arm64 pc_irpc test failure

[ikitayama]

On arm64, I don't seem to get the pc_irpc test passed.

$ ./test_driver -test pc_irpc -log -
Commencing test(s) ...
Sun Dec 11 23:03:40 CST 2016
Linux r2-a21 4.8.10-300.fc25.aarch64 #1 SMP Wed Nov 23 03:04:18 UTC 2016 aarch64 aarch64 aarch64 GNU/Linux
TESTDIR=/home/itaru.kitayama/builds/testsuite
[Tests with pc_irpc.dyn_g++_64_none_none]
-rwxr-xr-x 1 itaru.kitayama itaru.kitayama 111488 Dec 11 19:45 pc_irpc.dyn_g++_64_none_none*
	linux-vdso.so.1 (0x000003ff78760000)
	libdl.so.2 => /lib64/libdl.so.2 (0x000003ff78700000)
	libpthread.so.0 => /lib64/libpthread.so.0 (0x000003ff786c0000)
	libstdc++.so.6 => /lib64/libstdc++.so.6 (0x000003ff78510000)
	libm.so.6 => /lib64/libm.so.6 (0x000003ff78440000)
	libgcc_s.so.1 => /lib64/libgcc_s.so.1 (0x000003ff78410000)
	libc.so.6 => /lib64/libc.so.6 (0x000003ff78270000)
	/lib/ld-linux-aarch64.so.1 (0x000002aae0730000)

Running: allocation_mode=manual_allocate post_time=post_sequential post_to=post_to_proc rpc_sync=rpc_use_sync thread_start=rpc_start_stopped

[wrwilliams]

Can you capture a log with DYNINST_DEBUG_PROCCONTROL=1 set in your environment and post it as a gist? (It'll probably be huge, but it'll help the assorted experts track things down.)

[ikitayama]

The ptrace system is failing for some reason (directly calling it in my test program at least retrieves the regset) https://gist.github.com/ikitayama/93bbf999851d8fe606b57e6c63482565

Note that I'm testing programs together with the modified ProcControlAPI on a v4.9 kernel.

[wrwilliams]

Well, that's funny; why is ProcControl building without arch_aarch64 defined? That code clearly went through the old-style peekuser method of getting registers, which is not implemented in the kernel on ARM64. See the #ifdef at linux.C:2391 on master; as you tested manually, we should be using getregset or nothing.

[ikitayama]

I think it is executing ptrace with GETREGSET properly, not sure as to why EFAULT is returned.

[ikitayama]

FWIW, ptrace systemcall can't get regset with the 4.8.10-300.fc25.aarch64 kernel either.

[ikitayama]

I take it back. The ptrace system call does not always succeed.

[wrwilliams]

Fascinating. You see the log output that should only come from the PEEKUSER path, though?

Unified strace and debug logging would confirm what we did, as would a run under a debugger. And an upgrade from F21 is looking tricky, so I'm afraid I'm relying on you for reports from a current kernel for the moment.

[ikitayama]

@wrwilliams executing this

strace -o /tmp/strace.log ./test_driver -test pc_irpc

doesn't record the ptrace invocation with its parameters.

[ikitayama]

I've pushed another patch to my local repository. See PR #294.

[wrwilliams]

Is this resolved now that #294 is merged in?

[ikitayama]

In my test environment, yes.

$ ./test_driver -test pc_irpc -all
TEST                       COMP   OPT  ABI MODE     THREAD  LINK    PIC     RESULT
pc_irpc                    g++    none 64  create   SPST    dynamic nonPIC  PASSED
pc_irpc                    g++    none 64  create   MPST    dynamic nonPIC  PASSED
pc_irpc                    g++    none 64  create   SPMT    dynamic nonPIC  PASSED
pc_irpc                    g++    none 64  create   MPMT    dynamic nonPIC  PASSED
pc_irpc                    g++    none 64  attach   SPST    dynamic nonPIC  PASSED
pc_irpc                    g++    none 64  attach   MPST    dynamic nonPIC  PASSED
pc_irpc                    g++    none 64  attach   SPMT    dynamic nonPIC  PASSED
pc_irpc                    g++    none 64  attach   MPMT    dynamic nonPIC  PASSED
pc_irpc                    gcc    none 64  create   SPST    dynamic nonPIC  PASSED
pc_irpc                    gcc    none 64  create   MPST    dynamic nonPIC  PASSED
pc_irpc                    gcc    none 64  create   SPMT    dynamic nonPIC  PASSED
pc_irpc                    gcc    none 64  create   MPMT    dynamic nonPIC  PASSED
pc_irpc                    gcc    none 64  attach   SPST    dynamic nonPIC  PASSED
pc_irpc                    gcc    none 64  attach   MPST    dynamic nonPIC  PASSED
pc_irpc                    gcc    none 64  attach   SPMT    dynamic nonPIC  PASSED
pc_irpc                    gcc    none 64  attach   MPMT    dynamic nonPIC  PASSED
pc_irpc                    gcc    none 64  create   SPST    static  nonPIC  PASSED
pc_irpc                    gcc    none 64  create   MPST    static  nonPIC  PASSED
pc_irpc                    gcc    none 64  create   SPMT    static  nonPIC  PASSED
pc_irpc                    gcc    none 64  create   MPMT    static  nonPIC  PASSED
pc_irpc                    gcc    none 64  attach   SPST    static  nonPIC  PASSED
pc_irpc                    gcc    none 64  attach   MPST    static  nonPIC  PASSED
pc_irpc                    gcc    none 64  attach   SPMT    static  nonPIC  PASSED
pc_irpc                    gcc    none 64  attach   MPMT    static  nonPIC  PASSED

[wrwilliams]

Excellent. Feel free to close things you report, by the way; we can always reopen them if we find there's more to do.