Re: Cross Memory Attach v4

Re: Cross Memory Attach v4

[Daniel J Blueman]

On Jul 27, 11:30 am, Christopher Yeoh <cyeoh@au1.ibm.com> wrote:
> Hi Andrew,
>
> Here's an updated version of the Cross Memory Attach patch. Changes
> since the v3:
>
> - Adds x86_64 specific wire up
> - Changes behaviour so process_vm_readv and process_vm_writev return
>   the number of bytes successfully read or written even if an error
>   occurs
> - Adds more kernel doc interface comments
> - rename of some internal functions (process_vm_rw_check_iovecs,
>   process_vm_rw) so they make more sense.
>
> Still need to do benchmarking to see if the optimisation for small
> copies using a local on-stack array in process_vm_rw_core is worth it.
[]

A bit late to the party.

get_user_pages_fast may be a nice win here, since the pages will
almost always be present; it'll backoff to the locked get_user_pages
path to handle the pathological case of swapped pages.

Thanks,
  Daniel
-- 
Daniel J Blueman

Re: Cross Memory Attach v4

[Christopher Yeoh]

On Mon, 1 Aug 2011 10:06:31 +0800
Daniel J Blueman <daniel.blueman@gmail.com> wrote:

> On Jul 27, 11:30 am, Christopher Yeoh <cyeoh@au1.ibm.com> wrote:
> > Hi Andrew,
> >
> > Here's an updated version of the Cross Memory Attach patch. Changes
> > since the v3:
> >
> > - Adds x86_64 specific wire up
> > - Changes behaviour so process_vm_readv and process_vm_writev return
> >   the number of bytes successfully read or written even if an error
> >   occurs
> > - Adds more kernel doc interface comments
> > - rename of some internal functions (process_vm_rw_check_iovecs,
> >   process_vm_rw) so they make more sense.
> >
> > Still need to do benchmarking to see if the optimisation for small
> > copies using a local on-stack array in process_vm_rw_core is worth
> > it.
> []
> 
> A bit late to the party.
> 
> get_user_pages_fast may be a nice win here, since the pages will
> almost always be present; it'll backoff to the locked get_user_pages
> path to handle the pathological case of swapped pages.

Can't use get_user_pages_fast here as in this case we want to get pages
from a different process.

Chris
-- 
cyeoh@au.ibm.com