Re: [PATCH v2 3/3] ata: ahci: Skip 200 ms debounce delay for AMD 300 Series Chipset SATA Controller

[Damien Le Moal <damien.lemoal@opensource.wdc.com>]

On 3/31/22 23:42, Paul Menzel wrote:
> Dear Damien,
> 
> 
> Am 23.03.22 um 09:36 schrieb Paul Menzel:
> 
>> Am 23.03.22 um 09:24 schrieb Damien Le Moal:
>>> On 3/23/22 15:55, Paul Menzel wrote:
>>
>>>> Am 23.03.22 um 06:01 schrieb Damien Le Moal:
>>>>> On 3/22/22 06:51, Limonciello, Mario wrote:
>>
>>>>>>> -----Original Message-----
>>>>>>> From: Paul Menzel <pmenzel@molgen.mpg.de>
>>>>>>> Sent: Monday, March 21, 2022 16:25
>>>>
>>>> […]
>>>>
>>>>>> I seem to recall that we were talking about trying to drop the 
>>>>>> debounce delay for everything, weren't we?
>>>>>>
>>>>>> So perhaps it would be right to add a 4th patch in the series to do
>>>>>> just that.  Then If this turns out to be problematic for
>>>>>> anything other than the controllers in the series that you
>>>>>> identified as not problematic then that 4th patch can
>>>>>> potentially be reverted alone?
>>>>>
>>>>> Not quite everything :) But you are right, let's try to switch the 
>>>>> default to no delay. I will be posting patches today for that.
>>>>> With these patches, your patches are not necessary anymore as the AMD
>>>>> chipset falls under the default no-delay.
>>>>
>>>> I am all for improving the situation for all devices, but I am unable to
>>>> judge the regression potential of changing this, as it affects a lot of
>>>> devices. I guess it’d would go through the next tree, and hopefully the
>>>> company QA teams can give it a good spin. I hoped that my patches, as I
>>>> have tested them, and AMD will hopefully too, could go into the current
>>>> merge window.
>>>
>>> Yes, correct, the plan is to get the generic series queued as soon
>>> as rc1 so that it can spend plenty of time in linux-next for people
>>> to test. That will hopefully reduce the risk of breaking things in
>>> the field. Same for  the default LPM change.
>>
>> But 5.18 or 5.19? If 5.18, sounds good to me, if 5.19, I’d be great if 
>> my patches go into 5.18 cycle, as they have been tested, and it would 
>> mean the whole change gets tested more widely already.
>>
>>> With the default removal of the debounce delay, your patches addressing
>>> only the AMD adapter are not needed anymore: this adapter will not have a
>>> debounce delay unless the ATA_LFLAG_DEBOUNCE_DELAY flag is set.
>>
>> Yes, I understand.
> 
> The merge window for Linux 5.18 is going to close in three days this 
> Sunday. It’d be really great if my patches, tested on hardware, could go 
> into that.
> 
>>>>> It would be nice if you can test though.
>>>>
>>>> Of course, I am going to that either way.
>>>
>>> Series posted with you on CC. Please test !
>>
>> Thank you. I am going to test it in the coming days, and report back.
>>
>> Maybe more people should be put in Cc (Dell, Lenovo, IBM, x86 subsystem) 
>> with a request to test this?
> Thank you for the patches, which are a big improvement. Let’s hope, you 
> can re-roll them, so they get into Linux very soon for everyone’s benefit.

I am waiting for 5.18-rc1 to rebase the patches and re-post them. Given
reviewed-by and tested-by tags, I will queue them for 5.19. With that in
mind, I am not planning to apply your previous patches for 5.18, as they
would conflict and would only end up being churn since the delay removal
by default will undo your changes.


-- 
Damien Le Moal
Western Digital Research