On Tue, Jan 16, 2018 at 09:21:30AM -0600, Christopher Lameter wrote:
> > struct kmem_cache_attr {
> > const char name[32];
>
> Want to avoid the string reference mess that occurred in the past?
> Is that really necessary? But it would limit the size of the name.
I think that's a good thing! /proc/slabinfo really starts to get grotty
above 16 bytes. I'd like to chop off "_cache" from the name of every
single slab! If ext4_allocation_context has to become ext4_alloc_ctx,
I don't think we're going to lose any valuable information.
My real intent was to reduce the number of allocations; if we can make
it not necessary to kstrdup the name, I think that'd be appreciated by
our CONFIG_TINY friends.
> > (my rationale is that everything in attr should be const, but size, align
> > and flags all get modified by the slab code).
>
> Thought about putting all the parameters into the kmem_cache_attr struct.
>
> So
>
> struct kmem_cache_attr {
> char *name;
> size_t size;
> size_t align;
> slab_flags_t flags;
> unsigned int useroffset;
> unsinged int usersize;
> void (*ctor)(void *);
> kmem_isolate_func *isolate;
> kmem_migrate_func *migrate;
> ...
> }
In these slightly-more-security-conscious days, it's considered poor
practice to have function pointers in writable memory. That was why
I wanted to make the kmem_cache_attr const.
Also, there's no need for 'size' and 'align' to be size_t. Slab should
never support allocations above 4GB in size. I'm not even keen on seeing
allocations above 64kB, but I see my laptop has six 512kB allocations (!),
three 256kB allocations and seven 128kB allocations, so I must reluctantly
concede that using an unsigned int is necessary. If I were really into
bitshaving, I might force all allocations to be a multiple of 32-bytes
in size, and then we could use 16 bits to represent an allocation between
32 and 2MB, but I think that tips us beyond the complexity boundary.
