[PATCH] arm64: ptrace: hw_break_set take into account hardware breakpoints number

[Christopher Covington]

Hi Victor,

On 09/29/2014 01:49 PM, Victor Kamensky wrote:
> Hi Will,
> 
> On 29 September 2014 03:16, Will Deacon <will.deacon at arm.com> wrote:
>> Hi Victor,
>>
>> On Mon, Sep 29, 2014 at 09:04:02AM +0100, Victor Kamensky wrote:
>>> hw_break_set function that performs ptrace_regset for hardware
>>> breakpoints and watchpoints needs to take into account actual
>>> number of hardware breakpoints and watchpoints available in CPU.
>>>
>>> Current code iterates over all 16 entries of 'struct user_hwdebug_state'
>>> and tries to reserve hardware breakpoint for each index, which fails
>>> if CPU supports less than 16 hardware breakpoints. One manifestation of
>>> the issue is that gdb fails to debug multithreaded user land application
>>> and exits with 'Unexpected error setting hardware debug registers'
>>> error - ptrace system call for hardware breakpoints regset fails with
>>> ENOSPC.
>>
>> When does this happen? hw_break_set is driven by userspace, so if GDB is
>> asking for more registers than we actually have, then this is a GDB bug and
>> the kernel is doing the right thing.
>>
>> Have you reproduced this with the latest version of GDB?
> 
> Thank you for the reply. Just checked latest version of GDB, indeed it
> does not have my original issue. It turns out my gdb is based on some
> old linaro gdb version that unconditionally passes sizeof(user_hwdebug_state)
> as iov.len to PTRACE_SETREGSET with note type NT_ARM_HW_BREAK.
> Latest gdb adjusts iov.len based on available number of hardware breakpoints.
> I should have checked latest gdb before posting this. Naturally previously
> suggested patch is withdrawn.
> 
> Question is there any place where variable payload nature of
> PTRACE_SETREGSET with note type NT_ARM_HW_BREAK and
> NT_ARM_HW_WATCH documented? I've tried kernel Documentation
> directory, aarch64 abi document, http://infocenter.arm.com search, none
> of them mention NT_ARM_HW_BREAK. Just curious is there any place
> for that sort of information to look in the future. If answer that it is
> documented by existing kernel/gdb code :), please feel free to ignore
> this question.

While it doesn't address those specific note types (yet):

http://man7.org/linux/man-pages/man2/ptrace.2.html

       PTRACE_GETREGSET (since Linux 2.6.34)
              Read the tracee's registers.  addr specifies, in an
              architecture-dependent way, the type of registers to be read.
              NT_PRSTATUS (with numerical value 1) usually results in
              reading of general-purpose registers.  If the CPU has, for
              example, floating-point and/or vector registers, they can be
              retrieved by setting addr to the corresponding NT_foo
              constant.  data points to a struct iovec, which describes the
              destination buffer's location and length.  On return, the
              kernel modifies iov.len to indicate the actual number of bytes
              returned.

       PTRACE_SETREGSET (since Linux 2.6.34)
              Modify the tracee's registers.  The meaning of addr and data
              is analogous to PTRACE_GETREGSET.

Perhaps this could be expanded upon.

https://www.kernel.org/doc/man-pages/patches.html

Christopher

-- 
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by the Linux Foundation.