diff mbox series

[v3,3/3] iommu/iova: use named kmem_cache for iova magazines

Message ID dc5c51aaba50906a92b9ba1a5137ed462484a7be.1707144953.git.robin.murphy@arm.com (mailing list archive)
State New
Headers show
Series iommu/iova: use named kmem_cache for iova magazines | expand

Commit Message

Robin Murphy Feb. 5, 2024, 3:32 p.m. UTC 
From: Pasha Tatashin <pasha.tatashin@soleen.com>

The magazine buffers can take gigabytes of kmem memory, dominating all
other allocations. For observability purpose create named slab cache so
the iova magazine memory overhead can be clearly observed.

With this change:

> slabtop -o | head
 Active / Total Objects (% used)    : 869731 / 952904 (91.3%)
 Active / Total Slabs (% used)      : 103411 / 103974 (99.5%)
 Active / Total Caches (% used)     : 135 / 211 (64.0%)
 Active / Total Size (% used)       : 395389.68K / 411430.20K (96.1%)
 Minimum / Average / Maximum Object : 0.02K / 0.43K / 8.00K

OBJS ACTIVE  USE OBJ SIZE  SLABS OBJ/SLAB CACHE SIZE NAME
244412 244239 99%    1.00K  61103       4    244412K iommu_iova_magazine
 91636  88343 96%    0.03K    739     124      2956K kmalloc-32
 75744  74844 98%    0.12K   2367      32      9468K kernfs_node_cache

On this machine it is now clear that magazine use 242M of kmem memory.

Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Pasha Tatashin <pasha.tatashin@soleen.com>
[ rm: adjust to rework of iova_cache_{get,put} ]
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
---
 drivers/iommu/iova.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

Comments

John Garry Feb. 6, 2024, 11:24 a.m. UTC | #1 
On 05/02/2024 15:32, Robin Murphy wrote:
> From: Pasha Tatashin <pasha.tatashin@soleen.com>
> 
> The magazine buffers can take gigabytes of kmem memory, dominating all
> other allocations. For observability purpose create named slab cache so
> the iova magazine memory overhead can be clearly observed.
> 
> With this change:
> 
>> slabtop -o | head
>   Active / Total Objects (% used)    : 869731 / 952904 (91.3%)
>   Active / Total Slabs (% used)      : 103411 / 103974 (99.5%)
>   Active / Total Caches (% used)     : 135 / 211 (64.0%)
>   Active / Total Size (% used)       : 395389.68K / 411430.20K (96.1%)
>   Minimum / Average / Maximum Object : 0.02K / 0.43K / 8.00K
> 
> OBJS ACTIVE  USE OBJ SIZE  SLABS OBJ/SLAB CACHE SIZE NAME
> 244412 244239 99%    1.00K  61103       4    244412K iommu_iova_magazine
>   91636  88343 96%    0.03K    739     124      2956K kmalloc-32
>   75744  74844 98%    0.12K   2367      32      9468K kernfs_node_cache
> 
> On this machine it is now clear that magazine use 242M of kmem memory.

Those caches could do with a trimming ...

> 
> Acked-by: David Rientjes <rientjes@google.com>
> Signed-off-by: Pasha Tatashin <pasha.tatashin@soleen.com>
> [ rm: adjust to rework of iova_cache_{get,put} ]
> Signed-off-by: Robin Murphy <robin.murphy@arm.com>
> ---

FWIW:
Reviewed-by: John Garry <john.g.garry@oracle.com>
Robin Murphy Feb. 6, 2024, 11:32 a.m. UTC | #2 
On 06/02/2024 11:24 am, John Garry wrote:
> On 05/02/2024 15:32, Robin Murphy wrote:
>> From: Pasha Tatashin <pasha.tatashin@soleen.com>
>>
>> The magazine buffers can take gigabytes of kmem memory, dominating all
>> other allocations. For observability purpose create named slab cache so
>> the iova magazine memory overhead can be clearly observed.
>>
>> With this change:
>>
>>> slabtop -o | head
>>   Active / Total Objects (% used)    : 869731 / 952904 (91.3%)
>>   Active / Total Slabs (% used)      : 103411 / 103974 (99.5%)
>>   Active / Total Caches (% used)     : 135 / 211 (64.0%)
>>   Active / Total Size (% used)       : 395389.68K / 411430.20K (96.1%)
>>   Minimum / Average / Maximum Object : 0.02K / 0.43K / 8.00K
>>
>> OBJS ACTIVE  USE OBJ SIZE  SLABS OBJ/SLAB CACHE SIZE NAME
>> 244412 244239 99%    1.00K  61103       4    244412K iommu_iova_magazine
>>   91636  88343 96%    0.03K    739     124      2956K kmalloc-32
>>   75744  74844 98%    0.12K   2367      32      9468K kernfs_node_cache
>>
>> On this machine it is now clear that magazine use 242M of kmem memory.
> 
> Those caches could do with a trimming ...

See the discussion on v1 for more details:

https://lore.kernel.org/linux-iommu/20240201193014.2785570-1-tatashin@google.com/

but it seems this really is the idle baseline for lots of CPUs * lots of 
domains - if all those devices get going in anger it's likely that 
combined iova + iova_magazine usage truly will blow up into gigabytes.

Cheers,
Robin.

> 
>>
>> Acked-by: David Rientjes <rientjes@google.com>
>> Signed-off-by: Pasha Tatashin <pasha.tatashin@soleen.com>
>> [ rm: adjust to rework of iova_cache_{get,put} ]
>> Signed-off-by: Robin Murphy <robin.murphy@arm.com>
>> ---
> 
> FWIW:
> Reviewed-by: John Garry <john.g.garry@oracle.com>
Jerry Snitselaar Feb. 7, 2024, 4:43 p.m. UTC | #3 
On Mon, Feb 05, 2024 at 03:32:41PM +0000, Robin Murphy wrote:
> From: Pasha Tatashin <pasha.tatashin@soleen.com>
> 
> The magazine buffers can take gigabytes of kmem memory, dominating all
> other allocations. For observability purpose create named slab cache so
> the iova magazine memory overhead can be clearly observed.
> 
> With this change:
> 
> > slabtop -o | head
>  Active / Total Objects (% used)    : 869731 / 952904 (91.3%)
>  Active / Total Slabs (% used)      : 103411 / 103974 (99.5%)
>  Active / Total Caches (% used)     : 135 / 211 (64.0%)
>  Active / Total Size (% used)       : 395389.68K / 411430.20K (96.1%)
>  Minimum / Average / Maximum Object : 0.02K / 0.43K / 8.00K
> 
> OBJS ACTIVE  USE OBJ SIZE  SLABS OBJ/SLAB CACHE SIZE NAME
> 244412 244239 99%    1.00K  61103       4    244412K iommu_iova_magazine
>  91636  88343 96%    0.03K    739     124      2956K kmalloc-32
>  75744  74844 98%    0.12K   2367      32      9468K kernfs_node_cache
> 
> On this machine it is now clear that magazine use 242M of kmem memory.
> 
> Acked-by: David Rientjes <rientjes@google.com>
> Signed-off-by: Pasha Tatashin <pasha.tatashin@soleen.com>
> [ rm: adjust to rework of iova_cache_{get,put} ]
> Signed-off-by: Robin Murphy <robin.murphy@arm.com>
> ---

Reviewed-by: Jerry Snitselaar <jsnitsel@redhat.com>
diff mbox series

Patch

diff --git a/drivers/iommu/iova.c b/drivers/iommu/iova.c
index b5de865ee50b..d59d0ea2fd21 100644
--- a/drivers/iommu/iova.c
+++ b/drivers/iommu/iova.c
@@ -590,6 +590,8 @@  struct iova_rcache {
 	struct delayed_work work;
 };
 
+static struct kmem_cache *iova_magazine_cache;
+
 unsigned long iova_rcache_range(void)
 {
 	return PAGE_SIZE << (IOVA_RANGE_CACHE_MAX_SIZE - 1);
@@ -599,7 +601,7 @@  static struct iova_magazine *iova_magazine_alloc(gfp_t flags)
 {
 	struct iova_magazine *mag;
 
-	mag = kmalloc(sizeof(*mag), flags);
+	mag = kmem_cache_alloc(iova_magazine_cache, flags);
 	if (mag)
 		mag->size = 0;
 
@@ -608,7 +610,7 @@  static struct iova_magazine *iova_magazine_alloc(gfp_t flags)
 
 static void iova_magazine_free(struct iova_magazine *mag)
 {
-	kfree(mag);
+	kmem_cache_free(iova_magazine_cache, mag);
 }
 
 static void
@@ -953,6 +955,12 @@  int iova_cache_get(void)
 		if (!iova_cache)
 			goto out_err;
 
+		iova_magazine_cache = kmem_cache_create("iommu_iova_magazine",
+							sizeof(struct iova_magazine),
+							0, SLAB_HWCACHE_ALIGN, NULL);
+		if (!iova_magazine_cache)
+			goto out_err;
+
 		err = cpuhp_setup_state_multi(CPUHP_IOMMU_IOVA_DEAD, "iommu/iova:dead",
 					      NULL, iova_cpuhp_dead);
 		if (err) {
@@ -968,6 +976,7 @@  int iova_cache_get(void)
 
 out_err:
 	kmem_cache_destroy(iova_cache);
+	kmem_cache_destroy(iova_magazine_cache);
 	mutex_unlock(&iova_cache_mutex);
 	return err;
 }
@@ -984,6 +993,7 @@  void iova_cache_put(void)
 	if (!iova_cache_users) {
 		cpuhp_remove_multi_state(CPUHP_IOMMU_IOVA_DEAD);
 		kmem_cache_destroy(iova_cache);
+		kmem_cache_destroy(iova_magazine_cache);
 	}
 	mutex_unlock(&iova_cache_mutex);
 }