On Tue, Nov 26, 2024 at 12:11:34PM +0100, Martin Pieuchot wrote:
> uao_get() and uvn_get() are the two `pgo_get' implementations used by
> the lower layer fault handler.  First they are called with PGO_LOCKED
> to return resident pages, then if needed, they are called again with
> PGO_SYNCIO to fetch content of the page from disk (or cache).
> 
> The comment for uao_get() explains the following:
> 
>  /*
>   * uao_get: fetch me a page     
>   *
>   * we have three cases:
>   * 1: page is resident     -> just return the page.
>   * 2: page is zero-fill    -> allocate a new page and zero it.
>   * 3: page is swapped out  -> fetch the page from swap.
>   *
>   * cases 1 can be handled with PGO_LOCKED, cases 2 and 3 cannot.
> [...]
>  */ 
> 
> However case 2 is currently optimistically handled with PGO_LOCKED, if
> the NOWAIT allocation succeeds.
> 
> Diff below changes that and sync the PGO_LOCKED case with uvn_get() in
> preparation for faulting ahead resident pages in parallel.
> 
> The explanation for this change is that PGO_LOCKED will be soon a read-
> only operation, so we do not want any modification on the underlying UVM
> object.
> 
> While here reduce some cosmetic differences in sync with NetBSD.

ok tb