From: Yu Kuai <yukuai3@xxxxxxxxxx>
Due to known performance issues with md-bitmap and the unreasonable
implementations like following:
- self-managed pages, bitmap_storage->filemap;
- self-managed IO submitting like filemap_write_page();
- global spin_lock
...
I have decided not to continue optimizing based on the current bitmap
implementation, and plan to invent a new lock-less bitmap. And build
the bitmap as kernel module is a good way for isolation. However, we
still encourage anyone who wants to continue optimizing the current
implementation, the new bitmap will take some time to be ready.
Yu Kuai (13):
md/md-bitmap: remove the parameter 'init' for bitmap_ops->resize()
md/md-bitmap: merge md_bitmap_group into bitmap_operations
md/md-bitmap: add md_bitmap_registered/enabled() helper
md/md-bitmap: handle the case bitmap is not enabled before
start_sync()
md/md-bitmap: handle the case bitmap is not enabled before end_sync()
md/dm-raid: check if bitmap is registered in raid_ctr()
md/raid1: check bitmap before behind write
md/raid1: check before deferencing mddev->bitmap_ops
md/raid10: check before deferencing mddev->bitmap_ops
md/raid5: check before deferencing mddev->bitmap_ops
md: check before deferencing mddev->bitmap_ops
md: export some helpers
md/md-bitmap: support to build md-bitmap as kernel module
drivers/md/Kconfig | 18 ++++
drivers/md/Makefile | 4 +-
drivers/md/dm-raid.c | 5 +-
drivers/md/md-bitmap.c | 72 ++++++++--------
drivers/md/md-bitmap.h | 54 +++++++++++-
drivers/md/md-cluster.c | 2 +-
drivers/md/md.c | 183 +++++++++++++++++++++++++++++++---------
drivers/md/md.h | 12 ++-
drivers/md/raid1-10.c | 2 +-
drivers/md/raid1.c | 79 ++++++++++-------
drivers/md/raid10.c | 48 +++++------
drivers/md/raid5.c | 30 ++++---
12 files changed, 352 insertions(+), 157 deletions(-)
--
2.39.2
