On Tue, Jun 11, 2024 at 11:15:52AM -0700, Luis Chamberlain wrote: > On Tue, Jun 11, 2024 at 07:45:03AM -0700, Darrick J. Wong wrote: > > On Mon, Jun 10, 2024 at 08:02:02PM -0700, Luis Chamberlain wrote: > > > +# Requires CONFIG_DEBUGFS and truncation knobs > > > +_require_split_debugfs() > > > > Er... I thought "split" referred to debugfs itself. > > > > _require_split_huge_pages_knob? > > Much better, thanks. > > > > +# This aims at trying to reproduce a difficult to reproduce bug found with > > > +# min order. The issue was root caused to an xarray bug when we split folios > > > +# to another order other than 0. This functionality is used to support min > > > +# order. The crash: > > > +# > > > +# https://gist.github.com/mcgrof/d12f586ec6ebe32b2472b5d634c397df > > > > You might want to paste the stacktrace in here directly, in case the > > gist ever goes away. > > Its not a simple crash trace, it is pretty enourmous considering I > decoded it, and it has all locking candidates. Even including it after > the "---" lines of the patch might make someone go: TLDR. Thoughts? I'd paste it in, even if it's quite lengthy. I don't even think it's all that much if you remove some of the less useful bits of the unwind: "Crash excerpt is as follows: "BUG: kernel NULL pointer dereference, address: 0000000000000036 #PF: supervisor read access in kernel mode #PF: error_code(0x0000) - not-present page PGD 0 P4D 0 Oops: 0000 [#1] PREEMPT SMP NOPTI CPU: 7 PID: 2190 Comm: kworker/u38:5 Not tainted 6.9.0-rc5+ #14 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.3-debian-1.16.3-2 04/01/2014 Workqueue: writeback wb_workfn (flush-7:5) RIP: 0010:filemap_get_folios_tag+0xa9/0x200 Call Trace: <TASK> writeback_iter+0x17d/0x310 write_cache_pages+0x42/0xa0 iomap_writepages+0x33/0x50 xfs_vm_writepages+0x63/0x90 [xfs] do_writepages+0xcc/0x260 __writeback_single_inode+0x3d/0x340 writeback_sb_inodes+0x1ed/0x4b0 __writeback_inodes_wb+0x4c/0xe0 wb_writeback+0x267/0x2d0 wb_workfn+0x2a4/0x440 process_one_work+0x189/0x3b0 worker_thread+0x273/0x390 kthread+0xda/0x110 ret_from_fork+0x2d/0x50 ret_from_fork_asm+0x1a/0x30 </TASK>" --D > > > +if grep -q thp_split_page /proc/vmstat; then > > > + split_count_after=$(grep ^thp_split_page /proc/vmstat | head -1 | awk '{print $2}') > > > + split_count_failed_after=$(grep ^thp_split_page_failed /proc/vmstat | head -1 | awk '{print $2}') > > > > I think this ought to be a separate function for cleanliness? > > > > _proc_vmstat() > > { > > awk -v name="$1" '{if ($1 ~ name) {print($2)}}' /proc/vmstat > > } > > > Otherwise this test looks fine to me. > > Thanks! > > Luis >
