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 void notrace s390_kernel_write(void *dst, const void *src, size_t size) 59 { 60 unsigned long flags; 61 long copied; 62 63 spin_lock_irqsave(&s390_kernel_write_lock, flags); 64 while (size) { 65 copied = s390_kernel_write_odd(dst, src, size); 66 dst += copied; 67 src += copied; 68 size -= copied; 69 } 70 spin_unlock_irqrestore(&s390_kernel_write_lock, flags); 71 } 72 73 static int __no_sanitize_address __memcpy_real(void *dest, void *src, size_t count) 74 { 75 register unsigned long _dest asm("2") = (unsigned long) dest; 76 register unsigned long _len1 asm("3") = (unsigned long) count; 77 register unsigned long _src asm("4") = (unsigned long) src; 78 register unsigned long _len2 asm("5") = (unsigned long) count; 79 int rc = -EFAULT; 80 81 asm volatile ( 82 "0: mvcle %1,%2,0x0\n" 83 "1: jo 0b\n" 84 " lhi %0,0x0\n" 85 "2:\n" 86 EX_TABLE(1b,2b) 87 : "+d" (rc), "+d" (_dest), "+d" (_src), "+d" (_len1), 88 "+d" (_len2), "=m" (*((long *) dest)) 89 : "m" (*((long *) src)) 90 : "cc", "memory"); 91 return rc; 92 } 93 94 static unsigned long __no_sanitize_address _memcpy_real(unsigned long dest, 95 unsigned long src, 96 unsigned long count) 97 { 98 int irqs_disabled, rc; 99 unsigned long flags; 100 101 if (!count) 102 return 0; 103 flags = arch_local_irq_save(); 104 irqs_disabled = arch_irqs_disabled_flags(flags); 105 if (!irqs_disabled) 106 trace_hardirqs_off(); 107 __arch_local_irq_stnsm(0xf8); // disable DAT 108 rc = __memcpy_real((void *) dest, (void *) src, (size_t) count); 109 if (flags & PSW_MASK_DAT) 110 __arch_local_irq_stosm(0x04); // enable DAT 111 if (!irqs_disabled) 112 trace_hardirqs_on(); 113 __arch_local_irq_ssm(flags); 114 return rc; 115 } 116 117 /* 118 * Copy memory in real mode (kernel to kernel) 119 */ 120 int memcpy_real(void *dest, void *src, size_t count) 121 { 122 int rc; 123 124 if (S390_lowcore.nodat_stack != 0) { 125 preempt_disable(); 126 rc = CALL_ON_STACK(_memcpy_real, S390_lowcore.nodat_stack, 3, 127 dest, src, count); 128 preempt_enable(); 129 return rc; 130 } 131 /* 132 * This is a really early memcpy_real call, the stacks are 133 * not set up yet. Just call _memcpy_real on the early boot 134 * stack 135 */ 136 return _memcpy_real((unsigned long) dest,(unsigned long) src, 137 (unsigned long) count); 138 } 139 140 /* 141 * Copy memory in absolute mode (kernel to kernel) 142 */ 143 void memcpy_absolute(void *dest, void *src, size_t count) 144 { 145 unsigned long cr0, flags, prefix; 146 147 flags = arch_local_irq_save(); 148 __ctl_store(cr0, 0, 0); 149 __ctl_clear_bit(0, 28); /* disable lowcore protection */ 150 prefix = store_prefix(); 151 if (prefix) { 152 local_mcck_disable(); 153 set_prefix(0); 154 memcpy(dest, src, count); 155 set_prefix(prefix); 156 local_mcck_enable(); 157 } else { 158 memcpy(dest, src, count); 159 } 160 __ctl_load(cr0, 0, 0); 161 arch_local_irq_restore(flags); 162 } 163 164 /* 165 * Copy memory from kernel (real) to user (virtual) 166 */ 167 int copy_to_user_real(void __user *dest, void *src, unsigned long count) 168 { 169 int offs = 0, size, rc; 170 char *buf; 171 172 buf = (char *) __get_free_page(GFP_KERNEL); 173 if (!buf) 174 return -ENOMEM; 175 rc = -EFAULT; 176 while (offs < count) { 177 size = min(PAGE_SIZE, count - offs); 178 if (memcpy_real(buf, src + offs, size)) 179 goto out; 180 if (copy_to_user(dest + offs, buf, size)) 181 goto out; 182 offs += size; 183 } 184 rc = 0; 185 out: 186 free_page((unsigned long) buf); 187 return rc; 188 } 189 190 /* 191 * Check if physical address is within prefix or zero page 192 */ 193 static int is_swapped(unsigned long addr) 194 { 195 unsigned long lc; 196 int cpu; 197 198 if (addr < sizeof(struct lowcore)) 199 return 1; 200 for_each_online_cpu(cpu) { 201 lc = (unsigned long) lowcore_ptr[cpu]; 202 if (addr > lc + sizeof(struct lowcore) - 1 || addr < lc) 203 continue; 204 return 1; 205 } 206 return 0; 207 } 208 209 /* 210 * Convert a physical pointer for /dev/mem access 211 * 212 * For swapped prefix pages a new buffer is returned that contains a copy of 213 * the absolute memory. The buffer size is maximum one page large. 214 */ 215 void *xlate_dev_mem_ptr(phys_addr_t addr) 216 { 217 void *bounce = (void *) addr; 218 unsigned long size; 219 220 get_online_cpus(); 221 preempt_disable(); 222 if (is_swapped(addr)) { 223 size = PAGE_SIZE - (addr & ~PAGE_MASK); 224 bounce = (void *) __get_free_page(GFP_ATOMIC); 225 if (bounce) 226 memcpy_absolute(bounce, (void *) addr, size); 227 } 228 preempt_enable(); 229 put_online_cpus(); 230 return bounce; 231 } 232 233 /* 234 * Free converted buffer for /dev/mem access (if necessary) 235 */ 236 void unxlate_dev_mem_ptr(phys_addr_t addr, void *buf) 237 { 238 if ((void *) addr != buf) 239 free_page((unsigned long) buf); 240 } 241