On Mon, 9 Nov 2015, Dan Carpenter wrote: > On Sun, Nov 08, 2015 at 10:24:49PM +0000, Julia Lawall wrote: > > On Mon, 9 Nov 2015, Dan Carpenter wrote: > > > > > Cool. So, in grsec they use a GCC plugin to make these const > > > automatically since they only contain function pointers. There about > > > 100 struct types marked as __no_const. Kees would like to adopt the > > > grsec pluggin approach I expect. Do you have an idea how many structs > > > only contain function pointers or how many consts we would have to add > > > to get the same effect without the plugin? > > > > My list has 373 type names. In the list there are counts for good > > (already const) and bad (not const). The sum of the bad values is 2467. > > The list is below. > > > > julia > > Fantastic! Thanks. We could autogenerate the list of type names and > make checkpatch.pl complain if we declared those types as non const. > > I ran this command to find which functions grsec marks as __no_const. > > egrep '(^ struct |^@@|__no_const;)' grsecurity-3.1-4.2.5-201511021814.patch | grep __no_const -B1 | grep -v __no_const | grep -v '^--' | cut -d @ -f 5- | cut -b 9- | cut -d ' ' -f 1 > > There are 60 structs declared as __no_const. For some structs they > declare a no_const version and use it as needed. Like this: > typedef struct net_device_ops __no_const net_device_ops_no_const; > > grep __no_const grsecurity-3.1-4.2.5-201511021814.patch | grep typedef | cut -d ' ' -f 3 > > There are 32 of those. > > Then I compared to see if their structs were on your list. For some > reason there quite a few one their list which are not on yours. Out > of the first 10 about half weren't on your list. cpu_cache_fns, > outer_cache_fns, psci_operations, smp_operations, omap_hwmod_soc_ops, > smp_ops_t. These are mostly different arches? > > Also bit_table has in int has well as a function pointers but it is on > their list. I'm not sure why. Maybe they are marking structs const > that I don't know about. > > The other trick that they do is they define structs as __do_const if > they want them to be const by default, which is pretty neat. This feels > like it should be a standard GCC feature. In the meantime we could > mark things as __do_const and print a sparse warning if it was declared > as not const. > > I have attached the list of __no_const structs. Thanks. Architectures wouldn't matter for me, but an adjacent parsing problem or a strangely written type name could cause a problem. I will check your list. I would think though that the real problem wuld be things like the platform_driver structure, which to my recollection has one non-constant field, so the structure can't be const. julia -- To unsubscribe from this list: send the line "unsubscribe kernel-janitors" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html