]> asedeno.scripts.mit.edu Git - linux.git/commitdiff
iommu/hyper-v: Add Hyper-V stub IOMMU driver
authorLan Tianyu <Tianyu.Lan@microsoft.com>
Wed, 27 Feb 2019 14:54:04 +0000 (22:54 +0800)
committerJoerg Roedel <jroedel@suse.de>
Thu, 28 Feb 2019 10:12:16 +0000 (11:12 +0100)
On the bare metal, enabling X2APIC mode requires interrupt remapping
function which helps to deliver irq to cpu with 32-bit APIC ID.
Hyper-V doesn't provide interrupt remapping function so far and Hyper-V
MSI protocol already supports to deliver interrupt to the CPU whose
virtual processor index is more than 255. IO-APIC interrupt still has
8-bit APIC ID limitation.

This patch is to add Hyper-V stub IOMMU driver in order to enable
X2APIC mode successfully in Hyper-V Linux guest. The driver returns X2APIC
interrupt remapping capability when X2APIC mode is available. Otherwise,
it creates a Hyper-V irq domain to limit IO-APIC interrupts' affinity
and make sure cpus assigned with IO-APIC interrupt have 8-bit APIC ID.

Define 24 IO-APIC remapping entries because Hyper-V only expose one
single IO-APIC and one IO-APIC has 24 pins according IO-APIC spec(
https://pdos.csail.mit.edu/6.828/2016/readings/ia32/ioapic.pdf).

Reviewed-by: Michael Kelley <mikelley@microsoft.com>
Signed-off-by: Lan Tianyu <Tianyu.Lan@microsoft.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
drivers/iommu/Kconfig
drivers/iommu/Makefile
drivers/iommu/hyperv-iommu.c [new file with mode: 0644]
drivers/iommu/irq_remapping.c
drivers/iommu/irq_remapping.h

index d9a25715650e40565ea0ac99884a5f87b3317d06..34b066541491bc2e8296b1e3661da213d88b1b67 100644 (file)
@@ -435,4 +435,13 @@ config QCOM_IOMMU
        help
          Support for IOMMU on certain Qualcomm SoCs.
 
+config HYPERV_IOMMU
+       bool "Hyper-V x2APIC IRQ Handling"
+       depends on HYPERV
+       select IOMMU_API
+       default HYPERV
+       help
+         Stub IOMMU driver to handle IRQs as to allow Hyper-V Linux
+         guests to run with x2APIC mode enabled.
+
 endif # IOMMU_SUPPORT
index a158a68c8ea8af303bc4863b6c8a7086cf38f741..8c71a15e986b5dc4a068956c8454eb26f7ee05a6 100644 (file)
@@ -32,3 +32,4 @@ obj-$(CONFIG_EXYNOS_IOMMU) += exynos-iommu.o
 obj-$(CONFIG_FSL_PAMU) += fsl_pamu.o fsl_pamu_domain.o
 obj-$(CONFIG_S390_IOMMU) += s390-iommu.o
 obj-$(CONFIG_QCOM_IOMMU) += qcom_iommu.o
+obj-$(CONFIG_HYPERV_IOMMU) += hyperv-iommu.o
diff --git a/drivers/iommu/hyperv-iommu.c b/drivers/iommu/hyperv-iommu.c
new file mode 100644 (file)
index 0000000..a386b83
--- /dev/null
@@ -0,0 +1,196 @@
+// SPDX-License-Identifier: GPL-2.0
+
+/*
+ * Hyper-V stub IOMMU driver.
+ *
+ * Copyright (C) 2019, Microsoft, Inc.
+ *
+ * Author : Lan Tianyu <Tianyu.Lan@microsoft.com>
+ */
+
+#include <linux/types.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/iommu.h>
+#include <linux/module.h>
+
+#include <asm/apic.h>
+#include <asm/cpu.h>
+#include <asm/hw_irq.h>
+#include <asm/io_apic.h>
+#include <asm/irq_remapping.h>
+#include <asm/hypervisor.h>
+
+#include "irq_remapping.h"
+
+#ifdef CONFIG_IRQ_REMAP
+
+/*
+ * According 82093AA IO-APIC spec , IO APIC has a 24-entry Interrupt
+ * Redirection Table. Hyper-V exposes one single IO-APIC and so define
+ * 24 IO APIC remmapping entries.
+ */
+#define IOAPIC_REMAPPING_ENTRY 24
+
+static cpumask_t ioapic_max_cpumask = { CPU_BITS_NONE };
+static struct irq_domain *ioapic_ir_domain;
+
+static int hyperv_ir_set_affinity(struct irq_data *data,
+               const struct cpumask *mask, bool force)
+{
+       struct irq_data *parent = data->parent_data;
+       struct irq_cfg *cfg = irqd_cfg(data);
+       struct IO_APIC_route_entry *entry;
+       int ret;
+
+       /* Return error If new irq affinity is out of ioapic_max_cpumask. */
+       if (!cpumask_subset(mask, &ioapic_max_cpumask))
+               return -EINVAL;
+
+       ret = parent->chip->irq_set_affinity(parent, mask, force);
+       if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
+               return ret;
+
+       entry = data->chip_data;
+       entry->dest = cfg->dest_apicid;
+       entry->vector = cfg->vector;
+       send_cleanup_vector(cfg);
+
+       return 0;
+}
+
+static struct irq_chip hyperv_ir_chip = {
+       .name                   = "HYPERV-IR",
+       .irq_ack                = apic_ack_irq,
+       .irq_set_affinity       = hyperv_ir_set_affinity,
+};
+
+static int hyperv_irq_remapping_alloc(struct irq_domain *domain,
+                                    unsigned int virq, unsigned int nr_irqs,
+                                    void *arg)
+{
+       struct irq_alloc_info *info = arg;
+       struct irq_data *irq_data;
+       struct irq_desc *desc;
+       int ret = 0;
+
+       if (!info || info->type != X86_IRQ_ALLOC_TYPE_IOAPIC || nr_irqs > 1)
+               return -EINVAL;
+
+       ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
+       if (ret < 0)
+               return ret;
+
+       irq_data = irq_domain_get_irq_data(domain, virq);
+       if (!irq_data) {
+               irq_domain_free_irqs_common(domain, virq, nr_irqs);
+               return -EINVAL;
+       }
+
+       irq_data->chip = &hyperv_ir_chip;
+
+       /*
+        * If there is interrupt remapping function of IOMMU, setting irq
+        * affinity only needs to change IRTE of IOMMU. But Hyper-V doesn't
+        * support interrupt remapping function, setting irq affinity of IO-APIC
+        * interrupts still needs to change IO-APIC registers. But ioapic_
+        * configure_entry() will ignore value of cfg->vector and cfg->
+        * dest_apicid when IO-APIC's parent irq domain is not the vector
+        * domain.(See ioapic_configure_entry()) In order to setting vector
+        * and dest_apicid to IO-APIC register, IO-APIC entry pointer is saved
+        * in the chip_data and hyperv_irq_remapping_activate()/hyperv_ir_set_
+        * affinity() set vector and dest_apicid directly into IO-APIC entry.
+        */
+       irq_data->chip_data = info->ioapic_entry;
+
+       /*
+        * Hypver-V IO APIC irq affinity should be in the scope of
+        * ioapic_max_cpumask because no irq remapping support.
+        */
+       desc = irq_data_to_desc(irq_data);
+       cpumask_copy(desc->irq_common_data.affinity, &ioapic_max_cpumask);
+
+       return 0;
+}
+
+static void hyperv_irq_remapping_free(struct irq_domain *domain,
+                                unsigned int virq, unsigned int nr_irqs)
+{
+       irq_domain_free_irqs_common(domain, virq, nr_irqs);
+}
+
+static int hyperv_irq_remapping_activate(struct irq_domain *domain,
+                         struct irq_data *irq_data, bool reserve)
+{
+       struct irq_cfg *cfg = irqd_cfg(irq_data);
+       struct IO_APIC_route_entry *entry = irq_data->chip_data;
+
+       entry->dest = cfg->dest_apicid;
+       entry->vector = cfg->vector;
+
+       return 0;
+}
+
+static struct irq_domain_ops hyperv_ir_domain_ops = {
+       .alloc = hyperv_irq_remapping_alloc,
+       .free = hyperv_irq_remapping_free,
+       .activate = hyperv_irq_remapping_activate,
+};
+
+static int __init hyperv_prepare_irq_remapping(void)
+{
+       struct fwnode_handle *fn;
+       int i;
+
+       if (!hypervisor_is_type(X86_HYPER_MS_HYPERV) ||
+           !x2apic_supported())
+               return -ENODEV;
+
+       fn = irq_domain_alloc_named_id_fwnode("HYPERV-IR", 0);
+       if (!fn)
+               return -ENOMEM;
+
+       ioapic_ir_domain =
+               irq_domain_create_hierarchy(arch_get_ir_parent_domain(),
+                               0, IOAPIC_REMAPPING_ENTRY, fn,
+                               &hyperv_ir_domain_ops, NULL);
+
+       irq_domain_free_fwnode(fn);
+
+       /*
+        * Hyper-V doesn't provide irq remapping function for
+        * IO-APIC and so IO-APIC only accepts 8-bit APIC ID.
+        * Cpu's APIC ID is read from ACPI MADT table and APIC IDs
+        * in the MADT table on Hyper-v are sorted monotonic increasingly.
+        * APIC ID reflects cpu topology. There maybe some APIC ID
+        * gaps when cpu number in a socket is not power of two. Prepare
+        * max cpu affinity for IOAPIC irqs. Scan cpu 0-255 and set cpu
+        * into ioapic_max_cpumask if its APIC ID is less than 256.
+        */
+       for (i = min_t(unsigned int, num_possible_cpus() - 1, 255); i >= 0; i--)
+               if (cpu_physical_id(i) < 256)
+                       cpumask_set_cpu(i, &ioapic_max_cpumask);
+
+       return 0;
+}
+
+static int __init hyperv_enable_irq_remapping(void)
+{
+       return IRQ_REMAP_X2APIC_MODE;
+}
+
+static struct irq_domain *hyperv_get_ir_irq_domain(struct irq_alloc_info *info)
+{
+       if (info->type == X86_IRQ_ALLOC_TYPE_IOAPIC)
+               return ioapic_ir_domain;
+       else
+               return NULL;
+}
+
+struct irq_remap_ops hyperv_irq_remap_ops = {
+       .prepare                = hyperv_prepare_irq_remapping,
+       .enable                 = hyperv_enable_irq_remapping,
+       .get_ir_irq_domain      = hyperv_get_ir_irq_domain,
+};
+
+#endif
index b94ebd42edd851efd1bfce4e8c3c3a025e1a5a92..81cf2908c5314aae0985d70f8f2b26da53a9e03f 100644 (file)
@@ -103,6 +103,9 @@ int __init irq_remapping_prepare(void)
        else if (IS_ENABLED(CONFIG_AMD_IOMMU) &&
                 amd_iommu_irq_ops.prepare() == 0)
                remap_ops = &amd_iommu_irq_ops;
+       else if (IS_ENABLED(CONFIG_HYPERV_IOMMU) &&
+                hyperv_irq_remap_ops.prepare() == 0)
+               remap_ops = &hyperv_irq_remap_ops;
        else
                return -ENOSYS;
 
index 0afef6e43be40b96664e2f28cec97c38cbd0a683..f8609e9f1f5d3a8c890337f4da28dc054d8d33b2 100644 (file)
@@ -64,6 +64,7 @@ struct irq_remap_ops {
 
 extern struct irq_remap_ops intel_irq_remap_ops;
 extern struct irq_remap_ops amd_iommu_irq_ops;
+extern struct irq_remap_ops hyperv_irq_remap_ops;
 
 #else  /* CONFIG_IRQ_REMAP */