diff options
author | Simon Marchi <simon.marchi@efficios.com> | 2020-06-17 14:48:46 -0400 |
---|---|---|
committer | Simon Marchi <simon.marchi@efficios.com> | 2020-06-17 14:49:33 -0400 |
commit | efb763a5ea351f9d865cbe491909f03472ebf2d6 (patch) | |
tree | 6f6688b9a07c55804ca77b75706638fcf8f53acd /gdb/dwarf2 | |
parent | 2951f6c068f961a2ea1de892fc82cf0229af67da (diff) | |
download | gdb-efb763a5ea351f9d865cbe491909f03472ebf2d6.zip gdb-efb763a5ea351f9d865cbe491909f03472ebf2d6.tar.gz gdb-efb763a5ea351f9d865cbe491909f03472ebf2d6.tar.bz2 |
gdb: check for partial symtab presence in dwarf2_initialize_objfile
This patch fixes an internal error that is triggered when loading the
same binary twice with the index-cache on:
$ ./gdb -q -nx --data-directory=data-directory
(gdb) set index-cache on
(gdb) shell mktemp -d
/tmp/tmp.BLgouVoPq4
(gdb) set index-cache directory /tmp/tmp.BLgouVoPq4
(gdb) file a.out
Reading symbols from a.out...
(gdb) file a.out
Load new symbol table from "a.out"? (y or n) y
Reading symbols from a.out...
/home/smarchi/src/binutils-gdb/gdb/dwarf2/read.c:2540: internal-error: void create_cus_from_index(dwarf2_per_bfd*, const gdb_byte*, offset_type, const gdb_byte*, offset_type): Assertion `per_bfd->all_comp_units.empty ()' failed.
A problem internal to GDB has been detected,
further debugging may prove unreliable.
Quit this debugging session? (y or n)
This is what happens:
1. We load the binary the first time, partial symtabs are created,
per_bfd->all_comp_units is filled from those.
2. Because index-cache is on, we also generate an index in the cache.
3. We load the binary a second time, in dwarf2_initialize_objfile we
check: was an index already loaded for this BFD? No, so we try to
read the index and fill the per-bfd using it. We do find an index,
it's in the cache.
4. The function create_cus_from_index asserts (rightfully) that
per_cu->all_comp_units is empty, and the assertion fails.
This assertion verifies that we are not reading an index for a BFD for
which we have already built partial symtabs or read another index.
The index-cache gives a situation that isn't currently accounted for: a
BFD for which we have built the partial symtabs the first time, but has
an index the second time.
This patch addresses it by checking for the presence of partial symtabs
in dwarf2_initialize_objfile. If there are, we don't try reading the
index.
gdb/ChangeLog:
* dwarf2/read.c (dwarf2_initialize_objfile): Check for presence
of partial symtabs.
gdb/testsuite/ChangeLog:
* gdb.base/index-cache-load-twice.c: New.
* gdb.base/index-cache-load-twice.exp: New.
Change-Id: Ie05474c44823fcdff852b73170dd28dfd66cb6a2
Diffstat (limited to 'gdb/dwarf2')
-rw-r--r-- | gdb/dwarf2/read.c | 8 |
1 files changed, 8 insertions, 0 deletions
diff --git a/gdb/dwarf2/read.c b/gdb/dwarf2/read.c index 34915be..4dc9ad6 100644 --- a/gdb/dwarf2/read.c +++ b/gdb/dwarf2/read.c @@ -6027,6 +6027,14 @@ dwarf2_initialize_objfile (struct objfile *objfile, dw_index_kind *index_kind) return true; } + /* There might already be partial symtabs built for this BFD. This happens + when loading the same binary twice with the index-cache enabled. If so, + don't try to read an index. The objfile / per_objfile initialization will + be completed in dwarf2_build_psymtabs, in the standard partial symtabs + code path. */ + if (per_bfd->partial_symtabs != nullptr) + return false; + if (dwarf2_read_debug_names (per_objfile)) { *index_kind = dw_index_kind::DEBUG_NAMES; |