AIO for SG

AIO for SG

[Mark Lobo]

Guys,
Is there a patch available which adds the aio
interface to SG? (aio_read, aio_complete etc) 

TIA,
Mark

__________________________________
Do you Yahoo!?
Protect your identity with Yahoo! Mail AddressGuard
http://antispam.yahoo.com/whatsnewfree

Re: AIO for SG

[Douglas Gilbert]

Mark Lobo wrote:
> Guys,
> Is there a patch available which adds the aio
> interface to SG? (aio_read, aio_complete etc) 

Mark,
I have got one but it's against lk 2.5.69 . It is relatively
simple (i.e. supports aio_write and aio_complete but not
aio_read).

If I get time I'll send a patch against lk 2.6.0-test9 later
today.

Doug Gilbert

Re: AIO for SG

[Douglas Gilbert]

[-- Attachment #1: Type: text/plain, Size: 853 bytes --]

Mark Lobo wrote:
> Guys,
> Is there a patch available which adds the aio
> interface to SG? (aio_read, aio_complete etc) 

Mark,
I have ported the aio support from lk 2.5.69 to my latest
version of sg which I sent to this list a few days ago.

The 2 attachments:
   1) incremental against my patch a few days ago. This makes
      it simpler to see the changes associated with aio
   2) against the version of sg found in lk 2.6.0-test9 (so
      it includes my cdev and seqfile patches)

The sg3_utils-1.05 package found at http://www.torque.net/sg
has a file called sg_simple_aio.c ** in its "examples" directory.
You need libaio to link it. Links ok with libaio-0.3.92 but
libaio-0.3.93 gives undefines (seems to be related to i64, ppc
etc support added in 0.3.93, strange).

** beware of the "sleep(3)" before read() [best remove it].

Doug Gibert


[-- Attachment #2: sg_aio_260t9dpg.diff.gz --]
[-- Type: application/x-gzip, Size: 2832 bytes --]

[-- Attachment #3: sg_aio_260t9.diff.gz --]
[-- Type: application/x-gzip, Size: 8222 bytes --]

Re: AIO for SG

[Mark Lobo]

Douglas,
Thanks, I will definitely try it out.

I had one question about the aio calls though. I was
looking through the kernel sources after applying the
patch, and then looked at the examples you have
provided. How does the io_setcallback work? The kiocb
and the aiocb structure had no way to specify a
callback function in the kernel/usermode. What exactly
does aio_complete do and how is the callback done?
Atleast until SG, there was no way of calling back,
didnt look anywhere else.How does it work from the
kernel?

TIA
Mark

--- Douglas Gilbert <dougg@torque.net> wrote:
> Mark Lobo wrote:
> > Guys,
> > Is there a patch available which adds the aio
> > interface to SG? (aio_read, aio_complete etc) 
> 
> Mark,
> I have ported the aio support from lk 2.5.69 to my
> latest
> version of sg which I sent to this list a few days
> ago.
> 
> The 2 attachments:
>    1) incremental against my patch a few days ago.
> This makes
>       it simpler to see the changes associated with
> aio
>    2) against the version of sg found in lk
> 2.6.0-test9 (so
>       it includes my cdev and seqfile patches)
> 
> The sg3_utils-1.05 package found at
> http://www.torque.net/sg
> has a file called sg_simple_aio.c ** in its
> "examples" directory.
> You need libaio to link it. Links ok with
> libaio-0.3.92 but
> libaio-0.3.93 gives undefines (seems to be related
> to i64, ppc
> etc support added in 0.3.93, strange).
> 
> ** beware of the "sleep(3)" before read() [best
> remove it].
> 
> Doug Gibert
> 
> 

> ATTACHMENT part 2 application/x-gzip
name=sg_aio_260t9dpg.diff.gz


> ATTACHMENT part 3 application/x-gzip
name=sg_aio_260t9.diff.gz



__________________________________
Do you Yahoo!?
Protect your identity with Yahoo! Mail AddressGuard
http://antispam.yahoo.com/whatsnewfree