Merge pull request #34 from maxzhen/sync

Update source code

CMake/pkg.cmake

@@ -50,10 +50,17 @@ install(DIRECTORY ${AMDXDNA_BINS_DIR}/firmware/
   COMPONENT ${XDNA_COMPONENT}
   FILES_MATCHING
   PATTERN "*.sbin"
-  PATTERN "*.xclbin"
   PATTERN "download_raw" EXCLUDE
   )
 
+install(DIRECTORY ${AMDXDNA_BINS_DIR}/download_raw/xbutil_validate/bins/
+  DESTINATION xrt/${XDNA_COMPONENT}/bins
+  COMPONENT ${XDNA_COMPONENT}
+  FILES_MATCHING
+  PATTERN "*.xclbin"
+  PATTERN "*.txt"
+  )
+
 configure_file(
   ${CMAKE_CURRENT_SOURCE_DIR}/CMake/config/postinst.in
   ${CMAKE_CURRENT_BINARY_DIR}/package/postinst

CMakeLists.txt

@@ -1,5 +1,5 @@
 # SPDX-License-Identifier: Apache-2.0
-# Copyright (C) 2022-2023, Advanced Micro Devices, Inc. All rights reserved.
+# Copyright (C) 2022-2024, Advanced Micro Devices, Inc. All rights reserved.
 
 cmake_minimum_required(VERSION 3.19.0)
 project("amd-xdna")
@@ -29,6 +29,22 @@ add_subdirectory(src)
 # By default, build/build.sh downloads binaries to build/amdxdna_bins/
 set(AMDXDNA_BINS_DIR ${CMAKE_BINARY_DIR}/../amdxdna_bins)
 
+# Add common NPU support files
+install(DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/tools/bins/dpu_sequence/
+  DESTINATION xrt/${XDNA_COMPONENT}/bins/dpu_sequence
+  COMPONENT ${XDNA_COMPONENT}
+  FILES_MATCHING
+  PATTERN "*.txt" # dpu sequences are txt files
+  )
+
+# Add Phoenix NPU support files
+install(DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/tools/bins/
+  DESTINATION xrt/${XDNA_COMPONENT}/bins
+  COMPONENT ${XDNA_COMPONENT}
+  FILES_MATCHING
+  PATTERN "*.xclbin"
+  )
+
 set(XRT_SUBMOD_SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/xrt)
 set(XRT_SUBMOD_BINARY_DIR ${CMAKE_CURRENT_BINARY_DIR}/xrt)
 

README.md

@@ -105,7 +105,7 @@ A: Yes. For example, if you have 6.6-rc1 linux header install on your build mach
 ### Q: I'm developing amdxdna.ko driver module. How to enable XDNA_DBG() print?
 
 A: XDNA_DBG() relies on Linux's CONFIG_DYNAMIC_DEBUG framework, see Linux's [dynamic debug howto page](https://www.kernel.org/doc/html/v6.5/admin-guide/dynamic-debug-howto.html) for details.
-TL;DR, run `sudo insmod amdxdna.ko dyndbg==p` to enable XDNA_DBG() globally.
+TL;DR, run `sudo insmod amdxdna.ko dyndbg=+pf` to enable XDNA_DBG() globally, where +pf means enable debug printing and print the function name.
 
 ### Q: When install XRT plugin DEB package, apt-get/dpkg tool failed. What to do next?
 

build/build.sh

@@ -83,33 +83,6 @@ build_example()
 	cd ..
 }
 
-download_xclbins()
-{
-	#TODO: figure out the validate.xclbin location
-	local firmware_dir=${DOWNLOAD_BINS_DIR}/firmware
-
-	jq -c '.xclbins[]' "$INFO_JSON" |
-		while IFS= read -r line; do
-			local device=$(echo $line | jq -r '.device')
-			local pci_dev_id=$(echo $line | jq -r '.pci_device_id')
-			local pci_rev_id=$(echo $line | jq -r '.pci_revision_id')
-			local name=$(echo $line | jq -r '.name')
-			local url=$(echo $line | jq -r '.url')
-
-			if [[ -z "$url" ]]; then
-				echo "Empty URL for $device xclbin $name"
-				exit 1
-			fi
-
-			echo "Download $device XCLBIN $name :"
-			if [ ! -d "${firmware_dir}/${pci_dev_id}_${pci_rev_id}" ]; then
-				mkdir -p ${firmware_dir}/${pci_dev_id}_${pci_rev_id}
-			fi
-			wget -O ${firmware_dir}/${pci_dev_id}_${pci_rev_id}/$name $url
-
-		done
-}
-
 download_npufws()
 {
 	local firmware_dir=${DOWNLOAD_BINS_DIR}/firmware
@@ -129,9 +102,10 @@ download_npufws()
 			fi
 
 			echo "Download $device NPUFW version $version:"
-			if [ ! -d "${firmware_dir}/${pci_dev_id}_${pci_rev_id}" ]; then
-				mkdir -p ${firmware_dir}/${pci_dev_id}_${pci_rev_id}
+			if [ -d "${firmware_dir}/${pci_dev_id}_${pci_rev_id}" ]; then
+				rm ${firmware_dir}/${pci_dev_id}_${pci_rev_id}
 			fi
+			mkdir -p ${firmware_dir}/${pci_dev_id}_${pci_rev_id}
 			wget -O ${firmware_dir}/${pci_dev_id}_${pci_rev_id}/$fw_name $url
 
 		done
@@ -246,9 +220,13 @@ if [[ $clean == 1 ]]; then
 	exit 0
 fi
 
+if [[ $example == 1 ]]; then
+	build_example
+	exit 0
+fi
+
 if [[ $package == 1 ]]; then
 	download_npufws
-	download_xclbins
 	package_targets $DEBUG_BUILD_TYPE
 	package_targets $RELEASE_BUILD_TYPE
 	exit 0
@@ -262,8 +240,4 @@ if [[ $debug == 1 ]]; then
 	build_targets $DEBUG_BUILD_TYPE
 fi
 
-if [[ $example == 1 ]]; then
-	build_example
-fi
-
 # vim: ts=2 sw=2

src/driver/CMake/config/dkms.conf.in

@@ -1,6 +1,5 @@
 PACKAGE_NAME=@XDNA_DKMS_PKG_NAME@
 PACKAGE_VERSION="@XRT_PLUGIN_VERSION_STRING@"
-# Requires at least Linux 6.8 to compile
 BUILD_EXCLUSIVE_KERNEL_MIN=6.8
 MAKE[0]="cd driver/@XDNA_DRV_DIR@; make KERNEL_SRC=${kernel_source_dir}; cd ../.."
 CLEAN="cd driver/@XDNA_DRV_DIR@; make clean KERNEL_SRC=${kernel_source_dir}; cd ../.."

src/driver/amdxdna/amdxdna_ctx.c

@@ -241,9 +241,43 @@ int amdxdna_drm_config_hwctx_ioctl(struct drm_device *dev, void *data, struct dr
 	struct amdxdna_dev *xdna = to_xdna_dev(dev);
 	struct amdxdna_hwctx *hwctx;
 	int ret, idx;
+	u32 buf_size;
+	void *buf;
+	u64 val;
 
-	if (args->param_type >= DRM_AMDXDNA_HWCTX_CONFIG_NUM) {
-		XDNA_ERR(xdna, "Invalid HW context param type: %d", args->param_type);
+	if (!xdna->dev_info->ops->hwctx_config)
+		return -EOPNOTSUPP;
+
+	val = args->param_val;
+	buf_size = args->param_val_size;
+
+	switch (args->param_type) {
+	case DRM_AMDXDNA_HWCTX_CONFIG_CU:
+		/* For those types that param_val is pointer */
+		if (buf_size > PAGE_SIZE) {
+			XDNA_ERR(xdna, "Config CU param buffer too large");
+			return -E2BIG;
+		}
+
+		/* Hwctx needs to keep buf */
+		buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
+		if (!buf)
+			return -ENOMEM;
+
+		if (copy_from_user(buf, u64_to_user_ptr(val), buf_size)) {
+			kfree(buf);
+			return -EFAULT;
+		}
+
+		break;
+	case DRM_AMDXDNA_HWCTX_ASSIGN_DBG_BUF:
+	case DRM_AMDXDNA_HWCTX_REMOVE_DBG_BUF:
+		/* For those types that param_val is a value */
+		buf = NULL;
+		buf_size = 0;
+		break;
+	default:
+		XDNA_DBG(xdna, "Unknown HW context config type %d", args->param_type);
 		return -EINVAL;
 	}
 
@@ -255,17 +289,12 @@ int amdxdna_drm_config_hwctx_ioctl(struct drm_device *dev, void *data, struct dr
 		goto unlock_srcu;
 	}
 
-	if (!xdna->dev_info->ops->hwctx_config) {
-		ret = -EOPNOTSUPP;
-		goto unlock_srcu;
-	}
-
 	mutex_lock(&xdna->dev_lock);
-	ret = xdna->dev_info->ops->hwctx_config(hwctx, args->param_type, args->param_val,
-						args->param_val_size);
+	ret = xdna->dev_info->ops->hwctx_config(hwctx, args->param_type, val, buf, buf_size);
 	mutex_unlock(&xdna->dev_lock);
 
 unlock_srcu:
+	kfree(buf);
 	srcu_read_unlock(&client->hwctx_srcu, idx);
 	return ret;
 }
@@ -320,21 +349,21 @@ amdxdna_arg_bos_lookup(struct amdxdna_client *client,
 		}
 		abo = to_xdna_obj(gobj);
 
-		spin_lock(&abo->lock);
+		mutex_lock(&abo->lock);
 		if (abo->pinned) {
-			spin_unlock(&abo->lock);
+			mutex_unlock(&abo->lock);
 			job->bos[i] = gobj;
 			continue;
 		}
 
 		ret = amdxdna_gem_pin_nolock(abo);
 		if (ret) {
-			spin_unlock(&abo->lock);
+			mutex_unlock(&abo->lock);
 			drm_gem_object_put(gobj);
 			goto put_shmem_bo;
 		}
 		abo->pinned = true;
-		spin_unlock(&abo->lock);
+		mutex_unlock(&abo->lock);
 
 		job->bos[i] = gobj;
 	}
@@ -346,11 +375,12 @@ amdxdna_arg_bos_lookup(struct amdxdna_client *client,
 	return ret;
 }
 
-static int amdxdna_cmds_submit(struct amdxdna_client *client, u32 hwctx_id,
-			       struct amdxdna_gem_obj *cmd_bo, u32 cmd_bo_cnt,
-			       u32 *bo_hdls, u32 bo_cnt, u64 *seq)
+static int amdxdna_cmds_submit(struct amdxdna_client *client,
+			       struct amdxdna_gem_obj *cmd_bo, u32 *bo_hdls,
+			       struct amdxdna_drm_exec_cmd *args)
 {
 	struct amdxdna_dev *xdna = client->xdna;
+	u32 bo_cnt = args->arg_bo_count;
 	struct amdxdna_sched_job *job;
 	struct amdxdna_hwctx *hwctx;
 	u32 *cu_mask;
@@ -368,10 +398,10 @@ static int amdxdna_cmds_submit(struct amdxdna_client *client, u32 hwctx_id,
 	}
 
 	idx = srcu_read_lock(&client->hwctx_srcu);
-	hwctx = idr_find(&client->hwctx_idr, hwctx_id);
+	hwctx = idr_find(&client->hwctx_idr, args->hwctx);
 	if (!hwctx) {
 		XDNA_DBG(xdna, "PID %d failed to get hwctx %d",
-			 client->pid, hwctx_id);
+			 client->pid, args->hwctx);
 		ret = -EINVAL;
 		goto unlock_srcu;
 	}
@@ -421,7 +451,7 @@ static int amdxdna_cmds_submit(struct amdxdna_client *client, u32 hwctx_id,
 	}
 	kref_init(&job->refcnt);
 
-	ret = xdna->dev_info->ops->cmd_submit(hwctx, job, seq);
+	ret = xdna->dev_info->ops->cmd_submit(hwctx, job, &args->seq);
 	if (ret)
 		goto put_fence;
 
@@ -432,7 +462,7 @@ static int amdxdna_cmds_submit(struct amdxdna_client *client, u32 hwctx_id,
 	 * For here we can unlock SRCU.
 	 */
 	srcu_read_unlock(&client->hwctx_srcu, idx);
-	trace_xdna_job(job->hwctx->name, "job pushed", *seq);
+	trace_xdna_job(hwctx->name, "job pushed", args->seq);
 
 	return 0;
 
@@ -493,14 +523,13 @@ int amdxdna_drm_exec_cmd_ioctl(struct drm_device *dev, void *data, struct drm_fi
 		goto free_bo_hdls;
 	}
 
-	if(to_gobj(cmd_bo)->size < sizeof(struct amdxdna_cmd)) {
+	if (to_gobj(cmd_bo)->size < sizeof(struct amdxdna_cmd)) {
 		XDNA_DBG(xdna, "Bad cmd BO size: %ld", to_gobj(cmd_bo)->size);
 		ret = -EINVAL;
 		goto put_cmd_bo;
 	}
 
-	ret = amdxdna_cmds_submit(client, args->hwctx, cmd_bo, args->cmd_bo_count,
-				  bo_hdls, args->arg_bo_count, &args->seq);
+	ret = amdxdna_cmds_submit(client, cmd_bo, bo_hdls, args);
 	if (ret) {
 		XDNA_DBG(xdna, "Submit cmds failed, ret %d", ret);
 		goto put_cmd_bo;

src/driver/amdxdna/amdxdna_ctx.h

@@ -55,6 +55,8 @@ struct amdxdna_hwctx {
 	u32				num_tiles;
 	u32				mem_size;
 	u32				fw_ctx_id;
+	u32				col_list_len;
+	u32				*col_list;
 	u32				start_col;
 	u32				num_col;
 #define HWCTX_STAT_INIT  0
@@ -87,6 +89,12 @@ struct amdxdna_sched_job {
 	struct drm_gem_object	*bos[] __counted_by(bo_cnt);
 };
 
+static inline u32 amdxdna_hwctx_col_map(struct amdxdna_hwctx *hwctx)
+{
+	return GENMASK(hwctx->start_col + hwctx->num_col - 1,
+		       hwctx->start_col);
+}
+
 void amdxdna_job_put(struct amdxdna_sched_job *job);
 
 void amdxdna_hwctx_remove_all(struct amdxdna_client *client);

src/driver/amdxdna/amdxdna_drv.c

@@ -157,12 +157,33 @@ static int amdxdna_drm_get_info_ioctl(struct drm_device *dev, void *data, struct
 {
 	struct amdxdna_drm_get_info *args = data;
 	struct amdxdna_dev *xdna = to_xdna_dev(dev);
+	int ret;
 
 	if (!xdna->dev_info->ops->get_info)
 		return -EOPNOTSUPP;
 
 	XDNA_DBG(xdna, "Request parameter %u", args->param);
-	return xdna->dev_info->ops->get_info(xdna, args);
+	mutex_lock(&xdna->dev_lock);
+	ret = xdna->dev_info->ops->get_info(xdna, args);
+	mutex_unlock(&xdna->dev_lock);
+	return ret;
+}
+
+static int amdxdna_drm_set_state_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
+{
+	struct amdxdna_drm_set_state *args = data;
+	struct amdxdna_dev *xdna = to_xdna_dev(dev);
+	int ret = 0;
+
+	if (!xdna->dev_info->ops->set_state)
+		return -EOPNOTSUPP;
+
+	XDNA_DBG(xdna, "Request parameter %u", args->param);
+	mutex_lock(&xdna->dev_lock);
+	ret = xdna->dev_info->ops->set_state(xdna, args);
+	mutex_unlock(&xdna->dev_lock);
+
+	return ret;
 }
 
 static const struct drm_ioctl_desc amdxdna_drm_ioctls[] = {
@@ -179,6 +200,7 @@ static const struct drm_ioctl_desc amdxdna_drm_ioctls[] = {
 	DRM_IOCTL_DEF_DRV(AMDXDNA_WAIT_CMD, amdxdna_drm_wait_cmd_ioctl, 0),
 	/* Query */
 	DRM_IOCTL_DEF_DRV(AMDXDNA_GET_INFO, amdxdna_drm_get_info_ioctl, 0),
+	DRM_IOCTL_DEF_DRV(AMDXDNA_SET_STATE, amdxdna_drm_set_state_ioctl, DRM_ROOT_ONLY),
 };
 
 static const struct file_operations amdxdna_fops = {

src/driver/amdxdna/amdxdna_drv.h

@@ -40,12 +40,13 @@ struct amdxdna_dev_ops {
 	int (*resume)(struct amdxdna_dev *xdna);
 	void (*suspend)(struct amdxdna_dev *xdna);
 	int (*get_info)(struct amdxdna_dev *xdna, struct amdxdna_drm_get_info *args);
+	int (*set_state)(struct amdxdna_dev *xdna, struct amdxdna_drm_set_state *args);
 	int (*mmap)(struct amdxdna_dev *xdna, struct vm_area_struct *vma);
 	void (*debugfs)(struct amdxdna_dev *xdna);
 
 	int (*hwctx_init)(struct amdxdna_hwctx *hwctx);
 	void (*hwctx_fini)(struct amdxdna_hwctx *hwctx);
-	int (*hwctx_config)(struct amdxdna_hwctx *hwctx, u32 type, u64 value, u32 size);
+	int (*hwctx_config)(struct amdxdna_hwctx *hwctx, u32 type, u64 value, void *buf, u32 size);
 	void (*hwctx_suspend)(struct amdxdna_hwctx *hwctx);
 	void (*hwctx_resume)(struct amdxdna_hwctx *hwctx);
 	int (*cmd_submit)(struct amdxdna_hwctx *hwctx, struct amdxdna_sched_job *job, u64 *seq);