1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright IBM Corp. 2005, 2011 4 * 5 * Author(s): Rolf Adelsberger, 6 * Michael Holzheu <holzheu@linux.vnet.ibm.com> 7 */ 8 9 #include <linux/device.h> 10 #include <linux/mm.h> 11 #include <linux/kexec.h> 12 #include <linux/delay.h> 13 #include <linux/reboot.h> 14 #include <linux/ftrace.h> 15 #include <linux/debug_locks.h> 16 #include <asm/cio.h> 17 #include <asm/setup.h> 18 #include <asm/smp.h> 19 #include <asm/ipl.h> 20 #include <asm/diag.h> 21 #include <asm/elf.h> 22 #include <asm/asm-offsets.h> 23 #include <asm/cacheflush.h> 24 #include <asm/abs_lowcore.h> 25 #include <asm/os_info.h> 26 #include <asm/set_memory.h> 27 #include <asm/stacktrace.h> 28 #include <asm/switch_to.h> 29 #include <asm/nmi.h> 30 #include <asm/sclp.h> 31 32 typedef void (*relocate_kernel_t)(kimage_entry_t *, unsigned long, 33 unsigned long); 34 35 extern const unsigned char relocate_kernel[]; 36 extern const unsigned long long relocate_kernel_len; 37 38 #ifdef CONFIG_CRASH_DUMP 39 40 /* 41 * Reset the system, copy boot CPU registers to absolute zero, 42 * and jump to the kdump image 43 */ 44 static void __do_machine_kdump(void *image) 45 { 46 int (*start_kdump)(int); 47 unsigned long prefix; 48 49 /* store_status() saved the prefix register to lowcore */ 50 prefix = (unsigned long) S390_lowcore.prefixreg_save_area; 51 52 /* Now do the reset */ 53 s390_reset_system(); 54 55 /* 56 * Copy dump CPU store status info to absolute zero. 57 * This need to be done *after* s390_reset_system set the 58 * prefix register of this CPU to zero 59 */ 60 memcpy(absolute_pointer(__LC_FPREGS_SAVE_AREA), 61 (void *)(prefix + __LC_FPREGS_SAVE_AREA), 512); 62 63 __load_psw_mask(PSW_MASK_BASE | PSW_DEFAULT_KEY | PSW_MASK_EA | PSW_MASK_BA); 64 start_kdump = (void *)((struct kimage *) image)->start; 65 start_kdump(1); 66 67 /* Die if start_kdump returns */ 68 disabled_wait(); 69 } 70 71 /* 72 * Start kdump: create a LGR log entry, store status of all CPUs and 73 * branch to __do_machine_kdump. 74 */ 75 static noinline void __machine_kdump(void *image) 76 { 77 struct mcesa *mcesa; 78 union ctlreg2 cr2_old, cr2_new; 79 int this_cpu, cpu; 80 81 lgr_info_log(); 82 /* Get status of the other CPUs */ 83 this_cpu = smp_find_processor_id(stap()); 84 for_each_online_cpu(cpu) { 85 if (cpu == this_cpu) 86 continue; 87 if (smp_store_status(cpu)) 88 continue; 89 } 90 /* Store status of the boot CPU */ 91 mcesa = __va(S390_lowcore.mcesad & MCESA_ORIGIN_MASK); 92 if (MACHINE_HAS_VX) 93 save_vx_regs((__vector128 *) mcesa->vector_save_area); 94 if (MACHINE_HAS_GS) { 95 __ctl_store(cr2_old.val, 2, 2); 96 cr2_new = cr2_old; 97 cr2_new.gse = 1; 98 __ctl_load(cr2_new.val, 2, 2); 99 save_gs_cb((struct gs_cb *) mcesa->guarded_storage_save_area); 100 __ctl_load(cr2_old.val, 2, 2); 101 } 102 /* 103 * To create a good backchain for this CPU in the dump store_status 104 * is passed the address of a function. The address is saved into 105 * the PSW save area of the boot CPU and the function is invoked as 106 * a tail call of store_status. The backchain in the dump will look 107 * like this: 108 * restart_int_handler -> __machine_kexec -> __do_machine_kdump 109 * The call to store_status() will not return. 110 */ 111 store_status(__do_machine_kdump, image); 112 } 113 114 static unsigned long do_start_kdump(unsigned long addr) 115 { 116 struct kimage *image = (struct kimage *) addr; 117 int (*start_kdump)(int) = (void *)image->start; 118 int rc; 119 120 __arch_local_irq_stnsm(0xfb); /* disable DAT */ 121 rc = start_kdump(0); 122 __arch_local_irq_stosm(0x04); /* enable DAT */ 123 return rc; 124 } 125 126 #endif /* CONFIG_CRASH_DUMP */ 127 128 /* 129 * Check if kdump checksums are valid: We call purgatory with parameter "0" 130 */ 131 static bool kdump_csum_valid(struct kimage *image) 132 { 133 #ifdef CONFIG_CRASH_DUMP 134 int rc; 135 136 preempt_disable(); 137 rc = call_on_stack(1, S390_lowcore.nodat_stack, unsigned long, do_start_kdump, 138 unsigned long, (unsigned long)image); 139 preempt_enable(); 140 return rc == 0; 141 #else 142 return false; 143 #endif 144 } 145 146 #ifdef CONFIG_CRASH_DUMP 147 148 void crash_free_reserved_phys_range(unsigned long begin, unsigned long end) 149 { 150 unsigned long addr, size; 151 152 for (addr = begin; addr < end; addr += PAGE_SIZE) 153 free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT)); 154 size = begin - crashk_res.start; 155 if (size) 156 os_info_crashkernel_add(crashk_res.start, size); 157 else 158 os_info_crashkernel_add(0, 0); 159 } 160 161 static void crash_protect_pages(int protect) 162 { 163 unsigned long size; 164 165 if (!crashk_res.end) 166 return; 167 size = resource_size(&crashk_res); 168 if (protect) 169 set_memory_ro(crashk_res.start, size >> PAGE_SHIFT); 170 else 171 set_memory_rw(crashk_res.start, size >> PAGE_SHIFT); 172 } 173 174 void arch_kexec_protect_crashkres(void) 175 { 176 crash_protect_pages(1); 177 } 178 179 void arch_kexec_unprotect_crashkres(void) 180 { 181 crash_protect_pages(0); 182 } 183 184 #endif 185 186 /* 187 * Give back memory to hypervisor before new kdump is loaded 188 */ 189 static int machine_kexec_prepare_kdump(void) 190 { 191 #ifdef CONFIG_CRASH_DUMP 192 if (MACHINE_IS_VM) 193 diag10_range(PFN_DOWN(crashk_res.start), 194 PFN_DOWN(crashk_res.end - crashk_res.start + 1)); 195 return 0; 196 #else 197 return -EINVAL; 198 #endif 199 } 200 201 int machine_kexec_prepare(struct kimage *image) 202 { 203 void *reboot_code_buffer; 204 205 if (image->type == KEXEC_TYPE_CRASH) 206 return machine_kexec_prepare_kdump(); 207 208 /* We don't support anything but the default image type for now. */ 209 if (image->type != KEXEC_TYPE_DEFAULT) 210 return -EINVAL; 211 212 /* Get the destination where the assembler code should be copied to.*/ 213 reboot_code_buffer = (void *) page_to_phys(image->control_code_page); 214 215 /* Then copy it */ 216 memcpy(reboot_code_buffer, relocate_kernel, relocate_kernel_len); 217 return 0; 218 } 219 220 void machine_kexec_cleanup(struct kimage *image) 221 { 222 } 223 224 void arch_crash_save_vmcoreinfo(void) 225 { 226 struct lowcore *abs_lc; 227 228 VMCOREINFO_SYMBOL(lowcore_ptr); 229 VMCOREINFO_SYMBOL(high_memory); 230 VMCOREINFO_LENGTH(lowcore_ptr, NR_CPUS); 231 vmcoreinfo_append_str("SAMODE31=%lx\n", __samode31); 232 vmcoreinfo_append_str("EAMODE31=%lx\n", __eamode31); 233 vmcoreinfo_append_str("KERNELOFFSET=%lx\n", kaslr_offset()); 234 abs_lc = get_abs_lowcore(); 235 abs_lc->vmcore_info = paddr_vmcoreinfo_note(); 236 put_abs_lowcore(abs_lc); 237 } 238 239 void machine_shutdown(void) 240 { 241 } 242 243 void machine_crash_shutdown(struct pt_regs *regs) 244 { 245 set_os_info_reipl_block(); 246 } 247 248 /* 249 * Do normal kexec 250 */ 251 static void __do_machine_kexec(void *data) 252 { 253 unsigned long diag308_subcode; 254 relocate_kernel_t data_mover; 255 struct kimage *image = data; 256 257 s390_reset_system(); 258 data_mover = (relocate_kernel_t) page_to_phys(image->control_code_page); 259 260 __arch_local_irq_stnsm(0xfb); /* disable DAT - avoid no-execute */ 261 /* Call the moving routine */ 262 diag308_subcode = DIAG308_CLEAR_RESET; 263 if (sclp.has_iplcc) 264 diag308_subcode |= DIAG308_FLAG_EI; 265 (*data_mover)(&image->head, image->start, diag308_subcode); 266 267 /* Die if kexec returns */ 268 disabled_wait(); 269 } 270 271 /* 272 * Reset system and call either kdump or normal kexec 273 */ 274 static void __machine_kexec(void *data) 275 { 276 pfault_fini(); 277 tracing_off(); 278 debug_locks_off(); 279 #ifdef CONFIG_CRASH_DUMP 280 if (((struct kimage *) data)->type == KEXEC_TYPE_CRASH) 281 __machine_kdump(data); 282 #endif 283 __do_machine_kexec(data); 284 } 285 286 /* 287 * Do either kdump or normal kexec. In case of kdump we first ask 288 * purgatory, if kdump checksums are valid. 289 */ 290 void machine_kexec(struct kimage *image) 291 { 292 if (image->type == KEXEC_TYPE_CRASH && !kdump_csum_valid(image)) 293 return; 294 tracer_disable(); 295 smp_send_stop(); 296 smp_call_ipl_cpu(__machine_kexec, image); 297 } 298