/* We collect the basic block information including offset from module base,
* size, and num of instructions, and add it into a basic block table without
* instrumentation.
*/
static dr_emit_flags_t
event_basic_block_analysis(void *drcontext, void *tag, instrlist_t *bb,
bool for_trace, bool translating, OUT void **user_data)
{
per_thread_t *data;
instr_t *instr;
app_pc tag_pc, start_pc, end_pc;
/* do nothing for translation */
if (translating)
return DR_EMIT_DEFAULT;
data = (per_thread_t *)drmgr_get_tls_field(drcontext, tls_idx);
/* Collect the number of instructions and the basic block size,
* assuming the basic block does not have any elision on control
* transfer instructions, which is true for default options passed
* to DR but not for -opt_speed.
*/
/* We separate the tag from the instr pc ranges to handle displaced code
* such as for the vsyscall hook.
*/
tag_pc = dr_fragment_app_pc(tag);
start_pc = instr_get_app_pc(instrlist_first_app(bb));
end_pc = start_pc; /* for finding the size */
for (instr = instrlist_first_app(bb);
instr != NULL;
instr = instr_get_next_app(instr)) {
app_pc pc = instr_get_app_pc(instr);
int len = instr_length(drcontext, instr);
/* -opt_speed (elision) is not supported */
/* For rep str expansion pc may be one back from start pc but equal to the tag. */
ASSERT(pc != NULL && (pc >= start_pc || pc == tag_pc),
"-opt_speed is not supported");
if (pc + len > end_pc)
end_pc = pc + len;
}
/* We allow duplicated basic blocks for the following reasons:
* 1. Avoids handling issues like code cache consistency, e.g.,
* module load/unload, self-modifying code, etc.
* 2. Avoids the overhead on duplication check.
* 3. Stores more information on code cache events, e.g., trace building,
* repeated bb building, etc.
* 4. The duplication can be easily handled in a post-processing step,
* which is required anyway.
*/
bb_table_entry_add(drcontext, data, tag_pc, (uint)(end_pc - start_pc));
if (go_native)
return DR_EMIT_GO_NATIVE;
else
return DR_EMIT_DEFAULT;
}
static dr_emit_flags_t
event_bb_insert(void *drcontext, void *tag, instrlist_t *bb, instr_t *inst,
bool for_trace, bool translating, void *user_data)
{
static int freq;
reg_id_t reg1 = IF_X86_ELSE(DR_REG_XAX, DR_REG_R0);
reg_id_t reg2 = IF_X86_ELSE(DR_REG_XCX, DR_REG_R1);
CHECK(drmgr_is_first_instr(drcontext, instrlist_first_app(bb)), "first incorrect");
CHECK(!drmgr_is_first_instr(drcontext, instrlist_last(bb)) ||
instrlist_first_app(bb) == instrlist_last(bb), "first incorrect");
CHECK(drmgr_is_last_instr(drcontext, instrlist_last(bb)), "last incorrect");
CHECK(!drmgr_is_last_instr(drcontext, instrlist_first_app(bb)) ||
instrlist_first_app(bb) == instrlist_last(bb), "last incorrect");
/* hack to instrument every nth bb. assumes DR serializes bb events. */
freq++;
if (freq % 100 == 0 && inst == (instr_t*)user_data/*first instr*/) {
/* test read from cache */
dr_save_reg(drcontext, bb, inst, reg1, SPILL_SLOT_1);
drmgr_insert_read_tls_field(drcontext, tls_idx, bb, inst, reg1);
dr_insert_clean_call(drcontext, bb, inst, (void *)check_tls_from_cache,
false, 1, opnd_create_reg(reg1));
drmgr_insert_read_cls_field(drcontext, cls_idx, bb, inst, reg1);
dr_insert_clean_call(drcontext, bb, inst, (void *)check_cls_from_cache,
false, 1, opnd_create_reg(reg1));
dr_restore_reg(drcontext, bb, inst, reg1, SPILL_SLOT_1);
}
if (freq % 300 == 0 && inst == (instr_t*)user_data/*first instr*/) {
instr_t *first, *second;
/* test write from cache */
dr_save_reg(drcontext, bb, inst, reg1, SPILL_SLOT_1);
dr_save_reg(drcontext, bb, inst, reg2, SPILL_SLOT_2);
instrlist_insert_mov_immed_ptrsz(drcontext,
(ptr_int_t)MAGIC_NUMBER_FROM_CACHE,
opnd_create_reg(reg1),
bb, inst, &first, &second);
instr_set_meta(first);
if (second != NULL)
instr_set_meta(second);
drmgr_insert_write_tls_field(drcontext, tls_idx, bb, inst, reg1, reg2);
dr_insert_clean_call(drcontext, bb, inst, (void *)check_tls_write_from_cache,
false, 0);
drmgr_insert_write_cls_field(drcontext, cls_idx, bb, inst, reg1, reg2);
dr_insert_clean_call(drcontext, bb, inst, (void *)check_cls_write_from_cache,
false, 0);
dr_restore_reg(drcontext, bb, inst, reg2, SPILL_SLOT_2);
dr_restore_reg(drcontext, bb, inst, reg1, SPILL_SLOT_1);
}
return DR_EMIT_DEFAULT;
}
static dr_emit_flags_t
event_app_analysis(void *drcontext, void *tag, instrlist_t *bb,
bool for_trace, bool translating, OUT void **user_data)
{
instr_t *inst, *label;
bool prev_was_mov_const = false;
ptr_int_t val1, val2;
*user_data = NULL;
/* Look for duplicate mov immediates telling us which subtest we're in */
for (inst = instrlist_first_app(bb); inst != NULL; inst = instr_get_next_app(inst)) {
if (instr_is_mov_constant(inst, prev_was_mov_const ? &val2 : &val1)) {
if (prev_was_mov_const && val1 == val2 &&
val1 != 0 && /* rule out xor w/ self */
opnd_is_reg(instr_get_dst(inst, 0)) &&
opnd_get_reg(instr_get_dst(inst, 0)) == TEST_REG) {
*user_data = (void *) val1;
label = INSTR_CREATE_label(drcontext);
instr_set_translation(label, instr_get_app_pc(inst));
instrlist_meta_postinsert(bb, inst, label);
} else
prev_was_mov_const = true;
} else
prev_was_mov_const = false;
}
return DR_EMIT_DEFAULT;
}
static dr_emit_flags_t
event_basic_block(void *drcontext, void *tag, instrlist_t *bb,
bool for_trace, bool translating)
{
instr_t *instr, *next_instr;
#ifdef VERBOSE
dr_printf("in dr_basic_block(tag="PFX")\n", tag);
# if VERBOSE_VERBOSE
instrlist_disassemble(drcontext, tag, bb, STDOUT);
# endif
#endif
for (instr = instrlist_first_app(bb); instr != NULL; instr = next_instr) {
next_instr = instr_get_next_app(instr);
if (!instr_opcode_valid(instr))
continue;
/* instrument calls and returns -- ignore far calls/rets */
if (instr_is_call_direct(instr)) {
dr_insert_call_instrumentation(drcontext, bb, instr, (app_pc)at_call);
} else if (instr_is_call_indirect(instr)) {
dr_insert_mbr_instrumentation(drcontext, bb, instr, (app_pc)at_call_ind,
SPILL_SLOT_1);
} else if (instr_is_return(instr)) {
dr_insert_mbr_instrumentation(drcontext, bb, instr, (app_pc)at_return,
SPILL_SLOT_1);
}
}
return DR_EMIT_DEFAULT;
}
/* This event is passed the instruction list for the whole bb. */
static dr_emit_flags_t
event_analyze_bb(void *drcontext, void *tag, instrlist_t *bb, bool for_trace,
bool translating, void **user_data)
{
/* Count the instructions and pass the result to event_insert_instrumentation. */
instr_t *instr;
uint num_instrs;
for (instr = instrlist_first_app(bb), num_instrs = 0; instr != NULL;
instr = instr_get_next_app(instr)) {
num_instrs++;
}
*(uint *)user_data = num_instrs;
return DR_EMIT_DEFAULT;
}
static dr_emit_flags_t
event_analyze_bb(void *drcontext, void *tag, instrlist_t *bb,
bool for_trace, bool translating, void **user_data)
{
instr_t *instr;
trace_head_entry_t *e = NULL;
if (translating)
return DR_EMIT_DEFAULT;
for (instr = instrlist_first_app(bb);
instr != NULL;
instr = instr_get_next_app(instr)) {
/* Blocks containing calls are trace heads. */
if (instr_is_call(instr)) {
dr_mark_trace_head(drcontext, tag);
hashtable_lock(&head_table);
e = hashtable_lookup(&head_table, tag);
if (e == NULL) {
e = create_trace_head_entry(tag);
if (!hashtable_add(&head_table, tag, (void *)e))
DR_ASSERT(false);
} else
e->refcount++;
e->is_trace_head = true;
hashtable_unlock(&head_table);
#ifdef VERBOSE
dr_log(drcontext, DR_LOG_ALL, 3,
"inline: marking bb "PFX" as call trace head\n", tag);
#endif
/* Doesn't matter what's in rest of the bb. */
return DR_EMIT_DEFAULT;
} else if (instr_is_return(instr)) {
hashtable_lock(&head_table);
e = hashtable_lookup(&head_table, tag);
if (e == NULL) {
e = create_trace_head_entry(tag);
if (!hashtable_add(&head_table, tag, (void *)e))
DR_ASSERT(false);
} else
e->refcount++;
e->has_ret = true;
hashtable_unlock(&head_table);
#ifdef VERBOSE
dr_log(drcontext, DR_LOG_ALL, 3,
"inline: marking bb "PFX" as return trace head\n", tag);
#endif
}
}
return DR_EMIT_DEFAULT;
}
static dr_emit_flags_t
event_basic_block(void *drcontext, void *tag, instrlist_t *bb,
bool for_trace, bool translating)
{
instr_t *instr;
for (instr = instrlist_first_app(bb);
instr != NULL;
instr = instr_get_next_app(instr)) {
if (instr_is_cbr(instr)) {
dr_insert_cbr_instrumentation_ex
(drcontext, bb, instr, (void *)at_cbr,
OPND_CREATE_INTPTR(dr_fragment_app_pc(tag)));
}
}
return DR_EMIT_DEFAULT;
}
static dr_emit_flags_t
event_basic_block(void *drcontext, void *tag, instrlist_t *bb,
bool for_trace, bool translating)
{
instr_t *instr, *mbr = NULL;
uint num_instrs;
bool bb_in_app;
#ifdef VERBOSE
dr_printf("in dynamorio_basic_block(tag="PFX")\n", tag);
# ifdef VERBOSE_VERBOSE
instrlist_disassemble(drcontext, tag, bb, STDOUT);
# endif
#endif
for (instr = instrlist_first_app(bb), num_instrs = 0;
instr != NULL;
instr = instr_get_next_app(instr)) {
num_instrs++;
/* Assuming most of the transfers between app and lib are paired, we
* instrument indirect branches but not returns for better performance.
*/
if (instr_is_mbr(instr) && !instr_is_return(instr))
mbr = instr;
}
if (dr_fragment_app_pc(tag) >= app_base &&
dr_fragment_app_pc(tag) < app_end)
bb_in_app = true;
else
bb_in_app = false;
dr_insert_clean_call(drcontext, bb, instrlist_first(bb),
(void *)(bb_in_app ? app_update : lib_update),
false /* save fpstate */, 1,
OPND_CREATE_INT32(num_instrs));
if (mbr != NULL) {
dr_insert_mbr_instrumentation(drcontext, bb, mbr,
(void *)(bb_in_app ? app_mbr : lib_mbr),
SPILL_SLOT_1);
}
#if defined(VERBOSE) && defined(VERBOSE_VERBOSE)
dr_printf("Finished instrumenting dynamorio_basic_block(tag="PFX")\n", tag);
instrlist_disassemble(drcontext, tag, bb, STDOUT);
#endif
return DR_EMIT_DEFAULT;
}
static dr_emit_flags_t
event_app_instruction(void *drcontext, void *tag, instrlist_t *bb, instr_t *instr,
bool for_trace, bool translating, void *user_data)
{
uint num_instrs;
/* By default drmgr enables auto-predication, which predicates all instructions with
* the predicate of the current instruction on ARM.
* We disable it here because we want to unconditionally execute the following
* instrumentation.
*/
drmgr_disable_auto_predication(drcontext, bb);
if (!drmgr_is_first_instr(drcontext, instr))
return DR_EMIT_DEFAULT;
/* Only insert calls for in-app BBs */
if (user_data == NULL)
return DR_EMIT_DEFAULT;
/* Insert clean call */
num_instrs = (uint)(ptr_uint_t)user_data;
dr_insert_clean_call(drcontext, bb, instrlist_first_app(bb), (void *)inscount,
false /* save fpstate */, 1, OPND_CREATE_INT32(num_instrs));
return DR_EMIT_DEFAULT;
}
void
offline_instru_t::bb_analysis(void *drcontext, void *tag, void **bb_field,
instrlist_t *ilist, bool repstr_expanded)
{
instr_t *instr;
ptr_uint_t count = 0;
app_pc last_xl8 = NULL;
if (repstr_expanded) {
// The same-translation check below is not sufficient as drutil uses
// two different translations to deal with complexities.
// Thus we hardcode this.
count = 1;
} else {
for (instr = instrlist_first_app(ilist); instr != NULL;
instr = instr_get_next_app(instr)) {
// To deal with app2app changes, we do not double-count consecutive instrs
// with the same translation.
if (instr_get_app_pc(instr) != last_xl8)
++count;
last_xl8 = instr_get_app_pc(instr);
}
}
*bb_field = (void *)count;
}
/* This event is passed the instruction list for the whole bb. */
static dr_emit_flags_t
event_analyze_bb(void *drcontext, void *tag, instrlist_t *bb,
bool for_trace, bool translating, void **user_data)
{
/* Count the instructions and pass the result to event_insert_instrumentation. */
per_bb_data_t *per_bb = dr_thread_alloc(drcontext, sizeof(*per_bb));
instr_t *instr;
uint num_instrs = 0;
uint num_flops = 0;
uint num_syscalls = 0;
dr_fp_type_t fp_type;
for (instr = instrlist_first_app(bb);
instr != NULL;
instr = instr_get_next_app(instr)) {
num_instrs++;
if (instr_is_floating_ex(instr, &fp_type) &&
/* We exclude loads and stores (and reg-reg moves) and state preservation */
(fp_type == DR_FP_CONVERT || fp_type == DR_FP_MATH)) {
#ifdef VERBOSE
dr_print_instr(drcontext, STDOUT, instr, "Found flop: ");
#endif
num_flops++;
}
if (instr_is_syscall(instr)) {
num_syscalls++;
}
}
per_bb->num_instrs = num_instrs;
per_bb->num_flops = num_flops;
per_bb->num_syscalls = num_syscalls;
*(per_bb_data_t**)user_data = per_bb;
return DR_EMIT_DEFAULT;
}
static dr_emit_flags_t
event_basic_block(void *drcontext, void *tag, instrlist_t *bb,
bool for_trace, bool translating)
{
instr_t *instr, *next_instr;
#ifdef VERBOSE
dr_printf("in dynamorio_basic_block(tag="PFX")\n", tag);
# ifdef VERBOSE_VERBOSE
instrlist_disassemble(drcontext, tag, bb, STDOUT);
# endif
#endif
for (instr = instrlist_first_app(bb); instr != NULL; instr = next_instr) {
/* grab next now so we don't go over instructions we insert */
next_instr = instr_get_next_app(instr);
/* instrument calls and returns -- ignore far calls/rets */
if (instr_is_call_direct(instr)) {
insert_counter_update(drcontext, bb, instr,
offsetof(per_thread_t, num_direct_calls));
} else if (instr_is_call_indirect(instr)) {
insert_counter_update(drcontext, bb, instr,
offsetof(per_thread_t, num_indirect_calls));
} else if (instr_is_return(instr)) {
insert_counter_update(drcontext, bb, instr,
offsetof(per_thread_t, num_returns));
}
}
#if defined(VERBOSE) && defined(VERBOSE_VERBOSE)
dr_printf("Finished instrumenting dynamorio_basic_block(tag="PFX")\n", tag);
instrlist_disassemble(drcontext, tag, bb, STDOUT);
#endif
return DR_EMIT_DEFAULT;
}