Re: [RFC PATCH for 4.15 14/14] Restartable sequences: Provide self-tests

On Mon, Oct 16, 2017 at 02:00:33PM +1100, Michael Ellerman wrote:

Mathieu Desnoyers [off-list ref] writes:

quoted

Implements two basic tests of RSEQ functionality, and one more
exhaustive parameterizable test.

The first, "basic_test" only asserts that RSEQ works moderately
correctly.
E.g. that:
- The CPUID pointer works
- Code infinitely looping within a critical section will eventually be
  interrupted.
- Critical sections are interrupted by signals.

"basic_percpu_ops_test" is a slightly more "realistic" variant,
implementing a few simple per-cpu operations and testing their
correctness.

"param_test" is a parametrizable restartable sequences test. See
the "--help" output for usage.

As part of those tests, a helper library "rseq" implements a user-space
API around restartable sequences. It uses the cpu_opv system call as
fallback when single-stepped by a debugger. It exposes the instruction
pointer addresses where the rseq assembly blocks begin and end, as well
as the associated abort instruction pointer, in the __rseq_table
section. This section allows debuggers may know where to place
breakpoints when single-stepping through assembly blocks which may be
aborted at any point by the kernel.

The following rseq APIs are implemented in this helper library:
- rseq_register_current_thread()/rseq_unregister_current_thread():
    register/unregister current thread's use of rseq,
- rseq_current_cpu_raw():
    current CPU number,
- rseq_start():
    beginning of a restartable sequence,
- rseq_cpu_at_start():
    CPU number at start of restartable sequence,
- rseq_finish():
    End of restartable sequence made of zero or more loads, completed by
    a word-sized store,
- rseq_finish2():
    End of restartable sequence made of zero or more loads, one
    speculative word-sized store, completed by a word-sized store,
- rseq_finish2_release():
    End of restartable sequence made of zero or more loads, one
    speculative word-sized store, completed by a word-sized store with
    release semantic,
- rseq_finish_memcpy():
    End of restartable sequence made of zero or more loads, a
    speculative copy of a variable length memory region, completed by a
    word-sized store.
- rseq_finish_memcpy_release():
    End of restartable sequence made of zero or more loads, a
    speculative copy of a variable length memory region, completed by a
    word-sized store with release semantic.

PowerPC tests have been implemented by Boqun Feng.

Hi Boqun,

Hello Michael,

I'm having trouble testing these, I get:

  ~/linus/tools/testing/selftests/cpu-opv$ ./basic_cpu_opv_test
  Testing test_compare_eq same
  Testing test_compare_eq different
  Testing test_compare_ne same
  Testing test_compare_ne different
  Testing test_2compare_eq index
  Testing test_2compare_ne index
  Testing test_memcpy
  Testing test_memcpy_u32
  Testing test_add
  Testing test_two_add
  Testing test_or
  Testing test_and
  Testing test_xor
  Testing test_lshift
  Testing test_rshift
  Testing test_cmpxchg success
  Testing test_cmpxchg fail
  
  ~/linus/tools/testing/selftests/rseq$ ./basic_test
  testing current cpu
  testing critical section
  testing critical section is interrupted by signal

  ~/linus/tools/testing/selftests/rseq$ ./basic_percpu_ops_test
  ./basic_percpu_ops_test: error while loading shared libraries: R_PPC64_ADDR16_HI re10d8f10a0 for symbol `' out of range
  ~/linus/tools/testing/selftests/rseq$ ./param_test
  ./param_test: error while loading shared libraries: R_PPC64_ADDR16_HI re136251b48 for symbol `' out of range

I think this one is due to the same reason as:

	7998eb3dc700 ("powerpc: Fix 64 bit builds with binutils 2.24")

I have made the fix before, but seems forgot to send it to Mathieu...

so would this help?
	

diff --git a/tools/testing/selftests/rseq/rseq-ppc.h b/tools/testing/selftests/rseq/rseq-ppc.h
index bc78b4fd72b1..39cbabe89b0e 100644
--- a/tools/testing/selftests/rseq/rseq-ppc.h
+++ b/tools/testing/selftests/rseq/rseq-ppc.h

@@ -74,7 +74,7 @@ do {									\
 		"lis %%r17, (3b)@highest\n\t" \
 		"ori %%r17, %%r17, (3b)@higher\n\t" \
 		"rldicr %%r17, %%r17, 32, 31\n\t" \
-		"oris %%r17, %%r17, (3b)@h\n\t" \
+		"oris %%r17, %%r17, (3b)@high\n\t" \
 		"ori %%r17, %%r17, (3b)@l\n\t" \
 		"std %%r17, 0(%[rseq_cs])\n\t" \
 		RSEQ_INJECT_ASM(2) \

Regards,
Boqun

Any idea what's going on with the last two? I assume you don't see that
in your test setup :)

cheers