When there is direct IO (or other DMA write) running into a page, it is
not generally safe to submit this page for another IO (such as
writeback) because this can cause checksum failures or similar issues.
However sometimes we cannot avoid writing contents of these pages as
pages can be pinned for extensive amount of time (e.g. for RDMA). For
these cases we need to just bounce the pages if we really need to write
them out. Add support for this type of bouncing into the block layer
infrastructure.
Signed-off-by: Jan Kara <jack@xxxxxxx>
---
block/blk.h | 10 +++++++++-
block/bounce.c | 9 +++++++--
include/linux/blk_types.h | 1 +
mm/Kconfig | 8 ++++----
4 files changed, 21 insertions(+), 7 deletions(-)
diff --git a/block/blk.h b/block/blk.h
index 4c3b3325219a..def7ab8379bc 100644
--- a/block/blk.h
+++ b/block/blk.h
@@ -384,10 +384,18 @@ static inline bool blk_queue_may_bounce(struct request_queue *q)
max_low_pfn >= max_pfn;
}
+static inline bool bio_need_pin_bounce(struct bio *bio,
+ struct request_queue *q)
+{
+ return IS_ENABLED(CONFIG_BOUNCE) &&
+ bio->bi_flags & (1 << BIO_NEED_PIN_BOUNCE);
+}
+
static inline struct bio *blk_queue_bounce(struct bio *bio,
struct request_queue *q)
{
- if (unlikely(blk_queue_may_bounce(q) && bio_has_data(bio)))
+ if (unlikely((blk_queue_may_bounce(q) || bio_need_pin_bounce(bio, q)) &&
+ bio_has_data(bio)))
return __blk_queue_bounce(bio, q);
return bio;
}
diff --git a/block/bounce.c b/block/bounce.c
index 7cfcb242f9a1..ebda95953d58 100644
--- a/block/bounce.c
+++ b/block/bounce.c
@@ -207,12 +207,16 @@ struct bio *__blk_queue_bounce(struct bio *bio_orig, struct request_queue *q)
struct bvec_iter iter;
unsigned i = 0, bytes = 0;
bool bounce = false;
+ bool pinned_bounce = bio_orig->bi_flags & (1 << BIO_NEED_PIN_BOUNCE);
+ bool highmem_bounce = blk_queue_may_bounce(q);
int sectors;
bio_for_each_segment(from, bio_orig, iter) {
if (i++ < BIO_MAX_VECS)
bytes += from.bv_len;
- if (PageHighMem(from.bv_page))
+ if (highmem_bounce && PageHighMem(from.bv_page))
+ bounce = true;
+ if (pinned_bounce && page_maybe_dma_pinned(from.bv_page))
bounce = true;
}
if (!bounce)
@@ -241,7 +245,8 @@ struct bio *__blk_queue_bounce(struct bio *bio_orig, struct request_queue *q)
for (i = 0, to = bio->bi_io_vec; i < bio->bi_vcnt; to++, i++) {
struct page *bounce_page;
- if (!PageHighMem(to->bv_page))
+ if (!((highmem_bounce && PageHighMem(to->bv_page)) ||
+ (pinned_bounce && page_maybe_dma_pinned(to->bv_page))))
continue;
bounce_page = mempool_alloc(&page_pool, GFP_NOIO);
diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
index 99be590f952f..3aa1dc5d8dc6 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -321,6 +321,7 @@ enum {
BIO_NO_PAGE_REF, /* don't put release vec pages */
BIO_CLONED, /* doesn't own data */
BIO_BOUNCED, /* bio is a bounce bio */
+ BIO_NEED_PIN_BOUNCE, /* bio needs to bounce pinned pages */
BIO_QUIET, /* Make BIO Quiet */
BIO_CHAIN, /* chained bio, ->bi_remaining in effect */
BIO_REFFED, /* bio has elevated ->bi_cnt */
diff --git a/mm/Kconfig b/mm/Kconfig
index ff7b209dec05..eba075e959e8 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -659,11 +659,11 @@ config PHYS_ADDR_T_64BIT
config BOUNCE
bool "Enable bounce buffers"
default y
- depends on BLOCK && MMU && HIGHMEM
+ depends on BLOCK && MMU
help
- Enable bounce buffers for devices that cannot access the full range of
- memory available to the CPU. Enabled by default when HIGHMEM is
- selected, but you may say n to override this.
+ Enable bounce buffers. This is used for devices that cannot access
+ the full range of memory available to the CPU or when DMA can be
+ modifying pages while they are submitted for writeback.
config MMU_NOTIFIER
bool
--
2.35.3
