[Bug tdep/27369] ARC: Stepping over atomic instruction sequences loops infinitely

["cvs-commit at gcc dot gnu.org" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=27369

--- Comment #1 from cvs-commit at gcc dot gnu.org <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Shahab Vahedi <shahab@sourceware.org>:

https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=9b3e4b5d74a8170a8f9f224c36c651661fc26954

commit 9b3e4b5d74a8170a8f9f224c36c651661fc26954
Author: Shahab Vahedi <shahab@synopsys.com>
Date:   Thu Oct 31 17:33:08 2019 +0100

    gdb: Do not interrupt atomic sequences for ARC

    When stepping over thread-lock related codes (in uClibc), the inferior
process
    gets stuck and never manages to enter the critical section:

    ------8<-------
     1 size_t fwrite(const void * __restrict ptr, size_t size,
     2               size_t nmemb, register FILE * __restrict stream)
     3 {
     4     size_t retval;
     5     __STDIO_AUTO_THREADLOCK_VAR;
     6
     7 >   __STDIO_AUTO_THREADLOCK(stream);
     8
     9     retval = fwrite_unlocked(ptr, size, nmemb, stream);
    10
    11     __STDIO_AUTO_THREADUNLOCK(stream);
    12
    13     return retval;
    14 }
    ------>8-------

    Here, we are at line 7.  Using the "next" command leads no where.
    However, setting a breakpoint on line 9 and issuing "continue" works.

    Looking at the assembly instructions reveals that we're dealing with the
    critical section entry code [1] that should never be interrupted, in this
    case by the debugger's implicit breakpoints:

    ------8<-------
      ...
    1 add_s   r0,r13,0x38
    2 mov_s   r3,1
    3 llock   r2,[r0]        <-.
    4 brne.nt r2,0,14     --.  |
    5 scond   r3,[r0]       |  |
    6 bne     -10         --|--'
    7 brne_s  r2,0,84     <-'
      ...
    ------>8-------

    Lines 3 until 5 (inclusive) are supposed to be executed atomically.
    Therefore, GDB should never (implicitly) insert a breakpoint on lines
    4 and 5, else the program will try to acquire the lock again by jumping
    back to line 3 and gets stuck in an infinite loop.

    The solution is to make GDB aware of these patterns so it inserts
    breakpoints after the sequence -- line 6 in this example.

    [1]
   
https://cgit.uclibc-ng.org/cgi/cgit/uclibc-ng.git/tree/libc/sysdeps/linux/arc/bits/atomic.h#n46
    ------8<-------
      ({                                                                    \
            __typeof(oldval) prev;                                          \
                                                                            \
            __asm__ __volatile__(                                           \
            "1:     llock   %0, [%1]        \n"                             \
            "       brne    %0, %2, 2f      \n"                             \
            "       scond   %3, [%1]        \n"                             \
            "       bnz     1b              \n"                             \
            "2:                             \n"                             \
            : "=&r"(prev)                                                   \
            : "r"(mem), "ir"(oldval),                                       \
              "r"(newval) /* can't be "ir". scond can't take limm for "b" */\
            : "cc", "memory");                                              \
                                                                            \
            prev;                                                           \
      })
    ------>8-------
    "llock" (Load Locked) loads the 32-bit word pointed by the source
    operand.  If the load is completed without any interruption or
    exception, the physical address is remembered, in Lock Physical Address
    (LPA), and the Lock Flag (LF) is set to 1.  LF is a non-architecturally
    visible flag and is cleared whenever an interrupt or exception takes
    place.  LF is also cleared (atomically) whenever another process writes
    to the LPA.

    "scond" (Store Conditional) will write to the destination address if
    and only if the LF is set to 1.  When finished, with or without a write,
    it atomically copies the LF value to ZF (Zero Flag).

    These two instructions together provide the mechanism for entering a
    critical section.  The code snippet above comes from uClibc:
    -----------------------

    v3 (after Tom's remarks[2]):
     handle_atomic_sequence()
      - no need to initialize the std::vector with "{}"
      - fix typo in comments: "conditial" -> "conditional"
      - add braces to the body of "if" condition because of the comment line
     arc_linux_software_single_step()
      - make the performance slightly more efficient by moving a few
        variables after the likely "return" point.

    v2 (after Simon's remarks[3]):
    - handle_atomic_sequence() gets a copy of an instruction instead of
      a reference.
    - handle_atomic_sequence() asserts if the given instruction is an llock.

    [2]
    https://sourceware.org/pipermail/gdb-patches/2021-February/175805.html

    [3]
    https://sourceware.org/pipermail/gdb-patches/2021-January/175487.html

    gdb/ChangeLog:

            PR tdep/27369
            * arc-linux-tdep.c (handle_atomic_sequence): New.
            (arc_linux_software_single_step): Call handle_atomic_sequence().

-- 
You are receiving this mail because:
You are on the CC list for the bug.