Replace with calls to blockvector::blocks, and the appropriate method
call on the returned array_view.
Change-Id: I04d1f39603e4d4c21c96822421431d9a029d8ddd
In this review [1], Eli pointed out that we should be careful when
concatenating file names to avoid duplicated slashes. On Windows, a
double slash at the beginning of a file path has a special meaning. So
naively concatenating "/" and "foo/bar" would give "//foo/bar", which
would not give the desired results. We already have a few spots doing:
if (first_path ends with a slash)
path = first_path + second_path
else
path = first_path + slash + second_path
In general, I think it's nice to avoid superfluous slashes in file
paths, since they might end up visible to the user and look a bit
unprofessional.
Introduce the path_join function that can be used to join multiple path
components together (along with unit tests).
I initially wanted to make it possible to join two absolute paths, to
support the use case of prepending a sysroot path to a target file path,
or the prepending the debug-file-directory to a target file path. But
the code in solib_find_1 shows that it is more complex than this anyway
(for example, when the right hand side is a Windows path with a drive
letter). So I don't think we need to support that case in path_join.
That also keeps the implementation simpler.
Change a few spots to use path_join to show how it can be used. I
believe that all the spots I changed are guarded by some checks that
ensure the right hand side operand is not an absolute path.
Regression-tested on Ubuntu 18.04. Built-tested on Windows, and I also
ran the new unit-test there.
[1] https://sourceware.org/pipermail/gdb-patches/2022-April/187559.html
Change-Id: I0df889f7e3f644e045f42ff429277b732eb6c752
Change this field to an std::vector to facilitate memory management.
Since the linetable_entry array is copied into the symtab resulting from
the subfile, it is possible to change it without changing how symtab
stores the linetable entries (which would be a much larger change).
There is a small change in buildsym_compunit::record_line to avoid
accessing a now invalid linetable_entry. Before this patch, we keep a
pointer to the last linetable entry, pop it from the vector, and then
read last->line. It works with the manually-maintained array, but since
we now use std::vector::pop_back, I am afraid that it could be flagged
as an invalid access by the various static / dynamic analysis tools to
access the linetable_entry object after popping it from the vector.
Instead, record just the line number in an optional and use it.
There are substantial changes in xcoffread.c that simplify the code, but
I can't test them. I was hesitant to do this change because of that,
but I decided to send it anyway. I don't think that an almost dead
platform should hold back improving the code in the common parts of GDB.
The changes in xcoffread.c are:
- Make arrange_linetable "arrange" the linetable passed as a parameter,
instead of returning possibly a new one, possibly the same one.
- In the "Process main file's line numbers.", I'm not too sure what
happens. We get the lintable from "main_subfile", "arrange" it, but
then assign the result to the current subfile, obtained with
get_current_subfile. I assume that the current subfile is also the
main one, so now I just call arrange_linetable on the main subfile's
line table.
- Remove that weird "Useless if!!!" FIXME comment. It's been there
forever, but the "if" is still there, so I guess the "if" can stay
there.
Change-Id: I11799006fd85189e8cf5bd3a168f8f38c2c27a80
Change subfile::name to be a string, for easier memory management.
Change buildsym_compunit::m_comp_dir as well, since we move one in to
the other at some point in patch_subfile_names, so it's easier to do
both at the same time. There are various NULL checks for both fields
currently, replace them with empty checks, I think it ends up
equivalent.
I can't test the change in xcoffread.c, it's best-effort.
Change-Id: I62b5fb08b2089e096768a090627ac7617e90a016
Allocate struct subfile with new, initialize its fields instead of
memset-ing it to 0. Use a unique_ptr for the window after a subfile has
been allocated but before it is linked in the buildsym_compunit's list
of subfile (and therefore owned by the buildsym_compunit.
I can't test the change in xcoffread.c, it's best-effort. I couldn't
find where subfiles are freed in that file, I assume they were
intentionally (or not) leaked.
Change-Id: Ib3b6877de31b7e65bc466682f08dbf5840225f24
Add support for DW_LNS_set_prologue_end when building line-tables. This
attribute can be set by the compiler to indicate that an instruction is
an adequate place to set a breakpoint just after the prologue of a
function.
The compiler might set multiple prologue_end, but considering how
current skip_prologue_using_sal works, this commit modifies it to accept
the first instruction with this marker (if any) to be the place where a
breakpoint should be placed to be at the end of the prologue.
The need for this support came from a problematic usecase generated by
hipcc (i.e. clang). The problem is as follows: There's a function
(lets call it foo) which covers PC from 0xa800 to 0xa950. The body of
foo begins with a call to an inlined function, covering from 0xa800 to
0xa94c. The issue is that when placing a breakpoint at 'foo', GDB
inserts the breakpoint at 0xa818. The 0x18 offset is what GDB thinks is
foo's first address past the prologue.
Later, when hitting the breakpoint, GDB reports the stop within the
inlined function because the PC falls in its range while the user
expects to stop in FOO.
Looking at the line-table for this location, we have:
INDEX LINE ADDRESS IS-STMT
[...]
14 293 0x000000000000a66c Y
15 END 0x000000000000a6e0 Y
16 287 0x000000000000a800 Y
17 END 0x000000000000a818 Y
18 287 0x000000000000a824 Y
[...]
For comparison, let's look at llvm-dwarfdump's output for this CU:
Address Line Column File ISA Discriminator Flags
------------------ ------ ------ ------ --- ------------- -------------
[...]
0x000000000000a66c 293 12 2 0 0 is_stmt
0x000000000000a6e0 96 43 82 0 0 is_stmt
0x000000000000a6f8 102 18 82 0 0 is_stmt
0x000000000000a70c 102 24 82 0 0
0x000000000000a710 102 18 82 0 0
0x000000000000a72c 101 16 82 0 0 is_stmt
0x000000000000a73c 2915 50 83 0 0 is_stmt
0x000000000000a74c 110 1 1 0 0 is_stmt
0x000000000000a750 110 1 1 0 0 is_stmt end_sequence
0x000000000000a800 107 0 1 0 0 is_stmt
0x000000000000a800 287 12 2 0 0 is_stmt prologue_end
0x000000000000a818 114 59 81 0 0 is_stmt
0x000000000000a824 287 12 2 0 0 is_stmt
0x000000000000a828 100 58 82 0 0 is_stmt
[...]
The main difference we are interested in here is that llvm-dwarfdump's
output tells us that 0xa800 is an adequate place to place a breakpoint
past a function prologue. Since we know that foo covers from 0xa800 to
0xa94c, 0xa800 is the address at which the breakpoint should be placed
if the user wants to break in foo.
This commit proposes to add support for the prologue_end flag in the
line-program processing.
The processing of this prologue_end flag is made in skip_prologue_sal,
before it calls gdbarch_skip_prologue_noexcept. The intent is that if
the compiler gave information on where the prologue ends, we should use
this information and not try to rely on architecture dependent logic to
guess it.
The testsuite have been executed using this patch on GNU/Linux x86_64.
Testcases have been compiled with both gcc/g++ (verison 9.4.0) and
clang/clang++ (version 10.0.0) since at the time of writing GCC does not
set the prologue_end marker. Tests done with GCC 11.2.0 (not over the
entire testsuite) show that it does not emit this flag either.
No regression have been observed with GCC or Clang. Note that when
using Clang, this patch fixes a failure in
gdb.opt/inline-small-func.exp.
Change-Id: I720449a8a9b2e1fb45b54c6095d3b1e9da9152f8
Currently when recording a line entry (with
buildsym_compunit::record_line), a boolean argument argument is used to
indicate that the is_stmt flag should be set for this particular record.
As a later commit will add support for new flags, instead of adding a
parameter to record_line for each possible flag, transform the current
is_stmt parameter into a enum flag. This flags parameter will allow
greater flexibility in future commits.
This enum flags type is not propagated into the linetable_entry type as
this would require a lot of changes across the codebase for no practical
gain (it currently uses a bitfield where each interesting flag only
occupy 1 bit in the structure).
Tested on x86_64-linux, no regression observed.
Change-Id: I5d061fa67bdb34918742505ff983d37453839d6a
It's a bit confusing because we have both "compunit_symtab" and "symtab"
types, and many methods and functions containing "start_symtab" or
"end_symtab", which actually deal with compunit_symtabs. I believe this
comes from the time before compunit_symtab was introduced, where
symtab did the job of both.
Rename everything I found containing start_symtab or end_symtab to use
start_compunit_symtab or end_compunit_symtab.
Change-Id: If3849b156f6433640173085ad479b6a0b085ade2
Add a getter and a setter for a symbol's type. Remove the corresponding
macro and adjust all callers.
Change-Id: Ie1a137744c5bfe1df4d4f9ae5541c5299577c8de
Add a getter and a setter for whether a symbol is an argument. Remove
the corresponding macro and adjust all callers.
Change-Id: I71b4f0465f3dfd2ed8b9e140bd3f7d5eb8d9ee81
Add a getter and a setter for a symtab's language. Remove the
corresponding macro and adjust all callers.
Change-Id: I9f4d840b11c19f80f39bac1bce020fdd1739e11f
Add a getter and a setter for a symtab's linetable. Remove the
corresponding macro and adjust all callers.
Change-Id: I159183fc0ccd8e18ab937b3c2f09ef2244ec6e9c
Add a getter and a setter for a compunit_symtab's macro table. Remove the
corresponding macro and adjust all callers.
Change-Id: I00615ea72d5ac43d9a865e941cb2de0a979c173a
Add a getter and a setter for a compunit_symtab's block line section. Remove
the corresponding macro and adjust all callers.
Change-Id: I3eb1a323388ad55eae8bfa45f5bc4a08dc3df455
Add a getter and a setter for a compunit_symtab's blockvector. Remove
the corresponding macro and adjust all callers.
Change-Id: I99484c6619dcbbea7c5d89c72aa660316ca62f64
Add a getter and a setter for a compunit_symtab's dirname. Remove the
corresponding macro and adjust all callers.
Change-Id: If2f39b295fd26822586485e04a8b8b5aa5cc9b2e
Add a getter and a setter for a compunit_symtab's producer. Remove the
corresponding macro and adjust all callers.
Change-Id: Ia1d6d8a0e247a08a21af23819d71e49b37d8931b
Add a getter and a setter for a compunit_symtab's debugformat. Remove
the corresponding macro and adjust all callers.
Change-Id: I1667b02d5322346f8e23abd9f8a584afbcd75975
I think that most remaining uses of COMPUNIT_FILETABS intend to get the
primary filetab of the compunit_symtab specifically (and not to iterate
over all filetabs, for example, those cases would use compunit_filetabs,
which has been converted to compunit_symtab::filetabs), so replace mosts
uses with compunit_symtab::primary_filetab.
In jit.c, function finalize_symtab, we can save the symtab object
returned by allocate_symtab and use it, it makes things simpler.
Change-Id: I4e51d6d4b40759de8768b61292e5e13c8eae2e38
Add a method to set the primary filetab of the CU. This is currently
done in buildsym_compunit::end_symtab_with_blockvector.
Change-Id: I16c51a6b90a4cb4c0c5f183b0f2e12bc64b6fd74
This commit brings all the changes made by running gdb/copyright.py
as per GDB's Start of New Year Procedure.
For the avoidance of doubt, all changes in this commits were
performed by the script.
While working on PR26935 I noticed that when running test-case
gdb.base/morestack.exp with target board unix/-m32/-fPIE/-pie and ld linker,
I get this linetable fragment for morestack.S using readelf -wL:
...
CU: ../../../../libgcc/config/i386/morestack.S:
Line number Starting address View Stmt
109 0xc9c x
...
838 0xe03 x
- 0xe04
636 0 x
637 0x3 x
- 0x4
...
but with "maint info line-table" I get:
...
INDEX LINE ADDRESS IS-STMT
0 END 0x00000004 Y
1 109 0x00000c9c Y
...
110 838 0x00000e03 Y
111 END 0x00000e04 Y
...
So, apparently the entries with addresses 0x0 and 0x3 are filtered out
because the addresses are out of range, but the same doesn't happen with the
end-of-seq terminator.
Fix this by filtering out end-of-seq terminators that do not actually
terminate anything.
Tested on x86_64-linux.
gdb/ChangeLog:
2021-01-04 Tom de Vries <tdevries@suse.de>
* buildsym.c (buildsym_compunit::record_line): Filter out end-of-seq
terminators that do not terminate anything.
gdb/testsuite/ChangeLog:
2021-01-04 Tom de Vries <tdevries@suse.de>
* gdb.dwarf2/dw2-out-of-range-end-of-seq.exp: New file.
This commits the result of running gdb/copyright.py as per our Start
of New Year procedure...
gdb/ChangeLog
Update copyright year range in copyright header of all GDB files.
Consider test-case gdb.dwarf2/dw2-ranges-base.exp. It has (ignoring
non-sensical entries that are filtered out by buildsym_compunit::record_line)
a line-table for dw2-ranges-base.c like this:
...
Line Number Statements:
[0x0000014e] Extended opcode 2: set Address to 0x4004ba
[0x00000159] Advance Line by 10 to 11
[0x0000015b] Copy
[0x0000015c] Advance PC by 12 to 0x4004c6
[0x0000015e] Extended opcode 1: End of Sequence
[0x00000161] Extended opcode 2: set Address to 0x4004ae
[0x0000016c] Advance Line by 20 to 21
[0x0000016e] Copy
[0x0000016f] Advance PC by 12 to 0x4004ba
[0x00000171] Extended opcode 1: End of Sequence
[0x00000174] Extended opcode 2: set Address to 0x4004a7
[0x0000017f] Advance Line by 30 to 31
[0x00000181] Copy
[0x00000182] Advance PC by 7 to 0x4004ae
[0x00000184] Extended opcode 1: End of Sequence
...
If we disable the sorting in buildsym_compunit::end_symtab_with_blockvector,
we have the unsorted line table:
...
INDEX LINE ADDRESS IS-STMT
0 11 0x00000000004004ba Y
1 END 0x00000000004004c6 Y
2 21 0x00000000004004ae Y
3 END 0x00000000004004ba Y
4 31 0x00000000004004a7 Y
5 END 0x00000000004004ae Y
...
It contains 3 sequences, 11/END, 21/END and 31/END.
We want to sort the 3 sequences relative to each other, while sorting on
address, to get:
...
INDEX LINE ADDRESS IS-STMT
0 31 0x00000000004004a7 Y
1 END 0x00000000004004ae Y
2 21 0x00000000004004ae Y
3 END 0x00000000004004ba Y
4 11 0x00000000004004ba Y
5 END 0x00000000004004c6 Y
...
However, if we re-enable the sorting, we have instead:
...
INDEX LINE ADDRESS IS-STMT
0 31 0x00000000004004a7 Y
1 21 0x00000000004004ae Y
2 END 0x00000000004004ae Y
3 11 0x00000000004004ba Y
4 END 0x00000000004004ba Y
5 END 0x00000000004004c6 Y
...
This is a regression since commit 3d92a3e313 "gdb: Don't reorder line table
entries too much when sorting", that introduced sorting on address while
keeping entries with the same address in pre-sort order.
Indeed the entries 1 and 2 are in pre-sort order (they map to entries 2 and 5
in the unsorted line table), but entry 1 does not belong in the sequence
terminated by 2.
Fix this by handling End-Of-Sequence entries in the sorting function, such
that they are sorted before other entries with the same address.
Also, revert the find_pc_sect_line workaround introduced in commit 3d92a3e313,
since that's no longer necessary.
Tested on x86_64-linux.
gdb/ChangeLog:
2020-07-06 Tom de Vries <tdevries@suse.de>
* buildsym.c (buildsym_compunit::end_symtab_with_blockvector): Handle
End-Of-Sequence in lte_is_less_than.
* symtab.c (find_pc_sect_line): Revert change from commit 3d92a3e313
"gdb: Don't reorder line table entries too much when sorting".
gdb/testsuite/ChangeLog:
2020-07-06 Tom de Vries <tdevries@suse.de>
* gdb.dwarf2/dw2-ranges-base.exp: Test line-table order.
Add the `type` and `set_type` methods on `struct field`, in order to
remoremove the `FIELD_TYPE` macro. In this patch, the `FIELD_TYPE`
macro is changed to use `field::type`, so all the call sites that are
useused to set the field's type are changed to use `field::set_type`.
The next patch will remove `FIELD_TYPE` completely.
Note that because of the name clash between the existing field named
`type` and the new method, I renamed the field `m_type`. It is not
private per-se, because we can't make `struct field` a non-POD yet, but
it should be considered private anyway (not accessed outside `struct
field`).
gdb/ChangeLog:
* gdbtypes.h (struct field) <type, set_type>: New methods.
Rename `type` field to...
<m_type>: ... this. Change references throughout to use type or
set_type methods.
(FIELD_TYPE): Use field::type. Change call sites that modify
the field's type to use field::set_type instead.
Change-Id: Ie21f866e3b7f8a51ea49b722d07d272a724459a0
Add the `fields` and `set_fields` methods on `struct type`, in order to
remove the `TYPE_FIELDS` macro. In this patch, the `TYPE_FIELDS` macro
is changed to the `type::fields`, so all the call sites that use it to
set the fields array are changed to use `type::set_fields`. The next
patch will remove `TYPE_FIELDS` entirely.
gdb/ChangeLog:
* gdbtypes.h (struct type) <fields, set_fields>: New methods.
(TYPE_FIELDS): Use type::fields. Change all call sites that
modify the propery to use type::set_fields instead.
Change-Id: I05174ce68f2ce3fccdf5d8b469ff141f14886b33
Remove `TYPE_NFIELDS`, changing all the call sites to use
`type::num_fields` directly. This is quite a big diff, but this was
mostly done using sed and coccinelle. A few call sites were done by
hand.
gdb/ChangeLog:
* gdbtypes.h (TYPE_NFIELDS): Remove. Change all cal sites to use
type::num_fields instead.
Change-Id: Ib73be4c36f9e770e0f729bac3b5257d7cb2f9591
Add the `num_fields` and `set_num_fields` methods on `struct type`, in
order to remove the `TYPE_NFIELDS` macro. In this patch, the
`TYPE_NFIELDS` macro is changed to use `type::num_fields`, so all the
call sites that are used to set the number of fields are changed to use
`type::set_num_fields`. The next patch will remove `TYPE_NFIELDS`
completely.
I think that in the future, we should consider making the interface of
`struct type` better. For example, right now it's possible for the
number of fields property and the actual number of fields set to be out
of sync. However, I want to keep the existing behavior in this patch,
just translate from macros to methods.
gdb/ChangeLog:
* gdbtypes.h (struct type) <num_fields, set_num_fields>: New
methods.
(TYPE_NFIELDS): Use type::num_fields. Change all call sites
that modify the number of fields to use type::set_num_fields
instead.
Change-Id: I5ad9de5be4097feaf942d111077434bf91d13dc5
This reverts the following commit partially:
commit 64dc2d4bd2
Author: Bernd Edlinger <bernd.edlinger@hotmail.de>
Date: Thu Mar 12 11:52:34 2020 +0100
Fix an undefined behavior in record_line
Additionally do not completely remove symbols
at the same PC than the end marker, instead
make them non-is-stmt breakpoints.
We keep the undefined behavoir fix,
but have to restore the original behavior
regarding deletion of the line entries.
2020-04-09 Bernd Edlinger <bernd.edlinger@hotmail.de>
revert partially:
2020-04-01 Bernd Edlinger <bernd.edlinger@hotmail.de>
* buildsym.c (record_line): Fix undefined behavior and preserve
lines at eof.
In this commit:
commit 8c95582da8
Date: Mon Dec 30 21:04:51 2019 +0000
gdb: Add support for tracking the DWARF line table is-stmt field
A change was made in buildsym_compunit::record_line to remove
duplicate line table entries in some cases. This was an invalid
change, as these duplicate line table entries are used in _some_ cases
as part of prologue detection (see skip_prologue_using_sal).
It might be possible to identify those line table entries that are
required by skip_prologue_using_sal and only keep those duplicates
around, however, I have not done this here. The original duplicate
removal was done because (a) it was easy to implement, and (b) it
seemed obviously harmless.
As (b) is now known to be false, and implementation would be more
complex, and so (a) is also false. As such, it seems better to keep
all duplicates until an actual reason presents itself for why we
should remove any.
The original regression was spotted on RISC-V, which makes use of
skip_prologue_using_sal as part of riscv_skip_prologue. Originally I
created the test gdb.dwarf2/dw2-inline-small-func.exp, however, this
test will not compile on RISC-V as this target doesn't support
.uleb128 or .sleb128 assembler directives containing complex
expressions. As a result I added the gdb.opt/inline-small-func.exp
test, which exposes the bug on RISC-V, but obviously depends on the
compiler to produce specific DWARF information in order to expose the
bug. Still this test does ensure we always get the desired result,
even if the DWARF changes.
Originally the gdb.dwarf2/dw2-inline-small-func.exp test passed on
x86-64 even with the duplicate line table entries incorrectly
removed. The reason for this is that when a compilation unit doesn't
have a 'producer' string then skip_prologue_using_sal is not used,
instead the prologue is always skipped using analysis of the assembler
code.
However, for Clang on x86-64 skip_prologue_using_sal is used, so I
modified the gdb.dwarf2/dw2-inline-small-func.exp test to include a
'producer' string that names the Clang compiler. With this done the
test would fail on x86-64.
One thing to note is that the gdb.opt/inline-small-func.exp test might
fail on some targets. For example, if we compare sparc to risc-v by
looking at sparc32_skip_prologue we see that this function doesn't use
skip_prologue_using_sal, but instead uses find_pc_partial_function
directly. I don't know the full history behind why the code is like
it is, but it feels like sparc32_skip_prologue is an attempt to
duplicate some of the functionality of skip_prologue_using_sal, but
without all of the special cases. If this is true then the new test
could easily fail on this target, this would suggest that sparc should
consider switching to use skip_prologue_using_sal like risc-v does.
gdb/ChangeLog:
* buildsym.c (buildsym_compunit::record_line): Remove
deduplication code.
gdb/testsuite/ChangeLog:
* gdb.dwarf2/dw2-inline-small-func-lbls.c: New file.
* gdb.dwarf2/dw2-inline-small-func.c: New file.
* gdb.dwarf2/dw2-inline-small-func.exp: New file.
* gdb.dwarf2/dw2-inline-small-func.h: New file.
* gdb.opt/inline-small-func.c: New file.
* gdb.opt/inline-small-func.exp: New file.
* gdb.opt/inline-small-func.h: New file.
Additionally do not completely remove symbols
at the same PC than the end marker, instead
make them non-is-stmt breakpoints.
2020-04-01 Bernd Edlinger <bernd.edlinger@hotmail.de>
* buildsym.c (record_line): Fix undefined behavior and preserve
lines at eof.
This commit:
commit 8c95582da8
Date: Mon Dec 30 21:04:51 2019 +0000
gdb: Add support for tracking the DWARF line table is-stmt field
Introduced an invalid memory access, by reading outside the bounds of
an array.
This would cause this valgrind error:
==7633== Invalid read of size 4
==7633== at 0x4D002C: buildsym_compunit::record_line(subfile*, int, unsigned long, bool) (buildsym.c:688)
==7633== by 0x5F60A5: dwarf_record_line_1(gdbarch*, subfile*, unsigned int, unsigned long, bool, dwarf2_cu*) (read.c:19956)
==7633== by 0x5F63B0: lnp_state_machine::record_line(bool) (read.c:20024)
==7633== by 0x5F5DD5: lnp_state_machine::handle_special_opcode(unsigned char) (read.c:19851)
==7633== by 0x5F6706: dwarf_decode_lines_1(line_header*, dwarf2_cu*, int, unsigned long) (read.c:20135)
==7633== by 0x5F6C57: dwarf_decode_lines(line_header*, char const*, dwarf2_cu*, dwarf2_psymtab*, unsigned long, int) (read.c:20328)
==7633== by 0x5DF5F1: handle_DW_AT_stmt_list(die_info*, dwarf2_cu*, char const*, unsigned long) (read.c:10748)
==7633== by 0x5DF823: read_file_scope(die_info*, dwarf2_cu*) (read.c:10796)
==7633== by 0x5DDA63: process_die(die_info*, dwarf2_cu*) (read.c:9815)
==7633== by 0x5DD44A: process_full_comp_unit(dwarf2_per_cu_data*, language) (read.c:9580)
==7633== by 0x5DAB58: process_queue(dwarf2_per_objfile*) (read.c:8867)
==7633== by 0x5CB30E: dw2_do_instantiate_symtab(dwarf2_per_cu_data*, bool) (read.c:2374)
==7633== Address 0xa467f48 is 8 bytes before a block of size 16,024 alloc'd
==7633== at 0x4C2CDCB: malloc (vg_replace_malloc.c:299)
==7633== by 0x451FC4: xmalloc (alloc.c:60)
==7633== by 0x4CFFDF: buildsym_compunit::record_line(subfile*, int, unsigned long, bool) (buildsym.c:678)
==7633== by 0x5F60A5: dwarf_record_line_1(gdbarch*, subfile*, unsigned int, unsigned long, bool, dwarf2_cu*) (read.c:19956)
==7633== by 0x5F63B0: lnp_state_machine::record_line(bool) (read.c:20024)
==7633== by 0x5F5DD5: lnp_state_machine::handle_special_opcode(unsigned char) (read.c:19851)
==7633== by 0x5F6706: dwarf_decode_lines_1(line_header*, dwarf2_cu*, int, unsigned long) (read.c:20135)
==7633== by 0x5F6C57: dwarf_decode_lines(line_header*, char const*, dwarf2_cu*, dwarf2_psymtab*, unsigned long, int) (read.c:20328)
==7633== by 0x5DF5F1: handle_DW_AT_stmt_list(die_info*, dwarf2_cu*, char const*, unsigned long) (read.c:10748)
==7633== by 0x5DF823: read_file_scope(die_info*, dwarf2_cu*) (read.c:10796)
==7633== by 0x5DDA63: process_die(die_info*, dwarf2_cu*) (read.c:9815)
==7633== by 0x5DD44A: process_full_comp_unit(dwarf2_per_cu_data*, language) (read.c:9580)
gdb/ChangeLog:
* buildsyms.c (buildsym_compunit::record_line): Avoid accessing
previous item in the list, when the list has no items.
This commit brings support for the DWARF line table is_stmt field to
GDB. The is_stmt field is used by the compiler when a single source
line is split into multiple assembler instructions, especially if the
assembler instructions are interleaved with instruction from other
source lines.
The compiler will set the is_stmt flag false from some instructions
from the source lines, these instructions are not a good place to
insert a breakpoint in order to stop at the source line.
Instructions which are marked with the is_stmt flag true are a good
place to insert a breakpoint for that source line.
Currently GDB ignores all instructions for which is_stmt is false.
This is fine in a lot of cases, however, there are some cases where
this means the debug experience is not as good as it could be.
Consider stopping at a random instruction, currently this instruction
will be attributed to the last line table entry before this point for
which is_stmt was true - as these are the only line table entries that
GDB tracks. This can easily be incorrect in code with even a low
level of optimisation.
With is_stmt tracking in place, when stopping at a random instruction
we now attribute the instruction back to the real source line, even
when is_stmt is false for that instruction in the line table.
When inserting breakpoints we still select line table entries for
which is_stmt is true, so the breakpoint placing behaviour should not
change.
When stepping though code (at the line level, not the instruction
level) we will still stop at instruction where is_stmt is true, I
think this is more likely to be the desired behaviour.
Instruction stepping is, of course, unchanged, stepping one
instruction at a time, but we should now report more accurate line
table information with each instruction step.
The original motivation for this work was a patch posted by Bernd
here:
https://sourceware.org/ml/gdb-patches/2019-11/msg00792.html
As part of that thread it was suggested that many issues would be
resolved if GDB supported line table views, this isn't something I've
attempted in this patch, though reading the spec, it seems like this
would be a useful feature to support in GDB in the future. The spec
is here:
http://dwarfstd.org/ShowIssue.php?issue=170427.1
And Bernd gives a brief description of the benefits here:
https://sourceware.org/ml/gdb-patches/2020-01/msg00147.html
With that all said, I think that there is benefit to having proper
is_stmt support regardless of whether we have views support, so I
think we should consider getting this in first, and then building view
support on top of this.
The gdb.cp/step-and-next-inline.exp test is based off a test proposed
by Bernd Edlinger in this message:
https://sourceware.org/ml/gdb-patches/2019-12/msg00842.html
gdb/ChangeLog:
* buildsym-legacy.c (record_line): Pass extra parameter to
record_line.
* buildsym.c (buildsym_compunit::record_line): Take an extra
parameter, reduce duplication in the line table, and record the
is_stmt flag in the line table.
* buildsym.h (buildsym_compunit::record_line): Add extra
parameter.
* disasm.c (do_mixed_source_and_assembly_deprecated): Ignore
non-statement lines.
* dwarf2/read.c (dwarf_record_line_1): Add extra parameter, pass
this to the symtab builder.
(dwarf_finish_line): Pass extra parameter to dwarf_record_line_1.
(lnp_state_machine::record_line): Pass a suitable is_stmt flag
through to dwarf_record_line_1.
* infrun.c (process_event_stop_test): When stepping, don't stop at
a non-statement instruction, and only refresh the step info when
we land in the middle of a line's range. Also add an extra
comment.
* jit.c (jit_symtab_line_mapping_add_impl): Initialise is_stmt
field.
* record-btrace.c (btrace_find_line_range): Only record lines
marked as is-statement.
* stack.c (frame_show_address): Show the frame address if we are
in a non-statement sal.
* symmisc.c (dump_symtab_1): Print the is_stmt flag.
(maintenance_print_one_line_table): Print a header for the is_stmt
column, and include is_stmt information in the output.
* symtab.c (find_pc_sect_line): Find lines marked as statements in
preference to non-statements.
(find_pcs_for_symtab_line): Prefer is-statement entries.
(find_line_common): Likewise.
* symtab.h (struct linetable_entry): Add is_stmt field.
(struct symtab_and_line): Likewise.
* xcoffread.c (arrange_linetable): Initialise is_stmt field when
arranging the line table.
gdb/testsuite/ChangeLog:
* gdb.cp/step-and-next-inline.cc: New file.
* gdb.cp/step-and-next-inline.exp: New file.
* gdb.cp/step-and-next-inline.h: New file.
* gdb.dwarf2/dw2-is-stmt.c: New file.
* gdb.dwarf2/dw2-is-stmt.exp: New file.
* gdb.dwarf2/dw2-is-stmt-2.c: New file.
* gdb.dwarf2/dw2-is-stmt-2.exp: New file.
* gdb.dwarf2/dw2-ranges-base.exp: Update line table pattern.
This introduces a string cache on the per-BFD object, replacing the
macro and filename caches. Both of these caches just store strings,
so this consolidation by itself saves a little memory (about the size
of a bcache per objfile).
Then this patch switches some allocations on the objfile obstack to
use this bcache instead. This saves more space; and turns out to be a
bit faster as well.
Here are the before and after "maint time" + "maint space" results of
"file ./gdb":
Command execution time: 4.664021 (cpu), 4.728518 (wall)
Space used: 39190528 (+29212672 for this command)
Command execution time: 4.216209 (cpu), 4.107023 (wall)
Space used: 36667392 (+26689536 for this command)
The main interface to the string cache is a new pair of overloaded
methods, objfile::intern.
gdb/ChangeLog
2020-03-04 Tom Tromey <tom@tromey.com>
* symmisc.c (print_symbol_bcache_statistics)
(print_objfile_statistics): Update.
* symfile.c (allocate_symtab): Use intern.
* psymtab.c (partial_symtab::partial_symtab): Use intern.
* objfiles.h (struct objfile_per_bfd_storage) <filename_cache,
macro_cache>: Remove.
<string_cache>: New member.
(struct objfile) <intern>: New methods.
* elfread.c (elf_symtab_read): Use intern.
* dwarf2/read.c (fixup_go_packaging): Intern package name.
(dwarf2_compute_name, dwarf2_physname)
(create_dwo_unit_in_dwp_v1, create_dwo_unit_in_dwp_v2): Intern
names.
(guess_partial_die_structure_name): Update.
(partial_die_info::fixup): Intern name.
(dwarf2_canonicalize_name): Change parameter to objfile. Intern
name.
(dwarf2_name): Intern name. Update.
* buildsym.c (buildsym_compunit::get_macro_table): Use
string_cache.
Don't reorder line table entries for the same address when sorting the
line table, maintain the compiler given line order. Usually this will
reflect the order in which lines are conceptually encountered at a
given address.
Consider this example:
/* 1 */ volatile int global_var;
/* 2 */ int __attribute__ ((noinline))
/* 3 */ bar ()
/* 4 */ {
/* 5 */ return global_var;
/* 6 */ }
/* 7 */ static inline int __attribute__ ((always_inline))
/* 8 */ foo ()
/* 9 */ {
/* 10 */ return bar ();
/* 11 */ }
/* 12 */ int
/* 13 */ main ()
/* 14 */ {
/* 15 */ global_var = 0;
/* 16 */ return foo ();
/* 17 */ }
GCC 10 currently generates a line table like this (as shown by
objdump):
CU: ./test.c:
File name Line number Starting address
test.c 4 0x4004b0
test.c 5 0x4004b0
test.c 6 0x4004b6
test.c 6 0x4004b7
test.c 14 0x4003b0
test.c 15 0x4003b0
test.c 16 0x4003ba
test.c 10 0x4003ba
test.c 10 0x4003c1
The interesting entries are those for lines 16 and 10 at address
0x4003ba, these represent the call to foo and the inlined body of
foo.
With the current line table sorting GDB builds the line table like
this (as shown by 'maintenance info line-table'):
INDEX LINE ADDRESS
0 14 0x00000000004003b0
1 15 0x00000000004003b0
2 10 0x00000000004003ba
3 16 0x00000000004003ba
4 END 0x00000000004003c1
5 4 0x00000000004004b0
6 5 0x00000000004004b0
7 END 0x00000000004004b7
Notice that entries 2 and 3 for lines 10 and 16 are now in a different
order to the line table as given by the compiler. With this patch
applied the order is now:
INDEX LINE ADDRESS
0 14 0x00000000004003b0
1 15 0x00000000004003b0
2 16 0x00000000004003ba
3 10 0x00000000004003ba
4 END 0x00000000004003c1
5 4 0x00000000004004b0
6 5 0x00000000004004b0
7 END 0x00000000004004b7
Notice that entries 2 and 3 are now in their original order again.
The consequence of the incorrect ordering is that when stepping
through inlined functions GDB will display the wrong line for the
inner most frame. Here's a GDB session before this patch is applied:
Starting program: /home/andrew/tmp/inline/test
Temporary breakpoint 1, main () at test.c:15
15 /* 15 */ global_var = 0;
(gdb) step
16 /* 16 */ return foo ();
(gdb) step
foo () at test.c:16
16 /* 16 */ return foo ();
(gdb) step
bar () at test.c:5
5 /* 5 */ return global_var;
The step from line 15 to 16 was fine, but the next step should have
taken us to line 10, instead we are left at line 16. The final step
to line 5 is as expected.
With this patch applied the session goes better:
Starting program: /home/andrew/tmp/inline/test
Temporary breakpoint 1, main () at test.c:15
15 /* 15 */ global_var = 0;
(gdb) step
16 /* 16 */ return foo ();
(gdb) step
foo () at test.c:10
10 /* 10 */ return bar ();
(gdb) step
bar () at test.c:5
5 /* 5 */ return global_var;
We now visit the lines as 15, 16, 10, 5 as we would like.
The reason for this issue is that the inline frame unwinder is
detecting that foo is inlined in main. When we stop at the shared
address 0x4003ba the inline frame unwinder first shows us the outer
frame, this information is extracted from the DWARF's
DW_TAG_inlined_subroutine entries and passed via GDB's block data.
When we step again the inlined frame unwinder moves us up the call
stack to the inner most frame at which point the frame is displayed as
normal, with the location for the address being looked up in the line
table.
As GDB uses the last line table entry for an address as "the" line to
report for that address it is critical that GDB maintain the order of
the line table entries. In the first case, by reordering the line
table we report the wrong location.
I had to make a small adjustment in find_pc_sect_line in order to
correctly find the previous line in the line table. In some line
tables I was seeing an actual line entry and an end of sequence marker
at the same address, before this commit these would reorder to move
the end of sequence marker before the line entry (end of sequence has
line number 0). Now the end of sequence marker remains in its correct
location, and in order to find a previous line we should step backward
over any end of sequence markers.
As an example, the binary:
gdb/testsuite/outputs/gdb.dwarf2/dw2-ranges-func/dw2-ranges-func-lo-cold
Has this line table before the patch:
INDEX LINE ADDRESS
0 48 0x0000000000400487
1 END 0x000000000040048e
2 52 0x000000000040048e
3 54 0x0000000000400492
4 56 0x0000000000400497
5 END 0x000000000040049a
6 62 0x000000000040049a
7 END 0x00000000004004a1
8 66 0x00000000004004a1
9 68 0x00000000004004a5
10 70 0x00000000004004aa
11 72 0x00000000004004b9
12 END 0x00000000004004bc
13 76 0x00000000004004bc
14 78 0x00000000004004c0
15 80 0x00000000004004c5
16 END 0x00000000004004cc
And after this patch:
INDEX LINE ADDRESS
0 48 0x0000000000400487
1 52 0x000000000040048e
2 END 0x000000000040048e
3 54 0x0000000000400492
4 56 0x0000000000400497
5 END 0x000000000040049a
6 62 0x000000000040049a
7 66 0x00000000004004a1
8 END 0x00000000004004a1
9 68 0x00000000004004a5
10 70 0x00000000004004aa
11 72 0x00000000004004b9
12 END 0x00000000004004bc
13 76 0x00000000004004bc
14 78 0x00000000004004c0
15 80 0x00000000004004c5
16 END 0x00000000004004cc
When calling find_pc_sect_line with the address 0x000000000040048e, in
both cases we find entry #3, we then try to find the previous entry,
which originally found this entry '2 52 0x000000000040048e',
after the patch it finds '2 END 0x000000000040048e', which
cases the lookup to fail.
By skipping the END marker after this patch we get back to the correct
entry, which is now #1: '1 52 0x000000000040048e', and
everything works again.
gdb/ChangeLog:
* buildsym.c (lte_is_less_than): Delete.
(buildsym_compunit::end_symtab_with_blockvector): Create local
lambda function to sort line table entries, and use
std::stable_sort instead of std::sort.
* symtab.c (find_pc_sect_line): Skip backward over end of sequence
markers when looking for a previous line.
gdb/testsuite/ChangeLog:
* gdb.dwarf2/dw2-inline-stepping.c: New file.
* gdb.dwarf2/dw2-inline-stepping.exp: New file.
Change-Id: Ia0309494be4cfd9dcc554f30209477f5f040b21b
This removes code that was present from the very first git revisison
7b4ac7e1ed from 1988. It was in the
gdb/dbxread.c at the time (and makes more sense for dbx line info format
since line numbers are 16-bit entities in that debug format and debugging
files with more than 65535 lines would not work anyway) but moved from
there to gdb/buildsym.c which is used for dwarf line info as well, and
excluding an arbitrary line number does certainly not make sense nowadays.
Add a test case for line 65535
gdb:
2019-12-29 Bernd Edlinger <bernd.edlinger@hotmail.de>
* buildsym.c (buildsym_compunit::record_line): Do no longer ignore
line 65535.
gdb/testsuite:
2019-12-29 Bernd Edlinger <bernd.edlinger@hotmail.de>
* gdb.base/line65535.exp: New file.
* gdb.base/line65535.c: New file.
This also renames it to make it clearer that this is not a cheap
function (to compute_and_set_names). Also renames name to m_name
to make the implementation of the renamed function more readable.
Most of the places that access sym->m_name directly were also changed
to call linkage_name () instead, to make it clearer which name they
are accessing.
gdb/ChangeLog:
2019-12-26 Christian Biesinger <cbiesinger@google.com>
* ada-lang.c (ada_decode_symbol): Update.
* buildsym.c (add_symbol_to_list): Update.
* coffread.c (process_coff_symbol): Update.
* ctfread.c (ctf_add_enum_member_cb): Update.
(new_symbol): Update.
(ctf_add_var_cb): Update.
* dwarf2read.c (fixup_go_packaging): Update.
(dwarf2_compute_name): Update.
(new_symbol): Update.
* jit.c (finalize_symtab): Update.
* language.c (language_alloc_type_symbol): Update.
* mdebugread.c (new_symbol): Update.
* minsyms.c (minimal_symbol_reader::record_full): Update.
(minimal_symbol_reader::install): Update.
* psymtab.c (print_partial_symbols): Update.
(psymbol_hash): Update.
(psymbol_compare): Update.
(add_psymbol_to_bcache): Update.
(maintenance_check_psymtabs): Update.
* stabsread.c (define_symbol): Update.
* symtab.c (symbol_set_names): Rename to...
(general_symbol_info::compute_and_set_names): ...this.
(general_symbol_info::natural_name): Update.
(general_symbol_info::search_name): Update.
(fixup_section): Update.
* symtab.h (struct general_symbol_info) <name>: Rename to...
<m_name>: ...this.
<compute_and_set_names>: Rename from...
(symbol_set_names): ...this.
(SYMBOL_SET_NAMES): Remove.
(struct symbol) <ctor>: Update.
Change-Id: I8da1f10cab4e0b89f19d5750fa4e6e2ac8d2b24f
