drm/tegra: nvhost support for t264

- VIC RISC-V EB boot support - Programming sequence modification needed for Thor - Reloc block linear addressing not needed for t264 Bug 4132685 Signed-off-by: Santosh BS <santoshb@nvidia.com> Change-Id: I8ad47cce31cfd06020e33d3457a0d674a11e4d49
2025-12-22 17:25:35 +03:00 · 2023-08-08 15:17:56 +00:00
parent 340bd4418f
commit 26b1cb4a82
8 changed files with 416 additions and 59 deletions
--- a/drivers/gpu/drm/tegra/drm.c
+++ b/drivers/gpu/drm/tegra/drm.c
@@ -1457,6 +1457,7 @@ static const struct of_device_id host1x_drm_subdevs[] = {
 	{ .compatible = "nvidia,tegra234-nvjpg", },
 	{ .compatible = "nvidia,tegra234-ofa", },
 	{ .compatible = "nvidia,tegra234-host1x-virtual-engine", },
+	{ .compatible = "nvidia,tegra264-vic", },
 	{ /* sentinel */ }
 };

--- a/drivers/gpu/drm/tegra/drm.h
+++ b/drivers/gpu/drm/tegra/drm.h
@@ -95,6 +95,7 @@ struct tegra_drm_client_ops {
 	int (*get_streamid_offset)(struct tegra_drm_client *client, u32 *offset);
 	int (*can_use_memory_ctx)(struct tegra_drm_client *client, bool *supported);
 	int (*has_job_timestamping)(struct tegra_drm_client *client, bool *supported);
+	int (*skip_bl_swizzling)(struct tegra_drm_client *client, bool *skip);
 };

 int tegra_drm_submit(struct tegra_drm_context *context,
--- a/drivers/gpu/drm/tegra/riscv.c
+++ b/drivers/gpu/drm/tegra/riscv.c
@@ -12,6 +12,7 @@
 #include <linux/of.h>

 #include "riscv.h"
+#include "falcon.h"

 #define RISCV_CPUCTL					0x4388
 #define RISCV_CPUCTL_STARTCPU_TRUE			(1 << 0)
@@ -31,6 +32,13 @@
 #define RISCV_BCR_DMAADDR_FMCDATA_HI			0x4684
 #define RISCV_BCR_DMACFG_SEC				0x4694
 #define RISCV_BCR_DMACFG_SEC_GSCID(v)			((v) << 16)
+#define RISCV_BOOT_VECTOR_LO				0x1780
+#define RISCV_BOOT_VECTOR_HI				0x1784
+
+enum riscv_memory {
+	RISCV_MEMORY_IMEM,
+	RISCV_MEMORY_DATA,
+};

 static void riscv_writel(struct tegra_drm_riscv *riscv, u32 value, u32 offset)
 {
@@ -100,3 +108,141 @@ int tegra_drm_riscv_boot_bootrom(struct tegra_drm_riscv *riscv, phys_addr_t imag

 	return 0;
 }
+
+int tegra_drm_riscv_read_firmware(struct tegra_drm_riscv *riscv, const char *name)
+{
+	int err;
+
+	/* request_firmware prints error if it fails */
+	err = request_firmware(&riscv->firmware.firmware, name, riscv->dev);
+	if (err < 0)
+		return err;
+
+	riscv->firmware.size = riscv->firmware.firmware->size;
+
+	return 0;
+}
+
+int tegra_drm_riscv_load_firmware(struct tegra_drm_riscv *riscv)
+{
+	const struct firmware *firmware = riscv->firmware.firmware;
+	u32 *virt = riscv->firmware.virt;
+	size_t i;
+
+	/* Copy firmware image into local area taking into account endianness */
+	for (i = 0; i < firmware->size / sizeof(u32); i++)
+		virt[i] = le32_to_cpu(((__le32 *)firmware->data)[i]);
+
+	release_firmware(firmware);
+	riscv->firmware.firmware = NULL;
+
+	return 0;
+}
+
+int tegra_drm_riscv_init(struct tegra_drm_riscv *riscv)
+{
+	riscv->firmware.virt = NULL;
+
+	return 0;
+}
+
+void tegra_drm_riscv_exit(struct tegra_drm_riscv *riscv)
+{
+	if (riscv->firmware.firmware)
+		release_firmware(riscv->firmware.firmware);
+}
+
+static int tegra_drm_riscv_dma_wait_idle(struct tegra_drm_riscv *riscv)
+{
+	u32 value;
+
+	return readl_poll_timeout(riscv->regs + FALCON_DMATRFCMD, value,
+				  (value & FALCON_DMATRFCMD_IDLE), 10, 100000);
+}
+
+static int tegra_drm_riscv_copy_chunk(struct tegra_drm_riscv *riscv,
+			     phys_addr_t base,
+			     unsigned long offset,
+			     enum riscv_memory target)
+{
+	u32 cmd = FALCON_DMATRFCMD_SIZE_256B;
+
+	if (target == RISCV_MEMORY_IMEM)
+		cmd |= FALCON_DMATRFCMD_IMEM;
+
+	/*
+	 * Use second DMA context (i.e. the one for firmware). Strictly
+	 * speaking, at this point both DMA contexts point to the firmware
+	 * stream ID, but this register's value will be reused by the firmware
+	 * for later DMA transactions, so we need to use the correct value.
+	 */
+	cmd |= FALCON_DMATRFCMD_DMACTX(1);
+
+	riscv_writel(riscv, offset, FALCON_DMATRFMOFFS);
+	riscv_writel(riscv, base, FALCON_DMATRFFBOFFS);
+	riscv_writel(riscv, cmd, FALCON_DMATRFCMD);
+
+	return tegra_drm_riscv_dma_wait_idle(riscv);
+}
+
+int tegra_drm_riscv_boot_external(struct tegra_drm_riscv *riscv)
+{
+	unsigned long offset;
+	u32 value;
+	int err;
+
+	if (!riscv->firmware.virt)
+		return -EINVAL;
+
+	err = readl_poll_timeout(riscv->regs + FALCON_DMACTL, value,
+				 (value & (FALCON_DMACTL_IMEM_SCRUBBING |
+					   FALCON_DMACTL_DMEM_SCRUBBING)) == 0,
+				 10, 10000);
+	if (err < 0)
+		return err;
+
+	riscv_writel(riscv, 0, FALCON_DMACTL);
+
+	/* setup the address of the binary data so Falcon can access it later */
+	riscv_writel(riscv, riscv->firmware.iova >> 8, FALCON_DMATRFBASE);
+
+	/* copy the data segment into riscv internal memory */
+	for (offset = 0; offset < riscv->os_desc.data_size; offset += 256) {
+		err = tegra_drm_riscv_copy_chunk(riscv,
+				riscv->os_desc.data_offset + offset,
+				offset, RISCV_MEMORY_DATA);
+	}
+
+	/* copy the code segment into riscv internal memory */
+	for (offset = 0; offset < riscv->os_desc.code_size; offset += 256) {
+		err = tegra_drm_riscv_copy_chunk(riscv,
+				riscv->os_desc.code_offset + offset,
+				offset, RISCV_MEMORY_IMEM);
+	}
+
+	/* setup riscv interrupts */
+	riscv_writel(riscv, FALCON_IRQMSET_EXT(0xff) |
+			      FALCON_IRQMSET_SWGEN1 |
+			      FALCON_IRQMSET_SWGEN0 |
+			      FALCON_IRQMSET_EXTERR |
+			      FALCON_IRQMSET_HALT |
+			      FALCON_IRQMSET_WDTMR,
+		      FALCON_IRQMSET);
+	riscv_writel(riscv, FALCON_IRQDEST_EXT(0xff) |
+			      FALCON_IRQDEST_SWGEN1 |
+			      FALCON_IRQDEST_SWGEN0 |
+			      FALCON_IRQDEST_EXTERR |
+			      FALCON_IRQDEST_HALT,
+		      FALCON_IRQDEST);
+
+	/* enable interface */
+	riscv_writel(riscv, FALCON_ITFEN_MTHDEN |
+			      FALCON_ITFEN_CTXEN,
+		      FALCON_ITFEN);
+
+	/* boot riscv */
+	riscv_writel(riscv, 0x00000000, RISCV_BOOT_VECTOR_HI);
+	riscv_writel(riscv, 0x00100000, RISCV_BOOT_VECTOR_LO);
+
+	return 0;
+}
--- a/drivers/gpu/drm/tegra/riscv.h
+++ b/drivers/gpu/drm/tegra/riscv.h
@@ -14,11 +14,23 @@ struct tegra_drm_riscv_descriptor {
 	u32 data_size;
 };

+struct tegra_drm_riscv_firmware {
+	/* Firmware after it is read but not loaded */
+	const struct firmware *firmware;
+
+	/* Raw firmware data */
+	dma_addr_t iova;
+	dma_addr_t phys;
+	void *virt;
+	size_t size;
+};
+
 struct tegra_drm_riscv {
 	/* User initializes */
 	struct device *dev;
 	void __iomem *regs;

+	struct tegra_drm_riscv_firmware firmware;
 	struct tegra_drm_riscv_descriptor bl_desc;
 	struct tegra_drm_riscv_descriptor os_desc;
 };
@@ -26,5 +38,10 @@ struct tegra_drm_riscv {
 int tegra_drm_riscv_read_descriptors(struct tegra_drm_riscv *riscv);
 int tegra_drm_riscv_boot_bootrom(struct tegra_drm_riscv *riscv, phys_addr_t image_address,
 				 u32 gscid, const struct tegra_drm_riscv_descriptor *desc);
+int tegra_drm_riscv_init(struct tegra_drm_riscv *riscv);
+void tegra_drm_riscv_exit(struct tegra_drm_riscv *riscv);
+int tegra_drm_riscv_read_firmware(struct tegra_drm_riscv *riscv, const char *firmware_name);
+int tegra_drm_riscv_load_firmware(struct tegra_drm_riscv *riscv);
+int tegra_drm_riscv_boot_external(struct tegra_drm_riscv *riscv);

 #endif
--- a/drivers/gpu/drm/tegra/submit.c
+++ b/drivers/gpu/drm/tegra/submit.c
@@ -242,6 +242,11 @@ static int submit_write_reloc(struct tegra_drm_context *context, struct gather_b
 	/* TODO check that target_offset is within bounds */
 	dma_addr_t iova = buf->reloc.target_offset;
 	u32 written_ptr;
+	bool skip_bl_swizzling = false;
+
+	if (context->client->ops->skip_bl_swizzling &&
+	    context->client->ops->skip_bl_swizzling(context->client, &skip_bl_swizzling))
+		skip_bl_swizzling = false;

 	if (mapping->bo_map)
 		iova += mapping->iova;
@@ -249,7 +254,7 @@ static int submit_write_reloc(struct tegra_drm_context *context, struct gather_b
 		iova += mapping->ctx_map->mapping->phys;

 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
-	if (buf->flags & DRM_TEGRA_SUBMIT_RELOC_SECTOR_LAYOUT)
+	if ((buf->flags & DRM_TEGRA_SUBMIT_RELOC_SECTOR_LAYOUT) && !skip_bl_swizzling)
 		iova |= BIT_ULL(39);
 #endif

--- a/drivers/gpu/drm/tegra/vic.c
+++ b/drivers/gpu/drm/tegra/vic.c
@@ -12,6 +12,7 @@
 #include <linux/host1x-next.h>
 #include <linux/interconnect.h>
 #include <linux/iommu.h>
+#include <linux/iopoll.h>
 #include <linux/module.h>
 #include <linux/of.h>
 #include <linux/of_device.h>
@@ -30,18 +31,35 @@

 #include "drm.h"
 #include "falcon.h"
+#include "riscv.h"
 #include "util.h"
 #include "vic.h"
 #include "hwpm.h"

 #define VIC_SEC_INTF_CRC_CTRL               	0xe000

+#define VIC_RISCV_BCR_CTRL			0x1a68
+#define VIC_RISCV_BCR_CTRL_CORE_SELECT_RISCV	(1 << 4)
+#define VIC_FALCON_DEBUGINFO			0x1094
+#define VIC_DEBUGINFO_DUMMY			0xabcd1234
+#define VIC_DEBUGINFO_CLEAR			0x0
+#define VIC_RISCV_CPUCTL			0x1788
+#define VIC_RISCV_CPUCTL_STARTCPU_TRUE		(1 << 0)
+#define VIC_RISCV_CPUCTL_ACTIVE_STATE_V(x)	((x >> 7) & 0x1)
+#define VIC_RISCV_CPUCTL_ACTIVE_STATE_ACTIVE_V	0x1
+
 struct vic_config {
 	const char *firmware;
 	unsigned int version;
 	bool supports_sid;
 	bool supports_timestamping;
 	bool has_crc_enable;
+	bool has_riscv;
+	u32 transcfg_addr;
+	u32 actmon_active_mask;
+	u32 actmon_active_borps;
+	u32 actmon_active_weight;
+	bool skip_bl_swizzling;
 };

 struct vic {
@@ -63,6 +81,9 @@ struct vic {

 	/* Platform configuration */
 	const struct vic_config *config;
+
+	/* RISC-V specific data */
+	struct tegra_drm_riscv riscv;
 };

 static bool blf_write_allowed(u32 offset)
@@ -94,15 +115,12 @@ static void vic_writel(struct vic *vic, u32 value, unsigned int offset)
 	writel(value, vic->regs + offset);
 }

-static int vic_boot(struct vic *vic)
+static int vic_falcon_boot(struct vic *vic)
 {
 	u32 fce_ucode_size, fce_bin_data_offset;
 	void *hdr;
 	int err = 0;

-	if (vic->config->supports_sid)
-		tegra_drm_program_iommu_regs(vic->dev, vic->regs, VIC_TFBIF_TRANSCFG);
-
 	/* setup clockgating registers */
 	vic_writel(vic, CG_IDLE_CG_DLY_CNT(4) |
 			CG_IDLE_CG_EN |
@@ -139,6 +157,46 @@ static int vic_boot(struct vic *vic)
 	return 0;
 }

+static int vic_riscv_boot(struct vic *vic)
+{
+	int err;
+	u32 val;
+
+	vic_writel(vic, VIC_RISCV_BCR_CTRL_CORE_SELECT_RISCV, VIC_RISCV_BCR_CTRL);
+
+	/* Write a known pattern into DEBUGINFO register */
+	vic_writel(vic, VIC_DEBUGINFO_DUMMY, VIC_FALCON_DEBUGINFO);
+
+	err = tegra_drm_riscv_boot_external(&vic->riscv);
+	if (err < 0)
+		return err;
+
+	/* Kick start RISC-V */
+	vic_writel(vic, VIC_RISCV_CPUCTL_STARTCPU_TRUE, VIC_RISCV_CPUCTL);
+
+	err = readl_poll_timeout(
+		vic->regs + VIC_RISCV_CPUCTL, val,
+		VIC_RISCV_CPUCTL_ACTIVE_STATE_V(val) == VIC_RISCV_CPUCTL_ACTIVE_STATE_ACTIVE_V,
+		10, 100000);
+	if (err) {
+		dev_err(vic->dev, "cpuctl active state timeout! val=0x%x", val);
+		return err;
+	}
+
+	/* Check vic has reached a proper initialized state */
+	err  = readl_poll_timeout(
+		vic->regs + VIC_FALCON_DEBUGINFO, val,
+		(val == VIC_DEBUGINFO_CLEAR),
+		1000, 2000000);
+	if (err) {
+		dev_err(vic->dev, "not reached initialized state, timeout! val=0x%x", val);
+		return err;
+	}
+
+	return 0;
+}
+
+
 static int vic_set_rate(struct vic *vic, unsigned long rate)
 {
 	unsigned long dev_rate;
@@ -231,10 +289,14 @@ static int vic_devfreq_init(struct vic *vic)
 	struct devfreq_dev_profile *devfreq_profile;
 	struct devfreq *devfreq;

-	while (rate <= max_rate) {
+	margin = max(margin, 9000000UL);
+	do {
 		dev_pm_opp_add(vic->dev, rate, 0);
+		/* Overflow check */
+		if (margin > ULONG_MAX - rate)
+			break;
 		rate += margin;
-	}
+	} while (rate <= max_rate);

 	data = devm_kzalloc(vic->dev, sizeof(*data), GFP_KERNEL);
 	if (!data)
@@ -346,6 +408,11 @@ static int vic_exit(struct host1x_client *client)

 	vic->channel = NULL;

+	if (vic->config->has_riscv) {
+		dma_free_coherent(vic->dev, vic->riscv.firmware.size,
+				  vic->riscv.firmware.virt,
+				  vic->riscv.firmware.iova);
+	} else {
 		if (client->group) {
 			dma_unmap_single(vic->dev, vic->falcon.firmware.phys,
 					 vic->falcon.firmware.size, DMA_TO_DEVICE);
@@ -357,6 +424,7 @@ static int vic_exit(struct host1x_client *client)
 					  vic->falcon.firmware.virt,
 					  vic->falcon.firmware.iova);
 		}
+	}

 	return 0;
 }
@@ -411,7 +479,7 @@ static const struct host1x_client_ops vic_client_ops = {
 	.actmon_event = vic_actmon_event,
 };

-static int vic_load_firmware(struct vic *vic)
+static int vic_load_falcon_firmware(struct vic *vic)
 {
 	struct host1x_client *client = &vic->client.base;
 	struct tegra_drm *tegra = vic->client.drm;
@@ -506,29 +574,71 @@ cleanup:
 	return err;
 }

+static int vic_load_riscv_firmware(struct vic *vic)
+{
+	dma_addr_t iova;
+	size_t size;
+	void *virt;
+	int err;
+
+	if (vic->riscv.firmware.virt)
+		return 0;
+
+	err = tegra_drm_riscv_read_firmware(&vic->riscv, vic->config->firmware);
+	if (err < 0)
+		return err;
+
+	size = vic->riscv.firmware.size;
+
+	virt = dma_alloc_coherent(vic->dev, size, &iova, GFP_KERNEL);
+
+	err = dma_mapping_error(vic->dev, iova);
+	if (err < 0)
+		return err;
+
+	vic->riscv.firmware.virt = virt;
+	vic->riscv.firmware.iova = iova;
+
+	err = tegra_drm_riscv_load_firmware(&vic->riscv);
+	if (err < 0)
+		goto cleanup;
+
+	vic->can_use_context = true;
+
+	return 0;
+
+cleanup:
+	dma_free_coherent(vic->dev, size, virt, iova);
+
+	return err;
+}

 static void vic_actmon_reg_init(struct vic *vic)
 {
+	if (vic->config->actmon_active_mask)
 		vic_writel(vic,
 			VIC_TFBIF_ACTMON_ACTIVE_MASK_STARVED |
 			VIC_TFBIF_ACTMON_ACTIVE_MASK_STALLED |
 			VIC_TFBIF_ACTMON_ACTIVE_MASK_DELAYED,
-		   NV_PVIC_TFBIF_ACTMON_ACTIVE_MASK);
+			vic->config->actmon_active_mask);

+	if (vic->config->actmon_active_borps)
 		vic_writel(vic,
 			VIC_TFBIF_ACTMON_ACTIVE_BORPS_ACTIVE,
-		   NV_PVIC_TFBIF_ACTMON_ACTIVE_BORPS);
+			vic->config->actmon_active_borps);
 }

 static void vic_count_weight_init(struct vic *vic, unsigned long rate)
 {
+	const struct vic_config *config = vic->config;
 	struct host1x_client *client = &vic->client.base;
 	u32 weight = 0;

 	host1x_actmon_update_client_rate(client, rate, &weight);

-	if (weight)
-		vic_writel(vic, weight, NV_PVIC_TFBIF_ACTMON_ACTIVE_WEIGHT);
+	if (config->actmon_active_weight && weight)
+		vic_writel(vic, weight, config->actmon_active_weight);
+
 }

 static int __maybe_unused vic_runtime_resume(struct device *dev)
@@ -548,13 +658,26 @@ static int __maybe_unused vic_runtime_resume(struct device *dev)

 	usleep_range(10, 20);

-	err = vic_load_firmware(vic);
+	if (vic->config->supports_sid)
+		tegra_drm_program_iommu_regs(vic->dev, vic->regs, vic->config->transcfg_addr);
+
+	if (vic->config->has_riscv) {
+		err = vic_load_riscv_firmware(vic);
 		if (err < 0)
 			goto assert;

-	err = vic_boot(vic);
+		err = vic_riscv_boot(vic);
 		if (err < 0)
 			goto assert;
+	} else {
+		err = vic_load_falcon_firmware(vic);
+		if (err < 0)
+			goto assert;
+
+		err = vic_falcon_boot(vic);
+		if (err < 0)
+			goto assert;
+	}

 	/* Forcely set frequency as Fmax when device is resumed back */
 	vic->devfreq->resume_freq = vic->devfreq->scaling_max_freq;
@@ -638,12 +761,16 @@ static int vic_can_use_memory_ctx(struct tegra_drm_client *client, bool *support
 	struct vic *vic = to_vic(client);
 	int err;

+	if (vic->config->has_riscv) {
+		*supported = true;
+	} else {
 		/* This doesn't access HW so it's safe to call without powering up. */
-	err = vic_load_firmware(vic);
+		err = vic_load_falcon_firmware(vic);
 		if (err < 0)
 			return err;

 		*supported = vic->can_use_context;
+	}

 	return 0;
 }
@@ -657,6 +784,15 @@ static int vic_has_job_timestamping(struct tegra_drm_client *client, bool *suppo
 	return 0;
 }

+static int vic_skip_bl_swizzling(struct tegra_drm_client *client, bool *skip)
+{
+	struct vic *vic = to_vic(client);
+
+	*skip = vic->config->skip_bl_swizzling;
+
+	return 0;
+}
+
 static const struct tegra_drm_client_ops vic_ops = {
 	.open_channel = vic_open_channel,
 	.close_channel = vic_close_channel,
@@ -664,6 +800,7 @@ static const struct tegra_drm_client_ops vic_ops = {
 	.get_streamid_offset = tegra_drm_get_streamid_offset_thi,
 	.can_use_memory_ctx = vic_can_use_memory_ctx,
 	.has_job_timestamping = vic_has_job_timestamping,
+	.skip_bl_swizzling = vic_skip_bl_swizzling,
 };

 #define NVIDIA_TEGRA_124_VIC_FIRMWARE "nvidia/tegra124/vic03_ucode.bin"
@@ -688,6 +825,10 @@ static const struct vic_config vic_t186_config = {
 	.firmware = NVIDIA_TEGRA_186_VIC_FIRMWARE,
 	.version = 0x18,
 	.supports_sid = true,
+	.transcfg_addr = 0x2044,
+	.actmon_active_mask = 0x204c,
+	.actmon_active_borps = 0x2050,
+	.actmon_active_weight = 0x2054,
 };

 #define NVIDIA_TEGRA_194_VIC_FIRMWARE "nvidia/tegra194/vic.bin"
@@ -697,6 +838,10 @@ static const struct vic_config vic_t194_config = {
 	.version = 0x19,
 	.supports_sid = true,
 	.supports_timestamping = true,
+	.transcfg_addr = 0x2044,
+	.actmon_active_mask = 0x204c,
+	.actmon_active_borps = 0x2050,
+	.actmon_active_weight = 0x2054,
 };

 #define NVIDIA_TEGRA_234_VIC_FIRMWARE "nvidia/tegra234/vic.bin"
@@ -707,6 +852,25 @@ static const struct vic_config vic_t234_config = {
 	.supports_sid = true,
 	.supports_timestamping = true,
 	.has_crc_enable = true,
+	.transcfg_addr = 0x2044,
+	.actmon_active_mask = 0x204c,
+	.actmon_active_borps = 0x2050,
+	.actmon_active_weight = 0x2054,
+};
+
+#define NVIDIA_TEGRA_264_VIC_FIRMWARE "nvidia/tegra264/vic.bin"
+
+static const struct vic_config vic_t264_config = {
+	.firmware = NVIDIA_TEGRA_264_VIC_FIRMWARE,
+	.version = 0x26,
+	.supports_sid = true,
+	.supports_timestamping = true,
+	.has_riscv = true,
+	.transcfg_addr = 0x2244,
+	.actmon_active_mask = 0x224c,
+	.actmon_active_borps = 0x2250,
+	.actmon_active_weight = 0x2254,
+	.skip_bl_swizzling = true,
 };

 static const struct of_device_id tegra_vic_of_match[] = {
@@ -715,6 +879,7 @@ static const struct of_device_id tegra_vic_of_match[] = {
 	{ .compatible = "nvidia,tegra186-vic", .data = &vic_t186_config },
 	{ .compatible = "nvidia,tegra194-vic", .data = &vic_t194_config },
 	{ .compatible = "nvidia,tegra234-vic", .data = &vic_t234_config },
+	{ .compatible = "nvidia,tegra264-vic", .data = &vic_t264_config },
 	{ },
 };
 MODULE_DEVICE_TABLE(of, tegra_vic_of_match);
@@ -766,12 +931,26 @@ static int vic_probe(struct platform_device *pdev)
 		}
 	}

+	if (vic->config->has_riscv) {
+		vic->riscv.dev = dev;
+		vic->riscv.regs = vic->regs;
+
+		vic->riscv.os_desc.code_offset = 0x0;
+		vic->riscv.os_desc.code_size = 0x1200;
+		vic->riscv.os_desc.data_offset = 0x1200;
+		vic->riscv.os_desc.data_size = 0xa00;
+
+		err = tegra_drm_riscv_init(&vic->riscv);
+		if (err < 0)
+			return err;
+	} else {
 		vic->falcon.dev = dev;
 		vic->falcon.regs = vic->regs;

 		err = falcon_init(&vic->falcon);
 		if (err < 0)
 			return err;
+	}

 	platform_set_drvdata(pdev, vic);

@@ -790,7 +969,7 @@ static int vic_probe(struct platform_device *pdev)
 	err = host1x_client_register(&vic->client.base);
 	if (err < 0) {
 		dev_err(dev, "failed to register host1x client: %d\n", err);
-		goto exit_falcon;
+		goto exit_probe;
 	}

 	err = host1x_actmon_register(&vic->client.base);
@@ -801,13 +980,13 @@ static int vic_probe(struct platform_device *pdev)
 	err = clk_set_rate(vic->clk, ULONG_MAX);
 	if (err < 0) {
 		dev_err(&pdev->dev, "failed to set clock rate: %d\n", err);
-		goto exit_actmon;
+		goto exit_probe;
 	}

 	err = vic_devfreq_init(vic);
 	if (err < 0) {
 		dev_err(&pdev->dev, "failed to init devfreq: %d\n", err);
-		goto exit_actmon;
+		goto exit_probe;
 	}

 	if (vic->config->has_crc_enable) {
@@ -827,11 +1006,10 @@ static int vic_probe(struct platform_device *pdev)

 	return 0;

-exit_actmon:
-	host1x_actmon_unregister(&vic->client.base);
-	host1x_client_unregister(&vic->client.base);
-
-exit_falcon:
+exit_probe:
+	if (vic->config->has_riscv)
+		tegra_drm_riscv_exit(&vic->riscv);
+	else
 		falcon_exit(&vic->falcon);

 	return err;
@@ -852,6 +1030,9 @@ static int vic_remove(struct platform_device *pdev)

 	host1x_client_unregister(&vic->client.base);

+	if (vic->config->has_riscv)
+		tegra_drm_riscv_exit(&vic->riscv);
+	else
 		falcon_exit(&vic->falcon);

 	return 0;
@@ -905,3 +1086,6 @@ MODULE_FIRMWARE(NVIDIA_TEGRA_194_VIC_FIRMWARE);
 #if IS_ENABLED(CONFIG_ARCH_TEGRA_234_SOC)
 MODULE_FIRMWARE(NVIDIA_TEGRA_234_VIC_FIRMWARE);
 #endif
+#if IS_ENABLED(CONFIG_ARCH_TEGRA_264_SOC)
+MODULE_FIRMWARE(NVIDIA_TEGRA_264_VIC_FIRMWARE);
+#endif
--- a/drivers/gpu/drm/tegra/vic.h
+++ b/drivers/gpu/drm/tegra/vic.h
@@ -18,17 +18,11 @@
 #define CG_IDLE_CG_EN				(1 << 6)
 #define CG_WAKEUP_DLY_CNT(val)			((val & 0xf) << 16)

-#define NV_PVIC_TFBIF_ACTMON_ACTIVE_MASK	0x0000204c
-#define NV_PVIC_TFBIF_ACTMON_ACTIVE_BORPS	0x00002050
-#define NV_PVIC_TFBIF_ACTMON_ACTIVE_WEIGHT	0x00002054
-
 #define VIC_TFBIF_ACTMON_ACTIVE_MASK_STARVED	BIT(0)
 #define VIC_TFBIF_ACTMON_ACTIVE_MASK_STALLED	BIT(1)
 #define VIC_TFBIF_ACTMON_ACTIVE_MASK_DELAYED	BIT(2)
 #define VIC_TFBIF_ACTMON_ACTIVE_BORPS_ACTIVE	BIT(7)

-#define VIC_TFBIF_TRANSCFG	0x00002044
-
 /* Firmware offsets */

 #define VIC_UCODE_FCE_HEADER_OFFSET		(6*4)
--- a/drivers/gpu/drm/tegra/virt.c
+++ b/drivers/gpu/drm/tegra/virt.c
@@ -138,6 +138,14 @@ static int virt_engine_has_job_timestamping(struct tegra_drm_client *client, boo
 	return 0;
 }

+static int virt_skip_bl_swizzling(struct tegra_drm_client *client, bool *skip)
+{
+	if (of_machine_is_compatible("nvidia,tegra264"))
+		*skip = true;
+
+	return 0;
+}
+
 static const struct tegra_drm_client_ops virt_engine_ops = {
 	.open_channel = virt_engine_open_channel,
 	.close_channel = virt_engine_close_channel,
@@ -145,6 +153,7 @@ static const struct tegra_drm_client_ops virt_engine_ops = {
 	.get_streamid_offset = tegra_drm_get_streamid_offset_thi,
 	.can_use_memory_ctx = virt_engine_can_use_memory_ctx,
 	.has_job_timestamping = virt_engine_has_job_timestamping,
+	.skip_bl_swizzling = virt_skip_bl_swizzling,
 };

 static int virt_engine_setup_ivc(struct virt_engine *virt)