xref: /linux/arch/s390/mm/maccess.c (revision 3f0a50f345f78183f6e9b39c2f45ca5dcaa511ca)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Access kernel memory without faulting -- s390 specific implementation.
4  *
5  * Copyright IBM Corp. 2009, 2015
6  *
7  */
8 
9 #include <linux/uaccess.h>
10 #include <linux/kernel.h>
11 #include <linux/types.h>
12 #include <linux/errno.h>
13 #include <linux/gfp.h>
14 #include <linux/cpu.h>
15 #include <asm/asm-extable.h>
16 #include <asm/ctl_reg.h>
17 #include <asm/io.h>
18 #include <asm/stacktrace.h>
19 
20 static notrace long s390_kernel_write_odd(void *dst, const void *src, size_t size)
21 {
22 	unsigned long aligned, offset, count;
23 	char tmp[8];
24 
25 	aligned = (unsigned long) dst & ~7UL;
26 	offset = (unsigned long) dst & 7UL;
27 	size = min(8UL - offset, size);
28 	count = size - 1;
29 	asm volatile(
30 		"	bras	1,0f\n"
31 		"	mvc	0(1,%4),0(%5)\n"
32 		"0:	mvc	0(8,%3),0(%0)\n"
33 		"	ex	%1,0(1)\n"
34 		"	lg	%1,0(%3)\n"
35 		"	lra	%0,0(%0)\n"
36 		"	sturg	%1,%0\n"
37 		: "+&a" (aligned), "+&a" (count), "=m" (tmp)
38 		: "a" (&tmp), "a" (&tmp[offset]), "a" (src)
39 		: "cc", "memory", "1");
40 	return size;
41 }
42 
43 /*
44  * s390_kernel_write - write to kernel memory bypassing DAT
45  * @dst: destination address
46  * @src: source address
47  * @size: number of bytes to copy
48  *
49  * This function writes to kernel memory bypassing DAT and possible page table
50  * write protection. It writes to the destination using the sturg instruction.
51  * Therefore we have a read-modify-write sequence: the function reads eight
52  * bytes from destination at an eight byte boundary, modifies the bytes
53  * requested and writes the result back in a loop.
54  */
55 static DEFINE_SPINLOCK(s390_kernel_write_lock);
56 
57 notrace void *s390_kernel_write(void *dst, const void *src, size_t size)
58 {
59 	void *tmp = dst;
60 	unsigned long flags;
61 	long copied;
62 
63 	spin_lock_irqsave(&s390_kernel_write_lock, flags);
64 	if (!(flags & PSW_MASK_DAT)) {
65 		memcpy(dst, src, size);
66 	} else {
67 		while (size) {
68 			copied = s390_kernel_write_odd(tmp, src, size);
69 			tmp += copied;
70 			src += copied;
71 			size -= copied;
72 		}
73 	}
74 	spin_unlock_irqrestore(&s390_kernel_write_lock, flags);
75 
76 	return dst;
77 }
78 
79 static int __no_sanitize_address __memcpy_real(void *dest, void *src, size_t count)
80 {
81 	union register_pair _dst, _src;
82 	int rc = -EFAULT;
83 
84 	_dst.even = (unsigned long) dest;
85 	_dst.odd  = (unsigned long) count;
86 	_src.even = (unsigned long) src;
87 	_src.odd  = (unsigned long) count;
88 	asm volatile (
89 		"0:	mvcle	%[dst],%[src],0\n"
90 		"1:	jo	0b\n"
91 		"	lhi	%[rc],0\n"
92 		"2:\n"
93 		EX_TABLE(1b,2b)
94 		: [rc] "+&d" (rc), [dst] "+&d" (_dst.pair), [src] "+&d" (_src.pair)
95 		: : "cc", "memory");
96 	return rc;
97 }
98 
99 static unsigned long __no_sanitize_address _memcpy_real(unsigned long dest,
100 							unsigned long src,
101 							unsigned long count)
102 {
103 	int irqs_disabled, rc;
104 	unsigned long flags;
105 
106 	if (!count)
107 		return 0;
108 	flags = arch_local_irq_save();
109 	irqs_disabled = arch_irqs_disabled_flags(flags);
110 	if (!irqs_disabled)
111 		trace_hardirqs_off();
112 	__arch_local_irq_stnsm(0xf8); // disable DAT
113 	rc = __memcpy_real((void *) dest, (void *) src, (size_t) count);
114 	if (flags & PSW_MASK_DAT)
115 		__arch_local_irq_stosm(0x04); // enable DAT
116 	if (!irqs_disabled)
117 		trace_hardirqs_on();
118 	__arch_local_irq_ssm(flags);
119 	return rc;
120 }
121 
122 /*
123  * Copy memory in real mode (kernel to kernel)
124  */
125 int memcpy_real(void *dest, unsigned long src, size_t count)
126 {
127 	unsigned long _dest  = (unsigned long)dest;
128 	unsigned long _src   = (unsigned long)src;
129 	unsigned long _count = (unsigned long)count;
130 	int rc;
131 
132 	if (S390_lowcore.nodat_stack != 0) {
133 		preempt_disable();
134 		rc = call_on_stack(3, S390_lowcore.nodat_stack,
135 				   unsigned long, _memcpy_real,
136 				   unsigned long, _dest,
137 				   unsigned long, _src,
138 				   unsigned long, _count);
139 		preempt_enable();
140 		return rc;
141 	}
142 	/*
143 	 * This is a really early memcpy_real call, the stacks are
144 	 * not set up yet. Just call _memcpy_real on the early boot
145 	 * stack
146 	 */
147 	return _memcpy_real(_dest, _src, _count);
148 }
149 
150 /*
151  * Copy memory in absolute mode (kernel to kernel)
152  */
153 void memcpy_absolute(void *dest, void *src, size_t count)
154 {
155 	unsigned long cr0, flags, prefix;
156 
157 	flags = arch_local_irq_save();
158 	__ctl_store(cr0, 0, 0);
159 	__ctl_clear_bit(0, 28); /* disable lowcore protection */
160 	prefix = store_prefix();
161 	if (prefix) {
162 		local_mcck_disable();
163 		set_prefix(0);
164 		memcpy(dest, src, count);
165 		set_prefix(prefix);
166 		local_mcck_enable();
167 	} else {
168 		memcpy(dest, src, count);
169 	}
170 	__ctl_load(cr0, 0, 0);
171 	arch_local_irq_restore(flags);
172 }
173 
174 /*
175  * Copy memory from kernel (real) to user (virtual)
176  */
177 int copy_to_user_real(void __user *dest, unsigned long src, unsigned long count)
178 {
179 	int offs = 0, size, rc;
180 	char *buf;
181 
182 	buf = (char *) __get_free_page(GFP_KERNEL);
183 	if (!buf)
184 		return -ENOMEM;
185 	rc = -EFAULT;
186 	while (offs < count) {
187 		size = min(PAGE_SIZE, count - offs);
188 		if (memcpy_real(buf, src + offs, size))
189 			goto out;
190 		if (copy_to_user(dest + offs, buf, size))
191 			goto out;
192 		offs += size;
193 	}
194 	rc = 0;
195 out:
196 	free_page((unsigned long) buf);
197 	return rc;
198 }
199 
200 /*
201  * Check if physical address is within prefix or zero page
202  */
203 static int is_swapped(phys_addr_t addr)
204 {
205 	phys_addr_t lc;
206 	int cpu;
207 
208 	if (addr < sizeof(struct lowcore))
209 		return 1;
210 	for_each_online_cpu(cpu) {
211 		lc = virt_to_phys(lowcore_ptr[cpu]);
212 		if (addr > lc + sizeof(struct lowcore) - 1 || addr < lc)
213 			continue;
214 		return 1;
215 	}
216 	return 0;
217 }
218 
219 /*
220  * Convert a physical pointer for /dev/mem access
221  *
222  * For swapped prefix pages a new buffer is returned that contains a copy of
223  * the absolute memory. The buffer size is maximum one page large.
224  */
225 void *xlate_dev_mem_ptr(phys_addr_t addr)
226 {
227 	void *ptr = phys_to_virt(addr);
228 	void *bounce = ptr;
229 	unsigned long size;
230 
231 	cpus_read_lock();
232 	preempt_disable();
233 	if (is_swapped(addr)) {
234 		size = PAGE_SIZE - (addr & ~PAGE_MASK);
235 		bounce = (void *) __get_free_page(GFP_ATOMIC);
236 		if (bounce)
237 			memcpy_absolute(bounce, ptr, size);
238 	}
239 	preempt_enable();
240 	cpus_read_unlock();
241 	return bounce;
242 }
243 
244 /*
245  * Free converted buffer for /dev/mem access (if necessary)
246  */
247 void unxlate_dev_mem_ptr(phys_addr_t addr, void *ptr)
248 {
249 	if (addr != virt_to_phys(ptr))
250 		free_page((unsigned long)ptr);
251 }
252