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 if (S390_lowcore.nodat_stack != 0) 123 return CALL_ON_STACK(_memcpy_real, S390_lowcore.nodat_stack, 124 3, dest, src, count); 125 /* 126 * This is a really early memcpy_real call, the stacks are 127 * not set up yet. Just call _memcpy_real on the early boot 128 * stack 129 */ 130 return _memcpy_real((unsigned long) dest,(unsigned long) src, 131 (unsigned long) count); 132 } 133 134 /* 135 * Copy memory in absolute mode (kernel to kernel) 136 */ 137 void memcpy_absolute(void *dest, void *src, size_t count) 138 { 139 unsigned long cr0, flags, prefix; 140 141 flags = arch_local_irq_save(); 142 __ctl_store(cr0, 0, 0); 143 __ctl_clear_bit(0, 28); /* disable lowcore protection */ 144 prefix = store_prefix(); 145 if (prefix) { 146 local_mcck_disable(); 147 set_prefix(0); 148 memcpy(dest, src, count); 149 set_prefix(prefix); 150 local_mcck_enable(); 151 } else { 152 memcpy(dest, src, count); 153 } 154 __ctl_load(cr0, 0, 0); 155 arch_local_irq_restore(flags); 156 } 157 158 /* 159 * Copy memory from kernel (real) to user (virtual) 160 */ 161 int copy_to_user_real(void __user *dest, void *src, unsigned long count) 162 { 163 int offs = 0, size, rc; 164 char *buf; 165 166 buf = (char *) __get_free_page(GFP_KERNEL); 167 if (!buf) 168 return -ENOMEM; 169 rc = -EFAULT; 170 while (offs < count) { 171 size = min(PAGE_SIZE, count - offs); 172 if (memcpy_real(buf, src + offs, size)) 173 goto out; 174 if (copy_to_user(dest + offs, buf, size)) 175 goto out; 176 offs += size; 177 } 178 rc = 0; 179 out: 180 free_page((unsigned long) buf); 181 return rc; 182 } 183 184 /* 185 * Check if physical address is within prefix or zero page 186 */ 187 static int is_swapped(unsigned long addr) 188 { 189 unsigned long lc; 190 int cpu; 191 192 if (addr < sizeof(struct lowcore)) 193 return 1; 194 for_each_online_cpu(cpu) { 195 lc = (unsigned long) lowcore_ptr[cpu]; 196 if (addr > lc + sizeof(struct lowcore) - 1 || addr < lc) 197 continue; 198 return 1; 199 } 200 return 0; 201 } 202 203 /* 204 * Convert a physical pointer for /dev/mem access 205 * 206 * For swapped prefix pages a new buffer is returned that contains a copy of 207 * the absolute memory. The buffer size is maximum one page large. 208 */ 209 void *xlate_dev_mem_ptr(phys_addr_t addr) 210 { 211 void *bounce = (void *) addr; 212 unsigned long size; 213 214 get_online_cpus(); 215 preempt_disable(); 216 if (is_swapped(addr)) { 217 size = PAGE_SIZE - (addr & ~PAGE_MASK); 218 bounce = (void *) __get_free_page(GFP_ATOMIC); 219 if (bounce) 220 memcpy_absolute(bounce, (void *) addr, size); 221 } 222 preempt_enable(); 223 put_online_cpus(); 224 return bounce; 225 } 226 227 /* 228 * Free converted buffer for /dev/mem access (if necessary) 229 */ 230 void unxlate_dev_mem_ptr(phys_addr_t addr, void *buf) 231 { 232 if ((void *) addr != buf) 233 free_page((unsigned long) buf); 234 } 235