Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  *   Machine check handler
0004  *
0005  *    Copyright IBM Corp. 2000, 2009
0006  *    Author(s): Ingo Adlung <adlung@de.ibm.com>,
0007  *       Martin Schwidefsky <schwidefsky@de.ibm.com>,
0008  *       Cornelia Huck <cornelia.huck@de.ibm.com>,
0009  */
0010 
0011 #include <linux/kernel_stat.h>
0012 #include <linux/init.h>
0013 #include <linux/errno.h>
0014 #include <linux/entry-common.h>
0015 #include <linux/hardirq.h>
0016 #include <linux/log2.h>
0017 #include <linux/kprobes.h>
0018 #include <linux/kmemleak.h>
0019 #include <linux/time.h>
0020 #include <linux/module.h>
0021 #include <linux/sched/signal.h>
0022 
0023 #include <linux/export.h>
0024 #include <asm/lowcore.h>
0025 #include <asm/smp.h>
0026 #include <asm/stp.h>
0027 #include <asm/cputime.h>
0028 #include <asm/nmi.h>
0029 #include <asm/crw.h>
0030 #include <asm/switch_to.h>
0031 #include <asm/ctl_reg.h>
0032 #include <asm/asm-offsets.h>
0033 #include <asm/pai.h>
0034 
0035 #include <linux/kvm_host.h>
0036 
0037 struct mcck_struct {
0038     unsigned int kill_task : 1;
0039     unsigned int channel_report : 1;
0040     unsigned int warning : 1;
0041     unsigned int stp_queue : 1;
0042     unsigned long mcck_code;
0043 };
0044 
0045 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
0046 static struct kmem_cache *mcesa_cache;
0047 static unsigned long mcesa_origin_lc;
0048 
0049 static inline int nmi_needs_mcesa(void)
0050 {
0051     return MACHINE_HAS_VX || MACHINE_HAS_GS;
0052 }
0053 
0054 static inline unsigned long nmi_get_mcesa_size(void)
0055 {
0056     if (MACHINE_HAS_GS)
0057         return MCESA_MAX_SIZE;
0058     return MCESA_MIN_SIZE;
0059 }
0060 
0061 /*
0062  * The initial machine check extended save area for the boot CPU.
0063  * It will be replaced on the boot CPU reinit with an allocated
0064  * structure. The structure is required for machine check happening
0065  * early in the boot process.
0066  */
0067 static struct mcesa boot_mcesa __aligned(MCESA_MAX_SIZE);
0068 
0069 void __init nmi_alloc_mcesa_early(u64 *mcesad)
0070 {
0071     if (!nmi_needs_mcesa())
0072         return;
0073     *mcesad = __pa(&boot_mcesa);
0074     if (MACHINE_HAS_GS)
0075         *mcesad |= ilog2(MCESA_MAX_SIZE);
0076 }
0077 
0078 static void __init nmi_alloc_cache(void)
0079 {
0080     unsigned long size;
0081 
0082     if (!nmi_needs_mcesa())
0083         return;
0084     size = nmi_get_mcesa_size();
0085     if (size > MCESA_MIN_SIZE)
0086         mcesa_origin_lc = ilog2(size);
0087     /* create slab cache for the machine-check-extended-save-areas */
0088     mcesa_cache = kmem_cache_create("nmi_save_areas", size, size, 0, NULL);
0089     if (!mcesa_cache)
0090         panic("Couldn't create nmi save area cache");
0091 }
0092 
0093 int __ref nmi_alloc_mcesa(u64 *mcesad)
0094 {
0095     unsigned long origin;
0096 
0097     *mcesad = 0;
0098     if (!nmi_needs_mcesa())
0099         return 0;
0100     if (!mcesa_cache)
0101         nmi_alloc_cache();
0102     origin = (unsigned long) kmem_cache_alloc(mcesa_cache, GFP_KERNEL);
0103     if (!origin)
0104         return -ENOMEM;
0105     /* The pointer is stored with mcesa_bits ORed in */
0106     kmemleak_not_leak((void *) origin);
0107     *mcesad = __pa(origin) | mcesa_origin_lc;
0108     return 0;
0109 }
0110 
0111 void nmi_free_mcesa(u64 *mcesad)
0112 {
0113     if (!nmi_needs_mcesa())
0114         return;
0115     kmem_cache_free(mcesa_cache, __va(*mcesad & MCESA_ORIGIN_MASK));
0116 }
0117 
0118 static notrace void s390_handle_damage(void)
0119 {
0120     smp_emergency_stop();
0121     disabled_wait();
0122     while (1);
0123 }
0124 NOKPROBE_SYMBOL(s390_handle_damage);
0125 
0126 /*
0127  * Main machine check handler function. Will be called with interrupts disabled
0128  * and machine checks enabled.
0129  */
0130 void __s390_handle_mcck(void)
0131 {
0132     struct mcck_struct mcck;
0133 
0134     /*
0135      * Disable machine checks and get the current state of accumulated
0136      * machine checks. Afterwards delete the old state and enable machine
0137      * checks again.
0138      */
0139     local_mcck_disable();
0140     mcck = *this_cpu_ptr(&cpu_mcck);
0141     memset(this_cpu_ptr(&cpu_mcck), 0, sizeof(mcck));
0142     local_mcck_enable();
0143 
0144     if (mcck.channel_report)
0145         crw_handle_channel_report();
0146     /*
0147      * A warning may remain for a prolonged period on the bare iron.
0148      * (actually until the machine is powered off, or the problem is gone)
0149      * So we just stop listening for the WARNING MCH and avoid continuously
0150      * being interrupted.  One caveat is however, that we must do this per
0151      * processor and cannot use the smp version of ctl_clear_bit().
0152      * On VM we only get one interrupt per virtally presented machinecheck.
0153      * Though one suffices, we may get one interrupt per (virtual) cpu.
0154      */
0155     if (mcck.warning) { /* WARNING pending ? */
0156         static int mchchk_wng_posted = 0;
0157 
0158         /* Use single cpu clear, as we cannot handle smp here. */
0159         __ctl_clear_bit(14, 24);    /* Disable WARNING MCH */
0160         if (xchg(&mchchk_wng_posted, 1) == 0)
0161             kill_cad_pid(SIGPWR, 1);
0162     }
0163     if (mcck.stp_queue)
0164         stp_queue_work();
0165     if (mcck.kill_task) {
0166         local_irq_enable();
0167         printk(KERN_EMERG "mcck: Terminating task because of machine "
0168                "malfunction (code 0x%016lx).\n", mcck.mcck_code);
0169         printk(KERN_EMERG "mcck: task: %s, pid: %d.\n",
0170                current->comm, current->pid);
0171         make_task_dead(SIGSEGV);
0172     }
0173 }
0174 
0175 void noinstr s390_handle_mcck(struct pt_regs *regs)
0176 {
0177     trace_hardirqs_off();
0178     pai_kernel_enter(regs);
0179     __s390_handle_mcck();
0180     pai_kernel_exit(regs);
0181     trace_hardirqs_on();
0182 }
0183 /*
0184  * returns 0 if all required registers are available
0185  * returns 1 otherwise
0186  */
0187 static int notrace s390_validate_registers(union mci mci, int umode)
0188 {
0189     struct mcesa *mcesa;
0190     void *fpt_save_area;
0191     union ctlreg2 cr2;
0192     int kill_task;
0193     u64 zero;
0194 
0195     kill_task = 0;
0196     zero = 0;
0197 
0198     if (!mci.gr) {
0199         /*
0200          * General purpose registers couldn't be restored and have
0201          * unknown contents. Stop system or terminate process.
0202          */
0203         if (!umode)
0204             s390_handle_damage();
0205         kill_task = 1;
0206     }
0207     if (!mci.fp) {
0208         /*
0209          * Floating point registers can't be restored. If the
0210          * kernel currently uses floating point registers the
0211          * system is stopped. If the process has its floating
0212          * pointer registers loaded it is terminated.
0213          */
0214         if (S390_lowcore.fpu_flags & KERNEL_VXR_V0V7)
0215             s390_handle_damage();
0216         if (!test_cpu_flag(CIF_FPU))
0217             kill_task = 1;
0218     }
0219     fpt_save_area = &S390_lowcore.floating_pt_save_area;
0220     if (!mci.fc) {
0221         /*
0222          * Floating point control register can't be restored.
0223          * If the kernel currently uses the floating pointer
0224          * registers and needs the FPC register the system is
0225          * stopped. If the process has its floating pointer
0226          * registers loaded it is terminated. Otherwise the
0227          * FPC is just validated.
0228          */
0229         if (S390_lowcore.fpu_flags & KERNEL_FPC)
0230             s390_handle_damage();
0231         asm volatile(
0232             "   lfpc    %0\n"
0233             :
0234             : "Q" (zero));
0235         if (!test_cpu_flag(CIF_FPU))
0236             kill_task = 1;
0237     } else {
0238         asm volatile(
0239             "   lfpc    %0\n"
0240             :
0241             : "Q" (S390_lowcore.fpt_creg_save_area));
0242     }
0243 
0244     mcesa = __va(S390_lowcore.mcesad & MCESA_ORIGIN_MASK);
0245     if (!MACHINE_HAS_VX) {
0246         /* Validate floating point registers */
0247         asm volatile(
0248             "   ld  0,0(%0)\n"
0249             "   ld  1,8(%0)\n"
0250             "   ld  2,16(%0)\n"
0251             "   ld  3,24(%0)\n"
0252             "   ld  4,32(%0)\n"
0253             "   ld  5,40(%0)\n"
0254             "   ld  6,48(%0)\n"
0255             "   ld  7,56(%0)\n"
0256             "   ld  8,64(%0)\n"
0257             "   ld  9,72(%0)\n"
0258             "   ld  10,80(%0)\n"
0259             "   ld  11,88(%0)\n"
0260             "   ld  12,96(%0)\n"
0261             "   ld  13,104(%0)\n"
0262             "   ld  14,112(%0)\n"
0263             "   ld  15,120(%0)\n"
0264             :
0265             : "a" (fpt_save_area)
0266             : "memory");
0267     } else {
0268         /* Validate vector registers */
0269         union ctlreg0 cr0;
0270 
0271         /*
0272          * The vector validity must only be checked if not running a
0273          * KVM guest. For KVM guests the machine check is forwarded by
0274          * KVM and it is the responsibility of the guest to take
0275          * appropriate actions. The host vector or FPU values have been
0276          * saved by KVM and will be restored by KVM.
0277          */
0278         if (!mci.vr && !test_cpu_flag(CIF_MCCK_GUEST)) {
0279             /*
0280              * Vector registers can't be restored. If the kernel
0281              * currently uses vector registers the system is
0282              * stopped. If the process has its vector registers
0283              * loaded it is terminated. Otherwise just validate
0284              * the registers.
0285              */
0286             if (S390_lowcore.fpu_flags & KERNEL_VXR)
0287                 s390_handle_damage();
0288             if (!test_cpu_flag(CIF_FPU))
0289                 kill_task = 1;
0290         }
0291         cr0.val = S390_lowcore.cregs_save_area[0];
0292         cr0.afp = cr0.vx = 1;
0293         __ctl_load(cr0.val, 0, 0);
0294         asm volatile(
0295             "   la  1,%0\n"
0296             "   .word   0xe70f,0x1000,0x0036\n" /* vlm 0,15,0(1) */
0297             "   .word   0xe70f,0x1100,0x0c36\n" /* vlm 16,31,256(1) */
0298             :
0299             : "Q" (*(struct vx_array *)mcesa->vector_save_area)
0300             : "1");
0301         __ctl_load(S390_lowcore.cregs_save_area[0], 0, 0);
0302     }
0303     /* Validate access registers */
0304     asm volatile(
0305         "   lam 0,15,0(%0)\n"
0306         :
0307         : "a" (&S390_lowcore.access_regs_save_area)
0308         : "memory");
0309     if (!mci.ar) {
0310         /*
0311          * Access registers have unknown contents.
0312          * Terminating task.
0313          */
0314         kill_task = 1;
0315     }
0316     /* Validate guarded storage registers */
0317     cr2.val = S390_lowcore.cregs_save_area[2];
0318     if (cr2.gse) {
0319         if (!mci.gs) {
0320             /*
0321              * 2 cases:
0322              * - machine check in kernel or userspace
0323              * - machine check while running SIE (KVM guest)
0324              * For kernel or userspace the userspace values of
0325              * guarded storage control can not be recreated, the
0326              * process must be terminated.
0327              * For SIE the guest values of guarded storage can not
0328              * be recreated. This is either due to a bug or due to
0329              * GS being disabled in the guest. The guest will be
0330              * notified by KVM code and the guests machine check
0331              * handling must take care of this.  The host values
0332              * are saved by KVM and are not affected.
0333              */
0334             if (!test_cpu_flag(CIF_MCCK_GUEST))
0335                 kill_task = 1;
0336         } else {
0337             load_gs_cb((struct gs_cb *)mcesa->guarded_storage_save_area);
0338         }
0339     }
0340     /*
0341      * The getcpu vdso syscall reads CPU number from the programmable
0342      * field of the TOD clock. Disregard the TOD programmable register
0343      * validity bit and load the CPU number into the TOD programmable
0344      * field unconditionally.
0345      */
0346     set_tod_programmable_field(raw_smp_processor_id());
0347     /* Validate clock comparator register */
0348     set_clock_comparator(S390_lowcore.clock_comparator);
0349 
0350     if (!mci.ms || !mci.pm || !mci.ia)
0351         kill_task = 1;
0352 
0353     return kill_task;
0354 }
0355 NOKPROBE_SYMBOL(s390_validate_registers);
0356 
0357 /*
0358  * Backup the guest's machine check info to its description block
0359  */
0360 static void notrace s390_backup_mcck_info(struct pt_regs *regs)
0361 {
0362     struct mcck_volatile_info *mcck_backup;
0363     struct sie_page *sie_page;
0364 
0365     /* r14 contains the sie block, which was set in sie64a */
0366     struct kvm_s390_sie_block *sie_block =
0367             (struct kvm_s390_sie_block *) regs->gprs[14];
0368 
0369     if (sie_block == NULL)
0370         /* Something's seriously wrong, stop system. */
0371         s390_handle_damage();
0372 
0373     sie_page = container_of(sie_block, struct sie_page, sie_block);
0374     mcck_backup = &sie_page->mcck_info;
0375     mcck_backup->mcic = S390_lowcore.mcck_interruption_code &
0376                 ~(MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE);
0377     mcck_backup->ext_damage_code = S390_lowcore.external_damage_code;
0378     mcck_backup->failing_storage_address
0379             = S390_lowcore.failing_storage_address;
0380 }
0381 NOKPROBE_SYMBOL(s390_backup_mcck_info);
0382 
0383 #define MAX_IPD_COUNT   29
0384 #define MAX_IPD_TIME    (5 * 60 * USEC_PER_SEC) /* 5 minutes */
0385 
0386 #define ED_STP_ISLAND   6   /* External damage STP island check */
0387 #define ED_STP_SYNC 7   /* External damage STP sync check */
0388 
0389 #define MCCK_CODE_NO_GUEST  (MCCK_CODE_CP | MCCK_CODE_EXT_DAMAGE)
0390 
0391 /*
0392  * machine check handler.
0393  */
0394 int notrace s390_do_machine_check(struct pt_regs *regs)
0395 {
0396     static int ipd_count;
0397     static DEFINE_SPINLOCK(ipd_lock);
0398     static unsigned long long last_ipd;
0399     struct mcck_struct *mcck;
0400     unsigned long long tmp;
0401     irqentry_state_t irq_state;
0402     union mci mci;
0403     unsigned long mcck_dam_code;
0404     int mcck_pending = 0;
0405 
0406     irq_state = irqentry_nmi_enter(regs);
0407 
0408     if (user_mode(regs))
0409         update_timer_mcck();
0410     inc_irq_stat(NMI_NMI);
0411     mci.val = S390_lowcore.mcck_interruption_code;
0412     mcck = this_cpu_ptr(&cpu_mcck);
0413 
0414     /*
0415      * Reinject the instruction processing damages' machine checks
0416      * including Delayed Access Exception into the guest
0417      * instead of damaging the host if they happen in the guest.
0418      */
0419     if (mci.pd && !test_cpu_flag(CIF_MCCK_GUEST)) {
0420         if (mci.b) {
0421             /* Processing backup -> verify if we can survive this */
0422             u64 z_mcic, o_mcic, t_mcic;
0423             z_mcic = (1ULL<<63 | 1ULL<<59 | 1ULL<<29);
0424             o_mcic = (1ULL<<43 | 1ULL<<42 | 1ULL<<41 | 1ULL<<40 |
0425                   1ULL<<36 | 1ULL<<35 | 1ULL<<34 | 1ULL<<32 |
0426                   1ULL<<30 | 1ULL<<21 | 1ULL<<20 | 1ULL<<17 |
0427                   1ULL<<16);
0428             t_mcic = mci.val;
0429 
0430             if (((t_mcic & z_mcic) != 0) ||
0431                 ((t_mcic & o_mcic) != o_mcic)) {
0432                 s390_handle_damage();
0433             }
0434 
0435             /*
0436              * Nullifying exigent condition, therefore we might
0437              * retry this instruction.
0438              */
0439             spin_lock(&ipd_lock);
0440             tmp = get_tod_clock();
0441             if (((tmp - last_ipd) >> 12) < MAX_IPD_TIME)
0442                 ipd_count++;
0443             else
0444                 ipd_count = 1;
0445             last_ipd = tmp;
0446             if (ipd_count == MAX_IPD_COUNT)
0447                 s390_handle_damage();
0448             spin_unlock(&ipd_lock);
0449         } else {
0450             /* Processing damage -> stopping machine */
0451             s390_handle_damage();
0452         }
0453     }
0454     if (s390_validate_registers(mci, user_mode(regs))) {
0455         /*
0456          * Couldn't restore all register contents for the
0457          * user space process -> mark task for termination.
0458          */
0459         mcck->kill_task = 1;
0460         mcck->mcck_code = mci.val;
0461         mcck_pending = 1;
0462     }
0463 
0464     /*
0465      * Backup the machine check's info if it happens when the guest
0466      * is running.
0467      */
0468     if (test_cpu_flag(CIF_MCCK_GUEST))
0469         s390_backup_mcck_info(regs);
0470 
0471     if (mci.cd) {
0472         /* Timing facility damage */
0473         s390_handle_damage();
0474     }
0475     if (mci.ed && mci.ec) {
0476         /* External damage */
0477         if (S390_lowcore.external_damage_code & (1U << ED_STP_SYNC))
0478             mcck->stp_queue |= stp_sync_check();
0479         if (S390_lowcore.external_damage_code & (1U << ED_STP_ISLAND))
0480             mcck->stp_queue |= stp_island_check();
0481         mcck_pending = 1;
0482     }
0483 
0484     if (mci.cp) {
0485         /* Channel report word pending */
0486         mcck->channel_report = 1;
0487         mcck_pending = 1;
0488     }
0489     if (mci.w) {
0490         /* Warning pending */
0491         mcck->warning = 1;
0492         mcck_pending = 1;
0493     }
0494 
0495     /*
0496      * If there are only Channel Report Pending and External Damage
0497      * machine checks, they will not be reinjected into the guest
0498      * because they refer to host conditions only.
0499      */
0500     mcck_dam_code = (mci.val & MCIC_SUBCLASS_MASK);
0501     if (test_cpu_flag(CIF_MCCK_GUEST) &&
0502     (mcck_dam_code & MCCK_CODE_NO_GUEST) != mcck_dam_code) {
0503         /* Set exit reason code for host's later handling */
0504         *((long *)(regs->gprs[15] + __SF_SIE_REASON)) = -EINTR;
0505     }
0506     clear_cpu_flag(CIF_MCCK_GUEST);
0507 
0508     if (user_mode(regs) && mcck_pending) {
0509         irqentry_nmi_exit(regs, irq_state);
0510         return 1;
0511     }
0512 
0513     if (mcck_pending)
0514         schedule_mcck_handler();
0515 
0516     irqentry_nmi_exit(regs, irq_state);
0517     return 0;
0518 }
0519 NOKPROBE_SYMBOL(s390_do_machine_check);
0520 
0521 static int __init machine_check_init(void)
0522 {
0523     ctl_set_bit(14, 25);    /* enable external damage MCH */
0524     ctl_set_bit(14, 27);    /* enable system recovery MCH */
0525     ctl_set_bit(14, 24);    /* enable warning MCH */
0526     return 0;
0527 }
0528 early_initcall(machine_check_init);
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!