Re: [Qemu-ppc] [Qemu-devel] [PATCH 0/3] ppc: complete the new HV mode

[David Gibson]

On Sun, Jun 05, 2016 at 07:41:50PM +0200, Cédric Le Goater wrote:
> Hello Mark,
> 
> On 06/03/2016 03:52 PM, Mark Cave-Ayland wrote:
> > On 03/06/16 13:11, Cédric Le Goater wrote:
> > 
> >> This is follow up to complete the serie "ppc: preparing pnv landing
> >> (round 2)" plus a little fix on instruction privileges.
> >>
> >> Tested on a POWER8 pserie guest and on mac99.
> >>
> >> Benjamin Herrenschmidt (2):
> >>   ppc: Fix hreg_store_msr() so that non-HV mode cannot alter MSR:HV
> >>   ppc: Better figure out if processor has HV mode
> >>
> >> Cédric Le Goater (1):
> >>   ppc: fix hrfid, tlbia and slbia privilege
> >>
> >>  target-ppc/cpu.h            |  4 ++++
> >>  target-ppc/excp_helper.c    |  8 ++++++--
> >>  target-ppc/helper_regs.h    |  4 ++--
> >>  target-ppc/translate.c      | 10 ++++++----
> >>  target-ppc/translate_init.c | 19 +++++++++++++++----
> >>  5 files changed, 33 insertions(+), 12 deletions(-)
> > 
> > Hi Cédric,
> > 
> > I can confirm that this patchset fixes starting up OpenBIOS for both
> > g3beige/mac99 in my tests here. With the escc fix also applied, the only
> > outstanding issue is the removal of the tlb_flush() statements which
> > causes Darwin, MacOS X and HelenOS 0.60 to panic on boot
> > 
> > My only request is if it would be possible to move patch 2 "ppc: Better
> > figure out if processor has HV mode" to the front of this patchset which
> > will make the remaining patches bisectable for the Mac machines. With that:
> > 
> > Tested-by: Mark Cave-Ayland <address@hidden>
> > 
> > Does anyone know if Ben has any ideas as to why the MMU tlb_flush
> > changes patch is causing such problems?
> 
> 
> Here is a fix I think. Could you give it a try ? 

So, I had applied this to ppc-for-2.7, but I've now removed it again.
BenH correctly pointed out that it basically just removes any benefit
of his original tlb_flush() patch, in a slightly more subtle way that
the last "fix".  You just set the need_flush flag whenever IR or DR
are changed, whereas the whole point of BenH's patch is that the
translation on and off modes are now different MM contexts, which
should be flagged in qemu's TLB and so not require a full flush.  We
need to work out what the real problem is here.

-- 
David Gibson                    | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au  | minimalist, thank you.  NOT _the_ _other_
                                | _way_ _around_!
http://www.ozlabs.org/~dgibson

signature.asc
Description: PGP signature