kernel_optimize_test/arch/sh/mm/fault.c
Jason Baron df67b3daea [PATCH] make PROT_WRITE imply PROT_READ
Make PROT_WRITE imply PROT_READ for a number of architectures which don't
support write only in hardware.

While looking at this, I noticed that some architectures which do not
support write only mappings already take the exact same approach.  For
example, in arch/alpha/mm/fault.c:

"
        if (cause < 0) {
                if (!(vma->vm_flags & VM_EXEC))
                        goto bad_area;
        } else if (!cause) {
                /* Allow reads even for write-only mappings */
                if (!(vma->vm_flags & (VM_READ | VM_WRITE)))
                        goto bad_area;
        } else {
                if (!(vma->vm_flags & VM_WRITE))
                        goto bad_area;
        }
"

Thus, this patch brings other architectures which do not support write only
mappings in-line and consistent with the rest.  I've verified the patch on
ia64, x86_64 and x86.

Additional discussion:

Several architectures, including x86, can not support write-only mappings.
The pte for x86 reserves a single bit for protection and its two states are
read only or read/write.  Thus, write only is not supported in h/w.

Currently, if i 'mmap' a page write-only, the first read attempt on that page
creates a page fault and will SEGV.  That check is enforced in
arch/blah/mm/fault.c.  However, if i first write that page it will fault in
and the pte will be set to read/write.  Thus, any subsequent reads to the page
will succeed.  It is this inconsistency in behavior that this patch is
attempting to address.  Furthermore, if the page is swapped out, and then
brought back the first read will also cause a SEGV.  Thus, any arbitrary read
on a page can potentially result in a SEGV.

According to the SuSv3 spec, "if the application requests only PROT_WRITE, the
implementation may also allow read access." Also as mentioned, some
archtectures, such as alpha, shown above already take the approach that i am
suggesting.

The counter-argument to this raised by Arjan, is that the kernel is enforcing
the write only mapping the best it can given the h/w limitations.  This is
true, however Alan Cox, and myself would argue that the inconsitency in
behavior, that is applications can sometimes work/sometimes fails is highly
undesireable.  If you read through the thread, i think people, came to an
agreement on the last patch i posted, as nobody has objected to it...

Signed-off-by: Jason Baron <jbaron@redhat.com>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Roman Zippel <zippel@linux-m68k.org>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Andi Kleen <ak@muc.de>
Acked-by: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Acked-by: Paul Mundt <lethal@linux-sh.org>
Cc: Kazumoto Kojima <kkojima@rr.iij4u.or.jp>
Cc: Ian Molton <spyro@f2s.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-29 09:18:05 -07:00

264 lines
5.9 KiB
C

/*
* Page fault handler for SH with an MMU.
*
* Copyright (C) 1999 Niibe Yutaka
* Copyright (C) 2003 Paul Mundt
*
* Based on linux/arch/i386/mm/fault.c:
* Copyright (C) 1995 Linus Torvalds
*
* This file is subject to the terms and conditions of the GNU General Public
* License. See the file "COPYING" in the main directory of this archive
* for more details.
*/
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/hardirq.h>
#include <linux/kprobes.h>
#include <asm/system.h>
#include <asm/mmu_context.h>
#include <asm/kgdb.h>
extern void die(const char *,struct pt_regs *,long);
/*
* This routine handles page faults. It determines the address,
* and the problem, and then passes it off to one of the appropriate
* routines.
*/
asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long writeaccess,
unsigned long address)
{
struct task_struct *tsk;
struct mm_struct *mm;
struct vm_area_struct * vma;
unsigned long page;
#ifdef CONFIG_SH_KGDB
if (kgdb_nofault && kgdb_bus_err_hook)
kgdb_bus_err_hook();
#endif
tsk = current;
mm = tsk->mm;
/*
* If we're in an interrupt or have no user
* context, we must not take the fault..
*/
if (in_atomic() || !mm)
goto no_context;
down_read(&mm->mmap_sem);
vma = find_vma(mm, address);
if (!vma)
goto bad_area;
if (vma->vm_start <= address)
goto good_area;
if (!(vma->vm_flags & VM_GROWSDOWN))
goto bad_area;
if (expand_stack(vma, address))
goto bad_area;
/*
* Ok, we have a good vm_area for this memory access, so
* we can handle it..
*/
good_area:
if (writeaccess) {
if (!(vma->vm_flags & VM_WRITE))
goto bad_area;
} else {
if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
goto bad_area;
}
/*
* If for any reason at all we couldn't handle the fault,
* make sure we exit gracefully rather than endlessly redo
* the fault.
*/
survive:
switch (handle_mm_fault(mm, vma, address, writeaccess)) {
case VM_FAULT_MINOR:
tsk->min_flt++;
break;
case VM_FAULT_MAJOR:
tsk->maj_flt++;
break;
case VM_FAULT_SIGBUS:
goto do_sigbus;
case VM_FAULT_OOM:
goto out_of_memory;
default:
BUG();
}
up_read(&mm->mmap_sem);
return;
/*
* Something tried to access memory that isn't in our memory map..
* Fix it, but check if it's kernel or user first..
*/
bad_area:
up_read(&mm->mmap_sem);
if (user_mode(regs)) {
tsk->thread.address = address;
tsk->thread.error_code = writeaccess;
force_sig(SIGSEGV, tsk);
return;
}
no_context:
/* Are we prepared to handle this kernel fault? */
if (fixup_exception(regs))
return;
/*
* Oops. The kernel tried to access some bad page. We'll have to
* terminate things with extreme prejudice.
*
*/
if (address < PAGE_SIZE)
printk(KERN_ALERT "Unable to handle kernel NULL pointer dereference");
else
printk(KERN_ALERT "Unable to handle kernel paging request");
printk(" at virtual address %08lx\n", address);
printk(KERN_ALERT "pc = %08lx\n", regs->pc);
asm volatile("mov.l %1, %0"
: "=r" (page)
: "m" (__m(MMU_TTB)));
if (page) {
page = ((unsigned long *) page)[address >> 22];
printk(KERN_ALERT "*pde = %08lx\n", page);
if (page & _PAGE_PRESENT) {
page &= PAGE_MASK;
address &= 0x003ff000;
page = ((unsigned long *) __va(page))[address >> PAGE_SHIFT];
printk(KERN_ALERT "*pte = %08lx\n", page);
}
}
die("Oops", regs, writeaccess);
do_exit(SIGKILL);
/*
* We ran out of memory, or some other thing happened to us that made
* us unable to handle the page fault gracefully.
*/
out_of_memory:
up_read(&mm->mmap_sem);
if (current->pid == 1) {
yield();
down_read(&mm->mmap_sem);
goto survive;
}
printk("VM: killing process %s\n", tsk->comm);
if (user_mode(regs))
do_exit(SIGKILL);
goto no_context;
do_sigbus:
up_read(&mm->mmap_sem);
/*
* Send a sigbus, regardless of whether we were in kernel
* or user mode.
*/
tsk->thread.address = address;
tsk->thread.error_code = writeaccess;
tsk->thread.trap_no = 14;
force_sig(SIGBUS, tsk);
/* Kernel mode? Handle exceptions or die */
if (!user_mode(regs))
goto no_context;
}
#ifdef CONFIG_SH_STORE_QUEUES
/*
* This is a special case for the SH-4 store queues, as pages for this
* space still need to be faulted in before it's possible to flush the
* store queue cache for writeout to the remapped region.
*/
#define P3_ADDR_MAX (P4SEG_STORE_QUE + 0x04000000)
#else
#define P3_ADDR_MAX P4SEG
#endif
/*
* Called with interrupts disabled.
*/
asmlinkage int __kprobes __do_page_fault(struct pt_regs *regs,
unsigned long writeaccess,
unsigned long address)
{
pgd_t *pgd;
pud_t *pud;
pmd_t *pmd;
pte_t *pte;
pte_t entry;
struct mm_struct *mm = current->mm;
spinlock_t *ptl;
int ret = 1;
#ifdef CONFIG_SH_KGDB
if (kgdb_nofault && kgdb_bus_err_hook)
kgdb_bus_err_hook();
#endif
/*
* We don't take page faults for P1, P2, and parts of P4, these
* are always mapped, whether it be due to legacy behaviour in
* 29-bit mode, or due to PMB configuration in 32-bit mode.
*/
if (address >= P3SEG && address < P3_ADDR_MAX) {
pgd = pgd_offset_k(address);
mm = NULL;
} else {
if (unlikely(address >= TASK_SIZE || !mm))
return 1;
pgd = pgd_offset(mm, address);
}
pud = pud_offset(pgd, address);
if (pud_none_or_clear_bad(pud))
return 1;
pmd = pmd_offset(pud, address);
if (pmd_none_or_clear_bad(pmd))
return 1;
if (mm)
pte = pte_offset_map_lock(mm, pmd, address, &ptl);
else
pte = pte_offset_kernel(pmd, address);
entry = *pte;
if (unlikely(pte_none(entry) || pte_not_present(entry)))
goto unlock;
if (unlikely(writeaccess && !pte_write(entry)))
goto unlock;
if (writeaccess)
entry = pte_mkdirty(entry);
entry = pte_mkyoung(entry);
#ifdef CONFIG_CPU_SH4
/*
* ITLB is not affected by "ldtlb" instruction.
* So, we need to flush the entry by ourselves.
*/
__flush_tlb_page(get_asid(), address & PAGE_MASK);
#endif
set_pte(pte, entry);
update_mmu_cache(NULL, address, entry);
ret = 0;
unlock:
if (mm)
pte_unmap_unlock(pte, ptl);
return ret;
}