On Sun, Oct 02, 2022 at 01:16:33PM -0400, Tom Rini wrote:
> On Sun, Oct 02, 2022 at 03:52:00AM +0200, Heinrich Schuchardt wrote:
> > 
> > 
> > On 10/2/22 01:57, Tom Rini wrote:
> > > > > Any thoughts or comments ?
> > > > In our Docker container command
> > > > 
> > > > tools/buildman/buildman -o build -w -E -W  -e --board qemu-riscv32_spl
> > > > 
> > > > leads to build/toolchain with content
> > > > 
> > > > gcc /opt/gcc-11.1.0-nolibc/riscv64-linux/bin/riscv64-linux-gcc
> > > > path /opt/gcc-11.1.0-nolibc/riscv64-linux/bin
> > > > cross riscv64-linux-
> > > > arch riscv64
> > > > 
> > > > When compiling qemu-riscv32_defconfig with Alexandre's patch and
> > > > 
> > > >     export
> > > > CROSS_COMPILE=/opt/gcc-11.1.0-nolibc/riscv64-linux/bin/riscv64-linux-
> > > > 
> > > > I see undefined reference to `__ashldi3'.
> > > Another reason to port the generic *di3 option from modern Linux kernel.
> > 
> > Why should we use a 64bit toolchain for 32bit RISC-V when a 32bit toolchain
> > is available?
> > 
> > Fixing the Docker container and buildman seems to be the logical way
> > forward.
> > 
> > Cf.
> > https://source.denx.de/u-boot/custodians/u-boot-efi/-/commits/riscv
> > https://source.denx.de/u-boot/custodians/u-boot-efi/-/pipelines/13657
> 
> Because it's also yet another example of the problems we have because we
> haven't ported that functionality back, which improves on the state of
> what's in our tree today wrt "libgcc".

Which is not to say I disagree with v3 of your series, this just
reminded me of problems we have that I dug in to last week for other
reasons.

-- 
Tom