Re: [PATCH v4 2/2] efi: an sysfs interface for user to update efi firmware

[Andy Lutomirski]

On Wed, Apr 29, 2015 at 2:39 PM, James Bottomley
<James.Bottomley@xxxxxxxxxxxxxxxxxxxxx> wrote:
> On Wed, 2015-04-29 at 14:36 -0700, Andy Lutomirski wrote:
>> On Wed, Apr 29, 2015 at 2:35 PM, James Bottomley
>> <James.Bottomley@xxxxxxxxxxxxxxxxxxxxx> wrote:
>> > On Wed, 2015-04-29 at 11:23 +0000, Kweh, Hock Leong wrote:
>> >> I agree with James. Due to different people may have different needs. But
>> >> from our side, we would just like to have a simple interface for us to upload
>> >> the efi capsule and perform update. We do not have any use case or need
>> >> to get info from QueryCapsuleUpdate(). Let me give a suggestion here:
>> >> please allow me to focus on deliver this simple loading interface and
>> >> upstream it. Then later whoever has the actual use case or needs on the ioctl
>> >> implementation, he or she could enhance base on this simple loading interface.
>> >> What do you guys think?
>> >>
>> >> Let me summarize the latest design idea:
>> >> - No longer leverage on firmware class but use misc device
>> >> - Do not use platform device but use device_create()
>> >> - User just need to perform "cat file.bin > /sys/.../capsule_loader" in the shell
>> >> - File operation functions include: open(), read(), write() and flush()
>> >> - Perform mutex lock in open() then release the mutex in flush() for avoiding
>> >>    race condition / concurrent loading
>> >> - Perform the capsule update and error return at flush() function
>> >>
>> >> Is there anything I missed? Any one still have concern with this idea?
>> >> Thanks for providing the ideas as well as the review.
>> >
>> > I think that's pretty much it.
>> >
>> > Why don't you let me construct a straw man patch.  It's going to be a
>> > bit controversial because it involves adding flush operations to sysfs
>> > and kernfs, slicing apart firmware_class.c to extract the transaction
>> > handling stuff and creating an new efi update capsule file which makes
>> > use of it.
>> >
>> > Once we have code, we at least have something more concrete to argue
>> > over.
>>
>> Would it be worth checking whether busybox is also okay with it first?
>> (Sorry to be a naysayer.)
>>
>> It would be a shame if we do all this to keep the userspace footprint
>> light and then it doesn't work for non-coreutils userspace.
>
> I don't think so, because we can fix busybox if it's a problem.  The
> embedded people wanting this control the tool space, so they can decide
> to use the fixed version.
>
> So yes, someone should check and fix busybox cat if broken, but no, it's
> not a blocker.

It's still a bit unfortunate that:

#!/bin/sh

cat "$1" >/sys/whatever
if [ "$?" != "0" ]; then
   echo "It didn't work because" ...
   exit 1
fi

echo "It worked!  Go reboot if needed."
exit 0

will only work sometimes.  Will people really test this on their
target implementation of cat?  I agree that making this possible with
just shell is nice, but I'm less happy about it if it'll be
unreliable.

--Andy


-- 
Andy Lutomirski
AMA Capital Management, LLC
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/