s390/mm: rework memcpy_real() to avoid DAT-off mode

Function memcpy_real() is an univeral data mover that does not
require DAT mode to be able reading from a physical address.
Its advantage is an ability to read from any address, even
those for which no kernel virtual mapping exists.

Although memcpy_real() is interrupt-safe, there are no handlers
that make use of this function. The compiler instrumentation
have to be disabled and separate no-DAT stack used to allow
execution of the function once DAT mode is disabled.

Rework memcpy_real() to overcome these shortcomings. As result,
data copying (which is primarily reading out a crashed system
memory by a user process) is executed on a regular stack with
enabled interrupts. Also, use of memcpy_real_buf swap buffer
becomes unnecessary and the swapping is eliminated.

The above is achieved by using a fixed virtual address range
that spans a single page and remaps that page repeatedly when
memcpy_real() is called for a particular physical address.

Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

arch/s390/boot/startup.c

@@ -17,6 +17,7 @@
 
 unsigned long __bootdata_preserved(__kaslr_offset);
 unsigned long __bootdata_preserved(__abs_lowcore);
+unsigned long __bootdata_preserved(__memcpy_real_area);
 unsigned long __bootdata(__amode31_base);
 unsigned long __bootdata_preserved(VMALLOC_START);
 unsigned long __bootdata_preserved(VMALLOC_END);
@@ -182,7 +183,9 @@ static void setup_kernel_memory_layout(void)
 	/* force vmalloc and modules below kasan shadow */
 	vmax = min(vmax, KASAN_SHADOW_START);
 #endif
-	__abs_lowcore = round_down(vmax - ABS_LOWCORE_MAP_SIZE, sizeof(struct lowcore));
+	__memcpy_real_area = round_down(vmax - PAGE_SIZE, PAGE_SIZE);
+	__abs_lowcore = round_down(__memcpy_real_area - ABS_LOWCORE_MAP_SIZE,
+				   sizeof(struct lowcore));
 	MODULES_END = round_down(__abs_lowcore, _SEGMENT_SIZE);
 	MODULES_VADDR = MODULES_END - MODULES_LEN;
 	VMALLOC_END = MODULES_VADDR;

arch/s390/include/asm/maccess.h

@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef __ASM_S390_MACCESS_H
+#define __ASM_S390_MACCESS_H
+
+#include <linux/types.h>
+
+struct iov_iter;
+
+extern unsigned long __memcpy_real_area;
+void memcpy_real_init(void);
+size_t memcpy_real_iter(struct iov_iter *iter, unsigned long src, size_t count);
+int memcpy_real(void *dest, unsigned long src, size_t count);
+
+#endif /* __ASM_S390_MACCESS_H */

arch/s390/include/asm/pgtable.h

@@ -1780,6 +1780,7 @@ extern void vmem_remove_mapping(unsigned long start, unsigned long size);
 extern int __vmem_map_4k_page(unsigned long addr, unsigned long phys, pgprot_t prot, bool alloc);
 extern int vmem_map_4k_page(unsigned long addr, unsigned long phys, pgprot_t prot);
 extern void vmem_unmap_4k_page(unsigned long addr);
+extern pte_t *vmem_get_alloc_pte(unsigned long addr, bool alloc);
 extern int s390_enable_sie(void);
 extern int s390_enable_skey(void);
 extern void s390_reset_cmma(struct mm_struct *mm);

arch/s390/include/asm/processor.h

@@ -306,8 +306,6 @@ static __always_inline void __noreturn disabled_wait(void)
 
 #define ARCH_LOW_ADDRESS_LIMIT	0x7fffffffUL
 
-extern int memcpy_real(void *, unsigned long, size_t);
-
 extern int s390_isolate_bp(void);
 extern int s390_isolate_bp_guest(void);
 

arch/s390/kernel/crash_dump.c

@@ -21,6 +21,7 @@
 #include <asm/elf.h>
 #include <asm/ipl.h>
 #include <asm/sclp.h>
+#include <asm/maccess.h>
 
 #define PTR_ADD(x, y) (((char *) (x)) + ((unsigned long) (y)))
 #define PTR_SUB(x, y) (((char *) (x)) - ((unsigned long) (y)))
@@ -53,8 +54,6 @@ struct save_area {
 };
 
 static LIST_HEAD(dump_save_areas);
-static DEFINE_MUTEX(memcpy_real_mutex);
-static char memcpy_real_buf[PAGE_SIZE];
 
 /*
  * Allocate a save area
@@ -116,26 +115,6 @@ void __init save_area_add_vxrs(struct save_area *sa, __vector128 *vxrs)
 	memcpy(sa->vxrs_high, vxrs + 16, 16 * sizeof(__vector128));
 }
 
-static size_t copy_to_iter_real(struct iov_iter *iter, unsigned long src, size_t count)
-{
-	size_t len, copied, res = 0;
-
-	mutex_lock(&memcpy_real_mutex);
-	while (count) {
-		len = min(PAGE_SIZE, count);
-		if (memcpy_real(memcpy_real_buf, src, len))
-			break;
-		copied = copy_to_iter(memcpy_real_buf, len, iter);
-		count -= copied;
-		src += copied;
-		res += copied;
-		if (copied < len)
-			break;
-	}
-	mutex_unlock(&memcpy_real_mutex);
-	return res;
-}
-
 size_t copy_oldmem_iter(struct iov_iter *iter, unsigned long src, size_t count)
 {
 	size_t len, copied, res = 0;
@@ -156,7 +135,7 @@ size_t copy_oldmem_iter(struct iov_iter *iter, unsigned long src, size_t count)
 			} else {
 				len = count;
 			}
-			copied = copy_to_iter_real(iter, src, len);
+			copied = memcpy_real_iter(iter, src, len);
 		}
 		count -= copied;
 		src += copied;

arch/s390/kernel/setup.c

@@ -74,6 +74,7 @@
 #include <asm/alternative.h>
 #include <asm/nospec-branch.h>
 #include <asm/mem_detect.h>
+#include <asm/maccess.h>
 #include <asm/uv.h>
 #include <asm/asm-offsets.h>
 #include "entry.h"
@@ -1050,7 +1051,7 @@ void __init setup_arch(char **cmdline_p)
 	 * Create kernel page tables and switch to virtual addressing.
 	 */
         paging_init();
-
+	memcpy_real_init();
 	/*
 	 * After paging_init created the kernel page table, the new PSWs
 	 * in lowcore can now run with DAT enabled.

arch/s390/mm/maccess.c

@@ -12,12 +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>
 
+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)
 {
 	unsigned long aligned, offset, count;
@@ -77,75 +82,55 @@ 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;
+	memcpy_real_ptep = vmem_get_alloc_pte(__memcpy_real_area, true);
+	if (!memcpy_real_ptep)
+		panic("Couldn't setup memcpy real area");
 }
 
-static unsigned long __no_sanitize_address _memcpy_real(unsigned long dest,
-							unsigned long src,
-							unsigned long count)
+size_t memcpy_real_iter(struct iov_iter *iter, unsigned long src, size_t count)
 {
-	int irqs_disabled, rc;
-	unsigned long flags;
+	size_t len, copied, res = 0;
+	unsigned long phys, offset;
+	void *chunk;
+	pte_t pte;
 
-	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;
+	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;
+	}
+	return res;
 }
 
-/*
- * Copy memory in real mode (kernel to kernel)
- */
 int memcpy_real(void *dest, 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;
+	struct iov_iter iter;
+	struct kvec kvec;
 
-	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;
-	}
-	/*
-	 * 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);
+	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;
 }
 
 /*

arch/s390/mm/vmem.c

@@ -567,7 +567,7 @@ int vmem_add_mapping(unsigned long start, unsigned long size)
  * while traversing is an error, since the function is expected to be
  * called against virtual regions reserverd for 4KB mappings only.
  */
-static pte_t *vmem_get_alloc_pte(unsigned long addr, bool alloc)
+pte_t *vmem_get_alloc_pte(unsigned long addr, bool alloc)
 {
 	pte_t *ptep = NULL;
 	pgd_t *pgd;

drivers/s390/char/zcore.c

@@ -30,6 +30,7 @@
 #include <asm/checksum.h>
 #include <asm/os_info.h>
 #include <asm/switch_to.h>
+#include <asm/maccess.h>
 #include "sclp.h"
 
 #define TRACE(x...) debug_sprintf_event(zcore_dbf, 1, x)