[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <f2b82be4-2ff2-65f0-5b44-e6fbb7e610c9@redhat.com>
Date: Mon, 25 Jan 2021 12:23:12 +0100
From: David Hildenbrand <david@...hat.com>
To: Oscar Salvador <osalvador@...e.de>
Cc: Anshuman Khandual <anshuman.khandual@....com>,
akpm@...ux-foundation.org, mhocko@...nel.org,
linux-kernel@...r.kernel.org, linux-mm@...ck.org, vbabka@...e.cz,
pasha.tatashin@...een.com
Subject: Re: [PATCH 2/5] mm,memory_hotplug: Allocate memmap from the added
memory range
On 25.01.21 12:18, Oscar Salvador wrote:
> On Mon, Jan 25, 2021 at 11:57:20AM +0100, David Hildenbrand wrote:
>> I'm confused.
>>
>> 1. Assume we hotplug memory, online it to ZONE_MOVABLE. The vmemmap gets
>> allocated from altmap space.
>
> The vmemmap could have never been allocated from altmap in case hpage vmemmap
> feature is enabled.
>
> Have a look at [1].
> If is_hugetlb_free_vmemmap_enabled(), vmemmap_populate() ends up calling
> vmemmap_populate_basepages().
Oh, it calls "vmemmap_populate_basepages(start, end, node, NULL);"
the "NULL" part is the important bit.
>
> And since no memory was consumed from altmap, and hence altmap_alloc_block_buf()
> was never called, vmem_altmap->alloc will be 0, and memory_block->nr_vmemmap_pages
> will be 0 as well.
>
> But on a second though, true is that we will get in trouble if hpage vmemmap
> feature ever gets to work with vmemmap_populate_hugepages.
> I will queue that to look in a new future.
This seriously needs comments and documentation. The problem is where to
document as long as one of both series is not merged yet :)
At least in the cover letter, because this is not obvious how both
things will play along.
Thanks!
--
Thanks,
David / dhildenb
Powered by blists - more mailing lists