| Message ID | 1455519687-23873-3-git-send-email-ming.lei@canonical.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h > index 4571ef1..b8ff6a3 100644 > --- a/include/linux/blkdev.h > +++ b/include/linux/blkdev.h > @@ -1388,7 +1388,7 @@ static inline bool bvec_gap_to_prev(struct request_queue *q, > static inline bool bio_will_gap(struct request_queue *q, struct bio *prev, > struct bio *next) > { > - if (!bio_has_data(prev)) > + if (!bio_has_data(prev) || !queue_virt_boundary(q)) > return false; Can we not do that? bvec_gap_to_prev is already checking the virt_boundary and I'd sorta like to keep the motivation to optimize bio_get_last_bvec() to be O(1). -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, Feb 15, 2016 at 4:22 PM, Sagi Grimberg <sagig@dev.mellanox.co.il> wrote: > >> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h >> index 4571ef1..b8ff6a3 100644 >> --- a/include/linux/blkdev.h >> +++ b/include/linux/blkdev.h >> @@ -1388,7 +1388,7 @@ static inline bool bvec_gap_to_prev(struct >> request_queue *q, >> static inline bool bio_will_gap(struct request_queue *q, struct bio >> *prev, >> struct bio *next) >> { >> - if (!bio_has_data(prev)) >> + if (!bio_has_data(prev) || !queue_virt_boundary(q)) >> return false; > > > Can we not do that? Given there are only 3 drivers which set virt boundary, I think it is reasonable to do that. > > bvec_gap_to_prev is already checking the virt_boundary and I'd sorta > like to keep the motivation to optimize bio_get_last_bvec() to be O(1). Currently the approaches I thought of still need to iterate bvec by bvec, not sure if O(1) can be reached easily, but I am happy to discuss the optimized implementation. Thanks, Ming -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
>>> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h >>> index 4571ef1..b8ff6a3 100644 >>> --- a/include/linux/blkdev.h >>> +++ b/include/linux/blkdev.h >>> @@ -1388,7 +1388,7 @@ static inline bool bvec_gap_to_prev(struct >>> request_queue *q, >>> static inline bool bio_will_gap(struct request_queue *q, struct bio >>> *prev, >>> struct bio *next) >>> { >>> - if (!bio_has_data(prev)) >>> + if (!bio_has_data(prev) || !queue_virt_boundary(q)) >>> bio_integrity_add_page return false; >> >> >> Can we not do that? > > Given there are only 3 drivers which set virt boundary, I think > it is reasonable to do that. 3 drivers that are really performance critical. I don't think we should add optimized branching for some of the drivers especially when the drivers that do set virt_boundary *really* care about latency. >> bvec_gap_to_prev is already checking the virt_boundary and I'd sorta >> like to keep the motivation to optimize bio_get_last_bvec() to be O(1). > > Currently the approaches I thought of still need to iterate bvec by bvec, > not sure if O(1) can be reached easily, but I am happy to discuss the > optimized implementation. Me too. Note that I don't mind if the bio split code won't be optimized, but I do want req_gap_back_merge/req_gap_front_merge to be... Also, are the bvec_gap_to_prev usages in bio_add_pc_page and bio_integrity_add_page safe? I didn't test this stuff with integrity payloads... -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, Feb 16, 2016 at 4:27 AM, Sagi Grimberg <sagig@dev.mellanox.co.il> wrote: > >>>> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h >>>> index 4571ef1..b8ff6a3 100644 >>>> --- a/include/linux/blkdev.h >>>> +++ b/include/linux/blkdev.h >>>> @@ -1388,7 +1388,7 @@ static inline bool bvec_gap_to_prev(struct >>>> request_queue *q, >>>> static inline bool bio_will_gap(struct request_queue *q, struct bio >>>> *prev, >>>> struct bio *next) >>>> { >>>> - if (!bio_has_data(prev)) >>>> + if (!bio_has_data(prev) || !queue_virt_boundary(q)) >>>> bio_integrity_add_page return false; >>> >>> >>> >>> Can we not do that? >> >> >> Given there are only 3 drivers which set virt boundary, I think >> it is reasonable to do that. > > > 3 drivers that are really performance critical. I don't think we > should add optimized branching for some of the drivers especially > when the drivers that do set virt_boundary *really* care about latency. > >>> bvec_gap_to_prev is already checking the virt_boundary and I'd sorta >>> like to keep the motivation to optimize bio_get_last_bvec() to be O(1). >> >> >> Currently the approaches I thought of still need to iterate bvec by bvec, >> not sure if O(1) can be reached easily, but I am happy to discuss the >> optimized implementation. > > > Me too. Note that I don't mind if the bio split code won't be optimized, > but I do want req_gap_back_merge/req_gap_front_merge to be... > > Also, are the bvec_gap_to_prev usages in bio_add_pc_page and > bio_integrity_add_page safe? I didn't test this stuff with integrity Yes, because both are non-cloned bvec table. > payloads... Thanks, -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, Feb 16, 2016 at 4:27 AM, Sagi Grimberg <sagig@dev.mellanox.co.il> wrote: > >>>> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h >>>> index 4571ef1..b8ff6a3 100644 >>>> --- a/include/linux/blkdev.h >>>> +++ b/include/linux/blkdev.h >>>> @@ -1388,7 +1388,7 @@ static inline bool bvec_gap_to_prev(struct >>>> request_queue *q, >>>> static inline bool bio_will_gap(struct request_queue *q, struct bio >>>> *prev, >>>> struct bio *next) >>>> { >>>> - if (!bio_has_data(prev)) >>>> + if (!bio_has_data(prev) || !queue_virt_boundary(q)) >>>> bio_integrity_add_page return false; >>> >>> >>> >>> Can we not do that? >> >> >> Given there are only 3 drivers which set virt boundary, I think >> it is reasonable to do that. > > > 3 drivers that are really performance critical. I don't think we > should add optimized branching for some of the drivers especially > when the drivers that do set virt_boundary *really* care about latency. I don't think the extra check on bvec_gap_to_prev() can make any difference, but if you do care we can introduce __bvec_gap_to_prev() in which the check is moved into bio_will_gap(). Thanks, -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h index 4571ef1..b8ff6a3 100644 --- a/include/linux/blkdev.h +++ b/include/linux/blkdev.h @@ -1388,7 +1388,7 @@ static inline bool bvec_gap_to_prev(struct request_queue *q, static inline bool bio_will_gap(struct request_queue *q, struct bio *prev, struct bio *next) { - if (!bio_has_data(prev)) + if (!bio_has_data(prev) || !queue_virt_boundary(q)) return false; return bvec_gap_to_prev(q, &prev->bi_io_vec[prev->bi_vcnt - 1],
In the following patch, the way for figuring out the last bvec will be changed with a bit cost introduced, so return immediately if the queue doesn't have virt boundary limit. Actually most of devices have not this limit. Cc: Sagi Grimberg <sagig@dev.mellanox.co.il> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Ming Lei <ming.lei@canonical.com> --- include/linux/blkdev.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)