Re: [PATCH] arch:x86:coco:sev: Initialize ctxt variable

["Ragavendra B.N." <ragavendra.bn@gmail.com>]

On Mon, Nov 18, 2024 at 01:50:55PM -0600, Tom Lendacky wrote:
> On 11/18/24 13:43, Ragavendra B.N. wrote:
> > On Mon, Nov 18, 2024 at 08:53:04AM -0600, Tom Lendacky wrote:
> >> On 11/18/24 08:44, Tom Lendacky wrote:
> >>> On 11/15/24 16:55, Ard Biesheuvel wrote:
> >>>> On Fri, 15 Nov 2024 at 20:53, Ragavendra B.N. <ragavendra.bn@gmail.com> wrote:
> >>>>>
> >>>>> On Fri, Nov 15, 2024 at 12:02:27PM +0100, Ard Biesheuvel wrote:
> >>>>>> On Fri, 15 Nov 2024 at 12:01, Ingo Molnar <mingo@kernel.org> wrote:
> >>>>>>>
> >>>>>>>
> >>>>>>> * Ragavendra <ragavendra.bn@gmail.com> wrote:
> >>>>>>>
> >>>>>>>> Updating the ctxt value to NULL in the svsm_perform_ghcb_protocol as
> >>>>>>>> it was not initialized.
> >>>>>>>>
> >>>>>>>> Fixes: 2e1b3cc9d7f7 (grafted) Merge tag 'arm-fixes-6.12-2' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc
> >>>>>>>
> >>>>>>> This 'Fixes' tag looks bogus.
> >>>>>>>
> >>>>>>
> >>>>>> So does the patch itself - 'struct es_em_ctxt ctxt' is not a pointer.
> >>>>> Thank you very much for your response. I am relatively new to kernel development.
> >>>>>
> >>>>> I know we can use kmalloc for memory allocation. Please advice.
> >>>>>
> >>>>> struct es_em_ctxt ctxt = kmalloc(sizeof(struct es_em_ctxt), GFP_KERNEL);
> >>>>>
> >>>>> I am thinking to update like above, but like you mentioned, ctxt is not a pointer. I can update this to be a pointer if needed.
> >>>>>
> >>>>
> >>>> The code is fine as is. Let's end this thread here, shall we?
> >>>
> >>> I was assuming he got some kind of warning from some compiler options or
> >>> a static checker. Is that the case Ragavendra?
> >>>
> >>> When I look at the code, it is possible for ctxt->fi.error_code to be
> >>> left uninitialized. The simple fix is to just initialize ctxt as:
> >>>
> >>> 	struct es_em_ctxt ctxt = {};
> >>
> >> Although to cover all cases now and going forwared, the es_em_ctxt fi
> >> member should just be zeroed in verify_exception_info() when
> >> ES_EXCEPTION is going to be returned.
> >>
> >> Thanks,
> >> Tom
> >>
> >>>
> >>> Thanks,
> >>> Tom
> > 
> > Yes Tom, that is exactly the reason I worked on it the first place. The issue was reported by the Coverity tool.
> > 
> > I can send the below fix if that is fine.
> >>> 	struct es_em_ctxt ctxt = {};
> > 
> > For the es_em_ctxt fi member to be zeroed, I can go ahead and assign 0 to all the three long members like below in verify_exception_info()
> > 
> > 
> > 	if (info & SVM_EVTINJ_VALID_ERR) {
> > 		ctxt->fi.error_code = info >> 32;
> > 	} else {
> > 		ctxt->fi.error_code = 0;
> > 		ctxt->fi.vector = 0;
> > 		ctxt->fi.cr2 = 0;
> 
> But then the cr2 value isn't set/zeroed in the true path of the if
> statement. I think a simple memset() at the beginning of the if path
> that will return ES_EXCEPTION is simplest.
> 
> Thanks,
> Tom
> 
> > 
> > 	}
> > 
> > return ES_EXCEPTION;
> > 
> > Thanks,
> > Ragavendra N.

I am assuming something like below.

/* Check if exception information from hypervisor is sane. */
if ((info & SVM_EVTINJ_VALID) &&
		((v == X86_TRAP_GP) || (v == X86_TRAP_UD)) &&
		((info & SVM_EVTINJ_TYPE_MASK) == SVM_EVTINJ_TYPE_EXEPT)) {

	memset(ctxt->fi, 0, sizeof(es_fault_info));

	ctxt->fi.vector = v;

PS - My C skills is not that great as well, as I am from Java/ C# background.


Thanks,
Ragavendra N.