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>

arch/s390/include/asm/io.h

@@ -38,11 +38,8 @@ static inline void * phys_to_virt(unsigned long address)
 	return (void *) address;
 }
 
-/*
- * Convert a physical pointer to a virtual kernel pointer for /dev/mem
- * access
- */
-#define xlate_dev_mem_ptr(p)	__va(p)
+void *xlate_dev_mem_ptr(unsigned long phys);
+void unxlate_dev_mem_ptr(unsigned long phys, void *addr);
 
 /*
  * Convert a virtual cached pointer to an uncached pointer

arch/s390/mm/maccess.c

@@ -12,6 +12,7 @@
 #include <linux/types.h>
 #include <linux/errno.h>
 #include <linux/gfp.h>
+#include <linux/cpu.h>
 #include <asm/ctl_reg.h>
 
 /*
@@ -166,3 +167,69 @@ int copy_from_user_real(void *dest, void __user *src, size_t count)
 	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);
+}