cached PCS

cached PCS

[Jeff Garzik]

can you resurrect your cached PCS patch?

2.6.18-rc didn't fix the ghost device and long boot delay problems for 
everybody.

Re: cached PCS

[Tejun Heo]

Jeff Garzik wrote:
> can you resurrect your cached PCS patch?
> 
> 2.6.18-rc didn't fix the ghost device and long boot delay problems for 
> everybody.

The cached PCS patch was to fix device detection failure on some ICH5s 
where PCS is cleared while probing the first port.  Remaining ghost 
device and long boot delay are fixed by honor-PCS patch.  Do you want me 
to resurrect both?

-- 
tejun

Re: cached PCS

[Jeff Garzik]

Tejun Heo wrote:
> Jeff Garzik wrote:
>> can you resurrect your cached PCS patch?
>>
>> 2.6.18-rc didn't fix the ghost device and long boot delay problems for 
>> everybody.
> 
> The cached PCS patch was to fix device detection failure on some ICH5s 
> where PCS is cleared while probing the first port.  Remaining ghost 
> device and long boot delay are fixed by honor-PCS patch.  Do you want me 
> to resurrect both?

hrm.  Maybe just honor-PCS?

	Jeff

Re: cached PCS

[Tejun Heo]

Jeff Garzik wrote:
> Tejun Heo wrote:
>> Jeff Garzik wrote:
>>> can you resurrect your cached PCS patch?
>>>
>>> 2.6.18-rc didn't fix the ghost device and long boot delay problems 
>>> for everybody.
>>
>> The cached PCS patch was to fix device detection failure on some ICH5s 
>> where PCS is cleared while probing the first port.  Remaining ghost 
>> device and long boot delay are fixed by honor-PCS patch.  Do you want 
>> me to resurrect both?
> 
> hrm.  Maybe just honor-PCS?

Yeap, we need to verify whether the less-jealous-PCS-update patch cured 
the ICH5 problem.

-- 
tejun

Re: cached PCS

[Jeff Garzik]

Tejun Heo wrote:
> Jeff Garzik wrote:
>> Tejun Heo wrote:
>>> Jeff Garzik wrote:
>>>> can you resurrect your cached PCS patch?
>>>>
>>>> 2.6.18-rc didn't fix the ghost device and long boot delay problems 
>>>> for everybody.
>>>
>>> The cached PCS patch was to fix device detection failure on some 
>>> ICH5s where PCS is cleared while probing the first port.  Remaining 
>>> ghost device and long boot delay are fixed by honor-PCS patch.  Do 
>>> you want me to resurrect both?
>>
>> hrm.  Maybe just honor-PCS?
> 
> Yeap, we need to verify whether the less-jealous-PCS-update patch cured 
> the ICH5 problem.

If you are talking about the patch currently in 2.6.18-rc, I got several 
ACKs that it fixed their problems.

But OTOH, there was also
http://marc.theaimsgroup.com/?t=115346292700002&r=1&w=2

	Jeff

Re: cached PCS

[Tejun Heo]

Jeff Garzik wrote:
> Tejun Heo wrote:
>> Jeff Garzik wrote:
>>> Tejun Heo wrote:
>>>> Jeff Garzik wrote:
>>>>> can you resurrect your cached PCS patch?
>>>>>
>>>>> 2.6.18-rc didn't fix the ghost device and long boot delay problems 
>>>>> for everybody.
>>>>
>>>> The cached PCS patch was to fix device detection failure on some 
>>>> ICH5s where PCS is cleared while probing the first port.  Remaining 
>>>> ghost device and long boot delay are fixed by honor-PCS patch.  Do 
>>>> you want me to resurrect both?
>>>
>>> hrm.  Maybe just honor-PCS?
>>
>> Yeap, we need to verify whether the less-jealous-PCS-update patch 
>> cured the ICH5 problem.
> 
> If you are talking about the patch currently in 2.6.18-rc, I got several 
> ACKs that it fixed their problems.
> 
> But OTOH, there was also
> http://marc.theaimsgroup.com/?t=115346292700002&r=1&w=2

I was talking about bugzilla bug #6724.

http://bugzilla.kernel.org/show_bug.cgi?id=6724

It was a weird case of ICH5 clearing PCS present bits while the first 
port is being probed.  The clearing happens during actual probe, that 
is, the bits look okay till the end of the first prereset() but on entry 
to prereset() for the second port, enabled bits are gone.  IIRC, 
bug#6724 is different from all other reports.

I asked the reporter to test 2.6.18-rc4 and see whether the problem is 
gone with new PCS handling.  I'm a bit skeptical about the result.  If 
the problem remains with 2.6.18-rc4, we'll need that cached PCS to solve 
that particular case.

BTW, do you know what Keith Owen's chipset was?

Thanks.

-- 
tejun

Re: cached PCS

[Jeff Garzik]

Tejun Heo wrote:
> Jeff Garzik wrote:
>> Tejun Heo wrote:
>>> Jeff Garzik wrote:
>>>> Tejun Heo wrote:
>>>>> Jeff Garzik wrote:
>>>>>> can you resurrect your cached PCS patch?
>>>>>>
>>>>>> 2.6.18-rc didn't fix the ghost device and long boot delay problems 
>>>>>> for everybody.
>>>>>
>>>>> The cached PCS patch was to fix device detection failure on some 
>>>>> ICH5s where PCS is cleared while probing the first port.  Remaining 
>>>>> ghost device and long boot delay are fixed by honor-PCS patch.  Do 
>>>>> you want me to resurrect both?
>>>>
>>>> hrm.  Maybe just honor-PCS?
>>>
>>> Yeap, we need to verify whether the less-jealous-PCS-update patch 
>>> cured the ICH5 problem.
>>
>> If you are talking about the patch currently in 2.6.18-rc, I got 
>> several ACKs that it fixed their problems.
>>
>> But OTOH, there was also
>> http://marc.theaimsgroup.com/?t=115346292700002&r=1&w=2
> 
> I was talking about bugzilla bug #6724.
> 
> http://bugzilla.kernel.org/show_bug.cgi?id=6724
> 
> It was a weird case of ICH5 clearing PCS present bits while the first 
> port is being probed.  The clearing happens during actual probe, that 
> is, the bits look okay till the end of the first prereset() but on entry 
> to prereset() for the second port, enabled bits are gone.  IIRC, 
> bug#6724 is different from all other reports.
> 
> I asked the reporter to test 2.6.18-rc4 and see whether the problem is 
> gone with new PCS handling.  I'm a bit skeptical about the result.  If 
> the problem remains with 2.6.18-rc4, we'll need that cached PCS to solve 
> that particular case.
> 
> BTW, do you know what Keith Owen's chipset was?

(Keith CC'd)

I didn't see lspci output, but it looks like ICH5 from his drivers/ide 
dmesg output.

Did you look at some of the other messages?  He provided some debug 
traces in follow-up messages.

	Jeff

Re: cached PCS

[Tejun Heo]

Jeff Garzik wrote:
> Tejun Heo wrote:
>> Jeff Garzik wrote:
>>> Tejun Heo wrote:
>>>> Jeff Garzik wrote:
>>>>> Tejun Heo wrote:
>>>>>> Jeff Garzik wrote:
>>>>>>> can you resurrect your cached PCS patch?
>>>>>>>
>>>>>>> 2.6.18-rc didn't fix the ghost device and long boot delay 
>>>>>>> problems for everybody.
>>>>>>
>>>>>> The cached PCS patch was to fix device detection failure on some 
>>>>>> ICH5s where PCS is cleared while probing the first port.  
>>>>>> Remaining ghost device and long boot delay are fixed by honor-PCS 
>>>>>> patch.  Do you want me to resurrect both?
>>>>>
>>>>> hrm.  Maybe just honor-PCS?
>>>>
>>>> Yeap, we need to verify whether the less-jealous-PCS-update patch 
>>>> cured the ICH5 problem.
>>>
>>> If you are talking about the patch currently in 2.6.18-rc, I got 
>>> several ACKs that it fixed their problems.
>>>
>>> But OTOH, there was also
>>> http://marc.theaimsgroup.com/?t=115346292700002&r=1&w=2
>>
>> I was talking about bugzilla bug #6724.
>>
>> http://bugzilla.kernel.org/show_bug.cgi?id=6724
>>
>> It was a weird case of ICH5 clearing PCS present bits while the first 
>> port is being probed.  The clearing happens during actual probe, that 
>> is, the bits look okay till the end of the first prereset() but on 
>> entry to prereset() for the second port, enabled bits are gone.  IIRC, 
>> bug#6724 is different from all other reports.
>>
>> I asked the reporter to test 2.6.18-rc4 and see whether the problem is 
>> gone with new PCS handling.  I'm a bit skeptical about the result.  If 
>> the problem remains with 2.6.18-rc4, we'll need that cached PCS to 
>> solve that particular case.
>>
>> BTW, do you know what Keith Owen's chipset was?
> 
> (Keith CC'd)
> 
> I didn't see lspci output, but it looks like ICH5 from his drivers/ide 
> dmesg output.
> 
> Did you look at some of the other messages?  He provided some debug 
> traces in follow-up messages.

Yes, I have and I can't think of any other way than ignoring PCS to work 
around the problem.  If we set IGNORE_PCS for ich5 sata then we can also 
forget about cached PCS which is for ich5 sata.  But, we might see ghost 
device detection and accompanying long delays.

For ich 6/7/8, your recent change and honor-pcs patch should do the job. 
  For ICH5, well...  Just set IGNORE_PCS and wait for bug reports?

-- 
tejun

Re: cached PCS

[Tejun Heo]

Tejun Heo wrote:
>> Did you look at some of the other messages?  He provided some debug 
>> traces in follow-up messages.
> 
> Yes, I have and I can't think of any other way than ignoring PCS to work 
> around the problem.  If we set IGNORE_PCS for ich5 sata then we can also 
> forget about cached PCS which is for ich5 sata.  But, we might see ghost 
> device detection and accompanying long delays.
> 
> For ich 6/7/8, your recent change and honor-pcs patch should do the job. 
>  For ICH5, well...  Just set IGNORE_PCS and wait for bug reports?

[CC'ing Steve (the bug reporter of #6724), Hi!]

Steve just reported -rc4 doesn't fix the problem && he saw complete 
misdetection of *all* devices on soft reboots.  I think Steve and Keith 
are experiencing the same problem here.

PCS enabled bits are cleared at some point and cached PCS works around 
the problem by caching the bits early but for soft reboots it doesn't do 
any good.  One interesting thing is that BIOSen seem to recognize the 
drives fine (right?) on such reboots even though PCS enabled bits would 
be zero.  FWIW, IGNORE_PCS works fine for Steve.

Thanks.

-- 
tejun