Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / iommu / omap-iommu.c
diff --git a/kernel/drivers/iommu/omap-iommu.c b/kernel/drivers/iommu/omap-iommu.c
new file mode 100644 (file)
index 0000000..a22c33d
--- /dev/null
@@ -0,0 +1,1424 @@
+/*
+ * omap iommu: tlb and pagetable primitives
+ *
+ * Copyright (C) 2008-2010 Nokia Corporation
+ *
+ * Written by Hiroshi DOYU <Hiroshi.DOYU@nokia.com>,
+ *             Paul Mundt and Toshihiro Kobayashi
+ *
+ * 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/err.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/interrupt.h>
+#include <linux/ioport.h>
+#include <linux/platform_device.h>
+#include <linux/iommu.h>
+#include <linux/omap-iommu.h>
+#include <linux/mutex.h>
+#include <linux/spinlock.h>
+#include <linux/io.h>
+#include <linux/pm_runtime.h>
+#include <linux/of.h>
+#include <linux/of_iommu.h>
+#include <linux/of_irq.h>
+#include <linux/of_platform.h>
+
+#include <asm/cacheflush.h>
+
+#include <linux/platform_data/iommu-omap.h>
+
+#include "omap-iopgtable.h"
+#include "omap-iommu.h"
+
+#define to_iommu(dev)                                                  \
+       ((struct omap_iommu *)platform_get_drvdata(to_platform_device(dev)))
+
+#define for_each_iotlb_cr(obj, n, __i, cr)                             \
+       for (__i = 0;                                                   \
+            (__i < (n)) && (cr = __iotlb_read_cr((obj), __i), true);   \
+            __i++)
+
+/* bitmap of the page sizes currently supported */
+#define OMAP_IOMMU_PGSIZES     (SZ_4K | SZ_64K | SZ_1M | SZ_16M)
+
+/**
+ * struct omap_iommu_domain - omap iommu domain
+ * @pgtable:   the page table
+ * @iommu_dev: an omap iommu device attached to this domain. only a single
+ *             iommu device can be attached for now.
+ * @dev:       Device using this domain.
+ * @lock:      domain lock, should be taken when attaching/detaching
+ */
+struct omap_iommu_domain {
+       u32 *pgtable;
+       struct omap_iommu *iommu_dev;
+       struct device *dev;
+       spinlock_t lock;
+       struct iommu_domain domain;
+};
+
+#define MMU_LOCK_BASE_SHIFT    10
+#define MMU_LOCK_BASE_MASK     (0x1f << MMU_LOCK_BASE_SHIFT)
+#define MMU_LOCK_BASE(x)       \
+       ((x & MMU_LOCK_BASE_MASK) >> MMU_LOCK_BASE_SHIFT)
+
+#define MMU_LOCK_VICT_SHIFT    4
+#define MMU_LOCK_VICT_MASK     (0x1f << MMU_LOCK_VICT_SHIFT)
+#define MMU_LOCK_VICT(x)       \
+       ((x & MMU_LOCK_VICT_MASK) >> MMU_LOCK_VICT_SHIFT)
+
+struct iotlb_lock {
+       short base;
+       short vict;
+};
+
+static struct platform_driver omap_iommu_driver;
+static struct kmem_cache *iopte_cachep;
+
+/**
+ * to_omap_domain - Get struct omap_iommu_domain from generic iommu_domain
+ * @dom:       generic iommu domain handle
+ **/
+static struct omap_iommu_domain *to_omap_domain(struct iommu_domain *dom)
+{
+       return container_of(dom, struct omap_iommu_domain, domain);
+}
+
+/**
+ * omap_iommu_save_ctx - Save registers for pm off-mode support
+ * @dev:       client device
+ **/
+void omap_iommu_save_ctx(struct device *dev)
+{
+       struct omap_iommu *obj = dev_to_omap_iommu(dev);
+       u32 *p = obj->ctx;
+       int i;
+
+       for (i = 0; i < (MMU_REG_SIZE / sizeof(u32)); i++) {
+               p[i] = iommu_read_reg(obj, i * sizeof(u32));
+               dev_dbg(obj->dev, "%s\t[%02d] %08x\n", __func__, i, p[i]);
+       }
+}
+EXPORT_SYMBOL_GPL(omap_iommu_save_ctx);
+
+/**
+ * omap_iommu_restore_ctx - Restore registers for pm off-mode support
+ * @dev:       client device
+ **/
+void omap_iommu_restore_ctx(struct device *dev)
+{
+       struct omap_iommu *obj = dev_to_omap_iommu(dev);
+       u32 *p = obj->ctx;
+       int i;
+
+       for (i = 0; i < (MMU_REG_SIZE / sizeof(u32)); i++) {
+               iommu_write_reg(obj, p[i], i * sizeof(u32));
+               dev_dbg(obj->dev, "%s\t[%02d] %08x\n", __func__, i, p[i]);
+       }
+}
+EXPORT_SYMBOL_GPL(omap_iommu_restore_ctx);
+
+static void __iommu_set_twl(struct omap_iommu *obj, bool on)
+{
+       u32 l = iommu_read_reg(obj, MMU_CNTL);
+
+       if (on)
+               iommu_write_reg(obj, MMU_IRQ_TWL_MASK, MMU_IRQENABLE);
+       else
+               iommu_write_reg(obj, MMU_IRQ_TLB_MISS_MASK, MMU_IRQENABLE);
+
+       l &= ~MMU_CNTL_MASK;
+       if (on)
+               l |= (MMU_CNTL_MMU_EN | MMU_CNTL_TWL_EN);
+       else
+               l |= (MMU_CNTL_MMU_EN);
+
+       iommu_write_reg(obj, l, MMU_CNTL);
+}
+
+static int omap2_iommu_enable(struct omap_iommu *obj)
+{
+       u32 l, pa;
+
+       if (!obj->iopgd || !IS_ALIGNED((u32)obj->iopgd,  SZ_16K))
+               return -EINVAL;
+
+       pa = virt_to_phys(obj->iopgd);
+       if (!IS_ALIGNED(pa, SZ_16K))
+               return -EINVAL;
+
+       l = iommu_read_reg(obj, MMU_REVISION);
+       dev_info(obj->dev, "%s: version %d.%d\n", obj->name,
+                (l >> 4) & 0xf, l & 0xf);
+
+       iommu_write_reg(obj, pa, MMU_TTB);
+
+       if (obj->has_bus_err_back)
+               iommu_write_reg(obj, MMU_GP_REG_BUS_ERR_BACK_EN, MMU_GP_REG);
+
+       __iommu_set_twl(obj, true);
+
+       return 0;
+}
+
+static void omap2_iommu_disable(struct omap_iommu *obj)
+{
+       u32 l = iommu_read_reg(obj, MMU_CNTL);
+
+       l &= ~MMU_CNTL_MASK;
+       iommu_write_reg(obj, l, MMU_CNTL);
+
+       dev_dbg(obj->dev, "%s is shutting down\n", obj->name);
+}
+
+static int iommu_enable(struct omap_iommu *obj)
+{
+       int err;
+       struct platform_device *pdev = to_platform_device(obj->dev);
+       struct iommu_platform_data *pdata = dev_get_platdata(&pdev->dev);
+
+       if (pdata && pdata->deassert_reset) {
+               err = pdata->deassert_reset(pdev, pdata->reset_name);
+               if (err) {
+                       dev_err(obj->dev, "deassert_reset failed: %d\n", err);
+                       return err;
+               }
+       }
+
+       pm_runtime_get_sync(obj->dev);
+
+       err = omap2_iommu_enable(obj);
+
+       return err;
+}
+
+static void iommu_disable(struct omap_iommu *obj)
+{
+       struct platform_device *pdev = to_platform_device(obj->dev);
+       struct iommu_platform_data *pdata = dev_get_platdata(&pdev->dev);
+
+       omap2_iommu_disable(obj);
+
+       pm_runtime_put_sync(obj->dev);
+
+       if (pdata && pdata->assert_reset)
+               pdata->assert_reset(pdev, pdata->reset_name);
+}
+
+/*
+ *     TLB operations
+ */
+static inline int iotlb_cr_valid(struct cr_regs *cr)
+{
+       if (!cr)
+               return -EINVAL;
+
+       return cr->cam & MMU_CAM_V;
+}
+
+static u32 iotlb_cr_to_virt(struct cr_regs *cr)
+{
+       u32 page_size = cr->cam & MMU_CAM_PGSZ_MASK;
+       u32 mask = get_cam_va_mask(cr->cam & page_size);
+
+       return cr->cam & mask;
+}
+
+static u32 get_iopte_attr(struct iotlb_entry *e)
+{
+       u32 attr;
+
+       attr = e->mixed << 5;
+       attr |= e->endian;
+       attr |= e->elsz >> 3;
+       attr <<= (((e->pgsz == MMU_CAM_PGSZ_4K) ||
+                       (e->pgsz == MMU_CAM_PGSZ_64K)) ? 0 : 6);
+       return attr;
+}
+
+static u32 iommu_report_fault(struct omap_iommu *obj, u32 *da)
+{
+       u32 status, fault_addr;
+
+       status = iommu_read_reg(obj, MMU_IRQSTATUS);
+       status &= MMU_IRQ_MASK;
+       if (!status) {
+               *da = 0;
+               return 0;
+       }
+
+       fault_addr = iommu_read_reg(obj, MMU_FAULT_AD);
+       *da = fault_addr;
+
+       iommu_write_reg(obj, status, MMU_IRQSTATUS);
+
+       return status;
+}
+
+static void iotlb_lock_get(struct omap_iommu *obj, struct iotlb_lock *l)
+{
+       u32 val;
+
+       val = iommu_read_reg(obj, MMU_LOCK);
+
+       l->base = MMU_LOCK_BASE(val);
+       l->vict = MMU_LOCK_VICT(val);
+
+}
+
+static void iotlb_lock_set(struct omap_iommu *obj, struct iotlb_lock *l)
+{
+       u32 val;
+
+       val = (l->base << MMU_LOCK_BASE_SHIFT);
+       val |= (l->vict << MMU_LOCK_VICT_SHIFT);
+
+       iommu_write_reg(obj, val, MMU_LOCK);
+}
+
+static void iotlb_read_cr(struct omap_iommu *obj, struct cr_regs *cr)
+{
+       cr->cam = iommu_read_reg(obj, MMU_READ_CAM);
+       cr->ram = iommu_read_reg(obj, MMU_READ_RAM);
+}
+
+static void iotlb_load_cr(struct omap_iommu *obj, struct cr_regs *cr)
+{
+       iommu_write_reg(obj, cr->cam | MMU_CAM_V, MMU_CAM);
+       iommu_write_reg(obj, cr->ram, MMU_RAM);
+
+       iommu_write_reg(obj, 1, MMU_FLUSH_ENTRY);
+       iommu_write_reg(obj, 1, MMU_LD_TLB);
+}
+
+/* only used in iotlb iteration for-loop */
+static struct cr_regs __iotlb_read_cr(struct omap_iommu *obj, int n)
+{
+       struct cr_regs cr;
+       struct iotlb_lock l;
+
+       iotlb_lock_get(obj, &l);
+       l.vict = n;
+       iotlb_lock_set(obj, &l);
+       iotlb_read_cr(obj, &cr);
+
+       return cr;
+}
+
+#ifdef PREFETCH_IOTLB
+static struct cr_regs *iotlb_alloc_cr(struct omap_iommu *obj,
+                                     struct iotlb_entry *e)
+{
+       struct cr_regs *cr;
+
+       if (!e)
+               return NULL;
+
+       if (e->da & ~(get_cam_va_mask(e->pgsz))) {
+               dev_err(obj->dev, "%s:\twrong alignment: %08x\n", __func__,
+                       e->da);
+               return ERR_PTR(-EINVAL);
+       }
+
+       cr = kmalloc(sizeof(*cr), GFP_KERNEL);
+       if (!cr)
+               return ERR_PTR(-ENOMEM);
+
+       cr->cam = (e->da & MMU_CAM_VATAG_MASK) | e->prsvd | e->pgsz | e->valid;
+       cr->ram = e->pa | e->endian | e->elsz | e->mixed;
+
+       return cr;
+}
+
+/**
+ * load_iotlb_entry - Set an iommu tlb entry
+ * @obj:       target iommu
+ * @e:         an iommu tlb entry info
+ **/
+static int load_iotlb_entry(struct omap_iommu *obj, struct iotlb_entry *e)
+{
+       int err = 0;
+       struct iotlb_lock l;
+       struct cr_regs *cr;
+
+       if (!obj || !obj->nr_tlb_entries || !e)
+               return -EINVAL;
+
+       pm_runtime_get_sync(obj->dev);
+
+       iotlb_lock_get(obj, &l);
+       if (l.base == obj->nr_tlb_entries) {
+               dev_warn(obj->dev, "%s: preserve entries full\n", __func__);
+               err = -EBUSY;
+               goto out;
+       }
+       if (!e->prsvd) {
+               int i;
+               struct cr_regs tmp;
+
+               for_each_iotlb_cr(obj, obj->nr_tlb_entries, i, tmp)
+                       if (!iotlb_cr_valid(&tmp))
+                               break;
+
+               if (i == obj->nr_tlb_entries) {
+                       dev_dbg(obj->dev, "%s: full: no entry\n", __func__);
+                       err = -EBUSY;
+                       goto out;
+               }
+
+               iotlb_lock_get(obj, &l);
+       } else {
+               l.vict = l.base;
+               iotlb_lock_set(obj, &l);
+       }
+
+       cr = iotlb_alloc_cr(obj, e);
+       if (IS_ERR(cr)) {
+               pm_runtime_put_sync(obj->dev);
+               return PTR_ERR(cr);
+       }
+
+       iotlb_load_cr(obj, cr);
+       kfree(cr);
+
+       if (e->prsvd)
+               l.base++;
+       /* increment victim for next tlb load */
+       if (++l.vict == obj->nr_tlb_entries)
+               l.vict = l.base;
+       iotlb_lock_set(obj, &l);
+out:
+       pm_runtime_put_sync(obj->dev);
+       return err;
+}
+
+#else /* !PREFETCH_IOTLB */
+
+static int load_iotlb_entry(struct omap_iommu *obj, struct iotlb_entry *e)
+{
+       return 0;
+}
+
+#endif /* !PREFETCH_IOTLB */
+
+static int prefetch_iotlb_entry(struct omap_iommu *obj, struct iotlb_entry *e)
+{
+       return load_iotlb_entry(obj, e);
+}
+
+/**
+ * flush_iotlb_page - Clear an iommu tlb entry
+ * @obj:       target iommu
+ * @da:                iommu device virtual address
+ *
+ * Clear an iommu tlb entry which includes 'da' address.
+ **/
+static void flush_iotlb_page(struct omap_iommu *obj, u32 da)
+{
+       int i;
+       struct cr_regs cr;
+
+       pm_runtime_get_sync(obj->dev);
+
+       for_each_iotlb_cr(obj, obj->nr_tlb_entries, i, cr) {
+               u32 start;
+               size_t bytes;
+
+               if (!iotlb_cr_valid(&cr))
+                       continue;
+
+               start = iotlb_cr_to_virt(&cr);
+               bytes = iopgsz_to_bytes(cr.cam & 3);
+
+               if ((start <= da) && (da < start + bytes)) {
+                       dev_dbg(obj->dev, "%s: %08x<=%08x(%x)\n",
+                               __func__, start, da, bytes);
+                       iotlb_load_cr(obj, &cr);
+                       iommu_write_reg(obj, 1, MMU_FLUSH_ENTRY);
+                       break;
+               }
+       }
+       pm_runtime_put_sync(obj->dev);
+
+       if (i == obj->nr_tlb_entries)
+               dev_dbg(obj->dev, "%s: no page for %08x\n", __func__, da);
+}
+
+/**
+ * flush_iotlb_all - Clear all iommu tlb entries
+ * @obj:       target iommu
+ **/
+static void flush_iotlb_all(struct omap_iommu *obj)
+{
+       struct iotlb_lock l;
+
+       pm_runtime_get_sync(obj->dev);
+
+       l.base = 0;
+       l.vict = 0;
+       iotlb_lock_set(obj, &l);
+
+       iommu_write_reg(obj, 1, MMU_GFLUSH);
+
+       pm_runtime_put_sync(obj->dev);
+}
+
+#ifdef CONFIG_OMAP_IOMMU_DEBUG
+
+#define pr_reg(name)                                                   \
+       do {                                                            \
+               ssize_t bytes;                                          \
+               const char *str = "%20s: %08x\n";                       \
+               const int maxcol = 32;                                  \
+               bytes = snprintf(p, maxcol, str, __stringify(name),     \
+                                iommu_read_reg(obj, MMU_##name));      \
+               p += bytes;                                             \
+               len -= bytes;                                           \
+               if (len < maxcol)                                       \
+                       goto out;                                       \
+       } while (0)
+
+static ssize_t
+omap2_iommu_dump_ctx(struct omap_iommu *obj, char *buf, ssize_t len)
+{
+       char *p = buf;
+
+       pr_reg(REVISION);
+       pr_reg(IRQSTATUS);
+       pr_reg(IRQENABLE);
+       pr_reg(WALKING_ST);
+       pr_reg(CNTL);
+       pr_reg(FAULT_AD);
+       pr_reg(TTB);
+       pr_reg(LOCK);
+       pr_reg(LD_TLB);
+       pr_reg(CAM);
+       pr_reg(RAM);
+       pr_reg(GFLUSH);
+       pr_reg(FLUSH_ENTRY);
+       pr_reg(READ_CAM);
+       pr_reg(READ_RAM);
+       pr_reg(EMU_FAULT_AD);
+out:
+       return p - buf;
+}
+
+ssize_t omap_iommu_dump_ctx(struct omap_iommu *obj, char *buf, ssize_t bytes)
+{
+       if (!obj || !buf)
+               return -EINVAL;
+
+       pm_runtime_get_sync(obj->dev);
+
+       bytes = omap2_iommu_dump_ctx(obj, buf, bytes);
+
+       pm_runtime_put_sync(obj->dev);
+
+       return bytes;
+}
+
+static int
+__dump_tlb_entries(struct omap_iommu *obj, struct cr_regs *crs, int num)
+{
+       int i;
+       struct iotlb_lock saved;
+       struct cr_regs tmp;
+       struct cr_regs *p = crs;
+
+       pm_runtime_get_sync(obj->dev);
+       iotlb_lock_get(obj, &saved);
+
+       for_each_iotlb_cr(obj, num, i, tmp) {
+               if (!iotlb_cr_valid(&tmp))
+                       continue;
+               *p++ = tmp;
+       }
+
+       iotlb_lock_set(obj, &saved);
+       pm_runtime_put_sync(obj->dev);
+
+       return  p - crs;
+}
+
+/**
+ * iotlb_dump_cr - Dump an iommu tlb entry into buf
+ * @obj:       target iommu
+ * @cr:                contents of cam and ram register
+ * @buf:       output buffer
+ **/
+static ssize_t iotlb_dump_cr(struct omap_iommu *obj, struct cr_regs *cr,
+                            char *buf)
+{
+       char *p = buf;
+
+       /* FIXME: Need more detail analysis of cam/ram */
+       p += sprintf(p, "%08x %08x %01x\n", cr->cam, cr->ram,
+                                       (cr->cam & MMU_CAM_P) ? 1 : 0);
+
+       return p - buf;
+}
+
+/**
+ * omap_dump_tlb_entries - dump cr arrays to given buffer
+ * @obj:       target iommu
+ * @buf:       output buffer
+ **/
+size_t omap_dump_tlb_entries(struct omap_iommu *obj, char *buf, ssize_t bytes)
+{
+       int i, num;
+       struct cr_regs *cr;
+       char *p = buf;
+
+       num = bytes / sizeof(*cr);
+       num = min(obj->nr_tlb_entries, num);
+
+       cr = kcalloc(num, sizeof(*cr), GFP_KERNEL);
+       if (!cr)
+               return 0;
+
+       num = __dump_tlb_entries(obj, cr, num);
+       for (i = 0; i < num; i++)
+               p += iotlb_dump_cr(obj, cr + i, p);
+       kfree(cr);
+
+       return p - buf;
+}
+
+#endif /* CONFIG_OMAP_IOMMU_DEBUG */
+
+/*
+ *     H/W pagetable operations
+ */
+static void flush_iopgd_range(u32 *first, u32 *last)
+{
+       /* FIXME: L2 cache should be taken care of if it exists */
+       do {
+               asm("mcr        p15, 0, %0, c7, c10, 1 @ flush_pgd"
+                   : : "r" (first));
+               first += L1_CACHE_BYTES / sizeof(*first);
+       } while (first <= last);
+}
+
+static void flush_iopte_range(u32 *first, u32 *last)
+{
+       /* FIXME: L2 cache should be taken care of if it exists */
+       do {
+               asm("mcr        p15, 0, %0, c7, c10, 1 @ flush_pte"
+                   : : "r" (first));
+               first += L1_CACHE_BYTES / sizeof(*first);
+       } while (first <= last);
+}
+
+static void iopte_free(u32 *iopte)
+{
+       /* Note: freed iopte's must be clean ready for re-use */
+       if (iopte)
+               kmem_cache_free(iopte_cachep, iopte);
+}
+
+static u32 *iopte_alloc(struct omap_iommu *obj, u32 *iopgd, u32 da)
+{
+       u32 *iopte;
+
+       /* a table has already existed */
+       if (*iopgd)
+               goto pte_ready;
+
+       /*
+        * do the allocation outside the page table lock
+        */
+       spin_unlock(&obj->page_table_lock);
+       iopte = kmem_cache_zalloc(iopte_cachep, GFP_KERNEL);
+       spin_lock(&obj->page_table_lock);
+
+       if (!*iopgd) {
+               if (!iopte)
+                       return ERR_PTR(-ENOMEM);
+
+               *iopgd = virt_to_phys(iopte) | IOPGD_TABLE;
+               flush_iopgd_range(iopgd, iopgd);
+
+               dev_vdbg(obj->dev, "%s: a new pte:%p\n", __func__, iopte);
+       } else {
+               /* We raced, free the reduniovant table */
+               iopte_free(iopte);
+       }
+
+pte_ready:
+       iopte = iopte_offset(iopgd, da);
+
+       dev_vdbg(obj->dev,
+                "%s: da:%08x pgd:%p *pgd:%08x pte:%p *pte:%08x\n",
+                __func__, da, iopgd, *iopgd, iopte, *iopte);
+
+       return iopte;
+}
+
+static int iopgd_alloc_section(struct omap_iommu *obj, u32 da, u32 pa, u32 prot)
+{
+       u32 *iopgd = iopgd_offset(obj, da);
+
+       if ((da | pa) & ~IOSECTION_MASK) {
+               dev_err(obj->dev, "%s: %08x:%08x should aligned on %08lx\n",
+                       __func__, da, pa, IOSECTION_SIZE);
+               return -EINVAL;
+       }
+
+       *iopgd = (pa & IOSECTION_MASK) | prot | IOPGD_SECTION;
+       flush_iopgd_range(iopgd, iopgd);
+       return 0;
+}
+
+static int iopgd_alloc_super(struct omap_iommu *obj, u32 da, u32 pa, u32 prot)
+{
+       u32 *iopgd = iopgd_offset(obj, da);
+       int i;
+
+       if ((da | pa) & ~IOSUPER_MASK) {
+               dev_err(obj->dev, "%s: %08x:%08x should aligned on %08lx\n",
+                       __func__, da, pa, IOSUPER_SIZE);
+               return -EINVAL;
+       }
+
+       for (i = 0; i < 16; i++)
+               *(iopgd + i) = (pa & IOSUPER_MASK) | prot | IOPGD_SUPER;
+       flush_iopgd_range(iopgd, iopgd + 15);
+       return 0;
+}
+
+static int iopte_alloc_page(struct omap_iommu *obj, u32 da, u32 pa, u32 prot)
+{
+       u32 *iopgd = iopgd_offset(obj, da);
+       u32 *iopte = iopte_alloc(obj, iopgd, da);
+
+       if (IS_ERR(iopte))
+               return PTR_ERR(iopte);
+
+       *iopte = (pa & IOPAGE_MASK) | prot | IOPTE_SMALL;
+       flush_iopte_range(iopte, iopte);
+
+       dev_vdbg(obj->dev, "%s: da:%08x pa:%08x pte:%p *pte:%08x\n",
+                __func__, da, pa, iopte, *iopte);
+
+       return 0;
+}
+
+static int iopte_alloc_large(struct omap_iommu *obj, u32 da, u32 pa, u32 prot)
+{
+       u32 *iopgd = iopgd_offset(obj, da);
+       u32 *iopte = iopte_alloc(obj, iopgd, da);
+       int i;
+
+       if ((da | pa) & ~IOLARGE_MASK) {
+               dev_err(obj->dev, "%s: %08x:%08x should aligned on %08lx\n",
+                       __func__, da, pa, IOLARGE_SIZE);
+               return -EINVAL;
+       }
+
+       if (IS_ERR(iopte))
+               return PTR_ERR(iopte);
+
+       for (i = 0; i < 16; i++)
+               *(iopte + i) = (pa & IOLARGE_MASK) | prot | IOPTE_LARGE;
+       flush_iopte_range(iopte, iopte + 15);
+       return 0;
+}
+
+static int
+iopgtable_store_entry_core(struct omap_iommu *obj, struct iotlb_entry *e)
+{
+       int (*fn)(struct omap_iommu *, u32, u32, u32);
+       u32 prot;
+       int err;
+
+       if (!obj || !e)
+               return -EINVAL;
+
+       switch (e->pgsz) {
+       case MMU_CAM_PGSZ_16M:
+               fn = iopgd_alloc_super;
+               break;
+       case MMU_CAM_PGSZ_1M:
+               fn = iopgd_alloc_section;
+               break;
+       case MMU_CAM_PGSZ_64K:
+               fn = iopte_alloc_large;
+               break;
+       case MMU_CAM_PGSZ_4K:
+               fn = iopte_alloc_page;
+               break;
+       default:
+               fn = NULL;
+               BUG();
+               break;
+       }
+
+       prot = get_iopte_attr(e);
+
+       spin_lock(&obj->page_table_lock);
+       err = fn(obj, e->da, e->pa, prot);
+       spin_unlock(&obj->page_table_lock);
+
+       return err;
+}
+
+/**
+ * omap_iopgtable_store_entry - Make an iommu pte entry
+ * @obj:       target iommu
+ * @e:         an iommu tlb entry info
+ **/
+static int
+omap_iopgtable_store_entry(struct omap_iommu *obj, struct iotlb_entry *e)
+{
+       int err;
+
+       flush_iotlb_page(obj, e->da);
+       err = iopgtable_store_entry_core(obj, e);
+       if (!err)
+               prefetch_iotlb_entry(obj, e);
+       return err;
+}
+
+/**
+ * iopgtable_lookup_entry - Lookup an iommu pte entry
+ * @obj:       target iommu
+ * @da:                iommu device virtual address
+ * @ppgd:      iommu pgd entry pointer to be returned
+ * @ppte:      iommu pte entry pointer to be returned
+ **/
+static void
+iopgtable_lookup_entry(struct omap_iommu *obj, u32 da, u32 **ppgd, u32 **ppte)
+{
+       u32 *iopgd, *iopte = NULL;
+
+       iopgd = iopgd_offset(obj, da);
+       if (!*iopgd)
+               goto out;
+
+       if (iopgd_is_table(*iopgd))
+               iopte = iopte_offset(iopgd, da);
+out:
+       *ppgd = iopgd;
+       *ppte = iopte;
+}
+
+static size_t iopgtable_clear_entry_core(struct omap_iommu *obj, u32 da)
+{
+       size_t bytes;
+       u32 *iopgd = iopgd_offset(obj, da);
+       int nent = 1;
+
+       if (!*iopgd)
+               return 0;
+
+       if (iopgd_is_table(*iopgd)) {
+               int i;
+               u32 *iopte = iopte_offset(iopgd, da);
+
+               bytes = IOPTE_SIZE;
+               if (*iopte & IOPTE_LARGE) {
+                       nent *= 16;
+                       /* rewind to the 1st entry */
+                       iopte = iopte_offset(iopgd, (da & IOLARGE_MASK));
+               }
+               bytes *= nent;
+               memset(iopte, 0, nent * sizeof(*iopte));
+               flush_iopte_range(iopte, iopte + (nent - 1) * sizeof(*iopte));
+
+               /*
+                * do table walk to check if this table is necessary or not
+                */
+               iopte = iopte_offset(iopgd, 0);
+               for (i = 0; i < PTRS_PER_IOPTE; i++)
+                       if (iopte[i])
+                               goto out;
+
+               iopte_free(iopte);
+               nent = 1; /* for the next L1 entry */
+       } else {
+               bytes = IOPGD_SIZE;
+               if ((*iopgd & IOPGD_SUPER) == IOPGD_SUPER) {
+                       nent *= 16;
+                       /* rewind to the 1st entry */
+                       iopgd = iopgd_offset(obj, (da & IOSUPER_MASK));
+               }
+               bytes *= nent;
+       }
+       memset(iopgd, 0, nent * sizeof(*iopgd));
+       flush_iopgd_range(iopgd, iopgd + (nent - 1) * sizeof(*iopgd));
+out:
+       return bytes;
+}
+
+/**
+ * iopgtable_clear_entry - Remove an iommu pte entry
+ * @obj:       target iommu
+ * @da:                iommu device virtual address
+ **/
+static size_t iopgtable_clear_entry(struct omap_iommu *obj, u32 da)
+{
+       size_t bytes;
+
+       spin_lock(&obj->page_table_lock);
+
+       bytes = iopgtable_clear_entry_core(obj, da);
+       flush_iotlb_page(obj, da);
+
+       spin_unlock(&obj->page_table_lock);
+
+       return bytes;
+}
+
+static void iopgtable_clear_entry_all(struct omap_iommu *obj)
+{
+       int i;
+
+       spin_lock(&obj->page_table_lock);
+
+       for (i = 0; i < PTRS_PER_IOPGD; i++) {
+               u32 da;
+               u32 *iopgd;
+
+               da = i << IOPGD_SHIFT;
+               iopgd = iopgd_offset(obj, da);
+
+               if (!*iopgd)
+                       continue;
+
+               if (iopgd_is_table(*iopgd))
+                       iopte_free(iopte_offset(iopgd, 0));
+
+               *iopgd = 0;
+               flush_iopgd_range(iopgd, iopgd);
+       }
+
+       flush_iotlb_all(obj);
+
+       spin_unlock(&obj->page_table_lock);
+}
+
+/*
+ *     Device IOMMU generic operations
+ */
+static irqreturn_t iommu_fault_handler(int irq, void *data)
+{
+       u32 da, errs;
+       u32 *iopgd, *iopte;
+       struct omap_iommu *obj = data;
+       struct iommu_domain *domain = obj->domain;
+       struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
+
+       if (!omap_domain->iommu_dev)
+               return IRQ_NONE;
+
+       errs = iommu_report_fault(obj, &da);
+       if (errs == 0)
+               return IRQ_HANDLED;
+
+       /* Fault callback or TLB/PTE Dynamic loading */
+       if (!report_iommu_fault(domain, obj->dev, da, 0))
+               return IRQ_HANDLED;
+
+       iommu_disable(obj);
+
+       iopgd = iopgd_offset(obj, da);
+
+       if (!iopgd_is_table(*iopgd)) {
+               dev_err(obj->dev, "%s: errs:0x%08x da:0x%08x pgd:0x%p *pgd:px%08x\n",
+                               obj->name, errs, da, iopgd, *iopgd);
+               return IRQ_NONE;
+       }
+
+       iopte = iopte_offset(iopgd, da);
+
+       dev_err(obj->dev, "%s: errs:0x%08x da:0x%08x pgd:0x%p *pgd:0x%08x pte:0x%p *pte:0x%08x\n",
+                       obj->name, errs, da, iopgd, *iopgd, iopte, *iopte);
+
+       return IRQ_NONE;
+}
+
+static int device_match_by_alias(struct device *dev, void *data)
+{
+       struct omap_iommu *obj = to_iommu(dev);
+       const char *name = data;
+
+       pr_debug("%s: %s %s\n", __func__, obj->name, name);
+
+       return strcmp(obj->name, name) == 0;
+}
+
+/**
+ * omap_iommu_attach() - attach iommu device to an iommu domain
+ * @name:      name of target omap iommu device
+ * @iopgd:     page table
+ **/
+static struct omap_iommu *omap_iommu_attach(const char *name, u32 *iopgd)
+{
+       int err;
+       struct device *dev;
+       struct omap_iommu *obj;
+
+       dev = driver_find_device(&omap_iommu_driver.driver, NULL,
+                               (void *)name,
+                               device_match_by_alias);
+       if (!dev)
+               return ERR_PTR(-ENODEV);
+
+       obj = to_iommu(dev);
+
+       spin_lock(&obj->iommu_lock);
+
+       obj->iopgd = iopgd;
+       err = iommu_enable(obj);
+       if (err)
+               goto err_enable;
+       flush_iotlb_all(obj);
+
+       spin_unlock(&obj->iommu_lock);
+
+       dev_dbg(obj->dev, "%s: %s\n", __func__, obj->name);
+       return obj;
+
+err_enable:
+       spin_unlock(&obj->iommu_lock);
+       return ERR_PTR(err);
+}
+
+/**
+ * omap_iommu_detach - release iommu device
+ * @obj:       target iommu
+ **/
+static void omap_iommu_detach(struct omap_iommu *obj)
+{
+       if (!obj || IS_ERR(obj))
+               return;
+
+       spin_lock(&obj->iommu_lock);
+
+       iommu_disable(obj);
+       obj->iopgd = NULL;
+
+       spin_unlock(&obj->iommu_lock);
+
+       dev_dbg(obj->dev, "%s: %s\n", __func__, obj->name);
+}
+
+/*
+ *     OMAP Device MMU(IOMMU) detection
+ */
+static int omap_iommu_probe(struct platform_device *pdev)
+{
+       int err = -ENODEV;
+       int irq;
+       struct omap_iommu *obj;
+       struct resource *res;
+       struct iommu_platform_data *pdata = dev_get_platdata(&pdev->dev);
+       struct device_node *of = pdev->dev.of_node;
+
+       obj = devm_kzalloc(&pdev->dev, sizeof(*obj) + MMU_REG_SIZE, GFP_KERNEL);
+       if (!obj)
+               return -ENOMEM;
+
+       if (of) {
+               obj->name = dev_name(&pdev->dev);
+               obj->nr_tlb_entries = 32;
+               err = of_property_read_u32(of, "ti,#tlb-entries",
+                                          &obj->nr_tlb_entries);
+               if (err && err != -EINVAL)
+                       return err;
+               if (obj->nr_tlb_entries != 32 && obj->nr_tlb_entries != 8)
+                       return -EINVAL;
+               if (of_find_property(of, "ti,iommu-bus-err-back", NULL))
+                       obj->has_bus_err_back = MMU_GP_REG_BUS_ERR_BACK_EN;
+       } else {
+               obj->nr_tlb_entries = pdata->nr_tlb_entries;
+               obj->name = pdata->name;
+       }
+
+       obj->dev = &pdev->dev;
+       obj->ctx = (void *)obj + sizeof(*obj);
+
+       spin_lock_init(&obj->iommu_lock);
+       spin_lock_init(&obj->page_table_lock);
+
+       res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       obj->regbase = devm_ioremap_resource(obj->dev, res);
+       if (IS_ERR(obj->regbase))
+               return PTR_ERR(obj->regbase);
+
+       irq = platform_get_irq(pdev, 0);
+       if (irq < 0)
+               return -ENODEV;
+
+       err = devm_request_irq(obj->dev, irq, iommu_fault_handler, IRQF_SHARED,
+                              dev_name(obj->dev), obj);
+       if (err < 0)
+               return err;
+       platform_set_drvdata(pdev, obj);
+
+       pm_runtime_irq_safe(obj->dev);
+       pm_runtime_enable(obj->dev);
+
+       omap_iommu_debugfs_add(obj);
+
+       dev_info(&pdev->dev, "%s registered\n", obj->name);
+       return 0;
+}
+
+static int omap_iommu_remove(struct platform_device *pdev)
+{
+       struct omap_iommu *obj = platform_get_drvdata(pdev);
+
+       iopgtable_clear_entry_all(obj);
+       omap_iommu_debugfs_remove(obj);
+
+       pm_runtime_disable(obj->dev);
+
+       dev_info(&pdev->dev, "%s removed\n", obj->name);
+       return 0;
+}
+
+static const struct of_device_id omap_iommu_of_match[] = {
+       { .compatible = "ti,omap2-iommu" },
+       { .compatible = "ti,omap4-iommu" },
+       { .compatible = "ti,dra7-iommu" },
+       {},
+};
+MODULE_DEVICE_TABLE(of, omap_iommu_of_match);
+
+static struct platform_driver omap_iommu_driver = {
+       .probe  = omap_iommu_probe,
+       .remove = omap_iommu_remove,
+       .driver = {
+               .name   = "omap-iommu",
+               .of_match_table = of_match_ptr(omap_iommu_of_match),
+       },
+};
+
+static void iopte_cachep_ctor(void *iopte)
+{
+       clean_dcache_area(iopte, IOPTE_TABLE_SIZE);
+}
+
+static u32 iotlb_init_entry(struct iotlb_entry *e, u32 da, u32 pa, int pgsz)
+{
+       memset(e, 0, sizeof(*e));
+
+       e->da           = da;
+       e->pa           = pa;
+       e->valid        = MMU_CAM_V;
+       e->pgsz         = pgsz;
+       e->endian       = MMU_RAM_ENDIAN_LITTLE;
+       e->elsz         = MMU_RAM_ELSZ_8;
+       e->mixed        = 0;
+
+       return iopgsz_to_bytes(e->pgsz);
+}
+
+static int omap_iommu_map(struct iommu_domain *domain, unsigned long da,
+                        phys_addr_t pa, size_t bytes, int prot)
+{
+       struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
+       struct omap_iommu *oiommu = omap_domain->iommu_dev;
+       struct device *dev = oiommu->dev;
+       struct iotlb_entry e;
+       int omap_pgsz;
+       u32 ret;
+
+       omap_pgsz = bytes_to_iopgsz(bytes);
+       if (omap_pgsz < 0) {
+               dev_err(dev, "invalid size to map: %d\n", bytes);
+               return -EINVAL;
+       }
+
+       dev_dbg(dev, "mapping da 0x%lx to pa %pa size 0x%x\n", da, &pa, bytes);
+
+       iotlb_init_entry(&e, da, pa, omap_pgsz);
+
+       ret = omap_iopgtable_store_entry(oiommu, &e);
+       if (ret)
+               dev_err(dev, "omap_iopgtable_store_entry failed: %d\n", ret);
+
+       return ret;
+}
+
+static size_t omap_iommu_unmap(struct iommu_domain *domain, unsigned long da,
+                           size_t size)
+{
+       struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
+       struct omap_iommu *oiommu = omap_domain->iommu_dev;
+       struct device *dev = oiommu->dev;
+
+       dev_dbg(dev, "unmapping da 0x%lx size %u\n", da, size);
+
+       return iopgtable_clear_entry(oiommu, da);
+}
+
+static int
+omap_iommu_attach_dev(struct iommu_domain *domain, struct device *dev)
+{
+       struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
+       struct omap_iommu *oiommu;
+       struct omap_iommu_arch_data *arch_data = dev->archdata.iommu;
+       int ret = 0;
+
+       if (!arch_data || !arch_data->name) {
+               dev_err(dev, "device doesn't have an associated iommu\n");
+               return -EINVAL;
+       }
+
+       spin_lock(&omap_domain->lock);
+
+       /* only a single device is supported per domain for now */
+       if (omap_domain->iommu_dev) {
+               dev_err(dev, "iommu domain is already attached\n");
+               ret = -EBUSY;
+               goto out;
+       }
+
+       /* get a handle to and enable the omap iommu */
+       oiommu = omap_iommu_attach(arch_data->name, omap_domain->pgtable);
+       if (IS_ERR(oiommu)) {
+               ret = PTR_ERR(oiommu);
+               dev_err(dev, "can't get omap iommu: %d\n", ret);
+               goto out;
+       }
+
+       omap_domain->iommu_dev = arch_data->iommu_dev = oiommu;
+       omap_domain->dev = dev;
+       oiommu->domain = domain;
+
+out:
+       spin_unlock(&omap_domain->lock);
+       return ret;
+}
+
+static void _omap_iommu_detach_dev(struct omap_iommu_domain *omap_domain,
+                       struct device *dev)
+{
+       struct omap_iommu *oiommu = dev_to_omap_iommu(dev);
+       struct omap_iommu_arch_data *arch_data = dev->archdata.iommu;
+
+       /* only a single device is supported per domain for now */
+       if (omap_domain->iommu_dev != oiommu) {
+               dev_err(dev, "invalid iommu device\n");
+               return;
+       }
+
+       iopgtable_clear_entry_all(oiommu);
+
+       omap_iommu_detach(oiommu);
+
+       omap_domain->iommu_dev = arch_data->iommu_dev = NULL;
+       omap_domain->dev = NULL;
+       oiommu->domain = NULL;
+}
+
+static void omap_iommu_detach_dev(struct iommu_domain *domain,
+                                struct device *dev)
+{
+       struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
+
+       spin_lock(&omap_domain->lock);
+       _omap_iommu_detach_dev(omap_domain, dev);
+       spin_unlock(&omap_domain->lock);
+}
+
+static struct iommu_domain *omap_iommu_domain_alloc(unsigned type)
+{
+       struct omap_iommu_domain *omap_domain;
+
+       if (type != IOMMU_DOMAIN_UNMANAGED)
+               return NULL;
+
+       omap_domain = kzalloc(sizeof(*omap_domain), GFP_KERNEL);
+       if (!omap_domain) {
+               pr_err("kzalloc failed\n");
+               goto out;
+       }
+
+       omap_domain->pgtable = kzalloc(IOPGD_TABLE_SIZE, GFP_KERNEL);
+       if (!omap_domain->pgtable) {
+               pr_err("kzalloc failed\n");
+               goto fail_nomem;
+       }
+
+       /*
+        * should never fail, but please keep this around to ensure
+        * we keep the hardware happy
+        */
+       BUG_ON(!IS_ALIGNED((long)omap_domain->pgtable, IOPGD_TABLE_SIZE));
+
+       clean_dcache_area(omap_domain->pgtable, IOPGD_TABLE_SIZE);
+       spin_lock_init(&omap_domain->lock);
+
+       omap_domain->domain.geometry.aperture_start = 0;
+       omap_domain->domain.geometry.aperture_end   = (1ULL << 32) - 1;
+       omap_domain->domain.geometry.force_aperture = true;
+
+       return &omap_domain->domain;
+
+fail_nomem:
+       kfree(omap_domain);
+out:
+       return NULL;
+}
+
+static void omap_iommu_domain_free(struct iommu_domain *domain)
+{
+       struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
+
+       /*
+        * An iommu device is still attached
+        * (currently, only one device can be attached) ?
+        */
+       if (omap_domain->iommu_dev)
+               _omap_iommu_detach_dev(omap_domain, omap_domain->dev);
+
+       kfree(omap_domain->pgtable);
+       kfree(omap_domain);
+}
+
+static phys_addr_t omap_iommu_iova_to_phys(struct iommu_domain *domain,
+                                         dma_addr_t da)
+{
+       struct omap_iommu_domain *omap_domain = to_omap_domain(domain);
+       struct omap_iommu *oiommu = omap_domain->iommu_dev;
+       struct device *dev = oiommu->dev;
+       u32 *pgd, *pte;
+       phys_addr_t ret = 0;
+
+       iopgtable_lookup_entry(oiommu, da, &pgd, &pte);
+
+       if (pte) {
+               if (iopte_is_small(*pte))
+                       ret = omap_iommu_translate(*pte, da, IOPTE_MASK);
+               else if (iopte_is_large(*pte))
+                       ret = omap_iommu_translate(*pte, da, IOLARGE_MASK);
+               else
+                       dev_err(dev, "bogus pte 0x%x, da 0x%llx", *pte,
+                                                       (unsigned long long)da);
+       } else {
+               if (iopgd_is_section(*pgd))
+                       ret = omap_iommu_translate(*pgd, da, IOSECTION_MASK);
+               else if (iopgd_is_super(*pgd))
+                       ret = omap_iommu_translate(*pgd, da, IOSUPER_MASK);
+               else
+                       dev_err(dev, "bogus pgd 0x%x, da 0x%llx", *pgd,
+                                                       (unsigned long long)da);
+       }
+
+       return ret;
+}
+
+static int omap_iommu_add_device(struct device *dev)
+{
+       struct omap_iommu_arch_data *arch_data;
+       struct device_node *np;
+       struct platform_device *pdev;
+
+       /*
+        * Allocate the archdata iommu structure for DT-based devices.
+        *
+        * TODO: Simplify this when removing non-DT support completely from the
+        * IOMMU users.
+        */
+       if (!dev->of_node)
+               return 0;
+
+       np = of_parse_phandle(dev->of_node, "iommus", 0);
+       if (!np)
+               return 0;
+
+       pdev = of_find_device_by_node(np);
+       if (WARN_ON(!pdev)) {
+               of_node_put(np);
+               return -EINVAL;
+       }
+
+       arch_data = kzalloc(sizeof(*arch_data), GFP_KERNEL);
+       if (!arch_data) {
+               of_node_put(np);
+               return -ENOMEM;
+       }
+
+       arch_data->name = kstrdup(dev_name(&pdev->dev), GFP_KERNEL);
+       dev->archdata.iommu = arch_data;
+
+       of_node_put(np);
+
+       return 0;
+}
+
+static void omap_iommu_remove_device(struct device *dev)
+{
+       struct omap_iommu_arch_data *arch_data = dev->archdata.iommu;
+
+       if (!dev->of_node || !arch_data)
+               return;
+
+       kfree(arch_data->name);
+       kfree(arch_data);
+}
+
+static const struct iommu_ops omap_iommu_ops = {
+       .domain_alloc   = omap_iommu_domain_alloc,
+       .domain_free    = omap_iommu_domain_free,
+       .attach_dev     = omap_iommu_attach_dev,
+       .detach_dev     = omap_iommu_detach_dev,
+       .map            = omap_iommu_map,
+       .unmap          = omap_iommu_unmap,
+       .map_sg         = default_iommu_map_sg,
+       .iova_to_phys   = omap_iommu_iova_to_phys,
+       .add_device     = omap_iommu_add_device,
+       .remove_device  = omap_iommu_remove_device,
+       .pgsize_bitmap  = OMAP_IOMMU_PGSIZES,
+};
+
+static int __init omap_iommu_init(void)
+{
+       struct kmem_cache *p;
+       const unsigned long flags = SLAB_HWCACHE_ALIGN;
+       size_t align = 1 << 10; /* L2 pagetable alignement */
+       struct device_node *np;
+
+       np = of_find_matching_node(NULL, omap_iommu_of_match);
+       if (!np)
+               return 0;
+
+       of_node_put(np);
+
+       p = kmem_cache_create("iopte_cache", IOPTE_TABLE_SIZE, align, flags,
+                             iopte_cachep_ctor);
+       if (!p)
+               return -ENOMEM;
+       iopte_cachep = p;
+
+       bus_set_iommu(&platform_bus_type, &omap_iommu_ops);
+
+       omap_iommu_debugfs_init();
+
+       return platform_driver_register(&omap_iommu_driver);
+}
+/* must be ready before omap3isp is probed */
+subsys_initcall(omap_iommu_init);
+
+static void __exit omap_iommu_exit(void)
+{
+       kmem_cache_destroy(iopte_cachep);
+
+       platform_driver_unregister(&omap_iommu_driver);
+
+       omap_iommu_debugfs_exit();
+}
+module_exit(omap_iommu_exit);
+
+MODULE_DESCRIPTION("omap iommu: tlb and pagetable primitives");
+MODULE_ALIAS("platform:omap-iommu");
+MODULE_AUTHOR("Hiroshi DOYU, Paul Mundt and Toshihiro Kobayashi");
+MODULE_LICENSE("GPL v2");