Hi, David!
Thanks for the reply.
Actually I just got CC'd on the Fedora Bug. I haven't seen this either, so I can't provide more info...
What I was thinking was that maybe after a fork, vma->vm_file->f_mapping of the child process wasn't set to the same value as the parent...
/Thomas
On 04/09/2014 08:54 PM, David Herrmann wrote:
Hi Thomas
On Tue, Apr 8, 2014 at 7:11 AM, Thomas Hellstrom thellstrom@vmware.com wrote:
Hi, David,
Are there any dev_mapping changes in 3.15 that could cause this? Do we know what happens to vma->vm_file->f_mapping during fork?
Sorry, I was traveling. Yes, there have been changes, but I converted all drivers properly. We now set file->f_mapping in drm_fops.c to dev->anon_inode->i_mapping during open(). All the ttm stuff now sets ttm->dev_mapping to the same during ttm init so I have no idea why that assertion fails. The vmwgfx driver sets it correctly in vmw_driver_load() and I cannot find any code changing it.
I also cannot reproduce it. Can you provide some more information? Like what is vma->vm_file set to? What is ttm->dev_mapping? What is drm_device->anon_inode->i_mapping? What is &drm_device->anon_inode->i_data? All these should _always_ be the same.
Btw., grep'ping for '_mapping' usually shows all related code, which I reviewed and all looks good.
Thanks David