2003-05-24 17:24:09

by Christoph Hellwig

[permalink] [raw]
Subject: [PATCH] make vt_ioctl ix86isms explicit

sys_ioperm is only implemented on x86 (i386/x86_64). Make the
ifdefs in vt_ioctl.c more explicit so the other architectures can
get rid of their stubs in favour of just using sys_ni_syscall in
the syscall table. (Personally I still wonder why they added it
at all but that's another question..)


--- 1.23/drivers/char/vt_ioctl.c Mon May 12 16:12:47 2003
+++ edited/drivers/char/vt_ioctl.c Fri May 23 21:27:46 2003
@@ -59,7 +59,7 @@
*/
unsigned char keyboard_type = KB_101;

-#if !defined(__alpha__) && !defined(__ia64__) && !defined(__mips__) && !defined(__arm__) && !defined(__sh__)
+#ifdef CONFIG_X86
asmlinkage long sys_ioperm(unsigned long from, unsigned long num, int on);
#endif

@@ -424,11 +424,13 @@
ucval = keyboard_type;
goto setchar;

-#if !defined(__alpha__) && !defined(__ia64__) && !defined(__mips__) && !defined(__arm__) && !defined(__sh__)
/*
* These cannot be implemented on any machine that implements
- * ioperm() in user level (such as Alpha PCs).
+ * ioperm() in user level (such as Alpha PCs) or not at all.
+ *
+ * XXX: you should never use these, just call ioperm directly..
*/
+#ifdef CONFIG_X86
case KDADDIO:
case KDDELIO:
/*


2003-05-24 17:42:44

by Linus Torvalds

[permalink] [raw]
Subject: Re: [PATCH] make vt_ioctl ix86isms explicit


On Sat, 24 May 2003, Christoph Hellwig wrote:
>
> sys_ioperm is only implemented on x86 (i386/x86_64). Make the
> ifdefs in vt_ioctl.c more explicit so the other architectures can
> get rid of their stubs in favour of just using sys_ni_syscall in
> the syscall table. (Personally I still wonder why they added it
> at all but that's another question..)

They were added because this was how the X server got IO permissions a
million years ago. It comes from some random old UNIX/386 thing, it
predates Linux itself as far as I know.

I'm fairly certain that X itself no longer uses it at all, but there may
be other programs that still do (unlikely). Your patch looks sane,
although it might be equally sane to just remove the code altogether.

Linus

2003-05-24 17:50:12

by Christoph Hellwig

[permalink] [raw]
Subject: Re: [PATCH] make vt_ioctl ix86isms explicit

On Sat, May 24, 2003 at 10:55:41AM -0700, Linus Torvalds wrote:
> They were added because this was how the X server got IO permissions a
> million years ago. It comes from some random old UNIX/386 thing, it
> predates Linux itself as far as I know.
>
> I'm fairly certain that X itself no longer uses it at all, but there may
> be other programs that still do (unlikely). Your patch looks sane,
> although it might be equally sane to just remove the code altogether.

What about this one instead? If no one sees this messages we can just
rip it out in 2.7.<early>


--- 1.23/drivers/char/vt_ioctl.c Mon May 12 16:12:47 2003
+++ edited/drivers/char/vt_ioctl.c Fri May 23 22:13:19 2003
@@ -59,7 +59,7 @@
*/
unsigned char keyboard_type = KB_101;

-#if !defined(__alpha__) && !defined(__ia64__) && !defined(__mips__) && !defined(__arm__) && !defined(__sh__)
+#ifdef CONFIG_X86
asmlinkage long sys_ioperm(unsigned long from, unsigned long num, int on);
#endif

@@ -424,23 +424,29 @@
ucval = keyboard_type;
goto setchar;

-#if !defined(__alpha__) && !defined(__ia64__) && !defined(__mips__) && !defined(__arm__) && !defined(__sh__)
/*
* These cannot be implemented on any machine that implements
- * ioperm() in user level (such as Alpha PCs).
+ * ioperm() in user level (such as Alpha PCs) or not at all.
+ *
+ * XXX: you should never use these, just call ioperm directly..
*/
+#ifdef CONFIG_X86
case KDADDIO:
case KDDELIO:
/*
* KDADDIO and KDDELIO may be able to add ports beyond what
* we reject here, but to be safe...
*/
+ printk(KERN_INFO "[%s]: using obsolete KDADDIO/KDDELIO ioctl\n",
+ current->comm);
if (arg < GPFIRST || arg > GPLAST)
return -EINVAL;
return sys_ioperm(arg, 1, (cmd == KDADDIO)) ? -ENXIO : 0;

case KDENABIO:
case KDDISABIO:
+ printk(KERN_INFO "[%s]: using obsolete KDENABIO/KDDISABIO ioctl\n",
+ current->comm);
return sys_ioperm(GPFIRST, GPNUM,
(cmd == KDENABIO)) ? -ENXIO : 0;
#endif

2003-05-24 17:50:27

by James Simmons

[permalink] [raw]
Subject: Re: [PATCH] make vt_ioctl ix86isms explicit


> > sys_ioperm is only implemented on x86 (i386/x86_64). Make the
> > ifdefs in vt_ioctl.c more explicit so the other architectures can
> > get rid of their stubs in favour of just using sys_ni_syscall in
> > the syscall table. (Personally I still wonder why they added it
> > at all but that's another question..)
>
> They were added because this was how the X server got IO permissions a
> million years ago. It comes from some random old UNIX/386 thing, it
> predates Linux itself as far as I know.
>
> I'm fairly certain that X itself no longer uses it at all, but there may
> be other programs that still do (unlikely). Your patch looks sane,
> although it might be equally sane to just remove the code altogether.

I suggest we remove it. It was the old way to access the VGA register
ports.

2003-05-24 22:03:35

by Andries Brouwer

[permalink] [raw]
Subject: Re: [PATCH] make vt_ioctl ix86isms explicit

On Sat, May 24, 2003 at 08:03:17PM +0200, Christoph Hellwig wrote:

> What about this one instead? If no one sees this messages we can just
> rip it out in 2.7.<early>

I think you can just rip it out.

A grep on some random source tree shows zero occurrences.
XFree86 uses it only in DGUX and SYSV sections.
A larger search turns up one more small program dated 1990.

Andries