Re: [PATCH V7 3/3] arm64/mm: Enable memory hot remove

From: David Hildenbrand <david@redhat.com>
To: Catalin Marinas <catalin.marinas@arm.com>,
	Anshuman Khandual <anshuman.khandual@arm.com>
Cc: linux-mm@kvack.org, linux-kernel@vger.kernel.org,
	linux-arm-kernel@lists.infradead.org, akpm@linux-foundation.org,
	will@kernel.org, mark.rutland@arm.com, mhocko@suse.com,
	ira.weiny@intel.com, cai@lca.pw, logang@deltatee.com,
	cpandya@codeaurora.org, arunks@codeaurora.org,
	dan.j.williams@intel.com, mgorman@techsingularity.net,
	osalvador@suse.de, ard.biesheuvel@arm.com, steve.capper@arm.com,
	broonie@kernel.org, valentin.schneider@arm.com,
	Robin.Murphy@arm.com, steven.price@arm.com,
	suzuki.poulose@arm.com
Subject: Re: [PATCH V7 3/3] arm64/mm: Enable memory hot remove
Date: Wed, 11 Sep 2019 12:31:47 +0200	[thread overview]
Message-ID: <eede2498-b2fa-9905-9020-31337045b00d@redhat.com> (raw)
In-Reply-To: <20190910161759.GI14442@C02TF0J2HF1T.local>

On 10.09.19 18:17, Catalin Marinas wrote:
> On Tue, Sep 03, 2019 at 03:15:58PM +0530, Anshuman Khandual wrote:
>> @@ -770,6 +1022,28 @@ int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
>>  void vmemmap_free(unsigned long start, unsigned long end,
>>  		struct vmem_altmap *altmap)
>>  {
>> +#ifdef CONFIG_MEMORY_HOTPLUG
>> +	/*
>> +	 * FIXME: We should have called remove_pagetable(start, end, true).
>> +	 * vmemmap and vmalloc virtual range might share intermediate kernel
>> +	 * page table entries. Removing vmemmap range page table pages here
>> +	 * can potentially conflict with a concurrent vmalloc() allocation.
>> +	 *
>> +	 * This is primarily because vmalloc() does not take init_mm ptl for
>> +	 * the entire page table walk and it's modification. Instead it just
>> +	 * takes the lock while allocating and installing page table pages
>> +	 * via [p4d|pud|pmd|pte]_alloc(). A concurrently vanishing page table
>> +	 * entry via memory hot remove can cause vmalloc() kernel page table
>> +	 * walk pointers to be invalid on the fly which can cause corruption
>> +	 * or worst, a crash.
>> +	 *
>> +	 * So free_empty_tables() gets called where vmalloc and vmemmap range
>> +	 * do not overlap at any intermediate level kernel page table entry.
>> +	 */
>> +	unmap_hotplug_range(start, end, true);
>> +	if (!vmalloc_vmemmap_overlap)
>> +		free_empty_tables(start, end);
>> +#endif
>>  }
>>  #endif	/* CONFIG_SPARSEMEM_VMEMMAP */
> 
> I wonder whether we could simply ignore the vmemmap freeing altogether,
> just leave it around and not unmap it. This way, we could call
> unmap_kernel_range() for removing the linear map and we save some code.
> 
> For the linear map, I think we use just above 2MB of tables for 1GB of
> memory mapped (worst case with 4KB pages we need 512 pte pages). For
> vmemmap we'd use slightly above 2MB for a 64GB hotplugged memory. Do we
> expect such memory to be re-plugged again in the same range? If we do,
> then I shouldn't even bother with removing the vmmemmap.
> 

FWIW, I think we should do it cleanly.

> I don't fully understand the use-case for memory hotremove, so any
> additional info would be useful to make a decision here.
> 

Especially in virtual environment, hotremove will be relevant. For
physical environments - I have no idea how important that is for ARM.

-- 

Thanks,

David / dhildenb