On Mon, Oct 26, 2020 at 03:46:47PM +0100, Greg Kurz wrote:
> On Mon, 26 Oct 2020 14:43:08 +0100
> Philippe Mathieu-Daudé <philmd@redhat.com> wrote:
> 
> > On 10/26/20 1:40 PM, Greg Kurz wrote:
> > > qemu_memalign() aborts if OOM. Drop some dead code.
> > > 
> > > Signed-off-by: Greg Kurz <groug@kaod.org>
> > > ---
> > >   hw/ppc/spapr.c       |    6 ------
> > >   hw/ppc/spapr_hcall.c |    8 ++------
> > >   2 files changed, 2 insertions(+), 12 deletions(-)
> > > 
> > > diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
> > > index 0cc19b5863a4..f098d0ee6d98 100644
> > > --- a/hw/ppc/spapr.c
> > > +++ b/hw/ppc/spapr.c
> > > @@ -1521,12 +1521,6 @@ void spapr_reallocate_hpt(SpaprMachineState *spapr, int shift,
> > >           int i;
> > >   
> > >           spapr->htab = qemu_memalign(size, size);
> > > -        if (!spapr->htab) {
> > > -            error_setg_errno(errp, errno,
> > > -                             "Could not allocate HPT of order %d", shift);
> > > -            return;
> > 
> > Wasn't the idea to use qemu_try_memalign() here?
> > 
> 
> Well... I have mixed feeling around this. The HTAB was first
> introduced by commit:
> 
> commit f43e35255cffb6ac6230dd09d308f7909f823f96
> Author: David Gibson <david@gibson.dropbear.id.au>
> Date:   Fri Apr 1 15:15:22 2011 +1100
> 
>     Virtual hash page table handling on pSeries machine
> 
> using qemu_mallocz(), which was aborting on OOM. It then got
> replaced by g_malloc0() when qemu_mallocz() got deprecated
> and eventually by qemu_memalign() when KVM support was added.
> 
> Surviving OOM when allocating the HTAB never seemed to be an
> option until this commit that introduced the check:
> 
> commit c5f54f3e31bf693f70a98d4d73ea5dbe05689857
> Author: David Gibson <david@gibson.dropbear.id.au>
> Date:   Tue Feb 9 10:21:56 2016 +1000
> 
>     pseries: Move hash page table allocation to reset time
> 
> I don't really see in the patch and in the changelog an obvious
> desire to try to handle OOM.


This one is probably ok.  AFAICT all failures returned here would be
more or less fatal in the caller, one way or another (&error_fatal in
two cases, and failure to load an incoming migration stream in the
other).

> > > -        }
> > > -
> > >           memset(spapr->htab, 0, size);
> > >           spapr->htab_shift = shift;
> > >   
> > > diff --git a/hw/ppc/spapr_hcall.c b/hw/ppc/spapr_hcall.c
> > > index 607740150fa2..34e146f628fb 100644
> > > --- a/hw/ppc/spapr_hcall.c
> > > +++ b/hw/ppc/spapr_hcall.c
> > > @@ -361,12 +361,8 @@ static void *hpt_prepare_thread(void *opaque)
> > >       size_t size = 1ULL << pending->shift;
> > >   
> > >       pending->hpt = qemu_memalign(size, size);
> > > -    if (pending->hpt) {
> > > -        memset(pending->hpt, 0, size);
> > > -        pending->ret = H_SUCCESS;
> > > -    } else {
> > > -        pending->ret = H_NO_MEM;
> > 
> > Ditto.
> > 
> 
> This one was introduced by commit:
> 
> commit 0b0b831016ae93bc14698a5d7202eb77feafea75
> Author: David Gibson <david@gibson.dropbear.id.au>
> Date:   Fri May 12 15:46:49 2017 +1000
> 
>     pseries: Implement HPT resizing
> 
> I agree that maybe the intent here could have been to use qemu_try_memalign(),
> but again I don't quite see any strong justification to handle OOM in the
> changelog.
> 
> David,
> 
> Any insight to share ?

Aborting on an HPT resize failure is definitely not ok, though.  This
one needs to be a qemu_try_memalign().

-- 
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