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, 224 insertions, 0 deletions

diff --git a/firmware/57X-GSP/570.144/find_functions_v2.py b/firmware/57X-GSP/570.144/find_functions_v2.py
new file mode 100644
index 0000000..df9506d
--- /dev/null
+++ b/firmware/57X-GSP/570.144/find_functions_v2.py
@@ -0,0 +1,224 @@
+#!/usr/bin/env python3
+"""
+Enhanced RISC-V function finder for GSP firmware.
+Searches for multiple prologue patterns.
+"""
+
+import struct
+import json
+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
+
+def read_binary(path):
+    with open(path, 'rb') as f:
+        return f.read()
+
+def find_functions(data, base_addr):
+    functions = []
+    
+    # Pattern 1: Standard 32-bit prologue
+    # addi sp, sp, -imm (opcode=0x13, funct3=0, rd=2, rs1=2)
+    # sd ra, offset(sp) (opcode=0x23, funct3=3, rs1=2, rs2=1)
+    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)
+                    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": "std_prologue_addi_sd"
+                            })
+                            continue
+    
+    # Pattern 2: Compressed prologue
+    # c.addi sp, sp, imm
+    # c.sd ra, offset(sp)
+    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
+            if (inst1 & 0xE003) == 0x6001:
+                rd_prime = (inst1 >> 7) & 0x7
+                if rd_prime == 2:
+                    # c.sd ra, offset(sp)
+                    if (inst2 & 0xE003) == 0xE000:
+                        rs1_prime = (inst2 >> 7) & 0x7
+                        rs2_prime = (inst2 >> 10) & 0x7
+                        if rs1_prime == 2 and rs2_prime == 1:
+                            addr = base_addr + i
+                            functions.append({
+                                "address": f"0x{addr:08x}",
+                                "offset": i,
+                                "pattern": "compressed_prologue"
+                            })
+                            continue
+    
+    # Pattern 3: Standard prologue with s0 save
+    # addi sp, sp, -N; sd ra, N-8(sp); sd s0, N-16(sp); addi s0, sp, N
+    for i in range(0, len(data) - 16, 2):
+        if i + 16 <= len(data):
+            inst1 = struct.unpack('<I', data[i:i+4])[0]
+            inst2 = struct.unpack('<I', data[i+4:i+8])[0]
+            inst3 = struct.unpack('<I', data[i+8:i+12])[0]
+            inst4 = struct.unpack('<I', data[i+12:i+16])[0]
+            
+            # addi sp, sp, -N
+            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, N-8(sp)
+                    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:
+                            # sd s0, N-16(sp)
+                            if (inst3 & 0x707F) == 0x23:
+                                funct3_3 = (inst3 >> 12) & 0x7
+                                rs1_3 = (inst3 >> 15) & 0x1F
+                                rs2_3 = (inst3 >> 20) & 0x1F
+                                if funct3_3 == 3 and rs1_3 == 2 and rs2_3 == 8:  # s0 = x8
+                                    # addi s0, sp, N
+                                    if (inst4 & 0x707F) == 0x0013:
+                                        rd4 = (inst4 >> 7) & 0x1F
+                                        rs1_4 = (inst4 >> 15) & 0x1F
+                                        if rd4 == 8 and rs1_4 == 2:
+                                            addr = base_addr + i
+                                            functions.append({
+                                                "address": f"0x{addr:08x}",
+                                                "offset": i,
+                                                "stack_size": -imm,
+                                                "pattern": "full_prologue_ra_s0"
+                                            })
+                                            continue
+    
+    # Pattern 4: Just sd ra, offset(sp) followed by code
+    # Many functions might just save ra
+    for i in range(0, len(data) - 4, 2):
+        if i + 4 <= len(data):
+            inst = struct.unpack('<I', data[i:i+4])[0]
+            
+            # sd ra, offset(sp)
+            if (inst & 0x707F) == 0x23:
+                funct3 = (inst >> 12) & 0x7
+                rs1 = (inst >> 15) & 0x1F
+                rs2 = (inst >> 20) & 0x1F
+                if funct3 == 3 and rs1 == 2 and rs2 == 1:
+                    # Check if previous instruction was addi sp, sp (within 8 bytes)
+                    has_addi_sp = False
+                    stack_size = 0
+                    for j in range(max(0, i-8), i, 2):
+                        if j >= 0:
+                            prev = struct.unpack('<I', data[j:j+4])[0]
+                            if (prev & 0x707F) == 0x0013:
+                                rd = (prev >> 7) & 0x1F
+                                rs1_p = (prev >> 15) & 0x1F
+                                if rd == 2 and rs1_p == 2:
+                                    imm = (prev >> 20) & 0xFFF
+                                    if imm >= 0x800:
+                                        imm = imm - 0x1000
+                                    has_addi_sp = True
+                                    stack_size = -imm
+                                    break
+                    
+                    if has_addi_sp:
+                        addr = base_addr + i
+                        # Avoid duplicates
+                        dup = False
+                        for f in functions:
+                            if f["offset"] == i:
+                                dup = True
+                                break
+                        if not dup:
+                            functions.append({
+                                "address": f"0x{addr:08x}",
+                                "offset": i,
+                                "stack_size": stack_size,
+                                "pattern": "sd_ra_with_addi_sp"
+                            })
+    
+    # Pattern 5: Jump/return instructions that might indicate function boundaries
+    # ret = jalr zero, 0(ra) = 0x00008067
+    # jr ra = jalr zero, 0(ra) (same)
+    for i in range(0, len(data) - 4, 2):
+        if i + 4 <= len(data):
+            inst = struct.unpack('<I', data[i:i+4])[0]
+            if inst == 0x00008067:  # ret / jalr zero, 0(ra)
+                addr = base_addr + i
+                functions.append({
+                    "address": f"0x{addr:08x}",
+                    "offset": i,
+                    "pattern": "return_instruction"
+                })
+    
+    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_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)
+    
+    # Sort by address
+    unique.sort(key=lambda x: x["offset"])
+    
+    print(f"Unique functions: {len(unique)}")
+    
+    # Save results
+    with open(f"{OUTPUT_DIR}/found_functions_v2.json", "w") as f:
+        json.dump(unique, f, indent=2)
+    
+    with open(f"{OUTPUT_DIR}/function_addresses_v2.txt", "w") as f:
+        for func in unique:
+            ss = func.get("stack_size", "?")
+            f.write(f"{func['address']}  stack_size={ss}  pattern={func['pattern']}\n")
+    
+    print(f"Results saved to {OUTPUT_DIR}/found_functions_v2.json")
+    
+    # Show all
+    for func in unique:
+        ss = func.get("stack_size", "?")
+        print(f"  {func['address']}  stack_size={ss}  pattern={func['pattern']}")
+
+if __name__ == "__main__":
+    main()