Hello, It appeared that in my previous patch, a stupid thinko can lead to a crash when instrumenting some builtin functionsK. Fixed thus.
Tested against trunk on x86_64-unknown-linux-gnu, and a bootstrap is underway. gcc/ * asan.c (instrument_builtin_call): Really put the length of the second source argument into src1_len. gcc/testsuite/ * c-c++-common/asan/memcmp-2.c: New test. --- gcc/asan.c | 2 +- gcc/testsuite/c-c++-common/asan/memcmp-2.c | 10 ++++++++++ 2 files changed, 11 insertions(+), 1 deletion(-) create mode 100644 gcc/testsuite/c-c++-common/asan/memcmp-2.c diff --git a/gcc/asan.c b/gcc/asan.c index 3cb9511..7783723 100644 --- a/gcc/asan.c +++ b/gcc/asan.c @@ -1764,7 +1764,7 @@ instrument_builtin_call (gimple_stmt_iterator *iter) if (get_mem_refs_of_builtin_call (call, &src0, &src0_len, &src0_is_store, - &src1, &src0_len, &src1_is_store, + &src1, &src1_len, &src1_is_store, &dest, &dest_len, &dest_is_store, &dest_is_deref)) { diff --git a/gcc/testsuite/c-c++-common/asan/memcmp-2.c b/gcc/testsuite/c-c++-common/asan/memcmp-2.c new file mode 100644 index 0000000..be20cd8 --- /dev/null +++ b/gcc/testsuite/c-c++-common/asan/memcmp-2.c @@ -0,0 +1,10 @@ +/* { dg-do compile } */ + +int +foo () +{ + char s0[5] = {0}; + char s1[5] = {1}; + + return __builtin_memcmp (s0, s1, 2); +} -- Dodji