On 11/14/2023 11:21 AM, Michael D Kinney wrote:
Hi Ranbir,
First I want to recognize your efforts to collect Coverity issues and
propose changes to address
them.
I still disagree with adding CpuDealLoop() for any static analysis issues.
There have been previous discussions about adding a PANIC() or FATAL()
macro that would
perform platform specific actions if a condition is detected where the
boot of the platform
can not continue. A platform get to make the choice to log or reboot or
hang, etc. Not the
code that detected the condition.
https://edk2.groups.io/g/devel/message/86597
<https://edk2.groups.io/g/devel/message/86597>
After going through hundreds of edk2 static analysis findings, we found
a small number of cases where an interface such as PanicLib was useful
and recently added an implementation
https://github.com/microsoft/mu_basecore/blob/release/202302/MdePkg/Include/Library/PanicLib.h.
For example, the return value from calls to MpInitLibWhoAmI() in
exception related code often goes unchecked and it's been used there.
Being able to separate the library instance implementation linked to a
given module from a more broad library class like DebugLib for these
cases has also been helpful.
Unfortunately, in order to fix some of these static analysis issues
correctly, we are going
to have to identify the use of ASSERT() that really is a fatal condition
that can not continue
and introduce an implementation approach that provides a platform
handler and
satisfies the static analysis tools.
We also have to evaluate if a return error status with a DEBUG_ERROR
message would be a better
choice than an ASSERT() that can be filtered out by build options.
Best regards,
Mike
*From:* devel@edk2.groups.io <devel@edk2.groups.io> *On Behalf Of
*Ranbir Singh
*Sent:* Tuesday, November 14, 2023 7:08 AM
*To:* Laszlo Ersek <ler...@redhat.com>
*Cc:* devel@edk2.groups.io; Ni, Ray <ray...@intel.com>; Veeresh Sangolli
<veeresh.sango...@dellteam.com>
*Subject:* Re: [edk2-devel] [PATCH v2 4/5]
MdeModulePkg/Bus/Pci/PciBusDxe: Fix NULL_RETURNS Coverity issue
On Mon, Nov 13, 2023 at 4:58 PM Laszlo Ersek <ler...@redhat.com
<mailto:ler...@redhat.com>> wrote:
On 11/10/23 07:11, Ranbir Singh wrote:
> EFI_NOT_READY was listed as one of the error return values in the
> function header of StartPciDevices(). So, I considered it here.
>
> Maybe we can go by a dual approach, including CpuDeadLoop() in
> StartPciDevices() as well as add return value check at the call
site in
> PciBusDriverBindingStart().
I don't think this makes much sense, given that execution is not
supposed to continue past CpuDeadLoop(), so the new error check would
not be reached.
I think two options are realistic:
- replace the assert() with a CpuDeadLoop() -- this is my preference
- keep the assert, add the "if", and in the caller, handle the
EFI_NOT_READY error. This is workable too. (Keeping the assert is goog
because it shows that we don't expect the "if" to fire.)
Anyway, now that we have no way to verify the change against Coverity, I
don't know if this patch is worth the churn. :( As I said, this ASSERT()
is one of those few justified ones in edk2 core that can indeed never
fail, IMO.
Laszlo
See, Does the following change look acceptable to you ?
ASSERT (RootBridge != NULL);
+ if (RootBridge == NULL) {
+ CpuDeadLoop ();
+ return EFI_NOT_READY;
+ }
+
which retains the existing assert, adds the NULL pointer check and
includes CpuDeadLoop () within the if block. As a result of CpuDeadLoop
(), the return statement afterwards will never reach execution (=> no
need to handle this return value at the call sites), but will satisfy
static analysis tools as the "RootBridge" dereference scenario doesn't
arise thereafter.
>
> On Tue, Nov 7, 2023 at 10:18 PM Laszlo Ersek <ler...@redhat.com
<mailto:ler...@redhat.com>
> <mailto:ler...@redhat.com <mailto:ler...@redhat.com>>> wrote:
>
> On 11/7/23 07:19, Ranbir Singh wrote:
> > From: Ranbir Singh <ranbir.sin...@dell.com
<mailto:ranbir.sin...@dell.com>>
> >
> > The function StartPciDevices has a check
> >
> > ASSERT (RootBridge != NULL);
> >
> > but this comes into play only in DEBUG mode. In Release
mode, there
> > is no handling if the RootBridge value is NULL and the code
proceeds
> > to unconditionally dereference "RootBridge" which will lead
to CRASH.
> >
> > Hence, for safety add NULL pointer checks always and return
> > EFI_NOT_READY if RootBridge value is NULL which is one of
the return
> > values as mentioned in the function description header.
> >
> > REF: https://bugzilla.tianocore.org/show_bug.cgi?id=4239
<https://bugzilla.tianocore.org/show_bug.cgi?id=4239>
> <https://bugzilla.tianocore.org/show_bug.cgi?id=4239
<https://bugzilla.tianocore.org/show_bug.cgi?id=4239>>
> >
> > Cc: Ray Ni <ray...@intel.com <mailto:ray...@intel.com>
<mailto:ray...@intel.com <mailto:ray...@intel.com>>>
> > Co-authored-by: Veeresh Sangolli
<veeresh.sango...@dellteam.com <mailto:veeresh.sango...@dellteam.com>
> <mailto:veeresh.sango...@dellteam.com
<mailto:veeresh.sango...@dellteam.com>>>
> > Signed-off-by: Ranbir Singh <ranbir.sin...@dell.com
<mailto:ranbir.sin...@dell.com>>
> > Signed-off-by: Ranbir Singh <rsi...@ventanamicro.com
<mailto:rsi...@ventanamicro.com>
> <mailto:rsi...@ventanamicro.com
<mailto:rsi...@ventanamicro.com>>>
> > ---
> > MdeModulePkg/Bus/Pci/PciBusDxe/PciDeviceSupport.c | 5 ++++-
> > 1 file changed, 4 insertions(+), 1 deletion(-)
> >
> > diff --git a/MdeModulePkg/Bus/Pci/PciBusDxe/PciDeviceSupport.c
> b/MdeModulePkg/Bus/Pci/PciBusDxe/PciDeviceSupport.c
> > index 581e9075ad41..3de80d98370e 100644
> > --- a/MdeModulePkg/Bus/Pci/PciBusDxe/PciDeviceSupport.c
> > +++ b/MdeModulePkg/Bus/Pci/PciBusDxe/PciDeviceSupport.c
> > @@ -772,7 +772,10 @@ StartPciDevices (
> > LIST_ENTRY *CurrentLink;
> >
> > RootBridge = GetRootBridgeByHandle (Controller);
> > - ASSERT (RootBridge != NULL);
> > + if (RootBridge == NULL) {
> > + return EFI_NOT_READY;
> > + }
> > +
> > ThisHostBridge = RootBridge->PciRootBridgeIo->ParentHandle;
> >
> > CurrentLink = mPciDevicePool.ForwardLink;
>
> I don't think this is a good fix.
>
> There is one call site, namely in PciBusDriverBindingStart().
That call
> site does not check the return value. (Of course /s)
>
> I think that this ASSERT() can indeed never fail. Therefore I
suggest
> CpuDeadLoop() instead.
>
> If you insist that CpuDeadLoop() is "too risky" here, then
the patch is
> acceptable, but then the StartPciDevices() call site in
> PciBusDriverBindingStart() must check the error properly: we
must not
> install "gEfiPciEnumerationCompleteProtocolGuid", and the
function must
> propagate the error outwards.
>
> Laszlo
>
-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#111214): https://edk2.groups.io/g/devel/message/111214
Mute This Topic: https://groups.io/mt/102438320/21656
Group Owner: devel+ow...@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-
