[Tarantool-patches] [PATCH luajit] Fix maxslots when recording BC_VARG, part 3.

Sergey Kaplun skaplun at tarantool.org
Wed Aug 16 19:11:57 MSK 2023


Hi, Maxim!
Thanks for the review!
Fixed your comments and force-pushed the branch.

On 16.08.23, Maxim Kokryashkin wrote:
> Hi, Sergey!
> Thanks for the patch!
> Please consider my comments below.
> 
> On Tue, Aug 15, 2023 at 03:32:15PM +0300, Sergey Kaplun wrote:
> > From: Mike Pall <mike>
> > 
> > Thanks to Peter Cawley.
> > 
> > (cherry-picked from commit abb27c7771947e082c9d919d184ad5f5f03e2e32)
> > 
> > In case, when `BC_VARG` set the VARG slot to the non-top stack slot,
> Typo: s/set/sets/

Fixed.

> > `maxslot` value was unconditionally set to the destination slot, so some
> > top slots may be omitted in the snapshot entry. Since these slots are
> > omitted, they are not restored correctly, when restoring from snapshot
> Typo: s/snapshot/a snapshot/

Fixed.

> > for this side exit.
> > 
> > This patch adds the check for the aforementioned case, to avoid maxslot
> > shrinking.
> > 
> > Sergey Kaplun:
> > * added the description and the test for the problem
> > 
> > Part of tarantool/tarantool#8825
> > ---
> > 
> > Branch: https://github.com/tarantool/luajit/tree/skaplun/lj-1046-fix-bc-varg-recording
> > PR: https://github.com/tarantool/tarantool/pull/8986
> > Related issues:
> > * https://github.com/LuaJIT/LuaJIT/issues/1046
> > * https://github.com/tarantool/tarantool/issues/8825
> > 
> >  src/lj_record.c                               | 12 +++-
> >  .../lj-1046-fix-bc-varg-recording.test.lua    | 58 +++++++++++++++++++
> >  2 files changed, 67 insertions(+), 3 deletions(-)
> >  create mode 100644 test/tarantool-tests/lj-1046-fix-bc-varg-recording.test.lua
> > 
> > diff --git a/src/lj_record.c b/src/lj_record.c
> > index 34d1210a..6bcdb04c 100644
> > --- a/src/lj_record.c
> > +++ b/src/lj_record.c

<snipped>

> > diff --git a/test/tarantool-tests/lj-1046-fix-bc-varg-recording.test.lua b/test/tarantool-tests/lj-1046-fix-bc-varg-recording.test.lua
> > new file mode 100644
> > index 00000000..34c5c572
> > --- /dev/null
> > +++ b/test/tarantool-tests/lj-1046-fix-bc-varg-recording.test.lua
> > @@ -0,0 +1,58 @@
> > +local tap = require('tap')
> > +local test = tap.test('lj-1046-fix-bc-varg-recording'):skipcond({
> > +  ['Test requires JIT enabled'] = not jit.status(),
> > +})
> > +
> > +test:plan(2)
> > +
> > +jit.opt.start('hotloop=1')
> > +
> > +-- luacheck: ignore
> > +local anchor
> > +local N_ITER = 5
> > +local SIDE_ITER = N_ITER - 1
> > +for i = 1, N_ITER do
> > +  -- In case, when `BC_VARG` set the VARG slot to the non-top
> > +  -- stack slot, `maxslot` value was unconditionally set to the
> > +  -- destination slot, so the following snapshot is used:
> > +  -- SNAP   #4   [ ---- ---- ---- nil  ]
> > +  -- instead of:
> > +  -- SNAP   #4   [ ---- nil  ---- ---- 0009 0001 ---- 0009 ]
> Snapshot examples here ceratinly give the idea of what goes wrong,
> but the `0009` and `0001` are meaningless by themselves. I think it would be
> nice to include IRs here too.

Added.

> > +  -- Since these slots are omitted, they are not restored
> > +  -- correctly, when restoring from snapshot for this side exit.
> Please fix the same typos as in the commit message here.

Fixed.

> > +  anchor = ...
> > +  if i > SIDE_ITER then
> > +    -- XXX: Don't use `test:ok()` here to avoid double-running of
> I think better phrasing would be:
> | `test:ok()` is not used here ...

Fixed.

> > +    -- tests in case of `i` incorrect restoring from the snapshot.
> Typo: s/restoring/restoration/

Fixed.

> > +    assert(i > SIDE_ITER)
> > +  end
> > +end
> > +
> > +test:ok(true, 'BC_VARG recording 0th frame depth, 1 result')
> > +
> > +-- Now the same case, but with an additional frame, so VARG slots
> > +-- are defined on the trace.
> > +local function varg_frame(anchor, i, side_iter, ...)
> > +  anchor = ...
> > +  -- In case, when `BC_VARG` set the VARG slot to the non-top
> > +  -- stack slot, `maxslot` value was unconditionally set to the
> > +  -- destination slot, so the following snapshot is used:
> > +  -- SNAP   #4   [ <snipped> | nil  nil  nil  `varg_frame` | nil ]
> > +  -- instead of:
> > +  -- SNAP   #4   [ <snipped> | nil  nil  nil  `varg_frame` | nil 0009 0005 ]
> > +  -- Since these slots are omitted, they are not restored
> > +  -- correctly, when restoring from snapshot for this side exit.
> I guess we don't need to repeat the entire comment again.

Dropped then.

> > +  if i > side_iter then
> > +    -- XXX: Don't use `test:ok()` here to avoid double-running of
> > +    -- tests in case of `i` incorrect restoring from the snapshot.
> Same typos as in the comment above.

Fixed.

> > +    assert(i > side_iter)

See the iterative patch below. Branch is force-pushed.

===================================================================
diff --git a/test/tarantool-tests/lj-1046-fix-bc-varg-recording.test.lua b/test/tarantool-tests/lj-1046-fix-bc-varg-recording.test.lua
index 34c5c572..30a87e54 100644
--- a/test/tarantool-tests/lj-1046-fix-bc-varg-recording.test.lua
+++ b/test/tarantool-tests/lj-1046-fix-bc-varg-recording.test.lua
@@ -12,18 +12,39 @@ local anchor
 local N_ITER = 5
 local SIDE_ITER = N_ITER - 1
 for i = 1, N_ITER do
-  -- In case, when `BC_VARG` set the VARG slot to the non-top
+  -- This trace generates the following IRs:
+  -- 0001 >  int SLOAD  #7    CRI
+  -- 0002 >  int LE     0001  +2147483646
+  -- 0003    int SLOAD  #6    CI
+  -- 0004    int SLOAD  #0    FR
+  -- 0005 >  int LE     0004  +11
+  -- 0006 >  num SLOAD  #5    T
+  -- 0007    num CONV   0003  num.int
+  -- ....        SNAP   #1   [ ---- ---- ---- nil  ]
+  -- 0008 >  num ULE    0007  0006
+  -- 0009  + int ADD    0003  +1
+  -- ....        SNAP   #2   [ ---- ---- ---- nil  ---- ---- ]
+  -- 0010 >  int LE     0009  0001
+  -- ....        SNAP   #3   [ ---- ---- ---- nil  ---- ---- 0009 0001 ---- 0009 ]
+  -- 0011 ------ LOOP ------------
+  -- 0012    num CONV   0009  num.int
+  -- ....        SNAP   #4   [ ---- ---- ---- nil  ]
+  --
+  -- In case, when `BC_VARG` sets the VARG slot to the non-top
   -- stack slot, `maxslot` value was unconditionally set to the
-  -- destination slot, so the following snapshot is used:
-  -- SNAP   #4   [ ---- ---- ---- nil  ]
+  -- destination slot, so the following snapshot (same for the #1)
+  -- is used:
+  -- ....        SNAP   #4   [ ---- ---- ---- nil  ]
   -- instead of:
-  -- SNAP   #4   [ ---- nil  ---- ---- 0009 0001 ---- 0009 ]
+  -- ....        SNAP   #4   [ ---- ---- ---- nil  ---- ---- 0009 0001 ---- 0009 ]
   -- Since these slots are omitted, they are not restored
-  -- correctly, when restoring from snapshot for this side exit.
+  -- correctly, when restoring from the snapshot for this side
+  -- exit.
   anchor = ...
   if i > SIDE_ITER then
-    -- XXX: Don't use `test:ok()` here to avoid double-running of
-    -- tests in case of `i` incorrect restoring from the snapshot.
+    -- XXX: `test:ok()` isn't used here to avoid double-running of
+    -- tests in case of `i` incorrect restoration from the
+    -- snapshot.
     assert(i > SIDE_ITER)
   end
 end
@@ -34,17 +55,10 @@ test:ok(true, 'BC_VARG recording 0th frame depth, 1 result')
 -- are defined on the trace.
 local function varg_frame(anchor, i, side_iter, ...)
   anchor = ...
-  -- In case, when `BC_VARG` set the VARG slot to the non-top
-  -- stack slot, `maxslot` value was unconditionally set to the
-  -- destination slot, so the following snapshot is used:
-  -- SNAP   #4   [ <snipped> | nil  nil  nil  `varg_frame` | nil ]
-  -- instead of:
-  -- SNAP   #4   [ <snipped> | nil  nil  nil  `varg_frame` | nil 0009 0005 ]
-  -- Since these slots are omitted, they are not restored
-  -- correctly, when restoring from snapshot for this side exit.
   if i > side_iter then
-    -- XXX: Don't use `test:ok()` here to avoid double-running of
-    -- tests in case of `i` incorrect restoring from the snapshot.
+    -- XXX: `test:ok()` isn't used here to avoid double-running of
+    -- tests in case of `i` incorrect restoration from the
+    -- snapshot.
     assert(i > side_iter)
   end
 end
===================================================================

<snipped>

> > 

-- 
Best regards,
Sergey Kaplun


More information about the Tarantool-patches mailing list