Re: [PATCH 0/6] arm64/mm: Enable memory hot remove and ZONE_DEVICE

From: Dan Williams <dan.j.williams@intel.com>
To: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	linux-arm-kernel@lists.infradead.org,
	Linux MM <linux-mm@kvack.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Will Deacon <will.deacon@arm.com>,
	Catalin Marinas <catalin.marinas@arm.com>,
	Michal Hocko <mhocko@suse.com>,
	Mel Gorman <mgorman@techsingularity.net>,
	james.morse@arm.com, Mark Rutland <mark.rutland@arm.com>,
	Robin Murphy <robin.murphy@arm.com>,
	cpandya@codeaurora.org, arunks@codeaurora.org, osalvador@suse.de,
	Logan Gunthorpe <logang@deltatee.com>,
	Pavel Tatashin <pasha.tatashin@oracle.com>,
	David Hildenbrand <david@redhat.com>,
	cai@lca.pw
Subject: Re: [PATCH 0/6] arm64/mm: Enable memory hot remove and ZONE_DEVICE
Date: Wed, 3 Apr 2019 11:08:53 -0700	[thread overview]
Message-ID: <CAPcyv4gdz7L20nSMEBNLDWgUzr-GjaBhecF3i8Q4D_O=ug0qNw@mail.gmail.com> (raw)
In-Reply-To: <1554265806-11501-1-git-send-email-anshuman.khandual@arm.com>

On Tue, Apr 2, 2019 at 9:30 PM Anshuman Khandual
<anshuman.khandual@arm.com> wrote:
>
> This series enables memory hot remove on arm64, fixes a memblock removal
> ordering problem in generic __remove_memory(), enables sysfs memory probe
> interface on arm64. It also enables ZONE_DEVICE with struct vmem_altmap
> support.
>
> Testing:
>
> Tested hot remove on arm64 for all 4K, 16K, 64K page config options with
> all possible VA_BITS and PGTABLE_LEVELS combinations. Tested ZONE_DEVICE
> with ARM64_4K_PAGES through a dummy driver.
>
> Build tested on non arm64 platforms. I will appreciate if folks can test
> arch_remove_memory() re-ordering in __remove_memory() on other platforms.
>
> Dependency:
>
> V5 series in the thread (https://lkml.org/lkml/2019/2/14/1096) will make
> kernel linear mapping loose pgtable_page_ctor() init. When this happens
> the proposed functions free_pte|pmd|pud_table() in [PATCH 2/6] will have
> to stop calling pgtable_page_dtor().

Hi Anshuman,

I'd be interested to integrate this with the sub-section hotplug
support [1]. Otherwise the padding implementation in libnvdimm can't
be removed unless all ZONE_DEVICE capable archs also agree on the
minimum arch_add_memory() granularity. I'd prefer not to special case
which archs support which granularity, but it unfortunately
complicates what you're trying to achieve.

I think at a minimum we, mm hotplug co-travellers, need to come to a
consensus on whether sub-section support is viable for v5.2 and / or a
pre-requisite for new arch-ZONE_DEVICE implementations.