[PATCH v4 0/2] mm: use memmap_on_memory semantics for dax/kmem

[PATCH v4 0/2] mm: use memmap_on_memory semantics for dax/kmem

[Vishal Verma]

The dax/kmem driver can potentially hot-add large amounts of memory
originating from CXL memory expanders, or NVDIMMs, or other 'device
memories'. There is a chance there isn't enough regular system memory
available to fit the memmap for this new memory. It's therefore
desirable, if all other conditions are met, for the kmem managed memory
to place its memmap on the newly added memory itself.

The main hurdle for accomplishing this for kmem is that memmap_on_memory
can only be done if the memory being added is equal to the size of one
memblock. To overcome this, allow the hotplug code to split an add_memory()
request into memblock-sized chunks, and try_remove_memory() to also
expect and handle such a scenario.

Patch 1 teaches the memory_hotplug code to allow for splitting
add_memory() and remove_memory() requests over memblock sized chunks.

Patch 2 adds a sysfs control for the kmem driver that would
allow an opt-out of using memmap_on_memory for the memory being added.

Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
---
Changes in v4:
- Rebase to Aneesh's PPC64 memmap_on_memory series v8 [2].
- Tweak a goto / error path in add_memory_create_devices() (Jonathan)
- Retain the old behavior for dax devices, only default to
  memmap_on_memory for CXL (Jonathan)
- Link to v3: https://lore.kernel.org/r/20230801-vv-kmem_memmap-v3-0-406e9aaf5689@intel.com

[2]: https://lore.kernel.org/linux-mm/20230808091501.287660-1-aneesh.kumar@linux.ibm.com

Changes in v3:
- Rebase on Aneesh's patches [1]
- Drop Patch 1 - it is not needed since [1] allows for dynamic setting
  of the memmap_on_memory param (David)
- Link to v2: https://lore.kernel.org/r/20230720-vv-kmem_memmap-v2-0-88bdaab34993@intel.com

[1]: https://lore.kernel.org/r/20230801044116.10674-1-aneesh.kumar@linux.ibm.com

Changes in v2:
- Drop the patch to create an override path for the memmap_on_memory
  module param (David)
- Move the chunking into memory_hotplug.c so that any caller of
  add_memory() can request this behavior. (David)
- Handle remove_memory() too. (David, Ying)
- Add a sysfs control in the kmem driver for memmap_on_memory semantics
  (David, Jonathan)
- Add a #else case to define mhp_supports_memmap_on_memory() if
  CONFIG_MEMORY_HOTPLUG is unset. (0day report)
- Link to v1: https://lore.kernel.org/r/20230613-vv-kmem_memmap-v1-0-f6de9c6af2c6@intel.com

---
Vishal Verma (2):
      mm/memory_hotplug: split memmap_on_memory requests across memblocks
      dax/kmem: allow kmem to add memory with memmap_on_memory

 drivers/dax/bus.h         |   1 +
 drivers/dax/dax-private.h |   1 +
 drivers/dax/bus.c         |  38 +++++++++++
 drivers/dax/cxl.c         |   1 +
 drivers/dax/hmem/hmem.c   |   1 +
 drivers/dax/kmem.c        |   8 ++-
 drivers/dax/pmem.c        |   1 +
 mm/memory_hotplug.c       | 165 ++++++++++++++++++++++++++++------------------
 8 files changed, 150 insertions(+), 66 deletions(-)
---
base-commit: 25b5b1a0646c3d39e1d885e27c10be1c9e202bf2
change-id: 20230613-vv-kmem_memmap-5483c8d04279

Best regards,
-- 
Vishal Verma <vishal.l.verma@intel.com>

[PATCH v4 1/2] mm/memory_hotplug: split memmap_on_memory requests across memblocks

[Vishal Verma]

The MHP_MEMMAP_ON_MEMORY flag for hotplugged memory is restricted to
'memblock_size' chunks of memory being added. Adding a larger span of
memory precludes memmap_on_memory semantics.

For users of hotplug such as kmem, large amounts of memory might get
added from the CXL subsystem. In some cases, this amount may exceed the
available 'main memory' to store the memmap for the memory being added.
In this case, it is useful to have a way to place the memmap on the
memory being added, even if it means splitting the addition into
memblock-sized chunks.

Change add_memory_resource() to loop over memblock-sized chunks of
memory if caller requested memmap_on_memory, and if other conditions for
it are met. Teach try_remove_memory() to also expect that a memory
range being removed might have been split up into memblock sized chunks,
and to loop through those as needed.

Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Huang Ying <ying.huang@intel.com>
Suggested-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
---
 mm/memory_hotplug.c | 165 +++++++++++++++++++++++++++++++---------------------
 1 file changed, 100 insertions(+), 65 deletions(-)

diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index f8d3e7427e32..43dbd71a4910 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -1380,6 +1380,44 @@ static bool mhp_supports_memmap_on_memory(unsigned long size)
 	return arch_supports_memmap_on_memory(vmemmap_size);
 }
 
+static int add_memory_create_devices(int nid, struct memory_group *group,
+				     u64 start, u64 size, mhp_t mhp_flags)
+{
+	struct mhp_params params = { .pgprot = pgprot_mhp(PAGE_KERNEL) };
+	struct vmem_altmap mhp_altmap = {
+		.base_pfn =  PHYS_PFN(start),
+		.end_pfn  =  PHYS_PFN(start + size - 1),
+	};
+	int ret;
+
+	if ((mhp_flags & MHP_MEMMAP_ON_MEMORY)) {
+		mhp_altmap.free = memory_block_memmap_on_memory_pages();
+		params.altmap = kmalloc(sizeof(struct vmem_altmap), GFP_KERNEL);
+		if (!params.altmap)
+			return -ENOMEM;
+
+		memcpy(params.altmap, &mhp_altmap, sizeof(mhp_altmap));
+	}
+
+	/* call arch's memory hotadd */
+	ret = arch_add_memory(nid, start, size, &params);
+	if (ret < 0)
+		goto error;
+
+	/* create memory block devices after memory was added */
+	ret = create_memory_block_devices(start, size, params.altmap, group);
+	if (ret)
+		goto err_bdev;
+
+	return 0;
+
+err_bdev:
+	arch_remove_memory(start, size, NULL);
+error:
+	kfree(params.altmap);
+	return ret;
+}
+
 /*
  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
  * and online/offline operations (triggered e.g. by sysfs).
@@ -1388,14 +1426,10 @@ static bool mhp_supports_memmap_on_memory(unsigned long size)
  */
 int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)
 {
-	struct mhp_params params = { .pgprot = pgprot_mhp(PAGE_KERNEL) };
+	unsigned long memblock_size = memory_block_size_bytes();
 	enum memblock_flags memblock_flags = MEMBLOCK_NONE;
-	struct vmem_altmap mhp_altmap = {
-		.base_pfn =  PHYS_PFN(res->start),
-		.end_pfn  =  PHYS_PFN(res->end),
-	};
 	struct memory_group *group = NULL;
-	u64 start, size;
+	u64 start, size, cur_start;
 	bool new_node = false;
 	int ret;
 
@@ -1436,28 +1470,21 @@ int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)
 	/*
 	 * Self hosted memmap array
 	 */
-	if (mhp_flags & MHP_MEMMAP_ON_MEMORY) {
-		if (mhp_supports_memmap_on_memory(size)) {
-			mhp_altmap.free = memory_block_memmap_on_memory_pages();
-			params.altmap = kmalloc(sizeof(struct vmem_altmap), GFP_KERNEL);
-			if (!params.altmap)
+	if ((mhp_flags & MHP_MEMMAP_ON_MEMORY) &&
+	    mhp_supports_memmap_on_memory(memblock_size)) {
+		for (cur_start = start; cur_start < start + size;
+		     cur_start += memblock_size) {
+			ret = add_memory_create_devices(nid, group, cur_start,
+							memblock_size,
+							mhp_flags);
+			if (ret)
 				goto error;
-
-			memcpy(params.altmap, &mhp_altmap, sizeof(mhp_altmap));
 		}
-		/* fallback to not using altmap  */
-	}
-
-	/* call arch's memory hotadd */
-	ret = arch_add_memory(nid, start, size, &params);
-	if (ret < 0)
-		goto error_free;
-
-	/* create memory block devices after memory was added */
-	ret = create_memory_block_devices(start, size, params.altmap, group);
-	if (ret) {
-		arch_remove_memory(start, size, NULL);
-		goto error_free;
+	} else {
+		ret = add_memory_create_devices(nid, group, start, size,
+						mhp_flags);
+		if (ret)
+			goto error;
 	}
 
 	if (new_node) {
@@ -1494,8 +1521,6 @@ int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)
 		walk_memory_blocks(start, size, NULL, online_memory_block);
 
 	return ret;
-error_free:
-	kfree(params.altmap);
 error:
 	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
 		memblock_remove(start, size);
@@ -2146,47 +2171,20 @@ void try_offline_node(int nid)
 }
 EXPORT_SYMBOL(try_offline_node);
 
-static int __ref try_remove_memory(u64 start, u64 size)
+static void __ref __try_remove_memory(int nid, u64 start, u64 size)
 {
+	int rc = 0;
 	struct memory_block *mem;
-	int rc = 0, nid = NUMA_NO_NODE;
 	struct vmem_altmap *altmap = NULL;
 
-	BUG_ON(check_hotplug_memory_range(start, size));
-
-	/*
-	 * All memory blocks must be offlined before removing memory.  Check
-	 * whether all memory blocks in question are offline and return error
-	 * if this is not the case.
-	 *
-	 * While at it, determine the nid. Note that if we'd have mixed nodes,
-	 * we'd only try to offline the last determined one -- which is good
-	 * enough for the cases we care about.
-	 */
-	rc = walk_memory_blocks(start, size, &nid, check_memblock_offlined_cb);
-	if (rc)
-		return rc;
-
-	/*
-	 * We only support removing memory added with MHP_MEMMAP_ON_MEMORY in
-	 * the same granularity it was added - a single memory block.
-	 */
-	if (mhp_memmap_on_memory()) {
-		rc = walk_memory_blocks(start, size, &mem, test_has_altmap_cb);
-		if (rc) {
-			if (size != memory_block_size_bytes()) {
-				pr_warn("Refuse to remove %#llx - %#llx,"
-					"wrong granularity\n",
-					start, start + size);
-				return -EINVAL;
-			}
-			altmap = mem->altmap;
-			/*
-			 * Mark altmap NULL so that we can add a debug
-			 * check on memblock free.
-			 */
-			mem->altmap = NULL;
-		}
+	rc = walk_memory_blocks(start, size, &mem, test_has_altmap_cb);
+	if (rc) {
+		altmap = mem->altmap;
+		/*
+		 * Mark altmap NULL so that we can add a debug
+		 * check on memblock free.
+		 */
+		mem->altmap = NULL;
 	}
 
 	/* remove memmap entry */
@@ -2219,6 +2217,43 @@ static int __ref try_remove_memory(u64 start, u64 size)
 		try_offline_node(nid);
 
 	mem_hotplug_done();
+}
+
+static int __ref try_remove_memory(u64 start, u64 size)
+{
+	int rc, nid = NUMA_NO_NODE;
+
+	BUG_ON(check_hotplug_memory_range(start, size));
+
+	/*
+	 * All memory blocks must be offlined before removing memory.  Check
+	 * whether all memory blocks in question are offline and return error
+	 * if this is not the case.
+	 *
+	 * While at it, determine the nid. Note that if we'd have mixed nodes,
+	 * we'd only try to offline the last determined one -- which is good
+	 * enough for the cases we care about.
+	 */
+	rc = walk_memory_blocks(start, size, &nid, check_memblock_offlined_cb);
+	if (rc)
+		return rc;
+
+	/*
+	 * For memmap_on_memory, the altmaps could have been added on
+	 * a per-memblock basis. Loop through the entire range if so,
+	 * and remove each memblock and its altmap.
+	 */
+	if (mhp_memmap_on_memory()) {
+		unsigned long memblock_size = memory_block_size_bytes();
+		u64 cur_start;
+
+		for (cur_start = start; cur_start < start + size;
+		     cur_start += memblock_size)
+			__try_remove_memory(nid, cur_start, memblock_size);
+	} else {
+		__try_remove_memory(nid, start, size);
+	}
+
 	return 0;
 }
 

-- 
2.41.0

[PATCH v4 2/2] dax/kmem: allow kmem to add memory with memmap_on_memory

[Vishal Verma]

Large amounts of memory managed by the kmem driver may come in via CXL,
and it is often desirable to have the memmap for this memory on the new
memory itself.

Enroll kmem-managed memory for memmap_on_memory semantics if the dax
region originates via CXL. For non-CXL dax regions, retain the existing
default behavior of hot adding without memmap_on_memory semantics.

Add a sysfs override under the dax device to control this behavior and
override either default.

Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Huang Ying <ying.huang@intel.com>
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
---
 drivers/dax/bus.h         |  1 +
 drivers/dax/dax-private.h |  1 +
 drivers/dax/bus.c         | 38 ++++++++++++++++++++++++++++++++++++++
 drivers/dax/cxl.c         |  1 +
 drivers/dax/hmem/hmem.c   |  1 +
 drivers/dax/kmem.c        |  8 +++++++-
 drivers/dax/pmem.c        |  1 +
 7 files changed, 50 insertions(+), 1 deletion(-)

diff --git a/drivers/dax/bus.h b/drivers/dax/bus.h
index 1ccd23360124..cbbf64443098 100644
--- a/drivers/dax/bus.h
+++ b/drivers/dax/bus.h
@@ -23,6 +23,7 @@ struct dev_dax_data {
 	struct dev_pagemap *pgmap;
 	resource_size_t size;
 	int id;
+	bool memmap_on_memory;
 };
 
 struct dev_dax *devm_create_dev_dax(struct dev_dax_data *data);
diff --git a/drivers/dax/dax-private.h b/drivers/dax/dax-private.h
index 27cf2daaaa79..446617b73aea 100644
--- a/drivers/dax/dax-private.h
+++ b/drivers/dax/dax-private.h
@@ -70,6 +70,7 @@ struct dev_dax {
 	struct ida ida;
 	struct device dev;
 	struct dev_pagemap *pgmap;
+	bool memmap_on_memory;
 	int nr_range;
 	struct dev_dax_range {
 		unsigned long pgoff;
diff --git a/drivers/dax/bus.c b/drivers/dax/bus.c
index 0ee96e6fc426..81351eb69884 100644
--- a/drivers/dax/bus.c
+++ b/drivers/dax/bus.c
@@ -367,6 +367,7 @@ static ssize_t create_store(struct device *dev, struct device_attribute *attr,
 			.dax_region = dax_region,
 			.size = 0,
 			.id = -1,
+			.memmap_on_memory = false,
 		};
 		struct dev_dax *dev_dax = devm_create_dev_dax(&data);
 
@@ -1269,6 +1270,40 @@ static ssize_t numa_node_show(struct device *dev,
 }
 static DEVICE_ATTR_RO(numa_node);
 
+static ssize_t memmap_on_memory_show(struct device *dev,
+				     struct device_attribute *attr, char *buf)
+{
+	struct dev_dax *dev_dax = to_dev_dax(dev);
+
+	return sprintf(buf, "%d\n", dev_dax->memmap_on_memory);
+}
+
+static ssize_t memmap_on_memory_store(struct device *dev,
+				      struct device_attribute *attr,
+				      const char *buf, size_t len)
+{
+	struct dev_dax *dev_dax = to_dev_dax(dev);
+	struct dax_region *dax_region = dev_dax->region;
+	ssize_t rc;
+	bool val;
+
+	rc = kstrtobool(buf, &val);
+	if (rc)
+		return rc;
+
+	device_lock(dax_region->dev);
+	if (!dax_region->dev->driver) {
+		device_unlock(dax_region->dev);
+		return -ENXIO;
+	}
+
+	dev_dax->memmap_on_memory = val;
+
+	device_unlock(dax_region->dev);
+	return rc == 0 ? len : rc;
+}
+static DEVICE_ATTR_RW(memmap_on_memory);
+
 static umode_t dev_dax_visible(struct kobject *kobj, struct attribute *a, int n)
 {
 	struct device *dev = container_of(kobj, struct device, kobj);
@@ -1295,6 +1330,7 @@ static struct attribute *dev_dax_attributes[] = {
 	&dev_attr_align.attr,
 	&dev_attr_resource.attr,
 	&dev_attr_numa_node.attr,
+	&dev_attr_memmap_on_memory.attr,
 	NULL,
 };
 
@@ -1400,6 +1436,8 @@ struct dev_dax *devm_create_dev_dax(struct dev_dax_data *data)
 	dev_dax->align = dax_region->align;
 	ida_init(&dev_dax->ida);
 
+	dev_dax->memmap_on_memory = data->memmap_on_memory;
+
 	inode = dax_inode(dax_dev);
 	dev->devt = inode->i_rdev;
 	dev->bus = &dax_bus_type;
diff --git a/drivers/dax/cxl.c b/drivers/dax/cxl.c
index 8bc9d04034d6..c696837ab23c 100644
--- a/drivers/dax/cxl.c
+++ b/drivers/dax/cxl.c
@@ -26,6 +26,7 @@ static int cxl_dax_region_probe(struct device *dev)
 		.dax_region = dax_region,
 		.id = -1,
 		.size = range_len(&cxlr_dax->hpa_range),
+		.memmap_on_memory = true,
 	};
 
 	return PTR_ERR_OR_ZERO(devm_create_dev_dax(&data));
diff --git a/drivers/dax/hmem/hmem.c b/drivers/dax/hmem/hmem.c
index 5d2ddef0f8f5..b9da69f92697 100644
--- a/drivers/dax/hmem/hmem.c
+++ b/drivers/dax/hmem/hmem.c
@@ -36,6 +36,7 @@ static int dax_hmem_probe(struct platform_device *pdev)
 		.dax_region = dax_region,
 		.id = -1,
 		.size = region_idle ? 0 : range_len(&mri->range),
+		.memmap_on_memory = false,
 	};
 
 	return PTR_ERR_OR_ZERO(devm_create_dev_dax(&data));
diff --git a/drivers/dax/kmem.c b/drivers/dax/kmem.c
index c57acb73e3db..0aa6c45a4e5a 100644
--- a/drivers/dax/kmem.c
+++ b/drivers/dax/kmem.c
@@ -12,6 +12,7 @@
 #include <linux/mm.h>
 #include <linux/mman.h>
 #include <linux/memory-tiers.h>
+#include <linux/memory_hotplug.h>
 #include "dax-private.h"
 #include "bus.h"
 
@@ -56,6 +57,7 @@ static int dev_dax_kmem_probe(struct dev_dax *dev_dax)
 	unsigned long total_len = 0;
 	struct dax_kmem_data *data;
 	int i, rc, mapped = 0;
+	mhp_t mhp_flags;
 	int numa_node;
 
 	/*
@@ -136,12 +138,16 @@ static int dev_dax_kmem_probe(struct dev_dax *dev_dax)
 		 */
 		res->flags = IORESOURCE_SYSTEM_RAM;
 
+		mhp_flags = MHP_NID_IS_MGID;
+		if (dev_dax->memmap_on_memory)
+			mhp_flags |= MHP_MEMMAP_ON_MEMORY;
+
 		/*
 		 * Ensure that future kexec'd kernels will not treat
 		 * this as RAM automatically.
 		 */
 		rc = add_memory_driver_managed(data->mgid, range.start,
-				range_len(&range), kmem_name, MHP_NID_IS_MGID);
+				range_len(&range), kmem_name, mhp_flags);
 
 		if (rc) {
 			dev_warn(dev, "mapping%d: %#llx-%#llx memory add failed\n",
diff --git a/drivers/dax/pmem.c b/drivers/dax/pmem.c
index ae0cb113a5d3..f3c6c67b8412 100644
--- a/drivers/dax/pmem.c
+++ b/drivers/dax/pmem.c
@@ -63,6 +63,7 @@ static struct dev_dax *__dax_pmem_probe(struct device *dev)
 		.id = id,
 		.pgmap = &pgmap,
 		.size = range_len(&range),
+		.memmap_on_memory = false,
 	};
 
 	return devm_create_dev_dax(&data);

-- 
2.41.0

Re: [PATCH v4 2/2] dax/kmem: allow kmem to add memory with memmap_on_memory

[David Hildenbrand]

On 28.09.23 22:30, Vishal Verma wrote:
> Large amounts of memory managed by the kmem driver may come in via CXL,
> and it is often desirable to have the memmap for this memory on the new
> memory itself.
> 
> Enroll kmem-managed memory for memmap_on_memory semantics if the dax
> region originates via CXL. For non-CXL dax regions, retain the existing
> default behavior of hot adding without memmap_on_memory semantics.
> 
> Add a sysfs override under the dax device to control this behavior and
> override either default.
> 
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: David Hildenbrand <david@redhat.com>
> Cc: Michal Hocko <mhocko@suse.com>
> Cc: Oscar Salvador <osalvador@suse.de>
> Cc: Dan Williams <dan.j.williams@intel.com>
> Cc: Dave Jiang <dave.jiang@intel.com>
> Cc: Dave Hansen <dave.hansen@linux.intel.com>
> Cc: Huang Ying <ying.huang@intel.com>
> Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
> Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
> ---

Reviewed-by: David Hildenbrand <david@redhat.com>

-- 
Cheers,

David / dhildenb

Re: [PATCH v4 1/2] mm/memory_hotplug: split memmap_on_memory requests across memblocks

[David Hildenbrand]

> +
> +static int __ref try_remove_memory(u64 start, u64 size)
> +{
> +	int rc, nid = NUMA_NO_NODE;
> +
> +	BUG_ON(check_hotplug_memory_range(start, size));
> +
> +	/*
> +	 * All memory blocks must be offlined before removing memory.  Check
> +	 * whether all memory blocks in question are offline and return error
> +	 * if this is not the case.
> +	 *
> +	 * While at it, determine the nid. Note that if we'd have mixed nodes,
> +	 * we'd only try to offline the last determined one -- which is good
> +	 * enough for the cases we care about.
> +	 */
> +	rc = walk_memory_blocks(start, size, &nid, check_memblock_offlined_cb);
> +	if (rc)
> +		return rc;
> +
> +	/*
> +	 * For memmap_on_memory, the altmaps could have been added on
> +	 * a per-memblock basis. Loop through the entire range if so,
> +	 * and remove each memblock and its altmap.
> +	 */
> +	if (mhp_memmap_on_memory()) {
> +		unsigned long memblock_size = memory_block_size_bytes();
> +		u64 cur_start;
> +
> +		for (cur_start = start; cur_start < start + size;
> +		     cur_start += memblock_size)
> +			__try_remove_memory(nid, cur_start, memblock_size);
> +	} else {
> +		__try_remove_memory(nid, start, size);
> +	}
> +
>   	return 0;
>   }

Why is the firmware, memblock and nid handling not kept in this outer 
function?

We really shouldn't be doing per memory block what needs to be done per 
memblock: remove_memory_block_devices() and arch_remove_memory().


-- 
Cheers,

David / dhildenb

Re: [PATCH v4 2/2] dax/kmem: allow kmem to add memory with memmap_on_memory

[Aneesh Kumar K V]

On 9/29/23 2:00 AM, Vishal Verma wrote:
> Large amounts of memory managed by the kmem driver may come in via CXL,
> and it is often desirable to have the memmap for this memory on the new
> memory itself.
> 
> Enroll kmem-managed memory for memmap_on_memory semantics if the dax
> region originates via CXL. For non-CXL dax regions, retain the existing
> default behavior of hot adding without memmap_on_memory semantics.
> 

Are we not looking at doing altmap space for CXL DAX regions? Last discussion around
this was suggesting we look at doing this via altmap reservation so that
we get contigous space for device memory enabling us to map them
via 1G direct mapping entries? 



> Add a sysfs override under the dax device to control this behavior and
> override either default.
> 
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: David Hildenbrand <david@redhat.com>
> Cc: Michal Hocko <mhocko@suse.com>
> Cc: Oscar Salvador <osalvador@suse.de>
> Cc: Dan Williams <dan.j.williams@intel.com>
> Cc: Dave Jiang <dave.jiang@intel.com>
> Cc: Dave Hansen <dave.hansen@linux.intel.com>
> Cc: Huang Ying <ying.huang@intel.com>
> Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
> Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
> ---
>  drivers/dax/bus.h         |  1 +
>  drivers/dax/dax-private.h |  1 +
>  drivers/dax/bus.c         | 38 ++++++++++++++++++++++++++++++++++++++
>  drivers/dax/cxl.c         |  1 +
>  drivers/dax/hmem/hmem.c   |  1 +
>  drivers/dax/kmem.c        |  8 +++++++-
>  drivers/dax/pmem.c        |  1 +
>  7 files changed, 50 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/dax/bus.h b/drivers/dax/bus.h
> index 1ccd23360124..cbbf64443098 100644
> --- a/drivers/dax/bus.h
> +++ b/drivers/dax/bus.h
> @@ -23,6 +23,7 @@ struct dev_dax_data {
>  	struct dev_pagemap *pgmap;
>  	resource_size_t size;
>  	int id;
> +	bool memmap_on_memory;
>  };
>  
>  struct dev_dax *devm_create_dev_dax(struct dev_dax_data *data);
> diff --git a/drivers/dax/dax-private.h b/drivers/dax/dax-private.h
> index 27cf2daaaa79..446617b73aea 100644
> --- a/drivers/dax/dax-private.h
> +++ b/drivers/dax/dax-private.h
> @@ -70,6 +70,7 @@ struct dev_dax {
>  	struct ida ida;
>  	struct device dev;
>  	struct dev_pagemap *pgmap;
> +	bool memmap_on_memory;
>  	int nr_range;
>  	struct dev_dax_range {
>  		unsigned long pgoff;
> diff --git a/drivers/dax/bus.c b/drivers/dax/bus.c
> index 0ee96e6fc426..81351eb69884 100644
> --- a/drivers/dax/bus.c
> +++ b/drivers/dax/bus.c
> @@ -367,6 +367,7 @@ static ssize_t create_store(struct device *dev, struct device_attribute *attr,
>  			.dax_region = dax_region,
>  			.size = 0,
>  			.id = -1,
> +			.memmap_on_memory = false,
>  		};
>  		struct dev_dax *dev_dax = devm_create_dev_dax(&data);
>  
> @@ -1269,6 +1270,40 @@ static ssize_t numa_node_show(struct device *dev,
>  }
>  static DEVICE_ATTR_RO(numa_node);
>  
> +static ssize_t memmap_on_memory_show(struct device *dev,
> +				     struct device_attribute *attr, char *buf)
> +{
> +	struct dev_dax *dev_dax = to_dev_dax(dev);
> +
> +	return sprintf(buf, "%d\n", dev_dax->memmap_on_memory);
> +}
> +
> +static ssize_t memmap_on_memory_store(struct device *dev,
> +				      struct device_attribute *attr,
> +				      const char *buf, size_t len)
> +{
> +	struct dev_dax *dev_dax = to_dev_dax(dev);
> +	struct dax_region *dax_region = dev_dax->region;
> +	ssize_t rc;
> +	bool val;
> +
> +	rc = kstrtobool(buf, &val);
> +	if (rc)
> +		return rc;
> +
> +	device_lock(dax_region->dev);
> +	if (!dax_region->dev->driver) {
> +		device_unlock(dax_region->dev);
> +		return -ENXIO;
> +	}
> +
> +	dev_dax->memmap_on_memory = val;
> +
> +	device_unlock(dax_region->dev);
> +	return rc == 0 ? len : rc;
> +}
> +static DEVICE_ATTR_RW(memmap_on_memory);
> +
>  static umode_t dev_dax_visible(struct kobject *kobj, struct attribute *a, int n)
>  {
>  	struct device *dev = container_of(kobj, struct device, kobj);
> @@ -1295,6 +1330,7 @@ static struct attribute *dev_dax_attributes[] = {
>  	&dev_attr_align.attr,
>  	&dev_attr_resource.attr,
>  	&dev_attr_numa_node.attr,
> +	&dev_attr_memmap_on_memory.attr,
>  	NULL,
>  };
>  
> @@ -1400,6 +1436,8 @@ struct dev_dax *devm_create_dev_dax(struct dev_dax_data *data)
>  	dev_dax->align = dax_region->align;
>  	ida_init(&dev_dax->ida);
>  
> +	dev_dax->memmap_on_memory = data->memmap_on_memory;
> +
>  	inode = dax_inode(dax_dev);
>  	dev->devt = inode->i_rdev;
>  	dev->bus = &dax_bus_type;
> diff --git a/drivers/dax/cxl.c b/drivers/dax/cxl.c
> index 8bc9d04034d6..c696837ab23c 100644
> --- a/drivers/dax/cxl.c
> +++ b/drivers/dax/cxl.c
> @@ -26,6 +26,7 @@ static int cxl_dax_region_probe(struct device *dev)
>  		.dax_region = dax_region,
>  		.id = -1,
>  		.size = range_len(&cxlr_dax->hpa_range),
> +		.memmap_on_memory = true,
>  	};
>  
>  	return PTR_ERR_OR_ZERO(devm_create_dev_dax(&data));
> diff --git a/drivers/dax/hmem/hmem.c b/drivers/dax/hmem/hmem.c
> index 5d2ddef0f8f5..b9da69f92697 100644
> --- a/drivers/dax/hmem/hmem.c
> +++ b/drivers/dax/hmem/hmem.c
> @@ -36,6 +36,7 @@ static int dax_hmem_probe(struct platform_device *pdev)
>  		.dax_region = dax_region,
>  		.id = -1,
>  		.size = region_idle ? 0 : range_len(&mri->range),
> +		.memmap_on_memory = false,
>  	};
>  
>  	return PTR_ERR_OR_ZERO(devm_create_dev_dax(&data));
> diff --git a/drivers/dax/kmem.c b/drivers/dax/kmem.c
> index c57acb73e3db..0aa6c45a4e5a 100644
> --- a/drivers/dax/kmem.c
> +++ b/drivers/dax/kmem.c
> @@ -12,6 +12,7 @@
>  #include <linux/mm.h>
>  #include <linux/mman.h>
>  #include <linux/memory-tiers.h>
> +#include <linux/memory_hotplug.h>
>  #include "dax-private.h"
>  #include "bus.h"
>  
> @@ -56,6 +57,7 @@ static int dev_dax_kmem_probe(struct dev_dax *dev_dax)
>  	unsigned long total_len = 0;
>  	struct dax_kmem_data *data;
>  	int i, rc, mapped = 0;
> +	mhp_t mhp_flags;
>  	int numa_node;
>  
>  	/*
> @@ -136,12 +138,16 @@ static int dev_dax_kmem_probe(struct dev_dax *dev_dax)
>  		 */
>  		res->flags = IORESOURCE_SYSTEM_RAM;
>  
> +		mhp_flags = MHP_NID_IS_MGID;
> +		if (dev_dax->memmap_on_memory)
> +			mhp_flags |= MHP_MEMMAP_ON_MEMORY;
> +
>  		/*
>  		 * Ensure that future kexec'd kernels will not treat
>  		 * this as RAM automatically.
>  		 */
>  		rc = add_memory_driver_managed(data->mgid, range.start,
> -				range_len(&range), kmem_name, MHP_NID_IS_MGID);
> +				range_len(&range), kmem_name, mhp_flags);
>  
>  		if (rc) {
>  			dev_warn(dev, "mapping%d: %#llx-%#llx memory add failed\n",
> diff --git a/drivers/dax/pmem.c b/drivers/dax/pmem.c
> index ae0cb113a5d3..f3c6c67b8412 100644
> --- a/drivers/dax/pmem.c
> +++ b/drivers/dax/pmem.c
> @@ -63,6 +63,7 @@ static struct dev_dax *__dax_pmem_probe(struct device *dev)
>  		.id = id,
>  		.pgmap = &pgmap,
>  		.size = range_len(&range),
> +		.memmap_on_memory = false,
>  	};
>  
>  	return devm_create_dev_dax(&data);
> 

Re: [PATCH v4 1/2] mm/memory_hotplug: split memmap_on_memory requests across memblocks

[Verma, Vishal L]

On Mon, 2023-10-02 at 11:28 +0200, David Hildenbrand wrote:
> 
> > +
> > +static int __ref try_remove_memory(u64 start, u64 size)
> > +{
> > +       int rc, nid = NUMA_NO_NODE;
> > +
> > +       BUG_ON(check_hotplug_memory_range(start, size));
> > +
> > +       /*
> > +        * All memory blocks must be offlined before removing memory.  Check
> > +        * whether all memory blocks in question are offline and return error
> > +        * if this is not the case.
> > +        *
> > +        * While at it, determine the nid. Note that if we'd have mixed nodes,
> > +        * we'd only try to offline the last determined one -- which is good
> > +        * enough for the cases we care about.
> > +        */
> > +       rc = walk_memory_blocks(start, size, &nid, check_memblock_offlined_cb);
> > +       if (rc)
> > +               return rc;
> > +
> > +       /*
> > +        * For memmap_on_memory, the altmaps could have been added on
> > +        * a per-memblock basis. Loop through the entire range if so,
> > +        * and remove each memblock and its altmap.
> > +        */
> > +       if (mhp_memmap_on_memory()) {
> > +               unsigned long memblock_size = memory_block_size_bytes();
> > +               u64 cur_start;
> > +
> > +               for (cur_start = start; cur_start < start + size;
> > +                    cur_start += memblock_size)
> > +                       __try_remove_memory(nid, cur_start, memblock_size);
> > +       } else {
> > +               __try_remove_memory(nid, start, size);
> > +       }
> > +
> >         return 0;
> >   }
> 
> Why is the firmware, memblock and nid handling not kept in this outer
> function?
> 
> We really shouldn't be doing per memory block what needs to be done per 
> memblock: remove_memory_block_devices() and arch_remove_memory().


Ah yes makes sense since we only do create_memory_block_devices() and
arch_add_memory() in the per memory block inner loop during addition.

How should the locking work in this case though?

The original code holds the mem_hotplug_begin() lock over
arch_remove_memory() and all of the nid and memblock stuff. Should I
just hold the lock and release it in the inner loop for each memory
block, and then also acquire and release it separately for the memblock
and nid stuff in the outer function?

Here's the incremental diff for what I'm thinking:

---

diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index 43dbd71a4910..13380178173d 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -2171,7 +2171,7 @@ void try_offline_node(int nid)
 }
 EXPORT_SYMBOL(try_offline_node);
 
-static void __ref __try_remove_memory(int nid, u64 start, u64 size)
+static void __ref remove_memory_block_and_altmap(int nid, u64 start, u64 size)
 {
 	int rc = 0;
 	struct memory_block *mem;
@@ -2187,9 +2187,6 @@ static void __ref __try_remove_memory(int nid, u64 start, u64 size)
 		mem->altmap = NULL;
 	}
 
-	/* remove memmap entry */
-	firmware_map_remove(start, start + size, "System RAM");
-
 	/*
 	 * Memory block device removal under the device_hotplug_lock is
 	 * a barrier against racing online attempts.
@@ -2206,16 +2203,6 @@ static void __ref __try_remove_memory(int nid, u64 start, u64 size)
 		kfree(altmap);
 	}
 
-	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
-		memblock_phys_free(start, size);
-		memblock_remove(start, size);
-	}
-
-	release_mem_region_adjustable(start, size);
-
-	if (nid != NUMA_NO_NODE)
-		try_offline_node(nid);
-
 	mem_hotplug_done();
 }
 
@@ -2249,11 +2236,29 @@ static int __ref try_remove_memory(u64 start, u64 size)
 
 		for (cur_start = start; cur_start < start + size;
 		     cur_start += memblock_size)
-			__try_remove_memory(nid, cur_start, memblock_size);
+			remove_memory_block_and_altmap(nid, cur_start,
+						       memblock_size);
 	} else {
-		__try_remove_memory(nid, start, size);
+		remove_memory_block_and_altmap(nid, start, size);
 	}
 
+	/* remove memmap entry */
+	firmware_map_remove(start, start + size, "System RAM");
+
+	mem_hotplug_begin();
+
+	if (IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) {
+		memblock_phys_free(start, size);
+		memblock_remove(start, size);
+	}
+
+	release_mem_region_adjustable(start, size);
+
+	if (nid != NUMA_NO_NODE)
+		try_offline_node(nid);
+
+	mem_hotplug_done();
+
 	return 0;
 }

Re: [PATCH v4 2/2] dax/kmem: allow kmem to add memory with memmap_on_memory

[Verma, Vishal L]

On Tue, 2023-10-03 at 09:34 +0530, Aneesh Kumar K V wrote:
> On 9/29/23 2:00 AM, Vishal Verma wrote:
> > Large amounts of memory managed by the kmem driver may come in via CXL,
> > and it is often desirable to have the memmap for this memory on the new
> > memory itself.
> > 
> > Enroll kmem-managed memory for memmap_on_memory semantics if the dax
> > region originates via CXL. For non-CXL dax regions, retain the existing
> > default behavior of hot adding without memmap_on_memory semantics.
> > 
> 
> Are we not looking at doing altmap space for CXL DAX regions? Last discussion around
> this was suggesting we look at doing this via altmap reservation so that
> we get contigous space for device memory enabling us to map them
> via 1G direct mapping entries? 
> 
Hey Aneesh - was this on a previous posting or something - do you have
a link so I can refresh myself on what the discussion was?

If it is enabling something in CXL similar to the --map=mem mode for
pmem + device dax, that could be incremental to this.

Re: [PATCH v4 2/2] dax/kmem: allow kmem to add memory with memmap_on_memory

[Aneesh Kumar K V]

On 10/4/23 5:18 AM, Verma, Vishal L wrote:
> On Tue, 2023-10-03 at 09:34 +0530, Aneesh Kumar K V wrote:
>> On 9/29/23 2:00 AM, Vishal Verma wrote:
>>> Large amounts of memory managed by the kmem driver may come in via CXL,
>>> and it is often desirable to have the memmap for this memory on the new
>>> memory itself.
>>>
>>> Enroll kmem-managed memory for memmap_on_memory semantics if the dax
>>> region originates via CXL. For non-CXL dax regions, retain the existing
>>> default behavior of hot adding without memmap_on_memory semantics.
>>>
>>
>> Are we not looking at doing altmap space for CXL DAX regions? Last discussion around
>> this was suggesting we look at doing this via altmap reservation so that
>> we get contigous space for device memory enabling us to map them
>> via 1G direct mapping entries? 
>>
> Hey Aneesh - was this on a previous posting or something - do you have
> a link so I can refresh myself on what the discussion was?
> 
> If it is enabling something in CXL similar to the --map=mem mode for
> pmem + device dax, that could be incremental to this.

Yes. Agree that we can do that incremental to this. 

https://lore.kernel.org/linux-mm/b6753402-2de9-25b2-36e9-eacd49752b19@redhat.com

-aneesh

Re: [PATCH v4 1/2] mm/memory_hotplug: split memmap_on_memory requests across memblocks

[David Hildenbrand]

On 03.10.23 22:03, Verma, Vishal L wrote:
> On Mon, 2023-10-02 at 11:28 +0200, David Hildenbrand wrote:
>>
>>> +
>>> +static int __ref try_remove_memory(u64 start, u64 size)
>>> +{
>>> +       int rc, nid = NUMA_NO_NODE;
>>> +
>>> +       BUG_ON(check_hotplug_memory_range(start, size));
>>> +
>>> +       /*
>>> +        * All memory blocks must be offlined before removing memory.  Check
>>> +        * whether all memory blocks in question are offline and return error
>>> +        * if this is not the case.
>>> +        *
>>> +        * While at it, determine the nid. Note that if we'd have mixed nodes,
>>> +        * we'd only try to offline the last determined one -- which is good
>>> +        * enough for the cases we care about.
>>> +        */
>>> +       rc = walk_memory_blocks(start, size, &nid, check_memblock_offlined_cb);
>>> +       if (rc)
>>> +               return rc;
>>> +
>>> +       /*
>>> +        * For memmap_on_memory, the altmaps could have been added on
>>> +        * a per-memblock basis. Loop through the entire range if so,
>>> +        * and remove each memblock and its altmap.
>>> +        */
>>> +       if (mhp_memmap_on_memory()) {
>>> +               unsigned long memblock_size = memory_block_size_bytes();
>>> +               u64 cur_start;
>>> +
>>> +               for (cur_start = start; cur_start < start + size;
>>> +                    cur_start += memblock_size)
>>> +                       __try_remove_memory(nid, cur_start, memblock_size);
>>> +       } else {
>>> +               __try_remove_memory(nid, start, size);
>>> +       }
>>> +
>>>          return 0;
>>>    }
>>
>> Why is the firmware, memblock and nid handling not kept in this outer
>> function?
>>
>> We really shouldn't be doing per memory block what needs to be done per
>> memblock: remove_memory_block_devices() and arch_remove_memory().
> 
> 
> Ah yes makes sense since we only do create_memory_block_devices() and
> arch_add_memory() in the per memory block inner loop during addition.
> 
> How should the locking work in this case though?

Sorry, I had to process a family NMI the last couple of days.

> 
> The original code holds the mem_hotplug_begin() lock over
> arch_remove_memory() and all of the nid and memblock stuff. Should I
> just hold the lock and release it in the inner loop for each memory
> block, and then also acquire and release it separately for the memblock
> and nid stuff in the outer function?

I think we have to hold it over the whole operation.

I saw that you sent a v5, I'll comment there.


-- 
Cheers,

David / dhildenb