Re: sh4 pmap bug? (Re: port-dreamcast/34243)

[email protected] wrote:

> > If there is no objection, I'll commit this patch in this weekend.
> > 
> > It isn't perfect as I wrote in the another post, but at least it
> > could be a workaround for the PR, i.e. all (but not sure) binaries
> > work fine on patched kernels on my dreamcast.
> 
> Yes, please.  Also, please, add a big XXX comment with a summary of
> what's going on.
> 
> We also need to pull this up to 4.0.

Well, Chuck Silvers pointed out we should also have PMAP_PREFER()
for SH4 to avoid some possible aliases.
(I thought we already had it)

I'll take a look at it.

Note current sh3/sh4 pmap also uses P1SEG for pools
(see PMAP_{MAP,UNMAP}_POOLPAGE() in sh3/include/pmap.h),
but I'm afraid it could also cause possible aliases
like mips ports.

I'm also afraid we should have options SOSEND_NO_LOAN
to avoid another possible aliases by pmap_kenter_pa(9)
in sosend_loan().

Umm, we really need MI VM support to handle VIPT cache aliases...
---
Izumi Tsutsui