Re: [PATCH 1/4] virtio_console: copy pipe buffers in splice_write()
From: Christian Brauner
Date: Wed Nov 11 2026 - 13:59:53 EST
On Wed, Nov 11, 2026 at 08:43:53AM -0800, Linus Torvalds wrote:
> On Wed, 11 Nov 2026 at 05:42, Christian Brauner <brauner@xxxxxxxxxx> wrote:
> >
> > So the fix below is what we've done before for pipes and I think it's
> > the right thing to do here too. I put the following on vfs.fixes:
>
> Please no.
>
> > - src = kmap_local_page(buf->page);
> > - memcpy(page_address(page) + offset, src + buf->offset, len);
> > - kunmap_local(src);
>
> This boils down to one memcpy.
>
> HIGHMEM is dead. Nobody sane uses it. It *will* go away. It's already
> effectively never ever used.
>
> There are some broken 32-bit arm setups that use it, and legacy museum
> 32-bit x86.
>
> Don't make one memcpy() that can be inlined become this garbage:
>
> > + /* The buffer can span several pages of a compound page. */
> > + kv.iov_base = page_address(page) + offset;
> > + kv.iov_len = len;
> > + iov_iter_kvec(&to, ITER_DEST, &kv, 1, len);
> > + if (copy_page_to_iter(buf->page, buf->offset, len, &to) != len) {
> > + __free_page(page);
> > + return -EFAULT;
> > + }
>
> just because of broken shit that nobody has ever complained about or ever hit.
If that's the official new stance than I'm very happy.
I went by what we have historically done and very recent history still
has us fixing HIGHMEM shit.
So if this is an official thing that we can now say: "Fuck it, we don't
care about HIGHMEM and will ignore bugs because". Then why can't we just
completely kill it for say v7.5?