diff mbox series

[PULL,06/30] spapr: Do PHB hoplug sanity check at pre-plug

Message ID 20201214045807.41003-7-david@gibson.dropbear.id.au
State New
Headers show
Series [PULL,01/30] spapr/xive: Turn some sanity checks into assertions | expand

Commit Message

David Gibson Dec. 14, 2020, 4:57 a.m. UTC
From: Greg Kurz <groug@kaod.org>

We currently detect that a PHB index is already in use at plug time.
But this can be decteted at pre-plug in order to error out earlier.

This allows to pass &error_abort to spapr_drc_attach() and to end
up with a plug handler that doesn't need to report errors anymore.

Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <20201120234208.683521-8-groug@kaod.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
---
 hw/ppc/spapr.c | 17 +++++++++++------
 1 file changed, 11 insertions(+), 6 deletions(-)

Comments

Peter Maydell Dec. 15, 2020, 4:56 p.m. UTC | #1
On Mon, 14 Dec 2020 at 04:58, David Gibson <david@gibson.dropbear.id.au> wrote:
>
> From: Greg Kurz <groug@kaod.org>
>
> We currently detect that a PHB index is already in use at plug time.
> But this can be decteted at pre-plug in order to error out earlier.
>
> This allows to pass &error_abort to spapr_drc_attach() and to end
> up with a plug handler that doesn't need to report errors anymore.
>
> Signed-off-by: Greg Kurz <groug@kaod.org>
> Message-Id: <20201120234208.683521-8-groug@kaod.org>
> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

Hi; this change seems to have nudged one of Coverity's
heuristics into deciding that spapr_drc_by_id() can return
NULL (because its return value is checked here, I suspect),
so it reports CID 1437757, 1437758, where spapr_add_lmbs()
and spapr_memory_unplug_request() both take the return value
of spapr_drc_by_id() and pass it directly to spapr_drc_index(),
which will crash if it is passed a NULL pointer.

Is it impossible for spapr_drc_by_id() to return NULL in
those functions (ie Coverity false positive) or is there
a missing error check ?

thanks
-- PMM
Greg Kurz Dec. 15, 2020, 5:31 p.m. UTC | #2
On Tue, 15 Dec 2020 16:56:36 +0000
Peter Maydell <peter.maydell@linaro.org> wrote:

> On Mon, 14 Dec 2020 at 04:58, David Gibson <david@gibson.dropbear.id.au> wrote:
> >
> > From: Greg Kurz <groug@kaod.org>
> >
> > We currently detect that a PHB index is already in use at plug time.
> > But this can be decteted at pre-plug in order to error out earlier.
> >
> > This allows to pass &error_abort to spapr_drc_attach() and to end
> > up with a plug handler that doesn't need to report errors anymore.
> >
> > Signed-off-by: Greg Kurz <groug@kaod.org>
> > Message-Id: <20201120234208.683521-8-groug@kaod.org>
> > Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
> 
> Hi; this change seems to have nudged one of Coverity's
> heuristics into deciding that spapr_drc_by_id() can return
> NULL (because its return value is checked here, I suspect),
> so it reports CID 1437757, 1437758, where spapr_add_lmbs()
> and spapr_memory_unplug_request() both take the return value
> of spapr_drc_by_id() and pass it directly to spapr_drc_index(),
> which will crash if it is passed a NULL pointer.
> 
> Is it impossible for spapr_drc_by_id() to return NULL in
> those functions (ie Coverity false positive) or is there
> a missing error check ?
> 

No, all DRC objects are created before any of these two
functions are called. Each function happens to loop over
the full list of memory DRCs a few line above the offending
call sites and already assert spapr_drc_by_id() doesn't
return NULL. But I guess Coverity isn't smart enough to
detect that.

I'll post a patch to add some more assertions.

> thanks
> -- PMM
diff mbox series

Patch

diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
index 28d266f7a7..ac115b0987 100644
--- a/hw/ppc/spapr.c
+++ b/hw/ppc/spapr.c
@@ -3886,6 +3886,7 @@  static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
     SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(dev);
     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
     const unsigned windows_supported = spapr_phb_windows_supported(sphb);
+    SpaprDrc *drc;
 
     if (dev->hotplugged && !smc->dr_phb_enabled) {
         error_setg(errp, "PHB hotplug not supported for this machine");
@@ -3897,6 +3898,12 @@  static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
         return false;
     }
 
+    drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, sphb->index);
+    if (drc && drc->dev) {
+        error_setg(errp, "PHB %d already attached", sphb->index);
+        return false;
+    }
+
     /*
      * This will check that sphb->index doesn't exceed the maximum number of
      * PHBs for the current machine type.
@@ -3910,8 +3917,7 @@  static bool spapr_phb_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
                            errp);
 }
 
-static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
-                           Error **errp)
+static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev)
 {
     SpaprMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
     SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
@@ -3927,9 +3933,8 @@  static void spapr_phb_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
     /* hotplug hooks should check it's enabled before getting this far */
     assert(drc);
 
-    if (!spapr_drc_attach(drc, dev, errp)) {
-        return;
-    }
+    /* spapr_phb_pre_plug() already checked the DRC is attachable */
+    spapr_drc_attach(drc, dev, &error_abort);
 
     if (hotplugged) {
         spapr_hotplug_req_add_by_index(drc);
@@ -3997,7 +4002,7 @@  static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
         spapr_core_plug(hotplug_dev, dev, errp);
     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
-        spapr_phb_plug(hotplug_dev, dev, errp);
+        spapr_phb_plug(hotplug_dev, dev);
     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_TPM_PROXY)) {
         spapr_tpm_proxy_plug(hotplug_dev, dev, errp);
     }