From mboxrd@z Thu Jan  1 00:00:00 1970
From: joel@joelfernandes.org (Joel Fernandes)
Date: Thu, 11 Oct 2018 09:28:40 -0700
Subject: [Cocci] Searching for parameter usages of pte_alloc()
In-Reply-To: <alpine.DEB.2.20.1810111746210.3392@hadrien>
References: <20181010211125.GA128537@joelaf.mtv.corp.google.com>
 <93da55ff-c807-6587-7ef3-3d2af820117d@users.sourceforge.net>
 <20181011154346.GA186033@joelaf.mtv.corp.google.com>
 <alpine.DEB.2.20.1810111746210.3392@hadrien>
Message-ID: <20181011162840.GA199386@joelaf.mtv.corp.google.com>
To: cocci@systeme.lip6.fr
List-Id: cocci@systeme.lip6.fr

On Thu, Oct 11, 2018 at 05:51:41PM +0200, Julia Lawall wrote:
> 
> 
> On Thu, 11 Oct 2018, Joel Fernandes wrote:
> 
> > On Thu, Oct 11, 2018 at 03:26:47PM +0200, SF Markus Elfring wrote:
> > > > Interestingly, if I do the following then that works too without needing 'exists':
> > >
> > > How do you think about the following variant for a SmPL rule in your
> > > source code analysis approach?
> > >
> > >
> > > @pte_args depends on report@
> > > identifier I;
> > > type T;
> > > position p;
> > > @@
> > >  pte_alloc at p(..., T I)
> > >  {
> > >  <+... I ...+>
> > >  }
> >
> > The thing is I know the exact parameter to match, and they are fixed in
> > number so I don't really care about that. But I am struggling a little with
> > other things and would like help from you and Julia. There are 3 things (and
> > sorry if some of this is like, wishful stuff, but I thought its better to ask
> > than not and may be let you know what features would be useful):
> >
> > 1. How do I match function names partially? For example I want to do,
> > something like this but I get errors, so I have to write a rule for each
> > function name variant, like pte_alloc_one, or for __pte_alloc. That can be
> > error prone if I miss something.
> >
> > @pte_args depends on report@
> > identifier E1, E2;
> > type T1, T2;
> > position p;
> > @@
> >
> >  ...pte_alloc... at p(T1 E1, T2 E2)
> >  {
> > <+...
> >  E2
> > ...+>
> >  }
> 
> You can use regular expressions.  See demos/regexp.cocci.  A more reliable
> way though would be to consider what is the common properties of these
> functions.  Are they stored in some structure for example?  You can make a
> series of rules to address this, eg:

Awesome! I'll try this.

> 
> @r@
> identifier x,a,pte_alloc;
> @@
> 
> struct foo x { .a = pte_alloc, };
> 
> @@
> identifier r.pte_alloc;
> @@
> 
> pte_alloc(...) { ... } // do whatever you want with the pte_alloc function

Cool, this identifier stuff seems useful. In this case the functions are just
defined in architecture specific headers and called from the mm/ code. They
are not store in a structure. I am hoping I can using the regex say assign
the match to an identifier, and use it in other rules perhaps. I'll send you
what I have today on the -mm thread so if I'm doing something silly
you can let me know (and much appreciate your help!).

> >
> > 2. How do I write a rule that renames function names using the "(" and "|"
> > syntax? I can do it for callers, but for function names this technique fails
> > as shown below, so again I have to end up writing separate rules:
> >
> > // Options: --no-includes --include-headers
> > @pte_args1 depends on patch exists@
> > identifier E1, E2;
> > type T1, T2;
> > @@
> >
> > (
> > - pte_alloc(T1 E1, T2 E2)
> > + pte_alloc(T1 E1)
> > |
> > - pte_alloc_kernel(T1 E1, T2 E2)
> > + pte_alloc_kernel(T1 E1)
> > )
> > { ... }
> >
> > Also something like the partial match would be even better, so something like:
> > - ...pte_alloc... at p(T1 E1, T2 E2)
> > + ...pte_alloc... at p(T1 E1)
> > { ... }
> >
> > And I want to do the same thing for callers as well:
> > - ...pte_alloc... at p(T1 E1, T2 E2);
> > + ...pte_alloc... at p(T1 E1);
> > Atleast for callers, the "(" and "|" syntax works, but for function
> > definitions, nothing works so I have to end up writing separate rules.
> 
> Hopefully the answer to question 1 will make this question irrelevant.

Yes, I think so.

> > 3. How do I match macro definitions of pte_alloc defined using #define, and
> > apply rules on those?  (also perhaps match *pte_alloc*). I tried writing
> > various rules to no luck making the spatch happy.
> 
> Sorry, I'm not completely sure what you want here.  Please send some
> examples of what you want to match.

Sure, so if I have code like this:
#define pte_alloc_one(mm, vmaddr) ((pte_t *) page_table_alloc(mm))

I want to have a rule that does:
- #define pte_alloc_one(mm, vmaddr) ((pte_t *) page_table_alloc(mm))
+ #define pte_alloc_one(mm) ((pte_t *) page_table_alloc(mm))

So far everything I tried only works for functions so I was wondering how one
do this with macros.

thanks,

- Joel