[GIT PULL] u-boot-riscv/master

[Conor Dooley]

On Mon, May 26, 2025 at 09:02:46AM +0530, Mayuresh Chitale wrote:
> On Thu, May 22, 2025 at 10:11 PM Yao Zi <ziyao at disroot.org> wrote:
> >
> > On Thu, May 22, 2025 at 12:28:18PM +0100, Conor Dooley wrote:
> > > On Wed, May 21, 2025 at 12:39:50PM -0600, Tom Rini wrote:
> > > > On Wed, 21 May 2025 17:50:03 +0800, Leo Liang wrote:

> > > This PR seems to have made my CI blow up, and I'm not entirely sure if
> > > that's something intentional or not. I've not yet bisected, but since
> > > the error is "Image arch not compatible with host arch", I can only
> > > imagine the patch in question is:
> > > | Subject: [PATCH v2 1/3] riscv: image: Add new image type for RV64
> > > | Date: Fri,  4 Apr 2025 14:48:55 +0000       [thread overview]
> > > | Message-ID: <20250404144859.112313-2-mchitale at ventanamicro.com> (raw)
> > > | In-Reply-To: <20250404144859.112313-1-mchitale at ventanamicro.com>
> > > |
> > > | Similar to ARM and X86, introduce a new image type which allows u-boot
> > > | to distinguish between images built for 32-bit vs 64-bit Risc-V CPUs.
> > > |
> > > | Signed-off-by: Mayuresh Chitale <mchitale at ventanamicro.com>
> > > | Reviewed-by: Maxim Moskalets <maximmosk4 at gmail.com>
> > > | ---
> > > |  boot/image.c    | 3 ++-
> > > |  include/image.h | 3 ++-
> > > |  2 files changed, 4 insertions(+), 2 deletions(-)
> > > |
> > > | diff --git a/boot/image.c b/boot/image.c
> > > | index 139c5bd035a..45299a7dc33 100644
> > > | --- a/boot/image.c
> > > | +++ b/boot/image.c
> > > | @@ -92,7 +92,8 @@ static const table_entry_t uimage_arch[] = {
> > > |     {       IH_ARCH_ARC,            "arc",          "ARC",          },
> > > |     {       IH_ARCH_X86_64,         "x86_64",       "AMD x86_64",   },
> > > |     {       IH_ARCH_XTENSA,         "xtensa",       "Xtensa",       },
> > > | -   {       IH_ARCH_RISCV,          "riscv",        "RISC-V",       },
> > > | +   {       IH_ARCH_RISCV,          "riscv",        "RISC-V 32 Bit",},
> > > | +   {       IH_ARCH_RISCV64,        "riscv64",      "RISC-V 64 Bit",},
> > > |     {       -1,                     "",             "",             },
> > > |  };
> > > |
> > > | diff --git a/include/image.h b/include/image.h
> > > | index 07912606f33..411bfcd0877 100644
> > > | --- a/include/image.h
> > > | +++ b/include/image.h
> > > | @@ -138,7 +138,8 @@ enum {
> > > |     IH_ARCH_ARC,                    /* Synopsys DesignWare ARC */
> > > |     IH_ARCH_X86_64,                 /* AMD x86_64, Intel and Via */
> > > |     IH_ARCH_XTENSA,                 /* Xtensa       */
> > > | -   IH_ARCH_RISCV,                  /* RISC-V */
> > > | +   IH_ARCH_RISCV,                  /* RISC-V 32 bit*/
> > > | +   IH_ARCH_RISCV64,                /* RISC-V 64 bit*/
> > > |
> > > |     IH_ARCH_COUNT,
> > > |  };
> > > | --
> > > | 2.43.0
> > > |
> > > since it is changing the existing "riscv" image type to be the 32-bit
> > > image and requiring the new entry for 64-bit. My CI job uses the system
> > > mkimage to create the image that U-Boot is loading, so it doesn't know
> > > about the new define etc. Maybe it's not considered a problem if a new
> > > U-Boot cannot boot an old image, but the comment above the enum reads:
> > > |/*
> > > | * CPU Architecture Codes (supported by Linux)
> > > | *
> > > | * The following are exposed to uImage header.
> > > | * New IDs *MUST* be appended at the end of the list and *NEVER*
> > > | * inserted for backward compatibility.
> > > | */
> > > The overwhelming majority of existing supported boards in U-Boot are
> > > 64-bit platforms, and the 64-bit platforms are the ones that have been
> > > supported for longer, so my thought would be that the compatibility of
> > > 64-bit platforms should be prioritised over 32-bit? Or even add explicit
> > > 32-bit and 64-bit entries and the existing one is a catch-all for
> > > compatibility reasons?
> >
> > I've mentioned entries with bitwidth explicitly specified in my previous
> > reply (and there hasn't been any response).
> >
> > > I'll consider IH_ARCH_RISCV32 a better idea, instead of implying 32bit
> > > when no suffix attached. We (and the Linux kernel) mix 32-bit and 64-bit
> > > variants of RISC-V together, thus it's hard to tell the exact bitwidth
> > > of "IH_ARCH_RISCV" without inspecting the code around. To me, it sounds
> > > more like "RISC-V, but no bitwidth specified".
> > >
> > > It will be nice if we could avoid this kind of ambiguity.
> >
> > (referring my own reply[1])
> >
> > I'll second explicit 32-bit and 64-bit entries, and keeping
> > IH_ARCH_RISCV for compatibility consideration.
> >
> Are you suggesting something like this :
> 
> { IH_ARCH_RISCV, "riscv", "RISC-V",},
> { IH_ARCH_RISCV32, "riscv32", "RISC-V 32 Bit",},
> { IH_ARCH_RISCV64, "riscv64", "RISC-V 64 Bit",},
> 
> I tried to implement it like on ARM64. However the problem is that
> while on ARM64 it switches to aarch32 if it encounters an IH_ARCH_ARM
> kernel image, there is no equivalent mechanism for Risc-V yet.

But we don't even want that, since the current behaviour on a 64-bit
system is not to do anything 32-bit related, right? My 64-bit images
created with my system mkimage use IH_ARCH_RISCV.

> Another
> option could be to disable this check in the bootm flow for now until
> a similar mechanism is implemented on Risc-V.

Why not just allow IH_ARCH_RISCV to retain the current "dumb" behaviour
permanently, and only implement the strict checking when the explicit
width types are detected?

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 228 bytes
Desc: not available
URL: <https://lists.denx.de/pipermail/u-boot/attachments/20250526/60dd81ef/attachment.sig>