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
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
|
use crate::core::*;
use crate::cruby::*;
use crate::yjit::yjit_enabled_p;
#[cfg(feature = "disasm")]
use crate::asm::CodeBlock;
#[cfg(feature = "disasm")]
use crate::codegen::CodePtr;
#[cfg(feature = "disasm")]
use crate::options::DumpDisasm;
#[cfg(feature = "disasm")]
use std::fmt::Write;
/// Primitive called in yjit.rb
/// Produce a string representing the disassembly for an ISEQ
#[no_mangle]
pub extern "C" fn rb_yjit_disasm_iseq(_ec: EcPtr, _ruby_self: VALUE, iseqw: VALUE) -> VALUE {
#[cfg(not(feature = "disasm"))]
{
let _ = iseqw;
return Qnil;
}
#[cfg(feature = "disasm")]
{
// TODO:
//if unsafe { CLASS_OF(iseqw) != rb_cISeq } {
// return Qnil;
//}
if !yjit_enabled_p() {
return Qnil;
}
// Get the iseq pointer from the wrapper
let iseq = unsafe { rb_iseqw_to_iseq(iseqw) };
// This will truncate disassembly of methods with 10k+ bytecodes.
// That's a good thing - this prints to console.
let out_string = with_vm_lock(src_loc!(), || disasm_iseq_insn_range(iseq, 0, 9999));
return rust_str_to_ruby(&out_string);
}
}
/// Only call while holding the VM lock.
#[cfg(feature = "disasm")]
pub fn disasm_iseq_insn_range(iseq: IseqPtr, start_idx: u16, end_idx: u16) -> String {
let mut out = String::from("");
// Get a list of block versions generated for this iseq
let block_list = get_or_create_iseq_block_list(iseq);
let mut block_list: Vec<&Block> = block_list.into_iter().map(|blockref| {
// SAFETY: We have the VM lock here and all the blocks on iseqs are valid.
unsafe { blockref.as_ref() }
}).collect();
// Get a list of codeblocks relevant to this iseq
let global_cb = crate::codegen::CodegenGlobals::get_inline_cb();
// Sort the blocks by increasing start addresses
block_list.sort_by(|a, b| {
use std::cmp::Ordering;
// Get the start addresses for each block
let addr_a = a.get_start_addr().raw_ptr();
let addr_b = b.get_start_addr().raw_ptr();
if addr_a < addr_b {
Ordering::Less
} else if addr_a == addr_b {
Ordering::Equal
} else {
Ordering::Greater
}
});
// Compute total code size in bytes for all blocks in the function
let mut total_code_size = 0;
for blockref in &block_list {
total_code_size += blockref.code_size();
}
writeln!(out, "NUM BLOCK VERSIONS: {}", block_list.len()).unwrap();
writeln!(out, "TOTAL INLINE CODE SIZE: {} bytes", total_code_size).unwrap();
// For each block, sorted by increasing start address
for (block_idx, block) in block_list.iter().enumerate() {
let blockid = block.get_blockid();
if blockid.idx >= start_idx && blockid.idx < end_idx {
let end_idx = block.get_end_idx();
let start_addr = block.get_start_addr();
let end_addr = block.get_end_addr();
let code_size = block.code_size();
// Write some info about the current block
let blockid_idx = blockid.idx;
let block_ident = format!(
"BLOCK {}/{}, ISEQ RANGE [{},{}), {} bytes ",
block_idx + 1,
block_list.len(),
blockid_idx,
end_idx,
code_size
);
writeln!(out, "== {:=<60}", block_ident).unwrap();
// Disassemble the instructions
for (start_addr, end_addr) in global_cb.writable_addrs(start_addr, end_addr) {
out.push_str(&disasm_addr_range(global_cb, start_addr, end_addr));
writeln!(out).unwrap();
}
// If this is not the last block
if block_idx < block_list.len() - 1 {
// Compute the size of the gap between this block and the next
let next_block = block_list[block_idx + 1];
let next_start_addr = next_block.get_start_addr();
let gap_size = next_start_addr.into_usize() - end_addr.into_usize();
// Log the size of the gap between the blocks if nonzero
if gap_size > 0 {
writeln!(out, "... {} byte gap ...", gap_size).unwrap();
}
}
}
}
return out;
}
#[cfg(feature = "disasm")]
pub fn dump_disasm_addr_range(cb: &CodeBlock, start_addr: CodePtr, end_addr: CodePtr, dump_disasm: &DumpDisasm) {
use std::fs::File;
use std::io::Write;
for (start_addr, end_addr) in cb.writable_addrs(start_addr, end_addr) {
let disasm = disasm_addr_range(cb, start_addr, end_addr);
if disasm.len() > 0 {
match dump_disasm {
DumpDisasm::Stdout => println!("{disasm}"),
DumpDisasm::File(path) => {
let mut f = File::options().create(true).append(true).open(path).unwrap();
f.write_all(disasm.as_bytes()).unwrap();
}
};
}
}
}
#[cfg(feature = "disasm")]
pub fn disasm_addr_range(cb: &CodeBlock, start_addr: usize, end_addr: usize) -> String {
let mut out = String::from("");
// Initialize capstone
use capstone::prelude::*;
#[cfg(target_arch = "x86_64")]
let mut cs = Capstone::new()
.x86()
.mode(arch::x86::ArchMode::Mode64)
.syntax(arch::x86::ArchSyntax::Intel)
.build()
.unwrap();
#[cfg(target_arch = "aarch64")]
let mut cs = Capstone::new()
.arm64()
.mode(arch::arm64::ArchMode::Arm)
.detail(true)
.build()
.unwrap();
cs.set_skipdata(true).unwrap();
// Disassemble the instructions
let code_size = end_addr - start_addr;
let code_slice = unsafe { std::slice::from_raw_parts(start_addr as _, code_size) };
// Stabilize output for cargo test
#[cfg(test)]
let start_addr = 0;
let insns = cs.disasm_all(code_slice, start_addr as u64).unwrap();
// For each instruction in this block
for insn in insns.as_ref() {
// Comments for this block
if let Some(comment_list) = cb.comments_at(insn.address() as usize) {
for comment in comment_list {
if cb.outlined {
write!(&mut out, "\x1b[34m").unwrap(); // Make outlined code blue
}
writeln!(&mut out, " \x1b[1m# {comment}\x1b[22m").unwrap(); // Make comments bold
}
}
if cb.outlined {
write!(&mut out, "\x1b[34m").unwrap(); // Make outlined code blue
}
writeln!(&mut out, " {insn}").unwrap();
if cb.outlined {
write!(&mut out, "\x1b[0m").unwrap(); // Disable blue
}
}
return out;
}
/// Assert that CodeBlock has the code specified with hex. In addition, if tested with
/// `cargo test --all-features`, it also checks it generates the specified disasm.
#[cfg(test)]
macro_rules! assert_disasm {
($cb:expr, $hex:expr, $disasm:expr) => {
#[cfg(feature = "disasm")]
{
let disasm = disasm_addr_range(
&$cb,
$cb.get_ptr(0).raw_ptr() as usize,
$cb.get_write_ptr().raw_ptr() as usize,
);
assert_eq!(unindent(&disasm, false), unindent(&$disasm, true));
}
assert_eq!(format!("{:x}", $cb), $hex);
};
}
#[cfg(test)]
pub(crate) use assert_disasm;
/// Remove the minimum indent from every line, skipping the first line if `skip_first`.
#[cfg(all(feature = "disasm", test))]
pub fn unindent(string: &str, trim_lines: bool) -> String {
fn split_lines(string: &str) -> Vec<String> {
let mut result: Vec<String> = vec![];
let mut buf: Vec<u8> = vec![];
for byte in string.as_bytes().iter() {
buf.push(*byte);
if *byte == b'\n' {
result.push(String::from_utf8(buf).unwrap());
buf = vec![];
}
}
if !buf.is_empty() {
result.push(String::from_utf8(buf).unwrap());
}
result
}
// Break up a string into multiple lines
let mut lines = split_lines(string);
if trim_lines { // raw string literals come with extra lines
lines.remove(0);
lines.remove(lines.len() - 1);
}
// Count the minimum number of spaces
let spaces = lines.iter().filter_map(|line| {
for (i, ch) in line.as_bytes().iter().enumerate() {
if *ch != b' ' {
return Some(i);
}
}
None
}).min().unwrap_or(0);
// Join lines, removing spaces
let mut unindented: Vec<u8> = vec![];
for line in lines.iter() {
if line.len() > spaces {
unindented.extend_from_slice(&line.as_bytes()[spaces..]);
} else {
unindented.extend_from_slice(&line.as_bytes());
}
}
String::from_utf8(unindented).unwrap()
}
/// Primitive called in yjit.rb
/// Produce a list of instructions compiled for an isew
#[no_mangle]
pub extern "C" fn rb_yjit_insns_compiled(_ec: EcPtr, _ruby_self: VALUE, iseqw: VALUE) -> VALUE {
{
// TODO:
//if unsafe { CLASS_OF(iseqw) != rb_cISeq } {
// return Qnil;
//}
if !yjit_enabled_p() {
return Qnil;
}
// Get the iseq pointer from the wrapper
let iseq = unsafe { rb_iseqw_to_iseq(iseqw) };
// Get the list of instructions compiled
let insn_vec = insns_compiled(iseq);
unsafe {
let insn_ary = rb_ary_new_capa((insn_vec.len() * 2) as i64);
// For each instruction compiled
for idx in 0..insn_vec.len() {
let op_name = &insn_vec[idx].0;
let insn_idx = insn_vec[idx].1;
let op_sym = rust_str_to_sym(&op_name);
// Store the instruction index and opcode symbol
rb_ary_store(
insn_ary,
(2 * idx + 0) as i64,
VALUE::fixnum_from_usize(insn_idx as usize),
);
rb_ary_store(insn_ary, (2 * idx + 1) as i64, op_sym);
}
insn_ary
}
}
}
fn insns_compiled(iseq: IseqPtr) -> Vec<(String, u16)> {
let mut insn_vec = Vec::new();
// Get a list of block versions generated for this iseq
let block_list = get_or_create_iseq_block_list(iseq);
// For each block associated with this iseq
for blockref in &block_list {
// SAFETY: Called as part of a Ruby method, which ensures the graph is
// well connected for the given iseq.
let block = unsafe { blockref.as_ref() };
let start_idx = block.get_blockid().idx;
let end_idx = block.get_end_idx();
assert!(u32::from(end_idx) <= unsafe { get_iseq_encoded_size(iseq) });
// For each YARV instruction in the block
let mut insn_idx = start_idx;
while insn_idx < end_idx {
// Get the current pc and opcode
let pc = unsafe { rb_iseq_pc_at_idx(iseq, insn_idx.into()) };
// try_into() call below is unfortunate. Maybe pick i32 instead of usize for opcodes.
let opcode: usize = unsafe { rb_iseq_opcode_at_pc(iseq, pc) }
.try_into()
.unwrap();
// Get the mnemonic for this opcode
let op_name = insn_name(opcode);
// Add the instruction to the list
insn_vec.push((op_name, insn_idx));
// Move to the next instruction
insn_idx += insn_len(opcode) as u16;
}
}
return insn_vec;
}
|