warn when trying to encode an instrlist with labels without a second pass

Created by: toshipiazza

When creating a code cache with dr_nonheap_alloc that has a jump, I find that all jumps to labels are replaced with an infinite loop because the jump's operand points to itself.

void
insert_jmp(void *drcontext, instrlist_t *ilist, instr_t *where)
{
    instr_t *label = INSTR_CREATE_label(drcontext);
    instrlist_meta_preinsert(ilist, where, XINST_CREATE_jump
                             (drcontext,
                              opnd_create_instr(label)));
    instrlist_meta_preinsert(ilist, where, label);
}

static void *
code_cache_init(void *drcontext)
{
    void *code_cache = dr_nonheap_alloc(dr_page_size(), DR_MEMPROT_READ  |
                                                        DR_MEMPROT_WRITE |
                                                        DR_MEMPROT_EXEC);
    instrlist_t *ilist = instrlist_create(drcontext);
    instr_t *where = INSTR_CREATE_ret(drcontext);
    instrlist_meta_append(ilist, where);

    /* jumping to this label causes an infinite loop */
    insert_jmp(drcontext, ilist, where);

    instrlist_encode(drcontext, ilist, code_cache, false);
    instrlist_clear_and_destroy(drcontext, ilist);
    return code_cache;
}

static dr_emit_flags_t
event_basic_block(void *drcontext, void *tag, instrlist_t *bb,
                  bool for_trace, bool translating)
{
    instr_t *where = instrlist_first(bb);
    /* jumping to this label works... */
    insert_jmp(drcontext, bb, where);
    dr_insert_clean_call(drcontext, bb, where, code_cache_init(drcontext), false, 0);
    return DR_EMIT_DEFAULT;
}

This is similar to what we do in the memtrace_x86 sample, however we use an absolute jump as opposed to a clean call in that sample.

I would like to use a clean call here, but if this behavior is intended I will instead use dr_prepare_for_call followed by the absolute jump.