]> git.ipfire.org Git - thirdparty/binutils-gdb.git/commit - gdb/ChangeLog
detach and breakpoint removal
authorPedro Alves <pedro@palves.net>
Sun, 13 Dec 2020 01:35:05 +0000 (01:35 +0000)
committerPedro Alves <pedro@palves.net>
Wed, 3 Feb 2021 01:15:12 +0000 (01:15 +0000)
commite87f0fe82375ffe508da44172fcfe04e258ddc9c
tree68ef50776ce802fc617c839d03b33ddab7d89bb8
parent8ff531399b7e5e391a664d089520e6d17d006ea4
detach and breakpoint removal

A following patch will add a testcase that has a number of threads
constantly stepping over a breakpoint, and then has GDB detach the
process.  That testcase sometimes fails with the inferior crashing
with SIGTRAP after the detach because of the bug fixed by this patch,
when tested with the native target.

The problem is that target_detach removes breakpoints from the target
immediately, and that does not work with the native GNU/Linux target
(and probably no other native target) currently.  The test wouldn't
fail with this issue when testing against gdbserver, because gdbserver
does allow accessing memory while the current thread is running, by
transparently pausing all threads temporarily, without GDB noticing.
Implementing that in gdbserver was a lot of work, so I'm not looking
forward right now to do the same in the native target.  Instead, I
came up with a simpler solution -- push the breakpoints removal down
to the targets.  The Linux target conveniently already pauses all
threads before detaching them, since PTRACE_DETACH only works with
stopped threads, so we move removing breakpoints to after that.  Only
the remote and GNU/Linux targets support support async execution, so
no other target should really need this.

gdb/ChangeLog:

* linux-nat.c (linux_nat_target::detach): Remove breakpoints
here...
* remote.c (remote_target::remote_detach_1): ... and here ...
* target.c (target_detach): ... instead of here.
* target.h (target_ops::detach): Add comment.
gdb/ChangeLog
gdb/linux-nat.c
gdb/remote.c
gdb/target.c
gdb/target.h