[PATCH 00/13] efi: Tidy up confusion between pointers and addresses

[Simon Glass]

Hi Mark,

On Wed, 27 Nov 2024 at 06:52, Mark Kettenis <mark.kettenis at xs4all.nl> wrote:
>
> > From: Simon Glass <sjg at chromium.org>
> > Date: Wed, 27 Nov 2024 06:08:10 -0700
>
> Hi Simon,
>
> > Hi Tom,
> >
> > On Tue, 26 Nov 2024 at 09:08, Tom Rini <trini at konsulko.com> wrote:
> > >
> > > On Tue, Nov 26, 2024 at 08:39:05AM -0700, Simon Glass wrote:
> > > > Hi Heinrich,
> > > >
> > > > On Tue, 26 Nov 2024 at 02:10, Heinrich Schuchardt <xypron.glpk at gmx.de> wrote:
> > > > >
> > > > > On 25.11.24 21:44, Simon Glass wrote:
> > > > > > The EFI-loader implementation converts things back and forth between
> > > > > > addresses and pointers, with not much consistency in how this is done.
> > > > > >
> > > > > > Within most of U-Boot a pointer is a void * and an address is a ulong
> > > > >
> > > > > No. It is phys_addr_t that was introduced to handle the sandbox's
> > > > > virtual addresses.
> > > > >
> > > > > And we should keep this sandbox stuff out of the EFI code. They are only
> > > > > needed for the command line interface.
> > > >
> > > > No. My statement is correct and long predates sandbox. U-Boot has
> > > > always used ulong. Using an address is generally much easier than a
> > > > pointer, so I believe this design decision was (and remains) a good
> > > > one. I certainly would not support changing it. The phys_addr_t type
> > > > has nothing to do with sandbox.
> > >
> > > At the high level, I think parts of the problem are that when U-Boot
> > > went from 32bit architectures to 36bit architectures (assorted PowerPC
> > > Freescale parts) to 64bit architectures that no, we didn't make some
> > > concious and consistent statement that "ulong is for address (except
> > > when it's not). If you look at the linux kernel, phys_addr_t is physical
> > > addresses, ulong is virtual addresses.
> > >
> > > And so phys_addr_t was introduced, iirc, for PowerPC where we had a
> > > 32bit architecture with 36bit addresses and had to deal with that.
> >
> > OK, thanks.
> >
> > I'm a bit unsure what to do here...but am leaning towards having a
> > separate struct for the memory areas in struct efi_mem_list, rather
> > than using the same efi_mem_desc. Then I can have one use addresses
> > and the other pointers.
>
> That does not make sense to me.  Why would you want to use pointers to
> manage what is effectively a memory address map?

I sent a series with the idea. Basically it is to use addresses within
the efi_memory.c implementation, doing the pointer <-> address
conversion in efi_boottime.c

So efi_memory.c has a table with real addresses. The
address-pretending-to-be-pointer happens only in efi_boottime.c and
doesn't spread any further.

Regards,
Simon