mm: remove per-zone hashtable of bitlock waitqueues
The per-zone waitqueues exist because of a scalability issue with the
page waitqueues on some NUMA machines, but it turns out that they hurt
normal loads, and now with the vmalloced stacks they also end up
breaking gfs2 that uses a bit_wait on a stack object:
wait_on_bit(&gh->gh_iflags, HIF_WAIT, TASK_UNINTERRUPTIBLE)
where 'gh' can be a reference to the local variable 'mount_gh' on the
stack of fill_super().
The reason the per-zone hash table breaks for this case is that there is
no "zone" for virtual allocations, and trying to look up the physical
page to get at it will fail (with a BUG_ON()).
It turns out that I actually complained to the mm people about the
per-zone hash table for another reason just a month ago: the zone lookup
also hurts the regular use of "unlock_page()" a lot, because the zone
lookup ends up forcing several unnecessary cache misses and generates
horrible code.
As part of that earlier discussion, we had a much better solution for
the NUMA scalability issue - by just making the page lock have a
separate contention bit, the waitqueue doesn't even have to be looked at
for the normal case.
Peter Zijlstra already has a patch for that, but let's see if anybody
even notices. In the meantime, let's fix the actual gfs2 breakage by
simplifying the bitlock waitqueues and removing the per-zone issue.
Reported-by: Andreas Gruenbacher <agruenba@redhat.com>
Tested-by: Bob Peterson <rpeterso@redhat.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 2b3bf67..de7c6e4 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -4977,72 +4977,6 @@
}
/*
- * Helper functions to size the waitqueue hash table.
- * Essentially these want to choose hash table sizes sufficiently
- * large so that collisions trying to wait on pages are rare.
- * But in fact, the number of active page waitqueues on typical
- * systems is ridiculously low, less than 200. So this is even
- * conservative, even though it seems large.
- *
- * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
- * waitqueues, i.e. the size of the waitq table given the number of pages.
- */
-#define PAGES_PER_WAITQUEUE 256
-
-#ifndef CONFIG_MEMORY_HOTPLUG
-static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
-{
- unsigned long size = 1;
-
- pages /= PAGES_PER_WAITQUEUE;
-
- while (size < pages)
- size <<= 1;
-
- /*
- * Once we have dozens or even hundreds of threads sleeping
- * on IO we've got bigger problems than wait queue collision.
- * Limit the size of the wait table to a reasonable size.
- */
- size = min(size, 4096UL);
-
- return max(size, 4UL);
-}
-#else
-/*
- * A zone's size might be changed by hot-add, so it is not possible to determine
- * a suitable size for its wait_table. So we use the maximum size now.
- *
- * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
- *
- * i386 (preemption config) : 4096 x 16 = 64Kbyte.
- * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
- * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
- *
- * The maximum entries are prepared when a zone's memory is (512K + 256) pages
- * or more by the traditional way. (See above). It equals:
- *
- * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
- * ia64(16K page size) : = ( 8G + 4M)byte.
- * powerpc (64K page size) : = (32G +16M)byte.
- */
-static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
-{
- return 4096UL;
-}
-#endif
-
-/*
- * This is an integer logarithm so that shifts can be used later
- * to extract the more random high bits from the multiplicative
- * hash function before the remainder is taken.
- */
-static inline unsigned long wait_table_bits(unsigned long size)
-{
- return ffz(~size);
-}
-
-/*
* Initially all pages are reserved - free ones are freed
* up by free_all_bootmem() once the early boot process is
* done. Non-atomic initialization, single-pass.
@@ -5304,49 +5238,6 @@
alloc_percpu(struct per_cpu_nodestat);
}
-static noinline __ref
-int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
-{
- int i;
- size_t alloc_size;
-
- /*
- * The per-page waitqueue mechanism uses hashed waitqueues
- * per zone.
- */
- zone->wait_table_hash_nr_entries =
- wait_table_hash_nr_entries(zone_size_pages);
- zone->wait_table_bits =
- wait_table_bits(zone->wait_table_hash_nr_entries);
- alloc_size = zone->wait_table_hash_nr_entries
- * sizeof(wait_queue_head_t);
-
- if (!slab_is_available()) {
- zone->wait_table = (wait_queue_head_t *)
- memblock_virt_alloc_node_nopanic(
- alloc_size, zone->zone_pgdat->node_id);
- } else {
- /*
- * This case means that a zone whose size was 0 gets new memory
- * via memory hot-add.
- * But it may be the case that a new node was hot-added. In
- * this case vmalloc() will not be able to use this new node's
- * memory - this wait_table must be initialized to use this new
- * node itself as well.
- * To use this new node's memory, further consideration will be
- * necessary.
- */
- zone->wait_table = vmalloc(alloc_size);
- }
- if (!zone->wait_table)
- return -ENOMEM;
-
- for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
- init_waitqueue_head(zone->wait_table + i);
-
- return 0;
-}
-
static __meminit void zone_pcp_init(struct zone *zone)
{
/*
@@ -5367,10 +5258,7 @@
unsigned long size)
{
struct pglist_data *pgdat = zone->zone_pgdat;
- int ret;
- ret = zone_wait_table_init(zone, size);
- if (ret)
- return ret;
+
pgdat->nr_zones = zone_idx(zone) + 1;
zone->zone_start_pfn = zone_start_pfn;
@@ -5382,6 +5270,7 @@
zone_start_pfn, (zone_start_pfn + size));
zone_init_free_lists(zone);
+ zone->initialized = 1;
return 0;
}