On Wed, Jul 12, 2017 at 01:11:30PM +0200, Greg Kurz wrote:
> On Wed, 12 Jul 2017 15:52:37 +0530
> Bharata B Rao <bhar...@linux.vnet.ibm.com> wrote:
>
> > On Wed, Jul 12, 2017 at 11:48:39AM +0200, Greg Kurz wrote:
> > > Since commit 5c1da81215c7 ("spapr: Remove unnecessary differences between
> > > hotplug and coldplug paths"), the CPU DT for the DRC is always allocated.
> > > This causes a memory leak for pseries-2.6 and older machine types, that
> > > don't support CPU hotplug and don't allocate DRCs for CPUs.
> > >
> > > Reported-by: Bharata B Rao <bhar...@linux.vnet.ibm.com>
> > > Signed-off-by: Greg Kurz <gr...@kaod.org>
> > > ---
> > > hw/ppc/spapr.c | 9 +++++----
> > > 1 file changed, 5 insertions(+), 4 deletions(-)
> > >
> > > diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
> > > index 12b3f099d4c9..4a480e1c1dd9 100644
> > > --- a/hw/ppc/spapr.c
> > > +++ b/hw/ppc/spapr.c
> > > @@ -2993,8 +2993,6 @@ static void spapr_core_plug(HotplugHandler
> > > *hotplug_dev, DeviceState *dev,
> > > CPUState *cs = CPU(core->threads);
> > > sPAPRDRConnector *drc;
> > > Error *local_err = NULL;
> > > - void *fdt = NULL;
> > > - int fdt_offset = 0;
> > > int smt = kvmppc_smt_threads();
> > > CPUArchId *core_slot;
> > > int index;
> > > @@ -3009,9 +3007,12 @@ static void spapr_core_plug(HotplugHandler
> > > *hotplug_dev, DeviceState *dev,
> > >
> > > g_assert(drc || !mc->has_hotpluggable_cpus);
> > >
> > > - fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
> > > -
> > > if (drc) {
> > > + void *fdt;
> > > + int fdt_offset;
> > > +
> > > + fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
> > > +
> > > spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
> > > if (local_err) {
> > > g_free(fdt);
> >
> > You say this in the patch description already, but want to note explicitly
> > that this prevents double allocation for pseries-2.6 and ealier types and
> > not for newer machine types.
> >
>
> This DT node doesn't have the exact same use as the one allocated in
> spapr_populate_cpus_dt_node(). Here, it is needed by the DRC logic:
What I mean to say is that even after this fix, for machines types > 2.6, we
still allocate memory for FDT and go through the code that creates DT entries
for CPUs twice. What we create via spapr_populate_cpus_dt_node() is used
while the CPU DT entries created by spapr_populate_hotplug_cpu_dt() won't
be used and corresponding drc->fdt is never freed, afaics for boot time
and cold-plugged CPUs.
May be this can be fixed during DT code reorg effort that David is
planning.
Regards,
Bharata.
