Re: [Xen-devel] [PATCH v3] gcov: add new interface and 3.4 and 4.7 format support

[Wei Liu <wei.liu2@xxxxxxxxxx>]

On Wed, Oct 12, 2016 at 09:42:17AM -0600, Jan Beulich wrote:
> >>> On 12.10.16 at 17:33, <wei.liu2@xxxxxxxxxx> wrote:
> > On Wed, Oct 12, 2016 at 06:42:53AM -0600, Jan Beulich wrote:
> >> >>> On 11.10.16 at 12:31, <wei.liu2@xxxxxxxxxx> wrote:
> >> > --- /dev/null
> >> > +++ b/xen/common/gcov/gcc_4_7.c
> >> > @@ -0,0 +1,205 @@
> >> > +/*
> >> > + *  This code provides functions to handle gcc's profiling data format
> >> > + *  introduced with gcc 4.7.
> >> > + *
> >> > + *  This file is based heavily on gcc_3_4.c file.
> >> > + *
> >> > + *  For a better understanding, refer to gcc source:
> >> > + *  gcc/gcov-io.h
> >> > + *  libgcc/libgcov.c
> >> > + *
> >> > + *  Uses gcc-internal data definitions.
> >> > + *
> >> > + *  Imported from Linux and modified for Xen by
> >> > + *    Wei Liu <wei.liu2@xxxxxxxxxx>
> >> > + */
> >> > +
> >> > +#include <xen/string.h>
> >> > +
> >> > +#include "gcov.h"
> >> > +
> >> > +#if GCC_VERSION < 40700
> >> > +#error "Wrong version of GCC used to compile gcov"
> >> > +#endif
> >> > +
> >> > +#if (__GNUC__ > 5) || (__GNUC__ == 5 && __GNUC_MINOR__ >= 1)
> >> > +#define GCOV_COUNTERS                   10
> >> > +#elif __GNUC__ == 4 && __GNUC_MINOR__ >= 9
> >> > +#define GCOV_COUNTERS                   9
> >> > +#else
> >> > +#define GCOV_COUNTERS                   8
> >> > +#endif
> >> 
> >> I'm sorry for not having pointed this out on v2 (I had noticed it,
> >> but then didn't finish analyzing the situation), but I'm afraid this
> >> together with ...
> >> 
> >> > +struct gcov_info {
> >> > +    unsigned int version;
> >> > +    struct gcov_info *next;
> >> > +    unsigned int stamp;
> >> > +    const char *filename;
> >> > +    void (*merge[GCOV_COUNTERS])(gcov_type *, unsigned int);
> >> > +    unsigned int n_functions;
> >> > +    struct gcov_fn_info **functions;
> >> > +};
> >> 
> >> ... this structure's trailing fields actually getting used by the code
> >> won't work well when changing compiler versions without cleaning
> >> the tree. I think instead you need thin gcc_5.c and gcc_4_9.c
> >> #define-ing their GCOV_COUNTERS and then #include-ing this
> >> shared source file. Plus btw, I don't think gcc 5.0.x (the
> >> development variant of 5.x) would use anything different from
> >> 5.1.x or 5.2.x; in fact use of __GNUC_MINOR__ should not
> >> normally be necessary anymore with gcc 5+.
> >> 
> > 
> > I think you misread here: __GNUC_MINOR__ is the "x" part of 5.x.y, the
> > "y" part is __GNUC_PATCHLEVEL__.
> 
> No, I didn't. From 5.x onwards the information previously carried in
> __GNUC_PATCHLEVEL__ is now in __GNUC_MINOR__. And as much
> as previously you would not normally need to look at the former,
> with newer gcc you shouldn't need to look at the latter.
> 

I can't find relevant information in GCC cpp manual.

Specifically, I look at 4.9.4 and 5.4.0 doc:

https://gcc.gnu.org/onlinedocs/gcc-4.9.4/cpp/Common-Predefined-Macros.html#Common-Predefined-Macros
https://gcc.gnu.org/onlinedocs/gcc-5.4.0/cpp/Common-Predefined-Macros.html#Common-Predefined-Macros

The sections about __GNUC_* macros are identical, their semantics stay
the same.

What did I miss?

> > I've broken down things into several files as well as provided
> > corresponding Kconfig options:
> > 
> >  gcc_4_7_base.c: the body of what is now gcc_4_7.c, better name is
> >                  welcome
> 
> Why don't you keep it gcc_4_7.c, with its counter definition being
> conditional upon the symbol not already being defined?
> 

Fixed as discussed on IRC.

Wei.

> Jan
> 

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel