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, 312 insertions, 0 deletions

diff --git a/legacy-driver-guest/vf/gsp_vf.c b/legacy-driver-guest/vf/gsp_vf.c
new file mode 100644
index 0000000..951c84d
--- /dev/null
+++ b/legacy-driver-guest/vf/gsp_vf.c
@@ -0,0 +1,312 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * GSP vGPU-IO Driver - VF Management
+ */
+
+#include "gsp_core.h"
+#include <linux/slab.h>
+#include <linux/idr.h>
+#include <linux/list.h>
+
+/* VF Reference Counting */
+static void gsp_vf_release(struct kref *ref)
+{
+    struct gsp_vf *vf = container_of(ref, struct gsp_vf, ref);
+    kfree(vf);
+}
+
+struct gsp_vf *gsp_vf_get(struct gsp_device *gsp, uint32_t id)
+{
+    struct gsp_vf *vf;
+    
+    mutex_lock(&gsp->vf_lock);
+    vf = idr_find(&gsp->vf_idr, id);
+    if (vf && vf->active)
+        kref_get(&vf->ref);
+    else
+        vf = NULL;
+    mutex_unlock(&gsp->vf_lock);
+    
+    return vf;
+}
+
+void gsp_vf_put(struct gsp_vf *vf)
+{
+    if (vf)
+        kref_put(&vf->ref, gsp_vf_release);
+}
+
+/* VF Creation */
+int gsp_vf_create(struct gsp_device *gsp, const struct gsp_vf_config *config)
+{
+    struct gsp_vf *vf;
+    int ret, id;
+    
+    if (!config || config->id == 0 || config->id > gsp->max_vfs)
+        return -EINVAL;
+    
+    vf = kzalloc(sizeof(*vf), GFP_KERNEL);
+    if (!vf)
+        return -ENOMEM;
+    
+    vf->gsp = gsp;
+    vf->id = config->id;
+    kref_init(&vf->ref);
+    mutex_init(&vf->lock);
+    mutex_init(&vf->wpr_lock);
+    INIT_LIST_HEAD(&vf->wpr_list);
+    INIT_LIST_HEAD(&vf->list);
+    atomic_set(&vf->refcount, 1);
+    
+    /* Copy configuration */
+    vf->bar1_offset = config->bar1_offset;
+    vf->bar1_size = config->bar1_size;
+    vf->pt_base = config->pt_base;
+    vf->pt_size = config->pt_size;
+    vf->cmd_base = config->cmd_base;
+    vf->cmd_size = config->cmd_size;
+    
+    /* Allocate VF ID */
+    mutex_lock(&gsp->vf_lock);
+    id = idr_alloc(&gsp->vf_idr, vf, config->id, config->id + 1, GFP_KERNEL);
+    if (id < 0) {
+        ret = id;
+        mutex_unlock(&gsp->vf_lock);
+        goto err_free;
+    }
+    
+    list_add(&vf->list, &gsp->vf_list);
+    vf->active = true;
+    vf->state = GSP_VF_ACTIVE;
+    mutex_unlock(&gsp->vf_lock);
+    
+    /* Initialize subsystems */
+    ret = gsp_vf_wpr_init(vf);
+    if (ret) {
+        dev_err(gsp->dev, "VF %u: WPR init failed: %d\n", vf->id, ret);
+        goto err_wpr;
+    }
+    
+    ret = gsp_vf_bar_init(vf);
+    if (ret) {
+        dev_err(gsp->dev, "VF %u: BAR init failed: %d\n", vf->id, ret);
+        goto err_bar;
+    }
+    
+    dev_info(gsp->dev, "VF %u created (bar1=0x%llx-0x%llx, pt=0x%llx-0x%llx)\n",
+             vf->id, vf->bar1_offset, vf->bar1_offset + vf->bar1_size,
+             vf->pt_base, vf->pt_base + vf->pt_size);
+    
+    return 0;
+
+err_bar:
+    gsp_vf_wpr_fini(vf);
+err_wpr:
+    mutex_lock(&gsp->vf_lock);
+    list_del(&vf->list);
+    idr_remove(&gsp->vf_idr, vf->id);
+    vf->active = false;
+    mutex_unlock(&gsp->vf_lock);
+err_free:
+    kfree(vf);
+    return ret;
+}
+
+/* VF Destruction */
+void gsp_vf_destroy(struct gsp_vf *vf)
+{
+    struct gsp_device *gsp = vf->gsp;
+    uint32_t id = vf->id;
+    
+    if (!vf->active)
+        return;
+    
+    /* Cancel any ongoing migration */
+    mutex_lock(&vf->lock);
+    if (vf->migration) {
+        gsp_migration_cancel(vf->migration);
+        vf->migration = NULL;
+    }
+    vf->active = false;
+    vf->state = GSP_VF_DESTROYING;
+    mutex_unlock(&vf->lock);
+    
+    /* Remove from device lists */
+    mutex_lock(&gsp->vf_lock);
+    list_del(&vf->list);
+    idr_remove(&gsp->vf_idr, id);
+    mutex_unlock(&gsp->vf_lock);
+    
+    /* Cleanup subsystems */
+    gsp_vf_bar_fini(vf);
+    gsp_vf_wpr_fini(vf);
+    
+    dev_info(gsp->dev, "VF %u destroyed\n", id);
+    
+    gsp_vf_put(vf);
+}
+
+/* Memory Allocation */
+int gsp_vf_mem_alloc(struct gsp_vf *vf, uint64_t size, uint32_t flags,
+                     uint64_t *handle, uint64_t *va)
+{
+    struct gsp_alloc_payload payload = {
+        .size = size,
+        .type = flags & 0x3,  /* Low bits: mem type */
+        .alignment = 0x1000,
+        .va_hint = 0,
+        .flags = flags >> 2,
+        .pid = current->pid,
+        .vf_id = vf->id,
+    };
+    struct gsp_alloc_response resp;
+    size_t resp_size = sizeof(resp);
+    int ret;
+    
+    if (!handle || !va)
+        return -EINVAL;
+    
+    /* In emulation mode, handle locally */
+    if (emulation_mode) {
+        static uint64_t next_handle = 0x1000;
+        static uint64_t next_va = 0x100000000ULL;  /* 4GB base */
+        
+        *handle = next_handle++;
+        *va = next_va;
+        next_va += (size + 0xFFF) & ~0xFFF;  /* Align to 4KB */
+        
+        dev_dbg(vf->gsp->dev, "VF %u: alloc handle=0x%llx va=0x%llx size=0x%llx (emulation)\n",
+                vf->id, *handle, *va, size);
+        
+        return 0;
+    }
+    
+    ret = gsp_rpc_send_sync(vf->gsp->rpc, GSP_RPC_ALLOC,
+                            &payload, sizeof(payload),
+                            &resp, &resp_size);
+    if (ret)
+        return ret;
+    
+    *handle = resp.handle;
+    *va = resp.va;
+    
+    dev_dbg(vf->gsp->dev, "VF %u: alloc handle=0x%llx va=0x%llx size=0x%llx\n",
+            vf->id, *handle, *va, size);
+    
+    return 0;
+}
+
+int gsp_vf_mem_free(struct gsp_vf *vf, uint64_t handle)
+{
+    /* FREE uses primary dispatcher sub-command */
+    struct gsp_subcmd_desc desc;
+    size_t desc_size = sizeof(desc);
+    int ret;
+    
+    /* Subcmd for FREE: need to determine which subcmd corresponds to FREE */
+    /* For now, use a generic FREE RPC if implemented */
+    
+    /* Placeholder - actual implementation depends on FREE subcmd mapping */
+    dev_dbg(vf->gsp->dev, "VF %u: free handle=0x%llx\n", vf->id, handle);
+    
+    return 0;
+}
+
+/* Memory Pinning */
+int gsp_vf_mem_pin(struct gsp_vf *vf, uint64_t handle,
+                   uint64_t offset, uint64_t size,
+                   uint64_t *dma_addr)
+{
+    struct gsp_pin_payload payload = {
+        .handle = handle,
+        .offset = offset,
+        .size = size,
+        .vf_id = vf->id,
+        .flags = GSP_PIN_READ | GSP_PIN_WRITE,
+    };
+    struct gsp_pin_response resp;
+    size_t resp_size = sizeof(resp);
+    int ret;
+    
+    if (!dma_addr)
+        return -EINVAL;
+    
+    /* In emulation mode, handle locally */
+    if (emulation_mode) {
+        static uint64_t next_dma = 0x200000000ULL;  /* 8GB base */
+        *dma_addr = next_dma;
+        next_dma += (size + 0xFFF) & ~0xFFF;
+        
+        dev_dbg(vf->gsp->dev, "VF %u: pin handle=0x%llx offset=0x%llx size=0x%llx -> dma=0x%llx (emulation)\n",
+                vf->id, handle, offset, size, *dma_addr);
+        
+        return 0;
+    }
+    
+    ret = gsp_rpc_send_sync(vf->gsp->rpc, GSP_RPC_PIN,
+                            &payload, sizeof(payload),
+                            &resp, &resp_size);
+    if (ret)
+        return ret;
+    
+    *dma_addr = resp.dma_addr;
+    
+    dev_dbg(vf->gsp->dev, "VF %u: pin handle=0x%llx offset=0x%llx size=0x%llx -> dma=0x%llx\n",
+            vf->id, handle, offset, size, *dma_addr);
+    
+    return 0;
+}
+
+int gsp_vf_mem_unpin(struct gsp_vf *vf, uint64_t handle)
+{
+    dev_dbg(vf->gsp->dev, "VF %u: unpin handle=0x%llx\n", vf->id, handle);
+    return 0;
+}
+
+/* Memory Mapping */
+int gsp_vf_mem_map(struct gsp_vf *vf, uint64_t handle, uint64_t gpu_va,
+                   uint64_t size, uint32_t prot, uint32_t flags,
+                   uint64_t offset, uint64_t *mapped_va)
+{
+    /* In emulation mode, just return the requested VA */
+    if (emulation_mode) {
+        if (mapped_va)
+            *mapped_va = gpu_va;
+        
+        dev_dbg(vf->gsp->dev, "VF %u: map handle=0x%llx gpu_va=0x%llx size=0x%llx -> mapped_va=0x%llx (emulation)\n",
+                vf->id, handle, gpu_va, size, gpu_va);
+        
+        return 0;
+    }
+    
+    /* TODO: Implement GSP RPC for MAP */
+    dev_dbg(vf->gsp->dev, "VF %u: map handle=0x%llx gpu_va=0x%llx size=0x%llx\n",
+            vf->id, handle, gpu_va, size);
+    
+    if (mapped_va)
+        *mapped_va = gpu_va;
+    
+    return 0;
+}
+
+int gsp_vf_mem_unmap(struct gsp_vf *vf, uint64_t handle, uint64_t gpu_va, uint64_t size)
+{
+    dev_dbg(vf->gsp->dev, "VF %u: unmap handle=0x%llx gpu_va=0x%llx size=0x%llx\n",
+            vf->id, handle, gpu_va, size);
+    return 0;
+}
+
+/* Scheduling */
+int gsp_vf_schedule(struct gsp_vf *vf, uint32_t priority, uint64_t time_slice_ns)
+{
+    /* Use primary dispatcher SCHEDULE subcmd or secondary dispatcher */
+    dev_dbg(vf->gsp->dev, "VF %u: schedule prio=%u slice=%llu ns\n",
+            vf->id, priority, time_slice_ns);
+    return 0;
+}
+
+int gsp_vf_yield(struct gsp_vf *vf)
+{
+    dev_dbg(vf->gsp->dev, "VF %u: yield\n", vf->id);
+    return 0;
+}
\ No newline at end of file