Merge pull request tenstorrent#20 from joelsmithTT/joelsmith/sysfs-aiclk

Establish pattern for exposing attributes in sysfs
joelsmithTT · May 28, 2024 · d436790 · d436790
2 parents 041ca45 + c275909
commit d436790
Show file tree

Hide file tree

Showing 6 changed files with 123 additions and 0 deletions.
diff --git a/device.h b/device.h
@@ -36,6 +36,7 @@ struct tenstorrent_device {
 	DECLARE_BITMAP(resource_lock, TENSTORRENT_RESOURCE_LOCK_COUNT);
 
 	struct tt_hwmon_context hwmon_context;
+	struct tt_attribute_data *attributes;
 };
 
 struct tenstorrent_device_class {

diff --git a/enumerate.c b/enumerate.c
@@ -81,6 +81,7 @@ static int tenstorrent_pci_probe(struct pci_dev *dev, const struct pci_device_id
 	pci_enable_pcie_error_reporting(dev);
 
 	pci_set_drvdata(dev, tt_dev);
+	dev_set_drvdata(&tt_dev->dev, tt_dev);
 
 	tt_dev->interrupt_enabled = tenstorrent_enable_interrupts(tt_dev);
 
@@ -96,13 +97,26 @@ static int tenstorrent_pci_probe(struct pci_dev *dev, const struct pci_device_id
 		register_reboot_notifier(&tt_dev->reboot_notifier);
 	}
 
+	if (tt_dev->attributes) {
+		struct tt_attribute_data *data = tt_dev->attributes;
+		for (; data->attr.attr.name; data++)
+			device_create_file(&tt_dev->dev, &data->attr);
+	}
+
 	return 0;
 }
 
 static void tenstorrent_pci_remove(struct pci_dev *dev)
 {
 	struct tenstorrent_device *tt_dev = pci_get_drvdata(dev);
 
+	if (tt_dev->attributes) {
+		struct tt_attribute_data *data = tt_dev->attributes;
+		for (; data->attr.attr.name; data++)
+			device_remove_file(&tt_dev->dev, &data->attr);
+	}
+
+
 	// These remove child sysfs entries which must happen before remove returns.
 	tenstorrent_unregister_device(tt_dev);
 	tenstorrent_disable_interrupts(tt_dev);

diff --git a/grayskull.c b/grayskull.c
@@ -135,6 +135,18 @@ struct TLB_16M_REG {
 
 #define TLB_16M_SIZE_SHIFT 24
 
+
+static struct tt_attribute_data gs_attributes[] = {
+	{ __ATTR(tt_aiclk,  S_IRUGO, tt_show_attribute, NULL), 0x50, 0xFFFF },
+	{ __ATTR(tt_axiclk, S_IRUGO, tt_show_attribute, NULL), 0x54, 0xFFFF },
+	{ __ATTR(tt_arcclk, S_IRUGO, tt_show_attribute, NULL), 0x58, 0xFFFF },
+	{ __ATTR(tt_card_type, S_IRUGO, tt_show_card_type, NULL), 0x10, 0x0 },
+	{ __ATTR(tt_serial, S_IRUGO, tt_show_card_serial, NULL), 0x10, 0x0 },
+	{ __ATTR(tt_arc_fw_ver, S_IRUGO, tt_show_fw_ver, NULL), 0x18, 0xFFFFFF },
+	{ __ATTR(tt_ttflash_ver, S_IRUGO, tt_show_fw_ver, NULL), 0x98, 0x0 },
+	{ __ATTR_NULL, 0, 0 }
+};
+
 static u32 gs_arc_addr_to_sysreg(u32 arc_addr) {
 	return ARC_CSM_MEMORY_OFFSET + (arc_addr - 0x10000000);
 }
@@ -820,6 +832,8 @@ static bool grayskull_init_hardware(struct tenstorrent_device *tt_dev) {
 
 	grayskull_hwmon_init(gs_dev);
 
+	tt_dev->attributes = gs_attributes;
+
 	return true;
 }
 

diff --git a/hwmon.c b/hwmon.c
@@ -6,6 +6,7 @@
 #include <linux/stat.h>
 #include <asm/io.h>
 
+#include "device.h"
 #include "hwmon.h"
 
 static umode_t tt_hwmon_is_visible(const void *drvdata, enum hwmon_sensor_types type, u32 attr, int channel) {
@@ -69,3 +70,66 @@ const struct hwmon_ops tt_hwmon_ops = {
 	.read_string = tt_hwmon_read_string,
 };
 
+ssize_t tt_show_attribute(struct device *dev, struct device_attribute *attr, char *buf)
+{
+	struct tenstorrent_device *tt_dev = dev_get_drvdata(dev);
+	struct tt_attribute_data *data = container_of(attr, struct tt_attribute_data, attr);
+	struct tt_hwmon_context *ctx = &tt_dev->hwmon_context;
+	u32 value = ioread32(ctx->telemetry_base + data->reg_offset);
+	value &= data->mask;
+	return sprintf(buf, "%u\n", value);
+}
+
+ssize_t tt_show_card_type(struct device *dev, struct device_attribute *attr, char *buf) {
+	struct tenstorrent_device *tt_dev = dev_get_drvdata(dev);
+	struct tt_attribute_data *data = container_of(attr, struct tt_attribute_data, attr);
+	struct tt_hwmon_context *ctx = &tt_dev->hwmon_context;
+
+	u32 board_id_hi = ioread32(ctx->telemetry_base + data->reg_offset);
+	u16 card_type = (board_id_hi >> 4) & 0xFFFF;
+	char *card_name;
+	switch (card_type) {
+	case 0x3: card_name = "e150"; break;
+	case 0x7: card_name = "e75"; break;
+	case 0x14: card_name = "n300"; break;
+	case 0x18: card_name = "n150"; break;
+	default: card_name = "unknown"; break;
+	}
+
+	return scnprintf(buf, PAGE_SIZE, "%s\n", card_name);
+}
+
+ssize_t tt_show_card_serial(struct device *dev, struct device_attribute *attr, char *buf) {
+	struct tenstorrent_device *tt_dev = dev_get_drvdata(dev);
+	struct tt_attribute_data *data = container_of(attr, struct tt_attribute_data, attr);
+	struct tt_hwmon_context *ctx = &tt_dev->hwmon_context;
+
+	u32 board_id_hi = ioread32(ctx->telemetry_base + data->reg_offset);
+	u32 board_id_lo = ioread32(ctx->telemetry_base + data->reg_offset + 4);
+	return scnprintf(buf, PAGE_SIZE, "%08X%08X\n", board_id_hi, board_id_lo);
+}
+
+ssize_t tt_show_fw_ver(struct device *dev, struct device_attribute *attr, char *buf) {
+	struct tenstorrent_device *tt_dev = dev_get_drvdata(dev);
+	struct tt_attribute_data *data = container_of(attr, struct tt_attribute_data, attr);
+	struct tt_hwmon_context *ctx = &tt_dev->hwmon_context;
+
+	u32 fw_ver = ioread32(ctx->telemetry_base + data->reg_offset);
+	u32 major = (fw_ver >> 24) & 0xFF;
+	u32 minor = (fw_ver >> 16) & 0xFF;
+	u32 patch = (fw_ver >>  8) & 0xFF;
+	u32 ver = fw_ver & 0xFF;
+	return scnprintf(buf, PAGE_SIZE, "%u.%u.%u.%u\n", major, minor, patch, ver);
+}
+
+ssize_t tt_show_eth_fw_ver(struct device *dev, struct device_attribute *attr, char *buf) {
+	struct tenstorrent_device *tt_dev = dev_get_drvdata(dev);
+	struct tt_attribute_data *data = container_of(attr, struct tt_attribute_data, attr);
+	struct tt_hwmon_context *ctx = &tt_dev->hwmon_context;
+
+	u32 fw_ver = ioread32(ctx->telemetry_base + data->reg_offset);
+	u32 major = (fw_ver >> 16) & 0xFF;
+	u32 minor = (fw_ver >> 12) & 0xF;
+	u32 patch = (fw_ver >>  0) & 0xFFF;
+	return scnprintf(buf, PAGE_SIZE, "%u.%u.%u\n", major, minor, patch);
+}
diff --git a/hwmon.h b/hwmon.h
@@ -5,6 +5,7 @@
 #define TTDRIVER_HWMON_H_INCLUDED
 
 #include <linux/hwmon.h>
+#include <linux/device.h>
 
 // Sentinel for tt_hwmon_attr's reg_offset field.
 #define TT_HWMON_ATTR_END 0xffffffff
@@ -32,4 +33,17 @@ struct tt_hwmon_context {
 
 extern const struct hwmon_ops tt_hwmon_ops;
 
+struct tt_attribute_data {
+	struct device_attribute attr;
+	u32 reg_offset;
+	u32 mask;
+};
+
+ssize_t tt_show_attribute(struct device *dev, struct device_attribute *attr, char *buf);
+ssize_t tt_show_card_type(struct device *dev, struct device_attribute *attr, char *buf);
+ssize_t tt_show_card_serial(struct device *dev, struct device_attribute *attr, char *buf);
+ssize_t tt_show_fw_ver(struct device *dev, struct device_attribute *attr, char *buf);
+ssize_t tt_show_eth_fw_ver(struct device *dev, struct device_attribute *attr, char *buf);
+
+
 #endif
diff --git a/wormhole.c b/wormhole.c
@@ -49,6 +49,20 @@ static void write_iatu_reg(struct wormhole_device *wh_dev, unsigned direction,
 	iowrite32(value, wh_dev->bar2_mapping + offset);
 }
 
+static struct tt_attribute_data wh_attributes[] = {
+	{ __ATTR(tt_aiclk,  S_IRUGO, tt_show_attribute, NULL), 0x60, 0xFFFF },
+	{ __ATTR(tt_axiclk, S_IRUGO, tt_show_attribute, NULL), 0x64, 0xFFFF },
+	{ __ATTR(tt_arcclk, S_IRUGO, tt_show_attribute, NULL), 0x68, 0xFFFF },
+	{ __ATTR(tt_card_type, S_IRUGO, tt_show_card_type, NULL), 0x10, 0x0 },
+	{ __ATTR(tt_serial, S_IRUGO, tt_show_card_serial, NULL), 0x10, 0x0 },
+	{ __ATTR(tt_arc_fw_ver, S_IRUGO, tt_show_fw_ver, NULL), 0x18, 0x0 },
+	{ __ATTR(tt_eth_fw_ver, S_IRUGO, tt_show_eth_fw_ver, NULL), 0x2c, 0x0 },
+	{ __ATTR(tt_m3bl_fw_ver, S_IRUGO, tt_show_fw_ver, NULL), 0x30, 0x0 },
+	{ __ATTR(tt_m3app_fw_ver, S_IRUGO, tt_show_fw_ver, NULL), 0x34, 0x0 },
+	{ __ATTR(tt_ttflash_ver, S_IRUGO, tt_show_fw_ver, NULL), 0xb8, 0x0 },
+	{ __ATTR_NULL, 0, 0 }
+};
+
 static u32 wh_arc_addr_to_sysreg(u32 arc_addr) {
 	return ARC_CSM_START + (arc_addr - 0x10000000);
 }
@@ -166,6 +180,8 @@ static bool wormhole_init_hardware(struct tenstorrent_device *tt_dev) {
 
 	wormhole_hwmon_init(wh_dev);
 
+	tt_dev->attributes = wh_attributes;
+
 	return true;
 }