Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 15 Jan 2001 13:17:21 -0800 (PST)
From:      John Baldwin <jhb@FreeBSD.org>
To:        "Justin T. Gibbs" <gibbs@scsiguy.com>
Cc:        Peter Wemm <peter@netplex.com.au>, Poul-Henning Kamp <phk@critter.freebsd.dk>, Wilko Bulte <wkb@freebie.demon.nl>, cvs-committers@FreeBSD.org, cvs-all@FreeBSD.org, Bruce Evans <bde@zeta.org.au>
Subject:   Re: cvs commit: src/sys/i386/conf GENERIC
Message-ID:  <XFMail.010115131721.jhb@FreeBSD.org>
In-Reply-To: <200101152012.f0FKCns56756@aslan.scsiguy.com>

next in thread | previous in thread | raw e-mail | index | archive | help

On 15-Jan-01 Justin T. Gibbs wrote:
>>So are you ready to write the code in trap() to handle an illegal instruction
>>fault in userland that decodes and executes all variants of cmpxchg?  The new
>>threading code in libc will be using atomic_cmpset() from userland, which is
>>going to be the main hurdle to get over.
> 
> This is the wrong way to handle it.  Have atomic_cmpset() perform a fixup
> of the calling code on first entry and the result will be code as optimized
> as possible for the processor type the code is running on.  If the user
> decides to write their own code that uses cmpxchg, they get what they
> deserve, but the primitives should not require a *fault* to work correctly.

*sigh*

Go look at the 386 version of atomic_cmpset in atomic.h:

#if defined(I386_CPU)
static __inline int
atomic_cmpset_int(volatile u_int *dst, u_int exp, u_int src)
{
        int res = exp;

        __asm __volatile(
        "       pushfl ;                "
        "       cli ;                   "
        "       cmpl    %1,%3 ;         "
        "       jne     1f ;            "
        "       movl    %2,%3 ;         "
        "1:                             "
        "       sete    %%al;           "
        "       movzbl  %%al,%0 ;       "
        "       popfl ;                 "
        "# atomic_cmpset_int"
        : "=a" (res)                    /* 0 (result) */
        : "0" (exp),                    /* 1 */
          "r" (src),                    /* 2 */
          "m" (*(dst))                  /* 3 */
        : "memory");

        return (res);
}

See those 'cli' and 'popfl' instrucitons?  Those are _privileged_.  Userland
can't disable/enable interrupts, so we have to trap into the kernel to do this
no matter what.  If you want to patch the code to do a syscall instead of a
cmpxchg instruction, fine.  However, emulating atomic_cmpset in userland on a
386 requires a trap into the kernel.  Please assume for at least 1 minute that
the SMPng guys are not complete bumbling idiots and that we may have actually
thought about this for at least 5 minutes.

-- 

John Baldwin <jhb@FreeBSD.org> -- http://www.FreeBSD.org/~jhb/
PGP Key: http://www.baldwin.cx/~john/pgpkey.asc
"Power Users Use the Power to Serve!"  -  http://www.FreeBSD.org/


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe cvs-all" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?XFMail.010115131721.jhb>