From: Chris Wilson Date: Mon, 8 Nov 2010 01:12:29 +0000 (+0000) Subject: drm/i915: Avoid might_fault during pwrite whilst holding our mutex X-Git-Url: https://git.stricted.de/?a=commitdiff_plain;h=b47b30ccdaad5f2fc39a1a65921bffd150574a91;p=GitHub%2Fmoto-9609%2Fandroid_kernel_motorola_exynos9610.git drm/i915: Avoid might_fault during pwrite whilst holding our mutex ... and so prevent a potential circular reference: [ INFO: possible circular locking dependency detected ] 2.6.37-rc1-uwe1+ #4 ------------------------------------------------------- Xorg/1401 is trying to acquire lock: (&mm->mmap_sem){++++++}, at: [] might_fault+0x4b/0xa0 but task is already holding lock: (&dev->struct_mutex){+.+.+.}, at: [] i915_mutex_lock_interruptible+0x3c/0x60 [i915] which lock already depends on the new lock. When the locking around the pwrite ioctl was simplified, I did not spot that the phys path never took any locks and so we introduced this potential circular reference. Reported-by: Uwe Helm Signed-off-by: Chris Wilson --- diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index 984eb6e9db03..eba9b1615228 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -4878,17 +4878,24 @@ i915_gem_phys_pwrite(struct drm_device *dev, struct drm_gem_object *obj, struct drm_file *file_priv) { struct drm_i915_gem_object *obj_priv = to_intel_bo(obj); - void *obj_addr; - int ret; - char __user *user_data; + void *vaddr = obj_priv->phys_obj->handle->vaddr + args->offset; + char __user *user_data = (char __user *) (uintptr_t) args->data_ptr; - user_data = (char __user *) (uintptr_t) args->data_ptr; - obj_addr = obj_priv->phys_obj->handle->vaddr + args->offset; + DRM_DEBUG_DRIVER("vaddr %p, %lld\n", vaddr, args->size); - DRM_DEBUG_DRIVER("obj_addr %p, %lld\n", obj_addr, args->size); - ret = copy_from_user(obj_addr, user_data, args->size); - if (ret) - return -EFAULT; + if (__copy_from_user_inatomic_nocache(vaddr, user_data, args->size)) { + unsigned long unwritten; + + /* The physical object once assigned is fixed for the lifetime + * of the obj, so we can safely drop the lock and continue + * to access vaddr. + */ + mutex_unlock(&dev->struct_mutex); + unwritten = copy_from_user(vaddr, user_data, args->size); + mutex_lock(&dev->struct_mutex); + if (unwritten) + return -EFAULT; + } drm_agp_chipset_flush(dev); return 0;