[Prev][Next][Index][Thread]

[k-abe@cs: oskit/x86/proc_reg.h]




------- Forwarded Messages
Date:    Thu, 21 Dec 2000 17:40:12 -0700
From:    k-abe@cs
To:      oskit@fast
Subject: oskit/x86/proc_reg.h

In oskit/x86/proc_reg.h, CR4_PGE is defined as 0x20.
I believe it should be 0x80. 
(0x20 is `physical address extension' bit that enables 36bit physical
address space, should be defined as CR4_PAE).
However, no OSKit source uses this symbol.

/*
 * CR4
 */
#define CR4_VME	0x00000001		/* enable virtual intrs in v86 mode */
#define CR4_PVI	0x00000002		/* enable virtual intrs in pmode */
#define CR4_TSD	0x00000004		/* disable RDTSC in user mode */
#define CR4_DE	0x00000008		/* debug extensions (I/O breakpoints) *
/
#define CR4_PSE	0x00000010		/* page size extensions */
#define CR4_PGE	0x00000020		/* page global extensions */
#define CR4_MCE	0x00000040		/* machine check exception */
#define CR4_PCE	0x00000100		/* enable read perf counter instr */

- -- 
Kota Abe
Dept. of Computer Science, Univ. of Utah.
Media Center, Osaka City Univ.

-----------
Date: Thu, 21 Dec 2000 23:51:18 -0500 (EST)
From: Roland McGrath <roland@frob.com>
To: k-abe@cs
Cc: oskit@fast
Subject: Re: oskit/x86/proc_reg.h
In-Reply-To: k-abe@cs.utah.edu's message of  Thu, 21 December 2000 17:40:12 -0700 <814rzxp9ub.wl@localhost.cs.utah.edu>

> In oskit/x86/proc_reg.h, CR4_PGE is defined as 0x20.
> I believe it should be 0x80. 
> (0x20 is `physical address extension' bit that enables 36bit physical
> address space, should be defined as CR4_PAE).
> However, no OSKit source uses this symbol.

Auuughghgh!!  But I used it in oskit-mach, and several people spent some
time trying to figure out why it caused the kernel to crash.  I read the
specs backwards and forewards trying to figure out what I was doing wrong,
and compared to the usage in BSD and Linux and gave up.  It never occurred
to me that the constants might be wrong!  

I put this into Fluke at one point too, though I don't recall anyone ever
seeing any problems with it.  I do recall doing some benchmarks and
wondering why it never seemed to help as the manuals said it should.

------- End of Forwarded Messages