All of lore.kernel.org
 help / color / mirror / Atom feed
* block and nvme polling improvements V3
@ 2018-12-02 16:46 ` Christoph Hellwig
  0 siblings, 0 replies; 104+ messages in thread
From: Christoph Hellwig @ 2018-12-02 16:46 UTC (permalink / raw)
  To: Jens Axboe, Keith Busch, Sagi Grimberg
  Cc: Max Gurtovoy, linux-nvme, linux-block

Hi all,

this series optimizes a few bits in the block layer and nvme code
related to polling.

It starts by moving the queue types recently introduce entirely into
the block layer instead of requiring an indirect call for them.

It then switches nvme and the block layer to only allow polling
with separate poll queues, which allows us to realize the following
benefits:

 - poll queues can safely avoid disabling irqs on any locks
   (we already do that in NVMe, but it isn't 100% kosher as-is)
 - regular interrupt driven queues can drop the CQ lock entirely,
   as we won't race for completing CQs

Then we drop the NVMe RDMA code, as it doesn't follow the new mode,
and remove the nvme multipath polling code including the block hooks
for it, which didn't make much sense to start with given that we
started bypassing the multipath code for single controller subsystems
early on.  Last but not least we enable polling in the block layer
by default if the underlying driver has poll queues, as that already
requires explicit user action.

Note that it would be really nice to have polling back for RDMA with
dedicated poll queues, but that might take a while.  Also based on
Jens' polling aio patches we could now implement a model in nvmet
where we have a thread polling both the backend nvme device and
the RDMA CQs, which might give us some pretty nice performace
(I know Sagi looked into something similar a while ago).

A git tree is also available at:

    git://git.infradead.org/users/hch/block.git nvme-polling

Gitweb:

    http://git.infradead.org/users/hch/block.git/shortlog/refs/heads/nvme-polling

Changes since v2:
 - fix a changelog typo
 - report a string instead of an index from the type sysfs attribute
 - move to a per-queue completions for queue deletion
 - clear NVMEQ_DELETE_ERROR when initializing a queue

Changes since v1:
 - rebased to the latest block for-4.21 tree

^ permalink raw reply	[flat|nested] 104+ messages in thread
* block and nvme polling improvements V2
@ 2018-11-29 19:12 Christoph Hellwig
  2018-11-29 19:12   ` Christoph Hellwig
  0 siblings, 1 reply; 104+ messages in thread
From: Christoph Hellwig @ 2018-11-29 19:12 UTC (permalink / raw)
  To: Jens Axboe, Keith Busch, Sagi Grimberg
  Cc: Max Gurtovoy, linux-nvme, linux-block

Hi all,

this series optimizes a few bits in the block layer and nvme code
related to polling.

It starts by moving the queue types recently introduce entirely into
the block layer instead of requiring an indirect call for them.

It then switches nvme and the block layer to only allow polling
with separate poll queues, which allows us to realize the following
benefits:

 - poll queues can safely avoid disabling irqs on any locks
   (we already do that in NVMe, but it isn't 100% kosher as-is)
 - regular interrupt driven queues can drop the CQ lock entirely,
   as we won't race for completing CQs

Then we drop the NVMe RDMA code, as it doesn't follow the new mode,
and remove the nvme multipath polling code including the block hooks
for it, which didn't make much sense to start with given that we
started bypassing the multipath code for single controller subsystems
early on.  Last but not least we enable polling in the block layer
by default if the underlying driver has poll queues, as that already
requires explicit user action.

Note that it would be really nice to have polling back for RDMA with
dedicated poll queues, but that might take a while.  Also based on
Jens' polling aio patches we could now implement a model in nvmet
where we have a thread polling both the backend nvme device and
the RDMA CQs, which might give us some pretty nice performace
(I know Sagi looked into something similar a while ago).

A git tree is also available at:

    git://git.infradead.org/users/hch/block.git nvme-polling

Gitweb:

    http://git.infradead.org/users/hch/block.git/shortlog/refs/heads/nvme-polling

Changes since v2:
 - rebased to the latest block for-4.21 tree

^ permalink raw reply	[flat|nested] 104+ messages in thread
* block and nvme polling improvements
@ 2018-11-21 16:23 Christoph Hellwig
  2018-11-21 16:23   ` Christoph Hellwig
  0 siblings, 1 reply; 104+ messages in thread
From: Christoph Hellwig @ 2018-11-21 16:23 UTC (permalink / raw)
  To: Jens Axboe, Keith Busch, Sagi Grimberg
  Cc: Max Gurtovoy, linux-nvme, linux-block

Hi all,

this series optimizes a few bits in the block layer and nvme code
related to polling.

It starts by moving the queue types recently introduce entirely into
the block layer instead of requiring an indirect call for them.

It then switches nvme and the block layer to only allow polling
with separate poll queues, which allows us to realize the following
benefits:

 - poll queues can safely avoid disabling irqs on any locks
   (we already do that in NVMe, but it isn't 100% kosher as-is)
 - regular interrupt driven queues can drop the CQ lock entirely,
   as we won't race for completing CQs

Then we drop the NVMe RDMA code, as it doesn't follow the new mode,
and remove the nvme multipath polling code including the block hooks
for it, which didn't make much sense to start with given that we
started bypassing the multipath code for single controller subsystems
early on.  Last but not least we enable polling in the block layer
by default if the underlying driver has poll queues, as that already
requires explicit user action.

Note that it would be really nice to have polling back for RDMA with
dedicated poll queues, but that might take a while.  Also based on
Jens' polling aio patches we could now implement a model in nvmet
where we have a thread polling both the backend nvme device and
the RDMA CQs, which might give us some pretty nice performace
(I know Sagi looked into something similar a while ago).

A git tree is also available at:

    git://git.infradead.org/users/hch/misc.git nvme-polling

Gitweb:

    http://git.infradead.org/users/hch/block.git/shortlog/refs/heads/nvme-polling

^ permalink raw reply	[flat|nested] 104+ messages in thread

end of thread, other threads:[~2018-12-04 18:40 UTC | newest]

Thread overview: 104+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-12-02 16:46 block and nvme polling improvements V3 Christoph Hellwig
2018-12-02 16:46 ` Christoph Hellwig
2018-12-02 16:46 ` [PATCH 01/13] block: move queues types to the block layer Christoph Hellwig
2018-12-02 16:46   ` Christoph Hellwig
2018-12-04  0:49   ` Sagi Grimberg
2018-12-04  0:49     ` Sagi Grimberg
2018-12-04 15:00     ` Christoph Hellwig
2018-12-04 15:00       ` Christoph Hellwig
2018-12-04 17:08       ` Sagi Grimberg
2018-12-04 17:08         ` Sagi Grimberg
2018-12-02 16:46 ` [PATCH 02/13] nvme-pci: use atomic bitops to mark a queue enabled Christoph Hellwig
2018-12-02 16:46   ` Christoph Hellwig
2018-12-04  0:54   ` Sagi Grimberg
2018-12-04  0:54     ` Sagi Grimberg
2018-12-04 15:04     ` Christoph Hellwig
2018-12-04 15:04       ` Christoph Hellwig
2018-12-04 17:11       ` Sagi Grimberg
2018-12-04 17:11         ` Sagi Grimberg
2018-12-02 16:46 ` [PATCH 03/13] nvme-pci: cleanup SQ allocation a bit Christoph Hellwig
2018-12-02 16:46   ` Christoph Hellwig
2018-12-04  0:55   ` Sagi Grimberg
2018-12-04  0:55     ` Sagi Grimberg
2018-12-02 16:46 ` [PATCH 04/13] nvme-pci: only allow polling with separate poll queues Christoph Hellwig
2018-12-02 16:46   ` Christoph Hellwig
2018-12-03 18:23   ` Keith Busch
2018-12-03 18:23     ` Keith Busch
2018-12-04  0:56   ` Sagi Grimberg
2018-12-04  0:56     ` Sagi Grimberg
2018-12-02 16:46 ` [PATCH 05/13] nvme-pci: consolidate code for polling non-dedicated queues Christoph Hellwig
2018-12-02 16:46   ` Christoph Hellwig
2018-12-04  0:58   ` Sagi Grimberg
2018-12-04  0:58     ` Sagi Grimberg
2018-12-04 15:04     ` Christoph Hellwig
2018-12-04 15:04       ` Christoph Hellwig
2018-12-04 17:13       ` Sagi Grimberg
2018-12-04 17:13         ` Sagi Grimberg
2018-12-04 18:19         ` Jens Axboe
2018-12-04 18:19           ` Jens Axboe
2018-12-02 16:46 ` [PATCH 06/13] nvme-pci: refactor nvme_disable_io_queues Christoph Hellwig
2018-12-02 16:46   ` Christoph Hellwig
2018-12-04  1:00   ` Sagi Grimberg
2018-12-04  1:00     ` Sagi Grimberg
2018-12-04 15:05     ` Christoph Hellwig
2018-12-04 15:05       ` Christoph Hellwig
2018-12-04 18:19       ` Jens Axboe
2018-12-04 18:19         ` Jens Axboe
2018-12-02 16:46 ` [PATCH 07/13] nvme-pci: don't poll from irq context when deleting queues Christoph Hellwig
2018-12-02 16:46   ` Christoph Hellwig
2018-12-03 18:15   ` Keith Busch
2018-12-03 18:15     ` Keith Busch
2018-12-04  1:05   ` Sagi Grimberg
2018-12-04  1:05     ` Sagi Grimberg
2018-12-02 16:46 ` [PATCH 08/13] nvme-pci: remove the CQ lock for interrupt driven queues Christoph Hellwig
2018-12-02 16:46   ` Christoph Hellwig
2018-12-04  1:08   ` Sagi Grimberg
2018-12-04  1:08     ` Sagi Grimberg
2018-12-02 16:46 ` [PATCH 09/13] nvme-rdma: remove I/O polling support Christoph Hellwig
2018-12-02 16:46   ` Christoph Hellwig
2018-12-02 16:46 ` [PATCH 10/13] nvme-mpath: " Christoph Hellwig
2018-12-02 16:46   ` Christoph Hellwig
2018-12-03 18:22   ` Keith Busch
2018-12-03 18:22     ` Keith Busch
2018-12-04  1:11   ` Sagi Grimberg
2018-12-04  1:11     ` Sagi Grimberg
2018-12-04 15:07     ` Christoph Hellwig
2018-12-04 15:07       ` Christoph Hellwig
2018-12-04 17:18       ` Sagi Grimberg
2018-12-04 17:18         ` Sagi Grimberg
2018-12-02 16:46 ` [PATCH 11/13] block: remove ->poll_fn Christoph Hellwig
2018-12-02 16:46   ` Christoph Hellwig
2018-12-04  1:11   ` Sagi Grimberg
2018-12-04  1:11     ` Sagi Grimberg
2018-12-02 16:46 ` [PATCH 12/13] block: only allow polling if a poll queue_map exists Christoph Hellwig
2018-12-02 16:46   ` Christoph Hellwig
2018-12-04  1:14   ` Sagi Grimberg
2018-12-04  1:14     ` Sagi Grimberg
2018-12-02 16:46 ` [PATCH 13/13] block: enable polling by default if a poll map is initalized Christoph Hellwig
2018-12-02 16:46   ` Christoph Hellwig
2018-12-04  1:14   ` Sagi Grimberg
2018-12-04  1:14     ` Sagi Grimberg
2018-12-04 18:40 ` block and nvme polling improvements V3 Jens Axboe
2018-12-04 18:40   ` Jens Axboe
  -- strict thread matches above, loose matches on Subject: below --
2018-11-29 19:12 block and nvme polling improvements V2 Christoph Hellwig
2018-11-29 19:12 ` [PATCH 01/13] block: move queues types to the block layer Christoph Hellwig
2018-11-29 19:12   ` Christoph Hellwig
2018-11-29 19:50   ` Jens Axboe
2018-11-29 19:50     ` Jens Axboe
2018-11-30  7:56     ` Christoph Hellwig
2018-11-30  7:56       ` Christoph Hellwig
2018-11-30 15:20       ` Jens Axboe
2018-11-30 15:20         ` Jens Axboe
2018-11-30 15:21         ` Christoph Hellwig
2018-11-30 15:21           ` Christoph Hellwig
2018-11-29 20:19   ` Keith Busch
2018-11-29 20:19     ` Keith Busch
2018-11-29 20:25     ` Jens Axboe
2018-11-29 20:25       ` Jens Axboe
2018-11-30  8:00     ` Christoph Hellwig
2018-11-30  8:00       ` Christoph Hellwig
2018-11-30 14:40       ` Keith Busch
2018-11-30 14:40         ` Keith Busch
2018-11-30 15:20       ` Jens Axboe
2018-11-30 15:20         ` Jens Axboe
2018-11-21 16:23 block and nvme polling improvements Christoph Hellwig
2018-11-21 16:23 ` [PATCH 01/13] block: move queues types to the block layer Christoph Hellwig
2018-11-21 16:23   ` Christoph Hellwig

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.