xref: /linux/arch/s390/mm/maccess.c (revision 87c9c16317882dd6dbbc07e349bc3223e14f3244)
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  *   Author(s): Heiko Carstens <heiko.carstens@de.ibm.com>,
8  *
9  */
10 
11 #include <linux/uaccess.h>
12 #include <linux/kernel.h>
13 #include <linux/types.h>
14 #include <linux/errno.h>
15 #include <linux/gfp.h>
16 #include <linux/cpu.h>
17 #include <asm/ctl_reg.h>
18 #include <asm/io.h>
19 #include <asm/stacktrace.h>
20 
21 static notrace long s390_kernel_write_odd(void *dst, const void *src, size_t size)
22 {
23 	unsigned long aligned, offset, count;
24 	char tmp[8];
25 
26 	aligned = (unsigned long) dst & ~7UL;
27 	offset = (unsigned long) dst & 7UL;
28 	size = min(8UL - offset, size);
29 	count = size - 1;
30 	asm volatile(
31 		"	bras	1,0f\n"
32 		"	mvc	0(1,%4),0(%5)\n"
33 		"0:	mvc	0(8,%3),0(%0)\n"
34 		"	ex	%1,0(1)\n"
35 		"	lg	%1,0(%3)\n"
36 		"	lra	%0,0(%0)\n"
37 		"	sturg	%1,%0\n"
38 		: "+&a" (aligned), "+&a" (count), "=m" (tmp)
39 		: "a" (&tmp), "a" (&tmp[offset]), "a" (src)
40 		: "cc", "memory", "1");
41 	return size;
42 }
43 
44 /*
45  * s390_kernel_write - write to kernel memory bypassing DAT
46  * @dst: destination address
47  * @src: source address
48  * @size: number of bytes to copy
49  *
50  * This function writes to kernel memory bypassing DAT and possible page table
51  * write protection. It writes to the destination using the sturg instruction.
52  * Therefore we have a read-modify-write sequence: the function reads eight
53  * bytes from destination at an eight byte boundary, modifies the bytes
54  * requested and writes the result back in a loop.
55  */
56 static DEFINE_SPINLOCK(s390_kernel_write_lock);
57 
58 notrace void *s390_kernel_write(void *dst, const void *src, size_t size)
59 {
60 	void *tmp = dst;
61 	unsigned long flags;
62 	long copied;
63 
64 	spin_lock_irqsave(&s390_kernel_write_lock, flags);
65 	if (!(flags & PSW_MASK_DAT)) {
66 		memcpy(dst, src, size);
67 	} else {
68 		while (size) {
69 			copied = s390_kernel_write_odd(tmp, src, size);
70 			tmp += copied;
71 			src += copied;
72 			size -= copied;
73 		}
74 	}
75 	spin_unlock_irqrestore(&s390_kernel_write_lock, flags);
76 
77 	return dst;
78 }
79 
80 static int __no_sanitize_address __memcpy_real(void *dest, void *src, size_t count)
81 {
82 	register unsigned long _dest asm("2") = (unsigned long) dest;
83 	register unsigned long _len1 asm("3") = (unsigned long) count;
84 	register unsigned long _src  asm("4") = (unsigned long) src;
85 	register unsigned long _len2 asm("5") = (unsigned long) count;
86 	int rc = -EFAULT;
87 
88 	asm volatile (
89 		"0:	mvcle	%1,%2,0x0\n"
90 		"1:	jo	0b\n"
91 		"	lhi	%0,0x0\n"
92 		"2:\n"
93 		EX_TABLE(1b,2b)
94 		: "+d" (rc), "+d" (_dest), "+d" (_src), "+d" (_len1),
95 		  "+d" (_len2), "=m" (*((long *) dest))
96 		: "m" (*((long *) src))
97 		: "cc", "memory");
98 	return rc;
99 }
100 
101 static unsigned long __no_sanitize_address _memcpy_real(unsigned long dest,
102 							unsigned long src,
103 							unsigned long count)
104 {
105 	int irqs_disabled, rc;
106 	unsigned long flags;
107 
108 	if (!count)
109 		return 0;
110 	flags = arch_local_irq_save();
111 	irqs_disabled = arch_irqs_disabled_flags(flags);
112 	if (!irqs_disabled)
113 		trace_hardirqs_off();
114 	__arch_local_irq_stnsm(0xf8); // disable DAT
115 	rc = __memcpy_real((void *) dest, (void *) src, (size_t) count);
116 	if (flags & PSW_MASK_DAT)
117 		__arch_local_irq_stosm(0x04); // enable DAT
118 	if (!irqs_disabled)
119 		trace_hardirqs_on();
120 	__arch_local_irq_ssm(flags);
121 	return rc;
122 }
123 
124 /*
125  * Copy memory in real mode (kernel to kernel)
126  */
127 int memcpy_real(void *dest, void *src, size_t count)
128 {
129 	int rc;
130 
131 	if (S390_lowcore.nodat_stack != 0) {
132 		preempt_disable();
133 		rc = CALL_ON_STACK(_memcpy_real, S390_lowcore.nodat_stack, 3,
134 				   dest, src, count);
135 		preempt_enable();
136 		return rc;
137 	}
138 	/*
139 	 * This is a really early memcpy_real call, the stacks are
140 	 * not set up yet. Just call _memcpy_real on the early boot
141 	 * stack
142 	 */
143 	return _memcpy_real((unsigned long) dest,(unsigned long) src,
144 			    (unsigned long) count);
145 }
146 
147 /*
148  * Copy memory in absolute mode (kernel to kernel)
149  */
150 void memcpy_absolute(void *dest, void *src, size_t count)
151 {
152 	unsigned long cr0, flags, prefix;
153 
154 	flags = arch_local_irq_save();
155 	__ctl_store(cr0, 0, 0);
156 	__ctl_clear_bit(0, 28); /* disable lowcore protection */
157 	prefix = store_prefix();
158 	if (prefix) {
159 		local_mcck_disable();
160 		set_prefix(0);
161 		memcpy(dest, src, count);
162 		set_prefix(prefix);
163 		local_mcck_enable();
164 	} else {
165 		memcpy(dest, src, count);
166 	}
167 	__ctl_load(cr0, 0, 0);
168 	arch_local_irq_restore(flags);
169 }
170 
171 /*
172  * Copy memory from kernel (real) to user (virtual)
173  */
174 int copy_to_user_real(void __user *dest, void *src, unsigned long count)
175 {
176 	int offs = 0, size, rc;
177 	char *buf;
178 
179 	buf = (char *) __get_free_page(GFP_KERNEL);
180 	if (!buf)
181 		return -ENOMEM;
182 	rc = -EFAULT;
183 	while (offs < count) {
184 		size = min(PAGE_SIZE, count - offs);
185 		if (memcpy_real(buf, src + offs, size))
186 			goto out;
187 		if (copy_to_user(dest + offs, buf, size))
188 			goto out;
189 		offs += size;
190 	}
191 	rc = 0;
192 out:
193 	free_page((unsigned long) buf);
194 	return rc;
195 }
196 
197 /*
198  * Check if physical address is within prefix or zero page
199  */
200 static int is_swapped(unsigned long addr)
201 {
202 	unsigned long lc;
203 	int cpu;
204 
205 	if (addr < sizeof(struct lowcore))
206 		return 1;
207 	for_each_online_cpu(cpu) {
208 		lc = (unsigned long) lowcore_ptr[cpu];
209 		if (addr > lc + sizeof(struct lowcore) - 1 || addr < lc)
210 			continue;
211 		return 1;
212 	}
213 	return 0;
214 }
215 
216 /*
217  * Convert a physical pointer for /dev/mem access
218  *
219  * For swapped prefix pages a new buffer is returned that contains a copy of
220  * the absolute memory. The buffer size is maximum one page large.
221  */
222 void *xlate_dev_mem_ptr(phys_addr_t addr)
223 {
224 	void *bounce = (void *) addr;
225 	unsigned long size;
226 
227 	get_online_cpus();
228 	preempt_disable();
229 	if (is_swapped(addr)) {
230 		size = PAGE_SIZE - (addr & ~PAGE_MASK);
231 		bounce = (void *) __get_free_page(GFP_ATOMIC);
232 		if (bounce)
233 			memcpy_absolute(bounce, (void *) addr, size);
234 	}
235 	preempt_enable();
236 	put_online_cpus();
237 	return bounce;
238 }
239 
240 /*
241  * Free converted buffer for /dev/mem access (if necessary)
242  */
243 void unxlate_dev_mem_ptr(phys_addr_t addr, void *buf)
244 {
245 	if ((void *) addr != buf)
246 		free_page((unsigned long) buf);
247 }
248