drm/i915/userptr: Hold mmref whilst calling get-user-pages
authorChris Wilson <chris@chris-wilson.co.uk>
Tue, 5 Apr 2016 14:00:00 +0000 (15:00 +0100)
committerJani Nikula <jani.nikula@intel.com>
Mon, 18 Apr 2016 09:35:49 +0000 (12:35 +0300)
Holding a reference to the containing task_struct is not sufficient to
prevent the mm_struct from being reaped under memory pressure. If this
happens whilst we are calling get_user_pages(), explosions erupt -
sometimes an immediate GPF, sometimes page flag corruption. To prevent
the target mm from being reaped as we are reading from it, acquire a
reference before we begin.

Testcase: igt/gem_shrink/*userptr
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Cc: stable@vger.kernel.org
Reviewed-by: Michał Winiarski <michal.winiarski@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1459864801-28606-2-git-send-email-chris@chris-wilson.co.uk
(cherry picked from commit 40313f0cd0b711a7a5905e5182422799e157d8aa)
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
drivers/gpu/drm/i915/i915_gem_userptr.c

index 18ba8139e922898e89d11423899bc0b4176b73fe..4d30b60defda44adcbc24eea0be77d1e9278e804 100644 (file)
@@ -501,19 +501,24 @@ __i915_gem_userptr_get_pages_worker(struct work_struct *_work)
        if (pvec != NULL) {
                struct mm_struct *mm = obj->userptr.mm->mm;
 
-               down_read(&mm->mmap_sem);
-               while (pinned < npages) {
-                       ret = get_user_pages_remote(work->task, mm,
-                                       obj->userptr.ptr + pinned * PAGE_SIZE,
-                                       npages - pinned,
-                                       !obj->userptr.read_only, 0,
-                                       pvec + pinned, NULL);
-                       if (ret < 0)
-                               break;
-
-                       pinned += ret;
+               ret = -EFAULT;
+               if (atomic_inc_not_zero(&mm->mm_users)) {
+                       down_read(&mm->mmap_sem);
+                       while (pinned < npages) {
+                               ret = get_user_pages_remote
+                                       (work->task, mm,
+                                        obj->userptr.ptr + pinned * PAGE_SIZE,
+                                        npages - pinned,
+                                        !obj->userptr.read_only, 0,
+                                        pvec + pinned, NULL);
+                               if (ret < 0)
+                                       break;
+
+                               pinned += ret;
+                       }
+                       up_read(&mm->mmap_sem);
+                       mmput(mm);
                }
-               up_read(&mm->mmap_sem);
        }
 
        mutex_lock(&dev->struct_mutex);