[PATCH 0/3] numa allocations for devm and pmem

[PATCH 0/3] numa allocations for devm and pmem

[Dan Williams]

I noticed that pmem should be using blk_alloc_queue_node() and then I
wondered about the devm allocations...  So, here is a quick conversion
of devm to use dev_to_node() for node local allocations by default and
as a result pmem to allocate all its driver infrastructure near to the
device.

---

Dan Williams (3):
      devm: make allocations numa aware by default
      devm_memremap_pages: use numa_mem_id
      pmem, memremap: convert to numa aware allocations


 drivers/base/devres.c  |   19 ++++++++++---------
 drivers/nvdimm/pmem.c  |    5 +++--
 include/linux/device.h |   16 ++++++++++++----
 kernel/memremap.c      |    9 +++++----
 4 files changed, 30 insertions(+), 19 deletions(-)

[PATCH 1/3] devm: make allocations numa aware by default

[Dan Williams]

Given we already have a device just use dev_to_node() to provide hint
allocations for devres.  However, current devres_alloc() users will need
to explicitly opt-in with devres_alloc_node().

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/base/devres.c  |   19 ++++++++++---------
 include/linux/device.h |   16 ++++++++++++----
 2 files changed, 22 insertions(+), 13 deletions(-)

diff --git a/drivers/base/devres.c b/drivers/base/devres.c
index 875464690117..8fc654f0807b 100644
--- a/drivers/base/devres.c
+++ b/drivers/base/devres.c
@@ -82,12 +82,12 @@ static struct devres_group * node_to_group(struct devres_node *node)
 }
 
 static __always_inline struct devres * alloc_dr(dr_release_t release,
-						size_t size, gfp_t gfp)
+						size_t size, gfp_t gfp, int nid)
 {
 	size_t tot_size = sizeof(struct devres) + size;
 	struct devres *dr;
 
-	dr = kmalloc_track_caller(tot_size, gfp);
+	dr = kmalloc_node_track_caller(tot_size, gfp, nid);
 	if (unlikely(!dr))
 		return NULL;
 
@@ -106,24 +106,25 @@ static void add_dr(struct device *dev, struct devres_node *node)
 }
 
 #ifdef CONFIG_DEBUG_DEVRES
-void * __devres_alloc(dr_release_t release, size_t size, gfp_t gfp,
+void * __devres_alloc_node(dr_release_t release, size_t size, gfp_t gfp, int nid,
 		      const char *name)
 {
 	struct devres *dr;
 
-	dr = alloc_dr(release, size, gfp | __GFP_ZERO);
+	dr = alloc_dr(release, size, gfp | __GFP_ZERO, nid);
 	if (unlikely(!dr))
 		return NULL;
 	set_node_dbginfo(&dr->node, name, size);
 	return dr->data;
 }
-EXPORT_SYMBOL_GPL(__devres_alloc);
+EXPORT_SYMBOL_GPL(__devres_alloc_node);
 #else
 /**
  * devres_alloc - Allocate device resource data
  * @release: Release function devres will be associated with
  * @size: Allocation size
  * @gfp: Allocation flags
+ * @nid: NUMA node
  *
  * Allocate devres of @size bytes.  The allocated area is zeroed, then
  * associated with @release.  The returned pointer can be passed to
@@ -132,16 +133,16 @@ EXPORT_SYMBOL_GPL(__devres_alloc);
  * RETURNS:
  * Pointer to allocated devres on success, NULL on failure.
  */
-void * devres_alloc(dr_release_t release, size_t size, gfp_t gfp)
+void * devres_alloc_node(dr_release_t release, size_t size, gfp_t gfp, int nid)
 {
 	struct devres *dr;
 
-	dr = alloc_dr(release, size, gfp | __GFP_ZERO);
+	dr = alloc_dr(release, size, gfp | __GFP_ZERO, nid);
 	if (unlikely(!dr))
 		return NULL;
 	return dr->data;
 }
-EXPORT_SYMBOL_GPL(devres_alloc);
+EXPORT_SYMBOL_GPL(devres_alloc_node);
 #endif
 
 /**
@@ -776,7 +777,7 @@ void * devm_kmalloc(struct device *dev, size_t size, gfp_t gfp)
 	struct devres *dr;
 
 	/* use raw alloc_dr for kmalloc caller tracing */
-	dr = alloc_dr(devm_kmalloc_release, size, gfp);
+	dr = alloc_dr(devm_kmalloc_release, size, gfp, dev_to_node(dev));
 	if (unlikely(!dr))
 		return NULL;
 
diff --git a/include/linux/device.h b/include/linux/device.h
index 5d7bc6349930..b8f411b57dcb 100644
--- a/include/linux/device.h
+++ b/include/linux/device.h
@@ -604,13 +604,21 @@ typedef void (*dr_release_t)(struct device *dev, void *res);
 typedef int (*dr_match_t)(struct device *dev, void *res, void *match_data);
 
 #ifdef CONFIG_DEBUG_DEVRES
-extern void *__devres_alloc(dr_release_t release, size_t size, gfp_t gfp,
-			     const char *name);
+extern void *__devres_alloc_node(dr_release_t release, size_t size, gfp_t gfp,
+				 int nid, const char *name);
 #define devres_alloc(release, size, gfp) \
-	__devres_alloc(release, size, gfp, #release)
+	__devres_alloc_node(release, size, gfp, NUMA_NO_NODE, #release)
+#define devres_alloc_node(release, size, gfp, nid) \
+	__devres_alloc_node(release, size, gfp, nid, #release)
 #else
-extern void *devres_alloc(dr_release_t release, size_t size, gfp_t gfp);
+extern void *devres_alloc_node(dr_release_t release, size_t size, gfp_t gfp,
+			       int nid);
+static inline void *devres_alloc(dr_release_t release, size_t size, gfp_t gfp)
+{
+	return devres_alloc_node(release, size, gfp, NUMA_NO_NODE);
+}
 #endif
+
 extern void devres_for_each_res(struct device *dev, dr_release_t release,
 				dr_match_t match, void *match_data,
 				void (*fn)(struct device *, void *, void *),

Re: [PATCH 1/3] devm: make allocations numa aware by default

[Dan Williams]

Hi Tejun, any thoughts on this, or concerns about globally enabling
numa hints for devm allocations?

On Fri, Sep 25, 2015 at 6:17 PM, Dan Williams <dan.j.williams@intel.com> wrote:
> Given we already have a device just use dev_to_node() to provide hint
> allocations for devres.  However, current devres_alloc() users will need
> to explicitly opt-in with devres_alloc_node().
>
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> ---
>  drivers/base/devres.c  |   19 ++++++++++---------
>  include/linux/device.h |   16 ++++++++++++----
>  2 files changed, 22 insertions(+), 13 deletions(-)
>
> diff --git a/drivers/base/devres.c b/drivers/base/devres.c
> index 875464690117..8fc654f0807b 100644
> --- a/drivers/base/devres.c
> +++ b/drivers/base/devres.c
> @@ -82,12 +82,12 @@ static struct devres_group * node_to_group(struct devres_node *node)
>  }
>
>  static __always_inline struct devres * alloc_dr(dr_release_t release,
> -                                               size_t size, gfp_t gfp)
> +                                               size_t size, gfp_t gfp, int nid)
>  {
>         size_t tot_size = sizeof(struct devres) + size;
>         struct devres *dr;
>
> -       dr = kmalloc_track_caller(tot_size, gfp);
> +       dr = kmalloc_node_track_caller(tot_size, gfp, nid);
>         if (unlikely(!dr))
>                 return NULL;
>
> @@ -106,24 +106,25 @@ static void add_dr(struct device *dev, struct devres_node *node)
>  }
>
>  #ifdef CONFIG_DEBUG_DEVRES
> -void * __devres_alloc(dr_release_t release, size_t size, gfp_t gfp,
> +void * __devres_alloc_node(dr_release_t release, size_t size, gfp_t gfp, int nid,
>                       const char *name)
>  {
>         struct devres *dr;
>
> -       dr = alloc_dr(release, size, gfp | __GFP_ZERO);
> +       dr = alloc_dr(release, size, gfp | __GFP_ZERO, nid);
>         if (unlikely(!dr))
>                 return NULL;
>         set_node_dbginfo(&dr->node, name, size);
>         return dr->data;
>  }
> -EXPORT_SYMBOL_GPL(__devres_alloc);
> +EXPORT_SYMBOL_GPL(__devres_alloc_node);
>  #else
>  /**
>   * devres_alloc - Allocate device resource data
>   * @release: Release function devres will be associated with
>   * @size: Allocation size
>   * @gfp: Allocation flags
> + * @nid: NUMA node
>   *
>   * Allocate devres of @size bytes.  The allocated area is zeroed, then
>   * associated with @release.  The returned pointer can be passed to
> @@ -132,16 +133,16 @@ EXPORT_SYMBOL_GPL(__devres_alloc);
>   * RETURNS:
>   * Pointer to allocated devres on success, NULL on failure.
>   */
> -void * devres_alloc(dr_release_t release, size_t size, gfp_t gfp)
> +void * devres_alloc_node(dr_release_t release, size_t size, gfp_t gfp, int nid)
>  {
>         struct devres *dr;
>
> -       dr = alloc_dr(release, size, gfp | __GFP_ZERO);
> +       dr = alloc_dr(release, size, gfp | __GFP_ZERO, nid);
>         if (unlikely(!dr))
>                 return NULL;
>         return dr->data;
>  }
> -EXPORT_SYMBOL_GPL(devres_alloc);
> +EXPORT_SYMBOL_GPL(devres_alloc_node);
>  #endif
>
>  /**
> @@ -776,7 +777,7 @@ void * devm_kmalloc(struct device *dev, size_t size, gfp_t gfp)
>         struct devres *dr;
>
>         /* use raw alloc_dr for kmalloc caller tracing */
> -       dr = alloc_dr(devm_kmalloc_release, size, gfp);
> +       dr = alloc_dr(devm_kmalloc_release, size, gfp, dev_to_node(dev));
>         if (unlikely(!dr))
>                 return NULL;
>
> diff --git a/include/linux/device.h b/include/linux/device.h
> index 5d7bc6349930..b8f411b57dcb 100644
> --- a/include/linux/device.h
> +++ b/include/linux/device.h
> @@ -604,13 +604,21 @@ typedef void (*dr_release_t)(struct device *dev, void *res);
>  typedef int (*dr_match_t)(struct device *dev, void *res, void *match_data);
>
>  #ifdef CONFIG_DEBUG_DEVRES
> -extern void *__devres_alloc(dr_release_t release, size_t size, gfp_t gfp,
> -                            const char *name);
> +extern void *__devres_alloc_node(dr_release_t release, size_t size, gfp_t gfp,
> +                                int nid, const char *name);
>  #define devres_alloc(release, size, gfp) \
> -       __devres_alloc(release, size, gfp, #release)
> +       __devres_alloc_node(release, size, gfp, NUMA_NO_NODE, #release)
> +#define devres_alloc_node(release, size, gfp, nid) \
> +       __devres_alloc_node(release, size, gfp, nid, #release)
>  #else
> -extern void *devres_alloc(dr_release_t release, size_t size, gfp_t gfp);
> +extern void *devres_alloc_node(dr_release_t release, size_t size, gfp_t gfp,
> +                              int nid);
> +static inline void *devres_alloc(dr_release_t release, size_t size, gfp_t gfp)
> +{
> +       return devres_alloc_node(release, size, gfp, NUMA_NO_NODE);
> +}
>  #endif
> +
>  extern void devres_for_each_res(struct device *dev, dr_release_t release,
>                                 dr_match_t match, void *match_data,
>                                 void (*fn)(struct device *, void *, void *),
>
> _______________________________________________
> Linux-nvdimm mailing list
> Linux-nvdimm@lists.01.org
> https://lists.01.org/mailman/listinfo/linux-nvdimm

Re: [PATCH 1/3] devm: make allocations numa aware by default

[Tejun Heo]

On Tue, Sep 29, 2015 at 05:30:26PM -0700, Dan Williams wrote:
> Hi Tejun, any thoughts on this, or concerns about globally enabling
> numa hints for devm allocations?

Looks good to me.  Please feel free to add

Reviewed-by: Tejun Heo <tj@kernel.org>

Thanks.

-- 
tejun

[PATCH 2/3] devm_memremap_pages: use numa_mem_id

[Dan Williams]

Hint to closest numa node for the placement of newly allocated pages.
As that is where the device's other allocations will originate by
default when it does not specify a NUMA node.

Cc: Christoph Hellwig <hch@lst.de>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 kernel/memremap.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/memremap.c b/kernel/memremap.c
index 72b0c66628b6..86a5cdeb18f2 100644
--- a/kernel/memremap.c
+++ b/kernel/memremap.c
@@ -175,7 +175,7 @@ void *devm_memremap_pages(struct device *dev, struct resource *res)
 
 	nid = dev_to_node(dev);
 	if (nid < 0)
-		nid = 0;
+		nid = numa_mem_id();
 
 	error = arch_add_memory(nid, res->start, resource_size(res), true);
 	if (error) {

[PATCH 3/3] pmem, memremap: convert to numa aware allocations

[Dan Williams]

Given that pmem ranges come with numa-locality hints, arrange for the
resulting driver objects to be obtained from node-local memory.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/nvdimm/pmem.c |    5 +++--
 kernel/memremap.c     |    7 ++++---
 2 files changed, 7 insertions(+), 5 deletions(-)

diff --git a/drivers/nvdimm/pmem.c b/drivers/nvdimm/pmem.c
index b9525385c0dc..570764a7599e 100644
--- a/drivers/nvdimm/pmem.c
+++ b/drivers/nvdimm/pmem.c
@@ -177,9 +177,10 @@ static void pmem_detach_disk(struct pmem_device *pmem)
 static int pmem_attach_disk(struct device *dev,
 		struct nd_namespace_common *ndns, struct pmem_device *pmem)
 {
+	int nid = dev_to_node(dev);
 	struct gendisk *disk;
 
-	pmem->pmem_queue = blk_alloc_queue(GFP_KERNEL);
+	pmem->pmem_queue = blk_alloc_queue_node(GFP_KERNEL, nid);
 	if (!pmem->pmem_queue)
 		return -ENOMEM;
 
@@ -189,7 +190,7 @@ static int pmem_attach_disk(struct device *dev,
 	blk_queue_bounce_limit(pmem->pmem_queue, BLK_BOUNCE_ANY);
 	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, pmem->pmem_queue);
 
-	disk = alloc_disk(0);
+	disk = alloc_disk_node(0, nid);
 	if (!disk) {
 		blk_cleanup_queue(pmem->pmem_queue);
 		return -ENOMEM;
diff --git a/kernel/memremap.c b/kernel/memremap.c
index 86a5cdeb18f2..f257c4e66e60 100644
--- a/kernel/memremap.c
+++ b/kernel/memremap.c
@@ -114,7 +114,8 @@ void *devm_memremap(struct device *dev, resource_size_t offset,
 {
 	void **ptr, *addr;
 
-	ptr = devres_alloc(devm_memremap_release, sizeof(*ptr), GFP_KERNEL);
+	ptr = devres_alloc_node(devm_memremap_release, sizeof(*ptr), GFP_KERNEL,
+			dev_to_node(dev));
 	if (!ptr)
 		return NULL;
 
@@ -166,8 +167,8 @@ void *devm_memremap_pages(struct device *dev, struct resource *res)
 	if (is_ram == REGION_INTERSECTS)
 		return __va(res->start);
 
-	page_map = devres_alloc(devm_memremap_pages_release,
-			sizeof(*page_map), GFP_KERNEL);
+	page_map = devres_alloc_node(devm_memremap_pages_release,
+			sizeof(*page_map), GFP_KERNEL, dev_to_node(dev));
 	if (!page_map)
 		return ERR_PTR(-ENOMEM);
 

Re: [PATCH 0/3] numa allocations for devm and pmem

[Christoph Hellwig]

The whole series looks fine to me:

Reviewed-by: Christoph Hellwig <hch@lst.de>