[U-Boot] [PATCH v2 2/3] efi_loader: enumerate disk devices every time

[AKASHI Takahiro]

On Fri, Jan 25, 2019 at 10:31:20AM +0100, Alexander Graf wrote:
> 
> 
> On 25.01.19 10:18, AKASHI Takahiro wrote:
> > On Fri, Jan 25, 2019 at 09:52:31AM +0100, Alexander Graf wrote:
> >>
> >>
> >> On 25.01.19 09:27, AKASHI Takahiro wrote:
> >>> Alex,
> >>>
> >>> On Wed, Jan 23, 2019 at 10:51:29AM +0100, Alexander Graf wrote:
> >>>> On 01/22/2019 08:39 PM, Simon Glass wrote:
> >>>>> Hi Alex,
> >>>>>
> >>>>> On Tue, 22 Jan 2019 at 22:08, Alexander Graf <agraf at suse.de> wrote:
> >>>>>>
> >>>>>>
> >>>>>> On 22.01.19 09:29, AKASHI Takahiro wrote:
> >>>>>>> Alex, Simon,
> >>>>>>>
> >>>>>>> Apologies for my slow response on this matter,
> >>>>>>>
> >>>>>>> On Fri, Jan 11, 2019 at 08:57:05AM +0100, Alexander Graf wrote:
> >>>>>>>>
> >>>>>>>> On 11.01.19 05:29, AKASHI Takahiro wrote:
> >>>>>>>>> Alex, Heinrich and Simon,
> >>>>>>>>>
> >>>>>>>>> Thank you for your comments, they are all valuable but also make me
> >>>>>>>>> confused as different people have different requirements :)
> >>>>>>>>> I'm not sure that all of us share the same *ultimate* goal here.
> >>>>>>>> The shared ultimate goal is to "merge" (as Simon put it) dm and efi objects.
> >>>>>>> I don't still understand what "merge" means very well.
> >>>>>> It basically means that "struct efi_object" moves into "struct udevice".
> >>>>>> Every udevice instance of type UCLASS_BLK would expose the block and
> >>>>>> device_path protocols.
> >>>>>>
> >>>>>> This will be a slightly bigger rework, but eventually allows us to
> >>>>>> basically get rid of efi_init_obj_list() I think.
> >>>>> I envisaged something like:
> >>>>>
> >>>>> - EFI objects have their own UCLASS_EFI uclass
> >>>>
> >>>> ... and then we need to create our own sub object model around the
> >>>> UCLASS_EFI devices again. I' not convinced that's a great idea yet :). I
> >>>> really see little reason not to just expose every dm device as EFI handle.
> >>>> Things would plug in quite naturally I think.
> >>>
> >>> You said that the ultimate goal is to remove all efi_object data.
> >>> Do you think that all the existing efi_object can be mapped to
> >>> one of existing u-boot uclass devices?
> >>>
> >>> If so, what would be an real entity of a UEFI handle?
> >>> struct udevice *?
> >>>
> >>> But Simon seems not to agree to adding any UEFI-specific members
> >>> in struct udevice.
> >>
> >> I think we'll have to experiment with both approaches. I personally
> >> would like to have struct udevice * be the UEFI handle, yes.
> >>
> >>>
> >>>> But either way, someone would need to sit down and prototype things to be
> >>>> sure.
> >>>>
> >>>
> >>> The most simplest prototype would include
> >>> * event mechanism (just registration and execution of hook/handler)
> >>>     event: udevice creation (and deletion)
> >>> * efi_disk hook for udevice(UCLASS_BLK) creation
> >>> * modified block device's enumeration code, say, scsi_scan(),
> >>>   to add an event hook at udevice creation
> >>> * removing efi_disk_register() from efi_init_obj_list()
> >>> * Optionally(?) UCLASS_PARTITION
> >>>   (Partition udevices would be created in part_init().)
> >>
> >> Almost.
> >>
> >> The simplest prototype would be to add a struct efi_object into struct
> >> udevice. Then whenever we're looping over efi_obj_list in the code, we
> >> additionally loop over all udevices to find the handle.
> > 
> > Ah, yes. You're going further :)
> > 
> >> Then, we could slowly give the uclasses explicit knowledge of uefi
> >> protocols. So most of the logic of efi_disk_register() would move into
> >> (or get called by) drivers/block/blk-uclass.c:blk_create_device().
> > 
> > Via event? Otherwise, we cannot decouple u-boot and UEFI world.
> 
> For a prototype, just make it explicit and see how far that gets us.
> 
> >> Instead of creating diskobj and adding calling efi_add_handle(), we
> >> could then just use existing data structure from the udevice (and its
> >> platdata).
> > 
> > I don't have good confidence that we can remove struct efi_disk_obj,
> > at least, for the time being as some of its members are quite UEFI-specific.
> 
> Maybe we can move them into struct blk_desc? It's a matter of
> experimenting I guess.
> 
> > 
> >>
> >> Does this make sense? Less events, more implicity :).
> > 
> > I'll go for it.
> 
> Thanks a lot :). Feel free to pick an easier target for starters too if
> you prefer.

Prototyping is done :)
Since it was so easy and simple, now I'm thinking of implementing
UCLASS_PARTITION. But it is not so straightforward as I expected,
and it won't bring us lots of advantages.
(I think that blk_desc should also support a partition in its own.)

Once it gets working, may I send out a patch?

-Takahiro Akashi


> 
> Alex