1 | //! This crate provides a cross-platform library and binary for translating addresses into |
2 | //! function names, file names and line numbers. Given an address in an executable or an |
3 | //! offset in a section of a relocatable object, it uses the debugging information to |
4 | //! figure out which file name and line number are associated with it. |
5 | //! |
6 | //! When used as a library, files must first be loaded using the |
7 | //! [`object`](https://github.com/gimli-rs/object) crate. |
8 | //! A context can then be created with [`Context::new`](./struct.Context.html#method.new). |
9 | //! The context caches some of the parsed information so that multiple lookups are |
10 | //! efficient. |
11 | //! Location information is obtained with |
12 | //! [`Context::find_location`](./struct.Context.html#method.find_location) or |
13 | //! [`Context::find_location_range`](./struct.Context.html#method.find_location_range). |
14 | //! Function information is obtained with |
15 | //! [`Context::find_frames`](./struct.Context.html#method.find_frames), which returns |
16 | //! a frame for each inline function. Each frame contains both name and location. |
17 | //! |
18 | //! The crate has an example CLI wrapper around the library which provides some of |
19 | //! the functionality of the `addr2line` command line tool distributed with [GNU |
20 | //! binutils](https://www.gnu.org/software/binutils/). |
21 | //! |
22 | //! Currently this library only provides information from the DWARF debugging information, |
23 | //! which is parsed using [`gimli`](https://github.com/gimli-rs/gimli). The example CLI |
24 | //! wrapper also uses symbol table information provided by the `object` crate. |
25 | #![deny (missing_docs)] |
26 | #![no_std ] |
27 | |
28 | #[cfg (feature = "std" )] |
29 | extern crate std; |
30 | |
31 | #[allow (unused_imports)] |
32 | #[macro_use ] |
33 | extern crate alloc; |
34 | |
35 | #[cfg (feature = "fallible-iterator" )] |
36 | pub extern crate fallible_iterator; |
37 | pub extern crate gimli; |
38 | #[cfg (feature = "object" )] |
39 | pub extern crate object; |
40 | |
41 | use alloc::borrow::Cow; |
42 | use alloc::boxed::Box; |
43 | #[cfg (feature = "object" )] |
44 | use alloc::rc::Rc; |
45 | use alloc::string::{String, ToString}; |
46 | use alloc::sync::Arc; |
47 | use alloc::vec::Vec; |
48 | |
49 | use core::cmp::{self, Ordering}; |
50 | use core::iter; |
51 | use core::marker::PhantomData; |
52 | use core::mem; |
53 | use core::num::NonZeroU64; |
54 | use core::ops::ControlFlow; |
55 | use core::u64; |
56 | |
57 | use crate::function::{Function, Functions, InlinedFunction}; |
58 | use crate::lazy::LazyCell; |
59 | |
60 | #[cfg (feature = "smallvec" )] |
61 | mod maybe_small { |
62 | pub type Vec<T> = smallvec::SmallVec<[T; 16]>; |
63 | pub type IntoIter<T> = smallvec::IntoIter<[T; 16]>; |
64 | } |
65 | #[cfg (not(feature = "smallvec" ))] |
66 | mod maybe_small { |
67 | pub type Vec<T> = alloc::vec::Vec<T>; |
68 | pub type IntoIter<T> = alloc::vec::IntoIter<T>; |
69 | } |
70 | |
71 | #[cfg (all(feature = "std" , feature = "object" , feature = "memmap2" ))] |
72 | /// A simple builtin split DWARF loader. |
73 | pub mod builtin_split_dwarf_loader; |
74 | mod function; |
75 | mod lazy; |
76 | |
77 | type Error = gimli::Error; |
78 | |
79 | #[derive (Debug, Clone, Copy, PartialEq, Eq)] |
80 | enum DebugFile { |
81 | Primary, |
82 | Supplementary, |
83 | Dwo, |
84 | } |
85 | |
86 | /// Operations that consult debug information may require additional files |
87 | /// to be loaded if split DWARF is being used. This enum returns the result |
88 | /// of the operation in the `Break` variant, or information about the split |
89 | /// DWARF that is required and a continuation to invoke once it is available |
90 | /// in the `Continue` variant. |
91 | /// |
92 | /// This enum is intended to be used in a loop like so: |
93 | /// ```no_run |
94 | /// # use addr2line::*; |
95 | /// # use std::sync::Arc; |
96 | /// # let ctx: Context<gimli::EndianRcSlice<gimli::RunTimeEndian>> = todo!(); |
97 | /// # let do_split_dwarf_load = |load: SplitDwarfLoad<gimli::EndianRcSlice<gimli::RunTimeEndian>>| -> Option<Arc<gimli::Dwarf<gimli::EndianRcSlice<gimli::RunTimeEndian>>>> { None }; |
98 | /// const ADDRESS: u64 = 0xdeadbeef; |
99 | /// let mut r = ctx.find_frames(ADDRESS); |
100 | /// let result = loop { |
101 | /// match r { |
102 | /// LookupResult::Output(result) => break result, |
103 | /// LookupResult::Load { load, continuation } => { |
104 | /// let dwo = do_split_dwarf_load(load); |
105 | /// r = continuation.resume(dwo); |
106 | /// } |
107 | /// } |
108 | /// }; |
109 | /// ``` |
110 | pub enum LookupResult<L: LookupContinuation> { |
111 | /// The lookup requires split DWARF data to be loaded. |
112 | Load { |
113 | /// The information needed to find the split DWARF data. |
114 | load: SplitDwarfLoad<<L as LookupContinuation>::Buf>, |
115 | /// The continuation to resume with the loaded split DWARF data. |
116 | continuation: L, |
117 | }, |
118 | /// The lookup has completed and produced an output. |
119 | Output(<L as LookupContinuation>::Output), |
120 | } |
121 | |
122 | /// This trait represents a partially complete operation that can be resumed |
123 | /// once a load of needed split DWARF data is completed or abandoned by the |
124 | /// API consumer. |
125 | pub trait LookupContinuation: Sized { |
126 | /// The final output of this operation. |
127 | type Output; |
128 | /// The type of reader used. |
129 | type Buf: gimli::Reader; |
130 | |
131 | /// Resumes the operation with the provided data. |
132 | /// |
133 | /// After the caller loads the split DWARF data required, call this |
134 | /// method to resume the operation. The return value of this method |
135 | /// indicates if the computation has completed or if further data is |
136 | /// required. |
137 | /// |
138 | /// If the additional data cannot be located, or the caller does not |
139 | /// support split DWARF, `resume(None)` can be used to continue the |
140 | /// operation with the data that is available. |
141 | fn resume(self, input: Option<Arc<gimli::Dwarf<Self::Buf>>>) -> LookupResult<Self>; |
142 | } |
143 | |
144 | impl<L: LookupContinuation> LookupResult<L> { |
145 | /// Callers that do not handle split DWARF can call `skip_all_loads` |
146 | /// to fast-forward to the end result. This result is produced with |
147 | /// the data that is available and may be less accurate than the |
148 | /// the results that would be produced if the caller did properly |
149 | /// support split DWARF. |
150 | pub fn skip_all_loads(mut self) -> L::Output { |
151 | loop { |
152 | self = match self { |
153 | LookupResult::Output(t) => return t, |
154 | LookupResult::Load { continuation, .. } => continuation.resume(None), |
155 | }; |
156 | } |
157 | } |
158 | |
159 | fn map<T, F: FnOnce(L::Output) -> T>(self, f: F) -> LookupResult<MappedLookup<T, L, F>> { |
160 | match self { |
161 | LookupResult::Output(t) => LookupResult::Output(f(t)), |
162 | LookupResult::Load { load, continuation } => LookupResult::Load { |
163 | load, |
164 | continuation: MappedLookup { |
165 | original: continuation, |
166 | mutator: f, |
167 | }, |
168 | }, |
169 | } |
170 | } |
171 | |
172 | fn unwrap(self) -> L::Output { |
173 | match self { |
174 | LookupResult::Output(t) => t, |
175 | LookupResult::Load { .. } => unreachable!("Internal API misuse" ), |
176 | } |
177 | } |
178 | } |
179 | |
180 | /// The state necessary to perform address to line translation. |
181 | /// |
182 | /// Constructing a `Context` is somewhat costly, so users should aim to reuse `Context`s |
183 | /// when performing lookups for many addresses in the same executable. |
184 | pub struct Context<R: gimli::Reader> { |
185 | sections: Arc<gimli::Dwarf<R>>, |
186 | unit_ranges: Box<[UnitRange]>, |
187 | units: Box<[ResUnit<R>]>, |
188 | sup_units: Box<[SupUnit<R>]>, |
189 | } |
190 | |
191 | /// The type of `Context` that supports the `new` method. |
192 | #[cfg (feature = "std-object" )] |
193 | pub type ObjectContext = Context<gimli::EndianRcSlice<gimli::RunTimeEndian>>; |
194 | |
195 | #[cfg (feature = "std-object" )] |
196 | impl Context<gimli::EndianRcSlice<gimli::RunTimeEndian>> { |
197 | /// Construct a new `Context`. |
198 | /// |
199 | /// The resulting `Context` uses `gimli::EndianRcSlice<gimli::RunTimeEndian>`. |
200 | /// This means it is not thread safe, has no lifetime constraints (since it copies |
201 | /// the input data), and works for any endianity. |
202 | /// |
203 | /// Performance sensitive applications may want to use `Context::from_dwarf` |
204 | /// with a more specialised `gimli::Reader` implementation. |
205 | #[inline ] |
206 | pub fn new<'data: 'file, 'file, O: object::Object<'data, 'file>>( |
207 | file: &'file O, |
208 | ) -> Result<Self, Error> { |
209 | Self::new_with_sup(file, None) |
210 | } |
211 | |
212 | /// Construct a new `Context`. |
213 | /// |
214 | /// Optionally also use a supplementary object file. |
215 | /// |
216 | /// The resulting `Context` uses `gimli::EndianRcSlice<gimli::RunTimeEndian>`. |
217 | /// This means it is not thread safe, has no lifetime constraints (since it copies |
218 | /// the input data), and works for any endianity. |
219 | /// |
220 | /// Performance sensitive applications may want to use `Context::from_dwarf` |
221 | /// with a more specialised `gimli::Reader` implementation. |
222 | pub fn new_with_sup<'data: 'file, 'file, O: object::Object<'data, 'file>>( |
223 | file: &'file O, |
224 | sup_file: Option<&'file O>, |
225 | ) -> Result<Self, Error> { |
226 | let endian = if file.is_little_endian() { |
227 | gimli::RunTimeEndian::Little |
228 | } else { |
229 | gimli::RunTimeEndian::Big |
230 | }; |
231 | |
232 | fn load_section<'data: 'file, 'file, O, Endian>( |
233 | id: gimli::SectionId, |
234 | file: &'file O, |
235 | endian: Endian, |
236 | ) -> Result<gimli::EndianRcSlice<Endian>, Error> |
237 | where |
238 | O: object::Object<'data, 'file>, |
239 | Endian: gimli::Endianity, |
240 | { |
241 | use object::ObjectSection; |
242 | |
243 | let data = file |
244 | .section_by_name(id.name()) |
245 | .and_then(|section| section.uncompressed_data().ok()) |
246 | .unwrap_or(Cow::Borrowed(&[])); |
247 | Ok(gimli::EndianRcSlice::new(Rc::from(&*data), endian)) |
248 | } |
249 | |
250 | let mut dwarf = gimli::Dwarf::load(|id| load_section(id, file, endian))?; |
251 | if let Some(sup_file) = sup_file { |
252 | dwarf.load_sup(|id| load_section(id, sup_file, endian))?; |
253 | } |
254 | Context::from_dwarf(dwarf) |
255 | } |
256 | } |
257 | |
258 | impl<R: gimli::Reader> Context<R> { |
259 | /// Construct a new `Context` from DWARF sections. |
260 | /// |
261 | /// This method does not support using a supplementary object file. |
262 | pub fn from_sections( |
263 | debug_abbrev: gimli::DebugAbbrev<R>, |
264 | debug_addr: gimli::DebugAddr<R>, |
265 | debug_aranges: gimli::DebugAranges<R>, |
266 | debug_info: gimli::DebugInfo<R>, |
267 | debug_line: gimli::DebugLine<R>, |
268 | debug_line_str: gimli::DebugLineStr<R>, |
269 | debug_ranges: gimli::DebugRanges<R>, |
270 | debug_rnglists: gimli::DebugRngLists<R>, |
271 | debug_str: gimli::DebugStr<R>, |
272 | debug_str_offsets: gimli::DebugStrOffsets<R>, |
273 | default_section: R, |
274 | ) -> Result<Self, Error> { |
275 | Self::from_dwarf(gimli::Dwarf { |
276 | debug_abbrev, |
277 | debug_addr, |
278 | debug_aranges, |
279 | debug_info, |
280 | debug_line, |
281 | debug_line_str, |
282 | debug_str, |
283 | debug_str_offsets, |
284 | debug_types: default_section.clone().into(), |
285 | locations: gimli::LocationLists::new( |
286 | default_section.clone().into(), |
287 | default_section.into(), |
288 | ), |
289 | ranges: gimli::RangeLists::new(debug_ranges, debug_rnglists), |
290 | file_type: gimli::DwarfFileType::Main, |
291 | sup: None, |
292 | abbreviations_cache: gimli::AbbreviationsCache::new(), |
293 | }) |
294 | } |
295 | |
296 | /// Construct a new `Context` from an existing [`gimli::Dwarf`] object. |
297 | #[inline ] |
298 | pub fn from_dwarf(sections: gimli::Dwarf<R>) -> Result<Context<R>, Error> { |
299 | let sections = Arc::new(sections); |
300 | let (unit_ranges, units) = Context::parse_units(§ions)?; |
301 | let sup_units = if let Some(sup) = sections.sup.as_ref() { |
302 | Context::parse_sup(sup)? |
303 | } else { |
304 | Vec::new() |
305 | }; |
306 | Ok(Context { |
307 | sections, |
308 | unit_ranges: unit_ranges.into_boxed_slice(), |
309 | units: units.into_boxed_slice(), |
310 | sup_units: sup_units.into_boxed_slice(), |
311 | }) |
312 | } |
313 | |
314 | /// Finds the CUs for the function address given. |
315 | /// |
316 | /// There might be multiple CUs whose range contains this address. |
317 | /// Weak symbols have shown up in the wild which cause this to happen |
318 | /// but otherwise this can happen if the CU has non-contiguous functions |
319 | /// but only reports a single range. |
320 | /// |
321 | /// Consequently we return an iterator for all CUs which may contain the |
322 | /// address, and the caller must check if there is actually a function or |
323 | /// location in the CU for that address. |
324 | fn find_units(&self, probe: u64) -> impl Iterator<Item = &ResUnit<R>> { |
325 | self.find_units_range(probe, probe + 1) |
326 | .map(|(unit, _range)| unit) |
327 | } |
328 | |
329 | /// Finds the CUs covering the range of addresses given. |
330 | /// |
331 | /// The range is [low, high) (ie, the upper bound is exclusive). This can return multiple |
332 | /// ranges for the same unit. |
333 | #[inline ] |
334 | fn find_units_range( |
335 | &self, |
336 | probe_low: u64, |
337 | probe_high: u64, |
338 | ) -> impl Iterator<Item = (&ResUnit<R>, &gimli::Range)> { |
339 | // First up find the position in the array which could have our function |
340 | // address. |
341 | let pos = match self |
342 | .unit_ranges |
343 | .binary_search_by_key(&probe_high, |i| i.range.begin) |
344 | { |
345 | // Although unlikely, we could find an exact match. |
346 | Ok(i) => i + 1, |
347 | // No exact match was found, but this probe would fit at slot `i`. |
348 | // This means that slot `i` is bigger than `probe`, along with all |
349 | // indices greater than `i`, so we need to search all previous |
350 | // entries. |
351 | Err(i) => i, |
352 | }; |
353 | |
354 | // Once we have our index we iterate backwards from that position |
355 | // looking for a matching CU. |
356 | self.unit_ranges[..pos] |
357 | .iter() |
358 | .rev() |
359 | .take_while(move |i| { |
360 | // We know that this CU's start is beneath the probe already because |
361 | // of our sorted array. |
362 | debug_assert!(i.range.begin <= probe_high); |
363 | |
364 | // Each entry keeps track of the maximum end address seen so far, |
365 | // starting from the beginning of the array of unit ranges. We're |
366 | // iterating in reverse so if our probe is beyond the maximum range |
367 | // of this entry, then it's guaranteed to not fit in any prior |
368 | // entries, so we break out. |
369 | probe_low < i.max_end |
370 | }) |
371 | .filter_map(move |i| { |
372 | // If this CU doesn't actually contain this address, move to the |
373 | // next CU. |
374 | if probe_low >= i.range.end || probe_high <= i.range.begin { |
375 | return None; |
376 | } |
377 | Some((&self.units[i.unit_id], &i.range)) |
378 | }) |
379 | } |
380 | |
381 | /// Find the DWARF unit corresponding to the given virtual memory address. |
382 | pub fn find_dwarf_and_unit( |
383 | &self, |
384 | probe: u64, |
385 | ) -> LookupResult< |
386 | impl LookupContinuation<Output = Option<(&gimli::Dwarf<R>, &gimli::Unit<R>)>, Buf = R>, |
387 | > { |
388 | let mut units_iter = self.find_units(probe); |
389 | if let Some(unit) = units_iter.next() { |
390 | return LoopingLookup::new_lookup( |
391 | unit.find_function_or_location(probe, self), |
392 | move |r| { |
393 | ControlFlow::Break(match r { |
394 | Ok((Some(_), _)) | Ok((_, Some(_))) => { |
395 | let (_file, sections, unit) = unit |
396 | .dwarf_and_unit_dwo(self) |
397 | // We've already been through both error cases here to get to this point. |
398 | .unwrap() |
399 | .unwrap(); |
400 | Some((sections, unit)) |
401 | } |
402 | _ => match units_iter.next() { |
403 | Some(next_unit) => { |
404 | return ControlFlow::Continue( |
405 | next_unit.find_function_or_location(probe, self), |
406 | ); |
407 | } |
408 | None => None, |
409 | }, |
410 | }) |
411 | }, |
412 | ); |
413 | } |
414 | |
415 | LoopingLookup::new_complete(None) |
416 | } |
417 | |
418 | /// Find the source file and line corresponding to the given virtual memory address. |
419 | pub fn find_location(&self, probe: u64) -> Result<Option<Location<'_>>, Error> { |
420 | for unit in self.find_units(probe) { |
421 | if let Some(location) = unit.find_location(probe, &self.sections)? { |
422 | return Ok(Some(location)); |
423 | } |
424 | } |
425 | Ok(None) |
426 | } |
427 | |
428 | /// Return source file and lines for a range of addresses. For each location it also |
429 | /// returns the address and size of the range of the underlying instructions. |
430 | pub fn find_location_range( |
431 | &self, |
432 | probe_low: u64, |
433 | probe_high: u64, |
434 | ) -> Result<LocationRangeIter<'_, R>, Error> { |
435 | LocationRangeIter::new(self, probe_low, probe_high) |
436 | } |
437 | |
438 | /// Return an iterator for the function frames corresponding to the given virtual |
439 | /// memory address. |
440 | /// |
441 | /// If the probe address is not for an inline function then only one frame is |
442 | /// returned. |
443 | /// |
444 | /// If the probe address is for an inline function then the first frame corresponds |
445 | /// to the innermost inline function. Subsequent frames contain the caller and call |
446 | /// location, until an non-inline caller is reached. |
447 | pub fn find_frames( |
448 | &self, |
449 | probe: u64, |
450 | ) -> LookupResult<impl LookupContinuation<Output = Result<FrameIter<'_, R>, Error>, Buf = R>> |
451 | { |
452 | let mut units_iter = self.find_units(probe); |
453 | if let Some(unit) = units_iter.next() { |
454 | LoopingLookup::new_lookup(unit.find_function_or_location(probe, self), move |r| { |
455 | ControlFlow::Break(match r { |
456 | Err(e) => Err(e), |
457 | Ok((Some(function), location)) => { |
458 | let inlined_functions = function.find_inlined_functions(probe); |
459 | Ok(FrameIter(FrameIterState::Frames(FrameIterFrames { |
460 | unit, |
461 | sections: &self.sections, |
462 | function, |
463 | inlined_functions, |
464 | next: location, |
465 | }))) |
466 | } |
467 | Ok((None, Some(location))) => { |
468 | Ok(FrameIter(FrameIterState::Location(Some(location)))) |
469 | } |
470 | Ok((None, None)) => match units_iter.next() { |
471 | Some(next_unit) => { |
472 | return ControlFlow::Continue( |
473 | next_unit.find_function_or_location(probe, self), |
474 | ); |
475 | } |
476 | None => Ok(FrameIter(FrameIterState::Empty)), |
477 | }, |
478 | }) |
479 | }) |
480 | } else { |
481 | LoopingLookup::new_complete(Ok(FrameIter(FrameIterState::Empty))) |
482 | } |
483 | } |
484 | |
485 | /// Preload units for `probe`. |
486 | /// |
487 | /// The iterator returns pairs of `SplitDwarfLoad`s containing the |
488 | /// information needed to locate and load split DWARF for `probe` and |
489 | /// a matching callback to invoke once that data is available. |
490 | /// |
491 | /// If this method is called, and all of the returned closures are invoked, |
492 | /// addr2line guarantees that any future API call for the address `probe` |
493 | /// will not require the loading of any split DWARF. |
494 | /// |
495 | /// ```no_run |
496 | /// # use addr2line::*; |
497 | /// # use std::sync::Arc; |
498 | /// # let ctx: Context<gimli::EndianRcSlice<gimli::RunTimeEndian>> = todo!(); |
499 | /// # let do_split_dwarf_load = |load: SplitDwarfLoad<gimli::EndianRcSlice<gimli::RunTimeEndian>>| -> Option<Arc<gimli::Dwarf<gimli::EndianRcSlice<gimli::RunTimeEndian>>>> { None }; |
500 | /// const ADDRESS: u64 = 0xdeadbeef; |
501 | /// ctx.preload_units(ADDRESS).for_each(|(load, callback)| { |
502 | /// let dwo = do_split_dwarf_load(load); |
503 | /// callback(dwo); |
504 | /// }); |
505 | /// |
506 | /// let frames_iter = match ctx.find_frames(ADDRESS) { |
507 | /// LookupResult::Output(result) => result, |
508 | /// LookupResult::Load { .. } => unreachable!("addr2line promised we wouldn't get here" ), |
509 | /// }; |
510 | /// |
511 | /// // ... |
512 | /// ``` |
513 | pub fn preload_units( |
514 | &'_ self, |
515 | probe: u64, |
516 | ) -> impl Iterator< |
517 | Item = ( |
518 | SplitDwarfLoad<R>, |
519 | impl FnOnce(Option<Arc<gimli::Dwarf<R>>>) -> Result<(), gimli::Error> + '_, |
520 | ), |
521 | > { |
522 | self.find_units(probe) |
523 | .filter_map(move |unit| match unit.dwarf_and_unit_dwo(self) { |
524 | LookupResult::Output(_) => None, |
525 | LookupResult::Load { load, continuation } => Some((load, |result| { |
526 | continuation.resume(result).unwrap().map(|_| ()) |
527 | })), |
528 | }) |
529 | } |
530 | |
531 | /// Initialize all line data structures. This is used for benchmarks. |
532 | #[doc (hidden)] |
533 | pub fn parse_lines(&self) -> Result<(), Error> { |
534 | for unit in self.units.iter() { |
535 | unit.parse_lines(&self.sections)?; |
536 | } |
537 | Ok(()) |
538 | } |
539 | |
540 | /// Initialize all function data structures. This is used for benchmarks. |
541 | #[doc (hidden)] |
542 | pub fn parse_functions(&self) -> Result<(), Error> { |
543 | for unit in self.units.iter() { |
544 | unit.parse_functions(self).skip_all_loads()?; |
545 | } |
546 | Ok(()) |
547 | } |
548 | |
549 | /// Initialize all inlined function data structures. This is used for benchmarks. |
550 | #[doc (hidden)] |
551 | pub fn parse_inlined_functions(&self) -> Result<(), Error> { |
552 | for unit in self.units.iter() { |
553 | unit.parse_inlined_functions(self).skip_all_loads()?; |
554 | } |
555 | Ok(()) |
556 | } |
557 | } |
558 | |
559 | struct UnitRange { |
560 | unit_id: usize, |
561 | max_end: u64, |
562 | range: gimli::Range, |
563 | } |
564 | |
565 | struct ResUnit<R: gimli::Reader> { |
566 | offset: gimli::DebugInfoOffset<R::Offset>, |
567 | dw_unit: gimli::Unit<R>, |
568 | lang: Option<gimli::DwLang>, |
569 | lines: LazyCell<Result<Lines, Error>>, |
570 | funcs: LazyCell<Result<Functions<R>, Error>>, |
571 | dwo: LazyCell<Result<Option<Box<(Arc<gimli::Dwarf<R>>, gimli::Unit<R>)>>, Error>>, |
572 | } |
573 | |
574 | struct SupUnit<R: gimli::Reader> { |
575 | offset: gimli::DebugInfoOffset<R::Offset>, |
576 | dw_unit: gimli::Unit<R>, |
577 | } |
578 | |
579 | impl<R: gimli::Reader> Context<R> { |
580 | fn parse_units(sections: &gimli::Dwarf<R>) -> Result<(Vec<UnitRange>, Vec<ResUnit<R>>), Error> { |
581 | // Find all the references to compilation units in .debug_aranges. |
582 | // Note that we always also iterate through all of .debug_info to |
583 | // find compilation units, because .debug_aranges may be missing some. |
584 | let mut aranges = Vec::new(); |
585 | let mut headers = sections.debug_aranges.headers(); |
586 | while let Some(header) = headers.next()? { |
587 | aranges.push((header.debug_info_offset(), header.offset())); |
588 | } |
589 | aranges.sort_by_key(|i| i.0); |
590 | |
591 | let mut unit_ranges = Vec::new(); |
592 | let mut res_units = Vec::new(); |
593 | let mut units = sections.units(); |
594 | while let Some(header) = units.next()? { |
595 | let unit_id = res_units.len(); |
596 | let offset = match header.offset().as_debug_info_offset() { |
597 | Some(offset) => offset, |
598 | None => continue, |
599 | }; |
600 | // We mainly want compile units, but we may need to follow references to entries |
601 | // within other units for function names. We don't need anything from type units. |
602 | match header.type_() { |
603 | gimli::UnitType::Type { .. } | gimli::UnitType::SplitType { .. } => continue, |
604 | _ => {} |
605 | } |
606 | let dw_unit = match sections.unit(header) { |
607 | Ok(dw_unit) => dw_unit, |
608 | Err(_) => continue, |
609 | }; |
610 | |
611 | let mut lang = None; |
612 | let mut have_unit_range = false; |
613 | { |
614 | let mut entries = dw_unit.entries_raw(None)?; |
615 | |
616 | let abbrev = match entries.read_abbreviation()? { |
617 | Some(abbrev) => abbrev, |
618 | None => continue, |
619 | }; |
620 | |
621 | let mut ranges = RangeAttributes::default(); |
622 | for spec in abbrev.attributes() { |
623 | let attr = entries.read_attribute(*spec)?; |
624 | match attr.name() { |
625 | gimli::DW_AT_low_pc => match attr.value() { |
626 | gimli::AttributeValue::Addr(val) => ranges.low_pc = Some(val), |
627 | gimli::AttributeValue::DebugAddrIndex(index) => { |
628 | ranges.low_pc = Some(sections.address(&dw_unit, index)?); |
629 | } |
630 | _ => {} |
631 | }, |
632 | gimli::DW_AT_high_pc => match attr.value() { |
633 | gimli::AttributeValue::Addr(val) => ranges.high_pc = Some(val), |
634 | gimli::AttributeValue::DebugAddrIndex(index) => { |
635 | ranges.high_pc = Some(sections.address(&dw_unit, index)?); |
636 | } |
637 | gimli::AttributeValue::Udata(val) => ranges.size = Some(val), |
638 | _ => {} |
639 | }, |
640 | gimli::DW_AT_ranges => { |
641 | ranges.ranges_offset = |
642 | sections.attr_ranges_offset(&dw_unit, attr.value())?; |
643 | } |
644 | gimli::DW_AT_language => { |
645 | if let gimli::AttributeValue::Language(val) = attr.value() { |
646 | lang = Some(val); |
647 | } |
648 | } |
649 | _ => {} |
650 | } |
651 | } |
652 | |
653 | // Find the address ranges for the CU, using in order of preference: |
654 | // - DW_AT_ranges |
655 | // - .debug_aranges |
656 | // - DW_AT_low_pc/DW_AT_high_pc |
657 | // |
658 | // Using DW_AT_ranges before .debug_aranges is possibly an arbitrary choice, |
659 | // but the feeling is that DW_AT_ranges is more likely to be reliable or complete |
660 | // if it is present. |
661 | // |
662 | // .debug_aranges must be used before DW_AT_low_pc/DW_AT_high_pc because |
663 | // it has been observed on macOS that DW_AT_ranges was not emitted even for |
664 | // discontiguous CUs. |
665 | let i = match ranges.ranges_offset { |
666 | Some(_) => None, |
667 | None => aranges.binary_search_by_key(&offset, |x| x.0).ok(), |
668 | }; |
669 | if let Some(mut i) = i { |
670 | // There should be only one set per CU, but in practice multiple |
671 | // sets have been observed. This is probably a compiler bug, but |
672 | // either way we need to handle it. |
673 | while i > 0 && aranges[i - 1].0 == offset { |
674 | i -= 1; |
675 | } |
676 | for (_, aranges_offset) in aranges[i..].iter().take_while(|x| x.0 == offset) { |
677 | let aranges_header = sections.debug_aranges.header(*aranges_offset)?; |
678 | let mut aranges = aranges_header.entries(); |
679 | while let Some(arange) = aranges.next()? { |
680 | if arange.length() != 0 { |
681 | unit_ranges.push(UnitRange { |
682 | range: arange.range(), |
683 | unit_id, |
684 | max_end: 0, |
685 | }); |
686 | have_unit_range = true; |
687 | } |
688 | } |
689 | } |
690 | } else { |
691 | have_unit_range |= ranges.for_each_range(sections, &dw_unit, |range| { |
692 | unit_ranges.push(UnitRange { |
693 | range, |
694 | unit_id, |
695 | max_end: 0, |
696 | }); |
697 | })?; |
698 | } |
699 | } |
700 | |
701 | let lines = LazyCell::new(); |
702 | if !have_unit_range { |
703 | // The unit did not declare any ranges. |
704 | // Try to get some ranges from the line program sequences. |
705 | if let Some(ref ilnp) = dw_unit.line_program { |
706 | if let Ok(lines) = lines |
707 | .borrow_with(|| Lines::parse(&dw_unit, ilnp.clone(), sections)) |
708 | .as_ref() |
709 | { |
710 | for sequence in lines.sequences.iter() { |
711 | unit_ranges.push(UnitRange { |
712 | range: gimli::Range { |
713 | begin: sequence.start, |
714 | end: sequence.end, |
715 | }, |
716 | unit_id, |
717 | max_end: 0, |
718 | }) |
719 | } |
720 | } |
721 | } |
722 | } |
723 | |
724 | res_units.push(ResUnit { |
725 | offset, |
726 | dw_unit, |
727 | lang, |
728 | lines, |
729 | funcs: LazyCell::new(), |
730 | dwo: LazyCell::new(), |
731 | }); |
732 | } |
733 | |
734 | // Sort this for faster lookup in `find_unit_and_address` below. |
735 | unit_ranges.sort_by_key(|i| i.range.begin); |
736 | |
737 | // Calculate the `max_end` field now that we've determined the order of |
738 | // CUs. |
739 | let mut max = 0; |
740 | for i in unit_ranges.iter_mut() { |
741 | max = max.max(i.range.end); |
742 | i.max_end = max; |
743 | } |
744 | |
745 | Ok((unit_ranges, res_units)) |
746 | } |
747 | |
748 | fn parse_sup(sections: &gimli::Dwarf<R>) -> Result<Vec<SupUnit<R>>, Error> { |
749 | let mut sup_units = Vec::new(); |
750 | let mut units = sections.units(); |
751 | while let Some(header) = units.next()? { |
752 | let offset = match header.offset().as_debug_info_offset() { |
753 | Some(offset) => offset, |
754 | None => continue, |
755 | }; |
756 | let dw_unit = match sections.unit(header) { |
757 | Ok(dw_unit) => dw_unit, |
758 | Err(_) => continue, |
759 | }; |
760 | sup_units.push(SupUnit { dw_unit, offset }); |
761 | } |
762 | Ok(sup_units) |
763 | } |
764 | |
765 | // Find the unit containing the given offset, and convert the offset into a unit offset. |
766 | fn find_unit( |
767 | &self, |
768 | offset: gimli::DebugInfoOffset<R::Offset>, |
769 | file: DebugFile, |
770 | ) -> Result<(&gimli::Unit<R>, gimli::UnitOffset<R::Offset>), Error> { |
771 | let unit = match file { |
772 | DebugFile::Primary => { |
773 | match self |
774 | .units |
775 | .binary_search_by_key(&offset.0, |unit| unit.offset.0) |
776 | { |
777 | // There is never a DIE at the unit offset or before the first unit. |
778 | Ok(_) | Err(0) => return Err(gimli::Error::NoEntryAtGivenOffset), |
779 | Err(i) => &self.units[i - 1].dw_unit, |
780 | } |
781 | } |
782 | DebugFile::Supplementary => { |
783 | match self |
784 | .sup_units |
785 | .binary_search_by_key(&offset.0, |unit| unit.offset.0) |
786 | { |
787 | // There is never a DIE at the unit offset or before the first unit. |
788 | Ok(_) | Err(0) => return Err(gimli::Error::NoEntryAtGivenOffset), |
789 | Err(i) => &self.sup_units[i - 1].dw_unit, |
790 | } |
791 | } |
792 | DebugFile::Dwo => return Err(gimli::Error::NoEntryAtGivenOffset), |
793 | }; |
794 | |
795 | let unit_offset = offset |
796 | .to_unit_offset(&unit.header) |
797 | .ok_or(gimli::Error::NoEntryAtGivenOffset)?; |
798 | Ok((unit, unit_offset)) |
799 | } |
800 | } |
801 | |
802 | struct Lines { |
803 | files: Box<[String]>, |
804 | sequences: Box<[LineSequence]>, |
805 | } |
806 | |
807 | impl Lines { |
808 | fn parse<R: gimli::Reader>( |
809 | dw_unit: &gimli::Unit<R>, |
810 | ilnp: gimli::IncompleteLineProgram<R, R::Offset>, |
811 | sections: &gimli::Dwarf<R>, |
812 | ) -> Result<Self, Error> { |
813 | let mut sequences = Vec::new(); |
814 | let mut sequence_rows = Vec::<LineRow>::new(); |
815 | let mut rows = ilnp.rows(); |
816 | while let Some((_, row)) = rows.next_row()? { |
817 | if row.end_sequence() { |
818 | if let Some(start) = sequence_rows.first().map(|x| x.address) { |
819 | let end = row.address(); |
820 | let mut rows = Vec::new(); |
821 | mem::swap(&mut rows, &mut sequence_rows); |
822 | sequences.push(LineSequence { |
823 | start, |
824 | end, |
825 | rows: rows.into_boxed_slice(), |
826 | }); |
827 | } |
828 | continue; |
829 | } |
830 | |
831 | let address = row.address(); |
832 | let file_index = row.file_index(); |
833 | let line = row.line().map(NonZeroU64::get).unwrap_or(0) as u32; |
834 | let column = match row.column() { |
835 | gimli::ColumnType::LeftEdge => 0, |
836 | gimli::ColumnType::Column(x) => x.get() as u32, |
837 | }; |
838 | |
839 | if let Some(last_row) = sequence_rows.last_mut() { |
840 | if last_row.address == address { |
841 | last_row.file_index = file_index; |
842 | last_row.line = line; |
843 | last_row.column = column; |
844 | continue; |
845 | } |
846 | } |
847 | |
848 | sequence_rows.push(LineRow { |
849 | address, |
850 | file_index, |
851 | line, |
852 | column, |
853 | }); |
854 | } |
855 | sequences.sort_by_key(|x| x.start); |
856 | |
857 | let mut files = Vec::new(); |
858 | let header = rows.header(); |
859 | match header.file(0) { |
860 | Some(file) => files.push(render_file(dw_unit, file, header, sections)?), |
861 | None => files.push(String::from("" )), // DWARF version <= 4 may not have 0th index |
862 | } |
863 | let mut index = 1; |
864 | while let Some(file) = header.file(index) { |
865 | files.push(render_file(dw_unit, file, header, sections)?); |
866 | index += 1; |
867 | } |
868 | |
869 | Ok(Self { |
870 | files: files.into_boxed_slice(), |
871 | sequences: sequences.into_boxed_slice(), |
872 | }) |
873 | } |
874 | } |
875 | |
876 | fn render_file<R: gimli::Reader>( |
877 | dw_unit: &gimli::Unit<R>, |
878 | file: &gimli::FileEntry<R, R::Offset>, |
879 | header: &gimli::LineProgramHeader<R, R::Offset>, |
880 | sections: &gimli::Dwarf<R>, |
881 | ) -> Result<String, gimli::Error> { |
882 | let mut path = if let Some(ref comp_dir) = dw_unit.comp_dir { |
883 | comp_dir.to_string_lossy()?.into_owned() |
884 | } else { |
885 | String::new() |
886 | }; |
887 | |
888 | // The directory index 0 is defined to correspond to the compilation unit directory. |
889 | if file.directory_index() != 0 { |
890 | if let Some(directory) = file.directory(header) { |
891 | path_push( |
892 | &mut path, |
893 | sections |
894 | .attr_string(dw_unit, directory)? |
895 | .to_string_lossy()? |
896 | .as_ref(), |
897 | ); |
898 | } |
899 | } |
900 | |
901 | path_push( |
902 | &mut path, |
903 | sections |
904 | .attr_string(dw_unit, file.path_name())? |
905 | .to_string_lossy()? |
906 | .as_ref(), |
907 | ); |
908 | |
909 | Ok(path) |
910 | } |
911 | |
912 | struct LineSequence { |
913 | start: u64, |
914 | end: u64, |
915 | rows: Box<[LineRow]>, |
916 | } |
917 | |
918 | struct LineRow { |
919 | address: u64, |
920 | file_index: u64, |
921 | line: u32, |
922 | column: u32, |
923 | } |
924 | |
925 | /// This struct contains the information needed to find split DWARF data |
926 | /// and to produce a `gimli::Dwarf<R>` for it. |
927 | pub struct SplitDwarfLoad<R> { |
928 | /// The dwo id, for looking up in a DWARF package, or for |
929 | /// verifying an unpacked dwo found on the file system |
930 | pub dwo_id: gimli::DwoId, |
931 | /// The compilation directory `path` is relative to. |
932 | pub comp_dir: Option<R>, |
933 | /// A path on the filesystem, relative to `comp_dir` to find this dwo. |
934 | pub path: Option<R>, |
935 | /// Once the split DWARF data is loaded, the loader is expected |
936 | /// to call [make_dwo(parent)](gimli::read::Dwarf::make_dwo) before |
937 | /// returning the data. |
938 | pub parent: Arc<gimli::Dwarf<R>>, |
939 | } |
940 | |
941 | struct SimpleLookup<T, R, F> |
942 | where |
943 | F: FnOnce(Option<Arc<gimli::Dwarf<R>>>) -> T, |
944 | R: gimli::Reader, |
945 | { |
946 | f: F, |
947 | phantom: PhantomData<(T, R)>, |
948 | } |
949 | |
950 | impl<T, R, F> SimpleLookup<T, R, F> |
951 | where |
952 | F: FnOnce(Option<Arc<gimli::Dwarf<R>>>) -> T, |
953 | R: gimli::Reader, |
954 | { |
955 | fn new_complete(t: F::Output) -> LookupResult<SimpleLookup<T, R, F>> { |
956 | LookupResult::Output(t) |
957 | } |
958 | |
959 | fn new_needs_load(load: SplitDwarfLoad<R>, f: F) -> LookupResult<SimpleLookup<T, R, F>> { |
960 | LookupResult::Load { |
961 | load, |
962 | continuation: SimpleLookup { |
963 | f, |
964 | phantom: PhantomData, |
965 | }, |
966 | } |
967 | } |
968 | } |
969 | |
970 | impl<T, R, F> LookupContinuation for SimpleLookup<T, R, F> |
971 | where |
972 | F: FnOnce(Option<Arc<gimli::Dwarf<R>>>) -> T, |
973 | R: gimli::Reader, |
974 | { |
975 | type Output = T; |
976 | type Buf = R; |
977 | |
978 | fn resume(self, v: Option<Arc<gimli::Dwarf<Self::Buf>>>) -> LookupResult<Self> { |
979 | LookupResult::Output((self.f)(v)) |
980 | } |
981 | } |
982 | |
983 | struct MappedLookup<T, L, F> |
984 | where |
985 | L: LookupContinuation, |
986 | F: FnOnce(L::Output) -> T, |
987 | { |
988 | original: L, |
989 | mutator: F, |
990 | } |
991 | |
992 | impl<T, L, F> LookupContinuation for MappedLookup<T, L, F> |
993 | where |
994 | L: LookupContinuation, |
995 | F: FnOnce(L::Output) -> T, |
996 | { |
997 | type Output = T; |
998 | type Buf = L::Buf; |
999 | |
1000 | fn resume(self, v: Option<Arc<gimli::Dwarf<Self::Buf>>>) -> LookupResult<Self> { |
1001 | match self.original.resume(input:v) { |
1002 | LookupResult::Output(t: ::Output) => LookupResult::Output((self.mutator)(t)), |
1003 | LookupResult::Load { load: SplitDwarfLoad<::Buf>, continuation: L } => LookupResult::Load { |
1004 | load, |
1005 | continuation: MappedLookup { |
1006 | original: continuation, |
1007 | mutator: self.mutator, |
1008 | }, |
1009 | }, |
1010 | } |
1011 | } |
1012 | } |
1013 | |
1014 | /// Some functions (e.g. `find_frames`) require considering multiple |
1015 | /// compilation units, each of which might require their own split DWARF |
1016 | /// lookup (and thus produce a continuation). |
1017 | /// |
1018 | /// We store the underlying continuation here as well as a mutator function |
1019 | /// that will either a) decide that the result of this continuation is |
1020 | /// what is needed and mutate it to the final result or b) produce another |
1021 | /// `LookupResult`. `new_lookup` will in turn eagerly drive any non-continuation |
1022 | /// `LookupResult` with successive invocations of the mutator, until a new |
1023 | /// continuation or a final result is produced. And finally, the impl of |
1024 | /// `LookupContinuation::resume` will call `new_lookup` each time the |
1025 | /// computation is resumed. |
1026 | struct LoopingLookup<T, L, F> |
1027 | where |
1028 | L: LookupContinuation, |
1029 | F: FnMut(L::Output) -> ControlFlow<T, LookupResult<L>>, |
1030 | { |
1031 | continuation: L, |
1032 | mutator: F, |
1033 | } |
1034 | |
1035 | impl<T, L, F> LoopingLookup<T, L, F> |
1036 | where |
1037 | L: LookupContinuation, |
1038 | F: FnMut(L::Output) -> ControlFlow<T, LookupResult<L>>, |
1039 | { |
1040 | fn new_complete(t: T) -> LookupResult<Self> { |
1041 | LookupResult::Output(t) |
1042 | } |
1043 | |
1044 | fn new_lookup(mut r: LookupResult<L>, mut mutator: F) -> LookupResult<Self> { |
1045 | // Drive the loop eagerly so that we only ever have to represent one state |
1046 | // (the r == ControlFlow::Continue state) in LoopingLookup. |
1047 | loop { |
1048 | match r { |
1049 | LookupResult::Output(l) => match mutator(l) { |
1050 | ControlFlow::Break(t) => return LookupResult::Output(t), |
1051 | ControlFlow::Continue(r2) => { |
1052 | r = r2; |
1053 | } |
1054 | }, |
1055 | LookupResult::Load { load, continuation } => { |
1056 | return LookupResult::Load { |
1057 | load, |
1058 | continuation: LoopingLookup { |
1059 | continuation, |
1060 | mutator, |
1061 | }, |
1062 | }; |
1063 | } |
1064 | } |
1065 | } |
1066 | } |
1067 | } |
1068 | |
1069 | impl<T, L, F> LookupContinuation for LoopingLookup<T, L, F> |
1070 | where |
1071 | L: LookupContinuation, |
1072 | F: FnMut(L::Output) -> ControlFlow<T, LookupResult<L>>, |
1073 | { |
1074 | type Output = T; |
1075 | type Buf = L::Buf; |
1076 | |
1077 | fn resume(self, v: Option<Arc<gimli::Dwarf<Self::Buf>>>) -> LookupResult<Self> { |
1078 | let r: LookupResult = self.continuation.resume(input:v); |
1079 | LoopingLookup::new_lookup(r, self.mutator) |
1080 | } |
1081 | } |
1082 | |
1083 | impl<R: gimli::Reader> ResUnit<R> { |
1084 | fn dwarf_and_unit_dwo<'unit, 'ctx: 'unit>( |
1085 | &'unit self, |
1086 | ctx: &'ctx Context<R>, |
1087 | ) -> LookupResult< |
1088 | SimpleLookup< |
1089 | Result<(DebugFile, &'unit gimli::Dwarf<R>, &'unit gimli::Unit<R>), Error>, |
1090 | R, |
1091 | impl FnOnce( |
1092 | Option<Arc<gimli::Dwarf<R>>>, |
1093 | ) |
1094 | -> Result<(DebugFile, &'unit gimli::Dwarf<R>, &'unit gimli::Unit<R>), Error>, |
1095 | >, |
1096 | > { |
1097 | loop { |
1098 | break SimpleLookup::new_complete(match self.dwo.borrow() { |
1099 | Some(Ok(Some(v))) => Ok((DebugFile::Dwo, &*v.0, &v.1)), |
1100 | Some(Ok(None)) => Ok((DebugFile::Primary, &*ctx.sections, &self.dw_unit)), |
1101 | Some(Err(e)) => Err(*e), |
1102 | None => { |
1103 | let dwo_id = match self.dw_unit.dwo_id { |
1104 | None => { |
1105 | self.dwo.borrow_with(|| Ok(None)); |
1106 | continue; |
1107 | } |
1108 | Some(dwo_id) => dwo_id, |
1109 | }; |
1110 | |
1111 | let comp_dir = self.dw_unit.comp_dir.clone(); |
1112 | |
1113 | let dwo_name = self.dw_unit.dwo_name().and_then(|s| { |
1114 | if let Some(s) = s { |
1115 | Ok(Some(ctx.sections.attr_string(&self.dw_unit, s)?)) |
1116 | } else { |
1117 | Ok(None) |
1118 | } |
1119 | }); |
1120 | |
1121 | let path = match dwo_name { |
1122 | Ok(v) => v, |
1123 | Err(e) => { |
1124 | self.dwo.borrow_with(|| Err(e)); |
1125 | continue; |
1126 | } |
1127 | }; |
1128 | |
1129 | let process_dwo = move |dwo_dwarf: Option<Arc<gimli::Dwarf<R>>>| { |
1130 | let dwo_dwarf = match dwo_dwarf { |
1131 | None => return Ok(None), |
1132 | Some(dwo_dwarf) => dwo_dwarf, |
1133 | }; |
1134 | let mut dwo_units = dwo_dwarf.units(); |
1135 | let dwo_header = match dwo_units.next()? { |
1136 | Some(dwo_header) => dwo_header, |
1137 | None => return Ok(None), |
1138 | }; |
1139 | |
1140 | let mut dwo_unit = dwo_dwarf.unit(dwo_header)?; |
1141 | dwo_unit.copy_relocated_attributes(&self.dw_unit); |
1142 | Ok(Some(Box::new((dwo_dwarf, dwo_unit)))) |
1143 | }; |
1144 | |
1145 | return SimpleLookup::new_needs_load( |
1146 | SplitDwarfLoad { |
1147 | dwo_id, |
1148 | comp_dir, |
1149 | path, |
1150 | parent: ctx.sections.clone(), |
1151 | }, |
1152 | move |dwo_dwarf| match self.dwo.borrow_with(|| process_dwo(dwo_dwarf)) { |
1153 | Ok(Some(v)) => Ok((DebugFile::Dwo, &*v.0, &v.1)), |
1154 | Ok(None) => Ok((DebugFile::Primary, &*ctx.sections, &self.dw_unit)), |
1155 | Err(e) => Err(*e), |
1156 | }, |
1157 | ); |
1158 | } |
1159 | }); |
1160 | } |
1161 | } |
1162 | |
1163 | fn parse_lines(&self, sections: &gimli::Dwarf<R>) -> Result<Option<&Lines>, Error> { |
1164 | // NB: line information is always stored in the main debug file so this does not need |
1165 | // to handle DWOs. |
1166 | let ilnp = match self.dw_unit.line_program { |
1167 | Some(ref ilnp) => ilnp, |
1168 | None => return Ok(None), |
1169 | }; |
1170 | self.lines |
1171 | .borrow_with(|| Lines::parse(&self.dw_unit, ilnp.clone(), sections)) |
1172 | .as_ref() |
1173 | .map(Some) |
1174 | .map_err(Error::clone) |
1175 | } |
1176 | |
1177 | fn parse_functions_dwarf_and_unit( |
1178 | &self, |
1179 | unit: &gimli::Unit<R>, |
1180 | sections: &gimli::Dwarf<R>, |
1181 | ) -> Result<&Functions<R>, Error> { |
1182 | self.funcs |
1183 | .borrow_with(|| Functions::parse(unit, sections)) |
1184 | .as_ref() |
1185 | .map_err(Error::clone) |
1186 | } |
1187 | |
1188 | fn parse_functions<'unit, 'ctx: 'unit>( |
1189 | &'unit self, |
1190 | ctx: &'ctx Context<R>, |
1191 | ) -> LookupResult<impl LookupContinuation<Output = Result<&'unit Functions<R>, Error>, Buf = R>> |
1192 | { |
1193 | self.dwarf_and_unit_dwo(ctx).map(move |r| { |
1194 | let (_file, sections, unit) = r?; |
1195 | self.parse_functions_dwarf_and_unit(unit, sections) |
1196 | }) |
1197 | } |
1198 | fn parse_inlined_functions<'unit, 'ctx: 'unit>( |
1199 | &'unit self, |
1200 | ctx: &'ctx Context<R>, |
1201 | ) -> LookupResult<impl LookupContinuation<Output = Result<(), Error>, Buf = R> + 'unit> { |
1202 | self.dwarf_and_unit_dwo(ctx).map(move |r| { |
1203 | let (file, sections, unit) = r?; |
1204 | self.funcs |
1205 | .borrow_with(|| Functions::parse(unit, sections)) |
1206 | .as_ref() |
1207 | .map_err(Error::clone)? |
1208 | .parse_inlined_functions(file, unit, ctx, sections) |
1209 | }) |
1210 | } |
1211 | |
1212 | fn find_location( |
1213 | &self, |
1214 | probe: u64, |
1215 | sections: &gimli::Dwarf<R>, |
1216 | ) -> Result<Option<Location<'_>>, Error> { |
1217 | if let Some(mut iter) = LocationRangeUnitIter::new(self, sections, probe, probe + 1)? { |
1218 | match iter.next() { |
1219 | None => Ok(None), |
1220 | Some((_addr, _len, loc)) => Ok(Some(loc)), |
1221 | } |
1222 | } else { |
1223 | Ok(None) |
1224 | } |
1225 | } |
1226 | |
1227 | #[inline ] |
1228 | fn find_location_range( |
1229 | &self, |
1230 | probe_low: u64, |
1231 | probe_high: u64, |
1232 | sections: &gimli::Dwarf<R>, |
1233 | ) -> Result<Option<LocationRangeUnitIter<'_>>, Error> { |
1234 | LocationRangeUnitIter::new(self, sections, probe_low, probe_high) |
1235 | } |
1236 | |
1237 | fn find_function_or_location<'unit, 'ctx: 'unit>( |
1238 | &'unit self, |
1239 | probe: u64, |
1240 | ctx: &'ctx Context<R>, |
1241 | ) -> LookupResult< |
1242 | impl LookupContinuation< |
1243 | Output = Result<(Option<&'unit Function<R>>, Option<Location<'unit>>), Error>, |
1244 | Buf = R, |
1245 | >, |
1246 | > { |
1247 | self.dwarf_and_unit_dwo(ctx).map(move |r| { |
1248 | let (file, sections, unit) = r?; |
1249 | let functions = self.parse_functions_dwarf_and_unit(unit, sections)?; |
1250 | let function = match functions.find_address(probe) { |
1251 | Some(address) => { |
1252 | let function_index = functions.addresses[address].function; |
1253 | let (offset, ref function) = functions.functions[function_index]; |
1254 | Some( |
1255 | function |
1256 | .borrow_with(|| Function::parse(offset, file, unit, ctx, sections)) |
1257 | .as_ref() |
1258 | .map_err(Error::clone)?, |
1259 | ) |
1260 | } |
1261 | None => None, |
1262 | }; |
1263 | let location = self.find_location(probe, sections)?; |
1264 | Ok((function, location)) |
1265 | }) |
1266 | } |
1267 | } |
1268 | |
1269 | /// Iterator over `Location`s in a range of addresses, returned by `Context::find_location_range`. |
1270 | pub struct LocationRangeIter<'ctx, R: gimli::Reader> { |
1271 | unit_iter: Box<dyn Iterator<Item = (&'ctx ResUnit<R>, &'ctx gimli::Range)> + 'ctx>, |
1272 | iter: Option<LocationRangeUnitIter<'ctx>>, |
1273 | |
1274 | probe_low: u64, |
1275 | probe_high: u64, |
1276 | sections: &'ctx gimli::Dwarf<R>, |
1277 | } |
1278 | |
1279 | impl<'ctx, R: gimli::Reader> LocationRangeIter<'ctx, R> { |
1280 | #[inline ] |
1281 | fn new(ctx: &'ctx Context<R>, probe_low: u64, probe_high: u64) -> Result<Self, Error> { |
1282 | let sections = &ctx.sections; |
1283 | let unit_iter = ctx.find_units_range(probe_low, probe_high); |
1284 | Ok(Self { |
1285 | unit_iter: Box::new(unit_iter), |
1286 | iter: None, |
1287 | probe_low, |
1288 | probe_high, |
1289 | sections, |
1290 | }) |
1291 | } |
1292 | |
1293 | fn next_loc(&mut self) -> Result<Option<(u64, u64, Location<'ctx>)>, Error> { |
1294 | loop { |
1295 | let iter = self.iter.take(); |
1296 | match iter { |
1297 | None => match self.unit_iter.next() { |
1298 | Some((unit, range)) => { |
1299 | self.iter = unit.find_location_range( |
1300 | cmp::max(self.probe_low, range.begin), |
1301 | cmp::min(self.probe_high, range.end), |
1302 | self.sections, |
1303 | )?; |
1304 | } |
1305 | None => return Ok(None), |
1306 | }, |
1307 | Some(mut iter) => { |
1308 | if let item @ Some(_) = iter.next() { |
1309 | self.iter = Some(iter); |
1310 | return Ok(item); |
1311 | } |
1312 | } |
1313 | } |
1314 | } |
1315 | } |
1316 | } |
1317 | |
1318 | impl<'ctx, R> Iterator for LocationRangeIter<'ctx, R> |
1319 | where |
1320 | R: gimli::Reader + 'ctx, |
1321 | { |
1322 | type Item = (u64, u64, Location<'ctx>); |
1323 | |
1324 | #[inline ] |
1325 | fn next(&mut self) -> Option<Self::Item> { |
1326 | match self.next_loc() { |
1327 | Err(_) => None, |
1328 | Ok(loc: Option<(u64, u64, Location<'_>)>) => loc, |
1329 | } |
1330 | } |
1331 | } |
1332 | |
1333 | #[cfg (feature = "fallible-iterator" )] |
1334 | impl<'ctx, R> fallible_iterator::FallibleIterator for LocationRangeIter<'ctx, R> |
1335 | where |
1336 | R: gimli::Reader + 'ctx, |
1337 | { |
1338 | type Item = (u64, u64, Location<'ctx>); |
1339 | type Error = Error; |
1340 | |
1341 | #[inline ] |
1342 | fn next(&mut self) -> Result<Option<Self::Item>, Self::Error> { |
1343 | self.next_loc() |
1344 | } |
1345 | } |
1346 | |
1347 | struct LocationRangeUnitIter<'ctx> { |
1348 | lines: &'ctx Lines, |
1349 | seqs: &'ctx [LineSequence], |
1350 | seq_idx: usize, |
1351 | row_idx: usize, |
1352 | probe_high: u64, |
1353 | } |
1354 | |
1355 | impl<'ctx> LocationRangeUnitIter<'ctx> { |
1356 | fn new<R: gimli::Reader>( |
1357 | resunit: &'ctx ResUnit<R>, |
1358 | sections: &gimli::Dwarf<R>, |
1359 | probe_low: u64, |
1360 | probe_high: u64, |
1361 | ) -> Result<Option<Self>, Error> { |
1362 | let lines = resunit.parse_lines(sections)?; |
1363 | |
1364 | if let Some(lines) = lines { |
1365 | // Find index for probe_low. |
1366 | let seq_idx = lines.sequences.binary_search_by(|sequence| { |
1367 | if probe_low < sequence.start { |
1368 | Ordering::Greater |
1369 | } else if probe_low >= sequence.end { |
1370 | Ordering::Less |
1371 | } else { |
1372 | Ordering::Equal |
1373 | } |
1374 | }); |
1375 | let seq_idx = match seq_idx { |
1376 | Ok(x) => x, |
1377 | Err(0) => 0, // probe below sequence, but range could overlap |
1378 | Err(_) => lines.sequences.len(), |
1379 | }; |
1380 | |
1381 | let row_idx = if let Some(seq) = lines.sequences.get(seq_idx) { |
1382 | let idx = seq.rows.binary_search_by(|row| row.address.cmp(&probe_low)); |
1383 | match idx { |
1384 | Ok(x) => x, |
1385 | Err(0) => 0, // probe below sequence, but range could overlap |
1386 | Err(x) => x - 1, |
1387 | } |
1388 | } else { |
1389 | 0 |
1390 | }; |
1391 | |
1392 | Ok(Some(Self { |
1393 | lines, |
1394 | seqs: &*lines.sequences, |
1395 | seq_idx, |
1396 | row_idx, |
1397 | probe_high, |
1398 | })) |
1399 | } else { |
1400 | Ok(None) |
1401 | } |
1402 | } |
1403 | } |
1404 | |
1405 | impl<'ctx> Iterator for LocationRangeUnitIter<'ctx> { |
1406 | type Item = (u64, u64, Location<'ctx>); |
1407 | |
1408 | fn next(&mut self) -> Option<(u64, u64, Location<'ctx>)> { |
1409 | while let Some(seq) = self.seqs.get(self.seq_idx) { |
1410 | if seq.start >= self.probe_high { |
1411 | break; |
1412 | } |
1413 | |
1414 | match seq.rows.get(self.row_idx) { |
1415 | Some(row) => { |
1416 | if row.address >= self.probe_high { |
1417 | break; |
1418 | } |
1419 | |
1420 | let file = self |
1421 | .lines |
1422 | .files |
1423 | .get(row.file_index as usize) |
1424 | .map(String::as_str); |
1425 | let nextaddr = seq |
1426 | .rows |
1427 | .get(self.row_idx + 1) |
1428 | .map(|row| row.address) |
1429 | .unwrap_or(seq.end); |
1430 | |
1431 | let item = ( |
1432 | row.address, |
1433 | nextaddr - row.address, |
1434 | Location { |
1435 | file, |
1436 | line: if row.line != 0 { Some(row.line) } else { None }, |
1437 | column: if row.column != 0 { |
1438 | Some(row.column) |
1439 | } else { |
1440 | None |
1441 | }, |
1442 | }, |
1443 | ); |
1444 | self.row_idx += 1; |
1445 | |
1446 | return Some(item); |
1447 | } |
1448 | None => { |
1449 | self.seq_idx += 1; |
1450 | self.row_idx = 0; |
1451 | } |
1452 | } |
1453 | } |
1454 | None |
1455 | } |
1456 | } |
1457 | |
1458 | fn path_push(path: &mut String, p: &str) { |
1459 | if has_unix_root(p) || has_windows_root(p) { |
1460 | *path = p.to_string(); |
1461 | } else { |
1462 | let dir_separator: char = if has_windows_root(path.as_str()) { |
1463 | ' \\' |
1464 | } else { |
1465 | '/' |
1466 | }; |
1467 | |
1468 | if !path.is_empty() && !path.ends_with(dir_separator) { |
1469 | path.push(ch:dir_separator); |
1470 | } |
1471 | *path += p; |
1472 | } |
1473 | } |
1474 | |
1475 | /// Check if the path in the given string has a unix style root |
1476 | fn has_unix_root(p: &str) -> bool { |
1477 | p.starts_with('/' ) |
1478 | } |
1479 | |
1480 | /// Check if the path in the given string has a windows style root |
1481 | fn has_windows_root(p: &str) -> bool { |
1482 | p.starts_with(' \\' ) || p.get(1..3) == Some(": \\" ) |
1483 | } |
1484 | struct RangeAttributes<R: gimli::Reader> { |
1485 | low_pc: Option<u64>, |
1486 | high_pc: Option<u64>, |
1487 | size: Option<u64>, |
1488 | ranges_offset: Option<gimli::RangeListsOffset<<R as gimli::Reader>::Offset>>, |
1489 | } |
1490 | |
1491 | impl<R: gimli::Reader> Default for RangeAttributes<R> { |
1492 | fn default() -> Self { |
1493 | RangeAttributes { |
1494 | low_pc: None, |
1495 | high_pc: None, |
1496 | size: None, |
1497 | ranges_offset: None, |
1498 | } |
1499 | } |
1500 | } |
1501 | |
1502 | impl<R: gimli::Reader> RangeAttributes<R> { |
1503 | fn for_each_range<F: FnMut(gimli::Range)>( |
1504 | &self, |
1505 | sections: &gimli::Dwarf<R>, |
1506 | unit: &gimli::Unit<R>, |
1507 | mut f: F, |
1508 | ) -> Result<bool, Error> { |
1509 | let mut added_any = false; |
1510 | let mut add_range = |range: gimli::Range| { |
1511 | if range.begin < range.end { |
1512 | f(range); |
1513 | added_any = true |
1514 | } |
1515 | }; |
1516 | if let Some(ranges_offset) = self.ranges_offset { |
1517 | let mut range_list = sections.ranges(unit, ranges_offset)?; |
1518 | while let Some(range) = range_list.next()? { |
1519 | add_range(range); |
1520 | } |
1521 | } else if let (Some(begin), Some(end)) = (self.low_pc, self.high_pc) { |
1522 | add_range(gimli::Range { begin, end }); |
1523 | } else if let (Some(begin), Some(size)) = (self.low_pc, self.size) { |
1524 | add_range(gimli::Range { |
1525 | begin, |
1526 | end: begin + size, |
1527 | }); |
1528 | } |
1529 | Ok(added_any) |
1530 | } |
1531 | } |
1532 | |
1533 | /// An iterator over function frames. |
1534 | pub struct FrameIter<'ctx, R>(FrameIterState<'ctx, R>) |
1535 | where |
1536 | R: gimli::Reader; |
1537 | |
1538 | enum FrameIterState<'ctx, R> |
1539 | where |
1540 | R: gimli::Reader, |
1541 | { |
1542 | Empty, |
1543 | Location(Option<Location<'ctx>>), |
1544 | Frames(FrameIterFrames<'ctx, R>), |
1545 | } |
1546 | |
1547 | struct FrameIterFrames<'ctx, R> |
1548 | where |
1549 | R: gimli::Reader, |
1550 | { |
1551 | unit: &'ctx ResUnit<R>, |
1552 | sections: &'ctx gimli::Dwarf<R>, |
1553 | function: &'ctx Function<R>, |
1554 | inlined_functions: iter::Rev<maybe_small::IntoIter<&'ctx InlinedFunction<R>>>, |
1555 | next: Option<Location<'ctx>>, |
1556 | } |
1557 | |
1558 | impl<'ctx, R> FrameIter<'ctx, R> |
1559 | where |
1560 | R: gimli::Reader + 'ctx, |
1561 | { |
1562 | /// Advances the iterator and returns the next frame. |
1563 | pub fn next(&mut self) -> Result<Option<Frame<'ctx, R>>, Error> { |
1564 | let frames = match &mut self.0 { |
1565 | FrameIterState::Empty => return Ok(None), |
1566 | FrameIterState::Location(location) => { |
1567 | // We can't move out of a mutable reference, so use `take` instead. |
1568 | let location = location.take(); |
1569 | self.0 = FrameIterState::Empty; |
1570 | return Ok(Some(Frame { |
1571 | dw_die_offset: None, |
1572 | function: None, |
1573 | location, |
1574 | })); |
1575 | } |
1576 | FrameIterState::Frames(frames) => frames, |
1577 | }; |
1578 | |
1579 | let loc = frames.next.take(); |
1580 | let func = match frames.inlined_functions.next() { |
1581 | Some(func) => func, |
1582 | None => { |
1583 | let frame = Frame { |
1584 | dw_die_offset: Some(frames.function.dw_die_offset), |
1585 | function: frames.function.name.clone().map(|name| FunctionName { |
1586 | name, |
1587 | language: frames.unit.lang, |
1588 | }), |
1589 | location: loc, |
1590 | }; |
1591 | self.0 = FrameIterState::Empty; |
1592 | return Ok(Some(frame)); |
1593 | } |
1594 | }; |
1595 | |
1596 | let mut next = Location { |
1597 | file: None, |
1598 | line: if func.call_line != 0 { |
1599 | Some(func.call_line) |
1600 | } else { |
1601 | None |
1602 | }, |
1603 | column: if func.call_column != 0 { |
1604 | Some(func.call_column) |
1605 | } else { |
1606 | None |
1607 | }, |
1608 | }; |
1609 | if let Some(call_file) = func.call_file { |
1610 | if let Some(lines) = frames.unit.parse_lines(frames.sections)? { |
1611 | next.file = lines.files.get(call_file as usize).map(String::as_str); |
1612 | } |
1613 | } |
1614 | frames.next = Some(next); |
1615 | |
1616 | Ok(Some(Frame { |
1617 | dw_die_offset: Some(func.dw_die_offset), |
1618 | function: func.name.clone().map(|name| FunctionName { |
1619 | name, |
1620 | language: frames.unit.lang, |
1621 | }), |
1622 | location: loc, |
1623 | })) |
1624 | } |
1625 | } |
1626 | |
1627 | #[cfg (feature = "fallible-iterator" )] |
1628 | impl<'ctx, R> fallible_iterator::FallibleIterator for FrameIter<'ctx, R> |
1629 | where |
1630 | R: gimli::Reader + 'ctx, |
1631 | { |
1632 | type Item = Frame<'ctx, R>; |
1633 | type Error = Error; |
1634 | |
1635 | #[inline ] |
1636 | fn next(&mut self) -> Result<Option<Frame<'ctx, R>>, Error> { |
1637 | self.next() |
1638 | } |
1639 | } |
1640 | |
1641 | /// A function frame. |
1642 | pub struct Frame<'ctx, R: gimli::Reader> { |
1643 | /// The DWARF unit offset corresponding to the DIE of the function. |
1644 | pub dw_die_offset: Option<gimli::UnitOffset<R::Offset>>, |
1645 | /// The name of the function. |
1646 | pub function: Option<FunctionName<R>>, |
1647 | /// The source location corresponding to this frame. |
1648 | pub location: Option<Location<'ctx>>, |
1649 | } |
1650 | |
1651 | /// A function name. |
1652 | pub struct FunctionName<R: gimli::Reader> { |
1653 | /// The name of the function. |
1654 | pub name: R, |
1655 | /// The language of the compilation unit containing this function. |
1656 | pub language: Option<gimli::DwLang>, |
1657 | } |
1658 | |
1659 | impl<R: gimli::Reader> FunctionName<R> { |
1660 | /// The raw name of this function before demangling. |
1661 | pub fn raw_name(&self) -> Result<Cow<'_, str>, Error> { |
1662 | self.name.to_string_lossy() |
1663 | } |
1664 | |
1665 | /// The name of this function after demangling (if applicable). |
1666 | pub fn demangle(&self) -> Result<Cow<'_, str>, Error> { |
1667 | self.raw_name().map(|x: Cow<'_, str>| demangle_auto(name:x, self.language)) |
1668 | } |
1669 | } |
1670 | |
1671 | /// Demangle a symbol name using the demangling scheme for the given language. |
1672 | /// |
1673 | /// Returns `None` if demangling failed or is not required. |
1674 | #[allow (unused_variables)] |
1675 | pub fn demangle(name: &str, language: gimli::DwLang) -> Option<String> { |
1676 | match language { |
1677 | #[cfg (feature = "rustc-demangle" )] |
1678 | gimli::DW_LANG_Rust => rustc_demangle::try_demangle(name) |
1679 | .ok() |
1680 | .as_ref() |
1681 | .map(|x| format!(" {:#}" , x)), |
1682 | #[cfg (feature = "cpp_demangle" )] |
1683 | gimli::DW_LANG_C_plus_plus |
1684 | | gimli::DW_LANG_C_plus_plus_03 |
1685 | | gimli::DW_LANG_C_plus_plus_11 |
1686 | | gimli::DW_LANG_C_plus_plus_14 => cpp_demangle::Symbol::new(name) |
1687 | .ok() |
1688 | .and_then(|x| x.demangle(&Default::default()).ok()), |
1689 | _ => None, |
1690 | } |
1691 | } |
1692 | |
1693 | /// Apply 'best effort' demangling of a symbol name. |
1694 | /// |
1695 | /// If `language` is given, then only the demangling scheme for that language |
1696 | /// is used. |
1697 | /// |
1698 | /// If `language` is `None`, then heuristics are used to determine how to |
1699 | /// demangle the name. Currently, these heuristics are very basic. |
1700 | /// |
1701 | /// If demangling fails or is not required, then `name` is returned unchanged. |
1702 | pub fn demangle_auto(name: Cow<'_, str>, language: Option<gimli::DwLang>) -> Cow<'_, str> { |
1703 | match language { |
1704 | Some(language) => demangle(name.as_ref(), language), |
1705 | None => demangle(name.as_ref(), gimli::DW_LANG_Rust) |
1706 | .or_else(|| demangle(name.as_ref(), gimli::DW_LANG_C_plus_plus)), |
1707 | } |
1708 | .map(Cow::from) |
1709 | .unwrap_or(default:name) |
1710 | } |
1711 | |
1712 | /// A source location. |
1713 | pub struct Location<'a> { |
1714 | /// The file name. |
1715 | pub file: Option<&'a str>, |
1716 | /// The line number. |
1717 | pub line: Option<u32>, |
1718 | /// The column number. |
1719 | pub column: Option<u32>, |
1720 | } |
1721 | |
1722 | #[cfg (test)] |
1723 | mod tests { |
1724 | #[test ] |
1725 | fn context_is_send() { |
1726 | fn assert_is_send<T: Send>() {} |
1727 | assert_is_send::<crate::Context<gimli::read::EndianSlice<'_, gimli::LittleEndian>>>(); |
1728 | } |
1729 | } |
1730 | |