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 union register_pair _dst, _src; 83 int rc = -EFAULT; 84 85 _dst.even = (unsigned long) dest; 86 _dst.odd = (unsigned long) count; 87 _src.even = (unsigned long) src; 88 _src.odd = (unsigned long) count; 89 asm volatile ( 90 "0: mvcle %[dst],%[src],0\n" 91 "1: jo 0b\n" 92 " lhi %[rc],0\n" 93 "2:\n" 94 EX_TABLE(1b,2b) 95 : [rc] "+&d" (rc), [dst] "+&d" (_dst.pair), [src] "+&d" (_src.pair) 96 : : "cc", "memory"); 97 return rc; 98 } 99 100 static unsigned long __no_sanitize_address _memcpy_real(unsigned long dest, 101 unsigned long src, 102 unsigned long count) 103 { 104 int irqs_disabled, rc; 105 unsigned long flags; 106 107 if (!count) 108 return 0; 109 flags = arch_local_irq_save(); 110 irqs_disabled = arch_irqs_disabled_flags(flags); 111 if (!irqs_disabled) 112 trace_hardirqs_off(); 113 __arch_local_irq_stnsm(0xf8); // disable DAT 114 rc = __memcpy_real((void *) dest, (void *) src, (size_t) count); 115 if (flags & PSW_MASK_DAT) 116 __arch_local_irq_stosm(0x04); // enable DAT 117 if (!irqs_disabled) 118 trace_hardirqs_on(); 119 __arch_local_irq_ssm(flags); 120 return rc; 121 } 122 123 /* 124 * Copy memory in real mode (kernel to kernel) 125 */ 126 int memcpy_real(void *dest, void *src, size_t count) 127 { 128 unsigned long _dest = (unsigned long)dest; 129 unsigned long _src = (unsigned long)src; 130 unsigned long _count = (unsigned long)count; 131 int rc; 132 133 if (S390_lowcore.nodat_stack != 0) { 134 preempt_disable(); 135 rc = call_on_stack(3, S390_lowcore.nodat_stack, 136 unsigned long, _memcpy_real, 137 unsigned long, _dest, 138 unsigned long, _src, 139 unsigned long, _count); 140 preempt_enable(); 141 return rc; 142 } 143 /* 144 * This is a really early memcpy_real call, the stacks are 145 * not set up yet. Just call _memcpy_real on the early boot 146 * stack 147 */ 148 return _memcpy_real(_dest, _src, _count); 149 } 150 151 /* 152 * Copy memory in absolute mode (kernel to kernel) 153 */ 154 void memcpy_absolute(void *dest, void *src, size_t count) 155 { 156 unsigned long cr0, flags, prefix; 157 158 flags = arch_local_irq_save(); 159 __ctl_store(cr0, 0, 0); 160 __ctl_clear_bit(0, 28); /* disable lowcore protection */ 161 prefix = store_prefix(); 162 if (prefix) { 163 local_mcck_disable(); 164 set_prefix(0); 165 memcpy(dest, src, count); 166 set_prefix(prefix); 167 local_mcck_enable(); 168 } else { 169 memcpy(dest, src, count); 170 } 171 __ctl_load(cr0, 0, 0); 172 arch_local_irq_restore(flags); 173 } 174 175 /* 176 * Copy memory from kernel (real) to user (virtual) 177 */ 178 int copy_to_user_real(void __user *dest, void *src, unsigned long count) 179 { 180 int offs = 0, size, rc; 181 char *buf; 182 183 buf = (char *) __get_free_page(GFP_KERNEL); 184 if (!buf) 185 return -ENOMEM; 186 rc = -EFAULT; 187 while (offs < count) { 188 size = min(PAGE_SIZE, count - offs); 189 if (memcpy_real(buf, src + offs, size)) 190 goto out; 191 if (copy_to_user(dest + offs, buf, size)) 192 goto out; 193 offs += size; 194 } 195 rc = 0; 196 out: 197 free_page((unsigned long) buf); 198 return rc; 199 } 200 201 /* 202 * Check if physical address is within prefix or zero page 203 */ 204 static int is_swapped(unsigned long addr) 205 { 206 unsigned long lc; 207 int cpu; 208 209 if (addr < sizeof(struct lowcore)) 210 return 1; 211 for_each_online_cpu(cpu) { 212 lc = (unsigned long) lowcore_ptr[cpu]; 213 if (addr > lc + sizeof(struct lowcore) - 1 || addr < lc) 214 continue; 215 return 1; 216 } 217 return 0; 218 } 219 220 /* 221 * Convert a physical pointer for /dev/mem access 222 * 223 * For swapped prefix pages a new buffer is returned that contains a copy of 224 * the absolute memory. The buffer size is maximum one page large. 225 */ 226 void *xlate_dev_mem_ptr(phys_addr_t addr) 227 { 228 void *bounce = (void *) addr; 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, (void *) addr, 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 *buf) 248 { 249 if ((void *) addr != buf) 250 free_page((unsigned long) buf); 251 } 252