Once the unwinder is added (see bug 1232712), we should update the included frame filter to find the function and script info and return it. I'll attach a patch that adds function info and shows where the script info should be added. I'm not 100% sure this is correct but it has at least once printed information that was useful...

MozReview-Commit-ID: LJEr9sNtdMV

See https://bugzilla.mozilla.org/show_bug.cgi?id=1232712#c20 to understand what is wrong with this patch.

I've updated this patch (will attach soon) but it is still a bit wrong. It shows: #33 0x00007f0d1c475f61 in <<JitFrame_Exit>> () #34 0x00007f0cdadc5bd3 in <<JitFrame_BaselineStub>> () #35 0x00007f0c903720bc in <<JitFrame_BaselineJS "Handler.prototype.process">> () #36 0x00007f0cd8bd3e44 in <<JitFrame_BaselineStub>> () #37 0x00007f0c90483488 in <<JitFrame_BaselineJS "this.PromiseWalker.walkerLoop">> () #38 0x00007f0cd8bd3e44 in <<JitFrame_BaselineStub>> () #39 0x00007f0c90721072 in <<JitFrame_BaselineJS "bound ">> () #40 0x00007f0cd8bd3e44 in <<JitFrame_BaselineStub>> () #41 0x00007f0c90721072 in <<JitFrame_BaselineJS "bound bound ">> () #42 0x00007f0d1c474d94 in <<JitFrame_Entry "callFunctionWithAsyncStack">> () What would be correct here? I gather that these are generally off by one; does that include moving the function name from frame #35 onto frame #34? Also, "bound bound" seems pretty unfriendly and I'm wondering if there's a better way to extract the name. Currently it's using the "atom_" member of JSFunction.

Updated for changes in final unwinder patch. MozReview-Commit-ID: LJEr9sNtdMV

I also tried adding arguments here, but sometimes I get strange values for numActualArgs_. For example I can easily get a trace where this is 140737488348144. I did verify in this case that the callee token claims this is a function call. When the value is correct we can now print some arguments: #6 0x00007ffff7ff20f3 in <<JitFrame_BaselineStub "f1">> (this=JSVAL_VOID, arg1=$jsval(4700)) There's also this buglet: #7 0x00007ffff7fc096b in <<JitFrame_BaselineJS <error reading variable: Cannot access memory at address 0xfff9000000000000>>> ()

(In reply to Tom Tromey :tromey from comment #5) > When the value is correct we can now print some arguments: > > #6 0x00007ffff7ff20f3 in <<JitFrame_BaselineStub "f1">> (this=JSVAL_VOID, > arg1=$jsval(4700)) Only BaselineJS and IonJS should have a calleeToken associated to them. I will look at the patch to see what might be going wrong, but the BaselineStub header are not supposed to have a JitFrameLayout. > There's also this buglet: > > #7 0x00007ffff7fc096b in <<JitFrame_BaselineJS <error reading variable: > Cannot access memory at address 0xfff9000000000000>>> () 0xfff90… is a JS::Value which is "undefined", and this is probably because you are interpreting the content of the stack as a JitFrameLayout while it is not supposed to be one. I will look at the patch.

(In reply to Nicolas B. Pierron [:nbp] from comment #6) > Only BaselineJS and IonJS should have a calleeToken associated to them. This happened due to moving callee information to the next frame; which is wrong, but I don't really understand what is going on here. I'll attach my WIP patch that adds arguments as well.

MozReview-Commit-ID: LJEr9sNtdMV

The patch I'm about to attach adds function names, parameters, and filenames to the backtrace. It doesn't add line numbers, as that's more complicated; I'll file a follow-up bug. It also has one hack to work around bug 987069 for now; I'll file a follow-up for this as well.

Comment on attachment 8806004 [details] Bug 1254295 - add function name to gdb frame filter; https://reviewboard.mozilla.org/r/89578/#review89742 ::: js/src/gdb/mozilla/unwind.py:178 (Diff revision 1) > + except gdb.MemoryError: > + function = "(could not read function name)" > + script = fptr['u']['i']['s']['script_'] > + elif tag == self.cache.CalleeToken_Script: > + script = gdb.Value(calleetoken).cast(self.cache.JSScript) > + return [function, script] nit: using a dictionary would make this easier to extend as needed. ::: js/src/gdb/mozilla/unwind.py:223 (Diff revision 1) > + calleetoken = long(this_frame['calleeToken_']) > + tag = calleetoken & 3 > + if tag != self.cache.CalleeToken_Function and tag != self.cache.CalleeToken_FunctionConstructing: > + return FrameDecorator.frame_args(self) nit: Use _decode_jitframe and test if function is None.

Comment on attachment 8806005 [details] Bug 1254295 - fix a few typos in JS headers; https://reviewboard.mozilla.org/r/89580/#review89752

Comment on attachment 8806006 [details] Bug 1254295 - add "flushregs" advice to unwinder instructions; https://reviewboard.mozilla.org/r/89582/#review89754

Pushed by ttromey@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/1e2587f668f4 add function name to gdb frame filter; r=nbp https://hg.mozilla.org/integration/autoland/rev/52736ea528a9 fix a few typos in JS headers; r=nbp https://hg.mozilla.org/integration/autoland/rev/5ba29d0fdd6d add "flushregs" advice to unwinder instructions; r=nbp

https://hg.mozilla.org/mozilla-central/rev/1e2587f668f4 https://hg.mozilla.org/mozilla-central/rev/52736ea528a9 https://hg.mozilla.org/mozilla-central/rev/5ba29d0fdd6d