[PATCH v4 4/6] mm/hotplug: Allow pageblock alignment via altmap reservation

[David Hildenbrand]

On 24.07.23 17:16, Aneesh Kumar K V wrote:

>>
>> /*
>>   * In "forced" memmap_on_memory mode, we always align the vmemmap size up to cover
>>   * full pageblocks. That way, we can add memory even if the vmemmap size is not properly
>>   * aligned, however, we might waste memory.
>>   */
> 
> I am finding that confusing. We do want things to be pageblock_nr_pages aligned both ways.
> With MEMMAP_ON_MEMORY_FORCE, we do that by allocating more space for memmap and
> in the default case we do that by making sure only memory blocks of specific size supporting
> that alignment can use MEMMAP_ON_MEMORY feature.

See the usage inm hp_supports_memmap_on_memory(), I guess that makes 
sense then.

But if you have any ideas on how to clarify that (terminology), I'm all 
ears!

[...]

>>> +    return arch_supports_memmap_on_memory(size);
>>>    }
>>>      /*
>>> @@ -1311,7 +1391,11 @@ int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)
>>>    {
>>>        struct mhp_params params = { .pgprot = pgprot_mhp(PAGE_KERNEL) };
>>>        enum memblock_flags memblock_flags = MEMBLOCK_NONE;
>>> -    struct vmem_altmap mhp_altmap = {};
>>> +    struct vmem_altmap mhp_altmap = {
>>> +        .base_pfn =  PHYS_PFN(res->start),
>>> +        .end_pfn  =  PHYS_PFN(res->end),
>>> +        .reserve  = memory_block_align_base(resource_size(res)),
>>
>> Can you remind me why we have to set reserve here at all?
>>
>> IOW, can't we simply set
>>
>> .free = memory_block_memmap_on_memory_size();
>>
>> end then pass
>>
>> mhp_altmap.alloc + mhp_altmap.free
>>
>> to create_memory_block_devices() instead?
>>
> 
> But with the dax usage of altmap, altmap->reserve is what we use to reserve things to get
> the required alignment. One difference is where we allocate the struct page at. For this specific
> case it should not matter.
> 
> static unsigned long __meminit vmem_altmap_next_pfn(struct vmem_altmap *altmap)
> {
> 	return altmap->base_pfn + altmap->reserve + altmap->alloc
> 		+ altmap->align;
> }
> 
> And other is where we online a memory block
> 
> We find the start pfn using mem->altmap->alloc + mem->altmap->reserve;
> 
> Considering altmap->reserve is what dax pfn_dev use, is there a reason you want to use altmap->free for this?

"Reserve" is all about "reserving that much memory for driver usage".

We don't care about that. We simply want vmemmap allocations coming from 
the pageblock(s) we set aside. Where exactly, we don't care.

> I find it confusing to update free when we haven't allocated any altmap blocks yet.

"
@reserve: pages mapped, but reserved for driver use (relative to @base)"
@free: free pages set aside in the mapping for memmap storage
@alloc: track pages consumed, private to vmemmap_populate()
"

To me, that implies that we can ignore "reserve". We set @free to the 
aligned value and let the vmemmap get allocated from anything in there.

free + alloc should always sum up to our set-aside pageblock(s), no?


-- 
Cheers,

David / dhildenb