ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver
authorHiroshi DOYU <hdoyu@nvidia.com>
Thu, 17 Nov 2011 05:31:31 +0000 (07:31 +0200)
committerJoerg Roedel <joerg.roedel@amd.com>
Thu, 26 Jan 2012 12:59:02 +0000 (13:59 +0100)
Tegra 30 IOMMU H/W, SMMU (System Memory Management Unit). This patch
implements struct iommu_ops for SMMU for the upper IOMMU API.

This H/W module supports multiple virtual address spaces(domain x4),
and manages 2 level H/W translation pagetable.

Signed-off-by: Hiroshi DOYU <hdoyu@nvidia.com>
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
arch/arm/mach-tegra/include/mach/smmu.h [new file with mode: 0644]
drivers/iommu/Kconfig
drivers/iommu/Makefile
drivers/iommu/tegra-smmu.c [new file with mode: 0644]

diff --git a/arch/arm/mach-tegra/include/mach/smmu.h b/arch/arm/mach-tegra/include/mach/smmu.h
new file mode 100644 (file)
index 0000000..dad403a
--- /dev/null
@@ -0,0 +1,63 @@
+/*
+ * IOMMU API for SMMU in Tegra30
+ *
+ * Copyright (c) 2012, NVIDIA CORPORATION.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope 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.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#ifndef        MACH_SMMU_H
+#define        MACH_SMMU_H
+
+enum smmu_hwgrp {
+       HWGRP_AFI,
+       HWGRP_AVPC,
+       HWGRP_DC,
+       HWGRP_DCB,
+       HWGRP_EPP,
+       HWGRP_G2,
+       HWGRP_HC,
+       HWGRP_HDA,
+       HWGRP_ISP,
+       HWGRP_MPE,
+       HWGRP_NV,
+       HWGRP_NV2,
+       HWGRP_PPCS,
+       HWGRP_SATA,
+       HWGRP_VDE,
+       HWGRP_VI,
+
+       HWGRP_COUNT,
+
+       HWGRP_END = ~0,
+};
+
+#define HWG_AFI                (1 << HWGRP_AFI)
+#define HWG_AVPC       (1 << HWGRP_AVPC)
+#define HWG_DC         (1 << HWGRP_DC)
+#define HWG_DCB                (1 << HWGRP_DCB)
+#define HWG_EPP                (1 << HWGRP_EPP)
+#define HWG_G2         (1 << HWGRP_G2)
+#define HWG_HC         (1 << HWGRP_HC)
+#define HWG_HDA                (1 << HWGRP_HDA)
+#define HWG_ISP                (1 << HWGRP_ISP)
+#define HWG_MPE                (1 << HWGRP_MPE)
+#define HWG_NV         (1 << HWGRP_NV)
+#define HWG_NV2                (1 << HWGRP_NV2)
+#define HWG_PPCS       (1 << HWGRP_PPCS)
+#define HWG_SATA       (1 << HWGRP_SATA)
+#define HWG_VDE                (1 << HWGRP_VDE)
+#define HWG_VI         (1 << HWGRP_VI)
+
+#endif /* MACH_SMMU_H */
index 76c86da..3bd9fff 100644 (file)
@@ -152,4 +152,14 @@ config TEGRA_IOMMU_GART
          space through the GART (Graphics Address Relocation Table)
          hardware included on Tegra SoCs.
 
+config TEGRA_IOMMU_SMMU
+       bool "Tegra SMMU IOMMU Support"
+       depends on ARCH_TEGRA_3x_SOC
+       select IOMMU_API
+       help
+         Enables support for remapping discontiguous physical memory
+         shared with the operating system into contiguous I/O virtual
+         space through the SMMU (System Memory Management Unit)
+         hardware included on Tegra SoCs.
+
 endif # IOMMU_SUPPORT
index 3238a31..7ad7a3b 100644 (file)
@@ -9,3 +9,4 @@ obj-$(CONFIG_OMAP_IOMMU) += omap-iommu.o
 obj-$(CONFIG_OMAP_IOVMM) += omap-iovmm.o
 obj-$(CONFIG_OMAP_IOMMU_DEBUG) += omap-iommu-debug.o
 obj-$(CONFIG_TEGRA_IOMMU_GART) += tegra-gart.o
+obj-$(CONFIG_TEGRA_IOMMU_SMMU) += tegra-smmu.o
diff --git a/drivers/iommu/tegra-smmu.c b/drivers/iommu/tegra-smmu.c
new file mode 100644 (file)
index 0000000..eb93c82
--- /dev/null
@@ -0,0 +1,1034 @@
+/*
+ * IOMMU API for SMMU in Tegra30
+ *
+ * Copyright (c) 2011-2012, NVIDIA CORPORATION.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope 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.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#define pr_fmt(fmt)    "%s(): " fmt, __func__
+
+#include <linux/module.h>
+#include <linux/platform_device.h>
+#include <linux/spinlock.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/mm.h>
+#include <linux/pagemap.h>
+#include <linux/device.h>
+#include <linux/sched.h>
+#include <linux/iommu.h>
+#include <linux/io.h>
+
+#include <asm/page.h>
+#include <asm/cacheflush.h>
+
+#include <mach/iomap.h>
+#include <mach/smmu.h>
+
+/* bitmap of the page sizes currently supported */
+#define SMMU_IOMMU_PGSIZES     (SZ_4K)
+
+#define SMMU_CONFIG                            0x10
+#define SMMU_CONFIG_DISABLE                    0
+#define SMMU_CONFIG_ENABLE                     1
+
+#define SMMU_TLB_CONFIG                                0x14
+#define SMMU_TLB_CONFIG_STATS__MASK            (1 << 31)
+#define SMMU_TLB_CONFIG_STATS__ENABLE          (1 << 31)
+#define SMMU_TLB_CONFIG_HIT_UNDER_MISS__ENABLE (1 << 29)
+#define SMMU_TLB_CONFIG_ACTIVE_LINES__VALUE    0x10
+#define SMMU_TLB_CONFIG_RESET_VAL              0x20000010
+
+#define SMMU_PTC_CONFIG                                0x18
+#define SMMU_PTC_CONFIG_STATS__MASK            (1 << 31)
+#define SMMU_PTC_CONFIG_STATS__ENABLE          (1 << 31)
+#define SMMU_PTC_CONFIG_CACHE__ENABLE          (1 << 29)
+#define SMMU_PTC_CONFIG_INDEX_MAP__PATTERN     0x3f
+#define SMMU_PTC_CONFIG_RESET_VAL              0x2000003f
+
+#define SMMU_PTB_ASID                          0x1c
+#define SMMU_PTB_ASID_CURRENT_SHIFT            0
+
+#define SMMU_PTB_DATA                          0x20
+#define SMMU_PTB_DATA_RESET_VAL                        0
+#define SMMU_PTB_DATA_ASID_NONSECURE_SHIFT     29
+#define SMMU_PTB_DATA_ASID_WRITABLE_SHIFT      30
+#define SMMU_PTB_DATA_ASID_READABLE_SHIFT      31
+
+#define SMMU_TLB_FLUSH                         0x30
+#define SMMU_TLB_FLUSH_VA_MATCH_ALL            0
+#define SMMU_TLB_FLUSH_VA_MATCH_SECTION                2
+#define SMMU_TLB_FLUSH_VA_MATCH_GROUP          3
+#define SMMU_TLB_FLUSH_ASID_SHIFT              29
+#define SMMU_TLB_FLUSH_ASID_MATCH_DISABLE      0
+#define SMMU_TLB_FLUSH_ASID_MATCH_ENABLE       1
+#define SMMU_TLB_FLUSH_ASID_MATCH_SHIFT                31
+
+#define SMMU_PTC_FLUSH                         0x34
+#define SMMU_PTC_FLUSH_TYPE_ALL                        0
+#define SMMU_PTC_FLUSH_TYPE_ADR                        1
+#define SMMU_PTC_FLUSH_ADR_SHIFT               4
+
+#define SMMU_ASID_SECURITY                     0x38
+
+#define SMMU_STATS_TLB_HIT_COUNT               0x1f0
+#define SMMU_STATS_TLB_MISS_COUNT              0x1f4
+#define SMMU_STATS_PTC_HIT_COUNT               0x1f8
+#define SMMU_STATS_PTC_MISS_COUNT              0x1fc
+
+#define SMMU_TRANSLATION_ENABLE_0              0x228
+#define SMMU_TRANSLATION_ENABLE_1              0x22c
+#define SMMU_TRANSLATION_ENABLE_2              0x230
+
+#define SMMU_AFI_ASID  0x238   /* PCIE */
+#define SMMU_AVPC_ASID 0x23c   /* AVP */
+#define SMMU_DC_ASID   0x240   /* Display controller */
+#define SMMU_DCB_ASID  0x244   /* Display controller B */
+#define SMMU_EPP_ASID  0x248   /* Encoder pre-processor */
+#define SMMU_G2_ASID   0x24c   /* 2D engine */
+#define SMMU_HC_ASID   0x250   /* Host1x */
+#define SMMU_HDA_ASID  0x254   /* High-def audio */
+#define SMMU_ISP_ASID  0x258   /* Image signal processor */
+#define SMMU_MPE_ASID  0x264   /* MPEG encoder */
+#define SMMU_NV_ASID   0x268   /* (3D) */
+#define SMMU_NV2_ASID  0x26c   /* (3D) */
+#define SMMU_PPCS_ASID 0x270   /* AHB */
+#define SMMU_SATA_ASID 0x278   /* SATA */
+#define SMMU_VDE_ASID  0x27c   /* Video decoder */
+#define SMMU_VI_ASID   0x280   /* Video input */
+
+#define SMMU_PDE_NEXT_SHIFT            28
+
+/* AHB Arbiter Registers */
+#define AHB_XBAR_CTRL                          0xe0
+#define AHB_XBAR_CTRL_SMMU_INIT_DONE_DONE      1
+#define AHB_XBAR_CTRL_SMMU_INIT_DONE_SHIFT     17
+
+#define SMMU_NUM_ASIDS                         4
+#define SMMU_TLB_FLUSH_VA_SECTION__MASK                0xffc00000
+#define SMMU_TLB_FLUSH_VA_SECTION__SHIFT       12 /* right shift */
+#define SMMU_TLB_FLUSH_VA_GROUP__MASK          0xffffc000
+#define SMMU_TLB_FLUSH_VA_GROUP__SHIFT         12 /* right shift */
+#define SMMU_TLB_FLUSH_VA(iova, which) \
+       ((((iova) & SMMU_TLB_FLUSH_VA_##which##__MASK) >> \
+               SMMU_TLB_FLUSH_VA_##which##__SHIFT) |   \
+       SMMU_TLB_FLUSH_VA_MATCH_##which)
+#define SMMU_PTB_ASID_CUR(n)   \
+               ((n) << SMMU_PTB_ASID_CURRENT_SHIFT)
+#define SMMU_TLB_FLUSH_ASID_MATCH_disable              \
+               (SMMU_TLB_FLUSH_ASID_MATCH_DISABLE <<   \
+                       SMMU_TLB_FLUSH_ASID_MATCH_SHIFT)
+#define SMMU_TLB_FLUSH_ASID_MATCH__ENABLE              \
+               (SMMU_TLB_FLUSH_ASID_MATCH_ENABLE <<    \
+                       SMMU_TLB_FLUSH_ASID_MATCH_SHIFT)
+
+#define SMMU_PAGE_SHIFT 12
+#define SMMU_PAGE_SIZE (1 << SMMU_PAGE_SHIFT)
+
+#define SMMU_PDIR_COUNT        1024
+#define SMMU_PDIR_SIZE (sizeof(unsigned long) * SMMU_PDIR_COUNT)
+#define SMMU_PTBL_COUNT        1024
+#define SMMU_PTBL_SIZE (sizeof(unsigned long) * SMMU_PTBL_COUNT)
+#define SMMU_PDIR_SHIFT        12
+#define SMMU_PDE_SHIFT 12
+#define SMMU_PTE_SHIFT 12
+#define SMMU_PFN_MASK  0x000fffff
+
+#define SMMU_ADDR_TO_PFN(addr) ((addr) >> 12)
+#define SMMU_ADDR_TO_PDN(addr) ((addr) >> 22)
+#define SMMU_PDN_TO_ADDR(addr) ((pdn) << 22)
+
+#define _READABLE      (1 << SMMU_PTB_DATA_ASID_READABLE_SHIFT)
+#define _WRITABLE      (1 << SMMU_PTB_DATA_ASID_WRITABLE_SHIFT)
+#define _NONSECURE     (1 << SMMU_PTB_DATA_ASID_NONSECURE_SHIFT)
+#define _PDE_NEXT      (1 << SMMU_PDE_NEXT_SHIFT)
+#define _MASK_ATTR     (_READABLE | _WRITABLE | _NONSECURE)
+
+#define _PDIR_ATTR     (_READABLE | _WRITABLE | _NONSECURE)
+
+#define _PDE_ATTR      (_READABLE | _WRITABLE | _NONSECURE)
+#define _PDE_ATTR_N    (_PDE_ATTR | _PDE_NEXT)
+#define _PDE_VACANT(pdn)       (((pdn) << 10) | _PDE_ATTR)
+
+#define _PTE_ATTR      (_READABLE | _WRITABLE | _NONSECURE)
+#define _PTE_VACANT(addr)      (((addr) >> SMMU_PAGE_SHIFT) | _PTE_ATTR)
+
+#define SMMU_MK_PDIR(page, attr)       \
+               ((page_to_phys(page) >> SMMU_PDIR_SHIFT) | (attr))
+#define SMMU_MK_PDE(page, attr)                \
+               (unsigned long)((page_to_phys(page) >> SMMU_PDE_SHIFT) | (attr))
+#define SMMU_EX_PTBL_PAGE(pde)         \
+               pfn_to_page((unsigned long)(pde) & SMMU_PFN_MASK)
+#define SMMU_PFN_TO_PTE(pfn, attr)     (unsigned long)((pfn) | (attr))
+
+#define SMMU_ASID_ENABLE(asid) ((asid) | (1 << 31))
+#define SMMU_ASID_DISABLE      0
+#define SMMU_ASID_ASID(n)      ((n) & ~SMMU_ASID_ENABLE(0))
+
+#define smmu_client_enable_hwgrp(c, m) smmu_client_set_hwgrp(c, m, 1)
+#define smmu_client_disable_hwgrp(c)   smmu_client_set_hwgrp(c, 0, 0)
+#define __smmu_client_enable_hwgrp(c, m) __smmu_client_set_hwgrp(c, m, 1)
+#define __smmu_client_disable_hwgrp(c) __smmu_client_set_hwgrp(c, 0, 0)
+
+#define HWGRP_INIT(client) [HWGRP_##client] = SMMU_##client##_ASID
+
+static const u32 smmu_hwgrp_asid_reg[] = {
+       HWGRP_INIT(AFI),
+       HWGRP_INIT(AVPC),
+       HWGRP_INIT(DC),
+       HWGRP_INIT(DCB),
+       HWGRP_INIT(EPP),
+       HWGRP_INIT(G2),
+       HWGRP_INIT(HC),
+       HWGRP_INIT(HDA),
+       HWGRP_INIT(ISP),
+       HWGRP_INIT(MPE),
+       HWGRP_INIT(NV),
+       HWGRP_INIT(NV2),
+       HWGRP_INIT(PPCS),
+       HWGRP_INIT(SATA),
+       HWGRP_INIT(VDE),
+       HWGRP_INIT(VI),
+};
+#define HWGRP_ASID_REG(x) (smmu_hwgrp_asid_reg[x])
+
+/*
+ * Per client for address space
+ */
+struct smmu_client {
+       struct device           *dev;
+       struct list_head        list;
+       struct smmu_as          *as;
+       u32                     hwgrp;
+};
+
+/*
+ * Per address space
+ */
+struct smmu_as {
+       struct smmu_device      *smmu;  /* back pointer to container */
+       unsigned int            asid;
+       spinlock_t              lock;   /* for pagetable */
+       struct page             *pdir_page;
+       unsigned long           pdir_attr;
+       unsigned long           pde_attr;
+       unsigned long           pte_attr;
+       unsigned int            *pte_count;
+
+       struct list_head        client;
+       spinlock_t              client_lock; /* for client list */
+};
+
+/*
+ * Per SMMU device - IOMMU device
+ */
+struct smmu_device {
+       void __iomem    *regs, *regs_ahbarb;
+       unsigned long   iovmm_base;     /* remappable base address */
+       unsigned long   page_count;     /* total remappable size */
+       spinlock_t      lock;
+       char            *name;
+       struct device   *dev;
+       int             num_as;
+       struct smmu_as  *as;            /* Run-time allocated array */
+       struct page *avp_vector_page;   /* dummy page shared by all AS's */
+
+       /*
+        * Register image savers for suspend/resume
+        */
+       unsigned long translation_enable_0;
+       unsigned long translation_enable_1;
+       unsigned long translation_enable_2;
+       unsigned long asid_security;
+};
+
+static struct smmu_device *smmu_handle; /* unique for a system */
+
+/*
+ *     SMMU/AHB register accessors
+ */
+static inline u32 smmu_read(struct smmu_device *smmu, size_t offs)
+{
+       return readl(smmu->regs + offs);
+}
+static inline void smmu_write(struct smmu_device *smmu, u32 val, size_t offs)
+{
+       writel(val, smmu->regs + offs);
+}
+
+static inline u32 ahb_read(struct smmu_device *smmu, size_t offs)
+{
+       return readl(smmu->regs_ahbarb + offs);
+}
+static inline void ahb_write(struct smmu_device *smmu, u32 val, size_t offs)
+{
+       writel(val, smmu->regs_ahbarb + offs);
+}
+
+#define VA_PAGE_TO_PA(va, page)        \
+       (page_to_phys(page) + ((unsigned long)(va) & ~PAGE_MASK))
+
+#define FLUSH_CPU_DCACHE(va, page, size)       \
+       do {    \
+               unsigned long _pa_ = VA_PAGE_TO_PA(va, page);           \
+               __cpuc_flush_dcache_area((void *)(va), (size_t)(size)); \
+               outer_flush_range(_pa_, _pa_+(size_t)(size));           \
+       } while (0)
+
+/*
+ * Any interaction between any block on PPSB and a block on APB or AHB
+ * must have these read-back barriers to ensure the APB/AHB bus
+ * transaction is complete before initiating activity on the PPSB
+ * block.
+ */
+#define FLUSH_SMMU_REGS(smmu)  smmu_read(smmu, SMMU_CONFIG)
+
+#define smmu_client_hwgrp(c) (u32)((c)->dev->platform_data)
+
+static int __smmu_client_set_hwgrp(struct smmu_client *c,
+                                  unsigned long map, int on)
+{
+       int i;
+       struct smmu_as *as = c->as;
+       u32 val, offs, mask = SMMU_ASID_ENABLE(as->asid);
+       struct smmu_device *smmu = as->smmu;
+
+       WARN_ON(!on && map);
+       if (on && !map)
+               return -EINVAL;
+       if (!on)
+               map = smmu_client_hwgrp(c);
+
+       for_each_set_bit(i, &map, HWGRP_COUNT) {
+               offs = HWGRP_ASID_REG(i);
+               val = smmu_read(smmu, offs);
+               if (on) {
+                       if (WARN_ON(val & mask))
+                               goto err_hw_busy;
+                       val |= mask;
+               } else {
+                       WARN_ON((val & mask) == mask);
+                       val &= ~mask;
+               }
+               smmu_write(smmu, val, offs);
+       }
+       FLUSH_SMMU_REGS(smmu);
+       c->hwgrp = map;
+       return 0;
+
+err_hw_busy:
+       for_each_set_bit(i, &map, HWGRP_COUNT) {
+               offs = HWGRP_ASID_REG(i);
+               val = smmu_read(smmu, offs);
+               val &= ~mask;
+               smmu_write(smmu, val, offs);
+       }
+       return -EBUSY;
+}
+
+static int smmu_client_set_hwgrp(struct smmu_client *c, u32 map, int on)
+{
+       u32 val;
+       unsigned long flags;
+       struct smmu_as *as = c->as;
+       struct smmu_device *smmu = as->smmu;
+
+       spin_lock_irqsave(&smmu->lock, flags);
+       val = __smmu_client_set_hwgrp(c, map, on);
+       spin_unlock_irqrestore(&smmu->lock, flags);
+       return val;
+}
+
+/*
+ * Flush all TLB entries and all PTC entries
+ * Caller must lock smmu
+ */
+static void smmu_flush_regs(struct smmu_device *smmu, int enable)
+{
+       u32 val;
+
+       smmu_write(smmu, SMMU_PTC_FLUSH_TYPE_ALL, SMMU_PTC_FLUSH);
+       FLUSH_SMMU_REGS(smmu);
+       val = SMMU_TLB_FLUSH_VA_MATCH_ALL |
+               SMMU_TLB_FLUSH_ASID_MATCH_disable;
+       smmu_write(smmu, val, SMMU_TLB_FLUSH);
+
+       if (enable)
+               smmu_write(smmu, SMMU_CONFIG_ENABLE, SMMU_CONFIG);
+       FLUSH_SMMU_REGS(smmu);
+}
+
+static void smmu_setup_regs(struct smmu_device *smmu)
+{
+       int i;
+       u32 val;
+
+       for (i = 0; i < smmu->num_as; i++) {
+               struct smmu_as *as = &smmu->as[i];
+               struct smmu_client *c;
+
+               smmu_write(smmu, SMMU_PTB_ASID_CUR(as->asid), SMMU_PTB_ASID);
+               val = as->pdir_page ?
+                       SMMU_MK_PDIR(as->pdir_page, as->pdir_attr) :
+                       SMMU_PTB_DATA_RESET_VAL;
+               smmu_write(smmu, val, SMMU_PTB_DATA);
+
+               list_for_each_entry(c, &as->client, list)
+                       __smmu_client_set_hwgrp(c, c->hwgrp, 1);
+       }
+
+       smmu_write(smmu, smmu->translation_enable_0, SMMU_TRANSLATION_ENABLE_0);
+       smmu_write(smmu, smmu->translation_enable_1, SMMU_TRANSLATION_ENABLE_1);
+       smmu_write(smmu, smmu->translation_enable_2, SMMU_TRANSLATION_ENABLE_2);
+       smmu_write(smmu, smmu->asid_security, SMMU_ASID_SECURITY);
+       smmu_write(smmu, SMMU_TLB_CONFIG_RESET_VAL, SMMU_TLB_CONFIG);
+       smmu_write(smmu, SMMU_PTC_CONFIG_RESET_VAL, SMMU_PTC_CONFIG);
+
+       smmu_flush_regs(smmu, 1);
+
+       val = ahb_read(smmu, AHB_XBAR_CTRL);
+       val |= AHB_XBAR_CTRL_SMMU_INIT_DONE_DONE <<
+               AHB_XBAR_CTRL_SMMU_INIT_DONE_SHIFT;
+       ahb_write(smmu, val, AHB_XBAR_CTRL);
+}
+
+static void flush_ptc_and_tlb(struct smmu_device *smmu,
+                     struct smmu_as *as, dma_addr_t iova,
+                     unsigned long *pte, struct page *page, int is_pde)
+{
+       u32 val;
+       unsigned long tlb_flush_va = is_pde
+               ?  SMMU_TLB_FLUSH_VA(iova, SECTION)
+               :  SMMU_TLB_FLUSH_VA(iova, GROUP);
+
+       val = SMMU_PTC_FLUSH_TYPE_ADR | VA_PAGE_TO_PA(pte, page);
+       smmu_write(smmu, val, SMMU_PTC_FLUSH);
+       FLUSH_SMMU_REGS(smmu);
+       val = tlb_flush_va |
+               SMMU_TLB_FLUSH_ASID_MATCH__ENABLE |
+               (as->asid << SMMU_TLB_FLUSH_ASID_SHIFT);
+       smmu_write(smmu, val, SMMU_TLB_FLUSH);
+       FLUSH_SMMU_REGS(smmu);
+}
+
+static void free_ptbl(struct smmu_as *as, dma_addr_t iova)
+{
+       unsigned long pdn = SMMU_ADDR_TO_PDN(iova);
+       unsigned long *pdir = (unsigned long *)page_address(as->pdir_page);
+
+       if (pdir[pdn] != _PDE_VACANT(pdn)) {
+               dev_dbg(as->smmu->dev, "pdn: %lx\n", pdn);
+
+               ClearPageReserved(SMMU_EX_PTBL_PAGE(pdir[pdn]));
+               __free_page(SMMU_EX_PTBL_PAGE(pdir[pdn]));
+               pdir[pdn] = _PDE_VACANT(pdn);
+               FLUSH_CPU_DCACHE(&pdir[pdn], as->pdir_page, sizeof pdir[pdn]);
+               flush_ptc_and_tlb(as->smmu, as, iova, &pdir[pdn],
+                                 as->pdir_page, 1);
+       }
+}
+
+static void free_pdir(struct smmu_as *as)
+{
+       unsigned addr;
+       int count;
+       struct device *dev = as->smmu->dev;
+
+       if (!as->pdir_page)
+               return;
+
+       addr = as->smmu->iovmm_base;
+       count = as->smmu->page_count;
+       while (count-- > 0) {
+               free_ptbl(as, addr);
+               addr += SMMU_PAGE_SIZE * SMMU_PTBL_COUNT;
+       }
+       ClearPageReserved(as->pdir_page);
+       __free_page(as->pdir_page);
+       as->pdir_page = NULL;
+       devm_kfree(dev, as->pte_count);
+       as->pte_count = NULL;
+}
+
+/*
+ * Maps PTBL for given iova and returns the PTE address
+ * Caller must unmap the mapped PTBL returned in *ptbl_page_p
+ */
+static unsigned long *locate_pte(struct smmu_as *as,
+                                dma_addr_t iova, bool allocate,
+                                struct page **ptbl_page_p,
+                                unsigned int **count)
+{
+       unsigned long ptn = SMMU_ADDR_TO_PFN(iova);
+       unsigned long pdn = SMMU_ADDR_TO_PDN(iova);
+       unsigned long *pdir = page_address(as->pdir_page);
+       unsigned long *ptbl;
+
+       if (pdir[pdn] != _PDE_VACANT(pdn)) {
+               /* Mapped entry table already exists */
+               *ptbl_page_p = SMMU_EX_PTBL_PAGE(pdir[pdn]);
+               ptbl = page_address(*ptbl_page_p);
+       } else if (!allocate) {
+               return NULL;
+       } else {
+               int pn;
+               unsigned long addr = SMMU_PDN_TO_ADDR(pdn);
+
+               /* Vacant - allocate a new page table */
+               dev_dbg(as->smmu->dev, "New PTBL pdn: %lx\n", pdn);
+
+               *ptbl_page_p = alloc_page(GFP_ATOMIC);
+               if (!*ptbl_page_p) {
+                       dev_err(as->smmu->dev,
+                               "failed to allocate smmu_device page table\n");
+                       return NULL;
+               }
+               SetPageReserved(*ptbl_page_p);
+               ptbl = (unsigned long *)page_address(*ptbl_page_p);
+               for (pn = 0; pn < SMMU_PTBL_COUNT;
+                    pn++, addr += SMMU_PAGE_SIZE) {
+                       ptbl[pn] = _PTE_VACANT(addr);
+               }
+               FLUSH_CPU_DCACHE(ptbl, *ptbl_page_p, SMMU_PTBL_SIZE);
+               pdir[pdn] = SMMU_MK_PDE(*ptbl_page_p,
+                                       as->pde_attr | _PDE_NEXT);
+               FLUSH_CPU_DCACHE(&pdir[pdn], as->pdir_page, sizeof pdir[pdn]);
+               flush_ptc_and_tlb(as->smmu, as, iova, &pdir[pdn],
+                                 as->pdir_page, 1);
+       }
+       *count = &as->pte_count[pdn];
+
+       return &ptbl[ptn % SMMU_PTBL_COUNT];
+}
+
+#ifdef CONFIG_SMMU_SIG_DEBUG
+static void put_signature(struct smmu_as *as,
+                         dma_addr_t iova, unsigned long pfn)
+{
+       struct page *page;
+       unsigned long *vaddr;
+
+       page = pfn_to_page(pfn);
+       vaddr = page_address(page);
+       if (!vaddr)
+               return;
+
+       vaddr[0] = iova;
+       vaddr[1] = pfn << PAGE_SHIFT;
+       FLUSH_CPU_DCACHE(vaddr, page, sizeof(vaddr[0]) * 2);
+}
+#else
+static inline void put_signature(struct smmu_as *as,
+                                unsigned long addr, unsigned long pfn)
+{
+}
+#endif
+
+/*
+ * Caller must lock/unlock as
+ */
+static int alloc_pdir(struct smmu_as *as)
+{
+       unsigned long *pdir;
+       int pdn;
+       u32 val;
+       struct smmu_device *smmu = as->smmu;
+
+       if (as->pdir_page)
+               return 0;
+
+       as->pte_count = devm_kzalloc(smmu->dev,
+                    sizeof(as->pte_count[0]) * SMMU_PDIR_COUNT, GFP_KERNEL);
+       if (!as->pte_count) {
+               dev_err(smmu->dev,
+                       "failed to allocate smmu_device PTE cunters\n");
+               return -ENOMEM;
+       }
+       as->pdir_page = alloc_page(GFP_KERNEL | __GFP_DMA);
+       if (!as->pdir_page) {
+               dev_err(smmu->dev,
+                       "failed to allocate smmu_device page directory\n");
+               devm_kfree(smmu->dev, as->pte_count);
+               as->pte_count = NULL;
+               return -ENOMEM;
+       }
+       SetPageReserved(as->pdir_page);
+       pdir = page_address(as->pdir_page);
+
+       for (pdn = 0; pdn < SMMU_PDIR_COUNT; pdn++)
+               pdir[pdn] = _PDE_VACANT(pdn);
+       FLUSH_CPU_DCACHE(pdir, as->pdir_page, SMMU_PDIR_SIZE);
+       val = SMMU_PTC_FLUSH_TYPE_ADR | VA_PAGE_TO_PA(pdir, as->pdir_page);
+       smmu_write(smmu, val, SMMU_PTC_FLUSH);
+       FLUSH_SMMU_REGS(as->smmu);
+       val = SMMU_TLB_FLUSH_VA_MATCH_ALL |
+               SMMU_TLB_FLUSH_ASID_MATCH__ENABLE |
+               (as->asid << SMMU_TLB_FLUSH_ASID_SHIFT);
+       smmu_write(smmu, val, SMMU_TLB_FLUSH);
+       FLUSH_SMMU_REGS(as->smmu);
+
+       return 0;
+}
+
+static void __smmu_iommu_unmap(struct smmu_as *as, dma_addr_t iova)
+{
+       unsigned long *pte;
+       struct page *page;
+       unsigned int *count;
+
+       pte = locate_pte(as, iova, false, &page, &count);
+       if (WARN_ON(!pte))
+               return;
+
+       if (WARN_ON(*pte == _PTE_VACANT(iova)))
+               return;
+
+       *pte = _PTE_VACANT(iova);
+       FLUSH_CPU_DCACHE(pte, page, sizeof(*pte));
+       flush_ptc_and_tlb(as->smmu, as, iova, pte, page, 0);
+       if (!--(*count)) {
+               free_ptbl(as, iova);
+               smmu_flush_regs(as->smmu, 0);
+       }
+}
+
+static void __smmu_iommu_map_pfn(struct smmu_as *as, dma_addr_t iova,
+                                unsigned long pfn)
+{
+       struct smmu_device *smmu = as->smmu;
+       unsigned long *pte;
+       unsigned int *count;
+       struct page *page;
+
+       pte = locate_pte(as, iova, true, &page, &count);
+       if (WARN_ON(!pte))
+               return;
+
+       if (*pte == _PTE_VACANT(iova))
+               (*count)++;
+       *pte = SMMU_PFN_TO_PTE(pfn, as->pte_attr);
+       if (unlikely((*pte == _PTE_VACANT(iova))))
+               (*count)--;
+       FLUSH_CPU_DCACHE(pte, page, sizeof(*pte));
+       flush_ptc_and_tlb(smmu, as, iova, pte, page, 0);
+       put_signature(as, iova, pfn);
+}
+
+static int smmu_iommu_map(struct iommu_domain *domain, unsigned long iova,
+                         phys_addr_t pa, size_t bytes, int prot)
+{
+       struct smmu_as *as = domain->priv;
+       unsigned long pfn = __phys_to_pfn(pa);
+       unsigned long flags;
+
+       dev_dbg(as->smmu->dev, "[%d] %08lx:%08x\n", as->asid, iova, pa);
+
+       if (!pfn_valid(pfn))
+               return -ENOMEM;
+
+       spin_lock_irqsave(&as->lock, flags);
+       __smmu_iommu_map_pfn(as, iova, pfn);
+       spin_unlock_irqrestore(&as->lock, flags);
+       return 0;
+}
+
+static size_t smmu_iommu_unmap(struct iommu_domain *domain, unsigned long iova,
+                              size_t bytes)
+{
+       struct smmu_as *as = domain->priv;
+       unsigned long flags;
+
+       dev_dbg(as->smmu->dev, "[%d] %08lx\n", as->asid, iova);
+
+       spin_lock_irqsave(&as->lock, flags);
+       __smmu_iommu_unmap(as, iova);
+       spin_unlock_irqrestore(&as->lock, flags);
+       return SMMU_PAGE_SIZE;
+}
+
+static phys_addr_t smmu_iommu_iova_to_phys(struct iommu_domain *domain,
+                                          unsigned long iova)
+{
+       struct smmu_as *as = domain->priv;
+       unsigned long *pte;
+       unsigned int *count;
+       struct page *page;
+       unsigned long pfn;
+       unsigned long flags;
+
+       spin_lock_irqsave(&as->lock, flags);
+
+       pte = locate_pte(as, iova, true, &page, &count);
+       pfn = *pte & SMMU_PFN_MASK;
+       WARN_ON(!pfn_valid(pfn));
+       dev_dbg(as->smmu->dev,
+               "iova:%08lx pfn:%08lx asid:%d\n", iova, pfn, as->asid);
+
+       spin_unlock_irqrestore(&as->lock, flags);
+       return PFN_PHYS(pfn);
+}
+
+static int smmu_iommu_domain_has_cap(struct iommu_domain *domain,
+                                    unsigned long cap)
+{
+       return 0;
+}
+
+static int smmu_iommu_attach_dev(struct iommu_domain *domain,
+                                struct device *dev)
+{
+       struct smmu_as *as = domain->priv;
+       struct smmu_device *smmu = as->smmu;
+       struct smmu_client *client, *c;
+       u32 map;
+       int err;
+
+       client = devm_kzalloc(smmu->dev, sizeof(*c), GFP_KERNEL);
+       if (!client)
+               return -ENOMEM;
+       client->dev = dev;
+       client->as = as;
+       map = (unsigned long)dev->platform_data;
+       if (!map)
+               return -EINVAL;
+
+       err = smmu_client_enable_hwgrp(client, map);
+       if (err)
+               goto err_hwgrp;
+
+       spin_lock(&as->client_lock);
+       list_for_each_entry(c, &as->client, list) {
+               if (c->dev == dev) {
+                       dev_err(smmu->dev,
+                               "%s is already attached\n", dev_name(c->dev));
+                       err = -EINVAL;
+                       goto err_client;
+               }
+       }
+       list_add(&client->list, &as->client);
+       spin_unlock(&as->client_lock);
+
+       /*
+        * Reserve "page zero" for AVP vectors using a common dummy
+        * page.
+        */
+       if (map & HWG_AVPC) {
+               struct page *page;
+
+               page = as->smmu->avp_vector_page;
+               __smmu_iommu_map_pfn(as, 0, page_to_pfn(page));
+
+               pr_info("Reserve \"page zero\" for AVP vectors using a common dummy\n");
+       }
+
+       dev_dbg(smmu->dev, "%s is attached\n", dev_name(c->dev));
+       return 0;
+
+err_client:
+       smmu_client_disable_hwgrp(client);
+       spin_unlock(&as->client_lock);
+err_hwgrp:
+       devm_kfree(smmu->dev, client);
+       return err;
+}
+
+static void smmu_iommu_detach_dev(struct iommu_domain *domain,
+                                 struct device *dev)
+{
+       struct smmu_as *as = domain->priv;
+       struct smmu_device *smmu = as->smmu;
+       struct smmu_client *c;
+
+       spin_lock(&as->client_lock);
+
+       list_for_each_entry(c, &as->client, list) {
+               if (c->dev == dev) {
+                       smmu_client_disable_hwgrp(c);
+                       list_del(&c->list);
+                       devm_kfree(smmu->dev, c);
+                       c->as = NULL;
+                       dev_dbg(smmu->dev,
+                               "%s is detached\n", dev_name(c->dev));
+                       goto out;
+               }
+       }
+       dev_err(smmu->dev, "Couldn't find %s\n", dev_name(c->dev));
+out:
+       spin_unlock(&as->client_lock);
+}
+
+static int smmu_iommu_domain_init(struct iommu_domain *domain)
+{
+       int i;
+       unsigned long flags;
+       struct smmu_as *as;
+       struct smmu_device *smmu = smmu_handle;
+
+       /* Look for a free AS with lock held */
+       for  (i = 0; i < smmu->num_as; i++) {
+               struct smmu_as *tmp = &smmu->as[i];
+
+               spin_lock_irqsave(&tmp->lock, flags);
+               if (!tmp->pdir_page) {
+                       as = tmp;
+                       goto found;
+               }
+               spin_unlock_irqrestore(&tmp->lock, flags);
+       }
+       dev_err(smmu->dev, "no free AS\n");
+       return -ENODEV;
+
+found:
+       if (alloc_pdir(as) < 0)
+               goto err_alloc_pdir;
+
+       spin_lock(&smmu->lock);
+
+       /* Update PDIR register */
+       smmu_write(smmu, SMMU_PTB_ASID_CUR(as->asid), SMMU_PTB_ASID);
+       smmu_write(smmu,
+                  SMMU_MK_PDIR(as->pdir_page, as->pdir_attr), SMMU_PTB_DATA);
+       FLUSH_SMMU_REGS(smmu);
+
+       spin_unlock(&smmu->lock);
+
+       spin_unlock_irqrestore(&as->lock, flags);
+       domain->priv = as;
+
+       dev_dbg(smmu->dev, "smmu_as@%p\n", as);
+       return 0;
+
+err_alloc_pdir:
+       spin_unlock_irqrestore(&as->lock, flags);
+       return -ENODEV;
+}
+
+static void smmu_iommu_domain_destroy(struct iommu_domain *domain)
+{
+       struct smmu_as *as = domain->priv;
+       struct smmu_device *smmu = as->smmu;
+       unsigned long flags;
+
+       spin_lock_irqsave(&as->lock, flags);
+
+       if (as->pdir_page) {
+               spin_lock(&smmu->lock);
+               smmu_write(smmu, SMMU_PTB_ASID_CUR(as->asid), SMMU_PTB_ASID);
+               smmu_write(smmu, SMMU_PTB_DATA_RESET_VAL, SMMU_PTB_DATA);
+               FLUSH_SMMU_REGS(smmu);
+               spin_unlock(&smmu->lock);
+
+               free_pdir(as);
+       }
+
+       if (!list_empty(&as->client)) {
+               struct smmu_client *c;
+
+               list_for_each_entry(c, &as->client, list)
+                       smmu_iommu_detach_dev(domain, c->dev);
+       }
+
+       spin_unlock_irqrestore(&as->lock, flags);
+
+       domain->priv = NULL;
+       dev_dbg(smmu->dev, "smmu_as@%p\n", as);
+}
+
+static struct iommu_ops smmu_iommu_ops = {
+       .domain_init    = smmu_iommu_domain_init,
+       .domain_destroy = smmu_iommu_domain_destroy,
+       .attach_dev     = smmu_iommu_attach_dev,
+       .detach_dev     = smmu_iommu_detach_dev,
+       .map            = smmu_iommu_map,
+       .unmap          = smmu_iommu_unmap,
+       .iova_to_phys   = smmu_iommu_iova_to_phys,
+       .domain_has_cap = smmu_iommu_domain_has_cap,
+       .pgsize_bitmap  = SMMU_IOMMU_PGSIZES,
+};
+
+static int tegra_smmu_suspend(struct device *dev)
+{
+       struct smmu_device *smmu = dev_get_drvdata(dev);
+
+       smmu->translation_enable_0 = smmu_read(smmu, SMMU_TRANSLATION_ENABLE_0);
+       smmu->translation_enable_1 = smmu_read(smmu, SMMU_TRANSLATION_ENABLE_1);
+       smmu->translation_enable_2 = smmu_read(smmu, SMMU_TRANSLATION_ENABLE_2);
+       smmu->asid_security = smmu_read(smmu, SMMU_ASID_SECURITY);
+       return 0;
+}
+
+static int tegra_smmu_resume(struct device *dev)
+{
+       struct smmu_device *smmu = dev_get_drvdata(dev);
+       unsigned long flags;
+
+       spin_lock_irqsave(&smmu->lock, flags);
+       smmu_setup_regs(smmu);
+       spin_unlock_irqrestore(&smmu->lock, flags);
+       return 0;
+}
+
+static int tegra_smmu_probe(struct platform_device *pdev)
+{
+       struct smmu_device *smmu;
+       struct resource *regs, *regs2, *window;
+       struct device *dev = &pdev->dev;
+       int i, err = 0;
+
+       if (smmu_handle)
+               return -EIO;
+
+       BUILD_BUG_ON(PAGE_SHIFT != SMMU_PAGE_SHIFT);
+
+       regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       regs2 = platform_get_resource(pdev, IORESOURCE_MEM, 1);
+       window = platform_get_resource(pdev, IORESOURCE_MEM, 2);
+       if (!regs || !regs2 || !window) {
+               dev_err(dev, "No SMMU resources\n");
+               return -ENODEV;
+       }
+
+       smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
+       if (!smmu) {
+               dev_err(dev, "failed to allocate smmu_device\n");
+               return -ENOMEM;
+       }
+
+       smmu->dev = dev;
+       smmu->num_as = SMMU_NUM_ASIDS;
+       smmu->iovmm_base = (unsigned long)window->start;
+       smmu->page_count = resource_size(window) >> SMMU_PAGE_SHIFT;
+       smmu->regs = devm_ioremap(dev, regs->start, resource_size(regs));
+       smmu->regs_ahbarb = devm_ioremap(dev, regs2->start,
+                                        resource_size(regs2));
+       if (!smmu->regs || !smmu->regs_ahbarb) {
+               dev_err(dev, "failed to remap SMMU registers\n");
+               err = -ENXIO;
+               goto fail;
+       }
+
+       smmu->translation_enable_0 = ~0;
+       smmu->translation_enable_1 = ~0;
+       smmu->translation_enable_2 = ~0;
+       smmu->asid_security = 0;
+
+       smmu->as = devm_kzalloc(dev,
+                       sizeof(smmu->as[0]) * smmu->num_as, GFP_KERNEL);
+       if (!smmu->as) {
+               dev_err(dev, "failed to allocate smmu_as\n");
+               err = -ENOMEM;
+               goto fail;
+       }
+
+       for (i = 0; i < smmu->num_as; i++) {
+               struct smmu_as *as = &smmu->as[i];
+
+               as->smmu = smmu;
+               as->asid = i;
+               as->pdir_attr = _PDIR_ATTR;
+               as->pde_attr = _PDE_ATTR;
+               as->pte_attr = _PTE_ATTR;
+
+               spin_lock_init(&as->lock);
+               INIT_LIST_HEAD(&as->client);
+       }
+       spin_lock_init(&smmu->lock);
+       smmu_setup_regs(smmu);
+       platform_set_drvdata(pdev, smmu);
+
+       smmu->avp_vector_page = alloc_page(GFP_KERNEL);
+       if (!smmu->avp_vector_page)
+               goto fail;
+
+       smmu_handle = smmu;
+       return 0;
+
+fail:
+       if (smmu->avp_vector_page)
+               __free_page(smmu->avp_vector_page);
+       if (smmu->regs)
+               devm_iounmap(dev, smmu->regs);
+       if (smmu->regs_ahbarb)
+               devm_iounmap(dev, smmu->regs_ahbarb);
+       if (smmu && smmu->as) {
+               for (i = 0; i < smmu->num_as; i++) {
+                       if (smmu->as[i].pdir_page) {
+                               ClearPageReserved(smmu->as[i].pdir_page);
+                               __free_page(smmu->as[i].pdir_page);
+                       }
+               }
+               devm_kfree(dev, smmu->as);
+       }
+       devm_kfree(dev, smmu);
+       return err;
+}
+
+static int tegra_smmu_remove(struct platform_device *pdev)
+{
+       struct smmu_device *smmu = platform_get_drvdata(pdev);
+       struct device *dev = smmu->dev;
+
+       smmu_write(smmu, SMMU_CONFIG_DISABLE, SMMU_CONFIG);
+       platform_set_drvdata(pdev, NULL);
+       if (smmu->as) {
+               int i;
+
+               for (i = 0; i < smmu->num_as; i++)
+                       free_pdir(&smmu->as[i]);
+               devm_kfree(dev, smmu->as);
+       }
+       if (smmu->avp_vector_page)
+               __free_page(smmu->avp_vector_page);
+       if (smmu->regs)
+               devm_iounmap(dev, smmu->regs);
+       if (smmu->regs_ahbarb)
+               devm_iounmap(dev, smmu->regs_ahbarb);
+       devm_kfree(dev, smmu);
+       smmu_handle = NULL;
+       return 0;
+}
+
+const struct dev_pm_ops tegra_smmu_pm_ops = {
+       .suspend        = tegra_smmu_suspend,
+       .resume         = tegra_smmu_resume,
+};
+
+static struct platform_driver tegra_smmu_driver = {
+       .probe          = tegra_smmu_probe,
+       .remove         = tegra_smmu_remove,
+       .driver = {
+               .owner  = THIS_MODULE,
+               .name   = "tegra-smmu",
+               .pm     = &tegra_smmu_pm_ops,
+       },
+};
+
+static int __devinit tegra_smmu_init(void)
+{
+       bus_set_iommu(&platform_bus_type, &smmu_iommu_ops);
+       return platform_driver_register(&tegra_smmu_driver);
+}
+
+static void __exit tegra_smmu_exit(void)
+{
+       platform_driver_unregister(&tegra_smmu_driver);
+}
+
+subsys_initcall(tegra_smmu_init);
+module_exit(tegra_smmu_exit);
+
+MODULE_DESCRIPTION("IOMMU API for SMMU in Tegra30");
+MODULE_AUTHOR("Hiroshi DOYU <hdoyu@nvidia.com>");
+MODULE_LICENSE("GPL v2");