Re: [musl] Re: add loongarch64 port

[Rich Felker <dalias@libc.org>]

On Wed, Apr 20, 2022 at 03:33:25PM +0200, Christian Brauner wrote:
> On Wed, Apr 20, 2022 at 05:09:08PM +0800, 王洪亮 wrote:
> > 
> > 在 2022/4/14 下午5:36, Christian Brauner 写道:
> > > On Wed, Apr 13, 2022 at 10:09:31AM -0400, Rich Felker wrote:
> > > > On Wed, Apr 13, 2022 at 03:25:05PM +0200, Arnd Bergmann wrote:
> > > > > On Wed, Apr 13, 2022 at 11:04 AM Christian Brauner <brauner@kernel.org> wrote:
> > > > > > On Wed, Apr 13, 2022 at 10:26:06AM +0200, Arnd Bergmann wrote:
> > > > > > > The normal rule is that we don't define obsolete system calls in new
> > > > > > > architectures when an improved variant has been added, e.g. oldoldstat,
> > > > > > > oldstat, stat, newstat and stat64 have all been replaced by statx over
> > > > > > > the decades. I was expecting the same to be true for clone(), but if
> > > > > > > clone3() is not meant as a replacement, we can keep both around.
> > > > > > No, I agree with you on this and would like to only implement clone3()
> > > > > > on new architectures.
> > > > > > 
> > > > > > What I'm asking is whether removing the size == 0 check is enough to
> > > > > > unblock the missing behavior and whether you'd be on board with removing
> > > > > > the check?
> > > > > I think that's ok here, since we'd only rely on this for loongarch64 at the
> > > > > moment. It would probably need to be documented in the man page
> > > > > as a special case though.
> > > > I'm okay with removing the check for size==0 (so size==0 will be
> > > > allowed) and dropping __NR_clone on new archs, as long as it's noted
> > > > in comments/documentation that size==0 is explicitly allowed so nobody
> > > > breaks this in the future.
> > > Ok, I'll try to have a patch ready early next week since I'm currently
> > > out sick.
> > 
> > Hi,
> > 
> > I'm implementing  __NR_clone3 syscall within __clone().
> > 
> > I have another problem:CLONE_DETACHED
> > 
> > in musl,internal call __clone()(such as __pthread_create()),the input
> > parameter flags
> > 
> > has been set CLONE_DETACHED ,in kernel,there is a check in
> > clone3_args_valid(),
> > 
> > if the condition met,return false.
> > 
> > How to deal with this problem?
> 
> CLONE_DETACHED is meaningles since Linux on 2.6.2. There really should
> be <=2.6.1 living kernel anywhere where CLONE_DETACHED does anything.
> I've documented that in detail under [1] as:
> 
>         CLONE_DETACHED (historical)
>               For a while (during the Linux 2.5 development series)
>               there was a CLONE_DETACHED flag, which caused the parent
>               not to receive a signal when the child terminated.
>               Ultimately, the effect of this flag was subsumed under the
>               CLONE_THREAD flag and by the time Linux 2.6.0 was
>               released, this flag had no effect.  Starting in Linux
>               2.6.2, the need to give this flag together with
>               CLONE_THREAD disappeared.
> 
>               This flag is still defined, but it is usually ignored when
>               calling clone().  However, see the description of
>               CLONE_PIDFD for some exceptions.
> 
> [1]: https://man7.org/linux/man-pages/man2/clone.2.html
> 
> Would it be possible to drop this flag from musl's pthread_create()
> implementation? (Iirc, glibc dropped CLONE_DETACHED in 2004.)

I think __clone should just mask it on newer archs. We support Linux
2.6.0 and if lack of CLONE_DETACHED causes bogus signals on 2.6.0 we
should keep it. If it can be established that this doesn't happen and
that CLONE_DETACHED just affected non-thread clones, we can probably
safely drop it.

Rich