> On Jul 16, 2025, at 17:47, Jakub Jelinek <ja...@redhat.com> wrote: > > On Wed, Jul 16, 2025 at 09:22:19PM +0000, Qing Zhao wrote: >> Yes, the above solution could also resolve the undefined behavior issue. We >> can certainly go >> with this approach. > > Another option is to use .ACCESS_WITH_SIZE (with different flags compared to > the FAM cases) solely on reads from the pointer, and then objsz pass taking > the indirection into account.
A little confused here, For this option, do you mean the first argument will be the ADDRESS of the pointer? > I.e. when you see some pointer dereferenced > and the pointer originates in .ACCESS_WITH_SIZE with that flag, take it as > size of what the pointer points at rather than the FAM way. Yes, this is what I did in the 7th and 8th version of the patch. > Still, > .ACCESS_WITH_SIZE first argument should be the pointer value, not its > address, What’s the major issue with passing the address of the pointer value to .ACCESS_WITH_SIZE? > and you should never instrument stores to the pointer, only reads > from it. > So, f->p = whatever; doesn't get .ACCESS_WITH_SIZE, > _1 = f->p; gets it and then f->p[a] is Still my question: How to decide it’s a store or a read when we see f->p in C parser? Is there any good example in C FE I can refer? Thanks a lot. Qing > _2 = _1 p+ off; > MEM_REF[_2] > opr similar. > > Jakub >
