OSDN Git Service

drm/amd/amdgpu: Partial revert of iova debugfs
authorTom St Denis <tom.stdenis@amd.com>
Tue, 19 Sep 2017 15:29:04 +0000 (11:29 -0400)
committerAlex Deucher <alexander.deucher@amd.com>
Tue, 26 Sep 2017 19:14:15 +0000 (15:14 -0400)
We discovered that on some devices even with iommu enabled
you can access all of system memory through the iommu translation.

Therefore, we revert the read method to the translation only service
and drop the write method completely.

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Christan König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c

index 0e5f78f..ce435db 100644 (file)
@@ -1784,98 +1784,34 @@ static ssize_t amdgpu_iova_to_phys_read(struct file *f, char __user *buf,
                                   size_t size, loff_t *pos)
 {
        struct amdgpu_device *adev = file_inode(f)->i_private;
-       ssize_t result, n;
        int r;
        uint64_t phys;
-       void *ptr;
        struct iommu_domain *dom;
 
-       dom = iommu_get_domain_for_dev(adev->dev);
-       if (!dom)
-               return -EFAULT;
-
-       result = 0;
-       while (size) {
-               // get physical address and map
-               phys = iommu_iova_to_phys(dom, *pos);
-
-               // copy upto one page
-               if (size > PAGE_SIZE)
-                       n = PAGE_SIZE;
-               else
-                       n = size;
-
-               // to end of the page
-               if (((*pos & (PAGE_SIZE - 1)) + n) >= PAGE_SIZE)
-                       n = PAGE_SIZE - (*pos & (PAGE_SIZE - 1));
-
-               ptr = kmap(pfn_to_page(PFN_DOWN(phys)));
-               if (!ptr)
-                       return -EFAULT;
-
-               r = copy_to_user(buf, ptr, n);
-               kunmap(pfn_to_page(PFN_DOWN(phys)));
-               if (r)
-                       return -EFAULT;
-
-               *pos += n;
-               size -= n;
-               result += n;
-       }
-
-       return result;
-}
+       // always return 8 bytes
+       if (size != 8)
+               return -EINVAL;
 
-static ssize_t amdgpu_iova_to_phys_write(struct file *f, const char __user *buf,
-                                  size_t size, loff_t *pos)
-{
-       struct amdgpu_device *adev = file_inode(f)->i_private;
-       ssize_t result, n;
-       int r;
-       uint64_t phys;
-       void *ptr;
-       struct iommu_domain *dom;
+       // only accept page addresses
+       if (*pos & 0xFFF)
+               return -EINVAL;
 
        dom = iommu_get_domain_for_dev(adev->dev);
-       if (!dom)
-               return -EFAULT;
-
-       result = 0;
-       while (size) {
-               // get physical address and map
+       if (dom)
                phys = iommu_iova_to_phys(dom, *pos);
+       else
+               phys = *pos;
 
-               // copy upto one page
-               if (size > PAGE_SIZE)
-                       n = PAGE_SIZE;
-               else
-                       n = size;
-
-               // to end of the page
-               if (((*pos & (PAGE_SIZE - 1)) + n) >= PAGE_SIZE)
-                       n = PAGE_SIZE - (*pos & (PAGE_SIZE - 1));
-
-               ptr = kmap(pfn_to_page(PFN_DOWN(phys)));
-               if (!ptr)
-                       return -EFAULT;
-
-               r = copy_from_user(ptr, buf, n);
-               kunmap(pfn_to_page(PFN_DOWN(phys)));
-               if (r)
-                       return -EFAULT;
-
-               *pos += n;
-               size -= n;
-               result += n;
-       }
+       r = copy_to_user(buf, &phys, 8);
+       if (r)
+               return -EFAULT;
 
-       return result;
+       return 8;
 }
 
 static const struct file_operations amdgpu_ttm_iova_fops = {
        .owner = THIS_MODULE,
        .read = amdgpu_iova_to_phys_read,
-       .write = amdgpu_iova_to_phys_write,
        .llseek = default_llseek
 };