Re: [PATCH v2 1/4] kernfs: kvmalloc xattr value instead of kmalloc

[Joe Perches <joe@perches.com>]

On Mon, 2020-03-09 at 11:21 -0700, Daniel Xu wrote:
> Hi Joe,

Hello Daniel.

> On Fri Mar 6, 2020 at 12:49 AM, Joe Perches wrote:
> > On Thu, 2020-03-05 at 13:16 -0800, Daniel Xu wrote:
> > > It's not really necessary to have contiguous physical memory for xattr
> > > values. We no longer need to worry about higher order allocations
> > > failing with kvmalloc, especially because the xattr size limit is at
> > > 64K.
> > 
> > So why use vmalloc memory at all?
> > 
> > 
> > > diff --git a/fs/xattr.c b/fs/xattr.c
> > ']
> > > @@ -817,7 +817,7 @@ struct simple_xattr *simple_xattr_alloc(const void *value, size_t size)
> > >  	if (len < sizeof(*new_xattr))
> > >  		return NULL;
> > >  
> > > -	new_xattr = kmalloc(len, GFP_KERNEL);
> > > +	new_xattr = kvmalloc(len, GFP_KERNEL);
> > 
> > Why is this sensible?
> > vmalloc memory is a much more limited resource.
> 
> What would be the alternative? As Greg said, contiguous memory should be
> more scarce.

If the need is to allocate from a single block of memory,
perhaps you need a submemory allocator like gen_pool.
(gennalloc.h)

Dunno.  Maybe i just don't quite understand your need.