drivers: pva: Add support for next chip

- PVA initialization and programming sequence update for next chip.
- Update to VMEM regions and addresses.

Signed-off-by: Amruta Bhamidipati <abhamidipati@nvidia.com>
Change-Id: I25b0fae260c516d5a7521aabc994a87525555577
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvidia/+/2925454
Reviewed-by: svc_kernel_abi <svc_kernel_abi@nvidia.com>
Reviewed-by: Sreehari Mohan <sreeharim@nvidia.com>
Reviewed-by: Krish Agarwal <krisha@nvidia.com>
Reviewed-by: Omar Nemri <onemri@nvidia.com>
GVS: Gerrit_Virtual_Submit <buildbot_gerritrpt@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nv-oot/+/2999149
Tested-by: Omar Nemri <onemri@nvidia.com>

drivers/video/tegra/host/pva/hw_vmem_pva.h

@@ -6,8 +6,10 @@
 #ifndef _hw_vmem_pva_h_
 #define _hw_vmem_pva_h_
 
-#define NUM_HEM_GEN		2U
-#define VMEM_REGION_COUNT	3U
+#define NUM_HEM_GEN		3U
+#define VMEM_REGION_COUNT_T19x	3U
+#define VMEM_REGION_COUNT_T23x	3U
+
 #define T19X_VMEM0_START	0x40U
 #define T19X_VMEM0_END		0x10000U
 #define T19X_VMEM1_START	0x40000U

drivers/video/tegra/host/pva/nvpva_client.c

@@ -49,7 +49,7 @@ client_context_search_locked(struct platform_device *pdev,
 	c_node->pva = dev;
 	c_node->curr_sema_value = 0;
 	mutex_init(&c_node->sema_val_lock);
-	if (dev->version == PVA_HW_GEN2) {
+	if (dev->version != PVA_HW_GEN1) {
 		c_node->cntxt_dev =
 			nvpva_iommu_context_dev_allocate(NULL,
 							 0,
@@ -70,7 +70,7 @@ client_context_search_locked(struct platform_device *pdev,
 		dev_err(&dev->pdev->dev,
 			"failed to init nvhost buffer for client:%lu",
 			PTR_ERR(c_node->buffers));
-		if (dev->version == PVA_HW_GEN2)
+		if (dev->version != PVA_HW_GEN1)
 			nvpva_iommu_context_dev_release(c_node->cntxt_dev);
 		c_node = NULL;
 	}

drivers/video/tegra/host/pva/pva.c

@@ -61,6 +61,10 @@
 #include "pva-fw-address-map.h"
 #include "pva_sec_ec.h"
 
+#ifdef CONFIG_TEGRA_T26X_GRHOST_PVA
+#include "pva_t264.h"
+#endif
+
 /*
  * NO IOMMU set 0x60000000 as start address.
  * With IOMMU set 0x80000000(>2GB) as startaddress
@@ -82,6 +86,10 @@ static u32 vm_regs_sid_idx_t234[] = {1, 2, 3, 4, 5, 6, 7, 7,
 #endif
 static u32 vm_regs_reg_idx_t234[] = {0, 1, 2, 3, 4, 5, 6, 7,
 				     8, 8, 8, 9, 9, 0, 0, 0};
+#ifndef CONFIG_TEGRA_T26X_GRHOST_PVA
+static u32 *vm_regs_sid_idx_t264 = vm_regs_sid_idx_t234;
+static u32 *vm_regs_reg_idx_t264 = vm_regs_reg_idx_t234;
+#endif
 static char *aux_dev_name = "16000000.pva0:pva0_niso1_ctx7";
 static u32 aux_dev_name_len = 29;
 
@@ -197,6 +205,12 @@ static struct of_device_id tegra_pva_of_match[] = {
 		.name = "pva0",
 		.compatible = "nvidia,tegra234-pva-hv",
 		.data = (struct nvhost_device_data *)&t23x_pva0_info },
+#ifdef CONFIG_TEGRA_T26X_GRHOST_PVA
+	{
+		.name = "pva0",
+		.compatible = "nvidia,tegra264-pva",
+		.data = (struct nvhost_device_data *)&t264_pva0_info },
+#endif
 	{ },
 };
 
@@ -768,9 +782,12 @@ static int nvpva_write_hwid(struct platform_device *pdev)
 	if (pva->version == PVA_HW_GEN1) {
 		id_idx = vm_regs_sid_idx_t19x;
 		reg_idx = vm_regs_reg_idx_t19x;
-	} else {
+	} else if (pva->version == PVA_HW_GEN2) {
 		id_idx = vm_regs_sid_idx_t234;
 		reg_idx = vm_regs_reg_idx_t234;
+	} else {
+		id_idx = vm_regs_sid_idx_t264;
+		reg_idx = vm_regs_reg_idx_t264;
 	}
 
 	/* Go through the StreamIDs and assemble register values */
@@ -1040,7 +1057,9 @@ static int pva_probe(struct platform_device *pdev)
 	struct pva *pva;
 	int err = 0;
 	size_t i;
+#ifndef CONFIG_TEGRA_T26X_GRHOST_PVA
 	u32 offset;
+#endif
 
 #if !IS_ENABLED(CONFIG_TEGRA_GRHOST)
 	struct kobj_attribute *attr = NULL;
@@ -1069,7 +1088,7 @@ static int pva_probe(struct platform_device *pdev)
 #endif
 
 	if ((pdata->version != PVA_HW_GEN1)
-	     && !is_cntxt_initialized()) {
+	     && !is_cntxt_initialized(pdata->version)) {
 		dev_warn(&pdev->dev,
 			 "nvpva cntxt was not initialized, deferring probe.");
 		return -EPROBE_DEFER;
@@ -1096,8 +1115,15 @@ static int pva_probe(struct platform_device *pdev)
 	}
 
 	/* Initialize PVA private data */
-	if (pdata->version == PVA_HW_GEN2) {
+	if (pdata->version == PVA_HW_GEN3) {
+		pva->version = PVA_HW_GEN3;
+		pdata->firmware_name = "nvpva_030.fw";
+		pdata->firmware_not_in_subdir = true;
+		pva->submit_cmd_mode = PVA_SUBMIT_MODE_MMIO_CCQ;
+		pva->version_config = &pva_t23x_config;
+	} else if (pdata->version == PVA_HW_GEN2) {
 		pva->version = PVA_HW_GEN2;
+		dev_info(&pdev->dev, "pdata->version is HW_GEN2");
 		pdata->firmware_name = "nvpva_020.fw";
 		pdata->firmware_not_in_subdir = true;
 		pva->submit_cmd_mode = PVA_SUBMIT_MODE_MMIO_CCQ;
@@ -1192,19 +1218,29 @@ static int pva_probe(struct platform_device *pdev)
 		goto err_client_device_init;
 	}
 
-	if (pdata->version != PVA_HW_GEN1) {
+	dev_info(dev, "Completed nvhost_client_device_init\n");
+
+	if (pdata->version == PVA_HW_GEN1) {
+		pva->aux_pdev = pva->pdev;
+	} else if (pdata->version == PVA_HW_GEN2) {
 		pva->aux_pdev =
 			nvpva_iommu_context_dev_allocate(aux_dev_name,
 							 aux_dev_name_len,
 							 false);
+	} else {
+#ifdef CONFIG_TEGRA_T26X_GRHOST_PVA
+		pva->aux_pdev =
+			nvpva_iommu_context_dev_allocate(aux_dev_name_t264,
+							 aux_dev_name_len_t264,
+							 false);
+#endif
+	}
+
 	if (pva->aux_pdev == NULL) {
 		dev_err(&pva->pdev->dev,
 			"failed to allocate aux device");
 		goto err_context_alloc;
 	}
-	} else {
-		pva->aux_pdev = pva->pdev;
-	}
 
 	pva->pool = nvpva_queue_init(pdev,  pva->aux_pdev, &pva_queue_ops,
 				     MAX_PVA_QUEUE_COUNT);
@@ -1256,7 +1292,7 @@ static int pva_probe(struct platform_device *pdev)
 	pva->sid_count = 0;
 	err = nvpva_iommu_context_dev_get_sids(&pva->sids[1],
 					 &pva->sid_count,
-					 NVPVA_USER_VM_COUNT);
+					 pdata->version);
 	if (err)
 		goto err_iommu_ctxt_init;
 
@@ -1268,6 +1304,7 @@ static int pva_probe(struct platform_device *pdev)
 
 	++(pva->sid_count);
 
+#ifndef CONFIG_TEGRA_T26X_GRHOST_PVA
 	offset = hwpm_get_offset();
 
 	if ((UINT_MAX - offset) < pdev->resource[0].start) {
@@ -1282,6 +1319,7 @@ static int pva_probe(struct platform_device *pdev)
 	pva->hwpm_ip_ops.hwpm_ip_pm = &pva_hwpm_ip_pm;
 	pva->hwpm_ip_ops.hwpm_ip_reg_op = &pva_hwpm_ip_reg_op;
 	tegra_soc_hwpm_ip_register(&pva->hwpm_ip_ops);
+#endif
 
 #if !IS_ENABLED(CONFIG_TEGRA_GRHOST)
 	if (pdata->num_clks > 0) {

drivers/video/tegra/host/pva/pva.h

@@ -54,8 +54,6 @@ struct pva_version_info {
 #define MAX_PVA_TASK_COUNT_PER_QUEUE_SEG	\
 	(MAX_PVA_TASK_COUNT_PER_QUEUE/MAX_PVA_SEG_COUNT_PER_QUEUE)
 
-#define NVPVA_USER_VM_COUNT	MAX_PVA_CLIENTS
-
 /**
  * Maximum task count that a PVA engine can support
  */
@@ -254,6 +252,7 @@ struct pva_status_interface_registers {
 
 #define PVA_HW_GEN1 1
 #define PVA_HW_GEN2 2
+#define PVA_HW_GEN3 3
 
 /**
  * @brief		HW version specific configuration and functions

drivers/video/tegra/host/pva/pva_iommu_context_dev.c

@@ -16,10 +16,23 @@
 #include <linux/of.h>
 #include <linux/version.h>
 #include <linux/dma-buf.h>
+#include <linux/nvhost.h>
+#include <linux/platform_device.h>
 
 #include "pva_iommu_context_dev.h"
 #include "pva.h"
 
+#define NVPVA_CNTXT_DEV_NAME_LEN_T23X	(29U)
+#define NVPVA_CNTXT_DEVICE_CNT		(8U)
+
+#ifdef CONFIG_TEGRA_T26X_GRHOST_PVA
+#include "pva_cntxt_dev_name_t264.h"
+#include "pva_iommu_context_dev_t264.h"
+#else
+#define NVPVA_CNTXT_DEV_NAME_LEN NVPVA_CNTXT_DEV_NAME_LEN_T23X
+#define NVPVA_CNTXT_DEVICE_CNT_T264 NVPVA_CNTXT_DEVICE_CNT
+#endif
+
 static u32 cntxt_dev_count;
 static char *dev_names[] = {
 	"pva0_niso1_ctx0",
@@ -30,6 +43,9 @@ static char *dev_names[] = {
 	"pva0_niso1_ctx5",
 	"pva0_niso1_ctx6",
 	"pva0_niso1_ctx7",
+#ifdef CONFIG_TEGRA_T26X_GRHOST_PVA
+	PVA_CNTXT_DEV_NAME_T264
+#endif
 };
 
 static const struct of_device_id pva_iommu_context_dev_of_match[] = {
@@ -49,22 +65,31 @@ struct pva_iommu_ctx {
 static LIST_HEAD(pva_iommu_ctx_list);
 static DEFINE_MUTEX(pva_iommu_ctx_list_mutex);
 
-bool is_cntxt_initialized(void)
+bool is_cntxt_initialized(const int hw_gen)
 {
-	return (cntxt_dev_count == 8);
+	u32 pva_cntxt_dev_cnt = (hw_gen == PVA_HW_GEN3) ? NVPVA_CNTXT_DEVICE_CNT_T264
+					 : NVPVA_CNTXT_DEVICE_CNT;
+	return (cntxt_dev_count == pva_cntxt_dev_cnt);
 }
 
-int nvpva_iommu_context_dev_get_sids(int *hwids, int *count, int max_cnt)
+int nvpva_iommu_context_dev_get_sids(int *hwids, int *count, const int hw_gen)
 {
 	struct pva_iommu_ctx *ctx;
 	int err = 0;
 	int i;
+	u32 pva_cntxt_dev_cnt;
+
+	if (hw_gen == PVA_HW_GEN3)
+		pva_cntxt_dev_cnt = NVPVA_CNTXT_DEVICE_CNT_T264;
+	else
+		pva_cntxt_dev_cnt = NVPVA_CNTXT_DEVICE_CNT;
 
 	*count = 0;
 	mutex_lock(&pva_iommu_ctx_list_mutex);
-	for (i = 0; i < max_cnt; i++) {
+	for (i = 0; i < pva_cntxt_dev_cnt; i++) {
 		list_for_each_entry(ctx, &pva_iommu_ctx_list, list) {
-			if (strnstr(ctx->pdev->name, dev_names[i], 29) != NULL) {
+			if (strnstr(ctx->pdev->name, dev_names[i],
+			 NVPVA_CNTXT_DEV_NAME_LEN) != NULL) {
 				hwids[*count] = nvpva_get_device_hwid(ctx->pdev, 0);
 				if (hwids[*count] < 0) {
 					err = hwids[*count];
@@ -72,7 +97,7 @@ int nvpva_iommu_context_dev_get_sids(int *hwids, int *count, int max_cnt)
 				}
 
 				++(*count);
-				if (*count >= max_cnt)
+				if (*count >= pva_cntxt_dev_cnt)
 					break;
 			}
 		}
@@ -167,7 +192,7 @@ static int pva_iommu_context_dev_probe(struct platform_device *pdev)
 		return -ENOMEM;
 	}
 
-	if (strnstr(pdev->name, dev_names[7], 29) != NULL)
+	if (strnstr(pdev->name, dev_names[7], NVPVA_CNTXT_DEV_NAME_LEN) != NULL)
 		dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
 	else
 		dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(39));

drivers/video/tegra/host/pva/pva_iommu_context_dev.h

@@ -11,7 +11,7 @@
 struct platform_device
 *nvpva_iommu_context_dev_allocate(char *identifier, size_t len, bool shared);
 void nvpva_iommu_context_dev_release(struct platform_device *pdev);
-int nvpva_iommu_context_dev_get_sids(int *hwids, int *count, int max_cnt);
-bool is_cntxt_initialized(void);
+int nvpva_iommu_context_dev_get_sids(int *hwids, int *count, const int hw_gen);
+bool is_cntxt_initialized(const int hw_gen);
 
 #endif

drivers/video/tegra/host/pva/pva_queue.c

@@ -198,10 +198,14 @@ pva_task_pin_fence(struct pva_submit_task *task,
 		if (IS_ERR(mem)) {
 			task_err(task, "sempahore submit pin failed");
 			err = PTR_ERR(mem);
-		} else
+		} else {
 			*addr = mem->dma_addr + fence->obj.sem.mem.offset;
 
 			*serial_id = mem->serial_id;
+			nvpva_dbg_info(task->pva,
+			       "id = %d, semaphore addr = %llx",
+			       fence->obj.sem.mem.pin_id, *addr);
+		}
 		break;
 	}
 	case NVPVA_FENCE_OBJ_SYNCPT: {
@@ -1257,7 +1261,7 @@ set_task_parameters(const struct pva_submit_tasks *task_header)
 	 * thus the response should come in the correct CCQ
 	 */
 	if ((task->pva->submit_task_mode == PVA_SUBMIT_MODE_MMIO_CCQ)
-	   && (task_header->tasks[0]->pva->version == PVA_HW_GEN2))
+	   && (task_header->tasks[0]->pva->version != PVA_HW_GEN1))
 		status_interface = (task->queue->id + 1U);
 
 	for (idx = 0U; idx < task_header->num_tasks; idx++) {

drivers/video/tegra/host/pva/pva_vpu_exe.c

@@ -12,6 +12,12 @@
 #include "hw_vmem_pva.h"
 #include "pva_vpu_exe.h"
 
+#ifdef CONFIG_TEGRA_T26X_GRHOST_PVA
+#include "hw_vmem_pva_t264.h"
+#else
+#define VMEM_REGION_COUNT_T26x 4
+#endif
+
 #define ELF_MAXIMUM_SECTION_NAME 64
 #define ELF_EXPORTS_SECTION "EXPORTS"
 #define DATA_SECTION_ALIGNMENT 32
@@ -1123,16 +1129,22 @@ struct vmem_region {
 	uint32_t end;
 };
 
-struct vmem_region vmem_regions_tab[NUM_HEM_GEN + 1][VMEM_REGION_COUNT] = {
+struct vmem_region vmem_regions_tab[NUM_HEM_GEN + 1][VMEM_REGION_COUNT_T26x] = {
 	{{.start = 0, .end = 0},
+	 {.start = 0, .end = 0},
 	 {.start = 0, .end = 0},
 	 {.start = 0, .end = 0}},
 	{{.start = T19X_VMEM0_START, .end = T19X_VMEM0_END},
 	 {.start = T19X_VMEM1_START, .end = T19X_VMEM1_END},
-	 {.start = T19X_VMEM2_START, .end = T19X_VMEM2_END}},
+	 {.start = T19X_VMEM2_START, .end = T19X_VMEM2_END},
+	 {.start = 0xFFFFFFFF, .end = 0}},
 	{{.start = T23x_VMEM0_START, .end = T23x_VMEM0_END},
 	 {.start = T23x_VMEM1_START, .end = T23x_VMEM1_END},
-	 {.start = T23x_VMEM2_START, .end = T23x_VMEM2_END}},
+	 {.start = T23x_VMEM2_START, .end = T23x_VMEM2_END},
+	 {.start = 0xFFFFFFFF, .end = 0}},
+#ifdef CONFIG_TEGRA_T26X_GRHOST_PVA
+	#include "pva_vmem_regions_tab_t264.h"
+#endif
 };
 
 int32_t
@@ -1143,13 +1155,15 @@ nvpva_validate_vmem_offset(const uint32_t vmem_offset,
 
 	int i;
 	int32_t err = -EINVAL;
+	const u8 vmem_region_count = (hw_gen == PVA_HW_GEN3) ? VMEM_REGION_COUNT_T26x
+								 : VMEM_REGION_COUNT_T23x;
 
 	if (hw_gen < 0 || hw_gen > NUM_HEM_GEN) {
 		pr_err("invalid hw_gen index: %d", hw_gen);
 		return err;
 	}
 
-	for (i = VMEM_REGION_COUNT; i > 0; i--) {
+	for (i = vmem_region_count; i > 0; i--) {
 		if (vmem_offset >= vmem_regions_tab[hw_gen][i-1].start)
 			break;
 	}