Re: [PATCH rfc 6/9] mm: migrate: support poisoned recover from migrate folio

[Kefeng Wang <wangkefeng.wang@huawei.com>]

On 2024/2/2 11:04, Kefeng Wang wrote:
> 
> 
> On 2024/2/2 4:34, Matthew Wilcox wrote:
>> On Mon, Jan 29, 2024 at 03:09:31PM +0800, Kefeng Wang wrote:
>>> In order to support poisoned folio copy recover from migrate folio,
>>> let's use folio_mc_copy() and move it in the begin of the function
>>> of __migrate_folio(), which could simply error handling since there
>>> is no turning back if folio_migrate_mapping() return success, the
>>> downside is the folio copied even though folio_migrate_mapping()
>>> return fail, a small optimization is to check whether folio does
>>> not have extra refs before we do more work ahead in __migrate_folio(),
>>> which could help us avoid unnecessary folio copy.
>>
>> OK, I see why you've done it this way.
>>
>> Would it make more sense if we pulled the folio refcount freezing
>> out of folio_migrate_mapping() into its callers?  That way
>> folio_migrate_mapping() could never fail.
> 
Question, the folio ref freezing is under the xas_lock_irq(), it can't
be moved out of lock, and if with xas lock irq, we couldn't call
folio_mc_copy(), so the above way is not feasible, or maybe I missing
something?