[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <4665970A.6080501@zytor.com>
Date: Tue, 05 Jun 2007 10:02:02 -0700
From: "H. Peter Anvin" <hpa@...or.com>
To: Andi Kleen <ak@...e.de>
CC: Andrew Morton <akpm@...ux-foundation.org>,
Rusty Russell <rusty@...tcorp.com.au>,
Matt Mackall <mpm@...enic.com>, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] lguest-fix-divide-error-implement-sched_clock
Andi Kleen wrote:
>
> I don't think it's a good idea for the TSC. There are various
> setups where it is unreliable and also often simulators don't
> implement it correctly. And it's always a valuable workaround
> to be able to turn it off.
>
> Except possibly for the FPU only features used by the gcc output
> should be tested this way. For everything else it is better to
> test at runtime.
>
> That is x86-64 makes some more assumptions. But even it
> doesn't assume TSC.
>
> I added the mechanism to statically evaluate mostly to share cpufeatures.h
> between 32bit and 64bit at some point -- but didn't quite finish that work
> before the last merge.
>
Note that tsc_setup in i386 at least has:
static int __init tsc_setup(char *str)
{
printk(KERN_WARNING "notsc: Kernel compiled with CONFIG_X86_TSC,"
"cannot disable TSC.\n");
return 1;
}
If TSC isn't actually a compile-time feature then we should allow it to
be disabled on the command line!
This is what I have for i386:
#ifndef CONFIG_MATH_EMULATION
# define NEED_FPU (1<<(X86_FEATURE_FPU & 31))
#else
# define NEED_FPU 0
#endif
#ifdef CONFIG_X86_TSC
# define NEED_TSC (1<<(X86_FEATURE_TSC & 31))
#else
# define NEED_TSC 0
#endif
#ifdef CONFIG_X86_PAE
# define NEED_PAE (1<<(X86_FEATURE_PAE & 31))
#else
# define NEED_PAE 0
#endif
#ifdef CONFIG_X86_CMOV
# define NEED_CMOV (1<<(X86_FEATURE_CMOV & 31))
#else
# define NEED_CMOV 0
#endif
#ifdef CONFIG_X86_CMPXCHG64
# define NEED_CX8 (1<<(X86_FEATURE_CX8 & 31))
#else
# define NEED_CX8 0
#endif
#define REQUIRED_MASK0 (NEED_FPU|NEED_TSC|NEED_PAE|NEED_CMOV|NEED_CX8)
#ifdef CONFIG_X86_USE_3DNOW
# define NEED_3DNOW (1<<(X86_FEATURE_3DNOW & 31))
#else
# define NEED_3DNOW 0
#endif
#define REQUIRED_MASK1 (NEED_3DNOW)
And for x86-64:
/* x86-64 baseline features */
#define NEED_FPU (1<<(X86_FEATURE_FPU & 31))
#define NEED_PSE (1<<(X86_FEATURE_PSE & 31))
#define NEED_TSC (1<<(X86_FEATURE_TSC & 31))
#define NEED_MSR (1<<(X86_FEATURE_MSR & 31))
#define NEED_PAE (1<<(X86_FEATURE_PAE & 31))
#define NEED_CX8 (1<<(X86_FEATURE_CX8 & 31))
#define NEED_PGE (1<<(X86_FEATURE_PGE & 31))
#define NEED_FXSR (1<<(X86_FEATURE_FXSR & 31))
#define NEED_CMOV (1<<(X86_FEATURE_CMOV & 31))
#define NEED_XMM (1<<(X86_FEATURE_XMM & 31))
#define NEED_XMM2 (1<<(X86_FEATURE_XMM2 & 31))
#define REQUIRED_MASK0 (NEED_FPU|NEED_PSE|NEED_TSC|NEED_MSR|NEED_PAE|\
NEED_CX8|NEED_PGE|NEED_FXSR|NEED_CMOV|\
NEED_XMM|NEED_XMM2)
#define SSE_MASK (NEED_XMM|NEED_XMM2)
/* x86-64 baseline features */
#define NEED_LM (1<<(X86_FEATURE_LM & 31))
#ifdef CONFIG_X86_USE_3DNOW
# define NEED_3DNOW (1<<(X86_FEATURE_3DNOW & 31))
#else
# define NEED_3DNOW 0
#endif
#define REQUIRED_MASK1 (NEED_LM|NEED_3DNOW)
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists