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 <linux/uio.h>
16 #include <asm/asm-extable.h>
17 #include <asm/ctl_reg.h>
18 #include <asm/io.h>
19 #include <asm/abs_lowcore.h>
20 #include <asm/stacktrace.h>
21 #include <asm/maccess.h>
22 
23 unsigned long __bootdata_preserved(__memcpy_real_area);
24 static __ro_after_init pte_t *memcpy_real_ptep;
25 static DEFINE_MUTEX(memcpy_real_mutex);
26 
s390_kernel_write_odd(void * dst,const void * src,size_t size)27 static notrace long s390_kernel_write_odd(void *dst, const void *src, size_t size)
28 {
29 	unsigned long aligned, offset, count;
30 	char tmp[8];
31 
32 	aligned = (unsigned long) dst & ~7UL;
33 	offset = (unsigned long) dst & 7UL;
34 	size = min(8UL - offset, size);
35 	count = size - 1;
36 	asm volatile(
37 		"	bras	1,0f\n"
38 		"	mvc	0(1,%4),0(%5)\n"
39 		"0:	mvc	0(8,%3),0(%0)\n"
40 		"	ex	%1,0(1)\n"
41 		"	lg	%1,0(%3)\n"
42 		"	lra	%0,0(%0)\n"
43 		"	sturg	%1,%0\n"
44 		: "+&a" (aligned), "+&a" (count), "=m" (tmp)
45 		: "a" (&tmp), "a" (&tmp[offset]), "a" (src)
46 		: "cc", "memory", "1");
47 	return size;
48 }
49 
50 /*
51  * s390_kernel_write - write to kernel memory bypassing DAT
52  * @dst: destination address
53  * @src: source address
54  * @size: number of bytes to copy
55  *
56  * This function writes to kernel memory bypassing DAT and possible page table
57  * write protection. It writes to the destination using the sturg instruction.
58  * Therefore we have a read-modify-write sequence: the function reads eight
59  * bytes from destination at an eight byte boundary, modifies the bytes
60  * requested and writes the result back in a loop.
61  */
62 static DEFINE_SPINLOCK(s390_kernel_write_lock);
63 
s390_kernel_write(void * dst,const void * src,size_t size)64 notrace void *s390_kernel_write(void *dst, const void *src, size_t size)
65 {
66 	void *tmp = dst;
67 	unsigned long flags;
68 	long copied;
69 
70 	spin_lock_irqsave(&s390_kernel_write_lock, flags);
71 	if (!(flags & PSW_MASK_DAT)) {
72 		memcpy(dst, src, size);
73 	} else {
74 		while (size) {
75 			copied = s390_kernel_write_odd(tmp, src, size);
76 			tmp += copied;
77 			src += copied;
78 			size -= copied;
79 		}
80 	}
81 	spin_unlock_irqrestore(&s390_kernel_write_lock, flags);
82 
83 	return dst;
84 }
85 
memcpy_real_init(void)86 void __init memcpy_real_init(void)
87 {
88 	memcpy_real_ptep = vmem_get_alloc_pte(__memcpy_real_area, true);
89 	if (!memcpy_real_ptep)
90 		panic("Couldn't setup memcpy real area");
91 }
92 
memcpy_real_iter(struct iov_iter * iter,unsigned long src,size_t count)93 size_t memcpy_real_iter(struct iov_iter *iter, unsigned long src, size_t count)
94 {
95 	size_t len, copied, res = 0;
96 	unsigned long phys, offset;
97 	void *chunk;
98 	pte_t pte;
99 
100 	while (count) {
101 		phys = src & PAGE_MASK;
102 		offset = src & ~PAGE_MASK;
103 		chunk = (void *)(__memcpy_real_area + offset);
104 		len = min(count, PAGE_SIZE - offset);
105 		pte = mk_pte_phys(phys, PAGE_KERNEL_RO);
106 
107 		mutex_lock(&memcpy_real_mutex);
108 		if (pte_val(pte) != pte_val(*memcpy_real_ptep)) {
109 			__ptep_ipte(__memcpy_real_area, memcpy_real_ptep, 0, 0, IPTE_GLOBAL);
110 			set_pte(memcpy_real_ptep, pte);
111 		}
112 		copied = copy_to_iter(chunk, len, iter);
113 		mutex_unlock(&memcpy_real_mutex);
114 
115 		count -= copied;
116 		src += copied;
117 		res += copied;
118 		if (copied < len)
119 			break;
120 	}
121 	return res;
122 }
123 
memcpy_real(void * dest,unsigned long src,size_t count)124 int memcpy_real(void *dest, unsigned long src, size_t count)
125 {
126 	struct iov_iter iter;
127 	struct kvec kvec;
128 
129 	kvec.iov_base = dest;
130 	kvec.iov_len = count;
131 	iov_iter_kvec(&iter, WRITE, &kvec, 1, count);
132 	if (memcpy_real_iter(&iter, src, count) < count)
133 		return -EFAULT;
134 	return 0;
135 }
136 
137 /*
138  * Find CPU that owns swapped prefix page
139  */
get_swapped_owner(phys_addr_t addr)140 static int get_swapped_owner(phys_addr_t addr)
141 {
142 	phys_addr_t lc;
143 	int cpu;
144 
145 	for_each_online_cpu(cpu) {
146 		lc = virt_to_phys(lowcore_ptr[cpu]);
147 		if (addr > lc + sizeof(struct lowcore) - 1 || addr < lc)
148 			continue;
149 		return cpu;
150 	}
151 	return -1;
152 }
153 
154 /*
155  * Convert a physical pointer for /dev/mem access
156  *
157  * For swapped prefix pages a new buffer is returned that contains a copy of
158  * the absolute memory. The buffer size is maximum one page large.
159  */
xlate_dev_mem_ptr(phys_addr_t addr)160 void *xlate_dev_mem_ptr(phys_addr_t addr)
161 {
162 	void *ptr = phys_to_virt(addr);
163 	void *bounce = ptr;
164 	struct lowcore *abs_lc;
165 	unsigned long flags;
166 	unsigned long size;
167 	int this_cpu, cpu;
168 
169 	cpus_read_lock();
170 	this_cpu = get_cpu();
171 	if (addr >= sizeof(struct lowcore)) {
172 		cpu = get_swapped_owner(addr);
173 		if (cpu < 0)
174 			goto out;
175 	}
176 	bounce = (void *)__get_free_page(GFP_ATOMIC);
177 	if (!bounce)
178 		goto out;
179 	size = PAGE_SIZE - (addr & ~PAGE_MASK);
180 	if (addr < sizeof(struct lowcore)) {
181 		abs_lc = get_abs_lowcore(&flags);
182 		ptr = (void *)abs_lc + addr;
183 		memcpy(bounce, ptr, size);
184 		put_abs_lowcore(abs_lc, flags);
185 	} else if (cpu == this_cpu) {
186 		ptr = (void *)(addr - virt_to_phys(lowcore_ptr[cpu]));
187 		memcpy(bounce, ptr, size);
188 	} else {
189 		memcpy(bounce, ptr, size);
190 	}
191 out:
192 	put_cpu();
193 	cpus_read_unlock();
194 	return bounce;
195 }
196 
197 /*
198  * Free converted buffer for /dev/mem access (if necessary)
199  */
unxlate_dev_mem_ptr(phys_addr_t addr,void * ptr)200 void unxlate_dev_mem_ptr(phys_addr_t addr, void *ptr)
201 {
202 	if (addr != virt_to_phys(ptr))
203 		free_page((unsigned long)ptr);
204 }
205