drm/amd: Add detailed GFXOFF stats to debugfs

Add debugfs interface to log GFXOFF statistics:

- Read amdgpu_gfxoff_count to get the total GFXOFF entry count at the
time of query since system power-up

- Write 1 to amdgpu_gfxoff_residency to start logging, and 0 to stop.
Read it to get average GFXOFF residency % multiplied by 100
during the last logging interval.

Both features are designed to be keep the values persistent between
suspends.

Signed-off-by: André Almeida <[e-mail]>
Signed-off-by: Alex Deucher <[e-mail]>

drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c

···
 }
 
 /**
+ * amdgpu_debugfs_gfxoff_residency_read - Read GFXOFF residency
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ *
+ * Read the last residency value logged. It doesn't auto update, one needs to
+ * stop logging before getting the current value.
+ */
+static ssize_t amdgpu_debugfs_gfxoff_residency_read(struct file *f, char __user *buf,
+						    size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+
+	if (size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	while (size) {
+		uint32_t value;
+
+		r = amdgpu_get_gfx_off_residency(adev, &value);
+		if (r)
+			goto out;
+
+		r = put_user(value, (uint32_t *)buf);
+		if (r)
+			goto out;
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+	r = result;
+out:
+	pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+
+	return r;
+}
+
+/**
+ * amdgpu_debugfs_gfxoff_residency_write - Log GFXOFF Residency
+ *
+ * @f: open file handle
+ * @buf: User buffer to write data from
+ * @size: Number of bytes to write
+ * @pos:  Offset to seek to
+ *
+ * Write a 32-bit non-zero to start logging; write a 32-bit zero to stop
+ */
+static ssize_t amdgpu_debugfs_gfxoff_residency_write(struct file *f, const char __user *buf,
+						     size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+
+	if (size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	while (size) {
+		u32 value;
+
+		r = get_user(value, (uint32_t *)buf);
+		if (r)
+			goto out;
+
+		amdgpu_set_gfx_off_residency(adev, value ? true : false);
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+	r = result;
+out:
+	pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+
+	return r;
+}
+
+
+/**
+ * amdgpu_debugfs_gfxoff_count_read - Read GFXOFF entry count
+ *
+ * @f: open file handle
+ * @buf: User buffer to store read data in
+ * @size: Number of bytes to read
+ * @pos:  Offset to seek to
+ */
+static ssize_t amdgpu_debugfs_gfxoff_count_read(struct file *f, char __user *buf,
+						size_t size, loff_t *pos)
+{
+	struct amdgpu_device *adev = file_inode(f)->i_private;
+	ssize_t result = 0;
+	int r;
+
+	if (size & 0x3 || *pos & 0x3)
+		return -EINVAL;
+
+	r = pm_runtime_get_sync(adev_to_drm(adev)->dev);
+	if (r < 0) {
+		pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+		return r;
+	}
+
+	while (size) {
+		u64 value = 0;
+
+		r = amdgpu_get_gfx_off_entrycount(adev, &value);
+		if (r)
+			goto out;
+
+		r = put_user(value, (u64 *)buf);
+		if (r)
+			goto out;
+
+		result += 4;
+		buf += 4;
+		*pos += 4;
+		size -= 4;
+	}
+
+	r = result;
+out:
+	pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
+	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
+
+	return r;
+}
+
+/**
  * amdgpu_debugfs_gfxoff_write - Enable/disable GFXOFF
  *
  * @f: open file handle
···
 	.llseek = default_llseek
 };
 
+static const struct file_operations amdgpu_debugfs_gfxoff_count_fops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_debugfs_gfxoff_count_read,
+	.llseek = default_llseek
+};
+
+static const struct file_operations amdgpu_debugfs_gfxoff_residency_fops = {
+	.owner = THIS_MODULE,
+	.read = amdgpu_debugfs_gfxoff_residency_read,
+	.write = amdgpu_debugfs_gfxoff_residency_write,
+	.llseek = default_llseek
+};
+
 static const struct file_operations *debugfs_regs[] = {
 	&amdgpu_debugfs_regs_fops,
 	&amdgpu_debugfs_regs2_fops,
···
 	&amdgpu_debugfs_gpr_fops,
 	&amdgpu_debugfs_gfxoff_fops,
 	&amdgpu_debugfs_gfxoff_status_fops,
+	&amdgpu_debugfs_gfxoff_count_fops,
+	&amdgpu_debugfs_gfxoff_residency_fops,
 };
 
 static const char *debugfs_regs_names[] = {
···
 	"amdgpu_gpr",
 	"amdgpu_gfxoff",
 	"amdgpu_gfxoff_status",
+	"amdgpu_gfxoff_count",
+	"amdgpu_gfxoff_residency",
 };
 
 /**

drivers/gpu/drm/amd/amdgpu/amdgpu_device.c

···
 	INIT_WORK(&adev->xgmi_reset_work, amdgpu_device_xgmi_reset_func);
 
 	adev->gfx.gfx_off_req_count = 1;
+	adev->gfx.gfx_off_residency = 0;
+	adev->gfx.gfx_off_entrycount = 0;
 	adev->pm.ac_power = power_supply_is_system_supplied() > 0;
 
 	atomic_set(&adev->throttling_logging_enabled, 1);

drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c

···
 	mutex_unlock(&adev->gfx.gfx_off_mutex);
 }
 
+int amdgpu_set_gfx_off_residency(struct amdgpu_device *adev, bool value)
+{
+	int r = 0;
+
+	mutex_lock(&adev->gfx.gfx_off_mutex);
+
+	r = amdgpu_dpm_set_residency_gfxoff(adev, value);
+
+	mutex_unlock(&adev->gfx.gfx_off_mutex);
+
+	return r;
+}
+
+int amdgpu_get_gfx_off_residency(struct amdgpu_device *adev, u32 *value)
+{
+	int r = 0;
+
+	mutex_lock(&adev->gfx.gfx_off_mutex);
+
+	r = amdgpu_dpm_get_residency_gfxoff(adev, value);
+
+	mutex_unlock(&adev->gfx.gfx_off_mutex);
+
+	return r;
+}
+
+int amdgpu_get_gfx_off_entrycount(struct amdgpu_device *adev, u64 *value)
+{
+	int r = 0;
+
+	mutex_lock(&adev->gfx.gfx_off_mutex);
+
+	r = amdgpu_dpm_get_entrycount_gfxoff(adev, value);
+
+	mutex_unlock(&adev->gfx.gfx_off_mutex);
+
+	return r;
+}
+
 int amdgpu_get_gfx_off_status(struct amdgpu_device *adev, uint32_t *value)
 {
 

drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h

···
 	struct mutex                    gfx_off_mutex;
 	uint32_t                        gfx_off_req_count; /* default 1, enable gfx off: dec 1, disable gfx off: add 1 */
 	struct delayed_work             gfx_off_delay_work;
+	uint32_t                        gfx_off_residency;
+	uint64_t                        gfx_off_entrycount;
 
 	/* pipe reservation */
 	struct mutex			pipe_reserve_mutex;
···
 void amdgpu_gfx_off_ctrl(struct amdgpu_device *adev, bool enable);
 int amdgpu_get_gfx_off_status(struct amdgpu_device *adev, uint32_t *value);
 int amdgpu_gfx_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block);
+void amdgpu_gfx_ras_fini(struct amdgpu_device *adev);
+int amdgpu_get_gfx_off_entrycount(struct amdgpu_device *adev, u64 *value);
+int amdgpu_get_gfx_off_residency(struct amdgpu_device *adev, u32 *residency);
+int amdgpu_set_gfx_off_residency(struct amdgpu_device *adev, bool value);
 int amdgpu_gfx_process_ras_data_cb(struct amdgpu_device *adev,
 		void *err_data,
 		struct amdgpu_iv_entry *entry);

drivers/gpu/drm/amd/pm/amdgpu_dpm.c

···
 	return ret;
 }
 
+int amdgpu_dpm_set_residency_gfxoff(struct amdgpu_device *adev, bool value)
+{
+	struct smu_context *smu = adev->powerplay.pp_handle;
+	int ret = 0;
+
+	if (!is_support_sw_smu(adev))
+		return -EOPNOTSUPP;
+
+	mutex_lock(&adev->pm.mutex);
+	ret = smu_set_residency_gfxoff(smu, value);
+	mutex_unlock(&adev->pm.mutex);
+
+	return ret;
+}
+
+int amdgpu_dpm_get_residency_gfxoff(struct amdgpu_device *adev, u32 *value)
+{
+	struct smu_context *smu = adev->powerplay.pp_handle;
+	int ret = 0;
+
+	if (!is_support_sw_smu(adev))
+		return -EOPNOTSUPP;
+
+	mutex_lock(&adev->pm.mutex);
+	ret = smu_get_residency_gfxoff(smu, value);
+	mutex_unlock(&adev->pm.mutex);
+
+	return ret;
+}
+
+int amdgpu_dpm_get_entrycount_gfxoff(struct amdgpu_device *adev, u64 *value)
+{
+	struct smu_context *smu = adev->powerplay.pp_handle;
+	int ret = 0;
+
+	if (!is_support_sw_smu(adev))
+		return -EOPNOTSUPP;
+
+	mutex_lock(&adev->pm.mutex);
+	ret = smu_get_entrycount_gfxoff(smu, value);
+	mutex_unlock(&adev->pm.mutex);
+
+	return ret;
+}
+
 int amdgpu_dpm_get_status_gfxoff(struct amdgpu_device *adev, uint32_t *value)
 {
 	struct smu_context *smu = adev->powerplay.pp_handle;

drivers/gpu/drm/amd/pm/inc/amdgpu_dpm.h

···
 int amdgpu_dpm_write_watermarks_table(struct amdgpu_device *adev);
 int amdgpu_dpm_wait_for_event(struct amdgpu_device *adev, enum smu_event_type event,
 		       uint64_t event_arg);
+int amdgpu_dpm_get_residency_gfxoff(struct amdgpu_device *adev, u32 *value);
+int amdgpu_dpm_set_residency_gfxoff(struct amdgpu_device *adev, bool value);
+int amdgpu_dpm_get_entrycount_gfxoff(struct amdgpu_device *adev, u64 *value);
 int amdgpu_dpm_get_status_gfxoff(struct amdgpu_device *adev, uint32_t *value);
 uint64_t amdgpu_dpm_get_thermal_throttling_counter(struct amdgpu_device *adev);
 void amdgpu_dpm_gfx_state_change(struct amdgpu_device *adev,

drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c

···
 	return smu_set_pp_feature_mask(smu, new_mask);
 }
 
+int smu_set_residency_gfxoff(struct smu_context *smu, bool value)
+{
+	if (!smu->ppt_funcs->set_gfx_off_residency)
+		return -EINVAL;
+
+	return smu_set_gfx_off_residency(smu, value);
+}
+
+int smu_get_residency_gfxoff(struct smu_context *smu, u32 *value)
+{
+	if (!smu->ppt_funcs->get_gfx_off_residency)
+		return -EINVAL;
+
+	return smu_get_gfx_off_residency(smu, value);
+}
+
+int smu_get_entrycount_gfxoff(struct smu_context *smu, u64 *value)
+{
+	if (!smu->ppt_funcs->get_gfx_off_entrycount)
+		return -EINVAL;
+
+	return smu_get_gfx_off_entrycount(smu, value);
+}
+
 int smu_get_status_gfxoff(struct smu_context *smu, uint32_t *value)
 {
 	if (!smu->ppt_funcs->get_gfx_off_status)
···
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 	struct smu_context *smu = adev->powerplay.pp_handle;
 	int ret;
+	uint64_t count;
 
 	if (amdgpu_sriov_vf(adev)&& !amdgpu_sriov_is_pp_one_vf(adev))
 		return 0;
···
 	smu->watermarks_bitmap &= ~(WATERMARKS_LOADED);
 
 	smu_set_gfx_cgpg(smu, false);
+
+	/*
+	 * pwfw resets entrycount when device is suspended, so we save the
+	 * last value to be used when we resume to keep it consistent
+	 */
+	ret = smu_get_entrycount_gfxoff(smu, &count);
+	if (!ret)
+		adev->gfx.gfx_off_entrycount = count;
 
 	return 0;
 }

drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h

···
 	uint32_t (*get_gfx_off_status)(struct smu_context *smu);
 
 	/**
+	 * @gfx_off_entrycount: total GFXOFF entry count at the time of
+	 * query since system power-up
+	 */
+	u32 (*get_gfx_off_entrycount)(struct smu_context *smu, uint64_t *entrycount);
+
+	/**
+	 * @set_gfx_off_residency: set 1 to start logging, 0 to stop logging
+	 */
+	u32 (*set_gfx_off_residency)(struct smu_context *smu, bool start);
+
+	/**
+	 * @get_gfx_off_residency: Average GFXOFF residency % during the logging interval
+	 */
+	u32 (*get_gfx_off_residency)(struct smu_context *smu, uint32_t *residency);
+
+	/**
 	 * @register_irq_handler: Register interupt request handlers.
 	 */
 	int (*register_irq_handler)(struct smu_context *smu);
···
 int smu_set_ac_dc(struct smu_context *smu);
 
 int smu_allow_xgmi_power_down(struct smu_context *smu, bool en);
+
+int smu_get_entrycount_gfxoff(struct smu_context *smu, u64 *value);
+
+int smu_get_residency_gfxoff(struct smu_context *smu, u32 *value);
+
+int smu_set_residency_gfxoff(struct smu_context *smu, bool value);
 
 int smu_get_status_gfxoff(struct smu_context *smu, uint32_t *value);
 

drivers/gpu/drm/amd/pm/swsmu/smu_internal.h

···
 #define smu_notify_memory_pool_location(smu)				smu_ppt_funcs(notify_memory_pool_location, 0, smu)
 #define smu_gfx_off_control(smu, enable)				smu_ppt_funcs(gfx_off_control, 0, smu, enable)
 #define smu_get_gfx_off_status(smu)						smu_ppt_funcs(get_gfx_off_status, 0, smu)
+#define smu_get_gfx_off_entrycount(smu, value)						smu_ppt_funcs(get_gfx_off_entrycount, 0, smu, value)
+#define smu_get_gfx_off_residency(smu, value)						smu_ppt_funcs(get_gfx_off_residency, 0, smu, value)
+#define smu_set_gfx_off_residency(smu, value)						smu_ppt_funcs(set_gfx_off_residency, 0, smu, value)
 #define smu_set_last_dcef_min_deep_sleep_clk(smu)			smu_ppt_funcs(set_last_dcef_min_deep_sleep_clk, 0, smu)
 #define smu_system_features_control(smu, en)				smu_ppt_funcs(system_features_control, 0, smu, en)
 #define smu_init_max_sustainable_clocks(smu)				smu_ppt_funcs(init_max_sustainable_clocks, 0, smu)