> > are you talking about this comment?
> > 		>	PGO_LOCKED: fault data structures are locked
> > 
> 
> No, this one which is in the function arguments...
> 
>     struct vm_page **pps /* IN/OUT */,
>     int *npagesp /* IN (OUT if PGO_LOCKED)*/,
> 
> hmmm but that could be interpreted to mean something else.

it just means it's an input/output of the function. :)

> Still, I am puzzled by your concern - if all is well then the per page
> checking does not change anything.  If the pages mismatch then, I
> think, I follow the same process as the genfs code when the page in
> fails.

uvn_get(PGO_LOCKED) is usually called with spin locks held.
thus you can't sleep.  otoh, a function you are calling seems sleepable.
so i think it's broken.

YAMAMOTO Takashi