Skip to contentSkip to conversation

misc: Move AMD side band interface(SBI) functionality

PATCH Superseded Akshay Gupta 4 participants 17 messages Mar 3, 2025 – Mar 7, 2025

Subsystems (3)

v5 of 9 Superseded Latest is v9 →

lore
Share & tools
lore
#0 lore raw

v5 cover misc: Move AMD side band interface(SBI) functionality

At present, sbrmi driver under hwmon subsystem, is probed as an i2c driver, fetches data using APML specified protocol and reports through hwmon power sensor.

AMD provides additional information using custom protocols, which cannot be enumerated as hwmon sensors. Hence, move the existing functionality from hwmon/ to misc/ and add support for following custom protocols
- read Processor feature capabilities and configuration information through side band.
- read Machine Check Architecture(MCA) registers over sideband. The information is accessed for range of MCA registers by passing register address and thread ID to the protocol.

NOTE: AMD defines Advanced Platform Management Link (APML) interface which provides system management functionality access to the baseboard management controller (BMC).

This patchset is an attempt to keep all APML core functionality in one place, provide hwmon and IOCTL interface to user space
1. [Patch 1] Move the i2c client probe, hwmon sensors and sbrmi core functionality from drivers/hwmon to drivers/misc/
2. [Patch 2] Move sbrmi core functionality to new core file to export core functionality
3. [Patch 3] Move hwmon device sensor as separate entity
4. [Patch 4] Convert i2c to regmap which provides multiple benefits
over direct smbus APIs.
a. i2c/i3c support and
b. 1 byte/2 byte RMI register size addressing
5. [Patch 5] Optimize wait condition with regmap API regmap_read_poll_timeout as per suggestion from Arnd
6. [Patch 6, 7] Register a misc device which provides
a. An ioctl interface through node /dev/sbrmiX
b. Register sets is common across APML protocols. IOCTL is providing
synchronization among protocols as transactions may create
race condition.
7. [Subsequent patches 8, 9 and 10] add support for AMD custom protocols a. CPUID
b. MCAMSR
c. Register xfer
8. [Patch 11] AMD side band description document

Open-sourced and widely used [1]_ will continue to provide user-space programmable API.

.. [1] https://github.com/amd/esmi_oob_library

Akshay Gupta (11):
hwmon/misc: amd-sbi: Move core sbrmi from hwmon to misc
misc: amd-sbi: Move protocol functionality to core file
misc: amd-sbi: Move hwmon device sensor as separate entity
misc: amd-sbi: Use regmap subsystem
misc: amd-sbi: Optimize the wait condition for mailbox command
completion
misc: amd-sbi: Add support for AMD_SBI IOCTL
misc: amd-sbi: Add support for mailbox error codes
misc: amd-sbi: Add support for CPUID protocol
misc: amd-sbi: Add support for read MCA register protocol
misc: amd-sbi: Add support for register xfer
misc: amd-sbi: Add document for AMD SB IOCTL description

 Documentation/misc-devices/amd-sbi.rst        |  87 ++++
 Documentation/misc-devices/index.rst          |   1 +
 .../userspace-api/ioctl/ioctl-number.rst      |   2 +
 drivers/hwmon/Kconfig                         |  10 -
 drivers/hwmon/sbrmi.c                         | 357 --------------
 drivers/misc/Kconfig                          |   1 +
 drivers/misc/Makefile                         |   1 +
 drivers/misc/amd-sbi/Kconfig                  |   9 +
 drivers/misc/amd-sbi/Makefile                 |   3 +
 drivers/misc/amd-sbi/rmi-core.c               | 445 ++++++++++++++++++
 drivers/misc/amd-sbi/rmi-core.h               |  67 +++
 drivers/misc/amd-sbi/rmi-hwmon.c              | 122 +++++
 drivers/misc/amd-sbi/rmi-i2c.c                | 133 ++++++
 include/uapi/misc/amd-apml.h                  |  97 ++++
 14 files changed, 968 insertions(+), 367 deletions(-)
 create mode 100644 Documentation/misc-devices/amd-sbi.rst
 delete mode 100644 drivers/hwmon/sbrmi.c
 create mode 100644 drivers/misc/amd-sbi/Kconfig
 create mode 100644 drivers/misc/amd-sbi/Makefile
 create mode 100644 drivers/misc/amd-sbi/rmi-core.c
 create mode 100644 drivers/misc/amd-sbi/rmi-core.h
 create mode 100644 drivers/misc/amd-sbi/rmi-hwmon.c
 create mode 100644 drivers/misc/amd-sbi/rmi-i2c.c
 create mode 100644 include/uapi/misc/amd-apml.h
Signature · 1 line
2.25.1
#1 lore raw

PATCH v5 1/11 hwmon/misc: amd-sbi: Move core sbrmi from hwmon to misc

This is done to support other functionality provided by the SBRMI, which does not fit in the hwmon subsystem.

---
Changes since v4:
- Split in 3 new patches as per review comments
1. Bring drivers/hwmon to drivers/misc/amd-sbi
2. Move out the core functionality
3. Move out the hwmon functionality.

Changes since v3:
Rebase the patch
Added Acked-by

Changes since v2:
Rebase the patch

Changes since v1:
- File name update
- Add hwmon sensor registration in this patch
- Update Copyright year

 drivers/hwmon/Kconfig                   | 10 ----------
 drivers/misc/Kconfig                    |  1 +
 drivers/misc/Makefile                   |  1 +
 drivers/misc/amd-sbi/Kconfig            |  9 +++++++++
 drivers/misc/amd-sbi/Makefile           |  2 ++
 drivers/{hwmon => misc/amd-sbi}/sbrmi.c |  0
 6 files changed, 13 insertions(+), 10 deletions(-)
 create mode 100644 drivers/misc/amd-sbi/Kconfig
 create mode 100644 drivers/misc/amd-sbi/Makefile
 rename drivers/{hwmon => misc/amd-sbi}/sbrmi.c (100%)
drivers/hwmon/Kconfig +0 -10
index 4cbaba15d86e..5b53137f8b89 100644
@@ -1866,16 +1866,6 @@ config SENSORS_SBTSI
 	  This driver can also be built as a module. If so, the module will
 	  be called sbtsi_temp.
 
-config SENSORS_SBRMI
-	tristate "Emulated SB-RMI sensor"
-	depends on I2C
-	help
-	  If you say yes here you get support for emulated RMI
-	  sensors on AMD SoCs with APML interface connected to a BMC device.
-
-	  This driver can also be built as a module. If so, the module will
-	  be called sbrmi.
-
 config SENSORS_SHT15
 	tristate "Sensiron humidity and temperature sensors. SHT15 and compat."
 	depends on GPIOLIB || COMPILE_TEST
drivers/misc/Kconfig +1 -0
index 56bc72c7ce4a..b2792b9664ba 100644
@@ -649,4 +649,5 @@ source "drivers/misc/uacce/Kconfig"
 source "drivers/misc/pvpanic/Kconfig"
 source "drivers/misc/mchp_pci1xxxx/Kconfig"
 source "drivers/misc/keba/Kconfig"
+source "drivers/misc/amd-sbi/Kconfig"
 endmenu
drivers/misc/Makefile +1 -0
index 545aad06d088..fde23b0b4f0e 100644
@@ -75,3 +75,4 @@ lan966x-pci-objs		:= lan966x_pci.o
 lan966x-pci-objs		+= lan966x_pci.dtbo.o
 obj-$(CONFIG_MCHP_LAN966X_PCI)	+= lan966x-pci.o
 obj-y				+= keba/
+obj-y				+= amd-sbi/
drivers/misc/amd-sbi/Kconfig new file +9 -0
new file mode 100644
index 000000000000..be2d9e495eb7
@@ -0,0 +1,9 @@
+# SPDX-License-Identifier: GPL-2.0-only
+config AMD_SBRMI_I2C
+        tristate "AMD side band RMI support"
+        depends on I2C
+        help
+          Side band RMI over I2C support for AMD out of band management.
+
+	  This driver can also be built as a module. If so, the module will
+	  be called sbrmi-i2c.
drivers/misc/amd-sbi/Makefile new file +2 -0
new file mode 100644
index 000000000000..304394bf5e59
@@ -0,0 +1,2 @@
+# SPDX-License-Identifier: GPL-2.0-only
+obj-$(CONFIG_AMD_SBRMI_I2C)	+= sbrmi.o
drivers/misc/amd-sbi/sbrmi.c renamed from drivers/hwmon/sbrmi.c +0 -0
similarity index 100%
rename from drivers/hwmon/sbrmi.c
rename to drivers/misc/amd-sbi/sbrmi.c
Signature · 1 line
2.25.1
#3 lore raw

PATCH v5 2/11 misc: amd-sbi: Move protocol functionality to core file

In reply to Akshay Gupta

- This is done to utilize the protocol functionality into other domains.
- Increase the scalability of the module with different bus(i2c/i3c)

---

Changes since v4:
New patch: Patch split from v4 patch 1/9
- Update Copyright year to 2025

 drivers/misc/amd-sbi/Makefile               |   3 +-
 drivers/misc/amd-sbi/rmi-core.c             | 113 ++++++++++++++
 drivers/misc/amd-sbi/rmi-core.h             |  63 ++++++++
 drivers/misc/amd-sbi/{sbrmi.c => rmi-i2c.c} | 165 ++------------------
 4 files changed, 187 insertions(+), 157 deletions(-)
 create mode 100644 drivers/misc/amd-sbi/rmi-core.c
 create mode 100644 drivers/misc/amd-sbi/rmi-core.h
 rename drivers/misc/amd-sbi/{sbrmi.c => rmi-i2c.c} (53%)
drivers/misc/amd-sbi/Makefile +2 -1
index 304394bf5e59..7cd8e0a1aa5d 100644
@@ -1,2 +1,3 @@
 # SPDX-License-Identifier: GPL-2.0-only
-obj-$(CONFIG_AMD_SBRMI_I2C)	+= sbrmi.o
+sbrmi-i2c-objs  		:= rmi-i2c.o rmi-core.o
+obj-$(CONFIG_AMD_SBRMI_I2C)	+= sbrmi-i2c.o
drivers/misc/amd-sbi/rmi-core.c new file +113 -0

113 lines not on this page · raw message

drivers/misc/amd-sbi/rmi-core.h new file +63 -0

63 lines not on this page · raw message

drivers/misc/amd-sbi/rmi-i2c.c renamed from drivers/misc/amd-sbi/sbrmi.c +9 -156

211 lines not on this page · raw message

Signature · 1 line
2.25.1
#4 lore raw

Re: [PATCH v5 02/11] misc: amd-sbi: Move protocol functionality to core file

kernel test robot noticed the following build errors:

Hi Akshay,

kernel test robot noticed the following build errors:

[auto build test ERROR on char-misc/char-misc-testing]
[also build test ERROR on char-misc/char-misc-next char-misc/char-misc-linus groeck-staging/hwmon-next soc/for-next linus/master v6.14-rc5 next-20250305] [If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Akshay-Gupta/hwmon-misc-amd-sbi-Move-core-sbrmi-from-hwmon-to-misc/20250303-190830
base:   char-misc/char-misc-testing
patch link:    https://lore.kernel.org/r/20250303105902.215009-3-akshay.gupta%40amd.com
patch subject: [PATCH v5 02/11] misc: amd-sbi: Move protocol functionality to core file
config: i386-randconfig-004-20250306 (https://download.01.org/0day-ci/archive/20250306/[email protected]/config)
compiler: gcc-12 (Debian 12.2.0-14) 12.2.0
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20250306/[email protected]/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

All errors (new ones prefixed by >>, old ones prefixed by <<):

WARNING: modpost: missing MODULE_DESCRIPTION() in lib/slub_kunit.o

>> ERROR: modpost: "devm_hwmon_device_register_with_info" [drivers/misc/amd-sbi/sbrmi-i2c.ko] undefined!
Signature · 2 lines
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
#5 lore raw

PATCH v5 3/11 misc: amd-sbi: Move hwmon device sensor as separate entity

In reply to Akshay Gupta

- Move hwmon device sensor to misc as only power is reported through hwmon sensor.

---
Changes since v4:
New patch: Patch split from v4 patch 1/9
- Update Copyright year to 2025

 drivers/misc/amd-sbi/Makefile    |   2 +-
 drivers/misc/amd-sbi/rmi-core.h  |   1 +
 drivers/misc/amd-sbi/rmi-hwmon.c | 121 +++++++++++++++++++++++++++++++
 drivers/misc/amd-sbi/rmi-i2c.c   | 105 +--------------------------
 4 files changed, 125 insertions(+), 104 deletions(-)
 create mode 100644 drivers/misc/amd-sbi/rmi-hwmon.c
drivers/misc/amd-sbi/Makefile +1 -1
index 7cd8e0a1aa5d..eac90a7635de 100644
@@ -1,3 +1,3 @@
 # SPDX-License-Identifier: GPL-2.0-only
-sbrmi-i2c-objs  		:= rmi-i2c.o rmi-core.o
+sbrmi-i2c-objs  		:= rmi-i2c.o rmi-core.o rmi-hwmon.o
 obj-$(CONFIG_AMD_SBRMI_I2C)	+= sbrmi-i2c.o
drivers/misc/amd-sbi/rmi-core.h +1 -0
index 8e30a43ec714..27165d3bb20f 100644
@@ -60,4 +60,5 @@ struct sbrmi_mailbox_msg {
 };
 
 int rmi_mailbox_xfer(struct sbrmi_data *data, struct sbrmi_mailbox_msg *msg);
+int create_hwmon_sensor_device(struct device *dev, struct sbrmi_data *data);
 #endif /*_SBRMI_CORE_H_*/
drivers/misc/amd-sbi/rmi-hwmon.c new file +121 -0

121 lines not on this page · raw message

drivers/misc/amd-sbi/rmi-i2c.c +2 -103

135 lines not on this page · raw message

Signature · 1 line
2.25.1
#6 lore raw

Re: [PATCH v5 03/11] misc: amd-sbi: Move hwmon device sensor as separate entity

kernel test robot noticed the following build errors:

Hi Akshay,

kernel test robot noticed the following build errors:

[auto build test ERROR on char-misc/char-misc-testing]
[also build test ERROR on char-misc/char-misc-next char-misc/char-misc-linus groeck-staging/hwmon-next soc/for-next linus/master v6.14-rc5 next-20250305] [If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Akshay-Gupta/hwmon-misc-amd-sbi-Move-core-sbrmi-from-hwmon-to-misc/20250303-190830
base:   char-misc/char-misc-testing
patch link:    https://lore.kernel.org/r/20250303105902.215009-4-akshay.gupta%40amd.com
patch subject: [PATCH v5 03/11] misc: amd-sbi: Move hwmon device sensor as separate entity
config: x86_64-randconfig-078-20250306 (https://download.01.org/0day-ci/archive/20250306/[email protected]/config)
compiler: gcc-12 (Debian 12.2.0-14) 12.2.0
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20250306/[email protected]/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

All errors (new ones prefixed by >>):

ld: drivers/misc/amd-sbi/rmi-hwmon.o: in function `create_hwmon_sensor_device':

>> drivers/misc/amd-sbi/rmi-hwmon.c:118: undefined reference to `devm_hwmon_device_register_with_info'

vim +118 drivers/misc/amd-sbi/rmi-hwmon.c

113
114 int create_hwmon_sensor_device(struct device *dev, struct sbrmi_data *data)
115 {
116 struct device *hwmon_dev;
117

 > 118		hwmon_dev = devm_hwmon_device_register_with_info(dev, "sbrmi", data,
Signature · 2 lines
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
#7 lore raw

PATCH v5 4/11 misc: amd-sbi: Use regmap subsystem

In reply to Akshay Gupta

- regmap subsystem provides multiple benefits over direct smbus APIs
- subsystem adds another abstraction layer on top of struct i2c_client to make it easy to read or write registers.
- The subsystem can be helpful in following cases
- Different types of bus (i2c/i3c), we have plans to support i3c.
- Different Register address size (1byte/2byte)

---
Changes since v4:
- Previously patch 2
Changes since v1:
- Previously patch 3
- Remove "__packed" from data structure

 drivers/misc/amd-sbi/rmi-core.c | 29 ++++++++++++-----------------
 drivers/misc/amd-sbi/rmi-core.h |  3 ++-
 drivers/misc/amd-sbi/rmi-i2c.c  | 25 ++++++++++++++++---------
 3 files changed, 30 insertions(+), 27 deletions(-)
drivers/misc/amd-sbi/rmi-core.c +12 -17
index 74456756270c..663ab9176d95 100644
@@ -9,6 +9,7 @@
 #include <linux/err.h>
 #include <linux/i2c.h>
 #include <linux/mutex.h>
+#include <linux/regmap.h>
 #include "rmi-core.h"
 
 /* Mask for Status Register bit[1] */
@@ -21,6 +22,7 @@
 int rmi_mailbox_xfer(struct sbrmi_data *data,
 		     struct sbrmi_mailbox_msg *msg)
 {
+	unsigned int bytes;
 	int i, ret, retry = 10;
 	int sw_status;
 	u8 byte;
@@ -28,14 +30,12 @@ int rmi_mailbox_xfer(struct sbrmi_data *data,
 	mutex_lock(&data->lock);
 
 	/* Indicate firmware a command is to be serviced */
-	ret = i2c_smbus_write_byte_data(data->client,
-					SBRMI_INBNDMSG7, START_CMD);
+	ret = regmap_write(data->regmap, SBRMI_INBNDMSG7, START_CMD);
 	if (ret < 0)
 		goto exit_unlock;
 
 	/* Write the command to SBRMI::InBndMsg_inst0 */
-	ret = i2c_smbus_write_byte_data(data->client,
-					SBRMI_INBNDMSG0, msg->cmd);
+	ret = regmap_write(data->regmap, SBRMI_INBNDMSG0, msg->cmd);
 	if (ret < 0)
 		goto exit_unlock;
 
@@ -46,8 +46,7 @@ int rmi_mailbox_xfer(struct sbrmi_data *data,
 	 */
 	for (i = 0; i < 4; i++) {
 		byte = (msg->data_in >> i * 8) & 0xff;
-		ret = i2c_smbus_write_byte_data(data->client,
-						SBRMI_INBNDMSG1 + i, byte);
+		ret = regmap_write(data->regmap, SBRMI_INBNDMSG1 + i, byte);
 		if (ret < 0)
 			goto exit_unlock;
 	}
@@ -56,8 +55,7 @@ int rmi_mailbox_xfer(struct sbrmi_data *data,
 	 * Write 0x01 to SBRMI::SoftwareInterrupt to notify firmware to
 	 * perform the requested read or write command
 	 */
-	ret = i2c_smbus_write_byte_data(data->client,
-					SBRMI_SW_INTERRUPT, TRIGGER_MAILBOX);
+	ret = regmap_write(data->regmap, SBRMI_SW_INTERRUPT, TRIGGER_MAILBOX);
 	if (ret < 0)
 		goto exit_unlock;
 
@@ -67,8 +65,7 @@ int rmi_mailbox_xfer(struct sbrmi_data *data,
 	 * of the requested command
 	 */
 	do {
-		sw_status = i2c_smbus_read_byte_data(data->client,
-						     SBRMI_STATUS);
+		ret = regmap_read(data->regmap, SBRMI_STATUS, &sw_status);
 		if (sw_status < 0) {
 			ret = sw_status;
 			goto exit_unlock;
@@ -79,8 +76,6 @@ int rmi_mailbox_xfer(struct sbrmi_data *data,
 	} while (retry--);
 
 	if (retry < 0) {
-		dev_err(&data->client->dev,
-			"Firmware fail to indicate command completion\n");
 		ret = -EIO;
 		goto exit_unlock;
 	}
@@ -92,11 +87,11 @@ int rmi_mailbox_xfer(struct sbrmi_data *data,
 	 */
 	if (msg->read) {
 		for (i = 0; i < 4; i++) {
-			ret = i2c_smbus_read_byte_data(data->client,
-						       SBRMI_OUTBNDMSG1 + i);
+			ret = regmap_read(data->regmap,
+					  SBRMI_OUTBNDMSG1 + i, &bytes);
 			if (ret < 0)
 				goto exit_unlock;
-			msg->data_out |= ret << i * 8;
+			msg->data_out |= bytes << i * 8;
 		}
 	}
 
@@ -104,8 +99,8 @@ int rmi_mailbox_xfer(struct sbrmi_data *data,
 	 * BMC must write 1'b1 to SBRMI::Status[SwAlertSts] to clear the
 	 * ALERT to initiator
 	 */
-	ret = i2c_smbus_write_byte_data(data->client, SBRMI_STATUS,
-					sw_status | SW_ALERT_MASK);
+	ret = regmap_write(data->regmap, SBRMI_STATUS,
+			   sw_status | SW_ALERT_MASK);
 
 exit_unlock:
 	mutex_unlock(&data->lock);
drivers/misc/amd-sbi/rmi-core.h +2 -1
index 27165d3bb20f..bbb6bb1cefde 100644
@@ -9,6 +9,7 @@
 #include <linux/mutex.h>
 #include <linux/i2c.h>
 #include <linux/platform_device.h>
+#include <linux/regmap.h>
 
 /* SB-RMI registers */
 enum sbrmi_reg {
@@ -47,7 +48,7 @@ enum sbrmi_msg_id {
 
 /* Each client has this additional data */
 struct sbrmi_data {
-	struct i2c_client *client;
+	struct regmap *regmap;
 	struct mutex lock;
 	u32 pwr_limit_max;
 };
drivers/misc/amd-sbi/rmi-i2c.c +16 -9
index 9ad4c8093399..7a9801273a4c 100644
@@ -13,24 +13,24 @@
 #include <linux/module.h>
 #include <linux/mutex.h>
 #include <linux/of.h>
+#include <linux/regmap.h>
 #include "rmi-core.h"
 
-static int sbrmi_enable_alert(struct i2c_client *client)
+static int sbrmi_enable_alert(struct sbrmi_data *data)
 {
-	int ctrl;
+	int ctrl, ret;
 
 	/*
 	 * Enable the SB-RMI Software alert status
 	 * by writing 0 to bit 4 of Control register(0x1)
 	 */
-	ctrl = i2c_smbus_read_byte_data(client, SBRMI_CTRL);
-	if (ctrl < 0)
-		return ctrl;
+	ret = regmap_read(data->regmap, SBRMI_CTRL, &ctrl);
+	if (ret < 0)
+		return ret;
 
 	if (ctrl & 0x10) {
 		ctrl &= ~0x10;
-		return i2c_smbus_write_byte_data(client,
-						 SBRMI_CTRL, ctrl);
+		return regmap_write(data->regmap, SBRMI_CTRL, ctrl);
 	}
 
 	return 0;
@@ -55,17 +55,24 @@ static int sbrmi_i2c_probe(struct i2c_client *client)
 {
 	struct device *dev = &client->dev;
 	struct sbrmi_data *data;
+	struct regmap_config sbrmi_i2c_regmap_config = {
+		.reg_bits = 8,
+		.val_bits = 8,
+	};
 	int ret;
 
 	data = devm_kzalloc(dev, sizeof(struct sbrmi_data), GFP_KERNEL);
 	if (!data)
 		return -ENOMEM;
 
-	data->client = client;
 	mutex_init(&data->lock);
 
+	data->regmap = devm_regmap_init_i2c(client, &sbrmi_i2c_regmap_config);
+	if (IS_ERR(data->regmap))
+		return PTR_ERR(data->regmap);
+
 	/* Enable alert for SB-RMI sequence */
-	ret = sbrmi_enable_alert(client);
+	ret = sbrmi_enable_alert(data);
 	if (ret < 0)
 		return ret;
 
Signature · 1 line
2.25.1
#8 lore raw

PATCH v5 5/11 misc: amd-sbi: Optimize the wait condition for mailbox command completion

In reply to Akshay Gupta

- optimize the wait condition to indicate command completion by replacing the do while loop with regmap subsystem API regmap_read_poll_timeout()

---
Changes since v4:
- Previously patch 3

Changes since v3:
- New patch as per suggestion from Arnd

 drivers/misc/amd-sbi/rmi-core.c | 19 ++++---------------
 1 file changed, 4 insertions(+), 15 deletions(-)
drivers/misc/amd-sbi/rmi-core.c +4 -15
index 663ab9176d95..1d5e2556ab88 100644
@@ -23,7 +23,7 @@ int rmi_mailbox_xfer(struct sbrmi_data *data,
 		     struct sbrmi_mailbox_msg *msg)
 {
 	unsigned int bytes;
-	int i, ret, retry = 10;
+	int i, ret;
 	int sw_status;
 	u8 byte;
 
@@ -64,21 +64,10 @@ int rmi_mailbox_xfer(struct sbrmi_data *data,
 	 * an ALERT (if enabled) to initiator (BMC) to indicate completion
 	 * of the requested command
 	 */
-	do {
-		ret = regmap_read(data->regmap, SBRMI_STATUS, &sw_status);
-		if (sw_status < 0) {
-			ret = sw_status;
-			goto exit_unlock;
-		}
-		if (sw_status & SW_ALERT_MASK)
-			break;
-		usleep_range(50, 100);
-	} while (retry--);
-
-	if (retry < 0) {
-		ret = -EIO;
+	ret = regmap_read_poll_timeout(data->regmap, SBRMI_STATUS, sw_status,
+				       sw_status & SW_ALERT_MASK, 500, 2000000);
+	if (ret)
 		goto exit_unlock;
-	}
 
 	/*
 	 * For a read operation, the initiator (BMC) reads the firmware
Signature · 1 line
2.25.1
#9 lore raw

PATCH v5 6/11 misc: amd-sbi: Add support for AMD_SBI IOCTL

In reply to Akshay Gupta

The present sbrmi module only support reporting power via hwmon. However, AMD data center range of processors support various system management functionality using custom protocols defined in Advanced Platform Management Link (APML) specification.

Register a miscdevice, which creates a device /dev/sbrmiX with an IOCTL interface for the user space to invoke the APML Mailbox protocol, which is already defined in sbrmi_mailbox_xfer().

The APML protocols depend on a set of RMI registers. Having an IOCTL as a single entry point will help in providing synchronization among these protocols as multiple transactions on RMI register set may create race condition.
Support for other protocols will be added in subsequent patches.

Open-sourced and widely used https://github.com/amd/esmi_oob_library
will continue to provide user-space programmable API.

---
Changes since v4:
- Address review comment
- Address Greg review comments
- Not initialize ret
- return on error
- Previously patch 4
- Fix documentation warning

Changes since v3:
- Previously patch 3
- Documentation and comments changes

Changes since v2:
- update the MACROS name as per feedback

Changes since v1:
- Previously patch 5
- Add IOCTL description in ioctl-number.rst
- Split patch as per suggestion.

 Documentation/misc-devices/index.rst          |  1 +
 .../userspace-api/ioctl/ioctl-number.rst      |  2 +
 drivers/misc/amd-sbi/rmi-core.c               | 81 +++++++++++++++++--
 drivers/misc/amd-sbi/rmi-core.h               | 15 ++--
 drivers/misc/amd-sbi/rmi-hwmon.c              | 15 ++--
 drivers/misc/amd-sbi/rmi-i2c.c                | 25 +++++-
 include/uapi/misc/amd-apml.h                  | 66 +++++++++++++++
 7 files changed, 178 insertions(+), 27 deletions(-)
 create mode 100644 include/uapi/misc/amd-apml.h
Documentation/misc-devices/index.rst +1 -0
index 8c5b226d8313..081e79415e38 100644
@@ -12,6 +12,7 @@ fit into other categories.
    :maxdepth: 2
 
    ad525x_dpot
+   amd-sbi
    apds990x
    bh1770glc
    c2port
Documentation/userspace-api/ioctl/ioctl-number.rst +2 -0
index 6d1465315df3..5692b50b3c6f 100644
@@ -392,6 +392,8 @@ Code  Seq#    Include File                                           Comments
                                                                      <mailto:[email protected]>
 0xF8  all    arch/x86/include/uapi/asm/amd_hsmp.h                    AMD HSMP EPYC system management interface driver
                                                                      <mailto:[email protected]>
+0xF9  00-0F  uapi/misc/amd-apml.h		                     AMD side band system management interface driver
+                                                                     <mailto:[email protected]>
 0xFD  all    linux/dm-ioctl.h
 0xFE  all    linux/isst_if.h
 ====  =====  ======================================================= ================================================================
drivers/misc/amd-sbi/rmi-core.c +73 -8

116 lines not on this page · raw message

drivers/misc/amd-sbi/rmi-core.h +7 -8
index bbb6bb1cefde..e3a11575d19e 100644
@@ -6,10 +6,12 @@
 #ifndef _SBRMI_CORE_H_
 #define _SBRMI_CORE_H_
 
+#include <linux/miscdevice.h>
 #include <linux/mutex.h>
 #include <linux/i2c.h>
 #include <linux/platform_device.h>
 #include <linux/regmap.h>
+#include <uapi/misc/amd-apml.h>
 
 /* SB-RMI registers */
 enum sbrmi_reg {
@@ -48,18 +50,15 @@ enum sbrmi_msg_id {
 
 /* Each client has this additional data */
 struct sbrmi_data {
+	struct miscdevice sbrmi_misc_dev;
 	struct regmap *regmap;
+	/* Mutex locking */
 	struct mutex lock;
 	u32 pwr_limit_max;
+	u8 dev_static_addr;
 };
 
-struct sbrmi_mailbox_msg {
-	u8 cmd;
-	bool read;
-	u32 data_in;
-	u32 data_out;
-};
-
-int rmi_mailbox_xfer(struct sbrmi_data *data, struct sbrmi_mailbox_msg *msg);
+int rmi_mailbox_xfer(struct sbrmi_data *data, struct apml_message *msg);
 int create_hwmon_sensor_device(struct device *dev, struct sbrmi_data *data);
+int create_misc_rmi_device(struct sbrmi_data *data, struct device *dev);
 #endif /*_SBRMI_CORE_H_*/
drivers/misc/amd-sbi/rmi-hwmon.c +8 -7
index 720e800db1f0..ee0c3b72174c 100644
@@ -6,6 +6,7 @@
  */
 #include <linux/err.h>
 #include <linux/hwmon.h>
+#include <uapi/misc/amd-apml.h>
 #include "rmi-core.h"
 
 /* Do not allow setting negative power limit */
@@ -15,7 +16,7 @@ static int sbrmi_read(struct device *dev, enum hwmon_sensor_types type,
 		      u32 attr, int channel, long *val)
 {
 	struct sbrmi_data *data = dev_get_drvdata(dev);
-	struct sbrmi_mailbox_msg msg = { 0 };
+	struct apml_message msg = { 0 };
 	int ret;
 
 	if (!data)
@@ -24,7 +25,7 @@ static int sbrmi_read(struct device *dev, enum hwmon_sensor_types type,
 	if (type != hwmon_power)
 		return -EINVAL;
 
-	msg.read = true;
+	msg.data_in.reg_in[AMD_SBI_RD_FLAG_INDEX] = 1;
 	switch (attr) {
 	case hwmon_power_input:
 		msg.cmd = SBRMI_READ_PKG_PWR_CONSUMPTION;
@@ -35,7 +36,7 @@ static int sbrmi_read(struct device *dev, enum hwmon_sensor_types type,
 		ret = rmi_mailbox_xfer(data, &msg);
 		break;
 	case hwmon_power_cap_max:
-		msg.data_out = data->pwr_limit_max;
+		msg.data_out.mb_out[AMD_SBI_RD_WR_DATA_INDEX] = data->pwr_limit_max;
 		ret = 0;
 		break;
 	default:
@@ -44,7 +45,7 @@ static int sbrmi_read(struct device *dev, enum hwmon_sensor_types type,
 	if (ret < 0)
 		return ret;
 	/* hwmon power attributes are in microWatt */
-	*val = (long)msg.data_out * 1000;
+	*val = (long)msg.data_out.mb_out[AMD_SBI_RD_WR_DATA_INDEX] * 1000;
 	return ret;
 }
 
@@ -52,7 +53,7 @@ static int sbrmi_write(struct device *dev, enum hwmon_sensor_types type,
 		       u32 attr, int channel, long val)
 {
 	struct sbrmi_data *data = dev_get_drvdata(dev);
-	struct sbrmi_mailbox_msg msg = { 0 };
+	struct apml_message msg = { 0 };
 
 	if (!data)
 		return -ENODEV;
@@ -68,8 +69,8 @@ static int sbrmi_write(struct device *dev, enum hwmon_sensor_types type,
 	val = clamp_val(val, SBRMI_PWR_MIN, data->pwr_limit_max);
 
 	msg.cmd = SBRMI_WRITE_PKG_PWR_LIMIT;
-	msg.data_in = val;
-	msg.read = false;
+	msg.data_in.mb_in[AMD_SBI_RD_WR_DATA_INDEX] = val;
+	msg.data_in.reg_in[AMD_SBI_RD_FLAG_INDEX] = 0;
 
 	return rmi_mailbox_xfer(data, &msg);
 }
drivers/misc/amd-sbi/rmi-i2c.c +21 -4
index 7a9801273a4c..919ff7f61225 100644
@@ -38,15 +38,15 @@ static int sbrmi_enable_alert(struct sbrmi_data *data)
 
 static int sbrmi_get_max_pwr_limit(struct sbrmi_data *data)
 {
-	struct sbrmi_mailbox_msg msg = { 0 };
+	struct apml_message msg = { 0 };
 	int ret;
 
 	msg.cmd = SBRMI_READ_PKG_MAX_PWR_LIMIT;
-	msg.read = true;
+	msg.data_in.reg_in[AMD_SBI_RD_FLAG_INDEX] = 1;
 	ret = rmi_mailbox_xfer(data, &msg);
 	if (ret < 0)
 		return ret;
-	data->pwr_limit_max = msg.data_out;
+	data->pwr_limit_max = msg.data_out.mb_out[AMD_SBI_RD_WR_DATA_INDEX];
 
 	return ret;
 }
@@ -81,8 +81,24 @@ static int sbrmi_i2c_probe(struct i2c_client *client)
 	if (ret < 0)
 		return ret;
 
+	data->dev_static_addr = client->addr;
 	dev_set_drvdata(dev, data);
-	return create_hwmon_sensor_device(dev, data);
+	ret = create_hwmon_sensor_device(dev, data);
+	if (ret < 0)
+		return ret;
+	return create_misc_rmi_device(data, dev);
+}
+
+static void sbrmi_i2c_remove(struct i2c_client *client)
+{
+	struct sbrmi_data *data = dev_get_drvdata(&client->dev);
+
+	misc_deregister(&data->sbrmi_misc_dev);
+	/* Assign fops and parent of misc dev to NULL */
+	data->sbrmi_misc_dev.fops = NULL;
+	data->sbrmi_misc_dev.parent = NULL;
+	dev_info(&client->dev, "Removed sbrmi-i2c driver\n");
+	return;
 }
 
 static const struct i2c_device_id sbrmi_id[] = {
@@ -105,6 +121,7 @@ static struct i2c_driver sbrmi_driver = {
 		.of_match_table = of_match_ptr(sbrmi_of_match),
 	},
 	.probe = sbrmi_i2c_probe,
+	.remove = sbrmi_i2c_remove,
 	.id_table = sbrmi_id,
 };
 
include/uapi/misc/amd-apml.h new file +66 -0

66 lines not on this page · raw message

Signature · 1 line
2.25.1
maintainer
#10 lore raw

Re: [PATCH v5 06/11] misc: amd-sbi: Add support for AMD_SBI IOCTL

Quoted 10 lines · Akshay Gupta

+static long sbrmi_ioctl(struct file *fp, unsigned int cmd, unsigned long arg)

drivers/misc/amd-sbi/rmi-core.c +7 -0
+{
+	int __user *arguser = (int  __user *)arg;
+	struct apml_message msg = { 0 };
+	bool read = false;
+	int ret;
+
+	struct sbrmi_data *data = container_of(fp->private_data, struct 

sbrmi_data,

drivers/misc/amd-sbi/rmi-core.c int rmi_mailbox_xfer(struct sbrmi_data *data, · line 116

Quoted 8 lines
drivers/misc/amd-sbi/rmi-core.c +8 -0
+					       sbrmi_misc_dev);
+	if (!data)
+		return -ENODEV;
+
+	/* Copy the structure from user */
+	if (copy_struct_from_user(&msg, sizeof(msg), arguser,
+				  sizeof(struct apml_message)))
+		return -EFAULT;

This is not how ioctl commands work: you need to check the 'cmd' argument, which includes the length of the data.

copy_struct_from_user() makes no sense here since the length is fixed (for a given command).

drivers/misc/amd-sbi/rmi-core.c int rmi_mailbox_xfer(struct sbrmi_data *data, · line 126

Quoted 5 lines
drivers/misc/amd-sbi/rmi-core.c +5 -0
+	switch (msg.cmd) {
+	case 0 ... 0x999:
+		/* Mailbox protocol */
+		ret = rmi_mailbox_xfer(data, &msg);
+		break;

This looks like you are blindly passing through any command from userspace, which is generally not the preferred way.

Usually this should be a known set of high-level commands accepted by the driver.

drivers/misc/amd-sbi/rmi-core.c int rmi_mailbox_xfer(struct sbrmi_data *data, · line 142

Quoted 4 lines
drivers/misc/amd-sbi/rmi-core.c +4 -0
+static const struct file_operations sbrmi_fops = {
+	.owner		= THIS_MODULE,
+	.unlocked_ioctl	= sbrmi_ioctl,
+	.compat_ioctl	= sbrmi_ioctl,

Change this to

.compat_ioctl = compat_ptr_ioctl,

Quoted 4 lines
drivers/misc/amd-sbi/rmi-core.c +4 -0
+	data->sbrmi_misc_dev.name	= devm_kasprintf(dev,
+							 GFP_KERNEL,
+							 "sbrmi-%x",
+							 data->dev_static_addr);

drivers/misc/amd-sbi/rmi-core.c int rmi_mailbox_xfer(struct sbrmi_data *data, · line 161

Quoted 10 lines
drivers/misc/amd-sbi/rmi-core.c +10 -0
+	data->sbrmi_misc_dev.minor	= MISC_DYNAMIC_MINOR;
+	data->sbrmi_misc_dev.fops	= &sbrmi_fops;
+	data->sbrmi_misc_dev.parent	= dev;
+	data->sbrmi_misc_dev.nodename	= devm_kasprintf(dev,
+							 GFP_KERNEL,
+							 "sbrmi-%x",
+							 data->dev_static_addr);
+	data->sbrmi_misc_dev.mode	= 0600;
+
+	return misc_register(&data->sbrmi_misc_dev);

What is 'dev_static_addr'? Usually you want a miscdevice to have a constant name and a static structure definition, not dynamic allocation.

Are there multiple devices of this type in a given system?

Quoted 14 lines
include/uapi/misc/amd-apml.h new file +14 -0
+struct apml_message {
+	/* message ids:
+	 * Mailbox Messages:	0x0 ... 0x999
+	 */
+	__u32 cmd;
+
+	/*
+	 * 8 bit data for reg read,
+	 * 32 bit data in case of mailbox,
+	 */
+	union {
+		__u32 mb_out[2];
+		__u8 reg_out[8];
+	} data_out;

include/uapi/misc/amd-apml.h line 41

Quoted 10 lines
include/uapi/misc/amd-apml.h new file +10 -0
+
+	/*
+	 * [0]...[3] mailbox 32bit input
+	 * [7] read/write functionality
+	 */
+	union {
+		__u32 mb_in[2];
+		__u8 reg_in[8];
+	} data_in;
+} __attribute__((packed));

You normally want to have the in-kernel data aligned. Even if userspace has it at a misaligned offset, it will still work without the __packed.

Arnd

#11 lore raw

Re: [PATCH v5 06/11] misc: amd-sbi: Add support for AMD_SBI IOCTL

Quoted 17 lines · Arnd Bergmann

Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding.

Quoted 12 lines · Akshay Gupta

+static long sbrmi_ioctl(struct file *fp, unsigned int cmd, unsigned long arg)

drivers/misc/amd-sbi/rmi-core.c +7 -0
+{
+     int __user *arguser = (int  __user *)arg;
+     struct apml_message msg = { 0 };
+     bool read = false;
+     int ret;
+
+     struct sbrmi_data *data = container_of(fp->private_data, struct

sbrmi_data,

drivers/misc/amd-sbi/rmi-core.c +2 -0
+                                            sbrmi_misc_dev);
+     if (!data)
Quoted 8 lines

drivers/misc/amd-sbi/rmi-core.c int rmi_mailbox_xfer(struct sbrmi_data *data, · line 116

Quoted 6 lines
drivers/misc/amd-sbi/rmi-core.c +6 -0
+             return -ENODEV;
+
+     /* Copy the structure from user */
+     if (copy_struct_from_user(&msg, sizeof(msg), arguser,
+                               sizeof(struct apml_message)))
+             return -EFAULT;

This is not how ioctl commands work: you need to check the 'cmd' argument, which includes the length of the data.

will add check in ioctl for 'cmd'. Thank you.

Quoted 3 lines

copy_struct_from_user() makes no sense here since the length is fixed (for a given command).

will modify it to use copy_from_user and size using the _IOC_SIZE(cmd)

Quoted 11 lines

drivers/misc/amd-sbi/rmi-core.c int rmi_mailbox_xfer(struct sbrmi_data *data, · line 126

Quoted 6 lines
drivers/misc/amd-sbi/rmi-core.c +5 -0
+     switch (msg.cmd) {
+     case 0 ... 0x999:
+             /* Mailbox protocol */
+             ret = rmi_mailbox_xfer(data, &msg);
+             break;

This looks like you are blindly passing through any command from userspace, which is generally not the preferred way.

Usually this should be a known set of high-level commands accepted by the driver.

This will be addressed

Quoted 8 lines

drivers/misc/amd-sbi/rmi-core.c int rmi_mailbox_xfer(struct sbrmi_data *data, · line 142

Quoted 5 lines
drivers/misc/amd-sbi/rmi-core.c +4 -0
+static const struct file_operations sbrmi_fops = {
+     .owner          = THIS_MODULE,
+     .unlocked_ioctl = sbrmi_ioctl,
+     .compat_ioctl   = sbrmi_ioctl,

Change this to

.compat_ioctl = compat_ptr_ioctl,

sure.

Quoted 16 lines
Quoted 6 lines
drivers/misc/amd-sbi/rmi-core.c +4 -0
+     data->sbrmi_misc_dev.name       = devm_kasprintf(dev,
+                                                      GFP_KERNEL,
+                                                      "sbrmi-%x",
+                                                      data->dev_static_addr);

drivers/misc/amd-sbi/rmi-core.c int rmi_mailbox_xfer(struct sbrmi_data *data, · line 161

Quoted 10 lines
drivers/misc/amd-sbi/rmi-core.c +10 -0
+     data->sbrmi_misc_dev.minor      = MISC_DYNAMIC_MINOR;
+     data->sbrmi_misc_dev.fops       = &sbrmi_fops;
+     data->sbrmi_misc_dev.parent     = dev;
+     data->sbrmi_misc_dev.nodename   = devm_kasprintf(dev,
+                                                      GFP_KERNEL,
+                                                      "sbrmi-%x",
+                                                      data->dev_static_addr);
+     data->sbrmi_misc_dev.mode       = 0600;
+
+     return misc_register(&data->sbrmi_misc_dev);
Quoted 5 lines

What is 'dev_static_addr'? Usually you want a miscdevice to have a constant name and a static structure definition, not dynamic allocation.

Are there multiple devices of this type in a given system?

Yes, there can be multiple devices on the basis of number of nodes.

Quoted 29 lines
Quoted 15 lines
include/uapi/misc/amd-apml.h new file +14 -0
+struct apml_message {
+     /* message ids:
+      * Mailbox Messages:    0x0 ... 0x999
+      */
+     __u32 cmd;
+
+     /*
+      * 8 bit data for reg read,
+      * 32 bit data in case of mailbox,
+      */
+     union {
+             __u32 mb_out[2];
+             __u8 reg_out[8];
+     } data_out;

include/uapi/misc/amd-apml.h line 41

Quoted 10 lines
include/uapi/misc/amd-apml.h new file +10 -0
+
+     /*
+      * [0]...[3] mailbox 32bit input
+      * [7] read/write functionality
+      */
+     union {
+             __u32 mb_in[2];
+             __u8 reg_in[8];
+     } data_in;
+} __attribute__((packed));

You normally want to have the in-kernel data aligned. Even if userspace has it at a misaligned offset, it will still work without the __packed.

Quoted 1 line

Arnd

Thank you, will update.

#12 lore raw

PATCH v5 7/11 misc: amd-sbi: Add support for mailbox error codes

In reply to Akshay Gupta

APML mailbox protocol returns additional error codes written by SMU firmware in the out-bound register 0x37. These errors include, invalid core, message not supported over platform and others. This additional error codes can be used to provide more details to user space.

---
Changes since v4:
- Previously patch 5
- Fix Documentation warning

Changes since v3:
- update ioctl comment description

Changes since v1:
- bifurcated from previous patch 5

 drivers/misc/amd-sbi/rmi-core.c | 12 +++++++++++-
 include/uapi/misc/amd-apml.h    |  5 +++++
 2 files changed, 16 insertions(+), 1 deletion(-)
drivers/misc/amd-sbi/rmi-core.c +11 -1
index c39a29d90c27..642fe762cf6f 100644
@@ -25,13 +25,15 @@
 int rmi_mailbox_xfer(struct sbrmi_data *data,
 		     struct apml_message *msg)
 {
-	unsigned int bytes;
+	unsigned int bytes, ec;
 	int i, ret;
 	int sw_status;
 	u8 byte;
 
 	mutex_lock(&data->lock);
 
+	msg->fw_ret_code = 0;
+
 	/* Indicate firmware a command is to be serviced */
 	ret = regmap_write(data->regmap, SBRMI_INBNDMSG7, START_CMD);
 	if (ret < 0)
@@ -72,6 +74,9 @@ int rmi_mailbox_xfer(struct sbrmi_data *data,
 	if (ret)
 		goto exit_unlock;
 
+	ret = regmap_read(data->regmap, SBRMI_OUTBNDMSG7, &ec);
+	if (ret || ec)
+		goto exit_clear_alert;
 	/*
 	 * For a read operation, the initiator (BMC) reads the firmware
 	 * response Command Data Out[31:0] from SBRMI::OutBndMsg_inst[4:1]
@@ -87,12 +92,17 @@ int rmi_mailbox_xfer(struct sbrmi_data *data,
 		}
 	}
 
+exit_clear_alert:
 	/*
 	 * BMC must write 1'b1 to SBRMI::Status[SwAlertSts] to clear the
 	 * ALERT to initiator
 	 */
 	ret = regmap_write(data->regmap, SBRMI_STATUS,
 			   sw_status | SW_ALERT_MASK);
+	if (ec) {
+		ret = -EPROTOTYPE;
+		msg->fw_ret_code = ec;
+	}
 exit_unlock:
 	mutex_unlock(&data->lock);
 	return ret;
include/uapi/misc/amd-apml.h +5 -0
index 5721aaa0c6bd..47a057226300 100644
@@ -38,6 +38,10 @@ struct apml_message {
 		__u32 mb_in[2];
 		__u8 reg_in[8];
 	} data_in;
+	/*
+	 * Error code is returned in case of soft mailbox
+	 */
+	__u32 fw_ret_code;
 } __attribute__((packed));
 
 /*
@@ -60,6 +64,7 @@ struct apml_message {
  * The APML RMI module checks whether the cmd is
  * - Mailbox message read/write(0x0~0x999)
  * - returning "-EFAULT" if none of the above
+ * "-EPROTOTYPE" error is returned to provide additional error details
  */
 #define SBRMI_IOCTL_CMD		_IOWR(SB_BASE_IOCTL_NR, 0, struct apml_message)
 
Signature · 1 line
2.25.1
#13 lore raw

PATCH v5 8/11 misc: amd-sbi: Add support for CPUID protocol

In reply to Akshay Gupta

- AMD provides custom protocol to read Processor feature capabilities and configuration information through side band. The information is accessed by providing CPUID Function, extended function and thread ID to the protocol. Undefined function returns 0.

---
Changes since v4:
- Previously patch 6
- Address review comments

Changes since v3:
- Address review comments:
- update the #define to inline function
- pack the union inside the structure

Changes since v2:
- update the MACROS name as per feedback

Changes since v1:
- bifurcated from previous patch 5

 drivers/misc/amd-sbi/rmi-core.c | 148 ++++++++++++++++++++++++++++++++
 drivers/misc/amd-sbi/rmi-core.h |   5 +-
 include/uapi/misc/amd-apml.h    |  16 ++++
 3 files changed, 168 insertions(+), 1 deletion(-)
drivers/misc/amd-sbi/rmi-core.c +148 -0
index 642fe762cf6f..6fd6e8e579d5 100644
@@ -17,11 +17,156 @@
 
 /* Mask for Status Register bit[1] */
 #define SW_ALERT_MASK	0x2
+/* Mask to check H/W Alert status bit */
+#define HW_ALERT_MASK	0x80
 
 /* Software Interrupt for triggering */
 #define START_CMD	0x80
 #define TRIGGER_MAILBOX	0x01
 
+/* Default message lengths as per APML command protocol */
+/* CPUID */
+#define CPUID_RD_DATA_LEN	0x8
+#define CPUID_WR_DATA_LEN	0x8
+#define CPUID_RD_REG_LEN	0xa
+#define CPUID_WR_REG_LEN	0x9
+
+/* CPUID MSR Command Ids */
+#define CPUID_MCA_CMD	0x73
+#define RD_CPUID_CMD	0x91
+
+/* input for bulk write to CPUID protocol */
+struct cpu_msr_indata {
+	u8 wr_len;	/* const value */
+	u8 rd_len;	/* const value */
+	u8 proto_cmd;	/* const value */
+	u8 thread;	/* thread number */
+	union {
+		u8 reg_offset[4];	/* input value */
+		u32 value;
+	} __packed;
+	u8 ext; /* extended function */
+};
+
+/* output for bulk read from CPUID protocol */
+struct cpu_msr_outdata {
+	u8 num_bytes;	/* number of bytes return */
+	u8 status;	/* Protocol status code */
+	union {
+		u64 value;
+		u8 reg_data[8];
+	} __packed;
+};
+
+static inline void prepare_cpuid_input_message(struct cpu_msr_indata *input,
+					       u8 thread_id, u32 func,
+					       u8 ext_func)
+{
+	input->rd_len		= CPUID_RD_DATA_LEN;
+	input->wr_len		= CPUID_WR_DATA_LEN;
+	input->proto_cmd	= RD_CPUID_CMD;
+	input->thread		= thread_id << 1;
+	input->value		= func;
+	input->ext		= ext_func;
+}
+
+static int sbrmi_get_rev(struct sbrmi_data *data)
+{
+	struct apml_message msg = { 0 };
+	int ret;
+
+	msg.data_in.reg_in[AMD_SBI_REG_OFF_INDEX] = SBRMI_REV;
+	msg.data_in.reg_in[AMD_SBI_RD_FLAG_INDEX] = 1;
+	ret = regmap_read(data->regmap,
+			  msg.data_in.reg_in[AMD_SBI_REG_OFF_INDEX],
+			  &msg.data_out.mb_out[AMD_SBI_RD_WR_DATA_INDEX]);
+	if (ret < 0)
+		return ret;
+
+	data->rev = msg.data_out.reg_out[AMD_SBI_RD_WR_DATA_INDEX];
+	return 0;
+}
+
+/* Read CPUID function protocol */
+static int rmi_cpuid_read(struct sbrmi_data *data,
+			  struct apml_message *msg)
+{
+	struct cpu_msr_indata input = {0};
+	struct cpu_msr_outdata output = {0};
+	int val = 0;
+	int ret, hw_status;
+	u16 thread;
+
+	mutex_lock(&data->lock);
+	/* cache the rev value to identify if protocol is supported or not */
+	if (!data->rev) {
+		ret = sbrmi_get_rev(data);
+		if (ret < 0)
+			goto exit_unlock;
+	}
+	/* CPUID protocol for REV 0x10 is not supported*/
+	if (data->rev == 0x10) {
+		ret = -EOPNOTSUPP;
+		goto exit_unlock;
+	}
+
+	thread = msg->data_in.reg_in[AMD_SBI_THREAD_LOW_INDEX] |
+		 msg->data_in.reg_in[AMD_SBI_THREAD_HI_INDEX] << 8;
+
+	/* Thread > 127, Thread128 CS register, 1'b1 needs to be set to 1 */
+	if (thread > 127) {
+		thread -= 128;
+		val = 1;
+	}
+	ret = regmap_write(data->regmap, SBRMI_THREAD128CS, val);
+	if (ret < 0)
+		goto exit_unlock;
+
+	prepare_cpuid_input_message(&input, thread,
+				    msg->data_in.mb_in[AMD_SBI_RD_WR_DATA_INDEX],
+				    msg->data_in.reg_in[AMD_SBI_EXT_FUNC_INDEX]);
+
+	ret = regmap_bulk_write(data->regmap, CPUID_MCA_CMD,
+				&input, CPUID_WR_REG_LEN);
+	if (ret < 0)
+		goto exit_unlock;
+
+	/*
+	 * For RMI Rev 0x20, new h/w status bit is introduced. which is used
+	 * by firmware to indicate completion of commands (0x71, 0x72, 0x73).
+	 * wait for the status bit to be set by the hardware before
+	 * reading the data out.
+	 */
+	ret = regmap_read_poll_timeout(data->regmap, SBRMI_STATUS, hw_status,
+				       hw_status & HW_ALERT_MASK, 500, 2000000);
+	if (ret)
+		goto exit_unlock;
+
+	ret = regmap_bulk_read(data->regmap, CPUID_MCA_CMD,
+			       &output, CPUID_RD_REG_LEN);
+	if (ret < 0)
+		goto exit_unlock;
+
+	ret = regmap_write(data->regmap, SBRMI_STATUS,
+			   HW_ALERT_MASK);
+	if (ret < 0)
+		goto exit_unlock;
+
+	if (output.num_bytes != CPUID_RD_REG_LEN - 1) {
+		ret = -EMSGSIZE;
+		goto exit_unlock;
+	}
+	if (output.status) {
+		ret = -EPROTOTYPE;
+		msg->fw_ret_code = output.status;
+		goto exit_unlock;
+	}
+	msg->data_out.cpu_msr_out = output.value;
+exit_unlock:
+	mutex_unlock(&data->lock);
+	return ret;
+}
+
 int rmi_mailbox_xfer(struct sbrmi_data *data,
 		     struct apml_message *msg)
 {
@@ -134,6 +279,9 @@ static long sbrmi_ioctl(struct file *fp, unsigned int cmd, unsigned long arg)
 		/* Mailbox protocol */
 		ret = rmi_mailbox_xfer(data, &msg);
 		break;
+	case APML_CPUID:
+		ret = rmi_cpuid_read(data, &msg);
+		break;
 	default:
 		return -EINVAL;
 	}
drivers/misc/amd-sbi/rmi-core.h +4 -1
index e3a11575d19e..3f2c1c6b22d4 100644
@@ -15,7 +15,8 @@
 
 /* SB-RMI registers */
 enum sbrmi_reg {
-	SBRMI_CTRL		= 0x01,
+	SBRMI_REV,
+	SBRMI_CTRL,
 	SBRMI_STATUS,
 	SBRMI_OUTBNDMSG0	= 0x30,
 	SBRMI_OUTBNDMSG1,
@@ -34,6 +35,7 @@ enum sbrmi_reg {
 	SBRMI_INBNDMSG6,
 	SBRMI_INBNDMSG7,
 	SBRMI_SW_INTERRUPT,
+	SBRMI_THREAD128CS	= 0x4b,
 };
 
 /*
@@ -56,6 +58,7 @@ struct sbrmi_data {
 	struct mutex lock;
 	u32 pwr_limit_max;
 	u8 dev_static_addr;
+	u8 rev;
 };
 
 int rmi_mailbox_xfer(struct sbrmi_data *data, struct apml_message *msg);
include/uapi/misc/amd-apml.h +16 -0
index 47a057226300..847a83770ab0 100644
@@ -7,38 +7,53 @@
 
 #include <linux/types.h>
 
+/* command ID to identify CPUID protocol */
+#define APML_CPUID	0x1000
 /* These are byte indexes into data_in and data_out arrays */
 #define AMD_SBI_RD_WR_DATA_INDEX	0
 #define AMD_SBI_REG_OFF_INDEX		0
 #define AMD_SBI_REG_VAL_INDEX		4
 #define AMD_SBI_RD_FLAG_INDEX		7
+#define AMD_SBI_THREAD_LOW_INDEX	4
+#define AMD_SBI_THREAD_HI_INDEX		5
+#define AMD_SBI_EXT_FUNC_INDEX		6
 
 #define AMD_SBI_MB_DATA_SIZE		4
 
 struct apml_message {
 	/* message ids:
 	 * Mailbox Messages:	0x0 ... 0x999
+	 * APML_CPUID:		0x1000
 	 */
 	__u32 cmd;
 
 	/*
 	 * 8 bit data for reg read,
 	 * 32 bit data in case of mailbox,
+	 * up to 64 bit in case of cpuid
 	 */
 	union {
+		__u64 cpu_msr_out;
 		__u32 mb_out[2];
 		__u8 reg_out[8];
 	} data_out;
 
 	/*
 	 * [0]...[3] mailbox 32bit input
+	 *	     cpuid,
+	 * [4][5] cpuid: thread
+	 * [6] cpuid: ext function & read eax/ebx or ecx/edx
+	 *	[7:0] -> bits [7:4] -> ext function &
+	 *	bit [0] read eax/ebx or ecx/edx
 	 * [7] read/write functionality
 	 */
 	union {
+		__u64 cpu_msr_in;
 		__u32 mb_in[2];
 		__u8 reg_in[8];
 	} data_in;
 	/*
+	 * Status code is returned in case of CPUID access
 	 * Error code is returned in case of soft mailbox
 	 */
 	__u32 fw_ret_code;
@@ -63,6 +78,7 @@ struct apml_message {
  * The IOCTL provides userspace access to AMD sideband protocols
  * The APML RMI module checks whether the cmd is
  * - Mailbox message read/write(0x0~0x999)
+ * - CPUID read(0x1000)
  * - returning "-EFAULT" if none of the above
  * "-EPROTOTYPE" error is returned to provide additional error details
  */
Signature · 1 line
2.25.1
#14 lore raw

PATCH v5 9/11 misc: amd-sbi: Add support for read MCA register protocol

In reply to Akshay Gupta

- AMD provides custom protocol to read Machine Check Architecture(MCA) registers over sideband. The information is accessed for range of MCA registers by passing register address and thread ID to the protocol. MCA register read command using the register address to access Core::X86::Msr::MCG_CAP which determines the number of MCA banks. Access is read-only

---
Changes since v4:
- Previously patch 7
- Address review comment for documentation warning

Changes since v3:
- Address review comments:
- update the #define to inline function
- pack the union inside the structure

Changes since v2:
- update the MACROS name as per feedback

Changes since v1:
- bifurcated from previous patch 5

 drivers/misc/amd-sbi/rmi-core.c | 100 ++++++++++++++++++++++++++++++++
 include/uapi/misc/amd-apml.h    |  18 ++++--
 2 files changed, 112 insertions(+), 6 deletions(-)
drivers/misc/amd-sbi/rmi-core.c +100 -0
index 6fd6e8e579d5..662aa90980fc 100644
@@ -30,10 +30,16 @@
 #define CPUID_WR_DATA_LEN	0x8
 #define CPUID_RD_REG_LEN	0xa
 #define CPUID_WR_REG_LEN	0x9
+/* MSR */
+#define MSR_RD_REG_LEN		0xa
+#define MSR_WR_REG_LEN		0x8
+#define MSR_RD_DATA_LEN		0x8
+#define MSR_WR_DATA_LEN		0x7
 
 /* CPUID MSR Command Ids */
 #define CPUID_MCA_CMD	0x73
 #define RD_CPUID_CMD	0x91
+#define RD_MCA_CMD	0x86
 
 /* input for bulk write to CPUID protocol */
 struct cpu_msr_indata {
@@ -70,6 +76,16 @@ static inline void prepare_cpuid_input_message(struct cpu_msr_indata *input,
 	input->ext		= ext_func;
 }
 
+static inline void prepare_mca_msr_input_message(struct cpu_msr_indata *input,
+						 u8 thread_id, u32 data_in)
+{
+	input->rd_len		= MSR_RD_DATA_LEN;
+	input->wr_len		= MSR_WR_DATA_LEN;
+	input->proto_cmd	= RD_MCA_CMD;
+	input->thread		= thread_id << 1;
+	input->value		= data_in;
+}
+
 static int sbrmi_get_rev(struct sbrmi_data *data)
 {
 	struct apml_message msg = { 0 };
@@ -167,6 +183,86 @@ static int rmi_cpuid_read(struct sbrmi_data *data,
 	return ret;
 }
 
+/* MCA MSR protocol */
+static int rmi_mca_msr_read(struct sbrmi_data *data,
+			    struct apml_message *msg)
+{
+	struct cpu_msr_outdata output = {0};
+	struct cpu_msr_indata input = {0};
+	int ret, val = 0;
+	int hw_status;
+	u16 thread;
+
+	mutex_lock(&data->lock);
+	/* cache the rev value to identify if protocol is supported or not */
+	if (!data->rev) {
+		ret = sbrmi_get_rev(data);
+		if (ret < 0)
+			goto exit_unlock;
+	}
+	/* MCA MSR protocol for REV 0x10 is not supported*/
+	if (data->rev == 0x10) {
+		ret = -EOPNOTSUPP;
+		goto exit_unlock;
+	}
+
+	thread = msg->data_in.reg_in[AMD_SBI_THREAD_LOW_INDEX] |
+		 msg->data_in.reg_in[AMD_SBI_THREAD_HI_INDEX] << 8;
+
+	/* Thread > 127, Thread128 CS register, 1'b1 needs to be set to 1 */
+	if (thread > 127) {
+		thread -= 128;
+		val = 1;
+	}
+	ret = regmap_write(data->regmap, SBRMI_THREAD128CS, val);
+	if (ret < 0)
+		goto exit_unlock;
+
+	prepare_mca_msr_input_message(&input, thread,
+				      msg->data_in.mb_in[AMD_SBI_RD_WR_DATA_INDEX]);
+
+	ret = regmap_bulk_write(data->regmap, CPUID_MCA_CMD,
+				&input, MSR_WR_REG_LEN);
+	if (ret < 0)
+		goto exit_unlock;
+
+	/*
+	 * For RMI Rev 0x20, new h/w status bit is introduced. which is used
+	 * by firmware to indicate completion of commands (0x71, 0x72, 0x73).
+	 * wait for the status bit to be set by the hardware before
+	 * reading the data out.
+	 */
+	ret = regmap_read_poll_timeout(data->regmap, SBRMI_STATUS, hw_status,
+				       hw_status & HW_ALERT_MASK, 500, 2000000);
+	if (ret)
+		goto exit_unlock;
+
+	ret = regmap_bulk_read(data->regmap, CPUID_MCA_CMD,
+			       &output, MSR_RD_REG_LEN);
+	if (ret < 0)
+		goto exit_unlock;
+
+	ret = regmap_write(data->regmap, SBRMI_STATUS,
+			   HW_ALERT_MASK);
+	if (ret < 0)
+		goto exit_unlock;
+
+	if (output.num_bytes != MSR_RD_REG_LEN - 1) {
+		ret = -EMSGSIZE;
+		goto exit_unlock;
+	}
+	if (output.status) {
+		ret = -EPROTOTYPE;
+		msg->fw_ret_code = output.status;
+		goto exit_unlock;
+	}
+	msg->data_out.cpu_msr_out = output.value;
+
+exit_unlock:
+	mutex_unlock(&data->lock);
+	return ret;
+}
+
 int rmi_mailbox_xfer(struct sbrmi_data *data,
 		     struct apml_message *msg)
 {
@@ -282,6 +378,10 @@ static long sbrmi_ioctl(struct file *fp, unsigned int cmd, unsigned long arg)
 	case APML_CPUID:
 		ret = rmi_cpuid_read(data, &msg);
 		break;
+	case APML_MCA_MSR:
+		/* MCAMSR protocol */
+		ret = rmi_mca_msr_read(data, &msg);
+		break;
 	default:
 		return -EINVAL;
 	}
include/uapi/misc/amd-apml.h +12 -6
index 847a83770ab0..0a841809ca84 100644
@@ -7,8 +7,11 @@
 
 #include <linux/types.h>
 
-/* command ID to identify CPUID protocol */
-#define APML_CPUID	0x1000
+enum apml_protocol {
+	APML_CPUID	= 0x1000,
+	APML_MCA_MSR,
+};
+
 /* These are byte indexes into data_in and data_out arrays */
 #define AMD_SBI_RD_WR_DATA_INDEX	0
 #define AMD_SBI_REG_OFF_INDEX		0
@@ -24,13 +27,14 @@ struct apml_message {
 	/* message ids:
 	 * Mailbox Messages:	0x0 ... 0x999
 	 * APML_CPUID:		0x1000
+	 * APML_MCA_MSR:	0x1001
 	 */
 	__u32 cmd;
 
 	/*
 	 * 8 bit data for reg read,
 	 * 32 bit data in case of mailbox,
-	 * up to 64 bit in case of cpuid
+	 * up to 64 bit in case of cpuid and mca msr
 	 */
 	union {
 		__u64 cpu_msr_out;
@@ -40,8 +44,9 @@ struct apml_message {
 
 	/*
 	 * [0]...[3] mailbox 32bit input
-	 *	     cpuid,
-	 * [4][5] cpuid: thread
+	 *	     cpuid & mca msr,
+	 * [4][5] cpuid & mca msr: thread
+	 * [4] rmi reg wr: value
 	 * [6] cpuid: ext function & read eax/ebx or ecx/edx
 	 *	[7:0] -> bits [7:4] -> ext function &
 	 *	bit [0] read eax/ebx or ecx/edx
@@ -53,7 +58,7 @@ struct apml_message {
 		__u8 reg_in[8];
 	} data_in;
 	/*
-	 * Status code is returned in case of CPUID access
+	 * Status code is returned in case of CPUID/MCA access
 	 * Error code is returned in case of soft mailbox
 	 */
 	__u32 fw_ret_code;
@@ -79,6 +84,7 @@ struct apml_message {
  * The APML RMI module checks whether the cmd is
  * - Mailbox message read/write(0x0~0x999)
  * - CPUID read(0x1000)
+ * - MCAMSR read(0x1001)
  * - returning "-EFAULT" if none of the above
  * "-EPROTOTYPE" error is returned to provide additional error details
  */
Signature · 1 line
2.25.1
#15 lore raw

PATCH v5 10/11 misc: amd-sbi: Add support for register xfer

In reply to Akshay Gupta

- Provide user register access over IOCTL.
Both register read and write are supported.
- APML interface does not provide a synchronization method. By defining, a register access path, we use APML modules and library for all APML transactions. Without having to use external tools such as i2c-tools, which may cause race conditions.

---
Changes since v4:
- Previously patch 8
- Address review comment for documentation warning

Changes since v3:
- Add ioctl description comment

Changes since v2:
- update the MACROS name as per feedback

Changes since v1:
- bifurcated from previous patch 5

 drivers/misc/amd-sbi/rmi-core.c | 25 +++++++++++++++++++++++++
 include/uapi/misc/amd-apml.h    |  4 ++++
 2 files changed, 29 insertions(+)
drivers/misc/amd-sbi/rmi-core.c +25 -0
index 662aa90980fc..1c248dcb8ce9 100644
@@ -349,6 +349,27 @@ int rmi_mailbox_xfer(struct sbrmi_data *data,
 	return ret;
 }
 
+static int rmi_register_xfer(struct sbrmi_data *data,
+			     struct apml_message *msg)
+{
+	int ret;
+
+	if (WARN_ON(!msg))
+		return -EINVAL;
+
+	mutex_lock(&data->lock);
+	if (msg->data_in.reg_in[AMD_SBI_RD_FLAG_INDEX])
+		ret = regmap_read(data->regmap,
+				  msg->data_in.reg_in[AMD_SBI_REG_OFF_INDEX],
+				  &msg->data_out.mb_out[AMD_SBI_RD_WR_DATA_INDEX]);
+	else
+		ret = regmap_write(data->regmap,
+				   msg->data_in.reg_in[AMD_SBI_REG_OFF_INDEX],
+				   msg->data_in.reg_in[AMD_SBI_REG_VAL_INDEX]);
+	mutex_unlock(&data->lock);
+	return ret;
+}
+
 static long sbrmi_ioctl(struct file *fp, unsigned int cmd, unsigned long arg)
 {
 	int __user *arguser = (int  __user *)arg;
@@ -382,6 +403,10 @@ static long sbrmi_ioctl(struct file *fp, unsigned int cmd, unsigned long arg)
 		/* MCAMSR protocol */
 		ret = rmi_mca_msr_read(data, &msg);
 		break;
+	case APML_REG:
+		/* REG R/W */
+		ret = rmi_register_xfer(data, &msg);
+		break;
 	default:
 		return -EINVAL;
 	}
include/uapi/misc/amd-apml.h +4 -0
index 0a841809ca84..b18c20f9f809 100644
@@ -10,6 +10,7 @@
 enum apml_protocol {
 	APML_CPUID	= 0x1000,
 	APML_MCA_MSR,
+	APML_REG,
 };
 
 /* These are byte indexes into data_in and data_out arrays */
@@ -28,6 +29,7 @@ struct apml_message {
 	 * Mailbox Messages:	0x0 ... 0x999
 	 * APML_CPUID:		0x1000
 	 * APML_MCA_MSR:	0x1001
+	 * APML_REG:		0x1002
 	 */
 	__u32 cmd;
 
@@ -45,6 +47,7 @@ struct apml_message {
 	/*
 	 * [0]...[3] mailbox 32bit input
 	 *	     cpuid & mca msr,
+	 *	     rmi rd/wr: reg_offset
 	 * [4][5] cpuid & mca msr: thread
 	 * [4] rmi reg wr: value
 	 * [6] cpuid: ext function & read eax/ebx or ecx/edx
@@ -85,6 +88,7 @@ struct apml_message {
  * - Mailbox message read/write(0x0~0x999)
  * - CPUID read(0x1000)
  * - MCAMSR read(0x1001)
+ * - Register read/write(0x1002)
  * - returning "-EFAULT" if none of the above
  * "-EPROTOTYPE" error is returned to provide additional error details
  */
Signature · 1 line
2.25.1
#16 lore raw

PATCH v5 11/11 misc: amd-sbi: Add document for AMD SB IOCTL description

In reply to Akshay Gupta

- This document provides AMD side band IOCTL description defined for APML and its usage.
Multiple AMD custom protocols defined for side band system management uses this IOCTL.
User space C-APIs are made available by esmi_oob_library [1], which is provided by the E-SMS project [2].

---
Changes since v4:
- Previously patch 9
- Update description as per review comment
- Address the review comments for documentation warning

Changes since v3:
- Address the review comments

Changes since v2:
- update the MACROS name as per feedback

Changes since v1:
- New patch

 Documentation/misc-devices/amd-sbi.rst | 87 ++++++++++++++++++++++++++
 1 file changed, 87 insertions(+)
 create mode 100644 Documentation/misc-devices/amd-sbi.rst
Documentation/misc-devices/amd-sbi.rst new file +87 -0
new file mode 100644
index 000000000000..9fbb01b33032
@@ -0,0 +1,87 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+=======================
+AMD SIDE BAND interface
+=======================
+
+Some AMD Zen based processors supports system management
+functionality via side-band interface (SBI) called
+Advanced Platform Management Link (APML). APML is an I2C/I3C
+based 2-wire processor target interface. APML is used to
+communicate with the Remote Management Interface
+(SB Remote Management Interface (SB-RMI)
+and SB Temperature Sensor Interface (SB-TSI)).
+
+More details on the interface can be found in chapter
+"5 Advanced Platform Management Link (APML)" of the family/model PPR [1]_.
+
+.. [1] https://www.amd.com/content/dam/amd/en/documents/epyc-technical-docs/programmer-references/55898_B1_pub_0_50.zip
+
+
+SBRMI device
+============
+
+apml_sbrmi driver under the drivers/misc/amd-sbi creates miscdevice
+/dev/sbrmi-* to let user space programs run APML mailbox, CPUID,
+MCAMSR and register xfer commands.
+
+Register sets is common across APML protocols. IOCTL is providing synchronization
+among protocols as transactions may create race condition.
+
+$ ls -al /dev/sbrmi-3c
+crw-------    1 root     root       10,  53 Jul 10 11:13 /dev/sbrmi-3c
+
+apml_sbrmi driver registers hwmon sensors for monitoring power_cap_max,
+current power consumption and managing power_cap.
+
+Characteristics of the dev node:
+ * message ids are defined to run differnet xfer protocols:
+	* Mailbox:		0x0 ... 0x999
+	* CPUID:		0x1000
+	* MCA_MSR:		0x1001
+	* Register xfer:	0x1002
+
+Access restrictions:
+ * Only root user is allowed to open the file.
+ * APML Mailbox messages and Register xfer access are read-write,
+ * CPUID and MCA_MSR access is read-only.
+
+Driver IOCTLs
+=============
+
+.. c:macro:: SBRMI_IOCTL_CMD
+.. kernel-doc:: include/uapi/misc/amd-apml.h
+   :doc: SBRMI_IOCTL_CMD
+
+User-space usage
+================
+
+To access side band interface from a C program.
+First, user need to include the headers::
+
+  #include <uapi/misc/amd-apml.h>
+
+Which defines the supported IOCTL and data structure to be passed
+from the user space.
+
+Next thing, open the device file, as follows::
+
+  int file;
+
+  file = open("/dev/sbrmi-*", O_RDWR);
+  if (file < 0) {
+    /* ERROR HANDLING */
+    exit(1);
+  }
+
+The following IOCTL is defined:
+
+``#define SB_BASE_IOCTL_NR      0xF9``
+``#define SBRMI_IOCTL_CMD          _IOWR(SB_BASE_IOCTL_NR, 0, struct apml_message)``
+
+
+User space C-APIs are made available by esmi_oob_library, hosted at
+[2]_ which is provided by the E-SMS project [3]_.
+
+.. [2] https://github.com/amd/esmi_oob_library
+.. [3] https://www.amd.com/en/developer/e-sms.html
Signature · 1 line
2.25.1