On Fri, Aug 2, 2013 at 11:08 AM, Michael Schmitz
<schmitz@xxxxxxxxxxxxxxxxxxxxxxxxxx> wrote:
[ 0.160000] WARNING: at /root/linux-3.10.1/init/main.c:698
do_one_initcall+0x12e/0x13a()
[ 0.160000] initcall param_sysfs_init+0x0/0x1a4 returned with disabled
interrupts
This seems to happen for several initcalls. Geert et al,
can you have a look at them, they’re scary ☺
It only happens on multi-platform kernels, because of the following
definition
of ALLOWINT:
#if defined(MACH_ATARI_ONLY)
/* block out HSYNC = ipl 2 on the atari */
#define ALLOWINT (~0x500)
#else
/* portable version */
#define ALLOWINT (~0x700)
#endif /* machine compilation types */
On an Atari-only kernel, flags is either 0x00c02204 or 0x00c02214.
Hence "flags & ~ALLOWINT" is "flags & 0x500" is always zero.
On a multi-platform kernel, flags is one of 0x00c02004, 0x00c02014,
0x00c02204, or 0x00c02214.
Hence "flags & ~ALLOWINT" is "flags & 0x700" is sometimes non-zero,
triggering the warning.
Anyone who sees a solution that doesn't involve adding a variable to hold
ALLOWINT?
We're already having a check for Q40 in arch_local_irq_enable()
(in multi-platfom kernels only).
Didn't we have that sorted out earlier? I seem to recall this has surfaced
before.
You mean commit 94674cd5299e825cb31979c3b9a4c1a3e6074839
("m68k: Correct the Atari ALLOWINT definition")? That was a related but
slightly different problem.
What is the cause of the problem exactly - the hsync handler changing the
IPL to block out further interrupts, whenever it is called for the first
time after interrupts are enabled? We could stop doing that on
multi-platform kernels (taking all hsync interrupts will be a performance
hit but not stop the system from working).
AFAICS, it's indeed the hsync handler blocking further interrupts on
multi-platform kernels.
On multi-platform kernels, enabling interrupts enables all 7 levels
(interrupt enable/disable is either all-or-nothing in Linux, there are
no priorities).
When the hblank interrupt comes in, its handler (falcon_hblhandler())
disables level 2 interrupts again, to prevent more interrupts coming in,
which would slow down the system.
However, this causes arch_irqs_disabled_flags() to think interrupts are
completely disabled, causing the warnings in the logs.
We could ignore IPL2 when running on Atari (untested whitespace-damaged
patch):
--- a/arch/m68k/include/asm/irqflags.h
+++ b/arch/m68k/include/asm/irqflags.h
@@ -67,7 +67,11 @@ static inline void arch_local_irq_restore(unsigned long flags
static inline bool arch_irqs_disabled_flags(unsigned long flags)
{
- return (flags & ~ALLOWINT) != 0;
+ if (MACH_IS_ATARI) {
+ /* Ignore HSYNC = ipl 2 on Atari */
+ return (flags & ~(ALLOWINT | 0x200)) != 0;
+ } else
+ return (flags & ~ALLOWINT) != 0;
}
static inline bool arch_irqs_disabled(void)
or just ignore all priorities on all platforms, and consider interrupts disabled
iff all priorities are disabled:
--- a/arch/m68k/include/asm/irqflags.h
+++ b/arch/m68k/include/asm/irqflags.h
@@ -67,7 +67,7 @@ static inline void arch_local_irq_restore(unsigned long flags
static inline bool arch_irqs_disabled_flags(unsigned long flags)
{
- return (flags & ~ALLOWINT) != 0;
+ return (flags & ~ALLOWINT) != ~ALLOWINT;
}
static inline bool arch_irqs_disabled(void)
The former is safer but slower, the second is faster but will miss cases
where some interrupt priorities are disabled.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
--
To unsubscribe from this list: send the line "unsubscribe linux-m68k" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
