Add a new config parameter, CONFIG_BLK_USE_PIN_USER_PAGES_FOR_DIO, and dio_w_*() wrapper functions. Together, these allow the developer to choose between these sets of routines, for Direct IO code paths: a) pin_user_pages_fast() pin_user_page() unpin_user_page() b) get_user_pages_fast() get_page() put_page() CONFIG_BLK_USE_PIN_USER_PAGES_FOR_DIO is a temporary setting, and will be deleted once the conversion is complete. In the meantime, developers can enable this in order to try out each filesystem. More information: The Direct IO part of the block infrastructure is being changed to use pin_user_page*() and unpin_user_page*() calls, in place of a mix of get_user_pages_fast(), get_page(), and put_page(). These have to be changed over all at the same time, for block, bio, and all filesystems. While that changeover is in progress (but disabled via this new CONFIG option), kernel developers need a way to test their changes. The steps are: a) Enable CONFIG_BLK_USE_PIN_USER_PAGES_FOR_DIO b) Monitor these /proc/vmstat items: nr_foll_pin_acquired nr_foll_pin_released ...to ensure that they remain equal, when "at rest". Signed-off-by: John Hubbard <jhubbard@xxxxxxxxxx> --- block/Kconfig | 25 +++++++++++++++++++++++++ include/linux/bvec.h | 11 +++++++++++ 2 files changed, 36 insertions(+) diff --git a/block/Kconfig b/block/Kconfig index 168b873eb666..f6ca5e9597e4 100644 --- a/block/Kconfig +++ b/block/Kconfig @@ -50,6 +50,31 @@ config BLK_DEV_BSG_COMMON config BLK_ICQ bool +config BLK_USE_PIN_USER_PAGES_FOR_DIO + bool "DEVELOPERS ONLY: Enable pin_user_pages() for Direct IO" if EXPERT + default n + help + For Direct IO code, retain the pages via calls to + pin_user_pages_fast(), instead of via get_user_pages_fast(). + Likewise, use pin_user_page() instead of get_page(). And then + release such pages via unpin_user_page(), instead of + put_page(). + + This is a temporary setting, which will be deleted once the + conversion is completed, reviewed, and tested. In the meantime, + developers can enable this in order to try out each filesystem. + For that, it's best to monitor these /proc/vmstat items: + + nr_foll_pin_acquired + nr_foll_pin_released + + ...to ensure that they remain equal, when "at rest". + + Say yes here ONLY if are actively developing or testing the + block layer or filesystems with pin_user_pages_fast(). + Otherwise, this is just a way to throw off the refcounting of + pages in the system. + config BLK_DEV_BSGLIB bool "Block layer SG support v4 helper lib" select BLK_DEV_BSG_COMMON diff --git a/include/linux/bvec.h b/include/linux/bvec.h index 35c25dff651a..a96a68c687f6 100644 --- a/include/linux/bvec.h +++ b/include/linux/bvec.h @@ -241,4 +241,15 @@ static inline void *bvec_virt(struct bio_vec *bvec) return page_address(bvec->bv_page) + bvec->bv_offset; } +#ifdef CONFIG_BLK_USE_PIN_USER_PAGES_FOR_DIO +#define dio_w_pin_user_pages_fast(s, n, p, f) pin_user_pages_fast(s, n, p, f) +#define dio_w_pin_user_page(p) pin_user_page(p) +#define dio_w_unpin_user_page(p) unpin_user_page(p) + +#else +#define dio_w_pin_user_pages_fast(s, n, p, f) get_user_pages_fast(s, n, p, f) +#define dio_w_pin_user_page(p) get_page(p) +#define dio_w_unpin_user_page(p) put_page(p) +#endif + #endif /* __LINUX_BVEC_H */ -- 2.35.1
