Re: Kernel 6.15-rc2 unable to boot on 32bit x86 with PAE

[Mike Rapoport <rppt@kernel.org>]

On Tue, Apr 22, 2025 at 05:02:55PM +0930, Qu Wenruo wrote:
> 
> 
> 在 2025/4/22 16:30, Ingo Molnar 写道:
> > 
> > * Borislav Petkov <bp@alien8.de> wrote:
> > 
> > > On Fri, Apr 18, 2025 at 08:31:23PM +0930, Qu Wenruo wrote:
> > > > Hi,
> > > > 
> > > > Recently I'm testing a situation where highmem is involved, thus I'm
> > > > building the latest 32bit x86 with HIGHMEM and PAE, and run it inside a qemu
> > > > VM.
> > > 
> > > Does that fix it:
> > > 
> > > https://git.kernel.org/tip/1e07b9fad022e0e02215150ca1e20912e78e8ec1
> > > 
> > > ?
> > 
> > That commit caused other problems - the best fix we have right now is:
> > 
> > 	https://git.kernel.org/tip/83b2d345e1786fdab96fc2b52942eebde125e7cd
> 
> Thanks, I'll take a try again.
> 
> But considering how rare 32bit + highmem systems are nowadays, I still have
> one questions/concern:
> 
> - How to make sure we really got highmems?
>   The kernel docs only mention that highmems need temporary mapping,
>   and proper kmap/kunmap() handling.
> 
>   But I still didn't have a straightforward idea of how to verify,
>   without adding manual ftrace events when a highmem page is hit.
> 
>   E.g. free only shows 3GB memory even if I enabled highmem + PAE.

Without HIGHMEM it would be less than 900M.

You can look for "HIGHMEM available" and "Memory: n/m available" in the
kernel log to see how much highmem is there. 
 
>   Since HIGHMEM64G is removed, I'd expect to get some values more like
>   4G instead?

There are some gaps under 4G, so it's surely less that 4G of usable memory
there.
 
> Thanks,
> Qu

-- 
Sincerely yours,
Mike.