Andrew Morton wrote:
> On Wed, 20 Nov 2013 14:33:35 -0800 Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> > On Wed, Nov 20, 2013 at 9:47 AM, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
> > >
> > > BTW, something odd happened to mm/memory.c - either a mangled patch
> > > or a lost followup:
> > >
> > > commit ea1e7ed33708
> > > mm: create a separate slab for page->ptl allocation
> > >
> > > Fair enough, and yes, it does create that separate slab. The problem is,
> > > it's still using kmalloc/kfree for those beasts - page_ptl_cachep isn't
> > > used at all...
> >
> > Ok, it looks straightforward enough to just replace the kmalloc/kfree
> > with using a slab allocation using the page_ptl_cachep pointer. I'd do
> > it myself, but I would like to know how it got lost? Also, much
> > testing to make sure the cachep is initialized early enough.
>
> agh, I went through hell keeping that patch alive and it appears I lost
> some of it.
Actually, I've lost it while adding BLOATED_SPINLOCKS :(
> > Or should we just revert the commit that added the pointless/unused
> > slab pointer?
> >
> > Andrew, Kirill, comments?
>
> Let's just kill it please. We can try again for 3.14.
I'm okay with that.
Only side note: it's useful not only for debug case, but also for
PREEMPT_RT where spinlock_t is always bloated.
Fixed patch:
>From e624075b47caa2a15998225df7cec953d271b9ac Mon Sep 17 00:00:00 2001
From: "Kirill A. Shutemov" <kirill.shutemov@xxxxxxxxxxxxxxx>
Date: Thu, 14 Nov 2013 14:31:53 -0800
Subject: [PATCH] mm: create a separate slab for page->ptl allocation, try two
If DEBUG_SPINLOCK and DEBUG_LOCK_ALLOC are enabled spinlock_t on x86_64
is 72 bytes. For page->ptl they will be allocated from kmalloc-96 slab,
so we loose 24 on each. An average system can easily allocate few tens
thousands of page->ptl and overhead is significant.
Let's create a separate slab for page->ptl allocation to solve this.
To make sure that it really works this time, some numbers from my test
machine (just booted, no load):
Before:
# grep '^\(kmalloc-96\|page->ptl\)' /proc/slabinfo
kmalloc-96 31987 32190 128 30 1 : tunables 120 60 8 : slabdata 1073 1073 92
After:
# grep '^\(kmalloc-96\|page->ptl\)' /proc/slabinfo
page->ptl 27516 28143 72 53 1 : tunables 120 60 8 : slabdata 531 531 9
kmalloc-96 3853 5280 128 30 1 : tunables 120 60 8 : slabdata 176 176 0
Note that the patch is useful not only for debug case, but also for
PREEMPT_RT, where spinlock_t is always bloated.
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
---
include/linux/mm.h | 9 +++++++++
init/main.c | 2 +-
mm/memory.c | 11 +++++++++--
3 files changed, 19 insertions(+), 3 deletions(-)
diff --git a/include/linux/mm.h b/include/linux/mm.h
index 1cedd000cf29..0548eb201e05 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -1318,6 +1318,7 @@ static inline pmd_t *pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long a
#if USE_SPLIT_PTE_PTLOCKS
#if BLOATED_SPINLOCKS
+void __init ptlock_cache_init(void);
extern bool ptlock_alloc(struct page *page);
extern void ptlock_free(struct page *page);
@@ -1326,6 +1327,7 @@ static inline spinlock_t *ptlock_ptr(struct page *page)
return page->ptl;
}
#else /* BLOATED_SPINLOCKS */
+static inline void ptlock_cache_init(void) {}
static inline bool ptlock_alloc(struct page *page)
{
return true;
@@ -1378,10 +1380,17 @@ static inline spinlock_t *pte_lockptr(struct mm_struct *mm, pmd_t *pmd)
{
return &mm->page_table_lock;
}
+static inline void ptlock_cache_init(void) {}
static inline bool ptlock_init(struct page *page) { return true; }
static inline void pte_lock_deinit(struct page *page) {}
#endif /* USE_SPLIT_PTE_PTLOCKS */
+static inline void pgtable_init(void)
+{
+ ptlock_cache_init();
+ pgtable_cache_init();
+}
+
static inline bool pgtable_page_ctor(struct page *page)
{
inc_zone_page_state(page, NR_PAGETABLE);
diff --git a/init/main.c b/init/main.c
index febc511e078a..01573fdfa186 100644
--- a/init/main.c
+++ b/init/main.c
@@ -476,7 +476,7 @@ static void __init mm_init(void)
mem_init();
kmem_cache_init();
percpu_init_late();
- pgtable_cache_init();
+ pgtable_init();
vmalloc_init();
}
diff --git a/mm/memory.c b/mm/memory.c
index 5d9025f3b3e1..cf6098c10084 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -4272,11 +4272,18 @@ void copy_user_huge_page(struct page *dst, struct page *src,
#endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
#if USE_SPLIT_PTE_PTLOCKS && BLOATED_SPINLOCKS
+static struct kmem_cache *page_ptl_cachep;
+void __init ptlock_cache_init(void)
+{
+ page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0,
+ SLAB_PANIC, NULL);
+}
+
bool ptlock_alloc(struct page *page)
{
spinlock_t *ptl;
- ptl = kmalloc(sizeof(spinlock_t), GFP_KERNEL);
+ ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL);
if (!ptl)
return false;
page->ptl = ptl;
@@ -4285,6 +4292,6 @@ bool ptlock_alloc(struct page *page)
void ptlock_free(struct page *page)
{
- kfree(page->ptl);
+ kmem_cache_free(page_ptl_cachep, page->ptl);
}
#endif
--
Kirill A. Shutemov
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
