Re: struct mctelem_cookie missing definition

[Stefano Stabellini <sstabellini@xxxxxxxxxx>]

On Thu, 13 Feb 2025, Jan Beulich wrote:
> On 13.02.2025 03:00, Stefano Stabellini wrote:
> > On Wed, 12 Feb 2025, Stefano Stabellini wrote:
> >> On Thu, 13 Feb 2025, Andrew Cooper wrote:
> >>> On 13/02/2025 1:25 am, Stefano Stabellini wrote:
> >>>> Hi all,
> >>>>
> >>>> I am looking through the few remaining MISRA violations that we have
> >>>> left.  One of them is R11.2:
> >>>>
> >>>> https://saas.eclairit.com:3787/fs/var/local/eclair/xen-project.ecdf/xen-project/hardware/xen/ECLAIR_normal/staging/X86_64/9118578464/PROJECT.ecd;/by_service/MC3A2.R11.2.html#{%22select%22:true,%22selection%22:{%22hiddenAreaKinds%22:[],%22hiddenSubareaKinds%22:[],%22show%22:false,%22selector%22:{%22enabled%22:true,%22negated%22:true,%22kind%22:0,%22domain%22:%22kind%22,%22inputs%22:[{%22enabled%22:true,%22text%22:%22violation%22}]}}}
> >>>>
> >>>> Specifically, mctelem_cookie_t is a pointer to incomplete type and
> >>>> therefore COOKIE2MCTE triggers a "conversion between a pointer to an
> >>>> incomplete type and any other type".
> >>>>
> >>>> mctelem_cookie_t is defined as:
> >>>>
> >>>> typedef struct mctelem_cookie *mctelem_cookie_t;
> >>>>
> >>>> I am looking through the code and I genuinely cannot find the definition
> >>>> of struct mctelem_cookie.
> >>>>
> >>>> If mctelem_cookie_t is only used as a pointer, wouldn't it make more
> >>>> sense to do:
> >>>>
> >>>> typedef struct mctelem_ent *mctelem_cookie_t;
> >>>>
> >>>> ?
> >>>>
> >>>> What am I missing?
> >>>
> >>> Nothing.  Or perhaps the twisted thinking of the original author.
> >>>
> >>> It is genuinely a pointer type (== known size) which you can't deference
> >>> (because there is no definition), and can only operate on by casting to
> >>> an integer.  Except the code also requires it to be a uint64_t which is
> >>> why there's some fun disabling of relevant hypercalls for compat guests.
> >>>
> >>> If someone could find the time to file it in /dev/null and replace it
> >>> with literally anything else, I'd be very thankful.
> >>
> >> Are you OK with typedefing mctelem_cookie_t to uint64_t instead?
> > 
> > I confirm that the following resolves the MISRA violations
> > 
> > diff --git a/xen/arch/x86/cpu/mcheck/mctelem.h 
> > b/xen/arch/x86/cpu/mcheck/mctelem.h
> > index f4c5ff848d..2ccd490e5d 100644
> > --- a/xen/arch/x86/cpu/mcheck/mctelem.h
> > +++ b/xen/arch/x86/cpu/mcheck/mctelem.h
> > @@ -52,7 +52,7 @@
> >   * the element from the processing list.
> >   */
> >  
> > -typedef struct mctelem_cookie *mctelem_cookie_t;
> > +typedef uint64_t *mctelem_cookie_t;
> 
> Yet that makes it possible to de-reference the pointer. Which, as Andrew
> explained, is intended to be impossible. If this could be properly
> replaced (not exactly what Andrew indicated by "file it in /dev/null"),
> fine. Truly purging the code (i.e. as Andrew suggests) may still be an
> option, with appropriate justification. But simply adjusting the type
> and then moving on is too little, imo. Even if you used void * (to make
> de-referencing impossible) I'd view it as largely papering over an issue;
> then converting to other pointers (without explicit cast, and hence
> without making apparent the badness of doing so) would become possible.

What about converting to uintptr_t (not a pointer)?


In general, there are quite a few MISRA rules that we could mark as
blocking (clean) in our GitLab scan with just a few code changes like
this one. My goal is to make these rules blocking as soon as possible.
If I can improve the code in the process, that is even better, but it is
not mandatory. And I would rather spend one more hour marking a second
rule as blocking instead. 

What I mean is that I believe it would be acceptable to make some
compromises and accept non-perfect changes to the code if it helps us
enforce more rules as blocking in GitLab CI.