Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / irqchip / irq-gic-v3.c
diff --git a/kernel/drivers/irqchip/irq-gic-v3.c b/kernel/drivers/irqchip/irq-gic-v3.c
new file mode 100644 (file)
index 0000000..49875ad
--- /dev/null
@@ -0,0 +1,873 @@
+/*
+ * Copyright (C) 2013, 2014 ARM Limited, All Rights Reserved.
+ * Author: Marc Zyngier <marc.zyngier@arm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <linux/cpu.h>
+#include <linux/cpu_pm.h>
+#include <linux/delay.h>
+#include <linux/interrupt.h>
+#include <linux/of.h>
+#include <linux/of_address.h>
+#include <linux/of_irq.h>
+#include <linux/percpu.h>
+#include <linux/slab.h>
+
+#include <linux/irqchip/arm-gic-v3.h>
+
+#include <asm/cputype.h>
+#include <asm/exception.h>
+#include <asm/smp_plat.h>
+
+#include "irq-gic-common.h"
+#include "irqchip.h"
+
+struct redist_region {
+       void __iomem            *redist_base;
+       phys_addr_t             phys_base;
+};
+
+struct gic_chip_data {
+       void __iomem            *dist_base;
+       struct redist_region    *redist_regions;
+       struct rdists           rdists;
+       struct irq_domain       *domain;
+       u64                     redist_stride;
+       u32                     nr_redist_regions;
+       unsigned int            irq_nr;
+};
+
+static struct gic_chip_data gic_data __read_mostly;
+
+#define gic_data_rdist()               (this_cpu_ptr(gic_data.rdists.rdist))
+#define gic_data_rdist_rd_base()       (gic_data_rdist()->rd_base)
+#define gic_data_rdist_sgi_base()      (gic_data_rdist_rd_base() + SZ_64K)
+
+/* Our default, arbitrary priority value. Linux only uses one anyway. */
+#define DEFAULT_PMR_VALUE      0xf0
+
+static inline unsigned int gic_irq(struct irq_data *d)
+{
+       return d->hwirq;
+}
+
+static inline int gic_irq_in_rdist(struct irq_data *d)
+{
+       return gic_irq(d) < 32;
+}
+
+static inline void __iomem *gic_dist_base(struct irq_data *d)
+{
+       if (gic_irq_in_rdist(d))        /* SGI+PPI -> SGI_base for this CPU */
+               return gic_data_rdist_sgi_base();
+
+       if (d->hwirq <= 1023)           /* SPI -> dist_base */
+               return gic_data.dist_base;
+
+       return NULL;
+}
+
+static void gic_do_wait_for_rwp(void __iomem *base)
+{
+       u32 count = 1000000;    /* 1s! */
+
+       while (readl_relaxed(base + GICD_CTLR) & GICD_CTLR_RWP) {
+               count--;
+               if (!count) {
+                       pr_err_ratelimited("RWP timeout, gone fishing\n");
+                       return;
+               }
+               cpu_relax();
+               udelay(1);
+       };
+}
+
+/* Wait for completion of a distributor change */
+static void gic_dist_wait_for_rwp(void)
+{
+       gic_do_wait_for_rwp(gic_data.dist_base);
+}
+
+/* Wait for completion of a redistributor change */
+static void gic_redist_wait_for_rwp(void)
+{
+       gic_do_wait_for_rwp(gic_data_rdist_rd_base());
+}
+
+/* Low level accessors */
+static u64 __maybe_unused gic_read_iar(void)
+{
+       u64 irqstat;
+
+       asm volatile("mrs_s %0, " __stringify(ICC_IAR1_EL1) : "=r" (irqstat));
+       return irqstat;
+}
+
+static void __maybe_unused gic_write_pmr(u64 val)
+{
+       asm volatile("msr_s " __stringify(ICC_PMR_EL1) ", %0" : : "r" (val));
+}
+
+static void __maybe_unused gic_write_ctlr(u64 val)
+{
+       asm volatile("msr_s " __stringify(ICC_CTLR_EL1) ", %0" : : "r" (val));
+       isb();
+}
+
+static void __maybe_unused gic_write_grpen1(u64 val)
+{
+       asm volatile("msr_s " __stringify(ICC_GRPEN1_EL1) ", %0" : : "r" (val));
+       isb();
+}
+
+static void __maybe_unused gic_write_sgi1r(u64 val)
+{
+       asm volatile("msr_s " __stringify(ICC_SGI1R_EL1) ", %0" : : "r" (val));
+}
+
+static void gic_enable_sre(void)
+{
+       u64 val;
+
+       asm volatile("mrs_s %0, " __stringify(ICC_SRE_EL1) : "=r" (val));
+       val |= ICC_SRE_EL1_SRE;
+       asm volatile("msr_s " __stringify(ICC_SRE_EL1) ", %0" : : "r" (val));
+       isb();
+
+       /*
+        * Need to check that the SRE bit has actually been set. If
+        * not, it means that SRE is disabled at EL2. We're going to
+        * die painfully, and there is nothing we can do about it.
+        *
+        * Kindly inform the luser.
+        */
+       asm volatile("mrs_s %0, " __stringify(ICC_SRE_EL1) : "=r" (val));
+       if (!(val & ICC_SRE_EL1_SRE))
+               pr_err("GIC: unable to set SRE (disabled at EL2), panic ahead\n");
+}
+
+static void gic_enable_redist(bool enable)
+{
+       void __iomem *rbase;
+       u32 count = 1000000;    /* 1s! */
+       u32 val;
+
+       rbase = gic_data_rdist_rd_base();
+
+       val = readl_relaxed(rbase + GICR_WAKER);
+       if (enable)
+               /* Wake up this CPU redistributor */
+               val &= ~GICR_WAKER_ProcessorSleep;
+       else
+               val |= GICR_WAKER_ProcessorSleep;
+       writel_relaxed(val, rbase + GICR_WAKER);
+
+       if (!enable) {          /* Check that GICR_WAKER is writeable */
+               val = readl_relaxed(rbase + GICR_WAKER);
+               if (!(val & GICR_WAKER_ProcessorSleep))
+                       return; /* No PM support in this redistributor */
+       }
+
+       while (count--) {
+               val = readl_relaxed(rbase + GICR_WAKER);
+               if (enable ^ (val & GICR_WAKER_ChildrenAsleep))
+                       break;
+               cpu_relax();
+               udelay(1);
+       };
+       if (!count)
+               pr_err_ratelimited("redistributor failed to %s...\n",
+                                  enable ? "wakeup" : "sleep");
+}
+
+/*
+ * Routines to disable, enable, EOI and route interrupts
+ */
+static int gic_peek_irq(struct irq_data *d, u32 offset)
+{
+       u32 mask = 1 << (gic_irq(d) % 32);
+       void __iomem *base;
+
+       if (gic_irq_in_rdist(d))
+               base = gic_data_rdist_sgi_base();
+       else
+               base = gic_data.dist_base;
+
+       return !!(readl_relaxed(base + offset + (gic_irq(d) / 32) * 4) & mask);
+}
+
+static void gic_poke_irq(struct irq_data *d, u32 offset)
+{
+       u32 mask = 1 << (gic_irq(d) % 32);
+       void (*rwp_wait)(void);
+       void __iomem *base;
+
+       if (gic_irq_in_rdist(d)) {
+               base = gic_data_rdist_sgi_base();
+               rwp_wait = gic_redist_wait_for_rwp;
+       } else {
+               base = gic_data.dist_base;
+               rwp_wait = gic_dist_wait_for_rwp;
+       }
+
+       writel_relaxed(mask, base + offset + (gic_irq(d) / 32) * 4);
+       rwp_wait();
+}
+
+static void gic_mask_irq(struct irq_data *d)
+{
+       gic_poke_irq(d, GICD_ICENABLER);
+}
+
+static void gic_unmask_irq(struct irq_data *d)
+{
+       gic_poke_irq(d, GICD_ISENABLER);
+}
+
+static int gic_irq_set_irqchip_state(struct irq_data *d,
+                                    enum irqchip_irq_state which, bool val)
+{
+       u32 reg;
+
+       if (d->hwirq >= gic_data.irq_nr) /* PPI/SPI only */
+               return -EINVAL;
+
+       switch (which) {
+       case IRQCHIP_STATE_PENDING:
+               reg = val ? GICD_ISPENDR : GICD_ICPENDR;
+               break;
+
+       case IRQCHIP_STATE_ACTIVE:
+               reg = val ? GICD_ISACTIVER : GICD_ICACTIVER;
+               break;
+
+       case IRQCHIP_STATE_MASKED:
+               reg = val ? GICD_ICENABLER : GICD_ISENABLER;
+               break;
+
+       default:
+               return -EINVAL;
+       }
+
+       gic_poke_irq(d, reg);
+       return 0;
+}
+
+static int gic_irq_get_irqchip_state(struct irq_data *d,
+                                    enum irqchip_irq_state which, bool *val)
+{
+       if (d->hwirq >= gic_data.irq_nr) /* PPI/SPI only */
+               return -EINVAL;
+
+       switch (which) {
+       case IRQCHIP_STATE_PENDING:
+               *val = gic_peek_irq(d, GICD_ISPENDR);
+               break;
+
+       case IRQCHIP_STATE_ACTIVE:
+               *val = gic_peek_irq(d, GICD_ISACTIVER);
+               break;
+
+       case IRQCHIP_STATE_MASKED:
+               *val = !gic_peek_irq(d, GICD_ISENABLER);
+               break;
+
+       default:
+               return -EINVAL;
+       }
+
+       return 0;
+}
+
+static void gic_eoi_irq(struct irq_data *d)
+{
+       gic_write_eoir(gic_irq(d));
+}
+
+static int gic_set_type(struct irq_data *d, unsigned int type)
+{
+       unsigned int irq = gic_irq(d);
+       void (*rwp_wait)(void);
+       void __iomem *base;
+
+       /* Interrupt configuration for SGIs can't be changed */
+       if (irq < 16)
+               return -EINVAL;
+
+       /* SPIs have restrictions on the supported types */
+       if (irq >= 32 && type != IRQ_TYPE_LEVEL_HIGH &&
+                        type != IRQ_TYPE_EDGE_RISING)
+               return -EINVAL;
+
+       if (gic_irq_in_rdist(d)) {
+               base = gic_data_rdist_sgi_base();
+               rwp_wait = gic_redist_wait_for_rwp;
+       } else {
+               base = gic_data.dist_base;
+               rwp_wait = gic_dist_wait_for_rwp;
+       }
+
+       return gic_configure_irq(irq, type, base, rwp_wait);
+}
+
+static u64 gic_mpidr_to_affinity(u64 mpidr)
+{
+       u64 aff;
+
+       aff = (MPIDR_AFFINITY_LEVEL(mpidr, 3) << 32 |
+              MPIDR_AFFINITY_LEVEL(mpidr, 2) << 16 |
+              MPIDR_AFFINITY_LEVEL(mpidr, 1) << 8  |
+              MPIDR_AFFINITY_LEVEL(mpidr, 0));
+
+       return aff;
+}
+
+static asmlinkage void __exception_irq_entry gic_handle_irq(struct pt_regs *regs)
+{
+       u64 irqnr;
+
+       do {
+               irqnr = gic_read_iar();
+
+               if (likely(irqnr > 15 && irqnr < 1020) || irqnr >= 8192) {
+                       int err;
+                       err = handle_domain_irq(gic_data.domain, irqnr, regs);
+                       if (err) {
+                               WARN_ONCE(true, "Unexpected interrupt received!\n");
+                               gic_write_eoir(irqnr);
+                       }
+                       continue;
+               }
+               if (irqnr < 16) {
+                       gic_write_eoir(irqnr);
+#ifdef CONFIG_SMP
+                       handle_IPI(irqnr, regs);
+#else
+                       WARN_ONCE(true, "Unexpected SGI received!\n");
+#endif
+                       continue;
+               }
+       } while (irqnr != ICC_IAR1_EL1_SPURIOUS);
+}
+
+static void __init gic_dist_init(void)
+{
+       unsigned int i;
+       u64 affinity;
+       void __iomem *base = gic_data.dist_base;
+
+       /* Disable the distributor */
+       writel_relaxed(0, base + GICD_CTLR);
+       gic_dist_wait_for_rwp();
+
+       gic_dist_config(base, gic_data.irq_nr, gic_dist_wait_for_rwp);
+
+       /* Enable distributor with ARE, Group1 */
+       writel_relaxed(GICD_CTLR_ARE_NS | GICD_CTLR_ENABLE_G1A | GICD_CTLR_ENABLE_G1,
+                      base + GICD_CTLR);
+
+       /*
+        * Set all global interrupts to the boot CPU only. ARE must be
+        * enabled.
+        */
+       affinity = gic_mpidr_to_affinity(cpu_logical_map(smp_processor_id()));
+       for (i = 32; i < gic_data.irq_nr; i++)
+               writeq_relaxed(affinity, base + GICD_IROUTER + i * 8);
+}
+
+static int gic_populate_rdist(void)
+{
+       u64 mpidr = cpu_logical_map(smp_processor_id());
+       u64 typer;
+       u32 aff;
+       int i;
+
+       /*
+        * Convert affinity to a 32bit value that can be matched to
+        * GICR_TYPER bits [63:32].
+        */
+       aff = (MPIDR_AFFINITY_LEVEL(mpidr, 3) << 24 |
+              MPIDR_AFFINITY_LEVEL(mpidr, 2) << 16 |
+              MPIDR_AFFINITY_LEVEL(mpidr, 1) << 8 |
+              MPIDR_AFFINITY_LEVEL(mpidr, 0));
+
+       for (i = 0; i < gic_data.nr_redist_regions; i++) {
+               void __iomem *ptr = gic_data.redist_regions[i].redist_base;
+               u32 reg;
+
+               reg = readl_relaxed(ptr + GICR_PIDR2) & GIC_PIDR2_ARCH_MASK;
+               if (reg != GIC_PIDR2_ARCH_GICv3 &&
+                   reg != GIC_PIDR2_ARCH_GICv4) { /* We're in trouble... */
+                       pr_warn("No redistributor present @%p\n", ptr);
+                       break;
+               }
+
+               do {
+                       typer = readq_relaxed(ptr + GICR_TYPER);
+                       if ((typer >> 32) == aff) {
+                               u64 offset = ptr - gic_data.redist_regions[i].redist_base;
+                               gic_data_rdist_rd_base() = ptr;
+                               gic_data_rdist()->phys_base = gic_data.redist_regions[i].phys_base + offset;
+                               pr_info("CPU%d: found redistributor %llx region %d:%pa\n",
+                                       smp_processor_id(),
+                                       (unsigned long long)mpidr,
+                                       i, &gic_data_rdist()->phys_base);
+                               return 0;
+                       }
+
+                       if (gic_data.redist_stride) {
+                               ptr += gic_data.redist_stride;
+                       } else {
+                               ptr += SZ_64K * 2; /* Skip RD_base + SGI_base */
+                               if (typer & GICR_TYPER_VLPIS)
+                                       ptr += SZ_64K * 2; /* Skip VLPI_base + reserved page */
+                       }
+               } while (!(typer & GICR_TYPER_LAST));
+       }
+
+       /* We couldn't even deal with ourselves... */
+       WARN(true, "CPU%d: mpidr %llx has no re-distributor!\n",
+            smp_processor_id(), (unsigned long long)mpidr);
+       return -ENODEV;
+}
+
+static void gic_cpu_sys_reg_init(void)
+{
+       /* Enable system registers */
+       gic_enable_sre();
+
+       /* Set priority mask register */
+       gic_write_pmr(DEFAULT_PMR_VALUE);
+
+       /* EOI deactivates interrupt too (mode 0) */
+       gic_write_ctlr(ICC_CTLR_EL1_EOImode_drop_dir);
+
+       /* ... and let's hit the road... */
+       gic_write_grpen1(1);
+}
+
+static int gic_dist_supports_lpis(void)
+{
+       return !!(readl_relaxed(gic_data.dist_base + GICD_TYPER) & GICD_TYPER_LPIS);
+}
+
+static void gic_cpu_init(void)
+{
+       void __iomem *rbase;
+
+       /* Register ourselves with the rest of the world */
+       if (gic_populate_rdist())
+               return;
+
+       gic_enable_redist(true);
+
+       rbase = gic_data_rdist_sgi_base();
+
+       gic_cpu_config(rbase, gic_redist_wait_for_rwp);
+
+       /* Give LPIs a spin */
+       if (IS_ENABLED(CONFIG_ARM_GIC_V3_ITS) && gic_dist_supports_lpis())
+               its_cpu_init();
+
+       /* initialise system registers */
+       gic_cpu_sys_reg_init();
+}
+
+#ifdef CONFIG_SMP
+static int gic_secondary_init(struct notifier_block *nfb,
+                             unsigned long action, void *hcpu)
+{
+       if (action == CPU_STARTING || action == CPU_STARTING_FROZEN)
+               gic_cpu_init();
+       return NOTIFY_OK;
+}
+
+/*
+ * Notifier for enabling the GIC CPU interface. Set an arbitrarily high
+ * priority because the GIC needs to be up before the ARM generic timers.
+ */
+static struct notifier_block gic_cpu_notifier = {
+       .notifier_call = gic_secondary_init,
+       .priority = 100,
+};
+
+static u16 gic_compute_target_list(int *base_cpu, const struct cpumask *mask,
+                                  u64 cluster_id)
+{
+       int cpu = *base_cpu;
+       u64 mpidr = cpu_logical_map(cpu);
+       u16 tlist = 0;
+
+       while (cpu < nr_cpu_ids) {
+               /*
+                * If we ever get a cluster of more than 16 CPUs, just
+                * scream and skip that CPU.
+                */
+               if (WARN_ON((mpidr & 0xff) >= 16))
+                       goto out;
+
+               tlist |= 1 << (mpidr & 0xf);
+
+               cpu = cpumask_next(cpu, mask);
+               if (cpu >= nr_cpu_ids)
+                       goto out;
+
+               mpidr = cpu_logical_map(cpu);
+
+               if (cluster_id != (mpidr & ~0xffUL)) {
+                       cpu--;
+                       goto out;
+               }
+       }
+out:
+       *base_cpu = cpu;
+       return tlist;
+}
+
+#define MPIDR_TO_SGI_AFFINITY(cluster_id, level) \
+       (MPIDR_AFFINITY_LEVEL(cluster_id, level) \
+               << ICC_SGI1R_AFFINITY_## level ##_SHIFT)
+
+static void gic_send_sgi(u64 cluster_id, u16 tlist, unsigned int irq)
+{
+       u64 val;
+
+       val = (MPIDR_TO_SGI_AFFINITY(cluster_id, 3)     |
+              MPIDR_TO_SGI_AFFINITY(cluster_id, 2)     |
+              irq << ICC_SGI1R_SGI_ID_SHIFT            |
+              MPIDR_TO_SGI_AFFINITY(cluster_id, 1)     |
+              tlist << ICC_SGI1R_TARGET_LIST_SHIFT);
+
+       pr_debug("CPU%d: ICC_SGI1R_EL1 %llx\n", smp_processor_id(), val);
+       gic_write_sgi1r(val);
+}
+
+static void gic_raise_softirq(const struct cpumask *mask, unsigned int irq)
+{
+       int cpu;
+
+       if (WARN_ON(irq >= 16))
+               return;
+
+       /*
+        * Ensure that stores to Normal memory are visible to the
+        * other CPUs before issuing the IPI.
+        */
+       smp_wmb();
+
+       for_each_cpu(cpu, mask) {
+               u64 cluster_id = cpu_logical_map(cpu) & ~0xffUL;
+               u16 tlist;
+
+               tlist = gic_compute_target_list(&cpu, mask, cluster_id);
+               gic_send_sgi(cluster_id, tlist, irq);
+       }
+
+       /* Force the above writes to ICC_SGI1R_EL1 to be executed */
+       isb();
+}
+
+static void gic_smp_init(void)
+{
+       set_smp_cross_call(gic_raise_softirq);
+       register_cpu_notifier(&gic_cpu_notifier);
+}
+
+static int gic_set_affinity(struct irq_data *d, const struct cpumask *mask_val,
+                           bool force)
+{
+       unsigned int cpu = cpumask_any_and(mask_val, cpu_online_mask);
+       void __iomem *reg;
+       int enabled;
+       u64 val;
+
+       if (gic_irq_in_rdist(d))
+               return -EINVAL;
+
+       /* If interrupt was enabled, disable it first */
+       enabled = gic_peek_irq(d, GICD_ISENABLER);
+       if (enabled)
+               gic_mask_irq(d);
+
+       reg = gic_dist_base(d) + GICD_IROUTER + (gic_irq(d) * 8);
+       val = gic_mpidr_to_affinity(cpu_logical_map(cpu));
+
+       writeq_relaxed(val, reg);
+
+       /*
+        * If the interrupt was enabled, enabled it again. Otherwise,
+        * just wait for the distributor to have digested our changes.
+        */
+       if (enabled)
+               gic_unmask_irq(d);
+       else
+               gic_dist_wait_for_rwp();
+
+       return IRQ_SET_MASK_OK;
+}
+#else
+#define gic_set_affinity       NULL
+#define gic_smp_init()         do { } while(0)
+#endif
+
+#ifdef CONFIG_CPU_PM
+static int gic_cpu_pm_notifier(struct notifier_block *self,
+                              unsigned long cmd, void *v)
+{
+       if (cmd == CPU_PM_EXIT) {
+               gic_enable_redist(true);
+               gic_cpu_sys_reg_init();
+       } else if (cmd == CPU_PM_ENTER) {
+               gic_write_grpen1(0);
+               gic_enable_redist(false);
+       }
+       return NOTIFY_OK;
+}
+
+static struct notifier_block gic_cpu_pm_notifier_block = {
+       .notifier_call = gic_cpu_pm_notifier,
+};
+
+static void gic_cpu_pm_init(void)
+{
+       cpu_pm_register_notifier(&gic_cpu_pm_notifier_block);
+}
+
+#else
+static inline void gic_cpu_pm_init(void) { }
+#endif /* CONFIG_CPU_PM */
+
+static struct irq_chip gic_chip = {
+       .name                   = "GICv3",
+       .irq_mask               = gic_mask_irq,
+       .irq_unmask             = gic_unmask_irq,
+       .irq_eoi                = gic_eoi_irq,
+       .irq_set_type           = gic_set_type,
+       .irq_set_affinity       = gic_set_affinity,
+       .irq_get_irqchip_state  = gic_irq_get_irqchip_state,
+       .irq_set_irqchip_state  = gic_irq_set_irqchip_state,
+};
+
+#define GIC_ID_NR              (1U << gic_data.rdists.id_bits)
+
+static int gic_irq_domain_map(struct irq_domain *d, unsigned int irq,
+                             irq_hw_number_t hw)
+{
+       /* SGIs are private to the core kernel */
+       if (hw < 16)
+               return -EPERM;
+       /* Nothing here */
+       if (hw >= gic_data.irq_nr && hw < 8192)
+               return -EPERM;
+       /* Off limits */
+       if (hw >= GIC_ID_NR)
+               return -EPERM;
+
+       /* PPIs */
+       if (hw < 32) {
+               irq_set_percpu_devid(irq);
+               irq_domain_set_info(d, irq, hw, &gic_chip, d->host_data,
+                                   handle_percpu_devid_irq, NULL, NULL);
+               set_irq_flags(irq, IRQF_VALID | IRQF_NOAUTOEN);
+       }
+       /* SPIs */
+       if (hw >= 32 && hw < gic_data.irq_nr) {
+               irq_domain_set_info(d, irq, hw, &gic_chip, d->host_data,
+                                   handle_fasteoi_irq, NULL, NULL);
+               set_irq_flags(irq, IRQF_VALID | IRQF_PROBE);
+       }
+       /* LPIs */
+       if (hw >= 8192 && hw < GIC_ID_NR) {
+               if (!gic_dist_supports_lpis())
+                       return -EPERM;
+               irq_domain_set_info(d, irq, hw, &gic_chip, d->host_data,
+                                   handle_fasteoi_irq, NULL, NULL);
+               set_irq_flags(irq, IRQF_VALID);
+       }
+
+       return 0;
+}
+
+static int gic_irq_domain_xlate(struct irq_domain *d,
+                               struct device_node *controller,
+                               const u32 *intspec, unsigned int intsize,
+                               unsigned long *out_hwirq, unsigned int *out_type)
+{
+       if (d->of_node != controller)
+               return -EINVAL;
+       if (intsize < 3)
+               return -EINVAL;
+
+       switch(intspec[0]) {
+       case 0:                 /* SPI */
+               *out_hwirq = intspec[1] + 32;
+               break;
+       case 1:                 /* PPI */
+               *out_hwirq = intspec[1] + 16;
+               break;
+       case GIC_IRQ_TYPE_LPI:  /* LPI */
+               *out_hwirq = intspec[1];
+               break;
+       default:
+               return -EINVAL;
+       }
+
+       *out_type = intspec[2] & IRQ_TYPE_SENSE_MASK;
+       return 0;
+}
+
+static int gic_irq_domain_alloc(struct irq_domain *domain, unsigned int virq,
+                               unsigned int nr_irqs, void *arg)
+{
+       int i, ret;
+       irq_hw_number_t hwirq;
+       unsigned int type = IRQ_TYPE_NONE;
+       struct of_phandle_args *irq_data = arg;
+
+       ret = gic_irq_domain_xlate(domain, irq_data->np, irq_data->args,
+                                  irq_data->args_count, &hwirq, &type);
+       if (ret)
+               return ret;
+
+       for (i = 0; i < nr_irqs; i++)
+               gic_irq_domain_map(domain, virq + i, hwirq + i);
+
+       return 0;
+}
+
+static void gic_irq_domain_free(struct irq_domain *domain, unsigned int virq,
+                               unsigned int nr_irqs)
+{
+       int i;
+
+       for (i = 0; i < nr_irqs; i++) {
+               struct irq_data *d = irq_domain_get_irq_data(domain, virq + i);
+               irq_set_handler(virq + i, NULL);
+               irq_domain_reset_irq_data(d);
+       }
+}
+
+static const struct irq_domain_ops gic_irq_domain_ops = {
+       .xlate = gic_irq_domain_xlate,
+       .alloc = gic_irq_domain_alloc,
+       .free = gic_irq_domain_free,
+};
+
+static int __init gic_of_init(struct device_node *node, struct device_node *parent)
+{
+       void __iomem *dist_base;
+       struct redist_region *rdist_regs;
+       u64 redist_stride;
+       u32 nr_redist_regions;
+       u32 typer;
+       u32 reg;
+       int gic_irqs;
+       int err;
+       int i;
+
+       dist_base = of_iomap(node, 0);
+       if (!dist_base) {
+               pr_err("%s: unable to map gic dist registers\n",
+                       node->full_name);
+               return -ENXIO;
+       }
+
+       reg = readl_relaxed(dist_base + GICD_PIDR2) & GIC_PIDR2_ARCH_MASK;
+       if (reg != GIC_PIDR2_ARCH_GICv3 && reg != GIC_PIDR2_ARCH_GICv4) {
+               pr_err("%s: no distributor detected, giving up\n",
+                       node->full_name);
+               err = -ENODEV;
+               goto out_unmap_dist;
+       }
+
+       if (of_property_read_u32(node, "#redistributor-regions", &nr_redist_regions))
+               nr_redist_regions = 1;
+
+       rdist_regs = kzalloc(sizeof(*rdist_regs) * nr_redist_regions, GFP_KERNEL);
+       if (!rdist_regs) {
+               err = -ENOMEM;
+               goto out_unmap_dist;
+       }
+
+       for (i = 0; i < nr_redist_regions; i++) {
+               struct resource res;
+               int ret;
+
+               ret = of_address_to_resource(node, 1 + i, &res);
+               rdist_regs[i].redist_base = of_iomap(node, 1 + i);
+               if (ret || !rdist_regs[i].redist_base) {
+                       pr_err("%s: couldn't map region %d\n",
+                              node->full_name, i);
+                       err = -ENODEV;
+                       goto out_unmap_rdist;
+               }
+               rdist_regs[i].phys_base = res.start;
+       }
+
+       if (of_property_read_u64(node, "redistributor-stride", &redist_stride))
+               redist_stride = 0;
+
+       gic_data.dist_base = dist_base;
+       gic_data.redist_regions = rdist_regs;
+       gic_data.nr_redist_regions = nr_redist_regions;
+       gic_data.redist_stride = redist_stride;
+
+       /*
+        * Find out how many interrupts are supported.
+        * The GIC only supports up to 1020 interrupt sources (SGI+PPI+SPI)
+        */
+       typer = readl_relaxed(gic_data.dist_base + GICD_TYPER);
+       gic_data.rdists.id_bits = GICD_TYPER_ID_BITS(typer);
+       gic_irqs = GICD_TYPER_IRQS(typer);
+       if (gic_irqs > 1020)
+               gic_irqs = 1020;
+       gic_data.irq_nr = gic_irqs;
+
+       gic_data.domain = irq_domain_add_tree(node, &gic_irq_domain_ops,
+                                             &gic_data);
+       gic_data.rdists.rdist = alloc_percpu(typeof(*gic_data.rdists.rdist));
+
+       if (WARN_ON(!gic_data.domain) || WARN_ON(!gic_data.rdists.rdist)) {
+               err = -ENOMEM;
+               goto out_free;
+       }
+
+       set_handle_irq(gic_handle_irq);
+
+       if (IS_ENABLED(CONFIG_ARM_GIC_V3_ITS) && gic_dist_supports_lpis())
+               its_init(node, &gic_data.rdists, gic_data.domain);
+
+       gic_smp_init();
+       gic_dist_init();
+       gic_cpu_init();
+       gic_cpu_pm_init();
+
+       return 0;
+
+out_free:
+       if (gic_data.domain)
+               irq_domain_remove(gic_data.domain);
+       free_percpu(gic_data.rdists.rdist);
+out_unmap_rdist:
+       for (i = 0; i < nr_redist_regions; i++)
+               if (rdist_regs[i].redist_base)
+                       iounmap(rdist_regs[i].redist_base);
+       kfree(rdist_regs);
+out_unmap_dist:
+       iounmap(dist_base);
+       return err;
+}
+
+IRQCHIP_DECLARE(gic_v3, "arm,gic-v3", gic_of_init);