1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
|
# Pretty-printers for SpiderMonkey jsvals.
import gdb
import gdb.types
import mozilla.prettyprinters
from mozilla.prettyprinters import pretty_printer, ptr_pretty_printer
# Forget any printers from previous loads of this module.
mozilla.prettyprinters.clear_module_printers(__name__)
# Summary of the JS::Value (also known as jsval) type:
#
# Viewed abstractly, JS::Value is a 64-bit discriminated union, with
# JSString *, JSObject *, IEEE 64-bit floating-point, and 32-bit integer
# branches (and a few others). (It is not actually a C++ union;
# 'discriminated union' just describes the overall effect.) Note that
# JS::Value is always 64 bits long, even on 32-bit architectures.
#
# The ECMAScript standard specifies that ECMAScript numbers are IEEE 64-bit
# floating-point values. A JS::Value can represent any JavaScript number
# value directly, without referring to additional storage, or represent an
# object, string, or other ECMAScript value, and remember which type it is.
# This may seem surprising: how can a 64-bit type hold all the 64-bit IEEE
# values, and still distinguish them from objects, strings, and so on,
# which have 64-bit addresses?
#
# This is possible for two reasons:
#
# - First, ECMAScript implementations aren't required to distinguish all
# the values the IEEE 64-bit format can represent. The IEEE format
# specifies many bitstrings representing NaN values, while ECMAScript
# requires only a single NaN value. This means we can use one IEEE NaN to
# represent ECMAScript's NaN, and use all the other IEEE NaNs to
# represent the other ECMAScript values.
#
# (IEEE says that any floating-point value whose 11-bit exponent field is
# 0x7ff (all ones) and whose 52-bit fraction field is non-zero is a NaN.
# So as long as we ensure the fraction field is non-zero, and save a NaN
# for ECMAScript, we have 2^52 values to play with.)
#
# - Second, on the only 64-bit architecture we support, x86_64, only the
# lower 48 bits of an address are significant. The upper sixteen bits are
# required to be the sign-extension of bit 48. Furthermore, user code
# always runs in "positive addresses": those in which bit 48 is zero. So
# we only actually need 47 bits to store all possible object or string
# addresses, even on 64-bit platforms.
#
# With a 52-bit fraction field, and 47 bits needed for the 'payload', we
# have up to five bits left to store a 'tag' value, to indicate which
# branch of our discriminated union is live.
#
# Thus, we define JS::Value representations in terms of the IEEE 64-bit
# floating-point format:
#
# - Any bitstring that IEEE calls a number or an infinity represents that
# ECMAScript number.
#
# - Any bitstring that IEEE calls a NaN represents either an ECMAScript NaN
# or a non-number ECMAScript value, as determined by a tag field stored
# towards the most significant end of the fraction field (exactly where
# depends on the address size). If the tag field indicates that this
# JS::Value is an object, the fraction field's least significant end
# holds the address of a JSObject; if a string, the address of a
# JSString; and so on.
#
# On the only 64-bit platform we support, x86_64, only the lower 48 bits of
# an address are significant, and only those values whose top bit is zero
# are used for user-space addresses. This means that x86_64 addresses are
# effectively 47 bits long, and thus fit nicely in the available portion of
# the fraction field.
#
#
# In detail:
#
# - jsval (Value.h) is a typedef for JS::Value.
#
# - JS::Value (Value.h) is a class with a lot of methods and a single data
# member, of type jsval_layout.
#
# - jsval_layout (Value.h) is a helper type for picking apart values. This
# is always 64 bits long, with a variant for each address size (32 bits
# or 64 bits) and endianness (little- or big-endian).
#
# jsval_layout is a union with 'asBits', 'asDouble', and 'asPtr'
# branches, and an 's' branch, which is a struct that tries to break out
# the bitfields a little for the non-double types. On 64-bit machines,
# jsval_layout also has an 'asUIntPtr' branch.
#
# On 32-bit platforms, the 's' structure has a 'tag' member at the
# exponent end of the 's' struct, and a 'payload' union at the mantissa
# end. The 'payload' union's branches are things like JSString *,
# JSObject *, and so on: the natural representations of the tags.
#
# On 64-bit platforms, the payload is 47 bits long; since C++ doesn't let
# us declare bitfields that hold unions, we can't break it down so
# neatly. In this case, we apply bit-shifting tricks to the 'asBits'
# branch of the union to extract the tag.
class Box(object):
def __init__(self, asBits, jtc):
self.asBits = asBits
self.jtc = jtc
# jsval_layout::asBits is uint64, but somebody botches the sign bit, even
# though Python integers are arbitrary precision.
if self.asBits < 0:
self.asBits = self.asBits + (1 << 64)
# Return this value's type tag.
def tag(self): raise NotImplementedError
# Return this value as a 32-bit integer, double, or address.
def as_uint32(self): raise NotImplementedError
def as_double(self): raise NotImplementedError
def as_address(self): raise NotImplementedError
# Packed non-number boxing --- the format used on x86_64. It would be nice to simply
# call JSVAL_TO_INT, etc. here, but the debugger is likely to see many jsvals, and
# doing several inferior calls for each one seems like a bad idea.
class Punbox(Box):
FULL_WIDTH = 64
TAG_SHIFT = 47
PAYLOAD_MASK = (1 << TAG_SHIFT) - 1
TAG_MASK = (1 << (FULL_WIDTH - TAG_SHIFT)) - 1
TAG_MAX_DOUBLE = 0x1fff0
TAG_TYPE_MASK = 0x0000f
def tag(self):
tag = self.asBits >> Punbox.TAG_SHIFT
if tag <= Punbox.TAG_MAX_DOUBLE:
return self.jtc.DOUBLE
else:
return tag & Punbox.TAG_TYPE_MASK
def as_uint32(self): return int(self.asBits & ((1 << 32) - 1))
def as_address(self): return gdb.Value(self.asBits & Punbox.PAYLOAD_MASK)
class Nunbox(Box):
TAG_SHIFT = 32
TAG_CLEAR = 0xffff0000
PAYLOAD_MASK = 0xffffffff
TAG_TYPE_MASK = 0x0000000f
def tag(self):
tag = self.asBits >> Nunbox.TAG_SHIFT
if tag < Nunbox.TAG_CLEAR:
return self.jtc.DOUBLE
return tag & Nunbox.TAG_TYPE_MASK
def as_uint32(self): return int(self.asBits & Nunbox.PAYLOAD_MASK)
def as_address(self): return gdb.Value(self.asBits & Nunbox.PAYLOAD_MASK)
# Cache information about the jsval type for this objfile.
class jsvalTypeCache(object):
def __init__(self, cache):
# Capture the tag values.
d = gdb.types.make_enum_dict(gdb.lookup_type('JSValueType'))
self.DOUBLE = d['JSVAL_TYPE_DOUBLE']
self.INT32 = d['JSVAL_TYPE_INT32']
self.UNDEFINED = d['JSVAL_TYPE_UNDEFINED']
self.BOOLEAN = d['JSVAL_TYPE_BOOLEAN']
self.MAGIC = d['JSVAL_TYPE_MAGIC']
self.STRING = d['JSVAL_TYPE_STRING']
self.SYMBOL = d['JSVAL_TYPE_SYMBOL']
self.NULL = d['JSVAL_TYPE_NULL']
self.OBJECT = d['JSVAL_TYPE_OBJECT']
# Let self.magic_names be an array whose i'th element is the name of
# the i'th magic value.
d = gdb.types.make_enum_dict(gdb.lookup_type('JSWhyMagic'))
self.magic_names = list(range(max(d.values()) + 1))
for (k,v) in d.items(): self.magic_names[v] = k
# Choose an unboxing scheme for this architecture.
self.boxer = Punbox if cache.void_ptr_t.sizeof == 8 else Nunbox
@pretty_printer('jsval_layout')
class jsval_layout(object):
def __init__(self, value, cache):
# Save the generic typecache, and create our own, if we haven't already.
self.cache = cache
if not cache.mod_jsval:
cache.mod_jsval = jsvalTypeCache(cache)
self.jtc = cache.mod_jsval
self.value = value
self.box = self.jtc.boxer(value['asBits'], self.jtc)
def to_string(self):
tag = self.box.tag()
if tag == self.jtc.INT32:
value = self.box.as_uint32()
signbit = 1 << 31
value = (value ^ signbit) - signbit
elif tag == self.jtc.UNDEFINED:
return 'JSVAL_VOID'
elif tag == self.jtc.BOOLEAN:
return 'JSVAL_TRUE' if self.box.as_uint32() else 'JSVAL_FALSE'
elif tag == self.jtc.MAGIC:
value = self.box.as_uint32()
if 0 <= value and value < len(self.jtc.magic_names):
return '$jsmagic(%s)' % (self.jtc.magic_names[value],)
else:
return '$jsmagic(%d)' % (value,)
elif tag == self.jtc.STRING:
value = self.box.as_address().cast(self.cache.JSString_ptr_t)
elif tag == self.jtc.SYMBOL:
value = self.box.as_address().cast(self.cache.JSSymbol_ptr_t)
elif tag == self.jtc.NULL:
return 'JSVAL_NULL'
elif tag == self.jtc.OBJECT:
value = self.box.as_address().cast(self.cache.JSObject_ptr_t)
elif tag == self.jtc.DOUBLE:
value = self.value['asDouble']
else:
return '$jsval(unrecognized!)'
return '$jsval(%s)' % (value,)
@pretty_printer('JS::Value')
class JSValue(object):
def __new__(cls, value, cache):
return jsval_layout(value['data'], cache)
|
