[PATCH v8 00/12] mm: Sub-section memory hotplug support

[PATCH v8 00/12] mm: Sub-section memory hotplug support

[Dan Williams]

Changes since v7 [1]:

- Make subsection helpers pfn based rather than physical-address based
  (Oscar and Pavel)

- Make subsection bitmap definition scalable for different section and
  sub-section sizes across architectures. As a result:

      unsigned long map_active

  ...is converted to:

      DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION)

  ...and the helpers are renamed with a 'subsection' prefix. (Pavel)

- New in this version is a touch of arch/powerpc/include/asm/sparsemem.h
  in "[PATCH v8 01/12] mm/sparsemem: Introduce struct mem_section_usage"
  to define ARCH_SUBSECTION_SHIFT.

- Drop "mm/sparsemem: Introduce common definitions for the size and mask
  of a section" in favor of Robin's "mm/memremap: Rename and consolidate
  SECTION_SIZE" (Pavel)

- Collect some more Reviewed-by tags. Patches that still lack review
  tags: 1, 3, 9 - 12

[1]: https://lore.kernel.org/lkml/155677652226.2336373.8700273400832001094.stgit@dwillia2-desk3.amr.corp.intel.com/

---
[merge logistics]

Hi Andrew,

These are too late for v5.2, I'm posting this v8 during the merge window
to maintain the review momentum. 

---
[cover letter]

The memory hotplug section is an arbitrary / convenient unit for memory
hotplug. 'Section-size' units have bled into the user interface
('memblock' sysfs) and can not be changed without breaking existing
userspace. The section-size constraint, while mostly benign for typical
memory hotplug, has and continues to wreak havoc with 'device-memory'
use cases, persistent memory (pmem) in particular. Recall that pmem uses
devm_memremap_pages(), and subsequently arch_add_memory(), to allocate a
'struct page' memmap for pmem. However, it does not use the 'bottom
half' of memory hotplug, i.e. never marks pmem pages online and never
exposes the userspace memblock interface for pmem. This leaves an
opening to redress the section-size constraint.

To date, the libnvdimm subsystem has attempted to inject padding to
satisfy the internal constraints of arch_add_memory(). Beyond
complicating the code, leading to bugs [2], wasting memory, and limiting
configuration flexibility, the padding hack is broken when the platform
changes this physical memory alignment of pmem from one boot to the
next. Device failure (intermittent or permanent) and physical
reconfiguration are events that can cause the platform firmware to
change the physical placement of pmem on a subsequent boot, and device
failure is an everyday event in a data-center.

It turns out that sections are only a hard requirement of the
user-facing interface for memory hotplug and with a bit more
infrastructure sub-section arch_add_memory() support can be added for
kernel internal usages like devm_memremap_pages(). Here is an analysis
of the current design assumptions in the current code and how they are
addressed in the new implementation:

Current design assumptions:

- Sections that describe boot memory (early sections) are never
  unplugged / removed.

- pfn_valid(), in the CONFIG_SPARSEMEM_VMEMMAP=y, case devolves to a
  valid_section() check

- __add_pages() and helper routines assume all operations occur in
  PAGES_PER_SECTION units.

- The memblock sysfs interface only comprehends full sections

New design assumptions:

- Sections are instrumented with a sub-section bitmask to track (on x86)
  individual 2MB sub-divisions of a 128MB section.

- Partially populated early sections can be extended with additional
  sub-sections, and those sub-sections can be removed with
  arch_remove_memory(). With this in place we no longer lose usable memory
  capacity to padding.

- pfn_valid() is updated to look deeper than valid_section() to also check the
  active-sub-section mask. This indication is in the same cacheline as
  the valid_section() so the performance impact is expected to be
  negligible. So far the lkp robot has not reported any regressions.

- Outside of the core vmemmap population routines which are replaced,
  other helper routines like shrink_{zone,pgdat}_span() are updated to
  handle the smaller granularity. Core memory hotplug routines that deal
  with online memory are not touched.

- The existing memblock sysfs user api guarantees / assumptions are
  not touched since this capability is limited to !online
  !memblock-sysfs-accessible sections.

Meanwhile the issue reports continue to roll in from users that do not
understand when and how the 128MB constraint will bite them. The current
implementation relied on being able to support at least one misaligned
namespace, but that immediately falls over on any moderately complex
namespace creation attempt. Beyond the initial problem of 'System RAM'
colliding with pmem, and the unsolvable problem of physical alignment
changes, Linux is now being exposed to platforms that collide pmem
ranges with other pmem ranges by default [3]. In short,
devm_memremap_pages() has pushed the venerable section-size constraint
past the breaking point, and the simplicity of section-aligned
arch_add_memory() is no longer tenable.

These patches are exposed to the kbuild robot on my libnvdimm-pending
branch [4], and a preview of the unit test for this functionality is
available on the 'subsection-pending' branch of ndctl [5].

[2]: https://lore.kernel.org/r/155000671719.348031.2347363160141119237.stgit@dwillia2-desk3.amr.corp.intel.com
[3]: https://github.com/pmem/ndctl/issues/76
[4]: https://git.kernel.org/pub/scm/linux/kernel/git/djbw/nvdimm.git/log/?h=libnvdimm-pending
[5]: https://github.com/pmem/ndctl/commit/7c59b4867e1c

---

Dan Williams (11):
      mm/sparsemem: Introduce struct mem_section_usage
      mm/sparsemem: Add helpers track active portions of a section at boot
      mm/hotplug: Prepare shrink_{zone,pgdat}_span for sub-section removal
      mm/sparsemem: Convert kmalloc_section_memmap() to populate_section_memmap()
      mm/hotplug: Kill is_dev_zone() usage in __remove_pages()
      mm: Kill is_dev_zone() helper
      mm/sparsemem: Prepare for sub-section ranges
      mm/sparsemem: Support sub-section hotplug
      mm/devm_memremap_pages: Enable sub-section remap
      libnvdimm/pfn: Fix fsdax-mode namespace info-block zero-fields
      libnvdimm/pfn: Stop padding pmem namespaces to section alignment

Robin Murphy (1):
      mm/memremap: Rename and consolidate SECTION_SIZE


 arch/powerpc/include/asm/sparsemem.h |    3 
 arch/x86/mm/init_64.c                |    4 
 drivers/nvdimm/dax_devs.c            |    2 
 drivers/nvdimm/pfn.h                 |   15 -
 drivers/nvdimm/pfn_devs.c            |   95 +++------
 include/linux/memory_hotplug.h       |    7 -
 include/linux/mm.h                   |    4 
 include/linux/mmzone.h               |   93 +++++++--
 kernel/memremap.c                    |   63 ++----
 mm/hmm.c                             |    2 
 mm/memory_hotplug.c                  |  172 +++++++++-------
 mm/page_alloc.c                      |    8 -
 mm/sparse-vmemmap.c                  |   21 +-
 mm/sparse.c                          |  369 +++++++++++++++++++++++-----------
 14 files changed, 511 insertions(+), 347 deletions(-)

[PATCH v8 11/12] libnvdimm/pfn: Fix fsdax-mode namespace info-block zero-fields

[Dan Williams]

At namespace creation time there is the potential for the "expected to
be zero" fields of a 'pfn' info-block to be filled with indeterminate
data. While the kernel buffer is zeroed on allocation it is immediately
overwritten by nd_pfn_validate() filling it with the current contents of
the on-media info-block location. For fields like, 'flags' and the
'padding' it potentially means that future implementations can not rely
on those fields being zero.

In preparation to stop using the 'start_pad' and 'end_trunc' fields for
section alignment, arrange for fields that are not explicitly
initialized to be guaranteed zero. Bump the minor version to indicate it
is safe to assume the 'padding' and 'flags' are zero. Otherwise, this
corruption is expected to benign since all other critical fields are
explicitly initialized.

Fixes: 32ab0a3f5170 ("libnvdimm, pmem: 'struct page' for pmem")
Cc: <stable@vger.kernel.org>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/nvdimm/dax_devs.c |    2 +-
 drivers/nvdimm/pfn.h      |    1 +
 drivers/nvdimm/pfn_devs.c |   18 +++++++++++++++---
 3 files changed, 17 insertions(+), 4 deletions(-)

diff --git a/drivers/nvdimm/dax_devs.c b/drivers/nvdimm/dax_devs.c
index 0453f49dc708..326f02ffca81 100644
--- a/drivers/nvdimm/dax_devs.c
+++ b/drivers/nvdimm/dax_devs.c
@@ -126,7 +126,7 @@ int nd_dax_probe(struct device *dev, struct nd_namespace_common *ndns)
 	nvdimm_bus_unlock(&ndns->dev);
 	if (!dax_dev)
 		return -ENOMEM;
-	pfn_sb = devm_kzalloc(dev, sizeof(*pfn_sb), GFP_KERNEL);
+	pfn_sb = devm_kmalloc(dev, sizeof(*pfn_sb), GFP_KERNEL);
 	nd_pfn->pfn_sb = pfn_sb;
 	rc = nd_pfn_validate(nd_pfn, DAX_SIG);
 	dev_dbg(dev, "dax: %s\n", rc == 0 ? dev_name(dax_dev) : "<none>");
diff --git a/drivers/nvdimm/pfn.h b/drivers/nvdimm/pfn.h
index dde9853453d3..e901e3a3b04c 100644
--- a/drivers/nvdimm/pfn.h
+++ b/drivers/nvdimm/pfn.h
@@ -36,6 +36,7 @@ struct nd_pfn_sb {
 	__le32 end_trunc;
 	/* minor-version-2 record the base alignment of the mapping */
 	__le32 align;
+	/* minor-version-3 guarantee the padding and flags are zero */
 	u8 padding[4000];
 	__le64 checksum;
 };
diff --git a/drivers/nvdimm/pfn_devs.c b/drivers/nvdimm/pfn_devs.c
index 01f40672507f..a2406253eb70 100644
--- a/drivers/nvdimm/pfn_devs.c
+++ b/drivers/nvdimm/pfn_devs.c
@@ -420,6 +420,15 @@ static int nd_pfn_clear_memmap_errors(struct nd_pfn *nd_pfn)
 	return 0;
 }
 
+/**
+ * nd_pfn_validate - read and validate info-block
+ * @nd_pfn: fsdax namespace runtime state / properties
+ * @sig: 'devdax' or 'fsdax' signature
+ *
+ * Upon return the info-block buffer contents (->pfn_sb) are
+ * indeterminate when validation fails, and a coherent info-block
+ * otherwise.
+ */
 int nd_pfn_validate(struct nd_pfn *nd_pfn, const char *sig)
 {
 	u64 checksum, offset;
@@ -565,7 +574,7 @@ int nd_pfn_probe(struct device *dev, struct nd_namespace_common *ndns)
 	nvdimm_bus_unlock(&ndns->dev);
 	if (!pfn_dev)
 		return -ENOMEM;
-	pfn_sb = devm_kzalloc(dev, sizeof(*pfn_sb), GFP_KERNEL);
+	pfn_sb = devm_kmalloc(dev, sizeof(*pfn_sb), GFP_KERNEL);
 	nd_pfn = to_nd_pfn(pfn_dev);
 	nd_pfn->pfn_sb = pfn_sb;
 	rc = nd_pfn_validate(nd_pfn, PFN_SIG);
@@ -702,7 +711,7 @@ static int nd_pfn_init(struct nd_pfn *nd_pfn)
 	u64 checksum;
 	int rc;
 
-	pfn_sb = devm_kzalloc(&nd_pfn->dev, sizeof(*pfn_sb), GFP_KERNEL);
+	pfn_sb = devm_kmalloc(&nd_pfn->dev, sizeof(*pfn_sb), GFP_KERNEL);
 	if (!pfn_sb)
 		return -ENOMEM;
 
@@ -711,11 +720,14 @@ static int nd_pfn_init(struct nd_pfn *nd_pfn)
 		sig = DAX_SIG;
 	else
 		sig = PFN_SIG;
+
 	rc = nd_pfn_validate(nd_pfn, sig);
 	if (rc != -ENODEV)
 		return rc;
 
 	/* no info block, do init */;
+	memset(pfn_sb, 0, sizeof(*pfn_sb));
+
 	nd_region = to_nd_region(nd_pfn->dev.parent);
 	if (nd_region->ro) {
 		dev_info(&nd_pfn->dev,
@@ -768,7 +780,7 @@ static int nd_pfn_init(struct nd_pfn *nd_pfn)
 	memcpy(pfn_sb->uuid, nd_pfn->uuid, 16);
 	memcpy(pfn_sb->parent_uuid, nd_dev_to_uuid(&ndns->dev), 16);
 	pfn_sb->version_major = cpu_to_le16(1);
-	pfn_sb->version_minor = cpu_to_le16(2);
+	pfn_sb->version_minor = cpu_to_le16(3);
 	pfn_sb->start_pad = cpu_to_le32(start_pad);
 	pfn_sb->end_trunc = cpu_to_le32(end_trunc);
 	pfn_sb->align = cpu_to_le32(nd_pfn->align);

Re: [PATCH v8 00/12] mm: Sub-section memory hotplug support

[Mike Rapoport]

Hi Dan,

On Mon, May 06, 2019 at 04:39:26PM -0700, Dan Williams wrote:
> Changes since v7 [1]:

Sorry for jumping late, but presuming there will be v9, it'd be great if it
would also include include updates to
Documentation/admin-guide/mm/memory-hotplug.rst and
Documentation/vm/memory-model.rst
 
> - Make subsection helpers pfn based rather than physical-address based
>   (Oscar and Pavel)
> 
> - Make subsection bitmap definition scalable for different section and
>   sub-section sizes across architectures. As a result:
> 
>       unsigned long map_active
> 
>   ...is converted to:
> 
>       DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION)
> 
>   ...and the helpers are renamed with a 'subsection' prefix. (Pavel)
> 
> - New in this version is a touch of arch/powerpc/include/asm/sparsemem.h
>   in "[PATCH v8 01/12] mm/sparsemem: Introduce struct mem_section_usage"
>   to define ARCH_SUBSECTION_SHIFT.
> 
> - Drop "mm/sparsemem: Introduce common definitions for the size and mask
>   of a section" in favor of Robin's "mm/memremap: Rename and consolidate
>   SECTION_SIZE" (Pavel)
> 
> - Collect some more Reviewed-by tags. Patches that still lack review
>   tags: 1, 3, 9 - 12
> 
> [1]: https://lore.kernel.org/lkml/155677652226.2336373.8700273400832001094.stgit@dwillia2-desk3.amr.corp.intel.com/
> 
> ---
> [merge logistics]
> 
> Hi Andrew,
> 
> These are too late for v5.2, I'm posting this v8 during the merge window
> to maintain the review momentum. 
> 
> ---
> [cover letter]
> 
> The memory hotplug section is an arbitrary / convenient unit for memory
> hotplug. 'Section-size' units have bled into the user interface
> ('memblock' sysfs) and can not be changed without breaking existing
> userspace. The section-size constraint, while mostly benign for typical
> memory hotplug, has and continues to wreak havoc with 'device-memory'
> use cases, persistent memory (pmem) in particular. Recall that pmem uses
> devm_memremap_pages(), and subsequently arch_add_memory(), to allocate a
> 'struct page' memmap for pmem. However, it does not use the 'bottom
> half' of memory hotplug, i.e. never marks pmem pages online and never
> exposes the userspace memblock interface for pmem. This leaves an
> opening to redress the section-size constraint.
> 
> To date, the libnvdimm subsystem has attempted to inject padding to
> satisfy the internal constraints of arch_add_memory(). Beyond
> complicating the code, leading to bugs [2], wasting memory, and limiting
> configuration flexibility, the padding hack is broken when the platform
> changes this physical memory alignment of pmem from one boot to the
> next. Device failure (intermittent or permanent) and physical
> reconfiguration are events that can cause the platform firmware to
> change the physical placement of pmem on a subsequent boot, and device
> failure is an everyday event in a data-center.
> 
> It turns out that sections are only a hard requirement of the
> user-facing interface for memory hotplug and with a bit more
> infrastructure sub-section arch_add_memory() support can be added for
> kernel internal usages like devm_memremap_pages(). Here is an analysis
> of the current design assumptions in the current code and how they are
> addressed in the new implementation:
> 
> Current design assumptions:
> 
> - Sections that describe boot memory (early sections) are never
>   unplugged / removed.
> 
> - pfn_valid(), in the CONFIG_SPARSEMEM_VMEMMAP=y, case devolves to a
>   valid_section() check
> 
> - __add_pages() and helper routines assume all operations occur in
>   PAGES_PER_SECTION units.
> 
> - The memblock sysfs interface only comprehends full sections
> 
> New design assumptions:
> 
> - Sections are instrumented with a sub-section bitmask to track (on x86)
>   individual 2MB sub-divisions of a 128MB section.
> 
> - Partially populated early sections can be extended with additional
>   sub-sections, and those sub-sections can be removed with
>   arch_remove_memory(). With this in place we no longer lose usable memory
>   capacity to padding.
> 
> - pfn_valid() is updated to look deeper than valid_section() to also check the
>   active-sub-section mask. This indication is in the same cacheline as
>   the valid_section() so the performance impact is expected to be
>   negligible. So far the lkp robot has not reported any regressions.
> 
> - Outside of the core vmemmap population routines which are replaced,
>   other helper routines like shrink_{zone,pgdat}_span() are updated to
>   handle the smaller granularity. Core memory hotplug routines that deal
>   with online memory are not touched.
> 
> - The existing memblock sysfs user api guarantees / assumptions are
>   not touched since this capability is limited to !online
>   !memblock-sysfs-accessible sections.
> 
> Meanwhile the issue reports continue to roll in from users that do not
> understand when and how the 128MB constraint will bite them. The current
> implementation relied on being able to support at least one misaligned
> namespace, but that immediately falls over on any moderately complex
> namespace creation attempt. Beyond the initial problem of 'System RAM'
> colliding with pmem, and the unsolvable problem of physical alignment
> changes, Linux is now being exposed to platforms that collide pmem
> ranges with other pmem ranges by default [3]. In short,
> devm_memremap_pages() has pushed the venerable section-size constraint
> past the breaking point, and the simplicity of section-aligned
> arch_add_memory() is no longer tenable.
> 
> These patches are exposed to the kbuild robot on my libnvdimm-pending
> branch [4], and a preview of the unit test for this functionality is
> available on the 'subsection-pending' branch of ndctl [5].
> 
> [2]: https://lore.kernel.org/r/155000671719.348031.2347363160141119237.stgit@dwillia2-desk3.amr.corp.intel.com
> [3]: https://github.com/pmem/ndctl/issues/76
> [4]: https://git.kernel.org/pub/scm/linux/kernel/git/djbw/nvdimm.git/log/?h=libnvdimm-pending
> [5]: https://github.com/pmem/ndctl/commit/7c59b4867e1c
> 
> ---
> 
> Dan Williams (11):
>       mm/sparsemem: Introduce struct mem_section_usage
>       mm/sparsemem: Add helpers track active portions of a section at boot
>       mm/hotplug: Prepare shrink_{zone,pgdat}_span for sub-section removal
>       mm/sparsemem: Convert kmalloc_section_memmap() to populate_section_memmap()
>       mm/hotplug: Kill is_dev_zone() usage in __remove_pages()
>       mm: Kill is_dev_zone() helper
>       mm/sparsemem: Prepare for sub-section ranges
>       mm/sparsemem: Support sub-section hotplug
>       mm/devm_memremap_pages: Enable sub-section remap
>       libnvdimm/pfn: Fix fsdax-mode namespace info-block zero-fields
>       libnvdimm/pfn: Stop padding pmem namespaces to section alignment
> 
> Robin Murphy (1):
>       mm/memremap: Rename and consolidate SECTION_SIZE
> 
> 
>  arch/powerpc/include/asm/sparsemem.h |    3 
>  arch/x86/mm/init_64.c                |    4 
>  drivers/nvdimm/dax_devs.c            |    2 
>  drivers/nvdimm/pfn.h                 |   15 -
>  drivers/nvdimm/pfn_devs.c            |   95 +++------
>  include/linux/memory_hotplug.h       |    7 -
>  include/linux/mm.h                   |    4 
>  include/linux/mmzone.h               |   93 +++++++--
>  kernel/memremap.c                    |   63 ++----
>  mm/hmm.c                             |    2 
>  mm/memory_hotplug.c                  |  172 +++++++++-------
>  mm/page_alloc.c                      |    8 -
>  mm/sparse-vmemmap.c                  |   21 +-
>  mm/sparse.c                          |  369 +++++++++++++++++++++++-----------
>  14 files changed, 511 insertions(+), 347 deletions(-)
> 

-- 
Sincerely yours,
Mike.

Re: [PATCH v8 00/12] mm: Sub-section memory hotplug support

[Dan Williams]

On Mon, May 13, 2019 at 2:02 PM Mike Rapoport <rppt@linux.ibm.com> wrote:
>
> Hi Dan,
>
> On Mon, May 06, 2019 at 04:39:26PM -0700, Dan Williams wrote:
> > Changes since v7 [1]:
>
> Sorry for jumping late

No worries, it needs to be rebased on David's "mm/memory_hotplug:
Factor out memory block device handling" series which puts it firmly
in v5.3 territory.

> but presuming there will be v9, it'd be great if it
> would also include include updates to
> Documentation/admin-guide/mm/memory-hotplug.rst and

If I've done my job right this series should be irrelevant to
Documentation/admin-guide/mm/memory-hotplug.rst. The subsection
capability is strictly limited to arch_add_memory() users that never
online the memory, i.e. only ZONE_DEVICE / devm_memremap_pages()
users. So this isn't "memory-hotplug" as much as it is "support for
subsection vmemmap management".

> Documentation/vm/memory-model.rst

This looks more fitting and should probably include a paragraph on the
general ZONE_DEVICE / devm_memremap_pages() use case.

Re: [PATCH v8 00/12] mm: Sub-section memory hotplug support

[Oscar Salvador]

On Mon, May 06, 2019 at 04:39:26PM -0700, Dan Williams wrote:
> Changes since v7 [1]:
> 
> - Make subsection helpers pfn based rather than physical-address based
>   (Oscar and Pavel)
> 
> - Make subsection bitmap definition scalable for different section and
>   sub-section sizes across architectures. As a result:
> 
>       unsigned long map_active
> 
>   ...is converted to:
> 
>       DECLARE_BITMAP(subsection_map, SUBSECTIONS_PER_SECTION)
> 
>   ...and the helpers are renamed with a 'subsection' prefix. (Pavel)
> 
> - New in this version is a touch of arch/powerpc/include/asm/sparsemem.h
>   in "[PATCH v8 01/12] mm/sparsemem: Introduce struct mem_section_usage"
>   to define ARCH_SUBSECTION_SHIFT.
> 
> - Drop "mm/sparsemem: Introduce common definitions for the size and mask
>   of a section" in favor of Robin's "mm/memremap: Rename and consolidate
>   SECTION_SIZE" (Pavel)
> 
> - Collect some more Reviewed-by tags. Patches that still lack review
>   tags: 1, 3, 9 - 12

Hi Dan,

are you planning to send V10 anytime soon?

After you addressed comments from Patch#9, the general implementation looks
fine to me and nothing sticked out from the other patches.
But I would rather wait to see v10 with the comments addressed before stamping
my Reviewed-by.

I am planning to fire my vmemmap patchset again [1], and I would like to re-base
it on top of this work, otherwise we will face many unnecessary collisions.

Thanks

[1] https://patchwork.kernel.org/patch/10875025/

-- 
Oscar Salvador
SUSE L3