Enhanced printk() merged
Enhanced printk() merged
Posted Jul 17, 2008 8:59 UTC (Thu) by meuh (guest, #22042)Parent article: Enhanced printk() merged
Strange way to specify format.
According to printf() manual page:
Each conversion specification is introduced by the '%' character [...] after which the following appear in sequence:- Zero or more flags
- [...]
- A conversion specifier character that indicates the type of conversion to be applied.
So, the correct way to specify a struct inode pointer should be %Ip, where I is the flag and p the conversion specifier.
Did kernel developers read userland manual pages ? :)
Posted Jul 17, 2008 9:56 UTC (Thu)
by nix (subscriber, #2304)
[Link] (2 responses)
Posted Jul 17, 2008 16:41 UTC (Thu)
by meuh (guest, #22042)
[Link] (1 responses)
Posted Jul 17, 2008 17:41 UTC (Thu)
by nix (subscriber, #2304)
[Link]
Enhanced printk() merged
If they used that format, GCC would warn about it. The format chosen looks
to GCC (as to userland printf()) like a %p with unrelated characters after
it, so GCC doesn't check those unrelated characters because they're just
literal text as far as it knows.
Enhanced printk() merged
Extending GCC to support some new kind of format string is also possible,
sadly this can't be done dynamically, patching GCC is required.
GCC already knows about other format string, see:
http://gcc.gnu.org/onlinedocs/gcc-4.3.0/gcc/Target-Format...
http://gcc.gnu.org/onlinedocs/gcc-4.3.0/gcc/Function-Attr...
Enhanced printk() merged
Yes indeed, and printk() is marked up with the printf attribute. The trick
is to find a way to define new format characters that doesn't cause GCC to
warn about all of them.
There were attempts in the past to make the format attributes dynamically
redefinable, but if I recall correctly the consensus in the end was that
this was simply too damn complicated.
(I wonder if what we need is loose versions of the format attribute's
archetypes, which warn about incorrect numbers of parameters and type
mismatches for format letters GCC knows about, but does not check format
letters that GCC doesn't know? As long as nobody tries to reimplement
something like .* which changes the number of arguments consumed, this
should work fine.)