intel-iommu: Make "Unknown DMAR structure" message more informative
[safe/jmp/linux-2.6] / drivers / pci / intr_remapping.c
index 6961be8..0ed78a7 100644 (file)
@@ -5,13 +5,26 @@
 #include <linux/pci.h>
 #include <linux/irq.h>
 #include <asm/io_apic.h>
-#include "intel-iommu.h"
+#include <asm/smp.h>
+#include <asm/cpu.h>
+#include <linux/intel-iommu.h>
 #include "intr_remapping.h"
+#include <acpi/acpi.h>
+#include <asm/pci-direct.h>
+#include "pci.h"
 
 static struct ioapic_scope ir_ioapic[MAX_IO_APICS];
 static int ir_ioapic_num;
 int intr_remapping_enabled;
 
+static int disable_intremap;
+static __init int setup_nointremap(char *str)
+{
+       disable_intremap = 1;
+       return 0;
+}
+early_param("nointremap", setup_nointremap);
+
 struct irq_2_iommu {
        struct intel_iommu *iommu;
        u16 irte_index;
@@ -19,56 +32,128 @@ struct irq_2_iommu {
        u8  irte_mask;
 };
 
-#ifdef CONFIG_HAVE_DYNA_ARRAY
-static struct irq_2_iommu *irq_2_iommu;
-DEFINE_DYN_ARRAY(irq_2_iommu, sizeof(struct irq_2_iommu), nr_irqs, PAGE_SIZE, NULL);
-#else
-static struct irq_2_iommu irq_2_iommu[NR_IRQS];
+#ifdef CONFIG_GENERIC_HARDIRQS
+static struct irq_2_iommu *get_one_free_irq_2_iommu(int node)
+{
+       struct irq_2_iommu *iommu;
+
+       iommu = kzalloc_node(sizeof(*iommu), GFP_ATOMIC, node);
+       printk(KERN_DEBUG "alloc irq_2_iommu on node %d\n", node);
+
+       return iommu;
+}
+
+static struct irq_2_iommu *irq_2_iommu(unsigned int irq)
+{
+       struct irq_desc *desc;
+
+       desc = irq_to_desc(irq);
+
+       if (WARN_ON_ONCE(!desc))
+               return NULL;
+
+       return desc->irq_2_iommu;
+}
+
+static struct irq_2_iommu *irq_2_iommu_alloc(unsigned int irq)
+{
+       struct irq_desc *desc;
+       struct irq_2_iommu *irq_iommu;
+
+       desc = irq_to_desc(irq);
+       if (!desc) {
+               printk(KERN_INFO "can not get irq_desc for %d\n", irq);
+               return NULL;
+       }
+
+       irq_iommu = desc->irq_2_iommu;
+
+       if (!irq_iommu)
+               desc->irq_2_iommu = get_one_free_irq_2_iommu(irq_node(irq));
+
+       return desc->irq_2_iommu;
+}
+
+#else /* !CONFIG_SPARSE_IRQ */
+
+static struct irq_2_iommu irq_2_iommuX[NR_IRQS];
+
+static struct irq_2_iommu *irq_2_iommu(unsigned int irq)
+{
+       if (irq < nr_irqs)
+               return &irq_2_iommuX[irq];
+
+       return NULL;
+}
+static struct irq_2_iommu *irq_2_iommu_alloc(unsigned int irq)
+{
+       return irq_2_iommu(irq);
+}
 #endif
 
 static DEFINE_SPINLOCK(irq_2_ir_lock);
 
-int irq_remapped(int irq)
+static struct irq_2_iommu *valid_irq_2_iommu(unsigned int irq)
 {
-       if (irq > nr_irqs)
-               return 0;
+       struct irq_2_iommu *irq_iommu;
 
-       if (!irq_2_iommu[irq].iommu)
-               return 0;
+       irq_iommu = irq_2_iommu(irq);
 
-       return 1;
+       if (!irq_iommu)
+               return NULL;
+
+       if (!irq_iommu->iommu)
+               return NULL;
+
+       return irq_iommu;
+}
+
+int irq_remapped(int irq)
+{
+       return valid_irq_2_iommu(irq) != NULL;
 }
 
 int get_irte(int irq, struct irte *entry)
 {
        int index;
+       struct irq_2_iommu *irq_iommu;
+       unsigned long flags;
 
-       if (!entry || irq > nr_irqs)
+       if (!entry)
                return -1;
 
-       spin_lock(&irq_2_ir_lock);
-       if (!irq_2_iommu[irq].iommu) {
-               spin_unlock(&irq_2_ir_lock);
+       spin_lock_irqsave(&irq_2_ir_lock, flags);
+       irq_iommu = valid_irq_2_iommu(irq);
+       if (!irq_iommu) {
+               spin_unlock_irqrestore(&irq_2_ir_lock, flags);
                return -1;
        }
 
-       index = irq_2_iommu[irq].irte_index + irq_2_iommu[irq].sub_handle;
-       *entry = *(irq_2_iommu[irq].iommu->ir_table->base + index);
+       index = irq_iommu->irte_index + irq_iommu->sub_handle;
+       *entry = *(irq_iommu->iommu->ir_table->base + index);
 
-       spin_unlock(&irq_2_ir_lock);
+       spin_unlock_irqrestore(&irq_2_ir_lock, flags);
        return 0;
 }
 
 int alloc_irte(struct intel_iommu *iommu, int irq, u16 count)
 {
        struct ir_table *table = iommu->ir_table;
+       struct irq_2_iommu *irq_iommu;
        u16 index, start_index;
        unsigned int mask = 0;
+       unsigned long flags;
        int i;
 
        if (!count)
                return -1;
 
+#ifndef CONFIG_SPARSE_IRQ
+       /* protect irq_2_iommu_alloc later */
+       if (irq >= nr_irqs)
+               return -1;
+#endif
+
        /*
         * start the IRTE search from index 0.
         */
@@ -87,7 +172,7 @@ int alloc_irte(struct intel_iommu *iommu, int irq, u16 count)
                return -1;
        }
 
-       spin_lock(&irq_2_ir_lock);
+       spin_lock_irqsave(&irq_2_ir_lock, flags);
        do {
                for (i = index; i < index + count; i++)
                        if  (table->base[i].present)
@@ -99,7 +184,7 @@ int alloc_irte(struct intel_iommu *iommu, int irq, u16 count)
                index = (index + count) % INTR_REMAP_TABLE_ENTRIES;
 
                if (index == start_index) {
-                       spin_unlock(&irq_2_ir_lock);
+                       spin_unlock_irqrestore(&irq_2_ir_lock, flags);
                        printk(KERN_ERR "can't allocate an IRTE\n");
                        return -1;
                }
@@ -108,17 +193,24 @@ int alloc_irte(struct intel_iommu *iommu, int irq, u16 count)
        for (i = index; i < index + count; i++)
                table->base[i].present = 1;
 
-       irq_2_iommu[irq].iommu = iommu;
-       irq_2_iommu[irq].irte_index =  index;
-       irq_2_iommu[irq].sub_handle = 0;
-       irq_2_iommu[irq].irte_mask = mask;
+       irq_iommu = irq_2_iommu_alloc(irq);
+       if (!irq_iommu) {
+               spin_unlock_irqrestore(&irq_2_ir_lock, flags);
+               printk(KERN_ERR "can't allocate irq_2_iommu\n");
+               return -1;
+       }
 
-       spin_unlock(&irq_2_ir_lock);
+       irq_iommu->iommu = iommu;
+       irq_iommu->irte_index =  index;
+       irq_iommu->sub_handle = 0;
+       irq_iommu->irte_mask = mask;
+
+       spin_unlock_irqrestore(&irq_2_ir_lock, flags);
 
        return index;
 }
 
-static void qi_flush_iec(struct intel_iommu *iommu, int index, int mask)
+static int qi_flush_iec(struct intel_iommu *iommu, int index, int mask)
 {
        struct qi_desc desc;
 
@@ -126,106 +218,129 @@ static void qi_flush_iec(struct intel_iommu *iommu, int index, int mask)
                   | QI_IEC_SELECTIVE;
        desc.high = 0;
 
-       qi_submit_sync(&desc, iommu);
+       return qi_submit_sync(&desc, iommu);
 }
 
 int map_irq_to_irte_handle(int irq, u16 *sub_handle)
 {
        int index;
+       struct irq_2_iommu *irq_iommu;
+       unsigned long flags;
 
-       spin_lock(&irq_2_ir_lock);
-       if (irq >= nr_irqs || !irq_2_iommu[irq].iommu) {
-               spin_unlock(&irq_2_ir_lock);
+       spin_lock_irqsave(&irq_2_ir_lock, flags);
+       irq_iommu = valid_irq_2_iommu(irq);
+       if (!irq_iommu) {
+               spin_unlock_irqrestore(&irq_2_ir_lock, flags);
                return -1;
        }
 
-       *sub_handle = irq_2_iommu[irq].sub_handle;
-       index = irq_2_iommu[irq].irte_index;
-       spin_unlock(&irq_2_ir_lock);
+       *sub_handle = irq_iommu->sub_handle;
+       index = irq_iommu->irte_index;
+       spin_unlock_irqrestore(&irq_2_ir_lock, flags);
        return index;
 }
 
 int set_irte_irq(int irq, struct intel_iommu *iommu, u16 index, u16 subhandle)
 {
-       spin_lock(&irq_2_ir_lock);
-       if (irq >= nr_irqs || irq_2_iommu[irq].iommu) {
-               spin_unlock(&irq_2_ir_lock);
+       struct irq_2_iommu *irq_iommu;
+       unsigned long flags;
+
+       spin_lock_irqsave(&irq_2_ir_lock, flags);
+
+       irq_iommu = irq_2_iommu_alloc(irq);
+
+       if (!irq_iommu) {
+               spin_unlock_irqrestore(&irq_2_ir_lock, flags);
+               printk(KERN_ERR "can't allocate irq_2_iommu\n");
                return -1;
        }
 
-       irq_2_iommu[irq].iommu = iommu;
-       irq_2_iommu[irq].irte_index = index;
-       irq_2_iommu[irq].sub_handle = subhandle;
-       irq_2_iommu[irq].irte_mask = 0;
+       irq_iommu->iommu = iommu;
+       irq_iommu->irte_index = index;
+       irq_iommu->sub_handle = subhandle;
+       irq_iommu->irte_mask = 0;
 
-       spin_unlock(&irq_2_ir_lock);
+       spin_unlock_irqrestore(&irq_2_ir_lock, flags);
 
        return 0;
 }
 
 int clear_irte_irq(int irq, struct intel_iommu *iommu, u16 index)
 {
-       spin_lock(&irq_2_ir_lock);
-       if (irq >= nr_irqs || !irq_2_iommu[irq].iommu) {
-               spin_unlock(&irq_2_ir_lock);
+       struct irq_2_iommu *irq_iommu;
+       unsigned long flags;
+
+       spin_lock_irqsave(&irq_2_ir_lock, flags);
+       irq_iommu = valid_irq_2_iommu(irq);
+       if (!irq_iommu) {
+               spin_unlock_irqrestore(&irq_2_ir_lock, flags);
                return -1;
        }
 
-       irq_2_iommu[irq].iommu = NULL;
-       irq_2_iommu[irq].irte_index = 0;
-       irq_2_iommu[irq].sub_handle = 0;
-       irq_2_iommu[irq].irte_mask = 0;
+       irq_iommu->iommu = NULL;
+       irq_iommu->irte_index = 0;
+       irq_iommu->sub_handle = 0;
+       irq_2_iommu(irq)->irte_mask = 0;
 
-       spin_unlock(&irq_2_ir_lock);
+       spin_unlock_irqrestore(&irq_2_ir_lock, flags);
 
        return 0;
 }
 
 int modify_irte(int irq, struct irte *irte_modified)
 {
+       int rc;
        int index;
        struct irte *irte;
        struct intel_iommu *iommu;
+       struct irq_2_iommu *irq_iommu;
+       unsigned long flags;
 
-       spin_lock(&irq_2_ir_lock);
-       if (irq >= nr_irqs || !irq_2_iommu[irq].iommu) {
-               spin_unlock(&irq_2_ir_lock);
+       spin_lock_irqsave(&irq_2_ir_lock, flags);
+       irq_iommu = valid_irq_2_iommu(irq);
+       if (!irq_iommu) {
+               spin_unlock_irqrestore(&irq_2_ir_lock, flags);
                return -1;
        }
 
-       iommu = irq_2_iommu[irq].iommu;
+       iommu = irq_iommu->iommu;
 
-       index = irq_2_iommu[irq].irte_index + irq_2_iommu[irq].sub_handle;
+       index = irq_iommu->irte_index + irq_iommu->sub_handle;
        irte = &iommu->ir_table->base[index];
 
-       set_64bit((unsigned long *)irte, irte_modified->low | (1 << 1));
+       set_64bit((unsigned long *)&irte->low, irte_modified->low);
+       set_64bit((unsigned long *)&irte->high, irte_modified->high);
        __iommu_flush_cache(iommu, irte, sizeof(*irte));
 
-       qi_flush_iec(iommu, index, 0);
+       rc = qi_flush_iec(iommu, index, 0);
+       spin_unlock_irqrestore(&irq_2_ir_lock, flags);
 
-       spin_unlock(&irq_2_ir_lock);
-       return 0;
+       return rc;
 }
 
 int flush_irte(int irq)
 {
+       int rc;
        int index;
        struct intel_iommu *iommu;
+       struct irq_2_iommu *irq_iommu;
+       unsigned long flags;
 
-       spin_lock(&irq_2_ir_lock);
-       if (irq >= nr_irqs || !irq_2_iommu[irq].iommu) {
-               spin_unlock(&irq_2_ir_lock);
+       spin_lock_irqsave(&irq_2_ir_lock, flags);
+       irq_iommu = valid_irq_2_iommu(irq);
+       if (!irq_iommu) {
+               spin_unlock_irqrestore(&irq_2_ir_lock, flags);
                return -1;
        }
 
-       iommu = irq_2_iommu[irq].iommu;
+       iommu = irq_iommu->iommu;
 
-       index = irq_2_iommu[irq].irte_index + irq_2_iommu[irq].sub_handle;
+       index = irq_iommu->irte_index + irq_iommu->sub_handle;
 
-       qi_flush_iec(iommu, index, irq_2_iommu[irq].irte_mask);
-       spin_unlock(&irq_2_ir_lock);
+       rc = qi_flush_iec(iommu, index, irq_iommu->irte_mask);
+       spin_unlock_irqrestore(&irq_2_ir_lock, flags);
 
-       return 0;
+       return rc;
 }
 
 struct intel_iommu *map_ioapic_to_ir(int apic)
@@ -249,35 +364,135 @@ struct intel_iommu *map_dev_to_ir(struct pci_dev *dev)
        return drhd->iommu;
 }
 
-int free_irte(int irq)
+static int clear_entries(struct irq_2_iommu *irq_iommu)
 {
-       int index, i;
-       struct irte *irte;
+       struct irte *start, *entry, *end;
        struct intel_iommu *iommu;
+       int index;
+
+       if (irq_iommu->sub_handle)
+               return 0;
 
-       spin_lock(&irq_2_ir_lock);
-       if (irq >= nr_irqs || !irq_2_iommu[irq].iommu) {
-               spin_unlock(&irq_2_ir_lock);
+       iommu = irq_iommu->iommu;
+       index = irq_iommu->irte_index + irq_iommu->sub_handle;
+
+       start = iommu->ir_table->base + index;
+       end = start + (1 << irq_iommu->irte_mask);
+
+       for (entry = start; entry < end; entry++) {
+               set_64bit((unsigned long *)&entry->low, 0);
+               set_64bit((unsigned long *)&entry->high, 0);
+       }
+
+       return qi_flush_iec(iommu, index, irq_iommu->irte_mask);
+}
+
+int free_irte(int irq)
+{
+       int rc = 0;
+       struct irq_2_iommu *irq_iommu;
+       unsigned long flags;
+
+       spin_lock_irqsave(&irq_2_ir_lock, flags);
+       irq_iommu = valid_irq_2_iommu(irq);
+       if (!irq_iommu) {
+               spin_unlock_irqrestore(&irq_2_ir_lock, flags);
                return -1;
        }
 
-       iommu = irq_2_iommu[irq].iommu;
+       rc = clear_entries(irq_iommu);
 
-       index = irq_2_iommu[irq].irte_index + irq_2_iommu[irq].sub_handle;
-       irte = &iommu->ir_table->base[index];
+       irq_iommu->iommu = NULL;
+       irq_iommu->irte_index = 0;
+       irq_iommu->sub_handle = 0;
+       irq_iommu->irte_mask = 0;
+
+       spin_unlock_irqrestore(&irq_2_ir_lock, flags);
+
+       return rc;
+}
+
+/*
+ * source validation type
+ */
+#define SVT_NO_VERIFY          0x0  /* no verification is required */
+#define SVT_VERIFY_SID_SQ      0x1  /* verify using SID and SQ fiels */
+#define SVT_VERIFY_BUS         0x2  /* verify bus of request-id */
+
+/*
+ * source-id qualifier
+ */
+#define SQ_ALL_16      0x0  /* verify all 16 bits of request-id */
+#define SQ_13_IGNORE_1 0x1  /* verify most significant 13 bits, ignore
+                             * the third least significant bit
+                             */
+#define SQ_13_IGNORE_2 0x2  /* verify most significant 13 bits, ignore
+                             * the second and third least significant bits
+                             */
+#define SQ_13_IGNORE_3 0x3  /* verify most significant 13 bits, ignore
+                             * the least three significant bits
+                             */
+
+/*
+ * set SVT, SQ and SID fields of irte to verify
+ * source ids of interrupt requests
+ */
+static void set_irte_sid(struct irte *irte, unsigned int svt,
+                        unsigned int sq, unsigned int sid)
+{
+       irte->svt = svt;
+       irte->sq = sq;
+       irte->sid = sid;
+}
+
+int set_ioapic_sid(struct irte *irte, int apic)
+{
+       int i;
+       u16 sid = 0;
+
+       if (!irte)
+               return -1;
+
+       for (i = 0; i < MAX_IO_APICS; i++) {
+               if (ir_ioapic[i].id == apic) {
+                       sid = (ir_ioapic[i].bus << 8) | ir_ioapic[i].devfn;
+                       break;
+               }
+       }
 
-       if (!irq_2_iommu[irq].sub_handle) {
-               for (i = 0; i < (1 << irq_2_iommu[irq].irte_mask); i++)
-                       set_64bit((unsigned long *)irte, 0);
-               qi_flush_iec(iommu, index, irq_2_iommu[irq].irte_mask);
+       if (sid == 0) {
+               pr_warning("Failed to set source-id of IOAPIC (%d)\n", apic);
+               return -1;
        }
 
-       irq_2_iommu[irq].iommu = NULL;
-       irq_2_iommu[irq].irte_index = 0;
-       irq_2_iommu[irq].sub_handle = 0;
-       irq_2_iommu[irq].irte_mask = 0;
+       set_irte_sid(irte, 1, 0, sid);
+
+       return 0;
+}
 
-       spin_unlock(&irq_2_ir_lock);
+int set_msi_sid(struct irte *irte, struct pci_dev *dev)
+{
+       struct pci_dev *bridge;
+
+       if (!irte || !dev)
+               return -1;
+
+       /* PCIe device or Root Complex integrated PCI device */
+       if (dev->is_pcie || !dev->bus->parent) {
+               set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16,
+                            (dev->bus->number << 8) | dev->devfn);
+               return 0;
+       }
+
+       bridge = pci_find_upstream_pcie_bridge(dev);
+       if (bridge) {
+               if (bridge->is_pcie) /* this is a PCIE-to-PCI/PCIX bridge */
+                       set_irte_sid(irte, SVT_VERIFY_BUS, SQ_ALL_16,
+                               (bridge->bus->number << 8) | dev->bus->number);
+               else /* this is a legacy PCI bridge */
+                       set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16,
+                               (bridge->bus->number << 8) | bridge->devfn);
+       }
 
        return 0;
 }
@@ -285,7 +500,7 @@ int free_irte(int irq)
 static void iommu_set_intr_remapping(struct intel_iommu *iommu, int mode)
 {
        u64 addr;
-       u32 cmd, sts;
+       u32 sts;
        unsigned long flags;
 
        addr = virt_to_phys((void *)iommu->ir_table->base);
@@ -296,8 +511,8 @@ static void iommu_set_intr_remapping(struct intel_iommu *iommu, int mode)
                    (addr) | IR_X2APIC_MODE(mode) | INTR_REMAP_TABLE_REG_SIZE);
 
        /* Set interrupt-remapping table pointer */
-       cmd = iommu->gcmd | DMA_GCMD_SIRTP;
-       writel(cmd, iommu->reg + DMAR_GCMD_REG);
+       iommu->gcmd |= DMA_GCMD_SIRTP;
+       writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
 
        IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
                      readl, (sts & DMA_GSTS_IRTPS), sts);
@@ -312,9 +527,8 @@ static void iommu_set_intr_remapping(struct intel_iommu *iommu, int mode)
        spin_lock_irqsave(&iommu->register_lock, flags);
 
        /* Enable interrupt-remapping */
-       cmd = iommu->gcmd | DMA_GCMD_IRE;
        iommu->gcmd |= DMA_GCMD_IRE;
-       writel(cmd, iommu->reg + DMAR_GCMD_REG);
+       writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
 
        IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
                      readl, (sts & DMA_GSTS_IRES), sts);
@@ -329,12 +543,12 @@ static int setup_intr_remapping(struct intel_iommu *iommu, int mode)
        struct page *pages;
 
        ir_table = iommu->ir_table = kzalloc(sizeof(struct ir_table),
-                                            GFP_KERNEL);
+                                            GFP_ATOMIC);
 
        if (!iommu->ir_table)
                return -ENOMEM;
 
-       pages = alloc_pages(GFP_KERNEL | __GFP_ZERO, INTR_REMAP_PAGE_ORDER);
+       pages = alloc_pages(GFP_ATOMIC | __GFP_ZERO, INTR_REMAP_PAGE_ORDER);
 
        if (!pages) {
                printk(KERN_ERR "failed to allocate pages of order %d\n",
@@ -349,11 +563,93 @@ static int setup_intr_remapping(struct intel_iommu *iommu, int mode)
        return 0;
 }
 
+/*
+ * Disable Interrupt Remapping.
+ */
+static void iommu_disable_intr_remapping(struct intel_iommu *iommu)
+{
+       unsigned long flags;
+       u32 sts;
+
+       if (!ecap_ir_support(iommu->ecap))
+               return;
+
+       /*
+        * global invalidation of interrupt entry cache before disabling
+        * interrupt-remapping.
+        */
+       qi_global_iec(iommu);
+
+       spin_lock_irqsave(&iommu->register_lock, flags);
+
+       sts = dmar_readq(iommu->reg + DMAR_GSTS_REG);
+       if (!(sts & DMA_GSTS_IRES))
+               goto end;
+
+       iommu->gcmd &= ~DMA_GCMD_IRE;
+       writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
+
+       IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
+                     readl, !(sts & DMA_GSTS_IRES), sts);
+
+end:
+       spin_unlock_irqrestore(&iommu->register_lock, flags);
+}
+
+int __init intr_remapping_supported(void)
+{
+       struct dmar_drhd_unit *drhd;
+
+       if (disable_intremap)
+               return 0;
+
+       if (!dmar_ir_support())
+               return 0;
+
+       for_each_drhd_unit(drhd) {
+               struct intel_iommu *iommu = drhd->iommu;
+
+               if (!ecap_ir_support(iommu->ecap))
+                       return 0;
+       }
+
+       return 1;
+}
+
 int __init enable_intr_remapping(int eim)
 {
        struct dmar_drhd_unit *drhd;
        int setup = 0;
 
+       if (parse_ioapics_under_ir() != 1) {
+               printk(KERN_INFO "Not enable interrupt remapping\n");
+               return -1;
+       }
+
+       for_each_drhd_unit(drhd) {
+               struct intel_iommu *iommu = drhd->iommu;
+
+               /*
+                * If the queued invalidation is already initialized,
+                * shouldn't disable it.
+                */
+               if (iommu->qi)
+                       continue;
+
+               /*
+                * Clear previous faults.
+                */
+               dmar_fault(-1, iommu);
+
+               /*
+                * Disable intr remapping and queued invalidation, if already
+                * enabled prior to OS handover.
+                */
+               iommu_disable_intr_remapping(iommu);
+
+               dmar_disable_qi(iommu);
+       }
+
        /*
         * check for the Interrupt-remapping support
         */
@@ -415,6 +711,35 @@ error:
        return -1;
 }
 
+static void ir_parse_one_ioapic_scope(struct acpi_dmar_device_scope *scope,
+                                     struct intel_iommu *iommu)
+{
+       struct acpi_dmar_pci_path *path;
+       u8 bus;
+       int count;
+
+       bus = scope->bus;
+       path = (struct acpi_dmar_pci_path *)(scope + 1);
+       count = (scope->length - sizeof(struct acpi_dmar_device_scope))
+               / sizeof(struct acpi_dmar_pci_path);
+
+       while (--count > 0) {
+               /*
+                * Access PCI directly due to the PCI
+                * subsystem isn't initialized yet.
+                */
+               bus = read_pci_config_byte(bus, path->dev, path->fn,
+                                          PCI_SECONDARY_BUS);
+               path++;
+       }
+
+       ir_ioapic[ir_ioapic_num].bus   = bus;
+       ir_ioapic[ir_ioapic_num].devfn = PCI_DEVFN(path->dev, path->fn);
+       ir_ioapic[ir_ioapic_num].iommu = iommu;
+       ir_ioapic[ir_ioapic_num].id    = scope->enumeration_id;
+       ir_ioapic_num++;
+}
+
 static int ir_parse_ioapic_scope(struct acpi_dmar_header *header,
                                 struct intel_iommu *iommu)
 {
@@ -439,9 +764,7 @@ static int ir_parse_ioapic_scope(struct acpi_dmar_header *header,
                               " 0x%Lx\n", scope->enumeration_id,
                               drhd->address);
 
-                       ir_ioapic[ir_ioapic_num].iommu = iommu;
-                       ir_ioapic[ir_ioapic_num].id = scope->enumeration_id;
-                       ir_ioapic_num++;
+                       ir_parse_one_ioapic_scope(scope, iommu);
                }
                start += scope->length;
        }
@@ -477,3 +800,54 @@ int __init parse_ioapics_under_ir(void)
 
        return ir_supported;
 }
+
+void disable_intr_remapping(void)
+{
+       struct dmar_drhd_unit *drhd;
+       struct intel_iommu *iommu = NULL;
+
+       /*
+        * Disable Interrupt-remapping for all the DRHD's now.
+        */
+       for_each_iommu(iommu, drhd) {
+               if (!ecap_ir_support(iommu->ecap))
+                       continue;
+
+               iommu_disable_intr_remapping(iommu);
+       }
+}
+
+int reenable_intr_remapping(int eim)
+{
+       struct dmar_drhd_unit *drhd;
+       int setup = 0;
+       struct intel_iommu *iommu = NULL;
+
+       for_each_iommu(iommu, drhd)
+               if (iommu->qi)
+                       dmar_reenable_qi(iommu);
+
+       /*
+        * Setup Interrupt-remapping for all the DRHD's now.
+        */
+       for_each_iommu(iommu, drhd) {
+               if (!ecap_ir_support(iommu->ecap))
+                       continue;
+
+               /* Set up interrupt remapping for iommu.*/
+               iommu_set_intr_remapping(iommu, eim);
+               setup = 1;
+       }
+
+       if (!setup)
+               goto error;
+
+       return 0;
+
+error:
+       /*
+        * handle error condition gracefully here!
+        */
+       return -1;
+}
+