output emulatable .elf

7 files changed, 220 insertions, 56 deletions

diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..e6c512b
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,2 @@
+*.elf
+__pycache__
diff --git a/README.md b/README.md
new file mode 100644
index 0000000..df03318
--- /dev/null
+++ b/README.md
@@ -0,0 +1,22 @@
+SubV
+====
+
+This is a wip clone of [SubX][mu] for the RISC-V RV31I base ISA.
+
+    $ ./test.py | ./elf.py > out.elf
+    $ ./qemu.sh out.elf
+
+Debugging
+---------
+
+You can hook gdb into `qemu` using the `-s` flag, and make it halt on start
+using the `-S` flag. `gdb` can be attached using `target remote localhost:1234`:
+
+    $ ./qemu.sh out.elf -S -s
+    # in another terminal
+    $ riscv32-elf-gdb -iex "target remote localhost:1234"
+    layout asm # show assembly trace
+    nexti      # step forward one instruction
+    c          # free-run forward
+
+[mu]: https://github.com/akkartik/mu
diff --git a/elf.py b/elf.py
new file mode 100755
index 0000000..8990c88
--- /dev/null
+++ b/elf.py
@@ -0,0 +1,105 @@
+#!/usr/bin/env python3
+import sys
+from subx import clean, classify, parse_segment, parse_instr, format_instr
+
+"""
+Takes a packed & surveyed hex stream (with section headers) and packs it into a
+simulatable ELF file that can be run as a kernel:
+
+    qemu-system-riscv32 -nographic -machine sifive_u -nographic -kernel out.elf
+
+sifive-u has lots of memory mapped peripherals, see
+https://github.com/qemu/qemu/blob/master/hw/riscv/sifive_u.c#L66-L83
+
+Among them are two memory-mapped virtual UARTs at 0x10010000 and 0x10011000.
+Writing to UART0+0x0 writes to qemu's output TTY, and reading from UART0+0x4
+reads from it. The MSB in UART0+0x4 is set if there is nothing to read.
+See full documentation here:
+https://sifive.cdn.prismic.io/sifive/4d063bf8-3ae6-4db6-9843-ee9076ebadf7_fe310-g000.pdf
+"""
+
+segments = []
+segment = None
+for line in sys.stdin:
+  line = clean(line)
+  type = classify(line)
+
+  if type == 'segment':
+    (name, addr) = parse_segment(line)
+    segment = { 'name': name, 'addr': addr, 'content': [] }
+    segments.append(segment)
+  else:
+    if segment == None:
+      raise ValueError("label or code outside of segment!")
+
+    if type == 'label':
+      raise NotImplementedError()
+    else:
+      instr = parse_instr(line)
+      segment['content'] += instr
+
+entrypoint = 0x80000000
+cursor = 0
+
+def w(b):
+  global cursor
+  cursor += len(b)
+  sys.stdout.buffer.write(b)
+
+def wi(num, size=2):
+  global cursor
+  cursor += size
+  sys.stdout.buffer.write(num.to_bytes(size, byteorder='little'))
+
+def padto(offset):
+  missing = offset - cursor
+  if missing < 0:
+    raise ValueError("already past offset!")
+  w(b'\x00' * missing) 
+
+def write_elf_header():
+  w(b"\x7fELF")     # ELF magic
+  wi(0x010101, 4)   # 32bit, little endian
+  wi(0, 8)          # reserved
+  wi(0x02)          # e_type
+  wi(0xf3)          # e_machine
+  wi(0x01, 4)       # e_version
+  wi(entrypoint, 4) # e_entry
+  wi(0x34, 4)       # e_phoff (Program Header offset)
+  wi(0x00, 4)       # e_shoff (Section Header offset, unused)
+  wi(0x0004, 4)     # e_flags
+  wi(0x34)          # e_ehsize
+  wi(0x20)          # e_phentsize
+  wi(len(segments)) # e_phnum
+  wi(0x28)          # e_shentsize
+  wi(0x0)           # e_shnum
+  wi(0x0)           # e_shstrndx
+
+def write_program_header(segment):
+  align = 0x1000
+  offset = (1 + cursor // align) * align
+  start = segment['addr']
+  size = len(segment['content'])
+  segment['offset'] = offset
+
+  if offset % align != start % align:
+    print("{:02x} {:02x}".format(offset, offset % align))
+    print("{:02x} {:02x}".format(start, start % align))
+    raise ValueError("improper alignment")
+
+  wi(0x1, 4)    # p_type
+  wi(offset, 4) # p_offset
+  wi(start, 4)  # p_vaddr
+  wi(start, 4)  # p_paddr
+  wi(size, 4)   # p_filesz
+  wi(size, 4)   # p_memsz
+  wi(0x5, 4)    # p_flags, 0x5=rx, 0x6=rw
+  wi(align, 4)  # p_align
+
+write_elf_header()
+for seg in segments:
+  write_program_header(seg)
+for seg in segments:
+  padto(seg['offset'])
+  for part in seg['content']:
+    wi(part[0], 1)
diff --git a/qemu.sh b/qemu.sh
new file mode 100755
index 0000000..7cc3bde
--- /dev/null
+++ b/qemu.sh
@@ -0,0 +1,2 @@
+#!/bin/sh
+qemu-system-riscv32 -nographic -machine sifive_u -nographic -kernel "$@"
diff --git a/riscv.py b/riscv.py
new file mode 100644
index 0000000..b2af4cc
--- /dev/null
+++ b/riscv.py
@@ -0,0 +1,31 @@
+def format_r(op, rd, r1, r2, funct3, funct7):
+  # opcode[7] rd[5] funct3[3] rs1[5] rs2[5] funct7[7]
+  b0 = (rd & 0x1) << 7 | op
+  b1 = (r1 & 0x1) << 7 | funct3 << 4 | rd >> 1
+  b2 = (r2 & 0xf) << 4 | r1 >> 1
+  b3 =     funct7 << 1 | r2 >> 4
+  return [(b0,), (b1,), (b2,), (b3,)]
+
+def format_i(op, rd, r1, imm12, funct3):
+  # opcode[7] rd[5] funct3[3] rs1[5] imm[12]
+  b0 =    (rd & 0x1) << 7 | op
+  b1 =    (r1 & 0x1) << 7 | funct3 << 4 | rd >> 1
+  b2 = (imm12 & 0xf) << 4 | r1 >> 1
+  b3 =  imm12 >> 4
+  return [(b0,), (b1,), (b2,), (b3,)]
+
+def format_s(op, r1, r2, imm12, funct3):
+  # opcode[7] rd[5] funct3[3] rs1[5] rs2[5] funct7[7]
+  b0 = (imm12 & 0x1) << 7 | op
+  b1 =    (r1 & 0x1) << 7 | funct3 << 4 | (imm12 & 0x1f) >> 1
+  b2 =    (r2 & 0xf) << 4 | r1 >> 1
+  b3 = (imm12 & 0xf0)     | r2 >> 4
+  return [(b0,), (b1,), (b2,), (b3,)]
+
+def format_u(op, rd, imm20):
+  # opcode[7] rd[5] imm31:12[20]
+  b0 =    (rd & 0x1) << 7 | op
+  b1 = (imm20 & 0xf) << 4 | rd >> 1
+  b2 = (imm20 >> 4) & 0xff
+  b3 =  imm20 >> 12
+  return [(b0,), (b1,), (b2,), (b3,)]
diff --git a/subx.py b/subx.py
new file mode 100644
index 0000000..53f72ce
--- /dev/null
+++ b/subx.py
@@ -0,0 +1,40 @@
+import re
+white = re.compile('[ \t\.\n]+')
+hex = re.compile('^(0x)?[0-9a-f]+$')
+
+def parse_part(part):
+  part = part.split('/')
+  if hex.match(part[0]):
+    part[0] = int(part[0], 16)
+  return tuple(part)
+
+def parse_instr(line):
+  parts = white.split(line)
+  parts = [parse_part(part) for part in parts if part != '']
+  return parts
+
+def parse_segment(line):
+  parts = white.split(line)
+  return (parts[1], int(parts[2], 16))
+
+def format_part(part):
+  if not isinstance(part[0], str):
+    part = ('{:02x}'.format(part[0]),) + part[1:]
+  return '/'.join(part)
+
+def format_instr(inst, comment=None):
+  packed = ' '.join(format_part(part) for part in inst)
+  if comment:
+    packed = packed + '  # ' + comment
+  return packed
+
+def clean(line):
+  return line.strip().split('#')[0]
+
+def classify(line):
+  if line.startswith('=='): # segment
+    return 'segment'
+  elif line.endswith(':'): # label
+    return 'label'
+  else:
+    return 'instr'
diff --git a/test.py b/test.py
index 525e9ca..ba6ffd6 100644..100755
--- a/test.py
+++ b/test.py
@@ -1,59 +1,21 @@
 #!/usr/bin/env python3
-
-import re
-white = re.compile('[ \t\.\n]+')
-hex = re.compile('^(0x)?[0-9a-f]+$')
-
-def format_r(op, rd, r1, r2, funct3, funct7):
-  # opcode[7] rd[5] funct3[3] rs1[5] rs2[5] funct7[7]
-  b0 = (rd & 0x1) << 7 | op
-  b1 = (r1 & 0x1) << 7 | funct3 << 4 | rd >> 1
-  b2 = (r2 & 0xf) << 4 | r1 >> 1
-  b3 = funct7 << 1 | r2 >> 4
-  return [(b3,), (b2,), (b1,), (b0,)]
-
-def parse_part(part):
-  part = part.split('/')
-  if hex.match(part[0]):
-    part[0] = int(part[0], 16)
-  return tuple(part)
-
-def parse_instr(line):
-  parts = white.split(line)
-  parts = [parse_part(part) for part in parts if part != '']
-  return parts
-
-def translate_instr(orig):
-  return orig
-
-def format_part(part):
-  if not isinstance(part[0], str):
-    part = ('0x{:02x}'.format(part[0]),) + part[1:]
-  return '/'.join(part)
-
-def format_instr(inst, comment=None):
-  packed = ' '.join(format_part(part) for part in inst)
-  if comment:
-    packed = packed + '  # ' + comment
-  return packed
+from riscv import format_u, format_i, format_s
+from subx import format_instr
 
 if __name__ == "__main__":
-  import sys
-
-  seg = 'code'
-  for line in sys.stdin:
-    line = line.rstrip()
-    if line.startswith('=='): # segment
-      print(line)
-      seg = line.split(' ')[1] 
-    elif line.endswith(':'): # label
-      print(line)
-    elif seg != 'code':
-      print(line)
-    else:
-      parts = parse_instr(line)
-      risc = translate_instr(parts)
-      print(format_instr(risc, line))
-
-  # emit_instr(format_r(0x33, 0x5, 0x6, 0x7, 0, 0))
-  # emit_instr(format_r(0x33, 0x5, 0x6, 0x7, 0, 0), "add eax, 2")
+  t0 = 0x5
+  t1 = 0x6
+  print("== code 0x80000000")
+  print(format_instr(format_u(0x37, t0, 0x10010), "lui t0, 0x10010"))
+  print(format_instr(format_i(0x13, t1, 0, 72, 0x0), "addi t1, x0, 72"))
+  print(format_instr(format_s(0x23, t0, t1, 0, 0x2), "sw t1, 0(t0)"))
+  print(format_instr(format_i(0x13, t1, 0, 101, 0x0), "addi t1, x0, 101"))
+  print(format_instr(format_s(0x23, t0, t1, 0, 0x2), "sw t1, 0(t0)"))
+  print(format_instr(format_i(0x13, t1, 0, 108, 0x0), "addi t1, x0, 108"))
+  print(format_instr(format_s(0x23, t0, t1, 0, 0x2), "sw t1, 0(t0)"))
+  print(format_instr(format_i(0x13, t1, 0, 108, 0x0), "addi t1, x0, 108"))
+  print(format_instr(format_s(0x23, t0, t1, 0, 0x2), "sw t1, 0(t0)"))
+  print(format_instr(format_i(0x13, t1, 0, 111, 0x0), "addi t1, x0, 111"))
+  print(format_instr(format_s(0x23, t0, t1, 0, 0x2), "sw t1, 0(t0)"))
+  print(format_instr(format_i(0x13, t1, 0, 10, 0x0), "addi t1, x0, 10"))
+  print(format_instr(format_s(0x23, t0, t1, 0, 0x2), "sw t1, 0(t0)"))