On 11/13/19 2:06 PM, Dave Chinner wrote:
>> and in kmem.h:
>>
>> static inline void *
>> kmem_alloc_io(
>> size_t size,
>> gfp_t flags,
>> unsigned int align_mask)
>> {
>> trace_kmem_alloc_io(size, flags, align_mask, _RET_IP_);
>> return kmem_alloc(size, flags, align_mask);
>> }
> This should be able to go away soon, because the heap allocator will
> guarantee alignment soon. That means kmem.c is a single function,
> and kmem.h is a single function. I'd be looking to move the two
> helper functions into some other utility file at that point
> (fsops?)...
>
> ANother question: how much work is there to be done on the userspace
> side of things?
Right now userspace only uses kmem_alloc, with the 5.4 sync there is one
call to kmem_alloc_large which I just #defined to kmem_alloc in my patch
stack.
kmem_alloc_large is in libxfs so it'd just get fixed up if/when this patch is
merged across.
-Eric
