Open-IOV: initial public release

Open-IOV virtualizes NVIDIA GPUs so that unmodified guest drivers run with
real GPU functionality - CUDA first, 3D and video to follow - on unmodified
hosts, without a proprietary vGPU license and without patching any driver,
firmware or hypervisor.

Core principle: emulate the interface, pass through the work.

  kernel/     open-iov-vfio.ko - host VFIO mdev driver: virtual GPU device
              (PCI config space, BAR shadows, MSI-X), GSP boot emulation,
              GSP-RM message-queue/RPC emulation, memory model (GPU-VA to
              guest-PA discovery) and the semaphore-completion service that
              advances guest UVM progress.
  userspace/  libiov.so management API and examples.
  tools/      guest diagnostics, firmware analysis wrappers, RM probes.
  scripts/    stack manager, VM harness, build/load/test automation.
  config/     hardware/version profiles (the portability surface).
  docs/       documentation tree - start at docs/README.md.
  wiki/       long-form firmware and protocol deep dives.

Status: initialization works end to end for a stock 570.144 Linux guest on
Turing (TU104): driver probe, nvidia-smi, the full GSP-RM control battery,
object creation and the early CUDA/UVM start-up path. GPU work forwarding -
the step that makes CUDA kernels execute on real hardware - is the active
work item; see docs/04-implementation/01-known-knowns-and-unknowns.md.

License: BSD-3-Clause for the project; GPL-2.0 for kernel modules that link
GPL-only kernel interfaces. See LICENSE and docs/09-reference/legal-and-licensing.md.

1 files changed, 176 insertions, 0 deletions

diff --git a/userspace/tests/test_guest_init.c b/userspace/tests/test_guest_init.c
new file mode 100644
index 0000000..135c16f
--- /dev/null
+++ b/userspace/tests/test_guest_init.c
@@ -0,0 +1,176 @@
+/*
+ * Test for guest vGPU driver initialization
+ * This test verifies that the guest vGPU driver can initialize properly
+ * when passed through to a VM via VFIO/mdev
+ */
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <sys/ioctl.h>
+#include <sys/mman.h>
+#include <linux/vfio.h>
+#include <linux/mdev.h>
+#include <errno.h>
+
+#define NV_VGPU_IOCTL_MAGIC 'N'
+#define NV_VGPU_IOCTL_GET_INFO \
+	_IOWR(NV_VGPU_IOCTL_MAGIC, 0x01, struct nv_vgpu_ioctl_get_info)
+#define NV_VGPU_IOCTL_MAP_BAR \
+	_IOWR(NV_VGPU_IOCTL_MAGIC, 0x02, struct nv_vgpu_ioctl_map_bar)
+#define NV_VGPU_IOCTL_GSP_RPC \
+	_IOWR(NV_VGPU_IOCTL_MAGIC, 0x06, struct nv_vgpu_ioctl_gsp_rpc)
+
+struct nv_vgpu_ioctl_get_info {
+    uint32_t vgpu_type;
+    uint32_t vgpu_id;
+    uint32_t gpu_instance_id;
+    uint32_t compute_instance_id;
+    uint32_t state;
+    uint64_t fb_base;
+    uint64_t fb_size;
+    uint64_t bar0_config_base;
+    uint32_t num_bars;
+    struct {
+        uint32_t index;
+        uint64_t gpa;
+        uint64_t size;
+        uint64_t iova;
+        bool mapped;
+        bool trapped;
+    } bars[8];
+};
+
+struct nv_vgpu_ioctl_map_bar {
+    uint32_t bar_index;
+    uint64_t gpa;
+    uint64_t size;
+    uint64_t iova;
+    uint32_t flags;
+};
+
+struct nv_vgpu_ioctl_gsp_rpc {
+    uint32_t type;
+    uint32_t flags;
+    uint64_t data_ptr;
+    uint32_t data_size;
+    uint32_t result;
+};
+
+int main(int argc, char *argv[])
+{
+    int ret = 0;
+    int vfio_fd = -1;
+    int mdev_fd = -1;
+    char *mdev_path = NULL;
+    struct nv_vgpu_ioctl_get_info info = {0};
+    struct nv_vgpu_ioctl_map_bar map_bar = {0};
+    struct vfio_device_info dev_info = {0};
+
+    printf("Testing guest vGPU driver initialization...\n");
+
+    if (argc < 2) {
+        printf("Usage: %s <mdev_uuid>\n", argv[0]);
+        printf("Example: %s 00000000-0000-0000-0000-000000000000\n", argv[0]);
+        return 1;
+    }
+
+    mdev_path = argv[1];
+
+    vfio_fd = open("/dev/vfio/vfio", O_RDWR);
+    if (vfio_fd < 0) {
+        printf("FAIL: Cannot open /dev/vfio/vfio: %s\n", strerror(errno));
+        return 1;
+    }
+
+    dev_info.argsz = sizeof(dev_info);
+    if (ioctl(vfio_fd, VFIO_GET_API_VERSION, &dev_info) < 0) {
+        printf("FAIL: VFIO_GET_API_VERSION failed: %s\n", strerror(errno));
+        ret = 1;
+        goto cleanup;
+    }
+
+    if (dev_info.version != VFIO_API_VERSION) {
+        printf("FAIL: VFIO API version mismatch\n");
+        ret = 1;
+        goto cleanup;
+    }
+
+    printf("PASS: VFIO API version %d\n", dev_info.version);
+
+    char mdev_dev_path[256];
+    snprintf(mdev_dev_path, sizeof(mdev_dev_path), "/dev/vfio/%s", mdev_path);
+
+    mdev_fd = open(mdev_dev_path, O_RDWR);
+    if (mdev_fd < 0) {
+        printf("FAIL: Cannot open mdev device %s: %s\n", mdev_dev_path, strerror(errno));
+        ret = 1;
+        goto cleanup;
+    }
+
+    printf("PASS: Opened mdev device %s\n", mdev_dev_path);
+
+    dev_info.argsz = sizeof(dev_info);
+    if (ioctl(mdev_fd, VFIO_DEVICE_GET_INFO, &dev_info) < 0) {
+        printf("FAIL: VFIO_DEVICE_GET_INFO failed: %s\n", strerror(errno));
+        ret = 1;
+        goto cleanup;
+    }
+
+    printf("PASS: Device info - flags: 0x%x, num_regions: %d, num_irqs: %d\n",
+           dev_info.flags, dev_info.num_regions, dev_info.num_irqs);
+
+    if (ioctl(mdev_fd, NV_VGPU_IOCTL_GET_INFO, &info) < 0) {
+        printf("FAIL: NV_VGPU_IOCTL_GET_INFO failed: %s\n", strerror(errno));
+        ret = 1;
+        goto cleanup;
+    }
+
+    printf("PASS: vGPU info - type: %d, id: %d, state: %d\n",
+           info.vgpu_type, info.vgpu_id, info.state);
+    printf("       fb_base: 0x%lx, fb_size: 0x%lx\n", info.fb_base, info.fb_size);
+    printf("       bar0_config_base: 0x%lx\n", info.bar0_config_base);
+    printf("       num_bars: %d\n", info.num_bars);
+
+    for (int i = 0; i < info.num_bars; i++) {
+        printf("       BAR%d: gpa=0x%lx, size=0x%lx, iova=0x%lx, mapped=%d, trapped=%d\n",
+               info.bars[i].index, info.bars[i].gpa, info.bars[i].size,
+               info.bars[i].iova, info.bars[i].mapped, info.bars[i].trapped);
+    }
+
+    if (info.num_bars > 0) {
+        map_bar.bar_index = 0;
+        map_bar.gpa = info.bars[0].gpa;
+        map_bar.size = info.bars[0].size;
+        map_bar.flags = 0x1;
+
+        if (ioctl(mdev_fd, NV_VGPU_IOCTL_MAP_BAR, &map_bar) < 0) {
+            printf("FAIL: NV_VGPU_IOCTL_MAP_BAR failed: %s\n", strerror(errno));
+            ret = 1;
+            goto cleanup;
+        }
+
+        printf("PASS: Mapped BAR0 at IOVA 0x%lx\n", map_bar.iova);
+    }
+
+    struct nv_vgpu_ioctl_gsp_rpc rpc = {0};
+    rpc.type = 0x01;
+    rpc.data_size = 0;
+
+    if (ioctl(mdev_fd, NV_VGPU_IOCTL_GSP_RPC, &rpc) < 0) {
+        printf("WARN: NV_VGPU_IOCTL_GSP_RPC failed: %s (may not be implemented yet)\n", strerror(errno));
+    } else {
+        printf("PASS: GSP RPC test completed\n");
+    }
+
+    printf("\nAll guest initialization tests PASSED\n");
+
+cleanup:
+    if (mdev_fd >= 0)
+        close(mdev_fd);
+    if (vfio_fd >= 0)
+        close(vfio_fd);
+
+    return ret;
+}
\ No newline at end of file