On 10/1/22 00:03, Paolo Bonzini wrote:
On Sat, Oct 1, 2022 at 3:04 AM Richard Henderson
<richard.hender...@linaro.org> wrote:
On 9/21/22 06:09, Paolo Bonzini wrote:
On Tue, Sep 6, 2022 at 12:09 PM Richard Henderson
<richard.hender...@linaro.org> wrote:
+ gen_jcc1(s, b, l1);
+ gen_jmp_rel(s, ot, 0, 1);
+ gen_set_label(l1);
+ gen_jmp_rel(s, ot, diff, 0);
Might be worth a comment that jumps with 16-bit operand size truncate
EIP even if the jump is not taken.
Hmm. But is that correct? That's not reflected by the pseudocode for Jcc.
No, it's not:
int main() {
asm("clc; data16 jc 1f; 1:");
}
does not crash (it does with stc) on real hardware, but it does with
this series applied. So the various occurrences of gen_jmp_rel(s, ot,
0, 1) or gen_jmp_rel(s, MO_32, 0, 1) should stay as gen_jmp_tb(s,
s->pc - s->cs_base, 1).
Nice test. I had an idea this would be the case, so I had already added a helper to
perform the jump with truncation to the "current code size". It turned out that I needed
that in other places too, like rep.
New patch set coming up.
r~
