Kernel patch for AC notifier chain

Tippett, Matthew Matthew.Tippett at amd.com
Tue Sep 1 18:17:09 UTC 2009 

There are two reasons that the driver needs this.

1) To simplify driver event aggregation, we have the kernel portion of the driver gather various events and push it to the userland for touching the power-management driver code.  We expect long term to put this driver code in the kernel.  This driver code uses a set of different clock/voltage setting to control the power draw for the device.  The AC/DC values are typically set by the OEM into the BIOS based on their design.

2) There are some platform designs that the current draw of the GPU at high clocks may cause an overcurrent condition for the battery.  The window that has been communicated is measured in 10s and 100s of milliseconds to put the GPU into a lower powerstate.  Clearly posting a netlink event and then relying on a possibly non-running or misconfigured power daemon which then communicates with the driver through other mechanisms is inferior to the kernel side of the driver receiving the acpi event and immediately posting it for the userland side of the driver to take action.

Regards... Matthew


--
Blackberry-Sent
Matthew Tippett
Engineering Manager, Linux Core Engineering
Graphics Products Group, Advanced Micro Devices, Inc.
Ph: +1 905 882 2600 x8014 Cell: +1 416 671 0673  

----- Original Message -----
From: Stefan Bader <stefan.bader at canonical.com>
To: Kuehling, Felix
Cc: tim.gardner at canonical.com <tim.gardner at canonical.com>; kernel-team at lists.ubuntu.com <kernel-team at lists.ubuntu.com>; Tippett, Matthew; Marsan, Luugi; Bryce Harrington <bryce at canonical.com>
Sent: Tue Sep 01 13:46:24 2009
Subject: Re: Kernel patch for AC notifier chain

Felix Kuehling wrote:
> Really adding kernel-team at lists.ubuntu.com, moving Larry to BCC
> 
> Please consider the attached patch for inclusion in Ubuntu 9.10.
> 
> Without this patch the early-look fglrx driver that will be included in
> Ubuntu 9.10 will be unable to respond to AC/DC events. This may have
> detrimental effects on the power consumption on mobile devices.
> Depending on the power consumption and battery power, this may even
> result in hardware damage.
> 
> We need to submit (or not submit) our corresponding driver change for
> the driver release to Canonical by tonight. By then I need to know
> whether Ubuntu 9.10 is going to ship with the patch or not.
> 

As I have not been following the discussion this is more of a dump of thoughts...
I would be a little unsure about possible side effects as the patch causes the 
event being visible through multiple sources.
Not sure what the (ideological or other) reasons are for being reluctant to 
take that change, it seems the driver generates a netlink event with exactly 
the same information. So it is not clear why fglrx requires this...

> Regards,
>   Felix
> 
> On Tue, 2009-09-01 at 06:08 -0600, Tim Gardner wrote:
>> Felix,
>>
>> Please send your patch for review to kernel-team at lists.ubuntu.com and 
>> I'll get some eyeballs on it.
>>
>> rtg
>>
>> Pete Graner wrote:
>>> Felix,
>>>
>>> Sorry I have missed this. I have been traveling in Asia & Andy has been 
>>> on holiday (and still is for the remainder of the week).
>>>
>>> I'm bouncing this to my team lead Tim Gardner (cc'd). Tim should be able 
>>> to advise you.
>>>
>>> I will say that if upstream won't accept then it will be a hard sell for 
>>> Tim. Carrying out of band patches gets quickly expensive.
>>>
>>> I have not been following the discussion so I'm not sure what the reason 
>>> is but will you end up getting in included and if so whats the time 
>>> frame, that will sway Tim's thinking quite a bit.
>>>
>>> Thanks
>>>
>>> ~pete
>>>
>>> Felix Kuehling wrote:
>>>> Any update on this?
>>>>
>>>> Our deadline for putting in our side of the change is tomorrow night
>>>> eastern time.
>>>>
>>>> Regards,
>>>>    Felix
>>>>
>>>> On Wed, 2009-08-26 at 17:57 -0400, Felix Kuehling wrote:
>>>>> Hi Andy and Pete,
>>>>>
>>>>> We're having trouble getting the patch accepted upstream, not because
>>>>> it's technically wrong, but rather for ideological reasons. Would
>>>>> Canonical be willing to include the patch in Ubuntu 9.10 regardless of
>>>>> whether (or when) it's accepted upstream?
>>>>>
>>>>> Without this patch the early-look driver that will be included in Ubuntu
>>>>> 9.10 will be unable to respond to AC/DC events. This may have
>>>>> detrimental effects on the power consumption on mobile devices.
>>>>> Depending on the power consumption and battery power, this may even
>>>>> result in hardware damage.
>>>>>
>>>>> We need to submit (or not submit) our corresponding driver change for
>>>>> the driver release to Canonical by Tuesday next week. By then I need to
>>>>> know whether Ubuntu 9.10 is going to ship with the patch or not.
>>>>>
>>>>> Best regards,
>>>>>   Felix
>>>>>
>>>>> P.S.: I attached the patch again, for reference.
>>>>>
>>>>> On Tue, 2009-07-14 at 13:30 -0400, Pete Graner wrote:
>>>>>> Felix Kuehling wrote:
>>>>>>> Adding Matthew.
>>>>>>>
>>>>>>> We have a patch that we intend to propose for adoption upstream
>>>>>>> (attached). This would enable us to continue receiving 
>>>>>>> notifications in
>>>>>>> our kernel driver, whenever AC power is plugged or unplugged.
>>>>>>>
>>>>>>> Pete, Andrew, what's your feeling about this patch? Can you see any
>>>>>>> issues including this in Ubuntu 9.10? Do you foresee any trouble 
>>>>>>> getting
>>>>>>> this accepted upstream?
>>>>>> Felix,
>>>>>>
>>>>>> I'll let Andy comment on this.
>>>>>>
>>>>>>
>>>>>> Andy can you take a look pls.
>>>>>>
>>>>>> Thanks
>>>>>>
>>>>>> ~pete
>>>>>>
>>>>>>
>>>>>>
>>>>>>> On Thu, 2009-07-02 at 16:24 -0400, Pete Graner wrote:
>>>>>>>> Hi Bryce,
>>>>>>>>
>>>>>>>> Andy is on Holiday until next week, so I'll try and answer what I 
>>>>>>>> can...
>>>>>>>>
>>>>>>>> See responses inline...
>>>>>>>>
>>>>>>>> On Thu, Jul 02, 2009 at 12:40:12PM -0700, Bryce Harrington wrote:
>>>>>>>>> Hi Pete and Andy,
>>>>>>>>>
>>>>>>>>> There were a few kernel questions that came up at today's 
>>>>>>>>> AMD/Canonical
>>>>>>>>> call from Felix, who is one of AMD's fglrx development engineers 
>>>>>>>>> looking
>>>>>>>>> at getting the driver working well on Ubuntu.
>>>>>>>>>
>>>>>>>>> 1.  In testing recent Karmic, Felix noticed that we have the 
>>>>>>>>> intel drm
>>>>>>>>>     driver always being loaded now for KMS, and is wondering if 
>>>>>>>>> this is
>>>>>>>>>     going to be done in all cases going forward, or if logic will be
>>>>>>>>>     included in the kernel to only load it if on Intel graphics?
>>>>>>>>>
>>>>>>>>>     (And a special case to consider - if user has on-board Intel
>>>>>>>>>     graphics but an ATI add-in card present.)
>>>>>>>> Not sure on this, Andy will be able to follow up.
>>>>>>>>
>>>>>>>>> 2.  Related question is with proprietary drivers generally, if 
>>>>>>>>> KMS is
>>>>>>>>>     not supported by that driver, what mechanism will be used to 
>>>>>>>>> prevent
>>>>>>>>>     the FOSS drm driver being loaded.  There are brute force methods
>>>>>>>>>     such as passing a kernel param via grub's menu.lst or even just
>>>>>>>>>     removing the radeon module and rebuilding initrd on fglrx 
>>>>>>>>> install,
>>>>>>>>>     but we're wondering if we could do a more elegant method, 
>>>>>>>>> such as an
>>>>>>>>>     initrd parameter, or other toggle that could be flipped 
>>>>>>>>> during fglrx
>>>>>>>>>     install to let the kernel know to turn off KMS.
>>>>>>>>>
>>>>>>>> I think the later is the approach is what was decided, Andy can
>>>>>>>> confirm.
>>>>>>>>
>>>>>>>>
>>>>>>>>> 3.  An ACPI question - they've noticed some data structures have 
>>>>>>>>> been
>>>>>>>>>     moved from public headers to private locations.  This causes 
>>>>>>>>> a bit
>>>>>>>>>     of a problem for -fglrx to support certain hotkeys.  AMD has 
>>>>>>>>> a patch
>>>>>>>>>     they're going to be sending upstream to handle this, but it 
>>>>>>>>> probably
>>>>>>>>>     won't make it in time for .31.  Is there someone on the 
>>>>>>>>> kernel team
>>>>>>>>>     they could work with to review and/or include the patch for 
>>>>>>>>> karmic,
>>>>>>>>>     and to otherwise discuss the plan of attack with?
>>>>>>>> Andy can do the review and yes we can carry the patch.
>>>>>>>>
>>>>>>>>> Finally, since a number of kernel questions have been coming up, 
>>>>>>>>> we'd
>>>>>>>>> like to invite one Canonical kernel person to the next AMD call 
>>>>>>>>> in two
>>>>>>>>> weeks.  Can you recommend whom should be invited to this call?
>>>>>>>>>
>>>>>>>> Andy is your man :-) Send over the invite info and he can attend. Add
>>>>>>>> me on it as well so I have the info in case we ever need to send
>>>>>>>> someone else.
>>>
>>


-- 

When all other means of communication fail, try words!



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/kernel-team/attachments/20090901/dfd6b53b/attachment.html>

More information about the kernel-team mailing list