[Intel-wired-lan] [PATCH] e1000e: bump up timeout to wait when ME un-configure ULP mode

[Paul Menzel]

Dear Kai-Heng,


Am 26.03.20 um 12:29 schrieb Kai-Heng Feng:

>> On Mar 25, 2020, at 23:49, Paul Menzel <pmenzel at molgen.mpg.de> wrote:

>> Am 25.03.20 um 14:58 schrieb Neftin, Sasha:
>>> On 3/25/2020 08:43, Aaron Ma wrote:
>>
>>>> On 3/25/20 2:36 PM, Neftin, Sasha wrote:
>>>>> On 3/25/2020 06:17, Kai-Heng Feng wrote:
>>
>>>>>>> On Mar 24, 2020, at 03:16, Aaron Ma <aaron.ma at canonical.com> wrote:
>>>>>>>
>>>>>>> ME takes 2+ seconds to un-configure ULP mode done after resume
>>>>>>> from s2idle on some ThinkPad laptops.
>>>>>>> Without enough wait, reset and re-init will fail with error.
>>>>>>
>>>>>> Thanks, this patch solves the issue. We can drop the DMI quirk in
>>>>>> favor of this patch.
>>>>>>
>>>>>>> Fixes: f15bb6dde738cc8fa0 ("e1000e: Add support for S0ix")
>>>>>>> BugLink: https://bugs.launchpad.net/bugs/1865570
>>>>>>> Signed-off-by: Aaron Ma <aaron.ma at canonical.com>
>>>>>>> ---
>>>>>>> drivers/net/ethernet/intel/e1000e/ich8lan.c | 4 ++--
>>>>>>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>>>>>>
>>>>>>> diff --git a/drivers/net/ethernet/intel/e1000e/ich8lan.c
>>>>>>> b/drivers/net/ethernet/intel/e1000e/ich8lan.c
>>>>>>> index b4135c50e905..147b15a2f8b3 100644
>>>>>>> --- a/drivers/net/ethernet/intel/e1000e/ich8lan.c
>>>>>>> +++ b/drivers/net/ethernet/intel/e1000e/ich8lan.c
>>>>>>> @@ -1240,9 +1240,9 @@ static s32 e1000_disable_ulp_lpt_lp(struct
>>>>>>> e1000_hw *hw, bool force)
>>>>>>>               ew32(H2ME, mac_reg);
>>>>>>>           }
>>>>>>>
>>>>>>> -        /* Poll up to 300msec for ME to clear ULP_CFG_DONE. */
>>>>>>> +        /* Poll up to 2.5sec for ME to clear ULP_CFG_DONE. */
>>>>>>>           while (er32(FWSM) & E1000_FWSM_ULP_CFG_DONE) {
>>>>>>> -            if (i++ == 30) {
>>>>>>> +            if (i++ == 250) {
>>>>>>>                   ret_val = -E1000_ERR_PHY;
>>>>>>>                   goto out;
>>>>>>>               }
>>>>>>
>>>>>> The return value was not caught by the caller, so the error ends up
>>>>>> unnoticed.
>>>>>> Maybe let the caller check the return value of
>>>>>> e1000_disable_ulp_lpt_lp()?
>>
>>>>> I a bit confused. In our previous conversation you told ME not running.
>>>>> let me shimming in. Increasing delay won't be solve the problem and just
>>>>> mask it. We need to understand why ME take too much time. What is
>>>>> problem with this specific system?
>>>>> So, basically no ME system should works for you.
>>>>
>>>> Some laptops ME work that's why only reproduce issue on some laptops.
>>>> In this issue i219 is controlled by ME.
>>>
>>> Who can explain - why ME required too much time on this system?
>>> Probably need work with ME/BIOS vendor and understand it.
>>> Delay will just mask the real problem - we need to find root cause.
>>>> Quirk is only for 1 model type. But issue is reproduced by more models.
>>>> So it won't work.
>>
>> (Where is Aaron’s reply? It wasn’t delivered to me yet.)
>>
>> As this is clearly a regression, please revert the commit for now,
>> and then find a way to correctly implement S0ix support. Linux’
>> regression policy demands that as no fix has been found since
>> v5.5-rc1. Changing Intel ME settings, even if it would work around
>> the issue, is not an acceptable solution. Delaying the resume time
>> is also not a solution.
> 
> The s0ix patch can reduce power consumption, finally makes S2idle an
> acceptable sleep method. So I'd say it's a fix, albeit a regression
> was introduced.
> 
>> Regarding Intel Management Engine, only Intel knows what it does
>> and what the error is, as the ME firmware is proprietary and
>> closed.
>> 
>> Lastly, there is no way to fully disable the Intel Management
>> Engine. The HAP stuff claims to stop the Intel ME execution, but
>> nobody knows, if it’s successful.
>> 
>> Aaron, Kai-Hang, can you send the revert?
> 
> I consider that as an important fix for s2idle, I don't think
> reverting is appropriate.

If there is a fix with no other regression, I agree. But there has not 
been one, so please revert. It doesn’t matter if the commit introducing 
the regression fixes something else. It gets too complicated to decide, 
which regression is worth it or not. The Linux kernel guarantees its 
users, they can update any time without breaking user space (in this 
case suspend/resume).  Please read Linus’ several messages about that. 
His message [1] exactly addresses your arguments.

> Yeah, HELL NO!
> 
> Guess what? You're wrong. YOU ARE MISSING THE #1 KERNEL RULE.
> 
> We do not regress, and we do not regress exactly because your are 100% wrong.
> 
> And the reason you state for your opinion is in fact exactly *WHY* you
> are wrong.
> 
> Your "good reasons" are pure and utter garbage.
> 
> The whole point of "we do not regress" is so that people can upgrade
> the kernel and never have to worry about it.
> 
>> Kernel had a bug which has been fixed
> 
> That is *ENTIRELY* immaterial.
> 
> Guys, whether something was buggy or not DOES NOT MATTER.

So, please (also Intel developers), please adhere to this rule, and 
revert the commit.


Kind regards,

Paul


[1]: https://lkml.org/lkml/2018/8/3/621