On PA-RISC we have 36 signals defined for hpux compatibility. So M()
and T() in kernel/signal.c try to do (1UL << 33) which is garbage on 32-bit
architectures. How do people feel about this patch?
--- kernel/signal.c 21 Oct 2002 03:46:14 -0000 1.5
+++ kernel/signal.c 22 Oct 2002 21:25:29 -0000
@@ -96,7 +96,7 @@ int max_queued_signals = 1024;
#define M_SIGEMT 0
#endif
-#define M(sig) (1UL << (sig))
+#define M(sig) (1ULL << (sig))
#define SIG_USER_SPECIFIC_MASK (\
M(SIGILL) | M(SIGTRAP) | M(SIGABRT) | M(SIGBUS) | \
@@ -132,7 +132,7 @@ int max_queued_signals = 1024;
M(SIGXCPU) | M(SIGXFSZ) | M_SIGEMT )
#define T(sig, mask) \
- ((1UL << (sig)) & mask)
+ ((1ULL << (sig)) & mask)
#define sig_user_specific(sig) \
(((sig) < SIGRTMIN) && T(sig, SIG_USER_SPECIFIC_MASK))
(patch copied & pasted; whitespace is munged. This is an indication I
expect to get some pushback on this issue ;-)
--
Revolutions do not require corporate support.
On Tue, 2002-10-22 at 22:27, Matthew Wilcox wrote:
>
> On PA-RISC we have 36 signals defined for hpux compatibility. So M()
> and T() in kernel/signal.c try to do (1UL << 33) which is garbage on 32-bit
> architectures. How do people feel about this patch?
How does the compiler output look ?
On Tue, Oct 22, 2002 at 10:57:59PM +0100, Alan Cox wrote:
> On Tue, 2002-10-22 at 22:27, Matthew Wilcox wrote:
> >
> > On PA-RISC we have 36 signals defined for hpux compatibility. So M()
> > and T() in kernel/signal.c try to do (1UL << 33) which is garbage on 32-bit
> > architectures. How do people feel about this patch?
>
> How does the compiler output look ?
uhh.. 200 bytes extra on x86 ;-(
-rw-r--r-- 1 willy users 17956 Oct 22 14:44 kernel/signal.o
-rw-r--r-- 1 willy users 17748 Oct 22 06:50 kernel/signal.o_orig
--
Revolutions do not require corporate support.
On Tue, 2002-10-22 at 22:48, Matthew Wilcox wrote:
> On Tue, Oct 22, 2002 at 10:57:59PM +0100, Alan Cox wrote:
> > On Tue, 2002-10-22 at 22:27, Matthew Wilcox wrote:
> > >
> > > On PA-RISC we have 36 signals defined for hpux compatibility. So M()
> > > and T() in kernel/signal.c try to do (1UL << 33) which is garbage on 32-bit
> > > architectures. How do people feel about this patch?
> >
> > How does the compiler output look ?
>
> uhh.. 200 bytes extra on x86 ;-(
>
> -rw-r--r-- 1 willy users 17956 Oct 22 14:44 kernel/signal.o
> -rw-r--r-- 1 willy users 17748 Oct 22 06:50 kernel/signal.o_orig
Care to move the define into include/asm-foo then ?
On Wed, Oct 23, 2002 at 12:17:12AM +0100, Alan Cox wrote:
> Care to move the define into include/asm-foo then ?
How about this instead? All other arches define SIGRTMIN to be 32, so
this only affects PA.
Index: kernel/signal.c
===================================================================
RCS file: /var/cvs/linux-2.5/kernel/signal.c,v
retrieving revision 1.1.2.4
diff -u -p -r1.1.2.4 signal.c
--- kernel/signal.c 21 Oct 2002 03:07:20 -0000 1.1.2.4
+++ kernel/signal.c 23 Oct 2002 13:14:33 -0000
@@ -96,7 +96,12 @@ int max_queued_signals = 1024;
#define M_SIGEMT 0
#endif
+#if SIGRTMIN > 32
+#define M(sig) (1ULL << (sig))
+#else
#define M(sig) (1UL << (sig))
+#endif
+#define T(sig, mask) (M(sig) & mask)
#define SIG_USER_SPECIFIC_MASK (\
M(SIGILL) | M(SIGTRAP) | M(SIGABRT) | M(SIGBUS) | \
@@ -130,9 +135,6 @@ int max_queued_signals = 1024;
M(SIGQUIT) | M(SIGILL) | M(SIGTRAP) | M(SIGABRT) | \
M(SIGFPE) | M(SIGSEGV) | M(SIGBUS) | M(SIGSYS) | \
M(SIGXCPU) | M(SIGXFSZ) | M_SIGEMT )
-
-#define T(sig, mask) \
- ((1UL << (sig)) & mask)
#define sig_user_specific(sig) \
(((sig) < SIGRTMIN) && T(sig, SIG_USER_SPECIFIC_MASK))
--
Revolutions do not require corporate support.
On Wed, 2002-10-23 at 14:17, Matthew Wilcox wrote:
> +#if SIGRTMIN > 32
> +#define M(sig) (1ULL << (sig))
> +#else
> #define M(sig) (1UL << (sig))
> +#endif
Not >= ??
otherwise it looks very sensible
On Wed, Oct 23, 2002 at 02:48:31PM +0100, Alan Cox wrote:
> On Wed, 2002-10-23 at 14:17, Matthew Wilcox wrote:
> > +#if SIGRTMIN > 32
> > +#define M(sig) (1ULL << (sig))
> > +#else
> > #define M(sig) (1UL << (sig))
> > +#endif
>
> Not >= ??
No, definitely not >=. Realtime signals are tested for separately, and
SIGRTMIN is the number of the lowest RT signal, not the number of the
highest non-realtime signal. Take a look in include/asm-i386/signal.h:
#define SIGSYS 31
#define SIGUNUSED 31
/* These should not be considered constants from userland. */
#define SIGRTMIN 32
#define SIGRTMAX (_NSIG-1)
--
Revolutions do not require corporate support.