Tarantool development patches archive
 help / color / mirror / Atom feed
From: Igor Munkin <imun@tarantool.org>
To: Sergey Ostanevich <sergos@tarantool.org>,
	Vladislav Shpilevoy <v.shpilevoy@tarantool.org>,
	Kirill Yukhin <kyukhin@tarantool.org>
Cc: tarantool-patches@dev.tarantool.org
Subject: [Tarantool-patches] [PATCH luajit 2/2] jit: abort trace execution on JIT mode change
Date: Fri, 27 Mar 2020 13:47:51 +0300	[thread overview]
Message-ID: <6a753febbdde86642257c57f7e3a3b1700317ca3.1585304087.git.imun@tarantool.org> (raw)
In-Reply-To: <cover.1585304087.git.imun@tarantool.org>

Current luaJIT_setmode implementation aborts trace recording but nothing
prevents calling it on already compiled trace. E.g. if one conditionally
calls an FFI function having luaJIT_setmode with LUAJIT_MODE_FLUSH mode
underneath, the trace being executed can be purged and the return
address is invalidated as a result (since the mcode is released).

This changeset prohibits luaJIT_setmode call while mcode is being
executed leading to platform panic if the call occurs.

Signed-off-by: Igor Munkin <imun@tarantool.org>
---
 src/lj_dispatch.c                     |  5 +++++
 src/lj_errmsg.h                       |  1 +
 test/lj-flush-on-trace/CMakeLists.txt |  1 +
 test/lj-flush-on-trace/libflush.c     | 31 +++++++++++++++++++++++++++
 test/lj-flush-on-trace/test.lua       | 25 +++++++++++++++++++++
 5 files changed, 63 insertions(+)
 create mode 100644 test/lj-flush-on-trace/CMakeLists.txt
 create mode 100644 test/lj-flush-on-trace/libflush.c
 create mode 100644 test/lj-flush-on-trace/test.lua

diff --git a/src/lj_dispatch.c b/src/lj_dispatch.c
index 5d6795f..b3448c8 100644
--- a/src/lj_dispatch.c
+++ b/src/lj_dispatch.c
@@ -240,6 +240,11 @@ int luaJIT_setmode(lua_State *L, int idx, int mode)
 {
   global_State *g = G(L);
   int mm = mode & LUAJIT_MODE_MASK;
+  if (tvref(g->jit_base)) {
+    setstrV(L, L->top++, lj_err_str(L, LJ_ERR_JITMODE));
+    if (g->panic) g->panic(L);
+    exit(EXIT_FAILURE);
+  }
   lj_trace_abort(g);  /* Abort recording on any state change. */
   /* Avoid pulling the rug from under our own feet. */
   if ((g->hookmask & HOOK_GC))
diff --git a/src/lj_errmsg.h b/src/lj_errmsg.h
index 1580385..de7b867 100644
--- a/src/lj_errmsg.h
+++ b/src/lj_errmsg.h
@@ -113,6 +113,7 @@ ERRDEF(NOJIT,	"JIT compiler permanently disabled by build option")
 #endif
 ERRDEF(JITOPT,	"unknown or malformed optimization flag " LUA_QS)
 ERRDEF(JITCALL,	"Lua VM re-entrancy is detected while executing the trace")
+ERRDEF(JITMODE,	"JIT mode change is detected while executing the trace")
 
 /* Lexer/parser errors. */
 ERRDEF(XMODE,	"attempt to load chunk with wrong mode")
diff --git a/test/lj-flush-on-trace/CMakeLists.txt b/test/lj-flush-on-trace/CMakeLists.txt
new file mode 100644
index 0000000..a90452d
--- /dev/null
+++ b/test/lj-flush-on-trace/CMakeLists.txt
@@ -0,0 +1 @@
+build_lualib(libflush libflush.c)
diff --git a/test/lj-flush-on-trace/libflush.c b/test/lj-flush-on-trace/libflush.c
new file mode 100644
index 0000000..177409a
--- /dev/null
+++ b/test/lj-flush-on-trace/libflush.c
@@ -0,0 +1,31 @@
+#include <lua.h>
+#include <luajit.h>
+
+struct flush {
+	lua_State *L; /* Coroutine saved to change JIT mode */
+	int trigger;  /* Trigger for flushing all traces */
+};
+
+void flush(struct flush *state, int i)
+{
+	if (i < state->trigger)
+		return;
+
+	/* Trace flushing is triggered */
+	(void)luaJIT_setmode(state->L, 0, LUAJIT_MODE_ENGINE|LUAJIT_MODE_FLUSH);
+}
+
+static int init(lua_State *L)
+{
+	struct flush *state = lua_newuserdata(L, sizeof(struct flush));
+
+	state->L = L;
+	state->trigger = lua_tonumber(L, 1);
+	return 1;
+}
+
+LUA_API int luaopen_libflush(lua_State *L)
+{
+	lua_pushcfunction(L, init);
+	return 1;
+}
diff --git a/test/lj-flush-on-trace/test.lua b/test/lj-flush-on-trace/test.lua
new file mode 100644
index 0000000..ff6e0b6
--- /dev/null
+++ b/test/lj-flush-on-trace/test.lua
@@ -0,0 +1,25 @@
+local cfg = {
+  hotloop = arg[1] or 1,
+  trigger = arg[2] or 1,
+}
+
+local ffi = require('ffi')
+local ffiflush = ffi.load('libflush')
+ffi.cdef('void flush(struct flush *state, int i)')
+
+-- Save the current coroutine and set the value to trigger ipp
+-- call the Lua routine instead of pure C implementation.
+local flush = require('libflush')(cfg.trigger)
+
+-- Depending on trigger and hotloop values the following contexts
+-- are possible:
+-- * if trigger <= hotloop -> trace recording is aborted
+-- * if trigger >  hotloop -> trace is recorded but execution
+--   leads to panic
+jit.opt.start("3", string.format("hotloop=%d", cfg.hotloop))
+
+for i = 0, cfg.trigger + cfg.hotloop do
+  ffiflush.flush(flush, i)
+end
+-- Panic didn't occur earlier.
+print('OK')
-- 
2.25.0

  parent reply	other threads:[~2020-03-27 10:54 UTC|newest]

Thread overview: 20+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-03-27 10:47 [Tarantool-patches] [PATCH luajit 0/2] Trace abort on FFI sandwich or " Igor Munkin
2020-03-27 10:47 ` [Tarantool-patches] [PATCH luajit 1/2] jit: abort trace recording and execution for C API Igor Munkin
2020-03-28 16:33   ` Sergey Ostanevich
2020-03-28 20:30     ` Igor Munkin
2020-03-29  9:21       ` Sergey Ostanevich
2020-03-29 10:45         ` Igor Munkin
2020-03-30  8:58           ` Sergey Ostanevich
2020-03-30 14:25             ` Igor Munkin
2020-04-03 21:06               ` Sergey Ostanevich
2020-04-03 21:31                 ` Igor Munkin
2020-04-02 23:41   ` Vladislav Shpilevoy
2020-04-04 11:55     ` Igor Munkin
2020-04-04 21:37       ` Vladislav Shpilevoy
2020-04-07 21:16         ` Igor Munkin
2020-03-27 10:47 ` Igor Munkin [this message]
2020-03-28 19:36   ` [Tarantool-patches] [PATCH luajit 2/2] jit: abort trace execution on JIT mode change Sergey Ostanevich
2020-03-29 10:46     ` Igor Munkin
2020-04-02 23:41 ` [Tarantool-patches] [PATCH luajit 0/2] Trace abort on FFI sandwich or " Vladislav Shpilevoy
2020-04-03 21:32   ` Igor Munkin
2020-04-04 21:36     ` Vladislav Shpilevoy

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=6a753febbdde86642257c57f7e3a3b1700317ca3.1585304087.git.imun@tarantool.org \
    --to=imun@tarantool.org \
    --cc=kyukhin@tarantool.org \
    --cc=sergos@tarantool.org \
    --cc=tarantool-patches@dev.tarantool.org \
    --cc=v.shpilevoy@tarantool.org \
    --subject='Re: [Tarantool-patches] [PATCH luajit 2/2] jit: abort trace execution on JIT mode change' \
    /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