Date: Mon, 22 Aug 2016 02:49:07 +1000 (EST) From: Bruce Evans <brde@optusnet.com.au> To: Slawa Olhovchenkov <slw@zxy.spb.ru> Cc: Bruce Evans <brde@optusnet.com.au>, Ed Schouten <ed@freebsd.org>, src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: Re: svn commit: r304555 - head/sys/compat/cloudabi Message-ID: <20160822022832.Q3214@besplex.bde.org> In-Reply-To: <20160821135826.GB8192@zxy.spb.ru> References: <201608210741.u7L7fBnN075023@repo.freebsd.org> <20160821105207.GS22212@zxy.spb.ru> <20160821210751.J2219@besplex.bde.org> <20160821120016.GZ8192@zxy.spb.ru> <20160821223255.K2478@besplex.bde.org> <20160821131447.GA8192@zxy.spb.ru> <20160821232721.G2639@besplex.bde.org> <20160821135826.GB8192@zxy.spb.ru>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, 21 Aug 2016, Slawa Olhovchenkov wrote:
> On Sun, Aug 21, 2016 at 11:39:02PM +1000, Bruce Evans wrote:
>
>> On Sun, 21 Aug 2016, Slawa Olhovchenkov wrote:
>>> I am remeber about platforms with missaligment trap when
>>> accessing int16 by odd address. Now platforms like this do not exist
>>> anymore?
>>
>> i386 still exists, and it supports trapping on misalignement for at least
>> CPL 3 (not kernel CPL 0). IIRC, amd64 drops support for this.
>
> Someone enable and support this? I am don't see.
> May be PPC trap on this?
> Alpha trap on this, but support of Alpha is droped.
It is a 1-line change in asm (or a little more in C with #includes) to
enable the trap:
%%%
#include <sys/types.h>
#include <machine/cpufunc.h>
#include <machine/psl.h>
int
main(void)
{
char ch[5];
write_eflags(read_eflags() | PSL_AC);
*(int *)&ch[0] = 0;
*(int *)&ch[1] = 1;
/* NOTREACHED */
}
%%%
This works on amd64 too after s/eflags/rflags.
It is a trillion-line change to fix the compilers and applications to not
do misaligned accesses :-). I only tried to use this ~25 years ago. Then
the most obvious compiler bug was generating 32-bit acccesses to assign
large but misaligned structs. If the compiler just generated calls to
memcpy(), that might work, but in practice libraries also assume alignment.
>>>> There are also endianness problems. The old version was even more broken
>>>> on big endian systems. The current version needs some magic to reverse
>>>> the memcpy() of the bits. We already depend on this for some 64-bit
>>>> syscalls like lseek().
>>>
>>> Can you explain some more?
>>> This is not transfer over network and don't read from external media.
>>> Where is problem?
>>
>> It is similar to a network transfer. It needs a protocol to pass values
>> to applications. Type puns are fragile even within a single compilation
>> unit.
>
> Application ad kernel run with same byte order, not?
The application can do anything it wants, but has to translate if it uses
the kernel or a library written in another language.
Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20160822022832.Q3214>