1 files changed, 77 insertions, 98 deletions

diff --git a/arch/s390/mm/maccess.c b/arch/s390/mm/maccess.c
index d6d84e02f35a..1571cdcb0c50 100644
--- a/arch/s390/mm/maccess.c
+++ b/arch/s390/mm/maccess.c
@@ -12,10 +12,17 @@
 #include <linux/errno.h>
 #include <linux/gfp.h>
 #include <linux/cpu.h>
+#include <linux/uio.h>
 #include <asm/asm-extable.h>
 #include <asm/ctl_reg.h>
 #include <asm/io.h>
+#include <asm/abs_lowcore.h>
 #include <asm/stacktrace.h>
+#include <asm/maccess.h>
+
+unsigned long __bootdata_preserved(__memcpy_real_area);
+static __ro_after_init pte_t *memcpy_real_ptep;
+static DEFINE_MUTEX(memcpy_real_mutex);
 
 static notrace long s390_kernel_write_odd(void *dst, const void *src, size_t size)
 {
@@ -76,118 +83,72 @@ notrace void *s390_kernel_write(void *dst, const void *src, size_t size)
 	return dst;
 }
 
-static int __no_sanitize_address __memcpy_real(void *dest, void *src, size_t count)
+void __init memcpy_real_init(void)
 {
-	union register_pair _dst, _src;
-	int rc = -EFAULT;
-
-	_dst.even = (unsigned long) dest;
-	_dst.odd  = (unsigned long) count;
-	_src.even = (unsigned long) src;
-	_src.odd  = (unsigned long) count;
-	asm volatile (
-		"0:	mvcle	%[dst],%[src],0\n"
-		"1:	jo	0b\n"
-		"	lhi	%[rc],0\n"
-		"2:\n"
-		EX_TABLE(1b,2b)
-		: [rc] "+&d" (rc), [dst] "+&d" (_dst.pair), [src] "+&d" (_src.pair)
-		: : "cc", "memory");
-	return rc;
-}
-
-static unsigned long __no_sanitize_address _memcpy_real(unsigned long dest,
-							unsigned long src,
-							unsigned long count)
-{
-	int irqs_disabled, rc;
-	unsigned long flags;
-
-	if (!count)
-		return 0;
-	flags = arch_local_irq_save();
-	irqs_disabled = arch_irqs_disabled_flags(flags);
-	if (!irqs_disabled)
-		trace_hardirqs_off();
-	__arch_local_irq_stnsm(0xf8); // disable DAT
-	rc = __memcpy_real((void *) dest, (void *) src, (size_t) count);
-	if (flags & PSW_MASK_DAT)
-		__arch_local_irq_stosm(0x04); // enable DAT
-	if (!irqs_disabled)
-		trace_hardirqs_on();
-	__arch_local_irq_ssm(flags);
-	return rc;
+	memcpy_real_ptep = vmem_get_alloc_pte(__memcpy_real_area, true);
+	if (!memcpy_real_ptep)
+		panic("Couldn't setup memcpy real area");
 }
 
-/*
- * Copy memory in real mode (kernel to kernel)
- */
-int memcpy_real(void *dest, unsigned long src, size_t count)
+size_t memcpy_real_iter(struct iov_iter *iter, unsigned long src, size_t count)
 {
-	unsigned long _dest  = (unsigned long)dest;
-	unsigned long _src   = (unsigned long)src;
-	unsigned long _count = (unsigned long)count;
-	int rc;
-
-	if (S390_lowcore.nodat_stack != 0) {
-		preempt_disable();
-		rc = call_on_stack(3, S390_lowcore.nodat_stack,
-				   unsigned long, _memcpy_real,
-				   unsigned long, _dest,
-				   unsigned long, _src,
-				   unsigned long, _count);
-		preempt_enable();
-		return rc;
+	size_t len, copied, res = 0;
+	unsigned long phys, offset;
+	void *chunk;
+	pte_t pte;
+
+	while (count) {
+		phys = src & PAGE_MASK;
+		offset = src & ~PAGE_MASK;
+		chunk = (void *)(__memcpy_real_area + offset);
+		len = min(count, PAGE_SIZE - offset);
+		pte = mk_pte_phys(phys, PAGE_KERNEL_RO);
+
+		mutex_lock(&memcpy_real_mutex);
+		if (pte_val(pte) != pte_val(*memcpy_real_ptep)) {
+			__ptep_ipte(__memcpy_real_area, memcpy_real_ptep, 0, 0, IPTE_GLOBAL);
+			set_pte(memcpy_real_ptep, pte);
+		}
+		copied = copy_to_iter(chunk, len, iter);
+		mutex_unlock(&memcpy_real_mutex);
+
+		count -= copied;
+		src += copied;
+		res += copied;
+		if (copied < len)
+			break;
 	}
-	/*
-	 * This is a really early memcpy_real call, the stacks are
-	 * not set up yet. Just call _memcpy_real on the early boot
-	 * stack
-	 */
-	return _memcpy_real(_dest, _src, _count);
+	return res;
 }
 
-/*
- * Copy memory in absolute mode (kernel to kernel)
- */
-void memcpy_absolute(void *dest, void *src, size_t count)
+int memcpy_real(void *dest, unsigned long src, size_t count)
 {
-	unsigned long cr0, flags, prefix;
-
-	flags = arch_local_irq_save();
-	__ctl_store(cr0, 0, 0);
-	__ctl_clear_bit(0, 28); /* disable lowcore protection */
-	prefix = store_prefix();
-	if (prefix) {
-		local_mcck_disable();
-		set_prefix(0);
-		memcpy(dest, src, count);
-		set_prefix(prefix);
-		local_mcck_enable();
-	} else {
-		memcpy(dest, src, count);
-	}
-	__ctl_load(cr0, 0, 0);
-	arch_local_irq_restore(flags);
+	struct iov_iter iter;
+	struct kvec kvec;
+
+	kvec.iov_base = dest;
+	kvec.iov_len = count;
+	iov_iter_kvec(&iter, WRITE, &kvec, 1, count);
+	if (memcpy_real_iter(&iter, src, count) < count)
+		return -EFAULT;
+	return 0;
 }
 
 /*
- * Check if physical address is within prefix or zero page
+ * Find CPU that owns swapped prefix page
  */
-static int is_swapped(phys_addr_t addr)
+static int get_swapped_owner(phys_addr_t addr)
 {
 	phys_addr_t lc;
 	int cpu;
 
-	if (addr < sizeof(struct lowcore))
-		return 1;
 	for_each_online_cpu(cpu) {
 		lc = virt_to_phys(lowcore_ptr[cpu]);
 		if (addr > lc + sizeof(struct lowcore) - 1 || addr < lc)
 			continue;
-		return 1;
+		return cpu;
 	}
-	return 0;
+	return -1;
 }
 
 /*
@@ -200,17 +161,35 @@ void *xlate_dev_mem_ptr(phys_addr_t addr)
 {
 	void *ptr = phys_to_virt(addr);
 	void *bounce = ptr;
+	struct lowcore *abs_lc;
+	unsigned long flags;
 	unsigned long size;
+	int this_cpu, cpu;
 
 	cpus_read_lock();
-	preempt_disable();
-	if (is_swapped(addr)) {
-		size = PAGE_SIZE - (addr & ~PAGE_MASK);
-		bounce = (void *) __get_free_page(GFP_ATOMIC);
-		if (bounce)
-			memcpy_absolute(bounce, ptr, size);
+	this_cpu = get_cpu();
+	if (addr >= sizeof(struct lowcore)) {
+		cpu = get_swapped_owner(addr);
+		if (cpu < 0)
+			goto out;
+	}
+	bounce = (void *)__get_free_page(GFP_ATOMIC);
+	if (!bounce)
+		goto out;
+	size = PAGE_SIZE - (addr & ~PAGE_MASK);
+	if (addr < sizeof(struct lowcore)) {
+		abs_lc = get_abs_lowcore(&flags);
+		ptr = (void *)abs_lc + addr;
+		memcpy(bounce, ptr, size);
+		put_abs_lowcore(abs_lc, flags);
+	} else if (cpu == this_cpu) {
+		ptr = (void *)(addr - virt_to_phys(lowcore_ptr[cpu]));
+		memcpy(bounce, ptr, size);
+	} else {
+		memcpy(bounce, ptr, size);
 	}
-	preempt_enable();
+out:
+	put_cpu();
 	cpus_read_unlock();
 	return bounce;
 }