| Message ID | CE92B8B581B4584FB8CA86B51EDE8AC5243C568B04@wupexchange.riedel.net (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
On Tue, 2009-09-22 at 11:29 +0200, Weirich, Bernhard wrote: > Hello, > > I found the cause of the problem. > My architecture does not define the _PAGE_SPECIAL macro, thus in pte-common.h line 32, it is defined as 0 > And then in line 180, since _PAGE_SPECIAL is defined, the macro __HAVE_ARCH_PTE_SPECIAL is erroneously defined. > This makes any ppc architecture to act as if it has a pte special bit! > I suggest the following simple patch to fix this. Thank ! That looks indeed like a bug. Note that with 2.6.32, we should be able to use 0x20 as _PAGE_SPECIAL on 40x which would improve performances in some areas. Cheers, Ben. > Bernhard WEIRICH > Software Development > > > Signed-off-by: Bernhard Weirich <bernhard.weirich@riedel.net> > --- linux/arch/powerpc/include/asm/pte-common.h 2009-09-16 11:23:44.000000000 +0200 > +++ linux/arch/powerpc/include/asm/pte-common.h 2009-09-22 11:13:08.000000000 +0200 > @@ -176,7 +176,7 @@ > #define HAVE_PAGE_AGP > > /* Advertise support for _PAGE_SPECIAL */ > -#ifdef _PAGE_SPECIAL > +#if _PAGE_SPECIAL != 0 > #define __HAVE_ARCH_PTE_SPECIAL > #endif > > _______________________________________________ > Linuxppc-dev mailing list > Linuxppc-dev@lists.ozlabs.org > https://lists.ozlabs.org/listinfo/linuxppc-dev
On Tue, 2009-09-22 at 20:07 +1000, Benjamin Herrenschmidt wrote: > On Tue, 2009-09-22 at 11:29 +0200, Weirich, Bernhard wrote: > > Hello, > > > > I found the cause of the problem. > > My architecture does not define the _PAGE_SPECIAL macro, thus in pte-common.h line 32, it is defined as 0 > > And then in line 180, since _PAGE_SPECIAL is defined, the macro __HAVE_ARCH_PTE_SPECIAL is erroneously defined. > > This makes any ppc architecture to act as if it has a pte special bit! > > I suggest the following simple patch to fix this. > > Thank ! That looks indeed like a bug. > > Note that with 2.6.32, we should be able to use 0x20 as _PAGE_SPECIAL on > 40x which would improve performances in some areas. I had a quick look at it seems like 2.6.30 also has that bit free. In fact, 8xx also appears to have a few bits free. Maybe we should just define _PAGE_SPECIAL for all the "missing" processor types (I think that's really the only 2 ones) and remove the condition around #define __HAVE_ARCH_PTE_SPECIAL Care to send a patch ? :-) Cheers, Ben. > Cheers, > Ben. > > > Bernhard WEIRICH > > Software Development > > > > > > Signed-off-by: Bernhard Weirich <bernhard.weirich@riedel.net> > > --- linux/arch/powerpc/include/asm/pte-common.h 2009-09-16 11:23:44.000000000 +0200 > > +++ linux/arch/powerpc/include/asm/pte-common.h 2009-09-22 11:13:08.000000000 +0200 > > @@ -176,7 +176,7 @@ > > #define HAVE_PAGE_AGP > > > > /* Advertise support for _PAGE_SPECIAL */ > > -#ifdef _PAGE_SPECIAL > > +#if _PAGE_SPECIAL != 0 > > #define __HAVE_ARCH_PTE_SPECIAL > > #endif > > > > _______________________________________________ > > Linuxppc-dev mailing list > > Linuxppc-dev@lists.ozlabs.org > > https://lists.ozlabs.org/listinfo/linuxppc-dev
--- linux/arch/powerpc/include/asm/pte-common.h 2009-09-16 11:23:44.000000000 +0200 +++ linux/arch/powerpc/include/asm/pte-common.h 2009-09-22 11:13:08.000000000 +0200 @@ -176,7 +176,7 @@ #define HAVE_PAGE_AGP /* Advertise support for _PAGE_SPECIAL */ -#ifdef _PAGE_SPECIAL +#if _PAGE_SPECIAL != 0 #define __HAVE_ARCH_PTE_SPECIAL #endif
Hello, I found the cause of the problem. My architecture does not define the _PAGE_SPECIAL macro, thus in pte-common.h line 32, it is defined as 0 And then in line 180, since _PAGE_SPECIAL is defined, the macro __HAVE_ARCH_PTE_SPECIAL is erroneously defined. This makes any ppc architecture to act as if it has a pte special bit! I suggest the following simple patch to fix this. Bernhard WEIRICH Software Development Signed-off-by: Bernhard Weirich <bernhard.weirich@riedel.net>