On Sun, 30 Aug 2009, Jeff Smith wrote:
A patch follows. It corrects a typo in the comment, changes the function
name "implement" to "set_into_le_bitstream", changes the parameter name
"n" to "bitfield_size", printk's unsigned using %u not %d, and only does
a 32-bit read-modify-write when a 64-bit one is not necessary.
A patch that writes only the bytes that need to be changed, rather than
32 or 64-bit quantities that potentially access irrelevant memory
locations -- and that therefore require more complicated verification
logic -- would be a better approach. However, as we are at -rc8 already
and I don't fully understand the structures that are being changed, or
the reasons the code is as it is, I didn't feel confident about
presenting such a restructuring here.
Hi Jeff,
thanks for the patch, it looks correct on a quick glance. I would however
prefer the other approach you proposed.
We probably don't have to be nervous about being currently at -rc8, as
this will probably be fixed only in .32-rc1 anyway (as the unitialized
data is not used in any way, and therefore it doesn't require emergency
fix just to silence the kmemcheck warning).
Thanks,
--
Jiri Kosina
SUSE Labs