Re: [PATCH v4] tcg: Toggle page execution for Apple Silicon

[Roman Bolshakov <r.bolshakov@yadro.com>]

On Sat, Jan 23, 2021 at 07:33:37PM +0100, BALATON Zoltan wrote:
> On Sat, 23 Jan 2021, Roman Bolshakov wrote:
> > On Sat, Jan 23, 2021 at 02:53:49PM +0300, Roman Bolshakov wrote:
> > > On Thu, Jan 21, 2021 at 08:47:52AM -1000, Richard Henderson wrote:
> > > > From: Roman Bolshakov <r.bolshakov@yadro.com>
> > > > 
> > > > Pages can't be both write and executable at the same time on Apple
> > > > Silicon. macOS provides public API to switch write protection [1] for
> > > > JIT applications, like TCG.
> > > > 
> > > > 1. https://developer.apple.com/documentation/apple_silicon/porting_just-in-time_compilers_to_apple_silicon
> > > > 
> > > > Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
> > > > Message-Id: <20210113032806.18220-1-r.bolshakov@yadro.com>
> > > > [rth: Inline the qemu_thread_jit_* functions;
> > > >  drop the MAP_JIT change for a follow-on patch.]
> > > > Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> > > > ---
> > > > 
> > > > Supercedes: <20210113032806.18220-1-r.bolshakov@yadro.com>
> > > > 
> > > > This is the version of Roman's patch that I'm queuing to tcg-next.
> > > > What's missing from the full "Fix execution" patch is setting MAP_JIT
> > > > for !splitwx in alloc_code_gen_buffer().
> > > > 
> > > 
> > > Richard, thanks for updating the patch. I have no objections against
> > > moving the functions and inlining them. However I'm seeing an issue that
> > > wasn't present in v3:
> > > 
> > > Process 37109 stopped                                                                                                                          * thread #6, stop reason = EXC_BAD_ACCESS (code=1, address=0xfffffffffffffd4f)
> > >     frame #0: 0x00000001002f1c90 qemu-system-x86_64`tcg_emit_op(opc=INDEX_op_add_i64) at tcg.c:2531:5 [opt]                                       2528 TCGOp *tcg_emit_op(TCGOpcode opc)
> > >    2529 {                                                                                                                                         2530     TCGOp *op = tcg_op_alloc(opc);
> > > -> 2531     QTAILQ_INSERT_TAIL(&tcg_ctx->ops, op, link);
> > >    2532     return op;
> > >    2533 }
> > >    2534
> > > Target 0: (qemu-system-x86_64) stopped.
> > > (lldb) bt
> > > * thread #6, stop reason = EXC_BAD_ACCESS (code=1, address=0xfffffffffffffd4f)
> > >   * frame #0: 0x00000001002f1c90 qemu-system-x86_64`tcg_emit_op(opc=INDEX_op_add_i64) at tcg.c:2531:5 [opt]
> > >     frame #1: 0x000000010026f040 qemu-system-x86_64`tcg_gen_addi_i64 [inlined] tcg_gen_op3(opc=INDEX_op_add_i64, a1=4430334952, a2=4430333440,
> > > a3=4430361496) at tcg-op.c:60:17 [opt]
> > >     frame #2: 0x000000010026f038 qemu-system-x86_64`tcg_gen_addi_i64 [inlined] tcg_gen_op3_i64(opc=INDEX_op_add_i64, a1=<unavailable>, a2=<unav
> > > ailable>, a3=<unavailable>) at tcg-op.h:94 [opt]
> > >     frame #3: 0x000000010026f030 qemu-system-x86_64`tcg_gen_addi_i64 [inlined] tcg_gen_add_i64(ret=<unavailable>, arg1=<unavailable>, arg2=<una
> > > vailable>) at tcg-op.h:618 [opt]
> > >     frame #4: 0x000000010026f030 qemu-system-x86_64`tcg_gen_addi_i64(ret=<unavailable>, arg1=<unavailable>, arg2=<unavailable>) at tcg-op.c:123
> > > 5 [opt]
> > >     frame #5: 0x000000010021d1e0 qemu-system-x86_64`gen_lea_modrm_1(s=<unavailable>, a=(def_seg = 2, base = 5, index = -1, scale = 0, disp = -6
> > > 89)) at translate.c:2101:9 [opt]
> > >     frame #6: 0x000000010020eeec qemu-system-x86_64`disas_insn [inlined] gen_lea_modrm(env=0x0000000118610870, s=0x00000001700b6b00, modrm=<una
> > > vailable>) at translate.c:2111:15 [opt]
> > >     frame #7: 0x000000010020eec0 qemu-system-x86_64`disas_insn(s=0x00000001700b6b00, cpu=<unavailable>) at translate.c:5509 [opt]
> > >     frame #8: 0x000000010020bb44 qemu-system-x86_64`i386_tr_translate_insn(dcbase=0x00000001700b6b00, cpu=<unavailable>) at translate.c:8573:15
> > >  [opt]
> > >     frame #9: 0x00000001002fbcf8 qemu-system-x86_64`translator_loop(ops=0x0000000100b209c8, db=0x00000001700b6b00, cpu=0x0000000118608000, tb=0
> > > x0000000120017200, max_insns=512) at translator.c:0 [opt]
> > >     frame #10: 0x000000010020b73c qemu-system-x86_64`gen_intermediate_code(cpu=<unavailable>, tb=<unavailable>, max_insns=<unavailable>) at tra
> > > nslate.c:8635:5 [opt]
> > >     frame #11: 0x0000000100257970 qemu-system-x86_64`tb_gen_code(cpu=0x0000000118608000, pc=<unavailable>, cs_base=0, flags=4194483, cflags=-16
> > > 777216) at translate-all.c:1931:5 [opt]
> > >     frame #12: 0x00000001002deb90 qemu-system-x86_64`cpu_exec [inlined] tb_find(cpu=0x0000000118608000, last_tb=0x0000000000000000, tb_exit=<un
> > > available>, cf_mask=0) at cpu-exec.c:456:14 [opt]
> > >     frame #13: 0x00000001002deb54 qemu-system-x86_64`cpu_exec(cpu=0x0000000118608000) at cpu-exec.c:812 [opt]
> > >     frame #14: 0x00000001002bc0d0 qemu-system-x86_64`tcg_cpus_exec(cpu=0x0000000118608000) at tcg-cpus.c:57:11 [opt]
> > >     frame #15: 0x000000010024c2cc qemu-system-x86_64`rr_cpu_thread_fn(arg=<unavailable>) at tcg-cpus-rr.c:217:21 [opt]
> > >     frame #16: 0x00000001004b00b4 qemu-system-x86_64`qemu_thread_start(args=<unavailable>) at qemu-thread-posix.c:521:9 [opt]
> > >     frame #17: 0x0000000191c4d06c libsystem_pthread.dylib`_pthread_start + 320
> > > 
> > > I'm looking into the issue but perhaps we'll need v5.
> > > 
> > 
> > Nope. The issue is not directly related to the patch and W^X. I think it
> > can be applied.
> > 
> > tcg_ctx->ops is somehow getting corrupted despite it's initialized
> > properly during TCG start:
> > 
> > (lldb) p tcg_ctx->ops
> > (TCGContext::(anonymous union)) $18 = {
> >  tqh_first = 0x0000008401010000
> >  tqh_circ = {
> >    tql_next = 0x0000008401010000
> >    tql_prev = 0xfffffffffffffd4f
> >  }
> > }
> > 
> > I've bisected it (with v3 of Apple Silicon fix for TCG) to:
> > 
> > commit 8fe35e0444be88de4e3ab80a2a0e210a1f6d663d
> > Author: Richard Henderson <richard.henderson@linaro.org>
> > Date:   Mon Mar 30 20:42:43 2020 -0700
> > 
> >    tcg/optimize: Use tcg_constant_internal with constant folding
> 
> Yes, known problem, see https://bugs.launchpad.net/qemu/+bug/1912065
> 

Thanks for providing the link, Zoltan.

-Roman

> Regards,
> BALATON Zoltan
> 
> >    Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> > 
> > tcg/optimize.c | 108 ++++++++++++++++++++++++++-------------------------------
> > 1 file changed, 49 insertions(+), 59 deletions(-)
> > 
> > Although, it crashes with a bit different backtrace than provided above:
> > 
> > Process 17251 stopped
> > * thread #6, stop reason = EXC_BAD_ACCESS (code=1, address=0x10)
> >    frame #0: 0x00000001002727a8 qemu-system-x86_64`tcg_opt_gen_mov [inlined] ts_is_copy(ts=<unavailable>) at optimize.c:68:25 [opt]
> >   65
> >   66   static inline bool ts_is_copy(TCGTemp *ts)
> >   67   {
> > -> 68       return ts_info(ts)->next_copy != ts;
> >   69   }
> >   70
> >   71   /* Reset TEMP's state, possibly removing the temp for the list of copies.  */
> > Target 0: (qemu-system-x86_64) stopped.
> > (lldb) bt
> > * thread #6, stop reason = EXC_BAD_ACCESS (code=1, address=0x10)
> >  * frame #0: 0x00000001002727a8 qemu-system-x86_64`tcg_opt_gen_mov [inlined] ts_is_copy(ts=<unavailable>) at optimize.c:68:25 [opt]
> >    frame #1: 0x00000001002727a4 qemu-system-x86_64`tcg_opt_gen_mov [inlined] ts_are_copies(ts1=0x00000001202e8e50, ts2=0x00000001202ef398) at optimize.c:163 [opt]
> >    frame #2: 0x000000010027278c qemu-system-x86_64`tcg_opt_gen_mov(s=0x00000001202e8000, op=0x0000000119157710, dst=4834889296, src=4834915224) at optimize.c:191 [opt]
> >    frame #3: 0x0000000100271740 qemu-system-x86_64`tcg_optimize(s=<unavailable>) at optimize.c:0:9 [opt]
> >    frame #4: 0x00000001002f39c4 qemu-system-x86_64`tcg_gen_code(s=0x00000001202e8000, tb=0x0000000128020800) at tcg.c:4407:5 [opt]
> >    frame #5: 0x00000001002b2688 qemu-system-x86_64`tb_gen_code(cpu=0x0000000118428000, pc=<unavailable>, cs_base=0, flags=4194483, cflags=-16777216) at translate-all.c:1961:21 [opt]
> >    frame #6: 0x0000000100279460 qemu-system-x86_64`cpu_exec [inlined] tb_find(cpu=0x0000000118428000, last_tb=0x0000000000000000, tb_exit=<unavailable>, cf_mask=0) at cpu-exec.c:456:14 [opt]
> >    frame #7: 0x0000000100279424 qemu-system-x86_64`cpu_exec(cpu=0x0000000118428000) at cpu-exec.c:812 [opt]
> >    frame #8: 0x000000010026ea74 qemu-system-x86_64`tcg_cpus_exec(cpu=0x0000000118428000) at tcg-cpus.c:57:11 [opt]
> >    frame #9: 0x0000000100284efc qemu-system-x86_64`rr_cpu_thread_fn(arg=<unavailable>) at tcg-cpus-rr.c:217:21 [opt]
> >    frame #10: 0x00000001004cffe8 qemu-system-x86_64`qemu_thread_start(args=<unavailable>) at qemu-thread-posix.c:521:9 [opt]
> >    frame #11: 0x0000000191c4d06c libsystem_pthread.dylib`_pthread_start + 320
> > 
> > The problem is that ts2 in ts_are_copies() contains NULL state_ptr:
> > 
> > (lldb) p *ts2
> > (TCGTemp) $2 = {
> >  reg = TCG_REG_X0
> >  val_type = TEMP_VAL_DEAD
> >  base_type = TCG_TYPE_I64
> >  type = TCG_TYPE_I64
> >  kind = TEMP_CONST
> >  indirect_reg = 0
> >  indirect_base = 0
> >  mem_coherent = 0
> >  mem_allocated = 0
> >  temp_allocated = 1
> >  val = -690
> >  mem_base = 0x0000000000000000
> >  mem_offset = 0
> >  name = 0x0000000000000000
> >  state = 0
> >  state_ptr = 0x0000000000000000
> > }
> > 
> > -Roman
> > 
> >