gpu: nvgpu: gv10x therm boardobj support

- Added support for below multiple therm sensor device & defined macros
   GPC_TSOSC
   GPC SCI
   HBM2_SITE
   HBM2_COMBINED
- Added PMU interface for listed therm sensor device
- Added nvgpu interface for listed therm sensor device
- Added construct boardobj support for listed therm sensor device
  & called to update nvgpu interface.
- Updated devinit_get_therm_device_table() to read sensor info from
  therm device table from vbios table & construct respective
  therm device boardobj using construct_therm_device_*()
  based on class_id param read from vbios table.
- Updated RPC handler to handle THERM ack request
- Updated gv100 therm ops "get_internal_sensor_limits"
  to point to gp106_get_internal_sensor_limits()

Change-Id: I4b4ed501d0625cb8fc7b300c820622e40ae59fe6
Signed-off-by: Mahantesh Kumbar <mkumbar@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1676785
Reviewed-by: svc-misra-checker <svc-misra-checker@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vaikundanathan S <vaikuns@nvidia.com>
Tested-by: Vaikundanathan S <vaikuns@nvidia.com>
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

5 files changed, 249 insertions, 29 deletions

diff --git a/drivers/gpu/nvgpu/common/pmu/pmu_ipc.c b/drivers/gpu/nvgpu/common/pmu/pmu_ipc.c
index 68654a70..843a4551 100644
--- a/drivers/gpu/nvgpu/common/pmu/pmu_ipc.c
+++ b/drivers/gpu/nvgpu/common/pmu/pmu_ipc.c
@@ -841,8 +841,16 @@ static void pmu_rpc_handler(struct gk20a *g, struct pmu_msg *msg,
                 nvgpu_pmu_dbg(g, "reply PMU_UNIT_PERF");
                 break;
         case PMU_UNIT_THERM:
-                nvgpu_pmu_dbg(g, "reply PMU_UNIT_THERM");
-                break;
+                        switch (rpc.function) {
+                        case NV_PMU_RPC_ID_THERM_BOARD_OBJ_GRP_CMD:
+                                nvgpu_pmu_dbg(g,
+                                        "reply NV_PMU_RPC_ID_THERM_BOARD_OBJ_GRP_CMD");
+                                break;
+                        default:
+                                nvgpu_pmu_dbg(g, "reply PMU_UNIT_THERM");
+                                break;
+                        }
+                        break;
                 /* TBD case will be added */
         default:
                 nvgpu_err(g, " Invalid RPC response, stats 0x%x",
diff --git a/drivers/gpu/nvgpu/include/nvgpu/bios.h b/drivers/gpu/nvgpu/include/nvgpu/bios.h
index 0619fcb9..0323dce4 100644
--- a/drivers/gpu/nvgpu/include/nvgpu/bios.h
+++ b/drivers/gpu/nvgpu/include/nvgpu/bios.h
@@ -890,7 +890,12 @@ struct therm_device_1x_entry {
         u8 flags;
 } ;
 
-#define NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_GPU                               0x01
+#define NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_INVALID            0x00
+#define NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_GPU                        0x01
+#define NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_GPU_GPC_TSOSC      0x02
+#define NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_GPU_GPC_SCI        0x03
+#define NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_HBM2_SITE          0x70
+#define NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_HBM2_COMBINED      0x71
 
 #define NV_VBIOS_THERM_DEVICE_1X_ENTRY_PARAM0_I2C_DEVICE_INDEX_MASK        0xFF
 #define NV_VBIOS_THERM_DEVICE_1X_ENTRY_PARAM0_I2C_DEVICE_INDEX_SHIFT          0
diff --git a/drivers/gpu/nvgpu/include/nvgpu/pmuif/gpmuifthermsensor.h b/drivers/gpu/nvgpu/include/nvgpu/pmuif/gpmuifthermsensor.h
index 994cd5b7..8f2a8584 100644
--- a/drivers/gpu/nvgpu/include/nvgpu/pmuif/gpmuifthermsensor.h
+++ b/drivers/gpu/nvgpu/include/nvgpu/pmuif/gpmuifthermsensor.h
@@ -41,15 +41,37 @@ struct nv_pmu_therm_therm_device_boardobj_set {
         struct nv_pmu_boardobj super;
 };
 
+struct nv_pmu_therm_therm_device_gpu_gpc_tsosc_boardobj_set {
+        struct nv_pmu_therm_therm_device_boardobj_set super;
+        u8 gpc_tsosc_idx;
+};
+
+struct nv_pmu_therm_therm_device_gpu_sci_boardobj_set {
+        struct nv_pmu_therm_therm_device_boardobj_set super;
+};
+
 struct nv_pmu_therm_therm_device_i2c_boardobj_set {
         struct nv_pmu_therm_therm_device_boardobj_set super;
         u8 i2c_dev_idx;
 };
 
+struct nv_pmu_therm_therm_device_hbm2_site_boardobj_set {
+        struct nv_pmu_therm_therm_device_boardobj_set super;
+        u8 site_idx;
+};
+
+struct nv_pmu_therm_therm_device_hbm2_combined_boardobj_set {
+        struct nv_pmu_therm_therm_device_boardobj_set super;
+};
+
 union nv_pmu_therm_therm_device_boardobj_set_union {
         struct nv_pmu_boardobj board_obj;
         struct nv_pmu_therm_therm_device_boardobj_set therm_device;
+        struct nv_pmu_therm_therm_device_gpu_gpc_tsosc_boardobj_set gpu_gpc_tsosc;
+        struct nv_pmu_therm_therm_device_gpu_sci_boardobj_set gpu_sci;
         struct nv_pmu_therm_therm_device_i2c_boardobj_set i2c;
+        struct nv_pmu_therm_therm_device_hbm2_site_boardobj_set hbm2_site;
+        struct nv_pmu_therm_therm_device_hbm2_combined_boardobj_set hbm2_combined;
 };
 
 NV_PMU_BOARDOBJ_GRP_SET_MAKE_E32(therm, therm_device);
diff --git a/drivers/gpu/nvgpu/therm/thrmdev.c b/drivers/gpu/nvgpu/therm/thrmdev.c
index 7371ba25..aaf0a461 100644
--- a/drivers/gpu/nvgpu/therm/thrmdev.c
+++ b/drivers/gpu/nvgpu/therm/thrmdev.c
@@ -30,22 +30,6 @@
 #include "gp106/bios_gp106.h"
 #include "ctrl/ctrltherm.h"
 
-static struct boardobj *construct_therm_device(struct gk20a *g,
-                        void *pargs, u16 pargs_size, u8 type)
-{
-        struct boardobj *board_obj_ptr = NULL;
-        u32 status;
-
-        status = boardobj_construct_super(g, &board_obj_ptr,
-                pargs_size, pargs);
-        if (status)
-                return NULL;
-
-        nvgpu_log_info(g, " Done");
-
-        return board_obj_ptr;
-}
-
 static u32 _therm_device_pmudata_instget(struct gk20a *g,
                         struct nv_pmu_boardobjgrp *pmuboardobjgrp,
                         struct nv_pmu_boardobj **ppboardobjpmudata,
@@ -70,6 +54,168 @@ static u32 _therm_device_pmudata_instget(struct gk20a *g,
         return 0;
 }
 
+static u32 construct_therm_device(struct gk20a *g,
+        struct boardobj **ppboardobj, u16 size, void *pargs)
+{
+        return boardobj_construct_super(g, ppboardobj, size, pargs);
+}
+
+static u32 construct_therm_device_gpu(struct gk20a *g,
+        struct boardobj **ppboardobj, u16 size, void *pargs)
+{
+        return construct_therm_device(g, ppboardobj, size, pargs);
+}
+
+static u32 construct_therm_device_gpu_sci(struct gk20a *g,
+        struct boardobj **ppboardobj, u16 size, void *pargs)
+{
+        return construct_therm_device(g, ppboardobj, size, pargs);
+}
+
+
+static u32 therm_device_pmu_data_init_gpu_gpc_tsosc(struct gk20a *g,
+        struct boardobj *pboard_obj, struct nv_pmu_boardobj *ppmudata)
+{
+        u32 status = 0;
+        struct therm_device_gpu_gpc_tsosc *pdev = NULL;
+        struct nv_pmu_therm_therm_device_gpu_gpc_tsosc_boardobj_set *pset;
+
+        status = boardobj_pmudatainit_super(g, pboard_obj, ppmudata);
+        if (status != 0) {
+                goto exit;
+        }
+
+        pdev = (struct therm_device_gpu_gpc_tsosc *)(void *)pboard_obj;
+        pset = (struct nv_pmu_therm_therm_device_gpu_gpc_tsosc_boardobj_set *)
+                (void*) ppmudata;
+
+        pset->gpc_tsosc_idx = pdev->gpc_tsosc_idx;
+
+exit:
+        return status;
+}
+
+static u32 construct_therm_device_gpu_tsosc(struct gk20a *g,
+        struct boardobj **ppboardobj, u16 size, void *pargs)
+{
+        struct therm_device_gpu_gpc_tsosc *pdev = NULL;
+        struct therm_device_gpu_gpc_tsosc *ptmp_dev =
+                (struct therm_device_gpu_gpc_tsosc *)pargs;
+        u32 status = 0;
+
+        status = construct_therm_device(g, ppboardobj, size, pargs);
+        if (status != 0) {
+                return status;
+        }
+
+        pdev = (struct therm_device_gpu_gpc_tsosc *)(void *)*ppboardobj;
+
+        pdev->super.super.pmudatainit =
+                therm_device_pmu_data_init_gpu_gpc_tsosc;
+
+        pdev->gpc_tsosc_idx = ptmp_dev->gpc_tsosc_idx;
+
+        return status;
+}
+
+static u32 therm_device_pmu_data_init_hbm2_site(struct gk20a *g,
+        struct boardobj *pboard_obj, struct nv_pmu_boardobj *ppmudata)
+{
+        u32 status = 0;
+        struct therm_device_hbm2_site *pdev = NULL;
+        struct nv_pmu_therm_therm_device_hbm2_site_boardobj_set *pset;
+
+        status = boardobj_pmudatainit_super(g, pboard_obj, ppmudata);
+        if (status != 0) {
+                goto exit;
+        }
+
+        pdev = (struct therm_device_hbm2_site *)(void *)pboard_obj;
+        pset = (struct nv_pmu_therm_therm_device_hbm2_site_boardobj_set *)
+                (void *)ppmudata;
+
+        pset->site_idx = pdev->site_idx;
+
+exit:
+        return status;
+}
+
+static u32 construct_therm_device_hbm2_site(struct gk20a *g,
+        struct boardobj **ppboardobj, u16 size, void *pargs)
+{
+        struct therm_device_hbm2_site *pdev = NULL;
+        struct therm_device_hbm2_site *ptmp_dev =
+                (struct therm_device_hbm2_site *)pargs;
+        u32 status = 0;
+
+        status = construct_therm_device(g, ppboardobj, size, pargs);
+        if (status != 0) {
+                return status;
+        }
+
+        pdev = (struct therm_device_hbm2_site *)(void *)*ppboardobj;
+
+        pdev->super.super.pmudatainit =
+                therm_device_pmu_data_init_hbm2_site;
+
+        pdev->site_idx = ptmp_dev->site_idx;
+
+        return status;
+}
+
+static u32 construct_therm_device_hbm2_combined(struct gk20a *g,
+        struct boardobj **ppboardobj, u16 size, void *pargs)
+{
+        return construct_therm_device(g, ppboardobj, size, pargs);
+}
+
+
+static struct boardobj *therm_device_construct(struct gk20a *g,
+        void *pargs)
+{
+        struct boardobj *board_obj_ptr = NULL;
+        u32 status = 0;
+
+        switch (BOARDOBJ_GET_TYPE(pargs)) {
+        case NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_GPU:
+                status = construct_therm_device_gpu(g, &board_obj_ptr,
+                        sizeof(struct therm_device), pargs);
+                break;
+        case NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_GPU_GPC_SCI:
+                status = construct_therm_device_gpu_sci(g, &board_obj_ptr,
+                        sizeof(struct therm_device_gpu_sci), pargs);
+                break;
+        case NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_GPU_GPC_TSOSC:
+                status = construct_therm_device_gpu_tsosc(g, &board_obj_ptr,
+                        sizeof(struct therm_device_gpu_gpc_tsosc), pargs);
+                break;
+        case NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_HBM2_SITE:
+                status = construct_therm_device_hbm2_site(g, &board_obj_ptr,
+                        sizeof(struct therm_device_hbm2_site), pargs);
+                break;
+        case NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_HBM2_COMBINED:
+                status = construct_therm_device_hbm2_combined(g, &board_obj_ptr,
+                        sizeof(struct therm_device_hbm2_combined), pargs);
+                break;
+        default:
+                nvgpu_err(g,
+                        "unsupported therm_device class - 0x%x",
+                        BOARDOBJ_GET_TYPE(pargs));
+                break;
+        }
+
+        if(status) {
+                board_obj_ptr = NULL;
+                nvgpu_err(g,
+                        "could not allocate memory for therm_device");
+                if (board_obj_ptr != NULL)
+                        nvgpu_kfree(g, board_obj_ptr);
+        }
+
+
+        return board_obj_ptr;
+}
+
 static u32 devinit_get_therm_device_table(struct gk20a *g,
                                 struct therm_devices *pthermdeviceobjs)
 {
@@ -81,10 +227,14 @@ static u32 devinit_get_therm_device_table(struct gk20a *g,
         struct therm_device_1x_entry *therm_device_table_entry = NULL;
         u32 index;
         u32 obj_index = 0;
-        u16 therm_device_size = 0;
+        u8 class_id = 0;
         union {
                 struct boardobj boardobj;
                 struct therm_device therm_device;
+                struct therm_device_gpu_sci gpu_sci;
+                struct therm_device_gpu_gpc_tsosc gpu_gpc_tsosc;
+                struct therm_device_hbm2_site hbm2_site;
+                struct therm_device_hbm2_combined hbm2_combined;
         } therm_device_data;
 
         nvgpu_log_info(g, " ");
@@ -120,17 +270,34 @@ static u32 devinit_get_therm_device_table(struct gk20a *g,
                         (curr_therm_device_table_ptr +
                                 (therm_device_table_header.table_entry_size * index));
 
-                if (therm_device_table_entry->class_id !=
-                                NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_GPU) {
+                class_id = therm_device_table_entry->class_id;
+
+                switch (class_id) {
+                case NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_INVALID:
                         continue;
+                        break;
+                case NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_GPU:
+                case NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_GPU_GPC_SCI:
+                        break;
+                case NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_GPU_GPC_TSOSC:
+                        therm_device_data.gpu_gpc_tsosc.gpc_tsosc_idx =
+                                therm_device_table_entry->param0;
+                        break;
+                case NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_HBM2_SITE:
+                        therm_device_data.hbm2_site.site_idx =
+                                therm_device_table_entry->param0;
+                        break;
+                case NV_VBIOS_THERM_DEVICE_1X_ENTRY_CLASS_HBM2_COMBINED:
+                        break;
+                default:
+                        nvgpu_err(g,
+                                "Unknown thermal device class i - %x, class - %x",
+                                index, class_id);
+                        goto done;
                 }
 
-                therm_device_size = sizeof(struct therm_device);
-                therm_device_data.boardobj.type = CTRL_THERMAL_THERM_DEVICE_CLASS_GPU;
-
-                boardobj = construct_therm_device(g, &therm_device_data,
-                                        therm_device_size, therm_device_data.boardobj.type);
-
+                therm_device_data.boardobj.type = class_id;
+                boardobj = therm_device_construct(g, &therm_device_data);
                 if (!boardobj) {
                         nvgpu_err(g,
                                 "unable to create thermal device for %d type %d",
diff --git a/drivers/gpu/nvgpu/therm/thrmdev.h b/drivers/gpu/nvgpu/therm/thrmdev.h
index 83ff5cbf..af349c88 100644
--- a/drivers/gpu/nvgpu/therm/thrmdev.h
+++ b/drivers/gpu/nvgpu/therm/thrmdev.h
@@ -32,7 +32,25 @@ struct therm_devices {
 };
 
 struct therm_device {
-        struct therm_devices super;
+        struct boardobj super;
+};
+
+struct therm_device_gpu_sci {
+        struct therm_device super;
+};
+
+struct therm_device_gpu_gpc_tsosc {
+        struct therm_device super;
+        u8 gpc_tsosc_idx;
+};
+
+struct therm_device_hbm2_site {
+        struct therm_device super;
+        u8 site_idx;
+};
+
+struct therm_device_hbm2_combined {
+        struct therm_device super;
 };
 
 u32 therm_device_sw_setup(struct gk20a *g);