]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/commitdiff
drivers/perf: arm_pmu: manage interrupts per-cpu
authorMark Rutland <mark.rutland@arm.com>
Fri, 10 Mar 2017 10:46:14 +0000 (10:46 +0000)
committerThadeu Lima de Souza Cascardo <cascardo@canonical.com>
Wed, 17 May 2017 16:40:57 +0000 (13:40 -0300)
BugLink: https://bugs.launchpad.net/bugs/1689661
When requesting or freeing interrupts, we use platform_get_irq() to find
relevant irqs, backing this up with additional information in an
optional irq_affinity table.

This means that our irq request and free paths are tied to a
platform_device, and our request path must jump through a number of
hoops in order to determine the required affinity of each interrupt.

Given that the affinity must be static, we can compute the affinity once
up-front at probe time, simplifying the irq request and free paths. By
recording interrupts in a per-cpu data structure, we simplify a few
paths, and permit a subsequent rework of the request and free paths.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
[will: rename local nr_irqs variable to avoid conflict with global]
Signed-off-by: Will Deacon <will.deacon@arm.com>
(cherry picked from commit 7ed98e0168bd23d8ea3294e95254cc5b4000c948)
Signed-off-by: dann frazier <dann.frazier@canonical.com>
Acked-by: Seth Forshee <seth.forshee@canonical.com>
Acked-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com>
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com>
drivers/perf/arm_pmu.c
include/linux/perf/arm_pmu.h

index 9735b2222ac87c9dcc45d428322577eda0f20d81..419ed27e1cac2a442a8a92a6d1f5d8dae7d76a67 100644 (file)
@@ -616,94 +616,76 @@ static void cpu_pmu_disable_percpu_irq(void *data)
 
 static void cpu_pmu_free_irq(struct arm_pmu *cpu_pmu)
 {
-       int i, irq, irqs;
-       struct platform_device *pmu_device = cpu_pmu->plat_device;
+       int cpu;
        struct pmu_hw_events __percpu *hw_events = cpu_pmu->hw_events;
 
-       irqs = min(pmu_device->num_resources, num_possible_cpus());
-
-       irq = platform_get_irq(pmu_device, 0);
-       if (irq > 0 && irq_is_percpu(irq)) {
-               on_each_cpu_mask(&cpu_pmu->supported_cpus,
-                                cpu_pmu_disable_percpu_irq, &irq, 1);
-               free_percpu_irq(irq, &hw_events->percpu_pmu);
-       } else {
-               for (i = 0; i < irqs; ++i) {
-                       int cpu = i;
-
-                       if (cpu_pmu->irq_affinity)
-                               cpu = cpu_pmu->irq_affinity[i];
-
-                       if (!cpumask_test_and_clear_cpu(cpu, &cpu_pmu->active_irqs))
-                               continue;
-                       irq = platform_get_irq(pmu_device, i);
-                       if (irq > 0)
-                               free_irq(irq, per_cpu_ptr(&hw_events->percpu_pmu, cpu));
+       for_each_cpu(cpu, &cpu_pmu->supported_cpus) {
+               int irq = per_cpu(hw_events->irq, cpu);
+               if (!irq)
+                       continue;
+
+               if (irq_is_percpu(irq)) {
+                       on_each_cpu_mask(&cpu_pmu->supported_cpus,
+                                        cpu_pmu_disable_percpu_irq, &irq, 1);
+                       free_percpu_irq(irq, &hw_events->percpu_pmu);
+
+                       break;
                }
+
+               if (!cpumask_test_and_clear_cpu(cpu, &cpu_pmu->active_irqs))
+                       continue;
+
+               free_irq(irq, per_cpu_ptr(&hw_events->percpu_pmu, cpu));
        }
 }
 
 static int cpu_pmu_request_irq(struct arm_pmu *cpu_pmu, irq_handler_t handler)
 {
-       int i, err, irq, irqs;
-       struct platform_device *pmu_device = cpu_pmu->plat_device;
+       int cpu, err;
        struct pmu_hw_events __percpu *hw_events = cpu_pmu->hw_events;
 
-       if (!pmu_device)
-               return -ENODEV;
-
-       irqs = min(pmu_device->num_resources, num_possible_cpus());
-       if (irqs < 1) {
-               pr_warn_once("perf/ARM: No irqs for PMU defined, sampling events not supported\n");
-               return 0;
-       }
-
-       irq = platform_get_irq(pmu_device, 0);
-       if (irq > 0 && irq_is_percpu(irq)) {
-               err = request_percpu_irq(irq, handler, "arm-pmu",
-                                        &hw_events->percpu_pmu);
-               if (err) {
-                       pr_err("unable to request IRQ%d for ARM PMU counters\n",
-                               irq);
-                       return err;
-               }
-
-               on_each_cpu_mask(&cpu_pmu->supported_cpus,
-                                cpu_pmu_enable_percpu_irq, &irq, 1);
-       } else {
-               for (i = 0; i < irqs; ++i) {
-                       int cpu = i;
-
-                       err = 0;
-                       irq = platform_get_irq(pmu_device, i);
-                       if (irq < 0)
-                               continue;
-
-                       if (cpu_pmu->irq_affinity)
-                               cpu = cpu_pmu->irq_affinity[i];
-
-                       /*
-                        * If we have a single PMU interrupt that we can't shift,
-                        * assume that we're running on a uniprocessor machine and
-                        * continue. Otherwise, continue without this interrupt.
-                        */
-                       if (irq_set_affinity(irq, cpumask_of(cpu)) && irqs > 1) {
-                               pr_warn("unable to set irq affinity (irq=%d, cpu=%u)\n",
-                                       irq, cpu);
-                               continue;
-                       }
+       for_each_cpu(cpu, &cpu_pmu->supported_cpus) {
+               int irq = per_cpu(hw_events->irq, cpu);
+               if (!irq)
+                       continue;
 
-                       err = request_irq(irq, handler,
-                                         IRQF_NOBALANCING | IRQF_NO_THREAD, "arm-pmu",
-                                         per_cpu_ptr(&hw_events->percpu_pmu, cpu));
+               if (irq_is_percpu(irq)) {
+                       err = request_percpu_irq(irq, handler, "arm-pmu",
+                                                &hw_events->percpu_pmu);
                        if (err) {
                                pr_err("unable to request IRQ%d for ARM PMU counters\n",
                                        irq);
                                return err;
                        }
 
-                       cpumask_set_cpu(cpu, &cpu_pmu->active_irqs);
+                       on_each_cpu_mask(&cpu_pmu->supported_cpus,
+                                        cpu_pmu_enable_percpu_irq, &irq, 1);
+
+                       break;
                }
+
+               /*
+                * If we have a single PMU interrupt that we can't shift,
+                * assume that we're running on a uniprocessor machine and
+                * continue. Otherwise, continue without this interrupt.
+                */
+               if (irq_set_affinity(irq, cpumask_of(cpu)) &&
+                   num_possible_cpus() > 1) {
+                       pr_warn("unable to set irq affinity (irq=%d, cpu=%u)\n",
+                               irq, cpu);
+                       continue;
+               }
+
+               err = request_irq(irq, handler,
+                                 IRQF_NOBALANCING | IRQF_NO_THREAD, "arm-pmu",
+                                 per_cpu_ptr(&hw_events->percpu_pmu, cpu));
+               if (err) {
+                       pr_err("unable to request IRQ%d for ARM PMU counters\n",
+                               irq);
+                       return err;
+               }
+
+               cpumask_set_cpu(cpu, &cpu_pmu->active_irqs);
        }
 
        return 0;
@@ -845,10 +827,6 @@ static int cpu_pmu_init(struct arm_pmu *cpu_pmu)
                on_each_cpu_mask(&cpu_pmu->supported_cpus, cpu_pmu->reset,
                         cpu_pmu, 1);
 
-       /* If no interrupts available, set the corresponding capability flag */
-       if (!platform_get_irq(cpu_pmu->plat_device, 0))
-               cpu_pmu->pmu.capabilities |= PERF_PMU_CAP_NO_INTERRUPT;
-
        /*
         * This is a CPU PMU potentially in a heterogeneous configuration (e.g.
         * big.LITTLE). This is not an uncore PMU, and we have taken ctx
@@ -896,98 +874,133 @@ static int probe_current_pmu(struct arm_pmu *pmu,
        return ret;
 }
 
-static int of_pmu_irq_cfg(struct arm_pmu *pmu)
+static int pmu_parse_percpu_irq(struct arm_pmu *pmu, int irq)
 {
-       int *irqs, i = 0;
-       bool using_spi = false;
-       struct platform_device *pdev = pmu->plat_device;
+       int cpu, ret;
+       struct pmu_hw_events __percpu *hw_events = pmu->hw_events;
 
-       irqs = kcalloc(pdev->num_resources, sizeof(*irqs), GFP_KERNEL);
-       if (!irqs)
-               return -ENOMEM;
+       ret = irq_get_percpu_devid_partition(irq, &pmu->supported_cpus);
+       if (ret)
+               return ret;
 
-       do {
-               struct device_node *dn;
-               int cpu, irq;
+       for_each_cpu(cpu, &pmu->supported_cpus)
+               per_cpu(hw_events->irq, cpu) = irq;
 
-               /* See if we have an affinity entry */
-               dn = of_parse_phandle(pdev->dev.of_node, "interrupt-affinity", i);
-               if (!dn)
-                       break;
+       return 0;
+}
 
-               /* Check the IRQ type and prohibit a mix of PPIs and SPIs */
-               irq = platform_get_irq(pdev, i);
-               if (irq > 0) {
-                       bool spi = !irq_is_percpu(irq);
-
-                       if (i > 0 && spi != using_spi) {
-                               pr_err("PPI/SPI IRQ type mismatch for %s!\n",
-                                       dn->name);
-                               of_node_put(dn);
-                               kfree(irqs);
-                               return -EINVAL;
-                       }
+static bool pmu_has_irq_affinity(struct device_node *node)
+{
+       return !!of_find_property(node, "interrupt-affinity", NULL);
+}
 
-                       using_spi = spi;
-               }
+static int pmu_parse_irq_affinity(struct device_node *node, int i)
+{
+       struct device_node *dn;
+       int cpu;
 
-               /* Now look up the logical CPU number */
-               for_each_possible_cpu(cpu) {
-                       struct device_node *cpu_dn;
+       /*
+        * If we don't have an interrupt-affinity property, we guess irq
+        * affinity matches our logical CPU order, as we used to assume.
+        * This is fragile, so we'll warn in pmu_parse_irqs().
+        */
+       if (!pmu_has_irq_affinity(node))
+               return i;
 
-                       cpu_dn = of_cpu_device_node_get(cpu);
-                       of_node_put(cpu_dn);
+       dn = of_parse_phandle(node, "interrupt-affinity", i);
+       if (!dn) {
+               pr_warn("failed to parse interrupt-affinity[%d] for %s\n",
+                       i, node->name);
+               return -EINVAL;
+       }
 
-                       if (dn == cpu_dn)
-                               break;
-               }
+       /* Now look up the logical CPU number */
+       for_each_possible_cpu(cpu) {
+               struct device_node *cpu_dn;
+
+               cpu_dn = of_cpu_device_node_get(cpu);
+               of_node_put(cpu_dn);
 
-               if (cpu >= nr_cpu_ids) {
-                       pr_warn("Failed to find logical CPU for %s\n",
-                               dn->name);
-                       of_node_put(dn);
-                       cpumask_setall(&pmu->supported_cpus);
+               if (dn == cpu_dn)
                        break;
-               }
-               of_node_put(dn);
+       }
 
-               /* For SPIs, we need to track the affinity per IRQ */
-               if (using_spi) {
-                       if (i >= pdev->num_resources)
-                               break;
+       if (cpu >= nr_cpu_ids) {
+               pr_warn("failed to find logical CPU for %s\n", dn->name);
+       }
 
-                       irqs[i] = cpu;
-               }
+       of_node_put(dn);
 
-               /* Keep track of the CPUs containing this PMU type */
-               cpumask_set_cpu(cpu, &pmu->supported_cpus);
-               i++;
-       } while (1);
+       return cpu;
+}
+
+static int pmu_parse_irqs(struct arm_pmu *pmu)
+{
+       int i = 0, irqs;
+       struct platform_device *pdev = pmu->plat_device;
+       struct pmu_hw_events __percpu *hw_events = pmu->hw_events;
+
+       irqs = platform_irq_count(pdev);
+       if (irqs < 0) {
+               pr_err("unable to count PMU IRQs\n");
+               return irqs;
+       }
+
+       /*
+        * In this case we have no idea which CPUs are covered by the PMU.
+        * To match our prior behaviour, we assume all CPUs in this case.
+        */
+       if (irqs == 0) {
+               pr_warn("no irqs for PMU, sampling events not supported\n");
+               pmu->pmu.capabilities |= PERF_PMU_CAP_NO_INTERRUPT;
+               cpumask_setall(&pmu->supported_cpus);
+               return 0;
+       }
 
-       /* If we didn't manage to parse anything, try the interrupt affinity */
-       if (cpumask_weight(&pmu->supported_cpus) == 0) {
+       if (irqs == 1) {
                int irq = platform_get_irq(pdev, 0);
+               if (irq && irq_is_percpu(irq))
+                       return pmu_parse_percpu_irq(pmu, irq);
+       }
 
-               if (irq > 0 && irq_is_percpu(irq)) {
-                       /* If using PPIs, check the affinity of the partition */
-                       int ret;
+       if (!pmu_has_irq_affinity(pdev->dev.of_node)) {
+               pr_warn("no interrupt-affinity property for %s, guessing.\n",
+                       of_node_full_name(pdev->dev.of_node));
+       }
 
-                       ret = irq_get_percpu_devid_partition(irq, &pmu->supported_cpus);
-                       if (ret) {
-                               kfree(irqs);
-                               return ret;
-                       }
-               } else {
-                       /* Otherwise default to all CPUs */
-                       cpumask_setall(&pmu->supported_cpus);
+       /*
+        * Some platforms have all PMU IRQs OR'd into a single IRQ, with a
+        * special platdata function that attempts to demux them.
+        */
+       if (dev_get_platdata(&pdev->dev))
+               cpumask_setall(&pmu->supported_cpus);
+
+       for (i = 0; i < irqs; i++) {
+               int cpu, irq;
+
+               irq = platform_get_irq(pdev, i);
+               if (WARN_ON(irq <= 0))
+                       continue;
+
+               if (irq_is_percpu(irq)) {
+                       pr_warn("multiple PPIs or mismatched SPI/PPI detected\n");
+                       return -EINVAL;
                }
-       }
 
-       /* If we matched up the IRQ affinities, use them to route the SPIs */
-       if (using_spi && i == pdev->num_resources)
-               pmu->irq_affinity = irqs;
-       else
-               kfree(irqs);
+               cpu = pmu_parse_irq_affinity(pdev->dev.of_node, i);
+               if (cpu < 0)
+                       return cpu;
+               if (cpu >= nr_cpu_ids)
+                       continue;
+
+               if (per_cpu(hw_events->irq, cpu)) {
+                       pr_warn("multiple PMU IRQs for the same CPU detected\n");
+                       return -EINVAL;
+               }
+
+               per_cpu(hw_events->irq, cpu) = irq;
+               cpumask_set_cpu(cpu, &pmu->supported_cpus);
+       }
 
        return 0;
 }
@@ -1049,6 +1062,10 @@ int arm_pmu_device_probe(struct platform_device *pdev,
 
        pmu->plat_device = pdev;
 
+       ret = pmu_parse_irqs(pmu);
+       if (ret)
+               goto out_free;
+
        if (node && (of_id = of_match_node(of_table, pdev->dev.of_node))) {
                init_fn = of_id->data;
 
@@ -1061,9 +1078,7 @@ int arm_pmu_device_probe(struct platform_device *pdev,
                        pmu->secure_access = false;
                }
 
-               ret = of_pmu_irq_cfg(pmu);
-               if (!ret)
-                       ret = init_fn(pmu);
+               ret = init_fn(pmu);
        } else if (probe_table) {
                cpumask_setall(&pmu->supported_cpus);
                ret = probe_current_pmu(pmu, probe_table);
@@ -1096,7 +1111,6 @@ out_destroy:
 out_free:
        pr_info("%s: failed to register PMU devices!\n",
                of_node_full_name(node));
-       kfree(pmu->irq_affinity);
        armpmu_free(pmu);
        return ret;
 }
index 8462da2660899153552df914268ddaadd448fc2d..05a3eb447fc8e2b71bd01403b52f4cdc42e32b2d 100644 (file)
@@ -75,6 +75,8 @@ struct pmu_hw_events {
         * already have to allocate this struct per cpu.
         */
        struct arm_pmu          *percpu_pmu;
+
+       int irq;
 };
 
 enum armpmu_attr_groups {
@@ -88,7 +90,6 @@ struct arm_pmu {
        struct pmu      pmu;
        cpumask_t       active_irqs;
        cpumask_t       supported_cpus;
-       int             *irq_affinity;
        char            *name;
        irqreturn_t     (*handle_irq)(int irq_num, void *dev);
        void            (*enable)(struct perf_event *event);