From 00e54c13c7c9030c4d5f7151467f2d85a3730e29 Mon Sep 17 00:00:00 2001 From: Arthur Hanson Rasmusson Date: Sun, 13 Sep 2026 18:36:21 -0400 Subject: Initial commit: Open-IOV vGPU-IO complete stack - Firmware decompilation: GSP 570.144W (28.5MB ELF, RV64IMAC), 132 decompiled C files - Kernel modules: nvidia.ko (34MB, OpenRM), open-iov-vfio.ko (2.8MB, VFIO mdev), gsp-vgpu.ko (1.8MB, guest driver) - Userspace: libiov.so/libvgpu.so (76KB), example_vgpu (28KB), rpc_emulator.c - PCI config space emulation (256B with PM, MSI, MSI-X, PCIe caps) - 12 RPC types fully mapped from firmware decompilation - BAR trap, WPR, scheduling, notification, migration emulation - Helper scripts: load_vgpu_stack, create_mdev, launch_qemu_vgpu, test_vgpu_stack, setup_guest_driver, unload_vgpu_stack - Architecture documentation - Wiki: 10 technical documents - Git LFS for large binaries (.ko, .bin, .iso, .img, .o, ghidra repos) - nvidia-open-gpu-kernel-modules as submodule --- kernel/include/nvidia-vgpu-vfio.h | 194 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 194 insertions(+) create mode 100644 kernel/include/nvidia-vgpu-vfio.h (limited to 'kernel/include/nvidia-vgpu-vfio.h') diff --git a/kernel/include/nvidia-vgpu-vfio.h b/kernel/include/nvidia-vgpu-vfio.h new file mode 100644 index 0000000..7f778ee --- /dev/null +++ b/kernel/include/nvidia-vgpu-vfio.h @@ -0,0 +1,194 @@ +#ifndef _NVIDIA_VGPU_VFIO_H_ +#define _NVIDIA_VGPU_VFIO_H_ + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#define OPEN_IOV_VFIO_VERSION "0.1.0" +#define OPEN_IOV_VFIO_NAME "open-iov-vfio" + +#define NVIDIA_VGPU_VFIO_VERSION OPEN_IOV_VFIO_VERSION +#define NVIDIA_VGPU_VFIO_NAME OPEN_IOV_VFIO_NAME + +#define NVIDIA_VGPU_MAX_VF_COUNT 32 +#define NVIDIA_VGPU_MAX_REGIONS 8 +#define NVIDIA_VGPU_MAX_IRQS 16 + +#define NV_VGPU_MAX_REGIONS NVIDIA_VGPU_MAX_REGIONS +#define NV_VGPU_MAX_IRQS NVIDIA_VGPU_MAX_IRQS + +#define NV_VGPU_BAR0_SIZE (16 * 1024 * 1024) +#define NV_VGPU_BAR1_SIZE (256 * 1024 * 1024) +#define NV_VGPU_BAR2_SIZE (32 * 1024 * 1024) +#define NV_VGPU_BAR3_SIZE (32 * 1024 * 1024) + +enum nv_vgpu_type { + NV_VGPU_TYPE_UNKNOWN = 0, + NV_VGPU_TYPE_AMPERE, + NV_VGPU_TYPE_HOPPER, + NV_VGPU_TYPE_TURING, + NV_VGPU_TYPE_PASCAL, + NV_VGPU_TYPE_MAXWELL, + NV_VGPU_TYPE_KEPLER, +}; + +enum nv_vgpu_state { + NV_VGPU_STATE_UNINITIALIZED = 0, + NV_VGPU_STATE_INITIALIZING, + NV_VGPU_STATE_READY, + NV_VGPU_STATE_RUNNING, + NV_VGPU_STATE_SUSPENDED, + NV_VGPU_STATE_ERROR, + NV_VGPU_STATE_MIGRATING, +}; + +struct nv_vgpu_bar { + u64 gpa; + u64 size; + u64 iova; + void *vaddr; + bool mapped; + bool trapped; + struct mutex lock; +}; + +struct nv_vgpu_irq { + int virq; + int type; + bool masked; + struct eventfd_ctx *trigger; + struct irq_domain *domain; +}; + +struct nv_vgpu_region { + u64 offset; + u64 size; + u64 iova; + int fd; + void *vaddr; + bool readonly; +}; + +struct nv_vgpu_migration_ctx; + +struct nv_vgpu_migration_state { + u32 version; + u32 flags; + u64 timestamp; + u8 *data; + size_t data_size; + struct completion done; + int result; + struct nv_vgpu_migration_ctx *ctx; +}; + +struct nv_vgpu_device { + struct mdev_device *mdev; + struct pci_dev *pdev; + struct device dev; + struct mutex lock; + enum nv_vgpu_type type; + enum nv_vgpu_state state; + u32 vgpu_id; + u32 gpu_instance_id; + u32 compute_instance_id; + + struct nv_vgpu_bar bars[NV_VGPU_MAX_REGIONS]; + struct nv_vgpu_irq irqs[NV_VGPU_MAX_IRQS]; + struct nv_vgpu_region regions[NV_VGPU_MAX_REGIONS]; + + struct nv_vgpu_migration_state migration; + + struct work_struct init_work; + struct work_struct cleanup_work; + + void *gsp_comm_ctx; + void *bar_emul_ctx; + void *mm_ctx; + void *irq_ctx; + + u64 fb_base; + u64 fb_size; + u64 bar0_config_base; + + struct list_head list; + + bool dry_run; + + /* PCI Config Space for VFIO mdev */ + u8 pci_config[256]; + bool config_space_dirty; +}; + +struct nv_vgpu_driver_data { + struct mutex lock; + struct list_head devices; + u32 device_count; + struct class *class; + struct bus_type *bus; +}; + +extern struct nv_vgpu_driver_data nv_vgpu_driver; + +int nv_vgpu_device_create(struct mdev_device *mdev); +void nv_vgpu_device_destroy(struct mdev_device *mdev); +int nv_vgpu_device_init(struct nv_vgpu_device *vgpu); +void nv_vgpu_device_cleanup(struct nv_vgpu_device *vgpu); + +int nv_vgpu_bar_init(struct nv_vgpu_device *vgpu); +void nv_vgpu_bar_cleanup(struct nv_vgpu_device *vgpu); +int nv_vgpu_bar_trap(struct nv_vgpu_device *vgpu, int bar_idx); +void nv_vgpu_bar_untrap(struct nv_vgpu_device *vgpu, int bar_idx); + +int nv_vgpu_irq_init(struct nv_vgpu_device *vgpu); +void nv_vgpu_irq_cleanup(struct nv_vgpu_device *vgpu); +int nv_vgpu_irq_enable(struct nv_vgpu_device *vgpu, int irq_idx); +int nv_vgpu_irq_disable(struct nv_vgpu_device *vgpu, int irq_idx); + +int nv_vgpu_memory_init(struct nv_vgpu_device *vgpu); +void nv_vgpu_memory_cleanup(struct nv_vgpu_device *vgpu); +int nv_vgpu_memory_map(struct nv_vgpu_device *vgpu, u64 gpa, u64 size, u64 *iova); +int nv_vgpu_memory_unmap(struct nv_vgpu_device *vgpu, u64 iova, u64 size); + +int nv_vgpu_gsp_init(struct nv_vgpu_device *vgpu); +void nv_vgpu_gsp_cleanup(struct nv_vgpu_device *vgpu); +int nv_vgpu_gsp_send_rpc(struct nv_vgpu_device *vgpu, void *msg, size_t size); +int nv_vgpu_gsp_recv_rpc(struct nv_vgpu_device *vgpu, void *msg, size_t size, int timeout_ms); +int nv_vgpu_gsp_load_firmware(struct nv_vgpu_device *vgpu); +void nv_vgpu_gsp_release_firmware(struct nv_vgpu_device *vgpu); +int nv_vgpu_gsp_boot(struct nv_vgpu_device *vgpu); + +int nv_vgpu_migration_init(struct nv_vgpu_device *vgpu); +void nv_vgpu_migration_cleanup(struct nv_vgpu_device *vgpu); +int nv_vgpu_migration_save_state(struct nv_vgpu_device *vgpu, u8 **data, size_t *size); +int nv_vgpu_migration_restore_state(struct nv_vgpu_device *vgpu, u8 *data, size_t size); +int nv_vgpu_migration_prepare(struct nv_vgpu_device *vgpu); +int nv_vgpu_migration_complete(struct nv_vgpu_device *vgpu); +int nv_vgpu_migration_cancel(struct nv_vgpu_device *vgpu); +u8 *nv_vgpu_migration_get_data(struct nv_vgpu_device *vgpu, size_t *size); + +static inline struct nv_vgpu_device *mdev_to_vgpu(struct mdev_device *mdev) +{ + return dev_get_drvdata(&mdev->dev); +} + +#define nv_vgpu_dbg(vgpu, fmt, ...) \ + dev_dbg(&(vgpu)->dev, "%s: " fmt, __func__, ##__VA_ARGS__) + +#define nv_vgpu_info(vgpu, fmt, ...) \ + dev_info(&(vgpu)->dev, "%s: " fmt, __func__, ##__VA_ARGS__) + +#define nv_vgpu_warn(vgpu, fmt, ...) \ + dev_warn(&(vgpu)->dev, "%s: " fmt, __func__, ##__VA_ARGS__) + +#define nv_vgpu_err(vgpu, fmt, ...) \ + dev_err(&(vgpu)->dev, "%s: " fmt, __func__, ##__VA_ARGS__) + +#endif \ No newline at end of file -- cgit v1.2.3