* Odd problem with ACPI and i386 kernel
@ 2002-09-08 22:35 H. J. Lu
2002-09-08 22:55 ` Hugh Dickins
0 siblings, 1 reply; 3+ messages in thread
From: H. J. Lu @ 2002-09-08 22:35 UTC (permalink / raw)
To: linux kernel
I have a very strange problem with ACPI and i386 kernel. I built an
i386 kernel with ACPI for RedHat installation since my new P4 machines
needs ACPI to get IRQ. It works fine on my ASUS P4B533-E MB with Intel
845E chipset. However, on my Sony VAIO GRX560 which is a P4 1.6GHz
with Intel 845 chipset, the machine will reboot as soon as the kernel
starts to run. I tracked it down to CONFIG_X86_INVLPG. If I enable
it, kernel will be fine. Has anyone else seen this?
H.J.
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: Odd problem with ACPI and i386 kernel
2002-09-08 22:35 Odd problem with ACPI and i386 kernel H. J. Lu
@ 2002-09-08 22:55 ` Hugh Dickins
2002-09-08 23:13 ` H. J. Lu
0 siblings, 1 reply; 3+ messages in thread
From: Hugh Dickins @ 2002-09-08 22:55 UTC (permalink / raw)
To: H. J. Lu; +Cc: linux kernel
On Sun, 8 Sep 2002, H. J. Lu wrote:
> I have a very strange problem with ACPI and i386 kernel. I built an
> i386 kernel with ACPI for RedHat installation since my new P4 machines
> needs ACPI to get IRQ. It works fine on my ASUS P4B533-E MB with Intel
> 845E chipset. However, on my Sony VAIO GRX560 which is a P4 1.6GHz
> with Intel 845 chipset, the machine will reboot as soon as the kernel
> starts to run. I tracked it down to CONFIG_X86_INVLPG. If I enable
> it, kernel will be fine. Has anyone else seen this?
Yes, I sent Marcelo the patch below on 27th Aug, it's in 2.4.20-pre5.
I sent Linus a similar patch (copied to LKML) for the 2.5 tlbflush.h,
but he didn't care for its "cpu_has_pge" test, nor did he put in its
#define cpu_has_invlpg (boot_cpu_data.x86 > 3)
replacement: I'll resend.
CONFIG_M386 kernel running on PPro+ processor with X86_FEATURE_PGE may
set _PAGE_GLOBAL bit: then __flush_tlb_one must use invlpg instruction.
The need for this was shown by a recent HyperThreading discussion.
Marc Dietrich reported (LKML 22 Aug) that CONFIG_M386 CONFIG_SMP kernel
on P4 Xeon did not support HT: his dmesg showed acpi_tables_init failed
from bad table data due to unflushed TLB: he confirms patch fixes it.
No tears would be shed if CONFIG_M386 could not support HT, but bad TLB
is trouble. Same CONFIG_M386 bug affects CONFIG_HIGHMEM's kmap_atomic,
and potentially dmi_scan (now using set_fixmap via bt_ioremap). Though
it's true that none of these uses really needs _PAGE_GLOBAL bit itself.
Patch below to 2.4.20-pre4 or 2.4.20-pre4-ac2: please apply.
I'll mail Linus and Dave separately with the 2.5 version.
Hugh
--- 2.4.20-pre4/include/asm-i386/pgtable.h Thu Aug 22 20:59:51 2002
+++ linux/include/asm-i386/pgtable.h Fri Aug 23 00:11:39 2002
@@ -82,11 +82,19 @@
} while (0)
#endif
-#ifndef CONFIG_X86_INVLPG
-#define __flush_tlb_one(addr) __flush_tlb()
+#define __flush_tlb_single(addr) \
+ __asm__ __volatile__("invlpg %0": :"m" (*(char *) addr))
+
+#ifdef CONFIG_X86_INVLPG
+# define __flush_tlb_one(addr) __flush_tlb_single(addr)
#else
-#define __flush_tlb_one(addr) \
-__asm__ __volatile__("invlpg %0": :"m" (*(char *) addr))
+# define __flush_tlb_one(addr) \
+ do { \
+ if (cpu_has_pge) \
+ __flush_tlb_single(addr); \
+ else \
+ __flush_tlb(); \
+ } while (0)
#endif
/*
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: Odd problem with ACPI and i386 kernel
2002-09-08 22:55 ` Hugh Dickins
@ 2002-09-08 23:13 ` H. J. Lu
0 siblings, 0 replies; 3+ messages in thread
From: H. J. Lu @ 2002-09-08 23:13 UTC (permalink / raw)
To: Hugh Dickins; +Cc: linux kernel
On Sun, Sep 08, 2002 at 11:55:58PM +0100, Hugh Dickins wrote:
> On Sun, 8 Sep 2002, H. J. Lu wrote:
> > I have a very strange problem with ACPI and i386 kernel. I built an
> > i386 kernel with ACPI for RedHat installation since my new P4 machines
> > needs ACPI to get IRQ. It works fine on my ASUS P4B533-E MB with Intel
> > 845E chipset. However, on my Sony VAIO GRX560 which is a P4 1.6GHz
> > with Intel 845 chipset, the machine will reboot as soon as the kernel
> > starts to run. I tracked it down to CONFIG_X86_INVLPG. If I enable
> > it, kernel will be fine. Has anyone else seen this?
>
> Yes, I sent Marcelo the patch below on 27th Aug, it's in 2.4.20-pre5.
> I sent Linus a similar patch (copied to LKML) for the 2.5 tlbflush.h,
> but he didn't care for its "cpu_has_pge" test, nor did he put in its
> #define cpu_has_invlpg (boot_cpu_data.x86 > 3)
> replacement: I'll resend.
>
> CONFIG_M386 kernel running on PPro+ processor with X86_FEATURE_PGE may
> set _PAGE_GLOBAL bit: then __flush_tlb_one must use invlpg instruction.
>
THANKS! It does the trick.
H.J.
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2002-09-08 23:08 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2002-09-08 22:35 Odd problem with ACPI and i386 kernel H. J. Lu
2002-09-08 22:55 ` Hugh Dickins
2002-09-08 23:13 ` H. J. Lu
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox