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, 283 insertions, 0 deletions

diff --git a/userspace/tests/test_vgpu_userspace.c b/userspace/tests/test_vgpu_userspace.c
new file mode 100644
index 0000000..cd22111
--- /dev/null
+++ b/userspace/tests/test_vgpu_userspace.c
@@ -0,0 +1,283 @@
+/*
+ * GSP vGPU-IO Userspace Test
+ * Tests VFIO interface and driver functionality from userspace
+ */
+
+#define _GNU_SOURCE
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <sys/ioctl.h>
+#include <sys/mman.h>
+#include <linux/vfio.h>
+#include <errno.h>
+#include <stdint.h>
+#include <inttypes.h>
+
+#define GSP_VGPU_DEVICE "/dev/vfio/gsp-vgpu"
+
+#define TEST_PASS 0
+#define TEST_FAIL 1
+
+static int test_count = 0;
+static int pass_count = 0;
+static int fail_count = 0;
+
+#define TEST_ASSERT(cond, msg) \
+    do { \
+        test_count++; \
+        if (cond) { \
+            pass_count++; \
+            printf("PASS: %s\n", msg); \
+        } else { \
+            fail_count++; \
+            fprintf(stderr, "FAIL: %s\n", msg); \
+        } \
+    } while (0)
+
+#define TEST_ASSERT_EQ(a, b, msg) \
+    TEST_ASSERT((a) == (b), msg " (expected " #b ", got " #a ")")
+
+/* IOCTL definitions matching kernel driver */
+#define GSP_IOCTL_MAGIC 'G'
+#define GSP_IOCTL_CREATE_VF _IOW(GSP_IOCTL_MAGIC, 0x01, struct gsp_vf_config)
+#define GSP_IOCTL_DESTROY_VF _IOW(GSP_IOCTL_MAGIC, 0x02, uint32_t)
+#define GSP_IOCTL_ALLOC _IOWR(GSP_IOCTL_MAGIC, 0x10, struct gsp_alloc_args)
+#define GSP_IOCTL_FREE _IOW(GSP_IOCTL_MAGIC, 0x11, struct gsp_free_args)
+#define GSP_IOCTL_MAP _IOWR(GSP_IOCTL_MAGIC, 0x12, struct gsp_map_args)
+#define GSP_IOCTL_UNMAP _IOW(GSP_IOCTL_MAGIC, 0x13, struct gsp_unmap_args)
+#define GSP_IOCTL_PIN _IOWR(GSP_IOCTL_MAGIC, 0x14, struct gsp_pin_args)
+#define GSP_IOCTL_UNPIN _IOW(GSP_IOCTL_MAGIC, 0x15, struct gsp_unpin_args)
+#define GSP_IOCTL_BAR_TRAP _IOWR(GSP_IOCTL_MAGIC, 0x16, struct gsp_bar_trap_args)
+#define GSP_IOCTL_MIGRATE _IOWR(GSP_IOCTL_MAGIC, 0x17, struct gsp_migrate_args)
+#define GSP_IOCTL_WPR_PROTECT _IOWR(GSP_IOCTL_MAGIC, 0x18, struct gsp_wpr_args)
+#define GSP_IOCTL_WPR_UNPROTECT _IOW(GSP_IOCTL_MAGIC, 0x19, struct gsp_wpr_args)
+
+struct gsp_vf_config {
+    uint32_t id;
+    uint64_t bar1_offset;
+    uint64_t bar1_size;
+    uint64_t pt_base;
+    uint64_t pt_size;
+    uint64_t cmd_base;
+    uint64_t cmd_size;
+    uint32_t max_memory_mb;
+    uint32_t num_engines;
+};
+
+struct gsp_alloc_args {
+    uint32_t vf_id;
+    uint64_t size;
+    uint32_t type;
+    uint64_t alignment;
+    uint64_t va_hint;
+    uint32_t flags;
+    uint32_t pid;
+    uint64_t handle;
+    uint64_t va;
+    uint64_t actual_size;
+};
+
+struct gsp_free_args {
+    uint32_t vf_id;
+    uint64_t handle;
+};
+
+struct gsp_map_args {
+    uint32_t vf_id;
+    uint64_t handle;
+    uint64_t gpu_va;
+    uint64_t size;
+    uint32_t prot;
+    uint32_t flags;
+    uint64_t offset;
+    uint64_t mapped_va;
+};
+
+struct gsp_unmap_args {
+    uint32_t vf_id;
+    uint64_t handle;
+    uint64_t gpu_va;
+    uint64_t size;
+};
+
+struct gsp_pin_args {
+    uint32_t vf_id;
+    uint64_t handle;
+    uint64_t offset;
+    uint64_t size;
+    uint32_t flags;
+    uint64_t dma_addr;
+    uint32_t num_pages;
+};
+
+struct gsp_unpin_args {
+    uint32_t vf_id;
+    uint64_t handle;
+    uint64_t offset;
+    uint64_t size;
+};
+
+struct gsp_bar_trap_args {
+    uint32_t vf_id;
+    uint32_t bar_index;
+    uint32_t trap_type;
+    uint64_t offset;
+    uint64_t value;
+    uint32_t size;
+    uint64_t ret_value;
+};
+
+struct gsp_migrate_args {
+    uint32_t vf_id;
+    uint32_t phase;
+    uint32_t subcmd;
+    uint64_t src_vf;
+    uint64_t dst_vf;
+    uint64_t data_ptr;
+    uint64_t data_size;
+    uint32_t flags;
+    uint64_t token;
+};
+
+struct gsp_wpr_args {
+    uint32_t vf_id;
+    uint64_t base;
+    uint64_t size;
+    uint32_t flags;
+};
+
+int main(int argc, char *argv[])
+{
+    int fd, ret;
+    uint32_t vf_id = 1;
+
+    printf("========================================\n");
+    printf("GSP vGPU-IO Userspace Test Suite\n");
+    printf("========================================\n");
+
+    /* Open device */
+    fd = open(GSP_VGPU_DEVICE, O_RDWR);
+    if (fd < 0) {
+        fprintf(stderr, "Failed to open %s: %s\n", GSP_VGPU_DEVICE, strerror(errno));
+        fprintf(stderr, "Make sure driver is loaded and VFIO device created\n");
+        return TEST_FAIL;
+    }
+    printf("Opened device: %s (fd=%d)\n", GSP_VGPU_DEVICE, fd);
+
+    /* Test 1: Create VF */
+    {
+        struct gsp_vf_config config = {
+            .id = vf_id,
+            .bar1_offset = 0x10000000,
+            .bar1_size = 0x100000,
+            .pt_base = 0x20000000,
+            .pt_size = 0x100000,
+            .cmd_base = 0x30000000,
+            .cmd_size = 0x10000,
+            .max_memory_mb = 1024,
+            .num_engines = 4,
+        };
+
+        ret = ioctl(fd, GSP_IOCTL_CREATE_VF, &config);
+        TEST_ASSERT(ret == 0, "Create VF");
+        if (ret == 0)
+            printf("  VF %u created successfully\n", vf_id);
+    }
+
+    /* Test 2: Allocate memory */
+    {
+        struct gsp_alloc_args args = {
+            .vf_id = vf_id,
+            .size = 0x10000,  // 64KB
+            .type = 0,        // VRAM
+            .alignment = 0x1000,
+            .va_hint = 0,
+            .flags = 0,
+            .pid = getpid(),
+        };
+
+        ret = ioctl(fd, GSP_IOCTL_ALLOC, &args);
+        TEST_ASSERT(ret == 0, "Allocate memory");
+        if (ret == 0) {
+            printf("  Allocated: handle=0x%" PRIx64 ", va=0x%" PRIx64 ", size=0x%" PRIx64 "\n",
+                   args.handle, args.va, args.actual_size);
+        }
+    }
+
+    /* Test 3: Pin memory */
+    {
+        struct gsp_pin_args args = {
+            .vf_id = vf_id,
+            .handle = 0,  // Would use handle from alloc
+            .offset = 0,
+            .size = 0x1000,
+            .flags = 3,  // READ | WRITE
+        };
+
+        ret = ioctl(fd, GSP_IOCTL_PIN, &args);
+        TEST_ASSERT(ret == 0 || errno == ENODEV, "Pin memory (may fail without real allocation)");
+    }
+
+    /* Test 4: BAR Trap */
+    {
+        struct gsp_bar_trap_args args = {
+            .vf_id = vf_id,
+            .bar_index = 1,
+            .trap_type = 1,  // WRITE
+            .offset = 0x1000,
+            .value = 0xDEADBEEF,
+            .size = 4,
+        };
+
+        ret = ioctl(fd, GSP_IOCTL_BAR_TRAP, &args);
+        TEST_ASSERT(ret == 0 || errno == ENODEV, "BAR trap (may fail without real HW)");
+    }
+
+    /* Test 5: WPR Protect */
+    {
+        struct gsp_wpr_args args = {
+            .vf_id = vf_id,
+            .base = 0x10000000,
+            .size = 0x10000,
+            .flags = 3,  // WRITE | READ
+        };
+
+        ret = ioctl(fd, GSP_IOCTL_WPR_PROTECT, &args);
+        TEST_ASSERT(ret == 0 || errno == ENODEV, "WPR protect (may fail without real HW)");
+    }
+
+    /* Test 6: Migration */
+    {
+        struct gsp_migrate_args args = {
+            .vf_id = vf_id,
+            .phase = 1,  // PREPARE
+            .subcmd = 3,
+            .src_vf = vf_id,
+            .dst_vf = 2,
+            .flags = 1,  // LIVE
+        };
+
+        ret = ioctl(fd, GSP_IOCTL_MIGRATE, &args);
+        TEST_ASSERT(ret == 0 || errno == ENODEV, "Migration (may fail without real HW)");
+    }
+
+    /* Test 7: Destroy VF */
+    {
+        ret = ioctl(fd, GSP_IOCTL_DESTROY_VF, &vf_id);
+        TEST_ASSERT(ret == 0, "Destroy VF");
+        if (ret == 0)
+            printf("  VF %u destroyed successfully\n", vf_id);
+    }
+
+    close(fd);
+
+    printf("========================================\n");
+    printf("Userspace Test Results: %d total, %d passed, %d failed\n",
+           test_count, pass_count, fail_count);
+    printf("========================================\n");
+
+    return (fail_count > 0) ? TEST_FAIL : TEST_PASS;
+}
\ No newline at end of file