On Mon, Feb 10, 2020 at 12:01 PM Uros Bizjak <ubiz...@gmail.com> wrote: > > On Mon, Feb 10, 2020 at 8:53 PM H.J. Lu <hjl.to...@gmail.com> wrote: > > > > On Mon, Feb 10, 2020 at 11:40 AM Uros Bizjak <ubiz...@gmail.com> wrote: > > > > > > On Mon, Feb 10, 2020 at 8:22 PM H.J. Lu <hjl.to...@gmail.com> wrote: > > > > > > > > Since nested function isn't only called directly, there is ENDBR32 at > > > > function entry and we need to skip it for direct jump in trampoline. > > > > > > Hm, I'm afraid I don't understand this comment. Can you perhaps rephrase > > > it? > > > > > > > ix86_trampoline_init has > > > > /* Compute offset from the end of the jmp to the target function. > > In the case in which the trampoline stores the static chain on > > the stack, we need to skip the first insn which pushes the > > (call-saved) register static chain; this push is 1 byte. */ > > offset += 5; > > disp = expand_binop (SImode, sub_optab, fnaddr, > > plus_constant (Pmode, XEXP (m_tramp, 0), > > offset - (MEM_P (chain) ? 1 : 0)), > > NULL_RTX, 1, OPTAB_DIRECT); > > emit_move_insn (mem, disp); > > > > Without CET, we got > > > > 0000011 <bar.1878>: > > 11: 56 push %esi > > 12: 55 push %ebp <<<<<< trampoline jumps here. > > 13: 89 e5 mov %esp,%ebp > > 15: 83 ec 08 sub $0x8,%esp > > > > With CET, if bar isn't only called directly, we got > > > > 00000015 <bar.1878>: > > 15: f3 0f 1e fb endbr32 > > 19: 56 push %esi > > 1a: 55 push %ebp <<<<<<<< trampoline jumps here. > > 1b: 89 e5 mov %esp,%ebp > > 1d: 83 ec 08 sub $0x8,%esp > > > > We need to add 4 bytes for trampoline to skip endbr32. > > > > Here is the updated patch to check if nested function isn't only > > called directly, > > Please figure out the final patch. I don't want to waste my time > reviewing different version every half hour. Ping me in a couple of > days.
This is the final version: https://gcc.gnu.org/ml/gcc-patches/2020-02/msg00586.html You can try the testcase in the patch on any machine with CET binutils since ENDBR32 is nop on none-CET machines. Without this patch, the test will fail. Thanks. -- H.J.
