Abstract: Technologies for cryptographic separation of MMIO operations with an accelerator device include a computing device having a processor and an accelerator. The processor establishes a trusted execution environment. The accelerator determines, based on a target memory address, a first memory address range associated with the memory-mapped I/O transaction, generates a second authentication tag using a first cryptographic key from a set of cryptographic keys, wherein the first key is uniquely associated with the first memory address range. An accelerator validator determines whether the first authentication tag matches the second authentication tag, and a memory mapper commits the memory-mapped I/O transaction in response to a determination that the first authentication tag matches the second authentication tag. Other embodiments are described and claimed.

Abstract: Embodiments are directed to trusted local memory management in a virtualized GPU. An embodiment of an apparatus includes one or more processors including a trusted execution environment (TEE); a GPU including a trusted agent; and a memory, the memory including GPU local memory, the trusted agent to ensure proper allocation/deallocation of the local memory and verify translations between graphics physical addresses (PAs) and PAs for the apparatus, wherein the local memory is partitioned into protection regions including a protected region and an unprotected region, and wherein the protected region to store a memory permission table maintained by the trusted agent, the memory permission table to include any virtual function assigned to a trusted domain, a per process graphics translation table to translate between graphics virtual address (VA) to graphics guest PA (GPA), and a local memory translation table to translate between graphics GPAs and PAs for the local memory.

Abstract: Technologies for cryptographic separation of MMIO operations with an accelerator device include a computing device having a processor and an accelerator. The processor establishes a trusted execution environment. The accelerator determines, based on a target memory address, a first memory address range associated with the memory-mapped I/O transaction, generates a second authentication tag using a first cryptographic key from a set of cryptographic keys, wherein the first key is uniquely associated with the first memory address range. An accelerator validator determines whether the first authentication tag matches the second authentication tag, and a memory mapper commits the memory-mapped I/O transaction in response to a determination that the first authentication tag matches the second authentication tag. Other embodiments are described and claimed.

Abstract: A method comprises initializing, by an accelerator device of the computing device, an authentication tag in response to an initialization command from a trusted execution environment of the computing device, initiating a transfer, by the accelerator device, of data between a host memory and an accelerator device memory in response to a descriptor from the trusted execution environment, wherein the descriptor comprises a target memory address and is indicative of a transfer direction, comparing, in a memory range selection engine comprising at least one comparator to compare the target memory address with a plurality of address ranges and select a cryptographic key from the plurality of plurality of address range registers based on the target memory address, performing, by the accelerator device, a cryptographic operation with the data in response to transferring the data, updating, by the accelerator device, the authentication tag in response to transferring the data, and finalizing, by the accelerator device

Abstract: Embodiments are directed to trusted local memory management in a virtualized GPU. An embodiment of an apparatus includes one or more processors including a trusted execution environment (TEE); a GPU including a trusted agent; and a memory, the memory including GPU local memory, the trusted agent to ensure proper allocation/deallocation of the local memory and verify translations between graphics physical addresses (PAs) and PAs for the apparatus, wherein the local memory is partitioned into protection regions including a protected region and an unprotected region, and wherein the protected region to store a memory permission table maintained by the trusted agent, the memory permission table to include any virtual function assigned to a trusted domain, a per process graphics translation table to translate between graphics virtual address (VA) to graphics guest PA (GPA), and a local memory translation table to translate between graphics GPAs and PAs for the local memory.

Abstract: Embodiments are directed to trusted local memory management in a virtualized GPU. An embodiment of an apparatus includes one or more processors including a trusted execution environment (TEE); a GPU including a trusted agent; and a memory, the memory including GPU local memory, the trusted agent to ensure proper allocation/deallocation of the local memory and verify translations between graphics physical addresses (PAs) and PAs for the apparatus, wherein the local memory is partitioned into protection regions including a protected region and an unprotected region, and wherein the protected region to store a memory permission table maintained by the trusted agent, the memory permission table to include any virtual function assigned to a trusted domain, a per process graphics translation table to translate between graphics virtual address (VA) to graphics guest PA (GPA), and a local memory translation table to translate between graphics GPAs and PAs for the local memory.

Abstract: Technologies for secure I/O include a compute device, which further includes a processor, a memory, a trusted execution environment (TEE), one or more input/output (I/O) devices, and an I/O subsystem. The I/O subsystem includes a device memory access table (DMAT) programmed by the TEE to establish bindings between the TEE and one or more I/O devices that the TEE trusts and a memory ownership table (MOT) programmed by the TEE when a memory page is allocated to the TEE.

Abstract: Technologies for secure I/O data transfer includes a compute device, which includes a processor to execute a trusted application, an input/output (I/O) device, and an I/O subsystem. The I/O subsystem is configured to establish a secured channel between the I/O subsystem and a trusted application running on the compute device, and receive, in response to an establishment of the secured channel, I/O data from the I/O device via an unsecured channel. The I/O subsystem is further configured to encrypt, in response to a receipt of the I/O data, the I/O data using a security key associated with the trusted application that is to process the I/O data and transmit the encrypted I/O data to the trusted application via the secured channel, wherein the secured channel has a data transfer rate that is higher than a data transfer rate of the unsecured channel between the I/O device and the I/O subsystem.

Abstract: Embodiments are directed to trusted local memory management in a virtualized GPU. An embodiment of an apparatus includes one or more processors including a trusted execution environment (TEE); a GPU including a trusted agent; and a memory, the memory including GPU local memory, the trusted agent to ensure proper allocation/deallocation of the local memory and verify translations between graphics physical addresses (PAs) and PAs for the apparatus, wherein the local memory is partitioned into protection regions including a protected region and an unprotected region, and wherein the protected region to store a memory permission table maintained by the trusted agent, the memory permission table to include any virtual function assigned to a trusted domain, a per process graphics translation table to translate between graphics virtual address (VA) to graphics guest PA (GPA), and a local memory translation table to translate between graphics GPAs and PAs for the local memory.

Abstract: Technologies for secure I/O data transfer includes a compute device, which includes a processor to execute a trusted application, an input/output (I/O) device, and an I/O subsystem. The I/O subsystem is configured to establish a secured channel between the I/O subsystem and a trusted application running on the compute device, and receive, in response to an establishment of the secured channel, I/O data from the I/O device via an unsecured channel. The I/O subsystem is further configured to encrypt, in response to a receipt of the I/O data, the I/O data using a security key associated with the trusted application that is to process the I/O data and transmit the encrypted I/O data to the trusted application via the secured channel, wherein the secured channel has a data transfer rate that is higher than a data transfer rate of the unsecured channel between the I/O device and the I/O subsystem.

Abstract: A method comprises initializing, by an accelerator device of the computing device, an authentication tag in response to an initialization command from a trusted execution environment of the computing device, initiating a transfer, by the accelerator device, of data between a host memory and an accelerator device memory in response to a descriptor from the trusted execution environment, wherein the descriptor comprises a target memory address and is indicative of a transfer direction, comparing, in a memory range selection engine comprising at least one comparator to compare the target memory address with a plurality of address ranges and select a cryptographic key from the plurality of plurality of address range registers based on the target memory address, performing, by the accelerator device, a cryptographic operation with the data in response to transferring the data, updating, by the accelerator device, the authentication tag in response to transferring the data, and finalizing, by the accelerator device

Abstract: Technologies for cryptographic separation of MMIO operations with an accelerator device include a computing device having a processor and an accelerator. The processor establishes a trusted execution environment. The accelerator determines, based on a target memory address, a first memory address range associated with the memory-mapped I/O transaction, generates a second authentication tag using a first cryptographic key from a set of cryptographic keys, wherein the first key is uniquely associated with the first memory address range. An accelerator validator determines whether the first authentication tag matches the second authentication tag, and a memory mapper commits the memory-mapped I/O transaction in response to a determination that the first authentication tag matches the second authentication tag. Other embodiments are described and claimed.

Abstract: A method comprises initializing, by an accelerator device of the computing device, an authentication tag in response to an initialization command from a trusted execution environment of the computing device, initiating a transfer, by the accelerator device, of data between a host memory and an accelerator device memory in response to a descriptor from the trusted execution environment, wherein the descriptor comprises a target memory address and is indicative of a transfer direction, comparing, in a memory range selection engine comprising at least one comparator to compare the target memory address with a plurality of address ranges and select a cryptographic key from the plurality of plurality of address range registers based on the target memory address, performing, by the accelerator device, a cryptographic operation with the data in response to transferring the data, updating, by the accelerator device, the authentication tag in response to transferring the data, and finalizing, by the accelerator device

Abstract: Technologies for platform-targeted machine learning include a computing device to generate a machine learning algorithm model indicative of a plurality of classes between which a user input is to be classified and translate the machine learning algorithm model into hardware code for execution on the target platform. Example instructions cause a processor to obtain dataset features indicative of a plurality of characteristics of an input dataset, rank, using multiple ranking algorithms, the dataset features, identify feature subsets for respective ones of the ranked dataset features, predict performance metrics based on the feature subsets, and select a final subset based on the predicted performance metrics.

Abstract: Technologies for cryptographic separation of MMIO operations with an accelerator device include a computing device having a processor and an accelerator. The processor establishes a trusted execution environment. The accelerator determines, based on a target memory address, a first memory address range associated with the memory-mapped I/O transaction, generates a second authentication tag using a first cryptographic key from a set of cryptographic keys, wherein the first key is uniquely associated with the first memory address range. An accelerator validator determines whether the first authentication tag matches the second authentication tag, and a memory mapper commits the memory-mapped I/O transaction in response to a determination that the first authentication tag matches the second authentication tag. Other embodiments are described and claimed.

Abstract: Technologies for secure data transfer include a computing device having a processor, an accelerator, and a security engine, such as a direct memory access (DMA) engine or a memory-mapped I/O (MMIO) engine. The computing device initializes the security engine with an initialization vector and a secret key. During initialization, the security engine pre-fills block cipher pipelines and pre-computes hash subkeys. After initialization, the processor initiates a data transfer, such as a DMA transaction or an MMIO request, between the processor and the accelerator. The security engine performs an authenticated cryptographic operation for the data transfer operation. The authenticated cryptographic operation may be AES-GCM authenticated encryption or authenticated decryption. The security engine may perform encryption or decryption using multiple block cipher pipelines. The security engine may calculate an authentication tag using multiple Galois field multipliers. Other embodiments are described and claimed.

Abstract: Technologies for platform-targeted machine learning include a computing device to generate a machine learning algorithm model indicative of a plurality of classes between which a user input is to be classified and translate the machine learning algorithm model into hardware code for execution on the target platform. The user input is to be classified as being associated with a particular class based on an application of one or more features to the user input, and each of the one or more features has an associated implementation cost indicative of a cost to perform on a target platform on which the corresponding feature is to be applied to the user input.

Abstract: Technologies for secure data transfer include a computing device having a processor, an accelerator, and a security engine, such as a direct memory access (DMA) engine or a memory-mapped I/O (MMIO) engine. The computing device initializes the security engine with an initialization vector and a secret key. During initialization, the security engine pre-fills block cipher pipelines and pre-computes hash subkeys. After initialization, the processor initiates a data transfer, such as a DMA transaction or an MMIO request, between the processor and the accelerator. The security engine performs an authenticated cryptographic operation for the data transfer operation. The authenticated cryptographic operation may be AES-GCM authenticated encryption or authenticated decryption. The security engine may perform encryption or decryption using multiple block cipher pipelines. The security engine may calculate an authentication tag using multiple Galois field multipliers. Other embodiments are described and claimed.

Abstract: Technologies for secure I/O include a compute device, which further includes a processor, a memory, a trusted execution environment (TEE), one or more input/output (I/O) devices, and an I/O subsystem. The I/O subsystem includes a device memory access table (DMAT) programmed by the TEE to establish bindings between the TEE and one or more I/O devices that the TEE trusts and a memory ownership table (MOT) programmed by the TEE when a memory page is allocated to the TEE.

Abstract: Technologies for secure I/O include a compute device, which further includes a processor, a memory, a trusted execution environment (TEE), one or more input/output (I/O) devices, and an I/O subsystem. The I/O subsystem includes a device memory access table (DMAT) programmed by the TEE to establish bindings between the TEE and one or more I/O devices that the TEE trusts and a memory ownership table (MOT) programmed by the TEE when a memory page is allocated to the TEE.