iommu/arm: Add support for ARM Ltd. System MMU architecture
authorWill Deacon <will.deacon@arm.com>
Mon, 24 Jun 2013 17:31:25 +0000 (18:31 +0100)
committerJoerg Roedel <joro@8bytes.org>
Tue, 25 Jun 2013 21:34:20 +0000 (23:34 +0200)
This patch adds support for SMMUs implementing the ARM System MMU
architecture versions 1 or 2. Both arm and arm64 are supported, although
the v7s descriptor format is not used.

Cc: Rob Herring <robherring2@gmail.com>
Cc: Andreas Herrmann <andreas.herrmann@calxeda.com>
Cc: Olav Haugan <ohaugan@codeaurora.org>
Cc: Joerg Roedel <joro@8bytes.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Acked-by: Andreas Herrmann <andreas.herrmann@calxeda.com>
Signed-off-by: Joerg Roedel <joro@8bytes.org>
drivers/iommu/Kconfig
drivers/iommu/Makefile
drivers/iommu/arm-smmu.c [new file with mode: 0644]

index c332fb9..957cfd4 100644 (file)
@@ -261,4 +261,17 @@ config SHMOBILE_IOMMU_L1SIZE
        default 256 if SHMOBILE_IOMMU_ADDRSIZE_64MB
        default 128 if SHMOBILE_IOMMU_ADDRSIZE_32MB
 
+config ARM_SMMU
+       bool "ARM Ltd. System MMU (SMMU) Support"
+       depends on ARM64 || (ARM_LPAE && OF)
+       select IOMMU_API
+       select ARM_DMA_USE_IOMMU if ARM
+       help
+         Support for implementations of the ARM System MMU architecture
+         versions 1 and 2. The driver supports both v7l and v8l table
+         formats with 4k and 64k page sizes.
+
+         Say Y here if your SoC includes an IOMMU device implementing
+         the ARM SMMU architecture.
+
 endif # IOMMU_SUPPORT
index ef0e520..bbe7041 100644 (file)
@@ -3,6 +3,7 @@ obj-$(CONFIG_OF_IOMMU)  += of_iommu.o
 obj-$(CONFIG_MSM_IOMMU) += msm_iommu.o msm_iommu_dev.o
 obj-$(CONFIG_AMD_IOMMU) += amd_iommu.o amd_iommu_init.o
 obj-$(CONFIG_AMD_IOMMU_V2) += amd_iommu_v2.o
+obj-$(CONFIG_ARM_SMMU) += arm-smmu.o
 obj-$(CONFIG_DMAR_TABLE) += dmar.o
 obj-$(CONFIG_INTEL_IOMMU) += iova.o intel-iommu.o
 obj-$(CONFIG_IRQ_REMAP) += intel_irq_remapping.o irq_remapping.o
diff --git a/drivers/iommu/arm-smmu.c b/drivers/iommu/arm-smmu.c
new file mode 100644 (file)
index 0000000..ebd0a4c
--- /dev/null
@@ -0,0 +1,1969 @@
+/*
+ * IOMMU API for ARM architected SMMU implementations.
+ *
+ * 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, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * Copyright (C) 2013 ARM Limited
+ *
+ * Author: Will Deacon <will.deacon@arm.com>
+ *
+ * This driver currently supports:
+ *     - SMMUv1 and v2 implementations
+ *     - Stream-matching and stream-indexing
+ *     - v7/v8 long-descriptor format
+ *     - Non-secure access to the SMMU
+ *     - 4k and 64k pages, with contiguous pte hints.
+ *     - Up to 39-bit addressing
+ *     - Context fault reporting
+ */
+
+#define pr_fmt(fmt) "arm-smmu: " fmt
+
+#include <linux/delay.h>
+#include <linux/dma-mapping.h>
+#include <linux/err.h>
+#include <linux/interrupt.h>
+#include <linux/io.h>
+#include <linux/iommu.h>
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/platform_device.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+
+#include <linux/amba/bus.h>
+
+#include <asm/pgalloc.h>
+
+/* Maximum number of stream IDs assigned to a single device */
+#define MAX_MASTER_STREAMIDS           8
+
+/* Maximum number of context banks per SMMU */
+#define ARM_SMMU_MAX_CBS               128
+
+/* Maximum number of mapping groups per SMMU */
+#define ARM_SMMU_MAX_SMRS              128
+
+/* Number of VMIDs per SMMU */
+#define ARM_SMMU_NUM_VMIDS             256
+
+/* SMMU global address space */
+#define ARM_SMMU_GR0(smmu)             ((smmu)->base)
+#define ARM_SMMU_GR1(smmu)             ((smmu)->base + (smmu)->pagesize)
+
+/* Page table bits */
+#define ARM_SMMU_PTE_PAGE              (((pteval_t)3) << 0)
+#define ARM_SMMU_PTE_CONT              (((pteval_t)1) << 52)
+#define ARM_SMMU_PTE_AF                        (((pteval_t)1) << 10)
+#define ARM_SMMU_PTE_SH_NS             (((pteval_t)0) << 8)
+#define ARM_SMMU_PTE_SH_OS             (((pteval_t)2) << 8)
+#define ARM_SMMU_PTE_SH_IS             (((pteval_t)3) << 8)
+
+#if PAGE_SIZE == SZ_4K
+#define ARM_SMMU_PTE_CONT_ENTRIES      16
+#elif PAGE_SIZE == SZ_64K
+#define ARM_SMMU_PTE_CONT_ENTRIES      32
+#else
+#define ARM_SMMU_PTE_CONT_ENTRIES      1
+#endif
+
+#define ARM_SMMU_PTE_CONT_SIZE         (PAGE_SIZE * ARM_SMMU_PTE_CONT_ENTRIES)
+#define ARM_SMMU_PTE_CONT_MASK         (~(ARM_SMMU_PTE_CONT_SIZE - 1))
+#define ARM_SMMU_PTE_HWTABLE_SIZE      (PTRS_PER_PTE * sizeof(pte_t))
+
+/* Stage-1 PTE */
+#define ARM_SMMU_PTE_AP_UNPRIV         (((pteval_t)1) << 6)
+#define ARM_SMMU_PTE_AP_RDONLY         (((pteval_t)2) << 6)
+#define ARM_SMMU_PTE_ATTRINDX_SHIFT    2
+
+/* Stage-2 PTE */
+#define ARM_SMMU_PTE_HAP_FAULT         (((pteval_t)0) << 6)
+#define ARM_SMMU_PTE_HAP_READ          (((pteval_t)1) << 6)
+#define ARM_SMMU_PTE_HAP_WRITE         (((pteval_t)2) << 6)
+#define ARM_SMMU_PTE_MEMATTR_OIWB      (((pteval_t)0xf) << 2)
+#define ARM_SMMU_PTE_MEMATTR_NC                (((pteval_t)0x5) << 2)
+#define ARM_SMMU_PTE_MEMATTR_DEV       (((pteval_t)0x1) << 2)
+
+/* Configuration registers */
+#define ARM_SMMU_GR0_sCR0              0x0
+#define sCR0_CLIENTPD                  (1 << 0)
+#define sCR0_GFRE                      (1 << 1)
+#define sCR0_GFIE                      (1 << 2)
+#define sCR0_GCFGFRE                   (1 << 4)
+#define sCR0_GCFGFIE                   (1 << 5)
+#define sCR0_USFCFG                    (1 << 10)
+#define sCR0_VMIDPNE                   (1 << 11)
+#define sCR0_PTM                       (1 << 12)
+#define sCR0_FB                                (1 << 13)
+#define sCR0_BSU_SHIFT                 14
+#define sCR0_BSU_MASK                  0x3
+
+/* Identification registers */
+#define ARM_SMMU_GR0_ID0               0x20
+#define ARM_SMMU_GR0_ID1               0x24
+#define ARM_SMMU_GR0_ID2               0x28
+#define ARM_SMMU_GR0_ID3               0x2c
+#define ARM_SMMU_GR0_ID4               0x30
+#define ARM_SMMU_GR0_ID5               0x34
+#define ARM_SMMU_GR0_ID6               0x38
+#define ARM_SMMU_GR0_ID7               0x3c
+#define ARM_SMMU_GR0_sGFSR             0x48
+#define ARM_SMMU_GR0_sGFSYNR0          0x50
+#define ARM_SMMU_GR0_sGFSYNR1          0x54
+#define ARM_SMMU_GR0_sGFSYNR2          0x58
+#define ARM_SMMU_GR0_PIDR0             0xfe0
+#define ARM_SMMU_GR0_PIDR1             0xfe4
+#define ARM_SMMU_GR0_PIDR2             0xfe8
+
+#define ID0_S1TS                       (1 << 30)
+#define ID0_S2TS                       (1 << 29)
+#define ID0_NTS                                (1 << 28)
+#define ID0_SMS                                (1 << 27)
+#define ID0_PTFS_SHIFT                 24
+#define ID0_PTFS_MASK                  0x2
+#define ID0_PTFS_V8_ONLY               0x2
+#define ID0_CTTW                       (1 << 14)
+#define ID0_NUMIRPT_SHIFT              16
+#define ID0_NUMIRPT_MASK               0xff
+#define ID0_NUMSMRG_SHIFT              0
+#define ID0_NUMSMRG_MASK               0xff
+
+#define ID1_PAGESIZE                   (1 << 31)
+#define ID1_NUMPAGENDXB_SHIFT          28
+#define ID1_NUMPAGENDXB_MASK           7
+#define ID1_NUMS2CB_SHIFT              16
+#define ID1_NUMS2CB_MASK               0xff
+#define ID1_NUMCB_SHIFT                        0
+#define ID1_NUMCB_MASK                 0xff
+
+#define ID2_OAS_SHIFT                  4
+#define ID2_OAS_MASK                   0xf
+#define ID2_IAS_SHIFT                  0
+#define ID2_IAS_MASK                   0xf
+#define ID2_UBS_SHIFT                  8
+#define ID2_UBS_MASK                   0xf
+#define ID2_PTFS_4K                    (1 << 12)
+#define ID2_PTFS_16K                   (1 << 13)
+#define ID2_PTFS_64K                   (1 << 14)
+
+#define PIDR2_ARCH_SHIFT               4
+#define PIDR2_ARCH_MASK                        0xf
+
+/* Global TLB invalidation */
+#define ARM_SMMU_GR0_STLBIALL          0x60
+#define ARM_SMMU_GR0_TLBIVMID          0x64
+#define ARM_SMMU_GR0_TLBIALLNSNH       0x68
+#define ARM_SMMU_GR0_TLBIALLH          0x6c
+#define ARM_SMMU_GR0_sTLBGSYNC         0x70
+#define ARM_SMMU_GR0_sTLBGSTATUS       0x74
+#define sTLBGSTATUS_GSACTIVE           (1 << 0)
+#define TLB_LOOP_TIMEOUT               1000000 /* 1s! */
+
+/* Stream mapping registers */
+#define ARM_SMMU_GR0_SMR(n)            (0x800 + ((n) << 2))
+#define SMR_VALID                      (1 << 31)
+#define SMR_MASK_SHIFT                 16
+#define SMR_MASK_MASK                  0x7fff
+#define SMR_ID_SHIFT                   0
+#define SMR_ID_MASK                    0x7fff
+
+#define ARM_SMMU_GR0_S2CR(n)           (0xc00 + ((n) << 2))
+#define S2CR_CBNDX_SHIFT               0
+#define S2CR_CBNDX_MASK                        0xff
+#define S2CR_TYPE_SHIFT                        16
+#define S2CR_TYPE_MASK                 0x3
+#define S2CR_TYPE_TRANS                        (0 << S2CR_TYPE_SHIFT)
+#define S2CR_TYPE_BYPASS               (1 << S2CR_TYPE_SHIFT)
+#define S2CR_TYPE_FAULT                        (2 << S2CR_TYPE_SHIFT)
+
+/* Context bank attribute registers */
+#define ARM_SMMU_GR1_CBAR(n)           (0x0 + ((n) << 2))
+#define CBAR_VMID_SHIFT                        0
+#define CBAR_VMID_MASK                 0xff
+#define CBAR_S1_MEMATTR_SHIFT          12
+#define CBAR_S1_MEMATTR_MASK           0xf
+#define CBAR_S1_MEMATTR_WB             0xf
+#define CBAR_TYPE_SHIFT                        16
+#define CBAR_TYPE_MASK                 0x3
+#define CBAR_TYPE_S2_TRANS             (0 << CBAR_TYPE_SHIFT)
+#define CBAR_TYPE_S1_TRANS_S2_BYPASS   (1 << CBAR_TYPE_SHIFT)
+#define CBAR_TYPE_S1_TRANS_S2_FAULT    (2 << CBAR_TYPE_SHIFT)
+#define CBAR_TYPE_S1_TRANS_S2_TRANS    (3 << CBAR_TYPE_SHIFT)
+#define CBAR_IRPTNDX_SHIFT             24
+#define CBAR_IRPTNDX_MASK              0xff
+
+#define ARM_SMMU_GR1_CBA2R(n)          (0x800 + ((n) << 2))
+#define CBA2R_RW64_32BIT               (0 << 0)
+#define CBA2R_RW64_64BIT               (1 << 0)
+
+/* Translation context bank */
+#define ARM_SMMU_CB_BASE(smmu)         ((smmu)->base + ((smmu)->size >> 1))
+#define ARM_SMMU_CB(smmu, n)           ((n) * (smmu)->pagesize)
+
+#define ARM_SMMU_CB_SCTLR              0x0
+#define ARM_SMMU_CB_RESUME             0x8
+#define ARM_SMMU_CB_TTBCR2             0x10
+#define ARM_SMMU_CB_TTBR0_LO           0x20
+#define ARM_SMMU_CB_TTBR0_HI           0x24
+#define ARM_SMMU_CB_TTBCR              0x30
+#define ARM_SMMU_CB_S1_MAIR0           0x38
+#define ARM_SMMU_CB_FSR                        0x58
+#define ARM_SMMU_CB_FAR_LO             0x60
+#define ARM_SMMU_CB_FAR_HI             0x64
+#define ARM_SMMU_CB_FSYNR0             0x68
+
+#define SCTLR_S1_ASIDPNE               (1 << 12)
+#define SCTLR_CFCFG                    (1 << 7)
+#define SCTLR_CFIE                     (1 << 6)
+#define SCTLR_CFRE                     (1 << 5)
+#define SCTLR_E                                (1 << 4)
+#define SCTLR_AFE                      (1 << 2)
+#define SCTLR_TRE                      (1 << 1)
+#define SCTLR_M                                (1 << 0)
+#define SCTLR_EAE_SBOP                 (SCTLR_AFE | SCTLR_TRE)
+
+#define RESUME_RETRY                   (0 << 0)
+#define RESUME_TERMINATE               (1 << 0)
+
+#define TTBCR_EAE                      (1 << 31)
+
+#define TTBCR_PASIZE_SHIFT             16
+#define TTBCR_PASIZE_MASK              0x7
+
+#define TTBCR_TG0_4K                   (0 << 14)
+#define TTBCR_TG0_64K                  (1 << 14)
+
+#define TTBCR_SH0_SHIFT                        12
+#define TTBCR_SH0_MASK                 0x3
+#define TTBCR_SH_NS                    0
+#define TTBCR_SH_OS                    2
+#define TTBCR_SH_IS                    3
+
+#define TTBCR_ORGN0_SHIFT              10
+#define TTBCR_IRGN0_SHIFT              8
+#define TTBCR_RGN_MASK                 0x3
+#define TTBCR_RGN_NC                   0
+#define TTBCR_RGN_WBWA                 1
+#define TTBCR_RGN_WT                   2
+#define TTBCR_RGN_WB                   3
+
+#define TTBCR_SL0_SHIFT                        6
+#define TTBCR_SL0_MASK                 0x3
+#define TTBCR_SL0_LVL_2                        0
+#define TTBCR_SL0_LVL_1                        1
+
+#define TTBCR_T1SZ_SHIFT               16
+#define TTBCR_T0SZ_SHIFT               0
+#define TTBCR_SZ_MASK                  0xf
+
+#define TTBCR2_SEP_SHIFT               15
+#define TTBCR2_SEP_MASK                        0x7
+
+#define TTBCR2_PASIZE_SHIFT            0
+#define TTBCR2_PASIZE_MASK             0x7
+
+/* Common definitions for PASize and SEP fields */
+#define TTBCR2_ADDR_32                 0
+#define TTBCR2_ADDR_36                 1
+#define TTBCR2_ADDR_40                 2
+#define TTBCR2_ADDR_42                 3
+#define TTBCR2_ADDR_44                 4
+#define TTBCR2_ADDR_48                 5
+
+#define MAIR_ATTR_SHIFT(n)             ((n) << 3)
+#define MAIR_ATTR_MASK                 0xff
+#define MAIR_ATTR_DEVICE               0x04
+#define MAIR_ATTR_NC                   0x44
+#define MAIR_ATTR_WBRWA                        0xff
+#define MAIR_ATTR_IDX_NC               0
+#define MAIR_ATTR_IDX_CACHE            1
+#define MAIR_ATTR_IDX_DEV              2
+
+#define FSR_MULTI                      (1 << 31)
+#define FSR_SS                         (1 << 30)
+#define FSR_UUT                                (1 << 8)
+#define FSR_ASF                                (1 << 7)
+#define FSR_TLBLKF                     (1 << 6)
+#define FSR_TLBMCF                     (1 << 5)
+#define FSR_EF                         (1 << 4)
+#define FSR_PF                         (1 << 3)
+#define FSR_AFF                                (1 << 2)
+#define FSR_TF                         (1 << 1)
+
+#define FSR_IGN                                (FSR_AFF | FSR_ASF | FSR_TLBMCF |       \
+                                        FSR_TLBLKF)
+#define FSR_FAULT                      (FSR_MULTI | FSR_SS | FSR_UUT |         \
+                                        FSR_EF | FSR_PF | FSR_TF)
+
+#define FSYNR0_WNR                     (1 << 4)
+
+struct arm_smmu_smr {
+       u8                              idx;
+       u16                             mask;
+       u16                             id;
+};
+
+struct arm_smmu_master {
+       struct device_node              *of_node;
+
+       /*
+        * The following is specific to the master's position in the
+        * SMMU chain.
+        */
+       struct rb_node                  node;
+       int                             num_streamids;
+       u16                             streamids[MAX_MASTER_STREAMIDS];
+
+       /*
+        * We only need to allocate these on the root SMMU, as we
+        * configure unmatched streams to bypass translation.
+        */
+       struct arm_smmu_smr             *smrs;
+};
+
+struct arm_smmu_device {
+       struct device                   *dev;
+       struct device_node              *parent_of_node;
+
+       void __iomem                    *base;
+       unsigned long                   size;
+       unsigned long                   pagesize;
+
+#define ARM_SMMU_FEAT_COHERENT_WALK    (1 << 0)
+#define ARM_SMMU_FEAT_STREAM_MATCH     (1 << 1)
+#define ARM_SMMU_FEAT_TRANS_S1         (1 << 2)
+#define ARM_SMMU_FEAT_TRANS_S2         (1 << 3)
+#define ARM_SMMU_FEAT_TRANS_NESTED     (1 << 4)
+       u32                             features;
+       int                             version;
+
+       u32                             num_context_banks;
+       u32                             num_s2_context_banks;
+       DECLARE_BITMAP(context_map, ARM_SMMU_MAX_CBS);
+       atomic_t                        irptndx;
+
+       u32                             num_mapping_groups;
+       DECLARE_BITMAP(smr_map, ARM_SMMU_MAX_SMRS);
+
+       unsigned long                   input_size;
+       unsigned long                   s1_output_size;
+       unsigned long                   s2_output_size;
+
+       u32                             num_global_irqs;
+       u32                             num_context_irqs;
+       unsigned int                    *irqs;
+
+       DECLARE_BITMAP(vmid_map, ARM_SMMU_NUM_VMIDS);
+
+       struct list_head                list;
+       struct rb_root                  masters;
+};
+
+struct arm_smmu_cfg {
+       struct arm_smmu_device          *smmu;
+       u8                              vmid;
+       u8                              cbndx;
+       u8                              irptndx;
+       u32                             cbar;
+       pgd_t                           *pgd;
+};
+
+struct arm_smmu_domain {
+       /*
+        * A domain can span across multiple, chained SMMUs and requires
+        * all devices within the domain to follow the same translation
+        * path.
+        */
+       struct arm_smmu_device          *leaf_smmu;
+       struct arm_smmu_cfg             root_cfg;
+       phys_addr_t                     output_mask;
+
+       spinlock_t                      lock;
+};
+
+static DEFINE_SPINLOCK(arm_smmu_devices_lock);
+static LIST_HEAD(arm_smmu_devices);
+
+static struct arm_smmu_master *find_smmu_master(struct arm_smmu_device *smmu,
+                                               struct device_node *dev_node)
+{
+       struct rb_node *node = smmu->masters.rb_node;
+
+       while (node) {
+               struct arm_smmu_master *master;
+               master = container_of(node, struct arm_smmu_master, node);
+
+               if (dev_node < master->of_node)
+                       node = node->rb_left;
+               else if (dev_node > master->of_node)
+                       node = node->rb_right;
+               else
+                       return master;
+       }
+
+       return NULL;
+}
+
+static int insert_smmu_master(struct arm_smmu_device *smmu,
+                             struct arm_smmu_master *master)
+{
+       struct rb_node **new, *parent;
+
+       new = &smmu->masters.rb_node;
+       parent = NULL;
+       while (*new) {
+               struct arm_smmu_master *this;
+               this = container_of(*new, struct arm_smmu_master, node);
+
+               parent = *new;
+               if (master->of_node < this->of_node)
+                       new = &((*new)->rb_left);
+               else if (master->of_node > this->of_node)
+                       new = &((*new)->rb_right);
+               else
+                       return -EEXIST;
+       }
+
+       rb_link_node(&master->node, parent, new);
+       rb_insert_color(&master->node, &smmu->masters);
+       return 0;
+}
+
+static int register_smmu_master(struct arm_smmu_device *smmu,
+                               struct device *dev,
+                               struct of_phandle_args *masterspec)
+{
+       int i;
+       struct arm_smmu_master *master;
+
+       master = find_smmu_master(smmu, masterspec->np);
+       if (master) {
+               dev_err(dev,
+                       "rejecting multiple registrations for master device %s\n",
+                       masterspec->np->name);
+               return -EBUSY;
+       }
+
+       if (masterspec->args_count > MAX_MASTER_STREAMIDS) {
+               dev_err(dev,
+                       "reached maximum number (%d) of stream IDs for master device %s\n",
+                       MAX_MASTER_STREAMIDS, masterspec->np->name);
+               return -ENOSPC;
+       }
+
+       master = devm_kzalloc(dev, sizeof(*master), GFP_KERNEL);
+       if (!master)
+               return -ENOMEM;
+
+       master->of_node         = masterspec->np;
+       master->num_streamids   = masterspec->args_count;
+
+       for (i = 0; i < master->num_streamids; ++i)
+               master->streamids[i] = masterspec->args[i];
+
+       return insert_smmu_master(smmu, master);
+}
+
+static struct arm_smmu_device *find_parent_smmu(struct arm_smmu_device *smmu)
+{
+       struct arm_smmu_device *parent;
+
+       if (!smmu->parent_of_node)
+               return NULL;
+
+       spin_lock(&arm_smmu_devices_lock);
+       list_for_each_entry(parent, &arm_smmu_devices, list)
+               if (parent->dev->of_node == smmu->parent_of_node)
+                       goto out_unlock;
+
+       parent = NULL;
+       dev_warn(smmu->dev,
+                "Failed to find SMMU parent despite parent in DT\n");
+out_unlock:
+       spin_unlock(&arm_smmu_devices_lock);
+       return parent;
+}
+
+static int __arm_smmu_alloc_bitmap(unsigned long *map, int start, int end)
+{
+       int idx;
+
+       do {
+               idx = find_next_zero_bit(map, end, start);
+               if (idx == end)
+                       return -ENOSPC;
+       } while (test_and_set_bit(idx, map));
+
+       return idx;
+}
+
+static void __arm_smmu_free_bitmap(unsigned long *map, int idx)
+{
+       clear_bit(idx, map);
+}
+
+/* Wait for any pending TLB invalidations to complete */
+static void arm_smmu_tlb_sync(struct arm_smmu_device *smmu)
+{
+       int count = 0;
+       void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
+
+       writel_relaxed(0, gr0_base + ARM_SMMU_GR0_sTLBGSYNC);
+       while (readl_relaxed(gr0_base + ARM_SMMU_GR0_sTLBGSTATUS)
+              & sTLBGSTATUS_GSACTIVE) {
+               cpu_relax();
+               if (++count == TLB_LOOP_TIMEOUT) {
+                       dev_err_ratelimited(smmu->dev,
+                       "TLB sync timed out -- SMMU may be deadlocked\n");
+                       return;
+               }
+               udelay(1);
+       }
+}
+
+static irqreturn_t arm_smmu_context_fault(int irq, void *dev)
+{
+       int flags, ret;
+       u32 fsr, far, fsynr, resume;
+       unsigned long iova;
+       struct iommu_domain *domain = dev;
+       struct arm_smmu_domain *smmu_domain = domain->priv;
+       struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
+       struct arm_smmu_device *smmu = root_cfg->smmu;
+       void __iomem *cb_base;
+
+       cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, root_cfg->cbndx);
+       fsr = readl_relaxed(cb_base + ARM_SMMU_CB_FSR);
+
+       if (!(fsr & FSR_FAULT))
+               return IRQ_NONE;
+
+       if (fsr & FSR_IGN)
+               dev_err_ratelimited(smmu->dev,
+                                   "Unexpected context fault (fsr 0x%u)\n",
+                                   fsr);
+
+       fsynr = readl_relaxed(cb_base + ARM_SMMU_CB_FSYNR0);
+       flags = fsynr & FSYNR0_WNR ? IOMMU_FAULT_WRITE : IOMMU_FAULT_READ;
+
+       far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_LO);
+       iova = far;
+#ifdef CONFIG_64BIT
+       far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_HI);
+       iova |= ((unsigned long)far << 32);
+#endif
+
+       if (!report_iommu_fault(domain, smmu->dev, iova, flags)) {
+               ret = IRQ_HANDLED;
+               resume = RESUME_RETRY;
+       } else {
+               ret = IRQ_NONE;
+               resume = RESUME_TERMINATE;
+       }
+
+       /* Clear the faulting FSR */
+       writel(fsr, cb_base + ARM_SMMU_CB_FSR);
+
+       /* Retry or terminate any stalled transactions */
+       if (fsr & FSR_SS)
+               writel_relaxed(resume, cb_base + ARM_SMMU_CB_RESUME);
+
+       return ret;
+}
+
+static irqreturn_t arm_smmu_global_fault(int irq, void *dev)
+{
+       u32 gfsr, gfsynr0, gfsynr1, gfsynr2;
+       struct arm_smmu_device *smmu = dev;
+       void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
+
+       gfsr = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSR);
+       gfsynr0 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR0);
+       gfsynr1 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR1);
+       gfsynr2 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR2);
+
+       dev_err_ratelimited(smmu->dev,
+               "Unexpected global fault, this could be serious\n");
+       dev_err_ratelimited(smmu->dev,
+               "\tGFSR 0x%08x, GFSYNR0 0x%08x, GFSYNR1 0x%08x, GFSYNR2 0x%08x\n",
+               gfsr, gfsynr0, gfsynr1, gfsynr2);
+
+       writel(gfsr, gr0_base + ARM_SMMU_GR0_sGFSR);
+       return IRQ_NONE;
+}
+
+static void arm_smmu_init_context_bank(struct arm_smmu_domain *smmu_domain)
+{
+       u32 reg;
+       bool stage1;
+       struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
+       struct arm_smmu_device *smmu = root_cfg->smmu;
+       void __iomem *cb_base, *gr0_base, *gr1_base;
+
+       gr0_base = ARM_SMMU_GR0(smmu);
+       gr1_base = ARM_SMMU_GR1(smmu);
+       stage1 = root_cfg->cbar != CBAR_TYPE_S2_TRANS;
+       cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, root_cfg->cbndx);
+
+       /* CBAR */
+       reg = root_cfg->cbar |
+             (root_cfg->vmid << CBAR_VMID_SHIFT);
+       if (smmu->version == 1)
+             reg |= root_cfg->irptndx << CBAR_IRPTNDX_SHIFT;
+
+       /* Use the weakest memory type, so it is overridden by the pte */
+       if (stage1)
+               reg |= (CBAR_S1_MEMATTR_WB << CBAR_S1_MEMATTR_SHIFT);
+       writel_relaxed(reg, gr1_base + ARM_SMMU_GR1_CBAR(root_cfg->cbndx));
+
+       if (smmu->version > 1) {
+               /* CBA2R */
+#ifdef CONFIG_64BIT
+               reg = CBA2R_RW64_64BIT;
+#else
+               reg = CBA2R_RW64_32BIT;
+#endif
+               writel_relaxed(reg,
+                              gr1_base + ARM_SMMU_GR1_CBA2R(root_cfg->cbndx));
+
+               /* TTBCR2 */
+               switch (smmu->input_size) {
+               case 32:
+                       reg = (TTBCR2_ADDR_32 << TTBCR2_SEP_SHIFT);
+                       break;
+               case 36:
+                       reg = (TTBCR2_ADDR_36 << TTBCR2_SEP_SHIFT);
+                       break;
+               case 39:
+                       reg = (TTBCR2_ADDR_40 << TTBCR2_SEP_SHIFT);
+                       break;
+               case 42:
+                       reg = (TTBCR2_ADDR_42 << TTBCR2_SEP_SHIFT);
+                       break;
+               case 44:
+                       reg = (TTBCR2_ADDR_44 << TTBCR2_SEP_SHIFT);
+                       break;
+               case 48:
+                       reg = (TTBCR2_ADDR_48 << TTBCR2_SEP_SHIFT);
+                       break;
+               }
+
+               switch (smmu->s1_output_size) {
+               case 32:
+                       reg |= (TTBCR2_ADDR_32 << TTBCR2_PASIZE_SHIFT);
+                       break;
+               case 36:
+                       reg |= (TTBCR2_ADDR_36 << TTBCR2_PASIZE_SHIFT);
+                       break;
+               case 39:
+                       reg |= (TTBCR2_ADDR_40 << TTBCR2_PASIZE_SHIFT);
+                       break;
+               case 42:
+                       reg |= (TTBCR2_ADDR_42 << TTBCR2_PASIZE_SHIFT);
+                       break;
+               case 44:
+                       reg |= (TTBCR2_ADDR_44 << TTBCR2_PASIZE_SHIFT);
+                       break;
+               case 48:
+                       reg |= (TTBCR2_ADDR_48 << TTBCR2_PASIZE_SHIFT);
+                       break;
+               }
+
+               if (stage1)
+                       writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR2);
+       }
+
+       /* TTBR0 */
+       reg = __pa(root_cfg->pgd);
+#ifndef __BIG_ENDIAN
+       writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_LO);
+       reg = (phys_addr_t)__pa(root_cfg->pgd) >> 32;
+       writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_HI);
+#else
+       writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_HI);
+       reg = (phys_addr_t)__pa(root_cfg->pgd) >> 32;
+       writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_LO);
+#endif
+
+       /*
+        * TTBCR
+        * We use long descriptor, with inner-shareable WBWA tables in TTBR0.
+        */
+       if (smmu->version > 1) {
+               if (PAGE_SIZE == SZ_4K)
+                       reg = TTBCR_TG0_4K;
+               else
+                       reg = TTBCR_TG0_64K;
+
+               if (!stage1) {
+                       switch (smmu->s2_output_size) {
+                       case 32:
+                               reg |= (TTBCR2_ADDR_32 << TTBCR_PASIZE_SHIFT);
+                               break;
+                       case 36:
+                               reg |= (TTBCR2_ADDR_36 << TTBCR_PASIZE_SHIFT);
+                               break;
+                       case 40:
+                               reg |= (TTBCR2_ADDR_40 << TTBCR_PASIZE_SHIFT);
+                               break;
+                       case 42:
+                               reg |= (TTBCR2_ADDR_42 << TTBCR_PASIZE_SHIFT);
+                               break;
+                       case 44:
+                               reg |= (TTBCR2_ADDR_44 << TTBCR_PASIZE_SHIFT);
+                               break;
+                       case 48:
+                               reg |= (TTBCR2_ADDR_48 << TTBCR_PASIZE_SHIFT);
+                               break;
+                       }
+               } else {
+                       reg |= (64 - smmu->s1_output_size) << TTBCR_T0SZ_SHIFT;
+               }
+       } else {
+               reg = 0;
+       }
+
+       reg |= TTBCR_EAE |
+             (TTBCR_SH_IS << TTBCR_SH0_SHIFT) |
+             (TTBCR_RGN_WBWA << TTBCR_ORGN0_SHIFT) |
+             (TTBCR_RGN_WBWA << TTBCR_IRGN0_SHIFT) |
+             (TTBCR_SL0_LVL_1 << TTBCR_SL0_SHIFT);
+       writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR);
+
+       /* MAIR0 (stage-1 only) */
+       if (stage1) {
+               reg = (MAIR_ATTR_NC << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_NC)) |
+                     (MAIR_ATTR_WBRWA << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_CACHE)) |
+                     (MAIR_ATTR_DEVICE << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_DEV));
+               writel_relaxed(reg, cb_base + ARM_SMMU_CB_S1_MAIR0);
+       }
+
+       /* Nuke the TLB */
+       writel_relaxed(root_cfg->vmid, gr0_base + ARM_SMMU_GR0_TLBIVMID);
+       arm_smmu_tlb_sync(smmu);
+
+       /* SCTLR */
+       reg = SCTLR_CFCFG | SCTLR_CFIE | SCTLR_CFRE | SCTLR_M | SCTLR_EAE_SBOP;
+       if (stage1)
+               reg |= SCTLR_S1_ASIDPNE;
+#ifdef __BIG_ENDIAN
+       reg |= SCTLR_E;
+#endif
+       writel(reg, cb_base + ARM_SMMU_CB_SCTLR);
+}
+
+static int arm_smmu_init_domain_context(struct iommu_domain *domain,
+                                       struct device *dev)
+{
+       int irq, ret, start;
+       struct arm_smmu_domain *smmu_domain = domain->priv;
+       struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
+       struct arm_smmu_device *smmu, *parent;
+
+       /*
+        * Walk the SMMU chain to find the root device for this chain.
+        * We assume that no masters have translations which terminate
+        * early, and therefore check that the root SMMU does indeed have
+        * a StreamID for the master in question.
+        */
+       parent = dev->archdata.iommu;
+       smmu_domain->output_mask = -1;
+       do {
+               smmu = parent;
+               smmu_domain->output_mask &= (1ULL << smmu->s2_output_size) - 1;
+       } while ((parent = find_parent_smmu(smmu)));
+
+       if (!find_smmu_master(smmu, dev->of_node)) {
+               dev_err(dev, "unable to find root SMMU for device\n");
+               return -ENODEV;
+       }
+
+       ret = __arm_smmu_alloc_bitmap(smmu->vmid_map, 0, ARM_SMMU_NUM_VMIDS);
+       if (IS_ERR_VALUE(ret))
+               return ret;
+
+       root_cfg->vmid = ret;
+       if (smmu->features & ARM_SMMU_FEAT_TRANS_NESTED) {
+               /*
+                * We will likely want to change this if/when KVM gets
+                * involved.
+                */
+               root_cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS;
+               start = smmu->num_s2_context_banks;
+       } else if (smmu->features & ARM_SMMU_FEAT_TRANS_S2) {
+               root_cfg->cbar = CBAR_TYPE_S2_TRANS;
+               start = 0;
+       } else {
+               root_cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS;
+               start = smmu->num_s2_context_banks;
+       }
+
+       ret = __arm_smmu_alloc_bitmap(smmu->context_map, start,
+                                     smmu->num_context_banks);
+       if (IS_ERR_VALUE(ret))
+               goto out_free_vmid;
+
+       root_cfg->cbndx = ret;
+
+       if (smmu->version == 1) {
+               root_cfg->irptndx = atomic_inc_return(&smmu->irptndx);
+               root_cfg->irptndx %= smmu->num_context_irqs;
+       } else {
+               root_cfg->irptndx = root_cfg->cbndx;
+       }
+
+       irq = smmu->irqs[smmu->num_global_irqs + root_cfg->irptndx];
+       ret = request_irq(irq, arm_smmu_context_fault, IRQF_SHARED,
+                         "arm-smmu-context-fault", domain);
+       if (IS_ERR_VALUE(ret)) {
+               dev_err(smmu->dev, "failed to request context IRQ %d (%u)\n",
+                       root_cfg->irptndx, irq);
+               root_cfg->irptndx = -1;
+               goto out_free_context;
+       }
+
+       root_cfg->smmu = smmu;
+       arm_smmu_init_context_bank(smmu_domain);
+       return ret;
+
+out_free_context:
+       __arm_smmu_free_bitmap(smmu->context_map, root_cfg->cbndx);
+out_free_vmid:
+       __arm_smmu_free_bitmap(smmu->vmid_map, root_cfg->vmid);
+       return ret;
+}
+
+static void arm_smmu_destroy_domain_context(struct iommu_domain *domain)
+{
+       struct arm_smmu_domain *smmu_domain = domain->priv;
+       struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
+       struct arm_smmu_device *smmu = root_cfg->smmu;
+       int irq;
+
+       if (!smmu)
+               return;
+
+       if (root_cfg->irptndx != -1) {
+               irq = smmu->irqs[smmu->num_global_irqs + root_cfg->irptndx];
+               free_irq(irq, domain);
+       }
+
+       __arm_smmu_free_bitmap(smmu->vmid_map, root_cfg->vmid);
+       __arm_smmu_free_bitmap(smmu->context_map, root_cfg->cbndx);
+}
+
+static int arm_smmu_domain_init(struct iommu_domain *domain)
+{
+       struct arm_smmu_domain *smmu_domain;
+       pgd_t *pgd;
+
+       /*
+        * Allocate the domain and initialise some of its data structures.
+        * We can't really do anything meaningful until we've added a
+        * master.
+        */
+       smmu_domain = kzalloc(sizeof(*smmu_domain), GFP_KERNEL);
+       if (!smmu_domain)
+               return -ENOMEM;
+
+       pgd = kzalloc(PTRS_PER_PGD * sizeof(pgd_t), GFP_KERNEL);
+       if (!pgd)
+               goto out_free_domain;
+       smmu_domain->root_cfg.pgd = pgd;
+
+       spin_lock_init(&smmu_domain->lock);
+       domain->priv = smmu_domain;
+       return 0;
+
+out_free_domain:
+       kfree(smmu_domain);
+       return -ENOMEM;
+}
+
+static void arm_smmu_free_ptes(pmd_t *pmd)
+{
+       pgtable_t table = pmd_pgtable(*pmd);
+       pgtable_page_dtor(table);
+       __free_page(table);
+}
+
+static void arm_smmu_free_pmds(pud_t *pud)
+{
+       int i;
+       pmd_t *pmd, *pmd_base = pmd_offset(pud, 0);
+
+       pmd = pmd_base;
+       for (i = 0; i < PTRS_PER_PMD; ++i) {
+               if (pmd_none(*pmd))
+                       continue;
+
+               arm_smmu_free_ptes(pmd);
+               pmd++;
+       }
+
+       pmd_free(NULL, pmd_base);
+}
+
+static void arm_smmu_free_puds(pgd_t *pgd)
+{
+       int i;
+       pud_t *pud, *pud_base = pud_offset(pgd, 0);
+
+       pud = pud_base;
+       for (i = 0; i < PTRS_PER_PUD; ++i) {
+               if (pud_none(*pud))
+                       continue;
+
+               arm_smmu_free_pmds(pud);
+               pud++;
+       }
+
+       pud_free(NULL, pud_base);
+}
+
+static void arm_smmu_free_pgtables(struct arm_smmu_domain *smmu_domain)
+{
+       int i;
+       struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
+       pgd_t *pgd, *pgd_base = root_cfg->pgd;
+
+       /*
+        * Recursively free the page tables for this domain. We don't
+        * care about speculative TLB filling, because the TLB will be
+        * nuked next time this context bank is re-allocated and no devices
+        * currently map to these tables.
+        */
+       pgd = pgd_base;
+       for (i = 0; i < PTRS_PER_PGD; ++i) {
+               if (pgd_none(*pgd))
+                       continue;
+               arm_smmu_free_puds(pgd);
+               pgd++;
+       }
+
+       kfree(pgd_base);
+}
+
+static void arm_smmu_domain_destroy(struct iommu_domain *domain)
+{
+       struct arm_smmu_domain *smmu_domain = domain->priv;
+       arm_smmu_destroy_domain_context(domain);
+       arm_smmu_free_pgtables(smmu_domain);
+       kfree(smmu_domain);
+}
+
+static int arm_smmu_master_configure_smrs(struct arm_smmu_device *smmu,
+                                         struct arm_smmu_master *master)
+{
+       int i;
+       struct arm_smmu_smr *smrs;
+       void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
+
+       if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH))
+               return 0;
+
+       if (master->smrs)
+               return -EEXIST;
+
+       smrs = kmalloc(sizeof(*smrs) * master->num_streamids, GFP_KERNEL);
+       if (!smrs) {
+               dev_err(smmu->dev, "failed to allocate %d SMRs for master %s\n",
+                       master->num_streamids, master->of_node->name);
+               return -ENOMEM;
+       }
+
+       /* Allocate the SMRs on the root SMMU */
+       for (i = 0; i < master->num_streamids; ++i) {
+               int idx = __arm_smmu_alloc_bitmap(smmu->smr_map, 0,
+                                                 smmu->num_mapping_groups);
+               if (IS_ERR_VALUE(idx)) {
+                       dev_err(smmu->dev, "failed to allocate free SMR\n");
+                       goto err_free_smrs;
+               }
+
+               smrs[i] = (struct arm_smmu_smr) {
+                       .idx    = idx,
+                       .mask   = 0, /* We don't currently share SMRs */
+                       .id     = master->streamids[i],
+               };
+       }
+
+       /* It worked! Now, poke the actual hardware */
+       for (i = 0; i < master->num_streamids; ++i) {
+               u32 reg = SMR_VALID | smrs[i].id << SMR_ID_SHIFT |
+                         smrs[i].mask << SMR_MASK_SHIFT;
+               writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_SMR(smrs[i].idx));
+       }
+
+       master->smrs = smrs;
+       return 0;
+
+err_free_smrs:
+       while (--i >= 0)
+               __arm_smmu_free_bitmap(smmu->smr_map, smrs[i].idx);
+       kfree(smrs);
+       return -ENOSPC;
+}
+
+static void arm_smmu_master_free_smrs(struct arm_smmu_device *smmu,
+                                     struct arm_smmu_master *master)
+{
+       int i;
+       void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
+       struct arm_smmu_smr *smrs = master->smrs;
+
+       /* Invalidate the SMRs before freeing back to the allocator */
+       for (i = 0; i < master->num_streamids; ++i) {
+               u8 idx = smrs[i].idx;
+               writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(idx));
+               __arm_smmu_free_bitmap(smmu->smr_map, idx);
+       }
+
+       master->smrs = NULL;
+       kfree(smrs);
+}
+
+static void arm_smmu_bypass_stream_mapping(struct arm_smmu_device *smmu,
+                                          struct arm_smmu_master *master)
+{
+       int i;
+       void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
+
+       for (i = 0; i < master->num_streamids; ++i) {
+               u16 sid = master->streamids[i];
+               writel_relaxed(S2CR_TYPE_BYPASS,
+                              gr0_base + ARM_SMMU_GR0_S2CR(sid));
+       }
+}
+
+static int arm_smmu_domain_add_master(struct arm_smmu_domain *smmu_domain,
+                                     struct arm_smmu_master *master)
+{
+       int i, ret;
+       struct arm_smmu_device *parent, *smmu = smmu_domain->root_cfg.smmu;
+       void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
+
+       ret = arm_smmu_master_configure_smrs(smmu, master);
+       if (ret)
+               return ret;
+
+       /* Bypass the leaves */
+       smmu = smmu_domain->leaf_smmu;
+       while ((parent = find_parent_smmu(smmu))) {
+               /*
+                * We won't have a StreamID match for anything but the root
+                * smmu, so we only need to worry about StreamID indexing,
+                * where we must install bypass entries in the S2CRs.
+                */
+               if (smmu->features & ARM_SMMU_FEAT_STREAM_MATCH)
+                       continue;
+
+               arm_smmu_bypass_stream_mapping(smmu, master);
+               smmu = parent;
+       }
+
+       /* Now we're at the root, time to point at our context bank */
+       for (i = 0; i < master->num_streamids; ++i) {
+               u32 idx, s2cr;
+               idx = master->smrs ? master->smrs[i].idx : master->streamids[i];
+               s2cr = (S2CR_TYPE_TRANS << S2CR_TYPE_SHIFT) |
+                      (smmu_domain->root_cfg.cbndx << S2CR_CBNDX_SHIFT);
+               writel_relaxed(s2cr, gr0_base + ARM_SMMU_GR0_S2CR(idx));
+       }
+
+       return 0;
+}
+
+static void arm_smmu_domain_remove_master(struct arm_smmu_domain *smmu_domain,
+                                         struct arm_smmu_master *master)
+{
+       struct arm_smmu_device *smmu = smmu_domain->root_cfg.smmu;
+
+       /*
+        * We *must* clear the S2CR first, because freeing the SMR means
+        * that it can be re-allocated immediately.
+        */
+       arm_smmu_bypass_stream_mapping(smmu, master);
+       arm_smmu_master_free_smrs(smmu, master);
+}
+
+static int arm_smmu_attach_dev(struct iommu_domain *domain, struct device *dev)
+{
+       int ret = -EINVAL;
+       struct arm_smmu_domain *smmu_domain = domain->priv;
+       struct arm_smmu_device *device_smmu = dev->archdata.iommu;
+       struct arm_smmu_master *master;
+
+       if (!device_smmu) {
+               dev_err(dev, "cannot attach to SMMU, is it on the same bus?\n");
+               return -ENXIO;
+       }
+
+       /*
+        * Sanity check the domain. We don't currently support domains
+        * that cross between different SMMU chains.
+        */
+       spin_lock(&smmu_domain->lock);
+       if (!smmu_domain->leaf_smmu) {
+               /* Now that we have a master, we can finalise the domain */
+               ret = arm_smmu_init_domain_context(domain, dev);
+               if (IS_ERR_VALUE(ret))
+                       goto err_unlock;
+
+               smmu_domain->leaf_smmu = device_smmu;
+       } else if (smmu_domain->leaf_smmu != device_smmu) {
+               dev_err(dev,
+                       "cannot attach to SMMU %s whilst already attached to domain on SMMU %s\n",
+                       dev_name(smmu_domain->leaf_smmu->dev),
+                       dev_name(device_smmu->dev));
+               goto err_unlock;
+       }
+       spin_unlock(&smmu_domain->lock);
+
+       /* Looks ok, so add the device to the domain */
+       master = find_smmu_master(smmu_domain->leaf_smmu, dev->of_node);
+       if (!master)
+               return -ENODEV;
+
+       return arm_smmu_domain_add_master(smmu_domain, master);
+
+err_unlock:
+       spin_unlock(&smmu_domain->lock);
+       return ret;
+}
+
+static void arm_smmu_detach_dev(struct iommu_domain *domain, struct device *dev)
+{
+       struct arm_smmu_domain *smmu_domain = domain->priv;
+       struct arm_smmu_master *master;
+
+       master = find_smmu_master(smmu_domain->leaf_smmu, dev->of_node);
+       if (master)
+               arm_smmu_domain_remove_master(smmu_domain, master);
+}
+
+static void arm_smmu_flush_pgtable(struct arm_smmu_device *smmu, void *addr,
+                                  size_t size)
+{
+       unsigned long offset = (unsigned long)addr & ~PAGE_MASK;
+
+       /*
+        * If the SMMU can't walk tables in the CPU caches, treat them
+        * like non-coherent DMA since we need to flush the new entries
+        * all the way out to memory. There's no possibility of recursion
+        * here as the SMMU table walker will not be wired through another
+        * SMMU.
+        */
+       if (!(smmu->features & ARM_SMMU_FEAT_COHERENT_WALK))
+               dma_map_page(smmu->dev, virt_to_page(addr), offset, size,
+                            DMA_TO_DEVICE);
+}
+
+static bool arm_smmu_pte_is_contiguous_range(unsigned long addr,
+                                            unsigned long end)
+{
+       return !(addr & ~ARM_SMMU_PTE_CONT_MASK) &&
+               (addr + ARM_SMMU_PTE_CONT_SIZE <= end);
+}
+
+static int arm_smmu_alloc_init_pte(struct arm_smmu_device *smmu, pmd_t *pmd,
+                                  unsigned long addr, unsigned long end,
+                                  unsigned long pfn, int flags, int stage)
+{
+       pte_t *pte, *start;
+       pteval_t pteval = ARM_SMMU_PTE_PAGE | ARM_SMMU_PTE_AF;
+
+       if (pmd_none(*pmd)) {
+               /* Allocate a new set of tables */
+               pgtable_t table = alloc_page(PGALLOC_GFP);
+               if (!table)
+                       return -ENOMEM;
+
+               arm_smmu_flush_pgtable(smmu, page_address(table),
+                                      ARM_SMMU_PTE_HWTABLE_SIZE);
+               pgtable_page_ctor(table);
+               pmd_populate(NULL, pmd, table);
+               arm_smmu_flush_pgtable(smmu, pmd, sizeof(*pmd));
+       }
+
+       if (stage == 1) {
+               pteval |= ARM_SMMU_PTE_AP_UNPRIV;
+               if (!(flags & IOMMU_WRITE) && (flags & IOMMU_READ))
+                       pteval |= ARM_SMMU_PTE_AP_RDONLY;
+
+               if (flags & IOMMU_CACHE)
+                       pteval |= (MAIR_ATTR_IDX_CACHE <<
+                                  ARM_SMMU_PTE_ATTRINDX_SHIFT);
+       } else {
+               pteval |= ARM_SMMU_PTE_HAP_FAULT;
+               if (flags & IOMMU_READ)
+                       pteval |= ARM_SMMU_PTE_HAP_READ;
+               if (flags & IOMMU_WRITE)
+                       pteval |= ARM_SMMU_PTE_HAP_WRITE;
+               if (flags & IOMMU_CACHE)
+                       pteval |= ARM_SMMU_PTE_MEMATTR_OIWB;
+               else
+                       pteval |= ARM_SMMU_PTE_MEMATTR_NC;
+       }
+
+       /* If no access, create a faulting entry to avoid TLB fills */
+       if (!(flags & (IOMMU_READ | IOMMU_WRITE)))
+               pteval &= ~ARM_SMMU_PTE_PAGE;
+
+       pteval |= ARM_SMMU_PTE_SH_IS;
+       start = pmd_page_vaddr(*pmd) + pte_index(addr);
+       pte = start;
+
+       /*
+        * Install the page table entries. This is fairly complicated
+        * since we attempt to make use of the contiguous hint in the
+        * ptes where possible. The contiguous hint indicates a series
+        * of ARM_SMMU_PTE_CONT_ENTRIES ptes mapping a physically
+        * contiguous region with the following constraints:
+        *
+        *   - The region start is aligned to ARM_SMMU_PTE_CONT_SIZE
+        *   - Each pte in the region has the contiguous hint bit set
+        *
+        * This complicates unmapping (also handled by this code, when
+        * neither IOMMU_READ or IOMMU_WRITE are set) because it is
+        * possible, yet highly unlikely, that a client may unmap only
+        * part of a contiguous range. This requires clearing of the
+        * contiguous hint bits in the range before installing the new
+        * faulting entries.
+        *
+        * Note that re-mapping an address range without first unmapping
+        * it is not supported, so TLB invalidation is not required here
+        * and is instead performed at unmap and domain-init time.
+        */
+       do {
+               int i = 1;
+               pteval &= ~ARM_SMMU_PTE_CONT;
+
+               if (arm_smmu_pte_is_contiguous_range(addr, end)) {
+                       i = ARM_SMMU_PTE_CONT_ENTRIES;
+                       pteval |= ARM_SMMU_PTE_CONT;
+               } else if (pte_val(*pte) &
+                          (ARM_SMMU_PTE_CONT | ARM_SMMU_PTE_PAGE)) {
+                       int j;
+                       pte_t *cont_start;
+                       unsigned long idx = pte_index(addr);
+
+                       idx &= ~(ARM_SMMU_PTE_CONT_ENTRIES - 1);
+                       cont_start = pmd_page_vaddr(*pmd) + idx;
+                       for (j = 0; j < ARM_SMMU_PTE_CONT_ENTRIES; ++j)
+                               pte_val(*(cont_start + j)) &= ~ARM_SMMU_PTE_CONT;
+
+                       arm_smmu_flush_pgtable(smmu, cont_start,
+                                              sizeof(*pte) *
+                                              ARM_SMMU_PTE_CONT_ENTRIES);
+               }
+
+               do {
+                       *pte = pfn_pte(pfn, __pgprot(pteval));
+               } while (pte++, pfn++, addr += PAGE_SIZE, --i);
+       } while (addr != end);
+
+       arm_smmu_flush_pgtable(smmu, start, sizeof(*pte) * (pte - start));
+       return 0;
+}
+
+static int arm_smmu_alloc_init_pmd(struct arm_smmu_device *smmu, pud_t *pud,
+                                  unsigned long addr, unsigned long end,
+                                  phys_addr_t phys, int flags, int stage)
+{
+       int ret;
+       pmd_t *pmd;
+       unsigned long next, pfn = __phys_to_pfn(phys);
+
+#ifndef __PAGETABLE_PMD_FOLDED
+       if (pud_none(*pud)) {
+               pmd = pmd_alloc_one(NULL, addr);
+               if (!pmd)
+                       return -ENOMEM;
+       } else
+#endif
+               pmd = pmd_offset(pud, addr);
+
+       do {
+               next = pmd_addr_end(addr, end);
+               ret = arm_smmu_alloc_init_pte(smmu, pmd, addr, end, pfn,
+                                             flags, stage);
+               pud_populate(NULL, pud, pmd);
+               arm_smmu_flush_pgtable(smmu, pud, sizeof(*pud));
+               phys += next - addr;
+       } while (pmd++, addr = next, addr < end);
+
+       return ret;
+}
+
+static int arm_smmu_alloc_init_pud(struct arm_smmu_device *smmu, pgd_t *pgd,
+                                  unsigned long addr, unsigned long end,
+                                  phys_addr_t phys, int flags, int stage)
+{
+       int ret = 0;
+       pud_t *pud;
+       unsigned long next;
+
+#ifndef __PAGETABLE_PUD_FOLDED
+       if (pgd_none(*pgd)) {
+               pud = pud_alloc_one(NULL, addr);
+               if (!pud)
+                       return -ENOMEM;
+       } else
+#endif
+               pud = pud_offset(pgd, addr);
+
+       do {
+               next = pud_addr_end(addr, end);
+               ret = arm_smmu_alloc_init_pmd(smmu, pud, addr, next, phys,
+                                             flags, stage);
+               pgd_populate(NULL, pud, pgd);
+               arm_smmu_flush_pgtable(smmu, pgd, sizeof(*pgd));
+               phys += next - addr;
+       } while (pud++, addr = next, addr < end);
+
+       return ret;
+}
+
+static int arm_smmu_handle_mapping(struct arm_smmu_domain *smmu_domain,
+                                  unsigned long iova, phys_addr_t paddr,
+                                  size_t size, int flags)
+{
+       int ret, stage;
+       unsigned long end;
+       phys_addr_t input_mask, output_mask;
+       struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
+       pgd_t *pgd = root_cfg->pgd;
+       struct arm_smmu_device *smmu = root_cfg->smmu;
+
+       if (root_cfg->cbar == CBAR_TYPE_S2_TRANS) {
+               stage = 2;
+               output_mask = (1ULL << smmu->s2_output_size) - 1;
+       } else {
+               stage = 1;
+               output_mask = (1ULL << smmu->s1_output_size) - 1;
+       }
+
+       if (!pgd)
+               return -EINVAL;
+
+       if (size & ~PAGE_MASK)
+               return -EINVAL;
+
+       input_mask = (1ULL << smmu->input_size) - 1;
+       if ((phys_addr_t)iova & ~input_mask)
+               return -ERANGE;
+
+       if (paddr & ~output_mask)
+               return -ERANGE;
+
+       spin_lock(&smmu_domain->lock);
+       pgd += pgd_index(iova);
+       end = iova + size;
+       do {
+               unsigned long next = pgd_addr_end(iova, end);
+
+               ret = arm_smmu_alloc_init_pud(smmu, pgd, iova, next, paddr,
+                                             flags, stage);
+               if (ret)
+                       goto out_unlock;
+
+               paddr += next - iova;
+               iova = next;
+       } while (pgd++, iova != end);
+
+out_unlock:
+       spin_unlock(&smmu_domain->lock);
+
+       /* Ensure new page tables are visible to the hardware walker */
+       if (smmu->features & ARM_SMMU_FEAT_COHERENT_WALK)
+               dsb();
+
+       return ret;
+}
+
+static int arm_smmu_map(struct iommu_domain *domain, unsigned long iova,
+                       phys_addr_t paddr, size_t size, int flags)
+{
+       struct arm_smmu_domain *smmu_domain = domain->priv;
+       struct arm_smmu_device *smmu = smmu_domain->leaf_smmu;
+
+       if (!smmu_domain || !smmu)
+               return -ENODEV;
+
+       /* Check for silent address truncation up the SMMU chain. */
+       if ((phys_addr_t)iova & ~smmu_domain->output_mask)
+               return -ERANGE;
+
+       return arm_smmu_handle_mapping(smmu_domain, iova, paddr, size, flags);
+}
+
+static size_t arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
+                            size_t size)
+{
+       int ret;
+       struct arm_smmu_domain *smmu_domain = domain->priv;
+       struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
+       struct arm_smmu_device *smmu = root_cfg->smmu;
+       void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
+
+       ret = arm_smmu_handle_mapping(smmu_domain, iova, 0, size, 0);
+       writel_relaxed(root_cfg->vmid, gr0_base + ARM_SMMU_GR0_TLBIVMID);
+       arm_smmu_tlb_sync(smmu);
+       return ret ? ret : size;
+}
+
+static phys_addr_t arm_smmu_iova_to_phys(struct iommu_domain *domain,
+                                        dma_addr_t iova)
+{
+       pgd_t *pgd;
+       pud_t *pud;
+       pmd_t *pmd;
+       pte_t *pte;
+       struct arm_smmu_domain *smmu_domain = domain->priv;
+       struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
+       struct arm_smmu_device *smmu = root_cfg->smmu;
+
+       spin_lock(&smmu_domain->lock);
+       pgd = root_cfg->pgd;
+       if (!pgd)
+               goto err_unlock;
+
+       pgd += pgd_index(iova);
+       if (pgd_none_or_clear_bad(pgd))
+               goto err_unlock;
+
+       pud = pud_offset(pgd, iova);
+       if (pud_none_or_clear_bad(pud))
+               goto err_unlock;
+
+       pmd = pmd_offset(pud, iova);
+       if (pmd_none_or_clear_bad(pmd))
+               goto err_unlock;
+
+       pte = pmd_page_vaddr(*pmd) + pte_index(iova);
+       if (pte_none(pte))
+               goto err_unlock;
+
+       spin_unlock(&smmu_domain->lock);
+       return __pfn_to_phys(pte_pfn(*pte)) | (iova & ~PAGE_MASK);
+
+err_unlock:
+       spin_unlock(&smmu_domain->lock);
+       dev_warn(smmu->dev,
+                "invalid (corrupt?) page tables detected for iova 0x%llx\n",
+                (unsigned long long)iova);
+       return -EINVAL;
+}
+
+static int arm_smmu_domain_has_cap(struct iommu_domain *domain,
+                                  unsigned long cap)
+{
+       unsigned long caps = 0;
+       struct arm_smmu_domain *smmu_domain = domain->priv;
+
+       if (smmu_domain->root_cfg.smmu->features & ARM_SMMU_FEAT_COHERENT_WALK)
+               caps |= IOMMU_CAP_CACHE_COHERENCY;
+
+       return !!(cap & caps);
+}
+
+static int arm_smmu_add_device(struct device *dev)
+{
+       struct arm_smmu_device *child, *parent, *smmu;
+       struct arm_smmu_master *master = NULL;
+
+       spin_lock(&arm_smmu_devices_lock);
+       list_for_each_entry(parent, &arm_smmu_devices, list) {
+               smmu = parent;
+
+               /* Try to find a child of the current SMMU. */
+               list_for_each_entry(child, &arm_smmu_devices, list) {
+                       if (child->parent_of_node == parent->dev->of_node) {
+                               /* Does the child sit above our master? */
+                               master = find_smmu_master(child, dev->of_node);
+                               if (master) {
+                                       smmu = NULL;
+                                       break;
+                               }
+                       }
+               }
+
+               /* We found some children, so keep searching. */
+               if (!smmu) {
+                       master = NULL;
+                       continue;
+               }
+
+               master = find_smmu_master(smmu, dev->of_node);
+               if (master)
+                       break;
+       }
+       spin_unlock(&arm_smmu_devices_lock);
+
+       if (!master)
+               return -ENODEV;
+
+       dev->archdata.iommu = smmu;
+       return 0;
+}
+
+static void arm_smmu_remove_device(struct device *dev)
+{
+       dev->archdata.iommu = NULL;
+}
+
+static struct iommu_ops arm_smmu_ops = {
+       .domain_init    = arm_smmu_domain_init,
+       .domain_destroy = arm_smmu_domain_destroy,
+       .attach_dev     = arm_smmu_attach_dev,
+       .detach_dev     = arm_smmu_detach_dev,
+       .map            = arm_smmu_map,
+       .unmap          = arm_smmu_unmap,
+       .iova_to_phys   = arm_smmu_iova_to_phys,
+       .domain_has_cap = arm_smmu_domain_has_cap,
+       .add_device     = arm_smmu_add_device,
+       .remove_device  = arm_smmu_remove_device,
+       .pgsize_bitmap  = (SECTION_SIZE |
+                          ARM_SMMU_PTE_CONT_SIZE |
+                          PAGE_SIZE),
+};
+
+static void arm_smmu_device_reset(struct arm_smmu_device *smmu)
+{
+       void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
+       int i = 0;
+       u32 scr0 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sCR0);
+
+       /* Mark all SMRn as invalid and all S2CRn as bypass */
+       for (i = 0; i < smmu->num_mapping_groups; ++i) {
+               writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(i));
+               writel_relaxed(S2CR_TYPE_BYPASS, gr0_base + ARM_SMMU_GR0_S2CR(i));
+       }
+
+       /* Invalidate the TLB, just in case */
+       writel_relaxed(0, gr0_base + ARM_SMMU_GR0_STLBIALL);
+       writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLH);
+       writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLNSNH);
+
+       /* Enable fault reporting */
+       scr0 |= (sCR0_GFRE | sCR0_GFIE | sCR0_GCFGFRE | sCR0_GCFGFIE);
+
+       /* Disable TLB broadcasting. */
+       scr0 |= (sCR0_VMIDPNE | sCR0_PTM);
+
+       /* Enable client access, but bypass when no mapping is found */
+       scr0 &= ~(sCR0_CLIENTPD | sCR0_USFCFG);
+
+       /* Disable forced broadcasting */
+       scr0 &= ~sCR0_FB;
+
+       /* Don't upgrade barriers */
+       scr0 &= ~(sCR0_BSU_MASK << sCR0_BSU_SHIFT);
+
+       /* Push the button */
+       arm_smmu_tlb_sync(smmu);
+       writel(scr0, gr0_base + ARM_SMMU_GR0_sCR0);
+}
+
+static int arm_smmu_id_size_to_bits(int size)
+{
+       switch (size) {
+       case 0:
+               return 32;
+       case 1:
+               return 36;
+       case 2:
+               return 40;
+       case 3:
+               return 42;
+       case 4:
+               return 44;
+       case 5:
+       default:
+               return 48;
+       }
+}
+
+static int arm_smmu_device_cfg_probe(struct arm_smmu_device *smmu)
+{
+       unsigned long size;
+       void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
+       u32 id;
+
+       dev_notice(smmu->dev, "probing hardware configuration...\n");
+
+       /* Primecell ID */
+       id = readl_relaxed(gr0_base + ARM_SMMU_GR0_PIDR2);
+       smmu->version = ((id >> PIDR2_ARCH_SHIFT) & PIDR2_ARCH_MASK) + 1;
+       dev_notice(smmu->dev, "SMMUv%d with:\n", smmu->version);
+
+       /* ID0 */
+       id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID0);
+#ifndef CONFIG_64BIT
+       if (((id >> ID0_PTFS_SHIFT) & ID0_PTFS_MASK) == ID0_PTFS_V8_ONLY) {
+               dev_err(smmu->dev, "\tno v7 descriptor support!\n");
+               return -ENODEV;
+       }
+#endif
+       if (id & ID0_S1TS) {
+               smmu->features |= ARM_SMMU_FEAT_TRANS_S1;
+               dev_notice(smmu->dev, "\tstage 1 translation\n");
+       }
+
+       if (id & ID0_S2TS) {
+               smmu->features |= ARM_SMMU_FEAT_TRANS_S2;
+               dev_notice(smmu->dev, "\tstage 2 translation\n");
+       }
+
+       if (id & ID0_NTS) {
+               smmu->features |= ARM_SMMU_FEAT_TRANS_NESTED;
+               dev_notice(smmu->dev, "\tnested translation\n");
+       }
+
+       if (!(smmu->features &
+               (ARM_SMMU_FEAT_TRANS_S1 | ARM_SMMU_FEAT_TRANS_S2 |
+                ARM_SMMU_FEAT_TRANS_NESTED))) {
+               dev_err(smmu->dev, "\tno translation support!\n");
+               return -ENODEV;
+       }
+
+       if (id & ID0_CTTW) {
+               smmu->features |= ARM_SMMU_FEAT_COHERENT_WALK;
+               dev_notice(smmu->dev, "\tcoherent table walk\n");
+       }
+
+       if (id & ID0_SMS) {
+               u32 smr, sid, mask;
+
+               smmu->features |= ARM_SMMU_FEAT_STREAM_MATCH;
+               smmu->num_mapping_groups = (id >> ID0_NUMSMRG_SHIFT) &
+                                          ID0_NUMSMRG_MASK;
+               if (smmu->num_mapping_groups == 0) {
+                       dev_err(smmu->dev,
+                               "stream-matching supported, but no SMRs present!\n");
+                       return -ENODEV;
+               }
+
+               smr = SMR_MASK_MASK << SMR_MASK_SHIFT;
+               smr |= (SMR_ID_MASK << SMR_ID_SHIFT);
+               writel_relaxed(smr, gr0_base + ARM_SMMU_GR0_SMR(0));
+               smr = readl_relaxed(gr0_base + ARM_SMMU_GR0_SMR(0));
+
+               mask = (smr >> SMR_MASK_SHIFT) & SMR_MASK_MASK;
+               sid = (smr >> SMR_ID_SHIFT) & SMR_ID_MASK;
+               if ((mask & sid) != sid) {
+                       dev_err(smmu->dev,
+                               "SMR mask bits (0x%x) insufficient for ID field (0x%x)\n",
+                               mask, sid);
+                       return -ENODEV;
+               }
+
+               dev_notice(smmu->dev,
+                          "\tstream matching with %u register groups, mask 0x%x",
+                          smmu->num_mapping_groups, mask);
+       }
+
+       /* ID1 */
+       id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID1);
+       smmu->pagesize = (id & ID1_PAGESIZE) ? SZ_64K : SZ_4K;
+
+       /* Check that we ioremapped enough */
+       size = 1 << (((id >> ID1_NUMPAGENDXB_SHIFT) & ID1_NUMPAGENDXB_MASK) + 1);
+       size *= (smmu->pagesize << 1);
+       if (smmu->size < size)
+               dev_warn(smmu->dev,
+                        "device is 0x%lx bytes but only mapped 0x%lx!\n",
+                        size, smmu->size);
+
+       smmu->num_s2_context_banks = (id >> ID1_NUMS2CB_SHIFT) &
+                                     ID1_NUMS2CB_MASK;
+       smmu->num_context_banks = (id >> ID1_NUMCB_SHIFT) & ID1_NUMCB_MASK;
+       if (smmu->num_s2_context_banks > smmu->num_context_banks) {
+               dev_err(smmu->dev, "impossible number of S2 context banks!\n");
+               return -ENODEV;
+       }
+       dev_notice(smmu->dev, "\t%u context banks (%u stage-2 only)\n",
+                  smmu->num_context_banks, smmu->num_s2_context_banks);
+
+       /* ID2 */
+       id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID2);
+       size = arm_smmu_id_size_to_bits((id >> ID2_IAS_SHIFT) & ID2_IAS_MASK);
+
+       /*
+        * Stage-1 output limited by stage-2 input size due to pgd
+        * allocation (PTRS_PER_PGD).
+        */
+#ifdef CONFIG_64BIT
+       /* Current maximum output size of 39 bits */
+       smmu->s1_output_size = min(39UL, size);
+#else
+       smmu->s1_output_size = min(32UL, size);
+#endif
+
+       /* The stage-2 output mask is also applied for bypass */
+       size = arm_smmu_id_size_to_bits((id >> ID2_OAS_SHIFT) & ID2_OAS_MASK);
+       smmu->s2_output_size = min((unsigned long)PHYS_MASK_SHIFT, size);
+
+       if (smmu->version == 1) {
+               smmu->input_size = 32;
+       } else {
+#ifdef CONFIG_64BIT
+               size = (id >> ID2_UBS_SHIFT) & ID2_UBS_MASK;
+               size = min(39, arm_smmu_id_size_to_bits(size));
+#else
+               size = 32;
+#endif
+               smmu->input_size = size;
+
+               if ((PAGE_SIZE == SZ_4K && !(id & ID2_PTFS_4K)) ||
+                   (PAGE_SIZE == SZ_64K && !(id & ID2_PTFS_64K)) ||
+                   (PAGE_SIZE != SZ_4K && PAGE_SIZE != SZ_64K)) {
+                       dev_err(smmu->dev, "CPU page size 0x%lx unsupported\n",
+                               PAGE_SIZE);
+                       return -ENODEV;
+               }
+       }
+
+       dev_notice(smmu->dev,
+                  "\t%lu-bit VA, %lu-bit IPA, %lu-bit PA\n",
+                  smmu->input_size, smmu->s1_output_size, smmu->s2_output_size);
+       return 0;
+}
+
+static int arm_smmu_device_dt_probe(struct platform_device *pdev)
+{
+       struct resource *res;
+       struct arm_smmu_device *smmu;
+       struct device_node *dev_node;
+       struct device *dev = &pdev->dev;
+       struct rb_node *node;
+       struct of_phandle_args masterspec;
+       int num_irqs, i, err;
+
+       smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
+       if (!smmu) {
+               dev_err(dev, "failed to allocate arm_smmu_device\n");
+               return -ENOMEM;
+       }
+       smmu->dev = dev;
+
+       res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       if (!res) {
+               dev_err(dev, "missing base address/size\n");
+               return -ENODEV;
+       }
+
+       smmu->size = resource_size(res);
+       smmu->base = devm_request_and_ioremap(dev, res);
+       if (!smmu->base)
+               return -EADDRNOTAVAIL;
+
+       if (of_property_read_u32(dev->of_node, "#global-interrupts",
+                                &smmu->num_global_irqs)) {
+               dev_err(dev, "missing #global-interrupts property\n");
+               return -ENODEV;
+       }
+
+       num_irqs = 0;
+       while ((res = platform_get_resource(pdev, IORESOURCE_IRQ, num_irqs))) {
+               num_irqs++;
+               if (num_irqs > smmu->num_global_irqs)
+                       smmu->num_context_irqs++;
+       }
+
+       if (num_irqs < smmu->num_global_irqs) {
+               dev_warn(dev, "found %d interrupts but expected at least %d\n",
+                        num_irqs, smmu->num_global_irqs);
+               smmu->num_global_irqs = num_irqs;
+       }
+       smmu->num_context_irqs = num_irqs - smmu->num_global_irqs;
+
+       smmu->irqs = devm_kzalloc(dev, sizeof(*smmu->irqs) * num_irqs,
+                                 GFP_KERNEL);
+       if (!smmu->irqs) {
+               dev_err(dev, "failed to allocate %d irqs\n", num_irqs);
+               return -ENOMEM;
+       }
+
+       for (i = 0; i < num_irqs; ++i) {
+               int irq = platform_get_irq(pdev, i);
+               if (irq < 0) {
+                       dev_err(dev, "failed to get irq index %d\n", i);
+                       return -ENODEV;
+               }
+               smmu->irqs[i] = irq;
+       }
+
+       i = 0;
+       smmu->masters = RB_ROOT;
+       while (!of_parse_phandle_with_args(dev->of_node, "mmu-masters",
+                                          "#stream-id-cells", i,
+                                          &masterspec)) {
+               err = register_smmu_master(smmu, dev, &masterspec);
+               if (err) {
+                       dev_err(dev, "failed to add master %s\n",
+                               masterspec.np->name);
+                       goto out_put_masters;
+               }
+
+               i++;
+       }
+       dev_notice(dev, "registered %d master devices\n", i);
+
+       if ((dev_node = of_parse_phandle(dev->of_node, "smmu-parent", 0)))
+               smmu->parent_of_node = dev_node;
+
+       err = arm_smmu_device_cfg_probe(smmu);
+       if (err)
+               goto out_put_parent;
+
+       if (smmu->version > 1 &&
+           smmu->num_context_banks != smmu->num_context_irqs) {
+               dev_err(dev,
+                       "found only %d context interrupt(s) but %d required\n",
+                       smmu->num_context_irqs, smmu->num_context_banks);
+               goto out_put_parent;
+       }
+
+       arm_smmu_device_reset(smmu);
+
+       for (i = 0; i < smmu->num_global_irqs; ++i) {
+               err = request_irq(smmu->irqs[i],
+                                 arm_smmu_global_fault,
+                                 IRQF_SHARED,
+                                 "arm-smmu global fault",
+                                 smmu);
+               if (err) {
+                       dev_err(dev, "failed to request global IRQ %d (%u)\n",
+                               i, smmu->irqs[i]);
+                       goto out_free_irqs;
+               }
+       }
+
+       INIT_LIST_HEAD(&smmu->list);
+       spin_lock(&arm_smmu_devices_lock);
+       list_add(&smmu->list, &arm_smmu_devices);
+       spin_unlock(&arm_smmu_devices_lock);
+       return 0;
+
+out_free_irqs:
+       while (i--)
+               free_irq(smmu->irqs[i], smmu);
+
+out_put_parent:
+       if (smmu->parent_of_node)
+               of_node_put(smmu->parent_of_node);
+
+out_put_masters:
+       for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
+               struct arm_smmu_master *master;
+               master = container_of(node, struct arm_smmu_master, node);
+               of_node_put(master->of_node);
+       }
+
+       return err;
+}
+
+static int arm_smmu_device_remove(struct platform_device *pdev)
+{
+       int i;
+       struct device *dev = &pdev->dev;
+       struct arm_smmu_device *curr, *smmu = NULL;
+       struct rb_node *node;
+
+       spin_lock(&arm_smmu_devices_lock);
+       list_for_each_entry(curr, &arm_smmu_devices, list) {
+               if (curr->dev == dev) {
+                       smmu = curr;
+                       list_del(&smmu->list);
+                       break;
+               }
+       }
+       spin_unlock(&arm_smmu_devices_lock);
+
+       if (!smmu)
+               return -ENODEV;
+
+       if (smmu->parent_of_node)
+               of_node_put(smmu->parent_of_node);
+
+       for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
+               struct arm_smmu_master *master;
+               master = container_of(node, struct arm_smmu_master, node);
+               of_node_put(master->of_node);
+       }
+
+       if (!bitmap_empty(smmu->vmid_map, ARM_SMMU_NUM_VMIDS))
+               dev_err(dev, "removing device with active domains!\n");
+
+       for (i = 0; i < smmu->num_global_irqs; ++i)
+               free_irq(smmu->irqs[i], smmu);
+
+       /* Turn the thing off */
+       writel(sCR0_CLIENTPD, ARM_SMMU_GR0(smmu) + ARM_SMMU_GR0_sCR0);
+       return 0;
+}
+
+#ifdef CONFIG_OF
+static struct of_device_id arm_smmu_of_match[] = {
+       { .compatible = "arm,smmu-v1", },
+       { .compatible = "arm,smmu-v2", },
+       { .compatible = "arm,mmu-400", },
+       { .compatible = "arm,mmu-500", },
+       { },
+};
+MODULE_DEVICE_TABLE(of, arm_smmu_of_match);
+#endif
+
+static struct platform_driver arm_smmu_driver = {
+       .driver = {
+               .owner          = THIS_MODULE,
+               .name           = "arm-smmu",
+               .of_match_table = of_match_ptr(arm_smmu_of_match),
+       },
+       .probe  = arm_smmu_device_dt_probe,
+       .remove = arm_smmu_device_remove,
+};
+
+static int __init arm_smmu_init(void)
+{
+       int ret;
+
+       ret = platform_driver_register(&arm_smmu_driver);
+       if (ret)
+               return ret;
+
+       /* Oh, for a proper bus abstraction */
+       if (!iommu_present(&platform_bus_type));
+               bus_set_iommu(&platform_bus_type, &arm_smmu_ops);
+
+       if (!iommu_present(&amba_bustype));
+               bus_set_iommu(&amba_bustype, &arm_smmu_ops);
+
+       return 0;
+}
+
+static void __exit arm_smmu_exit(void)
+{
+       return platform_driver_unregister(&arm_smmu_driver);
+}
+
+module_init(arm_smmu_init);
+module_exit(arm_smmu_exit);
+
+MODULE_DESCRIPTION("IOMMU API for ARM architected SMMU implementations");
+MODULE_AUTHOR("Will Deacon <will.deacon@arm.com>");
+MODULE_LICENSE("GPL v2");