On Mon, 2009-10-05 at 05:01 -0700, Eric W. Biederman wrote: > Joe Perches <joe at perches.com> writes: > > Added #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt > > Converted printk(KERN_<level> to pr_<level>( > > Added KERN_ERR to allocation failure message > I'm dense and I haven't seen the discussions. What > is the point of adding a prefix string where none exists > into a bunch of printks? Hi Eric. There have been a few messages on lkml with little comment, so I thought I'd get a few more by submitting some patches. There is a [0/21] message that you might have received directly. http://lkml.org/lkml/2009/10/4/198 I believe these are some of the +/-'s of each approach: (copy/pasted from an earlier response) Current: o Allows some messages to not have a prefix at all o Prefixes can vary inside a specific compilation unit Proposed: o Consistent, smaller source code, with no typos for instance: acpi/apic typos were found/fixed kernel/power had messages without PM: mce used "MCE: " and "mce: " prefixes o Compatible with KMSG_COMPONENT o All logging messages should have a prefix so it could be easier to grep/categorize logs o Future: - Doesn't require each compilation unit to #define pr_fmt - Smaller objects without duplicated prefixes - Extensible via some dynamic_debug like mechanism to hide or show modname/__func__/offset without significant overhead or any increase in object size (printk would emit the prefix via some insertion mechanism after "<level>")
