]> gcc.gnu.org Git - gcc.git/commit
analyzer: fix escaping of pointer arithmetic [PR105264]
authorDavid Malcolm <dmalcolm@redhat.com>
Thu, 14 Apr 2022 13:52:00 +0000 (09:52 -0400)
committerDavid Malcolm <dmalcolm@redhat.com>
Thu, 14 Apr 2022 22:42:34 +0000 (18:42 -0400)
commita358e4b60815b41e27f3508014ceb592f86b9b45
treecd2325b0d8fb044595e2689284b3d83fc38b4d7b
parentaf27d545dc6132dcd67d1ee854372ea9cfd2a225
analyzer: fix escaping of pointer arithmetic [PR105264]

PR analyzer/105264 reports that the analyzer can fail to treat
(PTR + IDX) and PTR[IDX] as referring to the same memory under
some situations.

There are various ways in which this can happen when IDX is a
symbolic value, due to having several ways in which such memory
regions can be referred to symbolically.  I attempted to fix this by
being smarter when folding svalues and regions, but this fix
seems too fiddly to attempt in stage 4.

Instead, this less ambitious patch fixes a false positive from
-Wanalyzer-use-of-uninitialized-value by making the analyzer's escape
analysis smarter, so that it treats *PTR as escaping when
(PTR + OFFSET) is passed to an external function, and thus
it treats *PTR as possibly-initialized (the "passing &PTR[IDX]" case
was already working).

gcc/analyzer/ChangeLog:
PR analyzer/105264
* region-model-reachability.cc (reachable_regions::handle_parm):
Use maybe_get_deref_base_region rather than just region_svalue, to
handle pointer arithmetic also.
* svalue.cc (svalue::maybe_get_deref_base_region): New.
* svalue.h (svalue::maybe_get_deref_base_region): New decl.

gcc/testsuite/ChangeLog:
PR analyzer/105264
* gcc.dg/analyzer/torture/symbolic-10.c: New test.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>
gcc/analyzer/region-model-reachability.cc
gcc/analyzer/svalue.cc
gcc/analyzer/svalue.h
gcc/testsuite/gcc.dg/analyzer/torture/symbolic-10.c [new file with mode: 0644]
This page took 0.063319 seconds and 6 git commands to generate.