]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
irqchip/qcom: Add IRQ combiner driver
authorAgustin Vega-Frias <agustinv@codeaurora.org>
Thu, 2 Feb 2017 23:23:59 +0000 (18:23 -0500)
committerMarc Zyngier <marc.zyngier@arm.com>
Fri, 3 Feb 2017 15:03:49 +0000 (15:03 +0000)
Driver for interrupt combiners in the Top-level Control and Status
Registers (TCSR) hardware block in Qualcomm Technologies chips.

An interrupt combiner in this block combines a set of interrupts by
OR'ing the individual interrupt signals into a summary interrupt
signal routed to a parent interrupt controller, and provides read-
only, 32-bit registers to query the status of individual interrupts.
The status bit for IRQ n is bit (n % 32) within register (n / 32)
of the given combiner. Thus, each combiner can be described as a set
of register offsets and the number of IRQs managed.

Signed-off-by: Agustin Vega-Frias <agustinv@codeaurora.org>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
drivers/irqchip/Kconfig
drivers/irqchip/Makefile
drivers/irqchip/qcom-irq-combiner.c [new file with mode: 0644]

index ae96731cd2fbd4722d8c5f067592794b4c41bba1..125528f39e92c2377846f799a841e23ef1e98654 100644 (file)
@@ -283,3 +283,12 @@ config EZNPS_GIC
 config STM32_EXTI
        bool
        select IRQ_DOMAIN
+
+config QCOM_IRQ_COMBINER
+       bool "QCOM IRQ combiner support"
+       depends on ARCH_QCOM && ACPI
+       select IRQ_DOMAIN
+       select IRQ_DOMAIN_HIERARCHY
+       help
+         Say yes here to add support for the IRQ combiner devices embedded
+         in Qualcomm Technologies chips.
index 0e55d94065bf597c828aaa7d301132c7111d88f7..5a53186351efb7f695499e65fd6299b165e584d6 100644 (file)
@@ -75,3 +75,4 @@ obj-$(CONFIG_LS_SCFG_MSI)             += irq-ls-scfg-msi.o
 obj-$(CONFIG_EZNPS_GIC)                        += irq-eznps.o
 obj-$(CONFIG_ARCH_ASPEED)              += irq-aspeed-vic.o
 obj-$(CONFIG_STM32_EXTI)               += irq-stm32-exti.o
+obj-$(CONFIG_QCOM_IRQ_COMBINER)                += qcom-irq-combiner.o
diff --git a/drivers/irqchip/qcom-irq-combiner.c b/drivers/irqchip/qcom-irq-combiner.c
new file mode 100644 (file)
index 0000000..03251da
--- /dev/null
@@ -0,0 +1,296 @@
+/* Copyright (c) 2015-2016, The Linux Foundation. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 and
+ * only 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.
+ */
+
+/*
+ * Driver for interrupt combiners in the Top-level Control and Status
+ * Registers (TCSR) hardware block in Qualcomm Technologies chips.
+ * An interrupt combiner in this block combines a set of interrupts by
+ * OR'ing the individual interrupt signals into a summary interrupt
+ * signal routed to a parent interrupt controller, and provides read-
+ * only, 32-bit registers to query the status of individual interrupts.
+ * The status bit for IRQ n is bit (n % 32) within register (n / 32)
+ * of the given combiner. Thus, each combiner can be described as a set
+ * of register offsets and the number of IRQs managed.
+ */
+
+#define pr_fmt(fmt) "QCOM80B1:" fmt
+
+#include <linux/acpi.h>
+#include <linux/irqchip/chained_irq.h>
+#include <linux/irqdomain.h>
+#include <linux/platform_device.h>
+
+#define REG_SIZE 32
+
+struct combiner_reg {
+       void __iomem *addr;
+       unsigned long enabled;
+};
+
+struct combiner {
+       struct irq_domain   *domain;
+       int                 parent_irq;
+       u32                 nirqs;
+       u32                 nregs;
+       struct combiner_reg regs[0];
+};
+
+static inline int irq_nr(u32 reg, u32 bit)
+{
+       return reg * REG_SIZE + bit;
+}
+
+/*
+ * Handler for the cascaded IRQ.
+ */
+static void combiner_handle_irq(struct irq_desc *desc)
+{
+       struct combiner *combiner = irq_desc_get_handler_data(desc);
+       struct irq_chip *chip = irq_desc_get_chip(desc);
+       u32 reg;
+
+       chained_irq_enter(chip, desc);
+
+       for (reg = 0; reg < combiner->nregs; reg++) {
+               int virq;
+               int hwirq;
+               u32 bit;
+               u32 status;
+
+               bit = readl_relaxed(combiner->regs[reg].addr);
+               status = bit & combiner->regs[reg].enabled;
+               if (!status)
+                       pr_warn_ratelimited("Unexpected IRQ on CPU%d: (%08x %08lx %p)\n",
+                                           smp_processor_id(), bit,
+                                           combiner->regs[reg].enabled,
+                                           combiner->regs[reg].addr);
+
+               while (status) {
+                       bit = __ffs(status);
+                       status &= ~(1 << bit);
+                       hwirq = irq_nr(reg, bit);
+                       virq = irq_find_mapping(combiner->domain, hwirq);
+                       if (virq > 0)
+                               generic_handle_irq(virq);
+
+               }
+       }
+
+       chained_irq_exit(chip, desc);
+}
+
+static void combiner_irq_chip_mask_irq(struct irq_data *data)
+{
+       struct combiner *combiner = irq_data_get_irq_chip_data(data);
+       struct combiner_reg *reg = combiner->regs + data->hwirq / REG_SIZE;
+
+       clear_bit(data->hwirq % REG_SIZE, &reg->enabled);
+}
+
+static void combiner_irq_chip_unmask_irq(struct irq_data *data)
+{
+       struct combiner *combiner = irq_data_get_irq_chip_data(data);
+       struct combiner_reg *reg = combiner->regs + data->hwirq / REG_SIZE;
+
+       set_bit(data->hwirq % REG_SIZE, &reg->enabled);
+}
+
+static struct irq_chip irq_chip = {
+       .irq_mask = combiner_irq_chip_mask_irq,
+       .irq_unmask = combiner_irq_chip_unmask_irq,
+       .name = "qcom-irq-combiner"
+};
+
+static int combiner_irq_map(struct irq_domain *domain, unsigned int irq,
+                                  irq_hw_number_t hwirq)
+{
+       irq_set_chip_and_handler(irq, &irq_chip, handle_level_irq);
+       irq_set_chip_data(irq, domain->host_data);
+       irq_set_noprobe(irq);
+       return 0;
+}
+
+static void combiner_irq_unmap(struct irq_domain *domain, unsigned int irq)
+{
+       irq_domain_reset_irq_data(irq_get_irq_data(irq));
+}
+
+static int combiner_irq_translate(struct irq_domain *d, struct irq_fwspec *fws,
+                                 unsigned long *hwirq, unsigned int *type)
+{
+       struct combiner *combiner = d->host_data;
+
+       if (is_acpi_node(fws->fwnode)) {
+               if (WARN_ON((fws->param_count != 2) ||
+                           (fws->param[0] >= combiner->nirqs) ||
+                           (fws->param[1] & IORESOURCE_IRQ_LOWEDGE) ||
+                           (fws->param[1] & IORESOURCE_IRQ_HIGHEDGE)))
+                       return -EINVAL;
+
+               *hwirq = fws->param[0];
+               *type = fws->param[1];
+               return 0;
+       }
+
+       return -EINVAL;
+}
+
+static const struct irq_domain_ops domain_ops = {
+       .map = combiner_irq_map,
+       .unmap = combiner_irq_unmap,
+       .translate = combiner_irq_translate
+};
+
+static acpi_status count_registers_cb(struct acpi_resource *ares, void *context)
+{
+       int *count = context;
+
+       if (ares->type == ACPI_RESOURCE_TYPE_GENERIC_REGISTER)
+               ++(*count);
+       return AE_OK;
+}
+
+static int count_registers(struct platform_device *pdev)
+{
+       acpi_handle ahandle = ACPI_HANDLE(&pdev->dev);
+       acpi_status status;
+       int count = 0;
+
+       if (!acpi_has_method(ahandle, METHOD_NAME__CRS))
+               return -EINVAL;
+
+       status = acpi_walk_resources(ahandle, METHOD_NAME__CRS,
+                                    count_registers_cb, &count);
+       if (ACPI_FAILURE(status))
+               return -EINVAL;
+       return count;
+}
+
+struct get_registers_context {
+       struct device *dev;
+       struct combiner *combiner;
+       int err;
+};
+
+static acpi_status get_registers_cb(struct acpi_resource *ares, void *context)
+{
+       struct get_registers_context *ctx = context;
+       struct acpi_resource_generic_register *reg;
+       phys_addr_t paddr;
+       void __iomem *vaddr;
+
+       if (ares->type != ACPI_RESOURCE_TYPE_GENERIC_REGISTER)
+               return AE_OK;
+
+       reg = &ares->data.generic_reg;
+       paddr = reg->address;
+       if ((reg->space_id != ACPI_SPACE_MEM) ||
+           (reg->bit_offset != 0) ||
+           (reg->bit_width > REG_SIZE)) {
+               dev_err(ctx->dev, "Bad register resource @%pa\n", &paddr);
+               ctx->err = -EINVAL;
+               return AE_ERROR;
+       }
+
+       vaddr = devm_ioremap(ctx->dev, reg->address, REG_SIZE);
+       if (IS_ERR(vaddr)) {
+               dev_err(ctx->dev, "Can't map register @%pa\n", &paddr);
+               ctx->err = PTR_ERR(vaddr);
+               return AE_ERROR;
+       }
+
+       ctx->combiner->regs[ctx->combiner->nregs].addr = vaddr;
+       ctx->combiner->nirqs += reg->bit_width;
+       ctx->combiner->nregs++;
+       return AE_OK;
+}
+
+static int get_registers(struct platform_device *pdev, struct combiner *comb)
+{
+       acpi_handle ahandle = ACPI_HANDLE(&pdev->dev);
+       acpi_status status;
+       struct get_registers_context ctx;
+
+       if (!acpi_has_method(ahandle, METHOD_NAME__CRS))
+               return -EINVAL;
+
+       ctx.dev = &pdev->dev;
+       ctx.combiner = comb;
+       ctx.err = 0;
+
+       status = acpi_walk_resources(ahandle, METHOD_NAME__CRS,
+                                    get_registers_cb, &ctx);
+       if (ACPI_FAILURE(status))
+               return ctx.err;
+       return 0;
+}
+
+static int __init combiner_probe(struct platform_device *pdev)
+{
+       struct combiner *combiner;
+       size_t alloc_sz;
+       u32 nregs;
+       int err;
+
+       nregs = count_registers(pdev);
+       if (nregs <= 0) {
+               dev_err(&pdev->dev, "Error reading register resources\n");
+               return -EINVAL;
+       }
+
+       alloc_sz = sizeof(*combiner) + sizeof(struct combiner_reg) * nregs;
+       combiner = devm_kzalloc(&pdev->dev, alloc_sz, GFP_KERNEL);
+       if (!combiner)
+               return -ENOMEM;
+
+       err = get_registers(pdev, combiner);
+       if (err < 0)
+               return err;
+
+       combiner->parent_irq = platform_get_irq(pdev, 0);
+       if (combiner->parent_irq <= 0) {
+               dev_err(&pdev->dev, "Error getting IRQ resource\n");
+               return -EPROBE_DEFER;
+       }
+
+       combiner->domain = irq_domain_create_linear(pdev->dev.fwnode, combiner->nirqs,
+                                                   &domain_ops, combiner);
+       if (!combiner->domain)
+               /* Errors printed by irq_domain_create_linear */
+               return -ENODEV;
+
+       irq_set_chained_handler_and_data(combiner->parent_irq,
+                                        combiner_handle_irq, combiner);
+
+       dev_info(&pdev->dev, "Initialized with [p=%d,n=%d,r=%p]\n",
+                combiner->parent_irq, combiner->nirqs, combiner->regs[0].addr);
+       return 0;
+}
+
+static const struct acpi_device_id qcom_irq_combiner_ids[] = {
+       { "QCOM80B1", },
+       { }
+};
+
+static struct platform_driver qcom_irq_combiner_probe = {
+       .driver = {
+               .name = "qcom-irq-combiner",
+               .acpi_match_table = ACPI_PTR(qcom_irq_combiner_ids),
+       },
+       .probe = combiner_probe,
+};
+
+static int __init register_qcom_irq_combiner(void)
+{
+       return platform_driver_register(&qcom_irq_combiner_probe);
+}
+device_initcall(register_qcom_irq_combiner);