runtime/trace: TestTraceSymbolize broken on solaris/amd64 as of CL 463835

[bcmills]

https://build.golang.org/log/ec79810c83e1ddca4758df05fd4b8f0351fe462a:

solaris-amd64-oraclerel at ba71817390f78bf8c479dc65d1bc51db98d667a7
…
--- FAIL: TestTraceSymbolize (0.24s)
    trace_stack_test.go:285: Did not match event GoSysCall with stack
          syscall.read					 :0
          syscall.Read					 :0
          internal/poll.ignoringEINTRIO			 :0
          internal/poll.(*FD).Read			 :0
          os.(*File).read				 :0
          os.(*File).Read				 :0
          runtime/trace_test.TestTraceSymbolize.func11	 :0

        Seen 43 events of the type
…
        Offset 855
          syscall.Read					/go/src/syscall/syscall_unix.go:181
          internal/poll.ignoringEINTRIO			/go/src/internal/poll/fd_unix.go:793
          internal/poll.(*FD).Read			/go/src/internal/poll/fd_unix.go:163
          os.(*File).read				/go/src/os/file_posix.go:31
          os.(*File).Read				/go/src/os/file.go:118
          runtime/trace_test.TestTraceSymbolize.func11	/go/src/runtime/trace/trace_stack_test.go:105
…
        Offset 2672
          syscall.Read					/go/src/syscall/syscall_unix.go:181
          internal/poll.ignoringEINTRIO			/go/src/internal/poll/fd_unix.go:793
          internal/poll.(*FD).Read			/go/src/internal/poll/fd_unix.go:163
          os.(*File).read				/go/src/os/file_posix.go:31
          os.(*File).Read				/go/src/os/file.go:118
          runtime/trace_test.TestTraceSymbolize.func11	/go/src/runtime/trace/trace_stack_test.go:105

This failure mode seems to have been introduced by CL 463835. It isn't clear to me whether it affects illumos because the test results are blocked by #58967.

(attn @golang/solaris @felixge @prattmic)

[prattmic]

syscall.read calls syscall.sysvicall6, which is a simple jump to runtime.syscall_sysvicall6.

I'd guess that one of those latter frames doesn't have proper framepointer handling.

[bcmills]

Looks like that call sequence does also apply to illumos (attn @golang/illumos), so adding that tag.

[felixge]

syscall.read calls syscall.sysvicall6, which is a simple jump to runtime.syscall_sysvicall6.

I'd guess that one of those latter frames doesn't have proper framepointer handling.

Sounds plausible. It would probably nice to fix the framepointer handling.

But for now we could also try to special-case the syscall test expectations for solaris and illumos? I could probably send a patch for the latter quickly, the former might take more time.

[gopherbot]

Change https://go.dev/cl/481336 mentions this issue: runtime/trace: Fix TestTraceSymbolize on solaris and illumos

[felixge]

Change https://go.dev/cl/481336 mentions this issue: runtime/trace: Fix TestTraceSymbolize on solaris and illumos

This should fix the problem for now.

As far as I can tell the issue wasn't caused by frames with incorrect frame pointer handling, but rather by the different number of frames between the trace events and the actual syscalls on different platforms.