qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH 0/5] linux-user: Support extended clone(CLONE_VM)


From: Alex Bennée
Subject: Re: [PATCH 0/5] linux-user: Support extended clone(CLONE_VM)
Date: Sat, 13 Jun 2020 12:16:49 +0100
User-agent: mu4e 1.5.2; emacs 28.0.50

Josh Kunz <jkz@google.com> writes:

> This patch series implements extended support for the `clone` system
> call. As best I can tell, any option combination including `CLONE_VM`
> should be supported with the addition of this patch series. The
> implementation is described in greater detail in the patches themselves.
>
> Testing:
>
>   * All targets built on x86_64.
>   * `make check` and `make check-tcg` are passing. Additional tests have
>     been added to `linux-test.c` to validate clone behavior.
>
> Caveats:
>
>   * This series touches, but does not fix, several bits of code that are
>     racey (namely the sigact table and the fd trans table).
>   * `exit_group` does not perform the appropriate cleanup for non-thread
>     children created with `CLONE_VM`. CPUs for such children are never
>     cleaned up. The correct implementation of exit-group is non-trivial
>     (since it also needs to track/handle cleanup for threads in the
>     clone'd child process). Also, I don't fully understand the
>     interaction between QOM<->linux-user.

When the QOM object gets unrefed for the final time it should cause a
bunch of clean-up in the common vCPU management code where things like
plugin cleanup are done.

This was recently touched in 1f81ce90e31ef338ee53a0cea02344237bc470cc
where I removed linux-user messing around with the active cpu list and
left it to the core code to deal with. Previously it wasn't being
properly unrealized.

>     My naive implementation based
>     on the current implementation `exit(2)` was regularly crashing. If
>     maintainers have suggestions for better ways to handle exit_group,
>     they would be greatly appreciated. 
>   * execve does not clean up the CPUs of clone'd children, for the same
>     reasons as `exit_group`.
>
> Josh Kunz (5):
>   linux-user: Refactor do_fork to use new `qemu_clone`
>   linux-user: Make fd_trans task-specific.
>   linux-user: Make sigact_table part of the task state.
>   linux-user: Support CLONE_VM and extended clone options
>   linux-user: Add PDEATHSIG test for clone process hierarchy.
>
>  linux-user/Makefile.objs            |   2 +-
>  linux-user/clone.c                  | 565 ++++++++++++++++++++++++++++
>  linux-user/clone.h                  |  27 ++
>  linux-user/fd-trans-tbl.c           |  13 +
>  linux-user/fd-trans-type.h          |  17 +
>  linux-user/fd-trans.c               |   3 -
>  linux-user/fd-trans.h               |  75 ++--
>  linux-user/main.c                   |   1 +
>  linux-user/qemu.h                   |  49 +++
>  linux-user/signal.c                 |  84 ++++-
>  linux-user/syscall.c                | 452 ++++++++++++----------
>  tests/tcg/multiarch/Makefile.target |   3 +
>  tests/tcg/multiarch/linux-test.c    | 227 ++++++++++-
>  13 files changed, 1264 insertions(+), 254 deletions(-)
>  create mode 100644 linux-user/clone.c
>  create mode 100644 linux-user/clone.h
>  create mode 100644 linux-user/fd-trans-tbl.c
>  create mode 100644 linux-user/fd-trans-type.h


-- 
Alex Bennée



reply via email to

[Prev in Thread] Current Thread [Next in Thread]