Re: [PATCH -fixes] dma-mapping: add default implementation to arch_dma_{set|clear}_uncached

[Robin Murphy]

On 09/07/2024 12:39 pm, Yangyu Chen wrote:
> > On Jul 9, 2024, at 19:19, Christoph Hellwig <hch@xxxxxx> wrote:
> >
> > On Tue, Jul 09, 2024 at 05:25:29PM +0800, Yangyu Chen wrote:
> > > Currently, we have some code in kernel/dma/direct.c which references
> > > arch_dma_set_uncached and arch_dma_clear_uncached. However, many
> > > architectures do not provide these symbols, and the code currently
> > > relies on compiler optimization to cut the unnecessary code. When the
> > > compiler fails to optimize it, the code will reference the symbol and
> > > cause a link error. I found this bug when developing some new extensions
> > > for RISC-V on LLVM. The error message is shown below:
> >
> > Same comment as for the last one.  I think your compiler misbehaves,
> > and the typical reason for that would be if you disable all
> > optimizations.
>
> The reason is that some optimizations failed to apply after adding
> some passes. I will fix the compiler later. Whatever, we should not
> rely on this optimization to get the code being successfully compiled.

We rely on unreachable code being optimised out *all over* the place - 
it's kind of the main point of constructs like IS_ENABLED(), and why the 
kernel as a whole does not and will not support being compiled with -O0. 
Frankly I'm quite surprised that this one would be the only case you hit 
if your compiler is not doing that properly.

Thanks,
Robin.