securemsm-kernel: smmu-proxy: Add full drivers

Add the full drivers for the PVM and TVM.

Change-Id: I3046306fff91670928c750ffa31db02bd0240c9c
Signed-off-by: Chris Goldsworthy <[email protected]>

Android.mk

@@ -122,15 +122,17 @@ include $(DLKM_DIR)/Build_external_kernelmodule.mk
 endif #TARGET_ENABLE_QSEECOM OR TARGET_BOARD_AUTO
 ###################################################
 ###################################################
+ifeq ($(TARGET_USES_SMMU_PROXY), true)
 include $(CLEAR_VARS)
 #LOCAL_SRC_FILES           := $(SSG_SRC_FILES)
-LOCAL_EXPORT_KO_INCLUDE_DIRS := $(LOCAL_PATH)/smmu-proxy/
+LOCAL_EXPORT_KO_INCLUDE_DIRS := $(LOCAL_PATH)/smmu-proxy/ $(LOCAL_PATH)/
 LOCAL_MODULE              := smmu_proxy_dlkm.ko
 LOCAL_MODULE_KBUILD_NAME  := smmu_proxy_dlkm.ko
 LOCAL_MODULE_TAGS         := optional
 LOCAL_MODULE_DEBUG_ENABLE := true
 LOCAL_MODULE_PATH         := $(KERNEL_MODULES_OUT)
 include $(DLKM_DIR)/Build_external_kernelmodule.mk
+endif
 ###################################################
 ###################################################
 endif #COMPILE_SECUREMSM_DLKM check

Kbuild

@@ -40,12 +40,15 @@ hdcp_qseecom_dlkm-objs := hdcp/hdcp_qseecom.o
 obj-$(CONFIG_HW_RANDOM_MSM_LEGACY) += qrng_dlkm.o
 qrng_dlkm-objs := qrng/msm_rng.o
 
-include $(SSG_MODULE_ROOT)/config/sec-kernel_defconfig_smmu_proxy.conf
-LINUXINCLUDE += -include $(SSG_MODULE_ROOT)/config/sec-kernel_defconfig_smmu_proxy.h
-
-ifneq ($(CONFIG_ARCH_QTI_VM), y)
-
-obj-$(CONFIG_QTI_SMMU_PROXY) += smmu_proxy_dlkm.o
-smmu_proxy_dlkm-objs := smmu-proxy/qti-smmu-proxy.o
-
+ifneq (, $(filter y, $(ARCH_QTI_VM) $(CONFIG_ARCH_PINEAPPLE)))
+    include $(SSG_MODULE_ROOT)/config/sec-kernel_defconfig_smmu_proxy.conf
+    LINUXINCLUDE += -include $(SSG_MODULE_ROOT)/config/sec-kernel_defconfig_smmu_proxy.h
+
+    obj-$(CONFIG_QTI_SMMU_PROXY) += smmu_proxy_dlkm.o
+    smmu_proxy_dlkm-objs := smmu-proxy/qti-smmu-proxy-common.o
+    ifneq ($(CONFIG_ARCH_QTI_VM), y)
+    smmu_proxy_dlkm-objs += smmu-proxy/qti-smmu-proxy-pvm.o
+    else
+    smmu_proxy_dlkm-objs += smmu-proxy/qti-smmu-proxy-tvm.o
+    endif
 endif

securemsm_kernel_product_board.mk

@@ -16,7 +16,10 @@ PRODUCT_PACKAGES += qcrypto-msm_dlkm.ko
 PRODUCT_PACKAGES += hdcp_qseecom_dlkm.ko
 PRODUCT_PACKAGES += qrng_dlkm.ko
 PRODUCT_PACKAGES += smcinvoke_dlkm.ko
+
+ifeq ($(TARGET_USES_SMMU_PROXY), true)
 PRODUCT_PACKAGES += smmu_proxy_dlkm.ko
+endif
 
 #Enable Qseecom if TARGET_ENABLE_QSEECOM or TARGET_BOARD_AUTO is set to true
 ifneq (, $(filter true, $(TARGET_ENABLE_QSEECOM) $(TARGET_BOARD_AUTO)))

securemsm_kernel_vendor_board.mk

@@ -24,7 +24,9 @@ BOARD_VENDOR_KERNEL_MODULES += $(KERNEL_MODULES_OUT)/smcinvoke_dlkm.ko
 BOARD_VENDOR_RAMDISK_KERNEL_MODULES += $(KERNEL_MODULES_OUT)/smcinvoke_dlkm.ko
 BOARD_VENDOR_RAMDISK_RECOVERY_KERNEL_MODULES_LOAD += $(KERNEL_MODULES_OUT)/smcinvoke_dlkm.ko
 
+ifeq ($(TARGET_USES_SMMU_PROXY), true)
 BOARD_VENDOR_KERNEL_MODULES += $(KERNEL_MODULES_OUT)/smmu_proxy_dlkm.ko
+endif
 
 #Enable Qseecom if TARGET_ENABLE_QSEECOM or TARGET_BOARD_AUTO is set to true
 ifneq (, $(filter true, $(TARGET_ENABLE_QSEECOM) $(TARGET_BOARD_AUTO)))

smmu-proxy/linux/qti-smmu-proxy.h

@@ -5,7 +5,11 @@
 #ifndef __QTI_SMMU_PROXY_H_
 #define __QTI_SMMU_PROXY_H_
 
-#include "../uapi/linux/qti-smmu-proxy.h"
+#include <linux/dma-buf.h>
+#include <linux/scatterlist.h>
+#include <linux/align.h>
+
+#include <smmu-proxy/uapi/linux/qti-smmu-proxy.h>
 
 #define SMMU_PROXY_MEM_ALIGNMENT (1 << 21)
 

smmu-proxy/qti-smmu-proxy-common.c

@@ -0,0 +1,99 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (c) 2023 Qualcomm Innovation Center, Inc. All rights reserved.
+ */
+
+#include <linux/cdev.h>
+#include "qti-smmu-proxy-common.h"
+#include "../include/linux/smcinvoke.h"
+#include "../include/linux/ITrustedCameraDriver.h"
+#include "../include/linux/CTrustedCameraDriver.h"
+#include "../include/linux/IClientEnv.h"
+
+#define SMMU_PROXY_MAX_DEVS 1
+static dev_t smmu_proxy_dev_no;
+static struct class *smmu_proxy_class;
+static struct cdev smmu_proxy_char_dev;
+
+static struct csf_version cached_csf_version;
+
+int smmu_proxy_get_csf_version(struct csf_version *csf_version)
+{
+	int ret;
+	struct Object client_env = {0};
+	struct Object sc_object;
+
+	if (cached_csf_version.arch_ver != 0) {
+		csf_version->arch_ver = cached_csf_version.arch_ver;
+		csf_version->max_ver = cached_csf_version.max_ver;
+		csf_version->min_ver = cached_csf_version.min_ver;
+
+		return 0;
+	}
+
+	ret = get_client_env_object(&client_env);
+	if (ret) {
+		pr_err("%s: Failed to get env object rc: %d\n", __func__,
+		       ret);
+		return ret;
+	}
+
+	ret = IClientEnv_open(client_env, CTrustedCameraDriver_UID, &sc_object);
+	if (ret) {
+		pr_err("%s: Failed to get seccam object rc: %d\n", __func__,
+		       ret);
+		return ret;
+	}
+
+	ret = ITrustedCameraDriver_getVersion(sc_object, &csf_version->arch_ver,
+					      &csf_version->max_ver,
+					      &csf_version->min_ver);
+
+	Object_release(sc_object);
+	Object_release(client_env);
+
+	return ret;
+}
+EXPORT_SYMBOL(smmu_proxy_get_csf_version);
+
+int smmu_proxy_create_dev(const struct file_operations *fops)
+{
+	int ret;
+	struct device *class_dev;
+
+	ret = alloc_chrdev_region(&smmu_proxy_dev_no, 0, SMMU_PROXY_MAX_DEVS,
+				  "qti-smmu-proxy");
+	if (ret < 0)
+		return ret;
+
+	smmu_proxy_class = class_create(THIS_MODULE, "qti-smmu-proxy");
+	if (IS_ERR(smmu_proxy_class)) {
+		ret = PTR_ERR(smmu_proxy_class);
+		goto err_class_create;
+	}
+
+	cdev_init(&smmu_proxy_char_dev, fops);
+	ret = cdev_add(&smmu_proxy_char_dev, smmu_proxy_dev_no,
+		       SMMU_PROXY_MAX_DEVS);
+	if (ret < 0)
+		goto err_cdev_add;
+
+	class_dev = device_create(smmu_proxy_class, NULL, smmu_proxy_dev_no, NULL,
+				  "qti-smmu-proxy");
+	if (IS_ERR(class_dev)) {
+		ret = PTR_ERR(class_dev);
+		goto err_dev_create;
+	}
+
+	return 0;
+
+err_dev_create:
+	cdev_del(&smmu_proxy_char_dev);
+err_cdev_add:
+	class_destroy(smmu_proxy_class);
+err_class_create:
+	unregister_chrdev_region(smmu_proxy_dev_no, SMMU_PROXY_MAX_DEVS);
+
+	return ret;
+}
+

smmu-proxy/qti-smmu-proxy-common.h

@@ -0,0 +1,30 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2023 Qualcomm Innovation Center, Inc. All rights reserved.
+ */
+
+#ifndef __QTI_SMMU_PROXY_COMMON_H_
+#define __QTI_SMMU_PROXY_COMMON_H_
+
+#include <linux/module.h>
+#include <linux/of_platform.h>
+#include <linux/platform_device.h>
+#include <linux/cdev.h>
+#include <linux/dma-buf.h>
+
+#include <linux/mem-buf.h>
+#include <soc/qcom/secure_buffer.h>
+#include <linux/gunyah/gh_msgq.h>
+#include "qti-smmu-proxy-msgq.h"
+#include "linux/qti-smmu-proxy.h"
+
+union smmu_proxy_ioctl_arg {
+	struct csf_version csf_version;
+	struct smmu_proxy_acl_ctl acl_ctl;
+	struct smmu_proxy_wipe_buf_ctl wipe_buf_ctl;
+	struct smmu_proxy_get_dma_buf_ctl get_dma_buf_ctl;
+};
+
+int smmu_proxy_create_dev(const struct file_operations *fops);
+
+#endif /* __QTI_SMMU_PROXY_COMMON_H_ */

smmu-proxy/qti-smmu-proxy-msgq.h

@@ -0,0 +1,107 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (c) 2022 Qualcomm Innovation Center, Inc. All rights reserved.
+ */
+#ifndef SMMU_PROXY_MSGQ_H
+#define SMMU_PROXY_MSGQ_H
+
+#include <linux/gunyah/gh_rm_drv.h>
+
+
+/**
+ * enum smmu_proxy_msg_type: Message types used by the SMMU proxy driver for
+ * communication.
+ * @SMMU_PROXY_MAP: The message is a request to map memory into the VM's
+ * SMMU.
+ * @SMMU_PROXY_MAP_RESP: The message is a response from a remote VM to a
+ * mapping request issued by the receiving VM
+ * @SMMU_PROXY_UNMAP: The message is a request to unmap some previously
+ * SMMU-mapped memory from the VM
+ * @SMMU_PROXY_UNMAP_RESP: The message is a response from a remote VM to an
+ * unmapping request issued by the receiving VM
+ * @SMMU_PROXY_ERR_RESP: The message is a response from a remote VM to give
+ * a generic error response for a prior message sent to the remote VM
+ */
+enum smmu_proxy_msg_type {
+	SMMU_PROXY_MAP,
+	SMMU_PROXY_MAP_RESP,
+	SMMU_PROXY_UNMAP,
+	SMMU_PROXY_UNMAP_RESP,
+	SMMU_PROXY_ERR_RESP,
+	SMMU_PROXY_MSG_MAX,
+};
+
+/**
+ * struct smmu_proxy_msg_hdr: The header for SMMU proxy messages
+ * @msg_type: The type of message.
+ * @msg_size: The size of message.
+ */
+struct smmu_proxy_msg_hdr {
+	u32 msg_type;
+	u32 msg_size;
+} __packed;
+
+/**
+ * struct smmu_proxy_msg_hdr: The header for responses to SMMU proxy messages
+ * @msg_type: The type of message.
+ * @msg_size: The size of message.
+ * @ret: Return code from remote VM
+ */
+struct smmu_proxy_resp_hdr {
+	u32 msg_type;
+	u32 msg_size;
+	s32 ret;
+} __packed;
+
+/**
+ * struct smmu_proxy_map_req: The message format for an SMMU mapping request from
+ * another VM.
+ * @hdr: Message header
+ * @hdl: The memparcel handle associated with the memory to be mapped in the SMMU
+ * of the relevant VM
+ * @cb_id: Context bank ID that we will map the memory associated with @hdl to
+ * @acl_desc: A GH ACL descriptor that describes the VMIDs that will be
+ * accessing the memory, as well as what permissions each VMID will have.
+ */
+struct smmu_proxy_map_req {
+	struct smmu_proxy_msg_hdr hdr;
+	u32 hdl;
+	u32 cb_id;
+	struct gh_acl_desc acl_desc;
+} __packed;
+
+/**
+ * struct smmu_proxy_map_resp: The message format for an SMMU mapping
+ * request response.
+ * @hdr: Response header
+ * @iova: IOVA of mapped memory
+ * @mapping_len: Lenth of IOMMU IOVA mapping
+ */
+struct smmu_proxy_map_resp {
+	struct smmu_proxy_resp_hdr hdr;
+	u64 iova;
+	u64 mapping_len;
+} __packed;
+
+/**
+ * struct smmu_proxy_unmap_req: The message format for an SMMU unmapping request from
+ * another VM.
+ * @hdr: Message header
+ * @hdl: The memparcel handle associated with the memory to be mapped in the SMMU
+ * of the relevant VM
+ */
+struct smmu_proxy_unmap_req {
+	struct smmu_proxy_msg_hdr hdr;
+	u32 hdl;
+} __packed;
+
+/**
+ * struct smmu_proxy_unmap_resp: The message format for an SMMU unmapping
+ * request response.
+ * @hdr: Response header
+ */
+struct smmu_proxy_unmap_resp {
+	struct smmu_proxy_resp_hdr hdr;
+} __packed;
+
+#endif /* SMMU_PROXY_MSGQ_H */

smmu-proxy/qti-smmu-proxy-pvm.c

@@ -0,0 +1,323 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (c) 2023 Qualcomm Innovation Center, Inc. All rights reserved.
+ */
+
+#include "qti-smmu-proxy-common.h"
+
+#include <linux/qti-smmu-proxy-callbacks.h>
+#include <linux/qcom-dma-mapping.h>
+#include <linux/of.h>
+
+static void *msgq_hdl;
+
+DEFINE_MUTEX(sender_mutex);
+
+static const struct file_operations smmu_proxy_dev_fops;
+
+int smmu_proxy_unmap(void *data)
+{
+	struct dma_buf *dmabuf;
+	void *buf;
+	size_t size;
+	int ret;
+	struct smmu_proxy_unmap_req *req;
+	struct smmu_proxy_unmap_resp *resp;
+
+	mutex_lock(&sender_mutex);
+	buf = kzalloc(GH_MSGQ_MAX_MSG_SIZE_BYTES, GFP_KERNEL);
+	if (!buf) {
+		ret = -ENOMEM;
+		pr_err("%s: Failed to allocate memory!\n", __func__);
+		goto out;
+	}
+
+	req = buf;
+
+	dmabuf = data;
+	ret = mem_buf_dma_buf_get_memparcel_hdl(dmabuf, &req->hdl);
+	if (ret) {
+		pr_err("%s: Failed to get memparcel handle rc: %d\n", __func__, ret);
+		goto free_buf;
+	}
+
+	req->hdr.msg_type = SMMU_PROXY_UNMAP;
+	req->hdr.msg_size = sizeof(*req);
+
+	ret = gh_msgq_send(msgq_hdl, (void *) req, req->hdr.msg_size, 0);
+	if (ret < 0) {
+		pr_err("%s: failed to send message rc: %d\n", __func__, ret);
+		goto free_buf;
+	}
+
+	/*
+	 * No need to validate size -  gh_msgq_recv() ensures that sizeof(*resp) <
+	 * GH_MSGQ_MAX_MSG_SIZE_BYTES
+	 */
+	ret = gh_msgq_recv(msgq_hdl, buf, sizeof(*resp), &size, 0);
+	if (ret < 0) {
+		pr_err_ratelimited("%s: failed to receive message rc: %d\n", __func__, ret);
+		goto free_buf;
+	}
+
+	resp = buf;
+	if (resp->hdr.ret) {
+		ret = resp->hdr.ret;
+		pr_err("%s: Unmap call failed on remote VM, rc: %d\n", __func__,
+		       resp->hdr.ret);
+	}
+
+free_buf:
+	kfree(buf);
+out:
+	mutex_unlock(&sender_mutex);
+
+	return ret;
+}
+
+int smmu_proxy_map(struct device *client_dev, struct sg_table *proxy_iova,
+		   struct dma_buf *dmabuf)
+{
+	void *buf;
+	size_t size;
+	int ret = 0;
+	int n_acl_entries, i;
+	int vmids[2] = { VMID_TVM, VMID_OEMVM };
+	int perms[2] = { PERM_READ | PERM_WRITE, PERM_READ | PERM_WRITE};
+	struct csf_version csf_version;
+	struct mem_buf_lend_kernel_arg arg = {0};
+	struct smmu_proxy_map_req *req;
+	struct smmu_proxy_map_resp *resp;
+
+	ret = smmu_proxy_get_csf_version(&csf_version);
+	if (ret) {
+		return ret;
+	}
+
+	/*
+	 * We enter this function iff the CSF version is 2.5.* . If CSF 2.5.1
+	 * is in use, we set n_acl_entries to two, in order to assign this
+	 * memory to the TVM and OEM VM. If CSF 2.5.0 is in use, we just assign
+	 * it to the TVM.
+	 */
+	n_acl_entries = csf_version.min_ver == 1 ? 2 : 1;
+
+	mutex_lock(&sender_mutex);
+	buf = kzalloc(GH_MSGQ_MAX_MSG_SIZE_BYTES, GFP_KERNEL);
+	if (!buf) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	if (mem_buf_dma_buf_exclusive_owner(dmabuf)) {
+		arg.vmids = vmids;
+		arg.perms = perms;
+		arg.nr_acl_entries = n_acl_entries;
+
+		ret = mem_buf_lend(dmabuf, &arg);
+		if (ret) {
+			pr_err("%s: Failed to lend buf rc: %d\n", __func__, ret);
+			goto free_buf;
+		}
+	}
+
+	/* Prepare the message */
+	req = buf;
+	req->acl_desc.n_acl_entries = n_acl_entries;
+	for (i = 0; i < n_acl_entries; i++) {
+		req->acl_desc.acl_entries[i].vmid = vmids[i];
+		req->acl_desc.acl_entries[i].perms = perms[i];
+	}
+
+	ret = mem_buf_dma_buf_get_memparcel_hdl(dmabuf, &req->hdl);
+	if (ret) {
+		pr_err("%s: Failed to get memparcel handle rc: %d\n", __func__, ret);
+		goto free_buf;
+	}
+
+	ret = of_property_read_u32(client_dev->of_node,
+				   "qti,smmu-proxy-cb-id",
+				   &req->cb_id);
+	if (ret) {
+		dev_err(client_dev, "%s: Err reading 'qti,smmu-proxy-cb-id' rc: %d\n",
+			__func__, ret);
+		goto free_buf;
+	}
+
+	req->hdr.msg_type = SMMU_PROXY_MAP;
+	req->hdr.msg_size = offsetof(struct smmu_proxy_map_req,
+				acl_desc.acl_entries[n_acl_entries]);
+
+	ret = gh_msgq_send(msgq_hdl, (void *) req, req->hdr.msg_size, 0);
+	if (ret < 0) {
+		pr_err("%s: failed to send message rc: %d\n", __func__, ret);
+		goto free_buf;
+	}
+
+	/*
+	 * No need to validate size -  gh_msgq_recv() ensures that sizeof(*resp) <
+	 * GH_MSGQ_MAX_MSG_SIZE_BYTES
+	 */
+	ret = gh_msgq_recv(msgq_hdl, buf, sizeof(*resp), &size, 0);
+	if (ret < 0) {
+		pr_err_ratelimited("%s: failed to receive message rc: %d\n", __func__, ret);
+		goto free_buf;
+	}
+
+	resp = buf;
+
+	if (resp->hdr.ret) {
+		proxy_iova = ERR_PTR(resp->hdr.ret);
+		pr_err_ratelimited("%s: Map call failed on remote VM, rc: %d\n", __func__,
+				   resp->hdr.ret);
+		goto free_buf;
+	}
+
+	ret = mem_buf_dma_buf_set_destructor(dmabuf, smmu_proxy_unmap, dmabuf);
+	if (ret) {
+		pr_err_ratelimited("%s: Failed to set vmperm destructor, rc: %d\n",
+				   __func__, ret);
+		goto free_buf;
+	}
+
+	sg_dma_address(proxy_iova->sgl) = resp->iova;
+	sg_dma_len(proxy_iova->sgl) = resp->mapping_len;
+	/*
+	 * We set the number of entries to one here, as we only allow the mapping to go
+	 * through on the TVM if the sg_table returned by dma_buf_map_attachment has one
+	 * entry.
+	 */
+	proxy_iova->nents = 1;
+
+free_buf:
+	kfree(buf);
+out:
+	mutex_unlock(&sender_mutex);
+
+	return ret;
+}
+
+void smmu_proxy_unmap_nop(struct device *client_dev, struct sg_table *table,
+			  struct dma_buf *dmabuf)
+{
+
+}
+
+
+static long smmu_proxy_dev_ioctl(struct file *filp, unsigned int cmd,
+			      unsigned long arg)
+{
+	unsigned int dir = _IOC_DIR(cmd);
+	union smmu_proxy_ioctl_arg ioctl_arg;
+	int ret;
+
+	if (_IOC_SIZE(cmd) > sizeof(ioctl_arg))
+		return -EINVAL;
+
+	if (copy_from_user(&ioctl_arg, (void __user *)arg, _IOC_SIZE(cmd)))
+		return -EFAULT;
+
+	if (!(dir & _IOC_WRITE))
+		memset(&ioctl_arg, 0, sizeof(ioctl_arg));
+
+	switch (cmd) {
+	case QTI_SMMU_PROXY_GET_VERSION_IOCTL:
+	{
+		struct csf_version *csf_version =
+			&ioctl_arg.csf_version;
+
+		ret = smmu_proxy_get_csf_version(csf_version);
+		if(ret)
+			return ret;
+
+		break;
+	}
+
+	default:
+		return -ENOTTY;
+	}
+
+	if (dir & _IOC_READ) {
+		if (copy_to_user((void __user *)arg, &ioctl_arg,
+				 _IOC_SIZE(cmd)))
+			return -EFAULT;
+	}
+
+	return 0;
+}
+
+static const struct file_operations smmu_proxy_dev_fops = {
+	.unlocked_ioctl = smmu_proxy_dev_ioctl,
+	.compat_ioctl = compat_ptr_ioctl,
+};
+
+static int sender_probe_handler(struct platform_device *pdev)
+{
+	int ret;
+	struct csf_version csf_version;
+
+	msgq_hdl = gh_msgq_register(GH_MSGQ_LABEL_SMMU_PROXY);
+	if (IS_ERR(msgq_hdl)) {
+		ret = PTR_ERR(msgq_hdl);
+		pr_err("%s: Queue registration failed rc: %d!\n", __func__, PTR_ERR(msgq_hdl));
+		return ret;
+	}
+
+	ret = smmu_proxy_get_csf_version(&csf_version);
+	if (ret) {
+		pr_err("%s: Failed to get CSF version rc: %d\n", __func__, ret);
+		goto free_msgq;
+	}
+
+	if (csf_version.arch_ver == 2 && csf_version.max_ver == 0) {
+		ret = qti_smmu_proxy_register_callbacks(NULL, NULL);
+	} else if (csf_version.arch_ver == 2 && csf_version.max_ver == 5) {
+		ret = qti_smmu_proxy_register_callbacks(smmu_proxy_map, smmu_proxy_unmap_nop);
+	} else {
+		pr_err("%s: Invalid CSF version: %d.%d\n", __func__, csf_version.arch_ver,
+			csf_version.max_ver);
+		goto free_msgq;
+	}
+
+	if (ret) {
+		pr_err("%s: Failed to set SMMU proxy callbacks rc: %d\n", __func__, ret);
+		goto free_msgq;
+	}
+
+	ret = smmu_proxy_create_dev(&smmu_proxy_dev_fops);
+	if (ret) {
+		pr_err("%s: Failed to create character device rc: %d\n", __func__,
+		       ret);
+		goto set_callbacks_null;
+	}
+
+	return 0;
+
+set_callbacks_null:
+	qti_smmu_proxy_register_callbacks(NULL, NULL);
+free_msgq:
+	gh_msgq_unregister(msgq_hdl);
+	return ret;
+}
+
+static const struct of_device_id smmu_proxy_match_table[] = {
+	{.compatible = "smmu-proxy-sender"},
+	{},
+};
+
+static struct platform_driver smmu_proxy_driver = {
+	.probe = sender_probe_handler,
+	.driver = {
+		.name = "qti-smmu-proxy",
+		.of_match_table = smmu_proxy_match_table,
+	},
+};
+
+int __init init_smmu_proxy_driver(void)
+{
+	return platform_driver_register(&smmu_proxy_driver);
+}
+module_init(init_smmu_proxy_driver);
+
+MODULE_IMPORT_NS(DMA_BUF);
+MODULE_LICENSE("GPL v2");

smmu-proxy/qti-smmu-proxy-tvm.c

@@ -0,0 +1,738 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (c) 2023 Qualcomm Innovation Center, Inc. All rights reserved.
+ */
+
+#include <linux/kthread.h>
+
+#include <linux/qcom-iommu-util.h>
+#include <dt-bindings/arm/msm/qti-smmu-proxy-dt-ids.h>
+#include "qti-smmu-proxy-common.h"
+
+#define RECEIVER_COMPAT_STR "smmu-proxy-receiver"
+#define CB_COMPAT_STR "smmu-proxy-cb"
+
+static void *msgq_hdl;
+
+struct smmu_proxy_buffer_cb_info {
+	bool mapped;
+	struct dma_buf_attachment *attachment;
+	struct sg_table *sg_table;
+};
+
+struct smmu_proxy_buffer_state {
+	bool locked;
+	struct smmu_proxy_buffer_cb_info cb_info[QTI_SMMU_PROXY_CB_IDS_LEN];
+	struct dma_buf *dmabuf;
+};
+
+static DEFINE_MUTEX(buffer_state_lock);
+static DEFINE_XARRAY(buffer_state_arr);
+
+static unsigned int cb_map_counts[QTI_SMMU_PROXY_CB_IDS_LEN] = { 0 };
+struct device *cb_devices[QTI_SMMU_PROXY_CB_IDS_LEN] = { 0 };
+
+struct task_struct *receiver_msgq_handler_thread;
+
+static int iommu_unmap_and_relinquish(u32 hdl)
+{
+	int cb_id, ret = 0;
+	struct smmu_proxy_buffer_state *buf_state;
+
+	mutex_lock(&buffer_state_lock);
+	buf_state = xa_load(&buffer_state_arr, hdl);
+	if (!buf_state) {
+		pr_err("%s: handle 0x%llx unknown to proxy driver!\n", __func__, hdl);
+		ret = -EINVAL;
+		goto out;
+	}
+
+	if (buf_state->locked) {
+		pr_err("%s: handle 0x%llx is locked!\n", __func__, hdl);
+		ret = -EINVAL;
+		goto out;
+	}
+
+	for (cb_id = 0; cb_id < QTI_SMMU_PROXY_CB_IDS_LEN; cb_id++) {
+		if (buf_state->cb_info[cb_id].mapped) {
+			dma_buf_unmap_attachment(buf_state->cb_info[cb_id].attachment,
+						 buf_state->cb_info[cb_id].sg_table,
+						 DMA_BIDIRECTIONAL);
+			dma_buf_detach(buf_state->dmabuf,
+				       buf_state->cb_info[cb_id].attachment);
+			buf_state->cb_info[cb_id].mapped = false;
+
+			/* If nothing left is mapped for this CB, unprogram its SMR */
+			cb_map_counts[cb_id]--;
+			if (!cb_map_counts[cb_id]) {
+				ret = qcom_iommu_sid_switch(cb_devices[cb_id], SID_RELEASE);
+				if (ret) {
+					pr_err("%s: Failed to unprogram SMR for cb_id %d rc: %d\n",
+					       __func__, cb_id, ret);
+					break;
+				}
+			}
+		}
+	}
+
+	dma_buf_put(buf_state->dmabuf);
+	flush_delayed_fput();
+
+	xa_erase(&buffer_state_arr, hdl);
+	kfree(buf_state);
+out:
+	mutex_unlock(&buffer_state_lock);
+
+	return ret;
+}
+
+static int process_unmap_request(struct smmu_proxy_unmap_req *req, size_t size)
+{
+	struct smmu_proxy_unmap_resp *resp;
+	int ret = 0;
+
+	resp = kzalloc(sizeof(*resp), GFP_KERNEL);
+	if (!resp) {
+		pr_err("%s: Failed to allocate memory for response\n", __func__);
+		return -ENOMEM;
+	}
+
+	ret = iommu_unmap_and_relinquish(req->hdl);
+
+	resp->hdr.msg_type = SMMU_PROXY_UNMAP_RESP;
+	resp->hdr.msg_size = sizeof(*resp);
+	resp->hdr.ret = ret;
+
+	ret = gh_msgq_send(msgq_hdl, resp, resp->hdr.msg_size, 0);
+	if (ret < 0)
+		pr_err("%s: failed to send response to mapping request rc: %d\n", __func__, ret);
+	else
+		pr_debug("%s: response to mapping request sent\n", __func__);
+
+	kfree(resp);
+
+	return ret;
+}
+
+static
+inline
+struct sg_table *retrieve_and_iommu_map(struct mem_buf_retrieve_kernel_arg *retrieve_arg,
+					u32 cb_id)
+{
+	int ret;
+	struct dma_buf *dmabuf;
+	bool new_buf = false;
+	struct smmu_proxy_buffer_state *buf_state;
+	struct dma_buf_attachment *attachment;
+	struct sg_table *table;
+
+	if (cb_id >= QTI_SMMU_PROXY_CB_IDS_LEN) {
+		pr_err("%s: CB ID %d too large\n", __func__, cb_id);
+		return ERR_PTR(-EINVAL);
+	}
+
+	if (!cb_devices[cb_id]) {
+		pr_err("%s: CB of ID %d not defined\n", __func__, cb_id);
+		return ERR_PTR(-EINVAL);
+	}
+
+	mutex_lock(&buffer_state_lock);
+	buf_state = xa_load(&buffer_state_arr, retrieve_arg->memparcel_hdl);
+	if (buf_state) {
+		if (buf_state->locked) {
+			pr_err("%s: handle 0x%llx is locked!\n", __func__,
+			       retrieve_arg->memparcel_hdl);
+			ret = -EINVAL;
+			goto unlock_err;
+		}
+		if (buf_state->cb_info[cb_id].mapped) {
+			table = buf_state->cb_info[cb_id].sg_table;
+			goto unlock;
+		}
+
+		dmabuf = buf_state->dmabuf;
+	} else {
+		new_buf = true;
+		dmabuf = mem_buf_retrieve(retrieve_arg);
+		if (IS_ERR(dmabuf)) {
+			ret = PTR_ERR(dmabuf);
+			pr_err("%s: Failed to retrieve DMA-BUF rc: %d\n", __func__, ret);
+			goto unlock_err;
+		}
+
+		buf_state = kzalloc(sizeof(*buf_state), GFP_KERNEL);
+		if (!buf_state) {
+			pr_err("%s: Unable to allocate memory for buf_state\n",
+			       __func__);
+			ret = -ENOMEM;
+			goto free_buf;
+		}
+
+		buf_state->dmabuf = dmabuf;
+	}
+
+	attachment = dma_buf_attach(dmabuf, cb_devices[cb_id]);
+	if (IS_ERR(attachment)) {
+		ret = PTR_ERR(attachment);
+		pr_err("%s: Failed to attach rc: %d\n", __func__, ret);
+		goto free_buf_state;
+	}
+
+	table = dma_buf_map_attachment(attachment, DMA_BIDIRECTIONAL);
+	if (IS_ERR(table)) {
+		ret = PTR_ERR(table);
+		pr_err("%s: Failed to map rc: %d\n", __func__, ret);
+		goto detach;
+	}
+
+	if (table->nents != 1) {
+		ret = -EINVAL;
+		pr_err("%s: Buffer not mapped as one segment!\n", __func__);
+		goto unmap;
+	}
+
+	buf_state->cb_info[cb_id].mapped = true;
+	buf_state->cb_info[cb_id].attachment = attachment;
+	buf_state->cb_info[cb_id].sg_table = table;
+
+	if (!cb_map_counts[cb_id]) {
+		ret = qcom_iommu_sid_switch(cb_devices[cb_id], SID_ACQUIRE);
+		if (ret) {
+			pr_err("%s: Failed to program SMRs for cb_id %d rc: %d\n", __func__,
+			       cb_id, ret);
+			goto unmap;
+		}
+	}
+	cb_map_counts[cb_id]++;
+
+	ret = xa_err(xa_store(&buffer_state_arr, retrieve_arg->memparcel_hdl, buf_state,
+		     GFP_KERNEL));
+	if (ret < 0) {
+		pr_err("%s: Failed to store new buffer in xarray rc: %d\n", __func__,
+		       ret);
+		goto dec_cb_map_count;
+	}
+
+unlock:
+	mutex_unlock(&buffer_state_lock);
+
+	return table;
+
+dec_cb_map_count:
+	cb_map_counts[cb_id]--;
+	if (!cb_map_counts[cb_id]) {
+		ret = qcom_iommu_sid_switch(cb_devices[cb_id], SID_RELEASE);
+		if (ret)
+			pr_err("%s: Failed to unprogram SMR for cb_id %d rc: %d\n",
+			       __func__, cb_id, ret);
+	}
+unmap:
+	dma_buf_unmap_attachment(attachment, table, DMA_BIDIRECTIONAL);
+detach:
+	dma_buf_detach(dmabuf, attachment);
+free_buf_state:
+	if (new_buf)
+		kfree(buf_state);
+free_buf:
+	if (new_buf)
+		dma_buf_put(dmabuf);
+unlock_err:
+	mutex_unlock(&buffer_state_lock);
+
+	return ERR_PTR(ret);
+}
+
+static int process_map_request(struct smmu_proxy_map_req *req, size_t size)
+{
+	struct smmu_proxy_map_resp *resp;
+	int ret = 0;
+	u32 n_acl_entries = req->acl_desc.n_acl_entries;
+	size_t map_req_len = offsetof(struct smmu_proxy_map_req,
+				      acl_desc.acl_entries[n_acl_entries]);
+	struct mem_buf_retrieve_kernel_arg retrieve_arg = {0};
+	int i;
+	struct sg_table *table;
+
+	/*
+	 * Last entry of smmu_proxy_map_req is an array of arbitrary length.
+	 * Validate that the number of entries fits within the buffer given
+	 * to us by the message queue.
+	 */
+	if (map_req_len > size) {
+		pr_err("%s: Reported size of smmu_proxy_map_request (%d bytes) greater than message size given by message queue (%d bytes)\n",
+		       __func__, map_req_len, size);
+		return -EINVAL;
+	}
+
+	resp = kzalloc(sizeof(*resp), GFP_KERNEL);
+	if (!resp) {
+		pr_err("%s: Failed to allocate memory for response\n", __func__);
+		return -ENOMEM;
+	}
+
+	retrieve_arg.vmids = kmalloc_array(n_acl_entries, sizeof(*retrieve_arg.vmids), GFP_KERNEL);
+	if (!retrieve_arg.vmids) {
+		ret = -ENOMEM;
+		goto free_resp;
+	}
+
+	retrieve_arg.perms = kmalloc_array(n_acl_entries, sizeof(*retrieve_arg.perms), GFP_KERNEL);
+	if (!retrieve_arg.perms) {
+		ret = -ENOMEM;
+		goto free_vmids;
+	}
+
+	retrieve_arg.memparcel_hdl = req->hdl;
+	retrieve_arg.sender_vmid = VMID_HLOS;
+	retrieve_arg.nr_acl_entries = n_acl_entries;
+
+	for (i = 0; i < n_acl_entries; i++) {
+		retrieve_arg.vmids[i] = req->acl_desc.acl_entries[i].vmid;
+		retrieve_arg.perms[i] = req->acl_desc.acl_entries[i].perms;
+	}
+
+	table = retrieve_and_iommu_map(&retrieve_arg, req->cb_id);
+	if (IS_ERR(table)) {
+		ret = PTR_ERR(table);
+		goto free_perms;
+	}
+
+	resp->hdr.msg_type = SMMU_PROXY_MAP_RESP;
+	resp->hdr.msg_size = sizeof(*resp);
+	resp->hdr.ret = ret;
+	resp->iova = sg_dma_address(table->sgl);
+	resp->mapping_len = sg_dma_len(table->sgl);
+
+	ret = gh_msgq_send(msgq_hdl, resp, resp->hdr.msg_size, 0);
+	if (ret < 0) {
+		pr_err("%s: failed to send response to mapping request rc: %d\n", __func__, ret);
+		iommu_unmap_and_relinquish(req->hdl);
+	} else {
+		pr_debug("%s: response to mapping request sent\n", __func__);
+	}
+
+free_perms:
+	kfree(retrieve_arg.perms);
+free_vmids:
+	kfree(retrieve_arg.vmids);
+free_resp:
+	kfree(resp);
+
+	return ret;
+}
+
+static void smmu_proxy_process_msg(void *buf, size_t size)
+{
+	struct smmu_proxy_msg_hdr *msg_hdr = buf;
+	struct smmu_proxy_resp_hdr *resp;
+	int ret = -EINVAL;
+
+	pr_err("%s: smmu-proxy message received\n", __func__);
+	if (size < sizeof(*msg_hdr) || msg_hdr->msg_size != size) {
+		pr_err("%s: message received is not of a proper size: 0x%lx, 0x:%lx\n",
+		       __func__, size, msg_hdr->msg_size);
+		goto handle_err;
+	}
+
+	switch (msg_hdr->msg_type) {
+	case SMMU_PROXY_MAP:
+		ret = process_map_request(buf, size);
+		break;
+	case SMMU_PROXY_UNMAP:
+		ret = process_unmap_request(buf, size);
+		break;
+	default:
+		pr_err("%s: received message of unknown type: %d\n", __func__,
+		       msg_hdr->msg_type);
+	}
+
+	if (!ret)
+		return;
+
+handle_err:
+	resp = kzalloc(sizeof(resp), GFP_KERNEL);
+	if (!resp) {
+		pr_err("%s: Failed to allocate memory for response\n", __func__);
+		return;
+	}
+
+	resp->msg_type = SMMU_PROXY_ERR_RESP;
+	resp->msg_size = sizeof(resp);
+	resp->ret = ret;
+
+	ret = gh_msgq_send(msgq_hdl, resp, resp->msg_size, 0);
+	if (ret < 0)
+		pr_err("%s: failed to send error response rc: %d\n", __func__, ret);
+	else
+		pr_debug("%s: response to mapping request sent\n", __func__);
+
+	kfree(resp);
+
+}
+
+static int receiver_msgq_handler(void *msgq_hdl)
+{
+	void *buf;
+	size_t size;
+	int ret;
+
+	buf = kzalloc(GH_MSGQ_MAX_MSG_SIZE_BYTES, GFP_KERNEL);
+	if (!buf)
+		return -ENOMEM;
+
+	while (!kthread_should_stop()) {
+		ret = gh_msgq_recv(msgq_hdl, buf, GH_MSGQ_MAX_MSG_SIZE_BYTES, &size, 0);
+		if (ret < 0) {
+			pr_err_ratelimited("%s failed to receive message rc: %d\n", __func__, ret);
+		} else {
+			smmu_proxy_process_msg(buf, size);
+		}
+	}
+
+	kfree(buf);
+
+	return 0;
+}
+
+static int smmu_proxy_ac_lock_toggle(int dma_buf_fd, bool lock)
+{
+	int ret = 0;
+	struct smmu_proxy_buffer_state *buf_state;
+	struct dma_buf *dmabuf;
+	u32 handle;
+
+	dmabuf = dma_buf_get(dma_buf_fd);
+	if (IS_ERR(dmabuf)) {
+		pr_err("%s: unable to get dma-buf from FD %d, rc: %d", __func__,
+		       dma_buf_fd, PTR_ERR(dmabuf));
+		return PTR_ERR(dmabuf);
+	}
+
+	ret = mem_buf_dma_buf_get_memparcel_hdl(dmabuf, &handle);
+	if (ret) {
+		pr_err("%s: Failed to get memparcel handle rc: %d\n", __func__, ret);
+		goto free_buf;
+	}
+
+	mutex_lock(&buffer_state_lock);
+	buf_state = xa_load(&buffer_state_arr, handle);
+	if (!buf_state) {
+		pr_err("%s: handle 0x%llx unknown to proxy driver!\n", __func__, handle);
+		ret = -EINVAL;
+		goto out;
+	}
+
+	if (buf_state->locked == lock) {
+		pr_err("%s: handle 0x%llx already %s!\n", __func__, handle,
+		       lock ? "locked" : "unlocked");
+		ret = -EINVAL;
+		goto out;
+	}
+
+	buf_state->locked = lock;
+out:
+	mutex_unlock(&buffer_state_lock);
+free_buf:
+	dma_buf_put(dmabuf);
+
+	return ret;
+}
+
+/*
+ * Iterate over all buffers mapped to context bank @context_bank_id, and zero
+ * out the buffers. If there is a single error for any buffer, we bail out with
+ * an error and disregard the rest of the buffers mapped to @context_bank_id.
+ */
+int smmu_proxy_clear_all_buffers(void __user *context_bank_id_array,
+				 __u32 num_cb_ids)
+{
+	unsigned long handle;
+	struct smmu_proxy_buffer_state *buf_state;
+	struct iosys_map vmap_struct = {0};
+	__u32 cb_ids[QTI_SMMU_PROXY_CB_IDS_LEN];
+	int i, ret = 0;
+	bool found_mapped_cb;
+
+	/* Checking this allows us to keep cb_id_arr fixed in length */
+	if (num_cb_ids > QTI_SMMU_PROXY_CB_IDS_LEN) {
+		pr_err("%s: Invalid number of CB IDs: %u\n", __func__, num_cb_ids);
+		return -EINVAL;
+	}
+
+	ret = copy_struct_from_user(&cb_ids, sizeof(cb_ids), context_bank_id_array,
+				    sizeof(cb_ids));
+	if (ret) {
+		pr_err("%s: Failed to get CB IDs from user space rc %d\n", __func__, ret);
+		return ret;
+	}
+
+	for (i = 0; i < num_cb_ids; i++) {
+		if (cb_ids[i] >= QTI_SMMU_PROXY_CB_IDS_LEN) {
+			pr_err("%s: Invalid CB ID of %u at pos %d\n", __func__, cb_ids[i], i);
+			return -EINVAL;
+		}
+	}
+
+	mutex_lock(&buffer_state_lock);
+	xa_for_each(&buffer_state_arr, handle, buf_state) {
+		found_mapped_cb = false;
+		for (i = 0; i < num_cb_ids; i++) {
+			if (buf_state->cb_info[cb_ids[i]].mapped) {
+				found_mapped_cb = true;
+				break;
+			}
+		}
+		if (!found_mapped_cb)
+			continue;
+
+		ret = dma_buf_vmap(buf_state->dmabuf, &vmap_struct);
+		if (ret) {
+			pr_err("%s: dma_buf_vmap() failed with %d\n", __func__, ret);
+			goto unlock;
+		}
+
+		/* Use DMA_TO_DEVICE since we are not reading anything */
+		ret = dma_buf_begin_cpu_access(buf_state->dmabuf, DMA_TO_DEVICE);
+		if (ret) {
+			pr_err("%s: dma_buf_begin_cpu_access() failed with %d\n", __func__, ret);
+			goto unmap;
+		}
+
+		memset(vmap_struct.vaddr, 0, buf_state->dmabuf->size);
+		ret = dma_buf_end_cpu_access(buf_state->dmabuf, DMA_TO_DEVICE);
+		if (ret)
+			pr_err("%s: dma_buf_end_cpu_access() failed with %d\n", __func__, ret);
+unmap:
+		dma_buf_vunmap(buf_state->dmabuf, &vmap_struct);
+		if (ret)
+			break;
+	}
+
+unlock:
+	mutex_unlock(&buffer_state_lock);
+	return ret;
+}
+
+static int smmu_proxy_get_dma_buf(struct smmu_proxy_get_dma_buf_ctl *get_dma_buf_ctl)
+{
+	struct smmu_proxy_buffer_state *buf_state;
+	int fd, ret = 0;
+
+	mutex_lock(&buffer_state_lock);
+	buf_state = xa_load(&buffer_state_arr, get_dma_buf_ctl->memparcel_hdl);
+	if (!buf_state) {
+		pr_err("%s: handle 0x%llx unknown to proxy driver!\n", __func__,
+		       get_dma_buf_ctl->memparcel_hdl);
+		ret = -EINVAL;
+		goto out;
+	}
+
+	get_dma_buf(buf_state->dmabuf);
+	fd = dma_buf_fd(buf_state->dmabuf, O_RDWR | O_CLOEXEC);
+	if (fd < 0) {
+		ret = fd;
+		pr_err("%s: Failed to install FD for dma-buf rc: %d\n", __func__,
+		       ret);
+		dma_buf_put(buf_state->dmabuf);
+	} else {
+		get_dma_buf_ctl->dma_buf_fd = fd;
+	}
+out:
+	mutex_unlock(&buffer_state_lock);
+
+	return ret;
+}
+
+static long smmu_proxy_dev_ioctl(struct file *filp, unsigned int cmd,
+			      unsigned long arg)
+{
+	unsigned int dir = _IOC_DIR(cmd);
+	union smmu_proxy_ioctl_arg ioctl_arg;
+	int ret;
+
+	if (_IOC_SIZE(cmd) > sizeof(ioctl_arg))
+		return -EINVAL;
+
+	if (copy_from_user(&ioctl_arg, (void __user *)arg, _IOC_SIZE(cmd)))
+		return -EFAULT;
+
+	if (!(dir & _IOC_WRITE))
+		memset(&ioctl_arg, 0, sizeof(ioctl_arg));
+
+	switch (cmd) {
+	case QTI_SMMU_PROXY_AC_LOCK_BUFFER:
+	{
+		struct smmu_proxy_acl_ctl *acl_ctl =
+			&ioctl_arg.acl_ctl;
+
+		ret = smmu_proxy_ac_lock_toggle(acl_ctl->dma_buf_fd, true);
+		if (ret)
+			return ret;
+
+		break;
+	}
+	case QTI_SMMU_PROXY_AC_UNLOCK_BUFFER:
+	{
+		struct smmu_proxy_acl_ctl *acl_ctl =
+			&ioctl_arg.acl_ctl;
+
+		ret = smmu_proxy_ac_lock_toggle(acl_ctl->dma_buf_fd, false);
+		if (ret)
+			return ret;
+
+		break;
+	}
+	case QTI_SMMU_PROXY_WIPE_BUFFERS:
+	{
+		struct smmu_proxy_wipe_buf_ctl *wipe_buf_ctl =
+			&ioctl_arg.wipe_buf_ctl;
+
+		ret = smmu_proxy_clear_all_buffers((void *) wipe_buf_ctl->context_bank_id_array,
+						   wipe_buf_ctl->num_cb_ids);
+		break;
+	}
+	case QTI_SMMU_PROXY_GET_DMA_BUF:
+	{
+		ret = smmu_proxy_get_dma_buf(&ioctl_arg.get_dma_buf_ctl);
+		break;
+	}
+
+	default:
+		return -ENOTTY;
+	}
+
+	if (dir & _IOC_READ) {
+		if (copy_to_user((void __user *)arg, &ioctl_arg,
+				 _IOC_SIZE(cmd)))
+			return -EFAULT;
+	}
+
+	return 0;
+}
+
+static const struct file_operations smmu_proxy_dev_fops = {
+	.unlocked_ioctl = smmu_proxy_dev_ioctl,
+	.compat_ioctl = compat_ptr_ioctl,
+};
+
+static int receiver_probe_handler(struct device *dev)
+{
+	int ret = 0;
+
+	msgq_hdl = gh_msgq_register(GH_MSGQ_LABEL_SMMU_PROXY);
+	if (IS_ERR(msgq_hdl)) {
+		ret = PTR_ERR(msgq_hdl);
+		dev_err(dev, "Queue registration failed: %d!\n", PTR_ERR(msgq_hdl));
+		return ret;
+	}
+
+	receiver_msgq_handler_thread = kthread_run(receiver_msgq_handler, msgq_hdl,
+						   "smmu_proxy_msgq_handler");
+	if (IS_ERR(receiver_msgq_handler_thread)) {
+		ret = PTR_ERR(receiver_msgq_handler_thread);
+		dev_err(dev, "Failed to launch receiver_msgq_handler thread: %d\n",
+			PTR_ERR(receiver_msgq_handler_thread));
+		goto free_msgq;
+	}
+
+	ret = smmu_proxy_create_dev(&smmu_proxy_dev_fops);
+	if (ret) {
+		pr_err("Failed to create character device with error %d\n", ret);
+		goto free_kthread;
+	}
+
+	return 0;
+free_kthread:
+	kthread_stop(receiver_msgq_handler_thread);
+free_msgq:
+	gh_msgq_unregister(msgq_hdl);
+	return ret;
+}
+
+static int cb_probe_handler(struct device *dev)
+{
+	int ret;
+	unsigned int context_bank_id;
+
+	ret = of_property_read_u32(dev->of_node, "qti,cb-id", &context_bank_id);
+	if (ret) {
+		dev_err(dev, "Failed to read qti,cb-id property for device\n");
+		return -EINVAL;
+	}
+
+	if (context_bank_id >= QTI_SMMU_PROXY_CB_IDS_LEN) {
+		dev_err(dev, "Invalid CB ID: %u\n", context_bank_id);
+		return -EINVAL;
+	}
+
+	if (cb_devices[context_bank_id]) {
+		dev_err(dev, "Context bank %u is already populated\n", context_bank_id);
+		return -EINVAL;
+	}
+
+	ret = dma_set_max_seg_size(dev, DMA_BIT_MASK(32));
+	if (ret) {
+		dev_err(dev, "Failed to set segment size\n");
+		return ret;
+	}
+
+	ret = dma_coerce_mask_and_coherent(dev, DMA_BIT_MASK(64));
+	if (ret) {
+		dev_err(dev, "Failed to set DMA-MASK\n");
+		return ret;
+	}
+
+	cb_devices[context_bank_id] = dev;
+
+	return 0;
+}
+
+static int smmu_proxy_probe(struct platform_device *pdev)
+{
+	struct device *dev = &pdev->dev;
+
+	if (of_device_is_compatible(dev->of_node, CB_COMPAT_STR)) {
+		return cb_probe_handler(dev);
+	} else if (of_device_is_compatible(dev->of_node, RECEIVER_COMPAT_STR)) {
+		return  receiver_probe_handler(dev);
+	} else {
+		return -EINVAL;
+	}
+}
+
+static const struct of_device_id smmu_proxy_match_table[] = {
+	{.compatible = RECEIVER_COMPAT_STR},
+	{.compatible = CB_COMPAT_STR},
+	{},
+};
+
+static struct platform_driver smmu_proxy_driver = {
+	.probe = smmu_proxy_probe,
+	.driver = {
+		.name = "qti-smmu-proxy",
+		.of_match_table = smmu_proxy_match_table,
+	},
+};
+
+int __init init_smmu_proxy_driver(void)
+{
+	int ret;
+	struct csf_version csf_version;
+
+	ret = smmu_proxy_get_csf_version(&csf_version);
+	if (ret) {
+		pr_err("%s: Unable to get CSF version\n", __func__);
+		return ret;
+	}
+
+	if (csf_version.arch_ver == 2 && csf_version.max_ver == 0) {
+		pr_err("%s: CSF 2.5 not in use, not loading module\n", __func__);
+		return -EINVAL;
+	}
+
+	return platform_driver_register(&smmu_proxy_driver);
+}
+module_init(init_smmu_proxy_driver);
+
+MODULE_IMPORT_NS(DMA_BUF);
+MODULE_LICENSE("GPL v2");

smmu-proxy/qti-smmu-proxy.c

@@ -1,154 +0,0 @@
-// SPDX-License-Identifier: GPL-2.0-only
-/*
- * Copyright (c) 2023 Qualcomm Innovation Center, Inc. All rights reserved.
- */
-
-#include <linux/module.h>
-#include <linux/of_platform.h>
-#include <linux/platform_device.h>
-#include <linux/cdev.h>
-
-#include "linux/qti-smmu-proxy.h"
-
-#define SMMU_PROXY_MAX_DEVS 1
-static dev_t smmu_proxy_dev_no;
-static struct class *smmu_proxy_class;
-static struct cdev smmu_proxy_char_dev;
-
-union smmu_proxy_ioctl_arg {
-	struct csf_version csf_version;
-};
-
-int smmu_proxy_get_csf_version(struct csf_version *csf_version)
-{
-	csf_version->arch_ver = 2;
-	csf_version->max_ver = 0;
-	csf_version->min_ver = 0;
-
-	return 0;
-}
-EXPORT_SYMBOL(smmu_proxy_get_csf_version);
-
-static long smmu_proxy_dev_ioctl(struct file *filp, unsigned int cmd,
-			      unsigned long arg)
-{
-	unsigned int dir = _IOC_DIR(cmd);
-	union smmu_proxy_ioctl_arg ioctl_arg;
-
-	if (_IOC_SIZE(cmd) > sizeof(ioctl_arg))
-		return -EINVAL;
-
-	if (copy_from_user(&ioctl_arg, (void __user *)arg, _IOC_SIZE(cmd)))
-		return -EFAULT;
-
-	if (!(dir & _IOC_WRITE))
-		memset(&ioctl_arg, 0, sizeof(ioctl_arg));
-
-	switch (cmd) {
-	case QTI_SMMU_PROXY_GET_VERSION_IOCTL:
-	{
-		struct csf_version *csf_version =
-			&ioctl_arg.csf_version;
-		int ret;
-
-		ret = smmu_proxy_get_csf_version(csf_version);
-		if(ret)
-			return ret;
-
-		break;
-	}
-	default:
-		return -ENOTTY;
-	}
-
-	if (dir & _IOC_READ) {
-		if (copy_to_user((void __user *)arg, &ioctl_arg,
-				 _IOC_SIZE(cmd)))
-			return -EFAULT;
-	}
-
-	return 0;
-}
-
-
-static const struct file_operations smmu_proxy_dev_fops = {
-	.unlocked_ioctl = smmu_proxy_dev_ioctl,
-	.compat_ioctl = compat_ptr_ioctl,
-};
-
-static int smmu_proxy_create_dev(struct platform_device *pdev)
-{
-	int ret;
-	struct device *class_dev;
-
-	cdev_init(&smmu_proxy_char_dev, &smmu_proxy_dev_fops);
-	ret = cdev_add(&smmu_proxy_char_dev, smmu_proxy_dev_no,
-		       SMMU_PROXY_MAX_DEVS);
-	if (ret < 0)
-		return ret;
-
-	class_dev = device_create(smmu_proxy_class, NULL, smmu_proxy_dev_no, NULL,
-				  "qti-smmu-proxy");
-	if (IS_ERR(class_dev)) {
-		ret = PTR_ERR(class_dev);
-		goto err_dev_create;
-	}
-
-	return 0;
-
-err_dev_create:
-	cdev_del(&smmu_proxy_char_dev);
-	return ret;
-}
-
-static int smmu_proxy_probe(struct platform_device *pdev)
-{
-	return smmu_proxy_create_dev(pdev);
-}
-
-static const struct of_device_id smmu_proxy_match_table[] = {
-	{.compatible = "smmu-proxy-sender"},
-	{},
-};
-
-static struct platform_driver smmu_proxy_driver = {
-	.probe = smmu_proxy_probe,
-	.driver = {
-		.name = "qti-smmu-proxy",
-		.of_match_table = smmu_proxy_match_table,
-	},
-};
-
-static int __init init_smmu_proxy_driver(void)
-{
-	int ret;
-
-	ret = alloc_chrdev_region(&smmu_proxy_dev_no, 0, SMMU_PROXY_MAX_DEVS,
-				  "qti-smmu-proxy");
-	if (ret < 0)
-		goto err_chrdev_region;
-
-	smmu_proxy_class = class_create(THIS_MODULE, "qti-smmu-proxy");
-	if (IS_ERR(smmu_proxy_class)) {
-		ret = PTR_ERR(smmu_proxy_class);
-		goto err_class_create;
-	}
-
-	ret = platform_driver_register(&smmu_proxy_driver);
-	if (ret < 0)
-		goto err_platform_drvr_register;
-
-	return 0;
-
-err_platform_drvr_register:
-	class_destroy(smmu_proxy_class);
-err_class_create:
-	unregister_chrdev_region(smmu_proxy_dev_no, SMMU_PROXY_MAX_DEVS);
-err_chrdev_region:
-	return ret;
-
-}
-module_init(init_smmu_proxy_driver);
-
-MODULE_IMPORT_NS(DMA_BUF);
-MODULE_LICENSE("GPL v2");

smmu-proxy/uapi/linux/qti-smmu-proxy.h

@@ -34,7 +34,8 @@ struct smmu_proxy_acl_ctl {
 					     struct smmu_proxy_acl_ctl)
 
 struct smmu_proxy_wipe_buf_ctl {
-	__u32 smmu_context_bank_id;
+	__u64 context_bank_id_array;
+	__u32 num_cb_ids;
 	__u32 padding;
 };