On Fri, Feb 01, 2019 at 02:03:46PM -0800, Andrew Morton wrote: > On Thu, 31 Jan 2019 18:43:10 -0800 Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote: > > > On Fri, Feb 01, 2019 at 11:42:42AM +1100, Tobin C. Harding wrote: > > > Currently when displaying /proc/slabinfo if any cache names are too long > > > then the output columns are not aligned. We could do something fancy to > > > get the maximum length of any cache name in the system or we could just > > > increase the hardcoded width. Currently it is 17 characters. Monitors > > > are wide these days so lets just increase it to 30 characters. > > > > I had a proposal some time ago to turn the slab name from being kmalloced > > to being an inline 16 bytes (with some fun hacks for cgroups). I think > > that's a better approach than permitting such long names. For example, > > ext4_allocation_context could be shortened to ext4_alloc_ctx without > > losing any expressivity. > > > > There are some back-compatibility concerns here. I'm don't understand sorry what back-compatibility concerns (please see sentiment at end of email :) > And truncating long names might result in duplicates. So I thought I had a good idea - add a pr_warn() if cache name > 16 and patch all current intree calls to kmem_cache_create() called as such. This process very kindly lead me to the fact that this does *not* work because of the macro KMEM_CACHE (which uses the struct name as the cache name). So, back to the drawing board. I'm concerned that this may be a waste of peoples time, if so please say so and I'll move on to something else. thanks, Tobin.
