Failure to create namespaces; pfn failed to enable + kernel oops #76
Comments
patr-geary-smci
changed the title
|
There is a known limitation with requiring namespaces to be 128MB aligned, but obviously the kernel is not figuring out the proper alignment. Can you send the output of "cat /proc/iomem" on this system and "ndctl list -R" |
|
iomem: Before configuring any namespaces: After namespace creation failure and oops: |
|
Tried manual alignment to no avail. 4k and 2M. All failed with the same message; all on the same region. -Edit: Removed 128M; it wasn't supported when I tested. |
|
I started going through the source. 1G Alignment worked. |
|
As root can you send the output of: I fear this configuration may be placing two regions within 128MB of each other. |
|
|
Ugh, yes, those are 64MB alignments. That is why 1GB aligned namespaces works and 2MB does not. |
|
Is this something I should be pointing out to the ipmctl Intel folks, or is this going to be a kernel "issue"? I actually haven't delved into the spec for any of this stuff, so I'm not sitting in a good place to draw a conclusion. |
|
Linux needs 128MB alignment for each adjacent namespace. There isn't a fix because BIOS has no visibility or responsibility for Linux alignment constraints. Going forward Linux will eventually gain the capability to support fsdax mode with namespaces that collide within a section (128MB) until then the only workarounds are "raw" mode (not useful), or requiring fsdax namespaces to be created with "--align=1GB". We faced something similar with section collisions with System RAM, but in that case we could interrogate the collision ahead of time. As it stands we don't find out about this collision until its too late. I'll try to think of something more clever, but the solution may devolve to just teaching the tooling to require large alignments. |
|
Hold off on contacting ipmctl folks until I exhaust a possible kernel workaround for this issue... |
|
If it's BIOS related that's something I can fire up a conversation with our BIOS team members responsible for integrating Intel's code. Interestingly I did successfully configure goals and namespaces under the BIOS and I experienced no such issues, although the region size was.... different. I didn't prod at the region ranges though, I just saw size differences once allocated. I didn't dig into it since the whole starting point for me is handling validation of these things under Linux. |
|
Certainly the BIOS can fix it by making sure each NFIT SPA range is at least 128MB aligned. |
|
...there's just no specification that requires that. |
|
No spec = big problem I'll talk to some folks, do some reading and see if Intel specified the alignment. |
|
No Intel does not specify the alignment, this is an odd and specific quirk of the current Linux memory mapping implementation. This is why "raw" namespace do not have this problem. |
|
Let me ask around. Might be some time before I get a good answer, and even then it might only apply to SMCI. |
|
Can you give this kernel patch a try: Even if you get this changed in your particular BIOS I still think the kernel should carry this fix. Also I'd like to give you credit for the report if you're comfortable with your email address in the public kernel history. Just respond with a tag like: Reported-by: Your Name your.name@email.address Thanks! |
fengguang
pushed a commit
to 0day-ci/linux
that referenced
this issue
Nov 24, 2018
Commit cfe30b8 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b8 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
|
I'm already out there so there are no worries. I'll give it a run Monday.
On Nov 24, 2018, at 11:02 AM, Dan Williams <notifications@github.com<mailto:notifications@github.com>> wrote:
Can you give this kernel patch a try:
https://patchwork.kernel.org/patch/10696565/
Even if you get this changed in your particular BIOS I still think the kernel should carry this fix.
Also I'd like to give you credit for the report if you're comfortable with your email address in the public kernel history. Just respond with a tag like:
Reported-by: Your Name your.name@email.address<mailto:your.name@email.address>
Thanks!
-
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub [github.com]<https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_pmem_ndctl_issues_76-23issuecomment-2D441388772&d=DwMFaQ&c=4DxX-JX0i28X6V65hK0ftwVK1xnmwcYC0vo7GVya1JY&r=PqM_W6Cnc3d4hrZogsMkDosh7sQQDLNzFhCmjRxIq8k&m=iwPeAq9DzhnyyUcfwfLGEGkAgzDIAlCe0pu7Rpo3Wg0&s=QKzbVysy1ygpZGlsh1LSj9aGadBjuKI14Yq_TuGlpRg&e=>, or mute the thread [github.com]<https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_AV3KUYqDp67wf6wE-2DCLC0InAy5gVtGhoks5uyZfYgaJpZM4YqJEL&d=DwMFaQ&c=4DxX-JX0i28X6V65hK0ftwVK1xnmwcYC0vo7GVya1JY&r=PqM_W6Cnc3d4hrZogsMkDosh7sQQDLNzFhCmjRxIq8k&m=iwPeAq9DzhnyyUcfwfLGEGkAgzDIAlCe0pu7Rpo3Wg0&s=tBpaFxWYiLHcnAajQjrE9hOoHnD3IXRySAUatoO_EcQ&e=>.
|
|
Yeesh; look at that mess from my phone's mailer. Anyways; tested. Works fine. |
|
Thanks @patr-geary-smci much appreciated. Mind sharing your contact details so I can get the patch credit correct? All I can see on your github profile is "patr-geary-smci". |
avagin
pushed a commit
to avagin/linux
that referenced
this issue
Nov 30, 2018
Commit cfe30b8 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled the libnvdimm core to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b8 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Whissi
pushed a commit
to Whissi/linux-stable
that referenced
this issue
Dec 13, 2018
commit ae86cbf upstream. Commit cfe30b8 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b8 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
woodsts
pushed a commit
to woodsts/linux-stable
that referenced
this issue
Dec 13, 2018
commit ae86cbf upstream. Commit cfe30b8 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b8 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
isjerryxiao
pushed a commit
to isjerryxiao/Amlogic_s905-kernel
that referenced
this issue
Dec 15, 2018
commit ae86cbf upstream. Commit cfe30b8 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b8 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Emerald-Phoenix
pushed a commit
to Emerald-Phoenix/linux
that referenced
this issue
Jan 1, 2019
commit ae86cbf upstream. Commit cfe30b8 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b8 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
gregmarsden
pushed a commit
to oracle/linux-uek
that referenced
this issue
Jan 25, 2019
Commit cfe30b8 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b8 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> (cherry picked from commit ae86cbf) Orabug: 29168389 Signed-off-by: Jane Chu <jane.chu@oracle.com> Reviewed-by: Larry Bassel <larry.bassel@oracle.com>
jpuhlman
pushed a commit
to MontaVista-OpenSourceTechnology/linux-mvista-2.4
that referenced
this issue
Feb 4, 2019
Source: linux-mvista-2.4 MR: 96580, 00000 Type: Integration Disposition: Merged from linux-mvista-2.4 ChangeID: 4794d94b44c4903514a878b9097e99417a5e6dbe Description: commit ae86cbfef3818300f1972e52f67a93211acb0e24 upstream. Commit cfe30b8 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b8 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Armin Kuster <akuster@mvista.com> Signed-off-by: Jeremy Puhlman <jpuhlman@mvista.com>
jpuhlman
pushed a commit
to MontaVista-OpenSourceTechnology/linux-mvista-2.4
that referenced
this issue
Feb 7, 2019
Source: Kernel.org MR: 96580 Type: Security Fix Disposition: Backport from git://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable linux-4.14.y ChangeID: 4794d94b44c4903514a878b9097e99417a5e6dbe Description: commit ae86cbfef3818300f1972e52f67a93211acb0e24 upstream. Commit cfe30b8 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b8 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Armin Kuster <akuster@mvista.com>
jpuhlman
pushed a commit
to MontaVista-OpenSourceTechnology/linux-mvista-2.4
that referenced
this issue
Mar 5, 2019
Source: Kernel.org MR: 96580 Type: Security Fix Disposition: Backport from git://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable linux-4.14.y ChangeID: 4794d94b44c4903514a878b9097e99417a5e6dbe Description: commit ae86cbfef3818300f1972e52f67a93211acb0e24 upstream. Commit cfe30b8 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b8 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Armin Kuster <akuster@mvista.com>
randomblame
pushed a commit
to randomblame/android_kernel_xiaomi_sm8150
that referenced
this issue
Apr 13, 2019
commit ae86cbfef3818300f1972e52f67a93211acb0e24 upstream. Commit cfe30b8 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b8 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
randomblame
pushed a commit
to randomblame/android_kernel_xiaomi_sm8150
that referenced
this issue
Apr 27, 2019
commit ae86cbfef3818300f1972e52f67a93211acb0e24 upstream. Commit cfe30b8 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b8 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
rgushchin
pushed a commit
to rgushchin/linux
that referenced
this issue
Jun 11, 2019
Patch series "mm: Sub-section memory hotplug support", v9.
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]: pmem/ndctl#76
[4]: https://git.kernel.org/pub/scm/linux/kernel/git/djbw/nvdimm.git/log/?h=libnvdimm-pending
[5]: pmem/ndctl@7c59b4867e1c
This patch (of 12):
Towards enabling memory hotplug to track partial population of a section,
introduce 'struct mem_section_usage'.
A pointer to a 'struct mem_section_usage' instance replaces the existing
pointer to a 'pageblock_flags' bitmap. Effectively it adds one more
'unsigned long' beyond the 'pageblock_flags' (usemap) allocation to house
a new 'subsection_map' bitmap. The new bitmap enables the memory
hot{plug,remove} implementation to act on incremental sub-divisions of a
section.
The default SUBSECTION_SHIFT is chosen to keep the 'subsection_map' no
larger than a single 'unsigned long' on the major architectures.
Alternatively an architecture can define ARCH_SUBSECTION_SHIFT to override
the default PMD_SHIFT. Note that PowerPC needs to use
ARCH_SUBSECTION_SHIFT to workaround PMD_SHIFT being a non-constant
expression on PowerPC.
The primary motivation for this functionality is to support platforms that
mix "System RAM" and "Persistent Memory" within a single section, or
multiple PMEM ranges with different mapping lifetimes within a single
section. The section restriction for hotplug has caused an ongoing saga
of hacks and bugs for devm_memremap_pages() users.
Beyond the fixups to teach existing paths how to retrieve the 'usemap'
from a section, and updates to usemap allocation path, there are no
expected behavior changes.
Link: http://lkml.kernel.org/r/155977187407.2443951.16503493275720588454.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: David Hildenbrand <david@redhat.com>
Cc: Jane Chu <jane.chu@oracle.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
cdown
pushed a commit
to cdown/linux
that referenced
this issue
Jul 4, 2019
Patch series "mm: Sub-section memory hotplug support", v10.
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 a subsection-v10 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]: pmem/ndctl#76
[4]: https://git.kernel.org/pub/scm/linux/kernel/git/djbw/nvdimm.git/log/?h=subsection-v10
[5]: pmem/ndctl@7c59b4867e1c
This patch (of 13):
Towards enabling memory hotplug to track partial population of a section,
introduce 'struct mem_section_usage'.
A pointer to a 'struct mem_section_usage' instance replaces the existing
pointer to a 'pageblock_flags' bitmap. Effectively it adds one more
'unsigned long' beyond the 'pageblock_flags' (usemap) allocation to house
a new 'subsection_map' bitmap. The new bitmap enables the memory
hot{plug,remove} implementation to act on incremental sub-divisions of a
section.
SUBSECTION_SHIFT is defined as global constant instead of per-architecture
value like SECTION_SIZE_BITS in order to allow cross-arch compatibility of
subsection users. Specifically a common subsection size allows for the
possibility that persistent memory namespace configurations be made
compatible across architectures.
The primary motivation for this functionality is to support platforms that
mix "System RAM" and "Persistent Memory" within a single section, or
multiple PMEM ranges with different mapping lifetimes within a single
section. The section restriction for hotplug has caused an ongoing saga
of hacks and bugs for devm_memremap_pages() users.
Beyond the fixups to teach existing paths how to retrieve the 'usemap'
from a section, and updates to usemap allocation path, there are no
expected behavior changes.
Link: http://lkml.kernel.org/r/156092349845.979959.73333291612799019.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Wei Yang <richardw.yang@linux.intel.com>
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com> [ppc64]
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Jane Chu <jane.chu@oracle.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Qian Cai <cai@lca.pw>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
xdarklight
pushed a commit
to xdarklight/linux
that referenced
this issue
Jul 9, 2019
Patch series "mm: Sub-section memory hotplug support", v10.
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 a subsection-v10 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]: pmem/ndctl#76
[4]: https://git.kernel.org/pub/scm/linux/kernel/git/djbw/nvdimm.git/log/?h=subsection-v10
[5]: pmem/ndctl@7c59b4867e1c
This patch (of 13):
Towards enabling memory hotplug to track partial population of a section,
introduce 'struct mem_section_usage'.
A pointer to a 'struct mem_section_usage' instance replaces the existing
pointer to a 'pageblock_flags' bitmap. Effectively it adds one more
'unsigned long' beyond the 'pageblock_flags' (usemap) allocation to house
a new 'subsection_map' bitmap. The new bitmap enables the memory
hot{plug,remove} implementation to act on incremental sub-divisions of a
section.
SUBSECTION_SHIFT is defined as global constant instead of per-architecture
value like SECTION_SIZE_BITS in order to allow cross-arch compatibility of
subsection users. Specifically a common subsection size allows for the
possibility that persistent memory namespace configurations be made
compatible across architectures.
The primary motivation for this functionality is to support platforms that
mix "System RAM" and "Persistent Memory" within a single section, or
multiple PMEM ranges with different mapping lifetimes within a single
section. The section restriction for hotplug has caused an ongoing saga
of hacks and bugs for devm_memremap_pages() users.
Beyond the fixups to teach existing paths how to retrieve the 'usemap'
from a section, and updates to usemap allocation path, there are no
expected behavior changes.
Link: http://lkml.kernel.org/r/156092349845.979959.73333291612799019.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Wei Yang <richardw.yang@linux.intel.com>
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com> [ppc64]
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Jane Chu <jane.chu@oracle.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Qian Cai <cai@lca.pw>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
avagin
pushed a commit
to avagin/linux
that referenced
this issue
Jul 12, 2019
Patch series "mm: Sub-section memory hotplug support", v10.
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 a subsection-v10 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]: pmem/ndctl#76
[4]: https://git.kernel.org/pub/scm/linux/kernel/git/djbw/nvdimm.git/log/?h=subsection-v10
[5]: pmem/ndctl@7c59b4867e1c
This patch (of 13):
Towards enabling memory hotplug to track partial population of a section,
introduce 'struct mem_section_usage'.
A pointer to a 'struct mem_section_usage' instance replaces the existing
pointer to a 'pageblock_flags' bitmap. Effectively it adds one more
'unsigned long' beyond the 'pageblock_flags' (usemap) allocation to house
a new 'subsection_map' bitmap. The new bitmap enables the memory
hot{plug,remove} implementation to act on incremental sub-divisions of a
section.
SUBSECTION_SHIFT is defined as global constant instead of per-architecture
value like SECTION_SIZE_BITS in order to allow cross-arch compatibility of
subsection users. Specifically a common subsection size allows for the
possibility that persistent memory namespace configurations be made
compatible across architectures.
The primary motivation for this functionality is to support platforms that
mix "System RAM" and "Persistent Memory" within a single section, or
multiple PMEM ranges with different mapping lifetimes within a single
section. The section restriction for hotplug has caused an ongoing saga
of hacks and bugs for devm_memremap_pages() users.
Beyond the fixups to teach existing paths how to retrieve the 'usemap'
from a section, and updates to usemap allocation path, there are no
expected behavior changes.
Link: http://lkml.kernel.org/r/156092349845.979959.73333291612799019.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Wei Yang <richardw.yang@linux.intel.com>
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com> [ppc64]
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Jane Chu <jane.chu@oracle.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Qian Cai <cai@lca.pw>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jason Gunthorpe <jgg@mellanox.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
avagin
pushed a commit
to avagin/linux
that referenced
this issue
Jul 15, 2019
Patch series "mm: Sub-section memory hotplug support", v10.
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 a subsection-v10 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]: pmem/ndctl#76
[4]: https://git.kernel.org/pub/scm/linux/kernel/git/djbw/nvdimm.git/log/?h=subsection-v10
[5]: pmem/ndctl@7c59b4867e1c
This patch (of 13):
Towards enabling memory hotplug to track partial population of a section,
introduce 'struct mem_section_usage'.
A pointer to a 'struct mem_section_usage' instance replaces the existing
pointer to a 'pageblock_flags' bitmap. Effectively it adds one more
'unsigned long' beyond the 'pageblock_flags' (usemap) allocation to house
a new 'subsection_map' bitmap. The new bitmap enables the memory
hot{plug,remove} implementation to act on incremental sub-divisions of a
section.
SUBSECTION_SHIFT is defined as global constant instead of per-architecture
value like SECTION_SIZE_BITS in order to allow cross-arch compatibility of
subsection users. Specifically a common subsection size allows for the
possibility that persistent memory namespace configurations be made
compatible across architectures.
The primary motivation for this functionality is to support platforms that
mix "System RAM" and "Persistent Memory" within a single section, or
multiple PMEM ranges with different mapping lifetimes within a single
section. The section restriction for hotplug has caused an ongoing saga
of hacks and bugs for devm_memremap_pages() users.
Beyond the fixups to teach existing paths how to retrieve the 'usemap'
from a section, and updates to usemap allocation path, there are no
expected behavior changes.
Link: http://lkml.kernel.org/r/156092349845.979959.73333291612799019.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Wei Yang <richardw.yang@linux.intel.com>
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com> [ppc64]
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Jane Chu <jane.chu@oracle.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Qian Cai <cai@lca.pw>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jason Gunthorpe <jgg@mellanox.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
avagin
pushed a commit
to avagin/linux
that referenced
this issue
Jul 16, 2019
Patch series "mm: Sub-section memory hotplug support", v10.
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 a subsection-v10 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]: pmem/ndctl#76
[4]: https://git.kernel.org/pub/scm/linux/kernel/git/djbw/nvdimm.git/log/?h=subsection-v10
[5]: pmem/ndctl@7c59b4867e1c
This patch (of 13):
Towards enabling memory hotplug to track partial population of a section,
introduce 'struct mem_section_usage'.
A pointer to a 'struct mem_section_usage' instance replaces the existing
pointer to a 'pageblock_flags' bitmap. Effectively it adds one more
'unsigned long' beyond the 'pageblock_flags' (usemap) allocation to house
a new 'subsection_map' bitmap. The new bitmap enables the memory
hot{plug,remove} implementation to act on incremental sub-divisions of a
section.
SUBSECTION_SHIFT is defined as global constant instead of per-architecture
value like SECTION_SIZE_BITS in order to allow cross-arch compatibility of
subsection users. Specifically a common subsection size allows for the
possibility that persistent memory namespace configurations be made
compatible across architectures.
The primary motivation for this functionality is to support platforms that
mix "System RAM" and "Persistent Memory" within a single section, or
multiple PMEM ranges with different mapping lifetimes within a single
section. The section restriction for hotplug has caused an ongoing saga
of hacks and bugs for devm_memremap_pages() users.
Beyond the fixups to teach existing paths how to retrieve the 'usemap'
from a section, and updates to usemap allocation path, there are no
expected behavior changes.
Link: http://lkml.kernel.org/r/156092349845.979959.73333291612799019.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Wei Yang <richardw.yang@linux.intel.com>
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com> [ppc64]
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Jane Chu <jane.chu@oracle.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Qian Cai <cai@lca.pw>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jason Gunthorpe <jgg@mellanox.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
avagin
pushed a commit
to avagin/linux
that referenced
this issue
Jul 17, 2019
Patch series "mm: Sub-section memory hotplug support", v10.
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 a subsection-v10 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]: pmem/ndctl#76
[4]: https://git.kernel.org/pub/scm/linux/kernel/git/djbw/nvdimm.git/log/?h=subsection-v10
[5]: pmem/ndctl@7c59b4867e1c
This patch (of 13):
Towards enabling memory hotplug to track partial population of a section,
introduce 'struct mem_section_usage'.
A pointer to a 'struct mem_section_usage' instance replaces the existing
pointer to a 'pageblock_flags' bitmap. Effectively it adds one more
'unsigned long' beyond the 'pageblock_flags' (usemap) allocation to house
a new 'subsection_map' bitmap. The new bitmap enables the memory
hot{plug,remove} implementation to act on incremental sub-divisions of a
section.
SUBSECTION_SHIFT is defined as global constant instead of per-architecture
value like SECTION_SIZE_BITS in order to allow cross-arch compatibility of
subsection users. Specifically a common subsection size allows for the
possibility that persistent memory namespace configurations be made
compatible across architectures.
The primary motivation for this functionality is to support platforms that
mix "System RAM" and "Persistent Memory" within a single section, or
multiple PMEM ranges with different mapping lifetimes within a single
section. The section restriction for hotplug has caused an ongoing saga
of hacks and bugs for devm_memremap_pages() users.
Beyond the fixups to teach existing paths how to retrieve the 'usemap'
from a section, and updates to usemap allocation path, there are no
expected behavior changes.
Link: http://lkml.kernel.org/r/156092349845.979959.73333291612799019.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Wei Yang <richardw.yang@linux.intel.com>
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com> [ppc64]
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Jane Chu <jane.chu@oracle.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Qian Cai <cai@lca.pw>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jason Gunthorpe <jgg@mellanox.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
mrchapp
pushed a commit
to mrchapp/linux
that referenced
this issue
Jul 18, 2019
Patch series "mm: Sub-section memory hotplug support", v10.
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 a subsection-v10 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]: pmem/ndctl#76
[4]: https://git.kernel.org/pub/scm/linux/kernel/git/djbw/nvdimm.git/log/?h=subsection-v10
[5]: pmem/ndctl@7c59b4867e1c
This patch (of 13):
Towards enabling memory hotplug to track partial population of a section,
introduce 'struct mem_section_usage'.
A pointer to a 'struct mem_section_usage' instance replaces the existing
pointer to a 'pageblock_flags' bitmap. Effectively it adds one more
'unsigned long' beyond the 'pageblock_flags' (usemap) allocation to house
a new 'subsection_map' bitmap. The new bitmap enables the memory
hot{plug,remove} implementation to act on incremental sub-divisions of a
section.
SUBSECTION_SHIFT is defined as global constant instead of per-architecture
value like SECTION_SIZE_BITS in order to allow cross-arch compatibility of
subsection users. Specifically a common subsection size allows for the
possibility that persistent memory namespace configurations be made
compatible across architectures.
The primary motivation for this functionality is to support platforms that
mix "System RAM" and "Persistent Memory" within a single section, or
multiple PMEM ranges with different mapping lifetimes within a single
section. The section restriction for hotplug has caused an ongoing saga
of hacks and bugs for devm_memremap_pages() users.
Beyond the fixups to teach existing paths how to retrieve the 'usemap'
from a section, and updates to usemap allocation path, there are no
expected behavior changes.
Link: http://lkml.kernel.org/r/156092349845.979959.73333291612799019.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Wei Yang <richardw.yang@linux.intel.com>
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com> [ppc64]
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Jane Chu <jane.chu@oracle.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Qian Cai <cai@lca.pw>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jason Gunthorpe <jgg@mellanox.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
mrchapp
pushed a commit
to mrchapp/linux
that referenced
this issue
Jul 19, 2019
Patch series "mm: Sub-section memory hotplug support", v10.
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 a subsection-v10 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]: pmem/ndctl#76
[4]: https://git.kernel.org/pub/scm/linux/kernel/git/djbw/nvdimm.git/log/?h=subsection-v10
[5]: pmem/ndctl@7c59b4867e1c
This patch (of 13):
Towards enabling memory hotplug to track partial population of a section,
introduce 'struct mem_section_usage'.
A pointer to a 'struct mem_section_usage' instance replaces the existing
pointer to a 'pageblock_flags' bitmap. Effectively it adds one more
'unsigned long' beyond the 'pageblock_flags' (usemap) allocation to house
a new 'subsection_map' bitmap. The new bitmap enables the memory
hot{plug,remove} implementation to act on incremental sub-divisions of a
section.
SUBSECTION_SHIFT is defined as global constant instead of per-architecture
value like SECTION_SIZE_BITS in order to allow cross-arch compatibility of
subsection users. Specifically a common subsection size allows for the
possibility that persistent memory namespace configurations be made
compatible across architectures.
The primary motivation for this functionality is to support platforms that
mix "System RAM" and "Persistent Memory" within a single section, or
multiple PMEM ranges with different mapping lifetimes within a single
section. The section restriction for hotplug has caused an ongoing saga
of hacks and bugs for devm_memremap_pages() users.
Beyond the fixups to teach existing paths how to retrieve the 'usemap'
from a section, and updates to usemap allocation path, there are no
expected behavior changes.
Link: http://lkml.kernel.org/r/156092349845.979959.73333291612799019.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Wei Yang <richardw.yang@linux.intel.com>
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com> [ppc64]
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Jane Chu <jane.chu@oracle.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Qian Cai <cai@lca.pw>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jason Gunthorpe <jgg@mellanox.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
nathanchance
pushed a commit
to ClangBuiltLinux/linux
that referenced
this issue
Jul 19, 2019
Patch series "mm: Sub-section memory hotplug support", v10.
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 a subsection-v10 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]: pmem/ndctl#76
[4]: https://git.kernel.org/pub/scm/linux/kernel/git/djbw/nvdimm.git/log/?h=subsection-v10
[5]: pmem/ndctl@7c59b4867e1c
This patch (of 13):
Towards enabling memory hotplug to track partial population of a section,
introduce 'struct mem_section_usage'.
A pointer to a 'struct mem_section_usage' instance replaces the existing
pointer to a 'pageblock_flags' bitmap. Effectively it adds one more
'unsigned long' beyond the 'pageblock_flags' (usemap) allocation to house
a new 'subsection_map' bitmap. The new bitmap enables the memory
hot{plug,remove} implementation to act on incremental sub-divisions of a
section.
SUBSECTION_SHIFT is defined as global constant instead of per-architecture
value like SECTION_SIZE_BITS in order to allow cross-arch compatibility of
subsection users. Specifically a common subsection size allows for the
possibility that persistent memory namespace configurations be made
compatible across architectures.
The primary motivation for this functionality is to support platforms that
mix "System RAM" and "Persistent Memory" within a single section, or
multiple PMEM ranges with different mapping lifetimes within a single
section. The section restriction for hotplug has caused an ongoing saga
of hacks and bugs for devm_memremap_pages() users.
Beyond the fixups to teach existing paths how to retrieve the 'usemap'
from a section, and updates to usemap allocation path, there are no
expected behavior changes.
Link: http://lkml.kernel.org/r/156092349845.979959.73333291612799019.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Wei Yang <richardw.yang@linux.intel.com>
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com> [ppc64]
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Jane Chu <jane.chu@oracle.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Qian Cai <cai@lca.pw>
Cc: Logan Gunthorpe <logang@deltatee.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jason Gunthorpe <jgg@mellanox.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mdeejay
pushed a commit
to BeastRoms-Devices/kernel_xiaomi_raphael
that referenced
this issue
Aug 16, 2019
commit ae86cbfef3818300f1972e52f67a93211acb0e24 upstream. Commit cfe30b8 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b8 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
mdeejay
pushed a commit
to BeastRoms-Devices/kernel_xiaomi_cepheus
that referenced
this issue
Aug 22, 2019
commit ae86cbfef3818300f1972e52f67a93211acb0e24 upstream. Commit cfe30b8 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b8 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Flatty11
pushed a commit
to Flatty11/SM-G9750
that referenced
this issue
Oct 10, 2019
commit ae86cbfef3818300f1972e52f67a93211acb0e24 upstream. Commit cfe30b872058 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b872058 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
freeza-inc
pushed a commit
to freeza-inc/bm-galaxy-note10-sd855-pie
that referenced
this issue
Dec 24, 2019
commit ae86cbfef3818300f1972e52f67a93211acb0e24 upstream. Commit cfe30b872058 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b872058 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Inkypen79
pushed a commit
to Inkypen79/kernel_xiaomi_andromeda_old
that referenced
this issue
Jan 15, 2020
commit ae86cbf upstream. Commit cfe30b8 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b8 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
|
Is this problem solved? |
|
Yeah, the patch and upstream kernel stuff fixed it. Didn't see any alerts until your ask Friday, got memory holed I'm sure when I got sidetracked before closing. I'm surprised no one came and smacked me up for leaving an issue open for so long. |
krazey
pushed a commit
to krazey/android_kernel_motorola_exynos9610
that referenced
this issue
Apr 23, 2022
commit ae86cbfef3818300f1972e52f67a93211acb0e24 upstream. Commit cfe30b8 "libnvdimm, pmem: adjust for section collisions with 'System RAM'" enabled Linux to workaround occasions where platform firmware arranges for "System RAM" and "Persistent Memory" to collide within a single section boundary. Unfortunately, as reported in this issue [1], platform firmware can inflict the same collision between persistent memory regions. The approach of interrogating iomem_resource does not work in this case because platform firmware may merge multiple regions into a single iomem_resource range. Instead provide a method to interrogate regions that share the same parent bus. This is a stop-gap until the core-MM can grow support for hotplug on sub-section boundaries. [1]: pmem/ndctl#76 Fixes: cfe30b8 ("libnvdimm, pmem: adjust for section collisions with...") Cc: <stable@vger.kernel.org> Reported-by: Patrick Geary <patrickg@supermicro.com> Tested-by: Patrick Geary <patrickg@supermicro.com> Reviewed-by: Vishal Verma <vishal.l.verma@intel.com> Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
This is all against Intel DCP DIMMs what were provisioned via ipmctl. I've tried mainline ipmctl builds to no avail; as well as attempting to use the latest 4.18.18-200.fc28.x86_64 fedora kernel.
[root@redacted ~]# ndctl create-namespace -r region0
{
"dev":"namespace0.0",
"mode":"fsdax",
"map":"dev",
"size":"248.00 GiB (266.29 GB)",
"uuid":"f0ba1c10-0cf2-4572-bfec-7f8e5e4098f7",
"raw_uuid":"f9c4a0a8-df1c-4389-89d4-6d5c8bac80d7",
"sector_size":512,
"blockdev":"pmem0",
"numa_node":0
}
[root@redacted ~]# ndctl create-namespace -r region1
libndctl: ndctl_pfn_enable: pfn1.0: failed to enable
Error: namespace1.0: failed to enable
failed to create namespace: No such device or address
I'm seeing OOP's spit out by the kernel (This may not be exact; I just grabbed the first one out of messages with a matching pfn):
Very similar to issue 39 ; I will attempt it with the alignment forced once I swap all the hardware back in. I'm posting this since now, even lacking data since the closing of 39 ended with "We have alignment fixes."
Interestingly it does not do this for all dimms; it seems totally based in how I lay out the physical dimms.
This is all Intel Optane DCP, The 2-2-2 configurations I have don't seem to have this issue; but 2-1-1 symmetric population does. I've seen it happen on either IMC; but it seems to be more prolific on IMC0. Asymmetric population does not appear to have the issue either. Additionally, I've only seen this when the dimm is in 0% Memory Mode (All storage). 50% ratios do not have this problem.
I'll bump the issue once I swap all the hardware back in and try with forced alignment.
The text was updated successfully, but these errors were encountered: