On Fri, Jan 22, 2021 at 12:17:33PM +0100, Sebastian Andrzej Siewior wrote:
> On 2021-01-20 13:54:03 [-0800], Paul E. McKenney wrote:
> > > > +// Record ptr in a page managed by krcp, with the
> > > > pre-krc_this_cpu_lock()
> > > > +// state specified by flags. If can_alloc is true, the caller must
> > > > +// be schedulable and not be holding any locks or mutexes that might be
> > > > +// acquired by the memory allocator or anything that it might invoke.
> > > > +// Returns true if ptr was successfully recorded, else the caller must
> > > > +// use a fallback.
> > >
> > > The whole RCU department is getting swamped by the // comments. Can't we
> > > have proper kernel doc and /* */ style comments like the remaining part
> > > of the kernel?
> >
> > Because // comments are easier to type and take up less horizontal space.
>
> As for the typing I could try to sell you
> ab // /*
>
> for your .vimrc and then //<enter> would become /* ;) As for the
> horizontal space, I don't have currently anything in my shop. I'm sorry.
;-)
> > Also, this kvfree_call_rcu_add_ptr_to_bulk() function is local to
> > kvfree_rcu(), and we don't normally docbook-ify such functions.
>
> I didn't mean to promote using docbook to use every. For instance if you
> look at kernel/trace/trace.c, there are no // comments around, just /*
> style, even for things like tracing_selftest_running.
>
> Basically I was curious if I could learn where this // is coming and if
> I could stop it.
Because they are now allowed and because they make my life easier as
noted above. Also in-function comment blocks are either one line or two
lines shorter. Yeah, they look strange at first, but it is not that hard
to get used to them. After all, I did manage to get used to the /* */
comment style shortly after first encountering it. ;-)
> > > > static inline bool
> > > > -kvfree_call_rcu_add_ptr_to_bulk(struct kfree_rcu_cpu *krcp, void *ptr)
> > > > +add_ptr_to_bulk_krc_lock(struct kfree_rcu_cpu **krcp,
> > > > + unsigned long *flags, void *ptr, bool can_alloc)
> > > > {
> > > > struct kvfree_rcu_bulk_data *bnode;
> > > > int idx;
> > > >
> > > > - if (unlikely(!krcp->initialized))
> > > > + *krcp = krc_this_cpu_lock(flags);
> > > > + if (unlikely(!(*krcp)->initialized))
> > > > return false;
> > > >
> > > > - lockdep_assert_held(&krcp->lock);
> > > > idx = !!is_vmalloc_addr(ptr);
> > > >
> > > > /* Check if a new block is required. */
> > > > - if (!krcp->bkvhead[idx] ||
> > > > - krcp->bkvhead[idx]->nr_records ==
> > > > KVFREE_BULK_MAX_ENTR) {
> > > > - bnode = get_cached_bnode(krcp);
> > > > - /* Switch to emergency path. */
> > > > + if (!(*krcp)->bkvhead[idx] ||
> > > > + (*krcp)->bkvhead[idx]->nr_records ==
> > > > KVFREE_BULK_MAX_ENTR) {
> > > > + bnode = get_cached_bnode(*krcp);
> > > > + if (!bnode && can_alloc) {
> > > > + krc_this_cpu_unlock(*krcp, *flags);
> > > > + bnode = (struct kvfree_rcu_bulk_data *)
> > >
> > > There is no need for this cast.
> >
> > Without it, gcc version 7.5.0 says:
> >
> > warning: assignment makes pointer from integer without a cast
> >
>
> I'm sorry. I forgot the part where __get_free_page() does not return
> (void *).
> But maybe it should given that free_pages() casts that long back to
> (void *) and __get_free_pages() -> page_address() returns (void *)
> which is then casted long.
No argument here. Then again, I am not the one in need of convincing.
There are use cases like this from pte_alloc_one_kernel():
unsigned long page = __get_free_page(GFP_DMA);
But a quick look indicates that they are in the minority.
> > > > + __get_free_page(GFP_KERNEL |
> > > > __GFP_RETRY_MAYFAIL | __GFP_NOWARN);
> > > > + *krcp = krc_this_cpu_lock(flags);
> > >
> > > so if bnode is NULL you could retry get_cached_bnode() since it might
> > > have been filled (given preemption or CPU migration changed something).
> > > Judging from patch #3 you think that a CPU migration is a bad thing. But
> > > why?
> >
> > So that the later "(*krcp)->bkvhead[idx] = bnode" assignment associates
> > it with the correct CPU.
> >
> > Though now that you mention it, couldn't the following happen?
> >
> > o Task A on CPU 0 notices that allocation is needed, so it
> > drops the lock disables migration, and sleeps while
> > allocating.
> >
> > o Task B on CPU 0 does the same.
> >
> > o The two tasks wake up in some order, and the second one
> > causes trouble at the "(*krcp)->bkvhead[idx] = bnode"
> > assignment.
>
> Yes it could, good point.
> I would really recommend using migrate_disable() at a minimum and only
> if it is really needed. It is more expensive than preempt_disable() and
> it isn't exactly good in terms of scheduling since the task is run able
> but restricted to a specific CPU.
> If it is unavoidable it is unavoidable but in this case I wouldn't use
> migrate_disable() but re-evaluate the situation after the allocation.
I could imagine the following alternatives:
o Acquire the old CPU's lock despite having been migrated.
If the above race happened, put the extra page in the
per-CPU cache. As Uladzislau notes, this would require
some sort of periodic cleanup that would be good to avoid.
o As now, which can result in an unfilled page, though only
in an uncommon race condition. (Uladzislau convinced me
that this was a good approach some months ago, and the
fact that he cannot make it happen easily does add some
weight to his argument.)
o Use migrate_disable().
Other ideas?
Thanx, Paul
> > Uladzislau, do we need to recheck "!(*krcp)->bkvhead[idx]" just after
> > the migrate_enable()? Along with the KVFREE_BULK_MAX_ENTR check?
> >
> > Thanx, Paul
>
> Sebastian
