On 07.05.19 23:02, Dan Williams wrote:
> On Tue, May 7, 2019 at 11:38 AM David Hildenbrand <david@xxxxxxxxxx> wrote:
>>
>> Let's prepare for better error handling while adding memory by allowing
>> to use arch_remove_memory() and __remove_pages() even if
>> CONFIG_MEMORY_HOTREMOVE is not set. CONFIG_MEMORY_HOTREMOVE effectively
>> covers
>> - Offlining of system ram (memory block devices) - offline_pages()
>> - Unplug of system ram - remove_memory()
>> - Unplug/remap of device memory - devm_memremap()
>>
>> This allows e.g. for handling like
>>
>> arch_add_memory()
>> rc = do_something();
>> if (rc) {
>> arch_remove_memory();
>> }
>>
>> Whereby do_something() will for example be memory block device creation
>> after it has been factored out.
>
> What's left after this?
I tried to describe this above here:
- Offlining of system ram (memory block devices) - offline_pages()
- Unplug of system ram - remove_memory()
- Unplug/remap of device memory - devm_memremap()
So administrators cannot trigger offlining/removal of memory.
> Can we just get rid of CONFIG_MEMORY_HOTREMOVE
> option completely when CONFIG_MEMORY_HOTPLUG is enabled? It's not
> clear to me why there was ever the option to compile out the remove
> code when the add code is included.
I guess it was a configure option because initially, offlining/unplug
was extremely unstable. Now it's only slightly unstable :D
I would actually favor getting rid of CONFIG_MEMORY_HOTREMOVE
completely. After all, unplug always has to be triggered by an admin (HW
or software).
Opinions?
--
Thanks,
David / dhildenb
