Debugging compressed pointers in Blink

[Daniel Libby]

https://crrev.com/c/3835682 enabled pointer compression for Blink Member<> pointers. How are folks handling these while debugging (either live or crash dumps)? Is there some tooling available that will look up and apply the cage base? 

I didn't see anything mentioned in https://docs.google.com/document/d/1neGN8Jq-1JLrWK3cvwRIfrSjLgE0Srjv-uq7Ze38Iao/edit# but maybe there are better known tools/techniques from v8 (which IIUC has had compressed pointers for some time now).

[Kentaro Hara]

+Michael Lippautz 

2022年9月22日(木) 4:01 'Daniel Libby' via blink-dev <blin...@chromium.org>:

https://crrev.com/c/3835682 enabled pointer compression for Blink Member<> pointers. How are folks handling these while debugging (either live or crash dumps)? Is there some tooling available that will look up and apply the cage base? 

I didn't see anything mentioned in https://docs.google.com/document/d/1neGN8Jq-1JLrWK3cvwRIfrSjLgE0Srjv-uq7Ze38Iao/edit# but maybe there are better known tools/techniques from v8 (which IIUC has had compressed pointers for some time now).

--
You received this message because you are subscribed to the Google Groups "blink-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to blink-dev+...@chromium.org.
To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/4b3e5bbb-134a-4483-9a1e-8e33fbc6f38en%40chromium.org.

[Anton Bikineev]

We have plans to provide more debugging tooling for Oilpan. I haven't had a need to examine compressed pointers myself, however I see that some simple gdb/windbg function that'd follow pointers would be useful.

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CABg10jxSdx3JzWvTC2j7RUEcw01P9gTRhT%3DnftwehY3br4qNWA%40mail.gmail.com.

[Daniel Cheng]

Is there a callable C++ function that can turn a compressed pointer into the actual pointer value?

Daniel

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CABH6udYH6VGy8Ooz6ZLZWpWR_LjjQfu1xh-AMBvWoyAnArojbA%40mail.gmail.com.

[Anton Bikineev]

There is "cppgc::internal::CompressedPointer::Decompress(void*)". I would, however, hide it behind a simpler name in .gdbinit.

[Ian Kilpatrick]

Is there a bug to follow regarding the debuggability of the pointers?

Ian

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CABH6udZS_4iw6Vhb%3Dtf0J9qgV%3DbwbcWAu%2B--Xt9dYrDpqDecXw%40mail.gmail.com.

[Stefan Zager]

I ran into this today, and it's pretty frustrating:

(rr) p inner_node_                                                                              
$11 = {                                                                                          
  <cppgc::internal::MemberBase> = {                                                              
    raw_ = {                                                                                    
      static kCompressedSentinel = 1,                                                            
      value_ = 2148082696                                                                        
    }                                                                                            
  },                                                                                            
  <cppgc::internal::DisabledCheckingPolicy> = {<No data fields>}, <No data fields>}              
(rr) p inner_node_.Load()                                                                        
Couldn't find method blink::Member<blink::Node>::Load                                            
(rr) p inner_node_.raw_.Load()                                                                  
Cannot evaluate function -- may be inlined                                                      
(rr) p cppgc::internal::CompressedPointer::Decompress(inner_node_.raw_.value_)                  
Cannot evaluate function -- may be inlined                                                      
(rr)  

I found that I can avoid the issue with this in args.gn:

cppgc_enable_caged_heap = false

cppgc_enable_pointer_compression = false

... but I would prefer a better solution. 

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAJL3UpQ%2BiPOAMX6ONDxRm9UG0zFRkScw1%3DdALFGdKJOUiAUpEw%40mail.gmail.com.

[Daniel Cheng]

For now, I guess it should be sufficient to add something to v8/include/cppgc/internal/member-storage.h like

#ifndef NDEBUG  // DCHECK_IS_ON() would be nicer, but not sure what v8 uses

extern "C" void* DecompressPointerForDebugger(uint32_t value) {

  return cppgc::internal::CompressedPointer::Decompress(value);
}

#endif

Which should hopefully be callable from the debugger and not eliminated by the linker?

Daniel

[Stefan Zager]

Maybe add NOINLINE for good measure, but yeah, I think that should work.

[Anton Bikineev]

Which should hopefully be callable from the debugger and not eliminated by the linker?

We can apply __attribute__((used,retain)) to make sure that the linker doesn't strip the functions with --gc-sections, same as did for V8 helpers (a week ago :) ). I'll prepare a CL.

[Anton Bikineev]

I'm now submitting a CL that adds some quick gdb helpers:

  (gdb) source v8/tools/gdbinit

  # Print Member:

  (gdb) cpm <member-name>

  # Print Compressed Pointer:

  (gdb) cpcp <compressed-32-bit-value>

It currently erases the pointee type, but this can be improved in the future.

[Dave Tapuska]

Anton, can you add this to https://source.chromium.org/chromium/chromium/src/+/main:docs/linux/debugging.md ?

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CABH6udZ7rMMqmsZ98RbS21yG_DBXXtnoOVYnO4oV%3Dno3tFvNug%40mail.gmail.com.

[Christian Biesinger]

Perhaps we should add a prettyprinter for cppgc::internal::MemberBase which automatically calls that function?

Christian

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAHgVhZWN15bg8nzz8mygH%3D99ZHeEu5gGbyQc_xKAO%2BuGJ7WhcQ%40mail.gmail.com.

[Anton Bikineev]

Perhaps we should add a prettyprinter for cppgc::internal::MemberBase which automatically calls that function?

I just added a pretty-printer for (cppgc|blink)::(Weak|Untraced)Member. Any feedback is appreciated!

Anton, can you add this to https://source.chromium.org/chromium/chromium/src/+/main:docs/linux/debugging.md ?

Since simple 'print' should work now, I think that 'cpcp' and 'cpm' can be obsolete. A general note: we plan to add debugging support for Oilpan (in our OKRs). As soon as we agree on the interface, it would probably be worth updating debugging.md.