Re: [PATCH v7 5/5] x86/ioreq server: Synchronously reset outstanding p2m_ioreq_server entries when an ioreq server unmaps.

[Andrew Cooper <andrew.cooper3@citrix.com>]

On 08/03/17 15:33, Yu Zhang wrote:
> After an ioreq server has unmapped, the remaining p2m_ioreq_server
> entries need to be reset back to p2m_ram_rw. This patch does this
> synchronously by iterating the p2m table.
>
> The synchronous resetting is necessary because we need to guarantee
> the p2m table is clean before another ioreq server is mapped. And
> since the sweeping of p2m table could be time consuming, it is done
> with hypercall continuation.
>
> Signed-off-by: Yu Zhang <yu.c.zhang@linux.intel.com>
> ---
> Cc: Paul Durrant <paul.durrant@citrix.com>
> Cc: Jan Beulich <jbeulich@suse.com>
> Cc: Andrew Cooper <andrew.cooper3@citrix.com>
> Cc: George Dunlap <george.dunlap@eu.citrix.com>
>
> changes in v1: 
>   - This patch is splitted from patch 4 of last version.
>   - According to comments from Jan: update the gfn_start for
>     when use hypercall continuation to reset the p2m type.
>   - According to comments from Jan: use min() to compare gfn_end
>     and max mapped pfn in p2m_finish_type_change()
> ---
>  xen/arch/x86/hvm/dm.c          | 43 +++++++++++++++++++++++++++++++++++++-----
>  xen/arch/x86/mm/p2m.c          | 29 ++++++++++++++++++++++++++++
>  xen/include/asm-x86/p2m.h      |  5 +++++
>  xen/include/public/hvm/dm_op.h |  3 +--
>  4 files changed, 73 insertions(+), 7 deletions(-)
>
> diff --git a/xen/arch/x86/hvm/dm.c b/xen/arch/x86/hvm/dm.c
> index f97478b..a92d5d7 100644
> --- a/xen/arch/x86/hvm/dm.c
> +++ b/xen/arch/x86/hvm/dm.c
> @@ -288,6 +288,7 @@ static int inject_event(struct domain *d,
>      return 0;
>  }
>  
> +#define DMOP_op_mask 0xff
>  static int dm_op(domid_t domid,
>                   unsigned int nr_bufs,
>                   xen_dm_op_buf_t bufs[])
> @@ -315,10 +316,8 @@ static int dm_op(domid_t domid,
>      }
>  
>      rc = -EINVAL;
> -    if ( op.pad )
> -        goto out;
>  
> -    switch ( op.op )
> +    switch ( op.op & DMOP_op_mask )

Nack to changes like this.  HVMop continuations only existed in this
form because we had to retrofit it to longterm-stable ABIs in the past,
and there were literally no free bits anywhere else.

Currently, the DMOP ABI isn't set in stone, so you have until code
freeze in 4.9 to make changes.  (i.e. soon now.)  We should also
consider which other DMops might potentially need to become continuable,
and take preventative action before the freeze.


If we need to make similar changes once the ABI truely is frozen, there
are plenty of free bits in the end of the union which can be used
without breaking the ABI.

~Andrew.

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel