Message-ID: <cde0064d-83dd-4a7f-8921-053c25aae08b@kernel.org>
Date: Fri, 23 Aug 2024 14:30:58 +0300
From: Roger Quadros <rogerq@...nel.org>
To: "Anwar, Md Danish" <a0501179@...com>, MD Danish Anwar
 <danishanwar@...com>, Dan Carpenter <dan.carpenter@...aro.org>,
 Andrew Lunn <andrew@...n.ch>, Jan Kiszka <jan.kiszka@...mens.com>,
 Vignesh Raghavendra <vigneshr@...com>,
 Javier Carrasco <javier.carrasco.cruz@...il.com>,
 Jacob Keller <jacob.e.keller@...el.com>, Diogo Ivo <diogo.ivo@...mens.com>,
 Simon Horman <horms@...nel.org>, Richard Cochran <richardcochran@...il.com>,
 Paolo Abeni <pabeni@...hat.com>, Jakub Kicinski <kuba@...nel.org>,
 Eric Dumazet <edumazet@...gle.com>, "David S. Miller" <davem@...emloft.net>
Cc: linux-kernel@...r.kernel.org, netdev@...r.kernel.org,
 linux-arm-kernel@...ts.infradead.org, srk@...com
Subject: Re: [PATCH net-next v2 1/7] net: ti: icssg-prueth: Enable IEP1

Hi,

On 22/08/2024 15:12, Anwar, Md Danish wrote:
> 
> 
> On 8/22/2024 4:57 PM, Roger Quadros wrote:
>>
>>
>> On 22/08/2024 08:52, MD Danish Anwar wrote:
>>>
>>>
>>> On 21/08/24 5:23 pm, Roger Quadros wrote:
>>>>
>>>>
>>>> On 21/08/2024 14:33, Anwar, Md Danish wrote:
>>>>> Hi Roger,
>>>>>
>>>>> On 8/21/2024 4:57 PM, Roger Quadros wrote:
>>>>>> Hi,
>>>>>>
>>>>>> On 13/08/2024 10:42, MD Danish Anwar wrote:
>>>>>>> IEP1 is needed by firmware to enable FDB learning and FDB ageing.
>>>>>>
>>>>>> Required by which firmware?
>>>>>>
>>>>>
>>>>> IEP1 is needed by all ICSSG firmwares (Dual EMAC / Switch / HSR)
>>>>>
>>>>>> Does dual-emac firmware need this?
>>>>>>
>>>>>
>>>>> Yes, Dual EMAC firmware needs IEP1 to enabled.
>>>>
>>>> Then this need to be a bug fix?
>>>
>>> Correct, this is in fact a bug. But IEP1 is also needed by HSR firmware
>>> so I thought of keeping this patch with HSR series. As HSR will be
>>> completely broken if IEP1 is not enabled.
>>>
>>> I didn't want to post two patches one as bug fix to net and one part of
>>> HSR to net-next thus I thought of keeping this patch in this series only.
>>
>> Bug fixes need to be posted earlier as they can get accepted sooner and
>> even back-ported to stable. You also need to add the Fixes tag.
>>
> 
> Yes I understand that. The problem was I will need to send the patch
> twice once as bug fix to net/main, once as part of this series to
> net-next/main and drop it from this series once the patch gets merged to
> net and is synced to net-next. Since I cannot post this series without
> this patch as the HSR feature will get broken.

HSR feature is not yet there so nothing will be broken. You can mention
in cover letter that a separate patch is required for functionality.

> 
> Or I need to post this to net/main and wait for it to be part of
> net-next and then only I can re-post this series. So to avoid all these
> I thought of only posting it as part of this series. This is not a major
> bug and it will be okay from feature perspective if this bug gets merged
> via net-next.
> 

If there is no build dependency I don't see why you need to wait.

> What do you suggest now? Should I post the bug fix to net/main and post
> this seires without this patch or wait for the bug fix to sync then only
> post this series?
> 

please see below.

>>>
>>>> What is the impact if IEP1 is not enabled for dual emac.
>>>>
>>>
>>> Without IEP1 enabled, Crash is seen on AM64x 10M link when connecting /
>>> disconnecting multiple times. On AM65x IEP1 was always enabled because
>>
>> In that ase you need to enable quirk_10m_link_issue for AM64x platform.
> 
> Correct. But since for all ICSSG supported platform quirk_10m_link_issue
>  needs to be enabled. Which in turn will enable IEP1. I think it's
> better enable IEP1 directly without any condition.
> 
> IEP1 is also needed for Switch and HSR firmwares so I thought directly
> enabling it instead of enabling it inside the if check
> `prueth->pdata.quirk_10m_link_issue` would be better idea.
> 
> What do you suggest here? Will setting `quirk_10m_link_issue` as true
> for AM64x will be a better approach or always enabling IEP1 without any
> if check will be better approach for this?

I would suggest that you first send a bug fix patch for AM64x which sets
quirk_10m_link_issue for AM64x. This should make it into the next -rc
and eventually stable.

Then in your HSR series, you can decide if you want to conditionally
enable it for HSR/switch mode or permanently enable it regardless.

> 
>> I understand that IEP1 is not required for 100M/1G.
>>
>>> `prueth->pdata.quirk_10m_link_issue` was true. FDB learning and FDB
>>> ageing will also get impacted if IEP1 is not enabled.
>>
>> Is FDB learning and ageing involved in dual Emac mode?
>>
> 
> Yes FDB learning and ageing is involved in dual EMAC mode as well.
> 
>>>
>>>>>
>>>>>>> Always enable IEP1
>>>>>>>
>>>>>>> Signed-off-by: MD Danish Anwar <danishanwar@...com>
>>>>>>> ---
>>>>
>>>
>>
> 

-- 
cheers,
-roger

Powered by blists - more mailing lists