[Qemu-devel] [PATCH 0/6] block: byte-based AIO read/write

Eric Blake posted 6 patches 7 years, 7 months ago
Failed in applying to current master (apply log)
There is a newer version of this series
include/block/block_int.h |  8 +++---
include/block/raw-aio.h   |  2 +-
block/io.c                | 26 ++++++++++++-------
block/file-win32.c        | 36 +++++++++++++-------------
block/null.c              | 66 ++++++++++++++++++++++++++---------------------
block/rbd.c               | 36 ++++++++++++--------------
block/vxhs.c              | 36 +++++++++++---------------
block/win32-aio.c         |  5 ++--
8 files changed, 109 insertions(+), 106 deletions(-)
[Qemu-devel] [PATCH 0/6] block: byte-based AIO read/write
Posted by Eric Blake 7 years, 7 months ago
While we would prefer that block drivers use coroutines instead
of aio callbacks, it is a fairly easy exercise to prove that
all existing drivers with aio callbacks are merely scaling
from bytes into sectors and back to bytes.  So, even though I
am not set up to completely run (or even compile-test) this
full series, it seems pretty straightforward to change the
signature to quit playing games with pointless scaling.

Note that except for the null-aio driver, I intentionally did
NOT try and change the request_alignment from the block layer's
default of 512 (it defaults to 1 only for drivers that use
coroutines).

(And along the way, I got my docker-test-mingw@fedora working;
thanks to the help I got on IRC)

Eric Blake (6):
  block: Support byte-based aio callbacks
  file-win32: Switch to byte-based callbacks
  null: Switch to byte-based read/write
  rbd: Switch to byte-based callbacks
  vxhs: Switch to byte-based callbacks
  block: Drop last of the sector-based aio callbacks

 include/block/block_int.h |  8 +++---
 include/block/raw-aio.h   |  2 +-
 block/io.c                | 26 ++++++++++++-------
 block/file-win32.c        | 36 +++++++++++++-------------
 block/null.c              | 66 ++++++++++++++++++++++++++---------------------
 block/rbd.c               | 36 ++++++++++++--------------
 block/vxhs.c              | 36 +++++++++++---------------
 block/win32-aio.c         |  5 ++--
 8 files changed, 109 insertions(+), 106 deletions(-)

-- 
2.14.3


Re: [Qemu-devel] [PATCH 0/6] block: byte-based AIO read/write
Posted by Eric Blake 7 years, 5 months ago
On 02/15/2018 01:28 PM, Eric Blake wrote:
> While we would prefer that block drivers use coroutines instead
> of aio callbacks, it is a fairly easy exercise to prove that
> all existing drivers with aio callbacks are merely scaling
> from bytes into sectors and back to bytes.  So, even though I
> am not set up to completely run (or even compile-test) this
> full series, it seems pretty straightforward to change the
> signature to quit playing games with pointless scaling.
> 
> Note that except for the null-aio driver, I intentionally did
> NOT try and change the request_alignment from the block layer's
> default of 512 (it defaults to 1 only for drivers that use
> coroutines).
> 
> (And along the way, I got my docker-test-mingw@fedora working;
> thanks to the help I got on IRC)

ping

> 
> Eric Blake (6):
>   block: Support byte-based aio callbacks
>   file-win32: Switch to byte-based callbacks
>   null: Switch to byte-based read/write
>   rbd: Switch to byte-based callbacks
>   vxhs: Switch to byte-based callbacks
>   block: Drop last of the sector-based aio callbacks
> 
>  include/block/block_int.h |  8 +++---
>  include/block/raw-aio.h   |  2 +-
>  block/io.c                | 26 ++++++++++++-------
>  block/file-win32.c        | 36 +++++++++++++-------------
>  block/null.c              | 66 ++++++++++++++++++++++++++---------------------
>  block/rbd.c               | 36 ++++++++++++--------------
>  block/vxhs.c              | 36 +++++++++++---------------
>  block/win32-aio.c         |  5 ++--
>  8 files changed, 109 insertions(+), 106 deletions(-)
> 

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org

Re: [Qemu-devel] [PATCH 0/6] block: byte-based AIO read/write
Posted by John Snow 7 years, 5 months ago

On 04/24/2018 11:02 AM, Eric Blake wrote:
> On 02/15/2018 01:28 PM, Eric Blake wrote:
>> While we would prefer that block drivers use coroutines instead
>> of aio callbacks, it is a fairly easy exercise to prove that
>> all existing drivers with aio callbacks are merely scaling
>> from bytes into sectors and back to bytes.  So, even though I
>> am not set up to completely run (or even compile-test) this
>> full series, it seems pretty straightforward to change the
>> signature to quit playing games with pointless scaling.
>>
>> Note that except for the null-aio driver, I intentionally did
>> NOT try and change the request_alignment from the block layer's
>> default of 512 (it defaults to 1 only for drivers that use
>> coroutines).
>>
>> (And along the way, I got my docker-test-mingw@fedora working;
>> thanks to the help I got on IRC)
> 
> ping
> 

One of your replies to patch 3 suggests you were going to send a v2 --
is that not the case? Do you want deeper review of v1 first?

--js