Skip to contentSkip to conversation

xfs: fall back to vmalloc when allocation log vector buffers

PATCH In review Dave Chinner 2 participants 3 messages Feb 1, 2018 – Feb 5, 2018 XFS filesystem

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xfs: fall back to vmalloc when allocation log vector buffers

From: Dave Chinner <[email protected]>

When using large directory blocks, we regularly see memory allocations of >64k being made for the shadow log vector buffer. When we are under memory pressure, kmalloc() may not be able to find contiguous memory chunks large enough to satisfy these allocations easily, and if memory is fragmented we can potentially stall here.

TO avoid this problem, switch the log vector buffer allocation to use kmem_alloc_large(). This will allow failed allocations to fall back to vmalloc and so remove the dependency on large contiguous regions of memory being available. This should prevent slowdowns and potential stalls when memory is low and/or fragmented.

---
 fs/xfs/kmem.c        | 6 +++---
 fs/xfs/kmem.h        | 8 +++++++-
 fs/xfs/xfs_log_cil.c | 2 +-
 3 files changed, 11 insertions(+), 5 deletions(-)
fs/xfs/kmem.c +3 -3
index 393b6849aeb3..7bace03dc9dc 100644
@@ -46,13 +46,13 @@ kmem_alloc(size_t size, xfs_km_flags_t flags)
 }
 
 void *
-kmem_zalloc_large(size_t size, xfs_km_flags_t flags)
+kmem_alloc_large(size_t size, xfs_km_flags_t flags)
 {
 	unsigned nofs_flag = 0;
 	void	*ptr;
 	gfp_t	lflags;
 
-	ptr = kmem_zalloc(size, flags | KM_MAYFAIL);
+	ptr = kmem_alloc(size, flags | KM_MAYFAIL);
 	if (ptr)
 		return ptr;
 
@@ -67,7 +67,7 @@ kmem_zalloc_large(size_t size, xfs_km_flags_t flags)
 		nofs_flag = memalloc_nofs_save();
 
 	lflags = kmem_flags_convert(flags);
-	ptr = __vmalloc(size, lflags | __GFP_ZERO, PAGE_KERNEL);
+	ptr = __vmalloc(size, lflags, PAGE_KERNEL);
 
 	if (flags & KM_NOFS)
 		memalloc_nofs_restore(nofs_flag);
fs/xfs/kmem.h +7 -1
index 4b87472f35bc..6023b594ead7 100644
@@ -71,7 +71,7 @@ kmem_flags_convert(xfs_km_flags_t flags)
 }
 
 extern void *kmem_alloc(size_t, xfs_km_flags_t);
-extern void *kmem_zalloc_large(size_t size, xfs_km_flags_t);
+extern void *kmem_alloc_large(size_t size, xfs_km_flags_t);
 extern void *kmem_realloc(const void *, size_t, xfs_km_flags_t);
 static inline void  kmem_free(const void *ptr)
 {
@@ -85,6 +85,12 @@ kmem_zalloc(size_t size, xfs_km_flags_t flags)
 	return kmem_alloc(size, flags | KM_ZERO);
 }
 
+static inline void *
+kmem_zalloc_large(size_t size, xfs_km_flags_t flags)
+{
+	return kmem_alloc_large(size, flags | KM_ZERO);
+}
+
 /*
  * Zone interfaces
  */
fs/xfs/xfs_log_cil.c +1 -1
index 43aa42a3a5d3..61ab5c0a4c45 100644
@@ -202,7 +202,7 @@ xlog_cil_alloc_shadow_bufs(
 			 */
 			kmem_free(lip->li_lv_shadow);
 
-			lv = kmem_alloc(buf_size, KM_SLEEP|KM_NOFS);
+			lv = kmem_alloc_large(buf_size, KM_SLEEP|KM_NOFS);
 			memset(lv, 0, xlog_cil_iovec_space(niovecs));
 
 			lv->lv_item = lip;
Signature · 1 line
2.15.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 81 lines · Dave Chinner

From: Dave Chinner <[email protected]>

When using large directory blocks, we regularly see memory allocations of >64k being made for the shadow log vector buffer. When we are under memory pressure, kmalloc() may not be able to find contiguous memory chunks large enough to satisfy these allocations easily, and if memory is fragmented we can potentially stall here.

TO avoid this problem, switch the log vector buffer allocation to use kmem_alloc_large(). This will allow failed allocations to fall back to vmalloc and so remove the dependency on large contiguous regions of memory being available. This should prevent slowdowns and potential stalls when memory is low and/or fragmented.

---
 fs/xfs/kmem.c        | 6 +++---
 fs/xfs/kmem.h        | 8 +++++++-
 fs/xfs/xfs_log_cil.c | 2 +-
 3 files changed, 11 insertions(+), 5 deletions(-)
fs/xfs/kmem.c +3 -3
index 393b6849aeb3..7bace03dc9dc 100644
@@ -46,13 +46,13 @@ kmem_alloc(size_t size, xfs_km_flags_t flags)
 }
 
 void *
-kmem_zalloc_large(size_t size, xfs_km_flags_t flags)
+kmem_alloc_large(size_t size, xfs_km_flags_t flags)
 {
 	unsigned nofs_flag = 0;
 	void	*ptr;
 	gfp_t	lflags;
 
-	ptr = kmem_zalloc(size, flags | KM_MAYFAIL);
+	ptr = kmem_alloc(size, flags | KM_MAYFAIL);
 	if (ptr)
 		return ptr;
 
@@ -67,7 +67,7 @@ kmem_zalloc_large(size_t size, xfs_km_flags_t flags)
 		nofs_flag = memalloc_nofs_save();
 
 	lflags = kmem_flags_convert(flags);
-	ptr = __vmalloc(size, lflags | __GFP_ZERO, PAGE_KERNEL);
+	ptr = __vmalloc(size, lflags, PAGE_KERNEL);
 
 	if (flags & KM_NOFS)
 		memalloc_nofs_restore(nofs_flag);
fs/xfs/kmem.h +7 -1
index 4b87472f35bc..6023b594ead7 100644
@@ -71,7 +71,7 @@ kmem_flags_convert(xfs_km_flags_t flags)
 }
 
 extern void *kmem_alloc(size_t, xfs_km_flags_t);
-extern void *kmem_zalloc_large(size_t size, xfs_km_flags_t);
+extern void *kmem_alloc_large(size_t size, xfs_km_flags_t);
 extern void *kmem_realloc(const void *, size_t, xfs_km_flags_t);
 static inline void  kmem_free(const void *ptr)
 {
@@ -85,6 +85,12 @@ kmem_zalloc(size_t size, xfs_km_flags_t flags)
 	return kmem_alloc(size, flags | KM_ZERO);
 }
 
+static inline void *
+kmem_zalloc_large(size_t size, xfs_km_flags_t flags)
+{
+	return kmem_alloc_large(size, flags | KM_ZERO);
+}
+
 /*
  * Zone interfaces
  */
fs/xfs/xfs_log_cil.c +0 -0
index 43aa42a3a5d3..61ab5c0a4c45 100644

fs/xfs/xfs_log_cil.c xlog_cil_alloc_shadow_bufs( · line 205

Quoted 5 lines
fs/xfs/xfs_log_cil.c +1 -1
@@ -202,7 +202,7 @@ xlog_cil_alloc_shadow_bufs(
 			 */
 			kmem_free(lip->li_lv_shadow);
 
-			lv = kmem_alloc(buf_size, KM_SLEEP|KM_NOFS);
+			lv = kmem_alloc_large(buf_size, KM_SLEEP|KM_NOFS);

"KM_SLEEP | KM_NOFS", will fix it on the way in.

I wonder about expanding the uses of vmalloc space, but otoh stalling the log seems like a worse idea...

Looks ok enough to give it a spin,

--D

Quoted 10 lines
fs/xfs/xfs_log_cil.c +0 -0
 			memset(lv, 0, xlog_cil_iovec_space(niovecs));
 
 			lv->lv_item = lip;
Signature · 6 lines
2.15.1

--
To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#2 lore raw
Quoted 16 lines · Darrick J. Wong
Quoted 15 lines · Dave Chinner

From: Dave Chinner <[email protected]>

When using large directory blocks, we regularly see memory allocations of >64k being made for the shadow log vector buffer. When we are under memory pressure, kmalloc() may not be able to find contiguous memory chunks large enough to satisfy these allocations easily, and if memory is fragmented we can potentially stall here.

TO avoid this problem, switch the log vector buffer allocation to use kmem_alloc_large(). This will allow failed allocations to fall back to vmalloc and so remove the dependency on large contiguous regions of memory being available. This should prevent slowdowns and potential stalls when memory is low and/or fragmented.

.....

Quoted 8 lines
Quoted 8 lines
fs/xfs/xfs_log_cil.c +0 -0
index 43aa42a3a5d3..61ab5c0a4c45 100644
@@ -202,7 +202,7 @@ xlog_cil_alloc_shadow_bufs(
 			 */
 			kmem_free(lip->li_lv_shadow);
 
Quoted 4 lines

fs/xfs/xfs_log_cil.c xlog_cil_alloc_shadow_bufs( · line 205

Quoted 2 lines
fs/xfs/xfs_log_cil.c +1 -1
-			lv = kmem_alloc(buf_size, KM_SLEEP|KM_NOFS);
+			lv = kmem_alloc_large(buf_size, KM_SLEEP|KM_NOFS);

"KM_SLEEP | KM_NOFS", will fix it on the way in.

*nod*

Quoted 2 lines

I wonder about expanding the uses of vmalloc space, but otoh stalling the log seems like a worse idea...

We'll only see this happen when memory is really low and/or fragmented, so i don't think it's a problem at all. I could trigger warnings without this patch - they disappear with this patch and the system seemed more responsive under equivalent loading. That's entirely subjective, but it's so hard to consistently drive the system into the severe memory shortages and reclaim failures that trigger this problem that's the best I was able to do...

Quoted 2 lines

Looks ok enough to give it a spin,

Thanks!

-Dave.

Signature · 2 lines
Dave Chinner
[email protected]