On Fri, Apr 16, 2021 at 10:36:52AM -0400, Mike Marshall wrote: > So... I think all your stuff is working well from my perspective > and that I need to figure out why my orangefs_readahead patch > is causing the regressions I listed. My readahead implementation (via your > readahead_expand) is really fast, but it is bare-bones... I'm probably > leaving out some important stuff... I see other filesystem's > readahead implementations doing stuff like avoiding doing readahead > on pages that have yet to be written back for example. You do?! Actual readahead implementations, or people still implementing the old ->readpages() method? The ->readahead() method is _only_ called for pages which are freshly allocated, Locked and !Uptodate. If you ever see a page which is Dirty or Writeback, something has gone very wrong. Could you tell me which filesystem you saw that bogosity in? > The top two commits at https://github.com/hubcapsc/linux/tree/readahead_v3 > is the current state of my readahead implementation. > > Please do add > Tested-by: Mike Marshall <hubcap@xxxxxxxxxxxx> > > -Mike > > On Tue, Apr 13, 2021 at 11:08 AM David Howells <dhowells@xxxxxxxxxx> wrote: > > > > Mike Marshall <hubcap@xxxxxxxxxxxx> wrote: > > > > > Hi David... I've been gone on a motorcycle adventure, > > > sorry for the delay... here's my public branch... > > > > > > https://github.com/hubcapsc/linux/tree/readahead_v3 > > > > That seems to have all of my fscache-iter branch in it. I thought you'd said > > you'd dropped them because they were causing problems. > > > > Anyway, I've distilled the basic netfs lib patches, including the readahead > > expansion patch and ITER_XARRAY patch here: > > > > https://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs.git/log/?h=netfs-lib > > > > if that's of use to you? > > > > If you're using any of these patches, would it be possible to get a Tested-by > > for them that I can add? > > > > David > >
