On Tue, 27 Aug 2002, Linus Torvalds wrote:
>
> This test is senseless, in my opinion:
>
> > + if (cpu_has_pge) \
> > + __flush_tlb_single(addr); \
>
> The test _should_ be for something like
>
> if (cpu_has_invlpg)
> __flush_tlb_single(addr);
>
> since we want to use the invlpg instruction regardless of any PGE issues
New patch below defines cpu_has_invlpg as (boot_cpu_data.x86 > 3).
But I do feel safer with that original cpu_has_pge test, which was
using a decent capability flag, and only changed behaviour of the
CONFIG_M386 __flush_tlb_one when it's necessary.
Isn't CONFIG_M386 about maximum safe applicability, rather than speed?
Am I imagining it, or were there a few i386 + i486 SMP machines built?
Or might there be some i486 clone which didn't really implement invlpg,
which could be used with a CONFIG_M386 kernel before this change,
but not after? But perhaps I'm just dreaming up excuses for my
senselessness - your call.
Hugh
CONFIG_M386 kernel running on PPro+ processor with X86_FEATURE_PGE may
set _PAGE_GLOBAL bit: then __flush_tlb_one must use invlpg instruction.
--- 2.5.32/include/asm-i386/tlbflush.h Tue May 28 21:41:36 2002
+++ linux/include/asm-i386/tlbflush.h Wed Aug 28 20:48:44 2002
@@ -45,11 +45,21 @@
__flush_tlb(); \
} while (0)
-#ifndef CONFIG_X86_INVLPG
-#define __flush_tlb_one(addr) __flush_tlb()
+#define cpu_has_invlpg (boot_cpu_data.x86 > 3)
+
+#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_invlpg) \
+ __flush_tlb_single(addr); \
+ else \
+ __flush_tlb(); \
+ } while (0)
#endif
/*
-
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/
This archive was generated by hypermail 2b29 : Sat Aug 31 2002 - 22:00:24 EST