diff mbox

[v2,all,00/14] use of_platform_default_populate() to populate default bus

Message ID CAL_JsqK34MOSkdJb1Sqq3CzYSz5xGeEnuThS_GmfFXMHX-8d9A@mail.gmail.com (mailing list archive)
State New, archived
Headers show

Commit Message

Rob Herring April 19, 2016, 9:10 p.m. UTC
On Wed, Mar 16, 2016 at 2:35 AM, Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
> Use helper of_platform_default_populate() in linux/of_platform
> when possible, instead of calling of_platform_populate() with
> the default match table.
>
> Then it is possible for driver code build as a module, and no
> need to export of_default_bus_match_table anymore.
>
> This patchset is based on Linux 4.5.
>
> Changes since v1:
> - Add ack, test and review
> - Remove v1 patch 14/15, the similar patch has posted and applied,
>   see https://patchwork.kernel.org/patch/8096991/
>
> v1:
> - https://lkml.org/lkml/2016/1/26/1123

I was going to apply this, but I've thought about this some more and
it is a lot of tree wide churn for little gain. Most of the call sites
here I want to get rid of by making of_platform_populate callable
multiple times. The infrastructure is almost there, but I think we
just need something like this:


Then we can call of_platform_populate from the DT core and remove all
the default calls.

Rob

Comments

Kefeng Wang April 21, 2016, 7:55 a.m. UTC | #1
Hi Rob,

On 2016/4/20 5:10, Rob Herring wrote:
> On Wed, Mar 16, 2016 at 2:35 AM, Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>> Use helper of_platform_default_populate() in linux/of_platform
>> when possible, instead of calling of_platform_populate() with
>> the default match table.
>>
>> Then it is possible for driver code build as a module, and no
>> need to export of_default_bus_match_table anymore.
>>
>> This patchset is based on Linux 4.5.
>>
>> Changes since v1:
>> - Add ack, test and review
>> - Remove v1 patch 14/15, the similar patch has posted and applied,
>>   see https://patchwork.kernel.org/patch/8096991/
>>
>> v1:
>> - https://lkml.org/lkml/2016/1/26/1123
> 
> I was going to apply this, but I've thought about this some more and
> it is a lot of tree wide churn for little gain. Most of the call sites
> here I want to get rid of by making of_platform_populate callable
> multiple times. The infrastructure is almost there, but I think we
> just need something like this:
> 
> diff --git a/drivers/of/platform.c b/drivers/of/platform.c
> index 8d103e4..b25e710 100644
> --- a/drivers/of/platform.c
> +++ b/drivers/of/platform.c
> @@ -345,6 +345,9 @@ static int of_platform_bus_create(struct device_node *bus,
>                 return 0;
>         }
> 
> +       if (of_node_check_flag(bus, OF_POPULATED_BUS))
> +               return 0;
> +
>         auxdata = of_dev_lookup(lookup, bus);
>         if (auxdata) {
>                 bus_id = auxdata->name;
> 
> Then we can call of_platform_populate from the DT core and remove all
> the default calls.

OK, it seems that we can drop all the of_of_platform_default_populate in arch code
by this changes, will do it.

But replace of_platform_populate() with the default match table by of_platform_default_populate
is a cleanup, and especially for the driver, we can not need export of_default_bus_match_table
after that.

There is a v3 patchset[1], it is based 4.6-rc2 with some updated, should this be merge firstly,
then drop all the default of_platform_populate in arch code, and I think the of_iommu_init() in
arch code could do the same thing.

If not, I will repost the patches with new ways, but still exists the part of unexport
of_default_bus_match_table by helper of_platform_default_populate().

Hope the v3 patchset could be merge in advance, after all, it has been reviewed fully. Thanks a lot.

BRs,
Kefeng


[1] https://lkml.org/lkml/2016/4/5/55

> 
> Rob
> 
> .
>
Rob Herring April 21, 2016, 1:10 p.m. UTC | #2
On Thu, Apr 21, 2016 at 2:55 AM, Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
> Hi Rob,
>
> On 2016/4/20 5:10, Rob Herring wrote:
>> On Wed, Mar 16, 2016 at 2:35 AM, Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>>> Use helper of_platform_default_populate() in linux/of_platform
>>> when possible, instead of calling of_platform_populate() with
>>> the default match table.
>>>
>>> Then it is possible for driver code build as a module, and no
>>> need to export of_default_bus_match_table anymore.
>>>
>>> This patchset is based on Linux 4.5.
>>>
>>> Changes since v1:
>>> - Add ack, test and review
>>> - Remove v1 patch 14/15, the similar patch has posted and applied,
>>>   see https://patchwork.kernel.org/patch/8096991/
>>>
>>> v1:
>>> - https://lkml.org/lkml/2016/1/26/1123
>>
>> I was going to apply this, but I've thought about this some more and
>> it is a lot of tree wide churn for little gain. Most of the call sites
>> here I want to get rid of by making of_platform_populate callable
>> multiple times. The infrastructure is almost there, but I think we
>> just need something like this:
>>
>> diff --git a/drivers/of/platform.c b/drivers/of/platform.c
>> index 8d103e4..b25e710 100644
>> --- a/drivers/of/platform.c
>> +++ b/drivers/of/platform.c
>> @@ -345,6 +345,9 @@ static int of_platform_bus_create(struct device_node *bus,
>>                 return 0;
>>         }
>>
>> +       if (of_node_check_flag(bus, OF_POPULATED_BUS))
>> +               return 0;
>> +
>>         auxdata = of_dev_lookup(lookup, bus);
>>         if (auxdata) {
>>                 bus_id = auxdata->name;
>>
>> Then we can call of_platform_populate from the DT core and remove all
>> the default calls.
>
> OK, it seems that we can drop all the of_of_platform_default_populate in arch code
> by this changes, will do it.
>
> But replace of_platform_populate() with the default match table by of_platform_default_populate
> is a cleanup, and especially for the driver, we can not need export of_default_bus_match_table
> after that.
>
> There is a v3 patchset[1], it is based 4.6-rc2 with some updated, should this be merge firstly,
> then drop all the default of_platform_populate in arch code, and I think the of_iommu_init() in
> arch code could do the same thing.
>
> If not, I will repost the patches with new ways, but still exists the part of unexport
> of_default_bus_match_table by helper of_platform_default_populate().
>
> Hope the v3 patchset could be merge in advance, after all, it has been reviewed fully. Thanks a lot.

We'd be touching the same line twice in a bunch of places if we apply
this series first, so it makes more sense to remove all these calls
first and then use of_platform_default_populate where it is still
needed.

Rob
Kefeng Wang April 22, 2016, 2:47 a.m. UTC | #3
On 2016/4/21 21:10, Rob Herring wrote:
> On Thu, Apr 21, 2016 at 2:55 AM, Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>> Hi Rob,
>>
>> On 2016/4/20 5:10, Rob Herring wrote:
>>> On Wed, Mar 16, 2016 at 2:35 AM, Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>>>> Use helper of_platform_default_populate() in linux/of_platform
>>>> when possible, instead of calling of_platform_populate() with
>>>> the default match table.
>>>>
>>>> Then it is possible for driver code build as a module, and no
>>>> need to export of_default_bus_match_table anymore.
>>>>
>>>> This patchset is based on Linux 4.5.
>>>>
>>>> Changes since v1:
>>>> - Add ack, test and review
>>>> - Remove v1 patch 14/15, the similar patch has posted and applied,
>>>>   see https://patchwork.kernel.org/patch/8096991/
>>>>
>>>> v1:
>>>> - https://lkml.org/lkml/2016/1/26/1123
>>>
>>> I was going to apply this, but I've thought about this some more and
>>> it is a lot of tree wide churn for little gain. Most of the call sites
>>> here I want to get rid of by making of_platform_populate callable
>>> multiple times. The infrastructure is almost there, but I think we
>>> just need something like this:
>>>
>>> diff --git a/drivers/of/platform.c b/drivers/of/platform.c
>>> index 8d103e4..b25e710 100644
>>> --- a/drivers/of/platform.c
>>> +++ b/drivers/of/platform.c
>>> @@ -345,6 +345,9 @@ static int of_platform_bus_create(struct device_node *bus,
>>>                 return 0;
>>>         }
>>>
>>> +       if (of_node_check_flag(bus, OF_POPULATED_BUS))
>>> +               return 0;
>>> +
>>>         auxdata = of_dev_lookup(lookup, bus);
>>>         if (auxdata) {
>>>                 bus_id = auxdata->name;
>>>
>>> Then we can call of_platform_populate from the DT core and remove all
>>> the default calls.
>>
>> OK, it seems that we can drop all the of_of_platform_default_populate in arch code
>> by this changes, will do it.
>>
>> But replace of_platform_populate() with the default match table by of_platform_default_populate
>> is a cleanup, and especially for the driver, we can not need export of_default_bus_match_table
>> after that.
>>
>> There is a v3 patchset[1], it is based 4.6-rc2 with some updated, should this be merge firstly,
>> then drop all the default of_platform_populate in arch code, and I think the of_iommu_init() in
>> arch code could do the same thing.
>>
>> If not, I will repost the patches with new ways, but still exists the part of unexport
>> of_default_bus_match_table by helper of_platform_default_populate().
>>
>> Hope the v3 patchset could be merge in advance, after all, it has been reviewed fully. Thanks a lot.
> 
> We'd be touching the same line twice in a bunch of places if we apply
> this series first, so it makes more sense to remove all these calls
> first and then use of_platform_default_populate where it is still
> needed.

Agree, will repost them.

BRs,
Kefeng

> 
> Rob
> 
> .
>
diff mbox

Patch

diff --git a/drivers/of/platform.c b/drivers/of/platform.c
index 8d103e4..b25e710 100644
--- a/drivers/of/platform.c
+++ b/drivers/of/platform.c
@@ -345,6 +345,9 @@  static int of_platform_bus_create(struct device_node *bus,
                return 0;
        }

+       if (of_node_check_flag(bus, OF_POPULATED_BUS))
+               return 0;
+
        auxdata = of_dev_lookup(lookup, bus);
        if (auxdata) {
                bus_id = auxdata->name;