powerpc/kexec/core_64.c

0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  * PPC64 code to handle Linux booting another kernel.
0004  *
0005  * Copyright (C) 2004-2005, IBM Corp.
0006  *
0007  * Created by: Milton D Miller II
0008  */
0009
0010
0011 #include <linux/kexec.h>
0012 #include <linux/smp.h>
0013 #include <linux/thread_info.h>
0014 #include <linux/init_task.h>
0015 #include <linux/errno.h>
0016 #include <linux/kernel.h>
0017 #include <linux/cpu.h>
0018 #include <linux/hardirq.h>
0019 #include <linux/of.h>
0020
0021 #include <asm/page.h>
0022 #include <asm/current.h>
0023 #include <asm/machdep.h>
0024 #include <asm/cacheflush.h>
0025 #include <asm/firmware.h>
0026 #include <asm/paca.h>
0027 #include <asm/mmu.h>
0028 #include <asm/sections.h>   /* _end */
0029 #include <asm/smp.h>
0030 #include <asm/hw_breakpoint.h>
0031 #include <asm/svm.h>
0032 #include <asm/ultravisor.h>
0033
0034 int machine_kexec_prepare(struct kimage *image)
0035 {
0036     int i;
0037     unsigned long begin, end;   /* limits of segment */
0038     unsigned long low, high;    /* limits of blocked memory range */
0039     struct device_node *node;
0040     const unsigned long *basep;
0041     const unsigned int *sizep;
0042
0043     /*
0044      * Since we use the kernel fault handlers and paging code to
0045      * handle the virtual mode, we must make sure no destination
0046      * overlaps kernel static data or bss.
0047      */
0048     for (i = 0; i < image->nr_segments; i++)
0049         if (image->segment[i].mem < __pa(_end))
0050             return -ETXTBSY;
0051
0052     /* We also should not overwrite the tce tables */
0053     for_each_node_by_type(node, "pci") {
0054         basep = of_get_property(node, "linux,tce-base", NULL);
0055         sizep = of_get_property(node, "linux,tce-size", NULL);
0056         if (basep == NULL || sizep == NULL)
0057             continue;
0058
0059         low = *basep;
0060         high = low + (*sizep);
0061
0062         for (i = 0; i < image->nr_segments; i++) {
0063             begin = image->segment[i].mem;
0064             end = begin + image->segment[i].memsz;
0065
0066             if ((begin < high) && (end > low)) {
0067                 of_node_put(node);
0068                 return -ETXTBSY;
0069             }
0070         }
0071     }
0072
0073     return 0;
0074 }
0075
0076 /* Called during kexec sequence with MMU off */
0077 static notrace void copy_segments(unsigned long ind)
0078 {
0079     unsigned long entry;
0080     unsigned long *ptr;
0081     void *dest;
0082     void *addr;
0083
0084     /*
0085      * We rely on kexec_load to create a lists that properly
0086      * initializes these pointers before they are used.
0087      * We will still crash if the list is wrong, but at least
0088      * the compiler will be quiet.
0089      */
0090     ptr = NULL;
0091     dest = NULL;
0092
0093     for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
0094         addr = __va(entry & PAGE_MASK);
0095
0096         switch (entry & IND_FLAGS) {
0097         case IND_DESTINATION:
0098             dest = addr;
0099             break;
0100         case IND_INDIRECTION:
0101             ptr = addr;
0102             break;
0103         case IND_SOURCE:
0104             copy_page(dest, addr);
0105             dest += PAGE_SIZE;
0106         }
0107     }
0108 }
0109
0110 /* Called during kexec sequence with MMU off */
0111 notrace void kexec_copy_flush(struct kimage *image)
0112 {
0113     long i, nr_segments = image->nr_segments;
0114     struct  kexec_segment ranges[KEXEC_SEGMENT_MAX];
0115
0116     /* save the ranges on the stack to efficiently flush the icache */
0117     memcpy(ranges, image->segment, sizeof(ranges));
0118
0119     /*
0120      * After this call we may not use anything allocated in dynamic
0121      * memory, including *image.
0122      *
0123      * Only globals and the stack are allowed.
0124      */
0125     copy_segments(image->head);
0126
0127     /*
0128      * we need to clear the icache for all dest pages sometime,
0129      * including ones that were in place on the original copy
0130      */
0131     for (i = 0; i < nr_segments; i++)
0132         flush_icache_range((unsigned long)__va(ranges[i].mem),
0133             (unsigned long)__va(ranges[i].mem + ranges[i].memsz));
0134 }
0135
0136 #ifdef CONFIG_SMP
0137
0138 static int kexec_all_irq_disabled = 0;
0139
0140 static void kexec_smp_down(void *arg)
0141 {
0142     local_irq_disable();
0143     hard_irq_disable();
0144
0145     mb(); /* make sure our irqs are disabled before we say they are */
0146     get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
0147     while(kexec_all_irq_disabled == 0)
0148         cpu_relax();
0149     mb(); /* make sure all irqs are disabled before this */
0150     hw_breakpoint_disable();
0151     /*
0152      * Now every CPU has IRQs off, we can clear out any pending
0153      * IPIs and be sure that no more will come in after this.
0154      */
0155     if (ppc_md.kexec_cpu_down)
0156         ppc_md.kexec_cpu_down(0, 1);
0157
0158     reset_sprs();
0159
0160     kexec_smp_wait();
0161     /* NOTREACHED */
0162 }
0163
0164 static void kexec_prepare_cpus_wait(int wait_state)
0165 {
0166     int my_cpu, i, notified=-1;
0167
0168     hw_breakpoint_disable();
0169     my_cpu = get_cpu();
0170     /* Make sure each CPU has at least made it to the state we need.
0171      *
0172      * FIXME: There is a (slim) chance of a problem if not all of the CPUs
0173      * are correctly onlined.  If somehow we start a CPU on boot with RTAS
0174      * start-cpu, but somehow that CPU doesn't write callin_cpu_map[] in
0175      * time, the boot CPU will timeout.  If it does eventually execute
0176      * stuff, the secondary will start up (paca_ptrs[]->cpu_start was
0177      * written) and get into a peculiar state.
0178      * If the platform supports smp_ops->take_timebase(), the secondary CPU
0179      * will probably be spinning in there.  If not (i.e. pseries), the
0180      * secondary will continue on and try to online itself/idle/etc. If it
0181      * survives that, we need to find these
0182      * possible-but-not-online-but-should-be CPUs and chaperone them into
0183      * kexec_smp_wait().
0184      */
0185     for_each_online_cpu(i) {
0186         if (i == my_cpu)
0187             continue;
0188
0189         while (paca_ptrs[i]->kexec_state < wait_state) {
0190             barrier();
0191             if (i != notified) {
0192                 printk(KERN_INFO "kexec: waiting for cpu %d "
0193                        "(physical %d) to enter %i state\n",
0194                        i, paca_ptrs[i]->hw_cpu_id, wait_state);
0195                 notified = i;
0196             }
0197         }
0198     }
0199     mb();
0200 }
0201
0202 /*
0203  * We need to make sure each present CPU is online.  The next kernel will scan
0204  * the device tree and assume primary threads are online and query secondary
0205  * threads via RTAS to online them if required.  If we don't online primary
0206  * threads, they will be stuck.  However, we also online secondary threads as we
0207  * may be using 'cede offline'.  In this case RTAS doesn't see the secondary
0208  * threads as offline -- and again, these CPUs will be stuck.
0209  *
0210  * So, we online all CPUs that should be running, including secondary threads.
0211  */
0212 static void wake_offline_cpus(void)
0213 {
0214     int cpu = 0;
0215
0216     for_each_present_cpu(cpu) {
0217         if (!cpu_online(cpu)) {
0218             printk(KERN_INFO "kexec: Waking offline cpu %d.\n",
0219                    cpu);
0220             WARN_ON(add_cpu(cpu));
0221         }
0222     }
0223 }
0224
0225 static void kexec_prepare_cpus(void)
0226 {
0227     wake_offline_cpus();
0228     smp_call_function(kexec_smp_down, NULL, /* wait */0);
0229     local_irq_disable();
0230     hard_irq_disable();
0231
0232     mb(); /* make sure IRQs are disabled before we say they are */
0233     get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
0234
0235     kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF);
0236     /* we are sure every CPU has IRQs off at this point */
0237     kexec_all_irq_disabled = 1;
0238
0239     /*
0240      * Before removing MMU mappings make sure all CPUs have entered real
0241      * mode:
0242      */
0243     kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE);
0244
0245     /* after we tell the others to go down */
0246     if (ppc_md.kexec_cpu_down)
0247         ppc_md.kexec_cpu_down(0, 0);
0248
0249     put_cpu();
0250 }
0251
0252 #else /* ! SMP */
0253
0254 static void kexec_prepare_cpus(void)
0255 {
0256     /*
0257      * move the secondarys to us so that we can copy
0258      * the new kernel 0-0x100 safely
0259      *
0260      * do this if kexec in setup.c ?
0261      *
0262      * We need to release the cpus if we are ever going from an
0263      * UP to an SMP kernel.
0264      */
0265     smp_release_cpus();
0266     if (ppc_md.kexec_cpu_down)
0267         ppc_md.kexec_cpu_down(0, 0);
0268     local_irq_disable();
0269     hard_irq_disable();
0270 }
0271
0272 #endif /* SMP */
0273
0274 /*
0275  * kexec thread structure and stack.
0276  *
0277  * We need to make sure that this is 16384-byte aligned due to the
0278  * way process stacks are handled.  It also must be statically allocated
0279  * or allocated as part of the kimage, because everything else may be
0280  * overwritten when we copy the kexec image.  We piggyback on the
0281  * "init_task" linker section here to statically allocate a stack.
0282  *
0283  * We could use a smaller stack if we don't care about anything using
0284  * current, but that audit has not been performed.
0285  */
0286 static union thread_union kexec_stack __init_task_data =
0287     { };
0288
0289 /*
0290  * For similar reasons to the stack above, the kexecing CPU needs to be on a
0291  * static PACA; we switch to kexec_paca.
0292  */
0293 static struct paca_struct kexec_paca;
0294
0295 /* Our assembly helper, in misc_64.S */
0296 extern void kexec_sequence(void *newstack, unsigned long start,
0297                void *image, void *control,
0298                void (*clear_all)(void),
0299                bool copy_with_mmu_off) __noreturn;
0300
0301 /* too late to fail here */
0302 void default_machine_kexec(struct kimage *image)
0303 {
0304     bool copy_with_mmu_off;
0305
0306     /* prepare control code if any */
0307
0308     /*
0309         * If the kexec boot is the normal one, need to shutdown other cpus
0310         * into our wait loop and quiesce interrupts.
0311         * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
0312         * stopping other CPUs and collecting their pt_regs is done before
0313         * using debugger IPI.
0314         */
0315
0316     if (!kdump_in_progress())
0317         kexec_prepare_cpus();
0318
0319     printk("kexec: Starting switchover sequence.\n");
0320
0321     /* switch to a staticly allocated stack.  Based on irq stack code.
0322      * We setup preempt_count to avoid using VMX in memcpy.
0323      * XXX: the task struct will likely be invalid once we do the copy!
0324      */
0325     current_thread_info()->flags = 0;
0326     current_thread_info()->preempt_count = HARDIRQ_OFFSET;
0327
0328     /* We need a static PACA, too; copy this CPU's PACA over and switch to
0329      * it. Also poison per_cpu_offset and NULL lppaca to catch anyone using
0330      * non-static data.
0331      */
0332     memcpy(&kexec_paca, get_paca(), sizeof(struct paca_struct));
0333     kexec_paca.data_offset = 0xedeaddeadeeeeeeeUL;
0334 #ifdef CONFIG_PPC_PSERIES
0335     kexec_paca.lppaca_ptr = NULL;
0336 #endif
0337
0338     if (is_secure_guest() && !(image->preserve_context ||
0339                    image->type == KEXEC_TYPE_CRASH)) {
0340         uv_unshare_all_pages();
0341         printk("kexec: Unshared all shared pages.\n");
0342     }
0343
0344     paca_ptrs[kexec_paca.paca_index] = &kexec_paca;
0345
0346     setup_paca(&kexec_paca);
0347
0348     /*
0349      * The lppaca should be unregistered at this point so the HV won't
0350      * touch it. In the case of a crash, none of the lppacas are
0351      * unregistered so there is not much we can do about it here.
0352      */
0353
0354     /*
0355      * On Book3S, the copy must happen with the MMU off if we are either
0356      * using Radix page tables or we are not in an LPAR since we can
0357      * overwrite the page tables while copying.
0358      *
0359      * In an LPAR, we keep the MMU on otherwise we can't access beyond
0360      * the RMA. On BookE there is no real MMU off mode, so we have to
0361      * keep it enabled as well (but then we have bolted TLB entries).
0362      */
0363 #ifdef CONFIG_PPC_BOOK3E
0364     copy_with_mmu_off = false;
0365 #else
0366     copy_with_mmu_off = radix_enabled() ||
0367         !(firmware_has_feature(FW_FEATURE_LPAR) ||
0368           firmware_has_feature(FW_FEATURE_PS3_LV1));
0369 #endif
0370
0371     /* Some things are best done in assembly.  Finding globals with
0372      * a toc is easier in C, so pass in what we can.
0373      */
0374     kexec_sequence(&kexec_stack, image->start, image,
0375                page_address(image->control_code_page),
0376                mmu_cleanup_all, copy_with_mmu_off);
0377     /* NOTREACHED */
0378 }
0379
0380 #ifdef CONFIG_PPC_64S_HASH_MMU
0381 /* Values we need to export to the second kernel via the device tree. */
0382 static unsigned long htab_base;
0383 static unsigned long htab_size;
0384
0385 static struct property htab_base_prop = {
0386     .name = "linux,htab-base",
0387     .length = sizeof(unsigned long),
0388     .value = &htab_base,
0389 };
0390
0391 static struct property htab_size_prop = {
0392     .name = "linux,htab-size",
0393     .length = sizeof(unsigned long),
0394     .value = &htab_size,
0395 };
0396
0397 static int __init export_htab_values(void)
0398 {
0399     struct device_node *node;
0400
0401     /* On machines with no htab htab_address is NULL */
0402     if (!htab_address)
0403         return -ENODEV;
0404
0405     node = of_find_node_by_path("/chosen");
0406     if (!node)
0407         return -ENODEV;
0408
0409     /* remove any stale properties so ours can be found */
0410     of_remove_property(node, of_find_property(node, htab_base_prop.name, NULL));
0411     of_remove_property(node, of_find_property(node, htab_size_prop.name, NULL));
0412
0413     htab_base = cpu_to_be64(__pa(htab_address));
0414     of_add_property(node, &htab_base_prop);
0415     htab_size = cpu_to_be64(htab_size_bytes);
0416     of_add_property(node, &htab_size_prop);
0417
0418     of_node_put(node);
0419     return 0;
0420 }
0421 late_initcall(export_htab_values);
0422 #endif /* CONFIG_PPC_64S_HASH_MMU */