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, 172 insertions, 0 deletions

diff --git a/firmware/57X-GSP/570.144/find_functions.py b/firmware/57X-GSP/570.144/find_functions.py
new file mode 100644
index 0000000..b9f3232
--- /dev/null
+++ b/firmware/57X-GSP/570.144/find_functions.py
@@ -0,0 +1,172 @@
+#!/usr/bin/env python3
+"""
+Search for RISC-V function prologues in GSP firmware binary.
+RISC-V typical prologue:
+  addi sp, sp, -N     (0x1141 + offset)
+  sd ra, N-8(sp)      (0x80e2 + offset)  
+  sd s0, N-16(sp)     (0x84e2 + offset)
+  addi s0, sp, N      (0x1144 + offset)
+"""
+
+import struct
+import json
+import sys
+import os
+
+FWIMAGE_PATH = os.path.join(os.path.dirname(os.path.abspath(__file__)), "gsp-570.144.fwimage.bin")
+OUTPUT_DIR = os.path.dirname(os.path.abspath(__file__))
+BASE_ADDR = 0x04000000  # Load address from memory_blocks.json
+
+# RISC-V instruction patterns (compressed and uncompressed)
+# addi sp, sp, -imm (opcode 0x13, funct3 0, rd=sp=2, rs1=sp=2)
+# sd ra, offset(sp) (opcode 0x23, funct3 3, rs1=sp=2, rs2=ra=1)
+# sd s0, offset(sp) (opcode 0x23, funct3 3, rs1=sp=2, rs2=s0=8)
+# addi s0, sp, imm (opcode 0x13, funct3 0, rd=s0=8, rs1=sp=2)
+
+def read_binary(path):
+    with open(path, 'rb') as f:
+        return f.read()
+
+def find_prologues(data, base_addr):
+    """Find potential function prologues in the binary."""
+    functions = []
+    
+    # Search for standard RISC-V function prologue pattern
+    # Pattern: addi sp, sp, -N; sd ra, N-8(sp); sd s0, N-16(sp); addi s0, sp, N
+    # We'll look for the first two instructions as a signature
+    
+    for i in range(0, len(data) - 8, 2):  # 16-bit aligned
+        # Check for 32-bit instructions (not compressed)
+        if i + 8 <= len(data):
+            inst1 = struct.unpack('<I', data[i:i+4])[0]
+            inst2 = struct.unpack('<I', data[i+4:i+8])[0]
+            
+            # Check for addi sp, sp, -imm (opcode=0x13, rd=2, rs1=2, funct3=0)
+            # 0x00000013 = addi zero, zero, 0 (nop)
+            # addi sp, sp, -N: rd=2, rs1=2, funct3=0, opcode=0x13
+            # Encoding: imm[11:0] | rs1 | funct3 | rd | opcode
+            if (inst1 & 0x707F) == 0x0013:  # opcode=0x13, funct3=0
+                rd = (inst1 >> 7) & 0x1F
+                rs1 = (inst1 >> 15) & 0x1F
+                if rd == 2 and rs1 == 2:  # sp = x2
+                    # Potential addi sp, sp, -N
+                    imm = (inst1 >> 20) & 0xFFF
+                    if imm >= 0x800:  # Negative immediate (sign extended)
+                        imm = imm - 0x1000
+                    
+                    # Check next instruction: sd ra, offset(sp)
+                    if (inst2 & 0x707F) == 0x23:  # opcode=0x23 (store)
+                        funct3 = (inst2 >> 12) & 0x7
+                        rs1_2 = (inst2 >> 15) & 0x1F
+                        rs2 = (inst2 >> 20) & 0x1F
+                        if funct3 == 3 and rs1_2 == 2 and rs2 == 1:  # sd ra, offset(sp)
+                            # Found potential prologue
+                            addr = base_addr + i
+                            functions.append({
+                                "address": f"0x{addr:08x}",
+                                "offset": i,
+                                "stack_size": -imm,
+                                "type": "standard_prologue"
+                            })
+    
+    return functions
+
+def find_all_functions(data, base_addr):
+    """Find all potential functions using multiple patterns."""
+    functions = []
+    
+    # Pattern 1: Standard prologue with addi sp, sp, -N
+    for i in range(0, len(data) - 8, 2):
+        if i + 8 <= len(data):
+            inst1 = struct.unpack('<I', data[i:i+4])[0]
+            inst2 = struct.unpack('<I', data[i+4:i+8])[0]
+            
+            # addi sp, sp, -imm
+            if (inst1 & 0x707F) == 0x0013:
+                rd = (inst1 >> 7) & 0x1F
+                rs1 = (inst1 >> 15) & 0x1F
+                if rd == 2 and rs1 == 2:
+                    imm = (inst1 >> 20) & 0xFFF
+                    if imm >= 0x800:
+                        imm = imm - 0x1000
+                    
+                    # sd ra, offset(sp) - check multiple possible offsets
+                    if (inst2 & 0x707F) == 0x23:
+                        funct3 = (inst2 >> 12) & 0x7
+                        rs1_2 = (inst2 >> 15) & 0x1F
+                        rs2 = (inst2 >> 20) & 0x1F
+                        if funct3 == 3 and rs1_2 == 2 and rs2 == 1:
+                            addr = base_addr + i
+                            functions.append({
+                                "address": f"0x{addr:08x}",
+                                "offset": i,
+                                "stack_size": -imm,
+                                "pattern": "addi_sp_sd_ra"
+                            })
+    
+    # Pattern 2: Compressed instructions (16-bit)
+    # c.addi sp, sp, -N (opcode 0x6101 + offset)
+    # c.sd ra, offset(sp) (opcode 0xE006 + offset)
+    for i in range(0, len(data) - 4, 2):
+        if i + 4 <= len(data):
+            inst1 = struct.unpack('<H', data[i:i+2])[0]
+            inst2 = struct.unpack('<H', data[i+2:i+4])[0]
+            
+            # c.addi sp, sp, imm (rd'=sp=2, imm != 0)
+            # Format: 0x6101 | (imm[5] << 12) | (imm[4:0] << 7) | (rs1'/rd' << 7)
+            if (inst1 & 0xE003) == 0x6001:  # c.addi
+                rd_prime = (inst1 >> 7) & 0x7
+                if rd_prime == 2:  # sp
+                    # c.sd ra, offset(sp) - ra=x1, sp=x2
+                    # Format: 0xE000 | (offset[8:6] << 10) | (rs2' << 7) | (rs1' << 7) | offset[5:3]
+                    if (inst2 & 0xE003) == 0xE000:  # c.sd
+                        rs1_prime = (inst2 >> 7) & 0x7
+                        rs2_prime = (inst2 >> 10) & 0x7
+                        if rs1_prime == 2 and rs2_prime == 1:  # sp, ra
+                            addr = base_addr + i
+                            functions.append({
+                                "address": f"0x{addr:08x}",
+                                "offset": i,
+                                "pattern": "compressed_prologue"
+                            })
+    
+    return functions
+
+def main():
+    print(f"Reading firmware binary: {FWIMAGE_PATH}")
+    data = read_binary(FWIMAGE_PATH)
+    print(f"Binary size: {len(data)} bytes")
+    
+    print("Searching for function prologues...")
+    functions = find_all_functions(data, BASE_ADDR)
+    
+    print(f"Found {len(functions)} potential functions")
+    
+    # Deduplicate by address
+    seen = set()
+    unique = []
+    for f in functions:
+        if f["address"] not in seen:
+            seen.add(f["address"])
+            unique.append(f)
+    
+    print(f"Unique functions: {len(unique)}")
+    
+    # Save results
+    with open(f"{OUTPUT_DIR}/found_functions.json", "w") as f:
+        json.dump(unique, f, indent=2)
+    
+    # Also create a simple address list
+    with open(f"{OUTPUT_DIR}/function_addresses.txt", "w") as f:
+        for func in sorted(unique, key=lambda x: x["offset"]):
+            f.write(f"{func['address']}  stack_size={func.get('stack_size', '?')}  pattern={func['pattern']}\n")
+    
+    print(f"Results saved to {OUTPUT_DIR}/found_functions.json")
+    print(f"Address list saved to {OUTPUT_DIR}/function_addresses.txt")
+    
+    # Show first 50
+    for func in sorted(unique, key=lambda x: x["offset"])[:50]:
+        print(f"  {func['address']}  stack_size={func.get('stack_size', '?')}  pattern={func['pattern']}")
+
+if __name__ == "__main__":
+    main()