Merge tag 'armsoc-tee' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
authorLinus Torvalds <torvalds@linux-foundation.org>
Wed, 10 May 2017 18:20:09 +0000 (11:20 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 10 May 2017 18:20:09 +0000 (11:20 -0700)
Pull TEE driver infrastructure and OP-TEE drivers from Arnd Bergmann:
 "This introduces a generic TEE framework in the kernel, to handle
  trusted environemtns (security coprocessor or software implementations
  such as OP-TEE/TrustZone). I'm sending it separately from the other
  arm-soc driver changes to give it a little more visibility, once the
  subsystem is merged, we will likely keep this in the armâ‚‹soc drivers
  branch or have the maintainers submit pull requests directly,
  depending on the patch volume.

  I have reviewed earlier versions in the past, and have reviewed the
  latest version in person during Linaro Connect BUD17.

  Here is my overall assessment of the subsystem:

   - There is clearly demand for this, both for the generic
     infrastructure and the specific OP-TEE implementation.

   - The code has gone through a large number of reviews, and the review
     comments have all been addressed, but the reviews were not coming
     up with serious issues any more and nobody volunteered to vouch for
     the quality.

   - The user space ioctl interface is sufficient to work with the
     OP-TEE driver, and it should in principle work with other TEE
     implementations that follow the GlobalPlatform[1] standards, but it
     might need to be extended in minor ways depending on specific
     requirements of future TEE implementations

   - The main downside of the API to me is how the user space is tied to
     the TEE implementation in hardware or firmware, but uses a generic
     way to communicate with it. This seems to be an inherent problem
     with what it is trying to do, and I could not come up with any
     better solution than what is implemented here.

  For a detailed history of the patch series, see

    https://lkml.org/lkml/2017/3/10/1277"

* tag 'armsoc-tee' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc:
  arm64: dt: hikey: Add optee node
  Documentation: tee subsystem and op-tee driver
  tee: add OP-TEE driver
  tee: generic TEE subsystem
  dt/bindings: add bindings for optee

1  2 
Documentation/devicetree/bindings/vendor-prefixes.txt
Documentation/ioctl/ioctl-number.txt
MAINTAINERS
arch/arm64/boot/dts/hisilicon/hi6220-hikey.dts
drivers/Makefile
drivers/tee/tee_shm.c

Simple merge
diff --cc MAINTAINERS
Simple merge
Simple merge
index 0000000,0be1e3e..d356d7f
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,358 +1,358 @@@
 -static void *tee_shm_op_kmap_atomic(struct dma_buf *dmabuf, unsigned long pgnum)
+ /*
+  * Copyright (c) 2015-2016, Linaro Limited
+  *
+  * This software is licensed under the terms of the GNU General Public
+  * License version 2, as published by the Free Software Foundation, and
+  * may be copied, distributed, and modified under those terms.
+  *
+  * This program is distributed in the hope that 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.
+  *
+  */
+ #include <linux/device.h>
+ #include <linux/dma-buf.h>
+ #include <linux/fdtable.h>
+ #include <linux/idr.h>
+ #include <linux/sched.h>
+ #include <linux/slab.h>
+ #include <linux/tee_drv.h>
+ #include "tee_private.h"
+ 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);
+       if (shm->ctx)
+               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;
+       poolm->ops->free(poolm, shm);
+       kfree(shm);
+       tee_device_put(teedev);
+ }
+ static struct sg_table *tee_shm_op_map_dma_buf(struct dma_buf_attachment
+                       *attach, enum dma_data_direction dir)
+ {
+       return NULL;
+ }
+ static void tee_shm_op_unmap_dma_buf(struct dma_buf_attachment *attach,
+                                    struct sg_table *table,
+                                    enum dma_data_direction dir)
+ {
+ }
+ static void tee_shm_op_release(struct dma_buf *dmabuf)
+ {
+       struct tee_shm *shm = dmabuf->priv;
+       tee_shm_release(shm);
+ }
 -static void *tee_shm_op_kmap(struct dma_buf *dmabuf, unsigned long pgnum)
++static void *tee_shm_op_map_atomic(struct dma_buf *dmabuf, unsigned long pgnum)
+ {
+       return NULL;
+ }
 -      .kmap_atomic = tee_shm_op_kmap_atomic,
 -      .kmap = tee_shm_op_kmap,
++static void *tee_shm_op_map(struct dma_buf *dmabuf, unsigned long pgnum)
+ {
+       return NULL;
+ }
+ 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;
+       return remap_pfn_range(vma, vma->vm_start, shm->paddr >> PAGE_SHIFT,
+                              size, vma->vm_page_prot);
+ }
+ static struct dma_buf_ops tee_shm_dma_buf_ops = {
+       .map_dma_buf = tee_shm_op_map_dma_buf,
+       .unmap_dma_buf = tee_shm_op_unmap_dma_buf,
+       .release = tee_shm_op_release,
++      .map_atomic = tee_shm_op_map_atomic,
++      .map = tee_shm_op_map,
+       .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)
+ {
+       struct tee_device *teedev = ctx->teedev;
+       struct tee_shm_pool_mgr *poolm = NULL;
+       struct tee_shm *shm;
+       void *ret;
+       int rc;
+       if (!(flags & TEE_SHM_MAPPED)) {
+               dev_err(teedev->dev.parent,
+                       "only mapped allocations supported\n");
+               return ERR_PTR(-EINVAL);
+       }
+       if ((flags & ~(TEE_SHM_MAPPED | TEE_SHM_DMA_BUF))) {
+               dev_err(teedev->dev.parent, "invalid shm flags 0x%x", flags);
+               return ERR_PTR(-EINVAL);
+       }
+       if (!tee_device_get(teedev))
+               return ERR_PTR(-EINVAL);
+       if (!teedev->pool) {
+               /* teedev has been detached from driver */
+               ret = ERR_PTR(-EINVAL);
+               goto err_dev_put;
+       }
+       shm = kzalloc(sizeof(*shm), GFP_KERNEL);
+       if (!shm) {
+               ret = ERR_PTR(-ENOMEM);
+               goto err_dev_put;
+       }
+       shm->flags = flags;
+       shm->teedev = teedev;
+       shm->ctx = ctx;
+       if (flags & TEE_SHM_DMA_BUF)
+               poolm = &teedev->pool->dma_buf_mgr;
+       else
+               poolm = &teedev->pool->private_mgr;
+       rc = poolm->ops->alloc(poolm, shm, size);
+       if (rc) {
+               ret = ERR_PTR(rc);
+               goto err_kfree;
+       }
+       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_pool_free;
+       }
+       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);
+                       goto err_rem;
+               }
+       }
+       mutex_lock(&teedev->mutex);
+       list_add_tail(&shm->link, &ctx->list_shm);
+       mutex_unlock(&teedev->mutex);
+       return shm;
+ err_rem:
+       mutex_lock(&teedev->mutex);
+       idr_remove(&teedev->idr, shm->id);
+       mutex_unlock(&teedev->mutex);
+ err_pool_free:
+       poolm->ops->free(poolm, shm);
+ err_kfree:
+       kfree(shm);
+ err_dev_put:
+       tee_device_put(teedev);
+       return ret;
+ }
+ EXPORT_SYMBOL_GPL(tee_shm_alloc);
+ /**
+  * tee_shm_get_fd() - Increase reference count and return file descriptor
+  * @shm:      Shared memory handle
+  * @returns user space file descriptor to shared memory
+  */
+ 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)
+               return -EINVAL;
+       fd = dma_buf_fd(shm->dmabuf, O_CLOEXEC);
+       if (fd >= 0)
+               get_dma_buf(shm->dmabuf);
+       return fd;
+ }
+ /**
+  * tee_shm_free() - Free shared memory
+  * @shm:      Handle to shared memory to free
+  */
+ void tee_shm_free(struct tee_shm *shm)
+ {
+       /*
+        * dma_buf_put() decreases the dmabuf reference counter and will
+        * call tee_shm_release() when the last reference is gone.
+        *
+        * In the case of driver private memory we call tee_shm_release
+        * directly instead as it doesn't have a reference counter.
+        */
+       if (shm->flags & TEE_SHM_DMA_BUF)
+               dma_buf_put(shm->dmabuf);
+       else
+               tee_shm_release(shm);
+ }
+ EXPORT_SYMBOL_GPL(tee_shm_free);
+ /**
+  * tee_shm_va2pa() - Get physical address of a virtual address
+  * @shm:      Shared memory handle
+  * @va:               Virtual address to tranlsate
+  * @pa:               Returned physical address
+  * @returns 0 on success and < 0 on failure
+  */
+ int tee_shm_va2pa(struct tee_shm *shm, void *va, phys_addr_t *pa)
+ {
+       /* Check that we're in the range of the shm */
+       if ((char *)va < (char *)shm->kaddr)
+               return -EINVAL;
+       if ((char *)va >= ((char *)shm->kaddr + shm->size))
+               return -EINVAL;
+       return tee_shm_get_pa(
+                       shm, (unsigned long)va - (unsigned long)shm->kaddr, pa);
+ }
+ EXPORT_SYMBOL_GPL(tee_shm_va2pa);
+ /**
+  * tee_shm_pa2va() - Get virtual address of a physical address
+  * @shm:      Shared memory handle
+  * @pa:               Physical address to tranlsate
+  * @va:               Returned virtual address
+  * @returns 0 on success and < 0 on failure
+  */
+ int tee_shm_pa2va(struct tee_shm *shm, phys_addr_t pa, void **va)
+ {
+       /* Check that we're in the range of the shm */
+       if (pa < shm->paddr)
+               return -EINVAL;
+       if (pa >= (shm->paddr + shm->size))
+               return -EINVAL;
+       if (va) {
+               void *v = tee_shm_get_va(shm, pa - shm->paddr);
+               if (IS_ERR(v))
+                       return PTR_ERR(v);
+               *va = v;
+       }
+       return 0;
+ }
+ EXPORT_SYMBOL_GPL(tee_shm_pa2va);
+ /**
+  * tee_shm_get_va() - Get virtual address of a shared memory plus an offset
+  * @shm:      Shared memory handle
+  * @offs:     Offset from start of this shared memory
+  * @returns virtual address of the shared memory + offs if offs is within
+  *    the bounds of this shared memory, else an ERR_PTR
+  */
+ void *tee_shm_get_va(struct tee_shm *shm, size_t offs)
+ {
+       if (offs >= shm->size)
+               return ERR_PTR(-EINVAL);
+       return (char *)shm->kaddr + offs;
+ }
+ EXPORT_SYMBOL_GPL(tee_shm_get_va);
+ /**
+  * tee_shm_get_pa() - Get physical address of a shared memory plus an offset
+  * @shm:      Shared memory handle
+  * @offs:     Offset from start of this shared memory
+  * @pa:               Physical address to return
+  * @returns 0 if offs is within the bounds of this shared memory, else an
+  *    error code.
+  */
+ int tee_shm_get_pa(struct tee_shm *shm, size_t offs, phys_addr_t *pa)
+ {
+       if (offs >= shm->size)
+               return -EINVAL;
+       if (pa)
+               *pa = shm->paddr + offs;
+       return 0;
+ }
+ EXPORT_SYMBOL_GPL(tee_shm_get_pa);
+ /**
+  * tee_shm_get_from_id() - Find shared memory object and increase reference
+  * count
+  * @ctx:      Context owning the shared memory
+  * @id:               Id of shared memory object
+  * @returns a pointer to 'struct tee_shm' on success or an ERR_PTR on failure
+  */
+ struct tee_shm *tee_shm_get_from_id(struct tee_context *ctx, int id)
+ {
+       struct tee_device *teedev;
+       struct tee_shm *shm;
+       if (!ctx)
+               return ERR_PTR(-EINVAL);
+       teedev = ctx->teedev;
+       mutex_lock(&teedev->mutex);
+       shm = idr_find(&teedev->idr, id);
+       if (!shm || shm->ctx != ctx)
+               shm = ERR_PTR(-EINVAL);
+       else if (shm->flags & TEE_SHM_DMA_BUF)
+               get_dma_buf(shm->dmabuf);
+       mutex_unlock(&teedev->mutex);
+       return shm;
+ }
+ 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
+  */
+ void tee_shm_put(struct tee_shm *shm)
+ {
+       if (shm->flags & TEE_SHM_DMA_BUF)
+               dma_buf_put(shm->dmabuf);
+ }
+ EXPORT_SYMBOL_GPL(tee_shm_put);