Tarantool development patches archive
 help / color / mirror / Atom feed
From: Sergey Bronnikov via Tarantool-patches <tarantool-patches@dev.tarantool.org>
To: Sergey Kaplun <skaplun@tarantool.org>,
	Maxim Kokryashkin <m.kokryashkin@tarantool.org>
Cc: tarantool-patches@dev.tarantool.org
Subject: Re: [Tarantool-patches] [PATCH luajit] Prevent down-recursion for side traces.
Date: Tue, 26 Mar 2024 18:08:44 +0300	[thread overview]
Message-ID: <b27cb112-cef8-4f78-b516-2238b2d98da0@tarantool.org> (raw)
In-Reply-To: <20240319164148.22506-1-skaplun@tarantool.org>

Sergey,

thanks for the patch. LGTM with three minor comments below.

Sergey

On 3/19/24 19:41, Sergey Kaplun wrote:
> From: Mike Pall <mike>
>
> Thanks to Sergey Kaplun.
>
> (cherry picked from commit cae361187e7e1e3545353fb560c032cdace32d5f)
>
> Assume we have the root trace that uses some spill slots with the
> corresponding stack adjustment. Then its side trace will restore stack
s/stack/the stack/
> only at its tail. It may look like the following:
>
> | ---- TRACE 4 mcode 1247
> | 55557f7df953  mov rax, [r14-0xe28]
> | 55557f7df95a  mov rax, [rax+0x30]
> | 55557f7df95e  sub rax, rdx
> | 55557f7df961  cmp rax, +0x68
> | 55557f7df965  jb 0x55557f7d004c       ->0
> | 55557f7df96b  add rsp, -0x10
> | ...
> | 55557f6efa71  cmp dword [rdx+0x4], -0x05
> | 55557f6efa75  jnz 0x55557f6e004c      ->0
> | ...
> | 55557f7dfe29  add rsp, +0x10
> | 55557f7dfe2d  jmp 0x5555556fe573
> | ---- TRACE 4 stop -> stitch
> |
> | ---- TRACE 5 start 4/0
> | ---- TRACE 5 mcode 101
> | 55557f6ef9d4  mov dword [0x40000518], 0x5
> | ...
> | 55557f6efa30  add rsp, +0x10
> | 55557f6efa34  jmp 0x55557f6ef9d4
> | ---- TRACE 5 stop -> down-recursion
>
> Such side traces have no stack addjustment at their heads since their
s/addjustment/adjustment/
> stack addjustment is inherited from the parent trace. The issue occurs
> if the side trace has a down-recursion, as mentioned above. Before any
> exit, we can jump back to the start of the trace several times with
> growing `rsp`. In that case, the `rsp` is restored incorrectly after
> exiting from the trace.
>
> This patch forbids down-recursion for non-root traces.
>
> Sergey Kaplun:
> * added the description and the test for the problem
>
> Part of tarantool/tarantool#9595
> ---
>
> Branch: https://github.com/tarantool/luajit/tree/skaplun/lj-1169-down-rec-side
> Related issues:
> * https://github.com/tarantool/tarantool/issues/9595
> * https://github.com/LuaJIT/LuaJIT/issues/1169
>
>   src/lj_record.c                               |  2 +-
>   .../lj-1169-down-rec-side.test.lua            | 89 +++++++++++++++++++
>   2 files changed, 90 insertions(+), 1 deletion(-)
>   create mode 100644 test/tarantool-tests/lj-1169-down-rec-side.test.lua
>
> diff --git a/src/lj_record.c b/src/lj_record.c
> index a6c48747..7fa56834 100644
> --- a/src/lj_record.c
> +++ b/src/lj_record.c
> @@ -867,7 +867,7 @@ void lj_record_ret(jit_State *J, BCReg rbase, ptrdiff_t gotresults)
>       if ((pt->flags & PROTO_NOJIT))
>         lj_trace_err(J, LJ_TRERR_CJITOFF);
>       if (J->framedepth == 0 && J->pt && frame == J->L->base - 1) {
> -      if (check_downrec_unroll(J, pt)) {
> +      if (!J->cur.root && check_downrec_unroll(J, pt)) {
>   	J->maxslot = (BCReg)(rbase + gotresults);
>   	lj_snap_purge(J);
>   	lj_record_stop(J, LJ_TRLINK_DOWNREC, J->cur.traceno);  /* Down-rec. */
> diff --git a/test/tarantool-tests/lj-1169-down-rec-side.test.lua b/test/tarantool-tests/lj-1169-down-rec-side.test.lua
> new file mode 100644
> index 00000000..63f9925f
> --- /dev/null
> +++ b/test/tarantool-tests/lj-1169-down-rec-side.test.lua
> @@ -0,0 +1,89 @@
> +local tap = require('tap')
> +
> +-- Test file to demonstrate the LuaJIT misbehaviour when recording
> +-- and executing a side trace with down-recursion.
> +-- See also: https://github.com/LuaJIT/LuaJIT/issues/1169.
> +
> +local test = tap.test('lj-1169-down-rec-side'):skipcond({
> +  ['Test requires JIT enabled'] = not jit.status(),
> +})
> +
> +local ffi = require('ffi')
> +
> +-- XXX: simplify `jit.dump()` output.
> +local modf = math.modf
> +local ffi_new = ffi.new
> +
> +local int64_t = ffi.typeof('int64_t')
> +
> +test:plan(1)
> +
> +-- If a parent trace has more than the default amount of spill
> +-- slots, the `rsp` register is adjusted at the start of the trace
> +-- and restored after. If there is a side trace created, it
> +-- modifies the stack only at exit (since adjustment is inherited
> +-- from a parent trace). If the side trace has down-recursion (for
s/own-recursion/a down-recursion/
> +-- now only down-recursion to itself is used), `rsp` may be
> +-- modified several times before exit, so the host stack becomes
> +-- corrupted.
> +--
> +-- This test provides the example of a side trace (5) with
> +-- down-recursion.
> +
> +local function trace_ret(arg) -- TRACE 1 start.
> +  return arg -- TRACE 4 start 1/0; TRACE 5 start 4/0.
> +end
> +
> +local function extra_frame()
> +  -- Stitch the trace (4) to prevent early down-recursion.
> +  modf(1)
> +  -- Start the side trace (5) with a down-recursion.
> +  return trace_ret({})
> +end
> +
> +local call_counter = 0
> +local function recursive_f() -- TRACE 2 start.
> +  -- XXX: 4 calls are needed to record the necessary traces after
> +  -- return. With the 5th call, the traces are executed.
> +  if call_counter > 4 then return end -- TRACE 3 start 2/1.
> +  call_counter = call_counter + 1
> +
> +  -- Compile the root trace first.
> +  trace_ret(1)
> +  trace_ret(1)
> +
> +  recursive_f()
> +  -- Stop the side trace (3) here after exiting the trace (2) with
> +  -- up-recursion.
> +  modf(1)
> +
> +  -- Start the side trace (4).
> +  trace_ret('')
> +  -- luacheck: no unused
> +  -- Generate register pressure to force spills.
> +  -- The amount is well-suited for arm64 and x86_64.
> +  local l1 = ffi_new(int64_t, call_counter + 1)
> +  local l2 = ffi_new(int64_t, call_counter + 2)
> +  local l3 = ffi_new(int64_t, call_counter + 3)
> +  local l4 = ffi_new(int64_t, call_counter + 4)
> +  local l5 = ffi_new(int64_t, call_counter + 5)
> +  local l6 = ffi_new(int64_t, call_counter + 6)
> +  local l7 = ffi_new(int64_t, call_counter + 7)
> +  local l8 = ffi_new(int64_t, call_counter + 8)
> +  local l9 = ffi_new(int64_t, call_counter + 9)
> +  local l10 = ffi_new(int64_t, call_counter + 10)
> +  local l11 = ffi_new(int64_t, call_counter + 11)
> +  local l12 = ffi_new(int64_t, call_counter + 12)
> +  local l13 = ffi_new(int64_t, call_counter + 13)
> +  -- Simulate the return to the same function using the additional
> +  -- frame for down-recursion.
> +  return trace_ret(extra_frame())
> +end
> +
> +jit.opt.start('hotloop=1', 'hotexit=1', 'recunroll=1')
> +
> +recursive_f()
> +
> +test:ok(true, 'no crash during execution of a trace with down-recursion')
> +
> +test:done(true)

  parent reply	other threads:[~2024-03-26 15:08 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-03-19 16:41 Sergey Kaplun via Tarantool-patches
2024-03-21 10:41 ` Maxim Kokryashkin via Tarantool-patches
2024-03-26 15:08 ` Sergey Bronnikov via Tarantool-patches [this message]
2024-04-03  6:12   ` Sergey Kaplun via Tarantool-patches
2024-04-03 14:19     ` Sergey Bronnikov via Tarantool-patches
2024-04-11 17:03 ` Sergey Kaplun via Tarantool-patches

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=b27cb112-cef8-4f78-b516-2238b2d98da0@tarantool.org \
    --to=tarantool-patches@dev.tarantool.org \
    --cc=m.kokryashkin@tarantool.org \
    --cc=sergeyb@tarantool.org \
    --cc=skaplun@tarantool.org \
    --subject='Re: [Tarantool-patches] [PATCH luajit] Prevent down-recursion for side traces.' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox