machine_kexec.c (7125B)
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/os_info.h> 25#include <asm/set_memory.h> 26#include <asm/stacktrace.h> 27#include <asm/switch_to.h> 28#include <asm/nmi.h> 29#include <asm/sclp.h> 30 31typedef void (*relocate_kernel_t)(kimage_entry_t *, unsigned long, 32 unsigned long); 33 34extern const unsigned char relocate_kernel[]; 35extern const unsigned long long relocate_kernel_len; 36 37#ifdef CONFIG_CRASH_DUMP 38 39/* 40 * Reset the system, copy boot CPU registers to absolute zero, 41 * and jump to the kdump image 42 */ 43static void __do_machine_kdump(void *image) 44{ 45 int (*start_kdump)(int); 46 unsigned long prefix; 47 48 /* store_status() saved the prefix register to lowcore */ 49 prefix = (unsigned long) S390_lowcore.prefixreg_save_area; 50 51 /* Now do the reset */ 52 s390_reset_system(); 53 54 /* 55 * Copy dump CPU store status info to absolute zero. 56 * This need to be done *after* s390_reset_system set the 57 * prefix register of this CPU to zero 58 */ 59 memcpy(absolute_pointer(__LC_FPREGS_SAVE_AREA), 60 (void *)(prefix + __LC_FPREGS_SAVE_AREA), 512); 61 62 __load_psw_mask(PSW_MASK_BASE | PSW_DEFAULT_KEY | PSW_MASK_EA | PSW_MASK_BA); 63 start_kdump = (void *)((struct kimage *) image)->start; 64 start_kdump(1); 65 66 /* Die if start_kdump returns */ 67 disabled_wait(); 68} 69 70/* 71 * Start kdump: create a LGR log entry, store status of all CPUs and 72 * branch to __do_machine_kdump. 73 */ 74static noinline void __machine_kdump(void *image) 75{ 76 struct mcesa *mcesa; 77 union ctlreg2 cr2_old, cr2_new; 78 int this_cpu, cpu; 79 80 lgr_info_log(); 81 /* Get status of the other CPUs */ 82 this_cpu = smp_find_processor_id(stap()); 83 for_each_online_cpu(cpu) { 84 if (cpu == this_cpu) 85 continue; 86 if (smp_store_status(cpu)) 87 continue; 88 } 89 /* Store status of the boot CPU */ 90 mcesa = __va(S390_lowcore.mcesad & MCESA_ORIGIN_MASK); 91 if (MACHINE_HAS_VX) 92 save_vx_regs((__vector128 *) mcesa->vector_save_area); 93 if (MACHINE_HAS_GS) { 94 __ctl_store(cr2_old.val, 2, 2); 95 cr2_new = cr2_old; 96 cr2_new.gse = 1; 97 __ctl_load(cr2_new.val, 2, 2); 98 save_gs_cb((struct gs_cb *) mcesa->guarded_storage_save_area); 99 __ctl_load(cr2_old.val, 2, 2); 100 } 101 /* 102 * To create a good backchain for this CPU in the dump store_status 103 * is passed the address of a function. The address is saved into 104 * the PSW save area of the boot CPU and the function is invoked as 105 * a tail call of store_status. The backchain in the dump will look 106 * like this: 107 * restart_int_handler -> __machine_kexec -> __do_machine_kdump 108 * The call to store_status() will not return. 109 */ 110 store_status(__do_machine_kdump, image); 111} 112 113static unsigned long do_start_kdump(unsigned long addr) 114{ 115 struct kimage *image = (struct kimage *) addr; 116 int (*start_kdump)(int) = (void *)image->start; 117 int rc; 118 119 __arch_local_irq_stnsm(0xfb); /* disable DAT */ 120 rc = start_kdump(0); 121 __arch_local_irq_stosm(0x04); /* enable DAT */ 122 return rc; 123} 124 125#endif /* CONFIG_CRASH_DUMP */ 126 127/* 128 * Check if kdump checksums are valid: We call purgatory with parameter "0" 129 */ 130static bool kdump_csum_valid(struct kimage *image) 131{ 132#ifdef CONFIG_CRASH_DUMP 133 int rc; 134 135 preempt_disable(); 136 rc = call_on_stack(1, S390_lowcore.nodat_stack, unsigned long, do_start_kdump, 137 unsigned long, (unsigned long)image); 138 preempt_enable(); 139 return rc == 0; 140#else 141 return false; 142#endif 143} 144 145#ifdef CONFIG_CRASH_DUMP 146 147void crash_free_reserved_phys_range(unsigned long begin, unsigned long end) 148{ 149 unsigned long addr, size; 150 151 for (addr = begin; addr < end; addr += PAGE_SIZE) 152 free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT)); 153 size = begin - crashk_res.start; 154 if (size) 155 os_info_crashkernel_add(crashk_res.start, size); 156 else 157 os_info_crashkernel_add(0, 0); 158} 159 160static void crash_protect_pages(int protect) 161{ 162 unsigned long size; 163 164 if (!crashk_res.end) 165 return; 166 size = resource_size(&crashk_res); 167 if (protect) 168 set_memory_ro(crashk_res.start, size >> PAGE_SHIFT); 169 else 170 set_memory_rw(crashk_res.start, size >> PAGE_SHIFT); 171} 172 173void arch_kexec_protect_crashkres(void) 174{ 175 crash_protect_pages(1); 176} 177 178void arch_kexec_unprotect_crashkres(void) 179{ 180 crash_protect_pages(0); 181} 182 183#endif 184 185/* 186 * Give back memory to hypervisor before new kdump is loaded 187 */ 188static int machine_kexec_prepare_kdump(void) 189{ 190#ifdef CONFIG_CRASH_DUMP 191 if (MACHINE_IS_VM) 192 diag10_range(PFN_DOWN(crashk_res.start), 193 PFN_DOWN(crashk_res.end - crashk_res.start + 1)); 194 return 0; 195#else 196 return -EINVAL; 197#endif 198} 199 200int machine_kexec_prepare(struct kimage *image) 201{ 202 void *reboot_code_buffer; 203 204 if (image->type == KEXEC_TYPE_CRASH) 205 return machine_kexec_prepare_kdump(); 206 207 /* We don't support anything but the default image type for now. */ 208 if (image->type != KEXEC_TYPE_DEFAULT) 209 return -EINVAL; 210 211 /* Get the destination where the assembler code should be copied to.*/ 212 reboot_code_buffer = (void *) page_to_phys(image->control_code_page); 213 214 /* Then copy it */ 215 memcpy(reboot_code_buffer, relocate_kernel, relocate_kernel_len); 216 return 0; 217} 218 219void machine_kexec_cleanup(struct kimage *image) 220{ 221} 222 223void arch_crash_save_vmcoreinfo(void) 224{ 225 VMCOREINFO_SYMBOL(lowcore_ptr); 226 VMCOREINFO_SYMBOL(high_memory); 227 VMCOREINFO_LENGTH(lowcore_ptr, NR_CPUS); 228 vmcoreinfo_append_str("SAMODE31=%lx\n", __samode31); 229 vmcoreinfo_append_str("EAMODE31=%lx\n", __eamode31); 230 vmcoreinfo_append_str("KERNELOFFSET=%lx\n", kaslr_offset()); 231 put_abs_lowcore(vmcore_info, paddr_vmcoreinfo_note()); 232} 233 234void machine_shutdown(void) 235{ 236} 237 238void machine_crash_shutdown(struct pt_regs *regs) 239{ 240 set_os_info_reipl_block(); 241} 242 243/* 244 * Do normal kexec 245 */ 246static void __do_machine_kexec(void *data) 247{ 248 unsigned long diag308_subcode; 249 relocate_kernel_t data_mover; 250 struct kimage *image = data; 251 252 s390_reset_system(); 253 data_mover = (relocate_kernel_t) page_to_phys(image->control_code_page); 254 255 __arch_local_irq_stnsm(0xfb); /* disable DAT - avoid no-execute */ 256 /* Call the moving routine */ 257 diag308_subcode = DIAG308_CLEAR_RESET; 258 if (sclp.has_iplcc) 259 diag308_subcode |= DIAG308_FLAG_EI; 260 (*data_mover)(&image->head, image->start, diag308_subcode); 261 262 /* Die if kexec returns */ 263 disabled_wait(); 264} 265 266/* 267 * Reset system and call either kdump or normal kexec 268 */ 269static void __machine_kexec(void *data) 270{ 271 pfault_fini(); 272 tracing_off(); 273 debug_locks_off(); 274#ifdef CONFIG_CRASH_DUMP 275 if (((struct kimage *) data)->type == KEXEC_TYPE_CRASH) 276 __machine_kdump(data); 277#endif 278 __do_machine_kexec(data); 279} 280 281/* 282 * Do either kdump or normal kexec. In case of kdump we first ask 283 * purgatory, if kdump checksums are valid. 284 */ 285void machine_kexec(struct kimage *image) 286{ 287 if (image->type == KEXEC_TYPE_CRASH && !kdump_csum_valid(image)) 288 return; 289 tracer_disable(); 290 smp_send_stop(); 291 smp_call_ipl_cpu(__machine_kexec, image); 292}