mca.c

linux 内核源代码

/*
 * C portion of the OS INIT handler
 *
 * Called from ia64_os_init_dispatch
 *
 * Inputs: pointer to pt_regs where processor info was saved.  SAL/OS state for
 * this event.  This code is used for both monarch and slave INIT events, see
 * sos->monarch.
 *
 * All INIT events switch to the INIT stack and change the previous process to
 * blocked status.  If one of the INIT events is the monarch then we are
 * probably processing the nmi button/command.  Use the monarch cpu to dump all
 * the processes.  The slave INIT events all spin until the monarch cpu
 * returns.  We can also get INIT slave events for MCA, in which case the MCA
 * process is the monarch.
 */

void
ia64_init_handler(struct pt_regs *regs, struct switch_stack *sw,
		  struct ia64_sal_os_state *sos)
{
	static atomic_t slaves;
	static atomic_t monarchs;
	struct task_struct *previous_current;
	int cpu = smp_processor_id();
	struct ia64_mca_notify_die nd =
		{ .sos = sos, .monarch_cpu = &monarch_cpu };

	(void) notify_die(DIE_INIT_ENTER, "INIT", regs, (long)&nd, 0, 0);

	mprintk(KERN_INFO "Entered OS INIT handler. PSP=%lx cpu=%d monarch=%ld\n",
		sos->proc_state_param, cpu, sos->monarch);
	salinfo_log_wakeup(SAL_INFO_TYPE_INIT, NULL, 0, 0);

	previous_current = ia64_mca_modify_original_stack(regs, sw, sos, "INIT");
	sos->os_status = IA64_INIT_RESUME;

	/* FIXME: Workaround for broken proms that drive all INIT events as
	 * slaves.  The last slave that enters is promoted to be a monarch.
	 * Remove this code in September 2006, that gives platforms a year to
	 * fix their proms and get their customers updated.
	 */
	if (!sos->monarch && atomic_add_return(1, &slaves) == num_online_cpus()) {
		mprintk(KERN_WARNING "%s: Promoting cpu %d to monarch.\n",
		       __FUNCTION__, cpu);
		atomic_dec(&slaves);
		sos->monarch = 1;
	}

	/* FIXME: Workaround for broken proms that drive all INIT events as
	 * monarchs.  Second and subsequent monarchs are demoted to slaves.
	 * Remove this code in September 2006, that gives platforms a year to
	 * fix their proms and get their customers updated.
	 */
	if (sos->monarch && atomic_add_return(1, &monarchs) > 1) {
		mprintk(KERN_WARNING "%s: Demoting cpu %d to slave.\n",
			       __FUNCTION__, cpu);
		atomic_dec(&monarchs);
		sos->monarch = 0;
	}

	if (!sos->monarch) {
		ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_INIT;
		while (monarch_cpu == -1)
		       cpu_relax();	/* spin until monarch enters */
		if (notify_die(DIE_INIT_SLAVE_ENTER, "INIT", regs, (long)&nd, 0, 0)
				== NOTIFY_STOP)
			ia64_mca_spin(__FUNCTION__);
		if (notify_die(DIE_INIT_SLAVE_PROCESS, "INIT", regs, (long)&nd, 0, 0)
				== NOTIFY_STOP)
			ia64_mca_spin(__FUNCTION__);
		while (monarch_cpu != -1)
		       cpu_relax();	/* spin until monarch leaves */
		if (notify_die(DIE_INIT_SLAVE_LEAVE, "INIT", regs, (long)&nd, 0, 0)
				== NOTIFY_STOP)
			ia64_mca_spin(__FUNCTION__);
		mprintk("Slave on cpu %d returning to normal service.\n", cpu);
		set_curr_task(cpu, previous_current);
		ia64_mc_info.imi_rendez_checkin[cpu] = IA64_MCA_RENDEZ_CHECKIN_NOTDONE;
		atomic_dec(&slaves);
		return;
	}

	monarch_cpu = cpu;
	if (notify_die(DIE_INIT_MONARCH_ENTER, "INIT", regs, (long)&nd, 0, 0)
			== NOTIFY_STOP)
		ia64_mca_spin(__FUNCTION__);

	/*
	 * Wait for a bit.  On some machines (e.g., HP's zx2000 and zx6000, INIT can be
	 * generated via the BMC's command-line interface, but since the console is on the
	 * same serial line, the user will need some time to switch out of the BMC before
	 * the dump begins.
	 */
	mprintk("Delaying for 5 seconds...\n");
	udelay(5*1000000);
	ia64_wait_for_slaves(cpu, "INIT");
	/* If nobody intercepts DIE_INIT_MONARCH_PROCESS then we drop through
	 * to default_monarch_init_process() above and just print all the
	 * tasks.
	 */
	if (notify_die(DIE_INIT_MONARCH_PROCESS, "INIT", regs, (long)&nd, 0, 0)
			== NOTIFY_STOP)
		ia64_mca_spin(__FUNCTION__);
	if (notify_die(DIE_INIT_MONARCH_LEAVE, "INIT", regs, (long)&nd, 0, 0)
			== NOTIFY_STOP)
		ia64_mca_spin(__FUNCTION__);
	mprintk("\nINIT dump complete.  Monarch on cpu %d returning to normal service.\n", cpu);
	atomic_dec(&monarchs);
	set_curr_task(cpu, previous_current);
	monarch_cpu = -1;
	return;
}

static int __init
ia64_mca_disable_cpe_polling(char *str)
{
	cpe_poll_enabled = 0;
	return 1;
}

__setup("disable_cpe_poll", ia64_mca_disable_cpe_polling);

static struct irqaction cmci_irqaction = {
	.handler =	ia64_mca_cmc_int_handler,
	.flags =	IRQF_DISABLED,
	.name =		"cmc_hndlr"
};

static struct irqaction cmcp_irqaction = {
	.handler =	ia64_mca_cmc_int_caller,
	.flags =	IRQF_DISABLED,
	.name =		"cmc_poll"
};

static struct irqaction mca_rdzv_irqaction = {
	.handler =	ia64_mca_rendez_int_handler,
	.flags =	IRQF_DISABLED,
	.name =		"mca_rdzv"
};

static struct irqaction mca_wkup_irqaction = {
	.handler =	ia64_mca_wakeup_int_handler,
	.flags =	IRQF_DISABLED,
	.name =		"mca_wkup"
};

#ifdef CONFIG_ACPI
static struct irqaction mca_cpe_irqaction = {
	.handler =	ia64_mca_cpe_int_handler,
	.flags =	IRQF_DISABLED,
	.name =		"cpe_hndlr"
};

static struct irqaction mca_cpep_irqaction = {
	.handler =	ia64_mca_cpe_int_caller,
	.flags =	IRQF_DISABLED,
	.name =		"cpe_poll"
};
#endif /* CONFIG_ACPI */

/* Minimal format of the MCA/INIT stacks.  The pseudo processes that run on
 * these stacks can never sleep, they cannot return from the kernel to user
 * space, they do not appear in a normal ps listing.  So there is no need to
 * format most of the fields.
 */

static void __cpuinit
format_mca_init_stack(void *mca_data, unsigned long offset,
		const char *type, int cpu)
{
	struct task_struct *p = (struct task_struct *)((char *)mca_data + offset);
	struct thread_info *ti;
	memset(p, 0, KERNEL_STACK_SIZE);
	ti = task_thread_info(p);
	ti->flags = _TIF_MCA_INIT;
	ti->preempt_count = 1;
	ti->task = p;
	ti->cpu = cpu;
	p->stack = ti;
	p->state = TASK_UNINTERRUPTIBLE;
	cpu_set(cpu, p->cpus_allowed);
	INIT_LIST_HEAD(&p->tasks);
	p->parent = p->real_parent = p->group_leader = p;
	INIT_LIST_HEAD(&p->children);
	INIT_LIST_HEAD(&p->sibling);
	strncpy(p->comm, type, sizeof(p->comm)-1);
}

/* Caller prevents this from being called after init */
static void * __init_refok mca_bootmem(void)
{
	void *p;

	p = alloc_bootmem(sizeof(struct ia64_mca_cpu) * NR_CPUS +
	                  KERNEL_STACK_SIZE);
	return (void *)ALIGN((unsigned long)p, KERNEL_STACK_SIZE);
}

/* Do per-CPU MCA-related initialization.  */
void __cpuinit
ia64_mca_cpu_init(void *cpu_data)
{
	void *pal_vaddr;
	static int first_time = 1;

	if (first_time) {
		void *mca_data;
		int cpu;

		first_time = 0;
		mca_data = mca_bootmem();
		for (cpu = 0; cpu < NR_CPUS; cpu++) {
			format_mca_init_stack(mca_data,
					offsetof(struct ia64_mca_cpu, mca_stack),
					"MCA", cpu);
			format_mca_init_stack(mca_data,
					offsetof(struct ia64_mca_cpu, init_stack),
					"INIT", cpu);
			__per_cpu_mca[cpu] = __pa(mca_data);
			mca_data += sizeof(struct ia64_mca_cpu);
		}
	}

	/*
	 * The MCA info structure was allocated earlier and its
	 * physical address saved in __per_cpu_mca[cpu].  Copy that
	 * address * to ia64_mca_data so we can access it as a per-CPU
	 * variable.
	 */
	__get_cpu_var(ia64_mca_data) = __per_cpu_mca[smp_processor_id()];

	/*
	 * Stash away a copy of the PTE needed to map the per-CPU page.
	 * We may need it during MCA recovery.
	 */
	__get_cpu_var(ia64_mca_per_cpu_pte) =
		pte_val(mk_pte_phys(__pa(cpu_data), PAGE_KERNEL));

	/*
	 * Also, stash away a copy of the PAL address and the PTE
	 * needed to map it.
	 */
	pal_vaddr = efi_get_pal_addr();
	if (!pal_vaddr)
		return;
	__get_cpu_var(ia64_mca_pal_base) =
		GRANULEROUNDDOWN((unsigned long) pal_vaddr);
	__get_cpu_var(ia64_mca_pal_pte) = pte_val(mk_pte_phys(__pa(pal_vaddr),
							      PAGE_KERNEL));
}

static void __cpuinit ia64_mca_cmc_vector_adjust(void *dummy)
{
	unsigned long flags;

	local_irq_save(flags);
	if (!cmc_polling_enabled)
		ia64_mca_cmc_vector_enable(NULL);
	local_irq_restore(flags);
}

static int __cpuinit mca_cpu_callback(struct notifier_block *nfb,
				      unsigned long action,
				      void *hcpu)
{
	int hotcpu = (unsigned long) hcpu;

	switch (action) {
	case CPU_ONLINE:
	case CPU_ONLINE_FROZEN:
		smp_call_function_single(hotcpu, ia64_mca_cmc_vector_adjust,
					 NULL, 1, 0);
		break;
	}
	return NOTIFY_OK;
}

static struct notifier_block mca_cpu_notifier __cpuinitdata = {
	.notifier_call = mca_cpu_callback
};

/*
 * ia64_mca_init
 *
 *  Do all the system level mca specific initialization.
 *
 *	1. Register spinloop and wakeup request interrupt vectors
 *
 *	2. Register OS_MCA handler entry point
 *
 *	3. Register OS_INIT handler entry point
 *
 *  4. Initialize MCA/CMC/INIT related log buffers maintained by the OS.
 *
 *  Note that this initialization is done very early before some kernel
 *  services are available.
 *
 *  Inputs  :   None
 *
 *  Outputs :   None
 */
void __init
ia64_mca_init(void)
{
	ia64_fptr_t *init_hldlr_ptr_monarch = (ia64_fptr_t *)ia64_os_init_dispatch_monarch;
	ia64_fptr_t *init_hldlr_ptr_slave = (ia64_fptr_t *)ia64_os_init_dispatch_slave;
	ia64_fptr_t *mca_hldlr_ptr = (ia64_fptr_t *)ia64_os_mca_dispatch;
	int i;
	s64 rc;
	struct ia64_sal_retval isrv;
	u64 timeout = IA64_MCA_RENDEZ_TIMEOUT;	/* platform specific */
	static struct notifier_block default_init_monarch_nb = {
		.notifier_call = default_monarch_init_process,
		.priority = 0/* we need to notified last */
	};

	IA64_MCA_DEBUG("%s: begin\n", __FUNCTION__);

	/* Clear the Rendez checkin flag for all cpus */
	for(i = 0 ; i < NR_CPUS; i++)
		ia64_mc_info.imi_rendez_checkin[i] = IA64_MCA_RENDEZ_CHECKIN_NOTDONE;

	/*
	 * Register the rendezvous spinloop and wakeup mechanism with SAL
	 */

	/* Register the rendezvous interrupt vector with SAL */
	while (1) {
		isrv = ia64_sal_mc_set_params(SAL_MC_PARAM_RENDEZ_INT,
					      SAL_MC_PARAM_MECHANISM_INT,
					      IA64_MCA_RENDEZ_VECTOR,
					      timeout,
					      SAL_MC_PARAM_RZ_ALWAYS);
		rc = isrv.status;
		if (rc == 0)
			break;
		if (rc == -2) {
			printk(KERN_INFO "Increasing MCA rendezvous timeout from "
				"%ld to %ld milliseconds\n", timeout, isrv.v0);
			timeout = isrv.v0;
			(void) notify_die(DIE_MCA_NEW_TIMEOUT, "MCA", NULL, timeout, 0, 0);
			continue;
		}
		printk(KERN_ERR "Failed to register rendezvous interrupt "
		       "with SAL (status %ld)\n", rc);
		return;
	}

	/* Register the wakeup interrupt vector with SAL */
	isrv = ia64_sal_mc_set_params(SAL_MC_PARAM_RENDEZ_WAKEUP,
				      SAL_MC_PARAM_MECHANISM_INT,
				      IA64_MCA_WAKEUP_VECTOR,
				      0, 0);
	rc = isrv.status;
	if (rc) {
		printk(KERN_ERR "Failed to register wakeup interrupt with SAL "
		       "(status %ld)\n", rc);
		return;
	}

	IA64_MCA_DEBUG("%s: registered MCA rendezvous spinloop and wakeup mech.\n", __FUNCTION__);

	ia64_mc_info.imi_mca_handler        = ia64_tpa(mca_hldlr_ptr->fp);
	/*
	 * XXX - disable SAL checksum by setting size to 0; should be
	 *	ia64_tpa(ia64_os_mca_dispatch_end) - ia64_tpa(ia64_os_mca_dispatch);
	 */
	ia64_mc_info.imi_mca_handler_size	= 0;

	/* Register the os mca handler with SAL */
	if ((rc = ia64_sal_set_vectors(SAL_VECTOR_OS_MCA,
				       ia64_mc_info.imi_mca_handler,
				       ia64_tpa(mca_hldlr_ptr->gp),
				       ia64_mc_info.imi_mca_handler_size,
				       0, 0, 0)))
	{
		printk(KERN_ERR "Failed to register OS MCA handler with SAL "
		       "(status %ld)\n", rc);
		return;
	}

	IA64_MCA_DEBUG("%s: registered OS MCA handler with SAL at 0x%lx, gp = 0x%lx\n", __FUNCTION__,
		       ia64_mc_info.imi_mca_handler, ia64_tpa(mca_hldlr_ptr->gp));

	/*
	 * XXX - disable SAL checksum by setting size to 0, should be
	 * size of the actual init handler in mca_asm.S.
	 */
	ia64_mc_info.imi_monarch_init_handler		= ia64_tpa(init_hldlr_ptr_monarch->fp);
	ia64_mc_info.imi_monarch_init_handler_size	= 0;
	ia64_mc_info.imi_slave_init_handler		= ia64_tpa(init_hldlr_ptr_slave->fp);
	ia64_mc_info.imi_slave_init_handler_size	= 0;

	IA64_MCA_DEBUG("%s: OS INIT handler at %lx\n", __FUNCTION__,
		       ia64_mc_info.imi_monarch_init_handler);

	/* Register the os init handler with SAL */
	if ((rc = ia64_sal_set_vectors(SAL_VECTOR_OS_INIT,
				       ia64_mc_info.imi_monarch_init_handler,
				       ia64_tpa(ia64_getreg(_IA64_REG_GP)),
				       ia64_mc_info.imi_monarch_init_handler_size,
				       ia64_mc_info.imi_slave_init_handler,
				       ia64_tpa(ia64_getreg(_IA64_REG_GP)),
				       ia64_mc_info.imi_slave_init_handler_size)))
	{
		printk(KERN_ERR "Failed to register m/s INIT handlers with SAL "
		       "(status %ld)\n", rc);
		return;
	}
	if (register_die_notifier(&default_init_monarch_nb)) {
		printk(KERN_ERR "Failed to register default monarch INIT process\n");
		return;
	}

	IA64_MCA_DEBUG("%s: registered OS INIT handler with SAL\n", __FUNCTION__);

	/*
	 *  Configure the CMCI/P vector and handler. Interrupts for CMC are
	 *  per-processor, so AP CMC interrupts are setup in smp_callin() (smpboot.c).
	 */
	register_percpu_irq(IA64_CMC_VECTOR, &cmci_irqaction);
	register_percpu_irq(IA64_CMCP_VECTOR, &cmcp_irqaction);
	ia64_mca_cmc_vector_setup();       /* Setup vector on BSP */

	/* Setup the MCA rendezvous interrupt vector */
	register_percpu_irq(IA64_MCA_RENDEZ_VECTOR, &mca_rdzv_irqaction);

	/* Setup the MCA wakeup interrupt vector */
	register_percpu_irq(IA64_MCA_WAKEUP_VECTOR, &mca_wkup_irqaction);

#ifdef CONFIG_ACPI
	/* Setup the CPEI/P handler */
	register_percpu_irq(IA64_CPEP_VECTOR, &mca_cpep_irqaction);
#endif

	/* Initialize the areas set aside by the OS to buffer the
	 * platform/processor error states for MCA/INIT/CMC
	 * handling.
	 */
	ia64_log_init(SAL_INFO_TYPE_MCA);
	ia64_log_init(SAL_INFO_TYPE_INIT);
	ia64_log_init(SAL_INFO_TYPE_CMC);
	ia64_log_init(SAL_INFO_TYPE_CPE);

	mca_init = 1;
	printk(KERN_INFO "MCA related initialization done\n");
}

/*
 * ia64_mca_late_init
 *
 *	Opportunity to setup things that require initialization later
 *	than ia64_mca_init.  Setup a timer to poll for CPEs if the
 *	platform doesn't support an interrupt driven mechanism.
 *
 *  Inputs  :   None
 *  Outputs :   Status
 */
static int __init
ia64_mca_late_init(void)
{
	if (!mca_init)
		return 0;

	register_hotcpu_notifier(&mca_cpu_notifier);

	/* Setup the CMCI/P vector and handler */
	init_timer(&cmc_poll_timer);
	cmc_poll_timer.function = ia64_mca_cmc_poll;

	/* Unmask/enable the vector */
	cmc_polling_enabled = 0;
	schedule_work(&cmc_enable_work);

	IA64_MCA_DEBUG("%s: CMCI/P setup and enabled.\n", __FUNCTION__);

#ifdef CONFIG_ACPI
	/* Setup the CPEI/P vector and handler */
	cpe_vector = acpi_request_vector(ACPI_INTERRUPT_CPEI);
	init_timer(&cpe_poll_timer);
	cpe_poll_timer.function = ia64_mca_cpe_poll;

	{
		irq_desc_t *desc;
		unsigned int irq;

		if (cpe_vector >= 0) {
			/* If platform supports CPEI, enable the irq. */
			irq = local_vector_to_irq(cpe_vector);
			if (irq > 0) {
				cpe_poll_enabled = 0;
				desc = irq_desc + irq;
				desc->status |= IRQ_PER_CPU;
				setup_irq(irq, &mca_cpe_irqaction);
				ia64_cpe_irq = irq;
				ia64_mca_register_cpev(cpe_vector);
				IA64_MCA_DEBUG("%s: CPEI/P setup and enabled.\n",
					__FUNCTION__);
				return 0;
			}
			printk(KERN_ERR "%s: Failed to find irq for CPE "
					"interrupt handler, vector %d\n",
					__FUNCTION__, cpe_vector);
		}
		/* If platform doesn't support CPEI, get the timer going. */
		if (cpe_poll_enabled) {
			ia64_mca_cpe_poll(0UL);
			IA64_MCA_DEBUG("%s: CPEP setup and enabled.\n", __FUNCTION__);
		}
	}
#endif

	return 0;
}

device_initcall(ia64_mca_late_init);