On Wed, Nov 02 2016, Hannes Reinecke wrote: > On 11/02/2016 04:16 AM, NeilBrown wrote: >> This series removes all printk() calls from md code, preferring >> pr_warn(), pr_err() etc. >> >> All strings that were split over multiple lines are not joined >> back together because being able to search for a message is more >> important that not having long lines in the code. >> Lots of printk(KERN_DEBUG... are not pr_debug() which means the >> messages won't get printed unless they are explicitly disabled. >> >> I included some rough guidelines on which pr_* to choose in md.c. >> Many things became pr_debug(), most of the rest are pr_warn(). >> pr_err() and pr_crit() are used sparingly. >> >> I simplified some code in multipath.c too. >> >> A particular benefit of this is that the various "print_conf()" >> functions are not silent by default. >> On very large arrays (raid10 with hundreds of devices), these can >> be very noisy. >> > Any specify reason why you didn't move to use dev_*() style of errors? > Generally I prefer having the messages prefixed with the device which > caused them; making debugging _so_ much easier ... I was already a bit concerned about how much change was in the one set of patches: - rejoining large strings - changing printk to pr_* - revising log levels I wouldn't want to add more change. But as something to be added afterwards I support the idea in principle, but in practice..... What device would you pass to dev_*() ?? disk_to_dev(mddev->gendisk) ?? That wouldn't work when dm-raid is in the picture, as mddev->gendisk is NULL in that case. Is some cases we already include the md_name(). If there a specific places were it could be added, I would certainly support adding the dev name. I just don't think dev_*() is the way to do it for md. Thanks, NeilBrown
Attachment:
signature.asc
Description: PGP signature
