Re: [PATCH 06/18] compat: backport ASYNC_DOMAIN_EXCLUSIVE()

Re: [PATCH 06/18] compat: backport ASYNC_DOMAIN_EXCLUSIVE()

[Luis R. Rodriguez]

On Wed, Apr 10, 2013 at 10:20 AM, Johannes Berg
<johannes@sipsolutions.net> wrote:
> On Wed, 2013-04-10 at 10:13 -0700, Luis R. Rodriguez wrote:
>> On Wed, Apr 10, 2013 at 6:22 AM, Johannes Berg
>> <johannes@sipsolutions.net> wrote:
>> > On Wed, 2013-04-10 at 04:35 -0700, Luis R. Rodriguez wrote:
>> >> From: "Luis R. Rodriguez" <mcgrof@do-not-panic.com>
>> >>
>> >> For older kernels this is simply a matter of using
>> >> LIST_HEAD.
>> >
>> > Hmm, how is that useful if it will not have any effect?
>>
>> The core of the handling is done in the internal asynch code and
>> that's kept hidden from the users of this -- so far.
>
> I guess I'd have to review the async API,

Yeap, reviewing the commit noted would help too.

> but now this wouldn't do
> anything?

Nope.

> What's the use of just this when you don't have things like
> async_schedule_domain() and async_synchronize_full_domain(), regulator
> stuff wouldn't compile I think?

You mean is not having the full asynch that deals with all registered
domains likely to have an issue on the useres of
async_synchronize_full_domain() ? Lets better ask Dan.

> I guess I'm a bit confused what kind of effect this has.

Nothing, really. If we had a user of it in other ways we'd fail
compiling anyway but right now its a simple direct replacement.

  Luis

Re: [PATCH 06/18] compat: backport ASYNC_DOMAIN_EXCLUSIVE()

[Johannes Berg]

On Wed, 2013-04-10 at 10:26 -0700, Luis R. Rodriguez wrote:

> > I guess I'd have to review the async API,
> 
> Yeap, reviewing the commit noted would help too.

Yeah ... :)

> > What's the use of just this when you don't have things like
> > async_schedule_domain() and async_synchronize_full_domain(), regulator
> > stuff wouldn't compile I think?
> 
> You mean is not having the full asynch that deals with all registered
> domains likely to have an issue on the useres of
> async_synchronize_full_domain() ? Lets better ask Dan.

I don't know. However it seems that in order to have an ASYNC_DOMAIN()
or ASYNC_DOMAIN_EXCLUSIVE() you always need to *do* something with it,
so for that you'd also need the functions async_schedule_domain() and
async_synchronize_full_domain() or similar, at least, no?

The point here seems to be making boot faster by starting a bunch of
async probing inside a domain, and then you wait for the entire domain,
so everything that's in that domain can be done in parallel.

Say for example you have 20 SCSI drives. If you look at them serially
then you'd waste much time waiting for the drives. The point here
appears to be that you create a domain (using this macro), then add all
the drives to the domain and then wait for the domain to finish.

However, it seems entirely pointless to backport just a small part of
the API?

johannes

Re: [PATCH 06/18] compat: backport ASYNC_DOMAIN_EXCLUSIVE()

[Luis R. Rodriguez]

On Wed, Apr 10, 2013 at 11:20 AM, Johannes Berg
<johannes@sipsolutions.net> wrote:
> On Wed, 2013-04-10 at 10:26 -0700, Luis R. Rodriguez wrote:
>
>> > I guess I'd have to review the async API,
>>
>> Yeap, reviewing the commit noted would help too.
>
> Yeah ... :)
>
>> > What's the use of just this when you don't have things like
>> > async_schedule_domain() and async_synchronize_full_domain(), regulator
>> > stuff wouldn't compile I think?
>>
>> You mean is not having the full asynch that deals with all registered
>> domains likely to have an issue on the useres of
>> async_synchronize_full_domain() ? Lets better ask Dan.
>
> I don't know. However it seems that in order to have an ASYNC_DOMAIN()
> or ASYNC_DOMAIN_EXCLUSIVE() you always need to *do* something with it,
> so for that you'd also need the functions async_schedule_domain() and
> async_synchronize_full_domain() or similar, at least, no?
>
> The point here seems to be making boot faster by starting a bunch of
> async probing inside a domain, and then you wait for the entire domain,
> so everything that's in that domain can be done in parallel.
>
> Say for example you have 20 SCSI drives. If you look at them serially
> then you'd waste much time waiting for the drives. The point here
> appears to be that you create a domain (using this macro), then add all
> the drives to the domain and then wait for the domain to finish.
>
> However, it seems entirely pointless to backport just a small part of
> the API?

Oh I agree don't get me wrong, however porting kernel/async.c seems
like a rather separate effort worth considering. As-is though I have
not seen any negative impact though to keep older subsystems from
compiling, ie its a no-op for older kernels as I see it.

 Luis

Re: [PATCH 06/18] compat: backport ASYNC_DOMAIN_EXCLUSIVE()

[Johannes Berg]

On Wed, 2013-04-10 at 12:19 -0700, Luis R. Rodriguez wrote:

> > However, it seems entirely pointless to backport just a small part of
> > the API?
> 
> Oh I agree don't get me wrong, however porting kernel/async.c seems
> like a rather separate effort worth considering. As-is though I have
> not seen any negative impact though to keep older subsystems from
> compiling, ie its a no-op for older kernels as I see it.

I guess that's what I don't understand -- I don't see usages of
ASYNC_DOMAIN_EXCLUSIVE in any header files, and in e.g. regulator/core.c
you'd also need the functions async_schedule_domain() etc. So where does
this help even compiling?

johannes

Re: [PATCH 06/18] compat: backport ASYNC_DOMAIN_EXCLUSIVE()

[Luis R. Rodriguez]

On Wed, Apr 10, 2013 at 12:27 PM, Johannes Berg
<johannes@sipsolutions.net> wrote:
> On Wed, 2013-04-10 at 12:19 -0700, Luis R. Rodriguez wrote:
>
>> > However, it seems entirely pointless to backport just a small part of
>> > the API?
>>
>> Oh I agree don't get me wrong, however porting kernel/async.c seems
>> like a rather separate effort worth considering. As-is though I have
>> not seen any negative impact though to keep older subsystems from
>> compiling, ie its a no-op for older kernels as I see it.
>
> I guess that's what I don't understand -- I don't see usages of
> ASYNC_DOMAIN_EXCLUSIVE in any header files, and in e.g. regulator/core.c
> you'd also need the functions async_schedule_domain() etc. So where does
> this help even compiling?

You know what, sorry this was left over from when I tried to backport
the regulatory to the core of compat, and since I decided to not even
go there given that it relies on init sections on the vmlinux we can
safely discard this patch (although what I said still hold, just not
needed).

  Luis

Re: [PATCH 06/18] compat: backport ASYNC_DOMAIN_EXCLUSIVE()

[Johannes Berg]

On Wed, 2013-04-10 at 12:32 -0700, Luis R. Rodriguez wrote:

> >> Oh I agree don't get me wrong, however porting kernel/async.c seems
> >> like a rather separate effort worth considering. As-is though I have
> >> not seen any negative impact though to keep older subsystems from
> >> compiling, ie its a no-op for older kernels as I see it.
> >
> > I guess that's what I don't understand -- I don't see usages of
> > ASYNC_DOMAIN_EXCLUSIVE in any header files, and in e.g. regulator/core.c
> > you'd also need the functions async_schedule_domain() etc. So where does
> > this help even compiling?
> 
> You know what, sorry this was left over from when I tried to backport
> the regulatory to the core of compat, and since I decided to not even
> go there given that it relies on init sections on the vmlinux we can
> safely discard this patch (although what I said still hold, just not
> needed).

Ok. Yeah after looking at the users I actually do agree this won't
really hurt, but it seemed it doesn't help anything at all hence my
confusion... :)

johannes

Re: [PATCH 06/18] compat: backport ASYNC_DOMAIN_EXCLUSIVE()

[Luis R. Rodriguez]

On Wed, Apr 10, 2013 at 12:39 PM, Johannes Berg
<johannes@sipsolutions.net> wrote:
> On Wed, 2013-04-10 at 12:32 -0700, Luis R. Rodriguez wrote:
>
>> >> Oh I agree don't get me wrong, however porting kernel/async.c seems
>> >> like a rather separate effort worth considering. As-is though I have
>> >> not seen any negative impact though to keep older subsystems from
>> >> compiling, ie its a no-op for older kernels as I see it.
>> >
>> > I guess that's what I don't understand -- I don't see usages of
>> > ASYNC_DOMAIN_EXCLUSIVE in any header files, and in e.g. regulator/core.c
>> > you'd also need the functions async_schedule_domain() etc. So where does
>> > this help even compiling?
>>
>> You know what, sorry this was left over from when I tried to backport
>> the regulatory to the core of compat, and since I decided to not even
>> go there given that it relies on init sections on the vmlinux we can
>> safely discard this patch (although what I said still hold, just not
>> needed).
>
> Ok. Yeah after looking at the users I actually do agree this won't
> really hurt, but it seemed it doesn't help anything at all hence my
> confusion... :)

Yea its fine to drop for now.

  Luis