Re: [PATCH v2 06/13] s390/cmf: Use kmalloc() for the CMB area

[Mike Rapoport <rppt@kernel.org>]

On Mon, Sep 14, 2026 at 07:59:52PM +0200, Heiko Carstens wrote:
> On Thu, Sep 10, 2026 at 02:00:08PM +0300, Mike Rapoport (Microsoft) wrote:
> > alloc_cmb() allocates the channel measurement block area shared by
> > devices using the basic channel measurement format.
> ...
> > The measurement block origin must be 32-byte aligned. Each CMB is
> > 32 bytes, so kmalloc() provides the required alignment without rounding
> > the allocation to a power-of-two number of pages. Reject an empty area
> > before allocating it.
> > 
> > Replace use of __get_free_pages() with kmalloc() and free_pages() with
> > kfree().
> ...
> >  		spin_unlock(&cmb_area.lock);
> > -		mem = (void *)__get_free_pages(GFP_KERNEL, get_order(size));
> > +		mem = kmalloc(size, GFP_KERNEL);
> >  		spin_lock(&cmb_area.lock);
> >  
> >  		if (cmb_area.mem) {
> >  			/* ok, another thread was faster */
> > -			free_pages((unsigned long)mem, get_order(size));
> > +			kfree(mem);
> >  		} else if (!mem) {
> >  			/* no luck */
> >  			ret = -ENOMEM;
> 
> This is the only patch which made me raise an eyebrow, since the allocated
> area is smaller. But it looks all good according to the architecture.
> Since the allocation changed anyway, it might also make sense to switch to
> kzalloc() and get rid of the subsequent memset() (not in the context of the
> above diff).

I suppose separate alloc and memset() are there to save a couple of cycles
when another thread is the first to allocate, no idea how important this is
though :)

I can send a patch on top if you'd like.

> But... I doubt it is worth it to send yet another version just for this.
> 
> The whole series looks good to me. Vineeth, Peter, any thoughts?

-- 
Sincerely yours,
Mike.