On Mon, Nov 27, 2023 at 07:41:07AM +0100, Christoph Hellwig wrote: > On Mon, Nov 27, 2023 at 12:09:08PM +0530, Ritesh Harjani wrote: > > Nice cleanup. As you explained, iomap_do_writepage() never gets called > > from memory reclaim context. So it is unused code which can be removed. > > > > However, there was an instance when this WARN was hit by wrong > > usage of PF_MEMALLOC flag [1], which was caught due to WARN_ON_ONCE. > > > > [1]: https://lore.kernel.org/linux-xfs/20200309185714.42850-1-ebiggers@xxxxxxxxxx/ > > > > Maybe we can just have a WARN_ON_ONCE() and update the comments? > > We anyway don't require "goto redirty" anymore since we will never > > actually get called from reclaim context. > > Well, xfs/iomap never really cared about the flag, just that it didn't > get called from direct reclaim or kswapd. If we think such a WARN_ON > is still useful, the only caller of it in do_writepages might be a better > place. I think that's a good idea, given the enormous complexity of modern day XFS and the sometimes rough quality of others. :P --D
