PERFORCE change 109706 for review

[John Baldwin]

On Monday 13 November 2006 18:12, Jung-uk Kim wrote:
> On Monday 13 November 2006 04:06 pm, John Baldwin wrote:
> > On Friday 10 November 2006 18:00, Jung-uk Kim wrote:
> > > http://perforce.freebsd.org/chv.cgi?CH=109706
> > >
> > > Change 109706 by jkim at jkim_hammer on 2006/11/10 22:59:53
> > >
> > > 	Add (ugly) 32-bit msgsnd/msgrcv support for amd64.
> > > 	msgp points to msghdr and msghdr contains msg_type, which is
> > > long. When we copyin/copyout, we copy the correct msg_type and
> > > advance msgp by msg_type size.
> > >
> > > 	This fixes LTP test cases msgget01, msgrcv01, msgrcv02,
> > > msgrcv04, and msgsnd03.
> > >
> > > 	Note: There is only one test case blocked in msgwait state after
> > > this change, which seems to be arch-independent issue.
> > >
> > > Affected files ...
> > >
> > > ..
> > > //depot/projects/linuxolator/src/sys/compat/linux/linux_ipc.c#3
> > > edit .. //depot/projects/linuxolator/src/sys/kern/sysv_msg.c#5
> > > edit
> >
> > Why not add kern_msgfoo() functions and do the copyout/copyin in
> > other places instead?
> 
> Unfortunately there are two different types of copyin/copyout's in 
> each msgsnd/msgrcv, i.e., message type and actual message.  Only the 
> first one (i.e., message type) is affected.  It is quite hard to 
> separate them because of that reason.  I thought about using one 
> copyin/copyout instead of two.  I thought about passing function 
> pointer to a copy function.  I even thought about making 
> msgsnd32/msgrcv32 syscalls.  But these are overkill for this, I 
> think.  As I noted in the commit log, I don't like this but I think 
> this is the simplest fix without breaking existing APIs.

Use uio_seg's for each item.  Then the wrappers only have to explicitly copy 
in the one that is different from normal and use UIO_SYSSPACE for that one.

-- 
John Baldwin