Date: Mon, 31 Jan 2005 17:55:00 -0500 From: John Baldwin <jhb@FreeBSD.org> To: freebsd-amd64@FreeBSD.org, Joseph Koshy <joseph.koshy@gmail.com> Cc: Rob Watt <rob@hudson-trading.com> Subject: Re: 32-bit binary compatibility on 5.3-amd64 Message-ID: <200501311755.00391.jhb@FreeBSD.org> In-Reply-To: <84dead72050128204479414c88@mail.gmail.com> References: <20050127100818.M46092@daemon.mistermishap.net> <20050128105930.V54223@daemon.mistermishap.net> <84dead72050128204479414c88@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Friday 28 January 2005 11:44 pm, Joseph Koshy wrote:
> rw> a ktrace is attached for this sample code:
>
> Here is the culprit:
>
> src/lib/libc/i386/gen/_set_tp.c:
>
> sel = i386_set_ldt(LDT_AUTO_ALLOC, &ldt, 1);
> __asm __volatile("movl %0,%%gs" : : "rm" ((sel << 3) | 7));
>
> i386_set_ldt() will return -1 on the AMD64 since this
> function is not supported there. The lack of error checking
> means that -1 will be attempted to be loaded into %gs by the
> movl instruction, leading to a processor protection fault and
> a process SIGSEGV.
>
> Compiling your sample program with '-static' under a 4.X
> environment yields an executable that runs on the AMD64.
>
> On 5.X both statically linked and dynamically linked
> executables appear to call _init_tls() and _set_tp()
> and are consequently not usable in IA32 emulation
> mode.
>
> Clearly our IA32 emulation needs work. I've no idea
> why we don't support a sysarch(I386_{GET,SET}_LDT) operation
> for IA32 executables.
Because amd64 doesn't have LDTs. Peter has hacks in place in the dynamic
32-bit libraries that amd64 uses to use the different backing methods for
implementing TLS on amd64, but that stuff wasn't in 5.3. It might be in
5.3-STABLE.
--
John Baldwin <jhb@FreeBSD.org> <>< http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve" = http://www.FreeBSD.org
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200501311755.00391.jhb>