genirq: Add a new generic IPI reservation code to irq core
authorQais Yousef <qais.yousef@imgtec.com>
Tue, 8 Dec 2015 13:20:19 +0000 (13:20 +0000)
committerThomas Gleixner <tglx@linutronix.de>
Thu, 25 Feb 2016 09:56:56 +0000 (10:56 +0100)
Add a generic mechanism to dynamically allocate an IPI. Depending on the
underlying implementation this creates either a single Linux irq or a
consective range of Linux irqs. The Linux irq is used later to send IPIs to
other CPUs.

[ tglx: Massaged the code and removed the 'consecutive mask' restriction for
   the single IRQ case ]

Signed-off-by: Qais Yousef <qais.yousef@imgtec.com>
Cc: <jason@lakedaemon.net>
Cc: <marc.zyngier@arm.com>
Cc: <jiang.liu@linux.intel.com>
Cc: <ralf@linux-mips.org>
Cc: <linux-mips@linux-mips.org>
Cc: <lisa.parratt@imgtec.com>
Cc: Qais Yousef <qsyousef@gmail.com>
Link: http://lkml.kernel.org/r/1449580830-23652-9-git-send-email-qais.yousef@imgtec.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
include/linux/irq.h
include/linux/irqdomain.h
kernel/irq/Makefile
kernel/irq/ipi.c [new file with mode: 0644]

index a32b47fbf874a706a56b6c9549b16aa89a3a8ace..95f4f66f95f371700924fd8e14418cb2adabf079 100644 (file)
@@ -940,4 +940,7 @@ static inline u32 irq_reg_readl(struct irq_chip_generic *gc,
                return readl(gc->reg_base + reg_offset);
 }
 
+/* Contrary to Linux irqs, for hardware irqs the irq number 0 is valid */
+#define INVALID_HWIRQ  (~0UL)
+
 #endif /* _LINUX_IRQ_H */
index c466cc17b8e94a2f88752a3707d44adc7c64d1a7..ed48594e96d2562ddc1de9c1dc94559629f7a4da 100644 (file)
@@ -344,6 +344,11 @@ int irq_domain_xlate_onetwocell(struct irq_domain *d, struct device_node *ctrlr,
                        const u32 *intspec, unsigned int intsize,
                        irq_hw_number_t *out_hwirq, unsigned int *out_type);
 
+/* IPI functions */
+unsigned int irq_reserve_ipi(struct irq_domain *domain,
+                            const struct cpumask *dest);
+void irq_destroy_ipi(unsigned int irq);
+
 /* V2 interfaces to support hierarchy IRQ domains. */
 extern struct irq_data *irq_domain_get_irq_data(struct irq_domain *domain,
                                                unsigned int virq);
index 2fc9cbdf35b6221385ab1f8393098523cc4c43cd..2ee42e95a3ce182ce44bea1b1824e63b93436f37 100644 (file)
@@ -8,3 +8,4 @@ obj-$(CONFIG_GENERIC_PENDING_IRQ) += migration.o
 obj-$(CONFIG_GENERIC_IRQ_MIGRATION) += cpuhotplug.o
 obj-$(CONFIG_PM_SLEEP) += pm.o
 obj-$(CONFIG_GENERIC_MSI_IRQ) += msi.o
+obj-$(CONFIG_GENERIC_IRQ_IPI) += ipi.o
diff --git a/kernel/irq/ipi.c b/kernel/irq/ipi.c
new file mode 100644 (file)
index 0000000..340af27
--- /dev/null
@@ -0,0 +1,137 @@
+/*
+ * linux/kernel/irq/ipi.c
+ *
+ * Copyright (C) 2015 Imagination Technologies Ltd
+ * Author: Qais Yousef <qais.yousef@imgtec.com>
+ *
+ * This file contains driver APIs to the IPI subsystem.
+ */
+
+#define pr_fmt(fmt) "genirq/ipi: " fmt
+
+#include <linux/irqdomain.h>
+#include <linux/irq.h>
+
+/**
+ * irq_reserve_ipi() - Setup an IPI to destination cpumask
+ * @domain:    IPI domain
+ * @dest:      cpumask of cpus which can receive the IPI
+ *
+ * Allocate a virq that can be used to send IPI to any CPU in dest mask.
+ *
+ * On success it'll return linux irq number and 0 on failure
+ */
+unsigned int irq_reserve_ipi(struct irq_domain *domain,
+                            const struct cpumask *dest)
+{
+       unsigned int nr_irqs, offset;
+       struct irq_data *data;
+       int virq, i;
+
+       if (!domain ||!irq_domain_is_ipi(domain)) {
+               pr_warn("Reservation on a non IPI domain\n");
+               return 0;
+       }
+
+       if (!cpumask_subset(dest, cpu_possible_mask)) {
+               pr_warn("Reservation is not in possible_cpu_mask\n");
+               return 0;
+       }
+
+       nr_irqs = cpumask_weight(dest);
+       if (!nr_irqs) {
+               pr_warn("Reservation for empty destination mask\n");
+               return 0;
+       }
+
+       if (irq_domain_is_ipi_single(domain)) {
+               /*
+                * If the underlying implementation uses a single HW irq on
+                * all cpus then we only need a single Linux irq number for
+                * it. We have no restrictions vs. the destination mask. The
+                * underlying implementation can deal with holes nicely.
+                */
+               nr_irqs = 1;
+               offset = 0;
+       } else {
+               unsigned int next;
+
+               /*
+                * The IPI requires a seperate HW irq on each CPU. We require
+                * that the destination mask is consecutive. If an
+                * implementation needs to support holes, it can reserve
+                * several IPI ranges.
+                */
+               offset = cpumask_first(dest);
+               /*
+                * Find a hole and if found look for another set bit after the
+                * hole. For now we don't support this scenario.
+                */
+               next = cpumask_next_zero(offset, dest);
+               if (next < nr_cpu_ids)
+                       next = cpumask_next(next, dest);
+               if (next < nr_cpu_ids) {
+                       pr_warn("Destination mask has holes\n");
+                       return 0;
+               }
+       }
+
+       virq = irq_domain_alloc_descs(-1, nr_irqs, 0, NUMA_NO_NODE);
+       if (virq <= 0) {
+               pr_warn("Can't reserve IPI, failed to alloc descs\n");
+               return 0;
+       }
+
+       virq = __irq_domain_alloc_irqs(domain, virq, nr_irqs, NUMA_NO_NODE,
+                                      (void *) dest, true);
+
+       if (virq <= 0) {
+               pr_warn("Can't reserve IPI, failed to alloc hw irqs\n");
+               goto free_descs;
+       }
+
+       for (i = 0; i < nr_irqs; i++) {
+               data = irq_get_irq_data(virq + i);
+               cpumask_copy(data->common->affinity, dest);
+               data->common->ipi_offset = offset;
+       }
+       return virq;
+
+free_descs:
+       irq_free_descs(virq, nr_irqs);
+       return 0;
+}
+
+/**
+ * irq_destroy_ipi() - unreserve an IPI that was previously allocated
+ * @irq:       linux irq number to be destroyed
+ *
+ * Return the IPIs allocated with irq_reserve_ipi() to the system destroying
+ * all virqs associated with them.
+ */
+void irq_destroy_ipi(unsigned int irq)
+{
+       struct irq_data *data = irq_get_irq_data(irq);
+       struct cpumask *ipimask = data ? irq_data_get_affinity_mask(data) : NULL;
+       struct irq_domain *domain;
+       unsigned int nr_irqs;
+
+       if (!irq || !data || !ipimask)
+               return;
+
+       domain = data->domain;
+       if (WARN_ON(domain == NULL))
+               return;
+
+       if (!irq_domain_is_ipi(domain)) {
+               pr_warn("Trying to destroy a non IPI domain!\n");
+               return;
+       }
+
+       if (irq_domain_is_ipi_per_cpu(domain))
+               nr_irqs = cpumask_weight(ipimask);
+       else
+               nr_irqs = 1;
+
+       irq_domain_free_irqs(irq, nr_irqs);
+}