Re: [PATCH] mm: cma: support sysfs

[Minchan Kim <minchan@kernel.org>]

On Fri, Feb 05, 2021 at 02:55:26AM +0000, Matthew Wilcox wrote:
> On Wed, Feb 03, 2021 at 07:50:01AM -0800, Minchan Kim wrote:
> > +++ b/mm/Makefile
> > @@ -106,6 +106,7 @@ obj-$(CONFIG_ZSMALLOC)	+= zsmalloc.o
> >  obj-$(CONFIG_Z3FOLD)	+= z3fold.o
> >  obj-$(CONFIG_GENERIC_EARLY_IOREMAP) += early_ioremap.o
> >  obj-$(CONFIG_CMA)	+= cma.o
> > +obj-$(CONFIG_SYSFS)     += cma_sysfs.o
> 
> ehh ... if we have a kernel build with CMA=n, SYSFS=y, we'll get
> cma_sysfs built in with no cma to report on.

OMG. Let me fix it.

> 
> > +static ssize_t cma_alloc_attempt_show(struct kobject *kobj,
> > +			struct kobj_attribute *attr, char *buf)
> > +{
> > +	unsigned long val;
> > +	struct cma_stat *stat = container_of(kobj, struct cma_stat, kobj);
> > +
> > +	val = stat->alloc_attempt;
> > +
> > +	return sysfs_emit(buf, "%lu\n", val);
> 
> Why not more simply:
> 
> {
> 	struct cma_stat *stat = container_of(kobj, struct cma_stat, kobj);
> 	return sysfs_emit(buf, "%lu\n", stat->alloc_attempt);

It's a legacy when I used the lock there but removed finally.
Will follow your suggestion.

> }
> 
> > +	for (i = 0; i < cma_area_count; i++) {
> > +		cma = &cma_areas[i];
> > +		stat = kzalloc(sizeof(*stat), GFP_KERNEL);
> > +		if (!stat)
> > +			goto out;
> 
> How many cma areas are there going to be?  do we really want to allocate
> their stat individually?

I am not sure what could be in the end but at least, I have
5+ candidates (but could be shrink or extend) and yes,
want to keep track them individually.