Resent-Date: Fri, 20 Nov 1998 11:57:31 +0100 (MET) To: richard@sleepie.demon.co.uk Cc: linux-m68k@lists.linux-m68k.org Subject: Re: mk_pte_phys() References: <360F9C83.FFFF4C42@net-tel.co.uk> From: Jes Sorensen Date: 20 Nov 1998 11:56:55 +0100 In-Reply-To: Richard Hirst's message of "Mon, 28 Sep 1998 15:26:11 +0100" Resent-From: linux-m68k@phil.uni-sb.de >>>>> "Richard" == Richard Hirst writes: Richard> Hi, asm/pgtable.h has mk_pte() and mk_pte_phys() defined to Richard> both call VTOP() on the address. I think mk_pte_phys() Richard> should be passed a physical address anyway, and therefore Richard> shouldn't call VTOP(). It gets used in remap_page_range(), Richard> and causes me problems if a user process does an mmap() on a Richard> chunk on /dev/memnc to access some range of VME address Richard> space. There are no kernel mmu tables for the vme address Richard> space so mm_vtop() fails. You are damn right, why don't I just listen to you and fix things before they start biting everybody? Anyway this was actually what caused X to fail with 2.1.127+. mmap() hangs forever on the 040/060 when you try to map unmapped memory and if you map the Z2 area it blows up. People who wants to run X with the latest kernel probably wants to apply this patch. Thanks a lot for pointing this out. Jes --- /var/tmp/native-2.1.128/include/asm-m68k/pgtable.h Wed Sep 2 16:39:18 1998 +++ linux/include/asm-m68k/pgtable.h Fri Nov 20 11:12:53 1998 @@ -428,7 +428,7 @@ #define mk_pte(page, pgprot) \ ({ pte_t __pte; pte_val(__pte) = virt_to_phys((void *)page) + pgprot_val(pgprot); __pte; }) #define mk_pte_phys(physpage, pgprot) \ -({ pte_t __pte; pte_val(__pte) = virt_to_phys((void *)physpage) + pgprot_val(pgprot); __pte; }) +({ pte_t __pte; pte_val(__pte) = (unsigned long)physpage + pgprot_val(pgprot); __pte; }) extern inline pte_t pte_modify(pte_t pte, pgprot_t newprot) { pte_val(pte) = (pte_val(pte) & _PAGE_CHG_MASK) | pgprot_val(newprot); return pte; }