Re: [PATCH v2 0/3] Enable user helper interface for efi capsule update

From: Andy Lutomirski
Date: Mon Nov 03 2014 - 18:08:36 EST


On Mon, Nov 3, 2014 at 3:02 PM, Greg Kroah-Hartman
<gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> On Mon, Nov 03, 2014 at 01:32:46PM -0800, Andy Lutomirski wrote:
>> On Mon, Nov 3, 2014 at 1:27 PM, Greg Kroah-Hartman
>> <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
>> > On Mon, Nov 03, 2014 at 11:33:23AM -0800, Andy Lutomirski wrote:
>> >> On 11/02/2014 07:07 PM, Kweh Hock Leong wrote:
>> >> > From: "Kweh, Hock Leong" <hock.leong.kweh-ral2JQCrhuEAvxtiuMwx3w@xxxxxxxxxxxxxxxx>
>> >> >
>> >> > Hi Guys,
>> >> >
>> >> > This patchset is created on top of "efi: Capsule update support" patch:
>> >> > http://permalink.gmane.org/gmane.linux.kernel.efi/4837
>> >> >
>> >> > It leverages the request_firmware_nowait() to expose the user helper interface for user to upload the capsule binary and calling the
>> >> > efi_capsule_update() API to pass the binary to EFI firmware.
>> >>
>> >> I don't get it. Why is the firmware interface at all reasonable for
>> >> uploading capsules?
>> >
>> > Tradition dictates that BIOS updates go through the firmware interface,
>> > that way you don't have to write a new userspace tool, which is a good
>> > thing.
>> >
>> >> The firmware interface makes sense for nonvolatile firmware where
>> >> hotplugging something or otherwise loading a driver needs a blob.
>> >
>> > Or BIOS data. We've been doing it this way for a long time now.
>>
>> On what system? Dell?
>
> Yes.
>
>> IMO this sucks from a UI point of view. When I install wifi firmware,
>> I expect to stick it somewhere and have the driver find it, because
>> the driver knows exactly when it needs the firmware. When I update my
>> BIOS, I want to click a button or type a command and update my bios.
>
> I agree, it should be "triggered" by something, not just automagically
> loaded whenever the kernel randomly looks for it.
>
>> >> But uploading an EFI capsule is an *action*, not something that should
>> >> happen transparently. If there's an EFI firmware update available and
>> >> the user wants to install it, then the userspace tool should install it,
>> >> and it shouldn't hang around in /lib/firmware. In fact, you shouldn't
>> >> even need /lib to be on writable media to use this.
>> >
>> > What does /lib have to do with this?
>>
>> Where else does the file come from, given that udev no longer supports
>> userspace firmware loading? Is there really some pre-existing tool
>> that pokes it into the sysfs firmware class thing?
>
> Well, you can specify other locations than /lib/firmware/ for firmware
> updates, but yes, you are right, it should be in /lib somewhere. But
> /lib doesn't need to be writable, it's a read-only file.
>

I assume that whoever downloaded the firmware update will want to
install it, right? I don't really expect distros to ship EFI capsules
in packages that install to /lib/firmware. Won't there be userspace
code that either installs a capsule from some URL or uses some future
magical find-my-firmware service?

>> Since EFI capsules are apparently on their way to becoming a
>> ubiquitous mechanism, I think it might be time to rethink
>> request_firmware for this.
>
> What do you suggest instead? A "custom" sysfs file? What is going to
> trigger it to be loaded? A userspace script that someone else has to
> write? :)

Some ioctl on /dev/efi_capsule seems reasonable to me, or a new script
that uses a custom sysfs file. Isn't the Dell thing already a rather
custom script? You write to a custom sysfs file ("rbu_image_type", I
think) and then the handler for that file calls request_firmware.

I think that we can handle a very small C program or script that
uploads the EFI capsule.

Also, FWIW, I think that there are EFI capsules that aren't firmware
updates. For example, IIRC there's some mechansim that allows you to
pass data to the next OS that boots via a capsule. It's probably
buggy on every motherboard in existence, but if it ever worked, using
it through /lib/firmware would make no sense.

--Andy
--
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/