Merge tag 'ntb-4.16' of git://github.com/jonmason/ntb
[sfrench/cifs-2.6.git] / drivers / tee / tee_shm.c
index 4bc7956cefc4afcde8e724f47346c2937e45e3e3..556960a1bab3b4631b10a2803ab54167b06c376b 100644 (file)
@@ -23,7 +23,6 @@
 static void tee_shm_release(struct tee_shm *shm)
 {
        struct tee_device *teedev = shm->teedev;
-       struct tee_shm_pool_mgr *poolm;
 
        mutex_lock(&teedev->mutex);
        idr_remove(&teedev->idr, shm->id);
@@ -31,12 +30,32 @@ static void tee_shm_release(struct tee_shm *shm)
                list_del(&shm->link);
        mutex_unlock(&teedev->mutex);
 
-       if (shm->flags & TEE_SHM_DMA_BUF)
-               poolm = &teedev->pool->dma_buf_mgr;
-       else
-               poolm = &teedev->pool->private_mgr;
+       if (shm->flags & TEE_SHM_POOL) {
+               struct tee_shm_pool_mgr *poolm;
+
+               if (shm->flags & TEE_SHM_DMA_BUF)
+                       poolm = teedev->pool->dma_buf_mgr;
+               else
+                       poolm = teedev->pool->private_mgr;
+
+               poolm->ops->free(poolm, shm);
+       } else if (shm->flags & TEE_SHM_REGISTER) {
+               size_t n;
+               int rc = teedev->desc->ops->shm_unregister(shm->ctx, shm);
+
+               if (rc)
+                       dev_err(teedev->dev.parent,
+                               "unregister shm %p failed: %d", shm, rc);
+
+               for (n = 0; n < shm->num_pages; n++)
+                       put_page(shm->pages[n]);
+
+               kfree(shm->pages);
+       }
+
+       if (shm->ctx)
+               teedev_ctx_put(shm->ctx);
 
-       poolm->ops->free(poolm, shm);
        kfree(shm);
 
        tee_device_put(teedev);
@@ -76,6 +95,10 @@ static int tee_shm_op_mmap(struct dma_buf *dmabuf, struct vm_area_struct *vma)
        struct tee_shm *shm = dmabuf->priv;
        size_t size = vma->vm_end - vma->vm_start;
 
+       /* Refuse sharing shared memory provided by application */
+       if (shm->flags & TEE_SHM_REGISTER)
+               return -EINVAL;
+
        return remap_pfn_range(vma, vma->vm_start, shm->paddr >> PAGE_SHIFT,
                               size, vma->vm_page_prot);
 }
@@ -89,26 +112,20 @@ static const struct dma_buf_ops tee_shm_dma_buf_ops = {
        .mmap = tee_shm_op_mmap,
 };
 
-/**
- * tee_shm_alloc() - Allocate shared memory
- * @ctx:       Context that allocates the shared memory
- * @size:      Requested size of shared memory
- * @flags:     Flags setting properties for the requested shared memory.
- *
- * Memory allocated as global shared memory is automatically freed when the
- * TEE file pointer is closed. The @flags field uses the bits defined by
- * TEE_SHM_* in <linux/tee_drv.h>. TEE_SHM_MAPPED must currently always be
- * set. If TEE_SHM_DMA_BUF global shared memory will be allocated and
- * associated with a dma-buf handle, else driver private memory.
- */
-struct tee_shm *tee_shm_alloc(struct tee_context *ctx, size_t size, u32 flags)
+static struct tee_shm *__tee_shm_alloc(struct tee_context *ctx,
+                                      struct tee_device *teedev,
+                                      size_t size, u32 flags)
 {
-       struct tee_device *teedev = ctx->teedev;
        struct tee_shm_pool_mgr *poolm = NULL;
        struct tee_shm *shm;
        void *ret;
        int rc;
 
+       if (ctx && ctx->teedev != teedev) {
+               dev_err(teedev->dev.parent, "ctx and teedev mismatch\n");
+               return ERR_PTR(-EINVAL);
+       }
+
        if (!(flags & TEE_SHM_MAPPED)) {
                dev_err(teedev->dev.parent,
                        "only mapped allocations supported\n");
@@ -135,13 +152,13 @@ struct tee_shm *tee_shm_alloc(struct tee_context *ctx, size_t size, u32 flags)
                goto err_dev_put;
        }
 
-       shm->flags = flags;
+       shm->flags = flags | TEE_SHM_POOL;
        shm->teedev = teedev;
        shm->ctx = ctx;
        if (flags & TEE_SHM_DMA_BUF)
-               poolm = &teedev->pool->dma_buf_mgr;
+               poolm = teedev->pool->dma_buf_mgr;
        else
-               poolm = &teedev->pool->private_mgr;
+               poolm = teedev->pool->private_mgr;
 
        rc = poolm->ops->alloc(poolm, shm, size);
        if (rc) {
@@ -171,9 +188,13 @@ struct tee_shm *tee_shm_alloc(struct tee_context *ctx, size_t size, u32 flags)
                        goto err_rem;
                }
        }
-       mutex_lock(&teedev->mutex);
-       list_add_tail(&shm->link, &ctx->list_shm);
-       mutex_unlock(&teedev->mutex);
+
+       if (ctx) {
+               teedev_ctx_get(ctx);
+               mutex_lock(&teedev->mutex);
+               list_add_tail(&shm->link, &ctx->list_shm);
+               mutex_unlock(&teedev->mutex);
+       }
 
        return shm;
 err_rem:
@@ -188,8 +209,145 @@ err_dev_put:
        tee_device_put(teedev);
        return ret;
 }
+
+/**
+ * tee_shm_alloc() - Allocate shared memory
+ * @ctx:       Context that allocates the shared memory
+ * @size:      Requested size of shared memory
+ * @flags:     Flags setting properties for the requested shared memory.
+ *
+ * Memory allocated as global shared memory is automatically freed when the
+ * TEE file pointer is closed. The @flags field uses the bits defined by
+ * TEE_SHM_* in <linux/tee_drv.h>. TEE_SHM_MAPPED must currently always be
+ * set. If TEE_SHM_DMA_BUF global shared memory will be allocated and
+ * associated with a dma-buf handle, else driver private memory.
+ */
+struct tee_shm *tee_shm_alloc(struct tee_context *ctx, size_t size, u32 flags)
+{
+       return __tee_shm_alloc(ctx, ctx->teedev, size, flags);
+}
 EXPORT_SYMBOL_GPL(tee_shm_alloc);
 
+struct tee_shm *tee_shm_priv_alloc(struct tee_device *teedev, size_t size)
+{
+       return __tee_shm_alloc(NULL, teedev, size, TEE_SHM_MAPPED);
+}
+EXPORT_SYMBOL_GPL(tee_shm_priv_alloc);
+
+struct tee_shm *tee_shm_register(struct tee_context *ctx, unsigned long addr,
+                                size_t length, u32 flags)
+{
+       struct tee_device *teedev = ctx->teedev;
+       const u32 req_flags = TEE_SHM_DMA_BUF | TEE_SHM_USER_MAPPED;
+       struct tee_shm *shm;
+       void *ret;
+       int rc;
+       int num_pages;
+       unsigned long start;
+
+       if (flags != req_flags)
+               return ERR_PTR(-ENOTSUPP);
+
+       if (!tee_device_get(teedev))
+               return ERR_PTR(-EINVAL);
+
+       if (!teedev->desc->ops->shm_register ||
+           !teedev->desc->ops->shm_unregister) {
+               tee_device_put(teedev);
+               return ERR_PTR(-ENOTSUPP);
+       }
+
+       teedev_ctx_get(ctx);
+
+       shm = kzalloc(sizeof(*shm), GFP_KERNEL);
+       if (!shm) {
+               ret = ERR_PTR(-ENOMEM);
+               goto err;
+       }
+
+       shm->flags = flags | TEE_SHM_REGISTER;
+       shm->teedev = teedev;
+       shm->ctx = ctx;
+       shm->id = -1;
+       start = rounddown(addr, PAGE_SIZE);
+       shm->offset = addr - start;
+       shm->size = length;
+       num_pages = (roundup(addr + length, PAGE_SIZE) - start) / PAGE_SIZE;
+       shm->pages = kcalloc(num_pages, sizeof(*shm->pages), GFP_KERNEL);
+       if (!shm->pages) {
+               ret = ERR_PTR(-ENOMEM);
+               goto err;
+       }
+
+       rc = get_user_pages_fast(start, num_pages, 1, shm->pages);
+       if (rc > 0)
+               shm->num_pages = rc;
+       if (rc != num_pages) {
+               if (rc >= 0)
+                       rc = -ENOMEM;
+               ret = ERR_PTR(rc);
+               goto err;
+       }
+
+       mutex_lock(&teedev->mutex);
+       shm->id = idr_alloc(&teedev->idr, shm, 1, 0, GFP_KERNEL);
+       mutex_unlock(&teedev->mutex);
+
+       if (shm->id < 0) {
+               ret = ERR_PTR(shm->id);
+               goto err;
+       }
+
+       rc = teedev->desc->ops->shm_register(ctx, shm, shm->pages,
+                                            shm->num_pages, start);
+       if (rc) {
+               ret = ERR_PTR(rc);
+               goto err;
+       }
+
+       if (flags & TEE_SHM_DMA_BUF) {
+               DEFINE_DMA_BUF_EXPORT_INFO(exp_info);
+
+               exp_info.ops = &tee_shm_dma_buf_ops;
+               exp_info.size = shm->size;
+               exp_info.flags = O_RDWR;
+               exp_info.priv = shm;
+
+               shm->dmabuf = dma_buf_export(&exp_info);
+               if (IS_ERR(shm->dmabuf)) {
+                       ret = ERR_CAST(shm->dmabuf);
+                       teedev->desc->ops->shm_unregister(ctx, shm);
+                       goto err;
+               }
+       }
+
+       mutex_lock(&teedev->mutex);
+       list_add_tail(&shm->link, &ctx->list_shm);
+       mutex_unlock(&teedev->mutex);
+
+       return shm;
+err:
+       if (shm) {
+               size_t n;
+
+               if (shm->id >= 0) {
+                       mutex_lock(&teedev->mutex);
+                       idr_remove(&teedev->idr, shm->id);
+                       mutex_unlock(&teedev->mutex);
+               }
+               if (shm->pages) {
+                       for (n = 0; n < shm->num_pages; n++)
+                               put_page(shm->pages[n]);
+                       kfree(shm->pages);
+               }
+       }
+       kfree(shm);
+       teedev_ctx_put(ctx);
+       tee_device_put(teedev);
+       return ret;
+}
+EXPORT_SYMBOL_GPL(tee_shm_register);
+
 /**
  * tee_shm_get_fd() - Increase reference count and return file descriptor
  * @shm:       Shared memory handle
@@ -197,10 +355,9 @@ EXPORT_SYMBOL_GPL(tee_shm_alloc);
  */
 int tee_shm_get_fd(struct tee_shm *shm)
 {
-       u32 req_flags = TEE_SHM_MAPPED | TEE_SHM_DMA_BUF;
        int fd;
 
-       if ((shm->flags & req_flags) != req_flags)
+       if (!(shm->flags & TEE_SHM_DMA_BUF))
                return -EINVAL;
 
        fd = dma_buf_fd(shm->dmabuf, O_CLOEXEC);
@@ -238,6 +395,8 @@ EXPORT_SYMBOL_GPL(tee_shm_free);
  */
 int tee_shm_va2pa(struct tee_shm *shm, void *va, phys_addr_t *pa)
 {
+       if (!(shm->flags & TEE_SHM_MAPPED))
+               return -EINVAL;
        /* Check that we're in the range of the shm */
        if ((char *)va < (char *)shm->kaddr)
                return -EINVAL;
@@ -258,6 +417,8 @@ EXPORT_SYMBOL_GPL(tee_shm_va2pa);
  */
 int tee_shm_pa2va(struct tee_shm *shm, phys_addr_t pa, void **va)
 {
+       if (!(shm->flags & TEE_SHM_MAPPED))
+               return -EINVAL;
        /* Check that we're in the range of the shm */
        if (pa < shm->paddr)
                return -EINVAL;
@@ -284,6 +445,8 @@ EXPORT_SYMBOL_GPL(tee_shm_pa2va);
  */
 void *tee_shm_get_va(struct tee_shm *shm, size_t offs)
 {
+       if (!(shm->flags & TEE_SHM_MAPPED))
+               return ERR_PTR(-EINVAL);
        if (offs >= shm->size)
                return ERR_PTR(-EINVAL);
        return (char *)shm->kaddr + offs;
@@ -335,17 +498,6 @@ struct tee_shm *tee_shm_get_from_id(struct tee_context *ctx, int id)
 }
 EXPORT_SYMBOL_GPL(tee_shm_get_from_id);
 
-/**
- * tee_shm_get_id() - Get id of a shared memory object
- * @shm:       Shared memory handle
- * @returns id
- */
-int tee_shm_get_id(struct tee_shm *shm)
-{
-       return shm->id;
-}
-EXPORT_SYMBOL_GPL(tee_shm_get_id);
-
 /**
  * tee_shm_put() - Decrease reference count on a shared memory handle
  * @shm:       Shared memory handle