Re: [PING][PATCH v2 PR gdb/21870] aarch64: Leftover uncleared debug registers

[Alan Hayward <Alan dot Hayward at arm dot com>]

> On 12 Feb 2019, at 01:10, Weimin Pan <weimin.pan@oracle.com> wrote:
> 
> 
> On 2/11/2019 7:24 AM, Alan Hayward wrote:
>> 
>>>>>>>>>   diff --git a/gdb/testsuite/gdb.arch/aarch64-dbreg-contents.c b/gdb/testsuite/gdb.arch/aarch64-dbreg-contents.c
>>>>>>>>> new file mode 100644
>>>>>>>>> index 0000000..85d4a03
>>>>>>>>> --- /dev/null
>>>>>>>>> +++ b/gdb/testsuite/gdb.arch/aarch64-dbreg-contents.c
>>>>>>>>> @@ -0,0 +1,179 @@
>>>>>>>>> +/* Test case for setting a memory-write unaligned watchpoint on aarch64.
>>>>>>>>> +
>>>>>>>>> +  This software is provided 'as-is', without any express or implied
>>>>>>>>> +  warranty.  In no event will the authors be held liable for any damages
>>>>>>>>> +  arising from the use of this software.
>>>>>>>>> +
>>>>>>>>> +  Permission is granted to anyone to use this software for any purpose,
>>>>>>>>> +  including commercial applications, and to alter it and redistribute it
>>>>>>>>> +  freely.  */
>>>>>>>>> +
>>>>>>>>> +#define _GNU_SOURCE 1
>>>>>>>>> +#ifdef __ia64__
>>>>>>>>> +#define ia64_fpreg ia64_fpreg_DISABLE
>>>>>>>>> +#define pt_all_user_regs pt_all_user_regs_DISABLE
>>>>>>>>> +#endif  /* __ia64__ */
>>>>>>>>> +#include <sys/ptrace.h>
>>>>>>>>> +#ifdef __ia64__
>>>>>>>>> +#undef ia64_fpreg
>>>>>>>>> +#undef pt_all_user_regs
>>>>>>>>> +#endif  /* __ia64__ */
>>>>>>>>> +#include <linux/ptrace.h>
>>>>>>>>> +#include <sys/types.h>
>>>>>>>>> +#include <sys/user.h>
>>>>>>>>> +#if defined __i386__ || defined __x86_64__
>>>>>>>>> +#include <sys/debugreg.h>
>>>>>>>>> +#endif
>>>>>>>>> +
>> I’m not sure why you have all the x86 and IA64 checks.
>> The test will only be executed on AArch64 (because of the checks in the .exp file).
>> Could you remove all of those checks please.
> 
> The test case is likely to have been used for other targets as well. I've removed
> all non-aarch64 code and unused header files.
> 
>>>>>>>>> +#include <assert.h>
>>>>>>>>> +#include <unistd.h>
>>>>>>>>> +#include <sys/wait.h>
>>>>>>>>> +#include <stdio.h>
>>>>>>>>> +#include <stdlib.h>
>>>>>>>>> +#include <stddef.h>
>>>>>>>>> +#include <errno.h>
>>>>>>>>> +#include <sys/uio.h>
>>>>>>>>> +#include <elf.h>
>>>>>>>>> +#include <error.h>
>>>>>>>>> +
>>>>>>>>> +static __attribute__((unused)) pid_t child;
>>>>>>>>> +
>>>>>>>>> +static __attribute__((unused)) void
>> Why are these marked as "static __attribute__((unused))” ?
> 
> It instructs GCC not to produce a warning if the function is unused.

Now that you have removed the x86/ia64 code, you should be able to
remove the static attribute too.

> 
>>>>>>>>> +cleanup (void)
>>>>>>>>> +{
>>>>>>>>> +  if (child > 0)
>>>>>>>>> +    kill (child, SIGKILL);
>>>>>>>>> +  child = 0;
>>>>>>>>> +}
>>>>>>>>> +
>>>>>>>>> +static __attribute__((unused)) void
>> Same as above.
>> 
>>>>>>>>> +handler_fail (int signo)
>>>>>>>>> +{
>>>>>>>>> +  cleanup ();
>>>>>>>>> +  signal (signo, SIG_DFL);
>>>>>>>>> +  raise (signo);
>>>>>>>>> +}
>>>>>>>>> +
>>>>>>>>> +#ifdef __aarch64__
>> Again, as before, you shouldn’t need this check. If the test is only run
>> on AArch64 then it isn’t needed.
> Done.
>> 
>>>>>>>>> +
>>>>>>>>> +#define SET_WATCHPOINT set_watchpoint
>>>>>>>>> +
>>>>>>>>> +/* Macros to extract fields from the hardware debug information word.  */
>>>>>>>>> +#define AARCH64_DEBUG_NUM_SLOTS(x) ((x) & 0xff)
>>>>>>>>> +#define AARCH64_DEBUG_ARCH(x) (((x) >> 8) & 0xff)
>>>>>>>>> +/* Macro for the expected version of the ARMv8-A debug architecture.  */
>>>>>>>>> +#define AARCH64_DEBUG_ARCH_V8 0x6
>>>>>>>>> +#define DR_CONTROL_ENABLED(ctrl)        (((ctrl) & 0x1) == 1)
>>>>>>>>> +#define DR_CONTROL_LENGTH(ctrl)         (((ctrl) >> 5) & 0xff)
>>>>>>>>> +
>>>>>>>>> +static void
>>>>>>>>> +set_watchpoint (pid_t pid, volatile void *addr, unsigned len_mask)
>>>>>>>>> +{
>>>>>>>>> +  struct user_hwdebug_state dreg_state;
>>>>>>>>> +  struct iovec iov;
>>>>>>>>> +  long l;
>>>>>>>>> +
>>>>>>>>> +  assert (len_mask >= 0x01);
>>>>>>>>> +  assert (len_mask <= 0xff);
>>>>>>>>> +
>>>>>>>>> +  iov.iov_base = &dreg_state;
>>>>>>>>> +  iov.iov_len = sizeof (dreg_state);
>>>>>>>>> +  errno = 0;
>>>>>>>>> +  l = ptrace (PTRACE_GETREGSET, pid, NT_ARM_HW_WATCH, &iov);
>>>>>>>>> +  assert (l == 0);
>>>>>>>>> +  assert (AARCH64_DEBUG_ARCH (dreg_state.dbg_info) == AARCH64_DEBUG_ARCH_V8);
>>>>>>>>> +  assert (AARCH64_DEBUG_NUM_SLOTS (dreg_state.dbg_info) >= 1);
>>>>>>>>> +
>>>>>>>>> +  assert (!DR_CONTROL_ENABLED (dreg_state.dbg_regs[0].ctrl));
>>>>>>>>> +  dreg_state.dbg_regs[0].ctrl |= 1;
>>>>>>>>> +  assert ( DR_CONTROL_ENABLED (dreg_state.dbg_regs[0].ctrl));
>>>>>>>>> +
>>>>>>>>> +  assert (DR_CONTROL_LENGTH (dreg_state.dbg_regs[0].ctrl) == 0);
>>>>>>>>> +  dreg_state.dbg_regs[0].ctrl |= len_mask << 5;
>>>>>>>>> +  assert (DR_CONTROL_LENGTH (dreg_state.dbg_regs[0].ctrl) == len_mask);
>>>>>>>>> +
>>>>>>>>> +  dreg_state.dbg_regs[0].ctrl |= 2 << 3; // write
>>>>>>>>> +  dreg_state.dbg_regs[0].ctrl |= 2 << 1; // GDB: ???: enabled at el0
>>>>>>>>> +  //printf("ctrl=0x%x\n",dreg_state.dbg_regs[0].ctrl);
>> Remove the commented out code.
>> 
>>>>>>>>> +  dreg_state.dbg_regs[0].addr = (uintptr_t) addr;
>>>>>>>>> +
>>>>>>>>> +  iov.iov_base = &dreg_state;
>>>>>>>>> +  iov.iov_len = (offsetof (struct user_hwdebug_state, dbg_regs)
>>>>>>>>> +                 + sizeof (dreg_state.dbg_regs[0]));
>>>>>>>>> +  errno = 0;
>>>>>>>>> +  l = ptrace (PTRACE_SETREGSET, pid, NT_ARM_HW_WATCH, &iov);
>>>>>>>>> +  if (errno != 0)
>>>>>>>>> +    error (1, errno, "PTRACE_SETREGSET: NT_ARM_HW_WATCH");
>>>>>>>>> +  assert (l == 0);
>>>>>>>>> +}
>>>>>>>>> +
>>>>>>>>> +#endif
>>>>>>>>> +
>>>>>>>>> +#ifndef SET_WATCHPOINT
>>>>>>>>> +
>>>>>>>>> +int
>>>>>>>>> +main (void)
>>>>>>>>> +{
>>>>>>>>> +  return 77;
>>>>>>>>> +}
>>>>>>>>> +#else
>> Having the executable exit with error on not AArch64 is not useful.
>> Again, this can be cut.
>> 
>> 
>>>>>>>>> +
>>>>>>>>> +static volatile long long check;
>>>>>>>>> +
>>>>>>>>> +int
>>>>>>>>> +main (void)
>>>>>>>>> +{
>>>>>>>>> +  pid_t got_pid;
>>>>>>>>> +  int i, status;
>>>>>>>>> +  long l;
>>>>>>>>> +
>>>>>>>>> +  atexit (cleanup);
>>>>>>>>> +  signal (SIGABRT, handler_fail);
>>>>>>>>> +  signal (SIGINT, handler_fail);
>> I’m not sure on the point of the handler_fail?
>> Would the test be simpler without them?
> Yes, and the function is removed.
>> 
>>>>>>>>> +
>>>>>>>>> +  child = fork ();
>>>>>>>>> +  assert (child >= 0);
>>>>>>>>> +  if (child == 0)
>>>>>>>>> +    {
>>>>>>>>> +      l = ptrace (PTRACE_TRACEME, 0, NULL, NULL);
>>>>>>>>> +      assert (l == 0);
>>>>>>>>> +      i = raise (SIGUSR1);
>>>>>>>>> +      assert (i == 0);
>>>>>>>>> +      check = -1;
>>>>>>>>> +      i = raise (SIGUSR2);
>>>>>>>>> +      /* NOTREACHED */
>>>>>>>>> +      assert (0);
>>>>>>>>> +    }
>>>>>>>>> +
>>>>>>>>> +  got_pid = waitpid (child, &status, 0);
>>>>>>>>> +  assert (got_pid == child);
>>>>>>>>> +  assert (WIFSTOPPED (status));
>>>>>>>>> +  assert (WSTOPSIG (status) == SIGUSR1);
>>>>>>>>> +
>>>>>>>>> +  // PASS:
>>>>>>>>> +  //SET_WATCHPOINT (child, &check, 0xff);
>>>>>>>>> +  // FAIL:
>> Remove the commented out code.
>> 
>>>>>>>>> +  SET_WATCHPOINT (child, &check, 0x02);
>>>>>>>>> +
>>>>>>>>> +  errno = 0;
>>>>>>>>> +  l = ptrace (PTRACE_CONT, child, 0l, 0l);
>>>>>>>>> +  assert_perror (errno);
>>>>>>>>> +  assert (l == 0);
>>>>>>>>> +
>>>>>>>>> +  got_pid = waitpid (child, &status, 0);
>>>>>>>>> +  assert (got_pid == child);
>>>>>>>>> +  assert (WIFSTOPPED (status));
>>>>>>>>> +  if (WSTOPSIG (status) == SIGUSR2)
>>>>>>>>> +    {
>>>>>>>>> +      /* We missed the watchpoint - unsupported by hardware? */
>>>>>>>>> +      cleanup ();
>>>>>>>>> +      return 2;
>>>>>>>>> +    }
>>>>>>>>> +  assert (WSTOPSIG (status) == SIGTRAP);
>>>>>>>>> +
>> It’s not immediately clear to me what is going on above.
>> A few comments are probably needed to make it clear:
>> *Add a watchpoint to check.
>> *Restart the child. It will write to check.
>> *Check child has stopped on the watchpoint.

Could you add some comments to the C file please.


>> 
> <0001-Adding-a-test-case.patch>

Attached patch looks ok.
Happy for you to push if you make the two changes above.


Thanks,
Alan.