gpu: nvgpu: PMU init message read from FBQ support

-Added NVGPU_SUPPORT_PMU_RTOS_FBQ feature to enable
 FBQ support.
-Add support to read PMU RTOS init message from
 FBQ message queue to process init message &
 construct FBQ for further communication
 with PMU RTOS ucode.
-Added functions to init FB command/message queues
 as per init message inputs from PMU RTOS ucode.

JIRA NVGPU-1578

Change-Id: If2678d20f7195e6e8cba354b7dca5117003e3c29
Signed-off-by: Mahantesh Kumbar <mkumbar@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1964068
Reviewed-by: svc-mobile-coverity <svc-mobile-coverity@nvidia.com>
Reviewed-by: svc-mobile-misra <svc-mobile-misra@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Vijayakumar Subbu <vsubbu@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

drivers/gpu/nvgpu/common/pmu/pmu.c

@@ -365,6 +365,101 @@ exit:
 	return err;
 }
 
+static void pmu_read_init_msg_fb(struct gk20a *g, struct nvgpu_pmu *pmu,
+	u32 element_index, u32 size, void *buffer)
+{
+	u32 fbq_msg_queue_ss_offset = 0U;
+
+	fbq_msg_queue_ss_offset = (u32)offsetof(
+		struct nv_pmu_super_surface,
+		fbq.msg_queue.element[element_index]);
+
+	nvgpu_mem_rd_n(g, &pmu->super_surface_buf, fbq_msg_queue_ss_offset,
+		buffer, size);
+}
+
+static int pmu_process_init_msg_fb(struct gk20a *g, struct nvgpu_pmu *pmu,
+	struct pmu_msg *msg)
+{
+	u32 tail = 0U;
+	int err = 0;
+
+	nvgpu_log_fn(g, " ");
+
+	g->ops.pmu.pmu_msgq_tail(pmu, &tail, QUEUE_GET);
+
+	pmu_read_init_msg_fb(g, pmu, tail, PMU_MSG_HDR_SIZE,
+		(void *)&msg->hdr);
+
+	if (msg->hdr.unit_id != PMU_UNIT_INIT) {
+		nvgpu_err(g, "FB MSG Q: expecting init msg");
+		err = -EINVAL;
+		goto exit;
+	}
+
+	pmu_read_init_msg_fb(g, pmu, tail, msg->hdr.size,
+		(void *)&msg->hdr);
+
+	if (msg->msg.init.msg_type != PMU_INIT_MSG_TYPE_PMU_INIT) {
+		nvgpu_err(g, "FB MSG Q: expecting pmu init msg");
+		err = -EINVAL;
+		goto exit;
+	}
+
+	/* Queue is not yet constructed, so inline next element code here.*/
+	tail++;
+	if (tail >= NV_PMU_FBQ_MSG_NUM_ELEMENTS) {
+		tail = 0U;
+	}
+
+	g->ops.pmu.pmu_msgq_tail(pmu, &tail, QUEUE_SET);
+
+exit:
+	return err;
+}
+
+static int pmu_process_init_msg_dmem(struct gk20a *g, struct nvgpu_pmu *pmu,
+	struct pmu_msg *msg)
+{
+	u32 tail = 0U;
+	int err = 0;
+
+	nvgpu_log_fn(g, " ");
+
+	g->ops.pmu.pmu_msgq_tail(pmu, &tail, QUEUE_GET);
+
+	err = nvgpu_falcon_copy_from_dmem(pmu->flcn, tail,
+		(u8 *)&msg->hdr, PMU_MSG_HDR_SIZE, 0);
+	if (err != 0) {
+		nvgpu_err(g, "PMU falcon DMEM copy failed");
+		goto exit;
+	}
+	if (msg->hdr.unit_id != PMU_UNIT_INIT) {
+		nvgpu_err(g, "expecting init msg");
+		err = -EINVAL;
+		goto exit;
+	}
+
+	err = nvgpu_falcon_copy_from_dmem(pmu->flcn, tail + PMU_MSG_HDR_SIZE,
+		(u8 *)&msg->msg, (u32)msg->hdr.size - PMU_MSG_HDR_SIZE, 0);
+	if (err != 0) {
+		nvgpu_err(g, "PMU falcon DMEM copy failed");
+		goto exit;
+	}
+
+	if (msg->msg.init.msg_type != PMU_INIT_MSG_TYPE_PMU_INIT) {
+		nvgpu_err(g, "expecting pmu init msg");
+		err = -EINVAL;
+		goto exit;
+	}
+
+	tail += ALIGN(msg->hdr.size, PMU_DMEM_ALIGNMENT);
+	g->ops.pmu.pmu_msgq_tail(pmu, &tail, QUEUE_SET);
+
+exit:
+	return err;
+}
+
 int nvgpu_pmu_process_init_msg(struct nvgpu_pmu *pmu,
 			struct pmu_msg *msg)
 {
@@ -372,33 +467,25 @@ int nvgpu_pmu_process_init_msg(struct nvgpu_pmu *pmu,
 	struct pmu_v *pv = &g->ops.pmu_ver;
 	union pmu_init_msg_pmu *init;
 	struct pmu_sha1_gid_data gid_data;
-	u32 i, j, tail = 0;
-	int err;
+	int err = 0;
+	u32 i = 0U;
+	u32 j = 0U;
 
 	nvgpu_log_fn(g, " ");
 
 	nvgpu_pmu_dbg(g, "init received\n");
 
-	g->ops.pmu.pmu_msgq_tail(pmu, &tail, QUEUE_GET);
-
-	nvgpu_falcon_copy_from_dmem(pmu->flcn, tail,
-		(u8 *)&msg->hdr, PMU_MSG_HDR_SIZE, 0);
-	if (msg->hdr.unit_id != PMU_UNIT_INIT) {
-		nvgpu_err(g, "expecting init msg");
-		return -EINVAL;
+	if (nvgpu_is_enabled(g, NVGPU_SUPPORT_PMU_RTOS_FBQ)) {
+		err = pmu_process_init_msg_fb(g, pmu, msg);
+	} else {
+		err = pmu_process_init_msg_dmem(g, pmu, msg);
 	}
 
-	nvgpu_falcon_copy_from_dmem(pmu->flcn, tail + PMU_MSG_HDR_SIZE,
-		(u8 *)&msg->msg, (u32)msg->hdr.size - PMU_MSG_HDR_SIZE, 0);
-
-	if (msg->msg.init.msg_type != PMU_INIT_MSG_TYPE_PMU_INIT) {
-		nvgpu_err(g, "expecting init msg");
-		return -EINVAL;
+	/* error check for above init message process*/
+	if (err != 0) {
+		goto exit;
 	}
 
-	tail += ALIGN(msg->hdr.size, PMU_DMEM_ALIGNMENT);
-	g->ops.pmu.pmu_msgq_tail(pmu, &tail, QUEUE_SET);
-
 	init = pv->get_pmu_msg_pmu_init_msg_ptr(&(msg->msg.init));
 	if (!pmu->gid_info.valid) {
 		u32 *gid_hdr_data = (u32 *)(gid_data.signature);
@@ -422,14 +509,30 @@ int nvgpu_pmu_process_init_msg(struct nvgpu_pmu *pmu,
 		}
 	}
 
-	for (i = 0; i < PMU_QUEUE_COUNT; i++) {
-		err = nvgpu_pmu_queue_init(pmu, i, init);
-		if (err != 0) {
-			for (j = 0; j < i; j++) {
-				nvgpu_pmu_queue_free(pmu, j);
+
+	if (nvgpu_is_enabled(g, NVGPU_SUPPORT_PMU_RTOS_FBQ)) {
+		pmu->queue_type = QUEUE_TYPE_FB;
+		for (i = 0; i < PMU_QUEUE_COUNT; i++) {
+			err = nvgpu_pmu_queue_init_fb(pmu, i, init);
+			if (err != 0) {
+				for (j = 0; j < i; j++) {
+					nvgpu_pmu_queue_free(pmu, j);
+				}
+				nvgpu_err(g, "PMU queue init failed");
+				return err;
+			}
+		}
+	} else {
+		pmu->queue_type = QUEUE_TYPE_DMEM;
+		for (i = 0; i < PMU_QUEUE_COUNT; i++) {
+			err = nvgpu_pmu_queue_init(pmu, i, init);
+			if (err != 0) {
+				for (j = 0; j < i; j++) {
+					nvgpu_pmu_queue_free(pmu, j);
+				}
+				nvgpu_err(g, "PMU queue init failed");
+				return err;
 			}
-			nvgpu_err(g, "PMU queue init failed");
-			return err;
 		}
 	}
 
@@ -450,9 +553,9 @@ int nvgpu_pmu_process_init_msg(struct nvgpu_pmu *pmu,
 
 	nvgpu_pmu_state_change(g, PMU_STATE_INIT_RECEIVED, true);
 
-	nvgpu_pmu_dbg(g, "init received end\n");
-
-	return 0;
+exit:
+	nvgpu_pmu_dbg(g, "init received end, err %x", err);
+	return err;
 }
 
 static void pmu_setup_hw_enable_elpg(struct gk20a *g)

drivers/gpu/nvgpu/common/pmu/pmu_ipc.c

@@ -26,6 +26,7 @@
 #include <nvgpu/timers.h>
 #include <nvgpu/bug.h>
 #include <nvgpu/pmuif/nvgpu_gpmu_cmdif.h>
+#include <nvgpu/pmuif/gpmu_super_surf_if.h>
 #include <nvgpu/falcon.h>
 #include <nvgpu/gk20a.h>
 #include <nvgpu/string.h>
@@ -103,7 +104,89 @@ int nvgpu_pmu_mutex_release(struct nvgpu_pmu *pmu, u32 id, u32 *token)
 	return g->ops.pmu.pmu_mutex_release(pmu, id, token);
 }
 
-/* PMU falcon queue init */
+/* FB queue init */
+int nvgpu_pmu_queue_init_fb(struct nvgpu_pmu *pmu,
+		u32 id, union pmu_init_msg_pmu *init)
+{
+	struct gk20a *g = gk20a_from_pmu(pmu);
+	struct nvgpu_falcon_queue_params params = {0};
+	u32 oflag = 0;
+	int err = 0;
+	u32 tmp_id = id;
+
+	/* init queue parameters */
+	if (PMU_IS_COMMAND_QUEUE(id)) {
+
+		/* currently PMU FBQ support SW command queue only */
+		if (!PMU_IS_SW_COMMAND_QUEUE(id)) {
+			pmu->queue[id] = NULL;
+			err = 0;
+			goto exit;
+		}
+
+		/*
+		 * set OFLAG_WRITE for command queue
+		 * i.e, push from nvgpu &
+		 * pop form falcon ucode
+		 */
+		oflag = OFLAG_WRITE;
+
+		params.super_surface_mem =
+			&pmu->super_surface_buf;
+		params.fbq_offset = (u32)offsetof(
+			struct nv_pmu_super_surface,
+			fbq.cmd_queues.queue[id]);
+		params.size = NV_PMU_FBQ_CMD_NUM_ELEMENTS;
+		params.fbq_element_size = NV_PMU_FBQ_CMD_ELEMENT_SIZE;
+	} else if (PMU_IS_MESSAGE_QUEUE(id)) {
+		/*
+		 * set OFLAG_READ for message queue
+		 * i.e, push from falcon ucode &
+		 * pop form nvgpu
+		 */
+		oflag = OFLAG_READ;
+
+		params.super_surface_mem =
+				&pmu->super_surface_buf;
+		params.fbq_offset = (u32)offsetof(
+				struct nv_pmu_super_surface,
+				fbq.msg_queue);
+		params.size = NV_PMU_FBQ_MSG_NUM_ELEMENTS;
+		params.fbq_element_size = NV_PMU_FBQ_MSG_ELEMENT_SIZE;
+	} else {
+		nvgpu_err(g, "invalid queue-id %d", id);
+		err = -EINVAL;
+		goto exit;
+	}
+
+	params.id = id;
+	params.oflag = oflag;
+	params.queue_type = QUEUE_TYPE_FB;
+
+	if (tmp_id == PMU_COMMAND_QUEUE_HPQ) {
+		tmp_id = PMU_QUEUE_HPQ_IDX_FOR_V3;
+	} else if (tmp_id == PMU_COMMAND_QUEUE_LPQ) {
+		tmp_id = PMU_QUEUE_LPQ_IDX_FOR_V3;
+	} else if (tmp_id == PMU_MESSAGE_QUEUE) {
+		tmp_id = PMU_QUEUE_MSG_IDX_FOR_V5;
+	} else {
+		/* return if queue id not supported*/
+		goto exit;
+	}
+	params.index = init->v5.queue_index[tmp_id];
+
+	params.offset = init->v5.queue_offset;
+
+	err = nvgpu_falcon_queue_init(pmu->flcn, &pmu->queue[id], params);
+	if (err != 0) {
+		nvgpu_err(g, "queue-%d init failed", id);
+	}
+
+exit:
+	return err;
+}
+
+/* DMEM queue init */
 int nvgpu_pmu_queue_init(struct nvgpu_pmu *pmu,
 		u32 id, union pmu_init_msg_pmu *init)
 {

drivers/gpu/nvgpu/include/nvgpu/enabled.h

@@ -174,10 +174,13 @@ struct gk20a;
 /* SEC2 RTOS support*/
 #define NVGPU_SUPPORT_SEC2_RTOS			69
 
+/* PMU RTOS FBQ support*/
+#define NVGPU_SUPPORT_PMU_RTOS_FBQ		70
+
 /*
  * Must be greater than the largest bit offset in the above list.
  */
-#define NVGPU_MAX_ENABLED_BITS			70U
+#define NVGPU_MAX_ENABLED_BITS			71U
 
 /**
  * nvgpu_is_enabled - Check if the passed flag is enabled.

drivers/gpu/nvgpu/include/nvgpu/pmu.h

@@ -325,6 +325,7 @@ struct nvgpu_pmu {
 	struct pmu_sha1_gid gid_info;
 
 	struct nvgpu_falcon_queue *queue[PMU_QUEUE_COUNT];
+	u32 queue_type;
 
 	struct pmu_sequence *seq;
 	unsigned long pmu_seq_tbl[PMU_SEQ_TBL_SIZE];
@@ -431,6 +432,9 @@ int nvgpu_pmu_queue_init(struct nvgpu_pmu *pmu, u32 id,
 	union pmu_init_msg_pmu *init);
 void nvgpu_pmu_queue_free(struct nvgpu_pmu *pmu, u32 id);
 
+int nvgpu_pmu_queue_init_fb(struct nvgpu_pmu *pmu,
+	u32 id, union pmu_init_msg_pmu *init);
+
 /* send a cmd to pmu */
 int nvgpu_pmu_cmd_post(struct gk20a *g, struct pmu_cmd *cmd,
 		struct pmu_msg *msg, struct pmu_payload *payload,

drivers/gpu/nvgpu/include/nvgpu/pmuif/gpmuif_pmu.h

@@ -88,7 +88,7 @@ struct pmu_sha1_gid_data {
 
 /* PMU INIT MSG */
 enum {
-	PMU_INIT_MSG_TYPE_PMU_INIT = 0,
+	PMU_INIT_MSG_TYPE_PMU_INIT = 0U,
 };
 
 struct pmu_init_msg_pmu_v1 {

drivers/gpu/nvgpu/tu104/hal_tu104.c

@@ -1213,6 +1213,7 @@ int tu104_init_hal(struct gk20a *g)
 	nvgpu_set_enabled(g, NVGPU_SUPPORT_MULTIPLE_WPR, true);
 	nvgpu_set_enabled(g, NVGPU_FECS_TRACE_VA, true);
 	nvgpu_set_enabled(g, NVGPU_SUPPORT_SEC2_RTOS, true);
+	nvgpu_set_enabled(g, NVGPU_SUPPORT_PMU_RTOS_FBQ, false);
 
 	/* for now */
 	gops->clk.support_clk_freq_controller = false;