Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / iommu / shmobile-iommu.c
diff --git a/kernel/drivers/iommu/shmobile-iommu.c b/kernel/drivers/iommu/shmobile-iommu.c
new file mode 100644 (file)
index 0000000..a028751
--- /dev/null
@@ -0,0 +1,402 @@
+/*
+ * IOMMU for IPMMU/IPMMUI
+ * Copyright (C) 2012  Hideki EIRAKU
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ */
+
+#include <linux/dma-mapping.h>
+#include <linux/io.h>
+#include <linux/iommu.h>
+#include <linux/platform_device.h>
+#include <linux/sizes.h>
+#include <linux/slab.h>
+#include <asm/dma-iommu.h>
+#include "shmobile-ipmmu.h"
+
+#define L1_SIZE CONFIG_SHMOBILE_IOMMU_L1SIZE
+#define L1_LEN (L1_SIZE / 4)
+#define L1_ALIGN L1_SIZE
+#define L2_SIZE SZ_1K
+#define L2_LEN (L2_SIZE / 4)
+#define L2_ALIGN L2_SIZE
+
+struct shmobile_iommu_domain_pgtable {
+       uint32_t *pgtable;
+       dma_addr_t handle;
+};
+
+struct shmobile_iommu_archdata {
+       struct list_head attached_list;
+       struct dma_iommu_mapping *iommu_mapping;
+       spinlock_t attach_lock;
+       struct shmobile_iommu_domain *attached;
+       int num_attached_devices;
+       struct shmobile_ipmmu *ipmmu;
+};
+
+struct shmobile_iommu_domain {
+       struct shmobile_iommu_domain_pgtable l1, l2[L1_LEN];
+       spinlock_t map_lock;
+       spinlock_t attached_list_lock;
+       struct list_head attached_list;
+       struct iommu_domain domain;
+};
+
+static struct shmobile_iommu_archdata *ipmmu_archdata;
+static struct kmem_cache *l1cache, *l2cache;
+
+static struct shmobile_iommu_domain *to_sh_domain(struct iommu_domain *dom)
+{
+       return container_of(dom, struct shmobile_iommu_domain, domain);
+}
+
+static int pgtable_alloc(struct shmobile_iommu_domain_pgtable *pgtable,
+                        struct kmem_cache *cache, size_t size)
+{
+       pgtable->pgtable = kmem_cache_zalloc(cache, GFP_ATOMIC);
+       if (!pgtable->pgtable)
+               return -ENOMEM;
+       pgtable->handle = dma_map_single(NULL, pgtable->pgtable, size,
+                                        DMA_TO_DEVICE);
+       return 0;
+}
+
+static void pgtable_free(struct shmobile_iommu_domain_pgtable *pgtable,
+                        struct kmem_cache *cache, size_t size)
+{
+       dma_unmap_single(NULL, pgtable->handle, size, DMA_TO_DEVICE);
+       kmem_cache_free(cache, pgtable->pgtable);
+}
+
+static uint32_t pgtable_read(struct shmobile_iommu_domain_pgtable *pgtable,
+                            unsigned int index)
+{
+       return pgtable->pgtable[index];
+}
+
+static void pgtable_write(struct shmobile_iommu_domain_pgtable *pgtable,
+                         unsigned int index, unsigned int count, uint32_t val)
+{
+       unsigned int i;
+
+       for (i = 0; i < count; i++)
+               pgtable->pgtable[index + i] = val;
+       dma_sync_single_for_device(NULL, pgtable->handle + index * sizeof(val),
+                                  sizeof(val) * count, DMA_TO_DEVICE);
+}
+
+static struct iommu_domain *shmobile_iommu_domain_alloc(unsigned type)
+{
+       struct shmobile_iommu_domain *sh_domain;
+       int i, ret;
+
+       if (type != IOMMU_DOMAIN_UNMANAGED)
+               return NULL;
+
+       sh_domain = kzalloc(sizeof(*sh_domain), GFP_KERNEL);
+       if (!sh_domain)
+               return NULL;
+       ret = pgtable_alloc(&sh_domain->l1, l1cache, L1_SIZE);
+       if (ret < 0) {
+               kfree(sh_domain);
+               return NULL;
+       }
+       for (i = 0; i < L1_LEN; i++)
+               sh_domain->l2[i].pgtable = NULL;
+       spin_lock_init(&sh_domain->map_lock);
+       spin_lock_init(&sh_domain->attached_list_lock);
+       INIT_LIST_HEAD(&sh_domain->attached_list);
+       return &sh_domain->domain;
+}
+
+static void shmobile_iommu_domain_free(struct iommu_domain *domain)
+{
+       struct shmobile_iommu_domain *sh_domain = to_sh_domain(domain);
+       int i;
+
+       for (i = 0; i < L1_LEN; i++) {
+               if (sh_domain->l2[i].pgtable)
+                       pgtable_free(&sh_domain->l2[i], l2cache, L2_SIZE);
+       }
+       pgtable_free(&sh_domain->l1, l1cache, L1_SIZE);
+       kfree(sh_domain);
+}
+
+static int shmobile_iommu_attach_device(struct iommu_domain *domain,
+                                       struct device *dev)
+{
+       struct shmobile_iommu_archdata *archdata = dev->archdata.iommu;
+       struct shmobile_iommu_domain *sh_domain = to_sh_domain(domain);
+       int ret = -EBUSY;
+
+       if (!archdata)
+               return -ENODEV;
+       spin_lock(&sh_domain->attached_list_lock);
+       spin_lock(&archdata->attach_lock);
+       if (archdata->attached != sh_domain) {
+               if (archdata->attached)
+                       goto err;
+               ipmmu_tlb_set(archdata->ipmmu, sh_domain->l1.handle, L1_SIZE,
+                             0);
+               ipmmu_tlb_flush(archdata->ipmmu);
+               archdata->attached = sh_domain;
+               archdata->num_attached_devices = 0;
+               list_add(&archdata->attached_list, &sh_domain->attached_list);
+       }
+       archdata->num_attached_devices++;
+       ret = 0;
+err:
+       spin_unlock(&archdata->attach_lock);
+       spin_unlock(&sh_domain->attached_list_lock);
+       return ret;
+}
+
+static void shmobile_iommu_detach_device(struct iommu_domain *domain,
+                                        struct device *dev)
+{
+       struct shmobile_iommu_archdata *archdata = dev->archdata.iommu;
+       struct shmobile_iommu_domain *sh_domain = to_sh_domain(domain);
+
+       if (!archdata)
+               return;
+       spin_lock(&sh_domain->attached_list_lock);
+       spin_lock(&archdata->attach_lock);
+       archdata->num_attached_devices--;
+       if (!archdata->num_attached_devices) {
+               ipmmu_tlb_set(archdata->ipmmu, 0, 0, 0);
+               ipmmu_tlb_flush(archdata->ipmmu);
+               archdata->attached = NULL;
+               list_del(&archdata->attached_list);
+       }
+       spin_unlock(&archdata->attach_lock);
+       spin_unlock(&sh_domain->attached_list_lock);
+}
+
+static void domain_tlb_flush(struct shmobile_iommu_domain *sh_domain)
+{
+       struct shmobile_iommu_archdata *archdata;
+
+       spin_lock(&sh_domain->attached_list_lock);
+       list_for_each_entry(archdata, &sh_domain->attached_list, attached_list)
+               ipmmu_tlb_flush(archdata->ipmmu);
+       spin_unlock(&sh_domain->attached_list_lock);
+}
+
+static int l2alloc(struct shmobile_iommu_domain *sh_domain,
+                  unsigned int l1index)
+{
+       int ret;
+
+       if (!sh_domain->l2[l1index].pgtable) {
+               ret = pgtable_alloc(&sh_domain->l2[l1index], l2cache, L2_SIZE);
+               if (ret < 0)
+                       return ret;
+       }
+       pgtable_write(&sh_domain->l1, l1index, 1,
+                     sh_domain->l2[l1index].handle | 0x1);
+       return 0;
+}
+
+static void l2realfree(struct shmobile_iommu_domain_pgtable *l2)
+{
+       if (l2->pgtable)
+               pgtable_free(l2, l2cache, L2_SIZE);
+}
+
+static void l2free(struct shmobile_iommu_domain *sh_domain,
+                  unsigned int l1index,
+                  struct shmobile_iommu_domain_pgtable *l2)
+{
+       pgtable_write(&sh_domain->l1, l1index, 1, 0);
+       if (sh_domain->l2[l1index].pgtable) {
+               *l2 = sh_domain->l2[l1index];
+               sh_domain->l2[l1index].pgtable = NULL;
+       }
+}
+
+static int shmobile_iommu_map(struct iommu_domain *domain, unsigned long iova,
+                             phys_addr_t paddr, size_t size, int prot)
+{
+       struct shmobile_iommu_domain_pgtable l2 = { .pgtable = NULL };
+       struct shmobile_iommu_domain *sh_domain = to_sh_domain(domain);
+       unsigned int l1index, l2index;
+       int ret;
+
+       l1index = iova >> 20;
+       switch (size) {
+       case SZ_4K:
+               l2index = (iova >> 12) & 0xff;
+               spin_lock(&sh_domain->map_lock);
+               ret = l2alloc(sh_domain, l1index);
+               if (!ret)
+                       pgtable_write(&sh_domain->l2[l1index], l2index, 1,
+                                     paddr | 0xff2);
+               spin_unlock(&sh_domain->map_lock);
+               break;
+       case SZ_64K:
+               l2index = (iova >> 12) & 0xf0;
+               spin_lock(&sh_domain->map_lock);
+               ret = l2alloc(sh_domain, l1index);
+               if (!ret)
+                       pgtable_write(&sh_domain->l2[l1index], l2index, 0x10,
+                                     paddr | 0xff1);
+               spin_unlock(&sh_domain->map_lock);
+               break;
+       case SZ_1M:
+               spin_lock(&sh_domain->map_lock);
+               l2free(sh_domain, l1index, &l2);
+               pgtable_write(&sh_domain->l1, l1index, 1, paddr | 0xc02);
+               spin_unlock(&sh_domain->map_lock);
+               ret = 0;
+               break;
+       default:
+               ret = -EINVAL;
+       }
+       if (!ret)
+               domain_tlb_flush(sh_domain);
+       l2realfree(&l2);
+       return ret;
+}
+
+static size_t shmobile_iommu_unmap(struct iommu_domain *domain,
+                                  unsigned long iova, size_t size)
+{
+       struct shmobile_iommu_domain_pgtable l2 = { .pgtable = NULL };
+       struct shmobile_iommu_domain *sh_domain = to_sh_domain(domain);
+       unsigned int l1index, l2index;
+       uint32_t l2entry = 0;
+       size_t ret = 0;
+
+       l1index = iova >> 20;
+       if (!(iova & 0xfffff) && size >= SZ_1M) {
+               spin_lock(&sh_domain->map_lock);
+               l2free(sh_domain, l1index, &l2);
+               spin_unlock(&sh_domain->map_lock);
+               ret = SZ_1M;
+               goto done;
+       }
+       l2index = (iova >> 12) & 0xff;
+       spin_lock(&sh_domain->map_lock);
+       if (sh_domain->l2[l1index].pgtable)
+               l2entry = pgtable_read(&sh_domain->l2[l1index], l2index);
+       switch (l2entry & 3) {
+       case 1:
+               if (l2index & 0xf)
+                       break;
+               pgtable_write(&sh_domain->l2[l1index], l2index, 0x10, 0);
+               ret = SZ_64K;
+               break;
+       case 2:
+               pgtable_write(&sh_domain->l2[l1index], l2index, 1, 0);
+               ret = SZ_4K;
+               break;
+       }
+       spin_unlock(&sh_domain->map_lock);
+done:
+       if (ret)
+               domain_tlb_flush(sh_domain);
+       l2realfree(&l2);
+       return ret;
+}
+
+static phys_addr_t shmobile_iommu_iova_to_phys(struct iommu_domain *domain,
+                                              dma_addr_t iova)
+{
+       struct shmobile_iommu_domain *sh_domain = to_sh_domain(domain);
+       uint32_t l1entry = 0, l2entry = 0;
+       unsigned int l1index, l2index;
+
+       l1index = iova >> 20;
+       l2index = (iova >> 12) & 0xff;
+       spin_lock(&sh_domain->map_lock);
+       if (sh_domain->l2[l1index].pgtable)
+               l2entry = pgtable_read(&sh_domain->l2[l1index], l2index);
+       else
+               l1entry = pgtable_read(&sh_domain->l1, l1index);
+       spin_unlock(&sh_domain->map_lock);
+       switch (l2entry & 3) {
+       case 1:
+               return (l2entry & ~0xffff) | (iova & 0xffff);
+       case 2:
+               return (l2entry & ~0xfff) | (iova & 0xfff);
+       default:
+               if ((l1entry & 3) == 2)
+                       return (l1entry & ~0xfffff) | (iova & 0xfffff);
+               return 0;
+       }
+}
+
+static int find_dev_name(struct shmobile_ipmmu *ipmmu, const char *dev_name)
+{
+       unsigned int i, n = ipmmu->num_dev_names;
+
+       for (i = 0; i < n; i++) {
+               if (strcmp(ipmmu->dev_names[i], dev_name) == 0)
+                       return 1;
+       }
+       return 0;
+}
+
+static int shmobile_iommu_add_device(struct device *dev)
+{
+       struct shmobile_iommu_archdata *archdata = ipmmu_archdata;
+       struct dma_iommu_mapping *mapping;
+
+       if (!find_dev_name(archdata->ipmmu, dev_name(dev)))
+               return 0;
+       mapping = archdata->iommu_mapping;
+       if (!mapping) {
+               mapping = arm_iommu_create_mapping(&platform_bus_type, 0,
+                                                  L1_LEN << 20);
+               if (IS_ERR(mapping))
+                       return PTR_ERR(mapping);
+               archdata->iommu_mapping = mapping;
+       }
+       dev->archdata.iommu = archdata;
+       if (arm_iommu_attach_device(dev, mapping))
+               pr_err("arm_iommu_attach_device failed\n");
+       return 0;
+}
+
+static const struct iommu_ops shmobile_iommu_ops = {
+       .domain_alloc = shmobile_iommu_domain_alloc,
+       .domain_free = shmobile_iommu_domain_free,
+       .attach_dev = shmobile_iommu_attach_device,
+       .detach_dev = shmobile_iommu_detach_device,
+       .map = shmobile_iommu_map,
+       .unmap = shmobile_iommu_unmap,
+       .map_sg = default_iommu_map_sg,
+       .iova_to_phys = shmobile_iommu_iova_to_phys,
+       .add_device = shmobile_iommu_add_device,
+       .pgsize_bitmap = SZ_1M | SZ_64K | SZ_4K,
+};
+
+int ipmmu_iommu_init(struct shmobile_ipmmu *ipmmu)
+{
+       static struct shmobile_iommu_archdata *archdata;
+
+       l1cache = kmem_cache_create("shmobile-iommu-pgtable1", L1_SIZE,
+                                   L1_ALIGN, SLAB_HWCACHE_ALIGN, NULL);
+       if (!l1cache)
+               return -ENOMEM;
+       l2cache = kmem_cache_create("shmobile-iommu-pgtable2", L2_SIZE,
+                                   L2_ALIGN, SLAB_HWCACHE_ALIGN, NULL);
+       if (!l2cache) {
+               kmem_cache_destroy(l1cache);
+               return -ENOMEM;
+       }
+       archdata = kzalloc(sizeof(*archdata), GFP_KERNEL);
+       if (!archdata) {
+               kmem_cache_destroy(l1cache);
+               kmem_cache_destroy(l2cache);
+               return -ENOMEM;
+       }
+       spin_lock_init(&archdata->attach_lock);
+       archdata->ipmmu = ipmmu;
+       ipmmu_archdata = archdata;
+       bus_set_iommu(&platform_bus_type, &shmobile_iommu_ops);
+       return 0;
+}