On Sun, Dec 31, 2023 at 12:13:49PM -0800, Darrick J. Wong wrote: > + error = xfile_stat(xf, &sb); > + if (error) > + return error; > + > + printk(KERN_ALERT "xfile ino 0x%lx isize 0x%llx dump:", inode->i_ino, > + sb.size); > + pflags = memalloc_nofs_save(); Hm, why? What makes it a bad idea to call back into the filesysteam at this point? > + page = shmem_read_mapping_page_gfp(mapping, > + datapos >> PAGE_SHIFT, __GFP_NOWARN); This GFP flag looks wrong. Why can't we use GFP_KERNEL here? I'm also not thrilled about the use of page APIs instead of folio APIs, but given how long this patchset has been in development, I understand why you didn't start out with folio APIs. It's not a blocker by any means. I can come through and convert it later when I decide that it's finally time to get rid of shmem_read_mapping_page_gfp(), which is going to take a big gulp because it now means touching GPU drivers ...
