Skip to content

Commit

Permalink
i#5975: Add an invariant check for the number of memory read/write re…
Browse files Browse the repository at this point in the history 
…cords matching operands.

Add an invariant check to verify the number of memory read/write records
is matching the operands.

Mark x86 "rep" operands as predicated, and skip the check since the
number of records depends on run time register values.

Change check_sane_control_flow() to use gen_instr_encoded() when
encoding is expeced.

Fixes #5975
  • Loading branch information
@ivankyluk
ivankyluk committed Aug 28, 2023
1 parent 4b7245f commit cd10f6b
Show file tree
Hide file tree
Showing 9 changed files with 611 additions and 208 deletions.
9 changes: 9 additions & 0 deletions clients/drcachesim/common/trace_entry.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -634,6 +634,15 @@ type_is_data(const trace_type_t type)
type == TRACE_TYPE_DATA_FLUSH_END;
}

/**
* Returns whether the type represents a memory read access.
*/
static inline bool
type_is_read(const trace_type_t type)
{
return type_is_prefetch(type) || type == TRACE_TYPE_READ;
}

static inline bool
marker_type_is_function_marker(const trace_marker_type_t mark)
{
Expand Down
247 changes: 239 additions & 8 deletions clients/drcachesim/tests/invariant_checker_test.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -281,13 +281,15 @@ check_sane_control_flow()
gen_marker(TID, TRACE_MARKER_TYPE_FILETYPE, OFFLINE_FILE_TYPE_ENCODINGS),
# if defined(X86_64) || defined(X86_32)
// 0x74 is "je" with the 2nd byte the offset.
gen_branch_encoded(TID, 0x71019dbc, { 0x74, 0x32 }),
gen_instr_encoded(0x71019ded, { 0x01 }, TID),
gen_branch_encoded(TID, /*pc=*/0x71019dbc, /*encoding=*/ { 0x74, 0x32 }),
// 0x0x71019ded: 90 nop
gen_instr_encoded(/*pc=*/0x71019ded, /*encoding=*/ { 0x90 }, TID),
# elif defined(ARM_64)
// 71019dbc: 540001a1 b.ne 71019df0
// <__executable_start+0x19df0>
gen_branch_encoded(TID, 0x71019dbc, 0x540001a1),
gen_instr_encoded(0x71019ded, 0x01, TID),
gen_branch_encoded(TID, /*pc=*/0x71019dbc, /*encoding=*/0x540001a1),
// 71019ded: 1f2003d5 nop
gen_instr_encoded(/*pc=*/0x71019ded, /*encoding=*/0x1f2003d5, TID),
# else
// TODO i#5871: Add AArch32 (and RISC-V) encodings.
# endif
Expand All @@ -307,15 +309,18 @@ check_sane_control_flow()
gen_marker(TID, TRACE_MARKER_TYPE_FILETYPE, OFFLINE_FILE_TYPE_ENCODINGS),
# if defined(X86_64) || defined(X86_32)
// 0x74 is "je" with the 2nd byte the offset.
gen_branch_encoded(TID, 0x71019dbc, { 0x74, 0x32 }),
gen_branch_encoded(TID, /*pc=*/0x71019dbc, /*encoding=*/ { 0x74, 0x32 }),
// 0x71019df0: 90 nop
gen_instr_encoded(/*pc=*/0x71019df0, /*encoding=*/ { 0x90 }, TID),
# elif defined(ARM_64)
// 71019dbc: 540001a1 b.ne 71019df0
// <__executable_start+0x19df0>
gen_branch_encoded(TID, 0x71019dbc, 0x540001a1),
gen_branch_encoded(TID, /*pc=*/0x71019dbc, /*encoding=*/0x540001a1),
// 71019df0: 1f2003d5 nop
gen_instr_encoded(/*pc=*/0x71019df0, /*encoding=*/0x1f2003d5, TID),
# else
// TODO i#5871: Add AArch32 (and RISC-V) encodings.
# endif
gen_instr(TID, 0x71019df0),
};

if (!run_checker(memrefs, false)) {
Expand Down Expand Up @@ -1644,6 +1649,231 @@ check_timestamps_increase_monotonically(void)
return true;
}

bool
check_read_write_records_match_operands()
{
std::cerr << "Testing number of read/write records matching operands\n";
constexpr memref_tid_t TID = 1;
#if defined(X86_64) || defined(X86_32) || defined(ARM_64)
constexpr addr_t ADDR = 0x7fcf3b9d;
#endif

// Only the number of memory read and write records are checked against the
// operands. Address and size are not used.

// Correct: number of read records matches operand.
{
std::vector<memref_t> memrefs = {
gen_marker(TID, TRACE_MARKER_TYPE_FILETYPE, OFFLINE_FILE_TYPE_ENCODINGS),
#if defined(X86_64) || defined(X86_32)
// 0x7fcf3b9d: 83 3d 84 21 23 03 00 cmp <rel> 0x000000000443ddc0[4byte]
// $0x00000000
gen_instr_encoded(ADDR,
/*encoding=*/
{ 0x83, 0x3d, 0x84, 0x21, 0x23, 0x03, 0x00 }),
gen_data(TID, /*load=*/true, /*addr=*/0, /*size=*/0),
// 0x7fcf3ba4: 90 nop
gen_instr_encoded(ADDR + 7, { 0x90 }),
#elif defined(ARM_64)
// 0x7fcf3b9d: 40 c2 11 58 ldr x0, 0x1234
gen_instr_encoded(ADDR, /*encoding=*/0x40c21158),
gen_data(TID, /*load=*/true, /*addr=*/0, /*size=*/0),
// 0x7fcf3ba1: 1f 20 03 d5 nop
gen_instr_encoded(ADDR + 4, /*encoding=*/0x1f2003d5),
#else
// TODO: Add AArch32 (and RISC-V) encodings.
#endif
};
if (!run_checker(memrefs, false)) {
return false;
}
}
// Incorrect: too many read records.
{
std::vector<memref_t> memrefs = {
gen_marker(TID, TRACE_MARKER_TYPE_FILETYPE, OFFLINE_FILE_TYPE_ENCODINGS),
#if defined(X86_64) || defined(X86_32)
// 0x7fcf3b9d: 83 3d 84 21 23 03 00 cmp <rel> 0x000000000443ddc0[4byte]
// $0x00000000
gen_instr_encoded(ADDR,
/*encoding=*/
{ 0x83, 0x3d, 0x84, 0x21, 0x23, 0x03, 0x00 }),
gen_data(TID, /*load=*/true, /*addr=*/0, /*size=*/0),
gen_data(TID, /*load=*/true, /*addr=*/0, /*size=*/0),
// 0x7fcf3ba4: 90 nop
gen_instr_encoded(ADDR + 7, { 0x90 }),
#elif defined(ARM_64)
// 0x7fcf3b9d: 40 c2 11 58 ldr x0, 0x1234
gen_instr_encoded(ADDR, /*encoding=*/0x40c21158),
gen_data(TID, /*load=*/true, /*addr=*/0, /*size=*/0),
gen_data(TID, /*load=*/true, /*addr=*/0, /*size=*/0),
// 0x7fcf3ba1: 1f 20 03 d5 nop
gen_instr_encoded(ADDR + 4, /*encoding=*/0x1f2003d5),
#else
// TODO: Add AArch32 (and RISC-V) encodings.
#endif
};
if (!run_checker(memrefs, true,
{ "Too many read records", TID, /*ref_ordinal=*/4,
/*last_timestamp=*/0, /*instrs_since_last_timestamp=*/1 },
"Failed to catch too many read records"))
return false;
}
// Incorrect: missing read records.
{
std::vector<memref_t> memrefs = {
gen_marker(TID, TRACE_MARKER_TYPE_FILETYPE, OFFLINE_FILE_TYPE_ENCODINGS),
#if defined(X86_64) || defined(X86_32)
// 0x7fcf3b9d: 83 3d 84 21 23 03 00 cmp <rel> 0x000000000443ddc0[4byte]
// $0x00000000
gen_instr_encoded(ADDR,
/*encoding=*/
{ 0x83, 0x3d, 0x84, 0x21, 0x23, 0x03, 0x00 }),
// 0x7fcf3ba4: 90 nop
gen_instr_encoded(ADDR + 7, { 0x90 }),
#elif defined(ARM_64)
// 0x7fcf3b9d: 40 c2 11 58 ldr x0, 0x1234
gen_instr_encoded(ADDR, /*encoding=*/0x40c21158),
// 0x7fcf3ba1: 1f 20 03 d5 nop
gen_instr_encoded(ADDR + 4, /*encoding=*/0x1f2003d5),
#else
// TODO: Add AArch32 (and RISC-V) encodings.
#endif
};
if (!run_checker(memrefs, true,
{ "Missing read records", TID, /*ref_ordinal=*/3,
/*last_timestamp=*/0, /*instrs_since_last_timestamp=*/2 },
"Failed to catch missing read records"))
return false;
}
// Correct: number of write records match operand.
{
std::vector<memref_t> memrefs = {
gen_marker(TID, TRACE_MARKER_TYPE_FILETYPE, OFFLINE_FILE_TYPE_ENCODINGS),
#if defined(X86_64) || defined(X86_32)
// 0x7fcf3b9d: 89 47 70 mov %eax -> 0x70(%rdi)[4byte]
gen_instr_encoded(ADDR, /*encoding=*/ { 0x89, 0x47, 0x70 }),
gen_data(TID, /*load=*/false, /*addr=*/0, /*size=*/0),
// 0x7fcf3ba0: 90 nop
gen_instr_encoded(ADDR + 3, { 0x90 }),
#elif defined(ARM_64)
// 0x7fcf3b9d: 41 00 00 f9 str x1, [x2]
gen_instr_encoded(ADDR, /*encoding=*/0x410000f9),
gen_data(TID, /*load=*/false, /*addr=*/0, /*size=*/0),
// 0x7fcf3ba1: 1f 20 03 d5 nop
gen_instr_encoded(ADDR + 4, /*encoding=*/0x1f2003d5),
#else
// TODO: Add AArch32 (and RISC-V) encodings.
#endif
};
if (!run_checker(memrefs, false)) {
return false;
}
}
// Incorrect: too many write records.
{
std::vector<memref_t> memrefs = {
gen_marker(TID, TRACE_MARKER_TYPE_FILETYPE, OFFLINE_FILE_TYPE_ENCODINGS),
#if defined(X86_64) || defined(X86_32)
// 0x7fcf3b9d: 89 47 70 mov %eax -> 0x70(%rdi)[4byte]
gen_instr_encoded(ADDR, /*encoding=*/ { 0x89, 0x47, 0x70 }),
gen_data(TID, /*load=*/false, /*addr=*/0, /*size=*/0),
gen_data(TID, /*load=*/false, /*addr=*/0, /*size=*/0),
// 0x7fcf3ba0: 90 nop
gen_instr_encoded(ADDR + 3, { 0x90 }),
#elif defined(ARM_64)
// 0x7fcf3b9d: 41 00 00 f9 str x1, [x2]
gen_instr_encoded(ADDR, /*encoding=*/0x410000f9),
gen_data(TID, /*load=*/false, /*addr=*/0, /*size=*/0),
gen_data(TID, /*load=*/false, /*addr=*/0, /*size=*/0),
// 0x7fcf3ba1: 1f 20 03 d5 nop
gen_instr_encoded(ADDR + 4, /*encoding=*/0x1f2003d5),
#else
// TODO: Add AArch32 (and RISC-V) encodings.
#endif
};
if (!run_checker(memrefs, true,
{ "Too many write records", TID, /*ref_ordinal=*/4,
/*last_timestamp=*/0,
/*instrs_since_last_timestamp=*/1 },
"Failed to catch too many write records"))
return false;
}
// Incorrect: missing write records.
{
std::vector<memref_t> memrefs = {
gen_marker(TID, TRACE_MARKER_TYPE_FILETYPE, OFFLINE_FILE_TYPE_ENCODINGS),
#if defined(X86_64) || defined(X86_32)
// 0x7fcf3b9d: 89 47 70 mov %eax -> 0x70(%rdi)[4byte]
gen_instr_encoded(ADDR, /*encoding=*/ { 0x89, 0x47, 0x70 }),
// 0x7fcf3ba0: 90 nop
gen_instr_encoded(ADDR + 3, { 0x90 }),
#elif defined(ARM_64)
// 0x7fcf3b9d: 41 00 00 f9 str x1, [x2]
gen_instr_encoded(ADDR, /*encoding=*/0x410000f9),
// 0x7fcf3ba1: 1f 20 03 d5 nop
gen_instr_encoded(ADDR + 4, /*encoding=*/0x1f2003d5),
#else
// TODO: Add AArch32 (and RISC-V) encodings.
#endif
};
if (!run_checker(memrefs, true,
{ "Missing write records", TID, /*ref_ordinal=*/3,
/*last_timestamp=*/0,
/*instrs_since_last_timestamp=*/2 },
"Fail to catch missing write records"))
return false;
}
// Correct: ignore predicated operands which may not have memory access.
{
std::vector<memref_t> memrefs = {
gen_marker(TID, TRACE_MARKER_TYPE_FILETYPE, OFFLINE_FILE_TYPE_ENCODINGS),
#if defined(X86_64) || defined(X86_32)
// 0x7fcf3b9d: b9 00 00 00 00 mov $0x00000000, %ecx
gen_instr_encoded(ADDR,
/*encoding=*/
{ 0xb9, 0x00, 0x00, 0x00, 0x00 }),
// 0x7fcf3ba2: 8d 34 25 0e 20 40 00 lea 0x0040200e, %esi
gen_instr_encoded(ADDR + 5,
/*encoding=*/
{ 0x8d, 0x34, 0x25, 0x0e, 0x20, 0x40, 0x00 }),
// 0x7fcf3ba9: 8d 3c 25 00 20 40 00 lea 0x00402000, %edi
gen_instr_encoded(ADDR + 12,
/*encoding=*/
{ 0x8d, 0x3c, 0x25, 0x00, 0x20, 0x40, 0x00 }),
// 0x7fcf3bb0: fc cld
gen_instr_encoded(ADDR + 19, /*encoding=*/ { 0xfc }),
// 0x7fcf3bb1: f3 a5 rep movsd
gen_instr_encoded(ADDR + 20, /*encoding=*/ { 0xf3, 0xa5 }),
// 0x7fcf3bb3: 90 nop
gen_instr_encoded(ADDR + 22, { 0x90 }),
#endif
};
if (!run_checker(memrefs, false)) {
return false;
}
}
// Correct: ignore operands with opcode which do not have real memory
// access.
{
std::vector<memref_t> memrefs = {
gen_marker(TID, TRACE_MARKER_TYPE_FILETYPE, OFFLINE_FILE_TYPE_ENCODINGS),
#if defined(X86_64) || defined(X86_32)
// 0x7fcf3b9d: 4c 8d b7 00 01 00 00 lea 0x00000100(%rdi) -> %r14
gen_instr_encoded(ADDR,
/*encoding=*/
{ 0x4c, 0x8d, 0xb7, 0x00, 0x01, 0x00, 0x00 }),
// 0x7fcf3ba4: 90 nop
gen_instr_encoded(ADDR + 7, { 0x90 }),
#endif
};
if (!run_checker(memrefs, false)) {
return false;
}
}
return true;
}

int
test_main(int argc, const char *argv[])
{
Expand All @@ -1652,7 +1882,8 @@ test_main(int argc, const char *argv[])
check_duplicate_syscall_with_same_pc() && check_false_syscalls() &&
check_rseq_side_exit_discontinuity() && check_schedule_file() &&
check_branch_decoration() && check_filter_endpoint() &&
check_timestamps_increase_monotonically()) {
check_timestamps_increase_monotonically() &&
check_read_write_records_match_operands()) {
std::cerr << "invariant_checker_test passed\n";
return 0;
}
Expand Down
3 changes: 2 additions & 1 deletion clients/drcachesim/tests/memref_gen.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -121,7 +121,8 @@ gen_branch_encoded(memref_tid_t tid, addr_t pc, int encoding)

#elif defined(X86_64) || defined(X86_32)
inline memref_t
gen_instr_encoded(addr_t pc, const std::vector<char> &encoding, memref_tid_t tid = 1)
gen_instr_encoded(addr_t pc, const std::vector<unsigned char> &encoding,
memref_tid_t tid = 1)
{
memref_t memref = gen_instr_type(TRACE_TYPE_INSTR, tid, pc, encoding.size());
memcpy(memref.instr.encoding, encoding.data(), encoding.size());
Expand Down
Loading

0 comments on commit cd10f6b

Please sign in to comment.