11] platform/x86/amd/pmf: Add fan control support
2022-07-28 18:20 [PATCH v2 RESEND 00/11] platform/x86/amd/pmf: Introduce AMD PMF Driver Shyam Sundar S K
` (4 preceding siblings ...)
2022-07-28 18:20 ` [PATCH v2 RESEND 05/11] platform/x86/amd/pmf: Add heartbeat signal support Shyam Sundar S K
@ 2022-07-28 18:20 ` Shyam Sundar S K
2022-08-01 11:40 ` Hans de Goede
2022-08-01 12:51 ` Hans de Goede
2022-07-28 18:20 ` [PATCH v2 RESEND 07/11] platform/x86/amd/pmf: Get performance metrics from PMFW Shyam Sundar S K
` (4 subsequent siblings)
10 siblings, 2 replies; 33+ messages in thread
From: Shyam Sundar S K @ 2022-07-28 18:20 UTC (permalink / raw)
To: hdegoede, markgross; +Cc: platform-driver-x86, Patil.Reddy, Shyam Sundar S K
PMF has a generic interface defined via PMF ACPI fn9 for influencing fan
policy during mode switch.
PMF ACPI fn9 will normally be invoked when AMDPMF needs to change the fan
table index for the EC. When AMDPMF is loaded this function will be invoked
to change fan speed. OEM can also choose to report the actual fan table
index and fan RPM to PMF through OEM structure.
This information will be communicated by PMF driver to customer BIOS.
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
---
drivers/platform/x86/amd/pmf/acpi.c | 26 ++++++++++++++++++++++++++
drivers/platform/x86/amd/pmf/pmf.h | 17 +++++++++++++++++
2 files changed, 43 insertions(+)
diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
index 58e4893edea2..1814d48f8e44 100644
--- a/drivers/platform/x86/amd/pmf/acpi.c
+++ b/drivers/platform/x86/amd/pmf/acpi.c
@@ -99,6 +99,7 @@ static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
{
func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
func->static_slider_granular = mask & APMF_FUNC_STATIC_SLIDER_GRANULAR;
+ func->fan_table_idx = mask & APMF_FUNC_SET_FAN_IDX;
}
int is_apmf_func_supported(unsigned long index)
@@ -134,6 +135,31 @@ static void apmf_sbios_heartbeat_notify(struct work_struct *work)
kfree(info);
}
+int apmf_update_fan_idx(struct apmf_if *ampf_if, bool manual, u32 idx)
+{
+ union acpi_object *info;
+ struct apmf_fan_idx args;
+ struct acpi_buffer params;
+ int err = 0;
+
+ args.size = sizeof(args);
+ args.fan_ctl_mode = manual;
+ args.fan_ctl_idx = idx;
+
+ params.length = sizeof(args);
+ params.pointer = (void *)&args;
+
+ info = apmf_if_call(ampf_if, APMF_FUNC_SET_FAN_IDX, ¶ms);
+ if (!info) {
+ err = -EIO;
+ goto out;
+ }
+
+out:
+ kfree(info);
+ return err;
+}
+
static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
{
struct apmf_verify_interface output;
diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
index f546062a10a7..b6501a68aa4e 100644
--- a/drivers/platform/x86/amd/pmf/pmf.h
+++ b/drivers/platform/x86/amd/pmf/pmf.h
@@ -18,6 +18,7 @@
#define APMF_FUNC_VERIFY_INTERFACE 0
#define APMF_FUNC_GET_SYS_PARAMS 1
#define APMF_FUNC_SBIOS_HEARTBEAT 4
+#define APMF_FUNC_SET_FAN_IDX 7
#define APMF_FUNC_STATIC_SLIDER_GRANULAR 9
/* Message Definitions */
@@ -39,12 +40,16 @@
#define GET_STT_LIMIT_APU 0x20
#define GET_STT_LIMIT_HS2 0x21
+/* Fan Index for Auto Mode */
+#define FAN_INDEX_AUTO 0xFFFFFFFF
+
#define ARG_NONE 0
/* AMD PMF BIOS interfaces */
struct apmf_if_functions {
bool system_params;
bool sbios_heartbeat;
+ bool fan_table_idx;
bool static_slider_granular;
};
@@ -73,6 +78,12 @@ struct apmf_system_params {
u32 heartbeat_int;
} __packed;
+struct apmf_fan_idx {
+ u16 size;
+ u8 fan_ctl_mode;
+ u32 fan_ctl_idx;
+} __packed;
+
enum amd_stt_skin_temp {
STT_TEMP_APU,
STT_TEMP_HS2,
@@ -133,6 +144,11 @@ struct amd_pmf_static_slider_granular {
struct apmf_sps_prop_granular prop[POWER_SOURCE_MAX][POWER_MODE_MAX];
};
+struct fan_table_control {
+ bool manual;
+ unsigned long fan_id;
+};
+
/* Core Layer */
int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev);
@@ -150,4 +166,5 @@ int apmf_get_static_slider_granular(struct apmf_if *ampf_if,
struct apmf_static_slider_granular_output *output);
void amd_pmf_load_defaults_sps(struct amd_pmf_dev *dev);
+int apmf_update_fan_idx(struct apmf_if *ampf_if, bool manual, u32 idx);
#endif /* PMF_H */
--
2.25.1
^ permalink raw reply related [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 06/11] platform/x86/amd/pmf: Add fan control support
2022-07-28 18:20 ` [PATCH v2 RESEND 06/11] platform/x86/amd/pmf: Add fan control support Shyam Sundar S K
@ 2022-08-01 11:40 ` Hans de Goede
2022-08-01 12:51 ` Hans de Goede
1 sibling, 0 replies; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 11:40 UTC (permalink / raw)
To: Shyam Sundar S K, markgross; +Cc: platform-driver-x86, Patil.Reddy
Hi,
On 7/28/22 20:20, Shyam Sundar S K wrote:
> PMF has a generic interface defined via PMF ACPI fn9 for influencing fan
> policy during mode switch.
>
> PMF ACPI fn9 will normally be invoked when AMDPMF needs to change the fan
> table index for the EC. When AMDPMF is loaded this function will be invoked
> to change fan speed. OEM can also choose to report the actual fan table
> index and fan RPM to PMF through OEM structure.
>
> This information will be communicated by PMF driver to customer BIOS.
>
> Reviewed-by: Hans de Goede <hdegoede@redhat.com>
> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
Still looks good to me.
Regards,
Hans
> ---
> drivers/platform/x86/amd/pmf/acpi.c | 26 ++++++++++++++++++++++++++
> drivers/platform/x86/amd/pmf/pmf.h | 17 +++++++++++++++++
> 2 files changed, 43 insertions(+)
>
> diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
> index 58e4893edea2..1814d48f8e44 100644
> --- a/drivers/platform/x86/amd/pmf/acpi.c
> +++ b/drivers/platform/x86/amd/pmf/acpi.c
> @@ -99,6 +99,7 @@ static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
> {
> func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
> func->static_slider_granular = mask & APMF_FUNC_STATIC_SLIDER_GRANULAR;
> + func->fan_table_idx = mask & APMF_FUNC_SET_FAN_IDX;
> }
>
> int is_apmf_func_supported(unsigned long index)
> @@ -134,6 +135,31 @@ static void apmf_sbios_heartbeat_notify(struct work_struct *work)
> kfree(info);
> }
>
> +int apmf_update_fan_idx(struct apmf_if *ampf_if, bool manual, u32 idx)
> +{
> + union acpi_object *info;
> + struct apmf_fan_idx args;
> + struct acpi_buffer params;
> + int err = 0;
> +
> + args.size = sizeof(args);
> + args.fan_ctl_mode = manual;
> + args.fan_ctl_idx = idx;
> +
> + params.length = sizeof(args);
> + params.pointer = (void *)&args;
> +
> + info = apmf_if_call(ampf_if, APMF_FUNC_SET_FAN_IDX, ¶ms);
> + if (!info) {
> + err = -EIO;
> + goto out;
> + }
> +
> +out:
> + kfree(info);
> + return err;
> +}
> +
> static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
> {
> struct apmf_verify_interface output;
> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
> index f546062a10a7..b6501a68aa4e 100644
> --- a/drivers/platform/x86/amd/pmf/pmf.h
> +++ b/drivers/platform/x86/amd/pmf/pmf.h
> @@ -18,6 +18,7 @@
> #define APMF_FUNC_VERIFY_INTERFACE 0
> #define APMF_FUNC_GET_SYS_PARAMS 1
> #define APMF_FUNC_SBIOS_HEARTBEAT 4
> +#define APMF_FUNC_SET_FAN_IDX 7
> #define APMF_FUNC_STATIC_SLIDER_GRANULAR 9
>
> /* Message Definitions */
> @@ -39,12 +40,16 @@
> #define GET_STT_LIMIT_APU 0x20
> #define GET_STT_LIMIT_HS2 0x21
>
> +/* Fan Index for Auto Mode */
> +#define FAN_INDEX_AUTO 0xFFFFFFFF
> +
> #define ARG_NONE 0
>
> /* AMD PMF BIOS interfaces */
> struct apmf_if_functions {
> bool system_params;
> bool sbios_heartbeat;
> + bool fan_table_idx;
> bool static_slider_granular;
> };
>
> @@ -73,6 +78,12 @@ struct apmf_system_params {
> u32 heartbeat_int;
> } __packed;
>
> +struct apmf_fan_idx {
> + u16 size;
> + u8 fan_ctl_mode;
> + u32 fan_ctl_idx;
> +} __packed;
> +
> enum amd_stt_skin_temp {
> STT_TEMP_APU,
> STT_TEMP_HS2,
> @@ -133,6 +144,11 @@ struct amd_pmf_static_slider_granular {
> struct apmf_sps_prop_granular prop[POWER_SOURCE_MAX][POWER_MODE_MAX];
> };
>
> +struct fan_table_control {
> + bool manual;
> + unsigned long fan_id;
> +};
> +
> /* Core Layer */
> int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
> void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev);
> @@ -150,4 +166,5 @@ int apmf_get_static_slider_granular(struct apmf_if *ampf_if,
> struct apmf_static_slider_granular_output *output);
> void amd_pmf_load_defaults_sps(struct amd_pmf_dev *dev);
>
> +int apmf_update_fan_idx(struct apmf_if *ampf_if, bool manual, u32 idx);
> #endif /* PMF_H */
^ permalink raw reply [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 06/11] platform/x86/amd/pmf: Add fan control support
2022-07-28 18:20 ` [PATCH v2 RESEND 06/11] platform/x86/amd/pmf: Add fan control support Shyam Sundar S K
2022-08-01 11:40 ` Hans de Goede
@ 2022-08-01 12:51 ` Hans de Goede
1 sibling, 0 replies; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 12:51 UTC (permalink / raw)
To: Shyam Sundar S K, markgross; +Cc: platform-driver-x86, Patil.Reddy
Hi,
On 7/28/22 20:20, Shyam Sundar S K wrote:
> PMF has a generic interface defined via PMF ACPI fn9 for influencing fan
> policy during mode switch.
>
> PMF ACPI fn9 will normally be invoked when AMDPMF needs to change the fan
> table index for the EC. When AMDPMF is loaded this function will be invoked
> to change fan speed. OEM can also choose to report the actual fan table
> index and fan RPM to PMF through OEM structure.
>
> This information will be communicated by PMF driver to customer BIOS.
>
> Reviewed-by: Hans de Goede <hdegoede@redhat.com>
> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> ---
> drivers/platform/x86/amd/pmf/acpi.c | 26 ++++++++++++++++++++++++++
> drivers/platform/x86/amd/pmf/pmf.h | 17 +++++++++++++++++
> 2 files changed, 43 insertions(+)
>
> diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
> index 58e4893edea2..1814d48f8e44 100644
> --- a/drivers/platform/x86/amd/pmf/acpi.c
> +++ b/drivers/platform/x86/amd/pmf/acpi.c
> @@ -99,6 +99,7 @@ static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
> {
> func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
> func->static_slider_granular = mask & APMF_FUNC_STATIC_SLIDER_GRANULAR;
> + func->fan_table_idx = mask & APMF_FUNC_SET_FAN_IDX;
I just noticed that this is not correct, this should be:
func->system_params = mask & BIT(APMF_FUNC_... - 1);
Which does something completely different!
Also this information is duplicated by:
is_apmf_func_supported(APMF_FUNC_...) so please drop the apmf_if_functions
struct completely and use is_apmf_func_supported() everywhere.
Regards,
Hans
> }
>
> int is_apmf_func_supported(unsigned long index)
> @@ -134,6 +135,31 @@ static void apmf_sbios_heartbeat_notify(struct work_struct *work)
> kfree(info);
> }
>
> +int apmf_update_fan_idx(struct apmf_if *ampf_if, bool manual, u32 idx)
> +{
> + union acpi_object *info;
> + struct apmf_fan_idx args;
> + struct acpi_buffer params;
> + int err = 0;
> +
> + args.size = sizeof(args);
> + args.fan_ctl_mode = manual;
> + args.fan_ctl_idx = idx;
> +
> + params.length = sizeof(args);
> + params.pointer = (void *)&args;
> +
> + info = apmf_if_call(ampf_if, APMF_FUNC_SET_FAN_IDX, ¶ms);
> + if (!info) {
> + err = -EIO;
> + goto out;
> + }
> +
> +out:
> + kfree(info);
> + return err;
> +}
> +
> static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
> {
> struct apmf_verify_interface output;
> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
> index f546062a10a7..b6501a68aa4e 100644
> --- a/drivers/platform/x86/amd/pmf/pmf.h
> +++ b/drivers/platform/x86/amd/pmf/pmf.h
> @@ -18,6 +18,7 @@
> #define APMF_FUNC_VERIFY_INTERFACE 0
> #define APMF_FUNC_GET_SYS_PARAMS 1
> #define APMF_FUNC_SBIOS_HEARTBEAT 4
> +#define APMF_FUNC_SET_FAN_IDX 7
> #define APMF_FUNC_STATIC_SLIDER_GRANULAR 9
>
> /* Message Definitions */
> @@ -39,12 +40,16 @@
> #define GET_STT_LIMIT_APU 0x20
> #define GET_STT_LIMIT_HS2 0x21
>
> +/* Fan Index for Auto Mode */
> +#define FAN_INDEX_AUTO 0xFFFFFFFF
> +
> #define ARG_NONE 0
>
> /* AMD PMF BIOS interfaces */
> struct apmf_if_functions {
> bool system_params;
> bool sbios_heartbeat;
> + bool fan_table_idx;
> bool static_slider_granular;
> };
>
> @@ -73,6 +78,12 @@ struct apmf_system_params {
> u32 heartbeat_int;
> } __packed;
>
> +struct apmf_fan_idx {
> + u16 size;
> + u8 fan_ctl_mode;
> + u32 fan_ctl_idx;
> +} __packed;
> +
> enum amd_stt_skin_temp {
> STT_TEMP_APU,
> STT_TEMP_HS2,
> @@ -133,6 +144,11 @@ struct amd_pmf_static_slider_granular {
> struct apmf_sps_prop_granular prop[POWER_SOURCE_MAX][POWER_MODE_MAX];
> };
>
> +struct fan_table_control {
> + bool manual;
> + unsigned long fan_id;
> +};
> +
> /* Core Layer */
> int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
> void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev);
> @@ -150,4 +166,5 @@ int apmf_get_static_slider_granular(struct apmf_if *ampf_if,
> struct apmf_static_slider_granular_output *output);
> void amd_pmf_load_defaults_sps(struct amd_pmf_dev *dev);
>
> +int apmf_update_fan_idx(struct apmf_if *ampf_if, bool manual, u32 idx);
> #endif /* PMF_H */
^ permalink raw reply [flat|nested] 33+ messages in thread
* [PATCH v2 RESEND 07/11] platform/x86/amd/pmf: Get performance metrics from PMFW
2022-07-28 18:20 [PATCH v2 RESEND 00/11] platform/x86/amd/pmf: Introduce AMD PMF Driver Shyam Sundar S K
` (5 preceding siblings ...)
2022-07-28 18:20 ` [PATCH v2 RESEND 06/11] platform/x86/amd/pmf: Add fan control support Shyam Sundar S K
@ 2022-07-28 18:20 ` Shyam Sundar S K
2022-07-28 21:02 ` Limonciello, Mario
2022-08-01 11:50 ` Hans de Goede
2022-07-28 18:20 ` [PATCH v2 RESEND 08/11] platform/x86/amd/pmf: Add support for Auto mode feature Shyam Sundar S K
` (3 subsequent siblings)
10 siblings, 2 replies; 33+ messages in thread
From: Shyam Sundar S K @ 2022-07-28 18:20 UTC (permalink / raw)
To: hdegoede, markgross; +Cc: platform-driver-x86, Patil.Reddy, Shyam Sundar S K
PMF driver polls for metrics information from PMFW to understand the system
behavior, power consumption etc.
This metrics table information will be used the PMF features to tweak the
thermal heuristics. The poll duration can also be changed by the user
by changing the poll duration time.
Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
---
drivers/platform/x86/amd/pmf/core.c | 56 +++++++++++++++++++++++++++++
drivers/platform/x86/amd/pmf/pmf.h | 39 ++++++++++++++++++++
2 files changed, 95 insertions(+)
diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
index 6c1c5a89fe71..ede4eefc33a4 100644
--- a/drivers/platform/x86/amd/pmf/core.c
+++ b/drivers/platform/x86/amd/pmf/core.c
@@ -47,6 +47,11 @@
#define DELAY_MIN_US 2000
#define DELAY_MAX_US 3000
+/* override Metrics Table sample size time (in ms) */
+static int metrics_table_loop_ms = 1000;
+module_param(metrics_table_loop_ms, int, 0644);
+MODULE_PARM_DESC(metrics_table_loop_ms, " Metrics Table sample size time (default = 1000ms) ");
+
static int current_power_limits_show(struct seq_file *seq, void *unused)
{
struct amd_pmf_dev *dev = seq->private;
@@ -88,6 +93,29 @@ int amd_pmf_get_power_source(void)
return POWER_SOURCE_DC;
}
+static void amd_pmf_get_metrics(struct work_struct *work)
+{
+ struct amd_pmf_dev *dev = container_of(work, struct amd_pmf_dev, work_buffer.work);
+ enum platform_profile_option current_profile;
+ ktime_t time_elapsed_ms;
+ int socket_power;
+
+ /* Get the current profile information */
+ current_profile = READ_ONCE(dev->current_profile);
+
+ /* Transfer table contents */
+ memset(&dev->m_table, 0, sizeof(dev->m_table));
+ amd_pmf_send_cmd(dev, SET_TRANSFER_TABLE, 0, 7, NULL);
+ memcpy(&dev->m_table, dev->buf, sizeof(dev->m_table));
+
+ time_elapsed_ms = ktime_to_ms(ktime_get()) - dev->start_time;
+ /* Calculate the avg SoC power consumption */
+ socket_power = dev->m_table.apu_power + dev->m_table.dgpu_power;
+
+ dev->start_time = ktime_to_ms(ktime_get());
+ schedule_delayed_work(&dev->work_buffer, msecs_to_jiffies(metrics_table_loop_ms));
+}
+
static inline u32 amd_pmf_reg_read(struct amd_pmf_dev *dev, int reg_offset)
{
return ioread32(dev->regbase + reg_offset);
@@ -181,6 +209,34 @@ static const struct pci_device_id pmf_pci_ids[] = {
{ }
};
+int amd_pmf_init_metrics_table(struct amd_pmf_dev *dev)
+{
+ u64 phys_addr;
+ u32 hi, low;
+
+ INIT_DELAYED_WORK(&dev->work_buffer, amd_pmf_get_metrics);
+
+ /* Get Metrics Table Address */
+ dev->buf = kzalloc(sizeof(dev->m_table), GFP_KERNEL);
+ if (!dev->buf)
+ return -ENOMEM;
+
+ phys_addr = virt_to_phys(dev->buf);
+ hi = phys_addr >> 32;
+ low = phys_addr & GENMASK(31, 0);
+
+ amd_pmf_send_cmd(dev, SET_DRAM_ADDR_HIGH, 0, hi, NULL);
+ amd_pmf_send_cmd(dev, SET_DRAM_ADDR_LOW, 0, low, NULL);
+
+ /*
+ * Start collecting the metrics data after a small delay
+ * or else, we might end up getting stale values from PMFW.
+ */
+ schedule_delayed_work(&dev->work_buffer, msecs_to_jiffies(metrics_table_loop_ms * 3));
+
+ return 0;
+}
+
static void amd_pmf_init_features(struct amd_pmf_dev *dev)
{
/* Enable Static Slider */
diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
index b6501a68aa4e..49d3232ee2e0 100644
--- a/drivers/platform/x86/amd/pmf/pmf.h
+++ b/drivers/platform/x86/amd/pmf/pmf.h
@@ -84,6 +84,41 @@ struct apmf_fan_idx {
u32 fan_ctl_idx;
} __packed;
+struct smu_pmf_metrics {
+ u16 gfxclk_freq; /* in MHz */
+ u16 socclk_freq; /* in MHz */
+ u16 vclk_freq; /* in MHz */
+ u16 dclk_freq; /* in MHz */
+ u16 memclk_freq; /* in MHz */
+ u16 spare;
+ u16 gfx_activity; /* in Centi */
+ u16 uvd_activity; /* in Centi */
+ u16 voltage[2]; /* in mV */
+ u16 currents[2]; /* in mA */
+ u16 power[2];/* in mW */
+ u16 core_freq[8]; /* in MHz */
+ u16 core_power[8]; /* in mW */
+ u16 core_temp[8]; /* in centi-Celsius */
+ u16 l3_freq; /* in MHz */
+ u16 l3_temp; /* in centi-Celsius */
+ u16 gfx_temp; /* in centi-Celsius */
+ u16 soc_temp; /* in centi-Celsius */
+ u16 throttler_status;
+ u16 current_socketpower; /* in mW */
+ u16 stapm_orig_limit; /* in W */
+ u16 stapm_cur_limit; /* in W */
+ u32 apu_power; /* in mW */
+ u32 dgpu_power; /* in mW */
+ u16 vdd_tdc_val; /* in mA */
+ u16 soc_tdc_val; /* in mA */
+ u16 vdd_edc_val; /* in mA */
+ u16 soc_edcv_al; /* in mA */
+ u16 infra_cpu_maxfreq; /* in MHz */
+ u16 infra_gfx_maxfreq; /* in MHz */
+ u16 skin_temp; /* in centi-Celsius */
+ u16 device_state;
+};
+
enum amd_stt_skin_temp {
STT_TEMP_APU,
STT_TEMP_HS2,
@@ -121,6 +156,9 @@ struct amd_pmf_dev {
struct delayed_work heart_beat;
struct mutex lock; /* protects the PMF interface */
struct dentry *dbgfs_dir;
+ struct smu_pmf_metrics m_table;
+ struct delayed_work work_buffer;
+ ktime_t start_time;
};
struct apmf_sps_prop_granular {
@@ -154,6 +192,7 @@ int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev);
int is_apmf_func_supported(unsigned long index);
int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32 *data);
+int amd_pmf_init_metrics_table(struct amd_pmf_dev *dev);
int amd_pmf_get_power_source(void);
/* SPS Layer */
--
2.25.1
^ permalink raw reply related [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 07/11] platform/x86/amd/pmf: Get performance metrics from PMFW
2022-07-28 18:20 ` [PATCH v2 RESEND 07/11] platform/x86/amd/pmf: Get performance metrics from PMFW Shyam Sundar S K
@ 2022-07-28 21:02 ` Limonciello, Mario
2022-07-29 17:43 ` Shyam Sundar S K
2022-08-01 11:50 ` Hans de Goede
1 sibling, 1 reply; 33+ messages in thread
From: Limonciello, Mario @ 2022-07-28 21:02 UTC (permalink / raw)
To: Shyam Sundar S K, hdegoede, markgross; +Cc: platform-driver-x86, Patil.Reddy
On 7/28/2022 13:20, Shyam Sundar S K wrote:
> PMF driver polls for metrics information from PMFW to understand the system
> behavior, power consumption etc.
>
> This metrics table information will be used the PMF features to tweak the
> thermal heuristics. The poll duration can also be changed by the user
> by changing the poll duration time.
>
> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> ---
> drivers/platform/x86/amd/pmf/core.c | 56 +++++++++++++++++++++++++++++
> drivers/platform/x86/amd/pmf/pmf.h | 39 ++++++++++++++++++++
> 2 files changed, 95 insertions(+)
>
> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
> index 6c1c5a89fe71..ede4eefc33a4 100644
> --- a/drivers/platform/x86/amd/pmf/core.c
> +++ b/drivers/platform/x86/amd/pmf/core.c
> @@ -47,6 +47,11 @@
> #define DELAY_MIN_US 2000
> #define DELAY_MAX_US 3000
>
> +/* override Metrics Table sample size time (in ms) */
> +static int metrics_table_loop_ms = 1000;
> +module_param(metrics_table_loop_ms, int, 0644);
> +MODULE_PARM_DESC(metrics_table_loop_ms, " Metrics Table sample size time (default = 1000ms) ");
> +
> static int current_power_limits_show(struct seq_file *seq, void *unused)
> {
> struct amd_pmf_dev *dev = seq->private;
> @@ -88,6 +93,29 @@ int amd_pmf_get_power_source(void)
> return POWER_SOURCE_DC;
> }
>
> +static void amd_pmf_get_metrics(struct work_struct *work)
> +{
> + struct amd_pmf_dev *dev = container_of(work, struct amd_pmf_dev, work_buffer.work);
> + enum platform_profile_option current_profile;
> + ktime_t time_elapsed_ms;
> + int socket_power;
> +
> + /* Get the current profile information */
> + current_profile = READ_ONCE(dev->current_profile);
AFAICT this parameter isn't actually used/needed anymore.
> +
> + /* Transfer table contents */
> + memset(&dev->m_table, 0, sizeof(dev->m_table));
> + amd_pmf_send_cmd(dev, SET_TRANSFER_TABLE, 0, 7, NULL);
> + memcpy(&dev->m_table, dev->buf, sizeof(dev->m_table));
> +
> + time_elapsed_ms = ktime_to_ms(ktime_get()) - dev->start_time;
> + /* Calculate the avg SoC power consumption */
> + socket_power = dev->m_table.apu_power + dev->m_table.dgpu_power;
> +
> + dev->start_time = ktime_to_ms(ktime_get());
> + schedule_delayed_work(&dev->work_buffer, msecs_to_jiffies(metrics_table_loop_ms));
> +}
> +
> static inline u32 amd_pmf_reg_read(struct amd_pmf_dev *dev, int reg_offset)
> {
> return ioread32(dev->regbase + reg_offset);
> @@ -181,6 +209,34 @@ static const struct pci_device_id pmf_pci_ids[] = {
> { }
> };
>
> +int amd_pmf_init_metrics_table(struct amd_pmf_dev *dev)
> +{
> + u64 phys_addr;
> + u32 hi, low;
> +
> + INIT_DELAYED_WORK(&dev->work_buffer, amd_pmf_get_metrics);
> +
> + /* Get Metrics Table Address */
> + dev->buf = kzalloc(sizeof(dev->m_table), GFP_KERNEL);
> + if (!dev->buf)
> + return -ENOMEM;
> +
> + phys_addr = virt_to_phys(dev->buf);
> + hi = phys_addr >> 32;
> + low = phys_addr & GENMASK(31, 0);
> +
> + amd_pmf_send_cmd(dev, SET_DRAM_ADDR_HIGH, 0, hi, NULL);
> + amd_pmf_send_cmd(dev, SET_DRAM_ADDR_LOW, 0, low, NULL);
> +
> + /*
> + * Start collecting the metrics data after a small delay
> + * or else, we might end up getting stale values from PMFW.
> + */
> + schedule_delayed_work(&dev->work_buffer, msecs_to_jiffies(metrics_table_loop_ms * 3));
> +
> + return 0;
> +}
> +
> static void amd_pmf_init_features(struct amd_pmf_dev *dev)
> {
> /* Enable Static Slider */
> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
> index b6501a68aa4e..49d3232ee2e0 100644
> --- a/drivers/platform/x86/amd/pmf/pmf.h
> +++ b/drivers/platform/x86/amd/pmf/pmf.h
> @@ -84,6 +84,41 @@ struct apmf_fan_idx {
> u32 fan_ctl_idx;
> } __packed;
>
> +struct smu_pmf_metrics {
> + u16 gfxclk_freq; /* in MHz */
> + u16 socclk_freq; /* in MHz */
> + u16 vclk_freq; /* in MHz */
> + u16 dclk_freq; /* in MHz */
> + u16 memclk_freq; /* in MHz */
> + u16 spare;
> + u16 gfx_activity; /* in Centi */
> + u16 uvd_activity; /* in Centi */
> + u16 voltage[2]; /* in mV */
> + u16 currents[2]; /* in mA */
> + u16 power[2];/* in mW */
> + u16 core_freq[8]; /* in MHz */
> + u16 core_power[8]; /* in mW */
> + u16 core_temp[8]; /* in centi-Celsius */
> + u16 l3_freq; /* in MHz */
> + u16 l3_temp; /* in centi-Celsius */
> + u16 gfx_temp; /* in centi-Celsius */
> + u16 soc_temp; /* in centi-Celsius */
> + u16 throttler_status;
> + u16 current_socketpower; /* in mW */
> + u16 stapm_orig_limit; /* in W */
> + u16 stapm_cur_limit; /* in W */
> + u32 apu_power; /* in mW */
> + u32 dgpu_power; /* in mW */
> + u16 vdd_tdc_val; /* in mA */
> + u16 soc_tdc_val; /* in mA */
> + u16 vdd_edc_val; /* in mA */
> + u16 soc_edcv_al; /* in mA */
> + u16 infra_cpu_maxfreq; /* in MHz */
> + u16 infra_gfx_maxfreq; /* in MHz */
> + u16 skin_temp; /* in centi-Celsius */
> + u16 device_state;
> +};
> +
> enum amd_stt_skin_temp {
> STT_TEMP_APU,
> STT_TEMP_HS2,
> @@ -121,6 +156,9 @@ struct amd_pmf_dev {
> struct delayed_work heart_beat;
> struct mutex lock; /* protects the PMF interface */
> struct dentry *dbgfs_dir;
> + struct smu_pmf_metrics m_table;
> + struct delayed_work work_buffer;
> + ktime_t start_time;
> };
>
> struct apmf_sps_prop_granular {
> @@ -154,6 +192,7 @@ int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
> void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev);
> int is_apmf_func_supported(unsigned long index);
> int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32 *data);
> +int amd_pmf_init_metrics_table(struct amd_pmf_dev *dev);
> int amd_pmf_get_power_source(void);
>
> /* SPS Layer */
^ permalink raw reply [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 07/11] platform/x86/amd/pmf: Get performance metrics from PMFW
2022-07-28 21:02 ` Limonciello, Mario
@ 2022-07-29 17:43 ` Shyam Sundar S K
0 siblings, 0 replies; 33+ messages in thread
From: Shyam Sundar S K @ 2022-07-29 17:43 UTC (permalink / raw)
To: Limonciello, Mario, hdegoede, markgross; +Cc: platform-driver-x86, Patil.Reddy
Hi Mario,
On 7/29/2022 2:32 AM, Limonciello, Mario wrote:
> On 7/28/2022 13:20, Shyam Sundar S K wrote:
>> PMF driver polls for metrics information from PMFW to understand the
>> system
>> behavior, power consumption etc.
>>
>> This metrics table information will be used the PMF features to tweak the
>> thermal heuristics. The poll duration can also be changed by the user
>> by changing the poll duration time.
>>
>> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
>> ---
>> drivers/platform/x86/amd/pmf/core.c | 56 +++++++++++++++++++++++++++++
>> drivers/platform/x86/amd/pmf/pmf.h | 39 ++++++++++++++++++++
>> 2 files changed, 95 insertions(+)
>>
>> diff --git a/drivers/platform/x86/amd/pmf/core.c
>> b/drivers/platform/x86/amd/pmf/core.c
>> index 6c1c5a89fe71..ede4eefc33a4 100644
>> --- a/drivers/platform/x86/amd/pmf/core.c
>> +++ b/drivers/platform/x86/amd/pmf/core.c
>> @@ -47,6 +47,11 @@
>> #define DELAY_MIN_US 2000
>> #define DELAY_MAX_US 3000
>> +/* override Metrics Table sample size time (in ms) */
>> +static int metrics_table_loop_ms = 1000;
>> +module_param(metrics_table_loop_ms, int, 0644);
>> +MODULE_PARM_DESC(metrics_table_loop_ms, " Metrics Table sample size
>> time (default = 1000ms) ");
>> +
>> static int current_power_limits_show(struct seq_file *seq, void
>> *unused)
>> {
>> struct amd_pmf_dev *dev = seq->private;
>> @@ -88,6 +93,29 @@ int amd_pmf_get_power_source(void)
>> return POWER_SOURCE_DC;
>> }
>> +static void amd_pmf_get_metrics(struct work_struct *work)
>> +{
>> + struct amd_pmf_dev *dev = container_of(work, struct amd_pmf_dev,
>> work_buffer.work);
>> + enum platform_profile_option current_profile;
>> + ktime_t time_elapsed_ms;
>> + int socket_power;
>> +
>> + /* Get the current profile information */
>> + current_profile = READ_ONCE(dev->current_profile);
>
> AFAICT this parameter isn't actually used/needed anymore.
ah! good catch.
Will look forward for other comments and will make this change in the
next version.
Thanks,
Shyam
>
>> +
>> + /* Transfer table contents */
>> + memset(&dev->m_table, 0, sizeof(dev->m_table));
>> + amd_pmf_send_cmd(dev, SET_TRANSFER_TABLE, 0, 7, NULL);
>> + memcpy(&dev->m_table, dev->buf, sizeof(dev->m_table));
>> +
>> + time_elapsed_ms = ktime_to_ms(ktime_get()) - dev->start_time;
>> + /* Calculate the avg SoC power consumption */
>> + socket_power = dev->m_table.apu_power + dev->m_table.dgpu_power;
>> +
>> + dev->start_time = ktime_to_ms(ktime_get());
>> + schedule_delayed_work(&dev->work_buffer,
>> msecs_to_jiffies(metrics_table_loop_ms));
>> +}
>> +
>> static inline u32 amd_pmf_reg_read(struct amd_pmf_dev *dev, int
>> reg_offset)
>> {
>> return ioread32(dev->regbase + reg_offset);
>> @@ -181,6 +209,34 @@ static const struct pci_device_id pmf_pci_ids[] = {
>> { }
>> };
>> +int amd_pmf_init_metrics_table(struct amd_pmf_dev *dev)
>> +{
>> + u64 phys_addr;
>> + u32 hi, low;
>> +
>> + INIT_DELAYED_WORK(&dev->work_buffer, amd_pmf_get_metrics);
>> +
>> + /* Get Metrics Table Address */
>> + dev->buf = kzalloc(sizeof(dev->m_table), GFP_KERNEL);
>> + if (!dev->buf)
>> + return -ENOMEM;
>> +
>> + phys_addr = virt_to_phys(dev->buf);
>> + hi = phys_addr >> 32;
>> + low = phys_addr & GENMASK(31, 0);
>> +
>> + amd_pmf_send_cmd(dev, SET_DRAM_ADDR_HIGH, 0, hi, NULL);
>> + amd_pmf_send_cmd(dev, SET_DRAM_ADDR_LOW, 0, low, NULL);
>> +
>> + /*
>> + * Start collecting the metrics data after a small delay
>> + * or else, we might end up getting stale values from PMFW.
>> + */
>> + schedule_delayed_work(&dev->work_buffer,
>> msecs_to_jiffies(metrics_table_loop_ms * 3));
>> +
>> + return 0;
>> +}
>> +
>> static void amd_pmf_init_features(struct amd_pmf_dev *dev)
>> {
>> /* Enable Static Slider */
>> diff --git a/drivers/platform/x86/amd/pmf/pmf.h
>> b/drivers/platform/x86/amd/pmf/pmf.h
>> index b6501a68aa4e..49d3232ee2e0 100644
>> --- a/drivers/platform/x86/amd/pmf/pmf.h
>> +++ b/drivers/platform/x86/amd/pmf/pmf.h
>> @@ -84,6 +84,41 @@ struct apmf_fan_idx {
>> u32 fan_ctl_idx;
>> } __packed;
>> +struct smu_pmf_metrics {
>> + u16 gfxclk_freq; /* in MHz */
>> + u16 socclk_freq; /* in MHz */
>> + u16 vclk_freq; /* in MHz */
>> + u16 dclk_freq; /* in MHz */
>> + u16 memclk_freq; /* in MHz */
>> + u16 spare;
>> + u16 gfx_activity; /* in Centi */
>> + u16 uvd_activity; /* in Centi */
>> + u16 voltage[2]; /* in mV */
>> + u16 currents[2]; /* in mA */
>> + u16 power[2];/* in mW */
>> + u16 core_freq[8]; /* in MHz */
>> + u16 core_power[8]; /* in mW */
>> + u16 core_temp[8]; /* in centi-Celsius */
>> + u16 l3_freq; /* in MHz */
>> + u16 l3_temp; /* in centi-Celsius */
>> + u16 gfx_temp; /* in centi-Celsius */
>> + u16 soc_temp; /* in centi-Celsius */
>> + u16 throttler_status;
>> + u16 current_socketpower; /* in mW */
>> + u16 stapm_orig_limit; /* in W */
>> + u16 stapm_cur_limit; /* in W */
>> + u32 apu_power; /* in mW */
>> + u32 dgpu_power; /* in mW */
>> + u16 vdd_tdc_val; /* in mA */
>> + u16 soc_tdc_val; /* in mA */
>> + u16 vdd_edc_val; /* in mA */
>> + u16 soc_edcv_al; /* in mA */
>> + u16 infra_cpu_maxfreq; /* in MHz */
>> + u16 infra_gfx_maxfreq; /* in MHz */
>> + u16 skin_temp; /* in centi-Celsius */
>> + u16 device_state;
>> +};
>> +
>> enum amd_stt_skin_temp {
>> STT_TEMP_APU,
>> STT_TEMP_HS2,
>> @@ -121,6 +156,9 @@ struct amd_pmf_dev {
>> struct delayed_work heart_beat;
>> struct mutex lock; /* protects the PMF interface */
>> struct dentry *dbgfs_dir;
>> + struct smu_pmf_metrics m_table;
>> + struct delayed_work work_buffer;
>> + ktime_t start_time;
>> };
>> struct apmf_sps_prop_granular {
>> @@ -154,6 +192,7 @@ int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
>> void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev);
>> int is_apmf_func_supported(unsigned long index);
>> int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get,
>> u32 arg, u32 *data);
>> +int amd_pmf_init_metrics_table(struct amd_pmf_dev *dev);
>> int amd_pmf_get_power_source(void);
>> /* SPS Layer */
>
^ permalink raw reply [flat|nested] 33+ messages in thread
* Re: [PATCH v2 RESEND 07/11] platform/x86/amd/pmf: Get performance metrics from PMFW
2022-07-28 18:20 ` [PATCH v2 RESEND 07/11] platform/x86/amd/pmf: Get performance metrics from PMFW Shyam Sundar S K
2022-07-28 21:02 ` Limonciello, Mario
@ 2022-08-01 11:50 ` Hans de Goede
1 sibling, 0 replies; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 11:50 UTC (permalink / raw)
To: Shyam Sundar S K, markgross; +Cc: platform-driver-x86, Patil.Reddy
Hi,
On 7/28/22 20:20, Shyam Sundar S K wrote:
> PMF driver polls for metrics information from PMFW to understand the system
> behavior, power consumption etc.
>
> This metrics table information will be used the PMF features to tweak the
> thermal heuristics. The poll duration can also be changed by the user
> by changing the poll duration time.
>
> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> ---
> drivers/platform/x86/amd/pmf/core.c | 56 +++++++++++++++++++++++++++++
> drivers/platform/x86/amd/pmf/pmf.h | 39 ++++++++++++++++++++
> 2 files changed, 95 insertions(+)
>
> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
> index 6c1c5a89fe71..ede4eefc33a4 100644
> --- a/drivers/platform/x86/amd/pmf/core.c
> +++ b/drivers/platform/x86/amd/pmf/core.c
> @@ -47,6 +47,11 @@
> #define DELAY_MIN_US 2000
> #define DELAY_MAX_US 3000
>
> +/* override Metrics Table sample size time (in ms) */
> +static int metrics_table_loop_ms = 1000;
> +module_param(metrics_table_loop_ms, int, 0644);
> +MODULE_PARM_DESC(metrics_table_loop_ms, " Metrics Table sample size time (default = 1000ms) ");
> +
> static int current_power_limits_show(struct seq_file *seq, void *unused)
> {
> struct amd_pmf_dev *dev = seq->private;
> @@ -88,6 +93,29 @@ int amd_pmf_get_power_source(void)
> return POWER_SOURCE_DC;
> }
>
> +static void amd_pmf_get_metrics(struct work_struct *work)
> +{
> + struct amd_pmf_dev *dev = container_of(work, struct amd_pmf_dev, work_buffer.work);
> + enum platform_profile_option current_profile;
> + ktime_t time_elapsed_ms;
> + int socket_power;
> +
> + /* Get the current profile information */
> + current_profile = READ_ONCE(dev->current_profile);
As Mario already mentioned current_profile is no longer used, so this can be dropped now.
> +
> + /* Transfer table contents */
> + memset(&dev->m_table, 0, sizeof(dev->m_table));
You are memsetting dev->m_table here, and then 2 lines down completely
overwrite it with the contents of dev->buf. As already mentioned in my review
of v1 do you perhaps intend to memset dev->buf to 0 here ?
> + amd_pmf_send_cmd(dev, SET_TRANSFER_TABLE, 0, 7, NULL);
> + memcpy(&dev->m_table, dev->buf, sizeof(dev->m_table));
> +
> + time_elapsed_ms = ktime_to_ms(ktime_get()) - dev->start_time;
> + /* Calculate the avg SoC power consumption */
> + socket_power = dev->m_table.apu_power + dev->m_table.dgpu_power;
> +
> + dev->start_time = ktime_to_ms(ktime_get());
> + schedule_delayed_work(&dev->work_buffer, msecs_to_jiffies(metrics_table_loop_ms));
> +}
> +
> static inline u32 amd_pmf_reg_read(struct amd_pmf_dev *dev, int reg_offset)
> {
> return ioread32(dev->regbase + reg_offset);
> @@ -181,6 +209,34 @@ static const struct pci_device_id pmf_pci_ids[] = {
> { }
> };
>
> +int amd_pmf_init_metrics_table(struct amd_pmf_dev *dev)
> +{
> + u64 phys_addr;
> + u32 hi, low;
> +
> + INIT_DELAYED_WORK(&dev->work_buffer, amd_pmf_get_metrics);
> +
> + /* Get Metrics Table Address */
> + dev->buf = kzalloc(sizeof(dev->m_table), GFP_KERNEL);
> + if (!dev->buf)
> + return -ENOMEM;
> +
> + phys_addr = virt_to_phys(dev->buf);
> + hi = phys_addr >> 32;
> + low = phys_addr & GENMASK(31, 0);
> +
> + amd_pmf_send_cmd(dev, SET_DRAM_ADDR_HIGH, 0, hi, NULL);
> + amd_pmf_send_cmd(dev, SET_DRAM_ADDR_LOW, 0, low, NULL);
> +
> + /*
> + * Start collecting the metrics data after a small delay
> + * or else, we might end up getting stale values from PMFW.
> + */
> + schedule_delayed_work(&dev->work_buffer, msecs_to_jiffies(metrics_table_loop_ms * 3));
> +
> + return 0;
> +}
> +
> static void amd_pmf_init_features(struct amd_pmf_dev *dev)
> {
> /* Enable Static Slider */
> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
> index b6501a68aa4e..49d3232ee2e0 100644
> --- a/drivers/platform/x86/amd/pmf/pmf.h
> +++ b/drivers/platform/x86/amd/pmf/pmf.h
> @@ -84,6 +84,41 @@ struct apmf_fan_idx {
> u32 fan_ctl_idx;
> } __packed;
>
> +struct smu_pmf_metrics {
> + u16 gfxclk_freq; /* in MHz */
> + u16 socclk_freq; /* in MHz */
> + u16 vclk_freq; /* in MHz */
> + u16 dclk_freq; /* in MHz */
> + u16 memclk_freq; /* in MHz */
> + u16 spare;
> + u16 gfx_activity; /* in Centi */
> + u16 uvd_activity; /* in Centi */
> + u16 voltage[2]; /* in mV */
> + u16 currents[2]; /* in mA */
> + u16 power[2];/* in mW */
> + u16 core_freq[8]; /* in MHz */
> + u16 core_power[8]; /* in mW */
> + u16 core_temp[8]; /* in centi-Celsius */
> + u16 l3_freq; /* in MHz */
> + u16 l3_temp; /* in centi-Celsius */
> + u16 gfx_temp; /* in centi-Celsius */
> + u16 soc_temp; /* in centi-Celsius */
> + u16 throttler_status;
> + u16 current_socketpower; /* in mW */
> + u16 stapm_orig_limit; /* in W */
> + u16 stapm_cur_limit; /* in W */
> + u32 apu_power; /* in mW */
> + u32 dgpu_power; /* in mW */
> + u16 vdd_tdc_val; /* in mA */
> + u16 soc_tdc_val; /* in mA */
> + u16 vdd_edc_val; /* in mA */
> + u16 soc_edcv_al; /* in mA */
> + u16 infra_cpu_maxfreq; /* in MHz */
> + u16 infra_gfx_maxfreq; /* in MHz */
> + u16 skin_temp; /* in centi-Celsius */
> + u16 device_state;
> +};
> +
Since you are memcpy-ing this from memory read from the PMF,
you should use __packed here.
> enum amd_stt_skin_temp {
> STT_TEMP_APU,
> STT_TEMP_HS2,
> @@ -121,6 +156,9 @@ struct amd_pmf_dev {
> struct delayed_work heart_beat;
> struct mutex lock; /* protects the PMF interface */
> struct dentry *dbgfs_dir;
> + struct smu_pmf_metrics m_table;
> + struct delayed_work work_buffer;
> + ktime_t start_time;
> };
>
> struct apmf_sps_prop_granular {
> @@ -154,6 +192,7 @@ int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
> void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev);
> int is_apmf_func_supported(unsigned long index);
> int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32 *data);
> +int amd_pmf_init_metrics_table(struct amd_pmf_dev *dev);
> int amd_pmf_get_power_source(void);
>
> /* SPS Layer */
Regards,
Hans
^ permalink raw reply [flat|nested] 33+ messages in thread
* [PATCH v2 RESEND 08/11] platform/x86/amd/pmf: Add support for Auto mode feature
2022-07-28 18:20 [PATCH v2 RESEND 00/11] platform/x86/amd/pmf: Introduce AMD PMF Driver Shyam Sundar S K
` (6 preceding siblings ...)
2022-07-28 18:20 ` [PATCH v2 RESEND 07/11] platform/x86/amd/pmf: Get performance metrics from PMFW Shyam Sundar S K
@ 2022-07-28 18:20 ` Shyam Sundar S K
2022-08-01 12:55 ` Hans de Goede
2022-07-28 18:20 ` [PATCH v2 RESEND 09/11] platform/x86/amd/pmf: Handle AMT and CQL events for Auto mode Shyam Sundar S K
` (2 subsequent siblings)
10 siblings, 1 reply; 33+ messages in thread
From: Shyam Sundar S K @ 2022-07-28 18:20 UTC (permalink / raw)
To: hdegoede, markgross; +Cc: platform-driver-x86, Patil.Reddy, Shyam Sundar S K
This feature has 3 modes quiet, balanced, performance
The objective of this feature is to track the moving average of system
power over the time period specified and switch to the subsequent mode.
In order to do this, PMF driver will get the moving average of APU power
from PMFW and power threshold, time constants, system config parameters
from OEM inputs.
System power as read by PMF driver from PMFW is the filtered value over
the sampling window. Every sampling window, moving average of system power
is computed. At the end of the monitoring window, the moving average is
compared against the threshold for mode switch for decision making.
With AMD managing the system config limits, any mode switch within
auto-mode will result in limits of fPPT/sPPT/STAPM or STT being scaled
down.
When "auto mode" is enabled, the static slider control remains out of
the PMF driver, so the platform_profile registration would not
happen in PMF driver.
The transition to auto-mode only happens when the APMF fn5 is enabled
in BIOS, platform_profile set to "balanced" and a AMT
(Auto Mode transition) is received.
Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
---
drivers/platform/x86/amd/pmf/Makefile | 3 +-
drivers/platform/x86/amd/pmf/acpi.c | 7 +
drivers/platform/x86/amd/pmf/auto-mode.c | 317 +++++++++++++++++++++++
drivers/platform/x86/amd/pmf/core.c | 9 +
drivers/platform/x86/amd/pmf/pmf.h | 115 ++++++++
5 files changed, 450 insertions(+), 1 deletion(-)
create mode 100644 drivers/platform/x86/amd/pmf/auto-mode.c
diff --git a/drivers/platform/x86/amd/pmf/Makefile b/drivers/platform/x86/amd/pmf/Makefile
index 557521a80427..ef2b08c9174d 100644
--- a/drivers/platform/x86/amd/pmf/Makefile
+++ b/drivers/platform/x86/amd/pmf/Makefile
@@ -5,4 +5,5 @@
#
obj-$(CONFIG_AMD_PMF) += amd-pmf.o
-amd-pmf-objs := core.o acpi.o sps.o
+amd-pmf-objs := core.o acpi.o sps.o \
+ auto-mode.o
diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
index 1814d48f8e44..724150ec58fb 100644
--- a/drivers/platform/x86/amd/pmf/acpi.c
+++ b/drivers/platform/x86/amd/pmf/acpi.c
@@ -99,6 +99,7 @@ static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
{
func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
func->static_slider_granular = mask & APMF_FUNC_STATIC_SLIDER_GRANULAR;
+ func->auto_mode_def = mask & APMF_FUNC_AUTO_MODE;
func->fan_table_idx = mask & APMF_FUNC_SET_FAN_IDX;
}
@@ -160,6 +161,12 @@ int apmf_update_fan_idx(struct apmf_if *ampf_if, bool manual, u32 idx)
return err;
}
+int apmf_get_auto_mode_def(struct apmf_if *apmf_if, struct apmf_auto_mode *data)
+{
+ return apmf_if_call_store_buffer(apmf_if, APMF_FUNC_AUTO_MODE,
+ data, sizeof(*data));
+}
+
static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
{
struct apmf_verify_interface output;
diff --git a/drivers/platform/x86/amd/pmf/auto-mode.c b/drivers/platform/x86/amd/pmf/auto-mode.c
new file mode 100644
index 000000000000..954fde25e71e
--- /dev/null
+++ b/drivers/platform/x86/amd/pmf/auto-mode.c
@@ -0,0 +1,317 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * AMD Platform Management Framework Driver
+ *
+ * Copyright (c) 2022, Advanced Micro Devices, Inc.
+ * All Rights Reserved.
+ *
+ * Author: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
+ */
+
+#include <linux/acpi.h>
+#include <linux/list.h>
+#include <linux/workqueue.h>
+#include "pmf.h"
+
+#define AVG_SAMPLE_SIZE 3
+
+struct power_history {
+ int socket_power;
+ struct list_head list;
+ int avg;
+ int total;
+};
+
+struct list_pdata {
+ int total;
+ int avg;
+};
+
+static struct power_history power_list;
+static struct list_pdata pdata;
+
+static struct auto_mode_mode_config config_store;
+static const char *state_as_str(unsigned int state);
+
+static void amd_pmf_handle_automode(struct amd_pmf_dev *dev, bool op, int idx,
+ struct auto_mode_mode_config *table)
+{
+ if (op == SLIDER_OP_SET) {
+ struct power_table_control *pwr_ctrl = &config_store.mode_set[idx].power_control;
+
+ amd_pmf_send_cmd(dev, SET_SPL, false, pwr_ctrl->spl, NULL);
+ amd_pmf_send_cmd(dev, SET_FPPT, false, pwr_ctrl->fppt, NULL);
+ amd_pmf_send_cmd(dev, SET_SPPT, false, pwr_ctrl->sppt, NULL);
+ amd_pmf_send_cmd(dev, SET_SPPT_APU_ONLY, false, pwr_ctrl->sppt_apu_only, NULL);
+ amd_pmf_send_cmd(dev, SET_STT_MIN_LIMIT, false, pwr_ctrl->stt_min, NULL);
+ amd_pmf_send_cmd(dev, SET_STT_LIMIT_APU, false,
+ pwr_ctrl->stt_skin_temp[STT_TEMP_APU], NULL);
+ amd_pmf_send_cmd(dev, SET_STT_LIMIT_HS2, false,
+ pwr_ctrl->stt_skin_temp[STT_TEMP_HS2], NULL);
+ } else if (op == SLIDER_OP_GET) {
+ amd_pmf_send_cmd(dev, GET_SPL, true, ARG_NONE,
+ &table->mode_set[idx].power_control.spl);
+ amd_pmf_send_cmd(dev, GET_FPPT, true, ARG_NONE,
+ &table->mode_set[idx].power_control.fppt);
+ amd_pmf_send_cmd(dev, GET_SPPT, true, ARG_NONE,
+ &table->mode_set[idx].power_control.sppt);
+ amd_pmf_send_cmd(dev, GET_SPPT_APU_ONLY, true, ARG_NONE,
+ &table->mode_set[idx].power_control.sppt_apu_only);
+ amd_pmf_send_cmd(dev, GET_STT_MIN_LIMIT, true, ARG_NONE,
+ &table->mode_set[idx].power_control.stt_min);
+ amd_pmf_send_cmd(dev, GET_STT_LIMIT_APU, true, ARG_NONE,
+ &table->mode_set[idx].power_control.stt_skin_temp[STT_TEMP_APU]);
+ amd_pmf_send_cmd(dev, GET_STT_LIMIT_HS2, true, ARG_NONE,
+ &table->mode_set[idx].power_control.stt_skin_temp[STT_TEMP_HS2]);
+ }
+
+ if (dev->apmf_if->func.fan_table_idx)
+ apmf_update_fan_idx(dev->apmf_if, config_store.mode_set[idx].fan_control.manual,
+ config_store.mode_set[idx].fan_control.fan_id);
+}
+
+static int amd_pmf_get_moving_avg(int socket_power)
+{
+ struct power_history *tmp;
+ struct list_head *pos, *q;
+ static int count;
+
+ tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
+ tmp->socket_power = socket_power;
+ list_add_tail(&tmp->list, &power_list.list);
+
+ list_for_each_safe(pos, q, &power_list.list) {
+ if (count >= AVG_SAMPLE_SIZE) {
+ tmp = list_first_entry(pos, struct power_history, list);
+ list_del_init(pos);
+ goto next;
+ }
+ }
+
+next:
+ pdata.total = 0;
+ pdata.avg = 0;
+
+ list_for_each(pos, &power_list.list) {
+ tmp = list_entry(pos, struct power_history, list);
+ pdata.total += tmp->socket_power;
+ pdata.avg = pdata.total / AVG_SAMPLE_SIZE;
+ }
+
+ count++;
+ if (count >= AVG_SAMPLE_SIZE)
+ return pdata.avg;
+
+ return 0;
+}
+
+void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t time_elapsed_ms)
+{
+ int avg_power = 0;
+ bool update = false;
+ int i, j;
+
+ /* Get the average moving average computed by auto mode algorithm */
+ avg_power = amd_pmf_get_moving_avg(socket_power);
+
+ for (i = 0; i < AUTO_TRANSITION_MAX; i++) {
+ if ((config_store.transition[i].shifting_up && avg_power >=
+ config_store.transition[i].power_threshold) ||
+ (!config_store.transition[i].shifting_up && avg_power <=
+ config_store.transition[i].power_threshold)) {
+ if (config_store.transition[i].timer <
+ config_store.transition[i].time_constant)
+ config_store.transition[i].timer += time_elapsed_ms;
+ } else {
+ config_store.transition[i].timer = 0;
+ }
+
+ if (config_store.transition[i].timer >=
+ config_store.transition[i].time_constant &&
+ !config_store.transition[i].applied) {
+ config_store.transition[i].applied = true;
+ update = true;
+ } else if (config_store.transition[i].timer <=
+ config_store.transition[i].time_constant &&
+ config_store.transition[i].applied) {
+ config_store.transition[i].applied = false;
+ update = true;
+ }
+ }
+
+ dev_dbg(dev->dev, "[AUTO_MODE] avg power: %u mW mode: %s\n", avg_power,
+ state_as_str(config_store.current_mode));
+
+ if (update) {
+ for (j = 0; j < AUTO_TRANSITION_MAX; j++) {
+ /* Apply the mode with highest priority indentified */
+ if (config_store.transition[j].applied) {
+ if (config_store.current_mode !=
+ config_store.transition[j].target_mode) {
+ config_store.current_mode =
+ config_store.transition[j].target_mode;
+ dev_dbg(dev->dev, "[AUTO_MODE] moving to mode:%s\n",
+ state_as_str(config_store.current_mode));
+ amd_pmf_handle_automode(dev, SLIDER_OP_SET,
+ config_store.current_mode, NULL);
+ }
+ break;
+ }
+ }
+ }
+}
+
+static void amd_pmf_get_power_threshold(void)
+{
+ config_store.transition[AUTO_TRANSITION_TO_QUIET].power_threshold =
+ config_store.mode_set[AUTO_BALANCE].power_floor -
+ config_store.transition[AUTO_TRANSITION_TO_QUIET].power_delta;
+
+ config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].power_threshold =
+ config_store.mode_set[AUTO_BALANCE].power_floor -
+ config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].power_delta;
+
+ config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].power_threshold =
+ config_store.mode_set[AUTO_QUIET].power_floor -
+ config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].power_delta;
+
+ config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].power_threshold =
+ config_store.mode_set[AUTO_PERFORMANCE].power_floor -
+ config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].power_delta;
+}
+
+static const char *state_as_str(unsigned int state)
+{
+ switch (state) {
+ case AUTO_QUIET:
+ return "QUIET";
+ case AUTO_BALANCE:
+ return "BALANCED";
+ case AUTO_PERFORMANCE_ON_LAP:
+ return "ON_LAP";
+ case AUTO_PERFORMANCE:
+ return "PERFORMANCE";
+ default:
+ return "Unknown Auto Mode State";
+ }
+}
+
+void amd_pmf_load_defaults_auto_mode(struct amd_pmf_dev *dev)
+{
+ struct apmf_auto_mode output;
+ struct power_table_control *pwr_ctrl;
+ int i;
+
+ if (dev->apmf_if->func.auto_mode_def) {
+ apmf_get_auto_mode_def(dev->apmf_if, &output);
+ /* time constant */
+ config_store.transition[AUTO_TRANSITION_TO_QUIET].time_constant =
+ output.balanced_to_quiet;
+ config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].time_constant =
+ output.balanced_to_perf;
+ config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].time_constant =
+ output.quiet_to_balanced;
+ config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].time_constant =
+ output.perf_to_balanced;
+
+ /* power floor */
+ config_store.mode_set[AUTO_QUIET].power_floor = output.pfloor_quiet;
+ config_store.mode_set[AUTO_BALANCE].power_floor = output.pfloor_balanced;
+ config_store.mode_set[AUTO_PERFORMANCE].power_floor = output.pfloor_perf;
+ config_store.mode_set[AUTO_PERFORMANCE_ON_LAP].power_floor = output.pfloor_perf;
+
+ /* Power delta for mode change */
+ config_store.transition[AUTO_TRANSITION_TO_QUIET].power_delta =
+ output.pd_balanced_to_quiet;
+ config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].power_delta =
+ output.pd_balanced_to_perf;
+ config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].power_delta =
+ output.pd_quiet_to_balanced;
+ config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].power_delta =
+ output.pd_perf_to_balanced;
+
+ /* Power threshold */
+ amd_pmf_get_power_threshold();
+
+ /* skin temperature limits */
+ pwr_ctrl = &config_store.mode_set[AUTO_QUIET].power_control;
+ pwr_ctrl->spl = output.spl_quiet;
+ pwr_ctrl->sppt = output.sppt_quiet;
+ pwr_ctrl->fppt = output.fppt_quiet;
+ pwr_ctrl->sppt_apu_only = output.sppt_apu_only_quiet;
+ pwr_ctrl->stt_min = output.stt_min_limit_quiet;
+ pwr_ctrl->stt_skin_temp[STT_TEMP_APU] = output.stt_apu_quiet;
+ pwr_ctrl->stt_skin_temp[STT_TEMP_HS2] = output.stt_hs2_quiet;
+
+ pwr_ctrl = &config_store.mode_set[AUTO_BALANCE].power_control;
+ pwr_ctrl->spl = output.spl_balanced;
+ pwr_ctrl->sppt = output.sppt_balanced;
+ pwr_ctrl->fppt = output.fppt_balanced;
+ pwr_ctrl->sppt_apu_only = output.sppt_apu_only_balanced;
+ pwr_ctrl->stt_min = output.stt_min_limit_balanced;
+ pwr_ctrl->stt_skin_temp[STT_TEMP_APU] = output.stt_apu_balanced;
+ pwr_ctrl->stt_skin_temp[STT_TEMP_HS2] = output.stt_hs2_balanced;
+
+ pwr_ctrl = &config_store.mode_set[AUTO_PERFORMANCE].power_control;
+ pwr_ctrl->spl = output.spl_perf;
+ pwr_ctrl->sppt = output.sppt_perf;
+ pwr_ctrl->fppt = output.fppt_perf;
+ pwr_ctrl->sppt_apu_only = output.sppt_apu_only_perf;
+ pwr_ctrl->stt_min = output.stt_min_limit_perf;
+ pwr_ctrl->stt_skin_temp[STT_TEMP_APU] = output.stt_apu_perf;
+ pwr_ctrl->stt_skin_temp[STT_TEMP_HS2] = output.stt_hs2_perf;
+
+ pwr_ctrl = &config_store.mode_set[AUTO_PERFORMANCE_ON_LAP].power_control;
+ pwr_ctrl->spl = output.spl_perf_on_lap;
+ pwr_ctrl->sppt = output.sppt_perf_on_lap;
+ pwr_ctrl->fppt = output.fppt_perf_on_lap;
+ pwr_ctrl->sppt_apu_only = output.sppt_apu_only_perf_on_lap;
+ pwr_ctrl->stt_min = output.stt_min_limit_perf_on_lap;
+ pwr_ctrl->stt_skin_temp[STT_TEMP_APU] = output.stt_apu_perf_on_lap;
+ pwr_ctrl->stt_skin_temp[STT_TEMP_HS2] = output.stt_hs2_perf_on_lap;
+
+ /* Fan ID */
+ config_store.mode_set[AUTO_QUIET].fan_control.fan_id = output.fan_id_quiet;
+ config_store.mode_set[AUTO_BALANCE].fan_control.fan_id = output.fan_id_balanced;
+ config_store.mode_set[AUTO_PERFORMANCE].fan_control.fan_id = output.fan_id_perf;
+ config_store.mode_set[AUTO_PERFORMANCE_ON_LAP].fan_control.fan_id =
+ output.fan_id_perf;
+
+ config_store.transition[AUTO_TRANSITION_TO_QUIET].target_mode = AUTO_QUIET;
+ config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode =
+ AUTO_PERFORMANCE;
+ config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].target_mode =
+ AUTO_BALANCE;
+ config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].target_mode =
+ AUTO_BALANCE;
+
+ config_store.transition[AUTO_TRANSITION_TO_QUIET].shifting_up = false;
+ config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].shifting_up = true;
+ config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].shifting_up = true;
+ config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].shifting_up =
+ false;
+
+ for (i = 0 ; i < AUTO_MODE_MAX ; i++) {
+ if (config_store.mode_set[i].fan_control.fan_id == FAN_INDEX_AUTO)
+ config_store.mode_set[i].fan_control.manual = false;
+ else
+ config_store.mode_set[i].fan_control.manual = true;
+ }
+ }
+}
+
+void amd_pmf_deinit_auto_mode(struct amd_pmf_dev *dev)
+{
+ cancel_delayed_work_sync(&dev->work_buffer);
+}
+
+void amd_pmf_init_auto_mode(struct amd_pmf_dev *dev)
+{
+ INIT_LIST_HEAD(&power_list.list);
+
+ amd_pmf_init_metrics_table(dev);
+ amd_pmf_load_defaults_auto_mode(dev);
+
+ /* update the thermal for Automode */
+ amd_pmf_handle_automode(dev, SLIDER_OP_SET, config_store.current_mode, NULL);
+}
diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
index ede4eefc33a4..ef71f8326248 100644
--- a/drivers/platform/x86/amd/pmf/core.c
+++ b/drivers/platform/x86/amd/pmf/core.c
@@ -244,12 +244,21 @@ static void amd_pmf_init_features(struct amd_pmf_dev *dev)
amd_pmf_init_sps(dev);
dev_dbg(dev->dev, "SPS enabled and Platform Profiles registered\n");
}
+
+ /* Enable Auto Mode */
+ if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE)) {
+ amd_pmf_init_auto_mode(dev);
+ dev_dbg(dev->dev, "Auto Mode Init done\n");
+ }
}
static void amd_pmf_deinit_features(struct amd_pmf_dev *dev)
{
if (is_apmf_func_supported(APMF_FUNC_STATIC_SLIDER_GRANULAR))
amd_pmf_deinit_sps(dev);
+
+ if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE))
+ amd_pmf_deinit_auto_mode(dev);
}
static const struct acpi_device_id amd_pmf_acpi_ids[] = {
diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
index 49d3232ee2e0..4618ddc5a662 100644
--- a/drivers/platform/x86/amd/pmf/pmf.h
+++ b/drivers/platform/x86/amd/pmf/pmf.h
@@ -18,6 +18,7 @@
#define APMF_FUNC_VERIFY_INTERFACE 0
#define APMF_FUNC_GET_SYS_PARAMS 1
#define APMF_FUNC_SBIOS_HEARTBEAT 4
+#define APMF_FUNC_AUTO_MODE 5
#define APMF_FUNC_SET_FAN_IDX 7
#define APMF_FUNC_STATIC_SLIDER_GRANULAR 9
@@ -49,6 +50,7 @@
struct apmf_if_functions {
bool system_params;
bool sbios_heartbeat;
+ bool auto_mode_def;
bool fan_table_idx;
bool static_slider_granular;
};
@@ -187,6 +189,111 @@ struct fan_table_control {
unsigned long fan_id;
};
+struct power_table_control {
+ u32 spl;
+ u32 sppt;
+ u32 fppt;
+ u32 sppt_apu_only;
+ u32 stt_min;
+ u32 stt_skin_temp[STT_TEMP_COUNT];
+ u32 reserved[16];
+};
+
+/* Auto Mode Layer */
+enum auto_mode_transition_priority {
+ AUTO_TRANSITION_TO_PERFORMANCE, /* Any other mode to Performance Mode */
+ AUTO_TRANSITION_FROM_QUIET_TO_BALANCE, /* Quiet Mode to Balance Mode */
+ AUTO_TRANSITION_TO_QUIET, /* Any other mode to Quiet Mode */
+ AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE, /* Performance Mode to Balance Mode */
+ AUTO_TRANSITION_MAX,
+};
+
+enum auto_mode_mode {
+ AUTO_QUIET,
+ AUTO_BALANCE,
+ AUTO_PERFORMANCE_ON_LAP,
+ AUTO_PERFORMANCE,
+ AUTO_MODE_MAX,
+};
+
+struct auto_mode_trans_params {
+ u32 time_constant; /* minimum time required to switch to next mode */
+ u32 power_delta; /* delta power to shift mode */
+ u32 power_threshold;
+ u32 timer; /* elapsed time. if timer > TimeThreshold, it will move to next mode */
+ u32 applied;
+ enum auto_mode_mode target_mode;
+ u32 shifting_up;
+};
+
+struct auto_mode_mode_settings {
+ struct power_table_control power_control;
+ struct fan_table_control fan_control;
+ u32 power_floor;
+};
+
+struct auto_mode_mode_config {
+ struct auto_mode_trans_params transition[AUTO_TRANSITION_MAX];
+ struct auto_mode_mode_settings mode_set[AUTO_MODE_MAX];
+ enum auto_mode_mode current_mode;
+ bool on_lap;
+ bool better_perf;
+ u32 amt_enabled; /* Auto Mode Transition */
+ u32 avg_power;
+};
+
+struct apmf_auto_mode {
+ u16 size;
+ /* time constant */
+ u32 balanced_to_perf;
+ u32 perf_to_balanced;
+ u32 quiet_to_balanced;
+ u32 balanced_to_quiet;
+ /* power floor */
+ u32 pfloor_perf;
+ u32 pfloor_balanced;
+ u32 pfloor_quiet;
+ /* Power delta for mode change */
+ u32 pd_balanced_to_perf;
+ u32 pd_perf_to_balanced;
+ u32 pd_quiet_to_balanced;
+ u32 pd_balanced_to_quiet;
+ /* skin temperature limits */
+ u8 stt_apu_perf_on_lap; /* CQL ON */
+ u8 stt_hs2_perf_on_lap; /* CQL ON */
+ u8 stt_apu_perf;
+ u8 stt_hs2_perf;
+ u8 stt_apu_balanced;
+ u8 stt_hs2_balanced;
+ u8 stt_apu_quiet;
+ u8 stt_hs2_quiet;
+ u32 stt_min_limit_perf_on_lap; /* CQL ON */
+ u32 stt_min_limit_perf;
+ u32 stt_min_limit_balanced;
+ u32 stt_min_limit_quiet;
+ /* SPL based */
+ u32 fppt_perf_on_lap; /* CQL ON */
+ u32 sppt_perf_on_lap; /* CQL ON */
+ u32 spl_perf_on_lap; /* CQL ON */
+ u32 sppt_apu_only_perf_on_lap; /* CQL ON */
+ u32 fppt_perf;
+ u32 sppt_perf;
+ u32 spl_perf;
+ u32 sppt_apu_only_perf;
+ u32 fppt_balanced;
+ u32 sppt_balanced;
+ u32 spl_balanced;
+ u32 sppt_apu_only_balanced;
+ u32 fppt_quiet;
+ u32 sppt_quiet;
+ u32 spl_quiet;
+ u32 sppt_apu_only_quiet;
+ /* Fan ID */
+ u32 fan_id_perf;
+ u32 fan_id_balanced;
+ u32 fan_id_quiet;
+} __packed;
+
/* Core Layer */
int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev);
@@ -206,4 +313,12 @@ int apmf_get_static_slider_granular(struct apmf_if *ampf_if,
void amd_pmf_load_defaults_sps(struct amd_pmf_dev *dev);
int apmf_update_fan_idx(struct apmf_if *ampf_if, bool manual, u32 idx);
+
+/* Auto Mode Layer */
+void amd_pmf_load_defaults_auto_mode(struct amd_pmf_dev *dev);
+int apmf_get_auto_mode_def(struct apmf_if *ampf_if, struct apmf_auto_mode *data);
+void amd_pmf_init_auto_mode(struct amd_pmf_dev *dev);
+void amd_pmf_deinit_auto_mode(struct amd_pmf_dev *dev);
+void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t time_elapsed_ms);
+
#endif /* PMF_H */
--
2.25.1
^ permalink raw reply related [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 08/11] platform/x86/amd/pmf: Add support for Auto mode feature
2022-07-28 18:20 ` [PATCH v2 RESEND 08/11] platform/x86/amd/pmf: Add support for Auto mode feature Shyam Sundar S K
@ 2022-08-01 12:55 ` Hans de Goede
2022-08-01 13:20 ` Hans de Goede
0 siblings, 1 reply; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 12:55 UTC (permalink / raw)
To: Shyam Sundar S K, markgross; +Cc: platform-driver-x86, Patil.Reddy
Hi,
On 7/28/22 20:20, Shyam Sundar S K wrote:
> This feature has 3 modes quiet, balanced, performance
>
> The objective of this feature is to track the moving average of system
> power over the time period specified and switch to the subsequent mode.
>
> In order to do this, PMF driver will get the moving average of APU power
> from PMFW and power threshold, time constants, system config parameters
> from OEM inputs.
>
> System power as read by PMF driver from PMFW is the filtered value over
> the sampling window. Every sampling window, moving average of system power
> is computed. At the end of the monitoring window, the moving average is
> compared against the threshold for mode switch for decision making.
>
> With AMD managing the system config limits, any mode switch within
> auto-mode will result in limits of fPPT/sPPT/STAPM or STT being scaled
> down.
>
> When "auto mode" is enabled, the static slider control remains out of
> the PMF driver, so the platform_profile registration would not
> happen in PMF driver.
>
> The transition to auto-mode only happens when the APMF fn5 is enabled
> in BIOS, platform_profile set to "balanced" and a AMT
> (Auto Mode transition) is received.
>
> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> ---
> drivers/platform/x86/amd/pmf/Makefile | 3 +-
> drivers/platform/x86/amd/pmf/acpi.c | 7 +
> drivers/platform/x86/amd/pmf/auto-mode.c | 317 +++++++++++++++++++++++
> drivers/platform/x86/amd/pmf/core.c | 9 +
> drivers/platform/x86/amd/pmf/pmf.h | 115 ++++++++
> 5 files changed, 450 insertions(+), 1 deletion(-)
> create mode 100644 drivers/platform/x86/amd/pmf/auto-mode.c
>
> diff --git a/drivers/platform/x86/amd/pmf/Makefile b/drivers/platform/x86/amd/pmf/Makefile
> index 557521a80427..ef2b08c9174d 100644
> --- a/drivers/platform/x86/amd/pmf/Makefile
> +++ b/drivers/platform/x86/amd/pmf/Makefile
> @@ -5,4 +5,5 @@
> #
>
> obj-$(CONFIG_AMD_PMF) += amd-pmf.o
> -amd-pmf-objs := core.o acpi.o sps.o
> +amd-pmf-objs := core.o acpi.o sps.o \
> + auto-mode.o
> diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
> index 1814d48f8e44..724150ec58fb 100644
> --- a/drivers/platform/x86/amd/pmf/acpi.c
> +++ b/drivers/platform/x86/amd/pmf/acpi.c
> @@ -99,6 +99,7 @@ static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
> {
> func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
> func->static_slider_granular = mask & APMF_FUNC_STATIC_SLIDER_GRANULAR;
> + func->auto_mode_def = mask & APMF_FUNC_AUTO_MODE;
I just noticed that this is not correct, this should be:
func->system_params = mask & BIT(APMF_FUNC_... - 1);
Which does something completely different!
Also this information is duplicated by:
is_apmf_func_supported(APMF_FUNC_...) so please drop the apmf_if_functions
struct completely and use is_apmf_func_supported() everywhere.
> func->fan_table_idx = mask & APMF_FUNC_SET_FAN_IDX;
> }
>
> @@ -160,6 +161,12 @@ int apmf_update_fan_idx(struct apmf_if *ampf_if, bool manual, u32 idx)
> return err;
> }
>
> +int apmf_get_auto_mode_def(struct apmf_if *apmf_if, struct apmf_auto_mode *data)
> +{
> + return apmf_if_call_store_buffer(apmf_if, APMF_FUNC_AUTO_MODE,
> + data, sizeof(*data));
> +}
> +
> static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
> {
> struct apmf_verify_interface output;
> diff --git a/drivers/platform/x86/amd/pmf/auto-mode.c b/drivers/platform/x86/amd/pmf/auto-mode.c
> new file mode 100644
> index 000000000000..954fde25e71e
> --- /dev/null
> +++ b/drivers/platform/x86/amd/pmf/auto-mode.c
> @@ -0,0 +1,317 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * AMD Platform Management Framework Driver
> + *
> + * Copyright (c) 2022, Advanced Micro Devices, Inc.
> + * All Rights Reserved.
> + *
> + * Author: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> + */
> +
> +#include <linux/acpi.h>
> +#include <linux/list.h>
Please drop the list.h include, see below.
> +#include <linux/workqueue.h>
> +#include "pmf.h"
> +
> +#define AVG_SAMPLE_SIZE 3
> +
please replace the code from here.
> +struct power_history {
> + int socket_power;
> + struct list_head list;
> + int avg;
> + int total;
> +};
> +
> +struct list_pdata {
> + int total;
> + int avg;
> +};
> +
> +static struct power_history power_list;
> +static struct list_pdata pdata;
to here with:
static int socket_power_history[AVG_SAMPLE_SIZE];
static int socket_power_history_idx = -1;
> +
> +static struct auto_mode_mode_config config_store;
> +static const char *state_as_str(unsigned int state);
> +
> +static void amd_pmf_handle_automode(struct amd_pmf_dev *dev, bool op, int idx,
> + struct auto_mode_mode_config *table)
> +{
all callers use SLIDER_OP_GET for the "op" argument. Please rename this to
amd_pmf_set_automode(), drop the "op" argument and drop the SLIDER_OP_GET
handling code.
> + if (op == SLIDER_OP_SET) {
> + struct power_table_control *pwr_ctrl = &config_store.mode_set[idx].power_control;
> +
> + amd_pmf_send_cmd(dev, SET_SPL, false, pwr_ctrl->spl, NULL);
> + amd_pmf_send_cmd(dev, SET_FPPT, false, pwr_ctrl->fppt, NULL);
> + amd_pmf_send_cmd(dev, SET_SPPT, false, pwr_ctrl->sppt, NULL);
> + amd_pmf_send_cmd(dev, SET_SPPT_APU_ONLY, false, pwr_ctrl->sppt_apu_only, NULL);
> + amd_pmf_send_cmd(dev, SET_STT_MIN_LIMIT, false, pwr_ctrl->stt_min, NULL);
> + amd_pmf_send_cmd(dev, SET_STT_LIMIT_APU, false,
> + pwr_ctrl->stt_skin_temp[STT_TEMP_APU], NULL);
> + amd_pmf_send_cmd(dev, SET_STT_LIMIT_HS2, false,
> + pwr_ctrl->stt_skin_temp[STT_TEMP_HS2], NULL);
> + } else if (op == SLIDER_OP_GET) {
> + amd_pmf_send_cmd(dev, GET_SPL, true, ARG_NONE,
> + &table->mode_set[idx].power_control.spl);
> + amd_pmf_send_cmd(dev, GET_FPPT, true, ARG_NONE,
> + &table->mode_set[idx].power_control.fppt);
> + amd_pmf_send_cmd(dev, GET_SPPT, true, ARG_NONE,
> + &table->mode_set[idx].power_control.sppt);
> + amd_pmf_send_cmd(dev, GET_SPPT_APU_ONLY, true, ARG_NONE,
> + &table->mode_set[idx].power_control.sppt_apu_only);
> + amd_pmf_send_cmd(dev, GET_STT_MIN_LIMIT, true, ARG_NONE,
> + &table->mode_set[idx].power_control.stt_min);
> + amd_pmf_send_cmd(dev, GET_STT_LIMIT_APU, true, ARG_NONE,
> + &table->mode_set[idx].power_control.stt_skin_temp[STT_TEMP_APU]);
> + amd_pmf_send_cmd(dev, GET_STT_LIMIT_HS2, true, ARG_NONE,
> + &table->mode_set[idx].power_control.stt_skin_temp[STT_TEMP_HS2]);
> + }
> +
> + if (dev->apmf_if->func.fan_table_idx)
> + apmf_update_fan_idx(dev->apmf_if, config_store.mode_set[idx].fan_control.manual,
> + config_store.mode_set[idx].fan_control.fan_id);
> +}
> +
> +static int amd_pmf_get_moving_avg(int socket_power)
> +{
> + struct power_history *tmp;
> + struct list_head *pos, *q;
> + static int count;
Why is this here, but is power_list a global ?
Note that you can initialize a (static) list while declaring it. E.g. :
static LIST_HEAD(power_list);
But as I mention below this whole code can be mode much much simpler.
> +
> + tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
> + tmp->socket_power = socket_power;
> + list_add_tail(&tmp->list, &power_list.list);
> +
> + list_for_each_safe(pos, q, &power_list.list) {
> + if (count >= AVG_SAMPLE_SIZE) {
> + tmp = list_first_entry(pos, struct power_history, list);
> + list_del_init(pos);
> + goto next;
> + }
> + }
> +
> +next:
> + pdata.total = 0;
> + pdata.avg = 0;
> +
> + list_for_each(pos, &power_list.list) {
> + tmp = list_entry(pos, struct power_history, list);
> + pdata.total += tmp->socket_power;
> + pdata.avg = pdata.total / AVG_SAMPLE_SIZE;
> + }
> +
> + count++;
> + if (count >= AVG_SAMPLE_SIZE)
> + return pdata.avg;
> +
> + return 0;
> +}
Please replace all this list + alloc + free magic which is very hard to review
and needlessly complicated with this much simpler implementation:
static int amd_pmf_get_moving_avg(int socket_power)
{
int i, total = 0;
if (socket_power_history_idx == -1) {
for (i = 0; i < AVG_SAMPLE_SIZE; i++)
socket_power_history[i] = socket_power;
}
socket_power_history_idx = (socket_power_history_idx + 1) % AVG_SAMPLE_SIZE;
socket_power_history[socket_power_history_idx] = socket_power;
for (i = 0; i < AVG_SAMPLE_SIZE; i++)
total += socket_power_history[i];
return total / AVG_SAMPLE_SIZE;
}
> +
> +void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t time_elapsed_ms)
> +{
> + int avg_power = 0;
> + bool update = false;
> + int i, j;
> +
> + /* Get the average moving average computed by auto mode algorithm */
> + avg_power = amd_pmf_get_moving_avg(socket_power);
> +
> + for (i = 0; i < AUTO_TRANSITION_MAX; i++) {
> + if ((config_store.transition[i].shifting_up && avg_power >=
> + config_store.transition[i].power_threshold) ||
> + (!config_store.transition[i].shifting_up && avg_power <=
> + config_store.transition[i].power_threshold)) {
> + if (config_store.transition[i].timer <
> + config_store.transition[i].time_constant)
> + config_store.transition[i].timer += time_elapsed_ms;
> + } else {
> + config_store.transition[i].timer = 0;
> + }
> +
> + if (config_store.transition[i].timer >=
> + config_store.transition[i].time_constant &&
> + !config_store.transition[i].applied) {
> + config_store.transition[i].applied = true;
> + update = true;
> + } else if (config_store.transition[i].timer <=
> + config_store.transition[i].time_constant &&
> + config_store.transition[i].applied) {
> + config_store.transition[i].applied = false;
> + update = true;
> + }
> + }
> +
> + dev_dbg(dev->dev, "[AUTO_MODE] avg power: %u mW mode: %s\n", avg_power,
> + state_as_str(config_store.current_mode));
> +
> + if (update) {
> + for (j = 0; j < AUTO_TRANSITION_MAX; j++) {
> + /* Apply the mode with highest priority indentified */
> + if (config_store.transition[j].applied) {
> + if (config_store.current_mode !=
> + config_store.transition[j].target_mode) {
> + config_store.current_mode =
> + config_store.transition[j].target_mode;
> + dev_dbg(dev->dev, "[AUTO_MODE] moving to mode:%s\n",
> + state_as_str(config_store.current_mode));
> + amd_pmf_handle_automode(dev, SLIDER_OP_SET,
> + config_store.current_mode, NULL);
> + }
> + break;
> + }
> + }
> + }
> +}
> +
> +static void amd_pmf_get_power_threshold(void)
> +{
> + config_store.transition[AUTO_TRANSITION_TO_QUIET].power_threshold =
> + config_store.mode_set[AUTO_BALANCE].power_floor -
> + config_store.transition[AUTO_TRANSITION_TO_QUIET].power_delta;
> +
> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].power_threshold =
> + config_store.mode_set[AUTO_BALANCE].power_floor -
> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].power_delta;
> +
> + config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].power_threshold =
> + config_store.mode_set[AUTO_QUIET].power_floor -
> + config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].power_delta;
> +
> + config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].power_threshold =
> + config_store.mode_set[AUTO_PERFORMANCE].power_floor -
> + config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].power_delta;
> +}
> +
> +static const char *state_as_str(unsigned int state)
> +{
> + switch (state) {
> + case AUTO_QUIET:
> + return "QUIET";
> + case AUTO_BALANCE:
> + return "BALANCED";
> + case AUTO_PERFORMANCE_ON_LAP:
> + return "ON_LAP";
> + case AUTO_PERFORMANCE:
> + return "PERFORMANCE";
> + default:
> + return "Unknown Auto Mode State";
> + }
> +}
> +
> +void amd_pmf_load_defaults_auto_mode(struct amd_pmf_dev *dev)
> +{
> + struct apmf_auto_mode output;
> + struct power_table_control *pwr_ctrl;
> + int i;
> +
> + if (dev->apmf_if->func.auto_mode_def) {
This is already checked by the "if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE))"
check inside amd_pmf_init_features(), please drop this check.
> + apmf_get_auto_mode_def(dev->apmf_if, &output);
> + /* time constant */
> + config_store.transition[AUTO_TRANSITION_TO_QUIET].time_constant =
> + output.balanced_to_quiet;
> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].time_constant =
> + output.balanced_to_perf;
> + config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].time_constant =
> + output.quiet_to_balanced;
> + config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].time_constant =
> + output.perf_to_balanced;
> +
> + /* power floor */
> + config_store.mode_set[AUTO_QUIET].power_floor = output.pfloor_quiet;
> + config_store.mode_set[AUTO_BALANCE].power_floor = output.pfloor_balanced;
> + config_store.mode_set[AUTO_PERFORMANCE].power_floor = output.pfloor_perf;
> + config_store.mode_set[AUTO_PERFORMANCE_ON_LAP].power_floor = output.pfloor_perf;
> +
> + /* Power delta for mode change */
> + config_store.transition[AUTO_TRANSITION_TO_QUIET].power_delta =
> + output.pd_balanced_to_quiet;
> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].power_delta =
> + output.pd_balanced_to_perf;
> + config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].power_delta =
> + output.pd_quiet_to_balanced;
> + config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].power_delta =
> + output.pd_perf_to_balanced;
> +
> + /* Power threshold */
> + amd_pmf_get_power_threshold();
> +
> + /* skin temperature limits */
> + pwr_ctrl = &config_store.mode_set[AUTO_QUIET].power_control;
> + pwr_ctrl->spl = output.spl_quiet;
> + pwr_ctrl->sppt = output.sppt_quiet;
> + pwr_ctrl->fppt = output.fppt_quiet;
> + pwr_ctrl->sppt_apu_only = output.sppt_apu_only_quiet;
> + pwr_ctrl->stt_min = output.stt_min_limit_quiet;
> + pwr_ctrl->stt_skin_temp[STT_TEMP_APU] = output.stt_apu_quiet;
> + pwr_ctrl->stt_skin_temp[STT_TEMP_HS2] = output.stt_hs2_quiet;
> +
> + pwr_ctrl = &config_store.mode_set[AUTO_BALANCE].power_control;
> + pwr_ctrl->spl = output.spl_balanced;
> + pwr_ctrl->sppt = output.sppt_balanced;
> + pwr_ctrl->fppt = output.fppt_balanced;
> + pwr_ctrl->sppt_apu_only = output.sppt_apu_only_balanced;
> + pwr_ctrl->stt_min = output.stt_min_limit_balanced;
> + pwr_ctrl->stt_skin_temp[STT_TEMP_APU] = output.stt_apu_balanced;
> + pwr_ctrl->stt_skin_temp[STT_TEMP_HS2] = output.stt_hs2_balanced;
> +
> + pwr_ctrl = &config_store.mode_set[AUTO_PERFORMANCE].power_control;
> + pwr_ctrl->spl = output.spl_perf;
> + pwr_ctrl->sppt = output.sppt_perf;
> + pwr_ctrl->fppt = output.fppt_perf;
> + pwr_ctrl->sppt_apu_only = output.sppt_apu_only_perf;
> + pwr_ctrl->stt_min = output.stt_min_limit_perf;
> + pwr_ctrl->stt_skin_temp[STT_TEMP_APU] = output.stt_apu_perf;
> + pwr_ctrl->stt_skin_temp[STT_TEMP_HS2] = output.stt_hs2_perf;
> +
> + pwr_ctrl = &config_store.mode_set[AUTO_PERFORMANCE_ON_LAP].power_control;
> + pwr_ctrl->spl = output.spl_perf_on_lap;
> + pwr_ctrl->sppt = output.sppt_perf_on_lap;
> + pwr_ctrl->fppt = output.fppt_perf_on_lap;
> + pwr_ctrl->sppt_apu_only = output.sppt_apu_only_perf_on_lap;
> + pwr_ctrl->stt_min = output.stt_min_limit_perf_on_lap;
> + pwr_ctrl->stt_skin_temp[STT_TEMP_APU] = output.stt_apu_perf_on_lap;
> + pwr_ctrl->stt_skin_temp[STT_TEMP_HS2] = output.stt_hs2_perf_on_lap;
> +
> + /* Fan ID */
> + config_store.mode_set[AUTO_QUIET].fan_control.fan_id = output.fan_id_quiet;
> + config_store.mode_set[AUTO_BALANCE].fan_control.fan_id = output.fan_id_balanced;
> + config_store.mode_set[AUTO_PERFORMANCE].fan_control.fan_id = output.fan_id_perf;
> + config_store.mode_set[AUTO_PERFORMANCE_ON_LAP].fan_control.fan_id =
> + output.fan_id_perf;
> +
> + config_store.transition[AUTO_TRANSITION_TO_QUIET].target_mode = AUTO_QUIET;
> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode =
> + AUTO_PERFORMANCE;
> + config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].target_mode =
> + AUTO_BALANCE;
> + config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].target_mode =
> + AUTO_BALANCE;
> +
> + config_store.transition[AUTO_TRANSITION_TO_QUIET].shifting_up = false;
> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].shifting_up = true;
> + config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].shifting_up = true;
> + config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].shifting_up =
> + false;
> +
> + for (i = 0 ; i < AUTO_MODE_MAX ; i++) {
> + if (config_store.mode_set[i].fan_control.fan_id == FAN_INDEX_AUTO)
> + config_store.mode_set[i].fan_control.manual = false;
> + else
> + config_store.mode_set[i].fan_control.manual = true;
> + }
> + }
This function should also set config_store.current_mode to some sane default (probably balanced).
> +}
> +
> +void amd_pmf_deinit_auto_mode(struct amd_pmf_dev *dev)
> +{
> + cancel_delayed_work_sync(&dev->work_buffer);
> +}
> +
> +void amd_pmf_init_auto_mode(struct amd_pmf_dev *dev)
> +{
> + INIT_LIST_HEAD(&power_list.list);
> +
> + amd_pmf_init_metrics_table(dev);
amd_pmf_init_metrics_table(dev) starts the work, so this needs to be done as the last
step in this function.
> + amd_pmf_load_defaults_auto_mode(dev);
> +
> + /* update the thermal for Automode */
> + amd_pmf_handle_automode(dev, SLIDER_OP_SET, config_store.current_mode, NULL);
> +}
> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
> index ede4eefc33a4..ef71f8326248 100644
> --- a/drivers/platform/x86/amd/pmf/core.c
> +++ b/drivers/platform/x86/amd/pmf/core.c
> @@ -244,12 +244,21 @@ static void amd_pmf_init_features(struct amd_pmf_dev *dev)
> amd_pmf_init_sps(dev);
> dev_dbg(dev->dev, "SPS enabled and Platform Profiles registered\n");
> }
> +
> + /* Enable Auto Mode */
> + if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE)) {
> + amd_pmf_init_auto_mode(dev);
> + dev_dbg(dev->dev, "Auto Mode Init done\n");
> + }
> }
>
> static void amd_pmf_deinit_features(struct amd_pmf_dev *dev)
> {
> if (is_apmf_func_supported(APMF_FUNC_STATIC_SLIDER_GRANULAR))
> amd_pmf_deinit_sps(dev);
> +
> + if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE))
> + amd_pmf_deinit_auto_mode(dev);
> }
>
> static const struct acpi_device_id amd_pmf_acpi_ids[] = {
> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
> index 49d3232ee2e0..4618ddc5a662 100644
> --- a/drivers/platform/x86/amd/pmf/pmf.h
> +++ b/drivers/platform/x86/amd/pmf/pmf.h
> @@ -18,6 +18,7 @@
> #define APMF_FUNC_VERIFY_INTERFACE 0
> #define APMF_FUNC_GET_SYS_PARAMS 1
> #define APMF_FUNC_SBIOS_HEARTBEAT 4
> +#define APMF_FUNC_AUTO_MODE 5
> #define APMF_FUNC_SET_FAN_IDX 7
> #define APMF_FUNC_STATIC_SLIDER_GRANULAR 9
>
> @@ -49,6 +50,7 @@
> struct apmf_if_functions {
> bool system_params;
> bool sbios_heartbeat;
> + bool auto_mode_def;
> bool fan_table_idx;
> bool static_slider_granular;
> };
> @@ -187,6 +189,111 @@ struct fan_table_control {
> unsigned long fan_id;
> };
>
> +struct power_table_control {
> + u32 spl;
> + u32 sppt;
> + u32 fppt;
> + u32 sppt_apu_only;
> + u32 stt_min;
> + u32 stt_skin_temp[STT_TEMP_COUNT];
> + u32 reserved[16];
> +};
> +
> +/* Auto Mode Layer */
> +enum auto_mode_transition_priority {
> + AUTO_TRANSITION_TO_PERFORMANCE, /* Any other mode to Performance Mode */
> + AUTO_TRANSITION_FROM_QUIET_TO_BALANCE, /* Quiet Mode to Balance Mode */
> + AUTO_TRANSITION_TO_QUIET, /* Any other mode to Quiet Mode */
> + AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE, /* Performance Mode to Balance Mode */
> + AUTO_TRANSITION_MAX,
> +};
> +
> +enum auto_mode_mode {
> + AUTO_QUIET,
> + AUTO_BALANCE,
> + AUTO_PERFORMANCE_ON_LAP,
> + AUTO_PERFORMANCE,
> + AUTO_MODE_MAX,
> +};
> +
> +struct auto_mode_trans_params {
> + u32 time_constant; /* minimum time required to switch to next mode */
> + u32 power_delta; /* delta power to shift mode */
> + u32 power_threshold;
> + u32 timer; /* elapsed time. if timer > TimeThreshold, it will move to next mode */
> + u32 applied;
> + enum auto_mode_mode target_mode;
> + u32 shifting_up;
> +};
> +
> +struct auto_mode_mode_settings {
> + struct power_table_control power_control;
> + struct fan_table_control fan_control;
> + u32 power_floor;
> +};
> +
> +struct auto_mode_mode_config {
> + struct auto_mode_trans_params transition[AUTO_TRANSITION_MAX];
> + struct auto_mode_mode_settings mode_set[AUTO_MODE_MAX];
> + enum auto_mode_mode current_mode;
> + bool on_lap;
> + bool better_perf;
> + u32 amt_enabled; /* Auto Mode Transition */
> + u32 avg_power;
> +};
> +
> +struct apmf_auto_mode {
> + u16 size;
> + /* time constant */
> + u32 balanced_to_perf;
> + u32 perf_to_balanced;
> + u32 quiet_to_balanced;
> + u32 balanced_to_quiet;
> + /* power floor */
> + u32 pfloor_perf;
> + u32 pfloor_balanced;
> + u32 pfloor_quiet;
> + /* Power delta for mode change */
> + u32 pd_balanced_to_perf;
> + u32 pd_perf_to_balanced;
> + u32 pd_quiet_to_balanced;
> + u32 pd_balanced_to_quiet;
> + /* skin temperature limits */
> + u8 stt_apu_perf_on_lap; /* CQL ON */
> + u8 stt_hs2_perf_on_lap; /* CQL ON */
> + u8 stt_apu_perf;
> + u8 stt_hs2_perf;
> + u8 stt_apu_balanced;
> + u8 stt_hs2_balanced;
> + u8 stt_apu_quiet;
> + u8 stt_hs2_quiet;
> + u32 stt_min_limit_perf_on_lap; /* CQL ON */
> + u32 stt_min_limit_perf;
> + u32 stt_min_limit_balanced;
> + u32 stt_min_limit_quiet;
> + /* SPL based */
> + u32 fppt_perf_on_lap; /* CQL ON */
> + u32 sppt_perf_on_lap; /* CQL ON */
> + u32 spl_perf_on_lap; /* CQL ON */
> + u32 sppt_apu_only_perf_on_lap; /* CQL ON */
> + u32 fppt_perf;
> + u32 sppt_perf;
> + u32 spl_perf;
> + u32 sppt_apu_only_perf;
> + u32 fppt_balanced;
> + u32 sppt_balanced;
> + u32 spl_balanced;
> + u32 sppt_apu_only_balanced;
> + u32 fppt_quiet;
> + u32 sppt_quiet;
> + u32 spl_quiet;
> + u32 sppt_apu_only_quiet;
> + /* Fan ID */
> + u32 fan_id_perf;
> + u32 fan_id_balanced;
> + u32 fan_id_quiet;
> +} __packed;
> +
> /* Core Layer */
> int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
> void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev);
> @@ -206,4 +313,12 @@ int apmf_get_static_slider_granular(struct apmf_if *ampf_if,
> void amd_pmf_load_defaults_sps(struct amd_pmf_dev *dev);
>
> int apmf_update_fan_idx(struct apmf_if *ampf_if, bool manual, u32 idx);
> +
> +/* Auto Mode Layer */
> +void amd_pmf_load_defaults_auto_mode(struct amd_pmf_dev *dev);
> +int apmf_get_auto_mode_def(struct apmf_if *ampf_if, struct apmf_auto_mode *data);
> +void amd_pmf_init_auto_mode(struct amd_pmf_dev *dev);
> +void amd_pmf_deinit_auto_mode(struct amd_pmf_dev *dev);
> +void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t time_elapsed_ms);
> +
> #endif /* PMF_H */
Regards,
Hans
^ permalink raw reply [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 08/11] platform/x86/amd/pmf: Add support for Auto mode feature
2022-08-01 12:55 ` Hans de Goede
@ 2022-08-01 13:20 ` Hans de Goede
0 siblings, 0 replies; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 13:20 UTC (permalink / raw)
To: Shyam Sundar S K, markgross; +Cc: platform-driver-x86, Patil.Reddy
Hi,
On 8/1/22 14:55, Hans de Goede wrote:
> Hi,
>
> On 7/28/22 20:20, Shyam Sundar S K wrote:
>> This feature has 3 modes quiet, balanced, performance
>>
>> The objective of this feature is to track the moving average of system
>> power over the time period specified and switch to the subsequent mode.
>>
>> In order to do this, PMF driver will get the moving average of APU power
>> from PMFW and power threshold, time constants, system config parameters
>> from OEM inputs.
>>
>> System power as read by PMF driver from PMFW is the filtered value over
>> the sampling window. Every sampling window, moving average of system power
>> is computed. At the end of the monitoring window, the moving average is
>> compared against the threshold for mode switch for decision making.
>>
>> With AMD managing the system config limits, any mode switch within
>> auto-mode will result in limits of fPPT/sPPT/STAPM or STT being scaled
>> down.
>>
>> When "auto mode" is enabled, the static slider control remains out of
>> the PMF driver, so the platform_profile registration would not
>> happen in PMF driver.
>>
>> The transition to auto-mode only happens when the APMF fn5 is enabled
>> in BIOS, platform_profile set to "balanced" and a AMT
>> (Auto Mode transition) is received.
>>
>> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
>> ---
>> drivers/platform/x86/amd/pmf/Makefile | 3 +-
>> drivers/platform/x86/amd/pmf/acpi.c | 7 +
>> drivers/platform/x86/amd/pmf/auto-mode.c | 317 +++++++++++++++++++++++
>> drivers/platform/x86/amd/pmf/core.c | 9 +
>> drivers/platform/x86/amd/pmf/pmf.h | 115 ++++++++
>> 5 files changed, 450 insertions(+), 1 deletion(-)
>> create mode 100644 drivers/platform/x86/amd/pmf/auto-mode.c
>>
>> diff --git a/drivers/platform/x86/amd/pmf/Makefile b/drivers/platform/x86/amd/pmf/Makefile
>> index 557521a80427..ef2b08c9174d 100644
>> --- a/drivers/platform/x86/amd/pmf/Makefile
>> +++ b/drivers/platform/x86/amd/pmf/Makefile
>> @@ -5,4 +5,5 @@
>> #
>>
>> obj-$(CONFIG_AMD_PMF) += amd-pmf.o
>> -amd-pmf-objs := core.o acpi.o sps.o
>> +amd-pmf-objs := core.o acpi.o sps.o \
>> + auto-mode.o
>> diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
>> index 1814d48f8e44..724150ec58fb 100644
>> --- a/drivers/platform/x86/amd/pmf/acpi.c
>> +++ b/drivers/platform/x86/amd/pmf/acpi.c
>> @@ -99,6 +99,7 @@ static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
>> {
>> func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
>> func->static_slider_granular = mask & APMF_FUNC_STATIC_SLIDER_GRANULAR;
>> + func->auto_mode_def = mask & APMF_FUNC_AUTO_MODE;
>
> I just noticed that this is not correct, this should be:
>
> func->system_params = mask & BIT(APMF_FUNC_... - 1);
>
> Which does something completely different!
>
> Also this information is duplicated by:
>
> is_apmf_func_supported(APMF_FUNC_...) so please drop the apmf_if_functions
> struct completely and use is_apmf_func_supported() everywhere.
>
>> func->fan_table_idx = mask & APMF_FUNC_SET_FAN_IDX;
>> }
>>
>> @@ -160,6 +161,12 @@ int apmf_update_fan_idx(struct apmf_if *ampf_if, bool manual, u32 idx)
>> return err;
>> }
>>
>> +int apmf_get_auto_mode_def(struct apmf_if *apmf_if, struct apmf_auto_mode *data)
>> +{
>> + return apmf_if_call_store_buffer(apmf_if, APMF_FUNC_AUTO_MODE,
>> + data, sizeof(*data));
>> +}
>> +
>> static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
>> {
>> struct apmf_verify_interface output;
>> diff --git a/drivers/platform/x86/amd/pmf/auto-mode.c b/drivers/platform/x86/amd/pmf/auto-mode.c
>> new file mode 100644
>> index 000000000000..954fde25e71e
>> --- /dev/null
>> +++ b/drivers/platform/x86/amd/pmf/auto-mode.c
>> @@ -0,0 +1,317 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +/*
>> + * AMD Platform Management Framework Driver
>> + *
>> + * Copyright (c) 2022, Advanced Micro Devices, Inc.
>> + * All Rights Reserved.
>> + *
>> + * Author: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
>> + */
>> +
>> +#include <linux/acpi.h>
>> +#include <linux/list.h>
>
> Please drop the list.h include, see below.
>
>> +#include <linux/workqueue.h>
>> +#include "pmf.h"
>> +
>> +#define AVG_SAMPLE_SIZE 3
>> +
>
> please replace the code from here.
>
>> +struct power_history {
>> + int socket_power;
>> + struct list_head list;
>> + int avg;
>> + int total;
>> +};
>> +
>> +struct list_pdata {
>> + int total;
>> + int avg;
>> +};
>> +
>> +static struct power_history power_list;
>> +static struct list_pdata pdata;
>
> to here with:
>
> static int socket_power_history[AVG_SAMPLE_SIZE];
> static int socket_power_history_idx = -1;
>
>> +
>> +static struct auto_mode_mode_config config_store;
>> +static const char *state_as_str(unsigned int state);
>> +
>> +static void amd_pmf_handle_automode(struct amd_pmf_dev *dev, bool op, int idx,
>> + struct auto_mode_mode_config *table)
>> +{
>
> all callers use SLIDER_OP_GET for the "op" argument. Please rename this to
> amd_pmf_set_automode(), drop the "op" argument and drop the SLIDER_OP_GET
> handling code.
>
>> + if (op == SLIDER_OP_SET) {
>> + struct power_table_control *pwr_ctrl = &config_store.mode_set[idx].power_control;
>> +
>> + amd_pmf_send_cmd(dev, SET_SPL, false, pwr_ctrl->spl, NULL);
>> + amd_pmf_send_cmd(dev, SET_FPPT, false, pwr_ctrl->fppt, NULL);
>> + amd_pmf_send_cmd(dev, SET_SPPT, false, pwr_ctrl->sppt, NULL);
>> + amd_pmf_send_cmd(dev, SET_SPPT_APU_ONLY, false, pwr_ctrl->sppt_apu_only, NULL);
>> + amd_pmf_send_cmd(dev, SET_STT_MIN_LIMIT, false, pwr_ctrl->stt_min, NULL);
>> + amd_pmf_send_cmd(dev, SET_STT_LIMIT_APU, false,
>> + pwr_ctrl->stt_skin_temp[STT_TEMP_APU], NULL);
>> + amd_pmf_send_cmd(dev, SET_STT_LIMIT_HS2, false,
>> + pwr_ctrl->stt_skin_temp[STT_TEMP_HS2], NULL);
>> + } else if (op == SLIDER_OP_GET) {
>> + amd_pmf_send_cmd(dev, GET_SPL, true, ARG_NONE,
>> + &table->mode_set[idx].power_control.spl);
>> + amd_pmf_send_cmd(dev, GET_FPPT, true, ARG_NONE,
>> + &table->mode_set[idx].power_control.fppt);
>> + amd_pmf_send_cmd(dev, GET_SPPT, true, ARG_NONE,
>> + &table->mode_set[idx].power_control.sppt);
>> + amd_pmf_send_cmd(dev, GET_SPPT_APU_ONLY, true, ARG_NONE,
>> + &table->mode_set[idx].power_control.sppt_apu_only);
>> + amd_pmf_send_cmd(dev, GET_STT_MIN_LIMIT, true, ARG_NONE,
>> + &table->mode_set[idx].power_control.stt_min);
>> + amd_pmf_send_cmd(dev, GET_STT_LIMIT_APU, true, ARG_NONE,
>> + &table->mode_set[idx].power_control.stt_skin_temp[STT_TEMP_APU]);
>> + amd_pmf_send_cmd(dev, GET_STT_LIMIT_HS2, true, ARG_NONE,
>> + &table->mode_set[idx].power_control.stt_skin_temp[STT_TEMP_HS2]);
>> + }
>> +
>> + if (dev->apmf_if->func.fan_table_idx)
>> + apmf_update_fan_idx(dev->apmf_if, config_store.mode_set[idx].fan_control.manual,
>> + config_store.mode_set[idx].fan_control.fan_id);
>> +}
>> +
>> +static int amd_pmf_get_moving_avg(int socket_power)
>> +{
>> + struct power_history *tmp;
>> + struct list_head *pos, *q;
>> + static int count;
>
> Why is this here, but is power_list a global ?
>
> Note that you can initialize a (static) list while declaring it. E.g. :
>
> static LIST_HEAD(power_list);
>
> But as I mention below this whole code can be mode much much simpler.
>
>> +
>> + tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
>> + tmp->socket_power = socket_power;
>> + list_add_tail(&tmp->list, &power_list.list);
>> +
>> + list_for_each_safe(pos, q, &power_list.list) {
>> + if (count >= AVG_SAMPLE_SIZE) {
>> + tmp = list_first_entry(pos, struct power_history, list);
>> + list_del_init(pos);
>> + goto next;
>> + }
>> + }
>> +
>> +next:
>> + pdata.total = 0;
>> + pdata.avg = 0;
>> +
>> + list_for_each(pos, &power_list.list) {
>> + tmp = list_entry(pos, struct power_history, list);
>> + pdata.total += tmp->socket_power;
>> + pdata.avg = pdata.total / AVG_SAMPLE_SIZE;
>> + }
>> +
>> + count++;
>> + if (count >= AVG_SAMPLE_SIZE)
>> + return pdata.avg;
>> +
>> + return 0;
>> +}
>
> Please replace all this list + alloc + free magic which is very hard to review
> and needlessly complicated with this much simpler implementation:
>
> static int amd_pmf_get_moving_avg(int socket_power)
> {
> int i, total = 0;
>
> if (socket_power_history_idx == -1) {
> for (i = 0; i < AVG_SAMPLE_SIZE; i++)
> socket_power_history[i] = socket_power;
> }
>
> socket_power_history_idx = (socket_power_history_idx + 1) % AVG_SAMPLE_SIZE;
> socket_power_history[socket_power_history_idx] = socket_power;
>
> for (i = 0; i < AVG_SAMPLE_SIZE; i++)
> total += socket_power_history[i];
>
> return total / AVG_SAMPLE_SIZE;
> }
A quick follow up, please make the new version of this function take
a "struct amd_pmf_dev *pdev" argument and make the socket_power_history_idx
and socket_power_history[] variables members of amd_pmf_dev instead of making
them global variables.
Regards,
Hans
>
>> +
>> +void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t time_elapsed_ms)
>> +{
>> + int avg_power = 0;
>> + bool update = false;
>> + int i, j;
>> +
>> + /* Get the average moving average computed by auto mode algorithm */
>> + avg_power = amd_pmf_get_moving_avg(socket_power);
>> +
>> + for (i = 0; i < AUTO_TRANSITION_MAX; i++) {
>> + if ((config_store.transition[i].shifting_up && avg_power >=
>> + config_store.transition[i].power_threshold) ||
>> + (!config_store.transition[i].shifting_up && avg_power <=
>> + config_store.transition[i].power_threshold)) {
>> + if (config_store.transition[i].timer <
>> + config_store.transition[i].time_constant)
>> + config_store.transition[i].timer += time_elapsed_ms;
>> + } else {
>> + config_store.transition[i].timer = 0;
>> + }
>> +
>> + if (config_store.transition[i].timer >=
>> + config_store.transition[i].time_constant &&
>> + !config_store.transition[i].applied) {
>> + config_store.transition[i].applied = true;
>> + update = true;
>> + } else if (config_store.transition[i].timer <=
>> + config_store.transition[i].time_constant &&
>> + config_store.transition[i].applied) {
>> + config_store.transition[i].applied = false;
>> + update = true;
>> + }
>> + }
>> +
>> + dev_dbg(dev->dev, "[AUTO_MODE] avg power: %u mW mode: %s\n", avg_power,
>> + state_as_str(config_store.current_mode));
>> +
>> + if (update) {
>> + for (j = 0; j < AUTO_TRANSITION_MAX; j++) {
>> + /* Apply the mode with highest priority indentified */
>> + if (config_store.transition[j].applied) {
>> + if (config_store.current_mode !=
>> + config_store.transition[j].target_mode) {
>> + config_store.current_mode =
>> + config_store.transition[j].target_mode;
>> + dev_dbg(dev->dev, "[AUTO_MODE] moving to mode:%s\n",
>> + state_as_str(config_store.current_mode));
>> + amd_pmf_handle_automode(dev, SLIDER_OP_SET,
>> + config_store.current_mode, NULL);
>> + }
>> + break;
>> + }
>> + }
>> + }
>> +}
>> +
>> +static void amd_pmf_get_power_threshold(void)
>> +{
>> + config_store.transition[AUTO_TRANSITION_TO_QUIET].power_threshold =
>> + config_store.mode_set[AUTO_BALANCE].power_floor -
>> + config_store.transition[AUTO_TRANSITION_TO_QUIET].power_delta;
>> +
>> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].power_threshold =
>> + config_store.mode_set[AUTO_BALANCE].power_floor -
>> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].power_delta;
>> +
>> + config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].power_threshold =
>> + config_store.mode_set[AUTO_QUIET].power_floor -
>> + config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].power_delta;
>> +
>> + config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].power_threshold =
>> + config_store.mode_set[AUTO_PERFORMANCE].power_floor -
>> + config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].power_delta;
>> +}
>> +
>> +static const char *state_as_str(unsigned int state)
>> +{
>> + switch (state) {
>> + case AUTO_QUIET:
>> + return "QUIET";
>> + case AUTO_BALANCE:
>> + return "BALANCED";
>> + case AUTO_PERFORMANCE_ON_LAP:
>> + return "ON_LAP";
>> + case AUTO_PERFORMANCE:
>> + return "PERFORMANCE";
>> + default:
>> + return "Unknown Auto Mode State";
>> + }
>> +}
>> +
>> +void amd_pmf_load_defaults_auto_mode(struct amd_pmf_dev *dev)
>> +{
>> + struct apmf_auto_mode output;
>> + struct power_table_control *pwr_ctrl;
>> + int i;
>> +
>> + if (dev->apmf_if->func.auto_mode_def) {
>
> This is already checked by the "if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE))"
> check inside amd_pmf_init_features(), please drop this check.
>
>> + apmf_get_auto_mode_def(dev->apmf_if, &output);
>> + /* time constant */
>> + config_store.transition[AUTO_TRANSITION_TO_QUIET].time_constant =
>> + output.balanced_to_quiet;
>> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].time_constant =
>> + output.balanced_to_perf;
>> + config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].time_constant =
>> + output.quiet_to_balanced;
>> + config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].time_constant =
>> + output.perf_to_balanced;
>> +
>> + /* power floor */
>> + config_store.mode_set[AUTO_QUIET].power_floor = output.pfloor_quiet;
>> + config_store.mode_set[AUTO_BALANCE].power_floor = output.pfloor_balanced;
>> + config_store.mode_set[AUTO_PERFORMANCE].power_floor = output.pfloor_perf;
>> + config_store.mode_set[AUTO_PERFORMANCE_ON_LAP].power_floor = output.pfloor_perf;
>> +
>> + /* Power delta for mode change */
>> + config_store.transition[AUTO_TRANSITION_TO_QUIET].power_delta =
>> + output.pd_balanced_to_quiet;
>> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].power_delta =
>> + output.pd_balanced_to_perf;
>> + config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].power_delta =
>> + output.pd_quiet_to_balanced;
>> + config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].power_delta =
>> + output.pd_perf_to_balanced;
>> +
>> + /* Power threshold */
>> + amd_pmf_get_power_threshold();
>> +
>> + /* skin temperature limits */
>> + pwr_ctrl = &config_store.mode_set[AUTO_QUIET].power_control;
>> + pwr_ctrl->spl = output.spl_quiet;
>> + pwr_ctrl->sppt = output.sppt_quiet;
>> + pwr_ctrl->fppt = output.fppt_quiet;
>> + pwr_ctrl->sppt_apu_only = output.sppt_apu_only_quiet;
>> + pwr_ctrl->stt_min = output.stt_min_limit_quiet;
>> + pwr_ctrl->stt_skin_temp[STT_TEMP_APU] = output.stt_apu_quiet;
>> + pwr_ctrl->stt_skin_temp[STT_TEMP_HS2] = output.stt_hs2_quiet;
>> +
>> + pwr_ctrl = &config_store.mode_set[AUTO_BALANCE].power_control;
>> + pwr_ctrl->spl = output.spl_balanced;
>> + pwr_ctrl->sppt = output.sppt_balanced;
>> + pwr_ctrl->fppt = output.fppt_balanced;
>> + pwr_ctrl->sppt_apu_only = output.sppt_apu_only_balanced;
>> + pwr_ctrl->stt_min = output.stt_min_limit_balanced;
>> + pwr_ctrl->stt_skin_temp[STT_TEMP_APU] = output.stt_apu_balanced;
>> + pwr_ctrl->stt_skin_temp[STT_TEMP_HS2] = output.stt_hs2_balanced;
>> +
>> + pwr_ctrl = &config_store.mode_set[AUTO_PERFORMANCE].power_control;
>> + pwr_ctrl->spl = output.spl_perf;
>> + pwr_ctrl->sppt = output.sppt_perf;
>> + pwr_ctrl->fppt = output.fppt_perf;
>> + pwr_ctrl->sppt_apu_only = output.sppt_apu_only_perf;
>> + pwr_ctrl->stt_min = output.stt_min_limit_perf;
>> + pwr_ctrl->stt_skin_temp[STT_TEMP_APU] = output.stt_apu_perf;
>> + pwr_ctrl->stt_skin_temp[STT_TEMP_HS2] = output.stt_hs2_perf;
>> +
>> + pwr_ctrl = &config_store.mode_set[AUTO_PERFORMANCE_ON_LAP].power_control;
>> + pwr_ctrl->spl = output.spl_perf_on_lap;
>> + pwr_ctrl->sppt = output.sppt_perf_on_lap;
>> + pwr_ctrl->fppt = output.fppt_perf_on_lap;
>> + pwr_ctrl->sppt_apu_only = output.sppt_apu_only_perf_on_lap;
>> + pwr_ctrl->stt_min = output.stt_min_limit_perf_on_lap;
>> + pwr_ctrl->stt_skin_temp[STT_TEMP_APU] = output.stt_apu_perf_on_lap;
>> + pwr_ctrl->stt_skin_temp[STT_TEMP_HS2] = output.stt_hs2_perf_on_lap;
>> +
>> + /* Fan ID */
>> + config_store.mode_set[AUTO_QUIET].fan_control.fan_id = output.fan_id_quiet;
>> + config_store.mode_set[AUTO_BALANCE].fan_control.fan_id = output.fan_id_balanced;
>> + config_store.mode_set[AUTO_PERFORMANCE].fan_control.fan_id = output.fan_id_perf;
>> + config_store.mode_set[AUTO_PERFORMANCE_ON_LAP].fan_control.fan_id =
>> + output.fan_id_perf;
>> +
>> + config_store.transition[AUTO_TRANSITION_TO_QUIET].target_mode = AUTO_QUIET;
>> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode =
>> + AUTO_PERFORMANCE;
>> + config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].target_mode =
>> + AUTO_BALANCE;
>> + config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].target_mode =
>> + AUTO_BALANCE;
>> +
>> + config_store.transition[AUTO_TRANSITION_TO_QUIET].shifting_up = false;
>> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].shifting_up = true;
>> + config_store.transition[AUTO_TRANSITION_FROM_QUIET_TO_BALANCE].shifting_up = true;
>> + config_store.transition[AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE].shifting_up =
>> + false;
>> +
>> + for (i = 0 ; i < AUTO_MODE_MAX ; i++) {
>> + if (config_store.mode_set[i].fan_control.fan_id == FAN_INDEX_AUTO)
>> + config_store.mode_set[i].fan_control.manual = false;
>> + else
>> + config_store.mode_set[i].fan_control.manual = true;
>> + }
>> + }
>
> This function should also set config_store.current_mode to some sane default (probably balanced).
>
>> +}
>> +
>> +void amd_pmf_deinit_auto_mode(struct amd_pmf_dev *dev)
>> +{
>> + cancel_delayed_work_sync(&dev->work_buffer);
>> +}
>> +
>> +void amd_pmf_init_auto_mode(struct amd_pmf_dev *dev)
>> +{
>> + INIT_LIST_HEAD(&power_list.list);
>> +
>> + amd_pmf_init_metrics_table(dev);
>
> amd_pmf_init_metrics_table(dev) starts the work, so this needs to be done as the last
> step in this function.
>
>> + amd_pmf_load_defaults_auto_mode(dev);
>> +
>> + /* update the thermal for Automode */
>> + amd_pmf_handle_automode(dev, SLIDER_OP_SET, config_store.current_mode, NULL);
>> +}
>> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
>> index ede4eefc33a4..ef71f8326248 100644
>> --- a/drivers/platform/x86/amd/pmf/core.c
>> +++ b/drivers/platform/x86/amd/pmf/core.c
>> @@ -244,12 +244,21 @@ static void amd_pmf_init_features(struct amd_pmf_dev *dev)
>> amd_pmf_init_sps(dev);
>> dev_dbg(dev->dev, "SPS enabled and Platform Profiles registered\n");
>> }
>> +
>> + /* Enable Auto Mode */
>> + if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE)) {
>> + amd_pmf_init_auto_mode(dev);
>> + dev_dbg(dev->dev, "Auto Mode Init done\n");
>> + }
>> }
>>
>> static void amd_pmf_deinit_features(struct amd_pmf_dev *dev)
>> {
>> if (is_apmf_func_supported(APMF_FUNC_STATIC_SLIDER_GRANULAR))
>> amd_pmf_deinit_sps(dev);
>> +
>> + if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE))
>> + amd_pmf_deinit_auto_mode(dev);
>> }
>>
>> static const struct acpi_device_id amd_pmf_acpi_ids[] = {
>> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
>> index 49d3232ee2e0..4618ddc5a662 100644
>> --- a/drivers/platform/x86/amd/pmf/pmf.h
>> +++ b/drivers/platform/x86/amd/pmf/pmf.h
>> @@ -18,6 +18,7 @@
>> #define APMF_FUNC_VERIFY_INTERFACE 0
>> #define APMF_FUNC_GET_SYS_PARAMS 1
>> #define APMF_FUNC_SBIOS_HEARTBEAT 4
>> +#define APMF_FUNC_AUTO_MODE 5
>> #define APMF_FUNC_SET_FAN_IDX 7
>> #define APMF_FUNC_STATIC_SLIDER_GRANULAR 9
>>
>> @@ -49,6 +50,7 @@
>> struct apmf_if_functions {
>> bool system_params;
>> bool sbios_heartbeat;
>> + bool auto_mode_def;
>> bool fan_table_idx;
>> bool static_slider_granular;
>> };
>> @@ -187,6 +189,111 @@ struct fan_table_control {
>> unsigned long fan_id;
>> };
>>
>> +struct power_table_control {
>> + u32 spl;
>> + u32 sppt;
>> + u32 fppt;
>> + u32 sppt_apu_only;
>> + u32 stt_min;
>> + u32 stt_skin_temp[STT_TEMP_COUNT];
>> + u32 reserved[16];
>> +};
>> +
>> +/* Auto Mode Layer */
>> +enum auto_mode_transition_priority {
>> + AUTO_TRANSITION_TO_PERFORMANCE, /* Any other mode to Performance Mode */
>> + AUTO_TRANSITION_FROM_QUIET_TO_BALANCE, /* Quiet Mode to Balance Mode */
>> + AUTO_TRANSITION_TO_QUIET, /* Any other mode to Quiet Mode */
>> + AUTO_TRANSITION_FROM_PERFORMANCE_TO_BALANCE, /* Performance Mode to Balance Mode */
>> + AUTO_TRANSITION_MAX,
>> +};
>> +
>> +enum auto_mode_mode {
>> + AUTO_QUIET,
>> + AUTO_BALANCE,
>> + AUTO_PERFORMANCE_ON_LAP,
>> + AUTO_PERFORMANCE,
>> + AUTO_MODE_MAX,
>> +};
>> +
>> +struct auto_mode_trans_params {
>> + u32 time_constant; /* minimum time required to switch to next mode */
>> + u32 power_delta; /* delta power to shift mode */
>> + u32 power_threshold;
>> + u32 timer; /* elapsed time. if timer > TimeThreshold, it will move to next mode */
>> + u32 applied;
>> + enum auto_mode_mode target_mode;
>> + u32 shifting_up;
>> +};
>> +
>> +struct auto_mode_mode_settings {
>> + struct power_table_control power_control;
>> + struct fan_table_control fan_control;
>> + u32 power_floor;
>> +};
>> +
>> +struct auto_mode_mode_config {
>> + struct auto_mode_trans_params transition[AUTO_TRANSITION_MAX];
>> + struct auto_mode_mode_settings mode_set[AUTO_MODE_MAX];
>> + enum auto_mode_mode current_mode;
>> + bool on_lap;
>> + bool better_perf;
>> + u32 amt_enabled; /* Auto Mode Transition */
>> + u32 avg_power;
>> +};
>> +
>> +struct apmf_auto_mode {
>> + u16 size;
>> + /* time constant */
>> + u32 balanced_to_perf;
>> + u32 perf_to_balanced;
>> + u32 quiet_to_balanced;
>> + u32 balanced_to_quiet;
>> + /* power floor */
>> + u32 pfloor_perf;
>> + u32 pfloor_balanced;
>> + u32 pfloor_quiet;
>> + /* Power delta for mode change */
>> + u32 pd_balanced_to_perf;
>> + u32 pd_perf_to_balanced;
>> + u32 pd_quiet_to_balanced;
>> + u32 pd_balanced_to_quiet;
>> + /* skin temperature limits */
>> + u8 stt_apu_perf_on_lap; /* CQL ON */
>> + u8 stt_hs2_perf_on_lap; /* CQL ON */
>> + u8 stt_apu_perf;
>> + u8 stt_hs2_perf;
>> + u8 stt_apu_balanced;
>> + u8 stt_hs2_balanced;
>> + u8 stt_apu_quiet;
>> + u8 stt_hs2_quiet;
>> + u32 stt_min_limit_perf_on_lap; /* CQL ON */
>> + u32 stt_min_limit_perf;
>> + u32 stt_min_limit_balanced;
>> + u32 stt_min_limit_quiet;
>> + /* SPL based */
>> + u32 fppt_perf_on_lap; /* CQL ON */
>> + u32 sppt_perf_on_lap; /* CQL ON */
>> + u32 spl_perf_on_lap; /* CQL ON */
>> + u32 sppt_apu_only_perf_on_lap; /* CQL ON */
>> + u32 fppt_perf;
>> + u32 sppt_perf;
>> + u32 spl_perf;
>> + u32 sppt_apu_only_perf;
>> + u32 fppt_balanced;
>> + u32 sppt_balanced;
>> + u32 spl_balanced;
>> + u32 sppt_apu_only_balanced;
>> + u32 fppt_quiet;
>> + u32 sppt_quiet;
>> + u32 spl_quiet;
>> + u32 sppt_apu_only_quiet;
>> + /* Fan ID */
>> + u32 fan_id_perf;
>> + u32 fan_id_balanced;
>> + u32 fan_id_quiet;
>> +} __packed;
>> +
>> /* Core Layer */
>> int apmf_acpi_init(struct amd_pmf_dev *pmf_dev);
>> void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev);
>> @@ -206,4 +313,12 @@ int apmf_get_static_slider_granular(struct apmf_if *ampf_if,
>> void amd_pmf_load_defaults_sps(struct amd_pmf_dev *dev);
>>
>> int apmf_update_fan_idx(struct apmf_if *ampf_if, bool manual, u32 idx);
>> +
>> +/* Auto Mode Layer */
>> +void amd_pmf_load_defaults_auto_mode(struct amd_pmf_dev *dev);
>> +int apmf_get_auto_mode_def(struct apmf_if *ampf_if, struct apmf_auto_mode *data);
>> +void amd_pmf_init_auto_mode(struct amd_pmf_dev *dev);
>> +void amd_pmf_deinit_auto_mode(struct amd_pmf_dev *dev);
>> +void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t time_elapsed_ms);
>> +
>> #endif /* PMF_H */
>
> Regards,
>
> Hans
^ permalink raw reply [flat|nested] 33+ messages in thread
* [PATCH v2 RESEND 09/11] platform/x86/amd/pmf: Handle AMT and CQL events for Auto mode
2022-07-28 18:20 [PATCH v2 RESEND 00/11] platform/x86/amd/pmf: Introduce AMD PMF Driver Shyam Sundar S K
` (7 preceding siblings ...)
2022-07-28 18:20 ` [PATCH v2 RESEND 08/11] platform/x86/amd/pmf: Add support for Auto mode feature Shyam Sundar S K
@ 2022-07-28 18:20 ` Shyam Sundar S K
2022-08-01 13:55 ` Hans de Goede
2022-08-01 14:01 ` Hans de Goede
2022-07-28 18:20 ` [PATCH v2 RESEND 10/11] platform/x86/amd/pmf: Force load driver on older supported platforms Shyam Sundar S K
2022-07-28 18:20 ` [PATCH v2 RESEND 11/11] MAINTAINERS: Add AMD PMF driver entry Shyam Sundar S K
10 siblings, 2 replies; 33+ messages in thread
From: Shyam Sundar S K @ 2022-07-28 18:20 UTC (permalink / raw)
To: hdegoede, markgross
Cc: platform-driver-x86, Patil.Reddy, Shyam Sundar S K,
Mario Limonciello, Mark Pearson
The transition to auto-mode happens when the PMF driver receives
AMT (Auto Mode transition) event. transition logic will reside in the
PMF driver but the events would come from other supported drivers[1].
The thermal parameters would vary between when a performance "on-lap" mode
is detected and versus when not. The CQL event would get triggered from
other drivers, so that PMF driver would adjust the system thermal config
based on the ACPI inputs.
OEMs can control whether or not to enable AMT or CQL via other supported
drivers[1] but the actual transition logic resides in the AMD PMF driver.
When an AMT event is received the automatic mode transition RAPL algorithm
will run. When a CQL event is received an performance "on-lap" mode will
be enabled and thermal parameters will be adjusted accordingly.
[1]
Link: https://git.kernel.org/pub/scm/linux/kernel/git/pdx86/platform-drivers-x86.git/commit/?h=review-hans&id=755b249250df1b612d982f3b702c831b26ecdf73
Cc: Mario Limonciello <mario.limonciello@amd.com>
Cc: Mark Pearson <markpearson@lenovo.com>
Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
---
drivers/platform/x86/amd/pmf/acpi.c | 59 ++++++++++++++++++++++++
drivers/platform/x86/amd/pmf/auto-mode.c | 22 +++++++++
drivers/platform/x86/amd/pmf/core.c | 32 +++++++++++++
drivers/platform/x86/amd/pmf/pmf.h | 22 +++++++++
4 files changed, 135 insertions(+)
diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
index 724150ec58fb..dedaebf18d88 100644
--- a/drivers/platform/x86/amd/pmf/acpi.c
+++ b/drivers/platform/x86/amd/pmf/acpi.c
@@ -12,6 +12,8 @@
#include "pmf.h"
#define APMF_METHOD "\\_SB_.PMF_.APMF"
+#define APMF_CQL_NOTIFICATION 2
+#define APMF_AMT_NOTIFICATION 3
static unsigned long supported_func;
@@ -99,6 +101,7 @@ static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
{
func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
func->static_slider_granular = mask & APMF_FUNC_STATIC_SLIDER_GRANULAR;
+ func->sbios_requests = mask & APMF_FUNC_SBIOS_REQUESTS;
func->auto_mode_def = mask & APMF_FUNC_AUTO_MODE;
func->fan_table_idx = mask & APMF_FUNC_SET_FAN_IDX;
}
@@ -167,6 +170,44 @@ int apmf_get_auto_mode_def(struct apmf_if *apmf_if, struct apmf_auto_mode *data)
data, sizeof(*data));
}
+int apmf_get_sbios_requests(struct apmf_if *apmf_if, struct apmf_sbios_req *req)
+{
+ return apmf_if_call_store_buffer(apmf_if, APMF_FUNC_SBIOS_REQUESTS,
+ req, sizeof(*req));
+}
+
+static void apmf_event_handler(acpi_handle handle, u32 event, void *data)
+{
+ struct amd_pmf_dev *pmf_dev = data;
+ struct apmf_if *apmf_if = pmf_dev->apmf_if;
+ int ret;
+
+ if (apmf_if->func.sbios_requests) {
+ struct apmf_sbios_req req;
+
+ ret = apmf_get_sbios_requests(apmf_if, &req);
+ if (ret) {
+ dev_err(pmf_dev->dev, "Failed to get SBIOS requests:%d\n", ret);
+ return;
+ }
+
+ if (req.pending_req & BIT(APMF_AMT_NOTIFICATION)) {
+ pr_debug("PMF: AMT is supported and notifications %s\n",
+ req.amt_event ? "Enabled" : "Disabled");
+ pmf_dev->is_amt_event = !!req.amt_event;
+ }
+
+ if (req.pending_req & BIT(APMF_CQL_NOTIFICATION)) {
+ pr_debug("PMF: CQL is supported and notifications %s\n",
+ req.cql_event ? "Enabled" : "Disabled");
+ pmf_dev->is_cql_event = !!req.cql_event;
+
+ /* update the target mode information */
+ amd_pmf_update_2_cql(pmf_dev);
+ }
+ }
+}
+
static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
{
struct apmf_verify_interface output;
@@ -211,12 +252,19 @@ static int apmf_get_system_params(struct apmf_if *apmf_if)
void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev)
{
+ acpi_handle ahandle = ACPI_HANDLE(pmf_dev->dev);
+
if (pmf_dev->apmf_if->func.sbios_heartbeat)
cancel_delayed_work_sync(&pmf_dev->heart_beat);
+
+ if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE))
+ acpi_remove_notify_handler(ahandle, ACPI_ALL_NOTIFY,
+ apmf_event_handler);
}
int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
{
+ acpi_handle ahandle = ACPI_HANDLE(pmf_dev->dev);
struct apmf_notification_cfg *n;
acpi_handle apmf_if_handle;
struct apmf_if *apmf_if;
@@ -256,6 +304,17 @@ int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
schedule_delayed_work(&pmf_dev->heart_beat, 0);
}
+ /* Install the APMF Notify handler */
+ if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE)) {
+ status = acpi_install_notify_handler(ahandle,
+ ACPI_ALL_NOTIFY,
+ apmf_event_handler, pmf_dev);
+ if (ACPI_FAILURE(status)) {
+ dev_err(pmf_dev->dev, "failed to install notify handler\n");
+ return -ENODEV;
+ }
+ }
+
out:
return ret;
}
diff --git a/drivers/platform/x86/amd/pmf/auto-mode.c b/drivers/platform/x86/amd/pmf/auto-mode.c
index 954fde25e71e..a85ef4b95cdb 100644
--- a/drivers/platform/x86/amd/pmf/auto-mode.c
+++ b/drivers/platform/x86/amd/pmf/auto-mode.c
@@ -111,6 +111,13 @@ void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t t
bool update = false;
int i, j;
+ if (!dev->amt_running && dev->is_amt_event) {
+ dev_dbg(dev->dev, "setting AMT thermals\n");
+ amd_pmf_handle_automode(dev, SLIDER_OP_SET, config_store.current_mode, NULL);
+ dev->amt_running = true;
+ return;
+ }
+
/* Get the average moving average computed by auto mode algorithm */
avg_power = amd_pmf_get_moving_avg(socket_power);
@@ -161,6 +168,21 @@ void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t t
}
}
+void amd_pmf_update_2_cql(struct amd_pmf_dev *dev)
+{
+ config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode =
+ dev->is_cql_event ? AUTO_PERFORMANCE_ON_LAP : AUTO_PERFORMANCE;
+ if ((config_store.current_mode == AUTO_PERFORMANCE ||
+ config_store.current_mode == AUTO_PERFORMANCE_ON_LAP) &&
+ config_store.current_mode !=
+ config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode) {
+ config_store.current_mode =
+ config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode;
+ amd_pmf_handle_automode(dev, SLIDER_OP_SET, config_store.current_mode, NULL);
+ }
+ dev_dbg(dev->dev, "updated CQL thermals\n");
+}
+
static void amd_pmf_get_power_threshold(void)
{
config_store.transition[AUTO_TRANSITION_TO_QUIET].power_threshold =
diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
index ef71f8326248..bfae779ccc23 100644
--- a/drivers/platform/x86/amd/pmf/core.c
+++ b/drivers/platform/x86/amd/pmf/core.c
@@ -93,6 +93,27 @@ int amd_pmf_get_power_source(void)
return POWER_SOURCE_DC;
}
+static void amd_pmf_reset_amt(struct amd_pmf_dev *dev)
+{
+ pr_debug("resetting AMT thermals\n");
+ dev->amt_running = false;
+
+ /*
+ * OEM BIOS implementation guide says that if the auto mode is enabled
+ * the platform_profile registration shall be done by the OEM driver.
+ * There could be cases where both static slider and auto mode BIOS
+ * functions are enabled and we could end up in a race. To avoid that
+ * add a protection and touch the static slider only if that's enabled
+ * from the BIOS side.
+ */
+
+ if (is_apmf_func_supported(APMF_FUNC_STATIC_SLIDER_GRANULAR)) {
+ u8 mode = amd_pmf_get_pprof_modes(dev);
+
+ amd_pmf_update_slider(dev, SLIDER_OP_SET, mode, NULL);
+ }
+}
+
static void amd_pmf_get_metrics(struct work_struct *work)
{
struct amd_pmf_dev *dev = container_of(work, struct amd_pmf_dev, work_buffer.work);
@@ -103,6 +124,9 @@ static void amd_pmf_get_metrics(struct work_struct *work)
/* Get the current profile information */
current_profile = READ_ONCE(dev->current_profile);
+ if (!dev->is_amt_event)
+ dev_dbg(dev->dev, "amt event: %s\n", dev->is_amt_event ? "Enabled" : "Disabled");
+
/* Transfer table contents */
memset(&dev->m_table, 0, sizeof(dev->m_table));
amd_pmf_send_cmd(dev, SET_TRANSFER_TABLE, 0, 7, NULL);
@@ -112,6 +136,14 @@ static void amd_pmf_get_metrics(struct work_struct *work)
/* Calculate the avg SoC power consumption */
socket_power = dev->m_table.apu_power + dev->m_table.dgpu_power;
+ if (dev->is_amt_event) {
+ /* Apply the Auto Mode transition */
+ amd_pmf_trans_automode(dev, socket_power, time_elapsed_ms);
+ } else if (!dev->is_amt_event && dev->amt_running) {
+ /* reset to other mode if we receive a AMT disable event */
+ amd_pmf_reset_amt(dev);
+ }
+
dev->start_time = ktime_to_ms(ktime_get());
schedule_delayed_work(&dev->work_buffer, msecs_to_jiffies(metrics_table_loop_ms));
}
diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
index 4618ddc5a662..0f8b25524845 100644
--- a/drivers/platform/x86/amd/pmf/pmf.h
+++ b/drivers/platform/x86/amd/pmf/pmf.h
@@ -17,6 +17,7 @@
/* APMF Functions */
#define APMF_FUNC_VERIFY_INTERFACE 0
#define APMF_FUNC_GET_SYS_PARAMS 1
+#define APMF_FUNC_SBIOS_REQUESTS 2
#define APMF_FUNC_SBIOS_HEARTBEAT 4
#define APMF_FUNC_AUTO_MODE 5
#define APMF_FUNC_SET_FAN_IDX 7
@@ -49,6 +50,7 @@
/* AMD PMF BIOS interfaces */
struct apmf_if_functions {
bool system_params;
+ bool sbios_requests;
bool sbios_heartbeat;
bool auto_mode_def;
bool fan_table_idx;
@@ -80,6 +82,21 @@ struct apmf_system_params {
u32 heartbeat_int;
} __packed;
+struct apmf_sbios_req {
+ u16 size;
+ u32 pending_req;
+ u8 rsd;
+ u8 cql_event;
+ u8 amt_event;
+ u32 fppt;
+ u32 sppt;
+ u32 fppt_apu_only;
+ u32 spl;
+ u32 stt_min_limit;
+ u8 skin_temp_apu;
+ u8 skin_temp_hs2;
+} __packed;
+
struct apmf_fan_idx {
u16 size;
u8 fan_ctl_mode;
@@ -161,6 +178,9 @@ struct amd_pmf_dev {
struct smu_pmf_metrics m_table;
struct delayed_work work_buffer;
ktime_t start_time;
+ bool is_amt_event;
+ bool is_cql_event;
+ bool amt_running;
};
struct apmf_sps_prop_granular {
@@ -320,5 +340,7 @@ int apmf_get_auto_mode_def(struct apmf_if *ampf_if, struct apmf_auto_mode *data)
void amd_pmf_init_auto_mode(struct amd_pmf_dev *dev);
void amd_pmf_deinit_auto_mode(struct amd_pmf_dev *dev);
void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t time_elapsed_ms);
+int apmf_get_sbios_requests(struct apmf_if *ampf_if, struct apmf_sbios_req *req);
+void amd_pmf_update_2_cql(struct amd_pmf_dev *dev);
#endif /* PMF_H */
--
2.25.1
^ permalink raw reply related [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 09/11] platform/x86/amd/pmf: Handle AMT and CQL events for Auto mode
2022-07-28 18:20 ` [PATCH v2 RESEND 09/11] platform/x86/amd/pmf: Handle AMT and CQL events for Auto mode Shyam Sundar S K
@ 2022-08-01 13:55 ` Hans de Goede
2022-08-02 11:22 ` Shyam Sundar S K
2022-08-01 14:01 ` Hans de Goede
1 sibling, 1 reply; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 13:55 UTC (permalink / raw)
To: Shyam Sundar S K, markgross
Cc: platform-driver-x86, Patil.Reddy, Mario Limonciello, Mark Pearson
Hi,
On 7/28/22 20:20, Shyam Sundar S K wrote:
> The transition to auto-mode happens when the PMF driver receives
> AMT (Auto Mode transition) event. transition logic will reside in the
> PMF driver but the events would come from other supported drivers[1].
>
> The thermal parameters would vary between when a performance "on-lap" mode
> is detected and versus when not. The CQL event would get triggered from
> other drivers, so that PMF driver would adjust the system thermal config
> based on the ACPI inputs.
>
> OEMs can control whether or not to enable AMT or CQL via other supported
> drivers[1] but the actual transition logic resides in the AMD PMF driver.
> When an AMT event is received the automatic mode transition RAPL algorithm
> will run. When a CQL event is received an performance "on-lap" mode will
> be enabled and thermal parameters will be adjusted accordingly.
>
> [1]
> Link: https://git.kernel.org/pub/scm/linux/kernel/git/pdx86/platform-drivers-x86.git/commit/?h=review-hans&id=755b249250df1b612d982f3b702c831b26ecdf73
>
> Cc: Mario Limonciello <mario.limonciello@amd.com>
> Cc: Mark Pearson <markpearson@lenovo.com>
> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> ---
> drivers/platform/x86/amd/pmf/acpi.c | 59 ++++++++++++++++++++++++
> drivers/platform/x86/amd/pmf/auto-mode.c | 22 +++++++++
> drivers/platform/x86/amd/pmf/core.c | 32 +++++++++++++
> drivers/platform/x86/amd/pmf/pmf.h | 22 +++++++++
> 4 files changed, 135 insertions(+)
>
> diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
> index 724150ec58fb..dedaebf18d88 100644
> --- a/drivers/platform/x86/amd/pmf/acpi.c
> +++ b/drivers/platform/x86/amd/pmf/acpi.c
> @@ -12,6 +12,8 @@
> #include "pmf.h"
>
> #define APMF_METHOD "\\_SB_.PMF_.APMF"
> +#define APMF_CQL_NOTIFICATION 2
> +#define APMF_AMT_NOTIFICATION 3
>
> static unsigned long supported_func;
>
> @@ -99,6 +101,7 @@ static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
> {
> func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
> func->static_slider_granular = mask & APMF_FUNC_STATIC_SLIDER_GRANULAR;
> + func->sbios_requests = mask & APMF_FUNC_SBIOS_REQUESTS;
I just noticed that this is not correct, this should be:
func->system_params = mask & BIT(APMF_FUNC_... - 1);
Which does something completely different!
Also this information is duplicated by:
is_apmf_func_supported(APMF_FUNC_...) so please drop the apmf_if_functions
struct completely and use is_apmf_func_supported() everywhere.
> func->auto_mode_def = mask & APMF_FUNC_AUTO_MODE;
> func->fan_table_idx = mask & APMF_FUNC_SET_FAN_IDX;
> }
> @@ -167,6 +170,44 @@ int apmf_get_auto_mode_def(struct apmf_if *apmf_if, struct apmf_auto_mode *data)
> data, sizeof(*data));
> }
>
> +int apmf_get_sbios_requests(struct apmf_if *apmf_if, struct apmf_sbios_req *req)
> +{
> + return apmf_if_call_store_buffer(apmf_if, APMF_FUNC_SBIOS_REQUESTS,
> + req, sizeof(*req));
> +}
> +
> +static void apmf_event_handler(acpi_handle handle, u32 event, void *data)
> +{
> + struct amd_pmf_dev *pmf_dev = data;
> + struct apmf_if *apmf_if = pmf_dev->apmf_if;
> + int ret;
> +
> + if (apmf_if->func.sbios_requests) {
IMHO it would be better to change the test for [un]registering the handler to:
if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE) &&
is_apmf_func_supported(APMF_FUNC_SBIOS_REQUESTS))
and then the sbios_requests test here (in apmf_event_handler()) can be dropped.
> + struct apmf_sbios_req req;
> +
> + ret = apmf_get_sbios_requests(apmf_if, &req);
> + if (ret) {
> + dev_err(pmf_dev->dev, "Failed to get SBIOS requests:%d\n", ret);
> + return;
> + }
> +
> + if (req.pending_req & BIT(APMF_AMT_NOTIFICATION)) {
> + pr_debug("PMF: AMT is supported and notifications %s\n",
> + req.amt_event ? "Enabled" : "Disabled");
> + pmf_dev->is_amt_event = !!req.amt_event;
The is_amt_event variable is not storing if the last event was an amt_event but
rather wether AMT is enabled or not, so please rename it to amt_enabled.
Also by just setting the flag here you are introducing a race, at least on
AMT disabling.
1: amd_pmf_get_metrics() is running, has just passed the
"if (dev->is_amt_event)" check and is about to call amd_pmf_trans_automode().
2. apmf_event_handler() gets called to disable AMT
3. apmf_event_handler() finishes and the firmware now starts programming the
limits because thinkpad_acpi's platform_profile has been set to !balanced.
4. amd_pmf_get_metrics() continues and calls amd_pmf_trans_automode()
5. amd_pmf_trans_automode() reprograms the limits, overriding the ones
set by the firmware eventhough AMT is disabled at this point.
To avoid this race apmf_event_handler() and amd_pmf_get_metrics() should both
lock a newly added shared update_mutex mutex while running so that they cannot
both run at the same time.
###
Likewise restoring the static slider settings on AMT disable is also racy
and to fix that the amd_pmf_reset_amt(dev);should be moved here; and then for
consistency and cleanness the setting of the initial AMT values should also
be moved here.
This will result in adding the following code here:
dev_dbg(dev->dev, "amt enabled: %d\n", dev->amt_enabled);
if (pmf_dev->amt_enabled)
amd_pmf_handle_automode(dev, SLIDER_OP_SET, config_store.current_mode, NULL);
else
amd_pmf_reset_amt(dev);
Also be handling the flanks (enabled<->disabled transitions) here like this there
no longer is a need for a separate amt_running variable to detect the flanks in other
places.
> + }
> +
> + if (req.pending_req & BIT(APMF_CQL_NOTIFICATION)) {
> + pr_debug("PMF: CQL is supported and notifications %s\n",
> + req.cql_event ? "Enabled" : "Disabled");
> + pmf_dev->is_cql_event = !!req.cql_event;
is_cql_event is only used to pass req.cql_event to amd_pmf_update_2_cql()
please drop it from struct amd_pmf_dev and make it an argument to
amd_pmf_update_2_cql()
> +
> + /* update the target mode information */
> + amd_pmf_update_2_cql(pmf_dev);
> + }
> + }
> +}
> +
> static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
> {
> struct apmf_verify_interface output;
> @@ -211,12 +252,19 @@ static int apmf_get_system_params(struct apmf_if *apmf_if)
>
> void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev)
> {
> + acpi_handle ahandle = ACPI_HANDLE(pmf_dev->dev);
> +
Hmm, is this the handle for: "\\_SB_.PMF" ? as in the parent of:
#define APMF_METHOD "\\_SB_.PMF_.APMF"
?
In that case then apmf_if_call() can use:
acpi_handle ahandle = ACPI_HANDLE(pmf_dev->dev);
...
status = acpi_evaluate_object(ahandle, "APMF", &apmf_if_arg_list, &buffer);
and the whole:
+ status = acpi_get_handle(NULL, (acpi_string) APMF_METHOD, &apmf_if_handle);
+ if (ACPI_FAILURE(status))
+ return -ENODEV;
+
bit + storing the handle + also:
+#define APMF_METHOD "\\_SB_.PMF_.APMF"
can then all be dropped. That would be a nice cleanup and then calling
dev_err(...) for the ACPI errors also makes even more sense
> if (pmf_dev->apmf_if->func.sbios_heartbeat)
> cancel_delayed_work_sync(&pmf_dev->heart_beat);
> +
> + if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE))
As mentioned above IMHO it would make sense to make this check:
if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE) &&
is_apmf_func_supported(APMF_FUNC_SBIOS_REQUESTS))
> + acpi_remove_notify_handler(ahandle, ACPI_ALL_NOTIFY,
> + apmf_event_handler);
> }
>
> int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
> {
> + acpi_handle ahandle = ACPI_HANDLE(pmf_dev->dev);
> struct apmf_notification_cfg *n;
> acpi_handle apmf_if_handle;
> struct apmf_if *apmf_if;
> @@ -256,6 +304,17 @@ int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
> schedule_delayed_work(&pmf_dev->heart_beat, 0);
> }
>
> + /* Install the APMF Notify handler */
> + if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE)) {
As mentioned above IMHO it would make sense to make this check:
if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE) &&
is_apmf_func_supported(APMF_FUNC_SBIOS_REQUESTS))
What is missing here is detecting the initial AMT and CQL values and configuring
things accordingly. Maybe this can be done by calling apmf_event_handler
manually once before registering it ?
Note the initial state detection must be done before registering the handler
to avoid races.
> + status = acpi_install_notify_handler(ahandle,
> + ACPI_ALL_NOTIFY,
> + apmf_event_handler, pmf_dev);
> + if (ACPI_FAILURE(status)) {
> + dev_err(pmf_dev->dev, "failed to install notify handler\n");
> + return -ENODEV;
> + }
> + }
> +
> out:
> return ret;
> }
> diff --git a/drivers/platform/x86/amd/pmf/auto-mode.c b/drivers/platform/x86/amd/pmf/auto-mode.c
> index 954fde25e71e..a85ef4b95cdb 100644
> --- a/drivers/platform/x86/amd/pmf/auto-mode.c
> +++ b/drivers/platform/x86/amd/pmf/auto-mode.c
> @@ -111,6 +111,13 @@ void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t t
> bool update = false;
> int i, j;
>
> + if (!dev->amt_running && dev->is_amt_event) {
> + dev_dbg(dev->dev, "setting AMT thermals\n");
> + amd_pmf_handle_automode(dev, SLIDER_OP_SET, config_store.current_mode, NULL);
> + dev->amt_running = true;
> + return;
> + }
> +
As suggested above this is best done on receiving the amt_event.
> /* Get the average moving average computed by auto mode algorithm */
> avg_power = amd_pmf_get_moving_avg(socket_power);
>
> @@ -161,6 +168,21 @@ void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t t
> }
> }
>
> +void amd_pmf_update_2_cql(struct amd_pmf_dev *dev)
> +{
> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode =
> + dev->is_cql_event ? AUTO_PERFORMANCE_ON_LAP : AUTO_PERFORMANCE;
> + if ((config_store.current_mode == AUTO_PERFORMANCE ||
> + config_store.current_mode == AUTO_PERFORMANCE_ON_LAP) &&
> + config_store.current_mode !=
> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode) {
> + config_store.current_mode =
> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode;
> + amd_pmf_handle_automode(dev, SLIDER_OP_SET, config_store.current_mode, NULL);
> + }
> + dev_dbg(dev->dev, "updated CQL thermals\n");
> +}
> +
> static void amd_pmf_get_power_threshold(void)
> {
> config_store.transition[AUTO_TRANSITION_TO_QUIET].power_threshold =
> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
> index ef71f8326248..bfae779ccc23 100644
> --- a/drivers/platform/x86/amd/pmf/core.c
> +++ b/drivers/platform/x86/amd/pmf/core.c
> @@ -93,6 +93,27 @@ int amd_pmf_get_power_source(void)
> return POWER_SOURCE_DC;
> }
>
> +static void amd_pmf_reset_amt(struct amd_pmf_dev *dev)
> +{
> + pr_debug("resetting AMT thermals\n");
> + dev->amt_running = false;
> +
> + /*
> + * OEM BIOS implementation guide says that if the auto mode is enabled
> + * the platform_profile registration shall be done by the OEM driver.
> + * There could be cases where both static slider and auto mode BIOS
> + * functions are enabled and we could end up in a race. To avoid that
> + * add a protection and touch the static slider only if that's enabled
> + * from the BIOS side.
> + */
This comment seems odd. The is_apmf_func_supported() does not help to
avoid the race. This check is necessary to avoid amd_pmf_update_slider()
on systems where amd_pmf_load_defaults_sps() does not run. Otherwise
amd_pmf_update_slider() ends up writing uninitialized (all 0) limits.
To fix the race the amd_pmf_update_slider() call must be done before
apmf_event_handler() exits as suggested above. The feature check has
nothing to do with the race ...
> +
> + if (is_apmf_func_supported(APMF_FUNC_STATIC_SLIDER_GRANULAR)) {
> + u8 mode = amd_pmf_get_pprof_modes(dev);
> +
> + amd_pmf_update_slider(dev, SLIDER_OP_SET, mode, NULL);
> + }
> +}
> +
> static void amd_pmf_get_metrics(struct work_struct *work)
> {
> struct amd_pmf_dev *dev = container_of(work, struct amd_pmf_dev, work_buffer.work);
> @@ -103,6 +124,9 @@ static void amd_pmf_get_metrics(struct work_struct *work)
> /* Get the current profile information */
> current_profile = READ_ONCE(dev->current_profile);
>
> + if (!dev->is_amt_event)
> + dev_dbg(dev->dev, "amt event: %s\n", dev->is_amt_event ? "Enabled" : "Disabled");
> +
This is weird, you check for !is_amt_event so this will only ever log:
"amt event: Disabled" so the check in the dev_dbg is not necessary. Also this is
best done when receiving the amt-event, as suggested above so this can just be dropped.
> /* Transfer table contents */
> memset(&dev->m_table, 0, sizeof(dev->m_table));
> amd_pmf_send_cmd(dev, SET_TRANSFER_TABLE, 0, 7, NULL);
> @@ -112,6 +136,14 @@ static void amd_pmf_get_metrics(struct work_struct *work)
> /* Calculate the avg SoC power consumption */
> socket_power = dev->m_table.apu_power + dev->m_table.dgpu_power;
>
> + if (dev->is_amt_event) {
> + /* Apply the Auto Mode transition */
> + amd_pmf_trans_automode(dev, socket_power, time_elapsed_ms);
> + } else if (!dev->is_amt_event && dev->amt_running) {
> + /* reset to other mode if we receive a AMT disable event */
> + amd_pmf_reset_amt(dev);
> + }
> +
As discussed above the amd_pmf_reset_amt() call must be done from inside
apmf_event_handler(). So the whole "else if (...) {}" block can be dropped here.
> dev->start_time = ktime_to_ms(ktime_get());
> schedule_delayed_work(&dev->work_buffer, msecs_to_jiffies(metrics_table_loop_ms));
> }
> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
> index 4618ddc5a662..0f8b25524845 100644
> --- a/drivers/platform/x86/amd/pmf/pmf.h
> +++ b/drivers/platform/x86/amd/pmf/pmf.h
> @@ -17,6 +17,7 @@
> /* APMF Functions */
> #define APMF_FUNC_VERIFY_INTERFACE 0
> #define APMF_FUNC_GET_SYS_PARAMS 1
> +#define APMF_FUNC_SBIOS_REQUESTS 2
> #define APMF_FUNC_SBIOS_HEARTBEAT 4
> #define APMF_FUNC_AUTO_MODE 5
> #define APMF_FUNC_SET_FAN_IDX 7
> @@ -49,6 +50,7 @@
> /* AMD PMF BIOS interfaces */
> struct apmf_if_functions {
> bool system_params;
> + bool sbios_requests;
> bool sbios_heartbeat;
> bool auto_mode_def;
> bool fan_table_idx;
> @@ -80,6 +82,21 @@ struct apmf_system_params {
> u32 heartbeat_int;
> } __packed;
>
> +struct apmf_sbios_req {
> + u16 size;
> + u32 pending_req;
> + u8 rsd;
> + u8 cql_event;
> + u8 amt_event;
> + u32 fppt;
> + u32 sppt;
> + u32 fppt_apu_only;
> + u32 spl;
> + u32 stt_min_limit;
> + u8 skin_temp_apu;
> + u8 skin_temp_hs2;
> +} __packed;
> +
> struct apmf_fan_idx {
> u16 size;
> u8 fan_ctl_mode;
> @@ -161,6 +178,9 @@ struct amd_pmf_dev {
> struct smu_pmf_metrics m_table;
> struct delayed_work work_buffer;
> ktime_t start_time;
> + bool is_amt_event;
The is_amt_event variable is not storing if the last event was an amt_event but
rather wether AMT is enabled or not, so please rename it to amt_enabled.
> + bool is_cql_event;
> + bool amt_running;
As discussed both these struct members can be dropped.
> };
>
> struct apmf_sps_prop_granular {
> @@ -320,5 +340,7 @@ int apmf_get_auto_mode_def(struct apmf_if *ampf_if, struct apmf_auto_mode *data)
> void amd_pmf_init_auto_mode(struct amd_pmf_dev *dev);
> void amd_pmf_deinit_auto_mode(struct amd_pmf_dev *dev);
> void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t time_elapsed_ms);
> +int apmf_get_sbios_requests(struct apmf_if *ampf_if, struct apmf_sbios_req *req);
> +void amd_pmf_update_2_cql(struct amd_pmf_dev *dev);
>
> #endif /* PMF_H */
Regards,
Hans
^ permalink raw reply [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 09/11] platform/x86/amd/pmf: Handle AMT and CQL events for Auto mode
2022-08-01 13:55 ` Hans de Goede
@ 2022-08-02 11:22 ` Shyam Sundar S K
2022-08-02 12:20 ` Hans de Goede
0 siblings, 1 reply; 33+ messages in thread
From: Shyam Sundar S K @ 2022-08-02 11:22 UTC (permalink / raw)
To: Hans de Goede, markgross
Cc: platform-driver-x86, Patil.Reddy, Mario Limonciello, Mark Pearson
Hi Hans,
On 8/1/2022 7:25 PM, Hans de Goede wrote:
> Hi,
>
> On 7/28/22 20:20, Shyam Sundar S K wrote:
>> The transition to auto-mode happens when the PMF driver receives
>> AMT (Auto Mode transition) event. transition logic will reside in the
>> PMF driver but the events would come from other supported drivers[1].
>>
>> The thermal parameters would vary between when a performance "on-lap" mode
>> is detected and versus when not. The CQL event would get triggered from
>> other drivers, so that PMF driver would adjust the system thermal config
>> based on the ACPI inputs.
>>
>> OEMs can control whether or not to enable AMT or CQL via other supported
>> drivers[1] but the actual transition logic resides in the AMD PMF driver.
>> When an AMT event is received the automatic mode transition RAPL algorithm
>> will run. When a CQL event is received an performance "on-lap" mode will
>> be enabled and thermal parameters will be adjusted accordingly.
>>
>> [1]
>> Link: https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgit.kernel.org%2Fpub%2Fscm%2Flinux%2Fkernel%2Fgit%2Fpdx86%2Fplatform-drivers-x86.git%2Fcommit%2F%3Fh%3Dreview-hans%26id%3D755b249250df1b612d982f3b702c831b26ecdf73&data=05%7C01%7Cshyam-sundar.s-k%40amd.com%7C1297e4cd71784092e05008da73c59095%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637949589653252953%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=5QDwDnnn45ZCwq2MfPQjv6ll27%2BPrj0QzffQIWX0ATo%3D&reserved=0
>>
>> Cc: Mario Limonciello <mario.limonciello@amd.com>
>> Cc: Mark Pearson <markpearson@lenovo.com>
>> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
>> ---
>> drivers/platform/x86/amd/pmf/acpi.c | 59 ++++++++++++++++++++++++
>> drivers/platform/x86/amd/pmf/auto-mode.c | 22 +++++++++
>> drivers/platform/x86/amd/pmf/core.c | 32 +++++++++++++
>> drivers/platform/x86/amd/pmf/pmf.h | 22 +++++++++
>> 4 files changed, 135 insertions(+)
>>
>> diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
>> index 724150ec58fb..dedaebf18d88 100644
>> --- a/drivers/platform/x86/amd/pmf/acpi.c
>> +++ b/drivers/platform/x86/amd/pmf/acpi.c
>> @@ -12,6 +12,8 @@
>> #include "pmf.h"
>>
>> #define APMF_METHOD "\\_SB_.PMF_.APMF"
>> +#define APMF_CQL_NOTIFICATION 2
>> +#define APMF_AMT_NOTIFICATION 3
>>
>> static unsigned long supported_func;
>>
>> @@ -99,6 +101,7 @@ static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
>> {
>> func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
>> func->static_slider_granular = mask & APMF_FUNC_STATIC_SLIDER_GRANULAR;
>> + func->sbios_requests = mask & APMF_FUNC_SBIOS_REQUESTS;
>
> I just noticed that this is not correct, this should be:
>
> func->system_params = mask & BIT(APMF_FUNC_... - 1);
>
> Which does something completely different!
>
> Also this information is duplicated by:
>
> is_apmf_func_supported(APMF_FUNC_...) so please drop the apmf_if_functions
> struct completely and use is_apmf_func_supported() everywhere.
>
>> func->auto_mode_def = mask & APMF_FUNC_AUTO_MODE;
>> func->fan_table_idx = mask & APMF_FUNC_SET_FAN_IDX;
>> }
>> @@ -167,6 +170,44 @@ int apmf_get_auto_mode_def(struct apmf_if *apmf_if, struct apmf_auto_mode *data)
>> data, sizeof(*data));
>> }
>>
>> +int apmf_get_sbios_requests(struct apmf_if *apmf_if, struct apmf_sbios_req *req)
>> +{
>> + return apmf_if_call_store_buffer(apmf_if, APMF_FUNC_SBIOS_REQUESTS,
>> + req, sizeof(*req));
>> +}
>> +
>> +static void apmf_event_handler(acpi_handle handle, u32 event, void *data)
>> +{
>> + struct amd_pmf_dev *pmf_dev = data;
>> + struct apmf_if *apmf_if = pmf_dev->apmf_if;
>> + int ret;
>> +
>> + if (apmf_if->func.sbios_requests) {
>
> IMHO it would be better to change the test for [un]registering the handler to:
>
> if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE) &&
> is_apmf_func_supported(APMF_FUNC_SBIOS_REQUESTS))
>
> and then the sbios_requests test here (in apmf_event_handler()) can be dropped.
>
>> + struct apmf_sbios_req req;
>> +
>> + ret = apmf_get_sbios_requests(apmf_if, &req);
>> + if (ret) {
>> + dev_err(pmf_dev->dev, "Failed to get SBIOS requests:%d\n", ret);
>> + return;
>> + }
>> +
>> + if (req.pending_req & BIT(APMF_AMT_NOTIFICATION)) {
>> + pr_debug("PMF: AMT is supported and notifications %s\n",
>> + req.amt_event ? "Enabled" : "Disabled");
>> + pmf_dev->is_amt_event = !!req.amt_event;
>
> The is_amt_event variable is not storing if the last event was an amt_event but
> rather wether AMT is enabled or not, so please rename it to amt_enabled.
>
> Also by just setting the flag here you are introducing a race, at least on
> AMT disabling.
>
> 1: amd_pmf_get_metrics() is running, has just passed the
> "if (dev->is_amt_event)" check and is about to call amd_pmf_trans_automode().
>
> 2. apmf_event_handler() gets called to disable AMT
>
> 3. apmf_event_handler() finishes and the firmware now starts programming the
> limits because thinkpad_acpi's platform_profile has been set to !balanced.
>
> 4. amd_pmf_get_metrics() continues and calls amd_pmf_trans_automode()
> 5. amd_pmf_trans_automode() reprograms the limits, overriding the ones
> set by the firmware eventhough AMT is disabled at this point.
>
> To avoid this race apmf_event_handler() and amd_pmf_get_metrics() should both
> lock a newly added shared update_mutex mutex while running so that they cannot
> both run at the same time.
>
> ###
>
> Likewise restoring the static slider settings on AMT disable is also racy
> and to fix that the amd_pmf_reset_amt(dev);should be moved here; and then for
> consistency and cleanness the setting of the initial AMT values should also
> be moved here.
>
> This will result in adding the following code here:
>
> dev_dbg(dev->dev, "amt enabled: %d\n", dev->amt_enabled);
>
> if (pmf_dev->amt_enabled)
> amd_pmf_handle_automode(dev, SLIDER_OP_SET, config_store.current_mode, NULL);
> else
> amd_pmf_reset_amt(dev);
>
> Also be handling the flanks (enabled<->disabled transitions) here like this there
> no longer is a need for a separate amt_running variable to detect the flanks in other
> places.
>
>> + }
>> +
>> + if (req.pending_req & BIT(APMF_CQL_NOTIFICATION)) {
>> + pr_debug("PMF: CQL is supported and notifications %s\n",
>> + req.cql_event ? "Enabled" : "Disabled");
>> + pmf_dev->is_cql_event = !!req.cql_event;
>
> is_cql_event is only used to pass req.cql_event to amd_pmf_update_2_cql()
> please drop it from struct amd_pmf_dev and make it an argument to
> amd_pmf_update_2_cql()
>
>> +
>> + /* update the target mode information */
>> + amd_pmf_update_2_cql(pmf_dev);
>> + }
>> + }
>> +}
>> +
>> static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
>> {
>> struct apmf_verify_interface output;
>> @@ -211,12 +252,19 @@ static int apmf_get_system_params(struct apmf_if *apmf_if)
>>
>> void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev)
>> {
>> + acpi_handle ahandle = ACPI_HANDLE(pmf_dev->dev);
>> +
>
> Hmm, is this the handle for: "\\_SB_.PMF" ? as in the parent of:
>
> #define APMF_METHOD "\\_SB_.PMF_.APMF"
>
> ?
>
> In that case then apmf_if_call() can use:
>
> acpi_handle ahandle = ACPI_HANDLE(pmf_dev->dev);
>
> ...
>
> status = acpi_evaluate_object(ahandle, "APMF", &apmf_if_arg_list, &buffer);
>
> and the whole:
>
> + status = acpi_get_handle(NULL, (acpi_string) APMF_METHOD, &apmf_if_handle);
> + if (ACPI_FAILURE(status))
> + return -ENODEV;
> +
>
> bit + storing the handle + also:
>
> +#define APMF_METHOD "\\_SB_.PMF_.APMF"
>
> can then all be dropped. That would be a nice cleanup and then calling
> dev_err(...) for the ACPI errors also makes even more sense
>
>
>> if (pmf_dev->apmf_if->func.sbios_heartbeat)
>> cancel_delayed_work_sync(&pmf_dev->heart_beat);
>> +
>> + if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE))
>
> As mentioned above IMHO it would make sense to make this check:
>
> if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE) &&
> is_apmf_func_supported(APMF_FUNC_SBIOS_REQUESTS))
>
>> + acpi_remove_notify_handler(ahandle, ACPI_ALL_NOTIFY,
>> + apmf_event_handler);
>> }
>>
>> int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
>> {
>> + acpi_handle ahandle = ACPI_HANDLE(pmf_dev->dev);
>> struct apmf_notification_cfg *n;
>> acpi_handle apmf_if_handle;
>> struct apmf_if *apmf_if;
>> @@ -256,6 +304,17 @@ int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
>> schedule_delayed_work(&pmf_dev->heart_beat, 0);
>> }
>>
>> + /* Install the APMF Notify handler */
>> + if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE)) {
>
> As mentioned above IMHO it would make sense to make this check:
>
> if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE) &&
> is_apmf_func_supported(APMF_FUNC_SBIOS_REQUESTS))
>
> What is missing here is detecting the initial AMT and CQL values and configuring
> things accordingly. Maybe this can be done by calling apmf_event_handler
> manually once before registering it ?
Thank you for all the review remarks.
I think manually calling the apmf_event_handler() is not required as the
the structure has the info on the pending AMT/CQL requests.
I did attempt to make this change but it did not make any difference.
Leaving this apart, I tried to address all of the comments in v2. Kindly
have a look again.
Thanks,
Shyam
>
> Note the initial state detection must be done before registering the handler
> to avoid races.
>
>> + status = acpi_install_notify_handler(ahandle,
>> + ACPI_ALL_NOTIFY,
>> + apmf_event_handler, pmf_dev);
>> + if (ACPI_FAILURE(status)) {
>> + dev_err(pmf_dev->dev, "failed to install notify handler\n");
>> + return -ENODEV;
>> + }
>> + }
>> +
>> out:
>> return ret;
>> }
>> diff --git a/drivers/platform/x86/amd/pmf/auto-mode.c b/drivers/platform/x86/amd/pmf/auto-mode.c
>> index 954fde25e71e..a85ef4b95cdb 100644
>> --- a/drivers/platform/x86/amd/pmf/auto-mode.c
>> +++ b/drivers/platform/x86/amd/pmf/auto-mode.c
>> @@ -111,6 +111,13 @@ void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t t
>> bool update = false;
>> int i, j;
>>
>> + if (!dev->amt_running && dev->is_amt_event) {
>> + dev_dbg(dev->dev, "setting AMT thermals\n");
>> + amd_pmf_handle_automode(dev, SLIDER_OP_SET, config_store.current_mode, NULL);
>> + dev->amt_running = true;
>> + return;
>> + }
>> +
>
> As suggested above this is best done on receiving the amt_event.
>
>> /* Get the average moving average computed by auto mode algorithm */
>> avg_power = amd_pmf_get_moving_avg(socket_power);
>>
>> @@ -161,6 +168,21 @@ void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t t
>> }
>> }
>>
>> +void amd_pmf_update_2_cql(struct amd_pmf_dev *dev)
>> +{
>> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode =
>> + dev->is_cql_event ? AUTO_PERFORMANCE_ON_LAP : AUTO_PERFORMANCE;
>> + if ((config_store.current_mode == AUTO_PERFORMANCE ||
>> + config_store.current_mode == AUTO_PERFORMANCE_ON_LAP) &&
>> + config_store.current_mode !=
>> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode) {
>> + config_store.current_mode =
>> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode;
>> + amd_pmf_handle_automode(dev, SLIDER_OP_SET, config_store.current_mode, NULL);
>> + }
>> + dev_dbg(dev->dev, "updated CQL thermals\n");
>> +}
>> +
>> static void amd_pmf_get_power_threshold(void)
>> {
>> config_store.transition[AUTO_TRANSITION_TO_QUIET].power_threshold =
>> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
>> index ef71f8326248..bfae779ccc23 100644
>> --- a/drivers/platform/x86/amd/pmf/core.c
>> +++ b/drivers/platform/x86/amd/pmf/core.c
>> @@ -93,6 +93,27 @@ int amd_pmf_get_power_source(void)
>> return POWER_SOURCE_DC;
>> }
>>
>> +static void amd_pmf_reset_amt(struct amd_pmf_dev *dev)
>> +{
>> + pr_debug("resetting AMT thermals\n");
>> + dev->amt_running = false;
>> +
>> + /*
>> + * OEM BIOS implementation guide says that if the auto mode is enabled
>> + * the platform_profile registration shall be done by the OEM driver.
>> + * There could be cases where both static slider and auto mode BIOS
>> + * functions are enabled and we could end up in a race. To avoid that
>> + * add a protection and touch the static slider only if that's enabled
>> + * from the BIOS side.
>> + */
>
> This comment seems odd. The is_apmf_func_supported() does not help to
> avoid the race. This check is necessary to avoid amd_pmf_update_slider()
> on systems where amd_pmf_load_defaults_sps() does not run. Otherwise
> amd_pmf_update_slider() ends up writing uninitialized (all 0) limits.
>
> To fix the race the amd_pmf_update_slider() call must be done before
> apmf_event_handler() exits as suggested above. The feature check has
> nothing to do with the race ...
>
>
>
>> +
>> + if (is_apmf_func_supported(APMF_FUNC_STATIC_SLIDER_GRANULAR)) {
>> + u8 mode = amd_pmf_get_pprof_modes(dev);
>> +
>> + amd_pmf_update_slider(dev, SLIDER_OP_SET, mode, NULL);
>> + }
>> +}
>> +
>> static void amd_pmf_get_metrics(struct work_struct *work)
>> {
>> struct amd_pmf_dev *dev = container_of(work, struct amd_pmf_dev, work_buffer.work);
>> @@ -103,6 +124,9 @@ static void amd_pmf_get_metrics(struct work_struct *work)
>> /* Get the current profile information */
>> current_profile = READ_ONCE(dev->current_profile);
>>
>> + if (!dev->is_amt_event)
>> + dev_dbg(dev->dev, "amt event: %s\n", dev->is_amt_event ? "Enabled" : "Disabled");
>> +
>
> This is weird, you check for !is_amt_event so this will only ever log:
> "amt event: Disabled" so the check in the dev_dbg is not necessary. Also this is
> best done when receiving the amt-event, as suggested above so this can just be dropped.
>
>> /* Transfer table contents */
>> memset(&dev->m_table, 0, sizeof(dev->m_table));
>> amd_pmf_send_cmd(dev, SET_TRANSFER_TABLE, 0, 7, NULL);
>> @@ -112,6 +136,14 @@ static void amd_pmf_get_metrics(struct work_struct *work)
>> /* Calculate the avg SoC power consumption */
>> socket_power = dev->m_table.apu_power + dev->m_table.dgpu_power;
>>
>> + if (dev->is_amt_event) {
>> + /* Apply the Auto Mode transition */
>> + amd_pmf_trans_automode(dev, socket_power, time_elapsed_ms);
>> + } else if (!dev->is_amt_event && dev->amt_running) {
>> + /* reset to other mode if we receive a AMT disable event */
>> + amd_pmf_reset_amt(dev);
>> + }
>> +
>
> As discussed above the amd_pmf_reset_amt() call must be done from inside
> apmf_event_handler(). So the whole "else if (...) {}" block can be dropped here.
>
>> dev->start_time = ktime_to_ms(ktime_get());
>> schedule_delayed_work(&dev->work_buffer, msecs_to_jiffies(metrics_table_loop_ms));
>> }
>> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
>> index 4618ddc5a662..0f8b25524845 100644
>> --- a/drivers/platform/x86/amd/pmf/pmf.h
>> +++ b/drivers/platform/x86/amd/pmf/pmf.h
>> @@ -17,6 +17,7 @@
>> /* APMF Functions */
>> #define APMF_FUNC_VERIFY_INTERFACE 0
>> #define APMF_FUNC_GET_SYS_PARAMS 1
>> +#define APMF_FUNC_SBIOS_REQUESTS 2
>> #define APMF_FUNC_SBIOS_HEARTBEAT 4
>> #define APMF_FUNC_AUTO_MODE 5
>> #define APMF_FUNC_SET_FAN_IDX 7
>> @@ -49,6 +50,7 @@
>> /* AMD PMF BIOS interfaces */
>> struct apmf_if_functions {
>> bool system_params;
>> + bool sbios_requests;
>> bool sbios_heartbeat;
>> bool auto_mode_def;
>> bool fan_table_idx;
>> @@ -80,6 +82,21 @@ struct apmf_system_params {
>> u32 heartbeat_int;
>> } __packed;
>>
>> +struct apmf_sbios_req {
>> + u16 size;
>> + u32 pending_req;
>> + u8 rsd;
>> + u8 cql_event;
>> + u8 amt_event;
>> + u32 fppt;
>> + u32 sppt;
>> + u32 fppt_apu_only;
>> + u32 spl;
>> + u32 stt_min_limit;
>> + u8 skin_temp_apu;
>> + u8 skin_temp_hs2;
>> +} __packed;
>> +
>> struct apmf_fan_idx {
>> u16 size;
>> u8 fan_ctl_mode;
>> @@ -161,6 +178,9 @@ struct amd_pmf_dev {
>> struct smu_pmf_metrics m_table;
>> struct delayed_work work_buffer;
>> ktime_t start_time;
>> + bool is_amt_event;
>
> The is_amt_event variable is not storing if the last event was an amt_event but
> rather wether AMT is enabled or not, so please rename it to amt_enabled.
>
>> + bool is_cql_event;
>> + bool amt_running;
>
> As discussed both these struct members can be dropped.
>
>> };
>>
>> struct apmf_sps_prop_granular {
>> @@ -320,5 +340,7 @@ int apmf_get_auto_mode_def(struct apmf_if *ampf_if, struct apmf_auto_mode *data)
>> void amd_pmf_init_auto_mode(struct amd_pmf_dev *dev);
>> void amd_pmf_deinit_auto_mode(struct amd_pmf_dev *dev);
>> void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t time_elapsed_ms);
>> +int apmf_get_sbios_requests(struct apmf_if *ampf_if, struct apmf_sbios_req *req);
>> +void amd_pmf_update_2_cql(struct amd_pmf_dev *dev);
>>
>> #endif /* PMF_H */
>
> Regards,
>
> Hans
>
^ permalink raw reply [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 09/11] platform/x86/amd/pmf: Handle AMT and CQL events for Auto mode
2022-08-02 11:22 ` Shyam Sundar S K
@ 2022-08-02 12:20 ` Hans de Goede
0 siblings, 0 replies; 33+ messages in thread
From: Hans de Goede @ 2022-08-02 12:20 UTC (permalink / raw)
To: Shyam Sundar S K, markgross
Cc: platform-driver-x86, Patil.Reddy, Mario Limonciello, Mark Pearson
Hi,
On 8/2/22 13:22, Shyam Sundar S K wrote:
> Hi Hans,
>
> On 8/1/2022 7:25 PM, Hans de Goede wrote:
>> Hi,
>>
>> On 7/28/22 20:20, Shyam Sundar S K wrote:
>>> The transition to auto-mode happens when the PMF driver receives
>>> AMT (Auto Mode transition) event. transition logic will reside in the
>>> PMF driver but the events would come from other supported drivers[1].
>>>
>>> The thermal parameters would vary between when a performance "on-lap" mode
>>> is detected and versus when not. The CQL event would get triggered from
>>> other drivers, so that PMF driver would adjust the system thermal config
>>> based on the ACPI inputs.
>>>
>>> OEMs can control whether or not to enable AMT or CQL via other supported
>>> drivers[1] but the actual transition logic resides in the AMD PMF driver.
>>> When an AMT event is received the automatic mode transition RAPL algorithm
>>> will run. When a CQL event is received an performance "on-lap" mode will
>>> be enabled and thermal parameters will be adjusted accordingly.
>>>
>>> [1]
>>> Link: https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgit.kernel.org%2Fpub%2Fscm%2Flinux%2Fkernel%2Fgit%2Fpdx86%2Fplatform-drivers-x86.git%2Fcommit%2F%3Fh%3Dreview-hans%26id%3D755b249250df1b612d982f3b702c831b26ecdf73&data=05%7C01%7Cshyam-sundar.s-k%40amd.com%7C1297e4cd71784092e05008da73c59095%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637949589653252953%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=5QDwDnnn45ZCwq2MfPQjv6ll27%2BPrj0QzffQIWX0ATo%3D&reserved=0
>>>
>>> Cc: Mario Limonciello <mario.limonciello@amd.com>
>>> Cc: Mark Pearson <markpearson@lenovo.com>
>>> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
>>> ---
>>> drivers/platform/x86/amd/pmf/acpi.c | 59 ++++++++++++++++++++++++
>>> drivers/platform/x86/amd/pmf/auto-mode.c | 22 +++++++++
>>> drivers/platform/x86/amd/pmf/core.c | 32 +++++++++++++
>>> drivers/platform/x86/amd/pmf/pmf.h | 22 +++++++++
>>> 4 files changed, 135 insertions(+)
>>>
>>> diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
>>> index 724150ec58fb..dedaebf18d88 100644
>>> --- a/drivers/platform/x86/amd/pmf/acpi.c
>>> +++ b/drivers/platform/x86/amd/pmf/acpi.c
>>> @@ -12,6 +12,8 @@
>>> #include "pmf.h"
>>>
>>> #define APMF_METHOD "\\_SB_.PMF_.APMF"
>>> +#define APMF_CQL_NOTIFICATION 2
>>> +#define APMF_AMT_NOTIFICATION 3
>>>
>>> static unsigned long supported_func;
>>>
>>> @@ -99,6 +101,7 @@ static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
>>> {
>>> func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
>>> func->static_slider_granular = mask & APMF_FUNC_STATIC_SLIDER_GRANULAR;
>>> + func->sbios_requests = mask & APMF_FUNC_SBIOS_REQUESTS;
>>
>> I just noticed that this is not correct, this should be:
>>
>> func->system_params = mask & BIT(APMF_FUNC_... - 1);
>>
>> Which does something completely different!
>>
>> Also this information is duplicated by:
>>
>> is_apmf_func_supported(APMF_FUNC_...) so please drop the apmf_if_functions
>> struct completely and use is_apmf_func_supported() everywhere.
>>
>>> func->auto_mode_def = mask & APMF_FUNC_AUTO_MODE;
>>> func->fan_table_idx = mask & APMF_FUNC_SET_FAN_IDX;
>>> }
>>> @@ -167,6 +170,44 @@ int apmf_get_auto_mode_def(struct apmf_if *apmf_if, struct apmf_auto_mode *data)
>>> data, sizeof(*data));
>>> }
>>>
>>> +int apmf_get_sbios_requests(struct apmf_if *apmf_if, struct apmf_sbios_req *req)
>>> +{
>>> + return apmf_if_call_store_buffer(apmf_if, APMF_FUNC_SBIOS_REQUESTS,
>>> + req, sizeof(*req));
>>> +}
>>> +
>>> +static void apmf_event_handler(acpi_handle handle, u32 event, void *data)
>>> +{
>>> + struct amd_pmf_dev *pmf_dev = data;
>>> + struct apmf_if *apmf_if = pmf_dev->apmf_if;
>>> + int ret;
>>> +
>>> + if (apmf_if->func.sbios_requests) {
>>
>> IMHO it would be better to change the test for [un]registering the handler to:
>>
>> if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE) &&
>> is_apmf_func_supported(APMF_FUNC_SBIOS_REQUESTS))
>>
>> and then the sbios_requests test here (in apmf_event_handler()) can be dropped.
>>
>>> + struct apmf_sbios_req req;
>>> +
>>> + ret = apmf_get_sbios_requests(apmf_if, &req);
>>> + if (ret) {
>>> + dev_err(pmf_dev->dev, "Failed to get SBIOS requests:%d\n", ret);
>>> + return;
>>> + }
>>> +
>>> + if (req.pending_req & BIT(APMF_AMT_NOTIFICATION)) {
>>> + pr_debug("PMF: AMT is supported and notifications %s\n",
>>> + req.amt_event ? "Enabled" : "Disabled");
>>> + pmf_dev->is_amt_event = !!req.amt_event;
>>
>> The is_amt_event variable is not storing if the last event was an amt_event but
>> rather wether AMT is enabled or not, so please rename it to amt_enabled.
>>
>> Also by just setting the flag here you are introducing a race, at least on
>> AMT disabling.
>>
>> 1: amd_pmf_get_metrics() is running, has just passed the
>> "if (dev->is_amt_event)" check and is about to call amd_pmf_trans_automode().
>>
>> 2. apmf_event_handler() gets called to disable AMT
>>
>> 3. apmf_event_handler() finishes and the firmware now starts programming the
>> limits because thinkpad_acpi's platform_profile has been set to !balanced.
>>
>> 4. amd_pmf_get_metrics() continues and calls amd_pmf_trans_automode()
>> 5. amd_pmf_trans_automode() reprograms the limits, overriding the ones
>> set by the firmware eventhough AMT is disabled at this point.
>>
>> To avoid this race apmf_event_handler() and amd_pmf_get_metrics() should both
>> lock a newly added shared update_mutex mutex while running so that they cannot
>> both run at the same time.
>>
>> ###
>>
>> Likewise restoring the static slider settings on AMT disable is also racy
>> and to fix that the amd_pmf_reset_amt(dev);should be moved here; and then for
>> consistency and cleanness the setting of the initial AMT values should also
>> be moved here.
>>
>> This will result in adding the following code here:
>>
>> dev_dbg(dev->dev, "amt enabled: %d\n", dev->amt_enabled);
>>
>> if (pmf_dev->amt_enabled)
>> amd_pmf_handle_automode(dev, SLIDER_OP_SET, config_store.current_mode, NULL);
>> else
>> amd_pmf_reset_amt(dev);
>>
>> Also be handling the flanks (enabled<->disabled transitions) here like this there
>> no longer is a need for a separate amt_running variable to detect the flanks in other
>> places.
>>
>>> + }
>>> +
>>> + if (req.pending_req & BIT(APMF_CQL_NOTIFICATION)) {
>>> + pr_debug("PMF: CQL is supported and notifications %s\n",
>>> + req.cql_event ? "Enabled" : "Disabled");
>>> + pmf_dev->is_cql_event = !!req.cql_event;
>>
>> is_cql_event is only used to pass req.cql_event to amd_pmf_update_2_cql()
>> please drop it from struct amd_pmf_dev and make it an argument to
>> amd_pmf_update_2_cql()
>>
>>> +
>>> + /* update the target mode information */
>>> + amd_pmf_update_2_cql(pmf_dev);
>>> + }
>>> + }
>>> +}
>>> +
>>> static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
>>> {
>>> struct apmf_verify_interface output;
>>> @@ -211,12 +252,19 @@ static int apmf_get_system_params(struct apmf_if *apmf_if)
>>>
>>> void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev)
>>> {
>>> + acpi_handle ahandle = ACPI_HANDLE(pmf_dev->dev);
>>> +
>>
>> Hmm, is this the handle for: "\\_SB_.PMF" ? as in the parent of:
>>
>> #define APMF_METHOD "\\_SB_.PMF_.APMF"
>>
>> ?
>>
>> In that case then apmf_if_call() can use:
>>
>> acpi_handle ahandle = ACPI_HANDLE(pmf_dev->dev);
>>
>> ...
>>
>> status = acpi_evaluate_object(ahandle, "APMF", &apmf_if_arg_list, &buffer);
>>
>> and the whole:
>>
>> + status = acpi_get_handle(NULL, (acpi_string) APMF_METHOD, &apmf_if_handle);
>> + if (ACPI_FAILURE(status))
>> + return -ENODEV;
>> +
>>
>> bit + storing the handle + also:
>>
>> +#define APMF_METHOD "\\_SB_.PMF_.APMF"
>>
>> can then all be dropped. That would be a nice cleanup and then calling
>> dev_err(...) for the ACPI errors also makes even more sense
>>
>>
>>> if (pmf_dev->apmf_if->func.sbios_heartbeat)
>>> cancel_delayed_work_sync(&pmf_dev->heart_beat);
>>> +
>>> + if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE))
>>
>> As mentioned above IMHO it would make sense to make this check:
>>
>> if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE) &&
>> is_apmf_func_supported(APMF_FUNC_SBIOS_REQUESTS))
>>
>>> + acpi_remove_notify_handler(ahandle, ACPI_ALL_NOTIFY,
>>> + apmf_event_handler);
>>> }
>>>
>>> int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
>>> {
>>> + acpi_handle ahandle = ACPI_HANDLE(pmf_dev->dev);
>>> struct apmf_notification_cfg *n;
>>> acpi_handle apmf_if_handle;
>>> struct apmf_if *apmf_if;
>>> @@ -256,6 +304,17 @@ int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
>>> schedule_delayed_work(&pmf_dev->heart_beat, 0);
>>> }
>>>
>>> + /* Install the APMF Notify handler */
>>> + if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE)) {
>>
>> As mentioned above IMHO it would make sense to make this check:
>>
>> if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE) &&
>> is_apmf_func_supported(APMF_FUNC_SBIOS_REQUESTS))
>>
>> What is missing here is detecting the initial AMT and CQL values and configuring
>> things accordingly. Maybe this can be done by calling apmf_event_handler
>> manually once before registering it ?
>
> Thank you for all the review remarks.
>
> I think manually calling the apmf_event_handler() is not required as the
> the structure has the info on the pending AMT/CQL requests.
Right the structure has the pending requests flags. But to see those
you first need to read the structure, which is done by
the apmf_event_handler(). There is no notification to the ACPI
AML of a notifier handler getting registered, so the ACPI AML cannot
re-do any missed notifies once a notifier shows up.
> I did attempt to make this change but it did not make any difference.
> Leaving this apart, I tried to address all of the comments in v2. Kindly
> have a look again.
The scenario which I believe will break is:
1. A thinkpad is configured to balanced profile at boot
2. thinkpad_acpi loads first, sees this and enables AMT, the
DYTC_FUNCTION_AMT, DYTC_MODE_AMT_ENABLE command done by thinkpad_acpi
will cause an ACPI Notify() on the "\\_SB_.PMF" device, bot nothing
is listening
3. the AMD-PMF driver loads (after thinkpad_acpi) at this point the
amd_pmf_dev.amt_enabled flag is 0, even though thinkpad_acpi
has enabled it
4. Now for as long as no further platform_profile changes happen
the AMD-PMF driver will have the amd_pmf_dev.amt_enabled flag
set to 0 and won't run the auto-mode code, even though it should
run the automode code.
I would expect calling the apmf_event_handler() once manually
after the acpi_install_notify_handler() call to fix this.
Regards,
Hans
>
> Thanks,
> Shyam
>
>>
>> Note the initial state detection must be done before registering the handler
>> to avoid races.
>>
>>> + status = acpi_install_notify_handler(ahandle,
>>> + ACPI_ALL_NOTIFY,
>>> + apmf_event_handler, pmf_dev);
>>> + if (ACPI_FAILURE(status)) {
>>> + dev_err(pmf_dev->dev, "failed to install notify handler\n");
>>> + return -ENODEV;
>>> + }
>>> + }
>>> +
>>> out:
>>> return ret;
>>> }
>>> diff --git a/drivers/platform/x86/amd/pmf/auto-mode.c b/drivers/platform/x86/amd/pmf/auto-mode.c
>>> index 954fde25e71e..a85ef4b95cdb 100644
>>> --- a/drivers/platform/x86/amd/pmf/auto-mode.c
>>> +++ b/drivers/platform/x86/amd/pmf/auto-mode.c
>>> @@ -111,6 +111,13 @@ void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t t
>>> bool update = false;
>>> int i, j;
>>>
>>> + if (!dev->amt_running && dev->is_amt_event) {
>>> + dev_dbg(dev->dev, "setting AMT thermals\n");
>>> + amd_pmf_handle_automode(dev, SLIDER_OP_SET, config_store.current_mode, NULL);
>>> + dev->amt_running = true;
>>> + return;
>>> + }
>>> +
>>
>> As suggested above this is best done on receiving the amt_event.
>>
>>> /* Get the average moving average computed by auto mode algorithm */
>>> avg_power = amd_pmf_get_moving_avg(socket_power);
>>>
>>> @@ -161,6 +168,21 @@ void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t t
>>> }
>>> }
>>>
>>> +void amd_pmf_update_2_cql(struct amd_pmf_dev *dev)
>>> +{
>>> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode =
>>> + dev->is_cql_event ? AUTO_PERFORMANCE_ON_LAP : AUTO_PERFORMANCE;
>>> + if ((config_store.current_mode == AUTO_PERFORMANCE ||
>>> + config_store.current_mode == AUTO_PERFORMANCE_ON_LAP) &&
>>> + config_store.current_mode !=
>>> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode) {
>>> + config_store.current_mode =
>>> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode;
>>> + amd_pmf_handle_automode(dev, SLIDER_OP_SET, config_store.current_mode, NULL);
>>> + }
>>> + dev_dbg(dev->dev, "updated CQL thermals\n");
>>> +}
>>> +
>>> static void amd_pmf_get_power_threshold(void)
>>> {
>>> config_store.transition[AUTO_TRANSITION_TO_QUIET].power_threshold =
>>> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
>>> index ef71f8326248..bfae779ccc23 100644
>>> --- a/drivers/platform/x86/amd/pmf/core.c
>>> +++ b/drivers/platform/x86/amd/pmf/core.c
>>> @@ -93,6 +93,27 @@ int amd_pmf_get_power_source(void)
>>> return POWER_SOURCE_DC;
>>> }
>>>
>>> +static void amd_pmf_reset_amt(struct amd_pmf_dev *dev)
>>> +{
>>> + pr_debug("resetting AMT thermals\n");
>>> + dev->amt_running = false;
>>> +
>>> + /*
>>> + * OEM BIOS implementation guide says that if the auto mode is enabled
>>> + * the platform_profile registration shall be done by the OEM driver.
>>> + * There could be cases where both static slider and auto mode BIOS
>>> + * functions are enabled and we could end up in a race. To avoid that
>>> + * add a protection and touch the static slider only if that's enabled
>>> + * from the BIOS side.
>>> + */
>>
>> This comment seems odd. The is_apmf_func_supported() does not help to
>> avoid the race. This check is necessary to avoid amd_pmf_update_slider()
>> on systems where amd_pmf_load_defaults_sps() does not run. Otherwise
>> amd_pmf_update_slider() ends up writing uninitialized (all 0) limits.
>>
>> To fix the race the amd_pmf_update_slider() call must be done before
>> apmf_event_handler() exits as suggested above. The feature check has
>> nothing to do with the race ...
>>
>>
>>
>>> +
>>> + if (is_apmf_func_supported(APMF_FUNC_STATIC_SLIDER_GRANULAR)) {
>>> + u8 mode = amd_pmf_get_pprof_modes(dev);
>>> +
>>> + amd_pmf_update_slider(dev, SLIDER_OP_SET, mode, NULL);
>>> + }
>>> +}
>>> +
>>> static void amd_pmf_get_metrics(struct work_struct *work)
>>> {
>>> struct amd_pmf_dev *dev = container_of(work, struct amd_pmf_dev, work_buffer.work);
>>> @@ -103,6 +124,9 @@ static void amd_pmf_get_metrics(struct work_struct *work)
>>> /* Get the current profile information */
>>> current_profile = READ_ONCE(dev->current_profile);
>>>
>>> + if (!dev->is_amt_event)
>>> + dev_dbg(dev->dev, "amt event: %s\n", dev->is_amt_event ? "Enabled" : "Disabled");
>>> +
>>
>> This is weird, you check for !is_amt_event so this will only ever log:
>> "amt event: Disabled" so the check in the dev_dbg is not necessary. Also this is
>> best done when receiving the amt-event, as suggested above so this can just be dropped.
>>
>>> /* Transfer table contents */
>>> memset(&dev->m_table, 0, sizeof(dev->m_table));
>>> amd_pmf_send_cmd(dev, SET_TRANSFER_TABLE, 0, 7, NULL);
>>> @@ -112,6 +136,14 @@ static void amd_pmf_get_metrics(struct work_struct *work)
>>> /* Calculate the avg SoC power consumption */
>>> socket_power = dev->m_table.apu_power + dev->m_table.dgpu_power;
>>>
>>> + if (dev->is_amt_event) {
>>> + /* Apply the Auto Mode transition */
>>> + amd_pmf_trans_automode(dev, socket_power, time_elapsed_ms);
>>> + } else if (!dev->is_amt_event && dev->amt_running) {
>>> + /* reset to other mode if we receive a AMT disable event */
>>> + amd_pmf_reset_amt(dev);
>>> + }
>>> +
>>
>> As discussed above the amd_pmf_reset_amt() call must be done from inside
>> apmf_event_handler(). So the whole "else if (...) {}" block can be dropped here.
>>
>>> dev->start_time = ktime_to_ms(ktime_get());
>>> schedule_delayed_work(&dev->work_buffer, msecs_to_jiffies(metrics_table_loop_ms));
>>> }
>>> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
>>> index 4618ddc5a662..0f8b25524845 100644
>>> --- a/drivers/platform/x86/amd/pmf/pmf.h
>>> +++ b/drivers/platform/x86/amd/pmf/pmf.h
>>> @@ -17,6 +17,7 @@
>>> /* APMF Functions */
>>> #define APMF_FUNC_VERIFY_INTERFACE 0
>>> #define APMF_FUNC_GET_SYS_PARAMS 1
>>> +#define APMF_FUNC_SBIOS_REQUESTS 2
>>> #define APMF_FUNC_SBIOS_HEARTBEAT 4
>>> #define APMF_FUNC_AUTO_MODE 5
>>> #define APMF_FUNC_SET_FAN_IDX 7
>>> @@ -49,6 +50,7 @@
>>> /* AMD PMF BIOS interfaces */
>>> struct apmf_if_functions {
>>> bool system_params;
>>> + bool sbios_requests;
>>> bool sbios_heartbeat;
>>> bool auto_mode_def;
>>> bool fan_table_idx;
>>> @@ -80,6 +82,21 @@ struct apmf_system_params {
>>> u32 heartbeat_int;
>>> } __packed;
>>>
>>> +struct apmf_sbios_req {
>>> + u16 size;
>>> + u32 pending_req;
>>> + u8 rsd;
>>> + u8 cql_event;
>>> + u8 amt_event;
>>> + u32 fppt;
>>> + u32 sppt;
>>> + u32 fppt_apu_only;
>>> + u32 spl;
>>> + u32 stt_min_limit;
>>> + u8 skin_temp_apu;
>>> + u8 skin_temp_hs2;
>>> +} __packed;
>>> +
>>> struct apmf_fan_idx {
>>> u16 size;
>>> u8 fan_ctl_mode;
>>> @@ -161,6 +178,9 @@ struct amd_pmf_dev {
>>> struct smu_pmf_metrics m_table;
>>> struct delayed_work work_buffer;
>>> ktime_t start_time;
>>> + bool is_amt_event;
>>
>> The is_amt_event variable is not storing if the last event was an amt_event but
>> rather wether AMT is enabled or not, so please rename it to amt_enabled.
>>
>>> + bool is_cql_event;
>>> + bool amt_running;
>>
>> As discussed both these struct members can be dropped.
>>
>>> };
>>>
>>> struct apmf_sps_prop_granular {
>>> @@ -320,5 +340,7 @@ int apmf_get_auto_mode_def(struct apmf_if *ampf_if, struct apmf_auto_mode *data)
>>> void amd_pmf_init_auto_mode(struct amd_pmf_dev *dev);
>>> void amd_pmf_deinit_auto_mode(struct amd_pmf_dev *dev);
>>> void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t time_elapsed_ms);
>>> +int apmf_get_sbios_requests(struct apmf_if *ampf_if, struct apmf_sbios_req *req);
>>> +void amd_pmf_update_2_cql(struct amd_pmf_dev *dev);
>>>
>>> #endif /* PMF_H */
>>
>> Regards,
>>
>> Hans
>>
>
^ permalink raw reply [flat|nested] 33+ messages in thread
* Re: [PATCH v2 RESEND 09/11] platform/x86/amd/pmf: Handle AMT and CQL events for Auto mode
2022-07-28 18:20 ` [PATCH v2 RESEND 09/11] platform/x86/amd/pmf: Handle AMT and CQL events for Auto mode Shyam Sundar S K
2022-08-01 13:55 ` Hans de Goede
@ 2022-08-01 14:01 ` Hans de Goede
1 sibling, 0 replies; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 14:01 UTC (permalink / raw)
To: Shyam Sundar S K, markgross
Cc: platform-driver-x86, Patil.Reddy, Mario Limonciello, Mark Pearson
On 7/28/22 20:20, Shyam Sundar S K wrote:
> The transition to auto-mode happens when the PMF driver receives
> AMT (Auto Mode transition) event. transition logic will reside in the
> PMF driver but the events would come from other supported drivers[1].
>
> The thermal parameters would vary between when a performance "on-lap" mode
> is detected and versus when not. The CQL event would get triggered from
> other drivers, so that PMF driver would adjust the system thermal config
> based on the ACPI inputs.
>
> OEMs can control whether or not to enable AMT or CQL via other supported
> drivers[1] but the actual transition logic resides in the AMD PMF driver.
> When an AMT event is received the automatic mode transition RAPL algorithm
> will run. When a CQL event is received an performance "on-lap" mode will
> be enabled and thermal parameters will be adjusted accordingly.
>
> [1]
> Link: https://git.kernel.org/pub/scm/linux/kernel/git/pdx86/platform-drivers-x86.git/commit/?h=review-hans&id=755b249250df1b612d982f3b702c831b26ecdf73
>
> Cc: Mario Limonciello <mario.limonciello@amd.com>
> Cc: Mark Pearson <markpearson@lenovo.com>
> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> ---
> drivers/platform/x86/amd/pmf/acpi.c | 59 ++++++++++++++++++++++++
> drivers/platform/x86/amd/pmf/auto-mode.c | 22 +++++++++
> drivers/platform/x86/amd/pmf/core.c | 32 +++++++++++++
> drivers/platform/x86/amd/pmf/pmf.h | 22 +++++++++
> 4 files changed, 135 insertions(+)
>
> diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
> index 724150ec58fb..dedaebf18d88 100644
> --- a/drivers/platform/x86/amd/pmf/acpi.c
> +++ b/drivers/platform/x86/amd/pmf/acpi.c
> @@ -12,6 +12,8 @@
> #include "pmf.h"
>
> #define APMF_METHOD "\\_SB_.PMF_.APMF"
> +#define APMF_CQL_NOTIFICATION 2
> +#define APMF_AMT_NOTIFICATION 3
>
> static unsigned long supported_func;
>
> @@ -99,6 +101,7 @@ static void apmf_if_parse_functions(struct apmf_if_functions *func, u32 mask)
> {
> func->system_params = mask & APMF_FUNC_GET_SYS_PARAMS;
> func->static_slider_granular = mask & APMF_FUNC_STATIC_SLIDER_GRANULAR;
> + func->sbios_requests = mask & APMF_FUNC_SBIOS_REQUESTS;
> func->auto_mode_def = mask & APMF_FUNC_AUTO_MODE;
> func->fan_table_idx = mask & APMF_FUNC_SET_FAN_IDX;
> }
> @@ -167,6 +170,44 @@ int apmf_get_auto_mode_def(struct apmf_if *apmf_if, struct apmf_auto_mode *data)
> data, sizeof(*data));
> }
>
> +int apmf_get_sbios_requests(struct apmf_if *apmf_if, struct apmf_sbios_req *req)
> +{
> + return apmf_if_call_store_buffer(apmf_if, APMF_FUNC_SBIOS_REQUESTS,
> + req, sizeof(*req));
> +}
> +
> +static void apmf_event_handler(acpi_handle handle, u32 event, void *data)
> +{
> + struct amd_pmf_dev *pmf_dev = data;
> + struct apmf_if *apmf_if = pmf_dev->apmf_if;
> + int ret;
> +
> + if (apmf_if->func.sbios_requests) {
> + struct apmf_sbios_req req;
> +
> + ret = apmf_get_sbios_requests(apmf_if, &req);
> + if (ret) {
> + dev_err(pmf_dev->dev, "Failed to get SBIOS requests:%d\n", ret);
> + return;
> + }
> +
> + if (req.pending_req & BIT(APMF_AMT_NOTIFICATION)) {
> + pr_debug("PMF: AMT is supported and notifications %s\n",
> + req.amt_event ? "Enabled" : "Disabled");
> + pmf_dev->is_amt_event = !!req.amt_event;
> + }
> +
> + if (req.pending_req & BIT(APMF_CQL_NOTIFICATION)) {
> + pr_debug("PMF: CQL is supported and notifications %s\n",
> + req.cql_event ? "Enabled" : "Disabled");
> + pmf_dev->is_cql_event = !!req.cql_event;
> +
> + /* update the target mode information */
> + amd_pmf_update_2_cql(pmf_dev);
I forgot to mention that this needs an "if (pmf_dev->amt_enabled)" guard so
that the handler does not mess with the limits while AMT is disabled, even
if CQL events are received while AMT is disabled.
> + }
> + }
> +}
> +
> static int apmf_if_verify_interface(struct amd_pmf_dev *pdev, struct apmf_if *apmf_if)
> {
> struct apmf_verify_interface output;
> @@ -211,12 +252,19 @@ static int apmf_get_system_params(struct apmf_if *apmf_if)
>
> void apmf_acpi_deinit(struct amd_pmf_dev *pmf_dev)
> {
> + acpi_handle ahandle = ACPI_HANDLE(pmf_dev->dev);
> +
> if (pmf_dev->apmf_if->func.sbios_heartbeat)
> cancel_delayed_work_sync(&pmf_dev->heart_beat);
> +
> + if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE))
> + acpi_remove_notify_handler(ahandle, ACPI_ALL_NOTIFY,
> + apmf_event_handler);
> }
>
> int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
> {
> + acpi_handle ahandle = ACPI_HANDLE(pmf_dev->dev);
> struct apmf_notification_cfg *n;
> acpi_handle apmf_if_handle;
> struct apmf_if *apmf_if;
> @@ -256,6 +304,17 @@ int apmf_acpi_init(struct amd_pmf_dev *pmf_dev)
> schedule_delayed_work(&pmf_dev->heart_beat, 0);
> }
>
> + /* Install the APMF Notify handler */
> + if (is_apmf_func_supported(APMF_FUNC_AUTO_MODE)) {
> + status = acpi_install_notify_handler(ahandle,
> + ACPI_ALL_NOTIFY,
> + apmf_event_handler, pmf_dev);
> + if (ACPI_FAILURE(status)) {
> + dev_err(pmf_dev->dev, "failed to install notify handler\n");
> + return -ENODEV;
> + }
> + }
> +
> out:
> return ret;
> }
> diff --git a/drivers/platform/x86/amd/pmf/auto-mode.c b/drivers/platform/x86/amd/pmf/auto-mode.c
> index 954fde25e71e..a85ef4b95cdb 100644
> --- a/drivers/platform/x86/amd/pmf/auto-mode.c
> +++ b/drivers/platform/x86/amd/pmf/auto-mode.c
> @@ -111,6 +111,13 @@ void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t t
> bool update = false;
> int i, j;
>
> + if (!dev->amt_running && dev->is_amt_event) {
> + dev_dbg(dev->dev, "setting AMT thermals\n");
> + amd_pmf_handle_automode(dev, SLIDER_OP_SET, config_store.current_mode, NULL);
> + dev->amt_running = true;
> + return;
> + }
> +
> /* Get the average moving average computed by auto mode algorithm */
> avg_power = amd_pmf_get_moving_avg(socket_power);
>
> @@ -161,6 +168,21 @@ void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t t
> }
> }
>
> +void amd_pmf_update_2_cql(struct amd_pmf_dev *dev)
> +{
> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode =
> + dev->is_cql_event ? AUTO_PERFORMANCE_ON_LAP : AUTO_PERFORMANCE;
> + if ((config_store.current_mode == AUTO_PERFORMANCE ||
> + config_store.current_mode == AUTO_PERFORMANCE_ON_LAP) &&
> + config_store.current_mode !=
> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode) {
> + config_store.current_mode =
> + config_store.transition[AUTO_TRANSITION_TO_PERFORMANCE].target_mode;
> + amd_pmf_handle_automode(dev, SLIDER_OP_SET, config_store.current_mode, NULL);
> + }
> + dev_dbg(dev->dev, "updated CQL thermals\n");
> +}
> +
> static void amd_pmf_get_power_threshold(void)
> {
> config_store.transition[AUTO_TRANSITION_TO_QUIET].power_threshold =
> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
> index ef71f8326248..bfae779ccc23 100644
> --- a/drivers/platform/x86/amd/pmf/core.c
> +++ b/drivers/platform/x86/amd/pmf/core.c
> @@ -93,6 +93,27 @@ int amd_pmf_get_power_source(void)
> return POWER_SOURCE_DC;
> }
>
> +static void amd_pmf_reset_amt(struct amd_pmf_dev *dev)
> +{
> + pr_debug("resetting AMT thermals\n");
> + dev->amt_running = false;
> +
> + /*
> + * OEM BIOS implementation guide says that if the auto mode is enabled
> + * the platform_profile registration shall be done by the OEM driver.
> + * There could be cases where both static slider and auto mode BIOS
> + * functions are enabled and we could end up in a race. To avoid that
> + * add a protection and touch the static slider only if that's enabled
> + * from the BIOS side.
> + */
> +
> + if (is_apmf_func_supported(APMF_FUNC_STATIC_SLIDER_GRANULAR)) {
> + u8 mode = amd_pmf_get_pprof_modes(dev);
> +
> + amd_pmf_update_slider(dev, SLIDER_OP_SET, mode, NULL);
> + }
> +}
> +
> static void amd_pmf_get_metrics(struct work_struct *work)
> {
> struct amd_pmf_dev *dev = container_of(work, struct amd_pmf_dev, work_buffer.work);
> @@ -103,6 +124,9 @@ static void amd_pmf_get_metrics(struct work_struct *work)
> /* Get the current profile information */
> current_profile = READ_ONCE(dev->current_profile);
>
> + if (!dev->is_amt_event)
> + dev_dbg(dev->dev, "amt event: %s\n", dev->is_amt_event ? "Enabled" : "Disabled");
> +
> /* Transfer table contents */
> memset(&dev->m_table, 0, sizeof(dev->m_table));
> amd_pmf_send_cmd(dev, SET_TRANSFER_TABLE, 0, 7, NULL);
> @@ -112,6 +136,14 @@ static void amd_pmf_get_metrics(struct work_struct *work)
> /* Calculate the avg SoC power consumption */
> socket_power = dev->m_table.apu_power + dev->m_table.dgpu_power;
>
> + if (dev->is_amt_event) {
> + /* Apply the Auto Mode transition */
> + amd_pmf_trans_automode(dev, socket_power, time_elapsed_ms);
> + } else if (!dev->is_amt_event && dev->amt_running) {
> + /* reset to other mode if we receive a AMT disable event */
> + amd_pmf_reset_amt(dev);
> + }
> +
> dev->start_time = ktime_to_ms(ktime_get());
> schedule_delayed_work(&dev->work_buffer, msecs_to_jiffies(metrics_table_loop_ms));
> }
> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
> index 4618ddc5a662..0f8b25524845 100644
> --- a/drivers/platform/x86/amd/pmf/pmf.h
> +++ b/drivers/platform/x86/amd/pmf/pmf.h
> @@ -17,6 +17,7 @@
> /* APMF Functions */
> #define APMF_FUNC_VERIFY_INTERFACE 0
> #define APMF_FUNC_GET_SYS_PARAMS 1
> +#define APMF_FUNC_SBIOS_REQUESTS 2
> #define APMF_FUNC_SBIOS_HEARTBEAT 4
> #define APMF_FUNC_AUTO_MODE 5
> #define APMF_FUNC_SET_FAN_IDX 7
> @@ -49,6 +50,7 @@
> /* AMD PMF BIOS interfaces */
> struct apmf_if_functions {
> bool system_params;
> + bool sbios_requests;
> bool sbios_heartbeat;
> bool auto_mode_def;
> bool fan_table_idx;
> @@ -80,6 +82,21 @@ struct apmf_system_params {
> u32 heartbeat_int;
> } __packed;
>
> +struct apmf_sbios_req {
> + u16 size;
> + u32 pending_req;
> + u8 rsd;
> + u8 cql_event;
> + u8 amt_event;
> + u32 fppt;
> + u32 sppt;
> + u32 fppt_apu_only;
> + u32 spl;
> + u32 stt_min_limit;
> + u8 skin_temp_apu;
> + u8 skin_temp_hs2;
> +} __packed;
> +
> struct apmf_fan_idx {
> u16 size;
> u8 fan_ctl_mode;
> @@ -161,6 +178,9 @@ struct amd_pmf_dev {
> struct smu_pmf_metrics m_table;
> struct delayed_work work_buffer;
> ktime_t start_time;
> + bool is_amt_event;
> + bool is_cql_event;
> + bool amt_running;
> };
>
> struct apmf_sps_prop_granular {
> @@ -320,5 +340,7 @@ int apmf_get_auto_mode_def(struct apmf_if *ampf_if, struct apmf_auto_mode *data)
> void amd_pmf_init_auto_mode(struct amd_pmf_dev *dev);
> void amd_pmf_deinit_auto_mode(struct amd_pmf_dev *dev);
> void amd_pmf_trans_automode(struct amd_pmf_dev *dev, int socket_power, ktime_t time_elapsed_ms);
> +int apmf_get_sbios_requests(struct apmf_if *ampf_if, struct apmf_sbios_req *req);
> +void amd_pmf_update_2_cql(struct amd_pmf_dev *dev);
>
> #endif /* PMF_H */
^ permalink raw reply [flat|nested] 33+ messages in thread
* [PATCH v2 RESEND 10/11] platform/x86/amd/pmf: Force load driver on older supported platforms
2022-07-28 18:20 [PATCH v2 RESEND 00/11] platform/x86/amd/pmf: Introduce AMD PMF Driver Shyam Sundar S K
` (8 preceding siblings ...)
2022-07-28 18:20 ` [PATCH v2 RESEND 09/11] platform/x86/amd/pmf: Handle AMT and CQL events for Auto mode Shyam Sundar S K
@ 2022-07-28 18:20 ` Shyam Sundar S K
2022-08-01 13:58 ` Hans de Goede
2022-07-28 18:20 ` [PATCH v2 RESEND 11/11] MAINTAINERS: Add AMD PMF driver entry Shyam Sundar S K
10 siblings, 1 reply; 33+ messages in thread
From: Shyam Sundar S K @ 2022-07-28 18:20 UTC (permalink / raw)
To: hdegoede, markgross; +Cc: platform-driver-x86, Patil.Reddy, Shyam Sundar S K
Some of the older platforms with _HID "AMDI0100" PMF driver can be force
loaded by interested users but only for experimental purposes.
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
---
drivers/platform/x86/amd/pmf/core.c | 16 ++++++++++++++++
1 file changed, 16 insertions(+)
diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
index bfae779ccc23..c0e3780ae3d1 100644
--- a/drivers/platform/x86/amd/pmf/core.c
+++ b/drivers/platform/x86/amd/pmf/core.c
@@ -39,6 +39,7 @@
#define AMD_PMF_RESULT_FAILED 0xFF
/* List of supported CPU ids */
+#define AMD_CPU_ID_RMB 0x14b5
#define AMD_CPU_ID_PS 0x14e8
#define PMF_MSG_DELAY_MIN_US 50
@@ -52,6 +53,11 @@ static int metrics_table_loop_ms = 1000;
module_param(metrics_table_loop_ms, int, 0644);
MODULE_PARM_DESC(metrics_table_loop_ms, " Metrics Table sample size time (default = 1000ms) ");
+/* Force load on supported older platforms */
+static bool force_load;
+module_param(force_load, bool, 0444);
+MODULE_PARM_DESC(force_load, " Force load this driver on supported older platforms (experimental) ");
+
static int current_power_limits_show(struct seq_file *seq, void *unused)
{
struct amd_pmf_dev *dev = seq->private;
@@ -237,6 +243,7 @@ int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32
}
static const struct pci_device_id pmf_pci_ids[] = {
+ { PCI_DEVICE(PCI_VENDOR_ID_AMD, AMD_CPU_ID_RMB) },
{ PCI_DEVICE(PCI_VENDOR_ID_AMD, AMD_CPU_ID_PS) },
{ }
};
@@ -294,6 +301,7 @@ static void amd_pmf_deinit_features(struct amd_pmf_dev *dev)
}
static const struct acpi_device_id amd_pmf_acpi_ids[] = {
+ {"AMDI0100", 0x100},
{"AMDI0102", 0},
{ }
};
@@ -301,6 +309,7 @@ MODULE_DEVICE_TABLE(acpi, amd_pmf_acpi_ids);
static int amd_pmf_probe(struct platform_device *pdev)
{
+ const struct acpi_device_id *id;
struct amd_pmf_dev *dev;
struct pci_dev *rdev;
u32 base_addr_lo;
@@ -309,6 +318,13 @@ static int amd_pmf_probe(struct platform_device *pdev)
u32 val;
int err;
+ id = acpi_match_device(amd_pmf_acpi_ids, &pdev->dev);
+ if (!id)
+ return -ENODEV;
+
+ if (id->driver_data == 0x100 && !force_load)
+ return -ENODEV;
+
dev = devm_kzalloc(&pdev->dev, sizeof(*dev), GFP_KERNEL);
if (!dev)
return -ENOMEM;
--
2.25.1
^ permalink raw reply related [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 10/11] platform/x86/amd/pmf: Force load driver on older supported platforms
2022-07-28 18:20 ` [PATCH v2 RESEND 10/11] platform/x86/amd/pmf: Force load driver on older supported platforms Shyam Sundar S K
@ 2022-08-01 13:58 ` Hans de Goede
0 siblings, 0 replies; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 13:58 UTC (permalink / raw)
To: Shyam Sundar S K, markgross; +Cc: platform-driver-x86, Patil.Reddy
Hi,
On 7/28/22 20:20, Shyam Sundar S K wrote:
> Some of the older platforms with _HID "AMDI0100" PMF driver can be force
> loaded by interested users but only for experimental purposes.
>
> Reviewed-by: Hans de Goede <hdegoede@redhat.com>
> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
This patch still looks good to me.
Regards,
Hans
> ---
> drivers/platform/x86/amd/pmf/core.c | 16 ++++++++++++++++
> 1 file changed, 16 insertions(+)
>
> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
> index bfae779ccc23..c0e3780ae3d1 100644
> --- a/drivers/platform/x86/amd/pmf/core.c
> +++ b/drivers/platform/x86/amd/pmf/core.c
> @@ -39,6 +39,7 @@
> #define AMD_PMF_RESULT_FAILED 0xFF
>
> /* List of supported CPU ids */
> +#define AMD_CPU_ID_RMB 0x14b5
> #define AMD_CPU_ID_PS 0x14e8
>
> #define PMF_MSG_DELAY_MIN_US 50
> @@ -52,6 +53,11 @@ static int metrics_table_loop_ms = 1000;
> module_param(metrics_table_loop_ms, int, 0644);
> MODULE_PARM_DESC(metrics_table_loop_ms, " Metrics Table sample size time (default = 1000ms) ");
>
> +/* Force load on supported older platforms */
> +static bool force_load;
> +module_param(force_load, bool, 0444);
> +MODULE_PARM_DESC(force_load, " Force load this driver on supported older platforms (experimental) ");
> +
> static int current_power_limits_show(struct seq_file *seq, void *unused)
> {
> struct amd_pmf_dev *dev = seq->private;
> @@ -237,6 +243,7 @@ int amd_pmf_send_cmd(struct amd_pmf_dev *dev, u8 message, bool get, u32 arg, u32
> }
>
> static const struct pci_device_id pmf_pci_ids[] = {
> + { PCI_DEVICE(PCI_VENDOR_ID_AMD, AMD_CPU_ID_RMB) },
> { PCI_DEVICE(PCI_VENDOR_ID_AMD, AMD_CPU_ID_PS) },
> { }
> };
> @@ -294,6 +301,7 @@ static void amd_pmf_deinit_features(struct amd_pmf_dev *dev)
> }
>
> static const struct acpi_device_id amd_pmf_acpi_ids[] = {
> + {"AMDI0100", 0x100},
> {"AMDI0102", 0},
> { }
> };
> @@ -301,6 +309,7 @@ MODULE_DEVICE_TABLE(acpi, amd_pmf_acpi_ids);
>
> static int amd_pmf_probe(struct platform_device *pdev)
> {
> + const struct acpi_device_id *id;
> struct amd_pmf_dev *dev;
> struct pci_dev *rdev;
> u32 base_addr_lo;
> @@ -309,6 +318,13 @@ static int amd_pmf_probe(struct platform_device *pdev)
> u32 val;
> int err;
>
> + id = acpi_match_device(amd_pmf_acpi_ids, &pdev->dev);
> + if (!id)
> + return -ENODEV;
> +
> + if (id->driver_data == 0x100 && !force_load)
> + return -ENODEV;
> +
> dev = devm_kzalloc(&pdev->dev, sizeof(*dev), GFP_KERNEL);
> if (!dev)
> return -ENOMEM;
^ permalink raw reply [flat|nested] 33+ messages in thread
* [PATCH v2 RESEND 11/11] MAINTAINERS: Add AMD PMF driver entry
2022-07-28 18:20 [PATCH v2 RESEND 00/11] platform/x86/amd/pmf: Introduce AMD PMF Driver Shyam Sundar S K
` (9 preceding siblings ...)
2022-07-28 18:20 ` [PATCH v2 RESEND 10/11] platform/x86/amd/pmf: Force load driver on older supported platforms Shyam Sundar S K
@ 2022-07-28 18:20 ` Shyam Sundar S K
2022-08-01 13:58 ` Hans de Goede
10 siblings, 1 reply; 33+ messages in thread
From: Shyam Sundar S K @ 2022-07-28 18:20 UTC (permalink / raw)
To: hdegoede, markgross; +Cc: platform-driver-x86, Patil.Reddy, Shyam Sundar S K
Update the MAINTAINERS file with AMD PMF driver details.
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
---
MAINTAINERS | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/MAINTAINERS b/MAINTAINERS
index e01478062c56..d3f6cabcaab2 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -998,6 +998,13 @@ L: platform-driver-x86@vger.kernel.org
S: Maintained
F: drivers/platform/x86/amd/pmc.c
+AMD PMF DRIVER
+M: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
+L: platform-driver-x86@vger.kernel.org
+S: Maintained
+F: Documentation/ABI/testing/sysfs-amd-pmf
+F: drivers/platform/x86/amd/pmf/
+
AMD HSMP DRIVER
M: Naveen Krishna Chatradhi <naveenkrishna.chatradhi@amd.com>
R: Carlos Bilbao <carlos.bilbao@amd.com>
--
2.25.1
^ permalink raw reply related [flat|nested] 33+ messages in thread* Re: [PATCH v2 RESEND 11/11] MAINTAINERS: Add AMD PMF driver entry
2022-07-28 18:20 ` [PATCH v2 RESEND 11/11] MAINTAINERS: Add AMD PMF driver entry Shyam Sundar S K
@ 2022-08-01 13:58 ` Hans de Goede
0 siblings, 0 replies; 33+ messages in thread
From: Hans de Goede @ 2022-08-01 13:58 UTC (permalink / raw)
To: Shyam Sundar S K, markgross; +Cc: platform-driver-x86, Patil.Reddy
Hi,
On 7/28/22 20:20, Shyam Sundar S K wrote:
> Update the MAINTAINERS file with AMD PMF driver details.
>
> Reviewed-by: Hans de Goede <hdegoede@redhat.com>
> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
This patch still looks good to me.
Regards,
Hans
> ---
> MAINTAINERS | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/MAINTAINERS b/MAINTAINERS
> index e01478062c56..d3f6cabcaab2 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -998,6 +998,13 @@ L: platform-driver-x86@vger.kernel.org
> S: Maintained
> F: drivers/platform/x86/amd/pmc.c
>
> +AMD PMF DRIVER
> +M: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> +L: platform-driver-x86@vger.kernel.org
> +S: Maintained
> +F: Documentation/ABI/testing/sysfs-amd-pmf
> +F: drivers/platform/x86/amd/pmf/
> +
> AMD HSMP DRIVER
> M: Naveen Krishna Chatradhi <naveenkrishna.chatradhi@amd.com>
> R: Carlos Bilbao <carlos.bilbao@amd.com>
^ permalink raw reply [flat|nested] 33+ messages in thread