Re: [PATCH] firmware/efi: export a routine to retrieve efi-variables by GUID

[Dan Williams <dan.j.williams@intel.com>]

On 3/19/2011 6:13 PM, James Bottomley wrote:
> On Sat, 2011-03-19 at 17:14 -0700, Greg KH wrote:
>> On Fri, Mar 18, 2011 at 06:15:47PM -0700, Dan Williams wrote:
>>> On 3/18/2011 5:22 PM, Greg KH wrote:
>>>> On Fri, Mar 18, 2011 at 04:10:10PM -0700, Dan Williams wrote:
>>>>>> I needed all patches in linux-next _before_ the merge window opened to
>>>>>> be able to accept it.
>>>>>
>>>>> Yes, I know, and as dmaengine maintainer I also hate being ambushed by
>>>>> last minute patches, but now I am unfortunately one of those annoying
>>>>> people on the other side of the coin.
>>>>
>>>> Then you should know better than to try to go around the well-known
>>>> rules :)
>>>
>>> Yes...
>>>
>>> /me about to push his luck
>>
>> <snip>
>>
>>> As Jeff pointed out:
>>> "It seemed like this was turning into another driver that would get
>>> held outside the kernel until it's "perfect."  If that is the case,
>>> Linus has also made it clear we should get drivers for high volume,
>>> shipping hardware into the kernel, even if its staging, if the
>>> alternative is to deny users the driver."
>>
>> That's fine, _BUT_ you are trying to go around the rules for the merge
>> window, which isn't acceptable.  Also note that your driver isn't
>> self-contained, it needs this change at the least, right?  Any others?
>>
>>> So yes, we are targeting that exception.  I'm up for taking the heat
>>> directly if you want...  because the pull request will need to
>>> backed up with justification.
>>
>> No, sorry, I'll not take this for .40, all of my trees are merged with
>> Linus now for .40 and I'll only be sending him bugfixes until the .41
>> merge window opens up.
>>
>> Remember, it's only a 3 month wait, you knew about this _WAY_ in
>> advance, so it's not like this is something new, or out of the ordinary
>> at all.  Because of that, I fail to see why this is somehow not
>> expected.
>>
>> On a personal note, I'm going to be very scarse for the next 3 weeks due
>> to conferences and travel for spring break, so I physically don't have
>> the time to do any more merges like this with Linus.
>
> So, here's the deal:  You get this driver to a mergeable state, which
> means all of the problems Christoph, others and I have outlined
> completely fixed or well on the way to being in two months and I will
> take it under the merge window exception for new drivers.  However, I
> mean seriously cleaned up and shiny (and dumping tens of thousands of
> lines of flue code and other oddities), not cleaned up for staging.

Sounds fair.

> Further you're going to have to use standard methods to store and
> retrieve data, not esoteric efi ones that rely on changes outside SCSI
> (you aren't the first people to need to set a SAS address and other
> things, so we have a request firmware like method for it).

We do have a request_firmware() interface and a binary blob generation 
utility in drivers/scsi/isci/firmware/, but that is only for the 
fallback (unable to retrieve from platform-firmware) case and should 
probably be considered a bios bug.  The expectation is that globally 
unique sas address, phy signal-amplification, and default port 
configuration settings are a per-board property set by the OEM in the 
factory.

Ideally this would have been something that just showed up in an 
option-rom bar on the device, but that did not happen so it is left to 
software.  For legacy-bios we scan adapter rom space looking for our 
table, for efi it's comparatively cleaner we just grab this efi variable 
identified by its own GUID.

We could have userspace rebuild the firmware image based on what it 
finds in the platform firmware... but that becomes a messy management 
step and gets confusing if one ever wants to override the platform 
defaults with a custom blob.

Other options?

> If you can't make it in two months .. can you do it? I'll extend the
> offer to the .40 which makes the cutoff around 5 months, if you need
> extra time.

Also fair.  I suspect we can get through the backlog in that time.

> What I suggest is that you take about a month and a half to make all the
> changes and then do a patch code drop on linux-scsi.  If you're less
> sure about what you need to do, do one after three weeks and we'll redo
> the assessment.

Ok.  I'll continue to post "[GIT] isci" updates to announce the 
intermediate state, but will drop full patch sets to the mailing list 
when we have upstream merge candidates ready for review.

--
Dan