linux/arch/s390/mm/maccess.c
Michael Holzheu b2a68c2356 s390: allow absolute memory access for /dev/mem
Currently dev/mem for s390 provides only real memory access. This means
that the CPU prefix pages are swapped. The prefix swap for real memory
works as follows:

Each CPU owns a prefix register that points to a page aligned memory
location "P". If this CPU accesses the address range [0,0x1fff], it is
translated by the hardware to [P,P+0x1fff]. Accordingly if this CPU
accesses the address range [P,P+0x1fff], it is translated by the hardware
to [0,0x1fff].  Therefore, if [P,P+0x1fff] or [0,0x1fff] is read from
the current /dev/mem device, the incorrectly swapped memory content is
returned.

With this patch the /dev/mem architecture code is modified to provide
absolute memory access. This is done via the arch specific functions
xlate_dev_mem_ptr() and unxlate_dev_mem_ptr(). For swapped pages on
s390 the function xlate_dev_mem_ptr() now returns a new buffer with a
copy of the requested absolute memory. In case the buffer was allocated,
the unxlate_dev_mem_ptr() function frees it after /dev/mem code has
called copy_to_user().

Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2012-05-16 14:42:38 +02:00

236 lines
5.1 KiB
C

/*
* Access kernel memory without faulting -- s390 specific implementation.
*
* Copyright IBM Corp. 2009
*
* Author(s): Heiko Carstens <heiko.carstens@de.ibm.com>,
*
*/
#include <linux/uaccess.h>
#include <linux/kernel.h>
#include <linux/types.h>
#include <linux/errno.h>
#include <linux/gfp.h>
#include <linux/cpu.h>
#include <asm/ctl_reg.h>
/*
* This function writes to kernel memory bypassing DAT and possible
* write protection. It copies one to four bytes from src to dst
* using the stura instruction.
* Returns the number of bytes copied or -EFAULT.
*/
static long probe_kernel_write_odd(void *dst, const void *src, size_t size)
{
unsigned long count, aligned;
int offset, mask;
int rc = -EFAULT;
aligned = (unsigned long) dst & ~3UL;
offset = (unsigned long) dst & 3;
count = min_t(unsigned long, 4 - offset, size);
mask = (0xf << (4 - count)) & 0xf;
mask >>= offset;
asm volatile(
" bras 1,0f\n"
" icm 0,0,0(%3)\n"
"0: l 0,0(%1)\n"
" lra %1,0(%1)\n"
"1: ex %2,0(1)\n"
"2: stura 0,%1\n"
" la %0,0\n"
"3:\n"
EX_TABLE(0b,3b) EX_TABLE(1b,3b) EX_TABLE(2b,3b)
: "+d" (rc), "+a" (aligned)
: "a" (mask), "a" (src) : "cc", "memory", "0", "1");
return rc ? rc : count;
}
long probe_kernel_write(void *dst, const void *src, size_t size)
{
long copied = 0;
while (size) {
copied = probe_kernel_write_odd(dst, src, size);
if (copied < 0)
break;
dst += copied;
src += copied;
size -= copied;
}
return copied < 0 ? -EFAULT : 0;
}
static int __memcpy_real(void *dest, void *src, size_t count)
{
register unsigned long _dest asm("2") = (unsigned long) dest;
register unsigned long _len1 asm("3") = (unsigned long) count;
register unsigned long _src asm("4") = (unsigned long) src;
register unsigned long _len2 asm("5") = (unsigned long) count;
int rc = -EFAULT;
asm volatile (
"0: mvcle %1,%2,0x0\n"
"1: jo 0b\n"
" lhi %0,0x0\n"
"2:\n"
EX_TABLE(1b,2b)
: "+d" (rc), "+d" (_dest), "+d" (_src), "+d" (_len1),
"+d" (_len2), "=m" (*((long *) dest))
: "m" (*((long *) src))
: "cc", "memory");
return rc;
}
/*
* Copy memory in real mode (kernel to kernel)
*/
int memcpy_real(void *dest, void *src, size_t count)
{
unsigned long flags;
int rc;
if (!count)
return 0;
local_irq_save(flags);
__arch_local_irq_stnsm(0xfbUL);
rc = __memcpy_real(dest, src, count);
local_irq_restore(flags);
return rc;
}
/*
* Copy memory to absolute zero
*/
void copy_to_absolute_zero(void *dest, void *src, size_t count)
{
unsigned long cr0;
BUG_ON((unsigned long) dest + count >= sizeof(struct _lowcore));
preempt_disable();
__ctl_store(cr0, 0, 0);
__ctl_clear_bit(0, 28); /* disable lowcore protection */
memcpy_real(dest + store_prefix(), src, count);
__ctl_load(cr0, 0, 0);
preempt_enable();
}
/*
* Copy memory from kernel (real) to user (virtual)
*/
int copy_to_user_real(void __user *dest, void *src, size_t count)
{
int offs = 0, size, rc;
char *buf;
buf = (char *) __get_free_page(GFP_KERNEL);
if (!buf)
return -ENOMEM;
rc = -EFAULT;
while (offs < count) {
size = min(PAGE_SIZE, count - offs);
if (memcpy_real(buf, src + offs, size))
goto out;
if (copy_to_user(dest + offs, buf, size))
goto out;
offs += size;
}
rc = 0;
out:
free_page((unsigned long) buf);
return rc;
}
/*
* Copy memory from user (virtual) to kernel (real)
*/
int copy_from_user_real(void *dest, void __user *src, size_t count)
{
int offs = 0, size, rc;
char *buf;
buf = (char *) __get_free_page(GFP_KERNEL);
if (!buf)
return -ENOMEM;
rc = -EFAULT;
while (offs < count) {
size = min(PAGE_SIZE, count - offs);
if (copy_from_user(buf, src + offs, size))
goto out;
if (memcpy_real(dest + offs, buf, size))
goto out;
offs += size;
}
rc = 0;
out:
free_page((unsigned long) buf);
return rc;
}
/*
* Check if physical address is within prefix or zero page
*/
static int is_swapped(unsigned long addr)
{
unsigned long lc;
int cpu;
if (addr < sizeof(struct _lowcore))
return 1;
for_each_online_cpu(cpu) {
lc = (unsigned long) lowcore_ptr[cpu];
if (addr > lc + sizeof(struct _lowcore) - 1 || addr < lc)
continue;
return 1;
}
return 0;
}
/*
* Return swapped prefix or zero page address
*/
static unsigned long get_swapped(unsigned long addr)
{
unsigned long prefix = store_prefix();
if (addr < sizeof(struct _lowcore))
return addr + prefix;
if (addr >= prefix && addr < prefix + sizeof(struct _lowcore))
return addr - prefix;
return addr;
}
/*
* Convert a physical pointer for /dev/mem access
*
* For swapped prefix pages a new buffer is returned that contains a copy of
* the absolute memory. The buffer size is maximum one page large.
*/
void *xlate_dev_mem_ptr(unsigned long addr)
{
void *bounce = (void *) addr;
unsigned long size;
get_online_cpus();
preempt_disable();
if (is_swapped(addr)) {
size = PAGE_SIZE - (addr & ~PAGE_MASK);
bounce = (void *) __get_free_page(GFP_ATOMIC);
if (bounce)
memcpy_real(bounce, (void *) get_swapped(addr), size);
}
preempt_enable();
put_online_cpus();
return bounce;
}
/*
* Free converted buffer for /dev/mem access (if necessary)
*/
void unxlate_dev_mem_ptr(unsigned long addr, void *buf)
{
if ((void *) addr != buf)
free_page((unsigned long) buf);
}