Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / iommu / tegra-smmu.c
diff --git a/kernel/drivers/iommu/tegra-smmu.c b/kernel/drivers/iommu/tegra-smmu.c
new file mode 100644 (file)
index 0000000..c845d99
--- /dev/null
@@ -0,0 +1,747 @@
+/*
+ * Copyright (C) 2011-2014 NVIDIA CORPORATION.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/bitops.h>
+#include <linux/err.h>
+#include <linux/iommu.h>
+#include <linux/kernel.h>
+#include <linux/of.h>
+#include <linux/of_device.h>
+#include <linux/platform_device.h>
+#include <linux/slab.h>
+
+#include <soc/tegra/ahb.h>
+#include <soc/tegra/mc.h>
+
+struct tegra_smmu {
+       void __iomem *regs;
+       struct device *dev;
+
+       struct tegra_mc *mc;
+       const struct tegra_smmu_soc *soc;
+
+       unsigned long pfn_mask;
+
+       unsigned long *asids;
+       struct mutex lock;
+
+       struct list_head list;
+};
+
+struct tegra_smmu_as {
+       struct iommu_domain domain;
+       struct tegra_smmu *smmu;
+       unsigned int use_count;
+       struct page *count;
+       struct page *pd;
+       unsigned id;
+       u32 attr;
+};
+
+static struct tegra_smmu_as *to_smmu_as(struct iommu_domain *dom)
+{
+       return container_of(dom, struct tegra_smmu_as, domain);
+}
+
+static inline void smmu_writel(struct tegra_smmu *smmu, u32 value,
+                              unsigned long offset)
+{
+       writel(value, smmu->regs + offset);
+}
+
+static inline u32 smmu_readl(struct tegra_smmu *smmu, unsigned long offset)
+{
+       return readl(smmu->regs + offset);
+}
+
+#define SMMU_CONFIG 0x010
+#define  SMMU_CONFIG_ENABLE (1 << 0)
+
+#define SMMU_TLB_CONFIG 0x14
+#define  SMMU_TLB_CONFIG_HIT_UNDER_MISS (1 << 29)
+#define  SMMU_TLB_CONFIG_ROUND_ROBIN_ARBITRATION (1 << 28)
+#define  SMMU_TLB_CONFIG_ACTIVE_LINES(x) ((x) & 0x3f)
+
+#define SMMU_PTC_CONFIG 0x18
+#define  SMMU_PTC_CONFIG_ENABLE (1 << 29)
+#define  SMMU_PTC_CONFIG_REQ_LIMIT(x) (((x) & 0x0f) << 24)
+#define  SMMU_PTC_CONFIG_INDEX_MAP(x) ((x) & 0x3f)
+
+#define SMMU_PTB_ASID 0x01c
+#define  SMMU_PTB_ASID_VALUE(x) ((x) & 0x7f)
+
+#define SMMU_PTB_DATA 0x020
+#define  SMMU_PTB_DATA_VALUE(page, attr) (page_to_phys(page) >> 12 | (attr))
+
+#define SMMU_MK_PDE(page, attr) (page_to_phys(page) >> SMMU_PTE_SHIFT | (attr))
+
+#define SMMU_TLB_FLUSH 0x030
+#define  SMMU_TLB_FLUSH_VA_MATCH_ALL     (0 << 0)
+#define  SMMU_TLB_FLUSH_VA_MATCH_SECTION (2 << 0)
+#define  SMMU_TLB_FLUSH_VA_MATCH_GROUP   (3 << 0)
+#define  SMMU_TLB_FLUSH_ASID(x)          (((x) & 0x7f) << 24)
+#define  SMMU_TLB_FLUSH_VA_SECTION(addr) ((((addr) & 0xffc00000) >> 12) | \
+                                         SMMU_TLB_FLUSH_VA_MATCH_SECTION)
+#define  SMMU_TLB_FLUSH_VA_GROUP(addr)   ((((addr) & 0xffffc000) >> 12) | \
+                                         SMMU_TLB_FLUSH_VA_MATCH_GROUP)
+#define  SMMU_TLB_FLUSH_ASID_MATCH       (1 << 31)
+
+#define SMMU_PTC_FLUSH 0x034
+#define  SMMU_PTC_FLUSH_TYPE_ALL (0 << 0)
+#define  SMMU_PTC_FLUSH_TYPE_ADR (1 << 0)
+
+#define SMMU_PTC_FLUSH_HI 0x9b8
+#define  SMMU_PTC_FLUSH_HI_MASK 0x3
+
+/* per-SWGROUP SMMU_*_ASID register */
+#define SMMU_ASID_ENABLE (1 << 31)
+#define SMMU_ASID_MASK 0x7f
+#define SMMU_ASID_VALUE(x) ((x) & SMMU_ASID_MASK)
+
+/* page table definitions */
+#define SMMU_NUM_PDE 1024
+#define SMMU_NUM_PTE 1024
+
+#define SMMU_SIZE_PD (SMMU_NUM_PDE * 4)
+#define SMMU_SIZE_PT (SMMU_NUM_PTE * 4)
+
+#define SMMU_PDE_SHIFT 22
+#define SMMU_PTE_SHIFT 12
+
+#define SMMU_PD_READABLE       (1 << 31)
+#define SMMU_PD_WRITABLE       (1 << 30)
+#define SMMU_PD_NONSECURE      (1 << 29)
+
+#define SMMU_PDE_READABLE      (1 << 31)
+#define SMMU_PDE_WRITABLE      (1 << 30)
+#define SMMU_PDE_NONSECURE     (1 << 29)
+#define SMMU_PDE_NEXT          (1 << 28)
+
+#define SMMU_PTE_READABLE      (1 << 31)
+#define SMMU_PTE_WRITABLE      (1 << 30)
+#define SMMU_PTE_NONSECURE     (1 << 29)
+
+#define SMMU_PDE_ATTR          (SMMU_PDE_READABLE | SMMU_PDE_WRITABLE | \
+                                SMMU_PDE_NONSECURE)
+#define SMMU_PTE_ATTR          (SMMU_PTE_READABLE | SMMU_PTE_WRITABLE | \
+                                SMMU_PTE_NONSECURE)
+
+static inline void smmu_flush_ptc(struct tegra_smmu *smmu, struct page *page,
+                                 unsigned long offset)
+{
+       phys_addr_t phys = page ? page_to_phys(page) : 0;
+       u32 value;
+
+       if (page) {
+               offset &= ~(smmu->mc->soc->atom_size - 1);
+
+               if (smmu->mc->soc->num_address_bits > 32) {
+#ifdef CONFIG_PHYS_ADDR_T_64BIT
+                       value = (phys >> 32) & SMMU_PTC_FLUSH_HI_MASK;
+#else
+                       value = 0;
+#endif
+                       smmu_writel(smmu, value, SMMU_PTC_FLUSH_HI);
+               }
+
+               value = (phys + offset) | SMMU_PTC_FLUSH_TYPE_ADR;
+       } else {
+               value = SMMU_PTC_FLUSH_TYPE_ALL;
+       }
+
+       smmu_writel(smmu, value, SMMU_PTC_FLUSH);
+}
+
+static inline void smmu_flush_tlb(struct tegra_smmu *smmu)
+{
+       smmu_writel(smmu, SMMU_TLB_FLUSH_VA_MATCH_ALL, SMMU_TLB_FLUSH);
+}
+
+static inline void smmu_flush_tlb_asid(struct tegra_smmu *smmu,
+                                      unsigned long asid)
+{
+       u32 value;
+
+       value = SMMU_TLB_FLUSH_ASID_MATCH | SMMU_TLB_FLUSH_ASID(asid) |
+               SMMU_TLB_FLUSH_VA_MATCH_ALL;
+       smmu_writel(smmu, value, SMMU_TLB_FLUSH);
+}
+
+static inline void smmu_flush_tlb_section(struct tegra_smmu *smmu,
+                                         unsigned long asid,
+                                         unsigned long iova)
+{
+       u32 value;
+
+       value = SMMU_TLB_FLUSH_ASID_MATCH | SMMU_TLB_FLUSH_ASID(asid) |
+               SMMU_TLB_FLUSH_VA_SECTION(iova);
+       smmu_writel(smmu, value, SMMU_TLB_FLUSH);
+}
+
+static inline void smmu_flush_tlb_group(struct tegra_smmu *smmu,
+                                       unsigned long asid,
+                                       unsigned long iova)
+{
+       u32 value;
+
+       value = SMMU_TLB_FLUSH_ASID_MATCH | SMMU_TLB_FLUSH_ASID(asid) |
+               SMMU_TLB_FLUSH_VA_GROUP(iova);
+       smmu_writel(smmu, value, SMMU_TLB_FLUSH);
+}
+
+static inline void smmu_flush(struct tegra_smmu *smmu)
+{
+       smmu_readl(smmu, SMMU_CONFIG);
+}
+
+static int tegra_smmu_alloc_asid(struct tegra_smmu *smmu, unsigned int *idp)
+{
+       unsigned long id;
+
+       mutex_lock(&smmu->lock);
+
+       id = find_first_zero_bit(smmu->asids, smmu->soc->num_asids);
+       if (id >= smmu->soc->num_asids) {
+               mutex_unlock(&smmu->lock);
+               return -ENOSPC;
+       }
+
+       set_bit(id, smmu->asids);
+       *idp = id;
+
+       mutex_unlock(&smmu->lock);
+       return 0;
+}
+
+static void tegra_smmu_free_asid(struct tegra_smmu *smmu, unsigned int id)
+{
+       mutex_lock(&smmu->lock);
+       clear_bit(id, smmu->asids);
+       mutex_unlock(&smmu->lock);
+}
+
+static bool tegra_smmu_capable(enum iommu_cap cap)
+{
+       return false;
+}
+
+static struct iommu_domain *tegra_smmu_domain_alloc(unsigned type)
+{
+       struct tegra_smmu_as *as;
+       unsigned int i;
+       uint32_t *pd;
+
+       if (type != IOMMU_DOMAIN_UNMANAGED)
+               return NULL;
+
+       as = kzalloc(sizeof(*as), GFP_KERNEL);
+       if (!as)
+               return NULL;
+
+       as->attr = SMMU_PD_READABLE | SMMU_PD_WRITABLE | SMMU_PD_NONSECURE;
+
+       as->pd = alloc_page(GFP_KERNEL | __GFP_DMA);
+       if (!as->pd) {
+               kfree(as);
+               return NULL;
+       }
+
+       as->count = alloc_page(GFP_KERNEL);
+       if (!as->count) {
+               __free_page(as->pd);
+               kfree(as);
+               return NULL;
+       }
+
+       /* clear PDEs */
+       pd = page_address(as->pd);
+       SetPageReserved(as->pd);
+
+       for (i = 0; i < SMMU_NUM_PDE; i++)
+               pd[i] = 0;
+
+       /* clear PDE usage counters */
+       pd = page_address(as->count);
+       SetPageReserved(as->count);
+
+       for (i = 0; i < SMMU_NUM_PDE; i++)
+               pd[i] = 0;
+
+       /* setup aperture */
+       as->domain.geometry.aperture_start = 0;
+       as->domain.geometry.aperture_end = 0xffffffff;
+       as->domain.geometry.force_aperture = true;
+
+       return &as->domain;
+}
+
+static void tegra_smmu_domain_free(struct iommu_domain *domain)
+{
+       struct tegra_smmu_as *as = to_smmu_as(domain);
+
+       /* TODO: free page directory and page tables */
+       ClearPageReserved(as->pd);
+
+       kfree(as);
+}
+
+static const struct tegra_smmu_swgroup *
+tegra_smmu_find_swgroup(struct tegra_smmu *smmu, unsigned int swgroup)
+{
+       const struct tegra_smmu_swgroup *group = NULL;
+       unsigned int i;
+
+       for (i = 0; i < smmu->soc->num_swgroups; i++) {
+               if (smmu->soc->swgroups[i].swgroup == swgroup) {
+                       group = &smmu->soc->swgroups[i];
+                       break;
+               }
+       }
+
+       return group;
+}
+
+static void tegra_smmu_enable(struct tegra_smmu *smmu, unsigned int swgroup,
+                             unsigned int asid)
+{
+       const struct tegra_smmu_swgroup *group;
+       unsigned int i;
+       u32 value;
+
+       for (i = 0; i < smmu->soc->num_clients; i++) {
+               const struct tegra_mc_client *client = &smmu->soc->clients[i];
+
+               if (client->swgroup != swgroup)
+                       continue;
+
+               value = smmu_readl(smmu, client->smmu.reg);
+               value |= BIT(client->smmu.bit);
+               smmu_writel(smmu, value, client->smmu.reg);
+       }
+
+       group = tegra_smmu_find_swgroup(smmu, swgroup);
+       if (group) {
+               value = smmu_readl(smmu, group->reg);
+               value &= ~SMMU_ASID_MASK;
+               value |= SMMU_ASID_VALUE(asid);
+               value |= SMMU_ASID_ENABLE;
+               smmu_writel(smmu, value, group->reg);
+       }
+}
+
+static void tegra_smmu_disable(struct tegra_smmu *smmu, unsigned int swgroup,
+                              unsigned int asid)
+{
+       const struct tegra_smmu_swgroup *group;
+       unsigned int i;
+       u32 value;
+
+       group = tegra_smmu_find_swgroup(smmu, swgroup);
+       if (group) {
+               value = smmu_readl(smmu, group->reg);
+               value &= ~SMMU_ASID_MASK;
+               value |= SMMU_ASID_VALUE(asid);
+               value &= ~SMMU_ASID_ENABLE;
+               smmu_writel(smmu, value, group->reg);
+       }
+
+       for (i = 0; i < smmu->soc->num_clients; i++) {
+               const struct tegra_mc_client *client = &smmu->soc->clients[i];
+
+               if (client->swgroup != swgroup)
+                       continue;
+
+               value = smmu_readl(smmu, client->smmu.reg);
+               value &= ~BIT(client->smmu.bit);
+               smmu_writel(smmu, value, client->smmu.reg);
+       }
+}
+
+static int tegra_smmu_as_prepare(struct tegra_smmu *smmu,
+                                struct tegra_smmu_as *as)
+{
+       u32 value;
+       int err;
+
+       if (as->use_count > 0) {
+               as->use_count++;
+               return 0;
+       }
+
+       err = tegra_smmu_alloc_asid(smmu, &as->id);
+       if (err < 0)
+               return err;
+
+       smmu->soc->ops->flush_dcache(as->pd, 0, SMMU_SIZE_PD);
+       smmu_flush_ptc(smmu, as->pd, 0);
+       smmu_flush_tlb_asid(smmu, as->id);
+
+       smmu_writel(smmu, as->id & 0x7f, SMMU_PTB_ASID);
+       value = SMMU_PTB_DATA_VALUE(as->pd, as->attr);
+       smmu_writel(smmu, value, SMMU_PTB_DATA);
+       smmu_flush(smmu);
+
+       as->smmu = smmu;
+       as->use_count++;
+
+       return 0;
+}
+
+static void tegra_smmu_as_unprepare(struct tegra_smmu *smmu,
+                                   struct tegra_smmu_as *as)
+{
+       if (--as->use_count > 0)
+               return;
+
+       tegra_smmu_free_asid(smmu, as->id);
+       as->smmu = NULL;
+}
+
+static int tegra_smmu_attach_dev(struct iommu_domain *domain,
+                                struct device *dev)
+{
+       struct tegra_smmu *smmu = dev->archdata.iommu;
+       struct tegra_smmu_as *as = to_smmu_as(domain);
+       struct device_node *np = dev->of_node;
+       struct of_phandle_args args;
+       unsigned int index = 0;
+       int err = 0;
+
+       while (!of_parse_phandle_with_args(np, "iommus", "#iommu-cells", index,
+                                          &args)) {
+               unsigned int swgroup = args.args[0];
+
+               if (args.np != smmu->dev->of_node) {
+                       of_node_put(args.np);
+                       continue;
+               }
+
+               of_node_put(args.np);
+
+               err = tegra_smmu_as_prepare(smmu, as);
+               if (err < 0)
+                       return err;
+
+               tegra_smmu_enable(smmu, swgroup, as->id);
+               index++;
+       }
+
+       if (index == 0)
+               return -ENODEV;
+
+       return 0;
+}
+
+static void tegra_smmu_detach_dev(struct iommu_domain *domain, struct device *dev)
+{
+       struct tegra_smmu_as *as = to_smmu_as(domain);
+       struct device_node *np = dev->of_node;
+       struct tegra_smmu *smmu = as->smmu;
+       struct of_phandle_args args;
+       unsigned int index = 0;
+
+       while (!of_parse_phandle_with_args(np, "iommus", "#iommu-cells", index,
+                                          &args)) {
+               unsigned int swgroup = args.args[0];
+
+               if (args.np != smmu->dev->of_node) {
+                       of_node_put(args.np);
+                       continue;
+               }
+
+               of_node_put(args.np);
+
+               tegra_smmu_disable(smmu, swgroup, as->id);
+               tegra_smmu_as_unprepare(smmu, as);
+               index++;
+       }
+}
+
+static u32 *as_get_pte(struct tegra_smmu_as *as, dma_addr_t iova,
+                      struct page **pagep)
+{
+       u32 *pd = page_address(as->pd), *pt, *count;
+       u32 pde = (iova >> SMMU_PDE_SHIFT) & 0x3ff;
+       u32 pte = (iova >> SMMU_PTE_SHIFT) & 0x3ff;
+       struct tegra_smmu *smmu = as->smmu;
+       struct page *page;
+       unsigned int i;
+
+       if (pd[pde] == 0) {
+               page = alloc_page(GFP_KERNEL | __GFP_DMA);
+               if (!page)
+                       return NULL;
+
+               pt = page_address(page);
+               SetPageReserved(page);
+
+               for (i = 0; i < SMMU_NUM_PTE; i++)
+                       pt[i] = 0;
+
+               smmu->soc->ops->flush_dcache(page, 0, SMMU_SIZE_PT);
+
+               pd[pde] = SMMU_MK_PDE(page, SMMU_PDE_ATTR | SMMU_PDE_NEXT);
+
+               smmu->soc->ops->flush_dcache(as->pd, pde << 2, 4);
+               smmu_flush_ptc(smmu, as->pd, pde << 2);
+               smmu_flush_tlb_section(smmu, as->id, iova);
+               smmu_flush(smmu);
+       } else {
+               page = pfn_to_page(pd[pde] & smmu->pfn_mask);
+               pt = page_address(page);
+       }
+
+       *pagep = page;
+
+       /* Keep track of entries in this page table. */
+       count = page_address(as->count);
+       if (pt[pte] == 0)
+               count[pde]++;
+
+       return &pt[pte];
+}
+
+static void as_put_pte(struct tegra_smmu_as *as, dma_addr_t iova)
+{
+       u32 pde = (iova >> SMMU_PDE_SHIFT) & 0x3ff;
+       u32 pte = (iova >> SMMU_PTE_SHIFT) & 0x3ff;
+       u32 *count = page_address(as->count);
+       u32 *pd = page_address(as->pd), *pt;
+       struct page *page;
+
+       page = pfn_to_page(pd[pde] & as->smmu->pfn_mask);
+       pt = page_address(page);
+
+       /*
+        * When no entries in this page table are used anymore, return the
+        * memory page to the system.
+        */
+       if (pt[pte] != 0) {
+               if (--count[pde] == 0) {
+                       ClearPageReserved(page);
+                       __free_page(page);
+                       pd[pde] = 0;
+               }
+
+               pt[pte] = 0;
+       }
+}
+
+static int tegra_smmu_map(struct iommu_domain *domain, unsigned long iova,
+                         phys_addr_t paddr, size_t size, int prot)
+{
+       struct tegra_smmu_as *as = to_smmu_as(domain);
+       struct tegra_smmu *smmu = as->smmu;
+       unsigned long offset;
+       struct page *page;
+       u32 *pte;
+
+       pte = as_get_pte(as, iova, &page);
+       if (!pte)
+               return -ENOMEM;
+
+       *pte = __phys_to_pfn(paddr) | SMMU_PTE_ATTR;
+       offset = offset_in_page(pte);
+
+       smmu->soc->ops->flush_dcache(page, offset, 4);
+       smmu_flush_ptc(smmu, page, offset);
+       smmu_flush_tlb_group(smmu, as->id, iova);
+       smmu_flush(smmu);
+
+       return 0;
+}
+
+static size_t tegra_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
+                              size_t size)
+{
+       struct tegra_smmu_as *as = to_smmu_as(domain);
+       struct tegra_smmu *smmu = as->smmu;
+       unsigned long offset;
+       struct page *page;
+       u32 *pte;
+
+       pte = as_get_pte(as, iova, &page);
+       if (!pte)
+               return 0;
+
+       offset = offset_in_page(pte);
+       as_put_pte(as, iova);
+
+       smmu->soc->ops->flush_dcache(page, offset, 4);
+       smmu_flush_ptc(smmu, page, offset);
+       smmu_flush_tlb_group(smmu, as->id, iova);
+       smmu_flush(smmu);
+
+       return size;
+}
+
+static phys_addr_t tegra_smmu_iova_to_phys(struct iommu_domain *domain,
+                                          dma_addr_t iova)
+{
+       struct tegra_smmu_as *as = to_smmu_as(domain);
+       struct page *page;
+       unsigned long pfn;
+       u32 *pte;
+
+       pte = as_get_pte(as, iova, &page);
+       pfn = *pte & as->smmu->pfn_mask;
+
+       return PFN_PHYS(pfn);
+}
+
+static struct tegra_smmu *tegra_smmu_find(struct device_node *np)
+{
+       struct platform_device *pdev;
+       struct tegra_mc *mc;
+
+       pdev = of_find_device_by_node(np);
+       if (!pdev)
+               return NULL;
+
+       mc = platform_get_drvdata(pdev);
+       if (!mc)
+               return NULL;
+
+       return mc->smmu;
+}
+
+static int tegra_smmu_add_device(struct device *dev)
+{
+       struct device_node *np = dev->of_node;
+       struct of_phandle_args args;
+       unsigned int index = 0;
+
+       while (of_parse_phandle_with_args(np, "iommus", "#iommu-cells", index,
+                                         &args) == 0) {
+               struct tegra_smmu *smmu;
+
+               smmu = tegra_smmu_find(args.np);
+               if (smmu) {
+                       /*
+                        * Only a single IOMMU master interface is currently
+                        * supported by the Linux kernel, so abort after the
+                        * first match.
+                        */
+                       dev->archdata.iommu = smmu;
+                       break;
+               }
+
+               index++;
+       }
+
+       return 0;
+}
+
+static void tegra_smmu_remove_device(struct device *dev)
+{
+       dev->archdata.iommu = NULL;
+}
+
+static const struct iommu_ops tegra_smmu_ops = {
+       .capable = tegra_smmu_capable,
+       .domain_alloc = tegra_smmu_domain_alloc,
+       .domain_free = tegra_smmu_domain_free,
+       .attach_dev = tegra_smmu_attach_dev,
+       .detach_dev = tegra_smmu_detach_dev,
+       .add_device = tegra_smmu_add_device,
+       .remove_device = tegra_smmu_remove_device,
+       .map = tegra_smmu_map,
+       .unmap = tegra_smmu_unmap,
+       .map_sg = default_iommu_map_sg,
+       .iova_to_phys = tegra_smmu_iova_to_phys,
+
+       .pgsize_bitmap = SZ_4K,
+};
+
+static void tegra_smmu_ahb_enable(void)
+{
+       static const struct of_device_id ahb_match[] = {
+               { .compatible = "nvidia,tegra30-ahb", },
+               { }
+       };
+       struct device_node *ahb;
+
+       ahb = of_find_matching_node(NULL, ahb_match);
+       if (ahb) {
+               tegra_ahb_enable_smmu(ahb);
+               of_node_put(ahb);
+       }
+}
+
+struct tegra_smmu *tegra_smmu_probe(struct device *dev,
+                                   const struct tegra_smmu_soc *soc,
+                                   struct tegra_mc *mc)
+{
+       struct tegra_smmu *smmu;
+       size_t size;
+       u32 value;
+       int err;
+
+       /* This can happen on Tegra20 which doesn't have an SMMU */
+       if (!soc)
+               return NULL;
+
+       smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
+       if (!smmu)
+               return ERR_PTR(-ENOMEM);
+
+       /*
+        * This is a bit of a hack. Ideally we'd want to simply return this
+        * value. However the IOMMU registration process will attempt to add
+        * all devices to the IOMMU when bus_set_iommu() is called. In order
+        * not to rely on global variables to track the IOMMU instance, we
+        * set it here so that it can be looked up from the .add_device()
+        * callback via the IOMMU device's .drvdata field.
+        */
+       mc->smmu = smmu;
+
+       size = BITS_TO_LONGS(soc->num_asids) * sizeof(long);
+
+       smmu->asids = devm_kzalloc(dev, size, GFP_KERNEL);
+       if (!smmu->asids)
+               return ERR_PTR(-ENOMEM);
+
+       mutex_init(&smmu->lock);
+
+       smmu->regs = mc->regs;
+       smmu->soc = soc;
+       smmu->dev = dev;
+       smmu->mc = mc;
+
+       smmu->pfn_mask = BIT_MASK(mc->soc->num_address_bits - PAGE_SHIFT) - 1;
+       dev_dbg(dev, "address bits: %u, PFN mask: %#lx\n",
+               mc->soc->num_address_bits, smmu->pfn_mask);
+
+       value = SMMU_PTC_CONFIG_ENABLE | SMMU_PTC_CONFIG_INDEX_MAP(0x3f);
+
+       if (soc->supports_request_limit)
+               value |= SMMU_PTC_CONFIG_REQ_LIMIT(8);
+
+       smmu_writel(smmu, value, SMMU_PTC_CONFIG);
+
+       value = SMMU_TLB_CONFIG_HIT_UNDER_MISS |
+               SMMU_TLB_CONFIG_ACTIVE_LINES(0x20);
+
+       if (soc->supports_round_robin_arbitration)
+               value |= SMMU_TLB_CONFIG_ROUND_ROBIN_ARBITRATION;
+
+       smmu_writel(smmu, value, SMMU_TLB_CONFIG);
+
+       smmu_flush_ptc(smmu, NULL, 0);
+       smmu_flush_tlb(smmu);
+       smmu_writel(smmu, SMMU_CONFIG_ENABLE, SMMU_CONFIG);
+       smmu_flush(smmu);
+
+       tegra_smmu_ahb_enable();
+
+       err = bus_set_iommu(&platform_bus_type, &tegra_smmu_ops);
+       if (err < 0)
+               return ERR_PTR(err);
+
+       return smmu;
+}