Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Add support for a Hyper-V based vGPU implementation that exposes the DirectX API to Linux userspace. - Register a new device type with vmbus. - Switch to the VM bus protocol version 40 to allow async messages. 1. PCI driver registration is added. A PCI device is created for each virtual GPU (vGPU) device, projected by the host. The device vendor is PCI_VENDOR_ID_MICROSOFT and device id is PCI_DEVICE_ID_VIRTUAL_RENDER. dxg_pci_probe_device handles arrival of such devices and it creates dxgadapter objects. The PCI config space of the vGPU device has luid of the corresponding per GPU VM bus channel. This is how the adapters are linked to VM bus channels. 2. dxgadapter initialization is changed to wait until VM bus channels are created. A dxgadapter object represents a virtual GPU, projected to the VM by the host. This object can start functioning only when the global VM bus channel and the corresponding per vGPU VM bus channel are initialized in the guest. Notifications about arrival of vGPU PCI device and VM bus channels can happen in any order. Therefore, the initial dxgadapter object state is DXGADAPTER_STATE_WAITING_VMBUS. A list of VM bus channels and a list of dxgadapter objects are created. When dxgkrnl is notified about a VM bus channel arrival, if tries to start all adapters, which are not started yet. 2. VM bus interface version is exchanged by reading/writing the PCI config space of the vGPU device. 3. Implement the latest VM bus protocol version. The protocol version 40 added a new header to VM bus messages. All VM bus messages need to be changed to accomodate the new format. The new protocol allows asynchronouse messages to improve performance. When asynchronous messages are enabled, all messages need to be sent via the global VM bus channel to preserve correct order on the host. - Read the host virtual GPU (vGPU) luid from the PCI config space. A virtual GPU object in the host is created for every virtual GPU in the guest. It is identified by a luid. When asynchronous VM bus messages are enabled, all messages are sent via the global VM bus channel instead of a vGPU VM bus channel. Therefore, each VM bus vGPU message needs to have a host vGPU luid to which the message should be delivered.. - Remove the ISERROR macto. - Add comments to struct d3dkmthandle and struct ntstatus. - Removed the TRACE_DEBUG macro - Removed the dxgmem_alloc layer. - Remove dxgmutex and call Linux mutex APIs directly - Add CONFIG_DXGKRNL_DEBUG to control adding more debug information to the driver - Remove lock order tracking from released code - Add more comments for NTSTATUS and D3DKMTHANDLE - Remove unused NTSTATUS definitions - Replace refcount_t by struct kref - Added more comments - Fix comments in Makefile - Fixed the issue where incorrect size is used to copy standard allocation private driver data to the allocation. - Fixed re-initialization of the global VM bus channel. - Fixed false positive in detecting memory leak for DXGMEM_VMBUS. - Return correct total private driver size for a shared resource to the caller. - Remove the assume_not_in_use flag when destroying allocation. Using the flag could lead to a GPU page fault. - Fix setting the paravirtualized flag from query_adapter_info - Call vmbus_submit_packet in a loop with delay when there is no space in the ring buffer. - Optimize the dxgk_get_device_state implementation. The dxgdevice execution state is cached in the guest to avoid sending a messgage to the host for every call.. The DXGK_VMBCOMMAND_SETGUESTDATA VM bus message is sent from the host when the device state is changed. The next call to dxgk_get_device_state will query the state from the host. - Fixed the issue where the hardware process fence object handle is not assigned. As the result, D3DKMTWaitForSynchronizationObjectFromGpu with this fence handle will fail. - Removed timeout when waiting for internal CPU event in dxgk_wait_sync_object_cpu. By design, the WDDM API does not have a timeout. - Fixed validation when creating a standard allocation. - Coding style fixes. - Fixed incorrect reference to system memory pointer when creating an allocations - Increased the limit on number of allocations in MakeResident and other APIs - Remove support for sharing of DXG objects by using global handles. - Using the global handles is not secure. Sharing should be done using FDs (NT security sharing). Signed-off-by: Iouri Tarassov <iourit@microsoft.com>
- Loading branch information