On Fri, 9 Mar 2018 17:08:06 -0500 Pavel Tatashin <pasha.tatas...@oracle.com>
wrote:
> Vlastimil Babka reported about a window issue during which when deferred
> pages are initialized, and the current version of on-demand initialization
> is finished, allocations may fail. While this is highly unlikely scenario,
> since this kind of allocation request must be large, and must come from
> interrupt handler, we still want to cover it.
>
> We solve this by initializing deferred pages with interrupts disabled, and
> holding node_size_lock spin lock while pages in the node are being
> initialized. The on-demand deferred page initialization that comes later
> will use the same lock, and thus synchronize with deferred_init_memmap().
>
> It is unlikely for threads that initialize deferred pages to be
> interrupted. They run soon after smp_init(), but before modules are
> initialized, and long before user space programs. This is why there is no
> adverse effect of having these threads running with interrupts disabled.
>
> ...
>
> --- a/include/linux/memory_hotplug.h
> +++ b/include/linux/memory_hotplug.h
>
> +#if defined(CONFIG_MEMORY_HOTPLUG) ||
> defined(CONFIG_DEFERRED_STRUCT_PAGE_INIT)
> +/*
> + * pgdat resizing functions
> + */
> +static inline
> +void pgdat_resize_lock(struct pglist_data *pgdat, unsigned long *flags)
> +{
> + spin_lock_irqsave(&pgdat->node_size_lock, *flags);
> +}
> +static inline
> +void pgdat_resize_unlock(struct pglist_data *pgdat, unsigned long *flags)
> +{
> + spin_unlock_irqrestore(&pgdat->node_size_lock, *flags);
> +}
> +static inline
> +void pgdat_resize_init(struct pglist_data *pgdat)
> +{
> + spin_lock_init(&pgdat->node_size_lock);
> +}
> +
> +/* Disable interrupts and save previous IRQ state in flags before locking */
> +static inline
> +void pgdat_resize_lock_irq(struct pglist_data *pgdat, unsigned long *flags)
> +{
> + unsigned long tmp_flags;
> +
> + local_irq_save(*flags);
> + local_irq_disable();
> + pgdat_resize_lock(pgdat, &tmp_flags);
> +}
As far as I can tell, this ugly-looking thing is identical to
pgdat_resize_lock().
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -1506,7 +1506,6 @@ static void __init deferred_free_pages(int nid, int
> zid, unsigned long pfn,
> } else if (!(pfn & nr_pgmask)) {
> deferred_free_range(pfn - nr_free, nr_free);
> nr_free = 1;
> - cond_resched();
> } else {
> nr_free++;
And how can we simply remove these cond_resched()s? I assume this is
being done because interrupts are now disabled? But those were there
for a reason, weren't they?
