by Alex Elder

[permalink] [raw]

Subject: Re: [PATCH net-next 2/6] net: ipa: kill gsi->virt_raw

On 2/17/23 7:04 AM, Alex Elder wrote:
> On 2/17/23 5:57 AM, Alexander Lobakin wrote:
>>>> just devm_platform_ioremap_resource_byname() be used here for
>>>> simplicity?
>>> Previously, virt_raw would be the "real" re-mapped pointer, and then
>>> virt would be adjusted downward from that. It was a weird thing to
>>> do, because the result pointed to a non-mapped address. But all uses
>>> of the virt pointer added an offset that was enough to put the result
>>> into the mapped range.
>>>
>>> The new code updates all offsets to account for what the adjustment
>>> previously did. The test that got removed isn't necessary any more.
>> Yeah I got it, just asked that maybe you can now use
>> platform_ioremap_resource_byname() instead of
>> platform_get_resource_byname() + ioremap() :)
>
> Sorry, I focused on the "devm" part and not this part.
> Yes I like that, but let me do that as a follow-on
> patch, and I think I can do it in more than this
> spot (possibly three, but I have to look closely).

Looking at this today, the only OF functions that look up a
resource and I/O remap it in one call are devm_*() variants.
There is no platform_ioremap_resource_byname() function.

One that's available is devm_platform_ioremap_resource_byname(),
which could possibly be used in the two locations that call
platform_get_resource_byname() followed by ioremap().

As I said earlier, if I were to use any devm_*() function
calls the driver, I would want to convert *everything* to
use devm_*() variants, and I have no plans to do that at
this time.

So I will not be implementing your suggestion.

-Alex

> Thanks.
>
> -Alex
>
>

2023-03-06 10:32:55

by Alexander Lobakin

[permalink] [raw]

Subject: Re: [PATCH net-next 2/6] net: ipa: kill gsi->virt_raw

From: Alex Elder <[email protected]>
Date: Sun, 5 Mar 2023 10:58:19 -0600

> On 2/17/23 7:04 AM, Alex Elder wrote:
>> On 2/17/23 5:57 AM, Alexander Lobakin wrote:
>>>>> just devm_platform_ioremap_resource_byname() be used here for
>>>>> simplicity?
>>>> Previously, virt_raw would be the "real" re-mapped pointer, and then
>>>> virt would be adjusted downward from that. It was a weird thing to
>>>> do, because the result pointed to a non-mapped address. But all uses
>>>> of the virt pointer added an offset that was enough to put the result
>>>> into the mapped range.
>>>>
>>>> The new code updates all offsets to account for what the adjustment
>>>> previously did. The test that got removed isn't necessary any more.
>>> Yeah I got it, just asked that maybe you can now use
>>> platform_ioremap_resource_byname() instead of
>>> platform_get_resource_byname() + ioremap() :)
>>
>> Sorry, I focused on the "devm" part and not this part.
>> Yes I like that, but let me do that as a follow-on
>> patch, and I think I can do it in more than this
>> spot (possibly three, but I have to look closely).
>
> Looking at this today, the only OF functions that look up a
> resource and I/O remap it in one call are devm_*() variants.
> There is no platform_ioremap_resource_byname() function.
>
> One that's available is devm_platform_ioremap_resource_byname(),
> which could possibly be used in the two locations that call
> platform_get_resource_byname() followed by ioremap().
>
> As I said earlier, if I were to use any devm_*() function
> calls the driver, I would want to convert *everything* to
> use devm_*() variants, and I have no plans to do that at
> this time.
>
> So I will not be implementing your suggestion.

Sure. It's fully up to the developers as it doesn't make the code worse
in any way.

>
> -Alex
>
>> Thanks.
>>
>> -Alex
>>
>>
>
Thanks,
Olek