On Thu, Oct 17, 2024 at 10:34:44PM +0200, Michal Hocko wrote:
> On Thu 17-10-24 14:22:13, Kent Overstreet wrote:
> > Why's this being added to stable? It's definitely not a bugfix, and the
> > reasoning ("inherently unsafe") was also incorrect.
>
> I do agree this is not a stable material because this is not fixing
> any actual bug. This is more about future maintainability stuff.
>
> So I agree with Kent that backpotrint this patch to stable trees make
> very little sense.
Ease of future backports is a valid point.
But, the revert never should've gone in, in the first place. The
"inherently unsafe" claim was specious, it turned out that a good
portion of the GFP_NOFAIL allocations we care about do in fact already
have error paths.
Meanwhile, this was the only thing we needed to add malloc() =>
kmalloc(GFP_KERNEL); and the lack of that means that now the Rust folks
have to duplicate and change all their stdlib container code for the
kernel to plumb gfp flags - and that's really fatal for any future hopes
of code sharing betwene userspace and the kernel, which otherwise we
would easily be able to do in Rust.
