Re: [PATCH] make do_signal static on i386

Re: [PATCH] make do_signal static on i386

[Pavel Machek]

Hi!

> I am not sure whether we need the FASTCALL() or whether the change
> in the comment in asm-um/signal.h makes sense.  (Does UML work on
> x86_64, yet?)
>

I dont think it does... 

Re: [PATCH] make do_signal static on i386

[Andi Kleen]

Stephen Rothwell <sfr@canb.auug.org.au> writes:

> I am not sure whether we need the FASTCALL() or whether the change

The FASTCALL was only needed when it was still called from entry.S
(= in 2.4) 

> in the comment in asm-um/signal.h makes sense.  (Does UML work on
> x86_64, yet?)

I doubt that it works. At least I never tried it.

-Andi

[PATCH] make do_signal static on i386

[Stephen Rothwell]

Hi Linus,

This patch makes do_signal static in arch/i386/kernel/signal.c which
means its declaration can be removed from asm-i386/signal.h which may
help Jeff out with UML.

I am not sure whether we need the FASTCALL() or whether the change
in the comment in asm-um/signal.h makes sense.  (Does UML work on
x86_64, yet?)

-- 
Cheers,
Stephen Rothwell                    sfr@canb.auug.org.au
http://www.canb.auug.org.au/~sfr/

diff -ruN 2.5.41-1.715/arch/i386/kernel/signal.c 2.5.41-1.715-si.1/arch/i386/kernel/signal.c
--- 2.5.41-1.715/arch/i386/kernel/signal.c	2002-10-02 11:23:54.000000000 +1000
+++ 2.5.41-1.715-si.1/arch/i386/kernel/signal.c	2002-10-09 17:52:15.000000000 +1000
@@ -27,6 +27,8 @@
 
 #define _BLOCKABLE (~(sigmask(SIGKILL) | sigmask(SIGSTOP)))
 
+static int FASTCALL(do_signal(struct pt_regs *regs, sigset_t *oldset));
+
 /*
  * Atomically swap in the new signal mask, and wait for a signal.
  */
@@ -545,7 +547,7 @@
  * want to handle. Thus you cannot kill init even with a SIGKILL even by
  * mistake.
  */
-int do_signal(struct pt_regs *regs, sigset_t *oldset)
+static int do_signal(struct pt_regs *regs, sigset_t *oldset)
 {
 	siginfo_t info;
 	int signr;
diff -ruN 2.5.41-1.715/include/asm-i386/signal.h 2.5.41-1.715-si.1/include/asm-i386/signal.h
--- 2.5.41-1.715/include/asm-i386/signal.h	2002-01-31 07:12:46.000000000 +1100
+++ 2.5.41-1.715-si.1/include/asm-i386/signal.h	2002-10-09 17:54:28.000000000 +1000
@@ -2,7 +2,6 @@
 #define _ASMi386_SIGNAL_H
 
 #include <linux/types.h>
-#include <linux/linkage.h>
 
 /* Avoid too many header ordering problems.  */
 struct siginfo;
@@ -217,9 +216,6 @@
 	return word;
 }
 
-struct pt_regs;
-extern int FASTCALL(do_signal(struct pt_regs *regs, sigset_t *oldset));
-
 #endif /* __KERNEL__ */
 
 #endif
diff -ruN 2.5.41-1.715/include/asm-um/signal.h 2.5.41-1.715-si.1/include/asm-um/signal.h
--- 2.5.41-1.715/include/asm-um/signal.h	2002-09-16 13:40:57.000000000 +1000
+++ 2.5.41-1.715-si.1/include/asm-um/signal.h	2002-10-09 17:56:20.000000000 +1000
@@ -6,7 +6,7 @@
 #ifndef __UM_SIGNAL_H
 #define __UM_SIGNAL_H
 
-/* Need to kill the do_signal() declaration in the i386 signal.h */
+/* Need to kill the do_signal() declaration in the x86_64 signal.h */
 
 #define do_signal do_signal_renamed
 #include "asm/arch/signal.h"

Re: [PATCH] make do_signal static on i386

[Jeff Dike]

sfr@canb.auug.org.au said:
> This patch makes do_signal static in arch/i386/kernel/signal.c which
> means its declaration can be removed from asm-i386/signal.h which may
> help Jeff out with UML. 

Cool, anything which makes the other arch headers more UML-friendly is
good.

> (Does UML work on x86_64, yet?)

Not yet.

				Jeff

Re: [PATCH] make do_signal static on i386

[george anzinger]

This will cause problems for nano_sleep and
clock_nanosleep.  These system calls need to call
do_signal() in order to meet the POSIX standard which states
that they are interrupted only by signals that are delivered
to user code.  Other signals are not to interrupt the
sleep.  This is why do_signal() returns this information to
the caller.

I suppose one could argue that such functions should be in
signal.c, but save for this signal issue the code is
common.  Seems a waste to support the same code in N
platforms.

-g

Stephen Rothwell wrote:
> 
> Hi Linus,
> 
> This patch makes do_signal static in arch/i386/kernel/signal.c which
> means its declaration can be removed from asm-i386/signal.h which may
> help Jeff out with UML.
> 
> I am not sure whether we need the FASTCALL() or whether the change
> in the comment in asm-um/signal.h makes sense.  (Does UML work on
> x86_64, yet?)
> 
> --
> Cheers,
> Stephen Rothwell                    sfr@canb.auug.org.au
> http://www.canb.auug.org.au/~sfr/
> 
> diff -ruN 2.5.41-1.715/arch/i386/kernel/signal.c 2.5.41-1.715-si.1/arch/i386/kernel/signal.c
> --- 2.5.41-1.715/arch/i386/kernel/signal.c      2002-10-02 11:23:54.000000000 +1000
> +++ 2.5.41-1.715-si.1/arch/i386/kernel/signal.c 2002-10-09 17:52:15.000000000 +1000
> @@ -27,6 +27,8 @@
> 
>  #define _BLOCKABLE (~(sigmask(SIGKILL) | sigmask(SIGSTOP)))
> 
> +static int FASTCALL(do_signal(struct pt_regs *regs, sigset_t *oldset));
> +
>  /*
>   * Atomically swap in the new signal mask, and wait for a signal.
>   */
> @@ -545,7 +547,7 @@
>   * want to handle. Thus you cannot kill init even with a SIGKILL even by
>   * mistake.
>   */
> -int do_signal(struct pt_regs *regs, sigset_t *oldset)
> +static int do_signal(struct pt_regs *regs, sigset_t *oldset)
>  {
>         siginfo_t info;
>         int signr;
> diff -ruN 2.5.41-1.715/include/asm-i386/signal.h 2.5.41-1.715-si.1/include/asm-i386/signal.h
> --- 2.5.41-1.715/include/asm-i386/signal.h      2002-01-31 07:12:46.000000000 +1100
> +++ 2.5.41-1.715-si.1/include/asm-i386/signal.h 2002-10-09 17:54:28.000000000 +1000
> @@ -2,7 +2,6 @@
>  #define _ASMi386_SIGNAL_H
> 
>  #include <linux/types.h>
> -#include <linux/linkage.h>
> 
>  /* Avoid too many header ordering problems.  */
>  struct siginfo;
> @@ -217,9 +216,6 @@
>         return word;
>  }
> 
> -struct pt_regs;
> -extern int FASTCALL(do_signal(struct pt_regs *regs, sigset_t *oldset));
> -
>  #endif /* __KERNEL__ */
> 
>  #endif
> diff -ruN 2.5.41-1.715/include/asm-um/signal.h 2.5.41-1.715-si.1/include/asm-um/signal.h
> --- 2.5.41-1.715/include/asm-um/signal.h        2002-09-16 13:40:57.000000000 +1000
> +++ 2.5.41-1.715-si.1/include/asm-um/signal.h   2002-10-09 17:56:20.000000000 +1000
> @@ -6,7 +6,7 @@
>  #ifndef __UM_SIGNAL_H
>  #define __UM_SIGNAL_H
> 
> -/* Need to kill the do_signal() declaration in the i386 signal.h */
> +/* Need to kill the do_signal() declaration in the x86_64 signal.h */
> 
>  #define do_signal do_signal_renamed
>  #include "asm/arch/signal.h"
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

-- 
George Anzinger   george@mvista.com
High-res-timers: 
http://sourceforge.net/projects/high-res-timers/
Preemption patch:
http://www.kernel.org/pub/linux/kernel/people/rml

Re: [PATCH] make do_signal static on i386

[Daniel Jacobowitz]

On Wed, Oct 09, 2002 at 10:40:39AM -0700, george anzinger wrote:
> This will cause problems for nano_sleep and
> clock_nanosleep.  These system calls need to call
> do_signal() in order to meet the POSIX standard which states
> that they are interrupted only by signals that are delivered
> to user code.  Other signals are not to interrupt the
> sleep.  This is why do_signal() returns this information to
> the caller.
> 
> I suppose one could argue that such functions should be in
> signal.c, but save for this signal issue the code is
> common.  Seems a waste to support the same code in N
> platforms.

IMO, calling the architecture's do_signal function to handle that is
entirely the wrong way to go.  They don't even all have the same
arguments, and the wrappers hi-res-timers put around sys_nanosleep are
hideous.

All of this should be handled correctly in kernel/signal.c, and things
like triggering the debugger should be done from there, not duplicated
in each platform's signal delivery code.

Ideally we should even trigger the debugger without necessarily
knocking the sleeping process out of sleep.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: [PATCH] make do_signal static on i386

[george anzinger]

Daniel Jacobowitz wrote:
> 
> On Wed, Oct 09, 2002 at 10:40:39AM -0700, george anzinger wrote:
> > This will cause problems for nano_sleep and
> > clock_nanosleep.  These system calls need to call
> > do_signal() in order to meet the POSIX standard which states
> > that they are interrupted only by signals that are delivered
> > to user code.  Other signals are not to interrupt the
> > sleep.  This is why do_signal() returns this information to
> > the caller.
> >
> > I suppose one could argue that such functions should be in
> > signal.c, but save for this signal issue the code is
> > common.  Seems a waste to support the same code in N
> > platforms.
> 
> IMO, calling the architecture's do_signal function to handle that is
> entirely the wrong way to go.  They don't even all have the same
> arguments, and the wrappers hi-res-timers put around sys_nanosleep are
> hideous.

Ah, but they should, unless you mean the register structure
is different...
> 
> All of this should be handled correctly in kernel/signal.c, and things
> like triggering the debugger should be done from there, not duplicated
> in each platform's signal delivery code.

The other side of this nasty mess is getting a handle on the
registers.  They are all passed differently from the system
call stuff in entry.S (or what ever).
> 
> Ideally we should even trigger the debugger without necessarily
> knocking the sleeping process out of sleep.

Actually it is the other way round, the debugger is
triggering the task and asking it to go to a "collection"
point.

I would love to do it otherwise.  But this is what we have
today.  And the debugger is not the only case of signals
that do not get delivered to user code.  For the moment, I
would like to make sure the problem can still be addressed. 
At least today, a solution is possible.

As to how to deliver such events, what is needed is a way to
make a task arrive at a known point and to wait there.  Sure
sounds like a signal to me.  I don't see how this can be
done with out waking sleepers, but then there are lots of
things I don't see :)


-- 
George Anzinger   george@mvista.com
High-res-timers: 
http://sourceforge.net/projects/high-res-timers/
Preemption patch:
http://www.kernel.org/pub/linux/kernel/people/rml