linux-mips
[Top] [All Lists]

Re: [MIPS] Fix swap entry for MIPS32 36-bit physical address

To: linux-mips@linux-mips.org
Subject: Re: [MIPS] Fix swap entry for MIPS32 36-bit physical address
From: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Date: Sat, 27 May 2006 19:58:03 +0400
Cc: Ralf Baechle <ralf@linux-mips.org>, Jordan Crouse <jordan.crouse@amd.com>
In-reply-to: <S8133863AbWEZXEn/20060526230443Z+689@ftp.linux-mips.org>
Organization: MontaVista Software Inc.
Original-recipient: rfc822;linux-mips@linux-mips.org
References: <S8133863AbWEZXEn/20060526230443Z+689@ftp.linux-mips.org>
Sender: linux-mips-bounce@linux-mips.org
User-agent: Mozilla/5.0 (X11; U; Linux i686; rv:1.7.2) Gecko/20040803
Hello.

linux-mips@linux-mips.org wrote:
Author: Sergei Shtylyov <sshtylyov@ru.mvista.com> Wed Apr 5 00:24:40 2006 +0400
Comitter: Ralf Baechle <ralf@linux-mips.org> Sat May 27 00:01:23 2006 +0100
Commit: f826429f5ad38f40b96ff587cd44def138f99f3a
Gitweb: http://www.linux-mips.org/g/linux/f826429f
Branch: master

With 64-bit physical address enabled, 'swapon' was causing kernel oops
on Alchemy CPUs (MIPS32R1) because of the swap entry type field
corrupting the _PAGE_FILE bit in pte_low. So, change layout of the
swap entry to use all bits except _PAGE_PRESENT and _PAGE_FILE (the
hardware protection bits are loaded from pte_high which should be
cleared by __swp_entry_to_pte() macro) -- which gives 25 bits for the
swap entry offset.  Additionally, PTEs in MIPS32R2 should have the same
layout for 64-bit physical address case as in MIPS32R1, according to
the architecture manuals -- so, fix the #ifdef's.

Signed-off-by: Konstantin Baydarov <kbaidarov@ru.mvista.com>
Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

---

 include/asm-mips/pgtable-32.h   |   22 +++++++++++++++++++---
 include/asm-mips/pgtable-bits.h |    2 +-
 2 files changed, 20 insertions(+), 4 deletions(-)

Alas, Ralf have finally managed to commit the wrong patch -- both outdated and error-prone because it reuses _PAGE_GLOBAL which is bad for set_pte() and pte_clear() and probably for the CPU itself since it ANDs G-bit of both the even and odd PTEs to decide whether to compare ASID or not when doing TLB lookups. I wonder whether this commit can be undone (it's still the most recent one) or should I compose an incremental patch?

WBR, Sergei

<Prev in Thread] Current Thread [Next in Thread>
  • Re: [MIPS] Fix swap entry for MIPS32 36-bit physical address, Sergei Shtylyov <=