drm/amd/pm: do not expose implementation details to other blocks out of power
authorEvan Quan <evan.quan@amd.com>
Mon, 22 Nov 2021 02:57:20 +0000 (10:57 +0800)
committerAlex Deucher <alexander.deucher@amd.com>
Fri, 14 Jan 2022 22:51:14 +0000 (17:51 -0500)
Those implementation details(whether swsmu supported, some ppt_funcs supported,
accessing internal statistics ...)should be kept internally. It's not a good
practice and even error prone to expose implementation details.

Signed-off-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
13 files changed:
drivers/gpu/drm/amd/amdgpu/aldebaran.c
drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c
drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
drivers/gpu/drm/amd/include/kgd_pp_interface.h
drivers/gpu/drm/amd/pm/amdgpu_dpm.c
drivers/gpu/drm/amd/pm/inc/amdgpu_dpm.h
drivers/gpu/drm/amd/pm/inc/amdgpu_smu.h
drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c

index bcfdb63..a545df4 100644 (file)
@@ -260,7 +260,7 @@ static int aldebaran_mode2_restore_ip(struct amdgpu_device *adev)
        adev->gfx.rlc.funcs->resume(adev);
 
        /* Wait for FW reset event complete */
-       r = smu_wait_for_event(adev, SMU_EVENT_RESET_COMPLETE, 0);
+       r = amdgpu_dpm_wait_for_event(adev, SMU_EVENT_RESET_COMPLETE, 0);
        if (r) {
                dev_err(adev->dev,
                        "Failed to get response from firmware after reset\n");
index 25e2e5b..9aea1cc 100644 (file)
@@ -1585,22 +1585,25 @@ static int amdgpu_debugfs_sclk_set(void *data, u64 val)
                return ret;
        }
 
-       if (is_support_sw_smu(adev)) {
-               ret = smu_get_dpm_freq_range(&adev->smu, SMU_SCLK, &min_freq, &max_freq);
-               if (ret || val > max_freq || val < min_freq)
-                       return -EINVAL;
-               ret = smu_set_soft_freq_range(&adev->smu, SMU_SCLK, (uint32_t)val, (uint32_t)val);
-       } else {
-               return 0;
+       ret = amdgpu_dpm_get_dpm_freq_range(adev, PP_SCLK, &min_freq, &max_freq);
+       if (ret == -EOPNOTSUPP) {
+               ret = 0;
+               goto out;
        }
+       if (ret || val > max_freq || val < min_freq) {
+               ret = -EINVAL;
+               goto out;
+       }
+
+       ret = amdgpu_dpm_set_soft_freq_range(adev, PP_SCLK, (uint32_t)val, (uint32_t)val);
+       if (ret)
+               ret = -EINVAL;
 
+out:
        pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
        pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
 
-       if (ret)
-               return -EINVAL;
-
-       return 0;
+       return ret;
 }
 
 DEFINE_DEBUGFS_ATTRIBUTE(fops_ib_preempt, NULL,
index cf7fad8..596bb2f 100644 (file)
@@ -2624,7 +2624,7 @@ static int amdgpu_device_ip_late_init(struct amdgpu_device *adev)
        /* For passthrough configuration on arcturus and aldebaran, enable special handling SBR */
        if (amdgpu_passthrough(adev) && ((adev->asic_type == CHIP_ARCTURUS && adev->gmc.xgmi.num_physical_nodes > 1)||
                               adev->asic_type == CHIP_ALDEBARAN ))
-               smu_handle_passthrough_sbr(&adev->smu, true);
+               amdgpu_dpm_handle_passthrough_sbr(adev, true);
 
        if (adev->gmc.xgmi.num_physical_nodes > 1) {
                mutex_lock(&mgpu_info.mutex);
@@ -2881,7 +2881,7 @@ static int amdgpu_device_ip_suspend_phase2(struct amdgpu_device *adev)
        int i, r;
 
        if (adev->in_s0ix)
-               amdgpu_gfx_state_change_set(adev, sGpuChangeState_D3Entry);
+               amdgpu_dpm_gfx_state_change(adev, sGpuChangeState_D3Entry);
 
        for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
                if (!adev->ip_blocks[i].status.valid)
@@ -4044,7 +4044,7 @@ int amdgpu_device_resume(struct drm_device *dev, bool fbcon)
                return 0;
 
        if (adev->in_s0ix)
-               amdgpu_gfx_state_change_set(adev, sGpuChangeState_D0Entry);
+               amdgpu_dpm_gfx_state_change(adev, sGpuChangeState_D0Entry);
 
        /* post card */
        if (amdgpu_device_need_post(adev)) {
index 1916ec8..3d8f82d 100644 (file)
@@ -615,7 +615,7 @@ int amdgpu_get_gfx_off_status(struct amdgpu_device *adev, uint32_t *value)
 
        mutex_lock(&adev->gfx.gfx_off_mutex);
 
-       r = smu_get_status_gfxoff(adev, value);
+       r = amdgpu_dpm_get_status_gfxoff(adev, value);
 
        mutex_unlock(&adev->gfx.gfx_off_mutex);
 
@@ -852,19 +852,3 @@ int amdgpu_gfx_get_num_kcq(struct amdgpu_device *adev)
        }
        return amdgpu_num_kcq;
 }
-
-/* amdgpu_gfx_state_change_set - Handle gfx power state change set
- * @adev: amdgpu_device pointer
- * @state: gfx power state(1 -sGpuChangeState_D0Entry and 2 -sGpuChangeState_D3Entry)
- *
- */
-
-void amdgpu_gfx_state_change_set(struct amdgpu_device *adev, enum gfx_change_state state)
-{
-       mutex_lock(&adev->pm.mutex);
-       if (adev->powerplay.pp_funcs &&
-           adev->powerplay.pp_funcs->gfx_state_change_set)
-               ((adev)->powerplay.pp_funcs->gfx_state_change_set(
-                       (adev)->powerplay.pp_handle, state));
-       mutex_unlock(&adev->pm.mutex);
-}
index f851196..776c886 100644 (file)
@@ -47,12 +47,6 @@ enum amdgpu_gfx_pipe_priority {
        AMDGPU_GFX_PIPE_PRIO_HIGH = AMDGPU_RING_PRIO_2
 };
 
-/* Argument for PPSMC_MSG_GpuChangeState */
-enum gfx_change_state {
-       sGpuChangeState_D0Entry = 1,
-       sGpuChangeState_D3Entry,
-};
-
 #define AMDGPU_GFX_QUEUE_PRIORITY_MINIMUM  0
 #define AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM  15
 
@@ -410,5 +404,4 @@ int amdgpu_gfx_cp_ecc_error_irq(struct amdgpu_device *adev,
 uint32_t amdgpu_kiq_rreg(struct amdgpu_device *adev, uint32_t reg);
 void amdgpu_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v);
 int amdgpu_gfx_get_num_kcq(struct amdgpu_device *adev);
-void amdgpu_gfx_state_change_set(struct amdgpu_device *adev, enum gfx_change_state state);
 #endif
index 586a30a..4a99704 100644 (file)
@@ -901,7 +901,7 @@ static void amdgpu_ras_get_ecc_info(struct amdgpu_device *adev, struct ras_err_d
         * choosing right query method according to
         * whether smu support query error information
         */
-       ret = smu_get_ecc_info(&adev->smu, (void *)&(ras->umc_ecc));
+       ret = amdgpu_dpm_get_ecc_info(adev, (void *)&(ras->umc_ecc));
        if (ret == -EOPNOTSUPP) {
                if (adev->umc.ras_funcs &&
                        adev->umc.ras_funcs->query_ras_error_count)
@@ -2141,8 +2141,7 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev)
                if (ret)
                        goto free;
 
-               if (adev->smu.ppt_funcs && adev->smu.ppt_funcs->send_hbm_bad_pages_num)
-                       adev->smu.ppt_funcs->send_hbm_bad_pages_num(&adev->smu, con->eeprom_control.ras_num_recs);
+               amdgpu_dpm_send_hbm_bad_pages_num(adev, con->eeprom_control.ras_num_recs);
        }
 
 #ifdef CONFIG_X86_MCE_AMD
index 46264a4..25951b2 100644 (file)
@@ -33,7 +33,7 @@ static int amdgpu_umc_do_page_retirement(struct amdgpu_device *adev,
        int ret = 0;
 
        kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
-       ret = smu_get_ecc_info(&adev->smu, (void *)&(con->umc_ecc));
+       ret = amdgpu_dpm_get_ecc_info(adev, (void *)&(con->umc_ecc));
        if (ret == -EOPNOTSUPP) {
                if (adev->umc.ras_funcs &&
                    adev->umc.ras_funcs->query_ras_error_count)
@@ -96,8 +96,7 @@ static int amdgpu_umc_do_page_retirement(struct amdgpu_device *adev,
                                                err_data->err_addr_cnt);
                        amdgpu_ras_save_bad_pages(adev);
 
-                       if (adev->smu.ppt_funcs && adev->smu.ppt_funcs->send_hbm_bad_pages_num)
-                               adev->smu.ppt_funcs->send_hbm_bad_pages_num(&adev->smu, con->eeprom_control.ras_num_recs);
+                       amdgpu_dpm_send_hbm_bad_pages_num(adev, con->eeprom_control.ras_num_recs);
                }
 
                if (reset)
index deae12d..329a4c8 100644 (file)
@@ -222,7 +222,7 @@ void kfd_smi_event_update_thermal_throttling(struct kfd_dev *dev,
 
        len = snprintf(fifo_in, sizeof(fifo_in), "%x %llx:%llx\n",
                       KFD_SMI_EVENT_THERMAL_THROTTLE, throttle_bitmask,
-                      atomic64_read(&dev->adev->smu.throttle_int_counter));
+                      amdgpu_dpm_get_thermal_throttling_counter(dev->adev));
 
        add_event_to_kfifo(dev, KFD_SMI_EVENT_THERMAL_THROTTLE, fifo_in, len);
 }
index 5c0867e..2e295fa 100644 (file)
 
 extern const struct amdgpu_ip_block_version pp_smu_ip_block;
 
+enum smu_event_type {
+       SMU_EVENT_RESET_COMPLETE = 0,
+};
+
 struct amd_vce_state {
        /* vce clocks */
        u32 evclk;
index 08362d5..73f3d29 100644 (file)
@@ -1614,3 +1614,93 @@ int amdgpu_pm_load_smu_firmware(struct amdgpu_device *adev, uint32_t *smu_versio
 
        return 0;
 }
+
+int amdgpu_dpm_handle_passthrough_sbr(struct amdgpu_device *adev, bool enable)
+{
+       return smu_handle_passthrough_sbr(&adev->smu, enable);
+}
+
+int amdgpu_dpm_send_hbm_bad_pages_num(struct amdgpu_device *adev, uint32_t size)
+{
+       return smu_send_hbm_bad_pages_num(&adev->smu, size);
+}
+
+int amdgpu_dpm_get_dpm_freq_range(struct amdgpu_device *adev,
+                                 enum pp_clock_type type,
+                                 uint32_t *min,
+                                 uint32_t *max)
+{
+       if (!is_support_sw_smu(adev))
+               return -EOPNOTSUPP;
+
+       switch (type) {
+       case PP_SCLK:
+               return smu_get_dpm_freq_range(&adev->smu, SMU_SCLK, min, max);
+       default:
+               return -EINVAL;
+       }
+}
+
+int amdgpu_dpm_set_soft_freq_range(struct amdgpu_device *adev,
+                                  enum pp_clock_type type,
+                                  uint32_t min,
+                                  uint32_t max)
+{
+       if (!is_support_sw_smu(adev))
+               return -EOPNOTSUPP;
+
+       switch (type) {
+       case PP_SCLK:
+               return smu_set_soft_freq_range(&adev->smu, SMU_SCLK, min, max);
+       default:
+               return -EINVAL;
+       }
+}
+
+int amdgpu_dpm_wait_for_event(struct amdgpu_device *adev,
+                             enum smu_event_type event,
+                             uint64_t event_arg)
+{
+       if (!is_support_sw_smu(adev))
+               return -EOPNOTSUPP;
+
+       return smu_wait_for_event(&adev->smu, event, event_arg);
+}
+
+int amdgpu_dpm_get_status_gfxoff(struct amdgpu_device *adev, uint32_t *value)
+{
+       if (!is_support_sw_smu(adev))
+               return -EOPNOTSUPP;
+
+       return smu_get_status_gfxoff(&adev->smu, value);
+}
+
+uint64_t amdgpu_dpm_get_thermal_throttling_counter(struct amdgpu_device *adev)
+{
+       return atomic64_read(&adev->smu.throttle_int_counter);
+}
+
+/* amdgpu_dpm_gfx_state_change - Handle gfx power state change set
+ * @adev: amdgpu_device pointer
+ * @state: gfx power state(1 -sGpuChangeState_D0Entry and 2 -sGpuChangeState_D3Entry)
+ *
+ */
+void amdgpu_dpm_gfx_state_change(struct amdgpu_device *adev,
+                                enum gfx_change_state state)
+{
+       mutex_lock(&adev->pm.mutex);
+       if (adev->powerplay.pp_funcs &&
+           adev->powerplay.pp_funcs->gfx_state_change_set)
+               ((adev)->powerplay.pp_funcs->gfx_state_change_set(
+                       (adev)->powerplay.pp_handle, state));
+       mutex_unlock(&adev->pm.mutex);
+}
+
+int amdgpu_dpm_get_ecc_info(struct amdgpu_device *adev,
+                           void *umc_ecc)
+{
+       if (!is_support_sw_smu(adev))
+               return -EOPNOTSUPP;
+
+       return smu_get_ecc_info(&adev->smu, umc_ecc);
+}
index c464a04..b444937 100644 (file)
 #ifndef __AMDGPU_DPM_H__
 #define __AMDGPU_DPM_H__
 
+/* Argument for PPSMC_MSG_GpuChangeState */
+enum gfx_change_state {
+       sGpuChangeState_D0Entry = 1,
+       sGpuChangeState_D3Entry,
+};
+
 enum amdgpu_int_thermal_type {
        THERMAL_TYPE_NONE,
        THERMAL_TYPE_EXTERNAL,
@@ -582,5 +588,22 @@ void amdgpu_dpm_enable_vce(struct amdgpu_device *adev, bool enable);
 void amdgpu_dpm_enable_jpeg(struct amdgpu_device *adev, bool enable);
 void amdgpu_pm_print_power_states(struct amdgpu_device *adev);
 int amdgpu_pm_load_smu_firmware(struct amdgpu_device *adev, uint32_t *smu_version);
-
+int amdgpu_dpm_handle_passthrough_sbr(struct amdgpu_device *adev, bool enable);
+int amdgpu_dpm_send_hbm_bad_pages_num(struct amdgpu_device *adev, uint32_t size);
+int amdgpu_dpm_get_dpm_freq_range(struct amdgpu_device *adev,
+                                      enum pp_clock_type type,
+                                      uint32_t *min,
+                                      uint32_t *max);
+int amdgpu_dpm_set_soft_freq_range(struct amdgpu_device *adev,
+                                       enum pp_clock_type type,
+                                       uint32_t min,
+                                       uint32_t max);
+int amdgpu_dpm_wait_for_event(struct amdgpu_device *adev, enum smu_event_type event,
+                      uint64_t event_arg);
+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,
+                                enum gfx_change_state state);
+int amdgpu_dpm_get_ecc_info(struct amdgpu_device *adev,
+                           void *umc_ecc);
 #endif
index ba7565b..b90ed0e 100644 (file)
@@ -241,11 +241,6 @@ struct smu_user_dpm_profile {
        uint32_t clk_dependency;
 };
 
-enum smu_event_type {
-
-       SMU_EVENT_RESET_COMPLETE = 0,
-};
-
 #define SMU_TABLE_INIT(tables, table_id, s, a, d)      \
        do {                                            \
                tables[table_id].size = s;              \
@@ -1413,15 +1408,15 @@ int smu_set_ac_dc(struct smu_context *smu);
 
 int smu_allow_xgmi_power_down(struct smu_context *smu, bool en);
 
-int smu_get_status_gfxoff(struct amdgpu_device *adev, uint32_t *value);
+int smu_get_status_gfxoff(struct smu_context *smu, uint32_t *value);
 
 int smu_handle_passthrough_sbr(struct smu_context *smu, bool enable);
 
-int smu_wait_for_event(struct amdgpu_device *adev, enum smu_event_type event,
+int smu_wait_for_event(struct smu_context *smu, enum smu_event_type event,
                       uint64_t event_arg);
 int smu_get_ecc_info(struct smu_context *smu, void *umc_ecc);
 int smu_stb_collect_info(struct smu_context *smu, void *buff, uint32_t size);
 void amdgpu_smu_stb_debug_fs_init(struct amdgpu_device *adev);
-
+int smu_send_hbm_bad_pages_num(struct smu_context *smu, uint32_t size);
 #endif
 #endif
index d93d28c..9d5a5a1 100644 (file)
@@ -100,17 +100,14 @@ static int smu_sys_set_pp_feature_mask(void *handle,
        return ret;
 }
 
-int smu_get_status_gfxoff(struct amdgpu_device *adev, uint32_t *value)
+int smu_get_status_gfxoff(struct smu_context *smu, uint32_t *value)
 {
-       int ret = 0;
-       struct smu_context *smu = &adev->smu;
+       if (!smu->ppt_funcs->get_gfx_off_status)
+               return -EINVAL;
 
-       if (is_support_sw_smu(adev) && smu->ppt_funcs->get_gfx_off_status)
-               *value = smu_get_gfx_off_status(smu);
-       else
-               ret = -EINVAL;
+       *value = smu_get_gfx_off_status(smu);
 
-       return ret;
+       return 0;
 }
 
 int smu_set_soft_freq_range(struct smu_context *smu,
@@ -3165,11 +3162,10 @@ static const struct amd_pm_funcs swsmu_pm_funcs = {
        .get_smu_prv_buf_details = smu_get_prv_buffer_details,
 };
 
-int smu_wait_for_event(struct amdgpu_device *adev, enum smu_event_type event,
+int smu_wait_for_event(struct smu_context *smu, enum smu_event_type event,
                       uint64_t event_arg)
 {
        int ret = -EINVAL;
-       struct smu_context *smu = &adev->smu;
 
        if (smu->ppt_funcs->wait_for_event) {
                mutex_lock(&smu->mutex);
@@ -3283,3 +3279,13 @@ void amdgpu_smu_stb_debug_fs_init(struct amdgpu_device *adev)
 #endif
 
 }
+
+int smu_send_hbm_bad_pages_num(struct smu_context *smu, uint32_t size)
+{
+       int ret = 0;
+
+       if (smu->ppt_funcs && smu->ppt_funcs->send_hbm_bad_pages_num)
+               ret = smu->ppt_funcs->send_hbm_bad_pages_num(smu, size);
+
+       return ret;
+}