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 while (size) { 66 copied = s390_kernel_write_odd(tmp, src, size); 67 tmp += copied; 68 src += copied; 69 size -= copied; 70 } 71 spin_unlock_irqrestore(&s390_kernel_write_lock, flags); 72 73 return dst; 74 } 75 76 static int __no_sanitize_address __memcpy_real(void *dest, void *src, size_t count) 77 { 78 register unsigned long _dest asm("2") = (unsigned long) dest; 79 register unsigned long _len1 asm("3") = (unsigned long) count; 80 register unsigned long _src asm("4") = (unsigned long) src; 81 register unsigned long _len2 asm("5") = (unsigned long) count; 82 int rc = -EFAULT; 83 84 asm volatile ( 85 "0: mvcle %1,%2,0x0\n" 86 "1: jo 0b\n" 87 " lhi %0,0x0\n" 88 "2:\n" 89 EX_TABLE(1b,2b) 90 : "+d" (rc), "+d" (_dest), "+d" (_src), "+d" (_len1), 91 "+d" (_len2), "=m" (*((long *) dest)) 92 : "m" (*((long *) src)) 93 : "cc", "memory"); 94 return rc; 95 } 96 97 static unsigned long __no_sanitize_address _memcpy_real(unsigned long dest, 98 unsigned long src, 99 unsigned long count) 100 { 101 int irqs_disabled, rc; 102 unsigned long flags; 103 104 if (!count) 105 return 0; 106 flags = arch_local_irq_save(); 107 irqs_disabled = arch_irqs_disabled_flags(flags); 108 if (!irqs_disabled) 109 trace_hardirqs_off(); 110 __arch_local_irq_stnsm(0xf8); // disable DAT 111 rc = __memcpy_real((void *) dest, (void *) src, (size_t) count); 112 if (flags & PSW_MASK_DAT) 113 __arch_local_irq_stosm(0x04); // enable DAT 114 if (!irqs_disabled) 115 trace_hardirqs_on(); 116 __arch_local_irq_ssm(flags); 117 return rc; 118 } 119 120 /* 121 * Copy memory in real mode (kernel to kernel) 122 */ 123 int memcpy_real(void *dest, void *src, size_t count) 124 { 125 int rc; 126 127 if (S390_lowcore.nodat_stack != 0) { 128 preempt_disable(); 129 rc = CALL_ON_STACK(_memcpy_real, S390_lowcore.nodat_stack, 3, 130 dest, src, count); 131 preempt_enable(); 132 return rc; 133 } 134 /* 135 * This is a really early memcpy_real call, the stacks are 136 * not set up yet. Just call _memcpy_real on the early boot 137 * stack 138 */ 139 return _memcpy_real((unsigned long) dest,(unsigned long) src, 140 (unsigned long) count); 141 } 142 143 /* 144 * Copy memory in absolute mode (kernel to kernel) 145 */ 146 void memcpy_absolute(void *dest, void *src, size_t count) 147 { 148 unsigned long cr0, flags, prefix; 149 150 flags = arch_local_irq_save(); 151 __ctl_store(cr0, 0, 0); 152 __ctl_clear_bit(0, 28); /* disable lowcore protection */ 153 prefix = store_prefix(); 154 if (prefix) { 155 local_mcck_disable(); 156 set_prefix(0); 157 memcpy(dest, src, count); 158 set_prefix(prefix); 159 local_mcck_enable(); 160 } else { 161 memcpy(dest, src, count); 162 } 163 __ctl_load(cr0, 0, 0); 164 arch_local_irq_restore(flags); 165 } 166 167 /* 168 * Copy memory from kernel (real) to user (virtual) 169 */ 170 int copy_to_user_real(void __user *dest, void *src, unsigned long count) 171 { 172 int offs = 0, size, rc; 173 char *buf; 174 175 buf = (char *) __get_free_page(GFP_KERNEL); 176 if (!buf) 177 return -ENOMEM; 178 rc = -EFAULT; 179 while (offs < count) { 180 size = min(PAGE_SIZE, count - offs); 181 if (memcpy_real(buf, src + offs, size)) 182 goto out; 183 if (copy_to_user(dest + offs, buf, size)) 184 goto out; 185 offs += size; 186 } 187 rc = 0; 188 out: 189 free_page((unsigned long) buf); 190 return rc; 191 } 192 193 /* 194 * Check if physical address is within prefix or zero page 195 */ 196 static int is_swapped(unsigned long addr) 197 { 198 unsigned long lc; 199 int cpu; 200 201 if (addr < sizeof(struct lowcore)) 202 return 1; 203 for_each_online_cpu(cpu) { 204 lc = (unsigned long) lowcore_ptr[cpu]; 205 if (addr > lc + sizeof(struct lowcore) - 1 || addr < lc) 206 continue; 207 return 1; 208 } 209 return 0; 210 } 211 212 /* 213 * Convert a physical pointer for /dev/mem access 214 * 215 * For swapped prefix pages a new buffer is returned that contains a copy of 216 * the absolute memory. The buffer size is maximum one page large. 217 */ 218 void *xlate_dev_mem_ptr(phys_addr_t addr) 219 { 220 void *bounce = (void *) addr; 221 unsigned long size; 222 223 get_online_cpus(); 224 preempt_disable(); 225 if (is_swapped(addr)) { 226 size = PAGE_SIZE - (addr & ~PAGE_MASK); 227 bounce = (void *) __get_free_page(GFP_ATOMIC); 228 if (bounce) 229 memcpy_absolute(bounce, (void *) addr, size); 230 } 231 preempt_enable(); 232 put_online_cpus(); 233 return bounce; 234 } 235 236 /* 237 * Free converted buffer for /dev/mem access (if necessary) 238 */ 239 void unxlate_dev_mem_ptr(phys_addr_t addr, void *buf) 240 { 241 if ((void *) addr != buf) 242 free_page((unsigned long) buf); 243 } 244