Re: [RFC PATCH v2 1/2] target/ppc: Moved functions out of mmu-hash64

[Lucas Mateus Martins Araujo e Castro]

On 03/05/2021 01:24, David Gibson wrote:
> On Fri, Apr 30, 2021 at 03:40:46PM -0300, Lucas Mateus Castro (alqotel) wrote:
> > The functions ppc_store_lpcr, ppc_hash64_filter_pagesizes and
> > ppc_hash64_unmap_hptes have been moved to mmu-misc.h since they are
> > not needed in a !TCG context and mmu-hash64 should not be compiled
> > in such situation.
> >
> > ppc_store_lpcr and ppc_hash64_filter_pagesizes are used by multiple
> > functions, while ppc_hash64_unmap_hptes is used by rehash_hpt (in
> > spapr_hcall.c).
> Hmm.. looking at it, ppc_store_lpcr() (and helper_store_lpcr()) don't
> really belong in this file at all.  The LPCR has some things related
> to the hash MMU, but plenty of others that don't.  So, maybe
> misc_helper.c?  That might have to be moved again, since misc_helper
> itself should probably mostly not be used for !TCG.  But.. one thing
> at a time.

I tested here and compiling misc_helper.c with disable-tcg it's kind of

complicated and it would require many changes in it, so for this patch

just move it there and deal with it in a later patch?

> AFAICT the only user of ppc_hash64_filter_pagesizes() is in
> spapr_caps.c.  For now you can just move it next to the caller, it's
> debatable whether it belongs more to PAPR or MMU code.

Also I'm assuming the prototype should also be moved from

"target/ppc/mmu-hash64.h" to "include/hw/ppc/spapr.h" (or some other

spapr_*.h file), or should it be left in the original file?

> ppc_hash64_unmap_hptes() is definitely TCG only and should stay where
> it is.  The call from rehash_hpt() can be solved because rehash_hpt()
> itself is TCG only.  I've already suggested splitting the TCG (well,
> softmmu) only things out from spapr_hcall.c, so it might simplify
> things to tackle that first.