target/ppc: Implement core timebase state machine and TFMR
[qemu/ar7.git] / scripts / oss-fuzz / minimize_qtest_trace.py
blobd1f3990c16a7c9f72821236b37cf3d9a2c35fcfc
1 #!/usr/bin/env python3
2 # -*- coding: utf-8 -*-
4 """
5 This takes a crashing qtest trace and tries to remove superfluous operations
6 """
8 import sys
9 import os
10 import subprocess
11 import time
12 import struct
14 QEMU_ARGS = None
15 QEMU_PATH = None
16 TIMEOUT = 5
17 CRASH_TOKEN = None
19 # Minimization levels
20 M1 = False # try removing IO commands iteratively
21 M2 = False # try setting bits in operand of write/out to zero
23 write_suffix_lookup = {"b": (1, "B"),
24 "w": (2, "H"),
25 "l": (4, "L"),
26 "q": (8, "Q")}
28 def usage():
29 sys.exit("""\
30 Usage:
32 QEMU_PATH="/path/to/qemu" QEMU_ARGS="args" {} [Options] input_trace output_trace
34 By default, will try to use the second-to-last line in the output to identify
35 whether the crash occred. Optionally, manually set a string that idenitifes the
36 crash by setting CRASH_TOKEN=
38 Options:
40 -M1: enable a loop around the remove minimizer, which may help decrease some
41 timing dependent instructions. Off by default.
42 -M2: try setting bits in operand of write/out to zero. Off by default.
44 """.format((sys.argv[0])))
46 deduplication_note = """\n\
47 Note: While trimming the input, sometimes the mutated trace triggers a different
48 type crash but indicates the same bug. Under this situation, our minimizer is
49 incapable of recognizing and stopped from removing it. In the future, we may
50 use a more sophisticated crash case deduplication method.
51 \n"""
53 def check_if_trace_crashes(trace, path):
54 with open(path, "w") as tracefile:
55 tracefile.write("".join(trace))
57 rc = subprocess.Popen("timeout -s 9 {timeout}s {qemu_path} {qemu_args} 2>&1\
58 < {trace_path}".format(timeout=TIMEOUT,
59 qemu_path=QEMU_PATH,
60 qemu_args=QEMU_ARGS,
61 trace_path=path),
62 shell=True,
63 stdin=subprocess.PIPE,
64 stdout=subprocess.PIPE,
65 encoding="utf-8")
66 global CRASH_TOKEN
67 if CRASH_TOKEN is None:
68 try:
69 outs, _ = rc.communicate(timeout=5)
70 CRASH_TOKEN = " ".join(outs.splitlines()[-2].split()[0:3])
71 except subprocess.TimeoutExpired:
72 print("subprocess.TimeoutExpired")
73 return False
74 print("Identifying Crashes by this string: {}".format(CRASH_TOKEN))
75 global deduplication_note
76 print(deduplication_note)
77 return True
79 for line in iter(rc.stdout.readline, ""):
80 if "CLOSED" in line:
81 return False
82 if CRASH_TOKEN in line:
83 return True
85 print("\nWarning:")
86 print(" There is no 'CLOSED'or CRASH_TOKEN in the stdout of subprocess.")
87 print(" Usually this indicates a different type of crash.\n")
88 return False
91 # If previous write commands write the same length of data at the same
92 # interval, we view it as a hint.
93 def split_write_hint(newtrace, i):
94 HINT_LEN = 3 # > 2
95 if i <=(HINT_LEN-1):
96 return None
98 #find previous continuous write traces
99 k = 0
100 l = i-1
101 writes = []
102 while (k != HINT_LEN and l >= 0):
103 if newtrace[l].startswith("write "):
104 writes.append(newtrace[l])
105 k += 1
106 l -= 1
107 elif newtrace[l] == "":
108 l -= 1
109 else:
110 return None
111 if k != HINT_LEN:
112 return None
114 length = int(writes[0].split()[2], 16)
115 for j in range(1, HINT_LEN):
116 if length != int(writes[j].split()[2], 16):
117 return None
119 step = int(writes[0].split()[1], 16) - int(writes[1].split()[1], 16)
120 for j in range(1, HINT_LEN-1):
121 if step != int(writes[j].split()[1], 16) - \
122 int(writes[j+1].split()[1], 16):
123 return None
125 return (int(writes[0].split()[1], 16)+step, length)
128 def remove_lines(newtrace, outpath):
129 remove_step = 1
130 i = 0
131 while i < len(newtrace):
132 # 1.) Try to remove lines completely and reproduce the crash.
133 # If it works, we're done.
134 if (i+remove_step) >= len(newtrace):
135 remove_step = 1
136 prior = newtrace[i:i+remove_step]
137 for j in range(i, i+remove_step):
138 newtrace[j] = ""
139 print("Removing {lines} ...\n".format(lines=prior))
140 if check_if_trace_crashes(newtrace, outpath):
141 i += remove_step
142 # Double the number of lines to remove for next round
143 remove_step *= 2
144 continue
145 # Failed to remove multiple IOs, fast recovery
146 if remove_step > 1:
147 for j in range(i, i+remove_step):
148 newtrace[j] = prior[j-i]
149 remove_step = 1
150 continue
151 newtrace[i] = prior[0] # remove_step = 1
153 # 2.) Try to replace write{bwlq} commands with a write addr, len
154 # command. Since this can require swapping endianness, try both LE and
155 # BE options. We do this, so we can "trim" the writes in (3)
157 if (newtrace[i].startswith("write") and not
158 newtrace[i].startswith("write ")):
159 suffix = newtrace[i].split()[0][-1]
160 assert(suffix in write_suffix_lookup)
161 addr = int(newtrace[i].split()[1], 16)
162 value = int(newtrace[i].split()[2], 16)
163 for endianness in ['<', '>']:
164 data = struct.pack("{end}{size}".format(end=endianness,
165 size=write_suffix_lookup[suffix][1]),
166 value)
167 newtrace[i] = "write {addr} {size} 0x{data}\n".format(
168 addr=hex(addr),
169 size=hex(write_suffix_lookup[suffix][0]),
170 data=data.hex())
171 if(check_if_trace_crashes(newtrace, outpath)):
172 break
173 else:
174 newtrace[i] = prior[0]
176 # 3.) If it is a qtest write command: write addr len data, try to split
177 # it into two separate write commands. If splitting the data operand
178 # from length/2^n bytes to the left does not work, try to move the pivot
179 # to the right side, then add one to n, until length/2^n == 0. The idea
180 # is to prune unnecessary bytes from long writes, while accommodating
181 # arbitrary MemoryRegion access sizes and alignments.
183 # This algorithm will fail under some rare situations.
184 # e.g., xxxxxxxxxuxxxxxx (u is the unnecessary byte)
186 if newtrace[i].startswith("write "):
187 addr = int(newtrace[i].split()[1], 16)
188 length = int(newtrace[i].split()[2], 16)
189 data = newtrace[i].split()[3][2:]
190 if length > 1:
192 # Can we get a hint from previous writes?
193 hint = split_write_hint(newtrace, i)
194 if hint is not None:
195 hint_addr = hint[0]
196 hint_len = hint[1]
197 if hint_addr >= addr and hint_addr+hint_len <= addr+length:
198 newtrace[i] = "write {addr} {size} 0x{data}\n".format(
199 addr=hex(hint_addr),
200 size=hex(hint_len),
201 data=data[(hint_addr-addr)*2:\
202 (hint_addr-addr)*2+hint_len*2])
203 if check_if_trace_crashes(newtrace, outpath):
204 # next round
205 i += 1
206 continue
207 newtrace[i] = prior[0]
209 # Try splitting it using a binary approach
210 leftlength = int(length/2)
211 rightlength = length - leftlength
212 newtrace.insert(i+1, "")
213 power = 1
214 while leftlength > 0:
215 newtrace[i] = "write {addr} {size} 0x{data}\n".format(
216 addr=hex(addr),
217 size=hex(leftlength),
218 data=data[:leftlength*2])
219 newtrace[i+1] = "write {addr} {size} 0x{data}\n".format(
220 addr=hex(addr+leftlength),
221 size=hex(rightlength),
222 data=data[leftlength*2:])
223 if check_if_trace_crashes(newtrace, outpath):
224 break
225 # move the pivot to right side
226 if leftlength < rightlength:
227 rightlength, leftlength = leftlength, rightlength
228 continue
229 power += 1
230 leftlength = int(length/pow(2, power))
231 rightlength = length - leftlength
232 if check_if_trace_crashes(newtrace, outpath):
233 i -= 1
234 else:
235 newtrace[i] = prior[0]
236 del newtrace[i+1]
237 i += 1
240 def clear_bits(newtrace, outpath):
241 # try setting bits in operands of out/write to zero
242 i = 0
243 while i < len(newtrace):
244 if (not newtrace[i].startswith("write ") and not
245 newtrace[i].startswith("out")):
246 i += 1
247 continue
248 # write ADDR SIZE DATA
249 # outx ADDR VALUE
250 print("\nzero setting bits: {}".format(newtrace[i]))
252 prefix = " ".join(newtrace[i].split()[:-1])
253 data = newtrace[i].split()[-1]
254 data_bin = bin(int(data, 16))
255 data_bin_list = list(data_bin)
257 for j in range(2, len(data_bin_list)):
258 prior = newtrace[i]
259 if (data_bin_list[j] == '1'):
260 data_bin_list[j] = '0'
261 data_try = hex(int("".join(data_bin_list), 2))
262 # It seems qtest only accepts padded hex-values.
263 if len(data_try) % 2 == 1:
264 data_try = data_try[:2] + "0" + data_try[2:]
266 newtrace[i] = "{prefix} {data_try}\n".format(
267 prefix=prefix,
268 data_try=data_try)
270 if not check_if_trace_crashes(newtrace, outpath):
271 data_bin_list[j] = '1'
272 newtrace[i] = prior
273 i += 1
276 def minimize_trace(inpath, outpath):
277 global TIMEOUT
278 with open(inpath) as f:
279 trace = f.readlines()
280 start = time.time()
281 if not check_if_trace_crashes(trace, outpath):
282 sys.exit("The input qtest trace didn't cause a crash...")
283 end = time.time()
284 print("Crashed in {} seconds".format(end-start))
285 TIMEOUT = (end-start)*5
286 print("Setting the timeout for {} seconds".format(TIMEOUT))
288 newtrace = trace[:]
289 global M1, M2
291 # remove lines
292 old_len = len(newtrace) + 1
293 while(old_len > len(newtrace)):
294 old_len = len(newtrace)
295 print("trace length = ", old_len)
296 remove_lines(newtrace, outpath)
297 if not M1 and not M2:
298 break
299 newtrace = list(filter(lambda s: s != "", newtrace))
300 assert(check_if_trace_crashes(newtrace, outpath))
302 # set bits to zero
303 if M2:
304 clear_bits(newtrace, outpath)
305 assert(check_if_trace_crashes(newtrace, outpath))
308 if __name__ == '__main__':
309 if len(sys.argv) < 3:
310 usage()
311 if "-M1" in sys.argv:
312 M1 = True
313 if "-M2" in sys.argv:
314 M2 = True
315 QEMU_PATH = os.getenv("QEMU_PATH")
316 QEMU_ARGS = os.getenv("QEMU_ARGS")
317 if QEMU_PATH is None or QEMU_ARGS is None:
318 usage()
319 # if "accel" not in QEMU_ARGS:
320 # QEMU_ARGS += " -accel qtest"
321 CRASH_TOKEN = os.getenv("CRASH_TOKEN")
322 QEMU_ARGS += " -qtest stdio -monitor none -serial none "
323 minimize_trace(sys.argv[-2], sys.argv[-1])